RE: [EXT] Re: [PATCH v4 2/2] dt-bindings: media: imx-jpeg: Assign slot for imx jpeg encoder/decoder

[Ming Qian]

>From: Conor Dooley <conor@xxxxxxxxxx>
>Sent: 2023年9月26日 21:26
>To: Ming Qian <ming.qian@xxxxxxx>
>Cc: Mirela Rabulea (OSS) <mirela.rabulea@xxxxxxxxxxx>;
>robh+dt@xxxxxxxxxx; shawnguo@xxxxxxxxxx;
>krzysztof.kozlowski+dt@xxxxxxxxxx; conor+dt@xxxxxxxxxx;
>mchehab@xxxxxxxxxx; hverkuil-cisco@xxxxxxxxx; s.hauer@xxxxxxxxxxxxxx;
>kernel@xxxxxxxxxxxxxx; festevam@xxxxxxxxx; X.H. Bao
><xiahong.bao@xxxxxxx>; Eagle Zhou <eagle.zhou@xxxxxxx>; Tao Jiang
><tao.jiang_2@xxxxxxx>; dl-linux-imx <linux-imx@xxxxxxx>;
>devicetree@xxxxxxxxxxxxxxx; linux-media@xxxxxxxxxxxxxxx; linux-
>kernel@xxxxxxxxxxxxxxx; linux-arm-kernel@xxxxxxxxxxxxxxxxxxx
>Subject: [EXT] Re: [PATCH v4 2/2] dt-bindings: media: imx-jpeg: Assign slot for
>imx jpeg encoder/decoder
>
>Hi,
>
>On Tue, Sep 26, 2023 at 06:10:00PM +0800, Ming Qian wrote:
>> There are total 4 slots available in the IP, and we only need to use
>> one slot in one os, assign a single slot, configure interrupt and
>> power domain only for 1 slot, not for the all 4 slots.
>>
>> Signed-off-by: Ming Qian <ming.qian@xxxxxxx>
>> ---
>> v4
>> - improve commit message
>
>> - don't make an ABI break
>
>What does this mean? Can you please try to explain things a bit more clearly
>in your changelogs?
>
>Also, where is the code that actually makes use of these properties?

In v3 patch, I make this property required, make it an ABI break, so in v4, I remove it from required, and default to the previous behavior if it is missing.

The code patch is sent before, but the dts change is not applicable, so I send it separately. The code patch link is as below:
https://patchwork.linuxtv.org/project/linux-media/patch/cdadb4a23697fdc97def958c69b12cd00f547212.1685430841.git.ming.qian@xxxxxxx/

But in the patch, the property name is "slot", not "nxp,slot", I will make another patch to fix the property name after this patch is reviewed.

>
>> v3
>> - add vender prefix, change property slot to nxp,slot
>> - add type for property slot
>>
>>  .../bindings/media/nxp,imx8-jpeg.yaml         | 45 +++++++++----------
>>  1 file changed, 21 insertions(+), 24 deletions(-)
>>
>> diff --git
>> a/Documentation/devicetree/bindings/media/nxp,imx8-jpeg.yaml
>> b/Documentation/devicetree/bindings/media/nxp,imx8-jpeg.yaml
>> index 3d9d1db37040..4bcfc815c894 100644
>> --- a/Documentation/devicetree/bindings/media/nxp,imx8-jpeg.yaml
>> +++ b/Documentation/devicetree/bindings/media/nxp,imx8-jpeg.yaml
>> @@ -32,19 +32,26 @@ properties:
>>      maxItems: 1
>>
>>    interrupts:
>> -    description: |
>> -      There are 4 slots available in the IP, which the driver may use
>> -      If a certain slot is used, it should have an associated interrupt
>> -      The interrupt with index i is assumed to be for slot i
>> -    minItems: 1               # At least one slot is needed by the driver
>> -    maxItems: 4               # The IP has 4 slots available for use
>> +    description:
>> +      Interrupt number for slot
>> +    maxItems: 1
>
>So, what happens to users that already exist that have 2, 3 or 4 interrupts?
>Your commit message says:

And currently the answer is that is inappropriate behavior,  but there has been no real negative impact. As we now have no practical application to use other slots, now we just want to make all the 4 slots configurable and workable.

>> There are total 4 slots available in the IP, and we only need to use
>> one slot in one os, assign a single slot, configure interrupt and
>> power domain only for 1 slot, not for the all 4 slots.
>
>Note that "only need to use one" doesn't mean that a user might not want to
>use more than one. Is it possible to use more than one slot?
>
>Also, if the interrupt numbers map 1:1 to slots, as the previous binding states,
>why do you even need this new "nxp,slot" property?
>
>Thanks,
>Conor.
>

The purpose of 4 slots is to support multi vm or os, Multiple slots are time-sharing the same jpeg engine, there is no benefit to enabling multiple slots in one os, it won't improve the performance, but it just adds complexity to the code.
So we'd like to only use 1 slot in linux, Then, theoretically, we can support up to 3 VMS.

In previous implementation, the dts configured all the interrupts and power-domains for 4 slots, but in driver code, only slot 0 is used, the result is the slot 1,2,3 are not usable, otherwise it will cause conflict

If someone has to enable 2 slots in linux, he can define 2 device node with different slot index in dts.  But this is not what we expect.

This IP defined separate interrupt number for each slot, but it doesn't define separate register region for each slot. It simply defines a set of registers for slots, so driver need the slot index to configure register correctly.
This IP is reused on our other platform, so the interrupt number is not a fixed value for a certain slot. So I think it's not a good way to convert the interrupt number to slot index.

Best regards,
Ming


>>
>>    power-domains:
>>      description:
>>        List of phandle and PM domain specifier as documented in
>>        Documentation/devicetree/bindings/power/power_domain.txt
>> -    minItems: 2               # Wrapper and 1 slot
>> -    maxItems: 5               # Wrapper and 4 slots
>> +    minItems: 1               # VPUMIX
>> +    maxItems: 2               # Wrapper and 1 slot
>> +
>> +  nxp,slot:
>> +    $ref: /schemas/types.yaml#/definitions/uint32
>> +    description:
>> +      Integer number of slot index used. There are 4 slots available in the IP,
>> +      and driver can use a certain slot, it should have an associated interrupt
>> +      and power-domain. In theory, it supports 4 os or vm. If not specified, 0
>> +      is used by default.
>> +    minimum: 0
>> +    maximum: 3
>>
>>  required:
>>    - compatible
>> @@ -62,28 +69,18 @@ examples:
>>      jpegdec: jpegdec@58400000 {
>>          compatible = "nxp,imx8qxp-jpgdec";
>>          reg = <0x58400000 0x00050000 >;
>> -        interrupts = <GIC_SPI 309 IRQ_TYPE_LEVEL_HIGH>,
>> -                     <GIC_SPI 310 IRQ_TYPE_LEVEL_HIGH>,
>> -                     <GIC_SPI 311 IRQ_TYPE_LEVEL_HIGH>,
>> -                     <GIC_SPI 312 IRQ_TYPE_LEVEL_HIGH>;
>> +        interrupts = <GIC_SPI 309 IRQ_TYPE_LEVEL_HIGH>;
>>          power-domains = <&pd IMX_SC_R_MJPEG_DEC_MP>,
>> -                        <&pd IMX_SC_R_MJPEG_DEC_S0>,
>> -                        <&pd IMX_SC_R_MJPEG_DEC_S1>,
>> -                        <&pd IMX_SC_R_MJPEG_DEC_S2>,
>> -                        <&pd IMX_SC_R_MJPEG_DEC_S3>;
>> +                        <&pd IMX_SC_R_MJPEG_DEC_S0>;
>> +        nxp,slot = <0>;
>>      };
>>
>>      jpegenc: jpegenc@58450000 {
>>          compatible = "nxp,imx8qm-jpgenc", "nxp,imx8qxp-jpgenc";
>>          reg = <0x58450000 0x00050000 >;
>> -        interrupts = <GIC_SPI 305 IRQ_TYPE_LEVEL_HIGH>,
>> -                     <GIC_SPI 306 IRQ_TYPE_LEVEL_HIGH>,
>> -                     <GIC_SPI 307 IRQ_TYPE_LEVEL_HIGH>,
>> -                     <GIC_SPI 308 IRQ_TYPE_LEVEL_HIGH>;
>> +        interrupts = <GIC_SPI 305 IRQ_TYPE_LEVEL_HIGH>;
>>          power-domains = <&pd IMX_SC_R_MJPEG_ENC_MP>,
>> -                        <&pd IMX_SC_R_MJPEG_ENC_S0>,
>> -                        <&pd IMX_SC_R_MJPEG_ENC_S1>,
>> -                        <&pd IMX_SC_R_MJPEG_ENC_S2>,
>> -                        <&pd IMX_SC_R_MJPEG_ENC_S3>;
>> +                        <&pd IMX_SC_R_MJPEG_ENC_S0>;
>> +        nxp,slot = <0>;
>>      };
>>  ...
>> --
>> 2.38.1
>>