[PATCH v3 0/2] Fix dwc_pcie pmu driver issues

[PATCH v3 0/2] Fix dwc_pcie pmu driver issues

[Yunhui Cui]

v1:
Only sent the patch titled "fix duplicate pci_dev devices".

v2:
1. New patch: "fix the incorrect reference count".
2. The "fix duplicate pci_dev devices" change is made in two places:
One is to remove the modification of the PMU name. The other is to change
"for_each_pci_dev" to "pci_get_domain_bus_and_slot".

v3:
patch[1]: Remove the logic of "fixing the reference count".
patch[2]: When pr_err, also output the sddf.

Yunhui Cui (2):
  perf/dwc_pcie: fix some unreleased resources
  perf/dwc_pcie: fix duplicate pci_dev devices

 drivers/perf/dwc_pcie_pmu.c | 53 ++++++++++++++++++++++++-------------
 1 file changed, 35 insertions(+), 18 deletions(-)

-- 
2.39.2

[PATCH v3 1/2] perf/dwc_pcie: fix some unreleased resources

[Yunhui Cui]

Release leaked resources, such as plat_dev and dev_info.

Signed-off-by: Yunhui Cui <cuiyunhui@bytedance.com>
---
 drivers/perf/dwc_pcie_pmu.c | 33 ++++++++++++++++++++++-----------
 1 file changed, 22 insertions(+), 11 deletions(-)

diff --git a/drivers/perf/dwc_pcie_pmu.c b/drivers/perf/dwc_pcie_pmu.c
index cccecae9823f..19fa2ba8dd67 100644
--- a/drivers/perf/dwc_pcie_pmu.c
+++ b/drivers/perf/dwc_pcie_pmu.c
@@ -572,8 +572,10 @@ static int dwc_pcie_register_dev(struct pci_dev *pdev)
 		return PTR_ERR(plat_dev);
 
 	dev_info = kzalloc(sizeof(*dev_info), GFP_KERNEL);
-	if (!dev_info)
+	if (!dev_info) {
+		platform_device_unregister(plat_dev);
 		return -ENOMEM;
+	}
 
 	/* Cache platform device to handle pci device hotplug */
 	dev_info->plat_dev = plat_dev;
@@ -730,6 +732,15 @@ static struct platform_driver dwc_pcie_pmu_driver = {
 	.driver = {.name = "dwc_pcie_pmu",},
 };
 
+static void dwc_pcie_cleanup_devices(void)
+{
+	struct dwc_pcie_dev_info *dev_info, *tmp;
+
+	list_for_each_entry_safe(dev_info, tmp, &dwc_pcie_dev_info_head, dev_node) {
+		dwc_pcie_unregister_dev(dev_info);
+	}
+}
+
 static int __init dwc_pcie_pmu_init(void)
 {
 	struct pci_dev *pdev = NULL;
@@ -742,7 +753,7 @@ static int __init dwc_pcie_pmu_init(void)
 		ret = dwc_pcie_register_dev(pdev);
 		if (ret) {
 			pci_dev_put(pdev);
-			return ret;
+			goto err_cleanup;
 		}
 	}
 
@@ -751,35 +762,35 @@ static int __init dwc_pcie_pmu_init(void)
 				      dwc_pcie_pmu_online_cpu,
 				      dwc_pcie_pmu_offline_cpu);
 	if (ret < 0)
-		return ret;
+		goto err_cleanup;
 
 	dwc_pcie_pmu_hp_state = ret;
 
 	ret = platform_driver_register(&dwc_pcie_pmu_driver);
 	if (ret)
-		goto platform_driver_register_err;
+		goto err_remove_cpuhp;
 
 	ret = bus_register_notifier(&pci_bus_type, &dwc_pcie_pmu_nb);
 	if (ret)
-		goto platform_driver_register_err;
+		goto err_unregister_driver;
 	notify = true;
 
 	return 0;
 
-platform_driver_register_err:
+err_unregister_driver:
+	platform_driver_unregister(&dwc_pcie_pmu_driver);
+err_remove_cpuhp:
 	cpuhp_remove_multi_state(dwc_pcie_pmu_hp_state);
-
+err_cleanup:
+	dwc_pcie_cleanup_devices();
 	return ret;
 }
 
 static void __exit dwc_pcie_pmu_exit(void)
 {
-	struct dwc_pcie_dev_info *dev_info, *tmp;
-
 	if (notify)
 		bus_unregister_notifier(&pci_bus_type, &dwc_pcie_pmu_nb);
-	list_for_each_entry_safe(dev_info, tmp, &dwc_pcie_dev_info_head, dev_node)
-		dwc_pcie_unregister_dev(dev_info);
+	dwc_pcie_cleanup_devices();
 	platform_driver_unregister(&dwc_pcie_pmu_driver);
 	cpuhp_remove_multi_state(dwc_pcie_pmu_hp_state);
 }
-- 
2.39.2

Re: [PATCH v3 1/2] perf/dwc_pcie: fix some unreleased resources

[Shuai Xue]

在 2025/2/8 18:40, Yunhui Cui 写道:
> Release leaked resources, such as plat_dev and dev_info.
> 
> Signed-off-by: Yunhui Cui <cuiyunhui@bytedance.com>
> ---
>   drivers/perf/dwc_pcie_pmu.c | 33 ++++++++++++++++++++++-----------
>   1 file changed, 22 insertions(+), 11 deletions(-)
> 
> diff --git a/drivers/perf/dwc_pcie_pmu.c b/drivers/perf/dwc_pcie_pmu.c
> index cccecae9823f..19fa2ba8dd67 100644
> --- a/drivers/perf/dwc_pcie_pmu.c
> +++ b/drivers/perf/dwc_pcie_pmu.c
> @@ -572,8 +572,10 @@ static int dwc_pcie_register_dev(struct pci_dev *pdev)
>   		return PTR_ERR(plat_dev);
>   
>   	dev_info = kzalloc(sizeof(*dev_info), GFP_KERNEL);
> -	if (!dev_info)
> +	if (!dev_info) {
> +		platform_device_unregister(plat_dev);
>   		return -ENOMEM;
> +	}
>   
>   	/* Cache platform device to handle pci device hotplug */
>   	dev_info->plat_dev = plat_dev;
> @@ -730,6 +732,15 @@ static struct platform_driver dwc_pcie_pmu_driver = {
>   	.driver = {.name = "dwc_pcie_pmu",},
>   };
>   
> +static void dwc_pcie_cleanup_devices(void)
> +{
> +	struct dwc_pcie_dev_info *dev_info, *tmp;
> +
> +	list_for_each_entry_safe(dev_info, tmp, &dwc_pcie_dev_info_head, dev_node) {
> +		dwc_pcie_unregister_dev(dev_info);
> +	}
> +}
> +
>   static int __init dwc_pcie_pmu_init(void)
>   {
>   	struct pci_dev *pdev = NULL;
> @@ -742,7 +753,7 @@ static int __init dwc_pcie_pmu_init(void)
>   		ret = dwc_pcie_register_dev(pdev);
>   		if (ret) {
>   			pci_dev_put(pdev);

Should we get a reference count of pdev in dwc_pcie_register_dev and put it in
dwc_pcie_unregister_dev?

Thanks.
Shuai

Re: [External] Re: [PATCH v3 1/2] perf/dwc_pcie: fix some unreleased resources

[yunhui cui]

Hi Shuai,

On Tue, Feb 11, 2025 at 4:05 PM Shuai Xue <xueshuai@linux.alibaba.com> wrote:
>
>
>
> 在 2025/2/8 18:40, Yunhui Cui 写道:
> > Release leaked resources, such as plat_dev and dev_info.
> >
> > Signed-off-by: Yunhui Cui <cuiyunhui@bytedance.com>
> > ---
> >   drivers/perf/dwc_pcie_pmu.c | 33 ++++++++++++++++++++++-----------
> >   1 file changed, 22 insertions(+), 11 deletions(-)
> >
> > diff --git a/drivers/perf/dwc_pcie_pmu.c b/drivers/perf/dwc_pcie_pmu.c
> > index cccecae9823f..19fa2ba8dd67 100644
> > --- a/drivers/perf/dwc_pcie_pmu.c
> > +++ b/drivers/perf/dwc_pcie_pmu.c
> > @@ -572,8 +572,10 @@ static int dwc_pcie_register_dev(struct pci_dev *pdev)
> >               return PTR_ERR(plat_dev);
> >
> >       dev_info = kzalloc(sizeof(*dev_info), GFP_KERNEL);
> > -     if (!dev_info)
> > +     if (!dev_info) {
> > +             platform_device_unregister(plat_dev);
> >               return -ENOMEM;
> > +     }
> >
> >       /* Cache platform device to handle pci device hotplug */
> >       dev_info->plat_dev = plat_dev;
> > @@ -730,6 +732,15 @@ static struct platform_driver dwc_pcie_pmu_driver = {
> >       .driver = {.name = "dwc_pcie_pmu",},
> >   };
> >
> > +static void dwc_pcie_cleanup_devices(void)
> > +{
> > +     struct dwc_pcie_dev_info *dev_info, *tmp;
> > +
> > +     list_for_each_entry_safe(dev_info, tmp, &dwc_pcie_dev_info_head, dev_node) {
> > +             dwc_pcie_unregister_dev(dev_info);
> > +     }
> > +}
> > +
> >   static int __init dwc_pcie_pmu_init(void)
> >   {
> >       struct pci_dev *pdev = NULL;
> > @@ -742,7 +753,7 @@ static int __init dwc_pcie_pmu_init(void)
> >               ret = dwc_pcie_register_dev(pdev);
> >               if (ret) {
> >                       pci_dev_put(pdev);
>
> Should we get a reference count of pdev in dwc_pcie_register_dev and put it in
> dwc_pcie_unregister_dev?

Personally, it's not an issue because RP devices are generally not unloaded.

>
> Thanks.
> Shuai
>

Thanks,
Yunhui

[PATCH v3 2/2] perf/dwc_pcie: fix duplicate pci_dev devices

[Yunhui Cui]

During platform_device_register, wrongly using struct device
pci_dev as platform_data caused a kmemdup copy of pci_dev. Worse
still, accessing the duplicated device leads to list corruption as its
mutex content (e.g., list, magic) remains the same as the original.

Signed-off-by: Yunhui Cui <cuiyunhui@bytedance.com>
---
 drivers/perf/dwc_pcie_pmu.c | 20 +++++++++++++-------
 1 file changed, 13 insertions(+), 7 deletions(-)

diff --git a/drivers/perf/dwc_pcie_pmu.c b/drivers/perf/dwc_pcie_pmu.c
index 19fa2ba8dd67..4f6599e32bba 100644
--- a/drivers/perf/dwc_pcie_pmu.c
+++ b/drivers/perf/dwc_pcie_pmu.c
@@ -565,9 +565,7 @@ static int dwc_pcie_register_dev(struct pci_dev *pdev)
 	u32 sbdf;
 
 	sbdf = (pci_domain_nr(pdev->bus) << 16) | PCI_DEVID(pdev->bus->number, pdev->devfn);
-	plat_dev = platform_device_register_data(NULL, "dwc_pcie_pmu", sbdf,
-						 pdev, sizeof(*pdev));
-
+	plat_dev = platform_device_register_simple("platform_dwc_pcie", sbdf, NULL, 0);
 	if (IS_ERR(plat_dev))
 		return PTR_ERR(plat_dev);
 
@@ -616,18 +614,26 @@ static struct notifier_block dwc_pcie_pmu_nb = {
 
 static int dwc_pcie_pmu_probe(struct platform_device *plat_dev)
 {
-	struct pci_dev *pdev = plat_dev->dev.platform_data;
+	struct pci_dev *pdev;
 	struct dwc_pcie_pmu *pcie_pmu;
 	char *name;
 	u32 sbdf;
 	u16 vsec;
 	int ret;
 
+	sbdf = plat_dev->id;
+	pdev = pci_get_domain_bus_and_slot(sbdf >> 16, PCI_BUS_NUM(sbdf & 0xffff),
+					   sbdf & 0xff);
+	if (!pdev) {
+		pr_err("No pdev found for the sbdf 0x%x\n", sbdf);
+		return -ENODEV;
+	}
+
 	vsec = dwc_pcie_des_cap(pdev);
 	if (!vsec)
 		return -ENODEV;
 
-	sbdf = plat_dev->id;
+	pci_dev_put(pdev);
 	name = devm_kasprintf(&plat_dev->dev, GFP_KERNEL, "dwc_rootport_%x", sbdf);
 	if (!name)
 		return -ENOMEM;
@@ -642,7 +648,7 @@ static int dwc_pcie_pmu_probe(struct platform_device *plat_dev)
 	pcie_pmu->on_cpu = -1;
 	pcie_pmu->pmu = (struct pmu){
 		.name		= name,
-		.parent		= &pdev->dev,
+		.parent		= &plat_dev->dev,
 		.module		= THIS_MODULE,
 		.attr_groups	= dwc_pcie_attr_groups,
 		.capabilities	= PERF_PMU_CAP_NO_EXCLUDE,
@@ -729,7 +735,7 @@ static int dwc_pcie_pmu_offline_cpu(unsigned int cpu, struct hlist_node *cpuhp_n
 
 static struct platform_driver dwc_pcie_pmu_driver = {
 	.probe = dwc_pcie_pmu_probe,
-	.driver = {.name = "dwc_pcie_pmu",},
+	.driver = {.name = "platform_dwc_pcie",},
 };
 
 static void dwc_pcie_cleanup_devices(void)
-- 
2.39.2

Re: [PATCH v3 2/2] perf/dwc_pcie: fix duplicate pci_dev devices

[Shuai Xue]

在 2025/2/8 18:40, Yunhui Cui 写道:
> During platform_device_register, wrongly using struct device
> pci_dev as platform_data caused a kmemdup copy of pci_dev. Worse
> still, accessing the duplicated device leads to list corruption as its
> mutex content (e.g., list, magic) remains the same as the original.
> 
> Signed-off-by: Yunhui Cui <cuiyunhui@bytedance.com>
> ---
>   drivers/perf/dwc_pcie_pmu.c | 20 +++++++++++++-------
>   1 file changed, 13 insertions(+), 7 deletions(-)
> 
> diff --git a/drivers/perf/dwc_pcie_pmu.c b/drivers/perf/dwc_pcie_pmu.c
> index 19fa2ba8dd67..4f6599e32bba 100644
> --- a/drivers/perf/dwc_pcie_pmu.c
> +++ b/drivers/perf/dwc_pcie_pmu.c
> @@ -565,9 +565,7 @@ static int dwc_pcie_register_dev(struct pci_dev *pdev)
>   	u32 sbdf;
>   
>   	sbdf = (pci_domain_nr(pdev->bus) << 16) | PCI_DEVID(pdev->bus->number, pdev->devfn);
> -	plat_dev = platform_device_register_data(NULL, "dwc_pcie_pmu", sbdf,
> -						 pdev, sizeof(*pdev));
> -
> +	plat_dev = platform_device_register_simple("platform_dwc_pcie", sbdf, NULL, 0);
>   	if (IS_ERR(plat_dev))
>   		return PTR_ERR(plat_dev);
>   
> @@ -616,18 +614,26 @@ static struct notifier_block dwc_pcie_pmu_nb = {
>   
>   static int dwc_pcie_pmu_probe(struct platform_device *plat_dev)
>   {
> -	struct pci_dev *pdev = plat_dev->dev.platform_data;
> +	struct pci_dev *pdev;
>   	struct dwc_pcie_pmu *pcie_pmu;
>   	char *name;
>   	u32 sbdf;
>   	u16 vsec;
>   	int ret;
>   
> +	sbdf = plat_dev->id;
> +	pdev = pci_get_domain_bus_and_slot(sbdf >> 16, PCI_BUS_NUM(sbdf & 0xffff),
> +					   sbdf & 0xff);
> +	if (!pdev) {
> +		pr_err("No pdev found for the sbdf 0x%x\n", sbdf);
> +		return -ENODEV;
> +	}
> +
>   	vsec = dwc_pcie_des_cap(pdev);
>   	if (!vsec)
>   		return -ENODEV;

pci_dev_put(pdev) should move ahead to aovid return here.

>   
> -	sbdf = plat_dev->id;
> +	pci_dev_put(pdev);
>   	name = devm_kasprintf(&plat_dev->dev, GFP_KERNEL, "dwc_rootport_%x", sbdf);
>   	if (!name)
>   		return -ENOMEM;
> @@ -642,7 +648,7 @@ static int dwc_pcie_pmu_probe(struct platform_device *plat_dev)
>   	pcie_pmu->on_cpu = -1;
>   	pcie_pmu->pmu = (struct pmu){
>   		.name		= name,
> -		.parent		= &pdev->dev,
> +		.parent		= &plat_dev->dev,
>   		.module		= THIS_MODULE,
>   		.attr_groups	= dwc_pcie_attr_groups,
>   		.capabilities	= PERF_PMU_CAP_NO_EXCLUDE,
> @@ -729,7 +735,7 @@ static int dwc_pcie_pmu_offline_cpu(unsigned int cpu, struct hlist_node *cpuhp_n
>   
>   static struct platform_driver dwc_pcie_pmu_driver = {
>   	.probe = dwc_pcie_pmu_probe,
> -	.driver = {.name = "dwc_pcie_pmu",},
> +	.driver = {.name = "platform_dwc_pcie",},

Aha, it is very difficult to come up with a name that satisfies everyone. The
original name uses the '_pmu' suffix to follow the unwritten convention of
other PMU drivers.

Personally, I think the original name is more appropriate, but I'll leave the
decision to @Will.

Thanks.
Best Regards.
Shuai

Re: [External] Re: [PATCH v3 2/2] perf/dwc_pcie: fix duplicate pci_dev devices

[yunhui cui]

Hi Shuai,

On Tue, Feb 11, 2025 at 4:02 PM Shuai Xue <xueshuai@linux.alibaba.com> wrote:
>
>
>
> 在 2025/2/8 18:40, Yunhui Cui 写道:
> > During platform_device_register, wrongly using struct device
> > pci_dev as platform_data caused a kmemdup copy of pci_dev. Worse
> > still, accessing the duplicated device leads to list corruption as its
> > mutex content (e.g., list, magic) remains the same as the original.
> >
> > Signed-off-by: Yunhui Cui <cuiyunhui@bytedance.com>
> > ---
> >   drivers/perf/dwc_pcie_pmu.c | 20 +++++++++++++-------
> >   1 file changed, 13 insertions(+), 7 deletions(-)
> >
> > diff --git a/drivers/perf/dwc_pcie_pmu.c b/drivers/perf/dwc_pcie_pmu.c
> > index 19fa2ba8dd67..4f6599e32bba 100644
> > --- a/drivers/perf/dwc_pcie_pmu.c
> > +++ b/drivers/perf/dwc_pcie_pmu.c
> > @@ -565,9 +565,7 @@ static int dwc_pcie_register_dev(struct pci_dev *pdev)
> >       u32 sbdf;
> >
> >       sbdf = (pci_domain_nr(pdev->bus) << 16) | PCI_DEVID(pdev->bus->number, pdev->devfn);
> > -     plat_dev = platform_device_register_data(NULL, "dwc_pcie_pmu", sbdf,
> > -                                              pdev, sizeof(*pdev));
> > -
> > +     plat_dev = platform_device_register_simple("platform_dwc_pcie", sbdf, NULL, 0);
> >       if (IS_ERR(plat_dev))
> >               return PTR_ERR(plat_dev);
> >
> > @@ -616,18 +614,26 @@ static struct notifier_block dwc_pcie_pmu_nb = {
> >
> >   static int dwc_pcie_pmu_probe(struct platform_device *plat_dev)
> >   {
> > -     struct pci_dev *pdev = plat_dev->dev.platform_data;
> > +     struct pci_dev *pdev;
> >       struct dwc_pcie_pmu *pcie_pmu;
> >       char *name;
> >       u32 sbdf;
> >       u16 vsec;
> >       int ret;
> >
> > +     sbdf = plat_dev->id;
> > +     pdev = pci_get_domain_bus_and_slot(sbdf >> 16, PCI_BUS_NUM(sbdf & 0xffff),
> > +                                        sbdf & 0xff);
> > +     if (!pdev) {
> > +             pr_err("No pdev found for the sbdf 0x%x\n", sbdf);
> > +             return -ENODEV;
> > +     }
> > +
> >       vsec = dwc_pcie_des_cap(pdev);
> >       if (!vsec)
> >               return -ENODEV;
>
> pci_dev_put(pdev) should move ahead to aovid return here.
>
> >
> > -     sbdf = plat_dev->id;
> > +     pci_dev_put(pdev);
> >       name = devm_kasprintf(&plat_dev->dev, GFP_KERNEL, "dwc_rootport_%x", sbdf);
> >       if (!name)
> >               return -ENOMEM;
> > @@ -642,7 +648,7 @@ static int dwc_pcie_pmu_probe(struct platform_device *plat_dev)
> >       pcie_pmu->on_cpu = -1;
> >       pcie_pmu->pmu = (struct pmu){
> >               .name           = name,
> > -             .parent         = &pdev->dev,
> > +             .parent         = &plat_dev->dev,
> >               .module         = THIS_MODULE,
> >               .attr_groups    = dwc_pcie_attr_groups,
> >               .capabilities   = PERF_PMU_CAP_NO_EXCLUDE,
> > @@ -729,7 +735,7 @@ static int dwc_pcie_pmu_offline_cpu(unsigned int cpu, struct hlist_node *cpuhp_n
> >
> >   static struct platform_driver dwc_pcie_pmu_driver = {
> >       .probe = dwc_pcie_pmu_probe,
> > -     .driver = {.name = "dwc_pcie_pmu",},
> > +     .driver = {.name = "platform_dwc_pcie",},
>
> Aha, it is very difficult to come up with a name that satisfies everyone. The
> original name uses the '_pmu' suffix to follow the unwritten convention of
> other PMU drivers.
>
> Personally, I think the original name is more appropriate, but I'll leave the
> decision to @Will.

Since Will hasn't replied, I'll update to the next version to keep the
original name.

>
> Thanks.
> Best Regards.
> Shuai

Thanks,
Yunhui