]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blob - arch/x86/kernel/unwind_frame.c
x86/cpu/AMD: Fix erratum 1076 (CPB bit)
[mirror_ubuntu-artful-kernel.git] / arch / x86 / kernel / unwind_frame.c
1 #include <linux/sched.h>
2 #include <linux/sched/task.h>
3 #include <linux/sched/task_stack.h>
4 #include <linux/interrupt.h>
5 #include <asm/sections.h>
6 #include <asm/ptrace.h>
7 #include <asm/bitops.h>
8 #include <asm/stacktrace.h>
9 #include <asm/unwind.h>
10
11 #define FRAME_HEADER_SIZE (sizeof(long) * 2)
12
13 unsigned long unwind_get_return_address(struct unwind_state *state)
14 {
15 if (unwind_done(state))
16 return 0;
17
18 return __kernel_text_address(state->ip) ? state->ip : 0;
19 }
20 EXPORT_SYMBOL_GPL(unwind_get_return_address);
21
22 unsigned long *unwind_get_return_address_ptr(struct unwind_state *state)
23 {
24 if (unwind_done(state))
25 return NULL;
26
27 return state->regs ? &state->regs->ip : state->bp + 1;
28 }
29
30 static void unwind_dump(struct unwind_state *state)
31 {
32 static bool dumped_before = false;
33 bool prev_zero, zero = false;
34 unsigned long word, *sp;
35 struct stack_info stack_info = {0};
36 unsigned long visit_mask = 0;
37
38 if (dumped_before)
39 return;
40
41 dumped_before = true;
42
43 printk_deferred("unwind stack type:%d next_sp:%p mask:0x%lx graph_idx:%d\n",
44 state->stack_info.type, state->stack_info.next_sp,
45 state->stack_mask, state->graph_idx);
46
47 for (sp = state->orig_sp; sp; sp = PTR_ALIGN(stack_info.next_sp, sizeof(long))) {
48 if (get_stack_info(sp, state->task, &stack_info, &visit_mask))
49 break;
50
51 for (; sp < stack_info.end; sp++) {
52
53 word = READ_ONCE_NOCHECK(*sp);
54
55 prev_zero = zero;
56 zero = word == 0;
57
58 if (zero) {
59 if (!prev_zero)
60 printk_deferred("%p: %0*x ...\n",
61 sp, BITS_PER_LONG/4, 0);
62 continue;
63 }
64
65 printk_deferred("%p: %0*lx (%pB)\n",
66 sp, BITS_PER_LONG/4, word, (void *)word);
67 }
68 }
69 }
70
71 static size_t regs_size(struct pt_regs *regs)
72 {
73 /* x86_32 regs from kernel mode are two words shorter: */
74 if (IS_ENABLED(CONFIG_X86_32) && !user_mode(regs))
75 return sizeof(*regs) - 2*sizeof(long);
76
77 return sizeof(*regs);
78 }
79
80 static bool in_entry_code(unsigned long ip)
81 {
82 char *addr = (char *)ip;
83
84 if (addr >= __entry_text_start && addr < __entry_text_end)
85 return true;
86
87 if (addr >= __irqentry_text_start && addr < __irqentry_text_end)
88 return true;
89
90 return false;
91 }
92
93 static inline unsigned long *last_frame(struct unwind_state *state)
94 {
95 return (unsigned long *)task_pt_regs(state->task) - 2;
96 }
97
98 static bool is_last_frame(struct unwind_state *state)
99 {
100 return state->bp == last_frame(state);
101 }
102
103 #ifdef CONFIG_X86_32
104 #define GCC_REALIGN_WORDS 3
105 #else
106 #define GCC_REALIGN_WORDS 1
107 #endif
108
109 static inline unsigned long *last_aligned_frame(struct unwind_state *state)
110 {
111 return last_frame(state) - GCC_REALIGN_WORDS;
112 }
113
114 static bool is_last_aligned_frame(struct unwind_state *state)
115 {
116 unsigned long *last_bp = last_frame(state);
117 unsigned long *aligned_bp = last_aligned_frame(state);
118
119 /*
120 * GCC can occasionally decide to realign the stack pointer and change
121 * the offset of the stack frame in the prologue of a function called
122 * by head/entry code. Examples:
123 *
124 * <start_secondary>:
125 * push %edi
126 * lea 0x8(%esp),%edi
127 * and $0xfffffff8,%esp
128 * pushl -0x4(%edi)
129 * push %ebp
130 * mov %esp,%ebp
131 *
132 * <x86_64_start_kernel>:
133 * lea 0x8(%rsp),%r10
134 * and $0xfffffffffffffff0,%rsp
135 * pushq -0x8(%r10)
136 * push %rbp
137 * mov %rsp,%rbp
138 *
139 * After aligning the stack, it pushes a duplicate copy of the return
140 * address before pushing the frame pointer.
141 */
142 return (state->bp == aligned_bp && *(aligned_bp + 1) == *(last_bp + 1));
143 }
144
145 static bool is_last_ftrace_frame(struct unwind_state *state)
146 {
147 unsigned long *last_bp = last_frame(state);
148 unsigned long *last_ftrace_bp = last_bp - 3;
149
150 /*
151 * When unwinding from an ftrace handler of a function called by entry
152 * code, the stack layout of the last frame is:
153 *
154 * bp
155 * parent ret addr
156 * bp
157 * function ret addr
158 * parent ret addr
159 * pt_regs
160 * -----------------
161 */
162 return (state->bp == last_ftrace_bp &&
163 *state->bp == *(state->bp + 2) &&
164 *(state->bp + 1) == *(state->bp + 4));
165 }
166
167 static bool is_last_task_frame(struct unwind_state *state)
168 {
169 return is_last_frame(state) || is_last_aligned_frame(state) ||
170 is_last_ftrace_frame(state);
171 }
172
173 /*
174 * This determines if the frame pointer actually contains an encoded pointer to
175 * pt_regs on the stack. See ENCODE_FRAME_POINTER.
176 */
177 static struct pt_regs *decode_frame_pointer(unsigned long *bp)
178 {
179 unsigned long regs = (unsigned long)bp;
180
181 if (!(regs & 0x1))
182 return NULL;
183
184 return (struct pt_regs *)(regs & ~0x1);
185 }
186
187 #ifdef CONFIG_X86_32
188 #define KERNEL_REGS_SIZE (sizeof(struct pt_regs) - 2*sizeof(long))
189 #else
190 #define KERNEL_REGS_SIZE (sizeof(struct pt_regs))
191 #endif
192
193 static bool update_stack_state(struct unwind_state *state,
194 unsigned long *next_bp)
195 {
196 struct stack_info *info = &state->stack_info;
197 enum stack_type prev_type = info->type;
198 struct pt_regs *regs;
199 unsigned long *frame, *prev_frame_end, *addr_p, addr;
200 size_t len;
201
202 if (state->regs)
203 prev_frame_end = (void *)state->regs + regs_size(state->regs);
204 else
205 prev_frame_end = (void *)state->bp + FRAME_HEADER_SIZE;
206
207 /* Is the next frame pointer an encoded pointer to pt_regs? */
208 regs = decode_frame_pointer(next_bp);
209 if (regs) {
210 frame = (unsigned long *)regs;
211 len = KERNEL_REGS_SIZE;
212 state->got_irq = true;
213 } else {
214 frame = next_bp;
215 len = FRAME_HEADER_SIZE;
216 }
217
218 /*
219 * If the next bp isn't on the current stack, switch to the next one.
220 *
221 * We may have to traverse multiple stacks to deal with the possibility
222 * that info->next_sp could point to an empty stack and the next bp
223 * could be on a subsequent stack.
224 */
225 while (!on_stack(info, frame, len))
226 if (get_stack_info(info->next_sp, state->task, info,
227 &state->stack_mask))
228 return false;
229
230 /* Make sure it only unwinds up and doesn't overlap the prev frame: */
231 if (state->orig_sp && state->stack_info.type == prev_type &&
232 frame < prev_frame_end)
233 return false;
234
235 /*
236 * On 32-bit with user mode regs, make sure the last two regs are safe
237 * to access:
238 */
239 if (IS_ENABLED(CONFIG_X86_32) && regs && user_mode(regs) &&
240 !on_stack(info, frame, len + 2*sizeof(long)))
241 return false;
242
243 /* Move state to the next frame: */
244 if (regs) {
245 state->regs = regs;
246 state->bp = NULL;
247 } else {
248 state->bp = next_bp;
249 state->regs = NULL;
250 }
251
252 /* Save the return address: */
253 if (state->regs && user_mode(state->regs))
254 state->ip = 0;
255 else {
256 addr_p = unwind_get_return_address_ptr(state);
257 addr = READ_ONCE_TASK_STACK(state->task, *addr_p);
258 state->ip = ftrace_graph_ret_addr(state->task, &state->graph_idx,
259 addr, addr_p);
260 }
261
262 /* Save the original stack pointer for unwind_dump(): */
263 if (!state->orig_sp)
264 state->orig_sp = frame;
265
266 return true;
267 }
268
269 bool unwind_next_frame(struct unwind_state *state)
270 {
271 struct pt_regs *regs;
272 unsigned long *next_bp;
273
274 if (unwind_done(state))
275 return false;
276
277 /* Have we reached the end? */
278 if (state->regs && user_mode(state->regs))
279 goto the_end;
280
281 if (is_last_task_frame(state)) {
282 regs = task_pt_regs(state->task);
283
284 /*
285 * kthreads (other than the boot CPU's idle thread) have some
286 * partial regs at the end of their stack which were placed
287 * there by copy_thread_tls(). But the regs don't have any
288 * useful information, so we can skip them.
289 *
290 * This user_mode() check is slightly broader than a PF_KTHREAD
291 * check because it also catches the awkward situation where a
292 * newly forked kthread transitions into a user task by calling
293 * do_execve(), which eventually clears PF_KTHREAD.
294 */
295 if (!user_mode(regs))
296 goto the_end;
297
298 /*
299 * We're almost at the end, but not quite: there's still the
300 * syscall regs frame. Entry code doesn't encode the regs
301 * pointer for syscalls, so we have to set it manually.
302 */
303 state->regs = regs;
304 state->bp = NULL;
305 state->ip = 0;
306 return true;
307 }
308
309 /* Get the next frame pointer: */
310 if (state->regs)
311 next_bp = (unsigned long *)state->regs->bp;
312 else
313 next_bp = (unsigned long *)READ_ONCE_TASK_STACK(state->task, *state->bp);
314
315 /* Move to the next frame if it's safe: */
316 if (!update_stack_state(state, next_bp))
317 goto bad_address;
318
319 return true;
320
321 bad_address:
322 state->error = true;
323
324 /*
325 * When unwinding a non-current task, the task might actually be
326 * running on another CPU, in which case it could be modifying its
327 * stack while we're reading it. This is generally not a problem and
328 * can be ignored as long as the caller understands that unwinding
329 * another task will not always succeed.
330 */
331 if (state->task != current)
332 goto the_end;
333
334 /*
335 * Don't warn if the unwinder got lost due to an interrupt in entry
336 * code or in the C handler before the first frame pointer got set up:
337 */
338 if (state->got_irq && in_entry_code(state->ip))
339 goto the_end;
340 if (state->regs &&
341 state->regs->sp >= (unsigned long)last_aligned_frame(state) &&
342 state->regs->sp < (unsigned long)task_pt_regs(state->task))
343 goto the_end;
344
345 if (state->regs) {
346 printk_deferred_once(KERN_WARNING
347 "WARNING: kernel stack regs at %p in %s:%d has bad 'bp' value %p\n",
348 state->regs, state->task->comm,
349 state->task->pid, next_bp);
350 unwind_dump(state);
351 } else {
352 printk_deferred_once(KERN_WARNING
353 "WARNING: kernel stack frame pointer at %p in %s:%d has bad value %p\n",
354 state->bp, state->task->comm,
355 state->task->pid, next_bp);
356 unwind_dump(state);
357 }
358 the_end:
359 state->stack_info.type = STACK_TYPE_UNKNOWN;
360 return false;
361 }
362 EXPORT_SYMBOL_GPL(unwind_next_frame);
363
364 void __unwind_start(struct unwind_state *state, struct task_struct *task,
365 struct pt_regs *regs, unsigned long *first_frame)
366 {
367 unsigned long *bp;
368
369 memset(state, 0, sizeof(*state));
370 state->task = task;
371 state->got_irq = (regs);
372
373 /* Don't even attempt to start from user mode regs: */
374 if (regs && user_mode(regs)) {
375 state->stack_info.type = STACK_TYPE_UNKNOWN;
376 return;
377 }
378
379 bp = get_frame_pointer(task, regs);
380
381 /* Initialize stack info and make sure the frame data is accessible: */
382 get_stack_info(bp, state->task, &state->stack_info,
383 &state->stack_mask);
384 update_stack_state(state, bp);
385
386 /*
387 * The caller can provide the address of the first frame directly
388 * (first_frame) or indirectly (regs->sp) to indicate which stack frame
389 * to start unwinding at. Skip ahead until we reach it.
390 */
391 while (!unwind_done(state) &&
392 (!on_stack(&state->stack_info, first_frame, sizeof(long)) ||
393 state->bp < first_frame))
394 unwind_next_frame(state);
395 }
396 EXPORT_SYMBOL_GPL(__unwind_start);