xref: /linux/arch/x86/kernel/dumpstack_64.c (revision e26207a3819684e9b4450a2d30bdd065fa92d9c7)
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 #include "dumpstack.h"
20 
21 #define N_EXCEPTION_STACKS_END \
22 		(N_EXCEPTION_STACKS + DEBUG_STKSZ/EXCEPTION_STKSZ - 2)
23 
24 static char x86_stack_ids[][8] = {
25 		[ DEBUG_STACK-1			]	= "#DB",
26 		[ NMI_STACK-1			]	= "NMI",
27 		[ DOUBLEFAULT_STACK-1		]	= "#DF",
28 		[ STACKFAULT_STACK-1		]	= "#SS",
29 		[ MCE_STACK-1			]	= "#MC",
30 #if DEBUG_STKSZ > EXCEPTION_STKSZ
31 		[ N_EXCEPTION_STACKS ...
32 		  N_EXCEPTION_STACKS_END	]	= "#DB[?]"
33 #endif
34 };
35 
36 static unsigned long *in_exception_stack(unsigned cpu, unsigned long stack,
37 					 unsigned *usedp, char **idp)
38 {
39 	unsigned k;
40 
41 	/*
42 	 * Iterate over all exception stacks, and figure out whether
43 	 * 'stack' is in one of them:
44 	 */
45 	for (k = 0; k < N_EXCEPTION_STACKS; k++) {
46 		unsigned long end = per_cpu(orig_ist, cpu).ist[k];
47 		/*
48 		 * Is 'stack' above this exception frame's end?
49 		 * If yes then skip to the next frame.
50 		 */
51 		if (stack >= end)
52 			continue;
53 		/*
54 		 * Is 'stack' above this exception frame's start address?
55 		 * If yes then we found the right frame.
56 		 */
57 		if (stack >= end - EXCEPTION_STKSZ) {
58 			/*
59 			 * Make sure we only iterate through an exception
60 			 * stack once. If it comes up for the second time
61 			 * then there's something wrong going on - just
62 			 * break out and return NULL:
63 			 */
64 			if (*usedp & (1U << k))
65 				break;
66 			*usedp |= 1U << k;
67 			*idp = x86_stack_ids[k];
68 			return (unsigned long *)end;
69 		}
70 		/*
71 		 * If this is a debug stack, and if it has a larger size than
72 		 * the usual exception stacks, then 'stack' might still
73 		 * be within the lower portion of the debug stack:
74 		 */
75 #if DEBUG_STKSZ > EXCEPTION_STKSZ
76 		if (k == DEBUG_STACK - 1 && stack >= end - DEBUG_STKSZ) {
77 			unsigned j = N_EXCEPTION_STACKS - 1;
78 
79 			/*
80 			 * Black magic. A large debug stack is composed of
81 			 * multiple exception stack entries, which we
82 			 * iterate through now. Dont look:
83 			 */
84 			do {
85 				++j;
86 				end -= EXCEPTION_STKSZ;
87 				x86_stack_ids[j][4] = '1' +
88 						(j - N_EXCEPTION_STACKS);
89 			} while (stack < end - EXCEPTION_STKSZ);
90 			if (*usedp & (1U << j))
91 				break;
92 			*usedp |= 1U << j;
93 			*idp = x86_stack_ids[j];
94 			return (unsigned long *)end;
95 		}
96 #endif
97 	}
98 	return NULL;
99 }
100 
101 static inline int
102 in_irq_stack(unsigned long *stack, unsigned long *irq_stack,
103 	     unsigned long *irq_stack_end)
104 {
105 	return (stack >= irq_stack && stack < irq_stack_end);
106 }
107 
108 /*
109  * We are returning from the irq stack and go to the previous one.
110  * If the previous stack is also in the irq stack, then bp in the first
111  * frame of the irq stack points to the previous, interrupted one.
112  * Otherwise we have another level of indirection: We first save
113  * the bp of the previous stack, then we switch the stack to the irq one
114  * and save a new bp that links to the previous one.
115  * (See save_args())
116  */
117 static inline unsigned long
118 fixup_bp_irq_link(unsigned long bp, unsigned long *stack,
119 		  unsigned long *irq_stack, unsigned long *irq_stack_end)
120 {
121 #ifdef CONFIG_FRAME_POINTER
122 	struct stack_frame *frame = (struct stack_frame *)bp;
123 
124 	if (!in_irq_stack(stack, irq_stack, irq_stack_end))
125 		return (unsigned long)frame->next_frame;
126 #endif
127 	return bp;
128 }
129 
130 /*
131  * x86-64 can have up to three kernel stacks:
132  * process stack
133  * interrupt stack
134  * severe exception (double fault, nmi, stack fault, debug, mce) hardware stack
135  */
136 
137 void dump_trace(struct task_struct *task, struct pt_regs *regs,
138 		unsigned long *stack, unsigned long bp,
139 		const struct stacktrace_ops *ops, void *data)
140 {
141 	const unsigned cpu = get_cpu();
142 	unsigned long *irq_stack_end =
143 		(unsigned long *)per_cpu(irq_stack_ptr, cpu);
144 	unsigned used = 0;
145 	struct thread_info *tinfo;
146 	int graph = 0;
147 
148 	if (!task)
149 		task = current;
150 
151 	if (!stack) {
152 		unsigned long dummy;
153 		stack = &dummy;
154 		if (task && task != current)
155 			stack = (unsigned long *)task->thread.sp;
156 	}
157 
158 #ifdef CONFIG_FRAME_POINTER
159 	if (!bp) {
160 		if (task == current) {
161 			/* Grab bp right from our regs */
162 			get_bp(bp);
163 		} else {
164 			/* bp is the last reg pushed by switch_to */
165 			bp = *(unsigned long *) task->thread.sp;
166 		}
167 	}
168 #endif
169 
170 	/*
171 	 * Print function call entries in all stacks, starting at the
172 	 * current stack address. If the stacks consist of nested
173 	 * exceptions
174 	 */
175 	tinfo = task_thread_info(task);
176 	for (;;) {
177 		char *id;
178 		unsigned long *estack_end;
179 		estack_end = in_exception_stack(cpu, (unsigned long)stack,
180 						&used, &id);
181 
182 		if (estack_end) {
183 			if (ops->stack(data, id) < 0)
184 				break;
185 
186 			bp = ops->walk_stack(tinfo, stack, bp, ops,
187 					     data, estack_end, &graph);
188 			ops->stack(data, "<EOE>");
189 			/*
190 			 * We link to the next stack via the
191 			 * second-to-last pointer (index -2 to end) in the
192 			 * exception stack:
193 			 */
194 			stack = (unsigned long *) estack_end[-2];
195 			continue;
196 		}
197 		if (irq_stack_end) {
198 			unsigned long *irq_stack;
199 			irq_stack = irq_stack_end -
200 				(IRQ_STACK_SIZE - 64) / sizeof(*irq_stack);
201 
202 			if (in_irq_stack(stack, irq_stack, irq_stack_end)) {
203 				if (ops->stack(data, "IRQ") < 0)
204 					break;
205 				bp = print_context_stack(tinfo, stack, bp,
206 					ops, data, irq_stack_end, &graph);
207 				/*
208 				 * We link to the next stack (which would be
209 				 * the process stack normally) the last
210 				 * pointer (index -1 to end) in the IRQ stack:
211 				 */
212 				stack = (unsigned long *) (irq_stack_end[-1]);
213 				bp = fixup_bp_irq_link(bp, stack, irq_stack,
214 						       irq_stack_end);
215 				irq_stack_end = NULL;
216 				ops->stack(data, "EOI");
217 				continue;
218 			}
219 		}
220 		break;
221 	}
222 
223 	/*
224 	 * This handles the process stack:
225 	 */
226 	bp = print_context_stack(tinfo, stack, bp, ops, data, NULL, &graph);
227 	put_cpu();
228 }
229 EXPORT_SYMBOL(dump_trace);
230 
231 void
232 show_stack_log_lvl(struct task_struct *task, struct pt_regs *regs,
233 		   unsigned long *sp, unsigned long bp, char *log_lvl)
234 {
235 	unsigned long *irq_stack_end;
236 	unsigned long *irq_stack;
237 	unsigned long *stack;
238 	int cpu;
239 	int i;
240 
241 	preempt_disable();
242 	cpu = smp_processor_id();
243 
244 	irq_stack_end	= (unsigned long *)(per_cpu(irq_stack_ptr, cpu));
245 	irq_stack	= (unsigned long *)(per_cpu(irq_stack_ptr, cpu) - IRQ_STACK_SIZE);
246 
247 	/*
248 	 * Debugging aid: "show_stack(NULL, NULL);" prints the
249 	 * back trace for this cpu:
250 	 */
251 	if (sp == NULL) {
252 		if (task)
253 			sp = (unsigned long *)task->thread.sp;
254 		else
255 			sp = (unsigned long *)&sp;
256 	}
257 
258 	stack = sp;
259 	for (i = 0; i < kstack_depth_to_print; i++) {
260 		if (stack >= irq_stack && stack <= irq_stack_end) {
261 			if (stack == irq_stack_end) {
262 				stack = (unsigned long *) (irq_stack_end[-1]);
263 				printk(" <EOI> ");
264 			}
265 		} else {
266 		if (((long) stack & (THREAD_SIZE-1)) == 0)
267 			break;
268 		}
269 		if (i && ((i % STACKSLOTS_PER_LINE) == 0))
270 			printk("\n%s", log_lvl);
271 		printk(" %016lx", *stack++);
272 		touch_nmi_watchdog();
273 	}
274 	preempt_enable();
275 
276 	printk("\n");
277 	show_trace_log_lvl(task, regs, sp, bp, log_lvl);
278 }
279 
280 void show_registers(struct pt_regs *regs)
281 {
282 	int i;
283 	unsigned long sp;
284 	const int cpu = smp_processor_id();
285 	struct task_struct *cur = current;
286 
287 	sp = regs->sp;
288 	printk("CPU %d ", cpu);
289 	__show_regs(regs, 1);
290 	printk("Process %s (pid: %d, threadinfo %p, task %p)\n",
291 		cur->comm, cur->pid, task_thread_info(cur), cur);
292 
293 	/*
294 	 * When in-kernel, we also print out the stack and code at the
295 	 * time of the fault..
296 	 */
297 	if (!user_mode(regs)) {
298 		unsigned int code_prologue = code_bytes * 43 / 64;
299 		unsigned int code_len = code_bytes;
300 		unsigned char c;
301 		u8 *ip;
302 
303 		printk(KERN_EMERG "Stack:\n");
304 		show_stack_log_lvl(NULL, regs, (unsigned long *)sp,
305 				regs->bp, KERN_EMERG);
306 
307 		printk(KERN_EMERG "Code: ");
308 
309 		ip = (u8 *)regs->ip - code_prologue;
310 		if (ip < (u8 *)PAGE_OFFSET || probe_kernel_address(ip, c)) {
311 			/* try starting at IP */
312 			ip = (u8 *)regs->ip;
313 			code_len = code_len - code_prologue + 1;
314 		}
315 		for (i = 0; i < code_len; i++, ip++) {
316 			if (ip < (u8 *)PAGE_OFFSET ||
317 					probe_kernel_address(ip, c)) {
318 				printk(" Bad RIP value.");
319 				break;
320 			}
321 			if (ip == (u8 *)regs->ip)
322 				printk("<%02x> ", c);
323 			else
324 				printk("%02x ", c);
325 		}
326 	}
327 	printk("\n");
328 }
329 
330 int is_valid_bugaddr(unsigned long ip)
331 {
332 	unsigned short ud2;
333 
334 	if (__copy_from_user(&ud2, (const void __user *) ip, sizeof(ud2)))
335 		return 0;
336 
337 	return ud2 == 0x0b0f;
338 }
339