Hi
Well, I must confess, I'm surprised:-) It looks like you're right. This
leak, if indeed there is one, has been there since the very first version
of soc-camera. I've spent some time looking at the code and so far I don't
find an explanation for the missing videobuf_mmap_free() call. I'll have
another look and, unless I find an explanation, why it's not needed, I'll
make a patch.
Also keep in mind, that this bug is only relevant for videobuf(1) drivers,
which anyway have to be converted to videobuf2;-)
Thanks
Guennadi
On Wed, 7 Nov 2012, �~M�~^�V wrote:
> Dear Guennadi
>
> I'm sure it's a bug.In linux-2.6.x, we call open() will allocate a struct soc_camera_file which contains struct videobuf_queue;then usr will call request_buffer, soc_camera module will call videobuf_alloc_vb(q) which will be installed in q->bufs[i].
> My question is how to free q->bufs[i] which is allocated from vb = kzalloc(size + sizeof(*mem), GFP_KERNEL) if we use videobuf-dma-contig memory model?
> videobuf_mmap_free()->kfree(q->bufs[i]) should call at every call close();we can't call kfree(q->bufs[i]) at q->ops->buf_release which is called in stream_off(), because q->bufs[i] reserve struct videobuf_mapping, unmap() will can't free videobuf which is used to store video data. Also can't call videobuf_mmap_free()->kfree(q->bufs[i]) at last close(), because in linux-2.6.x once open() allocates a videobuf_queue.
> In linux-3.x.x, we should call videobuf_mmap_free()->kfree(q->bufs[i]) only once at module remove callbcak function.
> You say, videobuf mmap allocations will be freed automatically. I want to known soc_camera module how to free q->bufs[i] automatically.
> If is there no bug in soc camera module , i'm sure all device driver use soc camera module have bugs, such as sh_mobile_ceu_caera.c, mx1_caera.c, mx3_caera.c etc. all of them donn't call videobuf_mmap_free()->kfree(q->bufs[i]).
>
> Your reply will be higly appreciated!
>
>
> ------------------ 原始邮件 ------------------
> 发件人: "Guennadi Liakhovetski"<g.liakhovetski@xxxxxx>;
> 发送时间: 2012年11月6日(星期二) 晚上7:30
> 收件人: "再回首"<308123027@xxxxxx>;
> 抄送: "linux-media"<linux-media@xxxxxxxxxxxxxxx>;
> 主题: Re: soc camera driver module may case memory leak
>
>
> Hi
>
> On Mon, 5 Nov 2012, ~F~M ~[~^ ~V wrote:
>
> > Dear sir:
> > why not call "videobuf_mmap_free",when device close call "soc_camera_close" in linux-2.6.x;
>
> I haven't found any version, where this has been done. I don't think this
> is needed, because videobuf mmap allocations will be freed automatically
> upon the last close(). Please, dismiss your bugzilla entry.
>
> Thanks
> Guennadi
>
> > do the same in linux-3.x.x?
> > video capture flow:
> > 1)open
> > 2)set fmt
> > 3)request buffer-->__videobuf_mmap_setup-->videobuf_alloc_vb(q)
> > 4)mmap
> > 5)enqueue, dequeue
> > 6)unmap
> > 7)close--->soc_camera_close-->?should call:videobuf_mmap_free
> > NOTE:
> > I have reviewed all the code, found:soc_camera_driver device driver coders has no way(callback function) to call videobuf_mmap_free; it will case memory leak.N r y b X ǧv ^ ){.n + { bj) w*jg ݢj/ z ޖ 2 ޙ & )ߡ a G h j:+v w ٥
>
> ---
> Guennadi Liakhovetski, Ph.D.
> Freelance Open-Source Software Developer
> http://www.open-technology.de/
---
Guennadi Liakhovetski, Ph.D.
Freelance Open-Source Software Developer
http://www.open-technology.de/
--
To unsubscribe from this list: send the line "unsubscribe linux-media" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
