| Message ID | 20240605022445.105747-2-dlemoal@kernel.org (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | Fix DM zone resource limits stacking | expand |
improve is misspelled in the subject. > @@ -80,6 +80,10 @@ static int blk_validate_zoned_limits(struct queue_limits *lim) > if (WARN_ON_ONCE(!IS_ENABLED(CONFIG_BLK_DEV_ZONED))) > return -EINVAL; > > + if (lim->max_active_zones && > + WARN_ON_ONCE(lim->max_open_zones > lim->max_active_zones)) > + lim->max_open_zones = lim->max_active_zones; Given how active zones are defined this is an error condition, and should return -EINVAL. > diff --git a/block/blk-zoned.c b/block/blk-zoned.c > index 52abebf56027..2af4d5ca81d2 100644 > --- a/block/blk-zoned.c > +++ b/block/blk-zoned.c > @@ -1660,6 +1660,11 @@ static int disk_update_zone_resources(struct gendisk *disk, > lim = queue_limits_start_update(q); > > nr_seq_zones = disk->nr_zones - nr_conv_zones; > + if (WARN_ON_ONCE(lim.max_active_zones > nr_seq_zones)) > + lim.max_active_zones = 0; > + if (WARN_ON_ONCE(lim.max_open_zones > nr_seq_zones)) > + lim.max_open_zones = 0; Why would you warn about this? Offering an open/active limit larger than the number of sequential zones is a pretty natural condition for certain corner cases (e.g. create only a tiny namespace on a ZNS SSD). This could also use a code comment explaining why the limit is adjusted.
On 6/5/24 13:17, Christoph Hellwig wrote: > improve is misspelled in the subject. > >> @@ -80,6 +80,10 @@ static int blk_validate_zoned_limits(struct queue_limits *lim) >> if (WARN_ON_ONCE(!IS_ENABLED(CONFIG_BLK_DEV_ZONED))) >> return -EINVAL; >> >> + if (lim->max_active_zones && >> + WARN_ON_ONCE(lim->max_open_zones > lim->max_active_zones)) >> + lim->max_open_zones = lim->max_active_zones; > > Given how active zones are defined this is an error condition, and > should return -EINVAL. > >> diff --git a/block/blk-zoned.c b/block/blk-zoned.c >> index 52abebf56027..2af4d5ca81d2 100644 >> --- a/block/blk-zoned.c >> +++ b/block/blk-zoned.c >> @@ -1660,6 +1660,11 @@ static int disk_update_zone_resources(struct gendisk *disk, >> lim = queue_limits_start_update(q); >> >> nr_seq_zones = disk->nr_zones - nr_conv_zones; >> + if (WARN_ON_ONCE(lim.max_active_zones > nr_seq_zones)) >> + lim.max_active_zones = 0; >> + if (WARN_ON_ONCE(lim.max_open_zones > nr_seq_zones)) >> + lim.max_open_zones = 0; > > Why would you warn about this? Offering an open/active limit larger > than the number of sequential zones is a pretty natural condition > for certain corner cases (e.g. create only a tiny namespace on a ZNS > SSD). This could also use a code comment explaining why the limit > is adjusted. Right. I actually did not consider that case, which is indeed valid given that for nvme, the limits are per controller, not namespace (which is a very unfortunate design flaw...). I will remove the warn and add a comment.
diff --git a/block/blk-settings.c b/block/blk-settings.c index effeb9a639bb..a79c57376ef7 100644 --- a/block/blk-settings.c +++ b/block/blk-settings.c @@ -80,6 +80,10 @@ static int blk_validate_zoned_limits(struct queue_limits *lim) if (WARN_ON_ONCE(!IS_ENABLED(CONFIG_BLK_DEV_ZONED))) return -EINVAL; + if (lim->max_active_zones && + WARN_ON_ONCE(lim->max_open_zones > lim->max_active_zones)) + lim->max_open_zones = lim->max_active_zones; + if (lim->zone_write_granularity < lim->logical_block_size) lim->zone_write_granularity = lim->logical_block_size; diff --git a/block/blk-zoned.c b/block/blk-zoned.c index 52abebf56027..2af4d5ca81d2 100644 --- a/block/blk-zoned.c +++ b/block/blk-zoned.c @@ -1660,6 +1660,11 @@ static int disk_update_zone_resources(struct gendisk *disk, lim = queue_limits_start_update(q); nr_seq_zones = disk->nr_zones - nr_conv_zones; + if (WARN_ON_ONCE(lim.max_active_zones > nr_seq_zones)) + lim.max_active_zones = 0; + if (WARN_ON_ONCE(lim.max_open_zones > nr_seq_zones)) + lim.max_open_zones = 0; + pool_size = max(lim.max_open_zones, lim.max_active_zones); if (!pool_size) pool_size = min(BLK_ZONE_WPLUG_DEFAULT_POOL_SIZE, nr_seq_zones);
Make sure that the zone resource limits of a zoned block device are correct by checking that: (a) If the device has a max active zones limit, make sure that the max open zones limit is lower than the max active zones limit. (b) If the device has a max open zones or a max active zones limit, check that the limits are lower than the number of sequential zones of the device. For (a), a check is added to blk_validate_zoned_limits(). For (b), given that we need to number of sequential zones of the device, this check is added to disk_update_zone_resources(). This is safe to do as that function is executed with the queue frozen and the check executed after queue_limits_start_update() with the queue limits lock held. Signed-off-by: Damien Le Moal <dlemoal@kernel.org> --- block/blk-settings.c | 4 ++++ block/blk-zoned.c | 5 +++++ 2 files changed, 9 insertions(+)