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>
10 #define orc_warn(fmt, ...) \
11 printk_deferred_once(KERN_WARNING pr_fmt("WARNING: " fmt), ##__VA_ARGS__)
13 extern int __start_orc_unwind_ip[];
14 extern int __stop_orc_unwind_ip[];
15 extern struct orc_entry __start_orc_unwind[];
16 extern struct orc_entry __stop_orc_unwind[];
18 static DEFINE_MUTEX(sort_mutex);
19 int *cur_orc_ip_table = __start_orc_unwind_ip;
20 struct orc_entry *cur_orc_table = __start_orc_unwind;
22 unsigned int lookup_num_blocks;
25 static inline unsigned long orc_ip(const int *ip)
27 return (unsigned long)ip + *ip;
30 static struct orc_entry *__orc_find(int *ip_table, struct orc_entry *u_table,
31 unsigned int num_entries, unsigned long ip)
33 int *first = ip_table;
34 int *last = ip_table + num_entries - 1;
35 int *mid = first, *found = first;
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.
46 while (first <= last) {
47 mid = first + ((last - first) / 2);
49 if (orc_ip(mid) <= ip) {
56 return u_table + (found - ip_table);
60 static struct orc_entry *orc_module_find(unsigned long ip)
64 mod = __module_address(ip);
65 if (!mod || !mod->arch.orc_unwind || !mod->arch.orc_unwind_ip)
67 return __orc_find(mod->arch.orc_unwind_ip, mod->arch.orc_unwind,
68 mod->arch.num_orcs, ip);
71 static struct orc_entry *orc_module_find(unsigned long ip)
77 #ifdef CONFIG_DYNAMIC_FTRACE
78 static struct orc_entry *orc_find(unsigned long ip);
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.
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.
90 static struct orc_entry *orc_ftrace_find(unsigned long ip)
92 struct ftrace_ops *ops;
93 unsigned long tramp_addr, offset;
95 ops = ftrace_ops_trampoline(ip);
99 /* Set tramp_addr to the start of the code copied by the trampoline */
100 if (ops->flags & FTRACE_OPS_FL_SAVE_REGS)
101 tramp_addr = (unsigned long)ftrace_regs_caller;
103 tramp_addr = (unsigned long)ftrace_caller;
105 /* Now place tramp_addr to the location within the trampoline ip is at */
106 offset = ip - ops->trampoline;
107 tramp_addr += offset;
109 /* Prevent unlikely recursion */
110 if (ip == tramp_addr)
113 return orc_find(tramp_addr);
116 static struct orc_entry *orc_ftrace_find(unsigned long ip)
123 * If we crash with IP==0, the last successfully executed instruction
124 * was probably an indirect function call with a NULL function pointer,
125 * and we don't have unwind information for NULL.
126 * This hardcoded ORC entry for IP==0 allows us to unwind from a NULL function
127 * pointer into its parent and then continue normally from there.
129 static struct orc_entry null_orc_entry = {
130 .sp_offset = sizeof(long),
131 .sp_reg = ORC_REG_SP,
132 .bp_reg = ORC_REG_UNDEFINED,
133 .type = ORC_TYPE_CALL
136 /* Fake frame pointer entry -- used as a fallback for generated code */
137 static struct orc_entry orc_fp_entry = {
138 .type = ORC_TYPE_CALL,
139 .sp_reg = ORC_REG_BP,
141 .bp_reg = ORC_REG_PREV_SP,
146 static struct orc_entry *orc_find(unsigned long ip)
148 static struct orc_entry *orc;
151 return &null_orc_entry;
153 /* For non-init vmlinux addresses, use the fast lookup table: */
154 if (ip >= LOOKUP_START_IP && ip < LOOKUP_STOP_IP) {
155 unsigned int idx, start, stop;
157 idx = (ip - LOOKUP_START_IP) / LOOKUP_BLOCK_SIZE;
159 if (unlikely((idx >= lookup_num_blocks-1))) {
160 orc_warn("WARNING: bad lookup idx: idx=%u num=%u ip=%pB\n",
161 idx, lookup_num_blocks, (void *)ip);
165 start = orc_lookup[idx];
166 stop = orc_lookup[idx + 1] + 1;
168 if (unlikely((__start_orc_unwind + start >= __stop_orc_unwind) ||
169 (__start_orc_unwind + stop > __stop_orc_unwind))) {
170 orc_warn("WARNING: bad lookup value: idx=%u num=%u start=%u stop=%u ip=%pB\n",
171 idx, lookup_num_blocks, start, stop, (void *)ip);
175 return __orc_find(__start_orc_unwind_ip + start,
176 __start_orc_unwind + start, stop - start, ip);
179 /* vmlinux .init slow lookup: */
180 if (init_kernel_text(ip))
181 return __orc_find(__start_orc_unwind_ip, __start_orc_unwind,
182 __stop_orc_unwind_ip - __start_orc_unwind_ip, ip);
185 orc = orc_module_find(ip);
189 return orc_ftrace_find(ip);
192 static void orc_sort_swap(void *_a, void *_b, int size)
194 struct orc_entry *orc_a, *orc_b;
195 struct orc_entry orc_tmp;
196 int *a = _a, *b = _b, tmp;
199 /* Swap the .orc_unwind_ip entries: */
204 /* Swap the corresponding .orc_unwind entries: */
205 orc_a = cur_orc_table + (a - cur_orc_ip_table);
206 orc_b = cur_orc_table + (b - cur_orc_ip_table);
212 static int orc_sort_cmp(const void *_a, const void *_b)
214 struct orc_entry *orc_a;
215 const int *a = _a, *b = _b;
216 unsigned long a_val = orc_ip(a);
217 unsigned long b_val = orc_ip(b);
225 * The "weak" section terminator entries need to always be on the left
226 * to ensure the lookup code skips them in favor of real entries.
227 * These terminator entries exist to handle any gaps created by
228 * whitelisted .o files which didn't get objtool generation.
230 orc_a = cur_orc_table + (a - cur_orc_ip_table);
231 return orc_a->sp_reg == ORC_REG_UNDEFINED && !orc_a->end ? -1 : 1;
234 #ifdef CONFIG_MODULES
235 void unwind_module_init(struct module *mod, void *_orc_ip, size_t orc_ip_size,
236 void *_orc, size_t orc_size)
238 int *orc_ip = _orc_ip;
239 struct orc_entry *orc = _orc;
240 unsigned int num_entries = orc_ip_size / sizeof(int);
242 WARN_ON_ONCE(orc_ip_size % sizeof(int) != 0 ||
243 orc_size % sizeof(*orc) != 0 ||
244 num_entries != orc_size / sizeof(*orc));
247 * The 'cur_orc_*' globals allow the orc_sort_swap() callback to
248 * associate an .orc_unwind_ip table entry with its corresponding
249 * .orc_unwind entry so they can both be swapped.
251 mutex_lock(&sort_mutex);
252 cur_orc_ip_table = orc_ip;
254 sort(orc_ip, num_entries, sizeof(int), orc_sort_cmp, orc_sort_swap);
255 mutex_unlock(&sort_mutex);
257 mod->arch.orc_unwind_ip = orc_ip;
258 mod->arch.orc_unwind = orc;
259 mod->arch.num_orcs = num_entries;
263 void __init unwind_init(void)
265 size_t orc_ip_size = (void *)__stop_orc_unwind_ip - (void *)__start_orc_unwind_ip;
266 size_t orc_size = (void *)__stop_orc_unwind - (void *)__start_orc_unwind;
267 size_t num_entries = orc_ip_size / sizeof(int);
268 struct orc_entry *orc;
271 if (!num_entries || orc_ip_size % sizeof(int) != 0 ||
272 orc_size % sizeof(struct orc_entry) != 0 ||
273 num_entries != orc_size / sizeof(struct orc_entry)) {
274 orc_warn("WARNING: Bad or missing .orc_unwind table. Disabling unwinder.\n");
278 /* Sort the .orc_unwind and .orc_unwind_ip tables: */
279 sort(__start_orc_unwind_ip, num_entries, sizeof(int), orc_sort_cmp,
282 /* Initialize the fast lookup table: */
283 lookup_num_blocks = orc_lookup_end - orc_lookup;
284 for (i = 0; i < lookup_num_blocks-1; i++) {
285 orc = __orc_find(__start_orc_unwind_ip, __start_orc_unwind,
287 LOOKUP_START_IP + (LOOKUP_BLOCK_SIZE * i));
289 orc_warn("WARNING: Corrupt .orc_unwind table. Disabling unwinder.\n");
293 orc_lookup[i] = orc - __start_orc_unwind;
296 /* Initialize the ending block: */
297 orc = __orc_find(__start_orc_unwind_ip, __start_orc_unwind, num_entries,
300 orc_warn("WARNING: Corrupt .orc_unwind table. Disabling unwinder.\n");
303 orc_lookup[lookup_num_blocks-1] = orc - __start_orc_unwind;
308 unsigned long unwind_get_return_address(struct unwind_state *state)
310 if (unwind_done(state))
313 return __kernel_text_address(state->ip) ? state->ip : 0;
315 EXPORT_SYMBOL_GPL(unwind_get_return_address);
317 unsigned long *unwind_get_return_address_ptr(struct unwind_state *state)
319 if (unwind_done(state))
323 return &state->regs->ip;
326 return (unsigned long *)state->sp - 1;
331 static bool stack_access_ok(struct unwind_state *state, unsigned long _addr,
334 struct stack_info *info = &state->stack_info;
335 void *addr = (void *)_addr;
337 if (!on_stack(info, addr, len) &&
338 (get_stack_info(addr, state->task, info, &state->stack_mask)))
344 static bool deref_stack_reg(struct unwind_state *state, unsigned long addr,
347 if (!stack_access_ok(state, addr, sizeof(long)))
350 *val = READ_ONCE_NOCHECK(*(unsigned long *)addr);
354 static bool deref_stack_regs(struct unwind_state *state, unsigned long addr,
355 unsigned long *ip, unsigned long *sp)
357 struct pt_regs *regs = (struct pt_regs *)addr;
359 /* x86-32 support will be more complicated due to the ®s->sp hack */
360 BUILD_BUG_ON(IS_ENABLED(CONFIG_X86_32));
362 if (!stack_access_ok(state, addr, sizeof(struct pt_regs)))
365 *ip = READ_ONCE_NOCHECK(regs->ip);
366 *sp = READ_ONCE_NOCHECK(regs->sp);
370 static bool deref_stack_iret_regs(struct unwind_state *state, unsigned long addr,
371 unsigned long *ip, unsigned long *sp)
373 struct pt_regs *regs = (void *)addr - IRET_FRAME_OFFSET;
375 if (!stack_access_ok(state, addr, IRET_FRAME_SIZE))
378 *ip = READ_ONCE_NOCHECK(regs->ip);
379 *sp = READ_ONCE_NOCHECK(regs->sp);
384 * If state->regs is non-NULL, and points to a full pt_regs, just get the reg
385 * value from state->regs.
387 * Otherwise, if state->regs just points to IRET regs, and the previous frame
388 * had full regs, it's safe to get the value from the previous regs. This can
389 * happen when early/late IRQ entry code gets interrupted by an NMI.
391 static bool get_reg(struct unwind_state *state, unsigned int reg_off,
394 unsigned int reg = reg_off/8;
399 if (state->full_regs) {
400 *val = READ_ONCE_NOCHECK(((unsigned long *)state->regs)[reg]);
404 if (state->prev_regs) {
405 *val = READ_ONCE_NOCHECK(((unsigned long *)state->prev_regs)[reg]);
412 bool unwind_next_frame(struct unwind_state *state)
414 unsigned long ip_p, sp, tmp, orig_ip = state->ip, prev_sp = state->sp;
415 enum stack_type prev_type = state->stack_info.type;
416 struct orc_entry *orc;
417 bool indirect = false;
419 if (unwind_done(state))
422 /* Don't let modules unload while we're reading their ORC data. */
425 /* End-of-stack check for user tasks: */
426 if (state->regs && user_mode(state->regs))
430 * Find the orc_entry associated with the text address.
432 * For a call frame (as opposed to a signal frame), state->ip points to
433 * the instruction after the call. That instruction's stack layout
434 * could be different from the call instruction's layout, for example
435 * if the call was to a noreturn function. So get the ORC data for the
436 * call instruction itself.
438 orc = orc_find(state->signal ? state->ip : state->ip - 1);
441 * As a fallback, try to assume this code uses a frame pointer.
442 * This is useful for generated code, like BPF, which ORC
443 * doesn't know about. This is just a guess, so the rest of
444 * the unwind is no longer considered reliable.
450 /* End-of-stack check for kernel threads: */
451 if (orc->sp_reg == ORC_REG_UNDEFINED) {
458 /* Find the previous frame's stack: */
459 switch (orc->sp_reg) {
461 sp = state->sp + orc->sp_offset;
465 sp = state->bp + orc->sp_offset;
468 case ORC_REG_SP_INDIRECT:
469 sp = state->sp + orc->sp_offset;
473 case ORC_REG_BP_INDIRECT:
474 sp = state->bp + orc->sp_offset;
479 if (!get_reg(state, offsetof(struct pt_regs, r10), &sp)) {
480 orc_warn("missing regs for base reg R10 at ip %pB\n",
487 if (!get_reg(state, offsetof(struct pt_regs, r13), &sp)) {
488 orc_warn("missing regs for base reg R13 at ip %pB\n",
495 if (!get_reg(state, offsetof(struct pt_regs, di), &sp)) {
496 orc_warn("missing regs for base reg DI at ip %pB\n",
503 if (!get_reg(state, offsetof(struct pt_regs, dx), &sp)) {
504 orc_warn("missing regs for base reg DX at ip %pB\n",
511 orc_warn("unknown SP base reg %d for ip %pB\n",
512 orc->sp_reg, (void *)state->ip);
517 if (!deref_stack_reg(state, sp, &sp))
521 /* Find IP, SP and possibly regs: */
524 ip_p = sp - sizeof(long);
526 if (!deref_stack_reg(state, ip_p, &state->ip))
529 state->ip = ftrace_graph_ret_addr(state->task, &state->graph_idx,
530 state->ip, (void *)ip_p);
534 state->prev_regs = NULL;
535 state->signal = false;
539 if (!deref_stack_regs(state, sp, &state->ip, &state->sp)) {
540 orc_warn("can't dereference registers at %p for ip %pB\n",
541 (void *)sp, (void *)orig_ip);
545 state->regs = (struct pt_regs *)sp;
546 state->prev_regs = NULL;
547 state->full_regs = true;
548 state->signal = true;
551 case ORC_TYPE_REGS_IRET:
552 if (!deref_stack_iret_regs(state, sp, &state->ip, &state->sp)) {
553 orc_warn("can't dereference iret registers at %p for ip %pB\n",
554 (void *)sp, (void *)orig_ip);
558 if (state->full_regs)
559 state->prev_regs = state->regs;
560 state->regs = (void *)sp - IRET_FRAME_OFFSET;
561 state->full_regs = false;
562 state->signal = true;
566 orc_warn("unknown .orc_unwind entry type %d for ip %pB\n",
567 orc->type, (void *)orig_ip);
572 switch (orc->bp_reg) {
573 case ORC_REG_UNDEFINED:
574 if (get_reg(state, offsetof(struct pt_regs, bp), &tmp))
578 case ORC_REG_PREV_SP:
579 if (!deref_stack_reg(state, sp + orc->bp_offset, &state->bp))
584 if (!deref_stack_reg(state, state->bp + orc->bp_offset, &state->bp))
589 orc_warn("unknown BP base reg %d for ip %pB\n",
590 orc->bp_reg, (void *)orig_ip);
594 /* Prevent a recursive loop due to bad ORC data: */
595 if (state->stack_info.type == prev_type &&
596 on_stack(&state->stack_info, (void *)state->sp, sizeof(long)) &&
597 state->sp <= prev_sp) {
598 orc_warn("stack going in the wrong direction? ip=%pB\n",
611 state->stack_info.type = STACK_TYPE_UNKNOWN;
614 EXPORT_SYMBOL_GPL(unwind_next_frame);
616 void __unwind_start(struct unwind_state *state, struct task_struct *task,
617 struct pt_regs *regs, unsigned long *first_frame)
619 memset(state, 0, sizeof(*state));
626 * Refuse to unwind the stack of a task while it's executing on another
627 * CPU. This check is racy, but that's ok: the unwinder has other
628 * checks to prevent it from going off the rails.
630 if (task_on_another_cpu(task))
637 state->ip = regs->ip;
638 state->sp = regs->sp;
639 state->bp = regs->bp;
641 state->full_regs = true;
642 state->signal = true;
644 } else if (task == current) {
645 asm volatile("lea (%%rip), %0\n\t"
648 : "=r" (state->ip), "=r" (state->sp),
652 struct inactive_task_frame *frame = (void *)task->thread.sp;
654 state->sp = task->thread.sp + sizeof(*frame);
655 state->bp = READ_ONCE_NOCHECK(frame->bp);
656 state->ip = READ_ONCE_NOCHECK(frame->ret_addr);
657 state->signal = (void *)state->ip == ret_from_fork;
660 if (get_stack_info((unsigned long *)state->sp, state->task,
661 &state->stack_info, &state->stack_mask)) {
663 * We weren't on a valid stack. It's possible that
664 * we overflowed a valid stack into a guard page.
665 * See if the next page up is valid so that we can
666 * generate some kind of backtrace if this happens.
668 void *next_page = (void *)PAGE_ALIGN((unsigned long)state->sp);
670 if (get_stack_info(next_page, state->task, &state->stack_info,
676 * The caller can provide the address of the first frame directly
677 * (first_frame) or indirectly (regs->sp) to indicate which stack frame
678 * to start unwinding at. Skip ahead until we reach it.
681 /* When starting from regs, skip the regs frame: */
683 unwind_next_frame(state);
687 /* Otherwise, skip ahead to the user-specified starting frame: */
688 while (!unwind_done(state) &&
689 (!on_stack(&state->stack_info, first_frame, sizeof(long)) ||
690 state->sp <= (unsigned long)first_frame))
691 unwind_next_frame(state);
698 state->stack_info.type = STACK_TYPE_UNKNOWN;
700 EXPORT_SYMBOL_GPL(__unwind_start);