linux/Documentation/driver-api/dma-buf.rst

Buffer Sharing and Synchronization
==================================

The dma-buf subsystem provides the framework for sharing buffers for
hardware (DMA) access across multiple device drivers and subsystems, and
for synchronizing asynchronous hardware access.

This is used, for example, by drm "prime" multi-GPU support, but is of
course not limited to GPU use cases.

The three main components of this are: (1) dma-buf, representing a
sg_table and exposed to userspace as a file descriptor to allow passing
between devices, (2) fence, which provides a mechanism to signal when
one device has finished access, and (3) reservation, which manages the
shared or exclusive fence(s) associated with the buffer.

Shared DMA Buffers
------------------

This document serves as a guide to device-driver writers on what is the dma-buf
buffer sharing API, how to use it for exporting and using shared buffers.

Any device driver which wishes to be a part of DMA buffer sharing, can do so as
either the 'exporter' of buffers, or the 'user' or 'importer' of buffers.

Say a driver A wants to use buffers created by driver B, then we call B as the
exporter, and A as buffer-user/importer.

The exporter

 - implements and manages operations in :c:type:`struct dma_buf_ops
   <dma_buf_ops>` for the buffer,
 - allows other users to share the buffer by using dma_buf sharing APIs,
 - manages the details of buffer allocation, wrapped in a :c:type:`struct
   dma_buf <dma_buf>`,
 - decides about the actual backing storage where this allocation happens,
 - and takes care of any migration of scatterlist - for all (shared) users of
   this buffer.

The buffer-user

 - is one of (many) sharing users of the buffer.
 - doesn't need to worry about how the buffer is allocated, or where.
 - and needs a mechanism to get access to the scatterlist that makes up this
   buffer in memory, mapped into its own address space, so it can access the
   same area of memory. This interface is provided by :c:type:`struct
   dma_buf_attachment <dma_buf_attachment>`.

Any exporters or users of the dma-buf buffer sharing framework must have a
'select DMA_SHARED_BUFFER' in their respective Kconfigs.

Userspace Interface Notes
~~~~~~~~~~~~~~~~~~~~~~~~~

Mostly a DMA buffer file descriptor is simply an opaque object for userspace,
and hence the generic interface exposed is very minimal. There's a few things to
consider though:

- Since kernel 3.12 the dma-buf FD supports the llseek system call, but only
  with offset=0 and whence=SEEK_END|SEEK_SET. SEEK_SET is supported to allow
  the usual size discover pattern size = SEEK_END(0); SEEK_SET(0). Every other
  llseek operation will report -EINVAL.

  If llseek on dma-buf FDs isn't support the kernel will report -ESPIPE for all
  cases. Userspace can use this to detect support for discovering the dma-buf
  size using llseek.

- In order to avoid fd leaks on exec, the FD_CLOEXEC flag must be set
  on the file descriptor.  This is not just a resource leak, but a
  potential security hole.  It could give the newly exec'd application
  access to buffers, via the leaked fd, to which it should otherwise
  not be permitted access.

  The problem with doing this via a separate fcntl() call, versus doing it
  atomically when the fd is created, is that this is inherently racy in a
  multi-threaded app[3].  The issue is made worse when it is library code
  opening/creating the file descriptor, as the application may not even be
  aware of the fd's.

  To avoid this problem, userspace must have a way to request O_CLOEXEC
  flag be set when the dma-buf fd is created.  So any API provided by
  the exporting driver to create a dmabuf fd must provide a way to let
  userspace control setting of O_CLOEXEC flag passed in to dma_buf_fd().

- Memory mapping the contents of the DMA buffer is also supported. See the
  discussion below on `CPU Access to DMA Buffer Objects`_ for the full details.

- The DMA buffer FD is also pollable, see `Implicit Fence Poll Support`_ below for
  details.

Basic Operation and Device DMA Access
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

.. kernel-doc:: drivers/dma-buf/dma-buf.c
   :doc: dma buf device access

CPU Access to DMA Buffer Objects
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

.. kernel-doc:: drivers/dma-buf/dma-buf.c
   :doc: cpu access

Implicit Fence Poll Support
~~~~~~~~~~~~~~~~~~~~~~~~~~~

.. kernel-doc:: drivers/dma-buf/dma-buf.c
   :doc: implicit fence polling

Kernel Functions and Structures Reference
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

.. kernel-doc:: drivers/dma-buf/dma-buf.c
   :export:

.. kernel-doc:: include/linux/dma-buf.h
   :internal:

Buffer Mapping Helpers
~~~~~~~~~~~~~~~~~~~~~~

.. kernel-doc:: include/linux/dma-buf-map.h
   :doc: overview

.. kernel-doc:: include/linux/dma-buf-map.h
   :internal:

Reservation Objects
-------------------

.. kernel-doc:: drivers/dma-buf/dma-resv.c
   :doc: Reservation Object Overview

.. kernel-doc:: drivers/dma-buf/dma-resv.c
   :export:

.. kernel-doc:: include/linux/dma-resv.h
   :internal:

DMA Fences
----------

.. kernel-doc:: drivers/dma-buf/dma-fence.c
   :doc: DMA fences overview

DMA Fence Cross-Driver Contract
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

.. kernel-doc:: drivers/dma-buf/dma-fence.c
   :doc: fence cross-driver contract

DMA Fence Signalling Annotations
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

.. kernel-doc:: drivers/dma-buf/dma-fence.c
   :doc: fence signalling annotation

DMA Fences Functions Reference
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

.. kernel-doc:: drivers/dma-buf/dma-fence.c
   :export:

.. kernel-doc:: include/linux/dma-fence.h
   :internal:

Seqno Hardware Fences
~~~~~~~~~~~~~~~~~~~~~

.. kernel-doc:: include/linux/seqno-fence.h
   :internal:

DMA Fence Array
~~~~~~~~~~~~~~~

.. kernel-doc:: drivers/dma-buf/dma-fence-array.c
   :export:

.. kernel-doc:: include/linux/dma-fence-array.h
   :internal:

DMA Fence uABI/Sync File
~~~~~~~~~~~~~~~~~~~~~~~~

.. kernel-doc:: drivers/dma-buf/sync_file.c
   :export:

.. kernel-doc:: include/linux/sync_file.h
   :internal:

Indefinite DMA Fences
~~~~~~~~~~~~~~~~~~~~~

At various times struct dma_fence with an indefinite time until dma_fence_wait()
finishes have been proposed. Examples include:

* Future fences, used in HWC1 to signal when a buffer isn't used by the display
  any longer, and created with the screen update that makes the buffer visible.
  The time this fence completes is entirely under userspace's control.

* Proxy fences, proposed to handle &drm_syncobj for which the fence has not yet
  been set. Used to asynchronously delay command submission.

* Userspace fences or gpu futexes, fine-grained locking within a command buffer
  that userspace uses for synchronization across engines or with the CPU, which
  are then imported as a DMA fence for integration into existing winsys
  protocols.

* Long-running compute command buffers, while still using traditional end of
  batch DMA fences for memory management instead of context preemption DMA
  fences which get reattached when the compute job is rescheduled.

Common to all these schemes is that userspace controls the dependencies of these
fences and controls when they fire. Mixing indefinite fences with normal
in-kernel DMA fences does not work, even when a fallback timeout is included to
protect against malicious userspace:

* Only the kernel knows about all DMA fence dependencies, userspace is not aware
  of dependencies injected due to memory management or scheduler decisions.

* Only userspace knows about all dependencies in indefinite fences and when
  exactly they will complete, the kernel has no visibility.

Furthermore the kernel has to be able to hold up userspace command submission
for memory management needs, which means we must support indefinite fences being
dependent upon DMA fences. If the kernel also support indefinite fences in the
kernel like a DMA fence, like any of the above proposal would, there is the
potential for deadlocks.

.. kernel-render:: DOT
   :alt: Indefinite Fencing Dependency Cycle
   :caption: Indefinite Fencing Dependency Cycle

   digraph "Fencing Cycle" {
      node [shape=box bgcolor=grey style=filled]
      kernel [label="Kernel DMA Fences"]
      userspace [label="userspace controlled fences"]
      kernel -> userspace [label="memory management"]
      userspace -> kernel [label="Future fence, fence proxy, ..."]

      { rank=same; kernel userspace }
   }

This means that the kernel might accidentally create deadlocks
through memory management dependencies which userspace is unaware of, which
randomly hangs workloads until the timeout kicks in. Workloads, which from
userspace's perspective, do not contain a deadlock.  In such a mixed fencing
architecture there is no single entity with knowledge of all dependencies.
Thefore preventing such deadlocks from within the kernel is not possible.

The only solution to avoid dependencies loops is by not allowing indefinite
fences in the kernel. This means:

* No future fences, proxy fences or userspace fences imported as DMA fences,
  with or without a timeout.

* No DMA fences that signal end of batchbuffer for command submission where
  userspace is allowed to use userspace fencing or long running compute
  workloads. This also means no implicit fencing for shared buffers in these
  cases.

Recoverable Hardware Page Faults Implications
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

Modern hardware supports recoverable page faults, which has a lot of
implications for DMA fences.

First, a pending page fault obviously holds up the work that's running on the
accelerator and a memory allocation is usually required to resolve the fault.
But memory allocations are not allowed to gate completion of DMA fences, which
means any workload using recoverable page faults cannot use DMA fences for
synchronization. Synchronization fences controlled by userspace must be used
instead.

On GPUs this poses a problem, because current desktop compositor protocols on
Linux rely on DMA fences, which means without an entirely new userspace stack
built on top of userspace fences, they cannot benefit from recoverable page
faults. Specifically this means implicit synchronization will not be possible.
The exception is when page faults are only used as migration hints and never to
on-demand fill a memory request. For now this means recoverable page
faults on GPUs are limited to pure compute workloads.

Furthermore GPUs usually have shared resources between the 3D rendering and
compute side, like compute units or command submission engines. If both a 3D
job with a DMA fence and a compute workload using recoverable page faults are
pending they could deadlock:

- The 3D workload might need to wait for the compute job to finish and release
  hardware resources first.

- The compute workload might be stuck in a page fault, because the memory
  allocation is waiting for the DMA fence of the 3D workload to complete.

There are a few options to prevent this problem, one of which drivers need to
ensure:

- Compute workloads can always be preempted, even when a page fault is pending
  and not yet repaired. Not all hardware supports this.

- DMA fence workloads and workloads which need page fault handling have
  independent hardware resources to guarantee forward progress. This could be
  achieved through e.g. through dedicated engines and minimal compute unit
  reservations for DMA fence workloads.

- The reservation approach could be further refined by only reserving the
  hardware resources for DMA fence workloads when they are in-flight. This must
  cover the time from when the DMA fence is visible to other threads up to
  moment when fence is completed through dma_fence_signal().

- As a last resort, if the hardware provides no useful reservation mechanics,
  all workloads must be flushed from the GPU when switching between jobs
  requiring DMA fences or jobs requiring page fault handling: This means all DMA
  fences must complete before a compute job with page fault handling can be
  inserted into the scheduler queue. And vice versa, before a DMA fence can be
  made visible anywhere in the system, all compute workloads must be preempted
  to guarantee all pending GPU page faults are flushed.

- Only a fairly theoretical option would be to untangle these dependencies when
  allocating memory to repair hardware page faults, either through separate
  memory blocks or runtime tracking of the full dependency graph of all DMA
  fences. This results very wide impact on the kernel, since resolving the page
  on the CPU side can itself involve a page fault. It is much more feasible and
  robust to limit the impact of handling hardware page faults to the specific
  driver.

Note that workloads that run on independent hardware like copy engines or other
GPUs do not have any impact. This allows us to keep using DMA fences internally
in the kernel even for resolving hardware page faults, e.g. by using copy
engines to clear or copy memory needed to resolve the page fault.

In some ways this page fault problem is a special case of the `Infinite DMA
Fences` discussions: Infinite fences from compute workloads are allowed to
depend on DMA fences, but not the other way around. And not even the page fault
problem is new, because some other CPU thread in userspace might
hit a page fault which holds up a userspace fence - supporting page faults on
GPUs doesn't anything fundamentally new.