From mboxrd@z Thu Jan 1 00:00:00 1970 From: Paolo Bonzini Subject: Re: [PATCH 5/5] KVM: nVMX: Enable nested posted interrupt processing. Date: Tue, 20 Jan 2015 09:04:30 +0100 Message-ID: <54BE0C0E.3040902@redhat.com> References: <54BCEDD5.40301@redhat.com> <54BE04F7.7030402@redhat.com> Mime-Version: 1.0 Content-Type: text/plain; charset=utf-8 Content-Transfer-Encoding: 7bit Cc: gleb@kernel.org, yang.z.zhang@intel.com, kvm@vger.kernel.org, linux-kernel@vger.kernel.org, Wanpeng Li , Jan Kiszka To: Wincy Van Return-path: In-Reply-To: Sender: linux-kernel-owner@vger.kernel.org List-Id: kvm.vger.kernel.org On 20/01/2015 08:54, Wincy Van wrote: > On Tue, Jan 20, 2015 at 3:34 PM, Paolo Bonzini wrote: >>> Hence, we can disable local interrupts while delivering nested posted >>> interrupts to make sure >>> we are faster than the destination vcpu. This is a bit tricky but it >>> an avoid that race. I think we >>> do not need to add a spin lock here. RCU does not fit this case, since >>> it will introduce a >>> new race window between the rcu handler and handle_vmptr**. >>> >>> I am wondering that whether there is a better way : ) >> >> Why not just use a spinlock? >> > > Hmm.. it seems that using a spinlock is the best way. > I think we can drop the local_irq_save and use a spinlock instead. > I can send v2 if it is necessary, any more ideas? Yes, please send v2 of this patch only. Paolo