Skip to contentSkip to conversation

monitor: turn QMP and HMP into QOM objects

PATCH Superseded Daniel P. Berrangé 4 participants 71 messages Jun 24 – Jul 3

Subsystems (4)

v5 of 7 Superseded Latest is v7 →

lore
Share & tools
lore
#0 lore raw

v5 cover monitor: turn QMP and HMP into QOM objects

Conceptually -object and object_add/object_del should be sufficient for essentially all QEMU configuration....if only we ported all our internal custom backends/devices/etc to QOM. That is of course a big job which is why it hasn't happened.

This series started with the premise that the monitor is one of the easier areas to convert since we have no more than three classes, a common base, and QMP and HMP subclasses[1]. So why not give it a go and thus unlock the ability to dynamically create/delete monitors in QMP/HMP.

This series does the conversion in a great many small steps to better understand the implications at each stage.

The high level outcome of this series is

* HMP and QMP monitors are QOM objects, 'monitor-hmp' and 'monitor-qmp' respectively

* Both can be cold plugged and hot plugged. QMP only, can also be hot unplugged.

* '-mon' is obsolete, deprecated and replaced by '-object', but -monitor, -qmp and kept as high level syntax sugar

* QMP gains a concept of "close-action" which makes it possible to mark a monitor for auto-delete.

The monitor hot-unplug code and the qtest and functional testing code is heavily derived from a series sent by Christian Brauner which proposed new monitor_add/monitor_del commands:

https://lists.nongnu.org/archive/html/qemu-devel/2026-04/msg01349.html

I left Christian's authorship & SoB on the patches which were derived from his code, though the code has been refactored quite a bit in places, so bugs are quite possibly my own.

Note that Christian's series allowed the use of "monitor_del" commands against the current monitor session. ie a client could delete the very monitor it was using. This is an awkward concept as it needs special casing to delay the deletion to happen in the background, such that that the QMP response to 'monitor_del' could still be sent back. This also left the chardev was orphaned as there's no way to run 'chardev_dev' in that usage pattern.

To provide an alternative mechanism to address the same use case, this series introduces the 'close-action' concept mentioned above, that allows hotplugging a monitor to service a specific task, with the monitor being purged when the script closes its connection. This avoids the special casing that an explicit "self deletion" paradigm required.

While supporting hotplug of HMP was trivial, I didn't do any work to think about hotunplug of HMP, since IMHO it is of limited value given HMP's typical use cases.

[1] ~~~ we did all this not because it was easy, but because we thought it would be easy ~~~

Changed in v5:

- Address another race when looking up monitor by ID
- Refer to "line editting" when talking about HMP readline feature
- Misc typos
- Fix mis-placed setting of 'delete_pending'
- Add test case to validate reconnecting to QMP

Changed in v4:

- Hold reference into BH to avoid race with auto-delete

Changed in v3:

- Removed unused 'dead' struct field
- Make use of 'setup_pending' struct field across BH
- Add missing free of chardev_id
- Add missnig ERRP_GUARD in monitor_new_qmp
- Misc docs typos / rephrasing
- Moved docs patch to the end
- Fixed docs about the default QOM ID naming for legacy monitor syntax
- Fixed random sleep time calculation in tests

Christian Brauner (6):
monitor: convert from oneshot BH to persistent BH
monitor: reject attempts to delete the current monitor
monitor: protect qemu_chr_fe_accept_input with monitor lock
monitor: implement support for deleting QMP objects tests/qtest: add tests for dynamic monitor add/remove tests/functional: add e2e test for dynamic QMP monitor hotplug

Daniel P. Berrangé (29):
qom: replace 'can_be_deleted' with 'prepare_delete'
monitor: replace 'common' with 'parent_obj' in MonitorHMP
monitor: replace 'common' with 'parent_obj' in MonitorQMP
monitor: rename monitor_init* to monitor_new*
monitor: minimal conversion of monitors to QOM
monitor: add 'chardev' property to Monitor base class
monitor: add 'readline' property to HMP Monitor class
monitor: add 'pretty' property to QMP Monitor class
monitor: remove 'skip_flush' field
monitor: move monitor_data_(init|destroy) into QOM init/finalize
monitor: use class methods for monitor_vprintf
monitor: use class methods for monitor_qapi_event_emit
monitor: use class methods for monitor_accept_input
monitor: use class method for I/O thread request
monitor: use dynamic cast in monitor_qmp_requests_pop_any_with_lock
util: use dynamic cast in error vreport
monitor: drop unused monitor_cur_is_qmp
monitor: use dynamic cast in QMP commands
monitor: use dynamic cast in monitor_is_hmp_non_interactive
monitor: drop unused monitor_is_qmp method
monitor: eliminate monitor_is_hmp_non_interactive method
monitor: implement "user creatable" interface for adding monitors tests/functional: add a stress test for monitor hot unplug
qom: add method for getting the "id" of a QOM object
qom: add trace events for user creatable create/delete APIs
monitor: add support for auto-deleting monitors upon close
tests: switch from -mon to -object monitor-qmp
qemu-options: document new monitor-hmp and monitor-qmp objects
docs: mark '-mon' as deprecated in favour of -object

 MAINTAINERS                                   |   1 +
 backends/cryptodev.c                          |  10 +-
 backends/hostmem.c                            |   5 +-
 backends/iommufd.c                            |  10 +-
 block/throttle-groups.c                       |  10 +-
 chardev/char.c                                |   3 +-
 docs/about/deprecated.rst                     |  10 +
 docs/devel/writing-monitor-commands.rst       |   4 +-
 docs/system/arm/xenpvh.rst                    |   4 +-
 docs/system/i386/xen.rst                      |   3 +-
 docs/system/i386/xenpvh.rst                   |   4 +-
 event-loop-base.c                             |   8 +-
 gdbstub/system.c                              |   4 +-
 include/monitor/monitor.h                     |  23 +-
 include/qom/object.h                          |  10 +
 include/qom/object_interfaces.h               |  26 +-
 include/system/event-loop-base.h              |   2 +-
 migration/migration-hmp-cmds.c                |   5 +-
 monitor/hmp-cmds.c                            |   7 +-
 monitor/hmp.c                                 | 181 ++++++++--
 monitor/monitor-internal.h                    |  73 ++--
 monitor/monitor.c                             | 262 +++++++-------
 monitor/qmp-cmds-control.c                    |  12 +-
 monitor/qmp-cmds.c                            |  14 +-
 monitor/qmp.c                                 | 319 +++++++++++++++---
 net/can/can_core.c                            |   5 +-
 python/qemu/machine/machine.py                |   4 +-
 qapi/qom.json                                 |  62 ++++
 qemu-options.hx                               |  56 ++-
 qom/object.c                                  |  17 +
 qom/object_interfaces.c                       |  20 +-
 qom/trace-events                              |   5 +
 storage-daemon/qemu-storage-daemon.c          |   2 +-
 stubs/monitor-core.c                          |   5 -
 stubs/monitor-internal.c                      |   3 +-
 system/vl.c                                   |  12 +-
 tests/functional/generic/meson.build          |   1 +
 .../generic/test_monitor_hotplug.py           | 305 +++++++++++++++++
 tests/qemu-iotests/245                        |   4 +-
 tests/qtest/libqtest.c                        |   2 +-
 tests/qtest/qmp-test.c                        | 174 ++++++++++
 tests/unit/test-util-sockets.c                |   1 -
 tools/qemu-vnc/stubs.c                        |   5 -
 ui/ui-hmp-cmds.c                              |   2 +-
 util/error-report.c                           |  13 +-
 util/main-loop.c                              |   5 +-
 46 files changed, 1386 insertions(+), 327 deletions(-)
 create mode 100755 tests/functional/generic/test_monitor_hotplug.py
Signature · 1 line
2.54.0
#6 lore raw

PATCH v5 2/35 monitor: replace 'common' with 'parent_obj' in MonitorHMP

The field name 'parent_obj' is standard practice for QOM structs so align the HMP monitor.

---
 monitor/hmp-cmds.c         |  2 +-
 monitor/hmp.c              | 43 ++++++++++++++++++++------------------
 monitor/monitor-internal.h |  2 +-
 monitor/monitor.c          |  6 +++---
 monitor/qmp-cmds.c         | 10 ++++-----
 ui/ui-hmp-cmds.c           |  2 +-
 6 files changed, 34 insertions(+), 31 deletions(-)
monitor/hmp-cmds.c +1 -1
index 443b8c785d..e139caeba9 100644
@@ -287,7 +287,7 @@ void hmp_info_sync_profile(Monitor *mon, const QDict *qdict)
 
 void hmp_info_history(Monitor *mon, const QDict *qdict)
 {
-    MonitorHMP *hmp_mon = container_of(mon, MonitorHMP, common);
+    MonitorHMP *hmp_mon = container_of(mon, MonitorHMP, parent_obj);
     int i;
     const char *str;
 
monitor/hmp.c +23 -20

130 lines not on this page · raw message

monitor/monitor-internal.h +1 -1
index a5c4aba306..eb301ea796 100644
@@ -128,7 +128,7 @@ struct Monitor {
 };
 
 struct MonitorHMP {
-    Monitor common;
+    Monitor parent_obj;
     bool use_readline;
     /*
      * State used only in the thread "owning" the monitor.
monitor/monitor.c +3 -3
index 00b93ed612..2a1b511ff4 100644
@@ -132,7 +132,7 @@ static inline bool monitor_is_hmp_non_interactive(const Monitor *mon)
         return false;
     }
 
-    return !monitor_uses_readline(container_of(mon, MonitorHMP, common));
+    return !monitor_uses_readline(container_of(mon, MonitorHMP, parent_obj));
 }
 
 static gboolean monitor_unblocked(void *do_not_use, GIOCondition cond,
@@ -542,7 +542,7 @@ static void monitor_accept_input(void *opaque)
 
     qemu_mutex_lock(&mon->mon_lock);
     if (!monitor_is_qmp(mon) && mon->reset_seen) {
-        MonitorHMP *hmp_mon = container_of(mon, MonitorHMP, common);
+        MonitorHMP *hmp_mon = container_of(mon, MonitorHMP, parent_obj);
         assert(hmp_mon->rs);
         readline_restart(hmp_mon->rs);
         qemu_mutex_unlock(&mon->mon_lock);
@@ -627,7 +627,7 @@ void monitor_data_destroy(Monitor *mon)
     if (monitor_is_qmp(mon)) {
         monitor_data_destroy_qmp(container_of(mon, MonitorQMP, common));
     } else {
-        readline_free(container_of(mon, MonitorHMP, common)->rs);
+        readline_free(container_of(mon, MonitorHMP, parent_obj)->rs);
     }
     g_string_free(mon->outbuf, true);
     qemu_mutex_destroy(&mon->mon_lock);
monitor/qmp-cmds.c +5 -5
index 0c409c27dc..aa9ee8a391 100644
@@ -168,10 +168,10 @@ char *qmp_human_monitor_command(const char *command_line, bool has_cpu_index,
     char *output = NULL;
     MonitorHMP hmp = {};
 
-    monitor_data_init(&hmp.common, false, true, false);
+    monitor_data_init(&hmp.parent_obj, false, true, false);
 
     if (has_cpu_index) {
-        int ret = monitor_set_cpu(&hmp.common, cpu_index);
+        int ret = monitor_set_cpu(&hmp.parent_obj, cpu_index);
         if (ret < 0) {
             error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cpu-index",
                        "a CPU number");
@@ -181,12 +181,12 @@ char *qmp_human_monitor_command(const char *command_line, bool has_cpu_index,
 
     handle_hmp_command(&hmp, command_line);
 
-    WITH_QEMU_LOCK_GUARD(&hmp.common.mon_lock) {
-        output = g_strdup(hmp.common.outbuf->str);
+    WITH_QEMU_LOCK_GUARD(&hmp.parent_obj.mon_lock) {
+        output = g_strdup(hmp.parent_obj.outbuf->str);
     }
 
 out:
-    monitor_data_destroy(&hmp.common);
+    monitor_data_destroy(&hmp.parent_obj);
     return output;
 }
 
ui/ui-hmp-cmds.c +1 -1
index ee3e731d07..06f4030ce4 100644
@@ -343,7 +343,7 @@ void hmp_change_vnc(Monitor *mon, const char *device, const char *target,
         return;
     }
     if (!arg) {
-        MonitorHMP *hmp_mon = container_of(mon, MonitorHMP, common);
+        MonitorHMP *hmp_mon = container_of(mon, MonitorHMP, parent_obj);
         monitor_read_password(hmp_mon, hmp_change_read_arg, NULL);
     } else {
         qmp_change_vnc_password(arg, errp);
Signature · 1 line
2.54.0
#1 lore raw

PATCH v5 1/35 qom: replace 'can_be_deleted' with 'prepare_delete'

In reply to Daniel P. Berrangé

While most objects can perform all their cleanup in the finalizer method, there can be interactions with other resources / subsystems / threads which require that some cleanup be performed on an user creatable object before unparenting it and entering finalization.

The current 'can_be_deleted' method runs in the deletion path and is intended to be used to block deletion. While it could be used to perform cleanup tasks, its name suggests it should be free of side-effects.

Generalize this by renaming it to 'prepare_delete', explicitly allowing for cleanup to be provided. Existing users of 'can_be_deleted' are re-written, which provides them with more detailed/tailored error messages.

---
 backends/cryptodev.c             | 10 +++++++---
 backends/hostmem.c               |  5 +++--
 backends/iommufd.c               | 10 +++++++---
 block/throttle-groups.c          | 10 +++++++---
 event-loop-base.c                |  8 ++++----
 include/qom/object_interfaces.h  | 26 +++++++++++++++++---------
 include/system/event-loop-base.h |  2 +-
 net/can/can_core.c               |  5 +++--
 qom/object_interfaces.c          | 14 ++++++--------
 tests/qemu-iotests/245           |  4 ++--
 util/main-loop.c                 |  5 +++--
 11 files changed, 60 insertions(+), 39 deletions(-)
backends/cryptodev.c +7 -3
index 79f8882d3b..90110dc633 100644
@@ -454,9 +454,13 @@ bool cryptodev_backend_is_ready(CryptoDevBackend *backend)
 }
 
 static bool
-cryptodev_backend_can_be_deleted(UserCreatable *uc)
+cryptodev_backend_prepare_delete(UserCreatable *uc, Error **errp)
 {
-    return !cryptodev_backend_is_used(CRYPTODEV_BACKEND(uc));
+    if (cryptodev_backend_is_used(CRYPTODEV_BACKEND(uc))) {
+        error_setg(errp, "Cryptodev backend is still in use");
+        return false;
+    }
+    return true;
 }
 
 static void cryptodev_backend_instance_init(Object *obj)
@@ -613,7 +617,7 @@ cryptodev_backend_class_init(ObjectClass *oc, const void *data)
     UserCreatableClass *ucc = USER_CREATABLE_CLASS(oc);
 
     ucc->complete = cryptodev_backend_complete;
-    ucc->can_be_deleted = cryptodev_backend_can_be_deleted;
+    ucc->prepare_delete = cryptodev_backend_prepare_delete;
 
     QTAILQ_INIT(&crypto_clients);
     object_class_property_add(oc, "queues", "uint32",
backends/hostmem.c +3 -2
index cd2085fb3c..eb915c64bc 100644
@@ -434,9 +434,10 @@ host_memory_backend_memory_complete(UserCreatable *uc, Error **errp)
 }
 
 static bool
-host_memory_backend_can_be_deleted(UserCreatable *uc)
+host_memory_backend_prepare_delete(UserCreatable *uc, Error **errp)
 {
     if (host_memory_backend_is_mapped(MEMORY_BACKEND(uc))) {
+        error_setg(errp, "Host memory backend is still mapped");
         return false;
     } else {
         return true;
@@ -508,7 +509,7 @@ host_memory_backend_class_init(ObjectClass *oc, const void *data)
     UserCreatableClass *ucc = USER_CREATABLE_CLASS(oc);
 
     ucc->complete = host_memory_backend_memory_complete;
-    ucc->can_be_deleted = host_memory_backend_can_be_deleted;
+    ucc->prepare_delete = host_memory_backend_prepare_delete;
 
     object_class_property_add_bool(oc, "merge",
         host_memory_backend_get_merge,
backends/iommufd.c +7 -3
index 09624cd652..7af49f6de3 100644
@@ -63,11 +63,15 @@ static void iommufd_backend_set_fd(Object *obj, const char *str, Error **errp)
     trace_iommu_backend_set_fd(be->fd);
 }
 
-static bool iommufd_backend_can_be_deleted(UserCreatable *uc)
+static bool iommufd_backend_prepare_delete(UserCreatable *uc, Error **errp)
 {
     IOMMUFDBackend *be = IOMMUFD_BACKEND(uc);
 
-    return !be->users;
+    if (be->users) {
+        error_setg(errp, "IOMMUFD backend still has %d users", be->users);
+        return false;
+    }
+    return true;
 }
 
 static void iommufd_backend_complete(UserCreatable *uc, Error **errp)
@@ -92,7 +96,7 @@ static void iommufd_backend_class_init(ObjectClass *oc, const void *data)
 {
     UserCreatableClass *ucc = USER_CREATABLE_CLASS(oc);
 
-    ucc->can_be_deleted = iommufd_backend_can_be_deleted;
+    ucc->prepare_delete = iommufd_backend_prepare_delete;
     ucc->complete = iommufd_backend_complete;
 
     object_class_property_add_str(oc, "fd", NULL, iommufd_backend_set_fd);
block/throttle-groups.c +7 -3
index 4b1b1944c2..7836fc4c76 100644
@@ -960,9 +960,13 @@ static void throttle_group_get_limits(Object *obj, Visitor *v,
     visit_type_ThrottleLimits(v, name, &argp, errp);
 }
 
-static bool throttle_group_can_be_deleted(UserCreatable *uc)
+static bool throttle_group_prepare_delete(UserCreatable *uc, Error **errp)
 {
-    return OBJECT(uc)->ref == 1;
+    if (OBJECT(uc)->ref > 1) {
+        error_setg(errp, "Throttle group still has multiple references");
+        return false;
+    }
+    return true;
 }
 
 static void throttle_group_obj_class_init(ObjectClass *klass,
@@ -972,7 +976,7 @@ static void throttle_group_obj_class_init(ObjectClass *klass,
     UserCreatableClass *ucc = USER_CREATABLE_CLASS(klass);
 
     ucc->complete = throttle_group_obj_complete;
-    ucc->can_be_deleted = throttle_group_can_be_deleted;
+    ucc->prepare_delete = throttle_group_prepare_delete;
 
     /* individual properties */
     for (i = 0; i < sizeof(properties) / sizeof(ThrottleParamInfo); i++) {
event-loop-base.c +4 -4
index 8ca143bea4..ee5987ddbf 100644
@@ -85,13 +85,13 @@ static void event_loop_base_complete(UserCreatable *uc, Error **errp)
     }
 }
 
-static bool event_loop_base_can_be_deleted(UserCreatable *uc)
+static bool event_loop_base_prepare_delete(UserCreatable *uc, Error **errp)
 {
     EventLoopBaseClass *bc = EVENT_LOOP_BASE_GET_CLASS(uc);
     EventLoopBase *backend = EVENT_LOOP_BASE(uc);
 
-    if (bc->can_be_deleted) {
-        return bc->can_be_deleted(backend);
+    if (bc->prepare_delete) {
+        return bc->prepare_delete(backend, errp);
     }
 
     return true;
@@ -102,7 +102,7 @@ static void event_loop_base_class_init(ObjectClass *klass,
 {
     UserCreatableClass *ucc = USER_CREATABLE_CLASS(klass);
     ucc->complete = event_loop_base_complete;
-    ucc->can_be_deleted = event_loop_base_can_be_deleted;
+    ucc->prepare_delete = event_loop_base_prepare_delete;
 
     object_class_property_add(klass, "aio-max-batch", "int",
                               event_loop_base_get_param,
include/qom/object_interfaces.h +17 -9
index e2b8615617..afd0fb93b2 100644
@@ -20,8 +20,10 @@ typedef struct UserCreatable UserCreatable;
  * UserCreatableClass:
  * @parent_class: the base class
  * @complete: callback to be called after @obj's properties are set.
- * @can_be_deleted: callback to be called before an object is removed
- * to check if @obj can be removed safely.
+ * @prepare_delete: to be called before an attempt to delete @obj
+ * to validate whether the object can be deleted and trigger any
+ * cleanup of any resources which have to be dealt with before the
+ * object is unparented and enters finalization.
  *
  * Interface is designed to work with -object/object-add/object_add
  * commands.
@@ -36,7 +38,9 @@ typedef struct UserCreatable UserCreatable;
  * For objects created without using -object/object-add/object_add,
  * @user_creatable_complete() wrapper should be called manually if
  * object's type implements USER_CREATABLE interface and needs
- * complete() callback to be called.
+ * complete() callback to be called. Similarly @user_creatable_prepare_delete()
+ * should be called manually prior to an attempt to the delete the
+ * object.
  */
 struct UserCreatableClass {
     /* <private> */
@@ -44,7 +48,7 @@ struct UserCreatableClass {
 
     /* <public> */
     void (*complete)(UserCreatable *uc, Error **errp);
-    bool (*can_be_deleted)(UserCreatable *uc);
+    bool (*prepare_delete)(UserCreatable *uc, Error **errp);
 };
 
 /**
@@ -61,13 +65,17 @@ struct UserCreatableClass {
 bool user_creatable_complete(UserCreatable *uc, Error **errp);
 
 /**
- * user_creatable_can_be_deleted:
- * @uc: the object whose can_be_deleted() method is called if implemented
+ * user_creatable_prepare_delete:
+ * @uc: the user-creatable object whose prepare_delete() method is called
+ * @errp: if an error occurs, a pointer to an area to store the error
+ *
+ * Wrapper to call prepare_delete() class method if defined, otherwise
+ * does nothing.
  *
- * Wrapper to call can_be_deleted() method if one of types it's inherited
- * from implements USER_CREATABLE interface.
+ * Returns: %true on success or if prepare_delete() is not defined,
+ *          %false on failure.
  */
-bool user_creatable_can_be_deleted(UserCreatable *uc);
+bool user_creatable_prepare_delete(UserCreatable *uc, Error **errp);
 
 /**
  * user_creatable_add_qapi:
include/system/event-loop-base.h +1 -1
index 130629e7f3..1e1e427ac7 100644
@@ -24,7 +24,7 @@ struct EventLoopBaseClass {
 
     void (*init)(EventLoopBase *base, Error **errp);
     void (*update_params)(EventLoopBase *base, Error **errp);
-    bool (*can_be_deleted)(EventLoopBase *base);
+    bool (*prepare_delete)(EventLoopBase *base, Error **errp);
 };
 
 struct EventLoopBase {
net/can/can_core.c +3 -2
index 77fe2b8ba4..8df9375167 100644
@@ -143,8 +143,9 @@ int can_bus_client_set_filters(CanBusClientState *client,
 }
 
 
-static bool can_bus_can_be_deleted(UserCreatable *uc)
+static bool can_bus_prepare_delete(UserCreatable *uc, Error **errp)
 {
+    error_setg(errp, "Deleting can bus devices is not supported");
     return false;
 }
 
@@ -153,7 +154,7 @@ static void can_bus_class_init(ObjectClass *klass,
 {
     UserCreatableClass *uc_klass = USER_CREATABLE_CLASS(klass);
 
-    uc_klass->can_be_deleted = can_bus_can_be_deleted;
+    uc_klass->prepare_delete = can_bus_prepare_delete;
 }
 
 static const TypeInfo can_bus_info = {
qom/object_interfaces.c +6 -8
index 7080f85f95..6faa0b2fd9 100644
@@ -32,16 +32,15 @@ bool user_creatable_complete(UserCreatable *uc, Error **errp)
     return !*errp;
 }
 
-bool user_creatable_can_be_deleted(UserCreatable *uc)
+bool user_creatable_prepare_delete(UserCreatable *uc, Error **errp)
 {
-
     UserCreatableClass *ucc = USER_CREATABLE_GET_CLASS(uc);
+    ERRP_GUARD();
 
-    if (ucc->can_be_deleted) {
-        return ucc->can_be_deleted(uc);
-    } else {
-        return true;
+    if (ucc->prepare_delete) {
+        ucc->prepare_delete(uc, errp);
     }
+    return !*errp;
 }
 
 void user_creatable_add_qapi(ObjectOptions *options, Error **errp)
@@ -253,8 +252,7 @@ bool user_creatable_del(const char *id, Error **errp)
         return false;
     }
 
-    if (!user_creatable_can_be_deleted(USER_CREATABLE(obj))) {
-        error_setg(errp, "object '%s' is in use, can not be deleted", id);
+    if (!user_creatable_prepare_delete(USER_CREATABLE(obj), errp)) {
         return false;
     }
 
tests/qemu-iotests/245 +2 -2
index f96610f510..e161bcfda9 100755
@@ -801,7 +801,7 @@ class TestBlockdevReopen(iotests.QMPTestCase):
         # Now group1 is in use, it cannot be deleted
         result = self.vm.qmp('object-del', id = 'group1')
         self.assert_qmp(result, 'error/class', 'GenericError')
-        self.assert_qmp(result, 'error/desc', "object 'group1' is in use, can not be deleted")
+        self.assert_qmp(result, 'error/desc', "Throttle group still has multiple references")
 
         # Default options, this switches the group back to group0
         self.reopen(opts)
@@ -809,7 +809,7 @@ class TestBlockdevReopen(iotests.QMPTestCase):
         # So now we cannot delete group0
         result = self.vm.qmp('object-del', id = 'group0')
         self.assert_qmp(result, 'error/class', 'GenericError')
-        self.assert_qmp(result, 'error/desc', "object 'group0' is in use, can not be deleted")
+        self.assert_qmp(result, 'error/desc', "Throttle group still has multiple references")
 
         # But group1 is free this time, and it can be deleted
         self.vm.cmd('object-del', id = 'group1')
util/main-loop.c +3 -2
index ad8645c30a..67ee06c311 100644
@@ -218,8 +218,9 @@ static void main_loop_init(EventLoopBase *base, Error **errp)
     mloop = m;
 }
 
-static bool main_loop_can_be_deleted(EventLoopBase *base)
+static bool main_loop_prepare_delete(EventLoopBase *base, Error **errp)
 {
+    error_setg(errp, "Deleting main loop is not supported");
     return false;
 }
 
@@ -229,7 +230,7 @@ static void main_loop_class_init(ObjectClass *oc, const void *class_data)
 
     bc->init = main_loop_init;
     bc->update_params = main_loop_update_params;
-    bc->can_be_deleted = main_loop_can_be_deleted;
+    bc->prepare_delete = main_loop_prepare_delete;
 }
 
 static const TypeInfo main_loop_info = {
Signature · 1 line
2.54.0
#8 lore raw

PATCH v5 4/35 monitor: rename monitor_init* to monitor_new*

In reply to Daniel P. Berrangé

The current "monitor_init" functions will clash with the methods of the same name that are required by QOM. To ease the transition to QOM, rename them out of the way.

---
 chardev/char.c                       |  2 +-
 gdbstub/system.c                     |  2 +-
 include/monitor/monitor.h            |  8 ++++----
 monitor/hmp.c                        |  2 +-
 monitor/monitor.c                    | 10 +++++-----
 monitor/qmp.c                        |  2 +-
 storage-daemon/qemu-storage-daemon.c |  2 +-
 stubs/monitor-internal.c             |  2 +-
 system/vl.c                          |  2 +-
 9 files changed, 16 insertions(+), 16 deletions(-)
chardev/char.c +1 -1
index ca8b37ed8d..813c04d953 100644
@@ -805,7 +805,7 @@ static Chardev *qemu_chr_new_from_name(const char *label, const char *filename,
 
     if (qemu_opt_get_bool(opts, "mux", 0)) {
         assert(permit_mux_mon);
-        monitor_init_hmp(chr, true, &err);
+        monitor_new_hmp(chr, true, &err);
         if (err) {
             error_report_err(err);
             object_unparent(OBJECT(chr));
gdbstub/system.c +1 -1
index 2063b63b2f..18f6a62b2e 100644
@@ -389,7 +389,7 @@ bool gdbserver_start(const char *device, Error **errp)
         /* Initialize a monitor terminal for gdb */
         mon_chr = qemu_chardev_new(NULL, TYPE_CHARDEV_GDB,
                                    NULL, NULL, &error_abort);
-        monitor_init_hmp(mon_chr, false, &error_abort);
+        monitor_new_hmp(mon_chr, false, &error_abort);
     } else {
         qemu_chr_fe_deinit(&gdbserver_system_state.chr, true);
         mon_chr = gdbserver_system_state.mon_chr;
include/monitor/monitor.h +4 -4
index 55649a8664..b9642b58ba 100644
@@ -19,10 +19,10 @@ bool monitor_cur_is_qmp(void);
 
 void monitor_init_globals(void);
 void monitor_init_globals_core(void);
-void monitor_init_qmp(Chardev *chr, bool pretty, Error **errp);
-void monitor_init_hmp(Chardev *chr, bool use_readline, Error **errp);
-int monitor_init(MonitorOptions *opts, bool allow_hmp, Error **errp);
-int monitor_init_opts(QemuOpts *opts, Error **errp);
+void monitor_new_qmp(Chardev *chr, bool pretty, Error **errp);
+void monitor_new_hmp(Chardev *chr, bool use_readline, Error **errp);
+int monitor_new(MonitorOptions *opts, bool allow_hmp, Error **errp);
+int monitor_new_opts(QemuOpts *opts, Error **errp);
 void monitor_cleanup(void);
 
 int monitor_suspend(Monitor *mon);
monitor/hmp.c +1 -1
index 3dde52ddbe..4e4468424a 100644
@@ -1524,7 +1524,7 @@ static void monitor_readline_flush(void *opaque)
     monitor_flush(&mon->parent_obj);
 }
 
-void monitor_init_hmp(Chardev *chr, bool use_readline, Error **errp)
+void monitor_new_hmp(Chardev *chr, bool use_readline, Error **errp)
 {
     MonitorHMP *mon = g_new0(MonitorHMP, 1);
 
monitor/monitor.c +5 -5
index 0b7b464ebe..a87597e606 100644
@@ -715,7 +715,7 @@ void monitor_init_globals(void)
     aio_co_schedule(iohandler_get_aio_context(), qmp_dispatcher_co);
 }
 
-int monitor_init(MonitorOptions *opts, bool allow_hmp, Error **errp)
+int monitor_new(MonitorOptions *opts, bool allow_hmp, Error **errp)
 {
     ERRP_GUARD();
     Chardev *chr;
@@ -732,7 +732,7 @@ int monitor_init(MonitorOptions *opts, bool allow_hmp, Error **errp)
 
     switch (opts->mode) {
     case MONITOR_MODE_CONTROL:
-        monitor_init_qmp(chr, opts->pretty, errp);
+        monitor_new_qmp(chr, opts->pretty, errp);
         break;
     case MONITOR_MODE_READLINE:
         if (!allow_hmp) {
@@ -743,7 +743,7 @@ int monitor_init(MonitorOptions *opts, bool allow_hmp, Error **errp)
             error_setg(errp, "'pretty' is not compatible with HMP monitors");
             return -1;
         }
-        monitor_init_hmp(chr, true, errp);
+        monitor_new_hmp(chr, true, errp);
         break;
     default:
         g_assert_not_reached();
@@ -752,7 +752,7 @@ int monitor_init(MonitorOptions *opts, bool allow_hmp, Error **errp)
     return *errp ? -1 : 0;
 }
 
-int monitor_init_opts(QemuOpts *opts, Error **errp)
+int monitor_new_opts(QemuOpts *opts, Error **errp)
 {
     Visitor *v;
     MonitorOptions *options;
@@ -765,7 +765,7 @@ int monitor_init_opts(QemuOpts *opts, Error **errp)
         return -1;
     }
 
-    ret = monitor_init(options, true, errp);
+    ret = monitor_new(options, true, errp);
     qapi_free_MonitorOptions(options);
     return ret;
 }
monitor/qmp.c +1 -1
index 27934206db..cb28a95efd 100644
@@ -513,7 +513,7 @@ static void monitor_qmp_setup_handlers_bh(void *opaque)
     monitor_list_append(&mon->parent_obj);
 }
 
-void monitor_init_qmp(Chardev *chr, bool pretty, Error **errp)
+void monitor_new_qmp(Chardev *chr, bool pretty, Error **errp)
 {
     MonitorQMP *mon = g_new0(MonitorQMP, 1);
 
storage-daemon/qemu-storage-daemon.c +1 -1
index eb72561358..50dbfbd97a 100644
@@ -330,7 +330,7 @@ static void process_options(int argc, char *argv[], bool pre_init_pass)
                 visit_free(v);
 
                 /* TODO Catch duplicate monitor IDs */
-                monitor_init(monitor, false, &error_fatal);
+                monitor_new(monitor, false, &error_fatal);
                 qapi_free_MonitorOptions(monitor);
                 break;
             }
stubs/monitor-internal.c +1 -1
index 4fece49d53..23d58da184 100644
@@ -8,6 +8,6 @@ int monitor_get_fd(Monitor *mon, const char *name, Error **errp)
     return -1;
 }
 
-void monitor_init_hmp(Chardev *chr, bool use_readline, Error **errp)
+void monitor_new_hmp(Chardev *chr, bool use_readline, Error **errp)
 {
 }
system/vl.c +1 -1
index 1c0da7df29..fb36140aac 100644
@@ -1247,7 +1247,7 @@ static int fsdev_init_func(void *opaque, QemuOpts *opts, Error **errp)
 
 static int mon_init_func(void *opaque, QemuOpts *opts, Error **errp)
 {
-    return monitor_init_opts(opts, errp);
+    return monitor_new_opts(opts, errp);
 }
 
 static void monitor_parse(const char *str, const char *mode, bool pretty)
Signature · 1 line
2.54.0
#21 lore raw

PATCH v5 11/35 monitor: use class methods for monitor_vprintf

In reply to Daniel P. Berrangé

This removes the need for using monitor_is_qmp() to check the subclass type, which is an anti-pattern.

---
 monitor/hmp.c              | 13 +++++++++++++
 monitor/monitor-internal.h |  7 +++++++
 monitor/monitor.c          | 11 ++++-------
 3 files changed, 24 insertions(+), 7 deletions(-)
monitor/hmp.c +13 -0
index 2bb2333da5..08055839a4 100644
@@ -68,11 +68,18 @@ static void monitor_hmp_set_readline(Object *obj, bool val, Error **errp)
     mon->use_readline = val;
 }
 
+int monitor_hmp_vprintf(Monitor *mon, const char *fmt, va_list ap)
+    G_GNUC_PRINTF(2, 0);
+
 static void monitor_hmp_class_init(ObjectClass *cls, const void *data)
 {
+    MonitorClass *moncls = MONITOR_CLASS(cls);
+
     object_class_property_add_bool(cls, "readline",
                                    monitor_hmp_get_readline,
                                    monitor_hmp_set_readline);
+
+    moncls->vprintf = monitor_hmp_vprintf;
 }
 
 static void monitor_hmp_init(Object *obj)
@@ -87,6 +94,12 @@ static void monitor_hmp_init(Object *obj)
     hmp->use_readline = true;
 }
 
+int monitor_hmp_vprintf(Monitor *mon, const char *fmt, va_list ap)
+{
+    g_autofree char *buf = g_strdup_vprintf(fmt, ap);
+    return monitor_puts(mon, buf);
+}
+
 static void monitor_command_cb(void *opaque, const char *cmdline,
                                void *readline_opaque)
 {
monitor/monitor-internal.h +7 -0
index 6a4c3b0f37..23cef68153 100644
@@ -104,6 +104,13 @@ typedef struct HMPCommand {
 
 struct MonitorClass {
     ObjectClass parent_class;
+
+    /*
+     * If non-NULL, the monitor is able to print messages
+     * for attention of the client user
+     */
+    int (*vprintf)(Monitor *mon, const char *fmt, va_list ap)
+        G_GNUC_PRINTF(2, 0);
 };
 
 struct Monitor {
monitor/monitor.c +4 -7
index 76df379a2c..8b70380d64 100644
@@ -262,21 +262,18 @@ int monitor_puts(Monitor *mon, const char *str)
 
 int monitor_vprintf(Monitor *mon, const char *fmt, va_list ap)
 {
-    char *buf;
-    int n;
+    MonitorClass *moncls;
 
     if (!mon) {
         return -1;
     }
 
-    if (monitor_is_qmp(mon)) {
+    moncls = MONITOR_GET_CLASS(mon);
+    if (!moncls->vprintf) {
         return -1;
     }
 
-    buf = g_strdup_vprintf(fmt, ap);
-    n = monitor_puts(mon, buf);
-    g_free(buf);
-    return n;
+    return moncls->vprintf(mon, fmt, ap);
 }
 
 int monitor_printf(Monitor *mon, const char *fmt, ...)
Signature · 1 line
2.54.0
#7 lore raw

PATCH v5 3/35 monitor: replace 'common' with 'parent_obj' in MonitorQMP

In reply to Daniel P. Berrangé

The field name 'parent_obj' is standard practice for QOM structs so align the QMP monitor.

---
 monitor/monitor-internal.h |  2 +-
 monitor/monitor.c          |  4 ++--
 monitor/qmp-cmds-control.c |  4 ++--
 monitor/qmp.c              | 40 +++++++++++++++++++-------------------
 4 files changed, 25 insertions(+), 25 deletions(-)
monitor/monitor-internal.h +1 -1
index eb301ea796..4da2b2a677 100644
@@ -141,7 +141,7 @@ struct MonitorHMP {
 };
 
 typedef struct {
-    Monitor common;
+    Monitor parent_obj;
     JSONMessageParser parser;
     bool pretty;
     /*
monitor/monitor.c +2 -2
index 2a1b511ff4..0b7b464ebe 100644
@@ -308,7 +308,7 @@ static void monitor_qapi_event_emit(QAPIEvent event, QDict *qdict)
             continue;
         }
 
-        qmp_mon = container_of(mon, MonitorQMP, common);
+        qmp_mon = container_of(mon, MonitorQMP, parent_obj);
         {
             QEMU_LOCK_GUARD(&mon->mon_lock);
             if (qmp_mon->commands == &qmp_cap_negotiation_commands) {
@@ -625,7 +625,7 @@ void monitor_data_destroy(Monitor *mon)
     g_free(mon->mon_cpu_path);
     qemu_chr_fe_deinit(&mon->chr, false);
     if (monitor_is_qmp(mon)) {
-        monitor_data_destroy_qmp(container_of(mon, MonitorQMP, common));
+        monitor_data_destroy_qmp(container_of(mon, MonitorQMP, parent_obj));
     } else {
         readline_free(container_of(mon, MonitorHMP, parent_obj)->rs);
     }
monitor/qmp-cmds-control.c +2 -2
index 150ca9f5cb..af6a2a118b 100644
@@ -76,7 +76,7 @@ void qmp_qmp_capabilities(bool has_enable, QMPCapabilityList *enable,
     MonitorQMP *mon;
 
     assert(monitor_is_qmp(cur_mon));
-    mon = container_of(cur_mon, MonitorQMP, common);
+    mon = container_of(cur_mon, MonitorQMP, parent_obj);
 
     if (mon->commands == &qmp_commands) {
         error_set(errp, ERROR_CLASS_COMMAND_NOT_FOUND,
@@ -126,7 +126,7 @@ CommandInfoList *qmp_query_commands(Error **errp)
     MonitorQMP *mon;
 
     assert(monitor_is_qmp(cur_mon));
-    mon = container_of(cur_mon, MonitorQMP, common);
+    mon = container_of(cur_mon, MonitorQMP, parent_obj);
 
     qmp_for_each_command(mon->commands, query_commands_cb, &list);
 
monitor/qmp.c +20 -20
index 687019811f..27934206db 100644
@@ -80,7 +80,7 @@ static void monitor_qmp_caps_reset(MonitorQMP *mon)
 {
     memset(mon->capab_offered, 0, sizeof(mon->capab_offered));
     memset(mon->capab, 0, sizeof(mon->capab));
-    mon->capab_offered[QMP_CAPABILITY_OOB] = mon->common.use_io_thread;
+    mon->capab_offered[QMP_CAPABILITY_OOB] = mon->parent_obj.use_io_thread;
 }
 
 static void qmp_request_free(QMPRequest *req)
@@ -124,7 +124,7 @@ static void monitor_qmp_cleanup_queue_and_resume(MonitorQMP *mon)
          * when we get here while the monitor is suspended.  An
          * unfortunately timed CHR_EVENT_CLOSED can do the trick.
          */
-        monitor_resume(&mon->common);
+        monitor_resume(&mon->parent_obj);
     }
 
 }
@@ -139,7 +139,7 @@ void qmp_send_response(MonitorQMP *mon, const QDict *rsp)
     trace_monitor_qmp_respond(mon, json->str);
 
     g_string_append_c(json, '\n');
-    monitor_puts(&mon->common, json->str);
+    monitor_puts(&mon->parent_obj, json->str);
 
     g_string_free(json, true);
 }
@@ -166,7 +166,7 @@ static void monitor_qmp_dispatch(MonitorQMP *mon, QObject *req)
     QDict *error;
 
     rsp = qmp_dispatch(mon->commands, req, qmp_oob_enabled(mon),
-                       &mon->common);
+                       &mon->parent_obj);
 
     if (mon->commands == &qmp_cap_negotiation_commands) {
         error = qdict_get_qdict(rsp, "error");
@@ -207,7 +207,7 @@ static QMPRequest *monitor_qmp_requests_pop_any_with_lock(void)
             continue;
         }
 
-        qmp_mon = container_of(mon, MonitorQMP, common);
+        qmp_mon = container_of(mon, MonitorQMP, parent_obj);
         qemu_mutex_lock(&qmp_mon->qmp_queue_lock);
         req_obj = g_queue_pop_head(qmp_mon->qmp_requests);
         if (req_obj) {
@@ -302,7 +302,7 @@ void coroutine_fn monitor_qmp_dispatcher_co(void *data)
         oob_enabled = qmp_oob_enabled(mon);
         if (oob_enabled
             && mon->qmp_requests->length == QMP_REQ_QUEUE_LEN_MAX - 1) {
-            monitor_resume(&mon->common);
+            monitor_resume(&mon->parent_obj);
         }
 
         /*
@@ -343,7 +343,7 @@ void coroutine_fn monitor_qmp_dispatcher_co(void *data)
         }
 
         if (!oob_enabled) {
-            monitor_resume(&mon->common);
+            monitor_resume(&mon->parent_obj);
         }
 
         qmp_request_free(req_obj);
@@ -408,7 +408,7 @@ static void handle_qmp_command(void *opaque, QObject *req, Error *err)
          */
         if (!qmp_oob_enabled(mon) ||
             mon->qmp_requests->length == QMP_REQ_QUEUE_LEN_MAX - 1) {
-            monitor_suspend(&mon->common);
+            monitor_suspend(&mon->parent_obj);
         }
 
         /*
@@ -462,7 +462,7 @@ static void monitor_qmp_event(void *opaque, QEMUChrEvent event)
 
     switch (event) {
     case CHR_EVENT_OPENED:
-        WITH_QEMU_LOCK_GUARD(&mon->common.mon_lock) {
+        WITH_QEMU_LOCK_GUARD(&mon->parent_obj.mon_lock) {
             mon->commands = &qmp_cap_negotiation_commands;
             monitor_qmp_caps_reset(mon);
         }
@@ -504,27 +504,27 @@ static void monitor_qmp_setup_handlers_bh(void *opaque)
     MonitorQMP *mon = opaque;
     GMainContext *context;
 
-    assert(mon->common.use_io_thread);
+    assert(mon->parent_obj.use_io_thread);
     context = iothread_get_g_main_context(mon_iothread);
     assert(context);
-    qemu_chr_fe_set_handlers(&mon->common.chr, monitor_can_read,
+    qemu_chr_fe_set_handlers(&mon->parent_obj.chr, monitor_can_read,
                              monitor_qmp_read, monitor_qmp_event,
-                             NULL, &mon->common, context, true);
-    monitor_list_append(&mon->common);
+                             NULL, &mon->parent_obj, context, true);
+    monitor_list_append(&mon->parent_obj);
 }
 
 void monitor_init_qmp(Chardev *chr, bool pretty, Error **errp)
 {
     MonitorQMP *mon = g_new0(MonitorQMP, 1);
 
-    if (!qemu_chr_fe_init(&mon->common.chr, chr, errp)) {
+    if (!qemu_chr_fe_init(&mon->parent_obj.chr, chr, errp)) {
         g_free(mon);
         return;
     }
-    qemu_chr_fe_set_echo(&mon->common.chr, true);
+    qemu_chr_fe_set_echo(&mon->parent_obj.chr, true);
 
     /* Note: we run QMP monitor in I/O thread when @chr supports that */
-    monitor_data_init(&mon->common, true, false,
+    monitor_data_init(&mon->parent_obj, true, false,
                       qemu_chr_has_feature(chr, QEMU_CHAR_FEATURE_GCONTEXT));
 
     mon->pretty = pretty;
@@ -533,7 +533,7 @@ void monitor_init_qmp(Chardev *chr, bool pretty, Error **errp)
     mon->qmp_requests = g_queue_new();
 
     json_message_parser_init(&mon->parser, handle_qmp_command, mon, NULL);
-    if (mon->common.use_io_thread) {
+    if (mon->parent_obj.use_io_thread) {
         /*
          * Make sure the old iowatch is gone.  It's possible when
          * e.g. the chardev is in client mode, with wait=on.
@@ -553,9 +553,9 @@ void monitor_init_qmp(Chardev *chr, bool pretty, Error **errp)
                                 monitor_qmp_setup_handlers_bh, mon);
         /* The bottom half will add @mon to @mon_list */
     } else {
-        qemu_chr_fe_set_handlers(&mon->common.chr, monitor_can_read,
+        qemu_chr_fe_set_handlers(&mon->parent_obj.chr, monitor_can_read,
                                  monitor_qmp_read, monitor_qmp_event,
-                                 NULL, &mon->common, NULL, true);
-        monitor_list_append(&mon->common);
+                                 NULL, &mon->parent_obj, NULL, true);
+        monitor_list_append(&mon->parent_obj);
     }
 }
Signature · 1 line
2.54.0
#22 lore raw

PATCH v5 12/35 monitor: use class methods for monitor_qapi_event_emit

In reply to Daniel P. Berrangé

This removes the need for using monitor_is_qmp() to check the subclass type, which is an anti-pattern.

---
 include/monitor/monitor.h  |  1 +
 monitor/monitor-internal.h |  5 +++++
 monitor/monitor.c          | 15 +++------------
 monitor/qmp.c              | 20 ++++++++++++++++++++
 4 files changed, 29 insertions(+), 12 deletions(-)
include/monitor/monitor.h +1 -0
index a135b3a590..8cb3db677b 100644
@@ -3,6 +3,7 @@
 
 #include "block/block.h"
 #include "qapi/qapi-types-misc.h"
+#include "qapi/qapi-emit-events.h"
 #include "qemu/readline.h"
 #include "exec/hwaddr.h"
 #include "qom/object.h"
monitor/monitor-internal.h +5 -0
index 23cef68153..763b5c9625 100644
@@ -111,6 +111,11 @@ struct MonitorClass {
      */
     int (*vprintf)(Monitor *mon, const char *fmt, va_list ap)
         G_GNUC_PRINTF(2, 0);
+    /*
+     * If non-NULL, the monitor is able to send event
+     * notifications back to the client
+     */
+    void (*emit_event)(Monitor *mon, QAPIEvent event, QDict *qdict);
 };
 
 struct Monitor {
monitor/monitor.c +3 -12
index 8b70380d64..6f0353d9f1 100644
@@ -346,22 +346,13 @@ static inline QEMUClockType monitor_get_event_clock(void)
 static void monitor_qapi_event_emit(QAPIEvent event, QDict *qdict)
 {
     Monitor *mon;
-    MonitorQMP *qmp_mon;
 
     trace_monitor_protocol_event_emit(event, qdict);
     QTAILQ_FOREACH(mon, &mon_list, entry) {
-        if (!monitor_is_qmp(mon)) {
-            continue;
+        MonitorClass *cls = MONITOR_GET_CLASS(mon);
+        if (cls->emit_event) {
+            cls->emit_event(mon, event, qdict);
         }
-
-        qmp_mon = container_of(mon, MonitorQMP, parent_obj);
-        {
-            QEMU_LOCK_GUARD(&mon->mon_lock);
-            if (qmp_mon->commands == &qmp_cap_negotiation_commands) {
-                continue;
-            }
-        }
-        qmp_send_response(qmp_mon, qdict);
     }
 }
 
monitor/qmp.c +20 -0
index 694d303215..ec17050a70 100644
@@ -99,11 +99,17 @@ static void monitor_qmp_set_pretty(Object *obj, bool val, Error **errp)
     mon->pretty = val;
 }
 
+static void monitor_qmp_emit_event(Monitor *mon, QAPIEvent event, QDict *qdict);
+
 static void monitor_qmp_class_init(ObjectClass *cls, const void *data)
 {
+    MonitorClass *moncls = MONITOR_CLASS(cls);
+
     object_class_property_add_bool(cls, "pretty",
                                    monitor_qmp_get_pretty,
                                    monitor_qmp_set_pretty);
+
+    moncls->emit_event = monitor_qmp_emit_event;
 }
 
 static void handle_qmp_command(void *opaque, QObject *req, Error *err);
@@ -117,6 +123,20 @@ static void monitor_qmp_init(Object *obj)
     json_message_parser_init(&mon->parser, handle_qmp_command, mon, NULL);
 }
 
+static void monitor_qmp_emit_event(Monitor *mon, QAPIEvent event, QDict *qdict)
+{
+    MonitorQMP *qmp = MONITOR_QMP(mon);
+
+    WITH_QEMU_LOCK_GUARD(&mon->mon_lock) {
+        if (qmp->commands == &qmp_cap_negotiation_commands) {
+            return;
+        }
+    }
+
+    qmp_send_response(qmp, qdict);
+}
+
+
 static bool qmp_oob_enabled(MonitorQMP *mon)
 {
     return mon->capab[QMP_CAPABILITY_OOB];
Signature · 1 line
2.54.0
#25 lore raw

PATCH v5 15/35 monitor: use dynamic cast in monitor_qmp_requests_pop_any_with_lock

In reply to Daniel P. Berrangé

This eliminates a use of monitor_is_qmp() from the QMP coroutine dispatch path.

---
 monitor/qmp.c | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)
monitor/qmp.c +3 -2
index 93f576c19d..d2a03bb7e3 100644
@@ -277,11 +277,12 @@ static QMPRequest *monitor_qmp_requests_pop_any_with_lock(void)
     MonitorQMP *qmp_mon;
 
     QTAILQ_FOREACH(mon, &mon_list, entry) {
-        if (!monitor_is_qmp(mon)) {
+        qmp_mon = MONITOR_QMP(
+            object_dynamic_cast(OBJECT(mon), TYPE_MONITOR_QMP));
+        if (!qmp_mon) {
             continue;
         }
 
-        qmp_mon = container_of(mon, MonitorQMP, parent_obj);
         qemu_mutex_lock(&qmp_mon->qmp_queue_lock);
         req_obj = g_queue_pop_head(qmp_mon->qmp_requests);
         if (req_obj) {
Signature · 1 line
2.54.0
#20 lore raw

PATCH v5 10/35 monitor: move monitor_data_(init|destroy) into QOM init/finalize

In reply to Daniel P. Berrangé

Start to take advantage of QOM, by using object init and finalize methods to replace monitor_data_init and monitor_data_destroy.

A standalone helper is provided to enable the I/O thread for QMP where appropriate for the chardev backend.

---
 monitor/hmp.c              |  6 +++--
 monitor/monitor-internal.h |  3 +--
 monitor/monitor.c          | 46 +++++++++++++-------------------------
 monitor/qmp-cmds.c         |  3 ---
 monitor/qmp.c              | 34 +++++++++++++++-------------
 5 files changed, 40 insertions(+), 52 deletions(-)
monitor/hmp.c +4 -2
index 516aa25d4c..2bb2333da5 100644
@@ -48,6 +48,10 @@ OBJECT_DEFINE_TYPE(MonitorHMP, monitor_hmp, MONITOR_HMP, MONITOR);
 
 static void monitor_hmp_finalize(Object *obj)
 {
+    MonitorHMP *mon = MONITOR_HMP(obj);
+    if (mon->rs) {
+        readline_free(mon->rs);
+    }
 }
 
 static bool monitor_hmp_get_readline(Object *obj, Error **errp)
@@ -1590,8 +1594,6 @@ void monitor_new_hmp(const char *chardev_id, bool use_readline, Error **errp)
         return;
     }
 
-    monitor_data_init(&mon->parent_obj, false, false);
-
     if (mon->use_readline) {
         mon->rs = readline_init(monitor_readline_printf,
                                 monitor_readline_flush,
monitor/monitor-internal.h +1 -2
index d34f9be139..6a4c3b0f37 100644
@@ -193,8 +193,7 @@ extern QemuMutex monitor_lock;
 extern MonitorList mon_list;
 
 void monitor_complete(Monitor *mon, Error **errp);
-void monitor_data_init(Monitor *mon, bool is_qmp, bool use_io_thread);
-void monitor_data_destroy(Monitor *mon);
+void monitor_iothread_init(Monitor *mon);
 int monitor_can_read(void *opaque);
 void monitor_list_append(Monitor *mon);
 void monitor_fdsets_cleanup(void);
monitor/monitor.c +16 -30

88 lines not on this page · raw message

monitor/qmp-cmds.c +0 -3
index 0d9adad288..6cb0b587fb 100644
@@ -168,8 +168,6 @@ char *qmp_human_monitor_command(const char *command_line, bool has_cpu_index,
     char *output = NULL;
     MonitorHMP *hmp = MONITOR_HMP(object_new(TYPE_MONITOR_HMP));
 
-    monitor_data_init(&hmp->parent_obj, false, false);
-
     if (has_cpu_index) {
         int ret = monitor_set_cpu(&hmp->parent_obj, cpu_index);
         if (ret < 0) {
@@ -186,7 +184,6 @@ char *qmp_human_monitor_command(const char *command_line, bool has_cpu_index,
     }
 
 out:
-    monitor_data_destroy(&hmp->parent_obj);
     object_unref(hmp);
     return output;
 }
monitor/qmp.c +19 -15
index e3318b57db..694d303215 100644
@@ -73,8 +73,16 @@ QmpCommandList qmp_commands, qmp_cap_negotiation_commands;
 
 OBJECT_DEFINE_TYPE(MonitorQMP, monitor_qmp, MONITOR_QMP, MONITOR);
 
+static void monitor_qmp_cleanup_req_queue_locked(MonitorQMP *mon);
+
 static void monitor_qmp_finalize(Object *obj)
 {
+    MonitorQMP *mon = MONITOR_QMP(obj);
+
+    json_message_parser_destroy(&mon->parser);
+    qemu_mutex_destroy(&mon->qmp_queue_lock);
+    monitor_qmp_cleanup_req_queue_locked(mon);
+    g_queue_free(mon->qmp_requests);
 }
 
 static bool monitor_qmp_get_pretty(Object *obj, Error **errp)
@@ -98,8 +106,15 @@ static void monitor_qmp_class_init(ObjectClass *cls, const void *data)
                                    monitor_qmp_set_pretty);
 }
 
+static void handle_qmp_command(void *opaque, QObject *req, Error *err);
 static void monitor_qmp_init(Object *obj)
 {
+    MonitorQMP *mon = MONITOR_QMP(obj);
+
+    qemu_mutex_init(&mon->qmp_queue_lock);
+    mon->qmp_requests = g_queue_new();
+
+    json_message_parser_init(&mon->parser, handle_qmp_command, mon, NULL);
 }
 
 static bool qmp_oob_enabled(MonitorQMP *mon)
@@ -522,14 +537,6 @@ static void monitor_qmp_event(void *opaque, QEMUChrEvent event)
     }
 }
 
-void monitor_data_destroy_qmp(MonitorQMP *mon)
-{
-    json_message_parser_destroy(&mon->parser);
-    qemu_mutex_destroy(&mon->qmp_queue_lock);
-    monitor_qmp_cleanup_req_queue_locked(mon);
-    g_queue_free(mon->qmp_requests);
-}
-
 static void monitor_qmp_setup_handlers_bh(void *opaque)
 {
     MonitorQMP *mon = opaque;
@@ -572,14 +579,11 @@ void monitor_new_qmp(const char *chardev_id, bool pretty, Error **errp)
     qemu_chr_fe_set_echo(&mon->parent_obj.chr, true);
 
     /* Note: we run QMP monitor in I/O thread when @chr supports that */
-    monitor_data_init(&mon->parent_obj, true,
-                      qemu_chr_has_feature(mon->parent_obj.chr.chr,
-                                           QEMU_CHAR_FEATURE_GCONTEXT));
-
-    qemu_mutex_init(&mon->qmp_queue_lock);
-    mon->qmp_requests = g_queue_new();
+    if (qemu_chr_has_feature(mon->parent_obj.chr.chr,
+                             QEMU_CHAR_FEATURE_GCONTEXT)) {
+        monitor_iothread_init(&mon->parent_obj);
+    }
 
-    json_message_parser_init(&mon->parser, handle_qmp_command, mon, NULL);
     if (mon->parent_obj.use_io_thread) {
         /*
          * Make sure the old iowatch is gone.  It's possible when
Signature · 1 line
2.54.0
#19 lore raw

PATCH v5 9/35 monitor: remove 'skip_flush' field

In reply to Daniel P. Berrangé

The 'skip_flush' field is set on the dummy throwaway HMP monitor object created by QMP's 'human-monitor-command', as an indication not to try to write data to the chardev. Instead the QMP command impl will grab the data straight out of the in-memory buffer.

The flag is redundant, however, as the monitor code could instead simply check the 'fe_is_open' field on the CharFrontend, which will be false in the same scenarios that 'skip_flush' is true.

---
 monitor/hmp.c              |  2 +-
 monitor/monitor-internal.h |  4 +---
 monitor/monitor.c          | 11 +++++++----
 monitor/qmp-cmds.c         |  2 +-
 monitor/qmp.c              |  2 +-
 5 files changed, 11 insertions(+), 10 deletions(-)
monitor/hmp.c +1 -1
index d5905e2279..516aa25d4c 100644
@@ -1590,7 +1590,7 @@ void monitor_new_hmp(const char *chardev_id, bool use_readline, Error **errp)
         return;
     }
 
-    monitor_data_init(&mon->parent_obj, false, false, false);
+    monitor_data_init(&mon->parent_obj, false, false);
 
     if (mon->use_readline) {
         mon->rs = readline_init(monitor_readline_printf,
monitor/monitor-internal.h +1 -3
index 145d52fd71..d34f9be139 100644
@@ -112,7 +112,6 @@ struct Monitor {
     CharFrontend chr;
     int suspend_cnt;            /* Needs to be accessed atomically */
     bool is_qmp;
-    bool skip_flush;
     bool use_io_thread;
 
     char *mon_cpu_path;
@@ -194,8 +193,7 @@ extern QemuMutex monitor_lock;
 extern MonitorList mon_list;
 
 void monitor_complete(Monitor *mon, Error **errp);
-void monitor_data_init(Monitor *mon, bool is_qmp, bool skip_flush,
-                       bool use_io_thread);
+void monitor_data_init(Monitor *mon, bool is_qmp, bool use_io_thread);
 void monitor_data_destroy(Monitor *mon);
 int monitor_can_read(void *opaque);
 void monitor_list_append(Monitor *mon);
monitor/monitor.c +7 -4
index 747ca27fd5..e3e5623a4b 100644
@@ -188,7 +188,12 @@ void monitor_flush_locked(Monitor *mon)
     size_t len;
     const char *buf;
 
-    if (mon->skip_flush) {
+    /*
+     * When used by QMP human-monitor-command, no chardev
+     * will be connected, as we want to just collect the
+     * output in the buffer
+     */
+    if (!mon->chr.fe_is_open) {
         return;
     }
 
@@ -642,8 +647,7 @@ static void monitor_iothread_init(void)
     mon_iothread = iothread_create("mon_iothread", &error_abort);
 }
 
-void monitor_data_init(Monitor *mon, bool is_qmp, bool skip_flush,
-                       bool use_io_thread)
+void monitor_data_init(Monitor *mon, bool is_qmp, bool use_io_thread)
 {
     if (use_io_thread && !mon_iothread) {
         monitor_iothread_init();
@@ -651,7 +655,6 @@ void monitor_data_init(Monitor *mon, bool is_qmp, bool skip_flush,
     qemu_mutex_init(&mon->mon_lock);
     mon->is_qmp = is_qmp;
     mon->outbuf = g_string_new(NULL);
-    mon->skip_flush = skip_flush;
     mon->use_io_thread = use_io_thread;
 }
 
monitor/qmp-cmds.c +1 -1
index bfde769ef0..0d9adad288 100644
@@ -168,7 +168,7 @@ char *qmp_human_monitor_command(const char *command_line, bool has_cpu_index,
     char *output = NULL;
     MonitorHMP *hmp = MONITOR_HMP(object_new(TYPE_MONITOR_HMP));
 
-    monitor_data_init(&hmp->parent_obj, false, true, false);
+    monitor_data_init(&hmp->parent_obj, false, false);
 
     if (has_cpu_index) {
         int ret = monitor_set_cpu(&hmp->parent_obj, cpu_index);
monitor/qmp.c +1 -1
index 2a8532ef96..e3318b57db 100644
@@ -572,7 +572,7 @@ void monitor_new_qmp(const char *chardev_id, bool pretty, Error **errp)
     qemu_chr_fe_set_echo(&mon->parent_obj.chr, true);
 
     /* Note: we run QMP monitor in I/O thread when @chr supports that */
-    monitor_data_init(&mon->parent_obj, true, false,
+    monitor_data_init(&mon->parent_obj, true,
                       qemu_chr_has_feature(mon->parent_obj.chr.chr,
                                            QEMU_CHAR_FEATURE_GCONTEXT));
 
Signature · 1 line
2.54.0
#23 lore raw

PATCH v5 13/35 monitor: use class methods for monitor_accept_input

In reply to Daniel P. Berrangé

This removes the need for using monitor_is_qmp() to check the subclass type, which is an anti-pattern.

---
 monitor/hmp.c              | 16 ++++++++++++++++
 monitor/monitor-internal.h |  5 +++++
 monitor/monitor.c          | 12 +++---------
 3 files changed, 24 insertions(+), 9 deletions(-)
monitor/hmp.c +16 -0
index 08055839a4..1da9370ead 100644
@@ -70,6 +70,7 @@ static void monitor_hmp_set_readline(Object *obj, bool val, Error **errp)
 
 int monitor_hmp_vprintf(Monitor *mon, const char *fmt, va_list ap)
     G_GNUC_PRINTF(2, 0);
+static void monitor_hmp_accept_input(Monitor *mon);
 
 static void monitor_hmp_class_init(ObjectClass *cls, const void *data)
 {
@@ -80,6 +81,7 @@ static void monitor_hmp_class_init(ObjectClass *cls, const void *data)
                                    monitor_hmp_set_readline);
 
     moncls->vprintf = monitor_hmp_vprintf;
+    moncls->accept_input = monitor_hmp_accept_input;
 }
 
 static void monitor_hmp_init(Object *obj)
@@ -100,6 +102,20 @@ int monitor_hmp_vprintf(Monitor *mon, const char *fmt, va_list ap)
     return monitor_puts(mon, buf);
 }
 
+static void monitor_hmp_accept_input(Monitor *mon)
+{
+    qemu_mutex_lock(&mon->mon_lock);
+    if (mon->reset_seen) {
+        MonitorHMP *hmp = MONITOR_HMP(mon);
+        assert(hmp->rs);
+        readline_restart(hmp->rs);
+        qemu_mutex_unlock(&mon->mon_lock);
+        readline_show_prompt(hmp->rs);
+    } else {
+        qemu_mutex_unlock(&mon->mon_lock);
+    }
+}
+
 static void monitor_command_cb(void *opaque, const char *cmdline,
                                void *readline_opaque)
 {
monitor/monitor-internal.h +5 -0
index 763b5c9625..592f146331 100644
@@ -116,6 +116,11 @@ struct MonitorClass {
      * notifications back to the client
      */
     void (*emit_event)(Monitor *mon, QAPIEvent event, QDict *qdict);
+    /*
+     * If non-NULL, perform any actions needed to prepare
+     * the monitor to accept further client input
+     */
+    void (*accept_input)(Monitor *mon);
 };
 
 struct Monitor {
monitor/monitor.c +3 -9
index 6f0353d9f1..2f16cd3053 100644
@@ -576,16 +576,10 @@ int monitor_suspend(Monitor *mon)
 static void monitor_accept_input(void *opaque)
 {
     Monitor *mon = opaque;
+    MonitorClass *cls = MONITOR_GET_CLASS(mon);
 
-    qemu_mutex_lock(&mon->mon_lock);
-    if (!monitor_is_qmp(mon) && mon->reset_seen) {
-        MonitorHMP *hmp_mon = container_of(mon, MonitorHMP, parent_obj);
-        assert(hmp_mon->rs);
-        readline_restart(hmp_mon->rs);
-        qemu_mutex_unlock(&mon->mon_lock);
-        readline_show_prompt(hmp_mon->rs);
-    } else {
-        qemu_mutex_unlock(&mon->mon_lock);
+    if (cls->accept_input) {
+        cls->accept_input(mon);
     }
 
     qemu_chr_fe_accept_input(&mon->chr);
Signature · 1 line
2.54.0
#17 lore raw

PATCH v5 7/35 monitor: add 'readline' property to HMP Monitor class

In reply to Daniel P. Berrangé

This determines whether a human monitor runs with readline for interactive use, or without readline for non-interactive use by the GDB stub.

---
 monitor/hmp.c | 27 ++++++++++++++++++++++++++-
 1 file changed, 26 insertions(+), 1 deletion(-)
monitor/hmp.c +26 -1
index 4ab81b5904..d5905e2279 100644
@@ -50,12 +50,37 @@ static void monitor_hmp_finalize(Object *obj)
 {
 }
 
+static bool monitor_hmp_get_readline(Object *obj, Error **errp)
+{
+    MonitorHMP *mon = MONITOR_HMP(obj);
+
+    return mon->use_readline;
+}
+
+static void monitor_hmp_set_readline(Object *obj, bool val, Error **errp)
+{
+    MonitorHMP *mon = MONITOR_HMP(obj);
+
+    mon->use_readline = val;
+}
+
 static void monitor_hmp_class_init(ObjectClass *cls, const void *data)
 {
+    object_class_property_add_bool(cls, "readline",
+                                   monitor_hmp_get_readline,
+                                   monitor_hmp_set_readline);
 }
 
 static void monitor_hmp_init(Object *obj)
 {
+    MonitorHMP *hmp = MONITOR_HMP(obj);
+
+    /*
+     * Default to common case for external HMP use,
+     * as opposed to non-interactive internal use
+     * from gdbstub
+     */
+    hmp->use_readline = true;
 }
 
 static void monitor_command_cb(void *opaque, const char *cmdline,
@@ -1550,6 +1575,7 @@ void monitor_new_hmp(const char *chardev_id, bool use_readline, Error **errp)
                                         id,
                                         errp,
                                         "chardev", chardev_id,
+                                        "readline", use_readline ? "yes" : "no",
                                         NULL);
 
     if (!obj) {
@@ -1566,7 +1592,6 @@ void monitor_new_hmp(const char *chardev_id, bool use_readline, Error **errp)
 
     monitor_data_init(&mon->parent_obj, false, false, false);
 
-    mon->use_readline = use_readline;
     if (mon->use_readline) {
         mon->rs = readline_init(monitor_readline_printf,
                                 monitor_readline_flush,
Signature · 1 line
2.54.0
#16 lore raw

PATCH v5 6/35 monitor: add 'chardev' property to Monitor base class

In reply to Daniel P. Berrangé

This is associates both QMP and HMP monitors with a character device backend.

---
 chardev/char.c             |  3 ++-
 gdbstub/system.c           |  4 ++--
 include/monitor/monitor.h  |  4 ++--
 monitor/hmp.c              | 10 ++++++--
 monitor/monitor-internal.h |  2 ++
 monitor/monitor.c          | 47 ++++++++++++++++++++++++++++++--------
 monitor/qmp.c              | 18 ++++++++++-----
 stubs/monitor-internal.c   |  2 +-
 8 files changed, 67 insertions(+), 23 deletions(-)
chardev/char.c +2 -1
index 813c04d953..981fbb1aaa 100644
@@ -804,8 +804,9 @@ static Chardev *qemu_chr_new_from_name(const char *label, const char *filename,
     }
 
     if (qemu_opt_get_bool(opts, "mux", 0)) {
+        const char *chardev_id = qemu_opts_id(opts);
         assert(permit_mux_mon);
-        monitor_new_hmp(chr, true, &err);
+        monitor_new_hmp(chardev_id, true, &err);
         if (err) {
             error_report_err(err);
             object_unparent(OBJECT(chr));
gdbstub/system.c +2 -2
index 18f6a62b2e..80175e7fa2 100644
@@ -387,9 +387,9 @@ bool gdbserver_start(const char *device, Error **errp)
         qemu_add_vm_change_state_handler(gdb_vm_state_change, NULL);
 
         /* Initialize a monitor terminal for gdb */
-        mon_chr = qemu_chardev_new(NULL, TYPE_CHARDEV_GDB,
+        mon_chr = qemu_chardev_new("gdbchrdev0", TYPE_CHARDEV_GDB,
                                    NULL, NULL, &error_abort);
-        monitor_new_hmp(mon_chr, false, &error_abort);
+        monitor_new_hmp("gdbchrdev0", false, &error_abort);
     } else {
         qemu_chr_fe_deinit(&gdbserver_system_state.chr, true);
         mon_chr = gdbserver_system_state.mon_chr;
include/monitor/monitor.h +2 -2
index 2e9f9e12e9..a135b3a590 100644
@@ -28,8 +28,8 @@ bool monitor_cur_is_qmp(void);
 
 void monitor_init_globals(void);
 void monitor_init_globals_core(void);
-void monitor_new_qmp(Chardev *chr, bool pretty, Error **errp);
-void monitor_new_hmp(Chardev *chr, bool use_readline, Error **errp);
+void monitor_new_qmp(const char *chardev_id, bool pretty, Error **errp);
+void monitor_new_hmp(const char *chardev_id, bool use_readline, Error **errp);
 int monitor_new(MonitorOptions *opts, bool allow_hmp, Error **errp);
 int monitor_new_opts(QemuOpts *opts, Error **errp);
 void monitor_cleanup(void);
monitor/hmp.c +8 -2
index 81047d2513..4ab81b5904 100644
@@ -23,6 +23,7 @@
  */
 
 #include "qemu/osdep.h"
+#include "qapi/error.h"
 #include <dirent.h>
 #include "hw/core/qdev.h"
 #include "hw/core/sysemu-cpu-ops.h"
@@ -1538,8 +1539,9 @@ static void monitor_readline_flush(void *opaque)
     monitor_flush(&mon->parent_obj);
 }
 
-void monitor_new_hmp(Chardev *chr, bool use_readline, Error **errp)
+void monitor_new_hmp(const char *chardev_id, bool use_readline, Error **errp)
 {
+    ERRP_GUARD();
     MonitorHMP *mon;
     static int counter;
     g_autofree char *id = g_strdup_printf("hmpcompat%d", counter++);
@@ -1547,13 +1549,17 @@ void monitor_new_hmp(Chardev *chr, bool use_readline, Error **errp)
                                         object_get_objects_root(),
                                         id,
                                         errp,
+                                        "chardev", chardev_id,
                                         NULL);
+
     if (!obj) {
         return;
     }
+
     mon = MONITOR_HMP(obj);
 
-    if (!qemu_chr_fe_init(&mon->parent_obj.chr, chr, errp)) {
+    monitor_complete(MONITOR(mon), errp);
+    if (*errp) {
         object_unparent(OBJECT(mon));
         return;
     }
monitor/monitor-internal.h +2 -0
index 05c1f2f5e0..145d52fd71 100644
@@ -108,6 +108,7 @@ struct MonitorClass {
 
 struct Monitor {
     Object parent;
+    char *chardev_id;
     CharFrontend chr;
     int suspend_cnt;            /* Needs to be accessed atomically */
     bool is_qmp;
@@ -192,6 +193,7 @@ extern QmpCommandList qmp_commands, qmp_cap_negotiation_commands;
 extern QemuMutex monitor_lock;
 extern MonitorList mon_list;
 
+void monitor_complete(Monitor *mon, Error **errp);
 void monitor_data_init(Monitor *mon, bool is_qmp, bool skip_flush,
                        bool use_io_thread);
 void monitor_data_destroy(Monitor *mon);
monitor/monitor.c +38 -9

78 lines not on this page · raw message

monitor/qmp.c +12 -6
index 5231ed506a..0abf2677f9 100644
@@ -527,8 +527,9 @@ static void monitor_qmp_setup_handlers_bh(void *opaque)
     monitor_list_append(&mon->parent_obj);
 }
 
-void monitor_new_qmp(Chardev *chr, bool pretty, Error **errp)
+void monitor_new_qmp(const char *chardev_id, bool pretty, Error **errp)
 {
+    ERRP_GUARD();
     MonitorQMP *mon;
     static int counter;
     g_autofree char *id = g_strdup_printf("qmpcompat%d", counter++);
@@ -536,21 +537,26 @@ void monitor_new_qmp(Chardev *chr, bool pretty, Error **errp)
                                         object_get_objects_root(),
                                         id,
                                         errp,
+                                        "chardev", chardev_id,
                                         NULL);
+
     if (!obj) {
         return;
     }
-    mon = MONITOR_QMP(obj);
 
-    if (!qemu_chr_fe_init(&mon->parent_obj.chr, chr, errp)) {
+    mon = MONITOR_QMP(obj);
+    monitor_complete(MONITOR(mon), errp);
+    if (*errp) {
         object_unparent(OBJECT(mon));
         return;
     }
+
     qemu_chr_fe_set_echo(&mon->parent_obj.chr, true);
 
     /* Note: we run QMP monitor in I/O thread when @chr supports that */
     monitor_data_init(&mon->parent_obj, true, false,
-                      qemu_chr_has_feature(chr, QEMU_CHAR_FEATURE_GCONTEXT));
+                      qemu_chr_has_feature(mon->parent_obj.chr.chr,
+                                           QEMU_CHAR_FEATURE_GCONTEXT));
 
     mon->pretty = pretty;
 
@@ -563,12 +569,12 @@ void monitor_new_qmp(Chardev *chr, bool pretty, Error **errp)
          * Make sure the old iowatch is gone.  It's possible when
          * e.g. the chardev is in client mode, with wait=on.
          */
-        remove_fd_in_watch(chr);
+        remove_fd_in_watch(mon->parent_obj.chr.chr);
         /*
          * Clean up listener IO sources early to prevent racy fd
          * handling between the main thread and the I/O thread.
          */
-        remove_listener_fd_in_watch(chr);
+        remove_listener_fd_in_watch(mon->parent_obj.chr.chr);
         /*
          * We can't call qemu_chr_fe_set_handlers() directly here
          * since chardev might be running in the monitor I/O
stubs/monitor-internal.c +1 -1
index 23d58da184..51db7588b9 100644
@@ -8,6 +8,6 @@ int monitor_get_fd(Monitor *mon, const char *name, Error **errp)
     return -1;
 }
 
-void monitor_new_hmp(Chardev *chr, bool use_readline, Error **errp)
+void monitor_new_hmp(const char *chardev_id, bool use_readline, Error **errp)
 {
 }
Signature · 1 line
2.54.0
#18 lore raw

PATCH v5 8/35 monitor: add 'pretty' property to QMP Monitor class

In reply to Daniel P. Berrangé

This determines whether the QMP JSON responses are pretty printed with newlines and indentation, or compact with no extra whitespace.

---
 monitor/qmp.c | 20 ++++++++++++++++++--
 1 file changed, 18 insertions(+), 2 deletions(-)
monitor/qmp.c +18 -2
index 0abf2677f9..2a8532ef96 100644
@@ -77,8 +77,25 @@ static void monitor_qmp_finalize(Object *obj)
 {
 }
 
+static bool monitor_qmp_get_pretty(Object *obj, Error **errp)
+{
+    MonitorQMP *mon = MONITOR_QMP(obj);
+
+    return mon->pretty;
+}
+
+static void monitor_qmp_set_pretty(Object *obj, bool val, Error **errp)
+{
+    MonitorQMP *mon = MONITOR_QMP(obj);
+
+    mon->pretty = val;
+}
+
 static void monitor_qmp_class_init(ObjectClass *cls, const void *data)
 {
+    object_class_property_add_bool(cls, "pretty",
+                                   monitor_qmp_get_pretty,
+                                   monitor_qmp_set_pretty);
 }
 
 static void monitor_qmp_init(Object *obj)
@@ -538,6 +555,7 @@ void monitor_new_qmp(const char *chardev_id, bool pretty, Error **errp)
                                         id,
                                         errp,
                                         "chardev", chardev_id,
+                                        "pretty", pretty ? "yes" : "no",
                                         NULL);
 
     if (!obj) {
@@ -558,8 +576,6 @@ void monitor_new_qmp(const char *chardev_id, bool pretty, Error **errp)
                       qemu_chr_has_feature(mon->parent_obj.chr.chr,
                                            QEMU_CHAR_FEATURE_GCONTEXT));
 
-    mon->pretty = pretty;
-
     qemu_mutex_init(&mon->qmp_queue_lock);
     mon->qmp_requests = g_queue_new();
 
Signature · 1 line
2.54.0
#30 lore raw

PATCH v5 20/35 monitor: drop unused monitor_is_qmp method

In reply to Daniel P. Berrangé

The previous patch dropped the only remaining use of monitor_is_qmp.

---
 monitor/monitor-internal.h | 9 ---------
 monitor/monitor.c          | 1 -
 2 files changed, 10 deletions(-)
monitor/monitor-internal.h +0 -9
index efd21b8912..fe5703af6d 100644
@@ -134,7 +134,6 @@ struct Monitor {
     char *chardev_id;
     CharFrontend chr;
     int suspend_cnt;            /* Needs to be accessed atomically */
-    bool is_qmp;
 
     char *mon_cpu_path;
     QTAILQ_ENTRY(Monitor) entry;
@@ -197,14 +196,6 @@ struct MonitorQMP {
     GQueue *qmp_requests;
 };
 
-/**
- * Is @mon a QMP monitor?
- */
-static inline bool monitor_is_qmp(const Monitor *mon)
-{
-    return mon->is_qmp;
-}
-
 typedef QTAILQ_HEAD(MonitorList, Monitor) MonitorList;
 extern IOThread *mon_iothread;
 extern Coroutine *qmp_dispatcher_co;
monitor/monitor.c +0 -1
index 15e49d78d7..6ed586fc4f 100644
@@ -113,7 +113,6 @@ static void monitor_init(Object *obj)
     Monitor *mon = MONITOR(obj);
 
     qemu_mutex_init(&mon->mon_lock);
-    mon->is_qmp = !!object_dynamic_cast(obj, TYPE_MONITOR_QMP);
     mon->outbuf = g_string_new(NULL);
 }
 
Signature · 1 line
2.54.0
#35 lore raw

PATCH v5 24/35 monitor: reject attempts to delete the current monitor

In reply to Daniel P. Berrangé

From: Christian Brauner <[email protected]>

If an 'object_del' command for a QMP monitor arrives targetting the current monitor, reject this request. If the current monitor is deleted, it will be impossible to send any reply and the client won't be able to remove the corresponding chardev backend.

Note, it is not possible to rely on checking monitor_cur() because if 'object_del' is called via human-monitor-command, monitor_cur() will reflect the temporary HMP, not the QMP target that needs to be checked.

[DB: extracted monitor tracking from larger commit; added logic
to monitor_qmp_prepare_delete to reject request]

---
 monitor/qmp.c | 17 +++++++++++++++++
 1 file changed, 17 insertions(+)
monitor/qmp.c +17 -0
index 0e1c3eb52b..b0535dec68 100644
@@ -72,6 +72,9 @@ typedef struct QMPRequest QMPRequest;
 
 QmpCommandList qmp_commands, qmp_cap_negotiation_commands;
 
+/* Monitor being serviced by the dispatcher.  Protected by BQL. */
+static MonitorQMP *qmp_dispatcher_current_mon;
+
 OBJECT_DEFINE_TYPE(MonitorQMP, monitor_qmp, MONITOR_QMP, MONITOR);
 
 static void monitor_qmp_cleanup_req_queue_locked(MonitorQMP *mon);
@@ -369,6 +372,7 @@ void coroutine_fn monitor_qmp_dispatcher_co(void *data)
          */
 
         mon = req_obj->mon;
+        qmp_dispatcher_current_mon = mon;
 
         /*
          * We need to resume the monitor if handle_qmp_command()
@@ -429,6 +433,7 @@ void coroutine_fn monitor_qmp_dispatcher_co(void *data)
         }
 
         qmp_request_free(req_obj);
+        qmp_dispatcher_current_mon = NULL;
     }
     qatomic_set(&qmp_dispatcher_co, NULL);
 }
@@ -573,6 +578,11 @@ static void monitor_qmp_event(void *opaque, QEMUChrEvent event)
     }
 }
 
+static bool monitor_qmp_dispatcher_is_servicing(MonitorQMP *mon)
+{
+    return qmp_dispatcher_current_mon == mon;
+}
+
 static void monitor_qmp_setup_handlers_bh(void *opaque)
 {
     MonitorQMP *mon = opaque;
@@ -646,6 +656,13 @@ static void monitor_qmp_complete(UserCreatable *uc, Error **errp)
 
 static bool monitor_qmp_prepare_delete(UserCreatable *uc, Error **errp)
 {
+    MonitorQMP *qmp = MONITOR_QMP(uc);
+
+    if (monitor_qmp_dispatcher_is_servicing(qmp)) {
+        error_setg(errp, "Cannot delete the current QMP monitor");
+        return false;
+    }
+
     error_setg(errp, "Deleting QMP monitors is not supported");
     return false;
 }
Signature · 1 line
2.54.0
#45 lore raw

PATCH v5 32/35 monitor: add support for auto-deleting monitors upon close

In reply to Daniel P. Berrangé

The default monitor is usually a long lived object that will exist for the entire lifetime of the VM. A monitor can only service a single client at a time though, and so it might be desirable to hotplug additional monitors at runtime for specific tasks. If doing that, however, there is a need to remove the monitor when it is no longer needed.

Allowing a client to run "object-del" against its own monitor adds complex edge cases, as it would be desirable to send the QMP response despite the monitor sending it being deleted. Doing "object-del" alone will also result in orphaning a character device backend instance, as there is no opportunity to run the companion "chardev-del" command.

A simpler way to ensure cleanup is to add the concept of auto-deleting monitor objects. Specifically when the "CHR_EVENT_CLOSED" event is emitted, the equivalent of "object-del" + "chardev-del" can be run internally. Since the transient client has already droppped its monitor connection, there is no synchronization to be concerned about.

This is implemented via a new "close-action=none|delete" property on the 'monitor-qmp' object. This concept could be extended with further actions in future, for example:

* close-action=shutdown - graceful guest shutdown
* close-action=terminate - immediate guest poweroff
* close-action=stop - pause guest CPUs while the monitor is not connected to any client

This is left as an exercise for future interested contributors.

---
 monitor/monitor-internal.h                    |  3 +
 monitor/qmp.c                                 | 75 +++++++++++++++++
 qapi/qom.json                                 | 21 ++++-
 .../generic/test_monitor_hotplug.py           | 83 ++++++++++++++++++-
 4 files changed, 177 insertions(+), 5 deletions(-)
monitor/monitor-internal.h +3 -0
index 5522e05464..23829f32f9 100644
@@ -28,6 +28,7 @@
 #include "chardev/char-fe.h"
 #include "monitor/monitor.h"
 #include "qapi/qapi-types-control.h"
+#include "qapi/qapi-types-qom.h"
 #include "qapi/qmp-registry.h"
 #include "qobject/json-parser.h"
 #include "qemu/readline.h"
@@ -178,7 +179,9 @@ struct MonitorQMP {
     Monitor parent_obj;
     JSONMessageParser parser;
     bool pretty;
+    MonitorQMPCloseAction close_action;
     bool setup_pending; /* iothread BH has not yet set up chardev handlers */
+    bool delete_pending; /* close_action has started 'delete' process */
     /*
      * When a client connects, we're in capabilities negotiation mode.
      * @commands is &qmp_cap_negotiation_commands then.  When command
monitor/qmp.c +75 -0

110 lines not on this page · raw message

qapi/qom.json +20 -1
index 6affb70a59..b4c49a2e8b 100644
@@ -1213,18 +1213,37 @@
   'base': 'MonitorProperties',
   'data': { '*readline': 'bool' } }
 
+
+##
+# @MonitorQMPCloseAction:
+#
+# Action to take when the character device backend is
+# closed.
+#
+# @none: take no action (the default)
+# @delete: delete both the 'monitor-qmp' object and its associated
+#          character device backend object
+#
+# Since 11.1
+##
+{ 'enum' : 'MonitorQMPCloseAction',
+  'data': ['none', 'delete'] }
+
 ##
 # @MonitorQMPProperties:
 #
 # Properties for the QMP monitor
 #
 # @pretty: whether to pretty print JSON responses (default: disabled)
+# @close-action: action to take when the character device backend
+#                is closed (default: none)
 #
 # Since: 11.1
 ##
 { 'struct': 'MonitorQMPProperties',
   'base': 'MonitorProperties',
-  'data': { '*pretty': 'bool' } }
+  'data': { '*pretty': 'bool',
+            '*close-action': 'MonitorQMPCloseAction' } }
 
 ##
 # @ObjectType:
tests/functional/generic/test_monitor_hotplug.py +79 -4

101 lines not on this page · raw message

Signature · 1 line
2.54.0
#39 lore raw

PATCH v5 28/35 tests/functional: add e2e test for dynamic QMP monitor hotplug

In reply to Daniel P. Berrangé

From: Christian Brauner <[email protected]>

Add functional tests that exercise dynamic monitor hotplug with real socket connections:

- Hotplug cycle: chardev-add a unix socket, object-add, connect to the socket, receive the QMP greeting, negotiate capabilities, send query-version, disconnect, remove the monitor and chardev, then repeat the entire cycle a second time to verify cleanup and reuse.

- Self-removal: a dynamically-added monitor sends object-del targeting itself, verifying that the request is rejected

- Large response: send query-qmp-schema on a dynamic monitor to exercise the output buffer flush path with a large response payload.

- Events after negotiation: trigger STOP/RESUME events via the main monitor and verify they are delivered on the dynamic monitor.

This complements the qtest unit tests by verifying that a real QMP client can connect to a dynamically-added monitor and exchange messages.

[DB: modified to use object-add/object-del; adjust self-removal test
to validate rejection of request]

---
 MAINTAINERS                                   |   1 +
 tests/functional/generic/meson.build          |   1 +
 .../generic/test_monitor_hotplug.py           | 168 ++++++++++++++++++
 3 files changed, 170 insertions(+)
 create mode 100755 tests/functional/generic/test_monitor_hotplug.py
MAINTAINERS +1 -0
index 93df53d87f..c74ffe56ae 100644
@@ -3616,6 +3616,7 @@ F: docs/interop/*qmp-*
 F: scripts/qmp/
 F: tests/qtest/qmp-test.c
 F: tests/qtest/qmp-cmd-test.c
+F: tests/functional/generic/test_monitor_hotplug.py
 T: git https://repo.or.cz/qemu/armbru.git qapi-next
 
 qtest
tests/functional/generic/meson.build +1 -0
index 09763c5d22..c94105c62e 100644
@@ -4,6 +4,7 @@ tests_generic_system = [
   'empty_cpu_model',
   'info_usernet',
   'linters',
+  'monitor_hotplug',
   'version',
   'vnc',
 ]
tests/functional/generic/test_monitor_hotplug.py new file +168 -0

168 lines not on this page · raw message

Signature · 1 line
2.54.0
#41 lore raw

PATCH v5 30/35 qom: add method for getting the "id" of a QOM object

In reply to Daniel P. Berrangé

The 'user_creatable_del' method doesn't want the 'Object *', but instead the "id" of the object to be deleted. While most of its work could be done with the "Object *" alone, purging the QemuOpts values needs the "id".

The "id" is not recorded against an Object though, it is indirectly stored as a property linking the parent and child objects together. Add an accessor that can fetch this property name.

---
 include/qom/object.h | 10 ++++++++++
 qom/object.c         | 17 +++++++++++++++++
 2 files changed, 27 insertions(+)
include/qom/object.h +10 -0
index 11f55613fc..216d2a30f6 100644
@@ -1759,6 +1759,16 @@ ObjectProperty *object_property_try_add_child(Object *obj, const char *name,
 ObjectProperty *object_property_add_child(Object *obj, const char *name,
                                           Object *child);
 
+/**
+ * object_property_get_child_name:
+ * @obj: the object that owns the property
+ * @child: the object referenced by the child property
+ *
+ * Return the property name against which @child is registered
+ * with @obj, or NULL if @obj is not a parent of @child.
+ */
+char *object_property_get_child_name(Object *obj, Object *child);
+
 typedef enum {
     /* Unref the link pointer when the property is deleted */
     OBJ_PROP_LINK_STRONG = 0x1,
qom/object.c +17 -0
index f79b2cf361..66bbd1488b 100644
@@ -1952,6 +1952,23 @@ object_property_add_child(Object *obj, const char *name,
     return object_property_try_add_child(obj, name, child, &error_abort);
 }
 
+char *object_property_get_child_name(Object *obj, Object *child)
+{
+    ObjectProperty *prop;
+    GHashTableIter iter;
+    gpointer key, value;
+
+    g_hash_table_iter_init(&iter, obj->properties);
+    while (g_hash_table_iter_next(&iter, &key, &value)) {
+        prop = value;
+        if (object_property_is_child(prop) && prop->opaque == child) {
+            return g_strdup(prop->name);
+        }
+    }
+    return NULL;
+}
+
+
 void object_property_allow_set_link(const Object *obj, const char *name,
                                     Object *val, Error **errp)
 {
Signature · 1 line
2.54.0
#31 lore raw

PATCH v5 21/35 monitor: eliminate monitor_is_hmp_non_interactive method

In reply to Daniel P. Berrangé

The monitor_is_hmp_non_interactive method is used by monitor_suspend and monitor_resume, to make them a no-op if the HMP does not use readline.

There are only a handful of callers of suspend/resume and they can be made to skip the call when readline is not present.

---
 include/monitor/monitor.h      |  2 +-
 migration/migration-hmp-cmds.c |  5 ++++-
 monitor/hmp-cmds.c             |  5 ++++-
 monitor/hmp.c                  | 11 ++++++++---
 monitor/monitor.c              | 30 +-----------------------------
 5 files changed, 18 insertions(+), 35 deletions(-)
include/monitor/monitor.h +1 -1
index 3f730cffee..ffda343585 100644
@@ -34,7 +34,7 @@ int monitor_new(MonitorOptions *opts, bool allow_hmp, Error **errp);
 int monitor_new_opts(QemuOpts *opts, Error **errp);
 void monitor_cleanup(void);
 
-int monitor_suspend(Monitor *mon);
+void monitor_suspend(Monitor *mon);
 void monitor_resume(Monitor *mon);
 
 int monitor_get_fd(Monitor *mon, const char *fdname, Error **errp);
migration/migration-hmp-cmds.c +4 -1
index 8b385f560e..502ca704da 100644
@@ -19,6 +19,7 @@
 #include "monitor/hmp.h"
 #include "monitor/hmp-completion.h"
 #include "monitor/monitor.h"
+#include "monitor/monitor-internal.h"
 #include "qapi/error.h"
 #include "qapi/qapi-commands-migration.h"
 #include "qapi/qapi-visit-migration.h"
@@ -853,12 +854,14 @@ void hmp_migrate(Monitor *mon, const QDict *qdict)
 
     if (!detach) {
         HMPMigrationStatus *status;
+        MonitorHMP *hmp = MONITOR_HMP(mon);
 
-        if (monitor_suspend(mon) < 0) {
+        if (!hmp->use_readline) {
             monitor_printf(mon, "terminal does not allow synchronous "
                            "migration, continuing detached\n");
             return;
         }
+        monitor_suspend(mon);
 
         status = g_malloc0(sizeof(*status));
         status->mon = mon;
monitor/hmp-cmds.c +4 -1
index e139caeba9..e9fb8d827a 100644
@@ -129,7 +129,10 @@ void hmp_info_version(Monitor *mon, const QDict *qdict)
 
 void hmp_quit(Monitor *mon, const QDict *qdict)
 {
-    monitor_suspend(mon);
+    MonitorHMP *hmp = MONITOR_HMP(mon);
+    if (hmp->use_readline) {
+        monitor_suspend(mon);
+    }
     qmp_quit(NULL);
 }
 
monitor/hmp.c +8 -3
index 1da9370ead..b8cccdcde3 100644
@@ -1528,13 +1528,16 @@ static void monitor_read(void *opaque, const uint8_t *buf, int size)
 static void monitor_event(void *opaque, QEMUChrEvent event)
 {
     Monitor *mon = opaque;
+    MonitorHMP *hmp = MONITOR_HMP(mon);
 
     switch (event) {
     case CHR_EVENT_MUX_IN:
         qemu_mutex_lock(&mon->mon_lock);
         if (mon->mux_out) {
             mon->mux_out = 0;
-            monitor_resume(mon);
+            if (hmp->use_readline) {
+                monitor_resume(mon);
+            }
         }
         qemu_mutex_unlock(&mon->mon_lock);
         break;
@@ -1547,7 +1550,9 @@ static void monitor_event(void *opaque, QEMUChrEvent event)
             } else {
                 monitor_flush_locked(mon);
             }
-            monitor_suspend(mon);
+            if (hmp->use_readline) {
+                monitor_suspend(mon);
+            }
             mon->mux_out = 1;
         }
         qemu_mutex_unlock(&mon->mon_lock);
@@ -1558,7 +1563,7 @@ static void monitor_event(void *opaque, QEMUChrEvent event)
                        "information\n", QEMU_VERSION);
         qemu_mutex_lock(&mon->mon_lock);
         mon->reset_seen = 1;
-        if (!mon->mux_out) {
+        if (!mon->mux_out && hmp->use_readline) {
             /* Suspend-resume forces the prompt to be printed.  */
             monitor_suspend(mon);
             monitor_resume(mon);
monitor/monitor.c +1 -29
index 6ed586fc4f..f88364a574 100644
@@ -155,25 +155,6 @@ bool monitor_requires_iothread(const Monitor *mon)
     return cls->requires_iothread && cls->requires_iothread(mon);
 }
 
-/**
- * Is @mon is using readline?
- * Note: not all HMP monitors use readline, e.g., gdbserver has a
- * non-interactive HMP monitor, so readline is not used there.
- */
-static inline bool monitor_uses_readline(const MonitorHMP *mon)
-{
-    return mon->use_readline;
-}
-
-static inline bool monitor_is_hmp_non_interactive(const Monitor *mon)
-{
-    if (!object_dynamic_cast(OBJECT(mon), TYPE_MONITOR_HMP)) {
-        return false;
-    }
-
-    return !monitor_uses_readline(container_of(mon, MonitorHMP, parent_obj));
-}
-
 static gboolean monitor_unblocked(void *do_not_use, GIOCondition cond,
                                   void *opaque)
 {
@@ -548,12 +529,8 @@ static gboolean qapi_event_throttle_equal(const void *a, const void *b)
     return TRUE;
 }
 
-int monitor_suspend(Monitor *mon)
+void monitor_suspend(Monitor *mon)
 {
-    if (monitor_is_hmp_non_interactive(mon)) {
-        return -ENOTTY;
-    }
-
     qatomic_inc(&mon->suspend_cnt);
 
     if (monitor_requires_iothread(mon)) {
@@ -565,7 +542,6 @@ int monitor_suspend(Monitor *mon)
     }
 
     trace_monitor_suspend(mon, 1);
-    return 0;
 }
 
 static void monitor_accept_input(void *opaque)
@@ -582,10 +558,6 @@ static void monitor_accept_input(void *opaque)
 
 void monitor_resume(Monitor *mon)
 {
-    if (monitor_is_hmp_non_interactive(mon)) {
-        return;
-    }
-
     if (qatomic_dec_fetch(&mon->suspend_cnt) == 0) {
         AioContext *ctx;
 
Signature · 1 line
2.54.0
#9 lore raw

PATCH v5 5/35 monitor: minimal conversion of monitors to QOM

In reply to Daniel P. Berrangé

This introduces a Monitor QOM object, with MonitorHMP and MonitorQMP subclasses. This is the bare minimum conversion of just the type declarations and replacing g_new/g_free with object_new/object_unref.

---
 include/monitor/monitor.h  | 11 ++++++++++-
 monitor/hmp.c              | 29 +++++++++++++++++++++++++++--
 monitor/monitor-internal.h | 18 ++++++++++++++++--
 monitor/monitor.c          | 18 ++++++++++++++++--
 monitor/qmp-cmds.c         | 15 ++++++++-------
 monitor/qmp.c              | 29 +++++++++++++++++++++++++++--
 6 files changed, 104 insertions(+), 16 deletions(-)
include/monitor/monitor.h +10 -1
index b9642b58ba..2e9f9e12e9 100644
@@ -5,8 +5,17 @@
 #include "qapi/qapi-types-misc.h"
 #include "qemu/readline.h"
 #include "exec/hwaddr.h"
+#include "qom/object.h"
+
+#define TYPE_MONITOR "monitor"
+OBJECT_DECLARE_TYPE(Monitor, MonitorClass, MONITOR);
+
+#define TYPE_MONITOR_HMP "monitor-hmp"
+OBJECT_DECLARE_TYPE(MonitorHMP, MonitorHMPClass, MONITOR_HMP);
+
+#define TYPE_MONITOR_QMP "monitor-qmp"
+OBJECT_DECLARE_TYPE(MonitorQMP, MonitorQMPClass, MONITOR_QMP);
 
-typedef struct MonitorHMP MonitorHMP;
 typedef struct MonitorOptions MonitorOptions;
 
 #define QMP_REQ_QUEUE_LEN_MAX 8
monitor/hmp.c +27 -2
index 4e4468424a..81047d2513 100644
@@ -43,6 +43,20 @@
 #include "system/block-backend.h"
 #include "trace.h"
 
+OBJECT_DEFINE_TYPE(MonitorHMP, monitor_hmp, MONITOR_HMP, MONITOR);
+
+static void monitor_hmp_finalize(Object *obj)
+{
+}
+
+static void monitor_hmp_class_init(ObjectClass *cls, const void *data)
+{
+}
+
+static void monitor_hmp_init(Object *obj)
+{
+}
+
 static void monitor_command_cb(void *opaque, const char *cmdline,
                                void *readline_opaque)
 {
@@ -1526,10 +1540,21 @@ static void monitor_readline_flush(void *opaque)
 
 void monitor_new_hmp(Chardev *chr, bool use_readline, Error **errp)
 {
-    MonitorHMP *mon = g_new0(MonitorHMP, 1);
+    MonitorHMP *mon;
+    static int counter;
+    g_autofree char *id = g_strdup_printf("hmpcompat%d", counter++);
+    Object *obj = object_new_with_props(TYPE_MONITOR_HMP,
+                                        object_get_objects_root(),
+                                        id,
+                                        errp,
+                                        NULL);
+    if (!obj) {
+        return;
+    }
+    mon = MONITOR_HMP(obj);
 
     if (!qemu_chr_fe_init(&mon->parent_obj.chr, chr, errp)) {
-        g_free(mon);
+        object_unparent(OBJECT(mon));
         return;
     }
 
monitor/monitor-internal.h +16 -2
index 4da2b2a677..05c1f2f5e0 100644
@@ -101,7 +101,13 @@ typedef struct HMPCommand {
     bool coroutine;
 } HMPCommand;
 
+
+struct MonitorClass {
+    ObjectClass parent_class;
+};
+
 struct Monitor {
+    Object parent;
     CharFrontend chr;
     int suspend_cnt;            /* Needs to be accessed atomically */
     bool is_qmp;
@@ -127,6 +133,10 @@ struct Monitor {
     int reset_seen;
 };
 
+struct MonitorHMPClass {
+    MonitorClass parent_class;
+};
+
 struct MonitorHMP {
     Monitor parent_obj;
     bool use_readline;
@@ -140,7 +150,11 @@ struct MonitorHMP {
     ReadLineState *rs;
 };
 
-typedef struct {
+struct MonitorQMPClass {
+    MonitorClass parent_class;
+};
+
+struct MonitorQMP {
     Monitor parent_obj;
     JSONMessageParser parser;
     bool pretty;
@@ -160,7 +174,7 @@ typedef struct {
     QemuMutex qmp_queue_lock;
     /* Input queue that holds all the parsed QMP requests */
     GQueue *qmp_requests;
-} MonitorQMP;
+};
 
 /**
  * Is @mon a QMP monitor?
monitor/monitor.c +16 -2
index a87597e606..a497c25c54 100644
@@ -73,6 +73,20 @@ static GHashTable *coroutine_mon; /* Maps Coroutine* to Monitor* */
 MonitorList mon_list;
 static bool monitor_destroyed;
 
+OBJECT_DEFINE_TYPE(Monitor, monitor, MONITOR, OBJECT);
+
+static void monitor_finalize(Object *obj)
+{
+}
+
+static void monitor_class_init(ObjectClass *cls, const void *data)
+{
+}
+
+static void monitor_init(Object *obj)
+{
+}
+
 Monitor *monitor_cur(void)
 {
     Monitor *mon;
@@ -598,7 +612,7 @@ void monitor_list_append(Monitor *mon)
 
     if (mon) {
         monitor_data_destroy(mon);
-        g_free(mon);
+        object_unparent(OBJECT(mon));
     }
 }
 
@@ -680,7 +694,7 @@ void monitor_cleanup(void)
         monitor_flush(mon);
         monitor_data_destroy(mon);
         qemu_mutex_lock(&monitor_lock);
-        g_free(mon);
+        object_unparent(OBJECT(mon));
     }
     qemu_mutex_unlock(&monitor_lock);
 
monitor/qmp-cmds.c +8 -7
index aa9ee8a391..bfde769ef0 100644
@@ -166,12 +166,12 @@ char *qmp_human_monitor_command(const char *command_line, bool has_cpu_index,
                                 int64_t cpu_index, Error **errp)
 {
     char *output = NULL;
-    MonitorHMP hmp = {};
+    MonitorHMP *hmp = MONITOR_HMP(object_new(TYPE_MONITOR_HMP));
 
-    monitor_data_init(&hmp.parent_obj, false, true, false);
+    monitor_data_init(&hmp->parent_obj, false, true, false);
 
     if (has_cpu_index) {
-        int ret = monitor_set_cpu(&hmp.parent_obj, cpu_index);
+        int ret = monitor_set_cpu(&hmp->parent_obj, cpu_index);
         if (ret < 0) {
             error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cpu-index",
                        "a CPU number");
@@ -179,14 +179,15 @@ char *qmp_human_monitor_command(const char *command_line, bool has_cpu_index,
         }
     }
 
-    handle_hmp_command(&hmp, command_line);
+    handle_hmp_command(hmp, command_line);
 
-    WITH_QEMU_LOCK_GUARD(&hmp.parent_obj.mon_lock) {
-        output = g_strdup(hmp.parent_obj.outbuf->str);
+    WITH_QEMU_LOCK_GUARD(&hmp->parent_obj.mon_lock) {
+        output = g_strdup(hmp->parent_obj.outbuf->str);
     }
 
 out:
-    monitor_data_destroy(&hmp.parent_obj);
+    monitor_data_destroy(&hmp->parent_obj);
+    object_unref(hmp);
     return output;
 }
 
monitor/qmp.c +27 -2
index cb28a95efd..5231ed506a 100644
@@ -71,6 +71,20 @@ typedef struct QMPRequest QMPRequest;
 
 QmpCommandList qmp_commands, qmp_cap_negotiation_commands;
 
+OBJECT_DEFINE_TYPE(MonitorQMP, monitor_qmp, MONITOR_QMP, MONITOR);
+
+static void monitor_qmp_finalize(Object *obj)
+{
+}
+
+static void monitor_qmp_class_init(ObjectClass *cls, const void *data)
+{
+}
+
+static void monitor_qmp_init(Object *obj)
+{
+}
+
 static bool qmp_oob_enabled(MonitorQMP *mon)
 {
     return mon->capab[QMP_CAPABILITY_OOB];
@@ -515,10 +529,21 @@ static void monitor_qmp_setup_handlers_bh(void *opaque)
 
 void monitor_new_qmp(Chardev *chr, bool pretty, Error **errp)
 {
-    MonitorQMP *mon = g_new0(MonitorQMP, 1);
+    MonitorQMP *mon;
+    static int counter;
+    g_autofree char *id = g_strdup_printf("qmpcompat%d", counter++);
+    Object *obj = object_new_with_props(TYPE_MONITOR_QMP,
+                                        object_get_objects_root(),
+                                        id,
+                                        errp,
+                                        NULL);
+    if (!obj) {
+        return;
+    }
+    mon = MONITOR_QMP(obj);
 
     if (!qemu_chr_fe_init(&mon->parent_obj.chr, chr, errp)) {
-        g_free(mon);
+        object_unparent(OBJECT(mon));
         return;
     }
     qemu_chr_fe_set_echo(&mon->parent_obj.chr, true);
Signature · 1 line
2.54.0
#36 lore raw

PATCH v5 25/35 monitor: protect qemu_chr_fe_accept_input with monitor lock

In reply to Daniel P. Berrangé

From: Christian Brauner <[email protected]>

The monitor_accept_input API is called from a bottom half, and will invoke qemu_chr_fe_accept_input().

When a following patch introduces the ability to delete monitors, it will be neccesary to delete the bottom half. Protecting the call to qemu_chr_fe_accept_input with the monitor lock will allow for synchronization with the deletion process.

[DB: extracted from a larger commit and refactored to apply
to the new monitor class structure]

---
 monitor/hmp.c     | 2 ++
 monitor/monitor.c | 6 +-----
 monitor/qmp.c     | 9 +++++++++
 3 files changed, 12 insertions(+), 5 deletions(-)
monitor/hmp.c +2 -0
index 2cd508f1ee..7b084c6ff8 100644
@@ -116,9 +116,11 @@ static void monitor_hmp_accept_input(Monitor *mon)
         MonitorHMP *hmp = MONITOR_HMP(mon);
         assert(hmp->rs);
         readline_restart(hmp->rs);
+        qemu_chr_fe_accept_input(&mon->chr);
         qemu_mutex_unlock(&mon->mon_lock);
         readline_show_prompt(hmp->rs);
     } else {
+        qemu_chr_fe_accept_input(&mon->chr);
         qemu_mutex_unlock(&mon->mon_lock);
     }
 }
monitor/monitor.c +1 -5
index 312e069bc5..0e1c623555 100644
@@ -560,11 +560,7 @@ static void monitor_accept_input(void *opaque)
     Monitor *mon = opaque;
     MonitorClass *cls = MONITOR_GET_CLASS(mon);
 
-    if (cls->accept_input) {
-        cls->accept_input(mon);
-    }
-
-    qemu_chr_fe_accept_input(&mon->chr);
+    cls->accept_input(mon);
 }
 
 void monitor_resume(Monitor *mon)
monitor/qmp.c +9 -0
index b0535dec68..2131be82c7 100644
@@ -107,6 +107,7 @@ static void monitor_qmp_emit_event(Monitor *mon, QAPIEvent event, QDict *qdict);
 static bool monitor_qmp_requires_iothread(const Monitor *mon);
 static void monitor_qmp_complete(UserCreatable *uc, Error **errp);
 static bool monitor_qmp_prepare_delete(UserCreatable *uc, Error **errp);
+static void monitor_qmp_accept_input(Monitor *mon);
 
 static void monitor_qmp_class_init(ObjectClass *cls, const void *data)
 {
@@ -119,6 +120,7 @@ static void monitor_qmp_class_init(ObjectClass *cls, const void *data)
 
     moncls->emit_event = monitor_qmp_emit_event;
     moncls->requires_iothread = monitor_qmp_requires_iothread;
+    moncls->accept_input = monitor_qmp_accept_input;
 
     ucc->complete = monitor_qmp_complete;
     ucc->prepare_delete = monitor_qmp_prepare_delete;
@@ -666,3 +668,10 @@ static bool monitor_qmp_prepare_delete(UserCreatable *uc, Error **errp)
     error_setg(errp, "Deleting QMP monitors is not supported");
     return false;
 }
+
+static void monitor_qmp_accept_input(Monitor *mon)
+{
+    WITH_QEMU_LOCK_GUARD(&mon->mon_lock) {
+        qemu_chr_fe_accept_input(&mon->chr);
+    }
+}
Signature · 1 line
2.54.0
#26 lore raw

PATCH v5 16/35 util: use dynamic cast in error vreport

In reply to Daniel P. Berrangé

This eliminates a use of monitor_is_qmp() from the error reporting path.

---
 util/error-report.c | 13 ++++++-------
 1 file changed, 6 insertions(+), 7 deletions(-)
util/error-report.c +6 -7
index f832ad9b6b..f333af9249 100644
@@ -231,16 +231,15 @@ char *real_time_iso8601(void)
 G_GNUC_PRINTF(2, 0)
 static void vreport(report_type type, const char *fmt, va_list ap)
 {
-    Monitor *cur = monitor_cur();
-    gchar *timestr;
-
     /*
      * When current monitor is QMP, messages must go to stderr
-     * and have prefixes added
+     * and have prefixes added, so we cast to HMP, leaving 'cur'
+     * as NULL in QMP case
      */
-    if (monitor_cur_is_qmp()) {
-        cur = NULL;
-    }
+    Monitor *cur = MONITOR(
+        object_dynamic_cast(OBJECT(monitor_cur()), TYPE_MONITOR_HMP));
+    gchar *timestr;
+
     if (!cur) {
         qemu_flockfile(stderr);
     }
Signature · 1 line
2.54.0
#61 lore raw

PATCH v5 35/35 docs: mark '-mon' as deprecated in favour of -object

In reply to Daniel P. Berrangé

The high level -qmp and -monitor options can remain as convenience wrappers, but the low level -mon is completed obsoleted by the new -object support with 'monitor-qmp' and 'monitor-hmp' types.

---
 docs/about/deprecated.rst               | 10 ++++++++++
 docs/devel/writing-monitor-commands.rst |  4 ++--
 docs/system/arm/xenpvh.rst              |  4 ++--
 docs/system/i386/xen.rst                |  3 ++-
 docs/system/i386/xenpvh.rst             |  4 ++--
 qemu-options.hx                         | 11 ++++++-----
 system/vl.c                             |  2 ++
 7 files changed, 26 insertions(+), 12 deletions(-)
docs/about/deprecated.rst +10 -0
index 97750f5edc..8370fc5680 100644
@@ -61,6 +61,16 @@ The ``debug-threads`` option of the ``-name`` argument is now
 ignored. Thread naming is unconditionally enabled for all platforms
 where it is supported.
 
+``-mon`` option (since 11.1)
+''''''''''''''''''''''''''''
+
+The ``-mon`` option was the generic mechanism for creating monitor objects
+if the convenience ``-qmp`` or ``-monitor`` options were not flexible
+enough. The monitor objects have been converted to QOM, so ``-mon mode=readline``
+is replaced by ``-object monitor-hmp`` and ``-mon mode=control`` is replaced
+by ``-object monitor-qmp``. The short convenience options are not deprecated,
+only ``-mon``.
+
 QEMU Machine Protocol (QMP) commands
 ------------------------------------
 
docs/devel/writing-monitor-commands.rst +2 -2
index 930da5cd06..7ae7efe327 100644
@@ -52,8 +52,8 @@ shown here.
 First, QEMU should be started like this::
 
  # qemu-system-TARGET [...] \
-     -chardev socket,id=qmp,port=4444,host=localhost,server=on \
-     -mon chardev=qmp,mode=control,pretty=on
+     -chardev socket,id=chrqmp0,port=4444,host=localhost,server=on \
+     -object monitor-qmp,chardev=chrqmp0,pretty=on,id=qmp0
 
 Then, in a different terminal::
 
docs/system/arm/xenpvh.rst +2 -2
index 430ac2c02e..511af63e3d 100644
@@ -27,9 +27,9 @@ Sample QEMU xenpvh commands for running and connecting with Xen:
 
     qemu-system-aarch64 -xen-domid 1 \
       -chardev socket,id=libxl-cmd,path=qmp-libxl-1,server=on,wait=off \
-      -mon chardev=libxl-cmd,mode=control \
+      -object monitor-qmp,id=qmp0,chardev=libxl-cmd \
       -chardev socket,id=libxenstat-cmd,path=qmp-libxenstat-1,server=on,wait=off \
-      -mon chardev=libxenstat-cmd,mode=control \
+      -object monitor-qmp,id=qmp1,chardev=libxenstat-cmd \
       -xen-attach -name guest0 -vnc none -display none -nographic \
       -machine xenpvh -m 1301 \
       -chardev socket,id=chrtpm,path=tmp/vtpm2/swtpm-sock \
docs/system/i386/xen.rst +2 -1
index 46db5f34c1..6ff5cdb1e4 100644
@@ -79,7 +79,8 @@ of type ``xen-console`` to connect to it. For the Xen console equivalent of
 the handy ``-serial mon:stdio`` option, for example:
 
 .. parsed-literal::
-   -chardev stdio,mux=on,id=char0,signal=off -mon char0 \\
+   -chardev stdio,mux=on,id=char0,signal=off \\
+   -object monitor-hmp,chardev=char0,id=hmp0 \\
    -device xen-console,chardev=char0
 
 The Xen network device is ``xen-net-device``, which becomes the default NIC
docs/system/i386/xenpvh.rst +2 -2
index 354250f073..904778e3f5 100644
@@ -33,9 +33,9 @@ case you need to construct one manually:
 
     qemu-system-i386 -xen-domid 3 -no-shutdown        \
       -chardev socket,id=libxl-cmd,path=/var/run/xen/qmp-libxl-3,server=on,wait=off \
-      -mon chardev=libxl-cmd,mode=control             \
+      -object monitor-qmp,id=qmp0,chardev=libxl-cmd   \
       -chardev socket,id=libxenstat-cmd,path=/var/run/xen/qmp-libxenstat-3,server=on,wait=off \
-      -mon chardev=libxenstat-cmd,mode=control        \
+      -object monitor-qmp,id=qmp1,chardev=libxenstat-cmd \
       -nodefaults                                     \
       -no-user-config                                 \
       -xen-attach -name g0                            \
qemu-options.hx +6 -5
index a804d5b880..999dca7444 100644
@@ -4137,7 +4137,7 @@ The general form of a character device option is:
     ::
 
         -chardev stdio,mux=on,id=char0 \
-        -mon chardev=char0,mode=readline \
+        -object monitor-hmp,id=hmp0,chardev=char0 \
         -serial chardev:char0 \
         -serial chardev:char0
 
@@ -4149,7 +4149,7 @@ The general form of a character device option is:
     ::
 
         -chardev stdio,mux=on,id=char0 \
-        -mon chardev=char0,mode=readline \
+        -object monitor-hmp,id=hmp0,chardev=char0 \
         -parallel chardev:char0 \
         -chardev tcp,...,mux=on,id=char1 \
         -serial chardev:char1 \
@@ -4950,7 +4950,8 @@ SRST
         -qmp tcp:localhost:4444,server=on,wait=off
 
     Not all options are configurable via this syntax; for maximum
-    flexibility use the ``-mon`` option and an accompanying ``-chardev``.
+    flexibility use ``-object monitor-qmp`` and an accompanying
+    ``-chardev``.
 
 ERST
 DEF("qmp-pretty", HAS_ARG, QEMU_OPTION_qmp_pretty, \
@@ -4981,12 +4982,12 @@ SRST
 
     enables the QMP monitor on localhost port 4444 with pretty-printing.
 
-    The use of ``-mon mode=readline`` is historical syntax sugar
+    The use of ``-mon mode=readline`` is deprecated syntax sugar
     for the new ``-object monitor-hmp`` option, each use of which
     creates an object with the ID ``hmpcompatNNN`` where ``NNN`` is
     a counter starting from 0.
 
-    The use of ``-mon mode=control`` is historical syntax sugar
+    The use of ``-mon mode=control`` is deprecated syntax sugar
     for the new ``-object monitor-qmp`` option, each use of which
     creates an object with the ID ``qmpcompatNNN`` where ``NNN`` is
     a counter starting from 0.
system/vl.c +2 -0
index 79ed2be449..a7b6d37452 100644
@@ -3240,6 +3240,8 @@ void qemu_init(int argc, char **argv)
                 default_monitor = 0;
                 break;
             case QEMU_OPTION_mon:
+                warn_report_once("'-mon' is deprecated, use '-object' with "
+                                 "'monitor-hmp' or 'monitor-qmp' types instead");
                 if (!qemu_opts_parse_noisily(qemu_find_opts("mon"), optarg,
                                              true)) {
                     exit(1);
Signature · 1 line
2.54.0
#34 lore raw

PATCH v5 23/35 monitor: convert from oneshot BH to persistent BH

In reply to Daniel P. Berrangé

From: Christian Brauner <[email protected]>

Convert monitor_accept_input from a oneshot BH (aio_bh_schedule_oneshot) to a persistent BH (aio_bh_new + qemu_bh_schedule). Oneshot BHs cannot be cancelled, so monitor_resume() racing with destruction would schedule a callback against memory that monitor_qmp_destroy() is about to free. A persistent BH can be deleted during destruction, cancelling any pending schedule.

---
 monitor/monitor-internal.h |  2 +-
 monitor/monitor.c          | 25 ++++++++++++++-----------
 2 files changed, 15 insertions(+), 12 deletions(-)
monitor/monitor-internal.h +1 -1
index 012d442a20..a82e1aacb6 100644
@@ -134,7 +134,7 @@ struct Monitor {
     char *chardev_id;
     CharFrontend chr;
     int suspend_cnt;            /* Needs to be accessed atomically */
-
+    QEMUBH *accept_input_bh;    /* persistent BH for monitor_accept_input */
     char *mon_cpu_path;
     QTAILQ_ENTRY(Monitor) entry;
 
monitor/monitor.c +14 -11
index bf5e8a184a..312e069bc5 100644
@@ -81,6 +81,9 @@ static void monitor_finalize(Object *obj)
 {
     Monitor *mon = MONITOR(obj);
 
+    if (mon->accept_input_bh) {
+        qemu_bh_delete(mon->accept_input_bh);
+    }
     g_free(mon->chardev_id);
     g_free(mon->mon_cpu_path);
     qemu_chr_fe_deinit(&mon->chr, false);
@@ -567,15 +570,7 @@ static void monitor_accept_input(void *opaque)
 void monitor_resume(Monitor *mon)
 {
     if (qatomic_dec_fetch(&mon->suspend_cnt) == 0) {
-        AioContext *ctx;
-
-        if (monitor_requires_iothread(mon)) {
-            ctx = iothread_get_aio_context(mon_iothread);
-        } else {
-            ctx = qemu_get_aio_context();
-        }
-
-        aio_bh_schedule_oneshot(ctx, monitor_accept_input, mon);
+        qemu_bh_schedule(mon->accept_input_bh);
     }
 
     trace_monitor_suspend(mon, -1);
@@ -691,6 +686,7 @@ void monitor_init_globals(void)
 static void monitor_complete(UserCreatable *uc, Error **errp)
 {
     Monitor *mon = MONITOR(uc);
+    AioContext *ctx;
 
     if (mon->chardev_id) {
         Chardev *chr = qemu_chr_find(mon->chardev_id);
@@ -704,9 +700,16 @@ static void monitor_complete(UserCreatable *uc, Error **errp)
         }
     }
 
-    if (monitor_requires_iothread(mon) && !mon_iothread) {
-        mon_iothread = iothread_create("mon_iothread", &error_abort);
+    if (monitor_requires_iothread(mon)) {
+        if (!mon_iothread) {
+            mon_iothread = iothread_create("mon_iothread", &error_abort);
+        }
+
+        ctx = iothread_get_aio_context(mon_iothread);
+    } else {
+        ctx = qemu_get_aio_context();
     }
+    mon->accept_input_bh = aio_bh_new(ctx, monitor_accept_input, mon);
 }
 
 int monitor_new(MonitorOptions *opts, bool allow_hmp, Error **errp)
Signature · 1 line
2.54.0
#37 lore raw

PATCH v5 26/35 monitor: implement support for deleting QMP objects

In reply to Daniel P. Berrangé

From: Christian Brauner <[email protected]>

The removal sequence is:

1. Remove from mon_list under monitor_lock. This must happen before disconnecting chardev handlers to prevent event broadcast from calling monitor_flush_locked() after the gcontext reset, which would create an out_watch on the wrong GMainContext (see monitor_cancel_out_watch()).
2. Cancel any pending out_watch while gcontext still points to the correct context.
3. Disconnect chardev handlers, passing context=NULL and close the connection.
4. Drain pending requests from any in-flight monitor_qmp_read().
5. Destroy the monitor object

[DB: extracted from a larger commit and refactored to apply
to the new monitor class structure. Remove 'self delete' feature which requires complex special-case code]

---
 monitor/monitor-internal.h |  2 ++
 monitor/monitor.c          | 22 ++++++++++++++++
 monitor/qmp.c              | 54 ++++++++++++++++++++++++++++++++++++--
 3 files changed, 76 insertions(+), 2 deletions(-)
monitor/monitor-internal.h +2 -0
index a82e1aacb6..5522e05464 100644
@@ -178,6 +178,7 @@ struct MonitorQMP {
     Monitor parent_obj;
     JSONMessageParser parser;
     bool pretty;
+    bool setup_pending; /* iothread BH has not yet set up chardev handlers */
     /*
      * When a client connects, we're in capabilities negotiation mode.
      * @commands is &qmp_cap_negotiation_commands then.  When command
@@ -206,6 +207,7 @@ extern MonitorList mon_list;
 
 bool monitor_requires_iothread(const Monitor *mon);
 int monitor_can_read(void *opaque);
+void monitor_cancel_out_watch(Monitor *mon);
 void monitor_list_append(Monitor *mon);
 void monitor_fdsets_cleanup(void);
 int monitor_set_cpu(Monitor *mon, int cpu_index);
monitor/monitor.c +22 -0
index 0e1c623555..b155f58546 100644
@@ -177,6 +177,28 @@ static gboolean monitor_unblocked(void *do_not_use, GIOCondition cond,
     return G_SOURCE_REMOVE;
 }
 
+/* Cancel a pending out_watch GSource.  Caller must hold mon_lock. */
+void monitor_cancel_out_watch(Monitor *mon)
+{
+    if (mon->out_watch) {
+        GMainContext *ctx = NULL;
+        GSource *src;
+
+        if (monitor_requires_iothread(mon)) {
+            ctx = iothread_get_g_main_context(mon_iothread);
+        }
+        src = g_main_context_find_source_by_id(ctx, mon->out_watch);
+        if (!src && ctx) {
+            /* Handler disconnect may have reset gcontext to NULL. */
+            src = g_main_context_find_source_by_id(NULL, mon->out_watch);
+        }
+        if (src) {
+            g_source_destroy(src);
+        }
+        mon->out_watch = 0;
+    }
+}
+
 /* Caller must hold mon->mon_lock */
 void monitor_flush_locked(Monitor *mon)
 {
monitor/qmp.c +52 -2

86 lines not on this page · raw message

Signature · 1 line
2.54.0
#29 lore raw

PATCH v5 19/35 monitor: use dynamic cast in monitor_is_hmp_non_interactive

In reply to Daniel P. Berrangé

Rather than checking !monitor_is_qmp(), use a dynamic cast to check for HMP.

---
 monitor/monitor.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
monitor/monitor.c +1 -1
index 0db7903795..15e49d78d7 100644
@@ -168,7 +168,7 @@ static inline bool monitor_uses_readline(const MonitorHMP *mon)
 
 static inline bool monitor_is_hmp_non_interactive(const Monitor *mon)
 {
-    if (monitor_is_qmp(mon)) {
+    if (!object_dynamic_cast(OBJECT(mon), TYPE_MONITOR_HMP)) {
         return false;
     }
 
Signature · 1 line
2.54.0
#24 lore raw

PATCH v5 14/35 monitor: use class method for I/O thread request

In reply to Daniel P. Berrangé

Introducing a virtual "requires_iothread" method allows the code to automatically initialize the I/O thread during object completion.

---
 monitor/monitor-internal.h | 14 +++++++++-----
 monitor/monitor.c          | 22 ++++++++++++----------
 monitor/qmp.c              | 20 +++++++++++---------
 3 files changed, 32 insertions(+), 24 deletions(-)
monitor/monitor-internal.h +9 -5
index 592f146331..efd21b8912 100644
@@ -121,6 +121,12 @@ struct MonitorClass {
      * the monitor to accept further client input
      */
     void (*accept_input)(Monitor *mon);
+
+    /*
+     * If non-NULL and returns true, then an I/O thread
+     * is required for processing the monitor
+     */
+    bool (*requires_iothread)(const Monitor *mon);
 };
 
 struct Monitor {
@@ -129,7 +135,6 @@ struct Monitor {
     CharFrontend chr;
     int suspend_cnt;            /* Needs to be accessed atomically */
     bool is_qmp;
-    bool use_io_thread;
 
     char *mon_cpu_path;
     QTAILQ_ENTRY(Monitor) entry;
@@ -159,9 +164,8 @@ struct MonitorHMP {
     bool use_readline;
     /*
      * State used only in the thread "owning" the monitor.
-     * If @use_io_thread, this is @mon_iothread. (This does not actually happen
-     * in the current state of the code.)
-     * Else, it's the main thread.
+     * This is currently always the main thread, since
+     * HMP does not allow use of the I/O thread at this time.
      * These members can be safely accessed without locks.
      */
     ReadLineState *rs;
@@ -210,7 +214,7 @@ extern QemuMutex monitor_lock;
 extern MonitorList mon_list;
 
 void monitor_complete(Monitor *mon, Error **errp);
-void monitor_iothread_init(Monitor *mon);
+bool monitor_requires_iothread(const Monitor *mon);
 int monitor_can_read(void *opaque);
 void monitor_list_append(Monitor *mon);
 void monitor_fdsets_cleanup(void);
monitor/monitor.c +12 -10
index 2f16cd3053..b788669f1f 100644
@@ -160,6 +160,12 @@ bool monitor_cur_is_qmp(void)
     return cur_mon && monitor_is_qmp(cur_mon);
 }
 
+bool monitor_requires_iothread(const Monitor *mon)
+{
+    MonitorClass *cls = MONITOR_GET_CLASS(mon);
+    return cls->requires_iothread && cls->requires_iothread(mon);
+}
+
 /**
  * Is @mon is using readline?
  * Note: not all HMP monitors use readline, e.g., gdbserver has a
@@ -561,7 +567,7 @@ int monitor_suspend(Monitor *mon)
 
     qatomic_inc(&mon->suspend_cnt);
 
-    if (mon->use_io_thread) {
+    if (monitor_requires_iothread(mon)) {
         /*
          * Kick I/O thread to make sure this takes effect.  It'll be
          * evaluated again in prepare() of the watch object.
@@ -594,7 +600,7 @@ void monitor_resume(Monitor *mon)
     if (qatomic_dec_fetch(&mon->suspend_cnt) == 0) {
         AioContext *ctx;
 
-        if (mon->use_io_thread) {
+        if (monitor_requires_iothread(mon)) {
             ctx = iothread_get_aio_context(mon_iothread);
         } else {
             ctx = qemu_get_aio_context();
@@ -632,14 +638,6 @@ void monitor_list_append(Monitor *mon)
     }
 }
 
-void monitor_iothread_init(Monitor *mon)
-{
-    if (!mon_iothread) {
-        mon_iothread = iothread_create("mon_iothread", &error_abort);
-    }
-    mon->use_io_thread = true;
-}
-
 void monitor_cleanup(void)
 {
     /*
@@ -734,6 +732,10 @@ void monitor_complete(Monitor *mon, Error **errp)
             return;
         }
     }
+
+    if (monitor_requires_iothread(mon) && !mon_iothread) {
+        mon_iothread = iothread_create("mon_iothread", &error_abort);
+    }
 }
 
 int monitor_new(MonitorOptions *opts, bool allow_hmp, Error **errp)
monitor/qmp.c +11 -9
index ec17050a70..93f576c19d 100644
@@ -100,6 +100,7 @@ static void monitor_qmp_set_pretty(Object *obj, bool val, Error **errp)
 }
 
 static void monitor_qmp_emit_event(Monitor *mon, QAPIEvent event, QDict *qdict);
+static bool monitor_qmp_requires_iothread(const Monitor *mon);
 
 static void monitor_qmp_class_init(ObjectClass *cls, const void *data)
 {
@@ -110,6 +111,7 @@ static void monitor_qmp_class_init(ObjectClass *cls, const void *data)
                                    monitor_qmp_set_pretty);
 
     moncls->emit_event = monitor_qmp_emit_event;
+    moncls->requires_iothread = monitor_qmp_requires_iothread;
 }
 
 static void handle_qmp_command(void *opaque, QObject *req, Error *err);
@@ -136,6 +138,11 @@ static void monitor_qmp_emit_event(Monitor *mon, QAPIEvent event, QDict *qdict)
     qmp_send_response(qmp, qdict);
 }
 
+static bool monitor_qmp_requires_iothread(const Monitor *mon)
+{
+    return qemu_chr_has_feature(mon->chr.chr,
+                                QEMU_CHAR_FEATURE_GCONTEXT);
+}
 
 static bool qmp_oob_enabled(MonitorQMP *mon)
 {
@@ -146,7 +153,8 @@ static void monitor_qmp_caps_reset(MonitorQMP *mon)
 {
     memset(mon->capab_offered, 0, sizeof(mon->capab_offered));
     memset(mon->capab, 0, sizeof(mon->capab));
-    mon->capab_offered[QMP_CAPABILITY_OOB] = mon->parent_obj.use_io_thread;
+    mon->capab_offered[QMP_CAPABILITY_OOB] =
+        monitor_requires_iothread(MONITOR(mon));
 }
 
 static void qmp_request_free(QMPRequest *req)
@@ -562,7 +570,7 @@ static void monitor_qmp_setup_handlers_bh(void *opaque)
     MonitorQMP *mon = opaque;
     GMainContext *context;
 
-    assert(mon->parent_obj.use_io_thread);
+    assert(monitor_requires_iothread(MONITOR(mon)));
     context = iothread_get_g_main_context(mon_iothread);
     assert(context);
     qemu_chr_fe_set_handlers(&mon->parent_obj.chr, monitor_can_read,
@@ -598,13 +606,7 @@ void monitor_new_qmp(const char *chardev_id, bool pretty, Error **errp)
 
     qemu_chr_fe_set_echo(&mon->parent_obj.chr, true);
 
-    /* Note: we run QMP monitor in I/O thread when @chr supports that */
-    if (qemu_chr_has_feature(mon->parent_obj.chr.chr,
-                             QEMU_CHAR_FEATURE_GCONTEXT)) {
-        monitor_iothread_init(&mon->parent_obj);
-    }
-
-    if (mon->parent_obj.use_io_thread) {
+    if (monitor_requires_iothread(MONITOR(mon))) {
         /*
          * Make sure the old iowatch is gone.  It's possible when
          * e.g. the chardev is in client mode, with wait=on.
Signature · 1 line
2.54.0
#55 lore raw

PATCH v5 33/35 tests: switch from -mon to -object monitor-qmp

In reply to Daniel P. Berrangé

Use the new preferred low level option for configuring the QMP service in libqtest and the python Machine class used by tests. This will avoid triggering deprecation warnings after the subsequent commit.

---
 python/qemu/machine/machine.py | 4 ++--
 tests/qtest/libqtest.c         | 2 +-
 2 files changed, 3 insertions(+), 3 deletions(-)
python/qemu/machine/machine.py +2 -2
index ebb58d5b68..5f4845f5fb 100644
@@ -304,8 +304,8 @@ def _base_args(self) -> List[str]:
                 )
             else:
                 moncdev = f"socket,id=mon,path={self._monitor_address}"
-            args.extend(['-chardev', moncdev, '-mon',
-                         'chardev=mon,mode=control'])
+            args.extend(['-chardev', moncdev, '-object',
+                         'monitor-qmp,id=qmp,chardev=mon'])
 
         if self._machine is not None:
             args.extend(['-machine', self._machine])
tests/qtest/libqtest.c +1 -1
index 4e22c66b75..c33c799c92 100644
@@ -460,7 +460,7 @@ gchar *qtest_qemu_args(const char *extra_args)
                       "-qtest unix:%s "
                       "-qtest-log %s "
                       "-chardev socket,path=%s,id=char0 "
-                      "-mon chardev=char0,mode=control "
+                      "-object monitor-qmp,id=qmp0,chardev=char0 "
                       "-display none "
                       "-audio none "
                       "%s "
Signature · 1 line
2.54.0
#32 lore raw

PATCH v5 22/35 monitor: implement "user creatable" interface for adding monitors

In reply to Daniel P. Berrangé

Implement the user creatable QOM interface and define the monitor-qmp and monitor-hmp types in QAPI. This unlocks the ability to create them on the command line with -object or in HMP/QMP with object_add.

For example:

$QEMU -chardev stdio,id=monchr0 -object monitor-hmp,id=mon0,chrdev=monchr0

Initially the "prepare_delete" callback is hardcoded to return an error which means -object and object_add can be used, but object_del will fail. Support for deleting monitors will be introduced in subsequent commits.

---
 monitor/hmp.c              | 70 ++++++++++++++++++++++++--------------
 monitor/monitor-internal.h |  1 -
 monitor/monitor.c          | 14 ++++++--
 monitor/qmp.c              | 46 +++++++++++++++++--------
 qapi/qom.json              | 43 +++++++++++++++++++++++
 stubs/monitor-internal.c   |  1 +
 system/vl.c                |  8 ++++-
 7 files changed, 138 insertions(+), 45 deletions(-)
monitor/hmp.c +44 -26
index b8cccdcde3..2cd508f1ee 100644
@@ -40,6 +40,7 @@
 #include "qemu/base-arch-defs.h"
 #include "qemu/target-info.h"
 #include "qemu/units.h"
+#include "qom/object_interfaces.h"
 #include "exec/gdbstub.h"
 #include "system/block-backend.h"
 #include "trace.h"
@@ -71,10 +72,13 @@ static void monitor_hmp_set_readline(Object *obj, bool val, Error **errp)
 int monitor_hmp_vprintf(Monitor *mon, const char *fmt, va_list ap)
     G_GNUC_PRINTF(2, 0);
 static void monitor_hmp_accept_input(Monitor *mon);
+static void monitor_hmp_complete(UserCreatable *uc, Error **errp);
+static bool monitor_hmp_prepare_delete(UserCreatable *uc, Error **errp);
 
 static void monitor_hmp_class_init(ObjectClass *cls, const void *data)
 {
     MonitorClass *moncls = MONITOR_CLASS(cls);
+    UserCreatableClass *ucc = USER_CREATABLE_CLASS(cls);
 
     object_class_property_add_bool(cls, "readline",
                                    monitor_hmp_get_readline,
@@ -82,6 +86,9 @@ static void monitor_hmp_class_init(ObjectClass *cls, const void *data)
 
     moncls->vprintf = monitor_hmp_vprintf;
     moncls->accept_input = monitor_hmp_accept_input;
+
+    ucc->complete = monitor_hmp_complete;
+    ucc->prepare_delete = monitor_hmp_prepare_delete;
 }
 
 static void monitor_hmp_init(Object *obj)
@@ -1604,42 +1611,53 @@ static void monitor_readline_flush(void *opaque)
 
 void monitor_new_hmp(const char *chardev_id, bool use_readline, Error **errp)
 {
-    ERRP_GUARD();
-    MonitorHMP *mon;
     static int counter;
     g_autofree char *id = g_strdup_printf("hmpcompat%d", counter++);
-    Object *obj = object_new_with_props(TYPE_MONITOR_HMP,
-                                        object_get_objects_root(),
-                                        id,
-                                        errp,
-                                        "chardev", chardev_id,
-                                        "readline", use_readline ? "yes" : "no",
-                                        NULL);
-
-    if (!obj) {
-        return;
-    }
+    object_new_with_props(TYPE_MONITOR_HMP,
+                          object_get_objects_root(),
+                          id,
+                          errp,
+                          "chardev", chardev_id,
+                          "readline", use_readline ? "yes" : "no",
+                          NULL);
+}
 
-    mon = MONITOR_HMP(obj);
+static void monitor_hmp_complete(UserCreatable *uc, Error **errp)
+{
+    MonitorHMP *mon = MONITOR_HMP(uc);
+    UserCreatableClass *ucc_parent =
+        USER_CREATABLE_CLASS(
+            object_class_get_parent(
+                OBJECT_CLASS(MONITOR_HMP_GET_CLASS(mon))));
+    ERRP_GUARD();
 
-    monitor_complete(MONITOR(mon), errp);
+    ucc_parent->complete(uc, errp);
     if (*errp) {
-        object_unparent(OBJECT(mon));
         return;
     }
 
-    if (mon->use_readline) {
-        mon->rs = readline_init(monitor_readline_printf,
-                                monitor_readline_flush,
-                                mon,
-                                monitor_find_completion);
-        monitor_read_command(mon, 0);
+    if (mon->parent_obj.chardev_id) {
+        if (mon->use_readline) {
+            mon->rs = readline_init(monitor_readline_printf,
+                                    monitor_readline_flush,
+                                    mon,
+                                    monitor_find_completion);
+            monitor_read_command(mon, 0);
+        }
+
+        qemu_chr_fe_set_handlers(&mon->parent_obj.chr,
+                                 monitor_can_read,
+                                 monitor_read,
+                                 monitor_event, NULL,
+                                 &mon->parent_obj, NULL, true);
+        monitor_list_append(&mon->parent_obj);
     }
+}
 
-    qemu_chr_fe_set_handlers(&mon->parent_obj.chr,
-                             monitor_can_read, monitor_read, monitor_event,
-                             NULL, &mon->parent_obj, NULL, true);
-    monitor_list_append(&mon->parent_obj);
+static bool monitor_hmp_prepare_delete(UserCreatable *uc, Error **errp)
+{
+    error_setg(errp, "Deleting HMP monitors is not supported");
+    return false;
 }
 
 /**
monitor/monitor-internal.h +0 -1
index fe5703af6d..012d442a20 100644
@@ -204,7 +204,6 @@ extern QmpCommandList qmp_commands, qmp_cap_negotiation_commands;
 extern QemuMutex monitor_lock;
 extern MonitorList mon_list;
 
-void monitor_complete(Monitor *mon, Error **errp);
 bool monitor_requires_iothread(const Monitor *mon);
 int monitor_can_read(void *opaque);
 void monitor_list_append(Monitor *mon);
monitor/monitor.c +12 -2
index f88364a574..bf5e8a184a 100644
@@ -29,6 +29,7 @@
 #include "qapi/qapi-emit-events.h"
 #include "qapi/qapi-visit-control.h"
 #include "qobject/qdict.h"
+#include "qom/object_interfaces.h"
 #include "qemu/error-report.h"
 #include "qemu/option.h"
 #include "system/qtest.h"
@@ -73,7 +74,8 @@ static GHashTable *coroutine_mon; /* Maps Coroutine* to Monitor* */
 MonitorList mon_list;
 static bool monitor_destroyed;
 
-OBJECT_DEFINE_TYPE(Monitor, monitor, MONITOR, OBJECT);
+OBJECT_DEFINE_TYPE_EXTENDED(Monitor, monitor, MONITOR, OBJECT, true,
+                            { TYPE_USER_CREATABLE }, {});
 
 static void monitor_finalize(Object *obj)
 {
@@ -101,11 +103,17 @@ static void monitor_set_chardev_id(Object *obj, const char *str, Error **errp)
     mon->chardev_id = g_strdup(str);
 }
 
+static void monitor_complete(UserCreatable *uc, Error **errp);
+
 static void monitor_class_init(ObjectClass *cls, const void *data)
 {
+    UserCreatableClass *ucc = USER_CREATABLE_CLASS(cls);
+
     object_class_property_add_str(cls, "chardev",
                                   monitor_get_chardev_id,
                                   monitor_set_chardev_id);
+
+    ucc->complete = monitor_complete;
 }
 
 static void monitor_init(Object *obj)
@@ -680,8 +688,10 @@ void monitor_init_globals(void)
     aio_co_schedule(iohandler_get_aio_context(), qmp_dispatcher_co);
 }
 
-void monitor_complete(Monitor *mon, Error **errp)
+static void monitor_complete(UserCreatable *uc, Error **errp)
 {
+    Monitor *mon = MONITOR(uc);
+
     if (mon->chardev_id) {
         Chardev *chr = qemu_chr_find(mon->chardev_id);
         if (chr == NULL) {
monitor/qmp.c +31 -15

81 lines not on this page · raw message

qapi/qom.json +43 -0

61 lines not on this page · raw message

stubs/monitor-internal.c +1 -0
index 51db7588b9..59ccc4b35c 100644
@@ -10,4 +10,5 @@ int monitor_get_fd(Monitor *mon, const char *name, Error **errp)
 
 void monitor_new_hmp(const char *chardev_id, bool use_readline, Error **errp)
 {
+    g_assert_not_reached();
 }
system/vl.c +7 -1
index fb36140aac..79ed2be449 100644
@@ -1830,6 +1830,10 @@ static void object_option_add_visitor(Visitor *v)
 {
     ObjectOption *opt = g_new0(ObjectOption, 1);
     visit_type_ObjectOptions(v, NULL, &opt->opts, &error_fatal);
+    if (opt->opts->qom_type == OBJECT_TYPE_MONITOR_HMP ||
+        opt->opts->qom_type == OBJECT_TYPE_MONITOR_QMP) {
+        default_monitor = 0;
+    }
     QTAILQ_INSERT_TAIL(&object_opts, opt, next);
 }
 
@@ -1971,7 +1975,9 @@ static bool object_create_early(const char *type)
 
     /* Reason: property "chardev" */
     if (g_str_equal(type, "rng-egd") ||
-        g_str_equal(type, "qtest")) {
+        g_str_equal(type, "qtest") ||
+        g_str_equal(type, "monitor-hmp") ||
+        g_str_equal(type, "monitor-qmp")) {
         return false;
     }
 
Signature · 1 line
2.54.0
#58 lore raw

PATCH v5 34/35 qemu-options: document new monitor-hmp and monitor-qmp objects

In reply to Daniel P. Berrangé

Add new docs for the -object monitor-hmp and -object monitor-qmp options, updating -mon to state that it is legacy syntax sugar for the new -object args.

---
 qemu-options.hx | 49 +++++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 49 insertions(+)
qemu-options.hx +49 -0
index c799286153..a804d5b880 100644
@@ -4980,6 +4980,16 @@ SRST
       -mon chardev=mon1,mode=control,pretty=on
 
     enables the QMP monitor on localhost port 4444 with pretty-printing.
+
+    The use of ``-mon mode=readline`` is historical syntax sugar
+    for the new ``-object monitor-hmp`` option, each use of which
+    creates an object with the ID ``hmpcompatNNN`` where ``NNN`` is
+    a counter starting from 0.
+
+    The use of ``-mon mode=control`` is historical syntax sugar
+    for the new ``-object monitor-qmp`` option, each use of which
+    creates an object with the ID ``qmpcompatNNN`` where ``NNN`` is
+    a counter starting from 0.
 ERST
 
 DEF("debugcon", HAS_ARG, QEMU_OPTION_debugcon, \
@@ -5730,6 +5740,45 @@ SRST
     they are specified. Note that the 'id' property must be set. These
     objects are placed in the '/objects' path.
 
+    ``-object monitor-hmp,id=id,chardev=chardev_id,readline=on|off``
+        Set up a monitor running the Human Monitor Protocol,
+        connected to the chardev ``chardev_id``.
+
+        The ``id`` parameter is a unique ID that can be used
+        to dynamically delete the monitor at runtime. Note
+        that monitors created using the historical syntax
+        will be allocated IDs following the pattern ``hmpcompatNN``.
+        Mixing ``-object`` with the historical monitor syntax is
+        discouraged.
+
+        The ``readline`` parameter, which defaults to ``on``,
+        controls whether the monitor provides line editing.
+
+    ``-object monitor-qmp,id=id,chardev=chardev_id,pretty=on|off,close-action=none|delete``
+        Set up a monitor running the QEMU Monitor Protocol,
+        connected to the chardev ``chardev_id``.
+
+        The ``id`` parameter is a unique ID that can be used
+        to dynamically delete the monitor at runtime. Note
+        that monitors created using the historical syntax
+        will be allocated IDs following the pattern ``qmpcompatNN``.
+        Mixing ``-object`` with the historical monitor syntax is
+        discouraged.
+
+        The ``pretty`` parameter, which defaults to ``off``,
+        controls whether the monitor responses are pretty
+        printed as multi-line indented JSON, as opposed to
+        constrained to a single line without extraneous
+        whitespace.
+
+        The ``close-action`` parameter, which defaults to ``none``,
+        controls what happens when the connection to the monitor
+        is terminated by the user. If set to ``delete``, then the
+        ``monitor-qmp`` object and its associated character
+        device are both immediately deleted. This can be useful
+        if an extra monitor was hotplugged for a specific task
+        and should be unplugged when completed.
+
     ``-object memory-backend-file,id=id,size=size,mem-path=dir,share=on|off,discard-data=on|off,merge=on|off,dump=on|off,prealloc=on|off,host-nodes=host-nodes,policy=default|preferred|bind|interleave,align=align,offset=offset,readonly=on|off,rom=on|off|auto``
         Creates a memory file backend object, which can be used to back
         the guest RAM with huge pages.
Signature · 1 line
2.54.0
#40 lore raw

PATCH v5 29/35 tests/functional: add a stress test for monitor hot unplug

In reply to Daniel P. Berrangé

When unplugging a monitor there is a careful synchronization dance between the monitor handling the "object-del" command and the command processing for the monitor being deleted.

The stress test runs a busy loop of 'query-qmp-schema' on a second monitor, while the primary monitor requests its deletion.

---
 .../generic/test_monitor_hotplug.py           | 62 +++++++++++++++++++
 1 file changed, 62 insertions(+)
tests/functional/generic/test_monitor_hotplug.py +62 -0
index 5d8a159eb0..55af8a238f 100755
@@ -6,11 +6,16 @@
 #
 # Copyright (c) 2026 Christian Brauner
 
+import asyncio
 import os
+import random
+import threading
+import time
 
 from qemu_test import QemuSystemTest
 
 from qemu.qmp.legacy import QEMUMonitorProtocol
+from qemu.qmp import QMPClient
 
 
 class MonitorHotplug(QemuSystemTest):
@@ -163,6 +168,63 @@ def test_events_after_negotiation(self):
         qmp.close()
         self._remove_monitor()
 
+    def stress_mon(self, sock):
+        async def main():
+            qmp = QMPClient('testvm')
+            await qmp.connect(sock)
+            # Run query-version in a tight loop so that the
+            # monitor thread/dispatcher is very busy at the
+            # time we try to delete the monitor
+            while True:
+                try:
+                    # A command which returns a lot of data to make
+                    # it more likely we're in the I/O reply path
+                    # when deleting the monitor
+                    res = await qmp.execute('query-qmp-schema')
+                    # Some commands which generate async events
+                    # as those can trigger different code paths
+                    res = await qmp.execute('stop')
+                    res = await qmp.execute('cont')
+                except:
+                    # we'll get here if the monitor is terminated
+                    # by QEMU in which case we must disconnect
+                    # out side, but....
+                    try:
+                        await qmp.disconnect()
+                    except (ConnectionResetError, EOFError, BrokenPipeError):
+                        # ... disconnect() will probably see
+                        # errors too, but we must try to call it
+                        # regardless to cleanup asyncio state
+                        # and prevent python warnings at GC time
+                        pass
+                    return
+        asyncio.run(main())
+
+    def test_hotplug_stress(self):
+        """
+        Repeatedly hotplug and unplug a monitor, while another thread
+        concurrently issues commands on that monitor. This stresses
+        the synchronization with the monitor thread during cleanup
+        """
+        self.set_machine('none')
+        self.vm.add_args('-nodefaults')
+        self.vm.launch()
+
+        # Each loop sleeps at most 0.5 seconds, so this should
+        # give an upper bound of approx 5 seconds execution
+        # time which is reasonable to run by default
+        repeat = 10
+        for i in range(repeat):
+            # First cycle
+            sock = self._add_monitor()
+            print ("# stress cycle %02d/%02d" % (i, repeat))
+            stress = threading.Thread(target=self.stress_mon, args=[sock])
+            stress.start()
+            # Sleep upto 1/2 second to vary the races
+            time.sleep(random.random() / 2)
+            self._remove_monitor()
+            stress.join()
+
 
 if __name__ == '__main__':
     QemuSystemTest.main()
Signature · 1 line
2.54.0
#44 lore raw

PATCH v5 31/35 qom: add trace events for user creatable create/delete APIs

In reply to Daniel P. Berrangé
---
 qom/object_interfaces.c | 6 ++++++
 qom/trace-events        | 5 +++++
 2 files changed, 11 insertions(+)
qom/object_interfaces.c +6 -0
index 6faa0b2fd9..2e618a7623 100644
@@ -20,12 +20,14 @@
 #include "qapi/opts-visitor.h"
 #include "qemu/config-file.h"
 #include "qemu/keyval.h"
+#include "trace.h"
 
 bool user_creatable_complete(UserCreatable *uc, Error **errp)
 {
     UserCreatableClass *ucc = USER_CREATABLE_GET_CLASS(uc);
     ERRP_GUARD();
 
+    trace_user_creatable_complete(uc, object_get_typename(OBJECT(uc)));
     if (ucc->complete) {
         ucc->complete(uc, errp);
     }
@@ -37,9 +39,13 @@ bool user_creatable_prepare_delete(UserCreatable *uc, Error **errp)
     UserCreatableClass *ucc = USER_CREATABLE_GET_CLASS(uc);
     ERRP_GUARD();
 
+    trace_user_creatable_prepare_delete(uc, object_get_typename(OBJECT(uc)));
     if (ucc->prepare_delete) {
         ucc->prepare_delete(uc, errp);
     }
+    trace_user_creatable_prepare_delete_result(
+        uc, object_get_typename(OBJECT(uc)),
+        *errp ? error_get_pretty(*errp) : NULL);
     return !*errp;
 }
 
qom/trace-events +5 -0
index 44c63e72af..bac8472618 100644
@@ -11,3 +11,8 @@ object_property_del_child(void *obj, const char *type, void *child, const char *
 object_property_parse(void *obj, const char *type, const char *name, const char *value) "obj=%p type=%s prop=%s value=%s"
 object_class_dynamic_cast_assert(const char *type, const char *target, const char *file, int line, const char *func) "type=%s->%s (%s:%d:%s)"
 object_class_property_add(const char *type, const char *name, void *value) "type=%s name=%s value=%p"
+
+# object_interfaces.c
+user_creatable_complete(void *obj, const char *type) "obj=%p type=%s"
+user_creatable_prepare_delete(void *obj, const char *type) "obj=%p type=%s"
+user_creatable_prepare_delete_result(void *obj, const char *type, const char *msg) "obj=%p type=%s msg=%s"
Signature · 1 line
2.54.0
#38 lore raw

PATCH v5 27/35 tests/qtest: add tests for dynamic monitor add/remove

In reply to Daniel P. Berrangé

From: Christian Brauner <[email protected]>

Test the object-add/object-del QMP commands with the monitor-qmp object type.

- Basic lifecycle: chardev-add -> object-add -> object-del -> chardev-remove
- Error: object-add with nonexistent chardev
- Error: second monitor on same chardev (chardev already in use)
- Removal of CLI-created QMP monitor succeeds
- Error: object-remove on HMP monitor
- Re-add after remove: same id and chardev reusable after removal

[DB: modified to use object-add/object-del, removing redundant
scenarios already handled by object-add/del code]

---
 tests/qtest/qmp-test.c | 174 +++++++++++++++++++++++++++++++++++++++++
 1 file changed, 174 insertions(+)
tests/qtest/qmp-test.c +174 -0
index edf0886787..3a325a04e0 100644
@@ -337,6 +337,174 @@ static void test_qmp_missing_any_arg(void)
     qtest_quit(qts);
 }
 
+static void test_qmp_monitor_add_remove(void)
+{
+    QTestState *qts;
+    QDict *resp;
+
+    qts = qtest_init(common_args);
+
+    /* Create a null chardev for the dynamic monitor */
+    resp = qtest_qmp(qts,
+                     "{'execute': 'chardev-add',"
+                     " 'arguments': {'id': 'monitor-chardev',"
+                     "               'backend': {'type': 'null',"
+                     "                           'data': {}}}}");
+    g_assert(qdict_haskey(resp, "return"));
+    qobject_unref(resp);
+
+    /* Add a dynamic monitor */
+    resp = qtest_qmp(qts,
+                     "{'execute': 'object-add',"
+                     " 'arguments': {'qom-type': 'monitor-qmp',"
+                     "               'id': 'dyn-mon',"
+                     "               'chardev': 'monitor-chardev'}}");
+    g_assert(qdict_haskey(resp, "return"));
+    qobject_unref(resp);
+
+    /* Remove the dynamic monitor */
+    resp = qtest_qmp(qts,
+                     "{'execute': 'object-del',"
+                     " 'arguments': {'id': 'dyn-mon'}}");
+    g_assert(qdict_haskey(resp, "return"));
+    qobject_unref(resp);
+
+    /* Add again after remove -- same id and chardev should work */
+    resp = qtest_qmp(qts,
+                     "{'execute': 'object-add',"
+                     " 'arguments': {'qom-type': 'monitor-qmp',"
+                     "               'id': 'dyn-mon',"
+                     "               'chardev': 'monitor-chardev'}}");
+    g_assert(qdict_haskey(resp, "return"));
+    qobject_unref(resp);
+
+    /* Clean up */
+    resp = qtest_qmp(qts,
+                     "{'execute': 'object-del',"
+                     " 'arguments': {'id': 'dyn-mon'}}");
+    g_assert(qdict_haskey(resp, "return"));
+    qobject_unref(resp);
+
+    resp = qtest_qmp(qts,
+                     "{'execute': 'chardev-remove',"
+                     " 'arguments': {'id': 'monitor-chardev'}}");
+    g_assert(qdict_haskey(resp, "return"));
+    qobject_unref(resp);
+
+    qtest_quit(qts);
+}
+
+static void test_qmp_monitor_error_paths(void)
+{
+    QTestState *qts;
+    QDict *resp;
+
+    qts = qtest_init(common_args);
+
+    /* Error: chardev does not exist */
+    resp = qtest_qmp(qts,
+                     "{'execute': 'object-add',"
+                     " 'arguments': {'qom-type': 'monitor-qmp',"
+                     "               'id': 'bad-mon',"
+                     "               'chardev': 'nonexistent'}}");
+    qmp_expect_error_and_unref(resp, "GenericError");
+
+    qtest_quit(qts);
+}
+
+static void test_qmp_monitor_chardev_in_use(void)
+{
+    QTestState *qts;
+    QDict *resp;
+
+    qts = qtest_init(common_args);
+
+    /* Create a null chardev */
+    resp = qtest_qmp(qts,
+                     "{'execute': 'chardev-add',"
+                     " 'arguments': {'id': 'shared-chr',"
+                     "               'backend': {'type': 'null',"
+                     "                           'data': {}}}}");
+    g_assert(qdict_haskey(resp, "return"));
+    qobject_unref(resp);
+
+    /* Attach first monitor */
+    resp = qtest_qmp(qts,
+                     "{'execute': 'object-add',"
+                     " 'arguments': {'qom-type': 'monitor-qmp',"
+                     "               'id': 'mon-1',"
+                     "               'chardev': 'shared-chr'}}");
+    g_assert(qdict_haskey(resp, "return"));
+    qobject_unref(resp);
+
+    /* Error: second monitor on the same chardev */
+    resp = qtest_qmp(qts,
+                     "{'execute': 'object-add',"
+                     " 'arguments': {'qom-type': 'monitor-qmp',"
+                     "               'id': 'mon-2',"
+                     "               'chardev': 'shared-chr'}}");
+    qmp_expect_error_and_unref(resp, "GenericError");
+
+    /* Clean up */
+    resp = qtest_qmp(qts,
+                     "{'execute': 'object-del',"
+                     " 'arguments': {'id': 'mon-1'}}");
+    g_assert(qdict_haskey(resp, "return"));
+    qobject_unref(resp);
+
+    resp = qtest_qmp(qts,
+                     "{'execute': 'chardev-remove',"
+                     " 'arguments': {'id': 'shared-chr'}}");
+    g_assert(qdict_haskey(resp, "return"));
+    qobject_unref(resp);
+
+    qtest_quit(qts);
+}
+
+static void test_qmp_monitor_remove_cli(void)
+{
+    QTestState *qts;
+    QDict *resp;
+
+    /* Launch with a named CLI monitor on a null chardev */
+    qts = qtest_initf("%s -chardev null,id=cli-chr"
+                      " -object monitor-qmp,id=cli-mon,chardev=cli-chr",
+                      common_args);
+
+    /* CLI-created QMP monitors can be removed */
+    resp = qtest_qmp(qts,
+                     "{'execute': 'object-del',"
+                     " 'arguments': {'id': 'cli-mon'}}");
+    g_assert(qdict_haskey(resp, "return"));
+    qobject_unref(resp);
+
+    resp = qtest_qmp(qts,
+                     "{'execute': 'chardev-remove',"
+                     " 'arguments': {'id': 'cli-chr'}}");
+    g_assert(qdict_haskey(resp, "return"));
+    qobject_unref(resp);
+
+    qtest_quit(qts);
+}
+
+static void test_qmp_monitor_remove_hmp(void)
+{
+    QTestState *qts;
+    QDict *resp;
+
+    qts = qtest_initf("%s -chardev null,id=hmp-chr"
+                      " -object monitor-hmp,id=hmp-mon,chardev=hmp-chr",
+                      common_args);
+
+    /* Error: object_del must reject HMP monitors */
+    resp = qtest_qmp(qts,
+                     "{'execute': 'object-del',"
+                     " 'arguments': {'id': 'hmp-mon'}}");
+    qmp_expect_error_and_unref(resp, "GenericError");
+
+    qtest_quit(qts);
+}
+
 int main(int argc, char *argv[])
 {
     g_test_init(&argc, &argv, NULL);
@@ -348,6 +516,12 @@ int main(int argc, char *argv[])
 #endif
     qtest_add_func("qmp/preconfig", test_qmp_preconfig);
     qtest_add_func("qmp/missing-any-arg", test_qmp_missing_any_arg);
+    qtest_add_func("qmp/monitor-add-remove", test_qmp_monitor_add_remove);
+    qtest_add_func("qmp/monitor-error-paths", test_qmp_monitor_error_paths);
+    qtest_add_func("qmp/monitor-chardev-in-use",
+                    test_qmp_monitor_chardev_in_use);
+    qtest_add_func("qmp/monitor-remove-cli", test_qmp_monitor_remove_cli);
+    qtest_add_func("qmp/monitor-remove-hmp", test_qmp_monitor_remove_hmp);
 
     return g_test_run();
 }
Signature · 1 line
2.54.0
#27 lore raw

PATCH v5 17/35 monitor: drop unused monitor_cur_is_qmp

In reply to Daniel P. Berrangé

The previous patch dropped the only remaining use of monitor_cur_is_qmp.

---
 include/monitor/monitor.h      |  1 -
 monitor/monitor.c              | 10 ----------
 stubs/monitor-core.c           |  5 -----
 tests/unit/test-util-sockets.c |  1 -
 tools/qemu-vnc/stubs.c         |  5 -----
 5 files changed, 22 deletions(-)
include/monitor/monitor.h +0 -1
index 8cb3db677b..3f730cffee 100644
@@ -25,7 +25,6 @@ extern QemuOptsList qemu_mon_opts;
 
 Monitor *monitor_cur(void);
 Monitor *monitor_set_cur(Coroutine *co, Monitor *mon);
-bool monitor_cur_is_qmp(void);
 
 void monitor_init_globals(void);
 void monitor_init_globals_core(void);
monitor/monitor.c +0 -10
index b788669f1f..0db7903795 100644
@@ -150,16 +150,6 @@ Monitor *monitor_set_cur(Coroutine *co, Monitor *mon)
     return old_monitor;
 }
 
-/**
- * Is the current monitor, if any, a QMP monitor?
- */
-bool monitor_cur_is_qmp(void)
-{
-    Monitor *cur_mon = monitor_cur();
-
-    return cur_mon && monitor_is_qmp(cur_mon);
-}
-
 bool monitor_requires_iothread(const Monitor *mon)
 {
     MonitorClass *cls = MONITOR_GET_CLASS(mon);
stubs/monitor-core.c +0 -5
index 078a5012e9..a7c32297c9 100644
@@ -7,11 +7,6 @@ Monitor *monitor_cur(void)
     return NULL;
 }
 
-bool monitor_cur_is_qmp(void)
-{
-    return false;
-}
-
 Monitor *monitor_set_cur(Coroutine *co, Monitor *mon)
 {
     return NULL;
tests/unit/test-util-sockets.c +0 -1
index b9f2453e29..ee66d727c3 100644
@@ -74,7 +74,6 @@ int monitor_get_fd(Monitor *mon, const char *fdname, Error **errp)
 Monitor *monitor_cur(void) { return cur_mon; }
 Monitor *monitor_set_cur(Coroutine *co, Monitor *mon) { abort(); }
 int monitor_vprintf(Monitor *mon, const char *fmt, va_list ap) { abort(); }
-bool monitor_cur_is_qmp(void) { abort(); };
 
 #ifndef _WIN32
 static void test_socket_fd_pass_name_good(void)
tools/qemu-vnc/stubs.c +0 -5
index a865ce85f0..1c82d8cff4 100644
@@ -36,11 +36,6 @@ Monitor *monitor_cur(void)
     return NULL;
 }
 
-bool monitor_cur_is_qmp(void)
-{
-    return false;
-}
-
 Monitor *monitor_set_cur(Coroutine *co, Monitor *mon)
 {
     return NULL;
Signature · 1 line
2.54.0
#28 lore raw

PATCH v5 18/35 monitor: use dynamic cast in QMP commands

In reply to Daniel P. Berrangé

Rather than asserting monitor_is_qmp(), use a QOM cast via MONITOR_QMP which performs an assert already.

---
 monitor/qmp-cmds-control.c | 12 ++----------
 1 file changed, 2 insertions(+), 10 deletions(-)
monitor/qmp-cmds-control.c +2 -10
index af6a2a118b..8fe0876a3d 100644
@@ -72,11 +72,7 @@ static bool qmp_caps_accept(MonitorQMP *mon, QMPCapabilityList *list,
 void qmp_qmp_capabilities(bool has_enable, QMPCapabilityList *enable,
                           Error **errp)
 {
-    Monitor *cur_mon = monitor_cur();
-    MonitorQMP *mon;
-
-    assert(monitor_is_qmp(cur_mon));
-    mon = container_of(cur_mon, MonitorQMP, parent_obj);
+    MonitorQMP *mon = MONITOR_QMP(monitor_cur());
 
     if (mon->commands == &qmp_commands) {
         error_set(errp, ERROR_CLASS_COMMAND_NOT_FOUND,
@@ -122,11 +118,7 @@ static void query_commands_cb(const QmpCommand *cmd, void *opaque)
 CommandInfoList *qmp_query_commands(Error **errp)
 {
     CommandInfoList *list = NULL;
-    Monitor *cur_mon = monitor_cur();
-    MonitorQMP *mon;
-
-    assert(monitor_is_qmp(cur_mon));
-    mon = container_of(cur_mon, MonitorQMP, parent_obj);
+    MonitorQMP *mon = MONITOR_QMP(monitor_cur());
 
     qmp_for_each_command(mon->commands, query_commands_cb, &list);
 
Signature · 1 line
2.54.0
Reviewed-by on v5 32/35
#46 lore raw

Re: [PATCH v5 32/35] monitor: add support for auto-deleting monitors upon close

In reply to Daniel P. Berrangé
Quoted 44 lines

The default monitor is usually a long lived object that will exist for the entire lifetime of the VM. A monitor can only service a single client at a time though, and so it might be desirable to hotplug additional monitors at runtime for specific tasks. If doing that, however, there is a need to remove the monitor when it is no longer needed.

Allowing a client to run "object-del" against its own monitor adds complex edge cases, as it would be desirable to send the QMP response despite the monitor sending it being deleted. Doing "object-del" alone will also result in orphaning a character device backend instance, as there is no opportunity to run the companion "chardev-del" command.

A simpler way to ensure cleanup is to add the concept of auto-deleting monitor objects. Specifically when the "CHR_EVENT_CLOSED" event is emitted, the equivalent of "object-del" + "chardev-del" can be run internally. Since the transient client has already droppped its monitor connection, there is no synchronization to be concerned about.

This is implemented via a new "close-action=none|delete" property on the 'monitor-qmp' object. This concept could be extended with further actions in future, for example:

* close-action=shutdown - graceful guest shutdown
* close-action=terminate - immediate guest poweroff
* close-action=stop - pause guest CPUs while the monitor is not connected to any client

This is left as an exercise for future interested contributors.

Message-ID: <[email protected]>

qapi/qom.json +2 -0
index 6affb70a594..b4c49a2e8b9 100644
@@ -1213,18 +1213,37 @@
   'base': 'MonitorProperties',
   'data': { '*readline': 'bool' } }
 
+
+##

qapi/qom.json line 1227

Quoted 10 lines
qapi/qom.json +10 -0
+# @MonitorQMPCloseAction:
+#
+# Action to take when the character device backend is
+# closed.
+#
+# @none: take no action (the default)
+# @delete: delete both the 'monitor-qmp' object and its associated
+#          character device backend object
+#
+# Since 11.1

should be "Since: 11.1" (worth a new check perhaps)

qapi/qom.json line 1229

Quoted 2 lines
qapi/qom.json +2 -0
+##
+{ 'enum' : 'MonitorQMPCloseAction',

extra space, vs the usual '{ 'enum':'.

Quoted 260 lines
qapi/qom.json +6 -1
+  'data': ['none', 'delete'] }
+
 ##
 # @MonitorQMPProperties:
 #
 # Properties for the QMP monitor
 #
 # @pretty: whether to pretty print JSON responses (default: disabled)
+# @close-action: action to take when the character device backend
+#                is closed (default: none)
 #
 # Since: 11.1
 ##
 { 'struct': 'MonitorQMPProperties',
   'base': 'MonitorProperties',
-  'data': { '*pretty': 'bool' } }
+  'data': { '*pretty': 'bool',
+            '*close-action': 'MonitorQMPCloseAction' } }
 
 ##
 # @ObjectType:
monitor/monitor-internal.h +3 -0
index 5522e05464b..23829f32f9a 100644
@@ -28,6 +28,7 @@
 #include "chardev/char-fe.h"
 #include "monitor/monitor.h"
 #include "qapi/qapi-types-control.h"
+#include "qapi/qapi-types-qom.h"
 #include "qapi/qmp-registry.h"
 #include "qobject/json-parser.h"
 #include "qemu/readline.h"
@@ -178,7 +179,9 @@ struct MonitorQMP {
     Monitor parent_obj;
     JSONMessageParser parser;
     bool pretty;
+    MonitorQMPCloseAction close_action;
     bool setup_pending; /* iothread BH has not yet set up chardev handlers */
+    bool delete_pending; /* close_action has started 'delete' process */
     /*
      * When a client connects, we're in capabilities negotiation mode.
      * @commands is &qmp_cap_negotiation_commands then.  When command
monitor/qmp.c +75 -0
index 5301927f09f..1057b0d12ba 100644
@@ -28,6 +28,7 @@
 #include "monitor-internal.h"
 #include "qapi/error.h"
 #include "qapi/qapi-commands-control.h"
+#include "qapi/qapi-commands-char.h"
 #include "qobject/qdict.h"
 #include "qobject/qjson.h"
 #include "qobject/qlist.h"
@@ -103,6 +104,20 @@ static void monitor_qmp_set_pretty(Object *obj, bool val, Error **errp)
     mon->pretty = val;
 }
 
+static int monitor_qmp_get_close_action(Object *obj, Error **errp)
+{
+    MonitorQMP *mon = MONITOR_QMP(obj);
+
+    return mon->close_action;
+}
+
+static void monitor_qmp_set_close_action(Object *obj, int val, Error **errp)
+{
+    MonitorQMP *mon = MONITOR_QMP(obj);
+
+    mon->close_action = val;
+}
+
 static void monitor_qmp_emit_event(Monitor *mon, QAPIEvent event, QDict *qdict);
 static bool monitor_qmp_requires_iothread(const Monitor *mon);
 static void monitor_qmp_complete(UserCreatable *uc, Error **errp);
@@ -117,6 +132,11 @@ static void monitor_qmp_class_init(ObjectClass *cls, const void *data)
     object_class_property_add_bool(cls, "pretty",
                                    monitor_qmp_get_pretty,
                                    monitor_qmp_set_pretty);
+    object_class_property_add_enum(cls, "close-action",
+                                   "MonitorQMPCloseAction",
+                                   &MonitorQMPCloseAction_lookup,
+                                   monitor_qmp_get_close_action,
+                                   monitor_qmp_set_close_action);
 
     moncls->emit_event = monitor_qmp_emit_event;
     moncls->requires_iothread = monitor_qmp_requires_iothread;
@@ -550,11 +570,45 @@ static QDict *qmp_greeting(MonitorQMP *mon)
         ver, cap_list);
 }
 
+static void monitor_qmp_self_delete_bh(void *opaque)
+{
+    MonitorQMP *mon = opaque;
+    g_autofree char *mon_id = object_property_get_child_name(
+        object_get_objects_root(), OBJECT(mon));
+    g_autofree char *chardev_id = g_strdup(mon->parent_obj.chardev_id);
+    Error *local_error = NULL;
+
+    if (!mon_id) {
+        /* Another monitor raced & ran 'object-del' on 'mon'
+         * before this BH got scheduled, so we have a ref on
+         * mon but it is already unparented.
+         */
+        object_unref(mon);
+        return;
+    }
+
+    user_creatable_del(mon_id, &local_error);
+    object_unref(mon);
+    if (local_error != NULL) {
+        error_report_err(local_error);
+    } else {
+        qmp_chardev_remove(chardev_id, NULL);
+    }
+}
+
 static void monitor_qmp_event(void *opaque, QEMUChrEvent event)
 {
     QDict *data;
     MonitorQMP *mon = opaque;
 
+    /* Protect against race if a client drops & quickly
+     * reconnects - we'll have the delete BH scheduled
+     * so must not honour a new open request
+     */
+    if (mon->delete_pending) {
+        return;
+    }
+
     switch (event) {
     case CHR_EVENT_OPENED:
         WITH_QEMU_LOCK_GUARD(&mon->parent_obj.mon_lock) {
@@ -577,6 +631,27 @@ static void monitor_qmp_event(void *opaque, QEMUChrEvent event)
         json_message_parser_init(&mon->parser, handle_qmp_command,
                                  mon, NULL);
         monitor_fdsets_cleanup();
+        switch (mon->close_action) {
+        case MONITOR_QMP_CLOSE_ACTION_NONE:
+            break; /* nada */
+        case MONITOR_QMP_CLOSE_ACTION_DELETE:
+            mon->delete_pending = true;
+            /*
+             * Do NOT run in the AIO context associated with the
+             * monitor. We need to run in the default AIO context
+             * which is the same context in which 'qmp_object_del'
+             * will execute
+             *
+             * Hold an extra ref in case a separate monitor races
+             * with the BH by processing an explicit 'object-del'
+             */
+            object_ref(mon);
+            aio_bh_schedule_oneshot(qemu_get_aio_context(),
+                                    monitor_qmp_self_delete_bh, mon);
+            break;
+        default:
+            g_assert_not_reached();
+        }
         break;
     case CHR_EVENT_BREAK:
     case CHR_EVENT_MUX_IN:
tests/functional/generic/test_monitor_hotplug.py +72 -4
index 55af8a238f9..86092e781c9 100755
@@ -25,7 +25,7 @@ def setUp(self):
         sock_dir = self.socket_dir()
         self._sock_path = os.path.join(sock_dir.name, 'hotplug.sock')
 
-    def _add_monitor(self):
+    def _add_monitor(self, autodelete=False):
         """Create a chardev + monitor and return the socket path."""
         sock = self._sock_path
         self.vm.cmd('chardev-add', id='hotplug-chr', backend={
@@ -39,9 +39,15 @@ def _add_monitor(self):
                 'wait': False,
             }
         })
-        self.vm.cmd('object-add', id='hotplug-mon',
-                    qom_type='monitor-qmp',
-                    chardev='hotplug-chr')
+        if autodelete:
+            self.vm.cmd('object-add', id='hotplug-mon',
+                        qom_type='monitor-qmp',
+                        chardev='hotplug-chr',
+                        close_action='delete')
+        else:
+            self.vm.cmd('object-add', id='hotplug-mon',
+                        qom_type='monitor-qmp',
+                        chardev='hotplug-chr')
         return sock
 
     def _remove_monitor(self):
@@ -118,6 +124,75 @@ def test_self_removal(self):
         # Clean up the chardev
         self.vm.cmd('chardev-remove', id='hotplug-chr')
 
+    def test_auto_delete(self):
+        """
+        A dynamically-added monitor configured with 'close-action=delete'
+        should see itself deleted when the client is closed.
+        """
+        self.set_machine('none')
+        self.vm.add_args('-nodefaults')
+        self.vm.launch()
+
+        sock = self._add_monitor(autodelete=True)
+
+        cdevs = [c["label"] for c in self.vm.cmd('query-chardev')]
+        objs = [o["name"] for o in self.vm.cmd('qom-list', path='/objects')]
+        assert ('hotplug-chr' in cdevs)
+        assert ('hotplug-mon' in objs)
+
+        qmp = QEMUMonitorProtocol(sock)
+        greeting = qmp.connect(negotiate=True)
+        self.assertIn('QMP', greeting)
+
+        cdevs = [c["label"] for c in self.vm.cmd('query-chardev')]
+        objs = [o["name"] for o in self.vm.cmd('qom-list', path='/objects')]
+        assert ('hotplug-chr' in cdevs)
+        assert ('hotplug-mon' in objs)
+
+        qmp.close()
+
+        # Wait upto 10 seconds max for chardev to auto-delete, which
+        # is hopefully enough for reliability under high load
+        for i in range(int(10 / 0.2)):
+            cdevs = [c["label"] for c in self.vm.cmd('query-chardev')]
+            if 'hotplug-chr' not in cdevs:
+                break
+            # Wait a little more then try again
+            time.sleep(0.2)
+
+        cdevs = [c["label"] for c in self.vm.cmd('query-chardev')]
+        objs = [o["name"] for o in self.vm.cmd('qom-list', path='/objects')]
+        assert ('hotplug-chr' not in cdevs)
+        assert ('hotplug-mon' not in objs)
+
+    def test_reconnect(self):
+        """
+        A dynamically-added monitor configured without 'close-action'
+        should allow reconnects after the client is closed.
+        """
+        self.set_machine('none')
+        self.vm.add_args('-nodefaults')
+        self.vm.launch()
+
+        sock = self._add_monitor()
+
+        qmp = QEMUMonitorProtocol(sock)
+        qmp.connect(negotiate=True)
+
+        resp = qmp.cmd_obj({'execute': 'query-chardev'})
+        self.assertIn('return', resp)
+
+        qmp.close()
+
+        qmp = QEMUMonitorProtocol(sock)
+        qmp.connect(negotiate=True)

tests/functional/generic/test_monitor_hotplug.py def test_self_removal(self): · line 198

Quoted 10 lines
tests/functional/generic/test_monitor_hotplug.py +7 -0
+
+        resp = qmp.cmd_obj({'execute': 'query-chardev'})
+        self.assertIn('return', resp)
+
+        qmp.close()
+        self._remove_monitor()
+
     def test_large_response(self):
         """
         Send a command with a large response (query-qmp-schema) on a
Signature · 1 line
Marc-André Lureau <[email protected]>
#63 lore raw
In reply to Daniel P. Berrangé

Markus,

This series is fully reviewed now by multiple people.

Do you have any plans to do reviews and/or queue this as maintainer, or will you ack it for me to send a pull request.

I would like to get this merged before soft freeze which is fast approaching.

Quoted 164 lines · Daniel P. Berrangé

Conceptually -object and object_add/object_del should be sufficient for essentially all QEMU configuration....if only we ported all our internal custom backends/devices/etc to QOM. That is of course a big job which is why it hasn't happened.

This series started with the premise that the monitor is one of the easier areas to convert since we have no more than three classes, a common base, and QMP and HMP subclasses[1]. So why not give it a go and thus unlock the ability to dynamically create/delete monitors in QMP/HMP.

This series does the conversion in a great many small steps to better understand the implications at each stage.

The high level outcome of this series is

* HMP and QMP monitors are QOM objects, 'monitor-hmp' and 'monitor-qmp' respectively

* Both can be cold plugged and hot plugged. QMP only, can also be hot unplugged.

* '-mon' is obsolete, deprecated and replaced by '-object', but -monitor, -qmp and kept as high level syntax sugar

* QMP gains a concept of "close-action" which makes it possible to mark a monitor for auto-delete.

The monitor hot-unplug code and the qtest and functional testing code is heavily derived from a series sent by Christian Brauner which proposed new monitor_add/monitor_del commands:

https://lists.nongnu.org/archive/html/qemu-devel/2026-04/msg01349.html

I left Christian's authorship & SoB on the patches which were derived from his code, though the code has been refactored quite a bit in places, so bugs are quite possibly my own.

Note that Christian's series allowed the use of "monitor_del" commands against the current monitor session. ie a client could delete the very monitor it was using. This is an awkward concept as it needs special casing to delay the deletion to happen in the background, such that that the QMP response to 'monitor_del' could still be sent back. This also left the chardev was orphaned as there's no way to run 'chardev_dev' in that usage pattern.

To provide an alternative mechanism to address the same use case, this series introduces the 'close-action' concept mentioned above, that allows hotplugging a monitor to service a specific task, with the monitor being purged when the script closes its connection. This avoids the special casing that an explicit "self deletion" paradigm required.

While supporting hotplug of HMP was trivial, I didn't do any work to think about hotunplug of HMP, since IMHO it is of limited value given HMP's typical use cases.

[1] ~~~ we did all this not because it was easy, but because we thought it would be easy ~~~

Changed in v5:

- Address another race when looking up monitor by ID
- Refer to "line editting" when talking about HMP readline feature
- Misc typos
- Fix mis-placed setting of 'delete_pending'
- Add test case to validate reconnecting to QMP

Changed in v4:

- Hold reference into BH to avoid race with auto-delete

Changed in v3:

- Removed unused 'dead' struct field
- Make use of 'setup_pending' struct field across BH
- Add missing free of chardev_id
- Add missnig ERRP_GUARD in monitor_new_qmp
- Misc docs typos / rephrasing
- Moved docs patch to the end
- Fixed docs about the default QOM ID naming for legacy monitor syntax
- Fixed random sleep time calculation in tests

Christian Brauner (6):
monitor: convert from oneshot BH to persistent BH
monitor: reject attempts to delete the current monitor
monitor: protect qemu_chr_fe_accept_input with monitor lock
monitor: implement support for deleting QMP objects tests/qtest: add tests for dynamic monitor add/remove tests/functional: add e2e test for dynamic QMP monitor hotplug

Daniel P. Berrangé (29):
qom: replace 'can_be_deleted' with 'prepare_delete'
monitor: replace 'common' with 'parent_obj' in MonitorHMP
monitor: replace 'common' with 'parent_obj' in MonitorQMP
monitor: rename monitor_init* to monitor_new*
monitor: minimal conversion of monitors to QOM
monitor: add 'chardev' property to Monitor base class
monitor: add 'readline' property to HMP Monitor class
monitor: add 'pretty' property to QMP Monitor class
monitor: remove 'skip_flush' field
monitor: move monitor_data_(init|destroy) into QOM init/finalize
monitor: use class methods for monitor_vprintf
monitor: use class methods for monitor_qapi_event_emit
monitor: use class methods for monitor_accept_input
monitor: use class method for I/O thread request
monitor: use dynamic cast in monitor_qmp_requests_pop_any_with_lock
util: use dynamic cast in error vreport
monitor: drop unused monitor_cur_is_qmp
monitor: use dynamic cast in QMP commands
monitor: use dynamic cast in monitor_is_hmp_non_interactive
monitor: drop unused monitor_is_qmp method
monitor: eliminate monitor_is_hmp_non_interactive method
monitor: implement "user creatable" interface for adding monitors tests/functional: add a stress test for monitor hot unplug
qom: add method for getting the "id" of a QOM object
qom: add trace events for user creatable create/delete APIs
monitor: add support for auto-deleting monitors upon close
tests: switch from -mon to -object monitor-qmp
qemu-options: document new monitor-hmp and monitor-qmp objects
docs: mark '-mon' as deprecated in favour of -object

 MAINTAINERS                                   |   1 +
 backends/cryptodev.c                          |  10 +-
 backends/hostmem.c                            |   5 +-
 backends/iommufd.c                            |  10 +-
 block/throttle-groups.c                       |  10 +-
 chardev/char.c                                |   3 +-
 docs/about/deprecated.rst                     |  10 +
 docs/devel/writing-monitor-commands.rst       |   4 +-
 docs/system/arm/xenpvh.rst                    |   4 +-
 docs/system/i386/xen.rst                      |   3 +-
 docs/system/i386/xenpvh.rst                   |   4 +-
 event-loop-base.c                             |   8 +-
 gdbstub/system.c                              |   4 +-
 include/monitor/monitor.h                     |  23 +-
 include/qom/object.h                          |  10 +
 include/qom/object_interfaces.h               |  26 +-
 include/system/event-loop-base.h              |   2 +-
 migration/migration-hmp-cmds.c                |   5 +-
 monitor/hmp-cmds.c                            |   7 +-
 monitor/hmp.c                                 | 181 ++++++++--
 monitor/monitor-internal.h                    |  73 ++--
 monitor/monitor.c                             | 262 +++++++-------
 monitor/qmp-cmds-control.c                    |  12 +-
 monitor/qmp-cmds.c                            |  14 +-
 monitor/qmp.c                                 | 319 +++++++++++++++---
 net/can/can_core.c                            |   5 +-
 python/qemu/machine/machine.py                |   4 +-
 qapi/qom.json                                 |  62 ++++
 qemu-options.hx                               |  56 ++-
 qom/object.c                                  |  17 +
 qom/object_interfaces.c                       |  20 +-
 qom/trace-events                              |   5 +
 storage-daemon/qemu-storage-daemon.c          |   2 +-
 stubs/monitor-core.c                          |   5 -
 stubs/monitor-internal.c                      |   3 +-
 system/vl.c                                   |  12 +-
 tests/functional/generic/meson.build          |   1 +
 .../generic/test_monitor_hotplug.py           | 305 +++++++++++++++++
 tests/qemu-iotests/245                        |   4 +-
 tests/qtest/libqtest.c                        |   2 +-
Quoted 12 lines
 tests/qtest/qmp-test.c                        | 174 ++++++++++
 tests/unit/test-util-sockets.c                |   1 -
 tools/qemu-vnc/stubs.c                        |   5 -
 ui/ui-hmp-cmds.c                              |   2 +-
 util/error-report.c                           |  13 +-
 util/main-loop.c                              |   5 +-
 46 files changed, 1386 insertions(+), 327 deletions(-)
 create mode 100755 tests/functional/generic/test_monitor_hotplug.py
Signature · 1 line
2.54.0

With regards,
Daniel

Signature · 3 lines
|: https://berrange.com       ~~        https://hachyderm.io/@berrange :|
|: https://libvirt.org          ~~          https://entangle-photo.org :|
|: https://pixelfed.art/berrange   ~~    https://fstop138.berrange.com :|
#64 lore raw
Quoted 9 lines · Daniel P. Berrangé

Markus,

This series is fully reviewed now by multiple people.

Do you have any plans to do reviews and/or queue this as maintainer, or will you ack it for me to send a pull request.

I would like to get this merged before soft freeze which is fast approaching.

Soft freeze is in a week, and the heat wave left me seriously sleep-deprived.

I'd like to give your interface changes an eye-over. I probably won't be able to do more in time for the soft freeze.

#65 lore raw
Quoted 9 lines · Markus Armbruster
Quoted 7 lines · Daniel P. Berrangé

Markus,

This series is fully reviewed now by multiple people.

Do you have any plans to do reviews and/or queue this as maintainer, or will you ack it for me to send a pull request.

Quoted 8 lines
Quoted 2 lines

I would like to get this merged before soft freeze which is fast approaching.

Soft freeze is in a week, and the heat wave left me seriously sleep-deprived.

I'd like to give your interface changes an eye-over. I probably won't be able to do more in time for the soft freeze.

In terms of interfaces...

Patch 22 is where we define the QAPI for the new user-creatable QOM objects for HMP/QMP, to enable use with -object / object-add.

Patch 32 is where we define the QAPI for the new auto-delete-on-close concept for the monitor.

With regards,
Daniel

Signature · 3 lines
|: https://berrange.com       ~~        https://hachyderm.io/@berrange :|
|: https://libvirt.org          ~~          https://entangle-photo.org :|
|: https://pixelfed.art/berrange   ~~    https://fstop138.berrange.com :|
#2 lore raw

Re: [PATCH v5 01/35] qom: replace 'can_be_deleted' with 'prepare_delete'

In reply to Daniel P. Berrangé
Quoted 18 lines · Daniel P. Berrangé

While most objects can perform all their cleanup in the finalizer method, there can be interactions with other resources / subsystems / threads which require that some cleanup be performed on an user creatable object before unparenting it and entering finalization.

The current 'can_be_deleted' method runs in the deletion path and is intended to be used to block deletion. While it could be used to perform cleanup tasks, its name suggests it should be free of side-effects.

Generalize this by renaming it to 'prepare_delete', explicitly allowing for cleanup to be provided. Existing users of 'can_be_deleted' are re-written, which provides them with more detailed/tailored error messages.

[...]

Quoted 14 lines
qom/object_interfaces.c +2 -2
index 7080f85f95..6faa0b2fd9 100644
@@ -32,16 +32,15 @@ bool user_creatable_complete(UserCreatable *uc, Error **errp)
     return !*errp;
 }
 
-bool user_creatable_can_be_deleted(UserCreatable *uc)
+bool user_creatable_prepare_delete(UserCreatable *uc, Error **errp)
 {
-
     UserCreatableClass *ucc = USER_CREATABLE_GET_CLASS(uc);
+    ERRP_GUARD();

qom/object_interfaces.c bool user_creatable_complete(UserCreatable *uc, Error **errp) · line 44

Quoted 10 lines
qom/object_interfaces.c +3 -4
 
-    if (ucc->can_be_deleted) {
-        return ucc->can_be_deleted(uc);
-    } else {
-        return true;
+    if (ucc->prepare_delete) {
+        ucc->prepare_delete(uc, errp);
     }
+    return !*errp;
 }

Simpler:

       if (ucc->prepare_delete) {
           return ucc->prepare_delete(uc, errp);
       }

       return true;

qom/object_interfaces.c bool user_creatable_del(const char *id, Error **errp) · line 255

Quoted 9 lines
qom/object_interfaces.c +1 -2
 void user_creatable_add_qapi(ObjectOptions *options, Error **errp)
@@ -253,8 +252,7 @@ bool user_creatable_del(const char *id, Error **errp)
         return false;
     }
 
-    if (!user_creatable_can_be_deleted(USER_CREATABLE(obj))) {
-        error_setg(errp, "object '%s' is in use, can not be deleted", id);
+    if (!user_creatable_prepare_delete(USER_CREATABLE(obj), errp)) {

This changes error messages.

The old ones mention @id.

The new ones don't:

        error_setg(errp, "Cryptodev backend is still in use");
        error_setg(errp, "Host memory backend is still mapped");
        error_setg(errp, "IOMMUFD backend still has %d users", be->users);
        error_setg(errp, "Throttle group still has multiple references");
    error_setg(errp, "Deleting can bus devices is not supported");
    error_setg(errp, "Deleting main loop is not supported");

Visible in the diff to tests/qemu-iotests/245 below.

Because this runs within object-del, and @id is the argument provided by the user there, the error message is still sufficiently clear, I guess.

If this ever gets used where the object isn't obvious to the user from context, the error messages become sub-par. Observation, not a demand.

Please cover this change in the commit message.

Quoted 49 lines
qom/object_interfaces.c +0 -0
         return false;
     }
 
tests/qemu-iotests/245 +2 -2
index f96610f510..e161bcfda9 100755
@@ -801,7 +801,7 @@ class TestBlockdevReopen(iotests.QMPTestCase):
         # Now group1 is in use, it cannot be deleted
         result = self.vm.qmp('object-del', id = 'group1')
         self.assert_qmp(result, 'error/class', 'GenericError')
-        self.assert_qmp(result, 'error/desc', "object 'group1' is in use, can not be deleted")
+        self.assert_qmp(result, 'error/desc', "Throttle group still has multiple references")
 
         # Default options, this switches the group back to group0
         self.reopen(opts)
@@ -809,7 +809,7 @@ class TestBlockdevReopen(iotests.QMPTestCase):
         # So now we cannot delete group0
         result = self.vm.qmp('object-del', id = 'group0')
         self.assert_qmp(result, 'error/class', 'GenericError')
-        self.assert_qmp(result, 'error/desc', "object 'group0' is in use, can not be deleted")
+        self.assert_qmp(result, 'error/desc', "Throttle group still has multiple references")
 
         # But group1 is free this time, and it can be deleted
         self.vm.cmd('object-del', id = 'group1')
util/main-loop.c +3 -2
index ad8645c30a..67ee06c311 100644
@@ -218,8 +218,9 @@ static void main_loop_init(EventLoopBase *base, Error **errp)
     mloop = m;
 }
 
-static bool main_loop_can_be_deleted(EventLoopBase *base)
+static bool main_loop_prepare_delete(EventLoopBase *base, Error **errp)
 {
+    error_setg(errp, "Deleting main loop is not supported");
     return false;
 }
 
@@ -229,7 +230,7 @@ static void main_loop_class_init(ObjectClass *oc, const void *class_data)
 
     bc->init = main_loop_init;
     bc->update_params = main_loop_update_params;
-    bc->can_be_deleted = main_loop_can_be_deleted;
+    bc->prepare_delete = main_loop_prepare_delete;
 }
 
 static const TypeInfo main_loop_info = {
#3 lore raw

Re: [PATCH v5 01/35] qom: replace 'can_be_deleted' with 'prepare_delete'

Quoted 50 lines · Markus Armbruster
Quoted 18 lines · Daniel P. Berrangé

While most objects can perform all their cleanup in the finalizer method, there can be interactions with other resources / subsystems / threads which require that some cleanup be performed on an user creatable object before unparenting it and entering finalization.

The current 'can_be_deleted' method runs in the deletion path and is intended to be used to block deletion. While it could be used to perform cleanup tasks, its name suggests it should be free of side-effects.

Generalize this by renaming it to 'prepare_delete', explicitly allowing for cleanup to be provided. Existing users of 'can_be_deleted' are re-written, which provides them with more detailed/tailored error messages.

[...]

Quoted 14 lines
qom/object_interfaces.c +2 -2
index 7080f85f95..6faa0b2fd9 100644
@@ -32,16 +32,15 @@ bool user_creatable_complete(UserCreatable *uc, Error **errp)
     return !*errp;
 }
 
-bool user_creatable_can_be_deleted(UserCreatable *uc)
+bool user_creatable_prepare_delete(UserCreatable *uc, Error **errp)
 {
-
     UserCreatableClass *ucc = USER_CREATABLE_GET_CLASS(uc);
+    ERRP_GUARD();

qom/object_interfaces.c bool user_creatable_complete(UserCreatable *uc, Error **errp) · line 44

Quoted 10 lines
qom/object_interfaces.c +3 -4
 
-    if (ucc->can_be_deleted) {
-        return ucc->can_be_deleted(uc);
-    } else {
-        return true;
+    if (ucc->prepare_delete) {
+        ucc->prepare_delete(uc, errp);
     }
+    return !*errp;
 }

Simpler:

Quoted 5 lines
       if (ucc->prepare_delete) {
           return ucc->prepare_delete(uc, errp);
       }

       return true;

ok

Quoted 26 lines

qom/object_interfaces.c bool user_creatable_del(const char *id, Error **errp) · line 255

Quoted 10 lines
qom/object_interfaces.c +1 -2
 void user_creatable_add_qapi(ObjectOptions *options, Error **errp)
@@ -253,8 +252,7 @@ bool user_creatable_del(const char *id, Error **errp)
         return false;
     }
 
-    if (!user_creatable_can_be_deleted(USER_CREATABLE(obj))) {
-        error_setg(errp, "object '%s' is in use, can not be deleted", id);
+    if (!user_creatable_prepare_delete(USER_CREATABLE(obj), errp)) {

This changes error messages.

The old ones mention @id.

The new ones don't:

        error_setg(errp, "Cryptodev backend is still in use");
        error_setg(errp, "Host memory backend is still mapped");
        error_setg(errp, "IOMMUFD backend still has %d users", be->users);
        error_setg(errp, "Throttle group still has multiple references");
    error_setg(errp, "Deleting can bus devices is not supported");
    error_setg(errp, "Deleting main loop is not supported");

Visible in the diff to tests/qemu-iotests/245 below.

Quoted 7 lines

Because this runs within object-del, and @id is the argument provided by the user there, the error message is still sufficiently clear, I guess.

If this ever gets used where the object isn't obvious to the user from context, the error messages become sub-par. Observation, not a demand.

Please cover this change in the commit message.

Oh yes, I remember being slightly annoyed at the need to do this - I was surprised to realize that an Object does not actually know its own 'id'. The 'id' only exists as the name of the property in the parent.

We could add an object_get_id() API which traveres obj->parent and then iterates over child properties. Annoyingly O(n) time for that, but perhaps not a big enough problem to worry about ? At least for error reporting we're probably ok with that time complexity even if there are 1000 objects registered as children.

Quoted 43 lines
Quoted 43 lines
qom/object_interfaces.c +0 -0
         return false;
     }
 
tests/qemu-iotests/245 +2 -2
index f96610f510..e161bcfda9 100755
@@ -801,7 +801,7 @@ class TestBlockdevReopen(iotests.QMPTestCase):
         # Now group1 is in use, it cannot be deleted
         result = self.vm.qmp('object-del', id = 'group1')
         self.assert_qmp(result, 'error/class', 'GenericError')
-        self.assert_qmp(result, 'error/desc', "object 'group1' is in use, can not be deleted")
+        self.assert_qmp(result, 'error/desc', "Throttle group still has multiple references")
 
         # Default options, this switches the group back to group0
         self.reopen(opts)
@@ -809,7 +809,7 @@ class TestBlockdevReopen(iotests.QMPTestCase):
         # So now we cannot delete group0
         result = self.vm.qmp('object-del', id = 'group0')
         self.assert_qmp(result, 'error/class', 'GenericError')
-        self.assert_qmp(result, 'error/desc', "object 'group0' is in use, can not be deleted")
+        self.assert_qmp(result, 'error/desc', "Throttle group still has multiple references")
 
         # But group1 is free this time, and it can be deleted
         self.vm.cmd('object-del', id = 'group1')
util/main-loop.c +2 -1
index ad8645c30a..67ee06c311 100644
@@ -218,8 +218,9 @@ static void main_loop_init(EventLoopBase *base, Error **errp)
     mloop = m;
 }
 
-static bool main_loop_can_be_deleted(EventLoopBase *base)
+static bool main_loop_prepare_delete(EventLoopBase *base, Error **errp)
 {
+    error_setg(errp, "Deleting main loop is not supported");
     return false;
 }
 
@@ -229,7 +230,7 @@ static void main_loop_class_init(ObjectClass *oc, const void *class_data)
 
Quoted 8 lines

util/main-loop.c static void main_loop_class_init(ObjectClass *oc, const void *class_data) · line 236

Quoted 7 lines
util/main-loop.c +1 -1
     bc->init = main_loop_init;
     bc->update_params = main_loop_update_params;
-    bc->can_be_deleted = main_loop_can_be_deleted;
+    bc->prepare_delete = main_loop_prepare_delete;
 }
 
 static const TypeInfo main_loop_info = {

With regards,
Daniel

Signature · 3 lines
|: https://berrange.com       ~~        https://hachyderm.io/@berrange :|
|: https://libvirt.org          ~~          https://entangle-photo.org :|
|: https://pixelfed.art/berrange   ~~    https://fstop138.berrange.com :|
#10 lore raw

Re: [PATCH v5 05/35] monitor: minimal conversion of monitors to QOM

In reply to Daniel P. Berrangé
Quoted 66 lines · Daniel P. Berrangé

This introduces a Monitor QOM object, with MonitorHMP and MonitorQMP subclasses. This is the bare minimum conversion of just the type declarations and replacing g_new/g_free with object_new/object_unref.

---
 include/monitor/monitor.h  | 11 ++++++++++-
 monitor/hmp.c              | 29 +++++++++++++++++++++++++++--
 monitor/monitor-internal.h | 18 ++++++++++++++++--
 monitor/monitor.c          | 18 ++++++++++++++++--
 monitor/qmp-cmds.c         | 15 ++++++++-------
 monitor/qmp.c              | 29 +++++++++++++++++++++++++++--
 6 files changed, 104 insertions(+), 16 deletions(-)
include/monitor/monitor.h +10 -1
index b9642b58ba..2e9f9e12e9 100644
@@ -5,8 +5,17 @@
 #include "qapi/qapi-types-misc.h"
 #include "qemu/readline.h"
 #include "exec/hwaddr.h"
+#include "qom/object.h"
+
+#define TYPE_MONITOR "monitor"
+OBJECT_DECLARE_TYPE(Monitor, MonitorClass, MONITOR);
+
+#define TYPE_MONITOR_HMP "monitor-hmp"
+OBJECT_DECLARE_TYPE(MonitorHMP, MonitorHMPClass, MONITOR_HMP);
+
+#define TYPE_MONITOR_QMP "monitor-qmp"
+OBJECT_DECLARE_TYPE(MonitorQMP, MonitorQMPClass, MONITOR_QMP);
 
-typedef struct MonitorHMP MonitorHMP;
 typedef struct MonitorOptions MonitorOptions;
 
 #define QMP_REQ_QUEUE_LEN_MAX 8
monitor/hmp.c +14 -0
index 4e4468424a..81047d2513 100644
@@ -43,6 +43,20 @@
 #include "system/block-backend.h"
 #include "trace.h"
 
+OBJECT_DEFINE_TYPE(MonitorHMP, monitor_hmp, MONITOR_HMP, MONITOR);
+
+static void monitor_hmp_finalize(Object *obj)
+{
+}
+
+static void monitor_hmp_class_init(ObjectClass *cls, const void *data)
+{
+}
+
+static void monitor_hmp_init(Object *obj)
+{
+}
+
 static void monitor_command_cb(void *opaque, const char *cmdline,
                                void *readline_opaque)
 {

monitor/hmp.c static void monitor_readline_flush(void *opaque) · line 1545

Quoted 7 lines
monitor/hmp.c +3 -1
@@ -1526,10 +1540,21 @@ static void monitor_readline_flush(void *opaque)
 
 void monitor_new_hmp(Chardev *chr, bool use_readline, Error **errp)
 {
-    MonitorHMP *mon = g_new0(MonitorHMP, 1);
+    MonitorHMP *mon;
+    static int counter;
+    g_autofree char *id = g_strdup_printf("hmpcompat%d", counter++);

Hmm. The system picking IDs is problematic when they can clash with the user's IDs. If we had an ounce of common sense, we'd restrict both across the board so they cannot clash. But we don't.

We need an ID here, because we need to make the new object the child of something (actually: child of /objects/), which requires a child name.

Non-problem with -object / object-add, because @id is mandatory there.

Non-problem with -device / device_add, because we use separate parents for devices with and without @id (/machine/peripheral/ and /machine/peripheral-anon/, plus the /machine/unattached/ orphanage).

Example for an existing problem:

    $ qemu-system-x86_64 -nodefaults -monitor stdio -chardev null,id=chr0 -mon id=compat_monitor0,chardev=chr0
    qemu-system-x86_64: -mon id=compat_monitor0,chardev=chr0: Duplicate ID 'compat_monitor0' for mon

Example for a problem created by this series:

    $  qemu-system-x86_64 -nodefaults -display none -S -monitor stdio -chardev null,id=chr0 -object monitor-hmp,id=hmpcompat0,chardev=chr0
    qemu-system-x86_64: -monitor stdio: attempt to add duplicate property 'hmpcompat0' to object (type 'container')

I readily admit that these clashes are *unlikely*. Still, do we really want to define an interface that claims to let you pick any ID, then rejects some of them sometimes? Feels rather 1990s to me. At the very least, cover the wart in the commit message.

This existing problem example leads me to the next mess: interaction with monitors' *other* ID.

qemu-system-FOO's -mon accepts an optional "id" parameter. It goes into its QemuOpts, and from there into MonitorOptions member @id.

qemu-storage-daemon's --monitor is similar, except it bypasses QemuOpts.

qemu-system-FOO provides convenience options -monitor, -qmp, -qmp-pretty.

Their argument may refer to an existing chardev by ID, like "chardev:ID". This creates a monitor with that same[*] QemuOpts and MonitorOptions ID.

Else, their argument is character device configuration in legacy syntax, like "stdio". This creates both a monitor and a character device, with ID "compat_monitorN", where N counts up from zero. The character device is visible in "info chardev", as always.

Aside: in both cases we use the same ID for two different objects, which feels unadvisable.

Aside: we have code checking whether a QemuOpts or character device ID starts with "compat_monitor", which is horryfying.

Your series does not mess with this at all. Understandable; I stay out of this swamp when I can, too.

However, it results in monitors having two IDs, namely the one in MonitorOptions, and the one in /object/. This is confusing.

Perhaps we should we'd get rid of the one in MonitorOptions. May well be more trouble than it's worth.

Could we at least make the two IDs the same?

Quoted 102 lines
monitor/hmp.c +10 -1
+    Object *obj = object_new_with_props(TYPE_MONITOR_HMP,
+                                        object_get_objects_root(),
+                                        id,
+                                        errp,
+                                        NULL);
+    if (!obj) {
+        return;
+    }
+    mon = MONITOR_HMP(obj);
 
     if (!qemu_chr_fe_init(&mon->parent_obj.chr, chr, errp)) {
-        g_free(mon);
+        object_unparent(OBJECT(mon));
         return;
     }
 
monitor/monitor-internal.h +16 -2
index 4da2b2a677..05c1f2f5e0 100644
@@ -101,7 +101,13 @@ typedef struct HMPCommand {
     bool coroutine;
 } HMPCommand;
 
+
+struct MonitorClass {
+    ObjectClass parent_class;
+};
+
 struct Monitor {
+    Object parent;
     CharFrontend chr;
     int suspend_cnt;            /* Needs to be accessed atomically */
     bool is_qmp;
@@ -127,6 +133,10 @@ struct Monitor {
     int reset_seen;
 };
 
+struct MonitorHMPClass {
+    MonitorClass parent_class;
+};
+
 struct MonitorHMP {
     Monitor parent_obj;
     bool use_readline;
@@ -140,7 +150,11 @@ struct MonitorHMP {
     ReadLineState *rs;
 };
 
-typedef struct {
+struct MonitorQMPClass {
+    MonitorClass parent_class;
+};
+
+struct MonitorQMP {
     Monitor parent_obj;
     JSONMessageParser parser;
     bool pretty;
@@ -160,7 +174,7 @@ typedef struct {
     QemuMutex qmp_queue_lock;
     /* Input queue that holds all the parsed QMP requests */
     GQueue *qmp_requests;
-} MonitorQMP;
+};
 
 /**
  * Is @mon a QMP monitor?
monitor/monitor.c +15 -1
index a87597e606..a497c25c54 100644
@@ -73,6 +73,20 @@ static GHashTable *coroutine_mon; /* Maps Coroutine* to Monitor* */
 MonitorList mon_list;
 static bool monitor_destroyed;
 
+OBJECT_DEFINE_TYPE(Monitor, monitor, MONITOR, OBJECT);
+
+static void monitor_finalize(Object *obj)
+{
+}
+
+static void monitor_class_init(ObjectClass *cls, const void *data)
+{
+}
+
+static void monitor_init(Object *obj)
+{
+}
+
 Monitor *monitor_cur(void)
 {
     Monitor *mon;
@@ -598,7 +612,7 @@ void monitor_list_append(Monitor *mon)
 
     if (mon) {
         monitor_data_destroy(mon);
-        g_free(mon);
+        object_unparent(OBJECT(mon));
     }
 }
 

monitor/monitor.c void monitor_cleanup(void) · line 700

Quoted 8 lines
monitor/monitor.c +1 -1
@@ -680,7 +694,7 @@ void monitor_cleanup(void)
         monitor_flush(mon);
         monitor_data_destroy(mon);
         qemu_mutex_lock(&monitor_lock);
-        g_free(mon);
+        object_unparent(OBJECT(mon));
     }
     qemu_mutex_unlock(&monitor_lock);
 

Hmm...

    {"execute": "qom-list-types", "arguments": {"implements": "monitor"}}
    {"return": [{"name": "monitor-hmp", "parent": "monitor"}, {"name": "monitor-qmp", "parent": "monitor"}, {"name": "monitor", "parent": "object"}]}

Shouldn't type "monitor" be abstract?

Quoted 68 lines
monitor/qmp-cmds.c +8 -7
index aa9ee8a391..bfde769ef0 100644
@@ -166,12 +166,12 @@ char *qmp_human_monitor_command(const char *command_line, bool has_cpu_index,
                                 int64_t cpu_index, Error **errp)
 {
     char *output = NULL;
-    MonitorHMP hmp = {};
+    MonitorHMP *hmp = MONITOR_HMP(object_new(TYPE_MONITOR_HMP));
 
-    monitor_data_init(&hmp.parent_obj, false, true, false);
+    monitor_data_init(&hmp->parent_obj, false, true, false);
 
     if (has_cpu_index) {
-        int ret = monitor_set_cpu(&hmp.parent_obj, cpu_index);
+        int ret = monitor_set_cpu(&hmp->parent_obj, cpu_index);
         if (ret < 0) {
             error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cpu-index",
                        "a CPU number");
@@ -179,14 +179,15 @@ char *qmp_human_monitor_command(const char *command_line, bool has_cpu_index,
         }
     }
 
-    handle_hmp_command(&hmp, command_line);
+    handle_hmp_command(hmp, command_line);
 
-    WITH_QEMU_LOCK_GUARD(&hmp.parent_obj.mon_lock) {
-        output = g_strdup(hmp.parent_obj.outbuf->str);
+    WITH_QEMU_LOCK_GUARD(&hmp->parent_obj.mon_lock) {
+        output = g_strdup(hmp->parent_obj.outbuf->str);
     }
 
 out:
-    monitor_data_destroy(&hmp.parent_obj);
+    monitor_data_destroy(&hmp->parent_obj);
+    object_unref(hmp);
     return output;
 }
 
monitor/qmp.c +14 -0
index cb28a95efd..5231ed506a 100644
@@ -71,6 +71,20 @@ typedef struct QMPRequest QMPRequest;
 
 QmpCommandList qmp_commands, qmp_cap_negotiation_commands;
 
+OBJECT_DEFINE_TYPE(MonitorQMP, monitor_qmp, MONITOR_QMP, MONITOR);
+
+static void monitor_qmp_finalize(Object *obj)
+{
+}
+
+static void monitor_qmp_class_init(ObjectClass *cls, const void *data)
+{
+}
+
+static void monitor_qmp_init(Object *obj)
+{
+}
+
 static bool qmp_oob_enabled(MonitorQMP *mon)
 {
     return mon->capab[QMP_CAPABILITY_OOB];
@@ -515,10 +529,21 @@ static void monitor_qmp_setup_handlers_bh(void *opaque)
 
 void monitor_new_qmp(Chardev *chr, bool pretty, Error **errp)

monitor/qmp.c static void monitor_qmp_setup_handlers_bh(void *opaque) · line 539

Quoted 10 lines
monitor/qmp.c +8 -1
 {
-    MonitorQMP *mon = g_new0(MonitorQMP, 1);
+    MonitorQMP *mon;
+    static int counter;
+    g_autofree char *id = g_strdup_printf("qmpcompat%d", counter++);
+    Object *obj = object_new_with_props(TYPE_MONITOR_QMP,
+                                        object_get_objects_root(),
+                                        id,
+                                        errp,
+                                        NULL);

See discussion of monitor_new_hmp() above.

monitor/qmp.c static void monitor_qmp_setup_handlers_bh(void *opaque) · line 549

Quoted 11 lines
monitor/qmp.c +5 -1
+    if (!obj) {
+        return;
+    }
+    mon = MONITOR_QMP(obj);
 
     if (!qemu_chr_fe_init(&mon->parent_obj.chr, chr, errp)) {
-        g_free(mon);
+        object_unparent(OBJECT(mon));
         return;
     }
     qemu_chr_fe_set_echo(&mon->parent_obj.chr, true);

[*] Silently truncated to 31 characters. I kid you not.

#66 lore raw
In reply to Daniel P. Berrangé
Quoted 19 lines · Daniel P. Berrangé
Quoted 9 lines · Markus Armbruster
Quoted 7 lines · Daniel P. Berrangé

Markus,

This series is fully reviewed now by multiple people.

Do you have any plans to do reviews and/or queue this as maintainer, or will you ack it for me to send a pull request.

Quoted 8 lines
Quoted 2 lines

I would like to get this merged before soft freeze which is fast approaching.

Soft freeze is in a week, and the heat wave left me seriously sleep-deprived.

I'd like to give your interface changes an eye-over. I probably won't be able to do more in time for the soft freeze.

Quoted 7 lines

In terms of interfaces...

Patch 22 is where we define the QAPI for the new user-creatable QOM objects for HMP/QMP, to enable use with -object / object-add.

Patch 32 is where we define the QAPI for the new auto-delete-on-close concept for the monitor.

Thanks!

I'll try to also look at PATCH 34 and 35.

#11 lore raw

Re: [PATCH v5 05/35] monitor: minimal conversion of monitors to QOM

In reply to Markus Armbruster
Quoted 96 lines · Markus Armbruster
Quoted 66 lines · Daniel P. Berrangé

This introduces a Monitor QOM object, with MonitorHMP and MonitorQMP subclasses. This is the bare minimum conversion of just the type declarations and replacing g_new/g_free with object_new/object_unref.

---
 include/monitor/monitor.h  | 11 ++++++++++-
 monitor/hmp.c              | 29 +++++++++++++++++++++++++++--
 monitor/monitor-internal.h | 18 ++++++++++++++++--
 monitor/monitor.c          | 18 ++++++++++++++++--
 monitor/qmp-cmds.c         | 15 ++++++++-------
 monitor/qmp.c              | 29 +++++++++++++++++++++++++++--
 6 files changed, 104 insertions(+), 16 deletions(-)
include/monitor/monitor.h +10 -1
index b9642b58ba..2e9f9e12e9 100644
@@ -5,8 +5,17 @@
 #include "qapi/qapi-types-misc.h"
 #include "qemu/readline.h"
 #include "exec/hwaddr.h"
+#include "qom/object.h"
+
+#define TYPE_MONITOR "monitor"
+OBJECT_DECLARE_TYPE(Monitor, MonitorClass, MONITOR);
+
+#define TYPE_MONITOR_HMP "monitor-hmp"
+OBJECT_DECLARE_TYPE(MonitorHMP, MonitorHMPClass, MONITOR_HMP);
+
+#define TYPE_MONITOR_QMP "monitor-qmp"
+OBJECT_DECLARE_TYPE(MonitorQMP, MonitorQMPClass, MONITOR_QMP);
 
-typedef struct MonitorHMP MonitorHMP;
 typedef struct MonitorOptions MonitorOptions;
 
 #define QMP_REQ_QUEUE_LEN_MAX 8
monitor/hmp.c +14 -0
index 4e4468424a..81047d2513 100644
@@ -43,6 +43,20 @@
 #include "system/block-backend.h"
 #include "trace.h"
 
+OBJECT_DEFINE_TYPE(MonitorHMP, monitor_hmp, MONITOR_HMP, MONITOR);
+
+static void monitor_hmp_finalize(Object *obj)
+{
+}
+
+static void monitor_hmp_class_init(ObjectClass *cls, const void *data)
+{
+}
+
+static void monitor_hmp_init(Object *obj)
+{
+}
+
 static void monitor_command_cb(void *opaque, const char *cmdline,
                                void *readline_opaque)
 {

monitor/hmp.c static void monitor_readline_flush(void *opaque) · line 1545

Quoted 7 lines
monitor/hmp.c +3 -1
@@ -1526,10 +1540,21 @@ static void monitor_readline_flush(void *opaque)
 
 void monitor_new_hmp(Chardev *chr, bool use_readline, Error **errp)
 {
-    MonitorHMP *mon = g_new0(MonitorHMP, 1);
+    MonitorHMP *mon;
+    static int counter;
+    g_autofree char *id = g_strdup_printf("hmpcompat%d", counter++);

Hmm. The system picking IDs is problematic when they can clash with the user's IDs. If we had an ounce of common sense, we'd restrict both across the board so they cannot clash. But we don't.

We need an ID here, because we need to make the new object the child of something (actually: child of /objects/), which requires a child name.

Non-problem with -object / object-add, because @id is mandatory there.

Non-problem with -device / device_add, because we use separate parents for devices with and without @id (/machine/peripheral/ and /machine/peripheral-anon/, plus the /machine/unattached/ orphanage).

Example for an existing problem:

    $ qemu-system-x86_64 -nodefaults -monitor stdio -chardev null,id=chr0 -mon id=compat_monitor0,chardev=chr0
    qemu-system-x86_64: -mon id=compat_monitor0,chardev=chr0: Duplicate ID 'compat_monitor0' for mon

Example for a problem created by this series:

Quoted 7 lines
    $  qemu-system-x86_64 -nodefaults -display none -S -monitor stdio -chardev null,id=chr0 -object monitor-hmp,id=hmpcompat0,chardev=chr0
    qemu-system-x86_64: -monitor stdio: attempt to add duplicate property 'hmpcompat0' to object (type 'container')

I readily admit that these clashes are *unlikely*. Still, do we really want to define an interface that claims to let you pick any ID, then rejects some of them sometimes? Feels rather 1990s to me. At the very least, cover the wart in the commit message.

The way I looked at it was aything using -object with the new monitor-qmp/monitor-hmp types is new code. They can:

1. Trivially abide the warning about "hmpcompatNN" / "qmpcompatNN" being internal usage for compat syntax
2. Not use both -monitor and -object on the same QEMU instance so not have a clash between the two to begin with

The remaining danger where is some existing code using -object with a *non-monitor* type, and calling it "hmpcompatNN" / "qmpcompatNN", which would be insanity. Never say never, but I think that's an acceptable risk.

Quoted 5 lines

This existing problem example leads me to the next mess: interaction with monitors' *other* ID.

qemu-system-FOO's -mon accepts an optional "id" parameter. It goes into its QemuOpts, and from there into MonitorOptions member @id.

Arrrggggggghh. That is not documented for -mon at all AFAICT

  $ qemu-system-x86_64  -help 2>&1 | grep -- -mon
  -monitor dev    redirect the monitor to char device 'dev'
  -qmp dev        like -monitor but opens in 'control' mode
  -mon [chardev=]name[,mode=readline|control][,pretty[=on|off]]

but yeah as you say, it exists, and in fact libvirt even uses it despite it being undocumented.

Quoted 9 lines

qemu-storage-daemon's --monitor is similar, except it bypasses QemuOpts.

qemu-system-FOO provides convenience options -monitor, -qmp, -qmp-pretty.

Their argument may refer to an existing chardev by ID, like "chardev:ID". This creates a monitor with that same[*] QemuOpts and MonitorOptions ID.

Quoted 7 lines

Else, their argument is character device configuration in legacy syntax, like "stdio". This creates both a monitor and a character device, with ID "compat_monitorN", where N counts up from zero. The character device is visible in "info chardev", as always.

Aside: in both cases we use the same ID for two different objects, which feels unadvisable.

Agreed, that's awful.

Quoted 2 lines

Aside: we have code checking whether a QemuOpts or character device ID starts with "compat_monitor", which is horryfying.

Eww.

Quoted 2 lines

Your series does not mess with this at all. Understandable; I stay out of this swamp when I can, too.

I didn't realize the swap was there in this case !

Quoted 7 lines

However, it results in monitors having two IDs, namely the one in MonitorOptions, and the one in /object/. This is confusing.

Perhaps we should we'd get rid of the one in MonitorOptions. May well be more trouble than it's worth.

Could we at least make the two IDs the same?

Since -mon has an existing ID, we might as well pass it through to use as the child prop name for the objects.

Quoted 46 lines
Quoted 35 lines
monitor/monitor.c +15 -1
index a87597e606..a497c25c54 100644
@@ -73,6 +73,20 @@ static GHashTable *coroutine_mon; /* Maps Coroutine* to Monitor* */
 MonitorList mon_list;
 static bool monitor_destroyed;
 
+OBJECT_DEFINE_TYPE(Monitor, monitor, MONITOR, OBJECT);
+
+static void monitor_finalize(Object *obj)
+{
+}
+
+static void monitor_class_init(ObjectClass *cls, const void *data)
+{
+}
+
+static void monitor_init(Object *obj)
+{
+}
+
 Monitor *monitor_cur(void)
 {
     Monitor *mon;
@@ -598,7 +612,7 @@ void monitor_list_append(Monitor *mon)
 
     if (mon) {
         monitor_data_destroy(mon);
-        g_free(mon);
+        object_unparent(OBJECT(mon));
     }
 }
 

monitor/monitor.c void monitor_cleanup(void) · line 700

Quoted 8 lines
monitor/monitor.c +1 -1
@@ -680,7 +694,7 @@ void monitor_cleanup(void)
         monitor_flush(mon);
         monitor_data_destroy(mon);
         qemu_mutex_lock(&monitor_lock);
-        g_free(mon);
+        object_unparent(OBJECT(mon));
     }
     qemu_mutex_unlock(&monitor_lock);
 

Hmm...

Quoted 4 lines
    {"execute": "qom-list-types", "arguments": {"implements": "monitor"}}
    {"return": [{"name": "monitor-hmp", "parent": "monitor"}, {"name": "monitor-qmp", "parent": "monitor"}, {"name": "monitor", "parent": "object"}]}

Shouldn't type "monitor" be abstract?

Yes it absolutely should be, and I'd swear it *was* abstract at some point in my work, but I guess I lost it along the way.

With regards,
Daniel

Signature · 3 lines
|: https://berrange.com       ~~        https://hachyderm.io/@berrange :|
|: https://libvirt.org          ~~          https://entangle-photo.org :|
|: https://pixelfed.art/berrange   ~~    https://fstop138.berrange.com :|
Acked-by on v5 22/35
#33 lore raw

Re: [PATCH v5 22/35] monitor: implement "user creatable" interface for adding monitors

In reply to Daniel P. Berrangé
Quoted 15 lines · Daniel P. Berrangé

Implement the user creatable QOM interface and define the monitor-qmp and monitor-hmp types in QAPI. This unlocks the ability to create them on the command line with -object or in HMP/QMP with object_add.

For example:

$QEMU -chardev stdio,id=monchr0 -object monitor-hmp,id=mon0,chrdev=monchr0

Initially the "prepare_delete" callback is hardcoded to return an error which means -object and object_add can be used, but object_del will fail. Support for deleting monitors will be introduced in subsequent commits.

[...]

Quoted 17 lines
qapi/qom.json +9 -0
index dd45ac1087..6affb70a59 100644
@@ -1187,6 +1187,45 @@
   'data': { '*cpu-affinity': ['uint16'],
             '*node-affinity': ['uint16'] } }
 
+##
+# @MonitorProperties:
+#
+# Properties for all monitors
+#
+# @chardev: ID of the character device providing the monitor transport
+#
+# Since: 11.1
+##

qapi/qom.json line 1208

Quoted 10 lines
qapi/qom.json +10 -0
+{ 'struct': 'MonitorProperties',
+  'data': { 'chardev': 'str' }}
+
+##
+# @MonitorHMPProperties:
+#
+# Properties for the HMP monitor
+#
+# @readline: whether to enable readline for line editing
+#      (default: enabled)

The default value is 'true', and it means 'enabeld'. We pretty much universally have the default value in (default: ...), not the meaning.

qapi/qom.json line 1221

Quoted 13 lines
qapi/qom.json +13 -0
+#
+# Since: 11.1
+##
+{ 'struct': 'MonitorHMPProperties',
+  'base': 'MonitorProperties',
+  'data': { '*readline': 'bool' } }
+
+##
+# @MonitorQMPProperties:
+#
+# Properties for the QMP monitor
+#
+# @pretty: whether to pretty print JSON responses (default: disabled)

Likewise.

Quoted 20 lines
qapi/qom.json +9 -0
+#
+# Since: 11.1
+##
+{ 'struct': 'MonitorQMPProperties',
+  'base': 'MonitorProperties',
+  'data': { '*pretty': 'bool' } }
+
 ##
 # @ObjectType:
 #
@@ -1237,6 +1276,8 @@
     'memory-backend-ram',
     { 'name': 'memory-backend-shm',
       'if': 'CONFIG_POSIX' },
+    'monitor-hmp',
+    'monitor-qmp',
     'pef-guest',
     { 'name': 'pr-manager-helper',
       'if': 'CONFIG_LINUX' },

qapi/qom.json line 1360

Quoted 5 lines
qapi/qom.json +2 -0
@@ -1315,6 +1356,8 @@
       'memory-backend-ram':         'MemoryBackendProperties',
       'memory-backend-shm':         { 'type': 'MemoryBackendShmProperties',
                                       'if': 'CONFIG_POSIX' },
+      'monitor-hmp':                {'type': 'MonitorHMPProperties' },
+      'monitor-qmp':                {'type': 'MonitorQMPProperties' },

Use the short form when possible:

'monitor-hmp': 'MonitorHMPProperties',
'monitor-qmp': 'MonitorQMPProperties',

Quoted 18 lines
qapi/qom.json +0 -0
       'pr-manager-helper':          { 'type': 'PrManagerHelperProperties',
                                       'if': 'CONFIG_LINUX' },
       'qtest':                      'QtestProperties',
stubs/monitor-internal.c +1 -0
index 51db7588b9..59ccc4b35c 100644
@@ -10,4 +10,5 @@ int monitor_get_fd(Monitor *mon, const char *name, Error **errp)
 
 void monitor_new_hmp(const char *chardev_id, bool use_readline, Error **errp)
 {
+    g_assert_not_reached();
 }
system/vl.c +0 -0
index fb36140aac..79ed2be449 100644

system/vl.c static void object_option_add_visitor(Visitor *v) · line 1836

Quoted 7 lines
system/vl.c +4 -0
@@ -1830,6 +1830,10 @@ static void object_option_add_visitor(Visitor *v)
 {
     ObjectOption *opt = g_new0(ObjectOption, 1);
     visit_type_ObjectOptions(v, NULL, &opt->opts, &error_fatal);
+    if (opt->opts->qom_type == OBJECT_TYPE_MONITOR_HMP ||
+        opt->opts->qom_type == OBJECT_TYPE_MONITOR_QMP) {
+        default_monitor = 0;
+    }

Still okay, but if we pick up more, we better do it like default_driver_disable() does for devices.

system/vl.c static bool object_create_early(const char *type) · line 1982

Quoted 13 lines
system/vl.c +3 -1
     QTAILQ_INSERT_TAIL(&object_opts, opt, next);
 }
 
@@ -1971,7 +1975,9 @@ static bool object_create_early(const char *type)
 
     /* Reason: property "chardev" */
     if (g_str_equal(type, "rng-egd") ||
-        g_str_equal(type, "qtest")) {
+        g_str_equal(type, "qtest") ||
+        g_str_equal(type, "monitor-hmp") ||
+        g_str_equal(type, "monitor-qmp")) {
         return false;
     }

QAPI schema

#47 lore raw

Re: [PATCH v5 32/35] monitor: add support for auto-deleting monitors upon close

In reply to [email protected]
Quoted 6 lines · [email protected] writes:
Quoted 6 lines

The default monitor is usually a long lived object that will exist for the entire lifetime of the VM. A monitor can only service a single client at a time though, and so it might be desirable to hotplug additional monitors at runtime for specific tasks. If doing that, however, there is a need to remove the monitor when it is no longer needed.

Whatever adds the additional monitor can also delete it. The fact that you propose other means suggests you believe this would be cumbersome in practice. Why?

Quoted 7 lines
Quoted 7 lines

Allowing a client to run "object-del" against its own monitor adds complex edge cases, as it would be desirable to send the QMP response despite the monitor sending it being deleted. Doing "object-del" alone will also result in orphaning a character device backend instance, as there is no opportunity to run the companion "chardev-del" command.

Quoted 5 lines
Quoted 5 lines

A simpler way to ensure cleanup is to add the concept of auto-deleting monitor objects. Specifically when the "CHR_EVENT_CLOSED" event is emitted, the equivalent of "object-del" + "chardev-del" can be run internally. Since the transient client has already droppped its monitor connection, there is no synchronization to be concerned about.

If object-del or chardev-del fail, there's no way to report the error. Can they fail?

Do we always want to delete both monitor and character device?

Quoted 8 lines
Quoted 8 lines

This is implemented via a new "close-action=none|delete" property on the 'monitor-qmp' object. This concept could be extended with further actions in future, for example:

* close-action=shutdown - graceful guest shutdown
* close-action=terminate - immediate guest poweroff
* close-action=stop - pause guest CPUs while the monitor is not connected to any client

I'm not sure these would be useful and safe. Mentioning them here is okay regardless.

Quoted 15 lines
Quoted 15 lines

This is left as an exercise for future interested contributors.

Message-ID: <[email protected]>

qapi/qom.json +0 -0
index 6affb70a594..b4c49a2e8b9 100644
@@ -1213,18 +1213,37 @@
   'base': 'MonitorProperties',
   'data': { '*readline': 'bool' } }
 
Quoted 5 lines

qapi/qom.json line 1220

Quoted 5 lines
qapi/qom.json +5 -0
+
+##
+# @MonitorQMPCloseAction:
+#
+# Action to take when the character device backend is

No line break here.

Quoted 3 lines

qapi/qom.json line 1223

Quoted 3 lines
qapi/qom.json +3 -0
+# closed.
+#
+# @none: take no action (the default)

The type cannot specify a default, only its users can. Scratch (the default).

Blank line between member descriptions, please.

Quoted 2 lines

qapi/qom.json line 1225

Quoted 2 lines
qapi/qom.json +2 -0
+# @delete: delete both the 'monitor-qmp' object and its associated
+#          character device backend object

Format like

# @delete: delete both the 'monitor-qmp' object and its associated
# character device backend object

Quoted 4 lines

qapi/qom.json line 1227

Quoted 2 lines
qapi/qom.json +2 -0
+#
+# Since 11.1

should be "Since: 11.1" (worth a new check perhaps)

docs/devel/qapi-code-gen.rst:

    Section tags are case-sensitive and end with a colon.  They are only
    recognized after a blank line.  Good example::

     #
     # Since: 7.1

    Bad examples (all ordinary paragraphs)::

     # since: 7.1

     # Since 7.1

     # Since : 7.1

There are a few bad ones in master. I posted a patch:

    Subject: [PATCH] qapi: Fix misspelled section tags in doc comments
    Date: Wed, 1 Jul 2026 08:11:36 +0200
    Message-ID: <[email protected]>

Code to detect and reject such errors would have to carefully avoid false negatives.

Quoted 4 lines

qapi/qom.json line 1229

Quoted 2 lines
qapi/qom.json +2 -0
+##
+{ 'enum' : 'MonitorQMPCloseAction',

extra space, vs the usual '{ 'enum':'.

Yes.

Quoted 8 lines

qapi/qom.json line 1237

Quoted 8 lines
qapi/qom.json +2 -0
+  'data': ['none', 'delete'] }
+
 ##
 # @MonitorQMPProperties:
 #
 # Properties for the QMP monitor
 #
 # @pretty: whether to pretty print JSON responses (default: disabled)

(default: false)

See my review of PATCH 22.

Blank line between member descriptions, please.

Quoted 2 lines

qapi/qom.json line 1239

Quoted 2 lines
qapi/qom.json +2 -0
+# @close-action: action to take when the character device backend
+#                is closed (default: none)

Format like

# @close-action: action to take when the character device backend is
# closed (default: none)

Quoted 93 lines
Quoted 93 lines
qapi/qom.json +2 -1
 #
 # Since: 11.1
 ##
 { 'struct': 'MonitorQMPProperties',
   'base': 'MonitorProperties',
-  'data': { '*pretty': 'bool' } }
+  'data': { '*pretty': 'bool',
+            '*close-action': 'MonitorQMPCloseAction' } }
 
 ##
 # @ObjectType:
monitor/monitor-internal.h +3 -0
index 5522e05464b..23829f32f9a 100644
@@ -28,6 +28,7 @@
 #include "chardev/char-fe.h"
 #include "monitor/monitor.h"
 #include "qapi/qapi-types-control.h"
+#include "qapi/qapi-types-qom.h"
 #include "qapi/qmp-registry.h"
 #include "qobject/json-parser.h"
 #include "qemu/readline.h"
@@ -178,7 +179,9 @@ struct MonitorQMP {
     Monitor parent_obj;
     JSONMessageParser parser;
     bool pretty;
+    MonitorQMPCloseAction close_action;
     bool setup_pending; /* iothread BH has not yet set up chardev handlers */
+    bool delete_pending; /* close_action has started 'delete' process */
     /*
      * When a client connects, we're in capabilities negotiation mode.
      * @commands is &qmp_cap_negotiation_commands then.  When command
monitor/qmp.c +31 -0
index 5301927f09f..1057b0d12ba 100644
@@ -28,6 +28,7 @@
 #include "monitor-internal.h"
 #include "qapi/error.h"
 #include "qapi/qapi-commands-control.h"
+#include "qapi/qapi-commands-char.h"
 #include "qobject/qdict.h"
 #include "qobject/qjson.h"
 #include "qobject/qlist.h"
@@ -103,6 +104,20 @@ static void monitor_qmp_set_pretty(Object *obj, bool val, Error **errp)
     mon->pretty = val;
 }
 
+static int monitor_qmp_get_close_action(Object *obj, Error **errp)
+{
+    MonitorQMP *mon = MONITOR_QMP(obj);
+
+    return mon->close_action;
+}
+
+static void monitor_qmp_set_close_action(Object *obj, int val, Error **errp)
+{
+    MonitorQMP *mon = MONITOR_QMP(obj);
+
+    mon->close_action = val;
+}
+
 static void monitor_qmp_emit_event(Monitor *mon, QAPIEvent event, QDict *qdict);
 static bool monitor_qmp_requires_iothread(const Monitor *mon);
 static void monitor_qmp_complete(UserCreatable *uc, Error **errp);
@@ -117,6 +132,11 @@ static void monitor_qmp_class_init(ObjectClass *cls, const void *data)
     object_class_property_add_bool(cls, "pretty",
                                    monitor_qmp_get_pretty,
                                    monitor_qmp_set_pretty);
+    object_class_property_add_enum(cls, "close-action",
+                                   "MonitorQMPCloseAction",
+                                   &MonitorQMPCloseAction_lookup,
+                                   monitor_qmp_get_close_action,
+                                   monitor_qmp_set_close_action);
 
     moncls->emit_event = monitor_qmp_emit_event;
     moncls->requires_iothread = monitor_qmp_requires_iothread;
@@ -550,11 +570,45 @@ static QDict *qmp_greeting(MonitorQMP *mon)
         ver, cap_list);
 }
 
+static void monitor_qmp_self_delete_bh(void *opaque)
+{
+    MonitorQMP *mon = opaque;
+    g_autofree char *mon_id = object_property_get_child_name(
+        object_get_objects_root(), OBJECT(mon));
+    g_autofree char *chardev_id = g_strdup(mon->parent_obj.chardev_id);
+    Error *local_error = NULL;
+
+    if (!mon_id) {
+        /* Another monitor raced & ran 'object-del' on 'mon'
+         * before this BH got scheduled, so we have a ref on
Quoted 8 lines

monitor/qmp.c static QDict *qmp_greeting(MonitorQMP *mon) · line 591

Quoted 8 lines
monitor/qmp.c +8 -0
+         * mon but it is already unparented.
+         */
+        object_unref(mon);
+        return;
+    }
+
+    user_creatable_del(mon_id, &local_error);
+    object_unref(mon);

This object_unref() and the one above pair with the object_ref() in monitor_qmp_event() case MONITOR_QMP_CLOSE_ACTION_DELETE. Correct?

Is this worth a comment?

Quoted 12 lines
Quoted 12 lines
monitor/qmp.c +7 -0
+    if (local_error != NULL) {
+        error_report_err(local_error);
+    } else {
+        qmp_chardev_remove(chardev_id, NULL);
+    }
+}
+
 static void monitor_qmp_event(void *opaque, QEMUChrEvent event)
 {
     QDict *data;
     MonitorQMP *mon = opaque;
 
Quoted 7 lines

monitor/qmp.c static QDict *qmp_greeting(MonitorQMP *mon) · line 610

Quoted 7 lines
monitor/qmp.c +7 -0
+    /* Protect against race if a client drops & quickly
+     * reconnects - we'll have the delete BH scheduled
+     * so must not honour a new open request
+     */
+    if (mon->delete_pending) {
+        return;
+    }

Consider "something happens between setting @delete_pending and monitor destruction that triggers the event" and "same thing happens after destruction". Any observable differences in behavior?

Are there any other monitor interactions that might need similar special-casing when @delete_pending?

Quoted 11 lines

monitor/qmp.c static void monitor_qmp_event(void *opaque, QEMUChrEvent event) · line 636

Quoted 11 lines
monitor/qmp.c +4 -0
+
     switch (event) {
     case CHR_EVENT_OPENED:
         WITH_QEMU_LOCK_GUARD(&mon->parent_obj.mon_lock) {
@@ -577,6 +631,27 @@ static void monitor_qmp_event(void *opaque, QEMUChrEvent event)
         json_message_parser_init(&mon->parser, handle_qmp_command,
                                  mon, NULL);
         monitor_fdsets_cleanup();
+        switch (mon->close_action) {
+        case MONITOR_QMP_CLOSE_ACTION_NONE:
+            break; /* nada */

Useless comment :)

Quoted 7 lines

monitor/qmp.c static void monitor_qmp_event(void *opaque, QEMUChrEvent event) · line 643

Quoted 7 lines
monitor/qmp.c +7 -0
+        case MONITOR_QMP_CLOSE_ACTION_DELETE:
+            mon->delete_pending = true;
+            /*
+             * Do NOT run in the AIO context associated with the
+             * monitor. We need to run in the default AIO context
+             * which is the same context in which 'qmp_object_del'
+             * will execute

I believe you. I don't have the headspace right now to confirm this myself.

Quoted 3 lines

monitor/qmp.c static void monitor_qmp_event(void *opaque, QEMUChrEvent event) · line 646

Quoted 3 lines
monitor/qmp.c +3 -0
+             *
+             * Hold an extra ref in case a separate monitor races
+             * with the BH by processing an explicit 'object-del'

The extra ref ensures the object still lives when the BH runs. Correct?

Quoted 2 lines

monitor/qmp.c static void monitor_qmp_event(void *opaque, QEMUChrEvent event) · line 648

Quoted 2 lines
monitor/qmp.c +2 -0
+             */
+            object_ref(mon);

This is the object_ref() I referred to above.

Quoted 9 lines
Quoted 9 lines
monitor/qmp.c +6 -0
+            aio_bh_schedule_oneshot(qemu_get_aio_context(),
+                                    monitor_qmp_self_delete_bh, mon);
+            break;
+        default:
+            g_assert_not_reached();
+        }
         break;
     case CHR_EVENT_BREAK:
     case CHR_EVENT_MUX_IN:

[...]

Quoted 1 line
#59 lore raw

Re: [PATCH v5 34/35] qemu-options: document new monitor-hmp and monitor-qmp objects

In reply to Daniel P. Berrangé
Quoted 20 lines · Daniel P. Berrangé

Add new docs for the -object monitor-hmp and -object monitor-qmp options, updating -mon to state that it is legacy syntax sugar for the new -object args.

---
 qemu-options.hx | 49 +++++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 49 insertions(+)
qemu-options.hx +1 -0
index c799286153..a804d5b880 100644
@@ -4980,6 +4980,16 @@ SRST
       -mon chardev=mon1,mode=control,pretty=on
 
     enables the QMP monitor on localhost port 4444 with pretty-printing.
+

qemu-options.hx SRST · line 4993

Quoted 10 lines
qemu-options.hx +9 -0
+    The use of ``-mon mode=readline`` is historical syntax sugar
+    for the new ``-object monitor-hmp`` option, each use of which
+    creates an object with the ID ``hmpcompatNNN`` where ``NNN`` is
+    a counter starting from 0.
+
+    The use of ``-mon mode=control`` is historical syntax sugar
+    for the new ``-object monitor-qmp`` option, each use of which
+    creates an object with the ID ``qmpcompatNNN`` where ``NNN`` is
+    a counter starting from 0.
 ERST

Should we document how -monitor, -qmp, and -qmp-pretty map to -object and -chardev?

Quoted 48 lines
qemu-options.hx +39 -0
 DEF("debugcon", HAS_ARG, QEMU_OPTION_debugcon, \
@@ -5730,6 +5740,45 @@ SRST
     they are specified. Note that the 'id' property must be set. These
     objects are placed in the '/objects' path.
 
+    ``-object monitor-hmp,id=id,chardev=chardev_id,readline=on|off``
+        Set up a monitor running the Human Monitor Protocol,
+        connected to the chardev ``chardev_id``.
+
+        The ``id`` parameter is a unique ID that can be used
+        to dynamically delete the monitor at runtime. Note
+        that monitors created using the historical syntax
+        will be allocated IDs following the pattern ``hmpcompatNN``.
+        Mixing ``-object`` with the historical monitor syntax is
+        discouraged.
+
+        The ``readline`` parameter, which defaults to ``on``,
+        controls whether the monitor provides line editing.
+
+    ``-object monitor-qmp,id=id,chardev=chardev_id,pretty=on|off,close-action=none|delete``
+        Set up a monitor running the QEMU Monitor Protocol,
+        connected to the chardev ``chardev_id``.
+
+        The ``id`` parameter is a unique ID that can be used
+        to dynamically delete the monitor at runtime. Note
+        that monitors created using the historical syntax
+        will be allocated IDs following the pattern ``qmpcompatNN``.
+        Mixing ``-object`` with the historical monitor syntax is
+        discouraged.
+
+        The ``pretty`` parameter, which defaults to ``off``,
+        controls whether the monitor responses are pretty
+        printed as multi-line indented JSON, as opposed to
+        constrained to a single line without extraneous
+        whitespace.
+
+        The ``close-action`` parameter, which defaults to ``none``,
+        controls what happens when the connection to the monitor
+        is terminated by the user. If set to ``delete``, then the
+        ``monitor-qmp`` object and its associated character
+        device are both immediately deleted. This can be useful
+        if an extra monitor was hotplugged for a specific task
+        and should be unplugged when completed.
+
     ``-object memory-backend-file,id=id,size=size,mem-path=dir,share=on|off,discard-data=on|off,merge=on|off,dump=on|off,prealloc=on|off,host-nodes=host-nodes,policy=default|preferred|bind|interleave,align=align,offset=offset,readonly=on|off,rom=on|off|auto``
         Creates a memory file backend object, which can be used to back
         the guest RAM with huge pages.
#56 lore raw

Re: [PATCH v5 33/35] tests: switch from -mon to -object monitor-qmp

In reply to Daniel P. Berrangé
Quoted 33 lines · Daniel P. Berrangé

Use the new preferred low level option for configuring the QMP service in libqtest and the python Machine class used by tests. This will avoid triggering deprecation warnings after the subsequent commit.

---
 python/qemu/machine/machine.py | 4 ++--
 tests/qtest/libqtest.c         | 2 +-
 2 files changed, 3 insertions(+), 3 deletions(-)
python/qemu/machine/machine.py +2 -2
index ebb58d5b68..5f4845f5fb 100644
@@ -304,8 +304,8 @@ def _base_args(self) -> List[str]:
                 )
             else:
                 moncdev = f"socket,id=mon,path={self._monitor_address}"
-            args.extend(['-chardev', moncdev, '-mon',
-                         'chardev=mon,mode=control'])
+            args.extend(['-chardev', moncdev, '-object',
+                         'monitor-qmp,id=qmp,chardev=mon'])
 
         if self._machine is not None:
             args.extend(['-machine', self._machine])
tests/qtest/libqtest.c +0 -0
index 4e22c66b75..c33c799c92 100644

tests/qtest/libqtest.c gchar *qtest_qemu_args(const char *extra_args) · line 466

Quoted 8 lines
tests/qtest/libqtest.c +1 -1
@@ -460,7 +460,7 @@ gchar *qtest_qemu_args(const char *extra_args)
                       "-qtest unix:%s "
                       "-qtest-log %s "
                       "-chardev socket,path=%s,id=char0 "
-                      "-mon chardev=char0,mode=control "
+                      "-object monitor-qmp,id=qmp0,chardev=char0 "
                       "-display none "
                       "-audio none "
                       "%s "

Have you considered tests/functional/qemu_test/testcase.py

if sockpath is not None:
vm.add_args("-chardev",
f"socket,id=backdoor,path={sockpath},server=on,wait=off", "-mon", "chardev=backdoor,mode=control")

and tests/tcg/{aarch64,arm}/Makefile.softmmu-target

QEMU_SEMIHOST=-serial none -chardev stdio,mux=on,id=stdio0 -semihosting-config enable=on,chardev=stdio0 -mon chardev=stdio0,mode=readline

?

#67 lore raw
In reply to Daniel P. Berrangé
Quoted 16 lines · Daniel P. Berrangé

Conceptually -object and object_add/object_del should be sufficient for essentially all QEMU configuration....if only we ported all our internal custom backends/devices/etc to QOM. That is of course a big job which is why it hasn't happened.

This series started with the premise that the monitor is one of the easier areas to convert since we have no more than three classes, a common base, and QMP and HMP subclasses[1]. So why not give it a go and thus unlock the ability to dynamically create/delete monitors in QMP/HMP.

This series does the conversion in a great many small steps to better understand the implications at each stage.

The high level outcome of this series is

Quoted 8 lines

* HMP and QMP monitors are QOM objects, 'monitor-hmp' and 'monitor-qmp' respectively

* Both can be cold plugged and hot plugged. QMP only, can also be hot unplugged.

* '-mon' is obsolete, deprecated and replaced by '-object', but -monitor, -qmp and kept as high level syntax sugar

Possible additional work: change monitor_parse() to build a MonitorOptions instead of a QemuOpts, so it call monitor_new() directly instead of via monitor_new_opts(). Observation, not a demand.

Quoted 3 lines

* QMP gains a concept of "close-action" which makes it possible to mark a monitor for auto-delete.

I'm not quite convinced this is warranted. I'd like to hear more about use cases.

Is libvirt going to make use of it?

Quoted 25 lines

The monitor hot-unplug code and the qtest and functional testing code is heavily derived from a series sent by Christian Brauner which proposed new monitor_add/monitor_del commands:

https://lists.nongnu.org/archive/html/qemu-devel/2026-04/msg01349.html

I left Christian's authorship & SoB on the patches which were derived from his code, though the code has been refactored quite a bit in places, so bugs are quite possibly my own.

Note that Christian's series allowed the use of "monitor_del" commands against the current monitor session. ie a client could delete the very monitor it was using. This is an awkward concept as it needs special casing to delay the deletion to happen in the background, such that that the QMP response to 'monitor_del' could still be sent back. This also left the chardev was orphaned as there's no way to run 'chardev_dev' in that usage pattern.

To provide an alternative mechanism to address the same use case, this series introduces the 'close-action' concept mentioned above, that allows hotplugging a monitor to service a specific task, with the monitor being purged when the script closes its connection. This avoids the special casing that an explicit "self deletion" paradigm required.

Quoted 6 lines

While supporting hotplug of HMP was trivial, I didn't do any work to think about hotunplug of HMP, since IMHO it is of limited value given HMP's typical use cases.

[1] ~~~ we did all this not because it was easy, but because we thought it would be easy ~~~

Heh!

Fully addressing all review comments before the soft freeze may or may not be possible. Perhaps punting sufficiently harmless changes to a post-freeze fixup series could help. Use your judgement.

We could also leave the close-action feature for the next development cycle.

#48 lore raw

Re: [PATCH v5 32/35] monitor: add support for auto-deleting monitors upon close

In reply to Markus Armbruster
Quoted 9 lines · Markus Armbruster
Quoted 6 lines · [email protected] writes:
Quoted 6 lines

The default monitor is usually a long lived object that will exist for the entire lifetime of the VM. A monitor can only service a single client at a time though, and so it might be desirable to hotplug additional monitors at runtime for specific tasks. If doing that, however, there is a need to remove the monitor when it is no longer needed.

Quoted 3 lines

Whatever adds the additional monitor can also delete it. The fact that you propose other means suggests you believe this would be cumbersome in practice. Why?

The use is is that someone/something wants to spawn a script that does some job with the QEMU monitor. The thing that spawns the script adds the new monitor and launches the script. Having auto-delete means that you do not need to then keep track of that script to perform cleanup of the dynamically added monitor. It gives you "do the right thing" behaviour automatically when the script exits, closing its monitor connection.

The initial series proposed by Christian supported the ability to run "object-del' on the monitor itself - a "self delete" essentially. That is very awkward from the code POV, as it required special case hanlding to ensure the QMP response to the delete action got sent on the socket before the delete action took place. It also made it impossible to then delete the character device.

Auto-delete gives us a better solution with less code complexity.

Quoted 6 lines
Quoted 6 lines
Quoted 6 lines

Allowing a client to run "object-del" against its own monitor adds complex edge cases, as it would be desirable to send the QMP response despite the monitor sending it being deleted. Doing "object-del" alone will also result in orphaning a character device backend instance, as there is no opportunity to run the companion "chardev-del" command.

Quoted 8 lines
Quoted 5 lines
Quoted 5 lines

A simpler way to ensure cleanup is to add the concept of auto-deleting monitor objects. Specifically when the "CHR_EVENT_CLOSED" event is emitted, the equivalent of "object-del" + "chardev-del" can be run internally. Since the transient client has already droppped its monitor connection, there is no synchronization to be concerned about.

If object-del or chardev-del fail, there's no way to report the error. Can they fail?

object-del can fail if

* An object with the specified "id" does not exist. That shouldn't happen in this case but harmless if it odes.
* object_del command tries to delete the monitor that is servicing the object_del command. Cannot happen with auto-delete
* the monitor has not finished initializing it BH with chardev handlers. Cannot happen if we know we have a live connection already.

chardev-del can fail if

* The chardev with "id" does not exist. SHouldn't happen but is harmless if it does
* The chardev reports it is "busy" - aka the frontend is still connected - we just deleted it so cannot happen
* Record/replay is in use - a niche use case

So I don't think errors are a problem.

Quoted 1 line

Do we always want to delete both monitor and character device?

IMHO yes they are a pair whose lifetime should be tied together for normal use.

Quoted 11 lines
Quoted 8 lines
Quoted 8 lines

This is implemented via a new "close-action=none|delete" property on the 'monitor-qmp' object. This concept could be extended with further actions in future, for example:

* close-action=shutdown - graceful guest shutdown
* close-action=terminate - immediate guest poweroff
* close-action=stop - pause guest CPUs while the monitor is not connected to any client

I'm not sure these would be useful and safe. Mentioning them here is okay regardless.

I think they'd be interesting for some "embedded" usage of QEMU such as libguestfs where you want QEMU to go away entirely when the process disconnects. But we have a differnt mechanism via '-run-with exit-with-parent=on' these days so we probably don't need two mechanisms for the same concept.

Quoted 14 lines
Quoted 14 lines
Quoted 14 lines
monitor/qmp.c +14 -0
+static void monitor_qmp_self_delete_bh(void *opaque)
+{
+    MonitorQMP *mon = opaque;
+    g_autofree char *mon_id = object_property_get_child_name(
+        object_get_objects_root(), OBJECT(mon));
+    g_autofree char *chardev_id = g_strdup(mon->parent_obj.chardev_id);
+    Error *local_error = NULL;
+
+    if (!mon_id) {
+        /* Another monitor raced & ran 'object-del' on 'mon'
+         * before this BH got scheduled, so we have a ref on
+         * mon but it is already unparented.
+         */
+        object_unref(mon);
Quoted 8 lines
Quoted 5 lines

monitor/qmp.c static QDict *qmp_greeting(MonitorQMP *mon) · line 591

Quoted 5 lines
monitor/qmp.c +5 -0
+        return;
+    }
+
+    user_creatable_del(mon_id, &local_error);
+    object_unref(mon);

This object_unref() and the one above pair with the object_ref() in monitor_qmp_event() case MONITOR_QMP_CLOSE_ACTION_DELETE. Correct?

Yes.

Quoted 1 line

Is this worth a comment?

Sure.

Quoted 42 lines
Quoted 5 lines
Quoted 5 lines

static void monitor_qmp_event(void *opaque, QEMUChrEvent event)

 {
     QDict *data;
     MonitorQMP *mon = opaque;
Quoted 7 lines

monitor/qmp.c static QDict *qmp_greeting(MonitorQMP *mon) · line 610

Quoted 7 lines
monitor/qmp.c +7 -0
+    /* Protect against race if a client drops & quickly
+     * reconnects - we'll have the delete BH scheduled
+     * so must not honour a new open request
+     */
+    if (mon->delete_pending) {
+        return;
+    }

Consider "something happens between setting @delete_pending and monitor destruction that triggers the event" and "same thing happens after destruction". Any observable differences in behavior?

Are there any other monitor interactions that might need similar special-casing when @delete_pending?

Quoted 11 lines

monitor/qmp.c static void monitor_qmp_event(void *opaque, QEMUChrEvent event) · line 636

Quoted 11 lines
monitor/qmp.c +4 -0
+
     switch (event) {
     case CHR_EVENT_OPENED:
         WITH_QEMU_LOCK_GUARD(&mon->parent_obj.mon_lock) {
@@ -577,6 +631,27 @@ static void monitor_qmp_event(void *opaque, QEMUChrEvent event)
         json_message_parser_init(&mon->parser, handle_qmp_command,
                                  mon, NULL);
         monitor_fdsets_cleanup();
+        switch (mon->close_action) {
+        case MONITOR_QMP_CLOSE_ACTION_NONE:
+            break; /* nada */

Useless comment :)

Quoted 7 lines

monitor/qmp.c static void monitor_qmp_event(void *opaque, QEMUChrEvent event) · line 643

Quoted 7 lines
monitor/qmp.c +7 -0
+        case MONITOR_QMP_CLOSE_ACTION_DELETE:
+            mon->delete_pending = true;
+            /*
+             * Do NOT run in the AIO context associated with the
+             * monitor. We need to run in the default AIO context
+             * which is the same context in which 'qmp_object_del'
+             * will execute
Quoted 2 lines

I believe you. I don't have the headspace right now to confirm this myself.

When I got this wrong initially bad things happened :-)

Quoted 6 lines
Quoted 4 lines

monitor/qmp.c static void monitor_qmp_event(void *opaque, QEMUChrEvent event) · line 646

Quoted 4 lines
monitor/qmp.c +3 -0
+             *
+             * Hold an extra ref in case a separate monitor races
+             * with the BH by processing an explicit 'object-del'

The extra ref ensures the object still lives when the BH runs. Correct?

Yes.

Quoted 5 lines
Quoted 3 lines

monitor/qmp.c static void monitor_qmp_event(void *opaque, QEMUChrEvent event) · line 648

Quoted 3 lines
monitor/qmp.c +2 -0
+             */
+            object_ref(mon);

This is the object_ref() I referred to above.

Yes.

Quoted 11 lines
Quoted 10 lines

monitor/qmp.c static void monitor_qmp_event(void *opaque, QEMUChrEvent event) · line 657

Quoted 10 lines
monitor/qmp.c +6 -0
+            aio_bh_schedule_oneshot(qemu_get_aio_context(),
+                                    monitor_qmp_self_delete_bh, mon);
+            break;
+        default:
+            g_assert_not_reached();
+        }
         break;
     case CHR_EVENT_BREAK:
     case CHR_EVENT_MUX_IN:
Quoted 4 lines

[...]

Quoted 1 line

With regards,
Daniel

Signature · 3 lines
|: https://berrange.com       ~~        https://hachyderm.io/@berrange :|
|: https://libvirt.org          ~~          https://entangle-photo.org :|
|: https://pixelfed.art/berrange   ~~    https://fstop138.berrange.com :|
#57 lore raw

Re: [PATCH v5 33/35] tests: switch from -mon to -object monitor-qmp

In reply to Markus Armbruster
Quoted 51 lines · Markus Armbruster
Quoted 33 lines · Daniel P. Berrangé

Use the new preferred low level option for configuring the QMP service in libqtest and the python Machine class used by tests. This will avoid triggering deprecation warnings after the subsequent commit.

---
 python/qemu/machine/machine.py | 4 ++--
 tests/qtest/libqtest.c         | 2 +-
 2 files changed, 3 insertions(+), 3 deletions(-)
python/qemu/machine/machine.py +2 -2
index ebb58d5b68..5f4845f5fb 100644
@@ -304,8 +304,8 @@ def _base_args(self) -> List[str]:
                 )
             else:
                 moncdev = f"socket,id=mon,path={self._monitor_address}"
-            args.extend(['-chardev', moncdev, '-mon',
-                         'chardev=mon,mode=control'])
+            args.extend(['-chardev', moncdev, '-object',
+                         'monitor-qmp,id=qmp,chardev=mon'])
 
         if self._machine is not None:
             args.extend(['-machine', self._machine])
tests/qtest/libqtest.c +0 -0
index 4e22c66b75..c33c799c92 100644

tests/qtest/libqtest.c gchar *qtest_qemu_args(const char *extra_args) · line 466

Quoted 8 lines
tests/qtest/libqtest.c +1 -1
@@ -460,7 +460,7 @@ gchar *qtest_qemu_args(const char *extra_args)
                       "-qtest unix:%s "
                       "-qtest-log %s "
                       "-chardev socket,path=%s,id=char0 "
-                      "-mon chardev=char0,mode=control "
+                      "-object monitor-qmp,id=qmp0,chardev=char0 "
                       "-display none "
                       "-audio none "
                       "%s "

Have you considered tests/functional/qemu_test/testcase.py

if sockpath is not None:
vm.add_args("-chardev",
f"socket,id=backdoor,path={sockpath},server=on,wait=off", "-mon", "chardev=backdoor,mode=control")

Quoted 5 lines

and tests/tcg/{aarch64,arm}/Makefile.softmmu-target

QEMU_SEMIHOST=-serial none -chardev stdio,mux=on,id=stdio0 -semihosting-config enable=on,chardev=stdio0 -mon chardev=stdio0,mode=readline

?

I didn't think about them but the code allows for mixing of -mon and -object, so that CI passes shows they've not seen problems.

We could convert them to -object too if we like but it doesn't hugely matter except from the pov of "best practice"

With regards,
Daniel

Signature · 3 lines
|: https://berrange.com       ~~        https://hachyderm.io/@berrange :|
|: https://libvirt.org          ~~          https://entangle-photo.org :|
|: https://pixelfed.art/berrange   ~~    https://fstop138.berrange.com :|
#60 lore raw

Re: [PATCH v5 34/35] qemu-options: document new monitor-hmp and monitor-qmp objects

In reply to Markus Armbruster
Quoted 20 lines · Markus Armbruster
Quoted 18 lines · Daniel P. Berrangé

Add new docs for the -object monitor-hmp and -object monitor-qmp options, updating -mon to state that it is legacy syntax sugar for the new -object args.

---
 qemu-options.hx | 49 +++++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 49 insertions(+)
qemu-options.hx +0 -0
index c799286153..a804d5b880 100644
@@ -4980,6 +4980,16 @@ SRST
       -mon chardev=mon1,mode=control,pretty=on
 
Quoted 15 lines

qemu-options.hx SRST · line 4993

Quoted 12 lines
qemu-options.hx +10 -0
     enables the QMP monitor on localhost port 4444 with pretty-printing.
+
+    The use of ``-mon mode=readline`` is historical syntax sugar
+    for the new ``-object monitor-hmp`` option, each use of which
+    creates an object with the ID ``hmpcompatNNN`` where ``NNN`` is
+    a counter starting from 0.
+
+    The use of ``-mon mode=control`` is historical syntax sugar
+    for the new ``-object monitor-qmp`` option, each use of which
+    creates an object with the ID ``qmpcompatNNN`` where ``NNN`` is
+    a counter starting from 0.
 ERST

Should we document how -monitor, -qmp, and -qmp-pretty map to -object and -chardev?

I thought they already described how they were mapped to -mon, but I see I was mistaken. It is worth a followup patch to improve their docs, especially given your points around the implicit "id" value they have internally.

With regards,
Daniel

Signature · 3 lines
|: https://berrange.com       ~~        https://hachyderm.io/@berrange :|
|: https://libvirt.org          ~~          https://entangle-photo.org :|
|: https://pixelfed.art/berrange   ~~    https://fstop138.berrange.com :|
#68 lore raw
In reply to Markus Armbruster
Quoted 32 lines · Markus Armbruster
Quoted 16 lines · Daniel P. Berrangé

Conceptually -object and object_add/object_del should be sufficient for essentially all QEMU configuration....if only we ported all our internal custom backends/devices/etc to QOM. That is of course a big job which is why it hasn't happened.

This series started with the premise that the monitor is one of the easier areas to convert since we have no more than three classes, a common base, and QMP and HMP subclasses[1]. So why not give it a go and thus unlock the ability to dynamically create/delete monitors in QMP/HMP.

This series does the conversion in a great many small steps to better understand the implications at each stage.

The high level outcome of this series is

Quoted 8 lines

* HMP and QMP monitors are QOM objects, 'monitor-hmp' and 'monitor-qmp' respectively

* Both can be cold plugged and hot plugged. QMP only, can also be hot unplugged.

* '-mon' is obsolete, deprecated and replaced by '-object', but -monitor, -qmp and kept as high level syntax sugar

Possible additional work: change monitor_parse() to build a MonitorOptions instead of a QemuOpts, so it call monitor_new() directly instead of via monitor_new_opts(). Observation, not a demand.

Quoted 1 line
Quoted 5 lines
Quoted 2 lines

* QMP gains a concept of "close-action" which makes it possible to mark a monitor for auto-delete.

I'm not quite convinced this is warranted. I'd like to hear more about use cases.

Replied about that inline to the patch. IMHO without this facility the series does not fully address the needs of systemd that motivated the original patches:

https://lists.nongnu.org/archive/html/qemu-devel/2026-04/msg01349.html

Quoted 1 line

Is libvirt going to make use of it?

This wasn't motivated by libvirt's needs, but indeed it could be useful for libvirt. Currently if apps want to access QMP, they need to use libvirt's QMP passthrough, which means they need special client code to talk to libvirt APIs.

With this libvirt could dynamically add a monitor, with auto close, open it and pass the open FD back to the client app. libvirt would not need to watch for when the client app is gone as dropping the FD would cleanup the monitor.

Quoted 6 lines

Fully addressing all review comments before the soft freeze may or may not be possible. Perhaps punting sufficiently harmless changes to a post-freeze fixup series could help. Use your judgement.

We could also leave the close-action feature for the next development cycle.

With regards,
Daniel

Signature · 3 lines
|: https://berrange.com       ~~        https://hachyderm.io/@berrange :|
|: https://libvirt.org          ~~          https://entangle-photo.org :|
|: https://pixelfed.art/berrange   ~~    https://fstop138.berrange.com :|
#42 lore raw

Re: [PATCH v5 30/35] qom: add method for getting the "id" of a QOM object

In reply to Daniel P. Berrangé
Quoted 51 lines · Daniel P. Berrangé

The 'user_creatable_del' method doesn't want the 'Object *', but instead the "id" of the object to be deleted. While most of its work could be done with the "Object *" alone, purging the QemuOpts values needs the "id".

The "id" is not recorded against an Object though, it is indirectly stored as a property linking the parent and child objects together. Add an accessor that can fetch this property name.

---
 include/qom/object.h | 10 ++++++++++
 qom/object.c         | 17 +++++++++++++++++
 2 files changed, 27 insertions(+)
include/qom/object.h +10 -0
index 11f55613fc..216d2a30f6 100644
@@ -1759,6 +1759,16 @@ ObjectProperty *object_property_try_add_child(Object *obj, const char *name,
 ObjectProperty *object_property_add_child(Object *obj, const char *name,
                                           Object *child);
 
+/**
+ * object_property_get_child_name:
+ * @obj: the object that owns the property
+ * @child: the object referenced by the child property
+ *
+ * Return the property name against which @child is registered
+ * with @obj, or NULL if @obj is not a parent of @child.
+ */
+char *object_property_get_child_name(Object *obj, Object *child);
+
 typedef enum {
     /* Unref the link pointer when the property is deleted */
     OBJ_PROP_LINK_STRONG = 0x1,
qom/object.c +5 -0
index f79b2cf361..66bbd1488b 100644
@@ -1952,6 +1952,23 @@ object_property_add_child(Object *obj, const char *name,
     return object_property_try_add_child(obj, name, child, &error_abort);
 }
 
+char *object_property_get_child_name(Object *obj, Object *child)
+{
+    ObjectProperty *prop;
+    GHashTableIter iter;
+    gpointer key, value;

qom/object.c object_property_add_child(Object *obj, const char *name, · line 1969

Quoted 10 lines
qom/object.c +10 -0
+
+    g_hash_table_iter_init(&iter, obj->properties);
+    while (g_hash_table_iter_next(&iter, &key, &value)) {
+        prop = value;
+        if (object_property_is_child(prop) && prop->opaque == child) {
+            return g_strdup(prop->name);
+        }
+    }
+    return NULL;
+}

Isn't this a slight variation of object_get_canonical_path_component()?

Quoted 5 lines
qom/object.c +2 -0
+
+
 void object_property_allow_set_link(const Object *obj, const char *name,
                                     Object *val, Error **errp)
 {
#43 lore raw

Re: [PATCH v5 30/35] qom: add method for getting the "id" of a QOM object

Quoted 57 lines · Markus Armbruster
Quoted 55 lines · Daniel P. Berrangé

The 'user_creatable_del' method doesn't want the 'Object *', but instead the "id" of the object to be deleted. While most of its work could be done with the "Object *" alone, purging the QemuOpts values needs the "id".

The "id" is not recorded against an Object though, it is indirectly stored as a property linking the parent and child objects together. Add an accessor that can fetch this property name.

---
 include/qom/object.h | 10 ++++++++++
 qom/object.c         | 17 +++++++++++++++++
 2 files changed, 27 insertions(+)
include/qom/object.h +10 -0
index 11f55613fc..216d2a30f6 100644
@@ -1759,6 +1759,16 @@ ObjectProperty *object_property_try_add_child(Object *obj, const char *name,
 ObjectProperty *object_property_add_child(Object *obj, const char *name,
                                           Object *child);
 
+/**
+ * object_property_get_child_name:
+ * @obj: the object that owns the property
+ * @child: the object referenced by the child property
+ *
+ * Return the property name against which @child is registered
+ * with @obj, or NULL if @obj is not a parent of @child.
+ */
+char *object_property_get_child_name(Object *obj, Object *child);
+
 typedef enum {
     /* Unref the link pointer when the property is deleted */
     OBJ_PROP_LINK_STRONG = 0x1,
qom/object.c +9 -0
index f79b2cf361..66bbd1488b 100644
@@ -1952,6 +1952,23 @@ object_property_add_child(Object *obj, const char *name,
     return object_property_try_add_child(obj, name, child, &error_abort);
 }
 
+char *object_property_get_child_name(Object *obj, Object *child)
+{
+    ObjectProperty *prop;
+    GHashTableIter iter;
+    gpointer key, value;
+
+    g_hash_table_iter_init(&iter, obj->properties);
+    while (g_hash_table_iter_next(&iter, &key, &value)) {
+        prop = value;
Quoted 8 lines

qom/object.c object_property_add_child(Object *obj, const char *name, · line 1969

Quoted 6 lines
qom/object.c +6 -0
+        if (object_property_is_child(prop) && prop->opaque == child) {
+            return g_strdup(prop->name);
+        }
+    }
+    return NULL;
+}

Isn't this a slight variation of object_get_canonical_path_component()?

Oh yes, you're right. That long method name didn't grab me as the right functionality, but in fact it is.

We could also use it to retain the better error message in the earlier patch you commented on where I lost the "id" value from the errors.

Quoted 7 lines

qom/object.c object_property_add_child(Object *obj, const char *name, · line 1974

Quoted 6 lines
qom/object.c +2 -0
+
+
 void object_property_allow_set_link(const Object *obj, const char *name,
                                     Object *val, Error **errp)
 {

With regards,
Daniel

Signature · 3 lines
|: https://berrange.com       ~~        https://hachyderm.io/@berrange :|
|: https://libvirt.org          ~~          https://entangle-photo.org :|
|: https://pixelfed.art/berrange   ~~    https://fstop138.berrange.com :|
#62 lore raw

Re: [PATCH v5 35/35] docs: mark '-mon' as deprecated in favour of -object

In reply to Daniel P. Berrangé
Quoted 25 lines · Daniel P. Berrangé

The high level -qmp and -monitor options can remain as convenience wrappers, but the low level -mon is completed obsoleted by the new -object support with 'monitor-qmp' and 'monitor-hmp' types.

---
 docs/about/deprecated.rst               | 10 ++++++++++
 docs/devel/writing-monitor-commands.rst |  4 ++--
 docs/system/arm/xenpvh.rst              |  4 ++--
 docs/system/i386/xen.rst                |  3 ++-
 docs/system/i386/xenpvh.rst             |  4 ++--
 qemu-options.hx                         | 11 ++++++-----
 system/vl.c                             |  2 ++
 7 files changed, 26 insertions(+), 12 deletions(-)
docs/about/deprecated.rst +0 -0
index 97750f5edc..8370fc5680 100644
@@ -61,6 +61,16 @@ The ``debug-threads`` option of the ``-name`` argument is now
 ignored. Thread naming is unconditionally enabled for all platforms
 where it is supported.
 

docs/about/deprecated.rst The ``debug-threads`` option of the ``-name`` argument is now · line 73

Quoted 10 lines
docs/about/deprecated.rst +10 -0
+``-mon`` option (since 11.1)
+''''''''''''''''''''''''''''
+
+The ``-mon`` option was the generic mechanism for creating monitor objects
+if the convenience ``-qmp`` or ``-monitor`` options were not flexible
+enough. The monitor objects have been converted to QOM, so ``-mon mode=readline``
+is replaced by ``-object monitor-hmp`` and ``-mon mode=control`` is replaced
+by ``-object monitor-qmp``. The short convenience options are not deprecated,
+only ``-mon``.
+

Suggest to wrap lines around column 70.

Quoted 3 lines
docs/about/deprecated.rst +0 -0
 QEMU Machine Protocol (QMP) commands
 ------------------------------------
 

[...]

#4 lore raw

Re: [PATCH v5 01/35] qom: replace 'can_be_deleted' with 'prepare_delete'

In reply to Daniel P. Berrangé
Quoted 51 lines · Daniel P. Berrangé
Quoted 50 lines · Markus Armbruster
Quoted 18 lines · Daniel P. Berrangé

While most objects can perform all their cleanup in the finalizer method, there can be interactions with other resources / subsystems / threads which require that some cleanup be performed on an user creatable object before unparenting it and entering finalization.

The current 'can_be_deleted' method runs in the deletion path and is intended to be used to block deletion. While it could be used to perform cleanup tasks, its name suggests it should be free of side-effects.

Generalize this by renaming it to 'prepare_delete', explicitly allowing for cleanup to be provided. Existing users of 'can_be_deleted' are re-written, which provides them with more detailed/tailored error messages.

[...]

Quoted 14 lines
qom/object_interfaces.c +2 -2
index 7080f85f95..6faa0b2fd9 100644
@@ -32,16 +32,15 @@ bool user_creatable_complete(UserCreatable *uc, Error **errp)
     return !*errp;
 }
 
-bool user_creatable_can_be_deleted(UserCreatable *uc)
+bool user_creatable_prepare_delete(UserCreatable *uc, Error **errp)
 {
-
     UserCreatableClass *ucc = USER_CREATABLE_GET_CLASS(uc);
+    ERRP_GUARD();

qom/object_interfaces.c bool user_creatable_complete(UserCreatable *uc, Error **errp) · line 44

Quoted 10 lines
qom/object_interfaces.c +3 -4
 
-    if (ucc->can_be_deleted) {
-        return ucc->can_be_deleted(uc);
-    } else {
-        return true;
+    if (ucc->prepare_delete) {
+        ucc->prepare_delete(uc, errp);
     }
+    return !*errp;
 }

Simpler:

Quoted 30 lines
Quoted 5 lines
       if (ucc->prepare_delete) {
           return ucc->prepare_delete(uc, errp);
       }

       return true;

ok

Quoted 22 lines

qom/object_interfaces.c bool user_creatable_del(const char *id, Error **errp) · line 255

Quoted 10 lines
qom/object_interfaces.c +1 -2
 void user_creatable_add_qapi(ObjectOptions *options, Error **errp)
@@ -253,8 +252,7 @@ bool user_creatable_del(const char *id, Error **errp)
         return false;
     }
 
-    if (!user_creatable_can_be_deleted(USER_CREATABLE(obj))) {
-        error_setg(errp, "object '%s' is in use, can not be deleted", id);
+    if (!user_creatable_prepare_delete(USER_CREATABLE(obj), errp)) {

This changes error messages.

The old ones mention @id.

The new ones don't:

        error_setg(errp, "Cryptodev backend is still in use");
        error_setg(errp, "Host memory backend is still mapped");
        error_setg(errp, "IOMMUFD backend still has %d users", be->users);
        error_setg(errp, "Throttle group still has multiple references");
    error_setg(errp, "Deleting can bus devices is not supported");

Spel it CAN, please.

Quoted 17 lines
Quoted 11 lines
    error_setg(errp, "Deleting main loop is not supported");

Visible in the diff to tests/qemu-iotests/245 below.

Because this runs within object-del, and @id is the argument provided by the user there, the error message is still sufficiently clear, I guess.

If this ever gets used where the object isn't obvious to the user from context, the error messages become sub-par. Observation, not a demand.

Please cover this change in the commit message.

Oh yes, I remember being slightly annoyed at the need to do this - I was surprised to realize that an Object does not actually know its own 'id'. The 'id' only exists as the name of the property in the parent.

Quoted 5 lines

We could add an object_get_id() API which traveres obj->parent and then iterates over child properties. Annoyingly O(n) time for that, but perhaps not a big enough problem to worry about ? At least for error reporting we're probably ok with that time complexity even if there are 1000 objects registered as children.

Could you use object_get_canonical_path_component()?

[...]

#5 lore raw

Re: [PATCH v5 01/35] qom: replace 'can_be_deleted' with 'prepare_delete'

Quoted 103 lines · Markus Armbruster
Quoted 51 lines · Daniel P. Berrangé
Quoted 50 lines · Markus Armbruster
Quoted 18 lines · Daniel P. Berrangé

While most objects can perform all their cleanup in the finalizer method, there can be interactions with other resources / subsystems / threads which require that some cleanup be performed on an user creatable object before unparenting it and entering finalization.

The current 'can_be_deleted' method runs in the deletion path and is intended to be used to block deletion. While it could be used to perform cleanup tasks, its name suggests it should be free of side-effects.

Generalize this by renaming it to 'prepare_delete', explicitly allowing for cleanup to be provided. Existing users of 'can_be_deleted' are re-written, which provides them with more detailed/tailored error messages.

[...]

Quoted 14 lines
qom/object_interfaces.c +2 -2
index 7080f85f95..6faa0b2fd9 100644
@@ -32,16 +32,15 @@ bool user_creatable_complete(UserCreatable *uc, Error **errp)
     return !*errp;
 }
 
-bool user_creatable_can_be_deleted(UserCreatable *uc)
+bool user_creatable_prepare_delete(UserCreatable *uc, Error **errp)
 {
-
     UserCreatableClass *ucc = USER_CREATABLE_GET_CLASS(uc);
+    ERRP_GUARD();

qom/object_interfaces.c bool user_creatable_complete(UserCreatable *uc, Error **errp) · line 44

Quoted 10 lines
qom/object_interfaces.c +3 -4
 
-    if (ucc->can_be_deleted) {
-        return ucc->can_be_deleted(uc);
-    } else {
-        return true;
+    if (ucc->prepare_delete) {
+        ucc->prepare_delete(uc, errp);
     }
+    return !*errp;
 }

Simpler:

Quoted 30 lines
Quoted 5 lines
       if (ucc->prepare_delete) {
           return ucc->prepare_delete(uc, errp);
       }

       return true;

ok

Quoted 22 lines

qom/object_interfaces.c bool user_creatable_del(const char *id, Error **errp) · line 255

Quoted 10 lines
qom/object_interfaces.c +1 -2
 void user_creatable_add_qapi(ObjectOptions *options, Error **errp)
@@ -253,8 +252,7 @@ bool user_creatable_del(const char *id, Error **errp)
         return false;
     }
 
-    if (!user_creatable_can_be_deleted(USER_CREATABLE(obj))) {
-        error_setg(errp, "object '%s' is in use, can not be deleted", id);
+    if (!user_creatable_prepare_delete(USER_CREATABLE(obj), errp)) {

This changes error messages.

The old ones mention @id.

The new ones don't:

        error_setg(errp, "Cryptodev backend is still in use");
        error_setg(errp, "Host memory backend is still mapped");
        error_setg(errp, "IOMMUFD backend still has %d users", be->users);
        error_setg(errp, "Throttle group still has multiple references");
    error_setg(errp, "Deleting can bus devices is not supported");

Spel it CAN, please.

Quoted 17 lines
Quoted 11 lines
    error_setg(errp, "Deleting main loop is not supported");

Visible in the diff to tests/qemu-iotests/245 below.

Because this runs within object-del, and @id is the argument provided by the user there, the error message is still sufficiently clear, I guess.

If this ever gets used where the object isn't obvious to the user from context, the error messages become sub-par. Observation, not a demand.

Please cover this change in the commit message.

Oh yes, I remember being slightly annoyed at the need to do this - I was surprised to realize that an Object does not actually know its own 'id'. The 'id' only exists as the name of the property in the parent.

Quoted 7 lines
Quoted 5 lines

We could add an object_get_id() API which traveres obj->parent and then iterates over child properties. Annoyingly O(n) time for that, but perhaps not a big enough problem to worry about ? At least for error reporting we're probably ok with that time complexity even if there are 1000 objects registered as children.

Could you use object_get_canonical_path_component()?

Yes !

With regards,
Daniel

Signature · 3 lines
|: https://berrange.com       ~~        https://hachyderm.io/@berrange :|
|: https://libvirt.org          ~~          https://entangle-photo.org :|
|: https://pixelfed.art/berrange   ~~    https://fstop138.berrange.com :|
#12 lore raw

Re: [PATCH v5 05/35] monitor: minimal conversion of monitors to QOM

In reply to Daniel P. Berrangé
Quoted 105 lines · Daniel P. Berrangé
Quoted 96 lines · Markus Armbruster
Quoted 66 lines · Daniel P. Berrangé

This introduces a Monitor QOM object, with MonitorHMP and MonitorQMP subclasses. This is the bare minimum conversion of just the type declarations and replacing g_new/g_free with object_new/object_unref.

---
 include/monitor/monitor.h  | 11 ++++++++++-
 monitor/hmp.c              | 29 +++++++++++++++++++++++++++--
 monitor/monitor-internal.h | 18 ++++++++++++++++--
 monitor/monitor.c          | 18 ++++++++++++++++--
 monitor/qmp-cmds.c         | 15 ++++++++-------
 monitor/qmp.c              | 29 +++++++++++++++++++++++++++--
 6 files changed, 104 insertions(+), 16 deletions(-)
include/monitor/monitor.h +10 -1
index b9642b58ba..2e9f9e12e9 100644
@@ -5,8 +5,17 @@
 #include "qapi/qapi-types-misc.h"
 #include "qemu/readline.h"
 #include "exec/hwaddr.h"
+#include "qom/object.h"
+
+#define TYPE_MONITOR "monitor"
+OBJECT_DECLARE_TYPE(Monitor, MonitorClass, MONITOR);
+
+#define TYPE_MONITOR_HMP "monitor-hmp"
+OBJECT_DECLARE_TYPE(MonitorHMP, MonitorHMPClass, MONITOR_HMP);
+
+#define TYPE_MONITOR_QMP "monitor-qmp"
+OBJECT_DECLARE_TYPE(MonitorQMP, MonitorQMPClass, MONITOR_QMP);
 
-typedef struct MonitorHMP MonitorHMP;
 typedef struct MonitorOptions MonitorOptions;
 
 #define QMP_REQ_QUEUE_LEN_MAX 8
monitor/hmp.c +14 -0
index 4e4468424a..81047d2513 100644
@@ -43,6 +43,20 @@
 #include "system/block-backend.h"
 #include "trace.h"
 
+OBJECT_DEFINE_TYPE(MonitorHMP, monitor_hmp, MONITOR_HMP, MONITOR);
+
+static void monitor_hmp_finalize(Object *obj)
+{
+}
+
+static void monitor_hmp_class_init(ObjectClass *cls, const void *data)
+{
+}
+
+static void monitor_hmp_init(Object *obj)
+{
+}
+
 static void monitor_command_cb(void *opaque, const char *cmdline,
                                void *readline_opaque)
 {

monitor/hmp.c static void monitor_readline_flush(void *opaque) · line 1545

Quoted 7 lines
monitor/hmp.c +3 -1
@@ -1526,10 +1540,21 @@ static void monitor_readline_flush(void *opaque)
 
 void monitor_new_hmp(Chardev *chr, bool use_readline, Error **errp)
 {
-    MonitorHMP *mon = g_new0(MonitorHMP, 1);
+    MonitorHMP *mon;
+    static int counter;
+    g_autofree char *id = g_strdup_printf("hmpcompat%d", counter++);

Hmm. The system picking IDs is problematic when they can clash with the user's IDs. If we had an ounce of common sense, we'd restrict both across the board so they cannot clash. But we don't.

We need an ID here, because we need to make the new object the child of something (actually: child of /objects/), which requires a child name.

Non-problem with -object / object-add, because @id is mandatory there.

Non-problem with -device / device_add, because we use separate parents for devices with and without @id (/machine/peripheral/ and /machine/peripheral-anon/, plus the /machine/unattached/ orphanage).

Example for an existing problem:

    $ qemu-system-x86_64 -nodefaults -monitor stdio -chardev null,id=chr0 -mon id=compat_monitor0,chardev=chr0
    qemu-system-x86_64: -mon id=compat_monitor0,chardev=chr0: Duplicate ID 'compat_monitor0' for mon

Example for a problem created by this series:

Quoted 7 lines
    $  qemu-system-x86_64 -nodefaults -display none -S -monitor stdio -chardev null,id=chr0 -object monitor-hmp,id=hmpcompat0,chardev=chr0
    qemu-system-x86_64: -monitor stdio: attempt to add duplicate property 'hmpcompat0' to object (type 'container')

I readily admit that these clashes are *unlikely*. Still, do we really want to define an interface that claims to let you pick any ID, then rejects some of them sometimes? Feels rather 1990s to me. At the very least, cover the wart in the commit message.

Quoted 5 lines

The way I looked at it was aything using -object with the new monitor-qmp/monitor-hmp types is new code. They can:

1. Trivially abide the warning about "hmpcompatNN" / "qmpcompatNN" being internal usage for compat syntax

A wart you can easily cope with is still a wart. Oh well, we'll live. However, the wart should be documented.

Quoted 2 lines

2. Not use both -monitor and -object on the same QEMU instance so not have a clash between the two to begin with

Management applications aware of the monitor QOM types should use -object / object-add exclusively.

Human users will use -monitor to save typing. If they then need to plug an additional one at run time, they have no choice but use object-add. Again, we'll live.

Quoted 4 lines

The remaining danger where is some existing code using -object with a *non-monitor* type, and calling it "hmpcompatNN" / "qmpcompatNN", which would be insanity. Never say never, but I think that's an acceptable risk.

The risk is acceptable. The documentation complexity that comes with the wart bothers me.

Could we use id_generate()?

Quoted 38 lines
Quoted 5 lines

This existing problem example leads me to the next mess: interaction with monitors' *other* ID.

qemu-system-FOO's -mon accepts an optional "id" parameter. It goes into its QemuOpts, and from there into MonitorOptions member @id.

Arrrggggggghh. That is not documented for -mon at all AFAICT

  $ qemu-system-x86_64  -help 2>&1 | grep -- -mon
  -monitor dev    redirect the monitor to char device 'dev'
  -qmp dev        like -monitor but opens in 'control' mode
  -mon [chardev=]name[,mode=readline|control][,pretty[=on|off]]

but yeah as you say, it exists, and in fact libvirt even uses it despite it being undocumented.

Quoted 9 lines

qemu-storage-daemon's --monitor is similar, except it bypasses QemuOpts.

qemu-system-FOO provides convenience options -monitor, -qmp, -qmp-pretty.

Their argument may refer to an existing chardev by ID, like "chardev:ID". This creates a monitor with that same[*] QemuOpts and MonitorOptions ID.

Quoted 7 lines

Else, their argument is character device configuration in legacy syntax, like "stdio". This creates both a monitor and a character device, with ID "compat_monitorN", where N counts up from zero. The character device is visible in "info chardev", as always.

Aside: in both cases we use the same ID for two different objects, which feels unadvisable.

Agreed, that's awful.

Quoted 2 lines

Aside: we have code checking whether a QemuOpts or character device ID starts with "compat_monitor", which is horryfying.

Quoted 6 lines

Eww.

Quoted 2 lines

Your series does not mess with this at all. Understandable; I stay out of this swamp when I can, too.

I didn't realize the swap was there in this case !

Hope you brought insect repellent!

Quoted 10 lines
Quoted 7 lines

However, it results in monitors having two IDs, namely the one in MonitorOptions, and the one in /object/. This is confusing.

Perhaps we should we'd get rid of the one in MonitorOptions. May well be more trouble than it's worth.

Could we at least make the two IDs the same?

Since -mon has an existing ID, we might as well pass it through to use as the child prop name for the objects.

Makes sense.

[...]

#13 lore raw

Re: [PATCH v5 05/35] monitor: minimal conversion of monitors to QOM

Quoted 131 lines · Markus Armbruster
Quoted 105 lines · Daniel P. Berrangé
Quoted 96 lines · Markus Armbruster
Quoted 66 lines · Daniel P. Berrangé

This introduces a Monitor QOM object, with MonitorHMP and MonitorQMP subclasses. This is the bare minimum conversion of just the type declarations and replacing g_new/g_free with object_new/object_unref.

---
 include/monitor/monitor.h  | 11 ++++++++++-
 monitor/hmp.c              | 29 +++++++++++++++++++++++++++--
 monitor/monitor-internal.h | 18 ++++++++++++++++--
 monitor/monitor.c          | 18 ++++++++++++++++--
 monitor/qmp-cmds.c         | 15 ++++++++-------
 monitor/qmp.c              | 29 +++++++++++++++++++++++++++--
 6 files changed, 104 insertions(+), 16 deletions(-)
include/monitor/monitor.h +10 -1
index b9642b58ba..2e9f9e12e9 100644
@@ -5,8 +5,17 @@
 #include "qapi/qapi-types-misc.h"
 #include "qemu/readline.h"
 #include "exec/hwaddr.h"
+#include "qom/object.h"
+
+#define TYPE_MONITOR "monitor"
+OBJECT_DECLARE_TYPE(Monitor, MonitorClass, MONITOR);
+
+#define TYPE_MONITOR_HMP "monitor-hmp"
+OBJECT_DECLARE_TYPE(MonitorHMP, MonitorHMPClass, MONITOR_HMP);
+
+#define TYPE_MONITOR_QMP "monitor-qmp"
+OBJECT_DECLARE_TYPE(MonitorQMP, MonitorQMPClass, MONITOR_QMP);
 
-typedef struct MonitorHMP MonitorHMP;
 typedef struct MonitorOptions MonitorOptions;
 
 #define QMP_REQ_QUEUE_LEN_MAX 8
monitor/hmp.c +14 -0
index 4e4468424a..81047d2513 100644
@@ -43,6 +43,20 @@
 #include "system/block-backend.h"
 #include "trace.h"
 
+OBJECT_DEFINE_TYPE(MonitorHMP, monitor_hmp, MONITOR_HMP, MONITOR);
+
+static void monitor_hmp_finalize(Object *obj)
+{
+}
+
+static void monitor_hmp_class_init(ObjectClass *cls, const void *data)
+{
+}
+
+static void monitor_hmp_init(Object *obj)
+{
+}
+
 static void monitor_command_cb(void *opaque, const char *cmdline,
                                void *readline_opaque)
 {

monitor/hmp.c static void monitor_readline_flush(void *opaque) · line 1545

Quoted 7 lines
monitor/hmp.c +3 -1
@@ -1526,10 +1540,21 @@ static void monitor_readline_flush(void *opaque)
 
 void monitor_new_hmp(Chardev *chr, bool use_readline, Error **errp)
 {
-    MonitorHMP *mon = g_new0(MonitorHMP, 1);
+    MonitorHMP *mon;
+    static int counter;
+    g_autofree char *id = g_strdup_printf("hmpcompat%d", counter++);

Hmm. The system picking IDs is problematic when they can clash with the user's IDs. If we had an ounce of common sense, we'd restrict both across the board so they cannot clash. But we don't.

We need an ID here, because we need to make the new object the child of something (actually: child of /objects/), which requires a child name.

Non-problem with -object / object-add, because @id is mandatory there.

Non-problem with -device / device_add, because we use separate parents for devices with and without @id (/machine/peripheral/ and /machine/peripheral-anon/, plus the /machine/unattached/ orphanage).

Example for an existing problem:

    $ qemu-system-x86_64 -nodefaults -monitor stdio -chardev null,id=chr0 -mon id=compat_monitor0,chardev=chr0
    qemu-system-x86_64: -mon id=compat_monitor0,chardev=chr0: Duplicate ID 'compat_monitor0' for mon

Example for a problem created by this series:

Quoted 7 lines
    $  qemu-system-x86_64 -nodefaults -display none -S -monitor stdio -chardev null,id=chr0 -object monitor-hmp,id=hmpcompat0,chardev=chr0
    qemu-system-x86_64: -monitor stdio: attempt to add duplicate property 'hmpcompat0' to object (type 'container')

I readily admit that these clashes are *unlikely*. Still, do we really want to define an interface that claims to let you pick any ID, then rejects some of them sometimes? Feels rather 1990s to me. At the very least, cover the wart in the commit message.

Quoted 5 lines

The way I looked at it was aything using -object with the new monitor-qmp/monitor-hmp types is new code. They can:

1. Trivially abide the warning about "hmpcompatNN" / "qmpcompatNN" being internal usage for compat syntax

A wart you can easily cope with is still a wart. Oh well, we'll live. However, the wart should be documented.

Quoted 2 lines

2. Not use both -monitor and -object on the same QEMU instance so not have a clash between the two to begin with

Management applications aware of the monitor QOM types should use -object / object-add exclusively.

Human users will use -monitor to save typing. If they then need to plug an additional one at run time, they have no choice but use object-add. Again, we'll live.

Quoted 4 lines

The remaining danger where is some existing code using -object with a *non-monitor* type, and calling it "hmpcompatNN" / "qmpcompatNN", which would be insanity. Never say never, but I think that's an acceptable risk.

Quoted 4 lines

The risk is acceptable. The documentation complexity that comes with the wart bothers me.

Could we use id_generate()?

Yes, we could.

With regards,
Daniel

Signature · 3 lines
|: https://berrange.com       ~~        https://hachyderm.io/@berrange :|
|: https://libvirt.org          ~~          https://entangle-photo.org :|
|: https://pixelfed.art/berrange   ~~    https://fstop138.berrange.com :|
#49 lore raw

Re: [PATCH v5 32/35] monitor: add support for auto-deleting monitors upon close

In reply to Daniel P. Berrangé
Quoted 30 lines · Daniel P. Berrangé
Quoted 9 lines · Markus Armbruster
Quoted 6 lines · [email protected] writes:
Quoted 6 lines

The default monitor is usually a long lived object that will exist for the entire lifetime of the VM. A monitor can only service a single client at a time though, and so it might be desirable to hotplug additional monitors at runtime for specific tasks. If doing that, however, there is a need to remove the monitor when it is no longer needed.

Quoted 3 lines

Whatever adds the additional monitor can also delete it. The fact that you propose other means suggests you believe this would be cumbersome in practice. Why?

The use is is that someone/something wants to spawn a script that does some job with the QEMU monitor. The thing that spawns the script adds the new monitor and launches the script. Having auto-delete means that you do not need to then keep track of that script to perform cleanup of the dynamically added monitor. It gives you "do the right thing" behaviour automatically when the script exits, closing its monitor connection.

The initial series proposed by Christian supported the ability to run "object-del' on the monitor itself - a "self delete" essentially. That is very awkward from the code POV, as it required special case hanlding to ensure the QMP response to the delete action got sent on the socket before the delete action took place. It also made it impossible to then delete the character device.

Quoted 1 line

Auto-delete gives us a better solution with less code complexity.

I agree "self delete" is problematic.

I wonder how important the ability to fire and forget a script with a dedicated monitor is. I'd (naively?) expect whatever spawns the script to reap its exit status.

If it is important, what about fire and forget a script with a dedicated character device? Can't come up with a use case for that within ten seconds. However, we use character devices for all kinds of crap. Food for thought, not a demand.

Quoted 34 lines
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines

Allowing a client to run "object-del" against its own monitor adds complex edge cases, as it would be desirable to send the QMP response despite the monitor sending it being deleted. Doing "object-del" alone will also result in orphaning a character device backend instance, as there is no opportunity to run the companion "chardev-del" command.

Quoted 8 lines
Quoted 5 lines
Quoted 5 lines

A simpler way to ensure cleanup is to add the concept of auto-deleting monitor objects. Specifically when the "CHR_EVENT_CLOSED" event is emitted, the equivalent of "object-del" + "chardev-del" can be run internally. Since the transient client has already droppped its monitor connection, there is no synchronization to be concerned about.

If object-del or chardev-del fail, there's no way to report the error. Can they fail?

object-del can fail if

* An object with the specified "id" does not exist. That shouldn't happen in this case but harmless if it odes.
* object_del command tries to delete the monitor that is servicing the object_del command. Cannot happen with auto-delete
* the monitor has not finished initializing it BH with chardev handlers. Cannot happen if we know we have a live connection already.

chardev-del can fail if

* The chardev with "id" does not exist. SHouldn't happen but is harmless if it does
* The chardev reports it is "busy" - aka the frontend is still connected - we just deleted it so cannot happen
* Record/replay is in use - a niche use case

Quoted 6 lines

So I don't think errors are a problem.

Quoted 1 line

Do we always want to delete both monitor and character device?

IMHO yes they are a pair whose lifetime should be tied together for normal use.

Would we make monitor auto-delete delete its character device if character devices also had an auto-delete feature?

[...]

#69 lore raw
In reply to Daniel P. Berrangé
Quoted 52 lines · Daniel P. Berrangé
Quoted 32 lines · Markus Armbruster
Quoted 16 lines · Daniel P. Berrangé

Conceptually -object and object_add/object_del should be sufficient for essentially all QEMU configuration....if only we ported all our internal custom backends/devices/etc to QOM. That is of course a big job which is why it hasn't happened.

This series started with the premise that the monitor is one of the easier areas to convert since we have no more than three classes, a common base, and QMP and HMP subclasses[1]. So why not give it a go and thus unlock the ability to dynamically create/delete monitors in QMP/HMP.

This series does the conversion in a great many small steps to better understand the implications at each stage.

The high level outcome of this series is

Quoted 8 lines

* HMP and QMP monitors are QOM objects, 'monitor-hmp' and 'monitor-qmp' respectively

* Both can be cold plugged and hot plugged. QMP only, can also be hot unplugged.

* '-mon' is obsolete, deprecated and replaced by '-object', but -monitor, -qmp and kept as high level syntax sugar

Possible additional work: change monitor_parse() to build a MonitorOptions instead of a QemuOpts, so it call monitor_new() directly instead of via monitor_new_opts(). Observation, not a demand.

Quoted 1 line
Quoted 5 lines
Quoted 2 lines

* QMP gains a concept of "close-action" which makes it possible to mark a monitor for auto-delete.

I'm not quite convinced this is warranted. I'd like to hear more about use cases.

Replied about that inline to the patch. IMHO without this facility the series does not fully address the needs of systemd that motivated the original patches:

https://lists.nongnu.org/archive/html/qemu-devel/2026-04/msg01349.html

Quoted 1 line

Is libvirt going to make use of it?

This wasn't motivated by libvirt's needs, but indeed it could be useful for libvirt. Currently if apps want to access QMP, they need to use libvirt's QMP passthrough, which means they need special client code to talk to libvirt APIs.

Quoted 4 lines

With this libvirt could dynamically add a monitor, with auto close, open it and pass the open FD back to the client app. libvirt would not need to watch for when the client app is gone as dropping the FD would cleanup the monitor.

Alright, that's fair. The management application doesn't spawn something that wants a dedicated monitor, it makes one for a client.

Would this fit into a commit message somewhere?

[...]

#50 lore raw

Re: [PATCH v5 32/35] monitor: add support for auto-deleting monitors upon close

In reply to Markus Armbruster
Quoted 36 lines · Markus Armbruster
Quoted 30 lines · Daniel P. Berrangé
Quoted 9 lines · Markus Armbruster
Quoted 6 lines · [email protected] writes:
Quoted 6 lines

The default monitor is usually a long lived object that will exist for the entire lifetime of the VM. A monitor can only service a single client at a time though, and so it might be desirable to hotplug additional monitors at runtime for specific tasks. If doing that, however, there is a need to remove the monitor when it is no longer needed.

Quoted 3 lines

Whatever adds the additional monitor can also delete it. The fact that you propose other means suggests you believe this would be cumbersome in practice. Why?

The use is is that someone/something wants to spawn a script that does some job with the QEMU monitor. The thing that spawns the script adds the new monitor and launches the script. Having auto-delete means that you do not need to then keep track of that script to perform cleanup of the dynamically added monitor. It gives you "do the right thing" behaviour automatically when the script exits, closing its monitor connection.

The initial series proposed by Christian supported the ability to run "object-del' on the monitor itself - a "self delete" essentially. That is very awkward from the code POV, as it required special case hanlding to ensure the QMP response to the delete action got sent on the socket before the delete action took place. It also made it impossible to then delete the character device.

Quoted 1 line

Auto-delete gives us a better solution with less code complexity.

I agree "self delete" is problematic.

Quoted 8 lines

I wonder how important the ability to fire and forget a script with a dedicated monitor is. I'd (naively?) expect whatever spawns the script to reap its exit status.

If it is important, what about fire and forget a script with a dedicated character device? Can't come up with a use case for that within ten seconds. However, we use character devices for all kinds of crap. Food for thought, not a demand.

In a hypothetical libvirt usage of this functionality, libvirt would not be spawning the script or even know what its PID is. Libvirt would just create a auto-delete monitor, open an FD to it, and pass the FD back to the libvirt client & forget about it.

Quoted 36 lines
Quoted 36 lines
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines

Allowing a client to run "object-del" against its own monitor adds complex edge cases, as it would be desirable to send the QMP response despite the monitor sending it being deleted. Doing "object-del" alone will also result in orphaning a character device backend instance, as there is no opportunity to run the companion "chardev-del" command.

Quoted 8 lines
Quoted 5 lines
Quoted 5 lines

A simpler way to ensure cleanup is to add the concept of auto-deleting monitor objects. Specifically when the "CHR_EVENT_CLOSED" event is emitted, the equivalent of "object-del" + "chardev-del" can be run internally. Since the transient client has already droppped its monitor connection, there is no synchronization to be concerned about.

If object-del or chardev-del fail, there's no way to report the error. Can they fail?

object-del can fail if

* An object with the specified "id" does not exist. That shouldn't happen in this case but harmless if it odes.
* object_del command tries to delete the monitor that is servicing the object_del command. Cannot happen with auto-delete
* the monitor has not finished initializing it BH with chardev handlers. Cannot happen if we know we have a live connection already.

chardev-del can fail if

* The chardev with "id" does not exist. SHouldn't happen but is harmless if it does
* The chardev reports it is "busy" - aka the frontend is still connected - we just deleted it so cannot happen
* Record/replay is in use - a niche use case

So I don't think errors are a problem.

Quoted 7 lines
Quoted 4 lines
Quoted 1 line

Do we always want to delete both monitor and character device?

IMHO yes they are a pair whose lifetime should be tied together for normal use.

Would we make monitor auto-delete delete its character device if character devices also had an auto-delete feature?

That's a possibility.

We would have to accept that it allows users to create a scenario in which auto-delete will frequenty fail.

ie configuring the chardev for auto-delete, but not having the frontend auto-delete, which would trigger the "busy" check.

Not convinced that's nicer ?

With regards,
Daniel

Signature · 3 lines
|: https://berrange.com       ~~        https://hachyderm.io/@berrange :|
|: https://libvirt.org          ~~          https://entangle-photo.org :|
|: https://pixelfed.art/berrange   ~~    https://fstop138.berrange.com :|
#70 lore raw
In reply to Markus Armbruster
Quoted 59 lines · Markus Armbruster
Quoted 52 lines · Daniel P. Berrangé
Quoted 32 lines · Markus Armbruster
Quoted 16 lines · Daniel P. Berrangé

Conceptually -object and object_add/object_del should be sufficient for essentially all QEMU configuration....if only we ported all our internal custom backends/devices/etc to QOM. That is of course a big job which is why it hasn't happened.

This series started with the premise that the monitor is one of the easier areas to convert since we have no more than three classes, a common base, and QMP and HMP subclasses[1]. So why not give it a go and thus unlock the ability to dynamically create/delete monitors in QMP/HMP.

This series does the conversion in a great many small steps to better understand the implications at each stage.

The high level outcome of this series is

Quoted 8 lines

* HMP and QMP monitors are QOM objects, 'monitor-hmp' and 'monitor-qmp' respectively

* Both can be cold plugged and hot plugged. QMP only, can also be hot unplugged.

* '-mon' is obsolete, deprecated and replaced by '-object', but -monitor, -qmp and kept as high level syntax sugar

Possible additional work: change monitor_parse() to build a MonitorOptions instead of a QemuOpts, so it call monitor_new() directly instead of via monitor_new_opts(). Observation, not a demand.

Quoted 1 line
Quoted 5 lines
Quoted 2 lines

* QMP gains a concept of "close-action" which makes it possible to mark a monitor for auto-delete.

I'm not quite convinced this is warranted. I'd like to hear more about use cases.

Replied about that inline to the patch. IMHO without this facility the series does not fully address the needs of systemd that motivated the original patches:

https://lists.nongnu.org/archive/html/qemu-devel/2026-04/msg01349.html

Quoted 1 line

Is libvirt going to make use of it?

This wasn't motivated by libvirt's needs, but indeed it could be useful for libvirt. Currently if apps want to access QMP, they need to use libvirt's QMP passthrough, which means they need special client code to talk to libvirt APIs.

Quoted 4 lines

With this libvirt could dynamically add a monitor, with auto close, open it and pass the open FD back to the client app. libvirt would not need to watch for when the client app is gone as dropping the FD would cleanup the monitor.

Quoted 4 lines

Alright, that's fair. The management application doesn't spawn something that wants a dedicated monitor, it makes one for a client.

Would this fit into a commit message somewhere?

Sure, it fits in the commit message for the auto-delete patch.

With regards,
Daniel

Signature · 3 lines
|: https://berrange.com       ~~        https://hachyderm.io/@berrange :|
|: https://libvirt.org          ~~          https://entangle-photo.org :|
|: https://pixelfed.art/berrange   ~~    https://fstop138.berrange.com :|
#51 lore raw

Re: [PATCH v5 32/35] monitor: add support for auto-deleting monitors upon close

In reply to Daniel P. Berrangé
Quoted 95 lines · * Daniel P. Berrangé ([email protected]) wrote:
Quoted 36 lines · Markus Armbruster
Quoted 30 lines · Daniel P. Berrangé
Quoted 9 lines · Markus Armbruster
Quoted 6 lines · [email protected] writes:
Quoted 6 lines

The default monitor is usually a long lived object that will exist for the entire lifetime of the VM. A monitor can only service a single client at a time though, and so it might be desirable to hotplug additional monitors at runtime for specific tasks. If doing that, however, there is a need to remove the monitor when it is no longer needed.

Quoted 3 lines

Whatever adds the additional monitor can also delete it. The fact that you propose other means suggests you believe this would be cumbersome in practice. Why?

The use is is that someone/something wants to spawn a script that does some job with the QEMU monitor. The thing that spawns the script adds the new monitor and launches the script. Having auto-delete means that you do not need to then keep track of that script to perform cleanup of the dynamically added monitor. It gives you "do the right thing" behaviour automatically when the script exits, closing its monitor connection.

The initial series proposed by Christian supported the ability to run "object-del' on the monitor itself - a "self delete" essentially. That is very awkward from the code POV, as it required special case hanlding to ensure the QMP response to the delete action got sent on the socket before the delete action took place. It also made it impossible to then delete the character device.

Quoted 1 line

Auto-delete gives us a better solution with less code complexity.

I agree "self delete" is problematic.

Quoted 8 lines

I wonder how important the ability to fire and forget a script with a dedicated monitor is. I'd (naively?) expect whatever spawns the script to reap its exit status.

If it is important, what about fire and forget a script with a dedicated character device? Can't come up with a use case for that within ten seconds. However, we use character devices for all kinds of crap. Food for thought, not a demand.

In a hypothetical libvirt usage of this functionality, libvirt would not be spawning the script or even know what its PID is. Libvirt would just create a auto-delete monitor, open an FD to it, and pass the FD back to the libvirt client & forget about it.

Quoted 36 lines
Quoted 36 lines
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines

Allowing a client to run "object-del" against its own monitor adds complex edge cases, as it would be desirable to send the QMP response despite the monitor sending it being deleted. Doing "object-del" alone will also result in orphaning a character device backend instance, as there is no opportunity to run the companion "chardev-del" command.

Quoted 8 lines
Quoted 5 lines
Quoted 5 lines

A simpler way to ensure cleanup is to add the concept of auto-deleting monitor objects. Specifically when the "CHR_EVENT_CLOSED" event is emitted, the equivalent of "object-del" + "chardev-del" can be run internally. Since the transient client has already droppped its monitor connection, there is no synchronization to be concerned about.

If object-del or chardev-del fail, there's no way to report the error. Can they fail?

object-del can fail if

* An object with the specified "id" does not exist. That shouldn't happen in this case but harmless if it odes.
* object_del command tries to delete the monitor that is servicing the object_del command. Cannot happen with auto-delete
* the monitor has not finished initializing it BH with chardev handlers. Cannot happen if we know we have a live connection already.

chardev-del can fail if

* The chardev with "id" does not exist. SHouldn't happen but is harmless if it does
* The chardev reports it is "busy" - aka the frontend is still connected - we just deleted it so cannot happen
* Record/replay is in use - a niche use case

So I don't think errors are a problem.

Quoted 7 lines
Quoted 4 lines
Quoted 1 line

Do we always want to delete both monitor and character device?

IMHO yes they are a pair whose lifetime should be tied together for normal use.

Would we make monitor auto-delete delete its character device if character devices also had an auto-delete feature?

Quoted 7 lines

That's a possibility.

We would have to accept that it allows users to create a scenario in which auto-delete will frequenty fail.

ie configuring the chardev for auto-delete, but not having the frontend auto-delete, which would trigger the "busy" check.

This feels like it has some overlap with 'yank'? Where you can force a channel to get out of the way even if you don't delete it at that point.

Dave

Quoted 9 lines

Not convinced that's nicer ?

With regards,
Daniel

Signature · 3 lines
|: https://berrange.com       ~~        https://hachyderm.io/@berrange :|
|: https://libvirt.org          ~~          https://entangle-photo.org :|
|: https://pixelfed.art/berrange   ~~    https://fstop138.berrange.com :|
Signature · 4 lines
 -----Open up your eyes, open up your mind, open up your code -------   
/ Dr. David Alan Gilbert    |       Running GNU/Linux       | Happy  \ 
\        dave @ treblig.org |                               | In Hex /
 \ _________________________|_____ http://www.treblig.org   |_______/
#14 lore raw

Re: [PATCH v5 05/35] monitor: minimal conversion of monitors to QOM

In reply to Markus Armbruster
Quoted 131 lines · Markus Armbruster
Quoted 105 lines · Daniel P. Berrangé
Quoted 96 lines · Markus Armbruster
Quoted 66 lines · Daniel P. Berrangé

This introduces a Monitor QOM object, with MonitorHMP and MonitorQMP subclasses. This is the bare minimum conversion of just the type declarations and replacing g_new/g_free with object_new/object_unref.

---
 include/monitor/monitor.h  | 11 ++++++++++-
 monitor/hmp.c              | 29 +++++++++++++++++++++++++++--
 monitor/monitor-internal.h | 18 ++++++++++++++++--
 monitor/monitor.c          | 18 ++++++++++++++++--
 monitor/qmp-cmds.c         | 15 ++++++++-------
 monitor/qmp.c              | 29 +++++++++++++++++++++++++++--
 6 files changed, 104 insertions(+), 16 deletions(-)
include/monitor/monitor.h +10 -1
index b9642b58ba..2e9f9e12e9 100644
@@ -5,8 +5,17 @@
 #include "qapi/qapi-types-misc.h"
 #include "qemu/readline.h"
 #include "exec/hwaddr.h"
+#include "qom/object.h"
+
+#define TYPE_MONITOR "monitor"
+OBJECT_DECLARE_TYPE(Monitor, MonitorClass, MONITOR);
+
+#define TYPE_MONITOR_HMP "monitor-hmp"
+OBJECT_DECLARE_TYPE(MonitorHMP, MonitorHMPClass, MONITOR_HMP);
+
+#define TYPE_MONITOR_QMP "monitor-qmp"
+OBJECT_DECLARE_TYPE(MonitorQMP, MonitorQMPClass, MONITOR_QMP);
 
-typedef struct MonitorHMP MonitorHMP;
 typedef struct MonitorOptions MonitorOptions;
 
 #define QMP_REQ_QUEUE_LEN_MAX 8
monitor/hmp.c +14 -0
index 4e4468424a..81047d2513 100644
@@ -43,6 +43,20 @@
 #include "system/block-backend.h"
 #include "trace.h"
 
+OBJECT_DEFINE_TYPE(MonitorHMP, monitor_hmp, MONITOR_HMP, MONITOR);
+
+static void monitor_hmp_finalize(Object *obj)
+{
+}
+
+static void monitor_hmp_class_init(ObjectClass *cls, const void *data)
+{
+}
+
+static void monitor_hmp_init(Object *obj)
+{
+}
+
 static void monitor_command_cb(void *opaque, const char *cmdline,
                                void *readline_opaque)
 {

monitor/hmp.c static void monitor_readline_flush(void *opaque) · line 1545

Quoted 7 lines
monitor/hmp.c +3 -1
@@ -1526,10 +1540,21 @@ static void monitor_readline_flush(void *opaque)
 
 void monitor_new_hmp(Chardev *chr, bool use_readline, Error **errp)
 {
-    MonitorHMP *mon = g_new0(MonitorHMP, 1);
+    MonitorHMP *mon;
+    static int counter;
+    g_autofree char *id = g_strdup_printf("hmpcompat%d", counter++);

Hmm. The system picking IDs is problematic when they can clash with the user's IDs. If we had an ounce of common sense, we'd restrict both across the board so they cannot clash. But we don't.

We need an ID here, because we need to make the new object the child of something (actually: child of /objects/), which requires a child name.

Non-problem with -object / object-add, because @id is mandatory there.

Non-problem with -device / device_add, because we use separate parents for devices with and without @id (/machine/peripheral/ and /machine/peripheral-anon/, plus the /machine/unattached/ orphanage).

Example for an existing problem:

    $ qemu-system-x86_64 -nodefaults -monitor stdio -chardev null,id=chr0 -mon id=compat_monitor0,chardev=chr0
    qemu-system-x86_64: -mon id=compat_monitor0,chardev=chr0: Duplicate ID 'compat_monitor0' for mon

Example for a problem created by this series:

Quoted 7 lines
    $  qemu-system-x86_64 -nodefaults -display none -S -monitor stdio -chardev null,id=chr0 -object monitor-hmp,id=hmpcompat0,chardev=chr0
    qemu-system-x86_64: -monitor stdio: attempt to add duplicate property 'hmpcompat0' to object (type 'container')

I readily admit that these clashes are *unlikely*. Still, do we really want to define an interface that claims to let you pick any ID, then rejects some of them sometimes? Feels rather 1990s to me. At the very least, cover the wart in the commit message.

Quoted 5 lines

The way I looked at it was aything using -object with the new monitor-qmp/monitor-hmp types is new code. They can:

1. Trivially abide the warning about "hmpcompatNN" / "qmpcompatNN" being internal usage for compat syntax

A wart you can easily cope with is still a wart. Oh well, we'll live. However, the wart should be documented.

Quoted 2 lines

2. Not use both -monitor and -object on the same QEMU instance so not have a clash between the two to begin with

Management applications aware of the monitor QOM types should use -object / object-add exclusively.

Human users will use -monitor to save typing. If they then need to plug an additional one at run time, they have no choice but use object-add. Again, we'll live.

Quoted 4 lines

The remaining danger where is some existing code using -object with a *non-monitor* type, and calling it "hmpcompatNN" / "qmpcompatNN", which would be insanity. Never say never, but I think that's an acceptable risk.

Quoted 4 lines

The risk is acceptable. The documentation complexity that comes with the wart bothers me.

Could we use id_generate()?

Since we want to re-use the QemuOpts ID present in -monitor/-qmp, I'm going to stick with "compat_monitorNNN" and get rid of use of 'hmpcompatNN' and 'qmpcompatNN'. id_generate() does not help here because it forces a "#" prefix onto the ID which then (intentionally) violates id_wellformed() which object_new_xxx() validates.

With regards,
Daniel

Signature · 3 lines
|: https://berrange.com       ~~        https://hachyderm.io/@berrange :|
|: https://libvirt.org          ~~          https://entangle-photo.org :|
|: https://pixelfed.art/berrange   ~~    https://fstop138.berrange.com :|
#52 lore raw

Re: [PATCH v5 32/35] monitor: add support for auto-deleting monitors upon close

In reply to Markus Armbruster
Quoted 49 lines · Markus Armbruster
Quoted 6 lines · [email protected] writes:
Quoted 6 lines

The default monitor is usually a long lived object that will exist for the entire lifetime of the VM. A monitor can only service a single client at a time though, and so it might be desirable to hotplug additional monitors at runtime for specific tasks. If doing that, however, there is a need to remove the monitor when it is no longer needed.

Whatever adds the additional monitor can also delete it. The fact that you propose other means suggests you believe this would be cumbersome in practice. Why?

Quoted 7 lines
Quoted 7 lines

Allowing a client to run "object-del" against its own monitor adds complex edge cases, as it would be desirable to send the QMP response despite the monitor sending it being deleted. Doing "object-del" alone will also result in orphaning a character device backend instance, as there is no opportunity to run the companion "chardev-del" command.

Quoted 5 lines
Quoted 5 lines

A simpler way to ensure cleanup is to add the concept of auto-deleting monitor objects. Specifically when the "CHR_EVENT_CLOSED" event is emitted, the equivalent of "object-del" + "chardev-del" can be run internally. Since the transient client has already droppped its monitor connection, there is no synchronization to be concerned about.

If object-del or chardev-del fail, there's no way to report the error. Can they fail?

Do we always want to delete both monitor and character device?

Quoted 8 lines
Quoted 8 lines

This is implemented via a new "close-action=none|delete" property on the 'monitor-qmp' object. This concept could be extended with further actions in future, for example:

* close-action=shutdown - graceful guest shutdown
* close-action=terminate - immediate guest poweroff
* close-action=stop - pause guest CPUs while the monitor is not connected to any client

I'm not sure these would be useful and safe. Mentioning them here is okay regardless.

Quoted 6 lines
Quoted 6 lines

This is left as an exercise for future interested contributors.

Message-ID: <[email protected]>

snip

Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
monitor/monitor-internal.h +0 -0
index 5522e05464b..23829f32f9a 100644

snip

Quoted 49 lines
Quoted 11 lines
Quoted 11 lines
monitor/monitor-internal.h +11 -0
+static void monitor_qmp_self_delete_bh(void *opaque)
+{
+    MonitorQMP *mon = opaque;
+    g_autofree char *mon_id = object_property_get_child_name(
+        object_get_objects_root(), OBJECT(mon));
+    g_autofree char *chardev_id = g_strdup(mon->parent_obj.chardev_id);
+    Error *local_error = NULL;
+
+    if (!mon_id) {
+        /* Another monitor raced & ran 'object-del' on 'mon'
+         * before this BH got scheduled, so we have a ref on
Quoted 8 lines

monitor/monitor-internal.h

Quoted 8 lines
monitor/monitor-internal.h +8 -0
+         * mon but it is already unparented.
+         */
+        object_unref(mon);
+        return;
+    }
+
+    user_creatable_del(mon_id, &local_error);
+    object_unref(mon);

This object_unref() and the one above pair with the object_ref() in monitor_qmp_event() case MONITOR_QMP_CLOSE_ACTION_DELETE. Correct?

Is this worth a comment?

Quoted 12 lines
Quoted 12 lines
monitor/monitor-internal.h +7 -0
+    if (local_error != NULL) {
+        error_report_err(local_error);
+    } else {
+        qmp_chardev_remove(chardev_id, NULL);
+    }
+}
+
 static void monitor_qmp_event(void *opaque, QEMUChrEvent event)
 {
     QDict *data;
     MonitorQMP *mon = opaque;
Quoted 7 lines

monitor/monitor-internal.h

Quoted 7 lines
monitor/monitor-internal.h +7 -0
+    /* Protect against race if a client drops & quickly
+     * reconnects - we'll have the delete BH scheduled
+     * so must not honour a new open request
+     */
+    if (mon->delete_pending) {
+        return;
+    }

Consider "something happens between setting @delete_pending and monitor destruction that triggers the event" and "same thing happens after destruction". Any observable differences in behavior?

Quoted 2 lines

Are there any other monitor interactions that might need similar special-casing when @delete_pending?

When the BH deletes the monitor object, it will disconnect the chardev handlers. So there is no possibility of monitor_qmp_event being called after the monitor QOM destruction is complete. The only race is between this method and the start of the BH.

With regards,
Daniel

Signature · 3 lines
|: https://berrange.com       ~~        https://hachyderm.io/@berrange :|
|: https://libvirt.org          ~~          https://entangle-photo.org :|
|: https://pixelfed.art/berrange   ~~    https://fstop138.berrange.com :|
#53 lore raw

Re: [PATCH v5 32/35] monitor: add support for auto-deleting monitors upon close

Quoted 1 line · Daniel P. Berrangé

On Wed, Jul 01, 2026 at 08:32:06AM +0200, Markus Armbruster wrote:

[...]

Quoted 17 lines
Quoted 17 lines
Quoted 5 lines
Quoted 5 lines

static void monitor_qmp_event(void *opaque, QEMUChrEvent event)

 {
     QDict *data;
     MonitorQMP *mon = opaque;
Quoted 7 lines

monitor/qmp.c static QDict *qmp_greeting(MonitorQMP *mon) · line 610

Quoted 7 lines
monitor/qmp.c +7 -0
+    /* Protect against race if a client drops & quickly
+     * reconnects - we'll have the delete BH scheduled
+     * so must not honour a new open request
+     */
+    if (mon->delete_pending) {
+        return;
+    }

Consider "something happens between setting @delete_pending and monitor destruction that triggers the event" and "same thing happens after destruction". Any observable differences in behavior?

Quoted 7 lines
Quoted 2 lines

Are there any other monitor interactions that might need similar special-casing when @delete_pending?

When the BH deletes the monitor object, it will disconnect the chardev handlers. So there is no possibility of monitor_qmp_event being called after the monitor QOM destruction is complete. The only race is between this method and the start of the BH.

Add this completeness argument to the commit message?

#54 lore raw

Re: [PATCH v5 32/35] monitor: add support for auto-deleting monitors upon close

Quoted 26 lines · Markus Armbruster
Quoted 1 line · Daniel P. Berrangé

On Wed, Jul 01, 2026 at 08:32:06AM +0200, Markus Armbruster wrote:

[...]

Quoted 20 lines
Quoted 19 lines
Quoted 5 lines
Quoted 5 lines

static void monitor_qmp_event(void *opaque, QEMUChrEvent event)

 {
     QDict *data;
     MonitorQMP *mon = opaque;
Quoted 7 lines

monitor/qmp.c static QDict *qmp_greeting(MonitorQMP *mon) · line 610

Quoted 7 lines
monitor/qmp.c +7 -0
+    /* Protect against race if a client drops & quickly
+     * reconnects - we'll have the delete BH scheduled
+     * so must not honour a new open request
+     */
+    if (mon->delete_pending) {
+        return;
+    }

Consider "something happens between setting @delete_pending and monitor destruction that triggers the event" and "same thing happens after destruction". Any observable differences in behavior?

Are there any other monitor interactions that might need similar special-casing when @delete_pending?

Quoted 6 lines
Quoted 4 lines

When the BH deletes the monitor object, it will disconnect the chardev handlers. So there is no possibility of monitor_qmp_event being called after the monitor QOM destruction is complete. The only race is between this method and the start of the BH.

Add this completeness argument to the commit message?

Ok

With regards,
Daniel

Signature · 3 lines
|: https://berrange.com       ~~        https://hachyderm.io/@berrange :|
|: https://libvirt.org          ~~          https://entangle-photo.org :|
|: https://pixelfed.art/berrange   ~~    https://fstop138.berrange.com :|
#15 lore raw

Re: [PATCH v5 05/35] monitor: minimal conversion of monitors to QOM

In reply to Daniel P. Berrangé
Quoted 1 line · Daniel P. Berrangé

On Wed, Jul 01, 2026 at 03:19:53PM +0200, Markus Armbruster wrote:

[...]

Quoted 8 lines
Quoted 1 line

Could we use id_generate()?

Since we want to re-use the QemuOpts ID present in -monitor/-qmp, I'm going to stick with "compat_monitorNNN" and get rid of use of 'hmpcompatNN' and 'qmpcompatNN'. id_generate() does not help here because it forces a "#" prefix onto the ID which then (intentionally) violates id_wellformed() which object_new_xxx() validates.

A clean split between IDs provided by the user (must be id_wellformed()) and IDs conjured up by the system (must come from id_generate()) would be nice. But getting there would take some yak shaving. I'm not asking you to try. I'm okay with the solution in your v6.