Re: [PATCH] omap_vout: Added check in reqbuf & mmap for buf_size allocation

[Archit Taneja <archit@ti.com>]

Hi,

On Friday 17 June 2011 03:53 PM, Hiremath, Vaibhav wrote:
>> -----Original Message-----
>> From: Taneja, Archit
>> Sent: Friday, June 17, 2011 3:57 PM
>> To: Hiremath, Vaibhav
>> Cc: linux-media@vger.kernel.org; mchehab@redhat.com; hverkuil@xs4all.nl
>> Subject: Re: [PATCH] omap_vout: Added check in reqbuf&  mmap for buf_size
>> allocation
>>
>> Hi,
>>
>> On Friday 17 June 2011 03:33 PM, Hiremath, Vaibhav wrote:
>>>
>>>> -----Original Message-----
>>>> From: Taneja, Archit
>>>> Sent: Friday, June 17, 2011 2:16 PM
>>>> To: Hiremath, Vaibhav
>>>> Cc: linux-media@vger.kernel.org; mchehab@redhat.com; hverkuil@xs4all.nl
>>>> Subject: Re: [PATCH] omap_vout: Added check in reqbuf&   mmap for
>> buf_size
>>>> allocation
>>>>
>>>> Hi,
>>>>
>>>> On Friday 17 June 2011 01:44 AM, Hiremath, Vaibhav wrote:
>>>>> From: Vaibhav Hiremath<hvaibhav@ti.com>
>>>>>
>>>>> The usecase where, user allocates small size of buffer
>>>>> through bootargs (video1_bufsize/video2_bufsize) and later from
>>>> application
>>>>> tries to set the format which requires larger buffer size, driver
>>>> doesn't
>>>>> check for insufficient buffer size and allows application to map extra
>>>> buffer.
>>>>> This leads to kernel crash, when user application tries to access
>> memory
>>>>> beyond the allocation size.
>>>>
>>>> Query: Why do we pass the bufsize as bootargs in the first place? Is it
>>>> needed at probe time?
>>>>
>>> [Hiremath, Vaibhav] Yes, look out for variable
>> (video1_bufsize/video2_bufsize) in code.
>>
>> Yes, but why do we need to allocate some fixed size buffers at boot
>> time? Is it done because it makes our allocation happens faster during
>> reqbufs? Or is it required for VRFB?
>>
>> Could you explain the reason/startegy behind allocating buffers of a
>> particular size at boot time?
>>
> [Hiremath, Vaibhav] This is required to get rid of Linux memory fragmentation, user can reserve the memory based on usecase during boot time itself.

Ah okay, so if someone has passed this bootarg then he/she is certain 
about the max size needed for their use cases, now if they request for a 
larger buffer, its their fault, and hence we can return an error.
The patch makes sense now :)

I think we now need to work on what happens if the user doesn't enter 
video1_bufsize/video2_bufsize bootargs at all. "OMAP_VOUT_MAX_BUF_SIZE" 
doesn't look scalable I guess. Should we try not to allocate anything if 
video1_bufsize/video2_bufsize are not mentioned in bootargs? Another 
approach could be to replace OMAP_VOUT_MAX_BUF_SIZE with an inline 
function which returns the max buffer size based on what maximum 
dimensions the DSS for the current OMAP can support?

Thanks,
Archit

>
> Thanks,
> Vaibhav
>
>> Thanks,
>> Archit
>>
>>>
>>> Thanks,
>>> Vaibhav
>>>
>>>> Thanks,
>>>> Archit
>>>>
>>>>>
>>>>> Added check in both mmap and reqbuf call back function,
>>>>> and return error if the size of the buffer allocated by user through
>>>>> bootargs is less than the S_FMT size.
>>>>>
>>>>> Signed-off-by: Vaibhav Hiremath<hvaibhav@ti.com>
>>>>> ---
>>>>>     drivers/media/video/omap/omap_vout.c |   16 ++++++++++++++++
>>>>>     1 files changed, 16 insertions(+), 0 deletions(-)
>>>>>
>>>>> diff --git a/drivers/media/video/omap/omap_vout.c
>>>> b/drivers/media/video/omap/omap_vout.c
>>>>> index 3bc909a..343b50c 100644
>>>>> --- a/drivers/media/video/omap/omap_vout.c
>>>>> +++ b/drivers/media/video/omap/omap_vout.c
>>>>> @@ -678,6 +678,14 @@ static int omap_vout_buffer_setup(struct
>>>> videobuf_queue *q, unsigned int *count,
>>>>>     	startindex = (vout->vid == OMAP_VIDEO1) ?
>>>>>     		video1_numbuffers : video2_numbuffers;
>>>>>
>>>>> +	/* Check the size of the buffer */
>>>>> +	if (*size>    vout->buffer_size) {
>>>>> +		v4l2_err(&vout->vid_dev->v4l2_dev,
>>>>> +				"buffer allocation mismatch [%u] [%u]\n",
>>>>> +				*size, vout->buffer_size);
>>>>> +		return -ENOMEM;
>>>>> +	}
>>>>> +
>>>>>     	for (i = startindex; i<    *count; i++) {
>>>>>     		vout->buffer_size = *size;
>>>>>
>>>>> @@ -856,6 +864,14 @@ static int omap_vout_mmap(struct file *file,
>> struct
>>>> vm_area_struct *vma)
>>>>>     				(vma->vm_pgoff<<    PAGE_SHIFT));
>>>>>     		return -EINVAL;
>>>>>     	}
>>>>> +	/* Check the size of the buffer */
>>>>> +	if (size>    vout->buffer_size) {
>>>>> +		v4l2_err(&vout->vid_dev->v4l2_dev,
>>>>> +				"insufficient memory [%lu] [%u]\n",
>>>>> +				size, vout->buffer_size);
>>>>> +		return -ENOMEM;
>>>>> +	}
>>>>> +
>>>>>     	q->bufs[i]->baddr = vma->vm_start;
>>>>>
>>>>>     	vma->vm_flags |= VM_RESERVED;
>>>>> --
>>>>> 1.6.2.4
>>>>>
>>>>> --
>>>>> To unsubscribe from this list: send the line "unsubscribe linux-media"
>>>> in
>>>>> the body of a message to majordomo@vger.kernel.org
>>>>> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>>>>>
>>>
>>>
>
>