PRELIMINARY Am79C90 CMOS Local Area Network Controller for Ethernet (C-LANCE) DISTINCTIVE CHARACTERISTICS ■ Compatible with Ethernet and IEEE 802.3 10BASE-5 Type A, and 10BASE-2 Type B, “Cheapernet,” 10BASE-T ■ Easily interfaced with 80x86, 680x0, Am29000, Z8000™ microprocessors ■ On-board DMA and buffer management, 64-byte Receive, 48-byte Transmit FIFOs ■ 24-bit-wide linear addressing (Bus Master Mode) ■ Network and packet error reporting ■ Back-to-back packet reception with as little as 0.5 µs interframe spacing ■ Diagnostic Routines — Internal/external loopback — CRC logic check — Time domain reflectometer ■ Low power consumption for power-sensitive applications ■ Completely software- and hardware-compatible with AMD’s LANCE device (Am7990) (see Appendix A) GENERAL DESCRIPTION The Am79C90 CMOS Local Area Network Controller for Ethernet (C-LANCE) is a 48-pin VLSI device designed to greatly simplify interfacing a microcomputer or minicomputer to an IEEE 802.3/Ethernet Local Area Network. The C-LANCE, in conjunction with the Am7992B Serial Interface Adapter (SIA), Am7996 or Am79C98 and Am79C100 Transceiver, and closely coupled local memory and microprocessor, is intended to provide the user with a complete interface module for an Ethernet network. The Am79C90 is designed using a scalable CMOS technology and is compatible with a variety of microprocessors. On-board DMA, advanced buffer management, and extensive error reporting and diagnostics facilitate design and improve system performance. BLOCK DIAGRAM Local CPU Interface A23:16 INTR HOLD HLDA ALE/AS CS ADR DAS DALO DALI READ BM0/BYTE BM1/BUSAKO READY RESET Parallel Bus Interface DMA/Data Path Control C-LANCE/ CPU Control Bus Interface Station Address Detection Retry Logic Microprogram Store Serial I/O Interface RX RCLK TX TCLK CLSN TENA RENA Am7992B SIA Interface DAL15:0 17881C-1 Publication# 17881 Rev: C Amendment/0 Issue Date: January 1998 This document contains information on a product under development at Advanced Micro Devices. The information is intended to help you evaluate this product. AMD reserves the right to change or discontinue work on this proposed product without notice. 1 AMD PRELIMINARY RELATED AMD PRODUCTS Part No. Description Am7996 IEEE 802.3/Ethernet/Cheapernet Tap Transceiver Am79C100 Twisted-Pair Ethernet Transceiver Plus (TPEX+) Am79C900 Integrated Local Area Communications ControllerTM (ILACCTM) Am79C940 Media Access Controller for Ethernet (MACETM) 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 bus) Am79C974 PCnet-SCSI Combination Ethernet and SCSI Controller for PCI Systems Am79C98 Twisted-Pair Ethernet Transceiver (TPEX) Am79C981 Integrated Multiport Repeater PlusTM (IMR+TM) Am79C987 Hardware Implemented Management Information BaseTM (HIMIBTM) CONNECTION DIAGRAMS DAL11 DAL3 6 43 DAL12 DAL2 7 42 DAL13 DAL1 8 41 DAL14 DAL0 9 40 DAL15 READ 10 39 A16 INTR 11 38 A17 DALI 12 37 A18 DALO 13 36 A19 DAS 14 35 A20 BM0/BYTE 15 34 A21 BM1/BUSAKO HOLD/BUSRQ 16 33 A22 17 32 A23 ALE/AS 18 31 RX HLDA 19 30 RENA CS 20 29 TX ADR 21 28 CLSN READY 22 27 RCLK RESET 23 26 TENA VSS 24 25 TCLK 9 8 7 6 5 4 3 2 NC NC DAL2 DAL3 DAL4 DAL5 DAL6 DAL7 VSS VDD DAL8 DAL9 DAL10 DAL11 DAL12 NC NC NC 44 BM1/BUSRQ 5 HOLD/BUSRQ ALE/AS HLDA DAL4 1 68 67 66 65 64 63 62 61 10 11 12 13 14 15 16 17 18 19 20 21 60 59 58 57 56 55 54 53 52 51 50 49 22 48 23 47 24 46 25 45 26 44 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 A22 A23 RX RENA DAL10 NC 45 A21 4 NC DAL9 NC 46 A20 3 DAL5 NC DAS BM0/BYTE DAL6 A18 A19 NC DAL8 DALI DALO NC 47 DAL15 A16 A17 2 READ INTR VDD DAL7 DAL13 DAL14 48 NC 1 NC VSS DAL1 PLCC DAL0 DIP 17881B-3 Note: Pin 1 is marked for orientation. 2 17881B-2 Am79C90 NC CS NC NC ADR READY RESET NC NC VSS TCLK TENA RCLK CLSN TX NC NC P R E L I M I N A R Y TYPICAL ETHERNET/CHEAPERNET NODE MAU AUI Cable DTE Am7996 Transceiver Ethernet Local CPU TAP Am79C90 C-LANCE Local Memory Am7992B SIA Power Supply Local Bus Ethernet Coax DTE Cheapernet Local CPU Am79C90 C-LANCE Local Memory Am7996 Transceiver Am7992B SIA RG58A/U or RG58C/U BNC “T” Power Supply Local Bus Typical Ethernet 10BASE-T Node AUI Cable DTE Local CPU Local Memory Am79C90 C-LANCE Am7992B SIA MAU Am79C98 Am79C100 Twisted-Pair Power Supply Local Bus DTE Local CPU Local Memory Am79C90 C-LANCE Am7992B SIA Am79C98 Am79C100 Twisted-Pair Local Bus 17881C-4 AUI—Attachment Unit Interface DTE—Data Terminal Equipment Am79C90 MAU—Medium Attachment Unit 3 P R E L I M I N A R Y 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 the elements below. AM79C90 P C B OPTIONAL PROCESSING Blank = Standard Processing TR = Tape and Reel Packaging OPERATING CONDITIONS C = Commercial (0°C to +70°C) PACKAGE TYPE P = 48-Pin Plastic DIP (PD 048) J = 68-Pin Plastic Leaded Chip Carrier (PL 068) SPEED Not Applicable DEVICE NUMBER/DESCRIPTION Am79C90 CMOS Local Area Network Controller for Ethernet Valid Combinations Valid Combinations AM79C90 4 PC, JC, JCTR 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. Am79C90 P R E L I M I N A R Y PIN DESCRIPTION A16–A23 Byte selection using Byte Mask is done as described by the following table: High Order Address Bus (Output, Three-State) BM1 BM0 Selection Additional address bits to access a 24-bit address. These lines are driven as a Bus Master only. LOW LOW Whole Word LOW HIGH Upper Byte HlGH LOW Lower Byte HlGH HIGH None ADR Register Address Port Select (Input) When the C-LANCE is a Slave, ADR indicates which of the two register ports is selected. ADR LOW selects register data port; ADR HIGH selects register address port. ADR must be valid throughout the data portion of the bus cycle and is only used by the C-LANCE when CS is LOW. ALE/AS Address Latch Enable (Output, Three-State) Used to demultiplex the DAL lines and define the address portion of the bus cycle. This l/O pin is programmable through bit (01) of CSR3. BYTE, BUSAKO—If CSR3 (00) BCON = 1 PIN 15 = BYTE (Output, Three-State) (48-Pin DlPs) PIN 16 = BUSAKO (Output) (48-Pin DIPs) Byte selection may also be done using the BYTE line and DAL00 line, latched during the address portion of the bus cycle. The C-LANCE drives BYTE only as a Bus Master and ignores it when a Bus Slave selection is done (similar to BM0, BM1). Byte selection is done as outlined in the following table: As ALE (CSR3 (01), ACON = 0), the signal transitions from a HIGH to a LOW during the address portion of the transfer and remains LOW during the data portion. ALE can be used by a Slave device to control a latch on the bus address lines. When ALE is HIGH, the latch is open, and when ALE goes LOW, the latch is closed. As AS (CSR3 (01), ACON = 1), the signal pulses LOW during the address portion of the bus transaction. The LOW-to-HlGH transition of AS can be used by a Slave device to strobe the address into a register. The C-LANCE drives the ALE/AS line only as a Bus Master. BM0/BYTE, BM1/BUSAKO (Output, Three-State) The two pins are programmable through bit (00) of CSR3. BM0, BM1—If CSR3 (00) BCON = 0 PIN 15 = BM0 (Output, Three-State) (48-Pin DlPs) PIN 16 = BM1 (Output, Three-State) (48-Pin DlPs) BM0, BM1 (Byte Mask). This indicates that the byte(s) on the DAL are to be read or written during this bus transaction. The C-LANCE drives these lines only as a Bus Master. It ignores the Byte Mask lines when it is a Bus Slave and assumes word transfers. BYTE DAL00 LOW LOW Whole Word LOW HIGH Illegal Condition HlGH LOW Lower Byte HlGH HIGH Upper Byte Selection BUSAKO is a bus request daisy chain output. If the chip is not requesting the bus and it receives HLDA, BUSAKO will be driven LOW. If the C-LANCE is requesting the bus when it receives HLDA, BUSAKO will remain HIGH. Byte Swapping In order to be compatible with the variety of 16-bit microprocessors available to the designer, the C-LANCE may be programmed to swap the position of the upperand lower-order bytes on data involved in transfers with the internal FIFOs. Byte swapping is done when BSWP = 1. The most significant byte of the word in this case will appear on DAL lines 7–0 and the least significant byte on DAL lines 15–8. When BYTE = H (indicating a byte transfer) the table indicates on which part of the 16-bit data bus the actual data will appear. Whenever byte swap is activated, the only data that is swapped is data traveling to and from the Transmit/ Receive FIFOs. Am79C90 5 P R E L I M I N A R Y DAS Mode Bits BSWP = 0 and BCON = 1 BSWP = 1 and BCON = 1 BYTE = L and DAL00 = L Word Word BYTE = L and DAL00 = H Illegal Illegal BYTE = H and DAL00 = H Upper Byte Lower Byte BYTE = H and DAL00 = L Lower Byte Upper Byte Signal Line Data Strobe (Input/Output, Three-State) Defines the data portion of the bus transaction. DAS is high during the address portion of a bus transaction and low during the data portion. The LOW-to-HlGH transition can be used by a Slave device to strobe bus data into a register. DAS is driven only as a Bus Master. HLDA CLSN Collision (Input) A logical input that indicates that a collision is occurring on the channel. CS Bus Hold Acknowledge (Input) A response to HOLD. When HLDA is LOW in response to the chip’s assertion of HOLD, the chip is the Bus Master. During Bus Master operation, the C-LANCE waits for HLDA to be deasserted HIGH before reasserting HOLD LOW. This insures proper bus handshake under all situations. HOLD/BUSRQ Chip Select (Input) Indicates, when asserted, that the C-LANCE is the Slave device of the data transfer. CS must be valid throughout the data portion of the bus cycle. CS must not be asserted when HLDA is LOW. Bus Hold Request (Output, Open Drain) Asserted by the C-LANCE when it requires access to memory. HOLD is held LOW for the entire ensuing bus transaction. The function of this pin is programmed through bit (00) of CSR3. Bit (00) of CSR3 is cleared when RESET is asserted. DAL00–DAL15 When CSR3 (00) BCON = 0 Data/Address Lines (Input/Output, Three-State) The time multiplexed Address/Data bus. During the address portion of a memory transfer, DAL00–DAL15 contains the lower 16 bits of the memory address. The upper 8 bits of address are contained in A16–A23. PIN 17 = HOLD (Output Open Drain and input sense) (48-Pin DIPs) During the data portion of a memory transfer, DAL00– DAL15 contains the read or write data, depending on the type of transfer. If the C-LANCE wants to use the bus, it looks at HOLD/ BUSRQ; if it is HIGH the C-LANCE can pull it LOW and request the bus. If it is already LOW, the C-LANCE waits for it to go inactive-HlGH before requesting the bus. The C-LANCE drives these lines as a Bus Master and as a Bus Slave. PIN 17 = BUSRQ (I/O Sense, Open Drain) (48-Pin DlPs) INTR DALI Data/Address Line In (Output, Three-State) An external bus transceiver control line. DALI is asserted when the C-LANCE reads from the DAL lines. It will be LOW during the data portion of a READ transfer and remain HIGH for the entire transfer if it is a WRITE. DALI is driven only when C-LANCE is a Bus Master. Interrupt (Output, Open Drain) An attention signal that indicates, when active, that one or more of the following CSR0 status flags is set: BABL, MERR, MISS, RINT, TINT or IDON. INTR is enabled by bit 06 of CSR0 (INEA = 1). INTR remains asserted until the source of Interrupt is removed. RCLK DALO Data/Address Line Out (Output, Three-State) An external bus transceiver control line. DALO is asserted when the C-LANCE drives the DAL lines. DALO will be LOW only during the address portion if the transfer is a READ. It will be LOW for the entire transfer if the transfer is a WRITE. DALO is driven only when C-LANCE is a Bus Master. 6 When CSR3 (00) BCON = 1 Receive Clock (Input) A 10 MHz square wave synchronized to the Receive data and only active while receiving an Input Bit Stream. Am79C90 P R E L I M I N A R Y READ RESET (Input/Output, Three-State) Indicates the type of operation to be performed in the current bus cycle. This signal is an output when the C-LANCE is a Bus Master. Reset (Input) Reset causes the C-LANCE to cease operation, clear its internal logic, force all three-state buffers to the highimpedance state, and enter an idle state with the stop bit of CSR0 set. It is recommended that a 3.3 kΩ pullup resistor be connected to this pin. High — Data is taken off the DAL lines by the C-LANCE. Low — Data is placed on the DAL lines by the C-LANCE. The signal is an input when the C-LANCE is a Bus Slave. RX Receive (Input) Receive Input Bit Stream. TCLK High — Data is placed on the DAL lines by the C-LANCE. Transmit Clock (Input) Low — Data is taken off the DAL lines by the C-LANCE. TENA READY (Input/Output, Open Drain) When the C-LANCE is a Bus Master, READY is an asynchronous acknowledgment from the bus memory that it will accept data in a WRITE cycle or that it has put data on the DAL lines in a READ cycle. As a Bus Slave, the C-LANCE asserts READY when it has put data on the DAL lines during a READ cycle or is about to take data off the DAL lines during a write cycle. READY is a response to DAS and will return High after DAS has gone High. READY is an input when the C-LANCE is a Bus Master and an output when the C-LANCE is a Bus Slave. RENA Receive Enable (Input) A logical input that indicates the presence of carrier on the channel. 10 MHz clock. Transmit Enable (Output) Transmit Output Bit Stream enable. When asserted, it enables valid transmit output (TX). TX Transmit (Output) Transmit Output Bit Stream. VDD Power Supply Pin +5 V ±5% It is recommended that 0.1 µF and 10 µF decoupling capacitors be used between VDD and VSS. VSS Ground Pin 1 and 24 (48-Pin DlPs) should be connected together externally, as close to the chip as possible. Am79C90 7 AMD PRELIMINARY FUNCTIONAL DESCRIPTION The parallel interface of the CMOS Local Area Network Controller for Ethernet (C-LANCE) has been designed to be “friendly” or easy to interface to a variety of popular microprocessors. These microprocessors include the Am29000, 80x86, 680x0, Z8000 and LSI-11. The C-LANCE has a 24-bit wide linear address space when it is in the Bus Master Mode. A programmable mode of operation allows byte addressing in one of two ways: a Byte/Word control signal compatible with the 80x86 and Z8000 or an Upper Data Strobe and Lower Data Data and Address Bits 0–15 A16–A23 Strobe signal compatible with microprocessors such as the 68000. A programmable polarity on the Address Strobe signal eliminates the need for external logic. The C-LANCE interfaces with both multiplexed and demultiplexed data busses and features control signals for address/data bus transceivers. The C-LANCE is pin-forpin compatible with AMD’s LANCE device (Am7990). Please refer to Appendix B for a complete comparison between the C-LANCE and LANCE devices. Address Bits 16–23 Control Buffer CPU DAL0–DAL15 Buffer ALE A16–A23 C-LANCE DAL0 – DAL15 Buffer ALE ADR Latch ALE A16–A23 Decoder CS DAL0–DAL15 A16–A23 Control 17881B-5 Figure 1. C-LANCE/CPU Interfacing Multiplexed Bus 8 Am79C90 AMD PRELIMINARY Address Bus Data Bus Data/Address Bits 0-15 A0–A15 Latch ALE A16–A23 Address Bits 16-23 C-LANCE Buffer A0–A23 CS Decode 17881B-6 Figure 2. C-LANCE/CPU Interfacing Demultiplexed Bus During initialization, the CPU loads the starting address of the initialization block into two internal control registers. The C-LANCE has four internal control and status registers (CSR0, 1, 2, 3) which are used for various functions, such as the loading of the initialization block address, and programming different modes and status conditions. The host processor communicates with the C-LANCE during the initialization phase, for demand transmission, and periodically to read the status bits following interrupts. All other transfers to and from the memory are automatically handled as DMA. Interrupts to the microprocessor are generated by the C-LANCE upon: loads into the 48-byte Transmit FIFO; the C-LANCE then begins to transmit preamble while simultaneously loading the rest of the packet into Transmit FIFO for transmission. In the receive mode, packets are sent via the Am7992B SlA to the C-LANCE. The packets are loaded into the 64-byte Receive FIFO for preparation of automatic downloading into buffer memory. A CRC is calculated and compared with the CRC appended to the data packet. If the calculated CRC does not agree with the packet CRC, an error bit is set. Addressing Packets can be received using three different destination addressing schemes: physical, logical and promiscuous. completion of its initialization routine the reception of a packet the transmission of a packet transmitter timeout error a missed packet memory error The cause of the interrupt is ascertained by reading CSR0. Bit (06) of CSR0, (INEA), enables or disables interrupts to the microprocessor. In systems where polling is used in place of interrupts, bit (07) of CSR0, (INTR), indicates an interrupt condition. The basic operation of the C-LANCE consists of two distinct modes: transmit and receive. In the transmit mode, the C-LANCE chip directly accesses data (in a transmit buffer) in memory. It prefaces the data with a preamble, start frame delimiter (SFD), and calculates and appends a 32-bit CRC. On transmission, the first byte of data The first type is a full comparison of the 48-bit destination address in the packet with the node address that was programmed into the C-LANCE during an initialization cycle. There are two types of logical addresses. One is group type mask where the 48-bit address in the packet is put through a hash filter to map the 48-bit physical addresses into 1 of 64 logical groups. If any of these 64 groups have been preselected as the logical address, then the 48-bit address is stored in main memory. At this time, a look up is performed by the host computer comparing the 48-bit incoming address with the pre-stored 48-bit logical address. This mode can be useful if sending packets to all of a particular type of device simultaneously (i.e., send a packet to all file servers or all printer servers). Additional details on logical addressing can be found in the INITIALIZATION section Am79C90 9 AMD PRELIMINARY under “Logical Address Filter.” The second logical address is a broadcast address where all nodes on the network receive the packet. The last receive mode of operation is referred to as “promiscuous mode” in which a node will accept all packets on the medium regardless of their destination address. System errors include: Babbling Transmitter — Transmitter attempting to transmit more than 1518 bytes, excluding preamble and start frame delimiter Collision Collision Detection and Implementation The Ethernet and IEEE 802.3 CSMA/CD network access algorithms are implemented completely within the C-LANCE. In addition to listening for a clear medium before transmitting, Ethernet handles collisions in a predetermined way. Should two transmitters attempt to seize the medium at the same time, they will collide and the data on the medium will be garbled. The transmitting nodes listen while they transmit, detect the collision, then continue to transmit for a predetermined length of time to “jam” the network and ensure that all nodes have recognized the collision. The transmitting nodes then delay a random amount of time according to the Ethernet “truncated binary backoff” algorithm in order that the colliding nodes do not try to repeatedly access the network at the same time. The C-LANCE also offers a selectable Modified Backoff Algorithm for better performance on busy networks. Up to 16 attempts to access the network are made by the C-LANCE before reporting an error due to excessive collisions. Error Reporting and Diagnostics Extensive error reporting is provided by the C-LANCE. Error conditions reported relate either to the network as a whole or to individual data packets. Network-related errors are recorded as flags in the CSRs and are examined by the CPU following interrupt. Packet-related errors are written into descriptor entries corresponding to the packet. 10 — Collision detection circuitry nonfunctional Missed Packet — Insufficient buffer space Memory timeout — Memory response failure Packet-related errors: CRC — Invalid data Framing — Packet did not end on a byte boundary Overflow/Underflow — Indicates abnormal latency in servicing a DMA request Buffer — Insufficient buffer space available The C-LANCE performs several diagnostic routines which enhance the reliability and integrity of the system. These include a CRC check and two loop back modes (internal/external). Errors may be introduced into the system to check error detection logic. A Time Domain Reflectometer is incorporated into the C-LANCE to aid system designers in locating faults in the Ethernet physical medium. Shorts and opens manifest themselves in reflections which are sensed by the TDR. Am79C90 AMD PRELIMINARY Initialization Block Transmit Descriptor for 1st Data Buffer Transmit Descriptor for 2nd Data Buffer Transmit Descriptor for 3rd Data Buffer Transmit Descriptor Ring (4 words per entry) Transmit Descriptor for Nth Data Buffer Receive Descriptor for 1st Data Buffer Receive Descriptor for 2nd Data Buffer Receive Descriptor for 3rd Data Buffer Receive Descriptor Ring (4 words per entry) Receive Descriptor for Nth Data Buffer Transmit Data Buffer #1 Transmit Data Buffer #2 Transmit Data Buffer #3 Transmit Data Buffers Transmit Data Buffer #N Receive Data Buffer #1 Receive Data Buffer #2 Receive Data Buffer #3 Receive Data Buffers Receive Data Buffer #N 17881B-7 Figure 2-1. C-LANCE/Processor Memory Interface Am79C90 11 AMD PRELIMINARY Receive Descriptor Ring C-LANCE CSR Registers Pointer to Initialization Block Address of Receive Buffer 1 Buffer 1 Status Buffer 1 Byte Count Buffer 1 Message Count 2 2 2 2 Initialization Block Mode of Operation Receive Buffer Data Packet 1 Data Packet 2 Physical Address N N N N Logical Address Filter Pointer to Receive Ring Data Packet N Number of Receive Entries (N) Transmit Descriptor Ring Pointer to Transmit Ring Number of Transmit Entries (M) Address of Transmit Buffer 1 Buffer 1 Status Buffer 1 Byte Count Buffer 1 Error Status 2 2 2 2 M M M M Transmit Buffer Data Packet 1 Data Packet 2 Data Packet M 17881B-8 Figure 2-2. C-LANCE Memory Management Buffer Management A key feature of the C-LANCE and its on-board DMA channel is the flexibility and speed of communication between the C-LANCE and the host microprocessor through common memory locations. The basic organization of the buffer management is a circular queue of tasks in memory called descriptor rings as shown in Figures 2-1 and 2-2. There are separate descriptor rings to describe transmit and receive operations. Up to 128 tasks may be queued up on a descriptor ring awaiting execution by the C-LANCE. Each entry in a descriptor ring holds a pointer to a data memory buffer and an entry for the length of the data buffer. Data buffers can be chained or cascaded to handle a long packet in multiple data buffer areas. The C-LANCE searches the descriptor rings in a “lookahead” manner to determine the next empty buffer in order to chain buffers together or to handle back-to-back packets. As each buffer is filled, 12 the “own” bit is reset, allowing the host processor to process the data in the buffer. C-LANCE Interface CSR bits such as ACON, BCON and BSWP are used for programming the pin functions used for different interfacing schemes. For example, ACON is used to program the polarity of the Address Strobe signal (ALE/AS). BCON is used for programming the pins, for handling either the BYTE/WORD method for addressing word organized, byte addressable memories where the BYTE signal is decoded along with the least significant address bit to determine upper or lower byte, or an explicit scheme in which two signals labeled as BYTE MASK (BM0 and BM1) indicate which byte is addressed. When Am79C90 PRELIMINARY the BYTE scheme is chosen, the BM1 pin can be used for performing the function BUSAKO. BCON is also used to program pins for different DMA modes. In a daisy chain DMA scheme, 3 signals are used (BUSRQ, HLDA, BUSAKO). In systems using a DMA controller for arbitration, only HOLD and HLDA are used. C-LANCE in Bus Slave Mode AMD Write Sequence (Slave Mode) This cycle is similar to the read cycle, except that during this cycle, READ is not asserted (READ is LOW). The DAL buffers are tristated which configures these lines as inputs. The assertion of READY by C-LANCE indicates to the memory device that the data on the DAL lines have been stored by C-LANCE in its appropriate CSR register. CS, READ, DAS, ADR and DAL 15:00 must remain stable throughout the write cycle. Refer to Figure 4. The C-LANCE enters the Bus Slave Mode whenever CS becomes active. This mode must be entered whenever writing or reading the four status control registers (CSR0, CSR1, CSR2, and CSR3) and the Register Address Pointer (RAP). RAP and CSR0 may be read or written to at anytime, but the C-LANCE must be stopped (by setting the stop bit in CSR0) for CSR1, CSR2, and CSR3 access. Note: Setting the STOP bit in the C-LANCE will generate a C-LANCE reset, which will cause all bus control output signals (including READY) to float. To guarantee slave write timing when the STOP bit is being set in CSR0, the C-LANCE will latch the STOP bit and will wait for the slave cycle to complete before resetting itself and floating the output signals. Read Sequence (Slave Mode) C-LANCE in Bus Master Mode At the beginning of a read cycle, CS, READ, and DAS are asserted. ADR must be valid at this time. (If ADR is a “1,” the contents of RAP are placed on the DAL lines. Otherwise the contents of the CSR register addressed by RAP are placed on the DAL lines.) After the data on the DAL lines become valid, the C-LANCE asserts READY, CS, READ, DAS, and ADR must remain stable throughout the cycle. Refer to Figure 3. All data transfers from the C-LANCE in the bus Master mode are timed by ALE, DAS, and READY. The automatic adjustment of the C-LANCE cycle by the READY signal allows synchronization with variable cycle time memory due either to memory refresh or to dual port access. Transfers are a minimum of 600 ns in length except for the first transfer of a bus mastership period in which the minimum is 700 ns. Transfers can be increased in 100 ns increments. Am79C90 13 AMD PRELIMINARY DAL0–DAL15 Read Data See Note 1 DAS READ READY (Output from C-LANCE) O.D. HOLD CS ADR Note: 1. There are two types of delays which depend on which internal register is accessed. Type 1 refers to access of CSR0, CSR3 and RAP. Type 2 refers to access of CSR1 and CSR2 which are longer than Type 1 delay. Figure 3. Bus Slave Read Timing 14 Am79C90 17881B-9 AMD PRELIMINARY DAL0–DAL15 Write Data DAS READ READY (Output from C-LANCE) O.D. HOLD CS ADR 17881B-10 Figure 4. Bus Slave Write Timing Read Sequence (Master Mode) A read cycle is begun by placing a valid address on DAL00 – DAL15 and A16 – A23. The BYTE MASK signals are asserted to indicate a word, upper byte or lower byte memory reference. READ indicates the type of cycle. ALE or AS is pulsed, and the trailing edge of either can be used to latch addresses. DAL00 – DAL15 go into a 3-state mode, and DAS falls LOW to signal the beginning of the memory access. The memory responds by placing READY LOW to indicate that the DAL lines have valid data. The C-LANCE then latches memory data on the rising edge of DAS, which in turn ends the memory cycle and READY returns HIGH. Refer to Figure 5-1. The bus transceiver controls, DALI and DALO, are used to control the bus transceivers. DALI directs data toward the C-LANCE, and DALO directs data or addresses away from the C-LANCE. During a read cycle, DALO goes inactive before DALI becomes active to avoid “spiking” of the bus transceivers. Write Sequence (Master Mode) The write cycle is similar to the read cycle except that the DAL00 – DAL15 lines change from containing addresses to data after either ALE or AS goes inactive. After data is valid on the bus, DAS goes active. Data to memory is held valid after DAS goes inactive. Refer to Figure 5-2. Am79C90 15 AMD PRELIMINARY TWAIT T0 0 T1 100 T2 200 T3 T4 300 400 T5 500 T6 600 700 TCLK HOLD O.D. HLDA A16–A23 BM0, BM1 Address, BM0, BM1 ALE DAS READY DAL0–DAL15 (Read) Address Data In DALO (Read) DALI (Read) READ (Read) 17881B-11 Figure 5-1. Bus Master Read Timing (Single DMA Cycle) 16 Am79C90 AMD PRELIMINARY TWAIT T0 0 T1 100 T2 T3 200 T4 300 400 T5 500 T6 600 700 TCLK HOLD O.D. HLDA A16–A23 BM0, BM1 Address, BM0, BM1 ALE DAS READY DAL0–DAL15 (Write) Address Data Out DALO (Write) DALI (Write) READ (Write) 17881B-12 Figure 5-2. Bus Master Write Timing (Single DMA Cycle) Am79C90 17 AMD PRELIMINARY Differences Between Ethernet Versions 1 and 2 a. Version 2 specifies that the collision detect of the transceiver must be activated during the interpacket gap time. b. Version 2 specifies some network management functions, such as reporting the occurrence of collisions, retries and deferrals. c. Version 2 specifies that when transmission is terminated, the differential transmit lines are driven to 0 volt differentially (half step). Differences Between IEEE 802.3 and Ethernet A list of significant differences between Ethernet and IEEE 802.3 at the physical layer include the following: End of Transmission State Common Mode Voltage Ethernet Half Step Full Step (Rev 1) or Half Step (Rev 2) ±5.5 V Common Mode Current Less than 1 mA 0 – +5 V 1.6 mA ±40% Receive±, Collision± Input Threshold ±160 mV ±175 mV Fault Protection 16 V 0V a. IEEE 802.3 specifies a 2-byte length field rather than a type field. The length field (802.3) describes the actual amount of data in the frame. b. IEEE 802.3 allows the use of a PAD field in the data section of a frame, while Ethernet specifies the minimum packet size at 64 bytes. The use of a PAD allows the user to send and receive packets which have less than 46 bytes of data. 18 IEEE 802.3 Am79C90 AMD PRELIMINARY The C-LANCE loads itself with the information contained within the initialization block. PROGRAMMING This section defines the Control and Status Registers and the memory data structures required to program the Am79C90 (C-LANCE). The C-LANCE accesses the descriptor rings for packet handling. CONTROL AND STATUS REGISTERS Programming the Am79C90 (C-LANCE) The Am79C90 (C-LANCE) is designed to operate in an environment that includes close coupling with local memory and microprocessor (HOST). The Am79C90 C-LANCE is programmed by a combination of registers and data structures resident within the C-LANCE and memory registers. There are four Control and Status Registers (CSRs) within the C-LANCE which are programmed by the HOST device. Once enabled, the C-LANCE has the ability to access memory locations to acquire additional operating parameters. The Am79C90 has the ability to do independent buffer management as well as transfer data packets to and from the Ethernet. There are three memory structures accessed by the Chip: Initialization Block—12 words in contiguous memory starting on a word boundary. It also contains the operating parameters necessary for device operation. The initialization block is comprised of: — — — — Mode of Operation Physical Address Logical Address Mask Location to Receive and Transmit Descriptor Rings — Number of Entries in Receive and Transmit Descriptor Rings Receive and Transmit Descriptor Rings—Two ring structures, one for incoming and outgoing packets. Each entry in the rings is 4 words long and each entry must start on a quadword boundary. The Descriptor Rings are comprised of: — The address of a data buffer — The length of that data buffer — Status information associated with the buffer Data Buffers—Contiguous portions of memory reserved for packet buffering. Data buffers may begin on arbitrary byte boundaries. There are four Control and Status Registers (CSRs) on the chip. The CSRs are accessed through two bus addressable ports, an address port (RAP) and a data port (RDP). Accessing the Control and Status Registers The CSRs are read (or written) in a two step operation. The address of the CSR to be accessed is written into the RAP during a bus slave transaction. During a subsequent bus slave transaction, the data being read from (or written into) the RDP is read from (or written into) the CSR selected in the RAP. Once written, the address in RAP remains unchanged until rewritten. To distinguish the data port from the address port, a discrete input pin is provided. ADR Input Pin L H Port Register Data Port (RDP) Register Address Port (RAP) Register Data Port (RDP) 15 0 CSR DATA 17881B-13 Bit 15:00 In general, the programming sequence of the C-LANCE may be summarized as: Program the C-LANCE’s CSRs by a host device to locate an initialization block in memory. The byte control, byte address, and address latch enable modes are also defined here. Am79C90 Name CSR Data Description Writing data into RDP writes the data into the CSR selected in RAP. Reading the data from the RDP reads the data from the CSR selected in RAP. CSR1, CSR2 and CSR3 are accessible only when the STOP bit of CSR0 is set. If the STOP bit is not set while attempting to access CSR1, CSR2 or CSR3, the C-LANCE will return READY, but a READ operation will return undefined data. WRITE operation is ignored. 19 AMD PRELIMINARY Register Address Port (RAP) Bit Name Description 14 BABL 13 CERR 12 MISS BABBLE is a transmitter timeout error. It indicates that the transmitter has been on the channel longer than the time required to send the maximum length packet. BABL is a flag which indicates excessive length in the transmit buffer. It will be set after 1519 bytes have been transmitted, excluding preamble and start frame delimiter; the C-LANCE will continue to transmit until the whole packet is transmitted or until there is a failure before the whole packet is transmitted. When BABL error occurs, an interrupt will be generated if INEA = 1. BABL is READ/CLEAR ONLY and is set by the C-LANCE, and cleared by writing a “1” into the bit. Writing a ”0” has no effect. It is cleared by RESET or by setting the STOP bit. COLLISION ERROR indicates that the collision input to the C-LANCE was not asserted during the transmission, nor within 4.0 µs after the transmit completed. The collision after transmission is a transceiver test feature. This function is also known as heartbeat or SQE (Signal Quality Error) test. CERR is READ/CLEAR ONLY and is set by the C-LANCE and cleared by writing a “1” into the bit. Writing a “0” has no effect. It is cleared by RESET or by setting the STOP bit. CERR error will not cause an interrupt to occur (INTR = 0). MISSED PACKET is set when the receiver loses a packet because it does not own any receive buffer, indicating loss of data. FIFO overflow is not reported because there is no receive ring entry in which to write status. When MISS is set, an interrupt will be generated if INEA = 1. MISS is READ/CLEAR ONLY, and is set by the C-LANCE and cleared by writing a “1” into the bit. Writing a “0” has no effect. It is cleared by RESET or by setting the STOP bit. CSR 1:0 RES 17881B-14 Bit Name 15:02 RES 01:00 CSR(1:0) Description Reserved. Read as zeroes. Write as zeroes. CSR address select. READ/WRITE. Selects the CSR to be accessed through the RDP. RAP is cleared by Bus RESET. CSR(1 :0) CSR 00 01 10 11 CSR0 CSR1 CSR2 CSR3 Control and Status Register Definition Control and Status Register 0 (CSR0) 15 0 INIT ERR STRT BABL CERR STOP MISS TDMD MERR TXON RINT RXON TINT INEA IDON INTR The C-LANCE updates CSR0 by logical “ORing” the previous and present value of CSR0. 17881B-15 Bit 15 20 Name ERR Description ERROR summary is set by the “ORing” of BABL, CERR, MISS and MERR. ERR remains set as long as any of the error flags are true. ERR is read only; writing it has no effect. It is cleared by Bus RESET, setting the STOP bit, or clearing the individual error flags. Am79C90 AMD PRELIMINARY Bit Name Description Bit Name Description 11 MERR MEMORY ERROR is set when the C-LANCE is the Bus Master and has not received READY within 25.6 µs after asserting the address on the DAL lines. When a Memory Error is detected, the receiver and transmitter are turned off (CSR0, TXON = 0, RXON = 0) and an interrupt is generated if INEA = 1. MERR is READ/CLEAR ONLY, and is set by the C-LANCE and cleared by writing a “1” into the bit. Writing a “0” has no effect. It is cleared by RESET or by setting the STOP bit. RECEIVER INTERRUPT is set when the C-LANCE updates an entry in the Receive Descriptor Ring for the last buffer received or reception is stopped due to a failure. When RINT is set, an interrupt is generated if INEA = 1. RINT is READ/CLEAR ONLY, and is set by the C-LANCE and cleared by writing a “1” into the bit. Writing a “0” has no effect. It is cleared by RESET or by setting the STOP bit. TRANSMITTER INTERRUPT is set when the C-LANCE updates an entry in the transmit descriptor ring for the last buffer sent or transmission is stopped due to a failure. When TINT is set, an interrupt is generated if INEA = 1. TINT is READ/CLEAR ONLY and is set by the C-LANCE and cleared by writing a “1” into the bit. Writing a “0” has no effect. It is cleared by RESET or by setting the STOP bit. INITIALIZATION DONE indicates that the C-LANCE has completed the initialization procedure started by setting the INIT bit. When IDON is set, the C-LANCE has read the Initialization Block from memory and stored the new parameters. When IDON is set, an interrupt is generated if INEA = 1. 07 INTR 06 INEA 05 RXON 04 TXON INTERRUPT FLAG is set by the “ORing” of BABL, MISS, MERR, RINT, TINT and IDON. If INEA = 1 and INTR = 1, the INTR pin will be LOW. INTR is READ ONLY; writing this bit has no effect. INTR is cleared by RESET, by setting the STOP bit, or by clearing the condition causing the interrupt. INTERRUPT ENABLE allows the INTR pin to be driven LOW when the Interrupt Flag is set. If INEA = 1 and INTR = 1, the INTR pin will be Low. If INEA = 0, the INTR pin will be HIGH, regardless of the state of the Interrupt Flag. INEA is READ/WRITE and cleared by RESET or by setting the STOP bit. INEA can be set at any time, regardless of the state of the STOP bit. (reference Appendix B). RECEIVER ON indicates that the receiver is enabled. RXON is set when STRT is set if DRX = 0 in the MODE register in the initialization block and the initialization block has been read by the C-LANCE by setting the INIT bit. RXON is cleared when IDON is set from setting the INIT bit and DRX = 1 in the MODE register, or a memory error (MERR) has occurred. RXON is READ ONLY; writing this bit has no effect. RXON is cleared by RESET or by setting the STOP bit. TRANSMITTER ON indicates that the transmitter is enabled. TXON is set when STRT is set if DTX = 0 in the MODE register in the initialization block and the INIT bit has been set. TXON is cleared when IDON is set and DTX = 1 in the MODE register, or an error, such as MERR, UFLO or BUFF, has occurred during transmission. TXON is READ ONLY; writing this bit has no effect. TXON is cleared by RESET or by setting the STOP bit. 10 09 08 RINT TINT IDON IDON is READ/CLEAR ONLY, and is set by the C-LANCE and cleared by writing a “1” into the bit. Writing a “0” has no effect. It is cleared by RESET or by setting the STOP bit. Am79C90 21 AMD PRELIMINARY Bit Name Description 03 TDMD TRANSMIT DEMAND, when set, causes the C-LANCE to access the Transmit Descriptor Ring without waiting for the polltime interval to elapse. TDMD need not be set to transmit a packet; it merely hastens the C-LANCE’s response to a Transmit Descriptor Ring entry insertion by the host. TDMD is WRITE WITH ONE ONLY and is cleared by the microcode after it is used. It may read as a “1” for a short time after it is written because the microcode may have been busy when TDMD was set. It is also cleared by RESET or by setting the STOP bit. Writing a “0” in this bit has no effect. STOP disables the C-LANCE from all external activity when set and clears the internal logic. Setting STOP is the equivalent of asserting RESET. The C-LANCE remains inactive and STOP remains set until the STRT or INIT bit is set. If STRT, INIT and STOP are all set together, STOP will override the other bits and only STOP will be set. STOP is READ/WRITE WITH ONE ONLY and set by RESET. Writing a “0” to this bit has no effect. STOP is cleared by setting either INIT or STRT. CSR3 must be reloaded when the STOP bit is set. START enables the C-LANCE to send and receive packets, perform direct memory access, and do buffer management. The STOP bit must be set prior to setting the STRT bit. Setting STRT clears the STOP bit. STRT is READ/WRITE and is set with one only. Writing a “0” into this bit has no effect. STRT is cleared by RESET or by setting the STOP bit. INITIALIZE, when set, causes the C-LANCE to begin the initialization procedure and access the Initialization Block. The STOP bit must be set prior to setting the INIT bit. Setting INIT clears the STOP bit. INIT is READ/WRITE WITH “1” ONLY. Writing a “0” into this bit has no effect. INIT is cleared by RESET or by setting the STOP bit. Control and Status Register 1 (CSR1) READ/WRITE: 02 01 00 STOP STRT INIT 15 1 0 ‘0’ IADR (15:01) 17881B-16 Bit Name Description 15:01 IADR The low order 15 bits of the address of the first word (lowest address) in the Initialization Block. Must be zero. 00 Control and Status Register 2 (CSR2) READ/WRITE: 15 Accessible only when the STOP bit of CSR0 is a ONE and RAP = 10. The C-LANCE preserves the contents of CSR2 after STOP. 8 7 0 IADR (23:16) RES 17881B-17 Bit Name 15:08 RES 07:00 IADR The C-LANCE latches CSR0 during a slave read; therefore, the CSR0 status bits are guaranteed to be stable for the duration of the CSR0 access. 22 Accessible only when the STOP bit of CSR0 is a ONE and RAP = 01. The C-LANCE preserves the contents of CSR1 after STOP. Am79C90 Description Reserved. Read as zeroes. Write as zeroes. The high order 8 bits of the address of the first word (lowest address) in the initialization Block. AMD PRELIMINARY Control and Status Register 3 (CSR3) Initialization CSR3 allows redefinition of the Bus Master interface. Initialization Block READ/WRITE: Chip initialization includes the reading of the initialization block in memory to obtain the operating parameters. The following is a definition of the Initialization Block. 15 Accessible only when the STOP bit of CSR0 is ONE and RAP = 11. CSR3 is cleared by RESET or by setting the STOP bit in CSR0. 3 2 1 0 BCON ACON BSWP RES The Initialization Block is read by the C-LANCE when the INIT bit in CSR0 is set. The INIT bit should be set before or concurrent with the STRT bit to insure proper parameter initialization and chip operation. After the C-LANCE has read the Initialization Block, IDON is set in CSR0 and an interrupt is generated if INEA = 1. Higher Address 17881B-18 Bit 15:03 02 01 00 Name RES BSWP ACON BCON Description Reserved. Read as zeroes. Write as zeroes. BYTE SWAP allows the chip to operate in systems that consider bits (15:08) of data to be pointed at an even address and bits (07:00) to be pointed at an odd address. When BSWP = 1, the C-LANCE will swap the high and low bytes on DMA data transfers between the Receive FIFO and bus memory. Only data from the Receive FIFO transfers is swapped; the Initialization Block data and the Descriptor Ring entries are NOT swapped. BSWP is READ/WRITE and cleared by RESET or by setting the STOP bit in CSR0. TLEN–TDR (23:16) TDRA (15:00) RLEN–RDRA (23:16) RDRA (15:00) LADRF (63:48) LADRF (47:32) LADRF (31:16) LADRF (15:00) PADR (47:32) PADR (31:16) PADR (15:00) Base Address of Block MODE IADR +22 IADR +20 IADR +18 IADR +16 IADR +14 IADR +12 IADR +10 IADR +08 IADR +06 IADR +04 IADR +02 IADR +00 Mode The Mode Register allows alteration of the C-LANCE’s operating parameters. Normal operation is with the Mode Register clear. ALE CONTROL defines the assertive state of ALE when the C-LANCE is a Bus Master. ACON is READ/ WRITE and cleared by RESET and by setting the STOP bit in CSR0. 15 14 8 7 6 5 4 3 2 1 0 DRX DTX LOOP DTCR ACON ALE COLL 0 Asserted HIGH DRTY 1 Asserted LOW INTL BYTE CONTROL redefines the Byte Mask and Hold l/O pins. BCON is READ/WRITE and cleared by RESET or by setting the STOP bit in CSR0. BCON 0 1 Pin 16 Pin 15 Pin 17 EMBA RES PROM 17881B-19 BM1 BM0 HOLD BUSAKO BYTE BUSRQ All data transfers from the C-LANCE in the Bus Master mode are in words. However, the C-LANCE can handle odd address boundaries and/or packets with an odd number of bytes. Am79C90 23 AMD PRELIMINARY Bit Name Description Bit Name Description 15 PROM PROMISCUOUS mode. When PROM = 1, all incoming packets are accepted. RESERVED. Read as zeroes. Write as zeroes. Enable Modified Back-off Algorithm. When set (EMBA=1), enables the modified backoff algorithm. EMBA is cleared by activation of the RESET pin or setting the STOP bit. INTERNAL LOOPBACK is used with the LOOP bit to determine where the loopback is to be done. Internal loopback allows the chip to receive its own transmitted packet. Since this represents full duplex operation, the packet size is limited to 8–32 bytes. Internal loopback in the C-LANCE is operational when the packets are addressed to the node itself. The C-LANCE will not receive any packets externally when it is in internal loopback mode. EXTERNAL LOOPBACK allows the C-LANCE to transmit a packet through the SIA transceiver cable out to the Ethernet medium. It is used to determine the operability of all circuitry and connections between the C-LANCE and the physical medium. Multicast addressing in external loopback is valid only when DTCR = 1 (user needs to append the 4 bytes CRC). In external loopback, the C-LANCE also receives packets from other nodes. The FIFOs READ/WRITE pointers may misalign in the C-LANCE under heavy traffic. The packet could then be corrupted or not received. Therefore, the external loopback execution may need to be repeated. See specific discussion under “Loopback” in later section. INTL is only valid if LOOP = 1; otherwise, it is ignored. LOOP INTL LOOPBACK 04 COLL FORCE COLLISION. This bit allows the collision logic to be tested. The C-LANCE must be in internal loopback mode for COLL to be valid. If COLL = 1, a collision will be forced during the subsequent transmission attempt. This will result in 16 total transmission attempts with a retry error reported in TMD3. 03 DTCR 02 LOOP DISABLE TRANSMIT CRC. When DTCR = 0, the transmitter will generate and append a CRC to the transmitted packet. When DTCR = 1, the CRC logic is allocated to the receiver and no CRC is generated and sent with the transmitted packet. The ADD_FCS bit (bit 13, TMD1) can be used to override a DTCR=1 setting on a per packet basis. During loopback, DTCR = 0 will cause a CRC to be generated on the transmitted packet, but no CRC check will be done by the receiver since the CRC logic is shared and cannot generate and check CRC at the same time. The generated CRC will be written into memory with the data and can be checked by the host software. If DTCR = 1 during loopback, the host software must append a CRC value to the transmit data. The receiver will check the CRC on the received data and report any errors. LOOPBACK allows the C-LANCE to operate in full duplex mode for test purposes. The packet size is limited to 8–32 bytes.The received packet can be up to 36 bytes (32 + 4 bytes CRC) when DTCR = 0. During loopback, the runt packet filter is disabled because the maximum packet is forced to be smaller than the minimum size Ethernet packet (64 bytes). LOOP = 1 allows simultaneous transmission and reception for a message constrained to fit within the Transmit FIFO. The C-LANCE waits until the entire message is in the Transmit FIFO before serial transmission begins. The incoming data stream fills the Receive FIFO. Moving the received message out of the Receive FIFO to memory does not begin until reception has ceased. 14:08 07 06 05 24 RES EMBA INTL DRTY 0 X 1 0 No loopback, normal External 1 1 Internal DISABLE RETRY. When DRTY = 1, the C-LANCE will attempt only one transmission of a packet. If there is a collision on the first transmission attempt, a Retry Error (RTRY) will be reported in Transmit Message Descriptor 3 (TMD3). Am79C90 PRELIMINARY Bit Name 01 DTX 00 DRX Description In loopback mode, transmit data chaining is not possible. Receive data chaining is possible if receive buffers are 32 bytes long to allow time for lookahead. DISABLE THE TRANSMITTER causes the C-LANCE to not access the Transmitter Descriptor Ring, and therefore, no transmissions are attempted. DTX = 1 will clear the TXON bit in CSR0 when initialization is complete. DISABLE THE RECEIVER causes the C-LANCE to reject all incoming packets and not access the Receive Descriptor Ring. DRX = 1 will clear the RXON bit in the CSR0 when initialization is complete. 47 1 0 ‘0’ PADR (47:01) 17881B-20 47:00 PADR PHYSICAL ADDRESS is the unique 48-bit physical address assigned to the C-LANCE. PADR (0) must be zero. Logical Address Filter 63 0 LADRF 17881B-21 63:00 LADRF The 64-bit mask used by the C-LANCE to accept logical addresses. The purpose of logical (or group or multicast) addresses is to allow a group of nodes in a network to receive the same message. Each node can maintain a list of multicast addresses that it will respond to. The logical address filter mechanism in the C-LANCE is a hardware aide that reduces the average amount of host computer time required to determine whether or not an incoming packet with a multicast destination address should be accepted. The logical address filter hardware is an implementation of a hash code searching technique commonly used by software programmers. If the multicast bit of the destination address of an incoming packet is set, the AMD hardware maps this address into one of 64 categories which correspond to 64 bits in the Logical Address Filter Register. The hardware then accepts or rejects the packet depending on the state of the bit in the Logical Address Filter Register which corresponds to the selected category. For example, if the address maps into category 24, and bit 24 of the logical address filter register is set, the packet is accepted. A node can be made a member of several groups by setting the appropriate bits in the logical address filter register. The details of the hardware mapping algorithm are as follows: If the first bit of an incoming address is a “1” [PADR (0) =1], the address is deemed logical and is passed through the logical address filter. The logical address filter is a 64-bit mask composed of four sixteen-bit registers, LADRF (63:00) in the initialization block, that is used to accept incoming Logical Addresses. The incoming address is sent through the CRC circuit. After all 48 bits of the address have gone through the CRC circuit, the high order 6 bits of the resultant CRC (32-bit CRC) are strobed into a register. This register is used to select one of the 64-bit positions in the Logical Address Filter. If the selected filter bit is a “1,” the address is accepted and the packet will be put in memory. The logical address filter only assures that there is a possibility that the incoming logical address belongs to the node. To determine if it belongs to the node, the incoming logical address that is stored in main memory is compared by software to the list of logical addresses to be accepted by this node. The task of mapping a logical address to one of 64-bit positions requires a simple computer program (see Appendix A) which uses the same CRC algorithm (used in C-LANCE and defined per Ethernet) to calculate the HASH (see Figure 7). Driver software that manages a list of multicast addresses can work as follows. First the multicast address list and the logical address filter must be initialized. Some sort of management function such as the driver initialization routine passes to the driver a list of addresses. For each address in the list the driver uses a subroutine similar to the one listed in the appendix to set the appropriate bit in a software copy of the logical address filter register. When the complete list of addresses has been processed, the register is loaded. Later, when a packet is received, the driver first looks at the Individual/Group bit of the destination address of the packet to find out whether or not this is a multicast address. If it is, the driver must search the multicast address list to see if this address is in the list. If it is not in the list, the packet is discarded. Am79C90 25 AMD PRELIMINARY The Broadcast address, which consists of all ones is a special multicast address. Packets addressed to the broadcast address must be received by all nodes. Since broadcast packets are usually more common than other multicast packets, the broadcast address should be the first address in the multicast address list. Bit Name Description 31:29 RLEN RECEIVE RING LENGTH is the number of entries in the receive ring expressed as a power of two. RLEN Number of Entries 0 1 2 3 4 5 6 7 The Broadcast address does not go through the Logical Address Filter and is always enabled. If the Logical Address Filter is loaded with all zeroes, all incoming logical addresses except broadcast will be rejected. The multicast addressing in external loopback is operational only when DTCR in the mode register is set to 1. 28:24 RES 0 23:03 RDRA Logical Address Filter 63 0 02:00 32-Bit Resultant CRC 31 26 Destination Address 47 1 0 CRC Gen “1” Enable 1 2 4 8 16 32 64 128 RESERVED. Read as zeroes. Write as zeroes. RECEIVE DESCRIPTOR RING ADDRESS is the base address (lowest address) of the Receive Descriptor Ring. MUST BE ZEROES. These bits are RDRA (02:00) and must be zeroes because the Receive Ring is aligned on a quadword boundary. 64 Transmit Descriptor Ring Pointer MUX Match* 31 29 28 24 23 6 3 2 0 Select *Match - 1, the packet is accepted Match - 0, the packet is rejected 000 ‘(Quadword Boundary)’ RES TLEN 17881B-22 Figure 7. Logical Address Filter Operation TDRA (23:03) 17881B-24 Receive Descriptor Ring Pointer 31 29 28 24 23 RES RLEN 3 2 0 31:29 TLEN 0 1 2 3 4 5 6 7 000 ‘(Quadword Boundary)’ RDRA (23:03) 17881B-23 28:24 RES 23:03 TDRA 02:00 26 TRANSMIT RING LENGTH is the number of entries in the Transmit Ring expressed as a power of two. TLEN Number of Entries Am79C90 1 2 4 8 16 32 64 128 RESERVED. Read as zeroes. Write as zeroes. TRANSMIT DESCRIPTOR RING ADDRESS is the base address (lowest address) of the Transmit Descriptor Ring. MUST BE ZEROES. These bits are TDRA (02:00) and must be zeroes because the Transmit Ring is aligned on a quadword boundary. AMD PRELIMINARY Buffer Management Receive Message Descriptor 1 (RMD1) Buffer Management is accomplished through message descriptors organized in ring structures in memory. Each message descriptor entry is four words long. There are two rings allocated for the device: a Receive ring and a Transmit ring. The device is capable of polling each ring for buffers to either empty or fill with packets to or from the channel. The device is also capable of entering status information in the descriptor entry. C-LANCE polling is limited to looking one ahead of the descriptor entry the C-LANCE is currently working with. 15 87 0 HADR ENP STP BUFF CRC The location of the descriptor rings and their length are found in the initialization block, accessed during the initialization procedure by the C-LANCE. Writing a “ONE” into the STRT bit of CSR0 will cause the C-LANCE to start accessing the descriptor rings and enable it to send and receive packets. The C-LANCE communicates with a HOST device through the ring structures in memory. Each entry in the ring is either owned by the C-LANCE or the HOST. There is an ownership bit (OWN) in the message descriptor entry. Mutual exclusion is accomplished by a protocol which states that each device can only relinquish ownership of the descriptor entry to the other device; it can never take ownership, and no device can change the state of any field in any entry after it has relinquished ownership. OFLO FRAM ERR OWN 17881B-26 Bit Name Description 15 OWN 14 ERR 13 FRAM This bit indicates that the descriptor entry is owned by the host (OWN = 0) or by the C-LANCE (OWN = 1). The C-LANCE clears the OWN bit after filling the buffer pointed to by the descriptor entry. The host sets the OWN bit after emptying the buffer. Once the C-LANCE or host has relinquished ownership of a buffer, it must not change any field in the four words that comprise the descriptor entry. ERROR summary is the OR of FRAM, OFLO, CRC or BUFF. FRAMING ERROR indicates that the incoming packet contained a non-integer multiple of eight bits and there was a CRC error. If there was not a CRC error on the incoming packet, then FRAM will not be set even if there was a non-integer multiple of eight bits in the packet. FRAM is not valid in internal loopback mode. FRAM is valid only when ENP is set and OFLO is not. Descriptor Ring Each descriptor in a ring in memory is a 4-word entry. The following is the format of the receive and the transmit descriptors. Receive Message Descriptor Entry Receive Message Descriptor 0 (RMD0) 15 0 LADR 17881B-25 Bit Name Description 15:00 LADR The LOW ORDER 16 address bits of the buffer pointed to by this descriptor. LADR is written by the host and is not changed by the C-LANCE. Am79C90 27 AMD PRELIMINARY Bit Name Description 12 OFLO OVERFLOW error indicates that the receiver has lost all or part of the incoming packet due to an inability to store the packet in a memory buffer before the internal Receive FIFO overflowed. OFLO is valid only when ENP is not set. CRC indicates that the receiver has detected a CRC error on the incoming packet. CRC is valid only when ENP is set and OFLO is not. BUFFER ERROR is set any time the C-LANCE does not own the next buffer while data chaining a received packet. This can occur in either of two ways: 1) the OWN bit of the next buffer is zero, or 2) the Receive FIFO overflow occurred before the C-LANCE has performed a lookahead poll of the next receive descriptor. If a Buffer Error occurs, an Overflow Error may also occur internally in the Receive FIFO, but will not be reported in the descriptor status entry unless both BUFF and OFLO errors occur at the same time. START OF PACKET indicates that this is the first buffer used by the C-LANCE for this packet. It is used for data chaining buffers. END OF PACKET indicates that this is the last buffer used by the C-LANCE for this packet. It is used for data chaining buffers. If both STP and ENP are set, the packet fits into one buffer and there is no data chaining. The HIGH ORDER 8 address bits of the buffer pointed to by this descriptor. This field is written by the host and unchanged by the C-LANCE. Receive Message Descriptor 3 (RMD3) 15 11 CRC 10 BUFF 09 STP 08 ENP 07:00 HADR 12 11 0 MCNT RES 17881B-28 15:12 RES 11:00 MCNT RESERVED. Read as zeroes. Write as zeroes. MESSAGE BYTE COUNT is the length in bytes of the received message. MCNT is valid only when ERR is clear and ENP is set. MCNT is written by the chip and cleared by the host. Transmit Message Descriptor Entry Transmit Message Descriptor 0 (TMD0) 15 0 LADR 17881B-29 Bit Name Description 15:00 LADR The LOW ORDER 16 address bits of the buffer pointed to by this descriptor. LADR is written by the host and is not changed by the C-LANCE. Transmit Message Descriptor 1 (TMD1) 15 87 0 Receive Message Descriptor 2 (RMD2) 15 12 11 HADR 0 ENP STP BCNT Must be Ones 17881B-27 15:12 11:00 28 BCNT MUST BE ONES. This field is written by the host and is not changed by the C-LANCE. BUFFER BYTE COUNT is the length of the buffer pointed to by this descriptor, expressed as a two’s complement number. This field is written by the host and is not changed by the C-LANCE. Minimum buffer size is 64 bytes for the first buffer of packet. Am79C90 DEF ONE MORE ADD_FCS ERR OWN 17881B-30 AMD PRELIMINARY Bit Name Description 15 OWN This bit indicates that the descriptor entry is owned by the host (OWN = O) or by the C-LANCE (OWN = 1). The host sets the OWN bit after filling the buffer pointed to by this descriptor. The C-LANCE clears the OWN bit after transmitting the contents of the buffer. Neither the host nor the CLANCE may alter a descriptor entry after it has relinquished ownership. ERROR summary is the “OR” of LCOL, LCAR, UFLO or RTRY. Setting ADD_FCS=1, instructs the controller to append a CRC to this transmitted frame, regardless of the setting of the DTCR bit (bit 3 in the Mode Register). The ADD_FCS bit allows the controller to be configured to append CRC on a per packet basis, when DTCR=1. ADD_FCS is only valid when STP=1. MORE indicates that more than one retry was needed to transmit a packet. ONE indicates that exactly one retry was needed to transmit a packet. The ONE flag is not valid when LCOL is set. DEFERRED indicates that the C-LANCE had to defer while trying to transmit a packet. This condition occurs if the channel is busy when the C-LANCE is ready to transmit. START OF PACKET indicates that this is the first buffer to be used by the C-LANCE for this packet. It is used for data chaining buffers. STP is set by the host and is not changed by the C-LANCE. The STP bit must be set in the first buffer of the packet, or the C-LANCE will skip over this descriptor and poll the next descriptor(s) until the OWN and STP bits are set. END OF PACKET indicates that this is the last buffer to be used by the CLANCE for this packet. It is used for data chaining buffers. If both STP and ENP are set, the packet fits into one buffer and there is no data chaining. ENP is set by the host and is not changed by the C-LANCE. The HIGH ORDER 8 address bits of the buffer pointed to by this descriptor. This field is written by the host and is not changed by the C-LANCE. Transmit Message Descriptor 2 (TMD2) 15 14 ERR 13 ADD_FCS 12 MORE 11 ONE 10 DEF 09 STP 08 ENP 07:00 HADR 12 11 0 BCNT ONES 17881B-31 Bit Name Description 15:12 ONES 11:00 BCNT Must be ones. This field is set by the host and is not changed by the C-LANCE. BUFFER BYTE COUNT is the usable length in bytes of the buffer pointed to by this descriptor expressed as a negative two’s complement number. This is the number of bytes from this buffer that will be transmitted by the C-LANCE. This field is written by the host and is not changed by the C-LANCE. The first buffer of a packet has to be at least 100 bytes minimum when data chaining and 64 byte (DTCR = 1) or 60 bytes (DCTR = 0) when not data chaining. Am79C90 29 AMD PRELIMINARY 10 Transmit Message Descriptor 3 (TMD3) 15 10 9 0 TDR 09:00 RTRY LCAR LCOL RES UFLO BUFF 17881B-32 Bit Name Description 15 BUFF BUFFER ERROR is set by the C-LANCE during transmission when the C-LANCE does not find the ENP flag in the current buffer and does not own the next buffer. This can occur in either of two ways: either the OWN bit of the next buffer is zero, or Transmit FIFO underflow occurred before the C-LANCE has performed a lookahead poll of the next transmit descriptor. BUFF is set by the C-LANCE and cleared by the host. BUFF error will turn off the transmitter (CSR0, TXON = 0). 14 UFLO 13 RES 12 LCOL 11 LCAR 30 RTRY If a Buffer Error occurs, an Underflow Error will also occur. BUFF error is not valid when LCOL or RTRY error is set during TX data chaining. UNDERFLOW ERROR indicates that the transmitter has truncated a message due to data late from memory. UFLO indicates that the Transmit FIFO has emptied before the end of the packet was reached. Upon UFLO error, transmitter is turned off (CSR0, TXON = 0). RESERVED bit. The C-LANCE will write this bit with a “0.” LATE COLLISION indicates that a collision has occurred after the slot time of the channel has elapsed. The C-LANCE does not retry on late collisions. LOSS OF CARRIER is set when the carrier input (RENA) to the C-LANCE goes false during a C-LANCE-initiated transmission. The C-LANCE does not retry upon loss of carrier. It will continue to transmit the whole packet until done. LCAR is not valid in INTERNAL LOOPBACK MODE. TDR RETRY ERROR indicates that the transmitter has failed in 16 attempts to successfully transmit a message due to repeated collisions on the medium. If DRTY = 1 in the MODE register, RTRY will set after 1 failed transmission attempt. TIME DOMAIN REFLECTOMETRY reflects the state of an internal CLANCE counter that counts from the start of a transmission to the occurrence of a collision. This value is useful in determining the approximate distance to a cable fault. The TDR value is written by the C-LANCE and is valid only if RTRY is set. Ring Access Mechanism in the C-LANCE Once the C-LANCE is initialized through the initialization block and started, the CPU and the C-LANCE communicate via transmit and receive rings, for packet transmission and reception. There are 2 sets of RAM locations (four 16-bit register per set, corresponding to the 4 entries in each descriptor) in the C-LANCE. The first set points to the current buffer, and they are the working registers which are used for transferring the data for the packet. The second set contains the pointers to the next buffer in the ring which the C-LANCE obtained from the lookahead operation. There are three types of ring access in the C-LANCE. The first type is when the C-LANCE polls the rings to own a buffer. The second type is when the buffers are data chained. The C-LANCE does a lookahead operation between the time that it is transferring data to/from the Transmit/Receive FIFOs; this lookahead is done only once. The third type is when the C-LANCE tries to own the next descriptor in the ring when it clears the OWN bit for the current buffer. Transmit Ring Buffer Management When there is no Ethernet activity, the C-LANCE will automatically poll the transmit ring in the memory once it has started (CSR0, STRT = 1). This polling occurs every 1.6 ms, (CSR0 TDMD bit = 0) and consists of reading the status word of the transmit descriptor, TMD1, until the C-LANCE owns the descriptor. The C-LANCE will read TMD0 and TMD2 to get the rest of the buffer address and the buffer byte count when it owns the descriptor. Each of these memory reads is done separately with a new arbitration cycle for each transfer. If the transmit buffers are data chained (current buffer ENP = 0), the C-LANCE will look ahead to the next descriptor in the ring while transferring the current buffer into the Transmit FIFO (see Figure 8-1). The C-LANCE does this lookahead only once. If it does not own the next transmit Descriptor Table Entry (DTE) (2nd TX ring Am79C90 PRELIMINARY for this packet) it will transmit the current buffer and update the status of current Ring with the BUFF and UFLO error bits set. If the C-LANCE owns the 2nd DTE, it will also read the buffer address and the buffer byte count of this entry. Once the C-LANCE has finished emptying the current buffer, it clears the OWN bit for this buffer, and immediately starts loading the Transmit FIFO from the next (2nd) buffer. Between DMA bursts, starting from the 2nd buffer, the C-LANCE does a lookahead again to check if it owns the next (3rd) buffer. This activity goes on until the last transmit DTE indicates the end of the packet (TMD1, ENP = 1). Once the last part of the packet has been transmitted out from the Transmit FIFO to the medium, the C-LANCE will update the status in TMD1, TMD3 (TMD3 is updated only when there is an error) and will relinquish the last buffer to the CPU. The C-LANCE tries to own the next buffer (first buffer of the next packet), immediately after it relinquishes the last buffer of the current packet. This guarantees the backto-back transmission of the packets. If the C-LANCE does not own the next buffer, it then polls the TX ring every 1.6 ms. When an error occurs before all of the buffers get transmitted, the status, TMD3 , is updated in the current DTE, own bit is cleared in TMD1, and TINT bit is set in CSR0 which causes an interrupt if INEA = 1. The C-LANCE will then skip over the rest of the descriptors for this packet (clears the OWN bit and sets the TINT bit in CSR0) until it finds a buffer with both the STP and OWN bit being set (this indicates the first buffer for the next packet). When the transmit buffers are not data chained (current descriptor’s ENP = 1), the C-LANCE will not perform any lookahead operation. It will transmit the current buffer, update the TMD3 if any error, and then update the status and clear the OWN bit in TMD1 . The C-LANCE will then immediately check the next descriptor in the ring to see if it owns it. If it does, the C-LANCE will also read the rest of the entries from the descriptor table. If the C-LANCE does not own it, it will poll the ring once every 1.6 ms until it owns it. User may set the TDMD bit in CSR0 when it has relinquished a buffer to the C-LANCE. This will force the C-LANCE to check the OWN bit at this buffer without waiting for the polling time to elapse. Receive Ring Buffer Management Receive Ring access is similar to the transmit ring access. Once the receiver is enabled, the C-LANCE will always try to have a receive buffer available, should there be a packet addressed to this node for reception. Therefore, when the C-LANCE is idle, it will poll the receive ring entry once every 1.6 ms, until it owns the current receive DTE. Once the C-LANCE owns the buffer, it will read RMD0 and RMD2 to get the rest of buffer address and buffer byte count. When a packet arrives from the physical medium, after the Address Recognition Logic accepts the packet, the C-LANCE will immediately poll AMD the Receiver Ring once for a buffer. If it still does not own the buffer, it will set the MISS error in CSR0 and will not poll the receive ring until the packet ends. Assuming the C-LANCE owns a receive buffer when the packet arrives, it will perform a lookahead operation on the next DTE between periods when it is dumping the received data from the Receive FIFO to the first receive buffer in case the current buffer requires data chaining. When the C-LANCE owns the buffer, the lookahead operation consists of three separate single word DMA reads: RMD1, RMD0, and RMD2. When the C-LANCE does not own the next buffer, the lookahead operation consists of only one single DMA read, RMD1. Either lookahead operation is done only once. Following the lookahead operation, whether C-LANCE owns the next buffer or not, the C-LANCE will transfer the data from Receive FIFO to the first receive buffer for this packet in burst mode (8 word transfer per one DMA cycle arbitration). If the packet being received requires data chaining, and the C-LANCE does not own the second DTE, the C-LANCE will update the current buffer status, RMD1, with the BUFF and/or OFLO error bits set. If the C-LANCE does own the next buffer (second DTE) from previous lookahead, the C-LANCE will relinquish the current buffer and start filling up the second buffer for this packet. Between the time that the C-LANCE is transferring data from the Receive FIFO to the second buffer, it does a lookahead operation again to see if it owns the next (third) buffer. If the C-LANCE does own the third DTE, it will also read RMD0, and RMD2 to get the rest of buffer pointer address and buffer byte count. This activity continues on until the C-LANCE recognizes the end of the packet (physical medium is idle); it then updates the current buffer status with the end of packet bit (ENP) set. The C-LANCE will also update the message byte count (RMD3) with the total number of bytes received for this packet in the current buffer (the last buffer for this packet). The dual FIFOs in the C-LANCE are utilized by the internal microcode to guarantee that continuous receive activity does not prevent the servicing of pending transmit packets. The microcode includes a single transmit descriptor poll operation at the beginning of buffer DMA operations for an incoming receive packet. This single transmit descriptor poll is performed only once during the receive microcode routine for each packet that is received. If the OWN bit in the transmit descriptor is set, burst transfers to the Transmit FIFO are interleaved with burst transfers from the Receive FIFO. By interleaving the transmit buffer transfers with the receive buffer transfers, the beginning of the transmit packet is preloaded in the Transmit FIFO, ready to be transmitted immediately following the end of the receive packet on the wire. Am79C90 31 AMD PRELIMINARY 1 A (HOLD dwell time). Burst DMAs are always 8 transfer cycles unless there are fewer than 8 words left to be transferred to/from the Transmit/Receive FIFO, or if there are fewer than 8 words left to be transferred to/ from the RX/TX buffer. Transmit DMAs may be shorter than 8 words if a collision is detected during the DMA. B Single Word DMA Transfer 2 0 3 C 4 127 5 7 6 Output Packet A B C 17881B-33 Figure 8-1. Data Chaining (Transmit) C-LANCE 1 2 X 0 Y CPU Z 3 B (Note 1) W 2 1 CPU C 0 D 3 (Note 2) A 4 127 C-LANCE 127 4 5 7 6 Transmit 5 Receive 17881B-34 Notes: 1. W, X, Y, Z are the packets queued for transmission. 2. A, B, C, D are the packets received by the C-LANCE. Figure 8-2. Buffer Management Descriptor Rings C-LANCE DMA Transfer (Bus Master Mode) There are two types of DMA Transfers with the C-LANCE: Burst mode DMA FIFO Operation The dual FIFOs provide temporary buffer storage for data being transferred between the parallel bus l/O pins and serial I/O pins. The capacity of the Transmit FIFO is 48 bytes and the Receive FIFO is 64 bytes. Transmit Data is loaded into the Transmit FIFO under internal microprogram control. The Transmit FIFO has to have more than 16 bytes empty before the C-LANCE requests the bus (HOLD is asserted). The C-LANCE will start sending the preamble (if the line is idle) as soon as the first byte is loaded to the Transmit FIFO from memory. Receive Data is loaded into the Receive FIFO from the serial input shift register during reception. Data leaves the Receive FIFO under microprogram control. The C-LANCE microcode will wait until there are at least 16 bytes of data in the Receive FIFO before initiating a DMA burst transfer. Preamble and Start Frame Delimiter (SFD) are not loaded into the Receive FIFO. FIFOs – Memory Byte Alignment Single word DMA Burst Mode DMA Burst DMA is used for Transmission or Reception of the Packets, (Read/Write from/to Memory). The Burst Transfers are 8 consecutive word reads (transmit) or writes (receive) that are done in a single bus arbitration cycle. In other words, once the C-LANCE receives the bus acknowledge, (HLDA = LOW), it will do 8 word transfers (8 DMA cycle, min. at 600 ns per cycle) without releasing the bus request signal (HOLD = LOW). If there are more than 16 bytes empty in the Transmit FIFO, in transmit mode, or at least 16 bytes of data, in the Receive FIFO in receive mode, when the C-LANCE releases the bus (HOLD deasserted), the C-LANCE will request the bus again within 700 ns 32 The C-LANCE initiates single word DMA transfers to access the transmit and receive rings or the initialization block. The C-LANCE will not initiate any burst DMA transfers while reading the initialization block. The C-LANCE will not initiate any burst DMA transfers between the time that it discovers ownership of a descriptor and the time that it reads the buffer pointer and buffer byte count entries of that descriptor. Memory buffers may begin and end on arbitrary byte boundaries. Parallel data is byte aligned between the Transmit or Receive FIFO and DAL lines (DAL0–DAL15). Byte alignment can be reversed by setting the Byte Swap (BSWP) bit in CSR3. TRANSMISSION – WORD READ FROM EVEN MEMORY ADDRESS BSWP=0: FIFO BYTE n gets DAL <07:00> FIFO BYTE n + 1 gets DAL <15:08> BSWP=1: FIFO BYTE n gets DAL <15:08> FIFO BYTE n + 1 gets DAL <07:00> TRANSMISSION – BYTE READ FROM EVEN MEMORY ADDRESS Am79C90 AMD PRELIMINARY BSWP=0: FIFO BYTE n –don’t care gets DAL <07:00> gets DAL <15:08> BSWP=1: FIFO BYTE n –don’t care gets DAL <15:08> gets DAL <07:00> ACON, and BSWP are used; BCON, ACON, and BSWP default values are 0, 0, and 0 respectively). Only then the user may set the INIT bit in CSR0. It is recommended that the C-LANCE not be re-started, once it has been stopped (STOP = 1 in CSR0), by setting the STRT bit in CSR0 without reinitialization. Restarting the C-LANCE in this way puts the C-LANCE in operation in accordance with the parameters set up in the mode register, but the contents of the descriptor pointers in the C-LANCE will not be guaranteed. TRANSMISSION – BYTE READ FROM ODD MEMORY ADDRESS BSWP=0: FIFO BYTE n –don’t care gets DAL <15:08> gets DAL <07:00> BSWP=1: FIFO BYTE n –don’t care gets DAL <07:00> gets DAL <15:08> Frame Formatting RECEPTION – WORD WRITE TO EVEN MEMORY ADDRESS BSWP=0: DAL <07:00> DAL <15:08> gets FIFO BYTE n gets FIFO BYTE n + 1 BSWP=1: DAL <15:08> DAL <07:00> gets FIFO BYTE n gets FIFO BYTE n + 1 RECEPTION – BYTE WRITE TO EVEN MEMORY ADDRESS BSWP=0: DAL <07:00> DAL <15:08> gets FIFO BYTE n –undefined BSWP=1: DAL <15:08> DAL <07:00> gets FIFO BYTE n –undefined DAL <07:00> DAL <15:08> –undefined gets FIFO BYTE n BSWP=1: DAL <15:08> DAL <07:00> –undefined gets FIFO BYTE n Transmlt In transmit mode, the user must supply the destination address, source address, and Type Field (or Length Field) as a part of data in transmit data buffer memory. The C-LANCE will append the preamble, SFD, and CRC (FCS) to the frame as is shown in Figures 9-1 and 9-2. Receive RECEPTION – BYTE WRITE TO ODD MEMORY ADDRESS BSWP=0: The C-LANCE performs the encapsulation/decapsulation function of the data link layer (second layer of ISO model) as follows: The C-LANCE Recovery and Reinitialization In receive mode, the C-LANCE strips off the preamble and SFD and transfers the rest of the frame, including the CRC bytes (4 bytes), to the memory. The C-LANCE will discard packets with less than 64 bytes (runt packet) and will reuse the receive buffer for the next packet. This is the only case where the packet is discarded after the packet has been transferred to the receive buffer. A runt packet is normally the result of a collision. Preamble 1010 ... 1010 Synch 1 1 Dest. ADR Source ADR Type 62 Bits 2 Bits 6 Bytes 6 Bytes Data FCS 2 46–1500 4 Bytes Bytes Bytes The transmitter and receiver section of the C-LANCE are turned on via the initialization block (MODE REG: DRX, DTX bits). The state of the transmitter and the receiver are monitored through the CSR0 register (RXON, TXON bits). The C-LANCE must be reinitialized if the transmitter and/or the receiver has not been turned on during the original initialization, and later it is desired to have them turned on. When either the transmitter or receiver shuts off because an error (MERR, UFLO, TX BUFF error), it is necessary to reinitialize the C-LANCE to turn the transmitter and/or receiver back on again. The user should rearrange the descriptors in the transmit or receive ring prior to reinitialization. This is necessary since the transmit and receive descriptor pointers are reset to the beginning of the ring upon initialization. Framing Error (Dribbling Bits) To reinitialize the C-LANCE, the user must first stop the C-LANCE by setting the stop bit in CSR0. The user needs to reprogram CSR3 because its contents get cleared when the stop bit gets set (CSR3 reprogramming is not needed when default values of BCON, The C-LANCE can handle up to 7 dribbling bits when a received packet terminates; the input to the C-LANCE, RCLK, stops following the deassertion of RENA. During the reception, the CRC is generated on every serial bit (including the dribbling bits) coming from the medium, 17881B-35 Figure 9-1. Ethernet Frame Format Preamble SFD 1010 ... 1010 10101011 56 Bits 8 Bits Dest. ADR 6 Bytes Source ADR Length 6 Bytes 2 Bytes LLC Data PAD FCS 46–1500 Bytes 4 Bytes 17881B-36 Figure 9-2. IEEE 802.3 MAC Frame Format Am79C90 33 AMD PRELIMINARY and the CRC gets sampled internally on every byte boundary. The framing error is reported to the user as follows: If the number of the dribbling bits is 1 to 7 bits and there is no CRC error, then there is no Framing error (FRAM = 0). If the number of the dribbling bits is less than 8 and there is a CRC error, then there is also a Framing error (FRAM = 1). If the number of the dribbling bits = 0, then there is no Framing error. There may or may not be a CRC error. Interframe Spacing (IFS) The C-LANCE implements the two-part deferral algorithm following both receive and transmit activity, as specified as an option in the IEEE 802.3 Standard (ISO/ IEC 8802-3 1990). With two-part deferral, the interframe spacing, which begins immediately after the negation of RENA, is divided into two parts, IFS1 and IFS2. If RENA is asserted during IFS1, the interframe spacing counter is continually reset until RENA is deasserted (any pending transmissions will defer to the incoming receive traffic and the incoming frame may be received by the C-LANCE). Once the interframe spacing counter reaches IFS2, the counter proceeds, regardless of the state of RENA. When IFS2 expires, the C-LANCE may begin transmitting a frame if there is one pending. In the C-LANCE, IFS1 is 6.0 µs and IFS2 is 3.6 µs, making the minimum possible interframe spacing 9.6 µs. The 9.6 µs minimum interframe spacing complies with IEEE 802.3 specifications. Following each frame transmission, the C-LANCE blinds itself from any receive activity for the first 4.1 µs of the interframe spacing. The C-LANCE begins looking for the 011 start frame delimiter pattern after 800ns (8 bit times) of preamble has passed. Hence, if RENA is asserted during the first 4.1 µs of the interframe spacing, there must be at least 8 bits of preamble left following the end of the 4.1 µs window in order for the frame to be received correctly. Following each frame reception, the C-LANCE blinds itself from any receive activity for the first 0.5 µs of the interframe spacing. Collision Detection and Collision JAM Collisions are detected by monitoring the CLSN pin. If CLSN becomes asserted during a frame transmission, TENA will remain asserted for at least 32 (but not more than 40) additional bit times (including CLSN synchronization). This additional transmission after collision is referred to as COLLISION JAM. If collision occurs during the transmission of the preamble, the C-LANCE 34 continues to send the preamble, and sends the JAM pattern following the preamble. If collision occurs after the preamble, the C-LANCE will send the JAM pattern following the transmission of the current byte. The JAM pattern is any pattern except the CRC bytes. Receive Based Collision If CLSN becomes asserted during the reception of a packet, this reception is immediately terminated. Depending on the timing of COLLISION DETECTION, one of the following will occur. A collision that occurs within 6 byte times of the detection of the SFD (4.8 µs) will result in the packet being rejected because of an address mismatch; the Receive FIFO write pointer will be reset. A collision that occurs within 64 byte times (51.2 µs) will result in the packet being rejected since it is a runt packet. A collision that occurs after 64 byte times (late collision) will result in a truncated packet being written to the memory buffer with the CRC error bit most likely being set in the Status Word of the Receive Ring. Late collision error is not reported in receive mode. Transmit Based Collision When a transmission attempt has been terminated due to the assertion of CLSN, (a collision that occurs within 64 byte times), the C-LANCE will attempt to retry transmission 15 more times. The scheduling of the retransmissions is determined by a controlled randomized process called “truncated binary exponential backoff.” Upon the negation of the COLLISION JAM interval, the C-LANCE calculates a delay before retransmitting. The delay is an integral multiple of the SLOT TIME. The SLOT TIME is 512 bit times. The number of SLOT TIMES to delay before the nth retransmission is chosen as a uniformly distributed random integer in the range: 0≤ r ≤ 2k where k = min (n, 10). When the Modified Backoff Algorithm is enabled (EMBA), the backoff time may be longer than the minimum time specified above. Specifically, the backoff count will be suspended whenever a carrier is detected on the network. The backoff count will resume when the carrier drops. This behavior has the effect of making the backoff interval equal to the SUM of an integral number of SLOT TIMES plus the total duration of the carrier on the network during the backoff interval. If all 16 attempts fail, the C-LANCE sets the RTRY bit in the current Transmit Message Descriptor 3, TMD3, in memory, gives up ownership (sets the own bit to zero) for this packet, and processes the next packet in transmit ring for transmission. If there is a late collision (collision occurring after 64 byte times), the C-LANCE will not attempt to transmit this packet again; it will terminate the transmission, note the LCOL error in TMD3, and transmit the next packet in the ring. Am79C90 PRELIMINARY RECEPTION – MODE <02> LOOP = 0. The C-LANCE performs a check on the input bit stream from the first bit following the SFD to the last bit in the frame. The C-LANCE continually samples the state of the CRC check on framed byte boundaries, and, when the incoming bit stream stops, the last sample determines the state of the CRC error. Framing error (FRAM) is not reported if there is no CRC error. Collision—Microcode Interaction The microprogram uses the time provided by COLLISION JAM, INTERPACKET DELAY, and the backoff interval to restore the address and byte counts internally and starts loading the Transmit FIFO in anticipation of retransmission. It is important that C-LANCE be ready to transmit when the backoff interval elapses to utilize the channel properly. If, during the backoff interval, RENA and CLSN are never asserted (no wire activity), the C-LANCE does not re-poll the OWN bit and does not re-read the buffer address and byte count in the transmit descriptor before reloading the transmit data and retransmitting the transmit packet. However, if RENA or CLSN are asserted during the backoff interval, the C-LANCE must re-poll the OWN bit and re-read the buffer address and byte count in the transmit descriptor before starting the DMA access of the transmit buffer and performing the retry. Note that the re-polling of the transmit descriptor could be preceeded by receive DMA operations if an incoming packet arrives during the backoff interval and an address match is detected or when the C-LANCE is in promiscuous mode. Time Domain Reflectometry The C-LANCE contains a time domain reflectometry counter. The TDR counter is ten bits wide. It counts at a 10 MHz rate. It is cleared by the microprogram and counts upon the assertion of RENA during transmission. Counting ceases if CLSN becomes true, or RENA goes inactive. The counter does not wrap around. Once all ONEs are reached in the counter, the counter value is held until cleared. The value in the TDR is written into memory following the transmission of the packet. TDR is used to determine the location of suspected cable faults. LOOPBACK – MODE <02> LOOP =1, MODE <03> DTRC = 0. The C-LANCE generates and appends the CRC value to the outgoing bit stream as in Transmission but does not perform the CRC check of the incoming bit stream. LOOPBACK – MODE <02> LOOP = 1 MODE <03> DTRC = 1. C-LANCE performs the CRC check on the incoming bit stream as in Reception, but does not generate or append the CRC value to the outgoing bit stream during transmission. Loopback The normal operation of the C-LANCE is as a halfduplex device. However, to provide an on-line operational test of the C-LANCE, a pseudo-full duplex mode is provided. In this mode simultaneous transmission and reception of a loopback packet are enabled with the following constraints: The packet length must be no longer than 32 bytes, and no shorter than 8 bytes, exclusive of the CRC. Serial transmission does not begin until the Transmit FIFO contains the entire output packet. Moving the input packet from the Receive FIFO to the memory does not begin until the serial input bit stream terminates. Heartbeat CRC may be generated and appended to the output serial bit stream or may be checked on the input serial bit stream. CRC may not be used for both transmission and reception simultaneously. During the interpacket gap time following the negation of TENA, the CLSN input is asserted by some transceivers as a self-test. If the CLSN input is not asserted within 4 µs following the completion of transmission, then the C-LANCE will set the CERR bit in CSR0. CERR error will not cause an interrupt to occur (INTR = 0). In internal loopback, the packets should be addressed to the node itself. In external loopback, multicast addressing can be used only when DTCR = 1 is in the mode register. In this case, the user needs to append the CRC bytes. Cyclic Redundancy Check (CRC) The C-LANCE utilizes the 32-bit CRC function as described in the IEEE 802.3 standard section 3.2.8 to generate the Frame Check Sequence (FCS) field. The C-LANCE requirements for the CRC logic are the following: AMD Loopback is controlled by bits <06, 03, 02> INTL, DTCR, and LOOP of the MODE register. TRANSMISSION – MODE <02> LOOP = 0, MODE <03> DTCR = 0. The C-LANCE calculates the CRC from the first bit following the SFD to the last bit of the data field. The CRC value inverted is appended onto the transmission in one unbroken bit stream. Am79C90 35 AMD PRELIMINARY Serial Transmission Serial Reception Serial transmission consists of sending an unbroken bit stream from the TX output pin consisting of: Serial reception consists of receiving an unbroken bit stream on the RX input pin consisting of: Preamble/SFD: 56 alternating ONES and ZEROES terminating with the SFD byte (10101011). Preamble/SFD: Two ONES occurring a minimum of 8 bit times after the assertion of RENA. Data: The serialized bit stream from the Transmit FIFO Shifted out with LSB first. Destination Address: The 48 bits (6 bytes) following the SFD. CRC: The inverted 32-bit polynomial calculated from the data, address, and type field. CRC is not transmitted if: Data: The serial bit stream following the Destination Address. The last 4 complete bytes of data are the CRC. The Destination Address and the data are framed into bytes and enter the Receive FIFO. Source Address and Length field are part of the data which are transparent to the C-LANCE. — Transmission of the data field is truncated for any reason. — CLSN becomes asserted any time during transmission. — MODE <03> DTCR = 1 in a normal or loopback transmission mode, and ADD_FCS=0 in the transmit descriptor. Reception is indicated at the input pin by the assertion of RENA and the presence of clock on RCLK while TENA is inactive. The C-LANCE does not sample the received data until about 800 ns after RENA goes high. The Transmission is indicated at the output pin by the assertion of TENA with the first bit of the preamble and the negation of TENA after the last transmitted bit. The C-LANCE starts transmitting the preamble when the following are satisfied: There is at least one byte of data to be transmitted in the Transmit FIFO. The interpacket delay has elapsed. The backoff interval has elapsed, if doing a retransmission. 36 Am79C90 AMD PRELIMINARY ABSOLUTE MAXIMUM RATINGS OPERATING RANGES Storage Temperature . . . . . . . . . . . –65°C to +150°C Commercial (C) Devices Ambient Temperature with Power Applied . . . . . . . . . . . . . . . . . –25°C to +125°C Temperature (TA) . . . . . . . . . . . . . . . . . 0°C to +70°C Supply Voltages to Ground Potential Continuous . . . . . . . . . . . . . . . . . . . . . –0.3 V to +6 V VSS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 0 V Stresses above those listed under Absolute Maximum Ratings may cause permanent device failure. Functionality at or above these limits is not implied. Exposure to Absolute Maximum Ratings for extended periods may affect device reliability. Programming conditions may differ. Supply Voltage (VDD) . . . . . . . . . . +4.75 V to +5.25 V Operating ranges define those limits between which the functionality of the device is guaranteed. DC CHARACTERISTICS over operating ranges unless otherwise specified Parameter Symbol VIL Parameter Description Test Conditions VIH Input HIGH Voltage Output LOW Voltage IOL = 3.2 mA VOH Output HIGH Voltage IOH = –0.4 mA Input Leakage VIN = 0.4 V to VCC IIL Commercial Typ Max Input LOW Voltage VOL IDD* Min Unit 0.8 V 0.5 V 2 V 2.4 V Power Supply Current ±10 µA 50 mA *IDD is measured while running a functional pattern with spec. value IOH and IOL load applied. CAPACITANCE** (TA = 25°C; VDD = 0) Parameter Symbol CIN COUT CIO Parameter Description Test Conditions Max Unit Input Pin Capacitance f = 1 MHz Min Typ 10 pF Output Pin Capacitance f = 1 MHz 15 pF I/O Pin Capacitance f = 1 MHz 20 pF **Parameters are not tested. Am79C90 37 AMD PRELIMINARY SWITCHING CHARACTERISTICS No. Parameter Symbol Test Conditions Max Unit 1 tTCT TCLK Period 99 101 ns 2 tTCL TCLK LOW Time 45 55 ns 3 tTCH TCLK HIGH Time 45 55 ns Parameter Description Min Typ 4 tTCR Rise Time of TCLK (Note 3) 8 ns 5 tTCF Fall Time of TCLK (Note 3) 8 ns 6 tTEP TENA Propagation Delay After the Rising Edge of TCLK 60 ns 7 tTEH TENA Hold Time After the Rising Edge of TCLK 8 tTDP TX Data Propagation Delay After the Rising Edge of TCLK 9 tTDH TX Data Hold Time After the Rising Edge of TCLK 10 tRCT RCLK Period (Note 3) 85 11 tRCH RCLK HIGH Time (Note 2) 38 ns 12 tRCL RCLK LOW Time (Note 2) 38 ns 13 tRCR Rise Time of RCLK (Note 3) 14 tRCF Fall Time of RCLK (Note 3) 8 ns 15 tRDR RX Data Rise Time (Note 3) 8 ns 16 tRDF RX Data Fall Time (Note 3) 8 ns 17 tRDH RX Data Hold Time (RCLK to RX Data Change) (Note 2) 5 ns 18 tRDS RX Data Setup Time (RX Data Stable to the Rising Edge of RCLK) (Note 2) 35 ns 19 tDPL RENA LOW Time 1tTCT + 20 ns 20 tCPH CLSN HIGH Time 80 ns 21 tDOFF Bus Master Driver Disable After Rising Edge of HOLD 22 tDON Bus Master Driver Enable After Falling Edge of HLDA 50 23 tHHA Delay to Falling Edge of HLDA from Falling Edge of HOLD (Bus Master) 0 ns 24 tRW RESET Pulse Width LOW (Note 7) 2tTCT ns 25 tCYCLE Read/Write, Address/Data Cycle Time (Note 1) 6tTCT ns 26 tXAS Address Setup Time to the Falling Edge of ALE 75 ns 27 tXAH Address Hold Time After the Rising Edge of DAS 35 ns 28 tAS Address Setup Time to the Falling Edge of ALE 75 ns 29 tAH Address Hold Time After the Falling Edge of ALE 35 ns 30 tRDAS Data Setup Time to the Rising Edge of DAS (Bus Master Read) 40 ns 38 5 ns 60 5 Am79C90 ns ns 118 8 ns ns 50 ns 2tTCT + 50 ns AMD PRELIMINARY SWITCHING CHARACTERISTICS (continued) No. Parameter Symbol Parameter Description Test Conditions Min Typ Max Unit 31 tRDAH Data Hold Time After the Rising Edge of DAS (Bus Master Read) 0 ns 32 tDDAS Data Setup Time to the Falling Edge of DAS (Bus Master Write) 10 ns 33 tWDS Data Setup Time to the Rising Edge of DAS (Bus Master Write) 200 ns 34 tWDH Data Hold Time After the Rising Edge of DAS (Bus Master Write) 35 ns 35 tSD01 Data Driver Delay After the Falling Edge of DAS (Bus Slave Read) (CSR0, CSR3, RAP) (Note 6) 4tTCT ns 36 tSD02 Data Driver Delay After the Falling Edge of DAS (Bus Slave Read) (CSR1, 2) (Note 6) 12tTCT ns 37 tSRDH Data Hold Time After the Rising Edge of DAS (Bus Slave Read) 0 38 tSWDH Data Hold Time After the Rising Edge of DAS (Bus Slave Write) 0 ns 39 tSWDS Data Setup Time to the Falling Edge of DAS (Bus Slave Write) 0 ns 40 tALEW ALE Width HIGH 120 ns 41 tDALE Delay from Rising Edge of DAS to the Rising Edge of ALE 70 ns 55 ns 42 tDSW DAS Width LOW 200 43 tADAS Delay from the Falling Edge of ALE to the Falling Edge of DAS 80 44 tRIDF Delay from the Rising of DALO to the Falling Edge of DAS (Bus Master Read) 15 45 tRDYS Delay from the Falling Edge of READY to the Rising Edge of DAS 65 46 tROIF Delay from the Rising Edge of DALO to the Falling Edge of DALI (Bus Master Read) 15 ns 47 tRIS DALI Setup Time to the Rising Edge of DAS (Bus Master) 135 ns 48 tRIH DALI Hold Time After the Rising Edge of DAS (Bus Master Read) 0 ns 49 tRIOF Delay from the Rising Edge of DALI to the Falling Edge of DALO (Bus Master Read) 55 ns 50 tOS DALO and READ Setup Time to the Falling Edge of ALE (Bus Master Write and Read) 110 ns 51 tROH DALO Hold Time After the Falling Edge of ALE (Bus Master Read) 35 ns 52 tWDSI Delay from the Rising Edge of DAS to the Rising Edge of DALO (Bus Master Write) 35 ns 53 tCSH CS Hold Time After the Rising Edge of DAS (Bus Slave) 0 ns 54 tCSS CS Setup Time to the Falling Edge of DAS (Bus Slave) 0 ns Am79C90 ns 130 ns ns 250 ns 39 AMD PRELIMINARY SWITCHING CHARACTERISTICS (continued) No. 55 56 Parameter Symbol Parameter Description tSAH tSAS Test Conditions Min Typ Max Unit ADR Hold Time After the Rising Edge of DAS (Bus Slave) 0 ns ADR Setup Time to the Falling Edge of DAS (Bus Slave) 0 ns 57 tARYD Delay from the Falling Edge of ALE to the Falling Edge of READY to insure a Minimum Bus Cycle Time (600 ns) 58 tSRDS Data Setup Time to the Falling Edge of READY (Bus Slave Read) 75 ns 59 tRDYH READY Hold Time After the Rising Edge of DAS (Bus Master) 0 ns 80 ns (Note 5) 60 tSR01 READY Driver Turn On After the Falling Edge of DAS (Bus Slave) (CSR0, CSR3, RAP) (Notes 4, 6) 6tTCT ns 61 tSR02 READY Driver Turn On After the Falling Edge of DAS (Bus Slave) (CSR1, 2) (Note 6) 14tTCT ns 62 tSRYH READY Hold Time After the Rising Edge of DAS (Bus Slave) 0 35 ns 63 tSRH READ Hold Time After the Rising Edge of DAS (Bus Slave) 0 64 tSRS READ Setup Time to the Falling Edge of DAS (Bus Slave) 0 65 tCHL TCLK Rising Edge to HOLD LOW or High Delay 95 66 tCAV TCLK to Address Valid 100 ns 67 tCCA TCLK Rising Edge to Control Signals Active 75 ns 68 tCALE TCLK Falling Edge to ALE LOW 90 ns 69 tCDL TCLK Falling Edge to DAS Falling Edge 90 ns 70 tRCS Ready Setup Time to TCLK Falling Edge 71 tCDH TCLK Rising Edge to DAS HIGH 90 ns 72 tHCS HLDA Setup to TCLK Falling Edge 0 ns 73 tRENH RENA Hold Time After the Rising Edge of RCLK 0 ns 74 tCSR CS recovery time between deassertion of CS or HOLD and assertion of CS tTCT+60 ns (Note 5) ns ns 0 ns ns Notes: 1. Not shown in the timing diagrams, specifies the minimum bus cycle for a single DMA data transfer. Tested by functional data pattern. 2. Applicable parameters associated with Receive circuit are tested at tRCT (RCLK Period) = 100 ns, tTCT = 100 ns (TCLK Period). 3. Not tested. 4. CSR0 write access time (tSR01) when STOP bit is being set can be as long as 12tTCT. 5. It is guaranteed that no wait states will be added by the C-LANCE if either parameter #57 or #70 is met. 6. Parameter is for design reference only. 7. Reset must be asserted for at least two rising and two falling edges of TCLK for the device to be reset. If reset is deasserted before TCLK starts, the device behavior is undefined. 40 Am79C90 AMD PRELIMINARY IOL 1.5 V 100 pF IOH 17881B-37 A. Normal and Three-State Outputs IOL 1.5 V 100 pF 17881B-38 B. Open-Drain Outputs (INTR, HOLD/BUSRQ, READY) Am79C90 41 AMD PRELIMINARY KEY TO SWITCHING WAVEFORMS WAVEFORM INPUTS OUTPUTS Must Be Steady Will Be Steady May Change from H to L Will Be Changing from H to L May Change from L to H Will Be Changing from L to H Don’t Care, Any Change Permitted Changing, State Unknown Does Not Apply Center Line is HighImpedance “Off” State KS000010 SWITCHING WAVEFORMS (Note 1) CLSN 20 Serial Link Timing (Collision) 17881B-39 10 12 11 RCLK 18 16 13 14 RX 17 15 RENA 73 19 17881B-40 Serial Link Timing (Receive) 42 Am79C90 AMD PRELIMINARY SWITCHING WAVEFORMS 1 3 2 TCLK 8 9 4 5 TX TENA 7 6 * RENA 17881B-41 *During transmit, RENA input must be asserted (HIGH) and remain active-HIGH before TENA goes inactive (LOW). If RENA is deasserted before TENA is deasserted, LCAR will be reported in TMD3 after the transmission is completed by the C-LANCE. Serial Link Timing (Transmit) HOLD O.D. 21 HLDA 23 22 Bus Master Drivers Drivers Enabled 24 RESET 17881B-42 Note: 1. RESET is an asynchronous input to the C-LANCE and is not part of the Bus Acquisition timing. When RESET is asserted, the C-LANCE becomes a Bus Slave. Bus Acquisition Timing Am79C90 43 AMD PRELIMINARY SWITCHING WAVEFORMS DAL0–DAL15 Read Data 64 35 36 37 58 See Note 1 60 DAS 61 63 READ READY (Output from C-LANCE) 62 O.D 74 HOLD CS 53 54 ADR 56 55 17881B-45 Note: 1. There are two types of delays which depend on which internal register is accessed. Type 1 refers to access of CSR0 CSR3 and RAP. Type 2 refers to access of CSR1 and CSR2 which are longer than Type 1 delay. Bus Slave Read Timing 44 Am79C90 AMD PRELIMINARY SWITCHING WAVEFORMS DAL0–DAL15 Write Data 39 38 63 64 60 DAS 61 62 Read READY (Output from C-LANCE) O.D 74 HOLD 54 53 CS ADR 56 55 17881B-46 Bus Slave Write Timing Am79C90 45 46 Am79C90 READ (Read) DALI (Read) DALO (Read) DAL0–DAL15 (Read) READY DAS ALE A16–A23 BM0, BM1 HLDA HOLD TCLK 23 65 72 0 100 22 67 T0 1st DMA T2 68 300 50 Address 28 26 66 200 40 T1 T3 70 500 T5 600 29 48 59 41 T1 49 31 700 27 T6 50 69 300 T4 70 400 TWAIT T5 46 44 29 57 47 Data In 30 45 42 Address, BM0, BM1 43 T3 51 200 68 28 26 T2 Address 40 100 8th DMA Bus Master Read Timing (Burst DMA) 47 Data In 30 45 Address, BM0, BM1 400 T4 TWAIT 71 500 48 59 T6 27 31 65 600 O.D. 17881B-43 21 AMD PRELIMINARY SWITCHING WAVEFORMS Am79C90 READ (Write) DALI (Write) DALO (Write) DAL0–DAL15 (Write) READY DAS ALE A16–A23 BM0, BM1 HLDA 23 HOLD 65 TCLK 72 0 67 100 22 T0 1st DMA 66 T2 68 300 50 Address 28 26 200 40 T1 T3 70 500 T5 600 29 700 59 34 41 27 T6 T1 29 200 68 28 26 T2 Address 40 100 8th DMA Bus Master Write Timing (Burst DMA) 33 Data Out 45 Address, BM0, BM1 400 T4 TWAIT 69 300 T4 70 400 T5 57 32 71 500 33 Data Out 45 42 Address, BM0, BM1 43 T3 TWAIT 59 T6 52 65 600 O.D. 17881B-44 21 PRELIMINARY AMD SWITCHING WAVEFORMS 47 APPENDIX A Hash Filter Generation Programs for Logical Addressing 80x86 computer program example to generate the hash filter, for multicast addressing in the C-LANCE. 6 ; SUBROUTINE TO SET A BIT IN THE HASH FILTER FROM A 7 ; GIVEN ETHERNET LOGICAL ADDRESS 8 ; ON ENTRY Sl POINTS TO THE LOGICAL ADDRESS WITH LSB FIRST 9 ; 10 ; 11 ; 12 ; 13 Dl POINTS TO THE HASH FILTER WITH LSB FIRST ON RETURN Sl POINTS TO THE BYTE AFTER THE LOGICAL ADDRESS ALL OTHER REGISTERS ARE UNMODIFIED PUBLIC SETHASH 14 ASSUME CS:CSE61 15 ; 16 = 1DB6 POLYL EOU 1DB6H 17 = 04C1 POLYH EQU 04C1H 18 19 ;CRC POLYNOMINAL TERMS ; 0000 20 CSE61 SEGMENT PUBLIC ‘CODE’ ; 21 0000 22 0000 50 PUSH AX 23 0001 53 PUSH BX 24 0002 51 PUSH CX 25 0003 52 PUSH DX 26 0004 55 PUSH BP 27 SETHASH PROC NEAR ;SAVE ALL REGISTERS ; 28 0005 B8 FFFF MOV AX,0FFFFH 29 0008 BA FFFF MOV DX,0FFFFH ;PRESET CRC ACCUMULATOR TO ALL 1’S 30 000B B5 03 MOV CH,3 ;CH =WORD COUNTER MOV BP,[S1] ;GET A WORD OF ADDRESS ADD S1,2 ;POINT TO NEXT ADDRESS MOV CL,16 ;CL=BIT COUNTER MOV BX,DX ;GET HIGH WORD OF CRC ROL BX,1 ;PUT CRC31 TO LSB 31 ;AX,DX =CRC ACCUMULATOR ; 32 000D 8B 2C 33 000F 83 C6 02 34 0012 B1 10 35 SETH10: ; 36 0014 8B DA 37 0016 D1 C3 SETH20: 38 0018 33 DD XOR BX,BP ;COMBINE CRC31 WITH INCOMING BIT 39 001A D1 E0 SAL AX,1 ;LEFT SHIFT CRC ACCUMULATOR 40 001C D1 D2 RCL DX,1 41 001E 81 E3 0001 AND BX,0001H ;BX=CONTROL BIT 42 0022 74 07 JZ SETH30 ;DO NOT XOR IF CONTROL BIT = 0 43 ; 44 ; 48 PERFORM XOR OPERATION WHEN CONTROL BIT= 1 Am79C90 AMD 45 ; 46 0024 35 1D 86 XOR AX,POLYL 47 0027 81 F2 04C1 XOR DX,POLYH 48 ; 49 002B 0B C3 OR AX,BX ;PUT CONTROL BIT IN CRC0 50 002D D1 CD SETH30: ROR BP,1 ;ROTATE ADDRESS WORD 51 002F FE C9 DEC CL ;DECREMENT BIT COUNTER 52 0031 75 E1 JNZ SETH20 53 0033 FE CD DEC CH 54 0035 75 D6 JNZ SETH10 ;DECREMENT WORD COUNTER 55 ; FORMATION OF CRC COMPLETE, AL CONTAINS THE REVERSED HASH 56 ; CODE MOV CX,10 SETH40: SAL AL,1 ;REVERSE THE ORDER OF BITS IN AL ;AND PUT IT IN AH 58 0037 B9 000A 49 003A D0 E0 60 003C D0 DC RCR AH,1 61 003E E2 FA LOOP SETH40 62 63 ; 64 65 ; AH NOW CONTAINS THE HASH CODE 0040 8A DC MOV BL,AH ;BL = HASH CODE, BH IS ALREADY ZERO 66 0042 B1 03 MOV CL,3 ;DIVIDE HASH CODE BY 8 67 0044 D2 EB SHR BL,CL ;TO GET TO THE CORRECT BYTE 68 0046 B0 01 MOV AL,01H ;PRESET FILTER BIT 69 0048 80 E45 07 AND AH,7H ;EXTRACT BIT COUNT 70 004B 8A CC MOV CL,AH 71 004D D2 E0 SHL AL,CL ;SHIFT BIT TO CORRECT POSITION 72 004F 08 01 OR [Dl + BX],AL ;SET IN HASH FILTER 73 0051 5D POP BP 74 0052 5A POP DX 75 0053 59 POP CX 76 0054 5B POP BX 77 0055 58 POP AX 78 0056 C3 79 80 0057 81 82 RET ; SETHASH ENDP ; 0057 83 CSEG1 ENDS ; 84 END Program example in BASIC to generate the hash filter, for multicast addressing, in the C-LANCE. 100 REM 110 REM PROGRAM TO GENERATE A HASH NUMBER GIVEN AN ETHERNET ADDRESS 120 REM 130 DEFINT A–Z 140 DIM A(47): REM ETHERNET ADDRESS. 48 BITS. 150 DIM A$(6): REM INPUT FROM KEYBOARD 160 DIM C(32): REM CRC REGISTER–32 BITS Am79C90 49 AMD 170 PRINT “ENTER ETHERNET ADDRESS AS 6 HEXADECIMAL NUMBERS SEPARATED ” 180 PRINT “BY BLANKS. EACH NUMBER REPRESENTS ONE BYTE. THE LEAST ” 190 PRINT “SIGNIFICANT BIT OF THE FIRST BYTE IS THE FIRST BIT TRANSMITTED.” 200 PRINT “” 210 PRINT “ENTER ETHERNET ADDRESS”; 220 INPUT A$(0), A$(1), A$(2), A$(3), A$(4), A$(5) 240 REM 250 REM UNPACK ETHERNET ADDRESS INTO ADDRESS ARRAY 260 REM 270 M=0 280 FOR I = 0 TO 47: A(I) = 0: NEXT I 290 FOR I = 0 TO 5 300 IF LEN(A$(I)) = 1 THEN A$(I) = “0” + A$(I) 310 A$(I) = UCASE$(A$(I)) 320 FOR N = 2 TO 1 STEP –1 330 Y$ = MID$(A$(I), N, 1) 340 IF Y$ = “0” THEN 510 350 IF Y$ = “1” THEN A(M) = 1: GOTO 510 360 IF Y$ = “2” THEN A(M + 1) = 1: GOTO 510 370 IF Y$ = “3” THEN A(M + 1) = 1: A(M) = 1: GOTO 510 380 IF Y$ = “4” THEN A(M + 2) = 1: GOTO 510 390 IF Y$ = “5” THEN A(M + 2) = 1: A(M) = 1: GOTO 510 400 IF Y$ = “6” THEN A(M + 2) = 1: A(M + 1) = 1: GOTO 510 410 IF Y$ = “7” THEN A(M + 2) = 1: A(M + 1) = 1: A(M) = 1: GOTO 510 420 A(M + 3) = 1 430 IF Y$ = “8” THEN 510 440 IF Y$ = “9” THEN A(M) = 1: GOTO 510 450 IF Y$ = “A” THEN A(M + 1) = 1: GOTO 510 460 IF Y$ = “B” THEN A(M + 1) = 1: A(M) = 1: GOTO 510 470 IF Y$ = “C” THEN A(M + 2) = 1: GOTO 510 480 IF Y$ = “D” THEN A(M + 2) = 1: A(M) = 1: GOTO 510 490 IF Y$ = “E” THEN A(M + 2) = 1: A(M + 1) = 1: GOTO 510 500 IF Y$ = “F” THEN A(M + 2) = 1: A(M + 1) = 1: A(M) = 1 510 M=M+4 520 NEXT N 530 NEXT I 540 REM 550 REM PERFORM CRC ALGORITHM ON ARRAY A(0–47) 560 REM 570 FOR I = 0 TO 31: C(I) = 1: NEXT I 580 FOR N = 0 TO 47 590 REM SHIFT CRC REGISTER BY 1 600 FOR I = 32 TO 1 STEP –1: C(l) = C(I–1): NEXT I 610 C(0) = 0 620 T = C(32) XOR A(N): REM T = CONTROL BIT 630 IF T = 0 THEN 700: REM JUMP IF CONTROL BIT=0 50 Am79C90 AMD 640 C(1) = C(1) XOR 1: C(2) = C(2) XOR 1: C(4) = C(4) XOR 1 650 C(5) = C(5) XOR 1: C(7) = C(7) XOR 1: C(8) = C(8) XOR 1 660 C(10) = C(10) XOR 1: C(11) = C(11) XOR 1: C(12) = C(12) XOR 1 670 C(16) = C(16) XOR 1: C(22) = C(22) XOR 1: C(23) = C(23) XOR 1 680 C(26) = C(26) XOR 1 690 C(0) = 1 700 NEXT N 710 REM 720 REM CRC COMPUTATION COMPLETE, EXTRACT HASH NUMBER FROM C(0) TO C(5) 730 REM 740 HH=32*C(0)+16*C(1)+8*C(2)+4*C(3)+2*C(4)+C(5) 750 PRINT “THE HASH NUMBER FOR ”; 760 PRINT A$(0); “ ”; A$(1); “ ”; A$(2); “ ”; A$(3); “ ”; A$(4); “ ”; A$(5); 770 PRINT “IS”; HH 780 GOTO 210 Program example in C to generate the hash filter, for multicast addressing in the C-LANCE. /************************************************************ * hash.c Rev 0.1 * Generate a logical address filter value from a list of * Ethernet multicast addresses. * * Input: * User is prompted to enter an Ethernet address in * Ethernet hex format: First octet entered is the first * octet to appear on the line. LSB of most * significant octet is the first bit on the line. * Octets are separated by blanks. * After results are printed, user is prompted for * another address. * * (Note that the first octet transmitted is stored in * the C-LANCE as the least significant byte of the Physical * Address Register.) * Output: * After each address is entered, the program prints the * hash code for the last address and the cumulative * address filter function. The filter function is * printed as 8 hex bytes, least significant byte first. ****************************************************************/ #include <stdio.h> void updateCRC (int bit); int adr[6], /* Ethernet address */ ladrf[8], /* Logical address filter */ CRC[33], /* CRC register, 1 word/bit + extra control bit */ poly[] = /* CRC polynomial. poly[n] = coefficient of the x**n term of the CRC generator polynomial. */ {1,1,1,0, 1,1,0,1, 1,0,1,1, 1,0,0,0, 1,0,0,0, 0,0,1,1, 0,0,1,0, 0,0,0,0 }; void main() Am79C90 51 AMD { int k,i, byte; /* temporary array indices */ int hashcode; /* the object of this program */ char buf[80]; /* holds input characters */ for (i=0;i<8;i++) ladrf[i] = 0; /* clear log. adr. filter */ printf (”Enter Ethernet addresses as 6 octets separated by blanks.\n”); printf (”Each octet is one or two hex characters. The first octet \n”); printf (”entered is the first octet to be transmitted. The LSB of \n”); printf (”the first octet is the first bit transmitted. After each \n”); printf (”address is entered, the Logical Address Filter contents \n”); printf (”are displayed, least significant byte first, with the \n”); printf (”appropriate bits set for all addresses entered so far.\n”); printf (” To exit press the <Enter> key.\n\n”); while (1) { loop: printf (”\nEnter address: ”); /* If 1st character = CR, quit, otherwise read address. */ gets (buf); if ( buf[0] == ’\0’) break; if (sscanf (buf, ”%x %x %x %x %x %x”, &adr[0], &adr[1], &adr[2],&adr[3],&adr[4],&adr[5]) != 6) { printf (”Address must contain 6 octets separated by blanks.\n”); goto loop; } if ((adr[0] & 1) == 0) { printf (”First octet of multicast address ”); printf (”must be an odd number.\n”); goto loop; } /* Initialize CRC */ for (i=0; i<32; i++) CRC[i] = 1; /* Process each bit of the address in the order of transmission.*/ for (byte=0; byte<6; byte++) for (i=0; i<8; i++) updateCRC ((adr[byte] >> i) & 1); /* The hash code is the 6 least significant bits of the CRC in reverse order: CRC[0] = hash[5], CRC[1] = hash[4], etc. */ hashcode = 0; for (i=0; i<6; i++) hashcode = (hashcode << 1) + CRC[i]; /* Bits 3–5 of hashcode point to byte in address filter. Bits 0–2 point to bit within that byte. */ byte = hashcode >> 3; 52 Am79C90 AMD ladrf[byte] |= (1 << (hashcode & 7)); printf (”hashcode = %d (decimal) ladrf[0:63] = ”, hashcode); for (i=0; i<8; i++) printf (”%02X ”, ladrf[i]); printf (” (LSB first)\n”); } } void updateCRC (int bit) { int j; /* shift CRC and control bit (CRC[32]) */ for (j=32; j>0; j––) CRC[j] = CRC[j–1]; CRC[0] = 0; /* If bit XOR (control bit) = 1, set CRC = CRC XOR polynomial. */ if (bit ^ CRC[32]) for (j=0; j<32; j++) CRC[j] ^= poly[j]; } Am79C90 53 AMD Table A-1 “Mapping of Logical Address to Filter Mask” can be used to find a multicast address that maps into a particular address filter bit. For example, address BB 00 00 00 00 00 maps into bit 15. Therefore, any node that has bit 15 set in its logical address filter register will receive all packets addressed to BB 00 00 00 00 00. The table also shows that bit 15 is located in bit 7 of byte 1 of the Logical Address Filter Register. Addresses in this table are shown in the standard Ethernet format. The leftmost byte is the first byte to appear on the network with the least significant bit appearing first. Table A-1. Mapping of Logical Address to Filter Mask Byte Pos Bit Pos LAF Bit 0 0 0 85 00 00 00 00 0 1 1 A5 00 00 00 0 2 2 E5 00 00 00 0 3 3 C5 00 00 0 4 4 45 00 00 0 5 5 65 00 0 6 6 25 0 7 7 05 1 0 8 1 1 9 1 2 1 3 1 4 12 EB 00 00 00 00 00 5 4 44 4F 00 00 00 00 00 1 5 13 CB 00 00 00 00 00 5 5 45 6F 00 00 00 00 00 1 6 14 8B 00 00 00 00 00 5 6 46 2F 00 00 00 00 00 1 7 15 BB 00 00 00 00 00 5 7 47 0F 00 00 00 00 00 2 0 16 C7 00 00 00 00 00 6 0 48 63 00 00 00 00 00 2 1 17 E7 00 00 00 00 00 6 1 49 43 00 00 00 00 00 2 2 18 A7 00 00 00 00 00 6 2 50 03 00 00 00 00 00 2 3 19 87 00 00 00 00 00 6 3 51 23 00 00 00 00 00 2 4 20 07 00 00 00 00 00 6 4 52 A3 00 00 00 00 00 2 5 21 27 00 00 00 00 00 6 5 53 83 00 00 00 00 00 2 6 22 67 00 00 00 00 00 6 6 54 C3 00 00 00 00 00 2 7 23 47 00 00 00 00 00 6 7 55 E3 00 00 00 00 00 3 0 24 69 00 00 00 00 00 7 0 56 CD 00 00 00 00 00 3 1 25 49 00 00 00 00 00 7 1 57 ED 00 00 00 00 00 3 2 26 09 00 00 00 00 00 7 2 58 AD 00 00 00 00 00 3 3 27 29 00 00 00 00 00 7 3 59 8D 00 00 00 00 00 3 4 28 A9 00 00 00 00 00 7 4 60 0D 00 00 00 00 00 3 5 29 89 00 00 00 00 00 7 5 61 2D 00 00 00 00 00 3 6 30 C9 00 00 00 00 00 7 6 62 6D 00 00 00 00 00 3 7 31 E9 00 00 00 00 00 7 7 63 4D 00 00 00 00 00 54 Destination Address Accepted Byte Pos Bit Pos LAF Bit 00 4 0 32 21 00 00 00 00 00 00 00 4 1 33 01 00 00 00 00 00 00 00 4 2 34 41 00 00 00 00 00 00 00 00 4 3 35 71 00 00 00 00 00 00 00 00 4 4 36 E1 00 00 00 00 00 00 00 00 00 4 5 37 C1 00 00 00 00 00 00 00 00 00 00 4 6 38 81 00 00 00 00 00 00 00 00 00 00 4 7 39 A1 00 00 00 00 00 2B 00 00 00 00 00 5 0 40 8F 00 00 00 00 00 0B 00 00 00 00 00 5 1 41 BF 00 00 00 00 00 10 4B 00 00 00 00 00 5 2 42 EF 00 00 00 00 00 11 6B 00 00 00 00 00 5 3 43 CF 00 00 00 00 00 Am79C90 Destination Address Accepted APPENDIX B Comparison Between C-LANCE (Am79C90) and LANCE (Am7990) Devices OVERVIEW The Am79C90 C-LANCE device is a pin-for-pin equivalent for the Am7990 LANCE device. Using an advanced 0.8-micron CMOS process, the C-LANCE device consumes less power than the LANCE device, which is implemented in an outdated NMOS process. In addition to the inherent advantages provided by the advanced CMOS process, the C-LANCE device includes several functional enhancements over the LANCE device. The C-LANCE device is available in both 48-pin plastic DIP and 68-pin PLCC packages. These packages are socket-compatible with the LANCE packages. This document provides a comparison of the C-LANCE and LANCE devices. Table B-1 provides a summary of the comparison between the two devices. The remainder of the document gives details on each item listed in Table B-1. Am79C90 55 Table B-1. Comparison Summary of the C-LANCE and LANCE Devices Description Am79C90 C-LANCE Am7990 LANCE 1 Process/Power Consumption 0.8-micron CS-21S CMOS process ICC ≤ 50 mA NS-8B NMOS process ICC ≤ 270 mA 2 FIFOs Dual FIFOs: 48-byte TX, 64-byte RX Single FIFO: 48-byte TX/RX 3 Transmit Lockout Due to Receive Will not occur with dual FIFOs and enhanced microcode. May occur in high receive rate situations with “less than optimal” bus latencies. 4 Per-Packet FCS Transmit descriptor bit is used to allow per- No per-packet CRC control provided. packet addition of CRC when DTCR is set in the MODE register. 5 Backoff Algorithm Selectable Modified Backoff Algorithm or standard backoff algorithm. Only standard backoff algorithm available. 6 TX Descriptor Zero Buffer Byte Count Capability Allows TX buffer byte count of zero. No capability for TX buffer byte count of zero. 7 Interframe Spacing (IFS) Behavior a) Implements two-part deferral after transmit b) Part 1 of two-part deferral after receive is 6 µs c) Heartbeat window = 4 µs d) Receive blind time after receive less than 500 ns a) One-part deferral after transmit b) Part 1 of two-part deferral after receive is 4.1 µs c) Heartbeat window = 2 µs d) Receive blind time after receive = 4.1 µs 8 “Heartbeat OK” (No CERR) Heartbeat OK if collision is asserted at any Heartbeat OK if collision is asserted during Definition time from the beginning of the transmission the heartbeat window. to the end of the heartbeat window. 9 Receive Lockup Will not occur. May occur when bus latency is large. 10 ALE Behavior ALE may be driven HIGH at end of bus mastership when ACON is set to 0. When ACON is set to 1, ALE is not driven LOW at end of bus mastership period. ALE may be driven LOW at end of bus mastership when ACON is set to 1. When ACON is set to 0, ALE is not driven HIGH at end of bus mastership period. 11 External Loopback on a Live No problems. Network 12 Software Reset (STOP Bit) Handling a) STOP bit in CSR0 is latched. When STOP a) STOP bit in CSR0 not latched and will reset the device immediately when is set, the slave cycle is allowed to written. complete before the C-LANCE resets. b) CSR1 and CSR2 contents are preserved b) CSR1 and CSR2 are not preserved when the STOP bit is set to one. when the STOP bit is set to one. 13 CSR0 Slave Read Data Stability CSR0 latched during Slave reads to guarantee timing on DAL lines. CSR0 not latched during Slave read cycles (could give timing violations on DAL lines). 14 INEA Bit Behavior INEA bit can be set in CSR0 at any time, regardless of the state of the STOP bit. INEA cannot be set in CSR0 while the STOP bit is set. 15 Effect of Setting the STOP Bit on CSR0 Bits Setting the STOP bit in CSR0 when the STOP bit is already set does not affect any of the other bits in CSR0 (they are not cleared). Setting the STOP bit in CSR0 causes all of the other bits in CSR0 to clear, regardless of the previous state of the STOP bit. 16 AC Specification Changes #06 (tTEP) maximum = 60 ns #08 (tTDP) maximum = 60 ns #18 (tRDS) minimum = 35 ns #30 (tRDAS) minimum = 40 ns #45 (tRDYS) minimum = 65 ns #06 (tTEP) maximum = 70 ns #08 (tTDP) maximum = 70 ns #18 (tRDS) minimum = 40 ns #30 (tRDAS) minimum = 50 ns #45 (tRDYS) minimum = 75 ns 17 Burn-In Option The burn-in option for the C-LANCE is no longer available. 18 RX Descriptor Zero Buffer Byte Count Handling Unpredictable results when the RX Descriptor Buffer Byte Count is set to zero. 56 May receive invalid loopback failure indications. Am79C90 Interprets a BCNT field setting of zero in a receive descriptor as a 4096-byte buffer. Detailed Description of Enhancements 4. Per-Packet FCS 1. Process/Power Consumption In the LANCE device, addition of the Frame Check Sequence (FCS or CRC) to each transmit packet is controlled on a per-initialization basis. In other words, when the DTCR (Disable Transmit CRC) bit is set in the mode register at initialization, the only way that packets can subsequently be transmitted with an FCS attached is by re-initializing the device with the DTCR bit cleared. By using an advanced 0.8-micron CMOS process, the ICC specification for the C-LANCE device is reduced to 50 mA maximum, compared to the 270 mA maximum ICC specification for the LANCE device. 2. FIFOs The C-LANCE device incorporates a dual FIFO (48 bytes Transmit, 64 bytes Receive) architecture to help it compete for bandwidth on busy networks. The LANCE device’s single 48-byte FIFO architecture and its associated microcode has problems transmitting packets out on busy networks. This problem is known as the “Transmit Lockout Due to Receive” problem. It occurs when minimum or near-minimum IFS traffic is continually received by the LANCE device and bus latency is not “good” (“good” = latency < approximately 3 µs). In this situation, the LANCE device’s microcode and bus interface is locked servicing receive packets, and is not able to poll the pending transmit descriptor (until the receive traffic stops or does not pass address match). The C-LANCE device addresses this problem by including dual FIFOs and microcode that is modified to take advantage of the dual FIFOs. The microcode is changed so that a transmit descriptor poll operation occurs sometime early (exact time depends on bus latencies and whether the receive buffer was owned before the receive packet arrived) in the receive DMA operations for each packet. If the OWN bit in the TX descriptor is found set, transmit FIFO loading DMA is interleaved with the receive FIFO emptying DMA for the packet being received. The transmit packet is then ready to be transmitted immediately following the end of the receive packet on the wire. The dual FIFOs and microcode changes eliminate the possibility of transmit activity being locked out due to high receive activity. Interleaving the transmit DMA activity with receive DMA activity at the beginning of a reception has the effect of increasing the bus latency for receive DMA operations. To ensure that the C-LANCE device can tolerate the same bus latency as the LANCE device, the receive FIFO in the C-LANCE device is increased to 64 bytes. The transmit FIFO in the C-LANCE device holds 48 bytes. 3. Transmit Lockout Due to Receive As discussed in item 2, the dual FIFO architecture and modified microcode implemented in the C-LANCE device eliminates the possibility of Transmit Lockout Due to Receive occurring. The C-LANCE device provides the capability to override the DTCR setting on a per-packet basis. If DTCR was set in the mode register at initialization, the ADD_FCS bit in the transmit descriptor can be used to append FCS to transmitted packets on a per-packet basis, overriding the DTCR setting. If DTCR is cleared in the mode register, the ADD_FCS bit is a “don’t care.” The ADD_FCS bit is located in bit 13 of TMD1 in the C-LANCE device. This bit is RESERVED in the LANCE device. Table B-2 below summarizes the operation of the ADD_FCS bit. Note that the ADD_FCS bit is only meaningful in the first descriptor of a transmit buffer chain (STP = 1). Table B-2. ADD_FCS Bit Operation DTCR in Mode Reg. STP ADD_FCS FCS Added? 0 X X Yes 1 0 X N/A 1 1 0 No 1 1 1 Yes This feature should be compatible with existing implementations. Non-bridge nodes normally run with FCS enabled (DTCR cleared). Bridges run with FCS disabled. It is assumed that existing software in these applications do not set bit 13 of TMD1, which was previously RESERVED. The ADD_FCS bit is also implemented as bit 13 of TMD1 in the PCnet™-ISA (Am79C960) and operates identically to the way in which it operates in the C-LANCE device. As a side note, this feature can be used by software to distinguish the C-LANCE device from the LANCE device. The LANCE device writes bit 13 of TMD1 to zero when updating transmit status in the transmit descriptor. The C-LANCE device will write this bit with the value read, so if it is set to one it will be returned as a one. Am79C90 57 5. Backoff Algorithm A selectable Modified Backoff Algorithm is provided in the C-LANCE device that can improve throughput in busy networks. Bit 7 of the Mode register (EMBA bit) is used to enable the Modified Backoff Algorithm. This bit is RESERVED in the LANCE device. With the Modified Backoff Algorithm, counting of the IFS interval is suspended when receive carrier sense is detected. The count resumes when receive carrier sense goes away. This algorithm increases throughput in large networks with heavy traffic (many collisions). It can be considered an “adaptive” backoff algorithm. This mode should only be used in network segments in which all nodes are using this mode. Otherwise, the nodes that are using it will be at a disadvantage to those that are not. Note: This mode does not conflict with IEEE requirements for compliance. The IEEE 802.3 specification specifies only the minimum amount of time for the backoff interval. This leaves open the possibility of backing off more than the minimum, which is precisely how the Modified Backoff Algorithm works. The Modified Backoff Algorithm is included as an option in the MACE™ (Am79C940) and PCnet-ISA (Am79C960) devices. 6. TX Descriptor Zero Buffer Byte Count Capability The 12-bit BCNT field in the transmit descriptor of the LANCE and C-LANCE devices is loaded with the 2’s complement of the number of bytes that must be transmitted from the buffer. With the 2’s complement representation, a simple incrementer is used in the chip to count through the byte count as bytes are being read from the transmit buffer. When the 2’s complement number reaches all 0’s, the count has expired. The LANCE device does not check for the all 0’s case when the BCNT field is first loaded from the descriptor. Hence, the all 0’s case is interpreted by the LANCE device as a buffer count of 4096 (212), preventing zerolength TX buffers in the LANCE device. In addition, the LANCE device ignores the upper 4 bits in TMD2, which are adjacent to the BCNT field. These bits are indicated as “must be ones” in the LANCE data sheet. The C-LANCE device actually uses all 16 bits in TMD2 as the BCNT field. Compatibility with the LANCE device is preserved as long as the upper 4 bits in TMD2 are 1’s, as specified in the LANCE data sheet. The C-LANCE device checks for the case where all 16 bits in TMD2 are zero before starting any transmit DMA from the buffer. If all 16 bits are zero, a zero-length buffer is assumed, and the C-LANCE device immediately clears the OWN in the descriptor without starting any transmit activity on the network. Note that since all 16 bits are checked, compatibility with the LANCE device is preserved for non-Ethernet-compliant 58 implementations, which may use buffer lengths of 4096 bytes. Zero Transmit Buffer Byte Count Capability is included in the PCnet-ISA device. 7. Interframe Spacing (IFS) Behavior a. Two-Part Deferral After Transmit: Two-part deferral after receive has always been an option in the IEEE 802.3 specification. However, two-part deferral after transmit was recently added as an option in the 802.3 specification by the IEEE committee. With two-part deferral, the IFS is divided into two parts, IFS1 and IFS2. If there is activity on the wire during IFS1, the IFS counter is reset until the wire is clear again. The IFS counter is not reset once it enters IFS2. When the IFS counter expires, the chip will begin to transmit if it has anything to send. . The specification’s wording for two-part deferral after transmit is identical to the way that two-part deferral after receive has been worded all along. That is, the specification specifies that part 1 of the two parts can be anywhere from 0 to 2/3 of the IFS (9.6 µs). If part 1 = 0 (perfectly legal), it is equivalent to not implementing two-part deferral at all. Hence, the LANCE device, which implements two-part deferral after receive but not after transmit, complies with IEEE specifications. However, implementation of two-part deferral after both transmit and receive eliminates a possible scenario where packets cannot be received (due to very small or 0 IFS) but there is no indication of this fact through a collision indication at the transmitter. Therefore, although this scenario is very rare, the C-LANCE device implements two-part deferral after transmit in addition to after receive. b. The IEEE 802.3 specifications state that part 1 of two-part deferral can be anywhere from 0 to 2/3 of the IFS (9.6 µs). The LANCE device only implements two-part deferral after receive, with part 1 = 4.1 µ s and part 2 = 5.5 µ s (compliant). The C-LANCE device implements two-part deferral after both transmit and receive with part 1 = 6.0 µs and part 2 = 3.6 µs. Since the receiver is blinded following a transmit for 4.0 µs (see below), part 1 of twopart deferral after a transmit had to be extended beyond 4.1 µs or else part 1 would effectively be only from 4.0 µs to 4.1 µs during the IFS. Hence, in the C-LANCE device, part 1 of two-part deferral after transmit was set at 6.0 µs and the same value was used for part 1 following a receive. c. IEEE 802.3 specifications state that the Signal Quality Error (SQE) test window should be at least 4.0 µs and no more than 8.0 µs. The LANCE device implements a 2-µs window, which is not compliant with this specification. This generally turns out to be a non-issue because 802.3 also specifies that the Am79C90 MAU must generate the collision signal within 0.6 µs to 1.6 µs after the end of the transmit packet, which is typically early enough for the LANCE device to detect it, even with its non-compliant 2-µs window. However, to comply with IEEE standards, the C-LANCE device implements an SQE test window of 4 µs. d. IEEE specifications require that receive be blinded following transmit for 4 µs to prevent the controller from responding to any trash that may be generated by the MAU when it generates the SQE test signal. However, IEEE specifications do not state that the receiver should be blinded following a receive. The LANCE device implements a 4.1-µs blinding time following receive, violating IEEE specifications. This was erroneously implemented in the LANCE device, since it was thought to be a moot issue under the assumption that there should be no valid data on the wire within 4.1 µs of the end of a receive anyway. However, since two-part deferral after transmit and receive are both optional, as mentioned in 7a), there are rare situations where legal packets may arrive with an IFS of less than 4.1 µs. To better handle this situation, the C-LANCE device reduces the blind time following a receive to less than 500 ns. The blind time allows time to store and then clear the status that was generated by the ending reception. 8. “Heartbeat OK” (No CERR) Definition The heartbeat test or Signal Quality Error (SQE) test is performed to verify the ability of the AUI to pass the collision (SQE) indication to the DTE. The LANCE and C-LANCE devices indicate a heartbeat test failure by setting the CERR bit in CSR0 (bit 13). At the conclusion of each transmission, the DTE opens a time window during which it expects to see a collision indication. In the LANCE device, this window begins immediately when TENA deasserts and ends 2.0 µs after RENA deasserts. The heartbeat signal is expected by the LANCE device even if the packet being transmitted suffers a collision. This implementation violates IEEE requirements in three ways: 1. IEEE 802.3 specifications state that the heartbeat window should begin when the input becomes idle (RENA deasserts), not when the output becomes idle (TENA deasserts). 2. If a collision occurs, the IEEE 802.3 specifications indicate that the DTE should not look for the SQE test signal. 3. As mentioned in 7c), the window should end no earlier than 4.0 µs after RENA deasserts. The C-LANCE device implements the heartbeat test in full compliance with IEEE specifications. In the C-LANCE device, the heartbeat window begins when RENA deasserts and ends 4 µs later. In addition, the C-LANCE device does not look for the heartbeat signal whenever the packet being transmitted suffers a collision. The PCnet-ISA and MACE devices use the same heartbeat OK definition as the C-LANCE device. Details on the LANCE device’s violations of IEEE specifications: The consequences of the violations of the standard by the LANCE device are insignificant in practice. Item 1 (window begins when TENA deasserts, not RENA) actually prevents the LANCE from being penalized by Item 2 (heartbeat expected following a collision). That is, if the LANCE device did not violate Item 1 and started its window when RENA deasserted instead of TENA, then the LANCE device could get false CERR indications when a packet it is transmitting suffers a collision. This can happen as follows. In the event of a collision, the network may remain active for a while after one node stops transmitting its JAM sequence (other nodes involved in the collision may still have their JAM on the wire). At a node that ends its JAM sequence relatively early, the heartbeat signal can overlap with the collision or the end of the collision fragment, since the MAU times the heartbeat signal generation from when the controller stops transmitting. If this node uses a LANCE device as its controller, the LANCE device will see this heartbeat signal only because of the violation given in Item 1. If the LANCE device started its window when RENA deasserted instead of TENA, it would miss the heartbeat signal, since the heartbeat passes by while the collision is still on the wire. This would give false CERR indications. Hence, the violation of Item 1 in the LANCE device is not a problem. In fact, it makes the violation of Item 2 generally a non-issue. Although the violation of Item 1 masks the violation of Item 2 as just described, the violation of Item 2 (heartbeat still expected by the LANCE device when collision occurs) can still lead to false CERR indications when the LANCE device is used with a non-802.3-compliant MAU. The IEEE 802.3 specifications state that the MAU is to generate the SQE test signal after every transmit, even when the transmit suffers a collision. However, some MAUs on the market have been found not to comply with this requirement. When operating with a non-compliant MAU that does not generate the heartbeat signal after a collided transmission, the LANCE device can give false CERR indications. As mentioned in 7c), Item 3 is generally a non-issue. 9. Receive Lockup The LANCE device has an erratum in which the receiver locks up when the system bus latency is very high. This erratum is fixed in the C-LANCE device. Am79C90 59 10. ALE Behavior The LANCE device may drive the ALE pin LOW at the end of each bus mastership period when ACON = 1 (ALE/AS active low—AS mode). When the bus mastership period ends, the ALE pin is tri-stated; hence, if ALE is pulled HIGH by external logic, a glitch on ALE results. The glitch occurs about when the LANCE device is releasing the bus by bringing HOLD high. The C-LANCE device incorporates redesigned ALE logic to prevent this glitch from occurring. However, in the C-LANCE, when ACON = 0 (active high ALE), ALE is driven high before it is tri-stated at the end of every bus mastership period. In the LANCE, when ACON = 0 (active high ALE), ALE is not driven high before it is tri-stated at the end of every bus mastership period. This difference will not cause any problems in designs that set ACON = 1 (AS; active low ALE). It could cause problems in designs in which ACON = 0. The ALE signal is intended to provide a strobe signal for an external address latch. The rising edge, coupled with a subsequent falling edge that will occur if the pin is externally pulled down, will cause an invalid address to be strobed into the external address latch. However, since this occurs at the end of the bus mastership period, and further master cycles are not performed by the C-LANCE subsequent to the invalid address being strobed (until the next bus mastership period), the invalid address generally has no effect. A design could have problems with this if external logic is continuously decoding the latched address and taking some action on it even though the C-LANCE is not executing any master cycles. 11. External Loopback on a Live Network The LANCE device has an erratum that causes loopback failures when external loopback is run on a live network. This erratum is fixed in the C-LANCE device. 12. Software Reset (STOP Bit) Handling a. Latching of the STOP bit: In the LANCE device, writing the STOP bit in CSR0 causes all bus signals to immediately float. With READY pulled up externally (READY is open drain), this causes READY to deassert prematurely during the Slave cycle. If DAS and CS remain active, the LANCE device can erroneously start another Slave cycle. The C-LANCE device latches the STOP bit and, when it is set, allows the Slave cycle in progress to complete before resetting the part. b. Preservation of CSR1 and CSR2: The LANCE device does not preserve the contents of CSR1 and CSR2 during the initialization process. Hence, when the STOP bit is set, the contents of CSR1 and CSR2 are not the same as they were before initiali z a t i o n a n d t h ey mu s t b e r ew r i t t e n b e fo r e 60 re-initializing. This is not really a problem in the LANCE device, but it can add extra instructions to software. The C-LANCE device removes this software burden by preserving the contents of CSR1 and CSR2 during initialization so that when the STOP bit is set, they do not have to be reloaded before re-initializing. Note, however, that if the default values of CSR3 (defaults for BCON, ACON, and BSWP are 0, 0, and 0, respectively) are not used, CSR3 must still be reloaded after setting the STOP bit in the C-LANCE device, since CSR3 is cleared when the STOP bit is set. 13. CSR0 Slave Read Data Stability In the LANCE device, the status bit latches in CSR0 may change at any time, as governed by the occurrence of the external events they monitor. Hence, the ERR, BABL, CERR, MISS, IDON, and INTR bits in CSR0 may change during a Slave read cycle in which they are being accessed. This can cause timing violations on the DAL lines. In the C-LANCE device, CSR0 is latched in a shadow register during a read so that timing on the DAL lines is guaranteed. 14. INEA Bit Behavior With the C-LANCE device, an INEA bit can be set in CSR0 at any time, regardless of the state of the STOP bit. This actually removes a restriction that was present in the LANCE device, in which the INEA bit in CSR0 could be not be set while the STOP bit was set. This difference between the two devices does not affect normal device operation, but could disrupt diagnostic code written for the LANCE device. 15. Effect of Setting the STOP Bit on CSR0 Bits In the LANCE device, CSR0 is reset when the STOP bit in CSR0 is set. This reset happens even if the STOP bit was already set. When the reset occurs, all of the other bits in CSR0 are cleared. In the C-LANCE, CSR0 is reset when the STOP bit is set in CSR0 only if the STOP bit was not already set. This difference between the two devices does not affect normal device operation, but could disrupt diagnostic code written for the LANCE device. 16. AC Specification Changes The following differences in AC specification exist between the C-LANCE and the LANCE. C-LANCE LANCE #06 (tTEP) maximum 60 ns 70 ns #08 (tTDP) maximum 60 ns 70 ns #18 (tRDS) minimum 35 ns 40 ns #30 (tRDAS) minimum 40 ns 50 ns #45 (tRDYS) minimum 65 ns 75 ns Am79C90 17. Elimination of Burn-In Option The burn-in option for the C-LANCE is no longer available. Thus, the ordering part number Am79C90PCB is no longer valid (see page 4 of the C-LANCE data sheet). 18. RX Descriptor Zero Buffer Byte Count Handling The 12-bit BCNT field in the receive descriptor of the LANCE and C-LANCE devices is loaded with the 2’s complement of the number of bytes allocated to the associated receive buffer. In the LANCE device, when all 0’s are written to the BCNT field in a receive descriptor, a buffer length of 4096 (212) bytes is assumed. In the C-LANCE device, the case of all 0’s in the receive descriptor may produce unpredictable results. This difference should not cause problems in IEEE 802.3-compliant networks, because 802.3 has a maximum packet length specification of 1518 bytes. Am79C90 61 Trademarks Copyright © 1998 Advanced Micro Devices, Inc. All rights reserved. AMD, the AMD logo, and combinations thereof are trademarks of Advanced Micro Devices, Inc. Am186, Am386, Am486, Am29000, bIMR, eIMR, eIMR+, GigaPHY, HIMIB, ILACC, IMR, IMR+, IMR2, ISA-HUB, MACE, Magic Packet, PCnet, PCnet-FAST, PCnet-FAST+, PCnet-Mobile, QFEX, QFEXr, QuASI, QuEST, QuIET, TAXIchip, TPEX, and TPEX Plus are trademarks of Advanced Micro Devices, Inc. Microsoft is a registered trademark of Microsoft Corporation. Product names used in this publication are for identification purposes only and may be trademarks of their respective companies.