1 ======================
2 Function Tracer Design
3 ======================
4
5 :Author: Mike Frysinger
6
7 .. caution::
8 This document is out of date. Some of
9 match current implementation now.
10
11 Introduction
12 ------------
13
14 Here we will cover the architecture pieces tha
15 code relies on for proper functioning. Things
16 complexity so that you can start simple and at
17
18 Note that this focuses on architecture impleme
19 want more explanation of a feature in terms of
20 ftrace.txt file.
21
22 Ideally, everyone who wishes to retain perform
23 their kernel should make it all the way to dyn
24
25
26 Prerequisites
27 -------------
28
29 Ftrace relies on these features being implemen
30 - STACKTRACE_SUPPORT - implement save_stack_
31 - TRACE_IRQFLAGS_SUPPORT - implement include
32
33
34 HAVE_FUNCTION_TRACER
35 --------------------
36
37 You will need to implement the mcount and the
38
39 The exact mcount symbol name will depend on yo
40 "mcount", "_mcount", or even "__mcount". You
41 running something like::
42
43 $ echo 'main(){}' | gcc -x c -S -o - -
44 call mcount
45
46 We'll make the assumption below that the symbo
47 nice and simple in the examples.
48
49 Keep in mind that the ABI that is in effect in
50 *highly* architecture/toolchain specific. We
51 sorry. Dig up some old documentation and/or f
52 you to bang ideas off of. Typically, register
53 is a major issue at this point, especially in
54 mcount call (before/after function prologue).
55 how glibc has implemented the mcount function
56 be (semi-)relevant.
57
58 The mcount function should check the function
59 to see if it is set to ftrace_stub. If it is,
60 so return immediately. If it isn't, then call
61 the mcount function normally calls __mcount_in
62 the "frompc" while the second argument is the
63 size of the mcount call that is embedded in th
64
65 For example, if the function foo() calls bar()
66 mcount(), the arguments mcount() will pass to
67
68 - "frompc" - the address bar() will use to r
69 - "selfpc" - the address bar() (with mcount(
70
71 Also keep in mind that this mcount function wi
72 optimizing for the default case of no tracer w
73 your system when tracing is disabled. So the
74 typically the bare minimum with checking thing
75 means the code flow should usually be kept lin
76 case). This is of course an optimization and
77
78 Here is some pseudo code that should help (the
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
89
90 extern void (*ftrace_trace_fun
91 if (ftrace_trace_function != f
92 goto do_trace;
93
94 /* restore any bare state */
95
96 return;
97
98 do_trace:
99
100 /* save all state needed by th
101
102 unsigned long frompc = ...;
103 unsigned long selfpc = <return
104 ftrace_trace_function(frompc,
105
106 /* restore all state needed by
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. He
120 mcount function to check ftrace graph function
121 some functions to save (hijack) and restore th
122
123 The mcount function should check the function
124 (compare to ftrace_stub) and ftrace_graph_entr
125 ftrace_graph_entry_stub). If either of those
126 function, call the arch-specific function ftra
127 calls the arch-specific function prepare_ftrac
128 function names is strictly required, but you s
129 consistent across the architecture ports -- ea
130 things.
131
132 The arguments to prepare_ftrace_return are sli
133 passed to ftrace_trace_function. The second a
134 but the first argument should be a pointer to
135 located on the stack. This allows the functio
136 temporarily to have it point to the arch-speci
137 That function will simply call the common ftra
138 that will return the original return address w
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 != f
147 goto do_trace;
148
149 +#ifdef CONFIG_FUNCTION_GRAPH_TRACER
150 + extern void (*ftrace_graph_ret
151 + extern void (*ftrace_graph_ent
152 + if (ftrace_graph_return != ftr
153 + ftrace_graph_entry != ftra
154 + ftrace_graph_caller();
155 +#endif
156
157 /* restore any bare state */
158 ...
159
160 Here is the pseudo code for the new ftrace_gra
161
162 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
163 void ftrace_graph_caller(void)
164 {
165 /* save all state needed by th
166
167 unsigned long *frompc = &...;
168 unsigned long selfpc = <return
169 /* passing frame pointer up is
170 prepare_ftrace_return(frompc,
171
172 /* restore all state needed by
173 }
174 #endif
175
176 For information on how to implement prepare_ft
177 x86 version (the frame pointer passing is opti
178 more information). The only architecture-spec
179 the fault recovery table (the asm(...) code).
180 across architectures.
181
182 Here is the pseudo code for the new return_to_
183 that the ABI that applies here is different fr
184 code. Since you are returning from a function
185 be able to skimp on things saved/restored (usu
186 return values).
187 ::
188
189 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
190 void return_to_handler(void)
191 {
192 /* save all state needed by th
193
194 void (*original_return_point)(
195
196 /* restore all state needed by
197
198 /* this is usually either a re
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 poin
208 exiting of a function. On exit, the value is
209 match, then it will panic the kernel. This is
210 code generation with gcc. If gcc for your por
211 pointer under different optimization levels, t
212
213 However, adding support for it isn't terribly
214 that calls prepare_ftrace_return(), pass the f
215 Then in the C version of that function, do wha
216 along to ftrace_push_return_trace() instead of
217
218 Similarly, when you call ftrace_return_to_hand
219
220 HAVE_SYSCALL_TRACEPOINTS
221 ------------------------
222
223 You need very few things to get the syscalls t
224
225 - Support HAVE_ARCH_TRACEHOOK (see arch/Kcon
226 - Have a NR_syscalls variable in <asm/unistd
227 of syscalls supported by the arch.
228 - Support the TIF_SYSCALL_TRACEPOINT thread
229 - Put the trace_sys_enter() and trace_sys_ex
230 in the ptrace syscalls tracing path.
231 - If the system call table on this arch is m
232 of addresses of the system calls, implemen
233 the address of a given system call.
234 - If the symbol names of the system calls do
235 this arch, define ARCH_HAS_SYSCALL_MATCH_S
236 implement arch_syscall_match_sym_name with
237 true if the function name corresponds with
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. Ju
245 details for how to locate the addresses of mco
246 This option doesn't make much sense without al
247
248
249 HAVE_DYNAMIC_FTRACE
250 -------------------
251
252 You will first need HAVE_FTRACE_MCOUNT_RECORD
253 scroll your reader back up if you got over eag
254
255 Once those are out of the way, you will need t
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 deta
272
273 Define MCOUNT_ADDR as the address of your mcou
274
275 #define MCOUNT_ADDR ((unsigned long)mc
276
277 Since no one else will have a decl for that fu
278
279 extern void mcount(void);
280
281 You will also need the helper function ftrace_
282 will be able to stub it out like so::
283
284 static inline unsigned long ftrace_cal
285 {
286 return addr;
287 }
288
289 <details to be filled>
290
291 Lastly you will need the custom dyn_arch_ftrac
292 some extra state when runtime patching arbitra
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 ou
300 did already create a mcount() function earlier
301 stub function. This is because the mcount() w
302 and then all references to it will be patched
303 the guts of the old mcount() will be used to c
304 function. Because the two are hard to merge,
305 easier to have two separate definitions split
306 the ftrace_stub() as that will now be inlined
307
308 Before we get confused anymore, let's check ou
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 th
319
320 unsigned long frompc = ...;
321 unsigned long selfpc = <return
322
323 ftrace_call:
324 ftrace_stub(frompc, selfpc);
325
326 /* restore all state needed by
327
328 ftrace_stub:
329 return;
330 }
331
332 This might look a little odd at first, but kee
333 patching multiple things. First, only functio
334 will be patched to call ftrace_caller(). Seco
335 active at a time, we will patch the ftrace_cal
336 specific tracer in question. That is the poin
337
338 With that in mind, let's move on to the C code
339 runtime patching. You'll need a little knowle
340 order to make it through the next section.
341
342 Every arch has an init callback function. If
343 to initialize some state, this is the time to
344 function below should be sufficient for most p
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 ru
352 functions. The first is used to turn the mcou
353 is what helps us retain runtime performance wh
354 used to turn the mcount call site into a call
355 typically that is ftracer_caller()). See the
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
362 by the scripts/recordmcount.pl during build ti
363
364 The last function is used to do runtime patchi
365 will be modifying the assembly code at the loc
366 inside of the ftrace_caller() function. So yo
367 at that location to support the new function c
368 people will be using a "call" type instruction
369 "branch" type instruction. Specifically, the
370
371 ftrace_update_ftrace_func()
372
373
374 HAVE_DYNAMIC_FTRACE + HAVE_FUNCTION_GRAPH_TRAC
375 ----------------------------------------------
376
377 The function grapher needs a few tweaks in ord
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_c
386 - ftrace_disable_ftrace_graph_
387
388 <details to be filled>
389
390 Quick notes:
391
392 - add a nop stub after the ftrace_call
393 stub needs to be large enough to sup
394 - update ftrace_graph_caller() to work
395 ftrace_caller() since some semantics
396 - ftrace_enable_ftrace_graph_caller()
397 ftrace_graph_call location with a ca
398 - ftrace_disable_ftrace_graph_caller()
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.