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