Historically, a task that has been subjected to RCU priority boosting is deboosted at rcu_read_unlock() time. However, with the advent of deferred quiescent states, if the outermost rcu_read_unlock() was invoked with either bottom halves, interrupts, or preemption disabled, the deboosting will be delayed for some time. During this time, a low-priority process might be incorrectly running at a high real-time priority level. Fortunately, rcu_read_unlock_special() already provides mechanisms for forcing a minimal deferral of quiescent states, at least for kernels built with CONFIG_IRQ_WORK=y. These mechanisms are currently used when expedited grace periods are pending that might be blocked by the current task. This commit therefore causes those mechanisms to also be used in cases where the current task has been or might soon be subjected to RCU priority boosting. Note that this applies to all kernels built with CONFIG_RCU_BOOST=y, regardless of whether or not they are also built with CONFIG_PREEMPT_RT=y. This approach assumes that kernels build for use with aggressive real-time applications are built with CONFIG_IRQ_WORK=y. It is likely to be far simpler to enable CONFIG_IRQ_WORK=y than to implement a fast-deboosting scheme that works correctly in its absence. While in the area, alphabetize the rcu_preempt_deferred_qs_handler() function's local variables.
Showing the changes to kernel (1 of 1 file). Show all
Unified · Split
| @@ -598,9 +598,9 @@ static void rcu_preempt_deferred_qs_handler(struct irq_work *iwp) · 597 unmodified lines | |||
| 598 | static void rcu_read_unlock_special(struct task_struct *t) | 598 | static void rcu_read_unlock_special(struct task_struct *t) |
| 599 | { | 599 | { |
| 600 | unsigned long flags; | 600 | unsigned long flags; |
| 601 | bool irqs_were_disabled; | ||
| 601 | bool preempt_bh_were_disabled = | 602 | bool preempt_bh_were_disabled = |
| 602 | !!(preempt_count() & (PREEMPT_MASK | SOFTIRQ_MASK)); | 603 | !!(preempt_count() & (PREEMPT_MASK | SOFTIRQ_MASK)); |
| 603 | bool irqs_were_disabled; | ||
| 604 | 604 | ||
| 605 | /* NMI handlers cannot block and cannot safely manipulate state. */ | 605 | /* NMI handlers cannot block and cannot safely manipulate state. */ |
| 606 | if (in_nmi()) | 606 | if (in_nmi()) |
| @@ -609,30 +609,32 @@ static void rcu_read_unlock_special(struct task_struct *t) · 2 unmodified lines | |||
| 609 | local_irq_save(flags); | 609 | local_irq_save(flags); |
| 610 | irqs_were_disabled = irqs_disabled_flags(flags); | 610 | irqs_were_disabled = irqs_disabled_flags(flags); |
| 611 | if (preempt_bh_were_disabled || irqs_were_disabled) { | 611 | if (preempt_bh_were_disabled || irqs_were_disabled) { |
| 612 | bool exp; | 612 | bool expboost; // Expedited GP in flight or possible boosting. |
| 613 | struct rcu_data *rdp = this_cpu_ptr(&rcu_data); | 613 | struct rcu_data *rdp = this_cpu_ptr(&rcu_data); |
| 614 | struct rcu_node *rnp = rdp->mynode; | 614 | struct rcu_node *rnp = rdp->mynode; |
| 615 | 615 | ||
| 616 | exp = (t->rcu_blocked_node && | 616 | expboost = (t->rcu_blocked_node && READ_ONCE(t->rcu_blocked_node->exp_tasks)) || |
| 617 | READ_ONCE(t->rcu_blocked_node->exp_tasks)) || | 617 | (rdp->grpmask & READ_ONCE(rnp->expmask)) || |
| 618 | (rdp->grpmask & READ_ONCE(rnp->expmask)); | 618 | (IS_ENABLED(CONFIG_RCU_BOOST) && irqs_were_disabled && |
| 619 | t->rcu_blocked_node); | ||
| 619 | // Need to defer quiescent state until everything is enabled. | 620 | // Need to defer quiescent state until everything is enabled. |
| 620 | if (use_softirq && (in_irq() || (exp && !irqs_were_disabled))) { | 621 | if (use_softirq && (in_irq() || (expboost && !irqs_were_disabled))) { |
| 621 | // Using softirq, safe to awaken, and either the | 622 | // Using softirq, safe to awaken, and either the |
| 622 | // wakeup is free or there is an expedited GP. | 623 | // wakeup is free or there is either an expedited |
| 624 | // GP in flight or a potential need to deboost. | ||
| 623 | raise_softirq_irqoff(RCU_SOFTIRQ); | 625 | raise_softirq_irqoff(RCU_SOFTIRQ); |
| 624 | } else { | 626 | } else { |
| 625 | // Enabling BH or preempt does reschedule, so... | 627 | // Enabling BH or preempt does reschedule, so... |
| 626 | // Also if no expediting, slow is OK. | 628 | // Also if no expediting and no possible deboosting, |
| 627 | // Plus nohz_full CPUs eventually get tick enabled. | 629 | // slow is OK. Plus nohz_full CPUs eventually get |
| 630 | // tick enabled. | ||
| 628 | set_tsk_need_resched(current); | 631 | set_tsk_need_resched(current); |
| 629 | set_preempt_need_resched(); | 632 | set_preempt_need_resched(); |
| 630 | if (IS_ENABLED(CONFIG_IRQ_WORK) && irqs_were_disabled && | 633 | if (IS_ENABLED(CONFIG_IRQ_WORK) && irqs_were_disabled && |
| 631 | !rdp->defer_qs_iw_pending && exp && cpu_online(rdp->cpu)) { | 634 | expboost && !rdp->defer_qs_iw_pending && cpu_online(rdp->cpu)) { |
| 632 | // Get scheduler to re-evaluate and call hooks. | 635 | // Get scheduler to re-evaluate and call hooks. |
| 633 | // If !IRQ_WORK, FQS scan will eventually IPI. | 636 | // If !IRQ_WORK, FQS scan will eventually IPI. |
| 634 | init_irq_work(&rdp->defer_qs_iw, | 637 | init_irq_work(&rdp->defer_qs_iw, rcu_preempt_deferred_qs_handler); |
| 635 | rcu_preempt_deferred_qs_handler); | ||
| 636 | rdp->defer_qs_iw_pending = true; | 638 | rdp->defer_qs_iw_pending = true; |
| 637 | irq_work_queue_on(&rdp->defer_qs_iw, rdp->cpu); | 639 | irq_work_queue_on(&rdp->defer_qs_iw, rdp->cpu); |
| 638 | } | 640 | } |