Electrical computers and digital data processing systems: input/ – Intrasystem connection – Bus access regulation
Reexamination Certificate
1998-12-15
2001-07-03
Myers, Paul R. (Department: 2181)
Electrical computers and digital data processing systems: input/
Intrasystem connection
Bus access regulation
C710S100000, C710S263000, C710S268000
Reexamination Certificate
active
06256699
ABSTRACT:
BACKGROUND OF THE INVENTION
1. Field of the Invention
The present invention relates to maintaining data coherency and consistency (“transaction ordering”) in data transfer operations which occur asynchronously from their corresponding interrupt signals. More particularly, the present invention relates to reliable interrupt generation and reception over a buffered bus.
2. The Background
Asynchronous communication between computer system components, such as controlling the flow of data operations between peripheral devices (“peripherals”), and a host processor through a buffered bus using interrupt signals (“interrupts”) to advise the host processor of a request for a particular operation, are known in the art.
Referring now to
FIG. 1
, peripherals such as network interface modules (port adapters)
10
,
12
connect network media from networks
14
,
16
, respectively, to a first bus
18
which may be a PCI (peripheral component interchange) bus or a similar tppe of bus. Other peripherals such as a storage adapter
20
, display adapter
22
and audio adapter
24
, to name a few, may also be attached to first bus
18
as shown in FIG.
1
.
A second bus
26
, preferably using a first bus architecture such as a local bus architecture couples one or more host processors
28
with one or more instances of host memory
30
.
A bridge
32
with buffering capability is used to couple a first bus
18
(there may be one or more instances of first bus
18
and bridge
32
) to second bus
26
.
When a peripheral has data requiring processing by host processor
28
, two events occur asynchronously. In the first event, the peripheral performs a write transaction which includes sending the data to host memory
30
through first bus
18
, buffering bridge
32
and second bus
26
. The write transaction includes an address phase indicating a target location in host memory
30
and at least one data phase indicating the data to be stored in host memory
30
, thus data is typically sent in a number of data blocks including data and header information indicating address information. In the second event, the peripheral uses an interrupt request to notify host processor
28
, even when host processor is busy (such as when it is executing program code), that a data transaction needs to be processed. The interrupt reaches host processor
28
through first bus
18
, bridge
32
and second bus
26
. Host processor
28
then responds to the interrupt request by accessing the data from host memory
30
and sends an acknowledgement signal back to the originating peripheral over second bus
26
, bridge
32
and first bus
18
.
Using first bus
18
and bridge
32
to send data and interrupt requests to host processor
28
presents a number of disadvantages. First, a delay period (“latency”) is incurred between the time that the interrupt request is sent to first bus
18
and the time that it is received by host processor
28
. Second, this latency may be increased depending upon the number of peripherals arbitrating for the use of the first bus
18
. Furthermore, such an approach inherently creates a race condition between data to be processed by host processor
28
and the interrupt requests corresponding to that data. Such a race condition can sometimes result in the interrupt being received and acted upon by host processor
28
before the corresponding data has been received in host memory
30
. In such a situation, data may be lost.
Some solutions to the race condition problem exist. In accordance with one solution, the peripheral signaling the interrupt performs a read of the data written to host memory
30
to verify receipt in memory prior to sending the interrupt signal over the buffered bus to the host processor
28
. This solution is easy to implement. Having the peripheral perform the read is much better than having the host processor perform the read from a processor throughput point of view. A major disadvantage is that the peripheral must execute a read cycle over the bus which could take a relatively long time due to bus arbitration and time required to traverse all bridges (there may be more than one) in the path. Another problem with this solution is that it can have the side effect of providing the host processor
28
with a spurious interrupt This can happen when (1) the peripheral device writes block
1
of data and block
2
of data into host memory
30
; (2) the host memory
30
receives block
1
and block
2
; (3) the peripheral device begins reading the host memory to verify receipt and reads block
1
; (4) the peripheral device generates an interrupt for block
1
to the host processor
28
; (5) the receipt of block
2
by the peripheral device is delayed due to latency; (6) the host processor
28
receives the first interrupt, acknowledges it and acts on it by processing all of the contents of host memory
30
including block
1
and block
2
; (7) the read of block
2
by the peripheral device is completed and the peripheral device sends a second interrupt request to the host processor
28
; and (8) the host processor
28
acknowledges and acts on the second interrupt only to find that it has already processed the contents of host memory
30
.
In another solution the host processor
30
, or a device associated with it, performs a read operation to a register of the peripheral device in response to receipt of the interrupt request before accessing the data written by the peripheral into host memory. This guarantees that all of the data to be written to the host memory will necessarily have cleared the buffers before this step can be completed. This approach can increase latency because the read operation can be further delayed due to bus arbitration issues under loaded conditions.
In accordance with yet another solution shown in
FIG. 2
, a separate non-buffered path (
34
,
35
,
38
,
40
,
42
) is provided for each peripheral to signal the interrupt request to a special interrupt input register
44
in the host processor. This approach requires the provision of a separate path outside the bus structure for each such peripheral used for data communication and therefore requires special purpose hardware. This approach is also subject to a race condition because the interrupt can arrive before all data has cleared all intermediate buffers.
Accordingly, a need exists for a method and apparatus for reliable generation, transmission and reception of interrupt requests over buffered buses without the interrupt requests incurring significant delay due to buffering and without creating a race condition between the interrupt requests and corresponding data.
SUMMARY OF THE INVENTION
A method and apparatus for reliable interrupt reception over a buffered bus utilizes a mailbox register to receive interrupt request information sent after a data write transaction. The data is sent from an initiating peripheral device over the buffered bus to arrive with an arbitrary delay at the host memory. After completing the sending phase for the data the initiating peripheral device sends a mailbox register data block containing an interrupt request to a mailbox register associated with the host processor. Because the mailbox register data block will necessarily arrive after the receipt of the actual data in the host memory because it is following the actual data through the same buffered bus, the interrupt will be properly sequenced with the receipt of data.
REFERENCES:
patent: 4991169 (1991-02-01), Davis et al.
patent: 5122691 (1992-06-01), Balakrishnan
patent: 5301275 (1994-04-01), Vanbuskirk et al.
patent: 5594727 (1997-01-01), Kolbenson et al.
patent: 5659542 (1997-08-01), Bell et al.
patent: 5721828 (1998-02-01), Frisch
patent: 5732079 (1998-03-01), Castrigno
patent: 5742649 (1998-04-01), Muntz et al.
patent: 5796732 (1998-08-01), Mazzola et al.
patent: 5826093 (1998-10-01), Assouad et al.
patent: 5898848 (1999-04-01), Gulick
patent: 5913045 (1999-06-01), Gillespie et al.
patent: 5920572 (1999-07-01), Washington et al.
patent: 5925099 (1999-07-01), Futral et al.
patent: 5928338 (1999-07-01), Lawma
Cisco Technology Inc.
Myers Paul R.
Ritchie David B.
Thelen Reid & Priest LLP
LandOfFree
Reliable interrupt reception over buffered bus does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Reliable interrupt reception over buffered bus, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Reliable interrupt reception over buffered bus will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2543848