| Message ID | 20220308165349.231320-3-p.raghav@samsung.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | power_of_2 emulation support for NVMe ZNS devices | expand |
On 3/9/22 01:53, Pankaj Raghav wrote: > A new fops is added to block device which will be used to setup the > necessary hooks when a non-power_of_2 zone size is detected in a zoned > device. > > This fops will be called as a part of blk_revalidate_disk_zones. And what does this new hook do ? You are actually not explaining it, nor why it should be called from blk_revalidate_disk_zones(). Also, blk_revalidate_zone_cb() uses bit shift but this patch, nor the previous one fix that. > > The primary use case for this callback is to deal with zoned devices > that does not have a power_of_2 zone size such as ZNS drives.For e.g, > the current NVMe ZNS specification does not require zone size to be > power_of_2 but the linux block layer still expects a all zoned device to > have a power_of_2 zone size. > > Signed-off-by: Pankaj Raghav <p.raghav@samsung.com> > --- > block/blk-zoned.c | 3 +++ > include/linux/blkdev.h | 1 + > 2 files changed, 4 insertions(+) > > diff --git a/block/blk-zoned.c b/block/blk-zoned.c > index 602bef54c813..d3d821797559 100644 > --- a/block/blk-zoned.c > +++ b/block/blk-zoned.c > @@ -575,6 +575,9 @@ int blk_revalidate_disk_zones(struct gendisk *disk, > if (!get_capacity(disk)) > return -EIO; > > + if (disk->fops->npo2_zone_setup) > + disk->fops->npo2_zone_setup(disk); > + > /* > * Ensure that all memory allocations in this context are done as if > * GFP_NOIO was specified. > diff --git a/include/linux/blkdev.h b/include/linux/blkdev.h > index a12c031af887..08cf039c1622 100644 > --- a/include/linux/blkdev.h > +++ b/include/linux/blkdev.h > @@ -1472,6 +1472,7 @@ struct block_device_operations { > void (*swap_slot_free_notify) (struct block_device *, unsigned long); > int (*report_zones)(struct gendisk *, sector_t sector, > unsigned int nr_zones, report_zones_cb cb, void *data); > + void (*npo2_zone_setup)(struct gendisk *disk); > char *(*devnode)(struct gendisk *disk, umode_t *mode); > /* returns the length of the identifier or a negative errno: */ > int (*get_unique_id)(struct gendisk *disk, u8 id[16],
On 2022-03-09 04:46, Damien Le Moal wrote: > On 3/9/22 01:53, Pankaj Raghav wrote: >> A new fops is added to block device which will be used to setup the >> necessary hooks when a non-power_of_2 zone size is detected in a zoned >> device. >> >> This fops will be called as a part of blk_revalidate_disk_zones. > > And what does this new hook do ? You are actually not explaining it, nor > why it should be called from blk_revalidate_disk_zones(). I should have elaborated the "why" and "how" a bit more in my commit log. I will fix it in my next revision. The main idea why it was added and called as a part of blk_revalidate_disk_zones is this: As the block layer expects the zone sizes to be po2, this fops can be used by the driver to configure a npo2 device to be presented as a po2 device before the parameters such as zone bitmaps and chunk sectors are set. > Also, blk_revalidate_zone_cb() uses bit shift but this patch, nor the > previous one fix that. > The answer I gave to blkdev_nr_zones question for patch 1/6 applies here as well. The zone sizes used by blk_revalidate_zone_cb will be the emulated po2 zone size and not the actual device zone size which is npo2. So the math currently used in blk_revalidate_zone_cb is still applicable.
diff --git a/block/blk-zoned.c b/block/blk-zoned.c index 602bef54c813..d3d821797559 100644 --- a/block/blk-zoned.c +++ b/block/blk-zoned.c @@ -575,6 +575,9 @@ int blk_revalidate_disk_zones(struct gendisk *disk, if (!get_capacity(disk)) return -EIO; + if (disk->fops->npo2_zone_setup) + disk->fops->npo2_zone_setup(disk); + /* * Ensure that all memory allocations in this context are done as if * GFP_NOIO was specified. diff --git a/include/linux/blkdev.h b/include/linux/blkdev.h index a12c031af887..08cf039c1622 100644 --- a/include/linux/blkdev.h +++ b/include/linux/blkdev.h @@ -1472,6 +1472,7 @@ struct block_device_operations { void (*swap_slot_free_notify) (struct block_device *, unsigned long); int (*report_zones)(struct gendisk *, sector_t sector, unsigned int nr_zones, report_zones_cb cb, void *data); + void (*npo2_zone_setup)(struct gendisk *disk); char *(*devnode)(struct gendisk *disk, umode_t *mode); /* returns the length of the identifier or a negative errno: */ int (*get_unique_id)(struct gendisk *disk, u8 id[16],
A new fops is added to block device which will be used to setup the necessary hooks when a non-power_of_2 zone size is detected in a zoned device. This fops will be called as a part of blk_revalidate_disk_zones. The primary use case for this callback is to deal with zoned devices that does not have a power_of_2 zone size such as ZNS drives.For e.g, the current NVMe ZNS specification does not require zone size to be power_of_2 but the linux block layer still expects a all zoned device to have a power_of_2 zone size. Signed-off-by: Pankaj Raghav <p.raghav@samsung.com> --- block/blk-zoned.c | 3 +++ include/linux/blkdev.h | 1 + 2 files changed, 4 insertions(+)