From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from eggs.gnu.org ([2001:4830:134:3::10]:40340) by lists.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1XRLQ8-00037H-Et for qemu-devel@nongnu.org; Tue, 09 Sep 2014 09:25:26 -0400 Received: from Debian-exim by eggs.gnu.org with spam-scanned (Exim 4.71) (envelope-from ) id 1XRLQ3-0000pu-Hz for qemu-devel@nongnu.org; Tue, 09 Sep 2014 09:25:20 -0400 Received: from mx1.redhat.com ([209.132.183.28]:12777) by eggs.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1XRLQ3-0000pm-Au for qemu-devel@nongnu.org; Tue, 09 Sep 2014 09:25:15 -0400 From: Juan Quintela In-Reply-To: <1410265809-27247-4-git-send-email-pbonzini@redhat.com> (Paolo Bonzini's message of "Tue, 9 Sep 2014 14:30:02 +0200") References: <1410265809-27247-1-git-send-email-pbonzini@redhat.com> <1410265809-27247-4-git-send-email-pbonzini@redhat.com> Date: Tue, 09 Sep 2014 15:25:10 +0200 Message-ID: <87fvg12afd.fsf@troll.troll> MIME-Version: 1.0 Content-Type: text/plain Subject: Re: [Qemu-devel] [PATCH 03/10] cpu: init vmstate for ticks and clock offset Reply-To: quintela@redhat.com List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , To: Paolo Bonzini Cc: amit.shah@redhat.com, qemu-devel@nongnu.org, Pavel.Dovgaluk@ispras.ru, dgilbert@redhat.com Paolo Bonzini wrote: > From: Pavel Dovgalyuk > > Ticks and clock offset used by CPU timers have to be saved in vmstate. > But vmstate for these fields registered only in icount mode. > Missing registration leads to breaking the continuity when vmstate is loaded. > This patch introduces new initialization function which fixes this. > > Signed-off-by: Pavel Dovgalyuk > Signed-off-by: Paolo Bonzini Nack. This break migration when using with older machine types. Several questions before I propose an alternative: - are we this fields always needed (then why migration has worked until now) - after reding cpus.c the question is .... why it ever worked? After reading cpus.c, it appears that we only really need this functionality for qemu_clock_get_ns() when called with QEMU_CLOCK_VIRTUAL with icount disabled, right? further grep shows that it is used for acpi_pm_tmr_get_clock, whatever that is. When is this used, thanks? Later, Juan. > --- > cpus.c | 8 ++++++-- > include/qemu-common.h | 2 ++ > vl.c | 1 + > 3 files changed, 9 insertions(+), 2 deletions(-) > > diff --git a/cpus.c b/cpus.c > index 0f7d0ea..2a0e133 100644 > --- a/cpus.c > +++ b/cpus.c > @@ -493,13 +493,17 @@ static const VMStateDescription vmstate_timers = { > } > }; > > +void cpu_ticks_init(void) > +{ > + seqlock_init(&timers_state.vm_clock_seqlock, NULL); > + vmstate_register(NULL, 0, &vmstate_timers, &timers_state); > +} > + > void configure_icount(QemuOpts *opts, Error **errp) > { > const char *option; > char *rem_str = NULL; > > - seqlock_init(&timers_state.vm_clock_seqlock, NULL); > - vmstate_register(NULL, 0, &vmstate_timers, &timers_state); > option = qemu_opt_get(opts, "shift"); > if (!option) { > if (qemu_opt_get(opts, "align") != NULL) { > diff --git a/include/qemu-common.h b/include/qemu-common.h > index bcf7a6a..dcb57ab 100644 > --- a/include/qemu-common.h > +++ b/include/qemu-common.h > @@ -105,6 +105,8 @@ static inline char *realpath(const char *path, char *resolved_path) > } > #endif > > +void cpu_ticks_init(void); > + > /* icount */ > void configure_icount(QemuOpts *opts, Error **errp); > extern int use_icount; > diff --git a/vl.c b/vl.c > index 15aea95..5db0d08 100644 > --- a/vl.c > +++ b/vl.c > @@ -4334,6 +4334,7 @@ int main(int argc, char **argv, char **envp) > qemu_spice_init(); > #endif > > + cpu_ticks_init(); > if (icount_opts) { > if (kvm_enabled() || xen_enabled()) { > fprintf(stderr, "-icount is not allowed with kvm or xen\n");