char: xillybus: Replace deprecated MSI API

Replace deprecated pci_enable_msi() with pci_alloc_irq_vectors(). And add pci_free_irq_vectors() to free the allocated vectors before removing the device.

---
 drivers/char/xillybus/xillybus_pcie.c | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

index 9858711e3e79..fc7bffbd36ed 100644
@@ -76,7 +76,8 @@ static int xilly_probe(struct pci_dev *pdev,
 	pci_set_master(pdev);
 
 	/* Set up a single MSI interrupt */
-	if (pci_enable_msi(pdev)) {
+	rc = pci_alloc_irq_vectors(pdev, 1, 1, PCI_IRQ_ALL_TYPES);
+	if (rc < 0) {
 		dev_err(endpoint->dev,
 			"Failed to enable MSI interrupts. Aborting.\n");
 		return -ENODEV;

@@ -112,6 +113,8 @@ static void xilly_remove(struct pci_dev *pdev)
 {
 	struct xilly_endpoint *endpoint = pci_get_drvdata(pdev);
 
+	pci_free_irq_vectors(pdev);
+
 	xillybus_endpoint_remove(endpoint);
 }
 

2.43.0

Hello,

Thanks for your patch.

I can see that you've chosen PCI_IRQ_ALL_TYPES as the selector of IRQ types. As far as I understand, this changes the behavior of the driver, allowing interrupt types that it shouldn't...?

Another thing: I have to admit that I missed the need for a call to pci_disable_msi() when the device shuts down (which is what pci_free_irq_vectors() effectively does). Thanks for bringing that to my attention.

However, as this is needed, it must be assured that this call is made in any case that the device's initialization fails. This driver is using the managed device API, so I suppose the correct way is to add an devm action for pci_free_irq_vectors(). It appears to me that just adding it to the remove method won't be sufficient.

Regards,
Eli

Replace deprecated pci_enable_msi() with pci_alloc_irq_vectors(). And add devm action to free irq vectors.

---
Changes in v2:
- Replace PCI_IRQ_ALL_TYPES with PCI_IRQ_MSI
- Delete pci_free_irq_vectors(pdev) in remove function
- Add devm action that calls pci_free_irq_vectors(pdev)

 drivers/char/xillybus/xillybus_pcie.c | 17 ++++++++++++++++-
 1 file changed, 16 insertions(+), 1 deletion(-)

index 9858711e3e79..ed3b77cb8127 100644
@@ -32,6 +32,11 @@ static const struct pci_device_id xillyids[] = {
 	{ /* End: all zeroes */ }
 };
 
+static void xilly_pci_free_irq_vectors(void *data)
+{
+	pci_free_irq_vectors(data);
+}
+
 static int xilly_probe(struct pci_dev *pdev,
 		       const struct pci_device_id *ent)
 {

@@ -76,11 +81,21 @@ static int xilly_probe(struct pci_dev *pdev,
 	pci_set_master(pdev);
 
 	/* Set up a single MSI interrupt */
-	if (pci_enable_msi(pdev)) {
+	rc = pci_alloc_irq_vectors(pdev, 1, 1, PCI_IRQ_MSI);
+	if (rc < 0) {
 		dev_err(endpoint->dev,
 			"Failed to enable MSI interrupts. Aborting.\n");
 		return -ENODEV;
 	}
+	
+	rc = devm_add_action(&pdev->dev, xilly_pci_free_irq_vectors, pdev);
+       	if (rc) {
+		dev_err(endpoint->dev,
+			"Failed to add devm action. Aborting.\n");
+		pci_free_irq_vectors(pdev);
+		return -ENODEV;
+	}
+
 	rc = devm_request_irq(&pdev->dev, pdev->irq, xillybus_isr, 0,
 			      xillyname, endpoint);
 	if (rc) {

2.43.0

Replace deprecated pci_enable_msi() with pci_alloc_irq_vectors(). And add devm action to free irq vectors.

---
Changes in v3:
- Some checkpatch cleanups

Changes in v2:
- Replace PCI_IRQ_ALL_TYPES with PCI_IRQ_MSI
- Delete pci_free_irq_vectors(pdev) in remove function
- Add devm action that calls pci_free_irq_vectors(pdev)

 drivers/char/xillybus/xillybus_pcie.c | 17 ++++++++++++++++-
 1 file changed, 16 insertions(+), 1 deletion(-)

index 9858711e3e79..373b3ccd2e8f 100644
@@ -32,6 +32,11 @@ static const struct pci_device_id xillyids[] = {
 	{ /* End: all zeroes */ }
 };
 
+static void xilly_pci_free_irq_vectors(void *data)
+{
+	pci_free_irq_vectors(data);
+}
+
 static int xilly_probe(struct pci_dev *pdev,
 		       const struct pci_device_id *ent)
 {

@@ -76,11 +81,21 @@ static int xilly_probe(struct pci_dev *pdev,
 	pci_set_master(pdev);
 
 	/* Set up a single MSI interrupt */
-	if (pci_enable_msi(pdev)) {
+	rc = pci_alloc_irq_vectors(pdev, 1, 1, PCI_IRQ_MSI);
+	if (rc < 0) {
 		dev_err(endpoint->dev,
 			"Failed to enable MSI interrupts. Aborting.\n");
 		return -ENODEV;
 	}
+
+	rc = devm_add_action(&pdev->dev, xilly_pci_free_irq_vectors, pdev);
+	if (rc) {
+		dev_err(endpoint->dev,
+			"Failed to add devm action. Aborting.\n");
+		pci_free_irq_vectors(pdev);
+		return -ENODEV;
+	}
+
 	rc = devm_request_irq(&pdev->dev, pdev->irq, xillybus_isr, 0,
 			      xillyname, endpoint);
 	if (rc) {

2.43.0

Hello,

Replace deprecated pci_enable_msi() with pci_alloc_irq_vectors().

---
Changes in v4:
- Drop devm_add_action() since it is useless

Changes in v3:
- Some checkpatch cleanups

Changes in v2:
- Replace PCI_IRQ_ALL_TYPES with PCI_IRQ_MSI
- Delete pci_free_irq_vectors(pdev) in remove function
- Add devm action that calls pci_free_irq_vectors(pdev)

 drivers/char/xillybus/xillybus_pcie.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

index 9858711e3e79..c8b4cdfe695a 100644
@@ -76,7 +76,8 @@ static int xilly_probe(struct pci_dev *pdev,
 	pci_set_master(pdev);
 
 	/* Set up a single MSI interrupt */
-	if (pci_enable_msi(pdev)) {
+	rc = pci_alloc_irq_vectors(pdev, 1, 1, PCI_IRQ_MSI);
+	if (rc < 0) {
 		dev_err(endpoint->dev,
 			"Failed to enable MSI interrupts. Aborting.\n");
 		return -ENODEV;

2.43.0

Thanks, and sorry for the back-and-forth.

You are welcome

Best regards,
Salah Triki

This says what you are doing, but not _WHY_ you are doing it.

If this was a simple search/replace, why would it not have been done already? Are you _SURE_ this is correct, and if so, why? You need to prove it here...

thanks,

greg k-h

I did the replacement because pci_enable_msi() is deprecated, isn't that enough ?

Best regards,
Salah Triki

If it was a simple search/replace, why wouldn't have been done already?

Again, you need to prove why this is ok at all. pci_enable_msi() shouldn't be used in new drivers, but what's wrong with it being in existing drivers? Especially in ones that you can't test to verify it still works after changing the code?

thanks,

greg k-h

Do you agree that for someone who wants to contribute to the kernel by doing cleanups, it is better to do these cleanups on a part that is not directly related to the drivers, because it is necessary to have hardware to test the changes.

Best regards,
Salah Triki

I'm sorry, but I have no context here for what you are asking about.

Please always quote emails properly, and realize that some of us get hundreds, if not thousands, of emails a day to deal with.

thanks,

greg k-h