lkml.org 
[lkml]   [2005]   [Sep]   [9]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH 2.6.13-stable] cpuset semaphore double trip fix
On Fri, 9 Sep 2005 17:44:03 -0700 (PDT) Paul Jackson wrote:

> Code reading uncovered a potential deadlock on the global cpuset
> semaphore, cpuset_sem.
>
> ==> This patch is only useful in the 2.6.13-stable series.

Paul,

stable patch candidates should be sent to stable@kernel.org

Chris/Greg,
maybe that should be listed in MAINTAINERS as well as in
Documentation/stable_kernel_rules.txt (it's here already).


> (It's harmless, and useless, in the pre 2.6.14 fork)
>
> The pre-2.6.14 fork has already diverged, with an additional patch
> that further aggrevated this problem, and a more thorough overhaul
> of the cpuset locking, to fix the problems.
>
> All code paths in kernel/cpuset.c (2.6.13 or earlier) that first
> grab cpuset_sem and then allocate memory _must_ call the routine
> 'refresh_mems()', after getting cpuset_sem, before any possible
> allocation.
>
> If this refresh_mems() call is not done, then there is a risk that one
> of the cpuset_zone_allowed() calls made from within the page allocator
> (__alloc_pages) will find that the mems_generation of the current task
> doesn't match that of its cpuset, causing it to try to grab cpuset_sem.
> Since it already held cpuset_sem, this deadlocks that task, and any
> subsequent task wanting cpuset_sem.
>
> ==> The code paths leading to the kmalloc in check_for_release(), from
> cpuset_exit, cpuset_rmdir and attach_task (for the detached cpuset),
> fail to invoke refresh_mems() as required.
>
> The fix is easy enough - add the requisite refresh_mems() call.
>
> Unless someone is rapidly creating, modifying and destroying cpusets,
> they are unlikely to have any chance of encountering this deadlock.
> And even then, it is apparently difficult to do so.
>
> In the case we got here from cpuset_exit(), we have already torn
> down the tasks connection to this cpuset and current->cpuset is NULL.
> Don't call refresh_mems() in that case - it oops the kernel.
>
> Signed-off-by: Paul Jackson <pj@sgi.com>
>
> Index: linux-2.6.13-mem_exclusive_oom/kernel/cpuset.c
> ===================================================================
> --- linux-2.6.13-mem_exclusive_oom.orig/kernel/cpuset.c
> +++ linux-2.6.13-mem_exclusive_oom/kernel/cpuset.c
> @@ -458,7 +458,10 @@ static void check_for_release(struct cpu
> if (notify_on_release(cs) && atomic_read(&cs->count) == 0 &&
> list_empty(&cs->children)) {
> char *buf;
> + static void refresh_mems(void);
>
> + if (current->cpuset)
> + refresh_mems();
> buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
> if (!buf)
> return;
>

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

\
 
 \ /
  Last update: 2005-09-10 04:02    [W:0.037 / U:0.832 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site