[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <47069315.8030802@goop.org>
Date: Fri, 05 Oct 2007 12:40:05 -0700
From: Jeremy Fitzhardinge <jeremy@...p.org>
To: Rik van Riel <riel@...hat.com>
CC: Hugh Dickins <hugh@...itas.com>,
David Rientjes <rientjes@...gle.com>,
Zachary Amsden <zach@...are.com>,
Andrew Morton <akpm@...ux-foundation.org>,
Linus Torvalds <torvalds@...ux-foundation.org>,
Rusty Russell <rusty@...tcorp.com.au>, Andi Kleen <ak@...e.de>,
Keir Fraser <keir@...source.com>,
Linux Kernel Mailing List <linux-kernel@...r.kernel.org>
Subject: Re: race with page_referenced_one->ptep_test_and_clear_young and
pagetable setup/pulldown
Rik van Riel wrote:
> This makes for a narrow race window, during which ptep_test_and_clear_young
> cannot clear the referenced bit and may end up causing a crash. We do not
> care about it not clearing the referenced bit during that window, since it
> will be cleared during the next go-around and the race is very rare.
>
Is this the only possible case? It's just the one I found while testing
under high memory pressure.
> Hence, the only thing we need to fix is the crash.
>
> We can do that by adding an entry for ptep_test_and_clear_young to the
> exception table. This way we do not need to turn this into a new paravirt
> ops hook (since the fast path is exactly the same as x86 native) and there
> is no need for added complexity.
>
> Also, Xen would not conflict with SPLIT_PTLOCK_CPUS.
>
Well, isn't the correct fix to make Xen take all the pagetable locks
while pinning/unpinning? Adding exception handling to
test_and_clear_bit would solve this particular race, but are there
others (either now or potentially)? Seems fragile.
J
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
Powered by blists - more mailing lists