[PATCH v2 1/3] block: provide helpers for reading block count

[PATCH v2 1/3] block: provide helpers for reading block count

[Arnd Bergmann]

Several drivers use an expensive do_div() to compute the number
of logical or physical blocks in a blockdev, which can be done
more efficiently using a shift, since the blocksize is always
a power of two number.

Let's introduce bdev_logical_block_count() and bdev_physical_block_count()
helper functions mirroring the bdev_logical_block_size() and
bdev_physical_block_size() interfaces for the block size.

Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Suggested-by: Christoph Hellwig <hch@infradead.org>
---
 include/linux/blkdev.h | 14 ++++++++++++++
 1 file changed, 14 insertions(+)

diff --git a/include/linux/blkdev.h b/include/linux/blkdev.h
index 9d1e0a4650dc..ae8c408f6c22 100644
--- a/include/linux/blkdev.h
+++ b/include/linux/blkdev.h
@@ -1226,6 +1226,13 @@ static inline unsigned short bdev_logical_block_size(struct block_device *bdev)
 	return queue_logical_block_size(bdev_get_queue(bdev));
 }
 
+static inline sector_t bdev_logical_block_count(struct block_device *bdev)
+{
+	unsigned int block_shift = ilog2(bdev_logical_block_size(bdev));
+
+	return bdev->bd_inode->i_size >> block_shift;
+}
+
 static inline unsigned int queue_physical_block_size(struct request_queue *q)
 {
 	return q->limits.physical_block_size;
@@ -1236,6 +1243,13 @@ static inline unsigned int bdev_physical_block_size(struct block_device *bdev)
 	return queue_physical_block_size(bdev_get_queue(bdev));
 }
 
+static inline sector_t bdev_physical_block_count(struct block_device *bdev)
+{
+	unsigned int block_shift = ilog2(bdev_physical_block_size(bdev));
+
+	return bdev->bd_inode->i_size >> block_shift;
+}
+
 static inline unsigned int queue_io_min(struct request_queue *q)
 {
 	return q->limits.io_min;
-- 
2.9.0

[PATCH v2 1/3] block: provide helpers for reading block count

[Arnd Bergmann]

Several drivers use an expensive do_div() to compute the number
of logical or physical blocks in a blockdev, which can be done
more efficiently using a shift, since the blocksize is always
a power of two number.

Let's introduce bdev_logical_block_count() and bdev_physical_block_count()
helper functions mirroring the bdev_logical_block_size() and
bdev_physical_block_size() interfaces for the block size.

Signed-off-by: Arnd Bergmann <arnd-r2nGTMty4D4@public.gmane.org>
Suggested-by: Christoph Hellwig <hch-wEGCiKHe2LqWVfeAwA7xHQ@public.gmane.org>
---
 include/linux/blkdev.h | 14 ++++++++++++++
 1 file changed, 14 insertions(+)

diff --git a/include/linux/blkdev.h b/include/linux/blkdev.h
index 9d1e0a4650dc..ae8c408f6c22 100644
--- a/include/linux/blkdev.h
+++ b/include/linux/blkdev.h
@@ -1226,6 +1226,13 @@ static inline unsigned short bdev_logical_block_size(struct block_device *bdev)
 	return queue_logical_block_size(bdev_get_queue(bdev));
 }
 
+static inline sector_t bdev_logical_block_count(struct block_device *bdev)
+{
+	unsigned int block_shift = ilog2(bdev_logical_block_size(bdev));
+
+	return bdev->bd_inode->i_size >> block_shift;
+}
+
 static inline unsigned int queue_physical_block_size(struct request_queue *q)
 {
 	return q->limits.physical_block_size;
@@ -1236,6 +1243,13 @@ static inline unsigned int bdev_physical_block_size(struct block_device *bdev)
 	return queue_physical_block_size(bdev_get_queue(bdev));
 }
 
+static inline sector_t bdev_physical_block_count(struct block_device *bdev)
+{
+	unsigned int block_shift = ilog2(bdev_physical_block_size(bdev));
+
+	return bdev->bd_inode->i_size >> block_shift;
+}
+
 static inline unsigned int queue_io_min(struct request_queue *q)
 {
 	return q->limits.io_min;
-- 
2.9.0

[PATCH v2 2/3] partition/efi: use bdev_logical_block_count()

[Arnd Bergmann]

Enabling CONFIG_UBSAN_SANITIZE_ALL on ARM caused a link error:

last_lba.part.0':
:(.text+0xc3440): undefined reference to `____ilog2_NaN'
:(.text+0xc3538): undefined reference to `__aeabi_uldivmod'
:(.text+0xc38e8): undefined reference to `__aeabi_uldivmod'

This is caused by gcc not behaving in the expected ways with __builtin_constant_p(),
but it also points to somewhat inefficient code based on a 64-bit
division.

I have introduced a better bdev_logical_block_count() now, so we
can use that here.

Signed-off-by: Arnd Bergmann <arnd@arndb.de>
---
 block/partitions/efi.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/block/partitions/efi.c b/block/partitions/efi.c
index bcd86e5cd546..7b1a62073d34 100644
--- a/block/partitions/efi.c
+++ b/block/partitions/efi.c
@@ -149,8 +149,8 @@ static u64 last_lba(struct block_device *bdev)
 {
 	if (!bdev || !bdev->bd_inode)
 		return 0;
-	return div_u64(bdev->bd_inode->i_size,
-		       bdev_logical_block_size(bdev)) - 1ULL;
+
+	return bdev_logical_block_count(bdev) - 1;
 }
 
 static inline int pmbr_part_valid(gpt_mbr_record *part)
-- 
2.9.0

[PATCH v2 3/3] target/iblock: use bdev_logical_block_count()

[Arnd Bergmann]

Enabling CONFIG_UBSAN_SANITIZE_ALL on ARM caused a link error:

drivers/target/built-in.o: In function `iblock_emulate_read_cap_with_block_size.constprop.1':
target_core_iblock.c:(.text+0xc2774): undefined reference to `____ilog2_NaN'
target_core_iblock.c:(.text+0xc27f8): undefined reference to `__aeabi_uldivmod'
target_core_iblock.c:(.text+0xc299c): undefined reference to `__aeabi_uldivmod'

This is caused by gcc not behaving in the expected ways with __builtin_constant_p(),
but it also points to somewhat inefficient code based on an expensive 64-bit
division.

I have introduced a better bdev_logical_block_count() now, so we can use that here.

Signed-off-by: Arnd Bergmann <arnd@arndb.de>
---
 drivers/target/target_core_iblock.c | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)

diff --git a/drivers/target/target_core_iblock.c b/drivers/target/target_core_iblock.c
index 22af12f8b8eb..2dc0129553e1 100644
--- a/drivers/target/target_core_iblock.c
+++ b/drivers/target/target_core_iblock.c
@@ -201,9 +201,8 @@ static unsigned long long iblock_emulate_read_cap_with_block_size(
 	struct block_device *bd,
 	struct request_queue *q)
 {
-	unsigned long long blocks_long = (div_u64(i_size_read(bd->bd_inode),
-					bdev_logical_block_size(bd)) - 1);
 	u32 block_size = bdev_logical_block_size(bd);
+	unsigned long long blocks_long = bdev_logical_block_count(bd) - 1;
 
 	if (block_size == dev->dev_attrib.block_size)
 		return blocks_long;
-- 
2.9.0

Re: [PATCH v2 1/3] block: provide helpers for reading block count

[Christoph Hellwig]

Thanks Arnd,

the series looks fine to me:

Reviewed-by: Christoph Hellwig <hch@lst.de>

Re: [PATCH v2 1/3] block: provide helpers for reading block count

[Christoph Hellwig]

Thanks Arnd,

the series looks fine to me:

Reviewed-by: Christoph Hellwig <hch-jcswGhMUV9g@public.gmane.org>

Re: [PATCH v2 1/3] block: provide helpers for reading block count

[Sagi Grimberg]

Looks good, for the series:

Reviewed-by: Sagi Grimbeg <sagi@grimbeg.me>

Re: [PATCH v2 1/3] block: provide helpers for reading block count

[Davidlohr Bueso]

On Thu, 23 Jun 2016, Arnd Bergmann wrote:

>Several drivers use an expensive do_div() to compute the number
>of logical or physical blocks in a blockdev, which can be done
>more efficiently using a shift, since the blocksize is always
>a power of two number.
>
>Let's introduce bdev_logical_block_count() and bdev_physical_block_count()
>helper functions mirroring the bdev_logical_block_size() and
>bdev_physical_block_size() interfaces for the block size.

>@@ -1226,6 +1226,13 @@ static inline unsigned short bdev_logical_block_size(struct block_device *bdev)
> 	return queue_logical_block_size(bdev_get_queue(bdev));
> }
>
>+static inline sector_t bdev_logical_block_count(struct block_device *bdev)

Curious, why not just return u64 instead for all these instead of sector_t (ie
dealing with lba, it reads weird)?

Thanks,
Davidlohr

Re: [PATCH v2 1/3] block: provide helpers for reading block count

[Davidlohr Bueso]

On Thu, 23 Jun 2016, Arnd Bergmann wrote:

>Several drivers use an expensive do_div() to compute the number
>of logical or physical blocks in a blockdev, which can be done
>more efficiently using a shift, since the blocksize is always
>a power of two number.
>
>Let's introduce bdev_logical_block_count() and bdev_physical_block_count()
>helper functions mirroring the bdev_logical_block_size() and
>bdev_physical_block_size() interfaces for the block size.

>@@ -1226,6 +1226,13 @@ static inline unsigned short bdev_logical_block_size(struct block_device *bdev)
> 	return queue_logical_block_size(bdev_get_queue(bdev));
> }
>
>+static inline sector_t bdev_logical_block_count(struct block_device *bdev)

Curious, why not just return u64 instead for all these instead of sector_t (ie
dealing with lba, it reads weird)?

Thanks,
Davidlohr