[PATCHV5 1/3] tracing: add a possibility of exporting function trace to other places instead of ring buffer only

On 30 August 2016 at 22:30, Steven Rostedt [off-list ref] wrote:

On Tue, 30 Aug 2016 16:07:28 +0800
Chunyan Zhang [off-list ref] wrote:

quoted

Currently Function traces can be only exported to ring buffer, this
patch added trace_export concept which can process traces and export
them to a registered destination as an addition to the current only
output of Ftrace - i.e. ring buffer.

In this way, if we want Function traces to be sent to other destination
rather than ring buffer only, we just need to register a new
trace_export and implement its own .commit() callback or just use
'trace_generic_commit()' which this patch also added and hooks up its
own .write() function for writing traces to the storage.

With this patch, only Function trace (trace type is TRACE_FN)
is supported.

Signed-off-by: Chunyan Zhang <redacted>
---
 include/linux/trace.h |  35 ++++++++++++
 kernel/trace/trace.c  | 155 +++++++++++++++++++++++++++++++++++++++++++++++++-
 kernel/trace/trace.h  |   1 +
 3 files changed, 190 insertions(+), 1 deletion(-)
 create mode 100644 include/linux/trace.h

diff --git a/include/linux/trace.h b/include/linux/trace.h
new file mode 100644
index 0000000..30ded92
--- /dev/null
+++ b/include/linux/trace.h

@@ -0,0 +1,35 @@
+#ifndef _LINUX_TRACE_H
+#define _LINUX_TRACE_H
+
+#include <linux/ring_buffer.h>
+struct trace_array;
+
+#ifdef CONFIG_TRACING
+/*
+ * The trace export - an export of Ftrace. The trace_export can process
+ * traces and export them to a registered destination as an addition to
+ * the current only output of Ftrace - i.e. ring buffer.
+ *
+ * If you want traces to be sent to some other place rather than
+ * ring buffer only, just need to register a new trace_export and
+ * implement its own .commit() callback or just directly use
+ * 'trace_generic_commit()' and hooks up its own .write() function
+ * for writing traces to the storage.
+ *
+ * next              - pointer to the next trace_export
+ * commit    - commit the traces to the destination
+ * write     - copy traces which have been delt with ->commit() to
+ *             the destination
+ */
+struct trace_export {
+     struct trace_export __rcu       *next;
+     void (*commit)(struct trace_array *, struct ring_buffer_event *);
+     void (*write)(const char *, unsigned int);
+};
+
+int register_ftrace_export(struct trace_export *export);
+int unregister_ftrace_export(struct trace_export *export);
+
+#endif       /* CONFIG_TRACING */
+
+#endif       /* _LINUX_TRACE_H */

diff --git a/kernel/trace/trace.c b/kernel/trace/trace.c
index dade4c9..3163fa6 100644
--- a/kernel/trace/trace.c
+++ b/kernel/trace/trace.c

@@ -40,6 +40,7 @@
 #include <linux/poll.h>
 #include <linux/nmi.h>
 #include <linux/fs.h>
+#include <linux/trace.h>
 #include <linux/sched/rt.h>

 #include "trace.h"

@@ -2128,6 +2129,155 @@ void trace_buffer_unlock_commit_regs(struct trace_array *tr,
      ftrace_trace_userstack(buffer, flags, pc);
 }

+static DEFINE_STATIC_KEY_FALSE(ftrace_exports_enabled);
+
+static void ftrace_exports_enable(void)
+{
+     static_branch_enable(&ftrace_exports_enabled);
+}
+
+static void ftrace_exports_disable(void)
+{
+     static_branch_disable(&ftrace_exports_enabled);
+}
+
+static size_t trace_size[] = {
+     [TRACE_FN]              = sizeof(struct ftrace_entry),
+     [TRACE_CTX]             = sizeof(struct ctx_switch_entry),
+     [TRACE_WAKE]            = sizeof(struct ctx_switch_entry),
+     [TRACE_STACK]           = sizeof(struct stack_entry),
+     [TRACE_PRINT]           = sizeof(struct print_entry),
+     [TRACE_BPRINT]          = sizeof(struct bprint_entry),
+     [TRACE_MMIO_RW]         = sizeof(struct trace_mmiotrace_rw),
+     [TRACE_MMIO_MAP]        = sizeof(struct trace_mmiotrace_map),
+     [TRACE_BRANCH]          = sizeof(struct trace_branch),
+     [TRACE_GRAPH_RET]       = sizeof(struct ftrace_graph_ret_entry),
+     [TRACE_GRAPH_ENT]       = sizeof(struct ftrace_graph_ent_entry),
+     [TRACE_USER_STACK]      = sizeof(struct userstack_entry),
+     [TRACE_BPUTS]           = sizeof(struct bputs_entry),
+};
+
+static void
+trace_generic_commit(struct trace_array *tr,
+            struct ring_buffer_event *event)
+{
+     struct trace_entry *entry;
+     struct trace_export *export = tr->export;

Why not just pass in the export directly?

quoted

+     unsigned int size = 0;
+
+     entry = ring_buffer_event_data(event);
+
+     size = trace_size[entry->type];

BTW, it would make much more sense if you just did:

        size = ring_buffer_event_length(event);

and get rid of that silly array, as that is prone to bugs if you add a
new type.

quoted

+     if (!size)
+             return;
+
+     if (export && export->write)
+             export->write((char *)entry, size);
+}
+
+static DEFINE_MUTEX(ftrace_export_lock);
+
+static struct trace_export __rcu *ftrace_exports_list __read_mostly;
+
+static inline void
+ftrace_exports(struct trace_array *tr, struct ring_buffer_event *event)
+{
+     struct trace_export *export;
+
+     preempt_disable_notrace();
+
+     for (export = rcu_dereference_raw_notrace(ftrace_exports_list);
+          export && export->commit;
+          export = rcu_dereference_raw_notrace(export->next)) {
+             tr->export = export;
+             export->commit(tr, event);

Um, what's the commit for?

Why not just call the trace_generic_commit() above. Nothing has

The original thought was that allow other export (maybe added in the
future) to implement its own .commit().
But I accept your suggestion since there's no other trace_export so
far rather than stm_ftrace only, the modification should be kept as
simple as it can be.

Thanks for your review,
Chunyan

anything different. But change that name. Maybe call it
"process_export()" or something. And pass the export directly and get
rid of the tr->export.

quoted

+     }
+
+     preempt_enable_notrace();
+}
+
+static inline void
+add_trace_export(struct trace_export **list, struct trace_export *export)
+{
+     rcu_assign_pointer(export->next, *list);
+     /*
+      * We are entering export into the list but another
+      * CPU might be walking that list. We need to make sure
+      * the export->next pointer is valid before another CPU sees
+      * the export pointer included into the list.
+      */
+     rcu_assign_pointer(*list, export);
+}
+
+static inline int
+rm_trace_export(struct trace_export **list, struct trace_export *export)
+{
+     struct trace_export **p;
+
+     for (p = list; *p != NULL; p = &(*p)->next)
+             if (*p == export)
+                     break;
+
+     if (*p != export)
+             return -1;
+
+     rcu_assign_pointer(*p, (*p)->next);
+
+     return 0;
+}
+
+static inline void
+add_ftrace_export(struct trace_export **list, struct trace_export *export)
+{
+     if (*list == NULL)
+             ftrace_exports_enable();
+
+     add_trace_export(list, export);
+}
+
+static inline int
+rm_ftrace_export(struct trace_export **list, struct trace_export *export)
+{
+     int ret;
+
+     ret = rm_trace_export(list, export);
+     if (*list == NULL)
+             ftrace_exports_disable();
+
+     return ret;
+}
+
+int register_ftrace_export(struct trace_export *export)
+{
+     if (WARN_ON_ONCE(!export->write))
+             return -1;
+
+     mutex_lock(&ftrace_export_lock);
+
+     export->commit = trace_generic_commit;
+
+     add_ftrace_export(&ftrace_exports_list, export);
+
+     mutex_unlock(&ftrace_export_lock);
+
+     return 0;
+}
+EXPORT_SYMBOL_GPL(register_ftrace_export);
+
+int unregister_ftrace_export(struct trace_export *export)
+{
+     int ret;
+
+     mutex_lock(&ftrace_export_lock);
+
+     ret = rm_ftrace_export(&ftrace_exports_list, export);
+
+     mutex_unlock(&ftrace_export_lock);
+
+     return ret;
+}
+EXPORT_SYMBOL_GPL(unregister_ftrace_export);
+
 void
 trace_function(struct trace_array *tr,
             unsigned long ip, unsigned long parent_ip, unsigned long flags,

@@ -2146,8 +2296,11 @@ trace_function(struct trace_array *tr,
      entry->ip                       = ip;
      entry->parent_ip                = parent_ip;

-     if (!call_filter_check_discard(call, entry, buffer, event))
+     if (!call_filter_check_discard(call, entry, buffer, event)) {
+             if (static_branch_unlikely(&ftrace_exports_enabled))
+                     ftrace_exports(tr, event);
              __buffer_unlock_commit(buffer, event);

We do the generic commit here.

-- Steve

quoted

+     }
 }

 #ifdef CONFIG_STACKTRACE

diff --git a/kernel/trace/trace.h b/kernel/trace/trace.h
index f783df4..26a3088 100644
--- a/kernel/trace/trace.h
+++ b/kernel/trace/trace.h

@@ -260,6 +260,7 @@ struct trace_array {
      /* function tracing enabled */
      int                     function_enabled;
 #endif
+     struct trace_export     *export;
 };

 enum {