blob: 1d6698b545279e5232fbb2742bf725dbca723724 [file] [log] [blame]
Neil Horman878719e2008-10-23 10:40:06 -04001/*
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>
Ingo Molnarb17b0152017-02-08 18:51:35 +010013#include <linux/sched/debug.h>
Ingo Molnar68db0cf2017-02-08 18:51:37 +010014#include <linux/sched/task_stack.h>
Steven Rostedt712406a2009-02-09 10:54:03 -080015#include <linux/ftrace.h>
Neil Horman878719e2008-10-23 10:40:06 -040016#include <linux/kexec.h>
17#include <linux/bug.h>
18#include <linux/nmi.h>
19#include <linux/sysfs.h>
20
Thomas Gleixner92a0f812017-12-20 18:51:31 +010021#include <asm/cpu_entry_area.h>
Neil Horman878719e2008-10-23 10:40:06 -040022#include <asm/stacktrace.h>
Josh Poimboeufe18bccc2016-09-16 14:18:16 -050023#include <asm/unwind.h>
Neil Horman878719e2008-10-23 10:40:06 -040024
Borislav Petkov5d12f042018-04-17 18:11:16 +020025#define OPCODE_BUFSIZE 64
26
Neil Horman878719e2008-10-23 10:40:06 -040027int panic_on_unrecovered_nmi;
Kurt Garloff5211a242009-06-24 14:32:11 -070028int panic_on_io_nmi;
Neil Horman878719e2008-10-23 10:40:06 -040029static int die_counter;
30
Josh Poimboeufcb76c932016-09-14 21:07:42 -050031bool in_task_stack(unsigned long *stack, struct task_struct *task,
32 struct stack_info *info)
33{
34 unsigned long *begin = task_stack_page(task);
35 unsigned long *end = task_stack_page(task) + THREAD_SIZE;
36
37 if (stack < begin || stack >= end)
38 return false;
39
40 info->type = STACK_TYPE_TASK;
41 info->begin = begin;
42 info->end = end;
43 info->next_sp = NULL;
44
45 return true;
46}
47
Dave Hansen4fe2d8b2017-12-04 17:25:07 -080048bool in_entry_stack(unsigned long *stack, struct stack_info *info)
Andy Lutomirski33a2f1a2017-12-04 15:07:13 +010049{
Dave Hansen4fe2d8b2017-12-04 17:25:07 -080050 struct entry_stack *ss = cpu_entry_stack(smp_processor_id());
Andy Lutomirski33a2f1a2017-12-04 15:07:13 +010051
Andy Lutomirski0f9a4812017-12-04 15:07:28 +010052 void *begin = ss;
53 void *end = ss + 1;
Andy Lutomirski33a2f1a2017-12-04 15:07:13 +010054
55 if ((void *)stack < begin || (void *)stack >= end)
56 return false;
57
Dave Hansen4fe2d8b2017-12-04 17:25:07 -080058 info->type = STACK_TYPE_ENTRY;
Andy Lutomirski33a2f1a2017-12-04 15:07:13 +010059 info->begin = begin;
60 info->end = end;
61 info->next_sp = NULL;
62
63 return true;
64}
65
Adrien Schildknecht1fc7f612015-02-20 03:34:21 +010066static void printk_stack_address(unsigned long address, int reliable,
Josh Poimboeufd438f5f2016-08-24 11:50:16 -050067 char *log_lvl)
Neil Horman878719e2008-10-23 10:40:06 -040068{
Josh Poimboeufd438f5f2016-08-24 11:50:16 -050069 touch_nmi_watchdog();
Josh Poimboeufbb5e5ce2016-10-25 09:51:12 -050070 printk("%s %s%pB\n", log_lvl, reliable ? "" : "? ", (void *)address);
Jiri Slaby5f01c982013-10-25 15:06:58 +020071}
72
Borislav Petkovf0a1d7c2018-04-17 18:11:18 +020073static void show_opcodes(u8 *rip)
74{
Borislav Petkov9e4a90f2018-04-17 18:11:19 +020075 unsigned int code_prologue = OPCODE_BUFSIZE * 2 / 3;
76 u8 opcodes[OPCODE_BUFSIZE];
Borislav Petkovf0a1d7c2018-04-17 18:11:18 +020077 u8 *ip;
78 int i;
79
80 printk(KERN_DEFAULT "Code: ");
81
82 ip = (u8 *)rip - code_prologue;
Borislav Petkov9e4a90f2018-04-17 18:11:19 +020083 if (probe_kernel_read(opcodes, ip, OPCODE_BUFSIZE)) {
84 pr_cont("Bad RIP value.\n");
85 return;
Borislav Petkovf0a1d7c2018-04-17 18:11:18 +020086 }
Borislav Petkov9e4a90f2018-04-17 18:11:19 +020087
88 for (i = 0; i < OPCODE_BUFSIZE; i++, ip++) {
89 if (ip == rip)
90 pr_cont("<%02x> ", opcodes[i]);
Borislav Petkovf0a1d7c2018-04-17 18:11:18 +020091 else
Borislav Petkov9e4a90f2018-04-17 18:11:19 +020092 pr_cont("%02x ", opcodes[i]);
Borislav Petkovf0a1d7c2018-04-17 18:11:18 +020093 }
94 pr_cont("\n");
95}
96
Josh Poimboeufb02fcf92017-12-04 15:07:09 +010097void show_iret_regs(struct pt_regs *regs)
98{
99 printk(KERN_DEFAULT "RIP: %04x:%pS\n", (int)regs->cs, (void *)regs->ip);
100 printk(KERN_DEFAULT "RSP: %04x:%016lx EFLAGS: %08lx", (int)regs->ss,
101 regs->sp, regs->flags);
102}
103
Josh Poimboeufa9cdbe72017-12-31 10:18:06 -0600104static void show_regs_if_on_stack(struct stack_info *info, struct pt_regs *regs,
105 bool partial)
Josh Poimboeufb02fcf92017-12-04 15:07:09 +0100106{
Josh Poimboeufa9cdbe72017-12-31 10:18:06 -0600107 /*
108 * These on_stack() checks aren't strictly necessary: the unwind code
109 * has already validated the 'regs' pointer. The checks are done for
110 * ordering reasons: if the registers are on the next stack, we don't
111 * want to print them out yet. Otherwise they'll be shown as part of
112 * the wrong stack. Later, when show_trace_log_lvl() switches to the
113 * next stack, this function will be called again with the same regs so
114 * they can be printed in the right context.
115 */
116 if (!partial && on_stack(info, regs, sizeof(*regs))) {
Josh Poimboeufb02fcf92017-12-04 15:07:09 +0100117 __show_regs(regs, 0);
Josh Poimboeufa9cdbe72017-12-31 10:18:06 -0600118
119 } else if (partial && on_stack(info, (void *)regs + IRET_FRAME_OFFSET,
120 IRET_FRAME_SIZE)) {
Josh Poimboeufb02fcf92017-12-04 15:07:09 +0100121 /*
122 * When an interrupt or exception occurs in entry code, the
123 * full pt_regs might not have been saved yet. In that case
124 * just print the iret frame.
125 */
126 show_iret_regs(regs);
127 }
128}
129
Josh Poimboeufe18bccc2016-09-16 14:18:16 -0500130void show_trace_log_lvl(struct task_struct *task, struct pt_regs *regs,
131 unsigned long *stack, char *log_lvl)
Neil Horman878719e2008-10-23 10:40:06 -0400132{
Josh Poimboeufe18bccc2016-09-16 14:18:16 -0500133 struct unwind_state state;
134 struct stack_info stack_info = {0};
135 unsigned long visit_mask = 0;
136 int graph_idx = 0;
Arnd Bergmannebfc1502018-02-02 15:56:17 +0100137 bool partial = false;
Neil Horman878719e2008-10-23 10:40:06 -0400138
Neil Horman878719e2008-10-23 10:40:06 -0400139 printk("%sCall Trace:\n", log_lvl);
Josh Poimboeufe18bccc2016-09-16 14:18:16 -0500140
141 unwind_start(&state, task, regs, stack);
Josh Poimboeuff4474c92016-11-17 00:04:58 -0600142 stack = stack ? : get_stack_pointer(task, regs);
Josh Poimboeuf3ffdeb12017-12-31 10:18:07 -0600143 regs = unwind_get_entry_regs(&state, &partial);
Josh Poimboeufe18bccc2016-09-16 14:18:16 -0500144
145 /*
146 * Iterate through the stacks, starting with the current stack pointer.
147 * Each stack has a pointer to the next one.
148 *
149 * x86-64 can have several stacks:
150 * - task stack
151 * - interrupt stack
152 * - HW exception stacks (double fault, nmi, debug, mce)
Dave Hansen4fe2d8b2017-12-04 17:25:07 -0800153 * - entry stack
Josh Poimboeufe18bccc2016-09-16 14:18:16 -0500154 *
Andy Lutomirski6e60e582017-12-04 15:07:18 +0100155 * x86-32 can have up to four stacks:
Josh Poimboeufe18bccc2016-09-16 14:18:16 -0500156 * - task stack
157 * - softirq stack
158 * - hardirq stack
Dave Hansen4fe2d8b2017-12-04 17:25:07 -0800159 * - entry stack
Josh Poimboeufe18bccc2016-09-16 14:18:16 -0500160 */
Josh Poimboeuf3ffdeb12017-12-31 10:18:07 -0600161 for ( ; stack; stack = PTR_ALIGN(stack_info.next_sp, sizeof(long))) {
Josh Poimboeuf3d02a9c2016-11-18 11:46:23 -0600162 const char *stack_name;
Josh Poimboeufe18bccc2016-09-16 14:18:16 -0500163
Andy Lutomirski6e60e582017-12-04 15:07:18 +0100164 if (get_stack_info(stack, task, &stack_info, &visit_mask)) {
165 /*
166 * We weren't on a valid stack. It's possible that
167 * we overflowed a valid stack into a guard page.
168 * See if the next page up is valid so that we can
169 * generate some kind of backtrace if this happens.
170 */
171 stack = (unsigned long *)PAGE_ALIGN((unsigned long)stack);
172 if (get_stack_info(stack, task, &stack_info, &visit_mask))
173 break;
174 }
Josh Poimboeufe18bccc2016-09-16 14:18:16 -0500175
Josh Poimboeuf3d02a9c2016-11-18 11:46:23 -0600176 stack_name = stack_type_name(stack_info.type);
177 if (stack_name)
178 printk("%s <%s>\n", log_lvl, stack_name);
Josh Poimboeufe18bccc2016-09-16 14:18:16 -0500179
Josh Poimboeufb02fcf92017-12-04 15:07:09 +0100180 if (regs)
Josh Poimboeufa9cdbe72017-12-31 10:18:06 -0600181 show_regs_if_on_stack(&stack_info, regs, partial);
Josh Poimboeufb0529be2017-07-11 10:33:40 -0500182
Josh Poimboeufe18bccc2016-09-16 14:18:16 -0500183 /*
184 * Scan the stack, printing any text addresses we find. At the
185 * same time, follow proper stack frames with the unwinder.
186 *
187 * Addresses found during the scan which are not reported by
188 * the unwinder are considered to be additional clues which are
189 * sometimes useful for debugging and are prefixed with '?'.
190 * This also serves as a failsafe option in case the unwinder
191 * goes off in the weeds.
192 */
193 for (; stack < stack_info.end; stack++) {
194 unsigned long real_addr;
195 int reliable = 0;
Josh Poimboeuf91e08ab2016-11-17 09:57:24 -0600196 unsigned long addr = READ_ONCE_NOCHECK(*stack);
Josh Poimboeufe18bccc2016-09-16 14:18:16 -0500197 unsigned long *ret_addr_p =
198 unwind_get_return_address_ptr(&state);
199
200 if (!__kernel_text_address(addr))
201 continue;
202
Josh Poimboeuf3b3fa112016-10-20 11:34:43 -0500203 /*
204 * Don't print regs->ip again if it was already printed
Josh Poimboeufa9cdbe72017-12-31 10:18:06 -0600205 * by show_regs_if_on_stack().
Josh Poimboeuf3b3fa112016-10-20 11:34:43 -0500206 */
Josh Poimboeufb0529be2017-07-11 10:33:40 -0500207 if (regs && stack == &regs->ip)
208 goto next;
Josh Poimboeuf3b3fa112016-10-20 11:34:43 -0500209
Josh Poimboeufe18bccc2016-09-16 14:18:16 -0500210 if (stack == ret_addr_p)
211 reliable = 1;
212
213 /*
214 * When function graph tracing is enabled for a
215 * function, its return address on the stack is
216 * replaced with the address of an ftrace handler
217 * (return_to_handler). In that case, before printing
218 * the "real" address, we want to print the handler
219 * address as an "unreliable" hint that function graph
220 * tracing was involved.
221 */
222 real_addr = ftrace_graph_ret_addr(task, &graph_idx,
223 addr, stack);
224 if (real_addr != addr)
225 printk_stack_address(addr, 0, log_lvl);
226 printk_stack_address(real_addr, reliable, log_lvl);
227
228 if (!reliable)
229 continue;
230
Josh Poimboeufb0529be2017-07-11 10:33:40 -0500231next:
Josh Poimboeufe18bccc2016-09-16 14:18:16 -0500232 /*
233 * Get the next frame from the unwinder. No need to
234 * check for an error: if anything goes wrong, the rest
235 * of the addresses will just be printed as unreliable.
236 */
237 unwind_next_frame(&state);
Josh Poimboeuf3b3fa112016-10-20 11:34:43 -0500238
239 /* if the frame has entry regs, print them */
Josh Poimboeufa9cdbe72017-12-31 10:18:06 -0600240 regs = unwind_get_entry_regs(&state, &partial);
Josh Poimboeufb02fcf92017-12-04 15:07:09 +0100241 if (regs)
Josh Poimboeufa9cdbe72017-12-31 10:18:06 -0600242 show_regs_if_on_stack(&stack_info, regs, partial);
Josh Poimboeufe18bccc2016-09-16 14:18:16 -0500243 }
244
Josh Poimboeuf3d02a9c2016-11-18 11:46:23 -0600245 if (stack_name)
246 printk("%s </%s>\n", log_lvl, stack_name);
Josh Poimboeufe18bccc2016-09-16 14:18:16 -0500247 }
Neil Horman878719e2008-10-23 10:40:06 -0400248}
249
Neil Horman878719e2008-10-23 10:40:06 -0400250void show_stack(struct task_struct *task, unsigned long *sp)
251{
Josh Poimboeuf81539162016-09-16 08:05:20 -0500252 task = task ? : current;
253
Tejun Heoa77f2a42013-04-30 15:27:09 -0700254 /*
255 * Stack frames below this one aren't interesting. Don't show them
256 * if we're printing for %current.
257 */
Josh Poimboeufe18bccc2016-09-16 14:18:16 -0500258 if (!sp && task == current)
Josh Poimboeuf4b8afaf2016-08-24 11:50:17 -0500259 sp = get_stack_pointer(current, NULL);
Tejun Heoa77f2a42013-04-30 15:27:09 -0700260
Josh Poimboeuf0ee1dd92016-10-25 09:51:13 -0500261 show_trace_log_lvl(task, NULL, sp, KERN_DEFAULT);
Neil Horman878719e2008-10-23 10:40:06 -0400262}
263
Borislav Petkov81c29492016-07-05 00:31:27 +0200264void show_stack_regs(struct pt_regs *regs)
265{
Josh Poimboeuf0ee1dd92016-10-25 09:51:13 -0500266 show_trace_log_lvl(current, regs, NULL, KERN_DEFAULT);
Borislav Petkov81c29492016-07-05 00:31:27 +0200267}
268
Thomas Gleixneredc35bd2009-12-03 12:38:57 +0100269static arch_spinlock_t die_lock = __ARCH_SPIN_LOCK_UNLOCKED;
Neil Horman878719e2008-10-23 10:40:06 -0400270static int die_owner = -1;
271static unsigned int die_nest_count;
272
Masami Hiramatsu93266382014-04-17 17:18:14 +0900273unsigned long oops_begin(void)
Neil Horman878719e2008-10-23 10:40:06 -0400274{
275 int cpu;
276 unsigned long flags;
277
Neil Horman878719e2008-10-23 10:40:06 -0400278 oops_enter();
279
280 /* racy, but better than risking deadlock. */
281 raw_local_irq_save(flags);
282 cpu = smp_processor_id();
Thomas Gleixner0199c4e2009-12-02 20:01:25 +0100283 if (!arch_spin_trylock(&die_lock)) {
Neil Horman878719e2008-10-23 10:40:06 -0400284 if (cpu == die_owner)
285 /* nested oops. should stop eventually */;
286 else
Thomas Gleixner0199c4e2009-12-02 20:01:25 +0100287 arch_spin_lock(&die_lock);
Neil Horman878719e2008-10-23 10:40:06 -0400288 }
289 die_nest_count++;
290 die_owner = cpu;
291 console_verbose();
292 bust_spinlocks(1);
293 return flags;
294}
Masami Hiramatsu93266382014-04-17 17:18:14 +0900295NOKPROBE_SYMBOL(oops_begin);
Neil Horman878719e2008-10-23 10:40:06 -0400296
Andy Lutomirski2deb4be2016-07-14 13:22:55 -0700297void __noreturn rewind_stack_do_exit(int signr);
298
Masami Hiramatsu93266382014-04-17 17:18:14 +0900299void oops_end(unsigned long flags, struct pt_regs *regs, int signr)
Neil Horman878719e2008-10-23 10:40:06 -0400300{
301 if (regs && kexec_should_crash(current))
302 crash_kexec(regs);
303
304 bust_spinlocks(0);
305 die_owner = -1;
Rusty Russell373d4d02013-01-21 17:17:39 +1030306 add_taint(TAINT_DIE, LOCKDEP_NOW_UNRELIABLE);
Neil Horman878719e2008-10-23 10:40:06 -0400307 die_nest_count--;
308 if (!die_nest_count)
309 /* Nest count reaches zero, release the lock. */
Thomas Gleixner0199c4e2009-12-02 20:01:25 +0100310 arch_spin_unlock(&die_lock);
Neil Horman878719e2008-10-23 10:40:06 -0400311 raw_local_irq_restore(flags);
312 oops_exit();
313
314 if (!signr)
315 return;
316 if (in_interrupt())
317 panic("Fatal exception in interrupt");
318 if (panic_on_oops)
319 panic("Fatal exception");
Andy Lutomirski2deb4be2016-07-14 13:22:55 -0700320
321 /*
322 * We're not going to return, but we might be on an IST stack or
323 * have very little stack space left. Rewind the stack and kill
324 * the task.
325 */
326 rewind_stack_do_exit(signr);
Neil Horman878719e2008-10-23 10:40:06 -0400327}
Masami Hiramatsu93266382014-04-17 17:18:14 +0900328NOKPROBE_SYMBOL(oops_end);
Neil Horman878719e2008-10-23 10:40:06 -0400329
Masami Hiramatsu93266382014-04-17 17:18:14 +0900330int __die(const char *str, struct pt_regs *regs, long err)
Neil Horman878719e2008-10-23 10:40:06 -0400331{
332#ifdef CONFIG_X86_32
333 unsigned short ss;
334 unsigned long sp;
335#endif
Prarit Bhargavab0f4c4b2012-01-26 08:55:34 -0500336 printk(KERN_DEFAULT
Vlastimil Babka5f26d762017-12-19 22:33:46 +0100337 "%s: %04lx [#%d]%s%s%s%s%s\n", str, err & 0xffff, ++die_counter,
Rasmus Villemoes8fad7ec2016-03-26 21:40:16 +0100338 IS_ENABLED(CONFIG_PREEMPT) ? " PREEMPT" : "",
339 IS_ENABLED(CONFIG_SMP) ? " SMP" : "",
340 debug_pagealloc_enabled() ? " DEBUG_PAGEALLOC" : "",
Vlastimil Babka5f26d762017-12-19 22:33:46 +0100341 IS_ENABLED(CONFIG_KASAN) ? " KASAN" : "",
342 IS_ENABLED(CONFIG_PAGE_TABLE_ISOLATION) ?
343 (boot_cpu_has(X86_FEATURE_PTI) ? " PTI" : " NOPTI") : "");
Rasmus Villemoes8fad7ec2016-03-26 21:40:16 +0100344
Neil Horman878719e2008-10-23 10:40:06 -0400345 if (notify_die(DIE_OOPS, str, regs, err,
Srikar Dronamraju51e7dc72012-03-12 14:55:55 +0530346 current->thread.trap_nr, SIGSEGV) == NOTIFY_STOP)
Neil Horman878719e2008-10-23 10:40:06 -0400347 return 1;
348
Jan Beulich0fa0e2f2012-06-18 11:40:04 +0100349 print_modules();
Jan Beulich57da8b92012-05-09 08:47:37 +0100350 show_regs(regs);
Neil Horman878719e2008-10-23 10:40:06 -0400351#ifdef CONFIG_X86_32
Andy Lutomirskif39b6f02015-03-18 18:33:33 -0700352 if (user_mode(regs)) {
Neil Horman878719e2008-10-23 10:40:06 -0400353 sp = regs->sp;
Andy Lutomirski99504812017-07-28 06:00:32 -0700354 ss = regs->ss;
H. Peter Anvina343c752009-10-12 14:11:09 -0700355 } else {
356 sp = kernel_stack_pointer(regs);
357 savesegment(ss, ss);
Neil Horman878719e2008-10-23 10:40:06 -0400358 }
Josh Poimboeufbb5e5ce2016-10-25 09:51:12 -0500359 printk(KERN_EMERG "EIP: %pS SS:ESP: %04x:%08lx\n",
360 (void *)regs->ip, ss, sp);
Neil Horman878719e2008-10-23 10:40:06 -0400361#else
362 /* Executive summary in case the oops scrolled away */
Josh Poimboeufbb5e5ce2016-10-25 09:51:12 -0500363 printk(KERN_ALERT "RIP: %pS RSP: %016lx\n", (void *)regs->ip, regs->sp);
Neil Horman878719e2008-10-23 10:40:06 -0400364#endif
365 return 0;
366}
Masami Hiramatsu93266382014-04-17 17:18:14 +0900367NOKPROBE_SYMBOL(__die);
Neil Horman878719e2008-10-23 10:40:06 -0400368
369/*
370 * This is gone through when something in the kernel has done something bad
371 * and is about to be terminated:
372 */
373void die(const char *str, struct pt_regs *regs, long err)
374{
375 unsigned long flags = oops_begin();
376 int sig = SIGSEGV;
377
Neil Horman878719e2008-10-23 10:40:06 -0400378 if (__die(str, regs, err))
379 sig = 0;
380 oops_end(flags, regs, sig);
381}
382
Borislav Petkov16d1cb02018-03-06 10:49:14 +0100383void show_regs(struct pt_regs *regs)
384{
385 bool all = true;
Borislav Petkov16d1cb02018-03-06 10:49:14 +0100386
387 show_regs_print_info(KERN_DEFAULT);
388
389 if (IS_ENABLED(CONFIG_X86_32))
390 all = !user_mode(regs);
391
392 __show_regs(regs, all);
393
394 /*
395 * When in-kernel, we also print out the stack and code at the
396 * time of the fault..
397 */
398 if (!user_mode(regs)) {
Borislav Petkov16d1cb02018-03-06 10:49:14 +0100399 show_trace_log_lvl(current, regs, NULL, KERN_DEFAULT);
Borislav Petkov9e4a90f2018-04-17 18:11:19 +0200400
401 if (regs->ip < PAGE_OFFSET)
402 printk(KERN_DEFAULT "Code: Bad RIP value.\n");
403 else
404 show_opcodes((u8 *)regs->ip);
Borislav Petkov16d1cb02018-03-06 10:49:14 +0100405 }
Borislav Petkov16d1cb02018-03-06 10:49:14 +0100406}