Re: [OE-core] [PATCH v2] host_dmesg_logger: dmesg diff on BuildStarted and BuildCompleted

[Alexandre Belloni <alexandre.belloni@bootlin.com>]

On 24/06/2025 08:59:18+0100, Richard Purdie wrote:
> However there is an issue here:
> 
> https://autobuilder.yoctoproject.org/valkyrie/#/builders/43/builds/1900
> 
> stdio: WARNING: +[Mon Jun 23 23:24:20 2025] rustc[2793145]: segfault at 7d85a57fef60 ip 00007d85ace75881 sp 00007d85a57fef60 error 6 in librustc_driver-2c391d96288416bb.so[7d85a72c0000+6246000] likely on CPU 23 (core 53, socket 0)
> stdio: WARNING: +[Mon Jun 23 23:24:20 2025] Code: b5 13 03 48 89 df e8 be af 44 fa 66 2e 0f 1f 84 00 00 00 00 00 0f 1f 40 00 55 41 57 41 56 41 55 41 54 53 48 81 ec b8 00 00 00 <48> 89 34 24 49 89 ff 4c 8b 2f 49 83 fd 02 0f 84 dc 00 00 00 49 83
> stdio: WARNING: +[Mon Jun 23 23:24:20 2025] rustc[2793415]: segfault at 761e429fee20 ip 0000761e49d83872 sp 0000761e429fee10 error 6 in librustc_driver-2c391d96288416bb.so[761e444c0000+6246000] likely on CPU 65 (core 21, socket 0)
> stdio: WARNING: +[Mon Jun 23 23:24:20 2025] Code: 84 00 00 00 00 00 66 90 55 41 57 41 56 41 55 41 54 53 48 81 ec b8 06 00 00 48 8b 9e b8 00 00 00 80 4e 64 10 49 89 d4 49 89 fe <48> 89 74 24 10 0f b6 86 a8 00 00 00 83 f8 22 74 1e 83 f8 23 0f 85
> stdio: WARNING: +[Mon Jun 23 23:24:20 2025] rustc[2792864]: segfault at 7336b89fef50 ip 00007336bf403241 sp 00007336b89fef40 error 6 in librustc_driver-2c391d96288416bb.so[7336ba4c0000+6246000] likely on CPU 91 (core 41, socket 0)
> stdio: WARNING: +[Mon Jun 23 23:24:20 2025] Code: 48 8d 7c 24 40 e8 8f 9d ff ff 48 89 df e8 f7 d5 0b fb ff 15 49 98 d9 03 90 55 41 57 41 56 41 55 41 54 53 48 81 ec 18 02 00 00 <4c> 89 44 24 10 48 89 cb 89 94 24 b0 00 00 00 49 89 f7 48 89 7c 24
> stdio: WARNING: +[Mon Jun 23 23:24:21 2025] rustc[2792945]: segfault at 7b4c847fefe0 ip 00007b4c8b003241 sp 00007b4c847fefd0 error 6 in librustc_driver-2c391d96288416bb.so[7b4c860c0000+6246000] likely on CPU 46 (core 44, socket 0)
> stdio: WARNING: +[Mon Jun 23 23:24:21 2025] Code: 48 8d 7c 24 40 e8 8f 9d ff ff 48 89 df e8 f7 d5 0b fb ff 15 49 98 d9 03 90 55 41 57 41 56 41 55 41 54 53 48 81 ec 18 02 00 00 <4c> 89 44 24 10 48 89 cb 89 94 24 b0 00 00 00 49 89 f7 48 89 7c 24
> stdio: WARNING: +[Mon Jun 23 23:24:24 2025] rustc[2792678]: segfault at 7312ff7fff80 ip 000073130ad372c9 sp 00007312ff7fff80 error 6 in librustc_driver-2c391d96288416bb.so[7313058c0000+6246000] likely on CPU 64 (core 20, socket 0)
> stdio: WARNING: +[Mon Jun 23 23:24:24 2025] Code: 00 00 48 89 b4 24 c8 00 00 00 48 39 ca 0f 85 97 06 00 00 49 89 d6 49 89 f7 48 89 fb 0f b6 86 f5 00 00 00 0f b6 8e f7 00 00 00 <48> 89 14 24 88 4c 24 08 88 44 24 09 48 89 e6 4c 89 ff e8 e0 d1 fe
> 
> The question is therefore going to be how to filter this down to
> something useful, and/or how to save the data so we can work on
> reducing these issues over time?
> 
> The data is useful to have, and over time we should look for and fix
> these segfaults and so on but we can't have this level of warnings on
> the builds.

Seeing the actual output, I don't think having a human readable date is a good
idea as all the other dates we use are in UNIX timestamp format.

Maybe we could use -l to restrict the levels, something like dmesg --level=err+
or dmesg --level=warn+ I'm pretty sure this is going to filter the tap interface
message. The hrtimer message is a pr_warn_once so it would appear only after
reboot, I guess the build was juste after maintenance.

Or we could use --facility and filter out all the kernel builderssages as we are
probably more interested in what is happening in user space. maybe using -x
would be a good idea and we could then do this filtering and make it
configurable using a variable in the class.

Having a way to more precisely filter messages would be great but will also
require maintaining the list of filters.



-- 
Alexandre Belloni, co-owner and COO, Bootlin
Embedded Linux and Kernel engineering
https://bootlin.com