On 02/25/2015 02:26 PM, Radim Krčmář wrote:
> 2015-02-24 15:25-0600, Joel Schopp:
>>>> - clr_cr_intercept(svm, INTERCEPT_CR0_WRITE);
>>>> } else {
>>>> set_cr_intercept(svm, INTERCEPT_CR0_READ);
>>> (There is no point in checking fpu_active if cr0s are equal.)
>>>
>>>> - set_cr_intercept(svm, INTERCEPT_CR0_WRITE);
>>> KVM uses lazy FPU and the state is undefined before the first access.
>>> We set cr0.ts when !svm->vcpu.fpu_active to detect the first access, but
>>> if we allow the guest to clear cr0.ts without exiting, it can access FPU
>>> with undefined state.
>> Thanks for the valuable feedback. It's apparent I hadn't thought
>> through the interaction with lazy FPU and will need to go back and
>> rethink my approach here.
> I don't think we can gain much without sacrificing some laziness, like:
> when a guest with lazy FPU clears CR0.TS, it is going to use that FPU,
> so we could pre-load FPU in this case and drop the write intercept too;
> guests that unconditionally clear CR0.TS would perform worse though.
>
> It's going to take a lot of time, but two hunks in your patch, that made
> selective intercept benefit from decode assists, look useful even now.
>
> Would you post them separately?
I can re-post those separately. They are less useful, though probably
still worth doing, on their own because SVM_EXIT_WRITE_CR0 takes
precidence over SVM_EXIT_CR0_SEL_WRITE
--
To unsubscribe from this list: send the line "unsubscribe kvm" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
