| Message ID | 20201102143045.142121-1-zhangqilong3@huawei.com (mailing list archive) |
|---|---|
| State | New, archived |
| Delegated to: | Lorenzo Pieralisi |
| Headers | show |
| Series | PCI: dwc: fix reference leak in pex_ep_event_pex_rst_deassert | expand |
On 11/2/2020 8:00 PM, Zhang Qilong wrote: > External email: Use caution opening links or attachments > > > pm_runtime_get_sync will increment pm usage counter even it > failed. Forgetting to pm_runtime_put_noidle will result in > reference leak in pex_ep_event_pex_rst_deassert, so we should > fix it. > > Fixes: c57247f940e8e ("PCI: tegra: Add support for PCIe endpoint mode in Tegra194") > Signed-off-by: Zhang Qilong <zhangqilong3@huawei.com> > --- > drivers/pci/controller/dwc/pcie-tegra194.c | 1 + > 1 file changed, 1 insertion(+) > > diff --git a/drivers/pci/controller/dwc/pcie-tegra194.c b/drivers/pci/controller/dwc/pcie-tegra194.c > index f920e7efe118..936510b5c649 100644 > --- a/drivers/pci/controller/dwc/pcie-tegra194.c > +++ b/drivers/pci/controller/dwc/pcie-tegra194.c > @@ -1662,6 +1662,7 @@ static void pex_ep_event_pex_rst_deassert(struct tegra_pcie_dw *pcie) > > ret = pm_runtime_get_sync(dev); > if (ret < 0) { > + pm_runtime_put_noidle(dev); Why can't we call pm_runtime_put_sync(dev) as that is what is being called in failure cases anyway further down in this API? > dev_err(dev, "Failed to get runtime sync for PCIe dev: %d\n", > ret); > return; > -- > 2.17.1 >
Hi > > > On 11/2/2020 8:00 PM, Zhang Qilong wrote: > > External email: Use caution opening links or attachments > > > > > > pm_runtime_get_sync will increment pm usage counter even it failed. > > Forgetting to pm_runtime_put_noidle will result in reference leak in > > pex_ep_event_pex_rst_deassert, so we should fix it. > > > > Fixes: c57247f940e8e ("PCI: tegra: Add support for PCIe endpoint mode > > in Tegra194") > > Signed-off-by: Zhang Qilong <zhangqilong3@huawei.com> > > --- > > drivers/pci/controller/dwc/pcie-tegra194.c | 1 + > > 1 file changed, 1 insertion(+) > > > > diff --git a/drivers/pci/controller/dwc/pcie-tegra194.c > > b/drivers/pci/controller/dwc/pcie-tegra194.c > > index f920e7efe118..936510b5c649 100644 > > --- a/drivers/pci/controller/dwc/pcie-tegra194.c > > +++ b/drivers/pci/controller/dwc/pcie-tegra194.c > > @@ -1662,6 +1662,7 @@ static void pex_ep_event_pex_rst_deassert(struct > > tegra_pcie_dw *pcie) > > > > ret = pm_runtime_get_sync(dev); > > if (ret < 0) { > > + pm_runtime_put_noidle(dev); > Why can't we call pm_runtime_put_sync(dev) as that is what is being called in > failure cases anyway further down in this API? > Both of the two functions are OK, the difference is that, when pm_runtime_put_sync failed(runtime of the device has error, the device is in Inaccessible state or other error state...), it only increase the usage count of the power, and do nothing else. We merely need call pm_runtime_put_noidle to decrease the usage count. If we call pm_runtime_put_sync to reset it, it will notify device bus type if the device can be suspended, and that is meanless when pm_runtime_put_sync failed. Thanks, best wish! Zhang Qilong > > dev_err(dev, "Failed to get runtime sync for PCIe > dev: %d\n", > > ret); > > return; > > -- > > 2.17.1 > >
On Tue, Nov 03, 2020 at 03:14:21AM +0000, zhangqilong wrote: > Hi > > > > > > > On 11/2/2020 8:00 PM, Zhang Qilong wrote: > > > External email: Use caution opening links or attachments > > > > > > > > > pm_runtime_get_sync will increment pm usage counter even it failed. > > > Forgetting to pm_runtime_put_noidle will result in reference leak in > > > pex_ep_event_pex_rst_deassert, so we should fix it. > > > > > > Fixes: c57247f940e8e ("PCI: tegra: Add support for PCIe endpoint mode > > > in Tegra194") > > > Signed-off-by: Zhang Qilong <zhangqilong3@huawei.com> > > > --- > > > drivers/pci/controller/dwc/pcie-tegra194.c | 1 + > > > 1 file changed, 1 insertion(+) > > > > > > diff --git a/drivers/pci/controller/dwc/pcie-tegra194.c > > > b/drivers/pci/controller/dwc/pcie-tegra194.c > > > index f920e7efe118..936510b5c649 100644 > > > --- a/drivers/pci/controller/dwc/pcie-tegra194.c > > > +++ b/drivers/pci/controller/dwc/pcie-tegra194.c > > > @@ -1662,6 +1662,7 @@ static void pex_ep_event_pex_rst_deassert(struct > > > tegra_pcie_dw *pcie) > > > > > > ret = pm_runtime_get_sync(dev); > > > if (ret < 0) { > > > + pm_runtime_put_noidle(dev); > > Why can't we call pm_runtime_put_sync(dev) as that is what is being called in > > failure cases anyway further down in this API? > > > Both of the two functions are OK, the difference is that, when pm_runtime_put_sync failed(runtime of the device has error, the device is in Inaccessible state or other error state...), it only increase the usage count of the power, and do nothing else. We merely need call pm_runtime_put_noidle to decrease the usage count. If we call pm_runtime_put_sync to reset it, it will notify device bus type if the device can be suspended, and that is meanless when pm_runtime_put_sync failed. I don't understand. Vidya any feedback ? Lorenzo > Thanks, best wish! > > Zhang Qilong > > > dev_err(dev, "Failed to get runtime sync for PCIe > > dev: %d\n", > > > ret); > > > return; > > > -- > > > 2.17.1 > > >
On 03/11/2020 02:52, Vidya Sagar wrote: > > > On 11/2/2020 8:00 PM, Zhang Qilong wrote: >> External email: Use caution opening links or attachments >> >> >> pm_runtime_get_sync will increment pm usage counter even it >> failed. Forgetting to pm_runtime_put_noidle will result in >> reference leak in pex_ep_event_pex_rst_deassert, so we should >> fix it. >> >> Fixes: c57247f940e8e ("PCI: tegra: Add support for PCIe endpoint mode >> in Tegra194") >> Signed-off-by: Zhang Qilong <zhangqilong3@huawei.com> >> --- >> drivers/pci/controller/dwc/pcie-tegra194.c | 1 + >> 1 file changed, 1 insertion(+) >> >> diff --git a/drivers/pci/controller/dwc/pcie-tegra194.c >> b/drivers/pci/controller/dwc/pcie-tegra194.c >> index f920e7efe118..936510b5c649 100644 >> --- a/drivers/pci/controller/dwc/pcie-tegra194.c >> +++ b/drivers/pci/controller/dwc/pcie-tegra194.c >> @@ -1662,6 +1662,7 @@ static void pex_ep_event_pex_rst_deassert(struct >> tegra_pcie_dw *pcie) >> >> ret = pm_runtime_get_sync(dev); >> if (ret < 0) { >> + pm_runtime_put_noidle(dev); > Why can't we call pm_runtime_put_sync(dev) as that is what is being > called in failure cases anyway further down in this API? Simply because this is a failure case where the get_sync did not complete. So this change is correct, however, now we have pm_runtime_resume_and_get(), it is better/simpler just to replace the pm_runtime_get_sync with pm_runtime_resume_and_get. Cheers Jon
On 1/25/2021 10:04 PM, Lorenzo Pieralisi wrote: > External email: Use caution opening links or attachments > > > On Tue, Nov 03, 2020 at 03:14:21AM +0000, zhangqilong wrote: >> Hi >> >>> >>> >>> On 11/2/2020 8:00 PM, Zhang Qilong wrote: >>>> External email: Use caution opening links or attachments >>>> >>>> >>>> pm_runtime_get_sync will increment pm usage counter even it failed. >>>> Forgetting to pm_runtime_put_noidle will result in reference leak in >>>> pex_ep_event_pex_rst_deassert, so we should fix it. >>>> >>>> Fixes: c57247f940e8e ("PCI: tegra: Add support for PCIe endpoint mode >>>> in Tegra194") >>>> Signed-off-by: Zhang Qilong <zhangqilong3@huawei.com> >>>> --- >>>> drivers/pci/controller/dwc/pcie-tegra194.c | 1 + >>>> 1 file changed, 1 insertion(+) >>>> >>>> diff --git a/drivers/pci/controller/dwc/pcie-tegra194.c >>>> b/drivers/pci/controller/dwc/pcie-tegra194.c >>>> index f920e7efe118..936510b5c649 100644 >>>> --- a/drivers/pci/controller/dwc/pcie-tegra194.c >>>> +++ b/drivers/pci/controller/dwc/pcie-tegra194.c >>>> @@ -1662,6 +1662,7 @@ static void pex_ep_event_pex_rst_deassert(struct >>>> tegra_pcie_dw *pcie) >>>> >>>> ret = pm_runtime_get_sync(dev); >>>> if (ret < 0) { >>>> + pm_runtime_put_noidle(dev); >>> Why can't we call pm_runtime_put_sync(dev) as that is what is being called in >>> failure cases anyway further down in this API? >>> >> Both of the two functions are OK, the difference is that, when pm_runtime_put_sync failed(runtime of the device has error, the device is in Inaccessible state or other error state...), it only increase the usage count of the power, and do nothing else. We merely need call pm_runtime_put_noidle to decrease the usage count. If we call pm_runtime_put_sync to reset it, it will notify device bus type if the device can be suspended, and that is meanless when pm_runtime_put_sync failed. > > I don't understand. Vidya any feedback ? If I understand this correctly, we are assuming that nothing would have happened apart from just counter increment and hence don't want to run the idle check by calling pm_runtime_put_noidle(). On a different note, what is it that we are going to lose if we let the idle check also to run i.e. by calling pm_runtime_put_sync()? Thanks, Vidya Sagar > > Lorenzo > >> Thanks, best wish! >> >> Zhang Qilong >>>> dev_err(dev, "Failed to get runtime sync for PCIe >>> dev: %d\n", >>>> ret); >>>> return; >>>> -- >>>> 2.17.1 >>>>
On Mon, Jan 25, 2021 at 07:30:13PM +0000, Jon Hunter wrote: > > On 03/11/2020 02:52, Vidya Sagar wrote: > > > > > > On 11/2/2020 8:00 PM, Zhang Qilong wrote: > >> External email: Use caution opening links or attachments > >> > >> > >> pm_runtime_get_sync will increment pm usage counter even it > >> failed. Forgetting to pm_runtime_put_noidle will result in > >> reference leak in pex_ep_event_pex_rst_deassert, so we should > >> fix it. > >> > >> Fixes: c57247f940e8e ("PCI: tegra: Add support for PCIe endpoint mode > >> in Tegra194") > >> Signed-off-by: Zhang Qilong <zhangqilong3@huawei.com> > >> --- > >> drivers/pci/controller/dwc/pcie-tegra194.c | 1 + > >> 1 file changed, 1 insertion(+) > >> > >> diff --git a/drivers/pci/controller/dwc/pcie-tegra194.c > >> b/drivers/pci/controller/dwc/pcie-tegra194.c > >> index f920e7efe118..936510b5c649 100644 > >> --- a/drivers/pci/controller/dwc/pcie-tegra194.c > >> +++ b/drivers/pci/controller/dwc/pcie-tegra194.c > >> @@ -1662,6 +1662,7 @@ static void pex_ep_event_pex_rst_deassert(struct > >> tegra_pcie_dw *pcie) > >> > >> ret = pm_runtime_get_sync(dev); > >> if (ret < 0) { > >> + pm_runtime_put_noidle(dev); > > Why can't we call pm_runtime_put_sync(dev) as that is what is being > > called in failure cases anyway further down in this API? > > > Simply because this is a failure case where the get_sync did not > complete. So this change is correct, however, now we have > pm_runtime_resume_and_get(), it is better/simpler just to replace the > pm_runtime_get_sync with pm_runtime_resume_and_get. I think this patch slipped through the cracks, should I update the patch myself with the suggestion above and merge it or you guys prefer sending it ? Thanks, Lorenzo
diff --git a/drivers/pci/controller/dwc/pcie-tegra194.c b/drivers/pci/controller/dwc/pcie-tegra194.c index f920e7efe118..936510b5c649 100644 --- a/drivers/pci/controller/dwc/pcie-tegra194.c +++ b/drivers/pci/controller/dwc/pcie-tegra194.c @@ -1662,6 +1662,7 @@ static void pex_ep_event_pex_rst_deassert(struct tegra_pcie_dw *pcie) ret = pm_runtime_get_sync(dev); if (ret < 0) { + pm_runtime_put_noidle(dev); dev_err(dev, "Failed to get runtime sync for PCIe dev: %d\n", ret); return;
pm_runtime_get_sync will increment pm usage counter even it failed. Forgetting to pm_runtime_put_noidle will result in reference leak in pex_ep_event_pex_rst_deassert, so we should fix it. Fixes: c57247f940e8e ("PCI: tegra: Add support for PCIe endpoint mode in Tegra194") Signed-off-by: Zhang Qilong <zhangqilong3@huawei.com> --- drivers/pci/controller/dwc/pcie-tegra194.c | 1 + 1 file changed, 1 insertion(+)