Re: [PATCH] arm64: increase stack size for KASAN_EXTRA

[Arnd Bergmann]

On Tue, Dec 11, 2018 at 11:22 PM Qian Cai <cai@xxxxxx> wrote:
>
> On Tue, 2018-12-11 at 23:12 +0100, Arnd Bergmann wrote:
> > On Tue, Dec 11, 2018 at 10:59 PM Qian Cai <cai@xxxxxx> wrote:
> > >
> > > On Tue, 2018-12-11 at 22:56 +0100, Arnd Bergmann wrote:
> > > > On Tue, Dec 11, 2018 at 10:52 PM Qian Cai <cai@xxxxxx> wrote:
> > > > > On Tue, 2018-12-11 at 22:43 +0100, Arnd Bergmann wrote:
> > > > > > On Tue, Dec 11, 2018 at 6:18 PM Qian Cai <cai@xxxxxx> wrote:
> > > > >
> > > > > I am not too keen to do the version-check considering some LTS versions
> > > > > could
> > > > > just back-port those patches and the render the version-check
> > > > > incorrectly.
> > > >
> > > > I'm not following what the problem is. Do you mean distro versions gcc
> > > > with the compiler bugfix, or LTS kernel versions?
> > > >
> > >
> > > I mean distro versions of GCC where the version is still 8 but keep back-
> > > porting
> > > tons of patches.
> >
> > Ok, but in that case, checking the version would still be no worse
> > than your current patch, the only difference is that for users of a
> > fixed older gcc, the kernel would use more stack than it needs.
> >
>
> I am thinking about something it is probably best just waiting for those major
> distors to complete upgrading to GCC9 or back-porting those stack reduction
> patches first. Then, it is good time to tie up loose ends for those default
> stack sizes in all combinations.

I was basically trying to make sure we don't forget it when it gets to that.

Another alternative would be to just disable KASAN_EXTRA now
for gcc versions before 9, which essentially means for everyone,
but then we get it back once a working version gets released. As
I understand, this kasan option is actually fairly useless given its
cost, so very few people would miss it.

On a related note, I think we have to turn off asan-stack entirely
on all released clang versions. asan-stack in general is much more
useful than the use-after-scope check, but we clang produces some
very large stack frames with it and we probably can't even work
around it with KASAN_THREAD_SHIFT=2 but would need even
more than that otherwise.

         Arnd