On 2020/02/11 7:20, Alexey Dobriyan wrote:
> SK hynix is going to ship ZNS device with zone size not being power of 2.
>
> Signed-off-by: Alexey Dobriyan (SK hynix) <adobriyan@xxxxxxxxx>
In addition to Bart and Keith comments, I will add a clear NACK on this, at
least for this patch as is. Details below inline.
> ---
>
> v2: fixup one ">>ilog2"/div conversion :-/
>
> block/blk-settings.c | 4 +---
> block/blk-zoned.c | 10 +++++-----
> drivers/block/null_blk_main.c | 11 ++++++-----
> drivers/block/null_blk_zoned.c | 10 ++--------
> 4 files changed, 14 insertions(+), 21 deletions(-)
>
> --- a/block/blk-settings.c
> +++ b/block/blk-settings.c
> @@ -206,15 +206,13 @@ EXPORT_SYMBOL(blk_queue_max_hw_sectors);
> *
> * Description:
> * If a driver doesn't want IOs to cross a given chunk size, it can set
> - * this limit and prevent merging across chunks. Note that the chunk size
> - * must currently be a power-of-2 in sectors. Also note that the block
> + * this limit and prevent merging across chunks. Note that the block
> * layer must accept a page worth of data at any offset. So if the
> * crossing of chunks is a hard limitation in the driver, it must still be
> * prepared to split single page bios.
> **/
> void blk_queue_chunk_sectors(struct request_queue *q, unsigned int chunk_sectors)
> {
> - BUG_ON(!is_power_of_2(chunk_sectors));
> q->limits.chunk_sectors = chunk_sectors;
> }
> EXPORT_SYMBOL(blk_queue_chunk_sectors);
Chunk sectors is used by the entire bio/request merging and splitting code.
For zoned devices, this ensures that requests do not cross over zones
boundaries due to merges. All that code relies on power of 2 arithmetics
(bit shifts), so allowing a non power of 2 value would mean changing all of
that. That is the hot path ! adding multiplications and divisions will slow
down all devices, including non zoned ones.
Furthermore, chunk sectors is also used to indicate the stripe size of
software or hardware raid arrays for use by file systems and device mappers
to do optimizations and also avoid merging of bios/request accross device
stripe boundaries. Not enforcing a power of 2 here also potentially breaks
all of that.
> --- a/block/blk-zoned.c
> +++ b/block/blk-zoned.c
> @@ -83,7 +83,7 @@ unsigned int blkdev_nr_zones(struct gendisk *disk)
>
> if (!blk_queue_is_zoned(disk->queue))
> return 0;
> - return (get_capacity(disk) + zone_sectors - 1) >> ilog2(zone_sectors);
> + return div64_u64(get_capacity(disk) + zone_sectors - 1, zone_sectors);
> }
> EXPORT_SYMBOL_GPL(blkdev_nr_zones);
Sure, and this is the kind of modification that will be needed all over the
hot path for bio/request merging checks. Not nice at all. Slow down for
everything.
>
> @@ -363,14 +363,14 @@ static int blk_revalidate_zone_cb(struct blk_zone *zone, unsigned int idx,
> * smaller last zone.
> */
> if (zone->start == 0) {
> - if (zone->len == 0 || !is_power_of_2(zone->len)) {
> - pr_warn("%s: Invalid zoned device with non power of two zone size (%llu)\n",
> - disk->disk_name, zone->len);
> + if (zone->len == 0) {
> + pr_warn("%s: Invalid zoned device with length 0\n",
> + disk->disk_name);
> return -ENODEV;
> }
>
> args->zone_sectors = zone->len;
> - args->nr_zones = (capacity + zone->len - 1) >> ilog2(zone->len);
> + args->nr_zones = div64_u64(capacity + zone->len - 1, zone->len);
> } else if (zone->start + args->zone_sectors < capacity) {
> if (zone->len != args->zone_sectors) {
> pr_warn("%s: Invalid zoned device with non constant zone size\n",
You only have these 2 changes for the generic block layer part, but there
are a lot more to add. Anything that has ilog2(chunk_sectors) (e.g.
blk_queue_zone_no()) needs to be changed. mq-deadline and zone write
locking handling, and all bio/request split & merge checks handling. And
again, even with such changes, allowing chunk_sectors to not be a power of
2 will cause problems for non-zoned use cases such as RAID and in all file
systems relying on that value to be a power of 2.
> --- a/drivers/block/null_blk_main.c
> +++ b/drivers/block/null_blk_main.c
> @@ -187,7 +187,7 @@ MODULE_PARM_DESC(zoned, "Make device as a host-managed zoned block device. Defau
>
> static unsigned long g_zone_size = 256;
> module_param_named(zone_size, g_zone_size, ulong, S_IRUGO);
> -MODULE_PARM_DESC(zone_size, "Zone size in MB when block device is zoned. Must be power-of-two: Default: 256");
> +MODULE_PARM_DESC(zone_size, "Zone size in MB when block device is zoned. Default: 256");
>
> static unsigned int g_zone_nr_conv;
> module_param_named(zone_nr_conv, g_zone_nr_conv, uint, 0444);
> @@ -1641,10 +1641,11 @@ static int null_validate_conf(struct nullb_device *dev)
> if (dev->queue_mode == NULL_Q_BIO)
> dev->mbps = 0;
>
> - if (dev->zoned &&
> - (!dev->zone_size || !is_power_of_2(dev->zone_size))) {
> - pr_err("zone_size must be power-of-two\n");
> - return -EINVAL;
> + if (dev->zoned) {
> + if (dev->zone_size == 0) {
> + pr_err("zone_size must be positive\n");
> + return -EINVAL;
> + }
> }
>
> return 0;
> --- a/drivers/block/null_blk_zoned.c
> +++ b/drivers/block/null_blk_zoned.c
> @@ -7,7 +7,7 @@
>
> static inline unsigned int null_zone_no(struct nullb_device *dev, sector_t sect)
> {
> - return sect >> ilog2(dev->zone_size_sects);
> + return div64_u64(sect, dev->zone_size_sects);
> }
>
> int null_zone_init(struct nullb_device *dev)
> @@ -16,14 +16,8 @@ int null_zone_init(struct nullb_device *dev)
> sector_t sector = 0;
> unsigned int i;
>
> - if (!is_power_of_2(dev->zone_size)) {
> - pr_err("zone_size must be power-of-two\n");
> - return -EINVAL;
> - }
> -
> dev->zone_size_sects = dev->zone_size << ZONE_SIZE_SHIFT;
> - dev->nr_zones = dev_size >>
> - (SECTOR_SHIFT + ilog2(dev->zone_size_sects));
> + dev->nr_zones = div64_u64(dev_size, dev->zone_size_sects * SECTOR_SIZE);
> dev->zones = kvmalloc_array(dev->nr_zones, sizeof(struct blk_zone),
> GFP_KERNEL | __GFP_ZERO);
> if (!dev->zones)
>
And beyond null_blk, other drivers are affected:
* DM core, dm-linear, dm-flakey and dm-zoned
* f2fs and zonefs file systems will not work as is on such drive. Ongoing
brtfs work is affected too.
* probably some things get broken in the scsi stack too.
* And user space will suffer badly too, at the very least all user space
tools associated with file systems and DM format/check and sys-utils.
--
Damien Le Moal
Western Digital Research
