xref: /openbmc/linux/arch/x86/kernel/dumpstack_64.c (revision d78c317f)
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 		[ STACKFAULT_STACK-1		]	= "#SS",
28 		[ MCE_STACK-1			]	= "#MC",
29 #if DEBUG_STKSZ > EXCEPTION_STKSZ
30 		[ N_EXCEPTION_STACKS ...
31 		  N_EXCEPTION_STACKS_END	]	= "#DB[?]"
32 #endif
33 };
34 
35 static unsigned long *in_exception_stack(unsigned cpu, unsigned long stack,
36 					 unsigned *usedp, char **idp)
37 {
38 	unsigned k;
39 
40 	/*
41 	 * Iterate over all exception stacks, and figure out whether
42 	 * 'stack' is in one of them:
43 	 */
44 	for (k = 0; k < N_EXCEPTION_STACKS; k++) {
45 		unsigned long end = per_cpu(orig_ist, cpu).ist[k];
46 		/*
47 		 * Is 'stack' above this exception frame's end?
48 		 * If yes then skip to the next frame.
49 		 */
50 		if (stack >= end)
51 			continue;
52 		/*
53 		 * Is 'stack' above this exception frame's start address?
54 		 * If yes then we found the right frame.
55 		 */
56 		if (stack >= end - EXCEPTION_STKSZ) {
57 			/*
58 			 * Make sure we only iterate through an exception
59 			 * stack once. If it comes up for the second time
60 			 * then there's something wrong going on - just
61 			 * break out and return NULL:
62 			 */
63 			if (*usedp & (1U << k))
64 				break;
65 			*usedp |= 1U << k;
66 			*idp = x86_stack_ids[k];
67 			return (unsigned long *)end;
68 		}
69 		/*
70 		 * If this is a debug stack, and if it has a larger size than
71 		 * the usual exception stacks, then 'stack' might still
72 		 * be within the lower portion of the debug stack:
73 		 */
74 #if DEBUG_STKSZ > EXCEPTION_STKSZ
75 		if (k == DEBUG_STACK - 1 && stack >= end - DEBUG_STKSZ) {
76 			unsigned j = N_EXCEPTION_STACKS - 1;
77 
78 			/*
79 			 * Black magic. A large debug stack is composed of
80 			 * multiple exception stack entries, which we
81 			 * iterate through now. Dont look:
82 			 */
83 			do {
84 				++j;
85 				end -= EXCEPTION_STKSZ;
86 				x86_stack_ids[j][4] = '1' +
87 						(j - N_EXCEPTION_STACKS);
88 			} while (stack < end - EXCEPTION_STKSZ);
89 			if (*usedp & (1U << j))
90 				break;
91 			*usedp |= 1U << j;
92 			*idp = x86_stack_ids[j];
93 			return (unsigned long *)end;
94 		}
95 #endif
96 	}
97 	return NULL;
98 }
99 
100 static inline int
101 in_irq_stack(unsigned long *stack, unsigned long *irq_stack,
102 	     unsigned long *irq_stack_end)
103 {
104 	return (stack >= irq_stack && stack < irq_stack_end);
105 }
106 
107 /*
108  * x86-64 can have up to three kernel stacks:
109  * process stack
110  * interrupt stack
111  * severe exception (double fault, nmi, stack fault, debug, mce) hardware stack
112  */
113 
114 void dump_trace(struct task_struct *task, struct pt_regs *regs,
115 		unsigned long *stack, unsigned long bp,
116 		const struct stacktrace_ops *ops, void *data)
117 {
118 	const unsigned cpu = get_cpu();
119 	unsigned long *irq_stack_end =
120 		(unsigned long *)per_cpu(irq_stack_ptr, cpu);
121 	unsigned used = 0;
122 	struct thread_info *tinfo;
123 	int graph = 0;
124 	unsigned long dummy;
125 
126 	if (!task)
127 		task = current;
128 
129 	if (!stack) {
130 		if (regs)
131 			stack = (unsigned long *)regs->sp;
132 		else if (task != current)
133 			stack = (unsigned long *)task->thread.sp;
134 		else
135 			stack = &dummy;
136 	}
137 
138 	if (!bp)
139 		bp = stack_frame(task, regs);
140 	/*
141 	 * Print function call entries in all stacks, starting at the
142 	 * current stack address. If the stacks consist of nested
143 	 * exceptions
144 	 */
145 	tinfo = task_thread_info(task);
146 	for (;;) {
147 		char *id;
148 		unsigned long *estack_end;
149 		estack_end = in_exception_stack(cpu, (unsigned long)stack,
150 						&used, &id);
151 
152 		if (estack_end) {
153 			if (ops->stack(data, id) < 0)
154 				break;
155 
156 			bp = ops->walk_stack(tinfo, stack, bp, ops,
157 					     data, estack_end, &graph);
158 			ops->stack(data, "<EOE>");
159 			/*
160 			 * We link to the next stack via the
161 			 * second-to-last pointer (index -2 to end) in the
162 			 * exception stack:
163 			 */
164 			stack = (unsigned long *) estack_end[-2];
165 			continue;
166 		}
167 		if (irq_stack_end) {
168 			unsigned long *irq_stack;
169 			irq_stack = irq_stack_end -
170 				(IRQ_STACK_SIZE - 64) / sizeof(*irq_stack);
171 
172 			if (in_irq_stack(stack, irq_stack, irq_stack_end)) {
173 				if (ops->stack(data, "IRQ") < 0)
174 					break;
175 				bp = ops->walk_stack(tinfo, stack, bp,
176 					ops, data, irq_stack_end, &graph);
177 				/*
178 				 * We link to the next stack (which would be
179 				 * the process stack normally) the last
180 				 * pointer (index -1 to end) in the IRQ stack:
181 				 */
182 				stack = (unsigned long *) (irq_stack_end[-1]);
183 				irq_stack_end = NULL;
184 				ops->stack(data, "EOI");
185 				continue;
186 			}
187 		}
188 		break;
189 	}
190 
191 	/*
192 	 * This handles the process stack:
193 	 */
194 	bp = ops->walk_stack(tinfo, stack, bp, ops, data, NULL, &graph);
195 	put_cpu();
196 }
197 EXPORT_SYMBOL(dump_trace);
198 
199 void
200 show_stack_log_lvl(struct task_struct *task, struct pt_regs *regs,
201 		   unsigned long *sp, unsigned long bp, char *log_lvl)
202 {
203 	unsigned long *irq_stack_end;
204 	unsigned long *irq_stack;
205 	unsigned long *stack;
206 	int cpu;
207 	int i;
208 
209 	preempt_disable();
210 	cpu = smp_processor_id();
211 
212 	irq_stack_end	= (unsigned long *)(per_cpu(irq_stack_ptr, cpu));
213 	irq_stack	= (unsigned long *)(per_cpu(irq_stack_ptr, cpu) - IRQ_STACK_SIZE);
214 
215 	/*
216 	 * Debugging aid: "show_stack(NULL, NULL);" prints the
217 	 * back trace for this cpu:
218 	 */
219 	if (sp == NULL) {
220 		if (task)
221 			sp = (unsigned long *)task->thread.sp;
222 		else
223 			sp = (unsigned long *)&sp;
224 	}
225 
226 	stack = sp;
227 	for (i = 0; i < kstack_depth_to_print; i++) {
228 		if (stack >= irq_stack && stack <= irq_stack_end) {
229 			if (stack == irq_stack_end) {
230 				stack = (unsigned long *) (irq_stack_end[-1]);
231 				printk(KERN_CONT " <EOI> ");
232 			}
233 		} else {
234 		if (((long) stack & (THREAD_SIZE-1)) == 0)
235 			break;
236 		}
237 		if (i && ((i % STACKSLOTS_PER_LINE) == 0))
238 			printk(KERN_CONT "\n");
239 		printk(KERN_CONT " %016lx", *stack++);
240 		touch_nmi_watchdog();
241 	}
242 	preempt_enable();
243 
244 	printk(KERN_CONT "\n");
245 	show_trace_log_lvl(task, regs, sp, bp, log_lvl);
246 }
247 
248 void show_registers(struct pt_regs *regs)
249 {
250 	int i;
251 	unsigned long sp;
252 	const int cpu = smp_processor_id();
253 	struct task_struct *cur = current;
254 
255 	sp = regs->sp;
256 	printk("CPU %d ", cpu);
257 	print_modules();
258 	__show_regs(regs, 1);
259 	printk("Process %s (pid: %d, threadinfo %p, task %p)\n",
260 		cur->comm, cur->pid, task_thread_info(cur), cur);
261 
262 	/*
263 	 * When in-kernel, we also print out the stack and code at the
264 	 * time of the fault..
265 	 */
266 	if (!user_mode(regs)) {
267 		unsigned int code_prologue = code_bytes * 43 / 64;
268 		unsigned int code_len = code_bytes;
269 		unsigned char c;
270 		u8 *ip;
271 
272 		printk(KERN_DEFAULT "Stack:\n");
273 		show_stack_log_lvl(NULL, regs, (unsigned long *)sp,
274 				   0, KERN_DEFAULT);
275 
276 		printk(KERN_DEFAULT "Code: ");
277 
278 		ip = (u8 *)regs->ip - code_prologue;
279 		if (ip < (u8 *)PAGE_OFFSET || probe_kernel_address(ip, c)) {
280 			/* try starting at IP */
281 			ip = (u8 *)regs->ip;
282 			code_len = code_len - code_prologue + 1;
283 		}
284 		for (i = 0; i < code_len; i++, ip++) {
285 			if (ip < (u8 *)PAGE_OFFSET ||
286 					probe_kernel_address(ip, c)) {
287 				printk(KERN_CONT " Bad RIP value.");
288 				break;
289 			}
290 			if (ip == (u8 *)regs->ip)
291 				printk(KERN_CONT "<%02x> ", c);
292 			else
293 				printk(KERN_CONT "%02x ", c);
294 		}
295 	}
296 	printk(KERN_CONT "\n");
297 }
298 
299 int is_valid_bugaddr(unsigned long ip)
300 {
301 	unsigned short ud2;
302 
303 	if (__copy_from_user(&ud2, (const void __user *) ip, sizeof(ud2)))
304 		return 0;
305 
306 	return ud2 == 0x0b0f;
307 }
308