Method and Apparatus for Translating Signals Between Clock Domains of Different Frequencies

1. Field of the Invention

This invention relates generally to the field of data processing systems, and particularly to a method and apparatus for synchronizing signals in one clock domain with another clock domain.

2. Background Art

Most present day personal and workstation computers are organized as shown in FIG. 1. A central processing unit (CPU) 10 is coupled to a local CPU bus 12 as is the main memory 14. A cache memory 16 may also be coupled to bus 12. Communication with peripheral devices 22 is accomplished over an expansion bus 20 coupled to I/O port 18. A number of expansion bus standards have been developed. One such standard is the industry standard architecture (ISA) bus that is widely used in personal computer systems. More recently, the extended ISA (EISA) bus has been developed. This is a superset of the ISA bus that includes all ISA bus features, together with extensions to enhance system performance and capabilities. Both the ISA bus and EISA bus have relatively narrow bandwidths in comparison to the bandwidth available on local busses for state-of-the-art high performance CPUs. This results in I/O bottlenecks and limits the ability of the system to service high performance peripherals such as local area network (LAN), SCSI, graphics and video.

In order to deal with the inherent limitations of earlier generation expansion bus standards, a new bus standard designated the "Peripheral Component Interconnect" (PCI) bus has been developed. PCI is a fast local bus particularly adapted for servicing high performance peripheral units. A typical system employing a PCI bus is illustrated in FIG. 2. Here, CPU 10 is coupled to CPU bus 12 as in the previously described system. A PCI/host bridge unit 30 is coupled to CPU bus 12 to provide the interface with PCI bus 32. PCI/host bridge unit 30 preferably also includes the management functions for the main memory 14 and cache memory 16. PCI compatible peripheral units 34 are connected to the PCI bus.

In order to maintain system compatibility with peripheral units designed to operate with earlier generation ISA/EISA expansion busses, a PCI/expansion bus bridge unit 40 is coupled between the PCI bus and the ISA or EISA expansion bus 20. A system such as shown in FIG. 2 is thus able to take advantage of the increased performance capabilities of the PCI bus while maintaining access to the large existing base of ISA and EISA expansion cards and their corresponding software applications.

The PCI bus is particularly designed to address the growing need for a standardized local bus that is not directly dependent on the speed and size of the CPU bus. New generations of personal computer system software with sophisticated graphical interfaces, multi-tasking and multi-threading bring new requirements that traditional PC input/output architectures cannot satisfy. As indicated in FIG. 2, the input/output functions of the system are decoupled from the processor and memory assets, thereby enabling the input/output design of the system to remain stable over multiple generations of processor and/or memory technology. Regardless of new requirements or technology applied on the CPU side of the host bridge, the PCI side may remain unchanged, which allows re-usability of not only the remainder of the platform chip set, including the PCI/expansion bus bridge, but also all of the other I/O functions interfaced at both the PCI and expansion bus levels.

The PCI/expansion bus bridge unit 40 integrates several bus functions on both the PCI and expansion busses. For the PCI local bus, the functions include PCI local bus arbitration and default bus driver. In the case of an EISA expansion bus, the central functions include the EISA bus controller, EISA arbiter and EISA data swap logic. The PCI/EISA bridge also integrates system functions, including PCI parity and system error reporting, buffer coherency management protocol, PCI and EISA memory and I/O address mapping and decoding. For maximum flexibility, all of these functions are programmable, thereby allowing for a variety of optional features. A particular PCI/EISA bridge set has been developed for the Intel Pentium.TM. processor. This chip set is illustrated in FIG. 3 and comprises a PCI/EISA bridge (PCEB) and a EISA system component (ESC). These two components work in tandem to provide an EISA I/O sub-system interface for personal computer platforms based on the Pentium.TM. processor chip and PCI bus standard.

The hierarchy of the CPU bus as the execution bus, PCI local bus as a primary I/O bus and EISA bus as a secondary I/O bus allows concurrency for simultaneous operations in all three bus environments. Data buffering permits concurrency for operations that cross over from one bus environment to another. For example, a PCI device may post data into the PCEB, permitting the PCI local bus transaction to complete in a minimum time and freeing up the PCI local bus for further transactions. The PCI device need not wait for the transfer to complete to its final destination. Meanwhile, any ongoing EISA bus transactions are permitted to complete. The posted data will then be transferred to its EISA bus destination when the EISA bus is available. The PCEB-ESC chip set implements extensive buffering for PCI-to-EISA and EISA-to-PCI bus transactions. In addition to concurrency for the operation of the cross-bus environments, data buffering allows advantage to be taken of the fastest operations within a particular bus environment via PCI burst transfers and EISA burst transfers.

The PCI local bus with 132 MByte/second and EISA with 33 MByte/second peak data transfer rates represent bus environments with significantly different bandwidths. Without buffering, transfers that cross between bus environments would be performed at the speed of the slower bus. Data buffers provide a mechanism for data rate adoption so that the usable bandwidth of the fast bus environment (PCI) is not significantly impacted by the slower bus environment (EISA).

The PCEB can be either a master or slave on the PCI local bus and it is optimized for bus frequencies from 25 megahertz to 33 megahertz. For PCI-initiated transfers, the PCEB functions only as a slave. For EISA-initiated transfers to the PCI local bus, the PCEB is a PCI master. The PCEB contains a fully EISA master and slave interface. The PCEB directly drives eight EISA slots without external data or address buffering. The PCEB functions as a master or slave on the EISA bus only for transfers between the EISA bus and PCI local bus. For transfers confined to the EISA bus, the PCEB functions neither as a master nor slave. However, the data swap logic contained in the PCEB is involved in such transfers if data size translation is required.

The ESC implements system functions, such as timer/counter, DMA, and interrupt controller. The ESC also implements EISA sub-system control functions, such as EISA bus controller and EISA bus arbiter.

The present invention provides a mechanism for ensuring rapid synchronization of the PCEB and ESC units when the CPU commands a change in the PCI clock frequency and for translating signals in the EISA (slow) clock domain into signals that can be propagated in the PCI (fast) clock domain.

The PCI bus is snooped for access to a clock divisor control register. The value of the clock divisor controls the generation of the EISA clock as a submultiple of the PCI clock. The EISA clock is generated in the ESC unit. When acceptance of a new clock divisor is sensed by the snooping circuitry in the PCEB, the clock divisor is latched and the PCEB timing circuitry is adjusted accordingly to maintain synchronism with the ESC.

Synchronizer circuitry in the PCEB translates signals from the EISA clock domain into the PCI clock domain. A synchronizing pulse signal is developed at the frequency of the EISA clock, but phase synchronized to the PCI clock. The synchronizing pulse signal is employed to gate the EISA clock domain signal so that it is synchronized in the PCI clock domain.

FIG. 1 is a simplified block diagram of a prior art computer system.

FIG. 2 is a simplified block diagram of a computer system incorporating a peripheral component interconnect (PCI) bus.

FIG. 3 is a simplified block diagram of a PCI/expansion bus bridge.

FIG. 4 is a block diagram of a signal synchronizing circuit according to the present invention.

FIG. 5 is a timing diagram illustrating the operation of the circuit of FIG. 4.

In the following description, for purposes of explanation and not limitation, specific details are set forth in order to provide a thorough understanding of the present invention. However, it will be apparent to one skilled in the art that the present invention may be practiced in other embodiments that depart from these specific details. In certain instances, detailed descriptions of well-known data processing techniques, devices and circuits are omitted so as to not obscure the description of the present invention with unnecessary detail.

FIG. 4 illustrates the implementation of the present invention within the PCEB. There are two main functional blocks. These are the snooping mechanism block 100 and the synchronizer block 101. The CPU can change the frequency of its clock, and hence the frequency at which the PCI bus operates, on the fly. This requires that both the PCEB and ESC likewise change the frequency of their respective operations. This is accomplished by reprogramming a configuration register in the ESC. The PCEB and ESC work in tandem, and therefore must synchronize the change in frequency of their internal operations. Such synchronization is accomplished primarily within the PCEB by first snooping the frequency change request and then adjusting the internal timing circuitry to the newly requested frequency.

The CPU effects frequency changes by writing to I/O ports 22h and 23h. This loads the configuration registers in the ESC with the appropriate values. These I/O registers are used as an index address register (22h) and an index data register (23h). The index address register is used to write the configuration register address. The data (i.e., the configuration register address) in register 22h is used to select a particular configuration register. The selected configuration register is written to with a write operation to the index data register at I/O address 23h. The speed of the EISA bus is controlled by the EISA clock divisor in register location 4Dh. The EISA bus clock is divided down from the PCI clock by a factor of four (for a PCI clock frequency of 33.33 megahertz) or a factor of three (for a PCI clock frequency of 25 megahertz). Thus, in either case, the EISA clock frequency is maintained at 8.33 megahertz. It will be recognized that other embodiments may have additional primary I/O bus frequency options, in which case the clock divisor may assume any of a plurality of values as necessary to divide the clock frequency down to an appropriate value for operation of the secondary I/O bus.

The snooping mechanism block 100 permits the PCEB to sense a change in the EISA bus frequency concurrently with receipt of the frequency change request in the ESC configuration register. The ESC and the PCEB are thereby able to simultaneously change frequency after the I/O write to port 23h completes. Flip flop 102 receives bit 24 directly from the PCI bus. This bit carries the clock divisor bit when the EISA clock divisor register is written to. State machine 104 is idle until there is a write to I/O port 22h. State machine 104 then waits until there is a write to I/O 23h. It monitors the write and looks for successful completion as indicated as by the assertion of a target ready signal on the PCI bus. The assertion of the signal indicates that the ESC was able to latch the value of the clock divisor bit in the appropriate register. State machine 104 then asserts a mux selection signal on line 105 to latch the value of the clock divisor bit in flip flop 108. Prior to receipt of the target ready signal from the ESC, multiplexer 106 feeds the output of flip flop 102 through to the input of flip flop 108. Flip flop 108 is cleared by a system reset to force the clock divisor bit stored therein to the default value. The output of multiplexer 106 is designated DIVIDE.sub.-- BY.sub.-- BIT and constitutes a PCEB copy of the clock divisor bit latched in the ESC configuration register. This enables the PCEB to adjust its internal timing circuitry concurrently with adjustment of the ESC's internal timing circuitry so that the two units remain in synchronism.

DIVIDE.sub.-- BY.sub.-- BIT is asserted as an input at frequency shift state machine 110 of synchronizer block 101. This block synchronizes slow clock (BCLK) signals to the fast clock (FCLK) more quickly than conventional techniques. Proper operation of synchronizer block 101 requires that one edge of each clock maintains a stable relationship with the other and that both edges are in the same direction (i.e., one clock edge does not skew through the other clock edge and both edges must be either falling or rising). In addition, the frequency of BCLK must be an integer submultiple of the frequency of FCLK.

The output of state machine 110, designated as signal DIVIDE.sub.-- BY, is synchronized with the rising edge of the PCI clock signal (FCLK). Signal DIVIDE.sub.-- BY is asserted as an input to synchronizing state machine 112. State machine 112 also receives as inputs the PCI clock signal (FCLK) and the output FCLK.sub.-- BASE from reference block 114. Signal FCLK.sub.-- BASE has a frequency equal to one-half of the EISA clock frequency (BCLK), but is synchronized to FCLK. The timing relationship of FCLK.sub.-- BASE to the fast and slow clock signals FCLK and BCLK is shown on the top three lines of FIG. 5.

Synchronizing state machine 112 generates output signal SYNC.sub.-- PULSE based on the polarities of DIVIDE.sub.-- BY and FCLK.sub.-- BASE. A data signal in the EISA domain is designated BCLK.sub.-- SIG. Knowing when this signal is stable relative to FCLK.sub.-- BASE, the SYNC.sub.-- PULSE signal is generated in the appropriate FCLK cycle to allow a stable value on BCLK.sub.-- SIGNAL to pass into the PCI (FCLK) time domain. Thus, the value on BCLK.sub.-- SIGNAL is synchronized to the FCLK time domain by combinatorial block 116.

Referring again to FIG. 5, it will be seen that signal SYNC.sub.-- PULSE occurs at the rate of BCLK, but is synchronized with FCLK. The BCLK.sub.-- SIG changes state on the falling edge of BCLK and is stable at the time that signal SYNC.sub.-- PULSE is asserted. The BCLK.sub.-- SIG propagates as FCLK.sub.-- SIG during the interval of SYNC.sub.-- PULSE.

It will be recognized that the above described invention may be embodied in other specific forms without departing from the spirit or essential characteristics of the disclosure. Thus, it is understood that the invention is not to be limited by the foregoing illustrative details, but rather is to be defined by the appended claims.