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 /*
 108  * We are returning from the irq stack and go to the previous one.
 109  * If the previous stack is also in the irq stack, then bp in the first
 110  * frame of the irq stack points to the previous, interrupted one.
 111  * Otherwise we have another level of indirection: We first save
 112  * the bp of the previous stack, then we switch the stack to the irq one
 113  * and save a new bp that links to the previous one.
 114  * (See save_args())
 115  */
 116 static inline unsigned long
 117 fixup_bp_irq_link(unsigned long bp, unsigned long *stack,
 118                   unsigned long *irq_stack, unsigned long *irq_stack_end)
 119 {
 120 #ifdef CONFIG_FRAME_POINTER
 121         struct stack_frame *frame = (struct stack_frame *)bp;
 122         unsigned long next;
 123
 124         if (!in_irq_stack(stack, irq_stack, irq_stack_end)) {
 125                 if (!probe_kernel_address(&frame->next_frame, next))
 126                         return next;
 127                 else
 128                         WARN_ONCE(1, "Perf: bad frame pointer = %p in "
 129                                   "callchain\n", &frame->next_frame);
 130         }
 131 #endif
 132         return bp;
 133 }
 134
 135 /*
 136  * x86-64 can have up to three kernel stacks:
 137  * process stack
 138  * interrupt stack
 139  * severe exception (double fault, nmi, stack fault, debug, mce) hardware stack
 140  */
 141
 142 void dump_trace(struct task_struct *task, struct pt_regs *regs,
 143                 unsigned long *stack, unsigned long bp,
 144                 const struct stacktrace_ops *ops, void *data)
 145 {
 146         const unsigned cpu = get_cpu();
 147         unsigned long *irq_stack_end =
 148                 (unsigned long *)per_cpu(irq_stack_ptr, cpu);
 149         unsigned used = 0;
 150         struct thread_info *tinfo;
 151         int graph = 0;
 152         unsigned long dummy;
 153
 154         if (!task)
 155                 task = current;
 156
 157         if (!stack) {
 158                 if (regs)
 159                         stack = (unsigned long *)regs->sp;
 160                 else if (task && task != current)
 161                         stack = (unsigned long *)task->thread.sp;
 162                 else
 163                         stack = &dummy;
 164         }
 165
 166         if (!bp)
 167                 bp = stack_frame(task, regs);
 168         /*
 169          * Print function call entries in all stacks, starting at the
 170          * current stack address. If the stacks consist of nested
 171          * exceptions
 172          */
 173         tinfo = task_thread_info(task);
 174         for (;;) {
 175                 char *id;
 176                 unsigned long *estack_end;
 177                 estack_end = in_exception_stack(cpu, (unsigned long)stack,
 178                                                 &used, &id);
 179
 180                 if (estack_end) {
 181                         if (ops->stack(data, id) < 0)
 182                                 break;
 183
 184                         bp = ops->walk_stack(tinfo, stack, bp, ops,
 185                                              data, estack_end, &graph);
 186                         ops->stack(data, "<EOE>");
 187                         /*
 188                          * We link to the next stack via the
 189                          * second-to-last pointer (index -2 to end) in the
 190                          * exception stack:
 191                          */
 192                         stack = (unsigned long *) estack_end[-2];
 193                         continue;
 194                 }
 195                 if (irq_stack_end) {
 196                         unsigned long *irq_stack;
 197                         irq_stack = irq_stack_end -
 198                                 (IRQ_STACK_SIZE - 64) / sizeof(*irq_stack);
 199
 200                         if (in_irq_stack(stack, irq_stack, irq_stack_end)) {
 201                                 if (ops->stack(data, "IRQ") < 0)
 202                                         break;
 203                                 bp = ops->walk_stack(tinfo, stack, bp,
 204                                         ops, data, irq_stack_end, &graph);
 205                                 /*
 206                                  * We link to the next stack (which would be
 207                                  * the process stack normally) the last
 208                                  * pointer (index -1 to end) in the IRQ stack:
 209                                  */
 210                                 stack = (unsigned long *) (irq_stack_end[-1]);
 211                                 bp = fixup_bp_irq_link(bp, stack, irq_stack,
 212                                                        irq_stack_end);
 213                                 irq_stack_end = NULL;
 214                                 ops->stack(data, "EOI");
 215                                 continue;
 216                         }
 217                 }
 218                 break;
 219         }
 220
 221         /*
 222          * This handles the process stack:
 223          */
 224         bp = ops->walk_stack(tinfo, stack, bp, ops, data, NULL, &graph);
 225         put_cpu();
 226 }
 227 EXPORT_SYMBOL(dump_trace);
 228
 229 void
 230 show_stack_log_lvl(struct task_struct *task, struct pt_regs *regs,
 231                    unsigned long *sp, unsigned long bp, char *log_lvl)
 232 {
 233         unsigned long *irq_stack_end;
 234         unsigned long *irq_stack;
 235         unsigned long *stack;
 236         int cpu;
 237         int i;
 238
 239         preempt_disable();
 240         cpu = smp_processor_id();
 241
 242         irq_stack_end   = (unsigned long *)(per_cpu(irq_stack_ptr, cpu));
 243         irq_stack       = (unsigned long *)(per_cpu(irq_stack_ptr, cpu) - IRQ_STACK_SIZE);
 244
 245         /*
 246          * Debugging aid: "show_stack(NULL, NULL);" prints the
 247          * back trace for this cpu:
 248          */
 249         if (sp == NULL) {
 250                 if (task)
 251                         sp = (unsigned long *)task->thread.sp;
 252                 else
 253                         sp = (unsigned long *)&sp;
 254         }
 255
 256         stack = sp;
 257         for (i = 0; i < kstack_depth_to_print; i++) {
 258                 if (stack >= irq_stack && stack <= irq_stack_end) {
 259                         if (stack == irq_stack_end) {
 260                                 stack = (unsigned long *) (irq_stack_end[-1]);
 261                                 printk(KERN_CONT " <EOI> ");
 262                         }
 263                 } else {
 264                 if (((long) stack & (THREAD_SIZE-1)) == 0)
 265                         break;
 266                 }
 267                 if (i && ((i % STACKSLOTS_PER_LINE) == 0))
 268                         printk(KERN_CONT "\n");
 269                 printk(KERN_CONT " %016lx", *stack++);
 270                 touch_nmi_watchdog();
 271         }
 272         preempt_enable();
 273
 274         printk(KERN_CONT "\n");
 275         show_trace_log_lvl(task, regs, sp, bp, log_lvl);
 276 }
 277
 278 void show_registers(struct pt_regs *regs)
 279 {
 280         int i;
 281         unsigned long sp;
 282         const int cpu = smp_processor_id();
 283         struct task_struct *cur = current;
 284
 285         sp = regs->sp;
 286         printk("CPU %d ", cpu);
 287         print_modules();
 288         __show_regs(regs, 1);
 289         printk("Process %s (pid: %d, threadinfo %p, task %p)\n",
 290                 cur->comm, cur->pid, task_thread_info(cur), cur);
 291
 292         /*
 293          * When in-kernel, we also print out the stack and code at the
 294          * time of the fault..
 295          */
 296         if (!user_mode(regs)) {
 297                 unsigned int code_prologue = code_bytes * 43 / 64;
 298                 unsigned int code_len = code_bytes;
 299                 unsigned char c;
 300                 u8 *ip;
 301
 302                 printk(KERN_EMERG "Stack:\n");
 303                 show_stack_log_lvl(NULL, regs, (unsigned long *)sp,
 304                                    0, KERN_EMERG);
 305
 306                 printk(KERN_EMERG "Code: ");
 307
 308                 ip = (u8 *)regs->ip - code_prologue;
 309                 if (ip < (u8 *)PAGE_OFFSET || probe_kernel_address(ip, c)) {
 310                         /* try starting at IP */
 311                         ip = (u8 *)regs->ip;
 312                         code_len = code_len - code_prologue + 1;
 313                 }
 314                 for (i = 0; i < code_len; i++, ip++) {
 315                         if (ip < (u8 *)PAGE_OFFSET ||
 316                                         probe_kernel_address(ip, c)) {
 317                                 printk(" Bad RIP value.");
 318                                 break;
 319                         }
 320                         if (ip == (u8 *)regs->ip)
 321                                 printk("<%02x> ", c);
 322                         else
 323                                 printk("%02x ", c);
 324                 }
 325         }
 326         printk("\n");
 327 }
 328
 329 int is_valid_bugaddr(unsigned long ip)
 330 {
 331         unsigned short ud2;
 332
 333         if (__copy_from_user(&ud2, (const void __user *) ip, sizeof(ud2)))
 334                 return 0;
 335
 336         return ud2 == 0x0b0f;
 337 }