Bug 2208674

This bug was initially created as a copy of Bug #2185076

I am copying this bug because: 



Description of problem:

For using guests which require timely injection of PIT interrupts,
kvm-pit process should be pinned to same CPU as vcpu0. 

Version-Release number of selected component (if applicable):


How reproducible:


Steps to Reproduce:
1. start guest which uses PIT timer (RHEL-5 with clock=pit for example).
2. measure rate of interrupt injection with tracing.
3. 

Actual results:

kvm-pit running on pCPU-0 might not execute, due to additional load on CPU-0,
for longer periods of time.

          <idle>-0     [000] d.h. 102470.847586: hrtimer_expire_entry: hrtimer=00000000e3dea339 function=pit_timer_fn [kvm] now=102463903132805
          <idle>-0     [000] d.h. 102470.847587: sched_waking: comm=kvm-pit/67 pid=1212317 prio=120 target_cpu=001
          <idle>-0     [001] d... 102470.847590: sched_switch: prev_comm=swapper/1 prev_pid=0 prev_prio=120 prev_state=S ==> next_comm=kvm-pit/67 next_pid=1212317 next_prio=120
      kvm-pit/67-1212317 [001] d... 102470.847594: sched_switch: prev_comm=kvm-pit/67 prev_pid=1212317 prev_prio=120 prev_state=D ==> next_comm=swapper/1 next_pid=0 next_prio=120
       CPU 0/KVM-1212312 [006] d... 102470.847596: kvm_entry: vcpu 0, rip 0x35fd
       CPU 0/KVM-1212312 [006] d... 102470.847597: kvm_exit: vcpu 0 reason EOI_INDUCED rip 0x1bab info1 0x0000000000000022 info2 0x0000000000000000 intr_info 0x00000000 error_code 0x00000000
       CPU 0/KVM-1212312 [006] d... 102470.847598: kvm_entry: vcpu 0, rip 0x1bab
...
       CPU 0/KVM-1212312 [006] d... 102470.847622: kvm_exit: vcpu 0 reason HLT rip 0x35fc info1 0x0000000000000000 info2 0x0000000000000000 intr_info 0x00000000 error_code 0x00000000
           <...>-64244 [000] d.h. 102470.857590: hrtimer_expire_entry: hrtimer=00000000e3dea339 function=pit_timer_fn [kvm] now=102463913135899
           <...>-64244 [000] d.h. 102470.857599: sched_waking: comm=kvm-pit/67 pid=1212317 prio=120 target_cpu=001
  kube-apiserver-64649 [004] d.h. 102470.860004: hrtimer_expire_entry: hrtimer=00000000f5450c52 function=pit_timer_fn [kvm] now=102463915550678
           <...>-64245 [003] d.h. 102470.863029: hrtimer_expire_entry: hrtimer=000000004736fe96 function=pit_timer_fn [kvm] now=102463918575518
           <...>-64244 [000] d.h. 102470.867587: hrtimer_expire_entry: hrtimer=00000000e3dea339 function=pit_timer_fn [kvm] now=102463923133350
           <...>-64244 [000] d.h. 102470.869780: hrtimer_expire_entry: hrtimer=000000005deb41ca function=pit_timer_fn [kvm] now=102463925326017
           <...>-64244 [000] d... 102470.877465: sched_switch: prev_comm=kube-apiserver prev_pid=64244 prev_prio=120 prev_state=S ==> next_comm=kvm-pit/67 next_pid=1212317 next_prio=120
      kvm-pit/67-1212317 [000] d... 102470.877483: sched_switch: prev_comm=kvm-pit/67 prev_pid=1212317 prev_prio=120 prev_state=D ==> next_comm=cluster-policy- next_pid=60163 next_prio=120
       CPU 0/KVM-1212312 [006] d... 102470.877486: kvm_entry: vcpu 0, rip 0x35fd
       CPU 0/KVM-1212312 [006] d... 102470.877490: kvm_exit: vcpu 0 reason EOI_INDUCED rip 0x1bab info1 0x0000000000000022 info2 0x0000000000000000 intr_info 0x00000000 error_code 0x00000000
       CPU 0/KVM-1212312 [006] d... 102470.877491: sched_waking: comm=kvm-pit/67 pid=1212317 prio=120 target_cpu=000
       CPU 0/KVM-1212312 [006] d... 102470.877494: kvm_entry: vcpu 0, rip 0x1bab

Expected results:

PIT timer interrupt injection should not deviate more than 500us from expected (armed) time, when period is 10ms.


Additional info:

To fix this kubevirt should pin kvm-pit thread, for that VM, to pCPU of vCPU-0.