Method and apparatus for high-performance page-fault handling for multi-tenant scalable accelerators

Patent metadata

Field	Value
Publication number	US-12443477-B2
Application number	US-202117560170-A
Country	US
Kind code	B2
Filing date	Dec 22, 2021
Priority date	Jul 27, 2021
Publication date	Oct 14, 2025
Grant date	Oct 14, 2025

Abstract

Apparatus and method for high-performance page fault handling. For example, one embodiment of an apparatus comprises: one or more accelerator engines to process work descriptors submitted by clients to a plurality of work queues; fault processing hardware logic associated with the one or more accelerator engines, the fault processing hardware logic to implement a specified page fault handling mode for each work queue of the plurality of work queues, the page fault handling modes including a first page fault handling mode and a second page fault handling mode.

First claim

What is claimed is:

1. An apparatus comprising:
one or more accelerator engines to process work descriptors submitted by clients to a plurality of work queues; and
fault processing hardware logic associated with the one or more accelerator engines, the fault processing hardware logic to implement a specified page fault handling mode for each work queue of the plurality of work queues, page fault handling modes for the plurality of work queues including a first page fault handling mode and a second page fault handling mode, wherein upon a page fault occurring on an address associated with a completion record buffer, information associated with the page fault is submitted to an event log but not the completion record buffer.

2. The apparatus of claim 1 , wherein in the first page fault handling mode, the fault processing hardware logic is to notify a client and terminate processing of a work descriptor associated with the page fault.

3. The apparatus of claim 1 wherein the information is to allow a client to continue processing or re-submit work associated with a work descriptor if the work was only partially completed due to the page fault.

4. The apparatus of claim 3 wherein the information comprises a fault-address and a bytes completed field to indicate a number of source bytes processed before the page fault occurred.

5. The apparatus of claim 1 wherein the event log is to be associated with a particular accelerator engine of the one or more accelerator engines or a particular work queue of the plurality of work queues.

6. The apparatus of claim 1 wherein in the second page fault handling mode, processing of a work descriptor is to be paused and the fault processing hardware logic is to generate a page request to memory management hardware logic, the processing of the work descriptor to resume only after receiving a response from the memory management hardware logic.

7. A method comprising:
processing work descriptors by an accelerator engine, the work descriptors submitted by clients to a plurality of work queues; and
in response to detecting a page fault associated with a work descriptor in a work queue:
determining a page fault handling mode associated with the work queue, and
notifying a client who submitted the work descriptor and terminating processing of the work descriptor associated with the page fault if the page fault handling mode comprises a first page fault handling mode,
wherein if the page fault occurs on an address associated with a completion record buffer, information associated with the page fault is submitted to an event log but not the completion record buffer.

8. The method of claim 7 further comprising:
if the page fault handling mode comprises a second page fault handling mode, performing operations of: pausing work associated with the work descriptor and transmitting a page fault request to memory management hardware logic, and resuming the work associated with the work descriptor after receiving a response from the memory management hardware logic.

9. The method of claim 7 wherein, based on the information, the client is to continue processing or re-submit work associated with the work descriptor if the work was only partially completed due to the page fault.

10. The method of claim 9 wherein the information comprises a fault-address and a bytes completed field to indicate a number of source bytes processed before the page fault occurred.

11. The method of claim 7 wherein information associated with a page fault for a second work descriptor is also submitted to the event log but not the completion record buffer.

12. The method of claim 11 wherein at least one work descriptor comprises a batch descriptor comprising a descriptor list address identifying a memory location of a batch of descriptors or a drain descriptor to cause the accelerator engine to wait for completion of specified preceding descriptors in the work queue before processing the drain descriptor.

13. A non-transitory machine-readable medium having program code stored thereon which, when executed by a machine, causes the machine to perform:
processing work descriptors by an accelerator engine, the work descriptors submitted by clients to a plurality of work queues; and
in response to detecting a page fault associated with a work descriptor in a work queue:
determining a page fault handling mode associated with the work queue, and
notifying a client who submitted the work descriptor and terminating processing of the work descriptor associated with the page fault if the page fault handling mode comprises a first page fault handling mode,
wherein if the page fault occurs on an address associated with a completion record buffer, information associated with the page fault and usable by a client and/or system software to resolve the page fault is submitted to an event log but not the completion record buffer.

14. The non-transitory machine-readable medium of claim 13 , wherein the machine is caused to further perform:
if the page fault handling mode comprises a second page fault handling mode, performing: pausing work associated with the work descriptor and transmitting a page fault request to memory management hardware logic, and resuming the work associated with the work descriptor after receiving a response from the memory management hardware logic.

15. The non-transitory machine-readable medium of claim 13 wherein, based on the information, the client is to continue processing or re-submit work associated with the work descriptor if the work was only partially completed due to the page fault.

16. The non-transitory machine-readable medium of claim 15 wherein the information comprises a fault-address and a bytes completed field to indicate a number of source bytes processed before the page fault occurred.

17. The non-transitory machine-readable medium of claim 13 wherein information associated with a page fault for a second work descriptor is also submitted to the event log but not the completion record buffer.

18. The non-transitory machine-readable medium of claim 17 wherein at least one work descriptor comprises a batch descriptor comprising a descriptor list address identifying a memory location of a batch of descriptors or a drain descriptor to cause the accelerator engine to wait for completion of specified preceding descriptors in the work queue before processing the drain descriptor.

Assignees

• Intel Corp

Inventors

• Kakaiya Utkarsh Y
• Lantz Philip
• Kumar Sanjay
• Sankaran Rajesh
• Ranganathan Narayan
• Gayen Saurabh
• Joshi Dhananjay
• Rao Nikhil P

Classifications

• G06F11/0784

  Routing of error reports, e.g. with a specific transmission path or data flow · CPC title

• G06F11/0775

  Content or structure details of the error report, e.g. specific table structure, specific error fields · CPC title

• G06F12/1081

  for peripheral access to main memory, e.g. direct memory access [DMA] · CPC title

• G06F12/1009

  using page tables, e.g. page table structures · CPC title

• G06F11/079 Primary

  Root cause analysis, i.e. error or fault diagnosis (in a hardware test environment G06F11/22; in a software test environment G06F11/36) · CPC title