Allow the NONROT flag to be passed through to linear mappings if all
underlying device are non-rotational. Tools like ureadahead will
schedule IOs differently based on the rotational flag.

With this patch, I see boot time go from 7.75 s to 7.46 s on my device.

Suggested-by: J. Richard Barnette <jrbarnette@chromium.org>
Signed-off-by: Mandeep Singh Baines <msb@chromium.org>
Cc: Neil Brown <neilb@suse.de>
Cc: Mike Snitzer <snitzer@redhat.com>
Cc: Jens Axboe <jaxboe@fusionio.com>
Cc: Martin K. Petersen <martin.petersen@oracle.com>
Cc: Alasdair G Kergon <agk@redhat.com>
Cc: dm-devel@redhat.com
---
drivers/md/dm-table.c | 28 ++++++++++++++++++++++++++++
1 files changed, 28 insertions(+), 0 deletions(-)

diff --git a/drivers/md/dm-table.c b/drivers/md/dm-table.c
index cb8380c..40c6071 100644
--- a/drivers/md/dm-table.c
+++ b/drivers/md/dm-table.c
@@ -1177,6 +1177,30 @@ static void dm_table_set_integrity(struct dm_table *t)
blk_get_integrity(template_disk));
}

+static int device_nonrot(struct dm_target *ti, struct dm_dev *dev,
+ sector_t start, sector_t len, void *data)
+{
+ struct request_queue *q = bdev_get_queue(dev->bdev);
+
+ return q && blk_queue_nonrot(q);
+}
+
+static bool dm_table_all_nonrot(struct dm_table *t)
+{
+ unsigned i = 0;
+
+ /* Ensure that all underlying device are non rotational. */
+ while (i < dm_table_get_num_targets(t)) {
+ struct dm_target *ti = dm_table_get_target(t, i++);
+
+ if (!ti->type->iterate_devices ||
+ !ti->type->iterate_devices(ti, device_nonrot, NULL))
+ return false;
+ }
+
+ return true;
+}
+
void dm_table_set_restrictions(struct dm_table *t, struct request_queue *q,
struct queue_limits *limits)
{
@@ -1189,6 +1213,10 @@ void dm_table_set_restrictions(struct dm_table *t, struct request_queue *q,
queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q);
else
queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
+ if (!dm_table_all_nonrot(t))
+ queue_flag_clear_unlocked(QUEUE_FLAG_NONROT, q);
+ else
+ queue_flag_set_unlocked(QUEUE_FLAG_NONROT, q);

dm_table_set_integrity(t);

--
1.7.3.1

--
dm-devel mailing list
dm-devel@redhat.com
https://www.redhat.com/mailman/listinfo/dm-devel

On Wed, May 25 2011 at 4:50pm -0400,
Mandeep Singh Baines <msb@chromium.org> wrote:

> Allow the NONROT flag to be passed through to linear mappings if all
> underlying device are non-rotational. Tools like ureadahead will
> schedule IOs differently based on the rotational flag.
>
> With this patch, I see boot time go from 7.75 s to 7.46 s on my device.
>
> Suggested-by: J. Richard Barnette <jrbarnette@chromium.org>
> Signed-off-by: Mandeep Singh Baines <msb@chromium.org>
> Cc: Neil Brown <neilb@suse.de>
> Cc: Mike Snitzer <snitzer@redhat.com>
> Cc: Jens Axboe <jaxboe@fusionio.com>
> Cc: Martin K. Petersen <martin.petersen@oracle.com>
> Cc: Alasdair G Kergon <agk@redhat.com>
> Cc: dm-devel@redhat.com

Thanks for doing this, looks good in general.

Just a few small nits:
- rename device_nonrot to device_is_nonrot
- rename dm_table_all_nonrot to dm_table_is_nonrot
- add missing whitespace in dm_table_set_restrictions
- move ti declaration outside of dm_table_is_nonrot's while loop
- dm doesn't use 'true' or 'false' bool returns even though we have
functions that return bool -- odd but best to keep consistent for now

I'll respond with v2 that folds these changes in and my Signed-off-by
---
drivers/md/dm-table.c | 18 ++++++++++--------
1 files changed, 10 insertions(+), 8 deletions(-)

diff --git a/drivers/md/dm-table.c b/drivers/md/dm-table.c
index 40c6071..a173828 100644
--- a/drivers/md/dm-table.c
+++ b/drivers/md/dm-table.c
@@ -1177,28 +1177,29 @@ static void dm_table_set_integrity(struct dm_table *t)
blk_get_integrity(template_disk));
}

-static int device_nonrot(struct dm_target *ti, struct dm_dev *dev,
- sector_t start, sector_t len, void *data)
+static int device_is_nonrot(struct dm_target *ti, struct dm_dev *dev,
+ sector_t start, sector_t len, void *data)
{
struct request_queue *q = bdev_get_queue(dev->bdev);

return q && blk_queue_nonrot(q);
}

-static bool dm_table_all_nonrot(struct dm_table *t)
+static bool dm_table_is_nonrot(struct dm_table *t)
{
+ struct dm_target *ti;
unsigned i = 0;

/* Ensure that all underlying device are non rotational. */
while (i < dm_table_get_num_targets(t)) {
- struct dm_target *ti = dm_table_get_target(t, i++);
+ ti = dm_table_get_target(t, i++);

if (!ti->type->iterate_devices ||
- !ti->type->iterate_devices(ti, device_nonrot, NULL))
- return false;
+ !ti->type->iterate_devices(ti, device_is_nonrot, NULL))
+ return 0;
}

- return true;
+ return 1;
}

void dm_table_set_restrictions(struct dm_table *t, struct request_queue *q,
@@ -1213,7 +1214,8 @@ void dm_table_set_restrictions(struct dm_table *t, struct request_queue *q,
queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q);
else
queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
- if (!dm_table_all_nonrot(t))
+
+ if (!dm_table_is_nonrot(t))
queue_flag_clear_unlocked(QUEUE_FLAG_NONROT, q);
else
queue_flag_set_unlocked(QUEUE_FLAG_NONROT, q);

--
dm-devel mailing list
dm-devel@redhat.com
https://www.redhat.com/mailman/listinfo/dm-devel

>>>>> "Mike" == Mike Snitzer <snitzer@redhat.com> writes:

>> Allow the NONROT flag to be passed through to linear mappings if all
>> underlying device are non-rotational. Tools like ureadahead will
>> schedule IOs differently based on the rotational flag.

Mike> Thanks for doing this, looks good in general.

I'd rather we made the rotational property part of the topology instead
of having to special-case it in DM. I seem to recall posting a patch
that did that at some point but I can't seem to find it...

--
Martin K. Petersen Oracle Linux Engineering

--
dm-devel mailing list
dm-devel@redhat.com
https://www.redhat.com/mailman/listinfo/dm-devel

On Thu, May 26 2011 at 2:29pm -0400,
Martin K. Petersen <martin.petersen@oracle.com> wrote:

> >>>>> "Mike" == Mike Snitzer <snitzer@redhat.com> writes:
>
> >> Allow the NONROT flag to be passed through to linear mappings if all
> >> underlying device are non-rotational. Tools like ureadahead will
> >> schedule IOs differently based on the rotational flag.
>
> Mike> Thanks for doing this, looks good in general.
>
> I'd rather we made the rotational property part of the topology instead
> of having to special-case it in DM. I seem to recall posting a patch
> that did that at some point but I can't seem to find it...

Sure, that'd work too.

Similarly, couldn't the discard flag be made part of topology?

Mike

--
dm-devel mailing list
dm-devel@redhat.com
https://www.redhat.com/mailman/listinfo/dm-devel

>>>>> "Mike" == Mike Snitzer <snitzer@redhat.com> writes:

Mike> Similarly, couldn't the discard flag be made part of topology?

Actually, now that I think about it I'm fairly sure my patch nuked the
discard queue flag because I was tired of keeping the flag and the
topology in sync. I don't think I mucked with the rotational flag.

Let me grep my patch queue and see. It must be in there somewhere...

--
Martin K. Petersen Oracle Linux Engineering

--
dm-devel mailing list
dm-devel@redhat.com
https://www.redhat.com/mailman/listinfo/dm-devel

On 2011-05-26 20:29, Martin K. Petersen wrote:
>>>>>> "Mike" == Mike Snitzer <snitzer@redhat.com> writes:
>
>>> Allow the NONROT flag to be passed through to linear mappings if all
>>> underlying device are non-rotational. Tools like ureadahead will
>>> schedule IOs differently based on the rotational flag.
>
> Mike> Thanks for doing this, looks good in general.
>
> I'd rather we made the rotational property part of the topology instead
> of having to special-case it in DM. I seem to recall posting a patch
> that did that at some point but I can't seem to find it...

Agree, would be a better idea.

--
Jens Axboe

--
dm-devel mailing list
dm-devel@redhat.com
https://www.redhat.com/mailman/listinfo/dm-devel

Here's a tentative patch set for transitioning the non_rotational,
discard and secure discard queue flags to the topology information.

The first patch introduces a new function for setting the default limits
for stacking drivers like we discussed a few weeks ago. Wrt. to naming
I felt it was best to have a function that said "stacking". That's why
I introduced a new call instead of using blk_set_default_limits() even
though that function was originally motivated by device mapper.

The last two patches deal with non-rotational and the discard flags
respectively.

Let me know what you guys think...

--
Martin K. Petersen Oracle Linux Engineering

--
dm-devel mailing list
dm-devel@redhat.com
https://www.redhat.com/mailman/listinfo/dm-devel