arch/x86/kernel/dumpstack_64.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/hardirq.h>
   9 #include <linux/kdebug.h>
  10 #include <linux/module.h>
  11 #include <linux/ptrace.h>
  12 #include <linux/kexec.h>
  13 #include <linux/sysfs.h>
  14 #include <linux/bug.h>
  15 #include <linux/nmi.h>
  16
  17 #include <asm/stacktrace.h>
  18
  19
  20 #define N_EXCEPTION_STACKS_END \
  21                 (N_EXCEPTION_STACKS + DEBUG_STKSZ/EXCEPTION_STKSZ - 2)
  22
  23 static char x86_stack_ids[][8] = {
  24                 [ DEBUG_STACK-1                 ]       = "#DB",
  25                 [ NMI_STACK-1                   ]       = "NMI",
  26                 [ DOUBLEFAULT_STACK-1           ]       = "#DF",
  27                 [ STACKFAULT_STACK-1            ]       = "#SS",
  28                 [ MCE_STACK-1                   ]       = "#MC",
  29 #if DEBUG_STKSZ > EXCEPTION_STKSZ
  30                 [ N_EXCEPTION_STACKS ...
  31                   N_EXCEPTION_STACKS_END        ]       = "#DB[?]"
  32 #endif
  33 };
  34
  35 static unsigned long *in_exception_stack(unsigned cpu, unsigned long stack,
  36                                          unsigned *usedp, char **idp)
  37 {
  38         unsigned k;
  39
  40         /*
  41          * Iterate over all exception stacks, and figure out whether
  42          * 'stack' is in one of them:
  43          */
  44         for (k = 0; k < N_EXCEPTION_STACKS; k++) {
  45                 unsigned long end = per_cpu(orig_ist, cpu).ist[k];
  46                 /*
  47                  * Is 'stack' above this exception frame's end?
  48                  * If yes then skip to the next frame.
  49                  */
  50                 if (stack >= end)
  51                         continue;
  52                 /*
  53                  * Is 'stack' above this exception frame's start address?
  54                  * If yes then we found the right frame.
  55                  */
  56                 if (stack >= end - EXCEPTION_STKSZ) {
  57                         /*
  58                          * Make sure we only iterate through an exception
  59                          * stack once. If it comes up for the second time
  60                          * then there's something wrong going on - just
  61                          * break out and return NULL:
  62                          */
  63                         if (*usedp & (1U << k))
  64                                 break;
  65                         *usedp |= 1U << k;
  66                         *idp = x86_stack_ids[k];
  67                         return (unsigned long *)end;
  68                 }
  69                 /*
  70                  * If this is a debug stack, and if it has a larger size than
  71                  * the usual exception stacks, then 'stack' might still
  72                  * be within the lower portion of the debug stack:
  73                  */
  74 #if DEBUG_STKSZ > EXCEPTION_STKSZ
  75                 if (k == DEBUG_STACK - 1 && stack >= end - DEBUG_STKSZ) {
  76                         unsigned j = N_EXCEPTION_STACKS - 1;
  77
  78                         /*
  79                          * Black magic. A large debug stack is composed of
  80                          * multiple exception stack entries, which we
  81                          * iterate through now. Dont look:
  82                          */
  83                         do {
  84                                 ++j;
  85                                 end -= EXCEPTION_STKSZ;
  86                                 x86_stack_ids[j][4] = '1' +
  87                                                 (j - N_EXCEPTION_STACKS);
  88                         } while (stack < end - EXCEPTION_STKSZ);
  89                         if (*usedp & (1U << j))
  90                                 break;
  91                         *usedp |= 1U << j;
  92                         *idp = x86_stack_ids[j];
  93                         return (unsigned long *)end;
  94                 }
  95 #endif
  96         }
  97         return NULL;
  98 }
  99
 100 static inline int
 101 in_irq_stack(unsigned long *stack, unsigned long *irq_stack,
 102              unsigned long *irq_stack_end)
 103 {
 104         return (stack >= irq_stack && stack < irq_stack_end);
 105 }
 106
 107 static const unsigned long irq_stack_size =
 108         (IRQ_STACK_SIZE - 64) / sizeof(unsigned long);
 109
 110 enum stack_type {
 111         STACK_IS_UNKNOWN,
 112         STACK_IS_NORMAL,
 113         STACK_IS_EXCEPTION,
 114         STACK_IS_IRQ,
 115 };
 116
 117 static enum stack_type
 118 analyze_stack(int cpu, struct task_struct *task,
 119               unsigned long *stack, unsigned long **stack_end, char **id)
 120 {
 121         unsigned long *irq_stack;
 122         unsigned long addr;
 123         unsigned used = 0;
 124
 125         addr = ((unsigned long)stack & (~(THREAD_SIZE - 1)));
 126         if ((unsigned long)task_stack_page(task) == addr)
 127                 return STACK_IS_NORMAL;
 128
 129         *stack_end = in_exception_stack(cpu, (unsigned long)stack,
 130                                          &used, id);
 131         if (*stack_end)
 132                 return STACK_IS_EXCEPTION;
 133
 134         *stack_end = (unsigned long *)per_cpu(irq_stack_ptr, cpu);
 135         if (!*stack_end)
 136                 return STACK_IS_UNKNOWN;
 137
 138         irq_stack = *stack_end - irq_stack_size;
 139
 140         if (in_irq_stack(stack, irq_stack, *stack_end))
 141                 return STACK_IS_IRQ;
 142
 143         return STACK_IS_UNKNOWN;
 144 }
 145
 146 /*
 147  * x86-64 can have up to three kernel stacks:
 148  * process stack
 149  * interrupt stack
 150  * severe exception (double fault, nmi, stack fault, debug, mce) hardware stack
 151  */
 152
 153 void dump_trace(struct task_struct *task, struct pt_regs *regs,
 154                 unsigned long *stack, unsigned long bp,
 155                 const struct stacktrace_ops *ops, void *data)
 156 {
 157         const unsigned cpu = get_cpu();
 158         struct thread_info *tinfo;
 159         unsigned long *irq_stack;
 160         unsigned long dummy;
 161         int graph = 0;
 162         int done = 0;
 163
 164         if (!task)
 165                 task = current;
 166
 167         if (!stack) {
 168                 if (regs)
 169                         stack = (unsigned long *)regs->sp;
 170                 else if (task != current)
 171                         stack = (unsigned long *)task->thread.sp;
 172                 else
 173                         stack = &dummy;
 174         }
 175
 176         if (!bp)
 177                 bp = stack_frame(task, regs);
 178         /*
 179          * Print function call entries in all stacks, starting at the
 180          * current stack address. If the stacks consist of nested
 181          * exceptions
 182          */
 183         tinfo = task_thread_info(task);
 184         while (!done) {
 185                 unsigned long *stack_end;
 186                 enum stack_type stype;
 187                 char *id;
 188
 189                 stype = analyze_stack(cpu, task, stack, &stack_end, &id);
 190
 191                 /* Default finish unless specified to continue */
 192                 done = 1;
 193
 194                 switch (stype) {
 195
 196                 /* Break out early if we are on the thread stack */
 197                 case STACK_IS_NORMAL:
 198                         break;
 199
 200                 case STACK_IS_EXCEPTION:
 201
 202                         if (ops->stack(data, id) < 0)
 203                                 break;
 204
 205                         bp = ops->walk_stack(tinfo, stack, bp, ops,
 206                                              data, stack_end, &graph);
 207                         ops->stack(data, "<EOE>");
 208                         /*
 209                          * We link to the next stack via the
 210                          * second-to-last pointer (index -2 to end) in the
 211                          * exception stack:
 212                          */
 213                         stack = (unsigned long *) stack_end[-2];
 214                         done = 0;
 215                         break;
 216
 217                 case STACK_IS_IRQ:
 218
 219                         if (ops->stack(data, "IRQ") < 0)
 220                                 break;
 221                         bp = ops->walk_stack(tinfo, stack, bp,
 222                                      ops, data, stack_end, &graph);
 223                         /*
 224                          * We link to the next stack (which would be
 225                          * the process stack normally) the last
 226                          * pointer (index -1 to end) in the IRQ stack:
 227                          */
 228                         stack = (unsigned long *) (stack_end[-1]);
 229                         irq_stack = stack_end - irq_stack_size;
 230                         ops->stack(data, "EOI");
 231                         done = 0;
 232                         break;
 233
 234                 case STACK_IS_UNKNOWN:
 235                         ops->stack(data, "UNK");
 236                         break;
 237                 }
 238         }
 239
 240         /*
 241          * This handles the process stack:
 242          */
 243         bp = ops->walk_stack(tinfo, stack, bp, ops, data, NULL, &graph);
 244         put_cpu();
 245 }
 246 EXPORT_SYMBOL(dump_trace);
 247
 248 void
 249 show_stack_log_lvl(struct task_struct *task, struct pt_regs *regs,
 250                    unsigned long *sp, unsigned long bp, char *log_lvl)
 251 {
 252         unsigned long *irq_stack_end;
 253         unsigned long *irq_stack;
 254         unsigned long *stack;
 255         int cpu;
 256         int i;
 257
 258         preempt_disable();
 259         cpu = smp_processor_id();
 260
 261         irq_stack_end   = (unsigned long *)(per_cpu(irq_stack_ptr, cpu));
 262         irq_stack       = (unsigned long *)(per_cpu(irq_stack_ptr, cpu) - IRQ_STACK_SIZE);
 263
 264         /*
 265          * Debugging aid: "show_stack(NULL, NULL);" prints the
 266          * back trace for this cpu:
 267          */
 268         if (sp == NULL) {
 269                 if (task)
 270                         sp = (unsigned long *)task->thread.sp;
 271                 else
 272                         sp = (unsigned long *)&sp;
 273         }
 274
 275         stack = sp;
 276         for (i = 0; i < kstack_depth_to_print; i++) {
 277                 if (stack >= irq_stack && stack <= irq_stack_end) {
 278                         if (stack == irq_stack_end) {
 279                                 stack = (unsigned long *) (irq_stack_end[-1]);
 280                                 pr_cont(" <EOI> ");
 281                         }
 282                 } else {
 283                 if (((long) stack & (THREAD_SIZE-1)) == 0)
 284                         break;
 285                 }
 286                 if (i && ((i % STACKSLOTS_PER_LINE) == 0))
 287                         pr_cont("\n");
 288                 pr_cont(" %016lx", *stack++);
 289                 touch_nmi_watchdog();
 290         }
 291         preempt_enable();
 292
 293         pr_cont("\n");
 294         show_trace_log_lvl(task, regs, sp, bp, log_lvl);
 295 }
 296
 297 void show_regs(struct pt_regs *regs)
 298 {
 299         int i;
 300         unsigned long sp;
 301
 302         sp = regs->sp;
 303         show_regs_print_info(KERN_DEFAULT);
 304         __show_regs(regs, 1);
 305
 306         /*
 307          * When in-kernel, we also print out the stack and code at the
 308          * time of the fault..
 309          */
 310         if (!user_mode(regs)) {
 311                 unsigned int code_prologue = code_bytes * 43 / 64;
 312                 unsigned int code_len = code_bytes;
 313                 unsigned char c;
 314                 u8 *ip;
 315
 316                 printk(KERN_DEFAULT "Stack:\n");
 317                 show_stack_log_lvl(NULL, regs, (unsigned long *)sp,
 318                                    0, KERN_DEFAULT);
 319
 320                 printk(KERN_DEFAULT "Code: ");
 321
 322                 ip = (u8 *)regs->ip - code_prologue;
 323                 if (ip < (u8 *)PAGE_OFFSET || probe_kernel_address(ip, c)) {
 324                         /* try starting at IP */
 325                         ip = (u8 *)regs->ip;
 326                         code_len = code_len - code_prologue + 1;
 327                 }
 328                 for (i = 0; i < code_len; i++, ip++) {
 329                         if (ip < (u8 *)PAGE_OFFSET ||
 330                                         probe_kernel_address(ip, c)) {
 331                                 pr_cont(" Bad RIP value.");
 332                                 break;
 333                         }
 334                         if (ip == (u8 *)regs->ip)
 335                                 pr_cont("<%02x> ", c);
 336                         else
 337                                 pr_cont("%02x ", c);
 338                 }
 339         }
 340         pr_cont("\n");
 341 }
 342
 343 int is_valid_bugaddr(unsigned long ip)
 344 {
 345         unsigned short ud2;
 346
 347         if (__copy_from_user(&ud2, (const void __user *) ip, sizeof(ud2)))
 348                 return 0;
 349
 350         return ud2 == 0x0b0f;
 351 }