[U-Boot] [PATCH] mtd: nand: fix the written length when nand_write_skip_bad failed
When the data has been partially written into the NAND Flash, returning the written length instead of 0. The written length may be useful when the upper level decides to continue the writing after skipping the block causing the write failure.
Signed-off-by: Tao Hou hotforest@gmail.com Cc: Scott Wood scottwood@freescale.com Cc: Ben Gardiner bengardiner@nanometrics.ca Cc: Lei Wen leiwen@marvell.com --- drivers/mtd/nand/nand_util.c | 22 +++++++++++++++------- 1 file changed, 15 insertions(+), 7 deletions(-)
diff --git a/drivers/mtd/nand/nand_util.c b/drivers/mtd/nand/nand_util.c index de1d13e..f57d723 100644 --- a/drivers/mtd/nand/nand_util.c +++ b/drivers/mtd/nand/nand_util.c @@ -496,8 +496,10 @@ int nand_write_skip_bad(nand_info_t *nand, loff_t offset, size_t *length,
#ifdef CONFIG_CMD_NAND_YAFFS if (flags & WITH_YAFFS_OOB) { - if (flags & ~WITH_YAFFS_OOB) + if (flags & ~WITH_YAFFS_OOB) { + *length = 0; return -EINVAL; + }
int pages; pages = nand->erasesize / nand->writesize; @@ -505,6 +507,7 @@ int nand_write_skip_bad(nand_info_t *nand, loff_t offset, size_t *length, if (*length % (nand->writesize + nand->oobsize)) { printf("Attempt to write incomplete page" " in yaffs mode\n"); + *length = 0; return -EINVAL; } } else @@ -542,7 +545,6 @@ int nand_write_skip_bad(nand_info_t *nand, loff_t offset, size_t *length, if (rval == 0) return 0;
- *length = 0; printf("NAND write to offset %llx failed %d\n", offset, rval); return rval; @@ -550,7 +552,7 @@ int nand_write_skip_bad(nand_info_t *nand, loff_t offset, size_t *length,
while (left_to_write > 0) { size_t block_offset = offset & (nand->erasesize - 1); - size_t write_size, truncated_write_size; + size_t write_size, truncated_write_size, written_size;
WATCHDOG_RESET();
@@ -586,8 +588,10 @@ int nand_write_skip_bad(nand_info_t *nand, loff_t offset, size_t *length, ops.oobbuf = ops.datbuf + pagesize;
rval = nand->write_oob(nand, offset, &ops); - if (rval != 0) + if (rval != 0) { + written_size = pagesize_oob * page; break; + }
offset += pagesize; p_buffer += pagesize_oob; @@ -605,14 +609,18 @@ int nand_write_skip_bad(nand_info_t *nand, loff_t offset, size_t *length,
rval = nand_write(nand, offset, &truncated_write_size, p_buffer); - offset += write_size; - p_buffer += write_size; + if (rval == 0) { + offset += write_size; + p_buffer += write_size; + } else { + written_size = truncated_write_size; + } }
if (rval != 0) { printf("NAND write to offset %llx failed %d\n", offset, rval); - *length -= left_to_write; + *length -= left_to_write - written_size; return rval; }
On 03/02/2013 03:01:10 AM, Tao Hou wrote:
When the data has been partially written into the NAND Flash, returning the written length instead of 0. The written length may be useful when the upper level decides to continue the writing after skipping the block causing the write failure.
We already do that in some code paths.
Signed-off-by: Tao Hou hotforest@gmail.com Cc: Scott Wood scottwood@freescale.com Cc: Ben Gardiner bengardiner@nanometrics.ca Cc: Lei Wen leiwen@marvell.com
drivers/mtd/nand/nand_util.c | 22 +++++++++++++++------- 1 file changed, 15 insertions(+), 7 deletions(-)
Could you rebase this on top of this patch: http://patchwork.ozlabs.org/patch/224842/
BTW, are you actually using WITH_YAFFS_OOB? I think there are some other things wrong with it at the moment, as mentioned here: http://lists.denx.de/pipermail/u-boot/2013-March/148378.html
diff --git a/drivers/mtd/nand/nand_util.c b/drivers/mtd/nand/nand_util.c index de1d13e..f57d723 100644 --- a/drivers/mtd/nand/nand_util.c +++ b/drivers/mtd/nand/nand_util.c @@ -496,8 +496,10 @@ int nand_write_skip_bad(nand_info_t *nand, loff_t offset, size_t *length,
#ifdef CONFIG_CMD_NAND_YAFFS if (flags & WITH_YAFFS_OOB) {
if (flags & ~WITH_YAFFS_OOB)
if (flags & ~WITH_YAFFS_OOB) {*length = 0; return -EINVAL;}int pages; pages = nand->erasesize / nand->writesize;
@@ -505,6 +507,7 @@ int nand_write_skip_bad(nand_info_t *nand, loff_t offset, size_t *length, if (*length % (nand->writesize + nand->oobsize)) { printf("Attempt to write incomplete page" " in yaffs mode\n");
*length = 0; return -EINVAL;@@ -542,7 +545,6 @@ int nand_write_skip_bad(nand_info_t *nand, loff_t offset, size_t *length, if (rval == 0) return 0;
*length = 0;
OK so far...
@@ -550,7 +552,7 @@ int nand_write_skip_bad(nand_info_t *nand, loff_t offset, size_t *length,
while (left_to_write > 0) { size_t block_offset = offset & (nand->erasesize - 1);
size_t write_size, truncated_write_size;
size_t write_size, truncated_write_size, written_size;WATCHDOG_RESET();
@@ -586,8 +588,10 @@ int nand_write_skip_bad(nand_info_t *nand, loff_t offset, size_t *length, ops.oobbuf = ops.datbuf + pagesize;
rval = nand->write_oob(nand, offset,&ops);
if (rval != 0)
if (rval != 0) {written_size = pagesize_oob *page; break;
} offset += pagesize; p_buffer += pagesize_oob;@@ -605,14 +609,18 @@ int nand_write_skip_bad(nand_info_t *nand, loff_t offset, size_t *length,
rval = nand_write(nand, offset,&truncated_write_size, p_buffer);
offset += write_size;p_buffer += write_size;
if (rval == 0) {offset += write_size;p_buffer += write_size;} else {written_size = truncated_write_size;}}
if (rval != 0) { printf("NAND write to offset %llx failed %d\n", offset, rval);
*length -= left_to_write;
*length -= left_to_write - written_size; return rval;
...but I don't see why this part is needed (or correct). Why doesn't "*length -= left_to_write" already get you what you want?
-Scott
Hi, Scott
Thanks for your review.
On Tue, Mar 5, 2013 at 9:58 AM, Scott Wood scottwood@freescale.com wrote:
On 03/02/2013 03:01:10 AM, Tao Hou wrote:
When the data has been partially written into the NAND Flash, returning the written length instead of 0. The written length may be useful when the upper level decides to continue the writing after skipping the block causing the write failure.
We already do that in some code paths.
Signed-off-by: Tao Hou hotforest@gmail.com Cc: Scott Wood scottwood@freescale.com Cc: Ben Gardiner bengardiner@nanometrics.ca Cc: Lei Wen leiwen@marvell.com
drivers/mtd/nand/nand_util.c | 22 +++++++++++++++------- 1 file changed, 15 insertions(+), 7 deletions(-)
Could you rebase this on top of this patch: http://patchwork.ozlabs.org/patch/224842/
Do you mean a V2 patch ?
BTW, are you actually using WITH_YAFFS_OOB? I think there are some other things wrong with it at the moment, as mentioned here: http://lists.denx.de/pipermail/u-boot/2013-March/148378.html
No, I don't use it.
diff --git a/drivers/mtd/nand/nand_util.c b/drivers/mtd/nand/nand_util.c index de1d13e..f57d723 100644 --- a/drivers/mtd/nand/nand_util.c +++ b/drivers/mtd/nand/nand_util.c @@ -496,8 +496,10 @@ int nand_write_skip_bad(nand_info_t *nand, loff_t offset, size_t *length,
#ifdef CONFIG_CMD_NAND_YAFFS if (flags & WITH_YAFFS_OOB) {
if (flags & ~WITH_YAFFS_OOB)
if (flags & ~WITH_YAFFS_OOB) {*length = 0; return -EINVAL;} int pages; pages = nand->erasesize / nand->writesize;@@ -505,6 +507,7 @@ int nand_write_skip_bad(nand_info_t *nand, loff_t offset, size_t *length, if (*length % (nand->writesize + nand->oobsize)) { printf("Attempt to write incomplete page" " in yaffs mode\n");
*length = 0; return -EINVAL; } } else@@ -542,7 +545,6 @@ int nand_write_skip_bad(nand_info_t *nand, loff_t offset, size_t *length, if (rval == 0) return 0;
*length = 0; printf("NAND write to offset %llx failed %d\n", offset, rval); return rval;OK so far...
@@ -550,7 +552,7 @@ int nand_write_skip_bad(nand_info_t *nand, loff_t offset, size_t *length,
while (left_to_write > 0) { size_t block_offset = offset & (nand->erasesize - 1);
size_t write_size, truncated_write_size;
size_t write_size, truncated_write_size, written_size; WATCHDOG_RESET();@@ -586,8 +588,10 @@ int nand_write_skip_bad(nand_info_t *nand, loff_t offset, size_t *length, ops.oobbuf = ops.datbuf + pagesize;
rval = nand->write_oob(nand, offset,&ops);
if (rval != 0)
if (rval != 0) {written_size = pagesize_oob *page; break;
} offset += pagesize; p_buffer += pagesize_oob;@@ -605,14 +609,18 @@ int nand_write_skip_bad(nand_info_t *nand, loff_t offset, size_t *length,
rval = nand_write(nand, offset,&truncated_write_size, p_buffer);
offset += write_size;p_buffer += write_size;
if (rval == 0) {offset += write_size;p_buffer += write_size;} else {written_size = truncated_write_size;} } if (rval != 0) { printf("NAND write to offset %llx failed %d\n", offset, rval);
*length -= left_to_write;
*length -= left_to_write - written_size; return rval; }...but I don't see why this part is needed (or correct). Why doesn't "*length -= left_to_write" already get you what you want?
-Scott
I just use "*length -= left_to_write - written_size" to tell the upper level that what had been actually happened. For the current block, "written_size" has been written to the NAND Flash yet, so left_to_write should be subtracted by "written_size".
Cheers, Hou.
On 03/06/2013 08:56:56 AM, htbegin wrote:
Hi, Scott
Thanks for your review.
On Tue, Mar 5, 2013 at 9:58 AM, Scott Wood scottwood@freescale.com wrote:
On 03/02/2013 03:01:10 AM, Tao Hou wrote:
When the data has been partially written into the NAND Flash, returning the written length instead of 0. The written length may be useful when the upper level decides to continue the writing after skipping the block causing the write failure.
We already do that in some code paths.
Signed-off-by: Tao Hou hotforest@gmail.com Cc: Scott Wood scottwood@freescale.com Cc: Ben Gardiner bengardiner@nanometrics.ca Cc: Lei Wen leiwen@marvell.com
drivers/mtd/nand/nand_util.c | 22 +++++++++++++++------- 1 file changed, 15 insertions(+), 7 deletions(-)
Could you rebase this on top of this patch: http://patchwork.ozlabs.org/patch/224842/
Do you mean a V2 patch ?
Yes.
BTW, are you actually using WITH_YAFFS_OOB? I think there are some
other
things wrong with it at the moment, as mentioned here: http://lists.denx.de/pipermail/u-boot/2013-March/148378.html
No, I don't use it.
Changes to that code should be tested by someone...
if (rval != 0) { printf("NAND write to offset %llx failed%d\n",
offset, rval);
*length -= left_to_write;
*length -= left_to_write - written_size; return rval; }...but I don't see why this part is needed (or correct). Why
doesn't
"*length -= left_to_write" already get you what you want?
-Scott
I just use "*length -= left_to_write - written_size" to tell the upper level that what had been actually happened. For the current block, "written_size" has been written to the NAND Flash yet, so left_to_write should be subtracted by "written_size".
But left_to_write isn't decreased until after this error return, so that's already the case. Subtracting written_size from left_to_write has the effect of increasing length by written_size, so the return value will now look like the error page has been written.
-Scott
Hi, Scott
On Thu, Mar 7, 2013 at 2:22 AM, Scott Wood scottwood@freescale.com wrote:
On 03/06/2013 08:56:56 AM, htbegin wrote:
Hi, Scott
Thanks for your review.
On Tue, Mar 5, 2013 at 9:58 AM, Scott Wood scottwood@freescale.com wrote:
On 03/02/2013 03:01:10 AM, Tao Hou wrote:
When the data has been partially written into the NAND Flash, returning the written length instead of 0. The written length may be useful when the upper level decides to continue the writing after skipping the block causing the write failure.
We already do that in some code paths.
Signed-off-by: Tao Hou hotforest@gmail.com Cc: Scott Wood scottwood@freescale.com Cc: Ben Gardiner bengardiner@nanometrics.ca Cc: Lei Wen leiwen@marvell.com
drivers/mtd/nand/nand_util.c | 22 +++++++++++++++------- 1 file changed, 15 insertions(+), 7 deletions(-)
Could you rebase this on top of this patch: http://patchwork.ozlabs.org/patch/224842/
Do you mean a V2 patch ?
Yes.
I will send a V2 patch once we reach an agreement on the "written_length" problem.
BTW, are you actually using WITH_YAFFS_OOB? I think there are some other things wrong with it at the moment, as mentioned here: http://lists.denx.de/pipermail/u-boot/2013-March/148378.html
No, I don't use it.
Changes to that code should be tested by someone...
Sorry, I can't help.
if (rval != 0) { printf("NAND write to offset %llx failed %d\n", offset, rval);
*length -= left_to_write;
*length -= left_to_write - written_size; return rval; }...but I don't see why this part is needed (or correct). Why doesn't "*length -= left_to_write" already get you what you want?
-Scott
I just use "*length -= left_to_write - written_size" to tell the upper level that what had been actually happened. For the current block, "written_size" has been written to the NAND Flash yet, so left_to_write should be subtracted by "written_size".
But left_to_write isn't decreased until after this error return, so that's already the case. Subtracting written_size from left_to_write has the effect of increasing length by written_size, so the return value will now look like the error page has been written.
-Scott
No, the returned value doesn't include the length of the error page. In no-WITH_YAFFS_OOB case, when nand_write failed, truncated_write_size has been updated by nand_write to the length which has been successfully written , so it's OK to subtract written_size from left_to_write. In WITH_YAFFS_OOB case, when nand->write_oob failed, written_size is also the length which has been successfully written.
Cheers, Hou
On 03/07/2013 09:02:27 AM, htbegin wrote:
Hi, Scott
On Thu, Mar 7, 2013 at 2:22 AM, Scott Wood scottwood@freescale.com wrote:
On 03/06/2013 08:56:56 AM, htbegin wrote:
BTW, are you actually using WITH_YAFFS_OOB? I think there are
some
other things wrong with it at the moment, as mentioned here: http://lists.denx.de/pipermail/u-boot/2013-March/148378.html
No, I don't use it.
Changes to that code should be tested by someone...
Sorry, I can't help.
It's moot because I don't think this change should be made, but this is a case where you could enable it temporarily in your board config for some basic testing.
I just use "*length -= left_to_write - written_size" to tell the
upper
level that what had been actually happened. For the current block, "written_size"
has
been written to the NAND Flash yet, so left_to_write should be subtracted by "written_size".
But left_to_write isn't decreased until after this error return, so
that's
already the case. Subtracting written_size from left_to_write has
the
effect of increasing length by written_size, so the return value
will now
look like the error page has been written.
-Scott
No, the returned value doesn't include the length of the error page. In no-WITH_YAFFS_OOB case, when nand_write failed, truncated_write_size has been updated by nand_write to the length which has been successfully written , so it's OK to subtract written_size from left_to_write.
OK, but that doesn't explain why the change is needed. You said you wanted to find the block with the error. We only write one block at a time in the loop. Why do you need the specific page within the block that failed?
-Scott
Hi, Scott
On Fri, Mar 8, 2013 at 6:27 AM, Scott Wood scottwood@freescale.com wrote:
I just use "*length -= left_to_write - written_size" to tell the upper level that what had been actually happened. For the current block, "written_size" has been written to the NAND Flash yet, so left_to_write should be subtracted by "written_size".
But left_to_write isn't decreased until after this error return, so that's already the case. Subtracting written_size from left_to_write has the effect of increasing length by written_size, so the return value will now look like the error page has been written.
-Scott
No, the returned value doesn't include the length of the error page. In no-WITH_YAFFS_OOB case, when nand_write failed, truncated_write_size has been updated by nand_write to the length which has been successfully written , so it's OK to subtract written_size from left_to_write.
OK, but that doesn't explain why the change is needed. You said you wanted to find the block with the error. We only write one block at a time in the loop. Why do you need the specific page within the block that failed?
-Scott
Yes, you are right it's OK to ignore the written length of the write-failed block, but as I said before I just wanted to tell the upper level what had been actually written. So if you insist the subtraction of written_len is unnecessary, it's alright with me.
Thanks. Hou
On 03/09/2013 07:06:54 PM, htbegin wrote:
Hi, Scott
On Fri, Mar 8, 2013 at 6:27 AM, Scott Wood scottwood@freescale.com wrote:
I just use "*length -= left_to_write - written_size" to tell
the upper
level that what had been actually happened. For the current block,
"written_size" has
been written to the NAND Flash yet, so left_to_write should be subtracted by "written_size".
But left_to_write isn't decreased until after this error return,
so
that's already the case. Subtracting written_size from left_to_write
has the
effect of increasing length by written_size, so the return value
will
now look like the error page has been written.
-Scott
No, the returned value doesn't include the length of the error
page.
In no-WITH_YAFFS_OOB case, when nand_write failed, truncated_write_size has been updated by nand_write to the length which has been successfully written , so it's OK to subtract written_size from left_to_write.
OK, but that doesn't explain why the change is needed. You said
you wanted
to find the block with the error. We only write one block at a
time in the
loop. Why do you need the specific page within the block that
failed?
-Scott
Yes, you are right it's OK to ignore the written length of the write-failed block, but as I said before I just wanted to tell the upper level what had been actually written. So if you insist the subtraction of written_len is unnecessary, it's alright with me.
Thanks. I do insist -- it adds complexity.
-Scott
participants (3)
-
htbegin -
Scott Wood -
Tao Hou