Chapter 6. Changes

Table of Contents
6.1. Differences between V4L and V4L2
6.2. Changes of the V4L2 API
6.3. Relation of V4L2 to other Linux multimedia APIs
6.4. Experimental API Elements
6.5. Obsolete API Elements

The following chapters document the evolution of the V4L2 API, errata or extensions. They are also intended to help application and driver writers to port or update their code.

6.1. Differences between V4L and V4L2

The Video For Linux API was first introduced in Linux 2.1 to unify and replace various TV and radio device related interfaces, developed independently by driver writers in prior years. Starting with Linux 2.5 the much improved V4L2 API replaces the V4L API, although existing drivers will continue to support V4L applications in the future, either directly or through the V4L2 compatibility layer in the videodev kernel module translating ioctls on the fly. For a transition period not all drivers will support the V4L2 API.

6.1.1. Opening and Closing Devices

For compatibility reasons the character device file names recommended for V4L2 video capture, overlay, radio, teletext and raw vbi capture devices did not change from those used by V4L. They are listed in Chapter 4 and below in Table 6-1.

The V4L videodev module automatically assigns minor numbers to drivers in load order, depending on the registered device type. We recommend that V4L2 drivers by default register devices with the same numbers, but the system administrator can assign arbitrary minor numbers using driver module options. The major device number remains 81.

Table 6-1. V4L Device Types, Names and Numbers

Device TypeFile NameMinor Numbers
Video capture and overlay

/dev/video and /dev/bttv0a, /dev/video0 to /dev/video63

Radio receiver

/dev/radiob, /dev/radio0 to /dev/radio63

Teletext decoder

/dev/vtx, /dev/vtx0 to /dev/vtx31

Raw VBI capture

/dev/vbi, /dev/vbi0 to /dev/vbi31

a. According to Documentation/devices.txt these should be symbolic links to /dev/video0. Note the original bttv interface is not compatible with V4L or V4L2.
b. According to Documentation/devices.txt a symbolic link to /dev/radio0.

V4L prohibits (or used to prohibit) multiple opens of a device file. V4L2 drivers may support multiple opens, see Section 1.1 for details and consequences.

V4L drivers respond to V4L2 ioctls with an EINVAL error code. The compatibility layer in the V4L2 videodev module can translate V4L ioctl requests to their V4L2 counterpart, however a V4L2 driver usually needs more preparation to become fully V4L compatible. This is covered in more detail in Chapter 5.

6.1.2. Querying Capabilities

The V4L VIDIOCGCAP ioctl is equivalent to V4L2's VIDIOC_QUERYCAP.

The name field in struct video_capability became card in struct v4l2_capability, type was replaced by capabilities. Note V4L2 does not distinguish between device types like this, better think of basic video input, video output and radio devices supporting a set of related functions like video capturing, video overlay and VBI capturing. See Section 1.1 for an introduction.

struct video_capability typestruct v4l2_capability capabilities flagsPurpose
VID_TYPE_CAPTUREV4L2_CAP_VIDEO_CAPTUREThe video capture interface is supported.
VID_TYPE_TUNERV4L2_CAP_TUNERThe device has a tuner or modulator.
VID_TYPE_TELETEXTV4L2_CAP_VBI_CAPTUREThe raw VBI capture interface is supported.
VID_TYPE_OVERLAYV4L2_CAP_VIDEO_OVERLAYThe video overlay interface is supported.
VID_TYPE_CHROMAKEYV4L2_FBUF_CAP_CHROMAKEY in field capability of struct v4l2_framebufferWhether chromakey overlay is supported. For more information on overlay see Section 4.2.
VID_TYPE_CLIPPINGV4L2_FBUF_CAP_LIST_CLIPPING and V4L2_FBUF_CAP_BITMAP_CLIPPING in field capability of struct v4l2_framebufferWhether clipping the overlaid image is supported, see Section 4.2.
VID_TYPE_FRAMERAMV4L2_FBUF_CAP_EXTERNOVERLAY not set in field capability of struct v4l2_framebufferWhether overlay overwrites frame buffer memory, see Section 4.2.
VID_TYPE_SCALES-This flag indicates if the hardware can scale images. The V4L2 API implies the scale factor by setting the cropping dimensions and image size with the VIDIOC_S_CROP and VIDIOC_S_FMT ioctl, respectively. The driver returns the closest sizes possible. For more information on cropping and scaling see Section 1.11.
VID_TYPE_MONOCHROME-Applications can enumerate the supported image formats with the VIDIOC_ENUM_FMT ioctl to determine if the device supports grey scale capturing only. For more information on image formats see Chapter 2.
VID_TYPE_SUBCAPTURE-Applications can call the VIDIOC_G_CROP ioctl to determine if the device supports capturing a subsection of the full picture ("cropping" in V4L2). If not, the ioctl returns the EINVAL error code. For more information on cropping and scaling see Section 1.11.
VID_TYPE_MPEG_DECODER-Applications can enumerate the supported image formats with the VIDIOC_ENUM_FMT ioctl to determine if the device supports MPEG streams.

The audios field was replaced by capabilities flag V4L2_CAP_AUDIO, indicating if the device has any audio inputs or outputs. To determine their number applications can enumerate audio inputs with the VIDIOC_G_AUDIO ioctl. The audio ioctls are described in Section 1.5.

The maxwidth, maxheight, minwidth and minheight fields were removed. Calling the VIDIOC_S_FMT or VIDIOC_TRY_FMT ioctl with the desired dimensions returns the closest size possible, taking into account the current video standard, cropping and scaling limitations.

6.1.3. Video Sources

V4L provides the VIDIOCGCHAN and VIDIOCSCHAN ioctl using struct video_channel to enumerate the video inputs of a V4L device. The equivalent V4L2 ioctls are VIDIOC_ENUMINPUT, VIDIOC_G_INPUT and VIDIOC_S_INPUT using struct v4l2_input as discussed in Section 1.4.

The channel field counting inputs was renamed to index, the video input types were renamed as follows:

struct video_channel typestruct v4l2_input type

Unlike the tuners field expressing the number of tuners of this input, V4L2 assumes each video input is connected to at most one tuner. However a tuner can have more than one input, i. e. RF connectors, and a device can have multiple tuners. The index number of the tuner associated with the input, if any, is stored in field tuner of struct v4l2_input. Enumeration of tuners is discussed in Section 1.6.

The redundant VIDEO_VC_TUNER flag was dropped. Video inputs associated with a tuner are of type V4L2_INPUT_TYPE_TUNER. The VIDEO_VC_AUDIO flag was replaced by the audioset field. V4L2 considers devices with up to 32 audio inputs. Each set bit in the audioset field represents one audio input this video input combines with. For information about audio inputs and how to switch between them see Section 1.5.

The norm field describing the supported video standards was replaced by std. The V4L specification mentions a flag VIDEO_VC_NORM indicating whether the standard can be changed. This flag was a later addition together with the norm field and has been removed in the meantime. V4L2 has a similar, albeit more comprehensive approach to video standards, see Section 1.7 for more information.

6.1.4. Tuning

The V4L VIDIOCGTUNER and VIDIOCSTUNER ioctl and struct video_tuner can be used to enumerate the tuners of a V4L TV or radio device. The equivalent V4L2 ioctls are VIDIOC_G_TUNER and VIDIOC_S_TUNER using struct v4l2_tuner. Tuners are covered in Section 1.6.

The tuner field counting tuners was renamed to index. The fields name, rangelow and rangehigh remained unchanged.

The VIDEO_TUNER_PAL, VIDEO_TUNER_NTSC and VIDEO_TUNER_SECAM flags indicating the supported video standards were dropped. This information is now contained in the associated struct v4l2_input. No replacement exists for the VIDEO_TUNER_NORM flag indicating whether the video standard can be switched. The mode field to select a different video standard was replaced by a whole new set of ioctls and structures described in Section 1.7. Due to its ubiquity it should be mentioned the BTTV driver supports several standards in addition to the regular VIDEO_MODE_PAL (0), VIDEO_MODE_NTSC, VIDEO_MODE_SECAM and VIDEO_MODE_AUTO (3). Namely N/PAL Argentina, M/PAL, N/PAL, and NTSC Japan with numbers 3-6 (sic).

The VIDEO_TUNER_STEREO_ON flag indicating stereo reception became V4L2_TUNER_SUB_STEREO in field rxsubchans. This field also permits the detection of monaural and bilingual audio, see the definition of struct v4l2_tuner for details. Presently no replacement exists for the VIDEO_TUNER_RDS_ON and VIDEO_TUNER_MBS_ON flags.

The VIDEO_TUNER_LOW flag was renamed to V4L2_TUNER_CAP_LOW in the struct v4l2_tuner capability field.

The VIDIOCGFREQ and VIDIOCSFREQ ioctl to change the tuner frequency where renamed to VIDIOC_G_FREQUENCY and VIDIOC_S_FREQUENCY. They take a pointer to a struct v4l2_frequency instead of an unsigned long integer.

6.1.5. Image Properties

V4L2 has no equivalent of the VIDIOCGPICT and VIDIOCSPICT ioctl and struct video_picture. The following fields where replaced by V4L2 controls accessible with the VIDIOC_QUERYCTRL, VIDIOC_G_CTRL and VIDIOC_S_CTRL ioctls:

struct video_pictureV4L2 Control ID

The V4L picture controls are assumed to range from 0 to 65535 with no particular reset value. The V4L2 API permits arbitrary limits and defaults which can be queried with the VIDIOC_QUERYCTRL ioctl. For general information about controls see Section 1.8.

The depth (average number of bits per pixel) of a video image is implied by the selected image format. V4L2 does not explicitely provide such information assuming applications recognizing the format are aware of the image depth and others need not know. The palette field moved into the struct v4l2_pix_format:

struct video_picture palettestruct v4l2_pix_format pixfmt






























a. This is a custom format used by the BTTV driver, not one of the V4L2 standard formats.
b. Presumably all V4L RGB formats are little-endian, although some drivers might interpret them according to machine endianess. V4L2 defines little-endian, big-endian and red/blue swapped variants. For details see Section 2.4.
c. VIDEO_PALETTE_YUV422 and VIDEO_PALETTE_YUYV are the same formats. Some V4L drivers respond to one, some to the other.
d. Not to be confused with V4L2_PIX_FMT_YUV411P, which is a planar format.
e. V4L explains this as: "RAW capture (BT848)"
f. Not to be confused with V4L2_PIX_FMT_Y41P, which is a packed format.

V4L2 image formats are defined in Chapter 2. The image format can be selected with the VIDIOC_S_FMT ioctl.

6.1.6. Audio

The VIDIOCGAUDIO and VIDIOCSAUDIO ioctl and struct video_audio are used to enumerate the audio inputs of a V4L device. The equivalent V4L2 ioctls are VIDIOC_G_AUDIO and VIDIOC_S_AUDIO using struct v4l2_audio as discussed in Section 1.5.

The audio "channel number" field counting audio inputs was renamed to index.

On VIDIOCSAUDIO the mode field selects one of the VIDEO_SOUND_MONO, VIDEO_SOUND_STEREO, VIDEO_SOUND_LANG1 or VIDEO_SOUND_LANG2 audio demodulation modes. When the current audio standard is BTSC VIDEO_SOUND_LANG2 refers to SAP and VIDEO_SOUND_LANG1 is meaningless. Also undocumented in the V4L specification, there is no way to query the selected mode. On VIDIOCGAUDIO the driver returns the actually received audio programmes in this field. In the V4L2 API this information is stored in the struct v4l2_tuner rxsubchans and audmode fields, respectively. See Section 1.6 for more information on tuners. Related to audio modes struct v4l2_audio also reports if this is a mono or stereo input, regardless if the source is a tuner.

The following fields where replaced by V4L2 controls accessible with the VIDIOC_QUERYCTRL, VIDIOC_G_CTRL and VIDIOC_S_CTRL ioctls:

struct video_audioV4L2 Control ID

To determine which of these controls are supported by a driver V4L provides the flags VIDEO_AUDIO_VOLUME, VIDEO_AUDIO_BASS, VIDEO_AUDIO_TREBLE and VIDEO_AUDIO_BALANCE. In the V4L2 API the VIDIOC_QUERYCTRL ioctl reports if the respective control is supported. Accordingly the VIDEO_AUDIO_MUTABLE and VIDEO_AUDIO_MUTE flags where replaced by the boolean V4L2_CID_AUDIO_MUTE control.

All V4L2 controls have a step attribute replacing the struct video_audio step field. The V4L audio controls are assumed to range from 0 to 65535 with no particular reset value. The V4L2 API permits arbitrary limits and defaults which can be queried with the VIDIOC_QUERYCTRL ioctl. For general information about controls see Section 1.8.

6.1.7. Frame Buffer Overlay

The V4L2 ioctls equivalent to VIDIOCGFBUF and VIDIOCSFBUF are VIDIOC_G_FBUF and VIDIOC_S_FBUF. The base field of struct video_buffer remained unchanged, except V4L2 defines a flag to indicate non-destructive overlays instead of a NULL pointer. All other fields moved into the struct v4l2_pix_format fmt substructure of struct v4l2_framebuffer. The depth field was replaced by pixelformat. See Section 2.4 for a list of RGB formats and their respective color depths.

Instead of the special ioctls VIDIOCGWIN and VIDIOCSWIN V4L2 uses the general-purpose data format negotiation ioctls VIDIOC_G_FMT and VIDIOC_S_FMT. They take a pointer to a struct v4l2_format as argument. Here the win member of the fmt union is used, a struct v4l2_window.

The x, y, width and height fields of struct video_window moved into struct v4l2_rect substructure w of struct v4l2_window. The chromakey, clips, and clipcount fields remained unchanged. Struct video_clip was renamed to struct v4l2_clip, also containing a struct v4l2_rect, but the semantics are still the same.

The VIDEO_WINDOW_INTERLACE flag was dropped. Instead applications must set the field field to V4L2_FIELD_ANY or V4L2_FIELD_INTERLACED. The VIDEO_WINDOW_CHROMAKEY flag moved into struct v4l2_framebuffer, under the new name V4L2_FBUF_FLAG_CHROMAKEY.

In V4L, storing a bitmap pointer in clips and setting clipcount to VIDEO_CLIP_BITMAP (-1) requests bitmap clipping, using a fixed size bitmap of 1024 × 625 bits. Struct v4l2_window has a separate bitmap pointer field for this purpose and the bitmap size is determined by w.width and w.height.

The VIDIOCCAPTURE ioctl to enable or disable overlay was renamed to VIDIOC_OVERLAY.

6.1.8. Cropping

To capture only a subsection of the full picture V4L defines the VIDIOCGCAPTURE and VIDIOCSCAPTURE ioctls using struct video_capture. The equivalent V4L2 ioctls are VIDIOC_G_CROP and VIDIOC_S_CROP using struct v4l2_crop, and the related VIDIOC_CROPCAP ioctl. This is a rather complex matter, see Section 1.11 for details.

The x, y, width and height fields moved into struct v4l2_rect substructure c of struct v4l2_crop. The decimation field was dropped. In the V4L2 API the scaling factor is implied by the size of the cropping rectangle and the size of the captured or overlaid image.

The VIDEO_CAPTURE_ODD and VIDEO_CAPTURE_EVEN flags to capture only the odd or even field, respectively, were replaced by V4L2_FIELD_TOP and V4L2_FIELD_BOTTOM in the field named field of struct v4l2_pix_format and struct v4l2_window. These structures are used to select a capture or overlay format with the VIDIOC_S_FMT ioctl.

6.1.9. Reading Images, Memory Mapping Capturing using the read method

There is no essential difference between reading images from a V4L or V4L2 device using the read() function, however V4L2 drivers are not required to support this I/O method. Applications can determine if the function is available with the VIDIOC_QUERYCAP ioctl. All V4L2 devices exchanging data with applications must support the select() and poll() functions.

To select an image format and size, V4L provides the VIDIOCSPICT and VIDIOCSWIN ioctls. V4L2 uses the general-purpose data format negotiation ioctls VIDIOC_G_FMT and VIDIOC_S_FMT. They take a pointer to a struct v4l2_format as argument, here the struct v4l2_pix_format named pix of its fmt union is used.

For more information about the V4L2 read interface see Section 3.1. Capturing using memory mapping

Applications can read from V4L devices by mapping buffers in device memory, or more often just buffers allocated in DMA-able system memory, into their address space. This avoids the data copying overhead of the read method. V4L2 supports memory mapping as well, with a few differences.

 The image format must be selected before buffers are allocated, with the VIDIOC_S_FMT ioctl. When no format is selected the driver may use the last, possibly by another application requested format.

Applications cannot change the number of buffers. The it is built into the driver, unless it has a module option to change the number when the driver module is loaded.

The VIDIOC_REQBUFS ioctl allocates the desired number of buffers, this is a required step in the initialization sequence.

Drivers map all buffers as one contiguous range of memory. The VIDIOCGMBUF ioctl is available to query the number of buffers, the offset of each buffer from the start of the virtual file, and the overall amount of memory used, which can be used as arguments for the mmap() function.

Buffers are individually mapped. The offset and size of each buffer can be determined with the VIDIOC_QUERYBUF ioctl.

The VIDIOCMCAPTURE ioctl prepares a buffer for capturing. It also determines the image format for this buffer. The ioctl returns immediately, eventually with an EAGAIN error code if no video signal had been detected. When the driver supports more than one buffer applications can call the ioctl multiple times and thus have multiple outstanding capture requests.

The VIDIOCSYNC ioctl suspends execution until a particular buffer has been filled.

Drivers maintain an incoming and outgoing queue. VIDIOC_QBUF enqueues any empty buffer into the incoming queue. Filled buffers are dequeued from the outgoing queue with the VIDIOC_DQBUF ioctl. To wait until filled buffers become available this function, select() or poll() can be used. The VIDIOC_STREAMON ioctl must be called once after enqueuing one or more buffers to start capturing. Its counterpart VIDIOC_STREAMOFF stops capturing and dequeues all buffers from both queues. Applications can query the signal status, if known, with the VIDIOC_ENUMINPUT ioctl.

For a more in-depth discussion of memory mapping and examples, see Section 3.2.

6.1.10. Reading Raw VBI Data

Originally the V4L API did not specify a raw VBI capture interface, only the device file /dev/vbi was reserved for this purpose. The only driver supporting this interface was the BTTV driver, de-facto defining the V4L VBI interface. Reading from the device yields a raw VBI image with the following parameters:

struct v4l2_vbi_formatV4L, BTTV driver
sampling_rate28636363 Hz NTSC (or any other 525-line standard); 35468950 Hz PAL and SECAM (625-line standards)
sample_formatV4L2_PIX_FMT_GREY. The last four bytes (a machine endianess integer) contain a frame counter.
start[]10, 273 NTSC; 22, 335 PAL and SECAM

16, 16a

a. Old driver versions used different values, eventually the custom BTTV_VBISIZE ioctl was added to query the correct values.

Undocumented in the V4L specification, in Linux 2.3 the VIDIOCGVBIFMT and VIDIOCSVBIFMT ioctls using struct vbi_format were added to determine the VBI image parameters. These ioctls are only partially compatible with the V4L2 VBI interface specified in Section 4.7.

An offset field does not exist, sample_format is supposed to be VIDEO_PALETTE_RAW, equivalent to V4L2_PIX_FMT_GREY. The remaining fields are probably equivalent to struct v4l2_vbi_format.

Apparently only the Zoran (ZR 36120) driver implements these ioctls. The semantics differ from those specified for V4L2 in two ways. The parameters are reset on open() and VIDIOCSVBIFMT always returns an EINVAL error code if the parameters are invalid.

6.1.11. Miscellaneous

V4L2 has no equivalent of the VIDIOCGUNIT ioctl. Applications can find the VBI device associated with a video capture device (or vice versa) by reopening the device and requesting VBI data. For details see Section 1.1.

No replacement exists for VIDIOCKEY, and the V4L functions for microcode programming. A new interface for MPEG compression and playback devices is documented in Section 1.9.