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>
61 #include <asm/traps.h>
63 #include "mach_traps.h"
65 DECLARE_BITMAP(used_vectors, NR_VECTORS);
66 EXPORT_SYMBOL_GPL(used_vectors);
68 asmlinkage int system_call(void);
70 /* Do we ignore FPU interrupts ? */
74 * The IDT has to be page-aligned to simplify the Pentium
75 * F0 0F bug workaround.. We have a special link segment
78 gate_desc idt_table[256]
79 __attribute__((__section__(".data.idt"))) = { { { { 0, 0 } } }, };
81 int panic_on_unrecovered_nmi;
82 int kstack_depth_to_print = 24;
83 static unsigned int code_bytes = 64;
84 static int ignore_nmis;
85 static int die_counter;
87 void printk_address(unsigned long address, int reliable)
89 #ifdef CONFIG_KALLSYMS
90 unsigned long offset = 0;
91 unsigned long symsize;
95 char namebuf[KSYM_NAME_LEN];
98 symname = kallsyms_lookup(address, &symsize, &offset,
101 printk(" [<%08lx>]\n", address);
105 strcpy(reliab, "? ");
108 modname = delim = "";
109 printk(" [<%08lx>] %s%s%s%s%s+0x%lx/0x%lx\n",
110 address, reliab, delim, modname, delim, symname, offset, symsize);
112 printk(" [<%08lx>]\n", address);
116 static inline int valid_stack_ptr(struct thread_info *tinfo,
117 void *p, unsigned int size)
120 return p > t && p <= t + THREAD_SIZE - size;
123 /* The form of the top of the frame on the stack */
125 struct stack_frame *next_frame;
126 unsigned long return_address;
129 static inline unsigned long
130 print_context_stack(struct thread_info *tinfo,
131 unsigned long *stack, unsigned long bp,
132 const struct stacktrace_ops *ops, void *data)
134 struct stack_frame *frame = (struct stack_frame *)bp;
136 while (valid_stack_ptr(tinfo, stack, sizeof(*stack))) {
140 if (__kernel_text_address(addr)) {
141 if ((unsigned long) stack == bp + 4) {
142 ops->address(data, addr, 1);
143 frame = frame->next_frame;
144 bp = (unsigned long) frame;
146 ops->address(data, addr, bp == 0);
154 void dump_trace(struct task_struct *task, struct pt_regs *regs,
155 unsigned long *stack, unsigned long bp,
156 const struct stacktrace_ops *ops, void *data)
165 stack = (unsigned long *)task->thread.sp;
168 #ifdef CONFIG_FRAME_POINTER
170 if (task == current) {
171 /* Grab bp right from our regs */
172 asm("movl %%ebp, %0" : "=r" (bp) :);
174 /* bp is the last reg pushed by switch_to */
175 bp = *(unsigned long *) task->thread.sp;
181 struct thread_info *context;
183 context = (struct thread_info *)
184 ((unsigned long)stack & (~(THREAD_SIZE - 1)));
185 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:
191 if (ops->stack(data, "IRQ") < 0)
193 stack = (unsigned long *)context->previous_esp;
196 touch_nmi_watchdog();
199 EXPORT_SYMBOL(dump_trace);
202 print_trace_warning_symbol(void *data, char *msg, unsigned long symbol)
205 print_symbol(msg, symbol);
209 static void print_trace_warning(void *data, char *msg)
211 printk("%s%s\n", (char *)data, msg);
214 static int print_trace_stack(void *data, char *name)
220 * Print one address/symbol entries per line.
222 static void print_trace_address(void *data, unsigned long addr, int reliable)
224 printk("%s [<%08lx>] ", (char *)data, addr);
227 print_symbol("%s\n", addr);
228 touch_nmi_watchdog();
231 static const struct stacktrace_ops print_trace_ops = {
232 .warning = print_trace_warning,
233 .warning_symbol = print_trace_warning_symbol,
234 .stack = print_trace_stack,
235 .address = print_trace_address,
239 show_trace_log_lvl(struct task_struct *task, struct pt_regs *regs,
240 unsigned long *stack, unsigned long bp, char *log_lvl)
242 dump_trace(task, regs, stack, bp, &print_trace_ops, log_lvl);
243 printk("%s =======================\n", log_lvl);
246 void show_trace(struct task_struct *task, struct pt_regs *regs,
247 unsigned long *stack, unsigned long bp)
249 show_trace_log_lvl(task, regs, stack, bp, "");
253 show_stack_log_lvl(struct task_struct *task, struct pt_regs *regs,
254 unsigned long *sp, unsigned long bp, char *log_lvl)
256 unsigned long *stack;
261 sp = (unsigned long *)task->thread.sp;
263 sp = (unsigned long *)&sp;
267 for (i = 0; i < kstack_depth_to_print; i++) {
268 if (kstack_end(stack))
270 if (i && ((i % 8) == 0))
271 printk("\n%s ", log_lvl);
272 printk("%08lx ", *stack++);
274 printk("\n%sCall Trace:\n", log_lvl);
276 show_trace_log_lvl(task, regs, sp, bp, log_lvl);
279 void show_stack(struct task_struct *task, unsigned long *sp)
282 show_stack_log_lvl(task, NULL, sp, 0, "");
286 * The architecture-independent dump_stack generator
288 void dump_stack(void)
290 unsigned long bp = 0;
293 #ifdef CONFIG_FRAME_POINTER
295 asm("movl %%ebp, %0" : "=r" (bp):);
298 printk("Pid: %d, comm: %.20s %s %s %.*s\n",
299 current->pid, current->comm, print_tainted(),
300 init_utsname()->release,
301 (int)strcspn(init_utsname()->version, " "),
302 init_utsname()->version);
304 show_trace(current, NULL, &stack, bp);
307 EXPORT_SYMBOL(dump_stack);
309 void show_registers(struct pt_regs *regs)
314 __show_registers(regs, 0);
316 printk(KERN_EMERG "Process %.*s (pid: %d, ti=%p task=%p task.ti=%p)",
317 TASK_COMM_LEN, current->comm, task_pid_nr(current),
318 current_thread_info(), current, task_thread_info(current));
320 * When in-kernel, we also print out the stack and code at the
321 * time of the fault..
323 if (!user_mode_vm(regs)) {
324 unsigned int code_prologue = code_bytes * 43 / 64;
325 unsigned int code_len = code_bytes;
329 printk("\n" KERN_EMERG "Stack: ");
330 show_stack_log_lvl(NULL, regs, ®s->sp, 0, KERN_EMERG);
332 printk(KERN_EMERG "Code: ");
334 ip = (u8 *)regs->ip - code_prologue;
335 if (ip < (u8 *)PAGE_OFFSET || probe_kernel_address(ip, c)) {
336 /* try starting at EIP */
338 code_len = code_len - code_prologue + 1;
340 for (i = 0; i < code_len; i++, ip++) {
341 if (ip < (u8 *)PAGE_OFFSET ||
342 probe_kernel_address(ip, c)) {
343 printk(" Bad EIP value.");
346 if (ip == (u8 *)regs->ip)
347 printk("<%02x> ", c);
355 int is_valid_bugaddr(unsigned long ip)
359 if (ip < PAGE_OFFSET)
361 if (probe_kernel_address((unsigned short *)ip, ud2))
364 return ud2 == 0x0b0f;
367 static raw_spinlock_t die_lock = __RAW_SPIN_LOCK_UNLOCKED;
368 static int die_owner = -1;
369 static unsigned int die_nest_count;
371 unsigned __kprobes long oops_begin(void)
377 if (die_owner != raw_smp_processor_id()) {
379 raw_local_irq_save(flags);
380 __raw_spin_lock(&die_lock);
381 die_owner = smp_processor_id();
385 raw_local_irq_save(flags);
391 void __kprobes oops_end(unsigned long flags, struct pt_regs *regs, int signr)
395 add_taint(TAINT_DIE);
396 __raw_spin_unlock(&die_lock);
397 raw_local_irq_restore(flags);
402 if (kexec_should_crash(current))
406 panic("Fatal exception in interrupt");
409 panic("Fatal exception");
415 int __kprobes __die(const char *str, struct pt_regs *regs, long err)
420 printk(KERN_EMERG "%s: %04lx [#%d] ", str, err & 0xffff, ++die_counter);
421 #ifdef CONFIG_PREEMPT
427 #ifdef CONFIG_DEBUG_PAGEALLOC
428 printk("DEBUG_PAGEALLOC");
431 if (notify_die(DIE_OOPS, str, regs, err,
432 current->thread.trap_no, SIGSEGV) == NOTIFY_STOP)
435 show_registers(regs);
436 /* Executive summary in case the oops scrolled away */
437 sp = (unsigned long) (®s->sp);
439 if (user_mode(regs)) {
441 ss = regs->ss & 0xffff;
443 printk(KERN_EMERG "EIP: [<%08lx>] ", regs->ip);
444 print_symbol("%s", regs->ip);
445 printk(" SS:ESP %04x:%08lx\n", ss, sp);
450 * This is gone through when something in the kernel has done something bad
451 * and is about to be terminated:
453 void die(const char *str, struct pt_regs *regs, long err)
455 unsigned long flags = oops_begin();
457 if (die_nest_count < 3) {
458 report_bug(regs->ip, regs);
460 if (__die(str, regs, err))
463 printk(KERN_EMERG "Recursive die() failure, output suppressed\n");
466 oops_end(flags, regs, SIGSEGV);
470 die_if_kernel(const char *str, struct pt_regs *regs, long err)
472 if (!user_mode_vm(regs))
476 static void __kprobes
477 do_trap(int trapnr, int signr, char *str, int vm86, struct pt_regs *regs,
478 long error_code, siginfo_t *info)
480 struct task_struct *tsk = current;
482 if (regs->flags & X86_VM_MASK) {
488 if (!user_mode(regs))
493 * We want error_code and trap_no set for userspace faults and
494 * kernelspace faults which result in die(), but not
495 * kernelspace faults which are fixed up. die() gives the
496 * process no chance to handle the signal and notice the
497 * kernel fault information, so that won't result in polluting
498 * the information about previously queued, but not yet
499 * delivered, faults. See also do_general_protection below.
501 tsk->thread.error_code = error_code;
502 tsk->thread.trap_no = trapnr;
505 force_sig_info(signr, info, tsk);
507 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);
519 if (handle_vm86_trap((struct kernel_vm86_regs *) regs,
525 #define DO_ERROR(trapnr, signr, str, name) \
526 void do_##name(struct pt_regs *regs, long error_code) \
528 trace_hardirqs_fixup(); \
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 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 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 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 trace_hardirqs_fixup(); \
569 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
572 do_trap(trapnr, signr, str, 1, regs, error_code, &info); \
575 DO_VM86_ERROR_INFO(0, SIGFPE, "divide error", divide_error, FPE_INTDIV, regs->ip)
576 #ifndef CONFIG_KPROBES
577 DO_VM86_ERROR(3, SIGTRAP, "int3", int3)
579 DO_VM86_ERROR(4, SIGSEGV, "overflow", overflow)
580 DO_VM86_ERROR(5, SIGSEGV, "bounds", bounds)
581 DO_ERROR_INFO(6, SIGILL, "invalid opcode", invalid_op, ILL_ILLOPN, regs->ip, 0)
582 DO_ERROR(9, SIGFPE, "coprocessor segment overrun", coprocessor_segment_overrun)
583 DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
584 DO_ERROR(11, SIGBUS, "segment not present", segment_not_present)
585 DO_ERROR(12, SIGBUS, "stack segment", stack_segment)
586 DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0, 0)
587 DO_ERROR_INFO(32, SIGILL, "iret exception", iret_error, ILL_BADSTK, 0, 1)
590 do_general_protection(struct pt_regs *regs, long error_code)
592 struct task_struct *tsk;
593 struct thread_struct *thread;
594 struct tss_struct *tss;
598 tss = &per_cpu(init_tss, cpu);
599 thread = ¤t->thread;
602 * Perform the lazy TSS's I/O bitmap copy. If the TSS has an
603 * invalid offset set (the LAZY one) and the faulting thread has
604 * a valid I/O bitmap pointer, we copy the I/O bitmap in the TSS
605 * and we set the offset field correctly. Then we let the CPU to
606 * restart the faulting instruction.
608 if (tss->x86_tss.io_bitmap_base == INVALID_IO_BITMAP_OFFSET_LAZY &&
609 thread->io_bitmap_ptr) {
610 memcpy(tss->io_bitmap, thread->io_bitmap_ptr,
611 thread->io_bitmap_max);
613 * If the previously set map was extending to higher ports
614 * than the current one, pad extra space with 0xff (no access).
616 if (thread->io_bitmap_max < tss->io_bitmap_max) {
617 memset((char *) tss->io_bitmap +
618 thread->io_bitmap_max, 0xff,
619 tss->io_bitmap_max - thread->io_bitmap_max);
621 tss->io_bitmap_max = thread->io_bitmap_max;
622 tss->x86_tss.io_bitmap_base = IO_BITMAP_OFFSET;
623 tss->io_bitmap_owner = thread;
630 if (regs->flags & X86_VM_MASK)
634 if (!user_mode(regs))
637 tsk->thread.error_code = error_code;
638 tsk->thread.trap_no = 13;
640 if (show_unhandled_signals && unhandled_signal(tsk, SIGSEGV) &&
641 printk_ratelimit()) {
643 "%s[%d] general protection ip:%lx sp:%lx error:%lx",
644 tsk->comm, task_pid_nr(tsk),
645 regs->ip, regs->sp, error_code);
646 print_vma_addr(" in ", regs->ip);
650 force_sig(SIGSEGV, tsk);
655 handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
659 if (fixup_exception(regs))
662 tsk->thread.error_code = error_code;
663 tsk->thread.trap_no = 13;
664 if (notify_die(DIE_GPF, "general protection fault", regs,
665 error_code, 13, SIGSEGV) == NOTIFY_STOP)
667 die("general protection fault", regs, error_code);
670 static notrace __kprobes void
671 mem_parity_error(unsigned char reason, struct pt_regs *regs)
674 "Uhhuh. NMI received for unknown reason %02x on CPU %d.\n",
675 reason, smp_processor_id());
678 "You have some hardware problem, likely on the PCI bus.\n");
680 #if defined(CONFIG_EDAC)
681 if (edac_handler_set()) {
682 edac_atomic_assert_error();
687 if (panic_on_unrecovered_nmi)
688 panic("NMI: Not continuing");
690 printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
692 /* Clear and disable the memory parity error line. */
693 clear_mem_error(reason);
696 static notrace __kprobes void
697 io_check_error(unsigned char reason, struct pt_regs *regs)
701 printk(KERN_EMERG "NMI: IOCK error (debug interrupt?)\n");
702 show_registers(regs);
704 /* Re-enable the IOCK line, wait for a few seconds */
705 reason = (reason & 0xf) | 8;
716 static notrace __kprobes void
717 unknown_nmi_error(unsigned char reason, struct pt_regs *regs)
719 if (notify_die(DIE_NMIUNKNOWN, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP)
723 * Might actually be able to figure out what the guilty party
732 "Uhhuh. NMI received for unknown reason %02x on CPU %d.\n",
733 reason, smp_processor_id());
735 printk(KERN_EMERG "Do you have a strange power saving mode enabled?\n");
736 if (panic_on_unrecovered_nmi)
737 panic("NMI: Not continuing");
739 printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
742 static DEFINE_SPINLOCK(nmi_print_lock);
744 void notrace __kprobes die_nmi(char *str, struct pt_regs *regs, int do_panic)
746 if (notify_die(DIE_NMIWATCHDOG, str, regs, 0, 2, SIGINT) == NOTIFY_STOP)
749 spin_lock(&nmi_print_lock);
751 * We are in trouble anyway, lets at least try
752 * to get a message out:
755 printk(KERN_EMERG "%s", str);
756 printk(" on CPU%d, ip %08lx, registers:\n",
757 smp_processor_id(), regs->ip);
758 show_registers(regs);
760 panic("Non maskable interrupt");
762 spin_unlock(&nmi_print_lock);
766 * If we are in kernel we are probably nested up pretty bad
767 * and might aswell get out now while we still can:
769 if (!user_mode_vm(regs)) {
770 current->thread.trap_no = 2;
777 static notrace __kprobes void default_do_nmi(struct pt_regs *regs)
779 unsigned char reason = 0;
782 cpu = smp_processor_id();
784 /* Only the BSP gets external NMIs from the system. */
786 reason = get_nmi_reason();
788 if (!(reason & 0xc0)) {
789 if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 2, SIGINT)
792 #ifdef CONFIG_X86_LOCAL_APIC
794 * Ok, so this is none of the documented NMI sources,
795 * so it must be the NMI watchdog.
797 if (nmi_watchdog_tick(regs, reason))
799 if (!do_nmi_callback(regs, cpu))
800 unknown_nmi_error(reason, regs);
802 unknown_nmi_error(reason, regs);
807 if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP)
810 /* AK: following checks seem to be broken on modern chipsets. FIXME */
812 mem_parity_error(reason, regs);
814 io_check_error(reason, regs);
816 * Reassert NMI in case it became active meanwhile
817 * as it's edge-triggered:
822 notrace __kprobes void do_nmi(struct pt_regs *regs, long error_code)
828 cpu = smp_processor_id();
833 default_do_nmi(regs);
844 void restart_nmi(void)
850 #ifdef CONFIG_KPROBES
851 void __kprobes do_int3(struct pt_regs *regs, long error_code)
853 trace_hardirqs_fixup();
855 if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP)
859 * This is an interrupt gate, because kprobes wants interrupts
860 * disabled. Normal trap handlers don't.
862 restore_interrupts(regs);
864 do_trap(3, SIGTRAP, "int3", 1, regs, error_code, NULL);
869 * Our handling of the processor debug registers is non-trivial.
870 * We do not clear them on entry and exit from the kernel. Therefore
871 * it is possible to get a watchpoint trap here from inside the kernel.
872 * However, the code in ./ptrace.c has ensured that the user can
873 * only set watchpoints on userspace addresses. Therefore the in-kernel
874 * watchpoint trap can only occur in code which is reading/writing
875 * from user space. Such code must not hold kernel locks (since it
876 * can equally take a page fault), therefore it is safe to call
877 * force_sig_info even though that claims and releases locks.
879 * Code in ./signal.c ensures that the debug control register
880 * is restored before we deliver any signal, and therefore that
881 * user code runs with the correct debug control register even though
884 * Being careful here means that we don't have to be as careful in a
885 * lot of more complicated places (task switching can be a bit lazy
886 * about restoring all the debug state, and ptrace doesn't have to
887 * find every occurrence of the TF bit that could be saved away even
890 void __kprobes do_debug(struct pt_regs *regs, long error_code)
892 struct task_struct *tsk = current;
893 unsigned int condition;
895 trace_hardirqs_fixup();
897 get_debugreg(condition, 6);
900 * The processor cleared BTF, so don't mark that we need it set.
902 clear_tsk_thread_flag(tsk, TIF_DEBUGCTLMSR);
903 tsk->thread.debugctlmsr = 0;
905 if (notify_die(DIE_DEBUG, "debug", regs, condition, error_code,
906 SIGTRAP) == NOTIFY_STOP)
908 /* It's safe to allow irq's after DR6 has been saved */
909 if (regs->flags & X86_EFLAGS_IF)
912 /* Mask out spurious debug traps due to lazy DR7 setting */
913 if (condition & (DR_TRAP0|DR_TRAP1|DR_TRAP2|DR_TRAP3)) {
914 if (!tsk->thread.debugreg7)
918 if (regs->flags & X86_VM_MASK)
921 /* Save debug status register where ptrace can see it */
922 tsk->thread.debugreg6 = condition;
925 * Single-stepping through TF: make sure we ignore any events in
926 * kernel space (but re-enable TF when returning to user mode).
928 if (condition & DR_STEP) {
930 * We already checked v86 mode above, so we can
931 * check for kernel mode by just checking the CPL
934 if (!user_mode(regs))
935 goto clear_TF_reenable;
938 /* Ok, finally something we can handle */
939 send_sigtrap(tsk, regs, error_code);
942 * Disable additional traps. They'll be re-enabled when
943 * the signal is delivered.
950 handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code, 1);
954 set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
955 regs->flags &= ~X86_EFLAGS_TF;
960 * Note that we play around with the 'TS' bit in an attempt to get
961 * the correct behaviour even in the presence of the asynchronous
964 void math_error(void __user *ip)
966 struct task_struct *task;
968 unsigned short cwd, swd;
971 * Save the info for the exception handler and clear the error.
975 task->thread.trap_no = 16;
976 task->thread.error_code = 0;
977 info.si_signo = SIGFPE;
979 info.si_code = __SI_FAULT;
982 * (~cwd & swd) will mask out exceptions that are not set to unmasked
983 * status. 0x3f is the exception bits in these regs, 0x200 is the
984 * C1 reg you need in case of a stack fault, 0x040 is the stack
985 * fault bit. We should only be taking one exception at a time,
986 * so if this combination doesn't produce any single exception,
987 * then we have a bad program that isn't synchronizing its FPU usage
988 * and it will suffer the consequences since we won't be able to
989 * fully reproduce the context of the exception
991 cwd = get_fpu_cwd(task);
992 swd = get_fpu_swd(task);
993 switch (swd & ~cwd & 0x3f) {
994 case 0x000: /* No unmasked exception */
996 default: /* Multiple exceptions */
998 case 0x001: /* Invalid Op */
1000 * swd & 0x240 == 0x040: Stack Underflow
1001 * swd & 0x240 == 0x240: Stack Overflow
1002 * User must clear the SF bit (0x40) if set
1004 info.si_code = FPE_FLTINV;
1006 case 0x002: /* Denormalize */
1007 case 0x010: /* Underflow */
1008 info.si_code = FPE_FLTUND;
1010 case 0x004: /* Zero Divide */
1011 info.si_code = FPE_FLTDIV;
1013 case 0x008: /* Overflow */
1014 info.si_code = FPE_FLTOVF;
1016 case 0x020: /* Precision */
1017 info.si_code = FPE_FLTRES;
1020 force_sig_info(SIGFPE, &info, task);
1023 void do_coprocessor_error(struct pt_regs *regs, long error_code)
1026 math_error((void __user *)regs->ip);
1029 static void simd_math_error(void __user *ip)
1031 struct task_struct *task;
1033 unsigned short mxcsr;
1036 * Save the info for the exception handler and clear the error.
1039 save_init_fpu(task);
1040 task->thread.trap_no = 19;
1041 task->thread.error_code = 0;
1042 info.si_signo = SIGFPE;
1044 info.si_code = __SI_FAULT;
1047 * The SIMD FPU exceptions are handled a little differently, as there
1048 * is only a single status/control register. Thus, to determine which
1049 * unmasked exception was caught we must mask the exception mask bits
1050 * at 0x1f80, and then use these to mask the exception bits at 0x3f.
1052 mxcsr = get_fpu_mxcsr(task);
1053 switch (~((mxcsr & 0x1f80) >> 7) & (mxcsr & 0x3f)) {
1057 case 0x001: /* Invalid Op */
1058 info.si_code = FPE_FLTINV;
1060 case 0x002: /* Denormalize */
1061 case 0x010: /* Underflow */
1062 info.si_code = FPE_FLTUND;
1064 case 0x004: /* Zero Divide */
1065 info.si_code = FPE_FLTDIV;
1067 case 0x008: /* Overflow */
1068 info.si_code = FPE_FLTOVF;
1070 case 0x020: /* Precision */
1071 info.si_code = FPE_FLTRES;
1074 force_sig_info(SIGFPE, &info, task);
1077 void do_simd_coprocessor_error(struct pt_regs *regs, long error_code)
1080 /* Handle SIMD FPU exceptions on PIII+ processors. */
1082 simd_math_error((void __user *)regs->ip);
1086 * Handle strange cache flush from user space exception
1087 * in all other cases. This is undocumented behaviour.
1089 if (regs->flags & X86_VM_MASK) {
1090 handle_vm86_fault((struct kernel_vm86_regs *)regs, error_code);
1093 current->thread.trap_no = 19;
1094 current->thread.error_code = error_code;
1095 die_if_kernel("cache flush denied", regs, error_code);
1096 force_sig(SIGSEGV, current);
1099 void do_spurious_interrupt_bug(struct pt_regs *regs, long error_code)
1102 /* No need to warn about this any longer. */
1103 printk(KERN_INFO "Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
1107 unsigned long patch_espfix_desc(unsigned long uesp, unsigned long kesp)
1109 struct desc_struct *gdt = get_cpu_gdt_table(smp_processor_id());
1110 unsigned long base = (kesp - uesp) & -THREAD_SIZE;
1111 unsigned long new_kesp = kesp - base;
1112 unsigned long lim_pages = (new_kesp | (THREAD_SIZE - 1)) >> PAGE_SHIFT;
1113 __u64 desc = *(__u64 *)&gdt[GDT_ENTRY_ESPFIX_SS];
1115 /* Set up base for espfix segment */
1116 desc &= 0x00f0ff0000000000ULL;
1117 desc |= ((((__u64)base) << 16) & 0x000000ffffff0000ULL) |
1118 ((((__u64)base) << 32) & 0xff00000000000000ULL) |
1119 ((((__u64)lim_pages) << 32) & 0x000f000000000000ULL) |
1120 (lim_pages & 0xffff);
1121 *(__u64 *)&gdt[GDT_ENTRY_ESPFIX_SS] = desc;
1127 * 'math_state_restore()' saves the current math information in the
1128 * old math state array, and gets the new ones from the current task
1130 * Careful.. There are problems with IBM-designed IRQ13 behaviour.
1131 * Don't touch unless you *really* know how it works.
1133 * Must be called with kernel preemption disabled (in this case,
1134 * local interrupts are disabled at the call-site in entry.S).
1136 asmlinkage void math_state_restore(void)
1138 struct thread_info *thread = current_thread_info();
1139 struct task_struct *tsk = thread->task;
1141 if (!tsk_used_math(tsk)) {
1144 * does a slab alloc which can sleep
1146 if (init_fpu(tsk)) {
1148 * ran out of memory!
1150 do_group_exit(SIGKILL);
1153 local_irq_disable();
1156 clts(); /* Allow maths ops (or we recurse) */
1158 thread->status |= TS_USEDFPU; /* So we fnsave on switch_to() */
1161 EXPORT_SYMBOL_GPL(math_state_restore);
1163 #ifndef CONFIG_MATH_EMULATION
1165 asmlinkage void math_emulate(long arg)
1168 "math-emulation not enabled and no coprocessor found.\n");
1169 printk(KERN_EMERG "killing %s.\n", current->comm);
1170 force_sig(SIGFPE, current);
1174 #endif /* CONFIG_MATH_EMULATION */
1176 void __init trap_init(void)
1181 void __iomem *p = early_ioremap(0x0FFFD9, 4);
1183 if (readl(p) == 'E' + ('I'<<8) + ('S'<<16) + ('A'<<24))
1185 early_iounmap(p, 4);
1188 set_trap_gate(0, ÷_error);
1189 set_intr_gate(1, &debug);
1190 set_intr_gate(2, &nmi);
1191 set_system_intr_gate(3, &int3); /* int3 can be called from all */
1192 set_system_gate(4, &overflow); /* int4 can be called from all */
1193 set_trap_gate(5, &bounds);
1194 set_trap_gate(6, &invalid_op);
1195 set_trap_gate(7, &device_not_available);
1196 set_task_gate(8, GDT_ENTRY_DOUBLEFAULT_TSS);
1197 set_trap_gate(9, &coprocessor_segment_overrun);
1198 set_trap_gate(10, &invalid_TSS);
1199 set_trap_gate(11, &segment_not_present);
1200 set_trap_gate(12, &stack_segment);
1201 set_trap_gate(13, &general_protection);
1202 set_intr_gate(14, &page_fault);
1203 set_trap_gate(15, &spurious_interrupt_bug);
1204 set_trap_gate(16, &coprocessor_error);
1205 set_trap_gate(17, &alignment_check);
1206 #ifdef CONFIG_X86_MCE
1207 set_trap_gate(18, &machine_check);
1209 set_trap_gate(19, &simd_coprocessor_error);
1212 printk(KERN_INFO "Enabling fast FPU save and restore... ");
1213 set_in_cr4(X86_CR4_OSFXSR);
1218 "Enabling unmasked SIMD FPU exception support... ");
1219 set_in_cr4(X86_CR4_OSXMMEXCPT);
1223 set_system_gate(SYSCALL_VECTOR, &system_call);
1225 /* Reserve all the builtin and the syscall vector: */
1226 for (i = 0; i < FIRST_EXTERNAL_VECTOR; i++)
1227 set_bit(i, used_vectors);
1229 set_bit(SYSCALL_VECTOR, used_vectors);
1231 init_thread_xstate();
1233 * Should be a barrier for any external CPU state:
1240 static int __init kstack_setup(char *s)
1242 kstack_depth_to_print = simple_strtoul(s, NULL, 0);
1246 __setup("kstack=", kstack_setup);
1248 static int __init code_bytes_setup(char *s)
1250 code_bytes = simple_strtoul(s, NULL, 0);
1251 if (code_bytes > 8192)
1256 __setup("code_bytes=", code_bytes_setup);