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