[PATCH] x86_64: Disallow kprobes on NMI handlers
[deliverable/linux.git] / arch / x86_64 / kernel / traps.c
1 /*
2 * linux/arch/x86-64/traps.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 * Copyright (C) 2000, 2001, 2002 Andi Kleen, SuSE Labs
6 *
7 * Pentium III FXSR, SSE support
8 * Gareth Hughes <gareth@valinux.com>, May 2000
9 *
10 * $Id: traps.c,v 1.36 2002/03/24 11:09:10 ak Exp $
11 */
12
13 /*
14 * 'Traps.c' handles hardware traps and faults after we have saved some
15 * state in 'entry.S'.
16 */
17 #include <linux/config.h>
18 #include <linux/sched.h>
19 #include <linux/kernel.h>
20 #include <linux/string.h>
21 #include <linux/errno.h>
22 #include <linux/ptrace.h>
23 #include <linux/timer.h>
24 #include <linux/mm.h>
25 #include <linux/init.h>
26 #include <linux/delay.h>
27 #include <linux/spinlock.h>
28 #include <linux/interrupt.h>
29 #include <linux/module.h>
30 #include <linux/moduleparam.h>
31 #include <linux/nmi.h>
32 #include <linux/kprobes.h>
33
34 #include <asm/system.h>
35 #include <asm/uaccess.h>
36 #include <asm/io.h>
37 #include <asm/atomic.h>
38 #include <asm/debugreg.h>
39 #include <asm/desc.h>
40 #include <asm/i387.h>
41 #include <asm/kdebug.h>
42 #include <asm/processor.h>
43
44 #include <asm/smp.h>
45 #include <asm/pgalloc.h>
46 #include <asm/pda.h>
47 #include <asm/proto.h>
48 #include <asm/nmi.h>
49
50 extern struct gate_struct idt_table[256];
51
52 asmlinkage void divide_error(void);
53 asmlinkage void debug(void);
54 asmlinkage void nmi(void);
55 asmlinkage void int3(void);
56 asmlinkage void overflow(void);
57 asmlinkage void bounds(void);
58 asmlinkage void invalid_op(void);
59 asmlinkage void device_not_available(void);
60 asmlinkage void double_fault(void);
61 asmlinkage void coprocessor_segment_overrun(void);
62 asmlinkage void invalid_TSS(void);
63 asmlinkage void segment_not_present(void);
64 asmlinkage void stack_segment(void);
65 asmlinkage void general_protection(void);
66 asmlinkage void page_fault(void);
67 asmlinkage void coprocessor_error(void);
68 asmlinkage void simd_coprocessor_error(void);
69 asmlinkage void reserved(void);
70 asmlinkage void alignment_check(void);
71 asmlinkage void machine_check(void);
72 asmlinkage void spurious_interrupt_bug(void);
73
74 struct notifier_block *die_chain;
75 static DEFINE_SPINLOCK(die_notifier_lock);
76
77 int register_die_notifier(struct notifier_block *nb)
78 {
79 int err = 0;
80 unsigned long flags;
81 spin_lock_irqsave(&die_notifier_lock, flags);
82 err = notifier_chain_register(&die_chain, nb);
83 spin_unlock_irqrestore(&die_notifier_lock, flags);
84 return err;
85 }
86
87 static inline void conditional_sti(struct pt_regs *regs)
88 {
89 if (regs->eflags & X86_EFLAGS_IF)
90 local_irq_enable();
91 }
92
93 static int kstack_depth_to_print = 10;
94
95 #ifdef CONFIG_KALLSYMS
96 #include <linux/kallsyms.h>
97 int printk_address(unsigned long address)
98 {
99 unsigned long offset = 0, symsize;
100 const char *symname;
101 char *modname;
102 char *delim = ":";
103 char namebuf[128];
104
105 symname = kallsyms_lookup(address, &symsize, &offset, &modname, namebuf);
106 if (!symname)
107 return printk("[<%016lx>]", address);
108 if (!modname)
109 modname = delim = "";
110 return printk("<%016lx>{%s%s%s%s%+ld}",
111 address,delim,modname,delim,symname,offset);
112 }
113 #else
114 int printk_address(unsigned long address)
115 {
116 return printk("[<%016lx>]", address);
117 }
118 #endif
119
120 static unsigned long *in_exception_stack(unsigned cpu, unsigned long stack,
121 unsigned *usedp, const char **idp)
122 {
123 static char ids[][8] = {
124 [DEBUG_STACK - 1] = "#DB",
125 [NMI_STACK - 1] = "NMI",
126 [DOUBLEFAULT_STACK - 1] = "#DF",
127 [STACKFAULT_STACK - 1] = "#SS",
128 [MCE_STACK - 1] = "#MC",
129 #if DEBUG_STKSZ > EXCEPTION_STKSZ
130 [N_EXCEPTION_STACKS ... N_EXCEPTION_STACKS + DEBUG_STKSZ / EXCEPTION_STKSZ - 2] = "#DB[?]"
131 #endif
132 };
133 unsigned k;
134
135 for (k = 0; k < N_EXCEPTION_STACKS; k++) {
136 unsigned long end;
137
138 switch (k + 1) {
139 #if DEBUG_STKSZ > EXCEPTION_STKSZ
140 case DEBUG_STACK:
141 end = cpu_pda(cpu)->debugstack + DEBUG_STKSZ;
142 break;
143 #endif
144 default:
145 end = per_cpu(init_tss, cpu).ist[k];
146 break;
147 }
148 if (stack >= end)
149 continue;
150 if (stack >= end - EXCEPTION_STKSZ) {
151 if (*usedp & (1U << k))
152 break;
153 *usedp |= 1U << k;
154 *idp = ids[k];
155 return (unsigned long *)end;
156 }
157 #if DEBUG_STKSZ > EXCEPTION_STKSZ
158 if (k == DEBUG_STACK - 1 && stack >= end - DEBUG_STKSZ) {
159 unsigned j = N_EXCEPTION_STACKS - 1;
160
161 do {
162 ++j;
163 end -= EXCEPTION_STKSZ;
164 ids[j][4] = '1' + (j - N_EXCEPTION_STACKS);
165 } while (stack < end - EXCEPTION_STKSZ);
166 if (*usedp & (1U << j))
167 break;
168 *usedp |= 1U << j;
169 *idp = ids[j];
170 return (unsigned long *)end;
171 }
172 #endif
173 }
174 return NULL;
175 }
176
177 /*
178 * x86-64 can have upto three kernel stacks:
179 * process stack
180 * interrupt stack
181 * severe exception (double fault, nmi, stack fault, debug, mce) hardware stack
182 */
183
184 void show_trace(unsigned long *stack)
185 {
186 const unsigned cpu = safe_smp_processor_id();
187 unsigned long *irqstack_end = (unsigned long *)cpu_pda(cpu)->irqstackptr;
188 int i;
189 unsigned used = 0;
190
191 printk("\nCall Trace:");
192
193 #define HANDLE_STACK(cond) \
194 do while (cond) { \
195 unsigned long addr = *stack++; \
196 if (kernel_text_address(addr)) { \
197 if (i > 50) { \
198 printk("\n "); \
199 i = 0; \
200 } \
201 else \
202 i += printk(" "); \
203 /* \
204 * If the address is either in the text segment of the \
205 * kernel, or in the region which contains vmalloc'ed \
206 * memory, it *may* be the address of a calling \
207 * routine; if so, print it so that someone tracing \
208 * down the cause of the crash will be able to figure \
209 * out the call path that was taken. \
210 */ \
211 i += printk_address(addr); \
212 } \
213 } while (0)
214
215 for(i = 11; ; ) {
216 const char *id;
217 unsigned long *estack_end;
218 estack_end = in_exception_stack(cpu, (unsigned long)stack,
219 &used, &id);
220
221 if (estack_end) {
222 i += printk(" <%s>", id);
223 HANDLE_STACK (stack < estack_end);
224 i += printk(" <EOE>");
225 stack = (unsigned long *) estack_end[-2];
226 continue;
227 }
228 if (irqstack_end) {
229 unsigned long *irqstack;
230 irqstack = irqstack_end -
231 (IRQSTACKSIZE - 64) / sizeof(*irqstack);
232
233 if (stack >= irqstack && stack < irqstack_end) {
234 i += printk(" <IRQ>");
235 HANDLE_STACK (stack < irqstack_end);
236 stack = (unsigned long *) (irqstack_end[-1]);
237 irqstack_end = NULL;
238 i += printk(" <EOI>");
239 continue;
240 }
241 }
242 break;
243 }
244
245 HANDLE_STACK (((long) stack & (THREAD_SIZE-1)) != 0);
246 #undef HANDLE_STACK
247 printk("\n");
248 }
249
250 void show_stack(struct task_struct *tsk, unsigned long * rsp)
251 {
252 unsigned long *stack;
253 int i;
254 const int cpu = safe_smp_processor_id();
255 unsigned long *irqstack_end = (unsigned long *) (cpu_pda(cpu)->irqstackptr);
256 unsigned long *irqstack = (unsigned long *) (cpu_pda(cpu)->irqstackptr - IRQSTACKSIZE);
257
258 // debugging aid: "show_stack(NULL, NULL);" prints the
259 // back trace for this cpu.
260
261 if (rsp == NULL) {
262 if (tsk)
263 rsp = (unsigned long *)tsk->thread.rsp;
264 else
265 rsp = (unsigned long *)&rsp;
266 }
267
268 stack = rsp;
269 for(i=0; i < kstack_depth_to_print; i++) {
270 if (stack >= irqstack && stack <= irqstack_end) {
271 if (stack == irqstack_end) {
272 stack = (unsigned long *) (irqstack_end[-1]);
273 printk(" <EOI> ");
274 }
275 } else {
276 if (((long) stack & (THREAD_SIZE-1)) == 0)
277 break;
278 }
279 if (i && ((i % 4) == 0))
280 printk("\n ");
281 printk("%016lx ", *stack++);
282 touch_nmi_watchdog();
283 }
284 show_trace((unsigned long *)rsp);
285 }
286
287 /*
288 * The architecture-independent dump_stack generator
289 */
290 void dump_stack(void)
291 {
292 unsigned long dummy;
293 show_trace(&dummy);
294 }
295
296 EXPORT_SYMBOL(dump_stack);
297
298 void show_registers(struct pt_regs *regs)
299 {
300 int i;
301 int in_kernel = !user_mode(regs);
302 unsigned long rsp;
303 const int cpu = safe_smp_processor_id();
304 struct task_struct *cur = cpu_pda(cpu)->pcurrent;
305
306 rsp = regs->rsp;
307
308 printk("CPU %d ", cpu);
309 __show_regs(regs);
310 printk("Process %s (pid: %d, threadinfo %p, task %p)\n",
311 cur->comm, cur->pid, task_thread_info(cur), cur);
312
313 /*
314 * When in-kernel, we also print out the stack and code at the
315 * time of the fault..
316 */
317 if (in_kernel) {
318
319 printk("Stack: ");
320 show_stack(NULL, (unsigned long*)rsp);
321
322 printk("\nCode: ");
323 if(regs->rip < PAGE_OFFSET)
324 goto bad;
325
326 for(i=0;i<20;i++)
327 {
328 unsigned char c;
329 if(__get_user(c, &((unsigned char*)regs->rip)[i])) {
330 bad:
331 printk(" Bad RIP value.");
332 break;
333 }
334 printk("%02x ", c);
335 }
336 }
337 printk("\n");
338 }
339
340 void handle_BUG(struct pt_regs *regs)
341 {
342 struct bug_frame f;
343 long len;
344 const char *prefix = "";
345
346 if (user_mode(regs))
347 return;
348 if (__copy_from_user(&f, (const void __user *) regs->rip,
349 sizeof(struct bug_frame)))
350 return;
351 if (f.filename >= 0 ||
352 f.ud2[0] != 0x0f || f.ud2[1] != 0x0b)
353 return;
354 len = __strnlen_user((char *)(long)f.filename, PATH_MAX) - 1;
355 if (len < 0 || len >= PATH_MAX)
356 f.filename = (int)(long)"unmapped filename";
357 else if (len > 50) {
358 f.filename += len - 50;
359 prefix = "...";
360 }
361 printk("----------- [cut here ] --------- [please bite here ] ---------\n");
362 printk(KERN_ALERT "Kernel BUG at %s%.50s:%d\n", prefix, (char *)(long)f.filename, f.line);
363 }
364
365 #ifdef CONFIG_BUG
366 void out_of_line_bug(void)
367 {
368 BUG();
369 }
370 #endif
371
372 static DEFINE_SPINLOCK(die_lock);
373 static int die_owner = -1;
374
375 unsigned __kprobes long oops_begin(void)
376 {
377 int cpu = safe_smp_processor_id();
378 unsigned long flags;
379
380 /* racy, but better than risking deadlock. */
381 local_irq_save(flags);
382 if (!spin_trylock(&die_lock)) {
383 if (cpu == die_owner)
384 /* nested oops. should stop eventually */;
385 else
386 spin_lock(&die_lock);
387 }
388 die_owner = cpu;
389 console_verbose();
390 bust_spinlocks(1);
391 return flags;
392 }
393
394 void __kprobes oops_end(unsigned long flags)
395 {
396 die_owner = -1;
397 bust_spinlocks(0);
398 spin_unlock_irqrestore(&die_lock, flags);
399 if (panic_on_oops)
400 panic("Oops");
401 }
402
403 void __kprobes __die(const char * str, struct pt_regs * regs, long err)
404 {
405 static int die_counter;
406 printk(KERN_EMERG "%s: %04lx [%u] ", str, err & 0xffff,++die_counter);
407 #ifdef CONFIG_PREEMPT
408 printk("PREEMPT ");
409 #endif
410 #ifdef CONFIG_SMP
411 printk("SMP ");
412 #endif
413 #ifdef CONFIG_DEBUG_PAGEALLOC
414 printk("DEBUG_PAGEALLOC");
415 #endif
416 printk("\n");
417 notify_die(DIE_OOPS, str, regs, err, current->thread.trap_no, SIGSEGV);
418 show_registers(regs);
419 /* Executive summary in case the oops scrolled away */
420 printk(KERN_ALERT "RIP ");
421 printk_address(regs->rip);
422 printk(" RSP <%016lx>\n", regs->rsp);
423 }
424
425 void die(const char * str, struct pt_regs * regs, long err)
426 {
427 unsigned long flags = oops_begin();
428
429 handle_BUG(regs);
430 __die(str, regs, err);
431 oops_end(flags);
432 do_exit(SIGSEGV);
433 }
434
435 void __kprobes die_nmi(char *str, struct pt_regs *regs)
436 {
437 unsigned long flags = oops_begin();
438
439 /*
440 * We are in trouble anyway, lets at least try
441 * to get a message out.
442 */
443 printk(str, safe_smp_processor_id());
444 show_registers(regs);
445 if (panic_on_timeout || panic_on_oops)
446 panic("nmi watchdog");
447 printk("console shuts up ...\n");
448 oops_end(flags);
449 do_exit(SIGSEGV);
450 }
451
452 static void __kprobes do_trap(int trapnr, int signr, char *str,
453 struct pt_regs * regs, long error_code,
454 siginfo_t *info)
455 {
456 struct task_struct *tsk = current;
457
458 conditional_sti(regs);
459
460 tsk->thread.error_code = error_code;
461 tsk->thread.trap_no = trapnr;
462
463 if (user_mode(regs)) {
464 if (exception_trace && unhandled_signal(tsk, signr))
465 printk(KERN_INFO
466 "%s[%d] trap %s rip:%lx rsp:%lx error:%lx\n",
467 tsk->comm, tsk->pid, str,
468 regs->rip,regs->rsp,error_code);
469
470 if (info)
471 force_sig_info(signr, info, tsk);
472 else
473 force_sig(signr, tsk);
474 return;
475 }
476
477
478 /* kernel trap */
479 {
480 const struct exception_table_entry *fixup;
481 fixup = search_exception_tables(regs->rip);
482 if (fixup) {
483 regs->rip = fixup->fixup;
484 } else
485 die(str, regs, error_code);
486 return;
487 }
488 }
489
490 #define DO_ERROR(trapnr, signr, str, name) \
491 asmlinkage void do_##name(struct pt_regs * regs, long error_code) \
492 { \
493 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
494 == NOTIFY_STOP) \
495 return; \
496 do_trap(trapnr, signr, str, regs, error_code, NULL); \
497 }
498
499 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
500 asmlinkage void do_##name(struct pt_regs * regs, long error_code) \
501 { \
502 siginfo_t info; \
503 info.si_signo = signr; \
504 info.si_errno = 0; \
505 info.si_code = sicode; \
506 info.si_addr = (void __user *)siaddr; \
507 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
508 == NOTIFY_STOP) \
509 return; \
510 do_trap(trapnr, signr, str, regs, error_code, &info); \
511 }
512
513 DO_ERROR_INFO( 0, SIGFPE, "divide error", divide_error, FPE_INTDIV, regs->rip)
514 DO_ERROR( 4, SIGSEGV, "overflow", overflow)
515 DO_ERROR( 5, SIGSEGV, "bounds", bounds)
516 DO_ERROR_INFO( 6, SIGILL, "invalid opcode", invalid_op, ILL_ILLOPN, regs->rip)
517 DO_ERROR( 7, SIGSEGV, "device not available", device_not_available)
518 DO_ERROR( 9, SIGFPE, "coprocessor segment overrun", coprocessor_segment_overrun)
519 DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
520 DO_ERROR(11, SIGBUS, "segment not present", segment_not_present)
521 DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0)
522 DO_ERROR(18, SIGSEGV, "reserved", reserved)
523 DO_ERROR(12, SIGBUS, "stack segment", stack_segment)
524
525 asmlinkage void do_double_fault(struct pt_regs * regs, long error_code)
526 {
527 static const char str[] = "double fault";
528 struct task_struct *tsk = current;
529
530 /* Return not checked because double check cannot be ignored */
531 notify_die(DIE_TRAP, str, regs, error_code, 8, SIGSEGV);
532
533 tsk->thread.error_code = error_code;
534 tsk->thread.trap_no = 8;
535
536 /* This is always a kernel trap and never fixable (and thus must
537 never return). */
538 for (;;)
539 die(str, regs, error_code);
540 }
541
542 asmlinkage void __kprobes do_general_protection(struct pt_regs * regs,
543 long error_code)
544 {
545 struct task_struct *tsk = current;
546
547 conditional_sti(regs);
548
549 tsk->thread.error_code = error_code;
550 tsk->thread.trap_no = 13;
551
552 if (user_mode(regs)) {
553 if (exception_trace && unhandled_signal(tsk, SIGSEGV))
554 printk(KERN_INFO
555 "%s[%d] general protection rip:%lx rsp:%lx error:%lx\n",
556 tsk->comm, tsk->pid,
557 regs->rip,regs->rsp,error_code);
558
559 force_sig(SIGSEGV, tsk);
560 return;
561 }
562
563 /* kernel gp */
564 {
565 const struct exception_table_entry *fixup;
566 fixup = search_exception_tables(regs->rip);
567 if (fixup) {
568 regs->rip = fixup->fixup;
569 return;
570 }
571 if (notify_die(DIE_GPF, "general protection fault", regs,
572 error_code, 13, SIGSEGV) == NOTIFY_STOP)
573 return;
574 die("general protection fault", regs, error_code);
575 }
576 }
577
578 static __kprobes void
579 mem_parity_error(unsigned char reason, struct pt_regs * regs)
580 {
581 printk("Uhhuh. NMI received. Dazed and confused, but trying to continue\n");
582 printk("You probably have a hardware problem with your RAM chips\n");
583
584 /* Clear and disable the memory parity error line. */
585 reason = (reason & 0xf) | 4;
586 outb(reason, 0x61);
587 }
588
589 static __kprobes void
590 io_check_error(unsigned char reason, struct pt_regs * regs)
591 {
592 printk("NMI: IOCK error (debug interrupt?)\n");
593 show_registers(regs);
594
595 /* Re-enable the IOCK line, wait for a few seconds */
596 reason = (reason & 0xf) | 8;
597 outb(reason, 0x61);
598 mdelay(2000);
599 reason &= ~8;
600 outb(reason, 0x61);
601 }
602
603 static __kprobes void
604 unknown_nmi_error(unsigned char reason, struct pt_regs * regs)
605 { printk("Uhhuh. NMI received for unknown reason %02x.\n", reason);
606 printk("Dazed and confused, but trying to continue\n");
607 printk("Do you have a strange power saving mode enabled?\n");
608 }
609
610 /* Runs on IST stack. This code must keep interrupts off all the time.
611 Nested NMIs are prevented by the CPU. */
612 asmlinkage __kprobes void default_do_nmi(struct pt_regs *regs)
613 {
614 unsigned char reason = 0;
615 int cpu;
616
617 cpu = smp_processor_id();
618
619 /* Only the BSP gets external NMIs from the system. */
620 if (!cpu)
621 reason = get_nmi_reason();
622
623 if (!(reason & 0xc0)) {
624 if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 2, SIGINT)
625 == NOTIFY_STOP)
626 return;
627 #ifdef CONFIG_X86_LOCAL_APIC
628 /*
629 * Ok, so this is none of the documented NMI sources,
630 * so it must be the NMI watchdog.
631 */
632 if (nmi_watchdog > 0) {
633 nmi_watchdog_tick(regs,reason);
634 return;
635 }
636 #endif
637 unknown_nmi_error(reason, regs);
638 return;
639 }
640 if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP)
641 return;
642
643 /* AK: following checks seem to be broken on modern chipsets. FIXME */
644
645 if (reason & 0x80)
646 mem_parity_error(reason, regs);
647 if (reason & 0x40)
648 io_check_error(reason, regs);
649 }
650
651 /* runs on IST stack. */
652 asmlinkage void __kprobes do_int3(struct pt_regs * regs, long error_code)
653 {
654 if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP) == NOTIFY_STOP) {
655 return;
656 }
657 do_trap(3, SIGTRAP, "int3", regs, error_code, NULL);
658 return;
659 }
660
661 /* Help handler running on IST stack to switch back to user stack
662 for scheduling or signal handling. The actual stack switch is done in
663 entry.S */
664 asmlinkage __kprobes struct pt_regs *sync_regs(struct pt_regs *eregs)
665 {
666 struct pt_regs *regs = eregs;
667 /* Did already sync */
668 if (eregs == (struct pt_regs *)eregs->rsp)
669 ;
670 /* Exception from user space */
671 else if (user_mode(eregs))
672 regs = task_pt_regs(current);
673 /* Exception from kernel and interrupts are enabled. Move to
674 kernel process stack. */
675 else if (eregs->eflags & X86_EFLAGS_IF)
676 regs = (struct pt_regs *)(eregs->rsp -= sizeof(struct pt_regs));
677 if (eregs != regs)
678 *regs = *eregs;
679 return regs;
680 }
681
682 /* runs on IST stack. */
683 asmlinkage void __kprobes do_debug(struct pt_regs * regs,
684 unsigned long error_code)
685 {
686 unsigned long condition;
687 struct task_struct *tsk = current;
688 siginfo_t info;
689
690 get_debugreg(condition, 6);
691
692 if (notify_die(DIE_DEBUG, "debug", regs, condition, error_code,
693 SIGTRAP) == NOTIFY_STOP)
694 return;
695
696 conditional_sti(regs);
697
698 /* Mask out spurious debug traps due to lazy DR7 setting */
699 if (condition & (DR_TRAP0|DR_TRAP1|DR_TRAP2|DR_TRAP3)) {
700 if (!tsk->thread.debugreg7) {
701 goto clear_dr7;
702 }
703 }
704
705 tsk->thread.debugreg6 = condition;
706
707 /* Mask out spurious TF errors due to lazy TF clearing */
708 if (condition & DR_STEP) {
709 /*
710 * The TF error should be masked out only if the current
711 * process is not traced and if the TRAP flag has been set
712 * previously by a tracing process (condition detected by
713 * the PT_DTRACE flag); remember that the i386 TRAP flag
714 * can be modified by the process itself in user mode,
715 * allowing programs to debug themselves without the ptrace()
716 * interface.
717 */
718 if (!user_mode(regs))
719 goto clear_TF_reenable;
720 /*
721 * Was the TF flag set by a debugger? If so, clear it now,
722 * so that register information is correct.
723 */
724 if (tsk->ptrace & PT_DTRACE) {
725 regs->eflags &= ~TF_MASK;
726 tsk->ptrace &= ~PT_DTRACE;
727 }
728 }
729
730 /* Ok, finally something we can handle */
731 tsk->thread.trap_no = 1;
732 tsk->thread.error_code = error_code;
733 info.si_signo = SIGTRAP;
734 info.si_errno = 0;
735 info.si_code = TRAP_BRKPT;
736 info.si_addr = user_mode(regs) ? (void __user *)regs->rip : NULL;
737 force_sig_info(SIGTRAP, &info, tsk);
738
739 clear_dr7:
740 set_debugreg(0UL, 7);
741 return;
742
743 clear_TF_reenable:
744 set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
745 regs->eflags &= ~TF_MASK;
746 }
747
748 static int kernel_math_error(struct pt_regs *regs, const char *str, int trapnr)
749 {
750 const struct exception_table_entry *fixup;
751 fixup = search_exception_tables(regs->rip);
752 if (fixup) {
753 regs->rip = fixup->fixup;
754 return 1;
755 }
756 notify_die(DIE_GPF, str, regs, 0, trapnr, SIGFPE);
757 /* Illegal floating point operation in the kernel */
758 current->thread.trap_no = trapnr;
759 die(str, regs, 0);
760 return 0;
761 }
762
763 /*
764 * Note that we play around with the 'TS' bit in an attempt to get
765 * the correct behaviour even in the presence of the asynchronous
766 * IRQ13 behaviour
767 */
768 asmlinkage void do_coprocessor_error(struct pt_regs *regs)
769 {
770 void __user *rip = (void __user *)(regs->rip);
771 struct task_struct * task;
772 siginfo_t info;
773 unsigned short cwd, swd;
774
775 conditional_sti(regs);
776 if (!user_mode(regs) &&
777 kernel_math_error(regs, "kernel x87 math error", 16))
778 return;
779
780 /*
781 * Save the info for the exception handler and clear the error.
782 */
783 task = current;
784 save_init_fpu(task);
785 task->thread.trap_no = 16;
786 task->thread.error_code = 0;
787 info.si_signo = SIGFPE;
788 info.si_errno = 0;
789 info.si_code = __SI_FAULT;
790 info.si_addr = rip;
791 /*
792 * (~cwd & swd) will mask out exceptions that are not set to unmasked
793 * status. 0x3f is the exception bits in these regs, 0x200 is the
794 * C1 reg you need in case of a stack fault, 0x040 is the stack
795 * fault bit. We should only be taking one exception at a time,
796 * so if this combination doesn't produce any single exception,
797 * then we have a bad program that isn't synchronizing its FPU usage
798 * and it will suffer the consequences since we won't be able to
799 * fully reproduce the context of the exception
800 */
801 cwd = get_fpu_cwd(task);
802 swd = get_fpu_swd(task);
803 switch (swd & ~cwd & 0x3f) {
804 case 0x000:
805 default:
806 break;
807 case 0x001: /* Invalid Op */
808 /*
809 * swd & 0x240 == 0x040: Stack Underflow
810 * swd & 0x240 == 0x240: Stack Overflow
811 * User must clear the SF bit (0x40) if set
812 */
813 info.si_code = FPE_FLTINV;
814 break;
815 case 0x002: /* Denormalize */
816 case 0x010: /* Underflow */
817 info.si_code = FPE_FLTUND;
818 break;
819 case 0x004: /* Zero Divide */
820 info.si_code = FPE_FLTDIV;
821 break;
822 case 0x008: /* Overflow */
823 info.si_code = FPE_FLTOVF;
824 break;
825 case 0x020: /* Precision */
826 info.si_code = FPE_FLTRES;
827 break;
828 }
829 force_sig_info(SIGFPE, &info, task);
830 }
831
832 asmlinkage void bad_intr(void)
833 {
834 printk("bad interrupt");
835 }
836
837 asmlinkage void do_simd_coprocessor_error(struct pt_regs *regs)
838 {
839 void __user *rip = (void __user *)(regs->rip);
840 struct task_struct * task;
841 siginfo_t info;
842 unsigned short mxcsr;
843
844 conditional_sti(regs);
845 if (!user_mode(regs) &&
846 kernel_math_error(regs, "kernel simd math error", 19))
847 return;
848
849 /*
850 * Save the info for the exception handler and clear the error.
851 */
852 task = current;
853 save_init_fpu(task);
854 task->thread.trap_no = 19;
855 task->thread.error_code = 0;
856 info.si_signo = SIGFPE;
857 info.si_errno = 0;
858 info.si_code = __SI_FAULT;
859 info.si_addr = rip;
860 /*
861 * The SIMD FPU exceptions are handled a little differently, as there
862 * is only a single status/control register. Thus, to determine which
863 * unmasked exception was caught we must mask the exception mask bits
864 * at 0x1f80, and then use these to mask the exception bits at 0x3f.
865 */
866 mxcsr = get_fpu_mxcsr(task);
867 switch (~((mxcsr & 0x1f80) >> 7) & (mxcsr & 0x3f)) {
868 case 0x000:
869 default:
870 break;
871 case 0x001: /* Invalid Op */
872 info.si_code = FPE_FLTINV;
873 break;
874 case 0x002: /* Denormalize */
875 case 0x010: /* Underflow */
876 info.si_code = FPE_FLTUND;
877 break;
878 case 0x004: /* Zero Divide */
879 info.si_code = FPE_FLTDIV;
880 break;
881 case 0x008: /* Overflow */
882 info.si_code = FPE_FLTOVF;
883 break;
884 case 0x020: /* Precision */
885 info.si_code = FPE_FLTRES;
886 break;
887 }
888 force_sig_info(SIGFPE, &info, task);
889 }
890
891 asmlinkage void do_spurious_interrupt_bug(struct pt_regs * regs)
892 {
893 }
894
895 asmlinkage void __attribute__((weak)) smp_thermal_interrupt(void)
896 {
897 }
898
899 asmlinkage void __attribute__((weak)) mce_threshold_interrupt(void)
900 {
901 }
902
903 /*
904 * 'math_state_restore()' saves the current math information in the
905 * old math state array, and gets the new ones from the current task
906 *
907 * Careful.. There are problems with IBM-designed IRQ13 behaviour.
908 * Don't touch unless you *really* know how it works.
909 */
910 asmlinkage void math_state_restore(void)
911 {
912 struct task_struct *me = current;
913 clts(); /* Allow maths ops (or we recurse) */
914
915 if (!used_math())
916 init_fpu(me);
917 restore_fpu_checking(&me->thread.i387.fxsave);
918 task_thread_info(me)->status |= TS_USEDFPU;
919 }
920
921 void __init trap_init(void)
922 {
923 set_intr_gate(0,&divide_error);
924 set_intr_gate_ist(1,&debug,DEBUG_STACK);
925 set_intr_gate_ist(2,&nmi,NMI_STACK);
926 set_system_gate_ist(3,&int3,DEBUG_STACK); /* int3 can be called from all */
927 set_system_gate(4,&overflow); /* int4 can be called from all */
928 set_intr_gate(5,&bounds);
929 set_intr_gate(6,&invalid_op);
930 set_intr_gate(7,&device_not_available);
931 set_intr_gate_ist(8,&double_fault, DOUBLEFAULT_STACK);
932 set_intr_gate(9,&coprocessor_segment_overrun);
933 set_intr_gate(10,&invalid_TSS);
934 set_intr_gate(11,&segment_not_present);
935 set_intr_gate_ist(12,&stack_segment,STACKFAULT_STACK);
936 set_intr_gate(13,&general_protection);
937 set_intr_gate(14,&page_fault);
938 set_intr_gate(15,&spurious_interrupt_bug);
939 set_intr_gate(16,&coprocessor_error);
940 set_intr_gate(17,&alignment_check);
941 #ifdef CONFIG_X86_MCE
942 set_intr_gate_ist(18,&machine_check, MCE_STACK);
943 #endif
944 set_intr_gate(19,&simd_coprocessor_error);
945
946 #ifdef CONFIG_IA32_EMULATION
947 set_system_gate(IA32_SYSCALL_VECTOR, ia32_syscall);
948 #endif
949
950 /*
951 * Should be a barrier for any external CPU state.
952 */
953 cpu_init();
954 }
955
956
957 /* Actual parsing is done early in setup.c. */
958 static int __init oops_dummy(char *s)
959 {
960 panic_on_oops = 1;
961 return -1;
962 }
963 __setup("oops=", oops_dummy);
964
965 static int __init kstack_setup(char *s)
966 {
967 kstack_depth_to_print = simple_strtoul(s,NULL,0);
968 return 0;
969 }
970 __setup("kstack=", kstack_setup);
971
This page took 0.073958 seconds and 5 git commands to generate.