Re: frequent lockups in 3.18rc4

[Al Viro]

On Sat, Dec 13, 2014 at 03:38:57PM -0800, Linus Torvalds wrote:
> On Sat, Dec 13, 2014 at 3:35 PM, Al Viro <viro@xxxxxxxxxxxxxxxxxx> wrote:
> >
> > Er...  There's much more direct reason - suppose we get a timer interrupt
> > right in the middle of mnt_drop_write().  And lost the timeslice.
> 
> So?
> 
> You didn't have preemption disabled in *between* the mnt_want_write()
> and mnt_drop_write(), there's absolutely no reason to have it inside
> of them.
> 
> Nobody cares if you get preempted and go away for a while. It's
> exactly equivalent to sleeping while doing the write that the pair was
> protecting.
> 
> Seriously, the preemption disable looks like just voodoo code. It
> doesn't protect anything, it doesn't fix anything, it doesn't change
> anything. All it does is disable preemption over a random sequence of
> code.

Huh?  Sure, we can enable it after mnt_inc_writers() and disable just prior to
mnt_dec_writers(), but we absolutely *do* need it disabled during either.
Is that what you are talking about?  If so, yes, we can do that.

But that applies only to __mnt_want_write() - __mnt_drop_write() is pure
mnt_dec_writers() and we can't call that one with preemption enabled.
Seriously, look at the mnt_dec_writers():
static inline void mnt_dec_writers(struct mount *mnt)
{
#ifdef CONFIG_SMP
        this_cpu_dec(mnt->mnt_pcp->mnt_writers);  
#else
        mnt->mnt_writers--;
#endif
}
It's load/modify/store, without any kind of atomicity; get preempted in the
middle of that sequence by another caller of mnt_dec_writers() and obvious bad
things will happen...

Al, really confused by now...
--
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/