| Message ID | 1699900440-207345-3-git-send-email-steven.sistare@oracle.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | fix migration of suspended runstate | expand |
Steve Sistare <steven.sistare@oracle.com> writes: > A vm in the suspended state is not completely stopped. Is this a statement of a fact about VMs in the suspended state in general or is this describing what this patch is trying to fix? > The VCPUs have been paused, but the cpu clock still runs, and runstate > notifiers for the transition to stopped have not been called. ...it reads like the latter, but then why aren't we fixing this at the moment we put the VM in the suspend state? > Modify vm_stop_force_state to completely stop the vm if the current > state is suspended, to be called for live migration and snapshots. Hm, this changes the meaning of the "force" from: "force a state even if already stopped" into: "force a complete stop if already suspended, otherwise just set the state" I don't know what to make of this, shouldn't all vm_stops cause a complete stop? We need to at least resolve the overloading of the 'force' term. > > Suggested-by: Peter Xu <peterx@redhat.com> > Signed-off-by: Steve Sistare <steven.sistare@oracle.com> > --- > system/cpus.c | 8 ++++++-- > 1 file changed, 6 insertions(+), 2 deletions(-) > > diff --git a/system/cpus.c b/system/cpus.c > index f72c4be..c772708 100644 > --- a/system/cpus.c > +++ b/system/cpus.c > @@ -255,6 +255,8 @@ void cpu_interrupt(CPUState *cpu, int mask) > static int do_vm_stop(RunState state, bool send_stop, bool force) > { > int ret = 0; > + bool running = runstate_is_running(); > + bool suspended = runstate_check(RUN_STATE_SUSPENDED); > > if (qemu_in_vcpu_thread()) { > qemu_system_vmstop_request_prepare(); > @@ -267,10 +269,12 @@ static int do_vm_stop(RunState state, bool send_stop, bool force) > return 0; > } > > - if (runstate_is_running()) { > + if (running || (suspended && force)) { > runstate_set(state); > cpu_disable_ticks(); > - pause_all_vcpus(); > + if (running) { > + pause_all_vcpus(); > + } > vm_state_notify(0, state); > if (send_stop) { > qapi_event_send_stop();
On 11/20/2023 9:15 AM, Fabiano Rosas wrote: > Steve Sistare <steven.sistare@oracle.com> writes: > >> A vm in the suspended state is not completely stopped. > > Is this a statement of a fact about VMs in the suspended state in > general or is this describing what this patch is trying to fix? The former. >> The VCPUs have been paused, but the cpu clock still runs, and runstate >> notifiers for the transition to stopped have not been called. > > ...it reads like the latter, but then why aren't we fixing this at the > moment we put the VM in the suspend state? cpu_get_ticks() must continue to tick while the guest is suspended, so that QEMU_CLOCK_VIRTUAL continues to tick, so that timeouts based on that clock will fire. One example is timed wake from suspend, acpi_pm_tmr_timer. >> Modify vm_stop_force_state to completely stop the vm if the current >> state is suspended, to be called for live migration and snapshots. > > Hm, this changes the meaning of the "force" from: > > "force a state even if already stopped" > > into: > > "force a complete stop if already suspended, otherwise just set the > state" vm_stop_force_state has the same behavior as before for all states except suspended. If suspended, it also: - stops cpu ticks - calls runstate stopped handlers - sets a new runstate > I don't know what to make of this, shouldn't all vm_stops cause a > complete stop? We cannot stop cpu_get_ticks. We could maybe call the runstate stop handlers, but that requires a careful examination of every handler, and there is no obvious correctness or cleanliness reason to stop them immediately on vm_stop(), since cpu ticks still needs special handling later. > We need to at least resolve the overloading of the 'force' term. How about a more complete function header comment: /* * If the machine is running or suspended, completely stop it. * Force the new runstate to @state. * The current state is forgotten forever. */ - Steve >> Suggested-by: Peter Xu <peterx@redhat.com> >> Signed-off-by: Steve Sistare <steven.sistare@oracle.com> >> --- >> system/cpus.c | 8 ++++++-- >> 1 file changed, 6 insertions(+), 2 deletions(-) >> >> diff --git a/system/cpus.c b/system/cpus.c >> index f72c4be..c772708 100644 >> --- a/system/cpus.c >> +++ b/system/cpus.c >> @@ -255,6 +255,8 @@ void cpu_interrupt(CPUState *cpu, int mask) >> static int do_vm_stop(RunState state, bool send_stop, bool force) >> { >> int ret = 0; >> + bool running = runstate_is_running(); >> + bool suspended = runstate_check(RUN_STATE_SUSPENDED); >> >> if (qemu_in_vcpu_thread()) { >> qemu_system_vmstop_request_prepare(); >> @@ -267,10 +269,12 @@ static int do_vm_stop(RunState state, bool send_stop, bool force) >> return 0; >> } >> >> - if (runstate_is_running()) { >> + if (running || (suspended && force)) { >> runstate_set(state); >> cpu_disable_ticks(); >> - pause_all_vcpus(); >> + if (running) { >> + pause_all_vcpus(); >> + } >> vm_state_notify(0, state); >> if (send_stop) { >> qapi_event_send_stop();
On Mon, Nov 13, 2023 at 10:33:50AM -0800, Steve Sistare wrote: > A vm in the suspended state is not completely stopped. The VCPUs have been > paused, but the cpu clock still runs, and runstate notifiers for the > transition to stopped have not been called. Modify vm_stop_force_state to > completely stop the vm if the current state is suspended, to be called for > live migration and snapshots. > > Suggested-by: Peter Xu <peterx@redhat.com> > Signed-off-by: Steve Sistare <steven.sistare@oracle.com> > --- > system/cpus.c | 8 ++++++-- > 1 file changed, 6 insertions(+), 2 deletions(-) > > diff --git a/system/cpus.c b/system/cpus.c > index f72c4be..c772708 100644 > --- a/system/cpus.c > +++ b/system/cpus.c > @@ -255,6 +255,8 @@ void cpu_interrupt(CPUState *cpu, int mask) > static int do_vm_stop(RunState state, bool send_stop, bool force) > { > int ret = 0; > + bool running = runstate_is_running(); > + bool suspended = runstate_check(RUN_STATE_SUSPENDED); > > if (qemu_in_vcpu_thread()) { > qemu_system_vmstop_request_prepare(); > @@ -267,10 +269,12 @@ static int do_vm_stop(RunState state, bool send_stop, bool force) > return 0; > } > > - if (runstate_is_running()) { > + if (running || (suspended && force)) { > runstate_set(state); > cpu_disable_ticks(); Not directly relevant, but this is weird that I just notice. If we disable ticks before stopping vCPUs, IIUC it means vcpus can see stall ticks. I checked the vm_start() and indeed that one did it in the other way round: we'll stop vCPUs before stopping the ticks. > - pause_all_vcpus(); > + if (running) { > + pause_all_vcpus(); > + } > vm_state_notify(0, state); > if (send_stop) { > qapi_event_send_stop(); IIUC the "force" is not actually needed. It's only used when SUSPENDED, right? In general, considering all above, I'm wondering something like this would be much cleaner (and dropping force)? ===8<=== static int do_vm_stop(RunState state, bool send_stop) { + bool suspended = runstate_check(RUN_STATE_SUSPENDED); + bool running = runstate_is_running(); int ret = 0; - if (runstate_is_running()) { + /* + * RUNNING: VM and vCPUs are all running + * SUSPENDED: VM is running, VCPUs are stopped + * Others: VM and vCPUs are all stopped + */ + + /* Whether do we need to stop vCPUs? */ + if (running) { + pause_all_vcpus(); + } + + /* Whether do we need to stop the VM in general? */ + if (running || suspended) { runstate_set(state); cpu_disable_ticks(); - pause_all_vcpus(); vm_state_notify(0, state); if (send_stop) { qapi_event_send_stop();
Peter Xu <peterx@redhat.com> writes: > On Mon, Nov 13, 2023 at 10:33:50AM -0800, Steve Sistare wrote: >> A vm in the suspended state is not completely stopped. The VCPUs have been >> paused, but the cpu clock still runs, and runstate notifiers for the >> transition to stopped have not been called. Modify vm_stop_force_state to >> completely stop the vm if the current state is suspended, to be called for >> live migration and snapshots. >> >> Suggested-by: Peter Xu <peterx@redhat.com> >> Signed-off-by: Steve Sistare <steven.sistare@oracle.com> >> --- >> system/cpus.c | 8 ++++++-- >> 1 file changed, 6 insertions(+), 2 deletions(-) >> >> diff --git a/system/cpus.c b/system/cpus.c >> index f72c4be..c772708 100644 >> --- a/system/cpus.c >> +++ b/system/cpus.c >> @@ -255,6 +255,8 @@ void cpu_interrupt(CPUState *cpu, int mask) >> static int do_vm_stop(RunState state, bool send_stop, bool force) >> { >> int ret = 0; >> + bool running = runstate_is_running(); >> + bool suspended = runstate_check(RUN_STATE_SUSPENDED); >> >> if (qemu_in_vcpu_thread()) { >> qemu_system_vmstop_request_prepare(); >> @@ -267,10 +269,12 @@ static int do_vm_stop(RunState state, bool send_stop, bool force) >> return 0; >> } >> >> - if (runstate_is_running()) { >> + if (running || (suspended && force)) { >> runstate_set(state); >> cpu_disable_ticks(); > > Not directly relevant, but this is weird that I just notice. > > If we disable ticks before stopping vCPUs, IIUC it means vcpus can see > stall ticks. I checked the vm_start() and indeed that one did it in the > other way round: we'll stop vCPUs before stopping the ticks. > >> - pause_all_vcpus(); >> + if (running) { >> + pause_all_vcpus(); >> + } >> vm_state_notify(0, state); >> if (send_stop) { >> qapi_event_send_stop(); > > IIUC the "force" is not actually needed. It's only used when SUSPENDED, > right? That's the overloading I'm complaining about. We're using "force" to say both: "include suspended" and: "set the state". This is basically taking knowledge from the callsite being the migration code and encoding it in that flag. I'd prefer something like: static int do_vm_stop(RunState state, bool send_stop, bool set_state, bool include_suspended);
On 11/20/2023 2:59 PM, Peter Xu wrote: > On Mon, Nov 13, 2023 at 10:33:50AM -0800, Steve Sistare wrote: >> A vm in the suspended state is not completely stopped. The VCPUs have been >> paused, but the cpu clock still runs, and runstate notifiers for the >> transition to stopped have not been called. Modify vm_stop_force_state to >> completely stop the vm if the current state is suspended, to be called for >> live migration and snapshots. >> >> Suggested-by: Peter Xu <peterx@redhat.com> >> Signed-off-by: Steve Sistare <steven.sistare@oracle.com> >> --- >> system/cpus.c | 8 ++++++-- >> 1 file changed, 6 insertions(+), 2 deletions(-) >> >> diff --git a/system/cpus.c b/system/cpus.c >> index f72c4be..c772708 100644 >> --- a/system/cpus.c >> +++ b/system/cpus.c >> @@ -255,6 +255,8 @@ void cpu_interrupt(CPUState *cpu, int mask) >> static int do_vm_stop(RunState state, bool send_stop, bool force) >> { >> int ret = 0; >> + bool running = runstate_is_running(); >> + bool suspended = runstate_check(RUN_STATE_SUSPENDED); >> >> if (qemu_in_vcpu_thread()) { >> qemu_system_vmstop_request_prepare(); >> @@ -267,10 +269,12 @@ static int do_vm_stop(RunState state, bool send_stop, bool force) >> return 0; >> } >> >> - if (runstate_is_running()) { >> + if (running || (suspended && force)) { >> runstate_set(state); >> cpu_disable_ticks(); > > Not directly relevant, but this is weird that I just notice. > > If we disable ticks before stopping vCPUs, IIUC it means vcpus can see > stall ticks. I checked the vm_start() and indeed that one did it in the > other way round: we'll stop vCPUs before stopping the ticks. > >> - pause_all_vcpus(); >> + if (running) { >> + pause_all_vcpus(); >> + } >> vm_state_notify(0, state); >> if (send_stop) { >> qapi_event_send_stop(); > > IIUC the "force" is not actually needed. It's only used when SUSPENDED, > right? > > In general, considering all above, I'm wondering something like this would > be much cleaner (and dropping force)? If we drop force, then all calls to vm_stop will completely stop the suspended state, eg an hmp "stop" command. This causes two problems. First, that is a change in user-visible behavior for something that currently works, vs the migration code where we are fixing brokenness. Second, it does not quite work, because the state becomes RUN_STATE_PAUSED, so the suspended state is forgotten, and the hmp "cont" will try to set the running state. I could fix that by introducing a new state RUN_STATE_SUSPENDED_STOPPED, but again it is a user-visible change in existing behavior. (I even implemented that while developing, then I realized it was not needed to fix the migration bugs.) - Steve > ===8<=== > static int do_vm_stop(RunState state, bool send_stop) > { > + bool suspended = runstate_check(RUN_STATE_SUSPENDED); > + bool running = runstate_is_running(); > int ret = 0; > > - if (runstate_is_running()) { > + /* > + * RUNNING: VM and vCPUs are all running > + * SUSPENDED: VM is running, VCPUs are stopped > + * Others: VM and vCPUs are all stopped > + */ > + > + /* Whether do we need to stop vCPUs? */ > + if (running) { > + pause_all_vcpus(); > + } > + > + /* Whether do we need to stop the VM in general? */ > + if (running || suspended) { > runstate_set(state); > cpu_disable_ticks(); > - pause_all_vcpus(); > vm_state_notify(0, state); > if (send_stop) { > qapi_event_send_stop(); >
On 11/20/2023 3:47 PM, Fabiano Rosas wrote: > Peter Xu <peterx@redhat.com> writes: >> On Mon, Nov 13, 2023 at 10:33:50AM -0800, Steve Sistare wrote: >>> A vm in the suspended state is not completely stopped. The VCPUs have been >>> paused, but the cpu clock still runs, and runstate notifiers for the >>> transition to stopped have not been called. Modify vm_stop_force_state to >>> completely stop the vm if the current state is suspended, to be called for >>> live migration and snapshots. >>> >>> Suggested-by: Peter Xu <peterx@redhat.com> >>> Signed-off-by: Steve Sistare <steven.sistare@oracle.com> >>> --- >>> system/cpus.c | 8 ++++++-- >>> 1 file changed, 6 insertions(+), 2 deletions(-) >>> >>> diff --git a/system/cpus.c b/system/cpus.c >>> index f72c4be..c772708 100644 >>> --- a/system/cpus.c >>> +++ b/system/cpus.c >>> @@ -255,6 +255,8 @@ void cpu_interrupt(CPUState *cpu, int mask) >>> static int do_vm_stop(RunState state, bool send_stop, bool force) >>> { >>> int ret = 0; >>> + bool running = runstate_is_running(); >>> + bool suspended = runstate_check(RUN_STATE_SUSPENDED); >>> >>> if (qemu_in_vcpu_thread()) { >>> qemu_system_vmstop_request_prepare(); >>> @@ -267,10 +269,12 @@ static int do_vm_stop(RunState state, bool send_stop, bool force) >>> return 0; >>> } >>> >>> - if (runstate_is_running()) { >>> + if (running || (suspended && force)) { >>> runstate_set(state); >>> cpu_disable_ticks(); >> >> Not directly relevant, but this is weird that I just notice. >> >> If we disable ticks before stopping vCPUs, IIUC it means vcpus can see >> stall ticks. I checked the vm_start() and indeed that one did it in the >> other way round: we'll stop vCPUs before stopping the ticks. >> >>> - pause_all_vcpus(); >>> + if (running) { >>> + pause_all_vcpus(); >>> + } >>> vm_state_notify(0, state); >>> if (send_stop) { >>> qapi_event_send_stop(); >> >> IIUC the "force" is not actually needed. It's only used when SUSPENDED, >> right? When not suspended, the force flag causes a stopped state to be forced even if current is a different stopped state. > That's the overloading I'm complaining about. We're using "force" to say > both: "include suspended" and: "set the state". This is basically taking > knowledge from the callsite being the migration code and encoding it in > that flag. > > I'd prefer something like: > > static int do_vm_stop(RunState state, bool send_stop, bool set_state, > bool include_suspended); This function has always been tailored for use by migration code and no other callers. Migration would always pass set_state=true and include_suspended=true. We have no use case for other combinations and no test for them. To my mind, "force" naturally implies both behaviors. We force the machine into the specified stop state, completely stopping suspended execution. Perhaps renaming vm_stop_force_state would erase the old association of "force" with only forcing runstate, such as vm_stop_all(). - Steve
On Mon, Nov 20, 2023 at 03:55:54PM -0500, Steven Sistare wrote: > If we drop force, then all calls to vm_stop will completely stop the > suspended state, eg an hmp "stop" command. This causes two problems. > First, that is a change in user-visible behavior for something that > currently works, IMHO it depends on what should be the correct behavior. IOW, when VM is in SUSPENDED state and then the user sends "stop" QMP command, what should we expect? My understanding is we should expect to fully stop the VM, including the ticks, for example. Keeping the ticks running even after QMP "stop" doesn't sound right, isn't it? > vs the migration code where we are fixing brokenness. This is not a migration-only bug if above holds, IMO. > Second, it does not quite work, because the state becomes > RUN_STATE_PAUSED, so the suspended state is forgotten, and the hmp "cont" > will try to set the running state. I could fix that by introducing a new > state RUN_STATE_SUSPENDED_STOPPED, but again it is a user-visible change > in existing behavior. (I even implemented that while developing, then I > realized it was not needed to fix the migration bugs.) Good point. Now with above comments, what's your thoughts on whether we should change the user behavior? My answer is still a yes. Maybe SUSPENDED should not be a RunState at all? SUSPENDED is guest visible behavior, while something like QMP "stop" is not guest visible. Maybe we should remember it separately? It means qemu_system_suspend() could remember that in a separate field (as part of guest state), then when wakeup we should conditionally go back with/without vcpus running depending on the new "suspended" state.
On 11/20/2023 4:44 PM, Peter Xu wrote: > On Mon, Nov 20, 2023 at 03:55:54PM -0500, Steven Sistare wrote: >> If we drop force, then all calls to vm_stop will completely stop the >> suspended state, eg an hmp "stop" command. This causes two problems. >> First, that is a change in user-visible behavior for something that >> currently works, > > IMHO it depends on what should be the correct behavior. IOW, when VM is in > SUSPENDED state and then the user sends "stop" QMP command, what should we > expect? > > My understanding is we should expect to fully stop the VM, including the > ticks, for example. Keeping the ticks running even after QMP "stop" > doesn't sound right, isn't it? I agree, but others may not, and this decision would require approval from maintainers in other areas, including upper layers such as libvirt that are aware of the suspended state. It is a user-visible change, and may require a new libvirt release. >> vs the migration code where we are fixing brokenness. > > This is not a migration-only bug if above holds, IMO. > >> Second, it does not quite work, because the state becomes >> RUN_STATE_PAUSED, so the suspended state is forgotten, and the hmp "cont" >> will try to set the running state. I could fix that by introducing a new >> state RUN_STATE_SUSPENDED_STOPPED, but again it is a user-visible change >> in existing behavior. (I even implemented that while developing, then I >> realized it was not needed to fix the migration bugs.) > > Good point. > > Now with above comments, what's your thoughts on whether we should change > the user behavior? My answer is still a yes. > > Maybe SUSPENDED should not be a RunState at all? SUSPENDED is guest visible > behavior, while something like QMP "stop" is not guest visible. Maybe we > should remember it separately? > > It means qemu_system_suspend() could remember that in a separate field (as > part of guest state), then when wakeup we should conditionally go back > with/without vcpus running depending on the new "suspended" state. Regardless of how we remember it, the status command must still expose the suspended state to the user. The user must be able to see that a guest is suspended, and decide when to issue a wakeup command. If we change the stop command to completely stop a suspended vm, then we must allow the user to query whether a vm is suspended-running or suspended-stopped, because the command they must issue to resume is different: wakeup for the former, and cont for the latter. This change is visible to libvirt. Adding it will delay this entire patch series, and is not necessary for fixing the migration bug. There is no downside to drawing the line here for this series, and possibly changing stop semantics in the future. - Steve
On Tue, Nov 21, 2023 at 04:21:18PM -0500, Steven Sistare wrote: > On 11/20/2023 4:44 PM, Peter Xu wrote: > > On Mon, Nov 20, 2023 at 03:55:54PM -0500, Steven Sistare wrote: > >> If we drop force, then all calls to vm_stop will completely stop the > >> suspended state, eg an hmp "stop" command. This causes two problems. > >> First, that is a change in user-visible behavior for something that > >> currently works, > > > > IMHO it depends on what should be the correct behavior. IOW, when VM is in > > SUSPENDED state and then the user sends "stop" QMP command, what should we > > expect? > > > > My understanding is we should expect to fully stop the VM, including the > > ticks, for example. Keeping the ticks running even after QMP "stop" > > doesn't sound right, isn't it? > > I agree, but others may not, and this decision would require approval from > maintainers in other areas, including upper layers such as libvirt that are > aware of the suspended state. It is a user-visible change, and may require > a new libvirt release. $ ./scripts/get_maintainer.pl -f system/cpus.c Richard Henderson <richard.henderson@linaro.org> (maintainer:Overall TCG CPUs) Paolo Bonzini <pbonzini@redhat.com> (reviewer:Overall TCG CPUs) qemu-devel@nongnu.org (open list:All patches CC here) I'm also copying Richard, while Dan/Paolo is already in the loop, so we should have the "quorum" already. Let's see whether we can already get some comments from the maintainers.. > > >> vs the migration code where we are fixing brokenness. > > > > This is not a migration-only bug if above holds, IMO. > > > >> Second, it does not quite work, because the state becomes > >> RUN_STATE_PAUSED, so the suspended state is forgotten, and the hmp "cont" > >> will try to set the running state. I could fix that by introducing a new > >> state RUN_STATE_SUSPENDED_STOPPED, but again it is a user-visible change > >> in existing behavior. (I even implemented that while developing, then I > >> realized it was not needed to fix the migration bugs.) > > > > Good point. > > > > Now with above comments, what's your thoughts on whether we should change > > the user behavior? My answer is still a yes. > > > > Maybe SUSPENDED should not be a RunState at all? SUSPENDED is guest visible > > behavior, while something like QMP "stop" is not guest visible. Maybe we > > should remember it separately? > > > > It means qemu_system_suspend() could remember that in a separate field (as > > part of guest state), then when wakeup we should conditionally go back > > with/without vcpus running depending on the new "suspended" state. > > Regardless of how we remember it, the status command must still expose > the suspended state to the user. The user must be able to see that a > guest is suspended, and decide when to issue a wakeup command. Hmm, right, we may want to keep having the SUSPENDED state in RunState, even another separate "vm_suspended" boolean might still be required. > > If we change the stop command to completely stop a suspended vm, then we must > allow the user to query whether a vm is suspended-running or suspended-stopped, > because the command they must issue to resume is different: wakeup for the > former, and cont for the latter. If it's stopped, the user must need a "cont" anyway. And then if after "cont" the user still sees it's suspended, then would "system_wakeup" work here if necessary, after that "cont"? Let's consider the current QEMU with below sequence of operations: 1) vm running 2) guest triggers ACPI suspend -> vm suspended 3) admin triggers "stop" cmd -> vm suspended (ignored..) 4) admin triggers "cont" cmd -> vm suspended (ignored.. too) AFAICT both 2) and 3) are unwanted behavior, and after noticing 3) I feel stronger that this is not a migration issue alone. It also means after step 1)-3) if we got a wakeup elsewhere, the VM can actually be running! That's definitely unexpected after admin sends "stop" already. Isn't that another real bug? I'm slightly confused on why you said above that libvirt will need a new release. Could you elaborate? Especially on what scenario we need to maintain compatibility that still makes sense. > > This change is visible to libvirt. Adding it will delay this entire patch > series, and is not necessary for fixing the migration bug. There is no > downside to drawing the line here for this series, and possibly changing stop > semantics in the future. This series will need to wait for rc releases anyway until 8.2 all out: https://wiki.qemu.org/Planning/8.2 I think we still have time to even catch the earliest train right after 8.2 released, if we can reach a consensus soon in whatever form. Having a partial solution merged for migration is probably doable, but that will make the code even more complicated and harder to maintain. So before doing so, I'd at least like to understand better on what use case you were describing that will start to fall apart. Thanks,
On Mon, Nov 20, 2023 at 04:44:50PM -0500, Peter Xu wrote: > On Mon, Nov 20, 2023 at 03:55:54PM -0500, Steven Sistare wrote: > > If we drop force, then all calls to vm_stop will completely stop the > > suspended state, eg an hmp "stop" command. This causes two problems. > > First, that is a change in user-visible behavior for something that > > currently works, > > IMHO it depends on what should be the correct behavior. IOW, when VM is in > SUSPENDED state and then the user sends "stop" QMP command, what should we > expect? I would say that from a mgmtm app POV "stop" is initiating a state transition, from RUN_STATE_RUNNING to RUN_STATE_PAUSED and "cont" is doing the reverse from PAUSED to RUNNING. It is a little more complicated than that as there are some other states like INMIGRATE that are conceptually equiv to RUNNING, and states where the transition simply doesn't make sense. So my question is if we're in "SUSPENDED" and someone issues "stop", what state do we go into, and perhaps more importantly what state do we go to in a subsequent "cont". If you say SUSPENDED ---(stop)---> PAUSED ---(cont)---> SUSPENDED then we create a problem, because the decision for the transition out of PAUSED needs memory of the previous state. > My understanding is we should expect to fully stop the VM, including the > ticks, for example. Keeping the ticks running even after QMP "stop" > doesn't sound right, isn't it? The "stop" QMP command is documented as "Stop all guest VCPU execution" the devil is in the detail though, and we've not documented any detail. Whether or not timers keep running across stop/cont I think can be argued to be an impl detail, as long as the headline goal "vcpus don't execute" is satisfied. > > vs the migration code where we are fixing brokenness. > > This is not a migration-only bug if above holds, IMO. > > > Second, it does not quite work, because the state becomes > > RUN_STATE_PAUSED, so the suspended state is forgotten, and the hmp "cont" > > will try to set the running state. I could fix that by introducing a new > > state RUN_STATE_SUSPENDED_STOPPED, but again it is a user-visible change > > in existing behavior. (I even implemented that while developing, then I > > realized it was not needed to fix the migration bugs.) > > Good point. We have added new guest states periodically. It is a user visible change, but you could argue that it is implementing a new feature ie the ability to "stop" a "suspended" guest, and so is justified. S3 is so little used in virt, so I'm not surprised we're finding long standing edge cases that have never been thought about before. > Now with above comments, what's your thoughts on whether we should change > the user behavior? My answer is still a yes. > > Maybe SUSPENDED should not be a RunState at all? SUSPENDED is guest visible > behavior, while something like QMP "stop" is not guest visible. Maybe we > should remember it separately? Yes, every time I look at this area I come away thinking that the RunState enum is a mess, overloading too many different concepts onto the same single field. Specifically "SUSPENDED" vs "RUNNING" is a reflection of guest state (ie whether or not the VM is in S3), but pretty much all the others are a reflection of QEMU host state. I kind of feel that SUSPENDED (S3) probably shouldn't have been a RunState at all. I'd probably put guest-panicked into a separate thing too. But we're stuck with what we have. > It means qemu_system_suspend() could remember that in a separate field (as > part of guest state), then when wakeup we should conditionally go back > with/without vcpus running depending on the new "suspended" state. With regards, Daniel
On Wed, Nov 22, 2023 at 09:38:06AM +0000, Daniel P. Berrangé wrote: > On Mon, Nov 20, 2023 at 04:44:50PM -0500, Peter Xu wrote: > > On Mon, Nov 20, 2023 at 03:55:54PM -0500, Steven Sistare wrote: > > > If we drop force, then all calls to vm_stop will completely stop the > > > suspended state, eg an hmp "stop" command. This causes two problems. > > > First, that is a change in user-visible behavior for something that > > > currently works, > > > > IMHO it depends on what should be the correct behavior. IOW, when VM is in > > SUSPENDED state and then the user sends "stop" QMP command, what should we > > expect? > > I would say that from a mgmtm app POV "stop" is initiating a state > transition, from RUN_STATE_RUNNING to RUN_STATE_PAUSED and "cont" > is doing the reverse from PAUSED to RUNNING. > > It is a little more complicated than that as there are some other > states like INMIGRATE that are conceptually equiv to RUNNING, > and states where the transition simply doesn't make sense. In the qemu impl, INMIGRATE is, imho, more equiv of PAUSED - do_vm_stop() mostly ignores every state except RUNNING (putting bdrv operations aside). IOW, anything besides "running" is treated as "not running". But then Paolo fixed that in 1e9981465f ("qmp: handle stop/cont in INMIGRATE state"), wiring that to autostart. Now we seem to find that "suspended" should actually fall within (where "vm" is running, but "vcpu" is not), and it seems we should treat "vm" and "vcpu" differently. > > So my question is if we're in "SUSPENDED" and someone issues "stop", > what state do we go into, and perhaps more importantly what state > do we go to in a subsequent "cont". I think we must stop the "vm", not only the "vcpu". I discussed this bit in the other thread more or less: it's because qemu_system_wakeup_request() can be called in many places, e.g. acpi_pm_tmr_timer(). It means even after the VM is "stop"ped by the admin QMP (where qmp_stop() ignores SUSPENDED, keep the "vm" running), it can silently got waken up without admin even noticing it. I'm not sure what Libvirt will behave if it suddenly receives a QAPI_EVENT_WAKEUP randomly after a "stop". > > If you say SUSPENDED ---(stop)---> PAUSED ---(cont)---> SUSPENDED > then we create a problem, because the decision for the transition > out of PAUSED needs memory of the previous state. Right, that's why I think we at least need one more boolean to remember the suspended state, then when we switch from any STOP states into any RUN states, we know where to go. Here STOP states I meant anything except RUNNING and SUSPENDED, while RUN -> RUNNING or SUSPENDED. > > > My understanding is we should expect to fully stop the VM, including the > > ticks, for example. Keeping the ticks running even after QMP "stop" > > doesn't sound right, isn't it? > > The "stop" QMP command is documented as > > "Stop all guest VCPU execution" > > the devil is in the detail though, and we've not documented any detail. > > Whether or not timers keep running across stop/cont I think can be > argued to be an impl detail, as long as the headline goal "vcpus > don't execute" is satisfied. "stop" was since qemu v0.14, so I guess we can't blame the missing of details or any form of inaccuracy.. Obviously we do more than "stop vCPU executions" in the current implementation. But after we reach a consensus on how we should fix the current suspended problem, we may want to update the documentation to start containing more information. > > > > vs the migration code where we are fixing brokenness. > > > > This is not a migration-only bug if above holds, IMO. > > > > > Second, it does not quite work, because the state becomes > > > RUN_STATE_PAUSED, so the suspended state is forgotten, and the hmp "cont" > > > will try to set the running state. I could fix that by introducing a new > > > state RUN_STATE_SUSPENDED_STOPPED, but again it is a user-visible change > > > in existing behavior. (I even implemented that while developing, then I > > > realized it was not needed to fix the migration bugs.) > > > > Good point. > > We have added new guest states periodically. It is a user visible > change, but you could argue that it is implementing a new feature > ie the ability to "stop" a "suspended" guest, and so is justified. > > S3 is so little used in virt, so I'm not surprised we're finding > long standing edge cases that have never been thought about before. > > > Now with above comments, what's your thoughts on whether we should change > > the user behavior? My answer is still a yes. > > > > Maybe SUSPENDED should not be a RunState at all? SUSPENDED is guest visible > > behavior, while something like QMP "stop" is not guest visible. Maybe we > > should remember it separately? > > Yes, every time I look at this area I come away thinking that > the RunState enum is a mess, overloading too many different > concepts onto the same single field. > > Specifically "SUSPENDED" vs "RUNNING" is a reflection of guest > state (ie whether or not the VM is in S3), but pretty much all > the others are a reflection of QEMU host state. I kind of feel > that SUSPENDED (S3) probably shouldn't have been a RunState at > all. I'd probably put guest-panicked into a separate thing too. > > But we're stuck with what we have. IMO compatibility is only necessary if at least the existing code is running well. But now I see it a major flaw in suspended state and I can't see how it can go right if with current code base.. My concern is instead that after suspended will be used more (e.g., assuming CPR will rely on it) we can have more chance to confuse/oops a mgmt app like Libvirt, like I described above. In summary, I think a current solution to me is only to fix at least suspended state for good, by: - adding vm_suspended boolean to remember machine RUNNING / SUSPENDED state. When "cont" we need to switch to either RUNNING / SUSPENDED depending on the boolean. - keeping SUSPENDED state as RunState (for compatibility, otherwise we'll need another interface to fetch that boolean anyway), even though not query-able during any !RUNNING && !SUSPENDED state.. hopefully not a big deal - enrich ducumentation of qmp_stop/qmp_cont to describe what they really do - (with suspended working all right...) fix migration of SUSPENDED state I don't expect a lot of code changes is needed, maybe even less than the current series (because we don't need special knob to differenciate migration or non-migration callers of do_vm_stop()). IMHO this series is already doing some of that but just decided to ignore outside-migration states for suspeneded. We may want to add some test cases though to verify the suspended state transitions (maybe easier to put into migration-test with the new ACPI guest code), but optional. Thanks,
On 11/22/2023 11:21 AM, Peter Xu wrote: > On Wed, Nov 22, 2023 at 09:38:06AM +0000, Daniel P. Berrangé wrote: >> On Mon, Nov 20, 2023 at 04:44:50PM -0500, Peter Xu wrote: >>> On Mon, Nov 20, 2023 at 03:55:54PM -0500, Steven Sistare wrote: >>>> If we drop force, then all calls to vm_stop will completely stop the >>>> suspended state, eg an hmp "stop" command. This causes two problems. >>>> First, that is a change in user-visible behavior for something that >>>> currently works, >>> >>> IMHO it depends on what should be the correct behavior. IOW, when VM is in >>> SUSPENDED state and then the user sends "stop" QMP command, what should we >>> expect? >> >> I would say that from a mgmtm app POV "stop" is initiating a state >> transition, from RUN_STATE_RUNNING to RUN_STATE_PAUSED and "cont" >> is doing the reverse from PAUSED to RUNNING. >> >> It is a little more complicated than that as there are some other >> states like INMIGRATE that are conceptually equiv to RUNNING, >> and states where the transition simply doesn't make sense. > > In the qemu impl, INMIGRATE is, imho, more equiv of PAUSED - do_vm_stop() > mostly ignores every state except RUNNING (putting bdrv operations aside). > IOW, anything besides "running" is treated as "not running". > > But then Paolo fixed that in 1e9981465f ("qmp: handle stop/cont in > INMIGRATE state"), wiring that to autostart. > > Now we seem to find that "suspended" should actually fall within (where > "vm" is running, but "vcpu" is not), and it seems we should treat "vm" and > "vcpu" differently. > >> >> So my question is if we're in "SUSPENDED" and someone issues "stop", >> what state do we go into, and perhaps more importantly what state >> do we go to in a subsequent "cont". > > I think we must stop the "vm", not only the "vcpu". I discussed this bit > in the other thread more or less: it's because qemu_system_wakeup_request() > can be called in many places, e.g. acpi_pm_tmr_timer(). > > It means even after the VM is "stop"ped by the admin QMP (where qmp_stop() > ignores SUSPENDED, keep the "vm" running), it can silently got waken up > without admin even noticing it. I'm not sure what Libvirt will behave if > it suddenly receives a QAPI_EVENT_WAKEUP randomly after a "stop". > >> >> If you say SUSPENDED ---(stop)---> PAUSED ---(cont)---> SUSPENDED >> then we create a problem, because the decision for the transition >> out of PAUSED needs memory of the previous state. > > Right, that's why I think we at least need one more boolean to remember the > suspended state, then when we switch from any STOP states into any RUN > states, we know where to go. Here STOP states I meant anything except > RUNNING and SUSPENDED, while RUN -> RUNNING or SUSPENDED. > >> >>> My understanding is we should expect to fully stop the VM, including the >>> ticks, for example. Keeping the ticks running even after QMP "stop" >>> doesn't sound right, isn't it? >> >> The "stop" QMP command is documented as >> >> "Stop all guest VCPU execution" >> >> the devil is in the detail though, and we've not documented any detail. >> >> Whether or not timers keep running across stop/cont I think can be >> argued to be an impl detail, as long as the headline goal "vcpus >> don't execute" is satisfied. > > "stop" was since qemu v0.14, so I guess we can't blame the missing of > details or any form of inaccuracy.. Obviously we do more than "stop vCPU > executions" in the current implementation. > > But after we reach a consensus on how we should fix the current suspended > problem, we may want to update the documentation to start containing more > information. > >> >>>> vs the migration code where we are fixing brokenness. >>> >>> This is not a migration-only bug if above holds, IMO. >>> >>>> Second, it does not quite work, because the state becomes >>>> RUN_STATE_PAUSED, so the suspended state is forgotten, and the hmp "cont" >>>> will try to set the running state. I could fix that by introducing a new >>>> state RUN_STATE_SUSPENDED_STOPPED, but again it is a user-visible change >>>> in existing behavior. (I even implemented that while developing, then I >>>> realized it was not needed to fix the migration bugs.) >>> >>> Good point. >> >> We have added new guest states periodically. It is a user visible >> change, but you could argue that it is implementing a new feature >> ie the ability to "stop" a "suspended" guest, and so is justified. >> >> S3 is so little used in virt, so I'm not surprised we're finding >> long standing edge cases that have never been thought about before. >> >>> Now with above comments, what's your thoughts on whether we should change >>> the user behavior? My answer is still a yes. >>> >>> Maybe SUSPENDED should not be a RunState at all? SUSPENDED is guest visible >>> behavior, while something like QMP "stop" is not guest visible. Maybe we >>> should remember it separately? >> >> Yes, every time I look at this area I come away thinking that >> the RunState enum is a mess, overloading too many different >> concepts onto the same single field. >> >> Specifically "SUSPENDED" vs "RUNNING" is a reflection of guest >> state (ie whether or not the VM is in S3), but pretty much all >> the others are a reflection of QEMU host state. I kind of feel >> that SUSPENDED (S3) probably shouldn't have been a RunState at >> all. I'd probably put guest-panicked into a separate thing too. >> >> But we're stuck with what we have. > > IMO compatibility is only necessary if at least the existing code is > running well. But now I see it a major flaw in suspended state and I can't > see how it can go right if with current code base.. My concern is instead > that after suspended will be used more (e.g., assuming CPR will rely on it) > we can have more chance to confuse/oops a mgmt app like Libvirt, like I > described above. > > In summary, I think a current solution to me is only to fix at least > suspended state for good, by: > > - adding vm_suspended boolean to remember machine RUNNING / SUSPENDED > state. When "cont" we need to switch to either RUNNING / SUSPENDED > depending on the boolean. > > - keeping SUSPENDED state as RunState (for compatibility, otherwise we'll > need another interface to fetch that boolean anyway), even though not > query-able during any !RUNNING && !SUSPENDED state.. hopefully not a > big deal > > - enrich ducumentation of qmp_stop/qmp_cont to describe what they really do > > - (with suspended working all right...) fix migration of SUSPENDED state > > I don't expect a lot of code changes is needed, maybe even less than the > current series (because we don't need special knob to differenciate > migration or non-migration callers of do_vm_stop()). IMHO this series is > already doing some of that but just decided to ignore outside-migration > states for suspeneded. > > We may want to add some test cases though to verify the suspended state > transitions (maybe easier to put into migration-test with the new ACPI > guest code), but optional. FYI, here is a brief update before today's meeting. I have implemented this and I am testing libvirt and its various save + restore commands, when the guest is suspended running (RUN_STATE_SUSPENDED), and suspended stopped (RUN_STATE_PAUSED with vm_was_suspended = true). There are a few failures, and I am still investigating to see whether they can be fixed in qemu, or need a fix in libvirt. I will send more details later. - Steve
diff --git a/system/cpus.c b/system/cpus.c index f72c4be..c772708 100644 --- a/system/cpus.c +++ b/system/cpus.c @@ -255,6 +255,8 @@ void cpu_interrupt(CPUState *cpu, int mask) static int do_vm_stop(RunState state, bool send_stop, bool force) { int ret = 0; + bool running = runstate_is_running(); + bool suspended = runstate_check(RUN_STATE_SUSPENDED); if (qemu_in_vcpu_thread()) { qemu_system_vmstop_request_prepare(); @@ -267,10 +269,12 @@ static int do_vm_stop(RunState state, bool send_stop, bool force) return 0; } - if (runstate_is_running()) { + if (running || (suspended && force)) { runstate_set(state); cpu_disable_ticks(); - pause_all_vcpus(); + if (running) { + pause_all_vcpus(); + } vm_state_notify(0, state); if (send_stop) { qapi_event_send_stop();
A vm in the suspended state is not completely stopped. The VCPUs have been paused, but the cpu clock still runs, and runstate notifiers for the transition to stopped have not been called. Modify vm_stop_force_state to completely stop the vm if the current state is suspended, to be called for live migration and snapshots. Suggested-by: Peter Xu <peterx@redhat.com> Signed-off-by: Steve Sistare <steven.sistare@oracle.com> --- system/cpus.c | 8 ++++++-- 1 file changed, 6 insertions(+), 2 deletions(-)