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 #include "dumpstack.h"
  22
  23 int panic_on_unrecovered_nmi;
  24 int panic_on_io_nmi;
  25 unsigned int code_bytes = 64;
  26 int kstack_depth_to_print = 3 * STACKSLOTS_PER_LINE;
  27 static int die_counter;
  28
  29 void printk_address(unsigned long address, int reliable)
  30 {
  31         printk(" [<%p>] %s%pS\n", (void *) address,
  32                         reliable ? "" : "? ", (void *) address);
  33 }
  34
  35 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
  36 static void
  37 print_ftrace_graph_addr(unsigned long addr, void *data,
  38                         const struct stacktrace_ops *ops,
  39                         struct thread_info *tinfo, int *graph)
  40 {
  41         struct task_struct *task = tinfo->task;
  42         unsigned long ret_addr;
  43         int index = task->curr_ret_stack;
  44
  45         if (addr != (unsigned long)return_to_handler)
  46                 return;
  47
  48         if (!task->ret_stack || index < *graph)
  49                 return;
  50
  51         index -= *graph;
  52         ret_addr = task->ret_stack[index].ret;
  53
  54         ops->address(data, ret_addr, 1);
  55
  56         (*graph)++;
  57 }
  58 #else
  59 static inline void
  60 print_ftrace_graph_addr(unsigned long addr, void *data,
  61                         const struct stacktrace_ops *ops,
  62                         struct thread_info *tinfo, int *graph)
  63 { }
  64 #endif
  65
  66 /*
  67  * x86-64 can have up to three kernel stacks:
  68  * process stack
  69  * interrupt stack
  70  * severe exception (double fault, nmi, stack fault, debug, mce) hardware stack
  71  */
  72
  73 static inline int valid_stack_ptr(struct thread_info *tinfo,
  74                         void *p, unsigned int size, void *end)
  75 {
  76         void *t = tinfo;
  77         if (end) {
  78                 if (p < end && p >= (end-THREAD_SIZE))
  79                         return 1;
  80                 else
  81                         return 0;
  82         }
  83         return p > t && p < t + THREAD_SIZE - size;
  84 }
  85
  86 unsigned long
  87 print_context_stack(struct thread_info *tinfo,
  88                 unsigned long *stack, unsigned long bp,
  89                 const struct stacktrace_ops *ops, void *data,
  90                 unsigned long *end, int *graph)
  91 {
  92         struct stack_frame *frame = (struct stack_frame *)bp;
  93
  94         while (valid_stack_ptr(tinfo, stack, sizeof(*stack), end)) {
  95                 unsigned long addr;
  96
  97                 addr = *stack;
  98                 if (__kernel_text_address(addr)) {
  99                         if ((unsigned long) stack == bp + sizeof(long)) {
 100                                 ops->address(data, addr, 1);
 101                                 frame = frame->next_frame;
 102                                 bp = (unsigned long) frame;
 103                         } else {
 104                                 ops->address(data, addr, 0);
 105                         }
 106                         print_ftrace_graph_addr(addr, data, ops, tinfo, graph);
 107                 }
 108                 stack++;
 109         }
 110         return bp;
 111 }
 112 EXPORT_SYMBOL_GPL(print_context_stack);
 113
 114 unsigned long
 115 print_context_stack_bp(struct thread_info *tinfo,
 116                        unsigned long *stack, unsigned long bp,
 117                        const struct stacktrace_ops *ops, void *data,
 118                        unsigned long *end, int *graph)
 119 {
 120         struct stack_frame *frame = (struct stack_frame *)bp;
 121         unsigned long *ret_addr = &frame->return_address;
 122
 123         while (valid_stack_ptr(tinfo, ret_addr, sizeof(*ret_addr), end)) {
 124                 unsigned long addr = *ret_addr;
 125
 126                 if (__kernel_text_address(addr)) {
 127                         ops->address(data, addr, 1);
 128                         frame = frame->next_frame;
 129                         ret_addr = &frame->return_address;
 130                         print_ftrace_graph_addr(addr, data, ops, tinfo, graph);
 131                 }
 132         }
 133         return (unsigned long)frame;
 134 }
 135 EXPORT_SYMBOL_GPL(print_context_stack_bp);
 136
 137
 138 static void
 139 print_trace_warning_symbol(void *data, char *msg, unsigned long symbol)
 140 {
 141         printk(data);
 142         print_symbol(msg, symbol);
 143         printk("\n");
 144 }
 145
 146 static void print_trace_warning(void *data, char *msg)
 147 {
 148         printk("%s%s\n", (char *)data, msg);
 149 }
 150
 151 static int print_trace_stack(void *data, char *name)
 152 {
 153         printk("%s <%s> ", (char *)data, name);
 154         return 0;
 155 }
 156
 157 /*
 158  * Print one address/symbol entries per line.
 159  */
 160 static void print_trace_address(void *data, unsigned long addr, int reliable)
 161 {
 162         touch_nmi_watchdog();
 163         printk(data);
 164         printk_address(addr, reliable);
 165 }
 166
 167 static const struct stacktrace_ops print_trace_ops = {
 168         .warning                = print_trace_warning,
 169         .warning_symbol         = print_trace_warning_symbol,
 170         .stack                  = print_trace_stack,
 171         .address                = print_trace_address,
 172         .walk_stack             = print_context_stack,
 173 };
 174
 175 void
 176 show_trace_log_lvl(struct task_struct *task, struct pt_regs *regs,
 177                 unsigned long *stack, unsigned long bp, char *log_lvl)
 178 {
 179         printk("%sCall Trace:\n", log_lvl);
 180         dump_trace(task, regs, stack, bp, &print_trace_ops, log_lvl);
 181 }
 182
 183 void show_trace(struct task_struct *task, struct pt_regs *regs,
 184                 unsigned long *stack, unsigned long bp)
 185 {
 186         show_trace_log_lvl(task, regs, stack, bp, "");
 187 }
 188
 189 void show_stack(struct task_struct *task, unsigned long *sp)
 190 {
 191         show_stack_log_lvl(task, NULL, sp, 0, "");
 192 }
 193
 194 /*
 195  * The architecture-independent dump_stack generator
 196  */
 197 void dump_stack(void)
 198 {
 199         unsigned long bp = 0;
 200         unsigned long stack;
 201
 202 #ifdef CONFIG_FRAME_POINTER
 203         if (!bp)
 204                 get_bp(bp);
 205 #endif
 206
 207         printk("Pid: %d, comm: %.20s %s %s %.*s\n",
 208                 current->pid, current->comm, print_tainted(),
 209                 init_utsname()->release,
 210                 (int)strcspn(init_utsname()->version, " "),
 211                 init_utsname()->version);
 212         show_trace(NULL, NULL, &stack, bp);
 213 }
 214 EXPORT_SYMBOL(dump_stack);
 215
 216 static arch_spinlock_t die_lock = __ARCH_SPIN_LOCK_UNLOCKED;
 217 static int die_owner = -1;
 218 static unsigned int die_nest_count;
 219
 220 unsigned __kprobes long oops_begin(void)
 221 {
 222         int cpu;
 223         unsigned long flags;
 224
 225         /* notify the hw-branch tracer so it may disable tracing and
 226            add the last trace to the trace buffer -
 227            the earlier this happens, the more useful the trace. */
 228         trace_hw_branch_oops();
 229
 230         oops_enter();
 231
 232         /* racy, but better than risking deadlock. */
 233         raw_local_irq_save(flags);
 234         cpu = smp_processor_id();
 235         if (!arch_spin_trylock(&die_lock)) {
 236                 if (cpu == die_owner)
 237                         /* nested oops. should stop eventually */;
 238                 else
 239                         arch_spin_lock(&die_lock);
 240         }
 241         die_nest_count++;
 242         die_owner = cpu;
 243         console_verbose();
 244         bust_spinlocks(1);
 245         return flags;
 246 }
 247
 248 void __kprobes oops_end(unsigned long flags, struct pt_regs *regs, int signr)
 249 {
 250         if (regs && kexec_should_crash(current))
 251                 crash_kexec(regs);
 252
 253         bust_spinlocks(0);
 254         die_owner = -1;
 255         add_taint(TAINT_DIE);
 256         die_nest_count--;
 257         if (!die_nest_count)
 258                 /* Nest count reaches zero, release the lock. */
 259                 arch_spin_unlock(&die_lock);
 260         raw_local_irq_restore(flags);
 261         oops_exit();
 262
 263         if (!signr)
 264                 return;
 265         if (in_interrupt())
 266                 panic("Fatal exception in interrupt");
 267         if (panic_on_oops)
 268                 panic("Fatal exception");
 269         do_exit(signr);
 270 }
 271
 272 int __kprobes __die(const char *str, struct pt_regs *regs, long err)
 273 {
 274 #ifdef CONFIG_X86_32
 275         unsigned short ss;
 276         unsigned long sp;
 277 #endif
 278         printk(KERN_EMERG "%s: %04lx [#%d] ", str, err & 0xffff, ++die_counter);
 279 #ifdef CONFIG_PREEMPT
 280         printk("PREEMPT ");
 281 #endif
 282 #ifdef CONFIG_SMP
 283         printk("SMP ");
 284 #endif
 285 #ifdef CONFIG_DEBUG_PAGEALLOC
 286         printk("DEBUG_PAGEALLOC");
 287 #endif
 288         printk("\n");
 289         sysfs_printk_last_file();
 290         if (notify_die(DIE_OOPS, str, regs, err,
 291                         current->thread.trap_no, SIGSEGV) == NOTIFY_STOP)
 292                 return 1;
 293
 294         show_registers(regs);
 295 #ifdef CONFIG_X86_32
 296         if (user_mode_vm(regs)) {
 297                 sp = regs->sp;
 298                 ss = regs->ss & 0xffff;
 299         } else {
 300                 sp = kernel_stack_pointer(regs);
 301                 savesegment(ss, ss);
 302         }
 303         printk(KERN_EMERG "EIP: [<%08lx>] ", regs->ip);
 304         print_symbol("%s", regs->ip);
 305         printk(" SS:ESP %04x:%08lx\n", ss, sp);
 306 #else
 307         /* Executive summary in case the oops scrolled away */
 308         printk(KERN_ALERT "RIP ");
 309         printk_address(regs->ip, 1);
 310         printk(" RSP <%016lx>\n", regs->sp);
 311 #endif
 312         return 0;
 313 }
 314
 315 /*
 316  * This is gone through when something in the kernel has done something bad
 317  * and is about to be terminated:
 318  */
 319 void die(const char *str, struct pt_regs *regs, long err)
 320 {
 321         unsigned long flags = oops_begin();
 322         int sig = SIGSEGV;
 323
 324         if (!user_mode_vm(regs))
 325                 report_bug(regs->ip, regs);
 326
 327         if (__die(str, regs, err))
 328                 sig = 0;
 329         oops_end(flags, regs, sig);
 330 }
 331
 332 void notrace __kprobes
 333 die_nmi(char *str, struct pt_regs *regs, int do_panic)
 334 {
 335         unsigned long flags;
 336
 337         if (notify_die(DIE_NMIWATCHDOG, str, regs, 0, 2, SIGINT) == NOTIFY_STOP)
 338                 return;
 339
 340         /*
 341          * We are in trouble anyway, lets at least try
 342          * to get a message out.
 343          */
 344         flags = oops_begin();
 345         printk(KERN_EMERG "%s", str);
 346         printk(" on CPU%d, ip %08lx, registers:\n",
 347                 smp_processor_id(), regs->ip);
 348         show_registers(regs);
 349         oops_end(flags, regs, 0);
 350         if (do_panic || panic_on_oops)
 351                 panic("Non maskable interrupt");
 352         nmi_exit();
 353         local_irq_enable();
 354         do_exit(SIGBUS);
 355 }
 356
 357 static int __init oops_setup(char *s)
 358 {
 359         if (!s)
 360                 return -EINVAL;
 361         if (!strcmp(s, "panic"))
 362                 panic_on_oops = 1;
 363         return 0;
 364 }
 365 early_param("oops", oops_setup);
 366
 367 static int __init kstack_setup(char *s)
 368 {
 369         if (!s)
 370                 return -EINVAL;
 371         kstack_depth_to_print = simple_strtoul(s, NULL, 0);
 372         return 0;
 373 }
 374 early_param("kstack", kstack_setup);
 375
 376 static int __init code_bytes_setup(char *s)
 377 {
 378         code_bytes = simple_strtoul(s, NULL, 0);
 379         if (code_bytes > 8192)
 380                 code_bytes = 8192;
 381
 382         return 1;
 383 }
 384 __setup("code_bytes=", code_bytes_setup);