Re: [PATCH v4 8/10] mm/hwpoison: fix memory failure still holdreference count after unpoison empty zero page

From: Naoya Horiguchi
Date: Mon Aug 26 2013 - 20:47:21 EST


On Tue, Aug 27, 2013 at 08:21:05AM +0800, Wanpeng Li wrote:
> Hi Naoya,
> On Mon, Aug 26, 2013 at 08:12:29PM -0400, Naoya Horiguchi wrote:
> >Hi Wanpeng,
> >
> >On Tue, Aug 27, 2013 at 07:26:04AM +0800, Wanpeng Li wrote:
> >> Hi Naoya,
> >> On Mon, Aug 26, 2013 at 11:45:37AM -0400, Naoya Horiguchi wrote:
> >> >On Mon, Aug 26, 2013 at 04:46:12PM +0800, Wanpeng Li wrote:
> >> >> madvise hwpoison inject will poison the read-only empty zero page if there is
> >> >> no write access before poison. Empty zero page reference count will be increased
> >> >> for hwpoison, subsequent poison zero page will return directly since page has
> >> >> already been set PG_hwpoison, however, page reference count is still increased
> >> >> by get_user_pages_fast. The unpoison process will unpoison the empty zero page
> >> >> and decrease the reference count successfully for the fist time, however,
> >> >> subsequent unpoison empty zero page will return directly since page has already
> >> >> been unpoisoned and without decrease the page reference count of empty zero page.
> >> >> This patch fix it by decrease page reference count for empty zero page which has
> >> >> already been unpoisoned and page count > 1.
> >> >
> >> >I guess that fixing on the madvise side looks reasonable to me, because this
> >> >refcount mismatch happens only when we poison with madvise(). The root cause
> >> >is that we can get refcount multiple times on a page, even if memory_failure()
> >> >or soft_offline_page() can do its work only once.
> >> >
> >>
> >> I think this just happen in read-only before poison case against empty
> >> zero page.
> >
> >OK. I agree.
> >
> >> Hi Andrew,
> >>
> >> I see you have already merged the patch, which method you prefer?
> >>
> >> >How about making madvise_hwpoison() put a page and return immediately
> >> >(without calling memory_failure() or soft_offline_page()) when the page
> >> >is already hwpoisoned?
> >> >I hope it also helps us avoid meaningless printk flood.
> >> >
> >>
> >> Btw, Naoya, how about patch 10/10, any input are welcome! ;-)
> >
> >No objection if you (and Andrew) decide to go with current approach.
>
> Andrew prefer your method, I will resend the patch w/ your suggested-by. ;-)

Thanks you :)

> >But I think that if we shift to fix this problem in madvise(),
> >we don't need 10/10 any more. So it looks simpler to me.
>
> I don't think it's same issue. There is just one page in my test case.
> #define PAGES_TO_TEST 1
> If I miss something?

Ah, OK.

BTW, in my understanding, zero pages are not exist physically (I mean that
no real page is allocated to store 4096 bytes of 0.) So there can't happen
any real MCE SRAO on zero page. So one possible solution might be that we
completely ignore all of madvise(MADV_HWPOISON) over zero pages.

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/