[V4,1/8] null_blk: allow REQ_OP_WRITE_ZEROES
Commit Message
Add a helper function to enable the REQ_OP_WRITE_ZEROES operations
when null_blk.
Since write-zeroes is a non-trivial I/O operation we need this to
add a blktest so we can test the non-trivial I/O path from the
application to the block layer.
Signed-off-by: Chaitanya Kulkarni <kch@nvidia.com>
---
drivers/block/null_blk/main.c | 59 ++++++++++++++++++++++++++++++-
drivers/block/null_blk/null_blk.h | 1 +
2 files changed, 59 insertions(+), 1 deletion(-)
Comments
On Nov 29, 2022 / 15:28, Chaitanya Kulkarni wrote:
> Add a helper function to enable the REQ_OP_WRITE_ZEROES operations
> when null_blk.
>
> Since write-zeroes is a non-trivial I/O operation we need this to
> add a blktest so we can test the non-trivial I/O path from the
> application to the block layer.
>
> Signed-off-by: Chaitanya Kulkarni <kch@nvidia.com>
This motivation sounds good. I tried this patch. With a quick test it looks
working good for me. Please find minor comments in line.
[...]
> +static void null_zero_sector(struct nullb_device *d, sector_t sect,
> + sector_t nr_sects, bool cache)
> +{
> + struct radix_tree_root *root = cache ? &d->cache : &d->data;
> + struct nullb_page *t_page;
> + unsigned int offset;
> + void *dest;
> +
> + t_page = radix_tree_lookup(root, sect >> PAGE_SECTORS_SHIFT);
> + if (!t_page)
> + return;
> +
> + offset = (sect & SECTOR_MASK) << SECTOR_SHIFT;
> + dest = kmap_atomic(t_page->page);
> + memset(dest + offset, 0, SECTOR_SIZE * nr_sects);
> + kunmap_atomic(dest);
> +}
Did you consider to call null_lookup_page() for __null_lookup_page() from
null_zero_sector()? It may simplify this function a bit.
> +
> static struct nullb_page *null_radix_tree_insert(struct nullb *nullb, u64 idx,
> struct nullb_page *t_page, bool is_cache)
> {
> @@ -1186,6 +1211,27 @@ blk_status_t null_handle_discard(struct nullb_device *dev,
> return BLK_STS_OK;
> }
>
> +static blk_status_t null_handle_write_zeroes(struct nullb_device *dev,
> + sector_t sector, sector_t nr_sectors)
> +{
> + unsigned int bytes_left = nr_sectors << 9;
> + struct nullb *nullb = dev->nullb;
> + size_t curr_bytes;
> +
> + spin_lock_irq(&nullb->lock);
> + while (bytes_left > 0) {
> + curr_bytes = min_t(size_t, bytes_left, nullb->dev->blocksize);
> + nr_sectors = curr_bytes >> SECTOR_SHIFT;
> + null_zero_sector(nullb->dev, sector, nr_sectors, false);
> + if (null_cache_active(nullb))
> + null_zero_sector(nullb->dev, sector, nr_sectors, true);
> + sector += nr_sectors;
> + bytes_left -= curr_bytes;
> + }
> + spin_unlock_irq(&nullb->lock);
> + return BLK_STS_OK;
> +}
> +
> static int null_handle_flush(struct nullb *nullb)
> {
> int err;
> @@ -1352,6 +1398,9 @@ static inline blk_status_t null_handle_memory_backed(struct nullb_cmd *cmd,
> if (op == REQ_OP_DISCARD)
> return null_handle_discard(dev, sector, nr_sectors);
>
> + if (op == REQ_OP_WRITE_ZEROES)
> + return null_handle_write_zeroes(dev, sector, nr_sectors);
> +
> if (dev->queue_mode == NULL_Q_BIO)
> err = null_handle_bio(cmd);
> else
> @@ -1800,6 +1849,13 @@ static void null_config_discard(struct nullb *nullb)
> blk_queue_max_discard_sectors(nullb->q, UINT_MAX >> 9);
> }
>
> +static void null_config_write_zeroes(struct nullb *nullb)
> +{
> + if (!nullb->dev->write_zeroes)
> + return;
> + blk_queue_max_write_zeroes_sectors(nullb->q, UINT_MAX >> 9);
Just comment: this value UINT_MAX >> 9 sounds a bit weird, but probably ok. This
value was introduced by commit 306eb6b4ad4f ("nullb: support discard") to call
blk_queue_max_discard_sectors(). I guess you chose the same value for write
zeroes.
> +}
> +
> static const struct block_device_operations null_bio_ops = {
> .owner = THIS_MODULE,
> .submit_bio = null_submit_bio,
> @@ -2111,6 +2167,7 @@ static int null_add_dev(struct nullb_device *dev)
> blk_queue_virt_boundary(nullb->q, PAGE_SIZE - 1);
>
> null_config_discard(nullb);
> + null_config_write_zeroes(nullb);
>
> if (config_item_name(&dev->item)) {
> /* Use configfs dir name as the device name */
> diff --git a/drivers/block/null_blk/null_blk.h b/drivers/block/null_blk/null_blk.h
> index 94ff68052b1e..2c0c9c29158f 100644
> --- a/drivers/block/null_blk/null_blk.h
> +++ b/drivers/block/null_blk/null_blk.h
> @@ -111,6 +111,7 @@ struct nullb_device {
> bool power; /* power on/off the device */
> bool memory_backed; /* if data is stored in memory */
> bool discard; /* if support discard */
> + bool write_zeroes; /* if support write_zeroes */
> bool zoned; /* if device is zoned */
> bool virt_boundary; /* virtual boundary on/off for the device */
> bool no_sched; /* no IO scheduler for the device */
> --
> 2.29.0
>
>> Signed-off-by: Chaitanya Kulkarni <kch@nvidia.com>
>
> This motivation sounds good. I tried this patch. With a quick test it looks
> working good for me. Please find minor comments in line.
>
> [...]
>
>> +static void null_zero_sector(struct nullb_device *d, sector_t sect,
>> + sector_t nr_sects, bool cache)
>> +{
>> + struct radix_tree_root *root = cache ? &d->cache : &d->data;
>> + struct nullb_page *t_page;
>> + unsigned int offset;
>> + void *dest;
>> +
>> + t_page = radix_tree_lookup(root, sect >> PAGE_SECTORS_SHIFT);
>> + if (!t_page)
>> + return;
>> +
>> + offset = (sect & SECTOR_MASK) << SECTOR_SHIFT;
>> + dest = kmap_atomic(t_page->page);
>> + memset(dest + offset, 0, SECTOR_SIZE * nr_sects);
>> + kunmap_atomic(dest);
>> +}
>
> Did you consider to call null_lookup_page() for __null_lookup_page() from
> null_zero_sector()? It may simplify this function a bit.
>
I found this clear and easy than going over call chain when
debugging.
>> +
>> static struct nullb_page *null_radix_tree_insert(struct nullb *nullb, u64 idx,
[...]
>>
>> +static void null_config_write_zeroes(struct nullb *nullb)
>> +{
>> + if (!nullb->dev->write_zeroes)
>> + return;
>> + blk_queue_max_write_zeroes_sectors(nullb->q, UINT_MAX >> 9);
>
> Just comment: this value UINT_MAX >> 9 sounds a bit weird, but probably ok. This
> value was introduced by commit 306eb6b4ad4f ("nullb: support discard") to call
> blk_queue_max_discard_sectors(). I guess you chose the same value for write
> zeroes.
>
Yes indeed, plz have a look end patches to allow user to set this value.
-ck
On Nov 30, 2022 / 23:29, Chaitanya Kulkarni wrote:
> >> Signed-off-by: Chaitanya Kulkarni <kch@nvidia.com>
> >
> > This motivation sounds good. I tried this patch. With a quick test it looks
> > working good for me. Please find minor comments in line.
> >
> > [...]
> >
> >> +static void null_zero_sector(struct nullb_device *d, sector_t sect,
> >> + sector_t nr_sects, bool cache)
> >> +{
> >> + struct radix_tree_root *root = cache ? &d->cache : &d->data;
> >> + struct nullb_page *t_page;
> >> + unsigned int offset;
> >> + void *dest;
> >> +
> >> + t_page = radix_tree_lookup(root, sect >> PAGE_SECTORS_SHIFT);
> >> + if (!t_page)
> >> + return;
> >> +
> >> + offset = (sect & SECTOR_MASK) << SECTOR_SHIFT;
> >> + dest = kmap_atomic(t_page->page);
> >> + memset(dest + offset, 0, SECTOR_SIZE * nr_sects);
> >> + kunmap_atomic(dest);
> >> +}
> >
> > Did you consider to call null_lookup_page() for __null_lookup_page() from
> > null_zero_sector()? It may simplify this function a bit.
> >
>
> I found this clear and easy than going over call chain when
> debugging.
Okay, then the patch looks good to me :)
Reviewed-by: Shin'ichiro Kawasaki <shinichiro.kawasaki@wdc.com>
@@ -209,6 +209,10 @@ static bool g_discard;
module_param_named(discard, g_discard, bool, 0444);
MODULE_PARM_DESC(discard, "Support discard operations (requires memory-backed null_blk device). Default: false");
+static bool g_write_zeroes;
+module_param_named(write_zeroes, g_write_zeroes, bool, 0444);
+MODULE_PARM_DESC(write_zeroes, "Support write-zeores operations. Default: false");
+
static unsigned long g_cache_size;
module_param_named(cache_size, g_cache_size, ulong, 0444);
MODULE_PARM_DESC(mbps, "Cache size in MiB for memory-backed device. Default: 0 (none)");
@@ -416,6 +420,7 @@ NULLB_DEVICE_ATTR(blocking, bool, NULL);
NULLB_DEVICE_ATTR(use_per_node_hctx, bool, NULL);
NULLB_DEVICE_ATTR(memory_backed, bool, NULL);
NULLB_DEVICE_ATTR(discard, bool, NULL);
+NULLB_DEVICE_ATTR(write_zeroes, bool, NULL);
NULLB_DEVICE_ATTR(mbps, uint, NULL);
NULLB_DEVICE_ATTR(cache_size, ulong, NULL);
NULLB_DEVICE_ATTR(zoned, bool, NULL);
@@ -540,6 +545,7 @@ static struct configfs_attribute *nullb_device_attrs[] = {
&nullb_device_attr_power,
&nullb_device_attr_memory_backed,
&nullb_device_attr_discard,
+ &nullb_device_attr_write_zeroes,
&nullb_device_attr_mbps,
&nullb_device_attr_cache_size,
&nullb_device_attr_badblocks,
@@ -614,7 +620,7 @@ static ssize_t memb_group_features_show(struct config_item *item, char *page)
"poll_queues,power,queue_mode,shared_tag_bitmap,size,"
"submit_queues,use_per_node_hctx,virt_boundary,zoned,"
"zone_capacity,zone_max_active,zone_max_open,"
- "zone_nr_conv,zone_size\n");
+ "zone_nr_conv,zone_size,write_zeroes\n");
}
CONFIGFS_ATTR_RO(memb_group_, features);
@@ -678,6 +684,7 @@ static struct nullb_device *null_alloc_dev(void)
dev->blocking = g_blocking;
dev->memory_backed = g_memory_backed;
dev->discard = g_discard;
+ dev->write_zeroes = g_write_zeroes;
dev->cache_size = g_cache_size;
dev->mbps = g_mbps;
dev->use_per_node_hctx = g_use_per_node_hctx;
@@ -870,6 +877,24 @@ static void null_free_sector(struct nullb *nullb, sector_t sector,
}
}
+static void null_zero_sector(struct nullb_device *d, sector_t sect,
+ sector_t nr_sects, bool cache)
+{
+ struct radix_tree_root *root = cache ? &d->cache : &d->data;
+ struct nullb_page *t_page;
+ unsigned int offset;
+ void *dest;
+
+ t_page = radix_tree_lookup(root, sect >> PAGE_SECTORS_SHIFT);
+ if (!t_page)
+ return;
+
+ offset = (sect & SECTOR_MASK) << SECTOR_SHIFT;
+ dest = kmap_atomic(t_page->page);
+ memset(dest + offset, 0, SECTOR_SIZE * nr_sects);
+ kunmap_atomic(dest);
+}
+
static struct nullb_page *null_radix_tree_insert(struct nullb *nullb, u64 idx,
struct nullb_page *t_page, bool is_cache)
{
@@ -1186,6 +1211,27 @@ blk_status_t null_handle_discard(struct nullb_device *dev,
return BLK_STS_OK;
}
+static blk_status_t null_handle_write_zeroes(struct nullb_device *dev,
+ sector_t sector, sector_t nr_sectors)
+{
+ unsigned int bytes_left = nr_sectors << 9;
+ struct nullb *nullb = dev->nullb;
+ size_t curr_bytes;
+
+ spin_lock_irq(&nullb->lock);
+ while (bytes_left > 0) {
+ curr_bytes = min_t(size_t, bytes_left, nullb->dev->blocksize);
+ nr_sectors = curr_bytes >> SECTOR_SHIFT;
+ null_zero_sector(nullb->dev, sector, nr_sectors, false);
+ if (null_cache_active(nullb))
+ null_zero_sector(nullb->dev, sector, nr_sectors, true);
+ sector += nr_sectors;
+ bytes_left -= curr_bytes;
+ }
+ spin_unlock_irq(&nullb->lock);
+ return BLK_STS_OK;
+}
+
static int null_handle_flush(struct nullb *nullb)
{
int err;
@@ -1352,6 +1398,9 @@ static inline blk_status_t null_handle_memory_backed(struct nullb_cmd *cmd,
if (op == REQ_OP_DISCARD)
return null_handle_discard(dev, sector, nr_sectors);
+ if (op == REQ_OP_WRITE_ZEROES)
+ return null_handle_write_zeroes(dev, sector, nr_sectors);
+
if (dev->queue_mode == NULL_Q_BIO)
err = null_handle_bio(cmd);
else
@@ -1800,6 +1849,13 @@ static void null_config_discard(struct nullb *nullb)
blk_queue_max_discard_sectors(nullb->q, UINT_MAX >> 9);
}
+static void null_config_write_zeroes(struct nullb *nullb)
+{
+ if (!nullb->dev->write_zeroes)
+ return;
+ blk_queue_max_write_zeroes_sectors(nullb->q, UINT_MAX >> 9);
+}
+
static const struct block_device_operations null_bio_ops = {
.owner = THIS_MODULE,
.submit_bio = null_submit_bio,
@@ -2111,6 +2167,7 @@ static int null_add_dev(struct nullb_device *dev)
blk_queue_virt_boundary(nullb->q, PAGE_SIZE - 1);
null_config_discard(nullb);
+ null_config_write_zeroes(nullb);
if (config_item_name(&dev->item)) {
/* Use configfs dir name as the device name */
@@ -111,6 +111,7 @@ struct nullb_device {
bool power; /* power on/off the device */
bool memory_backed; /* if data is stored in memory */
bool discard; /* if support discard */
+ bool write_zeroes; /* if support write_zeroes */
bool zoned; /* if device is zoned */
bool virt_boundary; /* virtual boundary on/off for the device */
bool no_sched; /* no IO scheduler for the device */