Re: [PATCH v8 14/21] mm/mmap: Avoid zeroing vma tree in mmap_region()

From: Lorenzo Stoakes
Date: Tue Oct 01 2024 - 05:00:27 EST


On Tue, Oct 01, 2024 at 10:55:50AM GMT, Bert Karwatzki wrote:
> Am Dienstag, dem 01.10.2024 um 09:49 +0100 schrieb Lorenzo Stoakes:
> > On Tue, Oct 01, 2024 at 10:38:35AM GMT, Bert Karwatzki wrote:
> > > Am Dienstag, dem 01.10.2024 um 09:02 +0100 schrieb Lorenzo Stoakes:
> > > > On Tue, Oct 01, 2024 at 04:34:00AM GMT, Bert Karwatzki wrote:
> > > > > I just noticed (via a bisect between v6.11 and v6.12-rc1) that this patch
> > > > > (commit f8d112a4e657 in linux-next tree) leads to a severe memory corruption
> > > > > error under these (rather rare) circumstances:
> > > > > 1. Start a 32bit windows game via steam (which uses proton, steam's version of wine)
> > > > > 2. When starting the game you the proton version used has to be updated
> > > >
> > > > Yikes. Thanks for the report, very very much appreciated. Will look into
> > > > this as Liam is out until next week.
> > > >
> > > > How repro is this? Is it consistent?
> > >
> > > Reproducability is 100%, only the method is weird, you have to switch to an
> > > older version of proton in the steam settings of the game, start the game and
> > > then switch back to the new version and start the game again.
> > > It might also be possible using standard wine and repeatedly upgrading and
> > > downgrading wine and (I have not tried this, yet ...)
> > >
> >
> > OK that's good.
> >
> > Actually a quick one if you have a sec - could you try the same thing with tip
> > of Linus's tree?
> >
> > This will help eliminate any other possible cause.
> >
> > Thanks!
> >
> > >
> I first noticed the bug unsing linux-next-20240926, it's also present in v6.12-
> rc1 (hence the bisection), but I can try linux-next or linux-torvald master,
> too.

Next can have odd resolutions and 26th sep might not include all fixes, could
you check against Linus's tree just to rule out anything else?

Very likely it'll just repro again but be good to be sure before going too deep
on this, thanks!

>
> Bert Karwatzki