2 * Copyright (C) 1991, 1992 Linus Torvalds
3 * Copyright (C) 2000, 2001, 2002 Andi Kleen, SuSE Labs
5 * Pentium III FXSR, SSE support
6 * Gareth Hughes <gareth@valinux.com>, May 2000
10 * 'Traps.c' handles hardware traps and faults after we have saved some
13 #include <linux/interrupt.h>
14 #include <linux/kallsyms.h>
15 #include <linux/spinlock.h>
16 #include <linux/highmem.h>
17 #include <linux/kprobes.h>
18 #include <linux/uaccess.h>
19 #include <linux/utsname.h>
20 #include <linux/kdebug.h>
21 #include <linux/kernel.h>
22 #include <linux/module.h>
23 #include <linux/ptrace.h>
24 #include <linux/string.h>
25 #include <linux/unwind.h>
26 #include <linux/delay.h>
27 #include <linux/errno.h>
28 #include <linux/kexec.h>
29 #include <linux/sched.h>
30 #include <linux/timer.h>
31 #include <linux/init.h>
32 #include <linux/bug.h>
33 #include <linux/nmi.h>
37 #include <linux/ioport.h>
38 #include <linux/eisa.h>
42 #include <linux/mca.h>
45 #if defined(CONFIG_EDAC)
46 #include <linux/edac.h>
49 #include <asm/arch_hooks.h>
50 #include <asm/stacktrace.h>
51 #include <asm/processor.h>
52 #include <asm/debugreg.h>
53 #include <asm/atomic.h>
54 #include <asm/system.h>
55 #include <asm/unwind.h>
62 #include "mach_traps.h"
64 DECLARE_BITMAP(used_vectors, NR_VECTORS);
65 EXPORT_SYMBOL_GPL(used_vectors);
67 asmlinkage int system_call(void);
69 /* Do we ignore FPU interrupts ? */
73 * The IDT has to be page-aligned to simplify the Pentium
74 * F0 0F bug workaround.. We have a special link segment
77 gate_desc idt_table[256]
78 __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 panic_on_unrecovered_nmi;
101 int kstack_depth_to_print = 24;
102 static unsigned int code_bytes = 64;
103 static int ignore_nmis;
104 static int die_counter;
106 void printk_address(unsigned long address, int reliable)
108 #ifdef CONFIG_KALLSYMS
109 unsigned long offset = 0;
110 unsigned long symsize;
114 char namebuf[KSYM_NAME_LEN];
117 symname = kallsyms_lookup(address, &symsize, &offset,
120 printk(" [<%08lx>]\n", address);
124 strcpy(reliab, "? ");
127 modname = delim = "";
128 printk(" [<%08lx>] %s%s%s%s%s+0x%lx/0x%lx\n",
129 address, reliab, delim, modname, delim, symname, offset, symsize);
131 printk(" [<%08lx>]\n", address);
135 static inline int valid_stack_ptr(struct thread_info *tinfo,
136 void *p, unsigned int size)
139 return p > t && p <= t + THREAD_SIZE - size;
142 /* The form of the top of the frame on the stack */
144 struct stack_frame *next_frame;
145 unsigned long return_address;
148 static inline unsigned long
149 print_context_stack(struct thread_info *tinfo,
150 unsigned long *stack, unsigned long bp,
151 const struct stacktrace_ops *ops, void *data)
153 struct stack_frame *frame = (struct stack_frame *)bp;
155 while (valid_stack_ptr(tinfo, stack, sizeof(*stack))) {
159 if (__kernel_text_address(addr)) {
160 if ((unsigned long) stack == bp + 4) {
161 ops->address(data, addr, 1);
162 frame = frame->next_frame;
163 bp = (unsigned long) frame;
165 ops->address(data, addr, bp == 0);
173 void dump_trace(struct task_struct *task, struct pt_regs *regs,
174 unsigned long *stack, unsigned long bp,
175 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 || probe_kernel_address(ip, c)) {
355 /* try starting at EIP */
357 code_len = code_len - code_prologue + 1;
359 for (i = 0; i < code_len; i++, ip++) {
360 if (ip < (u8 *)PAGE_OFFSET ||
361 probe_kernel_address(ip, c)) {
362 printk(" Bad EIP value.");
365 if (ip == (u8 *)regs->ip)
366 printk("<%02x> ", c);
374 int is_valid_bugaddr(unsigned long ip)
378 if (ip < PAGE_OFFSET)
380 if (probe_kernel_address((unsigned short *)ip, ud2))
383 return ud2 == 0x0b0f;
386 static raw_spinlock_t die_lock = __RAW_SPIN_LOCK_UNLOCKED;
387 static int die_owner = -1;
388 static unsigned int die_nest_count;
390 unsigned __kprobes long oops_begin(void)
396 if (die_owner != raw_smp_processor_id()) {
398 raw_local_irq_save(flags);
399 __raw_spin_lock(&die_lock);
400 die_owner = smp_processor_id();
404 raw_local_irq_save(flags);
410 void __kprobes oops_end(unsigned long flags, struct pt_regs *regs, int signr)
414 add_taint(TAINT_DIE);
415 __raw_spin_unlock(&die_lock);
416 raw_local_irq_restore(flags);
421 if (kexec_should_crash(current))
425 panic("Fatal exception in interrupt");
428 panic("Fatal exception");
434 int __kprobes __die(const char *str, struct pt_regs *regs, long err)
439 printk(KERN_EMERG "%s: %04lx [#%d] ", str, err & 0xffff, ++die_counter);
440 #ifdef CONFIG_PREEMPT
446 #ifdef CONFIG_DEBUG_PAGEALLOC
447 printk("DEBUG_PAGEALLOC");
450 if (notify_die(DIE_OOPS, str, regs, err,
451 current->thread.trap_no, SIGSEGV) == NOTIFY_STOP)
454 show_registers(regs);
455 /* Executive summary in case the oops scrolled away */
456 sp = (unsigned long) (®s->sp);
458 if (user_mode(regs)) {
460 ss = regs->ss & 0xffff;
462 printk(KERN_EMERG "EIP: [<%08lx>] ", regs->ip);
463 print_symbol("%s", regs->ip);
464 printk(" SS:ESP %04x:%08lx\n", ss, sp);
469 * This is gone through when something in the kernel has done something bad
470 * and is about to be terminated:
472 void die(const char *str, struct pt_regs *regs, long err)
474 unsigned long flags = oops_begin();
476 if (die_nest_count < 3) {
477 report_bug(regs->ip, regs);
479 if (__die(str, regs, err))
482 printk(KERN_EMERG "Recursive die() failure, output suppressed\n");
485 oops_end(flags, regs, SIGSEGV);
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)
609 do_general_protection(struct pt_regs *regs, long error_code)
611 struct task_struct *tsk;
612 struct thread_struct *thread;
613 struct tss_struct *tss;
617 tss = &per_cpu(init_tss, cpu);
618 thread = ¤t->thread;
621 * Perform the lazy TSS's I/O bitmap copy. If the TSS has an
622 * invalid offset set (the LAZY one) and the faulting thread has
623 * a valid I/O bitmap pointer, we copy the I/O bitmap in the TSS
624 * and we set the offset field correctly. Then we let the CPU to
625 * restart the faulting instruction.
627 if (tss->x86_tss.io_bitmap_base == INVALID_IO_BITMAP_OFFSET_LAZY &&
628 thread->io_bitmap_ptr) {
629 memcpy(tss->io_bitmap, thread->io_bitmap_ptr,
630 thread->io_bitmap_max);
632 * If the previously set map was extending to higher ports
633 * than the current one, pad extra space with 0xff (no access).
635 if (thread->io_bitmap_max < tss->io_bitmap_max) {
636 memset((char *) tss->io_bitmap +
637 thread->io_bitmap_max, 0xff,
638 tss->io_bitmap_max - thread->io_bitmap_max);
640 tss->io_bitmap_max = thread->io_bitmap_max;
641 tss->x86_tss.io_bitmap_base = IO_BITMAP_OFFSET;
642 tss->io_bitmap_owner = thread;
649 if (regs->flags & X86_VM_MASK)
653 if (!user_mode(regs))
656 tsk->thread.error_code = error_code;
657 tsk->thread.trap_no = 13;
659 if (show_unhandled_signals && unhandled_signal(tsk, SIGSEGV) &&
660 printk_ratelimit()) {
662 "%s[%d] general protection ip:%lx sp:%lx error:%lx",
663 tsk->comm, task_pid_nr(tsk),
664 regs->ip, regs->sp, error_code);
665 print_vma_addr(" in ", regs->ip);
669 force_sig(SIGSEGV, tsk);
674 handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
678 if (fixup_exception(regs))
681 tsk->thread.error_code = error_code;
682 tsk->thread.trap_no = 13;
683 if (notify_die(DIE_GPF, "general protection fault", regs,
684 error_code, 13, SIGSEGV) == NOTIFY_STOP)
686 die("general protection fault", regs, error_code);
689 static notrace __kprobes void
690 mem_parity_error(unsigned char reason, struct pt_regs *regs)
693 "Uhhuh. NMI received for unknown reason %02x on CPU %d.\n",
694 reason, smp_processor_id());
697 "You have some hardware problem, likely on the PCI bus.\n");
699 #if defined(CONFIG_EDAC)
700 if (edac_handler_set()) {
701 edac_atomic_assert_error();
706 if (panic_on_unrecovered_nmi)
707 panic("NMI: Not continuing");
709 printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
711 /* Clear and disable the memory parity error line. */
712 clear_mem_error(reason);
715 static notrace __kprobes void
716 io_check_error(unsigned char reason, struct pt_regs *regs)
720 printk(KERN_EMERG "NMI: IOCK error (debug interrupt?)\n");
721 show_registers(regs);
723 /* Re-enable the IOCK line, wait for a few seconds */
724 reason = (reason & 0xf) | 8;
735 static notrace __kprobes void
736 unknown_nmi_error(unsigned char reason, struct pt_regs *regs)
738 if (notify_die(DIE_NMIUNKNOWN, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP)
742 * Might actually be able to figure out what the guilty party
751 "Uhhuh. NMI received for unknown reason %02x on CPU %d.\n",
752 reason, smp_processor_id());
754 printk(KERN_EMERG "Do you have a strange power saving mode enabled?\n");
755 if (panic_on_unrecovered_nmi)
756 panic("NMI: Not continuing");
758 printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
761 static DEFINE_SPINLOCK(nmi_print_lock);
763 void notrace __kprobes die_nmi(char *str, struct pt_regs *regs, int do_panic)
765 if (notify_die(DIE_NMIWATCHDOG, str, regs, 0, 2, SIGINT) == NOTIFY_STOP)
768 spin_lock(&nmi_print_lock);
770 * We are in trouble anyway, lets at least try
771 * to get a message out:
774 printk(KERN_EMERG "%s", str);
775 printk(" on CPU%d, ip %08lx, registers:\n",
776 smp_processor_id(), regs->ip);
777 show_registers(regs);
779 panic("Non maskable interrupt");
781 spin_unlock(&nmi_print_lock);
785 * If we are in kernel we are probably nested up pretty bad
786 * and might aswell get out now while we still can:
788 if (!user_mode_vm(regs)) {
789 current->thread.trap_no = 2;
796 static notrace __kprobes void default_do_nmi(struct pt_regs *regs)
798 unsigned char reason = 0;
801 cpu = smp_processor_id();
803 /* Only the BSP gets external NMIs from the system. */
805 reason = get_nmi_reason();
807 if (!(reason & 0xc0)) {
808 if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 2, SIGINT)
811 #ifdef CONFIG_X86_LOCAL_APIC
813 * Ok, so this is none of the documented NMI sources,
814 * so it must be the NMI watchdog.
816 if (nmi_watchdog_tick(regs, reason))
818 if (!do_nmi_callback(regs, cpu))
819 unknown_nmi_error(reason, regs);
821 unknown_nmi_error(reason, regs);
826 if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP)
829 /* AK: following checks seem to be broken on modern chipsets. FIXME */
831 mem_parity_error(reason, regs);
833 io_check_error(reason, regs);
835 * Reassert NMI in case it became active meanwhile
836 * as it's edge-triggered:
841 notrace __kprobes void do_nmi(struct pt_regs *regs, long error_code)
847 cpu = smp_processor_id();
852 default_do_nmi(regs);
863 void restart_nmi(void)
869 #ifdef CONFIG_KPROBES
870 void __kprobes do_int3(struct pt_regs *regs, long error_code)
872 trace_hardirqs_fixup();
874 if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP)
878 * This is an interrupt gate, because kprobes wants interrupts
879 * disabled. Normal trap handlers don't.
881 restore_interrupts(regs);
883 do_trap(3, SIGTRAP, "int3", 1, regs, error_code, NULL);
888 * Our handling of the processor debug registers is non-trivial.
889 * We do not clear them on entry and exit from the kernel. Therefore
890 * it is possible to get a watchpoint trap here from inside the kernel.
891 * However, the code in ./ptrace.c has ensured that the user can
892 * only set watchpoints on userspace addresses. Therefore the in-kernel
893 * watchpoint trap can only occur in code which is reading/writing
894 * from user space. Such code must not hold kernel locks (since it
895 * can equally take a page fault), therefore it is safe to call
896 * force_sig_info even though that claims and releases locks.
898 * Code in ./signal.c ensures that the debug control register
899 * is restored before we deliver any signal, and therefore that
900 * user code runs with the correct debug control register even though
903 * Being careful here means that we don't have to be as careful in a
904 * lot of more complicated places (task switching can be a bit lazy
905 * about restoring all the debug state, and ptrace doesn't have to
906 * find every occurrence of the TF bit that could be saved away even
909 void __kprobes do_debug(struct pt_regs *regs, long error_code)
911 struct task_struct *tsk = current;
912 unsigned int condition;
914 trace_hardirqs_fixup();
916 get_debugreg(condition, 6);
919 * The processor cleared BTF, so don't mark that we need it set.
921 clear_tsk_thread_flag(tsk, TIF_DEBUGCTLMSR);
922 tsk->thread.debugctlmsr = 0;
924 if (notify_die(DIE_DEBUG, "debug", regs, condition, error_code,
925 SIGTRAP) == NOTIFY_STOP)
927 /* It's safe to allow irq's after DR6 has been saved */
928 if (regs->flags & X86_EFLAGS_IF)
931 /* Mask out spurious debug traps due to lazy DR7 setting */
932 if (condition & (DR_TRAP0|DR_TRAP1|DR_TRAP2|DR_TRAP3)) {
933 if (!tsk->thread.debugreg7)
937 if (regs->flags & X86_VM_MASK)
940 /* Save debug status register where ptrace can see it */
941 tsk->thread.debugreg6 = condition;
944 * Single-stepping through TF: make sure we ignore any events in
945 * kernel space (but re-enable TF when returning to user mode).
947 if (condition & DR_STEP) {
949 * We already checked v86 mode above, so we can
950 * check for kernel mode by just checking the CPL
953 if (!user_mode(regs))
954 goto clear_TF_reenable;
957 /* Ok, finally something we can handle */
958 send_sigtrap(tsk, regs, error_code);
961 * Disable additional traps. They'll be re-enabled when
962 * the signal is delivered.
969 handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code, 1);
973 set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
974 regs->flags &= ~X86_EFLAGS_TF;
979 * Note that we play around with the 'TS' bit in an attempt to get
980 * the correct behaviour even in the presence of the asynchronous
983 void math_error(void __user *ip)
985 struct task_struct *task;
987 unsigned short cwd, swd;
990 * Save the info for the exception handler and clear the error.
994 task->thread.trap_no = 16;
995 task->thread.error_code = 0;
996 info.si_signo = SIGFPE;
998 info.si_code = __SI_FAULT;
1001 * (~cwd & swd) will mask out exceptions that are not set to unmasked
1002 * status. 0x3f is the exception bits in these regs, 0x200 is the
1003 * C1 reg you need in case of a stack fault, 0x040 is the stack
1004 * fault bit. We should only be taking one exception at a time,
1005 * so if this combination doesn't produce any single exception,
1006 * then we have a bad program that isn't synchronizing its FPU usage
1007 * and it will suffer the consequences since we won't be able to
1008 * fully reproduce the context of the exception
1010 cwd = get_fpu_cwd(task);
1011 swd = get_fpu_swd(task);
1012 switch (swd & ~cwd & 0x3f) {
1013 case 0x000: /* No unmasked exception */
1015 default: /* Multiple exceptions */
1017 case 0x001: /* Invalid Op */
1019 * swd & 0x240 == 0x040: Stack Underflow
1020 * swd & 0x240 == 0x240: Stack Overflow
1021 * User must clear the SF bit (0x40) if set
1023 info.si_code = FPE_FLTINV;
1025 case 0x002: /* Denormalize */
1026 case 0x010: /* Underflow */
1027 info.si_code = FPE_FLTUND;
1029 case 0x004: /* Zero Divide */
1030 info.si_code = FPE_FLTDIV;
1032 case 0x008: /* Overflow */
1033 info.si_code = FPE_FLTOVF;
1035 case 0x020: /* Precision */
1036 info.si_code = FPE_FLTRES;
1039 force_sig_info(SIGFPE, &info, task);
1042 void do_coprocessor_error(struct pt_regs *regs, long error_code)
1045 math_error((void __user *)regs->ip);
1048 static void simd_math_error(void __user *ip)
1050 struct task_struct *task;
1052 unsigned short mxcsr;
1055 * Save the info for the exception handler and clear the error.
1058 save_init_fpu(task);
1059 task->thread.trap_no = 19;
1060 task->thread.error_code = 0;
1061 info.si_signo = SIGFPE;
1063 info.si_code = __SI_FAULT;
1066 * The SIMD FPU exceptions are handled a little differently, as there
1067 * is only a single status/control register. Thus, to determine which
1068 * unmasked exception was caught we must mask the exception mask bits
1069 * at 0x1f80, and then use these to mask the exception bits at 0x3f.
1071 mxcsr = get_fpu_mxcsr(task);
1072 switch (~((mxcsr & 0x1f80) >> 7) & (mxcsr & 0x3f)) {
1076 case 0x001: /* Invalid Op */
1077 info.si_code = FPE_FLTINV;
1079 case 0x002: /* Denormalize */
1080 case 0x010: /* Underflow */
1081 info.si_code = FPE_FLTUND;
1083 case 0x004: /* Zero Divide */
1084 info.si_code = FPE_FLTDIV;
1086 case 0x008: /* Overflow */
1087 info.si_code = FPE_FLTOVF;
1089 case 0x020: /* Precision */
1090 info.si_code = FPE_FLTRES;
1093 force_sig_info(SIGFPE, &info, task);
1096 void do_simd_coprocessor_error(struct pt_regs *regs, long error_code)
1099 /* Handle SIMD FPU exceptions on PIII+ processors. */
1101 simd_math_error((void __user *)regs->ip);
1105 * Handle strange cache flush from user space exception
1106 * in all other cases. This is undocumented behaviour.
1108 if (regs->flags & X86_VM_MASK) {
1109 handle_vm86_fault((struct kernel_vm86_regs *)regs, error_code);
1112 current->thread.trap_no = 19;
1113 current->thread.error_code = error_code;
1114 die_if_kernel("cache flush denied", regs, error_code);
1115 force_sig(SIGSEGV, current);
1118 void do_spurious_interrupt_bug(struct pt_regs *regs, long error_code)
1121 /* No need to warn about this any longer. */
1122 printk(KERN_INFO "Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
1126 unsigned long patch_espfix_desc(unsigned long uesp, unsigned long kesp)
1128 struct desc_struct *gdt = get_cpu_gdt_table(smp_processor_id());
1129 unsigned long base = (kesp - uesp) & -THREAD_SIZE;
1130 unsigned long new_kesp = kesp - base;
1131 unsigned long lim_pages = (new_kesp | (THREAD_SIZE - 1)) >> PAGE_SHIFT;
1132 __u64 desc = *(__u64 *)&gdt[GDT_ENTRY_ESPFIX_SS];
1134 /* Set up base for espfix segment */
1135 desc &= 0x00f0ff0000000000ULL;
1136 desc |= ((((__u64)base) << 16) & 0x000000ffffff0000ULL) |
1137 ((((__u64)base) << 32) & 0xff00000000000000ULL) |
1138 ((((__u64)lim_pages) << 32) & 0x000f000000000000ULL) |
1139 (lim_pages & 0xffff);
1140 *(__u64 *)&gdt[GDT_ENTRY_ESPFIX_SS] = desc;
1146 * 'math_state_restore()' saves the current math information in the
1147 * old math state array, and gets the new ones from the current task
1149 * Careful.. There are problems with IBM-designed IRQ13 behaviour.
1150 * Don't touch unless you *really* know how it works.
1152 * Must be called with kernel preemption disabled (in this case,
1153 * local interrupts are disabled at the call-site in entry.S).
1155 asmlinkage void math_state_restore(void)
1157 struct thread_info *thread = current_thread_info();
1158 struct task_struct *tsk = thread->task;
1160 if (!tsk_used_math(tsk)) {
1163 * does a slab alloc which can sleep
1165 if (init_fpu(tsk)) {
1167 * ran out of memory!
1169 do_group_exit(SIGKILL);
1172 local_irq_disable();
1175 clts(); /* Allow maths ops (or we recurse) */
1177 thread->status |= TS_USEDFPU; /* So we fnsave on switch_to() */
1180 EXPORT_SYMBOL_GPL(math_state_restore);
1182 #ifndef CONFIG_MATH_EMULATION
1184 asmlinkage void math_emulate(long arg)
1187 "math-emulation not enabled and no coprocessor found.\n");
1188 printk(KERN_EMERG "killing %s.\n", current->comm);
1189 force_sig(SIGFPE, current);
1193 #endif /* CONFIG_MATH_EMULATION */
1195 void __init trap_init(void)
1200 void __iomem *p = early_ioremap(0x0FFFD9, 4);
1202 if (readl(p) == 'E' + ('I'<<8) + ('S'<<16) + ('A'<<24))
1204 early_iounmap(p, 4);
1207 set_trap_gate(0, ÷_error);
1208 set_intr_gate(1, &debug);
1209 set_intr_gate(2, &nmi);
1210 set_system_intr_gate(3, &int3); /* int3 can be called from all */
1211 set_system_gate(4, &overflow); /* int4 can be called from all */
1212 set_trap_gate(5, &bounds);
1213 set_trap_gate(6, &invalid_op);
1214 set_trap_gate(7, &device_not_available);
1215 set_task_gate(8, GDT_ENTRY_DOUBLEFAULT_TSS);
1216 set_trap_gate(9, &coprocessor_segment_overrun);
1217 set_trap_gate(10, &invalid_TSS);
1218 set_trap_gate(11, &segment_not_present);
1219 set_trap_gate(12, &stack_segment);
1220 set_trap_gate(13, &general_protection);
1221 set_intr_gate(14, &page_fault);
1222 set_trap_gate(15, &spurious_interrupt_bug);
1223 set_trap_gate(16, &coprocessor_error);
1224 set_trap_gate(17, &alignment_check);
1225 #ifdef CONFIG_X86_MCE
1226 set_trap_gate(18, &machine_check);
1228 set_trap_gate(19, &simd_coprocessor_error);
1231 printk(KERN_INFO "Enabling fast FPU save and restore... ");
1232 set_in_cr4(X86_CR4_OSFXSR);
1237 "Enabling unmasked SIMD FPU exception support... ");
1238 set_in_cr4(X86_CR4_OSXMMEXCPT);
1242 set_system_gate(SYSCALL_VECTOR, &system_call);
1244 /* Reserve all the builtin and the syscall vector: */
1245 for (i = 0; i < FIRST_EXTERNAL_VECTOR; i++)
1246 set_bit(i, used_vectors);
1248 set_bit(SYSCALL_VECTOR, used_vectors);
1250 init_thread_xstate();
1252 * Should be a barrier for any external CPU state:
1259 static int __init kstack_setup(char *s)
1261 kstack_depth_to_print = simple_strtoul(s, NULL, 0);
1265 __setup("kstack=", kstack_setup);
1267 static int __init code_bytes_setup(char *s)
1269 code_bytes = simple_strtoul(s, NULL, 0);
1270 if (code_bytes > 8192)
1275 __setup("code_bytes=", code_bytes_setup);