2 * Copyright (C) 1991, 1992 Linus Torvalds
3 * Copyright (C) 2000, 2001, 2002 Andi Kleen, SuSE Labs
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>
22 #include <asm/cpu_entry_area.h>
23 #include <asm/stacktrace.h>
24 #include <asm/unwind.h>
26 int panic_on_unrecovered_nmi;
28 static int die_counter;
30 static struct pt_regs exec_summary_regs;
32 bool noinstr in_task_stack(unsigned long *stack, struct task_struct *task,
33 struct stack_info *info)
35 unsigned long *begin = task_stack_page(task);
36 unsigned long *end = task_stack_page(task) + THREAD_SIZE;
38 if (stack < begin || stack >= end)
41 info->type = STACK_TYPE_TASK;
49 /* Called from get_stack_info_noinstr - so must be noinstr too */
50 bool noinstr in_entry_stack(unsigned long *stack, struct stack_info *info)
52 struct entry_stack *ss = cpu_entry_stack(smp_processor_id());
57 if ((void *)stack < begin || (void *)stack >= end)
60 info->type = STACK_TYPE_ENTRY;
68 static void printk_stack_address(unsigned long address, int reliable,
72 printk("%s %s%pBb\n", log_lvl, reliable ? "" : "? ", (void *)address);
75 static int copy_code(struct pt_regs *regs, u8 *buf, unsigned long src,
79 return copy_from_kernel_nofault(buf, (u8 *)src, nbytes);
81 /* The user space code from other tasks cannot be accessed. */
82 if (regs != task_pt_regs(current))
86 * Even if named copy_from_user_nmi() this can be invoked from
87 * other contexts and will not try to resolve a pagefault, which is
88 * the correct thing to do here as this code can be called from any
91 return copy_from_user_nmi(buf, (void __user *)src, nbytes);
95 * There are a couple of reasons for the 2/3rd prologue, courtesy of Linus:
97 * In case where we don't have the exact kernel image (which, if we did, we can
98 * simply disassemble and navigate to the RIP), the purpose of the bigger
99 * prologue is to have more context and to be able to correlate the code from
100 * the different toolchains better.
102 * In addition, it helps in recreating the register allocation of the failing
103 * kernel and thus make sense of the register dump.
105 * What is more, the additional complication of a variable length insn arch like
106 * x86 warrants having longer byte sequence before rIP so that the disassembler
107 * can "sync" up properly and find instruction boundaries when decoding the
110 * Thus, the 2/3rds prologue and 64 byte OPCODE_BUFSIZE is just a random
111 * guesstimate in attempt to achieve all of the above.
113 void show_opcodes(struct pt_regs *regs, const char *loglvl)
115 #define PROLOGUE_SIZE 42
116 #define EPILOGUE_SIZE 21
117 #define OPCODE_BUFSIZE (PROLOGUE_SIZE + 1 + EPILOGUE_SIZE)
118 u8 opcodes[OPCODE_BUFSIZE];
119 unsigned long prologue = regs->ip - PROLOGUE_SIZE;
121 switch (copy_code(regs, opcodes, prologue, sizeof(opcodes))) {
123 printk("%sCode: %" __stringify(PROLOGUE_SIZE) "ph <%02x> %"
124 __stringify(EPILOGUE_SIZE) "ph\n", loglvl, opcodes,
125 opcodes[PROLOGUE_SIZE], opcodes + PROLOGUE_SIZE + 1);
128 /* No access to the user space stack of other tasks. Ignore. */
131 printk("%sCode: Unable to access opcode bytes at RIP 0x%lx.\n",
137 void show_ip(struct pt_regs *regs, const char *loglvl)
140 printk("%sEIP: %pS\n", loglvl, (void *)regs->ip);
142 printk("%sRIP: %04x:%pS\n", loglvl, (int)regs->cs, (void *)regs->ip);
144 show_opcodes(regs, loglvl);
147 void show_iret_regs(struct pt_regs *regs, const char *log_lvl)
149 show_ip(regs, log_lvl);
150 printk("%sRSP: %04x:%016lx EFLAGS: %08lx", log_lvl, (int)regs->ss,
151 regs->sp, regs->flags);
154 static void show_regs_if_on_stack(struct stack_info *info, struct pt_regs *regs,
155 bool partial, const char *log_lvl)
158 * These on_stack() checks aren't strictly necessary: the unwind code
159 * has already validated the 'regs' pointer. The checks are done for
160 * ordering reasons: if the registers are on the next stack, we don't
161 * want to print them out yet. Otherwise they'll be shown as part of
162 * the wrong stack. Later, when show_trace_log_lvl() switches to the
163 * next stack, this function will be called again with the same regs so
164 * they can be printed in the right context.
166 if (!partial && on_stack(info, regs, sizeof(*regs))) {
167 __show_regs(regs, SHOW_REGS_SHORT, log_lvl);
169 } else if (partial && on_stack(info, (void *)regs + IRET_FRAME_OFFSET,
172 * When an interrupt or exception occurs in entry code, the
173 * full pt_regs might not have been saved yet. In that case
174 * just print the iret frame.
176 show_iret_regs(regs, log_lvl);
180 static void show_trace_log_lvl(struct task_struct *task, struct pt_regs *regs,
181 unsigned long *stack, const char *log_lvl)
183 struct unwind_state state;
184 struct stack_info stack_info = {0};
185 unsigned long visit_mask = 0;
187 bool partial = false;
189 printk("%sCall Trace:\n", log_lvl);
191 unwind_start(&state, task, regs, stack);
192 stack = stack ? : get_stack_pointer(task, regs);
193 regs = unwind_get_entry_regs(&state, &partial);
196 * Iterate through the stacks, starting with the current stack pointer.
197 * Each stack has a pointer to the next one.
199 * x86-64 can have several stacks:
202 * - HW exception stacks (double fault, nmi, debug, mce)
205 * x86-32 can have up to four stacks:
211 for ( ; stack; stack = PTR_ALIGN(stack_info.next_sp, sizeof(long))) {
212 const char *stack_name;
214 if (get_stack_info(stack, task, &stack_info, &visit_mask)) {
216 * We weren't on a valid stack. It's possible that
217 * we overflowed a valid stack into a guard page.
218 * See if the next page up is valid so that we can
219 * generate some kind of backtrace if this happens.
221 stack = (unsigned long *)PAGE_ALIGN((unsigned long)stack);
222 if (get_stack_info(stack, task, &stack_info, &visit_mask))
226 stack_name = stack_type_name(stack_info.type);
228 printk("%s <%s>\n", log_lvl, stack_name);
231 show_regs_if_on_stack(&stack_info, regs, partial, log_lvl);
234 * Scan the stack, printing any text addresses we find. At the
235 * same time, follow proper stack frames with the unwinder.
237 * Addresses found during the scan which are not reported by
238 * the unwinder are considered to be additional clues which are
239 * sometimes useful for debugging and are prefixed with '?'.
240 * This also serves as a failsafe option in case the unwinder
241 * goes off in the weeds.
243 for (; stack < stack_info.end; stack++) {
244 unsigned long real_addr;
246 unsigned long addr = READ_ONCE_NOCHECK(*stack);
247 unsigned long *ret_addr_p =
248 unwind_get_return_address_ptr(&state);
250 if (!__kernel_text_address(addr))
254 * Don't print regs->ip again if it was already printed
255 * by show_regs_if_on_stack().
257 if (regs && stack == ®s->ip)
260 if (stack == ret_addr_p)
264 * When function graph tracing is enabled for a
265 * function, its return address on the stack is
266 * replaced with the address of an ftrace handler
267 * (return_to_handler). In that case, before printing
268 * the "real" address, we want to print the handler
269 * address as an "unreliable" hint that function graph
270 * tracing was involved.
272 real_addr = ftrace_graph_ret_addr(task, &graph_idx,
274 if (real_addr != addr)
275 printk_stack_address(addr, 0, log_lvl);
276 printk_stack_address(real_addr, reliable, log_lvl);
283 * Get the next frame from the unwinder. No need to
284 * check for an error: if anything goes wrong, the rest
285 * of the addresses will just be printed as unreliable.
287 unwind_next_frame(&state);
289 /* if the frame has entry regs, print them */
290 regs = unwind_get_entry_regs(&state, &partial);
292 show_regs_if_on_stack(&stack_info, regs, partial, log_lvl);
296 printk("%s </%s>\n", log_lvl, stack_name);
300 void show_stack(struct task_struct *task, unsigned long *sp,
303 task = task ? : current;
306 * Stack frames below this one aren't interesting. Don't show them
307 * if we're printing for %current.
309 if (!sp && task == current)
310 sp = get_stack_pointer(current, NULL);
312 show_trace_log_lvl(task, NULL, sp, loglvl);
315 void show_stack_regs(struct pt_regs *regs)
317 show_trace_log_lvl(current, regs, NULL, KERN_DEFAULT);
320 static arch_spinlock_t die_lock = __ARCH_SPIN_LOCK_UNLOCKED;
321 static int die_owner = -1;
322 static unsigned int die_nest_count;
324 unsigned long oops_begin(void)
331 /* racy, but better than risking deadlock. */
332 raw_local_irq_save(flags);
333 cpu = smp_processor_id();
334 if (!arch_spin_trylock(&die_lock)) {
335 if (cpu == die_owner)
336 /* nested oops. should stop eventually */;
338 arch_spin_lock(&die_lock);
346 NOKPROBE_SYMBOL(oops_begin);
348 void __noreturn rewind_stack_and_make_dead(int signr);
350 void oops_end(unsigned long flags, struct pt_regs *regs, int signr)
352 if (regs && kexec_should_crash(current))
357 add_taint(TAINT_DIE, LOCKDEP_NOW_UNRELIABLE);
360 /* Nest count reaches zero, release the lock. */
361 arch_spin_unlock(&die_lock);
362 raw_local_irq_restore(flags);
365 /* Executive summary in case the oops scrolled away */
366 __show_regs(&exec_summary_regs, SHOW_REGS_ALL, KERN_DEFAULT);
371 panic("Fatal exception in interrupt");
373 panic("Fatal exception");
376 * We're not going to return, but we might be on an IST stack or
377 * have very little stack space left. Rewind the stack and kill
379 * Before we rewind the stack, we have to tell KASAN that we're going to
380 * reuse the task stack and that existing poisons are invalid.
382 kasan_unpoison_task_stack(current);
383 rewind_stack_and_make_dead(signr);
385 NOKPROBE_SYMBOL(oops_end);
387 static void __die_header(const char *str, struct pt_regs *regs, long err)
391 /* Save the regs of the first oops for the executive summary later. */
393 exec_summary_regs = *regs;
395 if (IS_ENABLED(CONFIG_PREEMPTION))
396 pr = IS_ENABLED(CONFIG_PREEMPT_RT) ? " PREEMPT_RT" : " PREEMPT";
399 "%s: %04lx [#%d]%s%s%s%s%s\n", str, err & 0xffff, ++die_counter,
401 IS_ENABLED(CONFIG_SMP) ? " SMP" : "",
402 debug_pagealloc_enabled() ? " DEBUG_PAGEALLOC" : "",
403 IS_ENABLED(CONFIG_KASAN) ? " KASAN" : "",
404 IS_ENABLED(CONFIG_PAGE_TABLE_ISOLATION) ?
405 (boot_cpu_has(X86_FEATURE_PTI) ? " PTI" : " NOPTI") : "");
407 NOKPROBE_SYMBOL(__die_header);
409 static int __die_body(const char *str, struct pt_regs *regs, long err)
414 if (notify_die(DIE_OOPS, str, regs, err,
415 current->thread.trap_nr, SIGSEGV) == NOTIFY_STOP)
420 NOKPROBE_SYMBOL(__die_body);
422 int __die(const char *str, struct pt_regs *regs, long err)
424 __die_header(str, regs, err);
425 return __die_body(str, regs, err);
427 NOKPROBE_SYMBOL(__die);
430 * This is gone through when something in the kernel has done something bad
431 * and is about to be terminated:
433 void die(const char *str, struct pt_regs *regs, long err)
435 unsigned long flags = oops_begin();
438 if (__die(str, regs, err))
440 oops_end(flags, regs, sig);
443 void die_addr(const char *str, struct pt_regs *regs, long err, long gp_addr)
445 unsigned long flags = oops_begin();
448 __die_header(str, regs, err);
450 kasan_non_canonical_hook(gp_addr);
451 if (__die_body(str, regs, err))
453 oops_end(flags, regs, sig);
456 void show_regs(struct pt_regs *regs)
458 enum show_regs_mode print_kernel_regs;
460 show_regs_print_info(KERN_DEFAULT);
462 print_kernel_regs = user_mode(regs) ? SHOW_REGS_USER : SHOW_REGS_ALL;
463 __show_regs(regs, print_kernel_regs, KERN_DEFAULT);
466 * When in-kernel, we also print out the stack at the time of the fault..
468 if (!user_mode(regs))
469 show_trace_log_lvl(current, regs, NULL, KERN_DEFAULT);