Put a workaround in for command timeout malfunctioning
ClosedPublic
Actions

Authored by imp on Oct 25 2018, 8:52 PM.

Details

Reviewers

jimharris
chuck

Commits

rS339775: Put a workaround in for command timeout malfunctioning

Summary

At least one NVMe drive has a bug that makeing the Command Time Out
PCIe feature unreliable. The workaround is to disable this
feature. The driver wouldn't deal correctly with a timeout anyway.
Only do this for drives that are known bad.

Diff Detail

Lint

Lint Passed

Unit

No Test Coverage

Build Status

Buildable 20435
Build 19873: arc lint + arc unit

Event Timeline

imp created this revision.Oct 25 2018, 8:52 PM

Harbormaster completed remote builds in B20432: Diff 49631.Oct 25 2018, 8:52 PM

imp added reviewers: jimharris, chuck.Oct 25 2018, 8:52 PM

I think you mean "completion timeout", not "command timeout" - in both the commit message and some of the comments.

sys/dev/nvme/nvme.c
289	DEVICE_CTL2 is 16 bits, not 32. This functionally still works, but I think a uint16_t would be more precise.
sys/dev/nvme/nvme_private.h
251	Whitespace looks inconsistent in general with the line before it.

This revision now requires changes to proceed.Oct 25 2018, 9:30 PM

jimharris added inline comments.Oct 25 2018, 9:31 PM

sys/dev/nvme/nvme_private.h
251	This comment should probably say "PCIe completion timeout" instead of "transaction timeout".

imp marked 2 inline comments as done.Oct 25 2018, 10:33 PM

imp added inline comments.

sys/dev/nvme/nvme.c
289	OK. I see that now. The vendor spec I got about this problem listed the upper 2 bytes as reserved anyway. The PCIe spec, however is clear that the next two bytes are for status (even though the spec defines it as MBZ).
sys/dev/nvme/nvme_private.h
251	OK to both.