2013-04-19 13:17:03

by Hein_Tibosch

[permalink] [raw]
Subject: RE: [PATCH] platform: fall-back to driver name check if there is no id found

Hi Andy, Mika,

On 8 Feb 2013, Andy Shevchenko wrote:

> Some of the platform devices rely on the name of their driver to match with. In
> the current implementation, if platform id table is needed, they have to add
> the name to the platform id table which sounds alogical. The patch adjustes the
> logic of the id table matching to make sure we will fall-back to match by the
> driver name. This will make it similar to the DT or ACPI cases.
>
> Signed-off-by: Andy Shevchenko <[email protected]>
> Reported-by: Mika Westerberg <[email protected]>
> Cc: Eric Miao <[email protected]>
> Cc: Greg Kroah-Hartman <[email protected]>
> ---
> drivers/base/platform.c | 4 ++--
> 1 file changed, 2 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/base/platform.c b/drivers/base/platform.c
> index c0b8df3..452ba4b 100644
> --- a/drivers/base/platform.c
> +++ b/drivers/base/platform.c
> @@ -732,8 +732,8 @@ static int platform_match(struct device *dev, struct device_driver *drv)
> return 1;
>
> /* Then try to match against the id table */
> - if (pdrv->id_table)
> - return platform_match_id(pdrv->id_table, pdev) != NULL;
> + if (pdrv->id_table && platform_match_id(pdrv->id_table, pdev))
> + return 1;
>
> /* fall-back to driver name match */
> return (strcmp(pdev->name, drv->name) == 0);

When I upgraded an avr32 system from 3.8 to a recent next release, I found it was
broken: DMA was not available because the dw_dma driver did not get probed anymore.

The dw_dma driver does have a id_table, but the boards in arch/avr32 are still expecting
driver identification by name.

As long as we want to support this simple identification-by-name, I'd say Andy's patch
should get quickly into stable release.


Hein


2013-04-19 13:17:26

by Eric Miao

[permalink] [raw]
Subject: Re: [PATCH] platform: fall-back to driver name check if there is no id found

On Fri, Apr 19, 2013 at 8:35 PM, Hein Tibosch <[email protected]> wrote:
> Hi Andy, Mika,
>
> On 8 Feb 2013, Andy Shevchenko wrote:
>
>> Some of the platform devices rely on the name of their driver to match with. In
>> the current implementation, if platform id table is needed, they have to add
>> the name to the platform id table which sounds alogical. The patch adjustes the
>> logic of the id table matching to make sure we will fall-back to match by the
>> driver name. This will make it similar to the DT or ACPI cases.
>>
>> Signed-off-by: Andy Shevchenko <[email protected]>
>> Reported-by: Mika Westerberg <[email protected]>
>> Cc: Eric Miao <[email protected]>
>> Cc: Greg Kroah-Hartman <[email protected]>
>> ---
>> drivers/base/platform.c | 4 ++--
>> 1 file changed, 2 insertions(+), 2 deletions(-)
>>
>> diff --git a/drivers/base/platform.c b/drivers/base/platform.c
>> index c0b8df3..452ba4b 100644
>> --- a/drivers/base/platform.c
>> +++ b/drivers/base/platform.c
>> @@ -732,8 +732,8 @@ static int platform_match(struct device *dev, struct device_driver *drv)
>> return 1;
>>
>> /* Then try to match against the id table */
>> - if (pdrv->id_table)
>> - return platform_match_id(pdrv->id_table, pdev) != NULL;
>> + if (pdrv->id_table && platform_match_id(pdrv->id_table, pdev))
>> + return 1;
>>
>> /* fall-back to driver name match */
>> return (strcmp(pdev->name, drv->name) == 0);
>
> When I upgraded an avr32 system from 3.8 to a recent next release, I found it was
> broken: DMA was not available because the dw_dma driver did not get probed anymore.
>
> The dw_dma driver does have a id_table, but the boards in arch/avr32 are still expecting
> driver identification by name.

I think this is a different philosophy here. I'm actually fine with either. The
questions are really:

1. will it be a bit inconsistent if the driver is using id_table,
while the device
is still using a legacy way?

2. instead of introducing a different logic in the platform driver core code,
is it possible this could be fixed at the board level?

>
> As long as we want to support this simple identification-by-name, I'd say Andy's patch
> should get quickly into stable release.

2013-04-19 13:20:31

by Eric Miao

[permalink] [raw]
Subject: Re: [PATCH] platform: fall-back to driver name check if there is no id found

On Fri, Apr 19, 2013 at 8:35 PM, Hein Tibosch <[email protected]> wrote:
> Hi Andy, Mika,
>
> On 8 Feb 2013, Andy Shevchenko wrote:
>
>> Some of the platform devices rely on the name of their driver to match with. In
>> the current implementation, if platform id table is needed, they have to add
>> the name to the platform id table which sounds alogical. The patch adjustes the
>> logic of the id table matching to make sure we will fall-back to match by the
>> driver name. This will make it similar to the DT or ACPI cases.
>>
>> Signed-off-by: Andy Shevchenko <[email protected]>
>> Reported-by: Mika Westerberg <[email protected]>
>> Cc: Eric Miao <[email protected]>
>> Cc: Greg Kroah-Hartman <[email protected]>
>> ---
>> drivers/base/platform.c | 4 ++--
>> 1 file changed, 2 insertions(+), 2 deletions(-)
>>
>> diff --git a/drivers/base/platform.c b/drivers/base/platform.c
>> index c0b8df3..452ba4b 100644
>> --- a/drivers/base/platform.c
>> +++ b/drivers/base/platform.c
>> @@ -732,8 +732,8 @@ static int platform_match(struct device *dev, struct device_driver *drv)
>> return 1;
>>
>> /* Then try to match against the id table */
>> - if (pdrv->id_table)
>> - return platform_match_id(pdrv->id_table, pdev) != NULL;
>> + if (pdrv->id_table && platform_match_id(pdrv->id_table, pdev))
>> + return 1;
>>
>> /* fall-back to driver name match */
>> return (strcmp(pdev->name, drv->name) == 0);
>
> When I upgraded an avr32 system from 3.8 to a recent next release, I found it was
> broken: DMA was not available because the dw_dma driver did not get probed anymore.
>
> The dw_dma driver does have a id_table, but the boards in arch/avr32 are still expecting
> driver identification by name.

I think this is a different philosophy here. I'm actually fine with either. The
questions are really:

1. will it be a bit inconsistent if the driver is using id_table,
while the device
is still using a legacy way?

2. instead of introducing a different logic in the platform driver core code,
is it possible this could be fixed at the board level?

>
> As long as we want to support this simple identification-by-name, I'd say Andy's patch
> should get quickly into stable release.

2013-04-19 14:44:44

by Andy Shevchenko

[permalink] [raw]
Subject: Re: [PATCH] platform: fall-back to driver name check if there is no id found

On Fri, 2013-04-19 at 21:20 +0800, Eric Miao wrote:
> On Fri, Apr 19, 2013 at 8:35 PM, Hein Tibosch <[email protected]> wrote:
> > Hi Andy, Mika,
> >
> > On 8 Feb 2013, Andy Shevchenko wrote:
> >
> >> Some of the platform devices rely on the name of their driver to match with. In
> >> the current implementation, if platform id table is needed, they have to add
> >> the name to the platform id table which sounds alogical. The patch adjustes the
> >> logic of the id table matching to make sure we will fall-back to match by the
> >> driver name. This will make it similar to the DT or ACPI cases.
> >>
> >> Signed-off-by: Andy Shevchenko <[email protected]>
> >> Reported-by: Mika Westerberg <[email protected]>
> >> Cc: Eric Miao <[email protected]>
> >> Cc: Greg Kroah-Hartman <[email protected]>
> >> ---
> >> drivers/base/platform.c | 4 ++--
> >> 1 file changed, 2 insertions(+), 2 deletions(-)
> >>
> >> diff --git a/drivers/base/platform.c b/drivers/base/platform.c
> >> index c0b8df3..452ba4b 100644
> >> --- a/drivers/base/platform.c
> >> +++ b/drivers/base/platform.c
> >> @@ -732,8 +732,8 @@ static int platform_match(struct device *dev, struct device_driver *drv)
> >> return 1;
> >>
> >> /* Then try to match against the id table */
> >> - if (pdrv->id_table)
> >> - return platform_match_id(pdrv->id_table, pdev) != NULL;
> >> + if (pdrv->id_table && platform_match_id(pdrv->id_table, pdev))
> >> + return 1;
> >>
> >> /* fall-back to driver name match */
> >> return (strcmp(pdev->name, drv->name) == 0);
> >
> > When I upgraded an avr32 system from 3.8 to a recent next release, I found it was
> > broken: DMA was not available because the dw_dma driver did not get probed anymore.
> >
> > The dw_dma driver does have a id_table, but the boards in arch/avr32 are still expecting
> > driver identification by name.
>
> I think this is a different philosophy here. I'm actually fine with either. The
> questions are really:
>
> 1. will it be a bit inconsistent if the driver is using id_table,
> while the device
> is still using a legacy way?

Legacy way may be considered as a fall-back.

> 2. instead of introducing a different logic in the platform driver core code,
> is it possible this could be fixed at the board level?

It might be fixed in the driver, though I think that is ugly approach,
by adding name of the driver into its id_table.

P.S. There are actually a few drivers in current mainline that uses
mentioned approach, but with this patch it will be not needed anymore.

--
Andy Shevchenko <[email protected]>
Intel Finland Oy