]> git.karo-electronics.de Git - karo-tx-linux.git/commitdiff
rcu: Add stall warnings to synchronize_sched_expedited()
authorPaul E. McKenney <paulmck@linux.vnet.ibm.com>
Tue, 30 Jun 2015 18:14:32 +0000 (11:14 -0700)
committerPaul E. McKenney <paulmck@linux.vnet.ibm.com>
Fri, 17 Jul 2015 21:59:01 +0000 (14:59 -0700)
Although synchronize_sched_expedited() historically has no RCU CPU stall
warnings, the availability of the rcupdate.rcu_expedited boot parameter
invalidates the old assumption that synchronize_sched()'s stall warnings
would suffice.  This commit therefore adds RCU CPU stall warnings to
synchronize_sched_expedited().

Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
kernel/rcu/tree.c
kernel/rcu/tree.h

index e45097fc39fae290cc5f43163da262bcdc5dc275..4b6594c7db5810ac73d0094ea02aed59a19736d4 100644 (file)
@@ -3369,16 +3369,65 @@ static struct rcu_node *exp_funnel_lock(struct rcu_state *rsp, unsigned long s)
        return rnp1;
 }
 
+/* Invoked on each online non-idle CPU for expedited quiescent state. */
 static int synchronize_sched_expedited_cpu_stop(void *data)
 {
-       struct rcu_state *rsp = data;
+       struct rcu_data *rdp = data;
+       struct rcu_state *rsp = rdp->rsp;
 
        /* We are here: If we are last, do the wakeup. */
+       rdp->exp_done = true;
        if (atomic_dec_and_test(&rsp->expedited_need_qs))
                wake_up(&rsp->expedited_wq);
        return 0;
 }
 
+static void synchronize_sched_expedited_wait(struct rcu_state *rsp)
+{
+       int cpu;
+       unsigned long jiffies_stall;
+       unsigned long jiffies_start;
+       struct rcu_data *rdp;
+       int ret;
+
+       jiffies_stall = rcu_jiffies_till_stall_check();
+       jiffies_start = jiffies;
+
+       for (;;) {
+               ret = wait_event_interruptible_timeout(
+                               rsp->expedited_wq,
+                               !atomic_read(&rsp->expedited_need_qs),
+                               jiffies_stall);
+               if (ret > 0)
+                       return;
+               if (ret < 0) {
+                       /* Hit a signal, disable CPU stall warnings. */
+                       wait_event(rsp->expedited_wq,
+                                  !atomic_read(&rsp->expedited_need_qs));
+                       return;
+               }
+               pr_err("INFO: %s detected expedited stalls on CPUs: {",
+                      rsp->name);
+               for_each_online_cpu(cpu) {
+                       rdp = per_cpu_ptr(rsp->rda, cpu);
+
+                       if (rdp->exp_done)
+                               continue;
+                       pr_cont(" %d", cpu);
+               }
+               pr_cont(" } %lu jiffies s: %lu\n",
+                       jiffies - jiffies_start, rsp->expedited_sequence);
+               for_each_online_cpu(cpu) {
+                       rdp = per_cpu_ptr(rsp->rda, cpu);
+
+                       if (rdp->exp_done)
+                               continue;
+                       dump_cpu_task(cpu);
+               }
+               jiffies_stall = 3 * rcu_jiffies_till_stall_check() + 3;
+       }
+}
+
 /**
  * synchronize_sched_expedited - Brute-force RCU-sched grace period
  *
@@ -3428,19 +3477,20 @@ void synchronize_sched_expedited(void)
                struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu);
                struct rcu_dynticks *rdtp = &per_cpu(rcu_dynticks, cpu);
 
+               rdp->exp_done = false;
+
                /* Skip our CPU and any idle CPUs. */
                if (raw_smp_processor_id() == cpu ||
                    !(atomic_add_return(0, &rdtp->dynticks) & 0x1))
                        continue;
                atomic_inc(&rsp->expedited_need_qs);
                stop_one_cpu_nowait(cpu, synchronize_sched_expedited_cpu_stop,
-                                   rsp, &rdp->exp_stop_work);
+                                   rdp, &rdp->exp_stop_work);
        }
 
        /* Remove extra count and, if necessary, wait for CPUs to stop. */
        if (!atomic_dec_and_test(&rsp->expedited_need_qs))
-               wait_event(rsp->expedited_wq,
-                          !atomic_read(&rsp->expedited_need_qs));
+               synchronize_sched_expedited_wait(rsp);
 
        rcu_exp_gp_seq_end(rsp);
        mutex_unlock(&rnp->exp_funnel_mutex);
index efee84ce1e081689757014fcb887cfebfce0874a..b3ae8d3cffbcbc7a91987c45b3534e5a2bb1e9f6 100644 (file)
@@ -370,6 +370,7 @@ struct rcu_data {
        struct rcu_head oom_head;
 #endif /* #ifdef CONFIG_RCU_FAST_NO_HZ */
        struct mutex exp_funnel_mutex;
+       bool exp_done;                  /* Expedited QS for this CPU? */
 
        /* 7) Callback offloading. */
 #ifdef CONFIG_RCU_NOCB_CPU