Re: AMD Ryzen 5 hangs on 13.x, fine on 12.4

On 4/19/23 08:00, Philip Homburg wrote:
> Hi,
> 
> I have a problem that my Ryzen 5 systems hangs on 13.x when running
> poudriere but works fine on 12.4. See
> https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=267782
> 
> Does anyone have any pointers how I would go debug this issue?
> 
> The problems is that usually it just hangs.

   Poudriere should have written logs of what was being worked on during 
the hang. Any other poudriere configurations may be helpful. Which ports 
are being built during the problem? How much RAM, swap partition space, 
and which Ryzen 5? Anything else special about the install like 
zfs+dedup or /etc/sysctl.conf (and tried without those tweaks) or 
/boot/loader.conf things adjusted?
   That PR made mention of BIOS updates. In my experience, there is a 
lot a BIOS update addresses that doesn't make it into the final notes 
average users see but has been done. I've seen many updates be important 
for stability and performance to where I apply them unless I have 
documentation telling me why I shouldn't, especially in recent years 
with AMD though I have been away from computer repair for a couple 
years. BIOS updates do have risk  on systems that don't have a backup 
BIOS or way to reflash it in case of failure. It may also turn out not 
at all related to the issue.
   If you can't go with what you last get from the terminal as it runs, 
from its own logs, and don't notice anything relevant like high RAM use, 
dmesg output pointing at anything drawing your attention, etc. then 
maybe kernel debugging type of steps are in order?
   I find it easy to lock up my i7-3820 with 32GB RAM with 4 or 8 
poudriere jobs running if I have no swap file. Adding swap goes from 
lockups to normally killing processes but sometimes hits more important 
ones bringing down X sessions in an unrecoverable way requiring a 
reboot. I do runs like `poudriere -J2` with multiple threads per job 
with much better success as long as other things like Firefox haven't 
hogged too much RAM.