mesa-drm/include/drm
Arvind Yadav 72d3140428 amdgpu: UAPI for AMDGPU usermode queues
This patch adds UAPI interface changes for AMDGPU usermode
queues, semaphore and new AMDGPU GEM domain for doorbells.
Usermode queues allow a userspace process to create
and submit its graphics/compute/sdma work directly to the GPU.

v2:(Marek)
 - Add csa support for SDMA queue.
 - Rename UAPI objects and struct as per UAPI review. (Shashank)

v3:(Yogesh)
 - Rename UAPI timeline* objects as per UAPI review. (Arvind)

v4: (Marek)
 - Drop AMDGPU_USERQ_BO_WRITE as this should not be a global option
   of the IOCTL, It should be option per buffer. Hence adding separate
   array for read and write BO handles. (Arun)
 - Modify num_fences to __u16, flags changed to __u16 and placed
   the num_fences next to flags for optimal padding and size. (Arun)

 - Fix 32-bit holes issue in sturct drm_amdgpu_gem_va as per
   UAPI review (Arvind).

v5:(Marek/Pierre-Eric)
  - add more detail params description for signal and wait IOCTL calls.
  - Remove the unused structure fields in signal and wait structs.
  - Add separate array of read and write for BO handles. (Arun)

  - Removes the unused flags parameter from the usermode queue UAPI structure
  - Clarify comments on top of drm_amdgpu_userq_in
  - Clarify comment for queue_id (in)
  - Clarify comment for mqd
  - Clarify comment for compute MQD size
  - Clarify comment for queue_id (out)
  - Adds padding variables in userqueue in/out structures. (Shashank)

v6:(Pierre-Eric)
  - Modify the function parameter names and struct
    field names as per the review comments. (Arun)

v7:(Marek)
  - Modify the structure field name and comments. (Arun)

  - Rename vm_timeline_syncobj and add comment for
    vm_timeline_point.
  - Remove GDS buffer support from MQD. (Arvind)

v8:(Pierre-Eric)
  - Modify the function parameter names.

Cc: Koenig, Christian <christian.koenig@amd.com>
Cc: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Signed-off-by: Shashank Sharma <shashank.sharma@amd.com>
Signed-off-by: Arvind Yadav <arvind.yadav@amd.com>
Signed-off-by: Arunpravin Paneer Selvam <Arunpravin.PaneerSelvam@amd.com>
2025-04-10 10:03:07 -04:00
..
amdgpu_drm.h amdgpu: UAPI for AMDGPU usermode queues 2025-04-10 10:03:07 -04:00
drm.h Sync headers with drm-next 2024-01-13 10:33:05 +01:00
drm_fourcc.h Sync headers with drm-next 2024-06-26 08:08:40 +00:00
drm_mode.h Sync headers with drm-next 2024-06-26 08:08:40 +00:00
drm_sarea.h headers: sync drm_sarea.h with airlied/drm-next 2017-06-03 23:39:22 +01:00
i915_drm.h include/drm/i915_drm.h: Update from Linux v6.0-rc7 2022-09-28 17:04:12 -07:00
mach64_drm.h Put mach64_drm.h back in to avoid breaking mesa build 2009-11-20 20:17:26 -05:00
mga_drm.h headers: Sync up mga_drm.h from drm-next. 2017-11-10 12:10:15 -08:00
msm_drm.h libdrm: add msm drm uapi header 2018-08-23 08:48:45 -07:00
nouveau_drm.h nouveau: sync up with nouveau_abi16.h and nouveau_drm.h 2020-05-06 23:17:48 +02:00
qxl_drm.h libdrm: headers: Sync with drm-next 2018-10-04 14:51:09 +01:00
r128_drm.h headers: Sync up some header guard changes from drm-next. 2017-11-10 12:10:12 -08:00
radeon_drm.h headers: Sync up kernel changes to use kernel types instead of stdint.h. 2017-11-10 12:10:17 -08:00
README include/drm/README: update drm-next link to use gitlab instead of cgit 2024-09-29 17:47:51 -07:00
savage_drm.h headers: Sync up some comment spelling and whitespace fixes from drm-next. 2017-11-10 12:10:13 -08:00
sis_drm.h headers: Sync up some header guard changes from drm-next. 2017-11-10 12:10:12 -08:00
tegra_drm.h tegra: Update for new UABI 2022-02-23 15:39:57 +01:00
vc4_drm.h libdrm: headers: Sync with drm-next 2018-10-04 14:51:09 +01:00
via_drm.h headers: Sync up some header guard changes from drm-next. 2017-11-10 12:10:12 -08:00
virtgpu_drm.h virtgpu: Update kernel header 2019-04-03 20:11:55 -04:00
vmwgfx_drm.h fix various typos 2019-04-17 18:23:25 +01:00

What are these headers ?
------------------------
This is the canonical source of drm headers that user space should use for
communicating with the kernel DRM subsystem.

They flow from the kernel, thus any changes must be merged there first.
Do _not_ attempt to "fix" these by deviating from the kernel ones !


Non-linux platforms - changes/patches
-------------------------------------
If your platform has local changes, please send them upstream for inclusion.
Even if your patches don't get accepted in their current form, devs will
give you feedback on how to address things properly.

git send-email --subject-prefix="PATCH libdrm" your patches to dri-devel
mailing list.

Before doing so, please consider the following:
 - Have the [libdrm vs kernel] headers on your platform deviated ?
Consider unifying them first.

 - Have you introduced additional ABI that's not available in Linux ?
Propose it for [Linux kernel] upstream inclusion.
If that doesn't work out (hopefully it never does), move it to another header
and/or keep the change(s) local ?

 - Are your changes DRI1/UMS specific ?
There is virtually no interest/power in keeping those legacy interfaces. They
are around due to the kernel "thou shalt not break existing user space" rule.

Consider porting the driver to DRI2/KMS - all (almost?) sensible hardware is
capable of supporting those.


Which headers go where ?
------------------------
A snipped from the, now removed, Makefile.am used to state:

  XXX airlied says, nothing besides *_drm.h and drm*.h should be necessary.
  however, r300 and via need their reg headers installed in order to build.
  better solutions are welcome.

Obviously the r300 and via headers are no longer around ;-)

Reason behind is that the drm headers can be used as a basic communications
channel with the respective kernel modules. If more advanced functionality is
required one can pull the specific libdrm_$driver which is free to pull
additional files from the kernel.

For example: nouveau has nouveau/nvif/*.h while vc4 has vc4/*.h

If your driver is still in prototyping/staging state, consider moving the
$driver_drm.h into $driver and _not_ installing it. An header providing opaque
definitions and access [via $driver_drmif.h or similar] would be better fit.


When and which headers to update
--------------------------------
Ideally all files will be synced (updated) with the latest released kernel on
each libdrm release. Sadly that's not yet possible since quite a few headers
differ significantly - see Outdated or Broken Headers section below.

That said, it's up-to the individual developers to sync with newer version
(from drm-next) as they see fit.


When and how to update these files
----------------------------------
Note: One should not do _any_ changes to the files apart from the steps below.

In order to update the files do the following:
 - Switch to a Linux kernel tree/branch which is not rebased.
   For example: drm-next (https://gitlab.freedesktop.org/drm/kernel/)
 - Install the headers via `make headers_install' to a separate location.
 - Copy the drm header[s] + git add + git commit.
 - Note: Your commit message must include:
   a) Brief summary on the delta. If there's any change that looks like an
API/ABI break one _must_ explicitly state why it's safe to do so.
   b) "Generated using make headers_install."
   c) "Generated from $tree/branch commit $sha"


Outdated or Broken Headers
--------------------------
This section contains a list of headers and the respective "issues" they might
have relative to their kernel equivalent.

Most UMS headers:
 - Not using fixed size integers - compat ioctls are broken.
Status: ?
Promote to fixed size ints, which match the current (32bit) ones.

nouveau_drm.h
 - Missing macros NOUVEAU_GETPARAM*, NOUVEAU_DRM_HEADER_PATCHLEVEL, structs,
enums
Status: Deliberate UABI choice; nouveau hides the exact kernel ABI behind libdrm

r128_drm.h
 - Broken compat ioctls.

radeon_drm.h
 - Missing RADEON_TILING_R600_NO_SCANOUT, CIK_TILE_MODE_*, broken UMS ioctls
 - Both kernel and libdrm: missing padding -
drm_radeon_gem_{create,{g,s}et_tiling,set_domain} others ?
Status: ?

savage_drm.h
 - Renamed ioctls - DRM_IOCTL_SAVAGE_{,BCI}_EVENT_EMIT, compat ioctls are broken.
Status: ?

sis_drm.h
 - Borken ioctls + libdrm uses int vs kernel long
Status: ?

via_drm.h
 - Borken ioctls - libdrm int vs kernel long
Status: ?


omap_drm.h (living in $TOP/omap)
 - License mismatch, missing DRM_IOCTL_OMAP_GEM_NEW and related struct
Status: ?

exynos_drm.h (living in $TOP/exynos)
 - License mismatch, now using fixed size ints (but not everywhere). Lots of
new stuff.
Status: ?