rcu: Mark accesses in tree_stall.h
authorPaul E. McKenney <paulmck@kernel.org>
Fri, 21 May 2021 04:08:38 +0000 (21:08 -0700)
committerPaul E. McKenney <paulmck@kernel.org>
Fri, 6 Aug 2021 20:41:48 +0000 (13:41 -0700)
This commit marks the accesses in tree_stall.h so as to both avoid
undesirable compiler optimizations and to keep KCSAN focused on the
accesses of the core algorithm.

Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
kernel/rcu/tree_stall.h

index 6dd6c9a..a8d0fcf 100644 (file)
@@ -465,9 +465,10 @@ static void rcu_check_gp_kthread_starvation(void)
                pr_err("%s kthread starved for %ld jiffies! g%ld f%#x %s(%d) ->state=%#x ->cpu=%d\n",
                       rcu_state.name, j,
                       (long)rcu_seq_current(&rcu_state.gp_seq),
-                      data_race(rcu_state.gp_flags),
-                      gp_state_getname(rcu_state.gp_state), rcu_state.gp_state,
-                      gpk ? gpk->__state : ~0, cpu);
+                      data_race(READ_ONCE(rcu_state.gp_flags)),
+                      gp_state_getname(rcu_state.gp_state),
+                      data_race(READ_ONCE(rcu_state.gp_state)),
+                      gpk ? data_race(READ_ONCE(gpk->__state)) : ~0, cpu);
                if (gpk) {
                        pr_err("\tUnless %s kthread gets sufficient CPU time, OOM is now expected behavior.\n", rcu_state.name);
                        pr_err("RCU grace-period kthread stack dump:\n");
@@ -510,7 +511,7 @@ static void rcu_check_gp_kthread_expired_fqs_timer(void)
                       (long)rcu_seq_current(&rcu_state.gp_seq),
                       data_race(rcu_state.gp_flags),
                       gp_state_getname(RCU_GP_WAIT_FQS), RCU_GP_WAIT_FQS,
-                      gpk->__state);
+                      data_race(READ_ONCE(gpk->__state)));
                pr_err("\tPossible timer handling issue on cpu=%d timer-softirq=%u\n",
                       cpu, kstat_softirqs_cpu(TIMER_SOFTIRQ, cpu));
        }
@@ -569,11 +570,11 @@ static void print_other_cpu_stall(unsigned long gp_seq, unsigned long gps)
                        pr_err("INFO: Stall ended before state dump start\n");
                } else {
                        j = jiffies;
-                       gpa = data_race(rcu_state.gp_activity);
+                       gpa = data_race(READ_ONCE(rcu_state.gp_activity));
                        pr_err("All QSes seen, last %s kthread activity %ld (%ld-%ld), jiffies_till_next_fqs=%ld, root ->qsmask %#lx\n",
                               rcu_state.name, j - gpa, j, gpa,
-                              data_race(jiffies_till_next_fqs),
-                              rcu_get_root()->qsmask);
+                              data_race(READ_ONCE(jiffies_till_next_fqs)),
+                              data_race(READ_ONCE(rcu_get_root()->qsmask)));
                }
        }
        /* Rewrite if needed in case of slow consoles. */
@@ -815,32 +816,34 @@ void show_rcu_gp_kthreads(void)
        struct task_struct *t = READ_ONCE(rcu_state.gp_kthread);
 
        j = jiffies;
-       ja = j - data_race(rcu_state.gp_activity);
-       jr = j - data_race(rcu_state.gp_req_activity);
-       js = j - data_race(rcu_state.gp_start);
-       jw = j - data_race(rcu_state.gp_wake_time);
+       ja = j - data_race(READ_ONCE(rcu_state.gp_activity));
+       jr = j - data_race(READ_ONCE(rcu_state.gp_req_activity));
+       js = j - data_race(READ_ONCE(rcu_state.gp_start));
+       jw = j - data_race(READ_ONCE(rcu_state.gp_wake_time));
        pr_info("%s: wait state: %s(%d) ->state: %#x ->rt_priority %u delta ->gp_start %lu ->gp_activity %lu ->gp_req_activity %lu ->gp_wake_time %lu ->gp_wake_seq %ld ->gp_seq %ld ->gp_seq_needed %ld ->gp_max %lu ->gp_flags %#x\n",
                rcu_state.name, gp_state_getname(rcu_state.gp_state),
-               rcu_state.gp_state, t ? t->__state : 0x1ffff, t ? t->rt_priority : 0xffU,
-               js, ja, jr, jw, (long)data_race(rcu_state.gp_wake_seq),
-               (long)data_race(rcu_state.gp_seq),
-               (long)data_race(rcu_get_root()->gp_seq_needed),
-               data_race(rcu_state.gp_max),
-               data_race(rcu_state.gp_flags));
+               data_race(READ_ONCE(rcu_state.gp_state)),
+               t ? data_race(READ_ONCE(t->__state)) : 0x1ffff, t ? t->rt_priority : 0xffU,
+               js, ja, jr, jw, (long)data_race(READ_ONCE(rcu_state.gp_wake_seq)),
+               (long)data_race(READ_ONCE(rcu_state.gp_seq)),
+               (long)data_race(READ_ONCE(rcu_get_root()->gp_seq_needed)),
+               data_race(READ_ONCE(rcu_state.gp_max)),
+               data_race(READ_ONCE(rcu_state.gp_flags)));
        rcu_for_each_node_breadth_first(rnp) {
                if (ULONG_CMP_GE(READ_ONCE(rcu_state.gp_seq), READ_ONCE(rnp->gp_seq_needed)) &&
-                   !data_race(rnp->qsmask) && !data_race(rnp->boost_tasks) &&
-                   !data_race(rnp->exp_tasks) && !data_race(rnp->gp_tasks))
+                   !data_race(READ_ONCE(rnp->qsmask)) && !data_race(READ_ONCE(rnp->boost_tasks)) &&
+                   !data_race(READ_ONCE(rnp->exp_tasks)) && !data_race(READ_ONCE(rnp->gp_tasks)))
                        continue;
                pr_info("\trcu_node %d:%d ->gp_seq %ld ->gp_seq_needed %ld ->qsmask %#lx %c%c%c%c ->n_boosts %ld\n",
                        rnp->grplo, rnp->grphi,
-                       (long)data_race(rnp->gp_seq), (long)data_race(rnp->gp_seq_needed),
-                       data_race(rnp->qsmask),
-                       ".b"[!!data_race(rnp->boost_kthread_task)],
-                       ".B"[!!data_race(rnp->boost_tasks)],
-                       ".E"[!!data_race(rnp->exp_tasks)],
-                       ".G"[!!data_race(rnp->gp_tasks)],
-                       data_race(rnp->n_boosts));
+                       (long)data_race(READ_ONCE(rnp->gp_seq)),
+                       (long)data_race(READ_ONCE(rnp->gp_seq_needed)),
+                       data_race(READ_ONCE(rnp->qsmask)),
+                       ".b"[!!data_race(READ_ONCE(rnp->boost_kthread_task))],
+                       ".B"[!!data_race(READ_ONCE(rnp->boost_tasks))],
+                       ".E"[!!data_race(READ_ONCE(rnp->exp_tasks))],
+                       ".G"[!!data_race(READ_ONCE(rnp->gp_tasks))],
+                       data_race(READ_ONCE(rnp->n_boosts)));
                if (!rcu_is_leaf_node(rnp))
                        continue;
                for_each_leaf_node_possible_cpu(rnp, cpu) {
@@ -850,12 +853,12 @@ void show_rcu_gp_kthreads(void)
                                         READ_ONCE(rdp->gp_seq_needed)))
                                continue;
                        pr_info("\tcpu %d ->gp_seq_needed %ld\n",
-                               cpu, (long)data_race(rdp->gp_seq_needed));
+                               cpu, (long)data_race(READ_ONCE(rdp->gp_seq_needed)));
                }
        }
        for_each_possible_cpu(cpu) {
                rdp = per_cpu_ptr(&rcu_data, cpu);
-               cbs += data_race(rdp->n_cbs_invoked);
+               cbs += data_race(READ_ONCE(rdp->n_cbs_invoked));
                if (rcu_segcblist_is_offloaded(&rdp->cblist))
                        show_rcu_nocb_state(rdp);
        }
@@ -937,11 +940,11 @@ void rcu_fwd_progress_check(unsigned long j)
 
        if (rcu_gp_in_progress()) {
                pr_info("%s: GP age %lu jiffies\n",
-                       __func__, jiffies - rcu_state.gp_start);
+                       __func__, jiffies - data_race(READ_ONCE(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);
+                       __func__, jiffies - data_race(READ_ONCE(rcu_state.gp_end)));
                preempt_disable();
                rdp = this_cpu_ptr(&rcu_data);
                rcu_check_gp_start_stall(rdp->mynode, rdp, j);