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/moduleparam.h>
14 #include <linux/interrupt.h>
15 #include <linux/kallsyms.h>
16 #include <linux/spinlock.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>
36 #if defined(CONFIG_EDAC)
37 #include <linux/edac.h>
40 #include <asm/stacktrace.h>
41 #include <asm/processor.h>
42 #include <asm/debugreg.h>
43 #include <asm/atomic.h>
44 #include <asm/system.h>
45 #include <asm/unwind.h>
51 #include <asm/pgalloc.h>
52 #include <asm/proto.h>
54 #include <asm/traps.h>
56 #include <mach_traps.h>
58 int panic_on_unrecovered_nmi;
59 int kstack_depth_to_print = 12;
60 static unsigned int code_bytes = 64;
61 static int ignore_nmis;
62 static int die_counter;
64 static inline void conditional_sti(struct pt_regs *regs)
66 if (regs->flags & X86_EFLAGS_IF)
70 static inline void preempt_conditional_sti(struct pt_regs *regs)
73 if (regs->flags & X86_EFLAGS_IF)
77 static inline void preempt_conditional_cli(struct pt_regs *regs)
79 if (regs->flags & X86_EFLAGS_IF)
81 /* Make sure to not schedule here because we could be running
82 on an exception stack. */
86 void printk_address(unsigned long address, int reliable)
88 printk(" [<%016lx>] %s%pS\n", address, reliable ? "": "? ", (void *) address);
91 static unsigned long *in_exception_stack(unsigned cpu, unsigned long stack,
92 unsigned *usedp, char **idp)
94 static char ids[][8] = {
95 [DEBUG_STACK - 1] = "#DB",
96 [NMI_STACK - 1] = "NMI",
97 [DOUBLEFAULT_STACK - 1] = "#DF",
98 [STACKFAULT_STACK - 1] = "#SS",
99 [MCE_STACK - 1] = "#MC",
100 #if DEBUG_STKSZ > EXCEPTION_STKSZ
101 [N_EXCEPTION_STACKS ... N_EXCEPTION_STACKS + DEBUG_STKSZ / EXCEPTION_STKSZ - 2] = "#DB[?]"
107 * Iterate over all exception stacks, and figure out whether
108 * 'stack' is in one of them:
110 for (k = 0; k < N_EXCEPTION_STACKS; k++) {
111 unsigned long end = per_cpu(orig_ist, cpu).ist[k];
113 * Is 'stack' above this exception frame's end?
114 * If yes then skip to the next frame.
119 * Is 'stack' above this exception frame's start address?
120 * If yes then we found the right frame.
122 if (stack >= end - EXCEPTION_STKSZ) {
124 * Make sure we only iterate through an exception
125 * stack once. If it comes up for the second time
126 * then there's something wrong going on - just
127 * break out and return NULL:
129 if (*usedp & (1U << k))
133 return (unsigned long *)end;
136 * If this is a debug stack, and if it has a larger size than
137 * the usual exception stacks, then 'stack' might still
138 * be within the lower portion of the debug stack:
140 #if DEBUG_STKSZ > EXCEPTION_STKSZ
141 if (k == DEBUG_STACK - 1 && stack >= end - DEBUG_STKSZ) {
142 unsigned j = N_EXCEPTION_STACKS - 1;
145 * Black magic. A large debug stack is composed of
146 * multiple exception stack entries, which we
147 * iterate through now. Dont look:
151 end -= EXCEPTION_STKSZ;
152 ids[j][4] = '1' + (j - N_EXCEPTION_STACKS);
153 } while (stack < end - EXCEPTION_STKSZ);
154 if (*usedp & (1U << j))
158 return (unsigned long *)end;
166 * x86-64 can have up to three kernel stacks:
169 * severe exception (double fault, nmi, stack fault, debug, mce) hardware stack
172 static inline int valid_stack_ptr(struct thread_info *tinfo,
173 void *p, unsigned int size, void *end)
177 if (p < end && p >= (end-THREAD_SIZE))
182 return p > t && p < t + THREAD_SIZE - size;
185 /* The form of the top of the frame on the stack */
187 struct stack_frame *next_frame;
188 unsigned long return_address;
191 static inline unsigned long
192 print_context_stack(struct thread_info *tinfo,
193 unsigned long *stack, unsigned long bp,
194 const struct stacktrace_ops *ops, void *data,
197 struct stack_frame *frame = (struct stack_frame *)bp;
199 while (valid_stack_ptr(tinfo, stack, sizeof(*stack), end)) {
203 if (__kernel_text_address(addr)) {
204 if ((unsigned long) stack == bp + 8) {
205 ops->address(data, addr, 1);
206 frame = frame->next_frame;
207 bp = (unsigned long) frame;
209 ops->address(data, addr, bp == 0);
217 void dump_trace(struct task_struct *task, struct pt_regs *regs,
218 unsigned long *stack, unsigned long bp,
219 const struct stacktrace_ops *ops, void *data)
221 const unsigned cpu = get_cpu();
222 unsigned long *irqstack_end = (unsigned long*)cpu_pda(cpu)->irqstackptr;
224 struct thread_info *tinfo;
232 if (task && task != current)
233 stack = (unsigned long *)task->thread.sp;
236 #ifdef CONFIG_FRAME_POINTER
238 if (task == current) {
239 /* Grab bp right from our regs */
240 asm("movq %%rbp, %0" : "=r" (bp) :);
242 /* bp is the last reg pushed by switch_to */
243 bp = *(unsigned long *) task->thread.sp;
249 * Print function call entries in all stacks, starting at the
250 * current stack address. If the stacks consist of nested
253 tinfo = task_thread_info(task);
256 unsigned long *estack_end;
257 estack_end = in_exception_stack(cpu, (unsigned long)stack,
261 if (ops->stack(data, id) < 0)
264 bp = print_context_stack(tinfo, stack, bp, ops,
266 ops->stack(data, "<EOE>");
268 * We link to the next stack via the
269 * second-to-last pointer (index -2 to end) in the
272 stack = (unsigned long *) estack_end[-2];
276 unsigned long *irqstack;
277 irqstack = irqstack_end -
278 (IRQSTACKSIZE - 64) / sizeof(*irqstack);
280 if (stack >= irqstack && stack < irqstack_end) {
281 if (ops->stack(data, "IRQ") < 0)
283 bp = print_context_stack(tinfo, stack, bp,
284 ops, data, irqstack_end);
286 * We link to the next stack (which would be
287 * the process stack normally) the last
288 * pointer (index -1 to end) in the IRQ stack:
290 stack = (unsigned long *) (irqstack_end[-1]);
292 ops->stack(data, "EOI");
300 * This handles the process stack:
302 bp = print_context_stack(tinfo, stack, bp, ops, data, NULL);
305 EXPORT_SYMBOL(dump_trace);
308 print_trace_warning_symbol(void *data, char *msg, unsigned long symbol)
310 print_symbol(msg, symbol);
314 static void print_trace_warning(void *data, char *msg)
319 static int print_trace_stack(void *data, char *name)
321 printk(" <%s> ", name);
325 static void print_trace_address(void *data, unsigned long addr, int reliable)
327 touch_nmi_watchdog();
328 printk_address(addr, reliable);
331 static const struct stacktrace_ops print_trace_ops = {
332 .warning = print_trace_warning,
333 .warning_symbol = print_trace_warning_symbol,
334 .stack = print_trace_stack,
335 .address = print_trace_address,
339 show_trace_log_lvl(struct task_struct *task, struct pt_regs *regs,
340 unsigned long *stack, unsigned long bp, char *log_lvl)
342 printk("\nCall Trace:\n");
343 dump_trace(task, regs, stack, bp, &print_trace_ops, log_lvl);
347 void show_trace(struct task_struct *task, struct pt_regs *regs,
348 unsigned long *stack, unsigned long bp)
350 show_trace_log_lvl(task, regs, stack, bp, "");
354 show_stack_log_lvl(struct task_struct *task, struct pt_regs *regs,
355 unsigned long *sp, unsigned long bp, char *log_lvl)
357 unsigned long *stack;
359 const int cpu = smp_processor_id();
360 unsigned long *irqstack_end = (unsigned long *) (cpu_pda(cpu)->irqstackptr);
361 unsigned long *irqstack = (unsigned long *) (cpu_pda(cpu)->irqstackptr - IRQSTACKSIZE);
363 // debugging aid: "show_stack(NULL, NULL);" prints the
364 // back trace for this cpu.
368 sp = (unsigned long *)task->thread.sp;
370 sp = (unsigned long *)&sp;
374 for (i = 0; i < kstack_depth_to_print; i++) {
375 if (stack >= irqstack && stack <= irqstack_end) {
376 if (stack == irqstack_end) {
377 stack = (unsigned long *) (irqstack_end[-1]);
381 if (((long) stack & (THREAD_SIZE-1)) == 0)
384 if (i && ((i % 4) == 0))
386 printk(" %016lx", *stack++);
387 touch_nmi_watchdog();
389 show_trace_log_lvl(task, regs, sp, bp, log_lvl);
392 void show_stack(struct task_struct *task, unsigned long *sp)
394 show_stack_log_lvl(task, NULL, sp, 0, "");
398 * The architecture-independent dump_stack generator
400 void dump_stack(void)
402 unsigned long bp = 0;
405 #ifdef CONFIG_FRAME_POINTER
407 asm("movq %%rbp, %0" : "=r" (bp):);
410 printk("Pid: %d, comm: %.20s %s %s %.*s\n",
411 current->pid, current->comm, print_tainted(),
412 init_utsname()->release,
413 (int)strcspn(init_utsname()->version, " "),
414 init_utsname()->version);
415 show_trace(NULL, NULL, &stack, bp);
418 EXPORT_SYMBOL(dump_stack);
420 void show_registers(struct pt_regs *regs)
424 const int cpu = smp_processor_id();
425 struct task_struct *cur = cpu_pda(cpu)->pcurrent;
428 printk("CPU %d ", cpu);
430 printk("Process %s (pid: %d, threadinfo %p, task %p)\n",
431 cur->comm, cur->pid, task_thread_info(cur), cur);
434 * When in-kernel, we also print out the stack and code at the
435 * time of the fault..
437 if (!user_mode(regs)) {
438 unsigned int code_prologue = code_bytes * 43 / 64;
439 unsigned int code_len = code_bytes;
444 show_stack_log_lvl(NULL, regs, (unsigned long *)sp,
448 printk(KERN_EMERG "Code: ");
450 ip = (u8 *)regs->ip - code_prologue;
451 if (ip < (u8 *)PAGE_OFFSET || probe_kernel_address(ip, c)) {
452 /* try starting at RIP */
454 code_len = code_len - code_prologue + 1;
456 for (i = 0; i < code_len; i++, ip++) {
457 if (ip < (u8 *)PAGE_OFFSET ||
458 probe_kernel_address(ip, c)) {
459 printk(" Bad RIP value.");
462 if (ip == (u8 *)regs->ip)
463 printk("<%02x> ", c);
471 int is_valid_bugaddr(unsigned long ip)
475 if (__copy_from_user(&ud2, (const void __user *) ip, sizeof(ud2)))
478 return ud2 == 0x0b0f;
481 static raw_spinlock_t die_lock = __RAW_SPIN_LOCK_UNLOCKED;
482 static int die_owner = -1;
483 static unsigned int die_nest_count;
485 unsigned __kprobes long oops_begin(void)
492 /* racy, but better than risking deadlock. */
493 raw_local_irq_save(flags);
494 cpu = smp_processor_id();
495 if (!__raw_spin_trylock(&die_lock)) {
496 if (cpu == die_owner)
497 /* nested oops. should stop eventually */;
499 __raw_spin_lock(&die_lock);
508 void __kprobes oops_end(unsigned long flags, struct pt_regs *regs, int signr)
514 /* Nest count reaches zero, release the lock. */
515 __raw_spin_unlock(&die_lock);
516 raw_local_irq_restore(flags);
522 panic("Fatal exception");
527 int __kprobes __die(const char *str, struct pt_regs *regs, long err)
529 printk(KERN_EMERG "%s: %04lx [%u] ", str, err & 0xffff, ++die_counter);
530 #ifdef CONFIG_PREEMPT
536 #ifdef CONFIG_DEBUG_PAGEALLOC
537 printk("DEBUG_PAGEALLOC");
540 if (notify_die(DIE_OOPS, str, regs, err,
541 current->thread.trap_no, SIGSEGV) == NOTIFY_STOP)
544 show_registers(regs);
545 add_taint(TAINT_DIE);
546 /* Executive summary in case the oops scrolled away */
547 printk(KERN_ALERT "RIP ");
548 printk_address(regs->ip, 1);
549 printk(" RSP <%016lx>\n", regs->sp);
550 if (kexec_should_crash(current))
555 void die(const char *str, struct pt_regs *regs, long err)
557 unsigned long flags = oops_begin();
559 if (!user_mode(regs))
560 report_bug(regs->ip, regs);
562 if (__die(str, regs, err))
564 oops_end(flags, regs, SIGSEGV);
567 notrace __kprobes void
568 die_nmi(char *str, struct pt_regs *regs, int do_panic)
572 if (notify_die(DIE_NMIWATCHDOG, str, regs, 0, 2, SIGINT) == NOTIFY_STOP)
575 flags = oops_begin();
577 * We are in trouble anyway, lets at least try
578 * to get a message out.
580 printk(KERN_EMERG "%s", str);
581 printk(" on CPU%d, ip %08lx, registers:\n",
582 smp_processor_id(), regs->ip);
583 show_registers(regs);
584 if (kexec_should_crash(current))
586 if (do_panic || panic_on_oops)
587 panic("Non maskable interrupt");
588 oops_end(flags, NULL, SIGBUS);
594 static void __kprobes
595 do_trap(int trapnr, int signr, char *str, struct pt_regs *regs,
596 long error_code, siginfo_t *info)
598 struct task_struct *tsk = current;
600 if (!user_mode(regs))
604 * We want error_code and trap_no set for userspace faults and
605 * kernelspace faults which result in die(), but not
606 * kernelspace faults which are fixed up. die() gives the
607 * process no chance to handle the signal and notice the
608 * kernel fault information, so that won't result in polluting
609 * the information about previously queued, but not yet
610 * delivered, faults. See also do_general_protection below.
612 tsk->thread.error_code = error_code;
613 tsk->thread.trap_no = trapnr;
615 if (show_unhandled_signals && unhandled_signal(tsk, signr) &&
616 printk_ratelimit()) {
618 "%s[%d] trap %s ip:%lx sp:%lx error:%lx",
619 tsk->comm, tsk->pid, str,
620 regs->ip, regs->sp, error_code);
621 print_vma_addr(" in ", regs->ip);
626 force_sig_info(signr, info, tsk);
628 force_sig(signr, tsk);
632 if (!fixup_exception(regs)) {
633 tsk->thread.error_code = error_code;
634 tsk->thread.trap_no = trapnr;
635 die(str, regs, error_code);
640 #define DO_ERROR(trapnr, signr, str, name) \
641 asmlinkage void do_##name(struct pt_regs * regs, long error_code) \
643 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
646 conditional_sti(regs); \
647 do_trap(trapnr, signr, str, regs, error_code, NULL); \
650 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
651 asmlinkage void do_##name(struct pt_regs * regs, long error_code) \
654 info.si_signo = signr; \
656 info.si_code = sicode; \
657 info.si_addr = (void __user *)siaddr; \
658 trace_hardirqs_fixup(); \
659 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
662 conditional_sti(regs); \
663 do_trap(trapnr, signr, str, regs, error_code, &info); \
666 DO_ERROR_INFO(0, SIGFPE, "divide error", divide_error, FPE_INTDIV, regs->ip)
667 DO_ERROR(4, SIGSEGV, "overflow", overflow)
668 DO_ERROR(5, SIGSEGV, "bounds", bounds)
669 DO_ERROR_INFO(6, SIGILL, "invalid opcode", invalid_op, ILL_ILLOPN, regs->ip)
670 DO_ERROR(9, SIGFPE, "coprocessor segment overrun", coprocessor_segment_overrun)
671 DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
672 DO_ERROR(11, SIGBUS, "segment not present", segment_not_present)
673 DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0)
675 /* Runs on IST stack */
676 asmlinkage void do_stack_segment(struct pt_regs *regs, long error_code)
678 if (notify_die(DIE_TRAP, "stack segment", regs, error_code,
679 12, SIGBUS) == NOTIFY_STOP)
681 preempt_conditional_sti(regs);
682 do_trap(12, SIGBUS, "stack segment", regs, error_code, NULL);
683 preempt_conditional_cli(regs);
686 asmlinkage void do_double_fault(struct pt_regs * regs, long error_code)
688 static const char str[] = "double fault";
689 struct task_struct *tsk = current;
691 /* Return not checked because double check cannot be ignored */
692 notify_die(DIE_TRAP, str, regs, error_code, 8, SIGSEGV);
694 tsk->thread.error_code = error_code;
695 tsk->thread.trap_no = 8;
697 /* This is always a kernel trap and never fixable (and thus must
700 die(str, regs, error_code);
703 asmlinkage void __kprobes
704 do_general_protection(struct pt_regs *regs, long error_code)
706 struct task_struct *tsk;
708 conditional_sti(regs);
711 if (!user_mode(regs))
714 tsk->thread.error_code = error_code;
715 tsk->thread.trap_no = 13;
717 if (show_unhandled_signals && unhandled_signal(tsk, SIGSEGV) &&
718 printk_ratelimit()) {
720 "%s[%d] general protection ip:%lx sp:%lx error:%lx",
722 regs->ip, regs->sp, error_code);
723 print_vma_addr(" in ", regs->ip);
727 force_sig(SIGSEGV, tsk);
731 if (fixup_exception(regs))
734 tsk->thread.error_code = error_code;
735 tsk->thread.trap_no = 13;
736 if (notify_die(DIE_GPF, "general protection fault", regs,
737 error_code, 13, SIGSEGV) == NOTIFY_STOP)
739 die("general protection fault", regs, error_code);
742 static notrace __kprobes void
743 mem_parity_error(unsigned char reason, struct pt_regs *regs)
745 printk(KERN_EMERG "Uhhuh. NMI received for unknown reason %02x.\n",
747 printk(KERN_EMERG "You have some hardware problem, likely on the PCI bus.\n");
749 #if defined(CONFIG_EDAC)
750 if (edac_handler_set()) {
751 edac_atomic_assert_error();
756 if (panic_on_unrecovered_nmi)
757 panic("NMI: Not continuing");
759 printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
761 /* Clear and disable the memory parity error line. */
762 reason = (reason & 0xf) | 4;
766 static notrace __kprobes void
767 io_check_error(unsigned char reason, struct pt_regs *regs)
769 printk("NMI: IOCK error (debug interrupt?)\n");
770 show_registers(regs);
772 /* Re-enable the IOCK line, wait for a few seconds */
773 reason = (reason & 0xf) | 8;
780 static notrace __kprobes void
781 unknown_nmi_error(unsigned char reason, struct pt_regs * regs)
783 if (notify_die(DIE_NMIUNKNOWN, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP)
785 printk(KERN_EMERG "Uhhuh. NMI received for unknown reason %02x.\n",
787 printk(KERN_EMERG "Do you have a strange power saving mode enabled?\n");
789 if (panic_on_unrecovered_nmi)
790 panic("NMI: Not continuing");
792 printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
795 /* Runs on IST stack. This code must keep interrupts off all the time.
796 Nested NMIs are prevented by the CPU. */
797 asmlinkage notrace __kprobes void default_do_nmi(struct pt_regs *regs)
799 unsigned char reason = 0;
802 cpu = smp_processor_id();
804 /* Only the BSP gets external NMIs from the system. */
806 reason = get_nmi_reason();
808 if (!(reason & 0xc0)) {
809 if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 2, SIGINT)
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);
823 if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP)
826 /* AK: following checks seem to be broken on modern chipsets. FIXME */
828 mem_parity_error(reason, regs);
830 io_check_error(reason, regs);
833 asmlinkage notrace __kprobes void
834 do_nmi(struct pt_regs *regs, long error_code)
838 add_pda(__nmi_count, 1);
841 default_do_nmi(regs);
852 void restart_nmi(void)
858 /* runs on IST stack. */
859 asmlinkage void __kprobes do_int3(struct pt_regs *regs, long error_code)
861 trace_hardirqs_fixup();
863 if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP)
867 preempt_conditional_sti(regs);
868 do_trap(3, SIGTRAP, "int3", regs, error_code, NULL);
869 preempt_conditional_cli(regs);
872 /* Help handler running on IST stack to switch back to user stack
873 for scheduling or signal handling. The actual stack switch is done in
875 asmlinkage __kprobes struct pt_regs *sync_regs(struct pt_regs *eregs)
877 struct pt_regs *regs = eregs;
878 /* Did already sync */
879 if (eregs == (struct pt_regs *)eregs->sp)
881 /* Exception from user space */
882 else if (user_mode(eregs))
883 regs = task_pt_regs(current);
884 /* Exception from kernel and interrupts are enabled. Move to
885 kernel process stack. */
886 else if (eregs->flags & X86_EFLAGS_IF)
887 regs = (struct pt_regs *)(eregs->sp -= sizeof(struct pt_regs));
893 /* runs on IST stack. */
894 asmlinkage void __kprobes do_debug(struct pt_regs * regs,
895 unsigned long error_code)
897 struct task_struct *tsk = current;
898 unsigned long condition;
901 trace_hardirqs_fixup();
903 get_debugreg(condition, 6);
906 * The processor cleared BTF, so don't mark that we need it set.
908 clear_tsk_thread_flag(tsk, TIF_DEBUGCTLMSR);
909 tsk->thread.debugctlmsr = 0;
911 if (notify_die(DIE_DEBUG, "debug", regs, condition, error_code,
912 SIGTRAP) == NOTIFY_STOP)
915 preempt_conditional_sti(regs);
917 /* Mask out spurious debug traps due to lazy DR7 setting */
918 if (condition & (DR_TRAP0|DR_TRAP1|DR_TRAP2|DR_TRAP3)) {
919 if (!tsk->thread.debugreg7)
923 tsk->thread.debugreg6 = condition;
926 * Single-stepping through TF: make sure we ignore any events in
927 * kernel space (but re-enable TF when returning to user mode).
929 if (condition & DR_STEP) {
930 if (!user_mode(regs))
931 goto clear_TF_reenable;
934 /* Ok, finally something we can handle */
935 tsk->thread.trap_no = 1;
936 tsk->thread.error_code = error_code;
937 info.si_signo = SIGTRAP;
939 info.si_code = TRAP_BRKPT;
940 info.si_addr = user_mode(regs) ? (void __user *)regs->ip : NULL;
941 force_sig_info(SIGTRAP, &info, tsk);
945 preempt_conditional_cli(regs);
949 set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
950 regs->flags &= ~X86_EFLAGS_TF;
951 preempt_conditional_cli(regs);
955 static int kernel_math_error(struct pt_regs *regs, const char *str, int trapnr)
957 if (fixup_exception(regs))
960 notify_die(DIE_GPF, str, regs, 0, trapnr, SIGFPE);
961 /* Illegal floating point operation in the kernel */
962 current->thread.trap_no = trapnr;
968 * Note that we play around with the 'TS' bit in an attempt to get
969 * the correct behaviour even in the presence of the asynchronous
972 asmlinkage void do_coprocessor_error(struct pt_regs *regs)
974 void __user *ip = (void __user *)(regs->ip);
975 struct task_struct *task;
977 unsigned short cwd, swd;
979 conditional_sti(regs);
980 if (!user_mode(regs) &&
981 kernel_math_error(regs, "kernel x87 math error", 16))
985 * Save the info for the exception handler and clear the error.
989 task->thread.trap_no = 16;
990 task->thread.error_code = 0;
991 info.si_signo = SIGFPE;
993 info.si_code = __SI_FAULT;
996 * (~cwd & swd) will mask out exceptions that are not set to unmasked
997 * status. 0x3f is the exception bits in these regs, 0x200 is the
998 * C1 reg you need in case of a stack fault, 0x040 is the stack
999 * fault bit. We should only be taking one exception at a time,
1000 * so if this combination doesn't produce any single exception,
1001 * then we have a bad program that isn't synchronizing its FPU usage
1002 * and it will suffer the consequences since we won't be able to
1003 * fully reproduce the context of the exception
1005 cwd = get_fpu_cwd(task);
1006 swd = get_fpu_swd(task);
1007 switch (swd & ~cwd & 0x3f) {
1008 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 asmlinkage void bad_intr(void)
1038 printk("bad interrupt");
1041 asmlinkage void do_simd_coprocessor_error(struct pt_regs *regs)
1043 void __user *ip = (void __user *)(regs->ip);
1044 struct task_struct *task;
1046 unsigned short mxcsr;
1048 conditional_sti(regs);
1049 if (!user_mode(regs) &&
1050 kernel_math_error(regs, "kernel simd math error", 19))
1054 * Save the info for the exception handler and clear the error.
1057 save_init_fpu(task);
1058 task->thread.trap_no = 19;
1059 task->thread.error_code = 0;
1060 info.si_signo = SIGFPE;
1062 info.si_code = __SI_FAULT;
1065 * The SIMD FPU exceptions are handled a little differently, as there
1066 * is only a single status/control register. Thus, to determine which
1067 * unmasked exception was caught we must mask the exception mask bits
1068 * at 0x1f80, and then use these to mask the exception bits at 0x3f.
1070 mxcsr = get_fpu_mxcsr(task);
1071 switch (~((mxcsr & 0x1f80) >> 7) & (mxcsr & 0x3f)) {
1075 case 0x001: /* Invalid Op */
1076 info.si_code = FPE_FLTINV;
1078 case 0x002: /* Denormalize */
1079 case 0x010: /* Underflow */
1080 info.si_code = FPE_FLTUND;
1082 case 0x004: /* Zero Divide */
1083 info.si_code = FPE_FLTDIV;
1085 case 0x008: /* Overflow */
1086 info.si_code = FPE_FLTOVF;
1088 case 0x020: /* Precision */
1089 info.si_code = FPE_FLTRES;
1092 force_sig_info(SIGFPE, &info, task);
1095 asmlinkage void do_spurious_interrupt_bug(struct pt_regs * regs)
1099 asmlinkage void __attribute__((weak)) smp_thermal_interrupt(void)
1103 asmlinkage void __attribute__((weak)) mce_threshold_interrupt(void)
1108 * 'math_state_restore()' saves the current math information in the
1109 * old math state array, and gets the new ones from the current task
1111 * Careful.. There are problems with IBM-designed IRQ13 behaviour.
1112 * Don't touch unless you *really* know how it works.
1114 asmlinkage void math_state_restore(void)
1116 struct task_struct *me = current;
1121 * does a slab alloc which can sleep
1125 * ran out of memory!
1127 do_group_exit(SIGKILL);
1130 local_irq_disable();
1133 clts(); /* Allow maths ops (or we recurse) */
1135 * Paranoid restore. send a SIGSEGV if we fail to restore the state.
1137 if (unlikely(restore_fpu_checking(me))) {
1139 force_sig(SIGSEGV, me);
1142 task_thread_info(me)->status |= TS_USEDFPU;
1145 EXPORT_SYMBOL_GPL(math_state_restore);
1147 void __init trap_init(void)
1149 set_intr_gate(0, ÷_error);
1150 set_intr_gate_ist(1, &debug, DEBUG_STACK);
1151 set_intr_gate_ist(2, &nmi, NMI_STACK);
1152 set_system_gate_ist(3, &int3, DEBUG_STACK); /* int3 can be called from all */
1153 set_system_gate(4, &overflow); /* int4 can be called from all */
1154 set_intr_gate(5, &bounds);
1155 set_intr_gate(6, &invalid_op);
1156 set_intr_gate(7, &device_not_available);
1157 set_intr_gate_ist(8, &double_fault, DOUBLEFAULT_STACK);
1158 set_intr_gate(9, &coprocessor_segment_overrun);
1159 set_intr_gate(10, &invalid_TSS);
1160 set_intr_gate(11, &segment_not_present);
1161 set_intr_gate_ist(12, &stack_segment, STACKFAULT_STACK);
1162 set_intr_gate(13, &general_protection);
1163 set_intr_gate(14, &page_fault);
1164 set_intr_gate(15, &spurious_interrupt_bug);
1165 set_intr_gate(16, &coprocessor_error);
1166 set_intr_gate(17, &alignment_check);
1167 #ifdef CONFIG_X86_MCE
1168 set_intr_gate_ist(18, &machine_check, MCE_STACK);
1170 set_intr_gate(19, &simd_coprocessor_error);
1172 #ifdef CONFIG_IA32_EMULATION
1173 set_system_gate(IA32_SYSCALL_VECTOR, ia32_syscall);
1176 * Should be a barrier for any external CPU state:
1181 static int __init oops_setup(char *s)
1185 if (!strcmp(s, "panic"))
1189 early_param("oops", oops_setup);
1191 static int __init kstack_setup(char *s)
1195 kstack_depth_to_print = simple_strtoul(s, NULL, 0);
1198 early_param("kstack", kstack_setup);
1200 static int __init code_bytes_setup(char *s)
1202 code_bytes = simple_strtoul(s, NULL, 0);
1203 if (code_bytes > 8192)
1208 __setup("code_bytes=", code_bytes_setup);