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 bool in_entry_code(unsigned long ip
)
75 char *addr
= (char *)ip
;
77 return addr
>= __entry_text_start
&& addr
< __entry_text_end
;
80 static inline unsigned long *last_frame(struct unwind_state
*state
)
82 return (unsigned long *)task_pt_regs(state
->task
) - 2;
85 static bool is_last_frame(struct unwind_state
*state
)
87 return state
->bp
== last_frame(state
);
91 #define GCC_REALIGN_WORDS 3
93 #define GCC_REALIGN_WORDS 1
96 static inline unsigned long *last_aligned_frame(struct unwind_state
*state
)
98 return last_frame(state
) - GCC_REALIGN_WORDS
;
101 static bool is_last_aligned_frame(struct unwind_state
*state
)
103 unsigned long *last_bp
= last_frame(state
);
104 unsigned long *aligned_bp
= last_aligned_frame(state
);
107 * GCC can occasionally decide to realign the stack pointer and change
108 * the offset of the stack frame in the prologue of a function called
109 * by head/entry code. Examples:
114 * and $0xfffffff8,%esp
119 * <x86_64_start_kernel>:
121 * and $0xfffffffffffffff0,%rsp
126 * After aligning the stack, it pushes a duplicate copy of the return
127 * address before pushing the frame pointer.
129 return (state
->bp
== aligned_bp
&& *(aligned_bp
+ 1) == *(last_bp
+ 1));
132 static bool is_last_ftrace_frame(struct unwind_state
*state
)
134 unsigned long *last_bp
= last_frame(state
);
135 unsigned long *last_ftrace_bp
= last_bp
- 3;
138 * When unwinding from an ftrace handler of a function called by entry
139 * code, the stack layout of the last frame is:
149 return (state
->bp
== last_ftrace_bp
&&
150 *state
->bp
== *(state
->bp
+ 2) &&
151 *(state
->bp
+ 1) == *(state
->bp
+ 4));
154 static bool is_last_task_frame(struct unwind_state
*state
)
156 return is_last_frame(state
) || is_last_aligned_frame(state
) ||
157 is_last_ftrace_frame(state
);
161 * This determines if the frame pointer actually contains an encoded pointer to
162 * pt_regs on the stack. See ENCODE_FRAME_POINTER.
165 static struct pt_regs
*decode_frame_pointer(unsigned long *bp
)
167 unsigned long regs
= (unsigned long)bp
;
172 return (struct pt_regs
*)(regs
& ~0x1);
175 static struct pt_regs
*decode_frame_pointer(unsigned long *bp
)
177 unsigned long regs
= (unsigned long)bp
;
179 if (regs
& 0x80000000)
182 return (struct pt_regs
*)(regs
| 0x80000000);
187 * While walking the stack, KMSAN may stomp on stale locals from other
188 * functions that were marked as uninitialized upon function exit, and
189 * now hold the call frame information for the current function (e.g. the frame
190 * pointer). Because KMSAN does not specifically mark call frames as
191 * initialized, false positive reports are possible. To prevent such reports,
192 * we mark the functions scanning the stack (here and below) with
196 static bool update_stack_state(struct unwind_state
*state
,
197 unsigned long *next_bp
)
199 struct stack_info
*info
= &state
->stack_info
;
200 enum stack_type prev_type
= info
->type
;
201 struct pt_regs
*regs
;
202 unsigned long *frame
, *prev_frame_end
, *addr_p
, addr
;
206 prev_frame_end
= (void *)state
->regs
+ sizeof(*state
->regs
);
208 prev_frame_end
= (void *)state
->bp
+ FRAME_HEADER_SIZE
;
210 /* Is the next frame pointer an encoded pointer to pt_regs? */
211 regs
= decode_frame_pointer(next_bp
);
213 frame
= (unsigned long *)regs
;
215 state
->got_irq
= true;
218 len
= FRAME_HEADER_SIZE
;
222 * If the next bp isn't on the current stack, switch to the next one.
224 * We may have to traverse multiple stacks to deal with the possibility
225 * that info->next_sp could point to an empty stack and the next bp
226 * could be on a subsequent stack.
228 while (!on_stack(info
, frame
, len
))
229 if (get_stack_info(info
->next_sp
, state
->task
, info
,
233 /* Make sure it only unwinds up and doesn't overlap the prev frame: */
234 if (state
->orig_sp
&& state
->stack_info
.type
== prev_type
&&
235 frame
< prev_frame_end
)
238 /* Move state to the next frame: */
247 /* Save the return address: */
248 if (state
->regs
&& user_mode(state
->regs
))
251 addr_p
= unwind_get_return_address_ptr(state
);
252 addr
= READ_ONCE_TASK_STACK(state
->task
, *addr_p
);
253 state
->ip
= unwind_recover_ret_addr(state
, addr
, addr_p
);
256 /* Save the original stack pointer for unwind_dump(): */
258 state
->orig_sp
= frame
;
264 bool unwind_next_frame(struct unwind_state
*state
)
266 struct pt_regs
*regs
;
267 unsigned long *next_bp
;
269 if (unwind_done(state
))
272 /* Have we reached the end? */
273 if (state
->regs
&& user_mode(state
->regs
))
276 if (is_last_task_frame(state
)) {
277 regs
= task_pt_regs(state
->task
);
280 * kthreads (other than the boot CPU's idle thread) have some
281 * partial regs at the end of their stack which were placed
282 * there by copy_thread(). But the regs don't have any
283 * useful information, so we can skip them.
285 * This user_mode() check is slightly broader than a PF_KTHREAD
286 * check because it also catches the awkward situation where a
287 * newly forked kthread transitions into a user task by calling
288 * kernel_execve(), which eventually clears PF_KTHREAD.
290 if (!user_mode(regs
))
294 * We're almost at the end, but not quite: there's still the
295 * syscall regs frame. Entry code doesn't encode the regs
296 * pointer for syscalls, so we have to set it manually.
304 /* Get the next frame pointer: */
305 if (state
->next_bp
) {
306 next_bp
= state
->next_bp
;
307 state
->next_bp
= NULL
;
308 } else if (state
->regs
) {
309 next_bp
= (unsigned long *)state
->regs
->bp
;
311 next_bp
= (unsigned long *)READ_ONCE_TASK_STACK(state
->task
, *state
->bp
);
314 /* Move to the next frame if it's safe: */
315 if (!update_stack_state(state
, next_bp
))
324 * When unwinding a non-current task, the task might actually be
325 * running on another CPU, in which case it could be modifying its
326 * stack while we're reading it. This is generally not a problem and
327 * can be ignored as long as the caller understands that unwinding
328 * another task will not always succeed.
330 if (state
->task
!= current
)
334 * Don't warn if the unwinder got lost due to an interrupt in entry
335 * code or in the C handler before the first frame pointer got set up:
337 if (state
->got_irq
&& in_entry_code(state
->ip
))
340 state
->regs
->sp
>= (unsigned long)last_aligned_frame(state
) &&
341 state
->regs
->sp
< (unsigned long)task_pt_regs(state
->task
))
345 * There are some known frame pointer issues on 32-bit. Disable
346 * unwinder warnings on 32-bit until it gets objtool support.
348 if (IS_ENABLED(CONFIG_X86_32
))
351 if (state
->task
!= current
)
355 printk_deferred_once(KERN_WARNING
356 "WARNING: kernel stack regs at %p in %s:%d has bad 'bp' value %p\n",
357 state
->regs
, state
->task
->comm
,
358 state
->task
->pid
, next_bp
);
361 printk_deferred_once(KERN_WARNING
362 "WARNING: kernel stack frame pointer at %p in %s:%d has bad value %p\n",
363 state
->bp
, state
->task
->comm
,
364 state
->task
->pid
, next_bp
);
368 state
->stack_info
.type
= STACK_TYPE_UNKNOWN
;
371 EXPORT_SYMBOL_GPL(unwind_next_frame
);
373 void __unwind_start(struct unwind_state
*state
, struct task_struct
*task
,
374 struct pt_regs
*regs
, unsigned long *first_frame
)
378 memset(state
, 0, sizeof(*state
));
380 state
->got_irq
= (regs
);
382 /* Don't even attempt to start from user mode regs: */
383 if (regs
&& user_mode(regs
)) {
384 state
->stack_info
.type
= STACK_TYPE_UNKNOWN
;
388 bp
= get_frame_pointer(task
, regs
);
391 * If we crash with IP==0, the last successfully executed instruction
392 * was probably an indirect function call with a NULL function pointer.
393 * That means that SP points into the middle of an incomplete frame:
394 * *SP is a return pointer, and *(SP-sizeof(unsigned long)) is where we
395 * would have written a frame pointer if we hadn't crashed.
396 * Pretend that the frame is complete and that BP points to it, but save
397 * the real BP so that we can use it when looking for the next frame.
399 if (regs
&& regs
->ip
== 0 && (unsigned long *)regs
->sp
>= first_frame
) {
401 bp
= ((unsigned long *)regs
->sp
) - 1;
404 /* Initialize stack info and make sure the frame data is accessible: */
405 get_stack_info(bp
, state
->task
, &state
->stack_info
,
407 update_stack_state(state
, bp
);
410 * The caller can provide the address of the first frame directly
411 * (first_frame) or indirectly (regs->sp) to indicate which stack frame
412 * to start unwinding at. Skip ahead until we reach it.
414 while (!unwind_done(state
) &&
415 (!on_stack(&state
->stack_info
, first_frame
, sizeof(long)) ||
416 (state
->next_bp
== NULL
&& state
->bp
< first_frame
)))
417 unwind_next_frame(state
);
419 EXPORT_SYMBOL_GPL(__unwind_start
);