Re: [PATCH 1/2] mtd: spi-nor: rework spi nor read and write.

Re: [PATCH 1/2] mtd: spi-nor: rework spi nor read and write.

[Marek Vasut]

On Tuesday, July 28, 2015 at 11:23:02 AM, Michal Suchanek wrote:
> The spi_nor read and write functions pass thru the mtd retlen to the
> chip-specific read and write function. This makes it difficult to check
> for errors in read and write functions and these errors are not checked.
> This leads to silent data corruption.
> 
> This patch styles the chip-specific read and write function as unix
> read(2) and write(2) and updates the retlen in the spi-nor generic driver.
> 
> This also makes it possible to check for write errors.
> When pl330 fails to transfer the flash data over SPI I get I/O error
> instead of 4M of zeroes.
> 
> I do not have sst and fsl hardware to test with.
> 
> Signed-off-by: Michal Suchanek <hramrach@gmail.com>
> ---
>  drivers/mtd/devices/m25p80.c      | 33 ++++++++++++++---------
>  drivers/mtd/spi-nor/fsl-quadspi.c | 29 ++++++++++----------
>  drivers/mtd/spi-nor/spi-nor.c     | 57
> ++++++++++++++++++++++++++++----------- include/linux/mtd/spi-nor.h      
> |  8 +++---
>  4 files changed, 81 insertions(+), 46 deletions(-)
> 
> diff --git a/drivers/mtd/devices/m25p80.c b/drivers/mtd/devices/m25p80.c
> index d313f948b..d8f064b 100644
> --- a/drivers/mtd/devices/m25p80.c
> +++ b/drivers/mtd/devices/m25p80.c
> @@ -75,14 +75,15 @@ static int m25p80_write_reg(struct spi_nor *nor, u8
> opcode, u8 *buf, int len, return spi_write(spi, flash->command, len + 1);
>  }
> 
> -static void m25p80_write(struct spi_nor *nor, loff_t to, size_t len,
> -			size_t *retlen, const u_char *buf)
> +static ssize_t m25p80_write(struct spi_nor *nor, loff_t to, size_t len,
> +			    const u_char *buf)
>  {
>  	struct m25p *flash = nor->priv;
>  	struct spi_device *spi = flash->spi;
>  	struct spi_transfer t[2] = {};
>  	struct spi_message m;
>  	int cmd_sz = m25p_cmdsz(nor);
> +	ssize_t ret;
> 
>  	spi_message_init(&m);
> 
> @@ -100,9 +101,14 @@ static void m25p80_write(struct spi_nor *nor, loff_t
> to, size_t len, t[1].len = len;
>  	spi_message_add_tail(&t[1], &m);
> 
> -	spi_sync(spi, &m);
> +	ret = spi_sync(spi, &m);
> +	if (ret)
> +		return ret;
> 
> -	*retlen += m.actual_length - cmd_sz;
> +	ret = m.actual_length - cmd_sz;
> +	if (ret < 0)
> +		return -EIO;
> +	return ret;

I'd prefer to just add the return value and keep the retlen to keep error
codes and transfer length separate.

btw. you change the transfer length from unsigned to signed type -- long
transfer might get interpreted as an error.

[...]

Re: [PATCH 1/2] mtd: spi-nor: rework spi nor read and write.

[Michal Suchanek]

On 28 July 2015 at 20:15, Marek Vasut <marex@denx.de> wrote:
> On Tuesday, July 28, 2015 at 11:23:02 AM, Michal Suchanek wrote:
>> The spi_nor read and write functions pass thru the mtd retlen to the
>> chip-specific read and write function. This makes it difficult to check
>> for errors in read and write functions and these errors are not checked.
>> This leads to silent data corruption.
>>
>> This patch styles the chip-specific read and write function as unix
>> read(2) and write(2) and updates the retlen in the spi-nor generic driver.
>>
>> This also makes it possible to check for write errors.
>> When pl330 fails to transfer the flash data over SPI I get I/O error
>> instead of 4M of zeroes.
>>
>> I do not have sst and fsl hardware to test with.
>>
>> Signed-off-by: Michal Suchanek <hramrach@gmail.com>
>> ---
>>  drivers/mtd/devices/m25p80.c      | 33 ++++++++++++++---------
>>  drivers/mtd/spi-nor/fsl-quadspi.c | 29 ++++++++++----------
>>  drivers/mtd/spi-nor/spi-nor.c     | 57
>> ++++++++++++++++++++++++++++----------- include/linux/mtd/spi-nor.h
>> |  8 +++---
>>  4 files changed, 81 insertions(+), 46 deletions(-)
>>
>> diff --git a/drivers/mtd/devices/m25p80.c b/drivers/mtd/devices/m25p80.c
>> index d313f948b..d8f064b 100644
>> --- a/drivers/mtd/devices/m25p80.c
>> +++ b/drivers/mtd/devices/m25p80.c
>> @@ -75,14 +75,15 @@ static int m25p80_write_reg(struct spi_nor *nor, u8
>> opcode, u8 *buf, int len, return spi_write(spi, flash->command, len + 1);
>>  }
>>
>> -static void m25p80_write(struct spi_nor *nor, loff_t to, size_t len,
>> -                     size_t *retlen, const u_char *buf)
>> +static ssize_t m25p80_write(struct spi_nor *nor, loff_t to, size_t len,
>> +                         const u_char *buf)
>>  {
>>       struct m25p *flash = nor->priv;
>>       struct spi_device *spi = flash->spi;
>>       struct spi_transfer t[2] = {};
>>       struct spi_message m;
>>       int cmd_sz = m25p_cmdsz(nor);
>> +     ssize_t ret;
>>
>>       spi_message_init(&m);
>>
>> @@ -100,9 +101,14 @@ static void m25p80_write(struct spi_nor *nor, loff_t
>> to, size_t len, t[1].len = len;
>>       spi_message_add_tail(&t[1], &m);
>>
>> -     spi_sync(spi, &m);
>> +     ret = spi_sync(spi, &m);
>> +     if (ret)
>> +             return ret;
>>
>> -     *retlen += m.actual_length - cmd_sz;
>> +     ret = m.actual_length - cmd_sz;
>> +     if (ret < 0)
>> +             return -EIO;
>> +     return ret;
>
> I'd prefer to just add the return value and keep the retlen to keep error
> codes and transfer length separate.

I prefer to not pass around retlen because passing it around

 - causes code duplication
 - makes the code harder to understand

>
> btw. you change the transfer length from unsigned to signed type -- long
> transfer might get interpreted as an error.

Note that ssize_t is supposed to be enough for write(2) so when it
does not suffice you have a real system-wide problem.

That aside NOR flash sizes tend to be in the order of megabytes so
this concern is very theoretical.

Thanks

Michal

Re: [PATCH 1/2] mtd: spi-nor: rework spi nor read and write.

[Marek Vasut]

On Wednesday, July 29, 2015 at 06:16:21 AM, Michal Suchanek wrote:
> On 28 July 2015 at 20:15, Marek Vasut <marex@denx.de> wrote:
> > On Tuesday, July 28, 2015 at 11:23:02 AM, Michal Suchanek wrote:
> >> The spi_nor read and write functions pass thru the mtd retlen to the
> >> chip-specific read and write function. This makes it difficult to check
> >> for errors in read and write functions and these errors are not checked.
> >> This leads to silent data corruption.
> >> 
> >> This patch styles the chip-specific read and write function as unix
> >> read(2) and write(2) and updates the retlen in the spi-nor generic
> >> driver.
> >> 
> >> This also makes it possible to check for write errors.
> >> When pl330 fails to transfer the flash data over SPI I get I/O error
> >> instead of 4M of zeroes.
> >> 
> >> I do not have sst and fsl hardware to test with.
> >> 
> >> Signed-off-by: Michal Suchanek <hramrach@gmail.com>
> >> ---
> >> 
> >>  drivers/mtd/devices/m25p80.c      | 33 ++++++++++++++---------
> >>  drivers/mtd/spi-nor/fsl-quadspi.c | 29 ++++++++++----------
> >>  drivers/mtd/spi-nor/spi-nor.c     | 57
> >> 
> >> ++++++++++++++++++++++++++++----------- include/linux/mtd/spi-nor.h
> >> 
> >> |  8 +++---
> >>  
> >>  4 files changed, 81 insertions(+), 46 deletions(-)
> >> 
> >> diff --git a/drivers/mtd/devices/m25p80.c b/drivers/mtd/devices/m25p80.c
> >> index d313f948b..d8f064b 100644
> >> --- a/drivers/mtd/devices/m25p80.c
> >> +++ b/drivers/mtd/devices/m25p80.c
> >> @@ -75,14 +75,15 @@ static int m25p80_write_reg(struct spi_nor *nor, u8
> >> opcode, u8 *buf, int len, return spi_write(spi, flash->command, len +
> >> 1);
> >> 
> >>  }
> >> 
> >> -static void m25p80_write(struct spi_nor *nor, loff_t to, size_t len,
> >> -                     size_t *retlen, const u_char *buf)
> >> +static ssize_t m25p80_write(struct spi_nor *nor, loff_t to, size_t len,
> >> +                         const u_char *buf)
> >> 
> >>  {
> >>  
> >>       struct m25p *flash = nor->priv;
> >>       struct spi_device *spi = flash->spi;
> >>       struct spi_transfer t[2] = {};
> >>       struct spi_message m;
> >>       int cmd_sz = m25p_cmdsz(nor);
> >> 
> >> +     ssize_t ret;
> >> 
> >>       spi_message_init(&m);
> >> 
> >> @@ -100,9 +101,14 @@ static void m25p80_write(struct spi_nor *nor,
> >> loff_t to, size_t len, t[1].len = len;
> >> 
> >>       spi_message_add_tail(&t[1], &m);
> >> 
> >> -     spi_sync(spi, &m);
> >> +     ret = spi_sync(spi, &m);
> >> +     if (ret)
> >> +             return ret;
> >> 
> >> -     *retlen += m.actual_length - cmd_sz;
> >> +     ret = m.actual_length - cmd_sz;
> >> +     if (ret < 0)
> >> +             return -EIO;
> >> +     return ret;
> > 
> > I'd prefer to just add the return value and keep the retlen to keep error
> > codes and transfer length separate.
> 
> I prefer to not pass around retlen because passing it around
> 
>  - causes code duplication
>  - makes the code harder to understand

Not sure I want to argue on this one, since it looks like a matter of taste.
Adding the error return first and removing the retlen second might make the
patch more readable though.

> > btw. you change the transfer length from unsigned to signed type -- long
> > transfer might get interpreted as an error.
> 
> Note that ssize_t is supposed to be enough for write(2) so when it
> does not suffice you have a real system-wide problem.
> 
> That aside NOR flash sizes tend to be in the order of megabytes so
> this concern is very theoretical.

I do hope so :)