Re: [PART1 RFC v2 10/10] svm: Manage vcpu load/unload when enable AVIC

From: Radim KrÄmÃÅ
Date: Thu Mar 10 2016 - 09:02:00 EST

2016-03-09 22:46+0100, Radim KrÄmÃÅ:
> 2016-03-04 14:46-0600, Suravee Suthikulpanit:
>> From: Suravee Suthikulpanit <suravee.suthikulpanit@xxxxxxx>
>> When a vcpu is loaded/unloaded to a physical core, we need to update
>> information in the Physical APIC-ID table accordingly.
>> Signed-off-by: Suravee Suthikulpanit <suravee.suthikulpanit@xxxxxxx>
>> ---
>> +static int avic_vcpu_load(struct kvm_vcpu *vcpu, int cpu, bool is_load)
> This function does a lot and there is only one thing that must be done
> in svm_vcpu_load: change host physical APIC ID if the CPU changed.
> The rest can be done elsewhere:
> - is_running when blocking.

Well, we haven't reached an agreement on is_running yet. The situation:
if we don't unset vcpu1.is_running when vcpu1 is scheduled out and vcpu2
gets scheduled on vcpu1's physical core, then vcpu2 would receive a
doorbell intended to vcpu1.

We'd like to keep is_running set when there is no reason to vmexit, but
not if a guest can negatively affect other guests.
How does receiving a stray doorbell affect the performance?


(Toggling is_running on load/put is definitely safer, so it's a good
choice for first version.)