1 /*
2 * Copyright (C) 1991, 1992 Linus Torvalds
3 * Copyright (C) 2000, 2001, 2002 Andi Kleen, SuSE Labs
4 *
5 * Pentium III FXSR, SSE support
6 * Gareth Hughes <gareth@valinux.com>, May 2000
7 */
8
9 /*
10 * Handle hardware traps and faults.
11 */
12 #include <linux/interrupt.h>
13 #include <linux/kallsyms.h>
14 #include <linux/spinlock.h>
15 #include <linux/kprobes.h>
16 #include <linux/uaccess.h>
17 #include <linux/utsname.h>
18 #include <linux/kdebug.h>
19 #include <linux/kernel.h>
20 #include <linux/module.h>
21 #include <linux/ptrace.h>
22 #include <linux/string.h>
23 #include <linux/delay.h>
24 #include <linux/errno.h>
25 #include <linux/kexec.h>
26 #include <linux/sched.h>
27 #include <linux/timer.h>
28 #include <linux/init.h>
29 #include <linux/bug.h>
30 #include <linux/nmi.h>
31 #include <linux/mm.h>
32 #include <linux/smp.h>
33 #include <linux/io.h>
34
35 #ifdef CONFIG_EISA
36 #include <linux/ioport.h>
37 #include <linux/eisa.h>
38 #endif
39
40 #ifdef CONFIG_MCA
41 #include <linux/mca.h>
42 #endif
43
44 #if defined(CONFIG_EDAC)
45 #include <linux/edac.h>
46 #endif
47
48 #include <asm/stacktrace.h>
49 #include <asm/processor.h>
50 #include <asm/debugreg.h>
51 #include <asm/atomic.h>
52 #include <asm/system.h>
53 #include <asm/traps.h>
54 #include <asm/desc.h>
55 #include <asm/i387.h>
56
57 #include <mach_traps.h>
58
59 #ifdef CONFIG_X86_64
60 #include <asm/pgalloc.h>
61 #include <asm/proto.h>
62 #include <asm/pda.h>
63 #else
64 #include <asm/processor-flags.h>
65 #include <asm/arch_hooks.h>
66 #include <asm/traps.h>
67
68 #include "cpu/mcheck/mce.h"
69
70 asmlinkage int system_call(void);
71
72 /* Do we ignore FPU interrupts ? */
73 char ignore_fpu_irq;
74
75 /*
76 * The IDT has to be page-aligned to simplify the Pentium
77 * F0 0F bug workaround.. We have a special link segment
78 * for this.
79 */
80 gate_desc idt_table[256]
81 __attribute__((__section__(".data.idt"))) = { { { { 0, 0 } } }, };
82 #endif
83
84 DECLARE_BITMAP(used_vectors, NR_VECTORS);
85 EXPORT_SYMBOL_GPL(used_vectors);
86
87 static int ignore_nmis;
88
conditional_sti(struct pt_regs * regs)89 static inline void conditional_sti(struct pt_regs *regs)
90 {
91 if (regs->flags & X86_EFLAGS_IF)
92 local_irq_enable();
93 }
94
preempt_conditional_sti(struct pt_regs * regs)95 static inline void preempt_conditional_sti(struct pt_regs *regs)
96 {
97 inc_preempt_count();
98 if (regs->flags & X86_EFLAGS_IF)
99 local_irq_enable();
100 }
101
conditional_cli(struct pt_regs * regs)102 static inline void conditional_cli(struct pt_regs *regs)
103 {
104 if (regs->flags & X86_EFLAGS_IF)
105 local_irq_disable();
106 }
107
preempt_conditional_cli(struct pt_regs * regs)108 static inline void preempt_conditional_cli(struct pt_regs *regs)
109 {
110 if (regs->flags & X86_EFLAGS_IF)
111 local_irq_disable();
112 dec_preempt_count();
113 }
114
115 #ifdef CONFIG_X86_32
116 static inline void
die_if_kernel(const char * str,struct pt_regs * regs,long err)117 die_if_kernel(const char *str, struct pt_regs *regs, long err)
118 {
119 if (!user_mode_vm(regs))
120 die(str, regs, err);
121 }
122
123 /*
124 * Perform the lazy TSS's I/O bitmap copy. If the TSS has an
125 * invalid offset set (the LAZY one) and the faulting thread has
126 * a valid I/O bitmap pointer, we copy the I/O bitmap in the TSS,
127 * we set the offset field correctly and return 1.
128 */
lazy_iobitmap_copy(void)129 static int lazy_iobitmap_copy(void)
130 {
131 struct thread_struct *thread;
132 struct tss_struct *tss;
133 int cpu;
134
135 cpu = get_cpu();
136 tss = &per_cpu(init_tss, cpu);
137 thread = ¤t->thread;
138
139 if (tss->x86_tss.io_bitmap_base == INVALID_IO_BITMAP_OFFSET_LAZY &&
140 thread->io_bitmap_ptr) {
141 memcpy(tss->io_bitmap, thread->io_bitmap_ptr,
142 thread->io_bitmap_max);
143 /*
144 * If the previously set map was extending to higher ports
145 * than the current one, pad extra space with 0xff (no access).
146 */
147 if (thread->io_bitmap_max < tss->io_bitmap_max) {
148 memset((char *) tss->io_bitmap +
149 thread->io_bitmap_max, 0xff,
150 tss->io_bitmap_max - thread->io_bitmap_max);
151 }
152 tss->io_bitmap_max = thread->io_bitmap_max;
153 tss->x86_tss.io_bitmap_base = IO_BITMAP_OFFSET;
154 tss->io_bitmap_owner = thread;
155 put_cpu();
156
157 return 1;
158 }
159 put_cpu();
160
161 return 0;
162 }
163 #endif
164
165 static void __kprobes
do_trap(int trapnr,int signr,char * str,struct pt_regs * regs,long error_code,siginfo_t * info)166 do_trap(int trapnr, int signr, char *str, struct pt_regs *regs,
167 long error_code, siginfo_t *info)
168 {
169 struct task_struct *tsk = current;
170
171 #ifdef CONFIG_X86_32
172 if (regs->flags & X86_VM_MASK) {
173 /*
174 * traps 0, 1, 3, 4, and 5 should be forwarded to vm86.
175 * On nmi (interrupt 2), do_trap should not be called.
176 */
177 if (trapnr < 6)
178 goto vm86_trap;
179 goto trap_signal;
180 }
181 #endif
182
183 if (!user_mode(regs))
184 goto kernel_trap;
185
186 #ifdef CONFIG_X86_32
187 trap_signal:
188 #endif
189 /*
190 * We want error_code and trap_no set for userspace faults and
191 * kernelspace faults which result in die(), but not
192 * kernelspace faults which are fixed up. die() gives the
193 * process no chance to handle the signal and notice the
194 * kernel fault information, so that won't result in polluting
195 * the information about previously queued, but not yet
196 * delivered, faults. See also do_general_protection below.
197 */
198 tsk->thread.error_code = error_code;
199 tsk->thread.trap_no = trapnr;
200
201 #ifdef CONFIG_X86_64
202 if (show_unhandled_signals && unhandled_signal(tsk, signr) &&
203 printk_ratelimit()) {
204 printk(KERN_INFO
205 "%s[%d] trap %s ip:%lx sp:%lx error:%lx",
206 tsk->comm, tsk->pid, str,
207 regs->ip, regs->sp, error_code);
208 print_vma_addr(" in ", regs->ip);
209 printk("\n");
210 }
211 #endif
212
213 if (info)
214 force_sig_info(signr, info, tsk);
215 else
216 force_sig(signr, tsk);
217 return;
218
219 kernel_trap:
220 if (!fixup_exception(regs)) {
221 tsk->thread.error_code = error_code;
222 tsk->thread.trap_no = trapnr;
223 die(str, regs, error_code);
224 }
225 return;
226
227 #ifdef CONFIG_X86_32
228 vm86_trap:
229 if (handle_vm86_trap((struct kernel_vm86_regs *) regs,
230 error_code, trapnr))
231 goto trap_signal;
232 return;
233 #endif
234 }
235
236 #define DO_ERROR(trapnr, signr, str, name) \
237 dotraplinkage void do_##name(struct pt_regs *regs, long error_code) \
238 { \
239 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
240 == NOTIFY_STOP) \
241 return; \
242 conditional_sti(regs); \
243 do_trap(trapnr, signr, str, regs, error_code, NULL); \
244 }
245
246 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
247 dotraplinkage void do_##name(struct pt_regs *regs, long error_code) \
248 { \
249 siginfo_t info; \
250 info.si_signo = signr; \
251 info.si_errno = 0; \
252 info.si_code = sicode; \
253 info.si_addr = (void __user *)siaddr; \
254 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
255 == NOTIFY_STOP) \
256 return; \
257 conditional_sti(regs); \
258 do_trap(trapnr, signr, str, regs, error_code, &info); \
259 }
260
261 DO_ERROR_INFO(0, SIGFPE, "divide error", divide_error, FPE_INTDIV, regs->ip)
262 DO_ERROR(4, SIGSEGV, "overflow", overflow)
263 DO_ERROR(5, SIGSEGV, "bounds", bounds)
264 DO_ERROR_INFO(6, SIGILL, "invalid opcode", invalid_op, ILL_ILLOPN, regs->ip)
265 DO_ERROR(9, SIGFPE, "coprocessor segment overrun", coprocessor_segment_overrun)
266 DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
267 DO_ERROR(11, SIGBUS, "segment not present", segment_not_present)
268 #ifdef CONFIG_X86_32
269 DO_ERROR(12, SIGBUS, "stack segment", stack_segment)
270 #endif
271 DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0)
272
273 #ifdef CONFIG_X86_64
274 /* Runs on IST stack */
do_stack_segment(struct pt_regs * regs,long error_code)275 dotraplinkage void do_stack_segment(struct pt_regs *regs, long error_code)
276 {
277 if (notify_die(DIE_TRAP, "stack segment", regs, error_code,
278 12, SIGBUS) == NOTIFY_STOP)
279 return;
280 preempt_conditional_sti(regs);
281 do_trap(12, SIGBUS, "stack segment", regs, error_code, NULL);
282 preempt_conditional_cli(regs);
283 }
284
do_double_fault(struct pt_regs * regs,long error_code)285 dotraplinkage void do_double_fault(struct pt_regs *regs, long error_code)
286 {
287 static const char str[] = "double fault";
288 struct task_struct *tsk = current;
289
290 /* Return not checked because double check cannot be ignored */
291 notify_die(DIE_TRAP, str, regs, error_code, 8, SIGSEGV);
292
293 tsk->thread.error_code = error_code;
294 tsk->thread.trap_no = 8;
295
296 /*
297 * This is always a kernel trap and never fixable (and thus must
298 * never return).
299 */
300 for (;;)
301 die(str, regs, error_code);
302 }
303 #endif
304
305 dotraplinkage void __kprobes
do_general_protection(struct pt_regs * regs,long error_code)306 do_general_protection(struct pt_regs *regs, long error_code)
307 {
308 struct task_struct *tsk;
309
310 conditional_sti(regs);
311
312 #ifdef CONFIG_X86_32
313 if (lazy_iobitmap_copy()) {
314 /* restart the faulting instruction */
315 return;
316 }
317
318 if (regs->flags & X86_VM_MASK)
319 goto gp_in_vm86;
320 #endif
321
322 tsk = current;
323 if (!user_mode(regs))
324 goto gp_in_kernel;
325
326 tsk->thread.error_code = error_code;
327 tsk->thread.trap_no = 13;
328
329 if (show_unhandled_signals && unhandled_signal(tsk, SIGSEGV) &&
330 printk_ratelimit()) {
331 printk(KERN_INFO
332 "%s[%d] general protection ip:%lx sp:%lx error:%lx",
333 tsk->comm, task_pid_nr(tsk),
334 regs->ip, regs->sp, error_code);
335 print_vma_addr(" in ", regs->ip);
336 printk("\n");
337 }
338
339 force_sig(SIGSEGV, tsk);
340 return;
341
342 #ifdef CONFIG_X86_32
343 gp_in_vm86:
344 local_irq_enable();
345 handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
346 return;
347 #endif
348
349 gp_in_kernel:
350 if (fixup_exception(regs))
351 return;
352
353 tsk->thread.error_code = error_code;
354 tsk->thread.trap_no = 13;
355 if (notify_die(DIE_GPF, "general protection fault", regs,
356 error_code, 13, SIGSEGV) == NOTIFY_STOP)
357 return;
358 die("general protection fault", regs, error_code);
359 }
360
361 static notrace __kprobes void
mem_parity_error(unsigned char reason,struct pt_regs * regs)362 mem_parity_error(unsigned char reason, struct pt_regs *regs)
363 {
364 printk(KERN_EMERG
365 "Uhhuh. NMI received for unknown reason %02x on CPU %d.\n",
366 reason, smp_processor_id());
367
368 printk(KERN_EMERG
369 "You have some hardware problem, likely on the PCI bus.\n");
370
371 #if defined(CONFIG_EDAC)
372 if (edac_handler_set()) {
373 edac_atomic_assert_error();
374 return;
375 }
376 #endif
377
378 if (panic_on_unrecovered_nmi)
379 panic("NMI: Not continuing");
380
381 printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
382
383 /* Clear and disable the memory parity error line. */
384 reason = (reason & 0xf) | 4;
385 outb(reason, 0x61);
386 }
387
388 static notrace __kprobes void
io_check_error(unsigned char reason,struct pt_regs * regs)389 io_check_error(unsigned char reason, struct pt_regs *regs)
390 {
391 unsigned long i;
392
393 printk(KERN_EMERG "NMI: IOCK error (debug interrupt?)\n");
394 show_registers(regs);
395
396 /* Re-enable the IOCK line, wait for a few seconds */
397 reason = (reason & 0xf) | 8;
398 outb(reason, 0x61);
399
400 i = 2000;
401 while (--i)
402 udelay(1000);
403
404 reason &= ~8;
405 outb(reason, 0x61);
406 }
407
408 static notrace __kprobes void
unknown_nmi_error(unsigned char reason,struct pt_regs * regs)409 unknown_nmi_error(unsigned char reason, struct pt_regs *regs)
410 {
411 if (notify_die(DIE_NMIUNKNOWN, "nmi", regs, reason, 2, SIGINT) ==
412 NOTIFY_STOP)
413 return;
414 #ifdef CONFIG_MCA
415 /*
416 * Might actually be able to figure out what the guilty party
417 * is:
418 */
419 if (MCA_bus) {
420 mca_handle_nmi();
421 return;
422 }
423 #endif
424 printk(KERN_EMERG
425 "Uhhuh. NMI received for unknown reason %02x on CPU %d.\n",
426 reason, smp_processor_id());
427
428 printk(KERN_EMERG "Do you have a strange power saving mode enabled?\n");
429 if (panic_on_unrecovered_nmi)
430 panic("NMI: Not continuing");
431
432 printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
433 }
434
default_do_nmi(struct pt_regs * regs)435 static notrace __kprobes void default_do_nmi(struct pt_regs *regs)
436 {
437 unsigned char reason = 0;
438 int cpu;
439
440 cpu = smp_processor_id();
441
442 /* Only the BSP gets external NMIs from the system. */
443 if (!cpu)
444 reason = get_nmi_reason();
445
446 if (!(reason & 0xc0)) {
447 if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 2, SIGINT)
448 == NOTIFY_STOP)
449 return;
450 #ifdef CONFIG_X86_LOCAL_APIC
451 /*
452 * Ok, so this is none of the documented NMI sources,
453 * so it must be the NMI watchdog.
454 */
455 if (nmi_watchdog_tick(regs, reason))
456 return;
457 if (!do_nmi_callback(regs, cpu))
458 unknown_nmi_error(reason, regs);
459 #else
460 unknown_nmi_error(reason, regs);
461 #endif
462
463 return;
464 }
465 if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP)
466 return;
467
468 /* AK: following checks seem to be broken on modern chipsets. FIXME */
469 if (reason & 0x80)
470 mem_parity_error(reason, regs);
471 if (reason & 0x40)
472 io_check_error(reason, regs);
473 #ifdef CONFIG_X86_32
474 /*
475 * Reassert NMI in case it became active meanwhile
476 * as it's edge-triggered:
477 */
478 reassert_nmi();
479 #endif
480 }
481
482 dotraplinkage notrace __kprobes void
do_nmi(struct pt_regs * regs,long error_code)483 do_nmi(struct pt_regs *regs, long error_code)
484 {
485 nmi_enter();
486
487 inc_irq_stat(__nmi_count);
488
489 if (!ignore_nmis)
490 default_do_nmi(regs);
491
492 nmi_exit();
493 }
494
stop_nmi(void)495 void stop_nmi(void)
496 {
497 acpi_nmi_disable();
498 ignore_nmis++;
499 }
500
restart_nmi(void)501 void restart_nmi(void)
502 {
503 ignore_nmis--;
504 acpi_nmi_enable();
505 }
506
507 /* May run on IST stack. */
do_int3(struct pt_regs * regs,long error_code)508 dotraplinkage void __kprobes do_int3(struct pt_regs *regs, long error_code)
509 {
510 #ifdef CONFIG_KPROBES
511 if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP)
512 == NOTIFY_STOP)
513 return;
514 #else
515 if (notify_die(DIE_TRAP, "int3", regs, error_code, 3, SIGTRAP)
516 == NOTIFY_STOP)
517 return;
518 #endif
519
520 preempt_conditional_sti(regs);
521 do_trap(3, SIGTRAP, "int3", regs, error_code, NULL);
522 preempt_conditional_cli(regs);
523 }
524
525 #ifdef CONFIG_X86_64
526 /*
527 * Help handler running on IST stack to switch back to user stack
528 * for scheduling or signal handling. The actual stack switch is done in
529 * entry.S
530 */
sync_regs(struct pt_regs * eregs)531 asmlinkage __kprobes struct pt_regs *sync_regs(struct pt_regs *eregs)
532 {
533 struct pt_regs *regs = eregs;
534 /* Did already sync */
535 if (eregs == (struct pt_regs *)eregs->sp)
536 ;
537 /* Exception from user space */
538 else if (user_mode(eregs))
539 regs = task_pt_regs(current);
540 /*
541 * Exception from kernel and interrupts are enabled. Move to
542 * kernel process stack.
543 */
544 else if (eregs->flags & X86_EFLAGS_IF)
545 regs = (struct pt_regs *)(eregs->sp -= sizeof(struct pt_regs));
546 if (eregs != regs)
547 *regs = *eregs;
548 return regs;
549 }
550 #endif
551
552 /*
553 * Our handling of the processor debug registers is non-trivial.
554 * We do not clear them on entry and exit from the kernel. Therefore
555 * it is possible to get a watchpoint trap here from inside the kernel.
556 * However, the code in ./ptrace.c has ensured that the user can
557 * only set watchpoints on userspace addresses. Therefore the in-kernel
558 * watchpoint trap can only occur in code which is reading/writing
559 * from user space. Such code must not hold kernel locks (since it
560 * can equally take a page fault), therefore it is safe to call
561 * force_sig_info even though that claims and releases locks.
562 *
563 * Code in ./signal.c ensures that the debug control register
564 * is restored before we deliver any signal, and therefore that
565 * user code runs with the correct debug control register even though
566 * we clear it here.
567 *
568 * Being careful here means that we don't have to be as careful in a
569 * lot of more complicated places (task switching can be a bit lazy
570 * about restoring all the debug state, and ptrace doesn't have to
571 * find every occurrence of the TF bit that could be saved away even
572 * by user code)
573 *
574 * May run on IST stack.
575 */
do_debug(struct pt_regs * regs,long error_code)576 dotraplinkage void __kprobes do_debug(struct pt_regs *regs, long error_code)
577 {
578 struct task_struct *tsk = current;
579 unsigned long condition;
580 int si_code;
581
582 get_debugreg(condition, 6);
583
584 /*
585 * The processor cleared BTF, so don't mark that we need it set.
586 */
587 clear_tsk_thread_flag(tsk, TIF_DEBUGCTLMSR);
588 tsk->thread.debugctlmsr = 0;
589
590 if (notify_die(DIE_DEBUG, "debug", regs, condition, error_code,
591 SIGTRAP) == NOTIFY_STOP)
592 return;
593
594 /* It's safe to allow irq's after DR6 has been saved */
595 preempt_conditional_sti(regs);
596
597 /* Mask out spurious debug traps due to lazy DR7 setting */
598 if (condition & (DR_TRAP0|DR_TRAP1|DR_TRAP2|DR_TRAP3)) {
599 if (!tsk->thread.debugreg7)
600 goto clear_dr7;
601 }
602
603 #ifdef CONFIG_X86_32
604 if (regs->flags & X86_VM_MASK)
605 goto debug_vm86;
606 #endif
607
608 /* Save debug status register where ptrace can see it */
609 tsk->thread.debugreg6 = condition;
610
611 /*
612 * Single-stepping through TF: make sure we ignore any events in
613 * kernel space (but re-enable TF when returning to user mode).
614 */
615 if (condition & DR_STEP) {
616 if (!user_mode(regs))
617 goto clear_TF_reenable;
618 }
619
620 si_code = get_si_code(condition);
621 /* Ok, finally something we can handle */
622 send_sigtrap(tsk, regs, error_code, si_code);
623
624 /*
625 * Disable additional traps. They'll be re-enabled when
626 * the signal is delivered.
627 */
628 clear_dr7:
629 set_debugreg(0, 7);
630 preempt_conditional_cli(regs);
631 return;
632
633 #ifdef CONFIG_X86_32
634 debug_vm86:
635 /* reenable preemption: handle_vm86_trap() might sleep */
636 dec_preempt_count();
637 handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code, 1);
638 conditional_cli(regs);
639 return;
640 #endif
641
642 clear_TF_reenable:
643 set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
644 regs->flags &= ~X86_EFLAGS_TF;
645 preempt_conditional_cli(regs);
646 return;
647 }
648
649 #ifdef CONFIG_X86_64
kernel_math_error(struct pt_regs * regs,const char * str,int trapnr)650 static int kernel_math_error(struct pt_regs *regs, const char *str, int trapnr)
651 {
652 if (fixup_exception(regs))
653 return 1;
654
655 notify_die(DIE_GPF, str, regs, 0, trapnr, SIGFPE);
656 /* Illegal floating point operation in the kernel */
657 current->thread.trap_no = trapnr;
658 die(str, regs, 0);
659 return 0;
660 }
661 #endif
662
663 /*
664 * Note that we play around with the 'TS' bit in an attempt to get
665 * the correct behaviour even in the presence of the asynchronous
666 * IRQ13 behaviour
667 */
math_error(void __user * ip)668 void math_error(void __user *ip)
669 {
670 struct task_struct *task;
671 siginfo_t info;
672 unsigned short cwd, swd, err;
673
674 /*
675 * Save the info for the exception handler and clear the error.
676 */
677 task = current;
678 save_init_fpu(task);
679 task->thread.trap_no = 16;
680 task->thread.error_code = 0;
681 info.si_signo = SIGFPE;
682 info.si_errno = 0;
683 info.si_addr = ip;
684 /*
685 * (~cwd & swd) will mask out exceptions that are not set to unmasked
686 * status. 0x3f is the exception bits in these regs, 0x200 is the
687 * C1 reg you need in case of a stack fault, 0x040 is the stack
688 * fault bit. We should only be taking one exception at a time,
689 * so if this combination doesn't produce any single exception,
690 * then we have a bad program that isn't synchronizing its FPU usage
691 * and it will suffer the consequences since we won't be able to
692 * fully reproduce the context of the exception
693 */
694 cwd = get_fpu_cwd(task);
695 swd = get_fpu_swd(task);
696
697 err = swd & ~cwd;
698
699 if (err & 0x001) { /* Invalid op */
700 /*
701 * swd & 0x240 == 0x040: Stack Underflow
702 * swd & 0x240 == 0x240: Stack Overflow
703 * User must clear the SF bit (0x40) if set
704 */
705 info.si_code = FPE_FLTINV;
706 } else if (err & 0x004) { /* Divide by Zero */
707 info.si_code = FPE_FLTDIV;
708 } else if (err & 0x008) { /* Overflow */
709 info.si_code = FPE_FLTOVF;
710 } else if (err & 0x012) { /* Denormal, Underflow */
711 info.si_code = FPE_FLTUND;
712 } else if (err & 0x020) { /* Precision */
713 info.si_code = FPE_FLTRES;
714 } else {
715 /*
716 * If we're using IRQ 13, or supposedly even some trap 16
717 * implementations, it's possible we get a spurious trap...
718 */
719 return; /* Spurious trap, no error */
720 }
721 force_sig_info(SIGFPE, &info, task);
722 }
723
do_coprocessor_error(struct pt_regs * regs,long error_code)724 dotraplinkage void do_coprocessor_error(struct pt_regs *regs, long error_code)
725 {
726 conditional_sti(regs);
727
728 #ifdef CONFIG_X86_32
729 ignore_fpu_irq = 1;
730 #else
731 if (!user_mode(regs) &&
732 kernel_math_error(regs, "kernel x87 math error", 16))
733 return;
734 #endif
735
736 math_error((void __user *)regs->ip);
737 }
738
simd_math_error(void __user * ip)739 static void simd_math_error(void __user *ip)
740 {
741 struct task_struct *task;
742 siginfo_t info;
743 unsigned short mxcsr;
744
745 /*
746 * Save the info for the exception handler and clear the error.
747 */
748 task = current;
749 save_init_fpu(task);
750 task->thread.trap_no = 19;
751 task->thread.error_code = 0;
752 info.si_signo = SIGFPE;
753 info.si_errno = 0;
754 info.si_code = __SI_FAULT;
755 info.si_addr = ip;
756 /*
757 * The SIMD FPU exceptions are handled a little differently, as there
758 * is only a single status/control register. Thus, to determine which
759 * unmasked exception was caught we must mask the exception mask bits
760 * at 0x1f80, and then use these to mask the exception bits at 0x3f.
761 */
762 mxcsr = get_fpu_mxcsr(task);
763 switch (~((mxcsr & 0x1f80) >> 7) & (mxcsr & 0x3f)) {
764 case 0x000:
765 default:
766 break;
767 case 0x001: /* Invalid Op */
768 info.si_code = FPE_FLTINV;
769 break;
770 case 0x002: /* Denormalize */
771 case 0x010: /* Underflow */
772 info.si_code = FPE_FLTUND;
773 break;
774 case 0x004: /* Zero Divide */
775 info.si_code = FPE_FLTDIV;
776 break;
777 case 0x008: /* Overflow */
778 info.si_code = FPE_FLTOVF;
779 break;
780 case 0x020: /* Precision */
781 info.si_code = FPE_FLTRES;
782 break;
783 }
784 force_sig_info(SIGFPE, &info, task);
785 }
786
787 dotraplinkage void
do_simd_coprocessor_error(struct pt_regs * regs,long error_code)788 do_simd_coprocessor_error(struct pt_regs *regs, long error_code)
789 {
790 conditional_sti(regs);
791
792 #ifdef CONFIG_X86_32
793 if (cpu_has_xmm) {
794 /* Handle SIMD FPU exceptions on PIII+ processors. */
795 ignore_fpu_irq = 1;
796 simd_math_error((void __user *)regs->ip);
797 return;
798 }
799 /*
800 * Handle strange cache flush from user space exception
801 * in all other cases. This is undocumented behaviour.
802 */
803 if (regs->flags & X86_VM_MASK) {
804 handle_vm86_fault((struct kernel_vm86_regs *)regs, error_code);
805 return;
806 }
807 current->thread.trap_no = 19;
808 current->thread.error_code = error_code;
809 die_if_kernel("cache flush denied", regs, error_code);
810 force_sig(SIGSEGV, current);
811 #else
812 if (!user_mode(regs) &&
813 kernel_math_error(regs, "kernel simd math error", 19))
814 return;
815 simd_math_error((void __user *)regs->ip);
816 #endif
817 }
818
819 dotraplinkage void
do_spurious_interrupt_bug(struct pt_regs * regs,long error_code)820 do_spurious_interrupt_bug(struct pt_regs *regs, long error_code)
821 {
822 conditional_sti(regs);
823 #if 0
824 /* No need to warn about this any longer. */
825 printk(KERN_INFO "Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
826 #endif
827 }
828
829 #ifdef CONFIG_X86_32
patch_espfix_desc(unsigned long uesp,unsigned long kesp)830 unsigned long patch_espfix_desc(unsigned long uesp, unsigned long kesp)
831 {
832 struct desc_struct *gdt = get_cpu_gdt_table(smp_processor_id());
833 unsigned long base = (kesp - uesp) & -THREAD_SIZE;
834 unsigned long new_kesp = kesp - base;
835 unsigned long lim_pages = (new_kesp | (THREAD_SIZE - 1)) >> PAGE_SHIFT;
836 __u64 desc = *(__u64 *)&gdt[GDT_ENTRY_ESPFIX_SS];
837
838 /* Set up base for espfix segment */
839 desc &= 0x00f0ff0000000000ULL;
840 desc |= ((((__u64)base) << 16) & 0x000000ffffff0000ULL) |
841 ((((__u64)base) << 32) & 0xff00000000000000ULL) |
842 ((((__u64)lim_pages) << 32) & 0x000f000000000000ULL) |
843 (lim_pages & 0xffff);
844 *(__u64 *)&gdt[GDT_ENTRY_ESPFIX_SS] = desc;
845
846 return new_kesp;
847 }
848 #else
smp_thermal_interrupt(void)849 asmlinkage void __attribute__((weak)) smp_thermal_interrupt(void)
850 {
851 }
852
mce_threshold_interrupt(void)853 asmlinkage void __attribute__((weak)) mce_threshold_interrupt(void)
854 {
855 }
856 #endif
857
858 /*
859 * 'math_state_restore()' saves the current math information in the
860 * old math state array, and gets the new ones from the current task
861 *
862 * Careful.. There are problems with IBM-designed IRQ13 behaviour.
863 * Don't touch unless you *really* know how it works.
864 *
865 * Must be called with kernel preemption disabled (in this case,
866 * local interrupts are disabled at the call-site in entry.S).
867 */
math_state_restore(void)868 asmlinkage void math_state_restore(void)
869 {
870 struct thread_info *thread = current_thread_info();
871 struct task_struct *tsk = thread->task;
872
873 if (!tsk_used_math(tsk)) {
874 local_irq_enable();
875 /*
876 * does a slab alloc which can sleep
877 */
878 if (init_fpu(tsk)) {
879 /*
880 * ran out of memory!
881 */
882 do_group_exit(SIGKILL);
883 return;
884 }
885 local_irq_disable();
886 }
887
888 clts(); /* Allow maths ops (or we recurse) */
889 #ifdef CONFIG_X86_32
890 restore_fpu(tsk);
891 #else
892 /*
893 * Paranoid restore. send a SIGSEGV if we fail to restore the state.
894 */
895 if (unlikely(restore_fpu_checking(tsk))) {
896 stts();
897 force_sig(SIGSEGV, tsk);
898 return;
899 }
900 #endif
901 thread->status |= TS_USEDFPU; /* So we fnsave on switch_to() */
902 tsk->fpu_counter++;
903 }
904 EXPORT_SYMBOL_GPL(math_state_restore);
905
906 #ifndef CONFIG_MATH_EMULATION
math_emulate(struct math_emu_info * info)907 void math_emulate(struct math_emu_info *info)
908 {
909 printk(KERN_EMERG
910 "math-emulation not enabled and no coprocessor found.\n");
911 printk(KERN_EMERG "killing %s.\n", current->comm);
912 force_sig(SIGFPE, current);
913 schedule();
914 }
915 #endif /* CONFIG_MATH_EMULATION */
916
do_device_not_available(struct pt_regs regs)917 dotraplinkage void __kprobes do_device_not_available(struct pt_regs regs)
918 {
919 #ifdef CONFIG_X86_32
920 if (read_cr0() & X86_CR0_EM) {
921 struct math_emu_info info = { };
922
923 conditional_sti(®s);
924
925 info.regs = ®s;
926 math_emulate(&info);
927 } else {
928 math_state_restore(); /* interrupts still off */
929 conditional_sti(®s);
930 }
931 #else
932 math_state_restore();
933 #endif
934 }
935
936 #ifdef CONFIG_X86_32
do_iret_error(struct pt_regs * regs,long error_code)937 dotraplinkage void do_iret_error(struct pt_regs *regs, long error_code)
938 {
939 siginfo_t info;
940 local_irq_enable();
941
942 info.si_signo = SIGILL;
943 info.si_errno = 0;
944 info.si_code = ILL_BADSTK;
945 info.si_addr = 0;
946 if (notify_die(DIE_TRAP, "iret exception",
947 regs, error_code, 32, SIGILL) == NOTIFY_STOP)
948 return;
949 do_trap(32, SIGILL, "iret exception", regs, error_code, &info);
950 }
951 #endif
952
trap_init(void)953 void __init trap_init(void)
954 {
955 int i;
956
957 #ifdef CONFIG_EISA
958 void __iomem *p = early_ioremap(0x0FFFD9, 4);
959
960 if (readl(p) == 'E' + ('I'<<8) + ('S'<<16) + ('A'<<24))
961 EISA_bus = 1;
962 early_iounmap(p, 4);
963 #endif
964
965 set_intr_gate(0, ÷_error);
966 set_intr_gate_ist(1, &debug, DEBUG_STACK);
967 set_intr_gate_ist(2, &nmi, NMI_STACK);
968 /* int3 can be called from all */
969 set_system_intr_gate_ist(3, &int3, DEBUG_STACK);
970 /* int4 can be called from all */
971 set_system_intr_gate(4, &overflow);
972 set_intr_gate(5, &bounds);
973 set_intr_gate(6, &invalid_op);
974 set_intr_gate(7, &device_not_available);
975 #ifdef CONFIG_X86_32
976 set_task_gate(8, GDT_ENTRY_DOUBLEFAULT_TSS);
977 #else
978 set_intr_gate_ist(8, &double_fault, DOUBLEFAULT_STACK);
979 #endif
980 set_intr_gate(9, &coprocessor_segment_overrun);
981 set_intr_gate(10, &invalid_TSS);
982 set_intr_gate(11, &segment_not_present);
983 set_intr_gate_ist(12, &stack_segment, STACKFAULT_STACK);
984 set_intr_gate(13, &general_protection);
985 set_intr_gate(14, &page_fault);
986 set_intr_gate(15, &spurious_interrupt_bug);
987 set_intr_gate(16, &coprocessor_error);
988 set_intr_gate(17, &alignment_check);
989 #ifdef CONFIG_X86_MCE
990 set_intr_gate_ist(18, &machine_check, MCE_STACK);
991 #endif
992 set_intr_gate(19, &simd_coprocessor_error);
993
994 #ifdef CONFIG_IA32_EMULATION
995 set_system_intr_gate(IA32_SYSCALL_VECTOR, ia32_syscall);
996 #endif
997
998 #ifdef CONFIG_X86_32
999 if (cpu_has_fxsr) {
1000 printk(KERN_INFO "Enabling fast FPU save and restore... ");
1001 set_in_cr4(X86_CR4_OSFXSR);
1002 printk("done.\n");
1003 }
1004 if (cpu_has_xmm) {
1005 printk(KERN_INFO
1006 "Enabling unmasked SIMD FPU exception support... ");
1007 set_in_cr4(X86_CR4_OSXMMEXCPT);
1008 printk("done.\n");
1009 }
1010
1011 set_system_trap_gate(SYSCALL_VECTOR, &system_call);
1012 #endif
1013
1014 /* Reserve all the builtin and the syscall vector: */
1015 for (i = 0; i < FIRST_EXTERNAL_VECTOR; i++)
1016 set_bit(i, used_vectors);
1017
1018 #ifdef CONFIG_X86_64
1019 set_bit(IA32_SYSCALL_VECTOR, used_vectors);
1020 #else
1021 set_bit(SYSCALL_VECTOR, used_vectors);
1022 #endif
1023 /*
1024 * Should be a barrier for any external CPU state:
1025 */
1026 cpu_init();
1027
1028 #ifdef CONFIG_X86_32
1029 trap_init_hook();
1030 #endif
1031 }
1032