rcu: Move forward-progress checkers into tree_stall.h
authorPaul E. McKenney <paulmck@linux.ibm.com>
Tue, 15 Jan 2019 15:01:33 +0000 (07:01 -0800)
committerPaul E. McKenney <paulmck@linux.ibm.com>
Tue, 26 Mar 2019 21:40:13 +0000 (14:40 -0700)
This commit further consolidates stall-warning functionality by moving
forward-progress checkers into kernel/rcu/tree_stall.h, updating a
comment or two while in the area.  More specifically, this commit moves
show_rcu_gp_kthreads(), rcu_check_gp_start_stall(), rcu_fwd_progress_check(),
sysrq_rcu, sysrq_show_rcu(), sysrq_rcudump_op, and rcu_sysrq_init() from
kernel/rcu/tree.c to kernel/rcu/tree_stall.h.

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

index 929531ed168c02f6f604869d7e4d5dde12d543b4..4cb1ebd93b0ce4bd50eb940dd8ca3aaf4014f323 100644 (file)
@@ -102,9 +102,6 @@ int rcu_num_lvls __read_mostly = RCU_NUM_LVLS;
 /* Number of rcu_nodes at specified level. */
 int num_rcu_lvl[] = NUM_RCU_LVL_INIT;
 int rcu_num_nodes __read_mostly = NUM_RCU_NODES; /* Total # rcu_nodes in use. */
-/* Commandeer a sysrq key to dump RCU's tree. */
-static bool sysrq_rcu;
-module_param(sysrq_rcu, bool, 0444);
 
 /*
  * The rcu_scheduler_active variable is initialized to the value
@@ -510,74 +507,6 @@ static const char *gp_state_getname(short gs)
        return gp_state_names[gs];
 }
 
-/*
- * Show the state of the grace-period kthreads.
- */
-void show_rcu_gp_kthreads(void)
-{
-       int cpu;
-       unsigned long j;
-       unsigned long ja;
-       unsigned long jr;
-       unsigned long jw;
-       struct rcu_data *rdp;
-       struct rcu_node *rnp;
-
-       j = jiffies;
-       ja = j - READ_ONCE(rcu_state.gp_activity);
-       jr = j - READ_ONCE(rcu_state.gp_req_activity);
-       jw = j - READ_ONCE(rcu_state.gp_wake_time);
-       pr_info("%s: wait state: %s(%d) ->state: %#lx delta ->gp_activity %lu ->gp_req_activity %lu ->gp_wake_time %lu ->gp_wake_seq %ld ->gp_seq %ld ->gp_seq_needed %ld ->gp_flags %#x\n",
-               rcu_state.name, gp_state_getname(rcu_state.gp_state),
-               rcu_state.gp_state,
-               rcu_state.gp_kthread ? rcu_state.gp_kthread->state : 0x1ffffL,
-               ja, jr, jw, (long)READ_ONCE(rcu_state.gp_wake_seq),
-               (long)READ_ONCE(rcu_state.gp_seq),
-               (long)READ_ONCE(rcu_get_root()->gp_seq_needed),
-               READ_ONCE(rcu_state.gp_flags));
-       rcu_for_each_node_breadth_first(rnp) {
-               if (ULONG_CMP_GE(rcu_state.gp_seq, rnp->gp_seq_needed))
-                       continue;
-               pr_info("\trcu_node %d:%d ->gp_seq %ld ->gp_seq_needed %ld\n",
-                       rnp->grplo, rnp->grphi, (long)rnp->gp_seq,
-                       (long)rnp->gp_seq_needed);
-               if (!rcu_is_leaf_node(rnp))
-                       continue;
-               for_each_leaf_node_possible_cpu(rnp, cpu) {
-                       rdp = per_cpu_ptr(&rcu_data, cpu);
-                       if (rdp->gpwrap ||
-                           ULONG_CMP_GE(rcu_state.gp_seq,
-                                        rdp->gp_seq_needed))
-                               continue;
-                       pr_info("\tcpu %d ->gp_seq_needed %ld\n",
-                               cpu, (long)rdp->gp_seq_needed);
-               }
-       }
-       /* sched_show_task(rcu_state.gp_kthread); */
-}
-EXPORT_SYMBOL_GPL(show_rcu_gp_kthreads);
-
-/* Dump grace-period-request information due to commandeered sysrq. */
-static void sysrq_show_rcu(int key)
-{
-       show_rcu_gp_kthreads();
-}
-
-static struct sysrq_key_op sysrq_rcudump_op = {
-       .handler = sysrq_show_rcu,
-       .help_msg = "show-rcu(y)",
-       .action_msg = "Show RCU tree",
-       .enable_mask = SYSRQ_ENABLE_DUMP,
-};
-
-static int __init rcu_sysrq_init(void)
-{
-       if (sysrq_rcu)
-               return register_sysrq_key('y', &sysrq_rcudump_op);
-       return 0;
-}
-early_initcall(rcu_sysrq_init);
-
 /*
  * Send along grace-period-related data for rcutorture diagnostics.
  */
@@ -2323,101 +2252,6 @@ void rcu_force_quiescent_state(void)
 }
 EXPORT_SYMBOL_GPL(rcu_force_quiescent_state);
 
-/*
- * This function checks for grace-period requests that fail to motivate
- * RCU to come out of its idle mode.
- */
-void
-rcu_check_gp_start_stall(struct rcu_node *rnp, struct rcu_data *rdp,
-                        const unsigned long gpssdelay)
-{
-       unsigned long flags;
-       unsigned long j;
-       struct rcu_node *rnp_root = rcu_get_root();
-       static atomic_t warned = ATOMIC_INIT(0);
-
-       if (!IS_ENABLED(CONFIG_PROVE_RCU) || rcu_gp_in_progress() ||
-           ULONG_CMP_GE(rnp_root->gp_seq, rnp_root->gp_seq_needed))
-               return;
-       j = jiffies; /* Expensive access, and in common case don't get here. */
-       if (time_before(j, READ_ONCE(rcu_state.gp_req_activity) + gpssdelay) ||
-           time_before(j, READ_ONCE(rcu_state.gp_activity) + gpssdelay) ||
-           atomic_read(&warned))
-               return;
-
-       raw_spin_lock_irqsave_rcu_node(rnp, flags);
-       j = jiffies;
-       if (rcu_gp_in_progress() ||
-           ULONG_CMP_GE(rnp_root->gp_seq, rnp_root->gp_seq_needed) ||
-           time_before(j, READ_ONCE(rcu_state.gp_req_activity) + gpssdelay) ||
-           time_before(j, READ_ONCE(rcu_state.gp_activity) + gpssdelay) ||
-           atomic_read(&warned)) {
-               raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
-               return;
-       }
-       /* Hold onto the leaf lock to make others see warned==1. */
-
-       if (rnp_root != rnp)
-               raw_spin_lock_rcu_node(rnp_root); /* irqs already disabled. */
-       j = jiffies;
-       if (rcu_gp_in_progress() ||
-           ULONG_CMP_GE(rnp_root->gp_seq, rnp_root->gp_seq_needed) ||
-           time_before(j, rcu_state.gp_req_activity + gpssdelay) ||
-           time_before(j, rcu_state.gp_activity + gpssdelay) ||
-           atomic_xchg(&warned, 1)) {
-               raw_spin_unlock_rcu_node(rnp_root); /* irqs remain disabled. */
-               raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
-               return;
-       }
-       WARN_ON(1);
-       if (rnp_root != rnp)
-               raw_spin_unlock_rcu_node(rnp_root);
-       raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
-       show_rcu_gp_kthreads();
-}
-
-/*
- * Do a forward-progress check for rcutorture.  This is normally invoked
- * due to an OOM event.  The argument "j" gives the time period during
- * which rcutorture would like progress to have been made.
- */
-void rcu_fwd_progress_check(unsigned long j)
-{
-       unsigned long cbs;
-       int cpu;
-       unsigned long max_cbs = 0;
-       int max_cpu = -1;
-       struct rcu_data *rdp;
-
-       if (rcu_gp_in_progress()) {
-               pr_info("%s: GP age %lu jiffies\n",
-                       __func__, jiffies - rcu_state.gp_start);
-               show_rcu_gp_kthreads();
-       } else {
-               pr_info("%s: Last GP end %lu jiffies ago\n",
-                       __func__, jiffies - rcu_state.gp_end);
-               preempt_disable();
-               rdp = this_cpu_ptr(&rcu_data);
-               rcu_check_gp_start_stall(rdp->mynode, rdp, j);
-               preempt_enable();
-       }
-       for_each_possible_cpu(cpu) {
-               cbs = rcu_get_n_cbs_cpu(cpu);
-               if (!cbs)
-                       continue;
-               if (max_cpu < 0)
-                       pr_info("%s: callbacks", __func__);
-               pr_cont(" %d: %lu", cpu, cbs);
-               if (cbs <= max_cbs)
-                       continue;
-               max_cbs = cbs;
-               max_cpu = cpu;
-       }
-       if (max_cpu >= 0)
-               pr_cont("\n");
-}
-EXPORT_SYMBOL_GPL(rcu_fwd_progress_check);
-
 /* Perform RCU core processing work for the current CPU.  */
 static __latent_entropy void rcu_core(struct softirq_action *unused)
 {
index f882ce3ca5a56d1e3084d0f444e6fed18952d01e..e253d11af3c496209354987c00697eacb8b6edd6 100644 (file)
@@ -445,3 +445,5 @@ static void rcu_dynticks_task_exit(void);
 static void record_gp_stall_check_time(void);
 static void rcu_iw_handler(struct irq_work *iwp);
 static void check_cpu_stall(struct rcu_data *rdp);
+static void rcu_check_gp_start_stall(struct rcu_node *rnp, struct rcu_data *rdp,
+                                    const unsigned long gpssdelay);
index 526e223e41ce2589304d50d9152304412348cdf1..9e3db08d02bc1e4370e96928403ba78958cc297d 100644 (file)
@@ -536,3 +536,174 @@ static void check_cpu_stall(struct rcu_data *rdp)
                print_other_cpu_stall(gs2);
        }
 }
+
+//////////////////////////////////////////////////////////////////////////////
+//
+// RCU forward-progress mechanisms, including of callback invocation.
+
+
+/*
+ * Show the state of the grace-period kthreads.
+ */
+void show_rcu_gp_kthreads(void)
+{
+       int cpu;
+       unsigned long j;
+       unsigned long ja;
+       unsigned long jr;
+       unsigned long jw;
+       struct rcu_data *rdp;
+       struct rcu_node *rnp;
+
+       j = jiffies;
+       ja = j - READ_ONCE(rcu_state.gp_activity);
+       jr = j - READ_ONCE(rcu_state.gp_req_activity);
+       jw = j - READ_ONCE(rcu_state.gp_wake_time);
+       pr_info("%s: wait state: %s(%d) ->state: %#lx delta ->gp_activity %lu ->gp_req_activity %lu ->gp_wake_time %lu ->gp_wake_seq %ld ->gp_seq %ld ->gp_seq_needed %ld ->gp_flags %#x\n",
+               rcu_state.name, gp_state_getname(rcu_state.gp_state),
+               rcu_state.gp_state,
+               rcu_state.gp_kthread ? rcu_state.gp_kthread->state : 0x1ffffL,
+               ja, jr, jw, (long)READ_ONCE(rcu_state.gp_wake_seq),
+               (long)READ_ONCE(rcu_state.gp_seq),
+               (long)READ_ONCE(rcu_get_root()->gp_seq_needed),
+               READ_ONCE(rcu_state.gp_flags));
+       rcu_for_each_node_breadth_first(rnp) {
+               if (ULONG_CMP_GE(rcu_state.gp_seq, rnp->gp_seq_needed))
+                       continue;
+               pr_info("\trcu_node %d:%d ->gp_seq %ld ->gp_seq_needed %ld\n",
+                       rnp->grplo, rnp->grphi, (long)rnp->gp_seq,
+                       (long)rnp->gp_seq_needed);
+               if (!rcu_is_leaf_node(rnp))
+                       continue;
+               for_each_leaf_node_possible_cpu(rnp, cpu) {
+                       rdp = per_cpu_ptr(&rcu_data, cpu);
+                       if (rdp->gpwrap ||
+                           ULONG_CMP_GE(rcu_state.gp_seq,
+                                        rdp->gp_seq_needed))
+                               continue;
+                       pr_info("\tcpu %d ->gp_seq_needed %ld\n",
+                               cpu, (long)rdp->gp_seq_needed);
+               }
+       }
+       /* sched_show_task(rcu_state.gp_kthread); */
+}
+EXPORT_SYMBOL_GPL(show_rcu_gp_kthreads);
+
+/*
+ * This function checks for grace-period requests that fail to motivate
+ * RCU to come out of its idle mode.
+ */
+static void rcu_check_gp_start_stall(struct rcu_node *rnp, struct rcu_data *rdp,
+                                    const unsigned long gpssdelay)
+{
+       unsigned long flags;
+       unsigned long j;
+       struct rcu_node *rnp_root = rcu_get_root();
+       static atomic_t warned = ATOMIC_INIT(0);
+
+       if (!IS_ENABLED(CONFIG_PROVE_RCU) || rcu_gp_in_progress() ||
+           ULONG_CMP_GE(rnp_root->gp_seq, rnp_root->gp_seq_needed))
+               return;
+       j = jiffies; /* Expensive access, and in common case don't get here. */
+       if (time_before(j, READ_ONCE(rcu_state.gp_req_activity) + gpssdelay) ||
+           time_before(j, READ_ONCE(rcu_state.gp_activity) + gpssdelay) ||
+           atomic_read(&warned))
+               return;
+
+       raw_spin_lock_irqsave_rcu_node(rnp, flags);
+       j = jiffies;
+       if (rcu_gp_in_progress() ||
+           ULONG_CMP_GE(rnp_root->gp_seq, rnp_root->gp_seq_needed) ||
+           time_before(j, READ_ONCE(rcu_state.gp_req_activity) + gpssdelay) ||
+           time_before(j, READ_ONCE(rcu_state.gp_activity) + gpssdelay) ||
+           atomic_read(&warned)) {
+               raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
+               return;
+       }
+       /* Hold onto the leaf lock to make others see warned==1. */
+
+       if (rnp_root != rnp)
+               raw_spin_lock_rcu_node(rnp_root); /* irqs already disabled. */
+       j = jiffies;
+       if (rcu_gp_in_progress() ||
+           ULONG_CMP_GE(rnp_root->gp_seq, rnp_root->gp_seq_needed) ||
+           time_before(j, rcu_state.gp_req_activity + gpssdelay) ||
+           time_before(j, rcu_state.gp_activity + gpssdelay) ||
+           atomic_xchg(&warned, 1)) {
+               raw_spin_unlock_rcu_node(rnp_root); /* irqs remain disabled. */
+               raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
+               return;
+       }
+       WARN_ON(1);
+       if (rnp_root != rnp)
+               raw_spin_unlock_rcu_node(rnp_root);
+       raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
+       show_rcu_gp_kthreads();
+}
+
+/*
+ * Do a forward-progress check for rcutorture.  This is normally invoked
+ * due to an OOM event.  The argument "j" gives the time period during
+ * which rcutorture would like progress to have been made.
+ */
+void rcu_fwd_progress_check(unsigned long j)
+{
+       unsigned long cbs;
+       int cpu;
+       unsigned long max_cbs = 0;
+       int max_cpu = -1;
+       struct rcu_data *rdp;
+
+       if (rcu_gp_in_progress()) {
+               pr_info("%s: GP age %lu jiffies\n",
+                       __func__, jiffies - rcu_state.gp_start);
+               show_rcu_gp_kthreads();
+       } else {
+               pr_info("%s: Last GP end %lu jiffies ago\n",
+                       __func__, jiffies - rcu_state.gp_end);
+               preempt_disable();
+               rdp = this_cpu_ptr(&rcu_data);
+               rcu_check_gp_start_stall(rdp->mynode, rdp, j);
+               preempt_enable();
+       }
+       for_each_possible_cpu(cpu) {
+               cbs = rcu_get_n_cbs_cpu(cpu);
+               if (!cbs)
+                       continue;
+               if (max_cpu < 0)
+                       pr_info("%s: callbacks", __func__);
+               pr_cont(" %d: %lu", cpu, cbs);
+               if (cbs <= max_cbs)
+                       continue;
+               max_cbs = cbs;
+               max_cpu = cpu;
+       }
+       if (max_cpu >= 0)
+               pr_cont("\n");
+}
+EXPORT_SYMBOL_GPL(rcu_fwd_progress_check);
+
+/* Commandeer a sysrq key to dump RCU's tree. */
+static bool sysrq_rcu;
+module_param(sysrq_rcu, bool, 0444);
+
+/* Dump grace-period-request information due to commandeered sysrq. */
+static void sysrq_show_rcu(int key)
+{
+       show_rcu_gp_kthreads();
+}
+
+static struct sysrq_key_op sysrq_rcudump_op = {
+       .handler = sysrq_show_rcu,
+       .help_msg = "show-rcu(y)",
+       .action_msg = "Show RCU tree",
+       .enable_mask = SYSRQ_ENABLE_DUMP,
+};
+
+static int __init rcu_sysrq_init(void)
+{
+       if (sysrq_rcu)
+               return register_sysrq_key('y', &sysrq_rcudump_op);
+       return 0;
+}
+early_initcall(rcu_sysrq_init);