From mboxrd@z Thu Jan 1 00:00:00 1970 From: Keir Fraser Subject: Re: credit scheduler svc->flags access race? Date: Tue, 03 Jan 2012 07:33:28 +0000 Message-ID: References: <1325300125668-5111504.post@n5.nabble.com> Mime-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit Return-path: In-Reply-To: <1325300125668-5111504.post@n5.nabble.com> List-Unsubscribe: , List-Post: List-Help: List-Subscribe: , Sender: xen-devel-bounces@lists.xensource.com Errors-To: xen-devel-bounces@lists.xensource.com To: "Liu.yi" , xen-devel@lists.xensource.com Cc: George Dunlap List-Id: xen-devel@lists.xenproject.org On 31/12/2011 02:55, "Liu.yi" wrote: > hi,all > In my HP DL380G7 server, csched_vcpu_yield and csched_acct may access > svc->flags at the same time, when this happens vms stop running because > csched_vcpu_yield overwrites svc->flags which csched_acct set to > CSCHED_FLAG_VCPU_PARKED (My vm's schedule cap values is not 0). > Vms run fine if I modified schedule_credit.c as follows: Cc'ing George Dunlap. This is probably a good bug fix. -- Keir > --- xen/common/sched_credit.c 2010-12-10 10:19:45.000000000 +0800 > +++ ../../xen-4.1.0/xen/common/sched_credit.c 2010-12-31 > 10:47:39.000000000 +0800 > @@ -135,7 +135,7 @@ struct csched_vcpu { > struct vcpu *vcpu; > atomic_t credit; > s_time_t start_time; /* When we were scheduled (used for credit) */ > - uint16_t flags; > + uint32_t flags; > int16_t pri; > #ifdef CSCHED_STATS > struct { > @@ -787,7 +787,7 @@ csched_vcpu_yield(const struct scheduler > if ( !sched_credit_default_yield ) > { > /* Let the scheduler know that this vcpu is trying to yield */ > - sv->flags |= CSCHED_FLAG_VCPU_YIELD; > + set_bit(1, &sv->flags); > } > } > > @@ -1086,7 +1086,7 @@ csched_acct(void* dummy) > { > CSCHED_STAT_CRANK(vcpu_park); > vcpu_pause_nosync(svc->vcpu); > - svc->flags |= CSCHED_FLAG_VCPU_PARKED; > + set_bit(0, &svc->flags); > } > > /* Lower bound on credits */ > @@ -1111,7 +1111,7 @@ csched_acct(void* dummy) > */ > CSCHED_STAT_CRANK(vcpu_unpark); > vcpu_unpause(svc->vcpu); > - svc->flags &= ~CSCHED_FLAG_VCPU_PARKED; > + clear_bit(0, &svc->flags); > } > > /* Upper bound on credits means VCPU stops earning */ > @@ -1337,7 +1337,7 @@ csched_schedule( > * Clear YIELD flag before scheduling out > */ > if ( scurr->flags & CSCHED_FLAG_VCPU_YIELD ) > - scurr->flags &= ~(CSCHED_FLAG_VCPU_YIELD); > + clear_bit(1, &scurr->flags); > > Are these modification correct? Another interesting thing is that vms run > fine on other server even if these vms's credit cap value is not 0. I don't > know why. > My xen version is 4.1.0, dom0 kernel version is 2.6.32.41 from jerry's git a > few months before. > > Thanks > > liuyi > > > -- > View this message in context: > http://xen.1045712.n5.nabble.com/credit-scheduler-svc-flags-access-race-tp5111 > 504p5111504.html > Sent from the Xen - Dev mailing list archive at Nabble.com. > > _______________________________________________ > Xen-devel mailing list > Xen-devel@lists.xensource.com > http://lists.xensource.com/xen-devel