Replacing get_zeroed_page/free_page/dma_map_single/dma_unmap_single()
with dma_alloc_noncoherent/dma_free_noncoherent() helps to reduce
code size, and simplify the code, and the hardware keep DMA coherent
itself.
Signed-off-by: Cai Huoqing <[email protected]>
---
v1->v2:
*Change to dma_alloc/free_noncoherent from dma_alloc/free_coherent.
*Update changelog.
drivers/scsi/ibmvscsi_tgt/ibmvscsi_tgt.c | 46 ++++++++----------------
1 file changed, 15 insertions(+), 31 deletions(-)
diff --git a/drivers/scsi/ibmvscsi_tgt/ibmvscsi_tgt.c b/drivers/scsi/ibmvscsi_tgt/ibmvscsi_tgt.c
index 61f06f6885a5..91199b969718 100644
--- a/drivers/scsi/ibmvscsi_tgt/ibmvscsi_tgt.c
+++ b/drivers/scsi/ibmvscsi_tgt/ibmvscsi_tgt.c
@@ -3007,20 +3007,13 @@ static long ibmvscsis_create_command_q(struct scsi_info *vscsi, int num_cmds)
vscsi->cmd_q.size = pages;
- vscsi->cmd_q.base_addr =
- (struct viosrp_crq *)get_zeroed_page(GFP_KERNEL);
- if (!vscsi->cmd_q.base_addr)
- return -ENOMEM;
-
vscsi->cmd_q.mask = ((uint)pages * CRQ_PER_PAGE) - 1;
- vscsi->cmd_q.crq_token = dma_map_single(&vdev->dev,
- vscsi->cmd_q.base_addr,
- PAGE_SIZE, DMA_BIDIRECTIONAL);
- if (dma_mapping_error(&vdev->dev, vscsi->cmd_q.crq_token)) {
- free_page((unsigned long)vscsi->cmd_q.base_addr);
+ vscsi->cmd_q.base_addr = dma_alloc_noncoherent(&vdev->dev, PAGE_SIZE,
+ &vscsi->cmd_q.crq_token,
+ DMA_BIDIRECTIONAL, GFP_KERNEL);
+ if (!vscsi->cmd_q.base_addr)
return -ENOMEM;
- }
return 0;
}
@@ -3036,9 +3029,9 @@ static long ibmvscsis_create_command_q(struct scsi_info *vscsi, int num_cmds)
*/
static void ibmvscsis_destroy_command_q(struct scsi_info *vscsi)
{
- dma_unmap_single(&vscsi->dma_dev->dev, vscsi->cmd_q.crq_token,
- PAGE_SIZE, DMA_BIDIRECTIONAL);
- free_page((unsigned long)vscsi->cmd_q.base_addr);
+ dma_free_noncoherent(&vscsi->dma_dev->dev,
+ PAGE_SIZE, vscsi->cmd_q.base_addr,
+ vscsi->cmd_q.crq_token, DMA_BIDIRECTIONAL);
vscsi->cmd_q.base_addr = NULL;
vscsi->state = NO_QUEUE;
}
@@ -3504,18 +3497,12 @@ static int ibmvscsis_probe(struct vio_dev *vdev,
goto free_timer;
}
- vscsi->map_buf = kzalloc(PAGE_SIZE, GFP_KERNEL);
+ vscsi->map_buf = dma_alloc_noncoherent(&vdev->dev,
+ PAGE_SIZE, &vscsi->map_ioba,
+ DMA_BIDIRECTIONAL, GFP_KERNEL);
if (!vscsi->map_buf) {
rc = -ENOMEM;
dev_err(&vscsi->dev, "probe: allocating cmd buffer failed\n");
- goto destroy_queue;
- }
-
- vscsi->map_ioba = dma_map_single(&vdev->dev, vscsi->map_buf, PAGE_SIZE,
- DMA_BIDIRECTIONAL);
- if (dma_mapping_error(&vdev->dev, vscsi->map_ioba)) {
- rc = -ENOMEM;
- dev_err(&vscsi->dev, "probe: error mapping command buffer\n");
goto free_buf;
}
@@ -3544,7 +3531,7 @@ static int ibmvscsis_probe(struct vio_dev *vdev,
if (!vscsi->work_q) {
rc = -ENOMEM;
dev_err(&vscsi->dev, "create_workqueue failed\n");
- goto unmap_buf;
+ goto destroy_queue;
}
rc = request_irq(vdev->irq, ibmvscsis_interrupt, 0, "ibmvscsis", vscsi);
@@ -3562,11 +3549,9 @@ static int ibmvscsis_probe(struct vio_dev *vdev,
destroy_WQ:
destroy_workqueue(vscsi->work_q);
-unmap_buf:
- dma_unmap_single(&vdev->dev, vscsi->map_ioba, PAGE_SIZE,
- DMA_BIDIRECTIONAL);
free_buf:
- kfree(vscsi->map_buf);
+ dma_free_noncoherent(&vdev->dev, PAGE_SIZE, vscsi->map_buf,
+ vscsi->map_ioba, DMA_BIDIRECTIONAL);
destroy_queue:
tasklet_kill(&vscsi->work_task);
ibmvscsis_unregister_command_q(vscsi);
@@ -3602,9 +3587,8 @@ static void ibmvscsis_remove(struct vio_dev *vdev)
vio_disable_interrupts(vdev);
free_irq(vdev->irq, vscsi);
destroy_workqueue(vscsi->work_q);
- dma_unmap_single(&vdev->dev, vscsi->map_ioba, PAGE_SIZE,
- DMA_BIDIRECTIONAL);
- kfree(vscsi->map_buf);
+ dma_free_noncoherent(&vdev->dev, PAGE_SIZE, vscsi->map_buf,
+ vscsi->map_ioba, DMA_BIDIRECTIONAL);
tasklet_kill(&vscsi->work_task);
ibmvscsis_destroy_command_q(vscsi);
ibmvscsis_freetimer(vscsi);
--
2.25.1
On Tue, Oct 12, 2021 at 11:21:09AM +0800, Cai Huoqing wrote:
> Replacing get_zeroed_page/free_page/dma_map_single/dma_unmap_single()
> with dma_alloc_noncoherent/dma_free_noncoherent() helps to reduce
> code size, and simplify the code, and the hardware keep DMA coherent
> itself.
It might also be worth to mention that this also avoids potential
bounce buffering. Although for pseries vio devices this probably can't
happen anyway.
> + vscsi->cmd_q.base_addr = dma_alloc_noncoherent(&vdev->dev, PAGE_SIZE,
> + &vscsi->cmd_q.crq_token,
> + DMA_BIDIRECTIONAL, GFP_KERNEL);
Please avoid the overly long line.
The same comments also apply to the other patch.
On Tue, 12 Oct 2021, Cai Huoqing wrote:
> Replacing get_zeroed_page/free_page/dma_map_single/dma_unmap_single()
> with dma_alloc_noncoherent/dma_free_noncoherent() helps to reduce
> code size, and simplify the code, and the hardware keep DMA coherent
> itself.
>
> Signed-off-by: Cai Huoqing <[email protected]>
> ---
> v1->v2:
> *Change to dma_alloc/free_noncoherent from dma_alloc/free_coherent.
> *Update changelog.
>
> drivers/scsi/ibmvscsi_tgt/ibmvscsi_tgt.c | 46 ++++++++----------------
> 1 file changed, 15 insertions(+), 31 deletions(-)
>
> diff --git a/drivers/scsi/ibmvscsi_tgt/ibmvscsi_tgt.c b/drivers/scsi/ibmvscsi_tgt/ibmvscsi_tgt.c
> index 61f06f6885a5..91199b969718 100644
> --- a/drivers/scsi/ibmvscsi_tgt/ibmvscsi_tgt.c
> +++ b/drivers/scsi/ibmvscsi_tgt/ibmvscsi_tgt.c
> @@ -3007,20 +3007,13 @@ static long ibmvscsis_create_command_q(struct scsi_info *vscsi, int num_cmds)
>
> vscsi->cmd_q.size = pages;
>
> - vscsi->cmd_q.base_addr =
> - (struct viosrp_crq *)get_zeroed_page(GFP_KERNEL);
> - if (!vscsi->cmd_q.base_addr)
> - return -ENOMEM;
> -
> vscsi->cmd_q.mask = ((uint)pages * CRQ_PER_PAGE) - 1;
>
> - vscsi->cmd_q.crq_token = dma_map_single(&vdev->dev,
> - vscsi->cmd_q.base_addr,
> - PAGE_SIZE, DMA_BIDIRECTIONAL);
> - if (dma_mapping_error(&vdev->dev, vscsi->cmd_q.crq_token)) {
> - free_page((unsigned long)vscsi->cmd_q.base_addr);
> + vscsi->cmd_q.base_addr = dma_alloc_noncoherent(&vdev->dev, PAGE_SIZE,
> + &vscsi->cmd_q.crq_token,
> + DMA_BIDIRECTIONAL, GFP_KERNEL);
> + if (!vscsi->cmd_q.base_addr)
> return -ENOMEM;
> - }
>
> return 0;
> }
> @@ -3036,9 +3029,9 @@ static long ibmvscsis_create_command_q(struct scsi_info *vscsi, int num_cmds)
> */
> static void ibmvscsis_destroy_command_q(struct scsi_info *vscsi)
> {
> - dma_unmap_single(&vscsi->dma_dev->dev, vscsi->cmd_q.crq_token,
> - PAGE_SIZE, DMA_BIDIRECTIONAL);
> - free_page((unsigned long)vscsi->cmd_q.base_addr);
> + dma_free_noncoherent(&vscsi->dma_dev->dev,
> + PAGE_SIZE, vscsi->cmd_q.base_addr,
> + vscsi->cmd_q.crq_token, DMA_BIDIRECTIONAL);
> vscsi->cmd_q.base_addr = NULL;
> vscsi->state = NO_QUEUE;
> }
> @@ -3504,18 +3497,12 @@ static int ibmvscsis_probe(struct vio_dev *vdev,
> goto free_timer;
> }
>
> - vscsi->map_buf = kzalloc(PAGE_SIZE, GFP_KERNEL);
> + vscsi->map_buf = dma_alloc_noncoherent(&vdev->dev,
> + PAGE_SIZE, &vscsi->map_ioba,
> + DMA_BIDIRECTIONAL, GFP_KERNEL);
> if (!vscsi->map_buf) {
> rc = -ENOMEM;
> dev_err(&vscsi->dev, "probe: allocating cmd buffer failed\n");
> - goto destroy_queue;
> - }
> -
> - vscsi->map_ioba = dma_map_single(&vdev->dev, vscsi->map_buf, PAGE_SIZE,
> - DMA_BIDIRECTIONAL);
> - if (dma_mapping_error(&vdev->dev, vscsi->map_ioba)) {
> - rc = -ENOMEM;
> - dev_err(&vscsi->dev, "probe: error mapping command buffer\n");
> goto free_buf;
Shouldn't that be goto destroy_queue?
> }
>
> @@ -3544,7 +3531,7 @@ static int ibmvscsis_probe(struct vio_dev *vdev,
> if (!vscsi->work_q) {
> rc = -ENOMEM;
> dev_err(&vscsi->dev, "create_workqueue failed\n");
> - goto unmap_buf;
> + goto destroy_queue;
And goto free_buf?
> }
>
> rc = request_irq(vdev->irq, ibmvscsis_interrupt, 0, "ibmvscsis", vscsi);
> @@ -3562,11 +3549,9 @@ static int ibmvscsis_probe(struct vio_dev *vdev,
>
> destroy_WQ:
> destroy_workqueue(vscsi->work_q);
> -unmap_buf:
> - dma_unmap_single(&vdev->dev, vscsi->map_ioba, PAGE_SIZE,
> - DMA_BIDIRECTIONAL);
> free_buf:
> - kfree(vscsi->map_buf);
> + dma_free_noncoherent(&vdev->dev, PAGE_SIZE, vscsi->map_buf,
> + vscsi->map_ioba, DMA_BIDIRECTIONAL);
> destroy_queue:
> tasklet_kill(&vscsi->work_task);
> ibmvscsis_unregister_command_q(vscsi);
> @@ -3602,9 +3587,8 @@ static void ibmvscsis_remove(struct vio_dev *vdev)
> vio_disable_interrupts(vdev);
> free_irq(vdev->irq, vscsi);
> destroy_workqueue(vscsi->work_q);
> - dma_unmap_single(&vdev->dev, vscsi->map_ioba, PAGE_SIZE,
> - DMA_BIDIRECTIONAL);
> - kfree(vscsi->map_buf);
> + dma_free_noncoherent(&vdev->dev, PAGE_SIZE, vscsi->map_buf,
> + vscsi->map_ioba, DMA_BIDIRECTIONAL);
> tasklet_kill(&vscsi->work_task);
> ibmvscsis_destroy_command_q(vscsi);
> ibmvscsis_freetimer(vscsi);
>