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 static void printk_stack_address(unsigned long address, int reliable,
  29                 void *data)
  30 {
  31         printk("%s [<%p>] %s%pB\n",
  32                 (char *)data, (void *)address, reliable ? "" : "? ",
  33                 (void *)address);
  34 }
  35
  36 void printk_address(unsigned long address)
  37 {
  38         pr_cont(" [<%p>] %pS\n", (void *)address, (void *)address);
  39 }
  40
  41 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
  42 static void
  43 print_ftrace_graph_addr(unsigned long addr, void *data,
  44                         const struct stacktrace_ops *ops,
  45                         struct task_struct *task, int *graph)
  46 {
  47         unsigned long ret_addr;
  48         int index;
  49
  50         if (addr != (unsigned long)return_to_handler)
  51                 return;
  52
  53         index = task->curr_ret_stack;
  54
  55         if (!task->ret_stack || index < *graph)
  56                 return;
  57
  58         index -= *graph;
  59         ret_addr = task->ret_stack[index].ret;
  60
  61         ops->address(data, ret_addr, 1);
  62
  63         (*graph)++;
  64 }
  65 #else
  66 static inline void
  67 print_ftrace_graph_addr(unsigned long addr, void *data,
  68                         const struct stacktrace_ops *ops,
  69                         struct task_struct *task, int *graph)
  70 { }
  71 #endif
  72
  73 /*
  74  * x86-64 can have up to three kernel stacks:
  75  * process stack
  76  * interrupt stack
  77  * severe exception (double fault, nmi, stack fault, debug, mce) hardware stack
  78  */
  79
  80 static inline int valid_stack_ptr(struct task_struct *task,
  81                         void *p, unsigned int size, void *end)
  82 {
  83         void *t = task_stack_page(task);
  84         if (end) {
  85                 if (p < end && p >= (end-THREAD_SIZE))
  86                         return 1;
  87                 else
  88                         return 0;
  89         }
  90         return p >= t && p < t + THREAD_SIZE - size;
  91 }
  92
  93 unsigned long
  94 print_context_stack(struct task_struct *task,
  95                 unsigned long *stack, unsigned long bp,
  96                 const struct stacktrace_ops *ops, void *data,
  97                 unsigned long *end, int *graph)
  98 {
  99         struct stack_frame *frame = (struct stack_frame *)bp;
 100
 101         /*
 102          * If we overflowed the stack into a guard page, jump back to the
 103          * bottom of the usable stack.
 104          */
 105         if ((unsigned long)task_stack_page(task) - (unsigned long)stack <
 106             PAGE_SIZE)
 107                 stack = (unsigned long *)task_stack_page(task);
 108
 109         while (valid_stack_ptr(task, stack, sizeof(*stack), end)) {
 110                 unsigned long addr;
 111
 112                 addr = *stack;
 113                 if (__kernel_text_address(addr)) {
 114                         if ((unsigned long) stack == bp + sizeof(long)) {
 115                                 ops->address(data, addr, 1);
 116                                 frame = frame->next_frame;
 117                                 bp = (unsigned long) frame;
 118                         } else {
 119                                 ops->address(data, addr, 0);
 120                         }
 121                         print_ftrace_graph_addr(addr, data, ops, task, graph);
 122                 }
 123                 stack++;
 124         }
 125         return bp;
 126 }
 127 EXPORT_SYMBOL_GPL(print_context_stack);
 128
 129 unsigned long
 130 print_context_stack_bp(struct task_struct *task,
 131                        unsigned long *stack, unsigned long bp,
 132                        const struct stacktrace_ops *ops, void *data,
 133                        unsigned long *end, int *graph)
 134 {
 135         struct stack_frame *frame = (struct stack_frame *)bp;
 136         unsigned long *ret_addr = &frame->return_address;
 137
 138         while (valid_stack_ptr(task, ret_addr, sizeof(*ret_addr), end)) {
 139                 unsigned long addr = *ret_addr;
 140
 141                 if (!__kernel_text_address(addr))
 142                         break;
 143
 144                 if (ops->address(data, addr, 1))
 145                         break;
 146                 frame = frame->next_frame;
 147                 ret_addr = &frame->return_address;
 148                 print_ftrace_graph_addr(addr, data, ops, task, graph);
 149         }
 150
 151         return (unsigned long)frame;
 152 }
 153 EXPORT_SYMBOL_GPL(print_context_stack_bp);
 154
 155 static int print_trace_stack(void *data, char *name)
 156 {
 157         printk("%s <%s> ", (char *)data, name);
 158         return 0;
 159 }
 160
 161 /*
 162  * Print one address/symbol entries per line.
 163  */
 164 static int print_trace_address(void *data, unsigned long addr, int reliable)
 165 {
 166         touch_nmi_watchdog();
 167         printk_stack_address(addr, reliable, data);
 168         return 0;
 169 }
 170
 171 static const struct stacktrace_ops print_trace_ops = {
 172         .stack                  = print_trace_stack,
 173         .address                = print_trace_address,
 174         .walk_stack             = print_context_stack,
 175 };
 176
 177 void
 178 show_trace_log_lvl(struct task_struct *task, struct pt_regs *regs,
 179                 unsigned long *stack, unsigned long bp, char *log_lvl)
 180 {
 181         printk("%sCall Trace:\n", log_lvl);
 182         dump_trace(task, regs, stack, bp, &print_trace_ops, log_lvl);
 183 }
 184
 185 void show_trace(struct task_struct *task, struct pt_regs *regs,
 186                 unsigned long *stack, unsigned long bp)
 187 {
 188         show_trace_log_lvl(task, regs, stack, bp, "");
 189 }
 190
 191 void show_stack(struct task_struct *task, unsigned long *sp)
 192 {
 193         unsigned long bp = 0;
 194         unsigned long stack;
 195
 196         /*
 197          * Stack frames below this one aren't interesting.  Don't show them
 198          * if we're printing for %current.
 199          */
 200         if (!sp && (!task || task == current)) {
 201                 sp = &stack;
 202                 bp = stack_frame(current, NULL);
 203         }
 204
 205         show_stack_log_lvl(task, NULL, sp, bp, "");
 206 }
 207
 208 static arch_spinlock_t die_lock = __ARCH_SPIN_LOCK_UNLOCKED;
 209 static int die_owner = -1;
 210 static unsigned int die_nest_count;
 211
 212 unsigned long oops_begin(void)
 213 {
 214         int cpu;
 215         unsigned long flags;
 216
 217         oops_enter();
 218
 219         /* racy, but better than risking deadlock. */
 220         raw_local_irq_save(flags);
 221         cpu = smp_processor_id();
 222         if (!arch_spin_trylock(&die_lock)) {
 223                 if (cpu == die_owner)
 224                         /* nested oops. should stop eventually */;
 225                 else
 226                         arch_spin_lock(&die_lock);
 227         }
 228         die_nest_count++;
 229         die_owner = cpu;
 230         console_verbose();
 231         bust_spinlocks(1);
 232         return flags;
 233 }
 234 EXPORT_SYMBOL_GPL(oops_begin);
 235 NOKPROBE_SYMBOL(oops_begin);
 236
 237 void __noreturn rewind_stack_do_exit(int signr);
 238
 239 void oops_end(unsigned long flags, struct pt_regs *regs, int signr)
 240 {
 241         if (regs && kexec_should_crash(current))
 242                 crash_kexec(regs);
 243
 244         bust_spinlocks(0);
 245         die_owner = -1;
 246         add_taint(TAINT_DIE, LOCKDEP_NOW_UNRELIABLE);
 247         die_nest_count--;
 248         if (!die_nest_count)
 249                 /* Nest count reaches zero, release the lock. */
 250                 arch_spin_unlock(&die_lock);
 251         raw_local_irq_restore(flags);
 252         oops_exit();
 253
 254         if (!signr)
 255                 return;
 256         if (in_interrupt())
 257                 panic("Fatal exception in interrupt");
 258         if (panic_on_oops)
 259                 panic("Fatal exception");
 260
 261         /*
 262          * We're not going to return, but we might be on an IST stack or
 263          * have very little stack space left.  Rewind the stack and kill
 264          * the task.
 265          */
 266         rewind_stack_do_exit(signr);
 267 }
 268 NOKPROBE_SYMBOL(oops_end);
 269
 270 int __die(const char *str, struct pt_regs *regs, long err)
 271 {
 272 #ifdef CONFIG_X86_32
 273         unsigned short ss;
 274         unsigned long sp;
 275 #endif
 276         printk(KERN_DEFAULT
 277                "%s: %04lx [#%d]%s%s%s%s\n", str, err & 0xffff, ++die_counter,
 278                IS_ENABLED(CONFIG_PREEMPT) ? " PREEMPT"         : "",
 279                IS_ENABLED(CONFIG_SMP)     ? " SMP"             : "",
 280                debug_pagealloc_enabled()  ? " DEBUG_PAGEALLOC" : "",
 281                IS_ENABLED(CONFIG_KASAN)   ? " KASAN"           : "");
 282
 283         if (notify_die(DIE_OOPS, str, regs, err,
 284                         current->thread.trap_nr, SIGSEGV) == NOTIFY_STOP)
 285                 return 1;
 286
 287         print_modules();
 288         show_regs(regs);
 289 #ifdef CONFIG_X86_32
 290         if (user_mode(regs)) {
 291                 sp = regs->sp;
 292                 ss = regs->ss & 0xffff;
 293         } else {
 294                 sp = kernel_stack_pointer(regs);
 295                 savesegment(ss, ss);
 296         }
 297         printk(KERN_EMERG "EIP: [<%08lx>] ", regs->ip);
 298         print_symbol("%s", regs->ip);
 299         printk(" SS:ESP %04x:%08lx\n", ss, sp);
 300 #else
 301         /* Executive summary in case the oops scrolled away */
 302         printk(KERN_ALERT "RIP ");
 303         printk_address(regs->ip);
 304         printk(" RSP <%016lx>\n", regs->sp);
 305 #endif
 306         return 0;
 307 }
 308 NOKPROBE_SYMBOL(__die);
 309
 310 /*
 311  * This is gone through when something in the kernel has done something bad
 312  * and is about to be terminated:
 313  */
 314 void die(const char *str, struct pt_regs *regs, long err)
 315 {
 316         unsigned long flags = oops_begin();
 317         int sig = SIGSEGV;
 318
 319         if (!user_mode(regs))
 320                 report_bug(regs->ip, regs);
 321
 322         if (__die(str, regs, err))
 323                 sig = 0;
 324         oops_end(flags, regs, sig);
 325 }
 326
 327 static int __init kstack_setup(char *s)
 328 {
 329         ssize_t ret;
 330         unsigned long val;
 331
 332         if (!s)
 333                 return -EINVAL;
 334
 335         ret = kstrtoul(s, 0, &val);
 336         if (ret)
 337                 return ret;
 338         kstack_depth_to_print = val;
 339         return 0;
 340 }
 341 early_param("kstack", kstack_setup);
 342
 343 static int __init code_bytes_setup(char *s)
 344 {
 345         ssize_t ret;
 346         unsigned long val;
 347
 348         if (!s)
 349                 return -EINVAL;
 350
 351         ret = kstrtoul(s, 0, &val);
 352         if (ret)
 353                 return ret;
 354
 355         code_bytes = val;
 356         if (code_bytes > 8192)
 357                 code_bytes = 8192;
 358
 359         return 1;
 360 }
 361 __setup("code_bytes=", code_bytes_setup);