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 DECLARE_BITMAP(used_vectors, NR_VECTORS);
67 EXPORT_SYMBOL_GPL(used_vectors);
69 asmlinkage int system_call(void);
71 /* Do we ignore FPU interrupts ? */
72 char ignore_fpu_irq = 0;
75 * The IDT has to be page-aligned to simplify the Pentium
76 * F0 0F bug workaround.. We have a special link segment
79 gate_desc idt_table[256]
80 __attribute__((__section__(".data.idt"))) = { { { { 0, 0 } } }, };
82 asmlinkage void divide_error(void);
83 asmlinkage void debug(void);
84 asmlinkage void nmi(void);
85 asmlinkage void int3(void);
86 asmlinkage void overflow(void);
87 asmlinkage void bounds(void);
88 asmlinkage void invalid_op(void);
89 asmlinkage void device_not_available(void);
90 asmlinkage void coprocessor_segment_overrun(void);
91 asmlinkage void invalid_TSS(void);
92 asmlinkage void segment_not_present(void);
93 asmlinkage void stack_segment(void);
94 asmlinkage void general_protection(void);
95 asmlinkage void page_fault(void);
96 asmlinkage void coprocessor_error(void);
97 asmlinkage void simd_coprocessor_error(void);
98 asmlinkage void alignment_check(void);
99 asmlinkage void spurious_interrupt_bug(void);
100 asmlinkage void machine_check(void);
102 int kstack_depth_to_print = 24;
103 static unsigned int code_bytes = 64;
105 static inline int valid_stack_ptr(struct thread_info *tinfo, void *p, unsigned size)
107 return p > (void *)tinfo &&
108 p <= (void *)tinfo + THREAD_SIZE - size;
111 /* The form of the top of the frame on the stack */
113 struct stack_frame *next_frame;
114 unsigned long return_address;
117 static inline unsigned long print_context_stack(struct thread_info *tinfo,
118 unsigned long *stack, unsigned long bp,
119 const struct stacktrace_ops *ops, void *data)
121 #ifdef CONFIG_FRAME_POINTER
122 struct stack_frame *frame = (struct stack_frame *)bp;
123 while (valid_stack_ptr(tinfo, frame, sizeof(*frame))) {
124 struct stack_frame *next;
127 addr = frame->return_address;
128 ops->address(data, addr);
130 * break out of recursive entries (such as
131 * end_of_stack_stop_unwind_function). Also,
132 * we can never allow a frame pointer to
135 next = frame->next_frame;
141 while (valid_stack_ptr(tinfo, stack, sizeof(*stack))) {
145 if (__kernel_text_address(addr))
146 ops->address(data, addr);
152 #define MSG(msg) ops->warning(data, msg)
154 void dump_trace(struct task_struct *task, struct pt_regs *regs,
155 unsigned long *stack,
156 const struct stacktrace_ops *ops, void *data)
158 unsigned long bp = 0;
167 stack = (unsigned long *)task->thread.sp;
170 #ifdef CONFIG_FRAME_POINTER
172 if (task == current) {
173 /* Grab bp right from our regs */
174 asm ("movl %%ebp, %0" : "=r" (bp) : );
176 /* bp is the last reg pushed by switch_to */
177 bp = *(unsigned long *) task->thread.sp;
183 struct thread_info *context;
184 context = (struct thread_info *)
185 ((unsigned long)stack & (~(THREAD_SIZE - 1)));
186 bp = print_context_stack(context, stack, bp, ops, data);
187 /* Should be after the line below, but somewhere
188 in early boot context comes out corrupted and we
189 can't reference it -AK */
190 if (ops->stack(data, "IRQ") < 0)
192 stack = (unsigned long*)context->previous_esp;
195 touch_nmi_watchdog();
198 EXPORT_SYMBOL(dump_trace);
201 print_trace_warning_symbol(void *data, char *msg, unsigned long symbol)
204 print_symbol(msg, symbol);
208 static void print_trace_warning(void *data, char *msg)
210 printk("%s%s\n", (char *)data, msg);
213 static int print_trace_stack(void *data, char *name)
219 * Print one address/symbol entries per line.
221 static void print_trace_address(void *data, unsigned long addr)
223 printk("%s [<%08lx>] ", (char *)data, addr);
224 print_symbol("%s\n", addr);
225 touch_nmi_watchdog();
228 static const struct stacktrace_ops print_trace_ops = {
229 .warning = print_trace_warning,
230 .warning_symbol = print_trace_warning_symbol,
231 .stack = print_trace_stack,
232 .address = print_trace_address,
236 show_trace_log_lvl(struct task_struct *task, struct pt_regs *regs,
237 unsigned long * stack, char *log_lvl)
239 dump_trace(task, regs, stack, &print_trace_ops, log_lvl);
240 printk("%s =======================\n", log_lvl);
243 void show_trace(struct task_struct *task, struct pt_regs *regs,
244 unsigned long * stack)
246 show_trace_log_lvl(task, regs, stack, "");
249 static void show_stack_log_lvl(struct task_struct *task, struct pt_regs *regs,
250 unsigned long *sp, char *log_lvl)
252 unsigned long *stack;
257 sp = (unsigned long*)task->thread.sp;
259 sp = (unsigned long *)&sp;
263 for(i = 0; i < kstack_depth_to_print; i++) {
264 if (kstack_end(stack))
266 if (i && ((i % 8) == 0))
267 printk("\n%s ", log_lvl);
268 printk("%08lx ", *stack++);
270 printk("\n%sCall Trace:\n", log_lvl);
271 show_trace_log_lvl(task, regs, sp, log_lvl);
274 void show_stack(struct task_struct *task, unsigned long *sp)
277 show_stack_log_lvl(task, NULL, sp, "");
281 * The architecture-independent dump_stack generator
283 void dump_stack(void)
287 printk("Pid: %d, comm: %.20s %s %s %.*s\n",
288 current->pid, current->comm, print_tainted(),
289 init_utsname()->release,
290 (int)strcspn(init_utsname()->version, " "),
291 init_utsname()->version);
292 show_trace(current, NULL, &stack);
295 EXPORT_SYMBOL(dump_stack);
297 void show_registers(struct pt_regs *regs)
302 __show_registers(regs, 0);
303 printk(KERN_EMERG "Process %.*s (pid: %d, ti=%p task=%p task.ti=%p)",
304 TASK_COMM_LEN, current->comm, task_pid_nr(current),
305 current_thread_info(), current, task_thread_info(current));
307 * When in-kernel, we also print out the stack and code at the
308 * time of the fault..
310 if (!user_mode_vm(regs)) {
312 unsigned int code_prologue = code_bytes * 43 / 64;
313 unsigned int code_len = code_bytes;
316 printk("\n" KERN_EMERG "Stack: ");
317 show_stack_log_lvl(NULL, regs, ®s->sp, KERN_EMERG);
319 printk(KERN_EMERG "Code: ");
321 ip = (u8 *)regs->ip - code_prologue;
322 if (ip < (u8 *)PAGE_OFFSET ||
323 probe_kernel_address(ip, c)) {
324 /* try starting at EIP */
326 code_len = code_len - code_prologue + 1;
328 for (i = 0; i < code_len; i++, ip++) {
329 if (ip < (u8 *)PAGE_OFFSET ||
330 probe_kernel_address(ip, c)) {
331 printk(" Bad EIP value.");
334 if (ip == (u8 *)regs->ip)
335 printk("<%02x> ", c);
343 int is_valid_bugaddr(unsigned long ip)
347 if (ip < PAGE_OFFSET)
349 if (probe_kernel_address((unsigned short *)ip, ud2))
352 return ud2 == 0x0b0f;
356 * This is gone through when something in the kernel has done something bad and
357 * is about to be terminated.
359 void die(const char * str, struct pt_regs * regs, long err)
364 int lock_owner_depth;
366 .lock = __RAW_SPIN_LOCK_UNLOCKED,
368 .lock_owner_depth = 0
370 static int die_counter;
375 if (die.lock_owner != raw_smp_processor_id()) {
377 raw_local_irq_save(flags);
378 __raw_spin_lock(&die.lock);
379 die.lock_owner = smp_processor_id();
380 die.lock_owner_depth = 0;
383 raw_local_irq_save(flags);
385 if (++die.lock_owner_depth < 3) {
389 report_bug(regs->ip, regs);
391 printk(KERN_EMERG "%s: %04lx [#%d] ", str, err & 0xffff,
393 #ifdef CONFIG_PREEMPT
399 #ifdef CONFIG_DEBUG_PAGEALLOC
400 printk("DEBUG_PAGEALLOC");
404 if (notify_die(DIE_OOPS, str, regs, err,
405 current->thread.trap_no, SIGSEGV) !=
407 show_registers(regs);
408 /* Executive summary in case the oops scrolled away */
409 sp = (unsigned long) (®s->sp);
411 if (user_mode(regs)) {
413 ss = regs->ss & 0xffff;
415 printk(KERN_EMERG "EIP: [<%08lx>] ", regs->ip);
416 print_symbol("%s", regs->ip);
417 printk(" SS:ESP %04x:%08lx\n", ss, sp);
422 printk(KERN_EMERG "Recursive die() failure, output suppressed\n");
426 add_taint(TAINT_DIE);
427 __raw_spin_unlock(&die.lock);
428 raw_local_irq_restore(flags);
433 if (kexec_should_crash(current))
437 panic("Fatal exception in interrupt");
440 panic("Fatal exception");
446 static inline void die_if_kernel(const char * str, struct pt_regs * regs, long err)
448 if (!user_mode_vm(regs))
452 static void __kprobes do_trap(int trapnr, int signr, char *str, int vm86,
453 struct pt_regs * regs, long error_code,
456 struct task_struct *tsk = current;
458 if (regs->flags & VM_MASK) {
464 if (!user_mode(regs))
469 * We want error_code and trap_no set for userspace faults and
470 * kernelspace faults which result in die(), but not
471 * kernelspace faults which are fixed up. die() gives the
472 * process no chance to handle the signal and notice the
473 * kernel fault information, so that won't result in polluting
474 * the information about previously queued, but not yet
475 * delivered, faults. See also do_general_protection below.
477 tsk->thread.error_code = error_code;
478 tsk->thread.trap_no = trapnr;
481 force_sig_info(signr, info, tsk);
483 force_sig(signr, tsk);
488 if (!fixup_exception(regs)) {
489 tsk->thread.error_code = error_code;
490 tsk->thread.trap_no = trapnr;
491 die(str, regs, error_code);
497 int ret = handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code, trapnr);
498 if (ret) goto trap_signal;
503 #define DO_ERROR(trapnr, signr, str, name) \
504 void do_##name(struct pt_regs * regs, long error_code) \
506 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
509 do_trap(trapnr, signr, str, 0, regs, error_code, NULL); \
512 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr, irq) \
513 void do_##name(struct pt_regs * regs, long error_code) \
517 local_irq_enable(); \
518 info.si_signo = signr; \
520 info.si_code = sicode; \
521 info.si_addr = (void __user *)siaddr; \
522 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
525 do_trap(trapnr, signr, str, 0, regs, error_code, &info); \
528 #define DO_VM86_ERROR(trapnr, signr, str, name) \
529 void do_##name(struct pt_regs * regs, long error_code) \
531 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
534 do_trap(trapnr, signr, str, 1, regs, error_code, NULL); \
537 #define DO_VM86_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
538 void do_##name(struct pt_regs * regs, long error_code) \
541 info.si_signo = signr; \
543 info.si_code = sicode; \
544 info.si_addr = (void __user *)siaddr; \
545 trace_hardirqs_fixup(); \
546 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
549 do_trap(trapnr, signr, str, 1, regs, error_code, &info); \
552 DO_VM86_ERROR_INFO( 0, SIGFPE, "divide error", divide_error, FPE_INTDIV, regs->ip)
553 #ifndef CONFIG_KPROBES
554 DO_VM86_ERROR( 3, SIGTRAP, "int3", int3)
556 DO_VM86_ERROR( 4, SIGSEGV, "overflow", overflow)
557 DO_VM86_ERROR( 5, SIGSEGV, "bounds", bounds)
558 DO_ERROR_INFO( 6, SIGILL, "invalid opcode", invalid_op, ILL_ILLOPN, regs->ip, 0)
559 DO_ERROR( 9, SIGFPE, "coprocessor segment overrun", coprocessor_segment_overrun)
560 DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
561 DO_ERROR(11, SIGBUS, "segment not present", segment_not_present)
562 DO_ERROR(12, SIGBUS, "stack segment", stack_segment)
563 DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0, 0)
564 DO_ERROR_INFO(32, SIGSEGV, "iret exception", iret_error, ILL_BADSTK, 0, 1)
566 void __kprobes do_general_protection(struct pt_regs * regs,
570 struct tss_struct *tss = &per_cpu(init_tss, cpu);
571 struct thread_struct *thread = ¤t->thread;
574 * Perform the lazy TSS's I/O bitmap copy. If the TSS has an
575 * invalid offset set (the LAZY one) and the faulting thread has
576 * a valid I/O bitmap pointer, we copy the I/O bitmap in the TSS
577 * and we set the offset field correctly. Then we let the CPU to
578 * restart the faulting instruction.
580 if (tss->x86_tss.io_bitmap_base == INVALID_IO_BITMAP_OFFSET_LAZY &&
581 thread->io_bitmap_ptr) {
582 memcpy(tss->io_bitmap, thread->io_bitmap_ptr,
583 thread->io_bitmap_max);
585 * If the previously set map was extending to higher ports
586 * than the current one, pad extra space with 0xff (no access).
588 if (thread->io_bitmap_max < tss->io_bitmap_max)
589 memset((char *) tss->io_bitmap +
590 thread->io_bitmap_max, 0xff,
591 tss->io_bitmap_max - thread->io_bitmap_max);
592 tss->io_bitmap_max = thread->io_bitmap_max;
593 tss->x86_tss.io_bitmap_base = IO_BITMAP_OFFSET;
594 tss->io_bitmap_owner = thread;
600 if (regs->flags & VM_MASK)
603 if (!user_mode(regs))
606 current->thread.error_code = error_code;
607 current->thread.trap_no = 13;
608 if (show_unhandled_signals && unhandled_signal(current, SIGSEGV) &&
611 "%s[%d] general protection ip:%lx sp:%lx error:%lx\n",
612 current->comm, task_pid_nr(current),
613 regs->ip, regs->sp, error_code);
615 force_sig(SIGSEGV, current);
620 handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
624 if (!fixup_exception(regs)) {
625 current->thread.error_code = error_code;
626 current->thread.trap_no = 13;
627 if (notify_die(DIE_GPF, "general protection fault", regs,
628 error_code, 13, SIGSEGV) == NOTIFY_STOP)
630 die("general protection fault", regs, error_code);
634 static __kprobes void
635 mem_parity_error(unsigned char reason, struct pt_regs * regs)
637 printk(KERN_EMERG "Uhhuh. NMI received for unknown reason %02x on "
638 "CPU %d.\n", reason, smp_processor_id());
639 printk(KERN_EMERG "You have some hardware problem, likely on the PCI bus.\n");
641 #if defined(CONFIG_EDAC)
642 if(edac_handler_set()) {
643 edac_atomic_assert_error();
648 if (panic_on_unrecovered_nmi)
649 panic("NMI: Not continuing");
651 printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
653 /* Clear and disable the memory parity error line. */
654 clear_mem_error(reason);
657 static __kprobes void
658 io_check_error(unsigned char reason, struct pt_regs * regs)
662 printk(KERN_EMERG "NMI: IOCK error (debug interrupt?)\n");
663 show_registers(regs);
665 /* Re-enable the IOCK line, wait for a few seconds */
666 reason = (reason & 0xf) | 8;
669 while (--i) udelay(1000);
674 static __kprobes void
675 unknown_nmi_error(unsigned char reason, struct pt_regs * regs)
678 /* Might actually be able to figure out what the guilty party
685 printk(KERN_EMERG "Uhhuh. NMI received for unknown reason %02x on "
686 "CPU %d.\n", reason, smp_processor_id());
687 printk(KERN_EMERG "Do you have a strange power saving mode enabled?\n");
688 if (panic_on_unrecovered_nmi)
689 panic("NMI: Not continuing");
691 printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
694 static DEFINE_SPINLOCK(nmi_print_lock);
696 void __kprobes die_nmi(struct pt_regs *regs, const char *msg)
698 if (notify_die(DIE_NMIWATCHDOG, msg, regs, 0, 2, SIGINT) ==
702 spin_lock(&nmi_print_lock);
704 * We are in trouble anyway, lets at least try
705 * to get a message out.
708 printk(KERN_EMERG "%s", msg);
709 printk(" on CPU%d, ip %08lx, registers:\n",
710 smp_processor_id(), regs->ip);
711 show_registers(regs);
713 spin_unlock(&nmi_print_lock);
716 /* If we are in kernel we are probably nested up pretty bad
717 * and might aswell get out now while we still can.
719 if (!user_mode_vm(regs)) {
720 current->thread.trap_no = 2;
727 static __kprobes void default_do_nmi(struct pt_regs * regs)
729 unsigned char reason = 0;
731 /* Only the BSP gets external NMIs from the system. */
732 if (!smp_processor_id())
733 reason = get_nmi_reason();
735 if (!(reason & 0xc0)) {
736 if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 2, SIGINT)
739 #ifdef CONFIG_X86_LOCAL_APIC
741 * Ok, so this is none of the documented NMI sources,
742 * so it must be the NMI watchdog.
744 if (nmi_watchdog_tick(regs, reason))
746 if (!do_nmi_callback(regs, smp_processor_id()))
748 unknown_nmi_error(reason, regs);
752 if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP)
755 mem_parity_error(reason, regs);
757 io_check_error(reason, regs);
759 * Reassert NMI in case it became active meanwhile
760 * as it's edge-triggered.
765 static int ignore_nmis;
767 __kprobes void do_nmi(struct pt_regs * regs, long error_code)
773 cpu = smp_processor_id();
778 default_do_nmi(regs);
789 void restart_nmi(void)
795 #ifdef CONFIG_KPROBES
796 void __kprobes do_int3(struct pt_regs *regs, long error_code)
798 trace_hardirqs_fixup();
800 if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP)
803 /* This is an interrupt gate, because kprobes wants interrupts
804 disabled. Normal trap handlers don't. */
805 restore_interrupts(regs);
806 do_trap(3, SIGTRAP, "int3", 1, regs, error_code, NULL);
811 * Our handling of the processor debug registers is non-trivial.
812 * We do not clear them on entry and exit from the kernel. Therefore
813 * it is possible to get a watchpoint trap here from inside the kernel.
814 * However, the code in ./ptrace.c has ensured that the user can
815 * only set watchpoints on userspace addresses. Therefore the in-kernel
816 * watchpoint trap can only occur in code which is reading/writing
817 * from user space. Such code must not hold kernel locks (since it
818 * can equally take a page fault), therefore it is safe to call
819 * force_sig_info even though that claims and releases locks.
821 * Code in ./signal.c ensures that the debug control register
822 * is restored before we deliver any signal, and therefore that
823 * user code runs with the correct debug control register even though
826 * Being careful here means that we don't have to be as careful in a
827 * lot of more complicated places (task switching can be a bit lazy
828 * about restoring all the debug state, and ptrace doesn't have to
829 * find every occurrence of the TF bit that could be saved away even
832 void __kprobes do_debug(struct pt_regs * regs, long error_code)
834 unsigned int condition;
835 struct task_struct *tsk = current;
837 trace_hardirqs_fixup();
839 get_debugreg(condition, 6);
842 * The processor cleared BTF, so don't mark that we need it set.
844 clear_tsk_thread_flag(tsk, TIF_DEBUGCTLMSR);
845 tsk->thread.debugctlmsr = 0;
847 if (notify_die(DIE_DEBUG, "debug", regs, condition, error_code,
848 SIGTRAP) == NOTIFY_STOP)
850 /* It's safe to allow irq's after DR6 has been saved */
851 if (regs->flags & X86_EFLAGS_IF)
854 /* Mask out spurious debug traps due to lazy DR7 setting */
855 if (condition & (DR_TRAP0|DR_TRAP1|DR_TRAP2|DR_TRAP3)) {
856 if (!tsk->thread.debugreg7)
860 if (regs->flags & VM_MASK)
863 /* Save debug status register where ptrace can see it */
864 tsk->thread.debugreg6 = condition;
867 * Single-stepping through TF: make sure we ignore any events in
868 * kernel space (but re-enable TF when returning to user mode).
870 if (condition & DR_STEP) {
872 * We already checked v86 mode above, so we can
873 * check for kernel mode by just checking the CPL
876 if (!user_mode(regs))
877 goto clear_TF_reenable;
880 /* Ok, finally something we can handle */
881 send_sigtrap(tsk, regs, error_code);
883 /* Disable additional traps. They'll be re-enabled when
884 * the signal is delivered.
891 handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code, 1);
895 set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
896 regs->flags &= ~TF_MASK;
901 * Note that we play around with the 'TS' bit in an attempt to get
902 * the correct behaviour even in the presence of the asynchronous
905 void math_error(void __user *ip)
907 struct task_struct * task;
909 unsigned short cwd, swd;
912 * Save the info for the exception handler and clear the error.
916 task->thread.trap_no = 16;
917 task->thread.error_code = 0;
918 info.si_signo = SIGFPE;
920 info.si_code = __SI_FAULT;
923 * (~cwd & swd) will mask out exceptions that are not set to unmasked
924 * status. 0x3f is the exception bits in these regs, 0x200 is the
925 * C1 reg you need in case of a stack fault, 0x040 is the stack
926 * fault bit. We should only be taking one exception at a time,
927 * so if this combination doesn't produce any single exception,
928 * then we have a bad program that isn't syncronizing its FPU usage
929 * and it will suffer the consequences since we won't be able to
930 * fully reproduce the context of the exception
932 cwd = get_fpu_cwd(task);
933 swd = get_fpu_swd(task);
934 switch (swd & ~cwd & 0x3f) {
935 case 0x000: /* No unmasked exception */
937 default: /* Multiple exceptions */
939 case 0x001: /* Invalid Op */
941 * swd & 0x240 == 0x040: Stack Underflow
942 * swd & 0x240 == 0x240: Stack Overflow
943 * User must clear the SF bit (0x40) if set
945 info.si_code = FPE_FLTINV;
947 case 0x002: /* Denormalize */
948 case 0x010: /* Underflow */
949 info.si_code = FPE_FLTUND;
951 case 0x004: /* Zero Divide */
952 info.si_code = FPE_FLTDIV;
954 case 0x008: /* Overflow */
955 info.si_code = FPE_FLTOVF;
957 case 0x020: /* Precision */
958 info.si_code = FPE_FLTRES;
961 force_sig_info(SIGFPE, &info, task);
964 void do_coprocessor_error(struct pt_regs * regs, long error_code)
967 math_error((void __user *)regs->ip);
970 static void simd_math_error(void __user *ip)
972 struct task_struct * task;
974 unsigned short mxcsr;
977 * Save the info for the exception handler and clear the error.
981 task->thread.trap_no = 19;
982 task->thread.error_code = 0;
983 info.si_signo = SIGFPE;
985 info.si_code = __SI_FAULT;
988 * The SIMD FPU exceptions are handled a little differently, as there
989 * is only a single status/control register. Thus, to determine which
990 * unmasked exception was caught we must mask the exception mask bits
991 * at 0x1f80, and then use these to mask the exception bits at 0x3f.
993 mxcsr = get_fpu_mxcsr(task);
994 switch (~((mxcsr & 0x1f80) >> 7) & (mxcsr & 0x3f)) {
998 case 0x001: /* Invalid Op */
999 info.si_code = FPE_FLTINV;
1001 case 0x002: /* Denormalize */
1002 case 0x010: /* Underflow */
1003 info.si_code = FPE_FLTUND;
1005 case 0x004: /* Zero Divide */
1006 info.si_code = FPE_FLTDIV;
1008 case 0x008: /* Overflow */
1009 info.si_code = FPE_FLTOVF;
1011 case 0x020: /* Precision */
1012 info.si_code = FPE_FLTRES;
1015 force_sig_info(SIGFPE, &info, task);
1018 void do_simd_coprocessor_error(struct pt_regs * regs,
1022 /* Handle SIMD FPU exceptions on PIII+ processors. */
1024 simd_math_error((void __user *)regs->ip);
1027 * Handle strange cache flush from user space exception
1028 * in all other cases. This is undocumented behaviour.
1030 if (regs->flags & VM_MASK) {
1031 handle_vm86_fault((struct kernel_vm86_regs *)regs,
1035 current->thread.trap_no = 19;
1036 current->thread.error_code = error_code;
1037 die_if_kernel("cache flush denied", regs, error_code);
1038 force_sig(SIGSEGV, current);
1042 void do_spurious_interrupt_bug(struct pt_regs * regs,
1046 /* No need to warn about this any longer. */
1047 printk("Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
1051 unsigned long patch_espfix_desc(unsigned long uesp,
1054 struct desc_struct *gdt = __get_cpu_var(gdt_page).gdt;
1055 unsigned long base = (kesp - uesp) & -THREAD_SIZE;
1056 unsigned long new_kesp = kesp - base;
1057 unsigned long lim_pages = (new_kesp | (THREAD_SIZE - 1)) >> PAGE_SHIFT;
1058 __u64 desc = *(__u64 *)&gdt[GDT_ENTRY_ESPFIX_SS];
1059 /* Set up base for espfix segment */
1060 desc &= 0x00f0ff0000000000ULL;
1061 desc |= ((((__u64)base) << 16) & 0x000000ffffff0000ULL) |
1062 ((((__u64)base) << 32) & 0xff00000000000000ULL) |
1063 ((((__u64)lim_pages) << 32) & 0x000f000000000000ULL) |
1064 (lim_pages & 0xffff);
1065 *(__u64 *)&gdt[GDT_ENTRY_ESPFIX_SS] = desc;
1070 * 'math_state_restore()' saves the current math information in the
1071 * old math state array, and gets the new ones from the current task
1073 * Careful.. There are problems with IBM-designed IRQ13 behaviour.
1074 * Don't touch unless you *really* know how it works.
1076 * Must be called with kernel preemption disabled (in this case,
1077 * local interrupts are disabled at the call-site in entry.S).
1079 asmlinkage void math_state_restore(void)
1081 struct thread_info *thread = current_thread_info();
1082 struct task_struct *tsk = thread->task;
1084 clts(); /* Allow maths ops (or we recurse) */
1085 if (!tsk_used_math(tsk))
1088 thread->status |= TS_USEDFPU; /* So we fnsave on switch_to() */
1091 EXPORT_SYMBOL_GPL(math_state_restore);
1093 #ifndef CONFIG_MATH_EMULATION
1095 asmlinkage void math_emulate(long arg)
1097 printk(KERN_EMERG "math-emulation not enabled and no coprocessor found.\n");
1098 printk(KERN_EMERG "killing %s.\n",current->comm);
1099 force_sig(SIGFPE,current);
1103 #endif /* CONFIG_MATH_EMULATION */
1106 void __init trap_init(void)
1111 void __iomem *p = ioremap(0x0FFFD9, 4);
1112 if (readl(p) == 'E'+('I'<<8)+('S'<<16)+('A'<<24)) {
1118 #ifdef CONFIG_X86_LOCAL_APIC
1119 init_apic_mappings();
1122 set_trap_gate(0,÷_error);
1123 set_intr_gate(1,&debug);
1124 set_intr_gate(2,&nmi);
1125 set_system_intr_gate(3, &int3); /* int3/4 can be called from all */
1126 set_system_gate(4,&overflow);
1127 set_trap_gate(5,&bounds);
1128 set_trap_gate(6,&invalid_op);
1129 set_trap_gate(7,&device_not_available);
1130 set_task_gate(8,GDT_ENTRY_DOUBLEFAULT_TSS);
1131 set_trap_gate(9,&coprocessor_segment_overrun);
1132 set_trap_gate(10,&invalid_TSS);
1133 set_trap_gate(11,&segment_not_present);
1134 set_trap_gate(12,&stack_segment);
1135 set_trap_gate(13,&general_protection);
1136 set_intr_gate(14,&page_fault);
1137 set_trap_gate(15,&spurious_interrupt_bug);
1138 set_trap_gate(16,&coprocessor_error);
1139 set_trap_gate(17,&alignment_check);
1140 #ifdef CONFIG_X86_MCE
1141 set_trap_gate(18,&machine_check);
1143 set_trap_gate(19,&simd_coprocessor_error);
1147 * Verify that the FXSAVE/FXRSTOR data will be 16-byte aligned.
1148 * Generates a compile-time "error: zero width for bit-field" if
1149 * the alignment is wrong.
1151 struct fxsrAlignAssert {
1152 int _:!(offsetof(struct task_struct,
1153 thread.i387.fxsave) & 15);
1156 printk(KERN_INFO "Enabling fast FPU save and restore... ");
1157 set_in_cr4(X86_CR4_OSFXSR);
1161 printk(KERN_INFO "Enabling unmasked SIMD FPU exception "
1163 set_in_cr4(X86_CR4_OSXMMEXCPT);
1167 set_system_gate(SYSCALL_VECTOR,&system_call);
1169 /* Reserve all the builtin and the syscall vector. */
1170 for (i = 0; i < FIRST_EXTERNAL_VECTOR; i++)
1171 set_bit(i, used_vectors);
1172 set_bit(SYSCALL_VECTOR, used_vectors);
1175 * Should be a barrier for any external CPU state.
1182 static int __init kstack_setup(char *s)
1184 kstack_depth_to_print = simple_strtoul(s, NULL, 0);
1187 __setup("kstack=", kstack_setup);
1189 static int __init code_bytes_setup(char *s)
1191 code_bytes = simple_strtoul(s, NULL, 0);
1192 if (code_bytes > 8192)
1197 __setup("code_bytes=", code_bytes_setup);