Re: [PATCH RFC v5 0/8] media: qcom: iris: re-organize catalog & add support for SM8650

[Vikash Garodia]

On 4/15/2025 5:25 PM, neil.armstrong@xxxxxxxxxx wrote:
> Hi,
> 
> On 15/04/2025 13:26, Vikash Garodia wrote:
>>
>> On 4/15/2025 1:54 PM, neil.armstrong@xxxxxxxxxx wrote:
>>> Hi,
>>>
>>> On 14/04/2025 21:48, Vikash Garodia wrote:
>>>>
>>>> On 4/14/2025 5:39 PM, neil.armstrong@xxxxxxxxxx wrote:
>>>>> Hi,
>>>>>
>>>>> On 14/04/2025 12:54, Vikash Garodia wrote:
>>>>>> Hi Neil,
>>>>>>
>>>>>> On 4/14/2025 1:05 PM, Neil Armstrong wrote:
>>>>>>> Hi Vikash, Dikshita,
>>>>>>>
>>>>>>> On 10/04/2025 18:29, Neil Armstrong wrote:
>>>>>>>> Re-organize the platform support core into a gen1 catalog C file
>>>>>>>> declaring common platform structure and include platform headers
>>>>>>>> containing platform specific entries and iris_platform_data
>>>>>>>> structure.
>>>>>>>>
>>>>>>>> The goal is to share most of the structure while having
>>>>>>>> clear and separate per-SoC catalog files.
>>>>>>>>
>>>>>>>> The organization is based on the curent drm/msm dpu1 catalog
>>>>>>>> entries.
>>>>>>>
>>>>>>> Any feedback on this patchset ?
>>>>>> Myself and Dikshita went through the approach you are bringing here, let me
>>>>>> update some context here:
>>>>>> - sm8550, sm8650, sm8775p, qcs8300 are all irisv3, while qcs8300 is the
>>>>>> scaled
>>>>>> down variant i.e have 2 PIPE vs others having 4. Similarly there are other
>>>>>> irisv3 having 1 pipe as well.
>>>>>> - With above variations, firmware and instance caps would change for the
>>>>>> variant
>>>>>> SOCs.
>>>>>> - Above these, few(less) bindings/connections specific delta would be there,
>>>>>> like there is reset delta in sm8550 and sm8650.
>>>>>>
>>>>>> Given above, xxx_gen1.c and xxx_gen2.c can have all binding specific
>>>>>> tables and
>>>>>> SOC platform data, i.e sm8650_data (for sm8650). On top of this,
>>>>>> individual SOC
>>>>>> specific .c file can have any delta, from xxx_gen1/2.c) like reset table or
>>>>>> preset register table, etc and export these delta structs in xxx_gen1.c or
>>>>>> xxx_gen2.c.
>>>>>>
>>>>>> Going with above approach, sm8650.c would have only one reset table for now.
>>>>>> Later if any delta is identified, the same can be added in it. All other
>>>>>> common
>>>>>> structs, can reside in xxx_gen2.c for now.
>>>>>
>>>>> Thanks for reviewing, but...
>>>>> Sorry I don't understand what you and Dmitry are asking me...
>>>>>
>>>>> If I try really hard, you would like to have:
>>>>>
>>>>> iris_catalog_sm8550.c
>>>>> - iris_set_sm8550_preset_registers
>>>>> - sm8550_icc_table
>>>>> - sm8550_clk_reset_table
>>>>> - sm8550_bw_table_dec
>>>>> - sm8550_pmdomain_table
>>>>> - sm8550_opp_pd_table
>>>>> - sm8550_clk_table
>>>> Move or rename existing 8550.c as xxx_gen2.c. This is with the existing
>>>> assumption that everything under 8550.c is common for all gen2 to come in
>>>> future.
>>>>>
>>>>> iris_catalog_sm8650.c
>>>>> - sm8650_clk_reset_table
>>>>> - sm8650_controller_reset_table
>>>> yes, since reset is the only delta.
>>>>>
>>>>> iris_catalog_gen2.c
>>>>> - iris_hfi_gen2_command_ops_init
>>>>> - iris_hfi_gen2_response_ops_init
>>>>> ...
>>>>> - sm8550_dec_op_int_buf_tbl
>>>>>
>>>>> and:
>>>>> - struct iris_platform_data sm8550_data
>>>>> - struct iris_platform_data sm8650_data
>>>> all this goes to xxx_gen2.c as well.
>>>
>>> Yeah so this is exactly my current approach, except it use .h files
>>> for each SoC for simplicity.
>>>
>>>>
>>>>> using data from iris_catalog_sm8550.c & iris_catalog_sm8550.c
>>>>>
>>>>> So this is basically what I _already_ propose except
>>>>> you move data in separate .c files for no reasons,
>>>>> please explain why you absolutely want distinct .c
>>>>> files per SoC. We are no more in the 1990's and we camn
>>>>> defintely have big .c files.
>>>> Its not about the size of file alone, it is easy to understand later what would
>>>> be the delta in the SOCs and what would common. For ex, just navigating through
>>>> sm8650.c, anyone can comment that reset is the delta.
>>>
>>> What's the problem with the current approach with .h file for each SoC ?
>>>
>>>>>
>>>>> And we still have a big issue, how to get the:
>>>>> - ARRAY_SIZE(sm8550_clk_reset_table)
>>>>> - ARRAY_SIZE(sm8550_bw_table_dec)
>>>>> - ARRAY_SIZE(sm8550_pmdomain_table)
>>>>> ...
>>>>>
>>>>> since they are declared in a separate .c file and you
>>>>> need a compile-time const value to fill all the _size
>>>>> attribute in iris_platform_data.
>>>> I have not tries this, but isn't extern-ing the soc structs (in your case reset
>>>> tables) into xxx_gen2.c enough here ? Also i think the tables you are pointing
>>>> here, lies in the xxx_gen2.c only, so i am sure above ones would not be an
>>>> issue
>>>> at all. The only delta struct is reset table, lets see if extern helps.
>>>
>>> No it doesn't, because I wrote C for the last 25 years, and I tried it already,
>>> I also tried to export a const int with the table size, and it also doesn't
>>> work.
>> Got it, i tried too, it didn't work.
>>>
>>> The 3 only ways are:
>>> 1) add defines with sizes for each table
>> This leaves manual update everytime.
>>
>>> 2) add a NULL entry at the end of each table, and update all code using those
>>> tables
>> Does not sound right to update the code, just to get the size.
>>
>>> 3) declare in the same scope, which is my current proposalThe proposal in the
>>> RFC is about moving the common structs to 8550.h, rather, it
>> can be kept in xxx_gen2.c.
>> 8550.h can have the delta part (i.e reset tables) and can be included in
>> xxx_gen2.c. sm8650_data can reside in xxx_gen2.c, soc headers just brings the
>> delta structs which can be overridden from common in xxx_gen2.c
>> I am good with the header approach which contains the delta over and above
>> xxx_gen2.c.
> 
> I'll try to do that, but now I don't see the point of the SoC header files if
> they only contain the reset tables.
We already know that preset registers are also coming in (it is still a mystery
though on why h264 dec works without it, i have not got chance to explore it
more), so it would be useful when you extend it later.

Regards,
Vikash
> 
> Neil
> 
>>
>> Regards,
>> Vikash
>>> Neil
>>>
>>>>
>>>> Regards,
>>>> Vikash
>>>>>
>>>>> So I recall my goal, I just want to add sm8650 support,
>>>>> and I'm not the owner of this driver, and I'm really happy
>>>>> to help, but giving me random ideas to solve your problem
>>>>> doesn't help us at all going forward.
>>>>>
>>>>> Neil
>>>>>
>>>>>>
>>>>>> Regards,
>>>>>> Vikash
>>>>>>>
>>>>>>> Thanks,
>>>>>>> Neil
>>>>>>>
>>>>>>>>
>>>>>>>> Add support for the IRIS accelerator for the SM8650
>>>>>>>> platform, which uses the iris33 hardware.
>>>>>>>>
>>>>>>>> The vpu33 requires a different reset & poweroff sequence
>>>>>>>> in order to properly get out of runtime suspend.
>>>>>>>>
>>>>>>>> Follow-up of [1]:
>>>>>>>> https://lore.kernel.org/all/20250409-topic-sm8x50-iris-v10-v4-0-40e411594285@xxxxxxxxxx/
>>>>>>>>
>>>>>>>> Signed-off-by: Neil Armstrong <neil.armstrong@xxxxxxxxxx>
>>>>>>>> ---
>>>>>>>> Changes in v4:
>>>>>>>> - Reorganized into catalog, rebased sm8650 support on top
>>>>>>>> - Link to v4:
>>>>>>>> https://lore.kernel.org/all/20250409-topic-sm8x50-iris-v10-v4-0-40e411594285@xxxxxxxxxx
>>>>>>>>
>>>>>>>> Changes in v4:
>>>>>>>> - collected tags
>>>>>>>> - un-split power_off in vpu3x
>>>>>>>> - removed useless function defines
>>>>>>>> - added back vpu3x disappeared rename commit
>>>>>>>> - Link to v3:
>>>>>>>> https://lore.kernel.org/r/20250407-topic-sm8x50-iris-v10-v3-0-63569f6d04aa@xxxxxxxxxx
>>>>>>>>
>>>>>>>> Changes in v3:
>>>>>>>> - Collected review tags
>>>>>>>> - Removed bulky reset_controller ops
>>>>>>>> - Removed iris_vpu_power_off_controller split
>>>>>>>> - Link to v2:
>>>>>>>> https://lore.kernel.org/r/20250305-topic-sm8x50-iris-v10-v2-0-bd65a3fc099e@xxxxxxxxxx
>>>>>>>>
>>>>>>>> Changes in v2:
>>>>>>>> - Collected bindings review
>>>>>>>> - Reworked rest handling by adding a secondary optional table to be used by
>>>>>>>> controller poweroff
>>>>>>>> - Reworked power_off_controller to be reused and extended by vpu33 support
>>>>>>>> - Removed useless and unneeded vpu33 init
>>>>>>>> - Moved vpu33 into vpu3x files to reuse code from vpu3
>>>>>>>> - Moved sm8650 data table into sm8550
>>>>>>>> - Link to v1:
>>>>>>>> https://lore.kernel.org/r/20250225-topic-sm8x50-iris-v10-v1-0-128ef05d9665@xxxxxxxxxx
>>>>>>>>
>>>>>>>> ---
>>>>>>>> Neil Armstrong (8):
>>>>>>>>           media: qcom: iris: move sm8250 to gen1 catalog
>>>>>>>>           media: qcom: iris: move sm8550 to gen2 catalog
>>>>>>>>           dt-bindings: media: qcom,sm8550-iris: document SM8650 IRIS
>>>>>>>> accelerator
>>>>>>>>           media: platform: qcom/iris: add power_off_controller to vpu_ops
>>>>>>>>           media: platform: qcom/iris: introduce optional controller_rst_tbl
>>>>>>>>           media: platform: qcom/iris: rename iris_vpu3 to iris_vpu3x
>>>>>>>>           media: platform: qcom/iris: add support for vpu33
>>>>>>>>           media: platform: qcom/iris: add sm8650 support
>>>>>>>>
>>>>>>>>      .../bindings/media/qcom,sm8550-iris.yaml           |  33 ++-
>>>>>>>>      drivers/media/platform/qcom/iris/Makefile          |   6 +-
>>>>>>>>      .../media/platform/qcom/iris/iris_catalog_gen1.c   |  83 +++++++
>>>>>>>>      ...{iris_platform_sm8550.c => iris_catalog_gen2.c} |  85 +------
>>>>>>>>      ...ris_platform_sm8250.c => iris_catalog_sm8250.h} |  80 +-----
>>>>>>>>      .../media/platform/qcom/iris/iris_catalog_sm8550.h |  91 +++++++
>>>>>>>>      .../media/platform/qcom/iris/iris_catalog_sm8650.h |  68 +++++
>>>>>>>>      drivers/media/platform/qcom/iris/iris_core.h       |   1 +
>>>>>>>>      .../platform/qcom/iris/iris_platform_common.h      |   3 +
>>>>>>>>      drivers/media/platform/qcom/iris/iris_probe.c      |  43 +++-
>>>>>>>>      drivers/media/platform/qcom/iris/iris_vpu2.c       |   1 +
>>>>>>>>      drivers/media/platform/qcom/iris/iris_vpu3.c       | 122 ---------
>>>>>>>>      drivers/media/platform/qcom/iris/iris_vpu3x.c      | 275
>>>>>>>> +++++++++++++++++++++
>>>>>>>>      drivers/media/platform/qcom/iris/iris_vpu_common.c |   4 +-
>>>>>>>>      drivers/media/platform/qcom/iris/iris_vpu_common.h |   3 +
>>>>>>>>      15 files changed, 598 insertions(+), 300 deletions(-)
>>>>>>>> ---
>>>>>>>> base-commit: 2bdde620f7f2bff2ff1cb7dc166859eaa0c78a7c
>>>>>>>> change-id: 20250410-topic-sm8x50-upstream-iris-catalog-3e2e4a033d6f
>>>>>>>>
>>>>>>>> Best regards,
>>>>>>>
>>>>>
>>>
>