[RFC 0/7] Promote GuC ABI headers to shared location

John Harrison john.c.harrison at intel.com
Wed Jun 12 21:18:04 UTC 2024


On 6/11/2024 14:45, Michal Wajdeczko wrote:
> On 11.06.2024 22:32, John Harrison wrote:
>> On 6/11/2024 07:30, Michal Wajdeczko wrote:
>>> There are many GuC ABI definitions named in the same way by the i915
>>> and Xe drivers, preventing proper generation of the documentation.
>>>
>>> Promote GuC ABI definitions to shared location that can be used by
>>> both drivers and can be included in documentation.
>> I still very strongly feel that this is the wrong place for such
>> documentation. We do not document any of the hardware registers in the
>> driver, nor the MI_ instructions, etc. Why is this any different? The
>> GuC ABI is not under the control of the Linux kernel driver, either i915
>> or Xe. It is effectively a hardware interface no different to any other
>> hardware interface. It is already fully documented by the owners of that
>> interface. Rather than just copying random chunks of that documentation
>> into the kernel driver, we should just be publishing the official
>> document itself. Same as we do for the rest of the hardware.
> so go publish this official document
My point is that we should be putting effort into making this happen. So 
far as I know, I am the only person that ever raises this topic. And I 
do keep raising it only to be told it is too low a priority. If more 
people actually pushed for it then maybe it would happen.

>
> in the meantime IMO it is useful to show, with really little effort, on
> what grounds the communication between i915/Xe and GuC works, so
> everyone, not just selected engineers, can understand and review our
> implementation and check its correctness
Simply providing a half sentence description of a bitfield definition 
does not tell you anything about the correctness or usage of an 
interface. Sure, you can tell that it is syntactically correct but the 
compiler can do that for you with the appropriate FIELD definitions. It 
does not tell you anything about whether you are using the interface 
properly or even if this is the proper interface to use. You need the 
full descriptive documentation for that. Are you going to copy all of 
that into the driver? Then you might as well just publish the document 
you are copying from and be done.

>
> furthermore, if you don't like any hw documentation then we should
> revisit what is already in gpu/i915 section and also reconsider all our
> efforts to document all non-static driver functions, as those functions
> are still internal to the driver, not to be used outside
I do not know what you are arguing for here.

Xe in general is being pushed to have much better documentation of the 
code than i915 did. But documentation of the driver is a very different 
topic to documentation of the hardware.

John.

>
>> John.
>>
>>
>>> Cc: Jani Nikula <jani.nikula at intel.com>
>>> Cc: Rodrigo Vivi <rodrigo.vivi at intel.com>
>>> Cc: Lucas De Marchi <lucas.demarchi at intel.com>
>>> Cc: Matthew Brost <matthew.brost at intel.com>
>>> Cc: Daniele Ceraolo Spurio <daniele.ceraolospurio at intel.com>
>>> Cc: John Harrison <John.C.Harrison at Intel.com>
>>>
>>> Michal Wajdeczko (7):
>>>     drm/xe/guc: Promote GuC ABI headers to shared location
>>>     Documentation/gpu: Separate GuC ABI section
>>>     Documentation/gpu: Switch to shared GuC ABI definitions
>>>     drm/intel/guc: Update CTB communication ABI
>>>     drm/intel/guc: Add new KLV definitions
>>>     drm/i915: Use shared GuC ABI definitions
>>>     drm/xe: Promote SR-IOV GuC ABI definitions to shared location
>>>
>>>    Documentation/gpu/drivers.rst                 |   1 +
>>>    Documentation/gpu/guc.rst                     |  23 ++
>>>    Documentation/gpu/i915.rst                    |   9 -
>>>    drivers/gpu/drm/i915/Makefile                 |   5 +
>>>    .../gt/uc/abi/guc_communication_ctb_abi.h     | 170 -----------
>>>    .../gt/uc/abi/guc_communication_mmio_abi.h    |  49 ----
>>>    drivers/gpu/drm/i915/gt/uc/abi/guc_klvs_abi.h | 112 --------
>>>    .../gpu/drm/i915/gt/uc/abi/guc_messages_abi.h | 264 ------------------
>>>    .../guc}/abi/guc_actions_sriov_abi.h          |   0
>>>    .../guc}/abi/guc_communication_ctb_abi.h      |   2 +
>>>    .../guc}/abi/guc_communication_mmio_abi.h     |   0
>>>    .../drm/{xe => intel/guc}/abi/guc_klvs_abi.h  |  18 +-
>>>    .../{xe => intel/guc}/abi/guc_messages_abi.h  |   0
>>>    .../guc}/abi/guc_relay_actions_abi.h          |   0
>>>    .../guc}/abi/guc_relay_communication_abi.h    |   0
>>>    drivers/gpu/drm/xe/Makefile                   |   5 +
>>>    16 files changed, 49 insertions(+), 609 deletions(-)
>>>    create mode 100644 Documentation/gpu/guc.rst
>>>    delete mode 100644
>>> drivers/gpu/drm/i915/gt/uc/abi/guc_communication_ctb_abi.h
>>>    delete mode 100644
>>> drivers/gpu/drm/i915/gt/uc/abi/guc_communication_mmio_abi.h
>>>    delete mode 100644 drivers/gpu/drm/i915/gt/uc/abi/guc_klvs_abi.h
>>>    delete mode 100644 drivers/gpu/drm/i915/gt/uc/abi/guc_messages_abi.h
>>>    rename drivers/gpu/drm/{xe => intel/guc}/abi/guc_actions_sriov_abi.h
>>> (100%)
>>>    rename drivers/gpu/drm/{xe =>
>>> intel/guc}/abi/guc_communication_ctb_abi.h (98%)
>>>    rename drivers/gpu/drm/{xe =>
>>> intel/guc}/abi/guc_communication_mmio_abi.h (100%)
>>>    rename drivers/gpu/drm/{xe => intel/guc}/abi/guc_klvs_abi.h (97%)
>>>    rename drivers/gpu/drm/{xe => intel/guc}/abi/guc_messages_abi.h (100%)
>>>    rename drivers/gpu/drm/{xe => intel/guc}/abi/guc_relay_actions_abi.h
>>> (100%)
>>>    rename drivers/gpu/drm/{xe =>
>>> intel/guc}/abi/guc_relay_communication_abi.h (100%)
>>>



More information about the Intel-gfx mailing list