Skip to contentSkip to conversation

virtio_pci: use irq to detect interrupt support

PATCH In review Michael S. Tsirkin 5 participants 6 messages Oct 12, 2022 – Oct 14, 2022 virtio core

lore
Share & tools
lore

Michael S. Tsirkin

author maintainer
#0 lore raw

PATCH 1/1 virtio_pci: use irq to detect interrupt support

commit 71491c54eafa ("virtio_pci: don't try to use intxif pin is zero") breaks virtio_pci on powerpc, when running as a qemu guest.

vp_find_vqs() bails out because pci_dev->pin == 0.

But pci_dev->irq is populated correctly, so vp_find_vqs_intx() would succeed if we called it - which is what the code used to do.

This seems to happen because pci_dev->pin is not populated in pci_assign_irq().

Which is absolutely a bug in the relevant PCI code, but it may also affect other platforms that use of_irq_parse_and_map_pci().

However Linus said:

	The correct way to check for "no irq" doesn't use NO_IRQ at all, it just does
		if (dev->irq) ...

so let's just check irq and be done with it.

Fixes 71491c54eafa ("virtio_pci: don't try to use intxif pin is zero")
Cc "Angus Chen" <[email protected]>

---

Build tested only - very late here. Angus any chance you could help test this? Thanks!

 drivers/virtio/virtio_pci_common.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
drivers/virtio/virtio_pci_common.c +2 -2
index 4df77eeb4d16..a6c86f916dbd 100644
@@ -409,8 +409,8 @@ int vp_find_vqs(struct virtio_device *vdev, unsigned int nvqs,
 	err = vp_find_vqs_msix(vdev, nvqs, vqs, callbacks, names, false, ctx, desc);
 	if (!err)
 		return 0;
-	/* Is there an interrupt pin? If not give up. */
-	if (!(to_vp_device(vdev)->pci_dev->pin))
+	/* Is there an interrupt? If not give up. */
+	if (!(to_vp_device(vdev)->pci_dev->irq))
 		return err;
 	/* Finally fall back to regular interrupts. */
 	return vp_find_vqs_intx(vdev, nvqs, vqs, callbacks, names, ctx);
Signature · 6 lines
MST

_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization
#1 lore raw
Quoted 3 lines · Michael S. Tsirkin

This seems to happen because pci_dev->pin is not populated in pci_assign_irq().

Yeah, I have to say, I wonder why it was looking at pci_dev->pin in the first place.

I was curious, as this is fairly odd.

Of course, I only did a fairly strange 'grep' for this thing, so I might have missed some other use:

    git grep -e '->pin\>' $(git grep -l 'struct pci_dev')

and the above will possibly find other uses of 'pin' as a member than the 'struct pci_dev', so I'm not going to claim the above is exhaustive, but looking at the results, I do see how ACPI has somewhat similar logic, and acpi_pci_irq_enable() does this:

        ...
        pin = dev->pin;
        if (!pin) {
                dev_dbg(&dev->dev, "No interrupt pin configured\n");
                return 0;
        }
        ...

but in the end that seems to be because it's then later using the pin to do the actual PCI IRQ routing table lookup, and then it uses that value to actually look up the IRQ number:

        dev->irq = rc;
        dev->irq_managed = 1;

and in fact all this code also has a "have I already looked this up" and then it doesn't do anything (but somewhat illogically, it does that *after* having tested for that "!pin" condition - I think it would make more sense to go "oh, I already have this interrupt mapped, I shouldn't care about the pin", but I suspect it doesn't matter in practice).

And I really think that that is basically the only time you should use that 'pci_dev->pin' thing: it basically exists not for "does this device have an IRQ", but for "what is the routing of this irq on this device".

There's also some testing of dev->pin in drives/pci/pci.c and drivers/pci/probe.c, but it seems to be very similar: it's actually doing the irq lookup, and the pin swizzling that goes along with it.

IOW, I think that yes, this patch makes sense, and virtio_pci was doing something odd. That virtio_pci driver at no point actually cares about the PCI pin, will not do any PCI irq routing lookup with it, it will just do

        err = request_irq(vp_dev->pci_dev->irq, [...]

using the pci_dev->irq that has already been looked up.

So the patch makes sense to me. If there is some problem with pci_dev->irq, then it's up to request_irq() to complain about it (ie we have things like IRQ_NOTCONNECTED etc, which is a more modern-day version of the old NO_IRQ thing). Again, not something that virtio_pci should check for itself, I think.

But I don't really know the virtio code. I can only say that "check the pin" pattern seems entirely wrong and should only be done by code that does actual irq routing, and "just check the irq" is what everybody else does.

Linus

Tested-by on v1 1/1
#2 lore raw
In reply to Michael S. Tsirkin
Quoted 24 lines · Michael S. Tsirkin

commit 71491c54eafa ("virtio_pci: don't try to use intxif pin is zero") breaks virtio_pci on powerpc, when running as a qemu guest.

vp_find_vqs() bails out because pci_dev->pin == 0.

But pci_dev->irq is populated correctly, so vp_find_vqs_intx() would succeed if we called it - which is what the code used to do.

This seems to happen because pci_dev->pin is not populated in pci_assign_irq().

Which is absolutely a bug in the relevant PCI code, but it may also affect other platforms that use of_irq_parse_and_map_pci().

However Linus said:

	The correct way to check for "no irq" doesn't use NO_IRQ at all, it just does
		if (dev->irq) ...

so let's just check irq and be done with it.

Fixes 71491c54eafa ("virtio_pci: don't try to use intxif pin is zero")
Cc "Angus Chen" <[email protected]>
Quoted 4 lines

---

Build tested only - very late here. Angus any chance you could help test this? Thanks!

This works for me on powerpc.

cheers

Quoted 17 lines
drivers/virtio/virtio_pci_common.c +2 -2
index 4df77eeb4d16..a6c86f916dbd 100644
@@ -409,8 +409,8 @@ int vp_find_vqs(struct virtio_device *vdev, unsigned int nvqs,
 	err = vp_find_vqs_msix(vdev, nvqs, vqs, callbacks, names, false, ctx, desc);
 	if (!err)
 		return 0;
-	/* Is there an interrupt pin? If not give up. */
-	if (!(to_vp_device(vdev)->pci_dev->pin))
+	/* Is there an interrupt? If not give up. */
+	if (!(to_vp_device(vdev)->pci_dev->irq))
 		return err;
 	/* Finally fall back to regular interrupts. */
 	return vp_find_vqs_intx(vdev, nvqs, vqs, callbacks, names, ctx);
Signature · 1 line
MST

_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

Acked-by on v1 1/1
#3 lore raw
In reply to Michael S. Tsirkin
Quoted 20 lines · Michael S. Tsirkin

commit 71491c54eafa ("virtio_pci: don't try to use intxif pin is zero") breaks virtio_pci on powerpc, when running as a qemu guest.

vp_find_vqs() bails out because pci_dev->pin == 0.

But pci_dev->irq is populated correctly, so vp_find_vqs_intx() would succeed if we called it - which is what the code used to do.

This seems to happen because pci_dev->pin is not populated in pci_assign_irq().

Which is absolutely a bug in the relevant PCI code, but it may also affect other platforms that use of_irq_parse_and_map_pci().

However Linus said:
The correct way to check for "no irq" doesn't use NO_IRQ at all, it just does
if (dev->irq) ...
so let's just check irq and be done with it.

Quoted 5 lines
Fixes 71491c54eafa ("virtio_pci: don't try to use intxif pin is zero")
Cc "Angus Chen" <[email protected]>
Quoted 26 lines

---

Build tested only - very late here. Angus any chance you could help test this? Thanks!

 drivers/virtio/virtio_pci_common.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
drivers/virtio/virtio_pci_common.c +2 -3
index 4df77eeb4d16..a6c86f916dbd 100644
@@ -409,8 +409,8 @@ int vp_find_vqs(struct virtio_device *vdev, unsigned int nvqs,
        err = vp_find_vqs_msix(vdev, nvqs, vqs, callbacks, names, false, ctx, desc);
        if (!err)
                return 0;
-       /* Is there an interrupt pin? If not give up. */
-       if (!(to_vp_device(vdev)->pci_dev->pin))
+       /* Is there an interrupt? If not give up. */
+       if (!(to_vp_device(vdev)->pci_dev->irq))
                return err;
        /* Finally fall back to regular interrupts. */
        return vp_find_vqs_intx(vdev, nvqs, vqs, callbacks, names, ctx);
--

MST

#4 lore raw
In reply to Michael S. Tsirkin

Hi Mst

Quoted 33 lines

-----Original Message-----
From: Michael S. Tsirkin <[email protected]>
Sent: Thursday, October 13, 2022 6:04 AM
To: [email protected]
Cc: Linus Torvalds <[email protected]>; Michael Ellerman <[email protected]>; Angus Chen <[email protected]>; Jason Wang <[email protected]>; [email protected]
Subject: [PATCH] virtio_pci: use irq to detect interrupt support

commit 71491c54eafa ("virtio_pci: don't try to use intxif pin is zero") breaks virtio_pci on powerpc, when running as a qemu guest.

vp_find_vqs() bails out because pci_dev->pin == 0.

But pci_dev->irq is populated correctly, so vp_find_vqs_intx() would succeed if we called it - which is what the code used to do.

This seems to happen because pci_dev->pin is not populated in pci_assign_irq().

Which is absolutely a bug in the relevant PCI code, but it may also affect other platforms that use of_irq_parse_and_map_pci().

However Linus said:

	The correct way to check for "no irq" doesn't use NO_IRQ at all, it just does
		if (dev->irq) ...

so let's just check irq and be done with it.

Fixes 71491c54eafa ("virtio_pci: don't try to use intxif pin is zero")
Cc "Angus Chen" <[email protected]>
Quoted 4 lines

---

Build tested only - very late here. Angus any chance you could help test this? Thanks!

I tested the patch ,it work well on x86。Inlcude legacy and modern driver. And I tested it on arm server also,and find some problem because of 0.9.5 limitation.
"platform bug: legacy virtio-pci must not be used with RAM above 0x%llxGB\n", But the error is not effected by our patch.with or without our patch ,it print the same.
And I test modern dirver,it work well also.
Sorry for the late reply.

Quoted 23 lines
 drivers/virtio/virtio_pci_common.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
drivers/virtio/virtio_pci_common.c +2 -2
index 4df77eeb4d16..a6c86f916dbd 100644
@@ -409,8 +409,8 @@ int vp_find_vqs(struct virtio_device *vdev, unsigned int
 nvqs,
 	err = vp_find_vqs_msix(vdev, nvqs, vqs, callbacks, names, false, ctx, desc);
 	if (!err)
 		return 0;
-	/* Is there an interrupt pin? If not give up. */
-	if (!(to_vp_device(vdev)->pci_dev->pin))
+	/* Is there an interrupt? If not give up. */
+	if (!(to_vp_device(vdev)->pci_dev->irq))
 		return err;
 	/* Finally fall back to regular interrupts. */

return vp_find_vqs_intx(vdev, nvqs, vqs, callbacks, names, ctx);
--
MST

author maintainer
#5 lore raw
Quoted 37 lines · Angus Chen

Hi Mst

Quoted 35 lines

-----Original Message-----
From: Michael S. Tsirkin <[email protected]>
Sent: Thursday, October 13, 2022 6:04 AM
To: [email protected]
Cc: Linus Torvalds <[email protected]>; Michael Ellerman <[email protected]>; Angus Chen <[email protected]>; Jason Wang <[email protected]>; [email protected]
Subject: [PATCH] virtio_pci: use irq to detect interrupt support

commit 71491c54eafa ("virtio_pci: don't try to use intxif pin is zero") breaks virtio_pci on powerpc, when running as a qemu guest.

vp_find_vqs() bails out because pci_dev->pin == 0.

But pci_dev->irq is populated correctly, so vp_find_vqs_intx() would succeed if we called it - which is what the code used to do.

This seems to happen because pci_dev->pin is not populated in pci_assign_irq().

Which is absolutely a bug in the relevant PCI code, but it may also affect other platforms that use of_irq_parse_and_map_pci().

However Linus said:

	The correct way to check for "no irq" doesn't use NO_IRQ at all, it just does
		if (dev->irq) ...

so let's just check irq and be done with it.

Fixes 71491c54eafa ("virtio_pci: don't try to use intxif pin is zero")
Cc "Angus Chen" <[email protected]>

---

Quoted 6 lines
Quoted 2 lines

Build tested only - very late here. Angus any chance you could help test this? Thanks!

I tested the patch ,it work well on x86。Inlcude legacy and modern driver. And I tested it on arm server also,and find some problem because of 0.9.5 limitation.
"platform bug: legacy virtio-pci must not be used with RAM above 0x%llxGB\n", But the error is not effected by our patch.with or without our patch ,it print the same.

Yes that's a limitation of 0.9.5 - just make a smaller VM to test legacy.

Quoted 2 lines

And I test modern dirver,it work well also.
Sorry for the late reply.

Great, thanks a lot!

Quoted 24 lines
Quoted 23 lines
 drivers/virtio/virtio_pci_common.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
drivers/virtio/virtio_pci_common.c +2 -2
index 4df77eeb4d16..a6c86f916dbd 100644
@@ -409,8 +409,8 @@ int vp_find_vqs(struct virtio_device *vdev, unsigned int
 nvqs,
 	err = vp_find_vqs_msix(vdev, nvqs, vqs, callbacks, names, false, ctx, desc);
 	if (!err)
 		return 0;
-	/* Is there an interrupt pin? If not give up. */
-	if (!(to_vp_device(vdev)->pci_dev->pin))
+	/* Is there an interrupt? If not give up. */
+	if (!(to_vp_device(vdev)->pci_dev->irq))
 		return err;
 	/* Finally fall back to regular interrupts. */

return vp_find_vqs_intx(vdev, nvqs, vqs, callbacks, names, ctx);
--
MST