xref: /freebsd/sys/kern/subr_trap.c (revision 77a0943ded95b9e6438f7db70c4a28e4d93946d4)
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  * $FreeBSD$
39  */
40 
41 /*
42  * 386 Trap and System call handling
43  */
44 
45 #include "opt_cpu.h"
46 #include "opt_ddb.h"
47 #include "opt_ktrace.h"
48 #include "opt_clock.h"
49 #include "opt_trap.h"
50 
51 #include <sys/param.h>
52 #include <sys/bus.h>
53 #include <sys/systm.h>
54 #include <sys/proc.h>
55 #include <sys/pioctl.h>
56 #include <sys/ipl.h>
57 #include <sys/kernel.h>
58 #include <sys/ktr.h>
59 #include <sys/mutex.h>
60 #include <sys/resourcevar.h>
61 #include <sys/signalvar.h>
62 #include <sys/syscall.h>
63 #include <sys/sysctl.h>
64 #include <sys/sysent.h>
65 #include <sys/uio.h>
66 #include <sys/vmmeter.h>
67 #ifdef KTRACE
68 #include <sys/ktrace.h>
69 #endif
70 
71 #include <vm/vm.h>
72 #include <vm/vm_param.h>
73 #include <sys/lock.h>
74 #include <vm/pmap.h>
75 #include <vm/vm_kern.h>
76 #include <vm/vm_map.h>
77 #include <vm/vm_page.h>
78 #include <vm/vm_extern.h>
79 
80 #include <machine/cpu.h>
81 #include <machine/md_var.h>
82 #include <machine/pcb.h>
83 #ifdef SMP
84 #include <machine/smp.h>
85 #endif
86 #include <machine/tss.h>
87 
88 #include <i386/isa/icu.h>
89 #include <i386/isa/intr_machdep.h>
90 
91 #ifdef POWERFAIL_NMI
92 #include <sys/syslog.h>
93 #include <machine/clock.h>
94 #endif
95 
96 #include <machine/vm86.h>
97 
98 #include <ddb/ddb.h>
99 
100 #include "isa.h"
101 #include "npx.h"
102 
103 #include <sys/sysctl.h>
104 
105 int (*pmath_emulate) __P((struct trapframe *));
106 
107 extern void trap __P((struct trapframe frame));
108 extern int trapwrite __P((unsigned addr));
109 extern void syscall2 __P((struct trapframe frame));
110 extern void ast __P((struct trapframe frame));
111 
112 static int trap_pfault __P((struct trapframe *, int, vm_offset_t));
113 static void trap_fatal __P((struct trapframe *, vm_offset_t));
114 void dblfault_handler __P((void));
115 
116 extern inthand_t IDTVEC(syscall);
117 
118 #define MAX_TRAP_MSG		28
119 static char *trap_msg[] = {
120 	"",					/*  0 unused */
121 	"privileged instruction fault",		/*  1 T_PRIVINFLT */
122 	"",					/*  2 unused */
123 	"breakpoint instruction fault",		/*  3 T_BPTFLT */
124 	"",					/*  4 unused */
125 	"",					/*  5 unused */
126 	"arithmetic trap",			/*  6 T_ARITHTRAP */
127 	"system forced exception",		/*  7 T_ASTFLT */
128 	"",					/*  8 unused */
129 	"general protection fault",		/*  9 T_PROTFLT */
130 	"trace trap",				/* 10 T_TRCTRAP */
131 	"",					/* 11 unused */
132 	"page fault",				/* 12 T_PAGEFLT */
133 	"",					/* 13 unused */
134 	"alignment fault",			/* 14 T_ALIGNFLT */
135 	"",					/* 15 unused */
136 	"",					/* 16 unused */
137 	"",					/* 17 unused */
138 	"integer divide fault",			/* 18 T_DIVIDE */
139 	"non-maskable interrupt trap",		/* 19 T_NMI */
140 	"overflow trap",			/* 20 T_OFLOW */
141 	"FPU bounds check fault",		/* 21 T_BOUND */
142 	"FPU device not available",		/* 22 T_DNA */
143 	"double fault",				/* 23 T_DOUBLEFLT */
144 	"FPU operand fetch fault",		/* 24 T_FPOPFLT */
145 	"invalid TSS fault",			/* 25 T_TSSFLT */
146 	"segment not present fault",		/* 26 T_SEGNPFLT */
147 	"stack fault",				/* 27 T_STKFLT */
148 	"machine check trap",			/* 28 T_MCHK */
149 };
150 
151 static __inline int userret __P((struct proc *p, struct trapframe *frame,
152 				  u_quad_t oticks, int have_giant));
153 
154 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
155 extern int has_f00f_bug;
156 #endif
157 
158 #ifdef DDB
159 static int ddb_on_nmi = 1;
160 SYSCTL_INT(_machdep, OID_AUTO, ddb_on_nmi, CTLFLAG_RW,
161 	&ddb_on_nmi, 0, "Go to DDB on NMI");
162 #endif
163 static int panic_on_nmi = 1;
164 SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RW,
165 	&panic_on_nmi, 0, "Panic on NMI");
166 
167 static __inline int
168 userret(p, frame, oticks, have_giant)
169 	struct proc *p;
170 	struct trapframe *frame;
171 	u_quad_t oticks;
172 	int have_giant;
173 {
174 	int sig, s;
175 
176 	while ((sig = CURSIG(p)) != 0) {
177 		if (have_giant == 0) {
178 			mtx_enter(&Giant, MTX_DEF);
179 			have_giant = 1;
180 		}
181 		postsig(sig);
182 	}
183 
184 	p->p_priority = p->p_usrpri;
185 	if (resched_wanted()) {
186 		/*
187 		 * Since we are curproc, clock will normally just change
188 		 * our priority without moving us from one queue to another
189 		 * (since the running process is not on a queue.)
190 		 * If that happened after we setrunqueue ourselves but before we
191 		 * mi_switch()'ed, we might not be on the queue indicated by
192 		 * our priority.
193 		 */
194 		s = splhigh();
195 		mtx_enter(&sched_lock, MTX_SPIN);
196 		DROP_GIANT_NOSWITCH();
197 		setrunqueue(p);
198 		p->p_stats->p_ru.ru_nivcsw++;
199 		mi_switch();
200 		mtx_exit(&sched_lock, MTX_SPIN);
201 		PICKUP_GIANT();
202 		splx(s);
203 		while ((sig = CURSIG(p)) != 0) {
204 			if (have_giant == 0) {
205 				mtx_enter(&Giant, MTX_DEF);
206 				have_giant = 1;
207 			}
208 			postsig(sig);
209 		}
210 	}
211 	/*
212 	 * Charge system time if profiling.
213 	 */
214 	if (p->p_flag & P_PROFIL) {
215 		if (have_giant == 0) {
216 			mtx_enter(&Giant, MTX_DEF);
217 			have_giant = 1;
218 		}
219 		addupc_task(p, frame->tf_eip,
220 			    (u_int)(p->p_sticks - oticks) * psratio);
221 	}
222 	curpriority = p->p_priority;
223 	return(have_giant);
224 }
225 
226 /*
227  * Exception, fault, and trap interface to the FreeBSD kernel.
228  * This common code is called from assembly language IDT gate entry
229  * routines that prepare a suitable stack frame, and restore this
230  * frame after the exception has been processed.
231  */
232 
233 void
234 trap(frame)
235 	struct trapframe frame;
236 {
237 	struct proc *p = curproc;
238 	u_quad_t sticks = 0;
239 	int i = 0, ucode = 0, type, code;
240 	vm_offset_t eva;
241 #ifdef POWERFAIL_NMI
242 	static int lastalert = 0;
243 #endif
244 
245 	atomic_add_int(&cnt.v_trap, 1);
246 
247 	if ((frame.tf_eflags & PSL_I) == 0) {
248 		/*
249 		 * Buggy application or kernel code has disabled
250 		 * interrupts and then trapped.  Enabling interrupts
251 		 * now is wrong, but it is better than running with
252 		 * interrupts disabled until they are accidentally
253 		 * enabled later.  XXX Consider whether is this still
254 		 * correct.
255 		 */
256 		type = frame.tf_trapno;
257 		if (ISPL(frame.tf_cs) == SEL_UPL || (frame.tf_eflags & PSL_VM))
258 			printf(
259 			    "pid %ld (%s): trap %d with interrupts disabled\n",
260 			    (long)curproc->p_pid, curproc->p_comm, type);
261 		else if (type != T_BPTFLT && type != T_TRCTRAP)
262 			/*
263 			 * XXX not quite right, since this may be for a
264 			 * multiple fault in user mode.
265 			 */
266 			printf("kernel trap %d with interrupts disabled\n",
267 			    type);
268 		enable_intr();
269 	}
270 
271 	eva = 0;
272 	if (frame.tf_trapno == T_PAGEFLT) {
273 		/*
274 		 * For some Cyrix CPUs, %cr2 is clobbered by
275 		 * interrupts.  This problem is worked around by using
276 		 * an interrupt gate for the pagefault handler.  We
277 		 * are finally ready to read %cr2 and then must
278 		 * reenable interrupts.
279 		 */
280 		eva = rcr2();
281 		enable_intr();
282 	}
283 
284 	mtx_enter(&Giant, MTX_DEF);
285 
286 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
287 restart:
288 #endif
289 
290 	type = frame.tf_trapno;
291 	code = frame.tf_err;
292 
293         if ((ISPL(frame.tf_cs) == SEL_UPL) ||
294 	    ((frame.tf_eflags & PSL_VM) && !in_vm86call)) {
295 		/* user trap */
296 
297 		sticks = p->p_sticks;
298 		p->p_md.md_regs = &frame;
299 
300 		switch (type) {
301 		case T_PRIVINFLT:	/* privileged instruction fault */
302 			ucode = type;
303 			i = SIGILL;
304 			break;
305 
306 		case T_BPTFLT:		/* bpt instruction fault */
307 		case T_TRCTRAP:		/* trace trap */
308 			frame.tf_eflags &= ~PSL_T;
309 			i = SIGTRAP;
310 			break;
311 
312 		case T_ARITHTRAP:	/* arithmetic trap */
313 			ucode = code;
314 			i = SIGFPE;
315 			break;
316 
317 			/*
318 			 * The following two traps can happen in
319 			 * vm86 mode, and, if so, we want to handle
320 			 * them specially.
321 			 */
322 		case T_PROTFLT:		/* general protection fault */
323 		case T_STKFLT:		/* stack fault */
324 			if (frame.tf_eflags & PSL_VM) {
325 				i = vm86_emulate((struct vm86frame *)&frame);
326 				if (i == 0)
327 					goto user;
328 				break;
329 			}
330 			/* FALL THROUGH */
331 
332 		case T_SEGNPFLT:	/* segment not present fault */
333 		case T_TSSFLT:		/* invalid TSS fault */
334 		case T_DOUBLEFLT:	/* double fault */
335 		default:
336 			ucode = code + BUS_SEGM_FAULT ;
337 			i = SIGBUS;
338 			break;
339 
340 		case T_PAGEFLT:		/* page fault */
341 			i = trap_pfault(&frame, TRUE, eva);
342 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
343 			if (i == -2) {
344 				/*
345 				 * f00f hack workaround has triggered, treat
346 				 * as illegal instruction not page fault.
347 				 */
348 				frame.tf_trapno = T_PRIVINFLT;
349 				goto restart;
350 			}
351 #endif
352 			if (i == -1)
353 				goto out;
354 			if (i == 0)
355 				goto user;
356 
357 			ucode = T_PAGEFLT;
358 			break;
359 
360 		case T_DIVIDE:		/* integer divide fault */
361 			ucode = FPE_INTDIV;
362 			i = SIGFPE;
363 			break;
364 
365 #if NISA > 0
366 		case T_NMI:
367 #ifdef POWERFAIL_NMI
368 #ifndef TIMER_FREQ
369 #  define TIMER_FREQ 1193182
370 #endif
371 			if (time_second - lastalert > 10) {
372 				log(LOG_WARNING, "NMI: power fail\n");
373 				sysbeep(TIMER_FREQ/880, hz);
374 				lastalert = time_second;
375 			}
376 			goto out;
377 #else /* !POWERFAIL_NMI */
378 			/* machine/parity/power fail/"kitchen sink" faults */
379 			if (isa_nmi(code) == 0) {
380 #ifdef DDB
381 				/*
382 				 * NMI can be hooked up to a pushbutton
383 				 * for debugging.
384 				 */
385 				if (ddb_on_nmi) {
386 					printf ("NMI ... going to debugger\n");
387 					kdb_trap (type, 0, &frame);
388 				}
389 #endif /* DDB */
390 				goto out;
391 			} else if (panic_on_nmi)
392 				panic("NMI indicates hardware failure");
393 			break;
394 #endif /* POWERFAIL_NMI */
395 #endif /* NISA > 0 */
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 #if NNPX > 0
409 			/* transparent fault (due to context switch "late") */
410 			if (npxdna())
411 				goto out;
412 #endif
413 			if (!pmath_emulate) {
414 				i = SIGFPE;
415 				ucode = FPE_FPU_NP_TRAP;
416 				break;
417 			}
418 			i = (*pmath_emulate)(&frame);
419 			if (i == 0) {
420 				if (!(frame.tf_eflags & PSL_T))
421 					goto out;
422 				frame.tf_eflags &= ~PSL_T;
423 				i = SIGTRAP;
424 			}
425 			/* else ucode = emulator_only_knows() XXX */
426 			break;
427 
428 		case T_FPOPFLT:		/* FPU operand fetch fault */
429 			ucode = T_FPOPFLT;
430 			i = SIGILL;
431 			break;
432 		}
433 	} else {
434 		/* kernel trap */
435 
436 		switch (type) {
437 		case T_PAGEFLT:			/* page fault */
438 			(void) trap_pfault(&frame, FALSE, eva);
439 			goto out;
440 
441 		case T_DNA:
442 #if NNPX > 0
443 			/*
444 			 * The kernel is apparently using npx for copying.
445 			 * XXX this should be fatal unless the kernel has
446 			 * registered such use.
447 			 */
448 			if (npxdna())
449 				goto out;
450 #endif
451 			break;
452 
453 			/*
454 			 * The following two traps can happen in
455 			 * vm86 mode, and, if so, we want to handle
456 			 * them specially.
457 			 */
458 		case T_PROTFLT:		/* general protection fault */
459 		case T_STKFLT:		/* stack fault */
460 			if (frame.tf_eflags & PSL_VM) {
461 				i = vm86_emulate((struct vm86frame *)&frame);
462 				if (i != 0)
463 					/*
464 					 * returns to original process
465 					 */
466 					vm86_trap((struct vm86frame *)&frame);
467 				goto out;
468 			}
469 			if (type == T_STKFLT)
470 				break;
471 
472 			/* FALL THROUGH */
473 
474 		case T_SEGNPFLT:	/* segment not present fault */
475 			if (in_vm86call)
476 				break;
477 
478 			if (intr_nesting_level != 0)
479 				break;
480 
481 			/*
482 			 * Invalid %fs's and %gs's can be created using
483 			 * procfs or PT_SETREGS or by invalidating the
484 			 * underlying LDT entry.  This causes a fault
485 			 * in kernel mode when the kernel attempts to
486 			 * switch contexts.  Lose the bad context
487 			 * (XXX) so that we can continue, and generate
488 			 * a signal.
489 			 */
490 			if (frame.tf_eip == (int)cpu_switch_load_gs) {
491 				curpcb->pcb_gs = 0;
492 				psignal(p, SIGBUS);
493 				goto out;
494 			}
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 (curpcb && curpcb->pcb_onfault) {
523 				frame.tf_eip = (int)curpcb->pcb_onfault;
524 				goto out;
525 			}
526 			break;
527 
528 		case T_TSSFLT:
529 			/*
530 			 * PSL_NT can be set in user mode and isn't cleared
531 			 * automatically when the kernel is entered.  This
532 			 * causes a TSS fault when the kernel attempts to
533 			 * `iret' because the TSS link is uninitialized.  We
534 			 * want to get this fault so that we can fix the
535 			 * problem here and not every time the kernel is
536 			 * entered.
537 			 */
538 			if (frame.tf_eflags & PSL_NT) {
539 				frame.tf_eflags &= ~PSL_NT;
540 				goto out;
541 			}
542 			break;
543 
544 		case T_TRCTRAP:	 /* trace trap */
545 			if (frame.tf_eip == (int)IDTVEC(syscall)) {
546 				/*
547 				 * We've just entered system mode via the
548 				 * syscall lcall.  Continue single stepping
549 				 * silently until the syscall handler has
550 				 * saved the flags.
551 				 */
552 				goto out;
553 			}
554 			if (frame.tf_eip == (int)IDTVEC(syscall) + 1) {
555 				/*
556 				 * The syscall handler has now saved the
557 				 * flags.  Stop single stepping it.
558 				 */
559 				frame.tf_eflags &= ~PSL_T;
560 				goto out;
561 			}
562 			/*
563 			 * Ignore debug register trace traps due to
564 			 * accesses in the user's address space, which
565 			 * can happen under several conditions such as
566 			 * if a user sets a watchpoint on a buffer and
567 			 * then passes that buffer to a system call.
568 			 * We still want to get TRCTRAPS for addresses
569 			 * in kernel space because that is useful when
570 			 * debugging the kernel.
571 			 */
572 			if (user_dbreg_trap() && !in_vm86call) {
573 				/*
574 				 * Reset breakpoint bits because the
575 				 * processor doesn't
576 				 */
577 				load_dr6(rdr6() & 0xfffffff0);
578 				goto out;
579 			}
580 			/*
581 			 * Fall through (TRCTRAP kernel mode, kernel address)
582 			 */
583 		case T_BPTFLT:
584 			/*
585 			 * If DDB is enabled, let it handle the debugger trap.
586 			 * Otherwise, debugger traps "can't happen".
587 			 */
588 #ifdef DDB
589 			if (kdb_trap (type, 0, &frame))
590 				goto out;
591 #endif
592 			break;
593 
594 #if NISA > 0
595 		case T_NMI:
596 #ifdef POWERFAIL_NMI
597 			if (time_second - lastalert > 10) {
598 				log(LOG_WARNING, "NMI: power fail\n");
599 				sysbeep(TIMER_FREQ/880, hz);
600 				lastalert = time_second;
601 			}
602 			goto out;
603 #else /* !POWERFAIL_NMI */
604 			/* machine/parity/power fail/"kitchen sink" faults */
605 			if (isa_nmi(code) == 0) {
606 #ifdef DDB
607 				/*
608 				 * NMI can be hooked up to a pushbutton
609 				 * for debugging.
610 				 */
611 				if (ddb_on_nmi) {
612 					printf ("NMI ... going to debugger\n");
613 					kdb_trap (type, 0, &frame);
614 				}
615 #endif /* DDB */
616 				goto out;
617 			} else if (panic_on_nmi == 0)
618 				goto out;
619 			/* FALL THROUGH */
620 #endif /* POWERFAIL_NMI */
621 #endif /* NISA > 0 */
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(p, 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(p, &frame, sticks, 1);
646 out:
647 	mtx_exit(&Giant, MTX_DEF);
648 }
649 
650 #ifdef notyet
651 /*
652  * This version doesn't allow a page fault to user space while
653  * in the kernel. The rest of the kernel needs to be made "safe"
654  * before this can be used. I think the only things remaining
655  * to be made safe are the iBCS2 code and the process tracing/
656  * debugging code.
657  */
658 static int
659 trap_pfault(frame, usermode, eva)
660 	struct trapframe *frame;
661 	int usermode;
662 	vm_offset_t eva;
663 {
664 	vm_offset_t va;
665 	struct vmspace *vm = NULL;
666 	vm_map_t map = 0;
667 	int rv = 0;
668 	vm_prot_t ftype;
669 	struct proc *p = curproc;
670 
671 	if (frame->tf_err & PGEX_W)
672 		ftype = VM_PROT_WRITE;
673 	else
674 		ftype = VM_PROT_READ;
675 
676 	va = trunc_page(eva);
677 	if (va < VM_MIN_KERNEL_ADDRESS) {
678 		vm_offset_t v;
679 		vm_page_t mpte;
680 
681 		if (p == NULL ||
682 		    (!usermode && va < VM_MAXUSER_ADDRESS &&
683 		     (intr_nesting_level != 0 || curpcb == NULL ||
684 		      curpcb->pcb_onfault == NULL))) {
685 			trap_fatal(frame, eva);
686 			return (-1);
687 		}
688 
689 		/*
690 		 * This is a fault on non-kernel virtual memory.
691 		 * vm is initialized above to NULL. If curproc is NULL
692 		 * or curproc->p_vmspace is NULL the fault is fatal.
693 		 */
694 		vm = p->p_vmspace;
695 		if (vm == NULL)
696 			goto nogo;
697 
698 		map = &vm->vm_map;
699 
700 		/*
701 		 * Keep swapout from messing with us during this
702 		 *	critical time.
703 		 */
704 		++p->p_lock;
705 
706 		/*
707 		 * Grow the stack if necessary
708 		 */
709 		/* grow_stack returns false only if va falls into
710 		 * a growable stack region and the stack growth
711 		 * fails.  It returns true if va was not within
712 		 * a growable stack region, or if the stack
713 		 * growth succeeded.
714 		 */
715 		if (!grow_stack (p, va)) {
716 			rv = KERN_FAILURE;
717 			--p->p_lock;
718 			goto nogo;
719 		}
720 
721 		/* Fault in the user page: */
722 		rv = vm_fault(map, va, ftype,
723 			      (ftype & VM_PROT_WRITE) ? VM_FAULT_DIRTY
724 						      : VM_FAULT_NORMAL);
725 
726 		--p->p_lock;
727 	} else {
728 		/*
729 		 * Don't allow user-mode faults in kernel address space.
730 		 */
731 		if (usermode)
732 			goto nogo;
733 
734 		/*
735 		 * Since we know that kernel virtual address addresses
736 		 * always have pte pages mapped, we just have to fault
737 		 * the page.
738 		 */
739 		rv = vm_fault(kernel_map, va, ftype, VM_FAULT_NORMAL);
740 	}
741 
742 	if (rv == KERN_SUCCESS)
743 		return (0);
744 nogo:
745 	if (!usermode) {
746 		if (intr_nesting_level == 0 && curpcb && curpcb->pcb_onfault) {
747 			frame->tf_eip = (int)curpcb->pcb_onfault;
748 			return (0);
749 		}
750 		trap_fatal(frame, eva);
751 		return (-1);
752 	}
753 
754 	/* kludge to pass faulting virtual address to sendsig */
755 	frame->tf_err = eva;
756 
757 	return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
758 }
759 #endif
760 
761 int
762 trap_pfault(frame, usermode, eva)
763 	struct trapframe *frame;
764 	int usermode;
765 	vm_offset_t eva;
766 {
767 	vm_offset_t va;
768 	struct vmspace *vm = NULL;
769 	vm_map_t map = 0;
770 	int rv = 0;
771 	vm_prot_t ftype;
772 	struct proc *p = curproc;
773 
774 	va = trunc_page(eva);
775 	if (va >= KERNBASE) {
776 		/*
777 		 * Don't allow user-mode faults in kernel address space.
778 		 * An exception:  if the faulting address is the invalid
779 		 * instruction entry in the IDT, then the Intel Pentium
780 		 * F00F bug workaround was triggered, and we need to
781 		 * treat it is as an illegal instruction, and not a page
782 		 * fault.
783 		 */
784 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
785 		if ((eva == (unsigned int)&idt[6]) && has_f00f_bug)
786 			return -2;
787 #endif
788 		if (usermode)
789 			goto nogo;
790 
791 		map = kernel_map;
792 	} else {
793 		/*
794 		 * This is a fault on non-kernel virtual memory.
795 		 * vm is initialized above to NULL. If curproc is NULL
796 		 * or curproc->p_vmspace is NULL the fault is fatal.
797 		 */
798 		if (p != NULL)
799 			vm = p->p_vmspace;
800 
801 		if (vm == NULL)
802 			goto nogo;
803 
804 		map = &vm->vm_map;
805 	}
806 
807 	if (frame->tf_err & PGEX_W)
808 		ftype = VM_PROT_WRITE;
809 	else
810 		ftype = VM_PROT_READ;
811 
812 	if (map != kernel_map) {
813 		/*
814 		 * Keep swapout from messing with us during this
815 		 *	critical time.
816 		 */
817 		++p->p_lock;
818 
819 		/*
820 		 * Grow the stack if necessary
821 		 */
822 		/* grow_stack returns false only if va falls into
823 		 * a growable stack region and the stack growth
824 		 * fails.  It returns true if va was not within
825 		 * a growable stack region, or if the stack
826 		 * growth succeeded.
827 		 */
828 		if (!grow_stack (p, va)) {
829 			rv = KERN_FAILURE;
830 			--p->p_lock;
831 			goto nogo;
832 		}
833 
834 		/* Fault in the user page: */
835 		rv = vm_fault(map, va, ftype,
836 			      (ftype & VM_PROT_WRITE) ? VM_FAULT_DIRTY
837 						      : VM_FAULT_NORMAL);
838 
839 		--p->p_lock;
840 	} else {
841 		/*
842 		 * Don't have to worry about process locking or stacks in the kernel.
843 		 */
844 		rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
845 	}
846 
847 	if (rv == KERN_SUCCESS)
848 		return (0);
849 nogo:
850 	if (!usermode) {
851 		if (intr_nesting_level == 0 && curpcb && curpcb->pcb_onfault) {
852 			frame->tf_eip = (int)curpcb->pcb_onfault;
853 			return (0);
854 		}
855 		trap_fatal(frame, eva);
856 		return (-1);
857 	}
858 
859 	/* kludge to pass faulting virtual address to sendsig */
860 	frame->tf_err = eva;
861 
862 	return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
863 }
864 
865 static void
866 trap_fatal(frame, eva)
867 	struct trapframe *frame;
868 	vm_offset_t eva;
869 {
870 	int code, type, ss, esp;
871 	struct soft_segment_descriptor softseg;
872 
873 	code = frame->tf_err;
874 	type = frame->tf_trapno;
875 	sdtossd(&gdt[IDXSEL(frame->tf_cs & 0xffff)].sd, &softseg);
876 
877 	if (type <= MAX_TRAP_MSG)
878 		printf("\n\nFatal trap %d: %s while in %s mode\n",
879 			type, trap_msg[type],
880         		frame->tf_eflags & PSL_VM ? "vm86" :
881 			ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel");
882 #ifdef SMP
883 	/* two seperate prints in case of a trap on an unmapped page */
884 	printf("cpuid = %d; ", cpuid);
885 	printf("lapic.id = %08x\n", lapic.id);
886 #endif
887 	if (type == T_PAGEFLT) {
888 		printf("fault virtual address	= 0x%x\n", eva);
889 		printf("fault code		= %s %s, %s\n",
890 			code & PGEX_U ? "user" : "supervisor",
891 			code & PGEX_W ? "write" : "read",
892 			code & PGEX_P ? "protection violation" : "page not present");
893 	}
894 	printf("instruction pointer	= 0x%x:0x%x\n",
895 	       frame->tf_cs & 0xffff, frame->tf_eip);
896         if ((ISPL(frame->tf_cs) == SEL_UPL) || (frame->tf_eflags & PSL_VM)) {
897 		ss = frame->tf_ss & 0xffff;
898 		esp = frame->tf_esp;
899 	} else {
900 		ss = GSEL(GDATA_SEL, SEL_KPL);
901 		esp = (int)&frame->tf_esp;
902 	}
903 	printf("stack pointer	        = 0x%x:0x%x\n", ss, esp);
904 	printf("frame pointer	        = 0x%x:0x%x\n", ss, frame->tf_ebp);
905 	printf("code segment		= base 0x%x, limit 0x%x, type 0x%x\n",
906 	       softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type);
907 	printf("			= DPL %d, pres %d, def32 %d, gran %d\n",
908 	       softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_def32,
909 	       softseg.ssd_gran);
910 	printf("processor eflags	= ");
911 	if (frame->tf_eflags & PSL_T)
912 		printf("trace trap, ");
913 	if (frame->tf_eflags & PSL_I)
914 		printf("interrupt enabled, ");
915 	if (frame->tf_eflags & PSL_NT)
916 		printf("nested task, ");
917 	if (frame->tf_eflags & PSL_RF)
918 		printf("resume, ");
919 	if (frame->tf_eflags & PSL_VM)
920 		printf("vm86, ");
921 	printf("IOPL = %d\n", (frame->tf_eflags & PSL_IOPL) >> 12);
922 	printf("current process		= ");
923 	if (curproc) {
924 		printf("%lu (%s)\n",
925 		    (u_long)curproc->p_pid, curproc->p_comm ?
926 		    curproc->p_comm : "");
927 	} else {
928 		printf("Idle\n");
929 	}
930 
931 #ifdef KDB
932 	if (kdb_trap(&psl))
933 		return;
934 #endif
935 #ifdef DDB
936 	if ((debugger_on_panic || db_active) && kdb_trap(type, 0, frame))
937 		return;
938 #endif
939 	printf("trap number		= %d\n", type);
940 	if (type <= MAX_TRAP_MSG)
941 		panic(trap_msg[type]);
942 	else
943 		panic("unknown/reserved trap");
944 }
945 
946 /*
947  * Double fault handler. Called when a fault occurs while writing
948  * a frame for a trap/exception onto the stack. This usually occurs
949  * when the stack overflows (such is the case with infinite recursion,
950  * for example).
951  *
952  * XXX Note that the current PTD gets replaced by IdlePTD when the
953  * task switch occurs. This means that the stack that was active at
954  * the time of the double fault is not available at <kstack> unless
955  * the machine was idle when the double fault occurred. The downside
956  * of this is that "trace <ebp>" in ddb won't work.
957  */
958 void
959 dblfault_handler()
960 {
961 	printf("\nFatal double fault:\n");
962 	printf("eip = 0x%x\n", common_tss.tss_eip);
963 	printf("esp = 0x%x\n", common_tss.tss_esp);
964 	printf("ebp = 0x%x\n", common_tss.tss_ebp);
965 #ifdef SMP
966 	/* two seperate prints in case of a trap on an unmapped page */
967 	printf("cpuid = %d; ", cpuid);
968 	printf("lapic.id = %08x\n", lapic.id);
969 #endif
970 	panic("double fault");
971 }
972 
973 /*
974  * Compensate for 386 brain damage (missing URKR).
975  * This is a little simpler than the pagefault handler in trap() because
976  * it the page tables have already been faulted in and high addresses
977  * are thrown out early for other reasons.
978  */
979 int trapwrite(addr)
980 	unsigned addr;
981 {
982 	struct proc *p;
983 	vm_offset_t va;
984 	struct vmspace *vm;
985 	int rv;
986 
987 	va = trunc_page((vm_offset_t)addr);
988 	/*
989 	 * XXX - MAX is END.  Changed > to >= for temp. fix.
990 	 */
991 	if (va >= VM_MAXUSER_ADDRESS)
992 		return (1);
993 
994 	p = curproc;
995 	vm = p->p_vmspace;
996 
997 	++p->p_lock;
998 
999 	if (!grow_stack (p, va)) {
1000 		--p->p_lock;
1001 		return (1);
1002 	}
1003 
1004 	/*
1005 	 * fault the data page
1006 	 */
1007 	rv = vm_fault(&vm->vm_map, va, VM_PROT_WRITE, VM_FAULT_DIRTY);
1008 
1009 	--p->p_lock;
1010 
1011 	if (rv != KERN_SUCCESS)
1012 		return 1;
1013 
1014 	return (0);
1015 }
1016 
1017 /*
1018  *	syscall2 -	MP aware system call request C handler
1019  *
1020  *	A system call is essentially treated as a trap except that the
1021  *	MP lock is not held on entry or return.  We are responsible for
1022  *	obtaining the MP lock if necessary and for handling ASTs
1023  *	(e.g. a task switch) prior to return.
1024  *
1025  *	In general, only simple access and manipulation of curproc and
1026  *	the current stack is allowed without having to hold MP lock.
1027  */
1028 void
1029 syscall2(frame)
1030 	struct trapframe frame;
1031 {
1032 	caddr_t params;
1033 	int i;
1034 	struct sysent *callp;
1035 	struct proc *p = curproc;
1036 	u_quad_t sticks;
1037 	int error;
1038 	int narg;
1039 	int args[8];
1040 	int have_giant = 0;
1041 	u_int code;
1042 
1043 	atomic_add_int(&cnt.v_syscall, 1);
1044 
1045 #ifdef DIAGNOSTIC
1046 	if (ISPL(frame.tf_cs) != SEL_UPL) {
1047 		mtx_enter(&Giant, MTX_DEF);
1048 		panic("syscall");
1049 		/* NOT REACHED */
1050 	}
1051 #endif
1052 
1053 	/*
1054 	 * handle atomicy by looping since interrupts are enabled and the
1055 	 * MP lock is not held.
1056 	 */
1057 	sticks = ((volatile struct proc *)p)->p_sticks;
1058 	while (sticks != ((volatile struct proc *)p)->p_sticks)
1059 		sticks = ((volatile struct proc *)p)->p_sticks;
1060 
1061 	p->p_md.md_regs = &frame;
1062 	params = (caddr_t)frame.tf_esp + sizeof(int);
1063 	code = frame.tf_eax;
1064 
1065 	if (p->p_sysent->sv_prepsyscall) {
1066 		/*
1067 		 * The prep code is not MP aware.
1068 		 */
1069 		mtx_enter(&Giant, MTX_DEF);
1070 		(*p->p_sysent->sv_prepsyscall)(&frame, args, &code, &params);
1071 		mtx_exit(&Giant, MTX_DEF);
1072 	} else {
1073 		/*
1074 		 * Need to check if this is a 32 bit or 64 bit syscall.
1075 		 * fuword is MP aware.
1076 		 */
1077 		if (code == SYS_syscall) {
1078 			/*
1079 			 * Code is first argument, followed by actual args.
1080 			 */
1081 			code = fuword(params);
1082 			params += sizeof(int);
1083 		} else if (code == SYS___syscall) {
1084 			/*
1085 			 * Like syscall, but code is a quad, so as to maintain
1086 			 * quad alignment for the rest of the arguments.
1087 			 */
1088 			code = fuword(params);
1089 			params += sizeof(quad_t);
1090 		}
1091 	}
1092 
1093  	if (p->p_sysent->sv_mask)
1094  		code &= p->p_sysent->sv_mask;
1095 
1096  	if (code >= p->p_sysent->sv_size)
1097  		callp = &p->p_sysent->sv_table[0];
1098   	else
1099  		callp = &p->p_sysent->sv_table[code];
1100 
1101 	narg = callp->sy_narg & SYF_ARGMASK;
1102 
1103 	/*
1104 	 * copyin is MP aware, but the tracing code is not
1105 	 */
1106 	if (params && (i = narg * sizeof(int)) &&
1107 	    (error = copyin(params, (caddr_t)args, (u_int)i))) {
1108 		mtx_enter(&Giant, MTX_DEF);
1109 		have_giant = 1;
1110 #ifdef KTRACE
1111 		if (KTRPOINT(p, KTR_SYSCALL))
1112 			ktrsyscall(p->p_tracep, code, narg, args);
1113 #endif
1114 		goto bad;
1115 	}
1116 
1117 	/*
1118 	 * Try to run the syscall without the MP lock if the syscall
1119 	 * is MP safe.  We have to obtain the MP lock no matter what if
1120 	 * we are ktracing
1121 	 */
1122 	if ((callp->sy_narg & SYF_MPSAFE) == 0) {
1123 		mtx_enter(&Giant, MTX_DEF);
1124 		have_giant = 1;
1125 	}
1126 
1127 #ifdef KTRACE
1128 	if (KTRPOINT(p, KTR_SYSCALL)) {
1129 		if (have_giant == 0) {
1130 			mtx_enter(&Giant, MTX_DEF);
1131 			have_giant = 1;
1132 		}
1133 		ktrsyscall(p->p_tracep, code, narg, args);
1134 	}
1135 #endif
1136 	p->p_retval[0] = 0;
1137 	p->p_retval[1] = frame.tf_edx;
1138 
1139 	STOPEVENT(p, S_SCE, narg);	/* MP aware */
1140 
1141 	error = (*callp->sy_call)(p, args);
1142 
1143 	/*
1144 	 * MP SAFE (we may or may not have the MP lock at this point)
1145 	 */
1146 	switch (error) {
1147 	case 0:
1148 		/*
1149 		 * Reinitialize proc pointer `p' as it may be different
1150 		 * if this is a child returning from fork syscall.
1151 		 */
1152 		p = curproc;
1153 		frame.tf_eax = p->p_retval[0];
1154 		frame.tf_edx = p->p_retval[1];
1155 		frame.tf_eflags &= ~PSL_C;
1156 		break;
1157 
1158 	case ERESTART:
1159 		/*
1160 		 * Reconstruct pc, assuming lcall $X,y is 7 bytes,
1161 		 * int 0x80 is 2 bytes. We saved this in tf_err.
1162 		 */
1163 		frame.tf_eip -= frame.tf_err;
1164 		break;
1165 
1166 	case EJUSTRETURN:
1167 		break;
1168 
1169 	default:
1170 bad:
1171  		if (p->p_sysent->sv_errsize) {
1172  			if (error >= p->p_sysent->sv_errsize)
1173   				error = -1;	/* XXX */
1174    			else
1175   				error = p->p_sysent->sv_errtbl[error];
1176 		}
1177 		frame.tf_eax = error;
1178 		frame.tf_eflags |= PSL_C;
1179 		break;
1180 	}
1181 
1182 	/*
1183 	 * Traced syscall.  trapsignal() is not MP aware.
1184 	 */
1185 	if ((frame.tf_eflags & PSL_T) && !(frame.tf_eflags & PSL_VM)) {
1186 		if (have_giant == 0) {
1187 			mtx_enter(&Giant, MTX_DEF);
1188 			have_giant = 1;
1189 		}
1190 		frame.tf_eflags &= ~PSL_T;
1191 		trapsignal(p, SIGTRAP, 0);
1192 	}
1193 
1194 	/*
1195 	 * Handle reschedule and other end-of-syscall issues
1196 	 */
1197 	have_giant = userret(p, &frame, sticks, have_giant);
1198 
1199 #ifdef KTRACE
1200 	if (KTRPOINT(p, KTR_SYSRET)) {
1201 		if (have_giant == 0) {
1202 			mtx_enter(&Giant, MTX_DEF);
1203 			have_giant = 1;
1204 		}
1205 		ktrsysret(p->p_tracep, code, error, p->p_retval[0]);
1206 	}
1207 #endif
1208 
1209 	/*
1210 	 * This works because errno is findable through the
1211 	 * register set.  If we ever support an emulation where this
1212 	 * is not the case, this code will need to be revisited.
1213 	 */
1214 	STOPEVENT(p, S_SCX, code);
1215 
1216 	/*
1217 	 * Release the MP lock if we had to get it
1218 	 */
1219 	if (have_giant)
1220 		mtx_exit(&Giant, MTX_DEF);
1221 
1222 	mtx_assert(&sched_lock, MA_NOTOWNED);
1223 	mtx_assert(&Giant, MA_NOTOWNED);
1224 }
1225 
1226 void
1227 ast(frame)
1228 	struct trapframe frame;
1229 {
1230 	struct proc *p = CURPROC;
1231 	u_quad_t sticks;
1232 
1233 	/*
1234 	 * handle atomicy by looping since interrupts are enabled and the
1235 	 * MP lock is not held.
1236 	 */
1237 	sticks = ((volatile struct proc *)p)->p_sticks;
1238 	while (sticks != ((volatile struct proc *)p)->p_sticks)
1239 		sticks = ((volatile struct proc *)p)->p_sticks;
1240 
1241 	astoff();
1242 	atomic_add_int(&cnt.v_soft, 1);
1243 	if (p->p_flag & P_OWEUPC) {
1244 		mtx_enter(&Giant, MTX_DEF);
1245 		p->p_flag &= ~P_OWEUPC;
1246 		addupc_task(p, p->p_stats->p_prof.pr_addr,
1247 			    p->p_stats->p_prof.pr_ticks);
1248 	}
1249 	if (p->p_flag & P_ALRMPEND) {
1250 		if (!mtx_owned(&Giant))
1251 			mtx_enter(&Giant, MTX_DEF);
1252 		p->p_flag &= ~P_ALRMPEND;
1253 		psignal(p, SIGVTALRM);
1254 	}
1255 	if (p->p_flag & P_PROFPEND) {
1256 		if (!mtx_owned(&Giant))
1257 			mtx_enter(&Giant, MTX_DEF);
1258 		p->p_flag &= ~P_PROFPEND;
1259 		psignal(p, SIGPROF);
1260 	}
1261 	if (userret(p, &frame, sticks, mtx_owned(&Giant)) != 0)
1262 		mtx_exit(&Giant, MTX_DEF);
1263 }
1264 
1265 /*
1266  * Simplified back end of syscall(), used when returning from fork()
1267  * directly into user mode.  Giant is not held on entry, and must not
1268  * be held on return.
1269  */
1270 void
1271 fork_return(p, frame)
1272 	struct proc *p;
1273 	struct trapframe frame;
1274 {
1275 	int	have_giant;
1276 
1277 	frame.tf_eax = 0;		/* Child returns zero */
1278 	frame.tf_eflags &= ~PSL_C;	/* success */
1279 	frame.tf_edx = 1;
1280 
1281 	have_giant = userret(p, &frame, 0, mtx_owned(&Giant));
1282 #ifdef KTRACE
1283 	if (KTRPOINT(p, KTR_SYSRET)) {
1284 		if (have_giant == 0) {
1285 			mtx_enter(&Giant, MTX_DEF);
1286 			have_giant = 1;
1287 		}
1288 		ktrsysret(p->p_tracep, SYS_fork, 0, 0);
1289 	}
1290 #endif
1291 	if (have_giant)
1292 		mtx_exit(&Giant, MTX_DEF);
1293 }
1294