* Re: [PATCH] Fix OOB_READ bug where hardware FIFO is not drained completely. Fix potential bug for non-ECC operations.
2017-12-12 13:20 ` Boris Brezillon
@ 2017-12-12 13:33 ` Sean Nyekjær
2017-12-18 7:44 ` Sean Nyekjær
2017-12-12 13:59 ` Greg Cook
2017-12-12 16:29 ` Ezequiel Garcia
2 siblings, 1 reply; 9+ messages in thread
From: Sean Nyekjær @ 2017-12-12 13:33 UTC (permalink / raw)
To: Boris Brezillon, Greg Cook
Cc: ezequiel.garcia, linux-mtd, miquel.raynal, Kasper Revsbech (KREV)
Hi Boris
> Hi Greg,
>
> Could you at least Cc the NAND maintainer when you submit something
> related to NAND!
>
> On Tue, 12 Dec 2017 16:38:26 +0800
> Greg Cook <greg@morpheus.ws> wrote:
>
> Please add a commit message explaining what your fixing.
>
>> Signed-off-by: Greg Cook <greg@morpheus.ws>
>> ---
>> drivers/mtd/nand/pxa3xx_nand.c | 9 +++++++--
>> 1 file changed, 7 insertions(+), 2 deletions(-)
>>
>> diff --git a/drivers/mtd/nand/pxa3xx_nand.c b/drivers/mtd/nand/pxa3xx_nand.c
>> index 021374f..cfa8c71 100644
>> --- a/drivers/mtd/nand/pxa3xx_nand.c
>> +++ b/drivers/mtd/nand/pxa3xx_nand.c
>> @@ -677,7 +677,7 @@ static void disable_int(struct pxa3xx_nand_info *info, uint32_t int_mask)
>>
>> static void drain_fifo(struct pxa3xx_nand_info *info, void *data, int len)
>> {
>> - if (info->ecc_bch) {
>> + if (info->use_ecc && info->ecc_bch) {
>> u32 val;
>> int ret;
>>
>> @@ -1023,7 +1023,12 @@ static int prepare_set_command(struct pxa3xx_nand_info *info, int command,
>>
>> if (info->cur_chunk < info->nfullchunks) {
>> info->step_chunk_size = info->chunk_size;
>> - info->step_spare_size = info->spare_size;
>> + if (info->use_ecc) {
>> + info->step_spare_size = info->spare_size;
>> + } else {
>> + info->step_spare_size = info->spare_size
>> + + info->ecc_size;
>> + }
> Curly braces are unneeded, and I'd prefer to have the + operator at the
> end of the line. Or it could be written like that:
>
> info->step_spare_size = info->spare_size;
> if (!info->use_ecc)
> info->step_spare_size += info->ecc_size;
>
> Anyway, I'm still not convince this is the appropriate fix. As
> mentioned in my previous, I wonder why we're not activating ECC when
> reading OOB bytes. There's a good reason for having 2 different hooks
> to read/write OOBs: one is doing it in raw mode (with the ECC engine
> disabled), the other one is doing it with the ECC engine enabled.
> This driver not only omit raw accessors (which is already a bad news
> for everyone that needs to debug the driver) but it also do things
> differently for OOB and page access: OOB bytes are read in raw mode
> while pages are always accessed with the ECC engine enabled.
>
> OOB accesses are broken and should be fixed, I think we all agree on
> that, but we should fix it correctly, don't you think?
> And by correctly, I mean we should try to make things consistent at
> least (read OOB bytes with the ECC engine enabled).
>
>> } else {
>> info->step_chunk_size = info->last_chunk_size;
>> info->step_spare_size = info->last_spare_size;
> Can you try with the below patch and see if it fixes the timeout issue?
>
> --->8---
> diff --git a/drivers/mtd/nand/pxa3xx_nand.c b/drivers/mtd/nand/pxa3xx_nand.c
> index 021374fe59dc..d1979c7dbe7e 100644
> --- a/drivers/mtd/nand/pxa3xx_nand.c
> +++ b/drivers/mtd/nand/pxa3xx_nand.c
> @@ -961,6 +961,7 @@ static void prepare_start_command(struct pxa3xx_nand_info *info, int command)
>
> switch (command) {
> case NAND_CMD_READ0:
> + case NAND_CMD_READOOB:
> case NAND_CMD_PAGEPROG:
> info->use_ecc = 1;
> break;
This fixes the timeout issue :-) Also without bbt...
root@output-module:~# nanddump -oa /dev/mtd1
ECC failed: 0
ECC corrected: 0
Number of bad blocks: 0
Number of bbt blocks: 8
Block size 131072, page size 2048, OOB size 64
Dumping data starting at 0x00000000 and ending at 0x0ff00000...
UBIXSUBI
/Sean
^ permalink raw reply [flat|nested] 9+ messages in thread* Re: [PATCH] Fix OOB_READ bug where hardware FIFO is not drained completely. Fix potential bug for non-ECC operations.
2017-12-12 13:33 ` Sean Nyekjær
@ 2017-12-18 7:44 ` Sean Nyekjær
0 siblings, 0 replies; 9+ messages in thread
From: Sean Nyekjær @ 2017-12-18 7:44 UTC (permalink / raw)
To: Boris Brezillon
Cc: Greg Cook, ezequiel.garcia, linux-mtd, miquel.raynal,
Kasper Revsbech (KREV)
Hi Boris
>>
>> Can you try with the below patch and see if it fixes the timeout issue?
>>
>> --->8---
>> diff --git a/drivers/mtd/nand/pxa3xx_nand.c
>> b/drivers/mtd/nand/pxa3xx_nand.c
>> index 021374fe59dc..d1979c7dbe7e 100644
>> --- a/drivers/mtd/nand/pxa3xx_nand.c
>> +++ b/drivers/mtd/nand/pxa3xx_nand.c
>> @@ -961,6 +961,7 @@ static void prepare_start_command(struct
>> pxa3xx_nand_info *info, int command)
>> switch (command) {
>> case NAND_CMD_READ0:
>> + case NAND_CMD_READOOB:
>> case NAND_CMD_PAGEPROG:
>> info->use_ecc = 1;
>> break;
> This fixes the timeout issue :-) Also without bbt...
>
> root@output-module:~# nanddump -oa /dev/mtd1
> ECC failed: 0
> ECC corrected: 0
> Number of bad blocks: 0
> Number of bbt blocks: 8
> Block size 131072, page size 2048, OOB size 64
> Dumping data starting at 0x00000000 and ending at 0x0ff00000...
> UBIXSUBI
>
Is this considered a valid patch for our 4.14 LTS kernel, pxa3xx driver?
Or is it only masking some other problems?
/Sean
^ permalink raw reply [flat|nested] 9+ messages in thread
* Re: [PATCH] Fix OOB_READ bug where hardware FIFO is not drained completely. Fix potential bug for non-ECC operations.
2017-12-12 13:20 ` Boris Brezillon
2017-12-12 13:33 ` Sean Nyekjær
@ 2017-12-12 13:59 ` Greg Cook
2017-12-12 16:29 ` Ezequiel Garcia
2 siblings, 0 replies; 9+ messages in thread
From: Greg Cook @ 2017-12-12 13:59 UTC (permalink / raw)
To: Boris Brezillon
Cc: Ezequiel Garcia, linux-mtd, Sean Nyekjær, Miquel RAYNAL
On 12 December 2017 at 21:20, Boris Brezillon
<boris.brezillon@free-electrons.com> wrote:
> Hi Greg,
>
> Could you at least Cc the NAND maintainer when you submit something
> related to NAND!
It's my first time submitting a patch and I was following Sean's
instructions. I see now the recommendation is to use
'./scripts/get_maintainer.pl --norolestats 000*', which would have
CC'd in a LOT more people.
>
> On Tue, 12 Dec 2017 16:38:26 +0800
> Greg Cook <greg@morpheus.ws> wrote:
>
> Please add a commit message explaining what your fixing.
>
I have just resent it with a better subject line and a description. No
doubt this is also wrong ;)
>> Signed-off-by: Greg Cook <greg@morpheus.ws>
>> ---
>> drivers/mtd/nand/pxa3xx_nand.c | 9 +++++++--
>> 1 file changed, 7 insertions(+), 2 deletions(-)
>>
>> diff --git a/drivers/mtd/nand/pxa3xx_nand.c b/drivers/mtd/nand/pxa3xx_nand.c
>> index 021374f..cfa8c71 100644
>> --- a/drivers/mtd/nand/pxa3xx_nand.c
>> +++ b/drivers/mtd/nand/pxa3xx_nand.c
>> @@ -677,7 +677,7 @@ static void disable_int(struct pxa3xx_nand_info *info, uint32_t int_mask)
>>
>> static void drain_fifo(struct pxa3xx_nand_info *info, void *data, int len)
>> {
>> - if (info->ecc_bch) {
>> + if (info->use_ecc && info->ecc_bch) {
>> u32 val;
>> int ret;
>>
>> @@ -1023,7 +1023,12 @@ static int prepare_set_command(struct pxa3xx_nand_info *info, int command,
>>
>> if (info->cur_chunk < info->nfullchunks) {
>> info->step_chunk_size = info->chunk_size;
>> - info->step_spare_size = info->spare_size;
>> + if (info->use_ecc) {
>> + info->step_spare_size = info->spare_size;
>> + } else {
>> + info->step_spare_size = info->spare_size
>> + + info->ecc_size;
>> + }
>
> Curly braces are unneeded, ...
I disagree strongly because it costs a couple of keystrokes to add
them and I think it improves readability. *But* I see it is in the
Kernel coding style guidelines, so I grudgingly accept. I'm not going
to promise it won't happen again though...
> ...and I'd prefer to have the + operator at the
> end of the line. Or it could be written like that:
>
> info->step_spare_size = info->spare_size;
> if (!info->use_ecc)
> info->step_spare_size += info->ecc_size;
Agreed. This is much nicer.
>
> Anyway, I'm still not convince this is the appropriate fix. As
> mentioned in my previous, I wonder why we're not activating ECC when
> reading OOB bytes. There's a good reason for having 2 different hooks
> to read/write OOBs: one is doing it in raw mode (with the ECC engine
> disabled), the other one is doing it with the ECC engine enabled.
> This driver not only omit raw accessors (which is already a bad news
> for everyone that needs to debug the driver) but it also do things
> differently for OOB and page access: OOB bytes are read in raw mode
> while pages are always accessed with the ECC engine enabled.
>
> OOB accesses are broken and should be fixed, I think we all agree on
> that, but we should fix it correctly, don't you think?
> And by correctly, I mean we should try to make things consistent at
> least (read OOB bytes with the ECC engine enabled).
All fair points. I assumed there was some deliberate reason why
READOOB was being done without ECC, so I made a point of not changing
that part. For an extra data-point, I should be able to try your
suggested fix on our 2nd-round prototypes in the new year (maybe
sooner, on one of my 1st round protos if I get time),
>
>> } else {
>> info->step_chunk_size = info->last_chunk_size;
>> info->step_spare_size = info->last_spare_size;
>
> Can you try with the below patch and see if it fixes the timeout issue?
>
> --->8---
> diff --git a/drivers/mtd/nand/pxa3xx_nand.c b/drivers/mtd/nand/pxa3xx_nand.c
> index 021374fe59dc..d1979c7dbe7e 100644
> --- a/drivers/mtd/nand/pxa3xx_nand.c
> +++ b/drivers/mtd/nand/pxa3xx_nand.c
> @@ -961,6 +961,7 @@ static void prepare_start_command(struct pxa3xx_nand_info *info, int command)
>
> switch (command) {
> case NAND_CMD_READ0:
> + case NAND_CMD_READOOB:
> case NAND_CMD_PAGEPROG:
> info->use_ecc = 1;
> break;
^ permalink raw reply [flat|nested] 9+ messages in thread* Re: [PATCH] Fix OOB_READ bug where hardware FIFO is not drained completely. Fix potential bug for non-ECC operations.
2017-12-12 13:20 ` Boris Brezillon
2017-12-12 13:33 ` Sean Nyekjær
2017-12-12 13:59 ` Greg Cook
@ 2017-12-12 16:29 ` Ezequiel Garcia
2017-12-12 16:39 ` Boris Brezillon
2 siblings, 1 reply; 9+ messages in thread
From: Ezequiel Garcia @ 2017-12-12 16:29 UTC (permalink / raw)
To: Boris Brezillon
Cc: Greg Cook, Miquèl Raynal, linux-mtd@lists.infradead.org,
Ezequiel Garcia, Sean Nyekjær
On 12 December 2017 at 10:20, Boris Brezillon
<boris.brezillon@free-electrons.com> wrote:
> Hi Greg,
>
> Could you at least Cc the NAND maintainer when you submit something
> related to NAND!
>
> On Tue, 12 Dec 2017 16:38:26 +0800
> Greg Cook <greg@morpheus.ws> wrote:
>
> Please add a commit message explaining what your fixing.
>
>> Signed-off-by: Greg Cook <greg@morpheus.ws>
>> ---
>> drivers/mtd/nand/pxa3xx_nand.c | 9 +++++++--
>> 1 file changed, 7 insertions(+), 2 deletions(-)
>>
>> diff --git a/drivers/mtd/nand/pxa3xx_nand.c b/drivers/mtd/nand/pxa3xx_nand.c
>> index 021374f..cfa8c71 100644
>> --- a/drivers/mtd/nand/pxa3xx_nand.c
>> +++ b/drivers/mtd/nand/pxa3xx_nand.c
>> @@ -677,7 +677,7 @@ static void disable_int(struct pxa3xx_nand_info *info, uint32_t int_mask)
>>
>> static void drain_fifo(struct pxa3xx_nand_info *info, void *data, int len)
>> {
>> - if (info->ecc_bch) {
>> + if (info->use_ecc && info->ecc_bch) {
>> u32 val;
>> int ret;
>>
>> @@ -1023,7 +1023,12 @@ static int prepare_set_command(struct pxa3xx_nand_info *info, int command,
>>
>> if (info->cur_chunk < info->nfullchunks) {
>> info->step_chunk_size = info->chunk_size;
>> - info->step_spare_size = info->spare_size;
>> + if (info->use_ecc) {
>> + info->step_spare_size = info->spare_size;
>> + } else {
>> + info->step_spare_size = info->spare_size
>> + + info->ecc_size;
>> + }
>
> Curly braces are unneeded, and I'd prefer to have the + operator at the
> end of the line. Or it could be written like that:
>
> info->step_spare_size = info->spare_size;
> if (!info->use_ecc)
> info->step_spare_size += info->ecc_size;
>
> Anyway, I'm still not convince this is the appropriate fix. As
> mentioned in my previous, I wonder why we're not activating ECC when
> reading OOB bytes. There's a good reason for having 2 different hooks
> to read/write OOBs: one is doing it in raw mode (with the ECC engine
> disabled), the other one is doing it with the ECC engine enabled.
> This driver not only omit raw accessors (which is already a bad news
> for everyone that needs to debug the driver) but it also do things
> differently for OOB and page access: OOB bytes are read in raw mode
> while pages are always accessed with the ECC engine enabled.
>
> OOB accesses are broken and should be fixed, I think we all agree on
> that, but we should fix it correctly, don't you think?
> And by correctly, I mean we should try to make things consistent at
> least (read OOB bytes with the ECC engine enabled).
>
>> } else {
>> info->step_chunk_size = info->last_chunk_size;
>> info->step_spare_size = info->last_spare_size;
>
> Can you try with the below patch and see if it fixes the timeout issue?
>
Well, the patch may very well fix the timeout, but it's not really
reading any OOB.
Like I said in my previous mail, accessing the OOB doesn't
work when ECC is enabled. This is specified in the specs,
copying from the Armada 370 spec:
"""
14.4.1
Error Checking and Correction (ECC)
[..] When ECC is enabled, the number of spare area bytes required for use by
ECC are shown in the middle column of Table 30. These bytes are not
available for use by the
system software and cannot be written to or read from.
"""
I seriously doubt you are reading valid data with this change.
--
Ezequiel García, VanguardiaSur
www.vanguardiasur.com.ar
^ permalink raw reply [flat|nested] 9+ messages in thread* Re: [PATCH] Fix OOB_READ bug where hardware FIFO is not drained completely. Fix potential bug for non-ECC operations.
2017-12-12 16:29 ` Ezequiel Garcia
@ 2017-12-12 16:39 ` Boris Brezillon
2017-12-12 17:05 ` Ezequiel Garcia
0 siblings, 1 reply; 9+ messages in thread
From: Boris Brezillon @ 2017-12-12 16:39 UTC (permalink / raw)
To: Ezequiel Garcia
Cc: Greg Cook, Miquèl Raynal, linux-mtd@lists.infradead.org,
Ezequiel Garcia, Sean Nyekjær
On Tue, 12 Dec 2017 13:29:11 -0300
Ezequiel Garcia <ezequiel@vanguardiasur.com.ar> wrote:
> On 12 December 2017 at 10:20, Boris Brezillon
> <boris.brezillon@free-electrons.com> wrote:
> > Hi Greg,
> >
> > Could you at least Cc the NAND maintainer when you submit something
> > related to NAND!
> >
> > On Tue, 12 Dec 2017 16:38:26 +0800
> > Greg Cook <greg@morpheus.ws> wrote:
> >
> > Please add a commit message explaining what your fixing.
> >
> >> Signed-off-by: Greg Cook <greg@morpheus.ws>
> >> ---
> >> drivers/mtd/nand/pxa3xx_nand.c | 9 +++++++--
> >> 1 file changed, 7 insertions(+), 2 deletions(-)
> >>
> >> diff --git a/drivers/mtd/nand/pxa3xx_nand.c b/drivers/mtd/nand/pxa3xx_nand.c
> >> index 021374f..cfa8c71 100644
> >> --- a/drivers/mtd/nand/pxa3xx_nand.c
> >> +++ b/drivers/mtd/nand/pxa3xx_nand.c
> >> @@ -677,7 +677,7 @@ static void disable_int(struct pxa3xx_nand_info *info, uint32_t int_mask)
> >>
> >> static void drain_fifo(struct pxa3xx_nand_info *info, void *data, int len)
> >> {
> >> - if (info->ecc_bch) {
> >> + if (info->use_ecc && info->ecc_bch) {
> >> u32 val;
> >> int ret;
> >>
> >> @@ -1023,7 +1023,12 @@ static int prepare_set_command(struct pxa3xx_nand_info *info, int command,
> >>
> >> if (info->cur_chunk < info->nfullchunks) {
> >> info->step_chunk_size = info->chunk_size;
> >> - info->step_spare_size = info->spare_size;
> >> + if (info->use_ecc) {
> >> + info->step_spare_size = info->spare_size;
> >> + } else {
> >> + info->step_spare_size = info->spare_size
> >> + + info->ecc_size;
> >> + }
> >
> > Curly braces are unneeded, and I'd prefer to have the + operator at the
> > end of the line. Or it could be written like that:
> >
> > info->step_spare_size = info->spare_size;
> > if (!info->use_ecc)
> > info->step_spare_size += info->ecc_size;
> >
> > Anyway, I'm still not convince this is the appropriate fix. As
> > mentioned in my previous, I wonder why we're not activating ECC when
> > reading OOB bytes. There's a good reason for having 2 different hooks
> > to read/write OOBs: one is doing it in raw mode (with the ECC engine
> > disabled), the other one is doing it with the ECC engine enabled.
> > This driver not only omit raw accessors (which is already a bad news
> > for everyone that needs to debug the driver) but it also do things
> > differently for OOB and page access: OOB bytes are read in raw mode
> > while pages are always accessed with the ECC engine enabled.
> >
> > OOB accesses are broken and should be fixed, I think we all agree on
> > that, but we should fix it correctly, don't you think?
> > And by correctly, I mean we should try to make things consistent at
> > least (read OOB bytes with the ECC engine enabled).
> >
> >> } else {
> >> info->step_chunk_size = info->last_chunk_size;
> >> info->step_spare_size = info->last_spare_size;
> >
> > Can you try with the below patch and see if it fixes the timeout issue?
> >
>
> Well, the patch may very well fix the timeout, but it's not really
> reading any OOB.
>
> Like I said in my previous mail, accessing the OOB doesn't
> work when ECC is enabled. This is specified in the specs,
> copying from the Armada 370 spec:
>
> """
> 14.4.1
> Error Checking and Correction (ECC)
> [..] When ECC is enabled, the number of spare area bytes required for use by
> ECC are shown in the middle column of Table 30. These bytes are not
> available for use by the
> system software and cannot be written to or read from.
> """
>
> I seriously doubt you are reading valid data with this change.
That's not true. Reading ECC bytes is impossible when the ECC engine
is enable, reading free OOB bytes is possible, and most of the time the
caller doesn't care about ECC bytes when asking for a non-raw OOB
access.
Actually, it's even better to read the free OOB bytes with ECC enabled
because they are ECC protected, so if you have bitflips in the free
OOB section they will be fixed or reported as uncorrectable.
^ permalink raw reply [flat|nested] 9+ messages in thread* Re: [PATCH] Fix OOB_READ bug where hardware FIFO is not drained completely. Fix potential bug for non-ECC operations.
2017-12-12 16:39 ` Boris Brezillon
@ 2017-12-12 17:05 ` Ezequiel Garcia
0 siblings, 0 replies; 9+ messages in thread
From: Ezequiel Garcia @ 2017-12-12 17:05 UTC (permalink / raw)
To: Boris Brezillon
Cc: Greg Cook, Miquèl Raynal, linux-mtd@lists.infradead.org,
Ezequiel Garcia, Sean Nyekjær
On 12 December 2017 at 13:39, Boris Brezillon
<boris.brezillon@free-electrons.com> wrote:
> On Tue, 12 Dec 2017 13:29:11 -0300
> Ezequiel Garcia <ezequiel@vanguardiasur.com.ar> wrote:
>
>> On 12 December 2017 at 10:20, Boris Brezillon
>> <boris.brezillon@free-electrons.com> wrote:
>> > Hi Greg,
>> >
>> > Could you at least Cc the NAND maintainer when you submit something
>> > related to NAND!
>> >
>> > On Tue, 12 Dec 2017 16:38:26 +0800
>> > Greg Cook <greg@morpheus.ws> wrote:
>> >
>> > Please add a commit message explaining what your fixing.
>> >
>> >> Signed-off-by: Greg Cook <greg@morpheus.ws>
>> >> ---
>> >> drivers/mtd/nand/pxa3xx_nand.c | 9 +++++++--
>> >> 1 file changed, 7 insertions(+), 2 deletions(-)
>> >>
>> >> diff --git a/drivers/mtd/nand/pxa3xx_nand.c b/drivers/mtd/nand/pxa3xx_nand.c
>> >> index 021374f..cfa8c71 100644
>> >> --- a/drivers/mtd/nand/pxa3xx_nand.c
>> >> +++ b/drivers/mtd/nand/pxa3xx_nand.c
>> >> @@ -677,7 +677,7 @@ static void disable_int(struct pxa3xx_nand_info *info, uint32_t int_mask)
>> >>
>> >> static void drain_fifo(struct pxa3xx_nand_info *info, void *data, int len)
>> >> {
>> >> - if (info->ecc_bch) {
>> >> + if (info->use_ecc && info->ecc_bch) {
>> >> u32 val;
>> >> int ret;
>> >>
>> >> @@ -1023,7 +1023,12 @@ static int prepare_set_command(struct pxa3xx_nand_info *info, int command,
>> >>
>> >> if (info->cur_chunk < info->nfullchunks) {
>> >> info->step_chunk_size = info->chunk_size;
>> >> - info->step_spare_size = info->spare_size;
>> >> + if (info->use_ecc) {
>> >> + info->step_spare_size = info->spare_size;
>> >> + } else {
>> >> + info->step_spare_size = info->spare_size
>> >> + + info->ecc_size;
>> >> + }
>> >
>> > Curly braces are unneeded, and I'd prefer to have the + operator at the
>> > end of the line. Or it could be written like that:
>> >
>> > info->step_spare_size = info->spare_size;
>> > if (!info->use_ecc)
>> > info->step_spare_size += info->ecc_size;
>> >
>> > Anyway, I'm still not convince this is the appropriate fix. As
>> > mentioned in my previous, I wonder why we're not activating ECC when
>> > reading OOB bytes. There's a good reason for having 2 different hooks
>> > to read/write OOBs: one is doing it in raw mode (with the ECC engine
>> > disabled), the other one is doing it with the ECC engine enabled.
>> > This driver not only omit raw accessors (which is already a bad news
>> > for everyone that needs to debug the driver) but it also do things
>> > differently for OOB and page access: OOB bytes are read in raw mode
>> > while pages are always accessed with the ECC engine enabled.
>> >
>> > OOB accesses are broken and should be fixed, I think we all agree on
>> > that, but we should fix it correctly, don't you think?
>> > And by correctly, I mean we should try to make things consistent at
>> > least (read OOB bytes with the ECC engine enabled).
>> >
>> >> } else {
>> >> info->step_chunk_size = info->last_chunk_size;
>> >> info->step_spare_size = info->last_spare_size;
>> >
>> > Can you try with the below patch and see if it fixes the timeout issue?
>> >
>>
>> Well, the patch may very well fix the timeout, but it's not really
>> reading any OOB.
>>
>> Like I said in my previous mail, accessing the OOB doesn't
>> work when ECC is enabled. This is specified in the specs,
>> copying from the Armada 370 spec:
>>
>> """
>> 14.4.1
>> Error Checking and Correction (ECC)
>> [..] When ECC is enabled, the number of spare area bytes required for use by
>> ECC are shown in the middle column of Table 30. These bytes are not
>> available for use by the
>> system software and cannot be written to or read from.
>> """
>>
>> I seriously doubt you are reading valid data with this change.
>
> That's not true. Reading ECC bytes is impossible when the ECC engine
> is enable, reading free OOB bytes is possible, and most of the time the
> caller doesn't care about ECC bytes when asking for a non-raw OOB
> access.
>
> Actually, it's even better to read the free OOB bytes with ECC enabled
> because they are ECC protected, so if you have bitflips in the free
> OOB section they will be fixed or reported as uncorrectable.
OK, then, maybe I'm wrong. We need mark some a couple blocks
as bad in the bootloader, and then have the kernel finds them
and builds the BBT properly, with this patch.
--
Ezequiel García, VanguardiaSur
www.vanguardiasur.com.ar
^ permalink raw reply [flat|nested] 9+ messages in thread