On Thu, Apr 21, 2011 at 3:59 PM, Daniel GlÃckner <daniel-gl@xxxxxxx> wrote:
> Hi Bob,
>
> On Thu, Apr 21, 2011 at 11:17:42AM +0800, Bob Liu wrote:
>> +#ifdef CONFIG_MMU
>> Â Â Â if (i == queue->count || size != queue->buf_size) {
>> +#else
>> + Â Â if (i == queue->count || PAGE_ALIGN(size) != queue->buf_size) {
>> +#endif
>
> on mmu systems do_mmap_pgoff contains a len = PAGE_ALIGN(len); line.
> If we depend on this behavior, why not do it here as well and get rid
> of the #ifdef?
>
If do it in do_mmap_pgoff() the whole system will be effected, I am
not sure whether
it's correct and needed for other subsystem.
>> +unsigned long uvc_queue_get_unmapped_area(struct uvc_video_queue *queue,
>> + Â Â Â Â Â Â unsigned long addr, unsigned long len, unsigned long pgoff)
>> +{
>> + Â Â struct uvc_buffer *buffer;
>> + Â Â unsigned int i;
>> + Â Â int ret = 0;
>
> You still didn't change ret to unsigned long.
>
Oh, Sorry. My fault.
>> + Â Â addr = (unsigned long)queue->mem + buffer->buf.m.offset;
>> + Â Â ret = addr;
>
> Why the intermediate step using addr?
>
If don't return addr, do_mmap_pgoff() will return failure and we can't
setup vma correctly.
See mm/nommu.c line 1386(add = file->f_op->get_unmmapped_area() ).
>> diff --git a/drivers/media/video/v4l2-dev.c b/drivers/media/video/v4l2-dev.c
>> index 498e674..221e73f 100644
>> --- a/drivers/media/video/v4l2-dev.c
>> +++ b/drivers/media/video/v4l2-dev.c
>> @@ -368,6 +368,23 @@ static int v4l2_mmap(struct file *filp, struct vm_area_struct *vm)
>> Â Â Â return ret;
>> Â}
>>
>> +#ifdef CONFIG_MMU
>> +#define v4l2_get_unmapped_area NULL
>> +#else
>> +static unsigned long v4l2_get_unmapped_area(struct file *filp,
>> + Â Â Â Â Â Â unsigned long addr, unsigned long len, unsigned long pgoff,
>> + Â Â Â Â Â Â unsigned long flags)
>> +{
>> + Â Â struct video_device *vdev = video_devdata(filp);
>> +
>> + Â Â if (!vdev->fops->get_unmapped_area)
>> + Â Â Â Â Â Â return -ENOSYS;
>> + Â Â if (!video_is_registered(vdev))
>> + Â Â Â Â Â Â return -ENODEV;
>> + Â Â return vdev->fops->get_unmapped_area(filp, addr, len, pgoff, flags);
>> +}
>> +#endif
>> +
>> Â/* Override for the open function */
>> Âstatic int v4l2_open(struct inode *inode, struct file *filp)
>> Â{
>> @@ -452,6 +469,7 @@ static const struct file_operations v4l2_fops = {
>> Â Â Â .write = v4l2_write,
>> Â Â Â .open = v4l2_open,
>> Â Â Â .mmap = v4l2_mmap,
>> + Â Â .get_unmapped_area = v4l2_get_unmapped_area,
>> Â Â Â .unlocked_ioctl = v4l2_ioctl,
>> Â#ifdef CONFIG_COMPAT
>> Â Â Â .compat_ioctl = v4l2_compat_ioctl32,
>> diff --git a/include/media/v4l2-dev.h b/include/media/v4l2-dev.h
>> index 8266d5a..0616a43 100644
>> --- a/include/media/v4l2-dev.h
>> +++ b/include/media/v4l2-dev.h
>> @@ -63,6 +63,8 @@ struct v4l2_file_operations {
>> Â Â Â long (*ioctl) (struct file *, unsigned int, unsigned long);
>> Â Â Â long (*unlocked_ioctl) (struct file *, unsigned int, unsigned long);
>> Â Â Â int (*mmap) (struct file *, struct vm_area_struct *);
>> + Â Â unsigned long (*get_unmapped_area) (struct file *, unsigned long,
>> + Â Â Â Â Â Â Â Â Â Â unsigned long, unsigned long, unsigned long);
>> Â Â Â int (*open) (struct file *);
>> Â Â Â int (*release) (struct file *);
>> Â};
>
> I'd prefer a git revert c29fcff3daafbf46d64a543c1950bbd206ad8c1c for
> this block instead of reverting it together with the UVC changes.
>
Okay, I will confirm that and do it.
Thanks a lot for your review.
--
Regards,
--Bob
--
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
