[Bug 286018] kern.smp.disabled=1 allows for interesting crashes with i915_kms and possibly more

https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=286018

            Bug ID: 286018
           Summary: kern.smp.disabled=1 allows for interesting crashes
                    with i915_kms and possibly more
           Product: Base System
           Version: 14.2-RELEASE
          Hardware: Any
                OS: Any
            Status: New
          Severity: Affects Only Me
          Priority: ---
         Component: kern
          Assignee: bugs@FreeBSD.org
          Reporter: h@hartzer.sh

While investigating this bug,
https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=285867 I found that on a
Thinkpad T400 I could only boot the system with kern.disabled.smp=1, as of 14.2
(14.1 was fine.)

Running the system with kern.smp.disabled=1, I had a couple of unusual crashes.
I wasn't sure if it was the fault of 14.2, kern.disabled.smp=1, or both.

I pulled the SSD and booted it on a Thinkpad R500. Very similar hardware,
except this one boots fine with 14.2-RELEASE with SMP enabled.

On the Thinkpad T400 with SMP disabled (because it won't boot with SMP enabled)
I encountered crashes 100% of the time when blanking the display by running
`swaymsg "output * power off"`. I also had two crashes with just a few minutes
of video playback.

On the R500 with SMP enabled (because it boots fine either way) I was able to
blank the display without incident. I also played video without any crashes or
oddities.

I then disabled SMP on the R500. It crashed immediately after running `swaymsg
"output * power off"`. Thus, I feel there is an obvious connection to SMP being
disabled and these crashes, or at least one of them. Both systems are using the
i915_kms video driver.

I don't know if 14.1 also crashes in this manner if SMP is disabled.

Hopefully others can chime in with other testing here.

I do want to note that kern.smp.disabled=1 appeared to result in a much less
responsive system. I understand that obviously, the system has half the
available CPU power. But even console interactivity in neovim appeared to be
much less, under circumstances that should make no difference. This may not be
related at all -- perhaps the scheduler is less than ideal without additional
cores?

I don't have any single core systems to test if disabling SMP is identical to
running on only one core, unfortunately.

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