Message ID | 20180103043937.28235-1-kxuanobj@gmail.com (mailing list archive) |
---|---|
State | New, archived |
Headers | show |
On 03/01/2018 12:39 PM, Zhai Zhaoxuan wrote: > The function cached_dev_make_request() and flash_dev_make_request() call > generic_start_io_acct() with (struct bcache_device)->disk when they start a > closure. Then the function bio_complete() calls generic_end_io_acct() with > (struct search)->orig_bio->bi_disk when the closure has done. > Since the `bi_disk` is not the bcache device, the generic_end_io_acct() is > called with a wrong device queue. > > It causes the "inflight" (in struct hd_struct) counter keep increasing > without decreasing. > > This patch fix the problem by calling generic_end_io_acct() with > (struct bcache_device)->disk. > > Signed-off-by: Zhai Zhaoxuan <kxuanobj@gmail.com> Hi Zhaoxuan, Nice catch, it makes sense. Thanks. Reviewed-by: Coly Li <colyli@suse.de> One more question, do you experience any problem when inflight counter increases only ? I just wondering. Coly > --- > drivers/md/bcache/request.c | 4 ++-- > 1 file changed, 2 insertions(+), 2 deletions(-) > > diff --git a/drivers/md/bcache/request.c b/drivers/md/bcache/request.c > index 643c3021624f..83de85fe4542 100644 > --- a/drivers/md/bcache/request.c > +++ b/drivers/md/bcache/request.c > @@ -611,8 +611,8 @@ static void request_endio(struct bio *bio) > static void bio_complete(struct search *s) > { > if (s->orig_bio) { > - struct request_queue *q = s->orig_bio->bi_disk->queue; > - generic_end_io_acct(q, bio_data_dir(s->orig_bio), > + generic_end_io_acct(s->d->disk->queue, > + bio_data_dir(s->orig_bio), > &s->d->disk->part0, s->start_time); > > trace_bcache_request_end(s->d, s->orig_bio); >
On 03/01/2018 13:45 PM, Coly Li wrote: > On 03/01/2018 12:39 PM, Zhai Zhaoxuan wrote: >> The function cached_dev_make_request() and flash_dev_make_request() call >> generic_start_io_acct() with (struct bcache_device)->disk when they start a >> closure. Then the function bio_complete() calls generic_end_io_acct() with >> (struct search)->orig_bio->bi_disk when the closure has done. >> Since the `bi_disk` is not the bcache device, the generic_end_io_acct() is >> called with a wrong device queue. >> >> It causes the "inflight" (in struct hd_struct) counter keep increasing >> without decreasing. >> >> This patch fix the problem by calling generic_end_io_acct() with >> (struct bcache_device)->disk. >> >> Signed-off-by: Zhai Zhaoxuan <kxuanobj@gmail.com> > Hi Zhaoxuan, > > Nice catch, it makes sense. Thanks. > > Reviewed-by: Coly Li <colyli@suse.de> > > One more question, do you experience any problem when inflight counter > increases only ? I just wondering. Hi Coly, Thanks for your review. There is a problem in `iostat -x`. `iostat` shows the bcache device keep 100.00% busy (100.00 in the "%util" column), even if the bcache is idle. [root@base ~]# iostat -x -d 1 Linux 4.15.0-rc6-ARCH (base) 01/03/18 _x86_64_ (2 CPU) Device: rrqm/s wrqm/s r/s w/s rkB/s wkB/s avgrq-sz avgqu-sz await r_await w_await svctm %util vda 7.85 1.07 69.98 13.20 1818.60 158.20 47.54 0.04 1.08 0.67 3.28 0.30 2.47 vdb 0.00 0.00 2.61 0.00 75.80 0.00 58.09 0.00 0.75 0.75 0.00 0.00 0.00 bcache0 0.00 0.00 1.98 0.00 52.00 0.00 52.64 77.73 1.06 1.06 0.00 485.93 96.00 Device: rrqm/s wrqm/s r/s w/s rkB/s wkB/s avgrq-sz avgqu-sz await r_await w_await svctm %util vda 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 vdb 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 bcache0 0.00 0.00 0.00 0.00 0.00 0.00 0.00 81.00 0.00 0.00 0.00 0.00 100.00 Device: rrqm/s wrqm/s r/s w/s rkB/s wkB/s avgrq-sz avgqu-sz await r_await w_await svctm %util vda 0.00 9.00 0.00 2.00 0.00 44.00 44.00 0.00 10.00 0.00 10.00 0.00 0.00 vdb 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 bcache0 0.00 0.00 0.00 0.00 0.00 0.00 0.00 81.00 0.00 0.00 0.00 0.00 100.00 Device: rrqm/s wrqm/s r/s w/s rkB/s wkB/s avgrq-sz avgqu-sz await r_await w_await svctm %util vda 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 vdb 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 bcache0 0.00 0.00 0.00 0.00 0.00 0.00 0.00 81.00 0.00 0.00 0.00 0.00 100.00 ^C [root@base ~]# bcache-super-show /dev/vdb sb.magic ok sb.first_sector 8 [match] sb.csum 54824DA53D2DC101 [match] sb.version 1 [backing device] dev.label (empty) dev.uuid c5e43e66-4b28-48d7-bd46-73fb6e1efdbb dev.sectors_per_block 1 dev.sectors_per_bucket 1024 dev.data.first_sector 16 dev.data.cache_mode 0 [writethrough] dev.data.cache_state 1 [clean] cset.uuid a79f8422-ed64-4b92-8688-692f48cd6d97 [root@base ~]# bcache-super-show /dev/vda2 sb.magic ok sb.first_sector 8 [match] sb.csum EF7AB1A8928862E6 [match] sb.version 3 [cache device] dev.label (empty) dev.uuid 5762b745-77f8-4831-81ce-9ee7866b8d84 dev.sectors_per_block 1 dev.sectors_per_bucket 1024 dev.cache.first_sector 1024 dev.cache.cache_sectors 2096128 dev.cache.total_sectors 2097152 dev.cache.ordered yes dev.cache.discard no dev.cache.pos 0 dev.cache.replacement 0 [lru] cset.uuid a79f8422-ed64-4b92-8688-692f48cd6d97 > Coly > >> --- >> drivers/md/bcache/request.c | 4 ++-- >> 1 file changed, 2 insertions(+), 2 deletions(-) >> >> diff --git a/drivers/md/bcache/request.c b/drivers/md/bcache/request.c >> index 643c3021624f..83de85fe4542 100644 >> --- a/drivers/md/bcache/request.c >> +++ b/drivers/md/bcache/request.c >> @@ -611,8 +611,8 @@ static void request_endio(struct bio *bio) >> static void bio_complete(struct search *s) >> { >> if (s->orig_bio) { >> - struct request_queue *q = s->orig_bio->bi_disk->queue; >> - generic_end_io_acct(q, bio_data_dir(s->orig_bio), >> + generic_end_io_acct(s->d->disk->queue, >> + bio_data_dir(s->orig_bio), >> &s->d->disk->part0, s->start_time); >> >> trace_bcache_request_end(s->d, s->orig_bio); >>
On 03/01/2018 3:05 PM, Zhaoxuan wrote: > > On 03/01/2018 13:45 PM, Coly Li wrote: >> On 03/01/2018 12:39 PM, Zhai Zhaoxuan wrote: >>> The function cached_dev_make_request() and flash_dev_make_request() call >>> generic_start_io_acct() with (struct bcache_device)->disk when they >>> start a >>> closure. Then the function bio_complete() calls generic_end_io_acct() >>> with >>> (struct search)->orig_bio->bi_disk when the closure has done. >>> Since the `bi_disk` is not the bcache device, the >>> generic_end_io_acct() is >>> called with a wrong device queue. >>> >>> It causes the "inflight" (in struct hd_struct) counter keep increasing >>> without decreasing. >>> >>> This patch fix the problem by calling generic_end_io_acct() with >>> (struct bcache_device)->disk. >>> >>> Signed-off-by: Zhai Zhaoxuan <kxuanobj@gmail.com> >> Hi Zhaoxuan, >> >> Nice catch, it makes sense. Thanks. >> >> Reviewed-by: Coly Li <colyli@suse.de> >> >> One more question, do you experience any problem when inflight counter >> increases only ? I just wondering. > > Hi Coly, > > Thanks for your review. > > There is a problem in `iostat -x`. `iostat` shows the bcache device keep > 100.00% > busy (100.00 in the "%util" column), even if the bcache is idle. > > [root@base ~]# iostat -x -d 1 > Linux 4.15.0-rc6-ARCH (base) 01/03/18 _x86_64_ (2 CPU) > > Device: rrqm/s wrqm/s r/s w/s rkB/s wkB/s > avgrq-sz avgqu-sz await r_await w_await svctm %util > vda 7.85 1.07 69.98 13.20 1818.60 158.20 > 47.54 0.04 1.08 0.67 3.28 0.30 2.47 > vdb 0.00 0.00 2.61 0.00 75.80 0.00 > 58.09 0.00 0.75 0.75 0.00 0.00 0.00 > bcache0 0.00 0.00 1.98 0.00 52.00 0.00 > 52.64 77.73 1.06 1.06 0.00 485.93 96.00 > > Device: rrqm/s wrqm/s r/s w/s rkB/s wkB/s > avgrq-sz avgqu-sz await r_await w_await svctm %util > vda 0.00 0.00 0.00 0.00 0.00 0.00 > 0.00 0.00 0.00 0.00 0.00 0.00 0.00 > vdb 0.00 0.00 0.00 0.00 0.00 0.00 > 0.00 0.00 0.00 0.00 0.00 0.00 0.00 > bcache0 0.00 0.00 0.00 0.00 0.00 0.00 > 0.00 81.00 0.00 0.00 0.00 0.00 100.00 > > Device: rrqm/s wrqm/s r/s w/s rkB/s wkB/s > avgrq-sz avgqu-sz await r_await w_await svctm %util > vda 0.00 9.00 0.00 2.00 0.00 44.00 > 44.00 0.00 10.00 0.00 10.00 0.00 0.00 > vdb 0.00 0.00 0.00 0.00 0.00 0.00 > 0.00 0.00 0.00 0.00 0.00 0.00 0.00 > bcache0 0.00 0.00 0.00 0.00 0.00 0.00 > 0.00 81.00 0.00 0.00 0.00 0.00 100.00 > > Device: rrqm/s wrqm/s r/s w/s rkB/s wkB/s > avgrq-sz avgqu-sz await r_await w_await svctm %util > vda 0.00 0.00 0.00 0.00 0.00 0.00 > 0.00 0.00 0.00 0.00 0.00 0.00 0.00 > vdb 0.00 0.00 0.00 0.00 0.00 0.00 > 0.00 0.00 0.00 0.00 0.00 0.00 0.00 > bcache0 0.00 0.00 0.00 0.00 0.00 0.00 > 0.00 81.00 0.00 0.00 0.00 0.00 100.00 > > ^C > [root@base ~]# bcache-super-show /dev/vdb > sb.magic ok > sb.first_sector 8 [match] > sb.csum 54824DA53D2DC101 [match] > sb.version 1 [backing device] > > dev.label (empty) > dev.uuid c5e43e66-4b28-48d7-bd46-73fb6e1efdbb > dev.sectors_per_block 1 > dev.sectors_per_bucket 1024 > dev.data.first_sector 16 > dev.data.cache_mode 0 [writethrough] > dev.data.cache_state 1 [clean] > > cset.uuid a79f8422-ed64-4b92-8688-692f48cd6d97 > [root@base ~]# bcache-super-show /dev/vda2 > sb.magic ok > sb.first_sector 8 [match] > sb.csum EF7AB1A8928862E6 [match] > sb.version 3 [cache device] > > dev.label (empty) > dev.uuid 5762b745-77f8-4831-81ce-9ee7866b8d84 > dev.sectors_per_block 1 > dev.sectors_per_bucket 1024 > dev.cache.first_sector 1024 > dev.cache.cache_sectors 2096128 > dev.cache.total_sectors 2097152 > dev.cache.ordered yes > dev.cache.discard no > dev.cache.pos 0 > dev.cache.replacement 0 [lru] > > cset.uuid a79f8422-ed64-4b92-8688-692f48cd6d97 > Hi Zhaoxuan, Cool, this is an already reported issue but I still have no idea how to fix it yet... Thank you for the fix and the above detailed information! Coly Li
On 01/02/2018 08:39 PM, Zhai Zhaoxuan wrote: > The function cached_dev_make_request() and flash_dev_make_request() call > generic_start_io_acct() with (struct bcache_device)->disk when they start a > closure. Then the function bio_complete() calls generic_end_io_acct() with > (struct search)->orig_bio->bi_disk when the closure has done. > Since the `bi_disk` is not the bcache device, the generic_end_io_acct() is > called with a wrong device queue. > > It causes the "inflight" (in struct hd_struct) counter keep increasing > without decreasing. > > This patch fix the problem by calling generic_end_io_acct() with > (struct bcache_device)->disk. > > Signed-off-by: Zhai Zhaoxuan <kxuanobj@gmail.com> Verified, looks good to me, added to my for-next staging branch. Reviewed-by: Michael Lyle <mlyle@lyle.org>
diff --git a/drivers/md/bcache/request.c b/drivers/md/bcache/request.c index 643c3021624f..83de85fe4542 100644 --- a/drivers/md/bcache/request.c +++ b/drivers/md/bcache/request.c @@ -611,8 +611,8 @@ static void request_endio(struct bio *bio) static void bio_complete(struct search *s) { if (s->orig_bio) { - struct request_queue *q = s->orig_bio->bi_disk->queue; - generic_end_io_acct(q, bio_data_dir(s->orig_bio), + generic_end_io_acct(s->d->disk->queue, + bio_data_dir(s->orig_bio), &s->d->disk->part0, s->start_time); trace_bcache_request_end(s->d, s->orig_bio);
The function cached_dev_make_request() and flash_dev_make_request() call generic_start_io_acct() with (struct bcache_device)->disk when they start a closure. Then the function bio_complete() calls generic_end_io_acct() with (struct search)->orig_bio->bi_disk when the closure has done. Since the `bi_disk` is not the bcache device, the generic_end_io_acct() is called with a wrong device queue. It causes the "inflight" (in struct hd_struct) counter keep increasing without decreasing. This patch fix the problem by calling generic_end_io_acct() with (struct bcache_device)->disk. Signed-off-by: Zhai Zhaoxuan <kxuanobj@gmail.com> --- drivers/md/bcache/request.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)