From mboxrd@z Thu Jan 1 00:00:00 1970 From: Paolo Bonzini Subject: Re: [PATCH 1/2] KVM: introduce __vmx_flush_tlb to handle specific vpid Date: Fri, 25 Sep 2015 10:01:53 +0200 Message-ID: <5604FF71.6090109@redhat.com> References: Mime-Version: 1.0 Content-Type: text/plain; charset=windows-1252 Content-Transfer-Encoding: 8bit Cc: Jan Kiszka , Wincy Van , kvm@vger.kernel.org, linux-kernel@vger.kernel.org To: Bandan Das , Wanpeng Li Return-path: In-Reply-To: Sender: linux-kernel-owner@vger.kernel.org List-Id: kvm.vger.kernel.org On 24/09/2015 18:12, Bandan Das wrote: > Not sure myself what's the right thing to do but this may be undesirable > in a nested environment. Assuming the processor supports global invalidation > only, this seems like a easy way for the nested guest to invalidate *all* > mappings - even the L1 specific mappings. It's not a great thing but it's already what happens if you do a global INVEPT (it calls vmx_flush_tlb, which results in a global INVVPID if the single-context variant is not supported). Even without nested virt a single guest could slow down all other guests just by triggering frequent TLB flushes (e.g. by moving around a ROM BAR thousands of times per second). It would help to know _which_ processors actually don't support single-context INVVPIDs... Paolo