arch/x86/kernel/dumpstack.c

   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/ftrace.h>
  15 #include <linux/kexec.h>
  16 #include <linux/bug.h>
  17 #include <linux/nmi.h>
  18 #include <linux/sysfs.h>
  19
  20 #include <asm/stacktrace.h>
  21 #include <asm/unwind.h>
  22
  23 int panic_on_unrecovered_nmi;
  24 int panic_on_io_nmi;
  25 unsigned int code_bytes = 64;
  26 static int die_counter;
  27
  28 bool in_task_stack(unsigned long *stack, struct task_struct *task,
  29                    struct stack_info *info)
  30 {
  31         unsigned long *begin = task_stack_page(task);
  32         unsigned long *end   = task_stack_page(task) + THREAD_SIZE;
  33
  34         if (stack < begin || stack >= end)
  35                 return false;
  36
  37         info->type      = STACK_TYPE_TASK;
  38         info->begin     = begin;
  39         info->end       = end;
  40         info->next_sp   = NULL;
  41
  42         return true;
  43 }
  44
  45 static void printk_stack_address(unsigned long address, int reliable,
  46                                  char *log_lvl)
  47 {
  48         touch_nmi_watchdog();
  49         printk("%s %s%pB\n", log_lvl, reliable ? "" : "? ", (void *)address);
  50 }
  51
  52 void show_trace_log_lvl(struct task_struct *task, struct pt_regs *regs,
  53                         unsigned long *stack, char *log_lvl)
  54 {
  55         struct unwind_state state;
  56         struct stack_info stack_info = {0};
  57         unsigned long visit_mask = 0;
  58         int graph_idx = 0;
  59
  60         printk("%sCall Trace:\n", log_lvl);
  61
  62         unwind_start(&state, task, regs, stack);
  63         stack = stack ? : get_stack_pointer(task, regs);
  64
  65         /*
  66          * Iterate through the stacks, starting with the current stack pointer.
  67          * Each stack has a pointer to the next one.
  68          *
  69          * x86-64 can have several stacks:
  70          * - task stack
  71          * - interrupt stack
  72          * - HW exception stacks (double fault, nmi, debug, mce)
  73          *
  74          * x86-32 can have up to three stacks:
  75          * - task stack
  76          * - softirq stack
  77          * - hardirq stack
  78          */
  79         for (regs = NULL; stack; stack = stack_info.next_sp) {
  80                 const char *stack_name;
  81
  82                 /*
  83                  * If we overflowed the task stack into a guard page, jump back
  84                  * to the bottom of the usable stack.
  85                  */
  86                 if (task_stack_page(task) - (void *)stack < PAGE_SIZE)
  87                         stack = task_stack_page(task);
  88
  89                 if (get_stack_info(stack, task, &stack_info, &visit_mask))
  90                         break;
  91
  92                 stack_name = stack_type_name(stack_info.type);
  93                 if (stack_name)
  94                         printk("%s <%s>\n", log_lvl, stack_name);
  95
  96                 /*
  97                  * Scan the stack, printing any text addresses we find.  At the
  98                  * same time, follow proper stack frames with the unwinder.
  99                  *
 100                  * Addresses found during the scan which are not reported by
 101                  * the unwinder are considered to be additional clues which are
 102                  * sometimes useful for debugging and are prefixed with '?'.
 103                  * This also serves as a failsafe option in case the unwinder
 104                  * goes off in the weeds.
 105                  */
 106                 for (; stack < stack_info.end; stack++) {
 107                         unsigned long real_addr;
 108                         int reliable = 0;
 109                         unsigned long addr = READ_ONCE_NOCHECK(*stack);
 110                         unsigned long *ret_addr_p =
 111                                 unwind_get_return_address_ptr(&state);
 112
 113                         if (!__kernel_text_address(addr))
 114                                 continue;
 115
 116                         /*
 117                          * Don't print regs->ip again if it was already printed
 118                          * by __show_regs() below.
 119                          */
 120                         if (regs && stack == &regs->ip) {
 121                                 unwind_next_frame(&state);
 122                                 continue;
 123                         }
 124
 125                         if (stack == ret_addr_p)
 126                                 reliable = 1;
 127
 128                         /*
 129                          * When function graph tracing is enabled for a
 130                          * function, its return address on the stack is
 131                          * replaced with the address of an ftrace handler
 132                          * (return_to_handler).  In that case, before printing
 133                          * the "real" address, we want to print the handler
 134                          * address as an "unreliable" hint that function graph
 135                          * tracing was involved.
 136                          */
 137                         real_addr = ftrace_graph_ret_addr(task, &graph_idx,
 138                                                           addr, stack);
 139                         if (real_addr != addr)
 140                                 printk_stack_address(addr, 0, log_lvl);
 141                         printk_stack_address(real_addr, reliable, log_lvl);
 142
 143                         if (!reliable)
 144                                 continue;
 145
 146                         /*
 147                          * Get the next frame from the unwinder.  No need to
 148                          * check for an error: if anything goes wrong, the rest
 149                          * of the addresses will just be printed as unreliable.
 150                          */
 151                         unwind_next_frame(&state);
 152
 153                         /* if the frame has entry regs, print them */
 154                         regs = unwind_get_entry_regs(&state);
 155                         if (regs)
 156                                 __show_regs(regs, 0);
 157                 }
 158
 159                 if (stack_name)
 160                         printk("%s </%s>\n", log_lvl, stack_name);
 161         }
 162 }
 163
 164 void show_stack(struct task_struct *task, unsigned long *sp)
 165 {
 166         task = task ? : current;
 167
 168         /*
 169          * Stack frames below this one aren't interesting.  Don't show them
 170          * if we're printing for %current.
 171          */
 172         if (!sp && task == current)
 173                 sp = get_stack_pointer(current, NULL);
 174
 175         show_trace_log_lvl(task, NULL, sp, KERN_DEFAULT);
 176 }
 177
 178 void show_stack_regs(struct pt_regs *regs)
 179 {
 180         show_trace_log_lvl(current, regs, NULL, KERN_DEFAULT);
 181 }
 182
 183 static arch_spinlock_t die_lock = __ARCH_SPIN_LOCK_UNLOCKED;
 184 static int die_owner = -1;
 185 static unsigned int die_nest_count;
 186
 187 unsigned long oops_begin(void)
 188 {
 189         int cpu;
 190         unsigned long flags;
 191
 192         oops_enter();
 193
 194         /* racy, but better than risking deadlock. */
 195         raw_local_irq_save(flags);
 196         cpu = smp_processor_id();
 197         if (!arch_spin_trylock(&die_lock)) {
 198                 if (cpu == die_owner)
 199                         /* nested oops. should stop eventually */;
 200                 else
 201                         arch_spin_lock(&die_lock);
 202         }
 203         die_nest_count++;
 204         die_owner = cpu;
 205         console_verbose();
 206         bust_spinlocks(1);
 207         return flags;
 208 }
 209 EXPORT_SYMBOL_GPL(oops_begin);
 210 NOKPROBE_SYMBOL(oops_begin);
 211
 212 void __noreturn rewind_stack_do_exit(int signr);
 213
 214 void oops_end(unsigned long flags, struct pt_regs *regs, int signr)
 215 {
 216         if (regs && kexec_should_crash(current))
 217                 crash_kexec(regs);
 218
 219         bust_spinlocks(0);
 220         die_owner = -1;
 221         add_taint(TAINT_DIE, LOCKDEP_NOW_UNRELIABLE);
 222         die_nest_count--;
 223         if (!die_nest_count)
 224                 /* Nest count reaches zero, release the lock. */
 225                 arch_spin_unlock(&die_lock);
 226         raw_local_irq_restore(flags);
 227         oops_exit();
 228
 229         if (!signr)
 230                 return;
 231         if (in_interrupt())
 232                 panic("Fatal exception in interrupt");
 233         if (panic_on_oops)
 234                 panic("Fatal exception");
 235
 236         /*
 237          * We're not going to return, but we might be on an IST stack or
 238          * have very little stack space left.  Rewind the stack and kill
 239          * the task.
 240          */
 241         rewind_stack_do_exit(signr);
 242 }
 243 NOKPROBE_SYMBOL(oops_end);
 244
 245 int __die(const char *str, struct pt_regs *regs, long err)
 246 {
 247 #ifdef CONFIG_X86_32
 248         unsigned short ss;
 249         unsigned long sp;
 250 #endif
 251         printk(KERN_DEFAULT
 252                "%s: %04lx [#%d]%s%s%s%s\n", str, err & 0xffff, ++die_counter,
 253                IS_ENABLED(CONFIG_PREEMPT) ? " PREEMPT"         : "",
 254                IS_ENABLED(CONFIG_SMP)     ? " SMP"             : "",
 255                debug_pagealloc_enabled()  ? " DEBUG_PAGEALLOC" : "",
 256                IS_ENABLED(CONFIG_KASAN)   ? " KASAN"           : "");
 257
 258         if (notify_die(DIE_OOPS, str, regs, err,
 259                         current->thread.trap_nr, SIGSEGV) == NOTIFY_STOP)
 260                 return 1;
 261
 262         print_modules();
 263         show_regs(regs);
 264 #ifdef CONFIG_X86_32
 265         if (user_mode(regs)) {
 266                 sp = regs->sp;
 267                 ss = regs->ss & 0xffff;
 268         } else {
 269                 sp = kernel_stack_pointer(regs);
 270                 savesegment(ss, ss);
 271         }
 272         printk(KERN_EMERG "EIP: %pS SS:ESP: %04x:%08lx\n",
 273                (void *)regs->ip, ss, sp);
 274 #else
 275         /* Executive summary in case the oops scrolled away */
 276         printk(KERN_ALERT "RIP: %pS RSP: %016lx\n", (void *)regs->ip, regs->sp);
 277 #endif
 278         return 0;
 279 }
 280 NOKPROBE_SYMBOL(__die);
 281
 282 /*
 283  * This is gone through when something in the kernel has done something bad
 284  * and is about to be terminated:
 285  */
 286 void die(const char *str, struct pt_regs *regs, long err)
 287 {
 288         unsigned long flags = oops_begin();
 289         int sig = SIGSEGV;
 290
 291         if (!user_mode(regs))
 292                 report_bug(regs->ip, regs);
 293
 294         if (__die(str, regs, err))
 295                 sig = 0;
 296         oops_end(flags, regs, sig);
 297 }
 298
 299 static int __init code_bytes_setup(char *s)
 300 {
 301         ssize_t ret;
 302         unsigned long val;
 303
 304         if (!s)
 305                 return -EINVAL;
 306
 307         ret = kstrtoul(s, 0, &val);
 308         if (ret)
 309                 return ret;
 310
 311         code_bytes = val;
 312         if (code_bytes > 8192)
 313                 code_bytes = 8192;
 314
 315         return 1;
 316 }
 317 __setup("code_bytes=", code_bytes_setup);