Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / kernel / trace / trace_functions.c
diff --git a/kernel/kernel/trace/trace_functions.c b/kernel/kernel/trace/trace_functions.c
new file mode 100644 (file)
index 0000000..fcd41a1
--- /dev/null
@@ -0,0 +1,689 @@
+/*
+ * ring buffer based function tracer
+ *
+ * Copyright (C) 2007-2008 Steven Rostedt <srostedt@redhat.com>
+ * Copyright (C) 2008 Ingo Molnar <mingo@redhat.com>
+ *
+ * Based on code from the latency_tracer, that is:
+ *
+ *  Copyright (C) 2004-2006 Ingo Molnar
+ *  Copyright (C) 2004 Nadia Yvette Chambers
+ */
+#include <linux/ring_buffer.h>
+#include <linux/debugfs.h>
+#include <linux/uaccess.h>
+#include <linux/ftrace.h>
+#include <linux/slab.h>
+#include <linux/fs.h>
+
+#include "trace.h"
+
+static void tracing_start_function_trace(struct trace_array *tr);
+static void tracing_stop_function_trace(struct trace_array *tr);
+static void
+function_trace_call(unsigned long ip, unsigned long parent_ip,
+                   struct ftrace_ops *op, struct pt_regs *pt_regs);
+static void
+function_stack_trace_call(unsigned long ip, unsigned long parent_ip,
+                         struct ftrace_ops *op, struct pt_regs *pt_regs);
+static struct tracer_flags func_flags;
+
+/* Our option */
+enum {
+       TRACE_FUNC_OPT_STACK    = 0x1,
+};
+
+static int allocate_ftrace_ops(struct trace_array *tr)
+{
+       struct ftrace_ops *ops;
+
+       ops = kzalloc(sizeof(*ops), GFP_KERNEL);
+       if (!ops)
+               return -ENOMEM;
+
+       /* Currently only the non stack verision is supported */
+       ops->func = function_trace_call;
+       ops->flags = FTRACE_OPS_FL_RECURSION_SAFE;
+
+       tr->ops = ops;
+       ops->private = tr;
+       return 0;
+}
+
+
+int ftrace_create_function_files(struct trace_array *tr,
+                                struct dentry *parent)
+{
+       int ret;
+
+       /*
+        * The top level array uses the "global_ops", and the files are
+        * created on boot up.
+        */
+       if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
+               return 0;
+
+       ret = allocate_ftrace_ops(tr);
+       if (ret)
+               return ret;
+
+       ftrace_create_filter_files(tr->ops, parent);
+
+       return 0;
+}
+
+void ftrace_destroy_function_files(struct trace_array *tr)
+{
+       ftrace_destroy_filter_files(tr->ops);
+       kfree(tr->ops);
+       tr->ops = NULL;
+}
+
+static int function_trace_init(struct trace_array *tr)
+{
+       ftrace_func_t func;
+
+       /*
+        * Instance trace_arrays get their ops allocated
+        * at instance creation. Unless it failed
+        * the allocation.
+        */
+       if (!tr->ops)
+               return -ENOMEM;
+
+       /* Currently only the global instance can do stack tracing */
+       if (tr->flags & TRACE_ARRAY_FL_GLOBAL &&
+           func_flags.val & TRACE_FUNC_OPT_STACK)
+               func = function_stack_trace_call;
+       else
+               func = function_trace_call;
+
+       ftrace_init_array_ops(tr, func);
+
+       tr->trace_buffer.cpu = get_cpu();
+       put_cpu();
+
+       tracing_start_cmdline_record();
+       tracing_start_function_trace(tr);
+       return 0;
+}
+
+static void function_trace_reset(struct trace_array *tr)
+{
+       tracing_stop_function_trace(tr);
+       tracing_stop_cmdline_record();
+       ftrace_reset_array_ops(tr);
+}
+
+static void function_trace_start(struct trace_array *tr)
+{
+       tracing_reset_online_cpus(&tr->trace_buffer);
+}
+
+static void
+function_trace_call(unsigned long ip, unsigned long parent_ip,
+                   struct ftrace_ops *op, struct pt_regs *pt_regs)
+{
+       struct trace_array *tr = op->private;
+       struct trace_array_cpu *data;
+       unsigned long flags;
+       int bit;
+       int cpu;
+       int pc;
+
+       if (unlikely(!tr->function_enabled))
+               return;
+
+       pc = preempt_count();
+       preempt_disable_notrace();
+
+       bit = trace_test_and_set_recursion(TRACE_FTRACE_START, TRACE_FTRACE_MAX);
+       if (bit < 0)
+               goto out;
+
+       cpu = smp_processor_id();
+       data = per_cpu_ptr(tr->trace_buffer.data, cpu);
+       if (!atomic_read(&data->disabled)) {
+               local_save_flags(flags);
+               trace_function(tr, ip, parent_ip, flags, pc);
+       }
+       trace_clear_recursion(bit);
+
+ out:
+       preempt_enable_notrace();
+}
+
+static void
+function_stack_trace_call(unsigned long ip, unsigned long parent_ip,
+                         struct ftrace_ops *op, struct pt_regs *pt_regs)
+{
+       struct trace_array *tr = op->private;
+       struct trace_array_cpu *data;
+       unsigned long flags;
+       long disabled;
+       int cpu;
+       int pc;
+
+       if (unlikely(!tr->function_enabled))
+               return;
+
+       /*
+        * Need to use raw, since this must be called before the
+        * recursive protection is performed.
+        */
+       local_irq_save(flags);
+       cpu = raw_smp_processor_id();
+       data = per_cpu_ptr(tr->trace_buffer.data, cpu);
+       disabled = atomic_inc_return(&data->disabled);
+
+       if (likely(disabled == 1)) {
+               pc = preempt_count();
+               trace_function(tr, ip, parent_ip, flags, pc);
+               /*
+                * skip over 5 funcs:
+                *    __ftrace_trace_stack,
+                *    __trace_stack,
+                *    function_stack_trace_call
+                *    ftrace_list_func
+                *    ftrace_call
+                */
+               __trace_stack(tr, flags, 5, pc);
+       }
+
+       atomic_dec(&data->disabled);
+       local_irq_restore(flags);
+}
+
+static struct tracer_opt func_opts[] = {
+#ifdef CONFIG_STACKTRACE
+       { TRACER_OPT(func_stack_trace, TRACE_FUNC_OPT_STACK) },
+#endif
+       { } /* Always set a last empty entry */
+};
+
+static struct tracer_flags func_flags = {
+       .val = 0, /* By default: all flags disabled */
+       .opts = func_opts
+};
+
+static void tracing_start_function_trace(struct trace_array *tr)
+{
+       tr->function_enabled = 0;
+       register_ftrace_function(tr->ops);
+       tr->function_enabled = 1;
+}
+
+static void tracing_stop_function_trace(struct trace_array *tr)
+{
+       tr->function_enabled = 0;
+       unregister_ftrace_function(tr->ops);
+}
+
+static int
+func_set_flag(struct trace_array *tr, u32 old_flags, u32 bit, int set)
+{
+       switch (bit) {
+       case TRACE_FUNC_OPT_STACK:
+               /* do nothing if already set */
+               if (!!set == !!(func_flags.val & TRACE_FUNC_OPT_STACK))
+                       break;
+
+               unregister_ftrace_function(tr->ops);
+
+               if (set) {
+                       tr->ops->func = function_stack_trace_call;
+                       register_ftrace_function(tr->ops);
+               } else {
+                       tr->ops->func = function_trace_call;
+                       register_ftrace_function(tr->ops);
+               }
+
+               break;
+       default:
+               return -EINVAL;
+       }
+
+       return 0;
+}
+
+static struct tracer function_trace __tracer_data =
+{
+       .name           = "function",
+       .init           = function_trace_init,
+       .reset          = function_trace_reset,
+       .start          = function_trace_start,
+       .flags          = &func_flags,
+       .set_flag       = func_set_flag,
+       .allow_instances = true,
+#ifdef CONFIG_FTRACE_SELFTEST
+       .selftest       = trace_selftest_startup_function,
+#endif
+};
+
+#ifdef CONFIG_DYNAMIC_FTRACE
+static void update_traceon_count(void **data, bool on)
+{
+       long *count = (long *)data;
+       long old_count = *count;
+
+       /*
+        * Tracing gets disabled (or enabled) once per count.
+        * This function can be called at the same time on multiple CPUs.
+        * It is fine if both disable (or enable) tracing, as disabling
+        * (or enabling) the second time doesn't do anything as the
+        * state of the tracer is already disabled (or enabled).
+        * What needs to be synchronized in this case is that the count
+        * only gets decremented once, even if the tracer is disabled
+        * (or enabled) twice, as the second one is really a nop.
+        *
+        * The memory barriers guarantee that we only decrement the
+        * counter once. First the count is read to a local variable
+        * and a read barrier is used to make sure that it is loaded
+        * before checking if the tracer is in the state we want.
+        * If the tracer is not in the state we want, then the count
+        * is guaranteed to be the old count.
+        *
+        * Next the tracer is set to the state we want (disabled or enabled)
+        * then a write memory barrier is used to make sure that
+        * the new state is visible before changing the counter by
+        * one minus the old counter. This guarantees that another CPU
+        * executing this code will see the new state before seeing
+        * the new counter value, and would not do anything if the new
+        * counter is seen.
+        *
+        * Note, there is no synchronization between this and a user
+        * setting the tracing_on file. But we currently don't care
+        * about that.
+        */
+       if (!old_count)
+               return;
+
+       /* Make sure we see count before checking tracing state */
+       smp_rmb();
+
+       if (on == !!tracing_is_on())
+               return;
+
+       if (on)
+               tracing_on();
+       else
+               tracing_off();
+
+       /* unlimited? */
+       if (old_count == -1)
+               return;
+
+       /* Make sure tracing state is visible before updating count */
+       smp_wmb();
+
+       *count = old_count - 1;
+}
+
+static void
+ftrace_traceon_count(unsigned long ip, unsigned long parent_ip, void **data)
+{
+       update_traceon_count(data, 1);
+}
+
+static void
+ftrace_traceoff_count(unsigned long ip, unsigned long parent_ip, void **data)
+{
+       update_traceon_count(data, 0);
+}
+
+static void
+ftrace_traceon(unsigned long ip, unsigned long parent_ip, void **data)
+{
+       if (tracing_is_on())
+               return;
+
+       tracing_on();
+}
+
+static void
+ftrace_traceoff(unsigned long ip, unsigned long parent_ip, void **data)
+{
+       if (!tracing_is_on())
+               return;
+
+       tracing_off();
+}
+
+/*
+ * Skip 4:
+ *   ftrace_stacktrace()
+ *   function_trace_probe_call()
+ *   ftrace_ops_list_func()
+ *   ftrace_call()
+ */
+#define STACK_SKIP 4
+
+static void
+ftrace_stacktrace(unsigned long ip, unsigned long parent_ip, void **data)
+{
+       trace_dump_stack(STACK_SKIP);
+}
+
+static void
+ftrace_stacktrace_count(unsigned long ip, unsigned long parent_ip, void **data)
+{
+       long *count = (long *)data;
+       long old_count;
+       long new_count;
+
+       /*
+        * Stack traces should only execute the number of times the
+        * user specified in the counter.
+        */
+       do {
+
+               if (!tracing_is_on())
+                       return;
+
+               old_count = *count;
+
+               if (!old_count)
+                       return;
+
+               /* unlimited? */
+               if (old_count == -1) {
+                       trace_dump_stack(STACK_SKIP);
+                       return;
+               }
+
+               new_count = old_count - 1;
+               new_count = cmpxchg(count, old_count, new_count);
+               if (new_count == old_count)
+                       trace_dump_stack(STACK_SKIP);
+
+       } while (new_count != old_count);
+}
+
+static int update_count(void **data)
+{
+       unsigned long *count = (long *)data;
+
+       if (!*count)
+               return 0;
+
+       if (*count != -1)
+               (*count)--;
+
+       return 1;
+}
+
+static void
+ftrace_dump_probe(unsigned long ip, unsigned long parent_ip, void **data)
+{
+       if (update_count(data))
+               ftrace_dump(DUMP_ALL);
+}
+
+/* Only dump the current CPU buffer. */
+static void
+ftrace_cpudump_probe(unsigned long ip, unsigned long parent_ip, void **data)
+{
+       if (update_count(data))
+               ftrace_dump(DUMP_ORIG);
+}
+
+static int
+ftrace_probe_print(const char *name, struct seq_file *m,
+                  unsigned long ip, void *data)
+{
+       long count = (long)data;
+
+       seq_printf(m, "%ps:%s", (void *)ip, name);
+
+       if (count == -1)
+               seq_puts(m, ":unlimited\n");
+       else
+               seq_printf(m, ":count=%ld\n", count);
+
+       return 0;
+}
+
+static int
+ftrace_traceon_print(struct seq_file *m, unsigned long ip,
+                        struct ftrace_probe_ops *ops, void *data)
+{
+       return ftrace_probe_print("traceon", m, ip, data);
+}
+
+static int
+ftrace_traceoff_print(struct seq_file *m, unsigned long ip,
+                        struct ftrace_probe_ops *ops, void *data)
+{
+       return ftrace_probe_print("traceoff", m, ip, data);
+}
+
+static int
+ftrace_stacktrace_print(struct seq_file *m, unsigned long ip,
+                       struct ftrace_probe_ops *ops, void *data)
+{
+       return ftrace_probe_print("stacktrace", m, ip, data);
+}
+
+static int
+ftrace_dump_print(struct seq_file *m, unsigned long ip,
+                       struct ftrace_probe_ops *ops, void *data)
+{
+       return ftrace_probe_print("dump", m, ip, data);
+}
+
+static int
+ftrace_cpudump_print(struct seq_file *m, unsigned long ip,
+                       struct ftrace_probe_ops *ops, void *data)
+{
+       return ftrace_probe_print("cpudump", m, ip, data);
+}
+
+static struct ftrace_probe_ops traceon_count_probe_ops = {
+       .func                   = ftrace_traceon_count,
+       .print                  = ftrace_traceon_print,
+};
+
+static struct ftrace_probe_ops traceoff_count_probe_ops = {
+       .func                   = ftrace_traceoff_count,
+       .print                  = ftrace_traceoff_print,
+};
+
+static struct ftrace_probe_ops stacktrace_count_probe_ops = {
+       .func                   = ftrace_stacktrace_count,
+       .print                  = ftrace_stacktrace_print,
+};
+
+static struct ftrace_probe_ops dump_probe_ops = {
+       .func                   = ftrace_dump_probe,
+       .print                  = ftrace_dump_print,
+};
+
+static struct ftrace_probe_ops cpudump_probe_ops = {
+       .func                   = ftrace_cpudump_probe,
+       .print                  = ftrace_cpudump_print,
+};
+
+static struct ftrace_probe_ops traceon_probe_ops = {
+       .func                   = ftrace_traceon,
+       .print                  = ftrace_traceon_print,
+};
+
+static struct ftrace_probe_ops traceoff_probe_ops = {
+       .func                   = ftrace_traceoff,
+       .print                  = ftrace_traceoff_print,
+};
+
+static struct ftrace_probe_ops stacktrace_probe_ops = {
+       .func                   = ftrace_stacktrace,
+       .print                  = ftrace_stacktrace_print,
+};
+
+static int
+ftrace_trace_probe_callback(struct ftrace_probe_ops *ops,
+                           struct ftrace_hash *hash, char *glob,
+                           char *cmd, char *param, int enable)
+{
+       void *count = (void *)-1;
+       char *number;
+       int ret;
+
+       /* hash funcs only work with set_ftrace_filter */
+       if (!enable)
+               return -EINVAL;
+
+       if (glob[0] == '!') {
+               unregister_ftrace_function_probe_func(glob+1, ops);
+               return 0;
+       }
+
+       if (!param)
+               goto out_reg;
+
+       number = strsep(&param, ":");
+
+       if (!strlen(number))
+               goto out_reg;
+
+       /*
+        * We use the callback data field (which is a pointer)
+        * as our counter.
+        */
+       ret = kstrtoul(number, 0, (unsigned long *)&count);
+       if (ret)
+               return ret;
+
+ out_reg:
+       ret = register_ftrace_function_probe(glob, ops, count);
+
+       return ret < 0 ? ret : 0;
+}
+
+static int
+ftrace_trace_onoff_callback(struct ftrace_hash *hash,
+                           char *glob, char *cmd, char *param, int enable)
+{
+       struct ftrace_probe_ops *ops;
+
+       /* we register both traceon and traceoff to this callback */
+       if (strcmp(cmd, "traceon") == 0)
+               ops = param ? &traceon_count_probe_ops : &traceon_probe_ops;
+       else
+               ops = param ? &traceoff_count_probe_ops : &traceoff_probe_ops;
+
+       return ftrace_trace_probe_callback(ops, hash, glob, cmd,
+                                          param, enable);
+}
+
+static int
+ftrace_stacktrace_callback(struct ftrace_hash *hash,
+                          char *glob, char *cmd, char *param, int enable)
+{
+       struct ftrace_probe_ops *ops;
+
+       ops = param ? &stacktrace_count_probe_ops : &stacktrace_probe_ops;
+
+       return ftrace_trace_probe_callback(ops, hash, glob, cmd,
+                                          param, enable);
+}
+
+static int
+ftrace_dump_callback(struct ftrace_hash *hash,
+                          char *glob, char *cmd, char *param, int enable)
+{
+       struct ftrace_probe_ops *ops;
+
+       ops = &dump_probe_ops;
+
+       /* Only dump once. */
+       return ftrace_trace_probe_callback(ops, hash, glob, cmd,
+                                          "1", enable);
+}
+
+static int
+ftrace_cpudump_callback(struct ftrace_hash *hash,
+                          char *glob, char *cmd, char *param, int enable)
+{
+       struct ftrace_probe_ops *ops;
+
+       ops = &cpudump_probe_ops;
+
+       /* Only dump once. */
+       return ftrace_trace_probe_callback(ops, hash, glob, cmd,
+                                          "1", enable);
+}
+
+static struct ftrace_func_command ftrace_traceon_cmd = {
+       .name                   = "traceon",
+       .func                   = ftrace_trace_onoff_callback,
+};
+
+static struct ftrace_func_command ftrace_traceoff_cmd = {
+       .name                   = "traceoff",
+       .func                   = ftrace_trace_onoff_callback,
+};
+
+static struct ftrace_func_command ftrace_stacktrace_cmd = {
+       .name                   = "stacktrace",
+       .func                   = ftrace_stacktrace_callback,
+};
+
+static struct ftrace_func_command ftrace_dump_cmd = {
+       .name                   = "dump",
+       .func                   = ftrace_dump_callback,
+};
+
+static struct ftrace_func_command ftrace_cpudump_cmd = {
+       .name                   = "cpudump",
+       .func                   = ftrace_cpudump_callback,
+};
+
+static int __init init_func_cmd_traceon(void)
+{
+       int ret;
+
+       ret = register_ftrace_command(&ftrace_traceoff_cmd);
+       if (ret)
+               return ret;
+
+       ret = register_ftrace_command(&ftrace_traceon_cmd);
+       if (ret)
+               goto out_free_traceoff;
+
+       ret = register_ftrace_command(&ftrace_stacktrace_cmd);
+       if (ret)
+               goto out_free_traceon;
+
+       ret = register_ftrace_command(&ftrace_dump_cmd);
+       if (ret)
+               goto out_free_stacktrace;
+
+       ret = register_ftrace_command(&ftrace_cpudump_cmd);
+       if (ret)
+               goto out_free_dump;
+
+       return 0;
+
+ out_free_dump:
+       unregister_ftrace_command(&ftrace_dump_cmd);
+ out_free_stacktrace:
+       unregister_ftrace_command(&ftrace_stacktrace_cmd);
+ out_free_traceon:
+       unregister_ftrace_command(&ftrace_traceon_cmd);
+ out_free_traceoff:
+       unregister_ftrace_command(&ftrace_traceoff_cmd);
+
+       return ret;
+}
+#else
+static inline int init_func_cmd_traceon(void)
+{
+       return 0;
+}
+#endif /* CONFIG_DYNAMIC_FTRACE */
+
+static __init int init_function_trace(void)
+{
+       init_func_cmd_traceon();
+       return register_tracer(&function_trace);
+}
+core_initcall(init_function_trace);