https://bugzilla.kernel.org/show_bug.cgi?id=221089
Bug ID: 221089
Summary: amdgpu: Regression in Linux 7.0 with RX-580: BUG:
unable to handle page fault for address:
fffffffffffffff1
Product: Drivers
Version: 2.5
Hardware: All
OS: Linux
Status: NEW
Severity: normal
Priority: P3
Component: Video(DRI - non Intel)
Assignee: [email protected]
Reporter: [email protected]
Regression: Yes
Bisected ba448f9ed62cf5a89603a738e6de91fc6c42ab35
commit-id:
With my particular setup commit ba448f9ed62cf5a89603a738e6de91fc6c42ab35 causes
the plasma-login-manager 6.6 Beta 2 (a.k.a. version 6.5.91) to only show a
black screen with mouse cursor stuck in the middle of the screen (default
position).
After booting into a known working kernel, the journald of the failed boot
indicates that the amdgpu driver segfaulted after the login manager started.
I'll append a kernel stack trace to this bug report in a moment.
As part of bisecting and then digging into this I also tried a non-RT kernel as
well as manually setting `gpu_sched.sched_policy=1` which should correspond to
the deadline GPU scheduler instead of the `gpu_sched.sched_policy=0` that I
usually set (and which, I believe, set the RR scheduler). However the only fix
I found was to revert the commit, which is what I'm now running on top of
current git master (commit 7449f86bafcdb588422bb14a4babfd285e723670).
All crashes seem very similar with the only difference being that the one
non-RT build I tested lacked the `? dc_stream_get_scanoutpos+0x116/0x140
[amdgpu]` line but otherwise they all look the same to my pleb eyes.
I tried CC'ing the relevant people but seems their AMD addresses are not known
to the kernel infrastructure
--
You may reply to this email to add a comment.
You are receiving this mail because:
You are watching the assignee of the bug.
