On Mon, 29 Jan 2024 at 15:22:31 +0100, Vincent Lefevre wrote:
> However, in my case,
> tracker-extract (gst-plugin-scan) was yielding a crash in nouveau
>
> 2023-12-19T03:10:02.176832+01:00 qaa tracker-extract-3[1967]:
> nvc0_screen_create:1077 - Error allocating PGRAPH context for M2MF: -16
> 2023-12-19T03:10:02.177550+01:00 qaa kernel: ------------[ cut here
> ]------------
> 2023-12-19T03:10:02.177568+01:00 qaa kernel: nouveau 0000:01:00.0: timeout
> [...]
> 2023-12-19T03:10:02.177710+01:00 qaa kernel: CPU: 3 PID: 1967 Comm:
> gst-plugin-scan Tainted: G W 6.5.0-5-amd64 #1 Debian 6.5.13-1
That's certainly a kernel/driver bug. In principle it should not be
possible for Tracker to cause a crash in kernel code (was this an OOPS
or a BUG message or what?), even if it wanted to do this intentionally.
> and gst-plugin-scan even survived SIGKILL:
>
> [...]
> 2023-12-19T03:13:06.240298+01:00 qaa systemd[2079]:
> tracker-extract-3.service: State 'final-sigterm' timed out. Killing.
> 2023-12-19T03:13:06.240435+01:00 qaa systemd[2079]:
> tracker-extract-3.service: Killing process 2150 (gst-plugin-scan) with signal
> SIGKILL.
> 2023-12-19T03:13:40.973708+01:00 qaa kernel: INFO: task gst-plugin-scan:2150
> blocked for more than 120 seconds.
If it has got sufficiently wedged in kernel code that even SIGKILL won't
delete the process, then there isn't going to be anything further that
either Tracker or systemd can do to resolve that.
smcv
