Is it safe to assume that v == current in hvm_do_resume()?

Is it safe to assume that v == current in hvm_do_resume()?

[Razvan Cojocaru]

Hello,

As far as I've understood the code, and based on several tests, it would
appear that in xen/arch/x86/hvm/hvm.c, the struct vcpu *v parameter to
hvm_do_resume() is always equal to current.

Is that a reasonable assumption to make? I could have missed some use
cases for the function.


Thanks,
Razvan

Re: Is it safe to assume that v == current in hvm_do_resume()?

[Andrew Cooper]

On 22/10/14 09:11, Razvan Cojocaru wrote:
> Hello,
>
> As far as I've understood the code, and based on several tests, it would
> appear that in xen/arch/x86/hvm/hvm.c, the struct vcpu *v parameter to
> hvm_do_resume() is always equal to current.
>
> Is that a reasonable assumption to make? I could have missed some use
> cases for the function.

The calls look like this:

{vmx,svm}_do_resume is the schedule_tail function pointer, which is a
no-return function and called on the way out of a context switch.

They call hvm_do_resume() followed by a reset_stack_and_jump() to
{vmx,svm}_asm_do_vmentry.

So yes - you can be guaranteed that the v in hvm_do_resume() is current.

If you do come to explicitly rely on this behaviour, and it is in any
way unclear, it might be worth introducing an ASSERT(v == current) near
to the expectation.

~Andrew

Re: Is it safe to assume that v == current in hvm_do_resume()?

[Razvan Cojocaru]

On 10/22/2014 12:04 PM, Andrew Cooper wrote:
> On 22/10/14 09:11, Razvan Cojocaru wrote:
>> Hello,
>>
>> As far as I've understood the code, and based on several tests, it would
>> appear that in xen/arch/x86/hvm/hvm.c, the struct vcpu *v parameter to
>> hvm_do_resume() is always equal to current.
>>
>> Is that a reasonable assumption to make? I could have missed some use
>> cases for the function.
> 
> The calls look like this:
> 
> {vmx,svm}_do_resume is the schedule_tail function pointer, which is a
> no-return function and called on the way out of a context switch.
> 
> They call hvm_do_resume() followed by a reset_stack_and_jump() to
> {vmx,svm}_asm_do_vmentry.
> 
> So yes - you can be guaranteed that the v in hvm_do_resume() is current.
> 
> If you do come to explicitly rely on this behaviour, and it is in any
> way unclear, it might be worth introducing an ASSERT(v == current) near
> to the expectation.

Thanks for the confirmation!

Indeed, I had placed an ASSERT() before the code that needs v == current.


Thanks,
Razvan