400 lines · plain
1======================2Function Tracer Design3======================4 5:Author: Mike Frysinger6 7.. caution::8 This document is out of date. Some of the description below doesn't9 match current implementation now.10 11Introduction12------------13 14Here we will cover the architecture pieces that the common function tracing15code relies on for proper functioning. Things are broken down into increasing16complexity so that you can start simple and at least get basic functionality.17 18Note that this focuses on architecture implementation details only. If you19want more explanation of a feature in terms of common code, review the common20ftrace.txt file.21 22Ideally, everyone who wishes to retain performance while supporting tracing in23their kernel should make it all the way to dynamic ftrace support.24 25 26Prerequisites27-------------28 29Ftrace relies on these features being implemented:30 - STACKTRACE_SUPPORT - implement save_stack_trace()31 - TRACE_IRQFLAGS_SUPPORT - implement include/asm/irqflags.h32 33 34HAVE_FUNCTION_TRACER35--------------------36 37You will need to implement the mcount and the ftrace_stub functions.38 39The exact mcount symbol name will depend on your toolchain. Some call it40"mcount", "_mcount", or even "__mcount". You can probably figure it out by41running something like::42 43 $ echo 'main(){}' | gcc -x c -S -o - - -pg | grep mcount44 call mcount45 46We'll make the assumption below that the symbol is "mcount" just to keep things47nice and simple in the examples.48 49Keep in mind that the ABI that is in effect inside of the mcount function is50*highly* architecture/toolchain specific. We cannot help you in this regard,51sorry. Dig up some old documentation and/or find someone more familiar than52you to bang ideas off of. Typically, register usage (argument/scratch/etc...)53is a major issue at this point, especially in relation to the location of the54mcount call (before/after function prologue). You might also want to look at55how glibc has implemented the mcount function for your architecture. It might56be (semi-)relevant.57 58The mcount function should check the function pointer ftrace_trace_function59to see if it is set to ftrace_stub. If it is, there is nothing for you to do,60so return immediately. If it isn't, then call that function in the same way61the mcount function normally calls __mcount_internal -- the first argument is62the "frompc" while the second argument is the "selfpc" (adjusted to remove the63size of the mcount call that is embedded in the function).64 65For example, if the function foo() calls bar(), when the bar() function calls66mcount(), 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 71Also keep in mind that this mcount function will be called *a lot*, so72optimizing for the default case of no tracer will help the smooth running of73your system when tracing is disabled. So the start of the mcount function is74typically the bare minimum with checking things before returning. That also75means the code flow should usually be kept linear (i.e. no branching in the nop76case). This is of course an optimization and not a hard requirement.77 78Here is some pseudo code that should help (these functions should actually be79implemented 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 109Don't forget to export mcount for modules !110::111 112 extern void mcount(void);113 EXPORT_SYMBOL(mcount);114 115 116HAVE_FUNCTION_GRAPH_TRACER117--------------------------118 119Deep breath ... time to do some real work. Here you will need to update the120mcount function to check ftrace graph function pointers, as well as implement121some functions to save (hijack) and restore the return address.122 123The mcount function should check the function pointers ftrace_graph_return124(compare to ftrace_stub) and ftrace_graph_entry (compare to125ftrace_graph_entry_stub). If either of those is not set to the relevant stub126function, call the arch-specific function ftrace_graph_caller which in turn127calls the arch-specific function prepare_ftrace_return. Neither of these128function names is strictly required, but you should use them anyway to stay129consistent across the architecture ports -- easier to compare & contrast130things.131 132The arguments to prepare_ftrace_return are slightly different than what are133passed to ftrace_trace_function. The second argument "selfpc" is the same,134but the first argument should be a pointer to the "frompc". Typically this is135located on the stack. This allows the function to hijack the return address136temporarily to have it point to the arch-specific function return_to_handler.137That function will simply call the common ftrace_return_to_handler function and138that will return the original return address with which you can return to the139original call site.140 141Here 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_TRACER150 + 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 +#endif156 157 /* restore any bare state */158 ...159 160Here is the pseudo code for the new ftrace_graph_caller assembly function::161 162 #ifdef CONFIG_FUNCTION_GRAPH_TRACER163 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 #endif175 176For information on how to implement prepare_ftrace_return(), simply look at the177x86 version (the frame pointer passing is optional; see the next section for178more information). The only architecture-specific piece in it is the setup of179the fault recovery table (the asm(...) code). The rest should be the same180across architectures.181 182Here is the pseudo code for the new return_to_handler assembly function. Note183that the ABI that applies here is different from what applies to the mcount184code. Since you are returning from a function (after the epilogue), you might185be able to skimp on things saved/restored (usually just registers used to pass186return values).187::188 189 #ifdef CONFIG_FUNCTION_GRAPH_TRACER190 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 #endif202 203 204HAVE_FUNCTION_GRAPH_FP_TEST205---------------------------206 207An arch may pass in a unique value (frame pointer) to both the entering and208exiting of a function. On exit, the value is compared and if it does not209match, then it will panic the kernel. This is largely a sanity check for bad210code generation with gcc. If gcc for your port sanely updates the frame211pointer under different optimization levels, then ignore this option.212 213However, adding support for it isn't terribly difficult. In your assembly code214that calls prepare_ftrace_return(), pass the frame pointer as the 3rd argument.215Then in the C version of that function, do what the x86 port does and pass it216along to ftrace_push_return_trace() instead of a stub value of 0.217 218Similarly, when you call ftrace_return_to_handler(), pass it the frame pointer.219 220HAVE_SYSCALL_TRACEPOINTS221------------------------222 223You 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 number227 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 ptrace230 in the ptrace syscalls tracing path.231 - If the system call table on this arch is more complicated than a simple array232 of addresses of the system calls, implement an arch_syscall_addr to return233 the address of a given system call.234 - If the symbol names of the system calls do not match the function names on235 this arch, define ARCH_HAS_SYSCALL_MATCH_SYM_NAME in asm/ftrace.h and236 implement arch_syscall_match_sym_name with the appropriate logic to return237 true if the function name corresponds with the symbol name.238 - Tag this arch as HAVE_SYSCALL_TRACEPOINTS.239 240 241HAVE_FTRACE_MCOUNT_RECORD242-------------------------243 244See scripts/recordmcount.pl for more info. Just fill in the arch-specific245details for how to locate the addresses of mcount call sites via objdump.246This option doesn't make much sense without also implementing dynamic ftrace.247 248 249HAVE_DYNAMIC_FTRACE250-------------------251 252You will first need HAVE_FTRACE_MCOUNT_RECORD and HAVE_FUNCTION_TRACER, so253scroll your reader back up if you got over eager.254 255Once those are out of the way, you will need to implement:256 - asm/ftrace.h:257 - MCOUNT_ADDR258 - 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 271First you will need to fill out some arch details in your asm/ftrace.h.272 273Define MCOUNT_ADDR as the address of your mcount symbol similar to::274 275 #define MCOUNT_ADDR ((unsigned long)mcount)276 277Since no one else will have a decl for that function, you will need to::278 279 extern void mcount(void);280 281You will also need the helper function ftrace_call_adjust(). Most people282will 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 291Lastly you will need the custom dyn_arch_ftrace structure. If you need292some extra state when runtime patching arbitrary call sites, this is the293place. For now though, create an empty struct::294 295 struct dyn_arch_ftrace {296 /* No extra data needed */297 };298 299With the header out of the way, we can fill out the assembly code. While we300did already create a mcount() function earlier, dynamic ftrace only wants a301stub function. This is because the mcount() will only be used during boot302and then all references to it will be patched out never to return. Instead,303the guts of the old mcount() will be used to create a new ftrace_caller()304function. Because the two are hard to merge, it will most likely be a lot305easier to have two separate definitions split up by #ifdefs. Same goes for306the ftrace_stub() as that will now be inlined in ftrace_caller().307 308Before we get confused anymore, let's check out some pseudo code so you can309implement 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 332This might look a little odd at first, but keep in mind that we will be runtime333patching multiple things. First, only functions that we actually want to trace334will be patched to call ftrace_caller(). Second, since we only have one tracer335active at a time, we will patch the ftrace_caller() function itself to call the336specific tracer in question. That is the point of the ftrace_call label.337 338With that in mind, let's move on to the C code that will actually be doing the339runtime patching. You'll need a little knowledge of your arch's opcodes in340order to make it through the next section.341 342Every arch has an init callback function. If you need to do something early on343to initialize some state, this is the time to do that. Otherwise, this simple344function below should be sufficient for most people::345 346 int __init ftrace_dyn_arch_init(void)347 {348 return 0;349 }350 351There are two functions that are used to do runtime patching of arbitrary352functions. The first is used to turn the mcount call site into a nop (which353is what helps us retain runtime performance when not tracing). The second is354used to turn the mcount call site into a call to an arbitrary location (but355typically that is ftracer_caller()). See the general function definition in356linux/ftrace.h for the functions::357 358 ftrace_make_nop()359 ftrace_make_call()360 361The rec->ip value is the address of the mcount call site that was collected362by the scripts/recordmcount.pl during build time.363 364The last function is used to do runtime patching of the active tracer. This365will be modifying the assembly code at the location of the ftrace_call symbol366inside of the ftrace_caller() function. So you should have sufficient padding367at that location to support the new function calls you'll be inserting. Some368people will be using a "call" type instruction while others will be using a369"branch" type instruction. Specifically, the function is::370 371 ftrace_update_ftrace_func()372 373 374HAVE_DYNAMIC_FTRACE + HAVE_FUNCTION_GRAPH_TRACER375------------------------------------------------376 377The function grapher needs a few tweaks in order to work with dynamic ftrace.378Basically, 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 390Quick 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 new395 ftrace_caller() since some semantics may have changed396 - ftrace_enable_ftrace_graph_caller() will runtime patch the397 ftrace_graph_call location with a call to ftrace_graph_caller()398 - ftrace_disable_ftrace_graph_caller() will runtime patch the399 ftrace_graph_call location with nops400