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("$FreeBSD$");
42
43 /*
44 * 386 Trap and System call handling
45 */
46
47 #include "opt_clock.h"
48 #include "opt_cpu.h"
49 #include "opt_isa.h"
50 #include "opt_ktrace.h"
51 #include "opt_npx.h"
52 #include "opt_trap.h"
53
54 #include <sys/param.h>
55 #include <sys/bus.h>
56 #include <sys/systm.h>
57 #include <sys/proc.h>
58 #include <sys/pioctl.h>
59 #include <sys/ptrace.h>
60 #include <sys/kdb.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 extern void trap(struct trapframe frame);
100 #ifdef I386_CPU
101 extern int trapwrite(unsigned addr);
102 #endif
103 extern void syscall(struct trapframe frame);
104
105 static int trap_pfault(struct trapframe *, int, vm_offset_t);
106 static void trap_fatal(struct trapframe *, vm_offset_t);
107 void dblfault_handler(void);
108
109 extern inthand_t IDTVEC(lcall_syscall);
110
111 #define MAX_TRAP_MSG 30
112 static char *trap_msg[] = {
113 "", /* 0 unused */
114 "privileged instruction fault", /* 1 T_PRIVINFLT */
115 "", /* 2 unused */
116 "breakpoint instruction fault", /* 3 T_BPTFLT */
117 "", /* 4 unused */
118 "", /* 5 unused */
119 "arithmetic trap", /* 6 T_ARITHTRAP */
120 "", /* 7 unused */
121 "", /* 8 unused */
122 "general protection fault", /* 9 T_PROTFLT */
123 "trace trap", /* 10 T_TRCTRAP */
124 "", /* 11 unused */
125 "page fault", /* 12 T_PAGEFLT */
126 "", /* 13 unused */
127 "alignment fault", /* 14 T_ALIGNFLT */
128 "", /* 15 unused */
129 "", /* 16 unused */
130 "", /* 17 unused */
131 "integer divide fault", /* 18 T_DIVIDE */
132 "non-maskable interrupt trap", /* 19 T_NMI */
133 "overflow trap", /* 20 T_OFLOW */
134 "FPU bounds check fault", /* 21 T_BOUND */
135 "FPU device not available", /* 22 T_DNA */
136 "double fault", /* 23 T_DOUBLEFLT */
137 "FPU operand fetch fault", /* 24 T_FPOPFLT */
138 "invalid TSS fault", /* 25 T_TSSFLT */
139 "segment not present fault", /* 26 T_SEGNPFLT */
140 "stack fault", /* 27 T_STKFLT */
141 "machine check trap", /* 28 T_MCHK */
142 "SIMD floating-point exception", /* 29 T_XMMFLT */
143 "reserved (unknown) fault", /* 30 T_RESERVED */
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 #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 KDB_STOP_NMI
192 /* Handler for NMI IPIs used for debugging */
193 if (type == T_NMI) {
194 if (ipi_nmi_handler() == 0)
195 goto out;
196 }
197 #endif /* KDB_STOP_NMI */
198
199 #ifdef KDB
200 if (kdb_active) {
201 kdb_reenter();
202 goto out;
203 }
204 #endif
205
206 if ((frame.tf_eflags & PSL_I) == 0) {
207 /*
208 * Buggy application or kernel code has disabled
209 * interrupts and then trapped. Enabling interrupts
210 * now is wrong, but it is better than running with
211 * interrupts disabled until they are accidentally
212 * enabled later.
213 */
214 if (ISPL(frame.tf_cs) == SEL_UPL || (frame.tf_eflags & PSL_VM))
215 printf(
216 "pid %ld (%s): trap %d with interrupts disabled\n",
217 (long)curproc->p_pid, curproc->p_comm, type);
218 else if (type != T_BPTFLT && type != T_TRCTRAP &&
219 frame.tf_eip != (int)cpu_switch_load_gs) {
220 /*
221 * XXX not quite right, since this may be for a
222 * multiple fault in user mode.
223 */
224 printf("kernel trap %d with interrupts disabled\n",
225 type);
226 /*
227 * Page faults need interrupts diasabled until later,
228 * and we shouldn't enable interrupts while in a
229 * critical section.
230 */
231 if (type != T_PAGEFLT && td->td_critnest == 0)
232 enable_intr();
233 }
234 }
235
236 eva = 0;
237 code = frame.tf_err;
238 if (type == T_PAGEFLT) {
239 /*
240 * For some Cyrix CPUs, %cr2 is clobbered by
241 * interrupts. This problem is worked around by using
242 * an interrupt gate for the pagefault handler. We
243 * are finally ready to read %cr2 and then must
244 * reenable interrupts.
245 *
246 * If we get a page fault while in a critical section, then
247 * it is most likely a fatal kernel page fault. The kernel
248 * is already going to panic trying to get a sleep lock to
249 * do the VM lookup, so just consider it a fatal trap so the
250 * kernel can print out a useful trap message and even get
251 * to the debugger.
252 */
253 eva = rcr2();
254 if (td->td_critnest == 0)
255 enable_intr();
256 else
257 trap_fatal(&frame, eva);
258 }
259
260 #ifdef DEVICE_POLLING
261 if (poll_in_trap)
262 ether_poll(poll_in_trap);
263 #endif /* DEVICE_POLLING */
264
265 if ((ISPL(frame.tf_cs) == SEL_UPL) ||
266 ((frame.tf_eflags & PSL_VM) &&
267 !(PCPU_GET(curpcb)->pcb_flags & PCB_VM86CALL))) {
268 /* user trap */
269
270 sticks = td->td_sticks;
271 td->td_frame = &frame;
272 if (td->td_ucred != p->p_ucred)
273 cred_update_thread(td);
274
275 switch (type) {
276 case T_PRIVINFLT: /* privileged instruction fault */
277 ucode = type;
278 i = SIGILL;
279 break;
280
281 case T_BPTFLT: /* bpt instruction fault */
282 case T_TRCTRAP: /* trace trap */
283 enable_intr();
284 frame.tf_eflags &= ~PSL_T;
285 i = SIGTRAP;
286 break;
287
288 case T_ARITHTRAP: /* arithmetic trap */
289 #ifdef DEV_NPX
290 ucode = npxtrap();
291 if (ucode == -1)
292 goto userout;
293 #else
294 ucode = code;
295 #endif
296 i = SIGFPE;
297 break;
298
299 /*
300 * The following two traps can happen in
301 * vm86 mode, and, if so, we want to handle
302 * them specially.
303 */
304 case T_PROTFLT: /* general protection fault */
305 case T_STKFLT: /* stack fault */
306 if (frame.tf_eflags & PSL_VM) {
307 i = vm86_emulate((struct vm86frame *)&frame);
308 if (i == 0)
309 goto user;
310 break;
311 }
312 /* FALLTHROUGH */
313
314 case T_SEGNPFLT: /* segment not present fault */
315 case T_TSSFLT: /* invalid TSS fault */
316 case T_DOUBLEFLT: /* double fault */
317 default:
318 ucode = code + BUS_SEGM_FAULT ;
319 i = SIGBUS;
320 break;
321
322 case T_PAGEFLT: /* page fault */
323 if (td->td_pflags & TDP_SA)
324 thread_user_enter(td);
325
326 i = trap_pfault(&frame, TRUE, eva);
327 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
328 if (i == -2) {
329 /*
330 * The f00f hack workaround has triggered, so
331 * treat the fault as an illegal instruction
332 * (T_PRIVINFLT) instead of a page fault.
333 */
334 type = frame.tf_trapno = T_PRIVINFLT;
335
336 /* Proceed as in that case. */
337 ucode = type;
338 i = SIGILL;
339 break;
340 }
341 #endif
342 if (i == -1)
343 goto userout;
344 if (i == 0)
345 goto user;
346
347 ucode = T_PAGEFLT;
348 break;
349
350 case T_DIVIDE: /* integer divide fault */
351 ucode = FPE_INTDIV;
352 i = SIGFPE;
353 break;
354
355 #ifdef DEV_ISA
356 case T_NMI:
357 #ifdef POWERFAIL_NMI
358 #ifndef TIMER_FREQ
359 # define TIMER_FREQ 1193182
360 #endif
361 mtx_lock(&Giant);
362 if (time_second - lastalert > 10) {
363 log(LOG_WARNING, "NMI: power fail\n");
364 sysbeep(TIMER_FREQ/880, hz);
365 lastalert = time_second;
366 }
367 mtx_unlock(&Giant);
368 goto userout;
369 #else /* !POWERFAIL_NMI */
370 /* machine/parity/power fail/"kitchen sink" faults */
371 /* XXX Giant */
372 if (isa_nmi(code) == 0) {
373 #ifdef KDB
374 /*
375 * NMI can be hooked up to a pushbutton
376 * for debugging.
377 */
378 if (kdb_on_nmi) {
379 printf ("NMI ... going to debugger\n");
380 kdb_trap(type, 0, &frame);
381 }
382 #endif /* KDB */
383 goto userout;
384 } else if (panic_on_nmi)
385 panic("NMI indicates hardware failure");
386 break;
387 #endif /* POWERFAIL_NMI */
388 #endif /* DEV_ISA */
389
390 case T_OFLOW: /* integer overflow fault */
391 ucode = FPE_INTOVF;
392 i = SIGFPE;
393 break;
394
395 case T_BOUND: /* bounds check fault */
396 ucode = FPE_FLTSUB;
397 i = SIGFPE;
398 break;
399
400 case T_DNA:
401 #ifdef DEV_NPX
402 /* transparent fault (due to context switch "late") */
403 if (npxdna())
404 goto userout;
405 #endif
406 i = SIGFPE;
407 ucode = FPE_FPU_NP_TRAP;
408 break;
409
410 case T_FPOPFLT: /* FPU operand fetch fault */
411 ucode = T_FPOPFLT;
412 i = SIGILL;
413 break;
414
415 case T_XMMFLT: /* SIMD floating-point exception */
416 ucode = 0; /* XXX */
417 i = SIGFPE;
418 break;
419 }
420 } else {
421 /* kernel trap */
422
423 KASSERT(cold || td->td_ucred != NULL,
424 ("kernel trap doesn't have ucred"));
425 switch (type) {
426 case T_PAGEFLT: /* page fault */
427 (void) trap_pfault(&frame, FALSE, eva);
428 goto out;
429
430 case T_DNA:
431 #ifdef DEV_NPX
432 /*
433 * The kernel is apparently using npx for copying.
434 * XXX this should be fatal unless the kernel has
435 * registered such use.
436 */
437 if (npxdna())
438 goto out;
439 #endif
440 break;
441
442 /*
443 * The following two traps can happen in
444 * vm86 mode, and, if so, we want to handle
445 * them specially.
446 */
447 case T_PROTFLT: /* general protection fault */
448 case T_STKFLT: /* stack fault */
449 if (frame.tf_eflags & PSL_VM) {
450 i = vm86_emulate((struct vm86frame *)&frame);
451 if (i != 0)
452 /*
453 * returns to original process
454 */
455 vm86_trap((struct vm86frame *)&frame);
456 goto out;
457 }
458 if (type == T_STKFLT)
459 break;
460
461 /* FALL THROUGH */
462
463 case T_SEGNPFLT: /* segment not present fault */
464 if (PCPU_GET(curpcb)->pcb_flags & PCB_VM86CALL)
465 break;
466
467 /*
468 * Invalid %fs's and %gs's can be created using
469 * procfs or PT_SETREGS or by invalidating the
470 * underlying LDT entry. This causes a fault
471 * in kernel mode when the kernel attempts to
472 * switch contexts. Lose the bad context
473 * (XXX) so that we can continue, and generate
474 * a signal.
475 */
476 if (frame.tf_eip == (int)cpu_switch_load_gs) {
477 PCPU_GET(curpcb)->pcb_gs = 0;
478 #if 0
479 PROC_LOCK(p);
480 psignal(p, SIGBUS);
481 PROC_UNLOCK(p);
482 #endif
483 goto out;
484 }
485
486 if (td->td_intr_nesting_level != 0)
487 break;
488
489 /*
490 * Invalid segment selectors and out of bounds
491 * %eip's and %esp's can be set up in user mode.
492 * This causes a fault in kernel mode when the
493 * kernel tries to return to user mode. We want
494 * to get this fault so that we can fix the
495 * problem here and not have to check all the
496 * selectors and pointers when the user changes
497 * them.
498 */
499 if (frame.tf_eip == (int)doreti_iret) {
500 frame.tf_eip = (int)doreti_iret_fault;
501 goto out;
502 }
503 if (frame.tf_eip == (int)doreti_popl_ds) {
504 frame.tf_eip = (int)doreti_popl_ds_fault;
505 goto out;
506 }
507 if (frame.tf_eip == (int)doreti_popl_es) {
508 frame.tf_eip = (int)doreti_popl_es_fault;
509 goto out;
510 }
511 if (frame.tf_eip == (int)doreti_popl_fs) {
512 frame.tf_eip = (int)doreti_popl_fs_fault;
513 goto out;
514 }
515 if (PCPU_GET(curpcb)->pcb_onfault != NULL) {
516 frame.tf_eip =
517 (int)PCPU_GET(curpcb)->pcb_onfault;
518 goto out;
519 }
520 break;
521
522 case T_TSSFLT:
523 /*
524 * PSL_NT can be set in user mode and isn't cleared
525 * automatically when the kernel is entered. This
526 * causes a TSS fault when the kernel attempts to
527 * `iret' because the TSS link is uninitialized. We
528 * want to get this fault so that we can fix the
529 * problem here and not every time the kernel is
530 * entered.
531 */
532 if (frame.tf_eflags & PSL_NT) {
533 frame.tf_eflags &= ~PSL_NT;
534 goto out;
535 }
536 break;
537
538 case T_TRCTRAP: /* trace trap */
539 if (frame.tf_eip == (int)IDTVEC(lcall_syscall)) {
540 /*
541 * We've just entered system mode via the
542 * syscall lcall. Continue single stepping
543 * silently until the syscall handler has
544 * saved the flags.
545 */
546 goto out;
547 }
548 if (frame.tf_eip == (int)IDTVEC(lcall_syscall) + 1) {
549 /*
550 * The syscall handler has now saved the
551 * flags. Stop single stepping it.
552 */
553 frame.tf_eflags &= ~PSL_T;
554 goto out;
555 }
556 /*
557 * Ignore debug register trace traps due to
558 * accesses in the user's address space, which
559 * can happen under several conditions such as
560 * if a user sets a watchpoint on a buffer and
561 * then passes that buffer to a system call.
562 * We still want to get TRCTRAPS for addresses
563 * in kernel space because that is useful when
564 * debugging the kernel.
565 */
566 /* XXX Giant */
567 if (user_dbreg_trap() &&
568 !(PCPU_GET(curpcb)->pcb_flags & PCB_VM86CALL)) {
569 /*
570 * Reset breakpoint bits because the
571 * processor doesn't
572 */
573 load_dr6(rdr6() & 0xfffffff0);
574 goto out;
575 }
576 /*
577 * FALLTHROUGH (TRCTRAP kernel mode, kernel address)
578 */
579 case T_BPTFLT:
580 /*
581 * If KDB is enabled, let it handle the debugger trap.
582 * Otherwise, debugger traps "can't happen".
583 */
584 #ifdef KDB
585 /* XXX Giant */
586 if (kdb_trap(type, 0, &frame))
587 goto out;
588 #endif
589 break;
590
591 #ifdef DEV_ISA
592 case T_NMI:
593 #ifdef POWERFAIL_NMI
594 mtx_lock(&Giant);
595 if (time_second - lastalert > 10) {
596 log(LOG_WARNING, "NMI: power fail\n");
597 sysbeep(TIMER_FREQ/880, hz);
598 lastalert = time_second;
599 }
600 mtx_unlock(&Giant);
601 goto out;
602 #else /* !POWERFAIL_NMI */
603 /* XXX Giant */
604 /* machine/parity/power fail/"kitchen sink" faults */
605 if (isa_nmi(code) == 0) {
606 #ifdef KDB
607 /*
608 * NMI can be hooked up to a pushbutton
609 * for debugging.
610 */
611 if (kdb_on_nmi) {
612 printf ("NMI ... going to debugger\n");
613 kdb_trap(type, 0, &frame);
614 }
615 #endif /* KDB */
616 goto out;
617 } else if (panic_on_nmi == 0)
618 goto out;
619 /* FALLTHROUGH */
620 #endif /* POWERFAIL_NMI */
621 #endif /* DEV_ISA */
622 }
623
624 trap_fatal(&frame, eva);
625 goto out;
626 }
627
628 /* Translate fault for emulators (e.g. Linux) */
629 if (*p->p_sysent->sv_transtrap)
630 i = (*p->p_sysent->sv_transtrap)(i, type);
631
632 trapsignal(td, i, ucode);
633
634 #ifdef DEBUG
635 if (type <= MAX_TRAP_MSG) {
636 uprintf("fatal process exception: %s",
637 trap_msg[type]);
638 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
639 uprintf(", fault VA = 0x%lx", (u_long)eva);
640 uprintf("\n");
641 }
642 #endif
643
644 user:
645 userret(td, &frame, sticks);
646 mtx_assert(&Giant, MA_NOTOWNED);
647 userout:
648 out:
649 return;
650 }
651
652 static int
653 trap_pfault(frame, usermode, eva)
654 struct trapframe *frame;
655 int usermode;
656 vm_offset_t eva;
657 {
658 vm_offset_t va;
659 struct vmspace *vm = NULL;
660 vm_map_t map = 0;
661 int rv = 0;
662 vm_prot_t ftype;
663 struct thread *td = curthread;
664 struct proc *p = td->td_proc;
665
666 va = trunc_page(eva);
667 if (va >= KERNBASE) {
668 /*
669 * Don't allow user-mode faults in kernel address space.
670 * An exception: if the faulting address is the invalid
671 * instruction entry in the IDT, then the Intel Pentium
672 * F00F bug workaround was triggered, and we need to
673 * treat it is as an illegal instruction, and not a page
674 * fault.
675 */
676 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
677 if ((eva == (unsigned int)&idt[6]) && has_f00f_bug)
678 return -2;
679 #endif
680 if (usermode)
681 goto nogo;
682
683 map = kernel_map;
684 } else {
685 /*
686 * This is a fault on non-kernel virtual memory.
687 * vm is initialized above to NULL. If curproc is NULL
688 * or curproc->p_vmspace is NULL the fault is fatal.
689 */
690 if (p != NULL)
691 vm = p->p_vmspace;
692
693 if (vm == NULL)
694 goto nogo;
695
696 map = &vm->vm_map;
697 }
698
699 if (frame->tf_err & PGEX_W)
700 ftype = VM_PROT_WRITE;
701 else
702 ftype = VM_PROT_READ;
703
704 if (map != kernel_map) {
705 /*
706 * Keep swapout from messing with us during this
707 * critical time.
708 */
709 PROC_LOCK(p);
710 ++p->p_lock;
711 PROC_UNLOCK(p);
712
713 /* Fault in the user page: */
714 rv = vm_fault(map, va, ftype,
715 (ftype & VM_PROT_WRITE) ? VM_FAULT_DIRTY
716 : VM_FAULT_NORMAL);
717
718 PROC_LOCK(p);
719 --p->p_lock;
720 PROC_UNLOCK(p);
721 } else {
722 /*
723 * Don't have to worry about process locking or stacks in the
724 * kernel.
725 */
726 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
727 }
728 if (rv == KERN_SUCCESS)
729 return (0);
730 nogo:
731 if (!usermode) {
732 if (td->td_intr_nesting_level == 0 &&
733 PCPU_GET(curpcb)->pcb_onfault != NULL) {
734 frame->tf_eip = (int)PCPU_GET(curpcb)->pcb_onfault;
735 return (0);
736 }
737 trap_fatal(frame, eva);
738 return (-1);
739 }
740
741 /* kludge to pass faulting virtual address to sendsig */
742 frame->tf_err = eva;
743
744 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
745 }
746
747 static void
748 trap_fatal(frame, eva)
749 struct trapframe *frame;
750 vm_offset_t eva;
751 {
752 int code, type, ss, esp;
753 struct soft_segment_descriptor softseg;
754 char *msg;
755
756 code = frame->tf_err;
757 type = frame->tf_trapno;
758 sdtossd(&gdt[IDXSEL(frame->tf_cs & 0xffff)].sd, &softseg);
759
760 if (type <= MAX_TRAP_MSG)
761 msg = trap_msg[type];
762 else
763 msg = "UNKNOWN";
764 printf("\n\nFatal trap %d: %s while in %s mode\n", type, msg,
765 frame->tf_eflags & PSL_VM ? "vm86" :
766 ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel");
767 #ifdef SMP
768 /* two separate prints in case of a trap on an unmapped page */
769 printf("cpuid = %d; ", PCPU_GET(cpuid));
770 printf("apic id = %02x\n", PCPU_GET(apic_id));
771 #endif
772 if (type == T_PAGEFLT) {
773 printf("fault virtual address = 0x%x\n", eva);
774 printf("fault code = %s %s, %s\n",
775 code & PGEX_U ? "user" : "supervisor",
776 code & PGEX_W ? "write" : "read",
777 code & PGEX_P ? "protection violation" : "page not present");
778 }
779 printf("instruction pointer = 0x%x:0x%x\n",
780 frame->tf_cs & 0xffff, frame->tf_eip);
781 if ((ISPL(frame->tf_cs) == SEL_UPL) || (frame->tf_eflags & PSL_VM)) {
782 ss = frame->tf_ss & 0xffff;
783 esp = frame->tf_esp;
784 } else {
785 ss = GSEL(GDATA_SEL, SEL_KPL);
786 esp = (int)&frame->tf_esp;
787 }
788 printf("stack pointer = 0x%x:0x%x\n", ss, esp);
789 printf("frame pointer = 0x%x:0x%x\n", ss, frame->tf_ebp);
790 printf("code segment = base 0x%x, limit 0x%x, type 0x%x\n",
791 softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type);
792 printf(" = DPL %d, pres %d, def32 %d, gran %d\n",
793 softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_def32,
794 softseg.ssd_gran);
795 printf("processor eflags = ");
796 if (frame->tf_eflags & PSL_T)
797 printf("trace trap, ");
798 if (frame->tf_eflags & PSL_I)
799 printf("interrupt enabled, ");
800 if (frame->tf_eflags & PSL_NT)
801 printf("nested task, ");
802 if (frame->tf_eflags & PSL_RF)
803 printf("resume, ");
804 if (frame->tf_eflags & PSL_VM)
805 printf("vm86, ");
806 printf("IOPL = %d\n", (frame->tf_eflags & PSL_IOPL) >> 12);
807 printf("current process = ");
808 if (curproc) {
809 printf("%lu (%s)\n",
810 (u_long)curproc->p_pid, curproc->p_comm ?
811 curproc->p_comm : "");
812 } else {
813 printf("Idle\n");
814 }
815
816 #ifdef KDB
817 if (kdb_trap(type, 0, frame))
818 return;
819 #endif
820 printf("trap number = %d\n", type);
821 if (type <= MAX_TRAP_MSG)
822 panic("%s", trap_msg[type]);
823 else
824 panic("unknown/reserved trap");
825 }
826
827 /*
828 * Double fault handler. Called when a fault occurs while writing
829 * a frame for a trap/exception onto the stack. This usually occurs
830 * when the stack overflows (such is the case with infinite recursion,
831 * for example).
832 *
833 * XXX Note that the current PTD gets replaced by IdlePTD when the
834 * task switch occurs. This means that the stack that was active at
835 * the time of the double fault is not available at <kstack> unless
836 * the machine was idle when the double fault occurred. The downside
837 * of this is that "trace <ebp>" in ddb won't work.
838 */
839 void
840 dblfault_handler()
841 {
842 printf("\nFatal double fault:\n");
843 printf("eip = 0x%x\n", PCPU_GET(common_tss.tss_eip));
844 printf("esp = 0x%x\n", PCPU_GET(common_tss.tss_esp));
845 printf("ebp = 0x%x\n", PCPU_GET(common_tss.tss_ebp));
846 #ifdef SMP
847 /* two separate prints in case of a trap on an unmapped page */
848 printf("cpuid = %d; ", PCPU_GET(cpuid));
849 printf("apic id = %02x\n", PCPU_GET(apic_id));
850 #endif
851 panic("double fault");
852 }
853
854 #ifdef I386_CPU
855 /*
856 * Compensate for 386 brain damage (missing URKR).
857 * This is a little simpler than the pagefault handler in trap() because
858 * it the page tables have already been faulted in and high addresses
859 * are thrown out early for other reasons.
860 */
861 int trapwrite(addr)
862 unsigned addr;
863 {
864 struct thread *td;
865 struct proc *p;
866 vm_offset_t va;
867 struct vmspace *vm;
868 int rv;
869
870 va = trunc_page((vm_offset_t)addr);
871 /*
872 * XXX - MAX is END. Changed > to >= for temp. fix.
873 */
874 if (va >= VM_MAXUSER_ADDRESS)
875 return (1);
876
877 td = curthread;
878 p = td->td_proc;
879 vm = p->p_vmspace;
880
881 PROC_LOCK(p);
882 ++p->p_lock;
883 PROC_UNLOCK(p);
884
885 /*
886 * fault the data page
887 */
888 rv = vm_fault(&vm->vm_map, va, VM_PROT_WRITE, VM_FAULT_DIRTY);
889
890 PROC_LOCK(p);
891 --p->p_lock;
892 PROC_UNLOCK(p);
893
894 if (rv != KERN_SUCCESS)
895 return 1;
896
897 return (0);
898 }
899 #endif
900
901 /*
902 * syscall - system call request C handler
903 *
904 * A system call is essentially treated as a trap.
905 */
906 void
907 syscall(frame)
908 struct trapframe frame;
909 {
910 caddr_t params;
911 struct sysent *callp;
912 struct thread *td = curthread;
913 struct proc *p = td->td_proc;
914 register_t orig_tf_eflags;
915 u_int sticks;
916 int error;
917 int narg;
918 int args[8];
919 u_int code;
920
921 /*
922 * note: PCPU_LAZY_INC() can only be used if we can afford
923 * occassional inaccuracy in the count.
924 */
925 PCPU_LAZY_INC(cnt.v_syscall);
926
927 #ifdef DIAGNOSTIC
928 if (ISPL(frame.tf_cs) != SEL_UPL) {
929 mtx_lock(&Giant); /* try to stabilize the system XXX */
930 panic("syscall");
931 /* NOT REACHED */
932 mtx_unlock(&Giant);
933 }
934 #endif
935
936 sticks = td->td_sticks;
937 td->td_frame = &frame;
938 if (td->td_ucred != p->p_ucred)
939 cred_update_thread(td);
940 if (p->p_flag & P_SA)
941 thread_user_enter(td);
942 params = (caddr_t)frame.tf_esp + sizeof(int);
943 code = frame.tf_eax;
944 orig_tf_eflags = frame.tf_eflags;
945
946 if (p->p_sysent->sv_prepsyscall) {
947 /*
948 * The prep code is MP aware.
949 */
950 (*p->p_sysent->sv_prepsyscall)(&frame, args, &code, ¶ms);
951 } else {
952 /*
953 * Need to check if this is a 32 bit or 64 bit syscall.
954 * fuword is MP aware.
955 */
956 if (code == SYS_syscall) {
957 /*
958 * Code is first argument, followed by actual args.
959 */
960 code = fuword(params);
961 params += sizeof(int);
962 } else if (code == SYS___syscall) {
963 /*
964 * Like syscall, but code is a quad, so as to maintain
965 * quad alignment for the rest of the arguments.
966 */
967 code = fuword(params);
968 params += sizeof(quad_t);
969 }
970 }
971
972 if (p->p_sysent->sv_mask)
973 code &= p->p_sysent->sv_mask;
974
975 if (code >= p->p_sysent->sv_size)
976 callp = &p->p_sysent->sv_table[0];
977 else
978 callp = &p->p_sysent->sv_table[code];
979
980 narg = callp->sy_narg & SYF_ARGMASK;
981
982 /*
983 * copyin and the ktrsyscall()/ktrsysret() code is MP-aware
984 */
985 if (params != NULL && narg != 0)
986 error = copyin(params, (caddr_t)args,
987 (u_int)(narg * sizeof(int)));
988 else
989 error = 0;
990
991 #ifdef KTRACE
992 if (KTRPOINT(td, KTR_SYSCALL))
993 ktrsyscall(code, narg, args);
994 #endif
995
996 CTR4(KTR_SYSC, "syscall enter thread %p pid %d proc %s code %d", td,
997 td->td_proc->p_pid, td->td_proc->p_comm, code);
998
999 /*
1000 * Try to run the syscall without Giant if the syscall
1001 * is MP safe.
1002 */
1003 if ((callp->sy_narg & SYF_MPSAFE) == 0)
1004 mtx_lock(&Giant);
1005
1006 if (error == 0) {
1007 td->td_retval[0] = 0;
1008 td->td_retval[1] = frame.tf_edx;
1009
1010 STOPEVENT(p, S_SCE, narg);
1011
1012 PTRACESTOP_SC(p, td, S_PT_SCE);
1013
1014 error = (*callp->sy_call)(td, args);
1015 }
1016
1017 switch (error) {
1018 case 0:
1019 frame.tf_eax = td->td_retval[0];
1020 frame.tf_edx = td->td_retval[1];
1021 frame.tf_eflags &= ~PSL_C;
1022 break;
1023
1024 case ERESTART:
1025 /*
1026 * Reconstruct pc, assuming lcall $X,y is 7 bytes,
1027 * int 0x80 is 2 bytes. We saved this in tf_err.
1028 */
1029 frame.tf_eip -= frame.tf_err;
1030 break;
1031
1032 case EJUSTRETURN:
1033 break;
1034
1035 default:
1036 if (p->p_sysent->sv_errsize) {
1037 if (error >= p->p_sysent->sv_errsize)
1038 error = -1; /* XXX */
1039 else
1040 error = p->p_sysent->sv_errtbl[error];
1041 }
1042 frame.tf_eax = error;
1043 frame.tf_eflags |= PSL_C;
1044 break;
1045 }
1046
1047 /*
1048 * Release Giant if we previously set it.
1049 */
1050 if ((callp->sy_narg & SYF_MPSAFE) == 0)
1051 mtx_unlock(&Giant);
1052
1053 /*
1054 * Traced syscall.
1055 */
1056 if ((orig_tf_eflags & PSL_T) && !(orig_tf_eflags & PSL_VM)) {
1057 frame.tf_eflags &= ~PSL_T;
1058 trapsignal(td, SIGTRAP, 0);
1059 }
1060
1061 /*
1062 * Handle reschedule and other end-of-syscall issues
1063 */
1064 userret(td, &frame, sticks);
1065
1066 CTR4(KTR_SYSC, "syscall exit thread %p pid %d proc %s code %d", td,
1067 td->td_proc->p_pid, td->td_proc->p_comm, code);
1068
1069 #ifdef KTRACE
1070 if (KTRPOINT(td, KTR_SYSRET))
1071 ktrsysret(code, error, td->td_retval[0]);
1072 #endif
1073
1074 /*
1075 * This works because errno is findable through the
1076 * register set. If we ever support an emulation where this
1077 * is not the case, this code will need to be revisited.
1078 */
1079 STOPEVENT(p, S_SCX, code);
1080
1081 PTRACESTOP_SC(p, td, S_PT_SCX);
1082
1083 WITNESS_WARN(WARN_PANIC, NULL, "System call %s returning",
1084 (code >= 0 && code < SYS_MAXSYSCALL) ? syscallnames[code] : "???");
1085 mtx_assert(&sched_lock, MA_NOTOWNED);
1086 mtx_assert(&Giant, MA_NOTOWNED);
1087 }
1088
Cache object: c10f1ee5c4bbb024b9f8121dfe15b5a1
|