Cc'ing a few more people. Daniel Henrique Barboza <danielhb@linux.ibm.com> writes:
When issuing the qmp/hmp 'system_wakeup' command, what happens in a nutshell is:
- qmp_system_wakeup_request set runstate to RUNNING, sets a wakeup_reason and notify the event - in the main_loop, all vcpus are paused, a system reset is issued, all subscribers of wakeup_notifiers receives a notification, vcpus are then resumed and the wake up QAPI event is fired
Note that this procedure alone doesn't ensure that the guest will awake from SUSPENDED state - the subscribers of the wake up event must take action to resume the guest, otherwise the guest will simply reboot.
At this moment there are only two subscribers of the wake up event: one in hw/acpi/core.c and another one in hw/i386/xen/xen-hvm.c. This means that system_wakeup does not work as intended with other architectures.
However, only the presence of 'system_wakeup' is required for QGA to support 'guest-suspend-ram' and 'guest-suspend-hybrid' at this moment. This means that the user/management will expect to suspend the guest using one of those suspend commands and then resume execution using system_wakeup, regardless of the support offered in system_wakeup in the first place.
This patch adds a new flag called 'wakeup-suspend-support' in TargetInfo that allows the caller to query if the guest supports wake up from suspend via system_wakeup. It goes over the subscribers of the wake up event and, if it's empty, it assumes that the guest does not support wake up from suspend (and thus, pm-suspend itself).
This is the expected output of query-target when running a x86 guest:
{"execute" : "query-target"} {"return": {"arch": "x86_64", "wakeup-suspend-support": true}}
This is the output when running a pseries guest:
{"execute" : "query-target"} {"return": {"arch": "ppc64", "wakeup-suspend-support": false}}
Given that the TargetInfo structure is read-only, adding a new flag to it is backwards compatible. There is no need to deprecate the old TargetInfo format.
With this extra tool, management can avoid situations where a guest that does not have proper suspend/wake capabilities ends up in inconsistent state (e.g. https://github.com/open-power-host-os/qemu/issues/31).
Reported-by: Balamuruhan S <bala24@linux.vnet.ibm.com> Signed-off-by: Daniel Henrique Barboza <danielhb@linux.ibm.com>
Is query-target is the right place to carry this flag? v7 is rather late for this kind of question; my sincere apologies. The flag is true after qemu_register_wakeup_notifier(). Callers so far: * piix4_pm_realize() via acpi_pm1_cnt_init() This is the realize method of device PIIX4_PM. It's an optional onboard device (suppressed by -no-acpi) of machine types pc-i440fx-VERSION, pc-VERSION, malta. * pc_q35_init() via ich9_lpc_pm_init(), ich9_pm_init(), acpi_pm1_cnt_init() This is the initialization method of machine types pc-q35-VERSION. Note that -no-acpi is not honored. * vt82c686b_pm_realize() via acpi_pm1_cnt_init() This is the realize method of device VT82C686B_PM. It's an onboard device of machine type fulong2e. Again, -no-acpi is not honored. * xen_hvm_init() This one gets called with -accel xen. I suspect the actual callback xen_wakeup_notifier() doesn't actually make wakeup work, unlike the acpi_notify_wakeup() callback registered by the other callers. Issue#1: this calls into question your assumption that the existence of a wake-up notifier implies wake-up works. It still holds if --accel xen is only accepted together with other configuration that triggers registration of acpi_notify_wakeup(). Is it? Stefano, Anthony? Issue#2: the flag isn't a property of the target. Due to -no-acpi, it's not even a property of the machine type. If it was, query-machines would be the natural owner of the flag. Perhaps query-machines is still the proper owner. The value of wakeup-suspend-support would have to depend on -no-acpi for the machine types that honor it. Not ideal; I'd prefer MachineInfo to be static. Tolerable? I guess that's also a libvirt question.