]> git.ipfire.org Git - thirdparty/linux.git/commitdiff
function_graph: Reverse the order of pushing the ret_stack and the callback
authorSteven Rostedt (VMware) <rostedt@goodmis.org>
Tue, 20 Nov 2018 17:40:25 +0000 (12:40 -0500)
committerSteven Rostedt (VMware) <rostedt@goodmis.org>
Wed, 28 Nov 2018 01:31:54 +0000 (20:31 -0500)
The function graph profiler uses the ret_stack to store the "subtime" and
reuse it by nested functions and also on the return. But the current logic
has the profiler callback called before the ret_stack is updated, and it is
just modifying the ret_stack that will later be allocated (it's just lucky
that the "subtime" is not touched when it is allocated).

This could also cause a crash if we are at the end of the ret_stack when
this happens.

By reversing the order of the allocating the ret_stack and then calling the
callbacks attached to a function being traced, the ret_stack entry is no
longer used before it is allocated.

Cc: stable@kernel.org
Fixes: 03274a3ffb449 ("tracing/fgraph: Adjust fgraph depth before calling trace return callback")
Reviewed-by: Masami Hiramatsu <mhiramat@kernel.org>
Signed-off-by: Steven Rostedt (VMware) <rostedt@goodmis.org>
kernel/trace/trace_functions_graph.c

index 4f0d72ae636228e1626cc4118255afdc48593582..2561460d7baf8fc3484938f0ac779c928e626be8 100644 (file)
@@ -188,15 +188,17 @@ int function_graph_enter(unsigned long ret, unsigned long func,
        trace.func = func;
        trace.depth = ++current->curr_ret_depth;
 
-       /* Only trace if the calling function expects to */
-       if (!ftrace_graph_entry(&trace))
-               goto out;
-
        if (ftrace_push_return_trace(ret, func,
                                     frame_pointer, retp))
                goto out;
 
+       /* Only trace if the calling function expects to */
+       if (!ftrace_graph_entry(&trace))
+               goto out_ret;
+
        return 0;
+ out_ret:
+       current->curr_ret_stack--;
  out:
        current->curr_ret_depth--;
        return -EBUSY;