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/6.1/sys/i386/i386/trap.c 158179 2006-04-30 16:44:43Z cvs2svn $");
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 30
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 "SIMD floating-point exception", /* 29 T_XMMFLT */
145 "reserved (unknown) fault", /* 30 T_RESERVED */
146 };
147
148 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
149 extern int has_f00f_bug;
150 #endif
151
152 #ifdef KDB
153 static int kdb_on_nmi = 1;
154 SYSCTL_INT(_machdep, OID_AUTO, kdb_on_nmi, CTLFLAG_RW,
155 &kdb_on_nmi, 0, "Go to KDB on NMI");
156 #endif
157 static int panic_on_nmi = 1;
158 SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RW,
159 &panic_on_nmi, 0, "Panic on NMI");
160
161 #ifdef WITNESS
162 extern char *syscallnames[];
163 #endif
164
165 /*
166 * Exception, fault, and trap interface to the FreeBSD kernel.
167 * This common code is called from assembly language IDT gate entry
168 * routines that prepare a suitable stack frame, and restore this
169 * frame after the exception has been processed.
170 */
171
172 void
173 trap(frame)
174 struct trapframe frame;
175 {
176 struct thread *td = curthread;
177 struct proc *p = td->td_proc;
178 u_int sticks = 0;
179 int i = 0, ucode = 0, type, code;
180 vm_offset_t eva;
181 #ifdef POWERFAIL_NMI
182 static int lastalert = 0;
183 #endif
184
185 PCPU_LAZY_INC(cnt.v_trap);
186 type = frame.tf_trapno;
187
188 #ifdef KDB_STOP_NMI
189 /* Handler for NMI IPIs used for debugging */
190 if (type == T_NMI) {
191 if (ipi_nmi_handler() == 0)
192 goto out;
193 }
194 #endif /* KDB_STOP_NMI */
195
196 #ifdef KDB
197 if (kdb_active) {
198 kdb_reenter();
199 goto out;
200 }
201 #endif
202
203 #ifdef HWPMC_HOOKS
204 /*
205 * CPU PMCs interrupt using an NMI so we check for that first.
206 * If the HWPMC module is active, 'pmc_hook' will point to
207 * the function to be called. A return value of '1' from the
208 * hook means that the NMI was handled by it and that we can
209 * return immediately.
210 */
211 if (type == T_NMI && pmc_intr &&
212 (*pmc_intr)(PCPU_GET(cpuid), (uintptr_t) frame.tf_eip,
213 TRAPF_USERMODE(&frame)))
214 goto out;
215 #endif
216
217 if ((frame.tf_eflags & PSL_I) == 0) {
218 /*
219 * Buggy application or kernel code has disabled
220 * interrupts and then trapped. Enabling interrupts
221 * now is wrong, but it is better than running with
222 * interrupts disabled until they are accidentally
223 * enabled later.
224 */
225 if (ISPL(frame.tf_cs) == SEL_UPL || (frame.tf_eflags & PSL_VM))
226 printf(
227 "pid %ld (%s): trap %d with interrupts disabled\n",
228 (long)curproc->p_pid, curproc->p_comm, type);
229 else if (type != T_BPTFLT && type != T_TRCTRAP &&
230 frame.tf_eip != (int)cpu_switch_load_gs) {
231 /*
232 * XXX not quite right, since this may be for a
233 * multiple fault in user mode.
234 */
235 printf("kernel trap %d with interrupts disabled\n",
236 type);
237 /*
238 * Page faults need interrupts disabled until later,
239 * and we shouldn't enable interrupts while in a
240 * critical section or if servicing an NMI.
241 */
242 if (type != T_NMI && type != T_PAGEFLT &&
243 td->td_critnest == 0)
244 enable_intr();
245 }
246 }
247
248 eva = 0;
249 code = frame.tf_err;
250 if (type == T_PAGEFLT) {
251 /*
252 * For some Cyrix CPUs, %cr2 is clobbered by
253 * interrupts. This problem is worked around by using
254 * an interrupt gate for the pagefault handler. We
255 * are finally ready to read %cr2 and then must
256 * reenable interrupts.
257 *
258 * If we get a page fault while in a critical section, then
259 * it is most likely a fatal kernel page fault. The kernel
260 * is already going to panic trying to get a sleep lock to
261 * do the VM lookup, so just consider it a fatal trap so the
262 * kernel can print out a useful trap message and even get
263 * to the debugger.
264 */
265 eva = rcr2();
266 if (td->td_critnest == 0)
267 enable_intr();
268 else
269 trap_fatal(&frame, eva);
270 }
271
272 if ((ISPL(frame.tf_cs) == SEL_UPL) ||
273 ((frame.tf_eflags & PSL_VM) &&
274 !(PCPU_GET(curpcb)->pcb_flags & PCB_VM86CALL))) {
275 /* user trap */
276
277 sticks = td->td_sticks;
278 td->td_frame = &frame;
279 if (td->td_ucred != p->p_ucred)
280 cred_update_thread(td);
281
282 switch (type) {
283 case T_PRIVINFLT: /* privileged instruction fault */
284 ucode = type;
285 i = SIGILL;
286 break;
287
288 case T_BPTFLT: /* bpt instruction fault */
289 case T_TRCTRAP: /* trace trap */
290 enable_intr();
291 frame.tf_eflags &= ~PSL_T;
292 i = SIGTRAP;
293 break;
294
295 case T_ARITHTRAP: /* arithmetic trap */
296 #ifdef DEV_NPX
297 ucode = npxtrap();
298 if (ucode == -1)
299 goto userout;
300 #else
301 ucode = code;
302 #endif
303 i = SIGFPE;
304 break;
305
306 /*
307 * The following two traps can happen in
308 * vm86 mode, and, if so, we want to handle
309 * them specially.
310 */
311 case T_PROTFLT: /* general protection fault */
312 case T_STKFLT: /* stack fault */
313 if (frame.tf_eflags & PSL_VM) {
314 i = vm86_emulate((struct vm86frame *)&frame);
315 if (i == 0)
316 goto user;
317 break;
318 }
319 /* FALLTHROUGH */
320
321 case T_SEGNPFLT: /* segment not present fault */
322 case T_TSSFLT: /* invalid TSS fault */
323 case T_DOUBLEFLT: /* double fault */
324 default:
325 ucode = code + BUS_SEGM_FAULT ;
326 i = SIGBUS;
327 break;
328
329 case T_PAGEFLT: /* page fault */
330 if (td->td_pflags & TDP_SA)
331 thread_user_enter(td);
332
333 i = trap_pfault(&frame, TRUE, eva);
334 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
335 if (i == -2) {
336 /*
337 * The f00f hack workaround has triggered, so
338 * treat the fault as an illegal instruction
339 * (T_PRIVINFLT) instead of a page fault.
340 */
341 type = frame.tf_trapno = T_PRIVINFLT;
342
343 /* Proceed as in that case. */
344 ucode = type;
345 i = SIGILL;
346 break;
347 }
348 #endif
349 if (i == -1)
350 goto userout;
351 if (i == 0)
352 goto user;
353
354 ucode = T_PAGEFLT;
355 break;
356
357 case T_DIVIDE: /* integer divide fault */
358 ucode = FPE_INTDIV;
359 i = SIGFPE;
360 break;
361
362 #ifdef DEV_ISA
363 case T_NMI:
364 #ifdef POWERFAIL_NMI
365 #ifndef TIMER_FREQ
366 # define TIMER_FREQ 1193182
367 #endif
368 mtx_lock(&Giant);
369 if (time_second - lastalert > 10) {
370 log(LOG_WARNING, "NMI: power fail\n");
371 sysbeep(TIMER_FREQ/880, hz);
372 lastalert = time_second;
373 }
374 mtx_unlock(&Giant);
375 goto userout;
376 #else /* !POWERFAIL_NMI */
377 /* machine/parity/power fail/"kitchen sink" faults */
378 /* XXX Giant */
379 if (isa_nmi(code) == 0) {
380 #ifdef KDB
381 /*
382 * NMI can be hooked up to a pushbutton
383 * for debugging.
384 */
385 if (kdb_on_nmi) {
386 printf ("NMI ... going to debugger\n");
387 kdb_trap(type, 0, &frame);
388 }
389 #endif /* KDB */
390 goto userout;
391 } else if (panic_on_nmi)
392 panic("NMI indicates hardware failure");
393 break;
394 #endif /* POWERFAIL_NMI */
395 #endif /* DEV_ISA */
396
397 case T_OFLOW: /* integer overflow fault */
398 ucode = FPE_INTOVF;
399 i = SIGFPE;
400 break;
401
402 case T_BOUND: /* bounds check fault */
403 ucode = FPE_FLTSUB;
404 i = SIGFPE;
405 break;
406
407 case T_DNA:
408 #ifdef DEV_NPX
409 /* transparent fault (due to context switch "late") */
410 if (npxdna())
411 goto userout;
412 #endif
413 i = SIGFPE;
414 ucode = FPE_FPU_NP_TRAP;
415 break;
416
417 case T_FPOPFLT: /* FPU operand fetch fault */
418 ucode = T_FPOPFLT;
419 i = SIGILL;
420 break;
421
422 case T_XMMFLT: /* SIMD floating-point exception */
423 ucode = 0; /* XXX */
424 i = SIGFPE;
425 break;
426 }
427 } else {
428 /* kernel trap */
429
430 KASSERT(cold || td->td_ucred != NULL,
431 ("kernel trap doesn't have ucred"));
432 switch (type) {
433 case T_PAGEFLT: /* page fault */
434 (void) trap_pfault(&frame, FALSE, eva);
435 goto out;
436
437 case T_DNA:
438 #ifdef DEV_NPX
439 /*
440 * The kernel is apparently using npx for copying.
441 * XXX this should be fatal unless the kernel has
442 * registered such use.
443 */
444 if (npxdna())
445 goto out;
446 #endif
447 break;
448
449 /*
450 * The following two traps can happen in
451 * vm86 mode, and, if so, we want to handle
452 * them specially.
453 */
454 case T_PROTFLT: /* general protection fault */
455 case T_STKFLT: /* stack fault */
456 if (frame.tf_eflags & PSL_VM) {
457 i = vm86_emulate((struct vm86frame *)&frame);
458 if (i != 0)
459 /*
460 * returns to original process
461 */
462 vm86_trap((struct vm86frame *)&frame);
463 goto out;
464 }
465 if (type == T_STKFLT)
466 break;
467
468 /* FALL THROUGH */
469
470 case T_SEGNPFLT: /* segment not present fault */
471 if (PCPU_GET(curpcb)->pcb_flags & PCB_VM86CALL)
472 break;
473
474 /*
475 * Invalid %fs's and %gs's can be created using
476 * procfs or PT_SETREGS or by invalidating the
477 * underlying LDT entry. This causes a fault
478 * in kernel mode when the kernel attempts to
479 * switch contexts. Lose the bad context
480 * (XXX) so that we can continue, and generate
481 * a signal.
482 */
483 if (frame.tf_eip == (int)cpu_switch_load_gs) {
484 PCPU_GET(curpcb)->pcb_gs = 0;
485 #if 0
486 PROC_LOCK(p);
487 psignal(p, SIGBUS);
488 PROC_UNLOCK(p);
489 #endif
490 goto out;
491 }
492
493 if (td->td_intr_nesting_level != 0)
494 break;
495
496 /*
497 * Invalid segment selectors and out of bounds
498 * %eip's and %esp's can be set up in user mode.
499 * This causes a fault in kernel mode when the
500 * kernel tries to return to user mode. We want
501 * to get this fault so that we can fix the
502 * problem here and not have to check all the
503 * selectors and pointers when the user changes
504 * them.
505 */
506 if (frame.tf_eip == (int)doreti_iret) {
507 frame.tf_eip = (int)doreti_iret_fault;
508 goto out;
509 }
510 if (frame.tf_eip == (int)doreti_popl_ds) {
511 frame.tf_eip = (int)doreti_popl_ds_fault;
512 goto out;
513 }
514 if (frame.tf_eip == (int)doreti_popl_es) {
515 frame.tf_eip = (int)doreti_popl_es_fault;
516 goto out;
517 }
518 if (frame.tf_eip == (int)doreti_popl_fs) {
519 frame.tf_eip = (int)doreti_popl_fs_fault;
520 goto out;
521 }
522 if (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 KDB is enabled, let it handle the debugger trap.
589 * Otherwise, debugger traps "can't happen".
590 */
591 #ifdef KDB
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 KDB
614 /*
615 * NMI can be hooked up to a pushbutton
616 * for debugging.
617 */
618 if (kdb_on_nmi) {
619 printf ("NMI ... going to debugger\n");
620 kdb_trap(type, 0, &frame);
621 }
622 #endif /* KDB */
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 out:
656 return;
657 }
658
659 static int
660 trap_pfault(frame, usermode, eva)
661 struct trapframe *frame;
662 int usermode;
663 vm_offset_t eva;
664 {
665 vm_offset_t va;
666 struct vmspace *vm = NULL;
667 vm_map_t map = 0;
668 int rv = 0;
669 vm_prot_t ftype;
670 struct thread *td = curthread;
671 struct proc *p = td->td_proc;
672
673 va = trunc_page(eva);
674 if (va >= KERNBASE) {
675 /*
676 * Don't allow user-mode faults in kernel address space.
677 * An exception: if the faulting address is the invalid
678 * instruction entry in the IDT, then the Intel Pentium
679 * F00F bug workaround was triggered, and we need to
680 * treat it is as an illegal instruction, and not a page
681 * fault.
682 */
683 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
684 if ((eva == (unsigned int)&idt[6]) && has_f00f_bug)
685 return -2;
686 #endif
687 if (usermode)
688 goto nogo;
689
690 map = kernel_map;
691 } else {
692 /*
693 * This is a fault on non-kernel virtual memory.
694 * vm is initialized above to NULL. If curproc is NULL
695 * or curproc->p_vmspace is NULL the fault is fatal.
696 */
697 if (p != NULL)
698 vm = p->p_vmspace;
699
700 if (vm == NULL)
701 goto nogo;
702
703 map = &vm->vm_map;
704 }
705
706 if (frame->tf_err & PGEX_W)
707 ftype = VM_PROT_WRITE;
708 else
709 ftype = VM_PROT_READ;
710
711 if (map != kernel_map) {
712 /*
713 * Keep swapout from messing with us during this
714 * critical time.
715 */
716 PROC_LOCK(p);
717 ++p->p_lock;
718 PROC_UNLOCK(p);
719
720 /* Fault in the user page: */
721 rv = vm_fault(map, va, ftype,
722 (ftype & VM_PROT_WRITE) ? VM_FAULT_DIRTY
723 : VM_FAULT_NORMAL);
724
725 PROC_LOCK(p);
726 --p->p_lock;
727 PROC_UNLOCK(p);
728 } else {
729 /*
730 * Don't have to worry about process locking or stacks in the
731 * kernel.
732 */
733 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
734 }
735 if (rv == KERN_SUCCESS)
736 return (0);
737 nogo:
738 if (!usermode) {
739 if (td->td_intr_nesting_level == 0 &&
740 PCPU_GET(curpcb)->pcb_onfault != NULL) {
741 frame->tf_eip = (int)PCPU_GET(curpcb)->pcb_onfault;
742 return (0);
743 }
744 trap_fatal(frame, eva);
745 return (-1);
746 }
747
748 /* kludge to pass faulting virtual address to sendsig */
749 frame->tf_err = eva;
750
751 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
752 }
753
754 static void
755 trap_fatal(frame, eva)
756 struct trapframe *frame;
757 vm_offset_t eva;
758 {
759 int code, type, ss, esp;
760 struct soft_segment_descriptor softseg;
761 char *msg;
762
763 code = frame->tf_err;
764 type = frame->tf_trapno;
765 sdtossd(&gdt[IDXSEL(frame->tf_cs & 0xffff)].sd, &softseg);
766
767 if (type <= MAX_TRAP_MSG)
768 msg = trap_msg[type];
769 else
770 msg = "UNKNOWN";
771 printf("\n\nFatal trap %d: %s while in %s mode\n", type, msg,
772 frame->tf_eflags & PSL_VM ? "vm86" :
773 ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel");
774 #ifdef SMP
775 /* two separate prints in case of a trap on an unmapped page */
776 printf("cpuid = %d; ", PCPU_GET(cpuid));
777 printf("apic id = %02x\n", PCPU_GET(apic_id));
778 #endif
779 if (type == T_PAGEFLT) {
780 printf("fault virtual address = 0x%x\n", eva);
781 printf("fault code = %s %s, %s\n",
782 code & PGEX_U ? "user" : "supervisor",
783 code & PGEX_W ? "write" : "read",
784 code & PGEX_P ? "protection violation" : "page not present");
785 }
786 printf("instruction pointer = 0x%x:0x%x\n",
787 frame->tf_cs & 0xffff, frame->tf_eip);
788 if ((ISPL(frame->tf_cs) == SEL_UPL) || (frame->tf_eflags & PSL_VM)) {
789 ss = frame->tf_ss & 0xffff;
790 esp = frame->tf_esp;
791 } else {
792 ss = GSEL(GDATA_SEL, SEL_KPL);
793 esp = (int)&frame->tf_esp;
794 }
795 printf("stack pointer = 0x%x:0x%x\n", ss, esp);
796 printf("frame pointer = 0x%x:0x%x\n", ss, frame->tf_ebp);
797 printf("code segment = base 0x%x, limit 0x%x, type 0x%x\n",
798 softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type);
799 printf(" = DPL %d, pres %d, def32 %d, gran %d\n",
800 softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_def32,
801 softseg.ssd_gran);
802 printf("processor eflags = ");
803 if (frame->tf_eflags & PSL_T)
804 printf("trace trap, ");
805 if (frame->tf_eflags & PSL_I)
806 printf("interrupt enabled, ");
807 if (frame->tf_eflags & PSL_NT)
808 printf("nested task, ");
809 if (frame->tf_eflags & PSL_RF)
810 printf("resume, ");
811 if (frame->tf_eflags & PSL_VM)
812 printf("vm86, ");
813 printf("IOPL = %d\n", (frame->tf_eflags & PSL_IOPL) >> 12);
814 printf("current process = ");
815 if (curproc) {
816 printf("%lu (%s)\n",
817 (u_long)curproc->p_pid, curproc->p_comm ?
818 curproc->p_comm : "");
819 } else {
820 printf("Idle\n");
821 }
822
823 #ifdef KDB
824 if (debugger_on_panic || kdb_active) {
825 register_t eflags;
826 eflags = intr_disable();
827 if (kdb_trap(type, 0, frame)) {
828 intr_restore(eflags);
829 return;
830 }
831 intr_restore(eflags);
832 }
833 #endif
834 printf("trap number = %d\n", type);
835 if (type <= MAX_TRAP_MSG)
836 panic("%s", trap_msg[type]);
837 else
838 panic("unknown/reserved trap");
839 }
840
841 /*
842 * Double fault handler. Called when a fault occurs while writing
843 * a frame for a trap/exception onto the stack. This usually occurs
844 * when the stack overflows (such is the case with infinite recursion,
845 * for example).
846 *
847 * XXX Note that the current PTD gets replaced by IdlePTD when the
848 * task switch occurs. This means that the stack that was active at
849 * the time of the double fault is not available at <kstack> unless
850 * the machine was idle when the double fault occurred. The downside
851 * of this is that "trace <ebp>" in ddb won't work.
852 */
853 void
854 dblfault_handler()
855 {
856 printf("\nFatal double fault:\n");
857 printf("eip = 0x%x\n", PCPU_GET(common_tss.tss_eip));
858 printf("esp = 0x%x\n", PCPU_GET(common_tss.tss_esp));
859 printf("ebp = 0x%x\n", PCPU_GET(common_tss.tss_ebp));
860 #ifdef SMP
861 /* two separate prints in case of a trap on an unmapped page */
862 printf("cpuid = %d; ", PCPU_GET(cpuid));
863 printf("apic id = %02x\n", PCPU_GET(apic_id));
864 #endif
865 panic("double fault");
866 }
867
868 /*
869 * syscall - system call request C handler
870 *
871 * A system call is essentially treated as a trap.
872 */
873 void
874 syscall(frame)
875 struct trapframe frame;
876 {
877 caddr_t params;
878 struct sysent *callp;
879 struct thread *td = curthread;
880 struct proc *p = td->td_proc;
881 register_t orig_tf_eflags;
882 u_int sticks;
883 int error;
884 int narg;
885 int args[8];
886 u_int code;
887
888 /*
889 * note: PCPU_LAZY_INC() can only be used if we can afford
890 * occassional inaccuracy in the count.
891 */
892 PCPU_LAZY_INC(cnt.v_syscall);
893
894 #ifdef DIAGNOSTIC
895 if (ISPL(frame.tf_cs) != SEL_UPL) {
896 mtx_lock(&Giant); /* try to stabilize the system XXX */
897 panic("syscall");
898 /* NOT REACHED */
899 mtx_unlock(&Giant);
900 }
901 #endif
902
903 sticks = td->td_sticks;
904 td->td_frame = &frame;
905 if (td->td_ucred != p->p_ucred)
906 cred_update_thread(td);
907 if (p->p_flag & P_SA)
908 thread_user_enter(td);
909 params = (caddr_t)frame.tf_esp + sizeof(int);
910 code = frame.tf_eax;
911 orig_tf_eflags = frame.tf_eflags;
912
913 if (p->p_sysent->sv_prepsyscall) {
914 /*
915 * The prep code is MP aware.
916 */
917 (*p->p_sysent->sv_prepsyscall)(&frame, args, &code, ¶ms);
918 } else {
919 /*
920 * Need to check if this is a 32 bit or 64 bit syscall.
921 * fuword is MP aware.
922 */
923 if (code == SYS_syscall) {
924 /*
925 * Code is first argument, followed by actual args.
926 */
927 code = fuword(params);
928 params += sizeof(int);
929 } else if (code == SYS___syscall) {
930 /*
931 * Like syscall, but code is a quad, so as to maintain
932 * quad alignment for the rest of the arguments.
933 */
934 code = fuword(params);
935 params += sizeof(quad_t);
936 }
937 }
938
939 if (p->p_sysent->sv_mask)
940 code &= p->p_sysent->sv_mask;
941
942 if (code >= p->p_sysent->sv_size)
943 callp = &p->p_sysent->sv_table[0];
944 else
945 callp = &p->p_sysent->sv_table[code];
946
947 narg = callp->sy_narg & SYF_ARGMASK;
948
949 /*
950 * copyin and the ktrsyscall()/ktrsysret() code is MP-aware
951 */
952 if (params != NULL && narg != 0)
953 error = copyin(params, (caddr_t)args,
954 (u_int)(narg * sizeof(int)));
955 else
956 error = 0;
957
958 #ifdef KTRACE
959 if (KTRPOINT(td, KTR_SYSCALL))
960 ktrsyscall(code, narg, args);
961 #endif
962
963 CTR4(KTR_SYSC, "syscall enter thread %p pid %d proc %s code %d", td,
964 td->td_proc->p_pid, td->td_proc->p_comm, code);
965
966 /*
967 * Try to run the syscall without Giant if the syscall
968 * is MP safe.
969 */
970 if ((callp->sy_narg & SYF_MPSAFE) == 0)
971 mtx_lock(&Giant);
972
973 if (error == 0) {
974 td->td_retval[0] = 0;
975 td->td_retval[1] = frame.tf_edx;
976
977 STOPEVENT(p, S_SCE, narg);
978
979 PTRACESTOP_SC(p, td, S_PT_SCE);
980
981 error = (*callp->sy_call)(td, args);
982 }
983
984 switch (error) {
985 case 0:
986 frame.tf_eax = td->td_retval[0];
987 frame.tf_edx = td->td_retval[1];
988 frame.tf_eflags &= ~PSL_C;
989 break;
990
991 case ERESTART:
992 /*
993 * Reconstruct pc, assuming lcall $X,y is 7 bytes,
994 * int 0x80 is 2 bytes. We saved this in tf_err.
995 */
996 frame.tf_eip -= frame.tf_err;
997 break;
998
999 case EJUSTRETURN:
1000 break;
1001
1002 default:
1003 if (p->p_sysent->sv_errsize) {
1004 if (error >= p->p_sysent->sv_errsize)
1005 error = -1; /* XXX */
1006 else
1007 error = p->p_sysent->sv_errtbl[error];
1008 }
1009 frame.tf_eax = error;
1010 frame.tf_eflags |= PSL_C;
1011 break;
1012 }
1013
1014 /*
1015 * Release Giant if we previously set it.
1016 */
1017 if ((callp->sy_narg & SYF_MPSAFE) == 0)
1018 mtx_unlock(&Giant);
1019
1020 /*
1021 * Traced syscall.
1022 */
1023 if ((orig_tf_eflags & PSL_T) && !(orig_tf_eflags & PSL_VM)) {
1024 frame.tf_eflags &= ~PSL_T;
1025 trapsignal(td, SIGTRAP, 0);
1026 }
1027
1028 /*
1029 * Handle reschedule and other end-of-syscall issues
1030 */
1031 userret(td, &frame, sticks);
1032
1033 CTR4(KTR_SYSC, "syscall exit thread %p pid %d proc %s code %d", td,
1034 td->td_proc->p_pid, td->td_proc->p_comm, code);
1035
1036 #ifdef KTRACE
1037 if (KTRPOINT(td, KTR_SYSRET))
1038 ktrsysret(code, error, td->td_retval[0]);
1039 #endif
1040
1041 /*
1042 * This works because errno is findable through the
1043 * register set. If we ever support an emulation where this
1044 * is not the case, this code will need to be revisited.
1045 */
1046 STOPEVENT(p, S_SCX, code);
1047
1048 PTRACESTOP_SC(p, td, S_PT_SCX);
1049
1050 WITNESS_WARN(WARN_PANIC, NULL, "System call %s returning",
1051 (code >= 0 && code < SYS_MAXSYSCALL) ? syscallnames[code] : "???");
1052 mtx_assert(&sched_lock, MA_NOTOWNED);
1053 mtx_assert(&Giant, MA_NOTOWNED);
1054 }
1055
Cache object: 1bef77e292e4c0963f1834a7a2120b29
|