[Bug 239351] [panic] spin lock held too long under heavy load (net/wireguard affected)

[bugzilla-noreply at freebsd.org]

https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=239351

Mark Johnston <markj at FreeBSD.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Assignee|net at FreeBSD.org             |bugs at FreeBSD.org

--- Comment #13 from Mark Johnston <markj at FreeBSD.org> ---
These problems don't appear to be network-related.

I'm not sure how the page daemon got stuck waiting for its TLB shootdown to
complete.  clang unfortunately optimizes away the variable which would show
which CPU(s) we were waiting for.

For the VOP_UNSET_TEXT panic, please try applying the attached patch; it should
give more debugging info when the problem happens again.  I think you can drop
the patch I supplied in comment 5.

Please also try applying the patch in https://reviews.freebsd.org/D20327 .  It
should allow you to boot without having to set hw.pci.mcfg=0.

Finally, are you willing to share the vmcores with me so that I can take a
closer look?

-- 
You are receiving this mail because:
You are the assignee for the bug.