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/ftrace.h>
  14 #include <linux/kexec.h>
  15 #include <linux/bug.h>
  16 #include <linux/nmi.h>
  17 #include <linux/sysfs.h>
  18
  19 #include <asm/stacktrace.h>
  20 #include <asm/unwind.h>
  21
  22 int panic_on_unrecovered_nmi;
  23 int panic_on_io_nmi;
  24 unsigned int code_bytes = 64;
  25 int kstack_depth_to_print = 3 * STACKSLOTS_PER_LINE;
  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 [<%p>] %s%pB\n",
  50                 log_lvl, (void *)address, reliable ? "" : "? ",
  51                 (void *)address);
  52 }
  53
  54 void printk_address(unsigned long address)
  55 {
  56         pr_cont(" [<%p>] %pS\n", (void *)address, (void *)address);
  57 }
  58
  59 /*
  60  * x86-64 can have up to three kernel stacks:
  61  * process stack
  62  * interrupt stack
  63  * severe exception (double fault, nmi, stack fault, debug, mce) hardware stack
  64  */
  65
  66 unsigned long
  67 print_context_stack(struct task_struct *task,
  68                 unsigned long *stack, unsigned long bp,
  69                 const struct stacktrace_ops *ops, void *data,
  70                 struct stack_info *info, int *graph)
  71 {
  72         struct stack_frame *frame = (struct stack_frame *)bp;
  73
  74         /*
  75          * If we overflowed the stack into a guard page, jump back to the
  76          * bottom of the usable stack.
  77          */
  78         if ((unsigned long)task_stack_page(task) - (unsigned long)stack <
  79             PAGE_SIZE)
  80                 stack = (unsigned long *)task_stack_page(task);
  81
  82         while (on_stack(info, stack, sizeof(*stack))) {
  83                 unsigned long addr = *stack;
  84
  85                 if (__kernel_text_address(addr)) {
  86                         unsigned long real_addr;
  87                         int reliable = 0;
  88
  89                         if ((unsigned long) stack == bp + sizeof(long)) {
  90                                 reliable = 1;
  91                                 frame = frame->next_frame;
  92                                 bp = (unsigned long) frame;
  93                         }
  94
  95                         /*
  96                          * When function graph tracing is enabled for a
  97                          * function, its return address on the stack is
  98                          * replaced with the address of an ftrace handler
  99                          * (return_to_handler).  In that case, before printing
 100                          * the "real" address, we want to print the handler
 101                          * address as an "unreliable" hint that function graph
 102                          * tracing was involved.
 103                          */
 104                         real_addr = ftrace_graph_ret_addr(task, graph, addr,
 105                                                           stack);
 106                         if (real_addr != addr)
 107                                 ops->address(data, addr, 0);
 108
 109                         ops->address(data, real_addr, reliable);
 110                 }
 111                 stack++;
 112         }
 113         return bp;
 114 }
 115 EXPORT_SYMBOL_GPL(print_context_stack);
 116
 117 unsigned long
 118 print_context_stack_bp(struct task_struct *task,
 119                        unsigned long *stack, unsigned long bp,
 120                        const struct stacktrace_ops *ops, void *data,
 121                        struct stack_info *info, int *graph)
 122 {
 123         struct stack_frame *frame = (struct stack_frame *)bp;
 124         unsigned long *retp = &frame->return_address;
 125
 126         while (on_stack(info, stack, sizeof(*stack) * 2)) {
 127                 unsigned long addr = *retp;
 128                 unsigned long real_addr;
 129
 130                 if (!__kernel_text_address(addr))
 131                         break;
 132
 133                 real_addr = ftrace_graph_ret_addr(task, graph, addr, retp);
 134                 if (ops->address(data, real_addr, 1))
 135                         break;
 136
 137                 frame = frame->next_frame;
 138                 retp = &frame->return_address;
 139         }
 140
 141         return (unsigned long)frame;
 142 }
 143 EXPORT_SYMBOL_GPL(print_context_stack_bp);
 144
 145 void show_trace_log_lvl(struct task_struct *task, struct pt_regs *regs,
 146                         unsigned long *stack, char *log_lvl)
 147 {
 148         struct unwind_state state;
 149         struct stack_info stack_info = {0};
 150         unsigned long visit_mask = 0;
 151         int graph_idx = 0;
 152
 153         printk("%sCall Trace:\n", log_lvl);
 154
 155         unwind_start(&state, task, regs, stack);
 156
 157         /*
 158          * Iterate through the stacks, starting with the current stack pointer.
 159          * Each stack has a pointer to the next one.
 160          *
 161          * x86-64 can have several stacks:
 162          * - task stack
 163          * - interrupt stack
 164          * - HW exception stacks (double fault, nmi, debug, mce)
 165          *
 166          * x86-32 can have up to three stacks:
 167          * - task stack
 168          * - softirq stack
 169          * - hardirq stack
 170          */
 171         for (; stack; stack = stack_info.next_sp) {
 172                 const char *str_begin, *str_end;
 173
 174                 /*
 175                  * If we overflowed the task stack into a guard page, jump back
 176                  * to the bottom of the usable stack.
 177                  */
 178                 if (task_stack_page(task) - (void *)stack < PAGE_SIZE)
 179                         stack = task_stack_page(task);
 180
 181                 if (get_stack_info(stack, task, &stack_info, &visit_mask))
 182                         break;
 183
 184                 stack_type_str(stack_info.type, &str_begin, &str_end);
 185                 if (str_begin)
 186                         printk("%s <%s> ", log_lvl, str_begin);
 187
 188                 /*
 189                  * Scan the stack, printing any text addresses we find.  At the
 190                  * same time, follow proper stack frames with the unwinder.
 191                  *
 192                  * Addresses found during the scan which are not reported by
 193                  * the unwinder are considered to be additional clues which are
 194                  * sometimes useful for debugging and are prefixed with '?'.
 195                  * This also serves as a failsafe option in case the unwinder
 196                  * goes off in the weeds.
 197                  */
 198                 for (; stack < stack_info.end; stack++) {
 199                         unsigned long real_addr;
 200                         int reliable = 0;
 201                         unsigned long addr = *stack;
 202                         unsigned long *ret_addr_p =
 203                                 unwind_get_return_address_ptr(&state);
 204
 205                         if (!__kernel_text_address(addr))
 206                                 continue;
 207
 208                         if (stack == ret_addr_p)
 209                                 reliable = 1;
 210
 211                         /*
 212                          * When function graph tracing is enabled for a
 213                          * function, its return address on the stack is
 214                          * replaced with the address of an ftrace handler
 215                          * (return_to_handler).  In that case, before printing
 216                          * the "real" address, we want to print the handler
 217                          * address as an "unreliable" hint that function graph
 218                          * tracing was involved.
 219                          */
 220                         real_addr = ftrace_graph_ret_addr(task, &graph_idx,
 221                                                           addr, stack);
 222                         if (real_addr != addr)
 223                                 printk_stack_address(addr, 0, log_lvl);
 224                         printk_stack_address(real_addr, reliable, log_lvl);
 225
 226                         if (!reliable)
 227                                 continue;
 228
 229                         /*
 230                          * Get the next frame from the unwinder.  No need to
 231                          * check for an error: if anything goes wrong, the rest
 232                          * of the addresses will just be printed as unreliable.
 233                          */
 234                         unwind_next_frame(&state);
 235                 }
 236
 237                 if (str_end)
 238                         printk("%s <%s> ", log_lvl, str_end);
 239         }
 240 }
 241
 242 void show_stack(struct task_struct *task, unsigned long *sp)
 243 {
 244         task = task ? : current;
 245
 246         /*
 247          * Stack frames below this one aren't interesting.  Don't show them
 248          * if we're printing for %current.
 249          */
 250         if (!sp && task == current)
 251                 sp = get_stack_pointer(current, NULL);
 252
 253         show_stack_log_lvl(current, NULL, sp, "");
 254 }
 255
 256 void show_stack_regs(struct pt_regs *regs)
 257 {
 258         show_stack_log_lvl(current, regs, NULL, "");
 259 }
 260
 261 static arch_spinlock_t die_lock = __ARCH_SPIN_LOCK_UNLOCKED;
 262 static int die_owner = -1;
 263 static unsigned int die_nest_count;
 264
 265 unsigned long oops_begin(void)
 266 {
 267         int cpu;
 268         unsigned long flags;
 269
 270         oops_enter();
 271
 272         /* racy, but better than risking deadlock. */
 273         raw_local_irq_save(flags);
 274         cpu = smp_processor_id();
 275         if (!arch_spin_trylock(&die_lock)) {
 276                 if (cpu == die_owner)
 277                         /* nested oops. should stop eventually */;
 278                 else
 279                         arch_spin_lock(&die_lock);
 280         }
 281         die_nest_count++;
 282         die_owner = cpu;
 283         console_verbose();
 284         bust_spinlocks(1);
 285         return flags;
 286 }
 287 EXPORT_SYMBOL_GPL(oops_begin);
 288 NOKPROBE_SYMBOL(oops_begin);
 289
 290 void __noreturn rewind_stack_do_exit(int signr);
 291
 292 void oops_end(unsigned long flags, struct pt_regs *regs, int signr)
 293 {
 294         if (regs && kexec_should_crash(current))
 295                 crash_kexec(regs);
 296
 297         bust_spinlocks(0);
 298         die_owner = -1;
 299         add_taint(TAINT_DIE, LOCKDEP_NOW_UNRELIABLE);
 300         die_nest_count--;
 301         if (!die_nest_count)
 302                 /* Nest count reaches zero, release the lock. */
 303                 arch_spin_unlock(&die_lock);
 304         raw_local_irq_restore(flags);
 305         oops_exit();
 306
 307         if (!signr)
 308                 return;
 309         if (in_interrupt())
 310                 panic("Fatal exception in interrupt");
 311         if (panic_on_oops)
 312                 panic("Fatal exception");
 313
 314         /*
 315          * We're not going to return, but we might be on an IST stack or
 316          * have very little stack space left.  Rewind the stack and kill
 317          * the task.
 318          */
 319         rewind_stack_do_exit(signr);
 320 }
 321 NOKPROBE_SYMBOL(oops_end);
 322
 323 int __die(const char *str, struct pt_regs *regs, long err)
 324 {
 325 #ifdef CONFIG_X86_32
 326         unsigned short ss;
 327         unsigned long sp;
 328 #endif
 329         printk(KERN_DEFAULT
 330                "%s: %04lx [#%d]%s%s%s%s\n", str, err & 0xffff, ++die_counter,
 331                IS_ENABLED(CONFIG_PREEMPT) ? " PREEMPT"         : "",
 332                IS_ENABLED(CONFIG_SMP)     ? " SMP"             : "",
 333                debug_pagealloc_enabled()  ? " DEBUG_PAGEALLOC" : "",
 334                IS_ENABLED(CONFIG_KASAN)   ? " KASAN"           : "");
 335
 336         if (notify_die(DIE_OOPS, str, regs, err,
 337                         current->thread.trap_nr, SIGSEGV) == NOTIFY_STOP)
 338                 return 1;
 339
 340         print_modules();
 341         show_regs(regs);
 342 #ifdef CONFIG_X86_32
 343         if (user_mode(regs)) {
 344                 sp = regs->sp;
 345                 ss = regs->ss & 0xffff;
 346         } else {
 347                 sp = kernel_stack_pointer(regs);
 348                 savesegment(ss, ss);
 349         }
 350         printk(KERN_EMERG "EIP: [<%08lx>] ", regs->ip);
 351         print_symbol("%s", regs->ip);
 352         printk(" SS:ESP %04x:%08lx\n", ss, sp);
 353 #else
 354         /* Executive summary in case the oops scrolled away */
 355         printk(KERN_ALERT "RIP ");
 356         printk_address(regs->ip);
 357         printk(" RSP <%016lx>\n", regs->sp);
 358 #endif
 359         return 0;
 360 }
 361 NOKPROBE_SYMBOL(__die);
 362
 363 /*
 364  * This is gone through when something in the kernel has done something bad
 365  * and is about to be terminated:
 366  */
 367 void die(const char *str, struct pt_regs *regs, long err)
 368 {
 369         unsigned long flags = oops_begin();
 370         int sig = SIGSEGV;
 371
 372         if (!user_mode(regs))
 373                 report_bug(regs->ip, regs);
 374
 375         if (__die(str, regs, err))
 376                 sig = 0;
 377         oops_end(flags, regs, sig);
 378 }
 379
 380 static int __init kstack_setup(char *s)
 381 {
 382         ssize_t ret;
 383         unsigned long val;
 384
 385         if (!s)
 386                 return -EINVAL;
 387
 388         ret = kstrtoul(s, 0, &val);
 389         if (ret)
 390                 return ret;
 391         kstack_depth_to_print = val;
 392         return 0;
 393 }
 394 early_param("kstack", kstack_setup);
 395
 396 static int __init code_bytes_setup(char *s)
 397 {
 398         ssize_t ret;
 399         unsigned long val;
 400
 401         if (!s)
 402                 return -EINVAL;
 403
 404         ret = kstrtoul(s, 0, &val);
 405         if (ret)
 406                 return ret;
 407
 408         code_bytes = val;
 409         if (code_bytes > 8192)
 410                 code_bytes = 8192;
 411
 412         return 1;
 413 }
 414 __setup("code_bytes=", code_bytes_setup);