Patchwork mtd: use ecc_strength to determine whether to perform bitflip_threshold test

login
register
mail settings
Submitter Mike Dunn
Date May 1, 2012, 7:56 p.m.
Message ID <1335902181-18110-1-git-send-email-mikedunn@newsguy.com>
Download mbox | patch
Permalink /patch/156202/
State New
Headers show

Comments

Mike Dunn - May 1, 2012, 7:56 p.m.
This is a minor change to the recent EUCLEAN patch set that I think makes it
more correct.  Currently, the test in mtd_read() that determines whether or not
to return -EUCLEAN is skipped if bitflip_threshold == 0.  This logic was added
(without much thought) to fix a bug in an earlier version of the patch set that
would have caused devices lacking ecc to always return -EUCLEAN [1].

This patch replaces bitflip_threshold with ecc_strength as the value used to
determine whether the device has ecc capability.  I think this is more correct
because bitflip_threshold can be changed via sysfs.  Currently, if the user sets
it to zero, -EUCLEAN is never returned.  This is also what happens if the user
sets it to a value exceeding ecc_strength.  With this patch, if the user sets it
to zero, -EUCLEAN will always be returned (provided ecc_strength > 0), which
makes more sense, and may even be useful occasionally.

[1] http://lists.infradead.org/pipermail/linux-mtd/2012-March/040468.html

Signed-off-by: Mike Dunn <mikedunn@newsguy.com>
---

Sorry for wanting to make tweaks after reviews and Ivan's testing.  It's
optional, low risk, and I did test it.  Thanks!

 Documentation/ABI/testing/sysfs-class-mtd |    7 +++++--
 drivers/mtd/mtdcore.c                     |    2 +-
 2 files changed, 6 insertions(+), 3 deletions(-)
Brian Norris - May 2, 2012, 1:33 a.m.
On Tue, May 1, 2012 at 12:56 PM, Mike Dunn <mikedunn@newsguy.com> wrote:
> This patch replaces bitflip_threshold with ecc_strength as the value used to
> determine whether the device has ecc capability.  I think this is more correct
> because bitflip_threshold can be changed via sysfs.

This isn't the only reason. More simply, there's the reason that
ecc_strength is the natural measure of "do I have ECC", not
bitflip_threshold, even if we can often say that ecc_strength != 0 if
and only if bitflip_threshold != 0.

> Sorry for wanting to make tweaks after reviews and Ivan's testing.  It's
> optional, low risk, and I did test it.  Thanks!

Aside from the review and testing, this kind of "fixup" usually makes
sense to just squash into the original patch ("mtd: driver _read()
returns max_bitflips; mtd_read() returns -EUCLEAN"), right?

> diff --git a/drivers/mtd/mtdcore.c b/drivers/mtd/mtdcore.c
> index b5bb628..5757307 100644
> --- a/drivers/mtd/mtdcore.c
> +++ b/drivers/mtd/mtdcore.c
> @@ -815,7 +815,7 @@ int mtd_read(struct mtd_info *mtd, loff_t from, size_t len, size_t *retlen,
>        ret_code = mtd->_read(mtd, from, len, retlen, buf);
>        if (unlikely(ret_code < 0))
>                return ret_code;
> -       if (mtd->bitflip_threshold == 0)
> +       if (mtd->ecc_strength == 0)
>                return 0;       /* device lacks ecc */
>        return ret_code >= mtd->bitflip_threshold ? -EUCLEAN : 0;
>  }

I had meant to suggest this on the original patch but never got around to it.

For this patch (on its own, or in combo with the patch it fixes up):

Reviewed-by: Brian Norris <computersforpeace@gmail.com>
Artem Bityutskiy - May 2, 2012, 12:13 p.m.
On Tue, 2012-05-01 at 12:56 -0700, Mike Dunn wrote:
> This is a minor change to the recent EUCLEAN patch set that I think makes it
> more correct.  Currently, the test in mtd_read() that determines whether or not
> to return -EUCLEAN is skipped if bitflip_threshold == 0.  This logic was added
> (without much thought) to fix a bug in an earlier version of the patch set that
> would have caused devices lacking ecc to always return -EUCLEAN [1].
> 
> This patch replaces bitflip_threshold with ecc_strength as the value used to
> determine whether the device has ecc capability.  I think this is more correct
> because bitflip_threshold can be changed via sysfs.  Currently, if the user sets
> it to zero, -EUCLEAN is never returned.  This is also what happens if the user
> sets it to a value exceeding ecc_strength.  With this patch, if the user sets it
> to zero, -EUCLEAN will always be returned (provided ecc_strength > 0), which
> makes more sense, and may even be useful occasionally.
> 
> [1] http://lists.infradead.org/pipermail/linux-mtd/2012-March/040468.html
> 
> Signed-off-by: Mike Dunn <mikedunn@newsguy.com>


Folded this patch into:

mtd: driver _read() returns max_bitflips; mtd_read() returns -EUCLEAN

and pushed to l2-mtd.git, thanks!

Patch

diff --git a/Documentation/ABI/testing/sysfs-class-mtd b/Documentation/ABI/testing/sysfs-class-mtd
index 7883508..db1ad7e 100644
--- a/Documentation/ABI/testing/sysfs-class-mtd
+++ b/Documentation/ABI/testing/sysfs-class-mtd
@@ -167,7 +167,10 @@  Description:
 		block degradation, but high enough to avoid the consequences of
 		a persistent return value of -EUCLEAN on devices where sticky
 		bitflips occur.  Note that if bitflip_threshold exceeds
-		ecc_strength, -EUCLEAN is never returned by the read functions.
+		ecc_strength, -EUCLEAN is never returned by mtd_read().
+		Conversely, if bitflip_threshold is zero, -EUCLEAN is always
+		returned, absent a hard error.
 
 		This is generally applicable only to NAND flash devices with ECC
-		capability.  It is ignored on devices lacking ECC capability.
+		capability.  It is ignored on devices lacking ECC capability;
+		i.e., devices for which ecc_strength is zero.
diff --git a/drivers/mtd/mtdcore.c b/drivers/mtd/mtdcore.c
index b5bb628..5757307 100644
--- a/drivers/mtd/mtdcore.c
+++ b/drivers/mtd/mtdcore.c
@@ -815,7 +815,7 @@  int mtd_read(struct mtd_info *mtd, loff_t from, size_t len, size_t *retlen,
 	ret_code = mtd->_read(mtd, from, len, retlen, buf);
 	if (unlikely(ret_code < 0))
 		return ret_code;
-	if (mtd->bitflip_threshold == 0)
+	if (mtd->ecc_strength == 0)
 		return 0;	/* device lacks ecc */
 	return ret_code >= mtd->bitflip_threshold ? -EUCLEAN : 0;
 }