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 * $FreeBSD: releng/5.1/sys/i386/i386/trap.c 115184 2003-05-20 20:50:33Z jhb $
39 */
40
41 /*
42 * 386 Trap and System call handling
43 */
44
45 #include "opt_clock.h"
46 #include "opt_cpu.h"
47 #include "opt_ddb.h"
48 #include "opt_isa.h"
49 #include "opt_ktrace.h"
50 #include "opt_npx.h"
51 #include "opt_trap.h"
52
53 #include <sys/param.h>
54 #include <sys/bus.h>
55 #include <sys/systm.h>
56 #include <sys/proc.h>
57 #include <sys/pioctl.h>
58 #include <sys/kernel.h>
59 #include <sys/ktr.h>
60 #include <sys/lock.h>
61 #include <sys/mutex.h>
62 #include <sys/resourcevar.h>
63 #include <sys/signalvar.h>
64 #include <sys/syscall.h>
65 #include <sys/sysctl.h>
66 #include <sys/sysent.h>
67 #include <sys/uio.h>
68 #include <sys/vmmeter.h>
69 #ifdef KTRACE
70 #include <sys/ktrace.h>
71 #endif
72
73 #include <vm/vm.h>
74 #include <vm/vm_param.h>
75 #include <vm/pmap.h>
76 #include <vm/vm_kern.h>
77 #include <vm/vm_map.h>
78 #include <vm/vm_page.h>
79 #include <vm/vm_extern.h>
80
81 #include <machine/cpu.h>
82 #include <machine/md_var.h>
83 #include <machine/pcb.h>
84 #ifdef SMP
85 #include <machine/smp.h>
86 #endif
87 #include <machine/tss.h>
88
89 #include <i386/isa/icu.h>
90 #include <i386/isa/intr_machdep.h>
91
92 #ifdef POWERFAIL_NMI
93 #include <sys/syslog.h>
94 #include <machine/clock.h>
95 #endif
96
97 #include <machine/vm86.h>
98
99 #include <ddb/ddb.h>
100
101 #include <sys/sysctl.h>
102
103 int (*pmath_emulate)(struct trapframe *);
104
105 extern void trap(struct trapframe frame);
106 #ifdef I386_CPU
107 extern int trapwrite(unsigned addr);
108 #endif
109 extern void syscall(struct trapframe frame);
110
111 static int trap_pfault(struct trapframe *, int, vm_offset_t);
112 static void trap_fatal(struct trapframe *, vm_offset_t);
113 void dblfault_handler(void);
114
115 extern inthand_t IDTVEC(lcall_syscall);
116
117 #define MAX_TRAP_MSG 28
118 static char *trap_msg[] = {
119 "", /* 0 unused */
120 "privileged instruction fault", /* 1 T_PRIVINFLT */
121 "", /* 2 unused */
122 "breakpoint instruction fault", /* 3 T_BPTFLT */
123 "", /* 4 unused */
124 "", /* 5 unused */
125 "arithmetic trap", /* 6 T_ARITHTRAP */
126 "", /* 7 unused */
127 "", /* 8 unused */
128 "general protection fault", /* 9 T_PROTFLT */
129 "trace trap", /* 10 T_TRCTRAP */
130 "", /* 11 unused */
131 "page fault", /* 12 T_PAGEFLT */
132 "", /* 13 unused */
133 "alignment fault", /* 14 T_ALIGNFLT */
134 "", /* 15 unused */
135 "", /* 16 unused */
136 "", /* 17 unused */
137 "integer divide fault", /* 18 T_DIVIDE */
138 "non-maskable interrupt trap", /* 19 T_NMI */
139 "overflow trap", /* 20 T_OFLOW */
140 "FPU bounds check fault", /* 21 T_BOUND */
141 "FPU device not available", /* 22 T_DNA */
142 "double fault", /* 23 T_DOUBLEFLT */
143 "FPU operand fetch fault", /* 24 T_FPOPFLT */
144 "invalid TSS fault", /* 25 T_TSSFLT */
145 "segment not present fault", /* 26 T_SEGNPFLT */
146 "stack fault", /* 27 T_STKFLT */
147 "machine check trap", /* 28 T_MCHK */
148 };
149
150 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
151 extern int has_f00f_bug;
152 #endif
153
154 #ifdef DDB
155 static int ddb_on_nmi = 1;
156 SYSCTL_INT(_machdep, OID_AUTO, ddb_on_nmi, CTLFLAG_RW,
157 &ddb_on_nmi, 0, "Go to DDB on NMI");
158 #endif
159 static int panic_on_nmi = 1;
160 SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RW,
161 &panic_on_nmi, 0, "Panic on NMI");
162
163 #ifdef WITNESS
164 extern char *syscallnames[];
165 #endif
166
167 #ifdef DEVICE_POLLING
168 extern u_int32_t poll_in_trap;
169 extern int ether_poll(int count);
170 #endif /* DEVICE_POLLING */
171
172 /*
173 * Exception, fault, and trap interface to the FreeBSD kernel.
174 * This common code is called from assembly language IDT gate entry
175 * routines that prepare a suitable stack frame, and restore this
176 * frame after the exception has been processed.
177 */
178
179 void
180 trap(frame)
181 struct trapframe frame;
182 {
183 struct thread *td = curthread;
184 struct proc *p = td->td_proc;
185 u_int sticks = 0;
186 int i = 0, ucode = 0, type, code;
187 vm_offset_t eva;
188 #ifdef POWERFAIL_NMI
189 static int lastalert = 0;
190 #endif
191
192 atomic_add_int(&cnt.v_trap, 1);
193 type = frame.tf_trapno;
194
195 #ifdef DDB
196 if (db_active) {
197 eva = (type == T_PAGEFLT ? rcr2() : 0);
198 trap_fatal(&frame, eva);
199 goto out;
200 }
201 #endif
202
203 if ((frame.tf_eflags & PSL_I) == 0) {
204 /*
205 * Buggy application or kernel code has disabled
206 * interrupts and then trapped. Enabling interrupts
207 * now is wrong, but it is better than running with
208 * interrupts disabled until they are accidentally
209 * enabled later.
210 */
211 if (ISPL(frame.tf_cs) == SEL_UPL || (frame.tf_eflags & PSL_VM))
212 printf(
213 "pid %ld (%s): trap %d with interrupts disabled\n",
214 (long)curproc->p_pid, curproc->p_comm, type);
215 else if (type != T_BPTFLT && type != T_TRCTRAP &&
216 frame.tf_eip != (int)cpu_switch_load_gs) {
217 /*
218 * XXX not quite right, since this may be for a
219 * multiple fault in user mode.
220 */
221 printf("kernel trap %d with interrupts disabled\n",
222 type);
223 /*
224 * Page faults need interrupts diasabled until later,
225 * and we shouldn't enable interrupts while in a
226 * critical section.
227 */
228 if (type != T_PAGEFLT && td->td_critnest == 0)
229 enable_intr();
230 }
231 }
232
233 eva = 0;
234 code = frame.tf_err;
235 if (type == T_PAGEFLT) {
236 /*
237 * For some Cyrix CPUs, %cr2 is clobbered by
238 * interrupts. This problem is worked around by using
239 * an interrupt gate for the pagefault handler. We
240 * are finally ready to read %cr2 and then must
241 * reenable interrupts.
242 *
243 * If we get a page fault while in a critical section, then
244 * it is most likely a fatal kernel page fault. The kernel
245 * is already going to panic trying to get a sleep lock to
246 * do the VM lookup, so just consider it a fatal trap so the
247 * kernel can print out a useful trap message and even get
248 * to the debugger.
249 */
250 eva = rcr2();
251 if (td->td_critnest == 0)
252 enable_intr();
253 else
254 trap_fatal(&frame, eva);
255 }
256
257 #ifdef DEVICE_POLLING
258 if (poll_in_trap)
259 ether_poll(poll_in_trap);
260 #endif /* DEVICE_POLLING */
261
262 if ((ISPL(frame.tf_cs) == SEL_UPL) ||
263 ((frame.tf_eflags & PSL_VM) &&
264 !(PCPU_GET(curpcb)->pcb_flags & PCB_VM86CALL))) {
265 /* user trap */
266
267 sticks = td->td_sticks;
268 td->td_frame = &frame;
269 if (td->td_ucred != p->p_ucred)
270 cred_update_thread(td);
271
272 switch (type) {
273 case T_PRIVINFLT: /* privileged instruction fault */
274 ucode = type;
275 i = SIGILL;
276 break;
277
278 case T_BPTFLT: /* bpt instruction fault */
279 case T_TRCTRAP: /* trace trap */
280 frame.tf_eflags &= ~PSL_T;
281 i = SIGTRAP;
282 break;
283
284 case T_ARITHTRAP: /* arithmetic trap */
285 #ifdef DEV_NPX
286 ucode = npxtrap();
287 if (ucode == -1)
288 goto userout;
289 #else
290 ucode = code;
291 #endif
292 i = SIGFPE;
293 break;
294
295 /*
296 * The following two traps can happen in
297 * vm86 mode, and, if so, we want to handle
298 * them specially.
299 */
300 case T_PROTFLT: /* general protection fault */
301 case T_STKFLT: /* stack fault */
302 if (frame.tf_eflags & PSL_VM) {
303 i = vm86_emulate((struct vm86frame *)&frame);
304 if (i == 0)
305 goto user;
306 break;
307 }
308 /* FALLTHROUGH */
309
310 case T_SEGNPFLT: /* segment not present fault */
311 case T_TSSFLT: /* invalid TSS fault */
312 case T_DOUBLEFLT: /* double fault */
313 default:
314 ucode = code + BUS_SEGM_FAULT ;
315 i = SIGBUS;
316 break;
317
318 case T_PAGEFLT: /* page fault */
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 if (!pmath_emulate) {
400 i = SIGFPE;
401 ucode = FPE_FPU_NP_TRAP;
402 break;
403 }
404 mtx_lock(&Giant);
405 i = (*pmath_emulate)(&frame);
406 mtx_unlock(&Giant);
407 if (i == 0) {
408 if (!(frame.tf_eflags & PSL_T))
409 goto userout;
410 frame.tf_eflags &= ~PSL_T;
411 i = SIGTRAP;
412 }
413 /* else ucode = emulator_only_knows() XXX */
414 break;
415
416 case T_FPOPFLT: /* FPU operand fetch fault */
417 ucode = T_FPOPFLT;
418 i = SIGILL;
419 break;
420
421 case T_XMMFLT: /* SIMD floating-point exception */
422 ucode = 0; /* XXX */
423 i = SIGFPE;
424 break;
425 }
426 } else {
427 /* kernel trap */
428
429 KASSERT(cold || td->td_ucred != NULL,
430 ("kernel trap doesn't have ucred"));
431 switch (type) {
432 case T_PAGEFLT: /* page fault */
433 (void) trap_pfault(&frame, FALSE, eva);
434 goto out;
435
436 case T_DNA:
437 #ifdef DEV_NPX
438 /*
439 * The kernel is apparently using npx for copying.
440 * XXX this should be fatal unless the kernel has
441 * registered such use.
442 */
443 if (npxdna())
444 goto out;
445 #endif
446 break;
447
448 /*
449 * The following two traps can happen in
450 * vm86 mode, and, if so, we want to handle
451 * them specially.
452 */
453 case T_PROTFLT: /* general protection fault */
454 case T_STKFLT: /* stack fault */
455 if (frame.tf_eflags & PSL_VM) {
456 i = vm86_emulate((struct vm86frame *)&frame);
457 if (i != 0)
458 /*
459 * returns to original process
460 */
461 vm86_trap((struct vm86frame *)&frame);
462 goto out;
463 }
464 if (type == T_STKFLT)
465 break;
466
467 /* FALL THROUGH */
468
469 case T_SEGNPFLT: /* segment not present fault */
470 if (PCPU_GET(curpcb)->pcb_flags & PCB_VM86CALL)
471 break;
472
473 /*
474 * Invalid %fs's and %gs's can be created using
475 * procfs or PT_SETREGS or by invalidating the
476 * underlying LDT entry. This causes a fault
477 * in kernel mode when the kernel attempts to
478 * switch contexts. Lose the bad context
479 * (XXX) so that we can continue, and generate
480 * a signal.
481 */
482 if (frame.tf_eip == (int)cpu_switch_load_gs) {
483 PCPU_GET(curpcb)->pcb_gs = 0;
484 #if 0
485 PROC_LOCK(p);
486 psignal(p, SIGBUS);
487 PROC_UNLOCK(p);
488 #endif
489 goto out;
490 }
491
492 if (td->td_intr_nesting_level != 0)
493 break;
494
495 /*
496 * Invalid segment selectors and out of bounds
497 * %eip's and %esp's can be set up in user mode.
498 * This causes a fault in kernel mode when the
499 * kernel tries to return to user mode. We want
500 * to get this fault so that we can fix the
501 * problem here and not have to check all the
502 * selectors and pointers when the user changes
503 * them.
504 */
505 if (frame.tf_eip == (int)doreti_iret) {
506 frame.tf_eip = (int)doreti_iret_fault;
507 goto out;
508 }
509 if (frame.tf_eip == (int)doreti_popl_ds) {
510 frame.tf_eip = (int)doreti_popl_ds_fault;
511 goto out;
512 }
513 if (frame.tf_eip == (int)doreti_popl_es) {
514 frame.tf_eip = (int)doreti_popl_es_fault;
515 goto out;
516 }
517 if (frame.tf_eip == (int)doreti_popl_fs) {
518 frame.tf_eip = (int)doreti_popl_fs_fault;
519 goto out;
520 }
521 if (PCPU_GET(curpcb) != NULL &&
522 PCPU_GET(curpcb)->pcb_onfault != NULL) {
523 frame.tf_eip =
524 (int)PCPU_GET(curpcb)->pcb_onfault;
525 goto out;
526 }
527 break;
528
529 case T_TSSFLT:
530 /*
531 * PSL_NT can be set in user mode and isn't cleared
532 * automatically when the kernel is entered. This
533 * causes a TSS fault when the kernel attempts to
534 * `iret' because the TSS link is uninitialized. We
535 * want to get this fault so that we can fix the
536 * problem here and not every time the kernel is
537 * entered.
538 */
539 if (frame.tf_eflags & PSL_NT) {
540 frame.tf_eflags &= ~PSL_NT;
541 goto out;
542 }
543 break;
544
545 case T_TRCTRAP: /* trace trap */
546 if (frame.tf_eip == (int)IDTVEC(lcall_syscall)) {
547 /*
548 * We've just entered system mode via the
549 * syscall lcall. Continue single stepping
550 * silently until the syscall handler has
551 * saved the flags.
552 */
553 goto out;
554 }
555 if (frame.tf_eip == (int)IDTVEC(lcall_syscall) + 1) {
556 /*
557 * The syscall handler has now saved the
558 * flags. Stop single stepping it.
559 */
560 frame.tf_eflags &= ~PSL_T;
561 goto out;
562 }
563 /*
564 * Ignore debug register trace traps due to
565 * accesses in the user's address space, which
566 * can happen under several conditions such as
567 * if a user sets a watchpoint on a buffer and
568 * then passes that buffer to a system call.
569 * We still want to get TRCTRAPS for addresses
570 * in kernel space because that is useful when
571 * debugging the kernel.
572 */
573 /* XXX Giant */
574 if (user_dbreg_trap() &&
575 !(PCPU_GET(curpcb)->pcb_flags & PCB_VM86CALL)) {
576 /*
577 * Reset breakpoint bits because the
578 * processor doesn't
579 */
580 load_dr6(rdr6() & 0xfffffff0);
581 goto out;
582 }
583 /*
584 * FALLTHROUGH (TRCTRAP kernel mode, kernel address)
585 */
586 case T_BPTFLT:
587 /*
588 * If DDB is enabled, let it handle the debugger trap.
589 * Otherwise, debugger traps "can't happen".
590 */
591 #ifdef DDB
592 /* XXX Giant */
593 if (kdb_trap (type, 0, &frame))
594 goto out;
595 #endif
596 break;
597
598 #ifdef DEV_ISA
599 case T_NMI:
600 #ifdef POWERFAIL_NMI
601 mtx_lock(&Giant);
602 if (time_second - lastalert > 10) {
603 log(LOG_WARNING, "NMI: power fail\n");
604 sysbeep(TIMER_FREQ/880, hz);
605 lastalert = time_second;
606 }
607 mtx_unlock(&Giant);
608 goto out;
609 #else /* !POWERFAIL_NMI */
610 /* XXX Giant */
611 /* machine/parity/power fail/"kitchen sink" faults */
612 if (isa_nmi(code) == 0) {
613 #ifdef DDB
614 /*
615 * NMI can be hooked up to a pushbutton
616 * for debugging.
617 */
618 if (ddb_on_nmi) {
619 printf ("NMI ... going to debugger\n");
620 kdb_trap (type, 0, &frame);
621 }
622 #endif /* DDB */
623 goto out;
624 } else if (panic_on_nmi == 0)
625 goto out;
626 /* FALLTHROUGH */
627 #endif /* POWERFAIL_NMI */
628 #endif /* DEV_ISA */
629 }
630
631 trap_fatal(&frame, eva);
632 goto out;
633 }
634
635 /* Translate fault for emulators (e.g. Linux) */
636 if (*p->p_sysent->sv_transtrap)
637 i = (*p->p_sysent->sv_transtrap)(i, type);
638
639 trapsignal(td, i, ucode);
640
641 #ifdef DEBUG
642 if (type <= MAX_TRAP_MSG) {
643 uprintf("fatal process exception: %s",
644 trap_msg[type]);
645 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
646 uprintf(", fault VA = 0x%lx", (u_long)eva);
647 uprintf("\n");
648 }
649 #endif
650
651 user:
652 userret(td, &frame, sticks);
653 mtx_assert(&Giant, MA_NOTOWNED);
654 userout:
655 #ifdef DIAGNOSTIC
656 cred_free_thread(td);
657 #endif
658 out:
659 return;
660 }
661
662 static int
663 trap_pfault(frame, usermode, eva)
664 struct trapframe *frame;
665 int usermode;
666 vm_offset_t eva;
667 {
668 vm_offset_t va;
669 struct vmspace *vm = NULL;
670 vm_map_t map = 0;
671 int rv = 0;
672 vm_prot_t ftype;
673 struct thread *td = curthread;
674 struct proc *p = td->td_proc;
675
676 va = trunc_page(eva);
677 if (va >= KERNBASE) {
678 /*
679 * Don't allow user-mode faults in kernel address space.
680 * An exception: if the faulting address is the invalid
681 * instruction entry in the IDT, then the Intel Pentium
682 * F00F bug workaround was triggered, and we need to
683 * treat it is as an illegal instruction, and not a page
684 * fault.
685 */
686 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
687 if ((eva == (unsigned int)&idt[6]) && has_f00f_bug)
688 return -2;
689 #endif
690 if (usermode)
691 goto nogo;
692
693 map = kernel_map;
694 } else {
695 /*
696 * This is a fault on non-kernel virtual memory.
697 * vm is initialized above to NULL. If curproc is NULL
698 * or curproc->p_vmspace is NULL the fault is fatal.
699 */
700 if (p != NULL)
701 vm = p->p_vmspace;
702
703 if (vm == NULL)
704 goto nogo;
705
706 map = &vm->vm_map;
707 }
708
709 if (frame->tf_err & PGEX_W)
710 ftype = VM_PROT_WRITE;
711 else
712 ftype = VM_PROT_READ;
713
714 if (map != kernel_map) {
715 /*
716 * Keep swapout from messing with us during this
717 * critical time.
718 */
719 PROC_LOCK(p);
720 ++p->p_lock;
721 PROC_UNLOCK(p);
722
723 /* Fault in the user page: */
724 rv = vm_fault(map, va, ftype,
725 (ftype & VM_PROT_WRITE) ? VM_FAULT_DIRTY
726 : VM_FAULT_NORMAL);
727
728 PROC_LOCK(p);
729 --p->p_lock;
730 PROC_UNLOCK(p);
731 } else {
732 /*
733 * Don't have to worry about process locking or stacks in the
734 * kernel.
735 */
736 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
737 }
738 if (rv == KERN_SUCCESS)
739 return (0);
740 nogo:
741 if (!usermode) {
742 if (td->td_intr_nesting_level == 0 &&
743 PCPU_GET(curpcb) != NULL &&
744 PCPU_GET(curpcb)->pcb_onfault != NULL) {
745 frame->tf_eip = (int)PCPU_GET(curpcb)->pcb_onfault;
746 return (0);
747 }
748 trap_fatal(frame, eva);
749 return (-1);
750 }
751
752 /* kludge to pass faulting virtual address to sendsig */
753 frame->tf_err = eva;
754
755 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
756 }
757
758 static void
759 trap_fatal(frame, eva)
760 struct trapframe *frame;
761 vm_offset_t eva;
762 {
763 int code, type, ss, esp;
764 struct soft_segment_descriptor softseg;
765
766 code = frame->tf_err;
767 type = frame->tf_trapno;
768 sdtossd(&gdt[IDXSEL(frame->tf_cs & 0xffff)].sd, &softseg);
769
770 if (type <= MAX_TRAP_MSG)
771 printf("\n\nFatal trap %d: %s while in %s mode\n",
772 type, trap_msg[type],
773 frame->tf_eflags & PSL_VM ? "vm86" :
774 ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel");
775 #ifdef SMP
776 /* two separate prints in case of a trap on an unmapped page */
777 printf("cpuid = %d; ", PCPU_GET(cpuid));
778 printf("lapic.id = %08x\n", lapic.id);
779 #endif
780 if (type == T_PAGEFLT) {
781 printf("fault virtual address = 0x%x\n", eva);
782 printf("fault code = %s %s, %s\n",
783 code & PGEX_U ? "user" : "supervisor",
784 code & PGEX_W ? "write" : "read",
785 code & PGEX_P ? "protection violation" : "page not present");
786 }
787 printf("instruction pointer = 0x%x:0x%x\n",
788 frame->tf_cs & 0xffff, frame->tf_eip);
789 if ((ISPL(frame->tf_cs) == SEL_UPL) || (frame->tf_eflags & PSL_VM)) {
790 ss = frame->tf_ss & 0xffff;
791 esp = frame->tf_esp;
792 } else {
793 ss = GSEL(GDATA_SEL, SEL_KPL);
794 esp = (int)&frame->tf_esp;
795 }
796 printf("stack pointer = 0x%x:0x%x\n", ss, esp);
797 printf("frame pointer = 0x%x:0x%x\n", ss, frame->tf_ebp);
798 printf("code segment = base 0x%x, limit 0x%x, type 0x%x\n",
799 softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type);
800 printf(" = DPL %d, pres %d, def32 %d, gran %d\n",
801 softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_def32,
802 softseg.ssd_gran);
803 printf("processor eflags = ");
804 if (frame->tf_eflags & PSL_T)
805 printf("trace trap, ");
806 if (frame->tf_eflags & PSL_I)
807 printf("interrupt enabled, ");
808 if (frame->tf_eflags & PSL_NT)
809 printf("nested task, ");
810 if (frame->tf_eflags & PSL_RF)
811 printf("resume, ");
812 if (frame->tf_eflags & PSL_VM)
813 printf("vm86, ");
814 printf("IOPL = %d\n", (frame->tf_eflags & PSL_IOPL) >> 12);
815 printf("current process = ");
816 if (curproc) {
817 printf("%lu (%s)\n",
818 (u_long)curproc->p_pid, curproc->p_comm ?
819 curproc->p_comm : "");
820 } else {
821 printf("Idle\n");
822 }
823
824 #ifdef KDB
825 if (kdb_trap(&psl))
826 return;
827 #endif
828 #ifdef DDB
829 if ((debugger_on_panic || db_active) && kdb_trap(type, 0, frame))
830 return;
831 #endif
832 printf("trap number = %d\n", type);
833 if (type <= MAX_TRAP_MSG)
834 panic("%s", trap_msg[type]);
835 else
836 panic("unknown/reserved trap");
837 }
838
839 /*
840 * Double fault handler. Called when a fault occurs while writing
841 * a frame for a trap/exception onto the stack. This usually occurs
842 * when the stack overflows (such is the case with infinite recursion,
843 * for example).
844 *
845 * XXX Note that the current PTD gets replaced by IdlePTD when the
846 * task switch occurs. This means that the stack that was active at
847 * the time of the double fault is not available at <kstack> unless
848 * the machine was idle when the double fault occurred. The downside
849 * of this is that "trace <ebp>" in ddb won't work.
850 */
851 void
852 dblfault_handler()
853 {
854 printf("\nFatal double fault:\n");
855 printf("eip = 0x%x\n", PCPU_GET(common_tss.tss_eip));
856 printf("esp = 0x%x\n", PCPU_GET(common_tss.tss_esp));
857 printf("ebp = 0x%x\n", PCPU_GET(common_tss.tss_ebp));
858 #ifdef SMP
859 /* two separate prints in case of a trap on an unmapped page */
860 printf("cpuid = %d; ", PCPU_GET(cpuid));
861 printf("lapic.id = %08x\n", lapic.id);
862 #endif
863 panic("double fault");
864 }
865
866 #ifdef I386_CPU
867 /*
868 * Compensate for 386 brain damage (missing URKR).
869 * This is a little simpler than the pagefault handler in trap() because
870 * it the page tables have already been faulted in and high addresses
871 * are thrown out early for other reasons.
872 */
873 int trapwrite(addr)
874 unsigned addr;
875 {
876 struct thread *td;
877 struct proc *p;
878 vm_offset_t va;
879 struct vmspace *vm;
880 int rv;
881
882 va = trunc_page((vm_offset_t)addr);
883 /*
884 * XXX - MAX is END. Changed > to >= for temp. fix.
885 */
886 if (va >= VM_MAXUSER_ADDRESS)
887 return (1);
888
889 td = curthread;
890 p = td->td_proc;
891 vm = p->p_vmspace;
892
893 PROC_LOCK(p);
894 ++p->p_lock;
895 PROC_UNLOCK(p);
896
897 /*
898 * fault the data page
899 */
900 rv = vm_fault(&vm->vm_map, va, VM_PROT_WRITE, VM_FAULT_DIRTY);
901
902 PROC_LOCK(p);
903 --p->p_lock;
904 PROC_UNLOCK(p);
905
906 if (rv != KERN_SUCCESS)
907 return 1;
908
909 return (0);
910 }
911 #endif
912
913 /*
914 * syscall - system call request C handler
915 *
916 * A system call is essentially treated as a trap.
917 */
918 void
919 syscall(frame)
920 struct trapframe frame;
921 {
922 caddr_t params;
923 struct sysent *callp;
924 struct thread *td = curthread;
925 struct proc *p = td->td_proc;
926 register_t orig_tf_eflags;
927 u_int sticks;
928 int error;
929 int narg;
930 int args[8];
931 u_int code;
932
933 /*
934 * note: PCPU_LAZY_INC() can only be used if we can afford
935 * occassional inaccuracy in the count.
936 */
937 PCPU_LAZY_INC(cnt.v_syscall);
938
939 #ifdef DIAGNOSTIC
940 if (ISPL(frame.tf_cs) != SEL_UPL) {
941 mtx_lock(&Giant); /* try to stabilize the system XXX */
942 panic("syscall");
943 /* NOT REACHED */
944 mtx_unlock(&Giant);
945 }
946 #endif
947
948 sticks = td->td_sticks;
949 td->td_frame = &frame;
950 if (td->td_ucred != p->p_ucred)
951 cred_update_thread(td);
952 if (p->p_flag & P_THREADED)
953 thread_user_enter(p, td);
954 params = (caddr_t)frame.tf_esp + sizeof(int);
955 code = frame.tf_eax;
956 orig_tf_eflags = frame.tf_eflags;
957
958 if (p->p_sysent->sv_prepsyscall) {
959 /*
960 * The prep code is MP aware.
961 */
962 (*p->p_sysent->sv_prepsyscall)(&frame, args, &code, ¶ms);
963 } else {
964 /*
965 * Need to check if this is a 32 bit or 64 bit syscall.
966 * fuword is MP aware.
967 */
968 if (code == SYS_syscall) {
969 /*
970 * Code is first argument, followed by actual args.
971 */
972 code = fuword(params);
973 params += sizeof(int);
974 } else if (code == SYS___syscall) {
975 /*
976 * Like syscall, but code is a quad, so as to maintain
977 * quad alignment for the rest of the arguments.
978 */
979 code = fuword(params);
980 params += sizeof(quad_t);
981 }
982 }
983
984 if (p->p_sysent->sv_mask)
985 code &= p->p_sysent->sv_mask;
986
987 if (code >= p->p_sysent->sv_size)
988 callp = &p->p_sysent->sv_table[0];
989 else
990 callp = &p->p_sysent->sv_table[code];
991
992 narg = callp->sy_narg & SYF_ARGMASK;
993
994 /*
995 * copyin and the ktrsyscall()/ktrsysret() code is MP-aware
996 */
997 if (params != NULL && narg != 0)
998 error = copyin(params, (caddr_t)args,
999 (u_int)(narg * sizeof(int)));
1000 else
1001 error = 0;
1002
1003 #ifdef KTRACE
1004 if (KTRPOINT(td, KTR_SYSCALL))
1005 ktrsyscall(code, narg, args);
1006 #endif
1007
1008 /*
1009 * Try to run the syscall without Giant if the syscall
1010 * is MP safe.
1011 */
1012 if ((callp->sy_narg & SYF_MPSAFE) == 0)
1013 mtx_lock(&Giant);
1014
1015 if (error == 0) {
1016 td->td_retval[0] = 0;
1017 td->td_retval[1] = frame.tf_edx;
1018
1019 STOPEVENT(p, S_SCE, narg);
1020
1021 error = (*callp->sy_call)(td, args);
1022 }
1023
1024 switch (error) {
1025 case 0:
1026 frame.tf_eax = td->td_retval[0];
1027 frame.tf_edx = td->td_retval[1];
1028 frame.tf_eflags &= ~PSL_C;
1029 break;
1030
1031 case ERESTART:
1032 /*
1033 * Reconstruct pc, assuming lcall $X,y is 7 bytes,
1034 * int 0x80 is 2 bytes. We saved this in tf_err.
1035 */
1036 frame.tf_eip -= frame.tf_err;
1037 break;
1038
1039 case EJUSTRETURN:
1040 break;
1041
1042 default:
1043 if (p->p_sysent->sv_errsize) {
1044 if (error >= p->p_sysent->sv_errsize)
1045 error = -1; /* XXX */
1046 else
1047 error = p->p_sysent->sv_errtbl[error];
1048 }
1049 frame.tf_eax = error;
1050 frame.tf_eflags |= PSL_C;
1051 break;
1052 }
1053
1054 /*
1055 * Release Giant if we previously set it.
1056 */
1057 if ((callp->sy_narg & SYF_MPSAFE) == 0)
1058 mtx_unlock(&Giant);
1059
1060 /*
1061 * Traced syscall.
1062 */
1063 if ((orig_tf_eflags & PSL_T) && !(orig_tf_eflags & PSL_VM)) {
1064 frame.tf_eflags &= ~PSL_T;
1065 trapsignal(td, SIGTRAP, 0);
1066 }
1067
1068 /*
1069 * Handle reschedule and other end-of-syscall issues
1070 */
1071 userret(td, &frame, sticks);
1072
1073 #ifdef KTRACE
1074 if (KTRPOINT(td, KTR_SYSRET))
1075 ktrsysret(code, error, td->td_retval[0]);
1076 #endif
1077
1078 /*
1079 * This works because errno is findable through the
1080 * register set. If we ever support an emulation where this
1081 * is not the case, this code will need to be revisited.
1082 */
1083 STOPEVENT(p, S_SCX, code);
1084
1085 #ifdef DIAGNOSTIC
1086 cred_free_thread(td);
1087 #endif
1088 WITNESS_WARN(WARN_PANIC, NULL, "System call %s returning",
1089 (code >= 0 && code < SYS_MAXSYSCALL) ? syscallnames[code] : "???");
1090 mtx_assert(&sched_lock, MA_NOTOWNED);
1091 mtx_assert(&Giant, MA_NOTOWNED);
1092 }
1093
Cache object: 3bfe0643d6b46e080304ac065bbad9ec
|