 |
|
| If you can't view the Datasheet, Please click here to try to view without PDF Reader . |
|
|
|
| If you can't view the Datasheet, Please click here to try to view without PDF Reader . |
|
|
| Datasheet File OCR Text: |
| publication# 18681 rev. b amendment /1 issue date: october 1994 this document contains information on a product under development at advanced micro devices, inc. the information is intended to help you to evaluate this product. amd reserves the right to change or discontinue work on this proposed product without notice. advanced micro devices AM79C974 pcnet tm -scsi combination ethernet and scsi controller for pci systems preliminary
distinctive characteristics pci features n direct glueless interface to 33 mhz, 32-bit pci local bus n 132 mbyte/s burst dma transfer rate n compliant to pci local bus specification revision 2.0 ethernet features n supports iso 8802-3 (ieee/ansi 802.3) and ethernet standards n high-performance bus master architecture with integrated dma buffer management unit for low cpu and bus utilization n individual 136-byte transmit and 128-byte receive fifos provide frame buffering for increased system latency n microwire tm eeprom interface supports jumperless design n integrated manchester encoder/decoder n provides integrated attachment unit interface (aui) and 10base-t transceiver with automatic port selection n automatic twisted-pair receive polarity detec- tion and automatic correction of the receive polarity n dynamic transmit fcs generation programma- ble on a frame-by-frame basis n internal/external loopback capabilities n supports the following types of network interfaces: aui to external 10base2, 10base5, 10base-t or 10base-f mau internal 10base-t transceiver with smart squelch to twisted-pair medium scsi features n compliant to ansi standards x3.131 C 1986 (scsi-1) and x3.131 C 199x (scsi-2) n fast 8-bit scsi-2 10 mbyte/s synchronous or 7 mbyte/s asynchronous data transfer rate n scsi specific bus mastering dma engine (32-bit address/data) n 96-byte dma fifo for low bus latency n on-chip state machine to control the scsi sequences in hardware n integrated industry standard fast scsi-2 core n single-ended 48 ma outputs to drive the scsi bus directly n support for scatter-gather dma data transfers n hooks in silicon and software to enable disk drive spin down for power savings general features n software compatible with amd's am79c960 pcnet-isa, am79c961 pcnet-isa+, am79c965 pcnet-32, am79c970 pcnet-pci register and descriptor architecture n plug-in and software compatible with amd's pc scsi family of scsi controllers for pci n nand tree test mode for connectivity testing on printed circuit boards n single +5 v power supply operation n low-power, cmos design with sleep modes for both ethernet and scsi controllers allows re- duced power consumption for critical battery powered applications and `green pcs' n fully static design for low frequency and power operation n 132-pin pqfp package general description the pcnet-scsi combination ethernet and 8-bit fast scsi controller with a 32-bit pci bus interface is a highly integrated ethernet-fast scsi system solution de- signed to address high-performance system application requirements. this single-chip is a flexible bus-master- ing device that can be used in many applications, includ- ing network- and scsi-ready pcs, printers, fax modems, and bridge/router designs. the bus-master architecture provides high data throughput in the sys- tem and low cpu and system bus utilization. the pcnet-scsi controller is fabricated with amds ad- vanced low-power cmos process to provide low oper- ating and standby current for power sensitive applications.
amd p r e l i m i n a r y 2 AM79C974 the pcnet-scsi is part of amds pci product family of plug-in and software compatible scsi and ethernet controllers. this product compatibility ensures a low cost system upgrade path and lower motherboard manufacturing costs. ethernet specific the pcnet-scsi controller includes a complete ether- net node integrated into a single vlsi device. it contains a bus interface unit, a dma buffer management unit, an ieee 802.3-defined media access control (mac) func- tion, individual 136-byte transmit and 128-byte receive fifos, an ieee 802.3-defined attachment unit inter- face (aui) and twisted-pair transceiver media attach- ment unit (10base-t mau), and a microwire eeprom interface. the pcnet-scsi controller is also register compatible with the lance (am7990) ethernet control- ler, the c-lance (am79c90) ethernet controller, the ilacc (am79c900) ethernet controller, and all ether- net controllers in the pcnet family, including the pcnet- isa controller (am79c960), the pcnet-isa+ controller (am79c961), and the pcnet-32 controller (am79c965). the buffer management unit supports the lance, ilacc, and pcnet descriptor software models. the pcnet-scsi controller is software compatible with the novell ne2100 and ne1500 ethernet adapter card architectures. in addition, a sleep function has been in- corporated to provide low standby current, excellent for notebooks and green pcs. the 32-bit multiplexed bus interface unit provides a di- rect interface to the pci local bus applications, simplify- ing the design of an ethernet node in a pc system. with its built-in support for both little and big endian byte alignment, this controller also addresses proprietary non-pc applications. the pcnet-scsi controller supports auto configuration in the pci configuration space. additional pcnet-scsi controller configuration parameters, including the unique ieee physical address, can be read from an ex- ternal non-volatile memory (serial eeprom) immedi- ately following system reset. the controller also has the capability to automatically select either the aui port or the twisted-pair trans- ceiver. only one interface is active at any one time. the individual transmit and receive fifos optimize system overhead, providing sufficient latency during frame transmission and reception, and minimizing intervention during normal network error recovery. the integrated manchester encoder/decoder (mendec) eliminates the need for an external serial interface adapter (sia) in the system. in addition, the device provides program- mable on-chip led drivers for transmit, receive, colli- sion, receive polarity, link integrity or jabber status. scsi specific the pcnet-scsi controller also includes a high- performance fast scsi controller with a glueless inter- face to the pci local bus. the pcnet-scsi integrates its own 32-bit bus mastering dma engine with an industry standard fast scsi-2 block. the dma engine and ac- companying 96 byte dma fifo allow 32-bit burst data transfers across the high bandwidth pci bus at speeds of up to 132 mbyte/s. full support for scatter-gather dma transfers optimize performance in multi-tasking system applications. the pcnet-scsis on-chip state machine controls scsi bus sequences in hardware and is coupled with the bus mastering dma engine to eliminate the need for an on- chip risc processor. this results in a smaller die size giving the AM79C974 superior price/performance ver- sus competitive offerings. amd supports the AM79C974 with a total system solu- tion which includes: a full suite of licensable scsi drivers and utilities fully tested under the following operating system environments: dos 5.0 C 6.0 windows 3.1 windows nt os/2 2.x netware 3.x, 4.x sco unix 3.2.4, odt 2.0 an int13h compatible scsi rom bios aspi compatibility complete hardware reference design kit for more detailed information on the pcnet-scsi refer to the technical manual, pid #18738a. p r e l i m i n a r y amd 3 AM79C974 high level block diagram 18681a-1 scsi data scsi control 10base-t, aui ports scsi sequences, scsi control, scsi registers 802.3 mac core scsi fifo dma registers rcv fifo fifo control xmt fifo dma registers dma control dma fifo 96 bytes fifo & dma control pci host control and interface pci data/address pci host control amd p r e l i m i n a r y 4 AM79C974 cpu control address data core logic pci to isa cache sram dram memory pcnet-scsi (AM79C974) video control pc-at isa bus super i/o ide/floppy ser/par keyboard control 18681a-2 pci bus AM79C974 in a pci system p r e l i m i n a r y amd 5 AM79C974 table of contents distinctive characteristics 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . general description 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . high level block diagram 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . related products 11 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . connection diagram 12 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ordering information 13 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . pin designations 14 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . listed by pin number 14 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . listed by pin name 15 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . quick reference pin description 16 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . listed by driver type 17 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . logic symbol 18 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . pin description 19 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . pci bus interface 19 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ethernet controller pins 22 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . board interface 22 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . microwire eeprom interface 23 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . attachment unit interface 24 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . twisted-pair interface 24 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . scsi controller pins 24 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . test interface 25 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . miscellaneous 25 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . power supply 25 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . basic functions 26 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . system bus interface function 26 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . software interface 26 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ethernet interfaces 26 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . scsi interfaces 26 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . detailed functions 27 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . bus interface unit (biu) 27 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . slave configuration transfers 27 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . slave i/o transfers 29 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . bus acquisition 31 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . bus master dma transfers 32 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . target initiated termination 37 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . master initiated termination 40 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ethernet controller 43 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . buffer management unit (bmu) 43 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . initialization 43 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . re-initialization 43 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . buffer management 43 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . descriptor rings 43 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . descriptor ring access mechanism 44 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . polling 46 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . transmit descriptor table entry (tdte) 46 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . receive descriptor table entry (rdte) 48 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . amd p r e l i m i n a r y 6 AM79C974 media access control 48 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . transmit and receive message data encapsulation 48 . . . . . . . . . . . . . . . . . . . . . . . . . . . media access management 50 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . manchester encoder/decoder (mendec) 52 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . external crystal characteristics 52 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . external clock drive characteristics 52 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mendec transmit path 52 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . transmitter timing and operation 52 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . receiver path 53 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . input signal conditioning 53 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . clock acquisition 53 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . pll tracking 54 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . carrier tracking and end of message 54 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . data decoding 54 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . differential input terminations 54 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . collision detection 55 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . jitter tolerance definition 55 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . attachment unit interface (aui) 55 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . twisted-pair transceiver (t-mau) 55 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . twisted-pair transmit function 55 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . twisted-pair receive function 55 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . link test function 56 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . polarity detection and reversal 56 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . twisted-pair interface status 56 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . collision detect function 57 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . signal quality error (sqe) test (heartbeat) function 57 . . . . . . . . . . . . . . . . . . . . . . . . . . . jabber function 57 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . power down 57 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10base-t interface connection 57 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ethernet power savings modes 58 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . software access 58 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ethernet pci configuration registers 58 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . i/o resources 59 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . i/o register access 61 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . hardware access 64 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . pcnet-scsi controller master accesses 64 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . slave access to i/o resources 64 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . eeprom microwire access 66 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . transmit operation 69 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . transmit function programming 69 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . automatic pad generation 69 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . transmit fcs generation 70 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . transmit exception conditions 70 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . receive operation 71 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . receive function programming 71 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . automatic pad stripping 71 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . receive fcs checking 72 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . receive exception conditions 72 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . loopback operation 72 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . led support 73 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . p r e l i m i n a r y amd 7 AM79C974 h_reset, s_reset, and stop 74 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . h_reset 74 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . s_reset 74 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . stop 74 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . scsi controller 75 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . scsi specific dma engine 75 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . dma fifo 76 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . dma blast command 76 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . funneling logic 76 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . scsi dma programming sequence 76 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mdl based dma programming 76 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . dma scatter-gather mechanism 78 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . memory descriptor list (mdl) 78 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . dma scatter-gather operation (4k aligned elements) 78 . . . . . . . . . . . . . . . . . . . . . . . . dma scatter-gather operation (non-4k aligned elements mdl not set) 80 . . . . . . . . . . interrupts 81 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . the fast scsi block 81 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . scsi block id 81 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . scsi fifo threshold 81 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . data transmission 81 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . req / ack control 81 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . parity 82 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . parity checking on the scsi bus 82 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . parity generating on the scsi bus 82 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . reset levels 82 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . hard reset: (h) 82 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . soft reset: (s) 82 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . disconnected reset: (d) 83 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . device commands 83 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . command stacking 84 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . invalid commands 84 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . command window 84 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . initiator commands 84 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . information transfer command 84 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . initiator command complete steps 85 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . message accepted command 85 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . transfer pad bytes command 85 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . set atn command 85 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . reset atn command 86 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . idle state commands 86 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . select without atn steps command 86 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . select with atn steps command 86 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . select with atn and stop steps command 86 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . enable selection/reselection command 86 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . disable selection/reselection command 87 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . select with atn 3 steps command 87 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . general commands 87 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . no operation command 87 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . clear fifo command 87 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . reset device command 87 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . reset scsi bus command 87 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . amd p r e l i m i n a r y 8 AM79C974 scsi power management features 87 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . scsi activity pin 87 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . reduced power mode 87 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . power down pin (pwdn pin) 88 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . software disk spin-down 88 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . nand tree testing 89 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . absolute maximum ratings 92 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . operating ranges 92 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . dc characteristics : pci bus and board interface 92 . . . . . . . . . . . . . . . . . . . . . . . . . . . dc characteristics : attachment unit interface 93 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . dc characteristics : 10base-t interface 93 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . dc characteristics : scsi interface 94 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . dc characteristics : capacitance, esd, and latch up 95 . . . . . . . . . . . . . . . . . . . . . . . . ac switching characteristics : pci bus and board interface 96 . . . . . . . . . . . . . . . . . ac switching characteristics : 10base-t interface 97 . . . . . . . . . . . . . . . . . . . . . . . . . ac switching characteristics : attachment unit interface 98 . . . . . . . . . . . . . . . . . . . ac switching characteristics : scsi interface 99 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . key to switching waveforms 101 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ac switching test circuits 101 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ac switching waveforms : system bus interface 103 . . . . . . . . . . . . . . . . . . . . . . . . . . . ac switching waveforms : 10base-t interface 105 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ac switching waveforms : attachment unit interface 107 . . . . . . . . . . . . . . . . . . . . . . . . ac switching waveforms : scsi interface 110 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . physical dimensions 112 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . appendix a C register summary 113 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ethernet controller control and status registers 114 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . bcr bus configuration registers 117 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . scsi controller scsi register map 118 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . dma register map 118 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . appendix b C pcnet-scsi compatible media interface modules 119 . . . . . . . . . . . . . . . . . . appendix c C recommendation for power and ground decoupling 121 . . . . . . . . . . . . . . . appendix d C alternative method for initialization of ethernet controller 123 . . . . . . . . . . . appendix e C scsi system considerations 124 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . appendix f C designing a single motherboard for amd pci family 132 . . . . . . . . . . . . . . . p r e l i m i n a r y amd 9 AM79C974 list of figures figure 1. slave configuration read 27 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . figure 2. slave configuration write 28 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . figure 3. slave i/o read 29 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . figure 4. slave i/o write 30 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . figure 5. bus acquisition 31 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . figure 6. non-burst read cycles with wait states 32 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . figure 7. non-burst read cycles without wait states 33 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . figure 8. non-burst read cycles with and without wait states 34 . . . . . . . . . . . . . . . . . . . . . . . . figure 9. burst read cycles 35 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . figure 10. burst write cycles 36 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . figure 11. disconnect with data transfer 37 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . figure 12. disconnect without data transfer 38 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . figure 13. target abort 39 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . figure 14. preemption when frame is deasserted 40 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . figure 15. preemption when frame is asserted 41 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . figure 16. master abort 42 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . figure 17. 16-bit data structures: initialization block and descriptor rings 44 . . . . . . . . . . . . . . . figure 18. 32-bit data structures: initialization block and descriptor rings 45 . . . . . . . . . . . . . . . figure 19. receiver block diagram 53 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . figure 20. differential input termination 54 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . figure 21. 10base-t interface connection 57 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . figure 22. iso 8802-3 (ieee/ansi 802.3) data frame 69 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . figure 23. 802.3 frame and length field transmission order 72 . . . . . . . . . . . . . . . . . . . . . . . . . figure 24. led control logic 74 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . figure 25. pci biu C dma engine C scsi block 75 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . figure 26. dma fifo to scsi fifo interface 76 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . figure 27. AM79C974 nand tree test structure 89 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . figure 28. nand tree waveform 91 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . figure 29. scsi clock input 110 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . figure 30. asynchronous initiator transmit 110 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . figure 31. asynchronous initiator receive 111 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . figure 32. synchronous initiator transmit 111 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . figure 33. synchronous initiator receive 111 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . figure e-1. ideal routing scheme for scsi 124 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . figure e-2. a poor routing scheme for scsi 125 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . figure e-3. motherboard layout C approach #1 126 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . figure e-4. motherboard layout C approach #2 127 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . figure e-5. decoupling capacitor placement 129 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . figure e-6. regulated termination 130 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . figure f-1. pci family connections 134 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . amd p r e l i m i n a r y 10 AM79C974 list of tables table 1. crystal specifications 52 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . table 2. clock drive characteristics 52 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . table 3. bus master accesses 64 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . table 4. bus slave accesses 65 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . table 5. eeprom contents 68 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . table 6. the dma registers 77 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . table 7. summary of scsi commands 83 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . table 8. nand tree configuration 90 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . p r e l i m i n a r y amd 11 AM79C974 related products part no. description am33c93a synchronous scsi controller am386 a high-performance 32-bit microprocessor am486 tm high-performance 32-bit microprocessor am53c94/96 high-performance scsi controller am53c974 pc scsi tm bus mastering fast scsi controller for pci systems am53cf94/96 enhanced fast scsi-2 controller am79c90 cmos local area network controller for ethernet (c-lance) am79c98 twisted-pair ethernet transceiver (tpex) am79c100 twisted-pair ethernet transceiver plus (tpex+) am79c900 integrated local area communications controller tm (ilacc tm ) am79c940 media acces controller for ethernet (mace tm ) am79c960 pcnet-isa single-chip ethernet controller (for isa bus) am79c961 pcnet-isa + single-chip ethernet controller (with microsoft plug n play support) am79c965 pcnet-32 single-chip 32-bit ethernet controller (for 386dx, 486 and vl buses) am79c970 pcnet-pci single-chip ethernet controller for pci local bus am79c981 integrated multiport repeater plus tm (imr+ tm ) am79c987 hardware implemented management information base tm (himib tm ) am7990 local area network controller for ethernet (lance) am7996 ieee 802.3/ethernet/cheapernet tap transceiver am85c30 enhanced serial communication controller amd p r e l i m i n a r y 12 AM79C974 connection diagram pin 1 is marked for orientation. reserve = dont connect. 18681a-3 132 1 ad28 ad29 131 v ssb 130 ad30 129 ad31 128 reqa 127 reqb 126 v ss 125 gnta 124 gntb 123 v dd 122 clk 121 rst 120 v ss 119 intb 118 inta 117 reserve 116 sleep 115 eecs 114 dv ss 113 eesk/ led1 112 eedi/ lnkst 111 eedo/ led3 110 dv dd 109 av dd2 108 ci+ 107 ci- 106 di+ 105 di- 104 av dd1 103 do+ 102 do- 101 av ss1 100 xtal2 99 av ss2 98 xtal1 97 av dd3 96 txd+ 95 txp+ 94 txd- 93 txp- 92 av dd4 91 rxd+ 90 rxd- 89 dv ss 88 i/o 87 c/d 86 msg 85 v dd 84 ack 83 v ssbs 82 req 81 sel 80 dv ss 79 sd p 78 sd 7 77 v ddbs 76 sd 6 75 sd 5 74 sd 4 73 v ssbs 72 sd 3 71 sd 2 70 sd 1 69 sd 0 68 v ssbs 67 34 par 35 c/ be 1 36 ad15 37 v ssb 38 ad14 39 ad13 40 ad12 41 ad11 42 ad10 43 v ssb 44 ad9 45 ad8 46 v ddb 47 c/ be 0 48 ad7 49 ad6 50 v ssb 51 ad5 52 ad4 53 ad3 54 ad2 55 v ssb 56 ad1 57 ad0 58 59 v dd 60 scsiclk 61 v ss 62 busy 63 v ss 64 bsy 65 atn 66 scsi^rst v ddb 2 ad27 3 ad26 4 v ssb 5 ad25 6 ad24 7 c/ be 3 8 v dd 9 idsela 10 idselb 11 v ss 12 ad23 13 ad22 14 v ssb 15 ad21 16 ad20 17 v ddb 18 ad19 19 ad18 20 v ssb 21 ad17 22 ad16 23 c/ be 2 24 frame 25 irdy 26 trdy 27 devsel 28 stop 29 lock 30 v ss 31 perr 32 serr 33 v ddb pwdn AM79C974 pcnet-scsi p r e l i m i n a r y amd 13 AM79C974 ordering information standard products amd standard products are available in several packages and operating ranges. the order number (valid combination) is formed by a combination of: temperature range c = commercial (0 c to +70 c) package type (per prod. nomenclature) k = plastic quad flat pack trimmed and formed (pqb132) speed option not applicable device number/description AM79C974 pcnet-scsi combination ethernet and scsi controller for pci systems AM79C974 valid combinations kc\w valid combinations alternate packaging option \w = trimmed and formed in a tray AM79C974 k c \w valid combinations list configurations planned to be supported in volume for this device. consult the local amd sales office to confirm availability of specific valid combinations and to check on newly released combinations. optional processing blank = standard processing amd p r e l i m i n a r y 14 AM79C974 pin designations listed by pin number pin no. pin name pin no. pin name pin no. pin name pin no. pin name 1v ddb 34 par 67 v ssbs 100 av ss1 2 ad27 35 c/ be 168 sd 0 101 doC 3 ad26 36 ad15 69 sd 1 102 do+ 4v ssb 37 v ssb 70 sd 2 103 av dd1 5 ad25 38 ad14 71 sd 3 104 diC 6 ad24 39 ad13 72 v ssbs 105 di+ 7c/ be 3 40 ad12 73 sd 4 106 ciC 8v dd 41 ad11 74 sd 5 107 ci+ 9 idsela 42 ad10 75 sd 6 108 av dd2 10 idselb 43 v ssb 76 v ddbs 109 dv dd 11 v ss 44 ad9 77 sd 7 110 eedo/ led3 12 ad23 45 ad8 78 sd p 111 eedi/ lnkst 13 ad22 46 v ddb 79 dv ss 112 eesk/ led1 14 v ssb 47 c/ be 080 sel 113 dv ss 15 ad21 48 ad7 81 req 114 eecs 16 ad20 49 ad6 82 v ssbs 115 sleep 17 v ddb 50 v ssb 83 ack 116 reserve 18 ad19 51 ad5 84 dv dd 117 inta 19 ad18 52 ad4 85 msg 118 intb 20 v ssb 53 ad3 86 c/d 119 v ss 21 ad17 54 ad2 87 i/o 120 rst 22 ad16 55 v ssb 88 dv ss 121 clk 23 c/ be 2 56 ad1 89 rxdC 122 v dd 24 frame 57 ad0 90 rxd+ 123 gntb 25 irdy 58 pwdn 91 av dd4 124 gnta 26 trdy 59 v dd 92 txpC 125 v ss 27 devsel 60 scsiclk 93 txdC 126 reqb 28 stop 61 v ss 94 txp+ 127 reqa 29 lock 62 busy 95 txd+ 128 ad31 30 v ss 63 v ss 96 av dd3 129 ad30 31 perr 64 bsy 97 xtal1 130 v ssb 32 serr 65 atn 98 av ss2 131 ad29 33 v ddb 66 scsi^rst 99 xtal2 132 ad28 p r e l i m i n a r y amd 15 AM79C974 pin designations listed by pin name pin name pin no. pin name pin no. pin name pin no. pin name pin no. ack 83 atn 65 gntb 123 stop 28 ad0 57 av dd1 103 idsela 9 trdy 26 ad1 56 av dd2 108 idsel 10 xtal1 97 ad2 54 av dd3 96 inta 117 xtal2 99 ad3 53 av dd4 91 intb 118 txdC 93 ad4 52 av ss1 100 i/o 87 txd+ 95 ad5 51 av ss2 98 irdy 25 txpC 92 ad6 49 bsy 64 lock 29 txp+ 94 ad7 48 busy 62 msg 85 v dd 8 ad8 45 c/ be 0 47 par 34 v dd 59 ad9 44 c/ be 135 perr 31 v dd 122 ad10 42 c/ be 2 23 pwdn 58 v ddb 1 ad11 41 c/ be 37 req 81 v ddb 17 ad12 40 c/d 86 reqa 127 v ddb 33 ad13 39 clk 121 reqb 126 v ddb 46 ad14 38 ciC 106 reserve 116 v ddbs 76 ad15 36 ci+ 107 rst 120 v ss 11 ad16 22 devsel 27 rxdC 89 v ss 30 ad17 21 diC 104 rxd+ 90 v ss 61 ad18 19 di+ 105 scsiclk 60 v ss 63 ad19 18 doC 101 scsi^rst 66 v ss 119 ad20 16 do+ 102 sd 068 v ss 125 ad21 15 dv dd 84 sd 169 v ssb 4 ad22 13 dv dd 109 sd 270 v ssb 14 ad23 12 dv ss 79 sd 371 v ssb 20 ad24 6 dv ss 88 sd 473 v ssb 37 ad25 5 dv ss 113 sd 574 v ssb 43 ad26 3 eecs 114 sd 675 v ssb 50 ad27 2 eedi/ lnkst 111 sd 777 v ssb 55 ad28 132 eedo/ led3 110 sd p78 v ssb 130 ad29 131 eesk/ led1 112 sel 80 v ssbs 67 ad30 129 frame 24 serr 32 v ssbs 72 ad31 128 gnta 124 sleep 115 v ssbs 82 amd p r e l i m i n a r y 16 AM79C974 pin designations quick reference pin description pin name description type driver # pins pci bus interface ad[31:00] address/data bus io ts3 32 c/ be [3:0] bus command/byte enable io ts3 4 clk bus clock i na 1 devsel device select io ts6 1 frame cycle frame io ts6 1 gnta , gntb bus grant i na 1 idsela, idselb initialization device select i na 1 inta , intb interrupt io od6 1 irdy initiator ready io ts6 1 lock bus lock io ts6 1 par parity io ts6 1 perr parity error io ts6 1 reqa , reqb bus request io ts3 1 rst reset i na 1 serr system error io od6 1 stop stop io ts6 1 trdy target ready io ts6 1 ethernet specific board interface eecs microwire serial prom chip select o o8 1 eedi/ lnkst microwire serial eeprom data in/link status o led 1 eedo/ led3 microwire aprom data out/led predriver io led 1 eesk/ led1 microwire serial prom clock/led1 io led 1 sleep sleep mode i na 1 xtal1C2 crystal input/output io na 2 attachment unit interface (aui) ci+/ciC aui collision differential pair i na 2 di+/diC aui data in differential pair i na 2 do+/doC aui data out differential pair o do 2 10base-t interface rxd+/rxdC receive differential pair i na 2 txd+/txdC transmit differential pair o tdo 2 txp+/txpC transmit pre-distortion differential pair o tpo 2 lnkst /eedi link status/microwire serial eeprom data in o led 1 p r e l i m i n a r y amd 17 AM79C974 pin designations (continued) quick reference pin description pin name description type driver # pins scsi specific scsi interface sd [7:0] scsi data io od48 8 sd p scsi data parity io od48 1 msg message i 1 c/d command/data i 1 i/o input/output i 1 atn attention o od48 1 bsy busy io od48 1 sel select io od48 1 scsi^rst scsi bus reset io od48 1 req request i 1 ack acknowledge o od48 1 miscellaneous scsi clk scsi core clock i 1 reserve reserved, do not connect i 1 power management pwdn power down indicator i 1 test interface busy nand tree test output o o3 1 power supplies av dd analog power p na 4 av ss analog ground p na 2 v dd /dv dd digital power p na 5 v ss /dv ss digital ground p na 9 v ddb /v ddbs i/o buffer power p na 5 v ssb /v ssbs i/o buffer ground p na 11 listed by driver type the following table describes the various types of drivers that are implemented in the pcnet-scsi controller. current is given as milliamperes: name type i ol (ma) i oh (ma) pf ts3 tri-state tm 3 C2.0 50 ts6 tri-state 6 C2.0 50 o3 totem pole 3 C0.4 50 o6 totem pole 6 C0.4 50 o8 totem pole 8 C0.4 50 od6 open drain 6 na 50 od48 open drain 48 na led led 12 C0.4 50 amd p r e l i m i n a r y 18 AM79C974 logic symbol ad [31:0] c/ be [3:0] par frame trdy irdy stop devsel idsela reqa gnta clk rst inta lock perr serr sd [7:0] sd p msg c/d i/o atn bsy sel scsi^rst req ack scsi clk reserve pwdn busy pcnet-scsi (AM79C974) 18248b-4 scsi power management signals pci interface v dd v ss ci+/C di+/C xtal1 xtal2 do+/C rxd+/C txd+/C txp+/C eedi/ linkst eecs eesk/ led1 eedo/ led3 ethernet test interface idselb intb reqb gntb p r e l i m i n a r y amd 19 AM79C974 pin description pci bus interface ad[31:00] address and data input/output, active high these signals are multiplexed on the same pci pins. during the first clock of a transaction ad[31:00] contain the physical byte address (32 bits). during the subse- quent clocks ad[31:00] contain data. byte ordering is lit- tle endian by default. ad[07:00] are defined as least significant byte and ad[31:24] are defined as the most significant byte. for fifo data transfers, the pcnet- scsi controller can be programmed for big endian byte ordering. see csr3, bit 2 (bswp) for more details. during the address phase of the transaction, when the pcnet-scsi controller is a bus master, ad[31:2] will ad- dress the active dword (double-word). the pcnet- scsi controller always drives ad[1:0] to 00 during the address phase indicating linear burst order. when the pcnet-scsi controller is not a bus master, the ad[31:00] lines are continuously monitored to deter- mine if an address match exists for i/o slave transfers. during the data phase of the transaction, ad[31:00] are driven by the pcnet-scsi controller when performing bus master writes and slave read operations. data on ad[31:00] is latched by the pcnet-scsi controller when performing bus master reads and slave write operations. when rst is active, ad[31:0] are inputs for nand tree testing. c/ be [3:0] bus command and byte enables input/output, active low these signals are multiplexed on the same pci pins. during the address phase of the transaction, c/ be [3:0] define the bus command. during the data phase c/ be [3:0] are used as byte enables. the byte enables define which physical byte lanes carry meaningful data. c/ be 0 applies to byte 0 (ad[07:00]) and c/ be 3 applies to byte 3 (ad[31:24]). the function of the byte enables is independent of the byte ordering mode (csr3, bit 2). when rst is active, c/ be [3:0] are inputs for nand tree testing. clk clock input this signal provides timing for all the transactions on the pci bus and all pci devices on the bus including the pcnet-scsi controller. all bus signals are sampled on the rising edge of clk and all parameters are defined with respect to this edge. the pcnet-scsi controller op- erates over a range of 0 to 33 mhz. when rst is active, clk is an input for nand tree testing. devsel device select input/output, active low this signal when actively driven by the pcnet-scsi controller as a slave device signals to the master device that the pcnet-scsi controller has decoded its address as the target of the current access. as an input it indi- cates whether any device on the bus has been selected. when rst is active, devsel is an input for nand tree testing. frame cycle frame input/output, active low this signal is driven by the pcnet-scsi controller when it is the bus master to indicate the beginning and dura- tion of the access. frame is asserted to indicate a bus transaction is beginning. frame is asserted while data transfers continue. frame is deasserted when the transaction is in the final data phase. when rst is active, frame is an input for nand tree testing. gnta bus grant input, active low this signal indicates that the access to the bus has been granted to the AM79C974s scsi controller. the AM79C974 controller supports bus parking. when the pci bus is idle and the system arbiter asserts gnta without an active reqa from the AM79C974 controller, the controller will actively drive the ad[31:00], c/ be [3:0], and par lines. when rst is active, gnta is an input for nand tree testing. gntb bus grant input, active low this signal indicates that the access to the bus has been granted to the AM79C974s ethernet controller. the AM79C974 controller supports bus parking. when the pci bus is idle and the system arbiter asserts gntb without an active reqb from the AM79C974 controller, the controller will actively drive the ad, c/ be and par lines. amd p r e l i m i n a r y 20 AM79C974 when rst is active, gntb is an input for nand tree testing. idsela initialization device select input, active high this signal is used as a scsi controller selection for the AM79C974 during configuration read and write transaction. when rst is active, idsela is an input for nand tree testing. idselb initialization device select input, active high this signal is used as an ethernet controller selection for the pcnet-scsi controller during configuration read and write transaction. when rst is active, idselb is an input for nand tree testing. inta interrupt request input/output, active low, open drain this signal combines the interrupt requests from both the scsi dma engine and the scsi core. the interrupt source can be determined by reading the scsi dma status register. it is cleared when the status register is read. when rst is active, inta is an input for nand tree test- ing. this is the only time inta is an input. intb interrupt request input/output, active low, open drain an asynchronous attention signal which indicates that one or more of the following status flags is set: babl, miss, merr, rint, idon, rcvcco, rpco, jab, mpco, or txstrt. each status flag has a mask bit which allows for suppression of intb assertion. the flags have the following meaning: babl babble rcvcco receive collision count overflow rpco runt packet count overflow jab jabber miss missed frame merr memory error mpco missed packet count overflow rint receive interrupt idon initialization done txstrt transmit start when rst is active, intb is an input for nand tree testing. this is the only time intb is an input. irdy initiator ready input/output, active low this signal indicates pcnet-scsi controllers ability, as a master device, to complete the current data phase of the transaction. irdy is used in conjunction with the trdy . a data phase is completed on any clock when both irdy and trdy are asserted. during a write irdy indicates that valid data is present on ad[31:00]. during a read irdy indicates that data is accepted by the pcnet-scsi controller as a bus master. wait states are inserted until both irdy and trdy are asserted simul- taneously. when rst is active, irdy is an input for nand tree testing. lock lock input, active low lock is used by the current bus master to indicate an atomic operation that may require multiple transfers. as a slave device, the pcnet-scsi controller can be locked by any master device. when another master at- tempts to access the pcnet-scsi while it is locked, the pcnet-scsi controller will respond by asserting devsel and stop with trdy deasserted (pci retry). the pcnet-scsi controller will never assert lock as a master. when rst is active, lock is an input for nand tree testing. par parity input/output, active high parity is even parity across ad[31:00] and c/ be [3:0]. when the pcnet-scsi controller is a bus master, it generates parity during the address and write data phases. it checks parity during read data phases. when the pcnet-scsi controller operates in slave mode and is the target of the current cycle, it generates parity dur- ing read data phases. it checks parity during address and write data phases. when rst is active, par is an input for nand tree testing. perr parity error input/output, active low, open drain this signal is asserted for one clk by the pcnet-scsi controller when it detects a parity error during any data phase when its ad[31:00] lines are inputs. the perr pin is only active when perren (bit 6) in the pci com- mand register is set. p r e l i m i n a r y amd 21 AM79C974 the pcnet-scsi controller monitors the perr input during a bus master write cycle. it will assert the data parity reported bit in the status register of the configu- ration space when a parity error is reported by the target device. when rst is active, perr is an input for nand tree testing. reqa bus request input/output, active low the AM79C974s scsi controller asserts reqa pin as a signal that it wishes to become a bus master. once as- serted, reqa remains active until gnta has become active. when rst is active, reqa is an input for nand tree testing. this is the only time reqa is an input. reqb bus request input/output, active low the AM79C974s ethernet controller asserts reqb pin as a signal that it wishes to become a bus master. once asserted, reqb remains active until gnt has become active, independent of subsequent assertion of sleep or setting of the stop bit or access to the s_reset port (offset 14h). when rst is active, reqb is an input for nand tree testing. this is the only time reqb is an input. rst reset input, active low when rst is asserted low, then the pcnet-scsi con- troller performs an internal system reset of the type h_reset (hardware_reset). rst must be held for a minimum of 30 clk periods. while in the h_re- set state, the pcnet-scsi controller will disable or deassert all outputs. rst may be asynchronous to the clk when asserted or deasserted. it is recommended that the deassertion be synchronous to guarantee a clean and bounce free edge. when rst is active, nand tree testing is enabled. all pci interface pins are in input mode. the result of the nand tree testing can be observed on the busy output (pin 62). serr system error input/output, active low, open drain this signal is asserted for one clk by the pcnet-scsi controller when it detects a parity error during the ad- dress phase when its ad[31:00] lines are inputs. the serr pin is only active when serren (bit 8) and perren (bit 6) in the pci command register are set. when rst is active, serr is an input for nand tree testing. stop stop input/output, active low in the slave role, the pcnet-scsi controller drives the stop signal to inform the bus master to stop the current transaction. in the bus master role, the pcnet-scsi controller receives the stop signal and stops the cur- rent transaction. when rst is active, stop is an input for nand tree testing. trdy target ready input/output, active low this signal indicates the pcnet-scsi controllers ability as a selected device to complete the current data phase of the transaction. trdy is used in conjunction with the irdy . a data phase is completed on any clock both trdy and irdy are asserted. during a read trdy indi- cates that valid data is present on ad[31:00]. during a write, trdy indicates that data has been accepted. wait states are inserted until both irdy and trdy are asserted simultaneously. when rst is active, trdy is an input for nand tree testing. ethernet controller pins board interface led1 led1 output this pin is shared with the eesk function. as led1 , the function and polarity of this pin are programmable through bcr5. by default, led1 is active low and it in- dicates receive activity on the network. the led1 output from the pcnet-scsi controller is capable of sinking the 12 ma of current necessary to drive an led directly. the led1 pin is also used during eeprom auto-detec- tion to determine whether or not an eeprom is present at the pcnet-scsi controller microwire interface. at the trailing edge of the rst pin, led1 is sampled to deter- mine the value of the eedet bit in bcr19. a sampled high value means that an eeprom is present, and eedet will be set to one. a sampled low value means that an eeprom is not present, and eedet will be set to zero. see the eeprom auto-detection sec- tion for more details. if no led circuit is to be attached to this pin, then a pull up or pull down resistor must be attached instead, in or- der to resolve the eedet setting. amd p r e l i m i n a r y 22 AM79C974 led3 led3 output this pin is shared with the eedo function of the microwire serial eeprom interface. when functioning as led3 , the signal on this pin is programmable through bcr7. by default, led3 is active low and it indicates transmit activity on the network. special attention must be given to the external circuitry attached to this pin. if an led circuit were directly attached to this pin, it would create an i ol requirement that could not be met by the serial eeprom that would also be attached to this pin. therefore, if this pin is to be used as an additional led output while an eeprom is used in the system, then buffering is required between the led3 pin and the led circuit. if no eeprom is included in the system design, then the led3 signal may be directly connected to an led without buffering. the led3 output from the pcnet-scsi controller is capable of sinking the 12 ma of current necessary to drive an led in this case. for more details regarding led connection, see the section on leds. lnkst link status output this pin provides 12 ma for driving an led. by default, it indicates an active link connection on the 10base-t in- terface. this pin can also be programmed to indicate other network status (see bcr4). the lnkst pin polarity is programmable, but by default, it is active low. note that this pin is multiplexed with the eedi function. sleep sleep input when sleep is asserted (active low), the pcnet- scsi controller performs an internal system reset of the s_reset type and then proceeds into a power savings mode. (the reset operation caused by sleep assertion will not affect bcr registers.) the pci interface section is not effected by sleep . in particular, access to the pci configuration space remains possible. none of the configuration registers will be reset by sleep . all i/o accesses to the pcnet-scsi controller will result in a pci target abort response. the pcnet-scsi controller will not assert req while in sleep mode. when sleep is asserted, all non-pci interface outputs will be placed in their normal s_reset condition. all non-pci inter- face inputs will be ignored except for the sleep pin it- self. de-assertion of sleep results in wake-up. the system must refrain from starting the network opera- tions of the pcnet-scsi device for 0.5 seconds follow- ing the deassertion of the sleep signal in order to allow internal analog circuits to stabilize. both clk and xtal1 inputs must have valid clock sig- nals present in order for the sleep command to take effect. if sleep is asserted while req is asserted, then the pcnet-scsi controller will wait for the assertion of gnt . when gnt is asserted, the req signal will be de- asserted and then the pcnet-scsi controller will pro- ceed to the power savings mode. the sleep pin should not be asserted during power supply ramp-up. if it is desired that sleep be asserted at power up time, then the system must delay the asser- tion of sleep until three clk cycles after the comple- tion of a valid pin rst operation. the sleep pin does not affect the scsi section. xtal1 crystal oscillator input input xtal2 crystal oscillator output output the crystal frequency determines the network data rate. the pcnet-scsi controller supports the use of quartz crystals to generate a 20 mhz frequency compatible with the iso 8802-3 (ieee/ansi 802.3) network fre- quency tolerance and jitter specifications. see the sec- tion external crystal characteristics (in section manchester encoder/decoder) for more detail. the network data rate is one-half of the crystal fre- quency. xtal1 may alternatively be driven using an ex- ternal cmos level source, in which case xtal2 must be left unconnected. note that when the pcnet-scsi controller is in coma mode, there is an internal 22 k w re- sistor from xtal1 to ground. if an external source drives xtal1, some power will be consumed driving this resis- tor. if xtal1 is driven low at this time power consump- tion will be minimized. in this case, xtal1 must remain active for at least 30 cycles after the assertion of sleep and deassertion of req . microwire eeprom interface eesk eeprom serial clock input/output the eesk signal is used to access the external iso 8802-3 (ieee/ansi 802.3) address prom. this pin is designed to directly interface to a serial eeprom that uses the microwire interface protocol. eesk is con- nected to the microwire eeproms clock pin. it is con- trolled by either the pcnet-scsi controller directly during a read of the entire eeprom, or indirectly by the host system by writing to bcr19, bit 1. the eesk pin is also used during eeprom auto-detec- tion to determine whether or not an eeprom is present p r e l i m i n a r y amd 23 AM79C974 at the pcnet-scsi controller microwire interface. at the trailing edge of the rst signal, eesk is sampled to de- termine the value of the eedet bit in bcr19. a sam- pled high value means that an eeprom is present, and eedet will be set to one. a sampled low value means that an eeprom is not present, and eedet will be set to zero. see the eeprom auto-detection sec- tion for more details. eesk is shared with the led1 function. if no led circuit is to be attached to this pin, then a pull up or pull down resistor must be attached instead, in order to resolve the eedet setting. eedo eeprom data out input the eedo signal is used to access the external iso 8802-3 (ieee/ansi 802.3) address prom. this pin is designed to directly interface to a serial eeprom that uses the microwire interface protocol. eedo is con- nected to the microwire eeproms data output pin. it is controlled by the eeprom during reads. it may be read by the host system by reading bcr19 bit 0. eedo is shared with the led3 function. eecs eeprom chip select output the function of the eecs signal is to indicate to the microwire eeprom device that it is being accessed. the eecs signal is active high. it is controlled by either the pcnet-scsi controller during command portions of a read of the entire eeprom, or indirectly by the host system by writing to bcr19 bit 2. eedi eeprom data in output the eedi signal is used to access the external iso 8802-3 (ieee/ansi 802.3) address prom. eedi func- tions as an output. this pin is designed to directly inter- face to a serial eeprom that uses the microwire interface protocol. eedi is connected to the microwire eeproms data input pin. it is controlled by either the pcnet-scsi controller during command portions of a read of the entire eeprom, or indirectly by the host sys- tem by writing to bcr19 bit 0. eedi is shared with the lnkst function. attachment unit interface ci collision in input a differential input pair signaling the pcnet-scsi con- troller that a collision has been detected on the network media, indicated by the ci inputs being driven with a 10 mhz pattern of sufficient amplitude and pulse width to meet iso 8802-3 (ieee/ansi 802.3) standards. op- erates at pseudo ecl levels. di data in input a differential input pair to the pcnet-scsi controller car- rying manchester encoded data from the network. oper- ates at pseudo ecl levels. do data out output a differential output pair from the pcnet-scsi controller for transmitting manchester encoded data to the net- work. operates at pseudo ecl levels. twisted-pair interface rxd 10base-t receive data input 10base-t port differential receivers. txd 10base-t transmit data output 10base-t port differential drivers. txp 10base-t pre-distortion control output these outputs provide transmit pre-distortion control in conjunction with the 10base-t port differential drivers. amd p r e l i m i n a r y 24 AM79C974 scsi controller pins scsi bus interface signals scsi bus pins sd [7:0] scsi data input/output, active low, open drain/active negation, schmitt trigger these pins are defined as bi-directional scsi data bus. sd p scsi data parity input/output, active low, open drain/active negation, schmitt trigger this pin is defined as bi-directional scsi data parity. msg message input, active low, schmitt trigger it is a schmitt trigger input in the initiator mode. c/d command/data input, schmitt trigger it is a schmitt trigger input in the initiator mode. i/o input/output input, schmitt trigger it is a schmitt trigger input in the initiator mode. atn attention output, active low, open drain this signal is a 48 ma output in the initiator mode. this signal will be asserted when the device detects a parity error; also, it can be asserted via certain commands. bsy busy input/output, active low, schmitt trigger, open drain as a scsi input signal it has a schmitt trigger and as an output signal it has a 48 ma drive. sel select input/output, active low, schmitt trigger, open drain as a scsi input signal it has a schmitt trigger and as an output signal it has a 48 ma drive. scsi^rst reset input/output , active low, schmitt trigger, open drain as a scsi input signal it has a schmitt trigger and as an output signal it has a 48 ma drive. req request input, active low, schmitt trigger this is a scsi input signal with a schmitt trigger in the initiator mode. ack acknowledge output , active low, open drain/active negation this is a scsi output signal with a 48 ma drive in the initiator mode. scsi clk scsi clock input the scsi clock signal is used to generate all internal de- vice timings. the maximum frequency of this input is 40 mhz and a minimum of 10 mhz is required to main- tain the scsi bus timings. note: a 40 mhz clock must be supplied at this input to achieve 10 mbyte/s synchronous fast scsi transfers. pwdn power down indicator input, active high this signal, when asserted, sets the pwdn status bit in the dma status register and sends an interrupt to the host. test interface busy nand tree out output, active low this signal is logically equivalent to the scsi bus signal bsy . it is duplicated so that external logic can be connected to monitor scsi bus activity. the results of the nand tree testing can be observed on the busy pin where rst is asserted; otherwise, busy will reflect the state of the scsi bus signal line bsy (pin 64). p r e l i m i n a r y amd 25 AM79C974 miscellaneous reserved reserved_do not connect input this pin (#116) is reserved for internal test logic. it must not be connected to anything for proper chip operation. its use is subject to change in future products. power supply pins analog power supply pins av dd analog power (4 pins) power there are four analog +5 v supply pins. special atten- tion should be paid to the printed circuit board layout to avoid excessive noise on these lines. refer to appendix c and the technical manual (pid #18738a) for details. av ss analog ground (2 pins) power there are two analog ground pins. special attention should be paid to the printed circuit board layout to avoid excessive noise on these lines. refer to appendix c and the technical manual (pid #18738a) for details. digital power supply pins v dd / dv dd digital power (5 pins) power there are 5 power supply pins that are used by the inter- nal digital circuitry. all v dd pins must be connected to a +5 v supply. v ddb / v ddb i/o buffer power (5 pins) power there are 5 power supply pins that are used by the pci bus input/output buffer drivers. all v ddb pins must be connected to a +5 v supply. v ss / dv ss digital ground (9 pins) ground there are 9 ground pins that are used by the internal digital circuitry. v ssb / v ssbs i/o buffer ground (11 pins) ground there are 11 ground pins that are used by the pci bus input/output buffer drivers. amd p r e l i m i n a r y 26 AM79C974 basic functions system bus interface function during normal operations the AM79C974 operates as a bus master with a few slave l/o accesses for status and control functions. the ethernet controller is initialized through a combina- tion of pci configuration space accesses, i/o space bus slave accesses, memory space bus master ac- cesses, and optional reads of an external serial eeprom. the eeprom is read through the microwire interface either automatically by the AM79C974 or indi- rectly by a series of bus slave accesses to one of the ethernet bus configuration registers (bcrs). the eeprom normally contains the iso 8802-3 (ieee/ ansi 802.3) ethernet node address and data to be loaded into some of the ethernet bcrs. the scsi controller is initialized by bus slave writes to scsi core and scsi dma registers. software interface the am79c794 uses four address spaces: ethernet pci configuration space, scsi pci configuration space, i/o space, and memory space. scsi pci configuration space is selected when the idsela pin is active. ethernet pci configuration space is selected when the idselb pin is active. the way that idsela and idselb are controlled depends on exter- nal hardware. section 3.6.4.1 of the pci specification recommends two methods of generating configuration cycles called configuration mechanism #1 and configu- ration mechanism #2. the pci configuration spaces are used by system soft- ware to identify the scsi and ethernet controllers and to set up device configuration without the use of jumpers. certain pci configuration registers have read-only infor- mation about the devices resource requirements. other registers are used as mail boxes that system configura- tion software uses to inform other software what re- sources have been allocated to the device. the only pci configuration registers that affect the operation of the am79c794 are the scsi and ethernet base address registers, which are found at offset 10h in each of the two configuration spaces, and the command registers at offset 4. writing to these registers establishes the base address of the scsi l/o space and the base ad- dress of the ethernet i/o space. the scsi controller registers occupy 96 bytes of l/o space that starts on whatever 128-byte boundary that is programmed into the base address register at offset 10h in the scsi pci configuration space. the ethernet controller registers occupy 32 bytes of l/o space that starts on whatever 32-byte boundary that is pro- grammed into the base address register at offset 10h in the ethernet pci configuration space. these registers are used to set up controller operating modes, to enable or disable various features, to start certain op- erations, and to monitor operating status. in addition to the registers in the l/o space, the ethernet controller uses certain data structures that are set up (typically by the host computer) in normal memory space. these data structures are (1) the initialization block that contains configuration data that the ethernet controller automatically loads into its configuration and status registers (csrs), (2) the receive and transmit descriptor rings, that contain pointers to receive and transmit buffers and status and control information about these buffers, and (3) the receive and transmit buffers. the ethernet controller uses bus master ac- cesses to read the locations of the buffers, to store frames received from the network into the receive buff- ers, and to transmit the contents of the transmit buffers. ethernet interfaces the AM79C974 controller can be connected to an 802.3 network via one of two network interfaces. the attach- ment unit interface (aui) provides an iso 8802-3 (ieee/ansi 802.3) compliant differential interface to a remote mau or an on-board transceiver. the 10base-t interface provides a twisted-pair ethernet port. while in auto-selection mode, the interface in use is determined by an auto-sensing mechanism which checks the link status on the 10base-t port. if there is no active link status, then the device assumes an aui connection. scsi interfaces the AM79C974 acts as a bridge between the pci and scsi buses. as the maximum data transfer rate on the pci bus is a very high 132 mbyte/s compared with the scsi bus 10 mbyte/s, buffering is required between the two buses. the buffering is provided by two fifos: a 16-byte (16x8 bits) scsi core fifo and an additional 96-byte (24x32 bits) dma fifo. these fifos provide a temporary storage for all command, data, status and message bytes as they are transferred between the 32-bit pci bus and the 8-bit scsi bus. the AM79C974s scsi core and dma registers are ad- dressed using the value in the base address register (offset 10h in the pci configuration space). the scsi registers occupy 16 double words and the dma engine registers occupy 8 double word locations. the i/o ad- dress map is as follows: start offset end offset block name size 0x0000 0x003f scsi core reg 16 dw/64b 0x0040 0x005f pci dma ccb 8 dw/32b the pci configuration space, ethernet controller and scsi controller are described in detail in the following sections. p r e l i m i n a r y amd 27 AM79C974 detailed functions bus interface unit (biu) the bus interface unit is built of several state machines that run synchronously to clk. one bus interface unit state machine handles accesses where the AM79C974 controller is the bus slave, and another handles ac- cesses where the AM79C974 controller is the bus mas- ter. all inputs are synchronously sampled. all outputs are synchronously generated on the rising edge of clk. in the descriptions that follow, gnt , req , int , and idsel are used to refer to the set of gnta , reqa , inta , and idsela for the scsi controller and to the set of gntb , reqb , intb , and idselb for the ethernet controller, respectively. slave configuration transfers the host can access the AM79C974 pci configuration space with a configuration read or write command. the AM79C974 controller will assert devsel if the idsel input is asserted during the address phase and if the ac- cess is a configuration cycle. devsel is asserted two clock cycles after the host has asserted frame . all configuration cycles are of fixed length. the AM79C974 controller will assert trdy on the 3rd clock of the data phase. slave configuration read the slave configuration read command is used by the host cpu to read the configuration space in the AM79C974 controller. this provides the host cpu with information concerning the device and its capabilities. this is a single cycle, non-burst 8-bit, 16-bit, or 32-bit transfer. 18681a-5 frame clk ad irdy trdy c/ be 123456 addr data 1010 be 's par par par devsel idsel stop figure 1. slave configuration read amd p r e l i m i n a r y 28 AM79C974 slave configuration write the slave configuration write command is used by the host cpu to write the configuration space in the AM79C974 controller. this allows the host cpu to control basic activity of the device, such as enable/dis- able, change i/o location, etc. this is a single cycle, non-burst 8-bit, 16-bit, or 32-bit transfer. 18681a-6 frame clk ad irdy trdy c/ be 123456 addr data 1011 be 's par par par devsel idsel stop figure 2. slave configuration write p r e l i m i n a r y amd 29 AM79C974 slave i/o transfers after the AM79C974 controller is configured as i/o de- vice (by setting ioen in the pci command register), it starts monitoring the pci bus for access to its internal registers. the AM79C974 controller will look for an ad- dress that falls within its i/o address space. the AM79C974 controller will assert devsel if it detects an address match and the access is an i/o cycle. devsel is asserted two clock cycles after the host has asserted frame . the AM79C974 controller will not assert dev- sel if it detects an address match, but the pci com- mand is not of the type i/o read or i/o write. the AM79C974 controller will suspend looking for i/o cycles while being a bus master. slave i/o read the slave i/o read command is used by the host cpu to read the AM79C974s csrs, bcrs and eeprom lo- cations and scsi and ccb registers. it is a single cycle, non-burst 8-bit,16-bit or 32-bit transfer which is initiated by the host cpu. the typical number of wait states added to a slave i/o read access on the part of the AM79C974 controller is 6 to 7 clock cycles. the AM79C974 controller will not produce slave i/o read commands while being a bus master. frame clk ad irdy trdy c/ be 12345678 addr data 0010 be 's 91011 par par par stop devsel 18681a-7 figure 3. slave i/o read amd p r e l i m i n a r y 30 AM79C974 slave i/o write the slave i/o write command is used by the host cpu to write to the AM79C974s csrs, bcrs and eeprom locations and scsi and ccb registers. it is a single cy- cle, non-burst 8-bit, 16Cbit, or 32-bit transfer which is initiated by the host cpu. the typical number of wait states added to a slave i/o write access on the part of the AM79C974 controller is 6 to 7 clock cycles. the AM79C974 controller will not produce slave i/o write commands while being a bus master. 18681a-8 frame clk ad irdy trdy c/ be 12345678 addr data 0011 be 's 91011 par par par stop devsel figure 4. slave i/o write p r e l i m i n a r y amd 31 AM79C974 bus acquisition the AM79C974 microcode (in the buffer management section) will determine when a dma transfer should be initiated. the first step in any AM79C974 bus master transfer is to acquire ownership of the bus. this task is handled by synchronous logic within the biu. bus own- ership is requested with the req signal and ownership is granted by the arbiter through the gnt signal. figure 5 shows the AM79C974 controller bus acquisi- tion. gnt is asserted at clock 3. the AM79C974 control- ler starts driving ad[31:00] and c/ be [3:0] prior to clock 4. frame is asserted at clock 5 indicating a valid ad- dress and command on ad[31:00] and c/ be [3:0]. adstep (bit 7) in the pci command register is set to one to indicated that the AM79C974 controller uses ad- dress stepping. address stepping is only used for the first address phase of a bus master period. 18681a-9 frame clk ad c/ be req gnt 12345 addr cmd figure 5. bus acquisition amd p r e l i m i n a r y 32 AM79C974 bus master dma transfers there are four primary types of dma transfers. the AM79C974 controller uses non-burst as well as burst cycles for read and write access to the main memory. basic non-burst read cycles all AM79C974 controller non-burst read accesses are of the pci command type memory read (type 6). note that during all non-burst read operations, the AM79C974 controller will always activate all byte enables, even though some byte lanes may not contain valid data as indicated by a buffer pointer value. in such instances, the AM79C974 controller will internally discard un- needed bytes. figure 6 shows a typical non-burst read access. the AM79C974 controller asserts irdy at clock 5 immedi- ately after the address phase and starts sampling devsel . the target extends the cycle by asserting devsel not until clock 6. additionally, the target inserts one wait state by asserting its ready ( trdy ) at clock 8. 18681a-10 frame clk ad irdy trdy c/ be devsel req gnt 12345678 0000 9 par par par data devsel is sampled by the AM79C974 controller. addr 0110 figure 6. non-burst read cycles with wait states p r e l i m i n a r y amd 33 AM79C974 figure 7 shows two non-burst read access within one ar- bitration cycle. the AM79C974 controller will drop frame between two consecutive non-burst read cy- cles. the AM79C974 controller will re-request the bus right again if it is preempted before starting the second access. the example below also shows a target that can respond to the AM79C974 controller read cycles without wait states. 18681a-11 frame clk ad irdy trdy c/ be devsel req gnt 12345678 addr 0000 0110 11 10 9 par par par data addr data 0000 0110 par par devsel is sampled by the AM79C974 controller. figure 7. non-burst read cycles without wait states amd p r e l i m i n a r y 34 AM79C974 basic non-burst write all AM79C974 controller non-burst write accesses are of the pci command type memory write (type 7). figure 8 shows two non-burst write access within one arbitration cycle. the AM79C974 controller will drop frame between two consecutive non-burst write cy- cles. the AM79C974 controller will re-request the bus immediately if it is preempted before starting the second access. the example below shows an extended cycle for the first access. the target asserts devsel 2 clock cycles after the address phase ( frame asserted) and adds one extra wait state by asserting trdy only on clock 7. the second write cycle in the example shows a zero wait state access. 18681a-12 frame clk ad irdy trdy c/ be devsel req gnt 12345678 addr be 's 0111 9 par par par data addr data be 's 0111 par par 10 11 devsel is sampled by the AM79C974 controller. figure 8. non-burst write cycles with and without wait states p r e l i m i n a r y amd 35 AM79C974 basic burst read cycles all AM79C974 controller burst read transfers are of the pci command type memory read line (type14). ad[1:0] will both be zero during the address phase in- dicating a linear burst order. all four byte enable signals will be zero during the data phase as the AM79C974 controller always reads a full 32-bit word when in burst mode. figure 9 shows a typical burst read access. the AM79C974 controller arbitrates for the bus, is granted access, and reads four 32-bit words (dword) from system memory and then releases the bus. all four data phases in this example take two clock cycles each, which is determined by the timing of trdy . 18681a-13 frame clk ad irdy trdy c/ be devsel req gnt 12345678 addr 0000 1110 11 10 9 12 par par data data data data par par par par devsel is sampled by the AM79C974 controller. 13 figure 9. burst read cycles amd p r e l i m i n a r y 36 AM79C974 basic burst write cycles all AM79C974 controller burst write transfers are of the pci command type memory write (type 7). ad[1:0] will both be zero during the address phase indicating a lin- ear burst order. all four byte enable signals will be zero during the data phase as the AM79C974 controller al- ways writes a full 32-bit word when in burst mode. figure 10 shows a typical burst write access. the AM79C974 controller arbitrates for the bus, is granted access, and writes four 32-bit words (dwords) from system memory and then releases the bus. in this ex- ample, the memory system extends the data phase of the first access by one wait state. the following three data phases take one clock cycle each, which is deter- mined by the timing of trdy . 18681a-14 frame clk ad irdy trdy c/ be devsel req gnt 12345678 addr data data data 0000 0111 10 9 par par par par par devsel is sampled by the AM79C974 controller. data par figure 10. burst write cycles p r e l i m i n a r y amd 37 AM79C974 transaction termination termination of a pci transaction may be initiated by either the master or the target. during termination, the master remains in control to bring all pci transactions to an orderly and systematic conclusion regardless of what caused the termination. all transactions are concluded when frame and irdy are both deasserted, indicating an idle cycle. target initiated termination when the AM79C974 controller is a bus master, the cy- cles it produces on the pci bus may be terminated by the target in one of three different ways: disconnect with data transfer, disconnect without data transfer, and tar- get abort. disconnect with data transfer figure 11 shows a disconnection in which one last data transfer occurs after the target asserted stop . stop is asserted on clock 4 to start the termination sequence. data is still transferred during this cycles, since both irdy and trdy are asserted. the AM79C974 control- ler terminates the current transfer with the deassertion of frame on clock 5 and then one clock cycle later with the deassertion of irdy . it finally releases the bus on clock 6. the AM79C974 controller will re-request the bus after 2 clock cycles, if it wants to transfer more data. the starting address of the new transfer will be the ad- dress of the next untransferred data. 18681a-15 frame clk ad irdy trdy c/ be devsel req gnt 1234567 8 data data 0000 0111 10 9 par par par par devsel is sampled by the AM79C974 controller. stop 0111 11 addr +8 i addr i figure 11. disconnect with data transfer amd p r e l i m i n a r y 38 AM79C974 disconnect without data transfer figure 12 shows a target disconnect sequence during which no data is transferred. stop is asserted on clock 4 without trdy being asserted at the same time. the AM79C974 controller terminates the current transfer with the deassertion of frame on clock 5 and one clock cycle later with the deassertion of irdy . it finally re- leases the bus on clock 6. the AM79C974 controller will re-request the bus after 2 clock cycles to retry the last transfer. the starting address of the new transfer will be the same address as of the last untransferred data. 18681a-16 frame clk ad irdy trdy c/ be devsel req gnt 1234567 8 data 0000 0111 10 9 par par par devsel is sampled by the AM79C974 controller. stop 0111 11 addr i addr i figure 12. disconnect without data transfer p r e l i m i n a r y amd 39 AM79C974 target abort figure 13 shows a target abort sequence. the target as- serts devsel for one clock. it then deasserts devsel and asserts stop on clock 4. a target can use the target abort sequence to indicate that it cannot service the data transfer and that it does not want the transaction to be retried. additionally, the AM79C974 controller cannot make any assumption about the success of the previous data transfers in the current transaction. the AM79C974 controller terminates the current transfer with the deassertion of frame on clock 5 and one clock cycle later with the deassertion of irdy . it finally re- leases the bus on clock 6. since data integrity is not guaranteed, the AM79C974 controller cannot recover from a target abort event. for ethernet, the AM79C974 controller will reset all csr and bcr locations to their h_reset values. any on- going network activity will be stopped immediately. the pci configuration registers will not be cleared. for scsi, when target aborts, inta will not be asserted, but the abort bit (bit 2 of the dma status register at offset 54h), is set. for either ethernet or scsi a target abort causes rtabort (bit 12) of the status register in the appropriate pci configuration space to be set. 18681a-17 frame clk ad irdy trdy c/ be devsel req gnt 123456 data 0000 0111 par par par devsel is sampled by the AM79C974 controller. stop addr figure 13. target abort amd p r e l i m i n a r y 40 AM79C974 master initiated termination there are three scenarios besides normal completion of a transaction where the AM79C974 controller will termi- nate the cycles it produces on the pci bus. these are preemption with and without frame assertion and master abort. preemption when frame is deasserted the AM79C974 controller can generate multiple ad- dress phases during a single bus ownership period when transferring data. frame is deasserted in be- tween two address phases. while frame is deass- erted, the central arbiter can remove gnt to the AM79C974 controller at any time to service another master. when gnt is removed, the AM79C974 control- ler will finish the current transfer and then release the bus. it will keep req asserted to regain bus ownership as soon as possible. 18681a-18 frame clk ad irdy trdy c/ be devsel req gnt 1234567 8 addr be 's 0111 par par par data devsel is sampled by the AM79C974 controller. figure 14. preemption when frame is deasserted p r e l i m i n a r y amd 41 AM79C974 preemption when frame is asserted the central arbiter can take gnt to the AM79C974 con- troller away if the current bus operation takes too long. this may happen, for example, when the AM79C974 controller tries to fill the whole ethernet transmit fifo and the target inserts extra wait states for every data phase. when gnt is taken away, the AM79C974 con- troller will finish the current transfer and then immedi- ately release the bus. the latency timer in pci configuration space of the AM79C974 controller is al- ways set to zero. the AM79C974 controller will keep req asserted to regain bus ownership as soon as pos- sible. 18681a-19 frame clk ad irdy trdy c/ be devsel req gnt 12345678 addr data data data 0000 0111 9 par par par par par devsel is sampled by the AM79C974 controller. figure 15. preemption when frame is asserted amd p r e l i m i n a r y 42 AM79C974 master abort the AM79C974 controller will terminate its cycle with a master abort sequence if devsel is not asserted within 4 clocks after frame is asserted. master abort is treated as a fatal error by the AM79C974 controller. for the ethernet, the AM79C974 controller will reset all csr and bcr locations to their h_reset values. any on-going network activity will be stopped immediately. the pci configuration registers will not be cleared. for scsi, when the master aborts, inta will not be as- serted, but the abort bit (bit 2 of the dma status regis- ter at offset 54h), is set. for either ethernet or scsi master abort causes rmabort (bit 13) of the status register in the appropriate pci configuration space to be set. 18681a-20 frame clk ad irdy trdy c/ be devsel req gnt 12345678 addr data 0000 0111 10 9 par par par devsel is sampled by the AM79C974 controller. figure 16. master abort p r e l i m i n a r y amd 43 AM79C974 ethernet controller buffer management unit (bmu) the buffer management unit is a micro-coded state ma- chine which implements the initialization procedure and manages the descriptors and buffers. the buffer man- agement unit operates at half the speed of the clk input. initialization AM79C974 initialization includes the reading of the in- itialization block in memory to obtain the operating pa- rameters. the initialization block is read when the init bit in csr0 is set. the init bit should be set before or concurrent with the strt bit to insure correct operation. two dwords are read during each period of bus ma- stership. when ssize32 = 1 (bcr20, bit 8), this results in a total of 4 arbitration cycles (3 arbitration cycles if ssize32 = 0). once the initialization block has been completely read in and internal registers have been up- dated, idon will be set in csr0, and an interrupt gener- ated (if iena is set). at this point, the bmu knows where the receive and transmit descriptor rings and hence, normal network operations will begin. the AM79C974 controller obtains the start address of the initialization block from the contents of csr1 (least significant 16 bits of address) and csr2 (most signifi- cant 16 bits of address). the host must write csr1 and csr2 before setting the init bit. the block contains the user defined conditions for AM79C974 operation, to- gether with the base addresses and length information of the transmit and receive descriptor rings. there is an alternative method to initialize the AM79C974 controller. instead of initialization via the in- itialization block in memory, data can be written directly into the appropriate registers. either method may be used at the discretion of the programmer. if the registers are written to directly, the init bit must not be set, or the initialization block will be read in, thus overwriting the previously written information. please refer to appendix c for details on this alternative method. if initialization is done by writing directly to registers, the polling interval register (csr47) must be initialized in addition to those registers that can be loaded automati- cally from the initialization block. re-initialization the transmitter and receiver sections of the AM79C974 controller can be turned on via the initialization block (mode register dtx, drx bits; csr15[1:0]). the states of the transmitter and receiver are monitored by the host through csr0 (rxon, txon bits). the AM79C974 controller should be reinitialized if the trans- mitter and/or the receiver were not turned on during the original initialization, and it was subsequently required to activate them or if either section was shut off due to the detection of certain error conditions (merr, uflo, tx buff error). reinitialization may be done via the initialization block or by setting the stop bit in csr0, followed by writing to csr15, and then setting the start bit in csr0. note that this form of restart will not perform the same in the AM79C974 controller as in the lance. in particular, upon restart, the AM79C974 controller reloads the transmit and receive descriptor pointers with their re- spective base addresses. this means that the software must clear the descriptor own bits and reset its descrip- tor ring pointers before the restart of the AM79C974 controller. the reload of descriptor base addresses is performed in the lance only after initialization, so a re- start of the lance without initialization leaves the lance pointing at the same descriptor locations as be- fore the restart. buffer management buffer management is accomplished through message descriptor entries organized as ring structures in mem- ory. there are two rings, a receive ring and a transmit ring. the size of a message descriptor entry is 4 dwords, or 16 bytes, when ssize32 = 1. the size of a message descriptor entry is 4 words, or 8 bytes, when ssize32 = 0. descriptor rings each descriptor ring must be organized in a contiguous area of memory. at initialization time (setting the init bit in csr0), the AM79C974 controller reads the user-de- fined base address for the transmit and receive descrip- tor rings, as well as the number of entries contained in the descriptor rings. descriptor ring base addresses must be on a 16-byte boundary when ssize32=1, and 8-byte boundary when ssize=0. a maximum of 128 (or 512, depending upon the value of ssize32) ring entries is allowed when the ring length is set through the tlen and rlen fields of the initialization block. however, the ring lengths can be set beyond this range (up to 65535) by writing the transmit and receive ring length registers (csr76, csr78) directly. each ring entry contains the following information: 1. the address of the actual message data buffer in user or host memory 2. the length of the message buffer 3. status information indicating the condition of the buffer to permit the queuing and de-queuing of message buff- ers, ownership of each buffer is allocated to either the AM79C974 controller or the host. the own bit within the descriptor status information, either tmd or rmd (see section on tmd or rmd), is used for this purpose. own = 1 signifies that the AM79C974 controller cur- amd p r e l i m i n a r y 44 AM79C974 rently has ownership of this ring descriptor and its associated buffer. only the owner is permitted to relin- quish ownership or to write to any field in the descriptor entry. a device that is not the current owner of a descrip- tor entry cannot assume ownership or change any field in the entry. a device may, however, read from a de- scriptor that it does not currently own. software should always read descriptor entries in sequential order. when software finds that the current descriptor is owned by the AM79C974 controller, then the software must not read ahead to the next descriptor. the software should wait at the unowned descriptor until ownership has been granted to the software (when lappen = 1 (csr3, bit 5), then this rule is modified. see the lap- pen description). strict adherence to these rules in- sures that deadly embrace conditions are avoided. descriptor ring access mechanism at initialization, the AM79C974 controller reads the base address of both the transmit and receive descriptor rings into csrs for use by the AM79C974 controller dur- ing subsequent operations. as the final step in the self-initialization process, the base address of each ring is loaded into each of the cur- rent descriptor address registers and the address of the next descriptor entry in the transmit and receive rings is computed and loaded into each of the next descriptor address registers. when ssize32 = 0, software data structures are 16 bits wide. the following diagram, figure 17, illustrates the relationship between the initialization base address, the initialization block, the receive and transmit de- scriptor ring base addresses, the receive and trans- mit descriptors and the receive and transmit data buffers, for the case of ssize32 = 0. initialization block 24-bit base address pointer to initialization block iadr[15:0] iadr[23:16] res csr1 csr2 tdra[15:0] mode padr[15:0] padr[31:16] padr[47:32] ladrf[15:0] ladrf[31:16] ladrf[47:32] ladrf[63:48] rdra[15:0] rlen res rdra[23:16] tlen res tdra[23:16] rcv buffers ? ? ? rmd0 rmd1 rmd2 rmd3 rcv descriptor ring n n n n ? ? ? 1st desc. start 2nd desc. start rmd0 xmt buffers ? ? ? rx descriptor rings tmd0 tmd1 tmd2 tmd3 rx descriptor rings xmt descriptor ring m m m m ? ? ? 1st desc. start 2nd desc. start tmd0 data buffer n data buffer 1 data buffer 2 data buffer m data buffer 1 data buffer 2 18681a-21 figure 17 . 16-bit data structures: initialization block and descriptor rings p r e l i m i n a r y amd 45 AM79C974 when ssize32 = 1, software data structures are 32 bits wide. the following diagram illustrates, figure 18, the relationship between the initialization base address, the initialization block, the receive and transmit descriptor ring base addresses, the receive and transmit descriptors and the receive and transmit data buffers, for the case of ssize32 = 1. initialization block 32-bit base address pointer to initialization block csr1 csr2 rcv buffers ? ? ? rmd0 rmd1 rmd2 rmd3 rcv descriptor ring n n n n ? ? ? 1st desc. start 2nd desc. start rmd0 xmt buffers ? ? ? rx descriptor rings tmd0 tmd1 tmd2 tmd3 rx descriptor rings xmt descriptor ring m m m m ? ? ? 1st desc. start 2nd desc. start tmd0 data buffer n data buffer 1 data buffer 2 data buffer m data buffer 2 data buffer 1 padr[31:0] iadr[31:16] iadr[15:0] tlen res rlen res mode padr[47:32] res ladrf[31:0] ladrf[63:32] rdra[31:0] tdra[31:0] 18681a-22 figure 18 . 32-bit data structures: initialization block and descriptor rings amd p r e l i m i n a r y 46 AM79C974 polling if there is no network channel activity and there is no pre- or post-receive or pre- or post-transmit activity be- ing performed by the AM79C974 controller, then the AM79C974 controller will periodically poll the current re- ceive and transmit descriptor entries in order to ascer- tain their ownership. if the dpoll bit in csr4 is set, then the transmit polling function is disabled. a typical polling operation consists of the following: the AM79C974 controller will use the current receive de- scriptor address stored internally to vector to the appro- priate receive descriptor table entry (rdte). it will then use the current transmit descriptor address (stored internally) to vector to the appropriate transmit descrip- tor table entry (tdte). the accesses will be made in the following order: rmd1, then rmd0 of the current rdte during one bus arbitration, and after that, tmd1, then tmd0 of the current tdte during a second bus ar- bitration. all information collected during polling activity will be stored internally in the appropriate csrs. (i.e. csr18, csr19, csr20, csr21, csr40, csr42, csr50, csr52). unowned descriptor status will be in- ternally ignored. a typical receive poll is the product of the following conditions: 1. AM79C974 controller does not possess ownership of the current rdte and the poll time has elapsed and rxon=1 (csr0, bit 5), or 2. AM79C974 controller does not possess ownership of the next rdte the poll time has elapsed and rxon=1. if rxon=0 the AM79C974 controller will never poll rdte locations. the ideal system should always have at least one rdte available for the possibility of an unpredictable receive event. (this condition is not a requirement. if this condi- tion is not met, it simply means that frames will be missed by the system because there was no buffer space available.) but the typical system usually has at least one or two rdtes available for the possibility of an unpredictable receive event. given that this condition is satisfied, the current and next rdte polls are rarely seen and hence, the typical poll operation simply con- sists of a check of the status of the current tdte. when there is only one rdte (because the rlen was set to zero), then there is no next rdte and ownership of next rdte cannot be checked. if there is at least one rdte, the rdte poll will rarely be seen and the typical poll operation simply consists of a check of the current tdte. a typical transmit poll is the product of the following conditions: 1. AM79C974 controller does not possess ownership of the current tdte and dpoll=0 (csr4, bit 2) and txon=1 (csr0, bit 4) and the poll time has elapsed, or 2. AM79C974 controller does not possess ownership of the current tdte and dpoll=0 and txon=1 and a frame has just been received, or 3. AM79C974 controller does not possess ownership of the current tdte and dpoll=0 and txon=1 and a frame has just been transmitted. setting the tdmd bit of csr0 will cause the microcode controller to exit the poll counting code and immediately perform a polling operation. if rdte ownership has not been previously established, then an rdte poll will be performed ahead of the tdte poll. if the microcode is not executing the poll counting code when the tdmd bit is set, then the demanded poll of the tdte will be de- layed until the microcode returns to the poll counting code. the user may change the poll time value from the de- fault of 65,536 clock periods by modifying the value in the polling interval register (csr47). note that if a nonC default value is desired, then a strict sequence of setting the init bit in csr0, waiting for the idon bit in csr0, then writing to csr47, and then setting strt in csr0 must be observed, otherwise the default value will not be overwritten. see the csr47 section for details. transmit descriptor table entry (tdte) if, after a tdte access, the AM79C974 controller finds that the own bit of that tdte is not set, then the AM79C974 controller resumes the poll time count and reexamines the same tdte at the next expiration of the poll time count. if the own bit of the tdte is set, but start of frame (stp) bit is not set, the AM79C974 controller will imme- diately request the bus in order to reset the own bit of this descriptor. (this condition would normally be found following a lcol or retry error that occurred in the middle of a transmit frame chain of buffers.) after reset- ting the own bit of this descriptor, the AM79C974 con- troller will again immediately request the bus in order to access the next tdte location in the ring. if the own bit is set and the buffer length is 0, the own bit will be reset. in the lance the buffer length of 0 is interpreted as a 4096-byte buffer. it is acceptable to have a 0 length buffer on transmit with stp = 1 or stp=1 and enp = 1. it is not acceptable to have 0 length buffer with stp = 0 and enp =1. if the own bit is set and the start of frame (stp) bit is set, then microcode control proceeds to a routine that will enable transmit data transfers to the fifo. the p r e l i m i n a r y amd 47 AM79C974 AM79C974 controller will look ahead to the next trans- mit descriptor after it has performed at least one trans- mit data transfer from the first buffer. (more than one transmit data transfer may possibly take place, depend- ing upon the state of the transmitter.) the contents of tmd0 and tmd1 will be stored in next xmt buffer ad- dress (csr64 and csr65), next xmt byte count (csr66) and next xmt status (csr67) regardless of the state of the own bit. this transmit descriptor lookahead operation is performed only once. if the AM79C974 controller does not own the next tdte (i.e. the second tdte for this frame), then it will com- plete transmission of the current buffer and then update the status of the current (first) tdte with the buff and uflo bits being set. this will cause the transmitter to be disabled (csr0, txon=0). the AM79C974 controller will have to be re-initialized to restore the transmit func- tion. the situation that matches this description implies that the system has not been able to stay ahead of the AM79C974 controller in the transmit descriptor ring and therefore, the condition is treated as a fatal error. (to avoid this situation, the system should always set the transmit chain descriptor own bits in reverse order.) if the AM79C974 controller does own the second tdte in a chain, it will gradually empty the contents of the first buffer (as the bytes are needed by the transmit opera- tion), perform a single-cycle dma transfer to update the status of the first descriptor (reset the own bit in tmd1), and then it may perform one data dma access on the second buffer in the chain before executing an- other lookahead operation. (i.e. a lookahead to the third descriptor.) the AM79C974 controller can queue up to two frames in the transmit fifo. call them frame x and frame y, where y is after x. assume that frame x is currently being transmitted. because the AM79C974 controller can perform lookahead data transfer past the enp of frame x, it is possible for the AM79C974 controller to completely transfer the data from a buffer belonging to frame y into the fifo even though frame x has not yet been completely transmitted. at the end of this y buffer data transfer, the AM79C974 controller will write intermediate status (change the own bit to a zero) for the y frame buffer, if frame y uses data chaining. the last tdte for the x frame (containing enp) has not yet been written, since the x frame has not yet been completely transmitted. note that the AM79C974 controller has, in this instance, returned ownership of a tdte to the host out of a normal sequence. for this reason, it becomes imperative that the host sys- tem should never read the transmit dte ownership bits out of order. software should always process buffers in sequence, waiting for the ownership before proceeding. there should be no problems for software which proc- esses buffers in sequence, waiting for ownership before proceeding. if an error occurs in the transmission before all of the bytes of the current buffer have been transferred, then tmd2 and tmd1 of the current buffer will be written; in such a case, data transfers from the next buffer will not commence. instead, following the tmd2/tmd1 update, the AM79C974 controller will go to the next transmit frame, if any, skipping over the rest of the frame which experienced an error, including chained buffers. this is done by returning to the polling microcode where AM79C974 controller will immediately access the next descriptor and find the condition own=1 and stp=0 as described earlier. as described for that case, the AM79C974 controller will reset the own bit for this de- scriptor and continue in like manner until a descriptor with own=0 (no more transmit frames in the ring) or own=1 and stp=1 (the first buffer of a new frame) is reached. at the end of any transmit operation, whether successful or with errors, immediately following the completion of the descriptor updates, the AM79C974 controller will al- ways perform another poll operation. as described ear- lier, this poll operation will begin with a check of the current rdte, unless the AM79C974 controller already owns that descriptor. then the AM79C974 controller will proceed to polling the next tdte. if the transmit descrip- tor own bit has a zero value, then the AM79C974 controller will resume poll time count incrementing. if the transmit descriptor own bit has a value of one, then the AM79C974 controller will begin filling the fifo with transmit data and initiate a transmission. this endCofC operation poll coupled with the tdte lookahead opera- tion allows the AM79C974 controller to avoid inserting poll time counts between successive transmit frames. whenever the AM79C974 controller completes a trans- mit frame (either with or without error) and writes the status information to the current descriptor, then the tint bit of csr0 is set to indicate the completion of a transmission. this causes an interrupt signal if the iena bit of csr0 has been set and the tintm bit of csr3 is reset. amd p r e l i m i n a r y 48 AM79C974 receive descriptor table entry (rdte) if the AM79C974 controller does not own both the cur- rent and the next receive descriptor table entry then the AM79C974 controller will continue to poll according to the polling sequence described above. if the receive descriptor ring length is 1, then there is no next descrip- tor to be polled. if a poll operation has revealed that the current and the next rdte belong to the AM79C974 controller then ad- ditional poll accesses are not necessary. future poll op- erations will not include rdte accesses as long as the AM79C974 controller retains ownership of the current and the next rdte. when receive activity is present on the channel, the AM79C974 controller waits for the complete address of the message to arrive. it then decides whether to accept or reject the frame based on all active addressing schemes. if the frame is accepted the AM79C974 con- troller checks the current receive buffer status register crst (csr41) to determine the ownership of the cur- rent buffer. if ownership is lacking, then the AM79C974 controller will immediately perform a (last ditch) poll of the current rdte. if ownership is still denied, then the AM79C974 controller has no buffer in which to store the incoming message. the missed frame count register (csr112) will be incremented and the miss bit will be set in csr0 and an interrupt will be generated if iena=1 (csr0) and missm=0 (csr3). another poll of the current rdte will not occur until the frame has finished. if the AM79C974 controller sees that the last poll (either a normal poll, or the last-ditch effort described in the above paragraph) of the current rdte shows valid own- ership, then it proceeds to a poll of the next rdte. fol- lowing this poll, and regardless of the outcome of this poll, transfers of receive data from the fifo may begin. regardless of ownership of the second receive descrip- tor, the AM79C974 controller will continue to perform re- ceive data dma transfers to the first buffer. if the frame length exceeds the length of the first buffer, and the AM79C974 controller does not own the second buffer, ownership of the current descriptor will be passed back to the system by writing a zero to the own bit of rmd1 and status will be written indicating buffer (buff=1) and possibly overflow (oflo=1) errors. if the frame length exceeds the length of the first (cur- rent) buffer, and the AM79C974 controller does own the second (next) buffer, ownership will be passed back to the system by writing a zero to the own bit of rmd1 when the first buffer is full. receive data transfers to the second buffer may occur before the AM79C974 control- ler proceeds to look ahead to the ownership of the third buffer. such action will depend upon the state of the fifo when the status has been updated on the first de- scriptor. in any case, lookahead will be performed to the third buffer and the information gathered will be stored in the chip, regardless of the state of the ownership bit. as in the transmit flow, lookahead operations are per- formed only once. this activity continues until the AM79C974 controller recognizes the completion of the frame (the last byte of this receive message has been removed from the fifo). the AM79C974 controller will subsequently up- date the current rdte status with the end of frame (enp) indication set, write the message byte count (mcnt) of the complete frame into rmd2 and overwrite the current entries in the csrs with the next entries. media access control the media access control engine incorporates the es- sential protocol requirements for operation of a compli- ant ethernet/802.3 node, and provides the interface between the fifo sub-system and the manchester en- coder/decoder (mendec). the mac engine is fully compliant to section 4 of iso/ iec 8802-3 (ansi/ieee standard 1990 second edition) and ansi/ieee 802.3 (1985). the mac engine provides programmable enhanced features designed to minimize host supervision, bus utilization, and pre- or post- message processing. these include the ability to disable retries after a colli- sion, dynamic fcs generation on a frame-by-frame ba- sis, and automatic pad field insertion and deletion to enforce minimum frame size attributes, automatic retransmission without reloading the fifo, automatic deletion of collision fragments, and reduces bus bandwidth use. the two primary attributes of the mac engine are: transmit and receive message data encapsulation. framing (frame boundary delimitation, frame synchronization). addressing (source and destination address handling). error detection (physical medium transmission errors). media access management. medium allocation (collision avoidance). contention resolution (collision handling). transmit and receive message data encapsulation the mac engine provides minimum frame size enforce- ment for transmit and receive frames. when apad_xmt = 1 (csr, bit 11), transmit messages will be padded with sufficient bytes (containing 00h) to en- sure that the receiving station will observe an informa- tion field (destination address, source address, length/type, data and fcs) of 64-bytes. when p r e l i m i n a r y amd 49 AM79C974 astrp_rcv = 1 (csr4, bit 10), the receiver will automatically strip pad bytes from the received mes- sage by observing the value in the length field, and strip- ping excess bytes if this value is below the minimum data size (46 bytes). both features can be independ- ently over-ridden to allow illegally short (less than 64 bytes of frame data) messages to be transmitted and/or received. the use of this feature reduces bus utilization because the pad bytes are not transferred into or out of main memory. framing (frame boundary delimitation, frame synchronization) the mac engine will autonomously handle the con- struction of the transmit frame. once the transmit fifo has been filled to the predetermined threshold (set by xmtsp in csr80), and providing access to the channel is currently permitted, the mac engine will commence the 7 byte preamble sequence (10101010b, where first bit transmitted is a 1). the mac engine will subse- quently append the start frame delimiter (sfd) byte (10101011b) followed by the serialized data from the transmit fifo. once the data has been completed, the mac engine will append the fcs (most significant bit first) which was computed on the entire data portion of the frame. the data portion of the frame consists of des- tination address, source address, length/type, and frame data. the user is responsible for the correct ordering and con- tent in each of the fields in the frame. the receive section of the mac engine will detect an in- coming preamble sequence and lock to the encoded clock. the internal mendec will decode the serial bit stream and present this to the mac engine. the mac will discard the first 8-bits of information before search- ing for the sfd sequence. once the sfd is detected, all subsequent bits are treated as part of the frame. the mac engine will inspect the length field to ensure mini- mum frame size, strip unnecessary pad characters (if enabled), and pass the remaining bytes through the re- ceive fifo to the host. if pad stripping is performed, the mac engine will also strip the received fcs bytes, al- though the normal fcs computation and checking will occur. note that apart from pad stripping, the frame will be passed unmodified to the host. if the length field has a value of 46 or greater, the mac engine will not attempt to validate the length against the number of bytes con- tained in the message. if the frame terminates or suffers a collision before 64-bytes of information (after sfd) have been received, the mac engine will automatically delete the frame from the receive fifo, without host intervention. the AM79C974 controller has the ability to accept runt pack- ets for diagnostics purposes and proprietary networks. addressing (source and destination address handling) the first 6-bytes of information after sfd will be inter- preted as the destination address field. the mac engine provides facilities for physical, logical (multicast) and broadcast address reception. error detection (physical medium transmission errors) the mac engine provides several facilities which report and recover from errors on the medium. in addition, the network is protected from gross errors due to inability of the host to keep pace with the mac engine activity. on completion of transmission, the following transmit status is available in the appropriate tmd and csr areas: the exact number of transmission retry attempts (one, more, rtry or trc). whether the mac engine had to defer (def) due to channel activity. excessive deferral (exdef), indicating that the transmitter has experienced excessive deferral on this transmit frame, where excessive deferral is defined in iso 8802-3 (ieee/ansi 802.3). loss of carrier (lcar), indicating that there was an interruption in the ability of the mac engine to monitor its own transmission. repeated lcar errors indicate a potentially faulty trans- ceiver or network connection. late collision (lcol) indicates that the transmis- sion suffered a collision after the slot time. this is indicative of a badly configured network. late colli- sions should not occur in a normal operating net- work. collision error (cerr) indicates that the trans- ceiver did not respond with an sqe test message within the predetermined time after a transmission completed. this may be due to a failed transceiver, disconnected or faulty transceiver drop cable, or the fact the transceiver does not support this fea- ture (or it is disabled). in addition to the reporting of network errors, the mac engine will also attempt to prevent the creation of any network error due to the inability of the host to service the mac engine. during transmission, if the host fails to keep the transmit fifo filled sufficiently, causing an un- derflow, the mac engine will guarantee the message is either sent as a runt packet (which will be deleted by the receiving station) or has an invalid fcs (which will also cause the receiver to reject the message). amd p r e l i m i n a r y 50 AM79C974 the status of each receive message is available in the appropriate rmd and csr areas. fcs and framing er- rors (fram) are reported, although the received frame is still passed to the host. the fram error will only be reported if an fcs error is detected and there are a non integral number of bytes in the message. the mac en- gine will ignore up to 7 additional bits at the end of a mes- sage (dribbling bits), which can occur under normal network operating conditions. the reception of 8 addi- tional bits will cause the mac engine to de-serialize the entire byte, and will result in the received message and fcs being modified. the AM79C974 controller can handle up to 7 dribbling bits when a received frame terminates. during the re- ception, the fcs is generated on every serial bit (includ- ing the dribbling bits) coming from the cable, although the internally saved fcs value is only updated on the eighth bit (on each byte boundary). the framing error is reported to the user as follows: if the number of dribbling bits are 1 to 7 and there is no crc (fcs) error, then there is no framing error (fram = 0). if the number of dribbling bits are 1 to 7 and there is a crc (fcs) error, then there is also a framing error (fram = 1). if the number of dribbling bits = 0, then there is no framing error. there may or may not be a crc (fcs) error. counters are provided to report the receive collision count and runt packet count for network statistics and utilization calculations. note that if the mac engine detects a received frame which has a 00b pattern in the preamble (after the first 8 bits which are ignored), the entire frame will be ignored. the mac engine will wait for the network to go inactive before attempting to receive additional frames. media access management the basic requirement for all stations on the network is to provide fairness of channel allocation. the 802.3/ethernet protocols define a media access mecha- nism which permits all stations to access the channel with equality. any node can attempt to contend for the channel by waiting for a predetermined time (inter pack- et gap internal) after the last activity, before transmitting on the media. the channel is a multidrop communica- tions media (with various topological configurations per- mitted) which allows a single station to transmit and all other stations to receive. if two nodes simultaneously contend for the channel, their signals will interact caus- ing loss of data, defined as a collision. it is the responsi- bility of the mac to attempt to avoid and recover from a collision, to guarantee data integrity for the end-to-end transmission to the receiving station. medium allocation the ieee/ansi 802.3 standard (iso/iec 8802-3 1990) requires that the csma/cd mac monitor the medium for traffic by watching for carrier activity. when carrier is detected, the media is considered busy, and the mac should defer to the existing message. the iso 8802-3 (ieee/ansi 802.3) standard also al- lows optional two part deferral after a receive message. see ansi/ieee std 802.3 C1990 edition, 4.2.3.2.1: note : it is possible for the pls carrier sense indication to fail to be asserted during a collision on the media. if the deference process simply times the interframe gap based on this indication it is possible for a short inter- frame gap to be generated, leading to a potential recep- tion failure of a subsequent frame. to enhance system robustness the following optional measures, as speci- fied in 4.2.8, are recommended when inter- framespacingpart1 is other than zero: 1. upon completing a transmission, start timing the interpacket gap, as soon as transmitting and car- rier sense are both false. 2. when timing an interframe gap following recep- tion, reset the interframe gap timing if carrier sense becomes true during the first 2/3 of the in- terframe gap timing interval. during the final 1/3 of the interval the timer shall not be reset to ensure fair access to the medium. an initial period shorter than 2/3 of the interval is permissible including zero. the mac engine implements the optional receive two part deferral algorithm, with a first part inter-frame-spac- ing time of 6.0 m s. the second part of the inter-frame- spacing interval is therefore 3.6 m s. the AM79C974 controller will perform the two part de- ferral algorithm as specified in section 4.2.8 (process deference). the inter packet gap (ipg) timer will start timing the 9.6 m s interframespacing after the receive carrier is de-asserted. during the first part deferral (in- terframespacingpart1 C ifs1) the AM79C974 control- ler will defer any pending transmit frame and respond to the receive message. the ipg counter will be reset to zero continuously until the carrier de-asserts, at which point the ipg counter will resume the 9.6 m s count once again. once the ifs1 period of 6.0 m s has elapsed, the AM79C974 controller will begin timing the second part deferral (interframe spacing part 2 C ifs2) of 3.6 m s. once ifs1 has completed, and ifs2 has commenced, the AM79C974 controller will not defer to a receive frame if a transmit frame is pending. this means that the AM79C974 controller will not attempt to receive the re- ceive frame, since it will start to transmit, and generate a collision at 9.6 m s. the AM79C974 controller will guar- antee to complete the preamble (64-bit) and jam (32-bit) p r e l i m i n a r y amd 51 AM79C974 sequence before ceasing transmission and invoking the random backoff algorithm. this transmit two part deferral algorithm is implemented as an option which can be disabled using the dxmt2pd bit in csr3. two part deferral after transmission is use- ful for ensuring that severe ipg shrinkage cannot occur in specific circumstances, causing a transmit message to follow a receive message so closely as to make them indistinguishable. during the time period immediately after a transmission has been completed, the external transceiver (in the case of a standard aui connected device), should gen- erate the sqe test message (a nominal 10 mhz burst of 5 C15 bit times duration) on the ci pair (within 0.6 m s C 1.6 m s after the transmission ceases). during the time period in which the sqe test message is expected the AM79C974 controller will not respond to receive carrier sense. see ansi/ieee std 802.31990 edition, 7.2.4.6 (1): at the conclusion of the output function, the dte opens a time window during which it expects to see the sig- nal_quality_error signal asserted on the control in cir- cuit. the time window begins when the carrier_status becomes carrier_off. if exe- cution of the output function does not cause car- rier_on to occur, no sqe test occurs in the dte. the duration of the window shall be at least 4.0 m s but no more than 8.0 m s. during the time window the carrier sense function is inhibited. the AM79C974 controller implements a carrier sense blinding period within 0 m s C 4.0 m s from de-assertion of carrier sense after transmission. this effectively means that when transmit two part deferral is enabled (dxmt2pd is cleared) the ifs1 time is from 4 m s to 6 m s after a transmission. however, since ipg shrinkage be- low 4 m s will rarely be encountered on a correctly config- ured networks, and since the fragment size will be larger than the 4 m s blinding window, then the ipg counter will be reset by a worst case ipg shrinkage/fragment sce- nario and the AM79C974 controller will defer its trans- mission. in addition, the AM79C974 controller will not restart the blinding period if carrier is detected within the 4.0 m s C 6.0 m s ifs1 period, but will commence tim- ing of the entire ifs1 period. contention resolution (collision handling) collision detection is performed and reported to the mac engine by the integrated manchester encoder/de- coder (mendec). if a collision is detected before the complete preamble/sfd sequence has been transmit- ted, the mac engine will complete the preamble/sfd before appending the jam sequence. if a collision is de- tected after the preamble/sfd has been completed, but prior to 512 bits being transmitted, the mac engine will abort the transmission, and append the jam sequence immediately. the jam sequence is a 32-bit all zeros pat- tern. the mac engine will attempt to transmit a frame a total of 16 times (initial attempt plus 15 retries) due to normal collisions (those within the slot time). detection of colli- sion will cause the transmission to be re-scheduled, de- pendent on the backoff time that the mac engine computes. if a single retry was required, the one bit will be set in the transmit frame status. if more than one retry was required, the more bit will be set. if all 16 at- tempts experienced collisions, the rtry bit will be set (one and more will be clear), and the transmit mes- sage will be flushed from the fifo. if retries have been disabled by setting the drty bit in csr15, the mac en- gine will abandon transmission of the frame on detection of the first collision. in this case, only the rtry bit will be set and the transmit message will be flushed from the fifo. if a collision is detected after 512 bit times have been transmitted, the collision is termed a late collision. the mac engine will abort the transmission, append the jam sequence and set the lcol bit. no retry attempt will be scheduled on detection of a late collision, and the trans- mit message will be flushed from the fifo. the iso 8802-3 (ieee/ansi 802.3) standard requires use of a truncated binary exponential backoff algo- rithm which provides a controlled pseudo random mechanism to enforce the collision backoff interval, be- fore re-transmission is attempted. see ansi/ieee std 802.31990 edition, 4.2.3.2.5: at the end of enforcing a collision (jamming), the csma/cd sublayer delays before attempting to re- transmit the frame. the delay is an integer multiple of slot time. the number of slot times to delay before the nth re-transmission attempt is chosen as a uniformly distributed random integer r in the range: 0 r <2k where k = min (n,10). the AM79C974 controller provides an alternative algo- rithm, which suspends the counting of the slot time/ipg during the time that receive carrier sense is detected. this aids in networks where large numbers of nodes are present, and numerous nodes can be in collision. it ef- fectively accelerates the increase in the backoff time in busy networks, and allows nodes not involved in the col- lision to access the channel whilst the colliding nodes await a reduction in channel activity. once channel ac- tivity is reduced, the nodes resolving the collision time out their slot time counters as normal. amd p r e l i m i n a r y 52 AM79C974 manchester encoder/decoder (mendec) the integrated manchester encoder/decoder provides the pls (physical layer signaling) functions required for a fully compliant iso 8802-3 (ieee/ansi 802.3) sta- tion. the mendec provides the encoding function for data to be transmitted on the network using the high ac- curacy on-board oscillator, driven by either the crystal oscillator or an external cmos level compatible clock. the mendec also provides the decoding function from data received from the network. the mendec contains a power on reset (por) circuit, which ensures that all analog portions of the AM79C974 controller are forced into their correct state during power up, and prevents er- roneous data transmission and/or reception during this time. external crystal characteristics when using a crystal to drive the oscillator, the following crystal specification may be used to ensure less than 0.5 ns jitter at do . see table 1 below. table 1. crystal specification parameter min nom max unit 1. parallel resonant frequency 20 mhz 2. resonant frequency error C50 +50 ppm 3. change in resonant frequency with respect to temperature (0 C 70 c) C40 +40 ppm 4. crystal load capacitance 20 50 pf 5. motional crystal capacitance (c1) 0.022 pf 6. series resistance 35 w 7 . shunt capacitance 7pf 8. drive level tbd mw external clock drive characteristics when driving the oscillator from a cmos level external clock source, xtal2 must be left floating (uncon- nected). an external clock having the following charac- teristics must be used to ensure less than 0.5 ns jitter at do . see table 2. table 2. clock drive characteristics clock frequency: 20 mhz 0.01% rise/fall time (t r /t f ): <= 6 ns from 0.5 v to v dd C0.5 v xtal1 high/low time (thigh/tlow): 20 ns min xtal1 falling edge to falling edge jitter: < 0.2 ns at 2.5 v input (v dd/2 ) clock frequency: 20 mhz 0.01% rise/fall time (t r /t f ): <= 6 ns from 0.5 v to v dd C0.5 v xtal1 high/low time (thigh/tlow): 20 ns min xtal1 falling edge to falling edge jitter: < 0.2 ns at 2.5 v input (v dd/2 ) mendec transmit path the transmit section encodes separate clock and nrz data input signals into a standard manchester encoded serial bit stream. the transmit outputs (do ) are de- signed to operate into terminated transmission lines. when operating into a 78 w terminated transmission line, the transmit signaling meets the required output levels and skew for cheapernet, ethernet and ieee-802.3. transmitter timing and operation a 20 mhz fundamental mode crystal oscillator provides the basic timing reference for the mendec portion of the AM79C974 controller. the crystal is divided by two, to create the internal transmit clock reference. both clocks are fed into the mendecs manchester encoder to generate the transitions in the encoded data stream. the internal transmit clock is used by the mendec to internally synchronize the internal transmit data (itxdat) from the controller and internal transmit en- able (itxen). the internal transmit clock is also used as a stable bit rate clock by the receive section of the men- dec and controller. p r e l i m i n a r y amd 53 AM79C974 the oscillator requires an external 0.01% timing refer- ence. the accuracy requirements, if an external crystal is used are tighter because allowance for the on-board parasitics must be made to deliver a final accuracy of 0.01%. transmission is enabled by the controller. as long as the itxen request remains active, the serial output of the controller will be manchester encoded and appear at do . when the internal request is dropped by the con- troller, the differential transmit outputs go to one of two idle states, dependent on tsel in the mode register (csr15, bit 9): tsel low: the idle state of do yields zero differential to operate transformer- coupled loads. tsel high: in this idle state, do+ is positive with respect to doC (logical high). receiver path the principal functions of the receiver are to signal the AM79C974 controller that there is information on the re- ceive pair, and separate the incoming manchester en- coded data stream into clock and nrz data. the receiver section (see receiver block diagram) consists of two parallel paths. the receive data path is a zero threshold, wide bandwidth line receiver. the car- rier path is an offset threshold bandpass detecting line receiver. both receivers share common bias networks to allow operation over a wide input common mode range. 18681a-23 noise reject filter data receiver carrier detect circuit manchester decoder irxdat* isrdclk* irxcrs* di *internal signal figure 19 . receiver block diagram input signal conditioning transient noise pulses at the input data stream are re- jected by the noise rejection filter. pulse width rejec- tion is proportional to transmit data rate, (which is fixed at 10 mhz for ethernet but could be different for proprie- tary networks). dc inputs more negative than minus 100 mv are also surpressed. the carrier detection circuitry detects the presence of an incoming data frame by discerning and rejecting noise from expected manchester data, and controls the stop and start of the phase-lock loop during clock acqui- sition. clock acquisition requires a valid manchester bit pattern of 1010b to lock onto the incoming message. when input amplitude and pulse width conditions are met at di , the internal enable signal from the mendec to controller (irxcrs) is asserted and a clock acquisi- tion cycle is initiated. clock acquisition when there is no activity at di (receiver is idle), the re- ceive oscillator is phase locked to internal transmit clock. the first negative clock transition (bit cell center of first valid manchester 0) after irxcrs is asserted in- terrupts the receive oscillator. the oscillator is then re- started at the second manchester 0 (bit time 4) and is phase locked to it. as a result, the mendec acquires the clock from the incoming manchester bit pattern in 4 bit times with a 1010b manchester bit pattern. the internal serial receive data clock, isrdclk and the internal received data, irxdat, are enabled 1/4 bit time after clock acquisition in bit cell 5. irxdat is at a high state when the receiver is idle (no isrdclk). irxdat however, is undefined when clock is acquired and may remain high or change to low state whenever isrdclk is enabled. at 1/4 bit time through bit cell 5, amd p r e l i m i n a r y 54 AM79C974 the controller portion of the AM79C974 controller sees the first isrdclk transition. this also strobes in the in- coming fifth bit to the mendec as manchester 1. irxdat may make a transition after the isrdclk rising edge in bit cell 5, but its state is still undefined. the manchester 1 at bit 5 is clocked to irxdat output at 1/4 bit time in bit cell 6. pll tracking after clock acquisition, the phase-locked clock is com- pared to the incoming transition at the bit cell center (bcc) and the resulting phase error is applied to a cor- rection circuit. this circuit ensures that the phase- locked clock remains locked on the received signal. individual bit cell phase corrections of the voltage con- trolled oscillator (vco) are limited to 10% of the phase difference between bcc and phase-locked clock. hence, input data jitter is reduced in isrdclk by 10 to 1. carrier tracking and end of message the carrier detection circuit monitors the di inputs after irxcrs is asserted for an end of message. irxcrs de-asserts 1 to 2 bit times after the last positive transi- tion on the incoming message. this initiates the end of reception cycle. the time delay from the last rising edge of the message to irxcrs de-assert allows the last bit to be strobed by isrdclk and transferred to the con- troller section, but prevents any extra bit(s) at the end of message. data decoding the data receiver is a comparator with clocked output to minimize noise sensitivity to the di inputs. input error is less than 35 mv to minimize sensitivity to input rise and fall time. isrdclk strobes the data receiver output at 1/4 bit time to determine the value of the manchester bit, and clocks the data out on irxdat on the following isrdclk. the data receiver also generates the signal used for phase detector comparison to the internal mendec voltage controlled oscillator (vco). differential input terminations the differential input for the manchester data (di ) should be externally terminated by two 40.2 w 1% re- sistors and one optional common-mode bypass capaci- tor, as shown in the differential input termination diagram below. the differential input impedance, z idf , and the common-mode input impedance, z icm , are specified so that the ethernet specification for cable ter- mination impedance is met using standard 1% resistor terminators. if sip devices are used, 39 w is the nearest usable value. the ci differential inputs are terminated in exactly the same way as the di pair. 18681a-24 AM79C974 di+ di- 40.2 w 40.2 w 0.01 m f to 0.1 m f aui isolation transformer figure 20. differential input termination p r e l i m i n a r y amd 55 AM79C974 collision detection a mau detects the collision condition on the network and generates a differential signal at the ci inputs. this collision signal passes through an input stage which de- tects signal levels and pulse duration. when the signal is detected by the mendec it sets the internal collision signal high. the condition continues for approximately 1.5 bit times after the last low-to-high transition on ci . jitter tolerance definition the mendec utilizes a clock capture circuit to align its internal data strobe with an incoming bit stream. the clock acquisition circuitry requires four valid bits with the values 1010b. clock is phase-locked to the negative transition at the bit cell center of the second 0 in the pattern. since data is strobed at 1/4 bit time, manchester transi- tions which shift from their nominal placement through 1/4 bit time will result in improperly decoded data. with this as the criteria for an error, a definition of jitter han- dling is: the peak deviation approaching or crossing 1/4 bit cell position from nominal input transition, for which the mendec section will properly decode data. attachment unit interface (aui) the aui is the pls (physical layer signaling) to pma (physical medium attachment) interface which effec- tively connects the dte to a mau. the differential inter- face provided by the AM79C974 controller is fully compliant to section 7 of iso 8802-3 (ansi/ieee 802.3). after the AM79C974 controller initiates a transmission it will expect to see data looped-back on the di pair (when the aui port is selected). this will internally gen- erate a carrier sense, indicating that the integrity of the data path to and from the mau is intact, and that the mau is operating correctly. this carrier sense signal must be asserted at least 6 bit times before the last transmitted bit on do (when using the aui port). if car- rier sense does not become active in response to the data transmission, or becomes inactive before the end of transmission, the loss of carrier (lcar) error bit will be set in the transmit descriptor ring (tmd2, bit 27) af- ter the frame has been transmitted. twisted-pair transceiver (t-mau) the t-mau implements the medium attachment unit (mau) functions for the twisted-pair medium, as speci- fied by the supplement to iso 8802-3 (ieee/ansi 802.3) standard (type 10base-t). the t-mau pro- vides twisted pair driver and receiver circuits, including on-board transmit digital predistortion and receiver squelch and a number of additional features including link status indication, automatic twisted-pair receive polarity detection/correction and indication, receive carrier sense, transmit active and collision present indication. twisted-pair transmit function the differential driver circuitry in the txd and txp pins provides the necessary electrical driving capability and the pre-distortion control for transmitting signals over maximum length twisted-pair cable, as specified by the 10base-t supplement to the iso 8802-3 (ieee/ ansi 802.3) standard. the transmit function for data output meets the propagation delays and jitter specified by the standard. twisted-pair receive function the receiver complies with the receiver specifications of the iso 8802-3 (ieee/ansi 802.3) 10base-t stan- dard, including noise immunity and received signal re- jection criteria (smart squelch). signals meeting these criteria appearing at the rxd differential input pair are routed to the mendec. the receiver function meets the propagation delays and jitter requirements specified by the standard. the receiver squelch level drops to half its threshold value after unsquelch to allow reception of minimum amplitude signals and to offset carrier fade in the event of worst case signal attenuation and crosstalk noise conditions. note that the 10base-t standard defines the receive input amplitude at the external media dependent inter- face (mdi). filter and transformer loss are not specified. the t-mau receiver squelch levels are defined to ac- count for a 1 db insertion loss at 10 mhz, which is typical for the type of receive filters/transformers employed. normal 10base-t compatible receive thresholds are employed when the lrt bit (csr15[9]) is low. when the lrt bit is set (high), the low receive threshold option is invoked, and the sensitivity of the t-mau re- ceiver is increased. this allows longer line lengths to be employed, exceeding the 100 m target distance of nor- mal 10base-t (assuming typical 24 awg cable). the increased receiver sensitivity compensates for the in- creased signal attenuation caused by the additional cable distance. however, making the receiver more sensitive means that it is also more susceptible to extraneous noise, pri- marily caused by coupling from co-resident services (crosstalk). for this reason, it is recommended that when using the low receive threshold option that the service should be installed on 4-pair cable only. multi- pair cables within the same outer sheath have lower crosstalk attenuation, and may allow noise emitted from adjacent pairs to couple into the receive pair, and be of sufficient amplitude to falsely unsquelch the t-mau. amd p r e l i m i n a r y 56 AM79C974 link test function the link test function is implemented as specified by 10base-t standard. during periods of transmit pair inactivity, link beat pulses will be periodically sent over the twisted pair medium to constantly monitor me- dium integrity. when the link test function is enabled (dlnktst bit in csr15 is cleared), the absence of link beat pulses and receive data on the rxd pair will cause the tmau to go into a link fail state. in the link fail state, data transmis- sion, data reception, data loopback and the collision de- tection functions are disabled, and remain disabled until valid data or >5 consecutive link pulses appear on the rxd pair. during link fail, the link status ( lnkst pin) signal is inactive. when the link is identified as func- tional, the link status signal is asserted. the lnkst pin displays the link status signal by default. transmission attempts during link fail state will pro- duce no network activity and will produce lcar and cerr error indications. in order to inter-operate with systems which do not im- plement link test, this function can be disabled by set- ting the dlnktst bit in csr15. with link test disabled, the data driver, receiver and loopback functions as well as collision detection remain enabled irrespective of the presence or absence of data or link pulses on the rxd pair. link test pulses continue to be sent regardless of the state of the dlnktst bit. polarity detection and reversal the t-mau receive function includes the ability to invert the polarity of the signals appearing at the rxd pair if the polarity of the received signal is reversed (such as in the case of a wiring error). this feature allows data frames received from a reverse wired rxd input pair to be corrected in the t-mau prior to transfer to the men- dec. the polarity detection function is activated follow- ing h_reset or link fail, and will reverse the receive polarity based on both the polarity of any previous link beat pulses and the polarity of subsequent frames with a valid end transmit delimiter (etd). when in the link fail state, the t-mau will recognize link beat pulses of either positive or negative polarity. exit from the link fail state is made due to the reception of 5 C 6 consecutive link beat pulses of identical polarity. on entry to the link pass state, the polarity of the last 5 link beat pulses is used to determine the initial receive polarity configuration and the receiver is reconfigured to subsequently recognize only link beat pulses of the pre- viously recognized polarity. positive link beat pulses are defined as received signal with a positive amplitude greater than 585 mv (lrt = high) with a pulse width of 60 ns C 200 ns. this positive excursion may be followed by a negative excursion. this definition is consistent with the expected received signal at a correctly wired receiver, when a link beat pulse which fits the template of figure 14-12 of the 10base-t standard is generated at a transmitter and passed through 100 m of twisted pair cable. negative link beat pulses are defined as received sig- nals with a negative amplitude greater than 585 mv with a pulse width of 60 ns C 200 ns. this negative excursion may be followed by a positive excursion. this definition is consistent with the expected received signal at a re- verse wired receiver, when a link beat pulse which fits the template of figure 14-12 in the 10base-t standard is generated at a transmitter and passed through 100 m of twisted pair cable. the polarity detection/correction algorithm will remain armed until two consecutive frames with valid etd of identical polarity are detected. when armed, the re- ceiver is capable of changing the initial or previous po- larity configuration based on the etd polarity. on receipt of the first frame with valid etd following h_reset or link fail, the t-mau will utilize the inferred polarity information to configure its rxd input, regard- less of its previous state. on receipt of a second frame with a valid etd with correct polarity, the detection/cor- rection algorithm will lock-in the received polarity. if the second (or subsequent) frame is not detected as con- firming the previous polarity decision, the most recently detected etd polarity will be used as the default. note that frames with invalid etd have no effect on updating the previous polarity decision. once two consecutive frames with valid etd have been received, the t-mau will disable the detection/correction algorithm until either a link fail condition occurs or h_reset is activated. during polarity reversal, an internal pol signal will be active. during normal polarity conditions, this internal pol signal is inactive. the state of this signal can be read by software and/or displayed by led when en- abled by the led control bits in the bus configuration registers (bcr4Cbcr7). twisted-pair interface status three internal signals (xmt, rcv and col) indicate whether the t-mau is transmitting, receiving, or in a col- lision state with both functions active simultaneously. these signals are internal signals and the behavior of the led outputs depends on how the led output cir- cuitry is programmed. the t-mau will power up in the link fail state and nor- mal algorithm will apply to allow it to enter the link pass state. in the link pass state, transmit or receive activity will be indicated by assertion of rcv signal going active. if t-mau is selected using the portsel bits in csr15, then when moving from aui to t-mau selection the t-mau will be forced into the link fail state. in the link fail state, xmt, rcv and col are inactive. p r e l i m i n a r y amd 57 AM79C974 collision detect function activity on both twisted pair signals rxd and txd constitutes a collision, thereby causing the col signal to be activated. (col is used by the led control circuits) col will remain active until one of the two colliding sig- nals changes from active to idle. however, transmission attempt in link fail state results in lcar and cerr in- dication. col stays active for 2 bit times at the end of a collision. signal quality error (sqe) test (heartbeat) function the sqe function is disabled when the 10base-t port is selected. jabber function the jabber function inhibits the twisted pair transmit function of the t-mau txd is active for an excessive period (20 ms C 150 ms). this prevents any one node from disrupting the network due to a stuck-on or faulty transmitter. if this maximum transmit time is exceeded, the t-mau transmitter circuitry is disabled, the jab bit is set (csr4, bit 1) the col signal is asserted. once the transmit data stream to the t-mau is removed, an un- jab time of 250 ms C 750 ms will elapse before the t-mau col and re-enables the transmit circuitry. power down the t-mau circuitry can be made to go into power sav- ings mode. this feature is useful in battery powered or low duty cycle systems. the t-mau will go into power down mode when h_reset is active, coma mode is ac- tive, or the t-mau is not selected. refer to the power savings modes section for descriptions of the various power down modes. any of the three conditions listed above resets the inter- nal logic of the t-mau and places the device into power down mode. in this mode, the twisted-pair driver pins (txd , txp ) are driven low, and the internal t-mau status signals ( lnkst , rcvpol, xmt, rcv and col) signals are inactive. once h_reset ends, coma mode is disabled, and the t-mau is selected. the t-mau will remain in the reset state for up to 10 m s. immediately after the reset condi- tion is removed, the t-mau will be forced into the link fail state. the t-mau will move to the link pass state only after 5 C 6 link beat pulses and/or a single received message is detected on the rd pair. in snooze mode, the t-mau receive circuitry will remain enabled even while the sleep pin is driven low. the t-mau circuitry will always go into power down mode if h_reset is asserted, coma mode is enabled, or the t-mau is not selected. 10base-t interface connection figure 21 shows the proper 10base-t network inter- face design. refer to the technical manual (pid #18738a) for more design details, and refer to appendix b for a list of compatible 10base-t filter/ transformer modules. note that the recommended resistor values and filter and transformer modules are the same as those used by the imr (am79c980) and the imr+ (am79c981). 18681a-25 xmt filter rcv filter rj45 connector filter & transformer module txp+ txd- txp- txd+ rxd+ rxd- AM79C974 td+ td- rd+ rd- 1 2 3 6 61.9 w 422 w 61.9 w 422 w 100 w 1.21 k w 1:1 1:1 figure 21. 10base-t interface connection amd p r e l i m i n a r y 58 AM79C974 ethernet power savings modes the AM79C974s ethernet controller supports two hard- ware power savings modes. both are entered by driving the sleep pin low. the pci interface section is not effected by sleep . in particular, access to the pci configuration space re- mains possible. none of the configuration registers will be reset by sleep . all i/o accesses to the AM79C974s ethernet controller will result in a pci target abort response. the first power saving mode is called coma mode. in coma mode, the AM79C974 controller has no means to use the network to automatically wake itself up. coma mode is enabled when the awake bit in bcr2 is reset. coma mode is the default power down mode. the second power saving mode is called snooze mode. in snooze mode, enabled by setting the awake bit in bcr2 and driving the sleep pin low, the t-mau re- ceive circuitry will remain enabled even while the sleep pin is driven low. the lnkst output is the only one of the led pins that continues to function. the lnkste bit must be set in bcr4 to enable indication of a good 10base-t link if there are link beat pulses or valid frames present. this lnkst pin can be used to drive an led and/or external hardware that directly con- trols the sleep pin of the AM79C974 controller. this configuration effectively wakes the system when there is any activity on the 10base-t link. snooze mode can be used only if the t-mau is the selected network port. link beat pulses are not transmitted during snooze mode. if the req output is active when the sleep pin is as- serted, then the AM79C974 controller will wait until the gnt input is asserted. next, the AM79C974 controller will deassert the req pin and finally, it will internally en- ter either the coma or snooze sleep mode. before the sleep mode is invoked, the AM79C974 con- troller will perform an internal s_reset. this s_reset operation will not affect the values of the bcr registers or the pci configuration space. the sleep pin should not be asserted during power supply ramp-up. if it is desired that sleep be asserted at power up time, then the system must delay the asser- tion of sleep until three clk cycles after the comple- tion of a valid pin rst operation. software access ethernet pci configuration registers the AM79C974 controller supports the 64-byte header portion of the configuration space as predefined by the pci specification revision 2.0. none of the device spe- cific registers in locations 64 C 255 are used by the ethernet controller. the layout of the configuration reg- isters in the header region is shown in the table below. all registers required to identify the AM79C974 control- ler and its function are implemented. additional regis- ters are used to setup the configuration of the AM79C974 controller in a system. p r e l i m i n a r y amd 59 AM79C974 device id vendor id 00h status command 04h base-class sub-class programming if revision id 08h reserved header type latency timer reserved 0ch base address 10h reserved 14h reserved 18h reserved 1ch reserved 20h reserved 24h reserved 28h reserved 2ch reserved 30h reserved 34h reserved 38h reserved reserved interrupt pin interrupt line 3ch 31 24 23 16 15 8 7 0 offset the configuration registers are accessible only by pci configuration cycles. they can be accessed right after the AM79C974 controller is powered-on, even if the read operation of the serial eeprom is still on-going. all multi-byte numeric fields follow little endian byte or- dering. the command register is the only register cleared by h_reset. s_reset as well as asserting sleep have no effect on the value of the pci configura- tion registers. all write accesses to reserved locations have no effect, reads from these locations will return a data value of zero. when the AM79C974 controller samples its idsela or idselb input asserted during a configuration cycle, it will acknowledge the cycle by asserting its devsel out- put. the content of ad[31:00] during the address phase of the configuration cycles must meet the format as shown below: dont dont care care 0 0 dword index 0 0 31 11 10 8 7 6 5 2 1 0 ad[1:0] must both be zeros, since the AM79C974 controller is not a bridge device. it only recognizes con- figuration cycles of type 0 (as defined by the pci speci- fication revision 2.0). ad[7:2] specify the selected dword in the configuration space. ad[7:6] must both be zero, since the AM79C974s ethernet controller does not implement any of the device specific registers in locations 64 C 255. since ad[1:0] and ad[7:6] must all be zero, the lower 8 bits of the address for a configura- tion cycle are equal to the offset of the dword counting from the beginning of the pci configuration space. ad[10:8] specify one of eight possible functions of a pci device. the AM79C974 controller functions as two sin- gle function devices, as indicated in the header type registers of both pci configuration spaces (bit 7, funct = 0). therefore, the AM79C974 controller ig- nores ad[10:8] during the address phase of a configura- tion cycle. ad[31:11] are typically used to generate the idsela or idselb signals. the AM79C974 controller ignores all upper address bits. pci configuration registers can be accessed with 8-bit, 16-bit or 32-bit transfers. the active bytes within a dword are determined by the byte enable signals. e.g. a read of the sub-class register can be performed by reading from offset 08h with only be 2 being active. i/o resources the AM79C974 controller uses two separate blocks of i/o space, one for the scsi controller and one for the ethernet controller. this section discusses the i/o ad- dress block used by the ethernet controller. pcnet-scsi's ethernet controller i/o resource mapping the AM79C974s ethernet controller has several i/o re- sources. these resources use 32 bytes of i/o space that begin at the AM79C974s ethernet controller i/o base address. the ethernet controller allows two modes of slave ac- cess. word i/o mode treats all ethernet controller i/o resources as two-byte entities spaced at two-byte ad- dress intervals. double word i/o mode treats all ether- net controller i/o resources as four-byte entities spaced at four-byte address intervals. the selection of wio or dwio mode is accomplished in any of several ways: amd p r e l i m i n a r y 60 AM79C974 h_reset function. eeprom programming of the dwio mode bit of bcr18. automatic determination of dwio mode due to dword (double-word) i/o write access to offset 10h. the ethernet controller i/o mode setting will default to wio after h_reset (i.e. dwio = 0). following the h_reset operation, a pread operation of the eeprom will be executed. if the eeprom is pro- grammed with a one in the dwio bit position and the eeprom checksum is correct, then the eeprom read will result in the setting of the dwio mode to a one. if the eeprom programming was absent, failed, or con- tained a zero in the dwio bit position, then the soft- ware may invoke dwio mode by performing a double word write access to the i/o location at offset 10h (rdp). note that even though the i/o resource mapping changes when the i/o mode setting changes, the rdp location offset is the same for both modes. 1-, 2- and 3-byte accesses to ethernet controller i/o re- sources are not allowed during dwio mode. the mapping of the ethernet controller resources into the 32-byte i/o space varies depending upon the setting of the dwio bit of bcr10. depending upon the setting of this variable, the 32-byte i/o space will be either word i/o mapped (wio) or double word i/o mapped (dwio). a dwio setting of 0 produces word i/o mode, while a dwio setting of 1 produces double word i/o mapping. dwio is automatically programmed as active when the system attempts a dword write access to offset 10h of the ethernet controller i/o space. as long as no dword write access to offset 10h of the ethernet con- troller i/o space is performed, the ethernet controller will use the value of dwio that was programmed from the eeprom read operation. if no eeprom is used in the system, then the power up reset value of dwio will be zero, and this value will be maintained until a dword access is performed to ethernet controller i/o space. therefore, if dwio mode is desired, it is imperative that the first access to the ethernet controller be a dword write access to offset 10h, unless the eeprom will be used to program the dwio bit. alternatively, if dwio mode is not desired, then it is im- perative that the software never executes a dword write access to offset 10h of the ethernet controller i/o space, and the eeprom programming of the dwio bit must be zero. once the dwio bit has been set to a one, only a hard- ware h_reset or a new read of the eeprom can re- set it to a zero. the dwio mode setting is unaffected by the s_reset or setting the stop bit. wio i/o resource map when the ethernet controller i/o space is mapped as word i/o, then the resources that are allotted to the ethernet controller occur on word boundaries that are offset from the ethernet controller i/o base address as shown in the table below: no. of offset bytes register 0h 2 aprom 2h 2 aprom 4h 2 aprom 6h 2 aprom 8h 2 aprom ah 2 aprom ch 2 aprom eh 2 aprom 10h 2 rdp 12h 2 rap (shared by rdp and bdp) 14h 2 reset register 16h 2 bdp 18h 2 vendor specific word 1ah 2 reserved 1ch 2 reserved 1eh 2 reserved when ethernet controller i/o space is word mapped, all i/o resources fall on word boundaries and all i/o re- sources are word quantities. however, while in word i/o mode, aprom locations may also be accessed as indi- vidual bytes either on odd or even byte addresses. attempts to write to any ethernet controller i/o re- sources (except to offset 10h, rdp) as 32 bit quantities while in word i/o mode are illegal and may cause unex- pected reprogramming of the ethernet controller control registers. attempts to read from any ethernet controller i/o resources as 32-bit quantities while in word i/o mode are illegal and will yield undefined values. an attempt to write to offset 10h (rdp) as a 32 bit quan- tity while in word i/o mode will cause the ethernet con- troller to exit wio mode and immediately thereafter, to enter dwio mode. word accesses to non word address boundaries are not allowed while in wio mode. (a write access may cause unexpected reprogramming of the ethernet controller control registers. a read access will yield undefined values.) p r e l i m i n a r y amd 61 AM79C974 accesses of non word quantities to any i/o resource are not allowed while in wio mode, with the exception of a read to aprom locations. (a write access may cause unexpected reprogramming of the ethernet controller control registers; a read access will yield undefined values.) the vendor specific word (vsw) is not implemented by the ethernet controller. this particular i/o address is re- served for customer use and will not be used by future amd ethernet controller products. dwio i/o resource map when the ethernet controller i/o space is mapped as double word i/o, then all of the resources that are allot- ted to the ethernet controller occur on dword bounda- ries that are offset from the ethernet controller i/o base address as shown in the table below: no. of offset bytes register 0h 4 aprom 4h 4 aprom 8h 4 aprom ch 4 aprom 10h 4 rdp 14h 4 rap (shared by rdp and bdp) 18h 4 reset register 1ch 4 bdp when ethernet i/o space is double word mapped, all i/o resources fall on dword boundaries. aprom re- sources are dword quantities in dwio mode. rdp, rap and bdp contain only two bytes of valid data; the other two bytes of these resources are reserved for fu- ture use. (note that csr88 is an exception to this rule.) the reserved bits must be written as zeros, and when read, are considered undefined. accesses to non-doubleword address boundaries are not allowed while in dwio mode. (a write access may cause unexpected reprogramming of the ethernet con- troller control registers; a read access will yield unde- fined values.) accesses of less than 4 bytes to any i/o resource are not allowed while in dwio mode. (a write access may cause unexpected reprogramming of the ethernet con- troller control registers; a read access will yield unde- fined values.) if an eeprom is not used to program the value of dwio, then a dword write access to the rdp offset of 10h will automatically program dwio mode. note that in all cases when i/o resource width is defined as 32 bits, the upper 16 bits of the i/o resource is reserved and written as zeros and read as undefined, except for the aprom locations and csr88. dwio mode is exited by asserting the rst pin or by forcing a re-read of the eeprom when the eeprom will program a zero into the dwio bit location of bcr10. assertion of s_reset or setting the stop bit of csr0 will have no effect on the dwio mode setting. i/o space comments the following statements apply to both wio and dwio mapping: the rap is shared by the rdp and the bdp. the ethernet controller does not respond to any ad- dresses outside of the offset range 0hC17h when dwio = 0 or 0hC1fh when dwio = 1. i/o offsets 18h through 1fh are not used by the ethernet controller when pro- grammed for dwio = 0 mode; locations 1ah through 1fh are reserved for future amd use and therefore should not be implemented by the user if upward com- patibility to future amd devices is desired. note that aprom accesses do not directly access the eeprom, but are redirected to a set of shadow regis- ters on board the ethernet controller that contain a copy of the eeprom contents that was obtained during the automatic eeprom read operation that follows the h_reset operation. AM79C974's ethernet controller i/o base address the ethernet pci configuration space base address register defines what i/o base address the ethernet controller uses. this register is typically programmed by the pci configuration utility after system power-up. the pci configuration utility must also set the ioen bit in the command register to enable i/o accesses to the ethernet controller. the contents of the ethernet i/o base address regis- ters (bcr16 and bcr17) are ignored. i/o register access all i/o resources are accessed with similar i/o bus cycles. i/o accesses to the ethernet controller begin with a valid frame signal, the c/ be [3:0] lines signaling an i/o read or i/o write operation and an address on the ad[31:00] lines that falls within the i/o space of the ethernet con- troller. the ethernet i/o space will be determined by the base address register in the pci configuration space. the ethernet controller will respond to an access to its i/o space by asserting the devsel signal and eventu- ally, by asserting the trdy signal. typical i/o access times are 6 or 7 clock cycles. amd p r e l i m i n a r y 62 AM79C974 aprom access the aprom space is a convenient place to store the value of the 48-bit ieee station address. this space is automatically loaded from the serial eeprom, if an eeprom is present. it can be overwritten by the host computer. its contents have no effect on the operation of the controller. the software must copy the station ad- dress from the aprom space to the initialization block or to csr12-14 in order for the receiver to accept unicast frames directed to this station. when programmed for wio mode, any byte or word ad- dress from an offset of 0h to an offset of fh may be read. an appropriate byte or word of aprom contents will be delivered by the AM79C974 controller in response to ac- cesses that fall within the aprom range of 0h to fh. when programmed for dwio mode, only dword ad- dresses from an offset of 0h to an offset of fh may be read. an appropriate dword of aprom contents will be delivered in response to accesses that fall within the aprom range of 0h to fh. accesses of non-dword quantities are not allowed in dwio mode, even though such an access may be prop- erly aligned to a dword address boundary. write access to any of the aprom locations is allowed, but only 4 bytes on dword boundaries in dwio mode or 2 bytes on word boundaries in wio mode (only read accesses to the aprom locations can be in 8-bit quanti- ties while in wio mode). the iesrwe bit (see bcr2) must be set in order to enable a write operation to the aprom. only the AM79C974 controller on-board ieee shadow registers are modified by writes to aprom lo- cations. the eeprom is unaffected by writes to aprom locations. note that the aprom locations occupy 16 bytes of space, yet the ieee station address requirement is for 6 bytes. the 6 bytes of ieee station address occupy the first 6 locations of the aprom space. the next six bytes are reserved. bytes 12 and 13 should match the value of the checksum of bytes 1 through 11 and 14 and 15. bytes 14 and 15 should each be ascii w (57h). the above requirements must be met in order to be compat- ible with amd driver software. rdp access (csr register space) rdp = register data port. the rdp is used with the rap to gain access to any of the AM79C974 controller csr locations. access to any of the csr locations of the AM79C974 controller is performed through the AM79C974 control- lers register data port (rdp). in order to access a par- ticular csr location, the register address port (rap) should first be written with the appropriate csr ad- dress. the rdp now points to the selected csr. a read of the rdp will yield the selected csrs data. a write to the rdp will write to the selected csr. when programmed for wio mode, the rdp has a width of 16 bits, hence, all csr locations have 16 bits of width. note that when accessing rdp, the upper two bytes of the data bus will be undefined since the byte masks will not be active for those bytes. if dwio mode has been invoked, then the rdp has a width of 32 bits, hence, all csr locations have 32 bits of width and the upper two bytes of the data bus will be ac- tive, as indicated by the byte mask. in this case, note that the upper 16 bits of all csr locations (except csr88) are reserved and written as zeros and read as undefined values. therefore, during rdp write opera- tions in dwio mode, the upper 16 bits of all csr loca- tions should be written as zeros. rap access rap = register address port. the rap is used with the rdp and with the bdp to gain access to any of the csr and bcr register locations, respectively. the rap con- tains the address pointer that will be used by an access to either the rdp or bdp. therefore, it is necessary to set the rap value before accessing a specific csr or bcr location. once the rap has been written with a value, the rap value remains unchanged until another rap write occurs, or until an h_reset or s_reset occurs. rap is set to all zeros when an h_reset or s_reset occurs. rap is unaffected by the stop bit. when programmed for wio mode, the rap has a width of 16 bits. note that when accessing rap, the upper two bytes of the data bus will be undefined since the byte masks will not be active for those bytes when programmed for dwio mode, the rap has a width of 32 bits. in dwio mode, the upper 16 bits of the rap are reserved and written as zeros and read as undefined. these bits should be written as zeros. bdp access (bcr register space) bdp = bus configuration register data port. the bdp is used with the rap to gain access to any of the AM79C974 controller bcr locations. access to any of the bcr locations of the AM79C974 controller is performed through the AM79C974 control- lers bcr data port (bdp ); in order to access a particu- lar bcr location, the register address port (rap) should first be written with the appropriate bcr ad- dress. the bdp now points to the selected bcr. a read of the bdp will yield the selected bcr s data. a write to the bdp will write to the selected bcr. when programmed for wio mode, the bdp has a width of 16 bits, hence, all bcr locations have 16 bits of width in wio mode. note that when operating in wio mode, p r e l i m i n a r y amd 63 AM79C974 the upper two bytes of the data bus will be undefined since the byte mask will not be active for those bytes. if dwio mode has been invoked, then the bdp has a width of 32 bits, hence, all bcr locations have 32 bits of width and the upper two bytes of the data bus will be ac- tive, as indicated by the byte mask. in this case, note that the upper 16 bits of all bcr locations are reserved and written as zeros and read as undefined. there- fore, during bdp write operations in dwio mode, the upper 16 bits of all bcr locations should be written as zeros. reset register (s_reset) a read of the reset register creates an internal s_re- set pulse in the AM79C974 controller. this read ac- cess cycle must be 16 bits wide in wio mode and 32 bits wide in dwio mode. the internal s_reset pulse that is generated by this access is different from both the as- sertion of the hardware rst pin (h_reset) and from the assertion of the software stop bit. specifically, the reset registers s_reset will be the equivalent of the assertion of the rst pin (h_reset) assertion for all csr locations, but s_reset will have no effect at all on the bcr or pci configuration space locations, and s_reset will not cause a deassertion of the req pin. the ne2100 lance based family of ethernet cards re- quires that a write access to the reset register follows each read access to the reset register. the AM79C974 controller does not have a similar requirement. the write access is not required but it does not have any harmful effects. write accesses to the reset register will have no effect on the AM79C974 controller. note that a read access of the reset register will take longer than the normal i/o access time of the AM79C974 controller. this is because an internal s_reset pulse will be generated due to this access, and the access will not be allowed to complete on the system bus until the internal s_reset operation has been completed. this is to avoid the problem of allowing a new i/o access to proceed while the s_reset opera- tion has not yet completed, which would result in errone- ous data being returned by (or written into) the AM79C974 controller. the length of a read of the reset register can be as long as 64 clock cycles. note that a read of the reset register will not cause a deassertion of the req signal, if it happens to be active at the time of the read of the reset register. the req sig- nal will remain active until the gnt signal is asserted. following the read of the reset register, on the next clock cycle after the gnt signal is asserted, the AM79C974 controller will deassert the req signal. no bus master accesses will have been performed during this brief bus ownership period. note that this behavior differs from that which occurs fol- lowing the assertion of a minimum-width pulse on the rst pin (h_reset). a rst pin assertion will cause the req signal to deassert within six clock cycles following the assertion. in the rst pin case, the AM79C974 con- troller will not wait for the assertion of the gnt signal be- fore deasserting the req signal. vendor specific word this i/o offset is reserved for use by the system de- signer. the AM79C974 controller will not respond to ac- cesses directed toward this offset. the vendor specific word is only available when the AM79C974 controller is programmed to word i/o mode (dwio = 0). if more than one vendor specific word is needed, it is suggested that the vsw location should be divided into a vsw register address pointer (vswrap) at one lo- cation (e.g. vswrap at byte location 18h or word loca- tion 30h, depending upon dwio state) and a vsw data port (vswdp) at the other location (e.g. vswdp at byte location 19h or word location 32h, depending upon dwio state). alternatively, the system may capture rap data accesses in parallel with the AM79C974 con- troller and therefore share the AM79C974 controller rap to allow expanded vsw space. AM79C974 con- troller will not respond to access to the vsw i/o ad- dress. reserved i/o space these locations are reserved for future use by amd. the AM79C974 controller does not respond to ac- cesses directed toward these locations, but future amd products that are intended to be upward compatible with the AM79C974 controller device may decode accesses to these locations. therefore, the system designer may not utilize these i/o locations. amd p r e l i m i n a r y 64 AM79C974 hardware access pcnet-scsi controller master accesses the AM79C974 controller has a bus interface compat- ible with pci specification revision 2.0. complete descriptions of the signals involved in bus master transactions for each mode may be found in the pin description section of this document. timing dia- grams for master accesses may be found in the block description section for the bus interface unit. this sec- tion simply lists the types of master accesses that will be performed by the AM79C974 controller with respect to data size and address information. the AM79C974 controller will support master accesses only to 32-bit peripherals. the AM79C974 controller does not support master accesses to 8-bit or 16-bit memory. the AM79C974 controller is not compatible with 8-bit systems, since there is no mode that supports AM79C974 controller accesses to 8-bit peripherals. table 3 describes all possible bus master accesses that the AM79C974 controller will perform. the right most column lists all operations that may execute the given access: table 3. bus master accesses access mode be [3:0] operation 4-byte read read 0000 descriptor read or initialization block read or transmit data buffer read 4-byte write write 0000 descriptor write or receive data buffer write 3-byte write write 1000 receive data buffer write 3-byte write write 0001 receive data buffer write 2-byte write write 1100 receive data buffer write 2-byte write write 1001* receive data buffer write 2-byte write write 0011 receive data buffer write 1-byte write write 1110 receive data buffer write 1-byte write write 1101* receive data buffer write 1-byte write write 1011* receive data buffer write 1-byte write write 0111 descriptor write or receive data buffer write * cases marked with an asterisk represent extreme boundary conditions that are the result of programming one- and two-byte buffer sizes, and therefore will not be seen under normal circumstances. note that all AM79C974 controller master read opera- tions will always activate all byte enables. therefore, no one-, two- or three-byte read operations are indicated in the table. in the instance where a transmit buffer pointer address begins on a non-dword boundary, the pointer will be truncated to the next dword boundary address that lies below the given pointer address and the first read access from the transmit buffer will be indicated on the byte enable signals as a four-byte read from this ad- dress. any data from byte lanes that lie outside of the boundary indicated by the buffer pointer will be dis- carded inside of the AM79C974 controller. similarly, if the end of a transmit buffer occurs on a non-dword boundary, then all byte lanes will be indicated as active by the byte enable signals, and any data from byte lanes that lie outside of the boundary indicated by the buffer pointer will be discarded inside of the AM79C974 controller. slave access to i/o resources the AM79C974 device is always a 32-bit peripheral on the system bus. however, the width of individual soft- ware resources on board the AM79C974 controller may be either 16-bits or 32-bits. the AM79C974 controller i/o resource widths are determined by the setting of the dwio bit as indicated in the following table: AM79C974 controller i/o dwio setting resource width example application dwio = 0 16-bit existing pcnet-isa driver that assumes 16-bit i/o mapping and 16-bit resource widths dwio = 1 32-bit new drivers written specifically for the AM79C974 controller p r e l i m i n a r y amd 65 AM79C974 note that when i/o resource width is defined as 32 bits (dwio mode), the upper 16 bits of the i/o resource is reserved and written as zeros and read as undefined, except for the aprom locations and csr88 . the aprom locations and csr88 are the only i/o re- sources for which all 32 bits will have defined values. however, this is true only when the AM79C974 control- ler is in dwio mode. configuring the AM79C974 controller for dwio mode is accomplished whenever there is any attempt to perform a 32-bit write access to the rdp location (offset 10h). see the dwio section for more details. table 4 describes all possible bus slave accesses that may be directed toward the AM79C974 controller. (i.e., the AM79C974 controller is the target device during the transfer.) the first column indicates the type of slave ac- cess. rd stands for read, wr for a write operation. the second column indicates the value of the c/ be [3:0] lines during the data phase of the transfer. the four byte columns (ad[31:24], ad[23:16], ad[15:8], ad[7:0]) in- dicate the value on the address/data bus during the data phase of the access. data indicates the position of the active bytes; copy indicates the positions of copies of the active bytes; undef indicates byte locations that are undefined during the transfer. table 4. bus slave accesses ad ad ad ad type be [3:0] [31:24] [23:16] [15:8] [7:0] comments rd 0000 data data data data dword access to dword address, e.g. 300h, 30ch, 310h (dwio mode only) rd 1100 undef undef data data word access to even word address, e.g. 300h, 30ch, 310h (wio mode only) rd 0011 data data copy copy word access to odd word address, e.g. 302h, 30eh, 312h (wio mode only) rd 1110 undef undef undef data byte access to lower byte of even word address, e.g. 300h, 304h (wio mode only, aprom accesses only) rd 1101 undef undef data undef byte access to upper byte of even word address, e.g. 301h, 305h (wio mode only, aprom accesses only) rd 1011 undef data undef copy byte access to lower byte of odd word address, e.g. 302h, 306h (wio mode only, aprom accesses only) rd 0111 data undef copy undef byte access to upper byte of odd word address, e.g. 303h, 307h (wio mode only, aprom accesses only) wr 0000 data data data data dword access to dword address, e.g. 300h, 30ch, 310h (dwio mode only) wr 1100 undef undef data data word access to even word address, e.g. 300h, 30ch, 310h (wio mode only) wr 0011 data data undef undef word access to odd word address, e.g. 302h, 30eh, 312h (wio mode only) amd p r e l i m i n a r y 66 AM79C974 eeprom microwire access the AM79C974 controller contains a built-in capability for reading and writing to an external eeprom. this built-in capability consists of a microwire interface for di- rect connection to a microwire compatible eeprom, an automatic eeprom read feature, and a user-program- mable register that allows direct access to the microwire interface pins. automatic eeprom read operation shortly after the deassertion of the rst pin, the AM79C974 controller will read the contents of the eeprom that is attached to the microwire interface. because of this automatic-read capability of the AM79C974 controller, an eeprom can be used to pro- gram many of the features of the AM79C974 controller at power-up, allowing system-dependent configuration information to be stored in the hardware, instead of in- side of operating code. if an eeprom exists on the microwire interface, the AM79C974 controller will read the eeprom contents at the end of the h_reset operation. the eeprom con- tents will be serially shifted into a temporary register and then sent to various register locations on board the AM79C974 controller. the host can access the pci configuration space during the eeprom read opera- tions. access to the AM79C974 i/o resources, however, is not possible during the eeprom read operation. the AM79C974 controller will terminate these i/o accesses with the assertion of devsel and stop while trdy is not asserted, signaling to the initiator to retry the access at a later time. a checksum verification is performed on the data that is read from the eeprom. if the checksum verification of the eeprom data fails, then at the end of the eeprom read sequence, the AM79C974 controller will force all eeprom-programmable bcr registers back to their h_reset default values. the content of the aprom locations (offsets 0h C fh from the i/o base address), however, will not be cleared. the 8-bit checksum for the entire 36 bytes of the eeprom should be ffh. if no eeprom is present at the time of the automatic read operation, then the AM79C974 controller will rec- ognize this condition and will abort the automatic read operation and reset both the pread and pvalid bits in bcr19. all eeprom-programmable bcr registers will be assigned their default values after h_reset. the content of the address prom locations (offsets 0h C fh from the i/o base address) will be undefined. if the user wishes to modify any of the configuration bits that are contained in the eeprom, then the seven com- mand, data and status bits of bcr19 can be used to write to the eeprom. after writing to the eeprom, the host should set the pread bit of bcr19. this action forces a AM79C974 controller re-read of the eeprom so that the new eeprom contents will be loaded into the eeprom-programmable registers on board the AM79C974 controller. (the eeprom-programmable registers may also be reprogrammed directly, but only information that is stored in the eeprom will be pre- served at system power-down.) when the pread bit of bcr19 is set, it will cause the AM79C974 controller to terminate further accesses to internal i/o resources with the pci retry cycle. accesses to the pci configuration space is still possible. eeprom auto-detection the AM79C974 controller uses the eesk/ led1 pin to determine if an eeprom is present in the system. at all rising clk edges during the assertion of the rst pin, the AM79C974 controller will sample the value of the eesk/ led1 pin. if the sampled value is a one, then the AM79C974 controller assumes that an eeprom is pre- sent, and the eeprom read operation begins shortly after the rst pin is deasserted. if the sampled value of eesk/ led1 is a zero, then the AM79C974 controller assumes that an external pulldown device is holding the eesk/ led1 pin low, and therefore, there is no eeprom in the system. note that if the designer cre- ates a system that contains an led circuit on the eesk/ led1 pin but has no eeprom present, then the eeprom auto-detection function will incorrectly con- clude that an eeprom is present in the system. how- ever, this will not pose a problem for the AM79C974 controller, since it will recognize the lack of an eeprom at the end of the read operation, when the checksum verification fails. systems without an eeprom some systems may be able to save the cost of an eeprom by storing the iso 8802-3 (ieee/ansi 802.3) station address and other configuration information somewhere else in the system. there are several de- sign choices: if the led1 is not needed in the system, then the sys- tem designer may connect the eesk/ led1 pin to a resistive pulldown device. this will indicate to the eeprom auto-detection function that no eeprom is present. if the led1 function is needed in the system, then the system designer will connect the eesk/ led1 pin to a resistive pullup device and the eeprom auto-de- tection function will incorrectly conclude that an eeprom is present in the system. however, this will not pose a problem for the AM79C974 controller, since it will recognize the lack of an eeprom at the end of the read operation, when the checksum verifi- cation fails. in either case, following the pci configuration, addi- tional information, including the iso 8802-3 (ieee/ansi 802.3) station address, may be loaded into the AM79C974 controller. note that the iesrwe bit (bit 8 of bcr2) must be set before the AM79C974 controller will p r e l i m i n a r y amd 67 AM79C974 accept writes to the aprom offsets within the AM79C974 i/o resources map. startup code in the sys- tem bios can perform the pci configuration accesses, the iesrwe bit write, and the aprom writes. direct access to the microwire interface the user may directly access the microwire port through the eeprom register, bcr19. this register contains bits that can be used to control the microwire interface pins. by performing an appropriate sequence of i/o ac- cesses to bcr19, the user can effectively write to and read from the eeprom. this feature may be used by a system configuration utility to program hardware con- figuration information into the eeprom. eeprom-programmable registers the following registers contain configuration informa- tion that will be programmed automatically during the eeprom read operation: 1. i/o offsets 0h C fh aprom locations 2. bcr2 miscellaneous configuration register 3. bcr16 not used in the AM79C974 controller 4. bcr17 not used in the AM79C974 controller 5. bcr18 burst size and bus control register 6. bcr21 not used if the pread bit (bcr19) is reset to zero and the pvalid bit (bcr19) is reset to zero, then the eeprom read has experienced a failure and the con- tents of the eeprom programmable bcr register will be set to default h_reset values. the content of the aprom locations, however, will not be cleared. note that accesses to the aprom i/o locations do not directly access the address eeprom itself. instead, these accesses are routed to a set of shadow registers on board the AM79C974 controller that are loaded with a copy of the eeprom contents during the automatic read operation that immediately follows the h_reset operation. amd p r e l i m i n a r y 68 AM79C974 eeprom map the automatic eeprom read operation will access 18 words (i.e. 36 bytes) of the eeprom. the format of the eeprom contents is shown in table 5, beginning with the byte that resides at the lowest eeprom address: table 5. eeprom contents eeprom word byte byte address addr. most significant byte addr. least significant byte 00h 01h 2nd byte of the iso 8802-3 00h first byte of the iso 8802-3 (lowest (ieee/ansi 802.3) station physical (ieee/ansi 802.3) station physical eeprom address for this node address for this node, where first byte address) refers to the first byte to appear on the 802.3 medium 01h 03h 4th byte of the node address 02h 3rd byte of the node address 02h 05h 6th byte of the node address 04h 5th byte of the node address 03h 07h reserved location: must be 00h 06h reserved location must be 00h 04h 09h hardware id: must be 11h if 08h reserved location must be 00h compatibility to amd drivers is desired 05h 0bh user programmable space 0ah user programmable space 06h 0dh msbyte of two-byte checksum, 0ch lsbyte of two-byte checksum, which is the which is the sum of bytes 00hC0bh sum of bytes 00hC0bh and bytes 0eh and 0fh and bytes 0eh and 0fh 07h 0fh must be ascii w (57h) if compatibility 0eh must be ascii w (57h) if compatibility to amd driver software is desired to amd driver software is desired 08h 11h bcr16[15:8] (not used) 10h bcr16[7:0] (not used) 09h 13h bcr17[15:8] (not used) 12h bcr17[7:0] (not used) 0ah 15h bcr18[15:8] (burst size and bus control) 14h bcr18[7:0] (burst size and bus control) 0bh 17h bcr2[15:8] (misc. configuration) 16h bcr2[7:0] (misc. configuration) 0ch 19h bcr21[15:8] (not used) 18h bcr21[7:0] (not used) 0dh 1bh reserved location must be 00h 1ah reserved location must be 00h 0eh 1dh reserved location must be 00h 1ch reserved location must be 00h 0fh 1fh checksum adjust byte for the first 36 bytes 1eh reserved location must be 00h of the eeprom contents; checksum of the first 36 bytes of the eeprom should total to ffh 10h 21h reserved location must be 00h 20h reserved location must be 00h 11h 23h user programmable byte locations 22h user programmable byte locations eeprom contents note that the first bit out of any word location in the eeprom is treated as the msb of the register that is be- ing programmed. for example, the first bit out of eeprom word location 08h will be written into bcr16[15], the second bit out of eeprom word loca- tion 08h will be written into bcr16[14], etc. there are two checksum locations within the eeprom. the first is required for the eeprom address. this checksum will be used by amd driver software to verify that the iso 8802-3 (ieee/ansi 802.3) station address has not been corrupted. the value of bytes 0ch and 0dh should match the sum of bytes 00h through 0bh and 0eh and 0fh. the second checksum location C byte 1fh C is not a checksum total, but is, instead, a check- sum adjustment. the value of this byte should be such that the total checksum for the entire 36 bytes of eeprom data equals the value ffh. the checksum ad- just byte is needed by the AM79C974 controller in order to verify that the eeprom contents have not been corrupted. p r e l i m i n a r y amd 69 AM79C974 transmit operation the transmit operation and features of the AM79C974 controller are controlled by programmable options. the AM79C974 controller offers a136-byte transmit fifo to provide frame buffering for increased system latency, automatic retransmission with no fifo reload, and automatic transmit padding. transmit function programming automatic transmit features such as retry on collision, fcs generation/transmission, and pad field insertion can all be programmed to provide flexibility in the (re-)transmission of messages. disable retry on collision (drty) is controlled by the drty bit of the mode register (csr15) in the initiali- zation block. automatic pad field insertion is controlled by the apad_xmt bit in csr4. if apad_xmt is set, auto- matic pad field insertion is enabled, the dxmtfcs fea- ture is over-ridden, and the 4-byte fcs will be added to the transmitted frame unconditionally. if apad_xmt is clear, no pad field insertion will take place and runt packet transmission is possible. the disable fcs generation/transmission feature can be programmed dynamically on a frame by frame basis. see the add_fcs description of tmd1. transmit fifo watermark (xmtfw) in csr80 sets the point at which the bmu requests more data from the transmit buffers for the fifo. a minimum of xmtfw empty spaces must be available in the transmit fifo be- fore the bmu will request the system bus in order to transfer transmit packet data into the transmit fifo. transmit start point (xmtsp) in csr80 sets the point when the transmitter actually attempts to transmit a frame onto the media. a minimum of xmtsp bytes must be written to the transmit fifo for the current frame be- fore transmission of the current frame will begin. (when automatically padded packets are being sent, it is con- ceivable that the xmtsp is not reached when all of the data has been transferred to the fifo. in this case, the transmission will begin when all of the packet data has been placed into the transmit fifo.) when the entire frame is in the fifo, attempts at trans- mission of preamble will commence regardless of the value in xmtsp. the default value of xmtsp is 10b, meaning there has to be 64 bytes in the transmit fifo to start a transmission. automatic pad generation transmit frames can be automatically padded to extend them to 64 data bytes (excluding preamble). this allows the minimum frame size of 64 bytes (512 bits) for 802.3/ethernet to be guaranteed with no software inter- vention from the host/controlling process. setting the apad_xmt bit in csr4 enables the auto- matic padding feature. the pad is placed between the llc data field and fcs field in the 802.3 frame.fcs is always added if the frame is padded, regardless of the state of dxmtfcs. the transmit frame will be padded by bytes with the value of 00h. the default value of apad_xmt is 0; this will disable auto pad generation af- ter h_reset. preamble 1010....1010 sync 10101011 destination address source address length llc data pad fcs 4 bytes 46 1500 bytes 2 bytes 6 bytes 6 bytes 8 bits 56 bits 18681a-26 figure 22 . iso 8802-3 (ieee/ansi 802.3) data frame it is the responsibility of upper layer software to correctly define the actual length field contained in the message to correspond to the total number of llc data bytes en- capsulated in the packet (length field as defined in the iso 8802-3 (ieee/ansi 802.3) standard). the length value contained in the message is not used by the AM79C974 controller to compute the actual number of pad bytes to be inserted. the AM79C974 controller will append pad bytes dependent on the actual number of bits transmitted onto the network. once the last data byte of the frame has completed, prior to appending the fcs, the AM79C974 controller will check to ensure that 544 bits have been transmitted. if not, pad bytes are added to extend the frame size to this value, and the fcs is then added. amd p r e l i m i n a r y 70 AM79C974 the 544 bit count is derived from the following: minimum frame size (excluding preamble, including fcs) 64 bytes 512 bits preamble/sfd size 8 bytes 64 bits fcs size 4 bytes 32 bits to be classed as a minimum size frame at the receiver, the transmitted frame must contain: preamble + (min frame size + fcs) bits at the point that fcs is to be appended, the transmitted frame should contain: preamble + (min frame size C fcs) bits 64 + (512 C 32) bits a minimum length transmit frame from the AM79C974 controller will therefore be 576 bits, after the fcs is appended. the ethernet specification assumes that minimum length messages will be at least 64 bytes in length. transmit fcs generation automatic generation and transmission of fcs for a transmit frame depends on the value of dxmtfcs bit in csr15. when dxmtfcs = 0 the transmitter will gener- ate and append the fcs to the transmitted frame. if the automatic padding feature is invoked (apad_xmt is set in csr4), the fcs will be appended by the AM79C974 controller regardless of the state of dxmtfcs. note that the calculated fcs is transmitted most significant bit first. the default value of dxmtfcs is 0 after h_reset. transmit exception conditions exception conditions for frame transmission fall into two distinct categories. those which are the result of normal network operation, and those which occur due to abnor- mal network and/or host related events. normal events which may occur and which are handled autonomously by the AM79C974 controller include colli- sions within the slot time with automatic retry. the AM79C974 controller will ensure that collisions which occur within 512 bit times from the start of transmission (including preamble) will be automatically retried with no host intervention. the transmit fifo ensures this by guaranteeing that data contained within the fifo will not be overwritten until at least 64 bytes (512 bits) of pre- amble plus address, length and data fields have been transmitted onto the network without encountering a collision. if 16 total attempts (initial attempt plus 15 retries) fail, the AM79C974 controller sets the rtry bit in the current transmit tdte in host memory (tmd2), gives up ownership (resets the own bit to zero) for this frame, and processes the next frame in the transmit ring for transmission. abnormal network conditions include: loss of carrier. late collision. sqe test error. (does not apply to 10base-t port) these should not occur on a correctly configured 802.3 network, and will be reported if they do. when an error occurs in the middle of a multi-buffer frame transmission, the error status will be written in the current descriptor. the own bit(s) in the subsequent descriptor(s) will be reset until the stp (the next frame) is found. loss of carrier a loss of carrier condition will be reported if the AM79C974 controller cannot observe receive activity whilst it is transmitting on the aui port. after the AM79C974 controller initiates a transmission it will ex- pect to see data looped-back on the di pair. this will internally generate a carrier sense, indicating that the integrity of the data path to and from the mau is intact, and that the mau is operating correctly. this carrier sense signal must be asserted about 6 bit times before the last transmitted bit on do . if carrier sense does not become active in response to the data transmission, or becomes inactive before the end of transmission, the loss of carrier (lcar) error bit will be set in tmd2 after the frame has been transmitted. the frame will not be re-tried on the basis of an lcar error. when the 10base-t port is selected, lcar will be re- ported for every packet transmitted during the link fail condition. late collision a late collision will be reported if a collision condition oc- curs after one slot time (512 bit times) after the transmit process was initiated (first bit of preamble commenced). the AM79C974 controller will abandon the transmit process for the particular frame, set late collision (lcol) in the associated tmd2, and process the next transmit frame in the ring. frames experiencing a late collision will not be re-tried. recovery from this condition must be performed by upper layer software. sqe test error during the inter packet gap time following the comple- tion of a transmitted message, the aui ci pair is as- serted by some transceivers as a self-test. the integral manchester encoder/decoder will expect the sqe test message (nominal 10 mhz sequence) to be returned via the ci pair, within a 40 network bit time period after di goes inactive (this does not apply if the 10base-t port is selected). if the ci input is not asserted within the 40 p r e l i m i n a r y amd 71 AM79C974 network bit time period following the completion of transmission, then the AM79C974 controller will set the cerr bit in csr0. cerr will be asserted in 10base-t mode after transmit if t-mau is in link fail state. cerr will never cause inta to be activated. it will, however, set the err bit csr0. receive operation the receive operation and features of the AM79C974 controller are controlled by programmable options. address matching the AM79C974 controller supports three types of ad- dress matching: unicast, multicast, and broadcast. the normal address matching procedure can be modified by programming three bits in the mode register (prom, drcvbc, and drcvpa). if the first bit received after the start of frame delimiter (the least significant bit of the first byte of the destination address field) is 0, the frame is unicast, which indicates that the frame is meant to be received by a single node. if the first bit received is 1, the frame is multicast, which indicates that the frame is meant to be received by a group of nodes. if the destination address field contains all ones, the frame is broadcast, which is a special type of multicast. frames with the broadcast address in the destination address field are meant to be received by all nodes on the local area network. when a unicast frame arrives at the AM79C974 control- ler, the controller will accept the frame if the destination address field of the incoming frame exactly matches the 6-byte station address stored in the padr registers (csr12, csr13, and csr14). the byte ordering is such that the first byte received from the network (after the sfd) must match the least significant byte of csr12 (padr[7:0]), and the sixth byte received must match the most significant byte of csr14 (padr[47:40]). if drcvpa (bit 13 in the mode register) is set. the AM79C974 controller will not accept unicast frames. if the incoming frame is multicast the AM79C974 con- troller performs a calculation on the contents of the des- tination address field to determine whether or not to accept the frame. this calculation is explained in the section that describes the logical address filter (ladrf). if all bits of the ladrf registers are 0 no multicast frames are accepted, except for broadcast frames. although broadcast frames are classified as special multicast frames, they are treated differently by the AM79C974 controller hardware. broadcast frames are always accepted, except when drcvbc (bit 14 in the mode register) is set. none of the address filtering described above applies when the AM79C974 controller is operating in the promiscuous mode. in the promiscuous mode, all prop- erly formed packets are received, regardless of the con- tents of their destination address fields. the promiscuous mode overrides the disable receive broadcast bit (drcvbc bit l4 in the mode register) and the disable receive physical address bit (drcvpa, bit 13 mode register). the AM79C974 controller operates in promiscuous mode when prom (bit 15 in the mode register) is set. receive function programming automatic pad field stripping is enabled by setting the astrp_rcv bit in csr4. this can provide flexibility in the reception of messages using the 802.3 frame format. all receive frames can be accepted by setting the prom bit in csr15. when prom is set, the AM79C974 con- troller will attempt to receive all messages, subject to minimum frame enforcement. promiscuous mode over rides the effect of the disable receive broadcast bit on receiving broadcast frames. the point at which the bmu will start to transfer data from the receive fifo to buffer memory is controlled by the rcvfw bits in csr80. the default established dur- ing h_reset is 10b which sets the threshold flag at 64 bytes empty. automatic pad stripping during reception of an 802.3 frame the pad field can be stripped automatically. astrp_rcv (csr4, bit 0) = 1 enables the automatic pad stripping feature. the pad field will be stripped before the frame is passed to the fifo, thus preserving fifo space for additional frames. the fcs field will also be stripped, since it is computed at the transmitting station based on the data and pad field characters, and will be invalid for a receive frame that has had the pad characters stripped. the number of bytes to be stripped is calculated from the embedded length field (as defined in the iso 8802-3 (ieee/ansi 802.3) definition) contained in the frame. the length indicates the actual number of llc data bytes contained in the message. any received frame which contains a length field less than 46 bytes will have the pad field stripped (if astrp_rcv is set). receive frames which have a length field of 46 bytes or greater will be passed to the host unmodified. since any valid ethernet type field value will always be greater than a normal 802.3 length field ( 3 46), the AM79C974 controller will not attempt to strip valid ethernet frames. note that for some network protocols, the value passed in the ethernet type and/or 802.3 length field is not compliant with either standard and may cause problems. amd p r e l i m i n a r y 72 AM79C974 figure 23 shows the byte/bit ordering of the received length field for an 802.3 compatible frame format. preamble 1010....1010 sync 10101011 destination address source address length llc data pad fcs 4 bytes 46 1500 bytes 2 bytes 6 bytes 6 bytes 8 bits 56 bits start of frame at time = 0 increasing time bit 0 bit 7 bit 0 bit 7 most significant byte least significant byte 1 1500 bytes 45 0 bytes 18681a-27 figure 23. 802.3 frame and length field transmission order receive fcs checking reception and checking of the received fcs is per- formed automatically by the AM79C974 controller. note that if the automatic pad stripping feature is enabled, the fcs for padded frames will be verified against the value computed for the incoming bit stream including pad characters, but the fcs value for a padded frame will not be passed to the host. if an fcs error is detected in any frame, the error will be reported in the crc bit in rmd1. receive exception conditions exception conditions for frame reception fall into two distinct categories; those which are the result of normal network operation, and those which occur due to abnor- mal network and/or host related events. normal events which may occur and which are handled autonomously by the AM79C974 controller are basically collisions within the slot time and automatic runt packet rejection. the AM79C974 controller will ensure that col- lisions which occur within 512 bit times from the start of reception (excluding preamble) will be automatically de- leted from the receive fifo with no host intervention. the receive fifo will delete any frame which is com- posed of fewer than 64 bytes provided that the runt packet accept (rpa bit in csr124) feature has not been enabled. this criterion will be met regardless of whether the receive frame was the first (or only) frame in the fifo or if the receive frame was queued behind a previously received message. abnormal network conditions include: fcs errors late collision host related receive exception conditions include miss, buff, and oflo. these are described in the bmu section. loopback operation loopback is a mode of operation intended for system di- agnostics. in this mode, the transmitter and receiver are both operating at the same time so that the controller re- ceives its own transmissions. the controller provides two types of internal loopback and one type of external loopback. in internal loopback mode, the transmitted data can be looped back to the receiver at one of two places inside the controller without actually transmitting any data to the external network. the receiver will move the received data to the next receive buffer, where it can be examined by software. alternatively, in external loop- back mode, data can be transmitted to and received from the external network. p r e l i m i n a r y amd 73 AM79C974 there are restrictions on loopback operation. the AM79C974 controller has only one fcs generator cir- cuit. the fcs generator can be used by the transmitter to generate the fcs to append to the frame, or it can be used by the receiver to verify the fcs of the received frame. it can not be used by the receiver and transmitter simultaneously. if the fcs generator is connected to the receiver, the transmitter will not append an fcs to the frame, but the receiver will check for one. the user can, however, cal- culate the fcs value for a frame and include this four- byte number in the transmit buffer. if the fcs generator is connected to the transmitter, the transmitter will append an fcs to the frame, but the re- ceiver will not check for the fcs. however, the user can verify the fcs by software. during loopback, the fcs logic can be allocated to the receiver by setting dxmtfcs = 1 in csr15. if dxmtfcs=0, the mac engine will calculate and ap- pend the fcs to the transmitted message. the receive message passed to the host will therefore contain an ad- ditional 4 bytes of fcs. in this loopback configuration, the receive circuitry cannot detect fcs errors if they occur. if dxmtfcs=1, the last four bytes of the transmit mes- sage must contain the (software generated) fcs com- puted for the transmit data preceding it. the mac engine will transmit the data without addition of an fcs field, and the fcs will be calculated and verified at the receiver. the loopback facilities of the mac engine allow full op- eration to be verified without disturbance to the network. loopback operation is also affected by the state of the loopback control bits (loop, mendecl, and intl) in csr15. this affects whether the internal mendec is considered part of the internal or external loopback path. the multicast address detection logic uses the fcs generator circuit. therefore, in the loopback mode(s), the multicast address detection feature of the mac en- gine, programmed by the contents of the logical ad- dress filter (ladrf [63:0] in csrs 8C11) can only be tested when dxmtfcs=1, allocating the fcs genera- tor to the receiver. all other features operate identically in loopback as in normal operation, such as automatic transmit padding and receive pad stripping. when performing an internal loopback, no frame will be transmitted to the network. however, when the AM79C974 controller is configured for internal loopback the receiver will not be able to detect network traffic. ex- ternal loopback tests will transmit frames onto the network if the aui port is selected, and the AM79C974 controller will receive network traffic while configured for external loopback when the aui port is selected. runt packet accept is automatically enabled when any loop- back mode is invoked. loopback mode can be performed with any frame size. runt packet accept is internally enabled (rpa bit in csr124 is not affected) when any loopback mode is in- voked. this is to be backwards compatible to the lance (am7990) software. when the 10base-t mau is selected in external loop- back mode, the collision detection is disabled. this is necessary, because a collision in a 10base-t system is defined as activity on the transmitter outputs and re- ceiver inputs at the same time, which is exactly what oc- curs during external loopback. since a 10base-t hub does not normally feed the sta- tions transmitter outputs back into the stations receiver inputs, the use of external loopback in a 10base-t sys- tem usually requires some sort of external hardware that connects the outputs of the 10base-t mau to its inputs. led support the AM79C974 controller can support up to 3 leds. led outputs lnkst and led1 allow for direct connec- tion of an led and its supporting pullup device. led out- put led3 may require an additional buffer between the AM79C974 controller output pin and the led and its supporting pullup device. because the led3 output is multiplexed with other AM79C974 controller functions, it may not always be possible to connect an led circuit directly to the led3 pin. for example, when an led circuit is directly con- nected to the eedo/ led3 pin, then it is not possible for most serial eeprom devices to sink enough i ol to maintain a valid low level on the eedo input to the AM79C974 controller. therefore, in applications that re- quire both an eeprom and a third led, then it is neces- sary to buffer the led3 circuit from the eeprom-pcnet-scsi connection. the led registers in the bcr resource space allow each led output to be programmed for either active high or active low opera- tion, so that both inverting and non-inverting buffering choices are possible. in applications where an eeprom is not needed, the led3 pin may be directly connected to an led circuit. the AM79C974 led3 pin driver will be able to sink enough current to properly drive the led circuit. by default, after h_reset, the 3 led outputs are con- figured in the following manner: amd p r e l i m i n a r y 74 AM79C974 led default default default output interpretation drive enable output polarity lnkst link status enabled active low led1 receive enabled active low led3 transmit enabled active low for each led register, each of the status signals is anded with its enable signal, and these signals are all ored together to form a combined status signal. each led pins combined status signal runs to a pulse stretcher, which consists of a 3-bit shift register clocked at 38 hz (26 ms). the data input of each shift register is normally at logic 0. the or gate output for each led register asynchronously sets all three bits of its shift reg- ister when the output becomes asserted. the inverted output of each shift register is used to control an led pin. thus the pulse stretcher provides 2C3 clocks of stretched led output, or 52 ms to 78 ms. rcv rcv e xmt xmt e lnk lnk e rxpol rxpol e jab jab e col col e rcvm rcvm e to pulse stretcher 18681a-28 figure 24 . led control logic the diagram above shows the led signal circuit that ex- ists for each led pin within the AM79C974 controller. h_reset, s_reset, and stop there are three different types of reset operations that may be performed on the AM79C974 device, h_reset, s_reset and stop. these names have been used throughout the document. the following is a description of each type of reset operation: h_reset h_reset= hardware_reset is a AM79C974 re- set operation that has been created by the proper as- sertion of the rst pin of the AM79C974 device. when the minimum pulse width timing as specified in the rst pin description has been satisfied, then an internal re- set operation will be performed. h_reset will reset all of or some portions of csr0, 3, 4, 15, 58, 80, 82, 100, 112, 114, 122, 124 and 126 to default values. h_reset will reset all of or some por- tions of bcr 2, 4, 5, 6, 7, 18, 19, 20, 21 to default values. h_reset will reset the command register in the pci configuration space. h_reset will cause the microcode program to jump to its reset state. follow- ing the end of the h_reset operation, the AM79C974 controller will attempt to read the eeprom device through the eeprom microwire interface. h_reset resets the t-mau into the link fail state. s_reset s_reset = software_reset is an ethernet con- troller reset operation that has been created by a read access to the reset register which is located at off- set 14hex from the AM79C974 i/o base address. s_reset will reset all of or some portions of csr0, 3, 4, 15, 80, 100 and 124 to default values. s_reset will not affect any of the bcr and pci configuration space locations. s_reset will cause the microcode program to jump to its reset state. following the end of the s_reset operation, the AM79C974 controller will not attempt to read the eeprom device. s_reset sets the t-mau into the link fail state. note that s_reset will not cause a deassertion of the req signal, if it happens to be active at the time of the read to the reset register. the req signal will remain active until the gnt signal is asserted. following the read of the reset register, on the next clock cycle after the gnt signal is asserted, the AM79C974 controller will deassert the req signal. no bus master accesses will have been performed during this brief bus owner- ship period. stop stop is an ethernet controller reset operation that has been created by the assertion of the stop bit in csr0. that is, a stop reset is generated by writing a one to the stop bit of csr0 when the stop bit cur- rently has a value of zero. if the stop bit value is cur- rently a one and a one is rewritten to the stop bit, then no stop reset will be generated. stop will reset all or some portions of csr0, 3, and 4 to default values. stop will not affect any of the bcr and pci configuration space locations. stop will cause the microcode program to jump to its reset state. fol- lowing the end of the stop operation, the AM79C974 controller will not attempt to read the eeprom device. for the identity of individual csrs and bit locations that are affected by stop, see the individual csr register descriptions. setting the stop bit does not affect the t-mau. p r e l i m i n a r y amd 75 AM79C974 scsi controller the primary function of the pcnet-scsi controller is to transfer data between the 4 byte-wide pci bus and 1 byte-wide scsi bus. the controller consists of two blocks: scsi and dma. the scsi block sits between the scsi bus and the dma block. it controls data flow to/from scsi bus. the dma block is located between the scsi block and the pci bus interface unit. it handles data flow to/from pci bus. the operation of each block is governed by a set of con- trol registers: 1. channel context block (ccb) registers control the dma block 2. scsi registers control the scsi block in a normal operation, both sets of registers must be pro- grammed with the specifics of the transfer, such as start- ing address, transfer count, etc. (for more information, refer to technical manual pid #18738a). scsi specific dma engine the scsi specific dma engine in the AM79C974 pro- vides bus-mastering capabilities to allow flexibility and performance advantages over slave pci-scsi devices. built into the engine is a 96-byte (24 dword) fifo and additional logic to handle the transition between the 32-bit pci bus and the 8-bit scsi bus. figure 25 illustrates the dma engine in relation to the pci interface and the scsi block. as its most basic func- tion, the dma engine acts as the dma controller in a bus master capacity on the pci bus, transferring data be- tween memory and the scsi block. all command, data, status, and message bytes pass through the dma fifo on their way to or from the scsi bus. however, for pro- grammed i/o (pio) accesses to the scsi registers, the dma fifo is bypassed as data moves directly from the scsi block to the pci interface. since pio operations do not pass through the funneling logic and dma fifo, data is transferred one byte at a time from the scsi block to the pci interface via the least significant byte lane. (the three most significant byte lanes will contain null data.) 18681a-29 pci bus interface unit funnel/alignment logic scsi fifo (16x9) dreq dack scsi block dma cntl full empty data path unit pci^gnt ad(3:0) cs rd wr ad(4:0), cs wr , rd , 32 16 32 data data data pci config space ad (4:0) c/ be (3:0) pci^req dma reg scsi reg dma engine 8 data dma fifo (24x32) figure 25. pci biu C dma engine C scsi block since the pci bus is 4 bytes wide and the scsi bus is only 1 byte wide, funneling logic is included in this en- gine to handle byte alignment and to ensure that data is properly transferred between the scsi bus and the amd p r e l i m i n a r y 76 AM79C974 wider pci bus. all boundary conditions are handled through hardware by the dma engine. the dma engine is also designed for block type (4 kbyte page) transfers to support scatter-gather opera- tions. implementation of this feature is described further in the dma scatter-gather mechanism section. dma fifo data transfers from the scsi fifo to the dma fifo take place each time the threshold of two bytes is reached on the scsi side. the transfer is initiated by the scsi block when the internal dreq is asserted, and continues with the dack handshaking which typically takes place in dma accesses. data is accumulated in the dma fifo until a threshold of 16 dword (64 bytes) is reached. data is then burst across the pci bus to memory. residue data which is less than the threshold in each fifo is sent in non-contiguous bursts. for mem- ory read operations, data is sent in burst mode to the dma fifo and continues through to the scsi fifo and onto the scsi bus. dma blast command this command is used to retrieve the contents of the dma fifo when the target disconnects during a dma write operation. this could happen for example if a scsi disk drive detected the end of a sector and de- cided to give up the bus while it was looking for the next sector. the target disconnect can leave some bytes of data in the dma fifo and some in the scsi fifo, while some bytes have yet to be transferred from the periph- eral device. when this happens, the controller will assert inta to interrupt the processor, the scsi state machine will continue to empty its contents into the dma fifo, but the dma fifo will not necessarily dump its contents into memory (unless the 64-byte dma threshold hap- pens to have been exceeded at this time). the blast command causes the contents of the dma fifo to be emptied into memory. there are some re- strictions on when this command should be used. first, the command should be used only to recover from an interrupted dma write operationnot a read operation. second, the command must not be issued until the scsi fifo has finished dumping its contents into the dma fifo. third, the command should never be issued when the dma fifo has already been emptied. this is indicated by the state of the done bit in the dma status register at ((b)+54h). (this is a test for a special case that can occur when a target disconnect leaves only 1 byte left to be trans- ferred from the scsi peripheral. in this case, if the origi- nal transfer count was even, an odd number of bytes will be left in the scsi fifo. since the scsi engine trans- fers data to the dma fifo two bytes at a time, the last transfer consists of one byte of valid data and one byte of garbage. the dma engine treats this final invalid byte as valid data and writes it to memory. when it does this, it decrements its working byte counter to zero and sets the done bit, even though 1 byte still needs to be re- trieved from the peripheral device.) the following procedure outlines the use of the blast command after an interrupt has occurred. note that the order of steps 2C4 is not critical. the order can be changed to tune the performance. also note that each register is read only once in this procedure even though several tests may be made on data from one register. 1. verify that int (bit 7 of the scsi status register at ((b)+10h) is set to indicate that a scsi interrupt is pending. 2. read the scsi current fifo count (bits 4:0 of the current fifo/internal state register at ((b)+1ch). if this value is not zero, wait for the scsi fifo to empty its contents into the dma fifo. 3. if bit 4 of the scsi status register (ctz) is set, stop here. the transfer is complete, and it is not necessary to execute the blast command. 4. verify that dir (bit 7 of dma command register at ((b)+40h) is set to one to indicate that the direction of transfer is from scsi to memory. 5. test the error bits in the dma status register and the scsi status register (statreg at (b)+10h) to verify that the contents of the dma and scsi fifos are not invalid. 6. test the dma done bit in the dma status register. if done is not set, write 01 to the dma command register to issue the blast command. this will move the remaining data from the dma fifo into memory. 7. wait until the blast complete (bcmplt) in the dma status register is set to indicate the com- pletion of the blast operation. 8. write 00 to the dma command register to issue the idle command to the dma engine. (note that the idle command does not generate an interrupt.) the above procedure insures that the data that has been transferred from the scsi peripheral does not get lost in the dma fifo when a target disconnect occurs. however, it does not complete the original transfer. the software must now read the scsi current transfer count register (ctcreg) to find out how many bytes have yet to be transferred from the scsi peripheral de- vice and must start a new transfer operation to get the rest of the data. (ctcreg consists of three bytes lo- cated at ((b)+00h, (b)+04h, and (b)+38h.) funneling logic figure 26 shows the internal dma logic interface with the scsi block. the dma fifo interfaces to the funnel p r e l i m i n a r y amd 77 AM79C974 logic block via a 32-bit data bus, and the funnel logic properly reduces this stream of data to a 16-bit stream to properly interface with the scsi fifo. 96-byte dma fifo funnel logic 16-byte scsi fifo 32 16 18681a-30 figure 26. dma fifo to scsi fifo interface scsi dma programming sequence the following section outlines the procedure for execut- ing scsi dma operations: 1. issue idle command to the dma engine 2. configure the scsi block registers (e.g. synchro- nous operation, offset values, etc.) 3. program the dma registers to set up address and transfer count 4. issue a transfer command to the scsi command registers 5. issue the start command to the dma engine 6. at the end of the dma transaction, issue the idle command to the dma engine memory descriptor list (mdl) based dma programming the following section outlines the use of the mdl for scatter-gather dma operations: 1. set up the mdl list 2. use the programming sequence defined earlier for initiating a scsi dma transfer dma registers the following is a summary of the dma register set or the dma channel context block (dma ccb). these registers control the specifics for dma operations such as transfer length and scatter-gather options. the three read-only working counter registers allow the system software and driver to monitor the dma transaction. each register address is represented by the pci con- figuration base address (b) and its corresponding offset value. the base address for the scsi controller is stored at register address (10h) in the scsi pci configu- ration space. table 6. the dma registers register acronym addr (hex) register description type cmd (b)+40 command (bits 31:8 reserved, bits 7:0 used) r/w stc (b)+44 starting transfer count (bits 31:24 reserved, bits 23:0 used) r/w spa (b)+48 starting physical address (bits 31:0 used) r/w wbc (b)+4c working byte counter r wac (b)+50 working address counter (bits 31:0 used) r status (b)+54 status register (bits 31:8 reserved, bits 7:0 used) r smdla (b)+58 starting memory descriptor list (mdl) address r/w wmac (b)+5c working mdl counter r amd p r e l i m i n a r y 78 AM79C974 command register (cmd) the upper 3 bytes of command register are reserved, the remaining (lsb) byte is defined as follows: address (b)+40h, lsb read/write dir inte_d inte_p mdl reserved reserved cmd1 cmd0 76543210 dir: data transfer direction bit. when this bit is set, the direc- tion of transfer is from scsi to memory. inte_d: dma transfer active interrupt enable bit. when this bit is set, the error or done condition will cause inta to be asserted. inte_p: page transfer active interrupt bit. mdl: memory descriptor list (mdl) spa enable bit. reserved: reserved for future expansion. the zero value must be written in these bits. cmd1-0: these two bits are encoded to represent four com- mands: idle, blast, start, and abort. cmd1 cmd0 command description 0 0 idle resets the dma block to the idle state. stops any current transfer. does not affect status bits or cause an interrupt. 0 1 blast empties all data bytes in dma fifo to memory during a dma write operation. upon completion, the bcmplt bit will be set in the dma status register. this command should not be used during a dma read operation. 1 0 abort terminates the current dma transfer. restores the dma engine to the idle state. sets the abort bit (bit 2) in the status register. note: this is only valid after a start command is issued. 1 1 start initiates a new dma transfer. these bits must remain set throughout the dma operation until the done bit in the dma status register is set. note: this command should be issued only after all other control bits have been initialized. dma scatter-gather mechanism the AM79C974 controller contains a scatter-gather translation mechanism which facilitates faster data transfers. this feature uses a m emory d escriptor l ist which is stored in system memory. use of the memory descriptor list allows a single scsi transfer to be read from (or written to) non-contiguous physical memory lo- cations. this mechanism avoids copying the transfer data and mdl list, which was previously required for conventional dma operations. memory descriptor list (mdl) the mdl is a non-terminated (no end of file marker) list of 32-bit page frame addresses, which is always aligned on a double word boundary. the format is shown below: 31 12 11 0 page frame address ignored dma scatter-gather operation (4k aligned elements) the scatter-gather mechanism described below assumes 4k page alignment and size for all mdl entries except the first and last entry. this feature is enabled by setting the mdl bit in the dma command register (bit 4, address (b)+40h). 1. a) prepare the memory descriptor list (mdl) through software and store it in system memory. b) load the address of the starting entry in the memory descriptor list (mdl) into the start memory descriptor list address (smdla) register. this value is automatically copied into the working mdl address counter (wmac). c) program the starting transfer count (stc) register with the total transfer length (i.e., # of bytes). also program the starting physical address (spa) register (bits 11:0) with the starting offset of the first entry. note : the value in the smdla register must be double word aligned. therefore, read/write transactions will al- ways begin on a double word boundary. p r e l i m i n a r y amd 79 AM79C974 18681a/1-31 0 31 0 31 smdla wmac 0 12 31 mdl page frame address #1 page frame address #2 page frame address #3 page frame address #4 ignored page frame address #n ignored ignored ignored ignored in this example, the contents of the wmac register is pointing to page frame address #1. when the first entry in the mdl is read (page frame address #1), the wmac register is incremented to point to the next page entry (page frame address #2). 2. issue the start dma command. the scsi control- ler reads the page frame address (bits 31:12) from the mdl entry and combines it with the first page offset value in the starting physical address (spa) register (bits 11:0). this 32-bit value is loaded into the working address counter (wac) register and becomes the physical address for page#1, as shown below. the wmac is then incremented to point to the next entry in the mdl. 0 12 31 wac 4k page #1 page frame address #1 starting offset from the mdl 0 12 31 spa xxxx starting offset programmed by the software data 18681a/1-32 3. when the wac register (bits 11:0) reaches the first 4k byte boundary, the scsi controller reads the second mdl entry and combines the page frame address (bits 31:12) from this entry of the mdl with bits 11:00 of the wac register. this be- comes the physical address for page#2. since the wac register (bits 11:0) has rolled over to 00h, the wac now points to the beginning of the page frame address #2 as shown below. the wmac is then incremented to point to the next entry in the mdl. 18681a/1-33 0 12 31 wac 4k page #2 0 page frame address #2 from the mdl data amd p r e l i m i n a r y 80 AM79C974 when wac (bits 11:0) again reaches the next 4k byte boundary, the next mdl entry is read into the wac. the operation continues in this way until wmac register reaches the last mdl entry (page frame address #n in this example). 4. the wac register points to the beginning of the last page and the dma operation continues until the byte count is exhausted in the working byte counter (wbc) register. when wbc=0, the chip stops incrementing the wac register. this is shown below. 18681a/1-34 0 12 31 wac 4k page #n page frame address #n from the mdl wbc = 0 0 data dma scatter-gather operation (non-4k aligned elements mdl not set) there is another way to implement a scatter-gather op- eration which does not force the data elements to be aligned on 4k boundaries. it assumes a traditional scatter-gather list of the following format: element 0 physical address byte count element 1 physical address byte count ... element n physical address byte count this second implementation is described as follows: 1. set the scsi start transfer count register ((b)+00h, (b)+04h, (b)+38h) to the byte count of the first scatter-gather element. 2. program the dma starting transfer count regis- ter ((b)+44h) to the byte count of the first scatter- gather element. 3. program the dma starting physical address reg- ister ((b)+48h) to the physical address of the first scatter-gather element. 4. start the scsi operation by issuing a scsi infor- mation transfer command. 5. start the dma engine with dma transfer interrupt enable (bit 6, (b)+40h). 6. when the scatter-gather elements byte count is exhausted, the dma engine will generate an interrupt. 7. reprogram the next scatter-gather elements byte count into the scsi start transfer count register and the dma starting transfer count register. 8. reprogram the dma starting physical address register ((b)+48h) to the physical address of the next scatter-gather element. 9. repeat steps 4C8 until the scatter-gather list is completed. interrupts interrupts may come from two sources: the dma engine or the scsi block. upon receipt of an interrupt (inta as- serted), the dma status register should be serviced first to identify the interrupt source(s). dma engine related interrupts are cleared when the related flags are read in the dma status register. however, scsi block inter- rupts will be cleared only when the scsi status, internal state, and interrupt status registers are read. interrupts are caused by: successful completion of a dma transfer request. (bit 6 in the dma command register ((b+40h) must be set to enable this interrupt) an address error occurred on the pci bus during a dma transfer (bit 6 in the dma command regis- ter ((b)+40h) must be set to enable this interrupt) the pwdn pin is first asserted after completion of each page transfer during mdl operations. (bit 5 in the dma command register ((b)+40h) must be set to enable this interrupt) an interrupt from the scsi block will automatically set bit 4 (scsiint) in the dma status register (address (b)+54h). the scsi block will generate an interrupt un- der the following conditions: scsi reset occurred p r e l i m i n a r y amd 81 AM79C974 illegal command code issued the target disconnects from the scsi bus scsi bus service request successful completion of a command the AM79C974 has been reselected the fast scsi block the functionality of the scsi block is described in the following section. topics to be covered are: scsi block id scsi fifo threshold data transmission req / ack control parity reset levels scsi block id the AM79C974 contains a scsi block id code which is stored in the msb of the current transfer count regis- ter. the code reflects the chips revision level and family code. this 8-bit code may be read when the following conditions are true. after power up or a chip reset has occurred before the current transfer counter ((b)+38h) is loaded the part-unique id code in register ((b)+38h) will read as follows: bit 7 bit 6 bit 5 bit 4 bit 3 bit 2 bit 1 bit 0 00010010 scsi fifo threshold the threshold value for the scsi fifo is two bytes (one word). when this threshold is reached, the scsi block will indicate to the dma engine that it is capable of re- ceiving or sending data bytes. data transmission data transmission rates will vary from system to system, depending on the number of devices configured on the scsi bus, as well as the transfer rates that each individ- ual device is capable of sustaining. transfer rates for the AM79C974 are controlled by the fastscsi and fastclk bits in control register three, as well by the extended timing feature in con- trol register one. to achieve 10 mbyte/s transmission rates, the following conditions must be true: a 40 mhz clock (50% duty cycle) must be supplied to scsiclk. the target must be able to sustain fast scsi timings. bits 3 (fastclk) and 4 (fastscsi) in control register three must be set to 1. the lower three bits of register ((b)+24h), the clock conversion factor register must be programmed to 000. the lower 5 bits of the synchronous transfer period register ((b)+18h) must be set to a value of 04h. the fastclk and fastscsi bits in control register three modify the scsi state machine to produce both fast and normal scsi timings. synchronous data transmission rates are dependent on the input clock fre- quency selected, as well as the transfer period. bits 4:0 in the synchronous transfer period register ((b)+18h) specify the period for synchronous data trans- fers. for programming information, refer to the techni- cal manual. req / ack control the assertion and deassertion time for the req and ack signals may be controlled via the synchronous off- set register ((b)+1ch). bits 7:6 control req / ack deassertion delay with respect to the time that data is valid, while bits 5:4 control req / ack assertion delay. the deassertion for req / ack may be moved ahead .5 clock cycles, or it may be delayed for up to 1.5 clock cy- cles. deassertion delay options depend on the status of the fastclk bit in control register three. assertion delay for req / ack can vary from 0 to 1.5 clock cycles. parity parity on the scsi bus is such that the total number of logical ones on data bus including the parity bit must be odd. parity checking features are implemented via two bits in the status register and control register one. parity checking can be implemented on data flowing in from the scsi bus. parity is always generated internally by the AM79C974 for data moving onto the scsi bus. feature bit name bit # register parity from parity error 4 control scsi reporting reg one ((b)+20h) parity status parity error 5 status register ((b)+10h) parity checking on the scsi bus the parity error reporting bit (bit 4, control register one) enables parity checking on all incoming bytes from the scsi bus. this feature is cleared to 0 by a hardware reset. amd p r e l i m i n a r y 82 AM79C974 when this feature is enabled, the AM79C974 will check parity on all data received from the scsi bus. any de- tected error will be flagged by setting bit 5 in the scsi status register, and atn will be asserted on the scsi bus. however, no interrupt will be generated. when this feature is disabled (bit 4 set to 0), no parity check is done on incoming bytes. note that the parity on the pci bus is generated internally and is distinct from the parity received from the scsi bus. parity generating on the scsi bus for each byte transferred to the scsi bus, parity gen- eration is done automatically. reset levels the am79c794 has two reset pins and two reset com- mands that affect the scsi block. the rst pin resets the whole chip including the scsi controller, the ether- net controller, and the pci interface. the reset device command causes almost the same effect on the scsi controller that the rst pin does. however, the reset device command has no effect on the ethernet controller or the pci interface. also, after the reset device command has been issued, the user must issue a nop command before another command can be executed. the action of the rst signal or the reset device com- mand is called hard reset. the scsi^rst pin is a bidirectional signal on the scsi bus that resets a portion of the scsi logic when it is as- serted by a device on the scsi bus. similarly the am79c794 can assert the scsi^rst signal to cause all of the other devices on the scsi bus to reset. the reset scsi command causes the same effect on the scsi controller that the scsi^rst pin does, except that this command also causes the scsi^rst pin to be asserted so that all other (external) devices on the scsi bus are also reset. once a scsi reset command has been executed, the scsi^rst signal will remain as- serted until a hard reset has occurred. the action of the scsi^rst signal and the reset scsi command is called soft reset. in addition there is a third type of reset, called discon- nected reset, that is caused by certain sequences on the scsi bus. these three types of reset are described in the following sections. hard resets: (h) this reset occurs at power up, when the rst pin is as- serted through external hardware, or when the reset device command is issued by writing 02h to the scsi command register at ((b)+0ch). hard reset causes all chip functions to halt and resets all internal state ma- chines. it leaves the scsi block in the disconnected state. it leaves all scsi registers in their default states. in addition, if the hard reset is caused by the assertion of the rst pin, the following actions occur: the command register in the pci configuration space is cleared to zero. (no other register in the configuration space is affected.) the dma ccb registers are set to their default values. soft reset: (s) this reset occurs either when the scsi^rst pin on the scsi bus is asserted or when the reset scsi bus com- mand is issued (by writing 03h to the scsi command register at ((b) = 0ch)). soft reset causes the following actions to occur: all scsi bus signals except scsi^rst are released. the chip is returned to the disconnected state. an interrupt is generated if bit 6 in control register one is enabled. disconnected reset: (d) disconnected reset occurs when either of the following conditions occur: the AM79C974 is the initiator and the scsi bus moves to a bus free state the selection command terminates due to selection time-out disconnected reset causes the following actions: all scsi signals except scsi^rst are deasserted. the scsi command register is initialized to empty. the is1 and is0 bits in the internal state register, ((b)+18h), are cleared to 0. please refer to the technical manual (pid #18738a) for the default values for all registers. device commands the device commands can be broadly divided into two categories, dma commands and non-dma commands. dma commands are those which cause data movement between the host memory and the scsi bus while non- dma commands are those that cause data movement between the scsi fifo and the scsi bus. the most significant bit of the command byte differentiates the dma from the non-dma commands. when a dma command is issued, the contents of the start transfer count register will be loaded into the current transfer count register. data transmission will p r e l i m i n a r y amd 83 AM79C974 continue until the current transfer count register dec- rements to zero. non-dma commands do not modify the current trans- fer count register and are unaffected by the value in the current transfer count register. for non-dma com- mands, the number of bytes transmitted depends solely on the operation in progress. when non-dma commands are used, the host com- puter must use programmed i/o to transfer the data be- tween the scsi fifo and the host memory. table 7. summary of commands initiator commands information transfer 10 90 initiator command complete steps 11 91 message accepted 12 C transfer pad bytes C 98 set atn *1aC reset atn *1bC idle state commands select without atn steps 41 c1 select with atn steps 42 c2 select with atn and stop steps 43 c3 enable selection/reselection* 44 c4 disable selection/reselection 45 C select with atn 3 steps 46 c6 general commands no operation* 00 80 clear fifo* 01 C reset device* 02 C reset scsi bus** 03 C non- dma mode dma mode code (hex.) command notes: * these commands do not generate interrupt. ** an interrupt is generated when scsi bus reset interrupt reporting is not disabled (see control register1/disr bit6). command stacking the microprocessor may stack commands in the com- mand register ((b)+0ch) since it functions as a two- byte deep fifo. non-dma commands may not be stacked, and commands which transfer data in oppos- ing directions should not be stacked together; other- wise, the results are unpredictable. if dma commands are queued together, the start transfer count must be written before the associated command is loaded into the command register. since multiple interrupts can occur when commands are stacked, it is recommended that the enf bit in control register two (bit 6) be set in order to latch the scsi phase bits in the scsi status register ((b)+10h) at the completion of a command. this allows the host to deter- mine the phase of the interrupting command without having to consider phase changes that occurred after the stacked command began execution. note : command stacking should only be used during scsi data in or data out transfers. invalid commands when an illegal command is written to the AM79C974, the invalid command bit (bit 6, register (b)+14h) will be set to 1, and an interrupt will be generated to the host. when this happens, the interrupt must be serviced be- fore another command may be written to the command register. an invalid command is defined as a command written to the AM79C974 that is either not supported, not allowed in the specified mode, or a command that has an unsup- ported command mode. the following conditions will also cause an invalid com- mand interrupt to occur: an initiator information transfer, transfer pad, or command complete is issued when ack is still asserted. a selection command is issued with the dma bit enabled, if the selection command was previously issued with the dma enabled. command window the window at the point where the disable selection/ reselection command (45h/c5h) has been loaded into the command register ((b)+0ch), and before bus-initi- ated selection begins, has been eliminated. this pre- vents a false successful operation interrupt from being generated when the selection sequence continues to completion after the disable command has been loaded. initiator commands initiator commands are executed by the device when it is in the initiator mode. if the device is not in the initiator mode and an initiator command is received the device will ignore the command, generate an invalid command interrupt and clear the command register. should the target disconnect from the scsi bus by deasserting the bsy signal line while the AM79C974 (initiator) is waiting for the target to assert req , a dis- connected interrupt will be issued 1.5 to 3.5 clock cycles following bsy going false. upon receipt of the last byte during message in phase, ack will remain asserted to prevent the target from is- suing any additional bytes, while the initiator decides to amd p r e l i m i n a r y 84 AM79C974 accept/reject the message. if non-dma commands are used, the last byte signals the scsi fifo is empty. if dma commands are used, the current transfer count signals the last byte. a reset scsi bus command (03h/83h) will force the AM79C974 to abort the current operation and discon- nect from the bus. if the disr bit is reset (bit 6, control register one (b)+20h)), the host processor will be inter- rupted with a scsi reset interrupt before the AM79C974 proceeds to disconnect. if parity checking is enabled in the initiator mode during the data-in phase and an error is detected, atn will be asserted for the erroneous byte before deasserting ack . information transfer command (command code 10h/90h) the information transfer command is used to transfer information bytes over the scsi bus. this command may be issued during any scsi information transfer phase. synchronous data transmission requires use of the dma mode. the device will continue to transfer information until it is terminated by any one of the following conditions: the target changes the scsi bus phase before the expected number of bytes are transferred. the AM79C974 clears the command register (cmdreg), and generates a service request interrupt when the target asserts req . transfer has successfully completed. if the phase is message out, the AM79C974 deasserts atn before asserting ack for the last byte of the mes- sage. when the target asserts req , a service request interrupt is generated. in the message in phase when the device receives the last byte. the AM79C974 keeps the ack sig- nal asserted and generates a successful opera- tion interrupt. during synchronous data transfers the target may send up to the maximum synchronous offset number of req pulses to the initiator. if it is the synchronous data- in phase then the target sends the data and the req pulses. these bytes are stored by the initiator in the fifo as they are received. the information transfer command, when issued dur- ing the following scsi phases and terminated in syn- chronous data phases, is handled as described below: message in/status phase C when a phase change to synchronous data-in or synchronous data-out is detected by the device, the command register is cleared and the dma interface is disabled to prevent any transfer of data (phase) bytes. if the phase change is to synchronous data-in and bad parity is detected on the data bytes coming in, it is not reported since the status register will re- port the status of the command just completed. the parity error flag and the atn signal will be asserted when the next information transfer com- mand begins execution. message out/command phase C when a phase change to synchronous data-in or synchronous data-out is detected by the device, the command register is cleared and the dma interface is dis- abled to prevent any transfer of data (phase) bytes. if the phase change is to synchronous data-in and bad parity is detected on the data bytes coming in, it is not reported since the status register will re- port the status of the command just completed. the parity error flag and the atn signal will be as- serted when the next information transfer com- mand begins execution. the scsi fifo register will be latched and will remain in that condition until the next command begins execution. the value in the scsi fifo register indicates the number of non-data bytes in the scsi fifo when the phase changed to syn- chronous data-in. these bytes are cleared from the fifo, and only incoming data bytes are re- tained. in the synchronous data-out phase, the threshold counter is incremented as req pulses are re- ceived. the transfer is completed when the fifo is empty and the current transfer count register is 0. the threshold counter will not be 0. in the synchronous data-in phase, the current transfer count register is decremented as bytes are read from the scsi fifo rather than when the bytes are being written to the scsi fifo. the transfer is completed when current transfer count register is 0. however, the scsi fifo may not be empty. initiator command complete steps (command code 11h/91h) the initiator command complete steps command is normally issued when the scsi bus is in the status in phase. one status byte followed by one message byte is transferred if this command completes normally. after receiving the message byte the device will keep the ack signal asserted to allow the initiator to examine the message and assert the atn signal if it is unacceptable. the command terminates early if the target does not p r e l i m i n a r y amd 85 AM79C974 switch to the message in phase or if the target discon- nects from the scsi bus. this command does not utilize the internal state register ((b)+18h). message accepted command (command code 12h) the message accepted command is used to release the ack signal. this command is normally used to com- plete a message in handshake. upon execution of this command the device generates a service request in- terrupt after req is asserted by the target. after the device has received the last byte of message, it keeps the ack signal asserted. this allows the device to either accept or reject the message. to accept the message, message accepted command is issued. to reject the message the atn signal must be asserted (with the help of the set atn command) before issuing the message accepted command. in either case, the message accepted command has to be issued to re- lease the ack signal. transfer pad bytes command (command code 18h/98h) the transfer pad bytes command is used to recover from an error condition. this command is similar to the information transfer command, only the information bytes consists of null data. it is used when the target ex- pects more data bytes than the initiator has to send. it is also used when the initiator receives more information than expected from the target. when sending data to the scsi bus, the scsi fifo is loaded with null bytes which are sent out to the scsi bus. although an actual dma request is not made, dma interface must be enabled when pad bytes are transmit- ted since the AM79C974 uses the current transfer count register to terminate transmission. this command terminates under the same conditions as the information transfer command, but the device does not keep the ack signal asserted during the last byte of the message in phase. should this command terminate prematurely due to a disconnect or a phase change be- fore the current transfer count register decrements to zero, the scsi fifo may contain residual pad bytes. set atn command (command code 1ah) the set atn command is used to drive the atn signal active on the scsi bus. an interrupt is not generated at the end of this command. the atn signal is deasserted before asserting the ack signal during the last byte of the message out phase. note: the atn signal is asserted by the device without this command in the following cases: if any select with atn command is issued and the arbitration is won. an initiator needs the targets attention to send a message. the atn signal is asserted before deasserting the ack signal. reset atn command (command code 1bh) the reset atn command is used to deassert the atn signal on the scsi bus. an interrupt is not generated at the end of this command. this command is used only when interfacing with devices that do not support the common command set (ccs). these older devices do not deassert their atn signal automatically on the last byte of the message out phase. this device does deas- sert its atn signal automatically on the last byte of the message out phase. idle state commands the idle state commands can be issued to the device only when the device is disconnected from the scsi bus. if these commands are issued to the device when it is logically connected to the scsi bus, the commands are ignored, an invalid command interrupt is generated, and the command register (cmdreg) is cleared. select without atn steps command (command code 41h/c1h) the select without atn steps command is used by the initiator to select a target. when this command is is- sued, the device arbitrates for the control of the scsi bus. when the device wins arbitration, it selects the tar- get device and transfers the command descriptor block (cdb). before issuing this command the scsi timeout register (stimreg), control register one (cntlreg1), and the scsi destination id register (sdidreg) must be set to the proper values. if dma is enabled, the start transfer count register (stcreg) must be set to the total length of the command. if dma is not enabled, the data must be loaded into the fifo be- fore issuing this command. this command will be termi- nated early if the scsi timeout register times out, if the target does not go to the command phase following the selection phase, or if the target exits the command phase prematurely. a successful operation interrupt will be generated following normal command execution. select with atn steps command (command code 42h/c2h) the select with atn steps command is used by the in- itiator to select a target. when this command is issued, the device arbitrates for the control of the scsi bus. when the device wins arbitration, it selects the target device with the atn signal asserted and transfers the command descriptor block (cdb) and a one byte mes- sage. before issuing this command the scsi timeout register (stimreg), control register one (cntlreg1) and the scsi destination id register (sdidreg) must be set to the proper values. if dma is enabled, the start transfer count register (stcreg) must be set to the total length of the command and mes- sage. if dma is not enabled, the data must be loaded amd p r e l i m i n a r y 86 AM79C974 into the fifo before issuing this command. this com- mand will be terminated early in the following situations: the scsi timeout register times out the target does not go to the message out phase following the selection phase the target exits the message phase early the target does not go to the command phase following the message out phase the target exits the command phase early a successful operation/service request interrupt is generated when this command is completed successfully. select with atn and stop steps command (command code 43h/c3h) the select with atn and stop steps command is used by the initiator to send messages with lengths other than 1 or 3 bytes. when this command is issued, the device executes the selection process, transfers the first mes- sage byte, then stops the sequence. atn is not de- asserted at this time, allowing the initiator to send additional message bytes after the id message. to send these additional bytes, the initiator must write the transfer counter with the number of bytes which will fol- low, then issue a transfer information command. (note: the target is still in the message out phase when this command is issued). atn will remain asserted until the current transfer count register decrements to zero. the scsi timeout register (stimreg), control regis- ter one (cntlreg1), and the scsi destination id register (sdidreg) must be set to the proper values before the initiator issues this command. this command will be terminated early if the stimreg times out or if the target does not go to the message out phase fol- lowing the selection phase. enable selection/reselection command (command code 44h/c4h) the enable selection/reselection command is used to respond to a bus-initiated selection or reselection. upon disconnecting from the bus the selection/ reselection circuit is automatically disabled by the de- vice. this circuit must be enabled for the AM79C974 to respond to subsequent reselection attempts and the en- able selection/reselection command is issued to do that. this command is normally issued within 250 ms (select/reselect timeout) after the device disconnects from the bus. if dma is enabled, the device loads the received data to the buffer memory. if the dma is dis- abled, the received data stays in the fifo. disable selection/reselection command (command code 45h) the disable selection/reselection command is used by the target to disable response to a bus-initiated reselection. when this command is issued before a bus-initiated selection or reselection is in progress, it resets the internal state bits previously set by the enable selection/reselection command. the device also gen- erates a successful operation interrupt to the proces- sor. if however, this command is issued after a bus-initiated selection/reselection has begun, this command and all incoming commands are ignored since the command register (cmdreg) is held reset. the AM79C974 also generates a selected or reselected interrupt when the sequence is complete. select with atn 3 steps command (command code 46h/c6h) the select with atn 3 steps command is used by the initiator to select a target. this command is similar to the select with atn steps command, except that it sends exactly three message bytes. when this com- mand is issued the AM79C974 arbitrates for control of the scsi bus. when the device wins arbitration, it se- lects the target device with the atn signal asserted and transfers the command descriptor block (cdb) and three message bytes. before issuing this command the scsi timeout register (stimreg), control register one (cntlreg1), and the scsi destination id regis- ter (sdidreg) must be set to the proper values. if dma is enabled, the start transfer count register (stcreg) must be set to the total length of the com- mand. if dma is not enabled, the data must be loaded into the fifo before issuing this command. this com- mand will be terminated early in the following situations: the scsi timeout register times out the target does not go to the message out phase following the selection phase the target removes command phase early the target does not go to the command phase following the message out phase the target exits the command out phase early a successful operation/service request interrupt is generated when this command is executed successfully. general commands no operation command (command code 00h/80h) the no operation command administers no operation, therefore an interrupt is not generated upon completion. this command is issued following the reset device command to clear the command register (cmdreg). a no operation command in the dma mode may be used to verify the contents of the start transfer count register (stcreg). after the stcreg is loaded with the transfer count and a dma no operation command is issued, reading the current transfer count register (ctcreg) returns the transfer count value. p r e l i m i n a r y amd 87 AM79C974 clear fifo command (command code 01h) the clear fifo command is used to initialize the scsi fifo to the empty condition. the current fifo register (cfisreg) reflects the empty fifo status and the bot- tom of the fifo is set to zero. no interrupt is generated at the end of this command. reset device command (command code 02h) the reset device command immediately stops any de- vice operation and resets all the functions of the device. additionally, it returns the device to the disconnected state and it generates a hard reset. the reset device command remains on the top of the command register fifo holding the device in the reset state until the no operation command is loaded. once loaded, the no operation command serves to re-enable the command register. reset scsi bus command (command code 03h) the reset scsi bus command forces the scsi^rst signal active for a period of 25 ms, and drives the chip to the disconnected state. an interrupt is not generated upon command completion, however, if bit 6 is set to 0 in control register one (cntlreg1), a scsi reset in- terrupt will be issued. scsi power management features as a leader in low-voltage technology, amd has incor- porated power-saving features into the AM79C974. through hardware and software, the AM79C974 can be powered down to reduce consumption during chip inactivity. scsi activity pin the scsi bus activity is reflected by the busy output line. this signal, when active, indicates that the scsi bus is in use, therefore the AM79C974 should not be powered down. this signal is the logical equivalent to the scsi bus signal bsy , however, it is not physically connected to the bsy signal on the bus. to correctly identify the bus free state on the scsi bus, this busy output line must be inactive for at least 250 ms (selec- tion timeout period). reduced power mode when there is no activity on the scsi bus, and there are no pending commands, the AM79C974 may be pow- ered down by turning off the input buffers on the scsi bus lines (set bit 5 in control register four (b)+34h). for further power reduction, the internal registers may be programmed to a predetermined state, and the input clock disconnected via external logic. power down pin (pwdn pin) when pwdn is driven active, it sets the pwdn bit in the status register (bit 0, dma status register (b)+54h), signaling the AM79C974 that the host would like to power down the scsi interface. an interrupt is gener- ated when this bit is first set. software disk spin-down incorporated into the scsi rom bios and certain de- vice drivers of amds software solution is a module which physically spins down scsi fixed disks when the host system elects to enact power management on the scsi system. the software module is activated upon the rom bios and/or other device drivers receipt of an interrupt caused by the pwdn pin being driven active. upon receipt of this interrupt, the current software proc- ess is suspended and software control is given to the power management module. when the power management module is activated, it checks the status of all scsi fixed disks on the system under its control. the scsi fixed disks that are idle are issued a command to spin down their media. all fixed disks that are active at the time will be scheduled for spin down upon completion of their pending commands. once the bios and/or drivers have detected the com- pletion of each fixed disks final pending commands, they are issued the command to spin down as well. to spin down the disk drives, the power management module issues a scsi command (1bC start/stop unit). when the command is received by the drive, it spins down and waits in an idle state. for multiple drives on the scsi bus, the power management driver spins down each drive individually. the drives remain in the idle state until the bios and/or driver receives a com- mand for the particular fixed disk. once this occurs, the drive is issued the command to spin up. when the drive has completely spun up and is ready, the pending com- mand is issued to the drive. only the particular fixed disk issued the command is instructed to spin up. all other drives will remain in the spun down state until a com- mand is issued to them. note : this sequence does not turn off the input buffers as described in the previous section. amd p r e l i m i n a r y 88 AM79C974 nand tree testing the AM79C974 controller provides a nand tree test mode to allow checking connectivity to the device on a printed circuit board. the nand tree is built on all pci bus signals (see figure 27 and table 8). the nand tree testing is enabled by asserting rst . all pci bus signals will become inputs when rst is as- serted. the result of the nand tree test can be ob- served on the busy pin. 18681a-35 a b o s mux rst (pin 120) inta (pin 117) intb (pin 118) clk (pin 121) ad2 (pin 54) ad1 (pin 56) ad0 (pin 57) pwdn (pin 58) pcnet-scsi core busy (pin 62) +5 v busy figure 27. AM79C974 nand tree test structure p r e l i m i n a r y amd 89 AM79C974 as shown in figure 27, pin 120 ( rst ) is the first input to the nand tree. pin 117 ( inta ) is the second input to the nand tree, followed by pin 118 ( intb ). all other pci bus signals follow, counterclockwise, with pin 58 (pwdn) being the last. ethernet and scsi specific pins and all power supply pins are not part of the nand tree. table 8 shows the complete list of pins connected to the nand tree. table 8. nand tree configuration nand nand nand tree tree tree input # pin # name input # pin # name input # pin # name 1 120 rst 20 12 ad23 39 36 ad15 2 117 inta 21 13 ad22 40 38 ad14 3 118 intb 22 15 ad21 41 39 ad13 4 121 clk 23 16 ad20 42 40 ad12 5 123 gntb 24 18 ad19 43 41 ad11 6 124 gnta 25 19 ad18 44 42 ad10 7 126 reqb 26 21 ad17 45 44 ad9 8 127 reqa 27 22 ad16 46 45 ad8 9 128 ad31 28 23 c/ be 247 47c/ be 0 10 129 ad30 29 24 frame 48 48 ad7 11 131 ad29 30 25 irdy 49 49 ad6 12 132 ad28 31 26 trdy 50 51 ad5 13 2 ad27 32 27 devsel 51 52 ad4 14 3 ad26 33 28 stop 52 53 ad3 15 5 ad25 34 29 lock 53 54 ad2 16 6 ad24 35 31 perr 54 56 ad1 17 7 c/ be 336 32 serr 55 57 ad0 18 9 idsela 37 34 par 56 58 pwdn 19 10 idselb 38 35 c/ be 1 amd p r e l i m i n a r y 90 AM79C974 rst must be asserted low to start a nand tree test se- quence. initially, all nand tree inputs except rst should be driven high. this will result in a high output at the busy pin. if the nand tree inputs are driven low in the same order as they are connected to build the nand tree, busy will toggle every time an additional input is driven low. busy will change to a zero, when inta is driven low and all other nand tree inputs stay high. busy will toggle back to high, when clk is additionally driven low. busy will continue toggling as long as the nand tree inputs are toggling in the sequence shown in figure 28. busy will be high, when all nand tree inputs are driven low. when testing is complete, deassert rst to exit this test mode. note that some of the pins connected to the nand tree are outputs in normal mode of operation. they must not be driven from an external source until the pcnet-scsi controller is configured for nand tree testing. 18681a-36 rst inta clk gntb reqa ad[31:0] c/ be [3:0] idsela frame irdy trdy devsel stop lock perr serr par ffffffff 31 0000ffff f 7 intb gnta reqb idselb busy pwdn figure 28. nand tree waveform p r e l i m i n a r y amd 91 AM79C974 absolute maximum ratings storage temperature C65 c to +150 c . . . . . . . . . . . ambient temperature under bias C55 c to +125 c . . . . . . . . . . . . . . . . . . . supply voltage to av ss or v ssb or v ssbs or dv ss (av dd , v dd , v ddb , v ddbs , dv dd ) C0.3 v to +6.0 v . . . stresses above those listed under absolute maximum rat- ings may cause permanent device failure. functionality at or above these limits is not implied. exposure to absolute maxi- mum ratings for extended periods may affect device reliability. operating ranges temperature (t a )0 c to +70 c . . . . . . . . . . . . . . . . . supply voltages (av dd , v dd , v ddb , v ddbs , dv dd ) +5 v 5% . . . . . . . . . all inputs within the range: av ss C 0.5 v v in av dd + 0.5 v, or v ss C 0.5 v v in v dd + 0.5 v, or v ssb C 0.5 v v in v ddb + 0.5 v or v ssbs C 0.5 v v in v ddbs + 0.5 v or dv ss C 0.5 v v in dv dd + 0.5 v operating ranges define those limits between which the func- tionality of the device is guaranteed. dc characteristics over commercial operating range unless otherwise specified pci and board interface parameter symbol parameter description test conditions min max unit digital input voltage v il input low voltage 0.8 v v ih input high voltage 2.0 v digital output voltage v ol output low voltage i ol1 = 3 ma 0.55 v i ol2 = 6 ma i ol3 = 12 ma 0.4 v (note 1) v oh output high voltage (note 2) i oh = C2 ma (note 5) 2.4 v digital input leakage current i il input low leakage current (note 3) v in = 0 C10 10 m a i ih input high leakage current (note 3) v in = v dd , v ddb C10 +10 m a digital output leakage current i ozl output low leakage current (note 4) v out = 0.4 v C10 +10 m a i ozh output high leakage current (note 4) v out = v dd , v ddb C10 +10 m a crystal input current v ilx xtal1 input low voltage threshold v in = external clock C0.5 0.8 v v ihx xtal1 input high voltage threshold v in = external clock v dd C 0.8 v dd + 0.5 v i ilx xtal1 input low current v in = external clock active C120 0 m a v in = v ss sleep C10 +10 m a i ihx xtal1 input high current v in = external clock active 0 120 m a v in = v dd sleep 400 m a amd p r e l i m i n a r y 92 AM79C974 dc characteristics (continued) attachment unit interface parameter symbol parameter description test conditions min max unit attachment unit interface (aui) i iaxd input current at di+ and diC C1 v < v in < av dd +0.5v C500 +500 m a i iaxc input current at ci+ and ciC C1 v < v in < av dd +0.5v C500 +500 m a v aod differential output voltage |(do+)C(doC)| r l = 78 w 630 1200 mv v aodoff transmit differential output idle voltage r l = 78 w (note 9) C40 40 mv i aodoff transmit differential output idle current r l = 78 w (note 8) C1 1 ma v cmt transmit output common mode voltage r l = 78 w 2.5 a vdd v v odi do transmit differential output r l = 78 w (note 7) 25 mv voltage imbalance v ath receive data differential input threshold C35 35 mv v asq di and ci differential input threshold C275 C160 mv (squelch) v irdvd di and ci differential mode input C1.5 1.5 v voltage range v icm di and ci input bias voltage i in = 0 ma av dd C3.0 av dd C1.0 v v opd do undershoot voltage at zero (note 9) C100 mv differential on transmit return to zero (etd) twisted pair interface (10base-t) i irxd input current at rxd av dd < v in < av dd C500 500 m a r rxd rxd differential input resistance 10 k w v tivb rxd , rxdC open circuit input i in = 0 ma av dd C3.0 av dd C1.5 v voltage (bias) v tidv differential mode input voltage av dd = 5.0 v C3.1 3.1 v range (rxd ) v tsq+ rxd positive squelch threshold (peak) sinusoid, 5 mhz f 10 mhz 300 520 mv v tsq- rxd negative squelch threshold (peak) sinusoid, 5 mhz f 10 mhz C520 C300 mv v ths+ rxd post-squelch positive threshold sinusoid, 5 mhz f 10 mhz 150 293 mv (peak) v ths- rxd post-squelch negative threshold sinusoid, 5 mhz f 10 mhz C293 C150 mv (peak) v ltsq+ rxd positive squelch threshold (peak) lrt = low 180 312 mv v ltsq- rxd negative squelch threshold (peak) lrt = low C312 C180 mv v lths+ rxd post-squelch positive threshold lrt = low 90 176 mv (peak) v lths- rxd post-squelch negative threshold lrt = low C176 C90 mv (peak) v rxdth rxd switching threshold (note 4) C35 35 mv v txh txd and txp output high voltage v ss = 0 v v dd C0.6 v dd v v txl txd and txp output low voltage v dd = 5 v v ss v ss +0.6 v v txi txd and txp differential output C40 40 mv voltage imbalance v tx off txd and txp idle output voltage 40 mv r tx txd , txp differential driver output (note 4) 40 80 w impedance p r e l i m i n a r y amd 93 AM79C974 dc characteristics (continued) scsi interface parameter symbol parameter description test conditions min max unit scsi and pwdn v ih input high voltage 2.0 v dd +v all scsi inputs, pwdn 0.5 v il input low voltage v ss C 0.8 v all scsi inputs, pwdn 0.5 v ihst input hysteresis (note 6) 4.75 v < v dd < 5.25 v 300 mv all scsi inputs v oh output high voltage (note 2) i oh = C2 ma (note 5) 2.4 v dd v v ol scsi output low voltage i ol = 48 ma v ss 0.5 v atn , bsy , sel , scsi^rst , ack, sd [7:0], sd p i il input low leakage 0.0 v < v in < 2.7 v C10 +10 m a all scsi inputs, pwdn i ih input high leakage 2.7 v < v in < v dd C10 +10 m a all scsi inputs, pwdn i oz high impedance leakage 0.4 v < v out < v dd C10 +10 m a all i/o pins power supply current i dd active power supply current xtal1 = 20 mhz, clk = 33 mhz 150 ma i ddcoma sleep mode power supply current sleep active tbd m a i ddsnooze auto wake mode power supply current awake bit set active tbd m a notes: 1. i ol1 applies to ad[31:00], c/ be [3:0], par, and req i ol2 applies to frame , irdy , trdy , devsel , stop , serr , perr , and lock i ol3 applies to eesk/ led1 , eedo/ led3 , and eedi/ lnkst 2. v oh does not apply to open-drain output pins. 3. i il and i ih apply to all input pins except xtal1. 4. i ozl and i ozh apply to all three-state output pins and bi-directional pins. 5. outputs are cmos and will be driven to rail if the load is not resistive. 6. these parameters are not 100% tested, but are evaluated at initial characterization and at any time the design is modified where these parameters may be affected. 7. tested, but to values in excess of limits. test accuracy not sufficient to allow screening guard bands. 8. correlated to other tested parameters C not tested directly. 9. test not implemented to data sheet specification. amd p r e l i m i n a r y 94 AM79C974 dc characteristics (continued) capacitance, esd, and latch up parameter symbol parameter description pin names test conditions min max unit pin capacitance (note 1) (v cc = 5.0 v, t a = 25 c, f = 1.0 mhz) c in input pins all scsi inputs v in = 0 v 10 pf all ethernet inputs pwdn all pci inputs except idsel idsel v in = 0 v 8 pf c i/o i/o or output pins all scsi, ethernet, v i/o = 0 v 12 pf pci output and i/o pins, busy c clk clock pins clk (pci) v in = 0 v 12 pf scsi clk esd (note 1) input static discharge pin-to-pin human body model: 2k v 100 pf at 1.5 k w latchup (note 1) i lu latchup current all i/o v lu 10 v C100 +100 ma note: 1. these parameters are not 100% tested, but are evaluated at initial characterization and at any time the design is modified where these parameters may be affected. p r e l i m i n a r y amd 95 AM79C974 ac switching characteristics over operating range unless otherwise specified pci bus interface and board interface parameter symbol parameter description test conditions min max unit clock timing clk frequency 0 33 mhz t cyc clk period @ 1.5 v 30 ns t high clk high time @ 2.0 v 12 ns t low clk low time @ 0.8 v 12 ns t fall clk fall time over 2 v p-p 1 4 v/ns t rise clk rise time over 2 v p-p 1 4 v/ns output and float delay timing t val ad[31:00], c/ be [3:0], par, 2 11 ns frame , irdy , trdy , stop , lock , devsel , perr , serr valid delay t val ( req ) req valid delay 1 12 ns f eesk eesk frequency (see note below) 650 khz t val (eedi) eedi valid output delay from eesk (see note below) 100 400 ns t val (eesk) eecs valid output delay from eesk (see note below) 0 400 ns t on ad[31:00], c/ be [3:0], par, frame , 2 11 ns irdy , trdy , stop , lock , devsel active delay t off ad[31:00], c/ be [3:0], par, frame ,28ns irdy , trdy , stop , lock , devsel float delay setup and hold timing t su ad[31:00], c/ be [3:0], par, frame ,7ns irdy , trdy , stop , lock , devsel , idsel setup time t h ad[31:00], c/ be [3:0], par, frame ,0ns irdy , trdy , stop , lock , devsel , idsel hold time t su ( gnt ) gnt setup time 10 ns t h ( gnt ) gnt hold time 0 ns t su (eedo) eedo setup time to eesk (see note below) 50 ns t h (eedo) eedo hold time from eesk (see note below) 0 ns note: parameter value is given for automatic eeprom read operation. when eeprom port (bcr19) is used to access the eeprom, software is responsible for meeting eeprom timing requirements. amd p r e l i m i n a r y 96 AM79C974 ac switching characteristics 10base-t interface parameter symbol parameter description test conditions min max unit transmit timing t tetd transmit start of idle 250 350 ns t tr transmitter rise time (10% to 90%) 5.5 ns t tf transmitter fall time (90% to 10%) 5.5 ns t tm transmitter rise and fall time mismatch (t tm = | t tr C t tf |) 1 ns t perlp idle signal period 8 24 ms t pwlp idle link pulse width (see note below) 75 120 ns t pwplp predistortion idle link pulse width (see note below) 45 55 ns t ja transmit jabber activation time 20 150 ms t jr transmit jabber reset time 250 750 ms t jrec transmit jabber recovery time 1.0 m s (minimum time gap between transmitted frames to prevent jabber activation) receiving timing t pwnrd rxd pulse width not to turn off v in > v ths (min) 136 ns internal carrier sense t pwroff rxd pulse width to turn off v in > v ths (min) 200 ns t retd receive start of idle 200 ns note: not tested; parameter guaranteed by characterization. p r e l i m i n a r y amd 97 AM79C974 ac switching characteristics attachment unit interface parameter symbol parameter description test conditions min max unit aui port t dotr do+, doC rise time (10% to 90%) 2.5 5.0 ns t dotf do+, doC fall time (10% to 90%) 2.5 5.0 ns t dorm do+, doC rise and fall time mismatch 1.0 ns t doetd do end of transmission 200 375 ns t pwodi di pulse width accept/reject threshold |v in | > |vasq| (note 1) 15 45 ns t pwkdi di pulse width maintain/turn-off |v in | > |vasq| (note 2) 136 200 ns threshold t pwoci ci pulse width accepth/reject threshold |v in | > |vasq| (note 3) 10 26 ns t pwkci ci pulse width maintain/turn-off |v in | > |vasq| (note 4) 90 160 ns threshold internal mendec clock timing tx1 xtal1 period v in = external clock 49.995 50.001 ns tx1h xtal1 high pulse width v in = external clock 20 ns tx1l xtal1 low pulse width v in = external clock 20 ns tx1r xtal1 rise time v in = external clock 5 ns tx1f xtal1 fall time v in = external clock 5 ns notes: 1. di pulses narrower than t pwodi (min) will be rejected; pulses wider than t pwodi (max) will turn internal di carrier sense on. 2. di pulses narrower than t pwkdi (min) will maintain internal di carrier sense on; pulses wider than t pwkdi (max) will turn internal di carrier sense off. 3. ci pulses narrower than t pwoci (min) will be rejected; pulses wider than t pwoci (max) will turn internal ci carrier sense on. 4. ci pulses narrower than t pwkci (min) will maintain internal ci carrier sense on; pulses wider than t pwkci (max) will turn internal ci carrier sense off. amd p r e l i m i n a r y 98 AM79C974 ac switching characteristics scsi interface fastclk disabled (control register three (0ch) bit 3=0), see figure 29 parameter no. symbol parameter description test conditions min max unit 1t pwl 1 clock pulse width low 14.58 0.65 t cp ns 2t cp clock period (1 ? clock frequency) 40 100 ns 3t l synchronization latency 54.58 t pwl + t cp ns 4t pwh 1 clock pulse width high 14.58 0.65 t cp ns 5t rise * clock rise time over 2 v p-p 1 4 v/ns 6t fall * clock fall time over 2 v p-p 1 4 v/ns notes: 1. for synchronous data transmissions, the following condi- tions must be true: 2t cp + t pwl 97.92 ns 2t cp + t pwh 97.92 ns clock frequency range for fast clk disabled. = 10 mhz to 25 mhz for asynchronous transmission = 12 mhz to 25 mhz for synchronous transmission * these parameters are not 100% tested, but are evalu- ated at initial characterization and at any time the design is modified where these parameters may be affected. fastclk enabled (control register three (0ch) bit 3=1), see figure 29 parameter no. symbol parameter description test conditions min max unit 1t pwl clock pulse width low 0.4 t cp 0.6 t cp ns 2t cp clock period (1 ? clock frequency) 25 50 ns 3a t l synchronization latency 54.58 2 t cp ns 4t pwh clock pulse width high 0.4 t cp 0.6 t cp ns 5t rise * clock rise time over 2 v p-p 1 4 v/ns 6t fall * clock fall time over 2 v p-p 1 4 v/ns notes: clock frequency range for fast clk enabled. = 20 mhz to 40 mhz for asynchronous transmission = 20 mhz to 40 mhz for synchronous transmission * these parameters are not 100% tested, but are evalu- ated at initial characterization and at any time the design is modified where these parameters may be affected. p r e l i m i n a r y amd 99 AM79C974 ac switching characteristics (continued) scsi interface parameter no. symbol parameter description test conditions min max unit single ended: asynchronous initiator transmit, see figure 30 7t s data to ack set up time 60 ns 8t pd req to data delay 5 ns 9t pd req to ack delay 50 ns 10 t pd req to ack delay 50 ns single ended: asynchronous initiator receive, see figure 31 11 t pd req to ack delay 50 ns 12 t pd req to ack delay 50 ns 13 t s data to req set up time 10 ns 14 t h req to data hold time 18 ns normal scsi: (single ended) synchronous initiator transmit, see figure 32 15 t s data to req or ack 55* ns set up time 16 t pwl req or ack pulse width low 90* ns 17 t pwh req or ack pulse width high 90* ns 18 t h ack or req to data hold time 100* ns fast scsi: (single ended) synchronous initiator transmit, see figure 32 15 t s data to req or ack 25* ns set up time 16 t pwl req or ack pulse width low 30* ns 17 t pwh req or ack pulse width high 30* ns 18 t h ack or req to data hold time 35* ns synchronous initiator receive, see figure 33 19 t pwl req pulse width low 27 ns 19 t pwl ack pulse width low 20 ns 20 t pwh req pulse width high 20 ns 20 t pwh ack pulse width high 20 ns 21 t s data to req or ack 10 ns set up time 22 t h req or ack to data hold time 15 ns scsi bus lines t rise ** rise time, 10% to 90% scsi termination 8 20 ns + 20 pf t fall ** fall time, 10% to 90% scsi termination 5 12 ns + 20 pf dv h /dt** slew rate, low to high scsi termination 0.15 0.50 v/ns + 20 pf dv l /dt** slew rate, high to low scsi termination 0.25 0.80 v/ns + 20 pf * the minimum values have a wide range since they depend on the synchronization latency. the synchronization latency, in turn, depends on the operating frequency of the device. **these parameters are not 100% tested, but are evaluated at initial characterization and at any time the design is modified where these parameters may be affected. amd p r e l i m i n a r y 100 AM79C974 key to switching waveforms must be steady may change from h to l may change from l to h does not apply dont care, any change permitted will be steady will be changing from h to l will be changing from l to h changing, state unknown center line is high- impedance off state waveform inputs outputs ac switching test circuits c l v threshold i ol i oh sense point 18681a-37 normal and three-state outputs p r e l i m i n a r y amd 101 AM79C974 ac switching test circuits av dd do+ 154 w 100 pf 18681a-38 doC av ss 52.3 w test point aui do switching test circuit dv dd txd+ 294 w 100 pf 18681a-39 txdC dv ss 294 w test point includes test jig capacitance txd switching test circuit dv dd txp+ 715 w 100 pf 18681a-40 txpC dv ss 715 w test point includes test jig capacitance txp outputs test circuit amd p r e l i m i n a r y 102 AM79C974 ac switching waveforms system bus interface 18681a-41 clk 0.8 v 1.5 v 2.0 v t high t fall t cyc t rise t low 0.8 v 0.4 v 2.4v 1.5 v 2.0 v clk waveform 18681a-42 clk t h ad[31:00], c/ be [3:0], par, frame , irdy , trdy , stop , lock , devsel , idsel t su gnt t h( gnt ) t su( gnt ) tx tx input setup and hold timing p r e l i m i n a r y amd 103 AM79C974 ac switching waveforms system bus interface 18681a-43 clk t val( req ) tx tx tx min max valid n valid n+1 req min max valid n valid n+1 t val ad[31:00] c/ be [3:0], par, frame , irdy , trdy , stop , lock , devsel , perr , serr output valid delay timing 18681a-44 clk tx tx tx ad[31:00], c/ be [3:0], par, frame , irdy , trdy , stop , lock , devsel ad[31:00], c/ be [3:0], par, frame , irdy , trdy , stop , lock , devsel valid n t off t on valid n output tri-state delay timing amd p r e l i m i n a r y 104 AM79C974 ac switching waveforms 10base-t interface txd+ txp+ txdC txpC xmt (note 1) t xmtoff t tr t tf t tetd t xmton 18681a-45 note: 1. internal signal and is shown for clarification only. transmit timing txd+ 18681a-46 t perlp t pwlp t pwplp txp+ txdC txpC idle link test pulse p r e l i m i n a r y amd 105 AM79C974 switching waveforms 10base-t interface 18681a-47 rxd v tsq+ v tsqC v thsC v ths+ tmau_rcv_lrt_hi receive thresholds (lrt=0) 18681a-48 rxd v ltsq+ v ltsqC v lthsC v lths+ tmau_rcv_lrt_hi receive thresholds (lrt=1) amd p r e l i m i n a r y 106 AM79C974 ac switching waveforms attachment unit interface 18681a-49 t xi t dotr t dotf t x1h t x1l t x1f t x1r 1 1 0 1 xtal1 istdclk (note 1) itxdat+ (note 1) do+ doC do 0 1 1 itxen (note 1) note: 1. internal signal and is shown for clarification only. transmit timing C start of frame 18681a-50 typical > 200 ns t doetd xtal1 istdclk (note 1) itxen (note 1) itxdat+ (note 1) do+ doC do 0 1 0 0 10 bit (nC2) bit (nC1) bit (n) 1 note: 1. internal signal and is shown for clarification only. transmit timing C end of frame (last bit = 0) p r e l i m i n a r y amd 107 AM79C974 switching waveforms attachment unit interface 18681a-51 typical > 250 ns t doetd xtal1 itxen (note 1) itxdat+ (note 1) do+ doC do 0 11 0 1 bit (nC2) bit (nC1) bit (n) 1 istdclk (note 1) note: 1. internal signal and is shown for clarification only. transmit timing C end of frame (last bit = 1) di+/C v asq t pwodi t pwkdi 18681a-52 t pwkdi receive timing amd p r e l i m i n a r y 108 AM79C974 ac switching waveforms attachment unit interface ci+/C v asq t pwoci t pwkci 18681a-53 t pwkci collision timing t doetd 18681a-54 do+/C 40 mv 100 mv max. 0 v 80 bit times port do etd waveform p r e l i m i n a r y amd 109 AM79C974 ac switching test waveforms scsi interface 3.0 v 0.0 v v oh 2.0 v v ol v oh C 0.3 v 2.0 v v ol + 0.3 v 2.0 v 0.8 v 1.5 v 0.8 v 2.3 v all open drain outputs hi-z outputs sd [7:0], sd p true data outputs sd [7:0], sd p all inputs v ol 18681a-55 18681a-56 scsi clk 2 4 1 3 3a 5 6 figure 29. clock input 18681a-57 req ack sd [7:0] sd p 8 7 9 10 figure 30. asynchronous initiator transmit amd p r e l i m i n a r y 110 AM79C974 ac switching test waveforms scsi interface 18681a-58 req ack sd [7:0] sd p 11 12 13 14 figure 31. asynchronous initiator receive 18681a-59 req ack sd [7:0] sd p 15 18 16 17 figure 32. synchronous initiator transmit 18681a-60 req ack sd [7:0] sd p 21 22 19 20 figure 33. synchronous initiator receive p r e l i m i n a r y amd 111 AM79C974 physical dimensions* pqb132 plastic quad flat pack trimmed and formed (measured in inches) top view 20010a cl85 08/27/93 mh 1.097 1.103 1.075 1.085 0.947 0.953 0.947 0.953 1.097 1.103 1.075 1.085 0.025 basic 0.160 0.180 0.020 0.040 side view 0.80 ref 0.008 0.012 0.008 0.016 pin 1 id pin 2 pin 33 pin 66 pin 99 pin 132 0.130 0.150 *for reference only. bsc is an ansi standard for basic space centering. trademarks copyright ? 1994 advanced micro devices, inc. all rights reserved. amd and the amd logo are registered trademarks of advanced micro devices, inc. product names used in this publication are for identification purposes only and may be trademarks of their respective companies. 112 AM79C974 register summary appendix a ethernet pci configuration registers note: ro = read only, rw = read/write, u = undefined value offset name width in bit access mode default value 00h vendor id 16 ro 1022h 02h device id 16 ro 2000h 04h command 16 rw 0uuu 06h status 16 rw uu00 08h revision id 8 ro 00h 09h programming if 8 ro 00h 0ah sub-class 8 ro 00h 0bh base-class 8 ro 02h 0dh latency timer 8 ro 00h 0eh header type 8 ro 00h 10h base address 32 rw uuuu uuuu 3ch interrupt line 8 rw uu 3dh interrupt pin 8 ro 02h scsi pci configuration registers note: ro = read only, rw = read/write, u = undefined value offset name width in bit access mode default value 00h vendor id 16 ro 1022h 02h device id 16 ro 2020h 04h command 16 rw 0080h 06h status 16 rw uuuu 08h revision id 8 ro 00h 09h programming if 8 ro 00h 0ah sub-class 8 ro 00h 0bh base-class 8 ro 01h 0dh latency timer 8 ro 00h 0eh header type 8 ro 00h 10h base address 32 rw uuuu uuuu 3ch interrupt line 8 rw uu 3dh interrupt pin 8 ro 01h 40h reserved for software 32 rw uuuu 44h reserved for software 32 rw uuuu 48h reserved for software 32 rw uuuu 4ch reserved for software 32 rw uuuu amd p r e l i m i n a r y 113 AM79C974 ethernet controller control and status registers note: u = undefined value, r = running register, s = setup register, t = test register default value after rap addr symbol h_reset comments use 00 csr0 uuuu 0004 pcnet-scsi status register r 01 csr1 uuuu uuuu iadr[15:0]: base address of init block lower s 02 csr2 uuuu uuuu iadr[31:16]: base address of init block upper s 03 csr3 uuuu 0000 interrupt masks and deferral control s 04 csr4 uuuu 0115 test and features control r 05 csr5 uuuu 0000 reserved t 06 csr6 uuuu uuuu rxtx: rx/tx descriptor table lengths t 07 csr7 uuuu 0000 reserved t 08 csr8 uuuu uuuu ladr0: logical address filter ladrf[15:0] t 09 csr9 uuuu uuuu ladr1: logical address filter ladrf[31:16] t 10 csr10 uuuu uuuu ladr2: logical address filter ladrf[47:32] t 11 csr11 uuuu uuuu ladr3: logical address filter ladrf[63:48] t 12 csr12 uuuu uuuu padr0: physical address register padr[15:0] t 13 csr13 uuuu uuuu padr1: physical address register padr[31:16] t 14 csr14 uuuu uuuu padr2: physical address register padr[47:32] t 15 csr15 see reg. desc. mode: mode register s 16 csr16 uuuu uuuu iadr[15:0]: alias of csr1 t 17 csr17 uuuu uuuu iadr[31:16]: alias of csr2 t 18 csr18 uuuu uuuu crbal: current rcv buffer address lower t 19 csr22 uuuu uuuu crbau: current rcv buffer address upper t 20 csr20 uuuu uuuu cxbal: current xmt buffer address lower t 21 csr21 uuuu uuuu cxbau: current xmt buffer address upper t amd p r e l i m i n a r y 114 AM79C974 control and status registers (continued) default value after rap addr symbol h_reset comments use 22 csr22 uuuu uuuu nrbal: next rcv buffer address lower t 23 csr23 uuuu uuuu nrbau: next rcv buffer address upper t 24 csr24 uuuu uuuu badrl: base address of rcv ring lower s 25 csr25 uuuu uuuu badru: base address of rcv ring upper s 26 csr26 uuuu uuuu nrdal: next rcv descriptor address lower t 27 csr27 uuuu uuuu nrdau: next rcv descriptor address upper t 28 csr28 uuuu uuuu crdal: current rcv descriptor address lower t 29 csr29 uuuu uuuu crdau: current rcv descriptor address upper t 30 csr30 uuuu uuuu badxl: base address of xmt ring lower s 31 csr31 uuuu uuuu badxu: base address of xmt ring upper s 32 csr32 uuuu uuuu nxdal: next xmt descriptor address lower t 33 csr33 uuuu uuuu nxdau: next xmt descriptor address upper t 34 csr34 uuuu uuuu cxdal: current xmt descriptor address lower t 35 csr35 uuuu uuuu cxdau: current xmt descriptor address upper t 36 csr36 uuuu uuuu nnrdal: next next receive descriptor address lower t 37 csr37 uuuu uuuu nnrdau: next next receive descriptor address upper t 38 csr38 uuuu uuuu nnxdal: next next transmit descriptor address lower t 39 csr39 uuuu uuuu nnxdau: next next transmit descriptor address upper t 40 csr40 uuuu uuuu crbc: current rcv byte count t 41 csr41 uuuu uuuu crst: current rcv status t 42 csr42 uuuu uuuu cxbc: current xmt byte count t 43 csr43 uuuu uuuu cxst: current xmt status t 44 csr44 uuuu uuuu nrbc: next rcv byte count t 45 csr45 uuuu uuuu nrst: next rcv status t 46 csr46 uuuu uuuu poll: poll time counter t 47 csr47 uuuu uuuu pollint: polling interval s 48 csr48 uuuu uuuu reserved t 49 csr49 uuuu uuuu reserved t 50 csr50 uuuu uuuu reserved t 51 csr51 uuuu uuuu reserved t 52 csr52 uuuu uuuu reserved t 53 csr53 uuuu uuuu reserved t 54 csr54 uuuu uuuu reserved t 55 csr55 uuuu uuuu reserved t 56 csr56 uuuu uuuu reserved t 57 csr57 uuuu uuuu reserved t 58 csr58 see reg. desc. sws: software style s 59 csr59 uuuu 0105 ir: ir register t 60 csr60 uuuu uuuu pxdal: previous xmt descriptor address lower t 61 csr61 uuuu uuuu pxdau: previous xmt descriptor address upper t 62 csr62 uuuu uuuu pxbc: previous xmt byte count t 63 csr63 uuuu uuuu pxst: previous xmt status t 64 csr64 uuuu uuuu nxba: next xmt buffer address lower t amd p r e l i m i n a r y 115 AM79C974 control and status registers (continued) default value after rap addr symbol h_reset comments use 65 csr65 uuuu uuuu nxbau: next xmt buffer address upper t 66 csr66 uuuu uuuu nxbc: next xmt byte count t 67 csr67 uuuu uuuu nxst: next xmt status t 68 csr68 uuuu uuuu reserved t 69 csr69 uuuu uuuu reserved t 70 csr70 uuuu uuuu reserved t 71 csr71 uuuu uuuu reserved t 72 csr72 uuuu uuuu rcvrc: rcv ring counter t 73 csr73 uuuu uuuu reserved t 74 csr74 uuuu uuuu xmtrc: xmt ring counter t 75 csr75 uuuu uuuu reserved t 76 csr76 uuuu uuuu rcvrl: rcv ring length s 77 csr77 uuuu uuuu reserved t 78 csr78 uuuu uuuu xmtrl: xmt ring length s 79 csr79 uuuu uuuu reserved t 80 csr80 uuuu e810 dmatcfw: dma transfer counter and fifo threshold s 81 csr81 uuuu uuuu reserved t 82 csr82 uuuu 0000 dmabat: bus activity timer s 83 csr83 uuuu uuuu reserved t 84 csr84 uuuu uuuu dmabal: dma address register lower t 85 csr85 uuuu uuuu dmabau: dma address register upper t 86 csr86 uuuu uuuu dmabc: buffer byte counter t 87 csr87 uuuu uuuu reserved t 88 csr88 0242 0003 chip id register lower t 89 csr89 uuuu 0242 chip id register upper t 90 csr90 uuuu uuuu raeo register s 91 csr91 uuuu uuuu reserved t 92 csr92 uuuu uuuu rcon: ring length conversion t 93 csr93 uuuu uuuu reserved t 94 csr94 uuuu 0000 xmttdr: transmit time domain reflectometry count t 95 csr95 uuuu uuuu reserved t 96 csr96 uuuu uuuu reserved t 97 csr97 uuuu uuuu reserved t 98 csr98 uuuu uuuu reserved t 99 csr99 uuuu uuuu reserved t 100 csr100 uuuu 0200 merrto: bus time-out s 101 csr101 uuuu uuuu reserved t 102 csr102 uuuu uuuu reserved t 103 csr103 uuuu 0105 reserved t 104 csr104 uuuu uuuu reserved t 105 csr105 uuuu uuuu reserved t 106 csr106 uuuu uuuu reserved t 107 csr107 uuuu uuuu reserved t amd p r e l i m i n a r y 116 AM79C974 control and status registers (continued) default value after rap addr symbol h_reset comments use 108 csr108 uuuu uuuu reserved t 109 csr109 uuuu uuuu reserved t 110 csr110 uuuu uuuu reserved t 111 csr111 uuuu uuuu reserved t 112 csr112 uuuu 0000 mfc: missed frame count r 113 csr113 uuuu uuuu reserved t 114 csr114 uuuu 0000 rcc: receive collision count r 115 csr115 uuuu uuuu reserved t 116 csr116 uuuu uuuu reserved t 117 csr117 uuuu uuuu reserved t 118 csr118 uuuu uuuu reserved t 119 csr119 uuuu uuuu reserved t 120 csr120 uuuu uuuu reserved t 121 csr121 uuuu uuuu reserved t 122 csr122 see reg. desc. receive frame alignment control s 123 csr123 uuuu uuuu reserved t 124 csr124 see reg. desc. test register 1 t 125 csr125 uuuu uuuu reserved t 126 csr126 uuuu uuuu reserved t 127 csr127 uuuu uuuu reserved t bcrbus configuration registers writes to those registers marked as reserved will have no effect. reads from these locations will produce undefined values. bcr mnemonic default description user eeprom 0 msrda 0005h reserved no no 1 mswra 0005h reserved no no 2 mc n/a* miscellaneous configuration yes yes 3 reserved n/a reserved no no 4 lnkst 00c0h link status (default) yes no 5 led1 0084h receive status (default) yes no 6 led2 0088h reserved yes no 7 led3 0090h transmit status (default) yes no 8C15 reserved n/a reserved no no 16 iobasel n/a* reserved yes yes 17 iobaseu n/a* reserved yes yes 18 bsbc 2101h burst size and bus control yes yes 19 eecas 0002h eeprom control and status yes no 20 sws 0000h software style yes no 21 intcon n/a* reserved yes yes programmability * registers marked with an * have no default value, since they are not observable without first being programmed through the eeprom read operation. therefore, the only observable values for these registers are those that have been programmed and a default value is not applicable. amd p r e l i m i n a r y 117 AM79C974 scsi controller scsi register map register acronym address (hex.) register description type ctcreg (b)+00 current transfer count register low read stcreg (b)+00 start transfer count register low write ctcreg (b)+04 current transfer count register middle read stcreg (b)+04 start transfer count register middle write ffreg (b)+08 scsi fifo register read/write cmdreg (b)+0c scsi command register read/write statreg (b)+10 scsi status register read sdidreg (b)+10 scsi destination id register write instreg (b)+14 interrupt status register read stimreg (b)+14 scsi timeout register write isreg (b)+18 internal state register read stpreg (b)+18 synchronous transfer period register write cfisreg (b)+1c current fifo/internal state register read sofreg1 (b)+1c synchronous offset register write cntlreg1 (b)+20 control register one read/write clkfreg (b)+24 clock factor register write res (b)+28 reserved write cntlreg2 (b)+2c control register two read/write cntlreg3 (b)+30 control register three read/write cntlreg4 (b)+34 control register four read/write ctcreg (b)+38 current transfer count register high/part-unique id code read stcreg (b)+38 start current transfer count register high write res (b)+3c reserved write dma register map register acronym address (hex.) register description type cmd (b)+40 command r/w stc (b)+44 starting transfer count r/w spa (b)+48 starting physical address r/w wbc (b)+4c working byte counter r wac (b)+50 working address counter r status (b)+54 status register r smdla (b)+58 starting memory descriptor list (mdl) address r/w wmac (b)+5c working mdl counter r 118 AM79C974 pcnet-scsi compatible media interface modules appendix b pcnet-scsi compatible 10base-t filters and transformers manufacturer part # package description bel fuse a556-2006-de 16 pin 0.3 dil transmit and receive filters and transformers. bel fuse 0556-2006-00 14-pin sip transmit and receive filters and transformers. bel fuse 0556-2006-01 14-pin sip transmit and receive filters, transformers and common mode chokes. bel fuse 0556-6392-00 16-pin 0.5 dil transmit and receive filters, transformers and common mode chokes. halo electronics fd02-101g 16-pin 0.3 dil transmit and receive filters and transformers. halo electronics fd12-101g 16-pin 0.3 dil transmit and receive filters and transformers, transmit common mode choke. halo electronics fd22-101g 16-pin 0.3 dil transmit and receive filters, transformers and common mode chokes. pca electronics epa1990a 16-pin 0.3 dil transmit and receive filters and transformers. pca electronics epa2013d 16-pin 0.3 dil transmit and receive filters and transformers, transmit common mode choke. pulse engineering pe-65434 10-pin sip transmit and receive filters, transformers, and common mode choke. pulse engineering pe-65445 16-pin 0.3 dil transmit and receive filters and transformers (for smt use pe-65446) pulse engineering pe65467 16-pin 0.3 dil transmit and receive filters, transformers, common mode chokes, and amd specified resistors. pulse engineering pe-65424 16-pin 0.3 dil transmit and receive filters, transformers, and common mode chokes. tdk tla 470 14-pin sip transmit and receive filters and transformers. tdk him3000 24-pin 0.6 dil transmit and receive filters, transformers and common mode chokes. valor electronics pt3877 16-pin 0.3 dil transmit and receive filters and transformers. valor electronics pt3983 8-pin 0.3 dil transmit and receive common mode chokes. valor electronics fl1012 16-pin 0.3 dil transmit and receive filters and transformers, transmit common mode choke. amd p r e l i m i n a r y 119 AM79C974 pcnet-scsi compatible aui isolation transformers manufacturer part # package description bel fuse a553-0506-ab 16-pin 0.3 dil 50 m h halo electronics td01-0756k 16-pin 0.3 dil 75 m h halo electronics tg01-0756w 16-pin 0.3 smd 75 m h pca electronics ep9531-4 16-pin 0.3 dil 50 m h pulse engineering pe64106 16-pin 0.3 dil 50 m h tdk tla 100-3e 16-pin 0.3 dil 100 m h valor electronics lt6031 16-pin 0.3 dil 50 m h manufacturer contact information contact the following companies for further information on their products: bel fuse phone: (317) 831-4226 fax: (317) 831-4547 halo electronics phone: (415) 989-7313 fax: (415) 367-7158 pca electronics phone: (408) 954-0400 fax: (408) 954-1440 pulse engineering phone: (619) 674-8218 fax: (619) 675-8262 tdk phone: (213) 530-9397 fax: (213) 530-8127 valor electronics phone: (619) 458-1471 fax: (619) 458-0875 120 AM79C974 recommendation for power and ground decoupling appendix c the pcnet-scsi controller is an integrated, combina- tion ethernet and fast scsi controller, which contains both digital and analog circuitry. the analog circuitry contains a high speed phase-locked loop (pll) and voltage controlled oscillator (vco). because of the mixed signal characteristics of this chip, some extra pre- cautions must be taken into account when designing with this device. digital decoupling the pcnet-scsi controller separates the digital power supply pins into two groups. the v ssb and v ddb pins sup- ply power to the i/o buffers that connect to the pci bus. the v ss and v dd pins supply power to all internal cir- cuitry. amd recommends that at least one low fre- quency bulk decoupling capacitor be used for each group of digital power pins. 22 m f capacitors have worked well for this. in addition, a total of four or five 0.1 m f capacitors have proven sufficient around v ssb and v ddb pins that supply the drives of the pci bus out- put pins. an additional two to the three 0.1 m f capacitors shoud be used around the v ss and v dd pins. the cmos technology used in fabricating the pcnet- scsi controller employs an n-type substrate. in this technology, all v dd and v ddb pins are electrically con- nected to each other internally. hence, in a multi-layer board, when decoupling between v dd /v ddb and critical v ss /v ssb pins, the specific v dd /v ddb pin that you connect to is not critical. in fact, the v dd /v ddb connection of the decoupling capacitor can be made directly to the power plane, near the closest v dd /v ddb pin to the v ss /v ssb pin of interest. however, we recommend that the v ss /v ssb connection of the decoupling capacitor be made directly to the v ss /v ssb pin of interest as shown. v dd /v ddb pin v ss /v ssb pin pcnet-scsi via to v dd plane via to v ss plane 18681a-61 amd recommends that at least one low-frequency bulk decoupling capacitor be used in the area of the pcnet- scsi controller. 22 m f capacitors have worked well for this. in addition, a total of four or five 0.1 m f capacitors have proven sufficient around the v ssb and v ddb pins that supply the drivers of the pci bus output pins. analog decoupling the most critical pins are the analog supply and ground pins. all of the analog supply and ground pins are lo- cated in one corner of the device. specific requirements of the analog supply pins are listed below. av ss1 (pin 100) and av dd3 (pin 96) these pins provide the power and ground for the twisted pair and aui drivers. hence, they are very noisy. a dedicated 0.1 m f capacitor between these pins is recommended. av ss2 (pin 98) and av dd2 (pin 108) these pins are the most critical pins on the pcnet-scsi controller because they provide the power and ground for the pll portion of the chip. the vco portion of the pll is sensitive to noise in the 60 khz C200 khz. range. to prevent noise in this frequency range from disrupting the vco, amd strongly recommends that the low-pass filter shown below be implemented on these pins. tests using this filter have shown significantly increased noise immunity and reduced bit error rate (ber) statistics in designs using the pcnet-scsi controller. pcnet-scsi av dd pin 108 av ss pin 98 v dd plane 33 m f to 6.8 m f r1 2.7 w to 20 w 18681a-62 amd p r e l i m i n a r y 121 AM79C974 to determine the value for the resistor and capacitor, the formula is: r * c 3 88 where r is in ohms and c is in microfarads. some pos- sible combinations are given below. to minimize the voltage drop across the resistor, the r value should not be more than 20 w . 2.7 w 33 m f 4.3 w 22 m f 6.8 w 15 m f 10 w 10 m f 20 w 6.8 m f rc av dd1 (pin 103) and av dd4 (pin 91) these pins provide power for the aui and twisted-pair receive circuitry. no specific decoupling has been nec- essary on these pins. 122 appendix d alternative method for initialization of ethernet controller appendix d the ethernet portion of the pcnet-scsi controller may be initialized by performing i/o writes only. that is, data can be written directly to the appropriate control and status registers (csr) instead of reading from the initialization block in memory. the registers that must be written are shown in the table below. these register writes are followed by writing the start bit in csr0. control and status register comment csr8 ladrf[15:0] csr9 ladrf[31:16] csr10 ladrf[47:32] csr11 ladrf[63:48] csr12 padr[15:0] csr13 padr[31:16] csr14 padr[47:32] csr15 mode csr24-25 badr csr30-31 badx csr47 pollint csr76 rcvrl csr78 xmtrl note : the init bit must not be set or the initialization block will be accessed instead. 123 AM79C974 scsi system considerations appendix e introduction this appendix covers motherboard design considera- tions which use the AM79C974. it discusses scsi com- ponent placement, signal routing, pci interface recommendations, noise considerations and termina- tion schemes. signal routing and scsi placement the main components for board design include the AM79C974 (whose maximum trace length from the pci speedway is 1.5 ), scsi connectors (either one or two depending internal/external support), a 40 mhz crystal oscillator and regulated terminators (on-board) which can be up to .1 m (3.937 in) away from the AM79C974. there are many ways to route scsi bus traces on a host adapter board or motherboard. ideally, traces from the internal scsi bus connector, from the AM79C974 and from the external scsi bus connector should all connect in series. care should be taken not to have any stubs in the scsi bus. (stubs are any extensions off of the main bus. the maximum scsi bus stub length allowed is .1 m). this routing scheme helps maintain signal integrity by reducing the possibility of signal reflections and other undesirable effects. auto-routing programs used for board layout may not follow this scheme, and may cre- ate non-ideal environments by routing internal and ex- ternal on-board connectors first instead of routing both sets of traces to the AM79C974. when peripherals are added either internally or externally, a three-pronged scsi bus will be created instead of a linear one. if this or any other stub problem occurs, changes should be made manually to follow the ideal scheme. refer to fig- ures e-1 and e-2. 18681a-63 connector for internal peripheral termination external scsi-2 bus cable external peripheral high-density scsi-2 connectors high-density scsi-2 termination AM79C974 to pci bus 25 linear bus- no stubs figure e-1. ideal routing scheme amd p r e l i m i n a r y 124 AM79C974 18681a-64 external scsi-2 bus cable external peripheral high-density scsi-2 connectors high-density scsi-2 termination AM79C974 to pci bus 25 connector for internal peripheral termination 3-pronged bus (not recommended) stub bracketed card joining ribbon cable with high- density scsi-2 cable figure e-2. a poor routing scheme the motherboard the following two layouts may be used as a guideline for the design of motherboards which incorporate the AM79C974. for both layouts, the scsi connectors should be placed as far away from the pci speedway as possible. each layout refers to termination considera- tions which are described in further detail in the termi- nation considerations section. layout #1 this approach avoids the cost of placing an external connector on the motherboard. in this configuration, the AM79C974 is always at one end of the scsi bus, there- fore, the regulated terminators remain active. this elimi- nates the problem of switching the regulated terminators on or off to accommodate peripheral con- figurations. this approach also preserves the ideal linear routing scheme. the following lists the require- ments for implementation, while figure e-3 illustrates this approach. one connector on the motherboard connected to one end of the internal bus ribbon cable and its components. the other end of the internal bus ribbon cable con- nected to one end of the external bus high density cable and its peripherals via a bracketed add-on card. the internal bus may also be crimped to a scsi connector mounted on a bracket. on board regulated terminators a maximum of 0.1 m (3.937 in) from the AM79C974. external terminators connected to the end of the external bus. amd p r e l i m i n a r y 125 AM79C974 18681a-65 cpu memory external peripheral high-density scsi-2 termination high-density scsi-2 connector keyboard connector bracketed card joining ribbon cable with high-density scsi-2 cable external scsi-2 bus cable internal scsi bus cable connector for internal peripheral to dc power ferrite bead to scsi clk1 and clk2 AM79C974 to pci bus always on 25 termination figure e-3. motherboard layout approach #1 layout #2 this approach uses a pizza-box type of motherboard, which has been incorporated into pc systems and work- stations. this design reduces the systems height so that its casing resembles a pizza box. this is partly the result of a riser card that enables cards to rest on their side instead of upright. this approach requires the fol- lowing and is illustrated in figure e-4: an external connector mounted on the mothe- rboard with routings that connect to the AM79C974. the AM79C974 must be as close as possible to this external connector since this part of the scsi bus consists of motherboard routings (not just rib- bon cable). an internal connector on the motherboard to ac- commodate internal drives. on-board regulated terminators for scsi drives not mounted within the system. however, should the user decide to connect a drive internally, termi- nators are not needed. if on-board regulated terminators are used, they should be placed within .1 m (3.937 in) from the AM79C974. amd p r e l i m i n a r y 126 AM79C974 18681a-66 this is optional. use accordingly to terminate both ends of the scsi bus. cpu memory internal drive (terminated) riser card high-density scsi-2 termination external scsi bus cable external peripheral high-density scsi-2 connectors keyboard connector to scsi clk1 and clk2 AM79C974 to dc power ferrite bead to pci bus termination 25 mounted high-density scsi-2 connector termination figure e-4. motherboard layout approach #2 motherboard designs which place an internal and exter- nal connector on either side of the AM79C974 are dis- couraged since: two scsi connectors are required on the mothe- rboardone more than what the other two ap- proaches call for. when the number of internal and external bus components increase, the on-board terminators would have to be turned off either through software or hardware, which may be undesirable to a board designer. the possibility for creating stubs exists if the con- nectors and AM79C974 are not routed correctly. see figure e-1 for the ideal routing scheme. noise considerations some areas of a pci motherboard or host adapter de- sign (which include the AM79C974) are more suscepti- ble to noise. they are: the 40 mhz crystal oscillator the scsi cables the dc power planes pins on the ics, specifically the AM79C974. electromagnetic interference (emi) there are several ways to reduce the amount of noise present in these areas: a 40 mhz crystal oscillator must be used in order to have a 10 mb/s scsi data rate. use of this 40 mhz crystal oscillator, which drives the scsi clk pin on the AM79C974, introduces the possibility of unwanted high frequency compo- nents, including harmonics, coupling with AM79C974 signals. to help prevent this, a resistor should be placed in series with the oscillator to form a low pass filter with the input capacitance (10 pf) of the scsi clk pin. this filter reduces the edge rate of the clock waveform, increasing both rise and fall times by 2 ns. this removes higher frequencies, specifically harmonics from the crystal oscillator waveform and thus reduces the amount of noise introduced into the AM79C974. a ferrite bead, which is essentially an inductor, may be used with the oscillator to prevent coupling of higher frequencies with dc power supply sig- nals. the ferrite bead blocks high frequency noise while acting like a short to the dc components. amd p r e l i m i n a r y 127 AM79C974 from an emi standpoint, scsi-2 high-density ca- bles should be used instead of a scsi-1 cables. unlike the scsi-1 flat ribbon cable, the scsi-2 cable is electrically more substantial. it is shielded and signal wires are strategically placed for better signal travel. decoupling methods as stated in appendix c, decoupling capacitors should be used across the v dd and v ss pins on the motherboard there are pairs of v dd and v ss pins on the AM79C974 that should each have their own decoupling capacitor. the following decoupling method should be used for the v dd /v ss pairs: connect the capacitor directly between a v dd /v ss pair of the AM79C974 so that it sits on the compo- nent side of the board. this configuration will allow the capacitor to filter undesired high frequency components directly at the AM79C974 and not only at the power planes. this not only minimizes the noise on the power planes that the chip sees, but it also filters any noise generated by the chip before it reaches the power planes. the leads to each end of the capacitor should be wide and may contain several feed-throughs to the v dd and v ss planes to reduce the inductance present. refer to figure e-5. amd p r e l i m i n a r y 129 AM79C974 decoupling methods which are not recommended include: connecting a capacitor only between the v dd and v ss planes so that it sits on the component side of the board. this doesnt allow the capacitor to re- duce noise directly at the chip, but it does allow for a reduction of power plane noise and of board components. (capacitors). connecting a capacitor only between the v dd and v ss planes so that it sits on the solder side of the board. this configuration also doesnt allow direct decoupling at the chip. it does save board space, but it requires an extra manufacturing step. termination considerations the use of active or regulated termination for termina- tors on the motherboard is recommended (see figure e-6), while external scsi terminators can be used to terminate other parts of the scsi bus. terminators must match the impedance seen by a signal at the end of the scsi bus to the characteristic impedance of the scsi bus. this impedance is typically 84 +/C 12 w , but can vary greatly with pc board characteristics and cabling. v in v out low dropout voltage regulator v adj c1 10 m f alum. or 4.7 m f tant. 15 v r1 121 w 1% 1/4w r2 154 w 1% 1/4w c2 150 m f alum. or 22 ? tant. 10 v c3 .1 m f ceramic 25 v r3-20 are 110 w 1% r3-11 r12 r14 r15 r16 r17 r18 r19 r20 r13 termpwr db (0-7,p) atn bsy ack rst msg sel c/d req i/o 2.85 v 18681a-68 figure e-6. regulated termination amd p r e l i m i n a r y 130 AM79C974 termination there are three general termination schemes that apply to motherboard or host adapter setups when using regu- lated terminators. each scheme recognizes that the scsi bus must be terminated on both ends. therefore, as more components and peripherals are added to the bus, terminators must be relocated accordingly. each scheme is also based on an ideal routing situation, that is one where the internal peripherals, external peripher- als and the AM79C974 chip are connected by a linear scsi bus. see figure e-1. scheme #1 in this case, the system uses only scsi internal periph- erals. the regulated terminators on board should be ac- tivated. peripherals can be added to the bus by connecting them to a 50-pin ribbon cable. a scsi termi- nator should be attached to the last peripheral to termi- nate the other end of the bus. scheme #2 in this case, the system uses only external scsi periph- erals. as in scheme #1, the on-board regulated termina- tors should be activated. in this scheme, a 50-pin high density scsi-2 cable should connect the external port on the motherboard or host adapter to the first external peripheral. peripherals may be added with more cables and the last peripheral should be terminated. scheme #3 in this case, both internal and external scsi peripherals are used. the regulated terminators should be deacti- vated (since the AM79C974 will sit in the middle of the scsi bus). this may be accomplished through hard- ware or software. the hardware approach involves de- veloping a mechanism to detect peripherals connected to the scsi bus. this mechanism must then activate a signal to turn the regulated terminators on or off accord- ingly. the software approach involves having the user tell the system interactively that the regulated termina- tors should be turned on or off. care should be taken to ensure that each end of the bus is terminated. other considerations the following are motherboard considerations for rout- ing and layout. they should be taken into account along with scsi considerations. motherboards 1. high speed signals should be referenced only to the ground plane or exclusively to one of the power planes. if not, the power planes should be decoupled. 2. for a pci clk of 33 mhz, the maximum round trip time of any shared mother board signal should be 10 ns. 131 AM79C974 designing a single motherboard for amd pci family appendix f three devices in the amd pci family, the am53c974, the am79c970, and the am79c794, were designed with very similar pin assignments so that a single mothe- rboard design could be used for three different prod- ucts: one with a built-in scsi controller, one with a built-in ethernet controller, and one with both a scsi controller and an ethernet controller. this discussion describes some implementation details. pin out differences the table below shows the pins that are not the same on the three devices. all pins that are shown as nc are not bonded to the die inside the device package. these pins can be driven by any signal without affecting the opera- tion of the device. those pins that are shown in the pci family data sheets as reserved are connected to test logic on the die. these pins are active only during stand- alone chip testing, and therefore they may also be con- nected to signals on the board without affecting the operation of the device. on the other hand, pin 116 which is shown as reserved-dnc (for do not con- nect) must not be connected to anything on the board. am79c970 (ethernet) am53c974 (scsi) AM79C974 (combination) pin numbers function function function 9 reserved idsel idsela 10 idsel nc idselb 58 reserved pwdn pwdn 60 nc scsiclk1 scsiclk 68, 69, 70, 71, 73, nc scsi data & parity bus scsi data & parity bus 74, 75, 77, 78 64, 65, 66, 80, 81, nc scsi control lines scsi control lines 83, 85, 86, 87 89, 90, 92, 93, 94, 95 10base-t lines nc 10base-t lines 97 xtal1 scsiclk2 scsiclk 99 xtal2 nc xtal2 101, 102, 104, 105, 106, 107 aui lines nc aui lines 110, 111, 112, 114 eeprom interface & nc eeprom interface & led pins led pins 115 sleep nc sleep 116 reserved_dnc reserved_dnc reserved_dnc 117 inta inta inta 118 nc nc intb 126 req nc reqb 127 reserved req reqa 123 gnt nc gntb 124 nc gnt gnta intx (pins 117 and 118) the am79c970 and am53c974 devices have only one interrupt output, which is connected to pin 117. the AM79C974, on the other hand, has two interrupt out- puts: inta , which is used for scsi interrupts, is connected to pin 117; while intb , which is used for ethernet interrupts, is connected to pin 118. the motherboard can connect these pins directly to dedi- cated inputs to its interrupt controller, or it can connect amd p r e l i m i n a r y 132 AM79C974 them to multiplex logic that can map these outputs to in- terrupt controller inputs by means of software. req and gnt (pins 123, 124, 126, and 127) the am53c974 uses pins 127 and 124 ( req and gnt ) for bus master arbitration, while the am79c790 uses pins 126 and 123. the AM79C974 uses pins 127 and 124 ( reqa and gnta ) for bus arbitration from the scsi controller, and it uses pins 126 and 123 ( reqb and gntb ) for arbitration for the ethernet controller. the motherboard can connect both pairs of signals to its bus arbitration logic through jumpers or zero ohm series re- sistors. one pair of resistors or jumpers would be omit- ted for a scsi only product, and the other pair would be omitted for an ethernet only product. both pairs would be inserted for a product with both scsi and ethernet. if the motherboard uses a pci controller that allows a limited number of bus masters, you may have to omit one pci slot or define one slot to be slave only for a prod- uct with both scsi and ethernet on the motherboard. idsel (pins 9 and 10) the am53c974 uses pin 9 for idsel, which is a chip se- lect for the pci configuration space, while the am79c790 uses pin 10 for idsel. the AM79C974, which has two separate configuration spaces, uses pin 9 (idsela) to access the configuration space for the scsi controller and pin 10 (idselb) for the ethernet controller. idsel is used only during configuration cy- cles, which are defined by the command on the c/ be lines during the address phase. a typical way to generate the idsel signals on a motherboard is to connect a separate address line to each of the idsel pins of the various pci devices and expansion slots. since a typical pci motherboard will not have more than 10 pci devices and slots, two of the high order address lines can be connected directly to pins 9 and 10 of the amd device. the am79c970 will respond only to configuration cycles when pin 10 is ac- tive and will ignore configuration cycles when pin 9 is ac- tive. the am53c794 will respond only to configuration cycles when pin 9 is active. clocks (pins 60, 97, and 99) the am53c974 requires two clock inputs: scsiclk1 (pin 60) and scsiclk2 (pin 97). these clock inputs can be driven by the same clock source. the am79c790 re- quires either a crystal connected to the xtal1 and xtal2 pins (97 and 99) or an external oscillator con- nected to xtal1. the AM79C974 requires both a clock input for scsiclk (pin 60) and either a crystal con- nected to the xtal1 and xtal2 pins (97 and 99) or an external oscillator connected to xtal1. to satisfy all of these requirements, the motherboard can connect pin 60 to pin 97 through a jumper or zero ohm resistor. for the scsi only product, the oscillator or crystal should be omitted and the jumper inserted. for the ethernet-only product the oscillator or crystal should be inserted and the jumper omitted. it does not matter whether or not the scsi oscillator is included. for the combination scsi plus ethernet product, the crystal and the scsi oscillator should be included, and the jumper must be omitted. figure f-1 illustrates these connections. amd p r e l i m i n a r y 133 AM79C974 99 xtal2 97 xtal1/scsiclk2 60 scsiclk1 127 reqa 124 gnta 126 reqb 123 gntb 9 idsela 10 idselb 117 inta 118 intb am79c970, am53c974, or AM79C974 crystal scsi osc. bus arbiter mux mux control irq15 irq1 irq2 . . . adx ady 18681a-69 figure f-1. pci family connections amd p r e l i m i n a r y 134 AM79C974 trademarks copyright ? 1994 advanced micro devices, inc. all rights reserved. amd, the amd logo, and am386 are registered trademarks of advanced micro devices, inc. glitch eater, pcnet, pc scsi , himib, mace, ilacc, imr+, and am486 are trademarks of advanced micro devices, inc. product names used in this publication are for identification purposes only and may be trademarks of their respective companies. for information on additional scsi software products contact: sequoia advanced technologies, inc. (415) 459-7978 (415) 459-7988 fax 71322, 1020 compuserve id publication# 18681 rev. a amendment /1 issue date: april 1994 advanced micro devices AM79C974 pcnet tm -scsi combination ethernet and scsi controller for pci systems amendment this amendment corrects a few minor inaccuracies and adds or clarifies a few sections. minor corrections should be made on the existing data sheets. however, for ease of use, pages with more than a word or two of corrections are printed with this amendment. details: page 12 (reprinted as page 5 of this amendment) connection diagram change the names of the following power pins: change from v dd3b to v ddb . change from v ss3b to v ssb . change from v ddb to v ddbs . change from v ssb to v ssbs . pages 14C15 (reprinted as pages 6, 7 of this amendment) pin designations, listed by pin number and pin name change the names of the following power pins: change from v dd3b to v ddb . change from v ss3b to v ssb . change from v ddb to v ddbs . change from v ssb to v ssbs . page 17 (reprinted as page 8 of this amendment) pin designations, quick reference pin description the number of each type of power pin is incorrect. the numbers should be: change from v dd to v dd /dv dd , digital power, 5 pins change from v ddb /v dd3b to v ddb /v ddbs , i/o buffer power, 5 pins change from v ss to v ss /dv ss , digital ground, 9 pins change from v ssb /v ss3b to v ssb /v ssbs , i/o buffer ground, 11 pins pin designations, listed by driver type change the i oh value for both ts3 and ts6 from C0.4 ma to C2.0 ma . amd a m e n d m e n t 2 AM79C974 page 19 (reprinted as page 9 of this amendment) pin description, gnta add after the second paragraph: the AM79C974 supports bus parking. when the pci bus is idle and the system arbiter asserts gnta without an active reqa from the AM79C974 controller, the am79c794 will actively drive the ad[31:00], c/ be [3:0], and par lines. page 20 note that inta and intb are both open drain pins. page 24 note that atn is open drain. page 25 the number of each type of power pin is incorrect. the numbers should be: v dd /dv dd , digital power, 5 pins v ddb /v ddbs , i/o buffer power, 5 pins v ss /dv ss digital ground, 9 pins v ssb /v ssbs , i/o buffer ground, 11 pins page 30 the second sentence should read, it is a single cycle, non-burst 8-bit, 16-bit, or 32-bit transfer which is initiated by the host cpu. (the original omitted 8-bit). page 32 (reprinted as page 10 of this amendment) in figure 6, data on the ad lines should be driven during clock 6 and the second instance of par should be driven during clock 7. in each case this is one cycle earlier than what is shown in the figure. page 35 line 2, change type 15 to type 14 . page 39 (reprinted as page 11 of this amendment) replace the last two sentences with the following: for scsi, when target aborts, inta will not be asserted, but the abort bit (bit 2 of the dma status register at offset 54h) is set. for either ethernet or scsi a target abort causes rtabort (bit 12) of the status register in the appropriate pci configuration space to be set. (the original stated that inta will be asserted.) page 42 (reprinted as page 12 of this amendment) replace the last sentence with the following: for scsi, when the master aborts, inta will not be asserted, but the abort bit (bit 2 of the dma status register at offset 54h) is set. for either ethernet or scsi a master abort causes rmabort (bit 13) of the status register in the appropriate pci configuration space to be set. (the original stated that inta will be asserted.) amd a m e n d m e n t 3 AM79C974 page 44 change sprinten to lappen in lines 12 and 13. page 54 line 17, change 100% to 10% . page 59 line 12 should read, when the AM79C974 controller samples its idsela or idselb input ... (the original ommitted idsela). page 62 column 2, line 30, change lower two bytes to upper two bytes . page 71 line 15, change drcbc to drcvpa . page 76 (reprinted as pages 13 and 14 of the amendment) dma blast command section has been extensively revised. page 77 (reprinted as page 15 of this amendment) the description of the cmd1C0 bits at the end of column 2 should read, these two bits are encoded to represent four commands: idle, blast, start, and abort. (the blast command was omitted in the original.) in the table at the bottom, the description of the idle command should read, resets the dma block to the idle state. stops any current transfer. does not affect status bits or cause an interrupt. pages 78C80 (reprinted as pages 16C18 of this amendment) dma scatter-gather operation (4k aligned elements) section has been extensively revised. page 81 delete lines 19 and 20, which read, the dma transfer request was aborted (abort command, or the master or target abort). page 92 absolute maximum ratings line 4: add or v ssbs or dv ss to read supply voltage to av ss or v ssb or v ssbs or dv ss line 5: add v ddbs , dv dd to read (av dd , v dd , v ddb , v ddbs , dv dd ) page 92 operating ranges line 3: add v ddbs , dv dd to read (av dd , v dd , v ddb , v ddbs , dv dd ) after line 7: add or v ssbs C0.5 v v in v ddbs +0.5 v and or dv ss C0.5 v v in dv dd +0.5 v in the dc characteristics table, v ol (max) for those pins to which i ol1 or i ol2 applies is 0.55 v (to comply with the pci specification) rather than 0.45 v. also, v ol (max) for those pins to which i ol3 applies is 0.4 v rather than 0.45 v. amd a m e n d m e n t 4 AM79C974 page 94 in the dc characteristics table, delete row 5, v sol1 . change the parameter symbol for row 6 from v sol2 to v ol . add sd [7:0] and sd p to the parameter description for row 6. page 95 in the dc characteristics table, move pwdn from the list of pin names in row 1 as a separate item from pci input exceptions. the entry in row 1, column 3 should read, all scsi inputs. all ethernet inputs. pwdn. all pci inputs except idsel. page 113 in the ethernet pci configuration registers table, the default value for the interrupt pin (row 13, column 5) should be 02h, not 01h. page 129 (reprinted as page 19 of this amendment) figure e-5 decoupling capacitor placement change the names of the following power pins: change from v dd3b to v ddb . change from v ss3b to v ssb . change from v ddb to v ddbs . change from v ssb to v ssbs . amd a m e n d m e n t 5 AM79C974 connection diagram pin 1 is marked for orientation. reserve = dont connect. 18681a/1-3 132 1 ad28 ad29 131 v ssb 130 ad30 129 ad31 128 reqa 127 reqb 126 v ss 125 gnta 124 gntb 123 v dd 122 clk 121 rst 120 v ss 119 intb 118 inta 117 reserve 116 sleep 115 eecs 114 dv ss 113 eesk/ led1 112 eedi/ lnkst 111 eedo/ led3 110 dv dd 109 av dd2 108 ci+ 107 ci- 106 di+ 105 di- 104 av dd1 103 do+ 102 do- 101 av ss1 100 xtal2 99 av ss2 98 xtal1 97 av dd3 96 txd+ 95 txp+ 94 txd- 93 txp- 92 av dd4 91 rxd+ 90 rxd- 89 dv ss 88 i/o 87 c/d 86 msg 85 v dd 84 ack 83 v ssbs 82 req 81 sel 80 dv ss 79 sd p 78 sd 7 77 v ddbs 76 sd 6 75 sd 5 74 sd 4 73 v ssbs 72 sd 3 71 sd 2 70 sd 1 69 sd 0 68 v ssbs 67 34 par 35 c/ be 1 36 ad15 37 v ssb 38 ad14 39 ad13 40 ad12 41 ad11 42 ad10 43 v ssb 44 ad9 45 ad8 46 v ddb 47 c/ be 0 48 ad7 49 ad6 50 v ssb 51 ad5 52 ad4 53 ad3 54 ad2 55 v ssb 56 ad1 57 ad0 58 59 v dd 60 scsiclk 61 v ss 62 busy 63 v ss 64 bsy 65 atn 66 scsi^rst v ddb 2 ad27 3 ad26 4 v ssb 5 ad25 6 ad24 7 c/ be 3 8 v dd 9 idsela 10 idselb 11 v ss 12 ad23 13 ad22 14 v ssb 15 ad21 16 ad20 17 v ddb 18 ad19 19 ad18 20 v ssb 21 ad17 22 ad16 23 c/ be 2 24 frame 25 irdy 26 trdy 27 devsel 28 stop 29 lock 30 v ss 31 perr 32 serr 33 v ddb pwdn AM79C974 pcnet-scsi amd a m e n d m e n t 6 AM79C974 pin designations listed by pin number pin no. pin name pin no. pin name pin no. pin name pin no. pin name 1v ddb 34 par 67 v ssbs 100 av ss1 2 ad27 35 c/ be 168 sd 0 101 doC 3 ad26 36 ad15 69 sd 1 102 do+ 4v ssb 37 v ssb 70 sd 2 103 av dd1 5 ad25 38 ad14 71 sd 3 104 diC 6 ad24 39 ad13 72 v ssbs 105 di+ 7c/ be 3 40 ad12 73 sd 4 106 ciC 8v dd 41 ad11 74 sd 5 107 ci+ 9 idsela 42 ad10 75 sd 6 108 av dd2 10 idselb 43 v ssb 76 v ddbs 109 dv dd 11 v ss 44 ad9 77 sd 7 110 eedo/ led3 12 ad23 45 ad8 78 sd p 111 eedi/ lnkst 13 ad22 46 v ddb 79 dv ss 112 eesk/ led1 14 v ssb 47 c/ be 080 sel 113 dv ss 15 ad21 48 ad7 81 req 114 eecs 16 ad20 49 ad6 82 v ssbs 115 sleep 17 v ddb 50 v ssb 83 ack 116 reserve 18 ad19 51 ad5 84 dv dd 117 inta 19 ad18 52 ad4 85 msg 118 intb 20 v ssb 53 ad3 86 c/d 119 v ss 21 ad17 54 ad2 87 i/o 120 rst 22 ad16 55 v ssb 88 dv ss 121 clk 23 c/ be 2 56 ad1 89 rxdC 122 v dd 24 frame 57 ad0 90 rxd+ 123 gntb 25 irdy 58 pwdn 91 av dd4 124 gnta 26 trdy 59 v dd 92 txpC 125 v ss 27 devsel 60 scsiclk 93 txdC 126 reqb 28 stop 61 v ss 94 txp+ 127 reqa 29 lock 62 busy 95 txd+ 128 ad31 30 v ss 63 v ss 96 av dd3 129 ad30 31 perr 64 bsy 97 xtal1 130 v ssb 32 serr 65 atn 98 av ss2 131 ad29 33 v ddb 66 scsi^rst 99 xtal2 132 ad28 amd a m e n d m e n t 7 AM79C974 pin designations listed by pin name pin name pin no. pin name pin no. pin name pin no. pin name pin no. ack 83 atn 65 gntb 123 stop 28 ad0 57 av dd1 103 idsela 9 trdy 26 ad1 56 av dd2 108 idsel 10 xtal1 97 ad2 54 av dd3 96 inta 117 xtal2 99 ad3 53 av dd4 91 intb 118 txdC 93 ad4 52 av ss1 100 i/o 87 txd+ 95 ad5 51 av ss2 98 irdy 25 txpC 92 ad6 49 bsy 64 lock 29 txp+ 94 ad7 48 busy 62 msg 85 v dd 8 ad8 45 c/ be 0 47 par 34 v dd 59 ad9 44 c/ be 135 perr 31 v dd 122 ad10 42 c/ be 2 23 pwdn 58 v ddb 1 ad11 41 c/ be 37 req 81 v ddb 17 ad12 40 c/d 86 reqa 127 v ddb 33 ad13 39 clk 121 reqb 126 v ddb 46 ad14 38 ciC 106 reserve 116 v ddbs 76 ad15 36 ci+ 107 rst 120 v ss 11 ad16 22 devsel 27 rxdC 89 v ss 30 ad17 21 diC 104 rxd+ 90 v ss 61 ad18 19 di+ 105 scsiclk 60 v ss 63 ad19 18 doC 101 scsi^rst 66 v ss 119 ad20 16 do+ 102 sd 068 v ss 125 ad21 15 dv dd 84 sd 169 v ssb 4 ad22 13 dv dd 109 sd 270 v ssb 14 ad23 12 dv ss 79 sd 371 v ssb 20 ad24 6 dv ss 88 sd 473 v ssb 37 ad25 5 dv ss 113 sd 574 v ssb 43 ad26 3 eecs 114 sd 675 v ssb 50 ad27 2 eedi/ lnkst 111 sd 777 v ssb 55 ad28 132 eedo/ led3 110 sd p78 v ssb 130 ad29 131 eesk/ led1 112 sel 80 v ssbs 67 ad30 129 frame 24 serr 32 v ssbs 72 ad31 128 gnta 124 sleep 115 v ssbs 82 amd a m e n d m e n t 8 AM79C974 pin designations (continued) quick reference pin description pin name description type driver # pins scsi specific scsi interface sd [7:0] scsi data io od48 8 sd p scsi data parity io od48 1 msg message i 1 c/d command/data i 1 i/o input/output i 1 atn attention o od48 1 bsy busy io od48 1 sel select io od48 1 scsi^rst scsi bus reset io od48 1 req request i 1 ack acknowledge o od48 1 miscellaneous scsi clk scsi core clock i 1 reserve reserved, do not connect i 1 power management pwdn power down indicator i 1 test interface busy nand tree test output o o3 1 power supplies av dd analog power p na 4 av ss analog ground p na 2 v dd , dv dd digital power p na 5 v ss , dv ss digital ground p na 9 v ddb , v ddbs i/o buffer power p na 5 v ssb , v ssbs i/o buffer ground p na 11 listed by driver type the following table describes the various types of drivers that are implemented in the pcnet-scsi controller. current is given as milliamperes: name type i ol (ma) i oh (ma) pf ts3 tri-state tm 3 C2.0 50 ts6 tri-state 6 C2.0 50 o3 totem pole 3 C0.4 50 o6 totem pole 6 C0.4 50 o8 totem pole 8 C0.4 50 od6 open drain 6 na 50 od48 open drain 48 na led led 12 C0.4 50 amd a m e n d m e n t 9 AM79C974 pin description pci bus interface ad[31:00] address and data input/output, active high these signals are multiplexed on the same pci pins. during the first clock of a transaction ad[31:00] contain the physical byte address (32 bits). during the subse- quent clocks ad[31:00] contain data. byte ordering is lit- tle endian by default. ad[07:00] are defined as least significant byte and ad[31:24] are defined as the most significant byte. for fifo data transfers, the pcnet- scsi controller can be programmed for big endian byte ordering. see csr3, bit 2 (bswp) for more details. during the address phase of the transaction, when the pcnet-scsi controller is a bus master, ad[31:2] will ad- dress the active dword (double-word). the pcnet- scsi controller always drives ad[1:0] to 00 during the address phase indicating linear burst order. when the pcnet-scsi controller is not a bus master, the ad[31:00] lines are continuously monitored to deter- mine if an address match exists for i/o slave transfers. during the data phase of the transaction, ad[31:00] are driven by the pcnet-scsi controller when performing bus master writes and slave read operations. data on ad[31:00] is latched by the pcnet-scsi controller when performing bus master reads and slave write operations. when rst is active, ad[31:0] are inputs for nand tree testing. c/ be [3:0] bus command and byte enables input/output, active low these signals are multiplexed on the same pci pins. during the address phase of the transaction, c/ be [3:0] define the bus command. during the data phase c/ be [3:0] are used as byte enables. the byte enables define which physical byte lanes carry meaningful data. c/ be 0 applies to byte 0 (ad[07:00]) and c/ be 3 applies to byte 3 (ad[31:24]). the function of the byte enables is independent of the byte ordering mode (csr3, bit 2). when rst is active, c/ be [3:0] are inputs for nand tree testing. clk clock input this signal provides timing for all the transactions on the pci bus and all pci devices on the bus including the pcnet-scsi controller. all bus signals are sampled on the rising edge of clk and all parameters are defined with respect to this edge. the pcnet-scsi controller op- erates over a range of 0 to 33 mhz. when rst is active, clk is an input for nand tree testing. devsel device select input/output, active low this signal when actively driven by the pcnet-scsi controller as a slave device signals to the master device that the pcnet-scsi controller has decoded its address as the target of the current access. as an input it indi- cates whether any device on the bus has been selected. when rst is active, devsel is an input for nand tree testing. frame cycle frame input/output, active low this signal is driven by the pcnet-scsi controller when it is the bus master to indicate the beginning and dura- tion of the access. frame is asserted to indicate a bus transaction is beginning. frame is asserted while data transfers continue. frame is deasserted when the transaction is in the final data phase. when rst is active, frame is an input for nand tree testing. gnta bus grant input, active low this signal indicates that the access to the bus has been granted to the AM79C974s scsi controller. the AM79C974 supports bus parking. when the pci bus is idle and the system arbiter asserts gnta without an active reqa from the AM79C974 controller, the AM79C974 will actively drive the ad[31:00], c/ be [3:0], and par lines. when rst is active, gnta is an input for nand tree testing. amd a m e n d m e n t 10 AM79C974 bus master dma transfers there are four primary types of dma transfers. the AM79C974 controller uses non-burst as well as burst cycles for read and write access to the main memory. basic non-burst read cycles all AM79C974 controller non-burst read accesses are of the pci command type memory read (type 6). note that during all non-burst read operations, the AM79C974 controller will always activate all byte enables, even though some byte lanes may not contain valid data as indicated by a buffer pointer value. in such instances, the AM79C974 controller will internally discard un- needed bytes. figure 6 shows a typical non-burst read access. the AM79C974 controller asserts irdy at clock 5 immedi- ately after the address phase and starts sampling devsel . the target extends the cycle by asserting devsel not until clock 6. additionally, the target inserts one wait state by asserting its ready ( trdy ) at clock 8. 18681a/1-10 frame clk ad irdy trdy c/ be devsel req gnt 12345678 0000 9 par par par data devsel is sampled by the AM79C974 controller. addr 0110 figure 6. non-burst read cycles with wait states amd a m e n d m e n t 11 AM79C974 target abort figure 13 shows a target abort sequence. the target as- serts devsel for one clock. it then deasserts devsel and asserts stop on clock 4. a target can use the target abort sequence to indicate that it cannot service the data transfer and that it does not want the transaction to be retried. additionally, the AM79C974 controller cannot make any assumption about the success of the previous data transfers in the current transaction. the AM79C974 controller terminates the current transfer with the deassertion of frame on clock 5 and one clock cycle later with the deassertion of irdy . it finally re- leases the bus on clock 6. since data integrity is not guaranteed, the AM79C974 controller cannot recover from a target abort event. for ethernet, the AM79C974 controller will reset all csr and bcr locations to their h_reset values. any on-going network activity will be stopped immediately. the pci configuration registers will not be cleared. for scsi, when target aborts, inta will not be asserted, but the abort bit (bit 2 of the dma status register at offset 54h), is set. for either ethernet or scsi a target abort causes rtabort (bit 12) of the status register in the appropriate pci configuration space to be set. 18681a/1-17 frame clk ad irdy trdy c/ be devsel req gnt 123456 data 0000 0111 par par par devsel is sampled by the AM79C974 controller. stop addr figure 13. target abort amd a m e n d m e n t 12 AM79C974 master abort the AM79C974 controller will terminate its cycle with a master abort sequence if devsel is not asserted within 4 clocks after frame is asserted. master abort is treated as a fatal error by the AM79C974 controller. for the ethernet, the AM79C974 controller will reset all csr and bcr locations to their h_reset values. any on-going network activity will be stopped immediately. the pci configuration registers will not be cleared. for scsi, when the master aborts, inta will not be as- serted, but the abort bit (bit 2 of the dma status regis- ter at offset 54h), is set. for either ethernet or scsi a master abort causes rmabort (bit 13) of the status register in the appropriate pci configuration space to be set. 18681a/1-20 frame clk ad irdy trdy c/ be devsel req gnt 12345678 addr data 0000 0111 10 9 par par par devsel is sampled by the AM79C974 controller. figure 16. master abort amd a m e n d m e n t 13 AM79C974 since the pci bus is 4 bytes wide and the scsi bus is only 1 byte wide, funneling logic is included in this en- gine to handle byte alignment and to ensure that data is properly transferred between the scsi bus and the wider pci bus. all boundary conditions are handled through hardware by the dma engine. the dma engine is also designed for block type (4 kbyte page) transfers to support scatter-gather opera- tions. implementation of this feature is described further in the dma scatter-gather mechanism section. dma fifo data transfers from the scsi fifo to the dma fifo take place each time the threshold of two bytes is reached on the scsi side. the transfer is initiated by the scsi block when the internal dreq is asserted, and continues with the dack handshaking which typically takes place in dma accesses. data is accumulated in the dma fifo until a threshold of 16 dword (64 bytes) is reached. data is then burst across the pci bus to memory. residue data which is less than the threshold in each fifo is sent in non-contiguous bursts. for mem- ory read operations, data is sent in burst mode to the dma fifo and continues through to the scsi fifo and onto the scsi bus. dma blast command this command is used to retrieve the contents of the dma fifo when the target disconnects during a dma write operation. this could happen for example if a scsi disk drive detected the end of a sector and de- cided to give up the bus while it was looking for the next sector. the target disconnect can leave some bytes of data in the dma fifo and some in the scsi fifo, while some bytes have yet to be transferred from the periph- eral device. when this happens, the controller will assert inta to interrupt the processor, the scsi state machine will continue to empty its contents into the dma fifo, but the dma fifo will not necessarily dump its contents into memory (unless the 64-byte dma threshold hap- pens to have been exceeded at this time). the blast command causes the contents of the dma fifo to be emptied into memory. there are some re- strictions on when this command should be used. first, the command should be used only to recover from an interrupted dma write operationnot a read operation. second, the command must not be issued until the scsi fifo has finished dumping its contents into the dma fifo. third, the command should never be issued when the dma fifo has already been emptied. this is indicated by the state of the done bit in the dma status register at ((b)+54h). (this is a test for a special case that can occur when a target disconnect leaves only 1 byte left to be trans- ferred from the scsi peripheral. in this case, if the origi- nal transfer count was even, an odd number of bytes will be left in the scsi fifo. since the scsi engine trans- fers data to the dma fifo two bytes at a time, the last transfer consists of one byte of valid data and one byte of garbage. the dma engine treats this final invalid byte as valid data and writes it to memory. when it does this, it decrements its working byte counter to zero and sets the done bit, even though 1 byte still needs to be re- trieved from the peripheral device.) the following procedure outlines the use of the blast command after an interrupt has occurred. note that the order of steps 2C4 is not critical. the order can be changed to tune the performance. also note that each register is read only once in this procedure even though several tests may be made on data from one register. 1. verify that int (bit 7 of the scsi status register at ((b)+10h) is set to indicate that a scsi interrupt is pending. 2. read the scsi current fifo count (bits 4:0 of the current fifo/internal state register at ((b)+1ch). if this value is not zero, wait for the scsi fifo to empty its contents into the dma fifo. 3. if bit 4 of the scsi status register (ctz) is set, stop here. the transfer is complete, and it is not necessary to execute the blast command. 4. verify that dir (bit 7 of dma command register at ((b)+40h) is set to one to indicate that the direction of transfer is from scsi to memory. 5. test the error bits in the dma status register and the scsi status register (statreg at ((b)+10h) to verify that the contents of the dma and scsi fifos are not invalid. 6. test the dma done bit in the dma status register. if done is not set, write 01 to the dma command register to issue the blast command. this will move the remaining data from the dma fifo into memory. 7. wait until the blast complete (bcmplt) in the dma status register is set to indicate the com- pletion of the blast operation. 8. write 00 to the dma command register to issue the idle command to the dma engine. (note that the idle command does not generate an interrupt.) the above procedure insures that the data that has been transferred from the scsi peripheral does not get lost in the dma fifo when a target disconnect occurs. however, it does not complete the original transfer. the software must now read the scsi current transfer count register (ctcreg) to find out how many bytes amd a m e n d m e n t 14 AM79C974 have yet to be transferred from the scsi peripheral de- vice and must start a new transfer operation to get the rest of the data. (ctcreg consists of three bytes lo- cated at ((b)+00h, (b)+04h, and (b)+38h.) funneling logic figure 26 shows the internal dma logic interface with the scsi block. the dma fifo interfaces to the funnel logic block via a 32-bit data bus, and the funnel logic properly reduces this stream of data to a 16-bit stream to properly interface with the scsi fifo. 18681a/1-30 figure 26. dma fifo to scsi fifo interface scsi dma programming sequence the following section outlines the procedure for execut- ing scsi dma operations: 1. issue idle command to the dma engine 2. configure the scsi block registers (e.g. synchro- nous operation, offset values, etc.) 3. program the dma registers to set up address and transfer count 4. issue a transfer command to the scsi command registers 5. issue the start command to the dma engine 6. at the end of the dma transaction, issue the idle command to the dma engine mdl based dma programming the following section outlines the procedure for execut- ing mdl based dma operation: 1. set up the mdl list 2. use the programming sequence defined earlier for initiating a scsi dma transfer amd a m e n d m e n t 15 AM79C974 dma registers the following is a summary of the dma register set or the dma channel context block (dma ccb). these registers control the specifics for dma operations such as transfer length and scatter-gather options. the three read-only working counter registers allow the system software and driver to monitor the dma transaction. each register address is represented by the pci con- figuration base address (b) and its corresponding offset value. the base address for the AM79C974 is stored at register address (10h) in the pci configuration space. table 6. the dma registers register acronym addr (hex) register description type cmd (b)+40 command (bits 31:8 reserved, bits 7:0 used) r/w stc (b)+44 starting transfer count (bits 31:24 reserved, bits 23:0 used) r/w spa (b)+48 starting physical address (bits 31:0 used) r/w wbc (b)+4c working byte counter r wac (b)+50 working address counter (bits 31:0 used) r status (b)+54 status register (bits 31:8 reserved, bits 7:0 used) r smdla (b)+58 starting memory descriptor list (mdl) address r/w wmac (b)+5c working mdl counter r command register (cmd) the upper 3 bytes of command register are reserved, the remaining (lsb) byte is defined as follows: address (b)+40h, lsb read/write dir inte_d inte_p mdl reserved reserved cmd1 cmd0 76543210 dir: data transfer direction bit. inte_d: dma transfer active interrupt bit. inte_p: page transfer active interrupt bit. mdl: memory descriptor list (mdl) spa enable bit. reserved: reserved for future expansion. the zero value must be written in these bits. cmd1-0: these two bits are encoded to represent four com- mands: idle, blast, start, and abort. cmd1 cmd0 command description 0 0 idle resets the dma block to the idle state. stops any current transfer. does not affect status bits or cause an interrupt. 0 1 blast empties all data bytes in dma fifo to memory during a dma write operation. upon completion, the bcmplt bit will be set in the dma status register. this command should not be used during a dma read operation. 1 0 abort terminates the current dma transfer. the dma engine should be restored to the idle state following execution of this command. note : this is only valid after a start command is issued. 1 1 start initiates a new dma transfer. these bits must remain set throughout the dma operation until the done bit in the dma status register is set. note: this command should be issued only after all other control bits have been initialized. amd a m e n d m e n t 16 AM79C974 dma scatter-gather mechanism the AM79C974 contains a scatter-gather translation mechanism which facilitates faster data transfers. this feature uses a m emory d escriptor l ist (a list of contigu- ous physical memory addresses) which is stored in sys- tem memory. use of the memory descriptor list allows a single scsi transfer to be read from (or written to) non- contiguous physical memory locations. this mecha- nism avoids copying the transfer data and mdl list, which was previously required for conventional dma operations. memory descriptor list (mdl) the mdl is a non-terminated (no end of file marker) list of 32-bit page frame addresses, which is always aligned on a double word boundary. the format is shown below: 31 12 11 0 page frame address ignored dma scatter-gather operation (4k aligned elements) the scatter-gather mechanism described below assumes 4k page alignment and size for all mdl entries except the first and last entry. this feature is enabled by setting the mdl bit in the dma command register (bit 4, address (b)+40h). 1. a) prepare the memory descriptor list (mdl) through software and store it in system memory. b) load the address of the starting entry in the memory descriptor list (mdl) into the start memory descriptor list address (smdla) register. this value is automatically copied into the working mdl address counter (wmac). c) program the starting transfer count (stc) register with the total transfer length (i.e., # of bytes). also program the starting physical address (spa) register (bits 11:0) with the starting offset of the first entry. note : the value in the smdla register must be double word aligned. therefore, read/write transactions will al- ways begin on a double word boundary. 18681a/1-31 0 31 0 31 smdla wmac 0 12 31 mdl page frame address #1 page frame address #2 page frame address #3 page frame address #4 ignored page frame address #n ignored ignored ignored ignored amd a m e n d m e n t 17 AM79C974 in this example, the contents of the wmac register is pointing to page frame address #1. when the first entry in the mdl in read (page frame address #1), the wmac register is incremented to point to the next page entry (page frame address #2). 2. issue the start dma command. the AM79C974 reads the page frame address (bits 31:12) from the mdl entry and combines it with the first page off- set value in the starting physical address (spa) register (bits 11:0). this 32-bit value is loaded into the working address counter (wac) register and becomes the physical address for page#1, as shown below. the wmac is then incremented to point to the next entry in the mdl. 0 12 31 wac 4k page #1 page frame address #1 starting offset from the mdl 0 12 31 spa xxxx starting offset programmed by the software data 18681a/1-32 3. when the wac register (bits 11:0) reaches the first 4k byte boundary, the AM79C974 reads the second mdl entry and combines the page frame address (bits 31:12) from this entry of the mdl with bits 11:00 of the wac register. this becomes the physical address for page#2. since the wac register (bits 11:0) has rolled over to 00h, the wac now points to the beginning of the page frame address #2 as shown below. the wmac is then incremented to point to the next entry in the mdl. 18681a/1-33 0 12 31 wac 4k page #2 0 page frame address #2 from the mdl data amd a m e n d m e n t 18 AM79C974 when wac (bits 11:0) again reaches the next 4k byte boundary, the next mdl entry is read into the wac. the operation continues in this way until wmac register reaches the last mdl entry (page frame address #n in this example). 4. the wac register points to the beginning of the last page#n and the dma operation continues until the byte count is exhausted in the working byte counter (wbc) register. when wbc=0, the chip stops incrementing the wac register. this is shown below. 18681a/1-34 0 12 31 wac 4k page #n page frame address #n from the mdl wbc = 0 0 data dma scatter-gather operation (non-4k aligned elements mdl not set) there is another way to implement a scatter-gather op- eration which does not force the data elements to be aligned on 4k boundaries. it assumes a traditional scatter-gather list of the following format: element 0 physical address byte count element 1 physical address byte count ... element n physical address byte count this second implementation is described as follows: 1. set the scsi start transfer count register ((b)+00h, (b)+04h, (b)+38h) to the byte count of the first scatter-gather element. 2. program the dma starting transfer count regis- ter ((b)+44h) to the byte count of the first scatter- gather element. 3. program the dma starting physical address reg- ister ((b)+48h) to the physical address of the first scatter-gather element. 4. start the scsi operation by issuing a scsi infor- mation transfer command. 5. start the dma engine with dma transfer interrupt enable (bit 6, (b)+40h). 6. when the scatter-gather elements byte count is exhausted, the dma engine will generate an interrupt. 7. reprogram the next scatter-gather elements byte count into the scsi start transfer count register and the dma starting transfer count register. 8. reprogram the dma starting physical address register ((b)+48h) to the physical address of the next scatter-gather element. 9. repeat steps 4C8 until the scatter-gather list is completed. amd a m e n d m e n t 19 AM79C974 where c1 C c9 are decoupling capacitors. 18681a/1-67 figure e-5. decoupling capacitor placement |
Price & Availability of AM79C974
 |
|
|
|
|
All Rights Reserved © IC-ON-LINE 2003 - 2022 |
| [Add Bookmark] [Contact Us] [Link exchange] [Privacy policy] |
|
Mirror Sites : [www.datasheet.hk]
[www.maxim4u.com] [www.ic-on-line.cn]
[www.ic-on-line.com] [www.ic-on-line.net]
[www.alldatasheet.com.cn]
[www.gdcy.com]
[www.gdcy.net] |