[PATCH v2 02/10] coresight: perf: Fix per cpu path management

From: Suzuki K Poulose <suzuki.poulose@arm.com>
To: linux-arm-kernel@lists.infradead.org
Cc: linux-kernel@vger.kernel.org, robert.walker@arm.com,
	mathieu.poirier@linaro.org, mike.leach@linaro.org,
	coresight@lists.linaro.org,
	Suzuki K Poulose <suzuki.poulose@arm.com>
Subject: [PATCH v2 02/10] coresight: perf: Fix per cpu path management
Date: Tue, 17 Jul 2018 18:11:33 +0100	[thread overview]
Message-ID: <1531847501-22226-3-git-send-email-suzuki.poulose@arm.com> (raw)
In-Reply-To: <1531847501-22226-1-git-send-email-suzuki.poulose@arm.com>

We create a coresight trace path for each online CPU when
we start the event. We rely on the number of online CPUs
and then go on to allocate an array matching the "number of
online CPUs" for holding the path and then uses normal
CPU id as the index to the array. This is problematic as
we could have some offline CPUs causing us to access beyond
the actual array size (e.g, on a dual SMP system, if CPU0 is
offline, CPU1 could be really accessing beyond the array).
The solution is to switch to per-cpu array for holding the path.
Also we get rid of the {get/put}_online_cpus() which is
not really needed.

Cc: Mathieu Poirier <mathieu.poirier@linaro.org>
Signed-off-by: Suzuki K Poulose <suzuki.poulose@arm.com>
---
 drivers/hwtracing/coresight/coresight-etm-perf.c | 59 ++++++++++++++++--------
 1 file changed, 39 insertions(+), 20 deletions(-)

diff --git a/drivers/hwtracing/coresight/coresight-etm-perf.c b/drivers/hwtracing/coresight/coresight-etm-perf.c
index 6776956..ab2a6e8 100644
--- a/drivers/hwtracing/coresight/coresight-etm-perf.c
+++ b/drivers/hwtracing/coresight/coresight-etm-perf.c
@@ -12,6 +12,7 @@
 #include <linux/mm.h>
 #include <linux/init.h>
 #include <linux/perf_event.h>
+#include <linux/percpu-defs.h>
 #include <linux/slab.h>
 #include <linux/types.h>
 #include <linux/workqueue.h>
@@ -33,7 +34,7 @@ struct etm_event_data {
 	struct work_struct work;
 	cpumask_t mask;
 	void *snk_config;
-	struct list_head **path;
+	struct list_head * __percpu *path;
 };
 
 static DEFINE_PER_CPU(struct perf_output_handle, ctx_handle);
@@ -61,6 +62,18 @@ static const struct attribute_group *etm_pmu_attr_groups[] = {
 	NULL,
 };
 
+static inline struct list_head **
+etm_event_cpu_path_ptr(struct etm_event_data *data, int cpu)
+{
+	return per_cpu_ptr(data->path, cpu);
+}
+
+static inline struct list_head *
+etm_event_cpu_path(struct etm_event_data *data, int cpu)
+{
+	return *etm_event_cpu_path_ptr(data, cpu);
+}
+
 static void etm_event_read(struct perf_event *event) {}
 
 static int etm_addr_filters_alloc(struct perf_event *event)
@@ -120,23 +133,26 @@ static void free_event_data(struct work_struct *work)
 	 */
 	if (event_data->snk_config) {
 		cpu = cpumask_first(mask);
-		sink = coresight_get_sink(event_data->path[cpu]);
+		sink = coresight_get_sink(etm_event_cpu_path(event_data, cpu));
 		if (sink_ops(sink)->free_buffer)
 			sink_ops(sink)->free_buffer(event_data->snk_config);
 	}
 
 	for_each_cpu(cpu, mask) {
-		if (!(IS_ERR_OR_NULL(event_data->path[cpu])))
-			coresight_release_path(event_data->path[cpu]);
+		struct list_head **ppath;
+
+		ppath = etm_event_cpu_path_ptr(event_data, cpu);
+		if (!(IS_ERR_OR_NULL(*ppath)))
+			coresight_release_path(*ppath);
+		*ppath = NULL;
 	}
 
-	kfree(event_data->path);
+	free_percpu(event_data->path);
 	kfree(event_data);
 }
 
 static void *alloc_event_data(int cpu)
 {
-	int size;
 	cpumask_t *mask;
 	struct etm_event_data *event_data;
 
@@ -145,17 +161,11 @@ static void *alloc_event_data(int cpu)
 	if (!event_data)
 		return NULL;
 
-	/* Make sure nothing disappears under us */
-	get_online_cpus();
-	size = num_online_cpus();
-
 	mask = &event_data->mask;
 	if (cpu != -1)
 		cpumask_set_cpu(cpu, mask);
 	else
 		cpumask_copy(mask, cpu_online_mask);
-	put_online_cpus();
-
 	/*
 	 * Each CPU has a single path between source and destination.  As such
 	 * allocate an array using CPU numbers as indexes.  That way a path
@@ -164,8 +174,8 @@ static void *alloc_event_data(int cpu)
 	 * unused memory when dealing with single CPU trace scenarios is small
 	 * compared to the cost of searching through an optimized array.
 	 */
-	event_data->path = kcalloc(size,
-				   sizeof(struct list_head *), GFP_KERNEL);
+	event_data->path = alloc_percpu(struct list_head *);
+
 	if (!event_data->path) {
 		kfree(event_data);
 		return NULL;
@@ -213,6 +223,7 @@ static void *etm_setup_aux(int event_cpu, void **pages,
 
 	/* Setup the path for each CPU in a trace session */
 	for_each_cpu(cpu, mask) {
+		struct list_head *path;
 		struct coresight_device *csdev;
 
 		csdev = per_cpu(csdev_src, cpu);
@@ -224,9 +235,10 @@ static void *etm_setup_aux(int event_cpu, void **pages,
 		 * list of devices from source to sink that can be
 		 * referenced later when the path is actually needed.
 		 */
-		event_data->path[cpu] = coresight_build_path(csdev, sink);
-		if (IS_ERR(event_data->path[cpu]))
+		path = coresight_build_path(csdev, sink);
+		if (IS_ERR(path))
 			goto err;
+		*etm_event_cpu_path_ptr(event_data, cpu) = path;
 	}
 
 	if (!sink_ops(sink)->alloc_buffer)
@@ -255,6 +267,7 @@ static void etm_event_start(struct perf_event *event, int flags)
 	struct etm_event_data *event_data;
 	struct perf_output_handle *handle = this_cpu_ptr(&ctx_handle);
 	struct coresight_device *sink, *csdev = per_cpu(csdev_src, cpu);
+	struct list_head *path;
 
 	if (!csdev)
 		goto fail;
@@ -267,8 +280,9 @@ static void etm_event_start(struct perf_event *event, int flags)
 	if (!event_data)
 		goto fail;
 
+	path = etm_event_cpu_path(event_data, cpu);
 	/* We need a sink, no need to continue without one */
-	sink = coresight_get_sink(event_data->path[cpu]);
+	sink = coresight_get_sink(path);
 	if (WARN_ON_ONCE(!sink || !sink_ops(sink)->set_buffer))
 		goto fail_end_stop;
 
@@ -278,7 +292,7 @@ static void etm_event_start(struct perf_event *event, int flags)
 		goto fail_end_stop;
 
 	/* Nothing will happen without a path */
-	if (coresight_enable_path(event_data->path[cpu], CS_MODE_PERF))
+	if (coresight_enable_path(path, CS_MODE_PERF))
 		goto fail_end_stop;
 
 	/* Tell the perf core the event is alive */
@@ -306,6 +320,7 @@ static void etm_event_stop(struct perf_event *event, int mode)
 	struct coresight_device *sink, *csdev = per_cpu(csdev_src, cpu);
 	struct perf_output_handle *handle = this_cpu_ptr(&ctx_handle);
 	struct etm_event_data *event_data = perf_get_aux(handle);
+	struct list_head *path;
 
 	if (event->hw.state == PERF_HES_STOPPED)
 		return;
@@ -313,7 +328,11 @@ static void etm_event_stop(struct perf_event *event, int mode)
 	if (!csdev)
 		return;
 
-	sink = coresight_get_sink(event_data->path[cpu]);
+	path = etm_event_cpu_path(event_data, cpu);
+	if (!path)
+		return;
+
+	sink = coresight_get_sink(path);
 	if (!sink)
 		return;
 
@@ -344,7 +363,7 @@ static void etm_event_stop(struct perf_event *event, int mode)
 	}
 
 	/* Disabling the path make its elements available to other sessions */
-	coresight_disable_path(event_data->path[cpu]);
+	coresight_disable_path(path);
 }
 
 static int etm_event_add(struct perf_event *event, int mode)
-- 
2.7.4

