Feature: add a method for binding shared virtual addressing

[michael-adler]

Description

Add fpgaBindSVA() to allocate and bind PASIDs to FPGA ports. Shared virtual addressing requires hardware support, both in the CPU and on the FPGA.

fpgaBindSVA() depends on /dev/dfl-pci-sva from the dfl-pci driver. IOVA-based pinned DMA remains available with older drivers and hardware that does not support PASID.

OPAE behavior is unchanged when fpgaBindSVA() is not called.

Collateral (docs, reports, design examples, case IDs):

• [ ] Document Update Required? (Specify FIM/AFU/Scripts)

Tests added:

Tests run:

[coveralls]

Pull Request Test Coverage Report for Build 6243520861

• 3 of 38 (7.89%) changed or added relevant lines in 3 files are covered.
• 15 unchanged lines in 4 files lost coverage.
• Overall coverage decreased (-0.2%) to 71.808%

---

Changes Missing Coverage	Covered Lines	Changed/Added Lines	%
libraries/libopae-c/api-shell.c	0	6	0.0%
libraries/plugins/vfio/opae_vfio.c	1	30	3.33%
<!--	Total:	3	38	7.89%	-->

Files with Coverage Reduction	New Missed Lines	%
libraries/plugins/xfpga/open.c	1	74.68%
libraries/plugins/uio/opae_uio.c	4	90.18%
libraries/plugins/vfio/opae_vfio.c	4	82.64%
libraries/plugins/xfpga/mmio.c	6	82.87%
<!--	Total:	15		-->

Totals
Change from base Build 6238677251:	-0.2%
Covered Lines:	15744
Relevant Lines:	21925

---

💛 - Coveralls

[michael-adler]

@rweight and @matthew-gerlach - These are the user space changes that go along with the dfl-pci driver updates.

Note that the changes here will work with old drivers. They won't find /dev/dfl-pci-sva and will be fine with that.

[michael-adler]

@tswhison - Thanks for going through this. I'll come back to it once the low-level device mechanism is set with @rweight. The general framework here will remain, but the details down inside opae_vfio.c are in flux.

[michael-adler]

@tswhison - I'm confused by the failed test at opae-sdk/tests/opae-v/test_opae_vfio_c.cpp:3246. I think I didn't change anything relevant.

I think the test causes vfio_fpgaDestroyEventHandle() to close fd 0. Seems like a bad idea.

[michael-adler]

@tswhison - I still think that opae-sdk/tests/opae-v/test_opae_vfio_c.cpp:3246 is a broken test since it closes fd 0. I exposed that bug by causing another test to do the same thing. My case is fixed in the latest commit by adding flags for pasid so that when the handle is initialized to 0 it doesn't look like there's a file open.

[tswhison]

Thanks for pointing this out. I will submit a fix.