Re: [PATCH v3] erofs: cap LZMA stream pool size

[Guenter Roeck]

On 8/11/26 12:27, Geert Uytterhoeven wrote:
> On Mon, 3 Aug 2026 at 11:12, Uwe Kleine-König
> <u.kleine-koenig@xxxxxxxxxxxx> wrote:
> > On Mon, Aug 03, 2026 at 10:09:06AM +0200, Geert Uytterhoeven wrote:
> > > On Tue, 28 Jul 2026 at 08:55, SJ Park <sj@xxxxxxxxxx> wrote:
> > > > I don't have a m68k testfarm, either.  I'm doing crossbuild [1].  Hopefully you
> > > > could also reproduce it.
> >
> > FTR, I don't have m68k hardware either, but as this is a build time
> > failure installing the Debian package gcc-m68k-linux-gnu and then doing
> > builds using
> >
> >          make ARCH=m68k CROSS_COMPILE=m68k-linux-gnu-
> >
> > is good enough to reproduce the failure.
> >
> > > I think you can reproduce some of the issues on most architectures
> > > with a non-SMP kernel, cfr. my reply[1] to Uwe's proposed fix.
> > > Unfortunately the Kconfig warning only shows up on architectures that
> > > don't define NR_CPUS (because they do not support SMP) at all,
> > > while ending up with zero streams is probably a runtime failure.
> > >
> > > [1] https://lore.kernel.org/CAMuHMdVGKX=3cfOrLWp1uSHX76xf1-=C0+GVCXkA8JTqNPg-ew@xxxxxxxxxxxxxx
> >
> > I guess EROFS_FS_ZIP_LZMA_DEFAULT_MAX_STREAMS should depend on SMP and
> > it's use should be conditionalized using something like
> >
> > diff --git a/fs/erofs/decompressor_lzma.c b/fs/erofs/decompressor_lzma.c
> > index 6b0cdb446c6a..c00e94f8206f 100644
> > --- a/fs/erofs/decompressor_lzma.c
> > +++ b/fs/erofs/decompressor_lzma.c
> > @@ -50,9 +50,13 @@ static int __init z_erofs_lzma_init(void)
> >          unsigned int i;
> >
> >          /* by default, use # of possible CPUs instead */
> > -       if (!z_erofs_lzma_nstrms)
> > -               z_erofs_lzma_nstrms = min_t(unsigned int, num_possible_cpus(),
> > -                               CONFIG_EROFS_FS_ZIP_LZMA_DEFAULT_MAX_STREAMS);
> > +       if (!z_erofs_lzma_nstrms) {
> > +               if (IS_ENABLED(CONFIG_SMP))
> > +                       z_erofs_lzma_nstrms = min_t(unsigned int, num_possible_cpus(),
> > +                                       CONFIG_EROFS_FS_ZIP_LZMA_DEFAULT_MAX_STREAMS);
> > +               else
> > +                       z_erofs_lzma_nstrms = 1;
> > +       }
> >
> >          for (i = 0; i < z_erofs_lzma_nstrms; ++i) {
> >                  struct z_erofs_lzma *strm = kzalloc_obj(*strm);
>
> LGTM.
>
> > I played a bit with the definition of
> > EROFS_FS_ZIP_LZMA_DEFAULT_MAX_STREAMS in fs/erofs/Kconfig, but didn't
> > find a way to define it that both uses NR_CPUS and doesn't generate a
> > warning with ARCH=m68k.
>
> Any progress with fixing this for v7.2?
>
> NR_CPUS depends on SMP on most architectures, so you get e.g.
> with ARCH=arm allmodconfig + CONFIG_SMP=n:
>
>      CONFIG_EROFS_FS_ZIP_LZMA_DEFAULT_MAX_STREAMS=0
>
> X86 defines NR_CPUS unconditionally, so there it works as expected.
> Thanks!
FWIW, m68k:allmodconfig now fails to build due to this problem all the
way back to v6.6.y.

Guenter