Patchwork omap : nand : fix subpage ecc issue with prefetch

login
register
mail settings
Submitter Kishore Kadiyala
Date May 2, 2011, 11:10 a.m.
Message ID <1304334638-2198-1-git-send-email-kishore.kadiyala@ti.com>
Download mbox | patch
Permalink /patch/93765/
State Accepted
Commit c5d8c0cae4af7d78823d32fcd1c458ee1a1b5489
Headers show

Comments

Kishore Kadiyala - May 2, 2011, 11:10 a.m.
For prefetch engine, read and write  got broken in commit '2c01946c'.
We never hit a scenario of not getting 'gpmc_prefetch_enable'
call success.
When reading/writing a subpage with a non divisible by 4 ecc number
of bytes, the mis-aligned bytes gets handled first before enabling
the Prefetch engine, then it reads/writes rest of the bytes.

Signed-off-by: Kishore Kadiyala <kishore.kadiyala@ti.com>
Signed-off-by: Vimal Singh <vimal.newwork@gmail.com>
Reported-by: Bryan DE FARIA <bdefaria@adeneo-embedded.com>
---
 drivers/mtd/nand/omap2.c |   12 +++++-------
 1 files changed, 5 insertions(+), 7 deletions(-)
vimal singh - May 5, 2011, 2:20 p.m.
Hi Artem,

Can you please give a look on this patch?
This patch fixes is required for anyone who wants to use this driver
without using HW ECC.

-- Vimal

On Mon, May 2, 2011 at 4:40 PM, Kishore Kadiyala
<kishore.kadiyala@ti.com> wrote:
> For prefetch engine, read and write  got broken in commit '2c01946c'.
> We never hit a scenario of not getting 'gpmc_prefetch_enable'
> call success.
> When reading/writing a subpage with a non divisible by 4 ecc number
> of bytes, the mis-aligned bytes gets handled first before enabling
> the Prefetch engine, then it reads/writes rest of the bytes.
>
> Signed-off-by: Kishore Kadiyala <kishore.kadiyala@ti.com>
> Signed-off-by: Vimal Singh <vimal.newwork@gmail.com>
> Reported-by: Bryan DE FARIA <bdefaria@adeneo-embedded.com>
> ---
>  drivers/mtd/nand/omap2.c |   12 +++++-------
>  1 files changed, 5 insertions(+), 7 deletions(-)
>
> diff --git a/drivers/mtd/nand/omap2.c b/drivers/mtd/nand/omap2.c
> index da9a351..2c8040f 100644
> --- a/drivers/mtd/nand/omap2.c
> +++ b/drivers/mtd/nand/omap2.c
> @@ -263,11 +263,10 @@ static void omap_read_buf_pref(struct mtd_info *mtd, u_char *buf, int len)
>        if (ret) {
>                /* PFPW engine is busy, use cpu copy method */
>                if (info->nand.options & NAND_BUSWIDTH_16)
> -                       omap_read_buf16(mtd, buf, len);
> +                       omap_read_buf16(mtd, (u_char *)p, len);
>                else
> -                       omap_read_buf8(mtd, buf, len);
> +                       omap_read_buf8(mtd, (u_char *)p, len);
>        } else {
> -               p = (u32 *) buf;
>                do {
>                        r_count = gpmc_read_status(GPMC_PREFETCH_FIFO_CNT);
>                        r_count = r_count >> 2;
> @@ -293,7 +292,7 @@ static void omap_write_buf_pref(struct mtd_info *mtd,
>                                                struct omap_nand_info, mtd);
>        uint32_t w_count = 0;
>        int i = 0, ret = 0;
> -       u16 *p;
> +       u16 *p = (u16 *)buf;
>        unsigned long tim, limit;
>
>        /* take care of subpage writes */
> @@ -309,11 +308,10 @@ static void omap_write_buf_pref(struct mtd_info *mtd,
>        if (ret) {
>                /* PFPW engine is busy, use cpu copy method */
>                if (info->nand.options & NAND_BUSWIDTH_16)
> -                       omap_write_buf16(mtd, buf, len);
> +                       omap_write_buf16(mtd, (u_char *)p, len);
>                else
> -                       omap_write_buf8(mtd, buf, len);
> +                       omap_write_buf8(mtd, (u_char *)p, len);
>        } else {
> -               p = (u16 *) buf;
>                while (len) {
>                        w_count = gpmc_read_status(GPMC_PREFETCH_FIFO_CNT);
>                        w_count = w_count >> 1;
> --
> 1.7.4.1
>
>
Artem Bityutskiy - May 5, 2011, 6:48 p.m.
On Mon, 2011-05-02 at 16:40 +0530, Kishore Kadiyala wrote:
> For prefetch engine, read and write  got broken in commit '2c01946c'.
> We never hit a scenario of not getting 'gpmc_prefetch_enable'
> call success.
> When reading/writing a subpage with a non divisible by 4 ecc number
> of bytes, the mis-aligned bytes gets handled first before enabling
> the Prefetch engine, then it reads/writes rest of the bytes.
> 
> Signed-off-by: Kishore Kadiyala <kishore.kadiyala@ti.com>
> Signed-off-by: Vimal Singh <vimal.newwork@gmail.com>
> Reported-by: Bryan DE FARIA <bdefaria@adeneo-embedded.com>

This needs a better commit message with more explanation and analysis of
the problem and how it was fixed.This commit message is not very
understandable. And then it needs also:

Cc: stable@kernel.org [2.6.36+]

Right? And then we could send it upstream.
vimal singh - May 6, 2011, 7:30 a.m.
On Fri, May 6, 2011 at 12:18 AM, Artem Bityutskiy <dedekind1@gmail.com> wrote:
> On Mon, 2011-05-02 at 16:40 +0530, Kishore Kadiyala wrote:
>> For prefetch engine, read and write  got broken in commit '2c01946c'.
>> We never hit a scenario of not getting 'gpmc_prefetch_enable'
>> call success.
>> When reading/writing a subpage with a non divisible by 4 ecc number
>> of bytes, the mis-aligned bytes gets handled first before enabling
>> the Prefetch engine, then it reads/writes rest of the bytes.
>>
>> Signed-off-by: Kishore Kadiyala <kishore.kadiyala@ti.com>
>> Signed-off-by: Vimal Singh <vimal.newwork@gmail.com>
>> Reported-by: Bryan DE FARIA <bdefaria@adeneo-embedded.com>
>
> This needs a better commit message with more explanation and analysis of
> the problem and how it was fixed.This commit message is not very
> understandable. And then it needs also:

How about a commit log like this:

When reading/writing a subpage (When HW ECC is not available/enable)
for number of bytes not aligned to 4, the mis-aligned bytes gets
handled first (by cpu copy mathod) before enabling the Prefetch
engin to/from 'p' (start of buffer 'buf').
Then it reads/writes rest of the bytes with the help of Prefetch
engine, if available, or again by cpu copy mathod. Currnetly,
reading/writing of rest of bytes, is not done correctly. Its trying to
read/write again to/from begining of buffer 'buf', overwriting the
mis-aligned bytes.

For prefetch engine read and write it got broken in commit '2c01946c'.
And we never hit the scenario of not getting 'gpmc_prefetch_enable'
call sucess. So, problem did not get caught up.

This patch fixes the issue.

> Cc: stable@kernel.org [2.6.36+]
>
> Right? And then we could send it upstream.

Yes, sure. I think kishore can take care of adding this in 'cc'
while posting next time.
Artem Bityutskiy - May 6, 2011, 6:56 p.m.
On Fri, 2011-05-06 at 13:00 +0530, Vimal Singh wrote:
> On Fri, May 6, 2011 at 12:18 AM, Artem Bityutskiy <dedekind1@gmail.com> wrote:
> > On Mon, 2011-05-02 at 16:40 +0530, Kishore Kadiyala wrote:
> >> For prefetch engine, read and write  got broken in commit '2c01946c'.
> >> We never hit a scenario of not getting 'gpmc_prefetch_enable'
> >> call success.
> >> When reading/writing a subpage with a non divisible by 4 ecc number
> >> of bytes, the mis-aligned bytes gets handled first before enabling
> >> the Prefetch engine, then it reads/writes rest of the bytes.
> >>
> >> Signed-off-by: Kishore Kadiyala <kishore.kadiyala@ti.com>
> >> Signed-off-by: Vimal Singh <vimal.newwork@gmail.com>
> >> Reported-by: Bryan DE FARIA <bdefaria@adeneo-embedded.com>
> >
> > This needs a better commit message with more explanation and analysis of
> > the problem and how it was fixed.This commit message is not very
> > understandable. And then it needs also:
> 
> How about a commit log like this:

Looks better, thanks. Waiting for the new patch.

> 
> When reading/writing a subpage (When HW ECC is not available/enable)
> for number of bytes not aligned to 4, the mis-aligned bytes gets
> handled first (by cpu copy mathod) before enabling the Prefetch
> engin to/from 'p' (start of buffer 'buf').
> Then it reads/writes rest of the bytes with the help of Prefetch
> engine, if available, or again by cpu copy mathod. Currnetly,
> reading/writing of rest of bytes, is not done correctly. Its trying to
> read/write again to/from begining of buffer 'buf', overwriting the
> mis-aligned bytes.
> 
> For prefetch engine read and write it got broken in commit '2c01946c'.
> And we never hit the scenario of not getting 'gpmc_prefetch_enable'
> call sucess. So, problem did not get caught up.
> 
> This patch fixes the issue.

Patch

diff --git a/drivers/mtd/nand/omap2.c b/drivers/mtd/nand/omap2.c
index da9a351..2c8040f 100644
--- a/drivers/mtd/nand/omap2.c
+++ b/drivers/mtd/nand/omap2.c
@@ -263,11 +263,10 @@  static void omap_read_buf_pref(struct mtd_info *mtd, u_char *buf, int len)
 	if (ret) {
 		/* PFPW engine is busy, use cpu copy method */
 		if (info->nand.options & NAND_BUSWIDTH_16)
-			omap_read_buf16(mtd, buf, len);
+			omap_read_buf16(mtd, (u_char *)p, len);
 		else
-			omap_read_buf8(mtd, buf, len);
+			omap_read_buf8(mtd, (u_char *)p, len);
 	} else {
-		p = (u32 *) buf;
 		do {
 			r_count = gpmc_read_status(GPMC_PREFETCH_FIFO_CNT);
 			r_count = r_count >> 2;
@@ -293,7 +292,7 @@  static void omap_write_buf_pref(struct mtd_info *mtd,
 						struct omap_nand_info, mtd);
 	uint32_t w_count = 0;
 	int i = 0, ret = 0;
-	u16 *p;
+	u16 *p = (u16 *)buf;
 	unsigned long tim, limit;
 
 	/* take care of subpage writes */
@@ -309,11 +308,10 @@  static void omap_write_buf_pref(struct mtd_info *mtd,
 	if (ret) {
 		/* PFPW engine is busy, use cpu copy method */
 		if (info->nand.options & NAND_BUSWIDTH_16)
-			omap_write_buf16(mtd, buf, len);
+			omap_write_buf16(mtd, (u_char *)p, len);
 		else
-			omap_write_buf8(mtd, buf, len);
+			omap_write_buf8(mtd, (u_char *)p, len);
 	} else {
-		p = (u16 *) buf;
 		while (len) {
 			w_count = gpmc_read_status(GPMC_PREFETCH_FIFO_CNT);
 			w_count = w_count >> 1;