[PATCH v1 1/2] PCI: dw-rockchip: Add remove callback for resource cleanup

[Shawn Lin]

在 2025/10/28 星期二 17:34, Anand Moon 写道:
> Hi Shawn,
> 
> Thanks for your review comments.
> 
> On Tue, 28 Oct 2025 at 05:56, Shawn Lin <shawn.lin at rock-chips.com> wrote:
>>
>> 在 2025/10/27 星期一 22:55, Anand Moon 写道:
>>> Introduce a .remove() callback to the Rockchip DesignWare PCIe
>>> controller driver to ensure proper resource deinitialization during
>>> device removal. This includes disabling clocks and deinitializing the
>>> PCIe PHY.
>>>
>>> Signed-off-by: Anand Moon <linux.amoon at gmail.com>
>>> ---
>>>    drivers/pci/controller/dwc/pcie-dw-rockchip.c | 11 +++++++++++
>>>    1 file changed, 11 insertions(+)
>>>
>>> diff --git a/drivers/pci/controller/dwc/pcie-dw-rockchip.c b/drivers/pci/controller/dwc/pcie-dw-rockchip.c
>>> index 87dd2dd188b4..b878ae8e2b3e 100644
>>> --- a/drivers/pci/controller/dwc/pcie-dw-rockchip.c
>>> +++ b/drivers/pci/controller/dwc/pcie-dw-rockchip.c
>>> @@ -717,6 +717,16 @@ static int rockchip_pcie_probe(struct platform_device *pdev)
>>>        return ret;
>>>    }
>>>
>>> +static void rockchip_pcie_remove(struct platform_device *pdev)
>>> +{
>>> +     struct device *dev = &pdev->dev;
>>> +     struct rockchip_pcie *rockchip = dev_get_drvdata(dev);
>>> +
>>> +     /* Perform other cleanups as necessary */
>>> +     clk_bulk_disable_unprepare(rockchip->clk_cnt, rockchip->clks);
>>> +     rockchip_pcie_phy_deinit(rockchip);
>>> +}
>>
>> Thanks for the patch.
>>
>> Dou you need to call dw_pcie_host_deinit()?
> I feel the rockchip_pcie_phy_deinit will power off the phy
>> And I think you should also try to mask PCIE_CLIENT_INTR_MASK_MISC and
>> remove the irq domain as well.
>>
>> if (rockchip->irq_domain) {
>>          int virq, j;
>>          for (j = 0; j < PCI_NUM_INTX; j++) {
>>                  virq = irq_find_mapping(rockchip->irq_domain, j);
>>                  if (virq > 0)
>>                          irq_dispose_mapping(virq);
>>           }
>>          irq_set_chained_handler_and_data(rockchip->irq, NULL, NULL);
>>          irq_domain_remove(rockchip->irq_domain);
>> }
>>
> I have implemented resource cleanup in rockchip_pcie_remove,
> which is invoked when the system is shutting down.
> Your feedback on the updated code is welcome.
> 
> static void rockchip_pcie_remove(struct platform_device *pdev)
> {
>          struct device *dev = &pdev->dev;
>          struct rockchip_pcie *rockchip = dev_get_drvdata(dev);
>          int irq;
> 
>          irq = of_irq_get_byname(dev->of_node, "legacy");
>          if (irq < 0)
>                  return;
> 
>          /* Perform other cleanups as necessary */
>          /* clear up INTR staatus register */
>          rockchip_pcie_writel_apb(rockchip, 0xffffffff,
>                                   PCIE_CLIENT_INTR_STATUS_MISC);
>          if (rockchip->irq_domain) {
>                  int virq, j;
>                  for (j = 0; j < PCI_NUM_INTX; j++) {
>                          virq = irq_find_mapping(rockchip->irq_domain, j);
>                          if (virq > 0)
>                                  irq_dispose_mapping(virq);
>                  }
>                  irq_set_chained_handler_and_data(irq, NULL, NULL);
>                  irq_domain_remove(rockchip->irq_domain);
>          }
> 
>          clk_bulk_disable_unprepare(rockchip->clk_cnt, rockchip->clks);
>          /* poweroff the phy */
>          rockchip_pcie_phy_deinit(rockchip);
>          /* release the reset */

release? Or "reset the controller"?

>          reset_control_assert(rockchip->rst);
>          pm_runtime_put_sync(dev);
>          pm_runtime_disable(dev);
>          pm_runtime_no_callbacks(dev);
> }
> 
>> Another thin I noticed is should we call pm_runtime_* here for hope that
>> genpd could be powered donw once removed?
>>
> I could not find 'genpd' (power domain) used in the PCIe driver
> If we have an example to use genpd I will update this.
 > > I am also looking into adding NOIRQ_SYSTEM_SLEEP_PM_OPS

That sounds good, you can pick up my patch[1] if you'd like to continue
addressing the comments that I haven't had time to think more.

[1] https://www.spinics.net/lists/linux-pci/msg171846.html

> 
> Thanks
> -Anand
>