FreeBSD/Linux Kernel Cross Reference
sys/i386/i386/trap.c
1 /*-
2 * Copyright (C) 1994, David Greenman
3 * Copyright (c) 1990, 1993
4 * The Regents of the University of California. All rights reserved.
5 *
6 * This code is derived from software contributed to Berkeley by
7 * the University of Utah, and William Jolitz.
8 *
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
11 * are met:
12 * 1. Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in the
16 * documentation and/or other materials provided with the distribution.
17 * 3. All advertising materials mentioning features or use of this software
18 * must display the following acknowledgement:
19 * This product includes software developed by the University of
20 * California, Berkeley and its contributors.
21 * 4. Neither the name of the University nor the names of its contributors
22 * may be used to endorse or promote products derived from this software
23 * without specific prior written permission.
24 *
25 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
26 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
27 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
28 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
29 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
30 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
31 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
32 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
33 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
34 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35 * SUCH DAMAGE.
36 *
37 * from: @(#)trap.c 7.4 (Berkeley) 5/13/91
38 */
39
40 #include <sys/cdefs.h>
41 __FBSDID("$FreeBSD: releng/6.0/sys/i386/i386/trap.c 151061 2005-10-07 14:00:06Z glebius $");
42
43 /*
44 * 386 Trap and System call handling
45 */
46
47 #include "opt_clock.h"
48 #include "opt_cpu.h"
49 #include "opt_hwpmc_hooks.h"
50 #include "opt_isa.h"
51 #include "opt_kdb.h"
52 #include "opt_ktrace.h"
53 #include "opt_npx.h"
54 #include "opt_trap.h"
55
56 #include <sys/param.h>
57 #include <sys/bus.h>
58 #include <sys/systm.h>
59 #include <sys/proc.h>
60 #include <sys/pioctl.h>
61 #include <sys/ptrace.h>
62 #include <sys/kdb.h>
63 #include <sys/kernel.h>
64 #include <sys/ktr.h>
65 #include <sys/lock.h>
66 #include <sys/mutex.h>
67 #include <sys/resourcevar.h>
68 #include <sys/signalvar.h>
69 #include <sys/syscall.h>
70 #include <sys/sysctl.h>
71 #include <sys/sysent.h>
72 #include <sys/uio.h>
73 #include <sys/vmmeter.h>
74 #ifdef KTRACE
75 #include <sys/ktrace.h>
76 #endif
77 #ifdef HWPMC_HOOKS
78 #include <sys/pmckern.h>
79 #endif
80
81 #include <vm/vm.h>
82 #include <vm/vm_param.h>
83 #include <vm/pmap.h>
84 #include <vm/vm_kern.h>
85 #include <vm/vm_map.h>
86 #include <vm/vm_page.h>
87 #include <vm/vm_extern.h>
88
89 #include <machine/cpu.h>
90 #include <machine/intr_machdep.h>
91 #include <machine/md_var.h>
92 #include <machine/pcb.h>
93 #ifdef SMP
94 #include <machine/smp.h>
95 #endif
96 #include <machine/tss.h>
97 #include <machine/vm86.h>
98
99 #ifdef POWERFAIL_NMI
100 #include <sys/syslog.h>
101 #include <machine/clock.h>
102 #endif
103
104 extern void trap(struct trapframe frame);
105 extern void syscall(struct trapframe frame);
106
107 static int trap_pfault(struct trapframe *, int, vm_offset_t);
108 static void trap_fatal(struct trapframe *, vm_offset_t);
109 void dblfault_handler(void);
110
111 extern inthand_t IDTVEC(lcall_syscall);
112
113 #define MAX_TRAP_MSG 28
114 static char *trap_msg[] = {
115 "", /* 0 unused */
116 "privileged instruction fault", /* 1 T_PRIVINFLT */
117 "", /* 2 unused */
118 "breakpoint instruction fault", /* 3 T_BPTFLT */
119 "", /* 4 unused */
120 "", /* 5 unused */
121 "arithmetic trap", /* 6 T_ARITHTRAP */
122 "", /* 7 unused */
123 "", /* 8 unused */
124 "general protection fault", /* 9 T_PROTFLT */
125 "trace trap", /* 10 T_TRCTRAP */
126 "", /* 11 unused */
127 "page fault", /* 12 T_PAGEFLT */
128 "", /* 13 unused */
129 "alignment fault", /* 14 T_ALIGNFLT */
130 "", /* 15 unused */
131 "", /* 16 unused */
132 "", /* 17 unused */
133 "integer divide fault", /* 18 T_DIVIDE */
134 "non-maskable interrupt trap", /* 19 T_NMI */
135 "overflow trap", /* 20 T_OFLOW */
136 "FPU bounds check fault", /* 21 T_BOUND */
137 "FPU device not available", /* 22 T_DNA */
138 "double fault", /* 23 T_DOUBLEFLT */
139 "FPU operand fetch fault", /* 24 T_FPOPFLT */
140 "invalid TSS fault", /* 25 T_TSSFLT */
141 "segment not present fault", /* 26 T_SEGNPFLT */
142 "stack fault", /* 27 T_STKFLT */
143 "machine check trap", /* 28 T_MCHK */
144 };
145
146 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
147 extern int has_f00f_bug;
148 #endif
149
150 #ifdef KDB
151 static int kdb_on_nmi = 1;
152 SYSCTL_INT(_machdep, OID_AUTO, kdb_on_nmi, CTLFLAG_RW,
153 &kdb_on_nmi, 0, "Go to KDB on NMI");
154 #endif
155 static int panic_on_nmi = 1;
156 SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RW,
157 &panic_on_nmi, 0, "Panic on NMI");
158
159 #ifdef WITNESS
160 extern char *syscallnames[];
161 #endif
162
163 /*
164 * Exception, fault, and trap interface to the FreeBSD kernel.
165 * This common code is called from assembly language IDT gate entry
166 * routines that prepare a suitable stack frame, and restore this
167 * frame after the exception has been processed.
168 */
169
170 void
171 trap(frame)
172 struct trapframe frame;
173 {
174 struct thread *td = curthread;
175 struct proc *p = td->td_proc;
176 u_int sticks = 0;
177 int i = 0, ucode = 0, type, code;
178 vm_offset_t eva;
179 #ifdef POWERFAIL_NMI
180 static int lastalert = 0;
181 #endif
182
183 PCPU_LAZY_INC(cnt.v_trap);
184 type = frame.tf_trapno;
185
186 #ifdef KDB_STOP_NMI
187 /* Handler for NMI IPIs used for debugging */
188 if (type == T_NMI) {
189 if (ipi_nmi_handler() == 0)
190 goto out;
191 }
192 #endif /* KDB_STOP_NMI */
193
194 #ifdef KDB
195 if (kdb_active) {
196 kdb_reenter();
197 goto out;
198 }
199 #endif
200
201 #ifdef HWPMC_HOOKS
202 /*
203 * CPU PMCs interrupt using an NMI so we check for that first.
204 * If the HWPMC module is active, 'pmc_hook' will point to
205 * the function to be called. A return value of '1' from the
206 * hook means that the NMI was handled by it and that we can
207 * return immediately.
208 */
209 if (type == T_NMI && pmc_intr &&
210 (*pmc_intr)(PCPU_GET(cpuid), (uintptr_t) frame.tf_eip,
211 TRAPF_USERMODE(&frame)))
212 goto out;
213 #endif
214
215 if ((frame.tf_eflags & PSL_I) == 0) {
216 /*
217 * Buggy application or kernel code has disabled
218 * interrupts and then trapped. Enabling interrupts
219 * now is wrong, but it is better than running with
220 * interrupts disabled until they are accidentally
221 * enabled later.
222 */
223 if (ISPL(frame.tf_cs) == SEL_UPL || (frame.tf_eflags & PSL_VM))
224 printf(
225 "pid %ld (%s): trap %d with interrupts disabled\n",
226 (long)curproc->p_pid, curproc->p_comm, type);
227 else if (type != T_BPTFLT && type != T_TRCTRAP &&
228 frame.tf_eip != (int)cpu_switch_load_gs) {
229 /*
230 * XXX not quite right, since this may be for a
231 * multiple fault in user mode.
232 */
233 printf("kernel trap %d with interrupts disabled\n",
234 type);
235 /*
236 * Page faults need interrupts disabled until later,
237 * and we shouldn't enable interrupts while in a
238 * critical section or if servicing an NMI.
239 */
240 if (type != T_NMI && type != T_PAGEFLT &&
241 td->td_critnest == 0)
242 enable_intr();
243 }
244 }
245
246 eva = 0;
247 code = frame.tf_err;
248 if (type == T_PAGEFLT) {
249 /*
250 * For some Cyrix CPUs, %cr2 is clobbered by
251 * interrupts. This problem is worked around by using
252 * an interrupt gate for the pagefault handler. We
253 * are finally ready to read %cr2 and then must
254 * reenable interrupts.
255 *
256 * If we get a page fault while in a critical section, then
257 * it is most likely a fatal kernel page fault. The kernel
258 * is already going to panic trying to get a sleep lock to
259 * do the VM lookup, so just consider it a fatal trap so the
260 * kernel can print out a useful trap message and even get
261 * to the debugger.
262 */
263 eva = rcr2();
264 if (td->td_critnest == 0)
265 enable_intr();
266 else
267 trap_fatal(&frame, eva);
268 }
269
270 if ((ISPL(frame.tf_cs) == SEL_UPL) ||
271 ((frame.tf_eflags & PSL_VM) &&
272 !(PCPU_GET(curpcb)->pcb_flags & PCB_VM86CALL))) {
273 /* user trap */
274
275 sticks = td->td_sticks;
276 td->td_frame = &frame;
277 if (td->td_ucred != p->p_ucred)
278 cred_update_thread(td);
279
280 switch (type) {
281 case T_PRIVINFLT: /* privileged instruction fault */
282 ucode = type;
283 i = SIGILL;
284 break;
285
286 case T_BPTFLT: /* bpt instruction fault */
287 case T_TRCTRAP: /* trace trap */
288 enable_intr();
289 frame.tf_eflags &= ~PSL_T;
290 i = SIGTRAP;
291 break;
292
293 case T_ARITHTRAP: /* arithmetic trap */
294 #ifdef DEV_NPX
295 ucode = npxtrap();
296 if (ucode == -1)
297 goto userout;
298 #else
299 ucode = code;
300 #endif
301 i = SIGFPE;
302 break;
303
304 /*
305 * The following two traps can happen in
306 * vm86 mode, and, if so, we want to handle
307 * them specially.
308 */
309 case T_PROTFLT: /* general protection fault */
310 case T_STKFLT: /* stack fault */
311 if (frame.tf_eflags & PSL_VM) {
312 i = vm86_emulate((struct vm86frame *)&frame);
313 if (i == 0)
314 goto user;
315 break;
316 }
317 /* FALLTHROUGH */
318
319 case T_SEGNPFLT: /* segment not present fault */
320 case T_TSSFLT: /* invalid TSS fault */
321 case T_DOUBLEFLT: /* double fault */
322 default:
323 ucode = code + BUS_SEGM_FAULT ;
324 i = SIGBUS;
325 break;
326
327 case T_PAGEFLT: /* page fault */
328 if (td->td_pflags & TDP_SA)
329 thread_user_enter(td);
330
331 i = trap_pfault(&frame, TRUE, eva);
332 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
333 if (i == -2) {
334 /*
335 * The f00f hack workaround has triggered, so
336 * treat the fault as an illegal instruction
337 * (T_PRIVINFLT) instead of a page fault.
338 */
339 type = frame.tf_trapno = T_PRIVINFLT;
340
341 /* Proceed as in that case. */
342 ucode = type;
343 i = SIGILL;
344 break;
345 }
346 #endif
347 if (i == -1)
348 goto userout;
349 if (i == 0)
350 goto user;
351
352 ucode = T_PAGEFLT;
353 break;
354
355 case T_DIVIDE: /* integer divide fault */
356 ucode = FPE_INTDIV;
357 i = SIGFPE;
358 break;
359
360 #ifdef DEV_ISA
361 case T_NMI:
362 #ifdef POWERFAIL_NMI
363 #ifndef TIMER_FREQ
364 # define TIMER_FREQ 1193182
365 #endif
366 mtx_lock(&Giant);
367 if (time_second - lastalert > 10) {
368 log(LOG_WARNING, "NMI: power fail\n");
369 sysbeep(TIMER_FREQ/880, hz);
370 lastalert = time_second;
371 }
372 mtx_unlock(&Giant);
373 goto userout;
374 #else /* !POWERFAIL_NMI */
375 /* machine/parity/power fail/"kitchen sink" faults */
376 /* XXX Giant */
377 if (isa_nmi(code) == 0) {
378 #ifdef KDB
379 /*
380 * NMI can be hooked up to a pushbutton
381 * for debugging.
382 */
383 if (kdb_on_nmi) {
384 printf ("NMI ... going to debugger\n");
385 kdb_trap(type, 0, &frame);
386 }
387 #endif /* KDB */
388 goto userout;
389 } else if (panic_on_nmi)
390 panic("NMI indicates hardware failure");
391 break;
392 #endif /* POWERFAIL_NMI */
393 #endif /* DEV_ISA */
394
395 case T_OFLOW: /* integer overflow fault */
396 ucode = FPE_INTOVF;
397 i = SIGFPE;
398 break;
399
400 case T_BOUND: /* bounds check fault */
401 ucode = FPE_FLTSUB;
402 i = SIGFPE;
403 break;
404
405 case T_DNA:
406 #ifdef DEV_NPX
407 /* transparent fault (due to context switch "late") */
408 if (npxdna())
409 goto userout;
410 #endif
411 i = SIGFPE;
412 ucode = FPE_FPU_NP_TRAP;
413 break;
414
415 case T_FPOPFLT: /* FPU operand fetch fault */
416 ucode = T_FPOPFLT;
417 i = SIGILL;
418 break;
419
420 case T_XMMFLT: /* SIMD floating-point exception */
421 ucode = 0; /* XXX */
422 i = SIGFPE;
423 break;
424 }
425 } else {
426 /* kernel trap */
427
428 KASSERT(cold || td->td_ucred != NULL,
429 ("kernel trap doesn't have ucred"));
430 switch (type) {
431 case T_PAGEFLT: /* page fault */
432 (void) trap_pfault(&frame, FALSE, eva);
433 goto out;
434
435 case T_DNA:
436 #ifdef DEV_NPX
437 /*
438 * The kernel is apparently using npx for copying.
439 * XXX this should be fatal unless the kernel has
440 * registered such use.
441 */
442 if (npxdna())
443 goto out;
444 #endif
445 break;
446
447 /*
448 * The following two traps can happen in
449 * vm86 mode, and, if so, we want to handle
450 * them specially.
451 */
452 case T_PROTFLT: /* general protection fault */
453 case T_STKFLT: /* stack fault */
454 if (frame.tf_eflags & PSL_VM) {
455 i = vm86_emulate((struct vm86frame *)&frame);
456 if (i != 0)
457 /*
458 * returns to original process
459 */
460 vm86_trap((struct vm86frame *)&frame);
461 goto out;
462 }
463 if (type == T_STKFLT)
464 break;
465
466 /* FALL THROUGH */
467
468 case T_SEGNPFLT: /* segment not present fault */
469 if (PCPU_GET(curpcb)->pcb_flags & PCB_VM86CALL)
470 break;
471
472 /*
473 * Invalid %fs's and %gs's can be created using
474 * procfs or PT_SETREGS or by invalidating the
475 * underlying LDT entry. This causes a fault
476 * in kernel mode when the kernel attempts to
477 * switch contexts. Lose the bad context
478 * (XXX) so that we can continue, and generate
479 * a signal.
480 */
481 if (frame.tf_eip == (int)cpu_switch_load_gs) {
482 PCPU_GET(curpcb)->pcb_gs = 0;
483 #if 0
484 PROC_LOCK(p);
485 psignal(p, SIGBUS);
486 PROC_UNLOCK(p);
487 #endif
488 goto out;
489 }
490
491 if (td->td_intr_nesting_level != 0)
492 break;
493
494 /*
495 * Invalid segment selectors and out of bounds
496 * %eip's and %esp's can be set up in user mode.
497 * This causes a fault in kernel mode when the
498 * kernel tries to return to user mode. We want
499 * to get this fault so that we can fix the
500 * problem here and not have to check all the
501 * selectors and pointers when the user changes
502 * them.
503 */
504 if (frame.tf_eip == (int)doreti_iret) {
505 frame.tf_eip = (int)doreti_iret_fault;
506 goto out;
507 }
508 if (frame.tf_eip == (int)doreti_popl_ds) {
509 frame.tf_eip = (int)doreti_popl_ds_fault;
510 goto out;
511 }
512 if (frame.tf_eip == (int)doreti_popl_es) {
513 frame.tf_eip = (int)doreti_popl_es_fault;
514 goto out;
515 }
516 if (frame.tf_eip == (int)doreti_popl_fs) {
517 frame.tf_eip = (int)doreti_popl_fs_fault;
518 goto out;
519 }
520 if (PCPU_GET(curpcb)->pcb_onfault != NULL) {
521 frame.tf_eip =
522 (int)PCPU_GET(curpcb)->pcb_onfault;
523 goto out;
524 }
525 break;
526
527 case T_TSSFLT:
528 /*
529 * PSL_NT can be set in user mode and isn't cleared
530 * automatically when the kernel is entered. This
531 * causes a TSS fault when the kernel attempts to
532 * `iret' because the TSS link is uninitialized. We
533 * want to get this fault so that we can fix the
534 * problem here and not every time the kernel is
535 * entered.
536 */
537 if (frame.tf_eflags & PSL_NT) {
538 frame.tf_eflags &= ~PSL_NT;
539 goto out;
540 }
541 break;
542
543 case T_TRCTRAP: /* trace trap */
544 if (frame.tf_eip == (int)IDTVEC(lcall_syscall)) {
545 /*
546 * We've just entered system mode via the
547 * syscall lcall. Continue single stepping
548 * silently until the syscall handler has
549 * saved the flags.
550 */
551 goto out;
552 }
553 if (frame.tf_eip == (int)IDTVEC(lcall_syscall) + 1) {
554 /*
555 * The syscall handler has now saved the
556 * flags. Stop single stepping it.
557 */
558 frame.tf_eflags &= ~PSL_T;
559 goto out;
560 }
561 /*
562 * Ignore debug register trace traps due to
563 * accesses in the user's address space, which
564 * can happen under several conditions such as
565 * if a user sets a watchpoint on a buffer and
566 * then passes that buffer to a system call.
567 * We still want to get TRCTRAPS for addresses
568 * in kernel space because that is useful when
569 * debugging the kernel.
570 */
571 /* XXX Giant */
572 if (user_dbreg_trap() &&
573 !(PCPU_GET(curpcb)->pcb_flags & PCB_VM86CALL)) {
574 /*
575 * Reset breakpoint bits because the
576 * processor doesn't
577 */
578 load_dr6(rdr6() & 0xfffffff0);
579 goto out;
580 }
581 /*
582 * FALLTHROUGH (TRCTRAP kernel mode, kernel address)
583 */
584 case T_BPTFLT:
585 /*
586 * If KDB is enabled, let it handle the debugger trap.
587 * Otherwise, debugger traps "can't happen".
588 */
589 #ifdef KDB
590 /* XXX Giant */
591 if (kdb_trap(type, 0, &frame))
592 goto out;
593 #endif
594 break;
595
596 #ifdef DEV_ISA
597 case T_NMI:
598 #ifdef POWERFAIL_NMI
599 mtx_lock(&Giant);
600 if (time_second - lastalert > 10) {
601 log(LOG_WARNING, "NMI: power fail\n");
602 sysbeep(TIMER_FREQ/880, hz);
603 lastalert = time_second;
604 }
605 mtx_unlock(&Giant);
606 goto out;
607 #else /* !POWERFAIL_NMI */
608 /* XXX Giant */
609 /* machine/parity/power fail/"kitchen sink" faults */
610 if (isa_nmi(code) == 0) {
611 #ifdef KDB
612 /*
613 * NMI can be hooked up to a pushbutton
614 * for debugging.
615 */
616 if (kdb_on_nmi) {
617 printf ("NMI ... going to debugger\n");
618 kdb_trap(type, 0, &frame);
619 }
620 #endif /* KDB */
621 goto out;
622 } else if (panic_on_nmi == 0)
623 goto out;
624 /* FALLTHROUGH */
625 #endif /* POWERFAIL_NMI */
626 #endif /* DEV_ISA */
627 }
628
629 trap_fatal(&frame, eva);
630 goto out;
631 }
632
633 /* Translate fault for emulators (e.g. Linux) */
634 if (*p->p_sysent->sv_transtrap)
635 i = (*p->p_sysent->sv_transtrap)(i, type);
636
637 trapsignal(td, i, ucode);
638
639 #ifdef DEBUG
640 if (type <= MAX_TRAP_MSG) {
641 uprintf("fatal process exception: %s",
642 trap_msg[type]);
643 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
644 uprintf(", fault VA = 0x%lx", (u_long)eva);
645 uprintf("\n");
646 }
647 #endif
648
649 user:
650 userret(td, &frame, sticks);
651 mtx_assert(&Giant, MA_NOTOWNED);
652 userout:
653 out:
654 return;
655 }
656
657 static int
658 trap_pfault(frame, usermode, eva)
659 struct trapframe *frame;
660 int usermode;
661 vm_offset_t eva;
662 {
663 vm_offset_t va;
664 struct vmspace *vm = NULL;
665 vm_map_t map = 0;
666 int rv = 0;
667 vm_prot_t ftype;
668 struct thread *td = curthread;
669 struct proc *p = td->td_proc;
670
671 va = trunc_page(eva);
672 if (va >= KERNBASE) {
673 /*
674 * Don't allow user-mode faults in kernel address space.
675 * An exception: if the faulting address is the invalid
676 * instruction entry in the IDT, then the Intel Pentium
677 * F00F bug workaround was triggered, and we need to
678 * treat it is as an illegal instruction, and not a page
679 * fault.
680 */
681 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
682 if ((eva == (unsigned int)&idt[6]) && has_f00f_bug)
683 return -2;
684 #endif
685 if (usermode)
686 goto nogo;
687
688 map = kernel_map;
689 } else {
690 /*
691 * This is a fault on non-kernel virtual memory.
692 * vm is initialized above to NULL. If curproc is NULL
693 * or curproc->p_vmspace is NULL the fault is fatal.
694 */
695 if (p != NULL)
696 vm = p->p_vmspace;
697
698 if (vm == NULL)
699 goto nogo;
700
701 map = &vm->vm_map;
702 }
703
704 if (frame->tf_err & PGEX_W)
705 ftype = VM_PROT_WRITE;
706 else
707 ftype = VM_PROT_READ;
708
709 if (map != kernel_map) {
710 /*
711 * Keep swapout from messing with us during this
712 * critical time.
713 */
714 PROC_LOCK(p);
715 ++p->p_lock;
716 PROC_UNLOCK(p);
717
718 /* Fault in the user page: */
719 rv = vm_fault(map, va, ftype,
720 (ftype & VM_PROT_WRITE) ? VM_FAULT_DIRTY
721 : VM_FAULT_NORMAL);
722
723 PROC_LOCK(p);
724 --p->p_lock;
725 PROC_UNLOCK(p);
726 } else {
727 /*
728 * Don't have to worry about process locking or stacks in the
729 * kernel.
730 */
731 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
732 }
733 if (rv == KERN_SUCCESS)
734 return (0);
735 nogo:
736 if (!usermode) {
737 if (td->td_intr_nesting_level == 0 &&
738 PCPU_GET(curpcb)->pcb_onfault != NULL) {
739 frame->tf_eip = (int)PCPU_GET(curpcb)->pcb_onfault;
740 return (0);
741 }
742 trap_fatal(frame, eva);
743 return (-1);
744 }
745
746 /* kludge to pass faulting virtual address to sendsig */
747 frame->tf_err = eva;
748
749 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
750 }
751
752 static void
753 trap_fatal(frame, eva)
754 struct trapframe *frame;
755 vm_offset_t eva;
756 {
757 int code, type, ss, esp;
758 struct soft_segment_descriptor softseg;
759
760 code = frame->tf_err;
761 type = frame->tf_trapno;
762 sdtossd(&gdt[IDXSEL(frame->tf_cs & 0xffff)].sd, &softseg);
763
764 if (type <= MAX_TRAP_MSG)
765 printf("\n\nFatal trap %d: %s while in %s mode\n",
766 type, trap_msg[type],
767 frame->tf_eflags & PSL_VM ? "vm86" :
768 ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel");
769 #ifdef SMP
770 /* two separate prints in case of a trap on an unmapped page */
771 printf("cpuid = %d; ", PCPU_GET(cpuid));
772 printf("apic id = %02x\n", PCPU_GET(apic_id));
773 #endif
774 if (type == T_PAGEFLT) {
775 printf("fault virtual address = 0x%x\n", eva);
776 printf("fault code = %s %s, %s\n",
777 code & PGEX_U ? "user" : "supervisor",
778 code & PGEX_W ? "write" : "read",
779 code & PGEX_P ? "protection violation" : "page not present");
780 }
781 printf("instruction pointer = 0x%x:0x%x\n",
782 frame->tf_cs & 0xffff, frame->tf_eip);
783 if ((ISPL(frame->tf_cs) == SEL_UPL) || (frame->tf_eflags & PSL_VM)) {
784 ss = frame->tf_ss & 0xffff;
785 esp = frame->tf_esp;
786 } else {
787 ss = GSEL(GDATA_SEL, SEL_KPL);
788 esp = (int)&frame->tf_esp;
789 }
790 printf("stack pointer = 0x%x:0x%x\n", ss, esp);
791 printf("frame pointer = 0x%x:0x%x\n", ss, frame->tf_ebp);
792 printf("code segment = base 0x%x, limit 0x%x, type 0x%x\n",
793 softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type);
794 printf(" = DPL %d, pres %d, def32 %d, gran %d\n",
795 softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_def32,
796 softseg.ssd_gran);
797 printf("processor eflags = ");
798 if (frame->tf_eflags & PSL_T)
799 printf("trace trap, ");
800 if (frame->tf_eflags & PSL_I)
801 printf("interrupt enabled, ");
802 if (frame->tf_eflags & PSL_NT)
803 printf("nested task, ");
804 if (frame->tf_eflags & PSL_RF)
805 printf("resume, ");
806 if (frame->tf_eflags & PSL_VM)
807 printf("vm86, ");
808 printf("IOPL = %d\n", (frame->tf_eflags & PSL_IOPL) >> 12);
809 printf("current process = ");
810 if (curproc) {
811 printf("%lu (%s)\n",
812 (u_long)curproc->p_pid, curproc->p_comm ?
813 curproc->p_comm : "");
814 } else {
815 printf("Idle\n");
816 }
817
818 #ifdef KDB
819 if (debugger_on_panic || kdb_active) {
820 register_t eflags;
821 eflags = intr_disable();
822 if (kdb_trap(type, 0, frame)) {
823 intr_restore(eflags);
824 return;
825 }
826 intr_restore(eflags);
827 }
828 #endif
829 printf("trap number = %d\n", type);
830 if (type <= MAX_TRAP_MSG)
831 panic("%s", trap_msg[type]);
832 else
833 panic("unknown/reserved trap");
834 }
835
836 /*
837 * Double fault handler. Called when a fault occurs while writing
838 * a frame for a trap/exception onto the stack. This usually occurs
839 * when the stack overflows (such is the case with infinite recursion,
840 * for example).
841 *
842 * XXX Note that the current PTD gets replaced by IdlePTD when the
843 * task switch occurs. This means that the stack that was active at
844 * the time of the double fault is not available at <kstack> unless
845 * the machine was idle when the double fault occurred. The downside
846 * of this is that "trace <ebp>" in ddb won't work.
847 */
848 void
849 dblfault_handler()
850 {
851 printf("\nFatal double fault:\n");
852 printf("eip = 0x%x\n", PCPU_GET(common_tss.tss_eip));
853 printf("esp = 0x%x\n", PCPU_GET(common_tss.tss_esp));
854 printf("ebp = 0x%x\n", PCPU_GET(common_tss.tss_ebp));
855 #ifdef SMP
856 /* two separate prints in case of a trap on an unmapped page */
857 printf("cpuid = %d; ", PCPU_GET(cpuid));
858 printf("apic id = %02x\n", PCPU_GET(apic_id));
859 #endif
860 panic("double fault");
861 }
862
863 /*
864 * syscall - system call request C handler
865 *
866 * A system call is essentially treated as a trap.
867 */
868 void
869 syscall(frame)
870 struct trapframe frame;
871 {
872 caddr_t params;
873 struct sysent *callp;
874 struct thread *td = curthread;
875 struct proc *p = td->td_proc;
876 register_t orig_tf_eflags;
877 u_int sticks;
878 int error;
879 int narg;
880 int args[8];
881 u_int code;
882
883 /*
884 * note: PCPU_LAZY_INC() can only be used if we can afford
885 * occassional inaccuracy in the count.
886 */
887 PCPU_LAZY_INC(cnt.v_syscall);
888
889 #ifdef DIAGNOSTIC
890 if (ISPL(frame.tf_cs) != SEL_UPL) {
891 mtx_lock(&Giant); /* try to stabilize the system XXX */
892 panic("syscall");
893 /* NOT REACHED */
894 mtx_unlock(&Giant);
895 }
896 #endif
897
898 sticks = td->td_sticks;
899 td->td_frame = &frame;
900 if (td->td_ucred != p->p_ucred)
901 cred_update_thread(td);
902 if (p->p_flag & P_SA)
903 thread_user_enter(td);
904 params = (caddr_t)frame.tf_esp + sizeof(int);
905 code = frame.tf_eax;
906 orig_tf_eflags = frame.tf_eflags;
907
908 if (p->p_sysent->sv_prepsyscall) {
909 /*
910 * The prep code is MP aware.
911 */
912 (*p->p_sysent->sv_prepsyscall)(&frame, args, &code, ¶ms);
913 } else {
914 /*
915 * Need to check if this is a 32 bit or 64 bit syscall.
916 * fuword is MP aware.
917 */
918 if (code == SYS_syscall) {
919 /*
920 * Code is first argument, followed by actual args.
921 */
922 code = fuword(params);
923 params += sizeof(int);
924 } else if (code == SYS___syscall) {
925 /*
926 * Like syscall, but code is a quad, so as to maintain
927 * quad alignment for the rest of the arguments.
928 */
929 code = fuword(params);
930 params += sizeof(quad_t);
931 }
932 }
933
934 if (p->p_sysent->sv_mask)
935 code &= p->p_sysent->sv_mask;
936
937 if (code >= p->p_sysent->sv_size)
938 callp = &p->p_sysent->sv_table[0];
939 else
940 callp = &p->p_sysent->sv_table[code];
941
942 narg = callp->sy_narg & SYF_ARGMASK;
943
944 /*
945 * copyin and the ktrsyscall()/ktrsysret() code is MP-aware
946 */
947 if (params != NULL && narg != 0)
948 error = copyin(params, (caddr_t)args,
949 (u_int)(narg * sizeof(int)));
950 else
951 error = 0;
952
953 #ifdef KTRACE
954 if (KTRPOINT(td, KTR_SYSCALL))
955 ktrsyscall(code, narg, args);
956 #endif
957
958 CTR4(KTR_SYSC, "syscall enter thread %p pid %d proc %s code %d", td,
959 td->td_proc->p_pid, td->td_proc->p_comm, code);
960
961 /*
962 * Try to run the syscall without Giant if the syscall
963 * is MP safe.
964 */
965 if ((callp->sy_narg & SYF_MPSAFE) == 0)
966 mtx_lock(&Giant);
967
968 if (error == 0) {
969 td->td_retval[0] = 0;
970 td->td_retval[1] = frame.tf_edx;
971
972 STOPEVENT(p, S_SCE, narg);
973
974 PTRACESTOP_SC(p, td, S_PT_SCE);
975
976 error = (*callp->sy_call)(td, args);
977 }
978
979 switch (error) {
980 case 0:
981 frame.tf_eax = td->td_retval[0];
982 frame.tf_edx = td->td_retval[1];
983 frame.tf_eflags &= ~PSL_C;
984 break;
985
986 case ERESTART:
987 /*
988 * Reconstruct pc, assuming lcall $X,y is 7 bytes,
989 * int 0x80 is 2 bytes. We saved this in tf_err.
990 */
991 frame.tf_eip -= frame.tf_err;
992 break;
993
994 case EJUSTRETURN:
995 break;
996
997 default:
998 if (p->p_sysent->sv_errsize) {
999 if (error >= p->p_sysent->sv_errsize)
1000 error = -1; /* XXX */
1001 else
1002 error = p->p_sysent->sv_errtbl[error];
1003 }
1004 frame.tf_eax = error;
1005 frame.tf_eflags |= PSL_C;
1006 break;
1007 }
1008
1009 /*
1010 * Release Giant if we previously set it.
1011 */
1012 if ((callp->sy_narg & SYF_MPSAFE) == 0)
1013 mtx_unlock(&Giant);
1014
1015 /*
1016 * Traced syscall.
1017 */
1018 if ((orig_tf_eflags & PSL_T) && !(orig_tf_eflags & PSL_VM)) {
1019 frame.tf_eflags &= ~PSL_T;
1020 trapsignal(td, SIGTRAP, 0);
1021 }
1022
1023 /*
1024 * Handle reschedule and other end-of-syscall issues
1025 */
1026 userret(td, &frame, sticks);
1027
1028 CTR4(KTR_SYSC, "syscall exit thread %p pid %d proc %s code %d", td,
1029 td->td_proc->p_pid, td->td_proc->p_comm, code);
1030
1031 #ifdef KTRACE
1032 if (KTRPOINT(td, KTR_SYSRET))
1033 ktrsysret(code, error, td->td_retval[0]);
1034 #endif
1035
1036 /*
1037 * This works because errno is findable through the
1038 * register set. If we ever support an emulation where this
1039 * is not the case, this code will need to be revisited.
1040 */
1041 STOPEVENT(p, S_SCX, code);
1042
1043 PTRACESTOP_SC(p, td, S_PT_SCX);
1044
1045 WITNESS_WARN(WARN_PANIC, NULL, "System call %s returning",
1046 (code >= 0 && code < SYS_MAXSYSCALL) ? syscallnames[code] : "???");
1047 mtx_assert(&sched_lock, MA_NOTOWNED);
1048 mtx_assert(&Giant, MA_NOTOWNED);
1049 }
1050
Cache object: 315373f2eeb15df086180b02e141bd16
|