xref: /linux/arch/x86/kernel/dumpstack_64.c (revision c0e297dc61f8d4453e07afbea1fa8d0e67cd4a34)
1 /*
2  *  Copyright (C) 1991, 1992  Linus Torvalds
3  *  Copyright (C) 2000, 2001, 2002 Andi Kleen, SuSE Labs
4  */
5 #include <linux/kallsyms.h>
6 #include <linux/kprobes.h>
7 #include <linux/uaccess.h>
8 #include <linux/hardirq.h>
9 #include <linux/kdebug.h>
10 #include <linux/module.h>
11 #include <linux/ptrace.h>
12 #include <linux/kexec.h>
13 #include <linux/sysfs.h>
14 #include <linux/bug.h>
15 #include <linux/nmi.h>
16 
17 #include <asm/stacktrace.h>
18 
19 
20 #define N_EXCEPTION_STACKS_END \
21 		(N_EXCEPTION_STACKS + DEBUG_STKSZ/EXCEPTION_STKSZ - 2)
22 
23 static char x86_stack_ids[][8] = {
24 		[ DEBUG_STACK-1			]	= "#DB",
25 		[ NMI_STACK-1			]	= "NMI",
26 		[ DOUBLEFAULT_STACK-1		]	= "#DF",
27 		[ MCE_STACK-1			]	= "#MC",
28 #if DEBUG_STKSZ > EXCEPTION_STKSZ
29 		[ N_EXCEPTION_STACKS ...
30 		  N_EXCEPTION_STACKS_END	]	= "#DB[?]"
31 #endif
32 };
33 
34 static unsigned long *in_exception_stack(unsigned cpu, unsigned long stack,
35 					 unsigned *usedp, char **idp)
36 {
37 	unsigned k;
38 
39 	/*
40 	 * Iterate over all exception stacks, and figure out whether
41 	 * 'stack' is in one of them:
42 	 */
43 	for (k = 0; k < N_EXCEPTION_STACKS; k++) {
44 		unsigned long end = per_cpu(orig_ist, cpu).ist[k];
45 		/*
46 		 * Is 'stack' above this exception frame's end?
47 		 * If yes then skip to the next frame.
48 		 */
49 		if (stack >= end)
50 			continue;
51 		/*
52 		 * Is 'stack' above this exception frame's start address?
53 		 * If yes then we found the right frame.
54 		 */
55 		if (stack >= end - EXCEPTION_STKSZ) {
56 			/*
57 			 * Make sure we only iterate through an exception
58 			 * stack once. If it comes up for the second time
59 			 * then there's something wrong going on - just
60 			 * break out and return NULL:
61 			 */
62 			if (*usedp & (1U << k))
63 				break;
64 			*usedp |= 1U << k;
65 			*idp = x86_stack_ids[k];
66 			return (unsigned long *)end;
67 		}
68 		/*
69 		 * If this is a debug stack, and if it has a larger size than
70 		 * the usual exception stacks, then 'stack' might still
71 		 * be within the lower portion of the debug stack:
72 		 */
73 #if DEBUG_STKSZ > EXCEPTION_STKSZ
74 		if (k == DEBUG_STACK - 1 && stack >= end - DEBUG_STKSZ) {
75 			unsigned j = N_EXCEPTION_STACKS - 1;
76 
77 			/*
78 			 * Black magic. A large debug stack is composed of
79 			 * multiple exception stack entries, which we
80 			 * iterate through now. Dont look:
81 			 */
82 			do {
83 				++j;
84 				end -= EXCEPTION_STKSZ;
85 				x86_stack_ids[j][4] = '1' +
86 						(j - N_EXCEPTION_STACKS);
87 			} while (stack < end - EXCEPTION_STKSZ);
88 			if (*usedp & (1U << j))
89 				break;
90 			*usedp |= 1U << j;
91 			*idp = x86_stack_ids[j];
92 			return (unsigned long *)end;
93 		}
94 #endif
95 	}
96 	return NULL;
97 }
98 
99 static inline int
100 in_irq_stack(unsigned long *stack, unsigned long *irq_stack,
101 	     unsigned long *irq_stack_end)
102 {
103 	return (stack >= irq_stack && stack < irq_stack_end);
104 }
105 
106 static const unsigned long irq_stack_size =
107 	(IRQ_STACK_SIZE - 64) / sizeof(unsigned long);
108 
109 enum stack_type {
110 	STACK_IS_UNKNOWN,
111 	STACK_IS_NORMAL,
112 	STACK_IS_EXCEPTION,
113 	STACK_IS_IRQ,
114 };
115 
116 static enum stack_type
117 analyze_stack(int cpu, struct task_struct *task, unsigned long *stack,
118 	      unsigned long **stack_end, unsigned long *irq_stack,
119 	      unsigned *used, char **id)
120 {
121 	unsigned long addr;
122 
123 	addr = ((unsigned long)stack & (~(THREAD_SIZE - 1)));
124 	if ((unsigned long)task_stack_page(task) == addr)
125 		return STACK_IS_NORMAL;
126 
127 	*stack_end = in_exception_stack(cpu, (unsigned long)stack,
128 					used, id);
129 	if (*stack_end)
130 		return STACK_IS_EXCEPTION;
131 
132 	if (!irq_stack)
133 		return STACK_IS_NORMAL;
134 
135 	*stack_end = irq_stack;
136 	irq_stack = irq_stack - irq_stack_size;
137 
138 	if (in_irq_stack(stack, irq_stack, *stack_end))
139 		return STACK_IS_IRQ;
140 
141 	return STACK_IS_UNKNOWN;
142 }
143 
144 /*
145  * x86-64 can have up to three kernel stacks:
146  * process stack
147  * interrupt stack
148  * severe exception (double fault, nmi, stack fault, debug, mce) hardware stack
149  */
150 
151 void dump_trace(struct task_struct *task, struct pt_regs *regs,
152 		unsigned long *stack, unsigned long bp,
153 		const struct stacktrace_ops *ops, void *data)
154 {
155 	const unsigned cpu = get_cpu();
156 	struct thread_info *tinfo;
157 	unsigned long *irq_stack = (unsigned long *)per_cpu(irq_stack_ptr, cpu);
158 	unsigned long dummy;
159 	unsigned used = 0;
160 	int graph = 0;
161 	int done = 0;
162 
163 	if (!task)
164 		task = current;
165 
166 	if (!stack) {
167 		if (regs)
168 			stack = (unsigned long *)regs->sp;
169 		else if (task != current)
170 			stack = (unsigned long *)task->thread.sp;
171 		else
172 			stack = &dummy;
173 	}
174 
175 	if (!bp)
176 		bp = stack_frame(task, regs);
177 	/*
178 	 * Print function call entries in all stacks, starting at the
179 	 * current stack address. If the stacks consist of nested
180 	 * exceptions
181 	 */
182 	tinfo = task_thread_info(task);
183 	while (!done) {
184 		unsigned long *stack_end;
185 		enum stack_type stype;
186 		char *id;
187 
188 		stype = analyze_stack(cpu, task, stack, &stack_end,
189 				      irq_stack, &used, &id);
190 
191 		/* Default finish unless specified to continue */
192 		done = 1;
193 
194 		switch (stype) {
195 
196 		/* Break out early if we are on the thread stack */
197 		case STACK_IS_NORMAL:
198 			break;
199 
200 		case STACK_IS_EXCEPTION:
201 
202 			if (ops->stack(data, id) < 0)
203 				break;
204 
205 			bp = ops->walk_stack(tinfo, stack, bp, ops,
206 					     data, stack_end, &graph);
207 			ops->stack(data, "<EOE>");
208 			/*
209 			 * We link to the next stack via the
210 			 * second-to-last pointer (index -2 to end) in the
211 			 * exception stack:
212 			 */
213 			stack = (unsigned long *) stack_end[-2];
214 			done = 0;
215 			break;
216 
217 		case STACK_IS_IRQ:
218 
219 			if (ops->stack(data, "IRQ") < 0)
220 				break;
221 			bp = ops->walk_stack(tinfo, stack, bp,
222 				     ops, data, stack_end, &graph);
223 			/*
224 			 * We link to the next stack (which would be
225 			 * the process stack normally) the last
226 			 * pointer (index -1 to end) in the IRQ stack:
227 			 */
228 			stack = (unsigned long *) (stack_end[-1]);
229 			irq_stack = NULL;
230 			ops->stack(data, "EOI");
231 			done = 0;
232 			break;
233 
234 		case STACK_IS_UNKNOWN:
235 			ops->stack(data, "UNK");
236 			break;
237 		}
238 	}
239 
240 	/*
241 	 * This handles the process stack:
242 	 */
243 	bp = ops->walk_stack(tinfo, stack, bp, ops, data, NULL, &graph);
244 	put_cpu();
245 }
246 EXPORT_SYMBOL(dump_trace);
247 
248 void
249 show_stack_log_lvl(struct task_struct *task, struct pt_regs *regs,
250 		   unsigned long *sp, unsigned long bp, char *log_lvl)
251 {
252 	unsigned long *irq_stack_end;
253 	unsigned long *irq_stack;
254 	unsigned long *stack;
255 	int cpu;
256 	int i;
257 
258 	preempt_disable();
259 	cpu = smp_processor_id();
260 
261 	irq_stack_end	= (unsigned long *)(per_cpu(irq_stack_ptr, cpu));
262 	irq_stack	= (unsigned long *)(per_cpu(irq_stack_ptr, cpu) - IRQ_STACK_SIZE);
263 
264 	/*
265 	 * Debugging aid: "show_stack(NULL, NULL);" prints the
266 	 * back trace for this cpu:
267 	 */
268 	if (sp == NULL) {
269 		if (task)
270 			sp = (unsigned long *)task->thread.sp;
271 		else
272 			sp = (unsigned long *)&sp;
273 	}
274 
275 	stack = sp;
276 	for (i = 0; i < kstack_depth_to_print; i++) {
277 		if (stack >= irq_stack && stack <= irq_stack_end) {
278 			if (stack == irq_stack_end) {
279 				stack = (unsigned long *) (irq_stack_end[-1]);
280 				pr_cont(" <EOI> ");
281 			}
282 		} else {
283 		if (kstack_end(stack))
284 			break;
285 		}
286 		if ((i % STACKSLOTS_PER_LINE) == 0) {
287 			if (i != 0)
288 				pr_cont("\n");
289 			printk("%s %016lx", log_lvl, *stack++);
290 		} else
291 			pr_cont(" %016lx", *stack++);
292 		touch_nmi_watchdog();
293 	}
294 	preempt_enable();
295 
296 	pr_cont("\n");
297 	show_trace_log_lvl(task, regs, sp, bp, log_lvl);
298 }
299 
300 void show_regs(struct pt_regs *regs)
301 {
302 	int i;
303 	unsigned long sp;
304 
305 	sp = regs->sp;
306 	show_regs_print_info(KERN_DEFAULT);
307 	__show_regs(regs, 1);
308 
309 	/*
310 	 * When in-kernel, we also print out the stack and code at the
311 	 * time of the fault..
312 	 */
313 	if (!user_mode(regs)) {
314 		unsigned int code_prologue = code_bytes * 43 / 64;
315 		unsigned int code_len = code_bytes;
316 		unsigned char c;
317 		u8 *ip;
318 
319 		printk(KERN_DEFAULT "Stack:\n");
320 		show_stack_log_lvl(NULL, regs, (unsigned long *)sp,
321 				   0, KERN_DEFAULT);
322 
323 		printk(KERN_DEFAULT "Code: ");
324 
325 		ip = (u8 *)regs->ip - code_prologue;
326 		if (ip < (u8 *)PAGE_OFFSET || probe_kernel_address(ip, c)) {
327 			/* try starting at IP */
328 			ip = (u8 *)regs->ip;
329 			code_len = code_len - code_prologue + 1;
330 		}
331 		for (i = 0; i < code_len; i++, ip++) {
332 			if (ip < (u8 *)PAGE_OFFSET ||
333 					probe_kernel_address(ip, c)) {
334 				pr_cont(" Bad RIP value.");
335 				break;
336 			}
337 			if (ip == (u8 *)regs->ip)
338 				pr_cont("<%02x> ", c);
339 			else
340 				pr_cont("%02x ", c);
341 		}
342 	}
343 	pr_cont("\n");
344 }
345 
346 int is_valid_bugaddr(unsigned long ip)
347 {
348 	unsigned short ud2;
349 
350 	if (__copy_from_user(&ud2, (const void __user *) ip, sizeof(ud2)))
351 		return 0;
352 
353 	return ud2 == 0x0b0f;
354 }
355