Re: 2.6.10-rc1-mm5 [u]

["Martin Schlemmer [c]" <azarah@nosferatu.za.org>]

[-- Attachment #1: Type: text/plain, Size: 2084 bytes --]

On Sat, 2004-11-13 at 15:24 -0800, Andrew Morton wrote:
> "Martin Schlemmer [c]" <azarah@nosferatu.za.org> wrote:
> >
> >  > Could you please try:
> >  > 
> >  > wget ftp://ftp.kernel.org/pub/linux/kernel/people/akpm/patches/2.6/2.6.10-rc1/2.6.10-rc1-mm5/broken-out/futex_wait-fix.patch
> >  > patch -R -p1 < futex_wait-fix.patch
> >  > 
> >  > the retest?
> > 
> >  Yep, this seems to fix it (usually one thread at least have already hung
> >  at start for evo, now fine after a few mail fetches).
> 
> OK, thanks.
> 
> >  Is this a regression, or as in the other thread an issue with evolution
> >  that should be fixed ?  (Note:  gnome-btdownload also seemed to hang
> >  overnight with -mm[45], but I do not know if its related)
> 
> Don't know.  It's hard to see why that patch would cause gross misbehaviour
> in evolution and apache.  We may have to just revert it and take another
> look at the problem which it fixes.
> 

I will be honest that I know next to nothing about relevant code, but
this looks fishy:

------
@@ -486,6 +486,8 @@
        if (unlikely(ret != 0))
                goto out_release_sem;

-       queue_me(&q, -1, NULL);
-
        /*
         * Access the page after the futex is queued.
         * We hold the mmap semaphore, so the mapping cannot have changed
------

The 'queue_me(&q, -1, NULL);' gets moved down, but just below where it
was we have a comment before a get_user() ... it is not possible that
because the patch cause the get_user() to be before queue_me(), that
we can have two (or more?) get_user()'s for the same futex at relatively
the same time?


> One thing I do note which is unrelated to this problem is that futex_wait()
> does get_user() inside down_read(mmap_sem).  But a fault will do a second
> down_read().  And doing down_read() twice from within the same thread is a
> bug, because an intervening down_write() from another thread will cause
> deadlock.
> 

I will rather leave this to somebody that now something about this =)


Thanks,

-- 
Martin Schlemmer


[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 189 bytes --]