On Sat, 14 Sep 2024 08:03:29 +0530, James young wrote:

I've hit an odd CIFS corruption issue, in certain contexts, on ARM, dependent on the Linux kernel version. I have a repeatable test. It looks like: * Debian 11, bullseye, kernel 6.1: ok * Debian 12, bookworm, kernel 6.6: impacted * v6.10, that I compiled: ok

I think I need to be *far* more specific before I go to the main kernel mailing list or report a regression.

I'm planning to test the mainline kernel, though I haven't noted significant differences from what Debian distributed. What's most useful here? * bisecting what broke between 6.1 and 6.6 - about 88k commits * bisecting the fix between 6.6 and 6.10 - about 66k commits * survey the major branches - 6.1, 6.2, ... 6.10 - my current WIP Each iteration can take hours, and these aren't fully unattended, so, I can't be too extensive, here.

You can reduce number of commits by looking at fs/smb for a start. Though depending on the repro you can narrow down further. Best bet would be to trace the repro on 6.6 to find problematic code. Thanks, Siddh