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/sched.h>
13 #include <linux/kernel.h>
14 #include <linux/string.h>
15 #include <linux/errno.h>
16 #include <linux/timer.h>
18 #include <linux/init.h>
19 #include <linux/delay.h>
20 #include <linux/spinlock.h>
21 #include <linux/interrupt.h>
22 #include <linux/highmem.h>
23 #include <linux/kallsyms.h>
24 #include <linux/ptrace.h>
25 #include <linux/utsname.h>
26 #include <linux/kprobes.h>
27 #include <linux/kexec.h>
28 #include <linux/unwind.h>
29 #include <linux/uaccess.h>
30 #include <linux/nmi.h>
31 #include <linux/bug.h>
34 #include <linux/ioport.h>
35 #include <linux/eisa.h>
39 #include <linux/mca.h>
42 #if defined(CONFIG_EDAC)
43 #include <linux/edac.h>
46 #include <asm/processor.h>
47 #include <asm/system.h>
49 #include <asm/atomic.h>
50 #include <asm/debugreg.h>
54 #include <asm/unwind.h>
56 #include <asm/arch_hooks.h>
57 #include <linux/kdebug.h>
58 #include <asm/stacktrace.h>
60 #include <linux/module.h>
62 #include "mach_traps.h"
64 int panic_on_unrecovered_nmi;
66 asmlinkage int system_call(void);
68 /* Do we ignore FPU interrupts ? */
69 char ignore_fpu_irq = 0;
72 * The IDT has to be page-aligned to simplify the Pentium
73 * F0 0F bug workaround.. We have a special link segment
76 struct desc_struct idt_table[256] __attribute__((__section__(".data.idt"))) = { {0, 0}, };
78 asmlinkage void divide_error(void);
79 asmlinkage void debug(void);
80 asmlinkage void nmi(void);
81 asmlinkage void int3(void);
82 asmlinkage void overflow(void);
83 asmlinkage void bounds(void);
84 asmlinkage void invalid_op(void);
85 asmlinkage void device_not_available(void);
86 asmlinkage void coprocessor_segment_overrun(void);
87 asmlinkage void invalid_TSS(void);
88 asmlinkage void segment_not_present(void);
89 asmlinkage void stack_segment(void);
90 asmlinkage void general_protection(void);
91 asmlinkage void page_fault(void);
92 asmlinkage void coprocessor_error(void);
93 asmlinkage void simd_coprocessor_error(void);
94 asmlinkage void alignment_check(void);
95 asmlinkage void spurious_interrupt_bug(void);
96 asmlinkage void machine_check(void);
98 int kstack_depth_to_print = 24;
99 static unsigned int code_bytes = 64;
101 static inline int valid_stack_ptr(struct thread_info *tinfo, void *p, unsigned size)
103 return p > (void *)tinfo &&
104 p <= (void *)tinfo + THREAD_SIZE - size;
107 /* The form of the top of the frame on the stack */
109 struct stack_frame *next_frame;
110 unsigned long return_address;
113 static inline unsigned long print_context_stack(struct thread_info *tinfo,
114 unsigned long *stack, unsigned long ebp,
115 const struct stacktrace_ops *ops, void *data)
117 #ifdef CONFIG_FRAME_POINTER
118 struct stack_frame *frame = (struct stack_frame *)ebp;
119 while (valid_stack_ptr(tinfo, frame, sizeof(*frame))) {
120 struct stack_frame *next;
123 addr = frame->return_address;
124 ops->address(data, addr);
126 * break out of recursive entries (such as
127 * end_of_stack_stop_unwind_function). Also,
128 * we can never allow a frame pointer to
131 next = frame->next_frame;
137 while (valid_stack_ptr(tinfo, stack, sizeof(*stack))) {
141 if (__kernel_text_address(addr))
142 ops->address(data, addr);
148 #define MSG(msg) ops->warning(data, msg)
150 void dump_trace(struct task_struct *task, struct pt_regs *regs,
151 unsigned long *stack,
152 const struct stacktrace_ops *ops, void *data)
154 unsigned long ebp = 0;
163 stack = (unsigned long *)task->thread.esp;
166 #ifdef CONFIG_FRAME_POINTER
168 if (task == current) {
169 /* Grab ebp right from our regs */
170 asm ("movl %%ebp, %0" : "=r" (ebp) : );
172 /* ebp is the last reg pushed by switch_to */
173 ebp = *(unsigned long *) task->thread.esp;
179 struct thread_info *context;
180 context = (struct thread_info *)
181 ((unsigned long)stack & (~(THREAD_SIZE - 1)));
182 ebp = print_context_stack(context, stack, ebp, ops, data);
183 /* Should be after the line below, but somewhere
184 in early boot context comes out corrupted and we
185 can't reference it -AK */
186 if (ops->stack(data, "IRQ") < 0)
188 stack = (unsigned long*)context->previous_esp;
191 touch_nmi_watchdog();
194 EXPORT_SYMBOL(dump_trace);
197 print_trace_warning_symbol(void *data, char *msg, unsigned long symbol)
200 print_symbol(msg, symbol);
204 static void print_trace_warning(void *data, char *msg)
206 printk("%s%s\n", (char *)data, msg);
209 static int print_trace_stack(void *data, char *name)
215 * Print one address/symbol entries per line.
217 static void print_trace_address(void *data, unsigned long addr)
219 printk("%s [<%08lx>] ", (char *)data, addr);
220 print_symbol("%s\n", addr);
221 touch_nmi_watchdog();
224 static const struct stacktrace_ops print_trace_ops = {
225 .warning = print_trace_warning,
226 .warning_symbol = print_trace_warning_symbol,
227 .stack = print_trace_stack,
228 .address = print_trace_address,
232 show_trace_log_lvl(struct task_struct *task, struct pt_regs *regs,
233 unsigned long * stack, char *log_lvl)
235 dump_trace(task, regs, stack, &print_trace_ops, log_lvl);
236 printk("%s =======================\n", log_lvl);
239 void show_trace(struct task_struct *task, struct pt_regs *regs,
240 unsigned long * stack)
242 show_trace_log_lvl(task, regs, stack, "");
245 static void show_stack_log_lvl(struct task_struct *task, struct pt_regs *regs,
246 unsigned long *esp, char *log_lvl)
248 unsigned long *stack;
253 esp = (unsigned long*)task->thread.esp;
255 esp = (unsigned long *)&esp;
259 for(i = 0; i < kstack_depth_to_print; i++) {
260 if (kstack_end(stack))
262 if (i && ((i % 8) == 0))
263 printk("\n%s ", log_lvl);
264 printk("%08lx ", *stack++);
266 printk("\n%sCall Trace:\n", log_lvl);
267 show_trace_log_lvl(task, regs, esp, log_lvl);
270 void show_stack(struct task_struct *task, unsigned long *esp)
273 show_stack_log_lvl(task, NULL, esp, "");
277 * The architecture-independent dump_stack generator
279 void dump_stack(void)
283 show_trace(current, NULL, &stack);
286 EXPORT_SYMBOL(dump_stack);
288 void show_registers(struct pt_regs *regs)
293 unsigned short ss, gs;
295 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 fs: %04x gs: %04x ss: %04x\n",
317 regs->xds & 0xffff, regs->xes & 0xffff, regs->xfs & 0xffff, gs, 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, task_thread_info(current));
322 * When in-kernel, we also print out the stack and code at the
323 * time of the fault..
327 unsigned int code_prologue = code_bytes * 43 / 64;
328 unsigned int code_len = code_bytes;
331 printk("\n" KERN_EMERG "Stack: ");
332 show_stack_log_lvl(NULL, regs, (unsigned long *)esp, KERN_EMERG);
334 printk(KERN_EMERG "Code: ");
336 eip = (u8 *)regs->eip - code_prologue;
337 if (eip < (u8 *)PAGE_OFFSET ||
338 probe_kernel_address(eip, c)) {
339 /* try starting at EIP */
340 eip = (u8 *)regs->eip;
341 code_len = code_len - code_prologue + 1;
343 for (i = 0; i < code_len; i++, eip++) {
344 if (eip < (u8 *)PAGE_OFFSET ||
345 probe_kernel_address(eip, c)) {
346 printk(" Bad EIP value.");
349 if (eip == (u8 *)regs->eip)
350 printk("<%02x> ", c);
358 int is_valid_bugaddr(unsigned long eip)
362 if (eip < PAGE_OFFSET)
364 if (probe_kernel_address((unsigned short *)eip, ud2))
367 return ud2 == 0x0b0f;
371 * This is gone through when something in the kernel has done something bad and
372 * is about to be terminated.
374 void die(const char * str, struct pt_regs * regs, long err)
379 int lock_owner_depth;
381 .lock = __SPIN_LOCK_UNLOCKED(die.lock),
383 .lock_owner_depth = 0
385 static int die_counter;
390 if (die.lock_owner != raw_smp_processor_id()) {
392 spin_lock_irqsave(&die.lock, flags);
393 die.lock_owner = smp_processor_id();
394 die.lock_owner_depth = 0;
398 local_save_flags(flags);
400 if (++die.lock_owner_depth < 3) {
404 report_bug(regs->eip, regs);
406 printk(KERN_EMERG "%s: %04lx [#%d] ", str, err & 0xffff,
408 #ifdef CONFIG_PREEMPT
414 #ifdef CONFIG_DEBUG_PAGEALLOC
415 printk("DEBUG_PAGEALLOC");
419 if (notify_die(DIE_OOPS, str, regs, err,
420 current->thread.trap_no, SIGSEGV) !=
422 show_registers(regs);
423 /* Executive summary in case the oops scrolled away */
424 esp = (unsigned long) (®s->esp);
426 if (user_mode(regs)) {
428 ss = regs->xss & 0xffff;
430 printk(KERN_EMERG "EIP: [<%08lx>] ", regs->eip);
431 print_symbol("%s", regs->eip);
432 printk(" SS:ESP %04x:%08lx\n", ss, esp);
437 printk(KERN_EMERG "Recursive die() failure, output suppressed\n");
441 add_taint(TAINT_DIE);
442 spin_unlock_irqrestore(&die.lock, flags);
447 if (kexec_should_crash(current))
451 panic("Fatal exception in interrupt");
454 panic("Fatal exception");
460 static inline void die_if_kernel(const char * str, struct pt_regs * regs, long err)
462 if (!user_mode_vm(regs))
466 static void __kprobes do_trap(int trapnr, int signr, char *str, int vm86,
467 struct pt_regs * regs, long error_code,
470 struct task_struct *tsk = current;
472 if (regs->eflags & VM_MASK) {
478 if (!user_mode(regs))
483 * We want error_code and trap_no set for userspace faults and
484 * kernelspace faults which result in die(), but not
485 * kernelspace faults which are fixed up. die() gives the
486 * process no chance to handle the signal and notice the
487 * kernel fault information, so that won't result in polluting
488 * the information about previously queued, but not yet
489 * delivered, faults. See also do_general_protection below.
491 tsk->thread.error_code = error_code;
492 tsk->thread.trap_no = trapnr;
495 force_sig_info(signr, info, tsk);
497 force_sig(signr, tsk);
502 if (!fixup_exception(regs)) {
503 tsk->thread.error_code = error_code;
504 tsk->thread.trap_no = trapnr;
505 die(str, regs, error_code);
511 int ret = handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code, trapnr);
512 if (ret) goto trap_signal;
517 #define DO_ERROR(trapnr, signr, str, name) \
518 fastcall void do_##name(struct pt_regs * regs, long error_code) \
520 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
523 do_trap(trapnr, signr, str, 0, regs, error_code, NULL); \
526 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr, irq) \
527 fastcall void do_##name(struct pt_regs * regs, long error_code) \
531 local_irq_enable(); \
532 info.si_signo = signr; \
534 info.si_code = sicode; \
535 info.si_addr = (void __user *)siaddr; \
536 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
539 do_trap(trapnr, signr, str, 0, regs, error_code, &info); \
542 #define DO_VM86_ERROR(trapnr, signr, str, name) \
543 fastcall void do_##name(struct pt_regs * regs, long error_code) \
545 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
548 do_trap(trapnr, signr, str, 1, regs, error_code, NULL); \
551 #define DO_VM86_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
552 fastcall void do_##name(struct pt_regs * regs, long error_code) \
555 info.si_signo = signr; \
557 info.si_code = sicode; \
558 info.si_addr = (void __user *)siaddr; \
559 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
562 do_trap(trapnr, signr, str, 1, regs, error_code, &info); \
565 DO_VM86_ERROR_INFO( 0, SIGFPE, "divide error", divide_error, FPE_INTDIV, regs->eip)
566 #ifndef CONFIG_KPROBES
567 DO_VM86_ERROR( 3, SIGTRAP, "int3", int3)
569 DO_VM86_ERROR( 4, SIGSEGV, "overflow", overflow)
570 DO_VM86_ERROR( 5, SIGSEGV, "bounds", bounds)
571 DO_ERROR_INFO( 6, SIGILL, "invalid opcode", invalid_op, ILL_ILLOPN, regs->eip, 0)
572 DO_ERROR( 9, SIGFPE, "coprocessor segment overrun", coprocessor_segment_overrun)
573 DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
574 DO_ERROR(11, SIGBUS, "segment not present", segment_not_present)
575 DO_ERROR(12, SIGBUS, "stack segment", stack_segment)
576 DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0, 0)
577 DO_ERROR_INFO(32, SIGSEGV, "iret exception", iret_error, ILL_BADSTK, 0, 1)
579 fastcall void __kprobes do_general_protection(struct pt_regs * regs,
583 struct tss_struct *tss = &per_cpu(init_tss, cpu);
584 struct thread_struct *thread = ¤t->thread;
587 * Perform the lazy TSS's I/O bitmap copy. If the TSS has an
588 * invalid offset set (the LAZY one) and the faulting thread has
589 * a valid I/O bitmap pointer, we copy the I/O bitmap in the TSS
590 * and we set the offset field correctly. Then we let the CPU to
591 * restart the faulting instruction.
593 if (tss->x86_tss.io_bitmap_base == INVALID_IO_BITMAP_OFFSET_LAZY &&
594 thread->io_bitmap_ptr) {
595 memcpy(tss->io_bitmap, thread->io_bitmap_ptr,
596 thread->io_bitmap_max);
598 * If the previously set map was extending to higher ports
599 * than the current one, pad extra space with 0xff (no access).
601 if (thread->io_bitmap_max < tss->io_bitmap_max)
602 memset((char *) tss->io_bitmap +
603 thread->io_bitmap_max, 0xff,
604 tss->io_bitmap_max - thread->io_bitmap_max);
605 tss->io_bitmap_max = thread->io_bitmap_max;
606 tss->x86_tss.io_bitmap_base = IO_BITMAP_OFFSET;
607 tss->io_bitmap_owner = thread;
613 if (regs->eflags & VM_MASK)
616 if (!user_mode(regs))
619 current->thread.error_code = error_code;
620 current->thread.trap_no = 13;
621 if (show_unhandled_signals && unhandled_signal(current, SIGSEGV) &&
624 "%s[%d] general protection eip:%lx esp:%lx error:%lx\n",
625 current->comm, current->pid,
626 regs->eip, regs->esp, error_code);
628 force_sig(SIGSEGV, current);
633 handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
637 if (!fixup_exception(regs)) {
638 current->thread.error_code = error_code;
639 current->thread.trap_no = 13;
640 if (notify_die(DIE_GPF, "general protection fault", regs,
641 error_code, 13, SIGSEGV) == NOTIFY_STOP)
643 die("general protection fault", regs, error_code);
647 static __kprobes void
648 mem_parity_error(unsigned char reason, struct pt_regs * regs)
650 printk(KERN_EMERG "Uhhuh. NMI received for unknown reason %02x on "
651 "CPU %d.\n", reason, smp_processor_id());
652 printk(KERN_EMERG "You have some hardware problem, likely on the PCI bus.\n");
654 #if defined(CONFIG_EDAC)
655 if(edac_handler_set()) {
656 edac_atomic_assert_error();
661 if (panic_on_unrecovered_nmi)
662 panic("NMI: Not continuing");
664 printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
666 /* Clear and disable the memory parity error line. */
667 clear_mem_error(reason);
670 static __kprobes void
671 io_check_error(unsigned char reason, struct pt_regs * regs)
675 printk(KERN_EMERG "NMI: IOCK error (debug interrupt?)\n");
676 show_registers(regs);
678 /* Re-enable the IOCK line, wait for a few seconds */
679 reason = (reason & 0xf) | 8;
682 while (--i) udelay(1000);
687 static __kprobes void
688 unknown_nmi_error(unsigned char reason, struct pt_regs * regs)
691 /* Might actually be able to figure out what the guilty party
698 printk(KERN_EMERG "Uhhuh. NMI received for unknown reason %02x on "
699 "CPU %d.\n", reason, smp_processor_id());
700 printk(KERN_EMERG "Do you have a strange power saving mode enabled?\n");
701 if (panic_on_unrecovered_nmi)
702 panic("NMI: Not continuing");
704 printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
707 static DEFINE_SPINLOCK(nmi_print_lock);
709 void __kprobes die_nmi(struct pt_regs *regs, const char *msg)
711 if (notify_die(DIE_NMIWATCHDOG, msg, regs, 0, 2, SIGINT) ==
715 spin_lock(&nmi_print_lock);
717 * We are in trouble anyway, lets at least try
718 * to get a message out.
721 printk(KERN_EMERG "%s", msg);
722 printk(" on CPU%d, eip %08lx, registers:\n",
723 smp_processor_id(), regs->eip);
724 show_registers(regs);
726 spin_unlock(&nmi_print_lock);
729 /* If we are in kernel we are probably nested up pretty bad
730 * and might aswell get out now while we still can.
732 if (!user_mode_vm(regs)) {
733 current->thread.trap_no = 2;
740 static __kprobes void default_do_nmi(struct pt_regs * regs)
742 unsigned char reason = 0;
744 /* Only the BSP gets external NMIs from the system. */
745 if (!smp_processor_id())
746 reason = get_nmi_reason();
748 if (!(reason & 0xc0)) {
749 if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 2, SIGINT)
752 #ifdef CONFIG_X86_LOCAL_APIC
754 * Ok, so this is none of the documented NMI sources,
755 * so it must be the NMI watchdog.
757 if (nmi_watchdog_tick(regs, reason))
759 if (!do_nmi_callback(regs, smp_processor_id()))
761 unknown_nmi_error(reason, regs);
765 if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP)
768 mem_parity_error(reason, regs);
770 io_check_error(reason, regs);
772 * Reassert NMI in case it became active meanwhile
773 * as it's edge-triggered.
778 static int ignore_nmis;
780 fastcall __kprobes void do_nmi(struct pt_regs * regs, long error_code)
786 cpu = smp_processor_id();
791 default_do_nmi(regs);
802 void restart_nmi(void)
808 #ifdef CONFIG_KPROBES
809 fastcall void __kprobes do_int3(struct pt_regs *regs, long error_code)
811 if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP)
814 /* This is an interrupt gate, because kprobes wants interrupts
815 disabled. Normal trap handlers don't. */
816 restore_interrupts(regs);
817 do_trap(3, SIGTRAP, "int3", 1, regs, error_code, NULL);
822 * Our handling of the processor debug registers is non-trivial.
823 * We do not clear them on entry and exit from the kernel. Therefore
824 * it is possible to get a watchpoint trap here from inside the kernel.
825 * However, the code in ./ptrace.c has ensured that the user can
826 * only set watchpoints on userspace addresses. Therefore the in-kernel
827 * watchpoint trap can only occur in code which is reading/writing
828 * from user space. Such code must not hold kernel locks (since it
829 * can equally take a page fault), therefore it is safe to call
830 * force_sig_info even though that claims and releases locks.
832 * Code in ./signal.c ensures that the debug control register
833 * is restored before we deliver any signal, and therefore that
834 * user code runs with the correct debug control register even though
837 * Being careful here means that we don't have to be as careful in a
838 * lot of more complicated places (task switching can be a bit lazy
839 * about restoring all the debug state, and ptrace doesn't have to
840 * find every occurrence of the TF bit that could be saved away even
843 fastcall void __kprobes do_debug(struct pt_regs * regs, long error_code)
845 unsigned int condition;
846 struct task_struct *tsk = current;
848 get_debugreg(condition, 6);
850 if (notify_die(DIE_DEBUG, "debug", regs, condition, error_code,
851 SIGTRAP) == NOTIFY_STOP)
853 /* It's safe to allow irq's after DR6 has been saved */
854 if (regs->eflags & X86_EFLAGS_IF)
857 /* Mask out spurious debug traps due to lazy DR7 setting */
858 if (condition & (DR_TRAP0|DR_TRAP1|DR_TRAP2|DR_TRAP3)) {
859 if (!tsk->thread.debugreg[7])
863 if (regs->eflags & VM_MASK)
866 /* Save debug status register where ptrace can see it */
867 tsk->thread.debugreg[6] = condition;
870 * Single-stepping through TF: make sure we ignore any events in
871 * kernel space (but re-enable TF when returning to user mode).
873 if (condition & DR_STEP) {
875 * We already checked v86 mode above, so we can
876 * check for kernel mode by just checking the CPL
879 if (!user_mode(regs))
880 goto clear_TF_reenable;
883 /* Ok, finally something we can handle */
884 send_sigtrap(tsk, regs, error_code);
886 /* Disable additional traps. They'll be re-enabled when
887 * the signal is delivered.
894 handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code, 1);
898 set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
899 regs->eflags &= ~TF_MASK;
904 * Note that we play around with the 'TS' bit in an attempt to get
905 * the correct behaviour even in the presence of the asynchronous
908 void math_error(void __user *eip)
910 struct task_struct * task;
912 unsigned short cwd, swd;
915 * Save the info for the exception handler and clear the error.
919 task->thread.trap_no = 16;
920 task->thread.error_code = 0;
921 info.si_signo = SIGFPE;
923 info.si_code = __SI_FAULT;
926 * (~cwd & swd) will mask out exceptions that are not set to unmasked
927 * status. 0x3f is the exception bits in these regs, 0x200 is the
928 * C1 reg you need in case of a stack fault, 0x040 is the stack
929 * fault bit. We should only be taking one exception at a time,
930 * so if this combination doesn't produce any single exception,
931 * then we have a bad program that isn't syncronizing its FPU usage
932 * and it will suffer the consequences since we won't be able to
933 * fully reproduce the context of the exception
935 cwd = get_fpu_cwd(task);
936 swd = get_fpu_swd(task);
937 switch (swd & ~cwd & 0x3f) {
938 case 0x000: /* No unmasked exception */
940 default: /* Multiple exceptions */
942 case 0x001: /* Invalid Op */
944 * swd & 0x240 == 0x040: Stack Underflow
945 * swd & 0x240 == 0x240: Stack Overflow
946 * User must clear the SF bit (0x40) if set
948 info.si_code = FPE_FLTINV;
950 case 0x002: /* Denormalize */
951 case 0x010: /* Underflow */
952 info.si_code = FPE_FLTUND;
954 case 0x004: /* Zero Divide */
955 info.si_code = FPE_FLTDIV;
957 case 0x008: /* Overflow */
958 info.si_code = FPE_FLTOVF;
960 case 0x020: /* Precision */
961 info.si_code = FPE_FLTRES;
964 force_sig_info(SIGFPE, &info, task);
967 fastcall void do_coprocessor_error(struct pt_regs * regs, long error_code)
970 math_error((void __user *)regs->eip);
973 static void simd_math_error(void __user *eip)
975 struct task_struct * task;
977 unsigned short mxcsr;
980 * Save the info for the exception handler and clear the error.
984 task->thread.trap_no = 19;
985 task->thread.error_code = 0;
986 info.si_signo = SIGFPE;
988 info.si_code = __SI_FAULT;
991 * The SIMD FPU exceptions are handled a little differently, as there
992 * is only a single status/control register. Thus, to determine which
993 * unmasked exception was caught we must mask the exception mask bits
994 * at 0x1f80, and then use these to mask the exception bits at 0x3f.
996 mxcsr = get_fpu_mxcsr(task);
997 switch (~((mxcsr & 0x1f80) >> 7) & (mxcsr & 0x3f)) {
1001 case 0x001: /* Invalid Op */
1002 info.si_code = FPE_FLTINV;
1004 case 0x002: /* Denormalize */
1005 case 0x010: /* Underflow */
1006 info.si_code = FPE_FLTUND;
1008 case 0x004: /* Zero Divide */
1009 info.si_code = FPE_FLTDIV;
1011 case 0x008: /* Overflow */
1012 info.si_code = FPE_FLTOVF;
1014 case 0x020: /* Precision */
1015 info.si_code = FPE_FLTRES;
1018 force_sig_info(SIGFPE, &info, task);
1021 fastcall void do_simd_coprocessor_error(struct pt_regs * regs,
1025 /* Handle SIMD FPU exceptions on PIII+ processors. */
1027 simd_math_error((void __user *)regs->eip);
1030 * Handle strange cache flush from user space exception
1031 * in all other cases. This is undocumented behaviour.
1033 if (regs->eflags & VM_MASK) {
1034 handle_vm86_fault((struct kernel_vm86_regs *)regs,
1038 current->thread.trap_no = 19;
1039 current->thread.error_code = error_code;
1040 die_if_kernel("cache flush denied", regs, error_code);
1041 force_sig(SIGSEGV, current);
1045 fastcall void do_spurious_interrupt_bug(struct pt_regs * regs,
1049 /* No need to warn about this any longer. */
1050 printk("Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
1054 fastcall unsigned long patch_espfix_desc(unsigned long uesp,
1057 struct desc_struct *gdt = __get_cpu_var(gdt_page).gdt;
1058 unsigned long base = (kesp - uesp) & -THREAD_SIZE;
1059 unsigned long new_kesp = kesp - base;
1060 unsigned long lim_pages = (new_kesp | (THREAD_SIZE - 1)) >> PAGE_SHIFT;
1061 __u64 desc = *(__u64 *)&gdt[GDT_ENTRY_ESPFIX_SS];
1062 /* Set up base for espfix segment */
1063 desc &= 0x00f0ff0000000000ULL;
1064 desc |= ((((__u64)base) << 16) & 0x000000ffffff0000ULL) |
1065 ((((__u64)base) << 32) & 0xff00000000000000ULL) |
1066 ((((__u64)lim_pages) << 32) & 0x000f000000000000ULL) |
1067 (lim_pages & 0xffff);
1068 *(__u64 *)&gdt[GDT_ENTRY_ESPFIX_SS] = desc;
1073 * 'math_state_restore()' saves the current math information in the
1074 * old math state array, and gets the new ones from the current task
1076 * Careful.. There are problems with IBM-designed IRQ13 behaviour.
1077 * Don't touch unless you *really* know how it works.
1079 * Must be called with kernel preemption disabled (in this case,
1080 * local interrupts are disabled at the call-site in entry.S).
1082 asmlinkage void math_state_restore(void)
1084 struct thread_info *thread = current_thread_info();
1085 struct task_struct *tsk = thread->task;
1087 clts(); /* Allow maths ops (or we recurse) */
1088 if (!tsk_used_math(tsk))
1091 thread->status |= TS_USEDFPU; /* So we fnsave on switch_to() */
1094 EXPORT_SYMBOL_GPL(math_state_restore);
1096 #ifndef CONFIG_MATH_EMULATION
1098 asmlinkage void math_emulate(long arg)
1100 printk(KERN_EMERG "math-emulation not enabled and no coprocessor found.\n");
1101 printk(KERN_EMERG "killing %s.\n",current->comm);
1102 force_sig(SIGFPE,current);
1106 #endif /* CONFIG_MATH_EMULATION */
1109 * This needs to use 'idt_table' rather than 'idt', and
1110 * thus use the _nonmapped_ version of the IDT, as the
1111 * Pentium F0 0F bugfix can have resulted in the mapped
1112 * IDT being write-protected.
1114 void set_intr_gate(unsigned int n, void *addr)
1116 _set_gate(n, DESCTYPE_INT, addr, __KERNEL_CS);
1120 * This routine sets up an interrupt gate at directory privilege level 3.
1122 static inline void set_system_intr_gate(unsigned int n, void *addr)
1124 _set_gate(n, DESCTYPE_INT | DESCTYPE_DPL3, addr, __KERNEL_CS);
1127 static void __init set_trap_gate(unsigned int n, void *addr)
1129 _set_gate(n, DESCTYPE_TRAP, addr, __KERNEL_CS);
1132 static void __init set_system_gate(unsigned int n, void *addr)
1134 _set_gate(n, DESCTYPE_TRAP | DESCTYPE_DPL3, addr, __KERNEL_CS);
1137 static void __init set_task_gate(unsigned int n, unsigned int gdt_entry)
1139 _set_gate(n, DESCTYPE_TASK, (void *)0, (gdt_entry<<3));
1143 void __init trap_init(void)
1146 void __iomem *p = ioremap(0x0FFFD9, 4);
1147 if (readl(p) == 'E'+('I'<<8)+('S'<<16)+('A'<<24)) {
1153 #ifdef CONFIG_X86_LOCAL_APIC
1154 init_apic_mappings();
1157 set_trap_gate(0,÷_error);
1158 set_intr_gate(1,&debug);
1159 set_intr_gate(2,&nmi);
1160 set_system_intr_gate(3, &int3); /* int3/4 can be called from all */
1161 set_system_gate(4,&overflow);
1162 set_trap_gate(5,&bounds);
1163 set_trap_gate(6,&invalid_op);
1164 set_trap_gate(7,&device_not_available);
1165 set_task_gate(8,GDT_ENTRY_DOUBLEFAULT_TSS);
1166 set_trap_gate(9,&coprocessor_segment_overrun);
1167 set_trap_gate(10,&invalid_TSS);
1168 set_trap_gate(11,&segment_not_present);
1169 set_trap_gate(12,&stack_segment);
1170 set_trap_gate(13,&general_protection);
1171 set_intr_gate(14,&page_fault);
1172 set_trap_gate(15,&spurious_interrupt_bug);
1173 set_trap_gate(16,&coprocessor_error);
1174 set_trap_gate(17,&alignment_check);
1175 #ifdef CONFIG_X86_MCE
1176 set_trap_gate(18,&machine_check);
1178 set_trap_gate(19,&simd_coprocessor_error);
1182 * Verify that the FXSAVE/FXRSTOR data will be 16-byte aligned.
1183 * Generates a compile-time "error: zero width for bit-field" if
1184 * the alignment is wrong.
1186 struct fxsrAlignAssert {
1187 int _:!(offsetof(struct task_struct,
1188 thread.i387.fxsave) & 15);
1191 printk(KERN_INFO "Enabling fast FPU save and restore... ");
1192 set_in_cr4(X86_CR4_OSFXSR);
1196 printk(KERN_INFO "Enabling unmasked SIMD FPU exception "
1198 set_in_cr4(X86_CR4_OSXMMEXCPT);
1202 set_system_gate(SYSCALL_VECTOR,&system_call);
1205 * Should be a barrier for any external CPU state.
1212 static int __init kstack_setup(char *s)
1214 kstack_depth_to_print = simple_strtoul(s, NULL, 0);
1217 __setup("kstack=", kstack_setup);
1219 static int __init code_bytes_setup(char *s)
1221 code_bytes = simple_strtoul(s, NULL, 0);
1222 if (code_bytes > 8192)
1227 __setup("code_bytes=", code_bytes_setup);