1 // SPDX-License-Identifier: GPL-2.0-only
2 #include <linux/sched.h>
3 #include <linux/sched/task.h>
4 #include <linux/sched/task_stack.h>
5 #include <linux/interrupt.h>
6 #include <asm/sections.h>
7 #include <asm/ptrace.h>
8 #include <asm/bitops.h>
9 #include <asm/stacktrace.h>
10 #include <asm/unwind.h>
12 #define FRAME_HEADER_SIZE (sizeof(long) * 2)
14 unsigned long unwind_get_return_address(struct unwind_state
*state
)
16 if (unwind_done(state
))
19 return __kernel_text_address(state
->ip
) ? state
->ip
: 0;
21 EXPORT_SYMBOL_GPL(unwind_get_return_address
);
23 unsigned long *unwind_get_return_address_ptr(struct unwind_state
*state
)
25 if (unwind_done(state
))
28 return state
->regs
? &state
->regs
->ip
: state
->bp
+ 1;
31 static void unwind_dump(struct unwind_state
*state
)
33 static bool dumped_before
= false;
34 bool prev_zero
, zero
= false;
35 unsigned long word
, *sp
;
36 struct stack_info stack_info
= {0};
37 unsigned long visit_mask
= 0;
44 printk_deferred("unwind stack type:%d next_sp:%p mask:0x%lx graph_idx:%d\n",
45 state
->stack_info
.type
, state
->stack_info
.next_sp
,
46 state
->stack_mask
, state
->graph_idx
);
48 for (sp
= PTR_ALIGN(state
->orig_sp
, sizeof(long)); sp
;
49 sp
= PTR_ALIGN(stack_info
.next_sp
, sizeof(long))) {
50 if (get_stack_info(sp
, state
->task
, &stack_info
, &visit_mask
))
53 for (; sp
< stack_info
.end
; sp
++) {
55 word
= READ_ONCE_NOCHECK(*sp
);
62 printk_deferred("%p: %0*x ...\n",
63 sp
, BITS_PER_LONG
/4, 0);
67 printk_deferred("%p: %0*lx (%pB)\n",
68 sp
, BITS_PER_LONG
/4, word
, (void *)word
);
73 static size_t regs_size(struct pt_regs
*regs
)
75 /* x86_32 regs from kernel mode are two words shorter: */
76 if (IS_ENABLED(CONFIG_X86_32
) && !user_mode(regs
))
77 return sizeof(*regs
) - 2*sizeof(long);
82 static bool in_entry_code(unsigned long ip
)
84 char *addr
= (char *)ip
;
86 if (addr
>= __entry_text_start
&& addr
< __entry_text_end
)
89 if (addr
>= __irqentry_text_start
&& addr
< __irqentry_text_end
)
95 static inline unsigned long *last_frame(struct unwind_state
*state
)
97 return (unsigned long *)task_pt_regs(state
->task
) - 2;
100 static bool is_last_frame(struct unwind_state
*state
)
102 return state
->bp
== last_frame(state
);
106 #define GCC_REALIGN_WORDS 3
108 #define GCC_REALIGN_WORDS 1
111 static inline unsigned long *last_aligned_frame(struct unwind_state
*state
)
113 return last_frame(state
) - GCC_REALIGN_WORDS
;
116 static bool is_last_aligned_frame(struct unwind_state
*state
)
118 unsigned long *last_bp
= last_frame(state
);
119 unsigned long *aligned_bp
= last_aligned_frame(state
);
122 * GCC can occasionally decide to realign the stack pointer and change
123 * the offset of the stack frame in the prologue of a function called
124 * by head/entry code. Examples:
129 * and $0xfffffff8,%esp
134 * <x86_64_start_kernel>:
136 * and $0xfffffffffffffff0,%rsp
141 * After aligning the stack, it pushes a duplicate copy of the return
142 * address before pushing the frame pointer.
144 return (state
->bp
== aligned_bp
&& *(aligned_bp
+ 1) == *(last_bp
+ 1));
147 static bool is_last_ftrace_frame(struct unwind_state
*state
)
149 unsigned long *last_bp
= last_frame(state
);
150 unsigned long *last_ftrace_bp
= last_bp
- 3;
153 * When unwinding from an ftrace handler of a function called by entry
154 * code, the stack layout of the last frame is:
164 return (state
->bp
== last_ftrace_bp
&&
165 *state
->bp
== *(state
->bp
+ 2) &&
166 *(state
->bp
+ 1) == *(state
->bp
+ 4));
169 static bool is_last_task_frame(struct unwind_state
*state
)
171 return is_last_frame(state
) || is_last_aligned_frame(state
) ||
172 is_last_ftrace_frame(state
);
176 * This determines if the frame pointer actually contains an encoded pointer to
177 * pt_regs on the stack. See ENCODE_FRAME_POINTER.
180 static struct pt_regs
*decode_frame_pointer(unsigned long *bp
)
182 unsigned long regs
= (unsigned long)bp
;
187 return (struct pt_regs
*)(regs
& ~0x1);
190 static struct pt_regs
*decode_frame_pointer(unsigned long *bp
)
192 unsigned long regs
= (unsigned long)bp
;
194 if (regs
& 0x80000000)
197 return (struct pt_regs
*)(regs
| 0x80000000);
202 #define KERNEL_REGS_SIZE (sizeof(struct pt_regs) - 2*sizeof(long))
204 #define KERNEL_REGS_SIZE (sizeof(struct pt_regs))
207 static bool update_stack_state(struct unwind_state
*state
,
208 unsigned long *next_bp
)
210 struct stack_info
*info
= &state
->stack_info
;
211 enum stack_type prev_type
= info
->type
;
212 struct pt_regs
*regs
;
213 unsigned long *frame
, *prev_frame_end
, *addr_p
, addr
;
217 prev_frame_end
= (void *)state
->regs
+ regs_size(state
->regs
);
219 prev_frame_end
= (void *)state
->bp
+ FRAME_HEADER_SIZE
;
221 /* Is the next frame pointer an encoded pointer to pt_regs? */
222 regs
= decode_frame_pointer(next_bp
);
224 frame
= (unsigned long *)regs
;
225 len
= KERNEL_REGS_SIZE
;
226 state
->got_irq
= true;
229 len
= FRAME_HEADER_SIZE
;
233 * If the next bp isn't on the current stack, switch to the next one.
235 * We may have to traverse multiple stacks to deal with the possibility
236 * that info->next_sp could point to an empty stack and the next bp
237 * could be on a subsequent stack.
239 while (!on_stack(info
, frame
, len
))
240 if (get_stack_info(info
->next_sp
, state
->task
, info
,
244 /* Make sure it only unwinds up and doesn't overlap the prev frame: */
245 if (state
->orig_sp
&& state
->stack_info
.type
== prev_type
&&
246 frame
< prev_frame_end
)
250 * On 32-bit with user mode regs, make sure the last two regs are safe
253 if (IS_ENABLED(CONFIG_X86_32
) && regs
&& user_mode(regs
) &&
254 !on_stack(info
, frame
, len
+ 2*sizeof(long)))
257 /* Move state to the next frame: */
266 /* Save the return address: */
267 if (state
->regs
&& user_mode(state
->regs
))
270 addr_p
= unwind_get_return_address_ptr(state
);
271 addr
= READ_ONCE_TASK_STACK(state
->task
, *addr_p
);
272 state
->ip
= ftrace_graph_ret_addr(state
->task
, &state
->graph_idx
,
276 /* Save the original stack pointer for unwind_dump(): */
278 state
->orig_sp
= frame
;
283 bool unwind_next_frame(struct unwind_state
*state
)
285 struct pt_regs
*regs
;
286 unsigned long *next_bp
;
288 if (unwind_done(state
))
291 /* Have we reached the end? */
292 if (state
->regs
&& user_mode(state
->regs
))
295 if (is_last_task_frame(state
)) {
296 regs
= task_pt_regs(state
->task
);
299 * kthreads (other than the boot CPU's idle thread) have some
300 * partial regs at the end of their stack which were placed
301 * there by copy_thread_tls(). But the regs don't have any
302 * useful information, so we can skip them.
304 * This user_mode() check is slightly broader than a PF_KTHREAD
305 * check because it also catches the awkward situation where a
306 * newly forked kthread transitions into a user task by calling
307 * do_execve(), which eventually clears PF_KTHREAD.
309 if (!user_mode(regs
))
313 * We're almost at the end, but not quite: there's still the
314 * syscall regs frame. Entry code doesn't encode the regs
315 * pointer for syscalls, so we have to set it manually.
323 /* Get the next frame pointer: */
324 if (state
->next_bp
) {
325 next_bp
= state
->next_bp
;
326 state
->next_bp
= NULL
;
327 } else if (state
->regs
) {
328 next_bp
= (unsigned long *)state
->regs
->bp
;
330 next_bp
= (unsigned long *)READ_ONCE_TASK_STACK(state
->task
, *state
->bp
);
333 /* Move to the next frame if it's safe: */
334 if (!update_stack_state(state
, next_bp
))
343 * When unwinding a non-current task, the task might actually be
344 * running on another CPU, in which case it could be modifying its
345 * stack while we're reading it. This is generally not a problem and
346 * can be ignored as long as the caller understands that unwinding
347 * another task will not always succeed.
349 if (state
->task
!= current
)
353 * Don't warn if the unwinder got lost due to an interrupt in entry
354 * code or in the C handler before the first frame pointer got set up:
356 if (state
->got_irq
&& in_entry_code(state
->ip
))
359 state
->regs
->sp
>= (unsigned long)last_aligned_frame(state
) &&
360 state
->regs
->sp
< (unsigned long)task_pt_regs(state
->task
))
364 * There are some known frame pointer issues on 32-bit. Disable
365 * unwinder warnings on 32-bit until it gets objtool support.
367 if (IS_ENABLED(CONFIG_X86_32
))
371 printk_deferred_once(KERN_WARNING
372 "WARNING: kernel stack regs at %p in %s:%d has bad 'bp' value %p\n",
373 state
->regs
, state
->task
->comm
,
374 state
->task
->pid
, next_bp
);
377 printk_deferred_once(KERN_WARNING
378 "WARNING: kernel stack frame pointer at %p in %s:%d has bad value %p\n",
379 state
->bp
, state
->task
->comm
,
380 state
->task
->pid
, next_bp
);
384 state
->stack_info
.type
= STACK_TYPE_UNKNOWN
;
387 EXPORT_SYMBOL_GPL(unwind_next_frame
);
389 void __unwind_start(struct unwind_state
*state
, struct task_struct
*task
,
390 struct pt_regs
*regs
, unsigned long *first_frame
)
394 memset(state
, 0, sizeof(*state
));
396 state
->got_irq
= (regs
);
398 /* Don't even attempt to start from user mode regs: */
399 if (regs
&& user_mode(regs
)) {
400 state
->stack_info
.type
= STACK_TYPE_UNKNOWN
;
404 bp
= get_frame_pointer(task
, regs
);
407 * If we crash with IP==0, the last successfully executed instruction
408 * was probably an indirect function call with a NULL function pointer.
409 * That means that SP points into the middle of an incomplete frame:
410 * *SP is a return pointer, and *(SP-sizeof(unsigned long)) is where we
411 * would have written a frame pointer if we hadn't crashed.
412 * Pretend that the frame is complete and that BP points to it, but save
413 * the real BP so that we can use it when looking for the next frame.
415 if (regs
&& regs
->ip
== 0 &&
416 (unsigned long *)kernel_stack_pointer(regs
) >= first_frame
) {
418 bp
= ((unsigned long *)kernel_stack_pointer(regs
)) - 1;
421 /* Initialize stack info and make sure the frame data is accessible: */
422 get_stack_info(bp
, state
->task
, &state
->stack_info
,
424 update_stack_state(state
, bp
);
427 * The caller can provide the address of the first frame directly
428 * (first_frame) or indirectly (regs->sp) to indicate which stack frame
429 * to start unwinding at. Skip ahead until we reach it.
431 while (!unwind_done(state
) &&
432 (!on_stack(&state
->stack_info
, first_frame
, sizeof(long)) ||
433 (state
->next_bp
== NULL
&& state
->bp
< first_frame
)))
434 unwind_next_frame(state
);
436 EXPORT_SYMBOL_GPL(__unwind_start
);