Re: [Qemu-devel] [PATCH V3] Guest stop notification

[Marcelo Tosatti <mtosatti@redhat.com>]

On Sat, Dec 03, 2011 at 12:25:37PM +0100, Jan Kiszka wrote:
> On 2011-12-03 12:19, Marcelo Tosatti wrote:
> > On Sat, Dec 03, 2011 at 10:06:56AM +0100, Jan Kiszka wrote:
> >> On 2011-12-02 22:27, Eric B Munson wrote:
> >>> On Fri, 02 Dec 2011, Jan Kiszka wrote:
> >>>
> >>>> On 2011-12-02 20:19, Eric B Munson wrote:
> >>>>> Often when a guest is stopped from the qemu console, it will report spurious
> >>>>> soft lockup warnings on resume.  There are kernel patches being discussed that
> >>>>> will give the host the ability to tell the guest that it is being stopped and
> >>>>> should ignore the soft lockup warning that generates.
> >>>>>
> >>>>> Signed-off-by: Eric B Munson <emunson@mgebm.net>
> >>>>> Cc: Avi Kivity <avi@redhat.com>
> >>>>> Cc: Marcelo Tosatti <mtosatti@redhat.com>
> >>>>> Cc: Jan Kiszka <jan.kiszka@siemens.com>
> >>>>> Cc: ryanh@linux.vnet.ibm.com
> >>>>> Cc: aliguori@us.ibm.com
> >>>>> Cc: kvm@vger.kernel.org
> >>>>>
> >>>>> ---
> >>>>> Changes from V2:
> >>>>>  Move ioctl into hw/kvmclock.c so as other arches can use it as it is
> >>>>> implemented
> >>>>>
> >>>>> Changes from V1:
> >>>>>  Remove unnecessary encapsulating function
> >>>>>
> >>>>>  hw/kvmclock.c |   24 ++++++++++++++++++++++++
> >>>>>  1 files changed, 24 insertions(+), 0 deletions(-)
> >>>>>
> >>>>> diff --git a/hw/kvmclock.c b/hw/kvmclock.c
> >>>>> index 5388bc4..756839f 100644
> >>>>> --- a/hw/kvmclock.c
> >>>>> +++ b/hw/kvmclock.c
> >>>>> @@ -16,6 +16,7 @@
> >>>>>  #include "sysbus.h"
> >>>>>  #include "kvm.h"
> >>>>>  #include "kvmclock.h"
> >>>>> +#include "cpu-all.h"
> >>>>>  
> >>>>>  #include <linux/kvm.h>
> >>>>>  #include <linux/kvm_para.h>
> >>>>> @@ -69,11 +70,34 @@ static void kvmclock_vm_state_change(void *opaque, int running,
> >>>>>      }
> >>>>>  }
> >>>>>  
> >>>>> +static void kvmclock_vm_state_change_vcpu(void *opaque, int running,
> >>>>> +                                          RunState state)
> >>>>> +{
> >>>>> +    int ret;
> >>>>> +    CPUState *penv = first_cpu;
> >>>>> +
> >>>>> +    if (running) {
> >>>>> +	while (penv) {
> >>>>
> >>>> or: for (cpu = first_cpu; cpu != NULL; cpu = cpu->next_cpu) {
> >>>>
> >>>
> >>> Functionally equivalent and I see both in the code, is there a standard?
> >>
> >> Not really. I once tried to introduce an iterator macro, but it was
> >> refused. The above is just more compact.
> >>
> >> But this is only a minor nit.
> >>
> >>>
> >>>>> +            ret = kvm_vcpu_ioctl(penv, KVM_GUEST_PAUSED, 0);
> >>>>> +            if (ret) {
> >>>>> +                if (ret != ENOSYS) {
> >>>>> +                    fprintf(stderr,
> >>>>> +                            "kvmclock_vm_state_change_vcpu: %s\n",
> >>>>> +                            strerror(-ret));
> >>>>> +                }
> >>>>> +                return;
> >>>>> +            }
> >>>>> +            penv = (CPUState *)penv->next_cpu;
> >>>>
> >>>> Unneeded cast.
> >>>>
> >>>
> >>> Also following an example seen elsewhere.
> >>
> >> Generally, we try to avoid those pointless casts.
> >>
> >>>
> >>>>> +        }
> >>>>> +    }
> >>>>> +}
> >>>>> +
> >>>>
> >>>> Again: please use checkpatch.pl.
> >>>>
> >>>
> >>> Sorry, tough to get used to hitting space bar that many times...
> >>>
> >>>>>  static int kvmclock_init(SysBusDevice *dev)
> >>>>>  {
> >>>>>      KVMClockState *s = FROM_SYSBUS(KVMClockState, dev);
> >>>>>  
> >>>>>      qemu_add_vm_change_state_handler(kvmclock_vm_state_change, s);
> >>>>> +    qemu_add_vm_change_state_handler(kvmclock_vm_state_change_vcpu, NULL);
> >>>>>      return 0;
> >>>>>  }
> >>>>>  
> >>>>
> >>>> Why not extend the existing handler?
> >>>
> >>> Because the new handler doesn't touch the KVMClockState object.  If this is
> >>> preferred, I have no objection.
> >>
> >> The separate registration looks strange to me. And the fact that you
> >> don't need to object doesn't justify a callback of its own.
> >>
> >>>
> >>>>
> >>>> I still wonder if the IOCTL interface is actually kvmclock specific. But
> >>>> Marcello asked for this, and we could still change it when some arch
> >>>> comes around that provides it independent of kvmclock.
> >>>
> >>> The flag itself is stored in the pvclock_vcpu_time_info structure, and anything
> >>> else that touches that structure uses ioctls.
> >>
> >> That's the host-guest interface. But I'm talking about the kvm-qemu
> >> interface here which has no relation to how the "was paused" information
> >> is transferred to the guest.
> >>
> >> Jan
> > 
> > It is one simple, rarely used command. I don't see why another interface
> > such as kvm_run would be beneficial for this case.
> > 
> 
> I was referring to the relation between the IOCTL and kvmclock, but
> IOCTL vs. kvm_run.
> 
> Jan

Ah, OK. Yes, we better characterize it as KVMCLOCK specific (a generic
"guest is paused" command is not the scope of this patch).

So appending KVMCLOCK_ to the ioctl definitions would make that more
explicit.