xref: /openbmc/linux/arch/x86/kernel/dumpstack.c (revision 2359ccdd)
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/utsname.h>
9 #include <linux/hardirq.h>
10 #include <linux/kdebug.h>
11 #include <linux/module.h>
12 #include <linux/ptrace.h>
13 #include <linux/sched/debug.h>
14 #include <linux/sched/task_stack.h>
15 #include <linux/ftrace.h>
16 #include <linux/kexec.h>
17 #include <linux/bug.h>
18 #include <linux/nmi.h>
19 #include <linux/sysfs.h>
20 
21 #include <asm/cpu_entry_area.h>
22 #include <asm/stacktrace.h>
23 #include <asm/unwind.h>
24 
25 int panic_on_unrecovered_nmi;
26 int panic_on_io_nmi;
27 static unsigned int code_bytes = 64;
28 static int die_counter;
29 
30 bool in_task_stack(unsigned long *stack, struct task_struct *task,
31 		   struct stack_info *info)
32 {
33 	unsigned long *begin = task_stack_page(task);
34 	unsigned long *end   = task_stack_page(task) + THREAD_SIZE;
35 
36 	if (stack < begin || stack >= end)
37 		return false;
38 
39 	info->type	= STACK_TYPE_TASK;
40 	info->begin	= begin;
41 	info->end	= end;
42 	info->next_sp	= NULL;
43 
44 	return true;
45 }
46 
47 bool in_entry_stack(unsigned long *stack, struct stack_info *info)
48 {
49 	struct entry_stack *ss = cpu_entry_stack(smp_processor_id());
50 
51 	void *begin = ss;
52 	void *end = ss + 1;
53 
54 	if ((void *)stack < begin || (void *)stack >= end)
55 		return false;
56 
57 	info->type	= STACK_TYPE_ENTRY;
58 	info->begin	= begin;
59 	info->end	= end;
60 	info->next_sp	= NULL;
61 
62 	return true;
63 }
64 
65 static void printk_stack_address(unsigned long address, int reliable,
66 				 char *log_lvl)
67 {
68 	touch_nmi_watchdog();
69 	printk("%s %s%pB\n", log_lvl, reliable ? "" : "? ", (void *)address);
70 }
71 
72 void show_iret_regs(struct pt_regs *regs)
73 {
74 	printk(KERN_DEFAULT "RIP: %04x:%pS\n", (int)regs->cs, (void *)regs->ip);
75 	printk(KERN_DEFAULT "RSP: %04x:%016lx EFLAGS: %08lx", (int)regs->ss,
76 		regs->sp, regs->flags);
77 }
78 
79 static void show_regs_if_on_stack(struct stack_info *info, struct pt_regs *regs,
80 				  bool partial)
81 {
82 	/*
83 	 * These on_stack() checks aren't strictly necessary: the unwind code
84 	 * has already validated the 'regs' pointer.  The checks are done for
85 	 * ordering reasons: if the registers are on the next stack, we don't
86 	 * want to print them out yet.  Otherwise they'll be shown as part of
87 	 * the wrong stack.  Later, when show_trace_log_lvl() switches to the
88 	 * next stack, this function will be called again with the same regs so
89 	 * they can be printed in the right context.
90 	 */
91 	if (!partial && on_stack(info, regs, sizeof(*regs))) {
92 		__show_regs(regs, 0);
93 
94 	} else if (partial && on_stack(info, (void *)regs + IRET_FRAME_OFFSET,
95 				       IRET_FRAME_SIZE)) {
96 		/*
97 		 * When an interrupt or exception occurs in entry code, the
98 		 * full pt_regs might not have been saved yet.  In that case
99 		 * just print the iret frame.
100 		 */
101 		show_iret_regs(regs);
102 	}
103 }
104 
105 void show_trace_log_lvl(struct task_struct *task, struct pt_regs *regs,
106 			unsigned long *stack, char *log_lvl)
107 {
108 	struct unwind_state state;
109 	struct stack_info stack_info = {0};
110 	unsigned long visit_mask = 0;
111 	int graph_idx = 0;
112 	bool partial = false;
113 
114 	printk("%sCall Trace:\n", log_lvl);
115 
116 	unwind_start(&state, task, regs, stack);
117 	stack = stack ? : get_stack_pointer(task, regs);
118 	regs = unwind_get_entry_regs(&state, &partial);
119 
120 	/*
121 	 * Iterate through the stacks, starting with the current stack pointer.
122 	 * Each stack has a pointer to the next one.
123 	 *
124 	 * x86-64 can have several stacks:
125 	 * - task stack
126 	 * - interrupt stack
127 	 * - HW exception stacks (double fault, nmi, debug, mce)
128 	 * - entry stack
129 	 *
130 	 * x86-32 can have up to four stacks:
131 	 * - task stack
132 	 * - softirq stack
133 	 * - hardirq stack
134 	 * - entry stack
135 	 */
136 	for ( ; stack; stack = PTR_ALIGN(stack_info.next_sp, sizeof(long))) {
137 		const char *stack_name;
138 
139 		if (get_stack_info(stack, task, &stack_info, &visit_mask)) {
140 			/*
141 			 * We weren't on a valid stack.  It's possible that
142 			 * we overflowed a valid stack into a guard page.
143 			 * See if the next page up is valid so that we can
144 			 * generate some kind of backtrace if this happens.
145 			 */
146 			stack = (unsigned long *)PAGE_ALIGN((unsigned long)stack);
147 			if (get_stack_info(stack, task, &stack_info, &visit_mask))
148 				break;
149 		}
150 
151 		stack_name = stack_type_name(stack_info.type);
152 		if (stack_name)
153 			printk("%s <%s>\n", log_lvl, stack_name);
154 
155 		if (regs)
156 			show_regs_if_on_stack(&stack_info, regs, partial);
157 
158 		/*
159 		 * Scan the stack, printing any text addresses we find.  At the
160 		 * same time, follow proper stack frames with the unwinder.
161 		 *
162 		 * Addresses found during the scan which are not reported by
163 		 * the unwinder are considered to be additional clues which are
164 		 * sometimes useful for debugging and are prefixed with '?'.
165 		 * This also serves as a failsafe option in case the unwinder
166 		 * goes off in the weeds.
167 		 */
168 		for (; stack < stack_info.end; stack++) {
169 			unsigned long real_addr;
170 			int reliable = 0;
171 			unsigned long addr = READ_ONCE_NOCHECK(*stack);
172 			unsigned long *ret_addr_p =
173 				unwind_get_return_address_ptr(&state);
174 
175 			if (!__kernel_text_address(addr))
176 				continue;
177 
178 			/*
179 			 * Don't print regs->ip again if it was already printed
180 			 * by show_regs_if_on_stack().
181 			 */
182 			if (regs && stack == &regs->ip)
183 				goto next;
184 
185 			if (stack == ret_addr_p)
186 				reliable = 1;
187 
188 			/*
189 			 * When function graph tracing is enabled for a
190 			 * function, its return address on the stack is
191 			 * replaced with the address of an ftrace handler
192 			 * (return_to_handler).  In that case, before printing
193 			 * the "real" address, we want to print the handler
194 			 * address as an "unreliable" hint that function graph
195 			 * tracing was involved.
196 			 */
197 			real_addr = ftrace_graph_ret_addr(task, &graph_idx,
198 							  addr, stack);
199 			if (real_addr != addr)
200 				printk_stack_address(addr, 0, log_lvl);
201 			printk_stack_address(real_addr, reliable, log_lvl);
202 
203 			if (!reliable)
204 				continue;
205 
206 next:
207 			/*
208 			 * Get the next frame from the unwinder.  No need to
209 			 * check for an error: if anything goes wrong, the rest
210 			 * of the addresses will just be printed as unreliable.
211 			 */
212 			unwind_next_frame(&state);
213 
214 			/* if the frame has entry regs, print them */
215 			regs = unwind_get_entry_regs(&state, &partial);
216 			if (regs)
217 				show_regs_if_on_stack(&stack_info, regs, partial);
218 		}
219 
220 		if (stack_name)
221 			printk("%s </%s>\n", log_lvl, stack_name);
222 	}
223 }
224 
225 void show_stack(struct task_struct *task, unsigned long *sp)
226 {
227 	task = task ? : current;
228 
229 	/*
230 	 * Stack frames below this one aren't interesting.  Don't show them
231 	 * if we're printing for %current.
232 	 */
233 	if (!sp && task == current)
234 		sp = get_stack_pointer(current, NULL);
235 
236 	show_trace_log_lvl(task, NULL, sp, KERN_DEFAULT);
237 }
238 
239 void show_stack_regs(struct pt_regs *regs)
240 {
241 	show_trace_log_lvl(current, regs, NULL, KERN_DEFAULT);
242 }
243 
244 static arch_spinlock_t die_lock = __ARCH_SPIN_LOCK_UNLOCKED;
245 static int die_owner = -1;
246 static unsigned int die_nest_count;
247 
248 unsigned long oops_begin(void)
249 {
250 	int cpu;
251 	unsigned long flags;
252 
253 	oops_enter();
254 
255 	/* racy, but better than risking deadlock. */
256 	raw_local_irq_save(flags);
257 	cpu = smp_processor_id();
258 	if (!arch_spin_trylock(&die_lock)) {
259 		if (cpu == die_owner)
260 			/* nested oops. should stop eventually */;
261 		else
262 			arch_spin_lock(&die_lock);
263 	}
264 	die_nest_count++;
265 	die_owner = cpu;
266 	console_verbose();
267 	bust_spinlocks(1);
268 	return flags;
269 }
270 EXPORT_SYMBOL_GPL(oops_begin);
271 NOKPROBE_SYMBOL(oops_begin);
272 
273 void __noreturn rewind_stack_do_exit(int signr);
274 
275 void oops_end(unsigned long flags, struct pt_regs *regs, int signr)
276 {
277 	if (regs && kexec_should_crash(current))
278 		crash_kexec(regs);
279 
280 	bust_spinlocks(0);
281 	die_owner = -1;
282 	add_taint(TAINT_DIE, LOCKDEP_NOW_UNRELIABLE);
283 	die_nest_count--;
284 	if (!die_nest_count)
285 		/* Nest count reaches zero, release the lock. */
286 		arch_spin_unlock(&die_lock);
287 	raw_local_irq_restore(flags);
288 	oops_exit();
289 
290 	if (!signr)
291 		return;
292 	if (in_interrupt())
293 		panic("Fatal exception in interrupt");
294 	if (panic_on_oops)
295 		panic("Fatal exception");
296 
297 	/*
298 	 * We're not going to return, but we might be on an IST stack or
299 	 * have very little stack space left.  Rewind the stack and kill
300 	 * the task.
301 	 */
302 	rewind_stack_do_exit(signr);
303 }
304 NOKPROBE_SYMBOL(oops_end);
305 
306 int __die(const char *str, struct pt_regs *regs, long err)
307 {
308 #ifdef CONFIG_X86_32
309 	unsigned short ss;
310 	unsigned long sp;
311 #endif
312 	printk(KERN_DEFAULT
313 	       "%s: %04lx [#%d]%s%s%s%s%s\n", str, err & 0xffff, ++die_counter,
314 	       IS_ENABLED(CONFIG_PREEMPT) ? " PREEMPT"         : "",
315 	       IS_ENABLED(CONFIG_SMP)     ? " SMP"             : "",
316 	       debug_pagealloc_enabled()  ? " DEBUG_PAGEALLOC" : "",
317 	       IS_ENABLED(CONFIG_KASAN)   ? " KASAN"           : "",
318 	       IS_ENABLED(CONFIG_PAGE_TABLE_ISOLATION) ?
319 	       (boot_cpu_has(X86_FEATURE_PTI) ? " PTI" : " NOPTI") : "");
320 
321 	if (notify_die(DIE_OOPS, str, regs, err,
322 			current->thread.trap_nr, SIGSEGV) == NOTIFY_STOP)
323 		return 1;
324 
325 	print_modules();
326 	show_regs(regs);
327 #ifdef CONFIG_X86_32
328 	if (user_mode(regs)) {
329 		sp = regs->sp;
330 		ss = regs->ss;
331 	} else {
332 		sp = kernel_stack_pointer(regs);
333 		savesegment(ss, ss);
334 	}
335 	printk(KERN_EMERG "EIP: %pS SS:ESP: %04x:%08lx\n",
336 	       (void *)regs->ip, ss, sp);
337 #else
338 	/* Executive summary in case the oops scrolled away */
339 	printk(KERN_ALERT "RIP: %pS RSP: %016lx\n", (void *)regs->ip, regs->sp);
340 #endif
341 	return 0;
342 }
343 NOKPROBE_SYMBOL(__die);
344 
345 /*
346  * This is gone through when something in the kernel has done something bad
347  * and is about to be terminated:
348  */
349 void die(const char *str, struct pt_regs *regs, long err)
350 {
351 	unsigned long flags = oops_begin();
352 	int sig = SIGSEGV;
353 
354 	if (__die(str, regs, err))
355 		sig = 0;
356 	oops_end(flags, regs, sig);
357 }
358 
359 static int __init code_bytes_setup(char *s)
360 {
361 	ssize_t ret;
362 	unsigned long val;
363 
364 	if (!s)
365 		return -EINVAL;
366 
367 	ret = kstrtoul(s, 0, &val);
368 	if (ret)
369 		return ret;
370 
371 	code_bytes = val;
372 	if (code_bytes > 8192)
373 		code_bytes = 8192;
374 
375 	return 1;
376 }
377 __setup("code_bytes=", code_bytes_setup);
378 
379 void show_regs(struct pt_regs *regs)
380 {
381 	bool all = true;
382 	int i;
383 
384 	show_regs_print_info(KERN_DEFAULT);
385 
386 	if (IS_ENABLED(CONFIG_X86_32))
387 		all = !user_mode(regs);
388 
389 	__show_regs(regs, all);
390 
391 	/*
392 	 * When in-kernel, we also print out the stack and code at the
393 	 * time of the fault..
394 	 */
395 	if (!user_mode(regs)) {
396 		unsigned int code_prologue = code_bytes * 43 / 64;
397 		unsigned int code_len = code_bytes;
398 		unsigned char c;
399 		u8 *ip;
400 
401 		show_trace_log_lvl(current, regs, NULL, KERN_DEFAULT);
402 
403 		printk(KERN_DEFAULT "Code: ");
404 
405 		ip = (u8 *)regs->ip - code_prologue;
406 		if (ip < (u8 *)PAGE_OFFSET || probe_kernel_address(ip, c)) {
407 			/* try starting at IP */
408 			ip = (u8 *)regs->ip;
409 			code_len = code_len - code_prologue + 1;
410 		}
411 		for (i = 0; i < code_len; i++, ip++) {
412 			if (ip < (u8 *)PAGE_OFFSET ||
413 					probe_kernel_address(ip, c)) {
414 				pr_cont(" Bad RIP value.");
415 				break;
416 			}
417 			if (ip == (u8 *)regs->ip)
418 				pr_cont("<%02x> ", c);
419 			else
420 				pr_cont("%02x ", c);
421 		}
422 	}
423 	pr_cont("\n");
424 }
425