1 ======================
2 Function Tracer Design
3 ======================
4
5 :Author: Mike Frysinger
6
7 .. caution::
8 This document is out of date. Some of the description below doesn't
9 match current implementation now.
10
11 Introduction
12 ------------
13
14 Here we will cover the architecture pieces that the common function tracing
15 code relies on for proper functioning. Things are broken down into increasing
16 complexity so that you can start simple and at least get basic functionality.
17
18 Note that this focuses on architecture implementation details only. If you
19 want more explanation of a feature in terms of common code, review the common
20 ftrace.txt file.
21
22 Ideally, everyone who wishes to retain performance while supporting tracing in
23 their kernel should make it all the way to dynamic ftrace support.
24
25
26 Prerequisites
27 -------------
28
29 Ftrace relies on these features being implemented:
30 - STACKTRACE_SUPPORT - implement save_stack_trace()
31 - TRACE_IRQFLAGS_SUPPORT - implement include/asm/irqflags.h
32
33
34 HAVE_FUNCTION_TRACER
35 --------------------
36
37 You will need to implement the mcount and the ftrace_stub functions.
38
39 The exact mcount symbol name will depend on your toolchain. Some call it
40 "mcount", "_mcount", or even "__mcount". You can probably figure it out by
41 running something like::
42
43 $ echo 'main(){}' | gcc -x c -S -o - - -pg | grep mcount
44 call mcount
45
46 We'll make the assumption below that the symbol is "mcount" just to keep things
47 nice and simple in the examples.
48
49 Keep in mind that the ABI that is in effect inside of the mcount function is
50 *highly* architecture/toolchain specific. We cannot help you in this regard,
51 sorry. Dig up some old documentation and/or find someone more familiar than
52 you to bang ideas off of. Typically, register usage (argument/scratch/etc...)
53 is a major issue at this point, especially in relation to the location of the
54 mcount call (before/after function prologue). You might also want to look at
55 how glibc has implemented the mcount function for your architecture. It might
56 be (semi-)relevant.
57
58 The mcount function should check the function pointer ftrace_trace_function
59 to see if it is set to ftrace_stub. If it is, there is nothing for you to do,
60 so return immediately. If it isn't, then call that function in the same way
61 the mcount function normally calls __mcount_internal -- the first argument is
62 the "frompc" while the second argument is the "selfpc" (adjusted to remove the
63 size of the mcount call that is embedded in the function).
64
65 For example, if the function foo() calls bar(), when the bar() function calls
66 mcount(), the arguments mcount() will pass to the tracer are:
67
68 - "frompc" - the address bar() will use to return to foo()
69 - "selfpc" - the address bar() (with mcount() size adjustment)
70
71 Also keep in mind that this mcount function will be called *a lot*, so
72 optimizing for the default case of no tracer will help the smooth running of
73 your system when tracing is disabled. So the start of the mcount function is
74 typically the bare minimum with checking things before returning. That also
75 means the code flow should usually be kept linear (i.e. no branching in the nop
76 case). This is of course an optimization and not a hard requirement.
77
78 Here is some pseudo code that should help (these functions should actually be
79 implemented in assembly)::
80
81 void ftrace_stub(void)
82 {
83 return;
84 }
85
86 void mcount(void)
87 {
88 /* save any bare state needed in order to do initial checking */
89
90 extern void (*ftrace_trace_function)(unsigned long, unsigned long);
91 if (ftrace_trace_function != ftrace_stub)
92 goto do_trace;
93
94 /* restore any bare state */
95
96 return;
97
98 do_trace:
99
100 /* save all state needed by the ABI (see paragraph above) */
101
102 unsigned long frompc = ...;
103 unsigned long selfpc = <return address> - MCOUNT_INSN_SIZE;
104 ftrace_trace_function(frompc, selfpc);
105
106 /* restore all state needed by the ABI */
107 }
108
109 Don't forget to export mcount for modules !
110 ::
111
112 extern void mcount(void);
113 EXPORT_SYMBOL(mcount);
114
115
116 HAVE_FUNCTION_GRAPH_TRACER
117 --------------------------
118
119 Deep breath ... time to do some real work. Here you will need to update the
120 mcount function to check ftrace graph function pointers, as well as implement
121 some functions to save (hijack) and restore the return address.
122
123 The mcount function should check the function pointers ftrace_graph_return
124 (compare to ftrace_stub) and ftrace_graph_entry (compare to
125 ftrace_graph_entry_stub). If either of those is not set to the relevant stub
126 function, call the arch-specific function ftrace_graph_caller which in turn
127 calls the arch-specific function prepare_ftrace_return. Neither of these
128 function names is strictly required, but you should use them anyway to stay
129 consistent across the architecture ports -- easier to compare & contrast
130 things.
131
132 The arguments to prepare_ftrace_return are slightly different than what are
133 passed to ftrace_trace_function. The second argument "selfpc" is the same,
134 but the first argument should be a pointer to the "frompc". Typically this is
135 located on the stack. This allows the function to hijack the return address
136 temporarily to have it point to the arch-specific function return_to_handler.
137 That function will simply call the common ftrace_return_to_handler function and
138 that will return the original return address with which you can return to the
139 original call site.
140
141 Here is the updated mcount pseudo code::
142
143 void mcount(void)
144 {
145 ...
146 if (ftrace_trace_function != ftrace_stub)
147 goto do_trace;
148
149 +#ifdef CONFIG_FUNCTION_GRAPH_TRACER
150 + extern void (*ftrace_graph_return)(...);
151 + extern void (*ftrace_graph_entry)(...);
152 + if (ftrace_graph_return != ftrace_stub ||
153 + ftrace_graph_entry != ftrace_graph_entry_stub)
154 + ftrace_graph_caller();
155 +#endif
156
157 /* restore any bare state */
158 ...
159
160 Here is the pseudo code for the new ftrace_graph_caller assembly function::
161
162 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
163 void ftrace_graph_caller(void)
164 {
165 /* save all state needed by the ABI */
166
167 unsigned long *frompc = &...;
168 unsigned long selfpc = <return address> - MCOUNT_INSN_SIZE;
169 /* passing frame pointer up is optional -- see below */
170 prepare_ftrace_return(frompc, selfpc, frame_pointer);
171
172 /* restore all state needed by the ABI */
173 }
174 #endif
175
176 For information on how to implement prepare_ftrace_return(), simply look at the
177 x86 version (the frame pointer passing is optional; see the next section for
178 more information). The only architecture-specific piece in it is the setup of
179 the fault recovery table (the asm(...) code). The rest should be the same
180 across architectures.
181
182 Here is the pseudo code for the new return_to_handler assembly function. Note
183 that the ABI that applies here is different from what applies to the mcount
184 code. Since you are returning from a function (after the epilogue), you might
185 be able to skimp on things saved/restored (usually just registers used to pass
186 return values).
187 ::
188
189 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
190 void return_to_handler(void)
191 {
192 /* save all state needed by the ABI (see paragraph above) */
193
194 void (*original_return_point)(void) = ftrace_return_to_handler();
195
196 /* restore all state needed by the ABI */
197
198 /* this is usually either a return or a jump */
199 original_return_point();
200 }
201 #endif
202
203
204 HAVE_FUNCTION_GRAPH_FP_TEST
205 ---------------------------
206
207 An arch may pass in a unique value (frame pointer) to both the entering and
208 exiting of a function. On exit, the value is compared and if it does not
209 match, then it will panic the kernel. This is largely a sanity check for bad
210 code generation with gcc. If gcc for your port sanely updates the frame
211 pointer under different optimization levels, then ignore this option.
212
213 However, adding support for it isn't terribly difficult. In your assembly code
214 that calls prepare_ftrace_return(), pass the frame pointer as the 3rd argument.
215 Then in the C version of that function, do what the x86 port does and pass it
216 along to ftrace_push_return_trace() instead of a stub value of 0.
217
218 Similarly, when you call ftrace_return_to_handler(), pass it the frame pointer.
219
220 HAVE_SYSCALL_TRACEPOINTS
221 ------------------------
222
223 You need very few things to get the syscalls tracing in an arch.
224
225 - Support HAVE_ARCH_TRACEHOOK (see arch/Kconfig).
226 - Have a NR_syscalls variable in <asm/unistd.h> that provides the number
227 of syscalls supported by the arch.
228 - Support the TIF_SYSCALL_TRACEPOINT thread flags.
229 - Put the trace_sys_enter() and trace_sys_exit() tracepoints calls from ptrace
230 in the ptrace syscalls tracing path.
231 - If the system call table on this arch is more complicated than a simple array
232 of addresses of the system calls, implement an arch_syscall_addr to return
233 the address of a given system call.
234 - If the symbol names of the system calls do not match the function names on
235 this arch, define ARCH_HAS_SYSCALL_MATCH_SYM_NAME in asm/ftrace.h and
236 implement arch_syscall_match_sym_name with the appropriate logic to return
237 true if the function name corresponds with the symbol name.
238 - Tag this arch as HAVE_SYSCALL_TRACEPOINTS.
239
240
241 HAVE_FTRACE_MCOUNT_RECORD
242 -------------------------
243
244 See scripts/recordmcount.pl for more info. Just fill in the arch-specific
245 details for how to locate the addresses of mcount call sites via objdump.
246 This option doesn't make much sense without also implementing dynamic ftrace.
247
248
249 HAVE_DYNAMIC_FTRACE
250 -------------------
251
252 You will first need HAVE_FTRACE_MCOUNT_RECORD and HAVE_FUNCTION_TRACER, so
253 scroll your reader back up if you got over eager.
254
255 Once those are out of the way, you will need to implement:
256 - asm/ftrace.h:
257 - MCOUNT_ADDR
258 - ftrace_call_adjust()
259 - struct dyn_arch_ftrace{}
260 - asm code:
261 - mcount() (new stub)
262 - ftrace_caller()
263 - ftrace_call()
264 - ftrace_stub()
265 - C code:
266 - ftrace_dyn_arch_init()
267 - ftrace_make_nop()
268 - ftrace_make_call()
269 - ftrace_update_ftrace_func()
270
271 First you will need to fill out some arch details in your asm/ftrace.h.
272
273 Define MCOUNT_ADDR as the address of your mcount symbol similar to::
274
275 #define MCOUNT_ADDR ((unsigned long)mcount)
276
277 Since no one else will have a decl for that function, you will need to::
278
279 extern void mcount(void);
280
281 You will also need the helper function ftrace_call_adjust(). Most people
282 will be able to stub it out like so::
283
284 static inline unsigned long ftrace_call_adjust(unsigned long addr)
285 {
286 return addr;
287 }
288
289 <details to be filled>
290
291 Lastly you will need the custom dyn_arch_ftrace structure. If you need
292 some extra state when runtime patching arbitrary call sites, this is the
293 place. For now though, create an empty struct::
294
295 struct dyn_arch_ftrace {
296 /* No extra data needed */
297 };
298
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().
307
308 Before we get confused anymore, let's check out some pseudo code so you can
309 implement your own stuff in assembly::
310
311 void mcount(void)
312 {
313 return;
314 }
315
316 void ftrace_caller(void)
317 {
318 /* save all state needed by the ABI (see paragraph above) */
319
320 unsigned long frompc = ...;
321 unsigned long selfpc = <return address> - MCOUNT_INSN_SIZE;
322
323 ftrace_call:
324 ftrace_stub(frompc, selfpc);
325
326 /* restore all state needed by the ABI */
327
328 ftrace_stub:
329 return;
330 }
331
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.
337
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.
341
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::
345
346 int __init ftrace_dyn_arch_init(void)
347 {
348 return 0;
349 }
350
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
358 ftrace_make_nop()
359 ftrace_make_call()
360
361 The rec->ip value is the address of the mcount call site that was collected
362 by the scripts/recordmcount.pl during build time.
363
364 The last function is used to do runtime patching of the active tracer. This
365 will be modifying the assembly code at the location of the ftrace_call symbol
366 inside of the ftrace_caller() function. So you should have sufficient padding
367 at that location to support the new function calls you'll be inserting. Some
368 people will be using a "call" type instruction while others will be using a
369 "branch" type instruction. Specifically, the function is::
370
371 ftrace_update_ftrace_func()
372
373
374 HAVE_DYNAMIC_FTRACE + HAVE_FUNCTION_GRAPH_TRACER
375 ------------------------------------------------
376
377 The function grapher needs a few tweaks in order to work with dynamic ftrace.
378 Basically, you will need to:
379
380 - update:
381 - ftrace_caller()
382 - ftrace_graph_call()
383 - ftrace_graph_caller()
384 - implement:
385 - ftrace_enable_ftrace_graph_caller()
386 - ftrace_disable_ftrace_graph_caller()
387
388 <details to be filled>
389
390 Quick notes:
391
392 - add a nop stub after the ftrace_call location named ftrace_graph_call;
393 stub needs to be large enough to support a call to ftrace_graph_caller()
394 - update ftrace_graph_caller() to work with being called by the new
395 ftrace_caller() since some semantics may have changed
396 - ftrace_enable_ftrace_graph_caller() will runtime patch the
397 ftrace_graph_call location with a call to ftrace_graph_caller()
398 - ftrace_disable_ftrace_graph_caller() will runtime patch the
399 ftrace_graph_call location with nops
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.