Drop Giant before sleeping in linux_wait_for_timeout_common().

[markj]

Summary

We may get here with Giant held during an ACPI resume.

Sleep routines are supposed to drop Giant before relinquishing
the CPU. _sleep() handles this, but the LinuxKPI calls the sleepqueue
functions directly.

Test Plan

Pete Wright reported a panic during resume that was addressed
by this patch.

[• hselasky]

Looks good.

[• hselasky]

Was it investigated if there are missing DROP/PICKUP GIANT macros in the pci resume / suspend device methods?

[markj]

In D10414#215899, @hselasky wrote:

Was it investigated if there are missing DROP/PICKUP GIANT macros in the pci resume / suspend device methods?

Yes, I looked through a number of existing device suspend/resume methods and did not find any special handling for Giant.

[• hselasky]

Can this issue also be fixed by dropping / picking up giant in the suspend/resume device methods? The other device methods, attach, probe, shutdown already do this.

--HPS

[markj]

In D10414#216016, @hselasky wrote:

Can this issue also be fixed by dropping / picking up giant in the suspend/resume device methods? The other device methods, attach, probe, shutdown already do this.

Presumably yes, but I'm not familiar with the newbus locking protocol. It seemed preferable to change the sleep code since it must drop Giant anyway to be correct, and I suspect that the existing methods drop Giant just to cover up this problem.

I hadn't noticed that the other methods in linux_pci.c drop Giant. I'll try removing those DROP/PICKUP_GIANT calls and see if any problems arise with this patch applied.

[markj]

Fix linux_wait_for_common() as well.

[markj]

I didn't observe any problems after removing the DROP/PICKUP_GIANT calls in the linuxkpi pci device methods, so I'm not sure what problem they solve.