fix build with LOCK_PROFILING but without KDTRACE_HOOKS

[kp]

Summary

Sponsored by: Rubicon Communications, LLC ("Netgate")

[kib]

sys/kern/kern_rwlock.c

476

I do not understand this block. LOCK_PROFILING does not depend on KDTRACE_HOOK. While your patch puts LOCK_PROFILING control inside KDTRACE_HOOK block.

[kp]

sys/kern/kern_rwlock.c

476

I think I misunderstood the intent behind doing_lockprof. It's to allow lock profiling to be enabled either unconditionally (with LOCK_PROFILING) or conditionally. (if KDTRACE_HOOKS and LOCKSTAT_PROFILE_ENABLED() returns true).

So the correct thing here would be:

#if defined(KDTRACE_HOOKS)
	uintptr_t state = 0;
#endif
#if defined(KDTRACE_HOOKS) || defined(LOCK_PROFILING)
	int doing_lockprof = 0;
#endif

(And then not have this #ifdef LOCK_PROFILING here, of course)

[kib]

sys/kern/kern_rwlock.c

476

Yes, this sounds more correct.

[kp]

review remarks

[kib]

sys/kern/kern_rwlock.c

465

Merge this block of ifdef with the previous one?

478

Remove the assignment?

962

Same?

sys/kern/kern_sx.c

586

Same, merge?

605

This line can be removed, the same assignment is happen right after the block.

1053

Same

[kp]

Review remarks

We now unconditionally assign 'state = v' (ifdef KDTRACE_HOOKS), but that should
be fine, because we only ever use 'state' if doing_lockprof is true, which is
when it was assigned prior to this change.
Updating D47822: fix build with LOCK_PROFILING but without KDTRACE_HOOKS

[markj]

sys/kern/kern_rwlock.c

478

I don't see how this assignment can be removed. Without we'd always have state = 0.

[kp]

sys/kern/kern_rwlock.c

478

It’ll be set if doing_lockprof == 1, which is the only case we actually use it.
Either because LOCKSTAT_PROFILE_ENABLED or LOCK_PROFILING.

[markj]

sys/kern/kern_rwlock.c

478

I think I misunderstood kib's comment because it applied to an old version of the file, so the comment appears in a misleading place.