Re: [PATCH] media: v4l2-fh: define v4l2_fh struct regardless of condition

[Hans Verkuil]

On 3/26/20 3:30 AM, Seungchul Kim wrote:
> On 3/25/20 5:54 PM, Hans Verkuil wrote:
>> On 3/23/20 3:30 AM, Seungchul Kim wrote:
>>> v4l2_fh struct define differently by CONFIG_V4L2_MEM2MEM_DEV.
>>> If some vendors use CONFIG_V4L2_MEM2MEM_DEV by module, it can make the
>>> mismatch of v4l2_fh sturct.
>>>
>>> By the mismatch, the following error occurs.
>>> ===============================
>>> [    7.533506] v4l2_mem2mem: disagrees about version of symbol
>> video_devdata
>>> [    7.533594] v4l2_mem2mem: Unknown symbol video_devdata (err -22)
>>> [    7.535319] v4l2_mem2mem: disagrees about version of symbol
>>> v4l2_event_pending
>>> [    7.542532] v4l2_mem2mem: Unknown symbol v4l2_event_pending (err -22)
>>> ===============================
>>>
>>> So v4l2_fh struct is modified to does not have dependency for
>>> CONFIG_V4L2_MEM2MEM_DEV.
>>>
>>> Signed-off-by: Seungchul Kim <sc377.kim@xxxxxxxxxxx>
>>> ---
>>>  include/media/v4l2-fh.h | 2 --
>>>  1 file changed, 2 deletions(-)
>>>
>>> diff --git a/include/media/v4l2-fh.h b/include/media/v4l2-fh.h index
>>> 53b4dbb..b5b3e00 100644
>>> --- a/include/media/v4l2-fh.h
>>> +++ b/include/media/v4l2-fh.h
>>> @@ -53,9 +53,7 @@ struct v4l2_fh {
>>>  	unsigned int		navailable;
>>>  	u32			sequence;
>>>
>>> -#if IS_ENABLED(CONFIG_V4L2_MEM2MEM_DEV)
>>>  	struct v4l2_m2m_ctx	*m2m_ctx;
>>> -#endif
>>
>> This is a good change, but please also remove the same #if from
>> v4l2_ioctl_get_lock() in drivers/media/v4l2-core/v4l2-ioctl.c. That is now
>> no longer needed there either and without removing that those vendor
>> drivers would be using the wrong lock.
>>
>> Regards,
>>
>> 	Hans
> 
> Thank you for your comment. I worried about the same thing with you.
> But vfh->m2m_ctx on v4l2_ioctl_get_lock() is always null
> without CONFIG_V4L2_MEM2MEM_DEV, because m2m_ctx of v4l2-fh
> is initialized only in v4l2_mem2mem.c.
> Therefore it doesn't have a problem using an wrong lock
> regardless of modification, so I did not fix it.

It is: if your out-of-tree driver sets CONFIG_V4L2_MEM2MEM_DEV, then
I assume it will set vfh->m2m_ctx, and so the v4l2 core should also
return the right lock for that out-of-tree driver.

But in any case, if m2m_ctx is always part of v4l2-fh.h, then that
#if in v4l2-ioctl.c has become pointless and should be removed in any
case.

Regards,

	Hans