rcu: Add grace-period, quiescent-state, and call_rcu trace events
Paul E. McKenney [Sat, 25 Jun 2011 13:36:56 +0000 (06:36 -0700)]
Add trace events to record grace-period start and end, quiescent states,
CPUs noticing grace-period start and end, grace-period initialization,
call_rcu() invocation, tasks blocking in RCU read-side critical sections,
tasks exiting those same critical sections, force_quiescent_state()
detection of dyntick-idle and offline CPUs, CPUs entering and leaving
dyntick-idle mode (except from NMIs), CPUs coming online and going
offline, and CPUs being kicked for staying in dyntick-idle mode for too
long (as in many weeks, even on 32-bit systems).

Signed-off-by: Paul E. McKenney <paul.mckenney@linaro.org>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>

rcu: Add the rcu flavor to callback trace events

The earlier trace events for registering RCU callbacks and for invoking
them did not include the RCU flavor (rcu_bh, rcu_preempt, or rcu_sched).
This commit adds the RCU flavor to those trace events.

Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>

include/trace/events/rcu.h
kernel/rcu.h
kernel/rcutiny.c
kernel/rcutree.c
kernel/rcutree.h
kernel/rcutree_plugin.h

index ac52aba..669fbd6 100644 (file)
@@ -24,7 +24,7 @@ TRACE_EVENT(rcu_utilization,
        TP_ARGS(s),
 
        TP_STRUCT__entry(
-               __field(char *, s)
+               __field(char *, s)
        ),
 
        TP_fast_assign(
@@ -34,6 +34,297 @@ TRACE_EVENT(rcu_utilization,
        TP_printk("%s", __entry->s)
 );
 
+#ifdef CONFIG_RCU_TRACE
+
+#if defined(CONFIG_TREE_RCU) || defined(CONFIG_TREE_PREEMPT_RCU)
+
+/*
+ * Tracepoint for grace-period events: starting and ending a grace
+ * period ("start" and "end", respectively), a CPU noting the start
+ * of a new grace period or the end of an old grace period ("cpustart"
+ * and "cpuend", respectively), a CPU passing through a quiescent
+ * state ("cpuqs"), a CPU coming online or going offline ("cpuonl"
+ * and "cpuofl", respectively), and a CPU being kicked for being too
+ * long in dyntick-idle mode ("kick").
+ */
+TRACE_EVENT(rcu_grace_period,
+
+       TP_PROTO(char *rcuname, unsigned long gpnum, char *gpevent),
+
+       TP_ARGS(rcuname, gpnum, gpevent),
+
+       TP_STRUCT__entry(
+               __field(char *, rcuname)
+               __field(unsigned long, gpnum)
+               __field(char *, gpevent)
+       ),
+
+       TP_fast_assign(
+               __entry->rcuname = rcuname;
+               __entry->gpnum = gpnum;
+               __entry->gpevent = gpevent;
+       ),
+
+       TP_printk("%s %lu %s",
+                 __entry->rcuname, __entry->gpnum, __entry->gpevent)
+);
+
+/*
+ * Tracepoint for grace-period-initialization events.  These are
+ * distinguished by the type of RCU, the new grace-period number, the
+ * rcu_node structure level, the starting and ending CPU covered by the
+ * rcu_node structure, and the mask of CPUs that will be waited for.
+ * All but the type of RCU are extracted from the rcu_node structure.
+ */
+TRACE_EVENT(rcu_grace_period_init,
+
+       TP_PROTO(char *rcuname, unsigned long gpnum, u8 level,
+                int grplo, int grphi, unsigned long qsmask),
+
+       TP_ARGS(rcuname, gpnum, level, grplo, grphi, qsmask),
+
+       TP_STRUCT__entry(
+               __field(char *, rcuname)
+               __field(unsigned long, gpnum)
+               __field(u8, level)
+               __field(int, grplo)
+               __field(int, grphi)
+               __field(unsigned long, qsmask)
+       ),
+
+       TP_fast_assign(
+               __entry->rcuname = rcuname;
+               __entry->gpnum = gpnum;
+               __entry->level = level;
+               __entry->grplo = grplo;
+               __entry->grphi = grphi;
+               __entry->qsmask = qsmask;
+       ),
+
+       TP_printk("%s %lu %u %d %d %lx",
+                 __entry->rcuname, __entry->gpnum, __entry->level,
+                 __entry->grplo, __entry->grphi, __entry->qsmask)
+);
+
+/*
+ * Tracepoint for tasks blocking within preemptible-RCU read-side
+ * critical sections.  Track the type of RCU (which one day might
+ * include SRCU), the grace-period number that the task is blocking
+ * (the current or the next), and the task's PID.
+ */
+TRACE_EVENT(rcu_preempt_task,
+
+       TP_PROTO(char *rcuname, int pid, unsigned long gpnum),
+
+       TP_ARGS(rcuname, pid, gpnum),
+
+       TP_STRUCT__entry(
+               __field(char *, rcuname)
+               __field(unsigned long, gpnum)
+               __field(int, pid)
+       ),
+
+       TP_fast_assign(
+               __entry->rcuname = rcuname;
+               __entry->gpnum = gpnum;
+               __entry->pid = pid;
+       ),
+
+       TP_printk("%s %lu %d",
+                 __entry->rcuname, __entry->gpnum, __entry->pid)
+);
+
+/*
+ * Tracepoint for tasks that blocked within a given preemptible-RCU
+ * read-side critical section exiting that critical section.  Track the
+ * type of RCU (which one day might include SRCU) and the task's PID.
+ */
+TRACE_EVENT(rcu_unlock_preempted_task,
+
+       TP_PROTO(char *rcuname, unsigned long gpnum, int pid),
+
+       TP_ARGS(rcuname, gpnum, pid),
+
+       TP_STRUCT__entry(
+               __field(char *, rcuname)
+               __field(unsigned long, gpnum)
+               __field(int, pid)
+       ),
+
+       TP_fast_assign(
+               __entry->rcuname = rcuname;
+               __entry->gpnum = gpnum;
+               __entry->pid = pid;
+       ),
+
+       TP_printk("%s %lu %d", __entry->rcuname, __entry->gpnum, __entry->pid)
+);
+
+/*
+ * Tracepoint for quiescent-state-reporting events.  These are
+ * distinguished by the type of RCU, the grace-period number, the
+ * mask of quiescent lower-level entities, the rcu_node structure level,
+ * the starting and ending CPU covered by the rcu_node structure, and
+ * whether there are any blocked tasks blocking the current grace period.
+ * All but the type of RCU are extracted from the rcu_node structure.
+ */
+TRACE_EVENT(rcu_quiescent_state_report,
+
+       TP_PROTO(char *rcuname, unsigned long gpnum,
+                unsigned long mask, unsigned long qsmask,
+                u8 level, int grplo, int grphi, int gp_tasks),
+
+       TP_ARGS(rcuname, gpnum, mask, qsmask, level, grplo, grphi, gp_tasks),
+
+       TP_STRUCT__entry(
+               __field(char *, rcuname)
+               __field(unsigned long, gpnum)
+               __field(unsigned long, mask)
+               __field(unsigned long, qsmask)
+               __field(u8, level)
+               __field(int, grplo)
+               __field(int, grphi)
+               __field(u8, gp_tasks)
+       ),
+
+       TP_fast_assign(
+               __entry->rcuname = rcuname;
+               __entry->gpnum = gpnum;
+               __entry->mask = mask;
+               __entry->qsmask = qsmask;
+               __entry->level = level;
+               __entry->grplo = grplo;
+               __entry->grphi = grphi;
+               __entry->gp_tasks = gp_tasks;
+       ),
+
+       TP_printk("%s %lu %lx>%lx %u %d %d %u",
+                 __entry->rcuname, __entry->gpnum,
+                 __entry->mask, __entry->qsmask, __entry->level,
+                 __entry->grplo, __entry->grphi, __entry->gp_tasks)
+);
+
+/*
+ * Tracepoint for quiescent states detected by force_quiescent_state().
+ * These trace events include the type of RCU, the grace-period number
+ * that was blocked by the CPU, the CPU itself, and the type of quiescent
+ * state, which can be "dti" for dyntick-idle mode, "ofl" for CPU offline,
+ * or "kick" when kicking a CPU that has been in dyntick-idle mode for
+ * too long.
+ */
+TRACE_EVENT(rcu_fqs,
+
+       TP_PROTO(char *rcuname, unsigned long gpnum, int cpu, char *qsevent),
+
+       TP_ARGS(rcuname, gpnum, cpu, qsevent),
+
+       TP_STRUCT__entry(
+               __field(char *, rcuname)
+               __field(unsigned long, gpnum)
+               __field(int, cpu)
+               __field(char *, qsevent)
+       ),
+
+       TP_fast_assign(
+               __entry->rcuname = rcuname;
+               __entry->gpnum = gpnum;
+               __entry->cpu = cpu;
+               __entry->qsevent = qsevent;
+       ),
+
+       TP_printk("%s %lu %d %s",
+                 __entry->rcuname, __entry->gpnum,
+                 __entry->cpu, __entry->qsevent)
+);
+
+#endif /* #if defined(CONFIG_TREE_RCU) || defined(CONFIG_TREE_PREEMPT_RCU) */
+
+/*
+ * Tracepoint for dyntick-idle entry/exit events.  These take a string
+ * as argument: "Start" for entering dyntick-idle mode and "End" for
+ * leaving it.
+ */
+TRACE_EVENT(rcu_dyntick,
+
+       TP_PROTO(char *polarity),
+
+       TP_ARGS(polarity),
+
+       TP_STRUCT__entry(
+               __field(char *, polarity)
+       ),
+
+       TP_fast_assign(
+               __entry->polarity = polarity;
+       ),
+
+       TP_printk("%s", __entry->polarity)
+);
+
+/*
+ * Tracepoint for the registration of a single RCU callback function.
+ * The first argument is the type of RCU, the second argument is
+ * a pointer to the RCU callback itself, and the third element is the
+ * new RCU callback queue length for the current CPU.
+ */
+TRACE_EVENT(rcu_callback,
+
+       TP_PROTO(char *rcuname, struct rcu_head *rhp, long qlen),
+
+       TP_ARGS(rcuname, rhp, qlen),
+
+       TP_STRUCT__entry(
+               __field(char *, rcuname)
+               __field(void *, rhp)
+               __field(void *, func)
+               __field(long, qlen)
+       ),
+
+       TP_fast_assign(
+               __entry->rcuname = rcuname;
+               __entry->rhp = rhp;
+               __entry->func = rhp->func;
+               __entry->qlen = qlen;
+       ),
+
+       TP_printk("%s rhp=%p func=%pf %ld",
+                 __entry->rcuname, __entry->rhp, __entry->func, __entry->qlen)
+);
+
+/*
+ * Tracepoint for the registration of a single RCU callback of the special
+ * kfree() form.  The first argument is the RCU type, the second argument
+ * is a pointer to the RCU callback, the third argument is the offset
+ * of the callback within the enclosing RCU-protected data structure,
+ * and the fourth argument is the new RCU callback queue length for the
+ * current CPU.
+ */
+TRACE_EVENT(rcu_kfree_callback,
+
+       TP_PROTO(char *rcuname, struct rcu_head *rhp, unsigned long offset,
+                long qlen),
+
+       TP_ARGS(rcuname, rhp, offset, qlen),
+
+       TP_STRUCT__entry(
+               __field(char *, rcuname)
+               __field(void *, rhp)
+               __field(unsigned long, offset)
+               __field(long, qlen)
+       ),
+
+       TP_fast_assign(
+               __entry->rcuname = rcuname;
+               __entry->rhp = rhp;
+               __entry->offset = offset;
+               __entry->qlen = qlen;
+       ),
+
+       TP_printk("%s rhp=%p func=%ld %ld",
+                 __entry->rcuname, __entry->rhp, __entry->offset,
+                 __entry->qlen)
+);
+
 /*
  * Tracepoint for marking the beginning rcu_do_batch, performed to start
  * RCU callback invocation.  The first argument is the RCU flavor,
@@ -65,50 +356,58 @@ TRACE_EVENT(rcu_batch_start,
 
 /*
  * Tracepoint for the invocation of a single RCU callback function.
- * The argument is a pointer to the RCU callback itself.
+ * The first argument is the type of RCU, and the second argument is
+ * a pointer to the RCU callback itself.
  */
 TRACE_EVENT(rcu_invoke_callback,
 
-       TP_PROTO(struct rcu_head *rhp),
+       TP_PROTO(char *rcuname, struct rcu_head *rhp),
 
-       TP_ARGS(rhp),
+       TP_ARGS(rcuname, rhp),
 
        TP_STRUCT__entry(
-               __field(void *, rhp)
-               __field(void *, func)
+               __field(char *, rcuname)
+               __field(void *, rhp)
+               __field(void *, func)
        ),
 
        TP_fast_assign(
+               __entry->rcuname = rcuname;
                __entry->rhp = rhp;
                __entry->func = rhp->func;
        ),
 
-       TP_printk("rhp=%p func=%pf", __entry->rhp, __entry->func)
+       TP_printk("%s rhp=%p func=%pf",
+                 __entry->rcuname, __entry->rhp, __entry->func)
 );
 
 /*
  * Tracepoint for the invocation of a single RCU callback of the special
- * kfree() form.  The first argument is a pointer to the RCU callback
- * and the second argument is the offset of the callback within the
- * enclosing RCU-protected data structure.
+ * kfree() form.  The first argument is the RCU flavor, the second
+ * argument is a pointer to the RCU callback, and the third argument
+ * is the offset of the callback within the enclosing RCU-protected
+ * data structure.
  */
 TRACE_EVENT(rcu_invoke_kfree_callback,
 
-       TP_PROTO(struct rcu_head *rhp, unsigned long offset),
+       TP_PROTO(char *rcuname, struct rcu_head *rhp, unsigned long offset),
 
-       TP_ARGS(rhp, offset),
+       TP_ARGS(rcuname, rhp, offset),
 
        TP_STRUCT__entry(
-               __field(void *, rhp)
+               __field(char *, rcuname)
+               __field(void *, rhp)
                __field(unsigned long, offset)
        ),
 
        TP_fast_assign(
+               __entry->rcuname = rcuname;
                __entry->rhp = rhp;
                __entry->offset = offset;
        ),
 
-       TP_printk("rhp=%p func=%ld", __entry->rhp, __entry->offset)
+       TP_printk("%s rhp=%p func=%ld",
+                 __entry->rcuname, __entry->rhp, __entry->offset)
 );
 
 /*
@@ -136,6 +435,24 @@ TRACE_EVENT(rcu_batch_end,
                  __entry->rcuname, __entry->callbacks_invoked)
 );
 
+#else /* #ifdef CONFIG_RCU_TRACE */
+
+#define trace_rcu_grace_period(rcuname, gpnum, gpevent) do { } while (0)
+#define trace_rcu_grace_period_init(rcuname, gpnum, level, grplo, grphi, qsmask) do { } while (0)
+#define trace_rcu_preempt_task(rcuname, pid, gpnum) do { } while (0)
+#define trace_rcu_unlock_preempted_task(rcuname, gpnum, pid) do { } while (0)
+#define trace_rcu_quiescent_state_report(rcuname, gpnum, mask, qsmask, level, grplo, grphi, gp_tasks) do { } while (0)
+#define trace_rcu_fqs(rcuname, gpnum, cpu, qsevent) do { } while (0)
+#define trace_rcu_dyntick(polarity) do { } while (0)
+#define trace_rcu_callback(rcuname, rhp, qlen) do { } while (0)
+#define trace_rcu_kfree_callback(rcuname, rhp, offset, qlen) do { } while (0)
+#define trace_rcu_batch_start(rcuname, qlen, blimit) do { } while (0)
+#define trace_rcu_invoke_callback(rcuname, rhp) do { } while (0)
+#define trace_rcu_invoke_kfree_callback(rcuname, rhp, offset) do { } while (0)
+#define trace_rcu_batch_end(rcuname, callbacks_invoked) do { } while (0)
+
+#endif /* #else #ifdef CONFIG_RCU_TRACE */
+
 #endif /* _TRACE_RCU_H */
 
 /* This part must be outside protection */
index d7f00ec..f600868 100644 (file)
@@ -69,15 +69,15 @@ static inline void debug_rcu_head_unqueue(struct rcu_head *head)
 
 extern void kfree(const void *);
 
-static inline void __rcu_reclaim(struct rcu_head *head)
+static inline void __rcu_reclaim(char *rn, struct rcu_head *head)
 {
        unsigned long offset = (unsigned long)head->func;
 
        if (__is_kfree_rcu_offset(offset)) {
-               RCU_TRACE(trace_rcu_invoke_kfree_callback(head, offset));
+               RCU_TRACE(trace_rcu_invoke_kfree_callback(rn, head, offset));
                kfree((void *)head - offset);
        } else {
-               RCU_TRACE(trace_rcu_invoke_callback(head));
+               RCU_TRACE(trace_rcu_invoke_callback(rn, head));
                head->func(head);
        }
 }
index c9321d8..da775c8 100644 (file)
@@ -147,6 +147,7 @@ void rcu_check_callbacks(int cpu, int user)
  */
 static void __rcu_process_callbacks(struct rcu_ctrlblk *rcp)
 {
+       char *rn = NULL;
        struct rcu_head *next, *list;
        unsigned long flags;
        RCU_TRACE(int cb_count = 0);
@@ -171,12 +172,13 @@ static void __rcu_process_callbacks(struct rcu_ctrlblk *rcp)
        local_irq_restore(flags);
 
        /* Invoke the callbacks on the local list. */
+       RCU_TRACE(rn = rcp->name);
        while (list) {
                next = list->next;
                prefetch(next);
                debug_rcu_head_unqueue(list);
                local_bh_disable();
-               __rcu_reclaim(list);
+               __rcu_reclaim(rn, list);
                local_bh_enable();
                list = next;
                RCU_TRACE(cb_count++);
index 4e24399..7e02829 100644 (file)
@@ -166,6 +166,8 @@ void rcu_sched_qs(int cpu)
 
        rdp->passed_quiesc_completed = rdp->gpnum - 1;
        barrier();
+       if (rdp->passed_quiesc == 0)
+               trace_rcu_grace_period("rcu_sched", rdp->gpnum, "cpuqs");
        rdp->passed_quiesc = 1;
 }
 
@@ -175,6 +177,8 @@ void rcu_bh_qs(int cpu)
 
        rdp->passed_quiesc_completed = rdp->gpnum - 1;
        barrier();
+       if (rdp->passed_quiesc == 0)
+               trace_rcu_grace_period("rcu_bh", rdp->gpnum, "cpuqs");
        rdp->passed_quiesc = 1;
 }
 
@@ -319,6 +323,7 @@ static int rcu_implicit_offline_qs(struct rcu_data *rdp)
         * trust its state not to change because interrupts are disabled.
         */
        if (cpu_is_offline(rdp->cpu)) {
+               trace_rcu_fqs(rdp->rsp->name, rdp->gpnum, rdp->cpu, "ofl");
                rdp->offline_fqs++;
                return 1;
        }
@@ -359,6 +364,7 @@ void rcu_enter_nohz(void)
                local_irq_restore(flags);
                return;
        }
+       trace_rcu_dyntick("Start");
        /* CPUs seeing atomic_inc() must see prior RCU read-side crit sects */
        smp_mb__before_atomic_inc();  /* See above. */
        atomic_inc(&rdtp->dynticks);
@@ -396,6 +402,7 @@ void rcu_exit_nohz(void)
        /* CPUs seeing atomic_inc() must see later RCU read-side crit sects */
        smp_mb__after_atomic_inc();  /* See above. */
        WARN_ON_ONCE(!(atomic_read(&rdtp->dynticks) & 0x1));
+       trace_rcu_dyntick("End");
        local_irq_restore(flags);
 }
 
@@ -501,6 +508,7 @@ static int rcu_implicit_dynticks_qs(struct rcu_data *rdp)
         * of the current RCU grace period.
         */
        if ((curr & 0x1) == 0 || ULONG_CMP_GE(curr, snap + 2)) {
+               trace_rcu_fqs(rdp->rsp->name, rdp->gpnum, rdp->cpu, "dti");
                rdp->dynticks_fqs++;
                return 1;
        }
@@ -683,6 +691,7 @@ static void __note_new_gpnum(struct rcu_state *rsp, struct rcu_node *rnp, struct
                 * go looking for one.
                 */
                rdp->gpnum = rnp->gpnum;
+               trace_rcu_grace_period(rsp->name, rdp->gpnum, "cpustart");
                if (rnp->qsmask & rdp->grpmask) {
                        rdp->qs_pending = 1;
                        rdp->passed_quiesc = 0;
@@ -746,6 +755,7 @@ __rcu_process_gp_end(struct rcu_state *rsp, struct rcu_node *rnp, struct rcu_dat
 
                /* Remember that we saw this grace-period completion. */
                rdp->completed = rnp->completed;
+               trace_rcu_grace_period(rsp->name, rdp->gpnum, "cpuend");
 
                /*
                 * If we were in an extended quiescent state, we may have
@@ -856,6 +866,7 @@ rcu_start_gp(struct rcu_state *rsp, unsigned long flags)
 
        /* Advance to a new grace period and initialize state. */
        rsp->gpnum++;
+       trace_rcu_grace_period(rsp->name, rsp->gpnum, "start");
        WARN_ON_ONCE(rsp->signaled == RCU_GP_INIT);
        rsp->signaled = RCU_GP_INIT; /* Hold off force_quiescent_state. */
        rsp->jiffies_force_qs = jiffies + RCU_JIFFIES_TILL_FORCE_QS;
@@ -870,6 +881,9 @@ rcu_start_gp(struct rcu_state *rsp, unsigned long flags)
                rsp->signaled = RCU_SIGNAL_INIT; /* force_quiescent_state OK. */
                rcu_start_gp_per_cpu(rsp, rnp, rdp);
                rcu_preempt_boost_start_gp(rnp);
+               trace_rcu_grace_period_init(rsp->name, rnp->gpnum,
+                                           rnp->level, rnp->grplo,
+                                           rnp->grphi, rnp->qsmask);
                raw_spin_unlock_irqrestore(&rnp->lock, flags);
                return;
        }
@@ -906,6 +920,9 @@ rcu_start_gp(struct rcu_state *rsp, unsigned long flags)
                if (rnp == rdp->mynode)
                        rcu_start_gp_per_cpu(rsp, rnp, rdp);
                rcu_preempt_boost_start_gp(rnp);
+               trace_rcu_grace_period_init(rsp->name, rnp->gpnum,
+                                           rnp->level, rnp->grplo,
+                                           rnp->grphi, rnp->qsmask);
                raw_spin_unlock(&rnp->lock);    /* irqs remain disabled. */
        }
 
@@ -939,6 +956,7 @@ static void rcu_report_qs_rsp(struct rcu_state *rsp, unsigned long flags)
        if (gp_duration > rsp->gp_max)
                rsp->gp_max = gp_duration;
        rsp->completed = rsp->gpnum;
+       trace_rcu_grace_period(rsp->name, rsp->completed, "end");
        rsp->signaled = RCU_GP_IDLE;
        rcu_start_gp(rsp, flags);  /* releases root node's rnp->lock. */
 }
@@ -967,6 +985,10 @@ rcu_report_qs_rnp(unsigned long mask, struct rcu_state *rsp,
                        return;
                }
                rnp->qsmask &= ~mask;
+               trace_rcu_quiescent_state_report(rsp->name, rnp->gpnum,
+                                                mask, rnp->qsmask, rnp->level,
+                                                rnp->grplo, rnp->grphi,
+                                                !!rnp->gp_tasks);
                if (rnp->qsmask != 0 || rcu_preempt_blocked_readers_cgp(rnp)) {
 
                        /* Other bits still set at this level, so done. */
@@ -1135,11 +1157,20 @@ static void __rcu_offline_cpu(int cpu, struct rcu_state *rsp)
                if (rnp->qsmaskinit != 0) {
                        if (rnp != rdp->mynode)
                                raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */
+                       else
+                               trace_rcu_grace_period(rsp->name,
+                                                      rnp->gpnum + 1 -
+                                                      !!(rnp->qsmask & mask),
+                                                      "cpuofl");
                        break;
                }
-               if (rnp == rdp->mynode)
+               if (rnp == rdp->mynode) {
+                       trace_rcu_grace_period(rsp->name,
+                                              rnp->gpnum + 1 -
+                                              !!(rnp->qsmask & mask),
+                                              "cpuofl");
                        need_report = rcu_preempt_offline_tasks(rsp, rnp, rdp);
-               else
+               } else
                        raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */
                mask = rnp->grpmask;
                rnp = rnp->parent;
@@ -1226,7 +1257,7 @@ static void rcu_do_batch(struct rcu_state *rsp, struct rcu_data *rdp)
                next = list->next;
                prefetch(next);
                debug_rcu_head_unqueue(list);
-               __rcu_reclaim(list);
+               __rcu_reclaim(rsp->name, list);
                list = next;
                if (++count >= bl)
                        break;
@@ -1552,6 +1583,12 @@ __call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu),
        rdp->nxttail[RCU_NEXT_TAIL] = &head->next;
        rdp->qlen++;
 
+       if (__is_kfree_rcu_offset((unsigned long)func))
+               trace_rcu_kfree_callback(rsp->name, head, (unsigned long)func,
+                                        rdp->qlen);
+       else
+               trace_rcu_callback(rsp->name, head, rdp->qlen);
+
        /* If interrupts were disabled, don't dive into RCU core. */
        if (irqs_disabled_flags(flags)) {
                local_irq_restore(flags);
@@ -1850,6 +1887,7 @@ rcu_boot_init_percpu_data(int cpu, struct rcu_state *rsp)
        rdp->dynticks = &per_cpu(rcu_dynticks, cpu);
 #endif /* #ifdef CONFIG_NO_HZ */
        rdp->cpu = cpu;
+       rdp->rsp = rsp;
        raw_spin_unlock_irqrestore(&rnp->lock, flags);
 }
 
@@ -1898,6 +1936,7 @@ rcu_init_percpu_data(int cpu, struct rcu_state *rsp, int preemptible)
                        rdp->gpnum = rnp->completed; /* if GP in progress... */
                        rdp->completed = rnp->completed;
                        rdp->passed_quiesc_completed = rnp->completed - 1;
+                       trace_rcu_grace_period(rsp->name, rdp->gpnum, "cpuonl");
                }
                raw_spin_unlock(&rnp->lock); /* irqs already disabled. */
                rnp = rnp->parent;
index eee6c94..d11a006 100644 (file)
@@ -299,6 +299,7 @@ struct rcu_data {
        unsigned long n_rp_need_nothing;
 
        int cpu;
+       struct rcu_state *rsp;
 };
 
 /* Values for signaled field in struct rcu_state. */
index 94d9ca1..bdb2e82 100644 (file)
@@ -124,6 +124,8 @@ static void rcu_preempt_qs(int cpu)
 
        rdp->passed_quiesc_completed = rdp->gpnum - 1;
        barrier();
+       if (rdp->passed_quiesc == 0)
+               trace_rcu_grace_period("rcu_preempt", rdp->gpnum, "cpuqs");
        rdp->passed_quiesc = 1;
        current->rcu_read_unlock_special &= ~RCU_READ_UNLOCK_NEED_QS;
 }
@@ -190,6 +192,11 @@ static void rcu_preempt_note_context_switch(int cpu)
                        if (rnp->qsmask & rdp->grpmask)
                                rnp->gp_tasks = &t->rcu_node_entry;
                }
+               trace_rcu_preempt_task(rdp->rsp->name,
+                                      t->pid,
+                                      (rnp->qsmask & rdp->grpmask)
+                                      ? rnp->gpnum
+                                      : rnp->gpnum + 1);
                raw_spin_unlock_irqrestore(&rnp->lock, flags);
        } else if (t->rcu_read_lock_nesting < 0 &&
                   t->rcu_read_unlock_special) {
@@ -344,6 +351,8 @@ static noinline void rcu_read_unlock_special(struct task_struct *t)
                smp_mb(); /* ensure expedited fastpath sees end of RCU c-s. */
                np = rcu_next_node_entry(t, rnp);
                list_del_init(&t->rcu_node_entry);
+               trace_rcu_unlock_preempted_task("rcu_preempt",
+                                               rnp->gpnum, t->pid);
                if (&t->rcu_node_entry == rnp->gp_tasks)
                        rnp->gp_tasks = np;
                if (&t->rcu_node_entry == rnp->exp_tasks)
@@ -364,10 +373,17 @@ static noinline void rcu_read_unlock_special(struct task_struct *t)
                 * we aren't waiting on any CPUs, report the quiescent state.
                 * Note that rcu_report_unblock_qs_rnp() releases rnp->lock.
                 */
-               if (empty)
-                       raw_spin_unlock_irqrestore(&rnp->lock, flags);
-               else
+               if (!empty && !rcu_preempt_blocked_readers_cgp(rnp)) {
+                       trace_rcu_quiescent_state_report("preempt_rcu",
+                                                        rnp->gpnum,
+                                                        0, rnp->qsmask,
+                                                        rnp->level,
+                                                        rnp->grplo,
+                                                        rnp->grphi,
+                                                        !!rnp->gp_tasks);
                        rcu_report_unblock_qs_rnp(rnp, flags);
+               } else
+                       raw_spin_unlock_irqrestore(&rnp->lock, flags);
 
 #ifdef CONFIG_RCU_BOOST
                /* Unboost if we were boosted. */