From 94d44776737266eccafee32b985fe31fd5e021ca Mon Sep 17 00:00:00 2001 From: Jisheng Zhang Date: Wed, 22 Jun 2016 17:19:27 +0800 Subject: rcu: Use rcu_gp_kthread_wake() to wake up grace period kthreads Commit abedf8e2419f ("rcu: Use simple wait queues where possible in rcutree") converts Tree RCU's wait queues to simple wait queues, but it incorrectly reverts the commit 2aa792e6faf1 ("rcu: Use rcu_gp_kthread_wake() to wake up grace period kthreads"). This can result in redundant self-wakeups. This commit therefore replaces the simple wait-queue wakeups with rcu_gp_kthread_wake(), thus avoiding the redundant wakeups. Signed-off-by: Jisheng Zhang Signed-off-by: Paul E. McKenney --- kernel/rcu/tree.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) (limited to 'kernel/rcu/tree.c') diff --git a/kernel/rcu/tree.c b/kernel/rcu/tree.c index 5d80925e7fc8..cc1779a7ec5f 100644 --- a/kernel/rcu/tree.c +++ b/kernel/rcu/tree.c @@ -2344,7 +2344,7 @@ static void rcu_report_qs_rsp(struct rcu_state *rsp, unsigned long flags) WARN_ON_ONCE(!rcu_gp_in_progress(rsp)); WRITE_ONCE(rsp->gp_flags, READ_ONCE(rsp->gp_flags) | RCU_GP_FLAG_FQS); raw_spin_unlock_irqrestore_rcu_node(rcu_get_root(rsp), flags); - swake_up(&rsp->gp_wq); /* Memory barrier implied by swake_up() path. */ + rcu_gp_kthread_wake(rsp); } /* @@ -2970,7 +2970,7 @@ static void force_quiescent_state(struct rcu_state *rsp) } WRITE_ONCE(rsp->gp_flags, READ_ONCE(rsp->gp_flags) | RCU_GP_FLAG_FQS); raw_spin_unlock_irqrestore_rcu_node(rnp_old, flags); - swake_up(&rsp->gp_wq); /* Memory barrier implied by swake_up() path. */ + rcu_gp_kthread_wake(rsp); } /* -- cgit v1.2.3 From e77b7041258e11ba198951553d3acf1e371a9053 Mon Sep 17 00:00:00 2001 From: Paul Gortmaker Date: Fri, 15 Jul 2016 12:19:41 -0400 Subject: rcu: Don't use modular infrastructure in non-modular code The Kconfig currently controlling compilation of tree.c is: init/Kconfig:config TREE_RCU init/Kconfig: bool ...and update.c and sync.c are "obj-y" meaning that none are ever built as a module by anyone. Since MODULE_ALIAS is a no-op for non-modular code, we can remove them from these files. We leave moduleparam.h behind since the files instantiate some boot time configuration parameters with module_param() still. Cc: "Paul E. McKenney" Cc: Josh Triplett Cc: Steven Rostedt Cc: Mathieu Desnoyers Cc: Lai Jiangshan Signed-off-by: Paul Gortmaker Signed-off-by: Paul E. McKenney --- kernel/rcu/tree.c | 2 -- kernel/rcu/update.c | 3 +-- 2 files changed, 1 insertion(+), 4 deletions(-) (limited to 'kernel/rcu/tree.c') diff --git a/kernel/rcu/tree.c b/kernel/rcu/tree.c index cc1779a7ec5f..e83446062f65 100644 --- a/kernel/rcu/tree.c +++ b/kernel/rcu/tree.c @@ -41,7 +41,6 @@ #include #include #include -#include #include #include #include @@ -60,7 +59,6 @@ #include "tree.h" #include "rcu.h" -MODULE_ALIAS("rcutree"); #ifdef MODULE_PARAM_PREFIX #undef MODULE_PARAM_PREFIX #endif diff --git a/kernel/rcu/update.c b/kernel/rcu/update.c index f0d8322bc3ec..f19271dce0a9 100644 --- a/kernel/rcu/update.c +++ b/kernel/rcu/update.c @@ -46,7 +46,7 @@ #include #include #include -#include +#include #include #include @@ -54,7 +54,6 @@ #include "rcu.h" -MODULE_ALIAS("rcupdate"); #ifdef MODULE_PARAM_PREFIX #undef MODULE_PARAM_PREFIX #endif -- cgit v1.2.3 From 3563a438f124cb0b8cfd350c86de2f26c63d8837 Mon Sep 17 00:00:00 2001 From: "Paul E. McKenney" Date: Thu, 28 Jul 2016 09:39:11 -0700 Subject: rcu: Avoid redundant quiescent-state chasing Currently, __note_gp_changes() checks to see if the CPU has slept through multiple grace periods. If it has, it resynchronizes that CPU's view of the grace-period state, which includes whether or not the current grace period needs a quiescent state from this CPU. The fact of this need (or lack thereof) needs to be in two places, rdp->cpu_no_qs.b.norm and rdp->core_needs_qs. The former tells RCU's context-switch code to go get a quiescent state and the latter says that it needs to be reported. The current code unconditionally sets the former to true, but correctly sets the latter. This does not result in failures, but it does unnecessarily increase the amount of work done on average at context-switch time. This commit therefore correctly sets both fields. Signed-off-by: Paul E. McKenney --- kernel/rcu/tree.c | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-) (limited to 'kernel/rcu/tree.c') diff --git a/kernel/rcu/tree.c b/kernel/rcu/tree.c index e83446062f65..733902c33dd2 100644 --- a/kernel/rcu/tree.c +++ b/kernel/rcu/tree.c @@ -1846,6 +1846,7 @@ static bool __note_gp_changes(struct rcu_state *rsp, struct rcu_node *rnp, struct rcu_data *rdp) { bool ret; + bool need_gp; /* Handle the ends of any preceding grace periods first. */ if (rdp->completed == rnp->completed && @@ -1872,9 +1873,10 @@ static bool __note_gp_changes(struct rcu_state *rsp, struct rcu_node *rnp, */ rdp->gpnum = rnp->gpnum; trace_rcu_grace_period(rsp->name, rdp->gpnum, TPS("cpustart")); - rdp->cpu_no_qs.b.norm = true; + need_gp = !!(rnp->qsmask & rdp->grpmask); + rdp->cpu_no_qs.b.norm = need_gp; rdp->rcu_qs_ctr_snap = __this_cpu_read(rcu_qs_ctr); - rdp->core_needs_qs = !!(rnp->qsmask & rdp->grpmask); + rdp->core_needs_qs = need_gp; zero_cpu_stall_ticks(rdp); WRITE_ONCE(rdp->gpwrap, false); } -- cgit v1.2.3 From 7ec99de36f402618ae44147ac7fa9a07e4757a5f Mon Sep 17 00:00:00 2001 From: "Paul E. McKenney" Date: Thu, 30 Jun 2016 13:58:26 -0700 Subject: rcu: Provide exact CPU-online tracking for RCU Up to now, RCU has assumed that the CPU-online process makes it from CPU_UP_PREPARE to set_cpu_online() within one jiffy. Given the recent rise of virtualized environments, this assumption is very clearly obsolete. Failing to meet this deadline can result in RCU paying attention to an incoming CPU for one jiffy, then ignoring it until the grace period following the one in which that CPU sets itself online. This situation might prove to be fatally disappointing to any RCU read-side critical sections that had the misfortune to execute during the time in which RCU was ignoring the slow-to-come-online CPU. This commit therefore updates RCU's internal CPU state-tracking information at notify_cpu_starting() time, thus providing RCU with an exact transition of the CPU's state from offline to online. Note that this means that incoming CPUs must not use RCU read-side critical section (other than those of SRCU) until notify_cpu_starting() time. Note also that the CPU_STARTING notifiers -are- allowed to use RCU read-side critical sections. (Of course, CPU-hotplug notifiers are rapidly becoming obsolete, so you need to act fast!) If a given architecture or CPU family needs to use RCU read-side critical sections earlier, the call to rcu_cpu_starting() from notify_cpu_starting() will need to be architecture-specific, with architectures that need early use being required to hand-place the call to rcu_cpu_starting() at some point preceding the call to notify_cpu_starting(). Signed-off-by: Paul E. McKenney --- include/linux/rcupdate.h | 1 + kernel/cpu.c | 1 + kernel/rcu/tree.c | 32 +++++++++++++++++++++++++++++--- 3 files changed, 31 insertions(+), 3 deletions(-) (limited to 'kernel/rcu/tree.c') diff --git a/include/linux/rcupdate.h b/include/linux/rcupdate.h index 1aa62e1a761b..321f9ed552a9 100644 --- a/include/linux/rcupdate.h +++ b/include/linux/rcupdate.h @@ -334,6 +334,7 @@ void rcu_sched_qs(void); void rcu_bh_qs(void); void rcu_check_callbacks(int user); void rcu_report_dead(unsigned int cpu); +void rcu_cpu_starting(unsigned int cpu); #ifndef CONFIG_TINY_RCU void rcu_end_inkernel_boot(void); diff --git a/kernel/cpu.c b/kernel/cpu.c index 341bf80f80bd..9482ceb928e0 100644 --- a/kernel/cpu.c +++ b/kernel/cpu.c @@ -889,6 +889,7 @@ void notify_cpu_starting(unsigned int cpu) struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); enum cpuhp_state target = min((int)st->target, CPUHP_AP_ONLINE); + rcu_cpu_starting(cpu); /* All CPU_STARTING notifiers can use RCU. */ while (st->state < target) { struct cpuhp_step *step; diff --git a/kernel/rcu/tree.c b/kernel/rcu/tree.c index 5d80925e7fc8..d2973fb85e8c 100644 --- a/kernel/rcu/tree.c +++ b/kernel/rcu/tree.c @@ -3792,8 +3792,6 @@ rcu_init_percpu_data(int cpu, struct rcu_state *rsp) rnp = rdp->mynode; mask = rdp->grpmask; raw_spin_lock_rcu_node(rnp); /* irqs already disabled. */ - rnp->qsmaskinitnext |= mask; - rnp->expmaskinitnext |= mask; if (!rdp->beenonline) WRITE_ONCE(rsp->ncpus, READ_ONCE(rsp->ncpus) + 1); rdp->beenonline = true; /* We have now been online. */ @@ -3860,6 +3858,32 @@ int rcutree_dead_cpu(unsigned int cpu) return 0; } +/* + * Mark the specified CPU as being online so that subsequent grace periods + * (both expedited and normal) will wait on it. Note that this means that + * incoming CPUs are not allowed to use RCU read-side critical sections + * until this function is called. Failing to observe this restriction + * will result in lockdep splats. + */ +void rcu_cpu_starting(unsigned int cpu) +{ + unsigned long flags; + unsigned long mask; + struct rcu_data *rdp; + struct rcu_node *rnp; + struct rcu_state *rsp; + + for_each_rcu_flavor(rsp) { + rdp = this_cpu_ptr(rsp->rda); + rnp = rdp->mynode; + mask = rdp->grpmask; + raw_spin_lock_irqsave_rcu_node(rnp, flags); + rnp->qsmaskinitnext |= mask; + rnp->expmaskinitnext |= mask; + raw_spin_unlock_irqrestore_rcu_node(rnp, flags); + } +} + #ifdef CONFIG_HOTPLUG_CPU /* * The CPU is exiting the idle loop into the arch_cpu_idle_dead() @@ -4209,8 +4233,10 @@ void __init rcu_init(void) * or the scheduler are operational. */ pm_notifier(rcu_pm_notify, 0); - for_each_online_cpu(cpu) + for_each_online_cpu(cpu) { rcutree_prepare_cpu(cpu); + rcu_cpu_starting(cpu); + } } #include "tree_exp.h" -- cgit v1.2.3