Emulating memory mapped I/O for coherent accelerators in error state

Patent metadata

Field	Value
Publication number	US-9690495-B2
Application number	US-201514931612-A
Country	US
Kind code	B2
Filing date	Nov 3, 2015
Priority date	Nov 3, 2015
Publication date	Jun 27, 2017
Grant date	Jun 27, 2017

Abstract

Embodiments disclose techniques for emulating memory mapped I/O (MMIO) for coherent accelerators in an error state. In one embodiment, once an operating system determines that a processor is unable to access a coherent accelerator via a MMIO operation, the operating system deletes one or more page table entries associated with MMIO of one or more hardware contexts of the coherent accelerator. After deleting the page table entries, the operating system can detect a page fault associated with execution of a process by the processor. Upon determining that the page fault was caused by the process attempting to access one of the deleted page table entries while executing a MMIO operation, the operating system emulates the execution of the MMIO operation for the faulting process, giving the process the illusion that its requested MMIO operation was successful.

First claim

What is claimed is:

1. A method for emulating a memory mapped I/O (MMIO) operation for a coherent accelerator, the method comprising:
upon determining that a processor is unable to access the coherent accelerator via a MMIO operation, deleting one or more page table entries associated with MMIO of one or more hardware contexts of the coherent accelerator;
detecting a page fault associated with execution of a process by the processor after deleting the page table entries; and
upon determining that the page fault was caused by the process attempting to access one of the deleted page table entries while executing a MMIO operation, emulating the execution of the MMIO operation for the faulting process.

2. The method of claim 1 , further comprising, upon detecting the page fault, receiving information that specifies at least one of an effective address that triggered the page fault, and a program counter for the faulting process.

3. The method of claim 2 , wherein determining that the page fault was caused by the process attempting to access one of the deleted page table entries comprises identifying that the effective address belongs to a range of addresses used for MMIO between the processor and the coherent accelerator.

4. The method of claim 2 , wherein emulating execution of the MMIO operation comprises:
locating an instruction for the MMIO operation based on the program counter; and
parsing a bit pattern for the instruction to determine one or more characteristics of the MMIO operation, wherein the one or more characteristics indicate at least one of a type of MMIO instruction, a source memory location, and a destination memory location.

5. The method of claim 4 , further comprising:
upon determining that the instruction is for a MMIO read operation to the coherent accelerator, returning an outcome of the emulated execution to the faulting process, wherein the outcome comprises invalid data, and incrementing the program counter to a subsequent instruction.

6. The method of claim 4 , further comprising:
upon determining that the instruction is for a MMIO write operation to the coherent accelerator, ignoring the instruction and incrementing the program counter to a subsequent instruction.

7. The method of claim 5 , wherein a value of the invalid data is negative one, and wherein the invalid data indicates to the faulting process that the coherent accelerator is inaccessible, via MMIO, from the processor.

8. The method of claim 7 , wherein the process ceases requests to perform MMIO operations to the coherent accelerator upon indication of the invalid data.

9. A system, comprising:
a processor;
a coherent accelerator; and
a memory storing program code, which, when executed on the processor, performs an operation comprising:
upon determining that the processor is unable to access the coherent accelerator via a MMIO operation, deleting one or more page table entries associated with MMIO of one or more hardware contexts of the coherent accelerator;
detecting a page fault associated with execution of a process by the processor after deleting the page table entries; and
upon determining that the page fault was caused by the process attempting to access one of the deleted page table entries while executing a MMIO operation, emulating the execution of the MMIO operation for the faulting process.

10. The system of claim 9 , wherein the operation further comprises:
upon detecting the page fault, receiving information that specifies at least one of an effective address that triggered the page fault, and a program counter for the faulting process.

11. The system of claim 10 , wherein determining that the page fault was caused by the process attempting to access one of the deleted page table entries comprises identifying that the effective address belongs to a range of addresses used for MMIO between the processor and the coherent accelerator.

12. The system of claim 10 , wherein emulating execution of the MMIO operation comprises:
locating an instruction for the MMIO operation based on the program counter; and
parsing a bit pattern for the instruction to determine one or more characteristics of the MMIO operation, wherein the one or more characteristics indicate at least one of a type of MMIO instruction, a source memory location, and a destination memory location.

13. The system of claim 12 , wherein the operation further comprises:
upon determining that the instruction is for a MMIO read operation to the coherent accelerator, returning an outcome of the emulated execution to the faulting process, wherein the outcome comprises invalid data, and incrementing the program counter to a subsequent instruction.

14. The system of claim 12 , wherein the operation further comprises:
upon determining that the instruction is for a MMIO write operation to the coherent accelerator, ignoring the instruction and incrementing the program counter to a subsequent instruction.

15. The system of claim 13 , wherein a value of the invalid data is negative one, wherein the invalid data indicates to the faulting process that the coherent accelerator is inaccessible, via MMIO, from the processor, and wherein the faulting process ceases requests to perform MMIO operations to the coherent accelerator upon indication of the invalid data.

16. A computer program product, comprising:
a non-transitory computer-readable storage medium having computer-readable program code embodied therewith, the computer-readable program code executable by one or more computer processors to perform an operation comprising:
upon determining that a processor is unable to access a coherent accelerator via a MMIO operation, deleting one or more page table entries associated with MMIO of one or more hardware contexts of the coherent accelerator;
detecting a page fault associated with execution of a process by the processor after deleting the page table entries; and
upon determining that the page fault was caused by the process attempting to access one of the deleted page table entries while executing a MMIO operation, emulating the execution of the MMIO operation for the faulting process.

17. The computer program product of claim 16 , wherein the operation further comprises:
upon detecting the page fault, receiving information that specifies at least one of an effective address that triggered the page fault, and a program counter for the faulting process.

18. The computer program product of claim 17 , wherein determining that the page fault was caused by the process attempting to access one of the deleted page table entries comprises identifying that the effective address belongs to a range of addresses used for MMIO between the processor and the coherent accelerator.

19. The computer program product of claim 16 , wherein emulating execution of the MMIO operation comprises:
locating an instruction for the MMIO operation based on the program counter; and
parsing a bit pattern for the instruction to determine one or more characteristics of the MMIO operation, wherein the one or more characteristics indicate at least one of a type of MMIO instruction, a source memory location, and a destination memory location.

20. The computer program product of claim 19 , the operation further comprising:
upon determining that the instruction is for a MMIO read operation to the coherent accelerator, returning an outcome of the emulated execution to the faulting process and incrementing the program counter to a subsequent instruction, wherein the outcome comprises i

Assignees

• IBM

Inventors

• Albot Andre L
• Aslot Vishal C
• Mealey Bruce
• STILWELL Nicholas

Classifications

• G06F3/0611 Primary

  in relation to response time · CPC title

• G06F3/0673

  Single storage device · CPC title

• G06F9/455 Primary

  Emulation; Interpretation; Software simulation, e.g. virtualisation or emulation of application or operating system execution engines · CPC title

• G06F3/0659

  Command handling arrangements, e.g. command buffers, queues, command scheduling · CPC title