Skip to contentSkip to conversation

hw/virtio: Add PCIe capability to virtio devices

RFC PATCH Superseded Marcel Apfelbaum 4 participants 6 messages Oct 12, 2015 – Oct 13, 2015

v1 of 5 Superseded Latest is v5 →

lore
Share & tools
lore
#0 lore raw

RFC PATCH 1/1 hw/virtio: Add PCIe capability to virtio devices

The virtio devices are converted to PCI-Express
if they are plugged into a PCI-Express bus and
the 'modern' protocol is enabled.

---

This is an RFC because all it does it adds the PCIe capability and nothing more. I post it early so I can get feedbacks on what is the best way to continue it.

Any comments would be appreciated,
Thanks,
Marcel

 hw/virtio/virtio-pci.c | 11 +++++++++++
 1 file changed, 11 insertions(+)
hw/virtio/virtio-pci.c +11 -0
index 6703806..f7c93d9 100644
@@ -1591,6 +1591,17 @@ static void virtio_pci_realize(PCIDevice *pci_dev, Error **errp)
 
     address_space_init(&proxy->modern_as, &proxy->modern_cfg, "virtio-pci-cfg-as");
 
+    if (!(proxy->flags & VIRTIO_PCI_FLAG_DISABLE_MODERN)
+        && pci_bus_is_express(pci_dev->bus)) {
+        int pos = pci_add_capability(pci_dev, PCI_CAP_ID_EXP, 0,
+                                     PCI_EXP_VER2_SIZEOF);
+
+        if (pos > 0) {
+            pci_dev->exp.exp_cap = pos;
+            pci_dev->cap_present |= QEMU_PCI_CAP_EXPRESS;
+        }
+    }
+
     virtio_pci_bus_new(&proxy->bus, sizeof(proxy->bus), proxy);
     if (k->realize) {
         k->realize(proxy, errp);
Signature · 1 line
2.1.0
#1 lore raw
Quoted 8 lines · Marcel Apfelbaum

The virtio devices are converted to PCI-Express
if they are plugged into a PCI-Express bus and
the 'modern' protocol is enabled.

---

This is an RFC because all it does it adds the PCIe capability and nothing more.

Express capability is easy.
But if you go over express space you will see that a bunch of other capabilities are required, such as PM capability etc. These might need more work.

Quoted 14 lines

I post it early so I can get feedbacks on what is the best way to continue it.

Any comments would be appreciated,
Thanks,
Marcel

 hw/virtio/virtio-pci.c | 11 +++++++++++
 1 file changed, 11 insertions(+)
hw/virtio/virtio-pci.c +0 -0
index 6703806..f7c93d9 100644

hw/virtio/virtio-pci.c static void virtio_pci_realize(PCIDevice *pci_dev, Error **errp) · line 1595

Quoted 5 lines
hw/virtio/virtio-pci.c +2 -0
@@ -1591,6 +1591,17 @@ static void virtio_pci_realize(PCIDevice *pci_dev, Error **errp)
 
     address_space_init(&proxy->modern_as, &proxy->modern_cfg, "virtio-pci-cfg-as");
 
+    if (!(proxy->flags & VIRTIO_PCI_FLAG_DISABLE_MODERN)
+        && pci_bus_is_express(pci_dev->bus)) {

One point: we probably want to avoid doing this for integrated devices on root bus. Does pci_bus_is_express return true there?

hw/virtio/virtio-pci.c static void virtio_pci_realize(PCIDevice *pci_dev, Error **errp) · line 1599

Quoted 4 lines
hw/virtio/virtio-pci.c +4 -0
+        int pos = pci_add_capability(pci_dev, PCI_CAP_ID_EXP, 0,
+                                     PCI_EXP_VER2_SIZEOF);
+
+        if (pos > 0) {

We probably want to assert on pos < 0 instead.
That implies a code bug.

Quoted 10 lines
hw/virtio/virtio-pci.c +5 -0
+            pci_dev->exp.exp_cap = pos;
+            pci_dev->cap_present |= QEMU_PCI_CAP_EXPRESS;
+        }
+    }
+
     virtio_pci_bus_new(&proxy->bus, sizeof(proxy->bus), proxy);
     if (k->realize) {
         k->realize(proxy, errp);
Signature · 1 line
2.1.0
#2 lore raw
Quoted 8 lines · Michael S. Tsirkin
Quoted 7 lines · Marcel Apfelbaum

The virtio devices are converted to PCI-Express
if they are plugged into a PCI-Express bus and
the 'modern' protocol is enabled.

---

Quoted 6 lines
Quoted 1 line

This is an RFC because all it does it adds the PCIe capability and nothing more.

Express capability is easy.
But if you go over express space you will see that a bunch of other capabilities are required, such as PM capability etc. These might need more work.

Also what about the legacy io bar? I guess we'd better avoid that for express devices.

Maybe it makes sense to add virtio-*-pcie devices (virtio-1.0 only, with pcie caps)?

cheers,
Gerd

#3 lore raw
Quoted 11 lines · Gerd Hoffmann
Quoted 10 lines · Michael S. Tsirkin
Quoted 8 lines · Marcel Apfelbaum

The virtio devices are converted to PCI-Express
if they are plugged into a PCI-Express bus and
the 'modern' protocol is enabled.

---

This is an RFC because all it does it adds the PCIe capability and nothing more.

Quoted 7 lines
Quoted 4 lines

Express capability is easy.
But if you go over express space you will see that a bunch of other capabilities are required, such as PM capability etc. These might need more work.

Also what about the legacy io bar? I guess we'd better avoid that for express devices.

This needs some thought. We definitely still want a way to enable it I think, e.g. for old guests.

Quoted 5 lines

Maybe it makes sense to add virtio-*-pcie devices (virtio-1.0 only, with pcie caps)?

cheers,
Gerd

Personally I think it's ugly enough to remember the need to specify -pci. OTOH the need to specify upstream and downstream ports is even uglier. Any idea how to address both issues at the same time?

Signature · 1 line
MST
#4 lore raw
In reply to Michael S. Tsirkin
Quoted 8 lines · Michael S. Tsirkin
Quoted 7 lines · Marcel Apfelbaum

The virtio devices are converted to PCI-Express
if they are plugged into a PCI-Express bus and
the 'modern' protocol is enabled.

Signed-off-by Marcel Apfelbaum <[email protected] <javascript:;>>

---

Quoted 7 lines
Quoted 1 line

This is an RFC because all it does it adds the PCIe capability and

nothing more.

Express capability is easy.
But if you go over express space you will see that a bunch of other capabilities are required, such as PM capability etc. These might need more work.

Sure, I'll look on the PCIe spec for the minimum requirements.

Quoted 18 lines
Quoted 2 lines

I post it early so I can get feedbacks on what is the best way to

continue it.

hw/virtio/virtio-pci.c static void virtio_pci_realize(PCIDevice

Quoted 13 lines

Any comments would be appreciated,
Thanks,
Marcel

 hw/virtio/virtio-pci.c | 11 +++++++++++
 1 file changed, 11 insertions(+)
hw/virtio/virtio-pci.c +0 -0
index 6703806..f7c93d9 100644
@@ -1591,6 +1591,17 @@ static void virtio_pci_realize(PCIDevice

*pci_dev, Error **errp)

Quoted 1 line
Quoted 8 lines
Quoted 1 line
     address_space_init(&proxy->modern_as, &proxy->modern_cfg,

"virtio-pci-cfg-as");

hw/virtio/virtio-pci.c static void virtio_pci_realize(PCIDevice *pci_dev, Error **errp) · line 1595

Quoted 3 lines
hw/virtio/virtio-pci.c +2 -0
+    if (!(proxy->flags & VIRTIO_PCI_FLAG_DISABLE_MODERN)
+        && pci_bus_is_express(pci_dev->bus)) {

One point: we probably want to avoid doing this for integrated devices on root bus. Does pci_bus_is_express return true there?

Hmm, I'll check, but I think it does. Is this a must?

Quoted 10 lines

hw/virtio/virtio-pci.c static void virtio_pci_realize(PCIDevice *pci_dev, Error **errp) · line 1599

Quoted 5 lines
hw/virtio/virtio-pci.c +4 -0
+        int pos = pci_add_capability(pci_dev, PCI_CAP_ID_EXP, 0,
+                                     PCI_EXP_VER2_SIZEOF);
+
+        if (pos > 0) {

We probably want to assert on pos < 0 instead.
That implies a code bug.

I was wondering what to do here, thanks for the idea!

Thanks,
Marcel

Quoted 13 lines
Quoted 11 lines
hw/virtio/virtio-pci.c +5 -1
+            pci_dev->exp.exp_cap = pos;
+            pci_dev->cap_present |= QEMU_PCI_CAP_EXPRESS;
+        }
+    }
+
     virtio_pci_bus_new(&proxy->bus, sizeof(proxy->bus), proxy);
     if (k->realize) {
         k->realize(proxy, errp);
--

2.1.0

#5 lore raw
Quoted 52 lines · Marcel Apfelbaum

On Monday, October 12, 2015, Michael S. Tsirkin <[email protected]> wrote:

On Mon, Oct 12, 2015 at 06:17:54PM +0300, Marcel Apfelbaum wrote:
> The virtio devices are converted to PCI-Express
> if they are plugged into a PCI-Express bus and
> the 'modern' protocol is enabled.
>
> Signed-off-by: Marcel Apfelbaum <[email protected]>
> ---
>
> This is an RFC because all it does it adds the PCIe capability and nothing more.

Express capability is easy.
But if you go over express space you will see that a bunch of other capabilities are required, such as PM capability etc. These might need more work.

Sure, I'll look on the PCIe spec for the minimum requirements.

Quoted 1 line

I post it early so I can get feedbacks on what is the best way to

    continue it.
    >
    > Any comments would be appreciated,
    > Thanks,
    > Marcel
    >
    >  hw/virtio/virtio-pci.c | 11 +++++++++++
    >  1 file changed, 11 insertions(+)
    >
    > diff --git a/hw/virtio/virtio-pci.c b/hw/virtio/virtio-pci.c
    > index 6703806..f7c93d9 100644
    > --- a/hw/virtio/virtio-pci.c
    > +++ b/hw/virtio/virtio-pci.c
    > @@ -1591,6 +1591,17 @@ static void virtio_pci_realize(PCIDevice *pci_dev,
    Error **errp)
    >
    >      address_space_init(&proxy->modern_as, &proxy->modern_cfg,
    "virtio-pci-cfg-as");
    >
    > +    if (!(proxy->flags & VIRTIO_PCI_FLAG_DISABLE_MODERN)
    > +        && pci_bus_is_express(pci_dev->bus)) {

    One point: we probably want to avoid doing this for integrated
    devices on root bus. Does pci_bus_is_express return true there?
Quoted 1 line

Hmm, I'll check, but I think it does. Is this a must?

It's probably a smart thing to do because you can't e.g. hot-unplug them. So supporting PCI E capability for integrated devices is probably more work than just making the PCI devices (which is not spec compliant but very popular so guests support this).

Quoted 32 lines
Quoted 6 lines

+        int pos = pci_add_capability(pci_dev, PCI_CAP_ID_EXP, 0, +                                     PCI_EXP_VER2_SIZEOF); +
+        if (pos > 0) {

We probably want to assert on pos < 0 instead. That implies a code bug.

I was wondering what to do here, thanks for the idea!

Thanks,
Marcel

Quoted 10 lines

+            pci_dev->exp.exp_cap = pos;
+            pci_dev->cap_present |= QEMU_PCI_CAP_EXPRESS; +        }
+    }
+
      virtio_pci_bus_new(&proxy->bus, sizeof(proxy->bus), proxy); if (k->realize) {
          k->realize(proxy, errp);
--
2.1.0