monitor_qmp_dispatcher_co() runs in the iohandler AioContext that is not
polled during nested event loops. The coroutine currently reschedules
itself in the main loop's qemu_aio_context AioContext, which is polled
during nested event loops. One known problem is that QMP device-add
calls drain_call_rcu(), which temporarily drops the BQL, leading to all
sorts of havoc like other vCPU threads re-entering device emulation code
while another vCPU thread is waiting in device emulation code with
aio_poll().
Paolo Bonzini suggested running non-coroutine QMP handlers in the
iohandler AioContext. This avoids trouble with nested event loops. His
original idea was to move coroutine rescheduling to
monitor_qmp_dispatch(), but I resorted to moving it to qmp_dispatch()
because we don't know if the QMP handler needs to run in coroutine
context in monitor_qmp_dispatch(). monitor_qmp_dispatch() would have
been nicer since it's associated with the monitor implementation and not
as general as qmp_dispatch(), which is also used by qemu-ga.
A number of qemu-iotests need updated .out files because the order of
QMP events vs QMP responses has changed.
Solves Issue #1933.
Cc: qemu-stable@nongnu.org
Fixes: 7bed89958b ("device_core: use drain_call_rcu in in qmp_device_add")
Buglink: https://bugzilla.redhat.com/show_bug.cgi?id=2215192
Buglink: https://bugzilla.redhat.com/show_bug.cgi?id=2214985
Buglink: https://issues.redhat.com/browse/RHEL-17369
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-ID: <20240118144823.1497953-4-stefanha@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Tested-by: Fiona Ebner <f.ebner@proxmox.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
qatomic_mb_read and qatomic_mb_set were the very first atomic primitives
introduced for QEMU; their semantics are unclear and they provide a false
sense of safety.
The last use of qatomic_mb_read() has been removed, so delete it.
qatomic_mb_set() instead can survive as an optimized
qatomic_set()+smp_mb(), similar to Linux's smp_store_mb(), but
rename it to qatomic_set_mb() to match the order of the two
operations.
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Instead of relying on magic memory barriers, document the pattern that
is being used. It is the one based on Dekker's algorithm, and in this
case it is embodied as follows:
enqueue request; sleeping = true;
smp_mb(); smp_mb();
if (sleeping) kick(); if (!have a request) yield();
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Use a continue statement so that "after going to sleep" is treated the same
way as "after processing a request". Pull the monitor_lock critical
section out of monitor_qmp_requests_pop_any_with_lock() and protect
qmp_dispatcher_co_shutdown with the monitor_lock.
The two changes are complex to separate because monitor_qmp_dispatcher_co()
previously had a complicated logic to check for shutdown both before
and after going to sleep.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Instead of overloading qmp_dispatcher_co_busy, make the coroutine
pointer NULL. This will make things break spectacularly if somebody
tries to start a request after monitor_cleanup().
AIO_WAIT_WHILE_UNLOCKED() does not need qatomic_mb_read(), because
the macro contains all the necessary memory barriers.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
The QMP dispatcher coroutine holds the qmp_queue_lock over a yield
point, where it expects to be rescheduled from the main context. If a
CHR_EVENT_CLOSED event is received just then, it can race and block the
main thread on the mutex in monitor_qmp_cleanup_queue_and_resume.
monitor_resume does not need to be called from main context, so we can
call it immediately after popping a request from the queue, which allows
us to drop the qmp_queue_lock mutex before yielding.
Suggested-by: Wolfgang Bumiller <w.bumiller@proxmox.com>
Signed-off-by: Stefan Reiter <s.reiter@proxmox.com>
Message-Id: <20210322154024.15011-1-s.reiter@proxmox.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Cc: qemu-stable@nongnu.org
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Removed various qemu_mutex_lock and their respective qemu_mutex_unlock
calls and used lock guard macros (QEMU_LOCK_GUARD and
WITH_QEMU_LOCK_GUARD). This simplifies the code by
eliminating qemu_mutex_unlock calls.
Signed-off-by: Mahmoud Mandour <ma.mandourr@gmail.com>
Message-Id: <20210311031538.5325-6-ma.mandourr@gmail.com>
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Before this patch, monitor_qmp_dispatcher_co() used to check whether
shutdown is requested only when it would have to wait for new requests.
If there were still some queued requests, it would try to execute all of
them before shutting down.
This can be surprising when the queued QMP commands take long or hang
because Ctrl-C may not actually exit QEMU as soon as possible.
Change monitor_qmp_dispatcher_co() so that it additionally checks
whether shutdown is request before it gets a new request from the queue.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20210212172028.288825-3-kwolf@redhat.com>
Tested-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
monitor_qmp_dispatcher_co() needs to resume the monitor if
handle_qmp_command() suspended it. Two cases:
1. OOB enabled: suspended if mon->qmp_requests has no more space
2. OOB disabled: suspended always
We resume only after we processed the request. Which can take a long
time.
Resume the monitor right when the queue has space to keep the monitor
available for out-of-band commands even in this corner case.
Leave the "OOB disabled" case alone.
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Message-Id: <20210201161504.1976989-4-armbru@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
[Trailing whitespace tidied up]
Add tracepoints for in-band request enqueue and dequeue, processing of
queued in-band errors, and responses.
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Message-Id: <20210201161504.1976989-3-armbru@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Commit 9ce44e2ce2 "qmp: Move dispatcher to a coroutine" replaced
monitor_qmp_bh_dispatcher() by monitor_qmp_dispatcher_co(), but
neglected to update comments. Do that now.
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Message-Id: <20210201161504.1976989-2-armbru@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
qobject_to_json() and qobject_to_json_pretty() build a GString, then
covert it to QString. Just one of the callers actually needs a
QString: qemu_rbd_parse_filename(). A few others need a string they
can modify: qmp_send_response(), qga's send_response(), to_json_str(),
and qmp_fd_vsend_fds(). The remainder just need a string.
Change qobject_to_json() and qobject_to_json_pretty() to return the
GString.
qemu_rbd_parse_filename() now has to convert to QString. All others
save a QString temporary. to_json_str() actually becomes a bit
simpler, because GString provides more convenient modification
functions.
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Message-Id: <20201211171152.146877-6-armbru@redhat.com>
This moves the QMP dispatcher to a coroutine and runs all QMP command
handlers that declare 'coroutine': true in coroutine context so they
can avoid blocking the main loop while doing I/O or waiting for other
events.
For commands that are not declared safe to run in a coroutine, the
dispatcher drops out of coroutine context by calling the QMP command
handler from a bottom half.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Message-Id: <20201005155855.256490-10-kwolf@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
The correct way to set the current monitor for a coroutine handler will
be different than for a blocking handler, so monitor_set_cur() needs to
be called in qmp_dispatch().
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20201005155855.256490-7-kwolf@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
monitor_qmp_dispatch() is never supposed to be called in the context of
another monitor, so assert that monitor_cur() is NULL instead of saving
and restoring it.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20201005155855.256490-6-kwolf@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
cur_mon really needs to be coroutine-local as soon as we move monitor
command handlers to coroutines and let them yield. As a first step, just
remove all direct accesses to cur_mon so that we can implement this in
the getter function later.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20201005155855.256490-4-kwolf@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
For QMP commands without arguments, gen_marshal() laboriously
generates a qmp_marshal_FOO() that copes with null @args. Turns
there's just one caller that passes null instead of an empty QDict.
Adjust that caller, and simplify gen_marshal().
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Message-Id: <20200424084338.26803-15-armbru@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Trying to attach a QMP monitor to a chardev that is already in use
results in a crash because monitor_init_qmp() passes &error_abort to
qemu_chr_fe_init():
$ ./x86_64-softmmu/qemu-system-x86_64 --chardev stdio,id=foo --mon foo,mode=control --mon foo,mode=control
Unexpected error in qemu_chr_fe_init() at chardev/char-fe.c:220:
qemu-system-x86_64: --mon foo,mode=control: Device 'foo' is in use
Abgebrochen (Speicherabzug geschrieben)
Fix this by allowing monitor_init_qmp() to return an error and passing
any error in qemu_chr_fe_init() to its caller instead of aborting.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20200224143008.13362-18-kwolf@redhat.com>
Acked-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
misc.json contains definitions that are related to the system emulator,
so it can't be used for other tools like the storage daemon. This patch
moves basic functionality that is shared between all tools (and mostly
related to the monitor itself) into a new control.json, which could be
used in tools as well.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Message-Id: <20200129102239.31435-3-kwolf@redhat.com>
[Commit message tweaked]
Signed-off-by: Markus Armbruster <armbru@redhat.com>
The Chardev events are listed in the QEMUChrEvent enum.
By using the enum in the IOEventHandler typedef we:
- make the IOEventHandler type more explicit (this handler
process out-of-band information, while the IOReadHandler
is in-band),
- help static code analyzers.
This patch was produced with the following spatch script:
@match@
expression backend, opaque, context, set_open;
identifier fd_can_read, fd_read, fd_event, be_change;
@@
qemu_chr_fe_set_handlers(backend, fd_can_read, fd_read, fd_event,
be_change, opaque, context, set_open);
@depends on match@
identifier opaque, event;
identifier match.fd_event;
@@
static
-void fd_event(void *opaque, int event)
+void fd_event(void *opaque, QEMUChrEvent event)
{
...
}
Then the typedef was modified manually in
include/chardev/char-fe.h.
Signed-off-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Acked-by: Corey Minyard <cminyard@mvista.com>
Acked-by: Cornelia Huck <cohuck@redhat.com>
Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Message-Id: <20191218172009.8868-15-philmd@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
The Chardev events are listed in the QEMUChrEvent enum. To be
able to use this enum in the IOEventHandler typedef, we need to
explicit all the events ignored by this frontend, to silent the
following GCC warning:
CC monitor/qmp.o
monitor/qmp.c: In function ‘monitor_qmp_event’:
monitor/qmp.c:345:5: error: enumeration value ‘CHR_EVENT_BREAK’ not handled in switch [-Werror=switch]
345 | switch (event) {
| ^~~~~~
monitor/qmp.c:345:5: error: enumeration value ‘CHR_EVENT_MUX_IN’ not handled in switch [-Werror=switch]
monitor/qmp.c:345:5: error: enumeration value ‘CHR_EVENT_MUX_OUT’ not handled in switch [-Werror=switch]
cc1: all warnings being treated as errors
Signed-off-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Message-Id: <20191218172009.8868-12-philmd@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
When a monitor's queue is filled up in handle_qmp_command()
it gets suspended. It's the dispatcher bh's job currently to
resume the monitor, which it does after processing an event
from the queue. However, it is possible for a
CHR_EVENT_CLOSED event to be processed before before the bh
is scheduled, which will clear the queue without resuming
the monitor, thereby preventing the dispatcher from reaching
the resume() call.
Any new connections to the qmp socket will be accept()ed and
show the greeting, but will not respond to any messages sent
afterwards (as they will not be read from the
still-suspended socket).
Fix this by resuming the monitor when clearing a queue which
was filled up.
Signed-off-by: Wolfgang Bumiller <w.bumiller@proxmox.com>
Message-Id: <20191115085914.21287-1-w.bumiller@proxmox.com>
Commit 4eaca8de26 dropped monitor_qmp_respond()'s parameter @id
without updating its function comment. Fix that.
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Message-Id: <20190816193305.12090-1-armbru@redhat.com>
Most callers know which monitor type they want to have. Instead of
calling monitor_init() with flags that can describe both types of
monitors, make monitor_init_{hmp,qmp}() public interfaces that take
specific bools instead of flags and call these functions directly.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20190613153405.24769-15-kwolf@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Monitor.flags contains three different flags: One to distinguish HMP
from QMP; one specific to HMP (MONITOR_USE_READLINE) that is ignored
with QMP; and another one specific to QMP (MONITOR_USE_PRETTY) that is
ignored with HMP.
Split the flags field into three bools and move them to the right
subclass. Flags are still in use for the monitor_init() interface.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20190613153405.24769-14-kwolf@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Move QMP infrastructure from monitor/misc.c to monitor/qmp.c. This is
code that can be shared for all targets, so compile it only once.
The amount of function and particularly extern variables in
monitor_int.h is probably a bit larger than it needs to be, but this way
no non-trivial code modifications are needed. The interfaces between QMP
and the monitor core can be cleaned up later.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Message-Id: <20190613153405.24769-11-kwolf@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
[monitor_is_qmp() tidied up to make checkpatch.pl happy,
superfluous #include dropped]
Signed-off-by: Markus Armbruster <armbru@redhat.com>
