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 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 ? */
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 void printk_address(unsigned long address, int reliable)
107 #ifdef CONFIG_KALLSYMS
108 char namebuf[KSYM_NAME_LEN];
109 unsigned long offset = 0;
110 unsigned long symsize;
116 symname = kallsyms_lookup(address, &symsize, &offset,
119 printk(" [<%08lx>]\n", address);
123 strcpy(reliab, "? ");
126 modname = delim = "";
127 printk(" [<%08lx>] %s%s%s%s%s+0x%lx/0x%lx\n",
128 address, reliab, delim, modname, delim, symname, offset, symsize);
130 printk(" [<%08lx>]\n", address);
134 static inline int valid_stack_ptr(struct thread_info *tinfo,
135 void *p, unsigned int size)
137 return p > (void *)tinfo &&
138 p <= (void *)tinfo + THREAD_SIZE - size;
141 /* The form of the top of the frame on the stack */
143 struct stack_frame *next_frame;
144 unsigned long return_address;
147 static inline unsigned long
148 print_context_stack(struct thread_info *tinfo,
149 unsigned long *stack, unsigned long bp,
150 const struct stacktrace_ops *ops, void *data)
152 struct stack_frame *frame = (struct stack_frame *)bp;
154 while (valid_stack_ptr(tinfo, stack, sizeof(*stack))) {
158 if (__kernel_text_address(addr)) {
159 if ((unsigned long) stack == bp + 4) {
160 ops->address(data, addr, 1);
161 frame = frame->next_frame;
162 bp = (unsigned long) frame;
164 ops->address(data, addr, bp == 0);
172 void dump_trace(struct task_struct *task, struct pt_regs *regs,
173 unsigned long *stack, unsigned long bp,
174 const struct stacktrace_ops *ops, void *data)
183 stack = (unsigned long *)task->thread.sp;
186 #ifdef CONFIG_FRAME_POINTER
188 if (task == current) {
189 /* Grab bp right from our regs */
190 asm("movl %%ebp, %0" : "=r" (bp) :);
192 /* bp is the last reg pushed by switch_to */
193 bp = *(unsigned long *) task->thread.sp;
199 struct thread_info *context;
201 context = (struct thread_info *)
202 ((unsigned long)stack & (~(THREAD_SIZE - 1)));
203 bp = print_context_stack(context, stack, bp, ops, data);
205 * Should be after the line below, but somewhere
206 * in early boot context comes out corrupted and we
207 * can't reference it:
209 if (ops->stack(data, "IRQ") < 0)
211 stack = (unsigned long *)context->previous_esp;
214 touch_nmi_watchdog();
217 EXPORT_SYMBOL(dump_trace);
220 print_trace_warning_symbol(void *data, char *msg, unsigned long symbol)
223 print_symbol(msg, symbol);
227 static void print_trace_warning(void *data, char *msg)
229 printk("%s%s\n", (char *)data, msg);
232 static int print_trace_stack(void *data, char *name)
238 * Print one address/symbol entries per line.
240 static void print_trace_address(void *data, unsigned long addr, int reliable)
242 printk("%s [<%08lx>] ", (char *)data, addr);
245 print_symbol("%s\n", addr);
246 touch_nmi_watchdog();
249 static const struct stacktrace_ops print_trace_ops = {
250 .warning = print_trace_warning,
251 .warning_symbol = print_trace_warning_symbol,
252 .stack = print_trace_stack,
253 .address = print_trace_address,
257 show_trace_log_lvl(struct task_struct *task, struct pt_regs *regs,
258 unsigned long *stack, unsigned long bp, char *log_lvl)
260 dump_trace(task, regs, stack, bp, &print_trace_ops, log_lvl);
261 printk("%s =======================\n", log_lvl);
264 void show_trace(struct task_struct *task, struct pt_regs *regs,
265 unsigned long *stack, unsigned long bp)
267 show_trace_log_lvl(task, regs, stack, bp, "");
271 show_stack_log_lvl(struct task_struct *task, struct pt_regs *regs,
272 unsigned long *sp, unsigned long bp, char *log_lvl)
274 unsigned long *stack;
279 sp = (unsigned long *)task->thread.sp;
281 sp = (unsigned long *)&sp;
285 for (i = 0; i < kstack_depth_to_print; i++) {
286 if (kstack_end(stack))
288 if (i && ((i % 8) == 0))
289 printk("\n%s ", log_lvl);
290 printk("%08lx ", *stack++);
292 printk("\n%sCall Trace:\n", log_lvl);
294 show_trace_log_lvl(task, regs, sp, bp, log_lvl);
297 void show_stack(struct task_struct *task, unsigned long *sp)
300 show_stack_log_lvl(task, NULL, sp, 0, "");
304 * The architecture-independent dump_stack generator
306 void dump_stack(void)
308 unsigned long bp = 0;
311 #ifdef CONFIG_FRAME_POINTER
313 asm("movl %%ebp, %0" : "=r" (bp):);
316 printk("Pid: %d, comm: %.20s %s %s %.*s\n",
317 current->pid, current->comm, print_tainted(),
318 init_utsname()->release,
319 (int)strcspn(init_utsname()->version, " "),
320 init_utsname()->version);
322 show_trace(current, NULL, &stack, bp);
325 EXPORT_SYMBOL(dump_stack);
327 void show_registers(struct pt_regs *regs)
332 __show_registers(regs, 0);
334 printk(KERN_EMERG "Process %.*s (pid: %d, ti=%p task=%p task.ti=%p)",
335 TASK_COMM_LEN, current->comm, task_pid_nr(current),
336 current_thread_info(), current, task_thread_info(current));
338 * When in-kernel, we also print out the stack and code at the
339 * time of the fault..
341 if (!user_mode_vm(regs)) {
342 unsigned int code_prologue = code_bytes * 43 / 64;
343 unsigned int code_len = code_bytes;
347 printk("\n" KERN_EMERG "Stack: ");
348 show_stack_log_lvl(NULL, regs, ®s->sp, 0, KERN_EMERG);
350 printk(KERN_EMERG "Code: ");
352 ip = (u8 *)regs->ip - code_prologue;
353 if (ip < (u8 *)PAGE_OFFSET || probe_kernel_address(ip, c)) {
354 /* try starting at EIP */
356 code_len = code_len - code_prologue + 1;
358 for (i = 0; i < code_len; i++, ip++) {
359 if (ip < (u8 *)PAGE_OFFSET ||
360 probe_kernel_address(ip, c)) {
361 printk(" Bad EIP value.");
364 if (ip == (u8 *)regs->ip)
365 printk("<%02x> ", c);
373 int is_valid_bugaddr(unsigned long ip)
377 if (ip < PAGE_OFFSET)
379 if (probe_kernel_address((unsigned short *)ip, ud2))
382 return ud2 == 0x0b0f;
385 static int die_counter;
387 int __kprobes __die(const char *str, struct pt_regs *regs, long err)
392 printk(KERN_EMERG "%s: %04lx [#%d] ", str, err & 0xffff, ++die_counter);
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) != NOTIFY_STOP) {
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);
426 * This is gone through when something in the kernel has done something bad
427 * and is about to be terminated:
429 void die(const char *str, struct pt_regs *regs, long err)
434 int lock_owner_depth;
436 .lock = __RAW_SPIN_LOCK_UNLOCKED,
438 .lock_owner_depth = 0
444 if (die.lock_owner != raw_smp_processor_id()) {
446 raw_local_irq_save(flags);
447 __raw_spin_lock(&die.lock);
448 die.lock_owner = smp_processor_id();
449 die.lock_owner_depth = 0;
452 raw_local_irq_save(flags);
455 if (++die.lock_owner_depth < 3) {
456 report_bug(regs->ip, regs);
458 if (__die(str, regs, err))
461 printk(KERN_EMERG "Recursive die() failure, output suppressed\n");
466 add_taint(TAINT_DIE);
467 __raw_spin_unlock(&die.lock);
468 raw_local_irq_restore(flags);
473 if (kexec_should_crash(current))
477 panic("Fatal exception in interrupt");
480 panic("Fatal exception");
487 die_if_kernel(const char *str, struct pt_regs *regs, long err)
489 if (!user_mode_vm(regs))
493 static void __kprobes
494 do_trap(int trapnr, int signr, char *str, int vm86, struct pt_regs *regs,
495 long error_code, siginfo_t *info)
497 struct task_struct *tsk = current;
499 if (regs->flags & X86_VM_MASK) {
505 if (!user_mode(regs))
510 * We want error_code and trap_no set for userspace faults and
511 * kernelspace faults which result in die(), but not
512 * kernelspace faults which are fixed up. die() gives the
513 * process no chance to handle the signal and notice the
514 * kernel fault information, so that won't result in polluting
515 * the information about previously queued, but not yet
516 * delivered, faults. See also do_general_protection below.
518 tsk->thread.error_code = error_code;
519 tsk->thread.trap_no = trapnr;
522 force_sig_info(signr, info, tsk);
524 force_sig(signr, tsk);
528 if (!fixup_exception(regs)) {
529 tsk->thread.error_code = error_code;
530 tsk->thread.trap_no = trapnr;
531 die(str, regs, error_code);
536 if (handle_vm86_trap((struct kernel_vm86_regs *) regs,
542 #define DO_ERROR(trapnr, signr, str, name) \
543 void do_##name(struct pt_regs *regs, long error_code) \
545 trace_hardirqs_fixup(); \
546 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
549 do_trap(trapnr, signr, str, 0, regs, error_code, NULL); \
552 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr, irq) \
553 void do_##name(struct pt_regs *regs, long error_code) \
557 local_irq_enable(); \
558 info.si_signo = signr; \
560 info.si_code = sicode; \
561 info.si_addr = (void __user *)siaddr; \
562 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
565 do_trap(trapnr, signr, str, 0, regs, error_code, &info); \
568 #define DO_VM86_ERROR(trapnr, signr, str, name) \
569 void do_##name(struct pt_regs *regs, long error_code) \
571 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
574 do_trap(trapnr, signr, str, 1, regs, error_code, NULL); \
577 #define DO_VM86_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
578 void do_##name(struct pt_regs *regs, long error_code) \
581 info.si_signo = signr; \
583 info.si_code = sicode; \
584 info.si_addr = (void __user *)siaddr; \
585 trace_hardirqs_fixup(); \
586 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
589 do_trap(trapnr, signr, str, 1, regs, error_code, &info); \
592 DO_VM86_ERROR_INFO(0, SIGFPE, "divide error", divide_error, FPE_INTDIV, regs->ip)
593 #ifndef CONFIG_KPROBES
594 DO_VM86_ERROR(3, SIGTRAP, "int3", int3)
596 DO_VM86_ERROR(4, SIGSEGV, "overflow", overflow)
597 DO_VM86_ERROR(5, SIGSEGV, "bounds", bounds)
598 DO_ERROR_INFO(6, SIGILL, "invalid opcode", invalid_op, ILL_ILLOPN, regs->ip, 0)
599 DO_ERROR(9, SIGFPE, "coprocessor segment overrun", coprocessor_segment_overrun)
600 DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
601 DO_ERROR(11, SIGBUS, "segment not present", segment_not_present)
602 DO_ERROR(12, SIGBUS, "stack segment", stack_segment)
603 DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0, 0)
604 DO_ERROR_INFO(32, SIGILL, "iret exception", iret_error, ILL_BADSTK, 0, 1)
606 void __kprobes do_general_protection(struct pt_regs *regs, long error_code)
608 struct thread_struct *thread;
609 struct tss_struct *tss;
613 tss = &per_cpu(init_tss, cpu);
614 thread = ¤t->thread;
617 * Perform the lazy TSS's I/O bitmap copy. If the TSS has an
618 * invalid offset set (the LAZY one) and the faulting thread has
619 * a valid I/O bitmap pointer, we copy the I/O bitmap in the TSS
620 * and we set the offset field correctly. Then we let the CPU to
621 * restart the faulting instruction.
623 if (tss->x86_tss.io_bitmap_base == INVALID_IO_BITMAP_OFFSET_LAZY &&
624 thread->io_bitmap_ptr) {
625 memcpy(tss->io_bitmap, thread->io_bitmap_ptr,
626 thread->io_bitmap_max);
628 * If the previously set map was extending to higher ports
629 * than the current one, pad extra space with 0xff (no access).
631 if (thread->io_bitmap_max < tss->io_bitmap_max) {
632 memset((char *) tss->io_bitmap +
633 thread->io_bitmap_max, 0xff,
634 tss->io_bitmap_max - thread->io_bitmap_max);
636 tss->io_bitmap_max = thread->io_bitmap_max;
637 tss->x86_tss.io_bitmap_base = IO_BITMAP_OFFSET;
638 tss->io_bitmap_owner = thread;
645 if (regs->flags & X86_VM_MASK)
648 if (!user_mode(regs))
651 current->thread.error_code = error_code;
652 current->thread.trap_no = 13;
654 if (show_unhandled_signals && unhandled_signal(current, SIGSEGV) &&
655 printk_ratelimit()) {
657 "%s[%d] general protection ip:%lx sp:%lx error:%lx",
658 current->comm, task_pid_nr(current),
659 regs->ip, regs->sp, error_code);
660 print_vma_addr(" in ", regs->ip);
664 force_sig(SIGSEGV, current);
669 handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
673 if (!fixup_exception(regs)) {
674 current->thread.error_code = error_code;
675 current->thread.trap_no = 13;
676 if (notify_die(DIE_GPF, "general protection fault", regs,
677 error_code, 13, SIGSEGV) == NOTIFY_STOP)
679 die("general protection fault", regs, error_code);
683 static notrace __kprobes void
684 mem_parity_error(unsigned char reason, struct pt_regs *regs)
687 "Uhhuh. NMI received for unknown reason %02x on CPU %d.\n",
688 reason, smp_processor_id());
691 "You have some hardware problem, likely on the PCI bus.\n");
693 #if defined(CONFIG_EDAC)
694 if (edac_handler_set()) {
695 edac_atomic_assert_error();
700 if (panic_on_unrecovered_nmi)
701 panic("NMI: Not continuing");
703 printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
705 /* Clear and disable the memory parity error line. */
706 clear_mem_error(reason);
709 static notrace __kprobes void
710 io_check_error(unsigned char reason, struct pt_regs *regs)
714 printk(KERN_EMERG "NMI: IOCK error (debug interrupt?)\n");
715 show_registers(regs);
717 /* Re-enable the IOCK line, wait for a few seconds */
718 reason = (reason & 0xf) | 8;
729 static notrace __kprobes void
730 unknown_nmi_error(unsigned char reason, struct pt_regs *regs)
732 if (notify_die(DIE_NMIUNKNOWN, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP)
736 * Might actually be able to figure out what the guilty party
745 "Uhhuh. NMI received for unknown reason %02x on CPU %d.\n",
746 reason, smp_processor_id());
748 printk(KERN_EMERG "Do you have a strange power saving mode enabled?\n");
749 if (panic_on_unrecovered_nmi)
750 panic("NMI: Not continuing");
752 printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
755 static DEFINE_SPINLOCK(nmi_print_lock);
757 void notrace __kprobes die_nmi(char *str, struct pt_regs *regs, int do_panic)
759 if (notify_die(DIE_NMIWATCHDOG, str, regs, 0, 2, SIGINT) == NOTIFY_STOP)
762 spin_lock(&nmi_print_lock);
764 * We are in trouble anyway, lets at least try
765 * to get a message out:
768 printk(KERN_EMERG "%s", str);
769 printk(" on CPU%d, ip %08lx, registers:\n",
770 smp_processor_id(), regs->ip);
771 show_registers(regs);
773 panic("Non maskable interrupt");
775 spin_unlock(&nmi_print_lock);
779 * If we are in kernel we are probably nested up pretty bad
780 * and might aswell get out now while we still can:
782 if (!user_mode_vm(regs)) {
783 current->thread.trap_no = 2;
790 static notrace __kprobes void default_do_nmi(struct pt_regs *regs)
792 unsigned char reason = 0;
794 /* Only the BSP gets external NMIs from the system: */
795 if (!smp_processor_id())
796 reason = get_nmi_reason();
798 if (!(reason & 0xc0)) {
799 if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 2, SIGINT)
802 #ifdef CONFIG_X86_LOCAL_APIC
804 * Ok, so this is none of the documented NMI sources,
805 * so it must be the NMI watchdog.
807 if (nmi_watchdog_tick(regs, reason))
809 if (!do_nmi_callback(regs, smp_processor_id()))
810 unknown_nmi_error(reason, regs);
812 unknown_nmi_error(reason, regs);
817 if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP)
820 /* AK: following checks seem to be broken on modern chipsets. FIXME */
822 mem_parity_error(reason, regs);
824 io_check_error(reason, regs);
826 * Reassert NMI in case it became active meanwhile
827 * as it's edge-triggered:
832 static int ignore_nmis;
834 notrace __kprobes void do_nmi(struct pt_regs *regs, long error_code)
840 cpu = smp_processor_id();
845 default_do_nmi(regs);
856 void restart_nmi(void)
862 #ifdef CONFIG_KPROBES
863 void __kprobes do_int3(struct pt_regs *regs, long error_code)
865 trace_hardirqs_fixup();
867 if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP)
871 * This is an interrupt gate, because kprobes wants interrupts
872 * disabled. Normal trap handlers don't.
874 restore_interrupts(regs);
876 do_trap(3, SIGTRAP, "int3", 1, regs, error_code, NULL);
881 * Our handling of the processor debug registers is non-trivial.
882 * We do not clear them on entry and exit from the kernel. Therefore
883 * it is possible to get a watchpoint trap here from inside the kernel.
884 * However, the code in ./ptrace.c has ensured that the user can
885 * only set watchpoints on userspace addresses. Therefore the in-kernel
886 * watchpoint trap can only occur in code which is reading/writing
887 * from user space. Such code must not hold kernel locks (since it
888 * can equally take a page fault), therefore it is safe to call
889 * force_sig_info even though that claims and releases locks.
891 * Code in ./signal.c ensures that the debug control register
892 * is restored before we deliver any signal, and therefore that
893 * user code runs with the correct debug control register even though
896 * Being careful here means that we don't have to be as careful in a
897 * lot of more complicated places (task switching can be a bit lazy
898 * about restoring all the debug state, and ptrace doesn't have to
899 * find every occurrence of the TF bit that could be saved away even
902 void __kprobes do_debug(struct pt_regs *regs, long error_code)
904 struct task_struct *tsk = current;
905 unsigned int condition;
907 trace_hardirqs_fixup();
909 get_debugreg(condition, 6);
912 * The processor cleared BTF, so don't mark that we need it set.
914 clear_tsk_thread_flag(tsk, TIF_DEBUGCTLMSR);
915 tsk->thread.debugctlmsr = 0;
917 if (notify_die(DIE_DEBUG, "debug", regs, condition, error_code,
918 SIGTRAP) == NOTIFY_STOP)
920 /* It's safe to allow irq's after DR6 has been saved */
921 if (regs->flags & X86_EFLAGS_IF)
924 /* Mask out spurious debug traps due to lazy DR7 setting */
925 if (condition & (DR_TRAP0|DR_TRAP1|DR_TRAP2|DR_TRAP3)) {
926 if (!tsk->thread.debugreg7)
930 if (regs->flags & X86_VM_MASK)
933 /* Save debug status register where ptrace can see it */
934 tsk->thread.debugreg6 = condition;
937 * Single-stepping through TF: make sure we ignore any events in
938 * kernel space (but re-enable TF when returning to user mode).
940 if (condition & DR_STEP) {
942 * We already checked v86 mode above, so we can
943 * check for kernel mode by just checking the CPL
946 if (!user_mode(regs))
947 goto clear_TF_reenable;
950 /* Ok, finally something we can handle */
951 send_sigtrap(tsk, regs, error_code);
954 * Disable additional traps. They'll be re-enabled when
955 * the signal is delivered.
962 handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code, 1);
966 set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
967 regs->flags &= ~X86_EFLAGS_TF;
972 * Note that we play around with the 'TS' bit in an attempt to get
973 * the correct behaviour even in the presence of the asynchronous
976 void math_error(void __user *ip)
978 struct task_struct *task;
984 * Save the info for the exception handler and clear the error.
988 task->thread.trap_no = 16;
989 task->thread.error_code = 0;
990 info.si_signo = SIGFPE;
992 info.si_code = __SI_FAULT;
995 * (~cwd & swd) will mask out exceptions that are not set to unmasked
996 * status. 0x3f is the exception bits in these regs, 0x200 is the
997 * C1 reg you need in case of a stack fault, 0x040 is the stack
998 * fault bit. We should only be taking one exception at a time,
999 * so if this combination doesn't produce any single exception,
1000 * then we have a bad program that isn't synchronizing its FPU usage
1001 * and it will suffer the consequences since we won't be able to
1002 * fully reproduce the context of the exception
1004 cwd = get_fpu_cwd(task);
1005 swd = get_fpu_swd(task);
1006 switch (swd & ~cwd & 0x3f) {
1007 case 0x000: /* No unmasked exception */
1009 default: /* Multiple exceptions */
1011 case 0x001: /* Invalid Op */
1013 * swd & 0x240 == 0x040: Stack Underflow
1014 * swd & 0x240 == 0x240: Stack Overflow
1015 * User must clear the SF bit (0x40) if set
1017 info.si_code = FPE_FLTINV;
1019 case 0x002: /* Denormalize */
1020 case 0x010: /* Underflow */
1021 info.si_code = FPE_FLTUND;
1023 case 0x004: /* Zero Divide */
1024 info.si_code = FPE_FLTDIV;
1026 case 0x008: /* Overflow */
1027 info.si_code = FPE_FLTOVF;
1029 case 0x020: /* Precision */
1030 info.si_code = FPE_FLTRES;
1033 force_sig_info(SIGFPE, &info, task);
1036 void do_coprocessor_error(struct pt_regs *regs, long error_code)
1039 math_error((void __user *)regs->ip);
1042 static void simd_math_error(void __user *ip)
1044 struct task_struct *task;
1045 unsigned short mxcsr;
1049 * Save the info for the exception handler and clear the error.
1052 save_init_fpu(task);
1053 task->thread.trap_no = 19;
1054 task->thread.error_code = 0;
1055 info.si_signo = SIGFPE;
1057 info.si_code = __SI_FAULT;
1060 * The SIMD FPU exceptions are handled a little differently, as there
1061 * is only a single status/control register. Thus, to determine which
1062 * unmasked exception was caught we must mask the exception mask bits
1063 * at 0x1f80, and then use these to mask the exception bits at 0x3f.
1065 mxcsr = get_fpu_mxcsr(task);
1066 switch (~((mxcsr & 0x1f80) >> 7) & (mxcsr & 0x3f)) {
1070 case 0x001: /* Invalid Op */
1071 info.si_code = FPE_FLTINV;
1073 case 0x002: /* Denormalize */
1074 case 0x010: /* Underflow */
1075 info.si_code = FPE_FLTUND;
1077 case 0x004: /* Zero Divide */
1078 info.si_code = FPE_FLTDIV;
1080 case 0x008: /* Overflow */
1081 info.si_code = FPE_FLTOVF;
1083 case 0x020: /* Precision */
1084 info.si_code = FPE_FLTRES;
1087 force_sig_info(SIGFPE, &info, task);
1090 void do_simd_coprocessor_error(struct pt_regs *regs, long error_code)
1093 /* Handle SIMD FPU exceptions on PIII+ processors. */
1095 simd_math_error((void __user *)regs->ip);
1099 * Handle strange cache flush from user space exception
1100 * in all other cases. This is undocumented behaviour.
1102 if (regs->flags & X86_VM_MASK) {
1103 handle_vm86_fault((struct kernel_vm86_regs *)regs, error_code);
1106 current->thread.trap_no = 19;
1107 current->thread.error_code = error_code;
1108 die_if_kernel("cache flush denied", regs, error_code);
1109 force_sig(SIGSEGV, current);
1112 void do_spurious_interrupt_bug(struct pt_regs *regs, long error_code)
1115 /* No need to warn about this any longer. */
1116 printk(KERN_INFO "Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
1120 unsigned long patch_espfix_desc(unsigned long uesp, unsigned long kesp)
1122 struct desc_struct *gdt = get_cpu_gdt_table(smp_processor_id());
1123 unsigned long base = (kesp - uesp) & -THREAD_SIZE;
1124 unsigned long new_kesp = kesp - base;
1125 unsigned long lim_pages = (new_kesp | (THREAD_SIZE - 1)) >> PAGE_SHIFT;
1126 __u64 desc = *(__u64 *)&gdt[GDT_ENTRY_ESPFIX_SS];
1128 /* Set up base for espfix segment */
1129 desc &= 0x00f0ff0000000000ULL;
1130 desc |= ((((__u64)base) << 16) & 0x000000ffffff0000ULL) |
1131 ((((__u64)base) << 32) & 0xff00000000000000ULL) |
1132 ((((__u64)lim_pages) << 32) & 0x000f000000000000ULL) |
1133 (lim_pages & 0xffff);
1134 *(__u64 *)&gdt[GDT_ENTRY_ESPFIX_SS] = desc;
1140 * 'math_state_restore()' saves the current math information in the
1141 * old math state array, and gets the new ones from the current task
1143 * Careful.. There are problems with IBM-designed IRQ13 behaviour.
1144 * Don't touch unless you *really* know how it works.
1146 * Must be called with kernel preemption disabled (in this case,
1147 * local interrupts are disabled at the call-site in entry.S).
1149 asmlinkage void math_state_restore(void)
1151 struct thread_info *thread = current_thread_info();
1152 struct task_struct *tsk = thread->task;
1154 if (!tsk_used_math(tsk)) {
1157 * does a slab alloc which can sleep
1159 if (init_fpu(tsk)) {
1161 * ran out of memory!
1163 do_group_exit(SIGKILL);
1166 local_irq_disable();
1169 clts(); /* Allow maths ops (or we recurse) */
1171 thread->status |= TS_USEDFPU; /* So we fnsave on switch_to() */
1174 EXPORT_SYMBOL_GPL(math_state_restore);
1176 #ifndef CONFIG_MATH_EMULATION
1178 asmlinkage void math_emulate(long arg)
1181 "math-emulation not enabled and no coprocessor found.\n");
1182 printk(KERN_EMERG "killing %s.\n", current->comm);
1183 force_sig(SIGFPE, current);
1187 #endif /* CONFIG_MATH_EMULATION */
1189 void __init trap_init(void)
1194 void __iomem *p = early_ioremap(0x0FFFD9, 4);
1196 if (readl(p) == 'E' + ('I'<<8) + ('S'<<16) + ('A'<<24))
1198 early_iounmap(p, 4);
1201 set_trap_gate(0, ÷_error);
1202 set_intr_gate(1, &debug);
1203 set_intr_gate(2, &nmi);
1204 set_system_intr_gate(3, &int3); /* int3 can be called from all */
1205 set_system_gate(4, &overflow); /* int4 can be called from all */
1206 set_trap_gate(5, &bounds);
1207 set_trap_gate(6, &invalid_op);
1208 set_trap_gate(7, &device_not_available);
1209 set_task_gate(8, GDT_ENTRY_DOUBLEFAULT_TSS);
1210 set_trap_gate(9, &coprocessor_segment_overrun);
1211 set_trap_gate(10, &invalid_TSS);
1212 set_trap_gate(11, &segment_not_present);
1213 set_trap_gate(12, &stack_segment);
1214 set_trap_gate(13, &general_protection);
1215 set_intr_gate(14, &page_fault);
1216 set_trap_gate(15, &spurious_interrupt_bug);
1217 set_trap_gate(16, &coprocessor_error);
1218 set_trap_gate(17, &alignment_check);
1219 #ifdef CONFIG_X86_MCE
1220 set_trap_gate(18, &machine_check);
1222 set_trap_gate(19, &simd_coprocessor_error);
1225 printk(KERN_INFO "Enabling fast FPU save and restore... ");
1226 set_in_cr4(X86_CR4_OSFXSR);
1231 "Enabling unmasked SIMD FPU exception support... ");
1232 set_in_cr4(X86_CR4_OSXMMEXCPT);
1236 set_system_gate(SYSCALL_VECTOR, &system_call);
1238 /* Reserve all the builtin and the syscall vector: */
1239 for (i = 0; i < FIRST_EXTERNAL_VECTOR; i++)
1240 set_bit(i, used_vectors);
1242 set_bit(SYSCALL_VECTOR, used_vectors);
1244 init_thread_xstate();
1246 * Should be a barrier for any external CPU state:
1253 static int __init kstack_setup(char *s)
1255 kstack_depth_to_print = simple_strtoul(s, NULL, 0);
1259 __setup("kstack=", kstack_setup);
1261 static int __init code_bytes_setup(char *s)
1263 code_bytes = simple_strtoul(s, NULL, 0);
1264 if (code_bytes > 8192)
1269 __setup("code_bytes=", code_bytes_setup);