[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <878s5t7xbz.fsf@vitty.brq.redhat.com>
Date: Thu, 08 Apr 2021 14:52:16 +0200
From: Vitaly Kuznetsov <vkuznets@...hat.com>
To: Paolo Bonzini <pbonzini@...hat.com>,
Siddharth Chandrasekaran <sidcha@...zon.de>
Cc: Alexander Graf <graf@...zon.com>,
Evgeny Iakovlev <eyakovl@...zon.de>, kvm@...r.kernel.org,
linux-kernel@...r.kernel.org,
Sean Christopherson <seanjc@...gle.com>,
Wanpeng Li <wanpengli@...cent.com>,
Jim Mattson <jmattson@...gle.com>,
Joerg Roedel <joro@...tes.org>,
Thomas Gleixner <tglx@...utronix.de>,
Ingo Molnar <mingo@...hat.com>, Borislav Petkov <bp@...en8.de>,
x86@...nel.org, "H. Peter Anvin" <hpa@...or.com>
Subject: Re: [PATCH 3/4] KVM: x86: kvm_hv_flush_tlb use inputs from XMM
registers
Paolo Bonzini <pbonzini@...hat.com> writes:
> On 08/04/21 14:01, Vitaly Kuznetsov wrote:
>>
>> Also, we can probably defer kvm_hv_hypercall_read_xmm() until we know
>> how many regs we actually need to not read them all (we will always
>> need xmm[0] I guess so we can as well read it here).
>
> The cost is get/put FPU, so I think there's not much to gain from that.
>
Maybe, I just think that in most cases we will only need xmm0. To make
the optimization work we can probably do kvm_get_fpu() once we figured
out that we're dealing with XMM hypercall and do kvm_put_fpu() when
we're done processing hypercall parameters. This way we don't need to do
get/put twice. We can certainly leave this idea to the (possible) future
optimizations.
--
Vitaly
Powered by blists - more mailing lists