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/ftrace.h>
  14 #include <linux/kexec.h>
  15 #include <linux/bug.h>
  16 #include <linux/nmi.h>
  17 #include <linux/sysfs.h>
  18 #include <linux/kasan.h>
  19
  20 #include <asm/stacktrace.h>
  21 #include <asm/unwind.h>
  22
  23 int panic_on_unrecovered_nmi;
  24 int panic_on_io_nmi;
  25 unsigned int code_bytes = 64;
  26 int kstack_depth_to_print = 3 * STACKSLOTS_PER_LINE;
  27 static int die_counter;
  28
  29 bool in_task_stack(unsigned long *stack, struct task_struct *task,
  30                    struct stack_info *info)
  31 {
  32         unsigned long *begin = task_stack_page(task);
  33         unsigned long *end   = task_stack_page(task) + THREAD_SIZE;
  34
  35         if (stack < begin || stack >= end)
  36                 return false;
  37
  38         info->type      = STACK_TYPE_TASK;
  39         info->begin     = begin;
  40         info->end       = end;
  41         info->next_sp   = NULL;
  42
  43         return true;
  44 }
  45
  46 static void printk_stack_address(unsigned long address, int reliable,
  47                                  char *log_lvl)
  48 {
  49         touch_nmi_watchdog();
  50         printk("%s [<%p>] %s%pB\n",
  51                 log_lvl, (void *)address, reliable ? "" : "? ",
  52                 (void *)address);
  53 }
  54
  55 void printk_address(unsigned long address)
  56 {
  57         pr_cont(" [<%p>] %pS\n", (void *)address, (void *)address);
  58 }
  59
  60 void show_trace_log_lvl(struct task_struct *task, struct pt_regs *regs,
  61                         unsigned long *stack, char *log_lvl)
  62 {
  63         struct unwind_state state;
  64         struct stack_info stack_info = {0};
  65         unsigned long visit_mask = 0;
  66         int graph_idx = 0;
  67
  68         printk("%sCall Trace:\n", log_lvl);
  69
  70         unwind_start(&state, task, regs, stack);
  71
  72         /*
  73          * Iterate through the stacks, starting with the current stack pointer.
  74          * Each stack has a pointer to the next one.
  75          *
  76          * x86-64 can have several stacks:
  77          * - task stack
  78          * - interrupt stack
  79          * - HW exception stacks (double fault, nmi, debug, mce)
  80          *
  81          * x86-32 can have up to three stacks:
  82          * - task stack
  83          * - softirq stack
  84          * - hardirq stack
  85          */
  86         for (; stack; stack = stack_info.next_sp) {
  87                 const char *str_begin, *str_end;
  88
  89                 /*
  90                  * If we overflowed the task stack into a guard page, jump back
  91                  * to the bottom of the usable stack.
  92                  */
  93                 if (task_stack_page(task) - (void *)stack < PAGE_SIZE)
  94                         stack = task_stack_page(task);
  95
  96                 if (get_stack_info(stack, task, &stack_info, &visit_mask))
  97                         break;
  98
  99                 stack_type_str(stack_info.type, &str_begin, &str_end);
 100                 if (str_begin)
 101                         printk("%s <%s> ", log_lvl, str_begin);
 102
 103                 /*
 104                  * Scan the stack, printing any text addresses we find.  At the
 105                  * same time, follow proper stack frames with the unwinder.
 106                  *
 107                  * Addresses found during the scan which are not reported by
 108                  * the unwinder are considered to be additional clues which are
 109                  * sometimes useful for debugging and are prefixed with '?'.
 110                  * This also serves as a failsafe option in case the unwinder
 111                  * goes off in the weeds.
 112                  */
 113                 for (; stack < stack_info.end; stack++) {
 114                         unsigned long real_addr;
 115                         int reliable = 0;
 116                         unsigned long addr = READ_ONCE_NOCHECK(*stack);
 117                         unsigned long *ret_addr_p =
 118                                 unwind_get_return_address_ptr(&state);
 119
 120                         if (!__kernel_text_address(addr))
 121                                 continue;
 122
 123                         if (stack == ret_addr_p)
 124                                 reliable = 1;
 125
 126                         /*
 127                          * When function graph tracing is enabled for a
 128                          * function, its return address on the stack is
 129                          * replaced with the address of an ftrace handler
 130                          * (return_to_handler).  In that case, before printing
 131                          * the "real" address, we want to print the handler
 132                          * address as an "unreliable" hint that function graph
 133                          * tracing was involved.
 134                          */
 135                         real_addr = ftrace_graph_ret_addr(task, &graph_idx,
 136                                                           addr, stack);
 137                         if (real_addr != addr)
 138                                 printk_stack_address(addr, 0, log_lvl);
 139                         printk_stack_address(real_addr, reliable, log_lvl);
 140
 141                         if (!reliable)
 142                                 continue;
 143
 144                         /*
 145                          * Get the next frame from the unwinder.  No need to
 146                          * check for an error: if anything goes wrong, the rest
 147                          * of the addresses will just be printed as unreliable.
 148                          */
 149                         unwind_next_frame(&state);
 150                 }
 151
 152                 if (str_end)
 153                         printk("%s <%s> ", log_lvl, str_end);
 154         }
 155 }
 156
 157 void show_stack(struct task_struct *task, unsigned long *sp)
 158 {
 159         task = task ? : current;
 160
 161         /*
 162          * Stack frames below this one aren't interesting.  Don't show them
 163          * if we're printing for %current.
 164          */
 165         if (!sp && task == current)
 166                 sp = get_stack_pointer(current, NULL);
 167
 168         show_stack_log_lvl(task, NULL, sp, "");
 169 }
 170
 171 void show_stack_regs(struct pt_regs *regs)
 172 {
 173         show_stack_log_lvl(current, regs, NULL, "");
 174 }
 175
 176 static arch_spinlock_t die_lock = __ARCH_SPIN_LOCK_UNLOCKED;
 177 static int die_owner = -1;
 178 static unsigned int die_nest_count;
 179
 180 unsigned long oops_begin(void)
 181 {
 182         int cpu;
 183         unsigned long flags;
 184
 185         oops_enter();
 186
 187         /* racy, but better than risking deadlock. */
 188         raw_local_irq_save(flags);
 189         cpu = smp_processor_id();
 190         if (!arch_spin_trylock(&die_lock)) {
 191                 if (cpu == die_owner)
 192                         /* nested oops. should stop eventually */;
 193                 else
 194                         arch_spin_lock(&die_lock);
 195         }
 196         die_nest_count++;
 197         die_owner = cpu;
 198         console_verbose();
 199         bust_spinlocks(1);
 200         return flags;
 201 }
 202 EXPORT_SYMBOL_GPL(oops_begin);
 203 NOKPROBE_SYMBOL(oops_begin);
 204
 205 void __noreturn rewind_stack_do_exit(int signr);
 206
 207 void oops_end(unsigned long flags, struct pt_regs *regs, int signr)
 208 {
 209         if (regs && kexec_should_crash(current))
 210                 crash_kexec(regs);
 211
 212         bust_spinlocks(0);
 213         die_owner = -1;
 214         add_taint(TAINT_DIE, LOCKDEP_NOW_UNRELIABLE);
 215         die_nest_count--;
 216         if (!die_nest_count)
 217                 /* Nest count reaches zero, release the lock. */
 218                 arch_spin_unlock(&die_lock);
 219         raw_local_irq_restore(flags);
 220         oops_exit();
 221
 222         if (!signr)
 223                 return;
 224         if (in_interrupt())
 225                 panic("Fatal exception in interrupt");
 226         if (panic_on_oops)
 227                 panic("Fatal exception");
 228
 229         /*
 230          * We're not going to return, but we might be on an IST stack or
 231          * have very little stack space left.  Rewind the stack and kill
 232          * the task.
 233          * Before we rewind the stack, we have to tell KASAN that we're going to
 234          * reuse the task stack and that existing poisons are invalid.
 235          */
 236         kasan_unpoison_task_stack(current);
 237         rewind_stack_do_exit(signr);
 238 }
 239 NOKPROBE_SYMBOL(oops_end);
 240
 241 int __die(const char *str, struct pt_regs *regs, long err)
 242 {
 243 #ifdef CONFIG_X86_32
 244         unsigned short ss;
 245         unsigned long sp;
 246 #endif
 247         printk(KERN_DEFAULT
 248                "%s: %04lx [#%d]%s%s%s%s\n", str, err & 0xffff, ++die_counter,
 249                IS_ENABLED(CONFIG_PREEMPT) ? " PREEMPT"         : "",
 250                IS_ENABLED(CONFIG_SMP)     ? " SMP"             : "",
 251                debug_pagealloc_enabled()  ? " DEBUG_PAGEALLOC" : "",
 252                IS_ENABLED(CONFIG_KASAN)   ? " KASAN"           : "");
 253
 254         if (notify_die(DIE_OOPS, str, regs, err,
 255                         current->thread.trap_nr, SIGSEGV) == NOTIFY_STOP)
 256                 return 1;
 257
 258         print_modules();
 259         show_regs(regs);
 260 #ifdef CONFIG_X86_32
 261         if (user_mode(regs)) {
 262                 sp = regs->sp;
 263                 ss = regs->ss & 0xffff;
 264         } else {
 265                 sp = kernel_stack_pointer(regs);
 266                 savesegment(ss, ss);
 267         }
 268         printk(KERN_EMERG "EIP: [<%08lx>] ", regs->ip);
 269         print_symbol("%s", regs->ip);
 270         printk(" SS:ESP %04x:%08lx\n", ss, sp);
 271 #else
 272         /* Executive summary in case the oops scrolled away */
 273         printk(KERN_ALERT "RIP ");
 274         printk_address(regs->ip);
 275         printk(" RSP <%016lx>\n", regs->sp);
 276 #endif
 277         return 0;
 278 }
 279 NOKPROBE_SYMBOL(__die);
 280
 281 /*
 282  * This is gone through when something in the kernel has done something bad
 283  * and is about to be terminated:
 284  */
 285 void die(const char *str, struct pt_regs *regs, long err)
 286 {
 287         unsigned long flags = oops_begin();
 288         int sig = SIGSEGV;
 289
 290         if (!user_mode(regs))
 291                 report_bug(regs->ip, regs);
 292
 293         if (__die(str, regs, err))
 294                 sig = 0;
 295         oops_end(flags, regs, sig);
 296 }
 297
 298 static int __init kstack_setup(char *s)
 299 {
 300         ssize_t ret;
 301         unsigned long val;
 302
 303         if (!s)
 304                 return -EINVAL;
 305
 306         ret = kstrtoul(s, 0, &val);
 307         if (ret)
 308                 return ret;
 309         kstack_depth_to_print = val;
 310         return 0;
 311 }
 312 early_param("kstack", kstack_setup);
 313
 314 static int __init code_bytes_setup(char *s)
 315 {
 316         ssize_t ret;
 317         unsigned long val;
 318
 319         if (!s)
 320                 return -EINVAL;
 321
 322         ret = kstrtoul(s, 0, &val);
 323         if (ret)
 324                 return ret;
 325
 326         code_bytes = val;
 327         if (code_bytes > 8192)
 328                 code_bytes = 8192;
 329
 330         return 1;
 331 }
 332 __setup("code_bytes=", code_bytes_setup);