Skip to contentSkip to conversation

Reversions to fix PCIe in virtio

PATCH In review Dr. David Alan Gilbert (git) 4 participants 11 messages Jul 29, 2019

lore
Share & tools
lore
#0 lore raw

v2 cover Reversions to fix PCIe in virtio

From: "Dr. David Alan Gilbert" <[email protected]>

Revert a couple of patches that break PCIe capabilities in virtio devices. The 'optional' revert is just reverted to make the main reversion trivial.

Symptom:
Loss of PCIe capabilities in virtio devices hung off PCIe bridges

Dr. David Alan Gilbert (2):
Revert "Revert "globals: Allow global properties to be optional"" Revert "hw: report invalid disable-legacy|modern usage for
virtio-1-only devs"

 hw/core/machine.c             | 23 +++--------------------
 hw/display/virtio-gpu-pci.c   |  4 +---
 hw/display/virtio-vga.c       |  4 +---
 hw/virtio/virtio-crypto-pci.c |  4 +---
 hw/virtio/virtio-input-pci.c  |  4 +---
 hw/virtio/virtio-pci.c        | 26 ++++++++++----------------
 hw/virtio/virtio-pci.h        | 31 ++++++-------------------------
 include/hw/qdev-core.h        |  3 +++
 qom/object.c                  |  3 +++
 9 files changed, 29 insertions(+), 73 deletions(-)
Signature · 1 line
2.21.0
#1 lore raw

PATCH v2 1/2 Revert "Revert "globals: Allow global properties to be optional""

From: "Dr. David Alan Gilbert" <[email protected]>

This reverts commit 8fa70dbd8bb478d9483c1da3e9976a2d86b3f9a0.

Because we're about to revert it's neighbour and thus uses an optional again.

---
 include/hw/qdev-core.h | 3 +++
 qom/object.c           | 3 +++
 2 files changed, 6 insertions(+)
include/hw/qdev-core.h +3 -0
index e157fc4acd..136df7774c 100644
@@ -252,6 +252,8 @@ struct PropertyInfo {
 /**
  * GlobalProperty:
  * @used: Set to true if property was used when initializing a device.
+ * @optional: If set to true, GlobalProperty will be skipped without errors
+ *            if the property doesn't exist.
  *
  * An error is fatal for non-hotplugged devices, when the global is applied.
  */
@@ -260,6 +262,7 @@ typedef struct GlobalProperty {
     const char *property;
     const char *value;
     bool used;
+    bool optional;
 } GlobalProperty;
 
 static inline void
qom/object.c +3 -0
index 3966a3d461..1555547727 100644
@@ -386,6 +386,9 @@ void object_apply_global_props(Object *obj, const GPtrArray *props, Error **errp
         if (object_dynamic_cast(obj, p->driver) == NULL) {
             continue;
         }
+        if (p->optional && !object_property_find(obj, p->property, NULL)) {
+            continue;
+        }
         p->used = true;
         object_property_parse(obj, p->value, p->property, &err);
         if (err != NULL) {
Signature · 1 line
2.21.0
Reviewed-by on v2 1/2
#2 lore raw

Re: [Qemu-devel] [PATCH v2 1/2] Revert "Revert "globals: Allow global properties to be optional""

Quoted 4 lines · Dr. David Alan Gilbert (git)

From: "Dr. David Alan Gilbert" <[email protected]>

This reverts commit 8fa70dbd8bb478d9483c1da3e9976a2d86b3f9a0.

Quoted 8 lines

Because we're about to revert it's neighbour and thus uses an optional again.

---
 include/hw/qdev-core.h | 3 +++
 qom/object.c           | 3 +++
 2 files changed, 6 insertions(+)

Regards,
Daniel

Signature · 3 lines
|: https://berrange.com      -o-    https://www.flickr.com/photos/dberrange :|
|: https://libvirt.org         -o-            https://fstop138.berrange.com :|
|: https://entangle-photo.org    -o-    https://www.instagram.com/dberrange :|
Reviewed-by on v2 1/2
#3 lore raw

Re: [Qemu-devel] [PATCH v2 1/2] Revert "Revert "globals: Allow global properties to be optional""

In reply to Dr. David Alan Gilbert (git)
Quoted 4 lines · On Mon, 29 Jul 2019 17:29:02 +0100 "Dr. David Alan Gilbert (git)"…

From: "Dr. David Alan Gilbert" <[email protected]>

This reverts commit 8fa70dbd8bb478d9483c1da3e9976a2d86b3f9a0.

Quoted 8 lines

Because we're about to revert it's neighbour and thus uses an optional again.

---
 include/hw/qdev-core.h | 3 +++
 qom/object.c           | 3 +++
 2 files changed, 6 insertions(+)

As discussed on IRC, the 'bugs' mentioned in that commit were related to the other patch that is being reverted; so I don't really mind it for 4.1 (IOW, either of the two revert approaches should be fine.)

#4 lore raw

PATCH v2 2/2 Revert "hw: report invalid disable-legacy|modern usage for virtio-1-only devs"

In reply to Dr. David Alan Gilbert (git)

From: "Dr. David Alan Gilbert" <[email protected]>

This reverts commit f2784eed306449c3d04a71a05ed6463b8289aedf since that accidentally removes the PCIe capabilities from virtio devices because virtio_pci_dc_realize is called before the new 'mode' flag is set.

---
 hw/core/machine.c             | 23 +++--------------------
 hw/display/virtio-gpu-pci.c   |  4 +---
 hw/display/virtio-vga.c       |  4 +---
 hw/virtio/virtio-crypto-pci.c |  4 +---
 hw/virtio/virtio-input-pci.c  |  4 +---
 hw/virtio/virtio-pci.c        | 26 ++++++++++----------------
 hw/virtio/virtio-pci.h        | 31 ++++++-------------------------
 7 files changed, 23 insertions(+), 73 deletions(-)
hw/core/machine.c +3 -20
index c58a8e594e..c4a2ab2282 100644
@@ -115,26 +115,9 @@ const size_t hw_compat_2_7_len = G_N_ELEMENTS(hw_compat_2_7);
 
 GlobalProperty hw_compat_2_6[] = {
     { "virtio-mmio", "format_transport_address", "off" },
-    /*
-     * don't include devices which are modern-only
-     * ie keyboard, mouse, tablet, gpu, vga & crypto
-     */
-    { "virtio-9p-pci", "disable-modern", "on" },
-    { "virtio-9p-pci", "disable-legacy", "off" },
-    { "virtio-balloon-pci", "disable-modern", "on" },
-    { "virtio-balloon-pci", "disable-legacy", "off" },
-    { "virtio-blk-pci", "disable-modern", "on" },
-    { "virtio-blk-pci", "disable-legacy", "off" },
-    { "virtio-input-host-pci", "disable-modern", "on" },
-    { "virtio-input-host-pci", "disable-legacy", "off" },
-    { "virtio-net-pci", "disable-modern", "on" },
-    { "virtio-net-pci", "disable-legacy", "off" },
-    { "virtio-rng-pci", "disable-modern", "on" },
-    { "virtio-rng-pci", "disable-legacy", "off" },
-    { "virtio-scsi-pci", "disable-modern", "on" },
-    { "virtio-scsi-pci", "disable-legacy", "off" },
-    { "virtio-serial-pci", "disable-modern", "on" },
-    { "virtio-serial-pci", "disable-legacy", "off" },
+    /* Optional because not all virtio-pci devices support legacy mode */
+    { "virtio-pci", "disable-modern", "on",  .optional = true },
+    { "virtio-pci", "disable-legacy", "off", .optional = true },
 };
 const size_t hw_compat_2_6_len = G_N_ELEMENTS(hw_compat_2_6);
 
hw/display/virtio-gpu-pci.c +1 -3
index d6f01b4a98..e4c7eb6193 100644
@@ -33,9 +33,7 @@ static void virtio_gpu_pci_base_realize(VirtIOPCIProxy *vpci_dev, Error **errp)
     Error *local_error = NULL;
 
     qdev_set_parent_bus(vdev, BUS(&vpci_dev->bus));
-    if (!virtio_pci_force_virtio_1(vpci_dev, errp)) {
-        return;
-    }
+    virtio_pci_force_virtio_1(vpci_dev);
     object_property_set_bool(OBJECT(vdev), true, "realized", &local_error);
 
     if (local_error) {
hw/display/virtio-vga.c +1 -3
index 416e7fec87..79a145e284 100644
@@ -137,9 +137,7 @@ static void virtio_vga_base_realize(VirtIOPCIProxy *vpci_dev, Error **errp)
 
     /* init virtio bits */
     qdev_set_parent_bus(DEVICE(g), BUS(&vpci_dev->bus));
-    if (!virtio_pci_force_virtio_1(vpci_dev, errp)) {
-        return;
-    }
+    virtio_pci_force_virtio_1(vpci_dev);
     object_property_set_bool(OBJECT(g), true, "realized", &err);
     if (err) {
         error_propagate(errp, err);
hw/virtio/virtio-crypto-pci.c +1 -3
index c8a2317a10..91d4446080 100644
@@ -53,9 +53,7 @@ static void virtio_crypto_pci_realize(VirtIOPCIProxy *vpci_dev, Error **errp)
     }
 
     qdev_set_parent_bus(vdev, BUS(&vpci_dev->bus));
-    if (!virtio_pci_force_virtio_1(vpci_dev, errp)) {
-        return;
-    }
+    virtio_pci_force_virtio_1(vpci_dev);
     object_property_set_bool(OBJECT(vdev), true, "realized", errp);
     object_property_set_link(OBJECT(vcrypto),
                  OBJECT(vcrypto->vdev.conf.cryptodev), "cryptodev",
hw/virtio/virtio-input-pci.c +1 -3
index 1c40292abc..ad7774e93e 100644
@@ -49,9 +49,7 @@ static void virtio_input_pci_realize(VirtIOPCIProxy *vpci_dev, Error **errp)
     DeviceState *vdev = DEVICE(&vinput->vdev);
 
     qdev_set_parent_bus(vdev, BUS(&vpci_dev->bus));
-    if (!virtio_pci_force_virtio_1(vpci_dev, errp)) {
-        return;
-    }
+    virtio_pci_force_virtio_1(vpci_dev);
     object_property_set_bool(OBJECT(vdev), true, "realized", errp);
 }
 
hw/virtio/virtio-pci.c +10 -16
index ce928f2429..f6d2223e78 100644
@@ -1723,22 +1723,16 @@ static void virtio_pci_realize(PCIDevice *pci_dev, Error **errp)
                        /* PCI BAR regions must be powers of 2 */
                        pow2ceil(proxy->notify.offset + proxy->notify.size));
 
-    if ((proxy->disable_legacy == ON_OFF_AUTO_ON) ||
-        ((proxy->disable_legacy == ON_OFF_AUTO_AUTO) && pcie_port)) {
-        if (proxy->disable_modern) {
-            error_setg(errp, "device cannot work as neither modern nor "
-                       "legacy mode is enabled");
-            error_append_hint(errp, "Set either disable-modern or "
-                              "disable-legacy to off\n");
-            return;
-        }
-        proxy->mode = VIRTIO_PCI_MODE_MODERN;
-    } else {
-        if (proxy->disable_modern) {
-            proxy->mode = VIRTIO_PCI_MODE_LEGACY;
-        } else {
-            proxy->mode = VIRTIO_PCI_MODE_TRANSITIONAL;
-        }
+    if (proxy->disable_legacy == ON_OFF_AUTO_AUTO) {
+        proxy->disable_legacy = pcie_port ? ON_OFF_AUTO_ON : ON_OFF_AUTO_OFF;
+    }
+
+    if (!virtio_pci_modern(proxy) && !virtio_pci_legacy(proxy)) {
+        error_setg(errp, "device cannot work as neither modern nor legacy mode"
+                   " is enabled");
+        error_append_hint(errp, "Set either disable-modern or disable-legacy"
+                          " to off\n");
+        return;
     }
 
     if (pcie_port && pci_is_express(pci_dev)) {
hw/virtio/virtio-pci.h +6 -25
index 619d9098c1..292275acb1 100644
@@ -15,7 +15,6 @@
 #ifndef QEMU_VIRTIO_PCI_H
 #define QEMU_VIRTIO_PCI_H
 
-#include "qapi/error.h"
 #include "hw/pci/msi.h"
 #include "hw/virtio/virtio-bus.h"
 
@@ -119,12 +118,6 @@ typedef struct VirtIOPCIQueue {
   uint32_t used[2];
 } VirtIOPCIQueue;
 
-typedef enum {
-    VIRTIO_PCI_MODE_LEGACY,
-    VIRTIO_PCI_MODE_TRANSITIONAL,
-    VIRTIO_PCI_MODE_MODERN,
-} VirtIOPCIMode;
-
 struct VirtIOPCIProxy {
     PCIDevice pci_dev;
     MemoryRegion bar;
@@ -149,7 +142,6 @@ struct VirtIOPCIProxy {
     bool disable_modern;
     bool ignore_backend_features;
     OnOffAuto disable_legacy;
-    VirtIOPCIMode mode;
     uint32_t class_code;
     uint32_t nvectors;
     uint32_t dfselect;
@@ -164,34 +156,23 @@ struct VirtIOPCIProxy {
 
 static inline bool virtio_pci_modern(VirtIOPCIProxy *proxy)
 {
-    return proxy->mode != VIRTIO_PCI_MODE_LEGACY;
+    return !proxy->disable_modern;
 }
 
 static inline bool virtio_pci_legacy(VirtIOPCIProxy *proxy)
 {
-    return proxy->mode != VIRTIO_PCI_MODE_MODERN;
+    return proxy->disable_legacy == ON_OFF_AUTO_OFF;
 }
 
-static inline bool virtio_pci_force_virtio_1(VirtIOPCIProxy *proxy,
-                                             Error **errp)
+static inline void virtio_pci_force_virtio_1(VirtIOPCIProxy *proxy)
 {
-    if (proxy->disable_legacy == ON_OFF_AUTO_OFF) {
-        error_setg(errp, "Unable to set disable-legacy=off on a virtio-1.0 "
-                   "only device");
-        return false;
-    }
-    if (proxy->disable_modern == true) {
-        error_setg(errp, "Unable to set disable-modern=on on a virtio-1.0 "
-                   "only device");
-        return false;
-    }
-    proxy->mode = VIRTIO_PCI_MODE_MODERN;
-    return true;
+    proxy->disable_modern = false;
+    proxy->disable_legacy = ON_OFF_AUTO_ON;
 }
 
 static inline void virtio_pci_disable_modern(VirtIOPCIProxy *proxy)
 {
-    proxy->mode = VIRTIO_PCI_MODE_LEGACY;
+    proxy->disable_modern = true;
 }
 
 /*
Signature · 1 line
2.21.0
Reviewed-by on v2 2/2
#5 lore raw

Re: [Qemu-devel] [PATCH v2 2/2] Revert "hw: report invalid disable-legacy|modern usage for virtio-1-only devs"

Quoted 8 lines · Dr. David Alan Gilbert (git)

From: "Dr. David Alan Gilbert" <[email protected]>

This reverts commit f2784eed306449c3d04a71a05ed6463b8289aedf since that accidentally removes the PCIe capabilities from virtio devices because virtio_pci_dc_realize is called before the new 'mode' flag is set.

Quoted 9 lines
---
 hw/core/machine.c             | 23 +++--------------------
 hw/display/virtio-gpu-pci.c   |  4 +---
 hw/display/virtio-vga.c       |  4 +---
 hw/virtio/virtio-crypto-pci.c |  4 +---
 hw/virtio/virtio-input-pci.c  |  4 +---
 hw/virtio/virtio-pci.c        | 26 ++++++++++----------------
 hw/virtio/virtio-pci.h        | 31 ++++++-------------------------
 7 files changed, 23 insertions(+), 73 deletions(-)

Regards,
Daniel

Signature · 3 lines
|: https://berrange.com      -o-    https://www.flickr.com/photos/dberrange :|
|: https://libvirt.org         -o-            https://fstop138.berrange.com :|
|: https://entangle-photo.org    -o-    https://www.instagram.com/dberrange :|
Reviewed-by on v2 2/2
#6 lore raw

Re: [Qemu-devel] [PATCH v2 2/2] Revert "hw: report invalid disable-legacy|modern usage for virtio-1-only devs"

In reply to Dr. David Alan Gilbert (git)
Quoted 8 lines · On Mon, 29 Jul 2019 17:29:03 +0100 "Dr. David Alan Gilbert (git)"…

From: "Dr. David Alan Gilbert" <[email protected]>

This reverts commit f2784eed306449c3d04a71a05ed6463b8289aedf since that accidentally removes the PCIe capabilities from virtio devices because virtio_pci_dc_realize is called before the new 'mode' flag is set.

Quoted 9 lines
---
 hw/core/machine.c             | 23 +++--------------------
 hw/display/virtio-gpu-pci.c   |  4 +---
 hw/display/virtio-vga.c       |  4 +---
 hw/virtio/virtio-crypto-pci.c |  4 +---
 hw/virtio/virtio-input-pci.c  |  4 +---
 hw/virtio/virtio-pci.c        | 26 ++++++++++----------------
 hw/virtio/virtio-pci.h        | 31 ++++++-------------------------
 7 files changed, 23 insertions(+), 73 deletions(-)
#7 lore raw
In reply to Dr. David Alan Gilbert (git)
Quoted 5 lines · On Mon, 29 Jul 2019 17:29:01 +0100 "Dr. David Alan Gilbert (git)"…

From: "Dr. David Alan Gilbert" <[email protected]>

Revert a couple of patches that break PCIe capabilities in virtio devices. The 'optional' revert is just reverted to make the main reversion trivial.

Don't want to spoil the party here; but wasn't the optional stuff removed because it was deemed to be a bad idea?

Quoted 23 lines

Symptom:
Loss of PCIe capabilities in virtio devices hung off PCIe bridges

Dr. David Alan Gilbert (2):
Revert "Revert "globals: Allow global properties to be optional"" Revert "hw: report invalid disable-legacy|modern usage for
virtio-1-only devs"

 hw/core/machine.c             | 23 +++--------------------
 hw/display/virtio-gpu-pci.c   |  4 +---
 hw/display/virtio-vga.c       |  4 +---
 hw/virtio/virtio-crypto-pci.c |  4 +---
 hw/virtio/virtio-input-pci.c  |  4 +---
 hw/virtio/virtio-pci.c        | 26 ++++++++++----------------
 hw/virtio/virtio-pci.h        | 31 ++++++-------------------------
 include/hw/qdev-core.h        |  3 +++
 qom/object.c                  |  3 +++
 9 files changed, 29 insertions(+), 73 deletions(-)
#8 lore raw
Quoted 11 lines · * Cornelia Huck ([email protected]) wrote:
Quoted 5 lines · On Mon, 29 Jul 2019 17:29:01 +0100 "Dr. David Alan Gilbert (git)"…

From: "Dr. David Alan Gilbert" <[email protected]>

Revert a couple of patches that break PCIe capabilities in virtio devices. The 'optional' revert is just reverted to make the main reversion trivial.

Don't want to spoil the party here; but wasn't the optional stuff removed because it was deemed to be a bad idea?

I'm perfectly happy to go either way with this; it maybe a bad idea but it's harmless I think.

Dave

Quoted 12 lines
Quoted 12 lines

Symptom:
Loss of PCIe capabilities in virtio devices hung off PCIe bridges

Dr. David Alan Gilbert (2):
Revert "Revert "globals: Allow global properties to be optional"" Revert "hw: report invalid disable-legacy|modern usage for
virtio-1-only devs"

Quoted 12 lines
Quoted 11 lines
 hw/core/machine.c             | 23 +++--------------------
 hw/display/virtio-gpu-pci.c   |  4 +---
 hw/display/virtio-vga.c       |  4 +---
 hw/virtio/virtio-crypto-pci.c |  4 +---
 hw/virtio/virtio-input-pci.c  |  4 +---
 hw/virtio/virtio-pci.c        | 26 ++++++++++----------------
 hw/virtio/virtio-pci.h        | 31 ++++++-------------------------
 include/hw/qdev-core.h        |  3 +++
 qom/object.c                  |  3 +++
 9 files changed, 29 insertions(+), 73 deletions(-)

--
Dr. David Alan Gilbert / [email protected] / Manchester, UK

#9 lore raw
Quoted 11 lines · Dr. David Alan Gilbert
Quoted 9 lines · * Cornelia Huck ([email protected]) wrote:
Quoted 5 lines · On Mon, 29 Jul 2019 17:29:01 +0100 "Dr. David Alan Gilbert (git)"…

From: "Dr. David Alan Gilbert" <[email protected]>

Revert a couple of patches that break PCIe capabilities in virtio devices. The 'optional' revert is just reverted to make the main reversion trivial.

Quoted 5 lines
Quoted 2 lines

Don't want to spoil the party here; but wasn't the optional stuff removed because it was deemed to be a bad idea?

I'm perfectly happy to go either way with this; it maybe a bad idea but it's harmless I think.

It seems like the original commits were:
* patch that does something
* patch that removes no-longer used functionality (optional globals)

so it makes sense to me that if we want to revert the 'patch that does something' we should first revert the patch that cleaned up unused-functionality (because now we need it again). Is that right?

If optional-globals are a bad idea then we should take another run at this for 4.2, but as a "revert stuff for 4.1" strategy it seems fine to me.

thanks
-- PMM

#10 lore raw
Quoted 23 lines · Peter Maydell
Quoted 11 lines · Dr. David Alan Gilbert
Quoted 9 lines · * Cornelia Huck ([email protected]) wrote:
Quoted 5 lines · On Mon, 29 Jul 2019 17:29:01 +0100 "Dr. David Alan Gilbert (git)"…

From: "Dr. David Alan Gilbert" <[email protected]>

Revert a couple of patches that break PCIe capabilities in virtio devices. The 'optional' revert is just reverted to make the main reversion trivial.

Quoted 5 lines
Quoted 2 lines

Don't want to spoil the party here; but wasn't the optional stuff removed because it was deemed to be a bad idea?

I'm perfectly happy to go either way with this; it maybe a bad idea but it's harmless I think.

It seems like the original commits were:
* patch that does something
* patch that removes no-longer used functionality (optional globals)

Quoted 8 lines

so it makes sense to me that if we want to revert the 'patch that does something' we should first revert the patch that cleaned up unused-functionality (because now we need it again). Is that right?

If optional-globals are a bad idea then we should take another run at this for 4.2, but as a "revert stuff for 4.1" strategy it seems fine to me.

Functionally both approaches are supposed to be identical, but given that we already found one last minute problem with the 2nd patch, the full revert of both feels ever so slightly safer.

Regards,
Daniel

Signature · 3 lines
|: https://berrange.com      -o-    https://www.flickr.com/photos/dberrange :|
|: https://libvirt.org         -o-            https://fstop138.berrange.com :|
|: https://entangle-photo.org    -o-    https://www.instagram.com/dberrange :|