Re: [PATCH 1/2] mm/hwpoison: fix fail to split THP w/ refcount held

[Wanpeng Li]

On 8/10/15 4:50 PM, Naoya Horiguchi wrote:
> On Mon, Aug 10, 2015 at 04:29:18PM +0800, Wanpeng Li wrote:
>> Hi Naoya,
>>
>> On 8/10/15 4:10 PM, Naoya Horiguchi wrote:
>>> On Mon, Aug 10, 2015 at 02:32:30PM +0800, Wanpeng Li wrote:
>>>> THP pages will get a refcount in madvise_hwpoison() w/ MF_COUNT_INCREASED
>>>> flag, however, the refcount is still held when fail to split THP pages.
>>>>
>>>> Fix it by reducing the refcount of THP pages when fail to split THP.
>>>>
>>>> Signed-off-by: Wanpeng Li <wanpeng.li@xxxxxxxxxxx>
>>> It seems that the same conditional put_page() would be added to
>>> "soft offline: %#lx page already poisoned" branch too, right?
>> PageHWPoison() is just called before the soft_offline_page() in
>> madvise_hwpoion(). I think the PageHWPosion()
>> check in soft_offline_page() makes more sense for the other
>> soft_offline_page() callsites which don't have the
>> refcount held.
> What I am worried is a race like below:
>
>   CPU0                              CPU1
>
>   madvise_hwpoison
>   get_user_pages_fast
>   PageHWPoison check (false)
>                                     memory_failure
>                                     TestSetPageHWPoison
>   soft_offline_page
>   PageHWPoison check (true)
>   return -EBUSY (without put_page)

Indeed, there is a race even through it is rared happen.

>
> It's rare and madvise_hwpoison() is testing feature, so this never causes
> real problems in production systems, so it's not a big deal.
> My suggestion is maybe just for code correctness thing ...

Thanks for your proposal, I will add your suggestion in v2 and post out
after we have a uniform solution for patch 2/2. :)

Regards,
Wanpeng Li

>
> Thanks,
> Naoya Horiguchi

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/