Skip to contentSkip to conversation

pc_piix: enable legacy hotplug for Xen

PATCH In review Michael S. Tsirkin 4 participants 11 messages Feb 4, 2014 – Feb 5, 2014

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 pc_piix: enable legacy hotplug for Xen

xenfv has no fwcfg and so does not load acpi from QEMU. as such new acpi features don't work.

---

Sander, could you pleas econfirm this works for you? Also, Xen really should switch to fw cfg some day ...

 hw/i386/pc_piix.c | 11 +++++++++++
 1 file changed, 11 insertions(+)
hw/i386/pc_piix.c +11 -0
index a327d71..1acd2b2 100644
@@ -793,6 +793,17 @@ static QEMUMachine xenfv_machine = {
     .max_cpus = HVM_MAX_VCPUS,
     .default_machine_opts = "accel=xen",
     .hot_add_cpu = pc_hot_add_cpu,
+    .compat_props = (GlobalProperty[]) {
+        /* xenfv has no fwcfg and so does not load acpi from QEMU.
+         * as such new acpi features don't work.
+         */
+        {
+            .driver   = "PIIX4_PM",
+            .property = "acpi-pci-hotplug-with-bridge-support",
+            .value    = "off",
+        },
+        { /* end of list */ }
+    },
 };
 #endif
 
Signature · 1 line
MST
#1 lore raw

Il 04/02/2014 22:32, Michael S. Tsirkin ha scritto:

Quoted 9 lines

xenfv has no fwcfg and so does not load acpi from QEMU. as such new acpi features don't work.

---

Sander, could you pleas econfirm this works for you? Also, Xen really should switch to fw cfg some day ...

Yeah, especially because nowadays "-M pc" works with Xen. It would be nice if fw_cfg could be enabled even for Xen on all machines except the legacy "-M xenfv".

Stefano, could you give it a shot now that you'll be soon done with 4.4?

Paolo

Quoted 26 lines
 hw/i386/pc_piix.c | 11 +++++++++++
 1 file changed, 11 insertions(+)
hw/i386/pc_piix.c +11 -0
index a327d71..1acd2b2 100644
@@ -793,6 +793,17 @@ static QEMUMachine xenfv_machine = {
     .max_cpus = HVM_MAX_VCPUS,
     .default_machine_opts = "accel=xen",
     .hot_add_cpu = pc_hot_add_cpu,
+    .compat_props = (GlobalProperty[]) {
+        /* xenfv has no fwcfg and so does not load acpi from QEMU.
+         * as such new acpi features don't work.
+         */
+        {
+            .driver   = "PIIX4_PM",
+            .property = "acpi-pci-hotplug-with-bridge-support",
+            .value    = "off",
+        },
+        { /* end of list */ }
+    },
 };
 #endif
 
#2 lore raw
Quoted 9 lines · you

Il 04/02/2014 22:32, Michael S. Tsirkin ha scritto:

Quoted 8 lines

xenfv has no fwcfg and so does not load acpi from QEMU. as such new acpi features don't work.

---

Sander, could you pleas econfirm this works for you?

Yes this works for me (still on top of the debug patches + pull of the branch igor mentoined), but without -global PIIX4_PM.acpi-pci-hotplug-with-bridge-support=off:

No fw cfg. Bailing out.
bsel: 0, bus: pci.0
bsel: 0, bus: pci.0
bsel: 0, bus: pci.0
bsel: 0, bus: pci.0

Thanks !
--
Sander

Quoted 36 lines
Quoted 1 line

Also, Xen really should switch to fw cfg some day ...

Yeah, especially because nowadays "-M pc" works with Xen. It would be nice if fw_cfg could be enabled even for Xen on all machines except the legacy "-M xenfv".

Stefano, could you give it a shot now that you'll be soon done with 4.4?

Paolo

Quoted 26 lines
 hw/i386/pc_piix.c | 11 +++++++++++
 1 file changed, 11 insertions(+)
hw/i386/pc_piix.c +11 -0
index a327d71..1acd2b2 100644
@@ -793,6 +793,17 @@ static QEMUMachine xenfv_machine = {
     .max_cpus = HVM_MAX_VCPUS,
     .default_machine_opts = "accel=xen",
     .hot_add_cpu = pc_hot_add_cpu,
+    .compat_props = (GlobalProperty[]) {
+        /* xenfv has no fwcfg and so does not load acpi from QEMU.
+         * as such new acpi features don't work.
+         */
+        {
+            .driver   = "PIIX4_PM",
+            .property = "acpi-pci-hotplug-with-bridge-support",
+            .value    = "off",
+        },
+        { /* end of list */ }
+    },
 };
 #endif
 
#3 lore raw
Quoted 12 lines · Sander Eikelenboom
Quoted 9 lines · you

Il 04/02/2014 22:32, Michael S. Tsirkin ha scritto:

Quoted 8 lines

xenfv has no fwcfg and so does not load acpi from QEMU. as such new acpi features don't work.

---

Sander, could you pleas econfirm this works for you?

Quoted 9 lines

Yes this works for me (still on top of the debug patches + pull of the branch igor mentoined), but without -global PIIX4_PM.acpi-pci-hotplug-with-bridge-support=off:

No fw cfg. Bailing out.
bsel: 0, bus: pci.0
bsel: 0, bus: pci.0
bsel: 0, bus: pci.0
bsel: 0, bus: pci.0

Sander, thanks for testing and reporting the issue in the first place.

Just to be clear: does the patch fix PCI passthrough completely or do you still need to pass -global PIIX4_PM.acpi-pci-hotplug-with-bridge-support=off?
Sorry but I couldn't quite understand your reply.

Quoted 43 lines

--
Sander

Quoted 36 lines
Quoted 1 line

Also, Xen really should switch to fw cfg some day ...

Yeah, especially because nowadays "-M pc" works with Xen. It would be nice if fw_cfg could be enabled even for Xen on all machines except the legacy "-M xenfv".

Stefano, could you give it a shot now that you'll be soon done with 4.4?

Paolo

Quoted 26 lines
 hw/i386/pc_piix.c | 11 +++++++++++
 1 file changed, 11 insertions(+)
hw/i386/pc_piix.c +11 -0
index a327d71..1acd2b2 100644
@@ -793,6 +793,17 @@ static QEMUMachine xenfv_machine = {
     .max_cpus = HVM_MAX_VCPUS,
     .default_machine_opts = "accel=xen",
     .hot_add_cpu = pc_hot_add_cpu,
+    .compat_props = (GlobalProperty[]) {
+        /* xenfv has no fwcfg and so does not load acpi from QEMU.
+         * as such new acpi features don't work.
+         */
+        {
+            .driver   = "PIIX4_PM",
+            .property = "acpi-pci-hotplug-with-bridge-support",
+            .value    = "off",
+        },
+        { /* end of list */ }
+    },
 };
 #endif
 
#4 lore raw
Quoted 23 lines · you
Quoted 12 lines · Sander Eikelenboom
Quoted 9 lines · you

Il 04/02/2014 22:32, Michael S. Tsirkin ha scritto:

Quoted 8 lines

xenfv has no fwcfg and so does not load acpi from QEMU. as such new acpi features don't work.

---

Sander, could you pleas econfirm this works for you?

Quoted 9 lines

Yes this works for me (still on top of the debug patches + pull of the branch igor mentoined), but without -global PIIX4_PM.acpi-pci-hotplug-with-bridge-support=off:

No fw cfg. Bailing out.
bsel: 0, bus: pci.0
bsel: 0, bus: pci.0
bsel: 0, bus: pci.0
bsel: 0, bus: pci.0

Quoted 6 lines

Sander, thanks for testing and reporting the issue in the first place.

Just to be clear: does the patch fix PCI passthrough completely or do you still need to pass -global PIIX4_PM.acpi-pci-hotplug-with-bridge-support=off?
Sorry but I couldn't quite understand your reply.

It's fixed with the patch, you don't need to pass that global parameter.
(i only tested it with igor's upcoming branch and the debug-patches (printk's only) applied though)

Quoted 43 lines
Quoted 43 lines

--
Sander

Quoted 36 lines
Quoted 1 line

Also, Xen really should switch to fw cfg some day ...

Yeah, especially because nowadays "-M pc" works with Xen. It would be nice if fw_cfg could be enabled even for Xen on all machines except the legacy "-M xenfv".

Stefano, could you give it a shot now that you'll be soon done with 4.4?

Paolo

Quoted 26 lines
 hw/i386/pc_piix.c | 11 +++++++++++
 1 file changed, 11 insertions(+)
hw/i386/pc_piix.c +11 -0
index a327d71..1acd2b2 100644
@@ -793,6 +793,17 @@ static QEMUMachine xenfv_machine = {
     .max_cpus = HVM_MAX_VCPUS,
     .default_machine_opts = "accel=xen",
     .hot_add_cpu = pc_hot_add_cpu,
+    .compat_props = (GlobalProperty[]) {
+        /* xenfv has no fwcfg and so does not load acpi from QEMU.
+         * as such new acpi features don't work.
+         */
+        {
+            .driver   = "PIIX4_PM",
+            .property = "acpi-pci-hotplug-with-bridge-support",
+            .value    = "off",
+        },
+        { /* end of list */ }
+    },
 };
 #endif
 
#5 lore raw
Quoted 25 lines · Sander Eikelenboom
Quoted 24 lines · Sander Eikelenboom
Quoted 12 lines
Quoted 9 lines · you

Il 04/02/2014 22:32, Michael S. Tsirkin ha scritto:

Quoted 8 lines

xenfv has no fwcfg and so does not load acpi from QEMU. as such new acpi features don't work.

---

Sander, could you pleas econfirm this works for you?

Quoted 9 lines

Yes this works for me (still on top of the debug patches + pull of the branch igor mentoined), but without -global PIIX4_PM.acpi-pci-hotplug-with-bridge-support=off:

No fw cfg. Bailing out.
bsel: 0, bus: pci.0
bsel: 0, bus: pci.0
bsel: 0, bus: pci.0
bsel: 0, bus: pci.0

Sander, thanks for testing and reporting the issue in the first place.

Quoted 7 lines
Quoted 4 lines

Just to be clear: does the patch fix PCI passthrough completely or do you still need to pass -global PIIX4_PM.acpi-pci-hotplug-with-bridge-support=off?
Sorry but I couldn't quite understand your reply.

It's fixed with the patch, you don't need to pass that global parameter.
(i only tested it with igor's upcoming branch and the debug-patches (printk's only) applied though)

OK, that's fine, thanks.

Quoted 45 lines
Quoted 43 lines
Quoted 43 lines

--
Sander

Quoted 36 lines
Quoted 1 line

Also, Xen really should switch to fw cfg some day ...

Yeah, especially because nowadays "-M pc" works with Xen. It would be nice if fw_cfg could be enabled even for Xen on all machines except the legacy "-M xenfv".

Stefano, could you give it a shot now that you'll be soon done with 4.4?

Paolo

Quoted 26 lines
 hw/i386/pc_piix.c | 11 +++++++++++
 1 file changed, 11 insertions(+)
hw/i386/pc_piix.c +11 -0
index a327d71..1acd2b2 100644
@@ -793,6 +793,17 @@ static QEMUMachine xenfv_machine = {
     .max_cpus = HVM_MAX_VCPUS,
     .default_machine_opts = "accel=xen",
     .hot_add_cpu = pc_hot_add_cpu,
+    .compat_props = (GlobalProperty[]) {
+        /* xenfv has no fwcfg and so does not load acpi from QEMU.
+         * as such new acpi features don't work.
+         */
+        {
+            .driver   = "PIIX4_PM",
+            .property = "acpi-pci-hotplug-with-bridge-support",
+            .value    = "off",
+        },
+        { /* end of list */ }
+    },
 };
 #endif
 
#6 lore raw
In reply to Paolo Bonzini
Quoted 8 lines · Paolo Bonzini

Il 04/02/2014 22:32, Michael S. Tsirkin ha scritto:

Quoted 7 lines

xenfv has no fwcfg and so does not load acpi from QEMU. as such new acpi features don't work.

---

Quoted 8 lines
Quoted 2 lines

Sander, could you pleas econfirm this works for you? Also, Xen really should switch to fw cfg some day ...

Yeah, especially because nowadays "-M pc" works with Xen. It would be nice if fw_cfg could be enabled even for Xen on all machines except the legacy "-M xenfv".

Stefano, could you give it a shot now that you'll be soon done with 4.4?

It might not be that straightforward as Xen emulates some devices in the hypervisor and patches have been recently posted to support multiple device models for a single VM.

Quoted 27 lines
Quoted 26 lines
 hw/i386/pc_piix.c | 11 +++++++++++
 1 file changed, 11 insertions(+)
hw/i386/pc_piix.c +11 -0
index a327d71..1acd2b2 100644
@@ -793,6 +793,17 @@ static QEMUMachine xenfv_machine = {
     .max_cpus = HVM_MAX_VCPUS,
     .default_machine_opts = "accel=xen",
     .hot_add_cpu = pc_hot_add_cpu,
+    .compat_props = (GlobalProperty[]) {
+        /* xenfv has no fwcfg and so does not load acpi from QEMU.
+         * as such new acpi features don't work.
+         */
+        {
+            .driver   = "PIIX4_PM",
+            .property = "acpi-pci-hotplug-with-bridge-support",
+            .value    = "off",
+        },
+        { /* end of list */ }
+    },
 };
 #endif
 
#7 lore raw

Il 05/02/2014 17:41, Stefano Stabellini ha scritto:

Quoted 9 lines
Quoted 5 lines

Yeah, especially because nowadays "-M pc" works with Xen. It would be nice if fw_cfg could be enabled even for Xen on all machines except the legacy "-M xenfv".

Stefano, could you give it a shot now that you'll be soon done with 4.4?

It might not be that straightforward as Xen emulates some devices in the hypervisor and patches have been recently posted to support multiple device models for a single VM.

I'm not sure how this is related to fw_cfg? fw_cfg is just a simple key-value store. If hvmloader doesn't use it, it should be a burden.

Paolo

#8 lore raw
Quoted 6 lines · Paolo Bonzini

Il 05/02/2014 17:41, Stefano Stabellini ha scritto:

Quoted 5 lines
Quoted 1 line

Yeah, especially because nowadays "-M pc" works with Xen. It would be

nice if

Quoted 3 lines

fw_cfg could be enabled even for Xen on all machines except the legacy "-M xenfv".

Quoted 8 lines
Quoted 5 lines
Quoted 1 line

Stefano, could you give it a shot now that you'll be soon done with 4.4?

It might not be that straightforward as Xen emulates some devices in the hypervisor and patches have been recently posted to support multiple device models for a single VM.

I'm not sure how this is related to fw_cfg? fw_cfg is just a simple key-value store. If hvmloader doesn't use it, it should be a burden.

I was referring to the fact that QEMU might not have all the info that need to be written to fw_cfg when running on Xen: some of the info might be in the hypervisor and some other in another QEMU instance. Or do you think that fw_cfg could be written by multiple entities simultaneously?

#9 lore raw

Il 05/02/2014 17:54, Stefano Stabellini ha scritto:

Quoted 7 lines
Quoted 2 lines
Quoted 2 lines

I'm not sure how this is related to fw_cfg? fw_cfg is just a simple key-value store. If hvmloader doesn't use it, it should be a burden.

I was referring to the fact that QEMU might not have all the info that need to be written to fw_cfg when running on Xen: some of the info might be in the hypervisor and some other in another QEMU instance. Or do you think that fw_cfg could be written by multiple entities simultaneously?

Wouldn't it be possible to fetch the necessary information via hypercalls or in xenstore?

It's possible that some particular fw_cfg entries will be skipped for Xen, but most of them should apply. fw_cfg support would also make it possible to use -kernel/-initrd in Xen HVM machines, for example.

Paolo

#10 lore raw
Quoted 4 lines · Paolo Bonzini

Il 05/02/2014 17:54, Stefano Stabellini ha scritto:

Quoted 3 lines
Quoted 3 lines
Quoted 1 line

I'm not sure how this is related to fw_cfg? fw_cfg is just a simple

key-value

Quoted 1 line

store. If hvmloader doesn't use it, it should be a burden.

Quoted 8 lines
Quoted 5 lines

I was referring to the fact that QEMU might not have all the info that need to be written to fw_cfg when running on Xen: some of the info might be in the hypervisor and some other in another QEMU instance. Or do you think that fw_cfg could be written by multiple entities simultaneously?

Wouldn't it be possible to fetch the necessary information via hypercalls or in xenstore?

Probably, at least for the single QEMU case.

Quoted 3 lines

It's possible that some particular fw_cfg entries will be skipped for Xen, but most of them should apply. fw_cfg support would also make it possible to use -kernel/-initrd in Xen HVM machines, for example.