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/5.2/sys/i386/i386/trap.c 121986 2003-11-03 21:53:38Z jhb $");
42
43 /*
44 * 386 Trap and System call handling
45 */
46
47 #include "opt_clock.h"
48 #include "opt_cpu.h"
49 #include "opt_ddb.h"
50 #include "opt_isa.h"
51 #include "opt_ktrace.h"
52 #include "opt_npx.h"
53 #include "opt_trap.h"
54
55 #include <sys/param.h>
56 #include <sys/bus.h>
57 #include <sys/systm.h>
58 #include <sys/proc.h>
59 #include <sys/pioctl.h>
60 #include <sys/ptrace.h>
61 #include <sys/kernel.h>
62 #include <sys/ktr.h>
63 #include <sys/lock.h>
64 #include <sys/mutex.h>
65 #include <sys/resourcevar.h>
66 #include <sys/signalvar.h>
67 #include <sys/syscall.h>
68 #include <sys/sysctl.h>
69 #include <sys/sysent.h>
70 #include <sys/uio.h>
71 #include <sys/vmmeter.h>
72 #ifdef KTRACE
73 #include <sys/ktrace.h>
74 #endif
75
76 #include <vm/vm.h>
77 #include <vm/vm_param.h>
78 #include <vm/pmap.h>
79 #include <vm/vm_kern.h>
80 #include <vm/vm_map.h>
81 #include <vm/vm_page.h>
82 #include <vm/vm_extern.h>
83
84 #include <machine/cpu.h>
85 #include <machine/intr_machdep.h>
86 #include <machine/md_var.h>
87 #include <machine/pcb.h>
88 #ifdef SMP
89 #include <machine/smp.h>
90 #endif
91 #include <machine/tss.h>
92 #include <machine/vm86.h>
93
94 #ifdef POWERFAIL_NMI
95 #include <sys/syslog.h>
96 #include <machine/clock.h>
97 #endif
98
99 #include <ddb/ddb.h>
100
101 extern void trap(struct trapframe frame);
102 #ifdef I386_CPU
103 extern int trapwrite(unsigned addr);
104 #endif
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 DDB
151 static int ddb_on_nmi = 1;
152 SYSCTL_INT(_machdep, OID_AUTO, ddb_on_nmi, CTLFLAG_RW,
153 &ddb_on_nmi, 0, "Go to DDB 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 #ifdef DEVICE_POLLING
164 extern u_int32_t poll_in_trap;
165 extern int ether_poll(int count);
166 #endif /* DEVICE_POLLING */
167
168 /*
169 * Exception, fault, and trap interface to the FreeBSD kernel.
170 * This common code is called from assembly language IDT gate entry
171 * routines that prepare a suitable stack frame, and restore this
172 * frame after the exception has been processed.
173 */
174
175 void
176 trap(frame)
177 struct trapframe frame;
178 {
179 struct thread *td = curthread;
180 struct proc *p = td->td_proc;
181 u_int sticks = 0;
182 int i = 0, ucode = 0, type, code;
183 vm_offset_t eva;
184 #ifdef POWERFAIL_NMI
185 static int lastalert = 0;
186 #endif
187
188 atomic_add_int(&cnt.v_trap, 1);
189 type = frame.tf_trapno;
190
191 #ifdef DDB
192 if (db_active) {
193 eva = (type == T_PAGEFLT ? rcr2() : 0);
194 trap_fatal(&frame, eva);
195 goto out;
196 }
197 #endif
198
199 if ((frame.tf_eflags & PSL_I) == 0) {
200 /*
201 * Buggy application or kernel code has disabled
202 * interrupts and then trapped. Enabling interrupts
203 * now is wrong, but it is better than running with
204 * interrupts disabled until they are accidentally
205 * enabled later.
206 */
207 if (ISPL(frame.tf_cs) == SEL_UPL || (frame.tf_eflags & PSL_VM))
208 printf(
209 "pid %ld (%s): trap %d with interrupts disabled\n",
210 (long)curproc->p_pid, curproc->p_comm, type);
211 else if (type != T_BPTFLT && type != T_TRCTRAP &&
212 frame.tf_eip != (int)cpu_switch_load_gs) {
213 /*
214 * XXX not quite right, since this may be for a
215 * multiple fault in user mode.
216 */
217 printf("kernel trap %d with interrupts disabled\n",
218 type);
219 /*
220 * Page faults need interrupts diasabled until later,
221 * and we shouldn't enable interrupts while in a
222 * critical section.
223 */
224 if (type != T_PAGEFLT && td->td_critnest == 0)
225 enable_intr();
226 }
227 }
228
229 eva = 0;
230 code = frame.tf_err;
231 if (type == T_PAGEFLT) {
232 /*
233 * For some Cyrix CPUs, %cr2 is clobbered by
234 * interrupts. This problem is worked around by using
235 * an interrupt gate for the pagefault handler. We
236 * are finally ready to read %cr2 and then must
237 * reenable interrupts.
238 *
239 * If we get a page fault while in a critical section, then
240 * it is most likely a fatal kernel page fault. The kernel
241 * is already going to panic trying to get a sleep lock to
242 * do the VM lookup, so just consider it a fatal trap so the
243 * kernel can print out a useful trap message and even get
244 * to the debugger.
245 */
246 eva = rcr2();
247 if (td->td_critnest == 0)
248 enable_intr();
249 else
250 trap_fatal(&frame, eva);
251 }
252
253 #ifdef DEVICE_POLLING
254 if (poll_in_trap)
255 ether_poll(poll_in_trap);
256 #endif /* DEVICE_POLLING */
257
258 if ((ISPL(frame.tf_cs) == SEL_UPL) ||
259 ((frame.tf_eflags & PSL_VM) &&
260 !(PCPU_GET(curpcb)->pcb_flags & PCB_VM86CALL))) {
261 /* user trap */
262
263 sticks = td->td_sticks;
264 td->td_frame = &frame;
265 if (td->td_ucred != p->p_ucred)
266 cred_update_thread(td);
267
268 switch (type) {
269 case T_PRIVINFLT: /* privileged instruction fault */
270 ucode = type;
271 i = SIGILL;
272 break;
273
274 case T_BPTFLT: /* bpt instruction fault */
275 case T_TRCTRAP: /* trace trap */
276 enable_intr();
277 frame.tf_eflags &= ~PSL_T;
278 i = SIGTRAP;
279 break;
280
281 case T_ARITHTRAP: /* arithmetic trap */
282 #ifdef DEV_NPX
283 ucode = npxtrap();
284 if (ucode == -1)
285 goto userout;
286 #else
287 ucode = code;
288 #endif
289 i = SIGFPE;
290 break;
291
292 /*
293 * The following two traps can happen in
294 * vm86 mode, and, if so, we want to handle
295 * them specially.
296 */
297 case T_PROTFLT: /* general protection fault */
298 case T_STKFLT: /* stack fault */
299 if (frame.tf_eflags & PSL_VM) {
300 i = vm86_emulate((struct vm86frame *)&frame);
301 if (i == 0)
302 goto user;
303 break;
304 }
305 /* FALLTHROUGH */
306
307 case T_SEGNPFLT: /* segment not present fault */
308 case T_TSSFLT: /* invalid TSS fault */
309 case T_DOUBLEFLT: /* double fault */
310 default:
311 ucode = code + BUS_SEGM_FAULT ;
312 i = SIGBUS;
313 break;
314
315 case T_PAGEFLT: /* page fault */
316 if (td->td_flags & TDF_SA)
317 thread_user_enter(p, td);
318
319 i = trap_pfault(&frame, TRUE, eva);
320 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
321 if (i == -2) {
322 /*
323 * The f00f hack workaround has triggered, so
324 * treat the fault as an illegal instruction
325 * (T_PRIVINFLT) instead of a page fault.
326 */
327 type = frame.tf_trapno = T_PRIVINFLT;
328
329 /* Proceed as in that case. */
330 ucode = type;
331 i = SIGILL;
332 break;
333 }
334 #endif
335 if (i == -1)
336 goto userout;
337 if (i == 0)
338 goto user;
339
340 ucode = T_PAGEFLT;
341 break;
342
343 case T_DIVIDE: /* integer divide fault */
344 ucode = FPE_INTDIV;
345 i = SIGFPE;
346 break;
347
348 #ifdef DEV_ISA
349 case T_NMI:
350 #ifdef POWERFAIL_NMI
351 #ifndef TIMER_FREQ
352 # define TIMER_FREQ 1193182
353 #endif
354 mtx_lock(&Giant);
355 if (time_second - lastalert > 10) {
356 log(LOG_WARNING, "NMI: power fail\n");
357 sysbeep(TIMER_FREQ/880, hz);
358 lastalert = time_second;
359 }
360 mtx_unlock(&Giant);
361 goto userout;
362 #else /* !POWERFAIL_NMI */
363 /* machine/parity/power fail/"kitchen sink" faults */
364 /* XXX Giant */
365 if (isa_nmi(code) == 0) {
366 #ifdef DDB
367 /*
368 * NMI can be hooked up to a pushbutton
369 * for debugging.
370 */
371 if (ddb_on_nmi) {
372 printf ("NMI ... going to debugger\n");
373 kdb_trap (type, 0, &frame);
374 }
375 #endif /* DDB */
376 goto userout;
377 } else if (panic_on_nmi)
378 panic("NMI indicates hardware failure");
379 break;
380 #endif /* POWERFAIL_NMI */
381 #endif /* DEV_ISA */
382
383 case T_OFLOW: /* integer overflow fault */
384 ucode = FPE_INTOVF;
385 i = SIGFPE;
386 break;
387
388 case T_BOUND: /* bounds check fault */
389 ucode = FPE_FLTSUB;
390 i = SIGFPE;
391 break;
392
393 case T_DNA:
394 #ifdef DEV_NPX
395 /* transparent fault (due to context switch "late") */
396 if (npxdna())
397 goto userout;
398 #endif
399 i = SIGFPE;
400 ucode = FPE_FPU_NP_TRAP;
401 break;
402
403 case T_FPOPFLT: /* FPU operand fetch fault */
404 ucode = T_FPOPFLT;
405 i = SIGILL;
406 break;
407
408 case T_XMMFLT: /* SIMD floating-point exception */
409 ucode = 0; /* XXX */
410 i = SIGFPE;
411 break;
412 }
413 } else {
414 /* kernel trap */
415
416 KASSERT(cold || td->td_ucred != NULL,
417 ("kernel trap doesn't have ucred"));
418 switch (type) {
419 case T_PAGEFLT: /* page fault */
420 (void) trap_pfault(&frame, FALSE, eva);
421 goto out;
422
423 case T_DNA:
424 #ifdef DEV_NPX
425 /*
426 * The kernel is apparently using npx for copying.
427 * XXX this should be fatal unless the kernel has
428 * registered such use.
429 */
430 if (npxdna())
431 goto out;
432 #endif
433 break;
434
435 /*
436 * The following two traps can happen in
437 * vm86 mode, and, if so, we want to handle
438 * them specially.
439 */
440 case T_PROTFLT: /* general protection fault */
441 case T_STKFLT: /* stack fault */
442 if (frame.tf_eflags & PSL_VM) {
443 i = vm86_emulate((struct vm86frame *)&frame);
444 if (i != 0)
445 /*
446 * returns to original process
447 */
448 vm86_trap((struct vm86frame *)&frame);
449 goto out;
450 }
451 if (type == T_STKFLT)
452 break;
453
454 /* FALL THROUGH */
455
456 case T_SEGNPFLT: /* segment not present fault */
457 if (PCPU_GET(curpcb)->pcb_flags & PCB_VM86CALL)
458 break;
459
460 /*
461 * Invalid %fs's and %gs's can be created using
462 * procfs or PT_SETREGS or by invalidating the
463 * underlying LDT entry. This causes a fault
464 * in kernel mode when the kernel attempts to
465 * switch contexts. Lose the bad context
466 * (XXX) so that we can continue, and generate
467 * a signal.
468 */
469 if (frame.tf_eip == (int)cpu_switch_load_gs) {
470 PCPU_GET(curpcb)->pcb_gs = 0;
471 #if 0
472 PROC_LOCK(p);
473 psignal(p, SIGBUS);
474 PROC_UNLOCK(p);
475 #endif
476 goto out;
477 }
478
479 if (td->td_intr_nesting_level != 0)
480 break;
481
482 /*
483 * Invalid segment selectors and out of bounds
484 * %eip's and %esp's can be set up in user mode.
485 * This causes a fault in kernel mode when the
486 * kernel tries to return to user mode. We want
487 * to get this fault so that we can fix the
488 * problem here and not have to check all the
489 * selectors and pointers when the user changes
490 * them.
491 */
492 if (frame.tf_eip == (int)doreti_iret) {
493 frame.tf_eip = (int)doreti_iret_fault;
494 goto out;
495 }
496 if (frame.tf_eip == (int)doreti_popl_ds) {
497 frame.tf_eip = (int)doreti_popl_ds_fault;
498 goto out;
499 }
500 if (frame.tf_eip == (int)doreti_popl_es) {
501 frame.tf_eip = (int)doreti_popl_es_fault;
502 goto out;
503 }
504 if (frame.tf_eip == (int)doreti_popl_fs) {
505 frame.tf_eip = (int)doreti_popl_fs_fault;
506 goto out;
507 }
508 if (PCPU_GET(curpcb) != NULL &&
509 PCPU_GET(curpcb)->pcb_onfault != NULL) {
510 frame.tf_eip =
511 (int)PCPU_GET(curpcb)->pcb_onfault;
512 goto out;
513 }
514 break;
515
516 case T_TSSFLT:
517 /*
518 * PSL_NT can be set in user mode and isn't cleared
519 * automatically when the kernel is entered. This
520 * causes a TSS fault when the kernel attempts to
521 * `iret' because the TSS link is uninitialized. We
522 * want to get this fault so that we can fix the
523 * problem here and not every time the kernel is
524 * entered.
525 */
526 if (frame.tf_eflags & PSL_NT) {
527 frame.tf_eflags &= ~PSL_NT;
528 goto out;
529 }
530 break;
531
532 case T_TRCTRAP: /* trace trap */
533 if (frame.tf_eip == (int)IDTVEC(lcall_syscall)) {
534 /*
535 * We've just entered system mode via the
536 * syscall lcall. Continue single stepping
537 * silently until the syscall handler has
538 * saved the flags.
539 */
540 goto out;
541 }
542 if (frame.tf_eip == (int)IDTVEC(lcall_syscall) + 1) {
543 /*
544 * The syscall handler has now saved the
545 * flags. Stop single stepping it.
546 */
547 frame.tf_eflags &= ~PSL_T;
548 goto out;
549 }
550 /*
551 * Ignore debug register trace traps due to
552 * accesses in the user's address space, which
553 * can happen under several conditions such as
554 * if a user sets a watchpoint on a buffer and
555 * then passes that buffer to a system call.
556 * We still want to get TRCTRAPS for addresses
557 * in kernel space because that is useful when
558 * debugging the kernel.
559 */
560 /* XXX Giant */
561 if (user_dbreg_trap() &&
562 !(PCPU_GET(curpcb)->pcb_flags & PCB_VM86CALL)) {
563 /*
564 * Reset breakpoint bits because the
565 * processor doesn't
566 */
567 load_dr6(rdr6() & 0xfffffff0);
568 goto out;
569 }
570 /*
571 * FALLTHROUGH (TRCTRAP kernel mode, kernel address)
572 */
573 case T_BPTFLT:
574 /*
575 * If DDB is enabled, let it handle the debugger trap.
576 * Otherwise, debugger traps "can't happen".
577 */
578 #ifdef DDB
579 /* XXX Giant */
580 if (kdb_trap (type, 0, &frame))
581 goto out;
582 #endif
583 break;
584
585 #ifdef DEV_ISA
586 case T_NMI:
587 #ifdef POWERFAIL_NMI
588 mtx_lock(&Giant);
589 if (time_second - lastalert > 10) {
590 log(LOG_WARNING, "NMI: power fail\n");
591 sysbeep(TIMER_FREQ/880, hz);
592 lastalert = time_second;
593 }
594 mtx_unlock(&Giant);
595 goto out;
596 #else /* !POWERFAIL_NMI */
597 /* XXX Giant */
598 /* machine/parity/power fail/"kitchen sink" faults */
599 if (isa_nmi(code) == 0) {
600 #ifdef DDB
601 /*
602 * NMI can be hooked up to a pushbutton
603 * for debugging.
604 */
605 if (ddb_on_nmi) {
606 printf ("NMI ... going to debugger\n");
607 kdb_trap (type, 0, &frame);
608 }
609 #endif /* DDB */
610 goto out;
611 } else if (panic_on_nmi == 0)
612 goto out;
613 /* FALLTHROUGH */
614 #endif /* POWERFAIL_NMI */
615 #endif /* DEV_ISA */
616 }
617
618 trap_fatal(&frame, eva);
619 goto out;
620 }
621
622 /* Translate fault for emulators (e.g. Linux) */
623 if (*p->p_sysent->sv_transtrap)
624 i = (*p->p_sysent->sv_transtrap)(i, type);
625
626 trapsignal(td, i, ucode);
627
628 #ifdef DEBUG
629 if (type <= MAX_TRAP_MSG) {
630 uprintf("fatal process exception: %s",
631 trap_msg[type]);
632 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
633 uprintf(", fault VA = 0x%lx", (u_long)eva);
634 uprintf("\n");
635 }
636 #endif
637
638 user:
639 userret(td, &frame, sticks);
640 mtx_assert(&Giant, MA_NOTOWNED);
641 userout:
642 #ifdef DIAGNOSTIC
643 cred_free_thread(td);
644 #endif
645 out:
646 return;
647 }
648
649 static int
650 trap_pfault(frame, usermode, eva)
651 struct trapframe *frame;
652 int usermode;
653 vm_offset_t eva;
654 {
655 vm_offset_t va;
656 struct vmspace *vm = NULL;
657 vm_map_t map = 0;
658 int rv = 0;
659 vm_prot_t ftype;
660 struct thread *td = curthread;
661 struct proc *p = td->td_proc;
662
663 va = trunc_page(eva);
664 if (va >= KERNBASE) {
665 /*
666 * Don't allow user-mode faults in kernel address space.
667 * An exception: if the faulting address is the invalid
668 * instruction entry in the IDT, then the Intel Pentium
669 * F00F bug workaround was triggered, and we need to
670 * treat it is as an illegal instruction, and not a page
671 * fault.
672 */
673 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
674 if ((eva == (unsigned int)&idt[6]) && has_f00f_bug)
675 return -2;
676 #endif
677 if (usermode)
678 goto nogo;
679
680 map = kernel_map;
681 } else {
682 /*
683 * This is a fault on non-kernel virtual memory.
684 * vm is initialized above to NULL. If curproc is NULL
685 * or curproc->p_vmspace is NULL the fault is fatal.
686 */
687 if (p != NULL)
688 vm = p->p_vmspace;
689
690 if (vm == NULL)
691 goto nogo;
692
693 map = &vm->vm_map;
694 }
695
696 if (frame->tf_err & PGEX_W)
697 ftype = VM_PROT_WRITE;
698 else
699 ftype = VM_PROT_READ;
700
701 if (map != kernel_map) {
702 /*
703 * Keep swapout from messing with us during this
704 * critical time.
705 */
706 PROC_LOCK(p);
707 ++p->p_lock;
708 PROC_UNLOCK(p);
709
710 /* Fault in the user page: */
711 rv = vm_fault(map, va, ftype,
712 (ftype & VM_PROT_WRITE) ? VM_FAULT_DIRTY
713 : VM_FAULT_NORMAL);
714
715 PROC_LOCK(p);
716 --p->p_lock;
717 PROC_UNLOCK(p);
718 } else {
719 /*
720 * Don't have to worry about process locking or stacks in the
721 * kernel.
722 */
723 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
724 }
725 if (rv == KERN_SUCCESS)
726 return (0);
727 nogo:
728 if (!usermode) {
729 if (td->td_intr_nesting_level == 0 &&
730 PCPU_GET(curpcb) != NULL &&
731 PCPU_GET(curpcb)->pcb_onfault != NULL) {
732 frame->tf_eip = (int)PCPU_GET(curpcb)->pcb_onfault;
733 return (0);
734 }
735 trap_fatal(frame, eva);
736 return (-1);
737 }
738
739 /* kludge to pass faulting virtual address to sendsig */
740 frame->tf_err = eva;
741
742 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
743 }
744
745 static void
746 trap_fatal(frame, eva)
747 struct trapframe *frame;
748 vm_offset_t eva;
749 {
750 int code, type, ss, esp;
751 struct soft_segment_descriptor softseg;
752
753 code = frame->tf_err;
754 type = frame->tf_trapno;
755 sdtossd(&gdt[IDXSEL(frame->tf_cs & 0xffff)].sd, &softseg);
756
757 if (type <= MAX_TRAP_MSG)
758 printf("\n\nFatal trap %d: %s while in %s mode\n",
759 type, trap_msg[type],
760 frame->tf_eflags & PSL_VM ? "vm86" :
761 ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel");
762 #ifdef SMP
763 /* two separate prints in case of a trap on an unmapped page */
764 printf("cpuid = %d; ", PCPU_GET(cpuid));
765 printf("apic id = %02x\n", PCPU_GET(apic_id));
766 #endif
767 if (type == T_PAGEFLT) {
768 printf("fault virtual address = 0x%x\n", eva);
769 printf("fault code = %s %s, %s\n",
770 code & PGEX_U ? "user" : "supervisor",
771 code & PGEX_W ? "write" : "read",
772 code & PGEX_P ? "protection violation" : "page not present");
773 }
774 printf("instruction pointer = 0x%x:0x%x\n",
775 frame->tf_cs & 0xffff, frame->tf_eip);
776 if ((ISPL(frame->tf_cs) == SEL_UPL) || (frame->tf_eflags & PSL_VM)) {
777 ss = frame->tf_ss & 0xffff;
778 esp = frame->tf_esp;
779 } else {
780 ss = GSEL(GDATA_SEL, SEL_KPL);
781 esp = (int)&frame->tf_esp;
782 }
783 printf("stack pointer = 0x%x:0x%x\n", ss, esp);
784 printf("frame pointer = 0x%x:0x%x\n", ss, frame->tf_ebp);
785 printf("code segment = base 0x%x, limit 0x%x, type 0x%x\n",
786 softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type);
787 printf(" = DPL %d, pres %d, def32 %d, gran %d\n",
788 softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_def32,
789 softseg.ssd_gran);
790 printf("processor eflags = ");
791 if (frame->tf_eflags & PSL_T)
792 printf("trace trap, ");
793 if (frame->tf_eflags & PSL_I)
794 printf("interrupt enabled, ");
795 if (frame->tf_eflags & PSL_NT)
796 printf("nested task, ");
797 if (frame->tf_eflags & PSL_RF)
798 printf("resume, ");
799 if (frame->tf_eflags & PSL_VM)
800 printf("vm86, ");
801 printf("IOPL = %d\n", (frame->tf_eflags & PSL_IOPL) >> 12);
802 printf("current process = ");
803 if (curproc) {
804 printf("%lu (%s)\n",
805 (u_long)curproc->p_pid, curproc->p_comm ?
806 curproc->p_comm : "");
807 } else {
808 printf("Idle\n");
809 }
810
811 #ifdef KDB
812 if (kdb_trap(&psl))
813 return;
814 #endif
815 #ifdef DDB
816 if ((debugger_on_panic || db_active) && kdb_trap(type, 0, frame))
817 return;
818 #endif
819 printf("trap number = %d\n", type);
820 if (type <= MAX_TRAP_MSG)
821 panic("%s", trap_msg[type]);
822 else
823 panic("unknown/reserved trap");
824 }
825
826 /*
827 * Double fault handler. Called when a fault occurs while writing
828 * a frame for a trap/exception onto the stack. This usually occurs
829 * when the stack overflows (such is the case with infinite recursion,
830 * for example).
831 *
832 * XXX Note that the current PTD gets replaced by IdlePTD when the
833 * task switch occurs. This means that the stack that was active at
834 * the time of the double fault is not available at <kstack> unless
835 * the machine was idle when the double fault occurred. The downside
836 * of this is that "trace <ebp>" in ddb won't work.
837 */
838 void
839 dblfault_handler()
840 {
841 printf("\nFatal double fault:\n");
842 printf("eip = 0x%x\n", PCPU_GET(common_tss.tss_eip));
843 printf("esp = 0x%x\n", PCPU_GET(common_tss.tss_esp));
844 printf("ebp = 0x%x\n", PCPU_GET(common_tss.tss_ebp));
845 #ifdef SMP
846 /* two separate prints in case of a trap on an unmapped page */
847 printf("cpuid = %d; ", PCPU_GET(cpuid));
848 printf("apic id = %02x\n", PCPU_GET(apic_id));
849 #endif
850 panic("double fault");
851 }
852
853 #ifdef I386_CPU
854 /*
855 * Compensate for 386 brain damage (missing URKR).
856 * This is a little simpler than the pagefault handler in trap() because
857 * it the page tables have already been faulted in and high addresses
858 * are thrown out early for other reasons.
859 */
860 int trapwrite(addr)
861 unsigned addr;
862 {
863 struct thread *td;
864 struct proc *p;
865 vm_offset_t va;
866 struct vmspace *vm;
867 int rv;
868
869 va = trunc_page((vm_offset_t)addr);
870 /*
871 * XXX - MAX is END. Changed > to >= for temp. fix.
872 */
873 if (va >= VM_MAXUSER_ADDRESS)
874 return (1);
875
876 td = curthread;
877 p = td->td_proc;
878 vm = p->p_vmspace;
879
880 PROC_LOCK(p);
881 ++p->p_lock;
882 PROC_UNLOCK(p);
883
884 /*
885 * fault the data page
886 */
887 rv = vm_fault(&vm->vm_map, va, VM_PROT_WRITE, VM_FAULT_DIRTY);
888
889 PROC_LOCK(p);
890 --p->p_lock;
891 PROC_UNLOCK(p);
892
893 if (rv != KERN_SUCCESS)
894 return 1;
895
896 return (0);
897 }
898 #endif
899
900 /*
901 * syscall - system call request C handler
902 *
903 * A system call is essentially treated as a trap.
904 */
905 void
906 syscall(frame)
907 struct trapframe frame;
908 {
909 caddr_t params;
910 struct sysent *callp;
911 struct thread *td = curthread;
912 struct proc *p = td->td_proc;
913 register_t orig_tf_eflags;
914 u_int sticks;
915 int error;
916 int narg;
917 int args[8];
918 u_int code;
919
920 /*
921 * note: PCPU_LAZY_INC() can only be used if we can afford
922 * occassional inaccuracy in the count.
923 */
924 PCPU_LAZY_INC(cnt.v_syscall);
925
926 #ifdef DIAGNOSTIC
927 if (ISPL(frame.tf_cs) != SEL_UPL) {
928 mtx_lock(&Giant); /* try to stabilize the system XXX */
929 panic("syscall");
930 /* NOT REACHED */
931 mtx_unlock(&Giant);
932 }
933 #endif
934
935 sticks = td->td_sticks;
936 td->td_frame = &frame;
937 if (td->td_ucred != p->p_ucred)
938 cred_update_thread(td);
939 if (p->p_flag & P_SA)
940 thread_user_enter(p, td);
941 params = (caddr_t)frame.tf_esp + sizeof(int);
942 code = frame.tf_eax;
943 orig_tf_eflags = frame.tf_eflags;
944
945 if (p->p_sysent->sv_prepsyscall) {
946 /*
947 * The prep code is MP aware.
948 */
949 (*p->p_sysent->sv_prepsyscall)(&frame, args, &code, ¶ms);
950 } else {
951 /*
952 * Need to check if this is a 32 bit or 64 bit syscall.
953 * fuword is MP aware.
954 */
955 if (code == SYS_syscall) {
956 /*
957 * Code is first argument, followed by actual args.
958 */
959 code = fuword(params);
960 params += sizeof(int);
961 } else if (code == SYS___syscall) {
962 /*
963 * Like syscall, but code is a quad, so as to maintain
964 * quad alignment for the rest of the arguments.
965 */
966 code = fuword(params);
967 params += sizeof(quad_t);
968 }
969 }
970
971 if (p->p_sysent->sv_mask)
972 code &= p->p_sysent->sv_mask;
973
974 if (code >= p->p_sysent->sv_size)
975 callp = &p->p_sysent->sv_table[0];
976 else
977 callp = &p->p_sysent->sv_table[code];
978
979 narg = callp->sy_narg & SYF_ARGMASK;
980
981 /*
982 * copyin and the ktrsyscall()/ktrsysret() code is MP-aware
983 */
984 if (params != NULL && narg != 0)
985 error = copyin(params, (caddr_t)args,
986 (u_int)(narg * sizeof(int)));
987 else
988 error = 0;
989
990 #ifdef KTRACE
991 if (KTRPOINT(td, KTR_SYSCALL))
992 ktrsyscall(code, narg, args);
993 #endif
994
995 /*
996 * Try to run the syscall without Giant if the syscall
997 * is MP safe.
998 */
999 if ((callp->sy_narg & SYF_MPSAFE) == 0)
1000 mtx_lock(&Giant);
1001
1002 if (error == 0) {
1003 td->td_retval[0] = 0;
1004 td->td_retval[1] = frame.tf_edx;
1005
1006 STOPEVENT(p, S_SCE, narg);
1007
1008 PTRACESTOP_SC(p, td, S_PT_SCE);
1009
1010 error = (*callp->sy_call)(td, args);
1011 }
1012
1013 switch (error) {
1014 case 0:
1015 frame.tf_eax = td->td_retval[0];
1016 frame.tf_edx = td->td_retval[1];
1017 frame.tf_eflags &= ~PSL_C;
1018 break;
1019
1020 case ERESTART:
1021 /*
1022 * Reconstruct pc, assuming lcall $X,y is 7 bytes,
1023 * int 0x80 is 2 bytes. We saved this in tf_err.
1024 */
1025 frame.tf_eip -= frame.tf_err;
1026 break;
1027
1028 case EJUSTRETURN:
1029 break;
1030
1031 default:
1032 if (p->p_sysent->sv_errsize) {
1033 if (error >= p->p_sysent->sv_errsize)
1034 error = -1; /* XXX */
1035 else
1036 error = p->p_sysent->sv_errtbl[error];
1037 }
1038 frame.tf_eax = error;
1039 frame.tf_eflags |= PSL_C;
1040 break;
1041 }
1042
1043 /*
1044 * Release Giant if we previously set it.
1045 */
1046 if ((callp->sy_narg & SYF_MPSAFE) == 0)
1047 mtx_unlock(&Giant);
1048
1049 /*
1050 * Traced syscall.
1051 */
1052 if ((orig_tf_eflags & PSL_T) && !(orig_tf_eflags & PSL_VM)) {
1053 frame.tf_eflags &= ~PSL_T;
1054 trapsignal(td, SIGTRAP, 0);
1055 }
1056
1057 /*
1058 * Handle reschedule and other end-of-syscall issues
1059 */
1060 userret(td, &frame, sticks);
1061
1062 #ifdef KTRACE
1063 if (KTRPOINT(td, KTR_SYSRET))
1064 ktrsysret(code, error, td->td_retval[0]);
1065 #endif
1066
1067 /*
1068 * This works because errno is findable through the
1069 * register set. If we ever support an emulation where this
1070 * is not the case, this code will need to be revisited.
1071 */
1072 STOPEVENT(p, S_SCX, code);
1073
1074 PTRACESTOP_SC(p, td, S_PT_SCX);
1075
1076 #ifdef DIAGNOSTIC
1077 cred_free_thread(td);
1078 #endif
1079 WITNESS_WARN(WARN_PANIC, NULL, "System call %s returning",
1080 (code >= 0 && code < SYS_MAXSYSCALL) ? syscallnames[code] : "???");
1081 mtx_assert(&sched_lock, MA_NOTOWNED);
1082 mtx_assert(&Giant, MA_NOTOWNED);
1083 }
1084
Cache object: effc05f266187519b051f119898c9e9f
|