trace.func should be set to the recorded ip of the mcount calling site in the __mcount_loc section to filter the function entries configured through the tracing/set_graph_function interface, but before, this is set to the self_ra(the return address of mcount), which has made set_graph_function not work as expected. This fixes it via calculating the right recorded ip in the __mcount_loc section and assign it to trace.func. Reported-by: Zhiping Zhong <xzhong86@xxxxxxx> Signed-off-by: Wu Zhangjin <wuzhangjin@xxxxxxxxx> --- arch/mips/kernel/ftrace.c | 11 +++++++++-- 1 files changed, 9 insertions(+), 2 deletions(-) diff --git a/arch/mips/kernel/ftrace.c b/arch/mips/kernel/ftrace.c index bc91e4a..94ca2b0 100644 --- a/arch/mips/kernel/ftrace.c +++ b/arch/mips/kernel/ftrace.c @@ -257,7 +257,7 @@ void prepare_ftrace_return(unsigned long *parent_ra_addr, unsigned long self_ra, struct ftrace_graph_ent trace; unsigned long return_hooker = (unsigned long) &return_to_handler; - int faulted; + int faulted, insns; if (unlikely(atomic_read(¤t->tracing_graph_pause))) return; @@ -304,7 +304,14 @@ void prepare_ftrace_return(unsigned long *parent_ra_addr, unsigned long self_ra, return; } - trace.func = self_ra; + /* + * Get the recorded ip of the current mcount calling site in the + * __mcount_loc section, which will be used to filter the function + * entries configured through the tracing/set_graph_function interface. + */ + + insns = in_kernel_space(self_ra) ? 2 : MCOUNT_OFFSET_INSNS + 1; + trace.func = self_ra - (MCOUNT_INSN_SIZE * insns); /* Only trace if the calling function expects to */ if (!ftrace_graph_entry(&trace)) { -- 1.7.1