Re: [PATCH 1/1] videobuf: Initialize lists in videobuf_buffer.

["Figo.zhang" <zhangtianfei@leadcoretech.com>]

On 11/17/2010 03:11 PM, Hans Verkuil wrote:
> On Wednesday, November 17, 2010 02:38:09 Andrew Chew wrote:
>>>> diff --git a/drivers/media/video/videobuf-dma-contig.c
>>> b/drivers/media/video/videobuf-dma-contig.c
>>>> index c969111..f7e0f86 100644
>>>> --- a/drivers/media/video/videobuf-dma-contig.c
>>>> +++ b/drivers/media/video/videobuf-dma-contig.c
>>>> @@ -193,6 +193,8 @@ static struct videobuf_buffer
>>> *__videobuf_alloc_vb(size_t size)
>>>>    	if (vb) {
>>>>    		mem = vb->priv = ((char *)vb) + size;
>>>>    		mem->magic = MAGIC_DC_MEM;
>>>> +		INIT_LIST_HEAD(&vb->stream);
>>>> +		INIT_LIST_HEAD(&vb->queue);
>>>
>>> i think it no need to be init, it just a list-entry.
>>
>> Okay, if that's really the case, then sh_mobile_ceu_camera.c, pxa_camera.c, mx1_camera.c, mx2_camera.c, and omap1_camera.c needs to be fixed to remove that WARN_ON(!list_empty(&vb->queue)); in their videobuf_prepare() methods, because those WARN_ON's are assuming that vb->queue is properly initialized as a list head.
>>
>> Which will it be?
>>
>
> These list entries need to be inited. It is bad form to have uninitialized
> list entries. It is not as if this is a big deal to initialize them properly.

in kernel source code, list entry are not often to be inited.

for example, see mm/vmscan.c register_shrinker(), no one init the 
shrinker->list.

another example: see mm/swapfile.c  add_swap_extent(), no one init the
new_se->list.