Re: [PATCH] bus: mhi: host: pci: Disable runtime PM for QDU100

[Vivek Pernamitta <quic_vpernami@quicinc.com>]

On 4/18/2025 3:04 PM, Vivek Pernamitta wrote:
> 
> 
> On 4/18/2025 2:06 PM, Manivannan Sadhasivam wrote:
>> On Fri, Apr 18, 2025 at 11:55:24AM +0530, Vivek Pernamitta wrote:
>>>
>>>
>>> On 4/17/2025 11:37 AM, Manivannan Sadhasivam wrote:
>>>> On Thu, Apr 17, 2025 at 10:00:38AM +0530, Veerabhadrarao Badiganti 
>>>> wrote:
>>>>>
>>>>> On 4/14/2025 1:17 PM, Vivek Pernamitta wrote:
>>>>>> The QDU100 device does not support the MHI M3 state, necessitating 
>>>>>> the
>>>>>> disabling of runtime PM for this device. Since the PCIe core 
>>>>>> framework
>>>>>> enables runtime PM by default for all clients, it is essential to 
>>>>>> disable
>>>>>> runtime PM if the device does not support Low Power Mode (LPM).
>>>>>>
>>
>> Not true... See below.
>>
>>>>>> Signed-off-by: Vivek Pernamitta<quic_vpernami@quicinc.com>
>>>>>> ---
>>>>>>     drivers/bus/mhi/host/pci_generic.c | 10 ++++++++++
>>>>>>     1 file changed, 10 insertions(+)
>>>>>>
>>>>>> diff --git a/drivers/bus/mhi/host/pci_generic.c b/drivers/bus/mhi/ 
>>>>>> host/pci_generic.c
>>>>>> index 
>>>>>> 03aa887952098661a488650053a357f883d1559b..a011fd2d48c57cf9d1aec74040153267a206d797 100644
>>>>>> --- a/drivers/bus/mhi/host/pci_generic.c
>>>>>> +++ b/drivers/bus/mhi/host/pci_generic.c
>>>>>> @@ -43,6 +43,7 @@
>>>>>>      * @mru_default: default MRU size for MBIM network packets
>>>>>>      * @sideband_wake: Devices using dedicated sideband GPIO for 
>>>>>> wakeup instead
>>>>>>      *           of inband wake support (such as sdx24)
>>>>>> + * @pm_disable: disables runtime PM (optional)
>>>>>>      */
>>>>>>     struct mhi_pci_dev_info {
>>>>>>         const struct mhi_controller_config *config;
>>>>>> @@ -54,6 +55,7 @@ struct mhi_pci_dev_info {
>>>>>>         unsigned int dma_data_width;
>>>>>>         unsigned int mru_default;
>>>>>>         bool sideband_wake;
>>>>>> +    bool pm_disable;
>>>>>>     };
>>>>>>     #define MHI_CHANNEL_CONFIG_UL(ch_num, ch_name, el_count, 
>>>>>> ev_ring) \
>>>>>> @@ -295,6 +297,7 @@ static const struct mhi_pci_dev_info 
>>>>>> mhi_qcom_qdu100_info = {
>>>>>>         .bar_num = MHI_PCI_DEFAULT_BAR_NUM,
>>>>>>         .dma_data_width = 32,
>>>>>>         .sideband_wake = false,
>>>>>> +    .pm_disable = true,
>>>>>
>>>>> |no_m3|orno_|m3_support|would be more suitable than|pm_disable|
>>>>
>>>> Yes!
>>>>
We named variable pm_disable because the M3 state is implicitly set 
during runtime PM suspend. Both are not needed and not fully supported 
for the QDU100 accelerator card, so we want to disable runtime suspend 
for it.
>>>> But does the device not supporting M3 only or D3Hot also? If the 
>>>> former, then we
>>>> should prevent MHI host to enter/exit M3 state in mhi_pm_suspend/ 
>>>> mhi_pm_resume.
>>>> There is an incentive in allowing D3Hot if the device supports it.
>>>>
>>>> Also, is there a way we could probe M3 support in the device so that 
>>>> we can
>>>> check it during runtime?
>>>>
>>>> - Mani
>>>>
>>> In QDU100 device does not support M3 state, D3_hot will be supported. As
>>> QDU100 is an accelerator card which needs high-throughout, LPM needs to
>>> disabled here. So we are trying to disable runtime PM here, without M3 ,
>>> D3_hot will not have effect in QDU100.
>>> Also in MHI we don't have provision to check device M3 capability, so 
>>> we are
>>> trying to disable runtime PM here.
>>>
>>
>> Even though PCI core is enabling runtime PM for devices, it also 
>> prevents the
>> devices from getting suspended unless the drivers explicitly asks for 
>> it. So
>> this driver is using autosuspend feature to allow the devices to be 
>> suspended
>> if PME is supported from D3Hot.
>>
>> If you want to prevent the device from getting runtime suspended, you 
>> should
>> just skip autosuspend.
>>
>> - Mani
>>
> Thank you, I will upload new patchset accordingly.
> 
>