DP83932C-20/25/33 MHz SONIC Systems-Oriented Network Interface Controller
DP83932C-20/25/33 MHz SONIC
TM
Systems-Oriented Network Interface Controller
General Description
The SONIC (Systems-Oriented Network Interface Controller) is a second-generation Ethernet Controller designed to
meet the demands of today’s high-speed 32- and 16-bit systems. Its system interface operates with a high speed DMA
that typically consumes less than 3% of the bus bandwidth
(25 MHz bus clock). Selectable bus modes provide both big
and little endian byte ordering and a clean interface to standard microprocessors. The linked-list buffer management
system of SONIC offers maximum flexibility in a variety of
environments from PC-oriented adapters to high-speed
motherboard designs. Furthermore, the SONIC integrates a
fully-compatible IEEE 802.3 Encoder/Decoder (ENDEC) allowing for a simple 2-chip solution for Ethernet when the
SONIC is paired with the DP8392 Coaxial Transceiver Interface or a 10BASE-T transceiver.
For increased performance, the SONIC implements a
unique buffer management scheme to efficiently process,
receive and transmit packets in system memory. No intermediate packet copy is necessary. The receive buffer management uses three areas in memory for (1) allocating additional resources, (2) indicating status information, and (3)
buffering packet data. During reception, the SONIC stores
packets in the buffer area, then indicates receive status and
control information in the descriptor area. The system allocates more memory resources to the SONIC by adding descriptors to the memory resource area. The transmit buffer
management uses two areas in memory: one for indicating
status and control information and the other for fetching
packet data. The system can create a transmit queue allowing multiple packets to be transmitted from a single transmit
command. The packet data can reside on any arbitrary byte
boundary and can exist in several non-contiguous locations.
Two independent 32-byte transmit and receive FIFOs
Y
Bus compatibility for all standard microprocessors
Y
Supports big and little endian formats
Y
Integrated IEEE 802.3 ENDEC
Y
Complete address filtering for up to 16 physical and/or
multicast addresses
Y
32-bit general-purpose timer
Y
Full-duplex loopback diagnostics
Y
Fabricated in low-power CMOS
Y
132 PQFP package
Y
Full network management facilities support the 802.3
layer management standard
Y
Integrated support for bridge and repeater applications
System Diagram
TL/F/10492– 2
TRI-STATEÉis a registered trademark of National Semiconductor Corporation.
TM
SONIC
is a trademark of National Semiconductor Corporation.
C
1995 National Semiconductor CorporationRRD-B30M105/Printed in U. S. A.
TL/F/10492
1.0 FUNCTIONAL DESCRIPTION
1.1 IEEE 802.3 ENDEC Unit
1.1.1 ENDEC Operation
1.1.2 Selecting an External ENDEC
1.2 MAC Unit
1.2.1 MAC Receive Section
1.2.2 MAC Transmit Section
1.3 Data Width and Byte Ordering
1.4 FIFO and Control Logic
1.4.1 Receive FIFO
1.4.2 Transmit FIFO
1.5 Status and Configuration Registers
1.6 Bus Interface
1.7 Loopback and Diagnostics
1.7.1 Loopback Procedure
1.8 Network Management Functions
2.0 TRANSMIT/RECEIVE IEEE 802.3
FRAME FORMAT
2.1 Preamble and Start Of Frame Delimiter (SFD)
2.2 Destination Address
2.3 Source Address
2.4 Length/Type Field
2.5 Data Field
2.6 FCS Field
2.7 MAC (Media Access Control) Conformance
3.0 BUFFER MANAGEMENT
3.1 Buffer Management Overview
3.2 Descriptor Areas
3.2.1 Naming Convention for Descriptors
3.2.2 Abbreviations
3.2.3 Buffer Management Base Addresses
3.3 Descriptor Data Alignment
3.4 Receive Buffer Management
3.4.1 Receive Resource Area (RRA)
3.4.2 Receive Buffer Area (RBA)
3.4.3 Receive Descriptor Area (RDA)
3.4.4 Receive Buffer Management Initialization
3.4.5 Beginning of Reception
3.4.6 End of Packet Processing
3.4.7 Overflow Conditions
3.5 Transmit Buffer Management
3.5.1 Transmit Descriptor Area (TDA)
3.5.2 Transmit Buffer Area (TBA)
3.5.3 Preparing to Transmit
3.5.4 Dynamically Adding TDA Descriptors
Table of Contents
4.0 SONIC REGISTERS
4.1 The CAM Unit
4.1.1 The Load CAM Command
4.2 Status/Control Registers
4.3 Register Description
4.3.1 Command Register
4.3.2 Data Configuration Register
4.3.3 Receive Control Register
4.3.4 Transmit Control Register
4.3.5 Interrupt Mask Register
4.3.6 Interrupt Status Register
4.3.7 Data Configuration Register 2
4.3.8 Transmit Registers
4.3.9 Receive Registers
4.3.10 CAM Registers
4.3.11 Tally Counters
4.3.12 General Purpose Timer
4.3.13 Silicon Revision Register
5.0 BUS INTERFACE
5.1 Pin Configurations
5.2 Pin Description
5.3 System Configuration
5.4 Bus Operations
5.4.1 Acquiring the Bus
5.4.2 Block Transfers
5.4.3 Bus Status
5.4.4 Bus Mode Compatibility
5.4.5 Master Mode Bus Cycles
5.4.6 Bus Exceptions (Bus Retry)
5.4.7 Slave Mode Bus Cycle
5.4.8 On-Chip Memory Arbiter
5.4.9 Chip Reset
6.0 NETWORK INTERFACING
6.1 Manchester Encoder and Differential Driver
6.1.1 Manchester Decoder
6.1.2 Collision Translator
6.1.3 Oscillator Inputs
6.1.4 Power Supply Considerations
7.0 AC AND DC SPECIFICATIONS
8.0 AC TIMING TEST CONDITIONS
2
1.0 Functional Description
The SONIC
(ENDEC) unit, media access control (MAC) unit, separate
receive and transmit FIFOs, a system buffer management
engine, and a user programmable system bus interface unit
on a single chip. SONIC is highly pipelined providing maximum system level performance. This section provides a
functional overview of SONIC.
1.1 IEEE 802.3 ENDEC UNIT
The ENDEC (Encoder/Decoder) unit is the interface between the Ethernet transceiver and the MAC unit. It provides the Manchester data encoding and decoding functions for IEEE 802.3 Ethernet/Thin-Ethernet type local area
networks. The ENDEC operations of SONIC are identical to
the DP83910A CMOS Serial Network Interface device. During transmission, the ENDEC unit combines non-return-zero
(NRZ) data from the MAC section and clock pulses into
Manchester data and sends the converted data differentially
to the transceiver. Conversely, during reception, an analog
PLL decodes the Manchester data to NRZ format and receive clock. The ENDEC unit is a functionally complete
Manchester encoder/decoder incorporating a balanced
driver and receiver, on-board crystal oscillator, collision signal translator, and a diagnostic loopback. The features include:
Compatible with Ethernet I and II, IEEE 802.3 10base5
#
and 10base2
10Mb/s Manchester encoding/decoding with receive
#
clock recovery
Requires no precision components
#
Loopback capability for diagnostics
#
Externally selectable half or full step modes of operation
#
at transmit output
Squelch circuitry at the receive and collision inputs reject
#
noise
Connects to the transceiver (AUI) cable via external
#
pulse transformer
(Figure 1-1 )
consists of an encoder/decoder
1.1.1 ENDEC Operation
The primary function of the ENDEC unit
perform the encoding and decoding necessary for compatibility between the differential pair Manchester encoded data
of the transceiver and the Non-Return-to-Zero (NRZ) serial
data of the MAC unit data line. In addition to encoding and
decoding the data stream, the ENDEC also supplies all the
necessary special signals (e.g., collision detect, carrier
sense, and clocks) to the MAC unit. The signals provided to
the MAC unit from the on-chip ENDEC are also provided as
outputs to the user.
Manchester Encoder and Differential Output Driver:
During transmission to the network, the ENDEC unit translates the NRZ serial data from the MAC unit into differential
pair Manchester encoded data on the Coaxial Transceiver
Interface (e.g., National’s DP8392) transmit pair. To perform
this operation the NRZ bit stream from the MAC unit is
passed through the Manchester encoder block of the
ENDEC unit. Once the bit stream is encoded, it is transmitted out differentially to the transmit differential pair through
the transmit driver.
Manchester Decoder: During reception from the network,
the differential receive data from the transceiver (e.g., the
DP8392) is converted from Manchester encoded data into
NRZ serial data and a receive clock, which are sent to the
receive data and clock inputs of the MAC unit. To perform
this operation the signal, once received by the differential
receiver, is passed to the phase locked loop (PLL) decoder
block. The PLL decodes the data and generates a data receive clock and a NRZ serial data stream to the MAC unit.
Special Signals: In addition to performing the Manchester
encoding and decoding function, the ENDEC unit provides
control and clocking signals to the MAC unit. The ENDEC
sends a carrier sense (CRS) signal that indicates to the
MAC unit that data is present from the network on the
ENDEC’s receive differential pair. The MAC unit is also provided with a collision detection signal (COL) that informs the
MAC unit that a collision is taking place somewhere on
(Figure 1-2 )
is to
FIGURE 1-1. SONIC Block Diagram
3
TL/F/10492– 1
1.0 Functional Description (Continued)
TL/F/10492– 3
FIGURE 1-2. Block Diagram of Ethernet ENDEC
4
1.0 Functional Description (Continued)
the network. The ENDEC section detects this when its collision receiver detects a 10 MHz signal on the differential
collision input pair. The ENDEC also provides both the receive and transmit clocks to the MAC unit. The transmit
clock is one half of the oscillator input. The receive clock is
extracted from the input data by the PLL.
Oscillator: The oscillator generates the 10 MHz transmit
clock signal for network timing. The oscillator is controlled
by a parallel resonant crystal or by an external clock (see
Section 6.1.3). The 20 MHz output of the oscillator is divided
by 2 to generate the 10 MHz transmit clock (TXC) for the
MAC section. The oscillator provides an internal clock signal
for the encoding and decoding circuits.
Loopback Functions: The SONIC provides three loopback
modes. These modes allow loopback testing at the MAC,
ENDEC and external transceiver level (see Section 1.7 for
details). It is important to note that when the SONIC is transmitting, the transmitted packet will always be looped back
by the external transceiver. The SONIC takes advantage of
this to monitor the transmitted packet. See the explanation
of the Receive State Machine in Section 1.2.1 for more information about monitoring transmitted packets.
1.1.2 Selecting An External ENDEC
An option is provided on SONIC to disable the on-chip
ENDEC unit and use an external ENDEC. The internal IEEE
802.3 ENDEC can be bypassed by connecting the EXT pin
to V
(EXTe1). In this mode the MAC signals are redirect-
CC
ed, allowing an external ENDEC to be used. See Section 5.2
for the alternate pin definitions.
1.2 MAC UNIT
The MAC (Media Access Control) unit performs the media
access control functions for transmitting and receiving packets over Ethernet. During transmission, the MAC unit frames
information from the transmit FIFO and supplies serialized
data to the ENDEC unit. During reception, the incoming information from the ENDEC unit is deserialized, the frame
checked for valid reception, and the data is transferred to
the receive FIFO. Control and status registers on the SONIC
govern the operation of the MAC unit.
1.2.1 MAC Receive Section
The receive section
operations during reception, loopback, and transmission.
During reception, the deserializer goes active after detecting
the one byte SFD (Start of Frame Delimiter) pattern (Section
2.1) consisting of a ‘‘10101011’’ sequence. It then frames
the incoming bits into octet boundaries and transfers the
(Figure 1-3 )
controls the MAC receive
data to the 32-byte receive FIFO. Concurrently the address
comparator compares the Destination Address Field to the
addresses stored in the chip’s CAM address registers (Content Addressable Memory cells). If a match occurs, the deserializer passes the remainder of the packet to the receive
FIFO. The packet is decapsulated when the carrier sense
input pin (CRS) goes inactive. At the end of reception the
receive section checks the following:
Ð Frame alignment errors
Ð CRC errors
Ð Length errors (runt packets)
The appropriate status is indicated in the Receive Control
register (Section 4.3.3). In loopback operations, the receive
section operates the same as during normal reception.
During transmission, the receive section remains active to
allow monitoring of the self-received packet. The CRC
checker operates as normal, and the Source Address field
is compared with the CAM address entries. Status of the
CRC check and the source address comparison is indicated
by the PMB bit in the Transmit Control register (Section
4.3.4). No data is written to the receive FIFO during transmit
operations.
The receive section consists of the following blocks detailed
below.
Receive State Machine (RSM): The RSM insures the proper sequencing for normal reception and self-reception during transmission. When the network is inactive, the RSM
remains in an idle state continually monitoring for network
activity. If the network becomes active, the RSM allows the
deserializer to write data into the receive FIFO. During this
state, the following conditions may prevent the complete
reception of the packet.
Ð FIFO OverrunÐThe receive FIFO has been completely
filled before the SONIC could buffer the data to memory.
Ð CAM Address MismatchÐThe packet is rejected be-
cause of a mismatch between the destination address of
the packet and the address in the CAM.
Ð Memory Resource ErrorÐThere are no more resources
(buffers) available for buffering the incoming packets.
Ð Collision or Other ErrorÐA collision occured on the net-
work or some other error, such as a CRC error, occurred
(this is true if the SONIC has been told to reject packets
on a collision, or reject packets with errors).
If these conditions do not occur, the RSM processes the
packet indicating the appropriate status in the Receive Control register.
FIGURE 1-3. MAC Receiver
5
TL/F/10492– 4
1.0 Functional Description (Continued)
During transmission of a packet from the SONIC, the external transceiver will always loop the packet back to the
SONIC. The SONIC will use this to monitor the packet as it
is being transmitted. The CRC and source address of the
looped back packet are checked with the CRC and source
address that were transmitted. If they do not match, an error
bit is set in the status of the transmitted packet (see Packet
Monitored Bad, PMB, in the Transmit Control Register, Section 4.3.4). Data is not written to the receive FIFO during this
monitoring process unless Transceiver Loopback mode has
been selected (see Section 1.7).
Receive Logic: The receive logic contains the command,
control, and status registers that govern the operations of
the receive section. It generates the control signals for writing data to the receive FIFO, processes error signals obtained from the CRC checker and the deserializer, activates
the ‘‘packet reject’’ signal to the RSM for rejecting packets,
and posts the applicable status in the Receive Control register.
Deserializer: This section deserializes the serial input data
stream and furnishes a byte clock for the address comparator and receive logic. It also synchronizes the CRC checker
to begin operation (after SFD is detected), and checks for
proper frame alignment with respect to CRS going inactive
at the end of reception.
Address Comparator: The address comparator latches the
Destination Address (during reception or loopback) or
Source Address (during transmission) and determines
whether the address matches one of the entries in the CAM
(Content Addressable Memory).
CRC Checker: The CRC checker calculates the 4-byte
Frame Check Sequence (FCS) field from the incoming data
stream and compares it with the last 4-bytes of the received
packet. The CRC checker is active for both normal reception and self-reception during transmission.
Content Addressable Memory (CAM): The CAM contains
16 user programmable entries and 1 pre-programmed
Broadcast address entry for complete filtering of received
packets. The CAM can be loaded with any combination of
Physical and Multicast Addresses (Section 2.2). See Section 4.1 for the procedure on loading the CAM registers.
1.2.2 MAC Transmit Section
The transmit section
data from the transmit FIFO and transmitting a serial data
(Figure 1-4 )
is responsible for reading
stream onto the network in conformance with the IEEE
802.3 CSMA/CD standard. The Transmit Section consists
of the following blocks.
Transmit State Machine (TSM): The TSM controls the
functions of the serializer, preamble generator, and JAM
generator. It determines the proper sequence of events that
the transmitter follows under various network conditions. If
no collision occurs, the transmitter prefixes a 7 byte preamble and 1 byte Start of Frame Delimiter (SFD) consisting of a
‘‘10101011’’ sequence at the beginning of each packet,
then sends the serialized data. At the end of the packet, an
optional 4-byte CRC pattern is appended. If a collision occurs, the transmitter switches from transmitting data to
sending a 4-byte Jam pattern to notify all nodes that a collision has occurred. Should the collision occur during the preamble, the transmitter waits for it to complete before jamming. After the transmission has completed, the transmitter
writes status in the Transmit Control register (Section 4.3.4).
Protocol State Machine: The protocol state machine assures that the SONIC obeys the CSMA/CD protocol. Before
transmitting, this state machine monitors the carrier sense
and collision signals for network activity. If another node(s)
is currently transmitting, the SONIC defers until the network
is quiet, then transmits after its Interframe Gap Timer
(9.6 ms) has expired. The Interframe Gap time is divided into
two portions. During the first 6.4 ms, network activity restarts
the Interframe Gap timer. Beyond this time, however, network activity is ignored and the state machine waits the remaining 3.2 ms before transmitting. If the SONIC experiences a collision during a transmission, the SONIC switches
from transmitting data to a 4-byte JAM pattern (4 bytes of all
1’s), before ceasing to transmit. The SONIC then waits a
random number of slot times (51.2 ms) determined by the
Truncated Binary Exponential Backoff Algorithm
reattempting another transmission. In this algorithm, the
number of slot times to delay before the nth retransmission
is chosen to be a random integer r in the range of:
where kemin(n,10)
If a collision occurs on the 16th transmit attempt, the SONIC
aborts transmitting the packet and reports an ‘‘Excessive
Collisions’’ error in the Transmit Control register.
0
srs
k
2
before
FIGURE 1-4. MAC Transmitter
6
TL/F/10492– 5
1.0 Functional Description (Continued)
Serializer: After data has been written into the 32-byte
transmit FIFO, the serializer reads byte wide data from the
FIFO and sends a NRZ data stream to the Manchester encoder. The rate at which data is transmitted is determined
by the transmit clock (TXC). The serialized data is transmitted after the SFD.
Preamble Generator: The preamble generator prefixes a 7
byte alternating ‘‘1,0’’ pattern and a 1 byte ‘‘10101011’’
SFD pattern at the beginning of each packet. This allows
receiving nodes to synchronize to the incoming data. The
preamble is always transmitted in its entirety even in the
event of a collision. This assures that the minimum collision
fragment is 96 bits (64 bits of normal preamble, and 4 bytes,
or rather 32 bits, of the JAM pattern).
CRC Generator: The CRC generator calculates the 4-byte
FCS field from the transmitted serial data stream. If enabled, the 4-byte FCS field is appended to the end of the
transmitted packet (Section 2.6).
For bridging or switched ethernet applications the CRC
Generator can be inhibited by setting bit 13 in the Transmit
Control Register (Section 4.3.4). This feature is used when
an ethernet segment has already received a packet with a
CRC appended and needs to forward it to another ethernet
segment.
Jam Generator: The Jam generator produces a 4-byte pattern of all 1’s to assure that all nodes on the network sense
the collision. When a collision occurs, the SONIC stops
transmitting data and enables the Jam generator. If a collision occurs during the preamble, the SONIC finishes transmitting the preamble before enabling the Jam generator
(see Preamble Generator above).
1.3 DATA WIDTH AND BYTE ORDERING
The SONIC can be programmed to operate with either
32-bit or 16-bit wide memory. The data width is configured
during initialization by programming the DW bit in the Data
Configuration Register (DCR, Section 4.3.2). If the 16-bit
data path is selected, data is driven on pins D15–D0. The
SONIC also provides both Little Endian and Big Endian
byte-ordering capability for compatibility with National/Intel
or Motorola microprocessors respectively by selecting the
proper level on the BMODE pin. The byte ordering is depicted below.
Little Endian mode (National/Intel, BMODE
byte orientation for received and transmitted data in the Receive Buffer Area (RBA) and Transmit Buffer Area (TBA) of
system memory is as follows:
16-Bit Word
15870
Byte 1Byte 0
MSBLSB
32-Bit Long Word
3124231615870
Byte 3Byte 2Byte 1Byte 0
MSBLSB
Big Endian mode (Motorola, BMODE
entation for received and transmitted data in the RBA and
TBA is as follows:
16-Bit Word
15870
Byte 0Byte 1
LSBMSB
32-Bit Long Word
3124231615870
Byte 0Byte 1Byte 2Byte 3
LSBMSB
e
e
1): The byte ori-
0): The
FIGURE 1-5. Receive FIFO
7
TL/F/10492– 6
1.0 Functional Description (Continued)
1.4 FIFO AND CONTROL LOGIC
The SONIC incorporates two independent 32-byte FIFOs
for transferring data to/from the system interface and from/
to the network. The FIFOs, providing temporary storage of
data, free the host system from the real-time demands on
the network.
The way in which the FIFOS are emptied and filled is controlled by the FIFO threshold values and the Block Mode
Select bits (BMS, Section 4.3.2). The threshold values determine how full or empty the FIFOs can be before the
SONIC will request the bus to get more data from memory
or buffer more data to memory. When block mode is set, the
number of bytes transferred is set by the threshold value.
For example, if the threshold for the receive FIFO is 4
words, then the SONIC will always transfer 4 words from the
receive FIFO to memory. If empty/fill mode is set, however,
the number of bytes transferred is the number required to fill
the transmit FIFO or empty the receive FIFO. More specific
information about how the threshold affects reception and
transmission of packets is discussed in Sections 1.4.1 and
1.4.2 below.
1.4.1 Receive FIFO
To accommodate the different transfer rates, the receive
FIFO
(Figure 1-5 )
work (deserializer) interface and the 16/32-bit system interface. The FIFO is arranged as a 4-byte wide by 8 deep
memory array (8 long words, or 32 bytes) controlled by
three sections of logic. During reception, the Byte Ordering
logic directs the byte stream from the deserializer into the
FIFO using one of four write pointers. Depending on the
selected byte-ordering mode, data is written either least significant byte first or most significant byte first to accommodate little or big endian byte-ordering formats respectively.
As data enters the FIFO, the Threshold Logic monitors the
number of bytes written in from the deserializer. The programmable threshold (RFT1,0 in the Data Configuration
Register) determines the number of words (or long words)
written into the FIFO from the MAC unit before a DMA request for system memory occurs. When the threshold is
reached, the Threshold Logic enables the Buffer Management Engine to read a programmed number of 16- or 32-bit
words (depending upon the selected data width) from the
FIFO and transfers them to the system interface (the system memory) using DMA. The threshold is reached when
the number of bytes in the receive FIFO is greater than the
value of the threshold. For example, if the threshold is 4
words (8 bytes), then the Threshold Logic will not cause the
Buffer Management Engine to write to memory until there
are more than 8 bytes in the FIFO.
The Buffer Management Engine reads either the upper or
lower half (16 bits) of the FIFO in 16-bit mode or reads the
complete long word (32 bits) in 32-bit mode. If, after the
transfer is complete, the number of bytes in the FIFO is less
then the threshold, then the SONIC is done. This is always
the case when the SONIC is in empty/fill mode. If, however,
for some reason (e.g. latency on the bus) the number of
bytes in the FIFO is still greater than the threshold value,
the Threshold Logic will cause the Buffer Management Engine to do a DMA request to write to memory again. This
later case is usually only possible when the SONIC is in
block mode.
When in block mode, each time the SONIC requests the
bus, only a number of bytes equal to the threshold value will
be transferred. The Threshold Logic continues to monitor
serves as a buffer between the 8-bit net-
the number of bytes written in from the deserializer and enables the Buffer Management Engine every time the threshold has been reached. This process continues until the end
of the packet.
Once the end of the packet has been reached, the serializer
will fill out the last word (16-bit mode) or long word (32-bit
mode) if the last byte did not end on a word or long word
boundary respectively. The fill byte will be 0FFh. Immediately after the last byte (or fill byte) in the FIFO, the received
packets status will be written into the FIFO. The entire packet, including any fill bytes and the received packet status will
be buffered to memory. When a packet is buffered to memory by the Buffer Management Engine, it is always taken
from the FIFO in words or long words and buffered to memory on word (16-bit mode) or long word (32-bit mode)
boundaries. Data from a packet cannot be buffered on odd
byte boundaries for 16-bit mode, and odd word boundaries
for 32-bit mode (see Section 3.3). For more information on
the receive packet buffering process, see Section 3.4.
1.4.2 Transmit FIFO
Similar to the Receive FIFO, the Transmit FIFO
serves as a buffer between the 16/32-bit system interface
and the network (serializer) interface. The Transmit FIFO is
also arranged as a 4 byte by 8 deep memory array (8 long
words or 32 bytes) controlled by three sections of logic.
Before transmission can begin, the Buffer Management Engine fetches a programmed number of 16- or 32-bit words
from memory and transfers them to the FIFO. The Buffer
Management Engine writes either the upper or lower half
(16 bits) into the FIFO for 16-bit mode or writes the complete long word (32 bits) during 32-bit mode.
The Threshold logic monitors the number of bytes as they
are written into the FIFO. When the threshold has been
reached, the Transmit Byte Ordering state machine begins
reading bytes from the FIFO to produce a continuous byte
stream for the serializer. The threshold is met when the
number of bytes in the FIFO is greater than the value of the
threshold. For example, if the transmit threshold is 4 words
(8 bytes), the Transmit Byte Ordering state machine will not
begin reading bytes from the FIFO until there are 9 or more
bytes in the buffer. The Buffer Management Engine continues replenishing the FIFO until the end of the packet. It
does this by making multiple DMA requests to the system
interface. Whenever the number of bytes in the FIFO is
equal to or less than the threshold value, the Buffer Management Engine will do a DMA request. If block mode is set,
then after each request has been granted by the system,
the Buffer Management Engine will transfer a number of
bytes equal to the threshold value into the FIFO. If empty/fill
mode is set, the FIFO will be completely filled in one DMA
request.
Since data may be organized in big or little endian byte ordering format, the Transmit Byte Ordering state machine
uses one of four read pointers to locate the proper byte
within the 4 byte wide FIFO. It also determines the valid
number of bytes in the FIFO. For packets which begin or
end at odd bytes in the FIFO, the Buffer Management Engine writes extraneous bytes into the FIFO. The Transmit
Byte Ordering state machine detects these bytes and only
transfers the valid bytes to the serializer. The Buffer Management Engine can read data from memory on any byte
boundary (see Section 3.3). See Section 3.5 for more information on transmit buffering.
8
(Figure 1-6 )
1.0 Functional Description (Continued)
FIGURE 1-6. Transmit FIFO
1.5 STATUS AND CONFIGURATION REGISTERS
The SONIC contains a set of status/control registers for
conveying status and control information to/from the host
system. The SONIC uses these registers for loading commands generated from the system, indicating transmit and
receive status, buffering data to/from memory, and providing interrupt control. Each register is 16 bits in length. See
Section 4.0 for a description of the registers.
1.6 BUS INTERFACE
The system interface
essary for interfacing to a variety of buses. It includes the
I/O drivers for the data and address lines, bus access control for standard microprocessors, ready logic for synchronous or asynchronous systems, slave access control, interrupt control, and shared-memory access control. The functional signal groups are shown in
5.0 for a complete description of the SONIC bus interface.
1.7 LOOPBACK AND DIAGNOSTICS
The SONIC furnishes three loopback modes for self-testing
from the controller interface to the transceiver interface.
The loopback function is provided to allow self-testing of the
chip’s internal transmit and receive operations. During loopback, transmitted packets are routed back to the receive
section of the SONIC where they are filtered by the address
recognition logic and buffered to memory if accepted.
Transmit and receive status and interrupts remain active
during loopback. This means that when using loopback, it is
as if the packet was transmitted and received by two separate chips that are connected to the same bus and memory.
MAC Loopback: Transmitted data is looped back at the
MAC. Data is not sent from the MAC to either the internal
ENDEC or an external ENDEC (the external ENDEC interface pins will not be driven), hence, data is not transmitted
from the chip. Even though the ENDEC is not used in MAC
loopback, the ENDEC clock (an oscillator or crystal for the
(Figure 1-7 )
consists of the pins nec-
Figure 1-7
. See Section
TL/F/10492– 7
internal ENDEC or TXC for an external ENDEC) must be
driven. Network activity, such as a collision, does not affect
MAC loopback. CSMA/CD MAC protocol is not completely
followed in MAC loopback.
ENDEC Loopback: Transmitted data is looped back at the
ENDEC. If the internal ENDEC is used, data is switched
from the transmit section of the ENDEC to the receive section (
Figure 1-2
the collision lines, CD
ty does not affect ENDEC loopback. The LBK signal from
the MAC tells the internal ENDEC to go into loopback mode.
If an external ENDEC is used, it should operate in loopback
mode when the LBK signal is asserted. CSMA/CD MAC
protocol is followed even though data is not transmitted
from the chip.
Transceiver Loopback: Transmitted data is looped back at
the external transceiver (which is always the case regardless of the SONIC’s loopback mode). CSMA/CD MAC protocol is followed since data will be transmitted from the chip.
This means that transceiver loopback is affected by network
activity. In normal operations, the SONIC only monitors the
packet that is looped back by the transceiver, but does not
fill the receive FIFO and buffer the packet.
1.7.1 Loopback Procedure
The following procedure describes the loopback operation.
1. Initialize the Transmit and Receive Area as described in
Sections 3.4 and 3.5.
2. Load one of the CAM address registers (see Section 4.1),
with the Destination Address of the packet if you are verifying the SONIC’s address recognition capability.
3. Load one of the CAM address registers with the Source
Address of the packet if it is different than the Destination
Address to avoid getting a Packet Monitored Bad (PMB)
error in the Transmit status (see Section 4.3.4).
). Data is not transmitted from the chip and
g
, are ignored, hence, network activi-
9
1.0 Functional Description (Continued)
4. Program the Receive Control register with the desired receive filter and the loopback mode (LB1, LB0).
5. Issue the transmit command (TXP) and enable the receiver (RXEN) in the Command register.
The SONIC completes the loopback operation after the
packet has been completely received (or rejected if there is
an address mismatch). The Transmit Control and Receive
Control registers treat the loopback packet as in normal operation and indicate status accordingly. Interrupts are also
generated if enabled in the Interrupt Mask register.
Note: For MAC Loopback, only one packet may be queued for proper oper-
ation. This restriction occurs because the transmit MAC section,
which does not generate an Interframe Gap time (IFG) between
transmitted packets, does not allow the receive MAC section to update receive status. There are no restrictions for the other loopback
modes.
1.8 NETWORK MANAGEMENT FUNCTIONS
The SONIC fully supports the Layer Management IEEE
802.3 standard to allow a node to monitor the overall performance of the network. These statistics are available on a
per packet basis at the end of reception or transmission. In
addition, the SONIC provides three tally counters to tabulate
CRC errors, Frame Alignment errors, and missed packets.
Table 1-1 shows the statistics indicated by the SONIC.
*Note: DSACK0,1 are used for both Bus and Slave Access Control and are bidirectional. SMACK is used for both Slave access and shared memory access. The
BMODE pin selects between National/Intel or Motorola type buses.
TL/F/10492– 8
FIGURE 1-7. SONIC Bus Interface Signals
10
1.0 Functional Description (Continued)
TABLE 1-1. Network Management Statistics
StatisticRegister UsedBits Used
Frames Transmitted OKTCR (Note)PTX
Single Collision Frames(Note)NC0–NC4
Multiple Collision Frames(Note)NC0–NC4
Collision Frames(Note)NC0–NC4
Frames with Deferred TransmissionsTCR (Note)DEF
Late CollisionsTCR (Note)OWC
Excessive CollisionsTCR (Note)EXC
Excessive DeferralTCR (Note)EXD
Internal MAC Transmit ErrorTCR (Note)BCM, FU
Frames Received OKRCR (Note)PRX
Multicast Frames Received OKRCR (Note)MC
Broadcast Frames Received OKRCR (Note)BC
Frame Check Sequence ErrorsCRCTAll
Alignment ErrorsFAETAll
Frame Lost Due to Internal MAC Receive ErrorMPTAll
Note: The number of collisions and the contents of the Transmit Control register are posted in the TXpkt.status field (see
Section 3.5.1.2). The contents of the Receive Control register are posted in the RXpkt.status field (see Section 3.4.3).
2.0 Transmit/Receive IEEE 802.3 Frame Format
A standard IEEE 802.3 packet
following fields: preamble, Start of Frame Delimiter (SFD),
destination address, source address, length, data and
Frame Check Sequence (FCS). The typical format is shown
in
Figure 2-1
decoded by the ENDEC unit and transferred serially to/from
the MAC unit using NRZ data with a clock. All fields are of
fixed length except for the data field. The SONIC generates
and appends the preamble, SFD and FCS field during transmission. The Preamble and SFD fields are stripped during
reception. (The CRC is passed through to buffer memory
during reception.)
. The packets are Manchester encoded and
(Figure 2-1 )
consists of the
2.1 PREAMBLE AND START OF FRAME DELIMITER
(SFD)
The Manchester encoded alternating 1,0 preamble field is
used by the ENDEC to acquire bit synchronization with an
incoming packet. When transmitted, each packet contains
62 bits of an alternating 1,0 preamble. Some of this preamble may be lost as the packet travels through the network.
Byte alignment is performed when the Start of Frame Delimiter (SFD) pattern, consisting of two consecutive 1’s, is detected.
2.2 DESTINATION ADDRESS
The destination address indicates the destination of the
packet on the network and is used to filter unwanted pack-
RCRCRC
RCRFAE
ISRRFO
Note: Bebytes
bebitsTL/F/10492– 9
FIGURE 2-1. IEEE 802.3 Packet Structure
11
2.0 Transmit/Receive IEEE 802.3 Frame Format (Continued)
ets from reaching a node. There are three types of address
formats supported by the SONIC: Physical, Multicast, and
Broadcast.
Physical Address: The physical address is a unique address that corresponds only to a single node. All physical
addresses have the LSB of the first byte of the address set
to ‘‘0’’. These addresses are compared to the internally
stored CAM (Content Addressable Memory) address entries. All bits in the destination address must match an entry
in the CAM in order for the SONIC to accept the packet.
Multicast Address: Multicast addresses, which have the
LSB of the first byte of the address set to ‘‘1’’, are treated
similarly as Physical addresses, i.e., they must match an
entry in the CAM. This allows perfect filtering of Multicast
packets and eliminates the need for a hashing algorithm for
mapping Multicast packets.
Broadcast Address: If the address consists of all 1’s, it is a
Broadcast address, indicating that the packet is intended for
all nodes.
The SONIC also provides a promiscuous mode which allows reception of all physical address packets. Physical,
Multicast, Broadcast, and promiscuous address modes can
be selected via the Receive Control register.
2.3 SOURCE ADDRESS
The source address is the physical address of the sending
node. Source addresses cannot be multicast or broadcast
addresses. This field must be passed to the SONIC’s transmit buffer from the system software. During transmission,
the SONIC compares the Source address with its internal
CAM address entries before monitoring the CRC of the selfreceived packet. If the source address of the packet transmitted does not match a value in the CAM, the packet monitored bad flag (PMB) will be set in the transmit status field of
the transmit descriptor (see Sections 3.5.1.2 and 4.3.4). The
SONIC does not provide Source Address insertion. However, a transmit descriptor fragment, containing only the
Source Address, may be created for each packet. (See Section 3.5.1.)
2.4 LENGTH/TYPE FIELD
For IEEE 802.3 type packets, this field indicates the number
of bytes that are contained in the data field of the packet.
For Ethernet I and II networks, this field indicates the type of
packet. The SONIC does not operate on this field.
2.5 DATA FIELD
The data field has a variable octet length ranging from 46 to
1500 bytes as defined by the Ethernet specification. Messages longer than 1500 bytes need to be broken into multiple packets for IEEE 802.3 networks. Data fields shorter
than 46 bytes require appending a pad to bring the complete frame length to 64 bytes. If the data field is padded,
the number of valid bytes are indicated in the length field.
The SONIC does not append pad bytes for short packets
during transmission, nor check for oversize packets during
reception. However, the user’s driver software can easily
append the pad by lengthening the TXpkt.pktÐsize field
and TXpkt.fragÐsize field(s) to at least 64 bytes (see Section 3.5.1). While the Ethernet specification defines the
maximum number of bytes in the data field the SONIC can
transmit and receive packets up to 64k bytes in length.
2.6 FCS FIELD
The Frame Check Sequence (FCS) is a 32-bit CRC field
calculated and appended to a packet during transmission to
allow detection of error-free packets. During reception, an
error-free packet results in a specific pattern in the CRC
generator. The AUTODIN II (X
16
12
a
X
1
a
X
11
a
X
a
X
1) polynomial is used for the CRC calculations. The
SONIC may optionally append the CRC sequence during
transmission, and checks the CRC both during normal reception and self-reception during a transmission (see Section 1.2.1).
2.7 MAC (MEDIA ACCESS CONTROL) CONFORMANCE
The SONIC is designed to be compliant to the IEEE 802.3
MAC Conformance specification. The SONIC implements
most MAC functions in silicon and provides hooks for the
user software to handle the remaining functions. The MAC
Conformance specifications are summarized in Table 2-1.
TABLE 2-1. MAC Conformance Specifications
Conformance
Test Name
Minimum Frame SizeX
Maximum Frame SizeXX1
Address GenerationXX2
Address RecognitionX
Pad Length GenerationXX3
Start Of Frame DelimiterX
Length FieldX
Preamble GenerationX
Order of Bit TransmissionX
Inconsistent Frame LengthXX1
Non-Integral Octet CountX
Incorrect Frame Check
Sequence
Frame AssemblyX
FCS Generation and InsertionX
Carrier DeferenceX
Interframe SpacingX
Collision DetectionX
Collision HandlingX
Collision Backoff and
Retransmission
FCS ValidationX
Frame DisassemblyX
Back-to-Back FramesX
Flow ControlX
Attempt LimitX
Jam Size (after SFD)X
Jam Size (in Preamble)X
Note 1: The SONIC provides the byte count of the entire packet in the
RXpkt.byteÐcount (see Section 3.4.3). The user’s driver software may perform further filtering of the packet based upon the byte count.
Note 2: The SONIC does not provide Source Address insertion; however, a
transmit descriptor fragment, containing only the Source Address, may be
created for each packet. See Section 3.5.1.
Note 3: The SONIC does not provide Pad generation; however, the user’s
driver software can easily append the Pad by lengthening the TXpkt.pkt
size field and TXpkt.fragÐsize field(s) to at least 64 bytes. See Section
3.5.1.
12
32
26
23
a
a
10
a
X
X
8
7
a
X
5
a
X
X
22
a
X
a
a
X
4
2
a
a
X
X
Support By
User Driver
SONIC
Software
Notes
X
X
Ð
3.0 Buffer Management
3.1 BUFFER MANAGEMENT OVERVIEW
The SONIC’s buffer management scheme is based on separate buffers and descriptors (
ets that are received or transmitted are placed in buffers
called the Receive Buffer Area (RBA) and the Transmit Buffer Area (TBA). The system keeps track of packets in these
buffers using the information in the Receive Descriptor Area
(RDA) and the Transmit Descriptor Area (TDA). A single
(TDA) points to a single TBA, but multiple RDAs can point to
a single RBA (one RDA per packet in the buffer). The Receive Resource Area (RRA), which is another form of descriptor, is used to keep track of the actual buffer.
When packets are transmitted, the system sets up the packets in one or more TBAs with a TDA pointing to each TBA.
There can only be one packet per TBA/TDA pair. A single
TBA, however, may be made up of several fragments of
data dispersed in memory. There is one TDA pointing to
each TBA which specifies information about the buffer’s
size, location in memory, number of fragments and status
after transmission. The TDAs are linked together in a linked
list. The system causes the SONIC to transmit the packets
by passing the first TDA to the SONIC and issuing the transmit command.
Before a packet can be received, an RDA and RBA must be
set up by the system. RDA’s are made up as a linked list
similar to TDAs. An RDA is not linked to a particular RBA,
though. Instead, an RDA is linked specifically to a packet
after it has been buffered into an RBA. More than one packet can be buffered into the same RBA, but each packet gets
its own RDA. A received packet can not be scattered into
fragments. The system only needs to tell the SONIC where
the first RDA and where the RBAs are. Since an RDA never
specifically points to an RBA, the RRA is used to keep track
of the RBAs. The RRA is a circular queue of pointers and
buffer sizes (not a linked list). When the SONIC receives a
packet, it is buffered into a RBA with a corresponding and
unique RDA that is written to so that it points to and describes the new packet. If the RBA does not have enough
space to buffer the next packet, a new RBA is obtained from
the RRA.
3.2 DESCRIPTOR AREAS
Descriptors are the basis of the buffer management scheme
used by the SONIC. A RDA points to a received packet
within a RBA, RRA points to a RBA and a TDA points to a
TBA which contains a packet to be transmitted. The conventions and registers used to describe these descriptors
are discussed in the next three sections.
3.2.1 Naming Convention for Descriptors
The fields which make up the descriptors are named in a
consistent manner to assist in remembering the usage of
each descriptor. Each descriptor name consists of three
components in the following format.
[
RX/TX][descriptor name].[field
The first two capital letters indicate whether the descriptor is
used for transmission (TX) or reception (RX), and is then
followed by the descriptor name having one of two names.
Figures 3-2
and
3-11
]
). Pack-
e
rsrc
Resource descriptor
e
pkt
Packet descriptor
The last component consists of a field name to distinguish it
from the other fields of a descriptor. The field name is separated from the descriptor name by a period (‘‘.’’). An example of a descriptor is shown below.
TL/F/10492– 86
3.2.2 Abbreviations
The abbreviations in Table 3.1 are used to describe the
SONIC registers and data structures in memory. The ‘‘0’’
and ‘‘1’’ in the abbreviations indicate the least and most
significant portions of the registers or descriptors. Table 3-1
lists the naming convention abbreviations for descriptors.
3.2.3 Buffer Management Base Addresses
The SONIC uses three areas in memory to store descriptor
information: the Transmit Descriptor Area (TDA), Receive
Descriptor Area (RDA), and the Receive Resource Area
(RRA). The SONIC accesses these areas by concatenating
a 16-bit base address register with a 16-bit offset register.
The base address register supplies a fixed upper 16 bits of
address and the offset registers provide the lower 16 bits of
address. The base address registers are the Upper Transmit Descriptor Address (UTDA), Upper Receive Descriptor
Address (URDA), and the Upper Receive Resource Address
(URRA) registers. The corresponding offset registers are
shown below.
Upper Address RegistersOffset Registers
See Table 3-1 for definition of register mnemonics.
Figure 3-1
Area and the Receive Descriptor Area being located by the
UTDA and URDA registers. The descriptor areas, RDA,
TDA, and RRA are allowed to have the same base address.
i.e., URRA
to prevent these areas from overwriting each other.
URRARSA, REA, RWP, RRP
URDACRDA
UTDACTDA
shows an example of the Transmit Descriptor
e
URDAeUTDA. Care, however, must be taken
13
3.0 Buffer Management (Continued)
TABLE 3-1. Descriptor Abbreviations
TRANSMIT AND RECEIVE AREAS
RRAReceive Resource Area
RDAReceive Descriptor Area
RBAReceive Buffer Area
TDATransmit Descriptor Area
TBATransmit Buffer Area
BUFFER MANAGEMENT REGISTERS
RSAResource Start Area Register
REAResource End Area Register
RRPResource Read Pointer Register
RWPResource Write Pointer Register
CRDACurrent Receive Descriptor
Address Register
CRBA0,1Current Receive Buffer Address
Register
TCBA0,1Temporary Current Buffer Address
Register
RBWC0,1Remaining Buffer Word Count
Register
TRBWC0,1Temporary Remaining Buffer Word
Count Register
EOBCEnd of Buffer Count Register
TPSTransmit Packet Size Register
TSA0,1Transmit Start Address Register
CTDACurrent Transmit Descriptor
Address Register
BUFFER MANAGEMENT REGISTERS (Continued)
TFCTransmit Fragment Count Register
TFSTransmit Fragment Size Register
UTDAUpper Transmit Descriptor
Address Register
URRAUpper Receive Resource Address
Register
URDAUpper Receive Descriptor Address
Register
TRANSMIT AND RECEIVE DESCRIPTORS
RXrsrc.buffÐptr0,1 Buffer Pointer Field in the RRA
RXrsrc.buffÐwc0,1 Buffer Word Count Fields in the
RRA
RXpkt.statusReceive Status Field in the RDA
RXpkt.byteÐcountPacket Byte Count Field in the
RDA
RXpkt.buffÐptr0,1Buffer Pointer Fields in the RDA
RXpkt.linkReceive Descriptor Link Field in
RDA
RXpkt.inÐuse‘‘In Use’’ Field in RDA
TXpkt.fragÐcountFragment Count Field in TDA
TXpkt.pktÐsizePacket Size Field in TDA
TXpkt.pktÐptr0,1Packet Pointer Fields in TDA
TXpkt.fragÐsizeFragment Size Field in TDA
TXpkt.linkTransmit Descriptor Link Field in
TDA
FIGURE 3-1. Transmit and Receive Descriptor Pointers
14
TL/F/10492– 10
3.0 Buffer Management (Continued)
3.3 DESCRIPTOR DATA ALIGNMENT
All fields used by descriptors (RXpkt.xxx, RXrsrc.xxx, and
TXpkt.xxx) are word quantities (16-bit) and must be aligned
to word boundaries (A0
word boundaries (A1,A0
ceive Buffer Area (RBA) must also be aligned to a word
boundary in 16-bit mode and a long word boundary in 32-bit
mode. The fragments in the Transmit Buffer Area (TBA),
however, may be aligned on any arbitrary byte boundary.
3.4 RECEIVE BUFFER MANAGEMENT
The Receive Buffer Management operates on three areas in
memory into which data, status, and control information are
written during reception
must be initialized (Section 3.4.4) before enabling the receiver (setting the RXEN bit in the Command register). The
receive resource area (RRA) contains descriptors that locate receive buffer areas in system memory. These descriptors are denoted by R1, R2, etc. in
noted by P1, P2, etc.) can then be buffered into the corresponding RBAs. Depending on the size of each buffer area
and the size of the packet(s), multiple or single packets are
buffered into each RBA. The receive descriptor area (RDA)
contains status and control information for each packet (D1,
D2, etc. in
packet (D1 goes with P1, D2 with P2, etc.).
When a packet arrives, the address recognition logic checks
the address for a Physical, Multicast, or Broadcast match
and if the packet is accepted, the SONIC buffers the packet
contiguously into the selected Receive Buffer Area (RBA).
Because of the previous end-of-packet processing, the
SONIC assures that the complete packet is written into a
single contiguous block. When the packet ends, the SONIC
writes the receive status, byte count, and location of the
packet into the Receive Descriptor Area (RDA). The SONIC
then updates its pointers to locate the next available descriptor and checks the remaining words available in the
RBA. If sufficient space remains, the SONIC buffers the
next packet immediately after the previous packet. If the
current buffer is out of space the SONIC fetches a Resource descriptor from the Receive Resource Area (RRA)
acquiring an additional buffer that has been previously allocated by the system.
Figure 3-2
e
0) for 16-bit memory and to long
e
0,0) for 32-bit memory. The Re-
(Figure 3-2 )
) corresponding to each received
. These three areas
Figure 3-2
. Packets (de-
3.4.1 Receive Resource Area (RRA)
As buffer memory is consumed by the SONIC for storing
data, the Receive Resource Area (RRA) provides a mechanism that allows the system to allocate additional buffer
space for the SONIC. The system loads this area with resource descriptors that the SONIC, in turn, reads as its current buffer space is used up. Each resource descriptor consists of a 32-bit buffer pointer locating the starting point of
the RBA and a 32-bit Word Count that indicates the size of
the buffer in words (2 bytes per word). The buffer pointer
and word count are contiguously located using the format
shown in
16-bit fields. The SONIC stores this information internally
and concatenates the corresponding fields to create 32-bit
long words for the buffer pointer and word count. Note that
in 32-bit mode the upper word (D
the SONIC. This area may be used for other purposes since
the SONIC never writes into the RRA.
The SONIC organizes the RRA as a circular queue for efficient processing of descriptors. Four registers define the
RRA. The first two, the Resource Start Area (RSA) and the
Resource End Area (REA) registers, determine the starting
and ending locations of the RRA, and the other two registers update the RRA. The system adds descriptors at the
address specified by the Resource Write Pointer (RWP),
and the SONIC reads the next descriptor designated by the
Resource Read Pointer (RRP). The RRP is advanced 4
words in 16-bit mode (4 long words in 32-bit mode) after the
SONIC finishes reading the RRA and automatically wraps
around to the beginning of the RRA once the end has been
reached. When a descriptor in the RRA is read, the
RXrsc.buffÐpt0,1 is loaded into the CRBA0,1 registers and
the RXrsc.buffÐwc0,1 is loaded into the RBWC0,1 registers.
The alignment of the RRA is confined to either word or long
word boundaries, depending upon the data width mode. In
16-bit mode, the RRA must be aligned to a word boundary
(A0 is always zero) and in 32-bit mode, the RRA is aligned
to a long word boundary (A0 and A1 are always zero).
Figure 3-3
with each component composed of
k
31:16l) is not used by
FIGURE 3-2. Overview of Receive Buffer Management
15
TL/F/10492– 11
3.0 Buffer Management (Continued)
3.4.2 Receive Buffer Area (RBA)
The SONIC stores the actual data of a received packet in
the RBA. The RBAs are designated by the resource descriptors in the RRA as described above. The RXrsrc.buff
wc0,1 fields of the RRA indicate the length of the RBA.
When the SONIC gets a RBA from the RRA, the
RXrsrc.buffÐwc0,1 values are loaded into the Remaining
Buffer Word Count registers (RBWC0,1). These registers
keep track of how much space (in words) is left in the buffer.
When a packet is buffered in a RBA, it is buffered contiguously (the SONIC will not scatter a packet into multiple buffers or fragments). Therefore, if there is not enough space
left in a RBA after buffering a packet to buffer at least one
more maximum sized packet (the maximum legal sized
packet expected to be received from the network), a new
buffer must be acquired. The End of Buffer Count (EOBC)
register is used to tell the SONIC the maximum packet size
that the SONIC will need to buffer.
3.4.2.1 End of Buffer Count (EOBC)
The EOBC is a boundary in the RBA based from the bottom
of the buffer. The value written into the EOBC is the maximum expected size (in words) of the network packet that
the SONIC will have to buffer. This word count creates a line
in the RBA that, when crossed, causes the SONIC to fetch a
new RBA resource from the RRA.
Note: The EOBC is a word count, not a byte count. Also, the value pro-
grammed into EOBC must be a double word (32-bit) quantity when
the SONIC is in 32-bit mode (e.g. in 32-bit mode, EOBC should be set
to 758 words, not 759 words even though the maximum size of an
Ð
IEEE 802.3 packet is 759 words).
3.4.2.2 Buffering the Last Packet in an RBA
At the start of reception, the SONIC stores the packet beginning at the Current Receive Buffer Address (CRBA0,1)
and continues until the reception is complete. Concurrent
with reception, the SONIC decrements the Remaining Buffer Word Count (RBWC0,1) by one in 16-bit mode or by two
in 32-bit mode. At the end of reception, if the packet has
crossed the EOBC boundary, the SONIC knows that the
next packet might not fit in the RBA. This check is done by
comparing the RBWC0,1 registers with the EOBC. If
RBWC0,1 is less than the EOBC (the last packet buffered
has crossed the EOBC boundary), the SONIC fetches the
next resource descriptor in the RRA. If RBWC0,1 is greater
than or equal to the EOBC (the EOBC boundary has not
been crossed) the next packet reception continues at the
present location pointed to by CRBA0,1 in the same RBA.
Figure 3-4
t
illustrates the SONIC’s actions for (1) RBWC0,1
EOBC and (2) RBWC0,1kEOBC. See Section 3.4.4.4
for specific information about setting the EOBC.
Note: It is important that the EOBC boundary be ‘‘crossed.’’ In other words,
Ý
1in
Figure 3-4
case
occurs without case
k
EOBC will not work properly and the SONIC will not fetch a new
buffer. The result of this will be a buffer overflow (RBAE in the Interrupt Status Register, Section 4.3.6).
must exist before caseÝ2 exists. If caseÝ2
Ý
1 having occurred first, the test for RBWC0,1
FIGURE 3-3. Receive Resource Area Format
CaseÝ1
t
Case
Case
(RBWC0,1
Ý
Ý
EOBC)
1: SONIC buffers next packet in same RBA.
2: SONIC detects an exhausted RBA and will buffer the next packet in another RBA.
Ý
Case
(RBWC0,1
2
k
EOBC)
FIGURE 3-4. Receive Buffer Area
16
TL/F/10492– 12
TL/F/10492– 13
3.0 Buffer Management (Continued)
3.4.3 Receive Descriptor Area (RDA)
After the SONIC buffers a packet to memory, it writes 6
words of status and control information into the RDA, reads
the link field to the next receive descriptor, and writes to the
in-use field of the current descriptor. In 32-bit mode, the
upper word, D
memory should not be used for other purposes, since the
SONIC may still write into these locations. Each receive descriptor consists of the following sections (
receive status: indicates status of the received packet. The
SONIC writes the Receive Control register into this field.
Figure 3-6
loaded from the contents of the Receive Control register.
Note that ERR, RNT, BRD, PRO, and AMC are configuration bits and are programmed during initialization. See Section 4.3.3 for the description of the Receive Control register.
15141312111098
ERR RNT BRD PRO AMC LB1 LB0MC
7654 3 2 10
BC LPKT CRS COL CRCR FAER LBK PRX
byte count: gives the length of the complete packet from
the start of Destination Address to the end of FCS.
packet pointer: a 32-bit pointer that locates the packet in
the RBA. The SONIC writes the contents of the CRBA0,1
registers into this field.
sequence numbers: this field displays the contents of two
8-bit counters (modulo 256) that sequence the RBAs used
and the packets buffered. These counters assist the system
in determining when an RBA has been completely processed. The sequence numbers allow the system to tally the
packets that have been processed within a particular RBA.
There are two sequence numbers that describe a packet:
the RBA Sequence Number and the Packet Sequence
Number. When a packet is buffered to memory, the SONIC
maintains a single RBA Sequence Number for all packets in
an RBA and sequences the Packet Number for succeeding
packets in the RBA. When the SONIC uses the next RBA, it
increments the RBA Sequence Number and clears the
Packet Sequence Number. The RBA’s sequence counter is
not incremented when the read RRA command is issued in
the Command register. The format of the Receive Sequence Numbers are shown in
are reset during hardware reset or by writing zero to them.
k
31:16l, is not used. This unused area in
Figure 3-5
FIGURE 3-5. Receive Descriptor Format
TL/F/10492– 14
shows the receive status format. This field is
FIGURE 3-6. Receive Status Format
Figure 3-7
. These counters
).
15870
RBA Sequence NumberPacket Sequence Number
(Modulo 256)(Modulo 256)
FIGURE 3-7. Receive Sequence Number Format
receive link field: a 15-bit pointer (A15–A1) that locates
the next receive descriptor. The LSB of this field is the End
Of List (EOL) bit, and indicates the last descriptor in the list.
(Initialized by the system.)
in use field: this field provides a handshake between the
system and the SONIC to indicate the ownership of the descriptor. When the system avails a descriptor to the SONIC,
it writes a non-zero value into this field. The SONIC, in turn,
sets this field to all ‘‘0’s’’ when it has finished processing the
descriptor. (That is, when the CRDA register has advanced
to the next receive descriptor.) Generally, the SONIC releases control after writing the status and control information
into the RDA. If, however, the SONIC has reached the last
descriptor in the list, it maintains ownership of the descriptor
until the system has appended additional descriptors to the
list. The SONIC then relinquishes control after receiving the
next packet. (See Section 3.4.6.1 for details on when the
SONIC writes to this field). The receive packet descriptor
format is shown in
Figure 3-5
.
3.4.4 Receive Buffer Management Initialization
The Receive Resource, Descriptor, and Buffer areas (RRA,
RDA, RBA) in memory and the appropriate SONIC registers
must be properly initialized before the SONIC begins buffering packets. This section describes the initialization process.
3.4.4.1 Initializing The Descriptor Page
All descriptor areas (RRA, RDA, and TDA) used by the
SONIC reside within areas up to 32k (word) or 16k (long
word) pages. This page may be placed anywhere within the
32-bit address range by loading the upper 16 address lines
into the UTDA, URDA, and URRA registers.
3.4.4.2 Initializing The RRA
The initialization of the RRA consists of loading the four
SONIC RRA registers and writing the resource descriptor
information to memory.
The RRA registers are loaded with the following values.
Resource Start Area (RSA) register: The RSA is loaded
with the lower 16-bit address of the beginning of the RRA.
Resource End Area (REA) register: The REA is loaded
with the lower 16-bit address of the end of the RRA. The
end of the RRA is defined as the address of the last
RXrsrc.ptr0 field in the RRA plus 4 words in 16-bit mode or 4
long words in 32-bit mode (
Figure 3-3
).
Resource Read Pointer (RRP) register: The RRP is loaded with the lower 16-bit address of the first resource descriptor the SONIC reads.
Resource Write Pointer (RWP) register: The RWP is loaded with the lower 16-bit address of the next vacant location
where a resource descriptor will be placed by the system.
Note: The RWP register must only point to either (1) the RXrsrc.ptr0 field of
one of the RRA Descriptors, (2) the memory address that the RSA
points to (the start of the RRA), or (3) the memory address that the
REA points to (the end of the RRA). When the RWP
son is made, it is performed after the complete RRA descriptor has
been read and not during the fetch. Failure to set the RWP to any of
the above values prevents the RWP
becoming true.
e
e
RRP compari-
RRP comparison from ever
17
3.0 Buffer Management (Continued)
All RRA registers are concatenated with the URRA register
for generating the full 32-bit address.
The resource descriptors that the system writes to the RRA
consists of four fields: (1) RXrsrc.buffÐptr0, (2)
RXrsrc.buffÐptr1,(3)RXrsrc.buffÐwc0,and(4)
RXrsrc.buffÐwc1. The fields must be contiguous (they cannot straddle the end points) and are written in the order
shown in
denote the least and most significant portions for the Buffer
Pointer and Word Count. The first two fields supply the 32bit starting location of the Receive Buffer Area (RBA), and
the second two define the number of 16-bit words that the
RBA occupies.
Note that two restrictions apply to the Buffer Pointer and
Word Count. First, in 32-bit mode, since the SONIC always
writes long words, an even count must be written to
RXrsrc.buffÐwc0. Second, the Buffer Pointer must either
be pointing to a word boundary in 16-bit mode (A0
long word boundary in 32-bit mode (A0,A1
that the descriptors must be properly aligned in the RRA as
discussed in Section 3.3.
Figure 3-8
. The ‘‘0’’ and ‘‘1’’ in the descriptors
e
e
0) or a
0,0). Note also
the SONIC to begin receive processing at the first descriptor. An example of two descriptors linked together is shown
in
Figure 3-9
played in bold type. The other fields are written by the
SONIC after a packet is accepted. The RXpkt.inÐuse field
is first written by the system, and then by the SONIC. Note
that the descriptors must be aligned properly as discussed
in Section 3.3. Also note that the URDA register is concatenated with the CRDA register to generate the full 32-bit address.
. The fields initialized by the system are dis-
FIGURE 3-8. RRA Initialization
After configuring the RRA, the RRA Read command (setting
RRRA bit in the Command register) may be given. This
command causes the SONIC to read the RRA descriptor in
a single block operation, and load the following registers
(see Section 4.2 for register mnemonics):
When the command has completed, the RRRA bit in the
Command register is reset to ‘‘0’’. Generally this command
is only issued during initialization. At all other times, the RRA
is automatically read as the SONIC finishes using an RBA.
3.4.4.3 Initializing The RDA
To accept multiple packets from the network, the receive
packet descriptors must be linked together via the
RXpkt.link fields. Each link field must be written with a 15-bit
(A15–A1) pointer to locate the beginning of the next descriptor in the list. The LSB of the RXpkt.link field is the End
of List (EOL) bit and is used to indicate the end of the descriptor list. EOL
the first or middle descriptors. The RXpkt.inÐuse field indicates whether the descriptor is owned by the SONIC. The
system writes a non-zero value to this field when the descriptor is available, and the SONIC writes all ‘‘0’s’’ when it
finishes using the descriptor. At startup, the Current Receive
Descriptor Address (CRDA) register must be loaded with the
address of the first RXpkt.status field in order for
w
RXrsrc.buffÐptr0
w
RXrsrc.buffÐptr1
w
RXrsrc.buffÐwc0
w
RXrsrc.buffÐwc1
e
1 for the last descriptor and EOLe0 for
TL/F/10492– 15
FIGURE 3-9. RDA Initialization Example
TL/F/10492– 16
3.4.4.4 Initializing the Lower Boundary of the RBA
A ‘‘false bottom’’ is set in the RBA by loading the End Of
Buffer Count (EOBC) register with a value equal to the maximum size packet in words (16 bits) that may be received.
This creates a lower boundary in the RBA. Whenever the
Remaining Buffer Word Count (RBWC0,1) registers decrement below the EOBC register, the SONIC buffers the next
packet into another RBA. This also guarantees that a packet is always contiguously buffered into a single Receive
Buffer Area (RBA). The SONIC does not buffer a packet into
multiple RBAs. Note that in 32-bit mode, the SONIC holds
the LSB always low so that it properly compares with the
RBWC0,1 registers.
After a hardware reset, the EOBC reset, the EOBC register
is automatically initialized to 2F8h (760 words or 1520
bytes). For 32-bit applications this is the suggested value for
EOBC. EOBC defaults to 760 words (1520 bytes) instead of
759 words (1518 bytes) because 1518 is not a double word
(32-bit) boundary (see Section 3.4.2.1). If the SONIC is used
in 16-bit mode, then EOBC should be set to 759 words
(1518 bytes) because 1518 is a word (16-bit) boundary.
Sometimes it may be desired to buffer a single packet per
RBA. When doing this, it is important to set EOBC and the
buffer size correctly. The suggested practice is to set EOBC
to a value that is at least 4 bytes, in 32-bit mode, or 2 bytes,
in 16-bit mode, less than the buffer size. An example of this
for 32-bit mode is to set EOBC to 760 words (1520 bytes)
18
3.0 Buffer Management (Continued)
and the buffer size to 762 words (1524 bytes). A similar
example for 16-bit mode would be EOBC
(1518 bytes) and the buffer size set to 760 words (1520
bytes). The buffer can be any size, but as long as the EOBC
is 2 words, for 32-bit mode, or 1 word, for 16-bit mode, less
than the buffer size, only one packet will be buffered in that
RBA.
Note 1: It is possible to filter out most oversized packets by setting the buff-
er size to 760 words (1520 bytes) in 32-bit mode or 759 words (1518
bytes) in 16-bit mode. EOBC would be set to 758 words (1516
bytes) for both cases. With this configuration, any packet over 1520
bytes, in 32-bit mode, or 1518 bytes, in 16-bit mode, will not be
completely buffered because the packet will overflow the buffer.
When a packet overflow occurs, a Receive Buffer Area Exceeded
interrupt (RBAE in the Interrupt Status Register, Section 4.3.6) will
occur.
Note 2: When buffering one packet per buffer, it is suggested that the val-
ues in Note 1 above be used. Since the minimum legal sized Ethernet packet is 64 bytes, however, it is possible to set EOBC as much
as 64 bytes less than the buffer size and still end up with one packet
per buffer.
Figure 3-10
shows this ‘‘range.’’
3.4.5 Beginning Of Reception
At the beginning of reception, the SONIC checks its internally stored EOL bit from the previous RXpkt.link field for a
‘‘1’’. If the SONIC finds EOL
e
1, it recognizes that after the
previous reception, there were no more remaining receive
packet descriptors. It re-reads the same RXpkt.link field to
check if the system has updated this field since the last
reception. If the SONIC still finds EOL
es. (See Section 3.5 for adding descriptors to the list.) Otherwise, the SONIC begins storing the packet in the RBA
starting at the Current Receive Buffer Address (CRBA0,1)
registers and continues until the packet has completed.
Concurrent with the packet reception, the Remaining Buffer
Word Count (RBWC0,1) registers are decremented after
each word is written to memory. This register determines
the remaining words in the RBA at the end of reception.
3.4.6 End Of Packet Processing
At the end of a reception, the SONIC enters its end of packet processing sequence to determine whether to accept or
reject the packet based on receive errors and packet size.
At the end of reception the SONIC enters one of the following two sequences:
Ð Successful reception sequence
Ð Buffer recovery for runt packets or packets with errors
e
759 words
e
1, reception ceas-
3.4.6.1 Successful Reception
If the SONIC accepts the packet, it first writes 5 words of
descriptor information in the RDA beginning at the address
pointed to by the Current Receive Descriptor Address
(CRDA) register. It then reads the RXpkt.link field to advance the CRDA register to the next receive descriptor. The
SONIC also checks the EOL bit for a ‘‘1’’ in this field. If
e
EOL
1, no more descriptors are available for the SONIC.
The SONIC recovers the address of the current RXpkt.link
field (from a temporary register) and generates a ‘‘Receive
Descriptors Exhausted’’ indication in the Interrupt Status
register. (See Section 3.4.7 on how to add descriptors.) The
SONIC maintains ownership of the descriptor by not writing
to the RXpkt.inÐuse field. Otherwise, if EOL
e
0, the SONIC
advances the CRDA register to the next descriptor and resets the RXpkt.inÐuse field to all ‘‘0’s’’.
The SONIC accesses the complete 7 word RDA descriptor
in a single block operation.
The SONIC also checks if there is remaining space in the
RBA. The SONIC compares the Remaining Buffer Word
Count (RBWC0,1) registers with the static End Of Buffer
Count (EOBC). If the RBWC is less than the EOBC, a maximum sized packet will no longer fit in the remaining space in
the RBA; hence, the SONIC fetches a resource descriptor
from the RRA and loads its registers with the pointer and
word count of the next available RBA.
3.4.6.2 Buffer Recovery For Runt Packets Or
Packets With Errors
If a runt packet (less than 64 bytes) or packet with errors
arrives and the Receive Control register has been configured to not accept these packets, the SONIC recovers its
pointers back to the original positions. The CRBA0,1 registers are not advanced and the RBWC0,1 registers are not
decremented. The SONIC recovers its pointers by maintaining a copy of the buffer address in the Temporary Receive
Buffer Address registers (TRBA0,1). The SONIC recovers
the value in the RBWC0,1 registers from the Temporary
Buffer Word Count registers (TBWC0,1).
3.4.7 Overflow Conditions
When an overflow condition occurs, the SONIC halts its
DMA operations to prevent writing into unauthorized memory. The SONIC uses the Interrupt Status register (ISR) to
indicate three possible overflow conditions that can occur
Range of EOBCe(RXrsrc.wc0,1b2 to RXrsrc.wc0,1b32)
FIGURE 3-10. Setting EOBC for Single Packet RBA
19
TL/F/10492– 17
3.0 Buffer Management (Continued)
when its receive resources have been exhausted. The system should respond by replenishing the resources that have
been exhausted. These overflow conditions (Descriptor Resources Exhausted, Buffer Resources Exhausted, and RBA
Limit Exceeded) are indicated in the Interrupt Status register
and are detailed as follows:
Descriptor Resources Exhausted: This occurs when the
SONIC has reached the last receive descriptor in the list,
meaning that the SONIC has detected EOL
must supply additional descriptors for continued reception.
The system can do this in one of two ways: 1) appending
descriptors to the existing list, or 2) creating a separate list.
1) Appending descriptors to the existing list. This is the easiest and preferred way. To do this, the system, after creating the new list, joins the new list to the existing list by
simply writing the beginning address of the new list into
the RXpkt.link field and setting EOL
reception, the SONIC re-reads the last RXpkt.link field,
and updates its CRDA register to point to the next descriptor.
2) Creating a separate list. This requires an additional step
because the lists are not joined together and requires
that the CRDA register be loaded with the address of the
RXpkt.link field in the new list.
During this overflow condition, the SONIC maintains ownership of the descriptor (RXpkt.inÐuse
the system to add additional descriptors to the list. When
the system appends more descriptors, the SONIC releases
ownership of the descriptor after writing 0000h to the
RXpkt.inÐuse field.
Buffer Resources Exhausted: This occurs when the
SONIC has detected that the Resource Read Pointer (RRP)
and Resource Write Pointer (RWP) registers are equal (i.e.,
all RRA descriptors have been exhausted). The RBE bit in
the Interrupt Status register is set when the SONIC finishes
using the second to last receive buffer and reads the last
RRA descriptor. Actually, the SONIC is not truly out of resources, but gives the system an early warning of an impending out of resources condition. To continue reception
after the last RBA is used, the system must supply additional RRA descriptor(s), update the RWP register, and clear
the RBE bit in the ISR. The SONIC rereads the RRA after
this bit is cleared.
RBA Limit Exceeded: This occurs when a packet does not
completely fit within the remaining space of the RBA. This
can occur if the EOBC register is not programmed to a value
greater than the largest packet that can be received. When
this situation occurs, the packet is truncated and the SONIC
reads the RRA to obtain another RBA. Indication of an RBA
limit being exceeded is signified by the Receive Buffer Area
Exceeded (RBAE) interrrupt being set (see Section 4.3.6).
An RDA will not be set up for the truncated packet and the
buffer space will not be re-used. To rectify this potential
overflow condition, the EOBC register must be loaded with a
value equal to or greater than the largest packet that can be
accepted. (See Section 3.4.2.)
3.5 TRANSMIT BUFFER MANAGEMENT
To begin transmission, the system software issues the
Transmit command (TXP
e
1 in the CR). The Transmit Buffer Management uses two areas in memory for transmitting
packets
(Figure 3-11),
the Transmit Descriptor Area (TDA)
e
1. The system
e
0. At the next
i
00h) and waits for
and the Transmit Buffer Area (TBA). During transmission,
the SONIC fetches control information from the TDA, loads
its appropriate registers, and then transmits the data from
the TBA. When the transmission is complete, the SONIC
writes the status information in the TDA. From a single
transmit command, packets can either be transmitted singly
or in groups if several descriptors have been linked together.
FIGURE 3-11. Overview of Transmit Buffer Management
TL/F/10492– 18
3.5.1 Transmit Descriptor Area (TDA)
The TDA contains descriptors that the system has generated to exchange status and control information. Each descriptor corresponds to a single packet and consists of the
following 16-bit fields.
TXpkt.status: This field is written by the SONIC and provides status of the transmitted packet. (See Section 3.5.1.2
for more details.)
TXpkt.config: This field allows programming the SONIC to
one of the various transmit modes. The SONIC reads this
field and loads the corresponding configuration bits (PINT,
POWC, CRCI, and EXDIS) into the Transmit Control register. (See Section 3.5.1.1 for more details.)
TXpkt.pktÐsize: This field contains the byte count of the
entire packet.
TXpkt.fragÐcount: This field contains the number of fragments the packet is segmented into.
TXpkt.fragÐptr0,1: This field contains a 32-bit pointer
which locates the packet fragment to be transmitted in the
Transmit Buffer Area (TBA). This pointer is not restricted to
any byte alignment.
TXpkt.fragÐsize: This field contains the byte count of the
packet fragment. The minimum fragment size is 1 byte.
TXpkt.link: This field contains a 15-bit pointer (A15 –A1) to
the next TDA descriptor. The LSB, the End Of List (EOL) bit,
indicates the last descriptor in the list when set to a ‘‘1’’.
When descriptors have been linked together, the SONIC
transmits back-to-back packets from a single transmit command.
The data of the packet does not need to be contiguous, but
can exist in several locations (fragments) in memory. In this
case, the TXpkt.fragÐcount field is greater than one, and
additional TXpkt.fragÐptr0,1 and TXpkt.fragÐsize fields
corresponding to each fragment are used. The descriptor
format is shown in
upper word, D
Figure 3-12.
k
31:16l, is not used.
Note that in 32-bit mode the
20
3.0 Buffer Management (Continued)
FIGURE 3-12. Transmit Descriptor Area
3.5.1.1 Transmit Configuration
The TXpkt.config field allows the SONIC to be programmed
into one of the transmit modes before each transmission. At
the beginning of each transmission, the SONIC reads this
field and loads the PINT, POWC, CRCI, and EXDIS bits into
the Transmit Control register (TCR). The configuration bits
in the TCR correspond directly with the bits in the
TXpkt.config field as shown in
4.3.4 for the description on the TCR.
15141312111098
PINT POWC CRCI EXDISXXXX
7654321 0
XXXXXXX X
Note: xedon’t care
3.5.1.2 Transmit Status
At the end of each transmission the SONIC writes the status
bits (
the number of collisions experienced during the transmission into the TXpkt.status field
served). Bits NC4-NC0 indicate the number of collisions
where NC4 is the MSB. See Section 4.3.4 for the description of the TCR.
15141312111098
NC4 NC3 NC2 NC1 NC0 EXD DEF NCRS
CRSL EXC OWC res PMB FUBCM PTX
3.5.2 Transmit Buffer Area (TBA)
The TBA contains the fragments of packets that are defined
by the descriptors in the TDA. A packet can consist of a
single fragment or several fragments, depending upon the
fragment count in the TDA descriptor. The fragments also
can reside anywhere within the full 32-bit address range,
and be aligned to any byte boundary. When an odd byte
boundary is given, the SONIC automatically begins reading
data at the corresponding word boundary in 16-bit mode or
a long word boundary in 32-bit mode. The SONIC ignores
the extraneous bytes which are written into the FIFO during
FIGURE 3-13. TXpkt.config Field
k
10:0l) of the Transmit Control Register (TCR) and
76543210
FIGURE 3-14. TXpkt.status Field
Figure 3-13.
(Figure 3-14
TL/F/10492– 19
See Section
, resere-
odd byte alignment fragments. The minimum allowed fragment size is 1 byte.
tween the TDA and the TBA for single and multi-fragmented
packets.
3.5.3 Preparing To Transmit
All fields in the TDA descriptor and the Current Transmit
Descriptor Address (CTDA) register of the SONIC must be
initialized before the Transmit Command (setting the TXP bit
in the Command register) can be issued. If more than one
packet is queued, the descriptors must be linked together
with the TXpkt.link field. The last descriptor must have
e
EOL
1 and all other descriptors must have EOLe0. To
begin transmission, the system loads the address of the first
TXpkt.status field into the CTDA register. Note that the upper 16-bits of address are loaded in the Upper Transmit
Descriptor (UTDA) register. The user performs the following
transmit initialization.
1) Initialize the TDA
2) Load the CTDA register with the address of the first
transmit descriptor
3) Issue the transmit command
Note that if the Source Address of the packet being transmitted is not in the CAM, the Packet Monitored Bad (PMB)
bit in the TXpxt.status field will be set (see Section 6.3.4).
3.5.3.1 Transmit Process
When the Transmit Command (TXP
register) is issued, the SONIC fetches the control information in the TDA descriptor, loads its appropriate registers
(shown below) and begins transmission. (See Section 4.2
for register mnemonics.)
TCR
w
TXpkt.config
TPS
w
TXpkt.pktÐsize
TFC
w
TXpkt.fragÐcount
TSA0
w
TSA1
TFS
CTDA
(CTDA is loaded after all fragments have been read and
successfully transmitted. If the halt transmit command is issued (HTX bit in the Command register is set) the CTDA
register is not loaded.)
During transmission, the SONIC reads the packet descriptor
in the TDA and transmits the data from the TBA. If
TXpkt.fragÐcount is greater than one, the SONIC, after finishing transmission of the fragment, fetches the next
TXpkt.fragÐptr0,1 and TXpkt.fragÐsize fields and transmits
the next fragment. This process continues until all fragments of a packet are transmitted. At the end of packet
transmission, status is written in to the TXpkt.status field.
The SONIC then reads the TXpkt.link field and checks if
EOL
and transmits the next packet. If EOL
erates a ‘‘Transmission Done’’ indication in the Interrupt
Status register and resets the TXP bit in the Command register.
In the event of a collision, the SONIC recovers its pointer in
the TDA and retransmits the packet up to 15 times. The
SONIC maintains a copy of the CTDA register in the Temporary Transmit Descriptor Address (TTDA) register.
The SONIC performs a block operation of 6, 3, or 2 accesses in the TDA, depending on where the SONIC is in the
transmit process. For the first fragment, it reads the
TXpkt.fragÐptr0
w
TXpkt.fragÐptr1
w
TXpkt.fragÐsize
w
TXpkt.link
e
0. If it is ‘‘0’’, the SONIC fetches the next descriptor
Figure
3-11 shows the relationship be-
e
1 in the Command
e
1 the SONIC gen-
21
3.0 Buffer Management (Continued)
TXpkt.config to TXpkt.fragÐsize (6 accesses). For the next
fragment, if any, it reads the next 3 fields from TXpkt.frag
ptr0 to TXpkt.fragÐsize (3 accesses). At the end of transmission it writes the status information to TXpkt.status and
reads the TXpkt.link field (2 accesses).
3.5.3.2 Transmit Completion
The SONIC stops transmitting under two conditions. In the
normal case, the SONIC transmits the complete list of descriptors in the TDA and stops after it detects EOL
the second case, certain transmit errors cause the SONIC
to abort transmission. If FIFO Underrun, Byte Count Mismatch, Excessive Collision, or Excessive Deferral (if enabled) errors occur, transmission ceases. The CTDA register points to the last packet transmitted. The system can
also halt transmission under software control by setting the
HTX bit in the Command register. Transmission halts after
the SONIC writes to the TXpkt.status field.
3.5.4 Dynamically Adding TDA Descriptors
Descriptors can be dynamically added during transmission
without halting the SONIC. The SONIC can also be guaranteed to transmit the complete list including newly appended
descriptors (barring any transmit abort conditions) by observing the following rule: The last TXpkt.link field must
point to the next location where a descriptor will be added
(see step 3 below and
Figure 3-15
). The procedure for ap-
pending descriptors consists of:
1. Creating a new descriptor with its TXpkt.link pointing to
the next vacant descriptor location and its EOL bit set to
a ‘‘1’’.
2. Resetting the EOL bit to a ‘‘0’’ of the previously last descriptor.
3. Re-issuing the Transmit command (setting the TXP bit in
the Command register).
Step 3 assures that the SONIC will transmit all the packets
in the list. If the SONIC is currently transmitting, the Transmit command has no effect and continues transmitting until
it detects EOL
e
1. If the SONIC had just finished transmitting, it continues transmitting from where it had previously
stopped.
FIGURE 3-15. Initializing Last Link Field
e
1. In
TL/F/10492– 20
4.0 SONIC Registers
The SONIC contains two sets of registers: The status/con-
Ð
trol registers and the CAM memory cells. The status/control
registers are used to configure, control, and monitor SONIC
operation. They are directly addressable registers and occupy 64 consecutive address locations in the system memory
space (selected by the RA5 –RA0 address pins). There are
a total of 64 status/control registers divided into the following categories:
User Registers: These registers are accessed by the user
to configure, control, and monitor SONIC operation. These
are the only SONIC registers the user needs to access.
ure 4-3
shows the programmer’s model and Table 4-1 lists
Fig-
the attributes of each register.
Internal Use Registers: These registers (Table 4-2) are
used by the SONIC during normal operation and are not
intended to be accessed by the user.
National Factory Test Registers: These registers (Table
4-3) are for National factory use only and should never be
accessed by the user. Accessing these registers during normal operation can cause improper functioning of the
SONIC.
4.1 THE CAM UNIT
The CAM unit memory cells are indirectly accessed by programming the CAM descriptor area in system memory and
issuing the LCAM command (setting the LCAM bit in the
Control register). The CAM cells do not occupy address locations in register space and, thus, are not accessible
through the RA5–RA0 address pins. The CAM control registers, however, are part of the user register set and must be
initialized before issuing the LCAM command (see Section
4.3.10).
The Content Addressable Memory (CAM) consists of sixteen 48-bit entries for complete address filtering
(Figure 4-1)
of network packets. Each entry corresponds to a 48-bit destination address that is user programmable and can contain
any combination of Multicast or Physical addresses. Each
entry is partitioned into three 16-bit CAM cells accessible
through CAM Address Ports (CAP 2, CAP 1 and CAP 0) with
CAP0 corresponding to the least significant 16 bits of the
Destination Address and CAP2 corresponding to the most
significant bits. The CAM is accessed in a two step process.
First, the CAM Entry Pointer is loaded to point to one of the
16 entries. Then, each of the CAM Address Ports is accessed to select the CAM cell. The 16 user programmable
CAM entries can be masked out with the CAM Enable register (see Section 4.3.10).
Note: It is not necessary to program a broadcast address into the CAM
when it is desired to accept broadcast packets. Instead, to accept
broadcast packets, set the BRD bit in the Receive Control register. If
the BRD bit has been set, the CAM is still active. This means that it is
possible to accept broadcast packets at the same time as accepting
packets that match physical addresses in the CAM.
4.1.1 The Load CAM Command
Because the SONIC uses the CAM for a relatively long period of time during reception, it can only be written to via the
CAM Descriptor Area (CDA) and is only readable when the
22
4.0 SONIC Registers (Continued)
FIGURE 4-1. CAM Organization
SONIC is in software reset. The CDA resides in the same
64k byte block of memory as the Receive Resource Area
(RRA) and contains descriptors for loading the CAM registers. These descriptors are contiguous and each descriptor
consists of four 16-bit fields
upper word, D
k
31:16l, is not used. The first field contains
(Figure 4-2).
In 32-bit mode the
the value to be loaded into the CAM Entry Pointer and the
remaining fields are for the three CAM Address Ports (see
Section 4.3.10). In addition, there is one more field after the
last descriptor containing the mask for the CAM Enable register. Each of the CAM descriptors are addressed by the
CAM Descriptor Pointer (CDP) register.
After the system has initialized the CDA, it can issue the
Load CAM command to program the SONIC to read the
CDA and load the CAM. The procedure for issuing the Load
CAM command is as follows.
1. Initialize the Upper Receive Resource Address (URRA)
register. Note that the CAM Descriptor Area must reside
within the same 64k Page as the Receive Resource
Area. (See Section 4.3.9).
TL/F/10492– 21
2. Initialize the CDA as described above.
3. Initialize the CAM Descriptor Count with the number of
CAM descriptors. Note, only the lower 5 bits are used in
this register. The other bits are don’t cares. (See Section
4.3.10).
4. Initialize the CAM Descriptor Pointer to locate the first
descriptor in the CDA. This register must be reloaded
each time a new Load CAM command is issued.
5. Issue the Load CAM command (LCAM) in the Command
register. (See Section 4.3.1).
If a transmission or reception is in progress, the CAM DMA
function will not occur until these operations are complete.
When the SONIC completes the Load CAM command, the
CDP register points to the next location after the CAM enable field and the CDC equals zero. The SONIC resets the
LCAM bit in the Command register and sets the Load CAM
Done (LCD) bit in the ISR.
FIGURE 4-2. CAM Descriptor Area Format
23
TL/F/10492– 22
4.0 SONIC Registers (Continued)
k
l
5:0
RA
0h Command RegisterStatus and C Control Fields
1 Data Configuration RegisterControl Fields
Status and
Control Registers
Transmit
Registers
Receive
Registers
CAM
Registers
2 Receive Control RegisterStatus and Control Fields
3 Transmit Control RegisterStatus and Control Fields
4 Interrupt Mask RegisterMask Fields
5 Interrupt Status RegisterStatus Fields
$
3F Data Configuration Register 2Control Fields
6 Upper Transmit Descriptor Address Register Upper 16-bit Address Base
7 Current Transmit Descriptor Address Register Lower 16-bit Address Offset
Ð
0D Upper Receive Descriptor Address Register Upper 16-bit Address Base
0E Current Receive Descriptor Address Register Lower 16-bit Address Offset
14 Upper Receive Resource Address RegisterUpper 16-bit Address Base
This set of registers is used to convey status/control information to/from the host system and to control the operation
of the SONIC. These registers are used for loading commands generated from the system, indicating transmit and
receive status, buffering data to/from memory, and provid-
ing interrupt control. The registers are selected by asserting
chip select to the SONIC and providing the necessary address on register address pins RA5–RA0. Tables 4-1, 4-2,
and 4-3 show the locations of all SONIC registers and
where information on the registers can be found in the data
sheet.
Description
(section)
25
4.0 SONIC Registers (Continued)
TABLE 4-1. User Registers (Continued)
RA5–RA0AccessRegisterSymbol
WATCHDOG COUNTERS
29R/WWatchdog Timer 0WT04.3.12
2AR/WWatchdog Timer 1WT14.3.12
SILICON REVISION
28RSilicon RevisionSR4.3.13
Note 1: These registers can only be read when the SONIC is in reset mode (RST bit in the CR is set). The SONIC gives invalid data when these registers are read in
non-reset mode.
Note 2: This register can only be written to when the SONIC is in reset mode. This register is normally only loaded by the Load CAM command.
Note 3: The Data Configuration registers, DCR and DCR2, can only be written to when the SONIC is in reset mode (RST bit in CR is set). Writing to these registers
while not in reset mode does not alter the registers.
Note 4: The data written to these registers is inverted before being latched. That is, if a value of FFFFh is written, these registers will contain and read back the
value of 0000h. Data is not inverted during a read operation.
TABLE 4-2. Internal Use Registers (Users should not write to these registers)
Note 1: The data that is read from these registers is the inversion of what has been written to them.
Note 2: The value that is written to this register is shifted once in 16-bit mode and shifted twice in 32-bit mode.
TABLE 4-3. Internal Use Registers (Users should not access these registers)
(RA5–RA0)AccessRegisterSymbol
30These registers are for factory use only. Users must not
#
R/Waddress these registers as improper SONIC operationnonenone
3Ecan occur.
Description
(section)
Description
(section)
Description
(section)
26
4.0 SONIC Registers (Continued)
4.3 REGISTER DESCRIPTION
4.3.1 Command Register
k
l
(RA
This register
ing bits for the function. For all bits, except for the RST bit, the SONIC resets the bit after the command is completed. With the
exception of RST, writing a ‘‘0’’ to any bit has no effect. Before any commands can be issued, the RST bit must first be reset to
‘‘0’’. This means that, if the RST bit is set, two writes to the Command Register are required to issue a command to the SONIC;
one to clear the RST bit, and one to issue the command.
This register also controls the general purpose 32-bit Watchdog Timer. After the Watchdog Timer register has been loaded, it
begins to decrement once the ST bit has been set to ‘‘1’’. An interrupt is issued when the count reaches zero if the Timer
Complete interrupt is enabled in the IMR.
During hardware reset, bits 7, 4, and 2 are set to a ‘‘1’’; all others are cleared. During software reset bits 9, 8, 1, and 0 are
cleared and bits 7 and 2 are set to a ‘‘1’’; all others are unaffected.
e
5:0
0h)
(Figure 4-4
) is used for issuing commands to the SONIC. These commands are issued by setting the correspond-
Setting this bit causes the SONIC to load the CAM with the descriptor that is pointed to by the CAM Descriptor
Pointer register.
Note: This bit must not be set during transmission (TXP is set). The SONIC will lock up if both bits are set simultaneously.
8RRRA: READ RRA
Setting this bit causes the SONIC to read the next RRA descriptor pointed to by the Resource Read Pointer (RRP)
register. Generally this bit is only set during initialization. Setting this bit during normal operation can cause improper
receive operation.
7RST: SOFTWARE RESET
Setting this bit resets all internal state machines. The CRC generator is disabled and the Tally counters are halted,
but not cleared. The SONIC becomes operational when this bit is reset to ‘‘0’’. A hardware reset sets this bit to a ‘‘1’’.
It must be reset to ‘‘0’’ before the SONIC becomes operational.
6Must be 0.
5ST: START TIMER
Setting this bit enables the general-purpose watchdog timer to begin counting or to resume counting after it has
been halted. This bit is reset when the timer is halted (i.e., STP is set). Setting this bit resets STP.
4STP: STOP TIMER
Setting this bit halts the general-purpose watchdog timer and resets the ST bit. The timer resumes when the ST bit is
set. This bit powers up as a ‘‘1’’. Note: Simultaneously setting bits ST and STP stops the timer.
27
4.0 SONIC Registers (Continued)
4.3 REGISTER DESCRIPTION (Continued)
4.3.1 Command Register (Continued)
BitDescription
3RXEN: RECEIVER ENABLE
Setting this bit enables the receive buffer management engine to begin buffering data to memory. Setting this bit
resets the RXDIS bit. Note: If this bit is set while the MAC unit is currently receiving a packet, both RXEN and RXDIS
are set until the network goes inactive (i.e., the SONIC will not start buffering in the middle of a packet being
received).
2RXDIS: RECEIVER DISABLE
Setting this bit disables the receiver from buffering data to memory or the Receive FIFO. If this bit is set during the
reception of a packet, the receiver is disabled only after the packet is processed. The RXEN bit is reset when the
receiver is disabled. Tally counters remain active regardless of the state of this bit.
Note: If this bit is set while the SONIC is currently receiving a packet, both RXEN and RXDIS are set until the packet is fully received. When both
RXEN and RXDIS are set, RXDIS could be cleared by writing zero to it.
1TXP: TRANSMIT PACKET(S)
Setting this bit causes the SONIC to transmit packets which have been set up in the Transmit Descriptor Area (TDA).
The SONIC loads its appropriate registers from the TDA, then begins transmission. The SONIC clears this bit after
any of the following conditions have occurred: (1) transmission had completed (i.e., after the SONIC has detected
e
EOL
1), (2) the Halt Transmission command (HTX) has taken effect, or (3) a transmit abort condition has
occurred. This condition occurs when any of the following bits in the TCR have been set: EXC, EXD, FU, or BCM.
This bit must not be set if a Load CAM operation is in progress (LCAM is set). The SONIC will lock up if both bits are
set simultaneously.
0HTX: HALT TRANSMISSION
Setting this bit halts the transmit command after the current transmission has completed. TXP is reset after
transmission has halted. The Current Transmit Descriptor Address (CTDA) register points to the last descriptor
transmitted. The SONIC samples this bit after writing to the TXpkt.status field.
28
4.0 SONIC Registers (Continued)
4.3.2 Data Configuration Register
k
l
(RA
This register
During a hardware reset, bits 15 and 13 are cleared; all other bits are unaffected. (Because of this, the first thing the driver
software does to the SONIC should be to set up this register.) All bits are unaffected by a software reset. This register must only
be accessed when the SONIC is in reset mode (i.e., the RST bit is set in the Command register).
BitsDescription
15EXBUS: EXTENDED BUS MODE
14Must be 0.
13LBR: LATCHED BUS RETRY
12, 11PO1, PO0: PROGRAMMABLE OUTPUTS
e
5:0
1h)
(Figure 4-5)
establishes the bus cycle options for reading/writing data to/from 16- or 32-bit memory systems.
EXBUSEXTENDED BUS MODE
LBRLATCHED BUS RETRY
PO0,PO1PROGRAMMABLE OUTPUTS
SBUSSYNCHRONOUS BUS MODE
USR0, USR1USER DEFINABLE PINS
WC0, WC1WAIT STATE CONTROL
DWDATA WIDTH SELECT
BMSBLOCK MODE SELECT FOR DMA
RFT0, RFT1RECEIVE FIFO THRESHOLD
TFT0, TFT1TRANSMIT FIFO THRESHOLD
Setting this bit enables the Extended Bus mode which enables the following:
1)Extended Programmable Outputs, EXUSR
external ENDEC interface into four programmable user outputs, EXUSR
k
USR
1:0l. These outputs are programed with bits 15-12 in the DCR2 (see Section 4.3.7). On hardware reset,
k
3:0l: This changes the TXD, LBK, RXC and RXD pins from the
k
3:0lrespectively, which are similar to
these four pins will be TRI-STATE and will remain that way until the DCR is changed. If EXBUS is enabled, then
these pins will remain TRI-STATE until the SONIC becomes a bus master, at which time they will be driven according
to the DCR2. If EXBUS is disabled, then these four pins work normally as external ENDEC interface pins.
2)Synchronous Termination, STERM
synchronous memory termination input for compatibility with Motorola style processors. This input is only useful
when Asynchronous Bus mode is selected (bit 10 below is set to ‘‘0’’) and BMODE
: This changes the TXC pin from the External ENDEC interface into a
e
1 (Motorola mode). On
hardware reset, this pin will be TRI-STATE and will remain that way until the DCR is changed. If EXBUS is enabled,
this pin will remain TRI-STATE until the SONIC becomes a bus master, at which time it will become the STERM
input. If EXBUS is disabled, then this pin works normally as the TXC pin for the external ENDEC interface.
3)Asynchronous Bus Retry: Causes BRT
to be clocked in asynchronously off the falling edge of bus clock. This only
applies, however, when the SONIC is operating in asynchronous mode (bit 10 below is set to ‘‘0’’). If EXBUS is not
set, XTO (BRT) is sampled synchronously off the rising edge of bus clock. (See Section 5.4.6.)
The LBR bit controls the mode of operation of the BRT
signal (see pin description). It allows the BUS Retry operation
to be latched or unlatched.
0:Unlatched mode: The assertion of BRT
The SONIC will retry the operation when BRT
forces the SONIC to finish the current DMA operation and get off the bus.
is deserted.
1:Latched mode: The assertion of BRT forces the SONIC to finish the current DMA operation as above, however, the
SONIC will not retry until BRT
is deasserted, the BR bit in the ISR (see Section 4.3.6) has been reset, and BRT is
deasserted. Hence, the mode has been latched on until the BR bit is cleared.
Note: Unless LBR is set to a ‘‘1’’, BRT must remain asserted at least until the SONIC has gone idle. See Section 5.4.6 and the timing for Bus Retry
in section 7.0.
The PO1,PO0 bits individually control the USR1,0 pins respectively when SONIC is a bus master (HLDA or BGACK
active). When PO1/PO0 are set to a 1 the USR1/USR0 pins are high during bus master operations and when these
bits are set to a 0 the USR1/USR0 pins are low during bus master operations.
is
29
4.0 SONIC Registers (Continued)
4.3.2 Data Configuration Register (Continued)
BitsDescription
10SBUS: SYNCHRONOUS BUS MODE
The SBUS bit is used to select the mode of system bus operation when SONIC is a bus master. This bit selects the internal
ready line to be either a synchronous or asynchronous input to SONIC during block transfer DMA operations.
0: Asynchronous mode. RDYi
at the falling edge of the bus clock (T2 of the DMA cycle). No setup or hold times need to be met with respect to this edge
to guarantee proper bus operation. The minimum memory cycle time is 3 bus clocks.
1: Synchronous mode. RDYi (BMODEe0) and DSACK0,1 (BMODEe1) must respectively meet the setup and
hold times with respect to the rising edge of T1 or T2 to guarantee proper bus operation.
9, 8 USR1,0: USER DEFINABLE PINS
The USR1,0 bits report the level of the USR1,0 signal pins, respectively, after a chip hardware reset. If the USR1,0 signal pins
are at a logical 1 (tied to V
to ground) during a hardware reset the USR1,0 bits are set to a 0. These bits are latched on the rising edge of RST
they remain set/reset until the next hardware reset.
7, 6 WC1,0: WAIT STATE CONTROL
These encoded bits determine the number of additional bus cycles (T2 states) that are added during each DMA cycle.
WC1WC0Bus Cycles Added
000
011
102
113
5DW: DATA WIDTH SELECT
These bits select the data path width for DMA operations.
DW Data Width
016-bit
132-bit
4BMS: BLOCK MODE SELECT FOR DMA
Determines how data is emptied or filled into the Receive or Transmit FIFO.
0: Empty/fill mode: All DMA transfers continue until either the Receive FIFO has emptied or the Transmit FIFO has
filled completely.
1: Block mode: All DMA transfers continue until the programmed number of bytes (RFT0, RFT1 during reception or TF0,
TF1 during transmission) have been transferred. (See note for TFT0, TFT1.)
3, 2 RFT1,RFT0: RECEIVE FIFO THRESHOLD
These encoded bits determine the number of words (or long words) that are written into the receive FIFO from the MAC unit
before a receive DMA request occurs. (See Section 1.4.)
LB1LB0Function
002 words or 1 long word (4 bytes)
014 words or 2 long words (8 bytes)
108 words or 4 long words (16 bytes)
1112 words or 6 long words (24 bytes)
Note: In block mode (BMS bite1), the receive FIFO threshold sets the number of words (or long words) written to memory during a receive DMA block cycle.
1, 0 TFT1,TFT0: TRANSMIT FIFO THRESHOLD
These encoded bits determine the minimum number of words (or long words) the DMA section maintains in the transmit
FIFO. A bus request occurs when the number of words drops below the transmit FIFO threshold. (See Section 1.4.)
LB1LB0Function
004 words or 2 long words (8 bytes)
018 words or 4 long words (16 bytes)
1012 words or 6 long words (24 bytes)
1114 words or 7 long words (28 bytes)
Note: In block mode (BMSe1), the number of bytes the SONIC reads in a single DMA burst equals the transmit FIFO threshold value. If the number of words
or long words needed to fill the FIFO is less than the threshold value, then only the number of reads required to fill the FIFO in a single DMA burst will be made.
Typically, with the FIFO threshold value set to 12 or 14 words, the number of memory reads needed is less than the FIFO threshold value.
(BMODEe0) or DSACK0,1 (BMODEe1) are respectively internally synchronized
) during a hardware reset the USR1,0 bits are set to a 1. If the USR1,0 pins are at a logical 0 (tied
CC
. Once set
30
Loading...
+ 68 hidden pages
You need points to download manuals.
1 point = 1 manual.
You can buy points or you can get point for every manual you upload.