2 * linux/arch/i386/traps.c
4 * Copyright (C) 1991, 1992 Linus Torvalds
6 * Pentium III FXSR, SSE support
7 * Gareth Hughes <gareth@valinux.com>, May 2000
11 * 'Traps.c' handles hardware traps and faults after we have saved some
14 #include <linux/sched.h>
15 #include <linux/kernel.h>
16 #include <linux/string.h>
17 #include <linux/errno.h>
18 #include <linux/timer.h>
20 #include <linux/init.h>
21 #include <linux/delay.h>
22 #include <linux/spinlock.h>
23 #include <linux/interrupt.h>
24 #include <linux/highmem.h>
25 #include <linux/kallsyms.h>
26 #include <linux/ptrace.h>
27 #include <linux/utsname.h>
28 #include <linux/kprobes.h>
29 #include <linux/kexec.h>
30 #include <linux/unwind.h>
31 #include <linux/uaccess.h>
32 #include <linux/nmi.h>
33 #include <linux/bug.h>
36 #include <linux/ioport.h>
37 #include <linux/eisa.h>
41 #include <linux/mca.h>
44 #include <asm/processor.h>
45 #include <asm/system.h>
47 #include <asm/atomic.h>
48 #include <asm/debugreg.h>
52 #include <asm/unwind.h>
54 #include <asm/arch_hooks.h>
55 #include <asm/kdebug.h>
56 #include <asm/stacktrace.h>
58 #include <linux/module.h>
60 #include "mach_traps.h"
62 int panic_on_unrecovered_nmi;
64 asmlinkage int system_call(void);
66 /* Do we ignore FPU interrupts ? */
67 char ignore_fpu_irq = 0;
70 * The IDT has to be page-aligned to simplify the Pentium
71 * F0 0F bug workaround.. We have a special link segment
74 struct desc_struct idt_table[256] __attribute__((__section__(".data.idt"))) = { {0, 0}, };
76 asmlinkage void divide_error(void);
77 asmlinkage void debug(void);
78 asmlinkage void nmi(void);
79 asmlinkage void int3(void);
80 asmlinkage void overflow(void);
81 asmlinkage void bounds(void);
82 asmlinkage void invalid_op(void);
83 asmlinkage void device_not_available(void);
84 asmlinkage void coprocessor_segment_overrun(void);
85 asmlinkage void invalid_TSS(void);
86 asmlinkage void segment_not_present(void);
87 asmlinkage void stack_segment(void);
88 asmlinkage void general_protection(void);
89 asmlinkage void page_fault(void);
90 asmlinkage void coprocessor_error(void);
91 asmlinkage void simd_coprocessor_error(void);
92 asmlinkage void alignment_check(void);
93 asmlinkage void spurious_interrupt_bug(void);
94 asmlinkage void machine_check(void);
96 int kstack_depth_to_print = 24;
97 ATOMIC_NOTIFIER_HEAD(i386die_chain);
99 int register_die_notifier(struct notifier_block *nb)
102 return atomic_notifier_chain_register(&i386die_chain, nb);
104 EXPORT_SYMBOL(register_die_notifier); /* used modular by kdb */
106 int unregister_die_notifier(struct notifier_block *nb)
108 return atomic_notifier_chain_unregister(&i386die_chain, nb);
110 EXPORT_SYMBOL(unregister_die_notifier); /* used modular by kdb */
112 static inline int valid_stack_ptr(struct thread_info *tinfo, void *p)
114 return p > (void *)tinfo &&
115 p < (void *)tinfo + THREAD_SIZE - 3;
118 static inline unsigned long print_context_stack(struct thread_info *tinfo,
119 unsigned long *stack, unsigned long ebp,
120 struct stacktrace_ops *ops, void *data)
124 #ifdef CONFIG_FRAME_POINTER
125 while (valid_stack_ptr(tinfo, (void *)ebp)) {
126 unsigned long new_ebp;
127 addr = *(unsigned long *)(ebp + 4);
128 ops->address(data, addr);
130 * break out of recursive entries (such as
131 * end_of_stack_stop_unwind_function). Also,
132 * we can never allow a frame pointer to
135 new_ebp = *(unsigned long *)ebp;
141 while (valid_stack_ptr(tinfo, stack)) {
143 if (__kernel_text_address(addr))
144 ops->address(data, addr);
150 #define MSG(msg) ops->warning(data, msg)
152 void dump_trace(struct task_struct *task, struct pt_regs *regs,
153 unsigned long *stack,
154 struct stacktrace_ops *ops, void *data)
156 unsigned long ebp = 0;
164 if (task && task != current)
165 stack = (unsigned long *)task->thread.esp;
168 #ifdef CONFIG_FRAME_POINTER
170 if (task == current) {
171 /* Grab ebp right from our regs */
172 asm ("movl %%ebp, %0" : "=r" (ebp) : );
174 /* ebp is the last reg pushed by switch_to */
175 ebp = *(unsigned long *) task->thread.esp;
181 struct thread_info *context;
182 context = (struct thread_info *)
183 ((unsigned long)stack & (~(THREAD_SIZE - 1)));
184 ebp = print_context_stack(context, stack, ebp, ops, data);
185 /* Should be after the line below, but somewhere
186 in early boot context comes out corrupted and we
187 can't reference it -AK */
188 if (ops->stack(data, "IRQ") < 0)
190 stack = (unsigned long*)context->previous_esp;
193 touch_nmi_watchdog();
196 EXPORT_SYMBOL(dump_trace);
199 print_trace_warning_symbol(void *data, char *msg, unsigned long symbol)
202 print_symbol(msg, symbol);
206 static void print_trace_warning(void *data, char *msg)
208 printk("%s%s\n", (char *)data, msg);
211 static int print_trace_stack(void *data, char *name)
217 * Print one address/symbol entries per line.
219 static void print_trace_address(void *data, unsigned long addr)
221 printk("%s [<%08lx>] ", (char *)data, addr);
222 print_symbol("%s\n", addr);
225 static struct stacktrace_ops print_trace_ops = {
226 .warning = print_trace_warning,
227 .warning_symbol = print_trace_warning_symbol,
228 .stack = print_trace_stack,
229 .address = print_trace_address,
233 show_trace_log_lvl(struct task_struct *task, struct pt_regs *regs,
234 unsigned long * stack, char *log_lvl)
236 dump_trace(task, regs, stack, &print_trace_ops, log_lvl);
237 printk("%s =======================\n", log_lvl);
240 void show_trace(struct task_struct *task, struct pt_regs *regs,
241 unsigned long * stack)
243 show_trace_log_lvl(task, regs, stack, "");
246 static void show_stack_log_lvl(struct task_struct *task, struct pt_regs *regs,
247 unsigned long *esp, char *log_lvl)
249 unsigned long *stack;
254 esp = (unsigned long*)task->thread.esp;
256 esp = (unsigned long *)&esp;
260 for(i = 0; i < kstack_depth_to_print; i++) {
261 if (kstack_end(stack))
263 if (i && ((i % 8) == 0))
264 printk("\n%s ", log_lvl);
265 printk("%08lx ", *stack++);
267 printk("\n%sCall Trace:\n", log_lvl);
268 show_trace_log_lvl(task, regs, esp, log_lvl);
271 void show_stack(struct task_struct *task, unsigned long *esp)
274 show_stack_log_lvl(task, NULL, esp, "");
278 * The architecture-independent dump_stack generator
280 void dump_stack(void)
284 show_trace(current, NULL, &stack);
287 EXPORT_SYMBOL(dump_stack);
289 void show_registers(struct pt_regs *regs)
296 esp = (unsigned long) (®s->esp);
298 if (user_mode_vm(regs)) {
301 ss = regs->xss & 0xffff;
304 printk(KERN_EMERG "CPU: %d\n"
305 KERN_EMERG "EIP: %04x:[<%08lx>] %s VLI\n"
306 KERN_EMERG "EFLAGS: %08lx (%s %.*s)\n",
307 smp_processor_id(), 0xffff & regs->xcs, regs->eip,
308 print_tainted(), regs->eflags, init_utsname()->release,
309 (int)strcspn(init_utsname()->version, " "),
310 init_utsname()->version);
311 print_symbol(KERN_EMERG "EIP is at %s\n", regs->eip);
312 printk(KERN_EMERG "eax: %08lx ebx: %08lx ecx: %08lx edx: %08lx\n",
313 regs->eax, regs->ebx, regs->ecx, regs->edx);
314 printk(KERN_EMERG "esi: %08lx edi: %08lx ebp: %08lx esp: %08lx\n",
315 regs->esi, regs->edi, regs->ebp, esp);
316 printk(KERN_EMERG "ds: %04x es: %04x ss: %04x\n",
317 regs->xds & 0xffff, regs->xes & 0xffff, ss);
318 printk(KERN_EMERG "Process %.*s (pid: %d, ti=%p task=%p task.ti=%p)",
319 TASK_COMM_LEN, current->comm, current->pid,
320 current_thread_info(), current, current->thread_info);
322 * When in-kernel, we also print out the stack and code at the
323 * time of the fault..
330 printk("\n" KERN_EMERG "Stack: ");
331 show_stack_log_lvl(NULL, regs, (unsigned long *)esp, KERN_EMERG);
333 printk(KERN_EMERG "Code: ");
335 eip = (u8 *)regs->eip - 43;
336 if (eip < (u8 *)PAGE_OFFSET ||
337 probe_kernel_address(eip, c)) {
338 /* try starting at EIP */
339 eip = (u8 *)regs->eip;
342 for (i = 0; i < code_bytes; i++, eip++) {
343 if (eip < (u8 *)PAGE_OFFSET ||
344 probe_kernel_address(eip, c)) {
345 printk(" Bad EIP value.");
348 if (eip == (u8 *)regs->eip)
349 printk("<%02x> ", c);
357 int is_valid_bugaddr(unsigned long eip)
361 if (eip < PAGE_OFFSET)
363 if (probe_kernel_address((unsigned short *)eip, ud2))
366 return ud2 == 0x0b0f;
370 * This is gone through when something in the kernel has done something bad and
371 * is about to be terminated.
373 void die(const char * str, struct pt_regs * regs, long err)
378 int lock_owner_depth;
380 .lock = __SPIN_LOCK_UNLOCKED(die.lock),
382 .lock_owner_depth = 0
384 static int die_counter;
389 if (die.lock_owner != raw_smp_processor_id()) {
391 spin_lock_irqsave(&die.lock, flags);
392 die.lock_owner = smp_processor_id();
393 die.lock_owner_depth = 0;
397 local_save_flags(flags);
399 if (++die.lock_owner_depth < 3) {
404 report_bug(regs->eip);
406 printk(KERN_EMERG "%s: %04lx [#%d]\n", str, err & 0xffff, ++die_counter);
407 #ifdef CONFIG_PREEMPT
408 printk(KERN_EMERG "PREEMPT ");
417 #ifdef CONFIG_DEBUG_PAGEALLOC
420 printk("DEBUG_PAGEALLOC");
425 if (notify_die(DIE_OOPS, str, regs, err,
426 current->thread.trap_no, SIGSEGV) !=
428 show_registers(regs);
429 /* Executive summary in case the oops scrolled away */
430 esp = (unsigned long) (®s->esp);
432 if (user_mode(regs)) {
434 ss = regs->xss & 0xffff;
436 printk(KERN_EMERG "EIP: [<%08lx>] ", regs->eip);
437 print_symbol("%s", regs->eip);
438 printk(" SS:ESP %04x:%08lx\n", ss, esp);
443 printk(KERN_EMERG "Recursive die() failure, output suppressed\n");
447 spin_unlock_irqrestore(&die.lock, flags);
452 if (kexec_should_crash(current))
456 panic("Fatal exception in interrupt");
459 panic("Fatal exception");
465 static inline void die_if_kernel(const char * str, struct pt_regs * regs, long err)
467 if (!user_mode_vm(regs))
471 static void __kprobes do_trap(int trapnr, int signr, char *str, int vm86,
472 struct pt_regs * regs, long error_code,
475 struct task_struct *tsk = current;
476 tsk->thread.error_code = error_code;
477 tsk->thread.trap_no = trapnr;
479 if (regs->eflags & VM_MASK) {
485 if (!user_mode(regs))
490 force_sig_info(signr, info, tsk);
492 force_sig(signr, tsk);
497 if (!fixup_exception(regs))
498 die(str, regs, error_code);
503 int ret = handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code, trapnr);
504 if (ret) goto trap_signal;
509 #define DO_ERROR(trapnr, signr, str, name) \
510 fastcall void do_##name(struct pt_regs * regs, long error_code) \
512 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
515 do_trap(trapnr, signr, str, 0, regs, error_code, NULL); \
518 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
519 fastcall void do_##name(struct pt_regs * regs, long error_code) \
522 info.si_signo = signr; \
524 info.si_code = sicode; \
525 info.si_addr = (void __user *)siaddr; \
526 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
529 do_trap(trapnr, signr, str, 0, regs, error_code, &info); \
532 #define DO_VM86_ERROR(trapnr, signr, str, name) \
533 fastcall void do_##name(struct pt_regs * regs, long error_code) \
535 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
538 do_trap(trapnr, signr, str, 1, regs, error_code, NULL); \
541 #define DO_VM86_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
542 fastcall void do_##name(struct pt_regs * regs, long error_code) \
545 info.si_signo = signr; \
547 info.si_code = sicode; \
548 info.si_addr = (void __user *)siaddr; \
549 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
552 do_trap(trapnr, signr, str, 1, regs, error_code, &info); \
555 DO_VM86_ERROR_INFO( 0, SIGFPE, "divide error", divide_error, FPE_INTDIV, regs->eip)
556 #ifndef CONFIG_KPROBES
557 DO_VM86_ERROR( 3, SIGTRAP, "int3", int3)
559 DO_VM86_ERROR( 4, SIGSEGV, "overflow", overflow)
560 DO_VM86_ERROR( 5, SIGSEGV, "bounds", bounds)
561 DO_ERROR_INFO( 6, SIGILL, "invalid opcode", invalid_op, ILL_ILLOPN, regs->eip)
562 DO_ERROR( 9, SIGFPE, "coprocessor segment overrun", coprocessor_segment_overrun)
563 DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
564 DO_ERROR(11, SIGBUS, "segment not present", segment_not_present)
565 DO_ERROR(12, SIGBUS, "stack segment", stack_segment)
566 DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0)
567 DO_ERROR_INFO(32, SIGSEGV, "iret exception", iret_error, ILL_BADSTK, 0)
569 fastcall void __kprobes do_general_protection(struct pt_regs * regs,
573 struct tss_struct *tss = &per_cpu(init_tss, cpu);
574 struct thread_struct *thread = ¤t->thread;
577 * Perform the lazy TSS's I/O bitmap copy. If the TSS has an
578 * invalid offset set (the LAZY one) and the faulting thread has
579 * a valid I/O bitmap pointer, we copy the I/O bitmap in the TSS
580 * and we set the offset field correctly. Then we let the CPU to
581 * restart the faulting instruction.
583 if (tss->io_bitmap_base == INVALID_IO_BITMAP_OFFSET_LAZY &&
584 thread->io_bitmap_ptr) {
585 memcpy(tss->io_bitmap, thread->io_bitmap_ptr,
586 thread->io_bitmap_max);
588 * If the previously set map was extending to higher ports
589 * than the current one, pad extra space with 0xff (no access).
591 if (thread->io_bitmap_max < tss->io_bitmap_max)
592 memset((char *) tss->io_bitmap +
593 thread->io_bitmap_max, 0xff,
594 tss->io_bitmap_max - thread->io_bitmap_max);
595 tss->io_bitmap_max = thread->io_bitmap_max;
596 tss->io_bitmap_base = IO_BITMAP_OFFSET;
597 tss->io_bitmap_owner = thread;
603 current->thread.error_code = error_code;
604 current->thread.trap_no = 13;
606 if (regs->eflags & VM_MASK)
609 if (!user_mode(regs))
612 current->thread.error_code = error_code;
613 current->thread.trap_no = 13;
614 force_sig(SIGSEGV, current);
619 handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
623 if (!fixup_exception(regs)) {
624 if (notify_die(DIE_GPF, "general protection fault", regs,
625 error_code, 13, SIGSEGV) == NOTIFY_STOP)
627 die("general protection fault", regs, error_code);
631 static __kprobes void
632 mem_parity_error(unsigned char reason, struct pt_regs * regs)
634 printk(KERN_EMERG "Uhhuh. NMI received for unknown reason %02x on "
635 "CPU %d.\n", reason, smp_processor_id());
636 printk(KERN_EMERG "You have some hardware problem, likely on the PCI bus.\n");
637 if (panic_on_unrecovered_nmi)
638 panic("NMI: Not continuing");
640 printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
642 /* Clear and disable the memory parity error line. */
643 clear_mem_error(reason);
646 static __kprobes void
647 io_check_error(unsigned char reason, struct pt_regs * regs)
651 printk(KERN_EMERG "NMI: IOCK error (debug interrupt?)\n");
652 show_registers(regs);
654 /* Re-enable the IOCK line, wait for a few seconds */
655 reason = (reason & 0xf) | 8;
658 while (--i) udelay(1000);
663 static __kprobes void
664 unknown_nmi_error(unsigned char reason, struct pt_regs * regs)
667 /* Might actually be able to figure out what the guilty party
674 printk(KERN_EMERG "Uhhuh. NMI received for unknown reason %02x on "
675 "CPU %d.\n", reason, smp_processor_id());
676 printk(KERN_EMERG "Do you have a strange power saving mode enabled?\n");
677 if (panic_on_unrecovered_nmi)
678 panic("NMI: Not continuing");
680 printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
683 static DEFINE_SPINLOCK(nmi_print_lock);
685 void __kprobes die_nmi(struct pt_regs *regs, const char *msg)
687 if (notify_die(DIE_NMIWATCHDOG, msg, regs, 0, 2, SIGINT) ==
691 spin_lock(&nmi_print_lock);
693 * We are in trouble anyway, lets at least try
694 * to get a message out.
697 printk(KERN_EMERG "%s", msg);
698 printk(" on CPU%d, eip %08lx, registers:\n",
699 smp_processor_id(), regs->eip);
700 show_registers(regs);
702 spin_unlock(&nmi_print_lock);
705 /* If we are in kernel we are probably nested up pretty bad
706 * and might aswell get out now while we still can.
708 if (!user_mode_vm(regs)) {
709 current->thread.trap_no = 2;
716 static __kprobes void default_do_nmi(struct pt_regs * regs)
718 unsigned char reason = 0;
720 /* Only the BSP gets external NMIs from the system. */
721 if (!smp_processor_id())
722 reason = get_nmi_reason();
724 if (!(reason & 0xc0)) {
725 if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 2, SIGINT)
728 #ifdef CONFIG_X86_LOCAL_APIC
730 * Ok, so this is none of the documented NMI sources,
731 * so it must be the NMI watchdog.
733 if (nmi_watchdog_tick(regs, reason))
735 if (!do_nmi_callback(regs, smp_processor_id()))
737 unknown_nmi_error(reason, regs);
741 if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP)
744 mem_parity_error(reason, regs);
746 io_check_error(reason, regs);
748 * Reassert NMI in case it became active meanwhile
749 * as it's edge-triggered.
754 fastcall __kprobes void do_nmi(struct pt_regs * regs, long error_code)
760 cpu = smp_processor_id();
764 default_do_nmi(regs);
769 #ifdef CONFIG_KPROBES
770 fastcall void __kprobes do_int3(struct pt_regs *regs, long error_code)
772 if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP)
775 /* This is an interrupt gate, because kprobes wants interrupts
776 disabled. Normal trap handlers don't. */
777 restore_interrupts(regs);
778 do_trap(3, SIGTRAP, "int3", 1, regs, error_code, NULL);
783 * Our handling of the processor debug registers is non-trivial.
784 * We do not clear them on entry and exit from the kernel. Therefore
785 * it is possible to get a watchpoint trap here from inside the kernel.
786 * However, the code in ./ptrace.c has ensured that the user can
787 * only set watchpoints on userspace addresses. Therefore the in-kernel
788 * watchpoint trap can only occur in code which is reading/writing
789 * from user space. Such code must not hold kernel locks (since it
790 * can equally take a page fault), therefore it is safe to call
791 * force_sig_info even though that claims and releases locks.
793 * Code in ./signal.c ensures that the debug control register
794 * is restored before we deliver any signal, and therefore that
795 * user code runs with the correct debug control register even though
798 * Being careful here means that we don't have to be as careful in a
799 * lot of more complicated places (task switching can be a bit lazy
800 * about restoring all the debug state, and ptrace doesn't have to
801 * find every occurrence of the TF bit that could be saved away even
804 fastcall void __kprobes do_debug(struct pt_regs * regs, long error_code)
806 unsigned int condition;
807 struct task_struct *tsk = current;
809 get_debugreg(condition, 6);
811 if (notify_die(DIE_DEBUG, "debug", regs, condition, error_code,
812 SIGTRAP) == NOTIFY_STOP)
814 /* It's safe to allow irq's after DR6 has been saved */
815 if (regs->eflags & X86_EFLAGS_IF)
818 /* Mask out spurious debug traps due to lazy DR7 setting */
819 if (condition & (DR_TRAP0|DR_TRAP1|DR_TRAP2|DR_TRAP3)) {
820 if (!tsk->thread.debugreg[7])
824 if (regs->eflags & VM_MASK)
827 /* Save debug status register where ptrace can see it */
828 tsk->thread.debugreg[6] = condition;
831 * Single-stepping through TF: make sure we ignore any events in
832 * kernel space (but re-enable TF when returning to user mode).
834 if (condition & DR_STEP) {
836 * We already checked v86 mode above, so we can
837 * check for kernel mode by just checking the CPL
840 if (!user_mode(regs))
841 goto clear_TF_reenable;
844 /* Ok, finally something we can handle */
845 send_sigtrap(tsk, regs, error_code);
847 /* Disable additional traps. They'll be re-enabled when
848 * the signal is delivered.
855 handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code, 1);
859 set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
860 regs->eflags &= ~TF_MASK;
865 * Note that we play around with the 'TS' bit in an attempt to get
866 * the correct behaviour even in the presence of the asynchronous
869 void math_error(void __user *eip)
871 struct task_struct * task;
873 unsigned short cwd, swd;
876 * Save the info for the exception handler and clear the error.
880 task->thread.trap_no = 16;
881 task->thread.error_code = 0;
882 info.si_signo = SIGFPE;
884 info.si_code = __SI_FAULT;
887 * (~cwd & swd) will mask out exceptions that are not set to unmasked
888 * status. 0x3f is the exception bits in these regs, 0x200 is the
889 * C1 reg you need in case of a stack fault, 0x040 is the stack
890 * fault bit. We should only be taking one exception at a time,
891 * so if this combination doesn't produce any single exception,
892 * then we have a bad program that isn't syncronizing its FPU usage
893 * and it will suffer the consequences since we won't be able to
894 * fully reproduce the context of the exception
896 cwd = get_fpu_cwd(task);
897 swd = get_fpu_swd(task);
898 switch (swd & ~cwd & 0x3f) {
899 case 0x000: /* No unmasked exception */
901 default: /* Multiple exceptions */
903 case 0x001: /* Invalid Op */
905 * swd & 0x240 == 0x040: Stack Underflow
906 * swd & 0x240 == 0x240: Stack Overflow
907 * User must clear the SF bit (0x40) if set
909 info.si_code = FPE_FLTINV;
911 case 0x002: /* Denormalize */
912 case 0x010: /* Underflow */
913 info.si_code = FPE_FLTUND;
915 case 0x004: /* Zero Divide */
916 info.si_code = FPE_FLTDIV;
918 case 0x008: /* Overflow */
919 info.si_code = FPE_FLTOVF;
921 case 0x020: /* Precision */
922 info.si_code = FPE_FLTRES;
925 force_sig_info(SIGFPE, &info, task);
928 fastcall void do_coprocessor_error(struct pt_regs * regs, long error_code)
931 math_error((void __user *)regs->eip);
934 static void simd_math_error(void __user *eip)
936 struct task_struct * task;
938 unsigned short mxcsr;
941 * Save the info for the exception handler and clear the error.
945 task->thread.trap_no = 19;
946 task->thread.error_code = 0;
947 info.si_signo = SIGFPE;
949 info.si_code = __SI_FAULT;
952 * The SIMD FPU exceptions are handled a little differently, as there
953 * is only a single status/control register. Thus, to determine which
954 * unmasked exception was caught we must mask the exception mask bits
955 * at 0x1f80, and then use these to mask the exception bits at 0x3f.
957 mxcsr = get_fpu_mxcsr(task);
958 switch (~((mxcsr & 0x1f80) >> 7) & (mxcsr & 0x3f)) {
962 case 0x001: /* Invalid Op */
963 info.si_code = FPE_FLTINV;
965 case 0x002: /* Denormalize */
966 case 0x010: /* Underflow */
967 info.si_code = FPE_FLTUND;
969 case 0x004: /* Zero Divide */
970 info.si_code = FPE_FLTDIV;
972 case 0x008: /* Overflow */
973 info.si_code = FPE_FLTOVF;
975 case 0x020: /* Precision */
976 info.si_code = FPE_FLTRES;
979 force_sig_info(SIGFPE, &info, task);
982 fastcall void do_simd_coprocessor_error(struct pt_regs * regs,
986 /* Handle SIMD FPU exceptions on PIII+ processors. */
988 simd_math_error((void __user *)regs->eip);
991 * Handle strange cache flush from user space exception
992 * in all other cases. This is undocumented behaviour.
994 if (regs->eflags & VM_MASK) {
995 handle_vm86_fault((struct kernel_vm86_regs *)regs,
999 current->thread.trap_no = 19;
1000 current->thread.error_code = error_code;
1001 die_if_kernel("cache flush denied", regs, error_code);
1002 force_sig(SIGSEGV, current);
1006 fastcall void do_spurious_interrupt_bug(struct pt_regs * regs,
1010 /* No need to warn about this any longer. */
1011 printk("Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
1015 fastcall unsigned long patch_espfix_desc(unsigned long uesp,
1018 int cpu = smp_processor_id();
1019 struct Xgt_desc_struct *cpu_gdt_descr = &per_cpu(cpu_gdt_descr, cpu);
1020 struct desc_struct *gdt = (struct desc_struct *)cpu_gdt_descr->address;
1021 unsigned long base = (kesp - uesp) & -THREAD_SIZE;
1022 unsigned long new_kesp = kesp - base;
1023 unsigned long lim_pages = (new_kesp | (THREAD_SIZE - 1)) >> PAGE_SHIFT;
1024 __u64 desc = *(__u64 *)&gdt[GDT_ENTRY_ESPFIX_SS];
1025 /* Set up base for espfix segment */
1026 desc &= 0x00f0ff0000000000ULL;
1027 desc |= ((((__u64)base) << 16) & 0x000000ffffff0000ULL) |
1028 ((((__u64)base) << 32) & 0xff00000000000000ULL) |
1029 ((((__u64)lim_pages) << 32) & 0x000f000000000000ULL) |
1030 (lim_pages & 0xffff);
1031 *(__u64 *)&gdt[GDT_ENTRY_ESPFIX_SS] = desc;
1036 * 'math_state_restore()' saves the current math information in the
1037 * old math state array, and gets the new ones from the current task
1039 * Careful.. There are problems with IBM-designed IRQ13 behaviour.
1040 * Don't touch unless you *really* know how it works.
1042 * Must be called with kernel preemption disabled (in this case,
1043 * local interrupts are disabled at the call-site in entry.S).
1045 asmlinkage void math_state_restore(void)
1047 struct thread_info *thread = current_thread_info();
1048 struct task_struct *tsk = thread->task;
1050 clts(); /* Allow maths ops (or we recurse) */
1051 if (!tsk_used_math(tsk))
1054 thread->status |= TS_USEDFPU; /* So we fnsave on switch_to() */
1058 #ifndef CONFIG_MATH_EMULATION
1060 asmlinkage void math_emulate(long arg)
1062 printk(KERN_EMERG "math-emulation not enabled and no coprocessor found.\n");
1063 printk(KERN_EMERG "killing %s.\n",current->comm);
1064 force_sig(SIGFPE,current);
1068 #endif /* CONFIG_MATH_EMULATION */
1070 #ifdef CONFIG_X86_F00F_BUG
1071 void __init trap_init_f00f_bug(void)
1073 __set_fixmap(FIX_F00F_IDT, __pa(&idt_table), PAGE_KERNEL_RO);
1076 * Update the IDT descriptor and reload the IDT so that
1077 * it uses the read-only mapped virtual address.
1079 idt_descr.address = fix_to_virt(FIX_F00F_IDT);
1080 load_idt(&idt_descr);
1085 * This needs to use 'idt_table' rather than 'idt', and
1086 * thus use the _nonmapped_ version of the IDT, as the
1087 * Pentium F0 0F bugfix can have resulted in the mapped
1088 * IDT being write-protected.
1090 void set_intr_gate(unsigned int n, void *addr)
1092 _set_gate(n, DESCTYPE_INT, addr, __KERNEL_CS);
1096 * This routine sets up an interrupt gate at directory privilege level 3.
1098 static inline void set_system_intr_gate(unsigned int n, void *addr)
1100 _set_gate(n, DESCTYPE_INT | DESCTYPE_DPL3, addr, __KERNEL_CS);
1103 static void __init set_trap_gate(unsigned int n, void *addr)
1105 _set_gate(n, DESCTYPE_TRAP, addr, __KERNEL_CS);
1108 static void __init set_system_gate(unsigned int n, void *addr)
1110 _set_gate(n, DESCTYPE_TRAP | DESCTYPE_DPL3, addr, __KERNEL_CS);
1113 static void __init set_task_gate(unsigned int n, unsigned int gdt_entry)
1115 _set_gate(n, DESCTYPE_TASK, (void *)0, (gdt_entry<<3));
1119 void __init trap_init(void)
1122 void __iomem *p = ioremap(0x0FFFD9, 4);
1123 if (readl(p) == 'E'+('I'<<8)+('S'<<16)+('A'<<24)) {
1129 #ifdef CONFIG_X86_LOCAL_APIC
1130 init_apic_mappings();
1133 set_trap_gate(0,÷_error);
1134 set_intr_gate(1,&debug);
1135 set_intr_gate(2,&nmi);
1136 set_system_intr_gate(3, &int3); /* int3/4 can be called from all */
1137 set_system_gate(4,&overflow);
1138 set_trap_gate(5,&bounds);
1139 set_trap_gate(6,&invalid_op);
1140 set_trap_gate(7,&device_not_available);
1141 set_task_gate(8,GDT_ENTRY_DOUBLEFAULT_TSS);
1142 set_trap_gate(9,&coprocessor_segment_overrun);
1143 set_trap_gate(10,&invalid_TSS);
1144 set_trap_gate(11,&segment_not_present);
1145 set_trap_gate(12,&stack_segment);
1146 set_trap_gate(13,&general_protection);
1147 set_intr_gate(14,&page_fault);
1148 set_trap_gate(15,&spurious_interrupt_bug);
1149 set_trap_gate(16,&coprocessor_error);
1150 set_trap_gate(17,&alignment_check);
1151 #ifdef CONFIG_X86_MCE
1152 set_trap_gate(18,&machine_check);
1154 set_trap_gate(19,&simd_coprocessor_error);
1158 * Verify that the FXSAVE/FXRSTOR data will be 16-byte aligned.
1159 * Generates a compile-time "error: zero width for bit-field" if
1160 * the alignment is wrong.
1162 struct fxsrAlignAssert {
1163 int _:!(offsetof(struct task_struct,
1164 thread.i387.fxsave) & 15);
1167 printk(KERN_INFO "Enabling fast FPU save and restore... ");
1168 set_in_cr4(X86_CR4_OSFXSR);
1172 printk(KERN_INFO "Enabling unmasked SIMD FPU exception "
1174 set_in_cr4(X86_CR4_OSXMMEXCPT);
1178 set_system_gate(SYSCALL_VECTOR,&system_call);
1181 * Should be a barrier for any external CPU state.
1188 static int __init kstack_setup(char *s)
1190 kstack_depth_to_print = simple_strtoul(s, NULL, 0);
1193 __setup("kstack=", kstack_setup);