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 #include <linux/kasan.h>
  21
  22 #include <asm/cpu_entry_area.h>
  23 #include <asm/stacktrace.h>
  24 #include <asm/unwind.h>
  25
  26 int panic_on_unrecovered_nmi;
  27 int panic_on_io_nmi;
  28 unsigned int code_bytes = 64;
  29 static int die_counter;
  30
  31 bool in_task_stack(unsigned long *stack, struct task_struct *task,
  32                    struct stack_info *info)
  33 {
  34         unsigned long *begin = task_stack_page(task);
  35         unsigned long *end   = task_stack_page(task) + THREAD_SIZE;
  36
  37         if (stack < begin || stack >= end)
  38                 return false;
  39
  40         info->type      = STACK_TYPE_TASK;
  41         info->begin     = begin;
  42         info->end       = end;
  43         info->next_sp   = NULL;
  44
  45         return true;
  46 }
  47
  48 bool in_entry_stack(unsigned long *stack, struct stack_info *info)
  49 {
  50         struct entry_stack *ss = cpu_entry_stack(smp_processor_id());
  51
  52         void *begin = ss;
  53         void *end = ss + 1;
  54
  55         if ((void *)stack < begin || (void *)stack >= end)
  56                 return false;
  57
  58         info->type      = STACK_TYPE_ENTRY;
  59         info->begin     = begin;
  60         info->end       = end;
  61         info->next_sp   = NULL;
  62
  63         return true;
  64 }
  65
  66 static void printk_stack_address(unsigned long address, int reliable,
  67                                  char *log_lvl)
  68 {
  69         touch_nmi_watchdog();
  70         printk("%s %s%pB\n", log_lvl, reliable ? "" : "? ", (void *)address);
  71 }
  72
  73 void show_iret_regs(struct pt_regs *regs)
  74 {
  75         printk(KERN_DEFAULT "RIP: %04x:%pS\n", (int)regs->cs, (void *)regs->ip);
  76         printk(KERN_DEFAULT "RSP: %04x:%016lx EFLAGS: %08lx", (int)regs->ss,
  77                 regs->sp, regs->flags);
  78 }
  79
  80 static void show_regs_if_on_stack(struct stack_info *info, struct pt_regs *regs,
  81                                   bool partial)
  82 {
  83         /*
  84          * These on_stack() checks aren't strictly necessary: the unwind code
  85          * has already validated the 'regs' pointer.  The checks are done for
  86          * ordering reasons: if the registers are on the next stack, we don't
  87          * want to print them out yet.  Otherwise they'll be shown as part of
  88          * the wrong stack.  Later, when show_trace_log_lvl() switches to the
  89          * next stack, this function will be called again with the same regs so
  90          * they can be printed in the right context.
  91          */
  92         if (!partial && on_stack(info, regs, sizeof(*regs))) {
  93                 __show_regs(regs, 0);
  94
  95         } else if (partial && on_stack(info, (void *)regs + IRET_FRAME_OFFSET,
  96                                        IRET_FRAME_SIZE)) {
  97                 /*
  98                  * When an interrupt or exception occurs in entry code, the
  99                  * full pt_regs might not have been saved yet.  In that case
 100                  * just print the iret frame.
 101                  */
 102                 show_iret_regs(regs);
 103         }
 104 }
 105
 106 void show_trace_log_lvl(struct task_struct *task, struct pt_regs *regs,
 107                         unsigned long *stack, char *log_lvl)
 108 {
 109         struct unwind_state state;
 110         struct stack_info stack_info = {0};
 111         unsigned long visit_mask = 0;
 112         int graph_idx = 0;
 113         bool partial = false;
 114
 115         printk("%sCall Trace:\n", log_lvl);
 116
 117         unwind_start(&state, task, regs, stack);
 118         stack = stack ? : get_stack_pointer(task, regs);
 119         regs = unwind_get_entry_regs(&state, &partial);
 120
 121         /*
 122          * Iterate through the stacks, starting with the current stack pointer.
 123          * Each stack has a pointer to the next one.
 124          *
 125          * x86-64 can have several stacks:
 126          * - task stack
 127          * - interrupt stack
 128          * - HW exception stacks (double fault, nmi, debug, mce)
 129          * - entry stack
 130          *
 131          * x86-32 can have up to four stacks:
 132          * - task stack
 133          * - softirq stack
 134          * - hardirq stack
 135          * - entry stack
 136          */
 137         for ( ; stack; stack = PTR_ALIGN(stack_info.next_sp, sizeof(long))) {
 138                 const char *stack_name;
 139
 140                 if (get_stack_info(stack, task, &stack_info, &visit_mask)) {
 141                         /*
 142                          * We weren't on a valid stack.  It's possible that
 143                          * we overflowed a valid stack into a guard page.
 144                          * See if the next page up is valid so that we can
 145                          * generate some kind of backtrace if this happens.
 146                          */
 147                         stack = (unsigned long *)PAGE_ALIGN((unsigned long)stack);
 148                         if (get_stack_info(stack, task, &stack_info, &visit_mask))
 149                                 break;
 150                 }
 151
 152                 stack_name = stack_type_name(stack_info.type);
 153                 if (stack_name)
 154                         printk("%s <%s>\n", log_lvl, stack_name);
 155
 156                 if (regs)
 157                         show_regs_if_on_stack(&stack_info, regs, partial);
 158
 159                 /*
 160                  * Scan the stack, printing any text addresses we find.  At the
 161                  * same time, follow proper stack frames with the unwinder.
 162                  *
 163                  * Addresses found during the scan which are not reported by
 164                  * the unwinder are considered to be additional clues which are
 165                  * sometimes useful for debugging and are prefixed with '?'.
 166                  * This also serves as a failsafe option in case the unwinder
 167                  * goes off in the weeds.
 168                  */
 169                 for (; stack < stack_info.end; stack++) {
 170                         unsigned long real_addr;
 171                         int reliable = 0;
 172                         unsigned long addr = READ_ONCE_NOCHECK(*stack);
 173                         unsigned long *ret_addr_p =
 174                                 unwind_get_return_address_ptr(&state);
 175
 176                         if (!__kernel_text_address(addr))
 177                                 continue;
 178
 179                         /*
 180                          * Don't print regs->ip again if it was already printed
 181                          * by show_regs_if_on_stack().
 182                          */
 183                         if (regs && stack == &regs->ip)
 184                                 goto next;
 185
 186                         if (stack == ret_addr_p)
 187                                 reliable = 1;
 188
 189                         /*
 190                          * When function graph tracing is enabled for a
 191                          * function, its return address on the stack is
 192                          * replaced with the address of an ftrace handler
 193                          * (return_to_handler).  In that case, before printing
 194                          * the "real" address, we want to print the handler
 195                          * address as an "unreliable" hint that function graph
 196                          * tracing was involved.
 197                          */
 198                         real_addr = ftrace_graph_ret_addr(task, &graph_idx,
 199                                                           addr, stack);
 200                         if (real_addr != addr)
 201                                 printk_stack_address(addr, 0, log_lvl);
 202                         printk_stack_address(real_addr, reliable, log_lvl);
 203
 204                         if (!reliable)
 205                                 continue;
 206
 207 next:
 208                         /*
 209                          * Get the next frame from the unwinder.  No need to
 210                          * check for an error: if anything goes wrong, the rest
 211                          * of the addresses will just be printed as unreliable.
 212                          */
 213                         unwind_next_frame(&state);
 214
 215                         /* if the frame has entry regs, print them */
 216                         regs = unwind_get_entry_regs(&state, &partial);
 217                         if (regs)
 218                                 show_regs_if_on_stack(&stack_info, regs, partial);
 219                 }
 220
 221                 if (stack_name)
 222                         printk("%s </%s>\n", log_lvl, stack_name);
 223         }
 224 }
 225
 226 void show_stack(struct task_struct *task, unsigned long *sp)
 227 {
 228         task = task ? : current;
 229
 230         /*
 231          * Stack frames below this one aren't interesting.  Don't show them
 232          * if we're printing for %current.
 233          */
 234         if (!sp && task == current)
 235                 sp = get_stack_pointer(current, NULL);
 236
 237         show_trace_log_lvl(task, NULL, sp, KERN_DEFAULT);
 238 }
 239
 240 void show_stack_regs(struct pt_regs *regs)
 241 {
 242         show_trace_log_lvl(current, regs, NULL, KERN_DEFAULT);
 243 }
 244
 245 static arch_spinlock_t die_lock = __ARCH_SPIN_LOCK_UNLOCKED;
 246 static int die_owner = -1;
 247 static unsigned int die_nest_count;
 248
 249 unsigned long oops_begin(void)
 250 {
 251         int cpu;
 252         unsigned long flags;
 253
 254         oops_enter();
 255
 256         /* racy, but better than risking deadlock. */
 257         raw_local_irq_save(flags);
 258         cpu = smp_processor_id();
 259         if (!arch_spin_trylock(&die_lock)) {
 260                 if (cpu == die_owner)
 261                         /* nested oops. should stop eventually */;
 262                 else
 263                         arch_spin_lock(&die_lock);
 264         }
 265         die_nest_count++;
 266         die_owner = cpu;
 267         console_verbose();
 268         bust_spinlocks(1);
 269         return flags;
 270 }
 271 EXPORT_SYMBOL_GPL(oops_begin);
 272 NOKPROBE_SYMBOL(oops_begin);
 273
 274 void __noreturn rewind_stack_do_exit(int signr);
 275
 276 void oops_end(unsigned long flags, struct pt_regs *regs, int signr)
 277 {
 278         if (regs && kexec_should_crash(current))
 279                 crash_kexec(regs);
 280
 281         bust_spinlocks(0);
 282         die_owner = -1;
 283         add_taint(TAINT_DIE, LOCKDEP_NOW_UNRELIABLE);
 284         die_nest_count--;
 285         if (!die_nest_count)
 286                 /* Nest count reaches zero, release the lock. */
 287                 arch_spin_unlock(&die_lock);
 288         raw_local_irq_restore(flags);
 289         oops_exit();
 290
 291         if (!signr)
 292                 return;
 293         if (in_interrupt())
 294                 panic("Fatal exception in interrupt");
 295         if (panic_on_oops)
 296                 panic("Fatal exception");
 297
 298         /*
 299          * We're not going to return, but we might be on an IST stack or
 300          * have very little stack space left.  Rewind the stack and kill
 301          * the task.
 302          * Before we rewind the stack, we have to tell KASAN that we're going to
 303          * reuse the task stack and that existing poisons are invalid.
 304          */
 305         kasan_unpoison_task_stack(current);
 306         rewind_stack_do_exit(signr);
 307 }
 308 NOKPROBE_SYMBOL(oops_end);
 309
 310 int __die(const char *str, struct pt_regs *regs, long err)
 311 {
 312 #ifdef CONFIG_X86_32
 313         unsigned short ss;
 314         unsigned long sp;
 315 #endif
 316         printk(KERN_DEFAULT
 317                "%s: %04lx [#%d]%s%s%s%s%s\n", str, err & 0xffff, ++die_counter,
 318                IS_ENABLED(CONFIG_PREEMPT) ? " PREEMPT"         : "",
 319                IS_ENABLED(CONFIG_SMP)     ? " SMP"             : "",
 320                debug_pagealloc_enabled()  ? " DEBUG_PAGEALLOC" : "",
 321                IS_ENABLED(CONFIG_KASAN)   ? " KASAN"           : "",
 322                IS_ENABLED(CONFIG_PAGE_TABLE_ISOLATION) ?
 323                (boot_cpu_has(X86_FEATURE_PTI) ? " PTI" : " NOPTI") : "");
 324
 325         if (notify_die(DIE_OOPS, str, regs, err,
 326                         current->thread.trap_nr, SIGSEGV) == NOTIFY_STOP)
 327                 return 1;
 328
 329         print_modules();
 330         show_regs(regs);
 331 #ifdef CONFIG_X86_32
 332         if (user_mode(regs)) {
 333                 sp = regs->sp;
 334                 ss = regs->ss;
 335         } else {
 336                 sp = kernel_stack_pointer(regs);
 337                 savesegment(ss, ss);
 338         }
 339         printk(KERN_EMERG "EIP: %pS SS:ESP: %04x:%08lx\n",
 340                (void *)regs->ip, ss, sp);
 341 #else
 342         /* Executive summary in case the oops scrolled away */
 343         printk(KERN_ALERT "RIP: %pS RSP: %016lx\n", (void *)regs->ip, regs->sp);
 344 #endif
 345         return 0;
 346 }
 347 NOKPROBE_SYMBOL(__die);
 348
 349 /*
 350  * This is gone through when something in the kernel has done something bad
 351  * and is about to be terminated:
 352  */
 353 void die(const char *str, struct pt_regs *regs, long err)
 354 {
 355         unsigned long flags = oops_begin();
 356         int sig = SIGSEGV;
 357
 358         if (__die(str, regs, err))
 359                 sig = 0;
 360         oops_end(flags, regs, sig);
 361 }
 362
 363 static int __init code_bytes_setup(char *s)
 364 {
 365         ssize_t ret;
 366         unsigned long val;
 367
 368         if (!s)
 369                 return -EINVAL;
 370
 371         ret = kstrtoul(s, 0, &val);
 372         if (ret)
 373                 return ret;
 374
 375         code_bytes = val;
 376         if (code_bytes > 8192)
 377                 code_bytes = 8192;
 378
 379         return 1;
 380 }
 381 __setup("code_bytes=", code_bytes_setup);