Loading...
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
25int panic_on_unrecovered_nmi;
26int panic_on_io_nmi;
27static unsigned int code_bytes = 64;
28static int die_counter;
29
30bool 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
47bool 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
65static 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
72void 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
79static 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
105void 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 == ®s->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
206next:
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
225void 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
239void show_stack_regs(struct pt_regs *regs)
240{
241 show_trace_log_lvl(current, regs, NULL, KERN_DEFAULT);
242}
243
244static arch_spinlock_t die_lock = __ARCH_SPIN_LOCK_UNLOCKED;
245static int die_owner = -1;
246static unsigned int die_nest_count;
247
248unsigned 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}
270EXPORT_SYMBOL_GPL(oops_begin);
271NOKPROBE_SYMBOL(oops_begin);
272
273void __noreturn rewind_stack_do_exit(int signr);
274
275void 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}
304NOKPROBE_SYMBOL(oops_end);
305
306int __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}
343NOKPROBE_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 */
349void 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
359static 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
379void 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}
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#include <linux/kasan.h>
21
22#include <asm/cpu_entry_area.h>
23#include <asm/stacktrace.h>
24#include <asm/unwind.h>
25
26int panic_on_unrecovered_nmi;
27int panic_on_io_nmi;
28static int die_counter;
29
30static struct pt_regs exec_summary_regs;
31
32bool in_task_stack(unsigned long *stack, struct task_struct *task,
33 struct stack_info *info)
34{
35 unsigned long *begin = task_stack_page(task);
36 unsigned long *end = task_stack_page(task) + THREAD_SIZE;
37
38 if (stack < begin || stack >= end)
39 return false;
40
41 info->type = STACK_TYPE_TASK;
42 info->begin = begin;
43 info->end = end;
44 info->next_sp = NULL;
45
46 return true;
47}
48
49bool in_entry_stack(unsigned long *stack, struct stack_info *info)
50{
51 struct entry_stack *ss = cpu_entry_stack(smp_processor_id());
52
53 void *begin = ss;
54 void *end = ss + 1;
55
56 if ((void *)stack < begin || (void *)stack >= end)
57 return false;
58
59 info->type = STACK_TYPE_ENTRY;
60 info->begin = begin;
61 info->end = end;
62 info->next_sp = NULL;
63
64 return true;
65}
66
67static void printk_stack_address(unsigned long address, int reliable,
68 const char *log_lvl)
69{
70 touch_nmi_watchdog();
71 printk("%s %s%pB\n", log_lvl, reliable ? "" : "? ", (void *)address);
72}
73
74static int copy_code(struct pt_regs *regs, u8 *buf, unsigned long src,
75 unsigned int nbytes)
76{
77 if (!user_mode(regs))
78 return copy_from_kernel_nofault(buf, (u8 *)src, nbytes);
79
80 /*
81 * Make sure userspace isn't trying to trick us into dumping kernel
82 * memory by pointing the userspace instruction pointer at it.
83 */
84 if (__chk_range_not_ok(src, nbytes, TASK_SIZE_MAX))
85 return -EINVAL;
86
87 return copy_from_user_nmi(buf, (void __user *)src, nbytes);
88}
89
90/*
91 * There are a couple of reasons for the 2/3rd prologue, courtesy of Linus:
92 *
93 * In case where we don't have the exact kernel image (which, if we did, we can
94 * simply disassemble and navigate to the RIP), the purpose of the bigger
95 * prologue is to have more context and to be able to correlate the code from
96 * the different toolchains better.
97 *
98 * In addition, it helps in recreating the register allocation of the failing
99 * kernel and thus make sense of the register dump.
100 *
101 * What is more, the additional complication of a variable length insn arch like
102 * x86 warrants having longer byte sequence before rIP so that the disassembler
103 * can "sync" up properly and find instruction boundaries when decoding the
104 * opcode bytes.
105 *
106 * Thus, the 2/3rds prologue and 64 byte OPCODE_BUFSIZE is just a random
107 * guesstimate in attempt to achieve all of the above.
108 */
109void show_opcodes(struct pt_regs *regs, const char *loglvl)
110{
111#define PROLOGUE_SIZE 42
112#define EPILOGUE_SIZE 21
113#define OPCODE_BUFSIZE (PROLOGUE_SIZE + 1 + EPILOGUE_SIZE)
114 u8 opcodes[OPCODE_BUFSIZE];
115 unsigned long prologue = regs->ip - PROLOGUE_SIZE;
116
117 if (copy_code(regs, opcodes, prologue, sizeof(opcodes))) {
118 printk("%sCode: Bad RIP value.\n", loglvl);
119 } else {
120 printk("%sCode: %" __stringify(PROLOGUE_SIZE) "ph <%02x> %"
121 __stringify(EPILOGUE_SIZE) "ph\n", loglvl, opcodes,
122 opcodes[PROLOGUE_SIZE], opcodes + PROLOGUE_SIZE + 1);
123 }
124}
125
126void show_ip(struct pt_regs *regs, const char *loglvl)
127{
128#ifdef CONFIG_X86_32
129 printk("%sEIP: %pS\n", loglvl, (void *)regs->ip);
130#else
131 printk("%sRIP: %04x:%pS\n", loglvl, (int)regs->cs, (void *)regs->ip);
132#endif
133 show_opcodes(regs, loglvl);
134}
135
136void show_iret_regs(struct pt_regs *regs, const char *log_lvl)
137{
138 show_ip(regs, log_lvl);
139 printk("%sRSP: %04x:%016lx EFLAGS: %08lx", log_lvl, (int)regs->ss,
140 regs->sp, regs->flags);
141}
142
143static void show_regs_if_on_stack(struct stack_info *info, struct pt_regs *regs,
144 bool partial, const char *log_lvl)
145{
146 /*
147 * These on_stack() checks aren't strictly necessary: the unwind code
148 * has already validated the 'regs' pointer. The checks are done for
149 * ordering reasons: if the registers are on the next stack, we don't
150 * want to print them out yet. Otherwise they'll be shown as part of
151 * the wrong stack. Later, when show_trace_log_lvl() switches to the
152 * next stack, this function will be called again with the same regs so
153 * they can be printed in the right context.
154 */
155 if (!partial && on_stack(info, regs, sizeof(*regs))) {
156 __show_regs(regs, SHOW_REGS_SHORT, log_lvl);
157
158 } else if (partial && on_stack(info, (void *)regs + IRET_FRAME_OFFSET,
159 IRET_FRAME_SIZE)) {
160 /*
161 * When an interrupt or exception occurs in entry code, the
162 * full pt_regs might not have been saved yet. In that case
163 * just print the iret frame.
164 */
165 show_iret_regs(regs, log_lvl);
166 }
167}
168
169void show_trace_log_lvl(struct task_struct *task, struct pt_regs *regs,
170 unsigned long *stack, const char *log_lvl)
171{
172 struct unwind_state state;
173 struct stack_info stack_info = {0};
174 unsigned long visit_mask = 0;
175 int graph_idx = 0;
176 bool partial = false;
177
178 printk("%sCall Trace:\n", log_lvl);
179
180 unwind_start(&state, task, regs, stack);
181 stack = stack ? : get_stack_pointer(task, regs);
182 regs = unwind_get_entry_regs(&state, &partial);
183
184 /*
185 * Iterate through the stacks, starting with the current stack pointer.
186 * Each stack has a pointer to the next one.
187 *
188 * x86-64 can have several stacks:
189 * - task stack
190 * - interrupt stack
191 * - HW exception stacks (double fault, nmi, debug, mce)
192 * - entry stack
193 *
194 * x86-32 can have up to four stacks:
195 * - task stack
196 * - softirq stack
197 * - hardirq stack
198 * - entry stack
199 */
200 for ( ; stack; stack = PTR_ALIGN(stack_info.next_sp, sizeof(long))) {
201 const char *stack_name;
202
203 if (get_stack_info(stack, task, &stack_info, &visit_mask)) {
204 /*
205 * We weren't on a valid stack. It's possible that
206 * we overflowed a valid stack into a guard page.
207 * See if the next page up is valid so that we can
208 * generate some kind of backtrace if this happens.
209 */
210 stack = (unsigned long *)PAGE_ALIGN((unsigned long)stack);
211 if (get_stack_info(stack, task, &stack_info, &visit_mask))
212 break;
213 }
214
215 stack_name = stack_type_name(stack_info.type);
216 if (stack_name)
217 printk("%s <%s>\n", log_lvl, stack_name);
218
219 if (regs)
220 show_regs_if_on_stack(&stack_info, regs, partial, log_lvl);
221
222 /*
223 * Scan the stack, printing any text addresses we find. At the
224 * same time, follow proper stack frames with the unwinder.
225 *
226 * Addresses found during the scan which are not reported by
227 * the unwinder are considered to be additional clues which are
228 * sometimes useful for debugging and are prefixed with '?'.
229 * This also serves as a failsafe option in case the unwinder
230 * goes off in the weeds.
231 */
232 for (; stack < stack_info.end; stack++) {
233 unsigned long real_addr;
234 int reliable = 0;
235 unsigned long addr = READ_ONCE_NOCHECK(*stack);
236 unsigned long *ret_addr_p =
237 unwind_get_return_address_ptr(&state);
238
239 if (!__kernel_text_address(addr))
240 continue;
241
242 /*
243 * Don't print regs->ip again if it was already printed
244 * by show_regs_if_on_stack().
245 */
246 if (regs && stack == ®s->ip)
247 goto next;
248
249 if (stack == ret_addr_p)
250 reliable = 1;
251
252 /*
253 * When function graph tracing is enabled for a
254 * function, its return address on the stack is
255 * replaced with the address of an ftrace handler
256 * (return_to_handler). In that case, before printing
257 * the "real" address, we want to print the handler
258 * address as an "unreliable" hint that function graph
259 * tracing was involved.
260 */
261 real_addr = ftrace_graph_ret_addr(task, &graph_idx,
262 addr, stack);
263 if (real_addr != addr)
264 printk_stack_address(addr, 0, log_lvl);
265 printk_stack_address(real_addr, reliable, log_lvl);
266
267 if (!reliable)
268 continue;
269
270next:
271 /*
272 * Get the next frame from the unwinder. No need to
273 * check for an error: if anything goes wrong, the rest
274 * of the addresses will just be printed as unreliable.
275 */
276 unwind_next_frame(&state);
277
278 /* if the frame has entry regs, print them */
279 regs = unwind_get_entry_regs(&state, &partial);
280 if (regs)
281 show_regs_if_on_stack(&stack_info, regs, partial, log_lvl);
282 }
283
284 if (stack_name)
285 printk("%s </%s>\n", log_lvl, stack_name);
286 }
287}
288
289void show_stack(struct task_struct *task, unsigned long *sp,
290 const char *loglvl)
291{
292 task = task ? : current;
293
294 /*
295 * Stack frames below this one aren't interesting. Don't show them
296 * if we're printing for %current.
297 */
298 if (!sp && task == current)
299 sp = get_stack_pointer(current, NULL);
300
301 show_trace_log_lvl(task, NULL, sp, loglvl);
302}
303
304void show_stack_regs(struct pt_regs *regs)
305{
306 show_trace_log_lvl(current, regs, NULL, KERN_DEFAULT);
307}
308
309static arch_spinlock_t die_lock = __ARCH_SPIN_LOCK_UNLOCKED;
310static int die_owner = -1;
311static unsigned int die_nest_count;
312
313unsigned long oops_begin(void)
314{
315 int cpu;
316 unsigned long flags;
317
318 oops_enter();
319
320 /* racy, but better than risking deadlock. */
321 raw_local_irq_save(flags);
322 cpu = smp_processor_id();
323 if (!arch_spin_trylock(&die_lock)) {
324 if (cpu == die_owner)
325 /* nested oops. should stop eventually */;
326 else
327 arch_spin_lock(&die_lock);
328 }
329 die_nest_count++;
330 die_owner = cpu;
331 console_verbose();
332 bust_spinlocks(1);
333 return flags;
334}
335NOKPROBE_SYMBOL(oops_begin);
336
337void __noreturn rewind_stack_do_exit(int signr);
338
339void oops_end(unsigned long flags, struct pt_regs *regs, int signr)
340{
341 if (regs && kexec_should_crash(current))
342 crash_kexec(regs);
343
344 bust_spinlocks(0);
345 die_owner = -1;
346 add_taint(TAINT_DIE, LOCKDEP_NOW_UNRELIABLE);
347 die_nest_count--;
348 if (!die_nest_count)
349 /* Nest count reaches zero, release the lock. */
350 arch_spin_unlock(&die_lock);
351 raw_local_irq_restore(flags);
352 oops_exit();
353
354 /* Executive summary in case the oops scrolled away */
355 __show_regs(&exec_summary_regs, SHOW_REGS_ALL, KERN_DEFAULT);
356
357 if (!signr)
358 return;
359 if (in_interrupt())
360 panic("Fatal exception in interrupt");
361 if (panic_on_oops)
362 panic("Fatal exception");
363
364 /*
365 * We're not going to return, but we might be on an IST stack or
366 * have very little stack space left. Rewind the stack and kill
367 * the task.
368 * Before we rewind the stack, we have to tell KASAN that we're going to
369 * reuse the task stack and that existing poisons are invalid.
370 */
371 kasan_unpoison_task_stack(current);
372 rewind_stack_do_exit(signr);
373}
374NOKPROBE_SYMBOL(oops_end);
375
376static void __die_header(const char *str, struct pt_regs *regs, long err)
377{
378 const char *pr = "";
379
380 /* Save the regs of the first oops for the executive summary later. */
381 if (!die_counter)
382 exec_summary_regs = *regs;
383
384 if (IS_ENABLED(CONFIG_PREEMPTION))
385 pr = IS_ENABLED(CONFIG_PREEMPT_RT) ? " PREEMPT_RT" : " PREEMPT";
386
387 printk(KERN_DEFAULT
388 "%s: %04lx [#%d]%s%s%s%s%s\n", str, err & 0xffff, ++die_counter,
389 pr,
390 IS_ENABLED(CONFIG_SMP) ? " SMP" : "",
391 debug_pagealloc_enabled() ? " DEBUG_PAGEALLOC" : "",
392 IS_ENABLED(CONFIG_KASAN) ? " KASAN" : "",
393 IS_ENABLED(CONFIG_PAGE_TABLE_ISOLATION) ?
394 (boot_cpu_has(X86_FEATURE_PTI) ? " PTI" : " NOPTI") : "");
395}
396NOKPROBE_SYMBOL(__die_header);
397
398static int __die_body(const char *str, struct pt_regs *regs, long err)
399{
400 show_regs(regs);
401 print_modules();
402
403 if (notify_die(DIE_OOPS, str, regs, err,
404 current->thread.trap_nr, SIGSEGV) == NOTIFY_STOP)
405 return 1;
406
407 return 0;
408}
409NOKPROBE_SYMBOL(__die_body);
410
411int __die(const char *str, struct pt_regs *regs, long err)
412{
413 __die_header(str, regs, err);
414 return __die_body(str, regs, err);
415}
416NOKPROBE_SYMBOL(__die);
417
418/*
419 * This is gone through when something in the kernel has done something bad
420 * and is about to be terminated:
421 */
422void die(const char *str, struct pt_regs *regs, long err)
423{
424 unsigned long flags = oops_begin();
425 int sig = SIGSEGV;
426
427 if (__die(str, regs, err))
428 sig = 0;
429 oops_end(flags, regs, sig);
430}
431
432void die_addr(const char *str, struct pt_regs *regs, long err, long gp_addr)
433{
434 unsigned long flags = oops_begin();
435 int sig = SIGSEGV;
436
437 __die_header(str, regs, err);
438 if (gp_addr)
439 kasan_non_canonical_hook(gp_addr);
440 if (__die_body(str, regs, err))
441 sig = 0;
442 oops_end(flags, regs, sig);
443}
444
445void show_regs(struct pt_regs *regs)
446{
447 enum show_regs_mode print_kernel_regs;
448
449 show_regs_print_info(KERN_DEFAULT);
450
451 print_kernel_regs = user_mode(regs) ? SHOW_REGS_USER : SHOW_REGS_ALL;
452 __show_regs(regs, print_kernel_regs, KERN_DEFAULT);
453
454 /*
455 * When in-kernel, we also print out the stack at the time of the fault..
456 */
457 if (!user_mode(regs))
458 show_trace_log_lvl(current, regs, NULL, KERN_DEFAULT);
459}