root/arch/x86/kernel/dumpstack.c

DEFINITIONS

1. in_task_stack
2. in_entry_stack
3. printk_stack_address
4. show_opcodes
5. show_ip
6. show_iret_regs
7. show_regs_if_on_stack
8. show_trace_log_lvl
9. show_stack
10. show_stack_regs
11. oops_begin
12. oops_end
13. __die
14. die
15. show_regs

   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 static int die_counter;
  29 
  30 static struct pt_regs exec_summary_regs;
  31 
  32 bool in_task_stack(unsigned long *stack, struct task_struct *task,
  33                    struct stack_info *info)
  34 {
  35         unsigned long *begin = task_stack_page(task);
  36         unsigned long *end   = task_stack_page(task) + THREAD_SIZE;
  37 
  38         if (stack < begin || stack >= end)
  39                 return false;
  40 
  41         info->type      = STACK_TYPE_TASK;
  42         info->begin     = begin;
  43         info->end       = end;
  44         info->next_sp   = NULL;
  45 
  46         return true;
  47 }
  48 
  49 bool in_entry_stack(unsigned long *stack, struct stack_info *info)
  50 {
  51         struct entry_stack *ss = cpu_entry_stack(smp_processor_id());
  52 
  53         void *begin = ss;
  54         void *end = ss + 1;
  55 
  56         if ((void *)stack < begin || (void *)stack >= end)
  57                 return false;
  58 
  59         info->type      = STACK_TYPE_ENTRY;
  60         info->begin     = begin;
  61         info->end       = end;
  62         info->next_sp   = NULL;
  63 
  64         return true;
  65 }
  66 
  67 static void printk_stack_address(unsigned long address, int reliable,
  68                                  char *log_lvl)
  69 {
  70         touch_nmi_watchdog();
  71         printk("%s %s%pB\n", log_lvl, reliable ? "" : "? ", (void *)address);
  72 }
  73 
  74 /*
  75  * There are a couple of reasons for the 2/3rd prologue, courtesy of Linus:
  76  *
  77  * In case where we don't have the exact kernel image (which, if we did, we can
  78  * simply disassemble and navigate to the RIP), the purpose of the bigger
  79  * prologue is to have more context and to be able to correlate the code from
  80  * the different toolchains better.
  81  *
  82  * In addition, it helps in recreating the register allocation of the failing
  83  * kernel and thus make sense of the register dump.
  84  *
  85  * What is more, the additional complication of a variable length insn arch like
  86  * x86 warrants having longer byte sequence before rIP so that the disassembler
  87  * can "sync" up properly and find instruction boundaries when decoding the
  88  * opcode bytes.
  89  *
  90  * Thus, the 2/3rds prologue and 64 byte OPCODE_BUFSIZE is just a random
  91  * guesstimate in attempt to achieve all of the above.
  92  */
  93 void show_opcodes(struct pt_regs *regs, const char *loglvl)
  94 {
  95 #define PROLOGUE_SIZE 42
  96 #define EPILOGUE_SIZE 21
  97 #define OPCODE_BUFSIZE (PROLOGUE_SIZE + 1 + EPILOGUE_SIZE)
  98         u8 opcodes[OPCODE_BUFSIZE];
  99         unsigned long prologue = regs->ip - PROLOGUE_SIZE;
 100         bool bad_ip;
 101 
 102         /*
 103          * Make sure userspace isn't trying to trick us into dumping kernel
 104          * memory by pointing the userspace instruction pointer at it.
 105          */
 106         bad_ip = user_mode(regs) &&
 107                 __chk_range_not_ok(prologue, OPCODE_BUFSIZE, TASK_SIZE_MAX);
 108 
 109         if (bad_ip || probe_kernel_read(opcodes, (u8 *)prologue,
 110                                         OPCODE_BUFSIZE)) {
 111                 printk("%sCode: Bad RIP value.\n", loglvl);
 112         } else {
 113                 printk("%sCode: %" __stringify(PROLOGUE_SIZE) "ph <%02x> %"
 114                        __stringify(EPILOGUE_SIZE) "ph\n", loglvl, opcodes,
 115                        opcodes[PROLOGUE_SIZE], opcodes + PROLOGUE_SIZE + 1);
 116         }
 117 }
 118 
 119 void show_ip(struct pt_regs *regs, const char *loglvl)
 120 {
 121 #ifdef CONFIG_X86_32
 122         printk("%sEIP: %pS\n", loglvl, (void *)regs->ip);
 123 #else
 124         printk("%sRIP: %04x:%pS\n", loglvl, (int)regs->cs, (void *)regs->ip);
 125 #endif
 126         show_opcodes(regs, loglvl);
 127 }
 128 
 129 void show_iret_regs(struct pt_regs *regs)
 130 {
 131         show_ip(regs, KERN_DEFAULT);
 132         printk(KERN_DEFAULT "RSP: %04x:%016lx EFLAGS: %08lx", (int)regs->ss,
 133                 regs->sp, regs->flags);
 134 }
 135 
 136 static void show_regs_if_on_stack(struct stack_info *info, struct pt_regs *regs,
 137                                   bool partial)
 138 {
 139         /*
 140          * These on_stack() checks aren't strictly necessary: the unwind code
 141          * has already validated the 'regs' pointer.  The checks are done for
 142          * ordering reasons: if the registers are on the next stack, we don't
 143          * want to print them out yet.  Otherwise they'll be shown as part of
 144          * the wrong stack.  Later, when show_trace_log_lvl() switches to the
 145          * next stack, this function will be called again with the same regs so
 146          * they can be printed in the right context.
 147          */
 148         if (!partial && on_stack(info, regs, sizeof(*regs))) {
 149                 __show_regs(regs, SHOW_REGS_SHORT);
 150 
 151         } else if (partial && on_stack(info, (void *)regs + IRET_FRAME_OFFSET,
 152                                        IRET_FRAME_SIZE)) {
 153                 /*
 154                  * When an interrupt or exception occurs in entry code, the
 155                  * full pt_regs might not have been saved yet.  In that case
 156                  * just print the iret frame.
 157                  */
 158                 show_iret_regs(regs);
 159         }
 160 }
 161 
 162 void show_trace_log_lvl(struct task_struct *task, struct pt_regs *regs,
 163                         unsigned long *stack, char *log_lvl)
 164 {
 165         struct unwind_state state;
 166         struct stack_info stack_info = {0};
 167         unsigned long visit_mask = 0;
 168         int graph_idx = 0;
 169         bool partial = false;
 170 
 171         printk("%sCall Trace:\n", log_lvl);
 172 
 173         unwind_start(&state, task, regs, stack);
 174         stack = stack ? : get_stack_pointer(task, regs);
 175         regs = unwind_get_entry_regs(&state, &partial);
 176 
 177         /*
 178          * Iterate through the stacks, starting with the current stack pointer.
 179          * Each stack has a pointer to the next one.
 180          *
 181          * x86-64 can have several stacks:
 182          * - task stack
 183          * - interrupt stack
 184          * - HW exception stacks (double fault, nmi, debug, mce)
 185          * - entry stack
 186          *
 187          * x86-32 can have up to four stacks:
 188          * - task stack
 189          * - softirq stack
 190          * - hardirq stack
 191          * - entry stack
 192          */
 193         for ( ; stack; stack = PTR_ALIGN(stack_info.next_sp, sizeof(long))) {
 194                 const char *stack_name;
 195 
 196                 if (get_stack_info(stack, task, &stack_info, &visit_mask)) {
 197                         /*
 198                          * We weren't on a valid stack.  It's possible that
 199                          * we overflowed a valid stack into a guard page.
 200                          * See if the next page up is valid so that we can
 201                          * generate some kind of backtrace if this happens.
 202                          */
 203                         stack = (unsigned long *)PAGE_ALIGN((unsigned long)stack);
 204                         if (get_stack_info(stack, task, &stack_info, &visit_mask))
 205                                 break;
 206                 }
 207 
 208                 stack_name = stack_type_name(stack_info.type);
 209                 if (stack_name)
 210                         printk("%s <%s>\n", log_lvl, stack_name);
 211 
 212                 if (regs)
 213                         show_regs_if_on_stack(&stack_info, regs, partial);
 214 
 215                 /*
 216                  * Scan the stack, printing any text addresses we find.  At the
 217                  * same time, follow proper stack frames with the unwinder.
 218                  *
 219                  * Addresses found during the scan which are not reported by
 220                  * the unwinder are considered to be additional clues which are
 221                  * sometimes useful for debugging and are prefixed with '?'.
 222                  * This also serves as a failsafe option in case the unwinder
 223                  * goes off in the weeds.
 224                  */
 225                 for (; stack < stack_info.end; stack++) {
 226                         unsigned long real_addr;
 227                         int reliable = 0;
 228                         unsigned long addr = READ_ONCE_NOCHECK(*stack);
 229                         unsigned long *ret_addr_p =
 230                                 unwind_get_return_address_ptr(&state);
 231 
 232                         if (!__kernel_text_address(addr))
 233                                 continue;
 234 
 235                         /*
 236                          * Don't print regs->ip again if it was already printed
 237                          * by show_regs_if_on_stack().
 238                          */
 239                         if (regs && stack == &regs->ip)
 240                                 goto next;
 241 
 242                         if (stack == ret_addr_p)
 243                                 reliable = 1;
 244 
 245                         /*
 246                          * When function graph tracing is enabled for a
 247                          * function, its return address on the stack is
 248                          * replaced with the address of an ftrace handler
 249                          * (return_to_handler).  In that case, before printing
 250                          * the "real" address, we want to print the handler
 251                          * address as an "unreliable" hint that function graph
 252                          * tracing was involved.
 253                          */
 254                         real_addr = ftrace_graph_ret_addr(task, &graph_idx,
 255                                                           addr, stack);
 256                         if (real_addr != addr)
 257                                 printk_stack_address(addr, 0, log_lvl);
 258                         printk_stack_address(real_addr, reliable, log_lvl);
 259 
 260                         if (!reliable)
 261                                 continue;
 262 
 263 next:
 264                         /*
 265                          * Get the next frame from the unwinder.  No need to
 266                          * check for an error: if anything goes wrong, the rest
 267                          * of the addresses will just be printed as unreliable.
 268                          */
 269                         unwind_next_frame(&state);
 270 
 271                         /* if the frame has entry regs, print them */
 272                         regs = unwind_get_entry_regs(&state, &partial);
 273                         if (regs)
 274                                 show_regs_if_on_stack(&stack_info, regs, partial);
 275                 }
 276 
 277                 if (stack_name)
 278                         printk("%s </%s>\n", log_lvl, stack_name);
 279         }
 280 }
 281 
 282 void show_stack(struct task_struct *task, unsigned long *sp)
 283 {
 284         task = task ? : current;
 285 
 286         /*
 287          * Stack frames below this one aren't interesting.  Don't show them
 288          * if we're printing for %current.
 289          */
 290         if (!sp && task == current)
 291                 sp = get_stack_pointer(current, NULL);
 292 
 293         show_trace_log_lvl(task, NULL, sp, KERN_DEFAULT);
 294 }
 295 
 296 void show_stack_regs(struct pt_regs *regs)
 297 {
 298         show_trace_log_lvl(current, regs, NULL, KERN_DEFAULT);
 299 }
 300 
 301 static arch_spinlock_t die_lock = __ARCH_SPIN_LOCK_UNLOCKED;
 302 static int die_owner = -1;
 303 static unsigned int die_nest_count;
 304 
 305 unsigned long oops_begin(void)
 306 {
 307         int cpu;
 308         unsigned long flags;
 309 
 310         oops_enter();
 311 
 312         /* racy, but better than risking deadlock. */
 313         raw_local_irq_save(flags);
 314         cpu = smp_processor_id();
 315         if (!arch_spin_trylock(&die_lock)) {
 316                 if (cpu == die_owner)
 317                         /* nested oops. should stop eventually */;
 318                 else
 319                         arch_spin_lock(&die_lock);
 320         }
 321         die_nest_count++;
 322         die_owner = cpu;
 323         console_verbose();
 324         bust_spinlocks(1);
 325         return flags;
 326 }
 327 NOKPROBE_SYMBOL(oops_begin);
 328 
 329 void __noreturn rewind_stack_do_exit(int signr);
 330 
 331 void oops_end(unsigned long flags, struct pt_regs *regs, int signr)
 332 {
 333         if (regs && kexec_should_crash(current))
 334                 crash_kexec(regs);
 335 
 336         bust_spinlocks(0);
 337         die_owner = -1;
 338         add_taint(TAINT_DIE, LOCKDEP_NOW_UNRELIABLE);
 339         die_nest_count--;
 340         if (!die_nest_count)
 341                 /* Nest count reaches zero, release the lock. */
 342                 arch_spin_unlock(&die_lock);
 343         raw_local_irq_restore(flags);
 344         oops_exit();
 345 
 346         /* Executive summary in case the oops scrolled away */
 347         __show_regs(&exec_summary_regs, SHOW_REGS_ALL);
 348 
 349         if (!signr)
 350                 return;
 351         if (in_interrupt())
 352                 panic("Fatal exception in interrupt");
 353         if (panic_on_oops)
 354                 panic("Fatal exception");
 355 
 356         /*
 357          * We're not going to return, but we might be on an IST stack or
 358          * have very little stack space left.  Rewind the stack and kill
 359          * the task.
 360          * Before we rewind the stack, we have to tell KASAN that we're going to
 361          * reuse the task stack and that existing poisons are invalid.
 362          */
 363         kasan_unpoison_task_stack(current);
 364         rewind_stack_do_exit(signr);
 365 }
 366 NOKPROBE_SYMBOL(oops_end);
 367 
 368 int __die(const char *str, struct pt_regs *regs, long err)
 369 {
 370         const char *pr = "";
 371 
 372         /* Save the regs of the first oops for the executive summary later. */
 373         if (!die_counter)
 374                 exec_summary_regs = *regs;
 375 
 376         if (IS_ENABLED(CONFIG_PREEMPTION))
 377                 pr = IS_ENABLED(CONFIG_PREEMPT_RT) ? " PREEMPT_RT" : " PREEMPT";
 378 
 379         printk(KERN_DEFAULT
 380                "%s: %04lx [#%d]%s%s%s%s%s\n", str, err & 0xffff, ++die_counter,
 381                pr,
 382                IS_ENABLED(CONFIG_SMP)     ? " SMP"             : "",
 383                debug_pagealloc_enabled()  ? " DEBUG_PAGEALLOC" : "",
 384                IS_ENABLED(CONFIG_KASAN)   ? " KASAN"           : "",
 385                IS_ENABLED(CONFIG_PAGE_TABLE_ISOLATION) ?
 386                (boot_cpu_has(X86_FEATURE_PTI) ? " PTI" : " NOPTI") : "");
 387 
 388         show_regs(regs);
 389         print_modules();
 390 
 391         if (notify_die(DIE_OOPS, str, regs, err,
 392                         current->thread.trap_nr, SIGSEGV) == NOTIFY_STOP)
 393                 return 1;
 394 
 395         return 0;
 396 }
 397 NOKPROBE_SYMBOL(__die);
 398 
 399 /*
 400  * This is gone through when something in the kernel has done something bad
 401  * and is about to be terminated:
 402  */
 403 void die(const char *str, struct pt_regs *regs, long err)
 404 {
 405         unsigned long flags = oops_begin();
 406         int sig = SIGSEGV;
 407 
 408         if (__die(str, regs, err))
 409                 sig = 0;
 410         oops_end(flags, regs, sig);
 411 }
 412 
 413 void show_regs(struct pt_regs *regs)
 414 {
 415         show_regs_print_info(KERN_DEFAULT);
 416 
 417         __show_regs(regs, user_mode(regs) ? SHOW_REGS_USER : SHOW_REGS_ALL);
 418 
 419         /*
 420          * When in-kernel, we also print out the stack at the time of the fault..
 421          */
 422         if (!user_mode(regs))
 423                 show_trace_log_lvl(current, regs, NULL, KERN_DEFAULT);
 424 }