Re: [PATCH linux-2.6.1-rc1-mm1] aiodio_fallback_bio_count.patch

[Suparna Bhattacharya <suparna _at_ xxxxxxxxxx>]

On Wed, Dec 31, 2003 at 03:46:48PM -0800, Andrew Morton wrote:
> Daniel McNeil <daniel _at_ osdl.org> wrote:
> >
> > This is an update of AIO fallback patch and with the bio_count race
> >  fixed by changing bio_list_lock into bio_lock and using that for all
> >  the bio fields.  I changed bio_count and bios_in_flight from atomics
> >  into int.  They are now proctected by the bio_lock.  I fixed the race,
> >  by in finished_one_bio() by leaving the bio_count at 1 until after the
> >  dio_complete() and then do the bio_count decrement and wakeup holding
> >  the bio_lock.
> 
> Sob.  Daniel, please assume that I have an extremely small brain and forget
> things very easily.  And that 2,000 patches have passed under my nose since
> last I contemplated the direct-io code, OK?
> 
> What bio_count race?
> 
> And the patch seems to be doing more than your description describes?
> --

We need to combine this with my original patch description of the
aio-dio-fallback + follow on patch which had the race that Daniel
fixed and then rolled together into one working patch.

Does the following sound better as complete description ?

-----------------------------------------------------------------

This patch ensures that when the DIO code falls back to buffered i/o 
after having submitted part of the i/o, then buffered i/o is issued only
for the remaining part of the request (i.e. the part not already
covered by DIO), rather than redo the entire i/o.

We need to careful not to access dio fields if its possible that 
the dio could already have been freed asynchronously during i/o 
completion. A tricky part of this involves plugging the window between 
the decrement of bio_count and accessing dio->waiter during i/o 
completion where the dio could get freed by the submission path. 
This potential "bio_count race" was tackled (by Daniel) by changing 
bio_list_lock into bio_lock and using that for all the bio fields. 
Now bio_count and bios_in_flight have been converted from atomics 
into int and are both protected by the bio_lock. The race in 
finished_one_bio() could thus be fixed by leaving the bio_count at 1 
until after the dio_complete() and then doing the bio_count decrement 
and wakeup holding the bio_lock. It appears that shifting to the
spin_lock instead of atomic_inc/decs is ok performance wise as 
well.

Regards
Suparna

-- 
Suparna Bhattacharya (suparna _at_ in.ibm.com)
Linux Technology Center
IBM Software Lab, India

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