From patchwork Mon May 6 06:56:32 2019 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: =?utf-8?b?SsO8cmdlbiBHcm/Dnw==?= X-Patchwork-Id: 10930559 Return-Path: Received: from mail.wl.linuxfoundation.org (pdx-wl-mail.web.codeaurora.org [172.30.200.125]) by pdx-korg-patchwork-2.web.codeaurora.org (Postfix) with ESMTP id 937651390 for ; Mon, 6 May 2019 06:59:08 +0000 (UTC) Received: from mail.wl.linuxfoundation.org (localhost [127.0.0.1]) by mail.wl.linuxfoundation.org (Postfix) with ESMTP id 80F4F28397 for ; Mon, 6 May 2019 06:59:08 +0000 (UTC) Received: by mail.wl.linuxfoundation.org (Postfix, from userid 486) id 753392860A; Mon, 6 May 2019 06:59:08 +0000 (UTC) X-Spam-Checker-Version: SpamAssassin 3.3.1 (2010-03-16) on pdx-wl-mail.web.codeaurora.org X-Spam-Level: X-Spam-Status: No, score=-5.2 required=2.0 tests=BAYES_00,MAILING_LIST_MULTI, RCVD_IN_DNSWL_MED autolearn=ham version=3.3.1 Received: from lists.xenproject.org (lists.xenproject.org [192.237.175.120]) (using TLSv1.2 with cipher AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by mail.wl.linuxfoundation.org (Postfix) with ESMTPS id 637D128397 for ; Mon, 6 May 2019 06:59:07 +0000 (UTC) Received: from localhost ([127.0.0.1] helo=lists.xenproject.org) by lists.xenproject.org with esmtp (Exim 4.89) (envelope-from ) id 1hNXYs-0002Vn-Mf; Mon, 06 May 2019 06:57:18 +0000 Received: from us1-rack-dfw2.inumbo.com ([104.130.134.6]) by lists.xenproject.org with esmtp (Exim 4.89) (envelope-from ) id 1hNXYk-0002Ch-4R for xen-devel@lists.xenproject.org; Mon, 06 May 2019 06:57:10 +0000 X-Inumbo-ID: 26aa5e9b-6fcc-11e9-843c-bc764e045a96 Received: from mx1.suse.de (unknown [195.135.220.15]) by us1-rack-dfw2.inumbo.com (Halon) with ESMTPS id 26aa5e9b-6fcc-11e9-843c-bc764e045a96; Mon, 06 May 2019 06:57:03 +0000 (UTC) X-Virus-Scanned: by amavisd-new at test-mx.suse.de Received: from relay2.suse.de (unknown [195.135.220.254]) by mx1.suse.de (Postfix) with ESMTP id A51F3AF38; Mon, 6 May 2019 06:56:57 +0000 (UTC) From: Juergen Gross To: xen-devel@lists.xenproject.org Date: Mon, 6 May 2019 08:56:32 +0200 Message-Id: <20190506065644.7415-34-jgross@suse.com> X-Mailer: git-send-email 2.16.4 In-Reply-To: <20190506065644.7415-1-jgross@suse.com> References: <20190506065644.7415-1-jgross@suse.com> Subject: [Xen-devel] [PATCH RFC V2 33/45] xen/sched: add code to sync scheduling of all vcpus of a sched item X-BeenThere: xen-devel@lists.xenproject.org X-Mailman-Version: 2.1.23 Precedence: list List-Id: Xen developer discussion List-Unsubscribe: , List-Post: List-Help: List-Subscribe: , Cc: Juergen Gross , Stefano Stabellini , Wei Liu , Konrad Rzeszutek Wilk , George Dunlap , Andrew Cooper , Ian Jackson , Tim Deegan , Julien Grall , Jan Beulich , Dario Faggioli , =?utf-8?q?Roger_Pau_Monn=C3=A9?= MIME-Version: 1.0 Errors-To: xen-devel-bounces@lists.xenproject.org Sender: "Xen-devel" X-Virus-Scanned: ClamAV using ClamSMTP When switching sched items synchronize all vcpus of the new item to be scheduled at the same time. A variable sched_granularity is added which holds the number of vcpus per schedule item. As tasklets require to schedule the idle item it is required to set the tasklet_work_scheduled parameter of do_schedule() to true if any cpu covered by the current schedule() call has any pending tasklet work. For joining other vcpus of the schedule item we need to add a new softirq SCHED_SLAVE_SOFTIRQ in order to have a way to initiate a context switch without calling the generic schedule() function selecting the vcpu to switch to, as we already know which vcpu we want to run. This has the other advantage not to loose any other concurrent SCHEDULE_SOFTIRQ events. Signed-off-by: Juergen Gross --- RFC V2: move syncing after context_switch() to schedule.c --- xen/arch/arm/domain.c | 2 +- xen/arch/x86/domain.c | 3 +- xen/common/schedule.c | 339 +++++++++++++++++++++++++++++++++------------ xen/common/softirq.c | 6 +- xen/include/xen/sched-if.h | 1 + xen/include/xen/sched.h | 16 ++- xen/include/xen/softirq.h | 1 + 7 files changed, 269 insertions(+), 99 deletions(-) diff --git a/xen/arch/arm/domain.c b/xen/arch/arm/domain.c index 6dc633ed50..f18fb35ac2 100644 --- a/xen/arch/arm/domain.c +++ b/xen/arch/arm/domain.c @@ -311,7 +311,7 @@ static void schedule_tail(struct vcpu *prev) local_irq_enable(); - context_saved(prev); + sched_context_switched(prev, current); if ( prev != current ) update_runstate_area(current); diff --git a/xen/arch/x86/domain.c b/xen/arch/x86/domain.c index 9eaa978ce5..1525ccd8e5 100644 --- a/xen/arch/x86/domain.c +++ b/xen/arch/x86/domain.c @@ -1714,7 +1714,6 @@ static void __context_switch(void) per_cpu(curr_vcpu, cpu) = n; } - void context_switch(struct vcpu *prev, struct vcpu *next) { unsigned int cpu = smp_processor_id(); @@ -1792,7 +1791,7 @@ void context_switch(struct vcpu *prev, struct vcpu *next) } } - context_saved(prev); + sched_context_switched(prev, next); if ( prev != next ) { diff --git a/xen/common/schedule.c b/xen/common/schedule.c index b5fb48c553..9d65586cae 100644 --- a/xen/common/schedule.c +++ b/xen/common/schedule.c @@ -54,6 +54,10 @@ boolean_param("sched_smt_power_savings", sched_smt_power_savings); * */ int sched_ratelimit_us = SCHED_DEFAULT_RATELIMIT_US; integer_param("sched_ratelimit_us", sched_ratelimit_us); + +/* Number of vcpus per struct sched_item. */ +static unsigned int sched_granularity = 1; + /* Various timer handlers. */ static void s_timer_fn(void *unused); static void vcpu_periodic_timer_fn(void *data); @@ -1564,134 +1568,291 @@ static void vcpu_periodic_timer_work(struct vcpu *v) set_timer(&v->periodic_timer, periodic_next_event); } -/* - * The main function - * - deschedule the current domain (scheduler independent). - * - pick a new domain (scheduler dependent). - */ -static void schedule(void) +static void sched_switch_items(struct sched_resource *sd, + struct sched_item *next, struct sched_item *prev, + s_time_t now) { - struct sched_item *prev = current->sched_item, *next = NULL; - s_time_t now; - struct scheduler *sched; - unsigned long *tasklet_work = &this_cpu(tasklet_work_to_do); - bool tasklet_work_scheduled = false; - struct sched_resource *sd; - spinlock_t *lock; - int cpu = smp_processor_id(); + sd->curr = next; - ASSERT_NOT_IN_ATOMIC(); + TRACE_3D(TRC_SCHED_SWITCH_INFPREV, prev->domain->domain_id, prev->item_id, + now - prev->state_entry_time); + TRACE_4D(TRC_SCHED_SWITCH_INFNEXT, next->domain->domain_id, next->item_id, + (next->vcpu->runstate.state == RUNSTATE_runnable) ? + (now - next->state_entry_time) : 0, prev->next_time); - SCHED_STAT_CRANK(sched_run); + ASSERT(prev->vcpu->runstate.state == RUNSTATE_running); - sd = this_cpu(sched_res); + TRACE_4D(TRC_SCHED_SWITCH, prev->domain->domain_id, prev->item_id, + next->domain->domain_id, next->item_id); - /* Update tasklet scheduling status. */ - switch ( *tasklet_work ) + sched_item_runstate_change(prev, false, now); + prev->last_run_time = now; + + ASSERT(next->vcpu->runstate.state != RUNSTATE_running); + sched_item_runstate_change(next, true, now); + + /* + * NB. Don't add any trace records from here until the actual context + * switch, else lost_records resume will not work properly. + */ + + ASSERT(!next->is_running); + next->vcpu->is_running = 1; + next->is_running = 1; +} + +static bool sched_tasklet_check(void) +{ + unsigned long *tasklet_work; + bool tasklet_work_scheduled = false; + const cpumask_t *mask = this_cpu(sched_res)->cpus; + int cpu; + + for_each_cpu ( cpu, mask ) { - case TASKLET_enqueued: - set_bit(_TASKLET_scheduled, tasklet_work); - /* fallthrough */ - case TASKLET_enqueued|TASKLET_scheduled: - tasklet_work_scheduled = true; - break; - case TASKLET_scheduled: - clear_bit(_TASKLET_scheduled, tasklet_work); - case 0: - /*tasklet_work_scheduled = false;*/ - break; - default: - BUG(); - } + tasklet_work = &per_cpu(tasklet_work_to_do, cpu); - lock = pcpu_schedule_lock_irq(cpu); + switch ( *tasklet_work ) + { + case TASKLET_enqueued: + set_bit(_TASKLET_scheduled, tasklet_work); + /* fallthrough */ + case TASKLET_enqueued|TASKLET_scheduled: + tasklet_work_scheduled = true; + break; + case TASKLET_scheduled: + clear_bit(_TASKLET_scheduled, tasklet_work); + case 0: + /*tasklet_work_scheduled = false;*/ + break; + default: + BUG(); + } + } - now = NOW(); + return tasklet_work_scheduled; +} - stop_timer(&sd->s_timer); +static struct sched_item *do_schedule(struct sched_item *prev, s_time_t now) +{ + struct scheduler *sched = this_cpu(scheduler); + struct sched_resource *sd = this_cpu(sched_res); + struct sched_item *next; /* get policy-specific decision on scheduling... */ - sched = this_cpu(scheduler); - sched->do_schedule(sched, prev, now, tasklet_work_scheduled); + sched->do_schedule(sched, prev, now, sched_tasklet_check()); next = prev->next_task; - sd->curr = next; - if ( prev->next_time >= 0 ) /* -ve means no limit */ set_timer(&sd->s_timer, now + prev->next_time); - if ( unlikely(prev == next) ) + if ( likely(prev != next) ) + sched_switch_items(sd, next, prev, now); + + return next; +} + +static void context_saved(struct vcpu *prev) +{ + struct sched_item *item = prev->sched_item; + + item->is_running = 0; + item->state_entry_time = NOW(); + + /* Check for migration request /after/ clearing running flag. */ + smp_mb(); + + sched_context_saved(vcpu_scheduler(prev), item); + + sched_item_migrate_finish(item); +} + +/* + * Rendezvous on end of context switch. + * As no lock is protecting this rendezvous function we need to use atomic + * access functions on the counter. + * The counter will be 0 in case no rendezvous is needed. For the rendezvous + * case it is initialised to the number of cpus to rendezvous plus 1. Each + * member entering decrements the counter. The last one will decrement it to + * 1 and perform the final needed action in that case (call of context_saved() + * if vcpu was switched), and then set the counter to zero. The other members + * will wait until the counter becomes zero until they proceed. + */ +void sched_context_switched(struct vcpu *vprev, struct vcpu *vnext) +{ + struct sched_item *next = vnext->sched_item; + + /* Clear running flag /after/ writing context to memory. */ + smp_wmb(); + + vprev->is_running = 0; + + if ( atomic_read(&next->rendezvous_out_cnt) ) + { + int cnt = atomic_dec_return(&next->rendezvous_out_cnt); + + /* Call context_saved() before releasing other waiters. */ + if ( cnt == 1 ) + { + if ( vprev != vnext ) + context_saved(vprev); + atomic_set(&next->rendezvous_out_cnt, 0); + } + else + while ( atomic_read(&next->rendezvous_out_cnt) ) + cpu_relax(); + } + else if ( vprev != vnext ) + context_saved(vprev); +} + +static void sched_context_switch(struct vcpu *vprev, struct vcpu *vnext, + s_time_t now) +{ + if ( unlikely(vprev == vnext) ) { - pcpu_schedule_unlock_irq(lock, cpu); TRACE_4D(TRC_SCHED_SWITCH_INFCONT, - next->domain->domain_id, next->item_id, - now - prev->state_entry_time, - prev->next_time); - trace_continue_running(next->vcpu); - return continue_running(prev->vcpu); + vnext->domain->domain_id, vnext->sched_item->item_id, + now - vprev->runstate.state_entry_time, + vprev->sched_item->next_time); + sched_context_switched(vprev, vnext); + trace_continue_running(vnext); + return continue_running(vprev); } - TRACE_3D(TRC_SCHED_SWITCH_INFPREV, - prev->domain->domain_id, prev->item_id, - now - prev->state_entry_time); - TRACE_4D(TRC_SCHED_SWITCH_INFNEXT, - next->domain->domain_id, next->item_id, - (next->vcpu->runstate.state == RUNSTATE_runnable) ? - (now - next->state_entry_time) : 0, - prev->next_time); + SCHED_STAT_CRANK(sched_ctx); - ASSERT(prev->vcpu->runstate.state == RUNSTATE_running); + stop_timer(&vprev->periodic_timer); - TRACE_4D(TRC_SCHED_SWITCH, - prev->domain->domain_id, prev->item_id, - next->domain->domain_id, next->item_id); + if ( vnext->sched_item->migrated ) + vcpu_move_irqs(vnext); - sched_item_runstate_change(prev, false, now); - prev->last_run_time = now; + vcpu_periodic_timer_work(vnext); - ASSERT(next->vcpu->runstate.state != RUNSTATE_running); - sched_item_runstate_change(next, true, now); + context_switch(vprev, vnext); +} - /* - * NB. Don't add any trace records from here until the actual context - * switch, else lost_records resume will not work properly. - */ +/* + * Rendezvous before taking a scheduling decision. + * Called with schedule lock held, so all accesses to the rendezvous counter + * can be normal ones (no atomic accesses needed). + * The counter is initialized to the number of cpus to rendezvous initially. + * Each cpu entering will decrement the counter. In case the counter becomes + * zero do_schedule() is called and the rendezvous counter for leaving + * context_switch() is set. All other members will wait until the counter is + * becoming zero, dropping the schedule lock in between. + */ +static struct sched_item *sched_wait_rendezvous_in(struct sched_item *prev, + spinlock_t *lock, int cpu, + s_time_t now) +{ + struct sched_item *next; - ASSERT(!next->is_running); - next->vcpu->is_running = 1; - next->is_running = 1; - next->state_entry_time = now; + if ( !--prev->rendezvous_in_cnt ) + { + next = do_schedule(prev, now); + atomic_set(&next->rendezvous_out_cnt, sched_granularity + 1); + return next; + } - pcpu_schedule_unlock_irq(lock, cpu); + while ( prev->rendezvous_in_cnt ) + { + pcpu_schedule_unlock_irq(lock, cpu); + cpu_relax(); + pcpu_schedule_lock_irq(cpu); + } - SCHED_STAT_CRANK(sched_ctx); + return prev->next_task; +} + +static void sched_slave(void) +{ + struct vcpu *vprev = current; + struct sched_item *prev = vprev->sched_item, *next; + s_time_t now; + spinlock_t *lock; + int cpu = smp_processor_id(); + + ASSERT_NOT_IN_ATOMIC(); + + lock = pcpu_schedule_lock_irq(cpu); + + now = NOW(); + + if ( !prev->rendezvous_in_cnt ) + { + pcpu_schedule_unlock_irq(lock, cpu); + return; + } - stop_timer(&prev->vcpu->periodic_timer); + stop_timer(&this_cpu(sched_res)->s_timer); - if ( next->migrated ) - vcpu_move_irqs(next->vcpu); + next = sched_wait_rendezvous_in(prev, lock, cpu, now); - vcpu_periodic_timer_work(next->vcpu); + pcpu_schedule_unlock_irq(lock, cpu); - context_switch(prev->vcpu, next->vcpu); + sched_context_switch(vprev, next->vcpu, now); } -void context_saved(struct vcpu *prev) +/* + * The main function + * - deschedule the current domain (scheduler independent). + * - pick a new domain (scheduler dependent). + */ +static void schedule(void) { - /* Clear running flag /after/ writing context to memory. */ - smp_wmb(); + struct vcpu *vnext, *vprev = current; + struct sched_item *prev = vprev->sched_item, *next = NULL; + s_time_t now; + struct sched_resource *sd; + spinlock_t *lock; + int cpu = smp_processor_id(); - prev->is_running = 0; - prev->sched_item->is_running = 0; - prev->sched_item->state_entry_time = NOW(); + ASSERT_NOT_IN_ATOMIC(); - /* Check for migration request /after/ clearing running flag. */ - smp_mb(); + SCHED_STAT_CRANK(sched_run); - sched_context_saved(vcpu_scheduler(prev), prev->sched_item); + sd = this_cpu(sched_res); + + lock = pcpu_schedule_lock_irq(cpu); + + if ( prev->rendezvous_in_cnt ) + { + /* + * We have a race: sched_slave() should be called, so raise a softirq + * in order to re-enter schedule() later and call sched_slave() now. + */ + pcpu_schedule_unlock_irq(lock, cpu); + + raise_softirq(SCHEDULE_SOFTIRQ); + return sched_slave(); + } + + now = NOW(); + + stop_timer(&sd->s_timer); + + if ( sched_granularity > 1 ) + { + cpumask_t mask; + + prev->rendezvous_in_cnt = sched_granularity; + cpumask_andnot(&mask, sd->cpus, cpumask_of(cpu)); + cpumask_raise_softirq(&mask, SCHED_SLAVE_SOFTIRQ); + next = sched_wait_rendezvous_in(prev, lock, cpu, now); + } + else + { + prev->rendezvous_in_cnt = 0; + next = do_schedule(prev, now); + atomic_set(&next->rendezvous_out_cnt, 0); + } + + pcpu_schedule_unlock_irq(lock, cpu); - sched_item_migrate_finish(prev->sched_item); + vnext = next->vcpu; + sched_context_switch(vprev, vnext, now); } /* The scheduler timer: force a run through the scheduler */ @@ -1733,6 +1894,7 @@ static int cpu_schedule_up(unsigned int cpu) if ( sd == NULL ) return -ENOMEM; sd->processor = cpu; + sd->cpus = cpumask_of(cpu); per_cpu(sched_res, cpu) = sd; per_cpu(scheduler, cpu) = &ops; @@ -1892,6 +2054,7 @@ void __init scheduler_init(void) int i; open_softirq(SCHEDULE_SOFTIRQ, schedule); + open_softirq(SCHED_SLAVE_SOFTIRQ, sched_slave); for ( i = 0; i < NUM_SCHEDULERS; i++) { diff --git a/xen/common/softirq.c b/xen/common/softirq.c index 83c3c09bd5..2d66193203 100644 --- a/xen/common/softirq.c +++ b/xen/common/softirq.c @@ -33,8 +33,8 @@ static void __do_softirq(unsigned long ignore_mask) for ( ; ; ) { /* - * Initialise @cpu on every iteration: SCHEDULE_SOFTIRQ may move - * us to another processor. + * Initialise @cpu on every iteration: SCHEDULE_SOFTIRQ or + * SCHED_SLAVE_SOFTIRQ may move us to another processor. */ cpu = smp_processor_id(); @@ -55,7 +55,7 @@ void process_pending_softirqs(void) { ASSERT(!in_irq() && local_irq_is_enabled()); /* Do not enter scheduler as it can preempt the calling context. */ - __do_softirq(1ul<curr) diff --git a/xen/include/xen/sched.h b/xen/include/xen/sched.h index 8bde790d27..5224f0aa70 100644 --- a/xen/include/xen/sched.h +++ b/xen/include/xen/sched.h @@ -295,6 +295,12 @@ struct sched_item { /* Next item to run. */ struct sched_item *next_task; s_time_t next_time; + + /* Number of vcpus not yet joined for context switch. */ + unsigned int rendezvous_in_cnt; + + /* Number of vcpus not yet finished with context switch. */ + atomic_t rendezvous_out_cnt; }; #define for_each_sched_item(d, e) \ @@ -698,10 +704,10 @@ void sync_local_execstate(void); /* * Called by the scheduler to switch to another VCPU. This function must - * call context_saved(@prev) when the local CPU is no longer running in - * @prev's context, and that context is saved to memory. Alternatively, if - * implementing lazy context switching, it suffices to ensure that invoking - * sync_vcpu_execstate() will switch and commit @prev's state. + * call sched_context_switched(@prev, @next) when the local CPU is no longer + * running in @prev's context, and that context is saved to memory. + * Alternatively, if implementing lazy context switching, it suffices to ensure + * that invoking sync_vcpu_execstate() will switch and commit @prev's state. */ void context_switch( struct vcpu *prev, @@ -713,7 +719,7 @@ void context_switch( * saved to memory. Alternatively, if implementing lazy context switching, * ensure that invoking sync_vcpu_execstate() will switch and commit @prev. */ -void context_saved(struct vcpu *prev); +void sched_context_switched(struct vcpu *prev, struct vcpu *vnext); /* Called by the scheduler to continue running the current VCPU. */ void continue_running( diff --git a/xen/include/xen/softirq.h b/xen/include/xen/softirq.h index c327c9b6cd..d7273b389b 100644 --- a/xen/include/xen/softirq.h +++ b/xen/include/xen/softirq.h @@ -4,6 +4,7 @@ /* Low-latency softirqs come first in the following list. */ enum { TIMER_SOFTIRQ = 0, + SCHED_SLAVE_SOFTIRQ, SCHEDULE_SOFTIRQ, NEW_TLBFLUSH_CLOCK_PERIOD_SOFTIRQ, RCU_SOFTIRQ,