2 * Performance events callchain code, extracted from core.c:
4 * Copyright (C) 2008 Thomas Gleixner <tglx@linutronix.de>
5 * Copyright (C) 2008-2011 Red Hat, Inc., Ingo Molnar
6 * Copyright (C) 2008-2011 Red Hat, Inc., Peter Zijlstra <pzijlstr@redhat.com>
7 * Copyright © 2009 Paul Mackerras, IBM Corp. <paulus@au1.ibm.com>
9 * For licensing details see kernel-base/COPYING
12 #include <linux/perf_event.h>
13 #include <linux/slab.h>
16 struct callchain_cpus_entries {
17 struct rcu_head rcu_head;
18 struct perf_callchain_entry *cpu_entries[0];
21 static DEFINE_PER_CPU(int, callchain_recursion[PERF_NR_CONTEXTS]);
22 static atomic_t nr_callchain_events;
23 static DEFINE_MUTEX(callchain_mutex);
24 static struct callchain_cpus_entries *callchain_cpus_entries;
27 __weak void perf_callchain_kernel(struct perf_callchain_entry *entry,
32 __weak void perf_callchain_user(struct perf_callchain_entry *entry,
37 static void release_callchain_buffers_rcu(struct rcu_head *head)
39 struct callchain_cpus_entries *entries;
42 entries = container_of(head, struct callchain_cpus_entries, rcu_head);
44 for_each_possible_cpu(cpu)
45 kfree(entries->cpu_entries[cpu]);
50 static void release_callchain_buffers(void)
52 struct callchain_cpus_entries *entries;
54 entries = callchain_cpus_entries;
55 rcu_assign_pointer(callchain_cpus_entries, NULL);
56 call_rcu(&entries->rcu_head, release_callchain_buffers_rcu);
59 static int alloc_callchain_buffers(void)
63 struct callchain_cpus_entries *entries;
66 * We can't use the percpu allocation API for data that can be
67 * accessed from NMI. Use a temporary manual per cpu allocation
68 * until that gets sorted out.
70 size = offsetof(struct callchain_cpus_entries, cpu_entries[nr_cpu_ids]);
72 entries = kzalloc(size, GFP_KERNEL);
76 size = sizeof(struct perf_callchain_entry) * PERF_NR_CONTEXTS;
78 for_each_possible_cpu(cpu) {
79 entries->cpu_entries[cpu] = kmalloc_node(size, GFP_KERNEL,
81 if (!entries->cpu_entries[cpu])
85 rcu_assign_pointer(callchain_cpus_entries, entries);
90 for_each_possible_cpu(cpu)
91 kfree(entries->cpu_entries[cpu]);
97 int get_callchain_buffers(void)
102 mutex_lock(&callchain_mutex);
104 count = atomic_inc_return(&nr_callchain_events);
105 if (WARN_ON_ONCE(count < 1)) {
111 /* If the allocation failed, give up */
112 if (!callchain_cpus_entries)
117 err = alloc_callchain_buffers();
120 atomic_dec(&nr_callchain_events);
122 mutex_unlock(&callchain_mutex);
127 void put_callchain_buffers(void)
129 if (atomic_dec_and_mutex_lock(&nr_callchain_events, &callchain_mutex)) {
130 release_callchain_buffers();
131 mutex_unlock(&callchain_mutex);
135 static struct perf_callchain_entry *get_callchain_entry(int *rctx)
138 struct callchain_cpus_entries *entries;
140 *rctx = get_recursion_context(this_cpu_ptr(callchain_recursion));
144 entries = rcu_dereference(callchain_cpus_entries);
148 cpu = smp_processor_id();
150 return &entries->cpu_entries[cpu][*rctx];
154 put_callchain_entry(int rctx)
156 put_recursion_context(this_cpu_ptr(callchain_recursion), rctx);
159 struct perf_callchain_entry *
160 perf_callchain(struct perf_event *event, struct pt_regs *regs)
163 struct perf_callchain_entry *entry;
165 int kernel = !event->attr.exclude_callchain_kernel;
166 int user = !event->attr.exclude_callchain_user;
168 if (!kernel && !user)
171 entry = get_callchain_entry(&rctx);
180 if (kernel && !user_mode(regs)) {
181 perf_callchain_store(entry, PERF_CONTEXT_KERNEL);
182 perf_callchain_kernel(entry, regs);
186 if (!user_mode(regs)) {
188 regs = task_pt_regs(current);
195 * Disallow cross-task user callchains.
197 if (event->ctx->task && event->ctx->task != current)
200 perf_callchain_store(entry, PERF_CONTEXT_USER);
201 perf_callchain_user(entry, regs);
206 put_callchain_entry(rctx);