On 09/03/2018 04:10 PM, Nikita Leshenko wrote:
> On September 2, 2018 5:21:15 AM, fengguang.wu@xxxxxxxxx wrote:
>> diff --git a/virt/kvm/kvm_main.c b/virt/kvm/kvm_main.c
>> index 8b47507faab5..0c483720de8d 100644
>> --- a/virt/kvm/kvm_main.c
>> +++ b/virt/kvm/kvm_main.c
>> @@ -3892,6 +3892,7 @@ static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm)
>> if (type == KVM_EVENT_CREATE_VM) {
>> add_uevent_var(env, "EVENT=create");
>> kvm->userspace_pid = task_pid_nr(current);
>> + current->kvm = kvm;
>
> Is it OK to store `kvm` on the task_struct? What if the thread that
> originally created the VM exits? From the documentation it seems
> like a VM is associated with an address space and not a specific
> thread, so maybe it should be stored on mm_struct?
Yes, ioctls accessing the kvm can happen from all threads.
>
> From Documentation/virtual/kvm/api.txt:
> Only run VM ioctls from the same process (address space) that was used
> to create the VM.
>
> -Nikita
>> } else if (type == KVM_EVENT_DESTROY_VM) {
>> add_uevent_var(env, "EVENT=destroy");
>> }
>> --
>> 2.15.0
>>
>>
>>
>
