xref: /freebsd/sys/kern/subr_trap.c (revision 40d5099441391b6a117252779c6fb798c5189fcb)
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  *	$Id: trap.c,v 1.106 1997/08/20 05:25:48 fsmp Exp $
39  */
40 
41 /*
42  * 386 Trap and System call handling
43  */
44 
45 #include "opt_ktrace.h"
46 #include "opt_ddb.h"
47 
48 #include <sys/param.h>
49 #include <sys/systm.h>
50 #include <sys/proc.h>
51 #include <sys/kernel.h>
52 #include <sys/syscall.h>
53 #include <sys/sysent.h>
54 #include <sys/vmmeter.h>
55 #ifdef KTRACE
56 #include <sys/ktrace.h>
57 #endif
58 
59 #include <vm/vm.h>
60 #include <vm/vm_param.h>
61 #include <vm/vm_prot.h>
62 #include <sys/lock.h>
63 #include <vm/pmap.h>
64 #include <vm/vm_kern.h>
65 #include <vm/vm_map.h>
66 #include <vm/vm_page.h>
67 #include <vm/vm_extern.h>
68 
69 #include <sys/user.h>
70 
71 #include <machine/cpu.h>
72 #include <machine/ipl.h>
73 #include <machine/md_var.h>
74 #include <machine/psl.h>
75 #include <machine/../isa/intr_machdep.h>
76 #include <machine/smp.h>
77 
78 #ifdef POWERFAIL_NMI
79 #include <sys/syslog.h>
80 #include <machine/clock.h>
81 #endif
82 
83 #include "isa.h"
84 #include "npx.h"
85 
86 extern struct i386tss common_tss;
87 
88 int (*vm86_emulate) __P((struct vm86frame *));
89 int (*pmath_emulate) __P((struct trapframe *));
90 
91 extern void trap __P((struct trapframe frame));
92 extern int trapwrite __P((unsigned addr));
93 extern void syscall __P((struct trapframe frame));
94 
95 static int trap_pfault __P((struct trapframe *, int));
96 static void trap_fatal __P((struct trapframe *));
97 void dblfault_handler __P((void));
98 
99 extern inthand_t IDTVEC(syscall);
100 
101 #define MAX_TRAP_MSG		28
102 static char *trap_msg[] = {
103 	"",					/*  0 unused */
104 	"privileged instruction fault",		/*  1 T_PRIVINFLT */
105 	"",					/*  2 unused */
106 	"breakpoint instruction fault",		/*  3 T_BPTFLT */
107 	"",					/*  4 unused */
108 	"",					/*  5 unused */
109 	"arithmetic trap",			/*  6 T_ARITHTRAP */
110 	"system forced exception",		/*  7 T_ASTFLT */
111 	"",					/*  8 unused */
112 	"general protection fault",		/*  9 T_PROTFLT */
113 	"trace trap",				/* 10 T_TRCTRAP */
114 	"",					/* 11 unused */
115 	"page fault",				/* 12 T_PAGEFLT */
116 	"",					/* 13 unused */
117 	"alignment fault",			/* 14 T_ALIGNFLT */
118 	"",					/* 15 unused */
119 	"",					/* 16 unused */
120 	"",					/* 17 unused */
121 	"integer divide fault",			/* 18 T_DIVIDE */
122 	"non-maskable interrupt trap",		/* 19 T_NMI */
123 	"overflow trap",			/* 20 T_OFLOW */
124 	"FPU bounds check fault",		/* 21 T_BOUND */
125 	"FPU device not available",		/* 22 T_DNA */
126 	"double fault",				/* 23 T_DOUBLEFLT */
127 	"FPU operand fetch fault",		/* 24 T_FPOPFLT */
128 	"invalid TSS fault",			/* 25 T_TSSFLT */
129 	"segment not present fault",		/* 26 T_SEGNPFLT */
130 	"stack fault",				/* 27 T_STKFLT */
131 	"machine check trap",			/* 28 T_MCHK */
132 };
133 
134 static void userret __P((struct proc *p, struct trapframe *frame,
135 			 u_quad_t oticks));
136 
137 static inline void
138 userret(p, frame, oticks)
139 	struct proc *p;
140 	struct trapframe *frame;
141 	u_quad_t oticks;
142 {
143 	int sig, s;
144 
145 	while ((sig = CURSIG(p)) != 0)
146 		postsig(sig);
147 
148 #if 0
149 	if (!want_resched &&
150 		(p->p_priority <= p->p_usrpri) &&
151 		(p->p_rtprio.type == RTP_PRIO_NORMAL)) {
152 		 int newpriority;
153 		 p->p_estcpu += 1;
154 		 newpriority = PUSER + p->p_estcpu / 4 + 2 * p->p_nice;
155 		 newpriority = min(newpriority, MAXPRI);
156 		 p->p_usrpri = newpriority;
157 	}
158 #endif
159 
160 	p->p_priority = p->p_usrpri;
161 	if (want_resched) {
162 		/*
163 		 * Since we are curproc, clock will normally just change
164 		 * our priority without moving us from one queue to another
165 		 * (since the running process is not on a queue.)
166 		 * If that happened after we setrunqueue ourselves but before we
167 		 * mi_switch()'ed, we might not be on the queue indicated by
168 		 * our priority.
169 		 */
170 		s = splhigh();
171 		setrunqueue(p);
172 		p->p_stats->p_ru.ru_nivcsw++;
173 		mi_switch();
174 		splx(s);
175 		while ((sig = CURSIG(p)) != 0)
176 			postsig(sig);
177 	}
178 	/*
179 	 * Charge system time if profiling.
180 	 */
181 	if (p->p_flag & P_PROFIL)
182 		addupc_task(p, frame->tf_eip,
183 			    (u_int)(p->p_sticks - oticks) * psratio);
184 
185 	curpriority = p->p_priority;
186 }
187 
188 /*
189  * Exception, fault, and trap interface to the FreeBSD kernel.
190  * This common code is called from assembly language IDT gate entry
191  * routines that prepare a suitable stack frame, and restore this
192  * frame after the exception has been processed.
193  */
194 
195 void
196 trap(frame)
197 	struct trapframe frame;
198 {
199 	struct proc *p = curproc;
200 	u_quad_t sticks = 0;
201 	int i = 0, ucode = 0, type, code;
202 #ifdef DEBUG
203 	u_long eva;
204 #endif
205 
206 	type = frame.tf_trapno;
207 	code = frame.tf_err;
208 
209         if ((ISPL(frame.tf_cs) == SEL_UPL) || (frame.tf_eflags & PSL_VM)) {
210 		/* user trap */
211 
212 		sticks = p->p_sticks;
213 		p->p_md.md_regs = &frame;
214 
215 		switch (type) {
216 		case T_PRIVINFLT:	/* privileged instruction fault */
217 			ucode = type;
218 			i = SIGILL;
219 			break;
220 
221 		case T_BPTFLT:		/* bpt instruction fault */
222 		case T_TRCTRAP:		/* trace trap */
223 			frame.tf_eflags &= ~PSL_T;
224 			i = SIGTRAP;
225 			break;
226 
227 		case T_ARITHTRAP:	/* arithmetic trap */
228 			ucode = code;
229 			i = SIGFPE;
230 			break;
231 
232 		case T_ASTFLT:		/* Allow process switch */
233 			astoff();
234 			cnt.v_soft++;
235 			if (p->p_flag & P_OWEUPC) {
236 				p->p_flag &= ~P_OWEUPC;
237 				addupc_task(p, p->p_stats->p_prof.pr_addr,
238 					    p->p_stats->p_prof.pr_ticks);
239 			}
240 			goto out;
241 
242 			/*
243 			 * The following two traps can happen in
244 			 * vm86 mode, and, if so, we want to handle
245 			 * them specially.
246 			 */
247 		case T_PROTFLT:		/* general protection fault */
248 		case T_STKFLT:		/* stack fault */
249 			if (vm86_emulate && (frame.tf_eflags & PSL_VM)) {
250 				i = (*vm86_emulate)((struct vm86frame *)&frame);
251 				if (i == 0)
252 					goto out;
253 				break;
254 			}
255 			/* FALL THROUGH */
256 
257 		case T_SEGNPFLT:	/* segment not present fault */
258 		case T_TSSFLT:		/* invalid TSS fault */
259 		case T_DOUBLEFLT:	/* double fault */
260 		default:
261 			ucode = code + BUS_SEGM_FAULT ;
262 			i = SIGBUS;
263 			break;
264 
265 		case T_PAGEFLT:		/* page fault */
266 			i = trap_pfault(&frame, TRUE);
267 			if (i == -1)
268 				return;
269 			if (i == 0)
270 				goto out;
271 
272 			ucode = T_PAGEFLT;
273 			break;
274 
275 		case T_DIVIDE:		/* integer divide fault */
276 			ucode = FPE_INTDIV_TRAP;
277 			i = SIGFPE;
278 			break;
279 
280 #if NISA > 0
281 		case T_NMI:
282 #ifdef POWERFAIL_NMI
283 			goto handle_powerfail;
284 #else /* !POWERFAIL_NMI */
285 #ifdef DDB
286 			/* NMI can be hooked up to a pushbutton for debugging */
287 			printf ("NMI ... going to debugger\n");
288 			if (kdb_trap (type, 0, &frame))
289 				return;
290 #endif /* DDB */
291 			/* machine/parity/power fail/"kitchen sink" faults */
292 			if (isa_nmi(code) == 0) return;
293 			panic("NMI indicates hardware failure");
294 #endif /* POWERFAIL_NMI */
295 #endif /* NISA > 0 */
296 
297 		case T_OFLOW:		/* integer overflow fault */
298 			ucode = FPE_INTOVF_TRAP;
299 			i = SIGFPE;
300 			break;
301 
302 		case T_BOUND:		/* bounds check fault */
303 			ucode = FPE_SUBRNG_TRAP;
304 			i = SIGFPE;
305 			break;
306 
307 		case T_DNA:
308 #if NNPX > 0
309 			/* if a transparent fault (due to context switch "late") */
310 			if (npxdna())
311 				return;
312 #endif
313 			if (!pmath_emulate) {
314 				i = SIGFPE;
315 				ucode = FPE_FPU_NP_TRAP;
316 				break;
317 			}
318 			i = (*pmath_emulate)(&frame);
319 			if (i == 0) {
320 				if (!(frame.tf_eflags & PSL_T))
321 					return;
322 				frame.tf_eflags &= ~PSL_T;
323 				i = SIGTRAP;
324 			}
325 			/* else ucode = emulator_only_knows() XXX */
326 			break;
327 
328 		case T_FPOPFLT:		/* FPU operand fetch fault */
329 			ucode = T_FPOPFLT;
330 			i = SIGILL;
331 			break;
332 		}
333 	} else {
334 		/* kernel trap */
335 
336 		switch (type) {
337 		case T_PAGEFLT:			/* page fault */
338 			(void) trap_pfault(&frame, FALSE);
339 			return;
340 
341 		case T_DNA:
342 #if NNPX > 0
343 			/*
344 			 * The kernel is apparently using npx for copying.
345 			 * XXX this should be fatal unless the kernel has
346 			 * registered such use.
347 			 */
348 			if (npxdna())
349 				return;
350 #endif
351 			break;
352 
353 		case T_PROTFLT:		/* general protection fault */
354 		case T_SEGNPFLT:	/* segment not present fault */
355 			/*
356 			 * Invalid segment selectors and out of bounds
357 			 * %eip's and %esp's can be set up in user mode.
358 			 * This causes a fault in kernel mode when the
359 			 * kernel tries to return to user mode.  We want
360 			 * to get this fault so that we can fix the
361 			 * problem here and not have to check all the
362 			 * selectors and pointers when the user changes
363 			 * them.
364 			 */
365 #define	MAYBE_DORETI_FAULT(where, whereto)				\
366 	do {								\
367 		if (frame.tf_eip == (int)where) {			\
368 			frame.tf_eip = (int)whereto;			\
369 			return;						\
370 		}							\
371 	} while (0)
372 
373 			if (intr_nesting_level == 0) {
374 				/*
375 				 * Invalid %fs's and %gs's can be created using
376 				 * procfs or PT_SETREGS or by invalidating the
377 				 * underlying LDT entry.  This causes a fault
378 				 * in kernel mode when the kernel attempts to
379 				 * switch contexts.  Lose the bad context
380 				 * (XXX) so that we can continue, and generate
381 				 * a signal.
382 				 */
383 				if (frame.tf_eip == (int)cpu_switch_load_fs) {
384 					curpcb->pcb_fs = 0;
385 					psignal(p, SIGBUS);
386 					return;
387 				}
388 				if (frame.tf_eip == (int)cpu_switch_load_gs) {
389 					curpcb->pcb_gs = 0;
390 					psignal(p, SIGBUS);
391 					return;
392 				}
393 				MAYBE_DORETI_FAULT(doreti_iret,
394 						   doreti_iret_fault);
395 				MAYBE_DORETI_FAULT(doreti_popl_ds,
396 						   doreti_popl_ds_fault);
397 				MAYBE_DORETI_FAULT(doreti_popl_es,
398 						   doreti_popl_es_fault);
399 				if (curpcb && curpcb->pcb_onfault) {
400 					frame.tf_eip = (int)curpcb->pcb_onfault;
401 					return;
402 				}
403 			}
404 			break;
405 
406 		case T_TSSFLT:
407 			/*
408 			 * PSL_NT can be set in user mode and isn't cleared
409 			 * automatically when the kernel is entered.  This
410 			 * causes a TSS fault when the kernel attempts to
411 			 * `iret' because the TSS link is uninitialized.  We
412 			 * want to get this fault so that we can fix the
413 			 * problem here and not every time the kernel is
414 			 * entered.
415 			 */
416 			if (frame.tf_eflags & PSL_NT) {
417 				frame.tf_eflags &= ~PSL_NT;
418 				return;
419 			}
420 			break;
421 
422 		case T_TRCTRAP:	 /* trace trap */
423 			if (frame.tf_eip == (int)IDTVEC(syscall)) {
424 				/*
425 				 * We've just entered system mode via the
426 				 * syscall lcall.  Continue single stepping
427 				 * silently until the syscall handler has
428 				 * saved the flags.
429 				 */
430 				return;
431 			}
432 			if (frame.tf_eip == (int)IDTVEC(syscall) + 1) {
433 				/*
434 				 * The syscall handler has now saved the
435 				 * flags.  Stop single stepping it.
436 				 */
437 				frame.tf_eflags &= ~PSL_T;
438 				return;
439 			}
440 			/*
441 			 * Fall through.
442 			 */
443 		case T_BPTFLT:
444 			/*
445 			 * If DDB is enabled, let it handle the debugger trap.
446 			 * Otherwise, debugger traps "can't happen".
447 			 */
448 #ifdef DDB
449 			if (kdb_trap (type, 0, &frame))
450 				return;
451 #endif
452 			break;
453 
454 #if NISA > 0
455 		case T_NMI:
456 #ifdef POWERFAIL_NMI
457 #ifndef TIMER_FREQ
458 #  define TIMER_FREQ 1193182
459 #endif
460 	handle_powerfail:
461 		{
462 		  static unsigned lastalert = 0;
463 
464 		  if(time.tv_sec - lastalert > 10)
465 		    {
466 		      log(LOG_WARNING, "NMI: power fail\n");
467 		      sysbeep(TIMER_FREQ/880, hz);
468 		      lastalert = time.tv_sec;
469 		    }
470 		  return;
471 		}
472 #else /* !POWERFAIL_NMI */
473 #ifdef DDB
474 			/* NMI can be hooked up to a pushbutton for debugging */
475 			printf ("NMI ... going to debugger\n");
476 			if (kdb_trap (type, 0, &frame))
477 				return;
478 #endif /* DDB */
479 			/* machine/parity/power fail/"kitchen sink" faults */
480 			if (isa_nmi(code) == 0) return;
481 			/* FALL THROUGH */
482 #endif /* POWERFAIL_NMI */
483 #endif /* NISA > 0 */
484 		}
485 
486 		trap_fatal(&frame);
487 		return;
488 	}
489 
490 	trapsignal(p, i, ucode);
491 
492 #ifdef DEBUG
493 	eva = rcr2();
494 	if (type <= MAX_TRAP_MSG) {
495 		uprintf("fatal process exception: %s",
496 			trap_msg[type]);
497 		if ((type == T_PAGEFLT) || (type == T_PROTFLT))
498 			uprintf(", fault VA = 0x%x", eva);
499 		uprintf("\n");
500 	}
501 #endif
502 
503 out:
504 	userret(p, &frame, sticks);
505 }
506 
507 #ifdef notyet
508 /*
509  * This version doesn't allow a page fault to user space while
510  * in the kernel. The rest of the kernel needs to be made "safe"
511  * before this can be used. I think the only things remaining
512  * to be made safe are the iBCS2 code and the process tracing/
513  * debugging code.
514  */
515 static int
516 trap_pfault(frame, usermode)
517 	struct trapframe *frame;
518 	int usermode;
519 {
520 	vm_offset_t va;
521 	struct vmspace *vm = NULL;
522 	vm_map_t map = 0;
523 	int rv = 0;
524 	vm_prot_t ftype;
525 	int eva;
526 	struct proc *p = curproc;
527 
528 	if (frame->tf_err & PGEX_W)
529 		ftype = VM_PROT_READ | VM_PROT_WRITE;
530 	else
531 		ftype = VM_PROT_READ;
532 
533 	eva = rcr2();
534 	va = trunc_page((vm_offset_t)eva);
535 
536 	if (va < VM_MIN_KERNEL_ADDRESS) {
537 		vm_offset_t v;
538 		vm_page_t mpte;
539 
540 		if (p == NULL ||
541 		    (!usermode && va < VM_MAXUSER_ADDRESS &&
542 		     (intr_nesting_level != 0 || curpcb == NULL ||
543 		      curpcb->pcb_onfault == NULL))) {
544 			trap_fatal(frame);
545 			return (-1);
546 		}
547 
548 		/*
549 		 * This is a fault on non-kernel virtual memory.
550 		 * vm is initialized above to NULL. If curproc is NULL
551 		 * or curproc->p_vmspace is NULL the fault is fatal.
552 		 */
553 		vm = p->p_vmspace;
554 		if (vm == NULL)
555 			goto nogo;
556 
557 		map = &vm->vm_map;
558 
559 		/*
560 		 * Keep swapout from messing with us during this
561 		 *	critical time.
562 		 */
563 		++p->p_lock;
564 
565 		/*
566 		 * Grow the stack if necessary
567 		 */
568 		if ((caddr_t)va > vm->vm_maxsaddr
569 		    && (caddr_t)va < (caddr_t)USRSTACK) {
570 			if (!grow(p, va)) {
571 				rv = KERN_FAILURE;
572 				--p->p_lock;
573 				goto nogo;
574 			}
575 		}
576 
577 		/* Fault in the user page: */
578 		rv = vm_fault(map, va, ftype,
579 			(ftype & VM_PROT_WRITE) ? VM_FAULT_DIRTY : 0);
580 
581 		--p->p_lock;
582 	} else {
583 		/*
584 		 * Don't allow user-mode faults in kernel address space.
585 		 */
586 		if (usermode)
587 			goto nogo;
588 
589 		/*
590 		 * Since we know that kernel virtual address addresses
591 		 * always have pte pages mapped, we just have to fault
592 		 * the page.
593 		 */
594 		rv = vm_fault(kernel_map, va, ftype, FALSE);
595 	}
596 
597 	if (rv == KERN_SUCCESS)
598 		return (0);
599 nogo:
600 	if (!usermode) {
601 		if (intr_nesting_level == 0 && curpcb && curpcb->pcb_onfault) {
602 			frame->tf_eip = (int)curpcb->pcb_onfault;
603 			return (0);
604 		}
605 		trap_fatal(frame);
606 		return (-1);
607 	}
608 
609 	/* kludge to pass faulting virtual address to sendsig */
610 	frame->tf_err = eva;
611 
612 	return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
613 }
614 #endif
615 
616 int
617 trap_pfault(frame, usermode)
618 	struct trapframe *frame;
619 	int usermode;
620 {
621 	vm_offset_t va;
622 	struct vmspace *vm = NULL;
623 	vm_map_t map = 0;
624 	int rv = 0;
625 	vm_prot_t ftype;
626 	int eva;
627 	struct proc *p = curproc;
628 
629 	eva = rcr2();
630 	va = trunc_page((vm_offset_t)eva);
631 
632 	if (va >= KERNBASE) {
633 		/*
634 		 * Don't allow user-mode faults in kernel address space.
635 		 */
636 		if (usermode)
637 			goto nogo;
638 
639 		map = kernel_map;
640 	} else {
641 		/*
642 		 * This is a fault on non-kernel virtual memory.
643 		 * vm is initialized above to NULL. If curproc is NULL
644 		 * or curproc->p_vmspace is NULL the fault is fatal.
645 		 */
646 		if (p != NULL)
647 			vm = p->p_vmspace;
648 
649 		if (vm == NULL)
650 			goto nogo;
651 
652 		map = &vm->vm_map;
653 	}
654 
655 	if (frame->tf_err & PGEX_W)
656 		ftype = VM_PROT_READ | VM_PROT_WRITE;
657 	else
658 		ftype = VM_PROT_READ;
659 
660 	if (map != kernel_map) {
661 		/*
662 		 * Keep swapout from messing with us during this
663 		 *	critical time.
664 		 */
665 		++p->p_lock;
666 
667 		/*
668 		 * Grow the stack if necessary
669 		 */
670 		if ((caddr_t)va > vm->vm_maxsaddr
671 		    && (caddr_t)va < (caddr_t)USRSTACK) {
672 			if (!grow(p, va)) {
673 				rv = KERN_FAILURE;
674 				--p->p_lock;
675 				goto nogo;
676 			}
677 		}
678 
679 		/* Fault in the user page: */
680 		rv = vm_fault(map, va, ftype,
681 			(ftype & VM_PROT_WRITE) ? VM_FAULT_DIRTY : 0);
682 
683 		--p->p_lock;
684 	} else {
685 		/*
686 		 * Don't have to worry about process locking or stacks in the kernel.
687 		 */
688 		rv = vm_fault(map, va, ftype, FALSE);
689 	}
690 
691 	if (rv == KERN_SUCCESS)
692 		return (0);
693 nogo:
694 	if (!usermode) {
695 		if (intr_nesting_level == 0 && curpcb && curpcb->pcb_onfault) {
696 			frame->tf_eip = (int)curpcb->pcb_onfault;
697 			return (0);
698 		}
699 		trap_fatal(frame);
700 		return (-1);
701 	}
702 
703 	/* kludge to pass faulting virtual address to sendsig */
704 	frame->tf_err = eva;
705 
706 	return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
707 }
708 
709 static void
710 trap_fatal(frame)
711 	struct trapframe *frame;
712 {
713 	int code, type, eva, ss, esp;
714 	struct soft_segment_descriptor softseg;
715 
716 	code = frame->tf_err;
717 	type = frame->tf_trapno;
718 	eva = rcr2();
719 	sdtossd(&gdt[IDXSEL(frame->tf_cs & 0xffff)].sd, &softseg);
720 
721 	if (type <= MAX_TRAP_MSG)
722 		printf("\n\nFatal trap %d: %s while in %s mode\n",
723 			type, trap_msg[type],
724         		frame->tf_eflags & PSL_VM ? "vm86" :
725 			ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel");
726 #ifdef SMP
727 	printf("cpuid = %d\n", cpuid);
728 #endif
729 	if (type == T_PAGEFLT) {
730 		printf("fault virtual address	= 0x%x\n", eva);
731 		printf("fault code		= %s %s, %s\n",
732 			code & PGEX_U ? "user" : "supervisor",
733 			code & PGEX_W ? "write" : "read",
734 			code & PGEX_P ? "protection violation" : "page not present");
735 	}
736 	printf("instruction pointer	= 0x%x:0x%x\n",
737 	       frame->tf_cs & 0xffff, frame->tf_eip);
738         if ((ISPL(frame->tf_cs) == SEL_UPL) || (frame->tf_eflags & PSL_VM)) {
739 		ss = frame->tf_ss & 0xffff;
740 		esp = frame->tf_esp;
741 	} else {
742 		ss = GSEL(GDATA_SEL, SEL_KPL);
743 		esp = (int)&frame->tf_esp;
744 	}
745 	printf("stack pointer	        = 0x%x:0x%x\n", ss, esp);
746 	printf("frame pointer	        = 0x%x:0x%x\n", ss, frame->tf_ebp);
747 	printf("code segment		= base 0x%x, limit 0x%x, type 0x%x\n",
748 	       softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type);
749 	printf("			= DPL %d, pres %d, def32 %d, gran %d\n",
750 	       softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_def32,
751 	       softseg.ssd_gran);
752 	printf("processor eflags	= ");
753 	if (frame->tf_eflags & PSL_T)
754 		printf("trace trap, ");
755 	if (frame->tf_eflags & PSL_I)
756 		printf("interrupt enabled, ");
757 	if (frame->tf_eflags & PSL_NT)
758 		printf("nested task, ");
759 	if (frame->tf_eflags & PSL_RF)
760 		printf("resume, ");
761 	if (frame->tf_eflags & PSL_VM)
762 		printf("vm86, ");
763 	printf("IOPL = %d\n", (frame->tf_eflags & PSL_IOPL) >> 12);
764 	printf("current process		= ");
765 	if (curproc) {
766 		printf("%lu (%s)\n",
767 		    (u_long)curproc->p_pid, curproc->p_comm ?
768 		    curproc->p_comm : "");
769 	} else {
770 		printf("Idle\n");
771 	}
772 	printf("interrupt mask		= ");
773 	if ((cpl & net_imask) == net_imask)
774 		printf("net ");
775 	if ((cpl & tty_imask) == tty_imask)
776 		printf("tty ");
777 	if ((cpl & bio_imask) == bio_imask)
778 		printf("bio ");
779 	if (cpl == 0)
780 		printf("none");
781 #ifdef SMP
782 /**
783  *  XXX FIXME:
784  *	we probably SHOULD have stopped the other CPUs before now!
785  *	another CPU COULD have been touching cpl at this moment...
786  */
787 	printf(" <- SMP: XXX");
788 #endif
789 	printf("\n");
790 
791 #ifdef KDB
792 	if (kdb_trap(&psl))
793 		return;
794 #endif
795 #ifdef DDB
796 	if (kdb_trap (type, 0, frame))
797 		return;
798 #endif
799 	printf("trap number		= %d\n", type);
800 	if (type <= MAX_TRAP_MSG)
801 		panic(trap_msg[type]);
802 	else
803 		panic("unknown/reserved trap");
804 }
805 
806 /*
807  * Double fault handler. Called when a fault occurs while writing
808  * a frame for a trap/exception onto the stack. This usually occurs
809  * when the stack overflows (such is the case with infinite recursion,
810  * for example).
811  *
812  * XXX Note that the current PTD gets replaced by IdlePTD when the
813  * task switch occurs. This means that the stack that was active at
814  * the time of the double fault is not available at <kstack> unless
815  * the machine was idle when the double fault occurred. The downside
816  * of this is that "trace <ebp>" in ddb won't work.
817  */
818 void
819 dblfault_handler()
820 {
821 	printf("\nFatal double fault:\n");
822 	printf("eip = 0x%x\n", common_tss.tss_eip);
823 	printf("esp = 0x%x\n", common_tss.tss_esp);
824 	printf("ebp = 0x%x\n", common_tss.tss_ebp);
825 #ifdef SMP
826 	printf("cpuid = %d\n", cpuid);
827 #endif
828 	panic("double fault");
829 }
830 
831 /*
832  * Compensate for 386 brain damage (missing URKR).
833  * This is a little simpler than the pagefault handler in trap() because
834  * it the page tables have already been faulted in and high addresses
835  * are thrown out early for other reasons.
836  */
837 int trapwrite(addr)
838 	unsigned addr;
839 {
840 	struct proc *p;
841 	vm_offset_t va;
842 	struct vmspace *vm;
843 	int rv;
844 
845 	va = trunc_page((vm_offset_t)addr);
846 	/*
847 	 * XXX - MAX is END.  Changed > to >= for temp. fix.
848 	 */
849 	if (va >= VM_MAXUSER_ADDRESS)
850 		return (1);
851 
852 	p = curproc;
853 	vm = p->p_vmspace;
854 
855 	++p->p_lock;
856 
857 	if ((caddr_t)va >= vm->vm_maxsaddr
858 	    && (caddr_t)va < (caddr_t)USRSTACK) {
859 		if (!grow(p, va)) {
860 			--p->p_lock;
861 			return (1);
862 		}
863 	}
864 
865 	/*
866 	 * fault the data page
867 	 */
868 	rv = vm_fault(&vm->vm_map, va, VM_PROT_READ|VM_PROT_WRITE, VM_FAULT_DIRTY);
869 
870 	--p->p_lock;
871 
872 	if (rv != KERN_SUCCESS)
873 		return 1;
874 
875 	return (0);
876 }
877 
878 /*
879  * System call request from POSIX system call gate interface to kernel.
880  * Like trap(), argument is call by reference.
881  */
882 void
883 syscall(frame)
884 	struct trapframe frame;
885 {
886 	caddr_t params;
887 	int i;
888 	struct sysent *callp;
889 	struct proc *p = curproc;
890 	u_quad_t sticks;
891 	int error;
892 	int args[8], rval[2];
893 	u_int code;
894 
895 	sticks = p->p_sticks;
896 	if (ISPL(frame.tf_cs) != SEL_UPL)
897 		panic("syscall");
898 
899 	p->p_md.md_regs = &frame;
900 	params = (caddr_t)frame.tf_esp + sizeof(int);
901 	code = frame.tf_eax;
902 	if (p->p_sysent->sv_prepsyscall) {
903 		(*p->p_sysent->sv_prepsyscall)(&frame, args, &code, &params);
904 	} else {
905 		/*
906 		 * Need to check if this is a 32 bit or 64 bit syscall.
907 		 */
908 		if (code == SYS_syscall) {
909 			/*
910 			 * Code is first argument, followed by actual args.
911 			 */
912 			code = fuword(params);
913 			params += sizeof(int);
914 		} else if (code == SYS___syscall) {
915 			/*
916 			 * Like syscall, but code is a quad, so as to maintain
917 			 * quad alignment for the rest of the arguments.
918 			 */
919 			code = fuword(params);
920 			params += sizeof(quad_t);
921 		}
922 	}
923 
924  	if (p->p_sysent->sv_mask)
925  		code &= p->p_sysent->sv_mask;
926 
927  	if (code >= p->p_sysent->sv_size)
928  		callp = &p->p_sysent->sv_table[0];
929   	else
930  		callp = &p->p_sysent->sv_table[code];
931 
932 	if (params && (i = callp->sy_narg * sizeof(int)) &&
933 	    (error = copyin(params, (caddr_t)args, (u_int)i))) {
934 #ifdef KTRACE
935 		if (KTRPOINT(p, KTR_SYSCALL))
936 			ktrsyscall(p->p_tracep, code, callp->sy_narg, args);
937 #endif
938 		goto bad;
939 	}
940 #ifdef KTRACE
941 	if (KTRPOINT(p, KTR_SYSCALL))
942 		ktrsyscall(p->p_tracep, code, callp->sy_narg, args);
943 #endif
944 	rval[0] = 0;
945 	rval[1] = frame.tf_edx;
946 
947 	error = (*callp->sy_call)(p, args, rval);
948 
949 	switch (error) {
950 
951 	case 0:
952 		/*
953 		 * Reinitialize proc pointer `p' as it may be different
954 		 * if this is a child returning from fork syscall.
955 		 */
956 		p = curproc;
957 		frame.tf_eax = rval[0];
958 		frame.tf_edx = rval[1];
959 		frame.tf_eflags &= ~PSL_C;
960 		break;
961 
962 	case ERESTART:
963 		/*
964 		 * Reconstruct pc, assuming lcall $X,y is 7 bytes,
965 		 * int 0x80 is 2 bytes. We saved this in tf_err.
966 		 */
967 		frame.tf_eip -= frame.tf_err;
968 		break;
969 
970 	case EJUSTRETURN:
971 		break;
972 
973 	default:
974 bad:
975  		if (p->p_sysent->sv_errsize)
976  			if (error >= p->p_sysent->sv_errsize)
977   				error = -1;	/* XXX */
978    			else
979   				error = p->p_sysent->sv_errtbl[error];
980 		frame.tf_eax = error;
981 		frame.tf_eflags |= PSL_C;
982 		break;
983 	}
984 
985 	if ((frame.tf_eflags & PSL_T) && !(frame.tf_eflags & PSL_VM)) {
986 		/* Traced syscall. */
987 		frame.tf_eflags &= ~PSL_T;
988 		trapsignal(p, SIGTRAP, 0);
989 	}
990 
991 	userret(p, &frame, sticks);
992 
993 #ifdef KTRACE
994 	if (KTRPOINT(p, KTR_SYSRET))
995 		ktrsysret(p->p_tracep, code, error, rval[0]);
996 #endif
997 }
998 
999 /*
1000  * Simplified back end of syscall(), used when returning from fork()
1001  * directly into user mode.
1002  */
1003 void
1004 fork_return(p, frame)
1005 	struct proc *p;
1006 	struct trapframe frame;
1007 {
1008 	frame.tf_eax = 0;		/* Child returns zero */
1009 	frame.tf_eflags &= ~PSL_C;	/* success */
1010 	frame.tf_edx = 1;
1011 
1012 	userret(p, &frame, 0);
1013 #ifdef KTRACE
1014 	if (KTRPOINT(p, KTR_SYSRET))
1015 		ktrsysret(p->p_tracep, SYS_fork, 0, 0);
1016 #endif
1017 }
1018