[PATCH v5 03/16] media: rkisp1: Add user space ABI definitions

Hans Verkuil hverkuil at xs4all.nl
Fri Mar 9 05:14:00 PST 2018


On 08/03/18 11:00, Jacob Chen wrote:
> Hi Hans,
> 
> 2018-02-07 20:00 GMT+08:00 Hans Verkuil <hverkuil at xs4all.nl>:
>> On 12/29/17 08:52, Shunqian Zheng wrote:
>>> From: Jeffy Chen <jeffy.chen at rock-chips.com>
>>>
>>> Add the header for userspace
>>
>> General note: I saw four cases where this documentation referred to the
>> datasheet. Three comments on that:
>>
>> 1) You don't say which datasheet.
>> 2) I assume the datasheet is under NDA?
> 
> This datasheet can't be got by customers, even under NDA.

OK. This should be mentioned somewhere.

> 
>> 3) You do need to give enough information so a reasonable default can be
>>    used. I mentioned in an earlier review that creating an initial params
>>    struct that can be used as a templete would be helpful (or even
>>    required), and that would be a good place to put such defaults.
>>
> 
> It don't need a default config
> For applcation writers, they can just init it with zero data, and only
> set value for the part they concerned.

This too should be clearly documented. This is very useful information for
users of this API.

> 
> As for ABI, i have checked there is no mismatches.
> Those structures is 32 bit aligned both in 64bit/32bit env, since
> there is no 64bit value.
> "__attribute__ ((packed))" can avoid mismatches happen when we add a
> 64bit value to those structures.
> 
> As robin said, enums and bools are not guaranteed to be consistent
> between different compiler, so it's a potential risk.
> I have replace bools with unsigned char and enums with unsigned int.

Excellent.

Regards,

	Hans



More information about the linux-arm-kernel mailing list