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