Well it took a while, but finally got through 15 rounds of git bisect. The first bad commit is de59b69932e6.
For each of these builds, I first copied a config from 6.6.15.
And I applied a small patch [1] for module compression according
to this bug report[2].
[1] https://lore.kernel.org/all/3d34a965-ab9c-d549-0c63-c717ab5d2edc@tweek.dk/
[2] https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=1050582
I ran rocminfo a looked for either my GPU or a crash. When
rocminfo worked, I also ran librocrand1-test, the basic test only,
and captured a dmesg.
I'm using rocminfo 5.7.1-3. Once again host is bookworm with a
backported kernel (6.9.7+bpo-amd64) and guest is unstable with
various kernels listed below.
Here's a link[3] to a tight range of commits towards the end of the process.
[3] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/log/?qt=range&q=fa1f1cc09d58
And a link[4] to the first bad commit.
| step | result | version |
| ---:| ---:| --- |
| 0 | good | v6.6.15 |
| 1 | good | v6.6.0-14 |
| 3 | good | v6.6.0-05843-g89ed67ef126c-16 |
| 6 | good | 6.6.0-rc2-00675-g614351f41e8c-dirty |
| 7 | good | 6.5.0-rc7-01718-g5710c6c0bda3-dirty |
| 10 | good | 6.5.0-rc7-01763-gd849434977a8-dirty |
| 13 | good | 6.5.0-rc7-01768-gcd956e753128-dirty |
| 16 | good | 6.5.0-rc7-01769-g29495d81457a-dirty |
| 15 | bad | 6.5.0-rc7-01770-gde59b69932e6-dirty |
| 14 | bad | 6.5.0-rc7-01771-g41801c6b3042-dirty |
| 12 | bad | 6.5.0-rc7-01774-gfa1f1cc09d58-dirty |
| 11 | bad | 6.5.0-rc7-01786-g3a41a3982b1e-dirty |
| 9 | bad | 6.5.0-rc7-01809-g0021d70a0654-dirty |
| 8 | bad | 6.5.0-rc7-01900-gb3e73b5a8f8a-dirty |
| 5 | bad | 6.6.0-rc2-01403-g53dd920c1f47-dirty |
| 4 | bad | v6.6.0-08649-g7d461b291e65-17 |
| 2 | bad | v6.6.0-111004-gedd8e84ae951-15 |
| 0 | bad | v6.7.12 |
Side note, I bought an nvme gen5 ssd.
This got builds from 40 minutes on sata down to 10 minutes.
Brian
-- Brian DeRocher