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