xref: /openbmc/linux/arch/x86/kernel/dumpstack_64.c (revision a09d2831)
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 int x86_is_stack_id(int id, char *name)
37 {
38 	return x86_stack_ids[id - 1] == name;
39 }
40 
41 static unsigned long *in_exception_stack(unsigned cpu, unsigned long stack,
42 					 unsigned *usedp, char **idp)
43 {
44 	unsigned k;
45 
46 	/*
47 	 * Iterate over all exception stacks, and figure out whether
48 	 * 'stack' is in one of them:
49 	 */
50 	for (k = 0; k < N_EXCEPTION_STACKS; k++) {
51 		unsigned long end = per_cpu(orig_ist, cpu).ist[k];
52 		/*
53 		 * Is 'stack' above this exception frame's end?
54 		 * If yes then skip to the next frame.
55 		 */
56 		if (stack >= end)
57 			continue;
58 		/*
59 		 * Is 'stack' above this exception frame's start address?
60 		 * If yes then we found the right frame.
61 		 */
62 		if (stack >= end - EXCEPTION_STKSZ) {
63 			/*
64 			 * Make sure we only iterate through an exception
65 			 * stack once. If it comes up for the second time
66 			 * then there's something wrong going on - just
67 			 * break out and return NULL:
68 			 */
69 			if (*usedp & (1U << k))
70 				break;
71 			*usedp |= 1U << k;
72 			*idp = x86_stack_ids[k];
73 			return (unsigned long *)end;
74 		}
75 		/*
76 		 * If this is a debug stack, and if it has a larger size than
77 		 * the usual exception stacks, then 'stack' might still
78 		 * be within the lower portion of the debug stack:
79 		 */
80 #if DEBUG_STKSZ > EXCEPTION_STKSZ
81 		if (k == DEBUG_STACK - 1 && stack >= end - DEBUG_STKSZ) {
82 			unsigned j = N_EXCEPTION_STACKS - 1;
83 
84 			/*
85 			 * Black magic. A large debug stack is composed of
86 			 * multiple exception stack entries, which we
87 			 * iterate through now. Dont look:
88 			 */
89 			do {
90 				++j;
91 				end -= EXCEPTION_STKSZ;
92 				x86_stack_ids[j][4] = '1' +
93 						(j - N_EXCEPTION_STACKS);
94 			} while (stack < end - EXCEPTION_STKSZ);
95 			if (*usedp & (1U << j))
96 				break;
97 			*usedp |= 1U << j;
98 			*idp = x86_stack_ids[j];
99 			return (unsigned long *)end;
100 		}
101 #endif
102 	}
103 	return NULL;
104 }
105 
106 static inline int
107 in_irq_stack(unsigned long *stack, unsigned long *irq_stack,
108 	     unsigned long *irq_stack_end)
109 {
110 	return (stack >= irq_stack && stack < irq_stack_end);
111 }
112 
113 /*
114  * We are returning from the irq stack and go to the previous one.
115  * If the previous stack is also in the irq stack, then bp in the first
116  * frame of the irq stack points to the previous, interrupted one.
117  * Otherwise we have another level of indirection: We first save
118  * the bp of the previous stack, then we switch the stack to the irq one
119  * and save a new bp that links to the previous one.
120  * (See save_args())
121  */
122 static inline unsigned long
123 fixup_bp_irq_link(unsigned long bp, unsigned long *stack,
124 		  unsigned long *irq_stack, unsigned long *irq_stack_end)
125 {
126 #ifdef CONFIG_FRAME_POINTER
127 	struct stack_frame *frame = (struct stack_frame *)bp;
128 
129 	if (!in_irq_stack(stack, irq_stack, irq_stack_end))
130 		return (unsigned long)frame->next_frame;
131 #endif
132 	return bp;
133 }
134 
135 /*
136  * x86-64 can have up to three kernel stacks:
137  * process stack
138  * interrupt stack
139  * severe exception (double fault, nmi, stack fault, debug, mce) hardware stack
140  */
141 
142 void dump_trace(struct task_struct *task, struct pt_regs *regs,
143 		unsigned long *stack, unsigned long bp,
144 		const struct stacktrace_ops *ops, void *data)
145 {
146 	const unsigned cpu = get_cpu();
147 	unsigned long *irq_stack_end =
148 		(unsigned long *)per_cpu(irq_stack_ptr, cpu);
149 	unsigned used = 0;
150 	struct thread_info *tinfo;
151 	int graph = 0;
152 
153 	if (!task)
154 		task = current;
155 
156 	if (!stack) {
157 		unsigned long dummy;
158 		stack = &dummy;
159 		if (task && task != current)
160 			stack = (unsigned long *)task->thread.sp;
161 	}
162 
163 #ifdef CONFIG_FRAME_POINTER
164 	if (!bp) {
165 		if (task == current) {
166 			/* Grab bp right from our regs */
167 			get_bp(bp);
168 		} else {
169 			/* bp is the last reg pushed by switch_to */
170 			bp = *(unsigned long *) task->thread.sp;
171 		}
172 	}
173 #endif
174 
175 	/*
176 	 * Print function call entries in all stacks, starting at the
177 	 * current stack address. If the stacks consist of nested
178 	 * exceptions
179 	 */
180 	tinfo = task_thread_info(task);
181 	for (;;) {
182 		char *id;
183 		unsigned long *estack_end;
184 		estack_end = in_exception_stack(cpu, (unsigned long)stack,
185 						&used, &id);
186 
187 		if (estack_end) {
188 			if (ops->stack(data, id) < 0)
189 				break;
190 
191 			bp = ops->walk_stack(tinfo, stack, bp, ops,
192 					     data, estack_end, &graph);
193 			ops->stack(data, "<EOE>");
194 			/*
195 			 * We link to the next stack via the
196 			 * second-to-last pointer (index -2 to end) in the
197 			 * exception stack:
198 			 */
199 			stack = (unsigned long *) estack_end[-2];
200 			continue;
201 		}
202 		if (irq_stack_end) {
203 			unsigned long *irq_stack;
204 			irq_stack = irq_stack_end -
205 				(IRQ_STACK_SIZE - 64) / sizeof(*irq_stack);
206 
207 			if (in_irq_stack(stack, irq_stack, irq_stack_end)) {
208 				if (ops->stack(data, "IRQ") < 0)
209 					break;
210 				bp = print_context_stack(tinfo, stack, bp,
211 					ops, data, irq_stack_end, &graph);
212 				/*
213 				 * We link to the next stack (which would be
214 				 * the process stack normally) the last
215 				 * pointer (index -1 to end) in the IRQ stack:
216 				 */
217 				stack = (unsigned long *) (irq_stack_end[-1]);
218 				bp = fixup_bp_irq_link(bp, stack, irq_stack,
219 						       irq_stack_end);
220 				irq_stack_end = NULL;
221 				ops->stack(data, "EOI");
222 				continue;
223 			}
224 		}
225 		break;
226 	}
227 
228 	/*
229 	 * This handles the process stack:
230 	 */
231 	bp = print_context_stack(tinfo, stack, bp, ops, data, NULL, &graph);
232 	put_cpu();
233 }
234 EXPORT_SYMBOL(dump_trace);
235 
236 void
237 show_stack_log_lvl(struct task_struct *task, struct pt_regs *regs,
238 		   unsigned long *sp, unsigned long bp, char *log_lvl)
239 {
240 	unsigned long *irq_stack_end;
241 	unsigned long *irq_stack;
242 	unsigned long *stack;
243 	int cpu;
244 	int i;
245 
246 	preempt_disable();
247 	cpu = smp_processor_id();
248 
249 	irq_stack_end	= (unsigned long *)(per_cpu(irq_stack_ptr, cpu));
250 	irq_stack	= (unsigned long *)(per_cpu(irq_stack_ptr, cpu) - IRQ_STACK_SIZE);
251 
252 	/*
253 	 * Debugging aid: "show_stack(NULL, NULL);" prints the
254 	 * back trace for this cpu:
255 	 */
256 	if (sp == NULL) {
257 		if (task)
258 			sp = (unsigned long *)task->thread.sp;
259 		else
260 			sp = (unsigned long *)&sp;
261 	}
262 
263 	stack = sp;
264 	for (i = 0; i < kstack_depth_to_print; i++) {
265 		if (stack >= irq_stack && stack <= irq_stack_end) {
266 			if (stack == irq_stack_end) {
267 				stack = (unsigned long *) (irq_stack_end[-1]);
268 				printk(" <EOI> ");
269 			}
270 		} else {
271 		if (((long) stack & (THREAD_SIZE-1)) == 0)
272 			break;
273 		}
274 		if (i && ((i % STACKSLOTS_PER_LINE) == 0))
275 			printk("\n%s", log_lvl);
276 		printk(" %016lx", *stack++);
277 		touch_nmi_watchdog();
278 	}
279 	preempt_enable();
280 
281 	printk("\n");
282 	show_trace_log_lvl(task, regs, sp, bp, log_lvl);
283 }
284 
285 void show_registers(struct pt_regs *regs)
286 {
287 	int i;
288 	unsigned long sp;
289 	const int cpu = smp_processor_id();
290 	struct task_struct *cur = current;
291 
292 	sp = regs->sp;
293 	printk("CPU %d ", cpu);
294 	__show_regs(regs, 1);
295 	printk("Process %s (pid: %d, threadinfo %p, task %p)\n",
296 		cur->comm, cur->pid, task_thread_info(cur), cur);
297 
298 	/*
299 	 * When in-kernel, we also print out the stack and code at the
300 	 * time of the fault..
301 	 */
302 	if (!user_mode(regs)) {
303 		unsigned int code_prologue = code_bytes * 43 / 64;
304 		unsigned int code_len = code_bytes;
305 		unsigned char c;
306 		u8 *ip;
307 
308 		printk(KERN_EMERG "Stack:\n");
309 		show_stack_log_lvl(NULL, regs, (unsigned long *)sp,
310 				regs->bp, KERN_EMERG);
311 
312 		printk(KERN_EMERG "Code: ");
313 
314 		ip = (u8 *)regs->ip - code_prologue;
315 		if (ip < (u8 *)PAGE_OFFSET || probe_kernel_address(ip, c)) {
316 			/* try starting at IP */
317 			ip = (u8 *)regs->ip;
318 			code_len = code_len - code_prologue + 1;
319 		}
320 		for (i = 0; i < code_len; i++, ip++) {
321 			if (ip < (u8 *)PAGE_OFFSET ||
322 					probe_kernel_address(ip, c)) {
323 				printk(" Bad RIP value.");
324 				break;
325 			}
326 			if (ip == (u8 *)regs->ip)
327 				printk("<%02x> ", c);
328 			else
329 				printk("%02x ", c);
330 		}
331 	}
332 	printk("\n");
333 }
334 
335 int is_valid_bugaddr(unsigned long ip)
336 {
337 	unsigned short ud2;
338 
339 	if (__copy_from_user(&ud2, (const void __user *) ip, sizeof(ud2)))
340 		return 0;
341 
342 	return ud2 == 0x0b0f;
343 }
344