2 * Copyright (C) 1991, 1992 Linus Torvalds
4 * Pentium III FXSR, SSE support
5 * Gareth Hughes <gareth@valinux.com>, May 2000
9 * 'Traps.c' handles hardware traps and faults after we have saved some
12 #include <linux/interrupt.h>
13 #include <linux/kallsyms.h>
14 #include <linux/spinlock.h>
15 #include <linux/highmem.h>
16 #include <linux/kprobes.h>
17 #include <linux/uaccess.h>
18 #include <linux/utsname.h>
19 #include <linux/kdebug.h>
20 #include <linux/kernel.h>
21 #include <linux/module.h>
22 #include <linux/ptrace.h>
23 #include <linux/string.h>
24 #include <linux/unwind.h>
25 #include <linux/delay.h>
26 #include <linux/errno.h>
27 #include <linux/kexec.h>
28 #include <linux/sched.h>
29 #include <linux/timer.h>
30 #include <linux/init.h>
31 #include <linux/bug.h>
32 #include <linux/nmi.h>
36 #include <linux/ioport.h>
37 #include <linux/eisa.h>
41 #include <linux/mca.h>
44 #if defined(CONFIG_EDAC)
45 #include <linux/edac.h>
48 #include <asm/arch_hooks.h>
49 #include <asm/stacktrace.h>
50 #include <asm/processor.h>
51 #include <asm/debugreg.h>
52 #include <asm/atomic.h>
53 #include <asm/system.h>
54 #include <asm/unwind.h>
61 #include "mach_traps.h"
63 int panic_on_unrecovered_nmi;
65 DECLARE_BITMAP(used_vectors, NR_VECTORS);
66 EXPORT_SYMBOL_GPL(used_vectors);
68 asmlinkage int system_call(void);
70 /* Do we ignore FPU interrupts ? */
74 * The IDT has to be page-aligned to simplify the Pentium
75 * F0 0F bug workaround.. We have a special link segment
78 gate_desc idt_table[256]
79 __attribute__((__section__(".data.idt"))) = { { { { 0, 0 } } }, };
81 asmlinkage void divide_error(void);
82 asmlinkage void debug(void);
83 asmlinkage void nmi(void);
84 asmlinkage void int3(void);
85 asmlinkage void overflow(void);
86 asmlinkage void bounds(void);
87 asmlinkage void invalid_op(void);
88 asmlinkage void device_not_available(void);
89 asmlinkage void coprocessor_segment_overrun(void);
90 asmlinkage void invalid_TSS(void);
91 asmlinkage void segment_not_present(void);
92 asmlinkage void stack_segment(void);
93 asmlinkage void general_protection(void);
94 asmlinkage void page_fault(void);
95 asmlinkage void coprocessor_error(void);
96 asmlinkage void simd_coprocessor_error(void);
97 asmlinkage void alignment_check(void);
98 asmlinkage void spurious_interrupt_bug(void);
99 asmlinkage void machine_check(void);
101 int kstack_depth_to_print = 24;
102 static unsigned int code_bytes = 64;
104 void printk_address(unsigned long address, int reliable)
106 #ifdef CONFIG_KALLSYMS
107 char namebuf[KSYM_NAME_LEN];
108 unsigned long offset = 0;
109 unsigned long symsize;
115 symname = kallsyms_lookup(address, &symsize, &offset,
118 printk(" [<%08lx>]\n", address);
122 strcpy(reliab, "? ");
125 modname = delim = "";
126 printk(" [<%08lx>] %s%s%s%s%s+0x%lx/0x%lx\n",
127 address, reliab, delim, modname, delim, symname, offset, symsize);
129 printk(" [<%08lx>]\n", address);
133 static inline int valid_stack_ptr(struct thread_info *tinfo, void *p, unsigned size)
135 return p > (void *)tinfo &&
136 p <= (void *)tinfo + THREAD_SIZE - size;
139 /* The form of the top of the frame on the stack */
141 struct stack_frame *next_frame;
142 unsigned long return_address;
145 static inline unsigned long
146 print_context_stack(struct thread_info *tinfo,
147 unsigned long *stack, unsigned long bp,
148 const struct stacktrace_ops *ops, void *data)
150 struct stack_frame *frame = (struct stack_frame *)bp;
152 while (valid_stack_ptr(tinfo, stack, sizeof(*stack))) {
156 if (__kernel_text_address(addr)) {
157 if ((unsigned long) stack == bp + 4) {
158 ops->address(data, addr, 1);
159 frame = frame->next_frame;
160 bp = (unsigned long) frame;
162 ops->address(data, addr, bp == 0);
170 #define MSG(msg) ops->warning(data, msg)
172 void dump_trace(struct task_struct *task, struct pt_regs *regs,
173 unsigned long *stack, unsigned long bp,
174 const struct stacktrace_ops *ops, void *data)
184 stack = (unsigned long *)task->thread.sp;
187 #ifdef CONFIG_FRAME_POINTER
189 if (task == current) {
190 /* Grab bp right from our regs */
191 asm("movl %%ebp, %0" : "=r" (bp) :);
193 /* bp is the last reg pushed by switch_to */
194 bp = *(unsigned long *) task->thread.sp;
200 struct thread_info *context;
202 context = (struct thread_info *)
203 ((unsigned long)stack & (~(THREAD_SIZE - 1)));
204 bp = print_context_stack(context, stack, bp, ops, data);
206 * Should be after the line below, but somewhere
207 * in early boot context comes out corrupted and we
208 * can't reference it:
210 if (ops->stack(data, "IRQ") < 0)
212 stack = (unsigned long *)context->previous_esp;
215 touch_nmi_watchdog();
218 EXPORT_SYMBOL(dump_trace);
221 print_trace_warning_symbol(void *data, char *msg, unsigned long symbol)
224 print_symbol(msg, symbol);
228 static void print_trace_warning(void *data, char *msg)
230 printk("%s%s\n", (char *)data, msg);
233 static int print_trace_stack(void *data, char *name)
239 * Print one address/symbol entries per line.
241 static void print_trace_address(void *data, unsigned long addr, int reliable)
243 printk("%s [<%08lx>] ", (char *)data, addr);
246 print_symbol("%s\n", addr);
247 touch_nmi_watchdog();
250 static const struct stacktrace_ops print_trace_ops = {
251 .warning = print_trace_warning,
252 .warning_symbol = print_trace_warning_symbol,
253 .stack = print_trace_stack,
254 .address = print_trace_address,
258 show_trace_log_lvl(struct task_struct *task, struct pt_regs *regs,
259 unsigned long *stack, unsigned long bp, char *log_lvl)
261 dump_trace(task, regs, stack, bp, &print_trace_ops, log_lvl);
262 printk("%s =======================\n", log_lvl);
265 void show_trace(struct task_struct *task, struct pt_regs *regs,
266 unsigned long *stack, unsigned long bp)
268 show_trace_log_lvl(task, regs, stack, bp, "");
272 show_stack_log_lvl(struct task_struct *task, struct pt_regs *regs,
273 unsigned long *sp, unsigned long bp, char *log_lvl)
275 unsigned long *stack;
280 sp = (unsigned long *)task->thread.sp;
282 sp = (unsigned long *)&sp;
286 for (i = 0; i < kstack_depth_to_print; i++) {
287 if (kstack_end(stack))
289 if (i && ((i % 8) == 0))
290 printk("\n%s ", log_lvl);
291 printk("%08lx ", *stack++);
293 printk("\n%sCall Trace:\n", log_lvl);
295 show_trace_log_lvl(task, regs, sp, bp, log_lvl);
298 void show_stack(struct task_struct *task, unsigned long *sp)
301 show_stack_log_lvl(task, NULL, sp, 0, "");
305 * The architecture-independent dump_stack generator
307 void dump_stack(void)
309 unsigned long bp = 0;
312 #ifdef CONFIG_FRAME_POINTER
314 asm("movl %%ebp, %0" : "=r" (bp):);
317 printk("Pid: %d, comm: %.20s %s %s %.*s\n",
318 current->pid, current->comm, print_tainted(),
319 init_utsname()->release,
320 (int)strcspn(init_utsname()->version, " "),
321 init_utsname()->version);
323 show_trace(current, NULL, &stack, bp);
326 EXPORT_SYMBOL(dump_stack);
328 void show_registers(struct pt_regs *regs)
333 __show_registers(regs, 0);
335 printk(KERN_EMERG "Process %.*s (pid: %d, ti=%p task=%p task.ti=%p)",
336 TASK_COMM_LEN, current->comm, task_pid_nr(current),
337 current_thread_info(), current, task_thread_info(current));
339 * When in-kernel, we also print out the stack and code at the
340 * time of the fault..
342 if (!user_mode_vm(regs)) {
343 unsigned int code_prologue = code_bytes * 43 / 64;
344 unsigned int code_len = code_bytes;
348 printk("\n" KERN_EMERG "Stack: ");
349 show_stack_log_lvl(NULL, regs, ®s->sp, 0, KERN_EMERG);
351 printk(KERN_EMERG "Code: ");
353 ip = (u8 *)regs->ip - code_prologue;
354 if (ip < (u8 *)PAGE_OFFSET ||
355 probe_kernel_address(ip, c)) {
356 /* try starting at EIP */
358 code_len = code_len - code_prologue + 1;
360 for (i = 0; i < code_len; i++, ip++) {
361 if (ip < (u8 *)PAGE_OFFSET ||
362 probe_kernel_address(ip, c)) {
363 printk(" Bad EIP value.");
366 if (ip == (u8 *)regs->ip)
367 printk("<%02x> ", c);
375 int is_valid_bugaddr(unsigned long ip)
379 if (ip < PAGE_OFFSET)
381 if (probe_kernel_address((unsigned short *)ip, ud2))
384 return ud2 == 0x0b0f;
387 static int die_counter;
389 int __kprobes __die(const char *str, struct pt_regs *regs, long err)
394 printk(KERN_EMERG "%s: %04lx [#%d] ", str, err & 0xffff, ++die_counter);
395 #ifdef CONFIG_PREEMPT
401 #ifdef CONFIG_DEBUG_PAGEALLOC
402 printk("DEBUG_PAGEALLOC");
406 if (notify_die(DIE_OOPS, str, regs, err,
407 current->thread.trap_no, SIGSEGV) != NOTIFY_STOP) {
409 show_registers(regs);
410 /* Executive summary in case the oops scrolled away */
411 sp = (unsigned long) (®s->sp);
413 if (user_mode(regs)) {
415 ss = regs->ss & 0xffff;
417 printk(KERN_EMERG "EIP: [<%08lx>] ", regs->ip);
418 print_symbol("%s", regs->ip);
419 printk(" SS:ESP %04x:%08lx\n", ss, sp);
428 * This is gone through when something in the kernel has done something bad
429 * and is about to be terminated:
431 void die(const char *str, struct pt_regs *regs, long err)
436 int lock_owner_depth;
438 .lock = __RAW_SPIN_LOCK_UNLOCKED,
440 .lock_owner_depth = 0
446 if (die.lock_owner != raw_smp_processor_id()) {
448 raw_local_irq_save(flags);
449 __raw_spin_lock(&die.lock);
450 die.lock_owner = smp_processor_id();
451 die.lock_owner_depth = 0;
454 raw_local_irq_save(flags);
457 if (++die.lock_owner_depth < 3) {
458 report_bug(regs->ip, regs);
460 if (__die(str, regs, err))
463 printk(KERN_EMERG "Recursive die() failure, output suppressed\n");
468 add_taint(TAINT_DIE);
469 __raw_spin_unlock(&die.lock);
470 raw_local_irq_restore(flags);
475 if (kexec_should_crash(current))
479 panic("Fatal exception in interrupt");
482 panic("Fatal exception");
489 die_if_kernel(const char *str, struct pt_regs *regs, long err)
491 if (!user_mode_vm(regs))
495 static void __kprobes
496 do_trap(int trapnr, int signr, char *str, int vm86, struct pt_regs *regs,
497 long error_code, siginfo_t *info)
499 struct task_struct *tsk = current;
501 if (regs->flags & X86_VM_MASK) {
507 if (!user_mode(regs))
512 * We want error_code and trap_no set for userspace faults and
513 * kernelspace faults which result in die(), but not
514 * kernelspace faults which are fixed up. die() gives the
515 * process no chance to handle the signal and notice the
516 * kernel fault information, so that won't result in polluting
517 * the information about previously queued, but not yet
518 * delivered, faults. See also do_general_protection below.
520 tsk->thread.error_code = error_code;
521 tsk->thread.trap_no = trapnr;
524 force_sig_info(signr, info, tsk);
526 force_sig(signr, tsk);
530 if (!fixup_exception(regs)) {
531 tsk->thread.error_code = error_code;
532 tsk->thread.trap_no = trapnr;
533 die(str, regs, error_code);
538 if (handle_vm86_trap((struct kernel_vm86_regs *) regs,
544 #define DO_ERROR(trapnr, signr, str, name) \
545 void do_##name(struct pt_regs *regs, long error_code) \
547 trace_hardirqs_fixup(); \
548 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
551 do_trap(trapnr, signr, str, 0, regs, error_code, NULL); \
554 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr, irq) \
555 void do_##name(struct pt_regs *regs, long error_code) \
559 local_irq_enable(); \
560 info.si_signo = signr; \
562 info.si_code = sicode; \
563 info.si_addr = (void __user *)siaddr; \
564 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
567 do_trap(trapnr, signr, str, 0, regs, error_code, &info); \
570 #define DO_VM86_ERROR(trapnr, signr, str, name) \
571 void do_##name(struct pt_regs *regs, long error_code) \
573 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
576 do_trap(trapnr, signr, str, 1, regs, error_code, NULL); \
579 #define DO_VM86_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
580 void do_##name(struct pt_regs *regs, long error_code) \
583 info.si_signo = signr; \
585 info.si_code = sicode; \
586 info.si_addr = (void __user *)siaddr; \
587 trace_hardirqs_fixup(); \
588 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
591 do_trap(trapnr, signr, str, 1, regs, error_code, &info); \
594 DO_VM86_ERROR_INFO(0, SIGFPE, "divide error", divide_error, FPE_INTDIV, regs->ip)
595 #ifndef CONFIG_KPROBES
596 DO_VM86_ERROR(3, SIGTRAP, "int3", int3)
598 DO_VM86_ERROR(4, SIGSEGV, "overflow", overflow)
599 DO_VM86_ERROR(5, SIGSEGV, "bounds", bounds)
600 DO_ERROR_INFO(6, SIGILL, "invalid opcode", invalid_op, ILL_ILLOPN, regs->ip, 0)
601 DO_ERROR(9, SIGFPE, "coprocessor segment overrun", coprocessor_segment_overrun)
602 DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
603 DO_ERROR(11, SIGBUS, "segment not present", segment_not_present)
604 DO_ERROR(12, SIGBUS, "stack segment", stack_segment)
605 DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0, 0)
606 DO_ERROR_INFO(32, SIGILL, "iret exception", iret_error, ILL_BADSTK, 0, 1)
608 void __kprobes do_general_protection(struct pt_regs *regs, long error_code)
610 struct thread_struct *thread;
611 struct tss_struct *tss;
615 tss = &per_cpu(init_tss, cpu);
616 thread = ¤t->thread;
619 * Perform the lazy TSS's I/O bitmap copy. If the TSS has an
620 * invalid offset set (the LAZY one) and the faulting thread has
621 * a valid I/O bitmap pointer, we copy the I/O bitmap in the TSS
622 * and we set the offset field correctly. Then we let the CPU to
623 * restart the faulting instruction.
625 if (tss->x86_tss.io_bitmap_base == INVALID_IO_BITMAP_OFFSET_LAZY &&
626 thread->io_bitmap_ptr) {
627 memcpy(tss->io_bitmap, thread->io_bitmap_ptr,
628 thread->io_bitmap_max);
630 * If the previously set map was extending to higher ports
631 * than the current one, pad extra space with 0xff (no access).
633 if (thread->io_bitmap_max < tss->io_bitmap_max) {
634 memset((char *) tss->io_bitmap +
635 thread->io_bitmap_max, 0xff,
636 tss->io_bitmap_max - thread->io_bitmap_max);
638 tss->io_bitmap_max = thread->io_bitmap_max;
639 tss->x86_tss.io_bitmap_base = IO_BITMAP_OFFSET;
640 tss->io_bitmap_owner = thread;
647 if (regs->flags & X86_VM_MASK)
650 if (!user_mode(regs))
653 current->thread.error_code = error_code;
654 current->thread.trap_no = 13;
656 if (show_unhandled_signals && unhandled_signal(current, SIGSEGV) &&
657 printk_ratelimit()) {
659 "%s[%d] general protection ip:%lx sp:%lx error:%lx",
660 current->comm, task_pid_nr(current),
661 regs->ip, regs->sp, error_code);
662 print_vma_addr(" in ", regs->ip);
666 force_sig(SIGSEGV, current);
671 handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
675 if (!fixup_exception(regs)) {
676 current->thread.error_code = error_code;
677 current->thread.trap_no = 13;
678 if (notify_die(DIE_GPF, "general protection fault", regs,
679 error_code, 13, SIGSEGV) == NOTIFY_STOP)
681 die("general protection fault", regs, error_code);
685 static notrace __kprobes void
686 mem_parity_error(unsigned char reason, struct pt_regs *regs)
689 "Uhhuh. NMI received for unknown reason %02x on CPU %d.\n",
690 reason, smp_processor_id());
693 "You have some hardware problem, likely on the PCI bus.\n");
695 #if defined(CONFIG_EDAC)
696 if (edac_handler_set()) {
697 edac_atomic_assert_error();
702 if (panic_on_unrecovered_nmi)
703 panic("NMI: Not continuing");
705 printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
707 /* Clear and disable the memory parity error line. */
708 clear_mem_error(reason);
711 static notrace __kprobes void
712 io_check_error(unsigned char reason, struct pt_regs *regs)
716 printk(KERN_EMERG "NMI: IOCK error (debug interrupt?)\n");
717 show_registers(regs);
719 /* Re-enable the IOCK line, wait for a few seconds */
720 reason = (reason & 0xf) | 8;
731 static notrace __kprobes void
732 unknown_nmi_error(unsigned char reason, struct pt_regs *regs)
734 if (notify_die(DIE_NMIUNKNOWN, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP)
738 * Might actually be able to figure out what the guilty party
747 "Uhhuh. NMI received for unknown reason %02x on CPU %d.\n",
748 reason, smp_processor_id());
750 printk(KERN_EMERG "Do you have a strange power saving mode enabled?\n");
751 if (panic_on_unrecovered_nmi)
752 panic("NMI: Not continuing");
754 printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
757 static DEFINE_SPINLOCK(nmi_print_lock);
759 void notrace __kprobes die_nmi(struct pt_regs *regs, const char *msg)
761 if (notify_die(DIE_NMIWATCHDOG, msg, regs, 0, 2, SIGINT) == NOTIFY_STOP)
764 spin_lock(&nmi_print_lock);
766 * We are in trouble anyway, lets at least try
767 * to get a message out:
770 printk(KERN_EMERG "%s", msg);
771 printk(" on CPU%d, ip %08lx, registers:\n",
772 smp_processor_id(), regs->ip);
773 show_registers(regs);
775 spin_unlock(&nmi_print_lock);
779 * If we are in kernel we are probably nested up pretty bad
780 * and might aswell get out now while we still can:
782 if (!user_mode_vm(regs)) {
783 current->thread.trap_no = 2;
790 static notrace __kprobes void default_do_nmi(struct pt_regs *regs)
792 unsigned char reason = 0;
794 /* Only the BSP gets external NMIs from the system: */
795 if (!smp_processor_id())
796 reason = get_nmi_reason();
798 if (!(reason & 0xc0)) {
799 if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 2, SIGINT)
802 #ifdef CONFIG_X86_LOCAL_APIC
804 * Ok, so this is none of the documented NMI sources,
805 * so it must be the NMI watchdog.
807 if (nmi_watchdog_tick(regs, reason))
809 if (!do_nmi_callback(regs, smp_processor_id()))
810 unknown_nmi_error(reason, regs);
812 unknown_nmi_error(reason, regs);
817 if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP)
820 mem_parity_error(reason, regs);
822 io_check_error(reason, regs);
824 * Reassert NMI in case it became active meanwhile
825 * as it's edge-triggered:
830 static int ignore_nmis;
832 notrace __kprobes void do_nmi(struct pt_regs *regs, long error_code)
838 cpu = smp_processor_id();
843 default_do_nmi(regs);
854 void restart_nmi(void)
860 #ifdef CONFIG_KPROBES
861 void __kprobes do_int3(struct pt_regs *regs, long error_code)
863 trace_hardirqs_fixup();
865 if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP)
869 * This is an interrupt gate, because kprobes wants interrupts
870 * disabled. Normal trap handlers don't.
872 restore_interrupts(regs);
874 do_trap(3, SIGTRAP, "int3", 1, regs, error_code, NULL);
879 * Our handling of the processor debug registers is non-trivial.
880 * We do not clear them on entry and exit from the kernel. Therefore
881 * it is possible to get a watchpoint trap here from inside the kernel.
882 * However, the code in ./ptrace.c has ensured that the user can
883 * only set watchpoints on userspace addresses. Therefore the in-kernel
884 * watchpoint trap can only occur in code which is reading/writing
885 * from user space. Such code must not hold kernel locks (since it
886 * can equally take a page fault), therefore it is safe to call
887 * force_sig_info even though that claims and releases locks.
889 * Code in ./signal.c ensures that the debug control register
890 * is restored before we deliver any signal, and therefore that
891 * user code runs with the correct debug control register even though
894 * Being careful here means that we don't have to be as careful in a
895 * lot of more complicated places (task switching can be a bit lazy
896 * about restoring all the debug state, and ptrace doesn't have to
897 * find every occurrence of the TF bit that could be saved away even
900 void __kprobes do_debug(struct pt_regs *regs, long error_code)
902 struct task_struct *tsk = current;
903 unsigned int condition;
905 trace_hardirqs_fixup();
907 get_debugreg(condition, 6);
910 * The processor cleared BTF, so don't mark that we need it set.
912 clear_tsk_thread_flag(tsk, TIF_DEBUGCTLMSR);
913 tsk->thread.debugctlmsr = 0;
915 if (notify_die(DIE_DEBUG, "debug", regs, condition, error_code,
916 SIGTRAP) == NOTIFY_STOP)
918 /* It's safe to allow irq's after DR6 has been saved */
919 if (regs->flags & X86_EFLAGS_IF)
922 /* Mask out spurious debug traps due to lazy DR7 setting */
923 if (condition & (DR_TRAP0|DR_TRAP1|DR_TRAP2|DR_TRAP3)) {
924 if (!tsk->thread.debugreg7)
928 if (regs->flags & X86_VM_MASK)
931 /* Save debug status register where ptrace can see it */
932 tsk->thread.debugreg6 = condition;
935 * Single-stepping through TF: make sure we ignore any events in
936 * kernel space (but re-enable TF when returning to user mode).
938 if (condition & DR_STEP) {
940 * We already checked v86 mode above, so we can
941 * check for kernel mode by just checking the CPL
944 if (!user_mode(regs))
945 goto clear_TF_reenable;
948 /* Ok, finally something we can handle */
949 send_sigtrap(tsk, regs, error_code);
952 * Disable additional traps. They'll be re-enabled when
953 * the signal is delivered.
960 handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code, 1);
964 set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
965 regs->flags &= ~X86_EFLAGS_TF;
970 * Note that we play around with the 'TS' bit in an attempt to get
971 * the correct behaviour even in the presence of the asynchronous
974 void math_error(void __user *ip)
976 struct task_struct *task;
982 * Save the info for the exception handler and clear the error.
986 task->thread.trap_no = 16;
987 task->thread.error_code = 0;
988 info.si_signo = SIGFPE;
990 info.si_code = __SI_FAULT;
993 * (~cwd & swd) will mask out exceptions that are not set to unmasked
994 * status. 0x3f is the exception bits in these regs, 0x200 is the
995 * C1 reg you need in case of a stack fault, 0x040 is the stack
996 * fault bit. We should only be taking one exception at a time,
997 * so if this combination doesn't produce any single exception,
998 * then we have a bad program that isn't syncronizing its FPU usage
999 * and it will suffer the consequences since we won't be able to
1000 * fully reproduce the context of the exception
1002 cwd = get_fpu_cwd(task);
1003 swd = get_fpu_swd(task);
1004 switch (swd & ~cwd & 0x3f) {
1005 case 0x000: /* No unmasked exception */
1007 default: /* Multiple exceptions */
1009 case 0x001: /* Invalid Op */
1011 * swd & 0x240 == 0x040: Stack Underflow
1012 * swd & 0x240 == 0x240: Stack Overflow
1013 * User must clear the SF bit (0x40) if set
1015 info.si_code = FPE_FLTINV;
1017 case 0x002: /* Denormalize */
1018 case 0x010: /* Underflow */
1019 info.si_code = FPE_FLTUND;
1021 case 0x004: /* Zero Divide */
1022 info.si_code = FPE_FLTDIV;
1024 case 0x008: /* Overflow */
1025 info.si_code = FPE_FLTOVF;
1027 case 0x020: /* Precision */
1028 info.si_code = FPE_FLTRES;
1031 force_sig_info(SIGFPE, &info, task);
1034 void do_coprocessor_error(struct pt_regs *regs, long error_code)
1037 math_error((void __user *)regs->ip);
1040 static void simd_math_error(void __user *ip)
1042 struct task_struct *task;
1043 unsigned short mxcsr;
1047 * Save the info for the exception handler and clear the error.
1050 save_init_fpu(task);
1051 task->thread.trap_no = 19;
1052 task->thread.error_code = 0;
1053 info.si_signo = SIGFPE;
1055 info.si_code = __SI_FAULT;
1058 * The SIMD FPU exceptions are handled a little differently, as there
1059 * is only a single status/control register. Thus, to determine which
1060 * unmasked exception was caught we must mask the exception mask bits
1061 * at 0x1f80, and then use these to mask the exception bits at 0x3f.
1063 mxcsr = get_fpu_mxcsr(task);
1064 switch (~((mxcsr & 0x1f80) >> 7) & (mxcsr & 0x3f)) {
1068 case 0x001: /* Invalid Op */
1069 info.si_code = FPE_FLTINV;
1071 case 0x002: /* Denormalize */
1072 case 0x010: /* Underflow */
1073 info.si_code = FPE_FLTUND;
1075 case 0x004: /* Zero Divide */
1076 info.si_code = FPE_FLTDIV;
1078 case 0x008: /* Overflow */
1079 info.si_code = FPE_FLTOVF;
1081 case 0x020: /* Precision */
1082 info.si_code = FPE_FLTRES;
1085 force_sig_info(SIGFPE, &info, task);
1088 void do_simd_coprocessor_error(struct pt_regs *regs, long error_code)
1091 /* Handle SIMD FPU exceptions on PIII+ processors. */
1093 simd_math_error((void __user *)regs->ip);
1097 * Handle strange cache flush from user space exception
1098 * in all other cases. This is undocumented behaviour.
1100 if (regs->flags & X86_VM_MASK) {
1101 handle_vm86_fault((struct kernel_vm86_regs *)regs, error_code);
1104 current->thread.trap_no = 19;
1105 current->thread.error_code = error_code;
1106 die_if_kernel("cache flush denied", regs, error_code);
1107 force_sig(SIGSEGV, current);
1110 void do_spurious_interrupt_bug(struct pt_regs *regs, long error_code)
1113 /* No need to warn about this any longer. */
1114 printk(KERN_INFO "Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
1118 unsigned long patch_espfix_desc(unsigned long uesp, unsigned long kesp)
1120 struct desc_struct *gdt = __get_cpu_var(gdt_page).gdt;
1121 unsigned long base = (kesp - uesp) & -THREAD_SIZE;
1122 unsigned long new_kesp = kesp - base;
1123 unsigned long lim_pages = (new_kesp | (THREAD_SIZE - 1)) >> PAGE_SHIFT;
1124 __u64 desc = *(__u64 *)&gdt[GDT_ENTRY_ESPFIX_SS];
1126 /* Set up base for espfix segment */
1127 desc &= 0x00f0ff0000000000ULL;
1128 desc |= ((((__u64)base) << 16) & 0x000000ffffff0000ULL) |
1129 ((((__u64)base) << 32) & 0xff00000000000000ULL) |
1130 ((((__u64)lim_pages) << 32) & 0x000f000000000000ULL) |
1131 (lim_pages & 0xffff);
1132 *(__u64 *)&gdt[GDT_ENTRY_ESPFIX_SS] = desc;
1138 * 'math_state_restore()' saves the current math information in the
1139 * old math state array, and gets the new ones from the current task
1141 * Careful.. There are problems with IBM-designed IRQ13 behaviour.
1142 * Don't touch unless you *really* know how it works.
1144 * Must be called with kernel preemption disabled (in this case,
1145 * local interrupts are disabled at the call-site in entry.S).
1147 asmlinkage void math_state_restore(void)
1149 struct thread_info *thread = current_thread_info();
1150 struct task_struct *tsk = thread->task;
1152 if (!tsk_used_math(tsk)) {
1155 * does a slab alloc which can sleep
1157 if (init_fpu(tsk)) {
1159 * ran out of memory!
1161 do_group_exit(SIGKILL);
1164 local_irq_disable();
1167 clts(); /* Allow maths ops (or we recurse) */
1169 thread->status |= TS_USEDFPU; /* So we fnsave on switch_to() */
1172 EXPORT_SYMBOL_GPL(math_state_restore);
1174 #ifndef CONFIG_MATH_EMULATION
1176 asmlinkage void math_emulate(long arg)
1179 "math-emulation not enabled and no coprocessor found.\n");
1180 printk(KERN_EMERG "killing %s.\n", current->comm);
1181 force_sig(SIGFPE, current);
1185 #endif /* CONFIG_MATH_EMULATION */
1187 void __init trap_init(void)
1192 void __iomem *p = early_ioremap(0x0FFFD9, 4);
1194 if (readl(p) == 'E' + ('I'<<8) + ('S'<<16) + ('A'<<24))
1196 early_iounmap(p, 4);
1199 #ifdef CONFIG_X86_LOCAL_APIC
1200 init_apic_mappings();
1202 set_trap_gate(0, ÷_error);
1203 set_intr_gate(1, &debug);
1204 set_intr_gate(2, &nmi);
1205 set_system_intr_gate(3, &int3); /* int3/4 can be called from all */
1206 set_system_gate(4, &overflow);
1207 set_trap_gate(5, &bounds);
1208 set_trap_gate(6, &invalid_op);
1209 set_trap_gate(7, &device_not_available);
1210 set_task_gate(8, GDT_ENTRY_DOUBLEFAULT_TSS);
1211 set_trap_gate(9, &coprocessor_segment_overrun);
1212 set_trap_gate(10, &invalid_TSS);
1213 set_trap_gate(11, &segment_not_present);
1214 set_trap_gate(12, &stack_segment);
1215 set_trap_gate(13, &general_protection);
1216 set_intr_gate(14, &page_fault);
1217 set_trap_gate(15, &spurious_interrupt_bug);
1218 set_trap_gate(16, &coprocessor_error);
1219 set_trap_gate(17, &alignment_check);
1220 #ifdef CONFIG_X86_MCE
1221 set_trap_gate(18, &machine_check);
1223 set_trap_gate(19, &simd_coprocessor_error);
1226 printk(KERN_INFO "Enabling fast FPU save and restore... ");
1227 set_in_cr4(X86_CR4_OSFXSR);
1232 "Enabling unmasked SIMD FPU exception support... ");
1233 set_in_cr4(X86_CR4_OSXMMEXCPT);
1237 set_system_gate(SYSCALL_VECTOR, &system_call);
1239 /* Reserve all the builtin and the syscall vector: */
1240 for (i = 0; i < FIRST_EXTERNAL_VECTOR; i++)
1241 set_bit(i, used_vectors);
1243 set_bit(SYSCALL_VECTOR, used_vectors);
1245 init_thread_xstate();
1247 * Should be a barrier for any external CPU state:
1254 static int __init kstack_setup(char *s)
1256 kstack_depth_to_print = simple_strtoul(s, NULL, 0);
1260 __setup("kstack=", kstack_setup);
1262 static int __init code_bytes_setup(char *s)
1264 code_bytes = simple_strtoul(s, NULL, 0);
1265 if (code_bytes > 8192)
1270 __setup("code_bytes=", code_bytes_setup);