From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1755531Ab0CCSbr (ORCPT ); Wed, 3 Mar 2010 13:31:47 -0500 Received: from claw.goop.org ([74.207.240.146]:41228 "EHLO claw.goop.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1754490Ab0CCSbp (ORCPT ); Wed, 3 Mar 2010 13:31:45 -0500 Message-ID: <4B8EAB0F.3000104@goop.org> Date: Wed, 03 Mar 2010 10:31:43 -0800 From: Jeremy Fitzhardinge User-Agent: Mozilla/5.0 (X11; U; Linux x86_64; en-US; rv:1.9.1.7) Gecko/20100120 Fedora/3.0.1-1.fc12 Lightning/1.0b2pre Thunderbird/3.0.1 MIME-Version: 1.0 To: Sheng Yang CC: Keir Fraser , Jeremy Fitzhardinge , Ian Pratt , linux-kernel@vger.kernel.org, xen-devel , Ian Campbell , Stefano Stabellini Subject: Re: [Xen-devel] [PATCH 5/7] xen: Make event channel work with PV extension of HVM References: <1267436315-24486-1-git-send-email-sheng@linux.intel.com> <1267436315-24486-6-git-send-email-sheng@linux.intel.com> <4B8C6C0D.3070005@goop.org> <201003021348.06360.sheng@linux.intel.com> In-Reply-To: <201003021348.06360.sheng@linux.intel.com> Content-Type: text/plain; charset=ISO-8859-15; format=flowed Content-Transfer-Encoding: 7bit Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On 03/01/2010 09:48 PM, Sheng Yang wrote: >> Presumably even if we don't have PV_EVTCHN available/enabled, the Xen >> clocksource would be available for getting time? >> > I think currently Xen pv clocksource and clockevent are binding... Not sure if > a single line "clocksource_register(&xen_clocksource)" can work. I would give > it a try, maybe add a new PV feature. > There should be no strong binding between them, but there may be some sloppy assumptions in xen/time.c which should be fixed. Linux itself treats clocksources and eventsources as completely distinct entities, and doesn't assume they're running on the same timebase (for example). Having a PV clocksource even if the timer interrupts are emulated would make sense and be useful. >>> xen_setup_vcpu_info_placement(); >>> } >>> @@ -480,3 +487,138 @@ void __init xen_smp_init(void) >>> xen_fill_possible_map(); >>> xen_init_spinlocks(); >>> } >>> + >>> +static __cpuinit void xen_hvm_pv_start_secondary(void) >>> +{ >>> + int cpu = smp_processor_id(); >>> + >>> + cpu_init(); >>> + touch_nmi_watchdog(); >>> + preempt_disable(); >>> + >>> + /* otherwise gcc will move up smp_processor_id before the cpu_init */ >>> + barrier(); >>> + /* >>> + * Check TSC synchronization with the BSP: >>> + */ >>> + check_tsc_sync_target(); >>> + >>> + /* Done in smp_callin(), move it here */ >>> + set_mtrr_aps_delayed_init(); >>> + smp_store_cpu_info(cpu); >>> + >>> + /* This must be done before setting cpu_online_mask */ >>> + set_cpu_sibling_map(cpu); >>> + wmb(); >>> + >>> + set_cpu_online(smp_processor_id(), true); >>> + per_cpu(cpu_state, smp_processor_id()) = CPU_ONLINE; >>> + >>> + /* enable local interrupts */ >>> + local_irq_enable(); >>> + >>> + xen_setup_cpu_clockevents(); >>> >> How much of this is necessary? At this point, isn't CPU bringup the >> same as PV? >> > Xen_enable_sysenter/syscall is not needed for this. And we have a TSC sync > here - but it seems unnecessary for PV. But set_mtrr_aps_delayed_init() is > needed. Reuse the cpu_bring_up() is fine? > Doesn't Xen arrange for the tscs to be synced anyway? >> Is the MMUEXT_TLB_FLUSH/INVLPG_MULTI hypercall not currently available >> to HVM? >> > I think they are different. These hypercalls flushed native's TLB, but HVM > want to flush guest one, especially when using shadow, HVM need do something > for it. > I see. J