bdev = &BDEV_I(file->f_mapping->host)->bdev
Getting struct block_device from a file requires 2 memory dereferences
as illustrated above, that takes a toll on performance, so cache it in
yet unused file->private_data. That gives a noticeable peak performance
improvement.
Signed-off-by: Pavel Begunkov <[email protected]>
---
block/fops.c | 36 +++++++++++++++++++-----------------
1 file changed, 19 insertions(+), 17 deletions(-)
diff --git a/block/fops.c b/block/fops.c
index 765086d51f8b..99e699427f31 100644
--- a/block/fops.c
+++ b/block/fops.c
@@ -17,11 +17,16 @@
#include <linux/fs.h>
#include "blk.h"
-static struct inode *bdev_file_inode(struct file *file)
+static inline struct inode *bdev_file_inode(struct file *file)
{
return file->f_mapping->host;
}
+static inline struct block_device *blkdev_get_bdev(struct file *file)
+{
+ return file->private_data;
+}
+
static int blkdev_get_block(struct inode *inode, sector_t iblock,
struct buffer_head *bh, int create)
{
@@ -54,8 +59,7 @@ static void blkdev_bio_end_io_simple(struct bio *bio)
static ssize_t __blkdev_direct_IO_simple(struct kiocb *iocb,
struct iov_iter *iter, unsigned int nr_pages)
{
- struct file *file = iocb->ki_filp;
- struct block_device *bdev = I_BDEV(bdev_file_inode(file));
+ struct block_device *bdev = blkdev_get_bdev(iocb->ki_filp);
struct bio_vec inline_vecs[DIO_INLINE_BIO_VECS], *vecs;
loff_t pos = iocb->ki_pos;
bool should_dirty = false;
@@ -143,7 +147,7 @@ static struct bio_set blkdev_dio_pool;
static int blkdev_iopoll(struct kiocb *kiocb, struct io_batch *ib, bool wait)
{
- struct block_device *bdev = I_BDEV(kiocb->ki_filp->f_mapping->host);
+ struct block_device *bdev = blkdev_get_bdev(kiocb->ki_filp);
struct request_queue *q = bdev_get_queue(bdev);
return blk_poll(q, READ_ONCE(kiocb->ki_cookie), ib, wait);
@@ -191,9 +195,7 @@ static void blkdev_bio_end_io(struct bio *bio)
static ssize_t __blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
unsigned int nr_pages)
{
- struct file *file = iocb->ki_filp;
- struct inode *inode = bdev_file_inode(file);
- struct block_device *bdev = I_BDEV(inode);
+ struct block_device *bdev = blkdev_get_bdev(iocb->ki_filp);
struct blk_plug plug;
struct blkdev_dio *dio;
struct bio *bio;
@@ -405,8 +407,7 @@ static loff_t blkdev_llseek(struct file *file, loff_t offset, int whence)
static int blkdev_fsync(struct file *filp, loff_t start, loff_t end,
int datasync)
{
- struct inode *bd_inode = bdev_file_inode(filp);
- struct block_device *bdev = I_BDEV(bd_inode);
+ struct block_device *bdev = blkdev_get_bdev(filp);
int error;
error = file_write_and_wait_range(filp, start, end);
@@ -448,6 +449,8 @@ static int blkdev_open(struct inode *inode, struct file *filp)
bdev = blkdev_get_by_dev(inode->i_rdev, filp->f_mode, filp);
if (IS_ERR(bdev))
return PTR_ERR(bdev);
+
+ filp->private_data = bdev;
filp->f_mapping = bdev->bd_inode->i_mapping;
filp->f_wb_err = filemap_sample_wb_err(filp->f_mapping);
return 0;
@@ -455,7 +458,7 @@ static int blkdev_open(struct inode *inode, struct file *filp)
static int blkdev_close(struct inode *inode, struct file *filp)
{
- struct block_device *bdev = I_BDEV(bdev_file_inode(filp));
+ struct block_device *bdev = blkdev_get_bdev(filp);
blkdev_put(bdev, filp->f_mode);
return 0;
@@ -463,7 +466,7 @@ static int blkdev_close(struct inode *inode, struct file *filp)
static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
{
- struct block_device *bdev = I_BDEV(bdev_file_inode(file));
+ struct block_device *bdev = blkdev_get_bdev(file);
fmode_t mode = file->f_mode;
/*
@@ -487,14 +490,14 @@ static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
*/
static ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from)
{
- struct file *file = iocb->ki_filp;
- struct inode *bd_inode = bdev_file_inode(file);
+ struct block_device *bdev = blkdev_get_bdev(iocb->ki_filp);
+ struct inode *bd_inode = bdev->bd_inode;
loff_t size = i_size_read(bd_inode);
struct blk_plug plug;
size_t shorted = 0;
ssize_t ret;
- if (bdev_read_only(I_BDEV(bd_inode)))
+ if (bdev_read_only(bdev))
return -EPERM;
if (IS_SWAPFILE(bd_inode) && !is_hibernate_resume_dev(bd_inode->i_rdev))
@@ -526,9 +529,8 @@ static ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from)
static ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to)
{
- struct file *file = iocb->ki_filp;
- struct inode *bd_inode = bdev_file_inode(file);
- loff_t size = i_size_read(bd_inode);
+ struct block_device *bdev = blkdev_get_bdev(iocb->ki_filp);
+ loff_t size = (loff_t)bdev->bd_nr_sectors << SECTOR_SHIFT;
loff_t pos = iocb->ki_pos;
size_t shorted = 0;
ssize_t ret;
--
2.33.0
On 10/9/21 6:25 AM, Pavel Begunkov wrote:
> bdev = &BDEV_I(file->f_mapping->host)->bdev
>
> Getting struct block_device from a file requires 2 memory dereferences
> as illustrated above, that takes a toll on performance, so cache it in
> yet unused file->private_data. That gives a noticeable peak performance
> improvement.
It's hilariously bad right now, so I really welcome this change. One
comment:
> +static inline struct block_device *blkdev_get_bdev(struct file *file)
> +{
> + return file->private_data;
> +}
Get rid of this and just use bdev = file->private_data where
appropriate. Easier to read, we don't need to hide this in a function.
--
Jens Axboe
On Sat, Oct 09, 2021 at 10:33:17AM -0600, Jens Axboe wrote:
> > +static inline struct block_device *blkdev_get_bdev(struct file *file)
> > +{
> > + return file->private_data;
> > +}
>
> Get rid of this and just use bdev = file->private_data where
> appropriate. Easier to read, we don't need to hide this in a function.
100% agreed.
On 10/11/21 09:26, Christoph Hellwig wrote:
> On Sat, Oct 09, 2021 at 10:33:17AM -0600, Jens Axboe wrote:
>>> +static inline struct block_device *blkdev_get_bdev(struct file *file)
>>> +{
>>> + return file->private_data;
>>> +}
>>
>> Get rid of this and just use bdev = file->private_data where
>> appropriate. Easier to read, we don't need to hide this in a function.
>
> 100% agreed.
The reasoning is as always, it's much easier to change if we change
what we store there. I don't agree, but don't care enough to stay
on the point, will resend with the change
--
Pavel Begunkov