Re: commit 1c29a32ce65f4cd0f1c causes Bad rss-counter state and firefox-esr crash in linux-next-20240613

[Bert Karwatzki <spasswolf@web.de>]

Am Freitag, dem 12.07.2024 um 11:38 -0400 schrieb Liam R. Howlett:
> * Bert Karwatzki <spasswolf@web.de> [240712 08:18]:
> > I did some experiments on the rss counter bug. The next patch is made for linux-next-20240613
> > with commit 1c29a32ce65f4cd0f1c0f9 reverted. Then I simply inlined the code of do_vmi_unmap()
> > and do_vmi_align_munmap() into mmap_region(). This version of the code works fine and does not
> > show the rss counter bug.
>
> Are you still working with v1 of this patch set?
>
> I root-caused the rss counter issue and seg fault to the fact that next
> or prev may be expanded and I was using the new start/end on munmap() in
> the completion.  This was fixed in subsequent patches.
>
> I've sent v4 recently, but will have to a v5 to back off the removal of
> arch_unmap() for PPC.
>
> ...
>
> Thanks,
> Liam

Sorry, that was a mistake when using git send-email while working on another bug
I accidently sent this old mail. I actually have tested the v3 and v4 patchsets
on linux-next up to 2024711 with no error. I havent't tested v4 on linux-
20240712, yet, because next-20240712 has a deadlock issue:
https://lore.kernel.org/all/20240712080414.GA47643@google.com/
(while sending a mail to this issue, I accidently send the old mail, too)

Bert Karwatzki