Behavior of the PM framework w.r.t the parent-child relationship

Behavior of the PM framework w.r.t the parent-child relationship

[Krishna Chaitanya Chundru]

Hi Rafael,

We are planning to add runtime PM support to the PCIe controller driver
and facing some issues with runtime PM. These are the details.

PCIe controller is a parent of the PCIe bridge dev node and PCIe
endpoint driver as shown below

PCIe controller(Top level parent & parent of host bridge)
                         |
                         v
         PCIe Host bridge(Parent of PCI-PCI bridge driver)
                         |
                         v
         PCI-PCI bridge(Parent of endpoint driver)
                         |
                         v
                PCIe endpoint driver

There is no driver associated with the PCIe Host bridge device node.
And the runtime pm for the bridge node is not enabled explicitly.

we are seeing PCIe controller driver is getting runtime suspended even
when the PCI-PCI bridge and PCIe endpoint driver are active.
Our expectation is controller driver runtime suspend will be called only
when the child drivers went to the runtime suspend as the controller
is parent device.

For the things to work properly we are calling pm_runtime_set_active() &
pm_runtime_enable() in [1]

As bridge device node doesn't have driver associated and it is kind of
integrated with the controller, So we are expecting when the controller
driver enables its runtime pm it should enable the bridge runtime pm
also and vice versa.

Can you please check this behavior and see if it is something which
needs to be fixed at the PM framework.

You can find more discussions on thread[1]

[1] 
https://lore.kernel.org/all/20240305-runtime_pm_enable-v2-1-a849b74091d1@quicinc.com/T/

Thanks & Regards,
Krishna Chaitanya.

Re: Behavior of the PM framework w.r.t the parent-child relationship

[Manivannan Sadhasivam]

+ linux-pci

On Mon, Apr 22, 2024 at 03:30:18PM +0530, Krishna Chaitanya Chundru wrote:
> Hi Rafael,
> 
> We are planning to add runtime PM support to the PCIe controller driver
> and facing some issues with runtime PM. These are the details.
> 
> PCIe controller is a parent of the PCIe bridge dev node and PCIe
> endpoint driver as shown below
> 
> PCIe controller(Top level parent & parent of host bridge)
>                         |
>                         v
>         PCIe Host bridge(Parent of PCI-PCI bridge driver)
>                         |
>                         v
>         PCI-PCI bridge(Parent of endpoint driver)
>                         |
>                         v
>                PCIe endpoint driver
> 
> There is no driver associated with the PCIe Host bridge device node.
> And the runtime pm for the bridge node is not enabled explicitly.
> 
> we are seeing PCIe controller driver is getting runtime suspended even
> when the PCI-PCI bridge and PCIe endpoint driver are active.
> Our expectation is controller driver runtime suspend will be called only
> when the child drivers went to the runtime suspend as the controller
> is parent device.
> 
> For the things to work properly we are calling pm_runtime_set_active() &
> pm_runtime_enable() in [1]
> 
> As bridge device node doesn't have driver associated and it is kind of
> integrated with the controller, So we are expecting when the controller
> driver enables its runtime pm it should enable the bridge runtime pm
> also and vice versa.
> 
> Can you please check this behavior and see if it is something which
> needs to be fixed at the PM framework.
> 

Hi Rafael,

Gentle ping on this thread!

- Mani

-- 
மணிவண்ணன் சதாசிவம்