Hello Debian PowerPC developers and users,
I am writing to report a severe issue with the graphics drivers on a
PowerMac G5 Quad, which appears to be related to endianness bugs in
the AMD GPU driver stack. I have been unable to get a stable,
hardware-accelerated desktop with either the radeon or the generic
modesetting driver.
System Summary:
*
Machine: Apple PowerMac G5 Quad (PowerMac11,2)
*
OS: Debian Unstable (Trixie) for PowerPC64
*
Kernel: 6.16.3+deb14-powerpc64
*
Primary GPU: AMD FirePro V7900 (Cayman)
*
Secondary (inactive) GPU: AMD Radeon X1950 XT (R580+)
The goal is to get a stable, accelerated desktop using the FirePro
V7900. The X1950 XT is disabled at boot using the pci-stub kernel
parameter.
Problem Progression & Symptoms:
*
Default Configuration: The system boots to a black screen.
*
radeon with Glamor (default): The X server crashes immediately with a
Segmentation fault.
*
radeon with EXA: The display initializes but shows completely garbled
graphics. The kernel log reveals continuous GPU lockups and soft
resets.
*
modesetting driver: The X server fails to start, crashing with an
AddScreen/ScreenInit failed error.
*
Working Configuration (No Acceleration): The only way to get a stable
desktop is by forcing the radeon driver with the NoAccel option. This
provides a usable, but extremely slow, software-rendered environment.
This behavior strongly suggests that the entire hardware acceleration
stack for this GPU—from the specific radeon driver to the generic
modesetting driver's use of the kernel's display layer—is affected
by bugs on this big-endian ppc64 hardware.
-------------------------
Hardware Details:
lspci -nn | grep -i vga output:
0000:0a:00.0 VGA compatible controller [0300]: Advanced Micro Devices,
Inc. [AMD/ATI] Cayman PRO GL [FirePro V7900] [1002:6704]
0001:06:00.0 VGA compatible controller [0300]: Advanced Micro Devices,
Inc. [AMD/ATI] R580+ [Radeon X1950 XT] [1002:7244]
-------------------------
Supporting Logs:
1. modesetting Driver Failure: This is the fatal error from
/var/log/Xorg.0.log when trying to use the modesetting driver.
[ 27.226] (II) modesetting: Driver for Modesetting Kernel Drivers:
kms
[ 27.227] (II) modeset(0): using drv /dev/dri/card0
...
[ 27.360] (II) modeset(0): glamor X acceleration enabled on AMD
CAYMAN (DRM 2.50.0 / 6.16.3+deb14-powerpc64, LLVM 19.1.7)
[ 27.360] (II) modeset(0): glamor initialized
...
[ 27.498] (EE)
Fatal server error:
[ 27.498] (EE) AddScreen/ScreenInit failed for driver 0
2. radeon with Glamor Segmentation Fault: From /var/log/Xorg.0.log
when radeon is used without AccelMethod specified.
[ 28.085] (EE) Segmentation fault at address 0xc
[ 28.086] (EE) Caught signal 11 (Segmentation fault). Server
aborting
3. radeon with EXA Kernel GPU Lockup: From dmesg when using Option
"AccelMethod" "exa". The log is flooded with these repeating messages.
[ 22.654960] radeon 0000:0a:00.0: GPU lockup (current fence id
0x0000000000000001 last fence id 0x0000000000000002 on ring 0)
[ 52.399474] radeon 0000:0a:00.0: GPU softreset: 0x00000008
[ 52.417165] radeon 0000:0a:00.0: GPU reset succeeded, trying to
resume
...
[ 86.835489] WARNING: CPU: 2 PID: 782 at
drivers/gpu/drm/radeon/radeon_object.c:62
radeon_ttm_bo_destroy+0x138/0x160 [radeon]
-------------------------
Successful (but unaccelerated) Workaround:
The following configuration in /etc/X11/xorg.conf.d/20-radeon.conf
provides a stable desktop:
Section "Device"
Identifier "FirePro V7900"
Driver "radeon"
BusID "PCI:10:0:0"
Option "NoAccel" "True"
EndSection
I hope this comprehensive report is useful for tracking down these
critical issues. It seems PowerPC users with this class of AMD
hardware are currently unable to get a usable, accelerated desktop. I
am happy to provide further logs or perform any tests that might help
in debugging.
Thank you,
Ramon Collet (R·V·E)