[PATCH v9 0/1] virtio-media: Add device specification

[PATCH v9 0/1] virtio-media: Add device specification

[Albert Esteve]

New attempt of including virtio-media
device specification.

v8->v9:
- Rebased to virtio-1.4

v7->v8:
- Merged DQBUF and DQEVENT into the same
  item for unsupported ioctls
- Rewrote/clarified a couple paragraphs
  based on previous version reviews.

Virtio-media came from a discussion on virtio-dev
mailing list, which lead to Alex Courbot presenting
virtio-v4l2[1] specification as an alternative to
virtio-video.

Later, virtio-v4l2 was renamed to virtio-media[2]
and published at:

https://github.com/chromeos/virtio-media

The repository above includes a virtio-media driver able
to pass v4l2-compliance when proxying the vivid/vicodec
virtual devices or an actual UVC camera using the
V4L2 vhost device (available in the repository).
It also includes a FFmpeg-based video encoder
device. Steps to reproduce are also detailed[3].

Recently, virtio-media has landed in AOSP[4].
Also the driver patch has been sent to the
kernel, currently in its v2 [5].

Furthermore, virtio-media got a proposal to reserve
device ID 48, which was finally approved for
inclusion in v1.4.

There is some overlap with virtio-video in regards
to which devices it can handle. However, they take
different approaches, potentially making them
the preferable choice for different scenarios.
Furthermore, virtio-media has matured since it
was first presented, and there are public efforts
to have upstream driver and devices.

But mainly, given that virtio-media will be the virtualization
solution for media devices for ChromeOS, Android, and
possibly others, I think is important that it gets
standardized and included in the specification,
despite the aforementioned overlap.

Full PDF: https://drive.google.com/file/d/1sDZXWVlvPbHKKguhekCXMzkpNZZ0g3hV/view?usp=sharing
PDF with the media section only: https://drive.google.com/file/d/1JjOIzecnkFFlpFBRjPQEUyAOD3BDHAdw/view?usp=sharing

[1] https://mail.google.com/mail/u/0?ui=2&ik=73ebd65ebd&attid=0.1&permmsgid=msg-f:1767388565327924962&th=1887068940754ee2&view=att&disp=inline&realattid=f_libalimc0
[2] https://www.mail-archive.com/virtio-dev@lists.oasis-open.org/msg12665.html
[3] https://github.com/chromeos/virtio-media/blob/main/TRY_IT_OUT.md
[4] https://cs.android.com/android/platform/superproject/main/+/main:external/virtio-media/
[5] https://lore.kernel.org/lkml/20250201-virtio-media-v2-1-ac840681452d@gmail.com/T/

Albert Esteve (1):
  virtio-media: Add virtio media device specification

 conformance.tex                           |   6 +
 content.tex                               |   1 +
 device-types/media/description.tex        | 639 ++++++++++++++++++++++
 device-types/media/device-conformance.tex |  15 +
 device-types/media/driver-conformance.tex |  11 +
 5 files changed, 672 insertions(+)
 create mode 100644 device-types/media/description.tex
 create mode 100644 device-types/media/device-conformance.tex
 create mode 100644 device-types/media/driver-conformance.tex

-- 
2.49.0

[PATCH v9 1/1] virtio-media: Add virtio media device specification

[Albert Esteve]

Virtio-media is an encapsulation of the V4L2 UAPI into
virtio, able to virtualize any video device supported
by V4L2.

Note that virtio-media does not require the use of a
V4L2 device driver on the host or guest side -
V4L2 is only used as a host-guest protocol,
and both sides are free to convert it from/to any
model that they wish to use.

Reviewed-by: Matias Ezequiel Vara Larsen <mvaralar@redhat.com>
Reviewed-by: Alexandre Courbot <gnurou@gmail.com>
Signed-off-by: Albert Esteve <aesteve@redhat.com>
---
 conformance.tex                           |   6 +
 content.tex                               |   1 +
 device-types/media/description.tex        | 639 ++++++++++++++++++++++
 device-types/media/device-conformance.tex |  15 +
 device-types/media/driver-conformance.tex |  11 +
 5 files changed, 672 insertions(+)
 create mode 100644 device-types/media/description.tex
 create mode 100644 device-types/media/device-conformance.tex
 create mode 100644 device-types/media/driver-conformance.tex

diff --git a/conformance.tex b/conformance.tex
index d92a236..7220614 100644
--- a/conformance.tex
+++ b/conformance.tex
@@ -36,6 +36,8 @@ \section{Conformance Targets}\label{sec:Conformance / Conformance Targets}
 \ref{sec:Conformance / Driver Conformance / PMEM Driver Conformance},
 \ref{sec:Conformance / Driver Conformance / CAN Driver Conformance} or
 \ref{sec:Conformance / Driver Conformance / SPI Controller Driver Conformance}.
+\ref{sec:Conformance / Driver Conformance / PMEM Driver Conformance} or
+\ref{sec:Conformance / Driver Conformance / Media Driver Conformance}.
 
     \item Clause \ref{sec:Conformance / Legacy Interface: Transitional Device and Transitional Driver Conformance}.
   \end{itemize}
@@ -65,6 +67,8 @@ \section{Conformance Targets}\label{sec:Conformance / Conformance Targets}
 \ref{sec:Conformance / Device Conformance / PMEM Device Conformance},
 \ref{sec:Conformance / Device Conformance / CAN Device Conformance} or
 \ref{sec:Conformance / Device Conformance / SPI Controller Device Conformance}.
+\ref{sec:Conformance / Device Conformance / PMEM Device Conformance} or
+\ref{sec:Conformance / Device Conformance / Media Device Conformance}.
 
     \item Clause \ref{sec:Conformance / Legacy Interface: Transitional Device and Transitional Driver Conformance}.
   \end{itemize}
@@ -162,6 +166,7 @@ \section{Conformance Targets}\label{sec:Conformance / Conformance Targets}
 \input{device-types/pmem/driver-conformance.tex}
 \input{device-types/can/driver-conformance.tex}
 \input{device-types/spi/driver-conformance.tex}
+\input{device-types/media/driver-conformance.tex}
 
 \conformance{\section}{Device Conformance}\label{sec:Conformance / Device Conformance}
 
@@ -254,6 +259,7 @@ \section{Conformance Targets}\label{sec:Conformance / Conformance Targets}
 \input{device-types/pmem/device-conformance.tex}
 \input{device-types/can/device-conformance.tex}
 \input{device-types/spi/device-conformance.tex}
+\input{device-types/media/device-conformance.tex}
 
 \conformance{\section}{Legacy Interface: Transitional Device and Transitional Driver Conformance}\label{sec:Conformance / Legacy Interface: Transitional Device and Transitional Driver Conformance}
 A conformant implementation MUST be either transitional or
diff --git a/content.tex b/content.tex
index 09628ca..56b2cb1 100644
--- a/content.tex
+++ b/content.tex
@@ -778,6 +778,7 @@ \chapter{Device Types}\label{sec:Device Types}
 \input{device-types/pmem/description.tex}
 \input{device-types/can/description.tex}
 \input{device-types/spi/description.tex}
+\input{device-types/media/description.tex}
 
 \chapter{Reserved Feature Bits}\label{sec:Reserved Feature Bits}
 
diff --git a/device-types/media/description.tex b/device-types/media/description.tex
new file mode 100644
index 0000000..0527da7
--- /dev/null
+++ b/device-types/media/description.tex
@@ -0,0 +1,639 @@
+\section{Media Device}\label{sec:Device Types / Media Device}
+
+The virtio media device follows the same model (and structures) as V4L2. It
+can be used to virtualize cameras, codec devices, or any other device
+supported by V4L2. The complete definition of V4L2 structures and ioctls can
+be found under the
+\href{https://www.kernel.org/doc/html/latest/userspace-api/media/index.html}{V4L2 UAPI documentation}.
+
+V4L2 is a UAPI that allows a less privileged entity (user-space) to use video
+hardware exposed by a more privileged entity (the kernel). Virtio-media is an
+encapsulation of this API into virtio, turning it into a virtualization API
+for all classes of video devices supported by V4L2, where the device plays the
+role of the kernel and the driver the role of user-space.
+
+The device is therefore responsible for presenting a virtual device that behaves
+like an actual V4L2 device, which the driver can control.
+
+Note that virtio-media does not require the use of a V4L2 device driver or of
+Linux on any side - V4L2 is only used as a transport protocol,
+and both sides are free to convert it from/to any model that they wish to use.
+
+\subsection{Device ID}\label{sec:Device Types / Media Device / Device ID}
+
+48
+
+\subsection{Virtqueues}\label{sec:Device Types / Media Device / Virtqueues}
+
+\begin{description}
+\item[0] commandq - used for driver commands and device responses to these
+commands.
+\item[1] eventq - used for events sent by the device to the driver.
+\end{description}
+
+\subsection{Feature Bits}\label{sec:Device Types / Media Device / Feature Bits}
+
+None
+
+\subsection{Device Configuration Layout}\label{sec:Device Types / Media Device / Device Configuration Layout}
+
+The video device configuration space uses the following layout:
+
+\begin{lstlisting}
+struct virtio_media_config {
+    le32 device_caps;
+    le32 device_type;
+    le8 card[32];
+};
+\end{lstlisting}
+
+\begin{description}
+\item[\field{device_caps}] (driver-read-only) flags representing the device
+capabilities as used in
+\href{https://www.kernel.org/doc/html/latest/userspace-api/media/v4l/vidioc-querycap.html#c.V4L.v4l2_capability}{struct v4l2_capability}.
+It corresponds with the \field{device_caps} field in the \textit{struct video_device}.
+\item[\field{device_type}] (driver-read-only) informs the driver of the type
+of the video device. It corresponds with the \field{vfl_devnode_type} field of the device.
+\item[\field{card}] (driver-read-only) name of the device, a NUL-terminated
+UTF-8 string. It corresponds with the \field{card} field of the \textit{struct v4l2_capability}.
+If all the characters of the field are used, it does not need to be NUL-terminated.
+\end{description}
+
+\subsection{Device Initialization}
+
+A driver executes the following sequence to initialize a device:
+
+\begin{enumerate}
+\item Read the \field{device_caps} and \field{device_type} fields
+from the configuration layout to identify the device.
+\item Set up the \field{commandq} and \field{eventq} virtqueues.
+\item May open a session (see Section \ref{sec:Device Types / Media Device / Device Operation / Open Device})
+to use the device and send V4L2 ioctls in order to receive more information
+about the device, such as supported formats or controls.
+\end{enumerate}
+
+\subsection{Device Operation}\label{sec:Device Types / Media Device / Device Operation}
+
+The driver enqueues commands in the command queue for the device to process.
+The errors returned by each command are standard
+\href{https://www.kernel.org/doc/html/latest/userspace-api/media/gen-errors.html}{Linux kernel error codes}.
+For instance, a driver sending a command that contains invalid options will
+receive \textit{EINVAL} in return, after the device tries to process it.
+
+The device enqueues events in the event queue for the driver to process.
+
+\subsubsection{Command Virtqueue}
+
+\paragraph{Device Operation: Command headers}
+
+\begin{lstlisting}
+#define VIRTIO_MEDIA_CMD_OPEN 1
+#define VIRTIO_MEDIA_CMD_CLOSE 2
+#define VIRTIO_MEDIA_CMD_IOCTL 3
+#define VIRTIO_MEDIA_CMD_MMAP 4
+#define VIRTIO_MEDIA_CMD_MUNMAP 5
+
+/* Header for all virtio commands from the driver to the device on the commandq. */
+struct virtio_media_cmd_header {
+	le32 cmd;
+	le32 __reserved;
+};
+
+/* Header for all virtio responses from the device to the driver on the commandq. */
+struct virtio_media_resp_header {
+	le32 status;
+	le32 __reserved;
+};
+\end{lstlisting}
+
+A command consists of a command header \textit{virtio_media_cmd_header}
+containing the following device-readable field:
+
+\begin{description}
+\item[\field{cmd}] specifies a device request type (VIRTIO_MEDIA_CMD_*).
+\end{description}
+
+A response consists of a response header \textit{virtio_media_resp_header}
+containing the following device-writable field:
+
+\begin{description}
+\item[\field{status}] indicates a device request status.
+\end{description}
+
+When the device executes the command successfully, the value of the status
+field is 0. Conversely, when the device fails to execute the command, the value
+of the status fields corresponds with one of the standard Linux error codes.
+
+\drivernormative{\paragraph}{Device Operation: Command Virtqueue: Sessions}{Device Types / Media Device / Device Operation / Command Virtqueue}
+
+Sessions are how the device is multiplexed, allowing several distinct works to
+take place simultaneously. Before starting operation, the driver needs to open
+a session. This is equivalent to opening the \textit{/dev/videoX} file of the
+V4L2 device. Each session gets a unique ID assigned, which can be then used
+to perform actions on it.
+
+\paragraph{Device Operation: Open device}\label{sec:Device Types / Media Device / Device Operation / Open Device}
+
+\textbf{VIRTIO_MEDIA_CMD_OPEN} Command for creating a new session.
+
+This is the equivalent of calling \textit{open} on a V4L2 device node.
+The driver uses \textit{virtio_media_cmd_open} to send an open request.
+
+\begin{lstlisting}
+struct virtio_media_cmd_open {
+    struct virtio_media_cmd_header hdr;
+};
+\end{lstlisting}
+
+The device responds to \textit{VIRTIO_MEDIA_CMD_OPEN} with \textit{virtio_media_resp_open}.
+
+\begin{lstlisting}
+struct virtio_media_resp_open {
+    struct virtio_media_resp_header hdr;
+    le32 session_id;
+    le32 __reserved;
+};
+\end{lstlisting}
+
+\begin{description}
+\item[\field{session_id}] identifies the current session, which is used for
+other commands, predominantly ioctls.
+\end{description}
+
+\devicenormative{\subparagraph}{Device Operation: Open device}{Device Types / Media Device / Device Operation / Open device}
+
+Upon success, the device MUST set a \field{session_id} in \textit{virtio_media_resp_open}
+to an integer that is NOT used by any other open session.
+
+\paragraph{Device Operation: Close device}
+
+\textbf{VIRTIO_MEDIA_CMD_CLOSE} Command for closing an active session.
+
+This is the equivalent of calling \textit{close} on a previously opened V4L2
+device node. All resources associated with this session will be freed.
+
+This command does not require a response from the device.
+
+\begin{lstlisting}
+struct virtio_media_cmd_close {
+    struct virtio_media_cmd_header hdr;
+    le32 session_id;
+    le32 __reserved;
+};
+\end{lstlisting}
+
+\begin{description}
+\item[\field{session_id}] identifies the session to close.
+\end{description}
+
+\drivernormative{\subparagraph}{Device Operation: Close device}{Device Types / Media Device / Device Operation / Close device}
+
+The session ID SHALL NOT be used again after queueing this command, until it has
+been obtained again through a subsequent \textit{VIRTIO_MEDIA_CMD_OPEN} call.
+
+\paragraph{Device Operation: V4L2 ioctls}
+
+\textbf{VIRTIO_MEDIA_CMD_IOCTL} Command for executing an ioctl on an open
+session.
+
+This command tells the device to run one of the `VIDIOC_*` ioctls on the
+session identified by \textit{session_id}.
+
+\begin{lstlisting}
+struct virtio_media_cmd_ioctl {
+    struct virtio_media_cmd_header hdr;
+    le32 session_id;
+    le32 code;
+    /* Followed by the relevant ioctl command payload as defined in the macro */
+};
+\end{lstlisting}
+
+\begin{description}
+\item[\field{session_id}] identifies the session to run the ioctl on.
+\item[\field{code}] specifies the code of the \field{VIDIOC_*} ioctl to run.
+\end{description}
+
+The code is extracted from the
+\href{https://www.kernel.org/doc/html/latest/userspace-api/media/v4l/videodev.html}{videodev2.h},
+header file. The file defines the ioctl's codes, type of payload, and
+direction. The code consists of the second argument of the \field{_IO*} macro.
+
+For example, the \textit{VIDIOC_G_FMT} is defined as follows:
+
+\begin{lstlisting}
+#define VIDIOC_G_FMT _IOWR('V',  4, struct v4l2_format)
+\end{lstlisting}
+
+This means that its ioctl code is \textit{4}, its payload is a
+\textit{struct v4l2_format}, and its direction is \textit{WR} (i.e., the
+payload is written by both the driver and the device).
+See Section \ref{sec:Device Types / Media Device / V4L2 ioctls / Ioctls payload}
+for more information about the direction of ioctls.
+
+The payload struct layout always matches the 64-bit, little-endian
+representation of the corresponding V4L2 structure.
+
+The device responds to \textit{VIRTIO_MEDIA_CMD_IOCTL} with \textit{virtio_media_resp_ioctl}.
+
+\begin{lstlisting}
+struct virtio_media_resp_ioctl {
+    struct virtio_media_resp_header hdr;
+    /* Followed by the ioctl response payload as defined in the macro */
+};
+\end{lstlisting}
+
+\subparagraph{Ioctls payload}\label{sec:Device Types / Media Device / V4L2 ioctls / Ioctls payload}
+
+Each ioctl has a payload, which is defined by the third argument of the
+\field{_IO*} macro. 
+
+The payload of an ioctl in the descriptor chain follows the command structure,
+the response structure, or both depending on the direction:
+
+\begin{itemize}
+\item \textbf{_IOR} is read-only for the driver, meaning the payload
+follows the response in the device-writable section of the descriptor chain.
+\item \textbf{_IOW} is read-only for the device, meaning the payload
+follows the command in the device-readable section of the descriptor chain.
+\item \textbf{_IOWR} is writable by both the device and driver,
+meaning the payload must follow both the command in the device-readable section
+of the descriptor chain, and the response in the device-writable section.
+\end{itemize}
+
+A possible optimization for \textit{WR} ioctls is to provide the payload using
+descriptors that both point to the same buffer. This mimics the behavior of
+V4L2 ioctls where the data is only passed once and used as both input and
+output by the kernel.
+
+\devicenormative{\subparagraph}{Device Operation: V4L2 ioctls}{Device Types / Media Device / Device Operation / V4L2 ioctls}
+
+In case of success of a device-writable ioctl, the device MUST always write the
+payload in the device-writable part of the descriptor chain.
+
+In case of failure of a device-writable ioctl, the device is free to write the
+payload in the device-writable part of the descriptor chain or not. Some errors
+may still result in the payload being updated, and in this case the device is
+expected to write the updated payload. 
+
+\drivernormative{\subparagraph}{Device Operation: V4L2 ioctls}{Device Types / Media Device / Device Operation / V4L2 ioctls}
+
+For most V4L2 structures, the size is identical for both 32 and 64 bits versions.
+If the payload struct layout size differs for 32 and 64 bits, the driver
+MUST translate them to its the 64-bit, little-endian representation.
+
+If the device has not written the payload after an error (i.e., only the
+header is returned), the driver MUST assume that the payload has not
+been modified.
+
+\subparagraph{Handling of pointers to data in ioctl payload}
+\label{sec:Device Types / Media Device / V4L2 ioctls / Pointers in V4l2 ioctls}
+
+A few structures used as ioctl payloads contain pointers to further
+data needed for the ioctl. There are notably:
+
+\begin{itemize}
+\item The \field{planes} pointer of
+\href{https://www.kernel.org/doc/html/latest/userspace-api/media/v4l/buffer.html#struct-v4l2-buffer}{struct v4l2_buffer},
+which size is determined by the length member.
+\item The \field{controls} pointer of
+\href{https://www.kernel.org/doc/html/latest/userspace-api/media/v4l/vidioc-g-ext-ctrls.html#c.V4L.v4l2_ext_controls}{struct v4l2_ext_controls},
+which size is determined by the count member.
+\end{itemize}
+
+If the size of the pointed area is non-zero, then the main
+payload is immediately followed by the pointed data in their order of
+appearance in the structure.
+
+\devicenormative{\subparagraph}{Handling of pointers to data in ioctl payload}{Device Types / Media Device / Device Operation / Pointers in V4l2 ioctls}
+
+The pointer value, when its area is non-zero, is ignored by the device,
+which MUST return the value initially passed by the driver.
+
+
+\subparagraph{Handling of pointers to userspace memory in ioctl payload}
+\label{sec:Device Types / Media Device / V4L2 ioctls / Userspace memory}
+
+A few pointers (used for \textit{SHARED_PAGES} memory type,
+see \ref{sec:Device Types / Media Device / Device Operation / Shared Pages})
+are special in that they point to userspace memory in the
+original V4L2 specification. They are:
+
+\begin{itemize}
+\item The \field{m.userptr} member of \textit{struct v4l2_buffer} and
+\href{https://www.kernel.org/doc/html/latest/userspace-api/media/v4l/buffer.html#struct-v4l2-plane}{struct v4l2_plane}
+(technically an unsigned long, but designated a userspace address).
+\item The \field{ptr} member of \textit{struct v4l2_ext_ctrl}.
+\end{itemize}
+
+These pointers may cover large areas of scattered memory, which has the
+potential to require more descriptors than the virtio queue can provide. For
+these particular pointers only, a list of \textit{struct virtio_media_sg_entry}
+that covers the needed amount of memory for the pointer is used instead of
+using descriptors to map the pointed memory directly.
+
+\begin{lstlisting}
+struct virtio_media_sg_entry {
+	le64 start;
+	le32 len;
+	le32 __reserved;
+};
+\end{lstlisting}
+
+For each such pointer to read, the device reads as many SG entries as needed
+to cover the length of the pointed buffer, as described by its parent
+structure (\field{length} member of \textit{struct v4l2_buffer} or
+\textit{struct v4l2_plane} for buffer memory, and \field{size} member of
+\textit{struct v4l2_ext_control} for control data).
+
+Since the device never needs to modify the list of SG entries, it is only
+provided by the driver in the device-readable section of the descriptor chain,
+and not repeated in the device-writable section, even for WR ioctls.
+
+\subparagraph{Unsupported ioctls}
+
+A few ioctls are replaced by other, more suitable mechanisms.
+
+\begin{itemize}
+\item \textit{VIDIOC_QUERYCAP} is replaced by reading the configuration area
+(see \ref{sec:Device Types / Media Device / Device Configuration Layout}).
+\item \textit{VIDIOC_DQBUF} and \textit{VIDIOC_DQEVENT} are replaced by a dedicated event
+(see \ref{sec:Device Types / Media Device / Device Operation / Event Virtqueue}).
+\item \textit{VIDIOC_G_JPEGCOMP} and \textit{VIDIOC_S_JPEGCOMP} are deprecated
+and replaced by the controls of the JPEG class.
+\item \textit{VIDIOC_LOG_STATUS} is a driver-only operation and shall not be
+implemented by the device.
+\end{itemize}
+
+\devicenormative{\subparagraph}{Device Operation: Unsupported ioctls}{Device Types / Media Device / Device Operation / Unsupported ioctls}
+
+When a request is not supported, the device MUST return \textit{ENOTTY},
+which corresponds to the response for unknown ioctls.
+
+\paragraph{Device Operation: Mapping a MMAP buffer}
+
+\textbf{VIRTIO_MEDIA_CMD_MMAP} Command for mapping a MMAP buffer into the
+driver's address space.
+
+Shared memory region ID 0 is used to map MMAP buffers with
+the \textit{VIRTIO_MEDIA_CMD_MMAP} command.
+
+\begin{lstlisting}
+#define VIRTIO_MEDIA_MMAP_FLAG_RW (1 << 0)
+
+struct virtio_media_cmd_mmap {
+	struct virtio_media_cmd_header hdr;
+	le32 session_id;
+	le32 flags;
+	le32 offset;
+};
+\end{lstlisting}
+
+\begin{description}
+\item[\field{session_id}] identifies the session which the mapped buffer pertains to.
+\item[\field{flags}] is the set of flags for the mapping. \field{VIRTIO_MEDIA_MMAP_FLAG_RW}
+can be set if a read-write mapping is desired. Without this flag the mapping
+will be read-only.
+\item[\field{offset}] corresponds to the \field{mem_offset} field of the
+\textit{union v4l2_plane} for the plane to map. This field can be obtained
+using the \textit{VIDIOC_QUERYBUF} ioctl.
+\end{description}
+
+The device responds to \textit{VIRTIO_MEDIA_CMD_MMAP} with \textit{virtio_media_resp_mmap}.
+
+\begin{lstlisting}
+struct virtio_media_resp_mmap {
+    struct virtio_media_resp_header hdr;
+    le64 driver_addr;
+    le64 len;
+};
+\end{lstlisting}
+
+\begin{description}
+\item[\field{driver_addr}] offset into SHM region ID 0 of the start of the mapping.
+\item[\field{len}] length of the mapping as indicated by the \textit{struct v4l2_plane}
+the buffer belongs to.
+\end{description}
+
+\devicenormative{\subparagraph}{Device Operation: Mapping a MMAP buffer}{Device Types / Media Device / Device Operation / Mapping a MMAP buffer}
+
+The \textit{len} parameter of the \textit{virtio_media_resp_mmap} response sent
+by the device MUST always be equal to the length of the buffer.
+
+\paragraph{Device Operation: Unmapping a MMAP buffer}
+
+\textbf{VIRTIO_MEDIA_CMD_MUNMAP} unmaps a MMAP buffer previously mapped using \field{VIRTIO_MEDIA_CMD_MMAP}.
+
+\begin{lstlisting}
+struct virtio_media_cmd_munmap {
+    struct virtio_media_cmd_header hdr;
+    le64 driver_addr;
+};
+\end{lstlisting}
+
+\begin{description}
+\item[\field{driver_addr}] offset into SHM region ID 0 previously returned by 
+\textit{VIRTIO_MEDIA_CMD_MMAP} at which the buffer has been previously mapped.
+\end{description}
+
+The device responds to \textit{VIRTIO_MEDIA_CMD_MUNMAP} with \textit{virtio_media_resp_munmap}.
+
+\begin{lstlisting}
+struct virtio_media_resp_munmap {
+    struct virtio_media_resp_header hdr;
+};
+\end{lstlisting}
+
+\devicenormative{\subparagraph}{Device Operation: Unmapping a MMAP buffer}{Device Types / Media Device / Device Operation / Unmapping a MMAP buffer}
+
+The device MUST keep mappings performed using \textit{VIRTIO_MEDIA_CMD_MMAP}
+valid until \textit{VIRTIO_MEDIA_CMD_MUNMAP} is called, even if the buffers or
+session they belong to are released or closed by the driver.
+
+\paragraph{Device Operation: Memory Types}
+
+The semantics of the three V4L2 memory types (\textit{MMAP}, \textit{USERPTR}
+and \textit{DMABUF}) can easily be mapped to both driver and device context.
+
+\begin{lstlisting}
+enum virtio_media_memory {
+    VIRTIO_MEDIA_MMAP = V4L2_MEMORY_MMAP,
+    VIRTIO_MEDIA_SHARED_PAGES = V4L2_MEMORY_USERPTR,
+    VIRTIO_MEDIA_OBJECT = V4L2_MEMORY_DMABUF,
+};
+\end{lstlisting}
+
+\subparagraph{MMAP}
+
+\textit{MMAP} memory type is the semantic equivalent of
+\textit{V4L2_MEMORY_MMAP} in regular V4L2. 
+
+In virtio-media, \textit{MMAP} buffers are provisioned by the device, just like
+they are by the kernel in regular V4L2. Similarly to how userspace can map a
+\textit{MMAP} buffer into its address space using mmap and munmap, the
+virtio-media driver can map device buffers into the driver space by queueing the
+\textit{struct virtio_media_cmd_mmap} and \textit{struct virtio_media_cmd_munmap}
+commands to the commandq.
+
+\subparagraph{SHARED_PAGES}
+\label{sec:Device Types / Media Device / Device Operation / Shared Pages}
+
+\textit{SHARED_PAGES} memory type is the semantic equivalent of
+\textit{V4L2_MEMORY_USERPTR} in regular V4L2. 
+
+In virtio-media, \textit{SHARED_PAGES} buffers are provisioned by the driver,
+and use guest physical addresses. Instances of \textit{struct v4l2_buffer}
+and \textit{struct v4l2_plane} of this memory type are followed by a list of
+\textit{struct virtio_media_sg_entry}. For more information, see
+\ref{sec:Device Types / Media Device / V4L2 ioctls / Userspace memory}
+
+\devicenormative{\subparagraph}{Device Operation: Shared Pages}{Device Types / Media Device / Device Operation / Shared Pages}
+
+The device MUST not alter the pointer values provided by the driver, i.e.
+the \field{m.userptr} member of \textit{struct v4l2_buffer} and
+\textit{struct v4l2_plane} MUST be returned to the driver with the same value
+as it was provided.
+
+\subparagraph{VIRTIO_OBJECT}
+\label{sec:Device Types / Media Device / Device Operation / Virtio Object}
+
+\textit{VIRTIO_OBJECT} memory type is the semantic equivalent of
+\textit{V4L2_MEMORY_DMABUF} in regular V4L2. 
+
+In virtio-media, \textit{VIRTIO_OBJECT} buffers are provisioned by a virtio
+object, just like they are by a \textit{DMABUF} in regular V4L2. Virtio objects
+are 16-bytes UUIDs and do not fit in the placeholders for file descriptors, so
+they follow their embedding data structure as needed. For example, in multi-planar
+buffers, \textit{struct v4l2_plane} structures are located after the
+\textit{struct v4l2_buffer} in the memory layout. Then, after the last plane,
+the memory contains an array of UUIDs in which the first element corresponds
+with the first plane, and so on.
+
+Conversely to \textit{SHARED_PAGES} buffers, which SG lists are never
+modified by the device, UUIDs of virtio objects need to be added in both
+the device-readable and device-writable section of the descriptor chain
+so the device can update them if needed.
+
+Device-allocated buffers with the \textit{VIRTIO_MEDIA_MMAP} memory type may also
+be exported as virtio objects for use with another virtio device using the
+\textit{VIDIOC_EXPBUF} ioctl. The fd placefolder of \textit{v4l2_exportbuffer}
+means that space for the UUID needs to be reserved right after that structure.
+
+\devicenormative{\subparagraph}{Device Operation: Virtio Object}{Device Types / Media Device / Device Operation / Virtio Object}
+
+The device MUST leave the \field{fd} placeholder of the V4L2 structure unchanged.
+
+\subsubsection{Event Virtqueue}
+\label{sec:Device Types / Media Device / Device Operation / Event Virtqueue}
+
+Events are asynchronous notifications to the driver. In virtio-media,
+they are used as a replacement for the \textit{VIDIOC_DQBUF} and
+\textit{VIDIOC_DQEVENT} ioctls and the polling mechanism, which would be
+impractical to implement on top of virtio.
+
+\paragraph{Device Operation: Event header}
+
+\begin{lstlisting}
+#define VIRTIO_MEDIA_EVT_ERROR 0
+#define VIRTIO_MEDIA_EVT_DQBUF 1
+#define VIRTIO_MEDIA_EVT_EVENT 2
+
+/* Header for events queued by the device for the driver on the eventq. */
+struct virtio_media_event_header {
+    le32 event;
+    le32 session_id;
+};
+\end{lstlisting}
+
+\begin{description}
+\item[\field{event}] one of \field{VIRTIO_MEDIA_EVT_*}.
+\item[\field{session_id}] ID of the session the event applies to.
+\end{description}
+
+\paragraph{Device Operation: Device-side error}
+
+\textbf{VIRTIO_MEDIA_EVT_ERROR} Unrecoverable session error. Upon
+emitting this event, the device considers the session mentioned in the
+header to be invalid, and returns an error to all future commands
+referring to it. Upon receiving this event, the driver stops using the
+session, and shall close it as soon as possible.
+
+\begin{lstlisting}
+struct virtio_media_event_error {
+    struct virtio_media_event_header hdr;
+    le32 errno;
+    le32 __reserved;
+};
+\end{lstlisting}
+
+\begin{description}
+\item[\field{errno}] error code describing the kind of error that occurred.
+\end{description}
+
+\devicenormative{\subparagraph}{Device Operation: Device-side error}{Device Types / Media Device / Device Operation / Device-side error}
+
+After an error is signaled, when the device considers the session as
+non-existing, the device MUST NOT recycle the session ID until the driver has
+explicitly closed it.
+
+\drivernormative{\subparagraph}{Device Operation: Device-side error}{Device Types / Media Device / Device Operation / Device-side error}
+
+Upon receiving an error event for a session, the driver MUST
+explicitly close the session using a \textit{VIRTIO_MEDIA_CMD_CLOSE} command.
+
+\paragraph{Device Operation: Dequeue buffer}
+\label{sec:Device Types / Media Device / Device Operation / Dequeue buffer}
+
+\textbf{VIRTIO_MEDIA_EVT_DQBUF} signals that a buffer is not being used anymore
+by the device and is returned to the driver.
+
+Every time a buffer previously queued (i.e., using the
+\textit{VIDIOC_QBUF} ioctl) is done being processed, the
+device queues a \textit{struct virtio_media_event_dqbuf} event on
+the eventq, signifying that the buffer may be used again by the driver.
+This is like an implicit \textit{VIDIOC_DQBUF} ioctl.
+
+\begin{lstlisting}
+#define VIRTIO_MEDIA_MAX_PLANES 8
+
+struct virtio_media_event_dqbuf {
+    struct virtio_media_event_header hdr;
+    struct v4l2_buffer buffer;
+    struct v4l2_plane planes[VIRTIO_MEDIA_MAX_PLANES];
+};
+\end{lstlisting}
+
+\begin{description}
+\item[\field{buffer}] \textit{struct v4l2_buffer} describing the buffer that has been dequeued.
+\item[\field{planes}] array of \textit{struct v4l2_plane} containing the plane information for multi-planar buffers.
+\end{description}
+
+Note that in the case of a \field{SHARED_PAGES} buffer, the \textit{struct v4l2_buffer}
+used as event payload is not followed by the buffer's SG entries: since that memory
+is the same that the driver submitted with the \textit{VIDIOC_QBUF}, it would
+be redundant to have it here.
+
+\drivernormative{\subparagraph}{Device Operation: Dequeue buffer}{Device Types / Media Device / Device Operation / Dequeue buffer}
+
+Pointer values in the \textit{struct v4l2_buffer} and \textit{struct v4l2_plane}
+are meaningless and MUST be ignored by the driver. It is recommended that the
+device sets them to NULL in order to avoid leaking potential device addresses.
+
+\paragraph{Device Operation: Emit an event}
+
+\textbf{VIRTIO_MEDIA_EVT_EVENT} Signals that a V4L2 event has been emitted for a session.
+
+Every time an event for which the driver has been previously subcribed to
+(i.e., using the \textit{VIDIOC_SUBSCRIBE_EVENT} ioctl) is signaled, the
+device queues a \textit{struct virtio_media_event_event} event on the eventq.
+This is like an implicit \textit{VIDIOC_DQEVENT} ioctl.
+
+\begin{lstlisting}
+struct virtio_media_event_event {
+    struct virtio_media_event_header hdr;
+    struct v4l2_event event;
+};
+\end{lstlisting}
+
+\begin{description}
+\item[\field{event}] \textit{struct v4l2_event} describing the event that occurred.
+\end{description}
diff --git a/device-types/media/device-conformance.tex b/device-types/media/device-conformance.tex
new file mode 100644
index 0000000..99112ff
--- /dev/null
+++ b/device-types/media/device-conformance.tex
@@ -0,0 +1,15 @@
+\conformance{\subsection}{Media Device Conformance}\label{sec:Conformance / Device Conformance / Media Device Conformance}
+
+A Media device MUST conform to the following normative statements:
+
+\begin{itemize}
+\item \ref{devicenormative:Device Types / Media Device / Device Operation / Open device}
+\item \ref{devicenormative:Device Types / Media Device / Device Operation / V4L2 ioctls}
+\item \ref{devicenormative:Device Types / Media Device / Device Operation / Pointers in V4l2 ioctls}
+\item \ref{devicenormative:Device Types / Media Device / Device Operation / Unsupported ioctls}
+\item \ref{devicenormative:Device Types / Media Device / Device Operation / Mapping a MMAP buffer}
+\item \ref{devicenormative:Device Types / Media Device / Device Operation / Unmapping a MMAP buffer}
+\item \ref{devicenormative:Device Types / Media Device / Device Operation / Shared Pages}
+\item \ref{devicenormative:Device Types / Media Device / Device Operation / Virtio Object}
+\item \ref{devicenormative:Device Types / Media Device / Device Operation / Device-side error}
+\end{itemize}
diff --git a/device-types/media/driver-conformance.tex b/device-types/media/driver-conformance.tex
new file mode 100644
index 0000000..eb380ad
--- /dev/null
+++ b/device-types/media/driver-conformance.tex
@@ -0,0 +1,11 @@
+\conformance{\subsection}{Media Driver Conformance}\label{sec:Conformance / Driver Conformance / Media Driver Conformance}
+
+A Media driver MUST conform to the following normative statements:
+
+\begin{itemize}
+\item \ref{drivernormative:Device Types / Media Device / Device Operation / Command Virtqueue}
+\item \ref{drivernormative:Device Types / Media Device / Device Operation / Close device}
+\item \ref{drivernormative:Device Types / Media Device / Device Operation / V4L2 ioctls}
+\item \ref{drivernormative:Device Types / Media Device / Device Operation / Device-side error}
+\item \ref{drivernormative:Device Types / Media Device / Device Operation / Dequeue buffer}
+\end{itemize}
-- 
2.49.0

Re: [PATCH v9 0/1] virtio-media: Add device specification

[Matias Ezequiel Vara Larsen]

On Wed, Jul 09, 2025 at 11:34:55AM +0200, Albert Esteve wrote:
> New attempt of including virtio-media
> device specification.
> 
> v8->v9:
> - Rebased to virtio-1.4

I wonder if it wouldn't be better to just send a patch that fixes the
series instead of the whole series again. Such a change would be a non
editorial change thus not requiring to vote again. The v8 has been
already voted and accepted.

Matias

RE: [PATCH v9 0/1] virtio-media: Add device specification

[Parav Pandit]

> From: Matias Ezequiel Vara Larsen <mvaralar@redhat.com>
> Sent: 09 July 2025 03:14 PM
> 
> On Wed, Jul 09, 2025 at 11:34:55AM +0200, Albert Esteve wrote:
> > New attempt of including virtio-media
> > device specification.
> >
> > v8->v9:
> > - Rebased to virtio-1.4
> 
> I wonder if it wouldn't be better to just send a patch that fixes the series
> instead of the whole series again. Such a change would be a non editorial
> change thus not requiring to vote again. The v8 has been already voted and
> accepted.
> 
> Matias
> 

Right. I was able to apply v9 without merge conflict.
Even though, I prefer what Matias suggested to send the fixup patch so that v9 does not need to go through the review and vote again.

Re: [PATCH v9 0/1] virtio-media: Add device specification

[Albert Esteve]

On Wed, Jul 9, 2025 at 11:49 AM Parav Pandit <parav@nvidia.com> wrote:
>
>
> > From: Matias Ezequiel Vara Larsen <mvaralar@redhat.com>
> > Sent: 09 July 2025 03:14 PM
> >
> > On Wed, Jul 09, 2025 at 11:34:55AM +0200, Albert Esteve wrote:
> > > New attempt of including virtio-media
> > > device specification.
> > >
> > > v8->v9:
> > > - Rebased to virtio-1.4
> >
> > I wonder if it wouldn't be better to just send a patch that fixes the series
> > instead of the whole series again. Such a change would be a non editorial
> > change thus not requiring to vote again. The v8 has been already voted and
> > accepted.
> >
> > Matias
> >
>
> Right. I was able to apply v9 without merge conflict.
> Even though, I prefer what Matias suggested to send the fixup patch so that v9 does not need to go through the review and vote again.

I'm ok with that. I assumed that as it had no changes from v8
whatsoever, it would not require a vote. If it does, let's disregard
this new version.

What do you mean for a fixup patch? New patch series with just the
changeset required to eliminate the conflict from v8?

>

RE: [PATCH v9 0/1] virtio-media: Add device specification

[Parav Pandit]

> From: Albert Esteve <aesteve@redhat.com>
> Sent: 09 July 2025 03:31 PM
> 
> On Wed, Jul 9, 2025 at 11:49 AM Parav Pandit <parav@nvidia.com> wrote:
> >
> >
> > > From: Matias Ezequiel Vara Larsen <mvaralar@redhat.com>
> > > Sent: 09 July 2025 03:14 PM
> > >
> > > On Wed, Jul 09, 2025 at 11:34:55AM +0200, Albert Esteve wrote:
> > > > New attempt of including virtio-media device specification.
> > > >
> > > > v8->v9:
> > > > - Rebased to virtio-1.4
> > >
> > > I wonder if it wouldn't be better to just send a patch that fixes
> > > the series instead of the whole series again. Such a change would be
> > > a non editorial change thus not requiring to vote again. The v8 has
> > > been already voted and accepted.
> > >
> > > Matias
> > >
> >
> > Right. I was able to apply v9 without merge conflict.
> > Even though, I prefer what Matias suggested to send the fixup patch so that
> v9 does not need to go through the review and vote again.
> 
> I'm ok with that. I assumed that as it had no changes from v8 whatsoever, it
> would not require a vote. If it does, let's disregard this new version.
> 
If its editorial change, it does not require a vote.

> What do you mean for a fixup patch? New patch series with just the
> changeset required to eliminate the conflict from v8?
> 
I was looking for something that I apply locally to resolve the merge conflict myself.
That way, its basically just v8 + fixup.

So basically a short diff that you did manually, which I can also do before merging it.

Re: [PATCH v9 0/1] virtio-media: Add device specification

[Demi Marie Obenour]

[-- Attachment #1.1.1: Type: text/plain, Size: 4243 bytes --]

On 7/9/25 05:34, Albert Esteve wrote:
> New attempt of including virtio-media
> device specification.
> 
> v8->v9:
> - Rebased to virtio-1.4
> 
> v7->v8:
> - Merged DQBUF and DQEVENT into the same
>   item for unsupported ioctls
> - Rewrote/clarified a couple paragraphs
>   based on previous version reviews.
> 
> Virtio-media came from a discussion on virtio-dev
> mailing list, which lead to Alex Courbot presenting
> virtio-v4l2[1] specification as an alternative to
> virtio-video.
> 
> Later, virtio-v4l2 was renamed to virtio-media[2]
> and published at:
> 
> https://github.com/chromeos/virtio-media
> 
> The repository above includes a virtio-media driver able
> to pass v4l2-compliance when proxying the vivid/vicodec
> virtual devices or an actual UVC camera using the
> V4L2 vhost device (available in the repository).
> It also includes a FFmpeg-based video encoder
> device. Steps to reproduce are also detailed[3].
> 
> Recently, virtio-media has landed in AOSP[4].
> Also the driver patch has been sent to the
> kernel, currently in its v2 [5].
> 
> Furthermore, virtio-media got a proposal to reserve
> device ID 48, which was finally approved for
> inclusion in v1.4.
> 
> There is some overlap with virtio-video in regards
> to which devices it can handle. However, they take
> different approaches, potentially making them
> the preferable choice for different scenarios.
> Furthermore, virtio-media has matured since it
> was first presented, and there are public efforts
> to have upstream driver and devices.
> 
> But mainly, given that virtio-media will be the virtualization
> solution for media devices for ChromeOS, Android, and
> possibly others, I think is important that it gets
> standardized and included in the specification,
> despite the aforementioned overlap.
> 
> Full PDF: https://drive.google.com/file/d/1sDZXWVlvPbHKKguhekCXMzkpNZZ0g3hV/view?usp=sharing
> PDF with the media section only: https://drive.google.com/file/d/1JjOIzecnkFFlpFBRjPQEUyAOD3BDHAdw/view?usp=sharing
> 
> [1] https://mail.google.com/mail/u/0?ui=2&ik=73ebd65ebd&attid=0.1&permmsgid=msg-f:1767388565327924962&th=1887068940754ee2&view=att&disp=inline&realattid=f_libalimc0
> [2] https://www.mail-archive.com/virtio-dev@lists.oasis-open.org/msg12665.html
> [3] https://github.com/chromeos/virtio-media/blob/main/TRY_IT_OUT.md
> [4] https://cs.android.com/android/platform/superproject/main/+/main:external/virtio-media/
> [5] https://lore.kernel.org/lkml/20250201-virtio-media-v2-1-ac840681452d@gmail.com/T/
> 
> Albert Esteve (1):
>   virtio-media: Add virtio media device specification
> 
>  conformance.tex                           |   6 +
>  content.tex                               |   1 +
>  device-types/media/description.tex        | 639 ++++++++++++++++++++++
>  device-types/media/device-conformance.tex |  15 +
>  device-types/media/driver-conformance.tex |  11 +
>  5 files changed, 672 insertions(+)
>  create mode 100644 device-types/media/description.tex
>  create mode 100644 device-types/media/device-conformance.tex
>  create mode 100644 device-types/media/driver-conformance.tex

It's probably too late, but I have some serious concerns about this
device.  Specifically:

1. I don't see a reasonable way to support untrusted virtio-media
   devices.  v4l2 has too many ioctls and I can't realistically
   see a way to enforce that the return values from all of them
   are consistent with each other.  It is possible to only allow
   a fixed format (such as uncompressed ARGB) and only allow the
   video source and/or sink to provide resolutions and frame rates,
   but this means losing a lot of performance and abandoning any
   attempt at zero-copy.

   The use-case for untrusted virtio-media devices is not
   confidential computing, but rather disaggregated systems where
   video sources (like webcams) may be attached to untrusted virtual
   machines.

2. v4l2 is a chatty protocol and using it implies a lot of
   guest <=> host round-trips.  This is bad for performance.
   Is this overhead actually significant, and if so, are there
   plans to reduce it?
-- 
Sincerely,
Demi Marie Obenour (she/her/hers)

[-- Attachment #1.1.2: OpenPGP public key --]
[-- Type: application/pgp-keys, Size: 7253 bytes --]

[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

Re: [PATCH v9 0/1] virtio-media: Add device specification

[Demi Marie Obenour]

[-- Attachment #1.1.1: Type: text/plain, Size: 3183 bytes --]

On 7/9/25 13:58, Nicolas Dufresne wrote:
> Hi Demi,
> 
> Le mercredi 09 juillet 2025 à 13:26 -0400, Demi Marie Obenour a écrit :
>> On 7/9/25 05:34, Albert Esteve wrote:
>>>
> [...]
> 
>>
>> It's probably too late, but I have some serious concerns about this
>> device.  Specifically:
>>
>> 1. I don't see a reasonable way to support untrusted virtio-media
>>    devices.  v4l2 has too many ioctls and I can't realistically
>>    see a way to enforce that the return values from all of them
>>    are consistent with each other.  It is possible to only allow
>>    a fixed format (such as uncompressed ARGB) and only allow the
>>    video source and/or sink to provide resolutions and frame rates,
>>    but this means losing a lot of performance and abandoning any
>>    attempt at zero-copy.
>>
>>    The use-case for untrusted virtio-media devices is not
>>    confidential computing, but rather disaggregated systems where
>>    video sources (like webcams) may be attached to untrusted virtual
>>    machines.
> 
> Please consider that you are posting broadly and understanding of what you mean
> by untrusted virtio-media can be difficult for non-virtio devs like me.

My use-case is this:

1. There is a VM that has all USB devices.
2. Some of those USB devices, like untrusted webcams, may be
   passed through to another VM via usbip.
3. The VM with the usbip device provides a video device to yet another VM.

The VM with the webcam and the VM receiving the video stream do not trust
each other.

> With the enclosed information, I believe implementer are not forced to map to a
> host V4L2 driver. You can implement a userspace layer in between, and possibly
> even implement reclaiming by simulation some hot-unplug of the devices, or by
> not scheduling any work for M2M.

I expect that any implementation that cares about security is going to do this.

> I don't really see the relation between "trust" and the pixel format selection.
> Nor why preventing zero-copy would help handling untrusted virtual machines. I
> do believe you should not attempt this if you don't have the required knowledge
> though, as its unlikely the end-result will be safe.

How can I validate the various parameters provided by a backend?

> For me the main issue for untrusted VM is the lack of control measure, such as
> cgroup for resource usage. This is being worked on though, Maxime Ripard have
> given an update at the Linux Media Summit in Nice.

Denial of service is not a concern in my use-cases.

>> 2. v4l2 is a chatty protocol and using it implies a lot of
>>    guest <=> host round-trips.  This is bad for performance.
>>    Is this overhead actually significant, and if so, are there
>>    plans to reduce it?
> 
> You should try and demonstrate this. Virtio Video tried to make their own and
> endup with something kind of similar with about the same level of back and
> forth. We already had few concerns with ioctl() overhead, and some measure have
> been implemented, notably s_ctrl_ext.
I'll leave this for later.
-- 
Sincerely,
Demi Marie Obenour (she/her/hers)

[-- Attachment #1.1.2: OpenPGP public key --]
[-- Type: application/pgp-keys, Size: 7253 bytes --]

[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 833 bytes --]