Re: [Xen-devel] [PATCH 7/7] xen/arm64: Implement a fast path for handling SMCCC_ARCH_WORKAROUND_1

[Volodymyr Babchuk <vlad.babchuk@xxxxxxxxx>]

Hi Julien,


On 6 February 2018 at 20:33, Julien Grall <julien.grall@xxxxxxx> wrote:
> Hi,
>
> On 02/06/2018 04:36 PM, Volodymyr Babchuk wrote:
>>
>> On 5 February 2018 at 15:20, Julien Grall <julien.grall@xxxxxxx> wrote:
>>>
>>> The function SMCCC_ARCH_WORKAROUND_1 will be called by the guest for
>>> hardening the branch predictor. So we want the handling to be as fast as
>>> possible.
>>>
>>> As the mitigation is applied on every guest exit, we can check for the
>>> call before saving all the context and return very early.
>>
>> Have you tried any benchmarks? How big is the benefit?
>
>
> I have benchmarked but I can't share the result. I can give you an idea on
> how this could benefits Xen.
>
> Linux will call the workaround on every context switch between process. So
> imagine for each context switch, you have will enter in Xen and in the
> following order:
>         1) enter Xen
>         2) apply the workaround which means calling EL3.
>         3) save part of the guest context
>         4) call enter_hypervisor_head that will sync the vGIC state
>         5) detect you actually call SMCCC_ARCH_WORKAROUND_1 that will do
> nothing
>         6) call leave_hypervisor_tail that will sync back the vGIC state and
> execute softirq (that could reschedule the vCPU)
>         7) restore the guest context
>         8) return to the guest
>
> So effectively, instead of executing hundreds (if not thousands)
> instructions each time, you will end up only executing less than 50
> instructions.
Sounds fine.

>
>>>
>>> For now, only provide a fast path for HVC64 call. Because the code rely
>>> on 2 registers, x0 and x1 are saved in advanced.
>>
>> Is there a typo? Should it be "advance"?
>>
>>>
>>> Signed-off-by: Julien Grall <julien.grall@xxxxxxx>
Provided that above typo is fixed:

Reviewed-by: Volodymyr Babchuk <volodymyr_babchuk@xxxxxxxx>
>>>
>>> ---
>>>      guest_sync only handle 64-bit guest, so I have only implemented the
>>>      64-bit side for now. We can discuss whether it is useful to
>>>      implement it for 32-bit guests.
>>>
>>>      We could also consider to implement the fast path for SMC64,
>>>      althought a guest should always use HVC.
>>
>> I can imagine a guest that know nothing about virtualization and use
>> SMC as a conduit. But I can't provide real world example, thou.
I'm okay with this.


-- 
WBR Volodymyr Babchuk aka lorc [+380976646013]
mailto: vlad.babchuk@xxxxxxxxx

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxxx
https://lists.xenproject.org/mailman/listinfo/xen-devel