On 2012-02-11 12:25, Blue Swirl wrote:
> On Sat, Feb 11, 2012 at 10:06, Jan Kiszka <jan.kiszka@xxxxxx> wrote:
>> On 2012-02-11 11:02, Blue Swirl wrote:
>>> On Fri, Feb 10, 2012 at 18:31, Jan Kiszka <jan.kiszka@xxxxxxxxxxx> wrote:
>>>> As we have thread-local cpu_single_env now and KVM uses exactly one
>>>> thread per VCPU, we can drop the cpu_single_env updates from the loop
>>>> and initialize this variable only once during setup.
>>>
>>> I don't think this is correct. Maybe you missed the part that sets
>>> cpu_single_env to NULL, which I think is to annoy broken code that
>>> assumes that some CPU state is always globally available. This is not
>>> true for monitor context.
>>
>> I did check this before changing, and I see no such need. Particularly
>> as this old debugging help prevents valid use case.
>
> It looks like monitor code is safe now. But in several places there
> are checks like this in pc.c:
> DeviceState *cpu_get_current_apic(void)
> {
> if (cpu_single_env) {
> return cpu_single_env->apic_state;
> } else {
> return NULL;
> }
> }
>
> In cpu-exec.c, there are these lines:
> /* fail safe : never use cpu_single_env outside cpu_exec() */
> cpu_single_env = NULL;
That's legacy stuff from the pre-io-thread times. Nowadays,
cpu_single_env is logically either always valid (KVM) or switching
seamlessly between the VCPUs (TCG).
>
> I think using cpu_single_env is an indication of a problem, like poor
> code, layering violation or poor API (vmport). What is your use case?
We have a few poor ABIs like the VMware stuff or the KVM VAPIC, so we
have to live with it already for that use cases. Moreover, cpus.c use it
internally, like for vm_stop, to find out the caller context.
Jan
Attachment:
signature.asc
Description: OpenPGP digital signature
