MXS nand fixes in SPL
Subsystems (5)
v2 cover MXS nand fixes in SPL
Those patches come after some testing of failing in factory on some unit. We found out that the bootrom imx loader was not able to handling badblock. This can be a limit of the implementation right now in imx8mn. Anyway not all the imx platform has the support of this loader. I found some problems on the implementation so I have fixed it up according the experience of Sitara. I tested only using a Fit Image as a flash container. This version add in the series the fix of cmd_nandbcb and the fix of spl_nand load. I can imagine that a lot of boards and users are affected. I have started to backport this changes in some older uboot and adapt it
Michael Trimarchi (4):
nand: raw: mxs_nand: Fix specific hook registration
mtd: nand: mxs_nand_spl: Fix bad block skipping
arm: mach-imx: cmd_nandbcb fix bad block handling
spl: spl_nand: Fix bad block handling in fitImage
arch/arm/mach-imx/cmd_nandbcb.c | 21 +++---- common/spl/spl_nand.c | 5 +- drivers/mtd/nand/raw/mxs_nand.c | 32 +++++----- drivers/mtd/nand/raw/mxs_nand_spl.c | 90 ++++++++++++++++------------- 4 files changed, 73 insertions(+), 75 deletions(-)
Signature · 1 line
2.25.1
PATCH v2 1/4 nand: raw: mxs_nand: Fix specific hook registration
Move the hook after nand_scan_tail is called. The hook must be replaced to the mxs specific one but those must to be assignment later in the probe function.
With this fix markbad is working again. Before this change:
nand markbad 0xDEC00
NXS NAND: Writing OOB isn't supported
NXS NAND: Writing OOB isn't supported
block 0x000dec00 NOT marked as bad! ERROR 0
---V1->V2:
- Adjust the commit message
- Add Cc Han Xu and Fabio
--- drivers/mtd/nand/raw/mxs_nand.c | 32 ++++++++++++++++---------------- 1 file changed, 16 insertions(+), 16 deletions(-)
drivers/mtd/nand/raw/mxs_nand.c +16 -16
@@ -1246,22 +1246,6 @@ int mxs_nand_setup_ecc(struct mtd_info *mtd)/* Enable BCH complete interrupt */ writel (BCH_CTRL_COMPLETE_IRQ_EN , &bch_regs->hw_bch_ctrl_set); -/* Hook some operations at the MTD level. */ -if (mtd->_read_oob != mxs_nand_hook_read_oob) { - nand_info->hooked_read_oob = mtd->_read_oob; - mtd->_read_oob = mxs_nand_hook_read_oob; - } - -if (mtd->_write_oob != mxs_nand_hook_write_oob) { - nand_info->hooked_write_oob = mtd->_write_oob; - mtd->_write_oob = mxs_nand_hook_write_oob; - } - -if (mtd->_block_markbad != mxs_nand_hook_block_markbad) { - nand_info->hooked_block_markbad = mtd->_block_markbad; - mtd->_block_markbad = mxs_nand_hook_block_markbad; - } -return 0 ; }
@@ -1467,6 +1451,22 @@ int mxs_nand_init_ctrl(struct mxs_nand_info *nand_info)if (err) goto err_free_buffers; +/* Hook some operations at the MTD level. */ +if (mtd->_read_oob != mxs_nand_hook_read_oob) { + nand_info->hooked_read_oob = mtd->_read_oob; + mtd->_read_oob = mxs_nand_hook_read_oob; + } + +if (mtd->_write_oob != mxs_nand_hook_write_oob) { + nand_info->hooked_write_oob = mtd->_write_oob; + mtd->_write_oob = mxs_nand_hook_write_oob; + } + +if (mtd->_block_markbad != mxs_nand_hook_block_markbad) { + nand_info->hooked_block_markbad = mtd->_block_markbad; + mtd->_block_markbad = mxs_nand_hook_block_markbad; + } + err =nand_register (0 , mtd);if (err) goto err_free_buffers;
Signature · 1 line
2.25.1
Re: [PATCH V2 1/4] nand: raw: mxs_nand: Fix specific hook registration
Quoted 6 lines · Michael Trimarchi
Move the hook after
nand_scan_tailis called. The hook must be replaced to the mxs specific one but those must to be assignment later in the probe function.With this fix markbad is working again. Before this change:
Quoted 8 lines
nand markbad
0xDEC00
NXS NAND: Writing OOB isn't supported
NXS NAND: Writing OOB isn't supported
block0x000dec00NOT marked as bad! ERROR 0Cc Han Xu <[email protected]>Cc Fabio Estevam <[email protected]>Signed-off-by Michael Trimarchi <[email protected]>
Quoted 61 lines
---
V1->V2:
- Adjust the commit message
- Add Cc Han Xu and Fabio--- drivers/mtd/nand/raw/mxs_nand.c | 32 ++++++++++++++++---------------- 1 file changed, 16 insertions(+), 16 deletions(-)drivers/mtd/nand/raw/mxs_nand.c +16
-16@@ -1246,22 +1246,6 @@ int mxs_nand_setup_ecc(struct mtd_info *mtd)/* Enable BCH complete interrupt */ writel (BCH_CTRL_COMPLETE_IRQ_EN , &bch_regs->hw_bch_ctrl_set); -/* Hook some operations at the MTD level. */ -if (mtd->_read_oob != mxs_nand_hook_read_oob) { - nand_info->hooked_read_oob = mtd->_read_oob; - mtd->_read_oob = mxs_nand_hook_read_oob; - } - -if (mtd->_write_oob != mxs_nand_hook_write_oob) { - nand_info->hooked_write_oob = mtd->_write_oob; - mtd->_write_oob = mxs_nand_hook_write_oob; - } - -if (mtd->_block_markbad != mxs_nand_hook_block_markbad) { - nand_info->hooked_block_markbad = mtd->_block_markbad; - mtd->_block_markbad = mxs_nand_hook_block_markbad; - } -return 0 ; }@@ -1467,6 +1451,22 @@ int mxs_nand_init_ctrl(struct mxs_nand_info *nand_info)if (err) goto err_free_buffers; +/* Hook some operations at the MTD level. */ +if (mtd->_read_oob != mxs_nand_hook_read_oob) { + nand_info->hooked_read_oob = mtd->_read_oob; + mtd->_read_oob = mxs_nand_hook_read_oob; + } + +if (mtd->_write_oob != mxs_nand_hook_write_oob) { + nand_info->hooked_write_oob = mtd->_write_oob; + mtd->_write_oob = mxs_nand_hook_write_oob; + } + +if (mtd->_block_markbad != mxs_nand_hook_block_markbad) { + nand_info->hooked_block_markbad = mtd->_block_markbad; + mtd->_block_markbad = mxs_nand_hook_block_markbad; + } + err =nand_register (0 , mtd);if (err) goto err_free_buffers;Signature · 1 line
2.25.1
PATCH v2 2/4 mtd: nand: mxs_nand_spl: Fix bad block skipping
In reply to Michael TrimarchiThe specific implementation was having bug. Those bugs are since the beginning of the implementation. Some manufactures can receive this bug in their SPL code. This bug start to be more visible on architecture that has complicated boot process like imx8mn. Older version of uboot has the same problem only if the bad block appear in correspoding of befine of u-boot image. In order to adjust the function we scan from the first block. The logic is not changed to have a simple way to fix without get regression.
The problematic part of old code was in this part:
while (is_badblock(mtd, offs, 1)) {
page = page + nand_page_per_block;
/* Check i we've reached the end of flash. */
if (page >= mtd->size >> chip->page_shift) {
free(page_buf);
return -ENOMEM;
}
}Even we fix it adding increment of the offset of one erase block size we don't fix the problem, because the first erase block where the image start is not checked. The code was tested on an imx8mn where the boot rom api was not able to skip it. Apart of that other architecure are using this code and all boards that has nand as boot device can be affected
---V1->V2:
- Adjust the commit message
- Add Cc Han Xu and Fabio
- fix size >= 0 to > 0
--- drivers/mtd/nand/raw/mxs_nand_spl.c | 90 ++++++++++++++++------------- 1 file changed, 49 insertions(+), 41 deletions(-)
drivers/mtd/nand/raw/mxs_nand_spl.c +49 -41
@@ -218,14 +218,14 @@ void nand_init(void)mxs_nand_setup_ecc (mtd); } -int nand_spl_load_image (uint32_t offs,unsigned int size,void *buf) +int nand_spl_load_image (uint32_t offs,unsigned int size,void *dst) { -struct nand_chip *chip; -unsigned int page; +unsigned int sz; +unsigned int block, lastblock; +unsigned int page, page_offset;unsigned int nand_page_per_block; -unsigned int sz =0 ; +struct nand_chip *chip;u8 *page_buf =NULL ; -u32 page_off; chip =mtd_to_nand (mtd);if (!chip->numchips)
@@ -235,47 +235,42 @@ int nand_spl_load_image(uint32_t offs, unsigned int size, void *buf)if (!page_buf)return -ENOMEM ; - page = offs >> chip->page_shift; - page_off = offs & (mtd->writesize -1 ); +/* offs has to be aligned to a page address! */ + block = offs / mtd->erasesize; + lastblock = (offs + size -1 ) / mtd->erasesize; + page = (offs % mtd->erasesize) / mtd->writesize; + page_offset = offs % mtd->writesize; nand_page_per_block = mtd->erasesize / mtd->writesize; -debug ("%s offset:0x%08x len:%d page:%x\n" , __func__, offs, size, page); - -while (size) { -if (mxs_read_page_ecc (mtd, page_buf, page) <0 ) -return -1 ; - -if (size > (mtd->writesize - page_off)) - sz = (mtd->writesize - page_off); -else - sz = size; - -memcpy (buf, page_buf + page_off, sz); - - offs += mtd->writesize; - page++; - buf += (mtd->writesize - page_off); - page_off =0 ; - size -= sz; - -/* -* Check if we have crossed a block boundary, and if so -* check for bad block. -*/ -if (!(page % nand_page_per_block)) { -/* -* Yes, new block. See if this block is good. If not, -* loop until we find a good block. -*/ -while (is_badblock (mtd, offs,1 )) { - page = page + nand_page_per_block; -/* Check i we've reached the end of flash. */ -if (page >= mtd->size >> chip->page_shift) { +while (block <= lastblock && size >0 ) { +if (!is_badblock (mtd, mtd->erasesize * block,1 )) { +/* Skip bad blocks */ +while (page < nand_page_per_block) { +int curr_page = nand_page_per_block * block + page; + +if (mxs_read_page_ecc (mtd, page_buf, curr_page) <0 ) {free (page_buf); -return -ENOMEM ; +return -EIO ; } + +if (size > (mtd->writesize - page_offset)) + sz = (mtd->writesize - page_offset); +else + sz = size; + +memcpy (dst, page_buf + page_offset, sz); + dst += sz; + size -= sz; + page_offset =0 ; + page++; } + + page =0 ; + }else { + lastblock++; } + + block++; }free (page_buf);
@@ -294,6 +289,19 @@ void nand_deselect(void)u32 nand_spl_adjust_offset (u32 sector,u32 offs) { -/* Handle the offset adjust in nand_spl_load_image,*/ +unsigned int block, lastblock; + + block = sector / mtd->erasesize; + lastblock = (sector + offs) / mtd->erasesize; + +while (block <= lastblock) { +if (is_badblock (mtd, block * mtd->erasesize,1 )) { + offs += mtd->erasesize; + lastblock++; + } + + block++; + } +return offs; }
Signature · 1 line
2.25.1
RE: [PATCH V2 2/4] mtd: nand: mxs_nand_spl: Fix bad block skipping
Quoted 33 lines
-----Original Message-----
From: Michael Trimarchi <[email protected]>
Sent: Wednesday, April 27, 2022 12:50 AM
To: Han Xu <[email protected]>; U-Boot-Denx <[email protected]>
Cc: Ye Li <[email protected]>; Stefano Babic <[email protected]>; Miquel Raynal <[email protected]>; Fabio Estevam <[email protected]>; Dario Binacchi <[email protected]>; Sean Anderson <[email protected]>; [email protected]; Jagan Teki <[email protected]>; Ariel D'Alessandro <[email protected]>; Fabio Estevam <[email protected]>
Subject: [PATCH V2 2/4] mtd: nand:mxs_nand_spl: Fix bad block skippingThe specific implementation was having bug. Those bugs are since the beginning of the implementation. Some manufactures can receive this bug in their SPL code. This bug start to be more visible on architecture that has complicated boot process like imx8mn. Older version of uboot has the same problem only if the bad block appear in correspoding of befine of u-boot image. In order to adjust the function we scan from the first block. The logic is not changed to have a simple way to fix without get regression.
The problematic part of old code was in this part:
while (is_badblock(mtd, offs, 1)) { page = page + nand_page_per_block; /* Check i we've reached the end of flash. */ if (page >= mtd->size >> chip->page_shift) { free(page_buf); return -ENOMEM; } }
Quoted 2 lines
Even we fix it adding increment of the offset of one erase block size we don't fix the problem, because the first erase block where the image start is not checked.
Could you please describe more details about your test? Thanks.
Quoted 144 lines
The code was tested on an imx8mn where the boot rom api was not able to skip it. Apart of that other architecure are using this code and all boards that has nand as boot device can be affected
Cc Han Xu <[email protected]>Cc Fabio Estevam <[email protected]>Signed-off-by Michael Trimarchi <[email protected]>---
V1->V2:
- Adjust the commit message
- Add Cc Han Xu and Fabio
- fix size >= 0 to > 0--- drivers/mtd/nand/raw/mxs_nand_spl.c | 90 ++++++++++++++++------------- 1 file changed, 49 insertions(+), 41 deletions(-)drivers/mtd/nand/raw/mxs_nand_spl.c +35
-40@@ -218,14 +218,14 @@ void nand_init(void)mxs_nand_setup_ecc (mtd); } -int nand_spl_load_image (uint32_t offs,unsigned int size,void *buf) +int nand_spl_load_image (uint32_t offs,unsigned int size,void *dst) { -struct nand_chip *chip; -unsigned int page; +unsigned int sz; +unsigned int block, lastblock; +unsigned int page, page_offset;unsigned int nand_page_per_block; -unsigned int sz =0 ; +struct nand_chip *chip;u8 *page_buf =NULL ; -u32 page_off; chip =mtd_to_nand (mtd);if (!chip->numchips)@@ -235,47 +235,42 @@ int nand_spl_load_image(uint32_t offs, unsigned int size, void *buf)if (!page_buf)return -ENOMEM ; - page = offs >> chip->page_shift; - page_off = offs & (mtd->writesize -1 ); +/* offs has to be aligned to a page address! */ + block = offs / mtd->erasesize; + lastblock = (offs + size -1 ) / mtd->erasesize; + page = (offs % mtd->erasesize) / mtd->writesize; + page_offset = offs % mtd->writesize; nand_page_per_block = mtd->erasesize / mtd->writesize; -debug ("%s offset:0x%08x len:%d page:%x\n" , __func__, offs, size, page); - -while (size) { -if (mxs_read_page_ecc (mtd, page_buf, page) <0 ) -return -1 ; - -if (size > (mtd->writesize - page_off)) - sz = (mtd->writesize - page_off); -else - sz = size; - -memcpy (buf, page_buf + page_off, sz); - - offs += mtd->writesize; - page++; - buf += (mtd->writesize - page_off); - page_off =0 ; - size -= sz; - -/* -* Check if we have crossed a block boundary, and if so -* check for bad block. -*/ -if (!(page % nand_page_per_block)) { -/* -* Yes, new block. See if this block is good. If not, -* loop until we find a good block. -*/ -while (is_badblock (mtd, offs,1 )) { - page = page + nand_page_per_block; -/* Check i we've reached the end of flash. */ -if (page >= mtd->size >> chip->page_shift) { +while (block <= lastblock && size >0 ) { +if (!is_badblock (mtd, mtd->erasesize * block,1 )) { +/* Skip bad blocks */ +while (page < nand_page_per_block) { +int curr_page = nand_page_per_block * block + page; + +if (mxs_read_page_ecc (mtd, page_buf, curr_page) <0 ) {free (page_buf); -return -ENOMEM ; +return -EIO ; } + +if (size > (mtd->writesize - page_offset)) + sz = (mtd->writesize - page_offset); +else + sz = size; + +memcpy (dst, page_buf + page_offset, sz); + dst += sz; + size -= sz; + page_offset =0 ; + page++; } + + page =0 ; + }else { + lastblock++; } + + block++;drivers/mtd/nand/raw/mxs_nand_spl.c +14
-2}free (page_buf);@@ -294,6 +289,19 @@ void nand_deselect(void)u32 nand_spl_adjust_offset (u32 sector,u32 offs) { -/* Handle the offset adjust in nand_spl_load_image,*/ +unsigned int block, lastblock; + + block = sector / mtd->erasesize; + lastblock = (sector + offs) / mtd->erasesize; + +while (block <= lastblock) { +if (is_badblock (mtd, block * mtd->erasesize,1 )) { + offs += mtd->erasesize; + lastblock++; + } + + block++; + } +return offs; } --2.25.1
Re: [PATCH V2 2/4] mtd: nand: mxs_nand_spl: Fix bad block skipping
Hi
Quoted 34 lines · Han Xu
Quoted 34 lines
-----Original Message-----
From: Michael Trimarchi <[email protected]>
Sent: Wednesday, April 27, 2022 12:50 AM
To: Han Xu <[email protected]>; U-Boot-Denx <[email protected]>
Cc: Ye Li <[email protected]>; Stefano Babic <[email protected]>; Miquel Raynal <[email protected]>; Fabio Estevam <[email protected]>; Dario Binacchi <[email protected]>; Sean Anderson <[email protected]>; [email protected]; Jagan Teki <[email protected]>; Ariel D'Alessandro <[email protected]>; Fabio Estevam <[email protected]>
Subject: [PATCH V2 2/4] mtd: nand:mxs_nand_spl: Fix bad block skippingThe specific implementation was having bug. Those bugs are since the beginning of the implementation. Some manufactures can receive this bug in their SPL code. This bug start to be more visible on architecture that has complicated boot process like imx8mn. Older version of uboot has the same problem only if the bad block appear in correspoding of befine of u-boot image. In order to adjust the function we scan from the first block. The logic is not changed to have a simple way to fix without get regression.
The problematic part of old code was in this part:
while (is_badblock(mtd, offs, 1)) { page = page + nand_page_per_block; /* Check i we've reached the end of flash. */ if (page >= mtd->size >> chip->page_shift) { free(page_buf); return -ENOMEM; } }
Quoted 4 lines
Quoted 2 lines
Even we fix it adding increment of the offset of one erase block size we don't fix the problem, because the first erase block where the image start is not checked.
Could you please describe more details about your test? Thanks.
Suppose you have a badblock on 5 or 6. Let's start to have only 6 and you write uboot from 5 and let's the uboot be enough big to cover 5, 6, 7, 8
Case 1)
When you write the block 6 the code will skip it as bad during programming. THe image of uboot (or flash.bin) will
be on 5 7 8 9, because the 6 is skipped. The while loop on spl will read (from raw offset the 5) and then he will found the
bad block on next erase block in the while loop and will exists at the end of the flash because the test
is done on the offset and not on the page that is not incremented
Case 2)
Now same example but let's suppose to have block 5 bad. So you write your image and it will start
from a raw offset 5 but it will be written starting from 6. The spl loader will fail because it will not skip
the first block and then will fail anyway to read the image. The patch try to fix the above behavior
Case 3) can be any combination
Michael
Quoted 47 lines
Quoted 47 lines
The code was tested on an imx8mn where the boot rom api was not able to skip it. Apart of that other architecure are using this code and all boards that has nand as boot device can be affected
Cc Han Xu <[email protected]>Cc Fabio Estevam <[email protected]>Signed-off-by Michael Trimarchi <[email protected]>---
V1->V2:
- Adjust the commit message
- Add Cc Han Xu and Fabio
- fix size >= 0 to > 0--- drivers/mtd/nand/raw/mxs_nand_spl.c | 90 ++++++++++++++++------------- 1 file changed, 49 insertions(+), 41 deletions(-)drivers/mtd/nand/raw/mxs_nand_spl.c +5
-5@@ -218,14 +218,14 @@ void nand_init(void)mxs_nand_setup_ecc (mtd); } -int nand_spl_load_image (uint32_t offs,unsigned int size,void *buf) +int nand_spl_load_image (uint32_t offs,unsigned int size,void *dst) { -struct nand_chip *chip; -unsigned int page; +unsigned int sz; +unsigned int block, lastblock; +unsigned int page, page_offset;unsigned int nand_page_per_block; -unsigned int sz =0 ; +struct nand_chip *chip;u8 *page_buf =NULL ; -u32 page_off; chip =mtd_to_nand (mtd);if (!chip->numchips)@@ -235,47 +235,42 @@ int nand_spl_load_image(uint32_t offs, unsigned int size, void *buf)if (!page_buf)return -ENOMEM ;
Quoted 8 lines
drivers/mtd/nand/raw/mxs_nand_spl.c int nand_spl_load_image(uint32_t offs, unsigned int size, void *buf)
Quoted 8 lines
drivers/mtd/nand/raw/mxs_nand_spl.c +5
-2- page = offs >> chip->page_shift; - page_off = offs & (mtd->writesize -1 ); +/* offs has to be aligned to a page address! */ + block = offs / mtd->erasesize; + lastblock = (offs + size -1 ) / mtd->erasesize; + page = (offs % mtd->erasesize) / mtd->writesize; + page_offset = offs % mtd->writesize; nand_page_per_block = mtd->e
Copy Link
MA
Copy Link
NA
Copy Link
rasesize / mtd->writesize;
Quoted 91 lines
Quoted 90 lines
drivers/mtd/nand/raw/mxs_nand_spl.c +25
-33-debug ("%s offset:0x%08x len:%d page:%x\n" , __func__, offs, size, page); - -while (size) { -if (mxs_read_page_ecc (mtd, page_buf, page) <0 ) -return -1 ; - -if (size > (mtd->writesize - page_off)) - sz = (mtd->writesize - page_off); -else - sz = size; - -memcpy (buf, page_buf + page_off, sz); - - offs += mtd->writesize; - page++; - buf += (mtd->writesize - page_off); - page_off =0 ; - size -= sz; - -/* -* Check if we have crossed a block boundary, and if so -* check for bad block. -*/ -if (!(page % nand_page_per_block)) { -/* -* Yes, new block. See if this block is good. If not, -* loop until we find a good block. -*/ -while (is_badblock (mtd, offs,1 )) { - page = page + nand_page_per_block; -/* Check i we've reached the end of flash. */ -if (page >= mtd->size >> chip->page_shift) { +while (block <= lastblock && size >0 ) { +if (!is_badblock (mtd, mtd->erasesize * block,1 )) { +/* Skip bad blocks */ +while (page < nand_page_per_block) { +int curr_page = nand_page_per_block * block + page; + +if (mxs_read_page_ecc (mtd, page_buf, curr_page) <0 ) {free (page_buf); -return -ENOMEM ; +return -EIO ; } + +if (size > (mtd->writesize - page_offset)) + sz = (mtd->writesize - page_offset); +else + sz = size; + +memcpy (dst, page_buf + page_offset, sz); + dst += sz; + size -= sz; + page_offset =0 ; + page++; } + + page =0 ; + }else { + lastblock++; } + + block++; }free(page_buf);drivers/mtd/nand/raw/mxs_nand_spl.c +14
-2@@ -294,6 +289,19 @@ void nand_deselect(void)u32 nand_spl_adjust_offset (u32 sector,u32 offs) { -/* Handle the offset adjust in nand_spl_load_image,*/ +unsigned int block, lastblock; + + block = sector / mtd->erasesize; + lastblock = (sector + offs) / mtd->erasesize; + +while (block <= lastblock) { +if (is_badblock (mtd, block * mtd->erasesize,1 )) { + offs += mtd->erasesize; + lastblock++; + } + + block++; + } +return offs; } --2.25.1
Signature · 11 lines
Michael Nazzareno Trimarchi Co-Founder & Chief Executive Officer M. +39 347 913 2170 [email protected] __________________________________ Amarula Solutions BV Joop Geesinkweg 125, 1114 AB, Amsterdam, NL T. +31 (0)85 111 9172 [email protected] www.amarulasolutions.com
Re: [PATCH V2 2/4] mtd: nand: mxs_nand_spl: Fix bad block skipping
Dear Han and Fabio
Quoted 58 lines · Michael Nazzareno Trimarchi
Hi
Quoted 34 lines · Han Xu
Quoted 34 lines
-----Original Message-----
From: Michael Trimarchi <[email protected]>
Sent: Wednesday, April 27, 2022 12:50 AM
To: Han Xu <[email protected]>; U-Boot-Denx <[email protected]>
Cc: Ye Li <[email protected]>; Stefano Babic <[email protected]>; Miquel Raynal <[email protected]>; Fabio Estevam <[email protected]>; Dario Binacchi <[email protected]>; Sean Anderson <[email protected]>; [email protected]; Jagan Teki <[email protected]>; Ariel D'Alessandro <[email protected]>; Fabio Estevam <[email protected]>
Subject: [PATCH V2 2/4] mtd: nand:mxs_nand_spl: Fix bad block skippingThe specific implementation was having bug. Those bugs are since the beginning of the implementation. Some manufactures can receive this bug in their SPL code. This bug start to be more visible on architecture that has complicated boot process like imx8mn. Older version of uboot has the same problem only if the bad block appear in correspoding of befine of u-boot image. In order to adjust the function we scan from the first block. The logic is not changed to have a simple way to fix without get regression.
The problematic part of old code was in this part:
while (is_badblock(mtd, offs, 1)) { page = page + nand_page_per_block; /* Check i we've reached the end of flash. */ if (page >= mtd->size >> chip->page_shift) { free(page_buf); return -ENOMEM; } }Quoted 4 lines
Quoted 2 lines
Even we fix it adding increment of the offset of one erase block size we don't fix the problem, because the first erase block where the image start is not checked.
Could you please describe more details about your test? Thanks.
Suppose you have a badblock on 5 or 6. Let's start to have only 6 and you write uboot from 5 and let's the uboot be enough big to cover 5, 6, 7, 8
Case 1)
When you write the block 6 the code will skip it as bad during programming. THe image of uboot (or flash.bin) will
be on 5 7 8 9, because the 6 is skipped. The while loop on spl will read (from raw offset the 5) and then he will found the
bad block on next erase block in the while loop and will exists at the end of the flash because the test
is done on the offset and not on the page that is not incrementedCase 2)
Quoted 9 lines
Now same example but let's suppose to have block 5 bad. So you write your image and it will start
from a raw offset 5 but it will be written starting from 6. The spl loader will fail because it will not skip
the first block and then will fail anyway to read the image. The patch try to fix the above behaviorCase 3) can be any combination
Do I need to resend v3?
Michael
Quoted 170 lines
Michael
Quoted 47 lines
Quoted 47 lines
The code was tested on an imx8mn where the boot rom api was not able to skip it. Apart of that other architecure are using this code and all boards that has nand as boot device can be affected
Cc Han Xu <[email protected]>Cc Fabio Estevam <[email protected]>Signed-off-by Michael Trimarchi <[email protected]>---
V1->V2:
- Adjust the commit message
- Add Cc Han Xu and Fabio
- fix size >= 0 to > 0--- drivers/mtd/nand/raw/mxs_nand_spl.c | 90 ++++++++++++++++------------- 1 file changed, 49 insertions(+), 41 deletions(-)drivers/mtd/nand/raw/mxs_nand_spl.c +5
-5@@ -218,14 +218,14 @@ void nand_init(void)mxs_nand_setup_ecc (mtd); } -int nand_spl_load_image (uint32_t offs,unsigned int size,void *buf) +int nand_spl_load_image (uint32_t offs,unsigned int size,void *dst) { -struct nand_chip *chip; -unsigned int page; +unsigned int sz; +unsigned int block, lastblock; +unsigned int page, page_offset;unsigned int nand_page_per_block; -unsigned int sz =0 ; +struct nand_chip *chip;u8 *page_buf =NULL ; -u32 page_off; chip =mtd_to_nand (mtd);if (!chip->numchips)@@ -235,47 +235,42 @@ int nand_spl_load_image(uint32_t offs, unsigned int size, void *buf)if (!page_buf)return -ENOMEM ;Quoted 8 lines
drivers/mtd/nand/raw/mxs_nand_spl.c int nand_spl_load_image(uint32_t offs, unsigned int size, void *buf)
Quoted 8 lines
drivers/mtd/nand/raw/mxs_nand_spl.c +5
-2- page = offs >> chip->page_shift; - page_off = offs & (mtd->writesize -1 ); +/* offs has to be aligned to a page address! */ + block = offs / mtd->erasesize; + lastblock = (offs + size -1 ) / mtd->erasesize; + page = (offs % mtd->erasesize) / mtd->writesize; + page_offset = offs % mtd->writesize; nand_page_per_block = mtd->eCopy Link
MA
Copy Link
NA
Copy Link
rasesize /mtd->writesize;Quoted 82 lines
Quoted 82 lines
drivers/mtd/nand/raw/mxs_nand_spl.c +25
-33-debug ("%s offset:0x%08x len:%d page:%x\n" , __func__, offs, size, page); - -while (size) { -if (mxs_read_page_ecc (mtd, page_buf, page) <0 ) -return -1 ; - -if (size > (mtd->writesize - page_off)) - sz = (mtd->writesize - page_off); -else - sz = size; - -memcpy (buf, page_buf + page_off, sz); - - offs += mtd->writesize; - page++; - buf += (mtd->writesize - page_off); - page_off =0 ; - size -= sz; - -/* -* Check if we have crossed a block boundary, and if so -* check for bad block. -*/ -if (!(page % nand_page_per_block)) { -/* -* Yes, new block. See if this block is good. If not, -* loop until we find a good block. -*/ -while (is_badblock (mtd, offs,1 )) { - page = page + nand_page_per_block; -/* Check i we've reached the end of flash. */ -if (page >= mtd->size >> chip->page_shift) { +while (block <= lastblock && size >0 ) { +if (!is_badblock (mtd, mtd->erasesize * block,1 )) { +/* Skip bad blocks */ +while (page < nand_page_per_block) { +int curr_page = nand_page_per_block * block + page; + +if (mxs_read_page_ecc (mtd, page_buf, curr_page) <0 ) {free (page_buf); -return -ENOMEM ; +return -EIO ; } + +if (size > (mtd->writesize - page_offset)) + sz = (mtd->writesize - page_offset); +else + sz = size; + +memcpy (dst, page_buf + page_offset, sz); + dst += sz; + size -= sz; + page_offset =0 ; + page++; } + + page =0 ; + }else { + lastblock++; } + + block++; }free(page_buf);drivers/mtd/nand/raw/mxs_nand_spl.c +10
-1@@ -294,6 +289,19 @@ void nand_deselect(void)u32 nand_spl_adjust_offset (u32 sector,u32 offs) { -/* Handle the offset adjust in nand_spl_load_image,*/ +unsigned int block, lastblock; + + block = sector / mtd->erasesize; + lastblock = (sector + offs) / mtd->erasesize; + +while (block <= lastblock) { +if (is_badblock (mtd, block * mtd->erasesize,1 )) { + offs += mtd->erasesize; + lastblock++; + }Quoted 9 lines
Quoted 8 lines
drivers/mtd/nand/raw/mxs_nand_spl.c +4
-1+ + block++; + } +return offs; } --2.25.1
--
Michael Nazzareno Trimarchi
Co-Founder & Chief Executive Officer
M. +39 347 913 2170
[email protected]
__________________________________Amarula Solutions BV
Joop Geesinkweg 125, 1114 AB, Amsterdam, NL
T. +31 (0)85 111 9172
[email protected]
www.amarulasolutions.com
Signature · 11 lines
Michael Nazzareno Trimarchi Co-Founder & Chief Executive Officer M. +39 347 913 2170 [email protected] __________________________________ Amarula Solutions BV Joop Geesinkweg 125, 1114 AB, Amsterdam, NL T. +31 (0)85 111 9172 [email protected] www.amarulasolutions.com
Re: [PATCH V2 2/4] mtd: nand: mxs_nand_spl: Fix bad block skipping
Quoted 69 lines · Michael Nazzareno Trimarchi
Dear Han and Fabio
Quoted 65 lines · Michael Nazzareno Trimarchi
Hi
Quoted 34 lines · Han Xu
Quoted 34 lines
-----Original Message-----
From: Michael Trimarchi <[email protected]>
Sent: Wednesday, April 27, 2022 12:50 AM
To: Han Xu <[email protected]>; U-Boot-Denx <[email protected]>
Cc: Ye Li <[email protected]>; Stefano Babic <[email protected]>; Miquel Raynal <[email protected]>; Fabio Estevam <[email protected]>; Dario Binacchi <[email protected]>; Sean Anderson <[email protected]>; [email protected]; Jagan Teki <[email protected]>; Ariel D'Alessandro <[email protected]>; Fabio Estevam <[email protected]>
Subject: [PATCH V2 2/4] mtd: nand:mxs_nand_spl: Fix bad block skippingThe specific implementation was having bug. Those bugs are since the beginning of the implementation. Some manufactures can receive this bug in their SPL code. This bug start to be more visible on architecture that has complicated boot process like imx8mn. Older version of uboot has the same problem only if the bad block appear in correspoding of befine of u-boot image. In order to adjust the function we scan from the first block. The logic is not changed to have a simple way to fix without get regression.
The problematic part of old code was in this part:
while (is_badblock(mtd, offs, 1)) { page = page + nand_page_per_block; /* Check i we've reached the end of flash. */ if (page >= mtd->size >> chip->page_shift) { free(page_buf); return -ENOMEM; } }Quoted 4 lines
Quoted 2 lines
Even we fix it adding increment of the offset of one erase block size we don't fix the problem, because the first erase block where the image start is not checked.
Could you please describe more details about your test? Thanks.
Suppose you have a badblock on 5 or 6. Let's start to have only 6 and you write uboot from 5 and let's the uboot be enough big to cover 5, 6, 7, 8
Case 1)
When you write the block 6 the code will skip it as bad during programming. THe image of uboot (or flash.bin) will
be on 5 7 8 9, because the 6 is skipped. The while loop on spl will read (from raw offset the 5) and then he will found the
bad block on next erase block in the while loop and will exists at the end of the flash because the test
is done on the offset and not on the page that is not incrementedCase 2)
Now same example but let's suppose to have block 5 bad. So you write your image and it will start
from a raw offset 5 but it will be written starting from 6. The spl loader will fail because it will not skip
the first block and then will fail anyway to read the image. The patch try to fix the above behavior
Quoted 6 lines
Quoted 2 lines
Case 3) can be any combination
Do I need to resend v3?
Michael
It's not about the v3. I need to discuss some details with ROM team but unfortunately they are all in vacation till May 5th. I will respond after the discussion.
Quoted 186 lines
Quoted 171 lines
Michael
Quoted 47 lines
Quoted 47 lines
The code was tested on an imx8mn where the boot rom api was not able to skip it. Apart of that other architecure are using this code and all boards that has nand as boot device can be affected
Cc Han Xu <[email protected]>Cc Fabio Estevam <[email protected]>Signed-off-by Michael Trimarchi <[email protected]>---
V1->V2:
- Adjust the commit message
- Add Cc Han Xu and Fabio
- fix size >= 0 to > 0--- drivers/mtd/nand/raw/mxs_nand_spl.c | 90 ++++++++++++++++------------- 1 file changed, 49 insertions(+), 41 deletions(-)drivers/mtd/nand/raw/mxs_nand_spl.c +5
-5@@ -218,14 +218,14 @@ void nand_init(void)mxs_nand_setup_ecc (mtd); } -int nand_spl_load_image (uint32_t offs,unsigned int size,void *buf) +int nand_spl_load_image (uint32_t offs,unsigned int size,void *dst) { -struct nand_chip *chip; -unsigned int page; +unsigned int sz; +unsigned int block, lastblock; +unsigned int page, page_offset;unsigned int nand_page_per_block; -unsigned int sz =0 ; +struct nand_chip *chip;u8 *page_buf =NULL ; -u32 page_off; chip =mtd_to_nand (mtd);if (!chip->numchips)@@ -235,47 +235,42 @@ int nand_spl_load_image(uint32_t offs, unsigned int size, void *buf)if (!page_buf)return -ENOMEM ;Quoted 8 lines
drivers/mtd/nand/raw/mxs_nand_spl.c int nand_spl_load_image(uint32_t offs, unsigned int size, void *buf)
Quoted 8 lines
drivers/mtd/nand/raw/mxs_nand_spl.c +5
-2- page = offs >> chip->page_shift; - page_off = offs & (mtd->writesize -1 ); +/* offs has to be aligned to a page address! */ + block = offs / mtd->erasesize; + lastblock = (offs + size -1 ) / mtd->erasesize; + page = (offs % mtd->erasesize) / mtd->writesize; + page_offset = offs % mtd->writesize; nand_page_per_block = mtd->eCopy Link
MA
Copy Link
NA
Copy Link
rasesize /mtd->writesize;Quoted 82 lines
Quoted 82 lines
drivers/mtd/nand/raw/mxs_nand_spl.c +25
-33-debug ("%s offset:0x%08x len:%d page:%x\n" , __func__, offs, size, page); - -while (size) { -if (mxs_read_page_ecc (mtd, page_buf, page) <0 ) -return -1 ; - -if (size > (mtd->writesize - page_off)) - sz = (mtd->writesize - page_off); -else - sz = size; - -memcpy (buf, page_buf + page_off, sz); - - offs += mtd->writesize; - page++; - buf += (mtd->writesize - page_off); - page_off =0 ; - size -= sz; - -/* -* Check if we have crossed a block boundary, and if so -* check for bad block. -*/ -if (!(page % nand_page_per_block)) { -/* -* Yes, new block. See if this block is good. If not, -* loop until we find a good block. -*/ -while (is_badblock (mtd, offs,1 )) { - page = page + nand_page_per_block; -/* Check i we've reached the end of flash. */ -if (page >= mtd->size >> chip->page_shift) { +while (block <= lastblock && size >0 ) { +if (!is_badblock (mtd, mtd->erasesize * block,1 )) { +/* Skip bad blocks */ +while (page < nand_page_per_block) { +int curr_page = nand_page_per_block * block + page; + +if (mxs_read_page_ecc (mtd, page_buf, curr_page) <0 ) {free (page_buf); -return -ENOMEM ; +return -EIO ; } + +if (size > (mtd->writesize - page_offset)) + sz = (mtd->writesize - page_offset); +else + sz = size; + +memcpy (dst, page_buf + page_offset, sz); + dst += sz; + size -= sz; + page_offset =0 ; + page++; } + + page =0 ; + }else { + lastblock++; } + + block++; }free(page_buf);drivers/mtd/nand/raw/mxs_nand_spl.c +10
-1@@ -294,6 +289,19 @@ void nand_deselect(void)u32 nand_spl_adjust_offset (u32 sector,u32 offs) { -/* Handle the offset adjust in nand_spl_load_image,*/ +unsigned int block, lastblock; + + block = sector / mtd->erasesize; + lastblock = (sector + offs) / mtd->erasesize; + +while (block <= lastblock) { +if (is_badblock (mtd, block * mtd->erasesize,1 )) { + offs += mtd->erasesize; + lastblock++; + }Quoted 9 lines
Quoted 8 lines
drivers/mtd/nand/raw/mxs_nand_spl.c +4
-1+ + block++; + } +return offs; } --2.25.1
--
Michael Nazzareno Trimarchi
Co-Founder & Chief Executive Officer
M. +39 347 913 2170
[email protected]
__________________________________Amarula Solutions BV
Joop Geesinkweg 125, 1114 AB, Amsterdam, NL
T. +31 (0)85 111 9172
[email protected]
https://eur01.safelinks.protection.outlook.com/?url=http%3A%2F%2Fwww.amarulasolutions.com%2F&data=05%7C01%7Chan.xu%40nxp.com%7C81091e4443d745ce460708da2b3cec6a%7C686ea1d3bc2b4c6fa92cd99c5c301635%7C0%7C0%7C637869837927785927%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000%7C%7C%7C&sdata=I5jvEx5qmQLzO8QrQT9aMYttDPZhdvbGi0Z2lk77bio%3D&reserved=0Signature · 11 lines
Michael Nazzareno Trimarchi Co-Founder & Chief Executive Officer M. +39 347 913 2170 [email protected] __________________________________ Amarula Solutions BV Joop Geesinkweg 125, 1114 AB, Amsterdam, NL T. +31 (0)85 111 9172 [email protected] https://eur01.safelinks.protection.outlook.com/?url=http%3A%2F%2Fwww.amarulasolutions.com%2F&data=05%7C01%7Chan.xu%40nxp.com%7C81091e4443d745ce460708da2b3cec6a%7C686ea1d3bc2b4c6fa92cd99c5c301635%7C0%7C0%7C637869837927785927%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000%7C%7C%7C&sdata=I5jvEx5qmQLzO8QrQT9aMYttDPZhdvbGi0Z2lk77bio%3D&reserved=0
Re: [PATCH V2 2/4] mtd: nand: mxs_nand_spl: Fix bad block skipping
Hi Han
Quoted 74 lines · Han Xu
Quoted 72 lines · Michael Nazzareno Trimarchi
Dear Han and Fabio
Quoted 67 lines · Michael Nazzareno Trimarchi
Hi
Quoted 34 lines · Han Xu
Quoted 34 lines
-----Original Message-----
From: Michael Trimarchi <[email protected]>
Sent: Wednesday, April 27, 2022 12:50 AM
To: Han Xu <[email protected]>; U-Boot-Denx <[email protected]>
Cc: Ye Li <[email protected]>; Stefano Babic <[email protected]>; Miquel Raynal <[email protected]>; Fabio Estevam <[email protected]>; Dario Binacchi <[email protected]>; Sean Anderson <[email protected]>; [email protected]; Jagan Teki <[email protected]>; Ariel D'Alessandro <[email protected]>; Fabio Estevam <[email protected]>
Subject: [PATCH V2 2/4] mtd: nand:mxs_nand_spl: Fix bad block skippingThe specific implementation was having bug. Those bugs are since the beginning of the implementation. Some manufactures can receive this bug in their SPL code. This bug start to be more visible on architecture that has complicated boot process like imx8mn. Older version of uboot has the same problem only if the bad block appear in correspoding of befine of u-boot image. In order to adjust the function we scan from the first block. The logic is not changed to have a simple way to fix without get regression.
The problematic part of old code was in this part:
while (is_badblock(mtd, offs, 1)) { page = page + nand_page_per_block; /* Check i we've reached the end of flash. */ if (page >= mtd->size >> chip->page_shift) { free(page_buf); return -ENOMEM; } }Quoted 4 lines
Quoted 2 lines
Even we fix it adding increment of the offset of one erase block size we don't fix the problem, because the first erase block where the image start is not checked.
Could you please describe more details about your test? Thanks.
Suppose you have a badblock on 5 or 6. Let's start to have only 6 and you write uboot from 5 and let's the uboot be enough big to cover 5, 6, 7, 8
Case 1)
When you write the block 6 the code will skip it as bad during programming. THe image of uboot (or flash.bin) will
be on 5 7 8 9, because the 6 is skipped. The while loop on spl will read (from raw offset the 5) and then he will found the
bad block on next erase block in the while loop and will exists at the end of the flash because the test
is done on the offset and not on the page that is not incrementedCase 2)
Now same example but let's suppose to have block 5 bad. So you write your image and it will start
from a raw offset 5 but it will be written starting from 6. The spl loader will fail because it will not skip
the first block and then will fail anyway to read the image. The patch try to fix the above behaviorCase 3) can be any combination
Quoted 8 lines
Quoted 3 lines
Do I need to resend v3?
Michael
It's not about the v3. I need to discuss some details with ROM team but unfortunately they are all in vacation till May 5th. I will respond after the discussion.
No problem I will wait, but the code is used by other nxp architectures that don't use the
rom api.
Michael
Definit
Quoted 186 lines
Quoted 186 lines
Quoted 166 lines
Michael
Quoted 47 lines
Quoted 47 lines
The code was tested on an imx8mn where the boot rom api was not able to skip it. Apart of that other architecure are using this code and all boards that has nand as boot device can be affected
Cc Han Xu <[email protected]>Cc Fabio Estevam <[email protected]>Signed-off-by Michael Trimarchi <[email protected]>---
V1->V2:
- Adjust the commit message
- Add Cc Han Xu and Fabio
- fix size >= 0 to > 0--- drivers/mtd/nand/raw/mxs_nand_spl.c | 90 ++++++++++++++++------------- 1 file changed, 49 insertions(+), 41 deletions(-)drivers/mtd/nand/raw/mxs_nand_spl.c +5
-5@@ -218,14 +218,14 @@ void nand_init(void)mxs_nand_setup_ecc (mtd); } -int nand_spl_load_image (uint32_t offs,unsigned int size,void *buf) +int nand_spl_load_image (uint32_t offs,unsigned int size,void *dst) { -struct nand_chip *chip; -unsigned int page; +unsigned int sz; +unsigned int block, lastblock; +unsigned int page, page_offset;unsigned int nand_page_per_block; -unsigned int sz =0 ; +struct nand_chip *chip;u8 *page_buf =NULL ; -u32 page_off; chip =mtd_to_nand (mtd);if (!chip->numchips)@@ -235,47 +235,42 @@ int nand_spl_load_image(uint32_t offs, unsigned int size, void *buf)if (!page_buf)return -ENOMEM ;Quoted 8 lines
drivers/mtd/nand/raw/mxs_nand_spl.c int nand_spl_load_image(uint32_t offs, unsigned int size, void *buf)
Quoted 8 lines
drivers/mtd/nand/raw/mxs_nand_spl.c +5
-2- page = offs >> chip->page_shift; - page_off = offs & (mtd->writesize -1 ); +/* offs has to be aligned to a page address! */ + block = offs / mtd->erasesize; + lastblock = (offs + size -1 ) / mtd->erasesize; + page = (offs % mtd->erasesize) / mtd->writesize; + page_offset = offs % mtd->writesize; nand_page_per_block = mtd->eCopy Link
MA
Copy Link
NA
Copy Link
rasesize /mtd->writesize;Quoted 82 lines
Quoted 82 lines
drivers/mtd/nand/raw/mxs_nand_spl.c +25
-33-debug ("%s offset:0x%08x len:%d page:%x\n" , __func__, offs, size, page); - -while (size) { -if (mxs_read_page_ecc (mtd, page_buf, page) <0 ) -return -1 ; - -if (size > (mtd->writesize - page_off)) - sz = (mtd->writesize - page_off); -else - sz = size; - -memcpy (buf, page_buf + page_off, sz); - - offs += mtd->writesize; - page++; - buf += (mtd->writesize - page_off); - page_off =0 ; - size -= sz; - -/* -* Check if we have crossed a block boundary, and if so -* check for bad block. -*/ -if (!(page % nand_page_per_block)) { -/* -* Yes, new block. See if this block is good. If not, -* loop until we find a good block. -*/ -while (is_badblock (mtd, offs,1 )) { - page = page + nand_page_per_block; -/* Check i we've reached the end of flash. */ -if (page >= mtd->size >> chip->page_shift) { +while (block <= lastblock && size >0 ) { +if (!is_badblock (mtd, mtd->erasesize * block,1 )) { +/* Skip bad blocks */ +while (page < nand_page_per_block) { +int curr_page = nand_page_per_block * block + page; + +if (mxs_read_page_ecc (mtd, page_buf, curr_page) <0 ) {free (page_buf); -return -ENOMEM ; +return -EIO ; } + +if (size > (mtd->writesize - page_offset)) + sz = (mtd->writesize - page_offset); +else + sz = size; + +memcpy (dst, page_buf + page_offset, sz); + dst += sz; + size -= sz; + page_offset =0 ; + page++; } + + page =0 ; + }else { + lastblock++; } + + block++; }free(page_buf);drivers/mtd/nand/raw/mxs_nand_spl.c +10
-1@@ -294,6 +289,19 @@ void nand_deselect(void)u32 nand_spl_adjust_offset (u32 sector,u32 offs) { -/* Handle the offset adjust in nand_spl_load_image,*/ +unsigned int block, lastblock; + + block = sector / mtd->erasesize; + lastblock = (sector + offs) / mtd->erasesize; + +while (block <= lastblock) { +if (is_badblock (mtd, block * mtd->erasesize,1 )) { + offs += mtd->erasesize; + lastblock++; + }Quoted 9 lines
Quoted 8 lines
drivers/mtd/nand/raw/mxs_nand_spl.c +4
-1+ + block++; + } +return offs; } --2.25.1
--
Michael Nazzareno Trimarchi
Co-Founder & Chief Executive Officer
M. +39 347 913 2170
[email protected]
__________________________________Quoted 5 lines
Amarula Solutions BV
Joop Geesinkweg 125, 1114 AB, Amsterdam, NL
T. +31 (0)85 111 9172
[email protected]
https://eur01.safelinks.protection.outlook.com/?url=http%3A%2F%2Fwww.amarulasolutions.com%2F&data=05%7C01%7Chan.xu%40nxp.com%7C81091e4443d745ce460708da2b3cec6a%7C686ea1d3bc2b4c6fa92cd99c5c301635%7C0%7C0%7C637869837927785927%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000%7C%7C%7C&sdata=I5jvEx5qmQLzO8QrQT9aMYttDPZhdvbGi0Z2lk77bio%3D&reserved=0--
Michael Nazzareno Trimarchi
Co-Founder & Chief Executive Officer
M. +39 347 913 2170
[email protected]
__________________________________Amarula Solutions BV
Joop Geesinkweg 125, 1114 AB, Amsterdam, NL
T. +31 (0)85 111 9172
[email protected]
https://eur01.safelinks.protection.outlook.com/?url=http%3A%2F%2Fwww.amarulasolutions.com%2F&data=05%7C01%7Chan.xu%40nxp.com%7C81091e4443d745ce460708da2b3cec6a%7C686ea1d3bc2b4c6fa92cd99c5c301635%7C0%7C0%7C637869837927785927%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000%7C%7C%7C&sdata=I5jvEx5qmQLzO8QrQT9aMYttDPZhdvbGi0Z2lk77bio%3D&reserved=0
Signature · 11 lines
Michael Nazzareno Trimarchi Co-Founder & Chief Executive Officer M. +39 347 913 2170 [email protected] __________________________________ Amarula Solutions BV Joop Geesinkweg 125, 1114 AB, Amsterdam, NL T. +31 (0)85 111 9172 [email protected] www.amarulasolutions.com
Re: [PATCH V2 2/4] mtd: nand: mxs_nand_spl: Fix bad block skipping
Hi Han
Any update?
Michael
Quoted 295 lines · Michael Nazzareno Trimarchi
Hi Han
Quoted 74 lines · Han Xu
Quoted 72 lines · Michael Nazzareno Trimarchi
Dear Han and Fabio
Quoted 67 lines · Michael Nazzareno Trimarchi
Hi
Quoted 34 lines · Han Xu
Quoted 34 lines
-----Original Message-----
From: Michael Trimarchi <[email protected]>
Sent: Wednesday, April 27, 2022 12:50 AM
To: Han Xu <[email protected]>; U-Boot-Denx <[email protected]>
Cc: Ye Li <[email protected]>; Stefano Babic <[email protected]>; Miquel Raynal <[email protected]>; Fabio Estevam <[email protected]>; Dario Binacchi <[email protected]>; Sean Anderson <[email protected]>; [email protected]; Jagan Teki <[email protected]>; Ariel D'Alessandro <[email protected]>; Fabio Estevam <[email protected]>
Subject: [PATCH V2 2/4] mtd: nand:mxs_nand_spl: Fix bad block skippingThe specific implementation was having bug. Those bugs are since the beginning of the implementation. Some manufactures can receive this bug in their SPL code. This bug start to be more visible on architecture that has complicated boot process like imx8mn. Older version of uboot has the same problem only if the bad block appear in correspoding of befine of u-boot image. In order to adjust the function we scan from the first block. The logic is not changed to have a simple way to fix without get regression.
The problematic part of old code was in this part:
while (is_badblock(mtd, offs, 1)) { page = page + nand_page_per_block; /* Check i we've reached the end of flash. */ if (page >= mtd->size >> chip->page_shift) { free(page_buf); return -ENOMEM; } }Quoted 4 lines
Quoted 2 lines
Even we fix it adding increment of the offset of one erase block size we don't fix the problem, because the first erase block where the image start is not checked.
Could you please describe more details about your test? Thanks.
Suppose you have a badblock on 5 or 6. Let's start to have only 6 and you write uboot from 5 and let's the uboot be enough big to cover 5, 6, 7, 8
Case 1)
When you write the block 6 the code will skip it as bad during programming. THe image of uboot (or flash.bin) will
be on 5 7 8 9, because the 6 is skipped. The while loop on spl will read (from raw offset the 5) and then he will found the
bad block on next erase block in the while loop and will exists at the end of the flash because the test
is done on the offset and not on the page that is not incrementedCase 2)
Now same example but let's suppose to have block 5 bad. So you write your image and it will start
from a raw offset 5 but it will be written starting from 6. The spl loader will fail because it will not skip
the first block and then will fail anyway to read the image. The patch try to fix the above behaviorCase 3) can be any combination
Quoted 8 lines
Quoted 3 lines
Do I need to resend v3?
Michael
It's not about the v3. I need to discuss some details with ROM team but unfortunately they are all in vacation till May 5th. I will respond after the discussion.
No problem I will wait, but the code is used by other nxp architectures that don't use the
rom api.Michael
Definit
Quoted 181 lines
Quoted 181 lines
Quoted 166 lines
Michael
Quoted 47 lines
Quoted 47 lines
The code was tested on an imx8mn where the boot rom api was not able to skip it. Apart of that other architecure are using this code and all boards that has nand as boot device can be affected
Cc Han Xu <[email protected]>Cc Fabio Estevam <[email protected]>Signed-off-by Michael Trimarchi <[email protected]>---
V1->V2:
- Adjust the commit message
- Add Cc Han Xu and Fabio
- fix size >= 0 to > 0--- drivers/mtd/nand/raw/mxs_nand_spl.c | 90 ++++++++++++++++------------- 1 file changed, 49 insertions(+), 41 deletions(-)drivers/mtd/nand/raw/mxs_nand_spl.c +5
-5@@ -218,14 +218,14 @@ void nand_init(void)mxs_nand_setup_ecc (mtd); } -int nand_spl_load_image (uint32_t offs,unsigned int size,void *buf) +int nand_spl_load_image (uint32_t offs,unsigned int size,void *dst) { -struct nand_chip *chip; -unsigned int page; +unsigned int sz; +unsigned int block, lastblock; +unsigned int page, page_offset;unsigned int nand_page_per_block; -unsigned int sz =0 ; +struct nand_chip *chip;u8 *page_buf =NULL ; -u32 page_off; chip =mtd_to_nand (mtd);if (!chip->numchips)@@ -235,47 +235,42 @@ int nand_spl_load_image(uint32_t offs, unsigned int size, void *buf)if (!page_buf)return -ENOMEM ;Quoted 8 lines
drivers/mtd/nand/raw/mxs_nand_spl.c int nand_spl_load_image(uint32_t offs, unsigned int size, void *buf)
Quoted 8 lines
drivers/mtd/nand/raw/mxs_nand_spl.c +5
-2- page = offs >> chip->page_shift; - page_off = offs & (mtd->writesize -1 ); +/* offs has to be aligned to a page address! */ + block = offs / mtd->erasesize; + lastblock = (offs + size -1 ) / mtd->erasesize; + page = (offs % mtd->erasesize) / mtd->writesize; + page_offset = offs % mtd->writesize; nand_page_per_block = mtd->eCopy Link
MA
Copy Link
NA
Copy Link
rasesize /mtd->writesize;Quoted 82 lines
Quoted 82 lines
drivers/mtd/nand/raw/mxs_nand_spl.c +25
-33-debug ("%s offset:0x%08x len:%d page:%x\n" , __func__, offs, size, page); - -while (size) { -if (mxs_read_page_ecc (mtd, page_buf, page) <0 ) -return -1 ; - -if (size > (mtd->writesize - page_off)) - sz = (mtd->writesize - page_off); -else - sz = size; - -memcpy (buf, page_buf + page_off, sz); - - offs += mtd->writesize; - page++; - buf += (mtd->writesize - page_off); - page_off =0 ; - size -= sz; - -/* -* Check if we have crossed a block boundary, and if so -* check for bad block. -*/ -if (!(page % nand_page_per_block)) { -/* -* Yes, new block. See if this block is good. If not, -* loop until we find a good block. -*/ -while (is_badblock (mtd, offs,1 )) { - page = page + nand_page_per_block; -/* Check i we've reached the end of flash. */ -if (page >= mtd->size >> chip->page_shift) { +while (block <= lastblock && size >0 ) { +if (!is_badblock (mtd, mtd->erasesize * block,1 )) { +/* Skip bad blocks */ +while (page < nand_page_per_block) { +int curr_page = nand_page_per_block * block + page; + +if (mxs_read_page_ecc (mtd, page_buf, curr_page) <0 ) {free (page_buf); -return -ENOMEM ; +return -EIO ; } + +if (size > (mtd->writesize - page_offset)) + sz = (mtd->writesize - page_offset); +else + sz = size; + +memcpy (dst, page_buf + page_offset, sz); + dst += sz; + size -= sz; + page_offset =0 ; + page++; } + + page =0 ; + }else { + lastblock++; } + + block++; }free(page_buf);drivers/mtd/nand/raw/mxs_nand_spl.c +10
-1@@ -294,6 +289,19 @@ void nand_deselect(void)u32 nand_spl_adjust_offset (u32 sector,u32 offs) { -/* Handle the offset adjust in nand_spl_load_image,*/ +unsigned int block, lastblock; + + block = sector / mtd->erasesize; + lastblock = (sector + offs) / mtd->erasesize; + +while (block <= lastblock) { +if (is_badblock (mtd, block * mtd->erasesize,1 )) { + offs += mtd->erasesize; + lastblock++; + }Quoted 9 lines
Quoted 8 lines
drivers/mtd/nand/raw/mxs_nand_spl.c +4
-1+ + block++; + } +return offs; } --2.25.1
--
Michael Nazzareno Trimarchi
Co-Founder & Chief Executive Officer
M. +39 347 913 2170
[email protected]
__________________________________Quoted 5 lines
Amarula Solutions BV
Joop Geesinkweg 125, 1114 AB, Amsterdam, NL
T. +31 (0)85 111 9172
[email protected]
https://eur01.safelinks.protection.outlook.com/?url=http%3A%2F%2Fwww.amarulasolutions.com%2F&data=05%7C01%7Chan.xu%40nxp.com%7C81091e4443d745ce460708da2b3cec6a%7C686ea1d3bc2b4c6fa92cd99c5c301635%7C0%7C0%7C637869837927785927%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000%7C%7C%7C&sdata=I5jvEx5qmQLzO8QrQT9aMYttDPZhdvbGi0Z2lk77bio%3D&reserved=0--
Michael Nazzareno Trimarchi
Co-Founder & Chief Executive Officer
M. +39 347 913 2170
[email protected]
__________________________________Quoted 5 lines
Quoted 5 lines
Amarula Solutions BV
Joop Geesinkweg 125, 1114 AB, Amsterdam, NL
T. +31 (0)85 111 9172
[email protected]
https://eur01.safelinks.protection.outlook.com/?url=http%3A%2F%2Fwww.amarulasolutions.com%2F&data=05%7C01%7Chan.xu%40nxp.com%7C81091e4443d745ce460708da2b3cec6a%7C686ea1d3bc2b4c6fa92cd99c5c301635%7C0%7C0%7C637869837927785927%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000%7C%7C%7C&sdata=I5jvEx5qmQLzO8QrQT9aMYttDPZhdvbGi0Z2lk77bio%3D&reserved=0--
Michael Nazzareno Trimarchi
Co-Founder & Chief Executive Officer
M. +39 347 913 2170
[email protected]
__________________________________Amarula Solutions BV
Joop Geesinkweg 125, 1114 AB, Amsterdam, NL
T. +31 (0)85 111 9172
[email protected]
www.amarulasolutions.com
Signature · 11 lines
Michael Nazzareno Trimarchi Co-Founder & Chief Executive Officer M. +39 347 913 2170 [email protected] __________________________________ Amarula Solutions BV Joop Geesinkweg 125, 1114 AB, Amsterdam, NL T. +31 (0)85 111 9172 [email protected] www.amarulasolutions.com
Re: [PATCH V2 2/4] mtd: nand: mxs_nand_spl: Fix bad block skipping
In reply to Michael TrimarchiQuoted 27 lines · Michael Trimarchi
The specific implementation was having bug. Those bugs are since the beginning of the implementation. Some manufactures can receive this bug in their SPL code. This bug start to be more visible on architecture that has complicated boot process like imx8mn. Older version of uboot has the same problem only if the bad block appear in correspoding of befine of u-boot image. In order to adjust the function we scan from the first block. The logic is not changed to have a simple way to fix without get regression.
The problematic part of old code was in this part:
while (is_badblock(mtd, offs, 1)) { page = page + nand_page_per_block; /* Check i we've reached the end of flash. */ if (page >= mtd->size >> chip->page_shift) { free(page_buf); return -ENOMEM; } }Even we fix it adding increment of the offset of one erase block size we don't fix the problem, because the first erase block where the image start is not checked. The code was tested on an imx8mn where the boot rom api was not able to skip it. Apart of that other architecure are using this code and all boards that has nand as boot device can be affected
Quoted 3 lines
Cc Han Xu <[email protected]>Cc Fabio Estevam <[email protected]>Signed-off-by Michael Trimarchi <[email protected]>
Quoted 135 lines
---
V1->V2:
- Adjust the commit message
- Add Cc Han Xu and Fabio
- fix size >= 0 to > 0--- drivers/mtd/nand/raw/mxs_nand_spl.c | 90 ++++++++++++++++------------- 1 file changed, 49 insertions(+), 41 deletions(-)drivers/mtd/nand/raw/mxs_nand_spl.c +49
-41@@ -218,14 +218,14 @@ void nand_init(void)mxs_nand_setup_ecc (mtd); } -int nand_spl_load_image (uint32_t offs,unsigned int size,void *buf) +int nand_spl_load_image (uint32_t offs,unsigned int size,void *dst) { -struct nand_chip *chip; -unsigned int page; +unsigned int sz; +unsigned int block, lastblock; +unsigned int page, page_offset;unsigned int nand_page_per_block; -unsigned int sz =0 ; +struct nand_chip *chip;u8 *page_buf =NULL ; -u32 page_off; chip =mtd_to_nand (mtd);if (!chip->numchips)@@ -235,47 +235,42 @@ int nand_spl_load_image(uint32_t offs, unsigned int size, void *buf)if (!page_buf)return -ENOMEM ; - page = offs >> chip->page_shift; - page_off = offs & (mtd->writesize -1 ); +/* offs has to be aligned to a page address! */ + block = offs / mtd->erasesize; + lastblock = (offs + size -1 ) / mtd->erasesize; + page = (offs % mtd->erasesize) / mtd->writesize; + page_offset = offs % mtd->writesize; nand_page_per_block = mtd->erasesize / mtd->writesize; -debug ("%s offset:0x%08x len:%d page:%x\n" , __func__, offs, size, page); - -while (size) { -if (mxs_read_page_ecc (mtd, page_buf, page) <0 ) -return -1 ; - -if (size > (mtd->writesize - page_off)) - sz = (mtd->writesize - page_off); -else - sz = size; - -memcpy (buf, page_buf + page_off, sz); - - offs += mtd->writesize; - page++; - buf += (mtd->writesize - page_off); - page_off =0 ; - size -= sz; - -/* -* Check if we have crossed a block boundary, and if so -* check for bad block. -*/ -if (!(page % nand_page_per_block)) { -/* -* Yes, new block. See if this block is good. If not, -* loop until we find a good block. -*/ -while (is_badblock (mtd, offs,1 )) { - page = page + nand_page_per_block; -/* Check i we've reached the end of flash. */ -if (page >= mtd->size >> chip->page_shift) { +while (block <= lastblock && size >0 ) { +if (!is_badblock (mtd, mtd->erasesize * block,1 )) { +/* Skip bad blocks */ +while (page < nand_page_per_block) { +int curr_page = nand_page_per_block * block + page; + +if (mxs_read_page_ecc (mtd, page_buf, curr_page) <0 ) {free (page_buf); -return -ENOMEM ; +return -EIO ; } + +if (size > (mtd->writesize - page_offset)) + sz = (mtd->writesize - page_offset); +else + sz = size; + +memcpy (dst, page_buf + page_offset, sz); + dst += sz; + size -= sz; + page_offset =0 ; + page++; } + + page =0 ; + }else { + lastblock++; } + + block++; }free (page_buf);@@ -294,6 +289,19 @@ void nand_deselect(void)u32 nand_spl_adjust_offset (u32 sector,u32 offs) { -/* Handle the offset adjust in nand_spl_load_image,*/ +unsigned int block, lastblock; + + block = sector / mtd->erasesize; + lastblock = (sector + offs) / mtd->erasesize; + +while (block <= lastblock) { +if (is_badblock (mtd, block * mtd->erasesize,1 )) { + offs += mtd->erasesize; + lastblock++; + } + + block++; + } +return offs; }Signature · 1 line
2.25.1
Re: [PATCH V2 2/4] mtd: nand: mxs_nand_spl: Fix bad block skipping
Hi
Thank you. What is the reply from bootrom team?
Michael
Quoted 170 lines · Han Xu
Quoted 27 lines · Michael Trimarchi
The specific implementation was having bug. Those bugs are since the beginning of the implementation. Some manufactures can receive this bug in their SPL code. This bug start to be more visible on architecture that has complicated boot process like imx8mn. Older version of uboot has the same problem only if the bad block appear in correspoding of befine of u-boot image. In order to adjust the function we scan from the first block. The logic is not changed to have a simple way to fix without get regression.
The problematic part of old code was in this part:
while (is_badblock(mtd, offs, 1)) { page = page + nand_page_per_block; /* Check i we've reached the end of flash. */ if (page >= mtd->size >> chip->page_shift) { free(page_buf); return -ENOMEM; } }Even we fix it adding increment of the offset of one erase block size we don't fix the problem, because the first erase block where the image start is not checked. The code was tested on an imx8mn where the boot rom api was not able to skip it. Apart of that other architecure are using this code and all boards that has nand as boot device can be affected
Quoted 3 lines
Cc Han Xu <[email protected]>Cc Fabio Estevam <[email protected]>Signed-off-by Michael Trimarchi <[email protected]>Acked-by Han Xu <[email protected]>Quoted 135 lines
---
V1->V2:
- Adjust the commit message
- Add Cc Han Xu and Fabio
- fix size >= 0 to > 0--- drivers/mtd/nand/raw/mxs_nand_spl.c | 90 ++++++++++++++++------------- 1 file changed, 49 insertions(+), 41 deletions(-)drivers/mtd/nand/raw/mxs_nand_spl.c +49
-42@@ -218,14 +218,14 @@ void nand_init(void)mxs_nand_setup_ecc (mtd); } -int nand_spl_load_image (uint32_t offs,unsigned int size,void *buf) +int nand_spl_load_image (uint32_t offs,unsigned int size,void *dst) { -struct nand_chip *chip; -unsigned int page; +unsigned int sz; +unsigned int block, lastblock; +unsigned int page, page_offset;unsigned int nand_page_per_block; -unsigned int sz =0 ; +struct nand_chip *chip;u8 *page_buf =NULL ; -u32 page_off; chip =mtd_to_nand (mtd);if (!chip->numchips)@@ -235,47 +235,42 @@ int nand_spl_load_image(uint32_t offs, unsigned int size, void *buf)if (!page_buf)return -ENOMEM ; - page = offs >> chip->page_shift; - page_off = offs & (mtd->writesize -1 ); +/* offs has to be aligned to a page address! */ + block = offs / mtd->erasesize; + lastblock = (offs + size -1 ) / mtd->erasesize; + page = (offs % mtd->erasesize) / mtd->writesize; + page_offset = offs % mtd->writesize; nand_page_per_block = mtd->erasesize / mtd->writesize; -debug ("%s offset:0x%08x len:%d page:%x\n" , __func__, offs, size, page); - -while (size) { -if (mxs_read_page_ecc (mtd, page_buf, page) <0 ) -return -1 ; - -if (size > (mtd->writesize - page_off)) - sz = (mtd->writesize - page_off); -else - sz = size; - -memcpy (buf, page_buf + page_off, sz); - - offs += mtd->writesize; - page++; - buf += (mtd->writesize - page_off); - page_off =0 ; - size -= sz; - -/* -* Check if we have crossed a block boundary, and if so -* check for bad block. -*/ -if (!(page % nand_page_per_block)) { -/* -* Yes, new block. See if this block is good. If not, -* loop until we find a good block. -*/ -while (is_badblock (mtd, offs,1 )) { - page = page + nand_page_per_block; -/* Check i we've reached the end of flash. */ -if (page >= mtd->size >> chip->page_shift) { +while (block <= lastblock && size >0 ) { +if (!is_badblock (mtd, mtd->erasesize * block,1 )) { +/* Skip bad blocks */ +while (page < nand_page_per_block) { +int curr_page = nand_page_per_block * block + page; + +if (mxs_read_page_ecc (mtd, page_buf, curr_page) <0 ) {free (page_buf); -return -ENOMEM ; +return -EIO ; } + +if (size > (mtd->writesize - page_offset)) + sz = (mtd->writesize - page_offset); +else + sz = size; + +memcpy (dst, page_buf + page_offset, sz); + dst += sz; + size -= sz; + page_offset =0 ; + page++; } + + page =0 ; + }else { + lastblock++; } + + block++; }free (page_buf);@@ -294,6 +289,19 @@ void nand_deselect(void)u32 nand_spl_adjust_offset (u32 sector,u32 offs) { -/* Handle the offset adjust in nand_spl_load_image,*/ +unsigned int block, lastblock; + + block = sector / mtd->erasesize; + lastblock = (sector + offs) / mtd->erasesize; + +while (block <= lastblock) { +if (is_badblock (mtd, block * mtd->erasesize,1 )) { + offs += mtd->erasesize; + lastblock++; + } + + block++; + } +return offs; } --2.25.1
Signature · 11 lines
Michael Nazzareno Trimarchi Co-Founder & Chief Executive Officer M. +39 347 913 2170 [email protected] __________________________________ Amarula Solutions BV Joop Geesinkweg 125, 1114 AB, Amsterdam, NL T. +31 (0)85 111 9172 [email protected] www.amarulasolutions.com
PATCH v2 3/4 arm: mach-imx: cmd_nandbcb fix bad block handling
In reply to Michael TrimarchiThe badblock should be skipped properly in reading and writing. Fix the logic. The bcb struct is written, skipping the bad block, so we need to read using the same logic. This was tested create bad block in the area and then flash it and read it back.
---V1->V2:
- Adjust the commit message
- Add Cc Han Xu and Fabio
- move out from RFC
--- arch/arm/mach-imx/cmd_nandbcb.c | 21 +++++++-------------- 1 file changed, 7 insertions(+), 14 deletions(-)
arch/arm/mach-imx/cmd_nandbcb.c +7 -14
@@ -506,10 +506,6 @@ static int read_fcb(struct boot_config *boot_cfg, struct fcb_block *fcb,int ret =0 ; mtd = boot_cfg->mtd; -if (mtd_block_isbad (mtd, off)) { -printf ("Block %d is bad, skipped\n" , (int )CONV_TO_BLOCKS (off)); -return 1 ; - } fcb_raw_page =kzalloc (mtd->writesize + mtd->oobsize,GFP_KERNEL );if (!fcb_raw_page) {
@@ -530,7 +526,7 @@ static int read_fcb(struct boot_config *boot_cfg, struct fcb_block *fcb,else if (plat_config.misc_flags & FCB_ENCODE_BCH_40b)mxs_nand_mode_fcb_40bit (mtd); - ret =nand_read (mtd, off, &size, (u_char *)fcb); + ret =nand_read_skip_bad (mtd, off, &size,NULL , mtd->size, (u_char *)fcb);/* switch BCH back */ mxs_nand_mode_normal (mtd);
@@ -617,6 +613,7 @@ static int write_fcb(struct boot_config *boot_cfg, struct fcb_block *fcb)for (i =0 ; i < g_boot_search_count; i++) {if (mtd_block_isbad (mtd, off)) {printf ("Block %d is bad, skipped\n" , i); + off += mtd->erasesize;continue ; }
@@ -676,20 +673,15 @@ static int read_dbbt(struct boot_config *boot_cfg, struct dbbt_block *dbbt, void *dbbt_data_page, loff_t off) { size_t size; +size_t actual_size;struct mtd_info *mtd;loff_t to;int ret; mtd = boot_cfg->mtd; -if (mtd_block_isbad (mtd, off)) { -printf ("Block %d is bad, skipped\n" , - (int )CONV_TO_BLOCKS (off)); -return 1 ; - } - size =sizeof (struct dbbt_block ); - ret =nand_read (mtd, off, &size, (u_char *)dbbt); + ret =nand_read_skip_bad (mtd, off, &size, &actual_size, mtd->size, (u_char *)dbbt);printf ("NAND DBBT read from 0x%llx offset 0x%zx read: %s\n" , off, size, ret ?"ERROR" :"OK" );if (ret)
@@ -697,9 +689,9 @@ static int read_dbbt(struct boot_config *boot_cfg, struct dbbt_block *dbbt,/* dbbtpages == 0 if no bad blocks */ if (dbbt->dbbtpages >0 ) { - to = off +4 * mtd->writesize; + to = off +4 * mtd->writesize + actual_size - size; size = mtd->writesize; - ret =nand_read (mtd, to, &size, dbbt_data_page); + ret =nand_read_skip_bad (mtd, to, &size,NULL , mtd->size, dbbt_data_page);printf ("DBBT data read from 0x%llx offset 0x%zx read: %s\n" , to, size, ret ?"ERROR" :"OK" );
@@ -729,6 +721,7 @@ static int write_dbbt(struct boot_config *boot_cfg, struct dbbt_block *dbbt,if (mtd_block_isbad (mtd, off)) {printf ("Block %d is bad, skipped\n" , (int )(i +CONV_TO_BLOCKS (off))); + off += mtd->erasesize;continue ; }
Signature · 1 line
2.25.1
Re: [PATCH V2 3/4] arm: mach-imx: cmd_nandbcb fix bad block handling
Quoted 8 lines · Michael Trimarchi
The badblock should be skipped properly in reading and writing. Fix the logic. The bcb struct is written, skipping the bad block, so we need to read using the same logic. This was tested create bad block in the area and then flash it and read it back.
Cc Han Xu <[email protected]>Cc Fabio Estevam <[email protected]>Signed-off-by Michael Trimarchi <[email protected]>
Quoted 87 lines
---
V1->V2:
- Adjust the commit message
- Add Cc Han Xu and Fabio
- move out from RFC--- arch/arm/mach-imx/cmd_nandbcb.c | 21 +++++++-------------- 1 file changed, 7 insertions(+), 14 deletions(-)arch/arm/mach-imx/cmd_nandbcb.c +7
-14@@ -506,10 +506,6 @@ static int read_fcb(struct boot_config *boot_cfg, struct fcb_block *fcb,int ret =0 ; mtd = boot_cfg->mtd; -if (mtd_block_isbad (mtd, off)) { -printf ("Block %d is bad, skipped\n" , (int )CONV_TO_BLOCKS (off)); -return 1 ; - } fcb_raw_page =kzalloc (mtd->writesize + mtd->oobsize,GFP_KERNEL );if (!fcb_raw_page) {@@ -530,7 +526,7 @@ static int read_fcb(struct boot_config *boot_cfg, struct fcb_block *fcb,else if (plat_config.misc_flags & FCB_ENCODE_BCH_40b)mxs_nand_mode_fcb_40bit (mtd); - ret =nand_read (mtd, off, &size, (u_char *)fcb); + ret =nand_read_skip_bad (mtd, off, &size,NULL , mtd->size, (u_char *)fcb);/* switch BCH back */ mxs_nand_mode_normal (mtd);@@ -617,6 +613,7 @@ static int write_fcb(struct boot_config *boot_cfg, struct fcb_block *fcb)for (i =0 ; i < g_boot_search_count; i++) {if (mtd_block_isbad (mtd, off)) {printf ("Block %d is bad, skipped\n" , i); + off += mtd->erasesize;continue ; }@@ -676,20 +673,15 @@ static int read_dbbt(struct boot_config *boot_cfg, struct dbbt_block *dbbt, void *dbbt_data_page, loff_t off) { size_t size; +size_t actual_size;struct mtd_info *mtd;loff_t to;int ret; mtd = boot_cfg->mtd; -if (mtd_block_isbad (mtd, off)) { -printf ("Block %d is bad, skipped\n" , - (int )CONV_TO_BLOCKS (off)); -return 1 ; - } - size =sizeof (struct dbbt_block ); - ret =nand_read (mtd, off, &size, (u_char *)dbbt); + ret =nand_read_skip_bad (mtd, off, &size, &actual_size, mtd->size, (u_char *)dbbt);printf ("NAND DBBT read from 0x%llx offset 0x%zx read: %s\n" , off, size, ret ?"ERROR" :"OK" );if (ret)@@ -697,9 +689,9 @@ static int read_dbbt(struct boot_config *boot_cfg, struct dbbt_block *dbbt,/* dbbtpages == 0 if no bad blocks */ if (dbbt->dbbtpages >0 ) { - to = off +4 * mtd->writesize; + to = off +4 * mtd->writesize + actual_size - size; size = mtd->writesize; - ret =nand_read (mtd, to, &size, dbbt_data_page); + ret =nand_read_skip_bad (mtd, to, &size,NULL , mtd->size, dbbt_data_page);printf ("DBBT data read from 0x%llx offset 0x%zx read: %s\n" , to, size, ret ?"ERROR" :"OK" );@@ -729,6 +721,7 @@ static int write_dbbt(struct boot_config *boot_cfg, struct dbbt_block *dbbt,if (mtd_block_isbad (mtd, off)) {printf ("Block %d is bad, skipped\n" , (int )(i +CONV_TO_BLOCKS (off))); + off += mtd->erasesize;continue ; }Signature · 1 line
2.25.1
Re: [PATCH V2 3/4] arm: mach-imx: cmd_nandbcb fix bad block handling
Hi Tim
Do you have an alternative board to test?
Michael
Quoted 100 lines · Han Xu
Quoted 8 lines · Michael Trimarchi
The badblock should be skipped properly in reading and writing. Fix the logic. The bcb struct is written, skipping the bad block, so we need to read using the same logic. This was tested create bad block in the area and then flash it and read it back.
Cc Han Xu <[email protected]>Cc Fabio Estevam <[email protected]>Signed-off-by Michael Trimarchi <[email protected]>Acked-by Han Xu <[email protected]>Quoted 87 lines
---
V1->V2:
- Adjust the commit message
- Add Cc Han Xu and Fabio
- move out from RFC--- arch/arm/mach-imx/cmd_nandbcb.c | 21 +++++++-------------- 1 file changed, 7 insertions(+), 14 deletions(-)arch/arm/mach-imx/cmd_nandbcb.c +7
-15@@ -506,10 +506,6 @@ static int read_fcb(struct boot_config *boot_cfg, struct fcb_block *fcb,int ret =0 ; mtd = boot_cfg->mtd; -if (mtd_block_isbad (mtd, off)) { -printf ("Block %d is bad, skipped\n" , (int )CONV_TO_BLOCKS (off)); -return 1 ; - } fcb_raw_page =kzalloc (mtd->writesize + mtd->oobsize,GFP_KERNEL );if (!fcb_raw_page) {@@ -530,7 +526,7 @@ static int read_fcb(struct boot_config *boot_cfg, struct fcb_block *fcb,else if (plat_config.misc_flags & FCB_ENCODE_BCH_40b)mxs_nand_mode_fcb_40bit (mtd); - ret =nand_read (mtd, off, &size, (u_char *)fcb); + ret =nand_read_skip_bad (mtd, off, &size,NULL , mtd->size, (u_char *)fcb);/* switch BCH back */ mxs_nand_mode_normal (mtd);@@ -617,6 +613,7 @@ static int write_fcb(struct boot_config *boot_cfg, struct fcb_block *fcb)for (i =0 ; i < g_boot_search_count; i++) {if (mtd_block_isbad (mtd, off)) {printf ("Block %d is bad, skipped\n" , i); + off += mtd->erasesize;continue ; }@@ -676,20 +673,15 @@ static int read_dbbt(struct boot_config *boot_cfg, struct dbbt_block *dbbt, void *dbbt_data_page, loff_t off) { size_t size; +size_t actual_size;struct mtd_info *mtd;loff_t to;int ret; mtd = boot_cfg->mtd; -if (mtd_block_isbad (mtd, off)) { -printf ("Block %d is bad, skipped\n" , - (int )CONV_TO_BLOCKS (off)); -return 1 ; - } - size =sizeof (struct dbbt_block ); - ret =nand_read (mtd, off, &size, (u_char *)dbbt); + ret =nand_read_skip_bad (mtd, off, &size, &actual_size, mtd->size, (u_char *)dbbt);printf ("NAND DBBT read from 0x%llx offset 0x%zx read: %s\n" , off, size, ret ?"ERROR" :"OK" );if (ret)@@ -697,9 +689,9 @@ static int read_dbbt(struct boot_config *boot_cfg, struct dbbt_block *dbbt,/* dbbtpages == 0 if no bad blocks */ if (dbbt->dbbtpages >0 ) { - to = off +4 * mtd->writesize; + to = off +4 * mtd->writesize + actual_size - size; size = mtd->writesize; - ret =nand_read (mtd, to, &size, dbbt_data_page); + ret =nand_read_skip_bad (mtd, to, &size,NULL , mtd->size, dbbt_data_page);printf ("DBBT data read from 0x%llx offset 0x%zx read: %s\n" , to, size, ret ?"ERROR" :"OK" );@@ -729,6 +721,7 @@ static int write_dbbt(struct boot_config *boot_cfg, struct dbbt_block *dbbt,if (mtd_block_isbad (mtd, off)) {printf ("Block %d is bad, skipped\n" , (int )(i +CONV_TO_BLOCKS (off))); + off += mtd->erasesize;continue ; } --2.25.1
Signature · 11 lines
Michael Nazzareno Trimarchi Co-Founder & Chief Executive Officer M. +39 347 913 2170 [email protected] __________________________________ Amarula Solutions BV Joop Geesinkweg 125, 1114 AB, Amsterdam, NL T. +31 (0)85 111 9172 [email protected] www.amarulasolutions.com
Re: [PATCH V2 3/4] arm: mach-imx: cmd_nandbcb fix bad block handling
Quoted 5 lines · Michael Nazzareno Trimarchi
Hi Tim
Do you have an alternative board to test?
Michael,
Yes sorry... I tested the series and had the reply in my draft folder. I sent it just now in response to the cover letter.
Best regards,
Tim
Quoted 118 lines
Michael
Quoted 97 lines · Han Xu
Quoted 8 lines · Michael Trimarchi
The badblock should be skipped properly in reading and writing. Fix the logic. The bcb struct is written, skipping the bad block, so we need to read using the same logic. This was tested create bad block in the area and then flash it and read it back.
Cc Han Xu <[email protected]>Cc Fabio Estevam <[email protected]>Signed-off-by Michael Trimarchi <[email protected]>Acked-by Han Xu <[email protected]>Quoted 84 lines
---
V1->V2:
- Adjust the commit message
- Add Cc Han Xu and Fabio
- move out from RFC--- arch/arm/mach-imx/cmd_nandbcb.c | 21 +++++++-------------- 1 file changed, 7 insertions(+), 14 deletions(-)arch/arm/mach-imx/cmd_nandbcb.c +7
-14@@ -506,10 +506,6 @@ static int read_fcb(struct boot_config *boot_cfg, struct fcb_block *fcb,int ret =0 ; mtd = boot_cfg->mtd; -if (mtd_block_isbad (mtd, off)) { -printf ("Block %d is bad, skipped\n" , (int )CONV_TO_BLOCKS (off)); -return 1 ; - } fcb_raw_page =kzalloc (mtd->writesize + mtd->oobsize,GFP_KERNEL );if (!fcb_raw_page) {@@ -530,7 +526,7 @@ static int read_fcb(struct boot_config *boot_cfg, struct fcb_block *fcb,else if (plat_config.misc_flags & FCB_ENCODE_BCH_40b)mxs_nand_mode_fcb_40bit (mtd); - ret =nand_read (mtd, off, &size, (u_char *)fcb); + ret =nand_read_skip_bad (mtd, off, &size,NULL , mtd->size, (u_char *)fcb);/* switch BCH back */ mxs_nand_mode_normal (mtd);@@ -617,6 +613,7 @@ static int write_fcb(struct boot_config *boot_cfg, struct fcb_block *fcb)for (i =0 ; i < g_boot_search_count; i++) {if (mtd_block_isbad (mtd, off)) {printf ("Block %d is bad, skipped\n" , i); + off += mtd->erasesize;continue ; }@@ -676,20 +673,15 @@ static int read_dbbt(struct boot_config *boot_cfg, struct dbbt_block *dbbt, void *dbbt_data_page, loff_t off) { size_t size; +size_t actual_size;struct mtd_info *mtd;loff_t to;int ret; mtd = boot_cfg->mtd; -if (mtd_block_isbad (mtd, off)) { -printf ("Block %d is bad, skipped\n" , - (int )CONV_TO_BLOCKS (off)); -return 1 ; - } - size =sizeof (struct dbbt_block ); - ret =nand_read (mtd, off, &size, (u_char *)dbbt); + ret =nand_read_skip_bad (mtd, off, &size, &actual_size, mtd->size, (u_char *)dbbt);printf ("NAND DBBT read from 0x%llx offset 0x%zx read: %s\n" , off, size, ret ?"ERROR" :"OK" );if (ret)@@ -697,9 +689,9 @@ static int read_dbbt(struct boot_config *boot_cfg, struct dbbt_block *dbbt,/* dbbtpages == 0 if no bad blocks */ if (dbbt->dbbtpages >0 ) { - to = off +4 * mtd->writesize; + to = off +4 * mtd->writesize + actual_size - size; size = mtd->writesize; - ret =nand_read (mtd, to, &size, dbbt_data_page); + ret =nand_read_skip_bad (mtd, to, &size,NULL , mtd->size, dbbt_data_page);printf ("DBBT data read from 0x%llx offset 0x%zx read: %s\n" , to, size, ret ?"ERROR" :"OK" );@@ -729,6 +721,7 @@ static int write_dbbt(struct boot_config *boot_cfg, struct dbbt_block *dbbt,if (mtd_block_isbad (mtd, off)) {printf ("Block %d is bad, skipped\n" , (int )(i +CONV_TO_BLOCKS (off))); + off += mtd->erasesize;continue ; }Quoted 3 lines
Quoted 3 lines
--
2.25.1--
Michael Nazzareno Trimarchi
Co-Founder & Chief Executive Officer
M. +39 347 913 2170
[email protected]
__________________________________Amarula Solutions BV
Joop Geesinkweg 125, 1114 AB, Amsterdam, NL
T. +31 (0)85 111 9172
[email protected]
www.amarulasolutions.com
PATCH v2 4/4 spl: spl_nand: Fix bad block handling in fitImage
In reply to Michael TrimarchiIf the fitImage has some bad block in fit image area, the offset must be recalulcated. This should be done always. After implementing it in mxs now is possible to call the function even for that platform.
---V1->V2:
- move out from RFC
--- common/spl/spl_nand.c | 5 +---- 1 file changed, 1 insertion(+), 4 deletions(-)
common/spl/spl_nand.c +1 -4
@@ -43,15 +43,12 @@ static ulong spl_nand_fit_read(struct spl_load_info *load, ulong offs, ulong size, void *dst) { int err; -#ifdef CONFIG_SYS_NAND_BLOCK_SIZE ulong sector; sector = *(int *)load->priv; - offs = sector +nand_spl_adjust_offset (sector, offs - sector); -#else offs *= load->bl_len; size *= load->bl_len; -#endif + offs = sector +nand_spl_adjust_offset (sector, offs - sector); err =nand_spl_load_image (offs, size, dst);if (err)return 0 ;
Signature · 1 line
2.25.1
Re: [PATCH 4/4] spl: spl_nand: Fix bad block handling in fitImage
Quoted 7 lines · Michael Trimarchi
If the fitImage has some bad block in fit image area, the offset must be recalulcated. This should be done always. After implementing it in mxs now is possible to call the function even for that platform.
Cc Fabio Estevam <[email protected]>Signed-off-by Michael Trimarchi <[email protected]>
Signature · 1 line
Tom
Re: [PATCH 4/4] spl: spl_nand: Fix bad block handling in fitImage
In reply to Michael TrimarchiQuoted 7 lines · Michael Trimarchi
If the fitImage has some bad block in fit image area, the offset must be recalulcated. This should be done always. After implementing it in mxs now is possible to call the function even for that platform.
Cc Fabio Estevam <[email protected]>Signed-off-by Michael Trimarchi <[email protected]>
Quoted 31 lines
---
V1->V2:
- move out from RFC--- common/spl/spl_nand.c | 5 +---- 1 file changed, 1 insertion(+), 4 deletions(-)common/spl/spl_nand.c +1
-4@@ -43,15 +43,12 @@ static ulong spl_nand_fit_read(struct spl_load_info *load, ulong offs, ulong size, void *dst) { int err; -#ifdef CONFIG_SYS_NAND_BLOCK_SIZE ulong sector; sector = *(int *)load->priv; - offs = sector +nand_spl_adjust_offset (sector, offs - sector); -#else offs *= load->bl_len; size *= load->bl_len; -#endif + offs = sector +nand_spl_adjust_offset (sector, offs - sector); err =nand_spl_load_image (offs, size, dst);if (err)return 0 ;Signature · 1 line
2.25.1
Quoted 18 lines · Michael Trimarchi
Those patches come after some testing of failing in factory on some unit. We found out that the bootrom imx loader was not able to handling badblock. This can be a limit of the implementation right now in imx8mn. Anyway not all the imx platform has the support of this loader. I found some problems on the implementation so I have fixed it up according the experience of Sitara. I tested only using a Fit Image as a flash container. This version add in the series the fix of
cmd_nandbcband the fix ofspl_nandload. I can imagine that a lot of boards and users are affected. I have started to backport this changes in some older uboot and adapt itMichael Trimarchi (4):
nand: raw:mxs_nand: Fix specific hook registration
mtd: nand:mxs_nand_spl: Fix bad block skipping
arm: mach-imx:cmd_nandbcbfix bad block handling
spl:spl_nand: Fix bad block handling in fitImage
Quoted 9 lines
arch/arm/mach-imx/cmd_nandbcb.c | 21 +++---- common/spl/spl_nand.c | 5 +- drivers/mtd/nand/raw/mxs_nand.c | 32 +++++----- drivers/mtd/nand/raw/mxs_nand_spl.c | 90 ++++++++++++++++------------- 4 files changed, 73 insertions(+), 75 deletions(-)--
2.25.1
Tested By: Tim Harvey <[email protected]>
#gwventana_nand_defconfig (imx6qdl NAND)
Best Regards,
Tim