diff mbox

[v3,1/6] arm64: Add ftrace support

Message ID 1391768336-5642-2-git-send-email-takahiro.akashi@linaro.org
State New
Headers show

Commit Message

AKASHI Takahiro Feb. 7, 2014, 10:18 a.m. UTC
This patch implements arm64 specific part to support function tracers,
such as function (CONFIG_FUNCTION_TRACER), function_graph
(CONFIG_FUNCTION_GRAPH_TRACER) and function profiler
(CONFIG_FUNCTION_PROFILER).

With 'function' tracer, all the functions in the kernel are traced with
timestamps in ${sysfs}/tracing/trace. If function_graph tracer is
specified, call graph is generated.

The kernel must be compiled with -pg option so that _mcount() is inserted
at the beginning of functions. This function is called on every function's
entry as long as tracing is enabled.
In addition, function_graph tracer also needs to be able to probe function's
exit. ftrace_graph_caller() & return_to_handler do this by faking link
register's value to intercept function's return path.

More details on architecture specific requirements are described in
Documentation/trace/ftrace-design.txt.

Signed-off-by: AKASHI Takahiro <takahiro.akashi@linaro.org>
---
 arch/arm64/Kconfig               |    2 +
 arch/arm64/include/asm/ftrace.h  |   23 +++++
 arch/arm64/kernel/Makefile       |    6 ++
 arch/arm64/kernel/arm64ksyms.c   |    4 +
 arch/arm64/kernel/entry-ftrace.S |  172 ++++++++++++++++++++++++++++++++++++++
 arch/arm64/kernel/ftrace.c       |   64 ++++++++++++++
 6 files changed, 271 insertions(+)
 create mode 100644 arch/arm64/include/asm/ftrace.h
 create mode 100644 arch/arm64/kernel/entry-ftrace.S
 create mode 100644 arch/arm64/kernel/ftrace.c

Comments

AKASHI Takahiro Feb. 12, 2014, 6:42 a.m. UTC | #1
On 11 February 2014 00:03, Arnd Bergmann <arnd@arndb.de> wrote:
>
> On Friday 07 February 2014, AKASHI Takahiro wrote:
> > @@ -0,0 +1,23 @@
>
> > +
> > +#ifdef CONFIG_FUNCTION_TRACER
> > +#define MCOUNT_ADDR          ((unsigned long)_mcount)
> > +#define MCOUNT_INSN_SIZE     4 /* sizeof mcount call */
> > +
> > +#ifndef __ASSEMBLY__
> > +extern void _mcount(unsigned long);
> > +#endif /* __ASSEMBLY__ */
> > +#endif /* CONFIG_FUNCTION_TRACER */
>
> We generally like to have as few #ifdef as possible, and I think the
> #ifdef CONFIG_FUNCTION_TRACER here is not needed.

I will remove it.

> > +#endif /* __ASM_FTRACE_H */
> > diff --git a/arch/arm64/kernel/Makefile b/arch/arm64/kernel/Makefile
> > index 2d4554b..ca921fb 100644
> > --- a/arch/arm64/kernel/Makefile
> > +++ b/arch/arm64/kernel/Makefile
> > @@ -5,6 +5,11 @@
> >  CPPFLAGS_vmlinux.lds := -DTEXT_OFFSET=$(TEXT_OFFSET)
> >  AFLAGS_head.o                := -DTEXT_OFFSET=$(TEXT_OFFSET)
> >
> > +ifdef CONFIG_FUNCTION_TRACER
> > +CFLAGS_REMOVE_ftrace.o = -pg
> > +CFLAGS_REMOVE_insn.o = -pg
> > +endif
>
> Same here. It never hurts to have the CFLAGS_REMOVE_* statements here,
since
> you will not want to build these files for profiling.

I will remove it.

> > diff --git a/arch/arm64/kernel/arm64ksyms.c
b/arch/arm64/kernel/arm64ksyms.c
> > index 338b568..7f0512f 100644
> > --- a/arch/arm64/kernel/arm64ksyms.c
> > +++ b/arch/arm64/kernel/arm64ksyms.c
> > @@ -56,3 +56,7 @@ EXPORT_SYMBOL(clear_bit);
> >  EXPORT_SYMBOL(test_and_clear_bit);
> >  EXPORT_SYMBOL(change_bit);
> >  EXPORT_SYMBOL(test_and_change_bit);
> > +
> > +#ifdef CONFIG_FUNCTION_TRACER
> > +EXPORT_SYMBOL(_mcount);
> > +#endif
>
> This one is clearly needed of course.

Sure.

> > +/*
> > + * Gcc with -pg will put the following code in the beginning of each
function:
> > + *      mov x0, x30
> > + *      bl _mcount
> > + *   [function's body ...]
> > + * "bl _mcount" may be replaced to "bl ftrace_caller" or NOP if dynamic
> > + * ftrace is enabled.
>
> Unrelated to your patch, I have run into problems with this on arm32,
maybe
> someone on Cc to this mail has an idea what to do about it:
>
> If I build a large "randconfig" kernel or "allyesconfig", the kernel image
> size exceeds 32MB, which means I can no longer link callers with a 26
> bit signed immediate argument like the "bl _mcount" here. For any other
> function calls, "gcc -mlong-calls" can be used to work around this, but
> this particular instance is created by inserting assembly statements
> into the output without considering the long-call options. Is there
> a way to get the kernel to link anyway?

I hope that this won't happen on arm64 because a branch offset range is
+/- 128MB.

> > +#ifdef CONFIG_FUNCTION_GRAPH_TRACER
>
> Here, again, you an remove the #ifdef by making the entire file built
> only for CONFIG_FUNCTION_GRAPH_TRACER.

You're right as far as patch [1/6] is concerned, but once patch [2/6] is
applied,
this file (ftrace.c) may be compiled even if CONFIG_FUNCTION_GRAPH_TRACER
is not enabled. (That is CONFIG_DYNAMIC_FTRACE.)
Sorry for any confusion by poorly-divided patchset.

-Takahiro AKASHI

>         Arnd
AKASHI Takahiro Feb. 19, 2014, 10:46 a.m. UTC | #2
On 02/18/2014 03:12 AM, Will Deacon wrote:
> On Fri, Feb 07, 2014 at 10:18:51AM +0000, AKASHI Takahiro wrote:
>> This patch implements arm64 specific part to support function tracers,
>> such as function (CONFIG_FUNCTION_TRACER), function_graph
>> (CONFIG_FUNCTION_GRAPH_TRACER) and function profiler
>> (CONFIG_FUNCTION_PROFILER).
>>
>> With 'function' tracer, all the functions in the kernel are traced with
>> timestamps in ${sysfs}/tracing/trace. If function_graph tracer is
>> specified, call graph is generated.
>>
>> The kernel must be compiled with -pg option so that _mcount() is inserted
>> at the beginning of functions. This function is called on every function's
>> entry as long as tracing is enabled.
>> In addition, function_graph tracer also needs to be able to probe function's
>> exit. ftrace_graph_caller() & return_to_handler do this by faking link
>> register's value to intercept function's return path.
>>
>> More details on architecture specific requirements are described in
>> Documentation/trace/ftrace-design.txt.
>
> [...]
>
>> diff --git a/arch/arm64/include/asm/ftrace.h b/arch/arm64/include/asm/ftrace.h
>> new file mode 100644
>> index 0000000..0d5dfdb
>> --- /dev/null
>> +++ b/arch/arm64/include/asm/ftrace.h
>> @@ -0,0 +1,23 @@
>> +/*
>> + * arch/arm64/include/asm/ftrace.h
>> + *
>> + * Copyright (C) 2013 Linaro Limited
>> + * Author: AKASHI Takahiro <takahiro.akashi@linaro.org>
>> + *
>> + * This program is free software; you can redistribute it and/or modify
>> + * it under the terms of the GNU General Public License version 2 as
>> + * published by the Free Software Foundation.
>> + */
>> +#ifndef __ASM_FTRACE_H
>> +#define __ASM_FTRACE_H
>> +
>> +#ifdef CONFIG_FUNCTION_TRACER
>> +#define MCOUNT_ADDR            ((unsigned long)_mcount)
>> +#define MCOUNT_INSN_SIZE       4 /* sizeof mcount call */
>
> You can use AARCH64_INSN_SIZE here.

OK, but MCOUNT_INSN_SIZE can't be removed here because it is also used in
kernel/trace/ftrace.c. So I will redefine it as
#define MCOUNT_INSN_SIZE AARCH64_INSN_SIZE

In this case, I need to add __ASSEMBLY__ in asm/insn.h because
asm/ftrace.h (and so asm/insn.h) is included in entry-ftrace.S.

Is it OK?


>> +#include <linux/linkage.h>
>> +#include <asm/ftrace.h>
>> +
>> +/*
>> + * Gcc with -pg will put the following code in the beginning of each function:
>> + *      mov x0, x30
>> + *      bl _mcount
>> + *     [function's body ...]
>> + * "bl _mcount" may be replaced to "bl ftrace_caller" or NOP if dynamic
>> + * ftrace is enabled.
>> + *
>> + * Please note that x0 as an argument will not be used here because we can
>> + * get lr(x30) of insturmented function at any time by winding up call stack
>
> instrumented

I  will fix two occurences of "insturmented" :-)
Just a question: Is there any better word than "instrument"?

>> + * as long as the kernel is compiled without -fomit-frame-pointer.
>> + * (or CONFIG_FRAME_POINTER, this is forced on arm64)
>> + *
>> + * stack layout after mcount_enter:
>> + *
>> + *   0 ---------------------------------------- sp of _mcount()
>> + *      x29: fp of instrumented function       fp is not winded
>> + *     --------------------
>> + *      x30: lr of _mcount() (= instrumented function's pc)
>> + * +16 ---------------------------------------- sp of instrumented function
>> + *
>> + *     ....
>> + *
>> + * +xx ---------------------------------------- fp of instrumented function
>> + *      x29: fp of parent
>> + *     --------------------
>> + *      x30: lr of insturmented function (= parent's pc)
>> + *     --------------------
>> + *      xxx
>> + */
>> +
>> +       .macro mcount_enter
>> +       stp     x29, x30, [sp, #-48]!
>> +       .endm
>
> Can you elaborate in your comment about where this 48 comes from please? I
> can't join it up with your ascii art.

Right. It should be -16 as shown in the stack layout.
When I removed an operation of saving/restoring x0-x3 at v3 patch, this value
should also have been changed.

-Takahiro AKASHI

> Will
>
diff mbox

Patch

diff --git a/arch/arm64/Kconfig b/arch/arm64/Kconfig
index dd4327f..332e50c 100644
--- a/arch/arm64/Kconfig
+++ b/arch/arm64/Kconfig
@@ -33,6 +33,8 @@  config ARM64
 	select HAVE_DMA_ATTRS
 	select HAVE_DMA_CONTIGUOUS
 	select HAVE_EFFICIENT_UNALIGNED_ACCESS
+	select HAVE_FUNCTION_TRACER
+	select HAVE_FUNCTION_GRAPH_TRACER
 	select HAVE_GENERIC_DMA_COHERENT
 	select HAVE_HW_BREAKPOINT if PERF_EVENTS
 	select HAVE_MEMBLOCK
diff --git a/arch/arm64/include/asm/ftrace.h b/arch/arm64/include/asm/ftrace.h
new file mode 100644
index 0000000..0d5dfdb
--- /dev/null
+++ b/arch/arm64/include/asm/ftrace.h
@@ -0,0 +1,23 @@ 
+/*
+ * arch/arm64/include/asm/ftrace.h
+ *
+ * Copyright (C) 2013 Linaro Limited
+ * Author: AKASHI Takahiro <takahiro.akashi@linaro.org>
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+#ifndef __ASM_FTRACE_H
+#define __ASM_FTRACE_H
+
+#ifdef CONFIG_FUNCTION_TRACER
+#define MCOUNT_ADDR		((unsigned long)_mcount)
+#define MCOUNT_INSN_SIZE	4 /* sizeof mcount call */
+
+#ifndef __ASSEMBLY__
+extern void _mcount(unsigned long);
+#endif /* __ASSEMBLY__ */
+#endif /* CONFIG_FUNCTION_TRACER */
+
+#endif /* __ASM_FTRACE_H */
diff --git a/arch/arm64/kernel/Makefile b/arch/arm64/kernel/Makefile
index 2d4554b..ca921fb 100644
--- a/arch/arm64/kernel/Makefile
+++ b/arch/arm64/kernel/Makefile
@@ -5,6 +5,11 @@ 
 CPPFLAGS_vmlinux.lds	:= -DTEXT_OFFSET=$(TEXT_OFFSET)
 AFLAGS_head.o		:= -DTEXT_OFFSET=$(TEXT_OFFSET)
 
+ifdef CONFIG_FUNCTION_TRACER
+CFLAGS_REMOVE_ftrace.o = -pg
+CFLAGS_REMOVE_insn.o = -pg
+endif
+
 # Object file lists.
 arm64-obj-y		:= cputable.o debug-monitors.o entry.o irq.o fpsimd.o	\
 			   entry-fpsimd.o process.o ptrace.o setup.o signal.o	\
@@ -13,6 +18,7 @@  arm64-obj-y		:= cputable.o debug-monitors.o entry.o irq.o fpsimd.o	\
 
 arm64-obj-$(CONFIG_COMPAT)		+= sys32.o kuser32.o signal32.o 	\
 					   sys_compat.o
+arm64-obj-$(CONFIG_FUNCTION_TRACER)	+= ftrace.o entry-ftrace.o
 arm64-obj-$(CONFIG_MODULES)		+= arm64ksyms.o module.o
 arm64-obj-$(CONFIG_SMP)			+= smp.o smp_spin_table.o
 arm64-obj-$(CONFIG_HW_PERF_EVENTS)	+= perf_event.o
diff --git a/arch/arm64/kernel/arm64ksyms.c b/arch/arm64/kernel/arm64ksyms.c
index 338b568..7f0512f 100644
--- a/arch/arm64/kernel/arm64ksyms.c
+++ b/arch/arm64/kernel/arm64ksyms.c
@@ -56,3 +56,7 @@  EXPORT_SYMBOL(clear_bit);
 EXPORT_SYMBOL(test_and_clear_bit);
 EXPORT_SYMBOL(change_bit);
 EXPORT_SYMBOL(test_and_change_bit);
+
+#ifdef CONFIG_FUNCTION_TRACER
+EXPORT_SYMBOL(_mcount);
+#endif
diff --git a/arch/arm64/kernel/entry-ftrace.S b/arch/arm64/kernel/entry-ftrace.S
new file mode 100644
index 0000000..065f426
--- /dev/null
+++ b/arch/arm64/kernel/entry-ftrace.S
@@ -0,0 +1,172 @@ 
+/*
+ * arch/arm64/kernel/entry-ftrace.S
+ *
+ * Copyright (C) 2013 Linaro Limited
+ * Author: AKASHI Takahiro <takahiro.akashi@linaro.org>
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#include <linux/linkage.h>
+#include <asm/ftrace.h>
+
+/*
+ * Gcc with -pg will put the following code in the beginning of each function:
+ *      mov x0, x30
+ *      bl _mcount
+ *	[function's body ...]
+ * "bl _mcount" may be replaced to "bl ftrace_caller" or NOP if dynamic
+ * ftrace is enabled.
+ *
+ * Please note that x0 as an argument will not be used here because we can
+ * get lr(x30) of insturmented function at any time by winding up call stack
+ * as long as the kernel is compiled without -fomit-frame-pointer.
+ * (or CONFIG_FRAME_POINTER, this is forced on arm64)
+ *
+ * stack layout after mcount_enter:
+ *
+ *   0 ---------------------------------------- sp of _mcount()
+ *      x29: fp of instrumented function	fp is not winded
+ *     --------------------
+ *      x30: lr of _mcount() (= instrumented function's pc)
+ * +16 ---------------------------------------- sp of instrumented function
+ *
+ *     ....
+ *
+ * +xx ---------------------------------------- fp of instrumented function
+ *      x29: fp of parent
+ *     --------------------
+ *      x30: lr of insturmented function (= parent's pc)
+ *     --------------------
+ *      xxx
+ */
+
+	.macro mcount_enter
+	stp	x29, x30, [sp, #-48]!
+	.endm
+
+	.macro mcount_exit
+	ldp	x29, x30, [sp], #48
+	ret
+	.endm
+
+	.macro mcount_adjust_addr rd, rn
+	sub	\rd, \rn, #MCOUNT_INSN_SIZE
+	.endm
+
+	/* for instrumented function's parent */
+	.macro mcount_get_parent_fp reg
+	ldr	\reg, [sp]
+	ldr	\reg, [\reg]
+	.endm
+
+	/* for instrumented function */
+	.macro mcount_get_pc0 reg
+	mcount_adjust_addr	\reg, x30
+	.endm
+
+	.macro mcount_get_pc reg
+	ldr	\reg, [sp, #8]
+	mcount_adjust_addr	\reg, \reg
+	.endm
+
+	.macro mcount_get_lr reg
+	ldr	\reg, [sp]
+	ldr	\reg, [\reg, #8]
+	mcount_adjust_addr	\reg, \reg
+	.endm
+
+	.macro mcount_get_saved_lr_addr reg
+	ldr	\reg, [sp]
+	add	\reg, \reg, #8
+	.endm
+
+/*
+ * void _mcount(unsigned long return_address)
+ * @return_address: return address to instrumented function
+ *
+ * This function makes calls, if enabled, to:
+ *     - tracer function to probe instrumented function's entry,
+ *     - ftrace_graph_caller to set up an exit hook
+ */
+ENTRY(_mcount)
+#ifdef CONFIG_HAVE_FUNCTION_TRACE_MCOUNT_TEST
+	ldr	x0, =ftrace_trace_stop
+	ldr	x0, [x0]		// if ftrace_trace_stop
+	ret				//   return;
+#endif
+	mcount_enter
+
+	ldr	x0, =ftrace_trace_function
+	ldr	x2, [x0]
+	adr	x0, ftrace_stub
+	cmp	x0, x2			// if (ftrace_trace_function
+	b.eq	skip_ftrace_call	//     != ftrace_stub) {
+
+	mcount_get_pc	x0		//       function's pc
+	mcount_get_lr	x1		//       function's lr (= parent's pc)
+	blr	x2			//   (*ftrace_trace_function)(pc, lr);
+
+#ifndef CONFIG_FUNCTION_GRAPH_TRACER
+skip_ftrace_call:			//   return;
+	mcount_exit			// }
+#else
+	mcount_exit			//   return;
+					// }
+skip_ftrace_call:
+	ldr	x1, =ftrace_graph_return
+	ldr	x2, [x1]		//   if ((ftrace_graph_return
+	cmp	x0, x2			//        != ftrace_stub)
+	b.ne	ftrace_graph_caller
+
+	ldr	x1, =ftrace_graph_entry	//     || (ftrace_graph_entry
+	ldr	x2, [x1]		//        != ftrace_graph_entry_stub))
+	ldr	x0, =ftrace_graph_entry_stub
+	cmp	x0, x2
+	b.ne	ftrace_graph_caller	//     ftrace_graph_caller();
+
+	mcount_exit
+#endif /* CONFIG_FUNCTION_GRAPH_TRACER */
+ENDPROC(_mcount)
+
+ENTRY(ftrace_stub)
+	ret
+ENDPROC(ftrace_stub)
+
+#ifdef CONFIG_FUNCTION_GRAPH_TRACER
+/*
+ * void ftrace_graph_caller(void)
+ *
+ * Called from _mcount() or ftrace_caller() when function_graph tracer is
+ * selected.
+ * This function w/ prepare_ftrace_return() fakes link register's value on
+ * the call stack in order to intercept instrumented function's return path
+ * and run return_to_handler() later on its exit.
+ */
+ENTRY(ftrace_graph_caller)
+	mcount_get_saved_lr_addr  x0	//     pointer to function's saved lr
+	mcount_get_pc		  x1	//     function's pc
+	mcount_get_parent_fp	  x2	//     parent's fp
+	bl	prepare_ftrace_return	// prepare_ftrace_return(&lr, pc, fp)
+
+	mcount_exit
+ENDPROC(ftrace_graph_caller)
+
+/*
+ * void return_to_handler(void)
+ *
+ * Run ftrace_return_to_handler() before going back to parent.
+ * @fp is checked against the value passed by ftrace_graph_caller()
+ * only when CONFIG_FUNCTION_GRAPH_FP_TEST is enabled.
+ */
+	.global return_to_handler
+return_to_handler:
+	str	x0, [sp, #-16]!
+	mov	x0, x29			//     parent's fp
+	bl	ftrace_return_to_handler// addr = ftrace_return_to_hander(fp);
+	mov	x30, x0			// restore the original return address
+	ldr	x0, [sp], #16
+	ret
+#endif /* CONFIG_FUNCTION_GRAPH_TRACER */
diff --git a/arch/arm64/kernel/ftrace.c b/arch/arm64/kernel/ftrace.c
new file mode 100644
index 0000000..a559ab8
--- /dev/null
+++ b/arch/arm64/kernel/ftrace.c
@@ -0,0 +1,64 @@ 
+/*
+ * arch/arm64/kernel/ftrace.c
+ *
+ * Copyright (C) 2013 Linaro Limited
+ * Author: AKASHI Takahiro <takahiro.akashi@linaro.org>
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#include <linux/ftrace.h>
+#include <linux/swab.h>
+#include <linux/uaccess.h>
+
+#include <asm/cacheflush.h>
+#include <asm/ftrace.h>
+#include <asm/insn.h>
+
+#ifdef CONFIG_FUNCTION_GRAPH_TRACER
+/*
+ * function_graph tracer expects ftrace_return_to_handler() to be called
+ * on the way back to parent. For this purpose, this function is called
+ * in _mcount() or ftrace_caller() to replace return address (*parent) on
+ * the call stack to return_to_handler.
+ *
+ * Note that @frame_pointer is used only for sanity check later.
+ */
+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;
+	struct ftrace_graph_ent trace;
+	int err;
+
+	if (unlikely(atomic_read(&current->tracing_graph_pause)))
+		return;
+
+	/*
+	 * Note:
+	 * No protection against faulting at *parent, which may be seen
+	 * on other archs. It's unlikely on AArch64.
+	 */
+	old = *parent;
+	*parent = return_hooker;
+
+	trace.func = self_addr;
+	trace.depth = current->curr_ret_stack + 1;
+
+	/* Only trace if the calling function expects to */
+	if (!ftrace_graph_entry(&trace)) {
+		*parent = old;
+		return;
+	}
+
+	err = ftrace_push_return_trace(old, self_addr, &trace.depth,
+				       frame_pointer);
+	if (err == -EBUSY) {
+		*parent = old;
+		return;
+	}
+}
+#endif /* CONFIG_FUNCTION_GRAPH_TRACER */