Open Saviq opened 3 years ago
When reserving a system with
job_queue: rpi3-001
provision_data:
url: http://cdimage.ubuntu.com/releases/focal/release/ubuntu-20.04.1-preinstalled-server-armhf+raspi.img.xz
reserve_data:
ssh_keys:
- lp:${YOU}
I get a Pi3 without KMS support enabled. Is that expected?
And when setting dtoverlay=vc4-fkms-v3d
in /boot/firmware/usercfg.txt
and rebooting the Pi is constantly spewing errors to dmesg and doesn't bring SSH back up?!
It is easier to reproduce #1802 and possible that the fix is common
dmesg when the crash happens:
[ 1940.430288] [drm:vc4_bo_create [vc4]] *ERROR* Failed to allocate from CMA:
[ 1940.441565] vc4-drm soc:gpu: [drm] V3D: 99840kb BOs (21)
[ 1940.441585] vc4-drm soc:gpu: [drm] V3D shader: 24kb BOs (6)
[ 1940.441601] vc4-drm soc:gpu: [drm] dumb: 4068kb BOs (2)
[ 1940.441617] vc4-drm soc:gpu: [drm] binner: 16384kb BOs (1)
[ 1940.441632] vc4-drm soc:gpu: [drm] kernel BO cache: 12kb BOs (2)
[ 1940.441649] vc4-drm soc:gpu: [drm] total purged BO: 640kb BOs (4)
We're exceeding the 128MB CMA region, and something is not handling out-of-GPU-memory cleanly.
I can confirm this still happens, and we need to talk what's next for bypass.
With this testflinger host:
Running this (with either
/dev
or/rc
PPAs):Will run forever. But take
MIR_SERVER_BYPASS
out and you'll seeglmark2-es-wayland
crashing real soon:Crashes quite early:
And the server log: