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 int x86_is_stack_id(int id, char *name)
  37 {
  38         return x86_stack_ids[id - 1] == name;
  39 }
  40
  41 static unsigned long *in_exception_stack(unsigned cpu, unsigned long stack,
  42                                          unsigned *usedp, char **idp)
  43 {
  44         unsigned k;
  45
  46         /*
  47          * Iterate over all exception stacks, and figure out whether
  48          * 'stack' is in one of them:
  49          */
  50         for (k = 0; k < N_EXCEPTION_STACKS; k++) {
  51                 unsigned long end = per_cpu(orig_ist, cpu).ist[k];
  52                 /*
  53                  * Is 'stack' above this exception frame's end?
  54                  * If yes then skip to the next frame.
  55                  */
  56                 if (stack >= end)
  57                         continue;
  58                 /*
  59                  * Is 'stack' above this exception frame's start address?
  60                  * If yes then we found the right frame.
  61                  */
  62                 if (stack >= end - EXCEPTION_STKSZ) {
  63                         /*
  64                          * Make sure we only iterate through an exception
  65                          * stack once. If it comes up for the second time
  66                          * then there's something wrong going on - just
  67                          * break out and return NULL:
  68                          */
  69                         if (*usedp & (1U << k))
  70                                 break;
  71                         *usedp |= 1U << k;
  72                         *idp = x86_stack_ids[k];
  73                         return (unsigned long *)end;
  74                 }
  75                 /*
  76                  * If this is a debug stack, and if it has a larger size than
  77                  * the usual exception stacks, then 'stack' might still
  78                  * be within the lower portion of the debug stack:
  79                  */
  80 #if DEBUG_STKSZ > EXCEPTION_STKSZ
  81                 if (k == DEBUG_STACK - 1 && stack >= end - DEBUG_STKSZ) {
  82                         unsigned j = N_EXCEPTION_STACKS - 1;
  83
  84                         /*
  85                          * Black magic. A large debug stack is composed of
  86                          * multiple exception stack entries, which we
  87                          * iterate through now. Dont look:
  88                          */
  89                         do {
  90                                 ++j;
  91                                 end -= EXCEPTION_STKSZ;
  92                                 x86_stack_ids[j][4] = '1' +
  93                                                 (j - N_EXCEPTION_STACKS);
  94                         } while (stack < end - EXCEPTION_STKSZ);
  95                         if (*usedp & (1U << j))
  96                                 break;
  97                         *usedp |= 1U << j;
  98                         *idp = x86_stack_ids[j];
  99                         return (unsigned long *)end;
 100                 }
 101 #endif
 102         }
 103         return NULL;
 104 }
 105
 106 static inline int
 107 in_irq_stack(unsigned long *stack, unsigned long *irq_stack,
 108              unsigned long *irq_stack_end)
 109 {
 110         return (stack >= irq_stack && stack < irq_stack_end);
 111 }
 112
 113 /*
 114  * We are returning from the irq stack and go to the previous one.
 115  * If the previous stack is also in the irq stack, then bp in the first
 116  * frame of the irq stack points to the previous, interrupted one.
 117  * Otherwise we have another level of indirection: We first save
 118  * the bp of the previous stack, then we switch the stack to the irq one
 119  * and save a new bp that links to the previous one.
 120  * (See save_args())
 121  */
 122 static inline unsigned long
 123 fixup_bp_irq_link(unsigned long bp, unsigned long *stack,
 124                   unsigned long *irq_stack, unsigned long *irq_stack_end)
 125 {
 126 #ifdef CONFIG_FRAME_POINTER
 127         struct stack_frame *frame = (struct stack_frame *)bp;
 128         unsigned long next;
 129
 130         if (!in_irq_stack(stack, irq_stack, irq_stack_end)) {
 131                 if (!probe_kernel_address(&frame->next_frame, next))
 132                         return next;
 133                 else
 134                         WARN_ONCE(1, "Perf: bad frame pointer = %p in "
 135                                   "callchain\n", &frame->next_frame);
 136         }
 137 #endif
 138         return bp;
 139 }
 140
 141 /*
 142  * x86-64 can have up to three kernel stacks:
 143  * process stack
 144  * interrupt stack
 145  * severe exception (double fault, nmi, stack fault, debug, mce) hardware stack
 146  */
 147
 148 void dump_trace(struct task_struct *task, struct pt_regs *regs,
 149                 unsigned long *stack, unsigned long bp,
 150                 const struct stacktrace_ops *ops, void *data)
 151 {
 152         const unsigned cpu = get_cpu();
 153         unsigned long *irq_stack_end =
 154                 (unsigned long *)per_cpu(irq_stack_ptr, cpu);
 155         unsigned used = 0;
 156         struct thread_info *tinfo;
 157         int graph = 0;
 158
 159         if (!task)
 160                 task = current;
 161
 162         if (!stack) {
 163                 unsigned long dummy;
 164                 stack = &dummy;
 165                 if (task && task != current)
 166                         stack = (unsigned long *)task->thread.sp;
 167         }
 168
 169 #ifdef CONFIG_FRAME_POINTER
 170         if (!bp) {
 171                 if (task == current) {
 172                         /* Grab bp right from our regs */
 173                         get_bp(bp);
 174                 } else {
 175                         /* bp is the last reg pushed by switch_to */
 176                         bp = *(unsigned long *) task->thread.sp;
 177                 }
 178         }
 179 #endif
 180
 181         /*
 182          * Print function call entries in all stacks, starting at the
 183          * current stack address. If the stacks consist of nested
 184          * exceptions
 185          */
 186         tinfo = task_thread_info(task);
 187         for (;;) {
 188                 char *id;
 189                 unsigned long *estack_end;
 190                 estack_end = in_exception_stack(cpu, (unsigned long)stack,
 191                                                 &used, &id);
 192
 193                 if (estack_end) {
 194                         if (ops->stack(data, id) < 0)
 195                                 break;
 196
 197                         bp = ops->walk_stack(tinfo, stack, bp, ops,
 198                                              data, estack_end, &graph);
 199                         ops->stack(data, "<EOE>");
 200                         /*
 201                          * We link to the next stack via the
 202                          * second-to-last pointer (index -2 to end) in the
 203                          * exception stack:
 204                          */
 205                         stack = (unsigned long *) estack_end[-2];
 206                         continue;
 207                 }
 208                 if (irq_stack_end) {
 209                         unsigned long *irq_stack;
 210                         irq_stack = irq_stack_end -
 211                                 (IRQ_STACK_SIZE - 64) / sizeof(*irq_stack);
 212
 213                         if (in_irq_stack(stack, irq_stack, irq_stack_end)) {
 214                                 if (ops->stack(data, "IRQ") < 0)
 215                                         break;
 216                                 bp = print_context_stack(tinfo, stack, bp,
 217                                         ops, data, irq_stack_end, &graph);
 218                                 /*
 219                                  * We link to the next stack (which would be
 220                                  * the process stack normally) the last
 221                                  * pointer (index -1 to end) in the IRQ stack:
 222                                  */
 223                                 stack = (unsigned long *) (irq_stack_end[-1]);
 224                                 bp = fixup_bp_irq_link(bp, stack, irq_stack,
 225                                                        irq_stack_end);
 226                                 irq_stack_end = NULL;
 227                                 ops->stack(data, "EOI");
 228                                 continue;
 229                         }
 230                 }
 231                 break;
 232         }
 233
 234         /*
 235          * This handles the process stack:
 236          */
 237         bp = print_context_stack(tinfo, stack, bp, ops, data, NULL, &graph);
 238         put_cpu();
 239 }
 240 EXPORT_SYMBOL(dump_trace);
 241
 242 void
 243 show_stack_log_lvl(struct task_struct *task, struct pt_regs *regs,
 244                    unsigned long *sp, unsigned long bp, char *log_lvl)
 245 {
 246         unsigned long *irq_stack_end;
 247         unsigned long *irq_stack;
 248         unsigned long *stack;
 249         int cpu;
 250         int i;
 251
 252         preempt_disable();
 253         cpu = smp_processor_id();
 254
 255         irq_stack_end   = (unsigned long *)(per_cpu(irq_stack_ptr, cpu));
 256         irq_stack       = (unsigned long *)(per_cpu(irq_stack_ptr, cpu) - IRQ_STACK_SIZE);
 257
 258         /*
 259          * Debugging aid: "show_stack(NULL, NULL);" prints the
 260          * back trace for this cpu:
 261          */
 262         if (sp == NULL) {
 263                 if (task)
 264                         sp = (unsigned long *)task->thread.sp;
 265                 else
 266                         sp = (unsigned long *)&sp;
 267         }
 268
 269         stack = sp;
 270         for (i = 0; i < kstack_depth_to_print; i++) {
 271                 if (stack >= irq_stack && stack <= irq_stack_end) {
 272                         if (stack == irq_stack_end) {
 273                                 stack = (unsigned long *) (irq_stack_end[-1]);
 274                                 printk(" <EOI> ");
 275                         }
 276                 } else {
 277                 if (((long) stack & (THREAD_SIZE-1)) == 0)
 278                         break;
 279                 }
 280                 if (i && ((i % STACKSLOTS_PER_LINE) == 0))
 281                         printk("\n%s", log_lvl);
 282                 printk(" %016lx", *stack++);
 283                 touch_nmi_watchdog();
 284         }
 285         preempt_enable();
 286
 287         printk("\n");
 288         show_trace_log_lvl(task, regs, sp, bp, log_lvl);
 289 }
 290
 291 void show_registers(struct pt_regs *regs)
 292 {
 293         int i;
 294         unsigned long sp;
 295         const int cpu = smp_processor_id();
 296         struct task_struct *cur = current;
 297
 298         sp = regs->sp;
 299         printk("CPU %d ", cpu);
 300         __show_regs(regs, 1);
 301         printk("Process %s (pid: %d, threadinfo %p, task %p)\n",
 302                 cur->comm, cur->pid, task_thread_info(cur), cur);
 303
 304         /*
 305          * When in-kernel, we also print out the stack and code at the
 306          * time of the fault..
 307          */
 308         if (!user_mode(regs)) {
 309                 unsigned int code_prologue = code_bytes * 43 / 64;
 310                 unsigned int code_len = code_bytes;
 311                 unsigned char c;
 312                 u8 *ip;
 313
 314                 printk(KERN_EMERG "Stack:\n");
 315                 show_stack_log_lvl(NULL, regs, (unsigned long *)sp,
 316                                 regs->bp, KERN_EMERG);
 317
 318                 printk(KERN_EMERG "Code: ");
 319
 320                 ip = (u8 *)regs->ip - code_prologue;
 321                 if (ip < (u8 *)PAGE_OFFSET || probe_kernel_address(ip, c)) {
 322                         /* try starting at IP */
 323                         ip = (u8 *)regs->ip;
 324                         code_len = code_len - code_prologue + 1;
 325                 }
 326                 for (i = 0; i < code_len; i++, ip++) {
 327                         if (ip < (u8 *)PAGE_OFFSET ||
 328                                         probe_kernel_address(ip, c)) {
 329                                 printk(" Bad RIP value.");
 330                                 break;
 331                         }
 332                         if (ip == (u8 *)regs->ip)
 333                                 printk("<%02x> ", c);
 334                         else
 335                                 printk("%02x ", c);
 336                 }
 337         }
 338         printk("\n");
 339 }
 340
 341 int is_valid_bugaddr(unsigned long ip)
 342 {
 343         unsigned short ud2;
 344
 345         if (__copy_from_user(&ud2, (const void __user *) ip, sizeof(ud2)))
 346                 return 0;
 347
 348         return ud2 == 0x0b0f;
 349 }