About Kernel Documentation Linux Kernel Contact Linux Resources Linux Blog

Documentation / trace / ftrace-design.txt

Custom Search

Based on kernel version 4.9. Page generated on 2016-12-21 14:37 EST.

1			function tracer guts
2			====================
3			By Mike Frysinger
5	Introduction
6	------------
8	Here we will cover the architecture pieces that the common function tracing
9	code relies on for proper functioning.  Things are broken down into increasing
10	complexity so that you can start simple and at least get basic functionality.
12	Note that this focuses on architecture implementation details only.  If you
13	want more explanation of a feature in terms of common code, review the common
14	ftrace.txt file.
16	Ideally, everyone who wishes to retain performance while supporting tracing in
17	their kernel should make it all the way to dynamic ftrace support.
20	Prerequisites
21	-------------
23	Ftrace relies on these features being implemented:
24	 STACKTRACE_SUPPORT - implement save_stack_trace()
25	 TRACE_IRQFLAGS_SUPPORT - implement include/asm/irqflags.h
29	--------------------
31	You will need to implement the mcount and the ftrace_stub functions.
33	The exact mcount symbol name will depend on your toolchain.  Some call it
34	"mcount", "_mcount", or even "__mcount".  You can probably figure it out by
35	running something like:
36		$ echo 'main(){}' | gcc -x c -S -o - - -pg | grep mcount
37		        call    mcount
38	We'll make the assumption below that the symbol is "mcount" just to keep things
39	nice and simple in the examples.
41	Keep in mind that the ABI that is in effect inside of the mcount function is
42	*highly* architecture/toolchain specific.  We cannot help you in this regard,
43	sorry.  Dig up some old documentation and/or find someone more familiar than
44	you to bang ideas off of.  Typically, register usage (argument/scratch/etc...)
45	is a major issue at this point, especially in relation to the location of the
46	mcount call (before/after function prologue).  You might also want to look at
47	how glibc has implemented the mcount function for your architecture.  It might
48	be (semi-)relevant.
50	The mcount function should check the function pointer ftrace_trace_function
51	to see if it is set to ftrace_stub.  If it is, there is nothing for you to do,
52	so return immediately.  If it isn't, then call that function in the same way
53	the mcount function normally calls __mcount_internal -- the first argument is
54	the "frompc" while the second argument is the "selfpc" (adjusted to remove the
55	size of the mcount call that is embedded in the function).
57	For example, if the function foo() calls bar(), when the bar() function calls
58	mcount(), the arguments mcount() will pass to the tracer are:
59		"frompc" - the address bar() will use to return to foo()
60		"selfpc" - the address bar() (with mcount() size adjustment)
62	Also keep in mind that this mcount function will be called *a lot*, so
63	optimizing for the default case of no tracer will help the smooth running of
64	your system when tracing is disabled.  So the start of the mcount function is
65	typically the bare minimum with checking things before returning.  That also
66	means the code flow should usually be kept linear (i.e. no branching in the nop
67	case).  This is of course an optimization and not a hard requirement.
69	Here is some pseudo code that should help (these functions should actually be
70	implemented in assembly):
72	void ftrace_stub(void)
73	{
74		return;
75	}
77	void mcount(void)
78	{
79		/* save any bare state needed in order to do initial checking */
81		extern void (*ftrace_trace_function)(unsigned long, unsigned long);
82		if (ftrace_trace_function != ftrace_stub)
83			goto do_trace;
85		/* restore any bare state */
87		return;
89	do_trace:
91		/* save all state needed by the ABI (see paragraph above) */
93		unsigned long frompc = ...;
94		unsigned long selfpc = <return address> - MCOUNT_INSN_SIZE;
95		ftrace_trace_function(frompc, selfpc);
97		/* restore all state needed by the ABI */
98	}
100	Don't forget to export mcount for modules !
101	extern void mcount(void);
102	EXPORT_SYMBOL(mcount);
106	--------------------------
108	Deep breath ... time to do some real work.  Here you will need to update the
109	mcount function to check ftrace graph function pointers, as well as implement
110	some functions to save (hijack) and restore the return address.
112	The mcount function should check the function pointers ftrace_graph_return
113	(compare to ftrace_stub) and ftrace_graph_entry (compare to
114	ftrace_graph_entry_stub).  If either of those is not set to the relevant stub
115	function, call the arch-specific function ftrace_graph_caller which in turn
116	calls the arch-specific function prepare_ftrace_return.  Neither of these
117	function names is strictly required, but you should use them anyway to stay
118	consistent across the architecture ports -- easier to compare & contrast
119	things.
121	The arguments to prepare_ftrace_return are slightly different than what are
122	passed to ftrace_trace_function.  The second argument "selfpc" is the same,
123	but the first argument should be a pointer to the "frompc".  Typically this is
124	located on the stack.  This allows the function to hijack the return address
125	temporarily to have it point to the arch-specific function return_to_handler.
126	That function will simply call the common ftrace_return_to_handler function and
127	that will return the original return address with which you can return to the
128	original call site.
130	Here is the updated mcount pseudo code:
131	void mcount(void)
132	{
133	...
134		if (ftrace_trace_function != ftrace_stub)
135			goto do_trace;
138	+	extern void (*ftrace_graph_return)(...);
139	+	extern void (*ftrace_graph_entry)(...);
140	+	if (ftrace_graph_return != ftrace_stub ||
141	+	    ftrace_graph_entry != ftrace_graph_entry_stub)
142	+		ftrace_graph_caller();
143	+#endif
145		/* restore any bare state */
146	...
148	Here is the pseudo code for the new ftrace_graph_caller assembly function:
150	void ftrace_graph_caller(void)
151	{
152		/* save all state needed by the ABI */
154		unsigned long *frompc = &...;
155		unsigned long selfpc = <return address> - MCOUNT_INSN_SIZE;
156		/* passing frame pointer up is optional -- see below */
157		prepare_ftrace_return(frompc, selfpc, frame_pointer);
159		/* restore all state needed by the ABI */
160	}
161	#endif
163	For information on how to implement prepare_ftrace_return(), simply look at the
164	x86 version (the frame pointer passing is optional; see the next section for
165	more information).  The only architecture-specific piece in it is the setup of
166	the fault recovery table (the asm(...) code).  The rest should be the same
167	across architectures.
169	Here is the pseudo code for the new return_to_handler assembly function.  Note
170	that the ABI that applies here is different from what applies to the mcount
171	code.  Since you are returning from a function (after the epilogue), you might
172	be able to skimp on things saved/restored (usually just registers used to pass
173	return values).
176	void return_to_handler(void)
177	{
178		/* save all state needed by the ABI (see paragraph above) */
180		void (*original_return_point)(void) = ftrace_return_to_handler();
182		/* restore all state needed by the ABI */
184		/* this is usually either a return or a jump */
185		original_return_point();
186	}
187	#endif
191	---------------------------
193	An arch may pass in a unique value (frame pointer) to both the entering and
194	exiting of a function.  On exit, the value is compared and if it does not
195	match, then it will panic the kernel.  This is largely a sanity check for bad
196	code generation with gcc.  If gcc for your port sanely updates the frame
197	pointer under different optimization levels, then ignore this option.
199	However, adding support for it isn't terribly difficult.  In your assembly code
200	that calls prepare_ftrace_return(), pass the frame pointer as the 3rd argument.
201	Then in the C version of that function, do what the x86 port does and pass it
202	along to ftrace_push_return_trace() instead of a stub value of 0.
204	Similarly, when you call ftrace_return_to_handler(), pass it the frame pointer.
207	--------------------------------
209	An arch may pass in a pointer to the return address on the stack.  This
210	prevents potential stack unwinding issues where the unwinder gets out of
211	sync with ret_stack and the wrong addresses are reported by
212	ftrace_graph_ret_addr().
214	Adding support for it is easy: just define the macro in asm/ftrace.h and
215	pass the return address pointer as the 'retp' argument to
216	ftrace_push_return_trace().
219	---------------------
221	If you can't trace NMI functions, then skip this option.
223	<details to be filled>
227	------------------------
229	You need very few things to get the syscalls tracing in an arch.
231	- Support HAVE_ARCH_TRACEHOOK (see arch/Kconfig).
232	- Have a NR_syscalls variable in <asm/unistd.h> that provides the number
233	  of syscalls supported by the arch.
234	- Support the TIF_SYSCALL_TRACEPOINT thread flags.
235	- Put the trace_sys_enter() and trace_sys_exit() tracepoints calls from ptrace
236	  in the ptrace syscalls tracing path.
237	- If the system call table on this arch is more complicated than a simple array
238	  of addresses of the system calls, implement an arch_syscall_addr to return
239	  the address of a given system call.
240	- If the symbol names of the system calls do not match the function names on
241	  this arch, define ARCH_HAS_SYSCALL_MATCH_SYM_NAME in asm/ftrace.h and
242	  implement arch_syscall_match_sym_name with the appropriate logic to return
243	  true if the function name corresponds with the symbol name.
244	- Tag this arch as HAVE_SYSCALL_TRACEPOINTS.
248	-------------------------
250	See scripts/recordmcount.pl for more info.  Just fill in the arch-specific
251	details for how to locate the addresses of mcount call sites via objdump.
252	This option doesn't make much sense without also implementing dynamic ftrace.
256	-------------------
259	scroll your reader back up if you got over eager.
261	Once those are out of the way, you will need to implement:
262		- asm/ftrace.h:
264			- ftrace_call_adjust()
265			- struct dyn_arch_ftrace{}
266		- asm code:
267			- mcount() (new stub)
268			- ftrace_caller()
269			- ftrace_call()
270			- ftrace_stub()
271		- C code:
272			- ftrace_dyn_arch_init()
273			- ftrace_make_nop()
274			- ftrace_make_call()
275			- ftrace_update_ftrace_func()
277	First you will need to fill out some arch details in your asm/ftrace.h.
279	Define MCOUNT_ADDR as the address of your mcount symbol similar to:
280		#define MCOUNT_ADDR ((unsigned long)mcount)
281	Since no one else will have a decl for that function, you will need to:
282		extern void mcount(void);
284	You will also need the helper function ftrace_call_adjust().  Most people
285	will be able to stub it out like so:
286		static inline unsigned long ftrace_call_adjust(unsigned long addr)
287		{
288			return addr;
289		}
290	<details to be filled>
292	Lastly you will need the custom dyn_arch_ftrace structure.  If you need
293	some extra state when runtime patching arbitrary call sites, this is the
294	place.  For now though, create an empty struct:
295		struct dyn_arch_ftrace {
296			/* No extra data needed */
297		};
299	With the header out of the way, we can fill out the assembly code.  While we
300	did already create a mcount() function earlier, dynamic ftrace only wants a
301	stub function.  This is because the mcount() will only be used during boot
302	and then all references to it will be patched out never to return.  Instead,
303	the guts of the old mcount() will be used to create a new ftrace_caller()
304	function.  Because the two are hard to merge, it will most likely be a lot
305	easier to have two separate definitions split up by #ifdefs.  Same goes for
306	the ftrace_stub() as that will now be inlined in ftrace_caller().
308	Before we get confused anymore, let's check out some pseudo code so you can
309	implement your own stuff in assembly:
311	void mcount(void)
312	{
313		return;
314	}
316	void ftrace_caller(void)
317	{
318		/* save all state needed by the ABI (see paragraph above) */
320		unsigned long frompc = ...;
321		unsigned long selfpc = <return address> - MCOUNT_INSN_SIZE;
323	ftrace_call:
324		ftrace_stub(frompc, selfpc);
326		/* restore all state needed by the ABI */
328	ftrace_stub:
329		return;
330	}
332	This might look a little odd at first, but keep in mind that we will be runtime
333	patching multiple things.  First, only functions that we actually want to trace
334	will be patched to call ftrace_caller().  Second, since we only have one tracer
335	active at a time, we will patch the ftrace_caller() function itself to call the
336	specific tracer in question.  That is the point of the ftrace_call label.
338	With that in mind, let's move on to the C code that will actually be doing the
339	runtime patching.  You'll need a little knowledge of your arch's opcodes in
340	order to make it through the next section.
342	Every arch has an init callback function.  If you need to do something early on
343	to initialize some state, this is the time to do that.  Otherwise, this simple
344	function below should be sufficient for most people:
346	int __init ftrace_dyn_arch_init(void)
347	{
348		return 0;
349	}
351	There are two functions that are used to do runtime patching of arbitrary
352	functions.  The first is used to turn the mcount call site into a nop (which
353	is what helps us retain runtime performance when not tracing).  The second is
354	used to turn the mcount call site into a call to an arbitrary location (but
355	typically that is ftracer_caller()).  See the general function definition in
356	linux/ftrace.h for the functions:
357		ftrace_make_nop()
358		ftrace_make_call()
359	The rec->ip value is the address of the mcount call site that was collected
360	by the scripts/recordmcount.pl during build time.
362	The last function is used to do runtime patching of the active tracer.  This
363	will be modifying the assembly code at the location of the ftrace_call symbol
364	inside of the ftrace_caller() function.  So you should have sufficient padding
365	at that location to support the new function calls you'll be inserting.  Some
366	people will be using a "call" type instruction while others will be using a
367	"branch" type instruction.  Specifically, the function is:
368		ftrace_update_ftrace_func()
372	------------------------------------------------
374	The function grapher needs a few tweaks in order to work with dynamic ftrace.
375	Basically, you will need to:
376		- update:
377			- ftrace_caller()
378			- ftrace_graph_call()
379			- ftrace_graph_caller()
380		- implement:
381			- ftrace_enable_ftrace_graph_caller()
382			- ftrace_disable_ftrace_graph_caller()
384	<details to be filled>
385	Quick notes:
386		- add a nop stub after the ftrace_call location named ftrace_graph_call;
387		  stub needs to be large enough to support a call to ftrace_graph_caller()
388		- update ftrace_graph_caller() to work with being called by the new
389		  ftrace_caller() since some semantics may have changed
390		- ftrace_enable_ftrace_graph_caller() will runtime patch the
391		  ftrace_graph_call location with a call to ftrace_graph_caller()
392		- ftrace_disable_ftrace_graph_caller() will runtime patch the
393		  ftrace_graph_call location with nops
Hide Line Numbers
About Kernel Documentation Linux Kernel Contact Linux Resources Linux Blog

Information is copyright its respective author. All material is available from the Linux Kernel Source distributed under a GPL License. This page is provided as a free service by mjmwired.net.