On Tue, 1 Aug 2000, Andrea Arcangeli wrote:
> On Mon, 31 Jul 2000, Linus Torvalds wrote:
>
> >do something like
> >
> > int wait;
> >
> > if (!(gfp_mask & __GFP_IO))
> > goto unlock_continue;
> > wait = nr_dirty-- < 0;
>
> We still want to recycle the clean buffers from a GFP_BUFFER allocation.
> The fix I prefer is to pass gfp_mask to try_to_free_buffers and to call
> sync_page_buffers only if __GFP_IO was set (checking for it inside
> try_to_free_buffers).
Linus's fix seems to do the trick. I beat on it _very hard_ for four
hours and couldn't lock it. I'll try your suggestion too. The only
thing I saw during pounding is that heavy swapping to a loop device
is dog slow.. and maybe a little silly :) but it ran without locking.
Linus: There is something very odd going on though...
If I mount -oloop /usr/local and try a make -j30, I see read corruption
(cpp carping) as soon as memory pressure hits. If I start with /usr/local
mounted, and mount -oremount,loop all is well. (repeatable) When in
corruption mode, it will nibble on the underlying fs, so writes are being
corrupted too. I've found blocks of zeros replacing various data blocks in
the recent past after tinker sessions. (no damage noted today)
That's why I agreed with DANGEROUS.. it can nibble an fs pretty badly if
you play with it long enough when it's in this.. bad mood.
-Mike
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Mon Aug 07 2000 - 21:00:06 EST