Panfrost d00c0000.gpu: Fatal error during GPU init

Hi all,
Can someone running VIM2 with recent mainline linux kernel with panfrost config post output of
dmesg | grep panfrost
here is mine:

kvim2-gentoo ~ # cat /proc/version
Linux version 5.3.1 (root@kvim2-gentoo) (gcc version 9.2.0 (Gentoo 9.2.0 p1)) #2 SMP PREEMPT Fri Sep 27 21:45:31 CEST 2019

kvim2-gentoo ~ # zcat /proc/config.gz | grep -i panfrost
CONFIG_DRM_PANFROST=m

kvim2-gentoo ~ # dmesg | grep panfrost
[ 13.597976] panfrost d00c0000.gpu: clock rate = 666666666
[ 13.598073] panfrost d00c0000.gpu: mali-t820 id 0x820 major 0x1 minor 0x0 status 0x0
[ 13.598080] panfrost d00c0000.gpu: features: 00000000,101e76ff, issues: 00000000,24040400
[ 13.598088] panfrost d00c0000.gpu: Features: L2:0x07110206 Shader:0x00000000 Tiler:0x00000809 Mem:0x1 MMU:0x00002821 AS:0xff JS:0x7
[ 13.598093] panfrost d00c0000.gpu: shader_present=0x7 l2_present=0x1
[ 13.598180] panfrost d00c0000.gpu: GPU Fault 0x000000ad (UNKNOWN) at 0x33c299f2c0d9d62d
[ 13.598185] panfrost d00c0000.gpu: There were multiple GPU faults - some have not been reported
[ 13.605895] panfrost d00c0000.gpu: error powering up gpu
[ 13.622361] panfrost d00c0000.gpu: Fatal error during GPU init
[ 13.632411] panfrost: probe of d00c0000.gpu failed with error -12

clearly the GPU reports some faults.
I am curious to know if people see similar of same faults.

fwiw another box shows same except
[ 14.022634] panfrost d00c0000.gpu: GPU Fault 0x00000088 (UNKNOWN) at 0x000000e7e733c1c0
so I guess Fault and Address are somewhat random

Thanks

Panfrost will not init on S912 boards with a vanilla kernel. Do you have these patches applied?

https://github.com/chewitt/linux/commit/d0a03a1870329ebd76ab6affcfa3f103183ea771 and https://github.com/chewitt/linux/commit/5c46c60ef67f0d417ea29602e21f435b5da19a3e

Some proper fixes for the underlying issues are in process for future kernels.

3 Likes

thanks a lot!
I will apply the patches and let you know how things go.

update
panfrost does now init upon modprobe
and
/sys/devices/platform/soc/d0000000.apb/d00c0000.gpu/driver/module/drivers/platform:panfrost
appears

But the ‘GPU Fault 0x… (UNKNOWN) at 0x…’ is still there upon boot.
Looking at kernel code this may relate to request for irq to not succeed.

Also, Xorg now fails to start, and it logs:
(EE) No devices detected.

I understand this is work in progress.

Thanks for your contribution!

not sure if here is right place to discuss if it is common to all s912

^ I think the last few commits on that branch are the proposed longer-term fixes; I haven’t had much time to test anything recently so I’ve not tested them. Applications and usage will be user/device specific but from a panfrost perspective all S912 hardware is identical. I’ve never used panfrost for desktop tasks (only Kodi) so have no knowledge of other use cases.

1 Like

thanks for all pointers

Going from mainline kernel 5.3.1 to 5.4.0, panfrost does now init upon modprobe but he ‘GPU Fault 0x… (UNKNOWN) at 0x…’ is still there upon boot.

Moreover, as when I applied d0a03a1870329ebd76ab6affcfa3f103183ea771 and commit/5c46c60ef67f0d417ea29602e21f435b5da19a3e to 5.3.1, Xorg unfortunately fails to start …

Note that Xorg has always worked for me.

Of course, now that panfrost issues are ‘fixed’, maybe I need to configure something extra for making X find the drivers, but then: also note I did never configure anything to make Xorg work before.

Any hints?

Keep you informed.

I still have https://github.com/chewitt/linux/commit/f9c67a903da1cdc3e965c331ed83e2bbc131826a in my 5.4 branch. I’ve no idea about Xorg, but Kodi is up and running.

1 Like

I will try that patch.
It’s strange though how a just and only a kernel upgrade stops X11 from starting.
Have a nice day