US20040225814A1 - Method and apparatus for constructing wired-AND bus systems - Google Patents

Abstract

A large multimaster I²C bus system is partitioned into smaller bus segments. The bus segments are connected by bridges that isolate the segments and direct selected transactions and commands between the segments. By programming address bitmaps that are internal to each bridge, transactions can pass through the bridges so that the various bus segments appear to be one logical bus. Because each bridge implements address filtering so that transactions are selectively forwarded from one side of the bridge to the other based on the contents of an internal address bitmap, I²C slave addresses can be arbitrarily populated on either side of the bridge. Duplicate I²C slave addresses can be also used on different segments of a single logical I²C bus system. Masters on one segment can reach devices connected to the same bus segment and can also reach devices with duplicate addresses on other bus segments by using a tunnel command addressed to a bridge.

Description

FIELD OF THE INVENTION
• This invention relates to buses using a wired-AND protocol and, more particularly, to methods and apparatus for constructing complex multimaster bus systems.[0001]
BACKGROUND OF THE INVENTION
• The I[0002]²C bus system is an electronic bus for carrying commands and data between compatible devices connected to the bus. The system was developed and marketed by Philips Semiconductor Corporation and is described in detail in theI²C Specification, revision 2.0, Philips Semiconductor Corporation 1998, which specification is hereby incorporated by reference in its entirety. In the I²C bus system, two wires, called a serial data (SDA) line and serial clock (SCL) line, carry information between the devices connected to the bus. Both the SDA and SCL lines are bi-directional lines, connected to a positive supply voltage via pull-up resistors as shown in FIG. 1 to form a “wired-AND” configuration. For example, in thebus configuration100 illustrated in FIG. 1, theSDA line108 and theSCL line110 are connected to the V_DDsupply line102 by pull-up resistors104 and106, respectively. Other buses, which use a similar protocol, include the SMBus, Access bus and the InfiniBand^SM management link. Collectively, this type of bus system will be termed a “wired-AND” bus system. The remainder of the discussion will focus on the I²C bus system with the understanding that the discussion applies to these other bus systems as well.
• When the[0003]bus101 is free, both theSDA line108 and theSCL line110 are pulled to a “HIGH” state by theresistors104 and106. The output stages of devices connected to the bus must have an open-drain or open-collector in order to form the wired-AND configuration. Twodevices112 and114 are shown schematically in FIG. 1.Device112 has a clock output stage which includesoutput transistor116 which is connected across theSCL line110 andground118. When a signal on thegate117 oftransistor116 turns the transistor on, it pulls theSCL line110 “LOW.” Clock signals on theSCL line110 can be detected by means ofbuffer120 whose output forms the “clock in”line122.
• Similarly,[0004]device112 has a data output stage which includesoutput transistor124 which is connected across theSDA line108 andground126. When a signal on thegate123 oftransistor124 turns the transistor on, it pulls theSDA line108 “LOW.” Data signals on theSDA line108 can be detected by means ofbuffer128 whose output forms the “data in”line130.Device114 also has aclock output transistor132 andclock input buffer134 and adata output transistor136 anddata input buffer138 for communication with the SDA and SCL lines,108 and110.
• Devices on the bus communicate by periodically pulling the SDA and[0005]SCL lines108 and110 LOW producing data and clock pulses on thelines108 and110. In accordance with the I²C protocol, the data on theSDA line108 must be stable when the clock line SCL110 is HIGH. Thus, the HIGH or LOW state of thedata line108 can only change when theclock line110 is LOW. Two unique situations arise, which situations are defined as START and STOP conditions. In particular, a HIGH to LOW transition on theSDA line108 while theSCL line110 is HIGH is defined as a START condition. A LOW to HIGH transition on theSDA line108 while theSCL line110 is HIGH is defined as a STOP condition.
• Each[0006]device112,114 on thebus101 has a unique address and can operate as either a data transmitter or a data receiver, depending on the function of the device. For example, an LCD driver is always a data receiver, whereas a memory can both receive and transmit data. In addition to being transmitters and receivers, devices can also be bus masters or slaves when performing data transfers. A bus master is the device that initiates a data transfer on the bus, generates the clock signals required for that transfer and terminates that data transfer. During this transfer, any other device to which data is sent, or from which data is received, is considered a slave. The bus master may transmit data to a slave or receive data from a slave. In both cases, the clock signals are generated by the bus master. Bus master and slave relationships are not permanent and depend on which device initiated the data transfer at a given time.
• More than one bus master device can be connected to[0007]bus101. Bus implementations with exactly one device capable of acting as a master are called single-master buses, while those with two or more devices capable of acting as bus masters are called multimaster buses. In a single-master bus system, the I²C protocol is very straightforward, with every transaction consisting of a START condition followed by one or more address and data phases, followed by a STOP condition. Thus, the START and STOP conditions frame all activity on the bus and hence define the duration during which the bus is busy.
• A slave device responds to an address or data phase generated by the master with either an acknowledgement (ACK) or a negative-acknowledgement (NAK) response. An ACK response is represented as a LOW signal level on the[0008]SDA line108 during the acknowledge bit time, which is defined as the ninth clock pulse of any address or data phase. A NAK response is represented as a HIGH signal level on theSDA line108 during the acknowledge bit time. Since the I²C bus is a wired-AND configuration where the buses are always HIGH unless pulled LOW by a device, a NAK response is equivalent to no response from a slave device. A NAK response during an address phase may indicate that the slave device is busy and unable to accept I²C transactions at this time, that it is non-functional or simply missing.
• While a simple single-master system using the I[0009]²C technology works well, the situation becomes much more complicated when much larger and more complex I²C subsystems involving multiple unique master devices and dozens of slave devices on a single bus are constructed. Perhaps the biggest challenge in the design of any large multi-master I²C subsystem is the difficulty of ensuring reliable operation in the presence of multiple master implementations designed by different vendors. Since most I²C devices are used in simple single-master bus implementations, many available I²C master devices which claim to support multi-master operation have not been tested and verified sufficiently to be used together reliably on a single I²C bus. Furthermore, the wired-AND nature of the I²C bus means that a failure of any one device can cause the entire bus to fail, leading to difficulties in isolating the cause and nature of the fault that caused the bus failure.
• In addition, one of the fundamental signal integrity challenges in any large I[0010]²C system design is meeting the rise-time specification of one microsecond on the SDA and SCL signals. Because these are open-collector signals, usually with a simple pull-up resistor to the supply rail, the rise time is proportional to the total capacitance on the bus. Further, the strength of the pull-up resistor is limited by the maximum current that the output cells on I²C components can sink, which is stated in the aforementioned I²C specification as 3 ma. Thus, the total capacitance that an I²C bus segment can tolerate is approximately 400 picofarads, beyond which the rise-time specification cannot be met. These design constraints limit both the number of master and slave devices on the bus as well as the physical length of the bus, making large I²C system designs very challenging.
• Therefore there is need for a method and apparatus to reliably construct complicated I[0011]²C systems and to isolate faults so that they can be eliminated.
SUMMARY OF THE INVENTION
• In accordance with the principles of the invention, a large multimaster I[0012]²C bus system is partitioned into smaller bus segments. The bus segments are connected by bridges that isolate the segments and direct selected transactions and commands between the segments. By programming address bitmaps that are internal to each bridge, transactions can pass through the bridges so that the various bus segments appear to be one logical bus. The bus topology can be designed to maximize the ability to isolate faults within a given segment, thereby improving the ability of technicians to diagnose problems in very large I²C implementations. In addition, since the bus segments are much smaller than the overall bus system, it becomes much easier to meet the required rise time specification.
• Because each bridge implements address filtering so that transactions are selectively forwarded from one side of the bridge to the other based on the contents of an internal address bitmap, I[0013]²C slave addresses can be arbitrarily populated on either side of the bridge.
• In accordance with one embodiment, duplicate I[0014]²C slave addresses can be used on different segments of a single logical I²C bus system. Masters on one segment can reach devices connected to the same bus segment and can also reach devices with duplicate addresses on other bus segments by using a tunnel command addressed to a bridge. The tunnel command contains an address and causes the bridge to which it is addressed to forward the contained address.
• In still another embodiment, complex I[0015]²C bus hierarchies can be constructed using a tree topology by programming a set of registers in the bridges. In such a topology, I²C masters can communicate with slave devices without regard to the relative location of the master and slave. In accordance with the inventive principles, is the transactions will be forwarded as needed across levels of bus hierarchy to complete a communication path between master and slave.
BRIEF DESCRIPTION OF THE DRAWINGS
• The above and further advantages of the invention may be better understood by referring to the following description in conjunction with the accompanying drawings in which:[0016]
• FIG. 1 is a schematic electrical diagram of the conventional I[0017]²C bus system illustrating the manner of driving information onto the bus system and receiving information from the bus system.
• FIG. 2A is a detailed block schematic diagram of an inventive unidirectional bridge apparatus implemented with a microcontroller and connected between two I[0018]²C domains.
• FIG. 2B is a block schematic diagram of the some of the contents of the RAM memory in the microcontroller of FIG. 2A.[0019]
• FIGS. 3A-3I shown the contents of various registers in the bridge apparatus.[0020]
• FIG. 4 is a block schematic diagram of two unidirectional bus bridge apparatus connected in parallel as a bidirectional bridge between two I[0021]²C systems.
• FIG. 5 is a time diagram showing the sequence of command signals generated by a bus master and used to control a bridge.[0022]
• FIG. 6 is a time diagram showing the sequence of command signals generated by a bus master during a write register access bridge command.[0023]
• FIG. 7 is a time diagram showing the sequence of command signals generated by a bus master during a read register access bridge command.[0024]
• FIG. 8 is a time diagram showing the sequence of command signals generated by a bus master during a tunnel bridge command.[0025]
• FIGS. 9A and 9B, when placed together, form a flowchart illustrating illustrative steps in a procedure for the configuration of bridges by a configuration host in an I[0026]²C system.
• FIG. 10 is a block schematic diagram showing an exemplary system that is configured in accordance with the procedure set forth in FIGS. 9A and 9B.[0027]
• FIGS. 11A, 11B and[0028]11C, when placed together, form a flowchart illustrating a detailed procedure for the configuration of the system shown in FIG. 10.
• FIG. 12 is a block schematic diagram showing the exemplary system of FIG. 10 after configuration in accordance with the procedure set forth in FIGS. 11A-11C.[0029]
• FIG. 13 is a flowchart illustrating illustrative steps in a procedure for completing the configuration of bridges by a configuration host in an I[0030]²C system.
DETAILED DESCRIPTION
• The invention concerns a bus bridge apparatus and method that buffers I[0031]²C transactions generated on one I²C bus by a bus master and retransmits the transactions on another bus segment. During this retransmission, the I²C address may also be filtered and translated. In one embodiment, the bridge is implemented by one or more unidirectional devices; there are several different configurations in which the bridge devices can be used. For example, a single bridge device could be used to provide electrical isolation and loading isolation between two I²C domains where the transaction flow from one domain to the other is always unidirectional. Alternatively, multiple bridge devices could be placed as “peers” on a top-level bus with the system's main I²C controller, connecting to a set of downstream I²C buses, thus implementing a tree topology.
• In a single unidirectional bridge application, two I[0032]²C domains are separated by a bridge constructed, as described below, with a single microcontroller chip. Since the bridge constructed in this manner is a unidirectional bridge, transactions may pass only in one direction—from a port-A side of the bridge to a port-B side. However, the direction of data flow can be bidirectional, allowing both reads and writes. Used in this manner, the bridge device can act as a sort of “firewall.” For example, suppose an I²C implementation contains multiple masters, and one of these masters is not multimaster capable. By placing the nonmultimaster-capable master by itself on the port-A bus, and connecting the other masters and all slave devices on the port-B bus, the nonmultimaster device on the port-A bus would be able to communicate with all the devices on the port-B bus (through the bridge), but would be freed from the burden of handling the many complexities of multimaster I²C. A detailed disclosure of a firewall device constructed in this manner is set forth in U.S. patent application Ser. No. 09/630,099, entitled METHOD AND APPARATUS FOR CONNECTING SINGLE MASTER DEVICES TO A MULTIMASTER WIRED-AND BUS ENVIRONMENT, filed on Aug. 1, 2000 by Joseph J. Ervin and Jorge E. Lach, the disclosure of which is hereby incorporated by reference in its entirety.
• As discussed above, in a preferred embodiment, the bridge device is implemented in a programmable microcontroller, but other implementations are possible. A microcontroller, which is suitable for use with the invention, is the device 87LPC764, manufactured and sold by Philips Semiconductor Corporation. This controller is programmed with firmware, and hence the preferred embodiment is a combined hardware and software implementation. In addition to other hardware resources, this microcontroller has one multimaster I[0033]²C interface. Clearly, however, the inventive apparatus needs two I²C interfaces to fulfill its function as a bridge. Thus, the microcontroller's built-in I²C interface is used for the port-B bus and the port-A bus is implemented using two General Purpose Input/Output (GPIO) pins on the microcontroller and a software “bit-bang” driver.
• FIG. 2A shows how a[0034]microcontroller201 is used to connect two I²C domains202 and204. The87LPC764 controller201 includes 4 KB of One-Time-Programmable (OTP) internal memory that is used as program code space. Thus, the bridge software is embedded within the87LPC764 controller201. The87LPC764 controller201 runs with a 20 MHZ input clock provided bycrystal222 andcapacitors224 and226, and has an internal machine cycle which is ⅙^thof the input clock, or approximately 3.3 MHz. The pin numbers of the 87LPC764 controller are denoted next to their respective leads. Power is applied to pin15, vialead236, and ground is applied topin5, vialead228.
• The[0035]87LPC764 microcontroller201 includes one multimaster-capable I²C interface. In the present application, this port is reserved for the port-B interface214. This I²C interface presents the software running on the 87LPC764 with a 1-bit wide data path onto the I²C bus, so software must interact with the register interface on a bit-by-bit basis. Although this requires more software to drive the I²C interface than with byte-oriented I²C devices, the bit-oriented interface is actually preferred in that it allows software a much finer level of control over thebus212,216. This is particularly useful during bus recovery procedures after a fatal protocol error that causes the bus to hang.
• The port-[0036]A interface208 receives transactions from the I²C domain #0202 via the SCL_0 andSDA_0 lines206 and210 that are normally held HIGH byresistor pair218. The port-B interface214 drives the SCL_1 andSDA_1 lines212 and216 that are normally held HIGH byresistor pair220 and communicate with I²C domain #1204.
• In order to reliably detect new START conditions on the port-[0037]A interface208 while finishing up the end of a previous I²C transaction on the port-B interface214, the port-A SDA_0 line210 is connected to themicrocontroller201 external interrupt input onpin8, vialead211, in addition to being connected to the appropriate port-A GPIO pin20. This interrupt is enabled only when the port-A bus206,210 is not busy, i.e. immediately after a reset, and between STOP and START conditions on the port-A bus206,210. Thus, the initial START condition of any port-A transaction is detected via this interrupt mechanism, while all other port-A activity for the remainder of a transaction is detected via polling.
• The bridge apparatus also incorporates an array located in[0038]RAM238. The array is used as an address bitmap indicating which I²C addresses reside on the port-A interface and which are on port-B interface, thereby permitting the bridge to implement address replication and filtering. Address replication and filtering involves receiving an address from a first bus segment and selectively transmitting the address on another bus segment under control of the address bitmap. Replication of slave addresses may be useful, for example, if a system implementation includes multiple masters, each with some number of identical slave devices. Replication can be used, for example, in a system with two intelligent masters, each with an identical I²C EEPROM device used for local nonvolatile storage. If the masters never need to communicate directly with the EEPROM of the other master, the system can be implemented with two I²C bus segments connected by a bidirectional bridge implemented as discussed below. By configuring the address bitmaps in the bridge correctly, the addresses of the EEPROM devices could be local to each bus segment, but the masters could still communicate with one another. Furthermore, any devices that have unique addresses could be placed on either segment, with access from the remote master provided through the bridge.
• FIG. 2B shows some of the contents of the[0039]RAM238 including theaddress bitmap240 and a set ofinternal registers242 that are described in detail below. Also included is acommand interpreter244, which receives and parses commands addressed to the bridge. This interpreter would normally be implemented in firmware and its operation is discussed in detail below.
• The address filter bitmap comprises 32 bytes. Each of the 32 bytes, in turn, comprises 256 bits, which gives one bit per I[0040]²C address. Using this bitmap, a special master called a “configuration host” can control on an address-by-address basis, which addresses are ignored by a given bridge, and which are forwarded transparently from the port-A bus to the port-B bus. The operation of the configuration host and the manner in which it controls the bridges is discussed in detail below.
• The relationship between bits in the bitmap and I[0041]²C address is that first byte in the bitmap maps I²C addresses 00h through 07h (“h” stands for hexadecimal notation), with bit<0>corresponding to address 00h, bit<1>corresponding to I²C address 01h, and so on. In accordance with conventional I²C specifications, even numbered bits correspond to write addresses, and odd numbered bits correspond to read addresses. In order to address a bridge itself, the address is either 62h or 63h depending on whether the transaction is a write or read.
• The bridge will forward a transaction if the corresponding bit in the bitmap is a “1.” However, the address filtering mechanism will not pass transactions addressed to the bridge's own I[0042]²C address of 62h or 63h. These addresses, however, may be reached on the other side of a bridge through a procedure called “transaction tunneling” that is discussed in detail below. Transaction tunneling allows a master to send a special command to a bridge address. The command contains another address. When a bridge receives this command it transmits the command to the address contained within the command. This arrangement allows software on a configuration master, for example, to tunnel a transaction through a bridge to probe for other bridges.
• The transaction tunneling mechanism provides a way for masters to communicate with devices on remote segments if the devices have addresses identical to a device on the masters local segment. For example, in some system implementations, it may be necessary to have multiple I[0043]²C devices at the same addresses and for one or more masters in the system to be able to uniquely access each of these devices. One such anticipated application of this type will be a computer chassis containing InfiniBand^SM expansion slots for plug-in Target Channel Adapters (TCAs). All TCA cards are required to have a number of specified slave devices such as temperature sensors and EEPROM devices, and an intelligent master device. The addresses of these devices are identical from TCA to TCA. The InfiniBand^SM application requires that the I²C master in the chassis have communication with each TCA, and for at least one TCA per chassis to be able to communicate with the host I²C master in the chassis. Furthermore, the TCA cards are all hot pluggable, meaning that they may be plugged in while the chassis is powered up, and possibly while I²C activity is present on the host's bus segment.
• The transaction tunneling mechanism allows a system designer to accommodate, for example, multiple plug-in cards or devices that include I[0044]²C slaves at the same addresses. By placing each of these plug-in devices behind a bridge, any master anywhere in the system can communicate specifically with any of these identically addressed slave devices. Furthermore, if bidirectional bridges are used, then each of the plug-in devices can also communicate through the bridge with any devices elsewhere in the system. When an I²C master wishes to communicate with a device behind a bridge via tunneling, it simply formats a message directly to the bridge indicating in the body of the message the address of the device with which it wants to communicate.
• In addition, each bridge contains a number of internal registers located in the[0045]RAM238 that are used by a configuration host to configure all the bridges in the bus hierarchy after coming out of a reset condition. These registers may also be used to enable and disable certain features of a bridge, such as whether address filtering is in use.
• The set of the internal registers is shown in FIGS. 3A-3I. The registers are accessed via special I[0046]²C transactions that are described below. The REVISION register, shown below in FIG. 3A, provides an indication of the revision of the bridge chip itself to software running on the configuration host. This information may be used to take advantage of features or changes on a revision-by-revision basis.Bits7 to4 indicate the major release level, andbits3 to0 indicate the minor release level. For example, a revision of 1.7 would be indicated by a hexadecimal value of 17h.
• The RESET register, shown in FIG. 3B, can be used to initiate a software reset of the bridge circuit. Writing to this register with[0047]bit7 equal to “1” will cause the bridge to undergo a full reset, equivalent to the assertion of the RESET input pin on the chip itself.
• The STATUS register, shown in FIG. 3C, provides the configuration host with visibility into the status of several of the pins on the bridge chip. This register is read-only-writes have no effect.[0048]Bits7 to4 are unused and always read as zero.
• The FLTR_CTL register, shown in FIG. 3D, allows the configuration host to select between three distinct modes of operation with regard to address filtering, controlled by the FILT_EN and TRANS_EN bits. The FILT_EN and TRANS_EN bits are cleared by a hardware or software reset. The three modes of operation are given below in Table 1.[0049]

  TABLE 1
  TRANS_EN	FILT_EN	Description
  0	0	Disabled. No addresses forwarded to Port-B
  		bus.
  0	1	Normal. Addresses filtered based on address
  		bitmap.
  1	X	Fully transparent. All addresses passed to
  		port-B, except the bridge address 62h.

• The description of each mode is given in the last column of the table. The transparent mode is equivalent to setting all the bits in the address bitmap and then enabling address filtering. This mode of operation is provided to simplify bridge configuration when the bridge is used being used in a “firewall” application.[0050]
• The DA_CTL register, shown in FIG. 3E, allows the configuration host to control whether a given bridge implements a deterministic arbitration (DA) protocol, as described in U.S. Pat. No. 6,175,887 and discussed below. This register consists of a single bit, which when written to “1”, enables the DA protocol, utilizing the arbitration ID given in the DA_ID register. The DA_EN bit is cleared by a hardware or software reset.[0051]
• The DA_ID register, shown in FIG. 3F, is used by the configuration host to program the ID driven by this bridge on its port-B bus when it is configured to use the DA protocol. The contents of this register are initialized to 20h at reset.[0052]Only bits7 to1 are programmable;bit0 is forced to a “0” to ensure that the ID corresponds to a write address. Software on the configuration host should program this register with the desired arbitration ID.
• The BR_ID register, shown in FIG. 3G, is the location where the configuration host programs the bridge ID of the bridge containing this register. The values 00h and 01h are reserved; the bridge will ignore attempts to program these values into this register. Following a hardware or software reset, this register is reset to 00h for a downstream bridge (DNSTR pin HIGH), and to 01h for an upstream bridge (DNSTR pin LOW).[0053]
• The aforementioned terms “upstream” and “downstream” refer to the location of the configuration host processor relative to the port-A bus and the port-B bus. The upstream bus is defined to be the bus with the fewest bridges in the path to the configuration host. The other bus is the downstream bus. This distinction is more apparent in larger tree configurations where there may be several levels of bus hierarchy between the configuration host and the most distant I[0054]²C devices.
• The BR_ID register is used in conjunction with the CFG_IN# and CFG_OUT# pins ([0055]232 and230, FIG. 2A) in order to allow a configuration host to configure multiple peer bridges connected to the same bus with their port-A interfaces. Peer bridges on a given bus segment are connected in a “daisy chain” with the CGF_OUT# pin of a bridge connected to the CFG_IN# of another bridge. Unconfigured downstream bridges come out of RESET with their bridge IDs set to 0. The CFG_IN#/CFG_OUT# daisy chain allows unconfigured peer bridges to determine which one of them will respond to configuration commands from the configuration host. The CFG_IN# pin is strapped LOW on the first bridge in the chain to indicate that it may respond to bridge commands addressed tobridge #0. After each bridge in the daisy chain is given a valid ID by the configuration host, it passes the LOW value to the next bridge in the chain.
• In particular, when the BR_ID register contains 00h, the bridge containing the register can respond to bridge commands only if its CFG_IN#[0056]daisy chain input232 is LOW. When the configuration host programs this register to a valid value (not 00h or 01h) theCFG_OUT# pin230 will automatically go LOW to pass the daisy chain signal to the next downstream bridge on this bus segment. This allows the next unconfigured bridge in the chain to respond to bridge commands from the configuration host. The UNCONFIG bridge command will return downstream bridges to an ID of 00h and upstream bridges to 01h, and will cause the CFG_OUT# pin to go HIGH
• The NXT_BRID register, shown in FIG. 3H, is the location where the configuration host programs the start of the range of bridge IDs that exist downstream of this bridge. The term “downstream” means in the direction of the port-B bus for a downstream bridge (DNSTR pin HIGH), and it means in the direction of the port-A bus for an upstream bridge (DNSTR pin LOW). This register is used together with the LST_BRID register to form a range of bridge IDs.[0057]
• When a bridge command is received by a bridge on its port-A bus, it first checks whether the specified bridge ID matches its bridge ID as programmed in the BR_ID register. If not, then downstream bridges check to see whether the specified ID falls within the range defined by the value in the NXT_BRID register to the value (inclusive) in the LST_BRID register. If the bridge ID is within this range, the bridge forwards the transaction to its port-B bus. Similarly, upstream bridges check to see whether the specified bridge ID falls outside the range defined by the NXT_BRID and LST_BRID registers. If so, then the transaction is similarly forwarded to its port-B bus. Following a hardware or software reset, this register is set to 00h.[0058]
• The LST_BRID register, shown in FIG. 3I, is the location where the configuration host programs the end of the range of bridge IDs that exist downstream of this bridge. Note that the term “downstream” means in the direction of the port-B bus for a downstream bridge (DNSTR pin HIGH), and it means in the direction of the port-A bus for an upstream bridge (DNSTR pin LOW). This register is used together with the NXT_BRID register to form a range of bridge IDs as discussed above. Following a hardware or software reset, this register is set to 00h.[0059]
• Returning to FIG. 2A, the[0060]DNSTR pin14 is pulled HIGH internally and, when it is left “floating”, it indicates to thebridge201 that the bridge is a “downstream bridge, i.e. its port-B interface214 faces away from the aforementioned configuration host. As described below, the DNSTR signal is used by the bridge in conjunction with the backoff (BOFF) signal on lead234 (generated at pin2) to prevent a deadlock condition that can occur when two bridges are wired together to form a bidirectional bridge.
• In particular, as illustrated in FIG. 4, two[0061]unidirectional bridges400 and402 may be wired in parallel, transmitting transactions in opposite directions, to implement a transparent, bidirectional bridge. Bi-directional bridges can be used to allow a large multimaster I²C bus to be divided into two or more different domains, where each domain contains one or more masters and the slave devices that the masters access most frequently. By grouping related master and slave devices together in this way, multiple transactions can be in progress simultaneously between a master and its local slave devices. Thus, the aggregate throughput of the I²C system is maximized. Also, by programming the bridges to block transmission of certain addresses, a host controller can configure an I²C bus to simultaneously use the same address on different multiple bus segments, with the replicated addresses available only to bus masters that are local to that bus segment.
• As shown in FIG. 4,[0062]bridge400 transmits I²C transactions in the direction ofarrow404 and bridge402 transmits I²C transactions in the direction ofarrow406.Bridges400 and402 connect two I²C buses. BUS0 represented by lines BUS0_SCL408 andBUS0_SDA410 is designated as the “upstream” bus and BUS1, represented by lines BUS1_SCL412 andBUS1_SDA414, is designated the “downstream” bus. BUS0 is connected to port-A of thebridge400 and to port-B of thebridge402, vialines418 and422. Similarly, BUS1 is connected to port-B of thebridge400 and to port-A of thebridge402, vialines426 and428.
• FIG. 4 shows the wiring of the DNSTR and BOFF signals for two bridges used in a bidirectional configuration. These signals work together to prevent a “deadlock” situation that would occur if a master on BUS[0063]0 attempted to access a slave on BUS1, while, at the same time, a master on BUS1 attempted to access a slave on BUS0. Because each bridge chip will stall its port-A bus while attempting to acquire its port-B bus, there arises a situation where both buses can be stalled indefinitely. In order to give priority to the configuration host in such deadlock situations, the inventive bridges are arranged to always defer access to the downstream bridge.
• For example, in the aforementioned deadlock situation, where transactions start simultaneously from masters on both sides of the bridge to slaves on opposite sides of the bridge, the bridge configured as the downstream bridge (with its DNSTR pin unconnected and floating HIGH) is constructed to drive the BOFF signal as an output to indicate that it is attempting to acquire the bus connected to its port-B interface. The upstream bridge (with its DNSTR pin tied LOW) is constructed to read the BOFF signal as an input. Thus, if the upstream bridge sees BOFF asserted HIGH while waiting to acquire the upstream bus (BUS[0064]0), it will abort the transaction by passing a NAK response back to the master on its port-A interface (BUS1).
• With this operation, it is necessary for the master on BUS[0065]1 to terminate its transaction with a STOP condition so that the downstream bridge can complete the transaction. Since masters attempting transactions through upstream facing bridges may receive a NAK as a response to an address byte, they must be designed to gracefully terminate their transaction with a STOP and then retry the upstream transaction. It should be noted that there is no fairness associated with this backoff mechanism; upstream masters sending transactions downstream will always win over the downstream masters sending transactions upstream.
• Any I[0066]²C system containing bridges constructed in accordance with the principles of the invention must have one or more bus masters defined as a “configuration host.” In general, there will be only a single configuration host, although there may be multiple configuration hosts so long as they exist on the same bus segment. This arrangement is required because the upstream/downstream strapping of all the bridges in the system is relative to the configuration host. Thus, the configuration host must exist upstream of all bridges in the system in order for some of the configuration operations to work correctly.
• For example, there is a configuration command called UNCONFIG, discussed in detail below, which is issued to all bridges simultaneously, via a broadcast addressing mechanism. Since these broadcast messages are only forwarded by downstream bridges, if there are any bridges upstream from the configuration host, they will not receive such broadcast messages.[0067]
• The mechanism for all communication between an I[0068]²C master, such as the configuration host, and the inventive bridges are bridge “commands” such as those described below. In normal operation, a configuration host would use these bridge commands to discover the bus topology and to configure the address bitmaps of all bridges in the topology. Following this configuration, the configuration host can enable address filtering in all bridges (for which filtering is desired) at which point the bridges will transparently forward I²C transactions up and down through the bus hierarchy, providing transparent access to all unique slave addresses to all masters in the system. For slave addresses that are not unique, remote masters in the system may use a procedure described below and called “transaction tunneling” to request proxy access to the addressed device via the bridge connected to that device's bus on its port-B interface.
• The bridge commands are specially formatted messages. The basic structure of a[0069]message500 constructed in accordance with one embodiment is shown in FIG. 5. Although a particular message arrangement is discussed below, it would be obvious to those skilled in the art that other arrangements could also be used in order to accomplish the same results. FIG. 5 illustrates the bits of a message, as they would appear on the SDA line in time with time increasing to the right. In this transaction diagram and the following transaction diagrams, the symbol “S” indicates the I²C START condition, “Sr” indicates a repeated START, “P” indicates a STOP condition, “A” indicates a LOW level on a bus during an acknowledge bit, and “N” indicates a HIGH level during an acknowledge bit.
• Like all I[0070]²C transactions, these messages start with an I²C address502. As previously mentioned, the address used by all inventive bridges is either 62h or 63h depending on whether the transaction is a write or read. In these addresses theLOW bit504 is a ‘0’ for writes and a ‘1’ for reads, in a manner similar to other conventional I²C devices. Bridges are uniquely accessed with a secondary-addressing scheme called a “bridge ID.” Thebridge ID506 appears as the first data byte in the I²C message500 following theaddress502. Following a hardware or software reset, all bridges are internally programmed to assume a default bridge ID of 00h for downstream bridges, and 01h for upstream bridges.
• As is common among I[0071]²C devices that have internal registers, read operations are performed via an atomic write-read I²C operation. The write portion of the write-read transaction accomplishes several things. First, it provides the bridge ID byte to identify which bridge in the system is being targeted. Second, the write portion of the write/read transaction provides thebridge command byte406, which tells the bridge what type of command the host is transmitting. Since the intended target bridge of the transaction could be on any bus segment in the system, the transaction will be forwarded throughout the bus hierarchy as required by all the bridges between the host and the target bridge.
• Following the write portion of the write-read, the master device driving the transaction issues a repeated START followed by the address 63h. At this point, all bridges in the system are programmed to remember what type of operation was in progress, and to which bridge the operation was targeted. Thus, all bridges in the bus hierarchy between the master and the targeted bridge continue forwarding the transaction, just as during the write portion of the transaction. Since the logical path from the master to the target bridge is given explicitly in the write portion, and implied in the read portion, it is imperative that a master device issue a STOP condition before attempting to communicate with a different slave device. This STOP condition lets all the bridges in the system know that the operation is complete, and that they can return to normal operation.[0072]
• In addition to specifying a unique bridge ID in the command syntax, the master responsible for configuring all the bridges, i.e. the configuration host, is allowed to issue a special bridge ID, called the “broadcast ID.” In an illustrative embodiment, the broadcast ID is given the value FFh, and will cause the bridge command specified in the transaction to be forwarded by all downstream bridges in the system. This addressing scheme is reserved for a single command type called the UNCONFIG command, which the configuration host can use to return all bridges to their default, unconfigured state.[0073]
• In this state, all downstream bridges assume an ID of 00h, all upstream bridges assume an ID of 01h, and address filtering and support for other operations are disabled. The configuration host may then rediscover the bus topology and reconfigure all bridges in the system.[0074]
• In order to allow a master to address a transaction to any bridge in the bus hierarchy without regard to the actual bus hierarchy configuration, each bridge knows explicitly the bridge lDs of all the bridges downstream from itself. Therefore, it can determine for each bridge command presented to its port-A interface whether the destination bridge is located upstream or downstream from itself. The mechanism that allows a bridge to determine whether it needs to forward a bridge transaction from its port-A interface to its port-B interface is based on the aforementioned register pair NXT_BRID and LST_BRID which define the range of bridge IDs that are located downstream from this bridge. The use of two registers to define a range of bridge IDs that need to be forwarded implies a certain relationship between the bus topology and the scheme used to assign IDs to bridges. This bridge ID assignment scheme is discussed below.[0075]
• A master can issue several command types to an inventive bridge. The first of these commands is a register access command and is issued by a master to read or write the internal registers (discussed above in connection with FIGS. 3A-3I) of a bridge. The syntax for the read and write transactions is shown in FIGS. 6 and 7 and is very similar to that used for other standard I[0076]²C components, such as a serial EEPROM. The essential difference is that the bridge transaction requires a bridge ID and a bridge command byte as the first two data bytes in the transaction.
• The register access write command is shown in FIG. 6 and starts with the bridge address of 62h ([0077]602), followed by thebridge ID604. Next, acommand byte606 of 00h is transmitted to indicate a register access command, and then theregister index608 is transmitted, followed by the byte or bytes610-614 to be written. The writes to te internal register array are controlled by an index pointer that selects the register that is being written. The index pointer automatically increments at the end of every byte written, so that successive bytes can be written to the register array in a single transaction. This is particularly helpful when writing the address filter bitmap.
• Although it is not obvious from register access transactions in FIGS. 6 and 7, when the registers of a bridge that is on a bus segment behind another bridge are accessed, the bridge transaction forwarding mechanism becomes involved. One of the artifacts of this mechanism is that the ACK bits returned to the master during the transaction actually come from different devices at different points in the transaction.[0078]
• For example, when the master initiates a transaction and issues the initial bridge address of 62h, ACK bits are returned by all bridges on the same bus segment as the master. The same is true for the bridge ID byte that follows. When the command byte then is sent, all bridges except the one that is in the path to the targeted bridge abort the transaction, leaving only the master and a single bridge to continue the transaction. If the transaction were to a bridge on the same segment as the master, the transaction completes locally.[0079]
• However, in the case of a multi-level bus topology where the target bridge is on a separate bus segment located behind a bridge on the master's bus segment, the transaction will be forwarded to the next bus along the path to the target bridge. The decision whether to forward the transaction is made by each bridge, by comparing the specified bridge ID to the range of bridge IDs defined by the NXT_BRID and LST_BRID registers. If the specified ID falls within this range, then the transaction will be forwarded by that bridge to its port-B bus. After the forwarding bridge has repeated the I[0080]²C address byte 62h, the bridge ID, and the command byte, it enters transparent mode of operation in which is simply passes bits back and forth between its port-A and port-B interfaces.
• This forwarding action takes place at each level of bus hierarchy between the initiating master and the target bridge. When the target bridge has been reached, it will return an ACK signal in response to the command byte, which is passed back up the hierarchy to the initiating master. Thus, the ACK signal returning from the command byte is the first ACK in the transaction that has actually come from the bridge intended as the target of the transaction.[0081]
• The register access read command is shown in FIG. 7 and starts with the bridge address of 62h ([0082]702), followed by thebridge ID704. Next, acommand byte706 of 00h is transmitted to indicate a register access command, and then theregister index708 is transmitted. At this point during a read transaction, a repeatedSTART condition710 is transmitted to reverse the data direction. Following this repeatedSTART710, the master issues the bridge readaddress712 of 63h. Note that no bridge ID or command byte is issued at this point in the transaction. Rather, all the bridges in the path to the target bridge retain enough internal state such that they can continue forwarding the transaction to the bridge specified in the write portion of the transaction.
• Thus, the repeated[0083]START710 and theread address712 of 63h are forwarded along the path to the target bridge followed by the read data bytes714-716. The target bridge is fully unaware that the read transaction has come from a distant bus segment connected through possibly several layers of bridges, and simply returns the read data.
• The bridges in the active path through the hierarchy pass the read bits from the target bridge back to the initiating master, and also the ACK bits from the master back to the target bridge.[0084]
• When the initiating master is finished reading bytes, it returns a NAK signal in response to the last byte as per standard I[0085]²C protocol and issues a STOP.
• The STOP is similarly forwarded down the path through the bus hierarchy to the target bridge, at which point the transaction has ended.[0086]
• The 32 bytes in the address bitmap comprise 256 bits, which allows one bit for every possible read/write address in an eight-bit I[0087]²C address space. At the beginning of an I²C transaction, a bridge absorbs the entire8-bit address (7-bit address plus the read/write bit) from its port-A interface and checks the corresponding bit in the address bitmap. If the bit is a ‘1’, then the bridge will acquire the bus connected to its port-B interface and re-drive the transaction.
• When the bridge receives an ACK or NAK bit from the port-B interface, it passes the bit back to the port-A interface and the proceeds with the rest of the transaction in a transparent manner, passing data and ACK/NAK bits back and forth between the two interfaces. When the port-A master terminates the transaction with a STOP condition, the bridge will, in turn, terminate the port-B transaction with a STOP.[0088]
• Another set of bridge commands that can be issued by a master are “tunnel” read/write commands. Transaction tunneling is a mechanism provided by the inventive bridge to complement the address filtering mechanism. As previously described, address filtering allows the 128 possible read/write addresses of the I[0089]²C bus to be distributed arbitrarily among the bus segments of an I²C bus topology. The address bitmap of each bridge allows transactions to be forwarded as required up and down through the bus hierarchy to establish a transparent communication path between a master and the addressed slave. In some cases, however, it may be necessary for some number of slave devices to be assigned the same I²C address. In this case, by placing such slave devices on separate bus segments and using the inventive bridge to connect these segments into an I²C topology, all slave devices can be uniquely addressed. The tunnel command allows a bus master to bypass the internal address bitmap in a bridge and directly specify which address will be driven by the bridge.
• The syntax for a tunnel read command is shown in FIG. 8 and starts with the bridge address of 62h ([0090]802), followed by thebridge ID804. Next, acommand byte806 of 01h is transmitted to indicate a tunnel command, and then thetarget address808 is transmitted. At this point during a read transaction, a repeatedSTART condition810 is transmitted to reverse the data direction. Following this repeatedSTART810, the master issues the bridge readaddress812 of 63h. Again, no bridge ID or command byte is issued at this point in the transaction. Rather, all the bridges in the path to the target bridge retain enough internal state such that they can continue forwarding the transaction to the bridge specified in the write portion of the transaction. Thus, the repeatedSTART810 and theread address 812 of 63h are forwarded along the path to the target bridge along with the read data bytes814-816.
• The tunnel write command is similar, but does not include the repeated[0091]START810. Instead, the write data bytes are driven by the master immediately following the target address byte in a manner consistent with the slave being accessed.
• In a manner similar to that described above for the register access command, a tunneled transaction is forwarded by the bridges in the system until the transaction reaches the target bridge as specified in the bridge ID byte. This bridge then recognizes the[0092]tunnel command byte806 and interprets the followingtarget address byte808 as the slave address to be driven on its port-B bus. Once the target bridge has driven a START and the specified slave address on its port-B bus, it begins transparently forwarding data and ACK bits between its port-A and port-B interfaces.
• Similarly, any other bridges in the path between the initiating master and the target slave device operate transparently at this point. As in the case of a bridge register access, all the bridges in the active path through the hierarchy maintain this transparent mode of operation even through the repeated[0093]START810 shown in FIG. 8. All bridges forward the initiating master's STOP condition at the end of the transaction, thus returning all the active bus segments to the idle state.
• A final command that can be issued by a master is the UNCONFIG command. This command is a special type of command issued by the configuration host to return all bridges to their default unconfigured state, i.e. with their bridge IDs set to 00h, and with address filtering and deterministic arbitration support disabled. The UNCONFIG command could be used, for example, after a new hot-plug device has been added into the system. If the new device includes any bridge chips, then all the bridges in the system will possibly need to be renumbered in order to preserve the correctness of the bridge command forwarding mechanism. The addition of new I[0094]²C slave addresses to the I²C hierarchy can generally be accommodated without the need to unconfigure/reconfigure the bridges; a simple modification to the address filter bitmaps of the bridges in the hierarchy is generally sufficient.
• The inventive bridge supports multimaster I[0095]²C operation on both its port-A and port-B interfaces. Since the bridge acts only as a slave on its port-A interface, there is nothing specific that it needs to do in order to support multimaster traffic on this interface. However, the port-B interface, acts as the master interface. Because of this, the bridge must handle all of the normal issues associated with multimaster I²C bus activity.
• Perhaps the most basic requirement for a multimaster capable I[0096]²C master is the ability to perform bus arbitration. According to the aforementioned I²C Specification from Philips Semiconductor, bus arbitration is based on collisions between competing master or slave devices. Any device driving the bus with a pattern of address, data, or ACK/NAK bits watches the bus while it transmits, checking that the data appearing on the bus is what the device intended to drive. Because of the open-collector nature of the I²C bus, a device driving a “0” will always win out over a device driving a “1”. When a device detects a “0” on the bus when it intended to drive a “1”, it immediately aborts its involvement in the current transfer and releases the SCL and SDA lines.
• A master device that “loses arbitration” in this manner may retry its transaction when the current transaction is terminated via a STOP condition. However, one problem with the I[0097]²C arbitration mechanism as designed by Philips Semiconductor is that, if multiple masters collide while driving the same address, such as if two masters attempt to write to the same device at the same time, then no collision will be detected, and both masters will proceed together on the bus, unaware of each other's presence.
• It is likely in this scenario that the masters will drive different data later in the write transaction, resulting in a loss of arbitration very late in a transaction. Handling this type of late loss of arbitration may be complex depending on the application, since some number of data bytes may have already been committed to the device before the error was detected.[0098]
• Another problem that may arise when two masters are unknowingly on the bus at the same time is that the I[0099]²C arbitration mechanism does not work during START or STOP conditions. Thus, if one of the colliding masters issues a STOP or repeated START, the other master will need to handle this asynchronous error event in the midst of its write transaction.
• In order to prevent these types of complex error scenarios, an extension to the I[0100]²C Specification was developed by Sun Microsystems called the “Deterministic Arbitration Protocol”, more commonly referred to as the DA protocol and described in the aforementioned U.S. Pat. No. 6,175,887. The concept of the DA protocol is straightforward. Following a bus-idle condition,1²C masters implementing the DA protocol always drive as their first address a master-specific bit pattern called an “arbitration ID.” By virtue of the fact that every master has a unique arbitration ID, this ensures that any masters colliding on the bus will detect the collision during this initial address phase. Following the transmission of the arbitration ID, the winning master proceeds with its intended I²C transaction via a repeated START. Thus, all the complexities associated with late collisions and asynchronous START or STOP conditions due to collisions are avoided.
• In accordance with the principles of the invention, the bridge provides some special facilities for handling the DA protocol. For example, consider a configuration including an I[0101]²C bus with a master, a few slave devices, and an inventive bridge behind which are more slave devices.
• Because the arbitration ID used by the master is used on every transaction, this initial address phase does not provide an indication of whether the intended transaction needs to be forwarded by the bridge. In fact, the bridge must be configured such that the bit-pattern used as the arbitration ID corresponds to an address that is not forwarded by the bridge. Otherwise, the bridge would claim and forward every transaction from the master, preventing the master from reliably communicating with the slave devices on the same bus segment with the master.[0102]
• Rather, by programming the bridge's bitmap with a ‘0’ in the bit position corresponding to the arbitration ID used by the master, the bridge will ignore the address issued as the DA protocol portion of the transaction. Following the transmission of the DA arbitration ID, the master will issue a repeated START and then the slave address of the intended target of the transaction.[0103]
• The bridge will again look up this new address in its internal bitmap, and, if the corresponding bit is a ‘1’, the remainder of the transaction will be forwarded to the port-B bus. Since by necessity the bridge ignored the DA portion of the transaction, the port-A masters DA arbitration ID does not pass to the port-B bus. Therefore, a mechanism has been provided in the bridge to cause it to use the DA protocol with a programmable DA arbitration ID when it forwards transactions to the port-B I[0104]²C bus.
• This is accomplished through two registers, called DA_EN and DA_ID. These registers are discussed above in connection with FIGS. 3E and 3F but basically the DA_ID register, shown in FIG. 3F, contains the arbitration ID to be driven by the bridge on the port-B bus when forwarding transactions, and the DA_EN register, shown in FIG. 3E, is used to enable or disable the DA protocol. The default state following a reset is for bridges to not use the DA protocol.[0105]
• In the process of forwarding transactions from the port-A interface to the port-B interface, collisions may occur if other masters exist on the port-B bus. If a bridge is configured to use the aforementioned DA protocol on the port-B bus, then these collisions should happen only in the first address phase of a new transaction, but nonetheless it is possible that a bridge will lose the arbitration process and be forced to relinquish the port-B bus. In the case where the arbitration loss occurs during the initial address, a bridge will automatically attempt retransmission of the forwarded transaction, up to a limit of ten (10) attempts per transaction. This automatic retry mechanism works only if the loss of arbitration occurs during the initial address phase. If the bridge loses arbitration ten times in a row while attempting to transmit the initial address phase on a given transaction, it will finally give up and return a NAK condition to the port-A master device.[0106]
• Because the port-B master port on an inventive bridge can be connected to a multimaster bus, there are a number of error scenarios that may present themselves on the port-B interface. These are briefly:[0107]
• 1.) Transaction retry limit of ten retries exhausted.[0108]
• 2.) Late loss of arbitration.[0109]
• 3.) Asynchronous START or STOP detected.[0110]
• 4.) Failure to acquire port-B bus within 200 milliseconds.[0111]
• 5.) Slave on Port-B bus stalls a transaction with SCL low for more than 200 milliseconds.[0112]
• 6.) Protocol hang on port-B bus (Bus remains “busy” but with no activity.)[0113]
• 7.) Port-B slave device holding SDA low due to loss of bit synchronization.[0114]
• Because the I[0115]²C bus does not provide any error reporting mechanisms aside from the NAK response, most of the errors listed above are signaled to the master connected to the port-A interface via NAK responses. The exceptions are items (6) and (7). The error scenarios described in items (6) and (7) above are recoverable situations in most cases. These recoverable port-B errors can be caused by a malfunctioning master or slave device on port-B, or if a master device, such as an inventive bridge, driving a transaction on the bus, is reset by the system in the middle of its transaction.
• When one of the following situations occurs, the bus can get into a state where master and slave devices are out of synchronization with regard to whether the bus is idle or busy, and also with regard to bit and byte synchronization.[0116]
• In a first situation, a bridge believes that the bus is currently busy, but there is actually no master on the bus currently driving a transaction. This can happen if a master (other than an inventive bridge) issues a START condition, but then aborts its transaction without issuing a STOP condition. In this case, all I[0117]²C devices will consider the bus busy, and since the I²C protocol has no timeouts defined with regard to this type of hang, it can persist indefinitely.
• In another situation, a master device aborts a transaction suddenly in the middle of reading a data byte from a slave, and if the slave happens to be driving a “0” on the bus, then the slave will continue to drive a “0” on the SDA line, waiting for the master to finish the transaction. Since the master has aborted the transaction, the bus is hung both from a protocol perspective as mentioned in the previous situation, but also because the slave is holding the SDA line LOW, which prevents masters from issuing new START or STOP conditions.[0118]
• In order to recover from these two situations, when an inventive bridge is attempting to acquire the port-B bus, if the bus remains busy for 200 milliseconds, it will assume that the bus has become hung, i.e. that a master aborted a transaction, leaving the bus in the “busy” state. The bridges will then look at the SDA line to see whether any slave device appears to be holding the SDA line LOW. If so, it will toggle the clock line up and down in an attempt to cause the slave to shift out bits on the assumption that when the slave reaches what it believes to be the ACK bit, it will release SDA.[0119]
• When the bridge senses the SDA line become HIGH, it will then issue a START/STOP on the SDA/SCL lines to bring all I[0120]²C devices back into byte synchronization and protocol synchronization.
• Note that, in the I[0121]²C bus definition, there is no recovery from the situation when a device hangs holding the SCL line LOW. If this happens to the port-B bus because of an errant device on that bus, then all transactions from a port-A master targeted to slave devices on the port-B bus will receive a NAK response from the bridge after a 200 millisecond timeout.
• In addition to the error handling capabilities on the port-B interface, the port-A interface also implements a watchdog timer. This timer watches activity on the port-A interface and will cause the bridge to reset itself in the event that the transaction on the port-A interface stalls and fails to make forward progress for more than one second. In this way, if there is some type of electrical problem on the port-A interface that looks like a START condition, such as could occur when installing a hot-pluggable device, the bridge will reset itself after one second and will subsequently be ready for new activity. It should be noted that any master on a bus segment that also connects to the port-A interface of an inventive bridge must not stall its own transactions for more than one second, as this will cause the bridge to reset. In general, problems will begin to occur if master or slave devices stall transactions for more than 200 ms, since this will affect any bridges with port-B interfaces connected to that bus. The watchdog reset does not alter any software-visible state within a bridge.[0122]
• The inventive bridge is intended to allow the system designer a great deal of flexibility with regard to I[0123]²C bus topology and the distribution of I²C addresses within that overall topology. This flexibility is achieved through the programmability of various aspects of the bridge. The proper configuration of the bridges in an I²C system is key to the correct operation of the system. The basic configuration procedure is illustrated in FIGS. 9A and 9B and begins instep900
• In[0124]step902, the procedure discovers the bus topology and the bridges that form that topology. Next instep904, bridge IDs are assigned to all bridges. Instep906 appropriate information is entered into the NXT_BRID and LST_BRID registers (shown in FIGS. 3H and 3I) of every bridge to order the upstream and downstream bridges.
• In[0125]step908, a determination is made whether address filtering is to be used. If so, the address bitmaps of each bridge are initialized instep910 and the FILT_EN bit is set in the FILT_CTL register (FIG. 3D) in order to enable address filtering. Alternatively, if instep908, it is determined that a bridge is being used as a “firewall”, then the TRANS_EN bit is set in the FILT_CTL register (FIG. 3D) instep912. In either case, the procedure proceeds, via off-page connectors914 and916 to step918.
• [0126]Instep918, a determination is made whether the deterministic arbitration protocol discussed above will be used. If so, the process proceeds to step920 in which DA_ID values are assigned to each bridge such that no two port-B bridge interfaces share the same DA_ID on a given bus, and DA protocol is enabled by setting the DA_EN bit in the DA_CTL register (FIG. 3E.) The procedure then finishes instep922. Alternatively, if, instep918, a determination is made that the deterministic arbitration protocol will not be used, the procedure finishes instep922.
• An illustrative set of bus topology rules and a procedure for configuring a bus that adheres to those topology rules is described below. These configuration rules are exemplary only and are not intended to be all encompassing or to place restrictions on what topologies are allowed. Instead, these topology rules and the associated configuration procedure are an illustration of how a configuration host, with no foreknowledge of the bus topology, could correctly discover and initialize a complex bus structure built using bridges constructed in accordance with the principles of the invention as discussed above.[0127]
• In accordance with one illustrative embodiment, bus topology rules allow a generic bus configuration procedure to discover the topology of the bus and correctly initialize the bridges that comprise that topology. This bus configuration procedure is performed by software running on a configuration host, which is an I[0128]²C master device located on the bus being configured.
• The preferred bus topology is a tree, with the configuration host located at the root bus of the tree. This type of topology allows the configuration host to probe down through the tree, discovering and numbering bridges as it goes. As previously mentioned, the NXT_BRID and LST_BRID registers are used by each bridge to define a range of bridge IDs that exist downstream of certain bridge. For this reason, it is necessary that all bridges downstream from a given bridge be given IDs that form a range. This allows upstream and downstream bridges to easily check whether a bridge transaction received on its port-A interface needs to be forwarded; downstream bridges forward bridge transactions addressed to IDs within the defined range, while upstream bridges forward transactions addressed to bridges outside the defined range. In this way, both the configuration host and masters at lower layers of the hierarchy are able to send bridge transactions to bridges throughout the system.[0129]
• An example bus topology is shown below in FIG. 10. There are a total of four[0130]bus segments1006,1008,1032 and1036 in this system. In addition, there are fivebridges1004,1014,1016,1020 and1022. Theconfiguration host1000 is directly connected to aunidirectional bridge1004 that faces downstream, passing transactions tobus segment1008, and acting as a “firewall” bridge. In this example system, the implementation of this hypothetical configuration host lacks a sophisticated I²C port. Instead, it has only a very simple I²C interface consisting of two General Purpose I/O (GPIO) ports which are software stimulated (bit-banged) to function as an I²C master interface onbus segment1006. Due to the bit-bang nature of this interface,host1000 lacks multimaster support.
• However,[0131]configuration host1000 needs to drive transactions onmultimaster bus segment1008 and must interact withseveral bridges1014,1016,1020 and1022, and anothermaster device1002 ataddress 30h that are also connected tobus1008.Bridge1004 is used as a firewall device to allow the non-multimastercapable host1000 to access themultimaster bus1008. Although not shown in this diagram,bridge1004 must have its CFG_IN# pin strapped LOW to allow configuration of this bridge.
• Bridge pairs[0132]1014/1016 and1020/1022 are connected in parallel as described above and used to provide a bidirectional connection to two pluggable interfaces for attaching plug-in boards. In the illustrative example, two such plug-inboards1024 and1034 are shown connected at the bottom of the bus hierarchy.
• In this example, the I[0133]²C addresses used by the devices1026-1030 and1038-1042 on the two plug-in boards,1024 and1034, respectively, are all unique. A unique address selection might be accomplished, for example, by one or two address select signals (not shown) that are present at the connector interface to the plug-inboards1024 and1034. These signals can be used to automatically select the device addresses for the I²C components on these boards. Thus, configuration of the bridges in this example system will consist of configuring the address bitmaps of all the bridges to allow transparent access to all slave devices by all masters.
• In an alternate system implementation, for example, an l/O expansion chassis for InfiniBand[0134]^SM target channel adapter (TCA) cards, the plug-in boards would all use the same I²C addresses. Such a system would require the use of the transaction tunneling procedure described above in order to provide access to all slaves by all masters.
• Each of the master and[0135]slave devices1002,1010,1012,1026,1028,1030,1038,1040 and1042, in FIG. 10 has its I²C slave address that is shown in the figure. The configuration host, being a master-only device, does not have a slave address. The numbers shown within eachbridge1004,1014,1016,1020 and1022 represent the bridge ID assumed by that bridge following a system reset. For example,bridge1004 would assume a bridge ID of 00h following a system reset.Bridge1014 would assume an ID of 01h, etc. Effectively, all downstream bridges assume a default ID of 00h, while all upstream bridges assume a default ID of 01h.
• As can be seen in FIG. 10,[0136]bridges1016 and1020 have asignal1018 called CFG_OUT# connecting them. This signal represents the CFG_OUT# output frombridge1020, which connects to the CFG_IN# input ofbridge1016. This signal is required during bridge configuration to control which of thebridges1016 and1020 connected tobus segment1008 can respond to a bridge transaction addressed to bridgeID0.Bridge1020 has its CFG_IN# pin strapped LOW, making it the default responder to bridge transactions onbus segment1008 addressed to bridge ID=0. Afterbridge1020 is assigned a valid ID (i.e. something other than 00h or 01h),bridge1020 will pass this daisy chain grant to bridge1016 via its CFG_OUT# pin.
• At reset time, none of the[0137]bridges1004,1014,1016,1020 and1022 will pass any transactions. During the configuration process, theconfiguration host1000 will keep a global population address bitmap (GPB) for the system. The GPB for a system is a 32-byte bitmap similar to the address bitmap contained within each bridge, and simply has a “1” bit indicating a populated slave address. This allows theconfiguration host1000 to keep track of populated slave addresses in the entire system as it configures the bus. As new slave addresses are discovered on each bus segment, the host will check these addresses against the GPB to check for replicated addresses.
• In addition to the GPB, the[0138]configuration host1000 will keep track of a local population address bitmap (LPB) for each bus segment. The collection of LPBs will be used to identify unused addresses on each segment that may be used as arbitration ID's for the DA protocol feature of the inventive bridge as discussed above.
• In order to number bridges in a manner consistent with the bridge transaction forwarding mechanism used by the inventive system, the configuration procedure follows a “depth-first” methodology. This means that, when a bridge is discovered at bus level “N”, all the bus hierarchy beneath that bridge is recursively probed and configured before the configuration of any peer bridges on bus level “N”. The steps in the configuration procedure of this example system are illustrated in FIGS. 11A, 11B and[0139]11C.
• The configuration procedure starts in[0140]step1100 and proceeds to step1102 wherehost1000 creates the GPB for the system. Instep1104, the host createsLPB#0 forbus segment1006, and, instep1106, the host probes the I²C address space looking for populated slaves onbus segment1006. Becausebus segment1006 is the root bus, probing is done via simple I²C read or write transactions, looking for an ACK bit returning from a slave after the I²C address phase. Instep1108, a determination is made whether a slave device has been found. In this case there are no slave devices and the procedure proceeds, via off-page connectors1118 and1122, to step1124 wherehost1000 probes for an upstream bridge on thisbus segment1006 by attempting to read the REVISION register of a bridge onbus segment1006 at bridge ID=1. Instep1126, a determination is made that no upstream bridges are found so that the procedure moves to step1130.
• In[0141]step1130host1000 probes for a downstream bridge onbus1006 by attempting to read the REVISION register of a bridge at ID=0. This transaction succeeds, indicating the presence ofbridge1004 instep1132. The procedure then moves to step1134 wherehost1000 providesbridge1004 with an ID of “2” by writing its BR_ID register with 02h. (ID assignment starts at “2” because IDs “0” and “1” are reserved IDs.)Bridge1004 then asserts its CFG_OUT# pin LOW, but with no effect sincebridge1004 has no peers.Bridge1004 is defined to be the “parent bridge” of the newly discoveredbus segment1008. Likewise,bus segment1008 is the child ofbridge1004. The use of a tree topology guarantees that a bus can have no more than one parent bridge. The NXT_BRID and LST_BRID registers forbridge1004 remain at 00h, indicating no downstream bridges discovered yet. The procedure then returns, via off-page connectors1120 and1116 to step1104 to configure any buses discovered behind the newly discovered bridge.
• In[0142]step1104host1000 createsLPB#1 forbus segment1008 and continues the bus configuration recursion. Instep1106,host1000 probesbus segment1008 looking for slave devices. Note that the address bitmap ofbridge1004 has not yet been filled in, so this probe is accomplished using the aforementioned transaction tunneling procedure. Specifically,host1000 sends tunnel commands to bridge1004 at bridge ID=2 to perform write or read transactions onbus segment1008. If a transaction completes normally, then a device exists at that address. This is how all probing behind bridges is accomplished.
• In the configuration illustrated in FIG. 10,[0143]host1000 will find slave devices at addresses A0h, A2h and 30h. When each device is found as determined instep1108, the procedure moves to step1110 where a determination is made by checking the addresses against addresses in the GPB for duplicate addresses. No duplicate addresses are present so that the procedure moves to step1112 where the slave address is added to the GPB and toLPB#1. Alternatively, if duplicate addresses had been detected instep1110,host1000 leaves the associated bit in the LPB for the current bus segment clear, and instead adds the I²C address to a list of addresses that will need to be accessed via tunneling as indicated instep1114. The parent bridge of the bus segment on which this address resides is also saved so that a tunnel transaction can be sent to the parent bridge to access this slave device.
• After the slave address has been entered into either the GBP and LPB or into the tunnel list, the procedure returns to step[0144]1106 to check for additional slave addresses. Steps1108-1114 are repeated until no further slave addresses are detected instep1108. Next the procedure proceeds, via off-page connectors1118 and1122 to step1124 wherehost1000 probes throughbridge1004 using a tunneled message, looking for a bridge at bridge ID=1. If present, this latter bridge would be the upstream bridge frombus segment1008 back tobus segment1006. There is no such bridge in the system shown in FIG. 10, so this latter bridge is not discovered as determined instep1126.
• The procedure then moves to step[0145]1130 wherehost1000 probes for a downstream bridge onbus1008 at bridge ID=0. Instep1132,bridge1020 is discovered.Bridge1020 is defined as parent bridge of newly discoveredbus segment1036. Instep1134,host1000 gives this downstream bridge a bridge ID of 03h, causingbridge1020 to assert itsCFG_OUT#1018 LOW, thereby enablingbridge1016 to respond to future bridge transactions addressed onbus1008 to bridge ID=0, which will happen later.
• The procedure returns to step[0146]1104, via off-page connectors1120 and1116, wherehost1000 createsLPB#2 forbus segment1036 and, instep1106,host1000 probesbus1036 looking for slave devices. Thehost1000 will find slaves at addresses A4h, 9Ch, and 32h and process them by repeating steps1108-1114.
• The procedure then moves, via off-[0147]page connectors1118 and1122 to step1124 wherehost1000 probes for upstream bridge onbus segment1036 at bridge ID=1 and findsbridgelo1022.Host1000 then assigns bridge1022 a bridge ID of4 instep1128 and proceeds to step1130. Instep1130host1000 probes for downstream bridges at bridge ID=0 onbus segment1036, but finds none. The software recursion has reached the bottom of this leaf of the bus hierarchy.
• Accordingly, from[0148]step1132 the procedure moves, via off-page connectors1136 and1140 to step1144 where a determination is made whether a parent bridge exists. If so, the procedure moves to step1148 where the address bitmap ofbus segment1036 parent bridge (in this illustration bridge1020) is filled in with the logical OR ofLPB#2 and the bitmaps of all downstream bridges connected tobus segment1036. In this case, there are no subordinate bridges, so thebridge1020 address bitmap is simply filled in with the entries inLPB#2. Address filtering can be enabled at this time forbridge1020 by setting the FILT_EN bit in its FILT_CTL register.
• Since there is no further unconfigured bus hierarchy below[0149]bus segment1036, the NXT_BRID and LST_BRID range registers for the parent bridge (bridge1020) ofsegment1036 are populated instep1150. Since theparent bridge1020 has an ID of3, and the next unused bridge ID to be assigned is 5, thehost1000 writes thebridge1020 NXT_BRID register and LST_BRID register both equal to 4, indicating that there is only a single bridge at bridge ID of4 downstream ofbridge1020. Thus,bridge1020 will forward bridge commands to bridge ID=4 tobus segment1036. Similarly, the same values are written to the range registers ofbridge1022. Sincebridge1022 is an upstream bridge, it does not forward bridge commands within this range, since they are downstream. Instep1152. The procedure returns up one bus level, back tobus segment1008 and the procedure returns via off-page connectors1142 and1138 to step1130.
• Instep[0150]1130,host1000 probes fora downstream bridge onbus segment1008 at bridge ID=0, and discoversbridge1016 instep1132.Instep1134,host1000 writes a bridge ID of5 to the BR_ID register ofbridge1016, causingbridge1016 to assert its CFG_OUT# pin LOW, but with no effect since there are no more unconfigured peer bridges onbus segment1008.Bridge1016 is parent bridge of newly discoveredbus segment1025. The procedure then proceeds, via off-page connectors1120 and1116 back tostep1104.
• In[0151]step1104,host1000 createsLPB#3 and, instep1106,host1000 probes slave addresses, finding slave devices at addresses A6h, 9Eh, and 34h. These addresses are processed by looping through steps1106-1114 as previously described.
• When no more slave devices are found in[0152]step1108, the process moves, via off-page connectors1118 and1122, to step1124 wherehost1000 probes for an upstream bridge onbus segment1025 at bridge ID=1 and findsbridge1014. Instep1128,host1000 assigns this bridge a bridge ID of6.
• Next, in[0153]step1130,host1000 probes for a downstream bridge at bridge ID=0 onbus seg ment1025, but finds none as determined instep1132. The software recursion has reached the bottom of this leaf of the bus hierarchy and theprocess 20 moves, via off-page connectors1136 and1140 to step1144.
• If a parent bridge exists as determined in[0154]step1144, the process proceeds to step1148. In this sample system, the parent bridge isbridge1016. Instep1148, the address bitmap of thisparent bridge1016 is filled in with the logical OR ofLPB#3 and the bitmaps of all subordinate downstream bridges connected to this bus segment. In this example, there are no downstream bridges onbus segment1025. Accordingly, the address bitmap ofbridge1016 is set to the contents ofLPB#3. Address filtering can be enabled at this time forbridge1016 by setting the FILT_EN bit of its FILT_CTL register.
• Since there is no further unconfigured bus hierarchy below this bus segment, the NXT_BRID and LST_BRID range registers for the[0155]parent bridge1016 are filled in. Sincebridge1016 has a bridge ID of5, and the next unused ID to be assigned is 7, thehost1000 writes a 6 into the NXT_BRID and LST_BRID registers ofbridge1016, indicating that there is only a single bridge at bridge ID of6 downstream ofbridge1016. Thus,bridge1016 will forward bridge commands addressed to bridge ID=6 onbus segment1025. Similarly, the same values are written to the range registers ofbridge1014. Sincebridge1014 is an upstream bridge, it does not forward bridge commands to bridges within this range, since they are downstream. Instep1152, the process returns up one level, back tobus segment1008 and then proceeds, via off-page connectors1142 and1138 to step1130.
• In[0156]step1130,host1000 probes for a downstream bridge at bridge ID=0 onbus segment1008 and finds none as determined instep1132. The illustrative process has configured all downstream bridges below this bus segment in the hierarchy. Therefore, the process proceeds, via off-page connectors1136 and1140, to step1144 where a determination is made whether a parent bridge exists for thisbus segment1008. In the example,parent bridge1004 exists forbus segment1008. The process then moves to step1148 where the address bitmap ofparent bridge1004 is filled in with the logical OR ofLPB#1 and the address bitmaps of all subordinate downstream bridges connected tobus segment1008. In this example, this calculation is (LPB#1) OR (bridge1020 address bitmap) OR (bridge1016 address bitmap). Address filtering can be enabled at this time forbridge1004 by setting the FILT_EN bit of its FILT_CTL register.
• Since there is no further unconfigured bus hierarchy below this[0157]bus segment1008, the process proceeds to step1150 where the NXT_BRID and LST_BRID range registers forparent bridge1004 are filled. Since theparent bridge1004 has an ID of2, and the next unused ID to be assigned is7, thehost1000 writes a3 into the NXT_BRID register and a6 into the LST_BRID register, indicating that there are4 bridges downstream frombridge1004, atbridge IDs3,4,5, and6. Thus,bridge1004 will forward bridge commands to bridgeIDs3 through6 tobus segment1008.
• Next the process moves up one level in the hierarchy in[0158]step1152 back tobus segment1006 and the process returns, via off-page connectors1142 and1138 to step1130.
• In[0159]step1130,host1000 probes for additional downstream bridges at bridge ID=0 onbus segment1006 and finds none as determined instep1132. The procedure has configured all downstream bridges below this bus segment in the hierarchy. The process then proceeds, via off-page connectors1136 and1140, to step1144 where a determination is made whether any parent bridge exists forbus segment1006. Sincebus1006 is the root bus segment, there is no parent bridge, so the downstream portion of the bridge configuration is complete and the process finishes instep1146.
• FIG. 12 below shows the state of the example system at this point. Items in FIG. 12 that correspond to items in FIG. 10 have been given corresponding numeral designations. For example,[0160]host1000 in FIG. 10 corresponds to host1200 in FIG. 12. In FIG. 12, all of thedownstream bridges1204,1216 and1220 have been assigned bridge IDs, and have had their address bitmaps initialized. Allupstream bridges1216 and1222 have been assigned IDs, but their address bitmaps remain uninitialized. Address filtering is operational for the downstream paths. Thus, theconfiguration host1200 and themaster1202 ataddress 30h onbus segment1208 can see all the devices in the system. However, themasters1226 and1238 on the two plug-inboards1224 and1234, cannot access any devices other than the ones (1228,1230 and1240,1242, respectively, on theirrespective bus segments1225 and1236.
• The configuration process is completed using the process shown in FIG. 13. This process uses the information in the GPB and LPBs created in the process illustrated in FIGS. 11A-11C to fill in the address bitmaps of the[0161]upstream bridges1214 and1222. Since any duplicate addresses have already been omitted from the LPBs for each bus segment, no checking for duplicates is required.
• Now that the entire bus hierarchy has been probed, it is a simple matter for the[0162]configuration host1200 to walk through the system, programming the address bitmaps of all theupstream bridges1214 and1222. This procedure starts instep1300 and proceeds to step1302 where theconfiguration host1200 uses the information regarding the bus topology to walk the bus hierarchy. The bus is walked in the same order in which it was initialized, going upstream bridge by upstream bridge.
• In step[0163]1304 a determination is made whether an upstream bridge has been found. If so, the process moves to step1306 where the address bitmap is initialized. In the example system in FIG. 12,bridge1222 is located first. The address bitmap forbridge1222 is initialized with the XOR of the GPB and the bitmap already programmed intobridge1220. This XOR function results in a bitmap that contains all the addresses in the system that are not onbus segment1208 or below. Address filtering is then enabled inbridge1222 by setting the FILT_EN bit in its FILT_CTL register.
• The process then returns to step[0164]1302 where thehost1200 continues walking the bus bridges in the system, this time stopping atbridge1214 as determined instep1304. Instep1306, the address bitmap forbridge1214 is initialized with the XOR of the GPB and the bitmap already programmed intobridge1216. This XOR function results in a bitmap that contains all the addresses in the system that are not onbus segment1225 or below. Address filtering is enabled inbridge1214 by setting the FILT_EN bit in its FILT_CTL register. The process then returns to step1302.
• In[0165]step1304,host1200 determines that no more upstream bridges remain. Bus configuration is complete and the process proceeds to step1308. Theconfiguration host1200 may now choose addresses to use on each bridge's port-B interface for use as an ID for Deterministic Arbitration, as discussed above. In general, it does not matter whether the address used as a DA arbitration ID corresponds to a populated slave address; the DA protocol will issue a null write (with no data), so the state of an addressed slave should remain unaltered. In the event that this is a concern, thehost1200 can ascertain for each bus segment the list of addresses already claimed for that segment, and then use addresses that are not otherwise in use on this segment or being forwarded to any other segment. This list is stored in that bus segment's parent bridge's address bitmap. In the event that any duplicate addresses were found, thehost1200 should also check the tunnel list. The host can then choose from the list of unclaimed addresses for each bus segment and assign them to the masters on that bus and to the parent bridge for that bus for use as a DA arbitration ID as set forth instep1308. The process then finishes instep1310.
• Although an exemplary embodiment of the invention has been disclosed, it will be apparent to those skilled in the art that various changes and modifications can be made which will achieve some of the advantages of the invention without departing from the spirit and scope of the invention. For example, it will be obvious to those reasonably skilled in the art that, in other implementations, a different microcontroller may be used. Other aspects, such as the specific state flows, as well as other modifications to the inventive concept are intended to be covered by the appended claims[0166]

Claims (20)

What is claimed is:

1. A method for constructing a wired-AND bus system, comprising:

(a) constructing a plurality of wired-AND bus segments, each bus segment having being small enough to avoid rise time problems;

(b) connecting pairs of bus segments together with bus bridges wherein each bus bridge selectively forwards transactions and commands from one bus segment to another; and

(c) connecting master devices and slave devices to the bus segments.

2. The method ofclaim 1 wherein each bridge comprises an address bitmap and wherein the bridge selectively forwards transactions based on information in the address bitmap.

3. The method ofclaim 1 wherein each bridge comprises a pair of range registers wherein values in the range registers determine which commands will be forwarded by the bridge.

4. The method ofclaim 1 wherein step (b) comprises connecting the bus segments into a tree hierarchy.

5. The method ofclaim 1 further comprising:

(d) programming one bus master to enter information into the address bitmaps and range registers of each bridge.

6. The method ofclaim 5 wherein the bus segments are connected into a tree hierarchy having a root level and the method comprises:

(e) locating the one bus master at the root level.

7. The method ofclaim 1 wherein at least some of the bridges are bidirectional bridges.

8. The method ofclaim 7 wherein each bidirectional bridge is comprised of two unidirectional bridges, each having a bridge ID.

9. The method ofclaim 8 wherein each unidirectional bridge has a different bridge ID.

10. The method ofclaim 1 wherein at least two slave devices have the same address and the method further comprises:

(f) sending a tunnel command from a bus master, the tunnel command containing data and a slave device address, to one of the bridges whereupon the bridge forwards the data to the slave device address.

11. Apparatus for constructing a wired-AND bus system, comprising:

a plurality of wired-AND bus segments, each bus segment having being small enough to avoid rise time problems;

bus bridges connecting pairs of bus segments together wherein each bus bridge selectively forwards transactions and commands from one bus segment to another; and

at least one master device and at least one slave device connected to the bus segments.

12. The apparatus ofclaim 11 wherein each bridge comprises an address bitmap and wherein the bridge selectively forwards transactions based on information in the address bitmap.

13. The apparatus ofclaim 11 wherein each bridge comprises a pair of range registers wherein values in the range registers determine which commands will be forwarded by the bridge.

14. The apparatus ofclaim 11 wherein the bus bridges connect the bus segments into a tree hierarchy.

15. The apparatus ofclaim 11 further comprising a configuration host that enters information into the address bitmaps and range registers of each bridge.

16. The apparatus ofclaim 15 wherein the bus segments are connected into a tree hierarchy having a root level and the configuration host is located at the root level.

17. The apparatus ofclaim 11 wherein at least some of the bridges are bidirectional bridges.

18. The apparatus ofclaim 17 wherein each bidirectional bridge is comprised of two unidirectional bridges, each having a bridge ID.

19. The apparatus ofclaim 18 wherein each unidirectional bridge has a different bridge ID.

20. The apparatus ofclaim 11 wherein at least two slave devices have the same address and the apparatus further comprises a bus master that sends a tunnel command containing data and a slave device address to one of the bridges whereupon the bridge forwards the data to the slave device address.

Images