From: Frieder Schrempf <[email protected]>
We were pointed to the issue of bad block markers not being saved to flash on
one of our boards with SPI NAND flash. After a bit of investigation it seems
like there are two overlapping bugs in the original framework that cause silent
failure when writing a bad block marker.
This set contains fixes for both of these issues and one more fix (patch 2) that
should not affect the actual behavior of the driver.
Changes in v3:
* Patch 1: Correct "SPI MEM" to "SPI NAND" in commit message
* Patch 3: Fix the subject line and the commit message
* Patch 3: Add Boris' R-b tag
Changes in v2:
* Patch 1: Incorporate small improvements proposed by Boris
* Patch 1: Add Boris' R-b tag
* Patch 2: Add Boris' R-b tag
* Patch 3: Instead of waiting for the erase operation to finish,
just don't do an erase at all, as it is not needed.
Frieder Schrempf (3):
mtd: spinand: Stop using spinand->oobbuf for buffering bad block
markers
mtd: spinand: Explicitly use MTD_OPS_RAW to write the bad block marker
to OOB
mtd: spinand: Do not erase the block before writing a bad block marker
drivers/mtd/nand/spi/core.c | 18 ++++++++----------
1 file changed, 8 insertions(+), 10 deletions(-)
--
2.17.1
From: Frieder Schrempf <[email protected]>
For reading and writing the bad block markers, spinand->oobbuf is
currently used as a buffer for the marker bytes. During the
underlying read and write operations to actually get/set the content
of the OOB area, the content of spinand->oobbuf is reused and changed
by accessing it through spinand->oobbuf and/or spinand->databuf.
This is a flaw in the original design of the SPI NAND core and at the
latest from 13c15e07eedf ("mtd: spinand: Handle the case where
PROGRAM LOAD does not reset the cache") on, it results in not having
the bad block marker written at all, as the spinand->oobbuf is
cleared to 0xff after setting the marker bytes to zero.
To fix it, we now just store the two bytes for the marker on the
stack and let the read/write operations copy it from/to the page
buffer later.
Fixes: 7529df465248 ("mtd: nand: Add core infrastructure to support SPI NANDs")
Cc: [email protected]
Signed-off-by: Frieder Schrempf <[email protected]>
Reviewed-by: Boris Brezillon <[email protected]>
---
Changes in v3:
* Correct "SPI MEM" to "SPI NAND" in commit message
Changes in v2:
* Incorporate small improvements proposed by Boris
* Add Boris' R-b tag
---
drivers/mtd/nand/spi/core.c | 14 +++++++-------
1 file changed, 7 insertions(+), 7 deletions(-)
diff --git a/drivers/mtd/nand/spi/core.c b/drivers/mtd/nand/spi/core.c
index 89f6beefb01c..de36cd7a5d7e 100644
--- a/drivers/mtd/nand/spi/core.c
+++ b/drivers/mtd/nand/spi/core.c
@@ -568,18 +568,18 @@ static int spinand_mtd_write(struct mtd_info *mtd, loff_t to,
static bool spinand_isbad(struct nand_device *nand, const struct nand_pos *pos)
{
struct spinand_device *spinand = nand_to_spinand(nand);
+ u8 marker[2] = { };
struct nand_page_io_req req = {
.pos = *pos,
- .ooblen = 2,
+ .ooblen = sizeof(marker),
.ooboffs = 0,
- .oobbuf.in = spinand->oobbuf,
+ .oobbuf.in = marker,
.mode = MTD_OPS_RAW,
};
- memset(spinand->oobbuf, 0, 2);
spinand_select_target(spinand, pos->target);
spinand_read_page(spinand, &req, false);
- if (spinand->oobbuf[0] != 0xff || spinand->oobbuf[1] != 0xff)
+ if (marker[0] != 0xff || marker[1] != 0xff)
return true;
return false;
@@ -603,11 +603,12 @@ static int spinand_mtd_block_isbad(struct mtd_info *mtd, loff_t offs)
static int spinand_markbad(struct nand_device *nand, const struct nand_pos *pos)
{
struct spinand_device *spinand = nand_to_spinand(nand);
+ u8 marker[2] = { };
struct nand_page_io_req req = {
.pos = *pos,
.ooboffs = 0,
- .ooblen = 2,
- .oobbuf.out = spinand->oobbuf,
+ .ooblen = sizeof(marker),
+ .oobbuf.out = marker,
};
int ret;
@@ -622,7 +623,6 @@ static int spinand_markbad(struct nand_device *nand, const struct nand_pos *pos)
spinand_erase_op(spinand, pos);
- memset(spinand->oobbuf, 0, 2);
return spinand_write_page(spinand, &req);
}
--
2.17.1
From: Frieder Schrempf <[email protected]>
When writing the bad block marker to the OOB area the access mode
should be set to MTD_OPS_RAW as it is done for reading the marker.
Currently this only works because req.mode is initialized to
MTD_OPS_PLACE_OOB (0) and spinand_write_to_cache_op() checks for
req.mode != MTD_OPS_AUTO_OOB.
Fix this by explicitly setting req.mode to MTD_OPS_RAW.
Fixes: 7529df465248 ("mtd: nand: Add core infrastructure to support SPI NANDs")
Signed-off-by: Frieder Schrempf <[email protected]>
Reviewed-by: Boris Brezillon <[email protected]>
---
Changes in v3:
* none
Changes in v2:
* Add Boris' R-b tag
---
drivers/mtd/nand/spi/core.c | 1 +
1 file changed, 1 insertion(+)
diff --git a/drivers/mtd/nand/spi/core.c b/drivers/mtd/nand/spi/core.c
index de36cd7a5d7e..a94287884453 100644
--- a/drivers/mtd/nand/spi/core.c
+++ b/drivers/mtd/nand/spi/core.c
@@ -609,6 +609,7 @@ static int spinand_markbad(struct nand_device *nand, const struct nand_pos *pos)
.ooboffs = 0,
.ooblen = sizeof(marker),
.oobbuf.out = marker,
+ .mode = MTD_OPS_RAW,
};
int ret;
--
2.17.1
From: Frieder Schrempf <[email protected]>
Currently when marking a block, we use spinand_erase_op() to erase
the block before writing the marker to the OOB area. Doing so without
waiting for the operation to finish can lead to the marking failing
silently and no bad block marker being written to the flash.
In fact we don't need to do an erase at all before writing the BBM.
The ECC is disabled for raw accesses to the OOB data and we don't
need to work around any issues with chips reporting ECC errors as it
is known to be the case for raw NAND.
Fixes: 7529df465248 ("mtd: nand: Add core infrastructure to support SPI NANDs")
Cc: [email protected]
Signed-off-by: Frieder Schrempf <[email protected]>
Reviewed-by: Boris Brezillon <[email protected]>
---
Changes in v3:
* Fix the subject line and the commit message
* Add Boris' R-b tag
Changes in v2:
* Instead of waiting for the erase operation to finish, just don't
do an erase at all, as it is not needed.
* Update the commit message
---
drivers/mtd/nand/spi/core.c | 3 ---
1 file changed, 3 deletions(-)
diff --git a/drivers/mtd/nand/spi/core.c b/drivers/mtd/nand/spi/core.c
index a94287884453..8dda51bbdd11 100644
--- a/drivers/mtd/nand/spi/core.c
+++ b/drivers/mtd/nand/spi/core.c
@@ -613,7 +613,6 @@ static int spinand_markbad(struct nand_device *nand, const struct nand_pos *pos)
};
int ret;
- /* Erase block before marking it bad. */
ret = spinand_select_target(spinand, pos->target);
if (ret)
return ret;
@@ -622,8 +621,6 @@ static int spinand_markbad(struct nand_device *nand, const struct nand_pos *pos)
if (ret)
return ret;
- spinand_erase_op(spinand, pos);
-
return spinand_write_page(spinand, &req);
}
--
2.17.1
On Tue, 2020-02-18 at 10:05:35 UTC, Schrempf Frieder wrote:
> From: Frieder Schrempf <[email protected]>
>
> Currently when marking a block, we use spinand_erase_op() to erase
> the block before writing the marker to the OOB area. Doing so without
> waiting for the operation to finish can lead to the marking failing
> silently and no bad block marker being written to the flash.
>
> In fact we don't need to do an erase at all before writing the BBM.
> The ECC is disabled for raw accesses to the OOB data and we don't
> need to work around any issues with chips reporting ECC errors as it
> is known to be the case for raw NAND.
>
> Fixes: 7529df465248 ("mtd: nand: Add core infrastructure to support SPI NANDs")
> Cc: [email protected]
> Signed-off-by: Frieder Schrempf <[email protected]>
> Reviewed-by: Boris Brezillon <[email protected]>
Applied to https://git.kernel.org/pub/scm/linux/kernel/git/mtd/linux.git nand/next, thanks.
Miquel
On Tue, 2020-02-18 at 10:05:14 UTC, Schrempf Frieder wrote:
> From: Frieder Schrempf <[email protected]>
>
> For reading and writing the bad block markers, spinand->oobbuf is
> currently used as a buffer for the marker bytes. During the
> underlying read and write operations to actually get/set the content
> of the OOB area, the content of spinand->oobbuf is reused and changed
> by accessing it through spinand->oobbuf and/or spinand->databuf.
>
> This is a flaw in the original design of the SPI NAND core and at the
> latest from 13c15e07eedf ("mtd: spinand: Handle the case where
> PROGRAM LOAD does not reset the cache") on, it results in not having
> the bad block marker written at all, as the spinand->oobbuf is
> cleared to 0xff after setting the marker bytes to zero.
>
> To fix it, we now just store the two bytes for the marker on the
> stack and let the read/write operations copy it from/to the page
> buffer later.
>
> Fixes: 7529df465248 ("mtd: nand: Add core infrastructure to support SPI NANDs")
> Cc: [email protected]
> Signed-off-by: Frieder Schrempf <[email protected]>
> Reviewed-by: Boris Brezillon <[email protected]>
Applied to https://git.kernel.org/pub/scm/linux/kernel/git/mtd/linux.git nand/next, thanks.
Miquel
On Tue, 2020-02-18 at 10:05:25 UTC, Schrempf Frieder wrote:
> From: Frieder Schrempf <[email protected]>
>
> When writing the bad block marker to the OOB area the access mode
> should be set to MTD_OPS_RAW as it is done for reading the marker.
> Currently this only works because req.mode is initialized to
> MTD_OPS_PLACE_OOB (0) and spinand_write_to_cache_op() checks for
> req.mode != MTD_OPS_AUTO_OOB.
>
> Fix this by explicitly setting req.mode to MTD_OPS_RAW.
>
> Fixes: 7529df465248 ("mtd: nand: Add core infrastructure to support SPI NANDs")
> Signed-off-by: Frieder Schrempf <[email protected]>
> Reviewed-by: Boris Brezillon <[email protected]>
Applied to https://git.kernel.org/pub/scm/linux/kernel/git/mtd/linux.git nand/next, thanks.
Miquel