linux/Documentation/trace/ftrace-design.txt
<<
>>
Prefs
   1                function tracer guts
   2                ====================
   3                By Mike Frysinger
   4
   5Introduction
   6------------
   7
   8Here we will cover the architecture pieces that the common function tracing
   9code relies on for proper functioning.  Things are broken down into increasing
  10complexity so that you can start simple and at least get basic functionality.
  11
  12Note that this focuses on architecture implementation details only.  If you
  13want more explanation of a feature in terms of common code, review the common
  14ftrace.txt file.
  15
  16Ideally, everyone who wishes to retain performance while supporting tracing in
  17their kernel should make it all the way to dynamic ftrace support.
  18
  19
  20Prerequisites
  21-------------
  22
  23Ftrace relies on these features being implemented:
  24 STACKTRACE_SUPPORT - implement save_stack_trace()
  25 TRACE_IRQFLAGS_SUPPORT - implement include/asm/irqflags.h
  26
  27
  28HAVE_FUNCTION_TRACER
  29--------------------
  30
  31You will need to implement the mcount and the ftrace_stub functions.
  32
  33The 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
  35running something like:
  36        $ echo 'main(){}' | gcc -x c -S -o - - -pg | grep mcount
  37                call    mcount
  38We'll make the assumption below that the symbol is "mcount" just to keep things
  39nice and simple in the examples.
  40
  41Keep 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,
  43sorry.  Dig up some old documentation and/or find someone more familiar than
  44you to bang ideas off of.  Typically, register usage (argument/scratch/etc...)
  45is a major issue at this point, especially in relation to the location of the
  46mcount call (before/after function prologue).  You might also want to look at
  47how glibc has implemented the mcount function for your architecture.  It might
  48be (semi-)relevant.
  49
  50The mcount function should check the function pointer ftrace_trace_function
  51to see if it is set to ftrace_stub.  If it is, there is nothing for you to do,
  52so return immediately.  If it isn't, then call that function in the same way
  53the mcount function normally calls __mcount_internal -- the first argument is
  54the "frompc" while the second argument is the "selfpc" (adjusted to remove the
  55size of the mcount call that is embedded in the function).
  56
  57For example, if the function foo() calls bar(), when the bar() function calls
  58mcount(), 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)
  61
  62Also keep in mind that this mcount function will be called *a lot*, so
  63optimizing for the default case of no tracer will help the smooth running of
  64your system when tracing is disabled.  So the start of the mcount function is
  65typically the bare minimum with checking things before returning.  That also
  66means the code flow should usually be kept linear (i.e. no branching in the nop
  67case).  This is of course an optimization and not a hard requirement.
  68
  69Here is some pseudo code that should help (these functions should actually be
  70implemented in assembly):
  71
  72void ftrace_stub(void)
  73{
  74        return;
  75}
  76
  77void mcount(void)
  78{
  79        /* save any bare state needed in order to do initial checking */
  80
  81        extern void (*ftrace_trace_function)(unsigned long, unsigned long);
  82        if (ftrace_trace_function != ftrace_stub)
  83                goto do_trace;
  84
  85        /* restore any bare state */
  86
  87        return;
  88
  89do_trace:
  90
  91        /* save all state needed by the ABI (see paragraph above) */
  92
  93        unsigned long frompc = ...;
  94        unsigned long selfpc = <return address> - MCOUNT_INSN_SIZE;
  95        ftrace_trace_function(frompc, selfpc);
  96
  97        /* restore all state needed by the ABI */
  98}
  99
 100Don't forget to export mcount for modules !
 101extern void mcount(void);
 102EXPORT_SYMBOL(mcount);
 103
 104
 105HAVE_FUNCTION_TRACE_MCOUNT_TEST
 106-------------------------------
 107
 108This is an optional optimization for the normal case when tracing is turned off
 109in the system.  If you do not enable this Kconfig option, the common ftrace
 110code will take care of doing the checking for you.
 111
 112To support this feature, you only need to check the function_trace_stop
 113variable in the mcount function.  If it is non-zero, there is no tracing to be
 114done at all, so you can return.
 115
 116This additional pseudo code would simply be:
 117void mcount(void)
 118{
 119        /* save any bare state needed in order to do initial checking */
 120
 121+       if (function_trace_stop)
 122+               return;
 123
 124        extern void (*ftrace_trace_function)(unsigned long, unsigned long);
 125        if (ftrace_trace_function != ftrace_stub)
 126...
 127
 128
 129HAVE_FUNCTION_GRAPH_TRACER
 130--------------------------
 131
 132Deep breath ... time to do some real work.  Here you will need to update the
 133mcount function to check ftrace graph function pointers, as well as implement
 134some functions to save (hijack) and restore the return address.
 135
 136The mcount function should check the function pointers ftrace_graph_return
 137(compare to ftrace_stub) and ftrace_graph_entry (compare to
 138ftrace_graph_entry_stub).  If either of those is not set to the relevant stub
 139function, call the arch-specific function ftrace_graph_caller which in turn
 140calls the arch-specific function prepare_ftrace_return.  Neither of these
 141function names is strictly required, but you should use them anyway to stay
 142consistent across the architecture ports -- easier to compare & contrast
 143things.
 144
 145The arguments to prepare_ftrace_return are slightly different than what are
 146passed to ftrace_trace_function.  The second argument "selfpc" is the same,
 147but the first argument should be a pointer to the "frompc".  Typically this is
 148located on the stack.  This allows the function to hijack the return address
 149temporarily to have it point to the arch-specific function return_to_handler.
 150That function will simply call the common ftrace_return_to_handler function and
 151that will return the original return address with which you can return to the
 152original call site.
 153
 154Here is the updated mcount pseudo code:
 155void mcount(void)
 156{
 157...
 158        if (ftrace_trace_function != ftrace_stub)
 159                goto do_trace;
 160
 161+#ifdef CONFIG_FUNCTION_GRAPH_TRACER
 162+       extern void (*ftrace_graph_return)(...);
 163+       extern void (*ftrace_graph_entry)(...);
 164+       if (ftrace_graph_return != ftrace_stub ||
 165+           ftrace_graph_entry != ftrace_graph_entry_stub)
 166+               ftrace_graph_caller();
 167+#endif
 168
 169        /* restore any bare state */
 170...
 171
 172Here is the pseudo code for the new ftrace_graph_caller assembly function:
 173#ifdef CONFIG_FUNCTION_GRAPH_TRACER
 174void ftrace_graph_caller(void)
 175{
 176        /* save all state needed by the ABI */
 177
 178        unsigned long *frompc = &...;
 179        unsigned long selfpc = <return address> - MCOUNT_INSN_SIZE;
 180        /* passing frame pointer up is optional -- see below */
 181        prepare_ftrace_return(frompc, selfpc, frame_pointer);
 182
 183        /* restore all state needed by the ABI */
 184}
 185#endif
 186
 187For information on how to implement prepare_ftrace_return(), simply look at the
 188x86 version (the frame pointer passing is optional; see the next section for
 189more information).  The only architecture-specific piece in it is the setup of
 190the fault recovery table (the asm(...) code).  The rest should be the same
 191across architectures.
 192
 193Here is the pseudo code for the new return_to_handler assembly function.  Note
 194that the ABI that applies here is different from what applies to the mcount
 195code.  Since you are returning from a function (after the epilogue), you might
 196be able to skimp on things saved/restored (usually just registers used to pass
 197return values).
 198
 199#ifdef CONFIG_FUNCTION_GRAPH_TRACER
 200void return_to_handler(void)
 201{
 202        /* save all state needed by the ABI (see paragraph above) */
 203
 204        void (*original_return_point)(void) = ftrace_return_to_handler();
 205
 206        /* restore all state needed by the ABI */
 207
 208        /* this is usually either a return or a jump */
 209        original_return_point();
 210}
 211#endif
 212
 213
 214HAVE_FUNCTION_GRAPH_FP_TEST
 215---------------------------
 216
 217An arch may pass in a unique value (frame pointer) to both the entering and
 218exiting of a function.  On exit, the value is compared and if it does not
 219match, then it will panic the kernel.  This is largely a sanity check for bad
 220code generation with gcc.  If gcc for your port sanely updates the frame
 221pointer under different optimization levels, then ignore this option.
 222
 223However, adding support for it isn't terribly difficult.  In your assembly code
 224that calls prepare_ftrace_return(), pass the frame pointer as the 3rd argument.
 225Then in the C version of that function, do what the x86 port does and pass it
 226along to ftrace_push_return_trace() instead of a stub value of 0.
 227
 228Similarly, when you call ftrace_return_to_handler(), pass it the frame pointer.
 229
 230
 231HAVE_FTRACE_NMI_ENTER
 232---------------------
 233
 234If you can't trace NMI functions, then skip this option.
 235
 236<details to be filled>
 237
 238
 239HAVE_SYSCALL_TRACEPOINTS
 240------------------------
 241
 242You need very few things to get the syscalls tracing in an arch.
 243
 244- Support HAVE_ARCH_TRACEHOOK (see arch/Kconfig).
 245- Have a NR_syscalls variable in <asm/unistd.h> that provides the number
 246  of syscalls supported by the arch.
 247- Support the TIF_SYSCALL_TRACEPOINT thread flags.
 248- Put the trace_sys_enter() and trace_sys_exit() tracepoints calls from ptrace
 249  in the ptrace syscalls tracing path.
 250- If the system call table on this arch is more complicated than a simple array
 251  of addresses of the system calls, implement an arch_syscall_addr to return
 252  the address of a given system call.
 253- If the symbol names of the system calls do not match the function names on
 254  this arch, define ARCH_HAS_SYSCALL_MATCH_SYM_NAME in asm/ftrace.h and
 255  implement arch_syscall_match_sym_name with the appropriate logic to return
 256  true if the function name corresponds with the symbol name.
 257- Tag this arch as HAVE_SYSCALL_TRACEPOINTS.
 258
 259
 260HAVE_FTRACE_MCOUNT_RECORD
 261-------------------------
 262
 263See scripts/recordmcount.pl for more info.  Just fill in the arch-specific
 264details for how to locate the addresses of mcount call sites via objdump.
 265This option doesn't make much sense without also implementing dynamic ftrace.
 266
 267
 268HAVE_DYNAMIC_FTRACE
 269-------------------
 270
 271You will first need HAVE_FTRACE_MCOUNT_RECORD and HAVE_FUNCTION_TRACER, so
 272scroll your reader back up if you got over eager.
 273
 274Once those are out of the way, you will need to implement:
 275        - asm/ftrace.h:
 276                - MCOUNT_ADDR
 277                - ftrace_call_adjust()
 278                - struct dyn_arch_ftrace{}
 279        - asm code:
 280                - mcount() (new stub)
 281                - ftrace_caller()
 282                - ftrace_call()
 283                - ftrace_stub()
 284        - C code:
 285                - ftrace_dyn_arch_init()
 286                - ftrace_make_nop()
 287                - ftrace_make_call()
 288                - ftrace_update_ftrace_func()
 289
 290First you will need to fill out some arch details in your asm/ftrace.h.
 291
 292Define MCOUNT_ADDR as the address of your mcount symbol similar to:
 293        #define MCOUNT_ADDR ((unsigned long)mcount)
 294Since no one else will have a decl for that function, you will need to:
 295        extern void mcount(void);
 296
 297You will also need the helper function ftrace_call_adjust().  Most people
 298will be able to stub it out like so:
 299        static inline unsigned long ftrace_call_adjust(unsigned long addr)
 300        {
 301                return addr;
 302        }
 303<details to be filled>
 304
 305Lastly you will need the custom dyn_arch_ftrace structure.  If you need
 306some extra state when runtime patching arbitrary call sites, this is the
 307place.  For now though, create an empty struct:
 308        struct dyn_arch_ftrace {
 309                /* No extra data needed */
 310        };
 311
 312With the header out of the way, we can fill out the assembly code.  While we
 313did already create a mcount() function earlier, dynamic ftrace only wants a
 314stub function.  This is because the mcount() will only be used during boot
 315and then all references to it will be patched out never to return.  Instead,
 316the guts of the old mcount() will be used to create a new ftrace_caller()
 317function.  Because the two are hard to merge, it will most likely be a lot
 318easier to have two separate definitions split up by #ifdefs.  Same goes for
 319the ftrace_stub() as that will now be inlined in ftrace_caller().
 320
 321Before we get confused anymore, let's check out some pseudo code so you can
 322implement your own stuff in assembly:
 323
 324void mcount(void)
 325{
 326        return;
 327}
 328
 329void ftrace_caller(void)
 330{
 331        /* implement HAVE_FUNCTION_TRACE_MCOUNT_TEST if you desire */
 332
 333        /* save all state needed by the ABI (see paragraph above) */
 334
 335        unsigned long frompc = ...;
 336        unsigned long selfpc = <return address> - MCOUNT_INSN_SIZE;
 337
 338ftrace_call:
 339        ftrace_stub(frompc, selfpc);
 340
 341        /* restore all state needed by the ABI */
 342
 343ftrace_stub:
 344        return;
 345}
 346
 347This might look a little odd at first, but keep in mind that we will be runtime
 348patching multiple things.  First, only functions that we actually want to trace
 349will be patched to call ftrace_caller().  Second, since we only have one tracer
 350active at a time, we will patch the ftrace_caller() function itself to call the
 351specific tracer in question.  That is the point of the ftrace_call label.
 352
 353With that in mind, let's move on to the C code that will actually be doing the
 354runtime patching.  You'll need a little knowledge of your arch's opcodes in
 355order to make it through the next section.
 356
 357Every arch has an init callback function.  If you need to do something early on
 358to initialize some state, this is the time to do that.  Otherwise, this simple
 359function below should be sufficient for most people:
 360
 361int __init ftrace_dyn_arch_init(void *data)
 362{
 363        /* return value is done indirectly via data */
 364        *(unsigned long *)data = 0;
 365
 366        return 0;
 367}
 368
 369There are two functions that are used to do runtime patching of arbitrary
 370functions.  The first is used to turn the mcount call site into a nop (which
 371is what helps us retain runtime performance when not tracing).  The second is
 372used to turn the mcount call site into a call to an arbitrary location (but
 373typically that is ftracer_caller()).  See the general function definition in
 374linux/ftrace.h for the functions:
 375        ftrace_make_nop()
 376        ftrace_make_call()
 377The rec->ip value is the address of the mcount call site that was collected
 378by the scripts/recordmcount.pl during build time.
 379
 380The last function is used to do runtime patching of the active tracer.  This
 381will be modifying the assembly code at the location of the ftrace_call symbol
 382inside of the ftrace_caller() function.  So you should have sufficient padding
 383at that location to support the new function calls you'll be inserting.  Some
 384people will be using a "call" type instruction while others will be using a
 385"branch" type instruction.  Specifically, the function is:
 386        ftrace_update_ftrace_func()
 387
 388
 389HAVE_DYNAMIC_FTRACE + HAVE_FUNCTION_GRAPH_TRACER
 390------------------------------------------------
 391
 392The function grapher needs a few tweaks in order to work with dynamic ftrace.
 393Basically, you will need to:
 394        - update:
 395                - ftrace_caller()
 396                - ftrace_graph_call()
 397                - ftrace_graph_caller()
 398        - implement:
 399                - ftrace_enable_ftrace_graph_caller()
 400                - ftrace_disable_ftrace_graph_caller()
 401
 402<details to be filled>
 403Quick notes:
 404        - add a nop stub after the ftrace_call location named ftrace_graph_call;
 405          stub needs to be large enough to support a call to ftrace_graph_caller()
 406        - update ftrace_graph_caller() to work with being called by the new
 407          ftrace_caller() since some semantics may have changed
 408        - ftrace_enable_ftrace_graph_caller() will runtime patch the
 409          ftrace_graph_call location with a call to ftrace_graph_caller()
 410        - ftrace_disable_ftrace_graph_caller() will runtime patch the
 411          ftrace_graph_call location with nops
 412
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.