On Thu, Oct 23 2008, Miquel van Smoorenburg wrote:
quoted text
> On Wed, 2008-10-22 at 19:17 +0200, Jens Axboe wrote:
> > On Wed, Oct 22 2008, Miquel van Smoorenburg wrote:
> > > On Mon, 2008-10-20 at 19:12 +0200, Jens Axboe wrote:
> > > > On Sun, Oct 19 2008, Miquel van Smoorenburg wrote:
> >
> > > > > I noticed that the "iostat -k -x 2" output does't make any sense.
> > > > > The number of reads/sec and number of writes/sec are about what I
> > > > > would expect, and so are the other fields, but rkB/sec and wkB/sec
> > > > > are  completely off-scale: gigabytes read/written per second.
> > > >
> > > 
> > > > Are the reported values in iostat any sort of multiple of the real
> > > > throughtput, or is is just insanely large?
> > > 
> > > It looks like it's a multiple, but it appears to vary between 128 and
> > > 512, and is different for reads and writes, so I do not know what to
> > > make of it.
> > > 
> 
> I added some debug statements to block/blk-core.c, and it appears that
> blk_end_io() is always called with nr_bytes == 16320 Kbytes (16711680).
> 
> Ofcourse I should have noticed earlier that iostat -x 2 always prints
> 32640 as "average request size" (in sectors).

You didn't happen to log a backtrace from that, did you? That's a weird
nr_bytes, it's 0xff0000.

I'm assuming this then fixes it, just wondering what the heck is going
on here.

diff --git a/block/blk-core.c b/block/blk-core.c
index 33c8d4b..a372618 100644
--- a/block/blk-core.c
+++ b/block/blk-core.c
@@ -1712,11 +1710,12 @@ static int __end_that_request_first(struct request *req, int error,
 	if (blk_fs_request(req) && req->rq_disk) {
 		const int rw = rq_data_dir(req);
 		struct hd_struct *part;
-		int cpu;
+		int cpu, sectors;
 
+		sectors = min_t(int, nr_bytes >> 9, req->hard_nr_sectors);
 		cpu = part_stat_lock();
 		part = disk_map_sector_rcu(req->rq_disk, req->sector);
-		part_stat_add(cpu, part, sectors[rw], nr_bytes >> 9);
+		part_stat_add(cpu, part, sectors[rw], sectors);
 		part_stat_unlock();
 	}
 

-- 
Jens Axboe

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