Fix hrtimer_active() in case of cancelllation.

[mav]

Summary

While trying drm-next suspend on Skylake I found it complaining "Timed out waiting for forcewake timers to finish". Investigation shown that hrtimer_active() return false positive after hrtimer_cancel() was used.

My patch fixes that bug while also switching to FreeBSD internal active status. If there is some good reason to use external flag please explain it to me, for which case I have alternative patch.

Test Plan

drm-next code no longer complains about this during suspend.

[• hselasky]

sys/compat/linuxkpi/common/src/linux_hrtimer.c

58 ↗

(On Diff #31666)

Can you upload patch with full context?

What locks are protecting callout_active() ?

[mav]

sys/compat/linuxkpi/common/src/linux_hrtimer.c

58 ↗

(On Diff #31666)

Uploaded.

What do you mean by "what locks"? According to man page it is protected by user lock, in this case -- hrtimer->mtx. But some cases seem also covered by callout internal lock.

[markj]

I can't quite remember why I made the flag external. I might have forgotten about the existence of the callout_deactivate() KPI.

Thanks for fixing this.

[• hselasky]

Looks good. Please add a comment describing the return value of hrtimer_cancel() before commit.

sys/compat/linuxkpi/common/src/linux_hrtimer.c

64 ↗

(On Diff #31669)

Can you add a comment above this function describing the return values.