Skip to contentSkip to conversation

mmc: dw_mmc: Fix the max_blk_count in IDMAC

PATCH New Yuvaraj Kumar C D 2 participants 3 messages Mar 24, 2014

Subsystems (3)
Arm Samsung Exynos MMC, SD and SDIO Synopsys DesignWare MMC/SD/SDIO driver
lore
Share & tools
lore

Yuvaraj Kumar C D

author
#0 lore raw

PATCH 1/1 mmc: dw_mmc: Fix the max_blk_count in IDMAC

From: Alim Akhtar <[email protected]>

max_blk_count is currently set incorrectly, and the correct value can be determined by looking at the maximum number of bytes which can be transferred and the block size. We use the maximum block size to determine the minimum number of blocks that the controller should support. It could probably do more with a smaller block size but this value should be sufficient for good performance.

This improves sequential performance by 82% on write and 6% read on one particular device.

TEST= with this patch

	time dd if=/dev/zero of=/usr/local/100 bs=1M count=100 conv=fdatasync
	100+0 records in
	100+0 records out
	104857600 bytes (105 MB) copied, 2.62541 s, 39.9 MB/s

	real    0m2.638s
	user    0m0.000s
	sys     0m0.645s

	without this patch:
	time dd if=/dev/zero of=/usr/local/100 bs=1M count=100 conv=fdatasync
	100+0 records in
	100+0 records out
	104857600 bytes (105 MB) copied, 3.25873 s, 32.2 MB/s

	real    0m3.265s
	user    0m0.005s
	sys     0m0.690s
Signed-off-by Seungwon Jeon <[email protected]>
Signed-off-by Alim Akhtar <[email protected]>
Signed-off-by Sonny Rao <[email protected]>
Signed-off-by Yuvaraj Kumar C D <[email protected]>
---
 drivers/mmc/host/dw_mmc.c |   18 ++++++++++++++++--
 1 file changed, 16 insertions(+), 2 deletions(-)
drivers/mmc/host/dw_mmc.c +16 -2
index 0c56faa..2fc4030 100644
@@ -2213,10 +2213,24 @@ static int dw_mci_init_slot(struct dw_mci *host, unsigned int id)
 		/* Useful defaults if platform data is unset. */
 #ifdef CONFIG_MMC_DW_IDMAC
 		mmc->max_segs = host->ring_size;
+
+		/* the BLKSIZ register is 16-bits wide */
 		mmc->max_blk_size = 65536;
-		mmc->max_blk_count = host->ring_size;
+
+		/*
+		 * This value is calculated by taking the size of the
+		 * 32-bit BYTCNT (byte count) register and dividing by the
+		 * BLKSIZ (block size) register.  This is the minimum number
+		 * of blocks which could be handled.
+		 */
+		mmc->max_blk_count = 0xFFFF;
 		mmc->max_seg_size = 0x1000;
-		mmc->max_req_size = mmc->max_seg_size * mmc->max_blk_count;
+		/*
+		 * Maximum request size should be total number of descriptors
+		 * times the maximum amount of data each can reference
+		 */
+
+		mmc->max_req_size = mmc->max_seg_size * mmc->max_segs;
 #else
 		mmc->max_segs = 64;
 		mmc->max_blk_size = 65536; /* BLKSIZ is 16 bits */
Signature · 1 line
1.7.10.4
maintainer
#1 lore raw

PATCH 1/1 mmc: dw_mmc: Fix the max_blk_count in IDMAC

Hi,

Did you know that Seungwon has sent the patch "[PATCH 1/7] mmc: dw_mmc: fix the max_blk_count in IDMAC"? I think this patch looks like same, isn't it?
Plz check it.

Best Regards,
Jaehoon Chung

Quoted 72 lines · Yuvaraj Kumar C D

From: Alim Akhtar <[email protected]>

max_blk_count is currently set incorrectly, and the correct value can be determined by looking at the maximum number of bytes which can be transferred and the block size. We use the maximum block size to determine the minimum number of blocks that the controller should support. It could probably do more with a smaller block size but this value should be sufficient for good performance.

This improves sequential performance by 82% on write and 6% read on one particular device.

TEST= with this patch

	time dd if=/dev/zero of=/usr/local/100 bs=1M count=100 conv=fdatasync
	100+0 records in
	100+0 records out
	104857600 bytes (105 MB) copied, 2.62541 s, 39.9 MB/s

	real    0m2.638s
	user    0m0.000s
	sys     0m0.645s

	without this patch:
	time dd if=/dev/zero of=/usr/local/100 bs=1M count=100 conv=fdatasync
	100+0 records in
	100+0 records out
	104857600 bytes (105 MB) copied, 3.25873 s, 32.2 MB/s

	real    0m3.265s
	user    0m0.005s
	sys     0m0.690s
Signed-off-by Seungwon Jeon <[email protected]>
Signed-off-by Alim Akhtar <[email protected]>
Signed-off-by Sonny Rao <[email protected]>
Signed-off-by Yuvaraj Kumar C D <[email protected]>
---
 drivers/mmc/host/dw_mmc.c |   18 ++++++++++++++++--
 1 file changed, 16 insertions(+), 2 deletions(-)
drivers/mmc/host/dw_mmc.c +16 -2
index 0c56faa..2fc4030 100644
@@ -2213,10 +2213,24 @@ static int dw_mci_init_slot(struct dw_mci *host, unsigned int id)
 		/* Useful defaults if platform data is unset. */
 #ifdef CONFIG_MMC_DW_IDMAC
 		mmc->max_segs = host->ring_size;
+
+		/* the BLKSIZ register is 16-bits wide */
 		mmc->max_blk_size = 65536;
-		mmc->max_blk_count = host->ring_size;
+
+		/*
+		 * This value is calculated by taking the size of the
+		 * 32-bit BYTCNT (byte count) register and dividing by the
+		 * BLKSIZ (block size) register.  This is the minimum number
+		 * of blocks which could be handled.
+		 */
+		mmc->max_blk_count = 0xFFFF;
 		mmc->max_seg_size = 0x1000;
-		mmc->max_req_size = mmc->max_seg_size * mmc->max_blk_count;
+		/*
+		 * Maximum request size should be total number of descriptors
+		 * times the maximum amount of data each can reference
+		 */
+
+		mmc->max_req_size = mmc->max_seg_size * mmc->max_segs;
 #else
 		mmc->max_segs = 64;
 		mmc->max_blk_size = 65536; /* BLKSIZ is 16 bits */
author
#2 lore raw

PATCH 1/1 mmc: dw_mmc: Fix the max_blk_count in IDMAC

Quoted 4 lines · Jaehoon Chung

Hi,

Did you know that Seungwon has sent the patch "[PATCH 1/7] mmc: dw_mmc: fix the max_blk_count in IDMAC"? I think this patch looks like same, isn't it?

Yes.Its the same one.Please discard this patch.

Quoted 79 lines

Plz check it.

Best Regards,
Jaehoon Chung

Quoted 72 lines · Yuvaraj Kumar C D

From: Alim Akhtar <[email protected]>

max_blk_count is currently set incorrectly, and the correct value can be determined by looking at the maximum number of bytes which can be transferred and the block size. We use the maximum block size to determine the minimum number of blocks that the controller should support. It could probably do more with a smaller block size but this value should be sufficient for good performance.

This improves sequential performance by 82% on write and 6% read on one particular device.

TEST= with this patch

      time dd if=/dev/zero of=/usr/local/100 bs=1M count=100 conv=fdatasync
      100+0 records in
      100+0 records out
      104857600 bytes (105 MB) copied, 2.62541 s, 39.9 MB/s

      real    0m2.638s
      user    0m0.000s
      sys     0m0.645s

      without this patch:
      time dd if=/dev/zero of=/usr/local/100 bs=1M count=100 conv=fdatasync
      100+0 records in
      100+0 records out
      104857600 bytes (105 MB) copied, 3.25873 s, 32.2 MB/s

      real    0m3.265s
      user    0m0.005s
      sys     0m0.690s
Signed-off-by Seungwon Jeon <[email protected]>
Signed-off-by Alim Akhtar <[email protected]>
Signed-off-by Sonny Rao <[email protected]>
Signed-off-by Yuvaraj Kumar C D <[email protected]>
---
 drivers/mmc/host/dw_mmc.c |   18 ++++++++++++++++--
 1 file changed, 16 insertions(+), 2 deletions(-)
drivers/mmc/host/dw_mmc.c +16 -2
index 0c56faa..2fc4030 100644
@@ -2213,10 +2213,24 @@ static int dw_mci_init_slot(struct dw_mci *host, unsigned int id)
              /* Useful defaults if platform data is unset. */
 #ifdef CONFIG_MMC_DW_IDMAC
              mmc->max_segs = host->ring_size;
+
+             /* the BLKSIZ register is 16-bits wide */
              mmc->max_blk_size = 65536;
-             mmc->max_blk_count = host->ring_size;
+
+             /*
+              * This value is calculated by taking the size of the
+              * 32-bit BYTCNT (byte count) register and dividing by the
+              * BLKSIZ (block size) register.  This is the minimum number
+              * of blocks which could be handled.
+              */
+             mmc->max_blk_count = 0xFFFF;
              mmc->max_seg_size = 0x1000;
-             mmc->max_req_size = mmc->max_seg_size * mmc->max_blk_count;
+             /*
+              * Maximum request size should be total number of descriptors
+              * times the maximum amount of data each can reference
+              */
+
+             mmc->max_req_size = mmc->max_seg_size * mmc->max_segs;
 #else
              mmc->max_segs = 64;
              mmc->max_blk_size = 65536; /* BLKSIZ is 16 bits */