[PATCH v2 0/2] mmc: renesas_internal_dmac: add pre_req and post_req support

[PATCH v2 0/2] mmc: renesas_internal_dmac: add pre_req and post_req support

[Yoshihiro Shimoda]

To improve performance, add pre_req and post_req support into
renesas_internal_dmac. Note that the patch [1/2] is a bug fix patch.

Changes from v1:
 - Add a bug fix patch for hardware limitation in patch 1.
 - Remove a redundant argument in the unmap function in patch 2.
 - Describe the reason why we should not use host->sg_ptr in patch 2.
 - Rename argument of struct mmc_request from req to mrq in patch 2.

Takeshi Saito (1):
  mmc: renesas_sdhi_internal_dmac: Fix DMA buffer alignment from 8 to
    128-bytes

Yoshihiro Shimoda (1):
  mmc: renesas_internal_dmac: add pre_req and post_req support

 drivers/mmc/host/renesas_sdhi_internal_dmac.c | 91 ++++++++++++++++++++++++---
 1 file changed, 83 insertions(+), 8 deletions(-)

-- 
2.7.4

[PATCH v2 1/2] mmc: renesas_sdhi_internal_dmac: Fix DMA buffer alignment from 8 to 128-bytes

[Yoshihiro Shimoda]

From: Takeshi Saito <takeshi.saito.xv@renesas.com>

According to the latest datasheet, the internal DMAC buffer alignment
R-Car Gen3 SDHI HW should be 128-bytes. So, fix it.

Signed-off-by: Takeshi Saito <takeshi.saito.xv@renesas.com>
[shimoda: revise commit description, rebase]
Fixes: 2a68ea7896e3 ("mmc: renesas-sdhi: add support for R-Car Gen3 SDHI DMAC")
Signed-off-by: Yoshihiro Shimoda <yoshihiro.shimoda.uh@renesas.com>
---
 drivers/mmc/host/renesas_sdhi_internal_dmac.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/drivers/mmc/host/renesas_sdhi_internal_dmac.c b/drivers/mmc/host/renesas_sdhi_internal_dmac.c
index fe13e1e..f3e76d6 100644
--- a/drivers/mmc/host/renesas_sdhi_internal_dmac.c
+++ b/drivers/mmc/host/renesas_sdhi_internal_dmac.c
@@ -186,8 +186,8 @@ renesas_sdhi_internal_dmac_start_dma(struct tmio_mmc_host *host,
 			mmc_get_dma_dir(data)))
 		goto force_pio;
 
-	/* This DMAC cannot handle if buffer is not 8-bytes alignment */
-	if (!IS_ALIGNED(sg_dma_address(sg), 8))
+	/* This DMAC cannot handle if buffer is not 128-bytes alignment */
+	if (!IS_ALIGNED(sg_dma_address(sg), 128))
 		goto force_pio_with_unmap;
 
 	if (data->flags & MMC_DATA_READ) {
-- 
2.7.4

Re: [PATCH v2 1/2] mmc: renesas_sdhi_internal_dmac: Fix DMA buffer alignment from 8 to 128-bytes

[Wolfram Sang]

[-- Attachment #1: Type: text/plain, Size: 649 bytes --]

On Wed, Dec 16, 2020 at 07:29:31PM +0900, Yoshihiro Shimoda wrote:
> From: Takeshi Saito <takeshi.saito.xv@renesas.com>
> 
> According to the latest datasheet, the internal DMAC buffer alignment
> R-Car Gen3 SDHI HW should be 128-bytes. So, fix it.
> 
> Signed-off-by: Takeshi Saito <takeshi.saito.xv@renesas.com>
> [shimoda: revise commit description, rebase]
> Fixes: 2a68ea7896e3 ("mmc: renesas-sdhi: add support for R-Car Gen3 SDHI DMAC")
> Signed-off-by: Yoshihiro Shimoda <yoshihiro.shimoda.uh@renesas.com>

Reviewed-by: Wolfram Sang <wsa+renesas@sang-engineering.com>
Tested-by: Wolfram Sang <wsa+renesas@sang-engineering.com>


[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

[PATCH v2 2/2] mmc: renesas_internal_dmac: add pre_req and post_req support

[Yoshihiro Shimoda]

Add pre_req and post_req support to improve performance.

Inspired by a patch in the BSP by Masaharu Hayakawa.

Signed-off-by: Yoshihiro Shimoda <yoshihiro.shimoda.uh@renesas.com>
---
 drivers/mmc/host/renesas_sdhi_internal_dmac.c | 91 ++++++++++++++++++++++++---
 1 file changed, 83 insertions(+), 8 deletions(-)

diff --git a/drivers/mmc/host/renesas_sdhi_internal_dmac.c b/drivers/mmc/host/renesas_sdhi_internal_dmac.c
index f3e76d6..ff97f15 100644
--- a/drivers/mmc/host/renesas_sdhi_internal_dmac.c
+++ b/drivers/mmc/host/renesas_sdhi_internal_dmac.c
@@ -56,6 +56,12 @@
 #define INFO2_DTRANERR1		BIT(17)
 #define INFO2_DTRANERR0		BIT(16)
 
+enum renesas_sdhi_dma_cookie {
+	COOKIE_UNMAPPED,
+	COOKIE_PRE_MAPPED,
+	COOKIE_MAPPED,
+};
+
 /*
  * Specification of this driver:
  * - host->chan_{rx,tx} will be used as a flag of enabling/disabling the dma
@@ -172,6 +178,50 @@ renesas_sdhi_internal_dmac_dataend_dma(struct tmio_mmc_host *host) {
 	tasklet_schedule(&priv->dma_priv.dma_complete);
 }
 
+/*
+ * renesas_sdhi_internal_dmac_map() will be called with two difference
+ * sg pointers in two mmc_data by .pre_req(), but tmio host can have a single
+ * sg_ptr only. So, renesas_sdhi_internal_dmac_{un}map() should use a sg
+ * pointer in a mmc_data instead of host->sg_ptr.
+ */
+static void
+renesas_sdhi_internal_dmac_unmap(struct tmio_mmc_host *host,
+				 struct mmc_data *data,
+				 enum renesas_sdhi_dma_cookie cookie)
+{
+	bool unmap = cookie == COOKIE_UNMAPPED ? (data->host_cookie != cookie) :
+						 (data->host_cookie == cookie);
+
+	if (unmap) {
+		dma_unmap_sg(&host->pdev->dev, data->sg, data->sg_len,
+			     mmc_get_dma_dir(data));
+		data->host_cookie = COOKIE_UNMAPPED;
+	}
+}
+
+static bool
+renesas_sdhi_internal_dmac_map(struct tmio_mmc_host *host,
+			       struct mmc_data *data,
+			       enum renesas_sdhi_dma_cookie cookie)
+{
+	if (data->host_cookie == COOKIE_PRE_MAPPED)
+		return true;
+
+	if (!dma_map_sg(&host->pdev->dev, data->sg, data->sg_len,
+			    mmc_get_dma_dir(data)))
+		return false;
+
+	data->host_cookie = cookie;
+
+	/* This DMAC cannot handle if buffer is not 128-bytes alignment */
+	if (!IS_ALIGNED(sg_dma_address(data->sg), 128)) {
+		renesas_sdhi_internal_dmac_unmap(host, data, cookie);
+		return false;
+	}
+
+	return true;
+}
+
 static void
 renesas_sdhi_internal_dmac_start_dma(struct tmio_mmc_host *host,
 				     struct mmc_data *data)
@@ -182,14 +232,9 @@ renesas_sdhi_internal_dmac_start_dma(struct tmio_mmc_host *host,
 	if (!test_bit(SDHI_INTERNAL_DMAC_ADDR_MODE_FIXED_ONLY, &global_flags))
 		dtran_mode |= DTRAN_MODE_ADDR_MODE;
 
-	if (!dma_map_sg(&host->pdev->dev, sg, host->sg_len,
-			mmc_get_dma_dir(data)))
+	if (!renesas_sdhi_internal_dmac_map(host, data, COOKIE_MAPPED))
 		goto force_pio;
 
-	/* This DMAC cannot handle if buffer is not 128-bytes alignment */
-	if (!IS_ALIGNED(sg_dma_address(sg), 128))
-		goto force_pio_with_unmap;
-
 	if (data->flags & MMC_DATA_READ) {
 		dtran_mode |= DTRAN_MODE_CH_NUM_CH1;
 		if (test_bit(SDHI_INTERNAL_DMAC_ONE_RX_ONLY, &global_flags) &&
@@ -212,7 +257,7 @@ renesas_sdhi_internal_dmac_start_dma(struct tmio_mmc_host *host,
 	return;
 
 force_pio_with_unmap:
-	dma_unmap_sg(&host->pdev->dev, sg, host->sg_len, mmc_get_dma_dir(data));
+	renesas_sdhi_internal_dmac_unmap(host, data, COOKIE_UNMAPPED);
 
 force_pio:
 	renesas_sdhi_internal_dmac_enable_dma(host, false);
@@ -245,7 +290,7 @@ static bool renesas_sdhi_internal_dmac_complete(struct tmio_mmc_host *host)
 		dir = DMA_TO_DEVICE;
 
 	renesas_sdhi_internal_dmac_enable_dma(host, false);
-	dma_unmap_sg(&host->pdev->dev, host->sg_ptr, host->sg_len, dir);
+	renesas_sdhi_internal_dmac_unmap(host, host->data, COOKIE_MAPPED);
 
 	if (dir == DMA_FROM_DEVICE)
 		clear_bit(SDHI_INTERNAL_DMAC_RX_IN_USE, &global_flags);
@@ -274,6 +319,32 @@ static void renesas_sdhi_internal_dmac_end_dma(struct tmio_mmc_host *host)
 		renesas_sdhi_internal_dmac_complete(host);
 }
 
+static void renesas_sdhi_internal_dmac_post_req(struct mmc_host *mmc,
+						struct mmc_request *mrq,
+						int err)
+{
+	struct tmio_mmc_host *host = mmc_priv(mmc);
+	struct mmc_data *data = mrq->data;
+
+	if (!data)
+		return;
+
+	renesas_sdhi_internal_dmac_unmap(host, data, COOKIE_UNMAPPED);
+}
+
+static void renesas_sdhi_internal_dmac_pre_req(struct mmc_host *mmc,
+					       struct mmc_request *mrq)
+{
+	struct tmio_mmc_host *host = mmc_priv(mmc);
+	struct mmc_data *data = mrq->data;
+
+	if (!data)
+		return;
+
+	data->host_cookie = COOKIE_UNMAPPED;
+	renesas_sdhi_internal_dmac_map(host, data, COOKIE_PRE_MAPPED);
+}
+
 static void
 renesas_sdhi_internal_dmac_request_dma(struct tmio_mmc_host *host,
 				       struct tmio_mmc_data *pdata)
@@ -295,6 +366,10 @@ renesas_sdhi_internal_dmac_request_dma(struct tmio_mmc_host *host,
 	tasklet_init(&host->dma_issue,
 		     renesas_sdhi_internal_dmac_issue_tasklet_fn,
 		     (unsigned long)host);
+
+	/* Add pre_req and post_req */
+	host->ops.pre_req = renesas_sdhi_internal_dmac_pre_req;
+	host->ops.post_req = renesas_sdhi_internal_dmac_post_req;
 }
 
 static void
-- 
2.7.4

Re: [PATCH v2 2/2] mmc: renesas_internal_dmac: add pre_req and post_req support

[Wolfram Sang]

[-- Attachment #1: Type: text/plain, Size: 812 bytes --]

On Wed, Dec 16, 2020 at 07:29:32PM +0900, Yoshihiro Shimoda wrote:
> Add pre_req and post_req support to improve performance.
> 
> Inspired by a patch in the BSP by Masaharu Hayakawa.
> 
> Signed-off-by: Yoshihiro Shimoda <yoshihiro.shimoda.uh@renesas.com>

So, I tested this on H3 ES2.0 and M3-N with interesting results. H3
ES2.0 has a significant speedup:

with: 254MB/s, without 207MB/s

M3-N speedup is smaller but general throughput is much better:

with: 262MB/s, without 250MB/s

Same kernel, both systems idle otherwise. I don't think it has to do
with the number of CPU cores, but more that the SDHI revision is a tad
newer maybe?

Patch looks good, too:

Reviewed-by: Wolfram Sang <wsa+renesas@sang-engineering.com>
Tested-by: Wolfram Sang <wsa+renesas@sang-engineering.com>


[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

Re: [PATCH v2 0/2] mmc: renesas_internal_dmac: add pre_req and post_req support

[Ulf Hansson]

On Wed, 16 Dec 2020 at 11:29, Yoshihiro Shimoda
<yoshihiro.shimoda.uh@renesas.com> wrote:
>
> To improve performance, add pre_req and post_req support into
> renesas_internal_dmac. Note that the patch [1/2] is a bug fix patch.
>
> Changes from v1:
>  - Add a bug fix patch for hardware limitation in patch 1.
>  - Remove a redundant argument in the unmap function in patch 2.
>  - Describe the reason why we should not use host->sg_ptr in patch 2.
>  - Rename argument of struct mmc_request from req to mrq in patch 2.
>
> Takeshi Saito (1):
>   mmc: renesas_sdhi_internal_dmac: Fix DMA buffer alignment from 8 to
>     128-bytes
>
> Yoshihiro Shimoda (1):
>   mmc: renesas_internal_dmac: add pre_req and post_req support
>
>  drivers/mmc/host/renesas_sdhi_internal_dmac.c | 91 ++++++++++++++++++++++++---
>  1 file changed, 83 insertions(+), 8 deletions(-)
>

Applied for next, thanks!

Kind regards
Uffe