Game: the crash report's graphics card has no driver, and no renderer when the crash is during the start-up shader preparation #1263
Labels
No labels
area/accessibility
area/audio
area/input
area/install
area/menu
area/renderer
area/saves
area/tooling
area/upscaling
beanstalk: approved
beanstalk: failed
beanstalk: needs approval
beanstalk: running
beanstalk: tested
ci: approved
good first issue
gpu/amd
gpu/igpu
gpu/intel
gpu/nvidia
help wanted
kind
bug
kind
chore
kind
docs
kind
feature
kind
performance
kind
question
needs measurements
platform/linux
platform/macos
platform/steam-deck
platform/windows
preview
priority
critical
priority
high
priority
low
regression
risk
high
risk
low
risk
medium
status
blocked
status
confirmed
status
duplicate
status
invalid
status
needs-info
status
wontfix
ci: approved
No milestone
No project
No assignees
2 participants
Notifications
Due date
No due date set.
Dependencies
No dependencies set
Reference
MagicBeanDev/Longshot#1263
Loading…
Reference in a new issue
No description provided.
Delete branch "%!s()"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
Longshot version
main at
17394b7111(the merge of #983). The files involved are unchanged on main2547a8f3dd.Channel
Built from source
What kind of problem
Something else
What happened
Two fields of the crash report that Report automatically (and Send crash reports automatically) posts are missing what the description and the docs say they hold.
gpuhas no driver.build_reportputsf->gpuin the JSON (put_field(json, ",\n ", "gpu", f->gpu, HOS_CRASH_MAX_GPU, s, devices, nd);vk: <name>, Vulkan ...line. The driver (f->driver, fromvk: driver:) is joined to it only for the manual report (line 1059). Thereport: gpuline, which the renderer writes for this purpose (gpu/vk/vk_main.cppline 1662) and which carries the name and the driver, is never read:hos_crash_log_linetakesreport: os,cpuandrenderer(if ((at = strstr(line, "] report: os "))) hos_crash_take(f->os, sizeof f->os, at + 13);AMD Radeon RX 7800 XT, RADV 26.0).rendererisnot knownfor a crash during the start-up shader preparation.report: renderer Vulkanis logged only after the back end'sstart()returns (host/rx_backend.cline 392), and the preparation runs insidestart()(started_up()ingpu/vk/vk_main.cpp).How I know:
vk: NVIDIA GeForce RTX 3090, Vulkan 1.4, ...; mesh shaders yes,vk: driver: NVIDIA 580.95.05andreport: gpu NVIDIA GeForce RTX 3090, NVIDIA 580.95.05, posts"gpu": "NVIDIA GeForce RTX 3090".NVIDIA GeForce RTX 3060 Laptop GPUwith no driver, mapPreparing shadersand renderernot known. The earlier ones (from the Python reporter, for example #81 (comment)) have..., NVIDIA 620.70.06, so the same card is now two differentgpuvalues in Prometheus. The log folded into that comment hasvk: driver: NVIDIA 620.70.06andreport: gpu NVIDIA GeForce RTX 3060 Laptop GPU, NVIDIA 620.70.06, and noreport: rendererline.tests/test_crash_dialog.pychecks the posted field names (line 1003) but not the value ofgpu.Why it matters: the driver is the first thing a GPU crash is split by, issue 81's plan asks for it, the bot's table says "Graphics card and driver", and crashes in the start stage are a common report.
Where in the game
The crash window: Report automatically, or the launcher's Send crash reports automatically.
How often
Every time
Graphics card and driver version
Any. Seen on an RTX 3060 Laptop GPU (NVIDIA 620.70.06) in the live comment, and reproduced with a test log for an RTX 3090 (NVIDIA 580.95.05). Nothing here depends on the card.
CPU, RAM and operating system
Linux; the code is the same on Windows.
Changes from the defaults
None.
longshot_log.txt
Not attached: read from the code, from the existing game-free test helper, and from the public comment above.
Screenshots or video
Not applicable.
What would fix it
report: gpuinhos_crash_log_lineinto the GPU field (fall back to the device name plusvk: driver:when it is missing), and assert the postedgpuin the send test oftests/test_crash_dialog.py.report: rendererwhen the back end is chosen, beforestart(), so a start-stage crash has one.driverfield, as the performance summary does, so the relay can cut it to major.minor and class it the same way for both streams (the relay ignores fields it does not know, so the game can ship first).Before you send it
Possibly related, for context: