In D41978#957120, @siderop1_netapp.com wrote:

This looks pretty good to me. FWIW, I have a test queued to verify this fixes the issue we were seeing; should have results by tomorrow (if not earlier).

Thanks!
--ap

This looks pretty good to me. FWIW, I have a test queued to verify this fixes the issue we were seeing; should have results by tomorrow (if not earlier).

Barf. Looks like this was already implemented with 913c07a0499, but I failed to sync past that in my Git repo. I'll revisit if I find any differences worth submitting.
--ap

In D27331#610723, @mjg wrote:

Can you share the hw spec? Most notably cores/threads, cpu model and socket count?

In D27331#610694, @mjg wrote:

Any changes here should come with several benchmarks, showcasing longest time to grab the lock and perhaps some fiddling with min/max for backoff.

update error message based on review feedback

In D25756#572421, @whu wrote:

#ifdef DIAGNOSTIC
Is this only for test purpose? When will this be defined? Thanks.

I am not saying these cannot coexist, just that SBCC doesn't require FreeBSD to maintain its own trace implementations (i.e. all the __sanitizer_cov_trace* routines). Again, I do not know if SBCC can be used for syzkaller; if it cannot, then this whole discussion is moot. But if we can leverage SBCC, it would be worth exploring.
--ap

We is NetApp; we use this internally to gather code coverage metrics as well as input to profile guided optimization of the kernel. It is not public, but I think we can push it upstream if there is interest.

I understand the main objective with this change is to enable syzkaller, but is Clang's coverage sanitizer the only way to implement KCOV for this purpose? I ask because we have (and use) Clang's SBCC to get coverage data out of the kernel. It would be nice to not have two KCOV implementations. :)
--ap