Re: [RFC] ext3/jbd race: releasing in-use journal_heads

[Stephen C. Tweedie]

Hi,

On Mon, 2005-03-07 at 21:11, Andrew Morton wrote:

> > I'm having trouble testing it, though --- I seem to be getting livelocks
> >  in O_DIRECT running 400 fsstress processes in parallel; ring any bells? 
> 
> Nope.  I dont think anyone has been that cruel to ext3 for a while.
> I assume this workload used to succeed?

I think so.  I remember getting a lockup a couple of months ago that I
couldn't get to the bottom of and that looked very similar, so it may
just be a matter of it being much more reproducible now.  But I seem to
be able to livelock the directIO bits in minutes now, quite reliably. 
I'll try to narrow things down.

altgr-scrlck is showing a range of EIPs all in ext3_direct_IO->
invalidate_inode_pages2_range().  I'm seeing

invalidate_inode_pages2_range()->pagevec_lookup()->find_get_pages()

as by far the most common trace, but also 

invalidate_inode_pages2_range()->__might_sleep()
invalidate_inode_pages2_range()->unlock_page()
and a few other similar paths.

invalidate_inode_pages2_range() does do a cond_resched(), so the machine
carries on despite having 316 fsstress tasks in system-mode R state at
the moment. :-)  The scheduler is doing a rather impressive job.

--Stephen

-
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/