Re: [sparc64] fails to boot, (was: Re: [PATCH memcg v6] net: set proper memcg for net_init hooks allocations)

[Shakeel Butt]

On Mon, Sep 26, 2022 at 10:28 AM Jakub Kicinski <kuba@xxxxxxxxxx> wrote:
>
> On Mon, 26 Sep 2022 16:06:08 +0300 Anatoly Pugachev wrote:
> > On Wed, Sep 21, 2022 at 8:03 PM Michal Koutný <mkoutny@xxxxxxxx> wrote:
> > > On Wed, Sep 21, 2022 at 05:44:56PM +0300, Anatoly Pugachev <matorola@xxxxxxxxx> wrote:
> > > > reverting this patch makes my sparc64 box boot successfully.
> > >
> > > The failed address falls into vmmemmap region (per your boot log
> > > output). It looks like the respective page/folio (of init_net struct) is
> > > unbacked there (and likely folio_test_slab fails dereferencing ->flags).
> > >
> > > Would you mind sharing your kernel's config?
> > > (I'm most curious about CONFIG_SPARSMEM_VMEMMAP, I'm not familiar with
> > > your arch at all though.)
> >
> > mator@ttip:~/dmesg$ zcat config-6.0.0-rc6-00010-gb7f0f527dc3c.gz | grep VMEMMAP
> > CONFIG_SPARSEMEM_VMEMMAP_ENABLE=y
> > CONFIG_SPARSEMEM_VMEMMAP=y
> >
> > I do upload config and boot logs to
> > https://github.com/mator/sparc64-dmesg
> >
> > building a new kernel version/releases as 'make olddefconfig && make -j'
> > current version of booted 6.0.0-rc6 is available as
> > https://github.com/mator/sparc64-dmesg/blob/master/config-6.0.0-rc6-00010-gb7f0f527dc3c.gz
>
> Forgive my uniformed chime-in but Linus seemed happy with the size of
> -rc7 and now I'm worried there won't be an -rc8. AFAICT this is a 6.0
> regression. Vasily, Shakeel, do we have a plan to fix this?

I was actually waiting for Vasily to respond. Anyways, I think the
easiest way to proceed is to revert the commit 1d0403d20f6c ("net: set
proper memcg for net_init hooks allocations"). We can debug the issue
in the next cycle.