[PATCH] PCI: dwc: keystone: Free IRQ in `ks_pcie_remove` and the error handling section of `ks_pcie_probe`

[PATCH] PCI: dwc: keystone: Free IRQ in `ks_pcie_remove` and the error handling section of `ks_pcie_probe`

[Xiangyi Zeng]

Smatch complains that:
drivers/pci/controller/dwc/pci-keystone.c:1303 ks_pcie_probe() warn:
'irq' from request_irq() not released on lines: 1183,1187,1303.

"ks-pcie-error-irq" was requested in the `ks_pcie_probe` function, but
was not freed neither in the error handling part of `ks_pcie_probe`
nor in the `ks_pcie_remove` function.

Fix this by adding `free_irq` in `ks_pcie_remove` and in a new error
handling label `err_alloc` after `err_link` in `ks_pcie_probe`. In
`ks_pcie_probe`, if `phy` or `link` memory allocation fails, we will
fall to `err_alloc`. If any other error occurs that leads to
`err_get_sync` or `err_link`, we end up going to `err_alloc`.

Fixes: 0790eb175ee0 ("PCI: keystone: Cleanup error_irq configuration")
Signed-off-by: Xiangyi Zeng <xyzeng@stu.xidian.edu.cn>
Reviewed-by: Dongliang Mu <dzm91@hust.edu.cn>
---
This patch is tested with compilation and passed Smatch.
---
 drivers/pci/controller/dwc/pci-keystone.c | 17 +++++++++++++----
 1 file changed, 13 insertions(+), 4 deletions(-)

diff --git a/drivers/pci/controller/dwc/pci-keystone.c b/drivers/pci/controller/dwc/pci-keystone.c
index 78818853af9e..f321bc2e8026 100644
--- a/drivers/pci/controller/dwc/pci-keystone.c
+++ b/drivers/pci/controller/dwc/pci-keystone.c
@@ -1179,12 +1179,16 @@ static int __init ks_pcie_probe(struct platform_device *pdev)
 		num_lanes = 1;
 
 	phy = devm_kzalloc(dev, sizeof(*phy) * num_lanes, GFP_KERNEL);
-	if (!phy)
-		return -ENOMEM;
+	if (!phy) {
+		ret = -ENOMEM;
+		goto err_alloc;
+	}
 
 	link = devm_kzalloc(dev, sizeof(*link) * num_lanes, GFP_KERNEL);
-	if (!link)
-		return -ENOMEM;
+	if (!link) {
+		ret = -ENOMEM;
+		goto err_alloc;
+	}
 
 	for (i = 0; i < num_lanes; i++) {
 		snprintf(name, sizeof(name), "pcie-phy%d", i);
@@ -1300,6 +1304,9 @@ static int __init ks_pcie_probe(struct platform_device *pdev)
 	while (--i >= 0 && link[i])
 		device_link_del(link[i]);
 
+err_alloc:
+	free_irq(irq, ks_pcie);
+
 	return ret;
 }
 
@@ -1309,12 +1316,14 @@ static int __exit ks_pcie_remove(struct platform_device *pdev)
 	struct device_link **link = ks_pcie->link;
 	int num_lanes = ks_pcie->num_lanes;
 	struct device *dev = &pdev->dev;
+	int irq = platform_get_irq(pdev, 0);
 
 	pm_runtime_put(dev);
 	pm_runtime_disable(dev);
 	ks_pcie_disable_phy(ks_pcie);
 	while (num_lanes--)
 		device_link_del(link[num_lanes]);
+	free_irq(irq, ks_pcie);
 
 	return 0;
 }
-- 
2.34.1

Re: [PATCH] PCI: dwc: keystone: Free IRQ in `ks_pcie_remove` and the error handling section of `ks_pcie_probe`

[Bjorn Helgaas]

On Tue, May 16, 2023 at 01:16:59PM +0800, Xiangyi Zeng wrote:
> Smatch complains that:
> drivers/pci/controller/dwc/pci-keystone.c:1303 ks_pcie_probe() warn:
> 'irq' from request_irq() not released on lines: 1183,1187,1303.

Make this the entire warning line from smatch with no extra newlines
inserted.

> "ks-pcie-error-irq" was requested in the `ks_pcie_probe` function, but
> was not freed neither in the error handling part of `ks_pcie_probe`
> nor in the `ks_pcie_remove` function.
> 
> Fix this by adding `free_irq` in `ks_pcie_remove` and in a new error
> handling label `err_alloc` after `err_link` in `ks_pcie_probe`. In
> `ks_pcie_probe`, if `phy` or `link` memory allocation fails, we will
> fall to `err_alloc`. If any other error occurs that leads to
> `err_get_sync` or `err_link`, we end up going to `err_alloc`.

I think the backticks (`) are markdown that makes these "code".
Personally I think ks_pcie_probe() is more readable than
`ks_pcie_probe` since most people (I think) read these in plain-ASCII
situations.  And using backticks for labels and local variables seems
like overkill.

> Fixes: 0790eb175ee0 ("PCI: keystone: Cleanup error_irq configuration")
> Signed-off-by: Xiangyi Zeng <xyzeng@stu.xidian.edu.cn>
> Reviewed-by: Dongliang Mu <dzm91@hust.edu.cn>

It's best if the Reviewed-by tag is not added until Dongliang sends
email with that tag directly to the mailing list.  Internal reviews
before posting to the mailing list aren't worth much.

> @@ -1309,12 +1316,14 @@ static int __exit ks_pcie_remove(struct platform_device *pdev)
>  	struct device_link **link = ks_pcie->link;
>  	int num_lanes = ks_pcie->num_lanes;
>  	struct device *dev = &pdev->dev;
> +	int irq = platform_get_irq(pdev, 0);

I think it's better to save the irq we looked up in ks_pcie_probe()
and free *that*.  It's probably the same thing you get by calling
platform_get_irq() again, but it seems cleaner to me to save what we
got in ks_pcie_probe().

>  	pm_runtime_put(dev);
>  	pm_runtime_disable(dev);
>  	ks_pcie_disable_phy(ks_pcie);
>  	while (num_lanes--)
>  		device_link_del(link[num_lanes]);
> +	free_irq(irq, ks_pcie);
>  
>  	return 0;
>  }
> -- 
> 2.34.1
> 

Re: [PATCH] PCI: dwc: keystone: Free IRQ in `ks_pcie_remove` and the error handling section of `ks_pcie_probe`

[曾祥翼]

On 17/5/2023 03:49, Bjorn Helgaas wrote:
> On Tue, May 16, 2023 at 01:16:59PM +0800, Xiangyi Zeng wrote:
>> Smatch complains that:
>> drivers/pci/controller/dwc/pci-keystone.c:1303 ks_pcie_probe() warn:
>> 'irq' from request_irq() not released on lines: 1183,1187,1303.
> Make this the entire warning line from smatch with no extra newlines
> inserted.
Thank you for the suggestion. I will put the warning in one line.
>> "ks-pcie-error-irq" was requested in the `ks_pcie_probe` function, but
>> was not freed neither in the error handling part of `ks_pcie_probe`
>> nor in the `ks_pcie_remove` function.
>>
>> Fix this by adding `free_irq` in `ks_pcie_remove` and in a new error
>> handling label `err_alloc` after `err_link` in `ks_pcie_probe`. In
>> `ks_pcie_probe`, if `phy` or `link` memory allocation fails, we will
>> fall to `err_alloc`. If any other error occurs that leads to
>> `err_get_sync` or `err_link`, we end up going to `err_alloc`.
> I think the backticks (`) are markdown that makes these "code".
> Personally I think ks_pcie_probe() is more readable than
> `ks_pcie_probe` since most people (I think) read these in plain-ASCII
> situations.  And using backticks for labels and local variables seems
> like overkill.
>
Sorry for my wrong usage of backticks. I agree that it would be more
readable to use plain-ASCII names for functions and variables. I will
make sure to update the comment message and the subject.
>> Fixes: 0790eb175ee0 ("PCI: keystone: Cleanup error_irq configuration")
>> Signed-off-by: Xiangyi Zeng <xyzeng@stu.xidian.edu.cn>
>> Reviewed-by: Dongliang Mu <dzm91@hust.edu.cn>
> It's best if the Reviewed-by tag is not added until Dongliang sends
> email with that tag directly to the mailing list.  Internal reviews
> before posting to the mailing list aren't worth much.
In our internal review process, only the patch with the Reviewed-by
tag can be submitted to the mailing list. You can check this in our
google group.
https://groups.google.com/g/hust-os-kernel-patches/c/bt397rzVL24/m/l52XYbG4AgAJ
We will consider omitting this tag when sending to the kernel mailing
list in the future.
>> @@ -1309,12 +1316,14 @@ static int __exit ks_pcie_remove(struct platform_device *pdev)
>>   	struct device_link **link = ks_pcie->link;
>>   	int num_lanes = ks_pcie->num_lanes;
>>   	struct device *dev = &pdev->dev;
>> +	int irq = platform_get_irq(pdev, 0);
> I think it's better to save the irq we looked up in ks_pcie_probe()
> and free *that*.  It's probably the same thing you get by calling
> platform_get_irq() again, but it seems cleaner to me to save what we
> got in ks_pcie_probe().
Thanks for your guidance. I agree with saving the irq to make code cleaner.
I will change it in the next version.
>>   	pm_runtime_put(dev);
>>   	pm_runtime_disable(dev);
>>   	ks_pcie_disable_phy(ks_pcie);
>>   	while (num_lanes--)
>>   		device_link_del(link[num_lanes]);
>> +	free_irq(irq, ks_pcie);
>>   
>>   	return 0;
>>   }
>> -- 
>> 2.34.1
>>

Re: [PATCH] PCI: dwc: keystone: Free IRQ in `ks_pcie_remove` and the error handling section of `ks_pcie_probe`

[Bjorn Helgaas]

On Mon, May 22, 2023 at 02:07:10PM +0800, 曾祥翼 wrote:
> On 17/5/2023 03:49, Bjorn Helgaas wrote:
> > On Tue, May 16, 2023 at 01:16:59PM +0800, Xiangyi Zeng wrote:
> > ...

> > > Fixes: 0790eb175ee0 ("PCI: keystone: Cleanup error_irq configuration")
> > > Signed-off-by: Xiangyi Zeng <xyzeng@stu.xidian.edu.cn>
> > > Reviewed-by: Dongliang Mu <dzm91@hust.edu.cn>
> >
> > It's best if the Reviewed-by tag is not added until Dongliang sends
> > email with that tag directly to the mailing list.  Internal reviews
> > before posting to the mailing list aren't worth much.
>
> In our internal review process, only the patch with the Reviewed-by
> tag can be submitted to the mailing list. You can check this in our
> google group.
> https://groups.google.com/g/hust-os-kernel-patches/c/bt397rzVL24/m/l52XYbG4AgAJ
> We will consider omitting this tag when sending to the kernel mailing
> list in the future.

It's great that Dongliang has already reviewed it; I'm not suggesting
you omit any of that internal review.  My point is that it's best if
Dongliang's email review appears on linux-pci in addition to whatever
internal email list you use.  The hust-os-kernel-patches Google Group
is not archived and searchable the same way linux-pci@vger.kernel.org
is on https://lore.kernel.org/linux-pci/.

Bjorn

Re: [PATCH] PCI: dwc: keystone: Free IRQ in `ks_pcie_remove` and the error handling section of `ks_pcie_probe`

[Dan Carpenter]

Heh.  Presumably you sent this by mistake.  (Xiangyi Zeng sent a
completely different version to the HUST list a few moments ago).

regards,
dan carpenter