Re: [Xen-devel] xc_hvm_inject_trap() races

[Razvan Cojocaru <rcojocaru@xxxxxxxxxxxxxxx>]

On 11/08/2016 10:15 AM, Jan Beulich wrote:
>>>> On 07.11.16 at 18:01, <rcojocaru@xxxxxxxxxxxxxxx> wrote:
>> On 11/07/2016 06:10 PM, Jan Beulich wrote:
>>>>>> On 07.11.16 at 16:24, <rcojocaru@xxxxxxxxxxxxxxx> wrote:
>>>> The one-shot vm_event does sound reasonable. I could set a flag
>>>> per-VCPU, basically similar to v->arch.hvm_vcpu.inject_trap.vector, and
>>>> fire a single event from hvm_inject_trap() if it's set (then unset it) -
>>>> the flag would be set via an xc_monitor_next_interrupt() call in libxc.
>>>
>>> Doing this in hvm_inject_trap() would not cover all cases afict.
>>> I'd suggest doing this from hvm_do_resume() _after_ the
>>> (conditional) call to hvm_inject_trap(), if there is _any_ event
>>> pending.
>>
>> But that would only cover the hypercall-injected traps. The condition in
>> hvm_do_resume() is "if ( v->arch.hvm_vcpu.inject_trap.vector != -1 )",
>> and inject_trap.vector seems to only ever be set by the hypercall:
>> [...]
>> So if the next interrupt is not caused by the hypercall, we'll never get
>> another event. Am I reading the code wrong?
> 
> No, maybe I expressed myself ambiguously: I meant to say that the
> event should be delivered from hvm_do_resume(), but _outside_ the
> conditional guarding the call to hvm_inject_trap(). Otherwise things
> would have been worse than when doing it inside hvm_inject_trap().

Right, definitely, I'll do that.


Thanks,
Razvan

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