Patchwork [U-Boot,nand] Implement nand_extent_skip_bad

login
register
mail settings
Submitter Pantelis Antoniou
Date Dec. 10, 2012, 3:24 p.m.
Message ID <1355153064-6008-1-git-send-email-panto@antoniou-consulting.com>
Download mbox | patch
Permalink /patch/204939/
State Changes Requested
Delegated to: Scott Wood
Headers show

Comments

Pantelis Antoniou - Dec. 10, 2012, 3:24 p.m.
When accessing nand any bad blocks encountered are skipped, with no
indication about the amount of bad blocks encountered.
While this is normally fine, when you have to write a large amount
of data in chunks, you need to account for the skipped amount due
to the presence of the bad blocks.

nand_extend_skip_bad() returns the offset where the next access
should occur.

Signed-off-by: Pantelis Antoniou <panto@antoniou-consulting.com>
---
 drivers/mtd/nand/nand_util.c | 50 ++++++++++++++++++++++++++++++++++++++++++++
 include/nand.h               |  2 ++
 2 files changed, 52 insertions(+)
Scott Wood - Dec. 10, 2012, 10:53 p.m.
On 12/10/2012 09:24:24 AM, Pantelis Antoniou wrote:
> When accessing nand any bad blocks encountered are skipped, with no
> indication about the amount of bad blocks encountered.
> While this is normally fine, when you have to write a large amount
> of data in chunks, you need to account for the skipped amount due
> to the presence of the bad blocks.
> 
> nand_extend_skip_bad() returns the offset where the next access
> should occur.

s/extend/extent/

> 
> Signed-off-by: Pantelis Antoniou <panto@antoniou-consulting.com>
> ---
>  drivers/mtd/nand/nand_util.c | 50  
> ++++++++++++++++++++++++++++++++++++++++++++
>  include/nand.h               |  2 ++
>  2 files changed, 52 insertions(+)
> 
> diff --git a/drivers/mtd/nand/nand_util.c  
> b/drivers/mtd/nand/nand_util.c
> index 2ba0c5e..a25a4cb 100644
> --- a/drivers/mtd/nand/nand_util.c
> +++ b/drivers/mtd/nand/nand_util.c
> @@ -684,6 +684,56 @@ int nand_read_skip_bad(nand_info_t *nand, loff_t  
> offset, size_t *length,
>  	return 0;
>  }
> 
> +/**
> + * nand_extent_skip_bad:
> + *
> + * Find the extent of a chunk, return the offset where it ends
> + * Blocks that are marked bad are skipped and the next block is  
> examined
> + * instead as long as the extend is short enough to fit even after  
> skipping the
> + * bad blocks.
> + *
> + * @param nand NAND device
> + * @param offset offset in flash
> + * @param length extend length
> + * @return next offset in case of success (loff_t)-1 on error
> + */

Would it be better to return this information from existing read/write  
functions -- either instead of or in addition to exporting this  
functionality?

> +loff_t nand_extent_skip_bad(nand_info_t *nand, loff_t offset, size_t  
> length)
> +{
> +	size_t block_len, block_off;
> +	loff_t block_start;
> +
> +	if ((offset & (nand->writesize - 1)) != 0) {
> +		printf ("%s: Attempt to check extend non page aligned  
> data\n",
> +				__func__);
> +		return (loff_t)-1;
> +	}
> +
> +	while (length > 0) {
> +
> +		if (offset >= nand->size) {
> +			printf("%s: offset >= nand->size (%llx >=  
> %llx)\n",
> +					__func__, offset, nand->size);
> +			return (loff_t)-1;
> +		}
> +
> +		block_start = offset & ~(loff_t)(nand->erasesize - 1);
> +		block_off = offset & (nand->erasesize - 1);
> +		block_len = nand->erasesize - block_off;
> +		if (block_len > length)		/* left over */
> +			block_len = length;
> +
> +		if (!nand_block_isbad(nand, block_start))
> +			length -= block_len;
> +		else
> +			debug("%s: bad block at %llx (left %x)\n",
> +					__func__, block_start, length);
> +
> +		offset += block_len;
> +	}
> +
> +	return offset;
> +}

This seems duplicative of check_skip_len().

-Scott
Pantelis Antoniou - Dec. 11, 2012, 9:40 a.m.
Hi Scott,

On Dec 11, 2012, at 12:53 AM, Scott Wood wrote:

> On 12/10/2012 09:24:24 AM, Pantelis Antoniou wrote:
>> When accessing nand any bad blocks encountered are skipped, with no
>> indication about the amount of bad blocks encountered.
>> While this is normally fine, when you have to write a large amount
>> of data in chunks, you need to account for the skipped amount due
>> to the presence of the bad blocks.
>> nand_extend_skip_bad() returns the offset where the next access
>> should occur.
> 
> s/extend/extent/
> 

Yeah.

>> Signed-off-by: Pantelis Antoniou <panto@antoniou-consulting.com>
>> ---
>> drivers/mtd/nand/nand_util.c | 50 ++++++++++++++++++++++++++++++++++++++++++++
>> include/nand.h               |  2 ++
>> 2 files changed, 52 insertions(+)
>> diff --git a/drivers/mtd/nand/nand_util.c b/drivers/mtd/nand/nand_util.c
>> index 2ba0c5e..a25a4cb 100644
>> --- a/drivers/mtd/nand/nand_util.c
>> +++ b/drivers/mtd/nand/nand_util.c
>> @@ -684,6 +684,56 @@ int nand_read_skip_bad(nand_info_t *nand, loff_t offset, size_t *length,
>> 	return 0;
>> }
>> +/**
>> + * nand_extent_skip_bad:
>> + *
>> + * Find the extent of a chunk, return the offset where it ends
>> + * Blocks that are marked bad are skipped and the next block is examined
>> + * instead as long as the extend is short enough to fit even after skipping the
>> + * bad blocks.
>> + *
>> + * @param nand NAND device
>> + * @param offset offset in flash
>> + * @param length extend length
>> + * @return next offset in case of success (loff_t)-1 on error
>> + */
> 
> Would it be better to return this information from existing read/write functions -- either instead of or in addition to exporting this functionality?
> 

Yes it would. However that would require modifying all callers, which would be a hard sell when there's only one user of it.

>> +loff_t nand_extent_skip_bad(nand_info_t *nand, loff_t offset, size_t length)
>> +{
>> +	size_t block_len, block_off;
>> +	loff_t block_start;
>> +
>> +	if ((offset & (nand->writesize - 1)) != 0) {
>> +		printf ("%s: Attempt to check extend non page aligned data\n",
>> +				__func__);
>> +		return (loff_t)-1;
>> +	}
>> +
>> +	while (length > 0) {
>> +
>> +		if (offset >= nand->size) {
>> +			printf("%s: offset >= nand->size (%llx >= %llx)\n",
>> +					__func__, offset, nand->size);
>> +			return (loff_t)-1;
>> +		}
>> +
>> +		block_start = offset & ~(loff_t)(nand->erasesize - 1);
>> +		block_off = offset & (nand->erasesize - 1);
>> +		block_len = nand->erasesize - block_off;
>> +		if (block_len > length)		/* left over */
>> +			block_len = length;
>> +
>> +		if (!nand_block_isbad(nand, block_start))
>> +			length -= block_len;
>> +		else
>> +			debug("%s: bad block at %llx (left %x)\n",
>> +					__func__, block_start, length);
>> +
>> +		offset += block_len;
>> +	}
>> +
>> +	return offset;
>> +}
> 
> This seems duplicative of check_skip_len().
> 

It is. check_skip_len doesn't return the information I need. I could modify check_skip_len with
an extra parameter if that's going to be OK with you.

> -Scott

Regards

-- Pantelis
Scott Wood - Dec. 11, 2012, 5:13 p.m.
On 12/11/2012 03:40:53 AM, Pantelis Antoniou wrote:
> Hi Scott,
> 
> On Dec 11, 2012, at 12:53 AM, Scott Wood wrote:
> 
> >> +/**
> >> + * nand_extent_skip_bad:
> >> + *
> >> + * Find the extent of a chunk, return the offset where it ends
> >> + * Blocks that are marked bad are skipped and the next block is  
> examined
> >> + * instead as long as the extend is short enough to fit even  
> after skipping the
> >> + * bad blocks.
> >> + *
> >> + * @param nand NAND device
> >> + * @param offset offset in flash
> >> + * @param length extend length
> >> + * @return next offset in case of success (loff_t)-1 on error
> >> + */
> >
> > Would it be better to return this information from existing  
> read/write functions -- either instead of or in addition to exporting  
> this functionality?
> >
> 
> Yes it would. However that would require modifying all callers, which  
> would be a hard sell when there's only one user of it.

There aren't that many callers, and it's all common code (so no issue  
with testing on obscure hardware).

> > This seems duplicative of check_skip_len().
> >
> 
> It is. check_skip_len doesn't return the information I need. I could  
> modify check_skip_len with
> an extra parameter if that's going to be OK with you.

Yes, please modify check_skip_len() instead.

-Scott
Pantelis Antoniou - Dec. 11, 2012, 5:16 p.m.
Hi Scott,

On Dec 11, 2012, at 7:13 PM, Scott Wood wrote:

> On 12/11/2012 03:40:53 AM, Pantelis Antoniou wrote:
>> Hi Scott,
>> On Dec 11, 2012, at 12:53 AM, Scott Wood wrote:
>> >> +/**
>> >> + * nand_extent_skip_bad:
>> >> + *
>> >> + * Find the extent of a chunk, return the offset where it ends
>> >> + * Blocks that are marked bad are skipped and the next block is examined
>> >> + * instead as long as the extend is short enough to fit even after skipping the
>> >> + * bad blocks.
>> >> + *
>> >> + * @param nand NAND device
>> >> + * @param offset offset in flash
>> >> + * @param length extend length
>> >> + * @return next offset in case of success (loff_t)-1 on error
>> >> + */
>> >
>> > Would it be better to return this information from existing read/write functions -- either instead of or in addition to exporting this functionality?
>> >
>> Yes it would. However that would require modifying all callers, which would be a hard sell when there's only one user of it.
> 
> There aren't that many callers, and it's all common code (so no issue with testing on obscure hardware).
> 
>> > This seems duplicative of check_skip_len().
>> >
>> It is. check_skip_len doesn't return the information I need. I could modify check_skip_len with
>> an extra parameter if that's going to be OK with you.
> 
> Yes, please modify check_skip_len() instead.
> 
> -Scott

Nice, hope I'll get around doing it today or tomorrow.

Regards

-- Pantelis

Patch

diff --git a/drivers/mtd/nand/nand_util.c b/drivers/mtd/nand/nand_util.c
index 2ba0c5e..a25a4cb 100644
--- a/drivers/mtd/nand/nand_util.c
+++ b/drivers/mtd/nand/nand_util.c
@@ -684,6 +684,56 @@  int nand_read_skip_bad(nand_info_t *nand, loff_t offset, size_t *length,
 	return 0;
 }
 
+/**
+ * nand_extent_skip_bad:
+ *
+ * Find the extent of a chunk, return the offset where it ends
+ * Blocks that are marked bad are skipped and the next block is examined
+ * instead as long as the extend is short enough to fit even after skipping the
+ * bad blocks.
+ *
+ * @param nand NAND device
+ * @param offset offset in flash
+ * @param length extend length
+ * @return next offset in case of success (loff_t)-1 on error
+ */
+loff_t nand_extent_skip_bad(nand_info_t *nand, loff_t offset, size_t length)
+{
+	size_t block_len, block_off;
+	loff_t block_start;
+
+	if ((offset & (nand->writesize - 1)) != 0) {
+		printf ("%s: Attempt to check extend non page aligned data\n",
+				__func__);
+		return (loff_t)-1;
+	}
+
+	while (length > 0) {
+
+		if (offset >= nand->size) {
+			printf("%s: offset >= nand->size (%llx >= %llx)\n",
+					__func__, offset, nand->size);
+			return (loff_t)-1;
+		}
+
+		block_start = offset & ~(loff_t)(nand->erasesize - 1);
+		block_off = offset & (nand->erasesize - 1);
+		block_len = nand->erasesize - block_off;
+		if (block_len > length)		/* left over */
+			block_len = length;
+
+		if (!nand_block_isbad(nand, block_start))
+			length -= block_len;
+		else
+			debug("%s: bad block at %llx (left %x)\n",
+					__func__, block_start, length);
+
+		offset += block_len;
+	}
+
+	return offset;
+}
+
 #ifdef CONFIG_CMD_NAND_TORTURE
 
 /**
diff --git a/include/nand.h b/include/nand.h
index dded4e2..710c11a 100644
--- a/include/nand.h
+++ b/include/nand.h
@@ -168,3 +168,5 @@  __attribute__((noreturn)) void nand_boot(void);
 #define ENV_OFFSET_SIZE 8
 int get_nand_env_oob(nand_info_t *nand, unsigned long *result);
 #endif
+
+loff_t nand_extent_skip_bad(nand_info_t *nand, loff_t offset, size_t length);