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/sched/debug.h>
  14 #include <linux/sched/task_stack.h>
  15 #include <linux/ftrace.h>
  16 #include <linux/kexec.h>
  17 #include <linux/bug.h>
  18 #include <linux/nmi.h>
  19 #include <linux/sysfs.h>
  20
  21 #include <asm/cpu_entry_area.h>
  22 #include <asm/stacktrace.h>
  23 #include <asm/unwind.h>
  24
  25 int panic_on_unrecovered_nmi;
  26 int panic_on_io_nmi;
  27 unsigned int code_bytes = 64;
  28 static int die_counter;
  29
  30 bool in_task_stack(unsigned long *stack, struct task_struct *task,
  31                    struct stack_info *info)
  32 {
  33         unsigned long *begin = task_stack_page(task);
  34         unsigned long *end   = task_stack_page(task) + THREAD_SIZE;
  35
  36         if (stack < begin || stack >= end)
  37                 return false;
  38
  39         info->type      = STACK_TYPE_TASK;
  40         info->begin     = begin;
  41         info->end       = end;
  42         info->next_sp   = NULL;
  43
  44         return true;
  45 }
  46
  47 bool in_entry_stack(unsigned long *stack, struct stack_info *info)
  48 {
  49         struct entry_stack *ss = cpu_entry_stack(smp_processor_id());
  50
  51         void *begin = ss;
  52         void *end = ss + 1;
  53
  54         if ((void *)stack < begin || (void *)stack >= end)
  55                 return false;
  56
  57         info->type      = STACK_TYPE_ENTRY;
  58         info->begin     = begin;
  59         info->end       = end;
  60         info->next_sp   = NULL;
  61
  62         return true;
  63 }
  64
  65 static void printk_stack_address(unsigned long address, int reliable,
  66                                  char *log_lvl)
  67 {
  68         touch_nmi_watchdog();
  69         printk("%s %s%pB\n", log_lvl, reliable ? "" : "? ", (void *)address);
  70 }
  71
  72 void show_iret_regs(struct pt_regs *regs)
  73 {
  74         printk(KERN_DEFAULT "RIP: %04x:%pS\n", (int)regs->cs, (void *)regs->ip);
  75         printk(KERN_DEFAULT "RSP: %04x:%016lx EFLAGS: %08lx", (int)regs->ss,
  76                 regs->sp, regs->flags);
  77 }
  78
  79 static void show_regs_safe(struct stack_info *info, struct pt_regs *regs)
  80 {
  81         if (on_stack(info, regs, sizeof(*regs)))
  82                 __show_regs(regs, 0);
  83         else if (on_stack(info, (void *)regs + IRET_FRAME_OFFSET,
  84                           IRET_FRAME_SIZE)) {
  85                 /*
  86                  * When an interrupt or exception occurs in entry code, the
  87                  * full pt_regs might not have been saved yet.  In that case
  88                  * just print the iret frame.
  89                  */
  90                 show_iret_regs(regs);
  91         }
  92 }
  93
  94 void show_trace_log_lvl(struct task_struct *task, struct pt_regs *regs,
  95                         unsigned long *stack, char *log_lvl)
  96 {
  97         struct unwind_state state;
  98         struct stack_info stack_info = {0};
  99         unsigned long visit_mask = 0;
 100         int graph_idx = 0;
 101
 102         printk("%sCall Trace:\n", log_lvl);
 103
 104         unwind_start(&state, task, regs, stack);
 105         stack = stack ? : get_stack_pointer(task, regs);
 106
 107         /*
 108          * Iterate through the stacks, starting with the current stack pointer.
 109          * Each stack has a pointer to the next one.
 110          *
 111          * x86-64 can have several stacks:
 112          * - task stack
 113          * - interrupt stack
 114          * - HW exception stacks (double fault, nmi, debug, mce)
 115          * - entry stack
 116          *
 117          * x86-32 can have up to four stacks:
 118          * - task stack
 119          * - softirq stack
 120          * - hardirq stack
 121          * - entry stack
 122          */
 123         for (regs = NULL; stack; stack = PTR_ALIGN(stack_info.next_sp, sizeof(long))) {
 124                 const char *stack_name;
 125
 126                 if (get_stack_info(stack, task, &stack_info, &visit_mask)) {
 127                         /*
 128                          * We weren't on a valid stack.  It's possible that
 129                          * we overflowed a valid stack into a guard page.
 130                          * See if the next page up is valid so that we can
 131                          * generate some kind of backtrace if this happens.
 132                          */
 133                         stack = (unsigned long *)PAGE_ALIGN((unsigned long)stack);
 134                         if (get_stack_info(stack, task, &stack_info, &visit_mask))
 135                                 break;
 136                 }
 137
 138                 stack_name = stack_type_name(stack_info.type);
 139                 if (stack_name)
 140                         printk("%s <%s>\n", log_lvl, stack_name);
 141
 142                 if (regs)
 143                         show_regs_safe(&stack_info, regs);
 144
 145                 /*
 146                  * Scan the stack, printing any text addresses we find.  At the
 147                  * same time, follow proper stack frames with the unwinder.
 148                  *
 149                  * Addresses found during the scan which are not reported by
 150                  * the unwinder are considered to be additional clues which are
 151                  * sometimes useful for debugging and are prefixed with '?'.
 152                  * This also serves as a failsafe option in case the unwinder
 153                  * goes off in the weeds.
 154                  */
 155                 for (; stack < stack_info.end; stack++) {
 156                         unsigned long real_addr;
 157                         int reliable = 0;
 158                         unsigned long addr = READ_ONCE_NOCHECK(*stack);
 159                         unsigned long *ret_addr_p =
 160                                 unwind_get_return_address_ptr(&state);
 161
 162                         if (!__kernel_text_address(addr))
 163                                 continue;
 164
 165                         /*
 166                          * Don't print regs->ip again if it was already printed
 167                          * by show_regs_safe() below.
 168                          */
 169                         if (regs && stack == &regs->ip)
 170                                 goto next;
 171
 172                         if (stack == ret_addr_p)
 173                                 reliable = 1;
 174
 175                         /*
 176                          * When function graph tracing is enabled for a
 177                          * function, its return address on the stack is
 178                          * replaced with the address of an ftrace handler
 179                          * (return_to_handler).  In that case, before printing
 180                          * the "real" address, we want to print the handler
 181                          * address as an "unreliable" hint that function graph
 182                          * tracing was involved.
 183                          */
 184                         real_addr = ftrace_graph_ret_addr(task, &graph_idx,
 185                                                           addr, stack);
 186                         if (real_addr != addr)
 187                                 printk_stack_address(addr, 0, log_lvl);
 188                         printk_stack_address(real_addr, reliable, log_lvl);
 189
 190                         if (!reliable)
 191                                 continue;
 192
 193 next:
 194                         /*
 195                          * Get the next frame from the unwinder.  No need to
 196                          * check for an error: if anything goes wrong, the rest
 197                          * of the addresses will just be printed as unreliable.
 198                          */
 199                         unwind_next_frame(&state);
 200
 201                         /* if the frame has entry regs, print them */
 202                         regs = unwind_get_entry_regs(&state);
 203                         if (regs)
 204                                 show_regs_safe(&stack_info, regs);
 205                 }
 206
 207                 if (stack_name)
 208                         printk("%s </%s>\n", log_lvl, stack_name);
 209         }
 210 }
 211
 212 void show_stack(struct task_struct *task, unsigned long *sp)
 213 {
 214         task = task ? : current;
 215
 216         /*
 217          * Stack frames below this one aren't interesting.  Don't show them
 218          * if we're printing for %current.
 219          */
 220         if (!sp && task == current)
 221                 sp = get_stack_pointer(current, NULL);
 222
 223         show_trace_log_lvl(task, NULL, sp, KERN_DEFAULT);
 224 }
 225
 226 void show_stack_regs(struct pt_regs *regs)
 227 {
 228         show_trace_log_lvl(current, regs, NULL, KERN_DEFAULT);
 229 }
 230
 231 static arch_spinlock_t die_lock = __ARCH_SPIN_LOCK_UNLOCKED;
 232 static int die_owner = -1;
 233 static unsigned int die_nest_count;
 234
 235 unsigned long oops_begin(void)
 236 {
 237         int cpu;
 238         unsigned long flags;
 239
 240         oops_enter();
 241
 242         /* racy, but better than risking deadlock. */
 243         raw_local_irq_save(flags);
 244         cpu = smp_processor_id();
 245         if (!arch_spin_trylock(&die_lock)) {
 246                 if (cpu == die_owner)
 247                         /* nested oops. should stop eventually */;
 248                 else
 249                         arch_spin_lock(&die_lock);
 250         }
 251         die_nest_count++;
 252         die_owner = cpu;
 253         console_verbose();
 254         bust_spinlocks(1);
 255         return flags;
 256 }
 257 EXPORT_SYMBOL_GPL(oops_begin);
 258 NOKPROBE_SYMBOL(oops_begin);
 259
 260 void __noreturn rewind_stack_do_exit(int signr);
 261
 262 void oops_end(unsigned long flags, struct pt_regs *regs, int signr)
 263 {
 264         if (regs && kexec_should_crash(current))
 265                 crash_kexec(regs);
 266
 267         bust_spinlocks(0);
 268         die_owner = -1;
 269         add_taint(TAINT_DIE, LOCKDEP_NOW_UNRELIABLE);
 270         die_nest_count--;
 271         if (!die_nest_count)
 272                 /* Nest count reaches zero, release the lock. */
 273                 arch_spin_unlock(&die_lock);
 274         raw_local_irq_restore(flags);
 275         oops_exit();
 276
 277         if (!signr)
 278                 return;
 279         if (in_interrupt())
 280                 panic("Fatal exception in interrupt");
 281         if (panic_on_oops)
 282                 panic("Fatal exception");
 283
 284         /*
 285          * We're not going to return, but we might be on an IST stack or
 286          * have very little stack space left.  Rewind the stack and kill
 287          * the task.
 288          */
 289         rewind_stack_do_exit(signr);
 290 }
 291 NOKPROBE_SYMBOL(oops_end);
 292
 293 int __die(const char *str, struct pt_regs *regs, long err)
 294 {
 295 #ifdef CONFIG_X86_32
 296         unsigned short ss;
 297         unsigned long sp;
 298 #endif
 299         printk(KERN_DEFAULT
 300                "%s: %04lx [#%d]%s%s%s%s\n", str, err & 0xffff, ++die_counter,
 301                IS_ENABLED(CONFIG_PREEMPT) ? " PREEMPT"         : "",
 302                IS_ENABLED(CONFIG_SMP)     ? " SMP"             : "",
 303                debug_pagealloc_enabled()  ? " DEBUG_PAGEALLOC" : "",
 304                IS_ENABLED(CONFIG_KASAN)   ? " KASAN"           : "");
 305
 306         if (notify_die(DIE_OOPS, str, regs, err,
 307                         current->thread.trap_nr, SIGSEGV) == NOTIFY_STOP)
 308                 return 1;
 309
 310         print_modules();
 311         show_regs(regs);
 312 #ifdef CONFIG_X86_32
 313         if (user_mode(regs)) {
 314                 sp = regs->sp;
 315                 ss = regs->ss;
 316         } else {
 317                 sp = kernel_stack_pointer(regs);
 318                 savesegment(ss, ss);
 319         }
 320         printk(KERN_EMERG "EIP: %pS SS:ESP: %04x:%08lx\n",
 321                (void *)regs->ip, ss, sp);
 322 #else
 323         /* Executive summary in case the oops scrolled away */
 324         printk(KERN_ALERT "RIP: %pS RSP: %016lx\n", (void *)regs->ip, regs->sp);
 325 #endif
 326         return 0;
 327 }
 328 NOKPROBE_SYMBOL(__die);
 329
 330 /*
 331  * This is gone through when something in the kernel has done something bad
 332  * and is about to be terminated:
 333  */
 334 void die(const char *str, struct pt_regs *regs, long err)
 335 {
 336         unsigned long flags = oops_begin();
 337         int sig = SIGSEGV;
 338
 339         if (__die(str, regs, err))
 340                 sig = 0;
 341         oops_end(flags, regs, sig);
 342 }
 343
 344 static int __init code_bytes_setup(char *s)
 345 {
 346         ssize_t ret;
 347         unsigned long val;
 348
 349         if (!s)
 350                 return -EINVAL;
 351
 352         ret = kstrtoul(s, 0, &val);
 353         if (ret)
 354                 return ret;
 355
 356         code_bytes = val;
 357         if (code_bytes > 8192)
 358                 code_bytes = 8192;
 359
 360         return 1;
 361 }
 362 __setup("code_bytes=", code_bytes_setup);