Skip to contentSkip to conversation

sdio: move temporary buffer

RFC PATCH In review Marc Pignat 2 participants 3 messages Dec 19, 2007 – Dec 20, 2007 MMC, SD and SDIO

lore
Share & tools
lore

Marc Pignat

author
#0 lore raw

RFC PATCH 1/1 sdio: move temporary buffer

Move the temporary buffer used by some sdio functions from sdio_func struct to the mmc_host struct and make it dma-safe.

Signed-off-by Marc Pignat <[email protected]>

---

There is no need to have this temporary buffer in every sdio_func, since all access to it are serialized by sdio_[claim/release]_host.

I also think it should be ____cacheline_aligned to make sure we have no data loss when doing dma syncing operations.

Comments welcome!

Regards

Marc

sdio-dma-tmpbuf-patch (against 2.6.24-rc5)

 drivers/mmc/core/sdio_io.c    |   20 ++++++++++++--------
 include/linux/mmc/host.h      |    3 +++
 include/linux/mmc/sdio_func.h |    2 --
 3 files changed, 15 insertions(+), 10 deletions(-)
drivers/mmc/core/sdio_io.c +12 -8
index 625b92c..e48663e 100644
@@ -389,18 +389,19 @@ unsigned short sdio_readw(struct sdio_func *func, unsigned int addr,
 	int *err_ret)
 {
 	int ret;
+	u8 *tmpbuf = func->card->host->tmpbuf;
 
 	if (err_ret)
 		*err_ret = 0;
 
-	ret = sdio_memcpy_fromio(func, func->tmpbuf, addr, 2);
+	ret = sdio_memcpy_fromio(func, tmpbuf, addr, 2);
 	if (ret) {
 		if (err_ret)
 			*err_ret = ret;
 		return 0xFFFF;
 	}
 
-	return le16_to_cpu(*(u16*)func->tmpbuf);
+	return le16_to_cpu(*(u16*)tmpbuf);
 }
 EXPORT_SYMBOL_GPL(sdio_readw);
 

@@ -419,10 +420,11 @@ void sdio_writew(struct sdio_func *func, unsigned short b, unsigned int addr,
 	int *err_ret)
 {
 	int ret;
+	u8 *tmpbuf = func->card->host->tmpbuf;
 
-	*(u16*)func->tmpbuf = cpu_to_le16(b);
+	*(u16*)tmpbuf = cpu_to_le16(b);
 
-	ret = sdio_memcpy_toio(func, addr, func->tmpbuf, 2);
+	ret = sdio_memcpy_toio(func, addr, tmpbuf, 2);
 	if (err_ret)
 		*err_ret = ret;
 }

@@ -443,18 +445,19 @@ unsigned long sdio_readl(struct sdio_func *func, unsigned int addr,
 	int *err_ret)
 {
 	int ret;
+	u8 *tmpbuf = func->card->host->tmpbuf;
 
 	if (err_ret)
 		*err_ret = 0;
 
-	ret = sdio_memcpy_fromio(func, func->tmpbuf, addr, 4);
+	ret = sdio_memcpy_fromio(func, tmpbuf, addr, 4);
 	if (ret) {
 		if (err_ret)
 			*err_ret = ret;
 		return 0xFFFFFFFF;
 	}
 
-	return le32_to_cpu(*(u32*)func->tmpbuf);
+	return le32_to_cpu(*(u32*)tmpbuf);
 }
 EXPORT_SYMBOL_GPL(sdio_readl);
 

@@ -473,10 +476,11 @@ void sdio_writel(struct sdio_func *func, unsigned long b, unsigned int addr,
 	int *err_ret)
 {
 	int ret;
+	u8 *tmpbuf = func->card->host->tmpbuf;
 
-	*(u32*)func->tmpbuf = cpu_to_le32(b);
+	*(u32*)tmpbuf = cpu_to_le32(b);
 
-	ret = sdio_memcpy_toio(func, addr, func->tmpbuf, 4);
+	ret = sdio_memcpy_toio(func, addr, tmpbuf, 4);
 	if (err_ret)
 		*err_ret = ret;
 }
include/linux/mmc/host.h +3 -0
index 125eee1..8edb796 100644
@@ -139,6 +139,9 @@ struct mmc_host {
 	struct led_trigger	*led;		/* activity led */
 #endif
 
+	/* DMA:able scratch buffer */
+	u8			tmpbuf[4] ____cacheline_aligned;
+
 	unsigned long		private[0] ____cacheline_aligned;
 };
include/linux/mmc/sdio_func.h +0 -2
index b050f4d..4a797a7 100644
@@ -49,8 +49,6 @@ struct sdio_func {
 	unsigned int		state;		/* function state */
 #define SDIO_STATE_PRESENT	(1<<0)		/* present in sysfs */
 
-	u8			tmpbuf[4];	/* DMA:able scratch buffer */
-
 	unsigned		num_info;	/* number of info strings */
 	const char		**info;		/* info strings */
#1 lore raw
Quoted 9 lines · On Wed, 19 Dec 2007 11:55:25 +0100 Marc Pignat <[email protected]> wrote:

Move the temporary buffer used by some sdio functions from sdio_func struct to the mmc_host struct and make it dma-safe.

Signed-off-by Marc Pignat <[email protected]>

---

There is no need to have this temporary buffer in every sdio_func, since all access to it are serialized by sdio_[claim/release]_host.

I had some crazy idea about per function locking, but until that time this will do just fine.

Quoted 2 lines

I also think it should be ____cacheline_aligned to make sure we have no data loss when doing dma syncing operations.

This is still voodoo to me, so I'll have to take your word for it. :)

Quoted 3 lines

Comments welcome!

Looks good to me.

Rgds

Signature · 5 lines
     -- Pierre Ossman

  Linux kernel, MMC maintainer        http://www.kernel.org
  PulseAudio, core developer          http://pulseaudio.org
  rdesktop, core developer          http://www.rdesktop.org
author
#2 lore raw
Quoted 1 line · Pierre Ossman

On Wed, 19 Dec 2007 11:55:25 +0100

...

Quoted 1 line

This is still voodoo to me, so I'll have to take your word for it. :)

And almost voodoo to me... this is why I hope an ack from a DMA mentor!

Marc