Commit v5.2.0-190-g0546c0609c ("vl: split various early command line
options to a separate function") moved the trace backend init code to
the qemu_process_early_options(). Which is now being called before
os_daemonize() via qemu_maybe_daemonize().
Turns out that this change of order causes a problem when executing
QEMU in daemon mode and with CONFIG_TRACE_SIMPLE. The trace thread
is now being created by the parent, and the parent is left waiting for
a trace file flush that was registered via st_init(). The result is
that the parent process never exits.
To reproduce, fire up a QEMU process with -daemonize and with
CONFIG_TRACE_SIMPLE enabled. Two QEMU process will be left in the
host:
$ sudo ./x86_64-softmmu/qemu-system-x86_64 -S -no-user-config -nodefaults \
-nographic -machine none,accel=kvm:tcg -daemonize
$ ps axf | grep qemu
529710 pts/3 S+ 0:00 | \_ grep --color=auto qemu
529697 ? Ssl 0:00 \_ ./x86_64-softmmu/qemu-system-x86_64 -S -no-user-config -nodefaults -nographic -machine none,accel=kvm:tcg -daemonize
529699 ? Sl 0:00 \_ ./x86_64-softmmu/qemu-system-x86_64 -S -no-user-config -nodefaults -nographic -machine none,accel=kvm:tcg -daemonize
The parent thread is hang in flush_trace_file:
$ sudo gdb ./x86_64-softmmu/qemu-system-x86_64 529697
(..)
(gdb) bt
#0 0x00007f9dac6a137d in syscall () at /lib64/libc.so.6
#1 0x00007f9dacc3c4f3 in g_cond_wait () at /lib64/libglib-2.0.so.0
#2 0x0000555d12f952da in flush_trace_file (wait=true) at ../trace/simple.c:140
#3 0x0000555d12f95b4c in st_flush_trace_buffer () at ../trace/simple.c:383
#4 0x00007f9dac5e43a7 in __run_exit_handlers () at /lib64/libc.so.6
#5 0x00007f9dac5e4550 in on_exit () at /lib64/libc.so.6
#6 0x0000555d12d454de in os_daemonize () at ../os-posix.c:255
#7 0x0000555d12d0bd5c in qemu_maybe_daemonize (pid_file=0x0) at ../softmmu/vl.c:2408
#8 0x0000555d12d0e566 in qemu_init (argc=8, argv=0x7fffc594d9b8, envp=0x7fffc594da00) at ../softmmu/vl.c:3459
#9 0x0000555d128edac1 in main (argc=8, argv=0x7fffc594d9b8, envp=0x7fffc594da00) at ../softmmu/main.c:49
(gdb)
Aside from the 'zombie' process in the host, this is directly impacting
Libvirt. Libvirt waits for the parent process to exit to be sure that the
QMP monitor is available in the daemonized process to fetch QEMU
capabilities, and as is now Libvirt hangs at daemon start waiting
for the parent thread to exit.
The fix is simple: just move the trace backend related code back to
be executed after daemonizing.
Fixes: 0546c0609c
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Message-Id: <20210105181437.538366-2-danielhb413@gmail.com>
Acked-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
The functionality of -writeconfig is limited and the code
does not even try to detect cases where it prints incorrect
syntax (for example if values have a quote in them, since
qemu_config_parse does not support any kind of escaping)
so remove it.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
v2
Dropped vmstate: Fix memory leak in vmstate_handle_alloc
Broke on Power
Added migration: only check page size match if RAM postcopy is enabled
-----BEGIN PGP SIGNATURE-----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=4tbe
-----END PGP SIGNATURE-----
Merge remote-tracking branch 'remotes/dgilbert/tags/pull-migration-20210208a' into staging
Migration pull 2021-02-08
v2
Dropped vmstate: Fix memory leak in vmstate_handle_alloc
Broke on Power
Added migration: only check page size match if RAM postcopy is enabled
# gpg: Signature made Mon 08 Feb 2021 11:28:14 GMT
# gpg: using RSA key 45F5C71B4A0CB7FB977A9FA90516331EBC5BFDE7
# gpg: Good signature from "Dr. David Alan Gilbert (RH2) <dgilbert@redhat.com>" [full]
# Primary key fingerprint: 45F5 C71B 4A0C B7FB 977A 9FA9 0516 331E BC5B FDE7
* remotes/dgilbert/tags/pull-migration-20210208a: (27 commits)
migration: only check page size match if RAM postcopy is enabled
migration: introduce snapshot-{save, load, delete} QMP commands
iotests: fix loading of common.config from tests/ subdir
iotests: add support for capturing and matching QMP events
migration: introduce a delete_snapshot wrapper
migration: wire up support for snapshot device selection
migration: control whether snapshots are ovewritten
block: rename and alter bdrv_all_find_snapshot semantics
block: allow specifying name of block device for vmstate storage
block: add ability to specify list of blockdevs during snapshot
migration: stop returning errno from load_snapshot()
migration: Make save_snapshot() return bool, not 0/-1
block: push error reporting into bdrv_all_*_snapshot functions
migration: Display the migration blockers
migration: Add blocker information
migration: Fix a few absurdly defective error messages
migration: Fix cache_init()'s "Failed to allocate" error messages
migration: Clean up signed vs. unsigned XBZRLE cache-size
migration: Fix migrate-set-parameters argument validation
migration: introduce 'userfaultfd-wrlat.py' script
...
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Modify load_snapshot/save_snapshot to accept the device list and vmstate
node name parameters previously added to the block layer.
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
Message-Id: <20210204124834.774401-9-berrange@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
None of the callers care about the errno value since there is a full
Error object populated. This gives consistency with save_snapshot()
which already just returns a boolean value.
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
[PMD: Return false/true instead of -1/0, document function]
Acked-by: Pavel Dovgalyuk <pavel.dovgalyuk@ispras.ru>
Signed-off-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Message-Id: <20210204124834.774401-4-berrange@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
The platform specific details of mechanisms for implementing
confidential guest support may require setup at various points during
initialization. Thus, it's not really feasible to have a single cgs
initialization hook, but instead each mechanism needs its own
initialization calls in arch or machine specific code.
However, to make it harder to have a bug where a mechanism isn't
properly initialized under some circumstances, we want to have a
common place, late in boot, where we verify that cgs has been
initialized if it was requested.
This patch introduces a ready flag to the ConfidentialGuestSupport
base type to accomplish this, which we verify in
qemu_machine_creation_done().
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Reviewed-by: Greg Kurz <groug@kaod.org>
Global properties have an @optional field, which allows to apply a given
property to a given type even if one of its subclasses doesn't support
it. This is especially used in the compat code when dealing with the
"disable-modern" and "disable-legacy" properties and the "virtio-pci"
type.
Allow object_register_sugar_prop() to set this field as well.
Signed-off-by: Greg Kurz <groug@kaod.org>
Message-Id: <159738953558.377274.16617742952571083440.stgit@bahia.lan>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Eduardo Habkost <ehabkost@redhat.com>
Reviewed-by: Cornelia Huck <cohuck@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
This will allow us to centralize the registration of
the cpus.c module accelerator operations (in accel/accel-softmmu.c),
and trigger it automatically using object hierarchy lookup from the
new accel_init_interfaces() initialization step, depending just on
which accelerators are available in the code.
Rename all tcg-cpus.c, kvm-cpus.c, etc to tcg-accel-ops.c,
kvm-accel-ops.c, etc, matching the object type names.
Signed-off-by: Claudio Fontana <cfontana@suse.de>
Message-Id: <20210204163931.7358-18-cfontana@suse.de>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
The -msg timestamp=on|off option controls whether a timestamp is printed
with error_report() messages. The "-msg" name suggests that this option
has a wider effect than just error_report(). The next patch extends it
to the 'log' trace backend, so rename the variable from
error_with_timestamp to message_with_timestamp.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Tested-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Tested-by: BALATON Zoltan <balaton@eik.bme.hu>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Message-id: 20210125113507.224287-2-stefanha@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Use qemu_opts_parse_noisily now that HMP does not call
vnc_parse anymore.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Gerd Hoffmann <kraxel@redhat.com>
Message-Id: <20210120144235.345983-4-pbonzini@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
The possible choices for panic, reset and watchdog actions are inconsistent.
"-action panic=poweroff" should be renamed to "-action panic=shutdown"
on the command line. This is because "-action panic=poweroff" and
"-action watchdog=poweroff" have slightly different semantics, the first
does an unorderly exit while the second goes through qemu_cleanup(). With
this change, -no-shutdown would not have to change "-action panic=pause"
"pause", just like it does not have to change the reset action.
"-action reboot=none" should be renamed to "-action reboot=reset".
This should be self explanatory, since for example "-action panic=none"
lets the guest proceed without taking any action.
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Due to the renumbering of text consoles when graphical consoles are
created, init_displaystate must be called after all QemuConsoles are
created, i.e. after devices are created.
vl.c calls it from qemu_init_displays, while qmp_x_exit_preconfig is
where devices are created. If qemu_init_displays is called before it,
the VGA graphical console does not come up.
Reported-by: Howard Spoelstra <hsp.cat7@gmail.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Silly patch extracted from the next one, which is already big enough.
Because there are already local variables named "accel", we will name
the global vl.c variable for "-M accel" accelerators instead. Rename
it already in configure_accelerators to be ready.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
It has been marked as deprecated since QEMU v5.0, replaced by the
corresponding parameter of the -display option.
Signed-off-by: Thomas Huth <thuth@redhat.com>
Message-Id: <20201210155808.233895-5-thuth@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
It has been marked as deprecated since QEMU v4.2, replaced by
the -overcommit option. Time to remove it now.
Signed-off-by: Thomas Huth <thuth@redhat.com>
Message-Id: <20201210155808.233895-4-thuth@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
The '-tb-size' option (replaced by '-accel tcg,tb-size') is
deprecated since 5.0 (commit fe17413247). Remove it.
Signed-off-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Message-Id: <20201202112714.1223783-1-philmd@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
Signed-off-by: Thomas Huth <thuth@redhat.com>
Message-Id: <20201210155808.233895-2-thuth@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
The current default action of pausing a guest after a panic event
is received leaves the responsibility to resume guest execution to the
management layer. The reasons for this behavior are discussed here:
https://lore.kernel.org/qemu-devel/52148F88.5000509@redhat.com/
However, in instances like the case of older guests (Linux and
Windows) using a pvpanic device but missing support for the
PVPANIC_CRASHLOADED event, and Windows guests using the hv-crash
enlightenment, it is desirable to allow the guests to continue
running after sending a PVPANIC_PANICKED event. This allows such
guests to proceed to capture a crash dump and automatically reboot
without intervention of a management layer.
Add an option to avoid stopping a VM after a panic event is received,
by passing:
-action panic=none
in the command line arguments, or during runtime by using an upcoming
QMP command.
Suggested-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Alejandro Jimenez <alejandro.j.jimenez@oracle.com>
Message-Id: <1607705564-26264-3-git-send-email-alejandro.j.jimenez@oracle.com>
[Do not fix panic action in the variable, instead modify -no-shutdown. - Paolo]
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Several command line options currently in use are meant to modify
the behavior of QEMU in response to certain guest events like:
-no-reboot, -no-shutdown, -watchdog-action.
These can be grouped into a single option of the form:
-action event=action
Which can be used to specify the existing options above in the
following format:
-action reboot=none|shutdown
-action shutdown=poweroff|pause
-action watchdog=reset|shutdown|poweroff|pause|debug|none|inject-nmi
This is done in preparation for adding yet another option of this
type, which modifies the QEMU behavior when a guest panic occurs.
Suggested-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Alejandro Jimenez <alejandro.j.jimenez@oracle.com>
Message-Id: <1607705564-26264-2-git-send-email-alejandro.j.jimenez@oracle.com>
[Use QemuOpts help support, invoke QMP command. - Paolo]
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Add a QMP command to allow for the behaviors specified by the
-no-reboot and -no-shutdown command line option to be set at runtime.
The new command is named set-action and takes optional arguments, named
after an event, that provide a corresponding action to take.
Example:
-> { "execute": "set-action",
"arguments": {
"reboot": "none",
"shutdown": "poweroff",
"watchdog": "debug" } }
<- { "return": {} }
Suggested-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Alejandro Jimenez <alejandro.j.jimenez@oracle.com>
Message-Id: <1607705564-26264-4-git-send-email-alejandro.j.jimenez@oracle.com>
[Split the series differently, with -action based on the QMP command. - Paolo]
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Compute the DIRTY_MEMORY_CODE bit in memory_region_get_dirty_log_mask
instead of memory_region_init_*. This makes it possible to allocate
memory backend objects at any time.
Reviewed-by: Igor Mammedov <imammedo@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
qemu_finish_machine_init currently can only exit QEMU if it fails.
Prepare for giving it proper error propagation, and possibly for
adding a plugin_add monitor command that calls an accelerator
method.
While at it, make all errors from plugin_load look the same.
Reviewed-by: Igor Mammedov <imammedo@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Machine options can be retrieved as properties of the machine object.
Encourage that by removing the "easy" accessor to machine options.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Generalize the qdev_hotplug variable to the different phases of
machine initialization. We would like to allow different
monitor commands depending on the phase.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
machine_init_done is not the right flag to check when preconfig
is taken into account; for example "./qemu-system-x86_64 -serial
mon:stdio -preconfig" does not print the QEMU monitor header until after
exit_preconfig. Add back a custom bool for mux character devices. This
partially undoes commit c7278b4355 ("chardev: introduce chr_machine_done
hook", 2018-03-12), but it keeps the cleaner logic using a function
pointer in ChardevClass.
Reviewed-by: Igor Mammedov <imammedo@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
qdev_machine_creation_done is only setting a flag now. Extend it to
move more code out of vl.c. Leave only consistency checks and gdbserver
processing in qemu_machine_creation_done.
gdbserver_start can be moved after qdev_machine_creation_done because
it only does listen on the socket and creates some internal data
structures; it does not send any data (e.g. guest state) over the socket.
Reviewed-by: Igor Mammedov <imammedo@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Now that there is no RUN_STATE_PRECONFIG anymore that can conflict with
RUN_STATE_INMIGRATE, we can allow -incoming defer with -preconfig.
Reviewed-by: Igor Mammedov <imammedo@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Move post-preconfig initialization to the x-exit-preconfig. If preconfig
is not requested, just exit preconfig mode immediately with the QMP
command.
As a result, the preconfig loop will run with accel_setup_post
and os_setup_post restrictions (xen_restrict, chroot, etc.)
already done.
Reviewed-by: Igor Mammedov <imammedo@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
The preconfig state is only used if -incoming is not specified, which
makes the RunState state machine more tricky than it need be. However
there is already an equivalent condition which works even with -incoming,
namely qdev_hotplug. Use it instead of a separate runstate.
Reviewed-by: Igor Mammedov <imammedo@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Move more of them into MachineState, in preparation for moving initialization
of the machine out of vl.c.
Reviewed-by: Igor Mammedov <imammedo@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
serial_hd(i) is NULL if and only if i >= serial_max_hds(). Test
serial_hd(i) instead of bounding the loop at serial_max_hds(),
thus removing one more function that vl.c is expected to export.
Reviewed-by: Igor Mammedov <imammedo@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Complement the previous patch by starting the VM with a QMP command.
The plan is that the user will be able to do the same, invoking two
commands "finish-machine-init" and "cont" instead of
"x-exit-preconfig".
Reviewed-by: Igor Mammedov <imammedo@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Make qemu_start_incoming_migration local to migration/migration.c.
By using the runstate instead of a separate flag, vl need not do
anything to setup deferred incoming migration.
qmp_migrate_incoming also does not need the deferred_incoming flag
anymore, because "-incoming PROTOCOL" will clear the "once" flag
before the main loop starts. Therefore, later invocations of
the migrate-incoming command will fail with the existing
"The incoming migration has already been started" error message.
Reviewed-by: Igor Mammedov <imammedo@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
The check has the same effect here, it only matters that it is performed
once all devices, both builtin and user-specified, have been created.
Reviewed-by: Igor Mammedov <imammedo@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Displays should be available before the monitor starts, so that
it is possible to use the graphical console to interact with
the monitor itself.
This patch is quite ugly, but all this is temporary. The double
call to qemu_init_displays will go away as soon we can unify machine
initialization between the preconfig and "normal" flows, and so will
the preconfig_exit_requested variable (that is only preconfig_requested
remains).
Reviewed-by: Igor Mammedov <imammedo@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
This is a bit nasty: the machine is storing a string and later
resolving it. We probably want to remove the memdev property
and instead make this a memory-set command. "-M memdev" can be
handled a legacy option that is special cased by
machine_set_property.
Reviewed-by: Igor Mammedov <imammedo@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
"Early" backends are created before the machine and can be used as
machine options.
Reviewed-by: Igor Mammedov <imammedo@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Move CHECKPOINT_INIT right before the machine initialization is
completed. Everything before is essentially an extension of
command line parsing.
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: Igor Mammedov <imammedo@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
There is no need to load plugins in the middle of default device processing,
move -plugin handling just before preconfig is entered.
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: Igor Mammedov <imammedo@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Create it together with other default backends, even though the processing is
done later.
Reviewed-by: Igor Mammedov <imammedo@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
