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
20 Ftrace relies on these features being implemented:
21 STACKTRACE_SUPPORT - implement save_stack_trace()
22 TRACE_IRQFLAGS_SUPPORT - implement include/asm/irqflags.h
28 You will need to implement the mcount and the ftrace_stub functions.
30 The exact mcount symbol name will depend on your toolchain. Some call it
31 "mcount", "_mcount", or even "__mcount". You can probably figure it out by
32 running something like:
33 $ echo 'main(){}' | gcc -x c -S -o - - -pg | grep mcount
35 We'll make the assumption below that the symbol is "mcount" just to keep things
36 nice and simple in the examples.
38 Keep in mind that the ABI that is in effect inside of the mcount function is
39 *highly* architecture/toolchain specific. We cannot help you in this regard,
40 sorry. Dig up some old documentation and/or find someone more familiar than
41 you to bang ideas off of. Typically, register usage (argument/scratch/etc...)
42 is a major issue at this point, especially in relation to the location of the
43 mcount call (before/after function prologue). You might also want to look at
44 how glibc has implemented the mcount function for your architecture. It might
47 The mcount function should check the function pointer ftrace_trace_function
48 to see if it is set to ftrace_stub. If it is, there is nothing for you to do,
49 so return immediately. If it isn't, then call that function in the same way
50 the mcount function normally calls __mcount_internal -- the first argument is
51 the "frompc" while the second argument is the "selfpc" (adjusted to remove the
52 size of the mcount call that is embedded in the function).
54 For example, if the function foo() calls bar(), when the bar() function calls
55 mcount(), the arguments mcount() will pass to the tracer are:
56 "frompc" - the address bar() will use to return to foo()
57 "selfpc" - the address bar() (with mcount() size adjustment)
59 Also keep in mind that this mcount function will be called *a lot*, so
60 optimizing for the default case of no tracer will help the smooth running of
61 your system when tracing is disabled. So the start of the mcount function is
62 typically the bare minimum with checking things before returning. That also
63 means the code flow should usually be kept linear (i.e. no branching in the nop
64 case). This is of course an optimization and not a hard requirement.
66 Here is some pseudo code that should help (these functions should actually be
67 implemented in assembly):
69 void ftrace_stub(void)
76 /* save any bare state needed in order to do initial checking */
78 extern void (*ftrace_trace_function)(unsigned long, unsigned long);
79 if (ftrace_trace_function != ftrace_stub)
82 /* restore any bare state */
88 /* save all state needed by the ABI (see paragraph above) */
90 unsigned long frompc = ...;
91 unsigned long selfpc = <return address> - MCOUNT_INSN_SIZE;
92 ftrace_trace_function(frompc, selfpc);
94 /* restore all state needed by the ABI */
97 Don't forget to export mcount for modules !
98 extern void mcount(void);
99 EXPORT_SYMBOL(mcount);
102 HAVE_FUNCTION_TRACE_MCOUNT_TEST
103 -------------------------------
105 This is an optional optimization for the normal case when tracing is turned off
106 in the system. If you do not enable this Kconfig option, the common ftrace
107 code will take care of doing the checking for you.
109 To support this feature, you only need to check the function_trace_stop
110 variable in the mcount function. If it is non-zero, there is no tracing to be
111 done at all, so you can return.
113 This additional pseudo code would simply be:
116 /* save any bare state needed in order to do initial checking */
118 + if (function_trace_stop)
121 extern void (*ftrace_trace_function)(unsigned long, unsigned long);
122 if (ftrace_trace_function != ftrace_stub)
126 HAVE_FUNCTION_GRAPH_TRACER
127 --------------------------
129 Deep breath ... time to do some real work. Here you will need to update the
130 mcount function to check ftrace graph function pointers, as well as implement
131 some functions to save (hijack) and restore the return address.
133 The mcount function should check the function pointers ftrace_graph_return
134 (compare to ftrace_stub) and ftrace_graph_entry (compare to
135 ftrace_graph_entry_stub). If either of those is not set to the relevant stub
136 function, call the arch-specific function ftrace_graph_caller which in turn
137 calls the arch-specific function prepare_ftrace_return. Neither of these
138 function names is strictly required, but you should use them anyway to stay
139 consistent across the architecture ports -- easier to compare & contrast
142 The arguments to prepare_ftrace_return are slightly different than what are
143 passed to ftrace_trace_function. The second argument "selfpc" is the same,
144 but the first argument should be a pointer to the "frompc". Typically this is
145 located on the stack. This allows the function to hijack the return address
146 temporarily to have it point to the arch-specific function return_to_handler.
147 That function will simply call the common ftrace_return_to_handler function and
148 that will return the original return address with which you can return to the
151 Here is the updated mcount pseudo code:
155 if (ftrace_trace_function != ftrace_stub)
158 +#ifdef CONFIG_FUNCTION_GRAPH_TRACER
159 + extern void (*ftrace_graph_return)(...);
160 + extern void (*ftrace_graph_entry)(...);
161 + if (ftrace_graph_return != ftrace_stub ||
162 + ftrace_graph_entry != ftrace_graph_entry_stub)
163 + ftrace_graph_caller();
166 /* restore any bare state */
169 Here is the pseudo code for the new ftrace_graph_caller assembly function:
170 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
171 void ftrace_graph_caller(void)
173 /* save all state needed by the ABI */
175 unsigned long *frompc = &...;
176 unsigned long selfpc = <return address> - MCOUNT_INSN_SIZE;
177 /* passing frame pointer up is optional -- see below */
178 prepare_ftrace_return(frompc, selfpc, frame_pointer);
180 /* restore all state needed by the ABI */
184 For information on how to implement prepare_ftrace_return(), simply look at the
185 x86 version (the frame pointer passing is optional; see the next section for
186 more information). The only architecture-specific piece in it is the setup of
187 the fault recovery table (the asm(...) code). The rest should be the same
188 across architectures.
190 Here is the pseudo code for the new return_to_handler assembly function. Note
191 that the ABI that applies here is different from what applies to the mcount
192 code. Since you are returning from a function (after the epilogue), you might
193 be able to skimp on things saved/restored (usually just registers used to pass
196 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
197 void return_to_handler(void)
199 /* save all state needed by the ABI (see paragraph above) */
201 void (*original_return_point)(void) = ftrace_return_to_handler();
203 /* restore all state needed by the ABI */
205 /* this is usually either a return or a jump */
206 original_return_point();
211 HAVE_FUNCTION_GRAPH_FP_TEST
212 ---------------------------
214 An arch may pass in a unique value (frame pointer) to both the entering and
215 exiting of a function. On exit, the value is compared and if it does not
216 match, then it will panic the kernel. This is largely a sanity check for bad
217 code generation with gcc. If gcc for your port sanely updates the frame
218 pointer under different opitmization levels, then ignore this option.
220 However, adding support for it isn't terribly difficult. In your assembly code
221 that calls prepare_ftrace_return(), pass the frame pointer as the 3rd argument.
222 Then in the C version of that function, do what the x86 port does and pass it
223 along to ftrace_push_return_trace() instead of a stub value of 0.
225 Similarly, when you call ftrace_return_to_handler(), pass it the frame pointer.
228 HAVE_FTRACE_NMI_ENTER
229 ---------------------
231 If you can't trace NMI functions, then skip this option.
233 <details to be filled>
236 HAVE_SYSCALL_TRACEPOINTS
237 ---------------------
239 You need very few things to get the syscalls tracing in an arch.
241 - Support HAVE_ARCH_TRACEHOOK (see arch/Kconfig).
242 - Have a NR_syscalls variable in <asm/unistd.h> that provides the number
243 of syscalls supported by the arch.
244 - Support the TIF_SYSCALL_TRACEPOINT thread flags.
245 - Put the trace_sys_enter() and trace_sys_exit() tracepoints calls from ptrace
246 in the ptrace syscalls tracing path.
247 - Tag this arch as HAVE_SYSCALL_TRACEPOINTS.
250 HAVE_FTRACE_MCOUNT_RECORD
251 -------------------------
253 See scripts/recordmcount.pl for more info.
255 <details to be filled>
259 ---------------------
261 <details to be filled>