Re: [PATCH] mfd: Fix MACRO for commonly declared MFD cell attributes

From: Rhyland Klein
Date: Fri Feb 26 2016 - 11:35:40 EST


On 2/19/2016 11:28 AM, Rhyland Klein wrote:
> On 2/19/2016 3:50 AM, Lee Jones wrote:
>> On Thu, 18 Feb 2016, Rhyland Klein wrote:
>>
>>> MFD_ARRAY_SIZE() would not accurately return 0 if the passed
>>> parameter was NULL. Fix this so that num_resources will
>>> accurately be 0 in the case that _res is NULL.
>>>
>>> cc: Lee Jones <lee.jones@xxxxxxxxxx>
>>> cc: Laxman Dewangan <ldewangan@xxxxxxxxxx>
>>> Signed-off-by: Rhyland Klein <rklein@xxxxxxxxxx>
>>> ---
>>> include/linux/mfd/core.h | 15 +++++++++------
>>> 1 file changed, 9 insertions(+), 6 deletions(-)
>>>
>>> diff --git a/include/linux/mfd/core.h b/include/linux/mfd/core.h
>>> index 1a5a87f3cd38..62136ccff1df 100644
>>> --- a/include/linux/mfd/core.h
>>> +++ b/include/linux/mfd/core.h
>>> @@ -18,11 +18,11 @@
>>>
>>> #define MFD_ARRAY_SIZE(arr) (sizeof(arr) / sizeof((arr)[0]))
>>>
>>> -#define MFD_CELL_ALL(_name, _res, _pdata, _id, _compat, _match) \
>>> +#define MFD_CELL_ALL(_name, _nres, _res, _pdata, _id, _compat, _match) \
>>> { \
>>> .name = (_name), \
>>> .resources = (_res), \
>>> - .num_resources = MFD_ARRAY_SIZE((_res)), \
>>> + .num_resources = (_nres), \
>>> .platform_data = (_pdata), \
>>> .pdata_size = MFD_ARRAY_SIZE((_pdata)), \
>>> .of_compatible = (_compat), \
>>> @@ -31,16 +31,19 @@
>>> }
>>>
>>> #define OF_MFD_CELL(_name, _res, _pdata, _id, _compat) \
>>> - MFD_CELL_ALL(_name, _res, _pdata, _id, _compat, NULL) \
>>> + MFD_CELL_ALL(_name, MFD_ARRAY_SIZE((_res)), _res, \
>>> + _pdata, _id, _compat, NULL) \
>>
>> I'm confused. Why would it be any different just by changing the call
>> site of MFD_ARRAY_SIZE?
>
> It isn't different, but for MFD_CELL_NAME, it explicitly passes 0
> instead of using MFD_ARRAY_SIZE, as its the only place that doesn't
> expect to have resources.
>
>>
>> And what about .platform_data?
>
> This crashed for me (without the change) at :
>
> mfd_add_device():
> for (r = 0; r < cell->num_resources; r++) {
> res[r].name = cell->resources[r].name;
> res[r].flags = cell->resources[r].flags;
>
> where dereferencing cell->resources[0] when there are no resources. I
> guess the platform_data could do the same, but I didn't run into it.
>
>>
>> How about this change instead?
>>
>> diff --git a/include/linux/mfd/core.h b/include/linux/mfd/core.h
>> index 1a5a87f..8440f42 100644
>> --- a/include/linux/mfd/core.h
>> +++ b/include/linux/mfd/core.h
>> @@ -16,7 +16,7 @@
>>
>> #include <linux/platform_device.h>
>>
>> -#define MFD_ARRAY_SIZE(arr) (sizeof(arr) / sizeof((arr)[0]))
>> +#define MFD_ARRAY_SIZE(arr) (arr ? (sizeof(arr) / sizeof((arr)[0])) : 0)
>>
>> #define MFD_CELL_ALL(_name, _res, _pdata, _id, _compat, _match) \
>> { \
>>
> That was my first thought too. However, I see this when I try to compile
> that:
>
> In file included from drivers/mfd/max77620.c:18:0:
> include/linux/mfd/core.h:19:34: warning: the address of âgpio_resourcesâ
> will always evaluate as âtrueâ [-Waddress]
> #define MFD_ARRAY_SIZE(arr) (arr ? (sizeof(arr) / sizeof((arr)[0])) : 0)
>
> 7 different times. This patch was the only way I seemed to be able to
> WAR around compile time warnings.
>
> -rhyland
>

Did you not see warnings like this when you compiled the kernel? Did you
find a different approach than what I proposed above to deal with it?
I'd like to get this in soon so that when the max77620 drivers are all
in and using it, it should be functional.

-rhyland

--
nvpublic