From mboxrd@z Thu Jan 1 00:00:00 1970 From: Paolo Bonzini Subject: Re: [patch 2/2] KVM: x86: add option to advance tscdeadline hrtimer expiration Date: Thu, 11 Dec 2014 19:58:47 +0100 Message-ID: <5489E967.3060301@redhat.com> References: <20141210205749.035440781@redhat.com> <20141210205904.415174860@redhat.com> <5488D955.1050002@redhat.com> <20141211030753.GA6358@amt.cnet> Mime-Version: 1.0 Content-Type: text/plain; charset=windows-1252 Content-Transfer-Encoding: 7bit Cc: kvm@vger.kernel.org, Luiz Capitulino , Rik van Riel , Radim Krcmar To: Marcelo Tosatti Return-path: Received: from mail-wg0-f45.google.com ([74.125.82.45]:56375 "EHLO mail-wg0-f45.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1758583AbaLKS6w (ORCPT ); Thu, 11 Dec 2014 13:58:52 -0500 Received: by mail-wg0-f45.google.com with SMTP id b13so7225509wgh.32 for ; Thu, 11 Dec 2014 10:58:50 -0800 (PST) In-Reply-To: <20141211030753.GA6358@amt.cnet> Sender: kvm-owner@vger.kernel.org List-ID: On 11/12/2014 04:07, Marcelo Tosatti wrote: >> > So, on APICv, if the interrupt is in PIR but it has bits 7:4 <= >> > PPR[7:4], you have a problem. :( There is no APICv hook that lets you >> > get a vmexit when the PPR becomes low enough. > Well, you simply exit earlier and busy spin for VM-exit > time. Okay, given how small the jitter is in your plots, a small busy wait is not the end of the world even if it caused a very short priority inversion. Can you add a tracepoint that triggers when you do the busy wait, and possibly a kvm stat that counts good (no busy wait) vs. bad (busy wait) invocations of wait_lapic_expire? Thanks, Paolo