Re: port-amd64/60873: amdgpu enabled Kernel crash

["Taylor R Campbell via gnats" <gnats-admin%NetBSD.org@localhost>]

The following reply was made to PR port-amd64/60873; it has been noted by GNATS.

From: Taylor R Campbell <riastradh%NetBSD.org@localhost>
To: gnats-bugs%netbsd.org@localhost
Cc: port-amd64-maintainer%netbsd.org@localhost, gnats-admin%netbsd.org@localhost,
	netbsd-bugs%netbsd.org@localhost, shep971%spectrum.net@localhost
Subject: Re: port-amd64/60873: amdgpu enabled Kernel crash
Date: Thu, 15 Oct 2026 22:25:46 +0000

 > Date: Thu, 15 Oct 2026 15:01:20 -0700
 > From: "Heppler, J. Scott" <shep971%spectrum.net@localhost>
 >=20
 > I'm not sure the hand-held photo of the crash was helpful.
 >=20
 > Here is the hand transcribed dmesg leading up to the crash:
 > "[drm] kiq ring mec 2 pipe 1 q 0
 > panic: fpudna from kernel . ip 0xffffffff807b459b, trapframe 0xffff868827=
 064ba0
 > cpu0: begin traceback
 > vpanic() at netbsd:vpanic+0x171
 > panic() at netbsd:panic+0x3c"
 >=20
 > If there is anything else that would be helpful, I'll try to generate
 > it.
 
 Are there any lines after that?
 
 What's happening is:
 
 1. Something in the kernel is trying to use floating-point/vector
    instructions when it is forbidden (`panic: fpudna').
 2. There is a stack trace showing what part of the kernel was doing
    this.
 3. The stack trace as you shared it stops point where the fpu trap
    handler called panic(), but I want to see what provoked the fpu
    trap handler -- most likely, it's just some small part of the
    amdgpu driver that's missing fpu_kern_enter/leave (we've had a few
    bugs like this before), but we need to find which part of the
    driver that is.
 
 Or, did you get a crash dump, e.g. /var/crash/netbsd.N.core?
 
 If so, you may be able to get more info out with:
 
 $ crash -N /netbsd -M /var/crash/netbsd.N.core
 
 or
 
 $ gdb /netbsd
 (gdb) target kvm /var/crash/netbsd.N.core
 (gdb) bt