Skip to contentSkip to conversation

pc: Don't make die-id mandatory unless necessary

PATCH In review Eduardo Habkost 5 participants 8 messages Aug 16, 2019 – Feb 3, 2021

lore
Share & tools
lore
#0 lore raw

PATCH v2 1/1 pc: Don't make die-id mandatory unless necessary

We have this issue reported when using libvirt to hotplug CPUs:
https://bugzilla.redhat.com/show_bug.cgi?id=1741451

Basically, libvirt is not copying die-id from query-hotpluggable-cpus, but die-id is now mandatory.

We could blame libvirt and say it is not following the documented interface, because we have this buried in the QAPI schema documentation:

Quoted 5 lines

Note: currently there are 5 properties that could be present but management should be prepared to pass through other properties with device_add command to allow for future interface extension. This also requires the filed names to be kept in sync with the properties passed to -device/device_add.

But I don't think this would be reasonable from us. We can just make QEMU more flexible and let die-id to be omitted when there's no ambiguity. This will allow us to keep compatibility with existing libvirt versions.

Test case included to ensure we don't break this again.

Fixes commit 176d2cda0dee ("i386/cpu: Consolidate die-id validity in smp context")

---
Changes v1 -> v2:
* v1 was "pc: Don't make CPU properties mandatory unless necessary"
* Make only die-id optional (Igor Mammedov)

---
 hw/i386/pc.c                             |  8 ++++++
 tests/acceptance/pc_cpu_hotplug_props.py | 35 ++++++++++++++++++++++++
 2 files changed, 43 insertions(+)
 create mode 100644 tests/acceptance/pc_cpu_hotplug_props.py
hw/i386/pc.c +8 -0
index 3ab4bcb3ca..9c3f6ae828 100644
@@ -2406,6 +2406,14 @@ static void pc_cpu_pre_plug(HotplugHandler *hotplug_dev,
         int max_socket = (ms->smp.max_cpus - 1) /
                                 smp_threads / smp_cores / pcms->smp_dies;
 
+        /*
+         * die-id was optional in QEMU 4.0 and older, so keep it optional
+         * if there's only one die per socket.
+         */
+        if (cpu->die_id < 0 && pcms->smp_dies == 1) {
+            cpu->die_id = 0;
+        }
+
         if (cpu->socket_id < 0) {
             error_setg(errp, "CPU socket-id is not set");
             return;
tests/acceptance/pc_cpu_hotplug_props.py new file +35 -0
new file mode 100644
index 0000000000..08b7e632c6
@@ -0,0 +1,35 @@
+#
+# Ensure CPU die-id can be omitted on -device
+#
+#  Copyright (c) 2019 Red Hat Inc
+#
+# Author:
+#  Eduardo Habkost <[email protected]>
+#
+# This library is free software; you can redistribute it and/or
+# modify it under the terms of the GNU Lesser General Public
+# License as published by the Free Software Foundation; either
+# version 2 of the License, or (at your option) any later version.
+#
+# This library is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+# Lesser General Public License for more details.
+#
+# You should have received a copy of the GNU Lesser General Public
+# License along with this library; if not, see <http://www.gnu.org/licenses/>.
+#
+
+from avocado_qemu import Test
+
+class OmittedCPUProps(Test):
+    """
+    :avocado: tags=arch:x86_64
+    """
+    def test_no_die_id(self):
+        self.vm.add_args('-nodefaults', '-S')
+        self.vm.add_args('-smp', '1,sockets=2,cores=2,threads=2,maxcpus=8')
+        self.vm.add_args('-cpu', 'qemu64')
+        self.vm.add_args('-device', 'qemu64-x86_64-cpu,socket-id=1,core-id=0,thread-id=0')
+        self.vm.launch()
+        self.assertEquals(len(self.vm.command('query-cpus')), 2)
Signature · 1 line
2.21.0
Reviewed-by on v2 1/1
#1 lore raw
Quoted 21 lines · On Fri, 16 Aug 2019 14:07:50 -0300 Eduardo Habkost <[email protected]> wrote:

We have this issue reported when using libvirt to hotplug CPUs:
https://bugzilla.redhat.com/show_bug.cgi?id=1741451

Basically, libvirt is not copying die-id from query-hotpluggable-cpus, but die-id is now mandatory.

We could blame libvirt and say it is not following the documented interface, because we have this buried in the QAPI schema documentation:

Quoted 5 lines

Note: currently there are 5 properties that could be present but management should be prepared to pass through other properties with device_add command to allow for future interface extension. This also requires the filed names to be kept in sync with the properties passed to -device/device_add.

But I don't think this would be reasonable from us. We can just make QEMU more flexible and let die-id to be omitted when there's no ambiguity. This will allow us to keep compatibility with existing libvirt versions.

Quoted 4 lines

Test case included to ensure we don't break this again.

Fixes commit 176d2cda0dee ("i386/cpu: Consolidate die-id validity in smp context")
Quoted 70 lines

---
Changes v1 -> v2:
* v1 was "pc: Don't make CPU properties mandatory unless necessary"
* Make only die-id optional (Igor Mammedov)

---
 hw/i386/pc.c                             |  8 ++++++
 tests/acceptance/pc_cpu_hotplug_props.py | 35 ++++++++++++++++++++++++
 2 files changed, 43 insertions(+)
 create mode 100644 tests/acceptance/pc_cpu_hotplug_props.py
hw/i386/pc.c +8 -0
index 3ab4bcb3ca..9c3f6ae828 100644
@@ -2406,6 +2406,14 @@ static void pc_cpu_pre_plug(HotplugHandler *hotplug_dev,
         int max_socket = (ms->smp.max_cpus - 1) /
                                 smp_threads / smp_cores / pcms->smp_dies;
 
+        /*
+         * die-id was optional in QEMU 4.0 and older, so keep it optional
+         * if there's only one die per socket.
+         */
+        if (cpu->die_id < 0 && pcms->smp_dies == 1) {
+            cpu->die_id = 0;
+        }
+
         if (cpu->socket_id < 0) {
             error_setg(errp, "CPU socket-id is not set");
             return;
tests/acceptance/pc_cpu_hotplug_props.py new file +35 -0
new file mode 100644
index 0000000000..08b7e632c6
@@ -0,0 +1,35 @@
+#
+# Ensure CPU die-id can be omitted on -device
+#
+#  Copyright (c) 2019 Red Hat Inc
+#
+# Author:
+#  Eduardo Habkost <[email protected]>
+#
+# This library is free software; you can redistribute it and/or
+# modify it under the terms of the GNU Lesser General Public
+# License as published by the Free Software Foundation; either
+# version 2 of the License, or (at your option) any later version.
+#
+# This library is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+# Lesser General Public License for more details.
+#
+# You should have received a copy of the GNU Lesser General Public
+# License along with this library; if not, see <http://www.gnu.org/licenses/>.
+#
+
+from avocado_qemu import Test
+
+class OmittedCPUProps(Test):
+    """
+    :avocado: tags=arch:x86_64
+    """
+    def test_no_die_id(self):
+        self.vm.add_args('-nodefaults', '-S')
+        self.vm.add_args('-smp', '1,sockets=2,cores=2,threads=2,maxcpus=8')
+        self.vm.add_args('-cpu', 'qemu64')
+        self.vm.add_args('-device', 'qemu64-x86_64-cpu,socket-id=1,core-id=0,thread-id=0')
+        self.vm.launch()
+        self.assertEquals(len(self.vm.command('query-cpus')), 2)
#2 lore raw
In reply to Eduardo Habkost
Quoted 40 lines · Eduardo Habkost

We have this issue reported when using libvirt to hotplug CPUs:
https://bugzilla.redhat.com/show_bug.cgi?id=1741451

Basically, libvirt is not copying die-id from query-hotpluggable-cpus, but die-id is now mandatory.

We could blame libvirt and say it is not following the documented interface, because we have this buried in the QAPI schema documentation:

Quoted 5 lines

Note: currently there are 5 properties that could be present but management should be prepared to pass through other properties with device_add command to allow for future interface extension. This also requires the filed names to be kept in sync with the properties passed to -device/device_add.

But I don't think this would be reasonable from us. We can just make QEMU more flexible and let die-id to be omitted when there's no ambiguity. This will allow us to keep compatibility with existing libvirt versions.

Test case included to ensure we don't break this again.

Fixes commit 176d2cda0dee ("i386/cpu: Consolidate die-id validity in smp context")

---
Changes v1 -> v2:
* v1 was "pc: Don't make CPU properties mandatory unless necessary"
* Make only die-id optional (Igor Mammedov)

---
 hw/i386/pc.c                             |  8 ++++++
 tests/acceptance/pc_cpu_hotplug_props.py | 35 ++++++++++++++++++++++++
 2 files changed, 43 insertions(+)
 create mode 100644 tests/acceptance/pc_cpu_hotplug_props.py
hw/i386/pc.c +0 -0
index 3ab4bcb3ca..9c3f6ae828 100644

hw/i386/pc.c static void pc_cpu_pre_plug(HotplugHandler *hotplug_dev, · line 2412

Quoted 7 lines
hw/i386/pc.c +4 -0
@@ -2406,6 +2406,14 @@ static void pc_cpu_pre_plug(HotplugHandler *hotplug_dev,
         int max_socket = (ms->smp.max_cpus - 1) /
                                 smp_threads / smp_cores / pcms->smp_dies;
 
+        /*
+         * die-id was optional in QEMU 4.0 and older, so keep it optional
+         * if there's only one die per socket.
+         */

Is this comment worth its keep?

Quoted 38 lines
hw/i386/pc.c +4 -0
+        if (cpu->die_id < 0 && pcms->smp_dies == 1) {
+            cpu->die_id = 0;
+        }
+
         if (cpu->socket_id < 0) {
             error_setg(errp, "CPU socket-id is not set");
             return;
tests/acceptance/pc_cpu_hotplug_props.py new file +25 -0
new file mode 100644
index 0000000000..08b7e632c6
@@ -0,0 +1,35 @@
+#
+# Ensure CPU die-id can be omitted on -device
+#
+#  Copyright (c) 2019 Red Hat Inc
+#
+# Author:
+#  Eduardo Habkost <[email protected]>
+#
+# This library is free software; you can redistribute it and/or
+# modify it under the terms of the GNU Lesser General Public
+# License as published by the Free Software Foundation; either
+# version 2 of the License, or (at your option) any later version.
+#
+# This library is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+# Lesser General Public License for more details.
+#
+# You should have received a copy of the GNU Lesser General Public
+# License along with this library; if not, see <http://www.gnu.org/licenses/>.
+#
+
+from avocado_qemu import Test
+
+class OmittedCPUProps(Test):

tests/acceptance/pc_cpu_hotplug_props.py line 35

Quoted 10 lines
tests/acceptance/pc_cpu_hotplug_props.py new file +10 -0
+    """
+    :avocado: tags=arch:x86_64
+    """
+    def test_no_die_id(self):
+        self.vm.add_args('-nodefaults', '-S')
+        self.vm.add_args('-smp', '1,sockets=2,cores=2,threads=2,maxcpus=8')
+        self.vm.add_args('-cpu', 'qemu64')
+        self.vm.add_args('-device', 'qemu64-x86_64-cpu,socket-id=1,core-id=0,thread-id=0')
+        self.vm.launch()
+        self.assertEquals(len(self.vm.command('query-cpus')), 2)

This tests intentionally violates the "a CPU hot-plug with -device must specify all the properties returned by query-hotpluggable-cpus" rule laid down in qapi/machine.json, and the change to pc.c permits it.

Missing: a suitable update to the rule.

In addition, I'd like us to either work on making the rule stick in the future (see my reply to Igor for an idea), or ditch the rule. But that's outside the scope of this regression fix.

#3 lore raw
Quoted 45 lines · Markus Armbruster
Quoted 43 lines · Eduardo Habkost

We have this issue reported when using libvirt to hotplug CPUs:
https://bugzilla.redhat.com/show_bug.cgi?id=1741451

Basically, libvirt is not copying die-id from query-hotpluggable-cpus, but die-id is now mandatory.

We could blame libvirt and say it is not following the documented interface, because we have this buried in the QAPI schema documentation:

Quoted 5 lines

Note: currently there are 5 properties that could be present but management should be prepared to pass through other properties with device_add command to allow for future interface extension. This also requires the filed names to be kept in sync with the properties passed to -device/device_add.

But I don't think this would be reasonable from us. We can just make QEMU more flexible and let die-id to be omitted when there's no ambiguity. This will allow us to keep compatibility with existing libvirt versions.

Test case included to ensure we don't break this again.

Fixes commit 176d2cda0dee ("i386/cpu: Consolidate die-id validity in smp context")

---
Changes v1 -> v2:
* v1 was "pc: Don't make CPU properties mandatory unless necessary"
* Make only die-id optional (Igor Mammedov)

---
 hw/i386/pc.c                             |  8 ++++++
 tests/acceptance/pc_cpu_hotplug_props.py | 35 ++++++++++++++++++++++++
 2 files changed, 43 insertions(+)
 create mode 100644 tests/acceptance/pc_cpu_hotplug_props.py
hw/i386/pc.c +0 -0
index 3ab4bcb3ca..9c3f6ae828 100644
@@ -2406,6 +2406,14 @@ static void pc_cpu_pre_plug(HotplugHandler *hotplug_dev,
         int max_socket = (ms->smp.max_cpus - 1) /
                                 smp_threads / smp_cores / pcms->smp_dies;
 
Quoted 6 lines

hw/i386/pc.c static void pc_cpu_pre_plug(HotplugHandler *hotplug_dev, · line 2412

Quoted 4 lines
hw/i386/pc.c +4 -0
+        /*
+         * die-id was optional in QEMU 4.0 and older, so keep it optional
+         * if there's only one die per socket.
+         */

Is this comment worth its keep?

It's not obvious from the code why die-id is optional and all the other properties are not. The purpose of the comment is to clarify that.

Quoted 54 lines
Quoted 39 lines
hw/i386/pc.c +4 -0
+        if (cpu->die_id < 0 && pcms->smp_dies == 1) {
+            cpu->die_id = 0;
+        }
+
         if (cpu->socket_id < 0) {
             error_setg(errp, "CPU socket-id is not set");
             return;
tests/acceptance/pc_cpu_hotplug_props.py new file +25 -0
new file mode 100644
index 0000000000..08b7e632c6
@@ -0,0 +1,35 @@
+#
+# Ensure CPU die-id can be omitted on -device
+#
+#  Copyright (c) 2019 Red Hat Inc
+#
+# Author:
+#  Eduardo Habkost <[email protected]>
+#
+# This library is free software; you can redistribute it and/or
+# modify it under the terms of the GNU Lesser General Public
+# License as published by the Free Software Foundation; either
+# version 2 of the License, or (at your option) any later version.
+#
+# This library is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+# Lesser General Public License for more details.
+#
+# You should have received a copy of the GNU Lesser General Public
+# License along with this library; if not, see <http://www.gnu.org/licenses/>.
+#
+
+from avocado_qemu import Test
+
+class OmittedCPUProps(Test):

tests/acceptance/pc_cpu_hotplug_props.py line 35

Quoted 10 lines
tests/acceptance/pc_cpu_hotplug_props.py new file +10 -0
+    """
+    :avocado: tags=arch:x86_64
+    """
+    def test_no_die_id(self):
+        self.vm.add_args('-nodefaults', '-S')
+        self.vm.add_args('-smp', '1,sockets=2,cores=2,threads=2,maxcpus=8')
+        self.vm.add_args('-cpu', 'qemu64')
+        self.vm.add_args('-device', 'qemu64-x86_64-cpu,socket-id=1,core-id=0,thread-id=0')
+        self.vm.launch()
+        self.assertEquals(len(self.vm.command('query-cpus')), 2)

This tests intentionally violates the "a CPU hot-plug with -device must specify all the properties returned by query-hotpluggable-cpus" rule laid down in qapi/machine.json, and the change to pc.c permits it.

Quoted 1 line

Missing: a suitable update to the rule.

Intentional, because I don't want decisions about documentation to block the regression fix from being merged.

Quoted 4 lines

In addition, I'd like us to either work on making the rule stick in the future (see my reply to Igor for an idea), or ditch the rule. But that's outside the scope of this regression fix.

I'd prefer to ditch the rule, or at least change it to be a suggestion instead of a requirement.

Signature · 1 line
Eduardo
#4 lore raw
Quoted 98 lines · On Wed, 28 Aug 2019 11:35:24 -0300 Eduardo Habkost <[email protected]> wrote:
Quoted 45 lines · Markus Armbruster
Quoted 43 lines · Eduardo Habkost

We have this issue reported when using libvirt to hotplug CPUs:
https://bugzilla.redhat.com/show_bug.cgi?id=1741451

Basically, libvirt is not copying die-id from query-hotpluggable-cpus, but die-id is now mandatory.

We could blame libvirt and say it is not following the documented interface, because we have this buried in the QAPI schema documentation:

Quoted 5 lines

Note: currently there are 5 properties that could be present but management should be prepared to pass through other properties with device_add command to allow for future interface extension. This also requires the filed names to be kept in sync with the properties passed to -device/device_add.

But I don't think this would be reasonable from us. We can just make QEMU more flexible and let die-id to be omitted when there's no ambiguity. This will allow us to keep compatibility with existing libvirt versions.

Test case included to ensure we don't break this again.

Fixes commit 176d2cda0dee ("i386/cpu: Consolidate die-id validity in smp context")

---
Changes v1 -> v2:
* v1 was "pc: Don't make CPU properties mandatory unless necessary"
* Make only die-id optional (Igor Mammedov)

---
 hw/i386/pc.c                             |  8 ++++++
 tests/acceptance/pc_cpu_hotplug_props.py | 35 ++++++++++++++++++++++++
 2 files changed, 43 insertions(+)
 create mode 100644 tests/acceptance/pc_cpu_hotplug_props.py
hw/i386/pc.c +0 -0
index 3ab4bcb3ca..9c3f6ae828 100644
@@ -2406,6 +2406,14 @@ static void pc_cpu_pre_plug(HotplugHandler *hotplug_dev,
         int max_socket = (ms->smp.max_cpus - 1) /
                                 smp_threads / smp_cores / pcms->smp_dies;
 
Quoted 6 lines

hw/i386/pc.c static void pc_cpu_pre_plug(HotplugHandler *hotplug_dev, · line 2412

Quoted 4 lines
hw/i386/pc.c +4 -0
+        /*
+         * die-id was optional in QEMU 4.0 and older, so keep it optional
+         * if there's only one die per socket.
+         */  

Is this comment worth its keep?

It's not obvious from the code why die-id is optional and all the other properties are not. The purpose of the comment is to clarify that.

Quoted 41 lines
Quoted 41 lines
hw/i386/pc.c +4 -0
+        if (cpu->die_id < 0 && pcms->smp_dies == 1) {
+            cpu->die_id = 0;
+        }
+
         if (cpu->socket_id < 0) {
             error_setg(errp, "CPU socket-id is not set");
             return;
tests/acceptance/pc_cpu_hotplug_props.py new file +27 -0
new file mode 100644
index 0000000000..08b7e632c6
@@ -0,0 +1,35 @@
+#
+# Ensure CPU die-id can be omitted on -device
+#
+#  Copyright (c) 2019 Red Hat Inc
+#
+# Author:
+#  Eduardo Habkost <[email protected]>
+#
+# This library is free software; you can redistribute it and/or
+# modify it under the terms of the GNU Lesser General Public
+# License as published by the Free Software Foundation; either
+# version 2 of the License, or (at your option) any later version.
+#
+# This library is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+# Lesser General Public License for more details.
+#
+# You should have received a copy of the GNU Lesser General Public
+# License along with this library; if not, see <http://www.gnu.org/licenses/>.
+#
+
+from avocado_qemu import Test
+
+class OmittedCPUProps(Test):
+    """
+    :avocado: tags=arch:x86_64
Quoted 8 lines
Quoted 8 lines

tests/acceptance/pc_cpu_hotplug_props.py line 35

Quoted 8 lines
tests/acceptance/pc_cpu_hotplug_props.py new file +8 -0
+    """
+    def test_no_die_id(self):
+        self.vm.add_args('-nodefaults', '-S')
+        self.vm.add_args('-smp', '1,sockets=2,cores=2,threads=2,maxcpus=8')
+        self.vm.add_args('-cpu', 'qemu64')
+        self.vm.add_args('-device', 'qemu64-x86_64-cpu,socket-id=1,core-id=0,thread-id=0')
+        self.vm.launch()
+        self.assertEquals(len(self.vm.command('query-cpus')), 2)  

test_plug_with_device_add_x86() is supposed to test device_add (currently broken), so do we really need more heavy avocado tests,
how about adding this test to tests/cpu-plug-test.c
('make check' is used much more often by developers)

Quoted 10 lines
Quoted 5 lines

This tests intentionally violates the "a CPU hot-plug with -device must specify all the properties returned by query-hotpluggable-cpus" rule laid down in qapi/machine.json, and the change to pc.c permits it.

Missing: a suitable update to the rule.

Intentional, because I don't want decisions about documentation to block the regression fix from being merged.

Quoted 1 line
Quoted 6 lines
Quoted 3 lines

In addition, I'd like us to either work on making the rule stick in the future (see my reply to Igor for an idea), or ditch the rule. But that's outside the scope of this regression fix.

I'd prefer to ditch the rule, or at least change it to be a suggestion instead of a requirement.

Perhaps someone reads docs and uses API as designed (libvirt is not the only user)

I'd prefer to allow implicit die-id in 4.1 and 'stable' as that ship has already sailed and make it mandatory since 4.2 as it is supposed to be (+opening bug on libvirt - hoping that API would be fixed properly this time).

Another related to die-id series bug:
We should hide die-id from query-hotpluggable-cpus output for 4.0 and older machine types as well, so it won't break migration for users that implement interface as documented as it won't be possible to start
old-qemu-4.0 -device cpufoo,die-id=0,...
since that "-device cpufoo,die-id=0,..." were used on new-qemu source.

PS:
Adding affected targets maintainers to the loop to see if we can drop restriction.

Even though it works fine for die-id and I don't see immediate problems with relaxing rule, I reluctant to do it, since instead of simple
"add all properties you were told to"
implicit rules would evolve into mess similar to smp_parse() over time.

Also if we would need to change implicit values logic down the road it would be a pain like with any default parameters in QEMU, which is a good reason against relaxing rule.

#5 lore raw
Quoted 151 lines · Igor Mammedov
Quoted 98 lines · On Wed, 28 Aug 2019 11:35:24 -0300 Eduardo Habkost <[email protected]> wrote:
Quoted 45 lines · Markus Armbruster
Quoted 43 lines · Eduardo Habkost

We have this issue reported when using libvirt to hotplug CPUs:
https://bugzilla.redhat.com/show_bug.cgi?id=1741451

Basically, libvirt is not copying die-id from query-hotpluggable-cpus, but die-id is now mandatory.

We could blame libvirt and say it is not following the documented interface, because we have this buried in the QAPI schema documentation:

Quoted 5 lines

Note: currently there are 5 properties that could be present but management should be prepared to pass through other properties with device_add command to allow for future interface extension. This also requires the filed names to be kept in sync with the properties passed to -device/device_add.

But I don't think this would be reasonable from us. We can just make QEMU more flexible and let die-id to be omitted when there's no ambiguity. This will allow us to keep compatibility with existing libvirt versions.

Test case included to ensure we don't break this again.

Fixes commit 176d2cda0dee ("i386/cpu: Consolidate die-id validity in smp context")

---
Changes v1 -> v2:
* v1 was "pc: Don't make CPU properties mandatory unless necessary"
* Make only die-id optional (Igor Mammedov)

---
 hw/i386/pc.c                             |  8 ++++++
 tests/acceptance/pc_cpu_hotplug_props.py | 35 ++++++++++++++++++++++++
 2 files changed, 43 insertions(+)
 create mode 100644 tests/acceptance/pc_cpu_hotplug_props.py
hw/i386/pc.c +0 -0
index 3ab4bcb3ca..9c3f6ae828 100644
@@ -2406,6 +2406,14 @@ static void pc_cpu_pre_plug(HotplugHandler *hotplug_dev,
         int max_socket = (ms->smp.max_cpus - 1) /
                                 smp_threads / smp_cores / pcms->smp_dies;
 
Quoted 6 lines

hw/i386/pc.c static void pc_cpu_pre_plug(HotplugHandler *hotplug_dev, · line 2412

Quoted 4 lines
hw/i386/pc.c +4 -0
+        /*
+         * die-id was optional in QEMU 4.0 and older, so keep it optional
+         * if there's only one die per socket.
+         */  

Is this comment worth its keep?

It's not obvious from the code why die-id is optional and all the other properties are not. The purpose of the comment is to clarify that.

Quoted 41 lines
Quoted 41 lines
hw/i386/pc.c +4 -0
+        if (cpu->die_id < 0 && pcms->smp_dies == 1) {
+            cpu->die_id = 0;
+        }
+
         if (cpu->socket_id < 0) {
             error_setg(errp, "CPU socket-id is not set");
             return;
tests/acceptance/pc_cpu_hotplug_props.py new file +27 -0
new file mode 100644
index 0000000000..08b7e632c6
@@ -0,0 +1,35 @@
+#
+# Ensure CPU die-id can be omitted on -device
+#
+#  Copyright (c) 2019 Red Hat Inc
+#
+# Author:
+#  Eduardo Habkost <[email protected]>
+#
+# This library is free software; you can redistribute it and/or
+# modify it under the terms of the GNU Lesser General Public
+# License as published by the Free Software Foundation; either
+# version 2 of the License, or (at your option) any later version.
+#
+# This library is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+# Lesser General Public License for more details.
+#
+# You should have received a copy of the GNU Lesser General Public
+# License along with this library; if not, see <http://www.gnu.org/licenses/>.
+#
+
+from avocado_qemu import Test
+
+class OmittedCPUProps(Test):
+    """
+    :avocado: tags=arch:x86_64
Quoted 8 lines
Quoted 8 lines

tests/acceptance/pc_cpu_hotplug_props.py line 35

Quoted 8 lines
tests/acceptance/pc_cpu_hotplug_props.py new file +8 -0
+    """
+    def test_no_die_id(self):
+        self.vm.add_args('-nodefaults', '-S')
+        self.vm.add_args('-smp', '1,sockets=2,cores=2,threads=2,maxcpus=8')
+        self.vm.add_args('-cpu', 'qemu64')
+        self.vm.add_args('-device', 'qemu64-x86_64-cpu,socket-id=1,core-id=0,thread-id=0')
+        self.vm.launch()
+        self.assertEquals(len(self.vm.command('query-cpus')), 2)  

test_plug_with_device_add_x86() is supposed to test device_add (currently broken), so do we really need more heavy avocado tests,
how about adding this test to tests/cpu-plug-test.c
('make check' is used much more often by developers)

Quoted 10 lines
Quoted 5 lines

This tests intentionally violates the "a CPU hot-plug with -device must specify all the properties returned by query-hotpluggable-cpus" rule laid down in qapi/machine.json, and the change to pc.c permits it.

Missing: a suitable update to the rule.

Intentional, because I don't want decisions about documentation to block the regression fix from being merged.

Quoted 1 line
Quoted 6 lines
Quoted 3 lines

In addition, I'd like us to either work on making the rule stick in the future (see my reply to Igor for an idea), or ditch the rule. But that's outside the scope of this regression fix.

I'd prefer to ditch the rule, or at least change it to be a suggestion instead of a requirement.

Perhaps someone reads docs and uses API as designed (libvirt is not the only user)

I'd prefer to allow implicit die-id in 4.1 and 'stable' as that ship has already sailed and make it mandatory since 4.2 as it is supposed to be (+opening bug on libvirt - hoping that API would be fixed properly this time).

Another related to die-id series bug:
We should hide die-id from query-hotpluggable-cpus output for 4.0 and older machine types as well, so it won't break migration for users that implement interface as documented as it won't be possible to start
old-qemu-4.0 -device cpufoo,die-id=0,...
since that "-device cpufoo,die-id=0,..." were used on new-qemu source.

PS:
Adding affected targets maintainers to the loop to see if we can drop restriction.

Quoted 8 lines

Even though it works fine for die-id and I don't see immediate problems with relaxing rule, I reluctant to do it, since instead of simple
"add all properties you were told to"
implicit rules would evolve into mess similar to smp_parse() over time.

Also if we would need to change implicit values logic down the road it would be a pain like with any default parameters in QEMU, which is a good reason against relaxing rule.

Hi,

after a discussion with Igor I'm preparing patches which take the 'props' verbatim and use it for hotplug. It's a fairly simple change in libvirt. Originally I've opted to reconstruct the props from internal data since it was simpler as the values were already parsed into internal structures anyways.

With some persuation we could even get it into libvirt during the freeze so that with new libvirt the released stuff will not break any more.

#6 lore raw
Quoted 2 lines · Peter Krempa
Quoted 1 line · Igor Mammedov

On Wed, 28 Aug 2019 11:35:24 -0300

[...]

Quoted 6 lines
Quoted 3 lines

Also if we would need to change implicit values logic down the road it would be a pain like with any default parameters in QEMU, which is a good reason against relaxing rule.

Hi,

Quoted 8 lines

after a discussion with Igor I'm preparing patches which take the 'props' verbatim and use it for hotplug. It's a fairly simple change in libvirt. Originally I've opted to reconstruct the props from internal data since it was simpler as the values were already parsed into internal structures anyways.

With some persuation we could even get it into libvirt during the freeze so that with new libvirt the released stuff will not break any more.

https://www.redhat.com/archives/libvir-list/2019-August/msg01322.html

#7 lore raw
In reply to Igor Mammedov

Cc'ing avocado-devel for test ideas.

Quoted 4 lines · Igor Mammedov
Quoted 1 line · On Wed, 28 Aug 2019 11:35:24 -0300 Eduardo Habkost <[email protected]> wrote:

On Wed, Aug 28, 2019 at 08:52:28AM +0200, Markus Armbruster wrote:

...

Quoted 36 lines
Quoted 6 lines
Quoted 3 lines

In addition, I'd like us to either work on making the rule stick in the future (see my reply to Igor for an idea), or ditch the rule. But that's outside the scope of this regression fix.

I'd prefer to ditch the rule, or at least change it to be a suggestion instead of a requirement.

Perhaps someone reads docs and uses API as designed (libvirt is not the only user)

I'd prefer to allow implicit die-id in 4.1 and 'stable' as that ship has already sailed and make it mandatory since 4.2 as it is supposed to be (+opening bug on libvirt - hoping that API would be fixed properly this time).

Another related to die-id series bug:
We should hide die-id from query-hotpluggable-cpus output for 4.0 and older machine types as well, so it won't break migration for users that implement interface as documented as it won't be possible to start
old-qemu-4.0 -device cpufoo,die-id=0,...
since that "-device cpufoo,die-id=0,..." were used on new-qemu source.

PS:
Adding affected targets maintainers to the loop to see if we can drop restriction.

Even though it works fine for die-id and I don't see immediate problems with relaxing rule, I reluctant to do it, since instead of simple
"add all properties you were told to"
implicit rules would evolve into mess similar to smp_parse() over time.

Also if we would need to change implicit values logic down the road it would be a pain like with any default parameters in QEMU, which is a good reason against relaxing rule.