Scheduler/hypervisor traps, vm_event_vcpu_pause - some details on how they work?

Scheduler/hypervisor traps, vm_event_vcpu_pause - some details on how they work?

[Corneliu ZUZU]

Hey all,

I'm trying to add a vm-event function that would execute just before a 
vCPU resumes execution. To do this, I need to know precisely how and 
when this happens.
I've been browsing the codebase to try and gain some insight on the 
matter, but I still need some clarifications as parts of it are not so 
straight-forward.

Currently my understanding is that a vCPU can be entered in 3 ways:

1) The first entry, when the vCPU is first created. On ARM, the PC of 
the vCPU is set to continue_new_vcpu.
     Q1. How does this happen on X86?

2) After a hypervisor trap (X86 VMENTER/ARM ERET).
     Q2. How does pausing of the current vCPU (vm_event_vcpu_pause) 
work, specifically where does the context switch to another vCPU happen? 
Is it reentered through the below mentioned scheduling functions?
     Q3. Can we resume on another vCPU than the one that trapped if it 
wasn't paused?
     Q4. Can a hypervisor trap be interrupted by a scheduling operation?

3) Scheduler interrupt -> functions context_switch/continue_running.

Q5. Are these 3 the only ways a vCPU can be entered?

Thank you very much,
Corneliu.

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
http://lists.xen.org/xen-devel