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 <linux/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 static unsigned int code_bytes = 64;
99 static inline int valid_stack_ptr(struct thread_info *tinfo, void *p)
101 return p > (void *)tinfo &&
102 p < (void *)tinfo + THREAD_SIZE - 3;
105 static inline unsigned long print_context_stack(struct thread_info *tinfo,
106 unsigned long *stack, unsigned long ebp,
107 struct stacktrace_ops *ops, void *data)
111 #ifdef CONFIG_FRAME_POINTER
112 while (valid_stack_ptr(tinfo, (void *)ebp)) {
113 unsigned long new_ebp;
114 addr = *(unsigned long *)(ebp + 4);
115 ops->address(data, addr);
117 * break out of recursive entries (such as
118 * end_of_stack_stop_unwind_function). Also,
119 * we can never allow a frame pointer to
122 new_ebp = *(unsigned long *)ebp;
128 while (valid_stack_ptr(tinfo, stack)) {
130 if (__kernel_text_address(addr))
131 ops->address(data, addr);
137 #define MSG(msg) ops->warning(data, msg)
139 void dump_trace(struct task_struct *task, struct pt_regs *regs,
140 unsigned long *stack,
141 struct stacktrace_ops *ops, void *data)
143 unsigned long ebp = 0;
151 if (task && task != current)
152 stack = (unsigned long *)task->thread.esp;
155 #ifdef CONFIG_FRAME_POINTER
157 if (task == current) {
158 /* Grab ebp right from our regs */
159 asm ("movl %%ebp, %0" : "=r" (ebp) : );
161 /* ebp is the last reg pushed by switch_to */
162 ebp = *(unsigned long *) task->thread.esp;
168 struct thread_info *context;
169 context = (struct thread_info *)
170 ((unsigned long)stack & (~(THREAD_SIZE - 1)));
171 ebp = print_context_stack(context, stack, ebp, ops, data);
172 /* Should be after the line below, but somewhere
173 in early boot context comes out corrupted and we
174 can't reference it -AK */
175 if (ops->stack(data, "IRQ") < 0)
177 stack = (unsigned long*)context->previous_esp;
180 touch_nmi_watchdog();
183 EXPORT_SYMBOL(dump_trace);
186 print_trace_warning_symbol(void *data, char *msg, unsigned long symbol)
189 print_symbol(msg, symbol);
193 static void print_trace_warning(void *data, char *msg)
195 printk("%s%s\n", (char *)data, msg);
198 static int print_trace_stack(void *data, char *name)
204 * Print one address/symbol entries per line.
206 static void print_trace_address(void *data, unsigned long addr)
208 printk("%s [<%08lx>] ", (char *)data, addr);
209 print_symbol("%s\n", addr);
212 static struct stacktrace_ops print_trace_ops = {
213 .warning = print_trace_warning,
214 .warning_symbol = print_trace_warning_symbol,
215 .stack = print_trace_stack,
216 .address = print_trace_address,
220 show_trace_log_lvl(struct task_struct *task, struct pt_regs *regs,
221 unsigned long * stack, char *log_lvl)
223 dump_trace(task, regs, stack, &print_trace_ops, log_lvl);
224 printk("%s =======================\n", log_lvl);
227 void show_trace(struct task_struct *task, struct pt_regs *regs,
228 unsigned long * stack)
230 show_trace_log_lvl(task, regs, stack, "");
233 static void show_stack_log_lvl(struct task_struct *task, struct pt_regs *regs,
234 unsigned long *esp, char *log_lvl)
236 unsigned long *stack;
241 esp = (unsigned long*)task->thread.esp;
243 esp = (unsigned long *)&esp;
247 for(i = 0; i < kstack_depth_to_print; i++) {
248 if (kstack_end(stack))
250 if (i && ((i % 8) == 0))
251 printk("\n%s ", log_lvl);
252 printk("%08lx ", *stack++);
254 printk("\n%sCall Trace:\n", log_lvl);
255 show_trace_log_lvl(task, regs, esp, log_lvl);
258 void show_stack(struct task_struct *task, unsigned long *esp)
261 show_stack_log_lvl(task, NULL, esp, "");
265 * The architecture-independent dump_stack generator
267 void dump_stack(void)
271 show_trace(current, NULL, &stack);
274 EXPORT_SYMBOL(dump_stack);
276 void show_registers(struct pt_regs *regs)
281 unsigned short ss, gs;
283 esp = (unsigned long) (®s->esp);
286 if (user_mode_vm(regs)) {
289 ss = regs->xss & 0xffff;
292 printk(KERN_EMERG "CPU: %d\n"
293 KERN_EMERG "EIP: %04x:[<%08lx>] %s VLI\n"
294 KERN_EMERG "EFLAGS: %08lx (%s %.*s)\n",
295 smp_processor_id(), 0xffff & regs->xcs, regs->eip,
296 print_tainted(), regs->eflags, init_utsname()->release,
297 (int)strcspn(init_utsname()->version, " "),
298 init_utsname()->version);
299 print_symbol(KERN_EMERG "EIP is at %s\n", regs->eip);
300 printk(KERN_EMERG "eax: %08lx ebx: %08lx ecx: %08lx edx: %08lx\n",
301 regs->eax, regs->ebx, regs->ecx, regs->edx);
302 printk(KERN_EMERG "esi: %08lx edi: %08lx ebp: %08lx esp: %08lx\n",
303 regs->esi, regs->edi, regs->ebp, esp);
304 printk(KERN_EMERG "ds: %04x es: %04x fs: %04x gs: %04x ss: %04x\n",
305 regs->xds & 0xffff, regs->xes & 0xffff, regs->xfs & 0xffff, gs, ss);
306 printk(KERN_EMERG "Process %.*s (pid: %d, ti=%p task=%p task.ti=%p)",
307 TASK_COMM_LEN, current->comm, current->pid,
308 current_thread_info(), current, task_thread_info(current));
310 * When in-kernel, we also print out the stack and code at the
311 * time of the fault..
315 unsigned int code_prologue = code_bytes * 43 / 64;
316 unsigned int code_len = code_bytes;
319 printk("\n" KERN_EMERG "Stack: ");
320 show_stack_log_lvl(NULL, regs, (unsigned long *)esp, KERN_EMERG);
322 printk(KERN_EMERG "Code: ");
324 eip = (u8 *)regs->eip - code_prologue;
325 if (eip < (u8 *)PAGE_OFFSET ||
326 probe_kernel_address(eip, c)) {
327 /* try starting at EIP */
328 eip = (u8 *)regs->eip;
329 code_len = code_len - code_prologue + 1;
331 for (i = 0; i < code_len; i++, eip++) {
332 if (eip < (u8 *)PAGE_OFFSET ||
333 probe_kernel_address(eip, c)) {
334 printk(" Bad EIP value.");
337 if (eip == (u8 *)regs->eip)
338 printk("<%02x> ", c);
346 int is_valid_bugaddr(unsigned long eip)
350 if (eip < PAGE_OFFSET)
352 if (probe_kernel_address((unsigned short *)eip, ud2))
355 return ud2 == 0x0b0f;
359 * This is gone through when something in the kernel has done something bad and
360 * is about to be terminated.
362 void die(const char * str, struct pt_regs * regs, long err)
367 int lock_owner_depth;
369 .lock = __SPIN_LOCK_UNLOCKED(die.lock),
371 .lock_owner_depth = 0
373 static int die_counter;
378 if (die.lock_owner != raw_smp_processor_id()) {
380 spin_lock_irqsave(&die.lock, flags);
381 die.lock_owner = smp_processor_id();
382 die.lock_owner_depth = 0;
386 local_save_flags(flags);
388 if (++die.lock_owner_depth < 3) {
393 report_bug(regs->eip);
395 printk(KERN_EMERG "%s: %04lx [#%d]\n", str, err & 0xffff, ++die_counter);
396 #ifdef CONFIG_PREEMPT
397 printk(KERN_EMERG "PREEMPT ");
406 #ifdef CONFIG_DEBUG_PAGEALLOC
409 printk("DEBUG_PAGEALLOC");
414 if (notify_die(DIE_OOPS, str, regs, err,
415 current->thread.trap_no, SIGSEGV) !=
417 show_registers(regs);
418 /* Executive summary in case the oops scrolled away */
419 esp = (unsigned long) (®s->esp);
421 if (user_mode(regs)) {
423 ss = regs->xss & 0xffff;
425 printk(KERN_EMERG "EIP: [<%08lx>] ", regs->eip);
426 print_symbol("%s", regs->eip);
427 printk(" SS:ESP %04x:%08lx\n", ss, esp);
432 printk(KERN_EMERG "Recursive die() failure, output suppressed\n");
436 spin_unlock_irqrestore(&die.lock, flags);
441 if (kexec_should_crash(current))
445 panic("Fatal exception in interrupt");
448 panic("Fatal exception");
454 static inline void die_if_kernel(const char * str, struct pt_regs * regs, long err)
456 if (!user_mode_vm(regs))
460 static void __kprobes do_trap(int trapnr, int signr, char *str, int vm86,
461 struct pt_regs * regs, long error_code,
464 struct task_struct *tsk = current;
466 if (regs->eflags & VM_MASK) {
472 if (!user_mode(regs))
477 * We want error_code and trap_no set for userspace faults and
478 * kernelspace faults which result in die(), but not
479 * kernelspace faults which are fixed up. die() gives the
480 * process no chance to handle the signal and notice the
481 * kernel fault information, so that won't result in polluting
482 * the information about previously queued, but not yet
483 * delivered, faults. See also do_general_protection below.
485 tsk->thread.error_code = error_code;
486 tsk->thread.trap_no = trapnr;
489 force_sig_info(signr, info, tsk);
491 force_sig(signr, tsk);
496 if (!fixup_exception(regs)) {
497 tsk->thread.error_code = error_code;
498 tsk->thread.trap_no = trapnr;
499 die(str, regs, error_code);
505 int ret = handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code, trapnr);
506 if (ret) goto trap_signal;
511 #define DO_ERROR(trapnr, signr, str, name) \
512 fastcall void do_##name(struct pt_regs * regs, long error_code) \
514 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
517 do_trap(trapnr, signr, str, 0, regs, error_code, NULL); \
520 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
521 fastcall void do_##name(struct pt_regs * regs, long error_code) \
524 info.si_signo = signr; \
526 info.si_code = sicode; \
527 info.si_addr = (void __user *)siaddr; \
528 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
531 do_trap(trapnr, signr, str, 0, regs, error_code, &info); \
534 #define DO_VM86_ERROR(trapnr, signr, str, name) \
535 fastcall void do_##name(struct pt_regs * regs, long error_code) \
537 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
540 do_trap(trapnr, signr, str, 1, regs, error_code, NULL); \
543 #define DO_VM86_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
544 fastcall void do_##name(struct pt_regs * regs, long error_code) \
547 info.si_signo = signr; \
549 info.si_code = sicode; \
550 info.si_addr = (void __user *)siaddr; \
551 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
554 do_trap(trapnr, signr, str, 1, regs, error_code, &info); \
557 DO_VM86_ERROR_INFO( 0, SIGFPE, "divide error", divide_error, FPE_INTDIV, regs->eip)
558 #ifndef CONFIG_KPROBES
559 DO_VM86_ERROR( 3, SIGTRAP, "int3", int3)
561 DO_VM86_ERROR( 4, SIGSEGV, "overflow", overflow)
562 DO_VM86_ERROR( 5, SIGSEGV, "bounds", bounds)
563 DO_ERROR_INFO( 6, SIGILL, "invalid opcode", invalid_op, ILL_ILLOPN, regs->eip)
564 DO_ERROR( 9, SIGFPE, "coprocessor segment overrun", coprocessor_segment_overrun)
565 DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
566 DO_ERROR(11, SIGBUS, "segment not present", segment_not_present)
567 DO_ERROR(12, SIGBUS, "stack segment", stack_segment)
568 DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0)
569 DO_ERROR_INFO(32, SIGSEGV, "iret exception", iret_error, ILL_BADSTK, 0)
571 fastcall void __kprobes do_general_protection(struct pt_regs * regs,
575 struct tss_struct *tss = &per_cpu(init_tss, cpu);
576 struct thread_struct *thread = ¤t->thread;
579 * Perform the lazy TSS's I/O bitmap copy. If the TSS has an
580 * invalid offset set (the LAZY one) and the faulting thread has
581 * a valid I/O bitmap pointer, we copy the I/O bitmap in the TSS
582 * and we set the offset field correctly. Then we let the CPU to
583 * restart the faulting instruction.
585 if (tss->x86_tss.io_bitmap_base == INVALID_IO_BITMAP_OFFSET_LAZY &&
586 thread->io_bitmap_ptr) {
587 memcpy(tss->io_bitmap, thread->io_bitmap_ptr,
588 thread->io_bitmap_max);
590 * If the previously set map was extending to higher ports
591 * than the current one, pad extra space with 0xff (no access).
593 if (thread->io_bitmap_max < tss->io_bitmap_max)
594 memset((char *) tss->io_bitmap +
595 thread->io_bitmap_max, 0xff,
596 tss->io_bitmap_max - thread->io_bitmap_max);
597 tss->io_bitmap_max = thread->io_bitmap_max;
598 tss->x86_tss.io_bitmap_base = IO_BITMAP_OFFSET;
599 tss->io_bitmap_owner = thread;
605 if (regs->eflags & VM_MASK)
608 if (!user_mode(regs))
611 current->thread.error_code = error_code;
612 current->thread.trap_no = 13;
613 force_sig(SIGSEGV, current);
618 handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
622 if (!fixup_exception(regs)) {
623 current->thread.error_code = error_code;
624 current->thread.trap_no = 13;
625 if (notify_die(DIE_GPF, "general protection fault", regs,
626 error_code, 13, SIGSEGV) == NOTIFY_STOP)
628 die("general protection fault", regs, error_code);
632 static __kprobes void
633 mem_parity_error(unsigned char reason, struct pt_regs * regs)
635 printk(KERN_EMERG "Uhhuh. NMI received for unknown reason %02x on "
636 "CPU %d.\n", reason, smp_processor_id());
637 printk(KERN_EMERG "You have some hardware problem, likely on the PCI bus.\n");
638 if (panic_on_unrecovered_nmi)
639 panic("NMI: Not continuing");
641 printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
643 /* Clear and disable the memory parity error line. */
644 clear_mem_error(reason);
647 static __kprobes void
648 io_check_error(unsigned char reason, struct pt_regs * regs)
652 printk(KERN_EMERG "NMI: IOCK error (debug interrupt?)\n");
653 show_registers(regs);
655 /* Re-enable the IOCK line, wait for a few seconds */
656 reason = (reason & 0xf) | 8;
659 while (--i) udelay(1000);
664 static __kprobes void
665 unknown_nmi_error(unsigned char reason, struct pt_regs * regs)
668 /* Might actually be able to figure out what the guilty party
675 printk(KERN_EMERG "Uhhuh. NMI received for unknown reason %02x on "
676 "CPU %d.\n", reason, smp_processor_id());
677 printk(KERN_EMERG "Do you have a strange power saving mode enabled?\n");
678 if (panic_on_unrecovered_nmi)
679 panic("NMI: Not continuing");
681 printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
684 static DEFINE_SPINLOCK(nmi_print_lock);
686 void __kprobes die_nmi(struct pt_regs *regs, const char *msg)
688 if (notify_die(DIE_NMIWATCHDOG, msg, regs, 0, 2, SIGINT) ==
692 spin_lock(&nmi_print_lock);
694 * We are in trouble anyway, lets at least try
695 * to get a message out.
698 printk(KERN_EMERG "%s", msg);
699 printk(" on CPU%d, eip %08lx, registers:\n",
700 smp_processor_id(), regs->eip);
701 show_registers(regs);
703 spin_unlock(&nmi_print_lock);
706 /* If we are in kernel we are probably nested up pretty bad
707 * and might aswell get out now while we still can.
709 if (!user_mode_vm(regs)) {
710 current->thread.trap_no = 2;
717 static __kprobes void default_do_nmi(struct pt_regs * regs)
719 unsigned char reason = 0;
721 /* Only the BSP gets external NMIs from the system. */
722 if (!smp_processor_id())
723 reason = get_nmi_reason();
725 if (!(reason & 0xc0)) {
726 if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 2, SIGINT)
729 #ifdef CONFIG_X86_LOCAL_APIC
731 * Ok, so this is none of the documented NMI sources,
732 * so it must be the NMI watchdog.
734 if (nmi_watchdog_tick(regs, reason))
737 if (notify_die(DIE_NMI_POST, "nmi_post", regs, reason, 2, 0)
740 #ifdef CONFIG_X86_LOCAL_APIC
741 if (!do_nmi_callback(regs, smp_processor_id()))
743 unknown_nmi_error(reason, regs);
747 if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP)
750 mem_parity_error(reason, regs);
752 io_check_error(reason, regs);
754 * Reassert NMI in case it became active meanwhile
755 * as it's edge-triggered.
760 fastcall __kprobes void do_nmi(struct pt_regs * regs, long error_code)
766 cpu = smp_processor_id();
770 default_do_nmi(regs);
775 #ifdef CONFIG_KPROBES
776 fastcall void __kprobes do_int3(struct pt_regs *regs, long error_code)
778 if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP)
781 /* This is an interrupt gate, because kprobes wants interrupts
782 disabled. Normal trap handlers don't. */
783 restore_interrupts(regs);
784 do_trap(3, SIGTRAP, "int3", 1, regs, error_code, NULL);
789 * Our handling of the processor debug registers is non-trivial.
790 * We do not clear them on entry and exit from the kernel. Therefore
791 * it is possible to get a watchpoint trap here from inside the kernel.
792 * However, the code in ./ptrace.c has ensured that the user can
793 * only set watchpoints on userspace addresses. Therefore the in-kernel
794 * watchpoint trap can only occur in code which is reading/writing
795 * from user space. Such code must not hold kernel locks (since it
796 * can equally take a page fault), therefore it is safe to call
797 * force_sig_info even though that claims and releases locks.
799 * Code in ./signal.c ensures that the debug control register
800 * is restored before we deliver any signal, and therefore that
801 * user code runs with the correct debug control register even though
804 * Being careful here means that we don't have to be as careful in a
805 * lot of more complicated places (task switching can be a bit lazy
806 * about restoring all the debug state, and ptrace doesn't have to
807 * find every occurrence of the TF bit that could be saved away even
810 fastcall void __kprobes do_debug(struct pt_regs * regs, long error_code)
812 unsigned int condition;
813 struct task_struct *tsk = current;
815 get_debugreg(condition, 6);
817 if (notify_die(DIE_DEBUG, "debug", regs, condition, error_code,
818 SIGTRAP) == NOTIFY_STOP)
820 /* It's safe to allow irq's after DR6 has been saved */
821 if (regs->eflags & X86_EFLAGS_IF)
824 /* Mask out spurious debug traps due to lazy DR7 setting */
825 if (condition & (DR_TRAP0|DR_TRAP1|DR_TRAP2|DR_TRAP3)) {
826 if (!tsk->thread.debugreg[7])
830 if (regs->eflags & VM_MASK)
833 /* Save debug status register where ptrace can see it */
834 tsk->thread.debugreg[6] = condition;
837 * Single-stepping through TF: make sure we ignore any events in
838 * kernel space (but re-enable TF when returning to user mode).
840 if (condition & DR_STEP) {
842 * We already checked v86 mode above, so we can
843 * check for kernel mode by just checking the CPL
846 if (!user_mode(regs))
847 goto clear_TF_reenable;
850 /* Ok, finally something we can handle */
851 send_sigtrap(tsk, regs, error_code);
853 /* Disable additional traps. They'll be re-enabled when
854 * the signal is delivered.
861 handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code, 1);
865 set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
866 regs->eflags &= ~TF_MASK;
871 * Note that we play around with the 'TS' bit in an attempt to get
872 * the correct behaviour even in the presence of the asynchronous
875 void math_error(void __user *eip)
877 struct task_struct * task;
879 unsigned short cwd, swd;
882 * Save the info for the exception handler and clear the error.
886 task->thread.trap_no = 16;
887 task->thread.error_code = 0;
888 info.si_signo = SIGFPE;
890 info.si_code = __SI_FAULT;
893 * (~cwd & swd) will mask out exceptions that are not set to unmasked
894 * status. 0x3f is the exception bits in these regs, 0x200 is the
895 * C1 reg you need in case of a stack fault, 0x040 is the stack
896 * fault bit. We should only be taking one exception at a time,
897 * so if this combination doesn't produce any single exception,
898 * then we have a bad program that isn't syncronizing its FPU usage
899 * and it will suffer the consequences since we won't be able to
900 * fully reproduce the context of the exception
902 cwd = get_fpu_cwd(task);
903 swd = get_fpu_swd(task);
904 switch (swd & ~cwd & 0x3f) {
905 case 0x000: /* No unmasked exception */
907 default: /* Multiple exceptions */
909 case 0x001: /* Invalid Op */
911 * swd & 0x240 == 0x040: Stack Underflow
912 * swd & 0x240 == 0x240: Stack Overflow
913 * User must clear the SF bit (0x40) if set
915 info.si_code = FPE_FLTINV;
917 case 0x002: /* Denormalize */
918 case 0x010: /* Underflow */
919 info.si_code = FPE_FLTUND;
921 case 0x004: /* Zero Divide */
922 info.si_code = FPE_FLTDIV;
924 case 0x008: /* Overflow */
925 info.si_code = FPE_FLTOVF;
927 case 0x020: /* Precision */
928 info.si_code = FPE_FLTRES;
931 force_sig_info(SIGFPE, &info, task);
934 fastcall void do_coprocessor_error(struct pt_regs * regs, long error_code)
937 math_error((void __user *)regs->eip);
940 static void simd_math_error(void __user *eip)
942 struct task_struct * task;
944 unsigned short mxcsr;
947 * Save the info for the exception handler and clear the error.
951 task->thread.trap_no = 19;
952 task->thread.error_code = 0;
953 info.si_signo = SIGFPE;
955 info.si_code = __SI_FAULT;
958 * The SIMD FPU exceptions are handled a little differently, as there
959 * is only a single status/control register. Thus, to determine which
960 * unmasked exception was caught we must mask the exception mask bits
961 * at 0x1f80, and then use these to mask the exception bits at 0x3f.
963 mxcsr = get_fpu_mxcsr(task);
964 switch (~((mxcsr & 0x1f80) >> 7) & (mxcsr & 0x3f)) {
968 case 0x001: /* Invalid Op */
969 info.si_code = FPE_FLTINV;
971 case 0x002: /* Denormalize */
972 case 0x010: /* Underflow */
973 info.si_code = FPE_FLTUND;
975 case 0x004: /* Zero Divide */
976 info.si_code = FPE_FLTDIV;
978 case 0x008: /* Overflow */
979 info.si_code = FPE_FLTOVF;
981 case 0x020: /* Precision */
982 info.si_code = FPE_FLTRES;
985 force_sig_info(SIGFPE, &info, task);
988 fastcall void do_simd_coprocessor_error(struct pt_regs * regs,
992 /* Handle SIMD FPU exceptions on PIII+ processors. */
994 simd_math_error((void __user *)regs->eip);
997 * Handle strange cache flush from user space exception
998 * in all other cases. This is undocumented behaviour.
1000 if (regs->eflags & VM_MASK) {
1001 handle_vm86_fault((struct kernel_vm86_regs *)regs,
1005 current->thread.trap_no = 19;
1006 current->thread.error_code = error_code;
1007 die_if_kernel("cache flush denied", regs, error_code);
1008 force_sig(SIGSEGV, current);
1012 fastcall void do_spurious_interrupt_bug(struct pt_regs * regs,
1016 /* No need to warn about this any longer. */
1017 printk("Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
1021 fastcall unsigned long patch_espfix_desc(unsigned long uesp,
1024 struct desc_struct *gdt = __get_cpu_var(gdt_page).gdt;
1025 unsigned long base = (kesp - uesp) & -THREAD_SIZE;
1026 unsigned long new_kesp = kesp - base;
1027 unsigned long lim_pages = (new_kesp | (THREAD_SIZE - 1)) >> PAGE_SHIFT;
1028 __u64 desc = *(__u64 *)&gdt[GDT_ENTRY_ESPFIX_SS];
1029 /* Set up base for espfix segment */
1030 desc &= 0x00f0ff0000000000ULL;
1031 desc |= ((((__u64)base) << 16) & 0x000000ffffff0000ULL) |
1032 ((((__u64)base) << 32) & 0xff00000000000000ULL) |
1033 ((((__u64)lim_pages) << 32) & 0x000f000000000000ULL) |
1034 (lim_pages & 0xffff);
1035 *(__u64 *)&gdt[GDT_ENTRY_ESPFIX_SS] = desc;
1040 * 'math_state_restore()' saves the current math information in the
1041 * old math state array, and gets the new ones from the current task
1043 * Careful.. There are problems with IBM-designed IRQ13 behaviour.
1044 * Don't touch unless you *really* know how it works.
1046 * Must be called with kernel preemption disabled (in this case,
1047 * local interrupts are disabled at the call-site in entry.S).
1049 asmlinkage void math_state_restore(void)
1051 struct thread_info *thread = current_thread_info();
1052 struct task_struct *tsk = thread->task;
1054 clts(); /* Allow maths ops (or we recurse) */
1055 if (!tsk_used_math(tsk))
1058 thread->status |= TS_USEDFPU; /* So we fnsave on switch_to() */
1062 #ifndef CONFIG_MATH_EMULATION
1064 asmlinkage void math_emulate(long arg)
1066 printk(KERN_EMERG "math-emulation not enabled and no coprocessor found.\n");
1067 printk(KERN_EMERG "killing %s.\n",current->comm);
1068 force_sig(SIGFPE,current);
1072 #endif /* CONFIG_MATH_EMULATION */
1074 #ifdef CONFIG_X86_F00F_BUG
1075 void __init trap_init_f00f_bug(void)
1077 __set_fixmap(FIX_F00F_IDT, __pa(&idt_table), PAGE_KERNEL_RO);
1080 * Update the IDT descriptor and reload the IDT so that
1081 * it uses the read-only mapped virtual address.
1083 idt_descr.address = fix_to_virt(FIX_F00F_IDT);
1084 load_idt(&idt_descr);
1089 * This needs to use 'idt_table' rather than 'idt', and
1090 * thus use the _nonmapped_ version of the IDT, as the
1091 * Pentium F0 0F bugfix can have resulted in the mapped
1092 * IDT being write-protected.
1094 void set_intr_gate(unsigned int n, void *addr)
1096 _set_gate(n, DESCTYPE_INT, addr, __KERNEL_CS);
1100 * This routine sets up an interrupt gate at directory privilege level 3.
1102 static inline void set_system_intr_gate(unsigned int n, void *addr)
1104 _set_gate(n, DESCTYPE_INT | DESCTYPE_DPL3, addr, __KERNEL_CS);
1107 static void __init set_trap_gate(unsigned int n, void *addr)
1109 _set_gate(n, DESCTYPE_TRAP, addr, __KERNEL_CS);
1112 static void __init set_system_gate(unsigned int n, void *addr)
1114 _set_gate(n, DESCTYPE_TRAP | DESCTYPE_DPL3, addr, __KERNEL_CS);
1117 static void __init set_task_gate(unsigned int n, unsigned int gdt_entry)
1119 _set_gate(n, DESCTYPE_TASK, (void *)0, (gdt_entry<<3));
1123 void __init trap_init(void)
1126 void __iomem *p = ioremap(0x0FFFD9, 4);
1127 if (readl(p) == 'E'+('I'<<8)+('S'<<16)+('A'<<24)) {
1133 #ifdef CONFIG_X86_LOCAL_APIC
1134 init_apic_mappings();
1137 set_trap_gate(0,÷_error);
1138 set_intr_gate(1,&debug);
1139 set_intr_gate(2,&nmi);
1140 set_system_intr_gate(3, &int3); /* int3/4 can be called from all */
1141 set_system_gate(4,&overflow);
1142 set_trap_gate(5,&bounds);
1143 set_trap_gate(6,&invalid_op);
1144 set_trap_gate(7,&device_not_available);
1145 set_task_gate(8,GDT_ENTRY_DOUBLEFAULT_TSS);
1146 set_trap_gate(9,&coprocessor_segment_overrun);
1147 set_trap_gate(10,&invalid_TSS);
1148 set_trap_gate(11,&segment_not_present);
1149 set_trap_gate(12,&stack_segment);
1150 set_trap_gate(13,&general_protection);
1151 set_intr_gate(14,&page_fault);
1152 set_trap_gate(15,&spurious_interrupt_bug);
1153 set_trap_gate(16,&coprocessor_error);
1154 set_trap_gate(17,&alignment_check);
1155 #ifdef CONFIG_X86_MCE
1156 set_trap_gate(18,&machine_check);
1158 set_trap_gate(19,&simd_coprocessor_error);
1162 * Verify that the FXSAVE/FXRSTOR data will be 16-byte aligned.
1163 * Generates a compile-time "error: zero width for bit-field" if
1164 * the alignment is wrong.
1166 struct fxsrAlignAssert {
1167 int _:!(offsetof(struct task_struct,
1168 thread.i387.fxsave) & 15);
1171 printk(KERN_INFO "Enabling fast FPU save and restore... ");
1172 set_in_cr4(X86_CR4_OSFXSR);
1176 printk(KERN_INFO "Enabling unmasked SIMD FPU exception "
1178 set_in_cr4(X86_CR4_OSXMMEXCPT);
1182 set_system_gate(SYSCALL_VECTOR,&system_call);
1185 * Should be a barrier for any external CPU state.
1192 static int __init kstack_setup(char *s)
1194 kstack_depth_to_print = simple_strtoul(s, NULL, 0);
1197 __setup("kstack=", kstack_setup);
1199 static int __init code_bytes_setup(char *s)
1201 code_bytes = simple_strtoul(s, NULL, 0);
1202 if (code_bytes > 8192)
1207 __setup("code_bytes=", code_bytes_setup);