Re: [PATCH v8 4/6] arm/vm_event: get/set registers

[Julien Grall <julien.grall@arm.com>]

On 06/07/2016 20:23, Tamas K Lengyel wrote:
> On Wed, Jul 6, 2016 at 1:43 AM, Jan Beulich <JBeulich@suse.com> wrote:
>>>>> On 05.07.16 at 20:37, <tamas.lengyel@zentific.com> wrote:
>>> +struct vm_event_regs_arm64 {
>>> +    uint64_t x0;
>>> +    uint64_t x1;
>>> +    uint64_t x2;
>>> +    uint64_t x3;
>>> +    uint64_t x4;
>>> +    uint64_t x5;
>>> +    uint64_t x6;
>>> +    uint64_t x7;
>>> +    uint64_t x8;
>>> +    uint64_t x9;
>>> +    uint64_t x10;
>>> +    uint64_t x11;
>>> +    uint64_t x12;
>>> +    uint64_t x13;
>>> +    uint64_t x14;
>>> +    uint64_t x15;
>>> +    uint64_t x16;
>>> +    uint64_t x17;
>>> +    uint64_t x18;
>>> +    uint64_t x19;
>>> +    uint64_t x20;
>>> +    uint64_t x21;
>>> +    uint64_t x22;
>>> +    uint64_t x23;
>>> +    uint64_t x24;
>>> +    uint64_t x25;
>>> +    uint64_t x26;
>>> +    uint64_t x27;
>>> +    uint64_t x28;
>>> +    uint64_t x29;
>>> +    uint64_t x30;
>>> +    uint64_t pc;
>>> +};
>>
>> Isn't the stack pointer a fully separate register in aarch64? Not
>> making available something as essential as that seems wrong to
>> me.
>>
>
> The register is available for access already, so unless there is an
> actual use-case that requires it to be transmitted through vm_event I
> don't see the point for transmitting it. So as I mentioned in my other
> response, I'm inclined to reduce this patch to the bare essentials my
> use-case requires at this point and leave the extensions up for the
> future when - and if - it will be of use. Since this patch is just an
> optimization, if transmitting such reduced set is not acceptable for
> some reason, I'll forgo this patch entirely.

Here we go with again the same argument: "this is not necessary for my 
use-case". This data structure is part of the ABI between the hypervisor 
and the vm-event app, i.e modifying this structure for adding ARM64/ARM 
registers will result to an incompatibility with a previous version of 
the hypervisor. Better to do it now than in a couple of years when 
vm-event will have more users. I agree that it is time consuming to get 
an ABI correct, but it will save users to recompile/ship another version 
of their vm-event app because of this incompatibility.

As mentioned in a previous thread, the main use-case for trapping an SMC 
is emulating the call, hence a vm-event app would want to have access to 
the general-purpose registers. And yes, I know that your use-case is 
different and does not require those registers, I already expressed my 
concerns about it.

Now, if you drop this patch, how will you retrieve the vCPU register? I 
guess via DOMCTL_getvcpucontext? If so, if the vCPU has not been paused, 
you will get a context which is different compare to the time the 
vm-event has occurred. And yes, I know that in your use-case the vCPU is 
paused. This call will always be more expensive than passing the 
registers with event.

Anyway, I really don't think we ask for something really difficult to 
accomplish.

Regards,

-- 
Julien Grall

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