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
  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 static int print_trace_stack(void *data, const char *name)
 146 {
 147         printk("%s <%s> ", (char *)data, name);
 148         return 0;
 149 }
 150
 151 /*
 152  * Print one address/symbol entries per line.
 153  */
 154 static int print_trace_address(void *data, unsigned long addr, int reliable)
 155 {
 156         printk_stack_address(addr, reliable, data);
 157         return 0;
 158 }
 159
 160 static const struct stacktrace_ops print_trace_ops = {
 161         .stack                  = print_trace_stack,
 162         .address                = print_trace_address,
 163         .walk_stack             = print_context_stack,
 164 };
 165
 166 void
 167 show_trace_log_lvl(struct task_struct *task, struct pt_regs *regs,
 168                 unsigned long *stack, unsigned long bp, char *log_lvl)
 169 {
 170         printk("%sCall Trace:\n", log_lvl);
 171         dump_trace(task, regs, stack, bp, &print_trace_ops, log_lvl);
 172 }
 173
 174 void show_stack(struct task_struct *task, unsigned long *sp)
 175 {
 176         unsigned long bp = 0;
 177
 178         task = task ? : current;
 179
 180         /*
 181          * Stack frames below this one aren't interesting.  Don't show them
 182          * if we're printing for %current.
 183          */
 184         if (!sp && task == current) {
 185                 sp = get_stack_pointer(current, NULL);
 186                 bp = (unsigned long)get_frame_pointer(current, NULL);
 187         }
 188
 189         show_stack_log_lvl(task, NULL, sp, bp, "");
 190 }
 191
 192 void show_stack_regs(struct pt_regs *regs)
 193 {
 194         show_stack_log_lvl(current, regs, NULL, 0, "");
 195 }
 196
 197 static arch_spinlock_t die_lock = __ARCH_SPIN_LOCK_UNLOCKED;
 198 static int die_owner = -1;
 199 static unsigned int die_nest_count;
 200
 201 unsigned long oops_begin(void)
 202 {
 203         int cpu;
 204         unsigned long flags;
 205
 206         oops_enter();
 207
 208         /* racy, but better than risking deadlock. */
 209         raw_local_irq_save(flags);
 210         cpu = smp_processor_id();
 211         if (!arch_spin_trylock(&die_lock)) {
 212                 if (cpu == die_owner)
 213                         /* nested oops. should stop eventually */;
 214                 else
 215                         arch_spin_lock(&die_lock);
 216         }
 217         die_nest_count++;
 218         die_owner = cpu;
 219         console_verbose();
 220         bust_spinlocks(1);
 221         return flags;
 222 }
 223 EXPORT_SYMBOL_GPL(oops_begin);
 224 NOKPROBE_SYMBOL(oops_begin);
 225
 226 void __noreturn rewind_stack_do_exit(int signr);
 227
 228 void oops_end(unsigned long flags, struct pt_regs *regs, int signr)
 229 {
 230         if (regs && kexec_should_crash(current))
 231                 crash_kexec(regs);
 232
 233         bust_spinlocks(0);
 234         die_owner = -1;
 235         add_taint(TAINT_DIE, LOCKDEP_NOW_UNRELIABLE);
 236         die_nest_count--;
 237         if (!die_nest_count)
 238                 /* Nest count reaches zero, release the lock. */
 239                 arch_spin_unlock(&die_lock);
 240         raw_local_irq_restore(flags);
 241         oops_exit();
 242
 243         if (!signr)
 244                 return;
 245         if (in_interrupt())
 246                 panic("Fatal exception in interrupt");
 247         if (panic_on_oops)
 248                 panic("Fatal exception");
 249
 250         /*
 251          * We're not going to return, but we might be on an IST stack or
 252          * have very little stack space left.  Rewind the stack and kill
 253          * the task.
 254          */
 255         rewind_stack_do_exit(signr);
 256 }
 257 NOKPROBE_SYMBOL(oops_end);
 258
 259 int __die(const char *str, struct pt_regs *regs, long err)
 260 {
 261 #ifdef CONFIG_X86_32
 262         unsigned short ss;
 263         unsigned long sp;
 264 #endif
 265         printk(KERN_DEFAULT
 266                "%s: %04lx [#%d]%s%s%s%s\n", str, err & 0xffff, ++die_counter,
 267                IS_ENABLED(CONFIG_PREEMPT) ? " PREEMPT"         : "",
 268                IS_ENABLED(CONFIG_SMP)     ? " SMP"             : "",
 269                debug_pagealloc_enabled()  ? " DEBUG_PAGEALLOC" : "",
 270                IS_ENABLED(CONFIG_KASAN)   ? " KASAN"           : "");
 271
 272         if (notify_die(DIE_OOPS, str, regs, err,
 273                         current->thread.trap_nr, SIGSEGV) == NOTIFY_STOP)
 274                 return 1;
 275
 276         print_modules();
 277         show_regs(regs);
 278 #ifdef CONFIG_X86_32
 279         if (user_mode(regs)) {
 280                 sp = regs->sp;
 281                 ss = regs->ss & 0xffff;
 282         } else {
 283                 sp = kernel_stack_pointer(regs);
 284                 savesegment(ss, ss);
 285         }
 286         printk(KERN_EMERG "EIP: [<%08lx>] ", regs->ip);
 287         print_symbol("%s", regs->ip);
 288         printk(" SS:ESP %04x:%08lx\n", ss, sp);
 289 #else
 290         /* Executive summary in case the oops scrolled away */
 291         printk(KERN_ALERT "RIP ");
 292         printk_address(regs->ip);
 293         printk(" RSP <%016lx>\n", regs->sp);
 294 #endif
 295         return 0;
 296 }
 297 NOKPROBE_SYMBOL(__die);
 298
 299 /*
 300  * This is gone through when something in the kernel has done something bad
 301  * and is about to be terminated:
 302  */
 303 void die(const char *str, struct pt_regs *regs, long err)
 304 {
 305         unsigned long flags = oops_begin();
 306         int sig = SIGSEGV;
 307
 308         if (!user_mode(regs))
 309                 report_bug(regs->ip, regs);
 310
 311         if (__die(str, regs, err))
 312                 sig = 0;
 313         oops_end(flags, regs, sig);
 314 }
 315
 316 static int __init kstack_setup(char *s)
 317 {
 318         ssize_t ret;
 319         unsigned long val;
 320
 321         if (!s)
 322                 return -EINVAL;
 323
 324         ret = kstrtoul(s, 0, &val);
 325         if (ret)
 326                 return ret;
 327         kstack_depth_to_print = val;
 328         return 0;
 329 }
 330 early_param("kstack", kstack_setup);
 331
 332 static int __init code_bytes_setup(char *s)
 333 {
 334         ssize_t ret;
 335         unsigned long val;
 336
 337         if (!s)
 338                 return -EINVAL;
 339
 340         ret = kstrtoul(s, 0, &val);
 341         if (ret)
 342                 return ret;
 343
 344         code_bytes = val;
 345         if (code_bytes > 8192)
 346                 code_bytes = 8192;
 347
 348         return 1;
 349 }
 350 __setup("code_bytes=", code_bytes_setup);