[Bug 225450] 11.1-* panics on AMD Opteron 2k due to EARLY_AP_STARTUP

[bugzilla-noreply at freebsd.org]

https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=225450

--- Comment #11 from John Baldwin <jhb at FreeBSD.org> ---
Hmm, I don't know why the previous simple lock didn't help.  One other possible
thing to try is placing 'while (1);' infinite loop in the init_secondary_tail()
function in sys/x86/x86/mp_x86.c and moving it around in the function to narrow
down when the APs are triggering the double fault (which is a stack overflow). 
If you put the while(1) before the smp_cpus++; the failure mode you should see
if the AP doesn't fault is a 'panic AP #x failed to start'.  After the
smp_cpus++ line you should at least no longer get the double fault panic if you
haven't hit the double fault yet.

Another thought is that it might be there is a missing MFC in 11 related to one
or more kthreads starting too early.  You could perhaps build a kernel with:

options KTR_COMPILE=KTR_PROC
options KTR_MASK=KTR_PROC
options KTR_VERBOSE

And see what messages are logged before the crash (to see if the APs are
starting to run other kthreads besides the idle thread).

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