Skip to contentSkip to conversation

mmc: align data size for host which only supports power-of-2 block

PATCH In review Bryan Wu 3 participants 8 messages Mar 6, 2009 – Apr 10, 2009 MMC, SD and SDIO

lore
Share & tools
lore

Bryan Wu

author
#0 lore raw

PATCH 1/1 mmc: align data size for host which only supports power-of-2 block

From: Cliff Cai <[email protected]>

---
 drivers/mmc/core/core.c  |    8 +++++++-
 include/linux/mmc/host.h |    1 +
 2 files changed, 8 insertions(+), 1 deletions(-)
drivers/mmc/core/core.c +7 -1
index df6ce4a..15119df 100644
@@ -321,7 +321,13 @@ unsigned int mmc_align_data_size(struct mmc_card *card, unsigned int sz)
 	 * the core about its problems yet, so for now we just 32-bit
 	 * align the size.
 	 */
-	sz = ((sz + 3) / 4) * 4;
+
+	/* Align size for host which only supports power-of-2 block */
+	if (card->host->powerof2_block) {
+		if (sz & (sz - 1))
+			sz = 1 << fls(sz);
+	} else
+		sz = ((sz + 3) / 4) * 4;
 
 	return sz;
 }
include/linux/mmc/host.h +1 -0
index 4e45725..7416ed1 100644
@@ -162,6 +162,7 @@ struct mmc_host {
 	struct dentry		*debugfs_root;
 
 	unsigned long		private[0] ____cacheline_aligned;
+	unsigned int		powerof2_block;	/* host only supports power-of-2 block */
 };
 
 extern struct mmc_host *mmc_alloc_host(int extra, struct device *);
Signature · 1 line
1.5.6.3
#1 lore raw
Quoted 5 lines · On Fri, 6 Mar 2009 11:15:21 +0800 Bryan Wu <[email protected]> wrote:

From: Cliff Cai <[email protected]>

---

This patch seems premature as there is no associated modification of any of the host drivers.

Quoted 13 lines
 drivers/mmc/core/core.c  |    8 +++++++-
 include/linux/mmc/host.h |    1 +
 2 files changed, 8 insertions(+), 1 deletions(-)
drivers/mmc/core/core.c +0 -1
index df6ce4a..15119df 100644
@@ -321,7 +321,13 @@ unsigned int mmc_align_data_size(struct mmc_card *card, unsigned int sz)
 	 * the core about its problems yet, so for now we just 32-bit
 	 * align the size.
 	 */
-	sz = ((sz + 3) / 4) * 4;

drivers/mmc/core/core.c unsigned int mmc_align_data_size(struct mmc_card *card, unsigned int sz) · line 333

Quoted 10 lines
drivers/mmc/core/core.c +7 -0
+
+	/* Align size for host which only supports power-of-2 block */
+	if (card->host->powerof2_block) {
+		if (sz & (sz - 1))
+			sz = 1 << fls(sz);
+	} else
+		sz = ((sz + 3) / 4) * 4;
 
 	return sz;
 }

At the very least, the comment at the top of this function must go. But really, if we want to improve this we should probably do it properly and have flags for the different limitations that are available.

Padding to a power of two size can also mean a rather large padding. We might need to check the host data limits after doing the adjustment.

include/linux/mmc/host.h struct mmc_host { · line 168

Quoted 12 lines
include/linux/mmc/host.h +1 -0
index 4e45725..7416ed1 100644
@@ -162,6 +162,7 @@ struct mmc_host {
 	struct dentry		*debugfs_root;
 
 	unsigned long		private[0] ____cacheline_aligned;
+	unsigned int		powerof2_block;	/* host only supports power-of-2 block */
 };
 
 extern struct mmc_host *mmc_alloc_host(int extra, struct device *);

This is just broken. Putting it after private completely breaks accesses to the host driver private data.

Also, there are a whole bunch of alignment issues that can occur. We need some kind of flags field or a bitfield for this.

Rgds

Signature · 9 lines
     -- Pierre Ossman

  Linux kernel, MMC maintainer        http://www.kernel.org
  rdesktop, core developer          http://www.rdesktop.org

  WARNING: This correspondence is being monitored by the
  Swedish government. Make sure your server uses encryption
  for SMTP traffic and consider using PGP for end-to-end
  encryption.
#2 lore raw
Quoted 16 lines

-----Original Message-----
From: Pierre Ossman [mailto:[email protected]]
Sent: Sunday, March 15, 2009 4:31 AM
To: Bryan Wu
Cc: [email protected]; Cai, Cliff; Bryan Wu
Subject: Re: [PATCH] mmc: align data size for host which only supports power-of-2 block

On Fri, 6 Mar 2009 11:15:21 +0800
Bryan Wu <[email protected]> wrote:

Quoted 8 lines
Quoted 5 lines

From: Cliff Cai <[email protected]>

---

This patch seems premature as there is no associated modification of any of the host drivers.

The SDH on Blackfin BF54x/BF51x only supports power-of-2 block transfer,we will soon
Send the sdh driver to mainline as well.

Quoted 23 lines

drivers/mmc/core/core.c unsigned int mmc_align_data_size(struct

Quoted 10 lines
 drivers/mmc/core/core.c  |    8 +++++++-
 include/linux/mmc/host.h |    1 +
 2 files changed, 8 insertions(+), 1 deletions(-)
drivers/mmc/core/core.c index renamed from drivers/mmc/core/core.c +0 -0

df6ce4a..15119df 100644

drivers/mmc/core/core.c +0 -0
@@ -321,7 +321,13 @@ unsigned int mmc_align_data_size(struct 

mmc_card *card, unsigned int sz)

Quoted 12 lines
drivers/mmc/core/core.c +7 -1
 	 * the core about its problems yet, so for now we just 32-bit
 	 * align the size.
 	 */
-	sz = ((sz + 3) / 4) * 4;
+
+	/* Align size for host which only supports power-of-2 block */
+	if (card->host->powerof2_block) {
+		if (sz & (sz - 1))
+			sz = 1 << fls(sz);
+	} else
+		sz = ((sz + 3) / 4) * 4;
 
Quoted 10 lines

drivers/mmc/core/core.c unsigned int mmc_align_data_size(struct mmc_card *card, unsigned int sz) · line 333

Quoted 2 lines
drivers/mmc/core/core.c +0 -0
 	return sz;
 }

At the very least, the comment at the top of this function must go. But really, if we want to improve this we should probably do it properly and have flags for the different limitations that are available.
Padding to a power of two size can also mean a rather large padding. We might need to check the host data limits after doing the adjustment.

Yes,but it's the higher level driver's job.

Quoted 12 lines

Review

Quoted 2 lines
(quoted diff) +0 -0

b/include/linux/mmc/host.h index

include/linux/mmc/host.h struct mmc_host { · line 165

Quoted 8 lines

4e45725..7416ed1 100644

include/linux/mmc/host.h +1 -0
@@ -162,6 +162,7 @@ struct mmc_host {
 	struct dentry		*debugfs_root;
 
 	unsigned long		private[0] ____cacheline_aligned;
+	unsigned int		powerof2_block;	/* host only 

supports power-of-2 block */

Quoted 6 lines

include/linux/mmc/host.h struct mmc_host { · line 168

Quoted 3 lines
include/linux/mmc/host.h +0 -0
 };
 
 extern struct mmc_host *mmc_alloc_host(int extra, struct device *);

This is just broken. Putting it after private completely breaks accesses to the host driver private data.

Sure,it should be added before private.

Thanks

Cliff

#3 lore raw
Quoted 8 lines · Cliff
Quoted 3 lines

This patch seems premature as there is no associated modification of any of the host drivers.

The SDH on Blackfin BF54x/BF51x only supports power-of-2 block transfer,we will soon
Send the sdh driver to mainline as well.

Ok. Looking forward to it. :)

Quoted 6 lines
Quoted 3 lines

Padding to a power of two size can also mean a rather large padding. We might need to check the host data limits after doing the adjustment.

Yes,but it's the higher level driver's job.

mmc_align_data_size() shouldn't be adjusting the size to something it can easily determine that it's invalid.

Rgds

Signature · 9 lines
     -- Pierre Ossman

  Linux kernel, MMC maintainer        http://www.kernel.org
  rdesktop, core developer          http://www.rdesktop.org

  WARNING: This correspondence is being monitored by the
  Swedish government. Make sure your server uses encryption
  for SMTP traffic and consider using PGP for end-to-end
  encryption.
#4 lore raw
Quoted 31 lines

-----Original Message-----
From: Pierre Ossman [mailto:[email protected]]
Sent: Thursday, March 26, 2009 3:24 AM
To: Cai, Cliff
Cc: Bryan Wu; [email protected]
Subject: Re: [PATCH] mmc: align data size for host which only supports power-of-2 block

Quoted 2 lines · Cliff
Quoted 2 lines

This patch seems premature as there is no associated

modification of

Quoted 5 lines
Quoted 1 line

any of the host drivers.

The SDH on Blackfin BF54x/BF51x only supports power-of-2 block transfer,we will soon Send the sdh driver to mainline as well.

Ok. Looking forward to it. :)

Quoted 4 lines
Quoted 3 lines

Padding to a power of two size can also mean a rather large padding. We might need to check the host data limits after doing the adjustment.

Quoted 5 lines
Quoted 2 lines

Yes,but it's the higher level driver's job.

mmc_align_data_size() shouldn't be adjusting the size to something it can easily determine that it's invalid.

Should I create a new function to do the power-of-2 adjustment? Or any ideal way to do it?
Otherwise,the SDH almost can't be used as sdio host.

Thanks

Cliff

#5 lore raw
Quoted 6 lines · Cliff
Quoted 3 lines

mmc_align_data_size() shouldn't be adjusting the size to something it can easily determine that it's invalid.

Should I create a new function to do the power-of-2 adjustment? Or any ideal way to do it?

That function is the correct place. My point was that we need to check after the adjustment that we didn't violate any of the other restrictions the driver has set. If we do, we should just return the original value.

Quoted 1 line

Otherwise,the SDH almost can't be used as sdio host.

Unfortunately I think this controller will be very unsuitable as an SDIO controller anyway with that restriction.

Rgds

Signature · 9 lines
     -- Pierre Ossman

  Linux kernel, MMC maintainer        http://www.kernel.org
  rdesktop, core developer          http://www.rdesktop.org

  WARNING: This correspondence is being monitored by the
  Swedish government. Make sure your server uses encryption
  for SMTP traffic and consider using PGP for end-to-end
  encryption.
#6 lore raw
Quoted 18 lines

-----Original Message-----
From: Pierre Ossman [mailto:[email protected]]
Sent: Monday, April 06, 2009 3:09 AM
To: Cai, Cliff
Cc: Bryan Wu; [email protected]
Subject: Re: [PATCH] mmc: align data size for host which only supports power-of-2 block

Quoted 2 lines · Cliff
Quoted 2 lines

mmc_align_data_size() shouldn't be adjusting the size to

something it

Quoted 2 lines
Quoted 1 line

can easily determine that it's invalid.

Quoted 7 lines
Quoted 2 lines

Should I create a new function to do the power-of-2 adjustment? Or any ideal way to do it?

That function is the correct place. My point was that we need to check after the adjustment that we didn't violate any of the other restrictions the driver has set. If we do, we should just return the original value.

What are other restrictions?I don't see any other restrictions on data size in current host driver
Except 4 byte alignment.

Quoted 4 lines
Quoted 1 line

Otherwise,the SDH almost can't be used as sdio host.

Unfortunately I think this controller will be very unsuitable as an SDIO controller anyway with that restriction.

Yes,But since we need to support some SDIO devices according to our customers'
Requests ,I think we have no other choice of this.

Thanks

Cliff

#7 lore raw
Quoted 21 lines · Cliff
Quoted 21 lines

-----Original Message-----
From: Pierre Ossman [mailto:[email protected]]
Sent: Monday, April 06, 2009 3:09 AM
To: Cai, Cliff
Cc: Bryan Wu; [email protected]
Subject: Re: [PATCH] mmc: align data size for host which only supports power-of-2 block

Quoted 2 lines · Cliff
Quoted 2 lines

mmc_align_data_size() shouldn't be adjusting the size to

something it

Quoted 4 lines
Quoted 1 line

can easily determine that it's invalid.

Should I create a new function to do the power-of-2 adjustment? Or any ideal way to do it?

Quoted 9 lines
Quoted 4 lines

That function is the correct place. My point was that we need to check after the adjustment that we didn't violate any of the other restrictions the driver has set. If we do, we should just return the original value.

What are other restrictions?I don't see any other restrictions on data size in current host driver
Except 4 byte alignment.

The ones that are in the mmc_host structure:

/* host specific block data */
unsigned int max_seg_size; /* see blk_queue_max_segment_size */
unsigned short max_hw_segs; /* see blk_queue_max_hw_segments */
unsigned short max_phys_segs; /* see blk_queue_max_phys_segments */
unsigned short unused;
unsigned int max_req_size; /* maximum number of bytes in one req */
unsigned int max_blk_size; /* maximum size of one mmc block */
unsigned int max_blk_count; /* maximum number of blocks in one req */

If you adjust a transfer from 1100 bytes to 2048, that could exceed the limit of 2047 bytes some controllers have.

Quoted 10 lines
Quoted 5 lines
Quoted 2 lines

Otherwise,the SDH almost can't be used as sdio host.

Unfortunately I think this controller will be very unsuitable as an SDIO controller anyway with that restriction.

Yes,But since we need to support some SDIO devices according to our customers'
Requests ,I think we have no other choice of this.

They should be prepared for the fact that many SDIO devices cannot be made to work with a controller with these bugs though. The function you're modifying is only to optimise the transfer and is not mandatory for SDIO drivers to call in any way.

Rgds

Signature · 10 lines
     -- Pierre Ossman

  Linux kernel, MMC maintainer        http://www.kernel.org
  rdesktop, core developer          http://www.rdesktop.org
  TigerVNC, core developer          http://www.tigervnc.org

  WARNING: This correspondence is being monitored by the
  Swedish government. Make sure your server uses encryption
  for SMTP traffic and consider using PGP for end-to-end
  encryption.