https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=288507

--- Comment #5 from Mark Millard <[email protected]> ---
(In reply to Mark Johnston from comment #4)

For now I certainly have access to the core dump
to look around with. I'll note that I'm not
literate for the data structures involved. So
you may have to be very explicit about the
commands that you want output from.

I'll note that the context here is AMD, not Intel.

But difficulties with timely future testing are likely:

I'll likely lose access to the system for a signficant
time starting sometime in August. Igoring that until
then . . .

Given the number of bulk -Ca runs that got the OOM
activity from sdl2_gpu's doxygen's dot use that did
not panic, doubt repeatability is reasonable to
expect for the type of context that I've been doing.
1 of 3 runs got the failure (so far). And the type of
activity that lead to the example (bulk -Ca runs) takes
days+ for each try.

The bulk -Ca testing that exposed the crash has lead
to my monthly data limit being nearly reached, where
I started the explorations on 2025-Jul-12. (I got my
first ever notice warning of approaching the limit
yesterday.)

May be just direct repeated builds of sdl2_gpu might
work for getting a detected failure more often. I've
no clue how much prior/other parallel build activity
might have contributed to having a failure context.

I've another bulk -Ca in process, with sdl2_gpu
indicated as BROKEN in order to avoid it messing up
what I was originally doing. It could be a couple
more days before that run finishes. Finding out how
long it takes to finish the main-amd64 build with the
kernel.GENERIC-NODEBUG kernel in use and with the
specific type of poudriere-jail-world build
is part of the point [WITH_MALLOC_PRODUCTION and
WITHOUT_LLVM_ASSERTIONS as a type of jail context].

(Even with the kernel.GENERIC-NODEBUG kernel, bulk -Ca
runs based on WITHOUT_MALLOC_PRODUCTION and
WITH_LLVM_ASSERTIONS for the jail take more like 157
hrs instead of over 48 Hrs but likely under 72 hrs
for this hardware and the types of example poudriere
configurations that I've been using. The failure
made it to or out past [2D:01:38:25] as of when it
failed, the last part of the log likely not having
been written to disk.)

Anyway, I'll see about getting more data extractions
from the core file.

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

Reply via email to