[PATCH 12/14] csky: ftrace call graph supported.

From: guoren
Date: Mon Dec 31 2018 - 10:35:23 EST

From: Guo Ren <ren_guo@xxxxxxxxx>

With csky-gcc -pg -mbacktrace, ftrace call graph supported.

Signed-off-by: Guo Ren <ren_guo@xxxxxxxxx>
---
arch/csky/Kconfig | 1 +
arch/csky/abiv2/mcount.S | 116 ++++++++++++++++++++++++++++++++++++++---
arch/csky/include/asm/ftrace.h | 6 ++-
arch/csky/kernel/ftrace.c | 48 +++++++++++++----
4 files changed, 151 insertions(+), 20 deletions(-)

diff --git a/arch/csky/Kconfig b/arch/csky/Kconfig
index 0b9a290..6e81477 100644
--- a/arch/csky/Kconfig
+++ b/arch/csky/Kconfig
@@ -30,6 +30,7 @@ config CSKY
select GENERIC_SMP_IDLE_THREAD
select HAVE_ARCH_TRACEHOOK
select HAVE_FUNCTION_TRACER
+ select HAVE_FUNCTION_GRAPH_TRACER
select HAVE_GENERIC_DMA_COHERENT
select HAVE_KERNEL_GZIP
select HAVE_KERNEL_LZO
diff --git a/arch/csky/abiv2/mcount.S b/arch/csky/abiv2/mcount.S
index 73377d5..c633379 100644
--- a/arch/csky/abiv2/mcount.S
+++ b/arch/csky/abiv2/mcount.S
@@ -2,23 +2,123 @@
// Copyright (C) 2018 Hangzhou C-SKY Microsystems co.,ltd.

#include <linux/linkage.h>
+#include <asm/ftrace.h>

-ENTRY (_mcount)
- subi sp, 20
+/*
+ * csky-gcc with -pg will put the following asm after prologue:
+ * push r15
+ * jsri _mcount
+ *
+ * stack layout after mcount_enter in _mcount():
+ *
+ * current sp => 0:+-------+
+ * | a0-a3 | -> must save all argument regs
+ * +16:+-------+
+ * | lr | -> _mcount lr (instrumente function's pc)
+ * +20:+-------+
+ * | fp=r8 | -> instrumented function fp
+ * +24:+-------+
+ * | plr | -> instrumented function lr (parent's pc)
+ * +-------+
+ */
+
+.macro mcount_enter
+ subi sp, 24
stw a0, (sp, 0)
stw a1, (sp, 4)
stw a2, (sp, 8)
stw a3, (sp, 12)
stw lr, (sp, 16)
- mov a1, lr
- ldw a0, (sp, 20)
- jsri csky_mcount
+ stw r8, (sp, 20)
+.endm
+
+.macro mcount_exit
ldw a0, (sp, 0)
ldw a1, (sp, 4)
ldw a2, (sp, 8)
ldw a3, (sp, 12)
ldw t1, (sp, 16)
- ldw lr, (sp, 20)
- addi sp, 24
+ ldw r8, (sp, 20)
+ ldw lr, (sp, 24)
+ addi sp, 28
jmp t1
-END (_mcount)
+.endm
+
+.macro save_return_regs
+ subi sp, 16
+ stw a0, (sp, 0)
+ stw a1, (sp, 4)
+ stw a2, (sp, 8)
+ stw a3, (sp, 12)
+.endm
+
+.macro restore_return_regs
+ mov lr, a0
+ ldw a0, (sp, 0)
+ ldw a1, (sp, 4)
+ ldw a2, (sp, 8)
+ ldw a3, (sp, 12)
+ addi sp, 16
+.endm
+
+ENTRY(ftrace_stub)
+ jmp lr
+END(ftrace_stub)
+
+ENTRY(_mcount)
+ mcount_enter
+
+ /* r26 is link register, only used with jsri translation */
+ lrw r26, ftrace_trace_function
+ ldw r26, (r26, 0)
+ lrw a1, ftrace_stub
+ cmpne r26, a1
+ bf skip_ftrace
+
+ mov a0, lr
+ subi a0, MCOUNT_INSN_SIZE
+ ldw a1, (sp, 24)
+
+ jsr r26
+
+#ifndef CONFIG_FUNCTION_GRAPH_TRACER
+skip_ftrace:
+ mcount_exit
+#else
+skip_ftrace:
+ lrw a0, ftrace_graph_return
+ ldw a0, (a0, 0)
+ lrw a1, ftrace_stub
+ cmpne a0, a1
+ bt ftrace_graph_caller
+
+ lrw a0, ftrace_graph_entry
+ ldw a0, (a0, 0)
+ lrw a1, ftrace_graph_entry_stub
+ cmpne a0, a1
+ bt ftrace_graph_caller
+
+ mcount_exit
+#endif
+END(_mcount)
+
+#ifdef CONFIG_FUNCTION_GRAPH_TRACER
+ENTRY(ftrace_graph_caller)
+ mov a0, sp
+ addi a0, 24
+ ldw a1, (sp, 16)
+ subi a1, MCOUNT_INSN_SIZE
+ mov a2, r8
+ lrw r26, prepare_ftrace_return
+ jsr r26
+ mcount_exit
+END(ftrace_graph_caller)
+
+ENTRY(return_to_handler)
+ save_return_regs
+ mov a0, r8
+ jsri ftrace_return_to_handler
+ restore_return_regs
+ jmp lr
+END(return_to_handler)
+#endif
diff --git a/arch/csky/include/asm/ftrace.h b/arch/csky/include/asm/ftrace.h
index 1d22a17..7547c45 100644
--- a/arch/csky/include/asm/ftrace.h
+++ b/arch/csky/include/asm/ftrace.h
@@ -4,6 +4,10 @@
#ifndef __ASM_CSKY_FTRACE_H
#define __ASM_CSKY_FTRACE_H

-extern void _mcount(unsigned long from_pc);
+#define MCOUNT_INSN_SIZE 4
+
+#define HAVE_FUNCTION_GRAPH_FP_TEST
+
+#define HAVE_FUNCTION_GRAPH_RET_ADDR_PTR

#endif /* __ASM_CSKY_FTRACE_H */
diff --git a/arch/csky/kernel/ftrace.c b/arch/csky/kernel/ftrace.c
index ad054f719..274c431 100644
--- a/arch/csky/kernel/ftrace.c
+++ b/arch/csky/kernel/ftrace.c
@@ -4,21 +4,47 @@
#include <linux/ftrace.h>
#include <linux/uaccess.h>

-extern void (*ftrace_trace_function)(unsigned long, unsigned long,
- struct ftrace_ops*, struct pt_regs*);
+#ifdef CONFIG_FUNCTION_GRAPH_TRACER
+void prepare_ftrace_return(unsigned long *parent, unsigned long self_addr,
+ unsigned long frame_pointer)
+{
+ unsigned long return_hooker = (unsigned long)&return_to_handler;
+ unsigned long old;

+ if (unlikely(atomic_read(&current->tracing_graph_pause)))
+ return;

-noinline void __naked ftrace_stub(unsigned long ip, unsigned long parent_ip,
- struct ftrace_ops *op, struct pt_regs *regs)
-{
- asm volatile ("\n");
-}
+ old = *parent;

-noinline void csky_mcount(unsigned long from_pc, unsigned long self_pc)
-{
- if (ftrace_trace_function != ftrace_stub)
- ftrace_trace_function(self_pc, from_pc, NULL, NULL);
+ if (!function_graph_enter(old, self_addr,
+ *(unsigned long *)frame_pointer, parent)) {
+ /*
+ * For csky-gcc function has sub-call:
+ * subi sp, sp, 8
+ * stw r8, (sp, 0)
+ * mov r8, sp
+ * st.w r15, (sp, 0x4)
+ * push r15
+ * jl _mcount
+ * We only need set *parent for resume
+ *
+ * For csky-gcc function has no sub-call:
+ * subi sp, sp, 4
+ * stw r8, (sp, 0)
+ * mov r8, sp
+ * push r15
+ * jl _mcount
+ * We need set *parent and *(frame_pointer + 4) for resume,
+ * because lr is resumed twice.
+ */
+ *parent = return_hooker;
+ frame_pointer += 4;
+ if (*(unsigned long *)frame_pointer == old)
+ *(unsigned long *)frame_pointer = return_hooker;
+ }
}
+#endif

/* _mcount is defined in abi's mcount.S */
+extern void _mcount(void);
EXPORT_SYMBOL(_mcount);
--
2.7.4