unwind_orc.c

  1// SPDX-License-Identifier: GPL-2.0-only
  2#include <linux/export.h>
  3#include <linux/module.h>
  4#include <linux/objtool.h>
  5#include <linux/sort.h>
  6#include <asm/exception.h>
  7#include <asm/orc_header.h>
  8#include <asm/orc_lookup.h>
  9#include <asm/orc_types.h>
 10#include <asm/ptrace.h>
 11#include <asm/setup.h>
 12#include <asm/stacktrace.h>
 13#include <asm/tlb.h>
 14#include <asm/unwind.h>
 15
 16ORC_HEADER;
 17
 18#define orc_warn(fmt, ...) \
 19	printk_deferred_once(KERN_WARNING "WARNING: " fmt, ##__VA_ARGS__)
 20
 21extern int __start_orc_unwind_ip[];
 22extern int __stop_orc_unwind_ip[];
 23extern struct orc_entry __start_orc_unwind[];
 24extern struct orc_entry __stop_orc_unwind[];
 25
 26static bool orc_init __ro_after_init;
 27static unsigned int lookup_num_blocks __ro_after_init;
 28
 29/* Fake frame pointer entry -- used as a fallback for generated code */
 30static struct orc_entry orc_fp_entry = {
 31	.sp_reg		= ORC_REG_FP,
 32	.sp_offset	= 16,
 33	.fp_reg		= ORC_REG_PREV_SP,
 34	.fp_offset	= -16,
 35	.ra_reg		= ORC_REG_PREV_SP,
 36	.ra_offset	= -8,
 37	.type		= ORC_TYPE_CALL
 38};
 39
 40/*
 41 * If we crash with IP==0, the last successfully executed instruction
 42 * was probably an indirect function call with a NULL function pointer,
 43 * and we don't have unwind information for NULL.
 44 * This hardcoded ORC entry for IP==0 allows us to unwind from a NULL function
 45 * pointer into its parent and then continue normally from there.
 46 */
 47static struct orc_entry orc_null_entry = {
 48	.sp_reg		= ORC_REG_SP,
 49	.sp_offset	= sizeof(long),
 50	.fp_reg		= ORC_REG_UNDEFINED,
 51	.type		= ORC_TYPE_CALL
 52};
 53
 54static inline unsigned long orc_ip(const int *ip)
 55{
 56	return (unsigned long)ip + *ip;
 57}
 58
 59static struct orc_entry *__orc_find(int *ip_table, struct orc_entry *u_table,
 60				    unsigned int num_entries, unsigned long ip)
 61{
 62	int *first = ip_table;
 63	int *mid = first, *found = first;
 64	int *last = ip_table + num_entries - 1;
 65
 66	if (!num_entries)
 67		return NULL;
 68
 69	/*
 70	 * Do a binary range search to find the rightmost duplicate of a given
 71	 * starting address.  Some entries are section terminators which are
 72	 * "weak" entries for ensuring there are no gaps.  They should be
 73	 * ignored when they conflict with a real entry.
 74	 */
 75	while (first <= last) {
 76		mid = first + ((last - first) / 2);
 77
 78		if (orc_ip(mid) <= ip) {
 79			found = mid;
 80			first = mid + 1;
 81		} else
 82			last = mid - 1;
 83	}
 84
 85	return u_table + (found - ip_table);
 86}
 87
 88#ifdef CONFIG_MODULES
 89static struct orc_entry *orc_module_find(unsigned long ip)
 90{
 91	struct module *mod;
 92
 93	mod = __module_address(ip);
 94	if (!mod || !mod->arch.orc_unwind || !mod->arch.orc_unwind_ip)
 95		return NULL;
 96
 97	return __orc_find(mod->arch.orc_unwind_ip, mod->arch.orc_unwind, mod->arch.num_orcs, ip);
 98}
 99#else
100static struct orc_entry *orc_module_find(unsigned long ip)
101{
102	return NULL;
103}
104#endif
105
106#ifdef CONFIG_DYNAMIC_FTRACE
107static struct orc_entry *orc_find(unsigned long ip);
108
109/*
110 * Ftrace dynamic trampolines do not have orc entries of their own.
111 * But they are copies of the ftrace entries that are static and
112 * defined in ftrace_*.S, which do have orc entries.
113 *
114 * If the unwinder comes across a ftrace trampoline, then find the
115 * ftrace function that was used to create it, and use that ftrace
116 * function's orc entry, as the placement of the return code in
117 * the stack will be identical.
118 */
119static struct orc_entry *orc_ftrace_find(unsigned long ip)
120{
121	struct ftrace_ops *ops;
122	unsigned long tramp_addr, offset;
123
124	ops = ftrace_ops_trampoline(ip);
125	if (!ops)
126		return NULL;
127
128	/* Set tramp_addr to the start of the code copied by the trampoline */
129	if (ops->flags & FTRACE_OPS_FL_SAVE_REGS)
130		tramp_addr = (unsigned long)ftrace_regs_caller;
131	else
132		tramp_addr = (unsigned long)ftrace_caller;
133
134	/* Now place tramp_addr to the location within the trampoline ip is at */
135	offset = ip - ops->trampoline;
136	tramp_addr += offset;
137
138	/* Prevent unlikely recursion */
139	if (ip == tramp_addr)
140		return NULL;
141
142	return orc_find(tramp_addr);
143}
144#else
145static struct orc_entry *orc_ftrace_find(unsigned long ip)
146{
147	return NULL;
148}
149#endif
150
151static struct orc_entry *orc_find(unsigned long ip)
152{
153	static struct orc_entry *orc;
154
155	if (ip == 0)
156		return &orc_null_entry;
157
158	/* For non-init vmlinux addresses, use the fast lookup table: */
159	if (ip >= LOOKUP_START_IP && ip < LOOKUP_STOP_IP) {
160		unsigned int idx, start, stop;
161
162		idx = (ip - LOOKUP_START_IP) / LOOKUP_BLOCK_SIZE;
163
164		if (unlikely((idx >= lookup_num_blocks-1))) {
165			orc_warn("WARNING: bad lookup idx: idx=%u num=%u ip=%pB\n",
166				 idx, lookup_num_blocks, (void *)ip);
167			return NULL;
168		}
169
170		start = orc_lookup[idx];
171		stop = orc_lookup[idx + 1] + 1;
172
173		if (unlikely((__start_orc_unwind + start >= __stop_orc_unwind) ||
174			     (__start_orc_unwind + stop > __stop_orc_unwind))) {
175			orc_warn("WARNING: bad lookup value: idx=%u num=%u start=%u stop=%u ip=%pB\n",
176				 idx, lookup_num_blocks, start, stop, (void *)ip);
177			return NULL;
178		}
179
180		return __orc_find(__start_orc_unwind_ip + start,
181				  __start_orc_unwind + start, stop - start, ip);
182	}
183
184	/* vmlinux .init slow lookup: */
185	if (is_kernel_inittext(ip))
186		return __orc_find(__start_orc_unwind_ip, __start_orc_unwind,
187				  __stop_orc_unwind_ip - __start_orc_unwind_ip, ip);
188
189	/* Module lookup: */
190	orc = orc_module_find(ip);
191	if (orc)
192		return orc;
193
194	return orc_ftrace_find(ip);
195}
196
197#ifdef CONFIG_MODULES
198
199static DEFINE_MUTEX(sort_mutex);
200static int *cur_orc_ip_table = __start_orc_unwind_ip;
201static struct orc_entry *cur_orc_table = __start_orc_unwind;
202
203static void orc_sort_swap(void *_a, void *_b, int size)
204{
205	int delta = _b - _a;
206	int *a = _a, *b = _b, tmp;
207	struct orc_entry *orc_a, *orc_b;
208
209	/* Swap the .orc_unwind_ip entries: */
210	tmp = *a;
211	*a = *b + delta;
212	*b = tmp - delta;
213
214	/* Swap the corresponding .orc_unwind entries: */
215	orc_a = cur_orc_table + (a - cur_orc_ip_table);
216	orc_b = cur_orc_table + (b - cur_orc_ip_table);
217	swap(*orc_a, *orc_b);
218}
219
220static int orc_sort_cmp(const void *_a, const void *_b)
221{
222	const int *a = _a, *b = _b;
223	unsigned long a_val = orc_ip(a);
224	unsigned long b_val = orc_ip(b);
225	struct orc_entry *orc_a;
226
227	if (a_val > b_val)
228		return 1;
229	if (a_val < b_val)
230		return -1;
231
232	/*
233	 * The "weak" section terminator entries need to always be first
234	 * to ensure the lookup code skips them in favor of real entries.
235	 * These terminator entries exist to handle any gaps created by
236	 * whitelisted .o files which didn't get objtool generation.
237	 */
238	orc_a = cur_orc_table + (a - cur_orc_ip_table);
239
240	return orc_a->type == ORC_TYPE_UNDEFINED ? -1 : 1;
241}
242
243void unwind_module_init(struct module *mod, void *_orc_ip, size_t orc_ip_size,
244			void *_orc, size_t orc_size)
245{
246	int *orc_ip = _orc_ip;
247	struct orc_entry *orc = _orc;
248	unsigned int num_entries = orc_ip_size / sizeof(int);
249
250	WARN_ON_ONCE(orc_ip_size % sizeof(int) != 0 ||
251		     orc_size % sizeof(*orc) != 0 ||
252		     num_entries != orc_size / sizeof(*orc));
253
254	/*
255	 * The 'cur_orc_*' globals allow the orc_sort_swap() callback to
256	 * associate an .orc_unwind_ip table entry with its corresponding
257	 * .orc_unwind entry so they can both be swapped.
258	 */
259	mutex_lock(&sort_mutex);
260	cur_orc_ip_table = orc_ip;
261	cur_orc_table = orc;
262	sort(orc_ip, num_entries, sizeof(int), orc_sort_cmp, orc_sort_swap);
263	mutex_unlock(&sort_mutex);
264
265	mod->arch.orc_unwind_ip = orc_ip;
266	mod->arch.orc_unwind = orc;
267	mod->arch.num_orcs = num_entries;
268}
269#endif
270
271void __init unwind_init(void)
272{
273	int i;
274	size_t orc_size = (void *)__stop_orc_unwind - (void *)__start_orc_unwind;
275	size_t orc_ip_size = (void *)__stop_orc_unwind_ip - (void *)__start_orc_unwind_ip;
276	size_t num_entries = orc_ip_size / sizeof(int);
277	struct orc_entry *orc;
278
279	if (!num_entries || orc_ip_size % sizeof(int) != 0 ||
280	    orc_size % sizeof(struct orc_entry) != 0 ||
281	    num_entries != orc_size / sizeof(struct orc_entry)) {
282		orc_warn("WARNING: Bad or missing .orc_unwind table.  Disabling unwinder.\n");
283		return;
284	}
285
286	/*
287	 * Note, the orc_unwind and orc_unwind_ip tables were already
288	 * sorted at build time via the 'sorttable' tool.
289	 * It's ready for binary search straight away, no need to sort it.
290	 */
291
292	/* Initialize the fast lookup table: */
293	lookup_num_blocks = orc_lookup_end - orc_lookup;
294	for (i = 0; i < lookup_num_blocks-1; i++) {
295		orc = __orc_find(__start_orc_unwind_ip, __start_orc_unwind,
296				 num_entries, LOOKUP_START_IP + (LOOKUP_BLOCK_SIZE * i));
297		if (!orc) {
298			orc_warn("WARNING: Corrupt .orc_unwind table.  Disabling unwinder.\n");
299			return;
300		}
301
302		orc_lookup[i] = orc - __start_orc_unwind;
303	}
304
305	/* Initialize the ending block: */
306	orc = __orc_find(__start_orc_unwind_ip, __start_orc_unwind, num_entries, LOOKUP_STOP_IP);
307	if (!orc) {
308		orc_warn("WARNING: Corrupt .orc_unwind table.  Disabling unwinder.\n");
309		return;
310	}
311	orc_lookup[lookup_num_blocks-1] = orc - __start_orc_unwind;
312
313	orc_init = true;
314}
315
316static inline bool on_stack(struct stack_info *info, unsigned long addr, size_t len)
317{
318	unsigned long begin = info->begin;
319	unsigned long end   = info->end;
320
321	return (info->type != STACK_TYPE_UNKNOWN &&
322		addr >= begin && addr < end && addr + len > begin && addr + len <= end);
323}
324
325static bool stack_access_ok(struct unwind_state *state, unsigned long addr, size_t len)
326{
327	struct stack_info *info = &state->stack_info;
328
329	if (on_stack(info, addr, len))
330		return true;
331
332	return !get_stack_info(addr, state->task, info) && on_stack(info, addr, len);
333}
334
335unsigned long unwind_get_return_address(struct unwind_state *state)
336{
337	return __unwind_get_return_address(state);
338}
339EXPORT_SYMBOL_GPL(unwind_get_return_address);
340
341void unwind_start(struct unwind_state *state, struct task_struct *task,
342		    struct pt_regs *regs)
343{
344	__unwind_start(state, task, regs);
345	state->type = UNWINDER_ORC;
346	if (!unwind_done(state) && !__kernel_text_address(state->pc))
347		unwind_next_frame(state);
348}
349EXPORT_SYMBOL_GPL(unwind_start);
350
351static inline unsigned long bt_address(unsigned long ra)
352{
353#if defined(CONFIG_NUMA) && !defined(CONFIG_PREEMPT_RT)
354	int cpu;
355	int vec_sz = sizeof(exception_handlers);
356
357	for_each_possible_cpu(cpu) {
358		if (!pcpu_handlers[cpu])
359			continue;
360
361		if (ra >= pcpu_handlers[cpu] &&
362		    ra < pcpu_handlers[cpu] + vec_sz) {
363			ra = ra + eentry - pcpu_handlers[cpu];
364			break;
365		}
366	}
367#endif
368
369	if (ra >= eentry && ra < eentry +  EXCCODE_INT_END * VECSIZE) {
370		unsigned long func;
371		unsigned long type = (ra - eentry) / VECSIZE;
372		unsigned long offset = (ra - eentry) % VECSIZE;
373
374		switch (type) {
375		case 0 ... EXCCODE_INT_START - 1:
376			func = (unsigned long)exception_table[type];
377			break;
378		case EXCCODE_INT_START ... EXCCODE_INT_END:
379			func = (unsigned long)handle_vint;
380			break;
381		default:
382			func = (unsigned long)handle_reserved;
383			break;
384		}
385
386		ra = func + offset;
387	}
388
389	if (__kernel_text_address(ra))
390		return ra;
391
392	return 0;
393}
394
395bool unwind_next_frame(struct unwind_state *state)
396{
397	unsigned long *p, pc;
398	struct pt_regs *regs;
399	struct orc_entry *orc;
400	struct stack_info *info = &state->stack_info;
401
402	if (unwind_done(state))
403		return false;
404
405	/* Don't let modules unload while we're reading their ORC data. */
406	guard(rcu)();
407
408	orc = orc_find(state->pc);
409	if (!orc) {
410		/*
411		 * As a fallback, try to assume this code uses a frame pointer.
412		 * This is useful for generated code, like BPF, which ORC
413		 * doesn't know about.  This is just a guess, so the rest of
414		 * the unwind is no longer considered reliable.
415		 */
416		orc = &orc_fp_entry;
417		state->error = true;
418	} else {
419		if (orc->type == ORC_TYPE_UNDEFINED)
420			goto err;
421
422		if (orc->type == ORC_TYPE_END_OF_STACK)
423			goto end;
424	}
425
426	switch (orc->sp_reg) {
427	case ORC_REG_SP:
428		if (info->type == STACK_TYPE_IRQ && state->sp == info->end)
429			orc->type = ORC_TYPE_REGS;
430		else
431			state->sp = state->sp + orc->sp_offset;
432		break;
433	case ORC_REG_FP:
434		state->sp = state->fp;
435		break;
436	default:
437		orc_warn("unknown SP base reg %d at %pB\n", orc->sp_reg, (void *)state->pc);
438		goto err;
439	}
440
441	switch (orc->fp_reg) {
442	case ORC_REG_PREV_SP:
443		p = (unsigned long *)(state->sp + orc->fp_offset);
444		if (!stack_access_ok(state, (unsigned long)p, sizeof(unsigned long)))
445			goto err;
446
447		state->fp = *p;
448		break;
449	case ORC_REG_UNDEFINED:
450		/* Nothing. */
451		break;
452	default:
453		orc_warn("unknown FP base reg %d at %pB\n", orc->fp_reg, (void *)state->pc);
454		goto err;
455	}
456
457	switch (orc->type) {
458	case ORC_TYPE_CALL:
459		if (orc->ra_reg == ORC_REG_PREV_SP) {
460			p = (unsigned long *)(state->sp + orc->ra_offset);
461			if (!stack_access_ok(state, (unsigned long)p, sizeof(unsigned long)))
462				goto err;
463
464			pc = unwind_graph_addr(state, *p, state->sp);
465			pc -= LOONGARCH_INSN_SIZE;
466		} else if (orc->ra_reg == ORC_REG_UNDEFINED) {
467			if (!state->ra || state->ra == state->pc)
468				goto err;
469
470			pc = unwind_graph_addr(state, state->ra, state->sp);
471			pc -=  LOONGARCH_INSN_SIZE;
472			state->ra = 0;
473		} else {
474			orc_warn("unknown ra base reg %d at %pB\n", orc->ra_reg, (void *)state->pc);
475			goto err;
476		}
477		break;
478	case ORC_TYPE_REGS:
479		if (info->type == STACK_TYPE_IRQ && state->sp == info->end)
480			regs = (struct pt_regs *)info->next_sp;
481		else
482			regs = (struct pt_regs *)state->sp;
483
484		if (!stack_access_ok(state, (unsigned long)regs, sizeof(*regs)))
485			goto err;
486
487		if ((info->end == (unsigned long)regs + sizeof(*regs)) &&
488		    !regs->regs[3] && !regs->regs[1])
489			goto end;
490
491		if (user_mode(regs))
492			goto end;
493
494		pc = regs->csr_era;
495		if (!__kernel_text_address(pc))
496			goto err;
497
498		state->sp = regs->regs[3];
499		state->ra = regs->regs[1];
500		state->fp = regs->regs[22];
501		get_stack_info(state->sp, state->task, info);
502
503		break;
504	default:
505		orc_warn("unknown .orc_unwind entry type %d at %pB\n", orc->type, (void *)state->pc);
506		goto err;
507	}
508
509	state->pc = bt_address(pc);
510	if (!state->pc) {
511		pr_err("cannot find unwind pc at %px\n", (void *)pc);
512		goto err;
513	}
514
515	return true;
516
517err:
518	state->error = true;
519
520end:
521	state->stack_info.type = STACK_TYPE_UNKNOWN;
522	return false;
523}
524EXPORT_SYMBOL_GPL(unwind_next_frame);