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