Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2#include <linux/module.h>
3#include <linux/sort.h>
4#include <asm/ptrace.h>
5#include <asm/stacktrace.h>
6#include <asm/unwind.h>
7#include <asm/orc_types.h>
8#include <asm/orc_lookup.h>
9
10#define orc_warn(fmt, ...) \
11 printk_deferred_once(KERN_WARNING pr_fmt("WARNING: " fmt), ##__VA_ARGS__)
12
13extern int __start_orc_unwind_ip[];
14extern int __stop_orc_unwind_ip[];
15extern struct orc_entry __start_orc_unwind[];
16extern struct orc_entry __stop_orc_unwind[];
17
18static DEFINE_MUTEX(sort_mutex);
19int *cur_orc_ip_table = __start_orc_unwind_ip;
20struct orc_entry *cur_orc_table = __start_orc_unwind;
21
22unsigned int lookup_num_blocks;
23bool orc_init;
24
25static inline unsigned long orc_ip(const int *ip)
26{
27 return (unsigned long)ip + *ip;
28}
29
30static struct orc_entry *__orc_find(int *ip_table, struct orc_entry *u_table,
31 unsigned int num_entries, unsigned long ip)
32{
33 int *first = ip_table;
34 int *last = ip_table + num_entries - 1;
35 int *mid = first, *found = first;
36
37 if (!num_entries)
38 return NULL;
39
40 /*
41 * Do a binary range search to find the rightmost duplicate of a given
42 * starting address. Some entries are section terminators which are
43 * "weak" entries for ensuring there are no gaps. They should be
44 * ignored when they conflict with a real entry.
45 */
46 while (first <= last) {
47 mid = first + ((last - first) / 2);
48
49 if (orc_ip(mid) <= ip) {
50 found = mid;
51 first = mid + 1;
52 } else
53 last = mid - 1;
54 }
55
56 return u_table + (found - ip_table);
57}
58
59#ifdef CONFIG_MODULES
60static struct orc_entry *orc_module_find(unsigned long ip)
61{
62 struct module *mod;
63
64 mod = __module_address(ip);
65 if (!mod || !mod->arch.orc_unwind || !mod->arch.orc_unwind_ip)
66 return NULL;
67 return __orc_find(mod->arch.orc_unwind_ip, mod->arch.orc_unwind,
68 mod->arch.num_orcs, ip);
69}
70#else
71static struct orc_entry *orc_module_find(unsigned long ip)
72{
73 return NULL;
74}
75#endif
76
77#ifdef CONFIG_DYNAMIC_FTRACE
78static struct orc_entry *orc_find(unsigned long ip);
79
80/*
81 * Ftrace dynamic trampolines do not have orc entries of their own.
82 * But they are copies of the ftrace entries that are static and
83 * defined in ftrace_*.S, which do have orc entries.
84 *
85 * If the unwinder comes across a ftrace trampoline, then find the
86 * ftrace function that was used to create it, and use that ftrace
87 * function's orc entry, as the placement of the return code in
88 * the stack will be identical.
89 */
90static struct orc_entry *orc_ftrace_find(unsigned long ip)
91{
92 struct ftrace_ops *ops;
93 unsigned long caller;
94
95 ops = ftrace_ops_trampoline(ip);
96 if (!ops)
97 return NULL;
98
99 if (ops->flags & FTRACE_OPS_FL_SAVE_REGS)
100 caller = (unsigned long)ftrace_regs_call;
101 else
102 caller = (unsigned long)ftrace_call;
103
104 /* Prevent unlikely recursion */
105 if (ip == caller)
106 return NULL;
107
108 return orc_find(caller);
109}
110#else
111static struct orc_entry *orc_ftrace_find(unsigned long ip)
112{
113 return NULL;
114}
115#endif
116
117/*
118 * If we crash with IP==0, the last successfully executed instruction
119 * was probably an indirect function call with a NULL function pointer,
120 * and we don't have unwind information for NULL.
121 * This hardcoded ORC entry for IP==0 allows us to unwind from a NULL function
122 * pointer into its parent and then continue normally from there.
123 */
124static struct orc_entry null_orc_entry = {
125 .sp_offset = sizeof(long),
126 .sp_reg = ORC_REG_SP,
127 .bp_reg = ORC_REG_UNDEFINED,
128 .type = ORC_TYPE_CALL
129};
130
131/* Fake frame pointer entry -- used as a fallback for generated code */
132static struct orc_entry orc_fp_entry = {
133 .type = ORC_TYPE_CALL,
134 .sp_reg = ORC_REG_BP,
135 .sp_offset = 16,
136 .bp_reg = ORC_REG_PREV_SP,
137 .bp_offset = -16,
138 .end = 0,
139};
140
141static struct orc_entry *orc_find(unsigned long ip)
142{
143 static struct orc_entry *orc;
144
145 if (!orc_init)
146 return NULL;
147
148 if (ip == 0)
149 return &null_orc_entry;
150
151 /* For non-init vmlinux addresses, use the fast lookup table: */
152 if (ip >= LOOKUP_START_IP && ip < LOOKUP_STOP_IP) {
153 unsigned int idx, start, stop;
154
155 idx = (ip - LOOKUP_START_IP) / LOOKUP_BLOCK_SIZE;
156
157 if (unlikely((idx >= lookup_num_blocks-1))) {
158 orc_warn("WARNING: bad lookup idx: idx=%u num=%u ip=%pB\n",
159 idx, lookup_num_blocks, (void *)ip);
160 return NULL;
161 }
162
163 start = orc_lookup[idx];
164 stop = orc_lookup[idx + 1] + 1;
165
166 if (unlikely((__start_orc_unwind + start >= __stop_orc_unwind) ||
167 (__start_orc_unwind + stop > __stop_orc_unwind))) {
168 orc_warn("WARNING: bad lookup value: idx=%u num=%u start=%u stop=%u ip=%pB\n",
169 idx, lookup_num_blocks, start, stop, (void *)ip);
170 return NULL;
171 }
172
173 return __orc_find(__start_orc_unwind_ip + start,
174 __start_orc_unwind + start, stop - start, ip);
175 }
176
177 /* vmlinux .init slow lookup: */
178 if (init_kernel_text(ip))
179 return __orc_find(__start_orc_unwind_ip, __start_orc_unwind,
180 __stop_orc_unwind_ip - __start_orc_unwind_ip, ip);
181
182 /* Module lookup: */
183 orc = orc_module_find(ip);
184 if (orc)
185 return orc;
186
187 return orc_ftrace_find(ip);
188}
189
190static void orc_sort_swap(void *_a, void *_b, int size)
191{
192 struct orc_entry *orc_a, *orc_b;
193 struct orc_entry orc_tmp;
194 int *a = _a, *b = _b, tmp;
195 int delta = _b - _a;
196
197 /* Swap the .orc_unwind_ip entries: */
198 tmp = *a;
199 *a = *b + delta;
200 *b = tmp - delta;
201
202 /* Swap the corresponding .orc_unwind entries: */
203 orc_a = cur_orc_table + (a - cur_orc_ip_table);
204 orc_b = cur_orc_table + (b - cur_orc_ip_table);
205 orc_tmp = *orc_a;
206 *orc_a = *orc_b;
207 *orc_b = orc_tmp;
208}
209
210static int orc_sort_cmp(const void *_a, const void *_b)
211{
212 struct orc_entry *orc_a;
213 const int *a = _a, *b = _b;
214 unsigned long a_val = orc_ip(a);
215 unsigned long b_val = orc_ip(b);
216
217 if (a_val > b_val)
218 return 1;
219 if (a_val < b_val)
220 return -1;
221
222 /*
223 * The "weak" section terminator entries need to always be on the left
224 * to ensure the lookup code skips them in favor of real entries.
225 * These terminator entries exist to handle any gaps created by
226 * whitelisted .o files which didn't get objtool generation.
227 */
228 orc_a = cur_orc_table + (a - cur_orc_ip_table);
229 return orc_a->sp_reg == ORC_REG_UNDEFINED && !orc_a->end ? -1 : 1;
230}
231
232#ifdef CONFIG_MODULES
233void unwind_module_init(struct module *mod, void *_orc_ip, size_t orc_ip_size,
234 void *_orc, size_t orc_size)
235{
236 int *orc_ip = _orc_ip;
237 struct orc_entry *orc = _orc;
238 unsigned int num_entries = orc_ip_size / sizeof(int);
239
240 WARN_ON_ONCE(orc_ip_size % sizeof(int) != 0 ||
241 orc_size % sizeof(*orc) != 0 ||
242 num_entries != orc_size / sizeof(*orc));
243
244 /*
245 * The 'cur_orc_*' globals allow the orc_sort_swap() callback to
246 * associate an .orc_unwind_ip table entry with its corresponding
247 * .orc_unwind entry so they can both be swapped.
248 */
249 mutex_lock(&sort_mutex);
250 cur_orc_ip_table = orc_ip;
251 cur_orc_table = orc;
252 sort(orc_ip, num_entries, sizeof(int), orc_sort_cmp, orc_sort_swap);
253 mutex_unlock(&sort_mutex);
254
255 mod->arch.orc_unwind_ip = orc_ip;
256 mod->arch.orc_unwind = orc;
257 mod->arch.num_orcs = num_entries;
258}
259#endif
260
261void __init unwind_init(void)
262{
263 size_t orc_ip_size = (void *)__stop_orc_unwind_ip - (void *)__start_orc_unwind_ip;
264 size_t orc_size = (void *)__stop_orc_unwind - (void *)__start_orc_unwind;
265 size_t num_entries = orc_ip_size / sizeof(int);
266 struct orc_entry *orc;
267 int i;
268
269 if (!num_entries || orc_ip_size % sizeof(int) != 0 ||
270 orc_size % sizeof(struct orc_entry) != 0 ||
271 num_entries != orc_size / sizeof(struct orc_entry)) {
272 orc_warn("WARNING: Bad or missing .orc_unwind table. Disabling unwinder.\n");
273 return;
274 }
275
276 /* Sort the .orc_unwind and .orc_unwind_ip tables: */
277 sort(__start_orc_unwind_ip, num_entries, sizeof(int), orc_sort_cmp,
278 orc_sort_swap);
279
280 /* Initialize the fast lookup table: */
281 lookup_num_blocks = orc_lookup_end - orc_lookup;
282 for (i = 0; i < lookup_num_blocks-1; i++) {
283 orc = __orc_find(__start_orc_unwind_ip, __start_orc_unwind,
284 num_entries,
285 LOOKUP_START_IP + (LOOKUP_BLOCK_SIZE * i));
286 if (!orc) {
287 orc_warn("WARNING: Corrupt .orc_unwind table. Disabling unwinder.\n");
288 return;
289 }
290
291 orc_lookup[i] = orc - __start_orc_unwind;
292 }
293
294 /* Initialize the ending block: */
295 orc = __orc_find(__start_orc_unwind_ip, __start_orc_unwind, num_entries,
296 LOOKUP_STOP_IP);
297 if (!orc) {
298 orc_warn("WARNING: Corrupt .orc_unwind table. Disabling unwinder.\n");
299 return;
300 }
301 orc_lookup[lookup_num_blocks-1] = orc - __start_orc_unwind;
302
303 orc_init = true;
304}
305
306unsigned long unwind_get_return_address(struct unwind_state *state)
307{
308 if (unwind_done(state))
309 return 0;
310
311 return __kernel_text_address(state->ip) ? state->ip : 0;
312}
313EXPORT_SYMBOL_GPL(unwind_get_return_address);
314
315unsigned long *unwind_get_return_address_ptr(struct unwind_state *state)
316{
317 if (unwind_done(state))
318 return NULL;
319
320 if (state->regs)
321 return &state->regs->ip;
322
323 if (state->sp)
324 return (unsigned long *)state->sp - 1;
325
326 return NULL;
327}
328
329static bool stack_access_ok(struct unwind_state *state, unsigned long _addr,
330 size_t len)
331{
332 struct stack_info *info = &state->stack_info;
333 void *addr = (void *)_addr;
334
335 if (!on_stack(info, addr, len) &&
336 (get_stack_info(addr, state->task, info, &state->stack_mask)))
337 return false;
338
339 return true;
340}
341
342static bool deref_stack_reg(struct unwind_state *state, unsigned long addr,
343 unsigned long *val)
344{
345 if (!stack_access_ok(state, addr, sizeof(long)))
346 return false;
347
348 *val = READ_ONCE_NOCHECK(*(unsigned long *)addr);
349 return true;
350}
351
352static bool deref_stack_regs(struct unwind_state *state, unsigned long addr,
353 unsigned long *ip, unsigned long *sp)
354{
355 struct pt_regs *regs = (struct pt_regs *)addr;
356
357 /* x86-32 support will be more complicated due to the ®s->sp hack */
358 BUILD_BUG_ON(IS_ENABLED(CONFIG_X86_32));
359
360 if (!stack_access_ok(state, addr, sizeof(struct pt_regs)))
361 return false;
362
363 *ip = regs->ip;
364 *sp = regs->sp;
365 return true;
366}
367
368static bool deref_stack_iret_regs(struct unwind_state *state, unsigned long addr,
369 unsigned long *ip, unsigned long *sp)
370{
371 struct pt_regs *regs = (void *)addr - IRET_FRAME_OFFSET;
372
373 if (!stack_access_ok(state, addr, IRET_FRAME_SIZE))
374 return false;
375
376 *ip = regs->ip;
377 *sp = regs->sp;
378 return true;
379}
380
381bool unwind_next_frame(struct unwind_state *state)
382{
383 unsigned long ip_p, sp, orig_ip = state->ip, prev_sp = state->sp;
384 enum stack_type prev_type = state->stack_info.type;
385 struct orc_entry *orc;
386 bool indirect = false;
387
388 if (unwind_done(state))
389 return false;
390
391 /* Don't let modules unload while we're reading their ORC data. */
392 preempt_disable();
393
394 /* End-of-stack check for user tasks: */
395 if (state->regs && user_mode(state->regs))
396 goto the_end;
397
398 /*
399 * Find the orc_entry associated with the text address.
400 *
401 * Decrement call return addresses by one so they work for sibling
402 * calls and calls to noreturn functions.
403 */
404 orc = orc_find(state->signal ? state->ip : state->ip - 1);
405 if (!orc) {
406 /*
407 * As a fallback, try to assume this code uses a frame pointer.
408 * This is useful for generated code, like BPF, which ORC
409 * doesn't know about. This is just a guess, so the rest of
410 * the unwind is no longer considered reliable.
411 */
412 orc = &orc_fp_entry;
413 state->error = true;
414 }
415
416 /* End-of-stack check for kernel threads: */
417 if (orc->sp_reg == ORC_REG_UNDEFINED) {
418 if (!orc->end)
419 goto err;
420
421 goto the_end;
422 }
423
424 /* Find the previous frame's stack: */
425 switch (orc->sp_reg) {
426 case ORC_REG_SP:
427 sp = state->sp + orc->sp_offset;
428 break;
429
430 case ORC_REG_BP:
431 sp = state->bp + orc->sp_offset;
432 break;
433
434 case ORC_REG_SP_INDIRECT:
435 sp = state->sp + orc->sp_offset;
436 indirect = true;
437 break;
438
439 case ORC_REG_BP_INDIRECT:
440 sp = state->bp + orc->sp_offset;
441 indirect = true;
442 break;
443
444 case ORC_REG_R10:
445 if (!state->regs || !state->full_regs) {
446 orc_warn("missing regs for base reg R10 at ip %pB\n",
447 (void *)state->ip);
448 goto err;
449 }
450 sp = state->regs->r10;
451 break;
452
453 case ORC_REG_R13:
454 if (!state->regs || !state->full_regs) {
455 orc_warn("missing regs for base reg R13 at ip %pB\n",
456 (void *)state->ip);
457 goto err;
458 }
459 sp = state->regs->r13;
460 break;
461
462 case ORC_REG_DI:
463 if (!state->regs || !state->full_regs) {
464 orc_warn("missing regs for base reg DI at ip %pB\n",
465 (void *)state->ip);
466 goto err;
467 }
468 sp = state->regs->di;
469 break;
470
471 case ORC_REG_DX:
472 if (!state->regs || !state->full_regs) {
473 orc_warn("missing regs for base reg DX at ip %pB\n",
474 (void *)state->ip);
475 goto err;
476 }
477 sp = state->regs->dx;
478 break;
479
480 default:
481 orc_warn("unknown SP base reg %d for ip %pB\n",
482 orc->sp_reg, (void *)state->ip);
483 goto err;
484 }
485
486 if (indirect) {
487 if (!deref_stack_reg(state, sp, &sp))
488 goto err;
489 }
490
491 /* Find IP, SP and possibly regs: */
492 switch (orc->type) {
493 case ORC_TYPE_CALL:
494 ip_p = sp - sizeof(long);
495
496 if (!deref_stack_reg(state, ip_p, &state->ip))
497 goto err;
498
499 state->ip = ftrace_graph_ret_addr(state->task, &state->graph_idx,
500 state->ip, (void *)ip_p);
501
502 state->sp = sp;
503 state->regs = NULL;
504 state->signal = false;
505 break;
506
507 case ORC_TYPE_REGS:
508 if (!deref_stack_regs(state, sp, &state->ip, &state->sp)) {
509 orc_warn("can't dereference registers at %p for ip %pB\n",
510 (void *)sp, (void *)orig_ip);
511 goto err;
512 }
513
514 state->regs = (struct pt_regs *)sp;
515 state->full_regs = true;
516 state->signal = true;
517 break;
518
519 case ORC_TYPE_REGS_IRET:
520 if (!deref_stack_iret_regs(state, sp, &state->ip, &state->sp)) {
521 orc_warn("can't dereference iret registers at %p for ip %pB\n",
522 (void *)sp, (void *)orig_ip);
523 goto err;
524 }
525
526 state->regs = (void *)sp - IRET_FRAME_OFFSET;
527 state->full_regs = false;
528 state->signal = true;
529 break;
530
531 default:
532 orc_warn("unknown .orc_unwind entry type %d for ip %pB\n",
533 orc->type, (void *)orig_ip);
534 break;
535 }
536
537 /* Find BP: */
538 switch (orc->bp_reg) {
539 case ORC_REG_UNDEFINED:
540 if (state->regs && state->full_regs)
541 state->bp = state->regs->bp;
542 break;
543
544 case ORC_REG_PREV_SP:
545 if (!deref_stack_reg(state, sp + orc->bp_offset, &state->bp))
546 goto err;
547 break;
548
549 case ORC_REG_BP:
550 if (!deref_stack_reg(state, state->bp + orc->bp_offset, &state->bp))
551 goto err;
552 break;
553
554 default:
555 orc_warn("unknown BP base reg %d for ip %pB\n",
556 orc->bp_reg, (void *)orig_ip);
557 goto err;
558 }
559
560 /* Prevent a recursive loop due to bad ORC data: */
561 if (state->stack_info.type == prev_type &&
562 on_stack(&state->stack_info, (void *)state->sp, sizeof(long)) &&
563 state->sp <= prev_sp) {
564 orc_warn("stack going in the wrong direction? ip=%pB\n",
565 (void *)orig_ip);
566 goto err;
567 }
568
569 preempt_enable();
570 return true;
571
572err:
573 state->error = true;
574
575the_end:
576 preempt_enable();
577 state->stack_info.type = STACK_TYPE_UNKNOWN;
578 return false;
579}
580EXPORT_SYMBOL_GPL(unwind_next_frame);
581
582void __unwind_start(struct unwind_state *state, struct task_struct *task,
583 struct pt_regs *regs, unsigned long *first_frame)
584{
585 memset(state, 0, sizeof(*state));
586 state->task = task;
587
588 /*
589 * Refuse to unwind the stack of a task while it's executing on another
590 * CPU. This check is racy, but that's ok: the unwinder has other
591 * checks to prevent it from going off the rails.
592 */
593 if (task_on_another_cpu(task))
594 goto done;
595
596 if (regs) {
597 if (user_mode(regs))
598 goto done;
599
600 state->ip = regs->ip;
601 state->sp = regs->sp;
602 state->bp = regs->bp;
603 state->regs = regs;
604 state->full_regs = true;
605 state->signal = true;
606
607 } else if (task == current) {
608 asm volatile("lea (%%rip), %0\n\t"
609 "mov %%rsp, %1\n\t"
610 "mov %%rbp, %2\n\t"
611 : "=r" (state->ip), "=r" (state->sp),
612 "=r" (state->bp));
613
614 } else {
615 struct inactive_task_frame *frame = (void *)task->thread.sp;
616
617 state->sp = task->thread.sp;
618 state->bp = READ_ONCE_NOCHECK(frame->bp);
619 state->ip = READ_ONCE_NOCHECK(frame->ret_addr);
620 }
621
622 if (get_stack_info((unsigned long *)state->sp, state->task,
623 &state->stack_info, &state->stack_mask)) {
624 /*
625 * We weren't on a valid stack. It's possible that
626 * we overflowed a valid stack into a guard page.
627 * See if the next page up is valid so that we can
628 * generate some kind of backtrace if this happens.
629 */
630 void *next_page = (void *)PAGE_ALIGN((unsigned long)state->sp);
631 if (get_stack_info(next_page, state->task, &state->stack_info,
632 &state->stack_mask))
633 return;
634 }
635
636 /*
637 * The caller can provide the address of the first frame directly
638 * (first_frame) or indirectly (regs->sp) to indicate which stack frame
639 * to start unwinding at. Skip ahead until we reach it.
640 */
641
642 /* When starting from regs, skip the regs frame: */
643 if (regs) {
644 unwind_next_frame(state);
645 return;
646 }
647
648 /* Otherwise, skip ahead to the user-specified starting frame: */
649 while (!unwind_done(state) &&
650 (!on_stack(&state->stack_info, first_frame, sizeof(long)) ||
651 state->sp <= (unsigned long)first_frame))
652 unwind_next_frame(state);
653
654 return;
655
656done:
657 state->stack_info.type = STACK_TYPE_UNKNOWN;
658 return;
659}
660EXPORT_SYMBOL_GPL(__unwind_start);
1// SPDX-License-Identifier: GPL-2.0-only
2#include <linux/objtool.h>
3#include <linux/module.h>
4#include <linux/sort.h>
5#include <asm/ptrace.h>
6#include <asm/stacktrace.h>
7#include <asm/unwind.h>
8#include <asm/orc_types.h>
9#include <asm/orc_lookup.h>
10#include <asm/orc_header.h>
11
12ORC_HEADER;
13
14#define orc_warn(fmt, ...) \
15 printk_deferred_once(KERN_WARNING "WARNING: " fmt, ##__VA_ARGS__)
16
17#define orc_warn_current(args...) \
18({ \
19 static bool dumped_before; \
20 if (state->task == current && !state->error) { \
21 orc_warn(args); \
22 if (unwind_debug && !dumped_before) { \
23 dumped_before = true; \
24 unwind_dump(state); \
25 } \
26 } \
27})
28
29extern int __start_orc_unwind_ip[];
30extern int __stop_orc_unwind_ip[];
31extern struct orc_entry __start_orc_unwind[];
32extern struct orc_entry __stop_orc_unwind[];
33
34static bool orc_init __ro_after_init;
35static bool unwind_debug __ro_after_init;
36static unsigned int lookup_num_blocks __ro_after_init;
37
38static int __init unwind_debug_cmdline(char *str)
39{
40 unwind_debug = true;
41
42 return 0;
43}
44early_param("unwind_debug", unwind_debug_cmdline);
45
46static void unwind_dump(struct unwind_state *state)
47{
48 static bool dumped_before;
49 unsigned long word, *sp;
50 struct stack_info stack_info = {0};
51 unsigned long visit_mask = 0;
52
53 if (dumped_before)
54 return;
55
56 dumped_before = true;
57
58 printk_deferred("unwind stack type:%d next_sp:%p mask:0x%lx graph_idx:%d\n",
59 state->stack_info.type, state->stack_info.next_sp,
60 state->stack_mask, state->graph_idx);
61
62 for (sp = __builtin_frame_address(0); sp;
63 sp = PTR_ALIGN(stack_info.next_sp, sizeof(long))) {
64 if (get_stack_info(sp, state->task, &stack_info, &visit_mask))
65 break;
66
67 for (; sp < stack_info.end; sp++) {
68
69 word = READ_ONCE_NOCHECK(*sp);
70
71 printk_deferred("%0*lx: %0*lx (%pB)\n", BITS_PER_LONG/4,
72 (unsigned long)sp, BITS_PER_LONG/4,
73 word, (void *)word);
74 }
75 }
76}
77
78static inline unsigned long orc_ip(const int *ip)
79{
80 return (unsigned long)ip + *ip;
81}
82
83static struct orc_entry *__orc_find(int *ip_table, struct orc_entry *u_table,
84 unsigned int num_entries, unsigned long ip)
85{
86 int *first = ip_table;
87 int *last = ip_table + num_entries - 1;
88 int *mid, *found = first;
89
90 if (!num_entries)
91 return NULL;
92
93 /*
94 * Do a binary range search to find the rightmost duplicate of a given
95 * starting address. Some entries are section terminators which are
96 * "weak" entries for ensuring there are no gaps. They should be
97 * ignored when they conflict with a real entry.
98 */
99 while (first <= last) {
100 mid = first + ((last - first) / 2);
101
102 if (orc_ip(mid) <= ip) {
103 found = mid;
104 first = mid + 1;
105 } else
106 last = mid - 1;
107 }
108
109 return u_table + (found - ip_table);
110}
111
112#ifdef CONFIG_MODULES
113static struct orc_entry *orc_module_find(unsigned long ip)
114{
115 struct module *mod;
116
117 mod = __module_address(ip);
118 if (!mod || !mod->arch.orc_unwind || !mod->arch.orc_unwind_ip)
119 return NULL;
120 return __orc_find(mod->arch.orc_unwind_ip, mod->arch.orc_unwind,
121 mod->arch.num_orcs, ip);
122}
123#else
124static struct orc_entry *orc_module_find(unsigned long ip)
125{
126 return NULL;
127}
128#endif
129
130#ifdef CONFIG_DYNAMIC_FTRACE
131static struct orc_entry *orc_find(unsigned long ip);
132
133/*
134 * Ftrace dynamic trampolines do not have orc entries of their own.
135 * But they are copies of the ftrace entries that are static and
136 * defined in ftrace_*.S, which do have orc entries.
137 *
138 * If the unwinder comes across a ftrace trampoline, then find the
139 * ftrace function that was used to create it, and use that ftrace
140 * function's orc entry, as the placement of the return code in
141 * the stack will be identical.
142 */
143static struct orc_entry *orc_ftrace_find(unsigned long ip)
144{
145 struct ftrace_ops *ops;
146 unsigned long tramp_addr, offset;
147
148 ops = ftrace_ops_trampoline(ip);
149 if (!ops)
150 return NULL;
151
152 /* Set tramp_addr to the start of the code copied by the trampoline */
153 if (ops->flags & FTRACE_OPS_FL_SAVE_REGS)
154 tramp_addr = (unsigned long)ftrace_regs_caller;
155 else
156 tramp_addr = (unsigned long)ftrace_caller;
157
158 /* Now place tramp_addr to the location within the trampoline ip is at */
159 offset = ip - ops->trampoline;
160 tramp_addr += offset;
161
162 /* Prevent unlikely recursion */
163 if (ip == tramp_addr)
164 return NULL;
165
166 return orc_find(tramp_addr);
167}
168#else
169static struct orc_entry *orc_ftrace_find(unsigned long ip)
170{
171 return NULL;
172}
173#endif
174
175/*
176 * If we crash with IP==0, the last successfully executed instruction
177 * was probably an indirect function call with a NULL function pointer,
178 * and we don't have unwind information for NULL.
179 * This hardcoded ORC entry for IP==0 allows us to unwind from a NULL function
180 * pointer into its parent and then continue normally from there.
181 */
182static struct orc_entry null_orc_entry = {
183 .sp_offset = sizeof(long),
184 .sp_reg = ORC_REG_SP,
185 .bp_reg = ORC_REG_UNDEFINED,
186 .type = ORC_TYPE_CALL
187};
188
189/* Fake frame pointer entry -- used as a fallback for generated code */
190static struct orc_entry orc_fp_entry = {
191 .type = ORC_TYPE_CALL,
192 .sp_reg = ORC_REG_BP,
193 .sp_offset = 16,
194 .bp_reg = ORC_REG_PREV_SP,
195 .bp_offset = -16,
196};
197
198static struct orc_entry *orc_find(unsigned long ip)
199{
200 static struct orc_entry *orc;
201
202 if (ip == 0)
203 return &null_orc_entry;
204
205 /* For non-init vmlinux addresses, use the fast lookup table: */
206 if (ip >= LOOKUP_START_IP && ip < LOOKUP_STOP_IP) {
207 unsigned int idx, start, stop;
208
209 idx = (ip - LOOKUP_START_IP) / LOOKUP_BLOCK_SIZE;
210
211 if (unlikely((idx >= lookup_num_blocks-1))) {
212 orc_warn("WARNING: bad lookup idx: idx=%u num=%u ip=%pB\n",
213 idx, lookup_num_blocks, (void *)ip);
214 return NULL;
215 }
216
217 start = orc_lookup[idx];
218 stop = orc_lookup[idx + 1] + 1;
219
220 if (unlikely((__start_orc_unwind + start >= __stop_orc_unwind) ||
221 (__start_orc_unwind + stop > __stop_orc_unwind))) {
222 orc_warn("WARNING: bad lookup value: idx=%u num=%u start=%u stop=%u ip=%pB\n",
223 idx, lookup_num_blocks, start, stop, (void *)ip);
224 return NULL;
225 }
226
227 return __orc_find(__start_orc_unwind_ip + start,
228 __start_orc_unwind + start, stop - start, ip);
229 }
230
231 /* vmlinux .init slow lookup: */
232 if (is_kernel_inittext(ip))
233 return __orc_find(__start_orc_unwind_ip, __start_orc_unwind,
234 __stop_orc_unwind_ip - __start_orc_unwind_ip, ip);
235
236 /* Module lookup: */
237 orc = orc_module_find(ip);
238 if (orc)
239 return orc;
240
241 return orc_ftrace_find(ip);
242}
243
244#ifdef CONFIG_MODULES
245
246static DEFINE_MUTEX(sort_mutex);
247static int *cur_orc_ip_table = __start_orc_unwind_ip;
248static struct orc_entry *cur_orc_table = __start_orc_unwind;
249
250static void orc_sort_swap(void *_a, void *_b, int size)
251{
252 struct orc_entry *orc_a, *orc_b;
253 int *a = _a, *b = _b, tmp;
254 int delta = _b - _a;
255
256 /* Swap the .orc_unwind_ip entries: */
257 tmp = *a;
258 *a = *b + delta;
259 *b = tmp - delta;
260
261 /* Swap the corresponding .orc_unwind entries: */
262 orc_a = cur_orc_table + (a - cur_orc_ip_table);
263 orc_b = cur_orc_table + (b - cur_orc_ip_table);
264 swap(*orc_a, *orc_b);
265}
266
267static int orc_sort_cmp(const void *_a, const void *_b)
268{
269 struct orc_entry *orc_a;
270 const int *a = _a, *b = _b;
271 unsigned long a_val = orc_ip(a);
272 unsigned long b_val = orc_ip(b);
273
274 if (a_val > b_val)
275 return 1;
276 if (a_val < b_val)
277 return -1;
278
279 /*
280 * The "weak" section terminator entries need to always be first
281 * to ensure the lookup code skips them in favor of real entries.
282 * These terminator entries exist to handle any gaps created by
283 * whitelisted .o files which didn't get objtool generation.
284 */
285 orc_a = cur_orc_table + (a - cur_orc_ip_table);
286 return orc_a->type == ORC_TYPE_UNDEFINED ? -1 : 1;
287}
288
289void unwind_module_init(struct module *mod, void *_orc_ip, size_t orc_ip_size,
290 void *_orc, size_t orc_size)
291{
292 int *orc_ip = _orc_ip;
293 struct orc_entry *orc = _orc;
294 unsigned int num_entries = orc_ip_size / sizeof(int);
295
296 WARN_ON_ONCE(orc_ip_size % sizeof(int) != 0 ||
297 orc_size % sizeof(*orc) != 0 ||
298 num_entries != orc_size / sizeof(*orc));
299
300 /*
301 * The 'cur_orc_*' globals allow the orc_sort_swap() callback to
302 * associate an .orc_unwind_ip table entry with its corresponding
303 * .orc_unwind entry so they can both be swapped.
304 */
305 mutex_lock(&sort_mutex);
306 cur_orc_ip_table = orc_ip;
307 cur_orc_table = orc;
308 sort(orc_ip, num_entries, sizeof(int), orc_sort_cmp, orc_sort_swap);
309 mutex_unlock(&sort_mutex);
310
311 mod->arch.orc_unwind_ip = orc_ip;
312 mod->arch.orc_unwind = orc;
313 mod->arch.num_orcs = num_entries;
314}
315#endif
316
317void __init unwind_init(void)
318{
319 size_t orc_ip_size = (void *)__stop_orc_unwind_ip - (void *)__start_orc_unwind_ip;
320 size_t orc_size = (void *)__stop_orc_unwind - (void *)__start_orc_unwind;
321 size_t num_entries = orc_ip_size / sizeof(int);
322 struct orc_entry *orc;
323 int i;
324
325 if (!num_entries || orc_ip_size % sizeof(int) != 0 ||
326 orc_size % sizeof(struct orc_entry) != 0 ||
327 num_entries != orc_size / sizeof(struct orc_entry)) {
328 orc_warn("WARNING: Bad or missing .orc_unwind table. Disabling unwinder.\n");
329 return;
330 }
331
332 /*
333 * Note, the orc_unwind and orc_unwind_ip tables were already
334 * sorted at build time via the 'sorttable' tool.
335 * It's ready for binary search straight away, no need to sort it.
336 */
337
338 /* Initialize the fast lookup table: */
339 lookup_num_blocks = orc_lookup_end - orc_lookup;
340 for (i = 0; i < lookup_num_blocks-1; i++) {
341 orc = __orc_find(__start_orc_unwind_ip, __start_orc_unwind,
342 num_entries,
343 LOOKUP_START_IP + (LOOKUP_BLOCK_SIZE * i));
344 if (!orc) {
345 orc_warn("WARNING: Corrupt .orc_unwind table. Disabling unwinder.\n");
346 return;
347 }
348
349 orc_lookup[i] = orc - __start_orc_unwind;
350 }
351
352 /* Initialize the ending block: */
353 orc = __orc_find(__start_orc_unwind_ip, __start_orc_unwind, num_entries,
354 LOOKUP_STOP_IP);
355 if (!orc) {
356 orc_warn("WARNING: Corrupt .orc_unwind table. Disabling unwinder.\n");
357 return;
358 }
359 orc_lookup[lookup_num_blocks-1] = orc - __start_orc_unwind;
360
361 orc_init = true;
362}
363
364unsigned long unwind_get_return_address(struct unwind_state *state)
365{
366 if (unwind_done(state))
367 return 0;
368
369 return __kernel_text_address(state->ip) ? state->ip : 0;
370}
371EXPORT_SYMBOL_GPL(unwind_get_return_address);
372
373unsigned long *unwind_get_return_address_ptr(struct unwind_state *state)
374{
375 if (unwind_done(state))
376 return NULL;
377
378 if (state->regs)
379 return &state->regs->ip;
380
381 if (state->sp)
382 return (unsigned long *)state->sp - 1;
383
384 return NULL;
385}
386
387static bool stack_access_ok(struct unwind_state *state, unsigned long _addr,
388 size_t len)
389{
390 struct stack_info *info = &state->stack_info;
391 void *addr = (void *)_addr;
392
393 if (on_stack(info, addr, len))
394 return true;
395
396 return !get_stack_info(addr, state->task, info, &state->stack_mask) &&
397 on_stack(info, addr, len);
398}
399
400static bool deref_stack_reg(struct unwind_state *state, unsigned long addr,
401 unsigned long *val)
402{
403 if (!stack_access_ok(state, addr, sizeof(long)))
404 return false;
405
406 *val = READ_ONCE_NOCHECK(*(unsigned long *)addr);
407 return true;
408}
409
410static bool deref_stack_regs(struct unwind_state *state, unsigned long addr,
411 unsigned long *ip, unsigned long *sp)
412{
413 struct pt_regs *regs = (struct pt_regs *)addr;
414
415 /* x86-32 support will be more complicated due to the ®s->sp hack */
416 BUILD_BUG_ON(IS_ENABLED(CONFIG_X86_32));
417
418 if (!stack_access_ok(state, addr, sizeof(struct pt_regs)))
419 return false;
420
421 *ip = READ_ONCE_NOCHECK(regs->ip);
422 *sp = READ_ONCE_NOCHECK(regs->sp);
423 return true;
424}
425
426static bool deref_stack_iret_regs(struct unwind_state *state, unsigned long addr,
427 unsigned long *ip, unsigned long *sp)
428{
429 struct pt_regs *regs = (void *)addr - IRET_FRAME_OFFSET;
430
431 if (!stack_access_ok(state, addr, IRET_FRAME_SIZE))
432 return false;
433
434 *ip = READ_ONCE_NOCHECK(regs->ip);
435 *sp = READ_ONCE_NOCHECK(regs->sp);
436 return true;
437}
438
439/*
440 * If state->regs is non-NULL, and points to a full pt_regs, just get the reg
441 * value from state->regs.
442 *
443 * Otherwise, if state->regs just points to IRET regs, and the previous frame
444 * had full regs, it's safe to get the value from the previous regs. This can
445 * happen when early/late IRQ entry code gets interrupted by an NMI.
446 */
447static bool get_reg(struct unwind_state *state, unsigned int reg_off,
448 unsigned long *val)
449{
450 unsigned int reg = reg_off/8;
451
452 if (!state->regs)
453 return false;
454
455 if (state->full_regs) {
456 *val = READ_ONCE_NOCHECK(((unsigned long *)state->regs)[reg]);
457 return true;
458 }
459
460 if (state->prev_regs) {
461 *val = READ_ONCE_NOCHECK(((unsigned long *)state->prev_regs)[reg]);
462 return true;
463 }
464
465 return false;
466}
467
468bool unwind_next_frame(struct unwind_state *state)
469{
470 unsigned long ip_p, sp, tmp, orig_ip = state->ip, prev_sp = state->sp;
471 enum stack_type prev_type = state->stack_info.type;
472 struct orc_entry *orc;
473 bool indirect = false;
474
475 if (unwind_done(state))
476 return false;
477
478 /* Don't let modules unload while we're reading their ORC data. */
479 preempt_disable();
480
481 /* End-of-stack check for user tasks: */
482 if (state->regs && user_mode(state->regs))
483 goto the_end;
484
485 /*
486 * Find the orc_entry associated with the text address.
487 *
488 * For a call frame (as opposed to a signal frame), state->ip points to
489 * the instruction after the call. That instruction's stack layout
490 * could be different from the call instruction's layout, for example
491 * if the call was to a noreturn function. So get the ORC data for the
492 * call instruction itself.
493 */
494 orc = orc_find(state->signal ? state->ip : state->ip - 1);
495 if (!orc) {
496 /*
497 * As a fallback, try to assume this code uses a frame pointer.
498 * This is useful for generated code, like BPF, which ORC
499 * doesn't know about. This is just a guess, so the rest of
500 * the unwind is no longer considered reliable.
501 */
502 orc = &orc_fp_entry;
503 state->error = true;
504 } else {
505 if (orc->type == ORC_TYPE_UNDEFINED)
506 goto err;
507
508 if (orc->type == ORC_TYPE_END_OF_STACK)
509 goto the_end;
510 }
511
512 state->signal = orc->signal;
513
514 /* Find the previous frame's stack: */
515 switch (orc->sp_reg) {
516 case ORC_REG_SP:
517 sp = state->sp + orc->sp_offset;
518 break;
519
520 case ORC_REG_BP:
521 sp = state->bp + orc->sp_offset;
522 break;
523
524 case ORC_REG_SP_INDIRECT:
525 sp = state->sp;
526 indirect = true;
527 break;
528
529 case ORC_REG_BP_INDIRECT:
530 sp = state->bp + orc->sp_offset;
531 indirect = true;
532 break;
533
534 case ORC_REG_R10:
535 if (!get_reg(state, offsetof(struct pt_regs, r10), &sp)) {
536 orc_warn_current("missing R10 value at %pB\n",
537 (void *)state->ip);
538 goto err;
539 }
540 break;
541
542 case ORC_REG_R13:
543 if (!get_reg(state, offsetof(struct pt_regs, r13), &sp)) {
544 orc_warn_current("missing R13 value at %pB\n",
545 (void *)state->ip);
546 goto err;
547 }
548 break;
549
550 case ORC_REG_DI:
551 if (!get_reg(state, offsetof(struct pt_regs, di), &sp)) {
552 orc_warn_current("missing RDI value at %pB\n",
553 (void *)state->ip);
554 goto err;
555 }
556 break;
557
558 case ORC_REG_DX:
559 if (!get_reg(state, offsetof(struct pt_regs, dx), &sp)) {
560 orc_warn_current("missing DX value at %pB\n",
561 (void *)state->ip);
562 goto err;
563 }
564 break;
565
566 default:
567 orc_warn("unknown SP base reg %d at %pB\n",
568 orc->sp_reg, (void *)state->ip);
569 goto err;
570 }
571
572 if (indirect) {
573 if (!deref_stack_reg(state, sp, &sp))
574 goto err;
575
576 if (orc->sp_reg == ORC_REG_SP_INDIRECT)
577 sp += orc->sp_offset;
578 }
579
580 /* Find IP, SP and possibly regs: */
581 switch (orc->type) {
582 case ORC_TYPE_CALL:
583 ip_p = sp - sizeof(long);
584
585 if (!deref_stack_reg(state, ip_p, &state->ip))
586 goto err;
587
588 state->ip = unwind_recover_ret_addr(state, state->ip,
589 (unsigned long *)ip_p);
590 state->sp = sp;
591 state->regs = NULL;
592 state->prev_regs = NULL;
593 break;
594
595 case ORC_TYPE_REGS:
596 if (!deref_stack_regs(state, sp, &state->ip, &state->sp)) {
597 orc_warn_current("can't access registers at %pB\n",
598 (void *)orig_ip);
599 goto err;
600 }
601 /*
602 * There is a small chance to interrupt at the entry of
603 * arch_rethook_trampoline() where the ORC info doesn't exist.
604 * That point is right after the RET to arch_rethook_trampoline()
605 * which was modified return address.
606 * At that point, the @addr_p of the unwind_recover_rethook()
607 * (this has to point the address of the stack entry storing
608 * the modified return address) must be "SP - (a stack entry)"
609 * because SP is incremented by the RET.
610 */
611 state->ip = unwind_recover_rethook(state, state->ip,
612 (unsigned long *)(state->sp - sizeof(long)));
613 state->regs = (struct pt_regs *)sp;
614 state->prev_regs = NULL;
615 state->full_regs = true;
616 break;
617
618 case ORC_TYPE_REGS_PARTIAL:
619 if (!deref_stack_iret_regs(state, sp, &state->ip, &state->sp)) {
620 orc_warn_current("can't access iret registers at %pB\n",
621 (void *)orig_ip);
622 goto err;
623 }
624 /* See ORC_TYPE_REGS case comment. */
625 state->ip = unwind_recover_rethook(state, state->ip,
626 (unsigned long *)(state->sp - sizeof(long)));
627
628 if (state->full_regs)
629 state->prev_regs = state->regs;
630 state->regs = (void *)sp - IRET_FRAME_OFFSET;
631 state->full_regs = false;
632 break;
633
634 default:
635 orc_warn("unknown .orc_unwind entry type %d at %pB\n",
636 orc->type, (void *)orig_ip);
637 goto err;
638 }
639
640 /* Find BP: */
641 switch (orc->bp_reg) {
642 case ORC_REG_UNDEFINED:
643 if (get_reg(state, offsetof(struct pt_regs, bp), &tmp))
644 state->bp = tmp;
645 break;
646
647 case ORC_REG_PREV_SP:
648 if (!deref_stack_reg(state, sp + orc->bp_offset, &state->bp))
649 goto err;
650 break;
651
652 case ORC_REG_BP:
653 if (!deref_stack_reg(state, state->bp + orc->bp_offset, &state->bp))
654 goto err;
655 break;
656
657 default:
658 orc_warn("unknown BP base reg %d for ip %pB\n",
659 orc->bp_reg, (void *)orig_ip);
660 goto err;
661 }
662
663 /* Prevent a recursive loop due to bad ORC data: */
664 if (state->stack_info.type == prev_type &&
665 on_stack(&state->stack_info, (void *)state->sp, sizeof(long)) &&
666 state->sp <= prev_sp) {
667 orc_warn_current("stack going in the wrong direction? at %pB\n",
668 (void *)orig_ip);
669 goto err;
670 }
671
672 preempt_enable();
673 return true;
674
675err:
676 state->error = true;
677
678the_end:
679 preempt_enable();
680 state->stack_info.type = STACK_TYPE_UNKNOWN;
681 return false;
682}
683EXPORT_SYMBOL_GPL(unwind_next_frame);
684
685void __unwind_start(struct unwind_state *state, struct task_struct *task,
686 struct pt_regs *regs, unsigned long *first_frame)
687{
688 memset(state, 0, sizeof(*state));
689 state->task = task;
690
691 if (!orc_init)
692 goto err;
693
694 /*
695 * Refuse to unwind the stack of a task while it's executing on another
696 * CPU. This check is racy, but that's ok: the unwinder has other
697 * checks to prevent it from going off the rails.
698 */
699 if (task_on_another_cpu(task))
700 goto err;
701
702 if (regs) {
703 if (user_mode(regs))
704 goto the_end;
705
706 state->ip = regs->ip;
707 state->sp = regs->sp;
708 state->bp = regs->bp;
709 state->regs = regs;
710 state->full_regs = true;
711 state->signal = true;
712
713 } else if (task == current) {
714 asm volatile("lea (%%rip), %0\n\t"
715 "mov %%rsp, %1\n\t"
716 "mov %%rbp, %2\n\t"
717 : "=r" (state->ip), "=r" (state->sp),
718 "=r" (state->bp));
719
720 } else {
721 struct inactive_task_frame *frame = (void *)task->thread.sp;
722
723 state->sp = task->thread.sp + sizeof(*frame);
724 state->bp = READ_ONCE_NOCHECK(frame->bp);
725 state->ip = READ_ONCE_NOCHECK(frame->ret_addr);
726 state->signal = (void *)state->ip == ret_from_fork;
727 }
728
729 if (get_stack_info((unsigned long *)state->sp, state->task,
730 &state->stack_info, &state->stack_mask)) {
731 /*
732 * We weren't on a valid stack. It's possible that
733 * we overflowed a valid stack into a guard page.
734 * See if the next page up is valid so that we can
735 * generate some kind of backtrace if this happens.
736 */
737 void *next_page = (void *)PAGE_ALIGN((unsigned long)state->sp);
738 state->error = true;
739 if (get_stack_info(next_page, state->task, &state->stack_info,
740 &state->stack_mask))
741 return;
742 }
743
744 /*
745 * The caller can provide the address of the first frame directly
746 * (first_frame) or indirectly (regs->sp) to indicate which stack frame
747 * to start unwinding at. Skip ahead until we reach it.
748 */
749
750 /* When starting from regs, skip the regs frame: */
751 if (regs) {
752 unwind_next_frame(state);
753 return;
754 }
755
756 /* Otherwise, skip ahead to the user-specified starting frame: */
757 while (!unwind_done(state) &&
758 (!on_stack(&state->stack_info, first_frame, sizeof(long)) ||
759 state->sp <= (unsigned long)first_frame))
760 unwind_next_frame(state);
761
762 return;
763
764err:
765 state->error = true;
766the_end:
767 state->stack_info.type = STACK_TYPE_UNKNOWN;
768}
769EXPORT_SYMBOL_GPL(__unwind_start);