vm: Remove kernel stack swapping support, part 7

[markj]

Summary

Remove some uses of PHOLD which were there only to prevent the process'
threads from being swapped out.

[imp]

sys/dev/nvme/nvme_ctrlr.c

1266

Why isn't this needed? The user buffer is in the user's process, and almost never on the kernel stack.

[markj]

sys/dev/nvme/nvme_ctrlr.c

1266

I'm not sure why it was needed even before. The current process won't exit while this thread is sleeping below.

[imp]

Any idea how to test this?
Will something else killing, eg, camcontrol which has issued the command before the I/O is complete be a good test?
I guess I'm nervous for reasons that maybe are completly irrational and looking for some way to test this...

sys/dev/nvme/nvme_ctrlr.c

1266

Yea, I've never gone and seen why it's needed, only to have people comment on it when I didn't do it elsewhere...
But maybe it was part of the cargo-cult and/or battle scars from the memory starved era where a process's stack could get swapped out while the I/O was pending...
Re-reading the PHOLD and related code suggests that this is fine... I didn't see any of the other side effects I assumed would happen...

[markj]

In D46118#1051156, @imp wrote:

Any idea how to test this?
Will something else killing, eg, camcontrol which has issued the command before the I/O is complete be a good test?
I guess I'm nervous for reasons that maybe are completly irrational and looking for some way to test this...

I will ask Peter to run stress2. I don't know of any good way to specifically test this change.