1 // SPDX-License-Identifier: GPL-2.0
3 * Dynamic function tracing support.
5 * Copyright (C) 2007-2008 Steven Rostedt <srostedt@redhat.com>
7 * Thanks goes to Ingo Molnar, for suggesting the idea.
8 * Mathieu Desnoyers, for suggesting postponing the modifications.
9 * Arjan van de Ven, for keeping me straight, and explaining to me
10 * the dangers of modifying code on the run.
13 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
15 #include <linux/spinlock.h>
16 #include <linux/hardirq.h>
17 #include <linux/uaccess.h>
18 #include <linux/ftrace.h>
19 #include <linux/percpu.h>
20 #include <linux/sched.h>
21 #include <linux/slab.h>
22 #include <linux/init.h>
23 #include <linux/list.h>
24 #include <linux/module.h>
25 #include <linux/memory.h>
26 #include <linux/vmalloc.h>
27 #include <linux/set_memory.h>
28 #include <linux/execmem.h>
30 #include <trace/syscall.h>
32 #include <asm/kprobes.h>
33 #include <asm/ftrace.h>
35 #include <asm/text-patching.h>
37 #ifdef CONFIG_DYNAMIC_FTRACE
39 static int ftrace_poke_late
= 0;
41 void ftrace_arch_code_modify_prepare(void)
42 __acquires(&text_mutex
)
45 * Need to grab text_mutex to prevent a race from module loading
46 * and live kernel patching from changing the text permissions while
47 * ftrace has it set to "read/write".
49 mutex_lock(&text_mutex
);
53 void ftrace_arch_code_modify_post_process(void)
54 __releases(&text_mutex
)
57 * ftrace_make_{call,nop}() may be called during
58 * module load, and we need to finish the text_poke_queue()
63 mutex_unlock(&text_mutex
);
66 static const char *ftrace_nop_replace(void)
71 static const char *ftrace_call_replace(unsigned long ip
, unsigned long addr
)
74 * No need to translate into a callthunk. The trampoline does
75 * the depth accounting itself.
77 return text_gen_insn(CALL_INSN_OPCODE
, (void *)ip
, (void *)addr
);
80 static int ftrace_verify_code(unsigned long ip
, const char *old_code
)
82 char cur_code
[MCOUNT_INSN_SIZE
];
86 * We are paranoid about modifying text, as if a bug was to happen, it
87 * could cause us to read or write to someplace that could cause harm.
88 * Carefully read and modify the code with probe_kernel_*(), and make
89 * sure what we read is what we expected it to be before modifying it.
91 /* read the text we want to modify */
92 if (copy_from_kernel_nofault(cur_code
, (void *)ip
, MCOUNT_INSN_SIZE
)) {
97 /* Make sure it is what we expect it to be */
98 if (memcmp(cur_code
, old_code
, MCOUNT_INSN_SIZE
) != 0) {
99 ftrace_expected
= old_code
;
108 * Marked __ref because it calls text_poke_early() which is .init.text. That is
109 * ok because that call will happen early, during boot, when .init sections are
113 ftrace_modify_code_direct(unsigned long ip
, const char *old_code
,
114 const char *new_code
)
116 int ret
= ftrace_verify_code(ip
, old_code
);
120 /* replace the text with the new text */
121 if (ftrace_poke_late
) {
122 text_poke_queue((void *)ip
, new_code
, MCOUNT_INSN_SIZE
, NULL
);
124 mutex_lock(&text_mutex
);
125 text_poke((void *)ip
, new_code
, MCOUNT_INSN_SIZE
);
126 mutex_unlock(&text_mutex
);
131 int ftrace_make_nop(struct module
*mod
, struct dyn_ftrace
*rec
, unsigned long addr
)
133 unsigned long ip
= rec
->ip
;
134 const char *new, *old
;
136 old
= ftrace_call_replace(ip
, addr
);
137 new = ftrace_nop_replace();
140 * On boot up, and when modules are loaded, the MCOUNT_ADDR
141 * is converted to a nop, and will never become MCOUNT_ADDR
142 * again. This code is either running before SMP (on boot up)
143 * or before the code will ever be executed (module load).
144 * We do not want to use the breakpoint version in this case,
145 * just modify the code directly.
147 if (addr
== MCOUNT_ADDR
)
148 return ftrace_modify_code_direct(ip
, old
, new);
151 * x86 overrides ftrace_replace_code -- this function will never be used
154 WARN_ONCE(1, "invalid use of ftrace_make_nop");
158 int ftrace_make_call(struct dyn_ftrace
*rec
, unsigned long addr
)
160 unsigned long ip
= rec
->ip
;
161 const char *new, *old
;
163 old
= ftrace_nop_replace();
164 new = ftrace_call_replace(ip
, addr
);
166 /* Should only be called when module is loaded */
167 return ftrace_modify_code_direct(rec
->ip
, old
, new);
171 * Should never be called:
172 * As it is only called by __ftrace_replace_code() which is called by
173 * ftrace_replace_code() that x86 overrides, and by ftrace_update_code()
174 * which is called to turn mcount into nops or nops into function calls
175 * but not to convert a function from not using regs to one that uses
176 * regs, which ftrace_modify_call() is for.
178 int ftrace_modify_call(struct dyn_ftrace
*rec
, unsigned long old_addr
,
185 int ftrace_update_ftrace_func(ftrace_func_t func
)
190 ip
= (unsigned long)(&ftrace_call
);
191 new = ftrace_call_replace(ip
, (unsigned long)func
);
192 text_poke_bp((void *)ip
, new, MCOUNT_INSN_SIZE
, NULL
);
194 ip
= (unsigned long)(&ftrace_regs_call
);
195 new = ftrace_call_replace(ip
, (unsigned long)func
);
196 text_poke_bp((void *)ip
, new, MCOUNT_INSN_SIZE
, NULL
);
201 void ftrace_replace_code(int enable
)
203 struct ftrace_rec_iter
*iter
;
204 struct dyn_ftrace
*rec
;
205 const char *new, *old
;
208 for_ftrace_rec_iter(iter
) {
209 rec
= ftrace_rec_iter_record(iter
);
211 switch (ftrace_test_record(rec
, enable
)) {
212 case FTRACE_UPDATE_IGNORE
:
216 case FTRACE_UPDATE_MAKE_CALL
:
217 old
= ftrace_nop_replace();
220 case FTRACE_UPDATE_MODIFY_CALL
:
221 case FTRACE_UPDATE_MAKE_NOP
:
222 old
= ftrace_call_replace(rec
->ip
, ftrace_get_addr_curr(rec
));
226 ret
= ftrace_verify_code(rec
->ip
, old
);
228 ftrace_expected
= old
;
229 ftrace_bug(ret
, rec
);
230 ftrace_expected
= NULL
;
235 for_ftrace_rec_iter(iter
) {
236 rec
= ftrace_rec_iter_record(iter
);
238 switch (ftrace_test_record(rec
, enable
)) {
239 case FTRACE_UPDATE_IGNORE
:
243 case FTRACE_UPDATE_MAKE_CALL
:
244 case FTRACE_UPDATE_MODIFY_CALL
:
245 new = ftrace_call_replace(rec
->ip
, ftrace_get_addr_new(rec
));
248 case FTRACE_UPDATE_MAKE_NOP
:
249 new = ftrace_nop_replace();
253 text_poke_queue((void *)rec
->ip
, new, MCOUNT_INSN_SIZE
, NULL
);
254 ftrace_update_record(rec
, enable
);
259 void arch_ftrace_update_code(int command
)
261 ftrace_modify_all_code(command
);
264 /* Currently only x86_64 supports dynamic trampolines */
267 static inline void *alloc_tramp(unsigned long size
)
269 return execmem_alloc(EXECMEM_FTRACE
, size
);
271 static inline void tramp_free(void *tramp
)
276 /* Defined as markers to the end of the ftrace default trampolines */
277 extern void ftrace_regs_caller_end(void);
278 extern void ftrace_caller_end(void);
279 extern void ftrace_caller_op_ptr(void);
280 extern void ftrace_regs_caller_op_ptr(void);
281 extern void ftrace_regs_caller_jmp(void);
283 /* movq function_trace_op(%rip), %rdx */
284 /* 0x48 0x8b 0x15 <offset-to-ftrace_trace_op (4 bytes)> */
285 #define OP_REF_SIZE 7
288 * The ftrace_ops is passed to the function callback. Since the
289 * trampoline only services a single ftrace_ops, we can pass in
292 * The ftrace_op_code_union is used to create a pointer to the
293 * ftrace_ops that will be passed to the callback function.
295 union ftrace_op_code_union
{
296 char code
[OP_REF_SIZE
];
300 } __attribute__((packed
));
304 (IS_ENABLED(CONFIG_MITIGATION_RETPOLINE) ? 5 : 1 + IS_ENABLED(CONFIG_MITIGATION_SLS))
307 create_trampoline(struct ftrace_ops
*ops
, unsigned int *tramp_size
)
309 unsigned long start_offset
;
310 unsigned long end_offset
;
311 unsigned long op_offset
;
312 unsigned long call_offset
;
313 unsigned long jmp_offset
;
314 unsigned long offset
;
315 unsigned long npages
;
320 /* 48 8b 15 <offset> is movq <offset>(%rip), %rdx */
321 unsigned const char op_ref
[] = { 0x48, 0x8b, 0x15 };
322 unsigned const char retq
[] = { RET_INSN_OPCODE
, INT3_INSN_OPCODE
};
323 union ftrace_op_code_union op_ptr
;
326 if (ops
->flags
& FTRACE_OPS_FL_SAVE_REGS
) {
327 start_offset
= (unsigned long)ftrace_regs_caller
;
328 end_offset
= (unsigned long)ftrace_regs_caller_end
;
329 op_offset
= (unsigned long)ftrace_regs_caller_op_ptr
;
330 call_offset
= (unsigned long)ftrace_regs_call
;
331 jmp_offset
= (unsigned long)ftrace_regs_caller_jmp
;
333 start_offset
= (unsigned long)ftrace_caller
;
334 end_offset
= (unsigned long)ftrace_caller_end
;
335 op_offset
= (unsigned long)ftrace_caller_op_ptr
;
336 call_offset
= (unsigned long)ftrace_call
;
340 size
= end_offset
- start_offset
;
343 * Allocate enough size to store the ftrace_caller code,
344 * the iret , as well as the address of the ftrace_ops this
345 * trampoline is used for.
347 trampoline
= alloc_tramp(size
+ RET_SIZE
+ sizeof(void *));
351 *tramp_size
= size
+ RET_SIZE
+ sizeof(void *);
352 npages
= DIV_ROUND_UP(*tramp_size
, PAGE_SIZE
);
354 /* Copy ftrace_caller onto the trampoline memory */
355 ret
= text_poke_copy(trampoline
, (void *)start_offset
, size
);
359 ip
= trampoline
+ size
;
360 if (cpu_feature_enabled(X86_FEATURE_RETHUNK
))
361 __text_gen_insn(ip
, JMP32_INSN_OPCODE
, ip
, x86_return_thunk
, JMP32_INSN_SIZE
);
363 text_poke_copy(ip
, retq
, sizeof(retq
));
365 /* No need to test direct calls on created trampolines */
366 if (ops
->flags
& FTRACE_OPS_FL_SAVE_REGS
) {
367 /* NOP the jnz 1f; but make sure it's a 2 byte jnz */
368 ip
= trampoline
+ (jmp_offset
- start_offset
);
369 if (WARN_ON(*(char *)ip
!= 0x75))
371 if (!text_poke_copy(ip
, x86_nops
[2], 2))
376 * The address of the ftrace_ops that is used for this trampoline
377 * is stored at the end of the trampoline. This will be used to
378 * load the third parameter for the callback. Basically, that
379 * location at the end of the trampoline takes the place of
380 * the global function_trace_op variable.
383 ptr
= (unsigned long *)(trampoline
+ size
+ RET_SIZE
);
384 text_poke_copy(ptr
, &ops
, sizeof(unsigned long));
386 op_offset
-= start_offset
;
387 memcpy(&op_ptr
, trampoline
+ op_offset
, OP_REF_SIZE
);
389 /* Are we pointing to the reference? */
390 if (WARN_ON(memcmp(op_ptr
.op
, op_ref
, 3) != 0))
393 /* Load the contents of ptr into the callback parameter */
394 offset
= (unsigned long)ptr
;
395 offset
-= (unsigned long)trampoline
+ op_offset
+ OP_REF_SIZE
;
397 op_ptr
.offset
= offset
;
399 /* put in the new offset to the ftrace_ops */
400 text_poke_copy(trampoline
+ op_offset
, &op_ptr
, OP_REF_SIZE
);
402 /* put in the call to the function */
403 mutex_lock(&text_mutex
);
404 call_offset
-= start_offset
;
406 * No need to translate into a callthunk. The trampoline does
407 * the depth accounting before the call already.
409 dest
= ftrace_ops_get_func(ops
);
410 text_poke_copy_locked(trampoline
+ call_offset
,
411 text_gen_insn(CALL_INSN_OPCODE
, trampoline
+ call_offset
, dest
),
412 CALL_INSN_SIZE
, false);
413 mutex_unlock(&text_mutex
);
415 /* ALLOC_TRAMP flags lets us know we created it */
416 ops
->flags
|= FTRACE_OPS_FL_ALLOC_TRAMP
;
418 set_memory_rox((unsigned long)trampoline
, npages
);
419 return (unsigned long)trampoline
;
421 tramp_free(trampoline
);
425 void set_ftrace_ops_ro(void)
427 struct ftrace_ops
*ops
;
428 unsigned long start_offset
;
429 unsigned long end_offset
;
430 unsigned long npages
;
433 do_for_each_ftrace_op(ops
, ftrace_ops_list
) {
434 if (!(ops
->flags
& FTRACE_OPS_FL_ALLOC_TRAMP
))
437 if (ops
->flags
& FTRACE_OPS_FL_SAVE_REGS
) {
438 start_offset
= (unsigned long)ftrace_regs_caller
;
439 end_offset
= (unsigned long)ftrace_regs_caller_end
;
441 start_offset
= (unsigned long)ftrace_caller
;
442 end_offset
= (unsigned long)ftrace_caller_end
;
444 size
= end_offset
- start_offset
;
445 size
= size
+ RET_SIZE
+ sizeof(void *);
446 npages
= DIV_ROUND_UP(size
, PAGE_SIZE
);
447 set_memory_ro((unsigned long)ops
->trampoline
, npages
);
448 } while_for_each_ftrace_op(ops
);
451 static unsigned long calc_trampoline_call_offset(bool save_regs
)
453 unsigned long start_offset
;
454 unsigned long call_offset
;
457 start_offset
= (unsigned long)ftrace_regs_caller
;
458 call_offset
= (unsigned long)ftrace_regs_call
;
460 start_offset
= (unsigned long)ftrace_caller
;
461 call_offset
= (unsigned long)ftrace_call
;
464 return call_offset
- start_offset
;
467 void arch_ftrace_update_trampoline(struct ftrace_ops
*ops
)
470 unsigned long offset
;
475 if (!ops
->trampoline
) {
476 ops
->trampoline
= create_trampoline(ops
, &size
);
477 if (!ops
->trampoline
)
479 ops
->trampoline_size
= size
;
484 * The ftrace_ops caller may set up its own trampoline.
485 * In such a case, this code must not modify it.
487 if (!(ops
->flags
& FTRACE_OPS_FL_ALLOC_TRAMP
))
490 offset
= calc_trampoline_call_offset(ops
->flags
& FTRACE_OPS_FL_SAVE_REGS
);
491 ip
= ops
->trampoline
+ offset
;
492 func
= ftrace_ops_get_func(ops
);
494 mutex_lock(&text_mutex
);
495 /* Do a safe modify in case the trampoline is executing */
496 new = ftrace_call_replace(ip
, (unsigned long)func
);
497 text_poke_bp((void *)ip
, new, MCOUNT_INSN_SIZE
, NULL
);
498 mutex_unlock(&text_mutex
);
501 /* Return the address of the function the trampoline calls */
502 static void *addr_from_call(void *ptr
)
504 union text_poke_insn call
;
507 ret
= copy_from_kernel_nofault(&call
, ptr
, CALL_INSN_SIZE
);
508 if (WARN_ON_ONCE(ret
< 0))
511 /* Make sure this is a call */
512 if (WARN_ON_ONCE(call
.opcode
!= CALL_INSN_OPCODE
)) {
513 pr_warn("Expected E8, got %x\n", call
.opcode
);
517 return ptr
+ CALL_INSN_SIZE
+ call
.disp
;
521 * If the ops->trampoline was not allocated, then it probably
522 * has a static trampoline func, or is the ftrace caller itself.
524 static void *static_tramp_func(struct ftrace_ops
*ops
, struct dyn_ftrace
*rec
)
526 unsigned long offset
;
527 bool save_regs
= rec
->flags
& FTRACE_FL_REGS_EN
;
530 if (ops
&& ops
->trampoline
) {
531 #if !defined(CONFIG_HAVE_DYNAMIC_FTRACE_WITH_ARGS) && \
532 defined(CONFIG_FUNCTION_GRAPH_TRACER)
534 * We only know about function graph tracer setting as static
537 if (ops
->trampoline
== FTRACE_GRAPH_ADDR
)
538 return (void *)prepare_ftrace_return
;
543 offset
= calc_trampoline_call_offset(save_regs
);
546 ptr
= (void *)FTRACE_REGS_ADDR
+ offset
;
548 ptr
= (void *)FTRACE_ADDR
+ offset
;
550 return addr_from_call(ptr
);
553 void *arch_ftrace_trampoline_func(struct ftrace_ops
*ops
, struct dyn_ftrace
*rec
)
555 unsigned long offset
;
557 /* If we didn't allocate this trampoline, consider it static */
558 if (!ops
|| !(ops
->flags
& FTRACE_OPS_FL_ALLOC_TRAMP
))
559 return static_tramp_func(ops
, rec
);
561 offset
= calc_trampoline_call_offset(ops
->flags
& FTRACE_OPS_FL_SAVE_REGS
);
562 return addr_from_call((void *)ops
->trampoline
+ offset
);
565 void arch_ftrace_trampoline_free(struct ftrace_ops
*ops
)
567 if (!ops
|| !(ops
->flags
& FTRACE_OPS_FL_ALLOC_TRAMP
))
570 tramp_free((void *)ops
->trampoline
);
574 #endif /* CONFIG_X86_64 */
575 #endif /* CONFIG_DYNAMIC_FTRACE */
577 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
579 #if defined(CONFIG_DYNAMIC_FTRACE) && !defined(CONFIG_HAVE_DYNAMIC_FTRACE_WITH_ARGS)
580 extern void ftrace_graph_call(void);
581 static const char *ftrace_jmp_replace(unsigned long ip
, unsigned long addr
)
583 return text_gen_insn(JMP32_INSN_OPCODE
, (void *)ip
, (void *)addr
);
586 static int ftrace_mod_jmp(unsigned long ip
, void *func
)
590 new = ftrace_jmp_replace(ip
, (unsigned long)func
);
591 text_poke_bp((void *)ip
, new, MCOUNT_INSN_SIZE
, NULL
);
595 int ftrace_enable_ftrace_graph_caller(void)
597 unsigned long ip
= (unsigned long)(&ftrace_graph_call
);
599 return ftrace_mod_jmp(ip
, &ftrace_graph_caller
);
602 int ftrace_disable_ftrace_graph_caller(void)
604 unsigned long ip
= (unsigned long)(&ftrace_graph_call
);
606 return ftrace_mod_jmp(ip
, &ftrace_stub
);
608 #endif /* CONFIG_DYNAMIC_FTRACE && !CONFIG_HAVE_DYNAMIC_FTRACE_WITH_ARGS */
611 * Hook the return address and push it in the stack of return addrs
612 * in current thread info.
614 void prepare_ftrace_return(unsigned long ip
, unsigned long *parent
,
615 unsigned long frame_pointer
)
617 unsigned long return_hooker
= (unsigned long)&return_to_handler
;
621 * When resuming from suspend-to-ram, this function can be indirectly
622 * called from early CPU startup code while the CPU is in real mode,
623 * which would fail miserably. Make sure the stack pointer is a
626 * This check isn't as accurate as virt_addr_valid(), but it should be
627 * good enough for this purpose, and it's fast.
629 if (unlikely((long)__builtin_frame_address(0) >= 0))
632 if (unlikely(ftrace_graph_is_dead()))
635 if (unlikely(atomic_read(¤t
->tracing_graph_pause
)))
638 bit
= ftrace_test_recursion_trylock(ip
, *parent
);
642 if (!function_graph_enter(*parent
, ip
, frame_pointer
, parent
))
643 *parent
= return_hooker
;
645 ftrace_test_recursion_unlock(bit
);
648 #ifdef CONFIG_HAVE_DYNAMIC_FTRACE_WITH_ARGS
649 void ftrace_graph_func(unsigned long ip
, unsigned long parent_ip
,
650 struct ftrace_ops
*op
, struct ftrace_regs
*fregs
)
652 struct pt_regs
*regs
= &arch_ftrace_regs(fregs
)->regs
;
653 unsigned long *stack
= (unsigned long *)kernel_stack_pointer(regs
);
655 prepare_ftrace_return(ip
, (unsigned long *)stack
, 0);
659 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */