[PATCH] unwinder: more sanity checks in Dwarf2 unwinder
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / arch / x86_64 / kernel / traps.c
blob4fdd162f0befdcd06863963675bc5c29a8b3f16e
1 /*
2 * linux/arch/x86-64/traps.c
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 * Copyright (C) 2000, 2001, 2002 Andi Kleen, SuSE Labs
7 * Pentium III FXSR, SSE support
8 * Gareth Hughes <gareth@valinux.com>, May 2000
9 */
12 * 'Traps.c' handles hardware traps and faults after we have saved some
13 * state in 'entry.S'.
15 #include <linux/sched.h>
16 #include <linux/kernel.h>
17 #include <linux/string.h>
18 #include <linux/errno.h>
19 #include <linux/ptrace.h>
20 #include <linux/timer.h>
21 #include <linux/mm.h>
22 #include <linux/init.h>
23 #include <linux/delay.h>
24 #include <linux/spinlock.h>
25 #include <linux/interrupt.h>
26 #include <linux/kallsyms.h>
27 #include <linux/module.h>
28 #include <linux/moduleparam.h>
29 #include <linux/nmi.h>
30 #include <linux/kprobes.h>
31 #include <linux/kexec.h>
32 #include <linux/unwind.h>
33 #include <linux/uaccess.h>
35 #include <asm/system.h>
36 #include <asm/io.h>
37 #include <asm/atomic.h>
38 #include <asm/debugreg.h>
39 #include <asm/desc.h>
40 #include <asm/i387.h>
41 #include <asm/kdebug.h>
42 #include <asm/processor.h>
43 #include <asm/unwind.h>
44 #include <asm/smp.h>
45 #include <asm/pgalloc.h>
46 #include <asm/pda.h>
47 #include <asm/proto.h>
48 #include <asm/nmi.h>
49 #include <asm/stacktrace.h>
51 asmlinkage void divide_error(void);
52 asmlinkage void debug(void);
53 asmlinkage void nmi(void);
54 asmlinkage void int3(void);
55 asmlinkage void overflow(void);
56 asmlinkage void bounds(void);
57 asmlinkage void invalid_op(void);
58 asmlinkage void device_not_available(void);
59 asmlinkage void double_fault(void);
60 asmlinkage void coprocessor_segment_overrun(void);
61 asmlinkage void invalid_TSS(void);
62 asmlinkage void segment_not_present(void);
63 asmlinkage void stack_segment(void);
64 asmlinkage void general_protection(void);
65 asmlinkage void page_fault(void);
66 asmlinkage void coprocessor_error(void);
67 asmlinkage void simd_coprocessor_error(void);
68 asmlinkage void reserved(void);
69 asmlinkage void alignment_check(void);
70 asmlinkage void machine_check(void);
71 asmlinkage void spurious_interrupt_bug(void);
73 ATOMIC_NOTIFIER_HEAD(die_chain);
74 EXPORT_SYMBOL(die_chain);
76 int register_die_notifier(struct notifier_block *nb)
78 vmalloc_sync_all();
79 return atomic_notifier_chain_register(&die_chain, nb);
81 EXPORT_SYMBOL(register_die_notifier); /* used modular by kdb */
83 int unregister_die_notifier(struct notifier_block *nb)
85 return atomic_notifier_chain_unregister(&die_chain, nb);
87 EXPORT_SYMBOL(unregister_die_notifier); /* used modular by kdb */
89 static inline void conditional_sti(struct pt_regs *regs)
91 if (regs->eflags & X86_EFLAGS_IF)
92 local_irq_enable();
95 static inline void preempt_conditional_sti(struct pt_regs *regs)
97 preempt_disable();
98 if (regs->eflags & X86_EFLAGS_IF)
99 local_irq_enable();
102 static inline void preempt_conditional_cli(struct pt_regs *regs)
104 if (regs->eflags & X86_EFLAGS_IF)
105 local_irq_disable();
106 /* Make sure to not schedule here because we could be running
107 on an exception stack. */
108 preempt_enable_no_resched();
111 int kstack_depth_to_print = 12;
112 #ifdef CONFIG_STACK_UNWIND
113 static int call_trace = 1;
114 #else
115 #define call_trace (-1)
116 #endif
118 #ifdef CONFIG_KALLSYMS
119 void printk_address(unsigned long address)
121 unsigned long offset = 0, symsize;
122 const char *symname;
123 char *modname;
124 char *delim = ":";
125 char namebuf[128];
127 symname = kallsyms_lookup(address, &symsize, &offset,
128 &modname, namebuf);
129 if (!symname) {
130 printk(" [<%016lx>]\n", address);
131 return;
133 if (!modname)
134 modname = delim = "";
135 printk(" [<%016lx>] %s%s%s%s+0x%lx/0x%lx\n",
136 address, delim, modname, delim, symname, offset, symsize);
138 #else
139 void printk_address(unsigned long address)
141 printk(" [<%016lx>]\n", address);
143 #endif
145 static unsigned long *in_exception_stack(unsigned cpu, unsigned long stack,
146 unsigned *usedp, char **idp)
148 static char ids[][8] = {
149 [DEBUG_STACK - 1] = "#DB",
150 [NMI_STACK - 1] = "NMI",
151 [DOUBLEFAULT_STACK - 1] = "#DF",
152 [STACKFAULT_STACK - 1] = "#SS",
153 [MCE_STACK - 1] = "#MC",
154 #if DEBUG_STKSZ > EXCEPTION_STKSZ
155 [N_EXCEPTION_STACKS ... N_EXCEPTION_STACKS + DEBUG_STKSZ / EXCEPTION_STKSZ - 2] = "#DB[?]"
156 #endif
158 unsigned k;
161 * Iterate over all exception stacks, and figure out whether
162 * 'stack' is in one of them:
164 for (k = 0; k < N_EXCEPTION_STACKS; k++) {
165 unsigned long end = per_cpu(orig_ist, cpu).ist[k];
167 * Is 'stack' above this exception frame's end?
168 * If yes then skip to the next frame.
170 if (stack >= end)
171 continue;
173 * Is 'stack' above this exception frame's start address?
174 * If yes then we found the right frame.
176 if (stack >= end - EXCEPTION_STKSZ) {
178 * Make sure we only iterate through an exception
179 * stack once. If it comes up for the second time
180 * then there's something wrong going on - just
181 * break out and return NULL:
183 if (*usedp & (1U << k))
184 break;
185 *usedp |= 1U << k;
186 *idp = ids[k];
187 return (unsigned long *)end;
190 * If this is a debug stack, and if it has a larger size than
191 * the usual exception stacks, then 'stack' might still
192 * be within the lower portion of the debug stack:
194 #if DEBUG_STKSZ > EXCEPTION_STKSZ
195 if (k == DEBUG_STACK - 1 && stack >= end - DEBUG_STKSZ) {
196 unsigned j = N_EXCEPTION_STACKS - 1;
199 * Black magic. A large debug stack is composed of
200 * multiple exception stack entries, which we
201 * iterate through now. Dont look:
203 do {
204 ++j;
205 end -= EXCEPTION_STKSZ;
206 ids[j][4] = '1' + (j - N_EXCEPTION_STACKS);
207 } while (stack < end - EXCEPTION_STKSZ);
208 if (*usedp & (1U << j))
209 break;
210 *usedp |= 1U << j;
211 *idp = ids[j];
212 return (unsigned long *)end;
214 #endif
216 return NULL;
219 struct ops_and_data {
220 struct stacktrace_ops *ops;
221 void *data;
224 static int dump_trace_unwind(struct unwind_frame_info *info, void *context)
226 struct ops_and_data *oad = (struct ops_and_data *)context;
227 int n = 0;
228 unsigned long sp = UNW_SP(info);
230 if (arch_unw_user_mode(info))
231 return -1;
232 while (unwind(info) == 0 && UNW_PC(info)) {
233 n++;
234 oad->ops->address(oad->data, UNW_PC(info));
235 if (arch_unw_user_mode(info))
236 break;
237 if ((sp & ~(PAGE_SIZE - 1)) == (UNW_SP(info) & ~(PAGE_SIZE - 1))
238 && sp > UNW_SP(info))
239 break;
240 sp = UNW_SP(info);
242 return n;
245 #define MSG(txt) ops->warning(data, txt)
248 * x86-64 can have upto three kernel stacks:
249 * process stack
250 * interrupt stack
251 * severe exception (double fault, nmi, stack fault, debug, mce) hardware stack
254 static inline int valid_stack_ptr(struct thread_info *tinfo, void *p)
256 void *t = (void *)tinfo;
257 return p > t && p < t + THREAD_SIZE - 3;
260 void dump_trace(struct task_struct *tsk, struct pt_regs *regs,
261 unsigned long *stack,
262 struct stacktrace_ops *ops, void *data)
264 const unsigned cpu = get_cpu();
265 unsigned long *irqstack_end = (unsigned long*)cpu_pda(cpu)->irqstackptr;
266 unsigned used = 0;
267 struct thread_info *tinfo;
269 if (!tsk)
270 tsk = current;
272 if (call_trace >= 0) {
273 int unw_ret = 0;
274 struct unwind_frame_info info;
275 struct ops_and_data oad = { .ops = ops, .data = data };
277 if (regs) {
278 if (unwind_init_frame_info(&info, tsk, regs) == 0)
279 unw_ret = dump_trace_unwind(&info, &oad);
280 } else if (tsk == current)
281 unw_ret = unwind_init_running(&info, dump_trace_unwind,
282 &oad);
283 else {
284 if (unwind_init_blocked(&info, tsk) == 0)
285 unw_ret = dump_trace_unwind(&info, &oad);
287 if (unw_ret > 0) {
288 if (call_trace == 1 && !arch_unw_user_mode(&info)) {
289 ops->warning_symbol(data,
290 "DWARF2 unwinder stuck at %s",
291 UNW_PC(&info));
292 if ((long)UNW_SP(&info) < 0) {
293 MSG("Leftover inexact backtrace:");
294 stack = (unsigned long *)UNW_SP(&info);
295 if (!stack)
296 goto out;
297 } else
298 MSG("Full inexact backtrace again:");
299 } else if (call_trace >= 1)
300 goto out;
301 else
302 MSG("Full inexact backtrace again:");
303 } else
304 MSG("Inexact backtrace:");
306 if (!stack) {
307 unsigned long dummy;
308 stack = &dummy;
309 if (tsk && tsk != current)
310 stack = (unsigned long *)tsk->thread.rsp;
313 * Align the stack pointer on word boundary, later loops
314 * rely on that (and corruption / debug info bugs can cause
315 * unaligned values here):
317 stack = (unsigned long *)((unsigned long)stack & ~(sizeof(long)-1));
320 * Print function call entries within a stack. 'cond' is the
321 * "end of stackframe" condition, that the 'stack++'
322 * iteration will eventually trigger.
324 #define HANDLE_STACK(cond) \
325 do while (cond) { \
326 unsigned long addr = *stack++; \
327 /* Use unlocked access here because except for NMIs \
328 we should be already protected against module unloads */ \
329 if (__kernel_text_address(addr)) { \
330 /* \
331 * If the address is either in the text segment of the \
332 * kernel, or in the region which contains vmalloc'ed \
333 * memory, it *may* be the address of a calling \
334 * routine; if so, print it so that someone tracing \
335 * down the cause of the crash will be able to figure \
336 * out the call path that was taken. \
337 */ \
338 ops->address(data, addr); \
340 } while (0)
343 * Print function call entries in all stacks, starting at the
344 * current stack address. If the stacks consist of nested
345 * exceptions
347 for (;;) {
348 char *id;
349 unsigned long *estack_end;
350 estack_end = in_exception_stack(cpu, (unsigned long)stack,
351 &used, &id);
353 if (estack_end) {
354 if (ops->stack(data, id) < 0)
355 break;
356 HANDLE_STACK (stack < estack_end);
357 ops->stack(data, "<EOE>");
359 * We link to the next stack via the
360 * second-to-last pointer (index -2 to end) in the
361 * exception stack:
363 stack = (unsigned long *) estack_end[-2];
364 continue;
366 if (irqstack_end) {
367 unsigned long *irqstack;
368 irqstack = irqstack_end -
369 (IRQSTACKSIZE - 64) / sizeof(*irqstack);
371 if (stack >= irqstack && stack < irqstack_end) {
372 if (ops->stack(data, "IRQ") < 0)
373 break;
374 HANDLE_STACK (stack < irqstack_end);
376 * We link to the next stack (which would be
377 * the process stack normally) the last
378 * pointer (index -1 to end) in the IRQ stack:
380 stack = (unsigned long *) (irqstack_end[-1]);
381 irqstack_end = NULL;
382 ops->stack(data, "EOI");
383 continue;
386 break;
390 * This handles the process stack:
392 tinfo = current_thread_info();
393 HANDLE_STACK (valid_stack_ptr(tinfo, stack));
394 #undef HANDLE_STACK
395 out:
396 put_cpu();
398 EXPORT_SYMBOL(dump_trace);
400 static void
401 print_trace_warning_symbol(void *data, char *msg, unsigned long symbol)
403 print_symbol(msg, symbol);
404 printk("\n");
407 static void print_trace_warning(void *data, char *msg)
409 printk("%s\n", msg);
412 static int print_trace_stack(void *data, char *name)
414 printk(" <%s> ", name);
415 return 0;
418 static void print_trace_address(void *data, unsigned long addr)
420 printk_address(addr);
423 static struct stacktrace_ops print_trace_ops = {
424 .warning = print_trace_warning,
425 .warning_symbol = print_trace_warning_symbol,
426 .stack = print_trace_stack,
427 .address = print_trace_address,
430 void
431 show_trace(struct task_struct *tsk, struct pt_regs *regs, unsigned long *stack)
433 printk("\nCall Trace:\n");
434 dump_trace(tsk, regs, stack, &print_trace_ops, NULL);
435 printk("\n");
438 static void
439 _show_stack(struct task_struct *tsk, struct pt_regs *regs, unsigned long *rsp)
441 unsigned long *stack;
442 int i;
443 const int cpu = smp_processor_id();
444 unsigned long *irqstack_end = (unsigned long *) (cpu_pda(cpu)->irqstackptr);
445 unsigned long *irqstack = (unsigned long *) (cpu_pda(cpu)->irqstackptr - IRQSTACKSIZE);
447 // debugging aid: "show_stack(NULL, NULL);" prints the
448 // back trace for this cpu.
450 if (rsp == NULL) {
451 if (tsk)
452 rsp = (unsigned long *)tsk->thread.rsp;
453 else
454 rsp = (unsigned long *)&rsp;
457 stack = rsp;
458 for(i=0; i < kstack_depth_to_print; i++) {
459 if (stack >= irqstack && stack <= irqstack_end) {
460 if (stack == irqstack_end) {
461 stack = (unsigned long *) (irqstack_end[-1]);
462 printk(" <EOI> ");
464 } else {
465 if (((long) stack & (THREAD_SIZE-1)) == 0)
466 break;
468 if (i && ((i % 4) == 0))
469 printk("\n");
470 printk(" %016lx", *stack++);
471 touch_nmi_watchdog();
473 show_trace(tsk, regs, rsp);
476 void show_stack(struct task_struct *tsk, unsigned long * rsp)
478 _show_stack(tsk, NULL, rsp);
482 * The architecture-independent dump_stack generator
484 void dump_stack(void)
486 unsigned long dummy;
487 show_trace(NULL, NULL, &dummy);
490 EXPORT_SYMBOL(dump_stack);
492 void show_registers(struct pt_regs *regs)
494 int i;
495 int in_kernel = !user_mode(regs);
496 unsigned long rsp;
497 const int cpu = smp_processor_id();
498 struct task_struct *cur = cpu_pda(cpu)->pcurrent;
500 rsp = regs->rsp;
502 printk("CPU %d ", cpu);
503 __show_regs(regs);
504 printk("Process %s (pid: %d, threadinfo %p, task %p)\n",
505 cur->comm, cur->pid, task_thread_info(cur), cur);
508 * When in-kernel, we also print out the stack and code at the
509 * time of the fault..
511 if (in_kernel) {
513 printk("Stack: ");
514 _show_stack(NULL, regs, (unsigned long*)rsp);
516 printk("\nCode: ");
517 if (regs->rip < PAGE_OFFSET)
518 goto bad;
520 for (i=0; i<20; i++) {
521 unsigned char c;
522 if (__get_user(c, &((unsigned char*)regs->rip)[i])) {
523 bad:
524 printk(" Bad RIP value.");
525 break;
527 printk("%02x ", c);
530 printk("\n");
533 void handle_BUG(struct pt_regs *regs)
535 struct bug_frame f;
536 long len;
537 const char *prefix = "";
539 if (user_mode(regs))
540 return;
541 if (__copy_from_user(&f, (const void __user *) regs->rip,
542 sizeof(struct bug_frame)))
543 return;
544 if (f.filename >= 0 ||
545 f.ud2[0] != 0x0f || f.ud2[1] != 0x0b)
546 return;
547 len = __strnlen_user((char *)(long)f.filename, PATH_MAX) - 1;
548 if (len < 0 || len >= PATH_MAX)
549 f.filename = (int)(long)"unmapped filename";
550 else if (len > 50) {
551 f.filename += len - 50;
552 prefix = "...";
554 printk("----------- [cut here ] --------- [please bite here ] ---------\n");
555 printk(KERN_ALERT "Kernel BUG at %s%.50s:%d\n", prefix, (char *)(long)f.filename, f.line);
558 #ifdef CONFIG_BUG
559 void out_of_line_bug(void)
561 BUG();
563 EXPORT_SYMBOL(out_of_line_bug);
564 #endif
566 static DEFINE_SPINLOCK(die_lock);
567 static int die_owner = -1;
568 static unsigned int die_nest_count;
570 unsigned __kprobes long oops_begin(void)
572 int cpu = smp_processor_id();
573 unsigned long flags;
575 oops_enter();
577 /* racy, but better than risking deadlock. */
578 local_irq_save(flags);
579 if (!spin_trylock(&die_lock)) {
580 if (cpu == die_owner)
581 /* nested oops. should stop eventually */;
582 else
583 spin_lock(&die_lock);
585 die_nest_count++;
586 die_owner = cpu;
587 console_verbose();
588 bust_spinlocks(1);
589 return flags;
592 void __kprobes oops_end(unsigned long flags)
594 die_owner = -1;
595 bust_spinlocks(0);
596 die_nest_count--;
597 if (die_nest_count)
598 /* We still own the lock */
599 local_irq_restore(flags);
600 else
601 /* Nest count reaches zero, release the lock. */
602 spin_unlock_irqrestore(&die_lock, flags);
603 if (panic_on_oops)
604 panic("Fatal exception");
605 oops_exit();
608 void __kprobes __die(const char * str, struct pt_regs * regs, long err)
610 static int die_counter;
611 printk(KERN_EMERG "%s: %04lx [%u] ", str, err & 0xffff,++die_counter);
612 #ifdef CONFIG_PREEMPT
613 printk("PREEMPT ");
614 #endif
615 #ifdef CONFIG_SMP
616 printk("SMP ");
617 #endif
618 #ifdef CONFIG_DEBUG_PAGEALLOC
619 printk("DEBUG_PAGEALLOC");
620 #endif
621 printk("\n");
622 notify_die(DIE_OOPS, str, regs, err, current->thread.trap_no, SIGSEGV);
623 show_registers(regs);
624 /* Executive summary in case the oops scrolled away */
625 printk(KERN_ALERT "RIP ");
626 printk_address(regs->rip);
627 printk(" RSP <%016lx>\n", regs->rsp);
628 if (kexec_should_crash(current))
629 crash_kexec(regs);
632 void die(const char * str, struct pt_regs * regs, long err)
634 unsigned long flags = oops_begin();
636 handle_BUG(regs);
637 __die(str, regs, err);
638 oops_end(flags);
639 do_exit(SIGSEGV);
642 void __kprobes die_nmi(char *str, struct pt_regs *regs, int do_panic)
644 unsigned long flags = oops_begin();
647 * We are in trouble anyway, lets at least try
648 * to get a message out.
650 printk(str, smp_processor_id());
651 show_registers(regs);
652 if (kexec_should_crash(current))
653 crash_kexec(regs);
654 if (do_panic || panic_on_oops)
655 panic("Non maskable interrupt");
656 oops_end(flags);
657 nmi_exit();
658 local_irq_enable();
659 do_exit(SIGSEGV);
662 static void __kprobes do_trap(int trapnr, int signr, char *str,
663 struct pt_regs * regs, long error_code,
664 siginfo_t *info)
666 struct task_struct *tsk = current;
668 tsk->thread.error_code = error_code;
669 tsk->thread.trap_no = trapnr;
671 if (user_mode(regs)) {
672 if (exception_trace && unhandled_signal(tsk, signr))
673 printk(KERN_INFO
674 "%s[%d] trap %s rip:%lx rsp:%lx error:%lx\n",
675 tsk->comm, tsk->pid, str,
676 regs->rip, regs->rsp, error_code);
678 if (info)
679 force_sig_info(signr, info, tsk);
680 else
681 force_sig(signr, tsk);
682 return;
686 /* kernel trap */
688 const struct exception_table_entry *fixup;
689 fixup = search_exception_tables(regs->rip);
690 if (fixup)
691 regs->rip = fixup->fixup;
692 else
693 die(str, regs, error_code);
694 return;
698 #define DO_ERROR(trapnr, signr, str, name) \
699 asmlinkage void do_##name(struct pt_regs * regs, long error_code) \
701 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
702 == NOTIFY_STOP) \
703 return; \
704 conditional_sti(regs); \
705 do_trap(trapnr, signr, str, regs, error_code, NULL); \
708 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
709 asmlinkage void do_##name(struct pt_regs * regs, long error_code) \
711 siginfo_t info; \
712 info.si_signo = signr; \
713 info.si_errno = 0; \
714 info.si_code = sicode; \
715 info.si_addr = (void __user *)siaddr; \
716 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
717 == NOTIFY_STOP) \
718 return; \
719 conditional_sti(regs); \
720 do_trap(trapnr, signr, str, regs, error_code, &info); \
723 DO_ERROR_INFO( 0, SIGFPE, "divide error", divide_error, FPE_INTDIV, regs->rip)
724 DO_ERROR( 4, SIGSEGV, "overflow", overflow)
725 DO_ERROR( 5, SIGSEGV, "bounds", bounds)
726 DO_ERROR_INFO( 6, SIGILL, "invalid opcode", invalid_op, ILL_ILLOPN, regs->rip)
727 DO_ERROR( 7, SIGSEGV, "device not available", device_not_available)
728 DO_ERROR( 9, SIGFPE, "coprocessor segment overrun", coprocessor_segment_overrun)
729 DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
730 DO_ERROR(11, SIGBUS, "segment not present", segment_not_present)
731 DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0)
732 DO_ERROR(18, SIGSEGV, "reserved", reserved)
734 /* Runs on IST stack */
735 asmlinkage void do_stack_segment(struct pt_regs *regs, long error_code)
737 if (notify_die(DIE_TRAP, "stack segment", regs, error_code,
738 12, SIGBUS) == NOTIFY_STOP)
739 return;
740 preempt_conditional_sti(regs);
741 do_trap(12, SIGBUS, "stack segment", regs, error_code, NULL);
742 preempt_conditional_cli(regs);
745 asmlinkage void do_double_fault(struct pt_regs * regs, long error_code)
747 static const char str[] = "double fault";
748 struct task_struct *tsk = current;
750 /* Return not checked because double check cannot be ignored */
751 notify_die(DIE_TRAP, str, regs, error_code, 8, SIGSEGV);
753 tsk->thread.error_code = error_code;
754 tsk->thread.trap_no = 8;
756 /* This is always a kernel trap and never fixable (and thus must
757 never return). */
758 for (;;)
759 die(str, regs, error_code);
762 asmlinkage void __kprobes do_general_protection(struct pt_regs * regs,
763 long error_code)
765 struct task_struct *tsk = current;
767 conditional_sti(regs);
769 tsk->thread.error_code = error_code;
770 tsk->thread.trap_no = 13;
772 if (user_mode(regs)) {
773 if (exception_trace && unhandled_signal(tsk, SIGSEGV))
774 printk(KERN_INFO
775 "%s[%d] general protection rip:%lx rsp:%lx error:%lx\n",
776 tsk->comm, tsk->pid,
777 regs->rip, regs->rsp, error_code);
779 force_sig(SIGSEGV, tsk);
780 return;
783 /* kernel gp */
785 const struct exception_table_entry *fixup;
786 fixup = search_exception_tables(regs->rip);
787 if (fixup) {
788 regs->rip = fixup->fixup;
789 return;
791 if (notify_die(DIE_GPF, "general protection fault", regs,
792 error_code, 13, SIGSEGV) == NOTIFY_STOP)
793 return;
794 die("general protection fault", regs, error_code);
798 static __kprobes void
799 mem_parity_error(unsigned char reason, struct pt_regs * regs)
801 printk(KERN_EMERG "Uhhuh. NMI received for unknown reason %02x.\n",
802 reason);
803 printk(KERN_EMERG "You have some hardware problem, likely on the PCI bus.\n");
805 if (panic_on_unrecovered_nmi)
806 panic("NMI: Not continuing");
808 printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
810 /* Clear and disable the memory parity error line. */
811 reason = (reason & 0xf) | 4;
812 outb(reason, 0x61);
815 static __kprobes void
816 io_check_error(unsigned char reason, struct pt_regs * regs)
818 printk("NMI: IOCK error (debug interrupt?)\n");
819 show_registers(regs);
821 /* Re-enable the IOCK line, wait for a few seconds */
822 reason = (reason & 0xf) | 8;
823 outb(reason, 0x61);
824 mdelay(2000);
825 reason &= ~8;
826 outb(reason, 0x61);
829 static __kprobes void
830 unknown_nmi_error(unsigned char reason, struct pt_regs * regs)
832 printk(KERN_EMERG "Uhhuh. NMI received for unknown reason %02x.\n",
833 reason);
834 printk(KERN_EMERG "Do you have a strange power saving mode enabled?\n");
836 if (panic_on_unrecovered_nmi)
837 panic("NMI: Not continuing");
839 printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
842 /* Runs on IST stack. This code must keep interrupts off all the time.
843 Nested NMIs are prevented by the CPU. */
844 asmlinkage __kprobes void default_do_nmi(struct pt_regs *regs)
846 unsigned char reason = 0;
847 int cpu;
849 cpu = smp_processor_id();
851 /* Only the BSP gets external NMIs from the system. */
852 if (!cpu)
853 reason = get_nmi_reason();
855 if (!(reason & 0xc0)) {
856 if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 2, SIGINT)
857 == NOTIFY_STOP)
858 return;
860 * Ok, so this is none of the documented NMI sources,
861 * so it must be the NMI watchdog.
863 if (nmi_watchdog_tick(regs,reason))
864 return;
865 if (!do_nmi_callback(regs,cpu))
866 unknown_nmi_error(reason, regs);
868 return;
870 if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP)
871 return;
873 /* AK: following checks seem to be broken on modern chipsets. FIXME */
875 if (reason & 0x80)
876 mem_parity_error(reason, regs);
877 if (reason & 0x40)
878 io_check_error(reason, regs);
881 /* runs on IST stack. */
882 asmlinkage void __kprobes do_int3(struct pt_regs * regs, long error_code)
884 if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP) == NOTIFY_STOP) {
885 return;
887 preempt_conditional_sti(regs);
888 do_trap(3, SIGTRAP, "int3", regs, error_code, NULL);
889 preempt_conditional_cli(regs);
892 /* Help handler running on IST stack to switch back to user stack
893 for scheduling or signal handling. The actual stack switch is done in
894 entry.S */
895 asmlinkage __kprobes struct pt_regs *sync_regs(struct pt_regs *eregs)
897 struct pt_regs *regs = eregs;
898 /* Did already sync */
899 if (eregs == (struct pt_regs *)eregs->rsp)
901 /* Exception from user space */
902 else if (user_mode(eregs))
903 regs = task_pt_regs(current);
904 /* Exception from kernel and interrupts are enabled. Move to
905 kernel process stack. */
906 else if (eregs->eflags & X86_EFLAGS_IF)
907 regs = (struct pt_regs *)(eregs->rsp -= sizeof(struct pt_regs));
908 if (eregs != regs)
909 *regs = *eregs;
910 return regs;
913 /* runs on IST stack. */
914 asmlinkage void __kprobes do_debug(struct pt_regs * regs,
915 unsigned long error_code)
917 unsigned long condition;
918 struct task_struct *tsk = current;
919 siginfo_t info;
921 get_debugreg(condition, 6);
923 if (notify_die(DIE_DEBUG, "debug", regs, condition, error_code,
924 SIGTRAP) == NOTIFY_STOP)
925 return;
927 preempt_conditional_sti(regs);
929 /* Mask out spurious debug traps due to lazy DR7 setting */
930 if (condition & (DR_TRAP0|DR_TRAP1|DR_TRAP2|DR_TRAP3)) {
931 if (!tsk->thread.debugreg7) {
932 goto clear_dr7;
936 tsk->thread.debugreg6 = condition;
938 /* Mask out spurious TF errors due to lazy TF clearing */
939 if (condition & DR_STEP) {
941 * The TF error should be masked out only if the current
942 * process is not traced and if the TRAP flag has been set
943 * previously by a tracing process (condition detected by
944 * the PT_DTRACE flag); remember that the i386 TRAP flag
945 * can be modified by the process itself in user mode,
946 * allowing programs to debug themselves without the ptrace()
947 * interface.
949 if (!user_mode(regs))
950 goto clear_TF_reenable;
952 * Was the TF flag set by a debugger? If so, clear it now,
953 * so that register information is correct.
955 if (tsk->ptrace & PT_DTRACE) {
956 regs->eflags &= ~TF_MASK;
957 tsk->ptrace &= ~PT_DTRACE;
961 /* Ok, finally something we can handle */
962 tsk->thread.trap_no = 1;
963 tsk->thread.error_code = error_code;
964 info.si_signo = SIGTRAP;
965 info.si_errno = 0;
966 info.si_code = TRAP_BRKPT;
967 info.si_addr = user_mode(regs) ? (void __user *)regs->rip : NULL;
968 force_sig_info(SIGTRAP, &info, tsk);
970 clear_dr7:
971 set_debugreg(0UL, 7);
972 preempt_conditional_cli(regs);
973 return;
975 clear_TF_reenable:
976 set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
977 regs->eflags &= ~TF_MASK;
978 preempt_conditional_cli(regs);
981 static int kernel_math_error(struct pt_regs *regs, const char *str, int trapnr)
983 const struct exception_table_entry *fixup;
984 fixup = search_exception_tables(regs->rip);
985 if (fixup) {
986 regs->rip = fixup->fixup;
987 return 1;
989 notify_die(DIE_GPF, str, regs, 0, trapnr, SIGFPE);
990 /* Illegal floating point operation in the kernel */
991 current->thread.trap_no = trapnr;
992 die(str, regs, 0);
993 return 0;
997 * Note that we play around with the 'TS' bit in an attempt to get
998 * the correct behaviour even in the presence of the asynchronous
999 * IRQ13 behaviour
1001 asmlinkage void do_coprocessor_error(struct pt_regs *regs)
1003 void __user *rip = (void __user *)(regs->rip);
1004 struct task_struct * task;
1005 siginfo_t info;
1006 unsigned short cwd, swd;
1008 conditional_sti(regs);
1009 if (!user_mode(regs) &&
1010 kernel_math_error(regs, "kernel x87 math error", 16))
1011 return;
1014 * Save the info for the exception handler and clear the error.
1016 task = current;
1017 save_init_fpu(task);
1018 task->thread.trap_no = 16;
1019 task->thread.error_code = 0;
1020 info.si_signo = SIGFPE;
1021 info.si_errno = 0;
1022 info.si_code = __SI_FAULT;
1023 info.si_addr = rip;
1025 * (~cwd & swd) will mask out exceptions that are not set to unmasked
1026 * status. 0x3f is the exception bits in these regs, 0x200 is the
1027 * C1 reg you need in case of a stack fault, 0x040 is the stack
1028 * fault bit. We should only be taking one exception at a time,
1029 * so if this combination doesn't produce any single exception,
1030 * then we have a bad program that isn't synchronizing its FPU usage
1031 * and it will suffer the consequences since we won't be able to
1032 * fully reproduce the context of the exception
1034 cwd = get_fpu_cwd(task);
1035 swd = get_fpu_swd(task);
1036 switch (swd & ~cwd & 0x3f) {
1037 case 0x000:
1038 default:
1039 break;
1040 case 0x001: /* Invalid Op */
1042 * swd & 0x240 == 0x040: Stack Underflow
1043 * swd & 0x240 == 0x240: Stack Overflow
1044 * User must clear the SF bit (0x40) if set
1046 info.si_code = FPE_FLTINV;
1047 break;
1048 case 0x002: /* Denormalize */
1049 case 0x010: /* Underflow */
1050 info.si_code = FPE_FLTUND;
1051 break;
1052 case 0x004: /* Zero Divide */
1053 info.si_code = FPE_FLTDIV;
1054 break;
1055 case 0x008: /* Overflow */
1056 info.si_code = FPE_FLTOVF;
1057 break;
1058 case 0x020: /* Precision */
1059 info.si_code = FPE_FLTRES;
1060 break;
1062 force_sig_info(SIGFPE, &info, task);
1065 asmlinkage void bad_intr(void)
1067 printk("bad interrupt");
1070 asmlinkage void do_simd_coprocessor_error(struct pt_regs *regs)
1072 void __user *rip = (void __user *)(regs->rip);
1073 struct task_struct * task;
1074 siginfo_t info;
1075 unsigned short mxcsr;
1077 conditional_sti(regs);
1078 if (!user_mode(regs) &&
1079 kernel_math_error(regs, "kernel simd math error", 19))
1080 return;
1083 * Save the info for the exception handler and clear the error.
1085 task = current;
1086 save_init_fpu(task);
1087 task->thread.trap_no = 19;
1088 task->thread.error_code = 0;
1089 info.si_signo = SIGFPE;
1090 info.si_errno = 0;
1091 info.si_code = __SI_FAULT;
1092 info.si_addr = rip;
1094 * The SIMD FPU exceptions are handled a little differently, as there
1095 * is only a single status/control register. Thus, to determine which
1096 * unmasked exception was caught we must mask the exception mask bits
1097 * at 0x1f80, and then use these to mask the exception bits at 0x3f.
1099 mxcsr = get_fpu_mxcsr(task);
1100 switch (~((mxcsr & 0x1f80) >> 7) & (mxcsr & 0x3f)) {
1101 case 0x000:
1102 default:
1103 break;
1104 case 0x001: /* Invalid Op */
1105 info.si_code = FPE_FLTINV;
1106 break;
1107 case 0x002: /* Denormalize */
1108 case 0x010: /* Underflow */
1109 info.si_code = FPE_FLTUND;
1110 break;
1111 case 0x004: /* Zero Divide */
1112 info.si_code = FPE_FLTDIV;
1113 break;
1114 case 0x008: /* Overflow */
1115 info.si_code = FPE_FLTOVF;
1116 break;
1117 case 0x020: /* Precision */
1118 info.si_code = FPE_FLTRES;
1119 break;
1121 force_sig_info(SIGFPE, &info, task);
1124 asmlinkage void do_spurious_interrupt_bug(struct pt_regs * regs)
1128 asmlinkage void __attribute__((weak)) smp_thermal_interrupt(void)
1132 asmlinkage void __attribute__((weak)) mce_threshold_interrupt(void)
1137 * 'math_state_restore()' saves the current math information in the
1138 * old math state array, and gets the new ones from the current task
1140 * Careful.. There are problems with IBM-designed IRQ13 behaviour.
1141 * Don't touch unless you *really* know how it works.
1143 asmlinkage void math_state_restore(void)
1145 struct task_struct *me = current;
1146 clts(); /* Allow maths ops (or we recurse) */
1148 if (!used_math())
1149 init_fpu(me);
1150 restore_fpu_checking(&me->thread.i387.fxsave);
1151 task_thread_info(me)->status |= TS_USEDFPU;
1152 me->fpu_counter++;
1155 void __init trap_init(void)
1157 set_intr_gate(0,&divide_error);
1158 set_intr_gate_ist(1,&debug,DEBUG_STACK);
1159 set_intr_gate_ist(2,&nmi,NMI_STACK);
1160 set_system_gate_ist(3,&int3,DEBUG_STACK); /* int3 can be called from all */
1161 set_system_gate(4,&overflow); /* int4 can be called from all */
1162 set_intr_gate(5,&bounds);
1163 set_intr_gate(6,&invalid_op);
1164 set_intr_gate(7,&device_not_available);
1165 set_intr_gate_ist(8,&double_fault, DOUBLEFAULT_STACK);
1166 set_intr_gate(9,&coprocessor_segment_overrun);
1167 set_intr_gate(10,&invalid_TSS);
1168 set_intr_gate(11,&segment_not_present);
1169 set_intr_gate_ist(12,&stack_segment,STACKFAULT_STACK);
1170 set_intr_gate(13,&general_protection);
1171 set_intr_gate(14,&page_fault);
1172 set_intr_gate(15,&spurious_interrupt_bug);
1173 set_intr_gate(16,&coprocessor_error);
1174 set_intr_gate(17,&alignment_check);
1175 #ifdef CONFIG_X86_MCE
1176 set_intr_gate_ist(18,&machine_check, MCE_STACK);
1177 #endif
1178 set_intr_gate(19,&simd_coprocessor_error);
1180 #ifdef CONFIG_IA32_EMULATION
1181 set_system_gate(IA32_SYSCALL_VECTOR, ia32_syscall);
1182 #endif
1185 * Should be a barrier for any external CPU state.
1187 cpu_init();
1191 static int __init oops_setup(char *s)
1193 if (!s)
1194 return -EINVAL;
1195 if (!strcmp(s, "panic"))
1196 panic_on_oops = 1;
1197 return 0;
1199 early_param("oops", oops_setup);
1201 static int __init kstack_setup(char *s)
1203 if (!s)
1204 return -EINVAL;
1205 kstack_depth_to_print = simple_strtoul(s,NULL,0);
1206 return 0;
1208 early_param("kstack", kstack_setup);
1210 #ifdef CONFIG_STACK_UNWIND
1211 static int __init call_trace_setup(char *s)
1213 if (!s)
1214 return -EINVAL;
1215 if (strcmp(s, "old") == 0)
1216 call_trace = -1;
1217 else if (strcmp(s, "both") == 0)
1218 call_trace = 0;
1219 else if (strcmp(s, "newfallback") == 0)
1220 call_trace = 1;
1221 else if (strcmp(s, "new") == 0)
1222 call_trace = 2;
1223 return 0;
1225 early_param("call_trace", call_trace_setup);
1226 #endif