[PATCH V2] blk-mq: add blk_rq_nr_bvec() helper

[PATCH V2] blk-mq: add blk_rq_nr_bvec() helper

[Chaitanya Kulkarni]

Add a new helper function blk_rq_nr_bvec() that returns the number of
bvecs in a request. This count represents the number of iterations
rq_for_each_bvec() would perform on a request.

Drivers need to pre-allocate bvec arrays before iterating through
a request's bvecs. Currently, they manually count bvecs using
rq_for_each_bvec() in a loop, which is repetitive. The new helper
centralizes this logic.

This pattern exists in loop and zloop drivers, where multi-bio requests
require copying bvecs into a contiguous array before creating
an iov_iter for file operations.

Update loop and zloop drivers to use the new helper, eliminating
duplicate code.

This patch also provides a clear API to avoid any potential misuse of
blk_nr_phys_segments() for calculating the bvecs since, one bvec can
have more than one segments and use of blk_nr_phys_segments() can
lead to extra memory allocation :-

[ 6155.673749] nullb_bio: 128K bio as ONE bvec: sector=0, size=131072
[ 6155.673846] null_blk: #### null_handle_data_transfer:1375
[ 6155.673850] null_blk: nr_bvec=1 blk_rq_nr_phys_segments=2
[ 6155.674263] null_blk: #### null_handle_data_transfer:1375
[ 6155.674267] null_blk: nr_bvec=1 blk_rq_nr_phys_segments=1

Signed-off-by: Chaitanya Kulkarni <ckulkarnilinux@gmail.com>
---

Hi,

During bio submission, the block layer may split a single bvec into
multiple physical segments based on device limits:
    submit_bio()
      -> submit_bio_noacct()
        -> __submit_bio_noacct()
          -> __submit_bio()
            -> blk_mq_submit_bio()
              -> __bio_split_to_limits()
                -> bio_split_rw()
                  -> bio_split_rw_at()
                    -> bio_split_io_at()
                      -> bio_for_each_bvec()
                        -> [Fast path] nsegs++ (1 bvec = 1 segment)
                        -> [Slow path] bvec_split_segs()

The bvec_split_segs() function handles the case where a single bvec must
be split into multiple segments:

    /**
     * bvec_split_segs - verify whether or not a bvec should be split in the
     *                   middle
     * ...
     * When splitting a bio, it can happen that a bvec is encountered that is
     * too big to fit in a single segment and hence that it has to be split in
     * the middle.
     */
    static bool bvec_split_segs(...)
    {
        while (len && *nsegs < max_segs) {
            seg_size = get_max_segment_size(...);
            (*nsegs)++;
            total_len += seg_size;
            len -= seg_size;
        }
        *bytes += total_len;
        return (len > 0);  // True if bvec was split
    }

Splitting occurs when a bvec exceeds:
- max_segment_size
- segment_boundary_mask (DMA boundary constraints)
- max_segments limit

Result after bio_split_io_at():
- nr_bvec (what rq_for_each_bvec iterates): **1**
- rq->nr_phys_segments: 2

*[ 6155.673749] nullb_bio: 128K bio as ONE bvec: sector=0, size=131072
*[ 6155.673846] null_blk: #### null_handle_data_transfer:1375*
*[ 6155.673850] null_blk: nr_bvec=1 blk_rq_nr_phys_segments=2*
*[ 6155.674263] null_blk: #### null_handle_data_transfer:1375*
*[ 6155.674267] null_blk: nr_bvec=1 blk_rq_nr_phys_segments=1*

-ck

---
 drivers/block/loop.c   |  5 ++---
 drivers/block/zloop.c  |  5 ++---
 include/linux/blk-mq.h | 18 ++++++++++++++++++
 3 files changed, 22 insertions(+), 6 deletions(-)

diff --git a/drivers/block/loop.c b/drivers/block/loop.c
index 13ce229d450c..7b716d759168 100644
--- a/drivers/block/loop.c
+++ b/drivers/block/loop.c
@@ -348,11 +348,10 @@ static int lo_rw_aio(struct loop_device *lo, struct loop_cmd *cmd,
 	struct file *file = lo->lo_backing_file;
 	struct bio_vec tmp;
 	unsigned int offset;
-	int nr_bvec = 0;
+	unsigned int nr_bvec;
 	int ret;
 
-	rq_for_each_bvec(tmp, rq, rq_iter)
-		nr_bvec++;
+	nr_bvec = blk_rq_nr_bvec(rq);
 
 	if (rq->bio != rq->biotail) {
 
diff --git a/drivers/block/zloop.c b/drivers/block/zloop.c
index 92be9f0af00a..857a8de61088 100644
--- a/drivers/block/zloop.c
+++ b/drivers/block/zloop.c
@@ -370,7 +370,7 @@ static void zloop_rw(struct zloop_cmd *cmd)
 	struct iov_iter iter;
 	struct bio_vec tmp;
 	sector_t zone_end;
-	int nr_bvec = 0;
+	unsigned int nr_bvec;
 	int ret;
 
 	atomic_set(&cmd->ref, 2);
@@ -437,8 +437,7 @@ static void zloop_rw(struct zloop_cmd *cmd)
 			zone->cond = BLK_ZONE_COND_FULL;
 	}
 
-	rq_for_each_bvec(tmp, rq, rq_iter)
-		nr_bvec++;
+	nr_bvec = blk_rq_nr_bvec(rq);
 
 	if (rq->bio != rq->biotail) {
 		struct bio_vec *bvec;
diff --git a/include/linux/blk-mq.h b/include/linux/blk-mq.h
index b25d12545f46..7cedc0eba561 100644
--- a/include/linux/blk-mq.h
+++ b/include/linux/blk-mq.h
@@ -1185,6 +1185,24 @@ static inline unsigned short blk_rq_nr_discard_segments(struct request *rq)
 	return max_t(unsigned short, rq->nr_phys_segments, 1);
 }
 
+/**
+ * blk_rq_nr_bvec - return number of bvecs in a request
+ * @rq: request to calculate bvecs for
+ *
+ * Returns the number of bvecs.
+ */
+static inline unsigned int blk_rq_nr_bvec(struct request *rq)
+{
+	struct req_iterator rq_iter;
+	struct bio_vec bv;
+	unsigned int nr_bvec = 0;
+
+	rq_for_each_bvec(bv, rq, rq_iter)
+		nr_bvec++;
+
+	return nr_bvec;
+}
+
 int __blk_rq_map_sg(struct request *rq, struct scatterlist *sglist,
 		struct scatterlist **last_sg);
 static inline int blk_rq_map_sg(struct request *rq, struct scatterlist *sglist)
-- 
2.40.0

Re: [PATCH V2] blk-mq: add blk_rq_nr_bvec() helper

[Keith Busch]

On Tue, Nov 11, 2025 at 03:22:52PM -0800, Chaitanya Kulkarni wrote:
> This patch also provides a clear API to avoid any potential misuse of
> blk_nr_phys_segments() for calculating the bvecs since, one bvec can
> have more than one segments and use of blk_nr_phys_segments() can
> lead to extra memory allocation :-

Perhaps blk_nr_phys_segments is misnamed as it represents device
segments, not physical host memory segments. Instead of rewalking to
calculate physical segments, maybe just introduce a new field into the
request so that we can save both the physical and device segments to
avoid having to repeatedly rewalk the same list. There is an ongoing
effort to avoid such repeated work.

Re: [PATCH V2] blk-mq: add blk_rq_nr_bvec() helper

[Chaitanya Kulkarni]

On 11/11/25 19:08, Keith Busch wrote:
> On Tue, Nov 11, 2025 at 03:22:52PM -0800, Chaitanya Kulkarni wrote:
>> This patch also provides a clear API to avoid any potential misuse of
>> blk_nr_phys_segments() for calculating the bvecs since, one bvec can
>> have more than one segments and use of blk_nr_phys_segments() can
>> lead to extra memory allocation :-
> Perhaps blk_nr_phys_segments is misnamed as it represents device
> segments, not physical host memory segments. Instead of rewalking to
> calculate physical segments, maybe just introduce a new field into the
> request so that we can save both the physical and device segments to
> avoid having to repeatedly rewalk the same list. There is an ongoing
> effort to avoid such repeated work.

I like the idea, how about something like this on the top of this patch.

Totally untested, incomplete, only good for conceptual discussion :-

  block/blk-merge.c      | 10 ++++++++--
  block/blk-mq.c         |  2 ++
  include/linux/blk-mq.h | 13 +++----------
  3 files changed, 13 insertions(+), 12 deletions(-)

diff --git a/block/blk-merge.c b/block/blk-merge.c
index c47d18587a0b..40b7ba3b6216 100644
--- a/block/blk-merge.c
+++ b/block/blk-merge.c
@@ -458,6 +458,7 @@ EXPORT_SYMBOL(bio_split_to_limits);
  unsigned int blk_recalc_rq_segments(struct request *rq)
  {
  	unsigned int nr_phys_segs = 0;
+	unsigned int nr_bvecs = 0;
  	unsigned int bytes = 0;
  	struct req_iterator iter;
  	struct bio_vec bv;
@@ -482,9 +483,12 @@ unsigned int blk_recalc_rq_segments(struct request *rq)
  		break;
  	}

-	rq_for_each_bvec(bv, rq, iter)
+	rq_for_each_bvec(bv, rq, iter) {
  		bvec_split_segs(&rq->q->limits, &bv, &nr_phys_segs, &bytes,
  				UINT_MAX, UINT_MAX);
+		nr_bvecs++;
+	}
+	rq->nr_bvecs = nr_bvecs;
  	return nr_phys_segs;
  }

@@ -531,6 +535,7 @@ static inline int ll_new_hw_segment(struct request *req, struct bio *bio,
  	 * counters.
  	 */
  	req->nr_phys_segments += nr_phys_segs;
+	req->nr_bvecs += bio->bi_vcnt;
  	if (bio_integrity(bio))
  		req->nr_integrity_segments += blk_rq_count_integrity_sg(req->q,
  									bio);
@@ -626,6 +631,7 @@ static int ll_merge_requests_fn(struct request_queue *q, struct request *req,

  	/* Merge is OK... */
  	req->nr_phys_segments = total_phys_segments;
+	req->nr_bvecs += next->nr_bvecs;
  	req->nr_integrity_segments += next->nr_integrity_segments;
  	return 1;
  }
diff --git a/block/blk-mq.c b/block/blk-mq.c
index d626d32f6e57..61cd0bd5b5af 100644
--- a/block/blk-mq.c
+++ b/block/blk-mq.c
@@ -436,6 +436,7 @@ static struct request *blk_mq_rq_ctx_init(struct blk_mq_alloc_data *data,
  	rq->stats_sectors = 0;
  	rq->nr_phys_segments = 0;
  	rq->nr_integrity_segments = 0;
+	rq->nr_bvecs = 0;
  	rq->end_io = NULL;
  	rq->end_io_data = NULL;

@@ -2675,6 +2676,7 @@ static void blk_mq_bio_to_request(struct request *rq, struct bio *bio,
  	rq->__sector = bio->bi_iter.bi_sector;
  	rq->__data_len = bio->bi_iter.bi_size;
  	rq->nr_phys_segments = nr_segs;
+	rq->nr_bvecs = bio->bi_vcnt;
  	if (bio_integrity(bio))
  		rq->nr_integrity_segments = blk_rq_count_integrity_sg(rq->q,
  								      bio);
diff --git a/include/linux/blk-mq.h b/include/linux/blk-mq.h
index 7cedc0eba561..c563255586c6 100644
--- a/include/linux/blk-mq.h
+++ b/include/linux/blk-mq.h
@@ -151,6 +151,7 @@ struct request {
  	 */
  	unsigned short nr_phys_segments;
  	unsigned short nr_integrity_segments;
+	unsigned short nr_bvecs;

  #ifdef CONFIG_BLK_INLINE_ENCRYPTION
  	struct bio_crypt_ctx *crypt_ctx;
@@ -1187,20 +1188,14 @@ static inline unsigned short blk_rq_nr_discard_segments(struct request *rq)

  /**
   * blk_rq_nr_bvec - return number of bvecs in a request
- * @rq: request to calculate bvecs for
+ * @rq: request to get bvec count for
   *
- * Returns the number of bvecs.
+ * Returns the number of bvecs in the request. This count is maintained
+ * during request initialization and merging to avoid repeated rewalking.
   */
  static inline unsigned int blk_rq_nr_bvec(struct request *rq)
  {
-	struct req_iterator rq_iter;
-	struct bio_vec bv;
-	unsigned int nr_bvec = 0;
-
-	rq_for_each_bvec(bv, rq, rq_iter)
-		nr_bvec++;
-
-	return nr_bvec;
+	return rq->nr_bvecs;
  }

  int __blk_rq_map_sg(struct request *rq, struct scatterlist *sglist,
--


-ck

Re: [PATCH V2] blk-mq: add blk_rq_nr_bvec() helper

[Ming Lei]

On Wed, Nov 12, 2025 at 12:11 PM Chaitanya Kulkarni
<chaitanyak@nvidia.com> wrote:
>
> On 11/11/25 19:08, Keith Busch wrote:
> > On Tue, Nov 11, 2025 at 03:22:52PM -0800, Chaitanya Kulkarni wrote:
> >> This patch also provides a clear API to avoid any potential misuse of
> >> blk_nr_phys_segments() for calculating the bvecs since, one bvec can
> >> have more than one segments and use of blk_nr_phys_segments() can
> >> lead to extra memory allocation :-
> > Perhaps blk_nr_phys_segments is misnamed as it represents device
> > segments, not physical host memory segments. Instead of rewalking to
> > calculate physical segments, maybe just introduce a new field into the
> > request so that we can save both the physical and device segments to
> > avoid having to repeatedly rewalk the same list. There is an ongoing
> > effort to avoid such repeated work.
>
> I like the idea, how about something like this on the top of this patch.
>
> Totally untested, incomplete, only good for conceptual discussion :-
>
>   block/blk-merge.c      | 10 ++++++++--
>   block/blk-mq.c         |  2 ++
>   include/linux/blk-mq.h | 13 +++----------
>   3 files changed, 13 insertions(+), 12 deletions(-)
>
> diff --git a/block/blk-merge.c b/block/blk-merge.c
> index c47d18587a0b..40b7ba3b6216 100644
> --- a/block/blk-merge.c
> +++ b/block/blk-merge.c
> @@ -458,6 +458,7 @@ EXPORT_SYMBOL(bio_split_to_limits);
>   unsigned int blk_recalc_rq_segments(struct request *rq)
>   {
>         unsigned int nr_phys_segs = 0;
> +       unsigned int nr_bvecs = 0;
>         unsigned int bytes = 0;
>         struct req_iterator iter;
>         struct bio_vec bv;
> @@ -482,9 +483,12 @@ unsigned int blk_recalc_rq_segments(struct request *rq)
>                 break;
>         }
>
> -       rq_for_each_bvec(bv, rq, iter)
> +       rq_for_each_bvec(bv, rq, iter) {
>                 bvec_split_segs(&rq->q->limits, &bv, &nr_phys_segs, &bytes,
>                                 UINT_MAX, UINT_MAX);
> +               nr_bvecs++;
> +       }
> +       rq->nr_bvecs = nr_bvecs;
>         return nr_phys_segs;
>   }
>
> @@ -531,6 +535,7 @@ static inline int ll_new_hw_segment(struct request *req, struct bio *bio,
>          * counters.
>          */
>         req->nr_phys_segments += nr_phys_segs;
> +       req->nr_bvecs += bio->bi_vcnt;
>         if (bio_integrity(bio))
>                 req->nr_integrity_segments += blk_rq_count_integrity_sg(req->q,
>                                                                         bio);
> @@ -626,6 +631,7 @@ static int ll_merge_requests_fn(struct request_queue *q, struct request *req,
>
>         /* Merge is OK... */
>         req->nr_phys_segments = total_phys_segments;
> +       req->nr_bvecs += next->nr_bvecs;
>         req->nr_integrity_segments += next->nr_integrity_segments;
>         return 1;
>   }
> diff --git a/block/blk-mq.c b/block/blk-mq.c
> index d626d32f6e57..61cd0bd5b5af 100644
> --- a/block/blk-mq.c
> +++ b/block/blk-mq.c
> @@ -436,6 +436,7 @@ static struct request *blk_mq_rq_ctx_init(struct blk_mq_alloc_data *data,
>         rq->stats_sectors = 0;
>         rq->nr_phys_segments = 0;
>         rq->nr_integrity_segments = 0;
> +       rq->nr_bvecs = 0;
>         rq->end_io = NULL;
>         rq->end_io_data = NULL;
>
> @@ -2675,6 +2676,7 @@ static void blk_mq_bio_to_request(struct request *rq, struct bio *bio,
>         rq->__sector = bio->bi_iter.bi_sector;
>         rq->__data_len = bio->bi_iter.bi_size;
>         rq->nr_phys_segments = nr_segs;
> +       rq->nr_bvecs = bio->bi_vcnt;

Please do not use bio->bi_vcnt, which should only be available for FS bio code.

It is obviously wrong to touch it for cloned bio...

Thanks,
Ming

Re: [PATCH V2] blk-mq: add blk_rq_nr_bvec() helper

[Chaitanya Kulkarni]

On 11/11/25 20:49, Ming Lei wrote:
>> @@ -2675,6 +2676,7 @@ static void blk_mq_bio_to_request(struct request *rq, struct bio *bio,
>>          rq->__sector = bio->bi_iter.bi_sector;
>>          rq->__data_len = bio->bi_iter.bi_size;
>>          rq->nr_phys_segments = nr_segs;
>> +       rq->nr_bvecs = bio->bi_vcnt;
> Please do not use bio->bi_vcnt, which should only be available for FS bio code.
>
> It is obviously wrong to touch it for cloned bio...
>
> Thanks,
> Ming

yes, it's wrong and above patch is not functionally correct, purely for 
conceptual discussion, to add a member to struct request if acceptable 
... it's going to increase the size of struct request, but if we can 
live with it without significant performance cost or adding a new 
cacheline access maybe it is worth it ? -ck

Re: [PATCH V2] blk-mq: add blk_rq_nr_bvec() helper

[Jens Axboe]

On 11/11/25 8:08 PM, Keith Busch wrote:
> On Tue, Nov 11, 2025 at 03:22:52PM -0800, Chaitanya Kulkarni wrote:
>> This patch also provides a clear API to avoid any potential misuse of
>> blk_nr_phys_segments() for calculating the bvecs since, one bvec can
>> have more than one segments and use of blk_nr_phys_segments() can
>> lead to extra memory allocation :-
> 
> Perhaps blk_nr_phys_segments is misnamed as it represents device
> segments, not physical host memory segments. Instead of rewalking to
> calculate physical segments, maybe just introduce a new field into the
> request so that we can save both the physical and device segments to
> avoid having to repeatedly rewalk the same list. There is an ongoing
> effort to avoid such repeated work.

Is it used in the fast path? From the initial patch, I only see
loop/zloop using it, and that's certainly not enough to warrant
further bloating struct request. Just have them iterate the
darn thing, it's not a huge deal.

-- 
Jens Axboe

Re: [PATCH V2] blk-mq: add blk_rq_nr_bvec() helper

[Chaitanya Kulkarni]

On 11/12/25 06:00, Jens Axboe wrote:
> On 11/11/25 8:08 PM, Keith Busch wrote:
>> On Tue, Nov 11, 2025 at 03:22:52PM -0800, Chaitanya Kulkarni wrote:
>>> This patch also provides a clear API to avoid any potential misuse of
>>> blk_nr_phys_segments() for calculating the bvecs since, one bvec can
>>> have more than one segments and use of blk_nr_phys_segments() can
>>> lead to extra memory allocation :-
>> Perhaps blk_nr_phys_segments is misnamed as it represents device
>> segments, not physical host memory segments. Instead of rewalking to
>> calculate physical segments, maybe just introduce a new field into the
>> request so that we can save both the physical and device segments to
>> avoid having to repeatedly rewalk the same list. There is an ongoing
>> effort to avoid such repeated work.
> Is it used in the fast path? From the initial patch, I only see
> loop/zloop using it, and that's certainly not enough to warrant
> further bloating struct request. Just have them iterate the
> darn thing, it's not a huge deal.
>
okay, will keep the V2 as is to get reviews.

-ck

Re: [PATCH V2] blk-mq: add blk_rq_nr_bvec() helper

[Niklas Cassel]

On Tue, Nov 11, 2025 at 03:22:52PM -0800, Chaitanya Kulkarni wrote:
> Add a new helper function blk_rq_nr_bvec() that returns the number of
> bvecs in a request. This count represents the number of iterations
> rq_for_each_bvec() would perform on a request.
> 
> Drivers need to pre-allocate bvec arrays before iterating through
> a request's bvecs. Currently, they manually count bvecs using
> rq_for_each_bvec() in a loop, which is repetitive. The new helper
> centralizes this logic.
> 
> This pattern exists in loop and zloop drivers, where multi-bio requests
> require copying bvecs into a contiguous array before creating
> an iov_iter for file operations.
> 
> Update loop and zloop drivers to use the new helper, eliminating
> duplicate code.
> 
> This patch also provides a clear API to avoid any potential misuse of
> blk_nr_phys_segments() for calculating the bvecs since, one bvec can
> have more than one segments and use of blk_nr_phys_segments() can
> lead to extra memory allocation :-
> 
> [ 6155.673749] nullb_bio: 128K bio as ONE bvec: sector=0, size=131072
> [ 6155.673846] null_blk: #### null_handle_data_transfer:1375
> [ 6155.673850] null_blk: nr_bvec=1 blk_rq_nr_phys_segments=2
> [ 6155.674263] null_blk: #### null_handle_data_transfer:1375
> [ 6155.674267] null_blk: nr_bvec=1 blk_rq_nr_phys_segments=1
> 
> Signed-off-by: Chaitanya Kulkarni <ckulkarnilinux@gmail.com>
> ---

What is the status of this patch?


FWIW, looks good to me:
Reviewed-by: Niklas Cassel <cassel@kernel.org>

Re: [PATCH V2] blk-mq: add blk_rq_nr_bvec() helper

[Chaitanya Kulkarni]

On 12/1/25 01:54, Niklas Cassel wrote:
>> This patch also provides a clear API to avoid any potential misuse of
>> blk_nr_phys_segments() for calculating the bvecs since, one bvec can
>> have more than one segments and use of blk_nr_phys_segments() can
>> lead to extra memory allocation :-
>>
>> [ 6155.673749] nullb_bio: 128K bio as ONE bvec: sector=0, size=131072
>> [ 6155.673846] null_blk: #### null_handle_data_transfer:1375
>> [ 6155.673850] null_blk: nr_bvec=1 blk_rq_nr_phys_segments=2
>> [ 6155.674263] null_blk: #### null_handle_data_transfer:1375
>> [ 6155.674267] null_blk: nr_bvec=1 blk_rq_nr_phys_segments=1
>>
>> Signed-off-by: Chaitanya Kulkarni<ckulkarnilinux@gmail.com>
>> ---
> What is the status of this patch?
>
>
> FWIW, looks good to me:
> Reviewed-by: Niklas Cassel<cassel@kernel.org>

Thanks Niklas, it's not been merged yet.

I'll send rabased V2.

-ck