GPU memory buffer pre-fetch and pre-back signaling to avoid page-fault

1. A method for demand paging in an input/output device, the method comprising:

tracking, by the input/output device, a usage of a first portion of mapped pages in a virtual memory system by an application executing on the input/output device, wherein at least the first portion of mapped pages is stored in a buffer, wherein the usage comprises a number of reads and writes to the buffer storing the first portion of mapped pages, and wherein the first portion of mapped pages represents a portion of a number of pages associated with the application, and wherein the first portion of mapped pages are backed into physical memory by a processor;

determining, by the input/output device, if a current location accessed in the buffer by the input/output device has crossed a watermark, wherein the watermark is a location within the buffer representing a percentage of usage of the first portion of mapped pages; and

in response to a determination that the current location accessed in the buffer by the input/output device has crossed the watermark, and without receiving a page fault, signaling the processor to back a second portion of mapped pages in physical memory, wherein the second portion of mapped pages represents a different portion of the number of pages associated with the application.

2. The method of claim 1 , further comprising:

backing, by the processor, the second portion of mapped pages to physical memory.

3. The method of claim 2 , further comprising:

receiving, at the input/output device, a signal from the processor indicating that the backing of the second portion of mapped pages is complete.

4. The method of claim 1 , wherein the buffer is one of a command stream buffer, a vertex buffer, a texture buffer, an instruction stream buffer, a rasterization buffer, and an intermediate stage flow buffer.

5. The method of claim 4 , wherein the buffer is the rasterization buffer and wherein the application is one of a block transfer (BLT) and a tiled renderer resolve.

6. The method of claim 1 , further comprising:

receiving the watermark from the processor.

7. The method of claim 1 , wherein the input/output device is a graphics processing unit.

8. The method of claim 1 , wherein the processor is a central processing unit.

9. An apparatus configured to perform demand paging for an input/output device comprising:

an input/output device configured to:

track a usage of a first portion of mapped pages in a virtual memory system by an application executing on the input/output device, wherein at least the first portion of mapped pages is stored in a buffer, wherein the usage comprises a number of reads and writes to the buffer storing the first portion of mapped pages, and wherein the first portion of mapped pages represents a portion of a number of pages associated with the application, and wherein the first portion of mapped pages are backed into physical memory by a processor;

determine if a current location accessed in the buffer by the input/output device has crossed a watermark, wherein the watermark is a location within the buffer representing a percentage of usage of the first portion of mapped pages; and

in response to a determination that the current location accessed in the buffer by the input/output device has crossed the watermark, and without receiving a page fault, signal the processor to back a second portion of mapped pages in physical memory, wherein the second portion of mapped pages represents a different portion of the number of pages associated with the application.

10. The apparatus of claim 9 , further comprising:

the processor, wherein the processor is configured to back the second portion of mapped pages to physical memory.

11. The apparatus of claim 10 , wherein the input/output device is further configured to:

receive a signal from the processor indicating that the backing of the second portion of mapped pages is complete.

12. The apparatus of claim 9 , wherein the buffer is one of a command stream buffer, a vertex buffer, a texture buffer, an instruction stream buffer, a rasterization buffer, and an intermediate stage flow buffer.

13. The apparatus of claim 12 , wherein the buffer is the rasterization buffer and wherein the application is one of a block transfer (BLT) and a tiled renderer resolve.

14. The apparatus of claim 9 , wherein the input/output device is further configured to:

receive the watermark from the processor.

15. The apparatus of claim 9 , wherein the input/output device is a graphics processing unit.

16. The apparatus of claim 9 , wherein the processor is a central processing unit.

17. The apparatus of claim 9 , wherein the input/output device and the processor are housed within a mobile device.

18. An apparatus configured to perform demand paging for an input/output device comprising:

means for tracking a usage of a first portion of mapped pages in a virtual memory system by an application executing on the input/output device, wherein at least the first portion of mapped pages is stored in a buffer, wherein the usage comprises a number of reads and writes to the buffer storing the first portion of mapped pages, and wherein the first portion of mapped pages represents a portion of a number of pages associated with the application, and wherein the first portion of mapped pages are backed into physical memory by a processor;

means for determining if a current location accessed in the buffer by the input/output device has crossed a watermark, wherein the watermark is a location within the buffer representing a percentage of usage of the first portion of mapped pages; and

means for signaling the processor to back a second portion of mapped pages in physical memory in response to a determination the current location accessed in the buffer by the input/output device has crossed the watermark, and without receiving a page fault, wherein the second portion of mapped pages represents a different portion of the number of pages associated with the application.

19. A non-transitory computer-readable storage medium storing instructions that, when executed, causes one or more processors configured to perform demand paging for an input/output device to:

track a usage of a first portion of mapped pages in a virtual memory system by an application executing on the input/output device, wherein at least the first portion of mapped pages is stored in a buffer, wherein the usage comprises a number of reads and writes to the buffer storing the first portion of mapped pages, and wherein the first portion of mapped pages represents a portion of a number of pages associated with the application, and wherein the first portion of mapped pages are backed into physical memory by a processor;

determine if a current location accessed in the buffer by the input/output device has crossed a watermark, wherein the watermark is a location within the buffer representing a percentage of usage of the first portion of mapped pages; and

in response to a determination that the current location accessed in the buffer by the input/output device has crossed the watermark, and without receiving a page fault, signal the processor to back a second portion of mapped pages in physical memory, wherein the second portion of mapped pages represents a different portion of the number of pages associated with the application.

20. A method for demand paging in an input/output device comprising:

tracking, by the input/output device, a currently accessed location in a page table of a first portion of mapped pages in a virtual memory system by an application executing on the input/output device, the page table mapping the first portion of mapped pages, and wherein the first portion of mapped pages represents a portion of a number of pages associated with the application, wherein page table entries for the first portion of mapped pages are stored in a memory management unit;

determining, by the input/output device, if the currently accessed location accessed in a last page of the page table by the input/output device has crossed a watermark; and

in response to a determination that the currently accessed location accessed in the last page of the page table by the input/output device has crossed the watermark, and without receiving a translation lookaside buffer (TLB) miss, signaling the memory management unit to fetch page table entries for a second portion of mapped pages, wherein the second portion of mapped pages represents a different portion of the number of pages associated with the application.

21. The method of claim 20 , further comprising:

fetching, by the memory management unit, page table entries for the second portion of mapped pages.

22. The method of claim 20 , further comprising:

receiving the watermark from a processor.

23. The method of claim 20 , wherein the input/output device is a graphics processing unit.

24. The method of claim 20 , wherein the memory management unit is an input/output memory management unit.

25. An apparatus configured to perform demand paging for an input/output device comprising:

an input/output device configured to:

track a currently accessed location in a page table of a first portion of mapped pages in a virtual memory system by an application executing on the input/output device, the page table mapping the first portion of mapped pages, wherein the first portion of mapped pages represents a portion of a number of pages associated with the application, wherein page table entries for the first portion of mapped pages are stored in a memory management unit;

determine if the currently accessed location accessed in a last page of the page table by the input/output device has crossed a watermark; and

in response to a determination that currently accessed location accessed in the last page of the page table by the input/output device has crossed the watermark, and without receiving a translation lookaside buffer (TLB) miss, signal the memory management unit to fetch page table entries for a second portion of mapped pages, wherein the second portion of mapped pages represents a different portion of the number of pages associated with the application.

26. The apparatus of claim 25 , further comprising:

the memory management unit, wherein the memory management unit is configured to:

fetch page table entries for the second portion of mapped pages.

27. The apparatus of claim 25 , wherein the input/output device is further configured to:

receive the watermark from a processor.

28. The apparatus of claim 25 , wherein the input/output device is a graphics processing unit.

29. The apparatus of claim 25 , wherein the memory management unit is an input/output memory management unit.

30. The apparatus of claim 25 , wherein the input/output device and the memory management unit are housed within a mobile device.

31. An apparatus configured to perform demand paging for an input/output device comprising:

means for tracking a currently accessed location in a page table of a first portion of mapped pages in a virtual memory system by an application executing on the input/output device, the page table mapping the first portion of mapped pages, wherein the first portion of mapped pages represents a portion of a number of pages associated with the application, and wherein page table entries for the first portion of mapped pages are stored in a memory management unit;

means for determining, by the input/output device, if the currently accessed location accessed in a last page of the page table by the input/output device has crossed a watermark; and

means for signaling the memory management unit to fetch page table entries for a second portion of mapped pages in response to a determination that the currently accessed location accessed in the last page of the page table by the input/output device has crossed the watermark, and without receiving a translation lookaside buffer (TLB) miss, wherein the second portion of mapped pages represents a different portion of the number of pages associated with the application.

32. A non-transitory computer-readable storage medium storing instructions that, when executed, causes one or more processors configured to perform demand paging for an input/output device to:

track a currently accessed location of a first portion of mapped pages in a virtual memory system by an application executing on the input/output device, the page table mapping the first portion of mapped pages, wherein the first portion of mapped pages represents a portion of a number of pages associated with the application, and wherein page table entries for the first portion of mapped pages are stored in a memory management unit;

determine if the currently accessed location accessed in a last page of the page table by the input/output device has crossed a watermark; and

in response to a determination that the currently accessed location accessed in the last page of the page table by the input/output device has crossed the watermark, signal the memory management unit to fetch page table entries for a second portion of mapped pages, and without receiving a translation lookaside buffer (TLB) miss, wherein the second portion of mapped pages represents a different portion of the number of pages associated with the application.