* and inputs from Rusty Russell, Andrea Arcangeli and Andi Kleen.
*
* For detailed explanation of Read-Copy Update mechanism see -
- * Documentation/RCU
+ * Documentation/RCU
*/
#include <linux/types.h>
#include <linux/kernel.h>
#include "rcutree.h"
-#ifdef CONFIG_DEBUG_LOCK_ALLOC
-static struct lock_class_key rcu_lock_key;
-struct lockdep_map rcu_lock_map =
- STATIC_LOCKDEP_MAP_INIT("rcu_read_lock", &rcu_lock_key);
-EXPORT_SYMBOL_GPL(rcu_lock_map);
-#endif
-
/* Data structures. */
#define RCU_STATE_INITIALIZER(name) { \
.gpnum = -300, \
.completed = -300, \
.onofflock = __SPIN_LOCK_UNLOCKED(&name.onofflock), \
+ .orphan_cbs_list = NULL, \
+ .orphan_cbs_tail = &name.orphan_cbs_list, \
+ .orphan_qlen = 0, \
.fqslock = __SPIN_LOCK_UNLOCKED(&name.fqslock), \
.n_force_qs = 0, \
.n_force_qs_ngp = 0, \
struct rcu_state rcu_bh_state = RCU_STATE_INITIALIZER(rcu_bh_state);
DEFINE_PER_CPU(struct rcu_data, rcu_bh_data);
-extern long rcu_batches_completed_sched(void);
-static struct rcu_node *rcu_get_root(struct rcu_state *rsp);
-static void cpu_quiet_msk(unsigned long mask, struct rcu_state *rsp,
- struct rcu_node *rnp, unsigned long flags);
-static void cpu_quiet_msk_finish(struct rcu_state *rsp, unsigned long flags);
-#ifdef CONFIG_HOTPLUG_CPU
-static void __rcu_offline_cpu(int cpu, struct rcu_state *rsp);
-#endif /* #ifdef CONFIG_HOTPLUG_CPU */
-static void __rcu_process_callbacks(struct rcu_state *rsp,
- struct rcu_data *rdp);
-static void __call_rcu(struct rcu_head *head,
- void (*func)(struct rcu_head *rcu),
- struct rcu_state *rsp);
-static int __rcu_pending(struct rcu_state *rsp, struct rcu_data *rdp);
-static void __cpuinit rcu_init_percpu_data(int cpu, struct rcu_state *rsp,
- int preemptable);
-#include "rcutree_plugin.h"
+/*
+ * Return true if an RCU grace period is in progress. The ACCESS_ONCE()s
+ * permit this function to be invoked without holding the root rcu_node
+ * structure's ->lock, but of course results can be subject to change.
+ */
+static int rcu_gp_in_progress(struct rcu_state *rsp)
+{
+ return ACCESS_ONCE(rsp->completed) != ACCESS_ONCE(rsp->gpnum);
+}
/*
* Note a quiescent state. Because we do not need to know
static int qhimark = 10000; /* If this many pending, ignore blimit. */
static int qlowmark = 100; /* Once only this many pending, use blimit. */
+module_param(blimit, int, 0);
+module_param(qhimark, int, 0);
+module_param(qlowmark, int, 0);
+
static void force_quiescent_state(struct rcu_state *rsp, int relaxed);
static int rcu_pending(int cpu);
static int
cpu_needs_another_gp(struct rcu_state *rsp, struct rcu_data *rdp)
{
- /* ACCESS_ONCE() because we are accessing outside of lock. */
- return *rdp->nxttail[RCU_DONE_TAIL] &&
- ACCESS_ONCE(rsp->completed) == ACCESS_ONCE(rsp->gpnum);
+ return *rdp->nxttail[RCU_DONE_TAIL] && !rcu_gp_in_progress(rsp);
}
/*
/*
* Snapshot the specified CPU's dynticks counter so that we can later
* credit them with an implicit quiescent state. Return 1 if this CPU
- * is already in a quiescent state courtesy of dynticks idle mode.
+ * is in dynticks idle mode, which is an extended quiescent state.
*/
static int dyntick_save_progress_counter(struct rcu_data *rdp)
{
long delta;
unsigned long flags;
struct rcu_node *rnp = rcu_get_root(rsp);
- struct rcu_node *rnp_cur = rsp->level[NUM_RCU_LVLS - 1];
- struct rcu_node *rnp_end = &rsp->node[NUM_RCU_NODES];
/* Only let one CPU complain about others per time interval. */
spin_lock_irqsave(&rnp->lock, flags);
delta = jiffies - rsp->jiffies_stall;
- if (delta < RCU_STALL_RAT_DELAY || rsp->gpnum == rsp->completed) {
+ if (delta < RCU_STALL_RAT_DELAY || !rcu_gp_in_progress(rsp)) {
spin_unlock_irqrestore(&rnp->lock, flags);
return;
}
rsp->jiffies_stall = jiffies + RCU_SECONDS_TILL_STALL_RECHECK;
+
+ /*
+ * Now rat on any tasks that got kicked up to the root rcu_node
+ * due to CPU offlining.
+ */
+ rcu_print_task_stall(rnp);
spin_unlock_irqrestore(&rnp->lock, flags);
/* OK, time to rat on our buddy... */
printk(KERN_ERR "INFO: RCU detected CPU stalls:");
- for (; rnp_cur < rnp_end; rnp_cur++) {
+ rcu_for_each_leaf_node(rsp, rnp) {
rcu_print_task_stall(rnp);
- if (rnp_cur->qsmask == 0)
+ if (rnp->qsmask == 0)
continue;
- for (cpu = 0; cpu <= rnp_cur->grphi - rnp_cur->grplo; cpu++)
- if (rnp_cur->qsmask & (1UL << cpu))
- printk(" %d", rnp_cur->grplo + cpu);
+ for (cpu = 0; cpu <= rnp->grphi - rnp->grplo; cpu++)
+ if (rnp->qsmask & (1UL << cpu))
+ printk(" %d", rnp->grplo + cpu);
}
printk(" (detected by %d, t=%ld jiffies)\n",
smp_processor_id(), (long)(jiffies - rsp->gp_start));
/* We haven't checked in, so go dump stack. */
print_cpu_stall(rsp);
- } else if (rsp->gpnum != rsp->completed &&
- delta >= RCU_STALL_RAT_DELAY) {
+ } else if (rcu_gp_in_progress(rsp) && delta >= RCU_STALL_RAT_DELAY) {
/* They had two time units to dump stack, so complain. */
print_other_cpu_stall(rsp);
{
struct rcu_data *rdp = rsp->rda[smp_processor_id()];
struct rcu_node *rnp = rcu_get_root(rsp);
- struct rcu_node *rnp_cur;
- struct rcu_node *rnp_end;
if (!cpu_needs_another_gp(rsp, rdp)) {
spin_unlock_irqrestore(&rnp->lock, flags);
note_new_gpnum(rsp, rdp);
/*
- * Because we are first, we know that all our callbacks will
- * be covered by this upcoming grace period, even the ones
- * that were registered arbitrarily recently.
+ * Because this CPU just now started the new grace period, we know
+ * that all of its callbacks will be covered by this upcoming grace
+ * period, even the ones that were registered arbitrarily recently.
+ * Therefore, advance all outstanding callbacks to RCU_WAIT_TAIL.
+ *
+ * Other CPUs cannot be sure exactly when the grace period started.
+ * Therefore, their recently registered callbacks must pass through
+ * an additional RCU_NEXT_READY stage, so that they will be handled
+ * by the next RCU grace period.
*/
rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL];
rdp->nxttail[RCU_WAIT_TAIL] = rdp->nxttail[RCU_NEXT_TAIL];
* one corresponding to this CPU, due to the fact that we have
* irqs disabled.
*/
- rnp_end = &rsp->node[NUM_RCU_NODES];
- for (rnp_cur = &rsp->node[0]; rnp_cur < rnp_end; rnp_cur++) {
- spin_lock(&rnp_cur->lock); /* irqs already disabled. */
+ rcu_for_each_node_breadth_first(rsp, rnp) {
+ spin_lock(&rnp->lock); /* irqs already disabled. */
rcu_preempt_check_blocked_tasks(rnp);
- rnp_cur->qsmask = rnp_cur->qsmaskinit;
+ rnp->qsmask = rnp->qsmaskinit;
rnp->gpnum = rsp->gpnum;
- spin_unlock(&rnp_cur->lock); /* irqs already disabled. */
+ spin_unlock(&rnp->lock); /* irqs already disabled. */
}
rsp->signaled = RCU_SIGNAL_INIT; /* force_quiescent_state now OK. */
* hold rnp->lock, as required by rcu_start_gp(), which will release it.
*/
static void cpu_quiet_msk_finish(struct rcu_state *rsp, unsigned long flags)
- __releases(rnp->lock)
+ __releases(rcu_get_root(rsp)->lock)
{
- WARN_ON_ONCE(rsp->completed == rsp->gpnum);
+ WARN_ON_ONCE(!rcu_gp_in_progress(rsp));
rsp->completed = rsp->gpnum;
rcu_process_gp_end(rsp, rsp->rda[smp_processor_id()]);
rcu_start_gp(rsp, flags); /* releases root node's rnp->lock. */
#ifdef CONFIG_HOTPLUG_CPU
/*
+ * Move a dying CPU's RCU callbacks to the ->orphan_cbs_list for the
+ * specified flavor of RCU. The callbacks will be adopted by the next
+ * _rcu_barrier() invocation or by the CPU_DEAD notifier, whichever
+ * comes first. Because this is invoked from the CPU_DYING notifier,
+ * irqs are already disabled.
+ */
+static void rcu_send_cbs_to_orphanage(struct rcu_state *rsp)
+{
+ int i;
+ struct rcu_data *rdp = rsp->rda[smp_processor_id()];
+
+ if (rdp->nxtlist == NULL)
+ return; /* irqs disabled, so comparison is stable. */
+ spin_lock(&rsp->onofflock); /* irqs already disabled. */
+ *rsp->orphan_cbs_tail = rdp->nxtlist;
+ rsp->orphan_cbs_tail = rdp->nxttail[RCU_NEXT_TAIL];
+ rdp->nxtlist = NULL;
+ for (i = 0; i < RCU_NEXT_SIZE; i++)
+ rdp->nxttail[i] = &rdp->nxtlist;
+ rsp->orphan_qlen += rdp->qlen;
+ rdp->qlen = 0;
+ spin_unlock(&rsp->onofflock); /* irqs remain disabled. */
+}
+
+/*
+ * Adopt previously orphaned RCU callbacks.
+ */
+static void rcu_adopt_orphan_cbs(struct rcu_state *rsp)
+{
+ unsigned long flags;
+ struct rcu_data *rdp;
+
+ spin_lock_irqsave(&rsp->onofflock, flags);
+ rdp = rsp->rda[smp_processor_id()];
+ if (rsp->orphan_cbs_list == NULL) {
+ spin_unlock_irqrestore(&rsp->onofflock, flags);
+ return;
+ }
+ *rdp->nxttail[RCU_NEXT_TAIL] = rsp->orphan_cbs_list;
+ rdp->nxttail[RCU_NEXT_TAIL] = rsp->orphan_cbs_tail;
+ rdp->qlen += rsp->orphan_qlen;
+ rsp->orphan_cbs_list = NULL;
+ rsp->orphan_cbs_tail = &rsp->orphan_cbs_list;
+ rsp->orphan_qlen = 0;
+ spin_unlock_irqrestore(&rsp->onofflock, flags);
+}
+
+/*
* Remove the outgoing CPU from the bitmasks in the rcu_node hierarchy
* and move all callbacks from the outgoing CPU to the current one.
*/
static void __rcu_offline_cpu(int cpu, struct rcu_state *rsp)
{
- int i;
unsigned long flags;
long lastcomp;
unsigned long mask;
struct rcu_data *rdp = rsp->rda[cpu];
- struct rcu_data *rdp_me;
struct rcu_node *rnp;
/* Exclude any attempts to start a new grace period. */
} while (rnp != NULL);
lastcomp = rsp->completed;
- spin_unlock(&rsp->onofflock); /* irqs remain disabled. */
+ spin_unlock_irqrestore(&rsp->onofflock, flags);
- /*
- * Move callbacks from the outgoing CPU to the running CPU.
- * Note that the outgoing CPU is now quiscent, so it is now
- * (uncharacteristically) safe to access its rcu_data structure.
- * Note also that we must carefully retain the order of the
- * outgoing CPU's callbacks in order for rcu_barrier() to work
- * correctly. Finally, note that we start all the callbacks
- * afresh, even those that have passed through a grace period
- * and are therefore ready to invoke. The theory is that hotplug
- * events are rare, and that if they are frequent enough to
- * indefinitely delay callbacks, you have far worse things to
- * be worrying about.
- */
- rdp_me = rsp->rda[smp_processor_id()];
- if (rdp->nxtlist != NULL) {
- *rdp_me->nxttail[RCU_NEXT_TAIL] = rdp->nxtlist;
- rdp_me->nxttail[RCU_NEXT_TAIL] = rdp->nxttail[RCU_NEXT_TAIL];
- rdp->nxtlist = NULL;
- for (i = 0; i < RCU_NEXT_SIZE; i++)
- rdp->nxttail[i] = &rdp->nxtlist;
- rdp_me->qlen += rdp->qlen;
- rdp->qlen = 0;
- }
- local_irq_restore(flags);
+ rcu_adopt_orphan_cbs(rsp);
}
/*
#else /* #ifdef CONFIG_HOTPLUG_CPU */
+static void rcu_send_cbs_to_orphanage(struct rcu_state *rsp)
+{
+}
+
+static void rcu_adopt_orphan_cbs(struct rcu_state *rsp)
+{
+}
+
static void rcu_offline_cpu(int cpu)
{
}
int cpu;
unsigned long flags;
unsigned long mask;
- struct rcu_node *rnp_cur = rsp->level[NUM_RCU_LVLS - 1];
- struct rcu_node *rnp_end = &rsp->node[NUM_RCU_NODES];
+ struct rcu_node *rnp;
- for (; rnp_cur < rnp_end; rnp_cur++) {
+ rcu_for_each_leaf_node(rsp, rnp) {
mask = 0;
- spin_lock_irqsave(&rnp_cur->lock, flags);
+ spin_lock_irqsave(&rnp->lock, flags);
if (rsp->completed != lastcomp) {
- spin_unlock_irqrestore(&rnp_cur->lock, flags);
+ spin_unlock_irqrestore(&rnp->lock, flags);
return 1;
}
- if (rnp_cur->qsmask == 0) {
- spin_unlock_irqrestore(&rnp_cur->lock, flags);
+ if (rnp->qsmask == 0) {
+ spin_unlock_irqrestore(&rnp->lock, flags);
continue;
}
- cpu = rnp_cur->grplo;
+ cpu = rnp->grplo;
bit = 1;
- for (; cpu <= rnp_cur->grphi; cpu++, bit <<= 1) {
- if ((rnp_cur->qsmask & bit) != 0 && f(rsp->rda[cpu]))
+ for (; cpu <= rnp->grphi; cpu++, bit <<= 1) {
+ if ((rnp->qsmask & bit) != 0 && f(rsp->rda[cpu]))
mask |= bit;
}
if (mask != 0 && rsp->completed == lastcomp) {
- /* cpu_quiet_msk() releases rnp_cur->lock. */
- cpu_quiet_msk(mask, rsp, rnp_cur, flags);
+ /* cpu_quiet_msk() releases rnp->lock. */
+ cpu_quiet_msk(mask, rsp, rnp, flags);
continue;
}
- spin_unlock_irqrestore(&rnp_cur->lock, flags);
+ spin_unlock_irqrestore(&rnp->lock, flags);
}
return 0;
}
struct rcu_node *rnp = rcu_get_root(rsp);
u8 signaled;
- if (ACCESS_ONCE(rsp->completed) == ACCESS_ONCE(rsp->gpnum))
+ if (!rcu_gp_in_progress(rsp))
return; /* No grace period in progress, nothing to force. */
if (!spin_trylock_irqsave(&rsp->fqslock, flags)) {
rsp->n_force_qs_lh++; /* Inexact, can lose counts. Tough! */
rdp->nxttail[RCU_NEXT_TAIL] = &head->next;
/* Start a new grace period if one not already started. */
- if (ACCESS_ONCE(rsp->completed) == ACCESS_ONCE(rsp->gpnum)) {
+ if (!rcu_gp_in_progress(rsp)) {
unsigned long nestflag;
struct rcu_node *rnp_root = rcu_get_root(rsp);
}
/* Has an RCU GP gone long enough to send resched IPIs &c? */
- if (ACCESS_ONCE(rsp->completed) != ACCESS_ONCE(rsp->gpnum) &&
+ if (rcu_gp_in_progress(rsp) &&
((long)(ACCESS_ONCE(rsp->jiffies_force_qs) - jiffies) < 0)) {
rdp->n_rp_need_fqs++;
return 1;
rcu_preempt_needs_cpu(cpu);
}
+static DEFINE_PER_CPU(struct rcu_head, rcu_barrier_head) = {NULL};
+static atomic_t rcu_barrier_cpu_count;
+static DEFINE_MUTEX(rcu_barrier_mutex);
+static struct completion rcu_barrier_completion;
+
+static void rcu_barrier_callback(struct rcu_head *notused)
+{
+ if (atomic_dec_and_test(&rcu_barrier_cpu_count))
+ complete(&rcu_barrier_completion);
+}
+
+/*
+ * Called with preemption disabled, and from cross-cpu IRQ context.
+ */
+static void rcu_barrier_func(void *type)
+{
+ int cpu = smp_processor_id();
+ struct rcu_head *head = &per_cpu(rcu_barrier_head, cpu);
+ void (*call_rcu_func)(struct rcu_head *head,
+ void (*func)(struct rcu_head *head));
+
+ atomic_inc(&rcu_barrier_cpu_count);
+ call_rcu_func = type;
+ call_rcu_func(head, rcu_barrier_callback);
+}
+
+/*
+ * Orchestrate the specified type of RCU barrier, waiting for all
+ * RCU callbacks of the specified type to complete.
+ */
+static void _rcu_barrier(struct rcu_state *rsp,
+ void (*call_rcu_func)(struct rcu_head *head,
+ void (*func)(struct rcu_head *head)))
+{
+ BUG_ON(in_interrupt());
+ /* Take mutex to serialize concurrent rcu_barrier() requests. */
+ mutex_lock(&rcu_barrier_mutex);
+ init_completion(&rcu_barrier_completion);
+ /*
+ * Initialize rcu_barrier_cpu_count to 1, then invoke
+ * rcu_barrier_func() on each CPU, so that each CPU also has
+ * incremented rcu_barrier_cpu_count. Only then is it safe to
+ * decrement rcu_barrier_cpu_count -- otherwise the first CPU
+ * might complete its grace period before all of the other CPUs
+ * did their increment, causing this function to return too
+ * early.
+ */
+ atomic_set(&rcu_barrier_cpu_count, 1);
+ preempt_disable(); /* stop CPU_DYING from filling orphan_cbs_list */
+ rcu_adopt_orphan_cbs(rsp);
+ on_each_cpu(rcu_barrier_func, (void *)call_rcu_func, 1);
+ preempt_enable(); /* CPU_DYING can again fill orphan_cbs_list */
+ if (atomic_dec_and_test(&rcu_barrier_cpu_count))
+ complete(&rcu_barrier_completion);
+ wait_for_completion(&rcu_barrier_completion);
+ mutex_unlock(&rcu_barrier_mutex);
+}
+
+/**
+ * rcu_barrier_bh - Wait until all in-flight call_rcu_bh() callbacks complete.
+ */
+void rcu_barrier_bh(void)
+{
+ _rcu_barrier(&rcu_bh_state, call_rcu_bh);
+}
+EXPORT_SYMBOL_GPL(rcu_barrier_bh);
+
+/**
+ * rcu_barrier_sched - Wait for in-flight call_rcu_sched() callbacks.
+ */
+void rcu_barrier_sched(void)
+{
+ _rcu_barrier(&rcu_sched_state, call_rcu_sched);
+}
+EXPORT_SYMBOL_GPL(rcu_barrier_sched);
+
/*
* Do boot-time initialization of a CPU's per-CPU RCU data.
*/
case CPU_UP_PREPARE_FROZEN:
rcu_online_cpu(cpu);
break;
+ case CPU_DYING:
+ case CPU_DYING_FROZEN:
+ /*
+ * preempt_disable() in _rcu_barrier() prevents stop_machine(),
+ * so when "on_each_cpu(rcu_barrier_func, (void *)type, 1);"
+ * returns, all online cpus have queued rcu_barrier_func().
+ * The dying CPU clears its cpu_online_mask bit and
+ * moves all of its RCU callbacks to ->orphan_cbs_list
+ * in the context of stop_machine(), so subsequent calls
+ * to _rcu_barrier() will adopt these callbacks and only
+ * then queue rcu_barrier_func() on all remaining CPUs.
+ */
+ rcu_send_cbs_to_orphanage(&rcu_bh_state);
+ rcu_send_cbs_to_orphanage(&rcu_sched_state);
+ rcu_preempt_send_cbs_to_orphanage();
+ break;
case CPU_DEAD:
case CPU_DEAD_FROZEN:
case CPU_UP_CANCELED:
cpustride *= rsp->levelspread[i];
rnp = rsp->level[i];
for (j = 0; j < rsp->levelcnt[i]; j++, rnp++) {
- spin_lock_init(&rnp->lock);
+ if (rnp != rcu_get_root(rsp))
+ spin_lock_init(&rnp->lock);
rnp->gpnum = 0;
rnp->qsmask = 0;
rnp->qsmaskinit = 0;
INIT_LIST_HEAD(&rnp->blocked_tasks[1]);
}
}
+ spin_lock_init(&rcu_get_root(rsp)->lock);
}
/*
*/
#define RCU_INIT_FLAVOR(rsp, rcu_data) \
do { \
+ int i; \
+ int j; \
+ struct rcu_node *rnp; \
+ \
rcu_init_one(rsp); \
rnp = (rsp)->level[NUM_RCU_LVLS - 1]; \
j = 0; \
} \
} while (0)
-#ifdef CONFIG_TREE_PREEMPT_RCU
-
-void __init __rcu_init_preempt(void)
-{
- int i; /* All used by RCU_INIT_FLAVOR(). */
- int j;
- struct rcu_node *rnp;
-
- RCU_INIT_FLAVOR(&rcu_preempt_state, rcu_preempt_data);
-}
-
-#else /* #ifdef CONFIG_TREE_PREEMPT_RCU */
-
-void __init __rcu_init_preempt(void)
-{
-}
-
-#endif /* #else #ifdef CONFIG_TREE_PREEMPT_RCU */
-
void __init __rcu_init(void)
{
- int i; /* All used by RCU_INIT_FLAVOR(). */
- int j;
- struct rcu_node *rnp;
-
rcu_bootup_announce();
#ifdef CONFIG_RCU_CPU_STALL_DETECTOR
printk(KERN_INFO "RCU-based detection of stalled CPUs is enabled.\n");
open_softirq(RCU_SOFTIRQ, rcu_process_callbacks);
}
-module_param(blimit, int, 0);
-module_param(qhimark, int, 0);
-module_param(qlowmark, int, 0);
+#include "rcutree_plugin.h"
git://ftp.safe.ca / safe / jmp / linux-2.6 / blobdiff