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 #if defined(CONFIG_EDAC)
45 #include <linux/edac.h>
48 #include <asm/processor.h>
49 #include <asm/system.h>
51 #include <asm/atomic.h>
52 #include <asm/debugreg.h>
56 #include <asm/unwind.h>
58 #include <asm/arch_hooks.h>
59 #include <linux/kdebug.h>
60 #include <asm/stacktrace.h>
62 #include <linux/module.h>
64 #include "mach_traps.h"
66 int panic_on_unrecovered_nmi;
68 asmlinkage int system_call(void);
70 /* Do we ignore FPU interrupts ? */
71 char ignore_fpu_irq = 0;
74 * The IDT has to be page-aligned to simplify the Pentium
75 * F0 0F bug workaround.. We have a special link segment
78 struct desc_struct idt_table[256] __attribute__((__section__(".data.idt"))) = { {0, 0}, };
80 asmlinkage void divide_error(void);
81 asmlinkage void debug(void);
82 asmlinkage void nmi(void);
83 asmlinkage void int3(void);
84 asmlinkage void overflow(void);
85 asmlinkage void bounds(void);
86 asmlinkage void invalid_op(void);
87 asmlinkage void device_not_available(void);
88 asmlinkage void coprocessor_segment_overrun(void);
89 asmlinkage void invalid_TSS(void);
90 asmlinkage void segment_not_present(void);
91 asmlinkage void stack_segment(void);
92 asmlinkage void general_protection(void);
93 asmlinkage void page_fault(void);
94 asmlinkage void coprocessor_error(void);
95 asmlinkage void simd_coprocessor_error(void);
96 asmlinkage void alignment_check(void);
97 asmlinkage void spurious_interrupt_bug(void);
98 asmlinkage void machine_check(void);
100 int kstack_depth_to_print = 24;
101 static unsigned int code_bytes = 64;
103 static inline int valid_stack_ptr(struct thread_info *tinfo, void *p, unsigned size)
105 return p > (void *)tinfo &&
106 p <= (void *)tinfo + THREAD_SIZE - size;
109 /* The form of the top of the frame on the stack */
111 struct stack_frame *next_frame;
112 unsigned long return_address;
115 static inline unsigned long print_context_stack(struct thread_info *tinfo,
116 unsigned long *stack, unsigned long ebp,
117 struct stacktrace_ops *ops, void *data)
119 #ifdef CONFIG_FRAME_POINTER
120 struct stack_frame *frame = (struct stack_frame *)ebp;
121 while (valid_stack_ptr(tinfo, frame, sizeof(*frame))) {
122 struct stack_frame *next;
125 addr = frame->return_address;
126 ops->address(data, addr);
128 * break out of recursive entries (such as
129 * end_of_stack_stop_unwind_function). Also,
130 * we can never allow a frame pointer to
133 next = frame->next_frame;
139 while (valid_stack_ptr(tinfo, stack, sizeof(*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;
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);
223 touch_nmi_watchdog();
226 static struct stacktrace_ops print_trace_ops = {
227 .warning = print_trace_warning,
228 .warning_symbol = print_trace_warning_symbol,
229 .stack = print_trace_stack,
230 .address = print_trace_address,
234 show_trace_log_lvl(struct task_struct *task, struct pt_regs *regs,
235 unsigned long * stack, char *log_lvl)
237 dump_trace(task, regs, stack, &print_trace_ops, log_lvl);
238 printk("%s =======================\n", log_lvl);
241 void show_trace(struct task_struct *task, struct pt_regs *regs,
242 unsigned long * stack)
244 show_trace_log_lvl(task, regs, stack, "");
247 static void show_stack_log_lvl(struct task_struct *task, struct pt_regs *regs,
248 unsigned long *esp, char *log_lvl)
250 unsigned long *stack;
255 esp = (unsigned long*)task->thread.esp;
257 esp = (unsigned long *)&esp;
261 for(i = 0; i < kstack_depth_to_print; i++) {
262 if (kstack_end(stack))
264 if (i && ((i % 8) == 0))
265 printk("\n%s ", log_lvl);
266 printk("%08lx ", *stack++);
268 printk("\n%sCall Trace:\n", log_lvl);
269 show_trace_log_lvl(task, regs, esp, log_lvl);
272 void show_stack(struct task_struct *task, unsigned long *esp)
275 show_stack_log_lvl(task, NULL, esp, "");
279 * The architecture-independent dump_stack generator
281 void dump_stack(void)
285 show_trace(current, NULL, &stack);
288 EXPORT_SYMBOL(dump_stack);
290 void show_registers(struct pt_regs *regs)
295 unsigned short ss, gs;
297 esp = (unsigned long) (®s->esp);
300 if (user_mode_vm(regs)) {
303 ss = regs->xss & 0xffff;
306 printk(KERN_EMERG "CPU: %d\n"
307 KERN_EMERG "EIP: %04x:[<%08lx>] %s VLI\n"
308 KERN_EMERG "EFLAGS: %08lx (%s %.*s)\n",
309 smp_processor_id(), 0xffff & regs->xcs, regs->eip,
310 print_tainted(), regs->eflags, init_utsname()->release,
311 (int)strcspn(init_utsname()->version, " "),
312 init_utsname()->version);
313 print_symbol(KERN_EMERG "EIP is at %s\n", regs->eip);
314 printk(KERN_EMERG "eax: %08lx ebx: %08lx ecx: %08lx edx: %08lx\n",
315 regs->eax, regs->ebx, regs->ecx, regs->edx);
316 printk(KERN_EMERG "esi: %08lx edi: %08lx ebp: %08lx esp: %08lx\n",
317 regs->esi, regs->edi, regs->ebp, esp);
318 printk(KERN_EMERG "ds: %04x es: %04x fs: %04x gs: %04x ss: %04x\n",
319 regs->xds & 0xffff, regs->xes & 0xffff, regs->xfs & 0xffff, gs, ss);
320 printk(KERN_EMERG "Process %.*s (pid: %d, ti=%p task=%p task.ti=%p)",
321 TASK_COMM_LEN, current->comm, current->pid,
322 current_thread_info(), current, task_thread_info(current));
324 * When in-kernel, we also print out the stack and code at the
325 * time of the fault..
329 unsigned int code_prologue = code_bytes * 43 / 64;
330 unsigned int code_len = code_bytes;
333 printk("\n" KERN_EMERG "Stack: ");
334 show_stack_log_lvl(NULL, regs, (unsigned long *)esp, KERN_EMERG);
336 printk(KERN_EMERG "Code: ");
338 eip = (u8 *)regs->eip - code_prologue;
339 if (eip < (u8 *)PAGE_OFFSET ||
340 probe_kernel_address(eip, c)) {
341 /* try starting at EIP */
342 eip = (u8 *)regs->eip;
343 code_len = code_len - code_prologue + 1;
345 for (i = 0; i < code_len; i++, eip++) {
346 if (eip < (u8 *)PAGE_OFFSET ||
347 probe_kernel_address(eip, c)) {
348 printk(" Bad EIP value.");
351 if (eip == (u8 *)regs->eip)
352 printk("<%02x> ", c);
360 int is_valid_bugaddr(unsigned long eip)
364 if (eip < PAGE_OFFSET)
366 if (probe_kernel_address((unsigned short *)eip, ud2))
369 return ud2 == 0x0b0f;
373 * This is gone through when something in the kernel has done something bad and
374 * is about to be terminated.
376 void die(const char * str, struct pt_regs * regs, long err)
381 int lock_owner_depth;
383 .lock = __SPIN_LOCK_UNLOCKED(die.lock),
385 .lock_owner_depth = 0
387 static int die_counter;
392 if (die.lock_owner != raw_smp_processor_id()) {
394 spin_lock_irqsave(&die.lock, flags);
395 die.lock_owner = smp_processor_id();
396 die.lock_owner_depth = 0;
400 local_save_flags(flags);
402 if (++die.lock_owner_depth < 3) {
407 report_bug(regs->eip, regs);
409 printk(KERN_EMERG "%s: %04lx [#%d]\n", str, err & 0xffff, ++die_counter);
410 #ifdef CONFIG_PREEMPT
411 printk(KERN_EMERG "PREEMPT ");
420 #ifdef CONFIG_DEBUG_PAGEALLOC
423 printk("DEBUG_PAGEALLOC");
428 if (notify_die(DIE_OOPS, str, regs, err,
429 current->thread.trap_no, SIGSEGV) !=
431 show_registers(regs);
432 /* Executive summary in case the oops scrolled away */
433 esp = (unsigned long) (®s->esp);
435 if (user_mode(regs)) {
437 ss = regs->xss & 0xffff;
439 printk(KERN_EMERG "EIP: [<%08lx>] ", regs->eip);
440 print_symbol("%s", regs->eip);
441 printk(" SS:ESP %04x:%08lx\n", ss, esp);
446 printk(KERN_EMERG "Recursive die() failure, output suppressed\n");
450 add_taint(TAINT_DIE);
451 spin_unlock_irqrestore(&die.lock, flags);
456 if (kexec_should_crash(current))
460 panic("Fatal exception in interrupt");
463 panic("Fatal exception");
469 static inline void die_if_kernel(const char * str, struct pt_regs * regs, long err)
471 if (!user_mode_vm(regs))
475 static void __kprobes do_trap(int trapnr, int signr, char *str, int vm86,
476 struct pt_regs * regs, long error_code,
479 struct task_struct *tsk = current;
481 if (regs->eflags & VM_MASK) {
487 if (!user_mode(regs))
492 * We want error_code and trap_no set for userspace faults and
493 * kernelspace faults which result in die(), but not
494 * kernelspace faults which are fixed up. die() gives the
495 * process no chance to handle the signal and notice the
496 * kernel fault information, so that won't result in polluting
497 * the information about previously queued, but not yet
498 * delivered, faults. See also do_general_protection below.
500 tsk->thread.error_code = error_code;
501 tsk->thread.trap_no = trapnr;
504 force_sig_info(signr, info, tsk);
506 force_sig(signr, tsk);
511 if (!fixup_exception(regs)) {
512 tsk->thread.error_code = error_code;
513 tsk->thread.trap_no = trapnr;
514 die(str, regs, error_code);
520 int ret = handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code, trapnr);
521 if (ret) goto trap_signal;
526 #define DO_ERROR(trapnr, signr, str, name) \
527 fastcall void do_##name(struct pt_regs * regs, long error_code) \
529 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
532 do_trap(trapnr, signr, str, 0, regs, error_code, NULL); \
535 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr, irq) \
536 fastcall void do_##name(struct pt_regs * regs, long error_code) \
540 local_irq_enable(); \
541 info.si_signo = signr; \
543 info.si_code = sicode; \
544 info.si_addr = (void __user *)siaddr; \
545 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
548 do_trap(trapnr, signr, str, 0, regs, error_code, &info); \
551 #define DO_VM86_ERROR(trapnr, signr, str, name) \
552 fastcall void do_##name(struct pt_regs * regs, long error_code) \
554 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
557 do_trap(trapnr, signr, str, 1, regs, error_code, NULL); \
560 #define DO_VM86_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
561 fastcall void do_##name(struct pt_regs * regs, long error_code) \
564 info.si_signo = signr; \
566 info.si_code = sicode; \
567 info.si_addr = (void __user *)siaddr; \
568 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
571 do_trap(trapnr, signr, str, 1, regs, error_code, &info); \
574 DO_VM86_ERROR_INFO( 0, SIGFPE, "divide error", divide_error, FPE_INTDIV, regs->eip)
575 #ifndef CONFIG_KPROBES
576 DO_VM86_ERROR( 3, SIGTRAP, "int3", int3)
578 DO_VM86_ERROR( 4, SIGSEGV, "overflow", overflow)
579 DO_VM86_ERROR( 5, SIGSEGV, "bounds", bounds)
580 DO_ERROR_INFO( 6, SIGILL, "invalid opcode", invalid_op, ILL_ILLOPN, regs->eip, 0)
581 DO_ERROR( 9, SIGFPE, "coprocessor segment overrun", coprocessor_segment_overrun)
582 DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
583 DO_ERROR(11, SIGBUS, "segment not present", segment_not_present)
584 DO_ERROR(12, SIGBUS, "stack segment", stack_segment)
585 DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0, 0)
586 DO_ERROR_INFO(32, SIGSEGV, "iret exception", iret_error, ILL_BADSTK, 0, 1)
588 fastcall void __kprobes do_general_protection(struct pt_regs * regs,
592 struct tss_struct *tss = &per_cpu(init_tss, cpu);
593 struct thread_struct *thread = ¤t->thread;
596 * Perform the lazy TSS's I/O bitmap copy. If the TSS has an
597 * invalid offset set (the LAZY one) and the faulting thread has
598 * a valid I/O bitmap pointer, we copy the I/O bitmap in the TSS
599 * and we set the offset field correctly. Then we let the CPU to
600 * restart the faulting instruction.
602 if (tss->x86_tss.io_bitmap_base == INVALID_IO_BITMAP_OFFSET_LAZY &&
603 thread->io_bitmap_ptr) {
604 memcpy(tss->io_bitmap, thread->io_bitmap_ptr,
605 thread->io_bitmap_max);
607 * If the previously set map was extending to higher ports
608 * than the current one, pad extra space with 0xff (no access).
610 if (thread->io_bitmap_max < tss->io_bitmap_max)
611 memset((char *) tss->io_bitmap +
612 thread->io_bitmap_max, 0xff,
613 tss->io_bitmap_max - thread->io_bitmap_max);
614 tss->io_bitmap_max = thread->io_bitmap_max;
615 tss->x86_tss.io_bitmap_base = IO_BITMAP_OFFSET;
616 tss->io_bitmap_owner = thread;
622 if (regs->eflags & VM_MASK)
625 if (!user_mode(regs))
628 current->thread.error_code = error_code;
629 current->thread.trap_no = 13;
630 if (show_unhandled_signals && unhandled_signal(current, SIGSEGV) &&
633 "%s[%d] general protection eip:%lx esp:%lx error:%lx\n",
634 current->comm, current->pid,
635 regs->eip, regs->esp, error_code);
637 force_sig(SIGSEGV, current);
642 handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
646 if (!fixup_exception(regs)) {
647 current->thread.error_code = error_code;
648 current->thread.trap_no = 13;
649 if (notify_die(DIE_GPF, "general protection fault", regs,
650 error_code, 13, SIGSEGV) == NOTIFY_STOP)
652 die("general protection fault", regs, error_code);
656 static __kprobes void
657 mem_parity_error(unsigned char reason, struct pt_regs * regs)
659 printk(KERN_EMERG "Uhhuh. NMI received for unknown reason %02x on "
660 "CPU %d.\n", reason, smp_processor_id());
661 printk(KERN_EMERG "You have some hardware problem, likely on the PCI bus.\n");
663 #if defined(CONFIG_EDAC)
664 if(edac_handler_set()) {
665 edac_atomic_assert_error();
670 if (panic_on_unrecovered_nmi)
671 panic("NMI: Not continuing");
673 printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
675 /* Clear and disable the memory parity error line. */
676 clear_mem_error(reason);
679 static __kprobes void
680 io_check_error(unsigned char reason, struct pt_regs * regs)
684 printk(KERN_EMERG "NMI: IOCK error (debug interrupt?)\n");
685 show_registers(regs);
687 /* Re-enable the IOCK line, wait for a few seconds */
688 reason = (reason & 0xf) | 8;
691 while (--i) udelay(1000);
696 static __kprobes void
697 unknown_nmi_error(unsigned char reason, struct pt_regs * regs)
700 /* Might actually be able to figure out what the guilty party
707 printk(KERN_EMERG "Uhhuh. NMI received for unknown reason %02x on "
708 "CPU %d.\n", reason, smp_processor_id());
709 printk(KERN_EMERG "Do you have a strange power saving mode enabled?\n");
710 if (panic_on_unrecovered_nmi)
711 panic("NMI: Not continuing");
713 printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
716 static DEFINE_SPINLOCK(nmi_print_lock);
718 void __kprobes die_nmi(struct pt_regs *regs, const char *msg)
720 if (notify_die(DIE_NMIWATCHDOG, msg, regs, 0, 2, SIGINT) ==
724 spin_lock(&nmi_print_lock);
726 * We are in trouble anyway, lets at least try
727 * to get a message out.
730 printk(KERN_EMERG "%s", msg);
731 printk(" on CPU%d, eip %08lx, registers:\n",
732 smp_processor_id(), regs->eip);
733 show_registers(regs);
735 spin_unlock(&nmi_print_lock);
738 /* If we are in kernel we are probably nested up pretty bad
739 * and might aswell get out now while we still can.
741 if (!user_mode_vm(regs)) {
742 current->thread.trap_no = 2;
749 static __kprobes void default_do_nmi(struct pt_regs * regs)
751 unsigned char reason = 0;
753 /* Only the BSP gets external NMIs from the system. */
754 if (!smp_processor_id())
755 reason = get_nmi_reason();
757 if (!(reason & 0xc0)) {
758 if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 2, SIGINT)
761 #ifdef CONFIG_X86_LOCAL_APIC
763 * Ok, so this is none of the documented NMI sources,
764 * so it must be the NMI watchdog.
766 if (nmi_watchdog_tick(regs, reason))
768 if (!do_nmi_callback(regs, smp_processor_id()))
770 unknown_nmi_error(reason, regs);
774 if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP)
777 mem_parity_error(reason, regs);
779 io_check_error(reason, regs);
781 * Reassert NMI in case it became active meanwhile
782 * as it's edge-triggered.
787 static int ignore_nmis;
789 fastcall __kprobes void do_nmi(struct pt_regs * regs, long error_code)
795 cpu = smp_processor_id();
800 default_do_nmi(regs);
811 void restart_nmi(void)
817 #ifdef CONFIG_KPROBES
818 fastcall void __kprobes do_int3(struct pt_regs *regs, long error_code)
820 if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP)
823 /* This is an interrupt gate, because kprobes wants interrupts
824 disabled. Normal trap handlers don't. */
825 restore_interrupts(regs);
826 do_trap(3, SIGTRAP, "int3", 1, regs, error_code, NULL);
831 * Our handling of the processor debug registers is non-trivial.
832 * We do not clear them on entry and exit from the kernel. Therefore
833 * it is possible to get a watchpoint trap here from inside the kernel.
834 * However, the code in ./ptrace.c has ensured that the user can
835 * only set watchpoints on userspace addresses. Therefore the in-kernel
836 * watchpoint trap can only occur in code which is reading/writing
837 * from user space. Such code must not hold kernel locks (since it
838 * can equally take a page fault), therefore it is safe to call
839 * force_sig_info even though that claims and releases locks.
841 * Code in ./signal.c ensures that the debug control register
842 * is restored before we deliver any signal, and therefore that
843 * user code runs with the correct debug control register even though
846 * Being careful here means that we don't have to be as careful in a
847 * lot of more complicated places (task switching can be a bit lazy
848 * about restoring all the debug state, and ptrace doesn't have to
849 * find every occurrence of the TF bit that could be saved away even
852 fastcall void __kprobes do_debug(struct pt_regs * regs, long error_code)
854 unsigned int condition;
855 struct task_struct *tsk = current;
857 get_debugreg(condition, 6);
859 if (notify_die(DIE_DEBUG, "debug", regs, condition, error_code,
860 SIGTRAP) == NOTIFY_STOP)
862 /* It's safe to allow irq's after DR6 has been saved */
863 if (regs->eflags & X86_EFLAGS_IF)
866 /* Mask out spurious debug traps due to lazy DR7 setting */
867 if (condition & (DR_TRAP0|DR_TRAP1|DR_TRAP2|DR_TRAP3)) {
868 if (!tsk->thread.debugreg[7])
872 if (regs->eflags & VM_MASK)
875 /* Save debug status register where ptrace can see it */
876 tsk->thread.debugreg[6] = condition;
879 * Single-stepping through TF: make sure we ignore any events in
880 * kernel space (but re-enable TF when returning to user mode).
882 if (condition & DR_STEP) {
884 * We already checked v86 mode above, so we can
885 * check for kernel mode by just checking the CPL
888 if (!user_mode(regs))
889 goto clear_TF_reenable;
892 /* Ok, finally something we can handle */
893 send_sigtrap(tsk, regs, error_code);
895 /* Disable additional traps. They'll be re-enabled when
896 * the signal is delivered.
903 handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code, 1);
907 set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
908 regs->eflags &= ~TF_MASK;
913 * Note that we play around with the 'TS' bit in an attempt to get
914 * the correct behaviour even in the presence of the asynchronous
917 void math_error(void __user *eip)
919 struct task_struct * task;
921 unsigned short cwd, swd;
924 * Save the info for the exception handler and clear the error.
928 task->thread.trap_no = 16;
929 task->thread.error_code = 0;
930 info.si_signo = SIGFPE;
932 info.si_code = __SI_FAULT;
935 * (~cwd & swd) will mask out exceptions that are not set to unmasked
936 * status. 0x3f is the exception bits in these regs, 0x200 is the
937 * C1 reg you need in case of a stack fault, 0x040 is the stack
938 * fault bit. We should only be taking one exception at a time,
939 * so if this combination doesn't produce any single exception,
940 * then we have a bad program that isn't syncronizing its FPU usage
941 * and it will suffer the consequences since we won't be able to
942 * fully reproduce the context of the exception
944 cwd = get_fpu_cwd(task);
945 swd = get_fpu_swd(task);
946 switch (swd & ~cwd & 0x3f) {
947 case 0x000: /* No unmasked exception */
949 default: /* Multiple exceptions */
951 case 0x001: /* Invalid Op */
953 * swd & 0x240 == 0x040: Stack Underflow
954 * swd & 0x240 == 0x240: Stack Overflow
955 * User must clear the SF bit (0x40) if set
957 info.si_code = FPE_FLTINV;
959 case 0x002: /* Denormalize */
960 case 0x010: /* Underflow */
961 info.si_code = FPE_FLTUND;
963 case 0x004: /* Zero Divide */
964 info.si_code = FPE_FLTDIV;
966 case 0x008: /* Overflow */
967 info.si_code = FPE_FLTOVF;
969 case 0x020: /* Precision */
970 info.si_code = FPE_FLTRES;
973 force_sig_info(SIGFPE, &info, task);
976 fastcall void do_coprocessor_error(struct pt_regs * regs, long error_code)
979 math_error((void __user *)regs->eip);
982 static void simd_math_error(void __user *eip)
984 struct task_struct * task;
986 unsigned short mxcsr;
989 * Save the info for the exception handler and clear the error.
993 task->thread.trap_no = 19;
994 task->thread.error_code = 0;
995 info.si_signo = SIGFPE;
997 info.si_code = __SI_FAULT;
1000 * The SIMD FPU exceptions are handled a little differently, as there
1001 * is only a single status/control register. Thus, to determine which
1002 * unmasked exception was caught we must mask the exception mask bits
1003 * at 0x1f80, and then use these to mask the exception bits at 0x3f.
1005 mxcsr = get_fpu_mxcsr(task);
1006 switch (~((mxcsr & 0x1f80) >> 7) & (mxcsr & 0x3f)) {
1010 case 0x001: /* Invalid Op */
1011 info.si_code = FPE_FLTINV;
1013 case 0x002: /* Denormalize */
1014 case 0x010: /* Underflow */
1015 info.si_code = FPE_FLTUND;
1017 case 0x004: /* Zero Divide */
1018 info.si_code = FPE_FLTDIV;
1020 case 0x008: /* Overflow */
1021 info.si_code = FPE_FLTOVF;
1023 case 0x020: /* Precision */
1024 info.si_code = FPE_FLTRES;
1027 force_sig_info(SIGFPE, &info, task);
1030 fastcall void do_simd_coprocessor_error(struct pt_regs * regs,
1034 /* Handle SIMD FPU exceptions on PIII+ processors. */
1036 simd_math_error((void __user *)regs->eip);
1039 * Handle strange cache flush from user space exception
1040 * in all other cases. This is undocumented behaviour.
1042 if (regs->eflags & VM_MASK) {
1043 handle_vm86_fault((struct kernel_vm86_regs *)regs,
1047 current->thread.trap_no = 19;
1048 current->thread.error_code = error_code;
1049 die_if_kernel("cache flush denied", regs, error_code);
1050 force_sig(SIGSEGV, current);
1054 fastcall void do_spurious_interrupt_bug(struct pt_regs * regs,
1058 /* No need to warn about this any longer. */
1059 printk("Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
1063 fastcall unsigned long patch_espfix_desc(unsigned long uesp,
1066 struct desc_struct *gdt = __get_cpu_var(gdt_page).gdt;
1067 unsigned long base = (kesp - uesp) & -THREAD_SIZE;
1068 unsigned long new_kesp = kesp - base;
1069 unsigned long lim_pages = (new_kesp | (THREAD_SIZE - 1)) >> PAGE_SHIFT;
1070 __u64 desc = *(__u64 *)&gdt[GDT_ENTRY_ESPFIX_SS];
1071 /* Set up base for espfix segment */
1072 desc &= 0x00f0ff0000000000ULL;
1073 desc |= ((((__u64)base) << 16) & 0x000000ffffff0000ULL) |
1074 ((((__u64)base) << 32) & 0xff00000000000000ULL) |
1075 ((((__u64)lim_pages) << 32) & 0x000f000000000000ULL) |
1076 (lim_pages & 0xffff);
1077 *(__u64 *)&gdt[GDT_ENTRY_ESPFIX_SS] = desc;
1082 * 'math_state_restore()' saves the current math information in the
1083 * old math state array, and gets the new ones from the current task
1085 * Careful.. There are problems with IBM-designed IRQ13 behaviour.
1086 * Don't touch unless you *really* know how it works.
1088 * Must be called with kernel preemption disabled (in this case,
1089 * local interrupts are disabled at the call-site in entry.S).
1091 asmlinkage void math_state_restore(void)
1093 struct thread_info *thread = current_thread_info();
1094 struct task_struct *tsk = thread->task;
1096 clts(); /* Allow maths ops (or we recurse) */
1097 if (!tsk_used_math(tsk))
1100 thread->status |= TS_USEDFPU; /* So we fnsave on switch_to() */
1103 EXPORT_SYMBOL_GPL(math_state_restore);
1105 #ifndef CONFIG_MATH_EMULATION
1107 asmlinkage void math_emulate(long arg)
1109 printk(KERN_EMERG "math-emulation not enabled and no coprocessor found.\n");
1110 printk(KERN_EMERG "killing %s.\n",current->comm);
1111 force_sig(SIGFPE,current);
1115 #endif /* CONFIG_MATH_EMULATION */
1117 #ifdef CONFIG_X86_F00F_BUG
1118 void __init trap_init_f00f_bug(void)
1120 __set_fixmap(FIX_F00F_IDT, __pa(&idt_table), PAGE_KERNEL_RO);
1123 * Update the IDT descriptor and reload the IDT so that
1124 * it uses the read-only mapped virtual address.
1126 idt_descr.address = fix_to_virt(FIX_F00F_IDT);
1127 load_idt(&idt_descr);
1132 * This needs to use 'idt_table' rather than 'idt', and
1133 * thus use the _nonmapped_ version of the IDT, as the
1134 * Pentium F0 0F bugfix can have resulted in the mapped
1135 * IDT being write-protected.
1137 void set_intr_gate(unsigned int n, void *addr)
1139 _set_gate(n, DESCTYPE_INT, addr, __KERNEL_CS);
1143 * This routine sets up an interrupt gate at directory privilege level 3.
1145 static inline void set_system_intr_gate(unsigned int n, void *addr)
1147 _set_gate(n, DESCTYPE_INT | DESCTYPE_DPL3, addr, __KERNEL_CS);
1150 static void __init set_trap_gate(unsigned int n, void *addr)
1152 _set_gate(n, DESCTYPE_TRAP, addr, __KERNEL_CS);
1155 static void __init set_system_gate(unsigned int n, void *addr)
1157 _set_gate(n, DESCTYPE_TRAP | DESCTYPE_DPL3, addr, __KERNEL_CS);
1160 static void __init set_task_gate(unsigned int n, unsigned int gdt_entry)
1162 _set_gate(n, DESCTYPE_TASK, (void *)0, (gdt_entry<<3));
1166 void __init trap_init(void)
1169 void __iomem *p = ioremap(0x0FFFD9, 4);
1170 if (readl(p) == 'E'+('I'<<8)+('S'<<16)+('A'<<24)) {
1176 #ifdef CONFIG_X86_LOCAL_APIC
1177 init_apic_mappings();
1180 set_trap_gate(0,÷_error);
1181 set_intr_gate(1,&debug);
1182 set_intr_gate(2,&nmi);
1183 set_system_intr_gate(3, &int3); /* int3/4 can be called from all */
1184 set_system_gate(4,&overflow);
1185 set_trap_gate(5,&bounds);
1186 set_trap_gate(6,&invalid_op);
1187 set_trap_gate(7,&device_not_available);
1188 set_task_gate(8,GDT_ENTRY_DOUBLEFAULT_TSS);
1189 set_trap_gate(9,&coprocessor_segment_overrun);
1190 set_trap_gate(10,&invalid_TSS);
1191 set_trap_gate(11,&segment_not_present);
1192 set_trap_gate(12,&stack_segment);
1193 set_trap_gate(13,&general_protection);
1194 set_intr_gate(14,&page_fault);
1195 set_trap_gate(15,&spurious_interrupt_bug);
1196 set_trap_gate(16,&coprocessor_error);
1197 set_trap_gate(17,&alignment_check);
1198 #ifdef CONFIG_X86_MCE
1199 set_trap_gate(18,&machine_check);
1201 set_trap_gate(19,&simd_coprocessor_error);
1205 * Verify that the FXSAVE/FXRSTOR data will be 16-byte aligned.
1206 * Generates a compile-time "error: zero width for bit-field" if
1207 * the alignment is wrong.
1209 struct fxsrAlignAssert {
1210 int _:!(offsetof(struct task_struct,
1211 thread.i387.fxsave) & 15);
1214 printk(KERN_INFO "Enabling fast FPU save and restore... ");
1215 set_in_cr4(X86_CR4_OSFXSR);
1219 printk(KERN_INFO "Enabling unmasked SIMD FPU exception "
1221 set_in_cr4(X86_CR4_OSXMMEXCPT);
1225 set_system_gate(SYSCALL_VECTOR,&system_call);
1228 * Should be a barrier for any external CPU state.
1235 static int __init kstack_setup(char *s)
1237 kstack_depth_to_print = simple_strtoul(s, NULL, 0);
1240 __setup("kstack=", kstack_setup);
1242 static int __init code_bytes_setup(char *s)
1244 code_bytes = simple_strtoul(s, NULL, 0);
1245 if (code_bytes > 8192)
1250 __setup("code_bytes=", code_bytes_setup);