Skip to contentSkip to conversation

virtio-blk: small cleanup

PATCH In review Chaitanya Kulkarni 3 participants 14 messages Apr 20, 2022 – Apr 25, 2022

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

cover virtio-blk: small cleanup

Hi,

This has some nit fixes and code cleanups along with removing deprecated API fir ida_simple_XXX().

-ck

Chaitanya Kulkarni (4):
virtio-blk: remove additional check in fast path
virtio-blk: don't add a new line
virtio-blk: avoid function call in the fast path
virtio-blk: remove deprecated ida_simple_XXX()

 drivers/block/virtio_blk.c | 38 +++++++++++++++++---------------------
 1 file changed, 17 insertions(+), 21 deletions(-)
Signature · 1 line
2.29.0
#1 lore raw

PATCH 1/4 virtio-blk: remove additional check in fast path

The function virtblk_setup_cmd() calls virtblk_setup_discard_write_zeroes() once we process the block layer request operation setup in the switch. Even though it saves duplicate call for REQ_OP_DISCARD and REQ_OP_WRITE_ZEROES it adds additional check in the fast path that is redundent since we already have a switch case for both REQ_OP_DISCARD and REQ_OP_WRITE_ZEROES.

Move the call virtblk_setup_discard_write_zeroes() into switch case label of REQ_OP_DISCARD and REQ_OP_WRITE_ZEROES and avoid duplicate branch in the fast path.

---
 drivers/block/virtio_blk.c | 9 ++++-----
 1 file changed, 4 insertions(+), 5 deletions(-)
drivers/block/virtio_blk.c +4 -5
index 6ccf15253dee..b77711e73422 100644
@@ -223,10 +223,14 @@ static blk_status_t virtblk_setup_cmd(struct virtio_device *vdev,
 		break;
 	case REQ_OP_DISCARD:
 		type = VIRTIO_BLK_T_DISCARD;
+		if (virtblk_setup_discard_write_zeroes(req, unmap))
+			return BLK_STS_RESOURCE;
 		break;
 	case REQ_OP_WRITE_ZEROES:
 		type = VIRTIO_BLK_T_WRITE_ZEROES;
 		unmap = !(req->cmd_flags & REQ_NOUNMAP);
+		if (virtblk_setup_discard_write_zeroes(req, unmap))
+			return BLK_STS_RESOURCE;
 		break;
 	case REQ_OP_DRV_IN:
 		type = VIRTIO_BLK_T_GET_ID;
@@ -239,11 +243,6 @@ static blk_status_t virtblk_setup_cmd(struct virtio_device *vdev,
 	vbr->out_hdr.type = cpu_to_virtio32(vdev, type);
 	vbr->out_hdr.ioprio = cpu_to_virtio32(vdev, req_get_ioprio(req));
 
-	if (type == VIRTIO_BLK_T_DISCARD || type == VIRTIO_BLK_T_WRITE_ZEROES) {
-		if (virtblk_setup_discard_write_zeroes(req, unmap))
-			return BLK_STS_RESOURCE;
-	}
-
 	return 0;
 }
 
Signature · 1 line
2.29.0
#4 lore raw

PATCH 2/4 virtio-blk: don't add a new line

In reply to Chaitanya Kulkarni

Don't split sector assignment line for REQ_OP_READ and REQ_OP_WRITE in the virtblk_setup_cmd() which fits in one line perfectly.

---
 drivers/block/virtio_blk.c | 6 ++----
 1 file changed, 2 insertions(+), 4 deletions(-)
drivers/block/virtio_blk.c +2 -4
index b77711e73422..d038800474c2 100644
@@ -210,13 +210,11 @@ static blk_status_t virtblk_setup_cmd(struct virtio_device *vdev,
 	switch (req_op(req)) {
 	case REQ_OP_READ:
 		type = VIRTIO_BLK_T_IN;
-		vbr->out_hdr.sector = cpu_to_virtio64(vdev,
-						      blk_rq_pos(req));
+		vbr->out_hdr.sector = cpu_to_virtio64(vdev, blk_rq_pos(req));
 		break;
 	case REQ_OP_WRITE:
 		type = VIRTIO_BLK_T_OUT;
-		vbr->out_hdr.sector = cpu_to_virtio64(vdev,
-						      blk_rq_pos(req));
+		vbr->out_hdr.sector = cpu_to_virtio64(vdev, blk_rq_pos(req));
 		break;
 	case REQ_OP_FLUSH:
 		type = VIRTIO_BLK_T_FLUSH;
Signature · 1 line
2.29.0
#7 lore raw

PATCH 3/4 virtio-blk: avoid function call in the fast path

In reply to Chaitanya Kulkarni

We can avoid a function call virtblk_map_data() in the fast path if block layer request has no physical segments by moving the call blk_rq_nr_phys_segments() from virtblk_map_data() to virtio_queue_rq().

---
 drivers/block/virtio_blk.c | 15 +++++++--------
 1 file changed, 7 insertions(+), 8 deletions(-)
drivers/block/virtio_blk.c +7 -8
index d038800474c2..74c3a48cd1e5 100644
@@ -178,9 +178,6 @@ static int virtblk_map_data(struct blk_mq_hw_ctx *hctx, struct request *req,
 {
 	int err;
 
-	if (!blk_rq_nr_phys_segments(req))
-		return 0;
-
 	vbr->sg_table.sgl = vbr->sg;
 	err = sg_alloc_table_chained(&vbr->sg_table,
 				     blk_rq_nr_phys_segments(req),
@@ -303,7 +300,7 @@ static blk_status_t virtio_queue_rq(struct blk_mq_hw_ctx *hctx,
 	struct request *req = bd->rq;
 	struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
 	unsigned long flags;
-	int num;
+	int num = 0;
 	int qid = hctx->queue_num;
 	bool notify = false;
 	blk_status_t status;
@@ -315,10 +312,12 @@ static blk_status_t virtio_queue_rq(struct blk_mq_hw_ctx *hctx,
 
 	blk_mq_start_request(req);
 
-	num = virtblk_map_data(hctx, req, vbr);
-	if (unlikely(num < 0)) {
-		virtblk_cleanup_cmd(req);
-		return BLK_STS_RESOURCE;
+	if (blk_rq_nr_phys_segments(req)) {
+		num = virtblk_map_data(hctx, req, vbr);
+		if (unlikely(num < 0)) {
+			virtblk_cleanup_cmd(req);
+			return BLK_STS_RESOURCE;
+		}
 	}
 
 	spin_lock_irqsave(&vblk->vqs[qid].lock, flags);
Signature · 1 line
2.29.0
#10 lore raw

PATCH 4/4 virtio-blk: remove deprecated ida_simple_XXX()

In reply to Chaitanya Kulkarni

Replace deprecated ida_simple_get() and ida_simple_remove() with ida_alloc_max() and ida_free().

---
 drivers/block/virtio_blk.c | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)
drivers/block/virtio_blk.c +4 -4
index 74c3a48cd1e5..e05748337dd1 100644
@@ -411,7 +411,7 @@ static void virtblk_free_disk(struct gendisk *disk)
 {
 	struct virtio_blk *vblk = disk->private_data;
 
-	ida_simple_remove(&vd_index_ida, vblk->index);
+	ida_free(&vd_index_ida, vblk->index);
 	mutex_destroy(&vblk->vdev_mutex);
 	kfree(vblk);
 }
@@ -720,8 +720,8 @@ static int virtblk_probe(struct virtio_device *vdev)
 		return -EINVAL;
 	}
 
-	err = ida_simple_get(&vd_index_ida, 0, minor_to_index(1 << MINORBITS),
-			     GFP_KERNEL);
+	err = ida_alloc_max(&vd_index_ida, minor_to_index(1 << MINORBITS),
+			    GFP_KERNEL);
 	if (err < 0)
 		goto out;
 	index = err;
@@ -911,7 +911,7 @@ static int virtblk_probe(struct virtio_device *vdev)
 out_free_vblk:
 	kfree(vblk);
 out_free_index:
-	ida_simple_remove(&vd_index_ida, index);
+	ida_free(&vd_index_ida, index);
 out:
 	return err;
 }
Signature · 1 line
2.29.0
#2 lore raw

Re: [PATCH 1/4] virtio-blk: remove additional check in fast path

In reply to Chaitanya Kulkarni
Quoted 7 lines · Chaitanya Kulkarni

The function virtblk_setup_cmd() calls virtblk_setup_discard_write_zeroes() once we process the block layer request operation setup in the switch. Even though it saves duplicate call for REQ_OP_DISCARD and REQ_OP_WRITE_ZEROES it adds additional check in the fast path that is redundent since we already have a switch case for both REQ_OP_DISCARD and REQ_OP_WRITE_ZEROES.

Quoted 8 lines

Move the call virtblk_setup_discard_write_zeroes() into switch case label of REQ_OP_DISCARD and REQ_OP_WRITE_ZEROES and avoid duplicate branch in the fast path.

---
 drivers/block/virtio_blk.c | 9 ++++-----
 1 file changed, 4 insertions(+), 5 deletions(-)

Is there data that shows the performance effect of moving the code out of the fast path?

drivers/block/virtio_blk.c static blk_status_t virtblk_setup_cmd(struct virtio_device *vdev, · line 226

Quoted 9 lines
drivers/block/virtio_blk.c +1 -0
index 6ccf15253dee..b77711e73422 100644
@@ -223,10 +223,14 @@ static blk_status_t virtblk_setup_cmd(struct virtio_device *vdev,
 		break;
 	case REQ_OP_DISCARD:
 		type = VIRTIO_BLK_T_DISCARD;
+		if (virtblk_setup_discard_write_zeroes(req, unmap))

unmap is never true here. The variable obscures what is going on:

s/unmap/false/

reviewer Reviewed-by on v1 2/4
#5 lore raw

Re: [PATCH 2/4] virtio-blk: don't add a new line

In reply to Chaitanya Kulkarni
Quoted 7 lines · Chaitanya Kulkarni

Don't split sector assignment line for REQ_OP_READ and REQ_OP_WRITE in the virtblk_setup_cmd() which fits in one line perfectly.

---
 drivers/block/virtio_blk.c | 6 ++----
 1 file changed, 2 insertions(+), 4 deletions(-)

There is a cost to patches: humans spend time reviewing them, CI systems consume energy running tests, downstream maintainers deal with backports, and git-blame(1) becomes harder to use when code changes.

What constitutes code churn is subjective. For me personally I prefer it when patches have a clear benefit that outweighs the costs. Nevertheless:

_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

#8 lore raw

Re: [PATCH 3/4] virtio-blk: avoid function call in the fast path

In reply to Chaitanya Kulkarni
Quoted 8 lines · Chaitanya Kulkarni

We can avoid a function call virtblk_map_data() in the fast path if block layer request has no physical segments by moving the call blk_rq_nr_phys_segments() from virtblk_map_data() to virtio_queue_rq().

---
 drivers/block/virtio_blk.c | 15 +++++++--------
 1 file changed, 7 insertions(+), 8 deletions(-)

virtblk_map_data() is a static function that is not called by anything else in virtio_blk.c. The disassembly on my x86_64 machine shows it has been inlined into virtio_queue_rq(). The compiler has already done what this patch is trying to do (and more).

Can you share performance data or some more background on why this code change is necessary?

Quoted 44 lines
drivers/block/virtio_blk.c +7 -8
index d038800474c2..74c3a48cd1e5 100644
@@ -178,9 +178,6 @@ static int virtblk_map_data(struct blk_mq_hw_ctx *hctx, struct request *req,
 {
 	int err;
 
-	if (!blk_rq_nr_phys_segments(req))
-		return 0;
-
 	vbr->sg_table.sgl = vbr->sg;
 	err = sg_alloc_table_chained(&vbr->sg_table,
 				     blk_rq_nr_phys_segments(req),
@@ -303,7 +300,7 @@ static blk_status_t virtio_queue_rq(struct blk_mq_hw_ctx *hctx,
 	struct request *req = bd->rq;
 	struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
 	unsigned long flags;
-	int num;
+	int num = 0;
 	int qid = hctx->queue_num;
 	bool notify = false;
 	blk_status_t status;
@@ -315,10 +312,12 @@ static blk_status_t virtio_queue_rq(struct blk_mq_hw_ctx *hctx,
 
 	blk_mq_start_request(req);
 
-	num = virtblk_map_data(hctx, req, vbr);
-	if (unlikely(num < 0)) {
-		virtblk_cleanup_cmd(req);
-		return BLK_STS_RESOURCE;
+	if (blk_rq_nr_phys_segments(req)) {
+		num = virtblk_map_data(hctx, req, vbr);
+		if (unlikely(num < 0)) {
+			virtblk_cleanup_cmd(req);
+			return BLK_STS_RESOURCE;
+		}
 	}
 
 	spin_lock_irqsave(&vblk->vqs[qid].lock, flags);
Signature · 1 line
2.29.0
reviewer Reviewed-by on v1 4/4
#11 lore raw

Re: [PATCH 4/4] virtio-blk: remove deprecated ida_simple_XXX()

In reply to Chaitanya Kulkarni
Quoted 7 lines · Chaitanya Kulkarni

Replace deprecated ida_simple_get() and ida_simple_remove() with ida_alloc_max() and ida_free().

---
 drivers/block/virtio_blk.c | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

#3 lore raw

Re: [PATCH 1/4] virtio-blk: remove additional check in fast path

In reply to Stefan Hajnoczi
Quoted 12 lines · Stefan Hajnoczi
Quoted 11 lines · Chaitanya Kulkarni

The function virtblk_setup_cmd() calls virtblk_setup_discard_write_zeroes() once we process the block layer request operation setup in the switch. Even though it saves duplicate call for REQ_OP_DISCARD and REQ_OP_WRITE_ZEROES it adds additional check in the fast path that is redundent since we already have a switch case for both REQ_OP_DISCARD and REQ_OP_WRITE_ZEROES.

Move the call virtblk_setup_discard_write_zeroes() into switch case label of REQ_OP_DISCARD and REQ_OP_WRITE_ZEROES and avoid duplicate branch in the fast path.

Quoted 8 lines
Quoted 4 lines
---
  drivers/block/virtio_blk.c | 9 ++++-----
  1 file changed, 4 insertions(+), 5 deletions(-)

Is there data that shows the performance effect of moving the code out of the fast path?

I don't have a data yet but trying to minimize fast path branches as I can when I was reading the code...

Quoted 12 lines

drivers/block/virtio_blk.c static blk_status_t virtblk_setup_cmd(struct virtio_device *vdev, · line 226

Quoted 9 lines
drivers/block/virtio_blk.c +1 -0
index 6ccf15253dee..b77711e73422 100644
@@ -223,10 +223,14 @@ static blk_status_t virtblk_setup_cmd(struct virtio_device *vdev,
  		break;
  	case REQ_OP_DISCARD:
  		type = VIRTIO_BLK_T_DISCARD;
+		if (virtblk_setup_discard_write_zeroes(req, unmap))

unmap is never true here. The variable obscures what is going on:

Quoted 2 lines

s/unmap/false/

yeah, I'll drop this patch from V2.

-ck

#6 lore raw

Re: [PATCH 2/4] virtio-blk: don't add a new line

In reply to Stefan Hajnoczi
Quoted 13 lines · Stefan Hajnoczi
Quoted 7 lines · Chaitanya Kulkarni

Don't split sector assignment line for REQ_OP_READ and REQ_OP_WRITE in the virtblk_setup_cmd() which fits in one line perfectly.

---
  drivers/block/virtio_blk.c | 6 ++----
  1 file changed, 2 insertions(+), 4 deletions(-)

There is a cost to patches: humans spend time reviewing them, CI systems consume energy running tests, downstream maintainers deal with backports, and git-blame(1) becomes harder to use when code changes.

Quoted 6 lines

What constitutes code churn is subjective. For me personally I prefer it when patches have a clear benefit that outweighs the costs. Nevertheless:

maybe we can just drop it then, for naked eye it looked a bit odd to add a new line..

-ck

#9 lore raw

Re: [PATCH 3/4] virtio-blk: avoid function call in the fast path

In reply to Stefan Hajnoczi
Quoted 10 lines · Stefan Hajnoczi
Quoted 8 lines · Chaitanya Kulkarni

We can avoid a function call virtblk_map_data() in the fast path if block layer request has no physical segments by moving the call blk_rq_nr_phys_segments() from virtblk_map_data() to virtio_queue_rq().

---
  drivers/block/virtio_blk.c | 15 +++++++--------
  1 file changed, 7 insertions(+), 8 deletions(-)
Quoted 9 lines

virtblk_map_data() is a static function that is not called by anything else in virtio_blk.c. The disassembly on my x86_64 machine shows it has been inlined into virtio_queue_rq(). The compiler has already done what this patch is trying to do (and more).

Can you share performance data or some more background on why this code change is necessary?

I don't have performance numbers, but when reading the code it takes to a function call especially in the fast path where read/write requests has phys segments and those are probably most frequent requests...

I'll drop it from V2, if I find performance numbers then I'll repost is with the quantitative data ...

-ck

#12 lore raw
In reply to Chaitanya Kulkarni
Quoted 7 lines · Chaitanya Kulkarni

Hi,

This has some nit fixes and code cleanups along with removing deprecated API fir ida_simple_XXX().

-ck

Quoted 9 lines

Chaitanya Kulkarni (4):
virtio-blk: remove additional check in fast path
virtio-blk: don't add a new line
virtio-blk: avoid function call in the fast path
virtio-blk: remove deprecated ida_simple_XXX()

  drivers/block/virtio_blk.c | 38 +++++++++++++++++---------------------
  1 file changed, 17 insertions(+), 21 deletions(-)

Thanks for the review, I'll send out a V2 and drop patches that lacks the quantitative data..

-ck

#13 lore raw
Quoted 14 lines · Chaitanya Kulkarni
Quoted 13 lines · Chaitanya Kulkarni

Hi,

This has some nit fixes and code cleanups along with removing deprecated API fir ida_simple_XXX().

-ck

Chaitanya Kulkarni (4):
virtio-blk: remove additional check in fast path
virtio-blk: don't add a new line
virtio-blk: avoid function call in the fast path
virtio-blk: remove deprecated ida_simple_XXX()

Quoted 6 lines
Quoted 3 lines
  drivers/block/virtio_blk.c | 38 +++++++++++++++++---------------------
  1 file changed, 17 insertions(+), 21 deletions(-)

Thanks for the review, I'll send out a V2 and drop patches that lacks the quantitative data..

Thank you!

Stefan

_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization