Re: [PATCH v2 2/7] media: platform: qcom/iris: split iris_vpu_power_off_controller in multiple steps

[Neil Armstrong <neil.armstrong@linaro.org>]

On 06/03/2025 13:34, Bryan O'Donoghue wrote:
> On 05/03/2025 19:05, Neil Armstrong wrote:
>> In order to support vpu33, the iris_vpu_power_off_controller needs to be
>> reused and extended, but the AON_WRAPPER_MVP_NOC_LPI_CONTROL cannot be
>> set from the power_off_controller sequence like vpu2 and vpu3 so
>> split the power_off_controller into 3 steps:
>> - iris_vpu_power_off_controller_begin
>> - iris_vpu_power_off_controller_end
>> - iris_vpu_power_off_controller_disable
>>
>> And use them in a common iris_vpu_power_off_controller() for
>> vpu2 and vpu3 based platforms.
>>
>> Signed-off-by: Neil Armstrong <neil.armstrong@linaro.org>
>> ---
>>   drivers/media/platform/qcom/iris/iris_vpu_common.c | 46 ++++++++++++++++------
>>   1 file changed, 33 insertions(+), 13 deletions(-)
>>
>> diff --git a/drivers/media/platform/qcom/iris/iris_vpu_common.c b/drivers/media/platform/qcom/iris/iris_vpu_common.c
>> index fe9896d66848cdcd8c67bd45bbf3b6ce4a01ab10..d6ce92f3c7544e44dccca26bf6a4c95a720f9229 100644
>> --- a/drivers/media/platform/qcom/iris/iris_vpu_common.c
>> +++ b/drivers/media/platform/qcom/iris/iris_vpu_common.c
>> @@ -211,33 +211,29 @@ int iris_vpu_prepare_pc(struct iris_core *core)
>>       return -EAGAIN;
>>   }
>>
>> -static int iris_vpu_power_off_controller(struct iris_core *core)
>> +static void iris_vpu_power_off_controller_begin(struct iris_core *core)
>>   {
>> -    u32 val = 0;
>> -    int ret;
>> -
>>       writel(MSK_SIGNAL_FROM_TENSILICA | MSK_CORE_POWER_ON, core->reg_base + CPU_CS_X2RPMH);
>> +}
>>
>> -    writel(REQ_POWER_DOWN_PREP, core->reg_base + AON_WRAPPER_MVP_NOC_LPI_CONTROL);
>> -
>> -    ret = readl_poll_timeout(core->reg_base + AON_WRAPPER_MVP_NOC_LPI_STATUS,
>> -                 val, val & BIT(0), 200, 2000);
>> -    if (ret)
>> -        goto disable_power;
>> +static int iris_vpu_power_off_controller_end(struct iris_core *core)
>> +{
>> +    u32 val = 0;
>> +    int ret;
>>
>>       writel(REQ_POWER_DOWN_PREP, core->reg_base + WRAPPER_IRIS_CPU_NOC_LPI_CONTROL);
>>
>>       ret = readl_poll_timeout(core->reg_base + WRAPPER_IRIS_CPU_NOC_LPI_STATUS,
>>                    val, val & BIT(0), 200, 2000);
>>       if (ret)
>> -        goto disable_power;
>> +        return ret;
>>
>>       writel(0x0, core->reg_base + WRAPPER_DEBUG_BRIDGE_LPI_CONTROL);
>>
>>       ret = readl_poll_timeout(core->reg_base + WRAPPER_DEBUG_BRIDGE_LPI_STATUS,
>>                    val, val == 0, 200, 2000);
>>       if (ret)
>> -        goto disable_power;
>> +        return ret;
>>
>>       writel(CTL_AXI_CLK_HALT | CTL_CLK_HALT,
>>              core->reg_base + WRAPPER_TZ_CTL_AXI_CLOCK_CONFIG);
>> @@ -245,10 +241,34 @@ static int iris_vpu_power_off_controller(struct iris_core *core)
>>       writel(0x0, core->reg_base + WRAPPER_TZ_QNS4PDXFIFO_RESET);
>>       writel(0x0, core->reg_base + WRAPPER_TZ_CTL_AXI_CLOCK_CONFIG);
>>
>> -disable_power:
>> +    return 0;
>> +}
>> +
>> +static void iris_vpu_power_off_controller_disable(struct iris_core *core)
>> +{
>>       iris_disable_unprepare_clock(core, IRIS_CTRL_CLK);
>>       iris_disable_unprepare_clock(core, IRIS_AXI_CLK);
>>       iris_disable_power_domains(core, core->pmdomain_tbl->pd_devs[IRIS_CTRL_POWER_DOMAIN]);
>> +}
>> +
>> +static int iris_vpu_power_off_controller(struct iris_core *core)
>> +{
>> +    u32 val = 0;
>> +    int ret;
>> +
>> +    iris_vpu_power_off_controller_begin(core);
>> +
>> +    writel(REQ_POWER_DOWN_PREP, core->reg_base + AON_WRAPPER_MVP_NOC_LPI_CONTROL);
>> +
>> +    ret = readl_poll_timeout(core->reg_base + AON_WRAPPER_MVP_NOC_LPI_STATUS,
>> +                 val, val & BIT(0), 200, 2000);
>> +    if (ret)
>> +        goto disable_power;
>> +
>> +    iris_vpu_power_off_controller_end(core);
>> +
>> +disable_power:
>> +    iris_vpu_power_off_controller_disable(core);
>>
>>       return 0;
>>   }
>>
>> -- 
>> 2.34.1
>>
>>
> 
> Have you checked that rb5/sm8250 still works after this change ?

I'm on it, but it doesn't add any functional changes.

Neil

> 
> ---
> bod