Move td_softdep_cleanup() from userret() to ast()

[trasz]

Summary

Move td_softdep_cleanup() from userret() to ast(); it's infrequent
at best. The schedule_cleanup() function already sets TDF_ASTPENDING.

[kib]

sys/kern/subr_trap.c

198 ↗

(On Diff #76814)

Leave the assert there.

[trasz]

sys/kern/subr_trap.c

198 ↗

(On Diff #76814)

Wouldn't it trip at the first possibility due to ast() being called from the syscall enter assembly code some time after userret() returns?

[kib]

sys/kern/subr_trap.c

198 ↗

(On Diff #76814)

I do not understand what are you saying there.

The assert can be triggered if userret() is executed before ast is handled, or if ast() 'forgot' to handle requested su cleanup. I do not believe that the first case is possible.

Or rather, if userret() triggers that, there is a bug in your patch.

[trasz]

sys/kern/subr_trap.c

198 ↗

(On Diff #76814)

Erm, are you sure about it? My understanding is that (eg) amd64_syscall() calls syscallret(), syscallret() calls userret(), and then amd64_syscall() returns; ast() gets only called afterwards.

[kib]

sys/kern/subr_trap.c

198 ↗

(On Diff #76814)

Yes you are right. I would say that this should be fixed, but actually your series of the patches makes it not matter at all.

[mckusick]

This change looks reasonable to me as at least for amd64 every path out of the kernel appears to check TDF_ASTPENDING and calls ast() if set.