Re: [Patch] jbd commit code deadloop when installing Linux

[Zou Nan hai]

On Wed, 2006-06-28 at 15:40, Andrew Morton wrote:
> On 28 Jun 2006 13:46:22 +0800
> Zou Nan hai <nanhai.zou@xxxxxxxxx> wrote:
> 
> > On Wed, 2006-06-28 at 14:55, Andrew Morton wrote:
> > > On Wed, 28 Jun 2006 08:38:59 +0200
> > > Ingo Molnar <mingo@xxxxxxx> wrote:
> > > 
> > > > 
> > > > * Andrew Morton <akpm@xxxxxxxx> wrote:
> > > > 
> > > > > > We see system hang in ext3 jbd code
> > > > > > when Linux install program anaconda copying 
> > > > > > packages. 
> > > > > > 
> > > > > > That is because anaconda is invoked from linuxrc 
> > > > > > in initrd when system_state is still SYSTEM_BOOTING.
> > > > 
> > > > [ argh ...! ]
> > > 
> > > That's what I thought  ;)
> > > 
> > > > > > Thus the cond_resched checks in  journal_commit_transaction 
> > > > > > will always return 1 without actually schedule, 
> > > > > > then the system fall into deadloop.
> > > > > 
> > > > > That's a bug in cond_resched().
> > > > > 
> > > > > Something like this..
> > > > 
> > > > Acked-by: Ingo Molnar <mingo@xxxxxxx>
> > > > 
> > > 
> > > Thanks.  Zou, it'd be great if you could test this in your setup, please. 
> > > I've tagged it as 2.6.17.x material.
> > 
> > Andrew, 
> >    I am building the env to test.
> >    The patch was my original idea, but I was afraid of breaking any code
> > that rely on the OLD wrong cond_sched semantic.
> 
> We prefer the "right" fix, however painful or risky that might be.
> 
> > However later I did a
> > grep found that there is very few code that checks the return value of
> > cond_resched. So the patch should be safe. 
> 
> Hope so.
> 
> > However I think cond_resched_lock and cond_resched_softirq also need fix
> > to make the semantic consistent.
> > 
> > Please check the following patch.
> > 
> 
> Ah.  I think the return value from these functions should mean "something
> disruptive happened", if you like.
> 
> See, the callers of cond_resched_lock() aren't interested in whether
> cond_resched_lock() actually called schedule().  They want to know whether
> cond_resched_lock() dropped the lock.  Because if the lock was dropped, the
> caller needs to take some special action, regardless of whether schedule()
> was finally called.
> 
> So I think the patch I queued is OK, agree?

  I am afraid the code like cond_resched_lock check in
fs/jbd/checkpoint.c log_do_checkpoint may fall into endless retry in
some condition, will it?
  Though I have not encountered that.

Zou Nan hai
-
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/