On Tue, Jun 18, 2019 at 09:12:59AM +0300, Mike Rapoport wrote:
> On Mon, Jun 17, 2019 at 05:36:30PM +0100, Will Deacon wrote:
> > On Mon, Jun 17, 2019 at 06:12:52PM +0300, Mike Rapoport wrote:
> > > Andrew, can you please add the patch below as an incremental fix?
> > >
> > > With this the arm64::pgd_alloc() should be in the right shape.
> > >
> > >
> > > From 1c1ef0bc04c655689c6c527bd03b140251399d87 Mon Sep 17 00:00:00 2001
> > > From: Mike Rapoport <rppt@xxxxxxxxxxxxx>
> > > Date: Mon, 17 Jun 2019 17:37:43 +0300
> > > Subject: [PATCH] arm64/mm: don't initialize pgd_cache twice
> > >
> > > When PGD_SIZE != PAGE_SIZE, arm64 uses kmem_cache for allocation of PGD
> > > memory. That cache was initialized twice: first through
> > > pgtable_cache_init() alias and then as an override for weak
> > > pgd_cache_init().
> > >
> > > After enabling accounting for the PGD memory, this created a confusion for
> > > memcg and slub sysfs code which resulted in the following errors:
> > >
> > > [ 90.608597] kobject_add_internal failed for pgd_cache(13:init.scope) (error: -2 parent: cgroup)
> > > [ 90.678007] kobject_add_internal failed for pgd_cache(13:init.scope) (error: -2 parent: cgroup)
> > > [ 90.713260] kobject_add_internal failed for pgd_cache(21:systemd-tmpfiles-setup.service) (error: -2 parent: cgroup)
> > >
> > > Removing the alias from pgtable_cache_init() and keeping the only pgd_cache
> > > initialization in pgd_cache_init() resolves the problem and allows
> > > accounting of PGD memory.
> > >
> > > Reported-by: Qian Cai <cai@xxxxxx>
> > > Signed-off-by: Mike Rapoport <rppt@xxxxxxxxxxxxx>
> > > ---
> > > arch/arm64/include/asm/pgtable.h | 3 +--
> > > arch/arm64/mm/pgd.c | 5 +----
> > > 2 files changed, 2 insertions(+), 6 deletions(-)
> >
> > Looks like this actually fixes caa841360134 ("x86/mm: Initialize PGD cache
> > during mm initialization") due to an unlucky naming conflict!
> >
> > In which case, I'd actually prefer to take this fix asap via the arm64
> > tree. Is that ok?
>
> I suppose so, it just won't apply as is. Would you like a patch against the
> current upstream?
Yes, please. I'm assuming it's a straightforward change (please shout if it
isn't).
Will
