* [PATCH 3/5] dma: optimize two stage transfer function
@ 2023-08-07 5:20 Kaiwei Liu
2023-08-07 8:27 ` Baolin Wang
0 siblings, 1 reply; 2+ messages in thread
From: Kaiwei Liu @ 2023-08-07 5:20 UTC (permalink / raw)
To: Vinod Koul, Orson Zhai, Baolin Wang, Chunyan Zhang
Cc: dmaengine, linux-kernel, kaiwei liu, Wenming Wu
The DMA hardware is updated to optimize two stages transmission
function, so modify relative register and logic. Two
stages transmission mode of dma allows one channel finished
transmission then start another channel transfer automatically.
Signed-off-by: Kaiwei Liu <kaiwei.liu@unisoc.com>
---
drivers/dma/sprd-dma.c | 124 ++++++++++++++++++++++++++++++-----------
1 file changed, 91 insertions(+), 33 deletions(-)
diff --git a/drivers/dma/sprd-dma.c b/drivers/dma/sprd-dma.c
index 0e146550dfbb..01053e106e8a 100644
--- a/drivers/dma/sprd-dma.c
+++ b/drivers/dma/sprd-dma.c
@@ -68,6 +68,7 @@
#define SPRD_DMA_GLB_TRANS_DONE_TRG BIT(18)
#define SPRD_DMA_GLB_BLOCK_DONE_TRG BIT(17)
#define SPRD_DMA_GLB_FRAG_DONE_TRG BIT(16)
+#define SPRD_DMA_GLB_TRG_MASK GENMASK(19, 16)
#define SPRD_DMA_GLB_TRG_OFFSET 16
#define SPRD_DMA_GLB_DEST_CHN_MASK GENMASK(13, 8)
#define SPRD_DMA_GLB_DEST_CHN_OFFSET 8
@@ -155,6 +156,13 @@
#define SPRD_DMA_SOFTWARE_UID 0
+#define SPRD_DMA_SRC_CHN0_INT 9
+#define SPRD_DMA_SRC_CHN1_INT 10
+#define SPRD_DMA_DST_CHN0_INT 11
+#define SPRD_DMA_DST_CHN1_INT 12
+#define SPRD_DMA_2STAGE_SET 1
+#define SPRD_DMA_2STAGE_CLEAR 0
+
/* dma data width values */
enum sprd_dma_datawidth {
SPRD_DMA_DATAWIDTH_1_BYTE,
@@ -212,7 +220,7 @@ struct sprd_dma_dev {
struct clk *ashb_clk;
int irq;
u32 total_chns;
- struct sprd_dma_chn channels[];
+ struct sprd_dma_chn channels[0];
};
static void sprd_dma_free_desc(struct virt_dma_desc *vd);
@@ -431,50 +439,90 @@ static enum sprd_dma_req_mode sprd_dma_get_req_type(struct sprd_dma_chn *schan)
return (frag_reg >> SPRD_DMA_REQ_MODE_OFFSET) & SPRD_DMA_REQ_MODE_MASK;
}
-static int sprd_dma_set_2stage_config(struct sprd_dma_chn *schan)
+static int sprd_dma_2stage_config(struct sprd_dma_chn *schan, u32 config_type)
{
struct sprd_dma_dev *sdev = to_sprd_dma_dev(&schan->vc.chan);
u32 val, chn = schan->chn_num + 1;
switch (schan->chn_mode) {
case SPRD_DMA_SRC_CHN0:
- val = chn & SPRD_DMA_GLB_SRC_CHN_MASK;
- val |= BIT(schan->trg_mode - 1) << SPRD_DMA_GLB_TRG_OFFSET;
- val |= SPRD_DMA_GLB_2STAGE_EN;
- if (schan->int_type != SPRD_DMA_NO_INT)
- val |= SPRD_DMA_GLB_SRC_INT;
-
- sprd_dma_glb_update(sdev, SPRD_DMA_GLB_2STAGE_GRP1, val, val);
+ if (config_type == SPRD_DMA_2STAGE_SET) {
+ val = chn & SPRD_DMA_GLB_SRC_CHN_MASK;
+ val |= BIT(schan->trg_mode - 1) << SPRD_DMA_GLB_TRG_OFFSET;
+ val |= SPRD_DMA_GLB_2STAGE_EN;
+ if (schan->int_type & SPRD_DMA_SRC_CHN0_INT)
+ val |= SPRD_DMA_GLB_SRC_INT;
+
+ sprd_dma_glb_update(sdev, SPRD_DMA_GLB_2STAGE_GRP1,
+ SPRD_DMA_GLB_SRC_INT |
+ SPRD_DMA_GLB_TRG_MASK |
+ SPRD_DMA_GLB_SRC_CHN_MASK, val);
+ } else {
+ sprd_dma_glb_update(sdev, SPRD_DMA_GLB_2STAGE_GRP1,
+ SPRD_DMA_GLB_SRC_INT |
+ SPRD_DMA_GLB_TRG_MASK |
+ SPRD_DMA_GLB_2STAGE_EN |
+ SPRD_DMA_GLB_SRC_CHN_MASK, 0);
+ }
break;
case SPRD_DMA_SRC_CHN1:
- val = chn & SPRD_DMA_GLB_SRC_CHN_MASK;
- val |= BIT(schan->trg_mode - 1) << SPRD_DMA_GLB_TRG_OFFSET;
- val |= SPRD_DMA_GLB_2STAGE_EN;
- if (schan->int_type != SPRD_DMA_NO_INT)
- val |= SPRD_DMA_GLB_SRC_INT;
-
- sprd_dma_glb_update(sdev, SPRD_DMA_GLB_2STAGE_GRP2, val, val);
+ if (config_type == SPRD_DMA_2STAGE_SET) {
+ val = chn & SPRD_DMA_GLB_SRC_CHN_MASK;
+ val |= BIT(schan->trg_mode - 1) << SPRD_DMA_GLB_TRG_OFFSET;
+ val |= SPRD_DMA_GLB_2STAGE_EN;
+ if (schan->int_type & SPRD_DMA_SRC_CHN1_INT)
+ val |= SPRD_DMA_GLB_SRC_INT;
+
+ sprd_dma_glb_update(sdev, SPRD_DMA_GLB_2STAGE_GRP2,
+ SPRD_DMA_GLB_SRC_INT |
+ SPRD_DMA_GLB_TRG_MASK |
+ SPRD_DMA_GLB_SRC_CHN_MASK, val);
+ } else {
+ sprd_dma_glb_update(sdev, SPRD_DMA_GLB_2STAGE_GRP2,
+ SPRD_DMA_GLB_SRC_INT |
+ SPRD_DMA_GLB_TRG_MASK |
+ SPRD_DMA_GLB_2STAGE_EN |
+ SPRD_DMA_GLB_SRC_CHN_MASK, 0);
+ }
break;
case SPRD_DMA_DST_CHN0:
- val = (chn << SPRD_DMA_GLB_DEST_CHN_OFFSET) &
- SPRD_DMA_GLB_DEST_CHN_MASK;
- val |= SPRD_DMA_GLB_2STAGE_EN;
- if (schan->int_type != SPRD_DMA_NO_INT)
- val |= SPRD_DMA_GLB_DEST_INT;
-
- sprd_dma_glb_update(sdev, SPRD_DMA_GLB_2STAGE_GRP1, val, val);
+ if (config_type == SPRD_DMA_2STAGE_SET) {
+ val = (chn << SPRD_DMA_GLB_DEST_CHN_OFFSET) &
+ SPRD_DMA_GLB_DEST_CHN_MASK;
+ val |= SPRD_DMA_GLB_2STAGE_EN;
+ if (schan->int_type & SPRD_DMA_DST_CHN0_INT)
+ val |= SPRD_DMA_GLB_DEST_INT;
+
+ sprd_dma_glb_update(sdev, SPRD_DMA_GLB_2STAGE_GRP1,
+ SPRD_DMA_GLB_DEST_INT |
+ SPRD_DMA_GLB_DEST_CHN_MASK, val);
+ } else {
+ sprd_dma_glb_update(sdev, SPRD_DMA_GLB_2STAGE_GRP1,
+ SPRD_DMA_GLB_DEST_INT |
+ SPRD_DMA_GLB_2STAGE_EN |
+ SPRD_DMA_GLB_DEST_CHN_MASK, 0);
+ }
break;
case SPRD_DMA_DST_CHN1:
- val = (chn << SPRD_DMA_GLB_DEST_CHN_OFFSET) &
- SPRD_DMA_GLB_DEST_CHN_MASK;
- val |= SPRD_DMA_GLB_2STAGE_EN;
- if (schan->int_type != SPRD_DMA_NO_INT)
- val |= SPRD_DMA_GLB_DEST_INT;
-
- sprd_dma_glb_update(sdev, SPRD_DMA_GLB_2STAGE_GRP2, val, val);
+ if (config_type == SPRD_DMA_2STAGE_SET) {
+ val = (chn << SPRD_DMA_GLB_DEST_CHN_OFFSET) &
+ SPRD_DMA_GLB_DEST_CHN_MASK;
+ val |= SPRD_DMA_GLB_2STAGE_EN;
+ if (schan->int_type & SPRD_DMA_DST_CHN1_INT)
+ val |= SPRD_DMA_GLB_DEST_INT;
+
+ sprd_dma_glb_update(sdev, SPRD_DMA_GLB_2STAGE_GRP2,
+ SPRD_DMA_GLB_DEST_INT |
+ SPRD_DMA_GLB_DEST_CHN_MASK, val);
+ } else {
+ sprd_dma_glb_update(sdev, SPRD_DMA_GLB_2STAGE_GRP2,
+ SPRD_DMA_GLB_DEST_INT |
+ SPRD_DMA_GLB_2STAGE_EN |
+ SPRD_DMA_GLB_DEST_CHN_MASK, 0);
+ }
break;
default:
@@ -545,7 +593,7 @@ static void sprd_dma_start(struct sprd_dma_chn *schan)
* Set 2-stage configuration if the channel starts one 2-stage
* transfer.
*/
- if (schan->chn_mode && sprd_dma_set_2stage_config(schan))
+ if (schan->chn_mode && sprd_dma_2stage_config(schan, SPRD_DMA_2STAGE_SET))
return;
/*
@@ -569,6 +617,12 @@ static void sprd_dma_stop(struct sprd_dma_chn *schan)
sprd_dma_set_pending(schan, false);
sprd_dma_unset_uid(schan);
sprd_dma_clear_int(schan);
+ /*
+ * If 2-stage transfer is used, the configuration must be clear
+ * when release DMA channel.
+ */
+ if (schan->chn_mode)
+ sprd_dma_2stage_config(schan, SPRD_DMA_2STAGE_CLEAR);
schan->cur_desc = NULL;
}
@@ -757,7 +811,9 @@ static int sprd_dma_fill_desc(struct dma_chan *chan,
phys_addr_t llist_ptr;
if (dir == DMA_MEM_TO_DEV) {
- src_step = sprd_dma_get_step(slave_cfg->src_addr_width);
+ src_step = slave_cfg->src_port_window_size ?
+ slave_cfg->src_port_window_size :
+ sprd_dma_get_step(slave_cfg->src_addr_width);
if (src_step < 0) {
dev_err(sdev->dma_dev.dev, "invalid source step\n");
return src_step;
@@ -773,7 +829,9 @@ static int sprd_dma_fill_desc(struct dma_chan *chan,
else
dst_step = SPRD_DMA_NONE_STEP;
} else {
- dst_step = sprd_dma_get_step(slave_cfg->dst_addr_width);
+ dst_step = slave_cfg->dst_port_window_size ?
+ slave_cfg->dst_port_window_size :
+ sprd_dma_get_step(slave_cfg->dst_addr_width);
if (dst_step < 0) {
dev_err(sdev->dma_dev.dev, "invalid destination step\n");
return dst_step;
--
2.17.1
^ permalink raw reply [flat|nested] 2+ messages in thread
* Re: [PATCH 3/5] dma: optimize two stage transfer function
2023-08-07 5:20 [PATCH 3/5] dma: optimize two stage transfer function Kaiwei Liu
@ 2023-08-07 8:27 ` Baolin Wang
0 siblings, 0 replies; 2+ messages in thread
From: Baolin Wang @ 2023-08-07 8:27 UTC (permalink / raw)
To: Kaiwei Liu, Vinod Koul, Orson Zhai, Chunyan Zhang
Cc: dmaengine, linux-kernel, kaiwei liu, Wenming Wu
On 8/7/2023 1:20 PM, Kaiwei Liu wrote:
> The DMA hardware is updated to optimize two stages transmission
> function, so modify relative register and logic. Two
> stages transmission mode of dma allows one channel finished
> transmission then start another channel transfer automatically.
>
> Signed-off-by: Kaiwei Liu <kaiwei.liu@unisoc.com>
> ---
> drivers/dma/sprd-dma.c | 124 ++++++++++++++++++++++++++++++-----------
> 1 file changed, 91 insertions(+), 33 deletions(-)
>
> diff --git a/drivers/dma/sprd-dma.c b/drivers/dma/sprd-dma.c
> index 0e146550dfbb..01053e106e8a 100644
> --- a/drivers/dma/sprd-dma.c
> +++ b/drivers/dma/sprd-dma.c
> @@ -68,6 +68,7 @@
> #define SPRD_DMA_GLB_TRANS_DONE_TRG BIT(18)
> #define SPRD_DMA_GLB_BLOCK_DONE_TRG BIT(17)
> #define SPRD_DMA_GLB_FRAG_DONE_TRG BIT(16)
> +#define SPRD_DMA_GLB_TRG_MASK GENMASK(19, 16)
> #define SPRD_DMA_GLB_TRG_OFFSET 16
> #define SPRD_DMA_GLB_DEST_CHN_MASK GENMASK(13, 8)
> #define SPRD_DMA_GLB_DEST_CHN_OFFSET 8
> @@ -155,6 +156,13 @@
>
> #define SPRD_DMA_SOFTWARE_UID 0
>
> +#define SPRD_DMA_SRC_CHN0_INT 9
> +#define SPRD_DMA_SRC_CHN1_INT 10
> +#define SPRD_DMA_DST_CHN0_INT 11
> +#define SPRD_DMA_DST_CHN1_INT 12
> +#define SPRD_DMA_2STAGE_SET 1
> +#define SPRD_DMA_2STAGE_CLEAR 0
This 2 are not hardware definition, and I don't think they are helpful,
just use a boolean parameter.
> +
> /* dma data width values */
> enum sprd_dma_datawidth {
> SPRD_DMA_DATAWIDTH_1_BYTE,
> @@ -212,7 +220,7 @@ struct sprd_dma_dev {
> struct clk *ashb_clk;
> int irq;
> u32 total_chns;
> - struct sprd_dma_chn channels[];
> + struct sprd_dma_chn channels[0];
Please remove redundant changes.
> };
>
> static void sprd_dma_free_desc(struct virt_dma_desc *vd);
> @@ -431,50 +439,90 @@ static enum sprd_dma_req_mode sprd_dma_get_req_type(struct sprd_dma_chn *schan)
> return (frag_reg >> SPRD_DMA_REQ_MODE_OFFSET) & SPRD_DMA_REQ_MODE_MASK;
> }
>
> -static int sprd_dma_set_2stage_config(struct sprd_dma_chn *schan)
> +static int sprd_dma_2stage_config(struct sprd_dma_chn *schan, u32 config_type)
Why change the function name? 'config_type' should be boolean.
> {
> struct sprd_dma_dev *sdev = to_sprd_dma_dev(&schan->vc.chan);
> u32 val, chn = schan->chn_num + 1;
>
> switch (schan->chn_mode) {
> case SPRD_DMA_SRC_CHN0:
> - val = chn & SPRD_DMA_GLB_SRC_CHN_MASK;
> - val |= BIT(schan->trg_mode - 1) << SPRD_DMA_GLB_TRG_OFFSET;
> - val |= SPRD_DMA_GLB_2STAGE_EN;
> - if (schan->int_type != SPRD_DMA_NO_INT)
> - val |= SPRD_DMA_GLB_SRC_INT;
> -
> - sprd_dma_glb_update(sdev, SPRD_DMA_GLB_2STAGE_GRP1, val, val);
> + if (config_type == SPRD_DMA_2STAGE_SET) {
> + val = chn & SPRD_DMA_GLB_SRC_CHN_MASK;
> + val |= BIT(schan->trg_mode - 1) << SPRD_DMA_GLB_TRG_OFFSET;
> + val |= SPRD_DMA_GLB_2STAGE_EN;
> + if (schan->int_type & SPRD_DMA_SRC_CHN0_INT)
can you explain why change the interrupt validation? ignore other
interrupt type?
> + val |= SPRD_DMA_GLB_SRC_INT;
> +
> + sprd_dma_glb_update(sdev, SPRD_DMA_GLB_2STAGE_GRP1,
> + SPRD_DMA_GLB_SRC_INT |
> + SPRD_DMA_GLB_TRG_MASK |
> + SPRD_DMA_GLB_SRC_CHN_MASK, val);
> + } else {
> + sprd_dma_glb_update(sdev, SPRD_DMA_GLB_2STAGE_GRP1,
> + SPRD_DMA_GLB_SRC_INT |
> + SPRD_DMA_GLB_TRG_MASK |
> + SPRD_DMA_GLB_2STAGE_EN |
> + SPRD_DMA_GLB_SRC_CHN_MASK, 0);
> + }
> break;
>
> case SPRD_DMA_SRC_CHN1:
> - val = chn & SPRD_DMA_GLB_SRC_CHN_MASK;
> - val |= BIT(schan->trg_mode - 1) << SPRD_DMA_GLB_TRG_OFFSET;
> - val |= SPRD_DMA_GLB_2STAGE_EN;
> - if (schan->int_type != SPRD_DMA_NO_INT)
> - val |= SPRD_DMA_GLB_SRC_INT;
> -
> - sprd_dma_glb_update(sdev, SPRD_DMA_GLB_2STAGE_GRP2, val, val);
> + if (config_type == SPRD_DMA_2STAGE_SET) {
> + val = chn & SPRD_DMA_GLB_SRC_CHN_MASK;
> + val |= BIT(schan->trg_mode - 1) << SPRD_DMA_GLB_TRG_OFFSET;
> + val |= SPRD_DMA_GLB_2STAGE_EN;
> + if (schan->int_type & SPRD_DMA_SRC_CHN1_INT)
> + val |= SPRD_DMA_GLB_SRC_INT;
> +
> + sprd_dma_glb_update(sdev, SPRD_DMA_GLB_2STAGE_GRP2,
> + SPRD_DMA_GLB_SRC_INT |
> + SPRD_DMA_GLB_TRG_MASK |
> + SPRD_DMA_GLB_SRC_CHN_MASK, val);
> + } else {
> + sprd_dma_glb_update(sdev, SPRD_DMA_GLB_2STAGE_GRP2,
> + SPRD_DMA_GLB_SRC_INT |
> + SPRD_DMA_GLB_TRG_MASK |
> + SPRD_DMA_GLB_2STAGE_EN |
> + SPRD_DMA_GLB_SRC_CHN_MASK, 0);
> + }
> break;
>
> case SPRD_DMA_DST_CHN0:
> - val = (chn << SPRD_DMA_GLB_DEST_CHN_OFFSET) &
> - SPRD_DMA_GLB_DEST_CHN_MASK;
> - val |= SPRD_DMA_GLB_2STAGE_EN;
> - if (schan->int_type != SPRD_DMA_NO_INT)
> - val |= SPRD_DMA_GLB_DEST_INT;
> -
> - sprd_dma_glb_update(sdev, SPRD_DMA_GLB_2STAGE_GRP1, val, val);
> + if (config_type == SPRD_DMA_2STAGE_SET) {
> + val = (chn << SPRD_DMA_GLB_DEST_CHN_OFFSET) &
> + SPRD_DMA_GLB_DEST_CHN_MASK;
> + val |= SPRD_DMA_GLB_2STAGE_EN;
> + if (schan->int_type & SPRD_DMA_DST_CHN0_INT)
> + val |= SPRD_DMA_GLB_DEST_INT;
> +
> + sprd_dma_glb_update(sdev, SPRD_DMA_GLB_2STAGE_GRP1,
> + SPRD_DMA_GLB_DEST_INT |
> + SPRD_DMA_GLB_DEST_CHN_MASK, val);
> + } else {
> + sprd_dma_glb_update(sdev, SPRD_DMA_GLB_2STAGE_GRP1,
> + SPRD_DMA_GLB_DEST_INT |
> + SPRD_DMA_GLB_2STAGE_EN |
> + SPRD_DMA_GLB_DEST_CHN_MASK, 0);
> + }
> break;
>
> case SPRD_DMA_DST_CHN1:
> - val = (chn << SPRD_DMA_GLB_DEST_CHN_OFFSET) &
> - SPRD_DMA_GLB_DEST_CHN_MASK;
> - val |= SPRD_DMA_GLB_2STAGE_EN;
> - if (schan->int_type != SPRD_DMA_NO_INT)
> - val |= SPRD_DMA_GLB_DEST_INT;
> -
> - sprd_dma_glb_update(sdev, SPRD_DMA_GLB_2STAGE_GRP2, val, val);
> + if (config_type == SPRD_DMA_2STAGE_SET) {
> + val = (chn << SPRD_DMA_GLB_DEST_CHN_OFFSET) &
> + SPRD_DMA_GLB_DEST_CHN_MASK;
> + val |= SPRD_DMA_GLB_2STAGE_EN;
> + if (schan->int_type & SPRD_DMA_DST_CHN1_INT)
> + val |= SPRD_DMA_GLB_DEST_INT;
> +
> + sprd_dma_glb_update(sdev, SPRD_DMA_GLB_2STAGE_GRP2,
> + SPRD_DMA_GLB_DEST_INT |
> + SPRD_DMA_GLB_DEST_CHN_MASK, val);
> + } else {
> + sprd_dma_glb_update(sdev, SPRD_DMA_GLB_2STAGE_GRP2,
> + SPRD_DMA_GLB_DEST_INT |
> + SPRD_DMA_GLB_2STAGE_EN |
> + SPRD_DMA_GLB_DEST_CHN_MASK, 0);
> + }
> break;
>
> default:
> @@ -545,7 +593,7 @@ static void sprd_dma_start(struct sprd_dma_chn *schan)
> * Set 2-stage configuration if the channel starts one 2-stage
> * transfer.
> */
> - if (schan->chn_mode && sprd_dma_set_2stage_config(schan))
> + if (schan->chn_mode && sprd_dma_2stage_config(schan, SPRD_DMA_2STAGE_SET))
> return;
>
> /*
> @@ -569,6 +617,12 @@ static void sprd_dma_stop(struct sprd_dma_chn *schan)
> sprd_dma_set_pending(schan, false);
> sprd_dma_unset_uid(schan);
> sprd_dma_clear_int(schan);
> + /*
> + * If 2-stage transfer is used, the configuration must be clear
> + * when release DMA channel.
Please explain why? not only 'what'.
> + */
> + if (schan->chn_mode)
> + sprd_dma_2stage_config(schan, SPRD_DMA_2STAGE_CLEAR);
How to ensure backward compatibility with previous SPRD DMA IP?
> schan->cur_desc = NULL;
> }
>
> @@ -757,7 +811,9 @@ static int sprd_dma_fill_desc(struct dma_chan *chan,
> phys_addr_t llist_ptr;
>
> if (dir == DMA_MEM_TO_DEV) {
> - src_step = sprd_dma_get_step(slave_cfg->src_addr_width);
> + src_step = slave_cfg->src_port_window_size ?
> + slave_cfg->src_port_window_size :
> + sprd_dma_get_step(slave_cfg->src_addr_width);
Please also explain why.
> if (src_step < 0) {
> dev_err(sdev->dma_dev.dev, "invalid source step\n");
> return src_step;
> @@ -773,7 +829,9 @@ static int sprd_dma_fill_desc(struct dma_chan *chan,
> else
> dst_step = SPRD_DMA_NONE_STEP;
> } else {
> - dst_step = sprd_dma_get_step(slave_cfg->dst_addr_width);
> + dst_step = slave_cfg->dst_port_window_size ?
> + slave_cfg->dst_port_window_size :
> + sprd_dma_get_step(slave_cfg->dst_addr_width);
> if (dst_step < 0) {
> dev_err(sdev->dma_dev.dev, "invalid destination step\n");
> return dst_step;
^ permalink raw reply [flat|nested] 2+ messages in thread
end of thread, other threads:[~2023-08-07 8:27 UTC | newest]
Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2023-08-07 5:20 [PATCH 3/5] dma: optimize two stage transfer function Kaiwei Liu
2023-08-07 8:27 ` Baolin Wang
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox
all inboxes | Powered by JetHome®