tracing: A couple of last minute fixes before pushing to 5.13

index 66a4ad93b5e9..e28d08905124 100644
@@ -2390,14 +2390,13 @@ static void tracing_stop_tr(struct trace_array *tr)
 
 static int trace_save_cmdline(struct task_struct *tsk)
 {
-	unsigned pid, idx;
+	unsigned tpid, idx;
 
 	/* treat recording of idle task as a success */
 	if (!tsk->pid)
 		return 1;
 
-	if (unlikely(tsk->pid > PID_MAX_DEFAULT))
-		return 0;
+	tpid = tsk->pid & (PID_MAX_DEFAULT - 1);
 
 	/*
 	 * It's not the end of the world if we don't get

@@ -2408,26 +2407,15 @@ static int trace_save_cmdline(struct task_struct *tsk)
 	if (!arch_spin_trylock(&trace_cmdline_lock))
 		return 0;
 
-	idx = savedcmd->map_pid_to_cmdline[tsk->pid];
+	idx = savedcmd->map_pid_to_cmdline[tpid];
 	if (idx == NO_CMDLINE_MAP) {
 		idx = (savedcmd->cmdline_idx + 1) % savedcmd->cmdline_num;
 
-		/*
-		 * Check whether the cmdline buffer at idx has a pid
-		 * mapped. We are going to overwrite that entry so we
-		 * need to clear the map_pid_to_cmdline. Otherwise we
-		 * would read the new comm for the old pid.
-		 */
-		pid = savedcmd->map_cmdline_to_pid[idx];
-		if (pid != NO_CMDLINE_MAP)
-			savedcmd->map_pid_to_cmdline[pid] = NO_CMDLINE_MAP;
-
-		savedcmd->map_cmdline_to_pid[idx] = tsk->pid;
-		savedcmd->map_pid_to_cmdline[tsk->pid] = idx;
-
+		savedcmd->map_pid_to_cmdline[tpid] = idx;
 		savedcmd->cmdline_idx = idx;
 	}
 
+	savedcmd->map_cmdline_to_pid[idx] = tsk->pid;
 	set_cmdline(idx, tsk->comm);
 
 	arch_spin_unlock(&trace_cmdline_lock);

@@ -2438,6 +2426,7 @@ static int trace_save_cmdline(struct task_struct *tsk)
 static void __trace_find_cmdline(int pid, char comm[])
 {
 	unsigned map;
+	int tpid;
 
 	if (!pid) {
 		strcpy(comm, "<idle>");

@@ -2449,16 +2438,16 @@ static void __trace_find_cmdline(int pid, char comm[])
 		return;
 	}
 
-	if (pid > PID_MAX_DEFAULT) {
-		strcpy(comm, "<...>");
-		return;
+	tpid = pid & (PID_MAX_DEFAULT - 1);
+	map = savedcmd->map_pid_to_cmdline[tpid];
+	if (map != NO_CMDLINE_MAP) {
+		tpid = savedcmd->map_cmdline_to_pid[map];
+		if (tpid == pid) {
+			strlcpy(comm, get_saved_cmdlines(map), TASK_COMM_LEN);
+			return;
+		}
 	}
-
-	map = savedcmd->map_pid_to_cmdline[pid];
-	if (map != NO_CMDLINE_MAP)
-		strlcpy(comm, get_saved_cmdlines(map), TASK_COMM_LEN);
-	else
-		strcpy(comm, "<...>");
+	strcpy(comm, "<...>");
 }
 
 void trace_find_cmdline(int pid, char comm[])

index aaf6793ededa..c1637f90c8a3 100644
@@ -95,33 +95,49 @@ u64 notrace trace_clock_global(void)
 {
 	unsigned long flags;
 	int this_cpu;
-	u64 now;
+	u64 now, prev_time;
 
 	raw_local_irq_save(flags);
 
 	this_cpu = raw_smp_processor_id();
-	now = sched_clock_cpu(this_cpu);
+
 	/*
-	 * If in an NMI context then dont risk lockups and return the
-	 * cpu_clock() time:
+	 * The global clock "guarantees" that the events are ordered
+	 * between CPUs. But if two events on two different CPUS call
+	 * trace_clock_global at roughly the same time, it really does
+	 * not matter which one gets the earlier time. Just make sure
+	 * that the same CPU will always show a monotonic clock.
+	 *
+	 * Use a read memory barrier to get the latest written
+	 * time that was recorded.
 	 */
-	if (unlikely(in_nmi()))
-		goto out;
+	smp_rmb();
+	prev_time = READ_ONCE(trace_clock_struct.prev_time);
+	now = sched_clock_cpu(this_cpu);
 
-	arch_spin_lock(&trace_clock_struct.lock);
+	/* Make sure that now is always greater than prev_time */
+	if ((s64)(now - prev_time) < 0)
+		now = prev_time + 1;
 
 	/*
-	 * TODO: if this happens often then maybe we should reset
-	 * my_scd->clock to prev_time+1, to make sure
-	 * we start ticking with the local clock from now on?
+	 * If in an NMI context then dont risk lockups and simply return
+	 * the current time.
 	 */
-	if ((s64)(now - trace_clock_struct.prev_time) < 0)
-		now = trace_clock_struct.prev_time + 1;
+	if (unlikely(in_nmi()))
+		goto out;
 
-	trace_clock_struct.prev_time = now;
+	/* Tracing can cause strange recursion, always use a try lock */
+	if (arch_spin_trylock(&trace_clock_struct.lock)) {
+		/* Reread prev_time in case it was already updated */
+		prev_time = READ_ONCE(trace_clock_struct.prev_time);
+		if ((s64)(now - prev_time) < 0)
+			now = prev_time + 1;
 
-	arch_spin_unlock(&trace_clock_struct.lock);
+		trace_clock_struct.prev_time = now;
 
+		/* The unlock acts as the wmb for the above rmb */
+		arch_spin_unlock(&trace_clock_struct.lock);
+	}
  out:
 	raw_local_irq_restore(flags);