Re: [PATCH 12/13] media: qcom: camss: Add CSID Gen3 support for sm8550

[Bryan O'Donoghue <bryan.odonoghue@linaro.org>]

On 15/08/2024 16:14, Depeng Shao wrote:
> Hi Bryan,
> 
> 
>>> ---
>>>   drivers/media/platform/qcom/camss/Makefile    |   1 +
>>>   .../platform/qcom/camss/camss-csid-gen3.c     | 339 ++++++++++++++++++
>>>   .../platform/qcom/camss/camss-csid-gen3.h     |  26 ++
>>
>>
>> So this "gen2" and "gen3" stuff would make sense if we had a number of 
>> SoCs based on gen2 and gen3 which were controlled from the upper-level 
>> gen2.c and gen3.c.
>>
>> What you're submitting here is csid-780 so the file should be named 
>> csid-780.
>>
>> When we add 680 or 880 then it makes sense to try to encapsulate a 
>> class of generation into one file - potentially.
>>
>> I'd guess that was the intent behind gen2.c.
>>
>> TL;DR please name your file csid-xxx.c
> 
> Sure, I will use csid-780.c
> 
>>> +
>>> +    writel(val, csid->base + CSID_CSI2_RX_CFG0);
>>> +
>>> +    val = 1 << CSI2_RX_CFG1_ECC_CORRECTION_EN;
>>> +    if (vc > 3)
>>> +        val |= 1 << CSI2_RX_CFG1_VC_MODE;
>>
>> So again these are needless bit-shifts.
>>
>> #define CSI2_RX_CFG1_PACKET_ECC_CORRECTION_EN BIT(0)
>>
>> val = CSI2_RX_CFG1_PACKET_ECC_CORRECTION_EN;
>>
> 
> You posted same comments in v3 series, I also replied it.
> https://lore.kernel.org/all/eeaf4f4e-5200-4b13-b38f-3f3385fc2a2b@quicinc.com/
> 
> Some of register bits which just need to be configured to 0 or 1, then 
> can use BIT(X), but some register bits need to configure a specific 
> value, e.g.,  CSID_RDI_CFG0 bits[22:26] need to configure a vc vaule, 
> bits[16:21] need to configure a dt value, then we can't use BIT(x) to 
> handle this.

Yes please use macros() to bury any _necessary_ bit shifts to populate a 
_bit_field_ away but as an example CSI2_RX_CFG1_PACKET_ECC_CORRECTION_EN 
is not a bit-field.


> 
> 
>>>       case MSM_CSID_PAD_SRC:
>>> -        if (csid->testgen_mode->cur.val == 0) {
>>> +        if (!csid->testgen_mode || csid->testgen_mode->cur.val == 0) {
>>
>> See my comments on adding new guards to core functionality.
>>
>> Is this sm8550 specific or generic ?
>>
> 
> It is sm8550 specific, since we don't have testgen mode in sm8550 csid, 
> so need to add some guards, the guards are added for similar reason.

Hmm, I see in my tree I just assigned testgen_mode to some dummy data. 
You're right, retain this, when we enable testgen as a standalone entity 
outside of CSID we can address this again.

> 
>>>               /* Test generator is disabled, */
>>>               /* keep pad formats in sync */
>>>               u32 code = fmt->code;
>>> @@ -1042,6 +1042,7 @@ static int csid_init_formats(struct v4l2_subdev 
>>> *sd, struct v4l2_subdev_fh *fh)
>>>   static int csid_set_test_pattern(struct csid_device *csid, s32 value)
>>>   {
>>>       struct csid_testgen_config *tg = &csid->testgen;
>>> +    const struct csid_hw_ops *hw_ops = csid->res->hw_ops;
>>>       /* If CSID is linked to CSIPHY, do not allow to enable test 
>>> generator */
>>>       if (value && media_pad_remote_pad_first(&csid- 
>>> >pads[MSM_CSID_PAD_SINK]))
>>> @@ -1049,7 +1050,10 @@ static int csid_set_test_pattern(struct 
>>> csid_device *csid, s32 value)
>>>       tg->enabled = !!value;
>>> -    return csid->res->hw_ops->configure_testgen_pattern(csid, value);
>>> +    if (hw_ops->configure_testgen_pattern)
>>> +        return -EOPNOTSUPP;
>>> +    else
>>> +        return hw_ops->configure_testgen_pattern(csid, value);
>>
>> If you just add a dummy configure_testgen_pattern we can get rid of 
>> this branching stuff.
>>
> 
> Do you mean add dummy function in csid-780/gen3.c? How about the other 
> ops in vfe_ops_780, add dummy function or use NULL? We need to guards if 
> we set it as NULL.

See above, you're right what you have is fine.

> 
> static int csid_configure_testgen_pattern(struct csid_device *csid, s32 
> val)
> {
>      return 0;
> }
> 
>>>   }
>>>   /*
>>> @@ -1121,6 +1125,19 @@ int msm_csid_subdev_init(struct camss *camss, 
>>> struct csid_device *csid,
>>>           csid->base = devm_platform_ioremap_resource_byname(pdev, 
>>> res->reg[0]);
>>>           if (IS_ERR(csid->base))
>>>               return PTR_ERR(csid->base);
>>> +
>>> +        /* CSID "top" is a new function in new version HW,
>>> +         * CSID can connect to VFE & SFE(Sensor Front End).
>>> +         * this connection is controlled by CSID "top" registers.
>>> +         * There is only one CSID "top" region for all CSIDs.
>>> +         */
>>> +        if (!csid_is_lite(csid) && res->reg[1] && !camss- 
>>> >csid_top_base) {
>>> +            camss->csid_top_base =
>>> +                devm_platform_ioremap_resource_byname(pdev, res- 
>>> >reg[1]);
>>
>> That's a complex clause.
>>
>> Let me send you a patch to do it a different way.
>>
> 
> I was also thinking to addd it in camss level, then I thought it is in 
> csid block, so I moved it to csid, but it is also fine to add it in 
> camss. Can I add your patch into this series? Just like the csiphy patches.

static const struct resources_wrapper csid_wrapper_res_sm8550 = {
         .reg = "csid_wrapper",
};

Yes go ahead, all you should need to do then is add 
"&csid_wrapper_res_sm8550" to your resources.

static const struct camss_resources sm8550_resources = {
         .version = CAMSS_SM8550,
         .pd_name = "top",
         .csiphy_res = csiphy_res_sm8550,
         .csid_res = csid_res_sm8550,
         .ispif_res = NULL,
         .vfe_res = vfe_res_sm8550,
         .csid_wrapper_res = &csid_wrapper_res_sm8550,
...
};

---
bod