Re: [PATCH v1] mm: hwpoison: disable memory error handling on 1GB hugepage

[Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>]

Hi Michal, Mike,

On Mon, Jan 29, 2018 at 10:08:53AM -0800, Mike Kravetz wrote:
> On 01/29/2018 01:54 AM, Michal Hocko wrote:
> > On Mon 29-01-18 06:30:55, Naoya Horiguchi wrote:
> >> My apology, I forgot to CC to the mailing lists.
> >>
> >> On Mon, Jan 29, 2018 at 03:28:03PM +0900, Naoya Horiguchi wrote:
> >>> Recently the following BUG was reported:
> >>>
> >>>     Injecting memory failure for pfn 0x3c0000 at process virtual address 0x7fe300000000
> >>>     Memory failure: 0x3c0000: recovery action for huge page: Recovered
> >>>     BUG: unable to handle kernel paging request at ffff8dfcc0003000
> >>>     IP: gup_pgd_range+0x1f0/0xc20
> >>>     PGD 17ae72067 P4D 17ae72067 PUD 0
> >>>     Oops: 0000 [#1] SMP PTI
> >>>     ...
> >>>     CPU: 3 PID: 5467 Comm: hugetlb_1gb Not tainted 4.15.0-rc8-mm1-abc+ #3
> >>>     Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.9.3-1.fc25 04/01/2014
> >>>
> >>> You can easily reproduce this by calling madvise(MADV_HWPOISON) twice on
> >>> a 1GB hugepage. This happens because get_user_pages_fast() is not aware
> >>> of a migration entry on pud that was created in the 1st madvise() event.
> > 
> > Do pgd size pages work properly?

PGD size is unsupported now too, and this patch is also disabling that size.

> 
> Adding Anshuman and Aneesh as they added pgd support for power.  And,
> this patch will disable that as well IIUC.

Thanks Mike, I want to have some feedback from PowerPC developers too.

> 
> This patch makes sense for x86.  My only concern/question is for other
> archs which may have huge page sizes defined which are > MAX_ORDER and
> < PUD_SIZE.  These would also be classified as gigantic and impacted
> by this patch.  Do these also have the same issue?

Maybe one clearer way is to use more explicit condition like "page size > PMD_SIZE".

> 
> -- 
> Mike Kravetz
> 
> >>> I think that conversion to pud-aligned migration entry is working,
> >>> but other MM code walking over page table isn't prepared for it.
> >>> We need some time and effort to make all this work properly, so
> >>> this patch avoids the reported bug by just disabling error handling
> >>> for 1GB hugepage.
> > 
> > Can we also get some documentation which would describe all requirements
> > for HWPoison pages to work properly please?

OK, I'll add this.

> > 
> >>> Signed-off-by: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
> > 
> > Acked-by: Michal Hocko <mhocko@suse.com>
> > 
> > We probably want a backport to stable as well. Although regular process
> > cannot get giga pages easily without admin help it is still not nice to
> > oops like this.

I'll add CC to stable.

Thanks,
Naoya Horiguchi