[PATCH 3/3] s390/pai: Support CPU hotplug for PMU PAI

The command 'perf stat -e pai_crypto/CRYPTO_ALL/ -- <command>'
crashes the kernel when CPUs are hotplug added during that run.

Root cause is the missing allocation of per-CPU data structures
for that new CPU. The allocation is dynamic and the first
event that has task context creates such a structure for
each online CPU. This is not sufficient. CPUs may be offline
during event creation and can be set online during the
perf run time. For example commands

 # echo 0 > /sys/devices/system/cpu/cpu1/online
 # perf stat -e cycles -i -- stress-ng -t10s --matrix X
 # sleep 1
 # echo 1 > /sys/devices/system/cpu/cpu1/online

Currently without a CPU hotplug handler, that new CPU has no
per-CPU data infrastructure. The scheduler runs PMU call back
function pai_add() to install the PMU support for that CPU before
the task is being scheduled on that new CPU.
In pai_add() instructions

    mp = this_cpu_ptr(pai_root[idx].mapptr);
    cpump = mp->mapptr;

return a NULL pointer and the result is a kernel panic as variable
cpump is used inside that function.

Add CPU hotplug support for CPU add and delete and create
the necessary per-CPU data infrastructure during CPU hotplug
add processing. Same for CPU hotplug remove.
This is done when the CPU is offline to ensure the data structures
are available when CPU is made online and tasks are schedules on it.

 #Cc: [email protected] # v6.19
Fixes: 582cc1b28e8c ("s390/pai_ext: Enable per-task and system-wide sampling event")
Fixes: 9f66572f2889 ("s390/pai_crypto: Enable per-task and system-wide sampling event")
Signed-off-by: Thomas Richter <[email protected]>
---
 arch/s390/include/asm/pai.h |   1 -
 arch/s390/kernel/perf_pai.c | 160 ++++++++++++++++++++++++++----------
 2 files changed, 116 insertions(+), 45 deletions(-)

diff --git a/arch/s390/include/asm/pai.h b/arch/s390/include/asm/pai.h
index 534d0320e2aa..a3456a36aaa7 100644
--- a/arch/s390/include/asm/pai.h
+++ b/arch/s390/include/asm/pai.h
@@ -76,7 +76,6 @@ static __always_inline void pai_kernel_exit(struct pt_regs *regs)
 }
 
 #define PAI_SAVE_AREA(x)	((x)->hw.event_base)
-#define PAI_CPU_MASK(x)		((x)->hw.addr_filters)
 #define PAI_PMU_IDX(x)		((x)->hw.last_tag)
 #define PAI_SWLIST(x)		(&(x)->hw.tp_list)
 
diff --git a/arch/s390/kernel/perf_pai.c b/arch/s390/kernel/perf_pai.c
index 52d9f654346a..04a9dafa8f37 100644
--- a/arch/s390/kernel/perf_pai.c
+++ b/arch/s390/kernel/perf_pai.c
@@ -67,6 +67,7 @@ struct pai_mapptr {
 
 static struct pai_root {		/* Anchor to per CPU data */
 	refcount_t refcnt;		/* Overall active events */
+	atomic_t tskctx;		/* Overall per-task events */
 	struct pai_mapptr __percpu *mapptr;
 } pai_root[PAI_PMU_MAX];
 
@@ -93,14 +94,15 @@ struct pai_pmu {			/* Define PAI PMU characteristics */
 static struct pai_pmu pai_pmu[];	/* Forward declaration */
 
 /* Free per CPU data when the last event is removed. */
-static void pai_root_free(int idx)
+static void pai_root_free(int idx, int tasks)
 {
-	if (refcount_dec_and_test(&pai_root[idx].refcnt)) {
+	if (refcount_sub_and_test(tasks, &pai_root[idx].refcnt)) {
 		free_percpu(pai_root[idx].mapptr);
 		pai_root[idx].mapptr = NULL;
 	}
-	debug_sprintf_event(paidbg, 5, "%s root[%d].refcount %d\n", __func__,
-			    idx, refcount_read(&pai_root[idx].refcnt));
+	debug_sprintf_event(paidbg, 5, "%s root[%d].refcount %d tskctx %d\n",
+			    __func__, idx, refcount_read(&pai_root[idx].refcnt),
+			    atomic_read(&pai_root[idx].tskctx));
 }
 
 /*
@@ -137,20 +139,36 @@ static void pai_free(struct pai_mapptr *mp)
 	mp->mapptr = NULL;
 }
 
-/* Adjust usage counters and remove allocated memory when all users are
- * gone. Called under mutex_lock.
- */
-static void pai_event_destroy_cpu(int idx, int cpu)
+/* Called under mutex_lock */
+static void pai_event_destroy_cpu(int idx, int cpu, bool hotplug)
 {
-	struct pai_mapptr *mp = per_cpu_ptr(pai_root[idx].mapptr, cpu);
-	struct pai_map *cpump = mp->mapptr;
+	struct pai_mapptr *mp;
+	struct pai_map *cpump;
+	int tasks = 1;
 
-	debug_sprintf_event(paidbg, 5, "%s users %d refcnt %u\n",
-			    __func__, cpump->active_events,
-			    refcount_read(&cpump->refcnt));
-	if (refcount_dec_and_test(&cpump->refcnt))
+	/* Check reference count and return when all gone.
+	 * 1. An event is installed on online CPU X.
+	 * 2. CPU x is offlined and the per-CPU data is removed.
+	 * 3. Event is destroyed via close system call.
+	 */
+	if (!refcount_read(&pai_root[idx].refcnt))
+		return;			/* No events at all */
+	mp = per_cpu_ptr(pai_root[idx].mapptr, cpu);
+	if (!mp || !mp->mapptr)		/* No events on that CPU */
+		return;
+
+	/* When hotplug is true, invocation is from CPU hotplug callback.
+	 * Delete per-CPU resource and adjust refcnt when per-task events
+	 * are currently active. This can be more than one.
+	 * In this case adjust counters.
+	 */
+	if (hotplug)
+		tasks = atomic_read(&pai_root[idx].tskctx);
+
+	cpump = mp->mapptr;
+	if (refcount_sub_and_test(tasks, &cpump->refcnt))
 		pai_free(mp);
-	pai_root_free(idx);
+	pai_root_free(idx, tasks);
 }
 
 static void pai_event_destroy(struct perf_event *event)
@@ -160,13 +178,11 @@ static void pai_event_destroy(struct perf_event *event)
 	free_page(PAI_SAVE_AREA(event));
 	mutex_lock(&pai_reserve_mutex);
 	if (event->cpu == -1) {
-		struct cpumask *mask = PAI_CPU_MASK(event);
-
-		for_each_cpu(cpu, mask)
-			pai_event_destroy_cpu(idx, cpu);
-		kfree(mask);
+		atomic_dec(&pai_root[idx].tskctx);
+		for_each_online_cpu(cpu)
+			pai_event_destroy_cpu(idx, cpu, false);
 	} else {
-		pai_event_destroy_cpu(idx, event->cpu);
+		pai_event_destroy_cpu(idx, event->cpu, false);
 	}
 	mutex_unlock(&pai_reserve_mutex);
 }
@@ -232,17 +248,25 @@ static u64 paicrypt_getall(struct perf_event *event)
 	return sum;
 }
 
-/* Allocate all per-CPU data structures. This function is called in
- * process context and can block. In case of error all partly allocated
- * memory is released and the reference counters adjusted correctly.
- * Called under mutex_lock.
- */
-static int pai_alloc_cpu(int idx, int cpu)
+/* Called under mutex_lock */
+static int pai_alloc_cpu(int idx, int cpu, bool hotplug)
 {
 	struct pai_map *cpump = NULL;
 	bool need_paiext_cb = false;
 	struct pai_mapptr *mp;
-	int rc;
+	int tasks = 1, rc = 0;
+
+	/* When hotplug is true, invocation is from CPU hotplug callback.
+	 * Allocate per-CPU resource when per-task events are currently active.
+	 * This can be more than one. In this case adjust all reference
+	 * counters. Otherwise return, this ensures memory is only allocated
+	 * when needed.
+	 */
+	if (hotplug) {
+		tasks = atomic_read(&pai_root[idx].tskctx);
+		if (!tasks)
+			goto out;
+	}
 
 	/* Allocate root node */
 	rc = pai_root_alloc(idx);
@@ -291,26 +315,42 @@ static int pai_alloc_cpu(int idx, int cpu)
 			goto undo;
 		}
 		INIT_LIST_HEAD(&cpump->syswide_list);
-		refcount_set(&cpump->refcnt, 1);
+		refcount_set(&cpump->refcnt, tasks);
 		rc = 0;
 	} else {
-		refcount_inc(&cpump->refcnt);
+		refcount_add(tasks, &cpump->refcnt);
 	}
+	/* If tasks is greater than 1, we are called from CPU hotplug path
+	 * and need to adjust the pai_root[idx].refcnt by the number of
+	 * per-process events. Function pai_root_alloc(idx) already
+	 * incremented by one. Adjust for the rest.
+	 */
+	if (tasks > 1)
+		refcount_add(tasks - 1, &pai_root[idx].refcnt);
 
 undo:
 	if (rc) {
 		/* Error in allocation of event, decrement anchor. Since
 		 * the event in not created, its destroy() function is never
 		 * invoked. Adjust the reference counter for the anchor.
+		 * The failure happened in the case of variable
+		 * cpump == NULL branch above. The pai_root[XXX].refcnt has
+		 * been incremented by one. Then the per-CPU allocation
+		 * failed, so decrement it by one, regardless of tasks.
 		 */
-		pai_root_free(idx);
+		pai_root_free(idx, 1);
 	}
 out:
 	/* If rc is non-zero, no increment of counter/sampler was done. */
 	return rc;
 }
 
-/* Called under mutex_lock */
+/* Check concurrent access of counting and sampling for PAI events.
+ * This function is called in process context and it is save to block.
+ * When the event initialization functions fails, no other call back will
+ * be invoked.
+ * Called under mutex_lock.
+ */
 static int pai_alloc(struct perf_event *event)
 {
 	int idx = PAI_PMU_IDX(event);
@@ -322,24 +362,20 @@ static int pai_alloc(struct perf_event *event)
 		goto out;
 
 	for_each_online_cpu(cpu) {
-		rc = pai_alloc_cpu(idx, cpu);
+		rc = pai_alloc_cpu(idx, cpu, false);
 		if (rc) {
 			for_each_cpu(cpu, maskptr)
-				pai_event_destroy_cpu(idx, cpu);
-			kfree(maskptr);
-			goto out;
+				pai_event_destroy_cpu(idx, cpu, false);
+			goto undo;
 		}
 		cpumask_set_cpu(cpu, maskptr);
 	}
 
-	/*
-	 * On error all cpumask are freed and all events have been destroyed.
-	 * Save of which CPUs data structures have been allocated for.
-	 * Release them in pai_event_destroy call back function
-	 * for this event.
-	 */
-	PAI_CPU_MASK(event) = maskptr;
 	rc = 0;
+	/* Trace per-task events for CPU hotplug. */
+	atomic_inc(&pai_root[idx].tskctx);
+undo:
+	kfree(maskptr);
 out:
 	return rc;
 }
@@ -389,7 +425,7 @@ static int pai_event_init(struct perf_event *event, int idx)
 
 	mutex_lock(&pai_reserve_mutex);
 	if (event->cpu >= 0)
-		rc = pai_alloc_cpu(idx, event->cpu);
+		rc = pai_alloc_cpu(idx, event->cpu, false);
 	else
 		rc = pai_alloc(event);
 	mutex_unlock(&pai_reserve_mutex);
@@ -1216,8 +1252,34 @@ static int __init paipmu_setup(void)
 	return install_ok;
 }
 
+static int pai_online_cpu(unsigned int cpu)
+{
+	int rc;
+
+	mutex_lock(&pai_reserve_mutex);
+	rc = pai_alloc_cpu(PAI_PMU_CRYPTO, cpu, true);
+	if (!rc) {
+		rc = pai_alloc_cpu(PAI_PMU_EXT, cpu, true);
+		if (rc)
+			pai_event_destroy_cpu(PAI_PMU_CRYPTO, cpu, true);
+	}
+	mutex_unlock(&pai_reserve_mutex);
+	return rc;
+}
+
+static int pai_offline_cpu(unsigned int cpu)
+{
+	mutex_lock(&pai_reserve_mutex);
+	pai_event_destroy_cpu(PAI_PMU_CRYPTO, cpu, true);
+	pai_event_destroy_cpu(PAI_PMU_EXT, cpu, true);
+	mutex_unlock(&pai_reserve_mutex);
+	return 0;
+}
+
 static int __init pai_init(void)
 {
+	int rc;
+
 	/* Setup s390dbf facility */
 	paidbg = debug_register("pai", 32, 256, 128);
 	if (!paidbg) {
@@ -1226,7 +1288,17 @@ static int __init pai_init(void)
 	}
 	debug_register_view(paidbg, &debug_sprintf_view);
 
+	/* CPUHP_BP_PREPARE_DYN --> before CPU is brought online */
+	rc = cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "perf/pai:prepare",
+			       pai_online_cpu, pai_offline_cpu);
+	if (rc < 0) {
+		debug_unregister_view(paidbg, &debug_sprintf_view);
+		debug_unregister(paidbg);
+		return rc;
+	}
+
 	if (!paipmu_setup()) {
+		cpuhp_remove_state(rc);
 		/* No PMU registration, no need for debug buffer */
 		debug_unregister_view(paidbg, &debug_sprintf_view);
 		debug_unregister(paidbg);
-- 
2.55.0