xref: /freebsd/sys/kern/subr_trap.c (revision ce834215a70ff69e7e222827437116eee2f9ac6f)
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 1.99 1997/06/07 04:36:10 bde 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/acct.h>
52 #include <sys/kernel.h>
53 #include <sys/syscall.h>
54 #include <sys/sysent.h>
55 #include <sys/queue.h>
56 #include <sys/vmmeter.h>
57 #ifdef KTRACE
58 #include <sys/ktrace.h>
59 #endif
60 
61 #include <vm/vm.h>
62 #include <vm/vm_param.h>
63 #include <vm/vm_prot.h>
64 #include <sys/lock.h>
65 #include <vm/pmap.h>
66 #include <vm/vm_kern.h>
67 #include <vm/vm_map.h>
68 #include <vm/vm_page.h>
69 #include <vm/vm_extern.h>
70 
71 #include <sys/user.h>
72 
73 #include <machine/cpu.h>
74 #include <machine/ipl.h>
75 #include <machine/md_var.h>
76 #include <machine/psl.h>
77 #include <machine/reg.h>
78 #include <machine/trap.h>
79 #include <machine/../isa/intr_machdep.h>
80 #include <machine/smp.h>
81 
82 #ifdef POWERFAIL_NMI
83 #include <sys/syslog.h>
84 #include <machine/clock.h>
85 #endif
86 
87 #include "isa.h"
88 #include "npx.h"
89 
90 extern struct i386tss common_tss;
91 
92 int (*pmath_emulate) __P((struct trapframe *));
93 
94 extern void trap __P((struct trapframe frame));
95 extern int trapwrite __P((unsigned addr));
96 extern void syscall __P((struct trapframe frame));
97 
98 static int trap_pfault __P((struct trapframe *, int));
99 static void trap_fatal __P((struct trapframe *));
100 void dblfault_handler __P((void));
101 
102 extern inthand_t IDTVEC(syscall);
103 
104 #define MAX_TRAP_MSG		28
105 static char *trap_msg[] = {
106 	"",					/*  0 unused */
107 	"privileged instruction fault",		/*  1 T_PRIVINFLT */
108 	"",					/*  2 unused */
109 	"breakpoint instruction fault",		/*  3 T_BPTFLT */
110 	"",					/*  4 unused */
111 	"",					/*  5 unused */
112 	"arithmetic trap",			/*  6 T_ARITHTRAP */
113 	"system forced exception",		/*  7 T_ASTFLT */
114 	"",					/*  8 unused */
115 	"general protection fault",		/*  9 T_PROTFLT */
116 	"trace trap",				/* 10 T_TRCTRAP */
117 	"",					/* 11 unused */
118 	"page fault",				/* 12 T_PAGEFLT */
119 	"",					/* 13 unused */
120 	"alignment fault",			/* 14 T_ALIGNFLT */
121 	"",					/* 15 unused */
122 	"",					/* 16 unused */
123 	"",					/* 17 unused */
124 	"integer divide fault",			/* 18 T_DIVIDE */
125 	"non-maskable interrupt trap",		/* 19 T_NMI */
126 	"overflow trap",			/* 20 T_OFLOW */
127 	"FPU bounds check fault",		/* 21 T_BOUND */
128 	"FPU device not available",		/* 22 T_DNA */
129 	"double fault",				/* 23 T_DOUBLEFLT */
130 	"FPU operand fetch fault",		/* 24 T_FPOPFLT */
131 	"invalid TSS fault",			/* 25 T_TSSFLT */
132 	"segment not present fault",		/* 26 T_SEGNPFLT */
133 	"stack fault",				/* 27 T_STKFLT */
134 	"machine check trap",			/* 28 T_MCHK */
135 };
136 
137 static void userret __P((struct proc *p, struct trapframe *frame,
138 			 u_quad_t oticks));
139 
140 static inline void
141 userret(p, frame, oticks)
142 	struct proc *p;
143 	struct trapframe *frame;
144 	u_quad_t oticks;
145 {
146 	int sig, s;
147 
148 	while ((sig = CURSIG(p)) != 0)
149 		postsig(sig);
150 	p->p_priority = p->p_usrpri;
151 	if (want_resched) {
152 		/*
153 		 * Since we are curproc, clock will normally just change
154 		 * our priority without moving us from one queue to another
155 		 * (since the running process is not on a queue.)
156 		 * If that happened after we setrunqueue ourselves but before we
157 		 * mi_switch()'ed, we might not be on the queue indicated by
158 		 * our priority.
159 		 */
160 		s = splhigh();
161 		setrunqueue(p);
162 		p->p_stats->p_ru.ru_nivcsw++;
163 		mi_switch();
164 		splx(s);
165 		while ((sig = CURSIG(p)) != 0)
166 			postsig(sig);
167 	}
168 	/*
169 	 * Charge system time if profiling.
170 	 */
171 	if (p->p_flag & P_PROFIL)
172 		addupc_task(p, frame->tf_eip,
173 			    (u_int)(p->p_sticks - oticks) * psratio);
174 
175 	curpriority = p->p_priority;
176 }
177 
178 /*
179  * Exception, fault, and trap interface to the FreeBSD kernel.
180  * This common code is called from assembly language IDT gate entry
181  * routines that prepare a suitable stack frame, and restore this
182  * frame after the exception has been processed.
183  */
184 
185 void
186 trap(frame)
187 	struct trapframe frame;
188 {
189 	struct proc *p = curproc;
190 	u_quad_t sticks = 0;
191 	int i = 0, ucode = 0, type, code;
192 #ifdef DEBUG
193 	u_long eva;
194 #endif
195 
196 	type = frame.tf_trapno;
197 	code = frame.tf_err;
198 
199 	if (ISPL(frame.tf_cs) == SEL_UPL) {
200 		/* user trap */
201 
202 		sticks = p->p_sticks;
203 		p->p_md.md_regs = &frame;
204 
205 		switch (type) {
206 		case T_PRIVINFLT:	/* privileged instruction fault */
207 			ucode = type;
208 			i = SIGILL;
209 			break;
210 
211 		case T_BPTFLT:		/* bpt instruction fault */
212 		case T_TRCTRAP:		/* trace trap */
213 			frame.tf_eflags &= ~PSL_T;
214 			i = SIGTRAP;
215 			break;
216 
217 		case T_ARITHTRAP:	/* arithmetic trap */
218 			ucode = code;
219 			i = SIGFPE;
220 			break;
221 
222 		case T_ASTFLT:		/* Allow process switch */
223 			astoff();
224 			cnt.v_soft++;
225 			if (p->p_flag & P_OWEUPC) {
226 				p->p_flag &= ~P_OWEUPC;
227 				addupc_task(p, p->p_stats->p_prof.pr_addr,
228 					    p->p_stats->p_prof.pr_ticks);
229 			}
230 			goto out;
231 
232 		case T_PROTFLT:		/* general protection fault */
233 		case T_SEGNPFLT:	/* segment not present fault */
234 		case T_STKFLT:		/* stack fault */
235 		case T_TSSFLT:		/* invalid TSS fault */
236 		case T_DOUBLEFLT:	/* double fault */
237 		default:
238 			ucode = code + BUS_SEGM_FAULT ;
239 			i = SIGBUS;
240 			break;
241 
242 		case T_PAGEFLT:		/* page fault */
243 			i = trap_pfault(&frame, TRUE);
244 			if (i == -1)
245 				return;
246 			if (i == 0)
247 				goto out;
248 
249 			ucode = T_PAGEFLT;
250 			break;
251 
252 		case T_DIVIDE:		/* integer divide fault */
253 			ucode = FPE_INTDIV_TRAP;
254 			i = SIGFPE;
255 			break;
256 
257 #if NISA > 0
258 		case T_NMI:
259 #ifdef POWERFAIL_NMI
260 			goto handle_powerfail;
261 #else /* !POWERFAIL_NMI */
262 #ifdef DDB
263 			/* NMI can be hooked up to a pushbutton for debugging */
264 			printf ("NMI ... going to debugger\n");
265 			if (kdb_trap (type, 0, &frame))
266 				return;
267 #endif /* DDB */
268 			/* machine/parity/power fail/"kitchen sink" faults */
269 			if (isa_nmi(code) == 0) return;
270 			panic("NMI indicates hardware failure");
271 #endif /* POWERFAIL_NMI */
272 #endif /* NISA > 0 */
273 
274 		case T_OFLOW:		/* integer overflow fault */
275 			ucode = FPE_INTOVF_TRAP;
276 			i = SIGFPE;
277 			break;
278 
279 		case T_BOUND:		/* bounds check fault */
280 			ucode = FPE_SUBRNG_TRAP;
281 			i = SIGFPE;
282 			break;
283 
284 		case T_DNA:
285 #if NNPX > 0
286 			/* if a transparent fault (due to context switch "late") */
287 			if (npxdna())
288 				return;
289 #endif
290 			if (!pmath_emulate) {
291 				i = SIGFPE;
292 				ucode = FPE_FPU_NP_TRAP;
293 				break;
294 			}
295 			i = (*pmath_emulate)(&frame);
296 			if (i == 0) {
297 				if (!(frame.tf_eflags & PSL_T))
298 					return;
299 				frame.tf_eflags &= ~PSL_T;
300 				i = SIGTRAP;
301 			}
302 			/* else ucode = emulator_only_knows() XXX */
303 			break;
304 
305 		case T_FPOPFLT:		/* FPU operand fetch fault */
306 			ucode = T_FPOPFLT;
307 			i = SIGILL;
308 			break;
309 		}
310 	} else {
311 		/* kernel trap */
312 
313 		switch (type) {
314 		case T_PAGEFLT:			/* page fault */
315 			(void) trap_pfault(&frame, FALSE);
316 			return;
317 
318 		case T_DNA:
319 #if NNPX > 0
320 			/*
321 			 * The kernel is apparently using npx for copying.
322 			 * XXX this should be fatal unless the kernel has
323 			 * registered such use.
324 			 */
325 			if (npxdna())
326 				return;
327 #endif
328 			break;
329 
330 		case T_PROTFLT:		/* general protection fault */
331 		case T_SEGNPFLT:	/* segment not present fault */
332 			/*
333 			 * Invalid segment selectors and out of bounds
334 			 * %eip's and %esp's can be set up in user mode.
335 			 * This causes a fault in kernel mode when the
336 			 * kernel tries to return to user mode.  We want
337 			 * to get this fault so that we can fix the
338 			 * problem here and not have to check all the
339 			 * selectors and pointers when the user changes
340 			 * them.
341 			 */
342 #define	MAYBE_DORETI_FAULT(where, whereto)				\
343 	do {								\
344 		if (frame.tf_eip == (int)where) {			\
345 			frame.tf_eip = (int)whereto;			\
346 			return;						\
347 		}							\
348 	} while (0)
349 
350 			if (intr_nesting_level == 0) {
351 				/*
352 				 * Invalid %fs's and %gs's can be created using
353 				 * procfs or PT_SETREGS or by invalidating the
354 				 * underlying LDT entry.  This causes a fault
355 				 * in kernel mode when the kernel attempts to
356 				 * switch contexts.  Lose the bad context
357 				 * (XXX) so that we can continue, and generate
358 				 * a signal.
359 				 */
360 				if (frame.tf_eip == (int)cpu_switch_load_fs) {
361 					curpcb->pcb_fs = 0;
362 					psignal(p, SIGBUS);
363 					return;
364 				}
365 				if (frame.tf_eip == (int)cpu_switch_load_gs) {
366 					curpcb->pcb_gs = 0;
367 					psignal(p, SIGBUS);
368 					return;
369 				}
370 				MAYBE_DORETI_FAULT(doreti_iret,
371 						   doreti_iret_fault);
372 				MAYBE_DORETI_FAULT(doreti_popl_ds,
373 						   doreti_popl_ds_fault);
374 				MAYBE_DORETI_FAULT(doreti_popl_es,
375 						   doreti_popl_es_fault);
376 				if (curpcb && curpcb->pcb_onfault) {
377 					frame.tf_eip = (int)curpcb->pcb_onfault;
378 					return;
379 				}
380 			}
381 			break;
382 
383 		case T_TSSFLT:
384 			/*
385 			 * PSL_NT can be set in user mode and isn't cleared
386 			 * automatically when the kernel is entered.  This
387 			 * causes a TSS fault when the kernel attempts to
388 			 * `iret' because the TSS link is uninitialized.  We
389 			 * want to get this fault so that we can fix the
390 			 * problem here and not every time the kernel is
391 			 * entered.
392 			 */
393 			if (frame.tf_eflags & PSL_NT) {
394 				frame.tf_eflags &= ~PSL_NT;
395 				return;
396 			}
397 			break;
398 
399 		case T_TRCTRAP:	 /* trace trap */
400 			if (frame.tf_eip == (int)IDTVEC(syscall)) {
401 				/*
402 				 * We've just entered system mode via the
403 				 * syscall lcall.  Continue single stepping
404 				 * silently until the syscall handler has
405 				 * saved the flags.
406 				 */
407 				return;
408 			}
409 			if (frame.tf_eip == (int)IDTVEC(syscall) + 1) {
410 				/*
411 				 * The syscall handler has now saved the
412 				 * flags.  Stop single stepping it.
413 				 */
414 				frame.tf_eflags &= ~PSL_T;
415 				return;
416 			}
417 			/*
418 			 * Fall through.
419 			 */
420 		case T_BPTFLT:
421 			/*
422 			 * If DDB is enabled, let it handle the debugger trap.
423 			 * Otherwise, debugger traps "can't happen".
424 			 */
425 #ifdef DDB
426 			if (kdb_trap (type, 0, &frame))
427 				return;
428 #endif
429 			break;
430 
431 #if NISA > 0
432 		case T_NMI:
433 #ifdef POWERFAIL_NMI
434 #ifndef TIMER_FREQ
435 #  define TIMER_FREQ 1193182
436 #endif
437 	handle_powerfail:
438 		{
439 		  static unsigned lastalert = 0;
440 
441 		  if(time.tv_sec - lastalert > 10)
442 		    {
443 		      log(LOG_WARNING, "NMI: power fail\n");
444 		      sysbeep(TIMER_FREQ/880, hz);
445 		      lastalert = time.tv_sec;
446 		    }
447 		  return;
448 		}
449 #else /* !POWERFAIL_NMI */
450 #ifdef DDB
451 			/* NMI can be hooked up to a pushbutton for debugging */
452 			printf ("NMI ... going to debugger\n");
453 			if (kdb_trap (type, 0, &frame))
454 				return;
455 #endif /* DDB */
456 			/* machine/parity/power fail/"kitchen sink" faults */
457 			if (isa_nmi(code) == 0) return;
458 			/* FALL THROUGH */
459 #endif /* POWERFAIL_NMI */
460 #endif /* NISA > 0 */
461 		}
462 
463 		trap_fatal(&frame);
464 		return;
465 	}
466 
467 	trapsignal(p, i, ucode);
468 
469 #ifdef DEBUG
470 	eva = rcr2();
471 	if (type <= MAX_TRAP_MSG) {
472 		uprintf("fatal process exception: %s",
473 			trap_msg[type]);
474 		if ((type == T_PAGEFLT) || (type == T_PROTFLT))
475 			uprintf(", fault VA = 0x%x", eva);
476 		uprintf("\n");
477 	}
478 #endif
479 
480 out:
481 	userret(p, &frame, sticks);
482 }
483 
484 #ifdef notyet
485 /*
486  * This version doesn't allow a page fault to user space while
487  * in the kernel. The rest of the kernel needs to be made "safe"
488  * before this can be used. I think the only things remaining
489  * to be made safe are the iBCS2 code and the process tracing/
490  * debugging code.
491  */
492 static int
493 trap_pfault(frame, usermode)
494 	struct trapframe *frame;
495 	int usermode;
496 {
497 	vm_offset_t va;
498 	struct vmspace *vm = NULL;
499 	vm_map_t map = 0;
500 	int rv = 0;
501 	vm_prot_t ftype;
502 	int eva;
503 	struct proc *p = curproc;
504 
505 	if (frame->tf_err & PGEX_W)
506 		ftype = VM_PROT_READ | VM_PROT_WRITE;
507 	else
508 		ftype = VM_PROT_READ;
509 
510 	eva = rcr2();
511 	va = trunc_page((vm_offset_t)eva);
512 
513 	if (va < VM_MIN_KERNEL_ADDRESS) {
514 		vm_offset_t v;
515 		vm_page_t mpte;
516 
517 		if (p == NULL ||
518 		    (!usermode && va < VM_MAXUSER_ADDRESS &&
519 		     (intr_nesting_level != 0 || curpcb == NULL ||
520 		      curpcb->pcb_onfault == NULL))) {
521 			trap_fatal(frame);
522 			return (-1);
523 		}
524 
525 		/*
526 		 * This is a fault on non-kernel virtual memory.
527 		 * vm is initialized above to NULL. If curproc is NULL
528 		 * or curproc->p_vmspace is NULL the fault is fatal.
529 		 */
530 		vm = p->p_vmspace;
531 		if (vm == NULL)
532 			goto nogo;
533 
534 		map = &vm->vm_map;
535 
536 		/*
537 		 * Keep swapout from messing with us during this
538 		 *	critical time.
539 		 */
540 		++p->p_lock;
541 
542 		/*
543 		 * Grow the stack if necessary
544 		 */
545 		if ((caddr_t)va > vm->vm_maxsaddr
546 		    && (caddr_t)va < (caddr_t)USRSTACK) {
547 			if (!grow(p, va)) {
548 				rv = KERN_FAILURE;
549 				--p->p_lock;
550 				goto nogo;
551 			}
552 		}
553 
554 		/* Fault in the user page: */
555 		rv = vm_fault(map, va, ftype,
556 			(ftype & VM_PROT_WRITE) ? VM_FAULT_DIRTY : 0);
557 
558 		--p->p_lock;
559 	} else {
560 		/*
561 		 * Don't allow user-mode faults in kernel address space.
562 		 */
563 		if (usermode)
564 			goto nogo;
565 
566 		/*
567 		 * Since we know that kernel virtual address addresses
568 		 * always have pte pages mapped, we just have to fault
569 		 * the page.
570 		 */
571 		rv = vm_fault(kernel_map, va, ftype, FALSE);
572 	}
573 
574 	if (rv == KERN_SUCCESS)
575 		return (0);
576 nogo:
577 	if (!usermode) {
578 		if (intr_nesting_level == 0 && curpcb && curpcb->pcb_onfault) {
579 			frame->tf_eip = (int)curpcb->pcb_onfault;
580 			return (0);
581 		}
582 		trap_fatal(frame);
583 		return (-1);
584 	}
585 
586 	/* kludge to pass faulting virtual address to sendsig */
587 	frame->tf_err = eva;
588 
589 	return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
590 }
591 #endif
592 
593 int
594 trap_pfault(frame, usermode)
595 	struct trapframe *frame;
596 	int usermode;
597 {
598 	vm_offset_t va;
599 	struct vmspace *vm = NULL;
600 	vm_map_t map = 0;
601 	int rv = 0;
602 	vm_prot_t ftype;
603 	int eva;
604 	struct proc *p = curproc;
605 
606 	eva = rcr2();
607 	va = trunc_page((vm_offset_t)eva);
608 
609 	if (va >= KERNBASE) {
610 		/*
611 		 * Don't allow user-mode faults in kernel address space.
612 		 */
613 		if (usermode)
614 			goto nogo;
615 
616 		map = kernel_map;
617 	} else {
618 		/*
619 		 * This is a fault on non-kernel virtual memory.
620 		 * vm is initialized above to NULL. If curproc is NULL
621 		 * or curproc->p_vmspace is NULL the fault is fatal.
622 		 */
623 		if (p != NULL)
624 			vm = p->p_vmspace;
625 
626 		if (vm == NULL)
627 			goto nogo;
628 
629 		map = &vm->vm_map;
630 	}
631 
632 	if (frame->tf_err & PGEX_W)
633 		ftype = VM_PROT_READ | VM_PROT_WRITE;
634 	else
635 		ftype = VM_PROT_READ;
636 
637 	if (map != kernel_map) {
638 		/*
639 		 * Keep swapout from messing with us during this
640 		 *	critical time.
641 		 */
642 		++p->p_lock;
643 
644 		/*
645 		 * Grow the stack if necessary
646 		 */
647 		if ((caddr_t)va > vm->vm_maxsaddr
648 		    && (caddr_t)va < (caddr_t)USRSTACK) {
649 			if (!grow(p, va)) {
650 				rv = KERN_FAILURE;
651 				--p->p_lock;
652 				goto nogo;
653 			}
654 		}
655 
656 		/* Fault in the user page: */
657 		rv = vm_fault(map, va, ftype,
658 			(ftype & VM_PROT_WRITE) ? VM_FAULT_DIRTY : 0);
659 
660 		--p->p_lock;
661 	} else {
662 		/*
663 		 * Don't have to worry about process locking or stacks in the kernel.
664 		 */
665 		rv = vm_fault(map, va, ftype, FALSE);
666 	}
667 
668 	if (rv == KERN_SUCCESS)
669 		return (0);
670 nogo:
671 	if (!usermode) {
672 		if (intr_nesting_level == 0 && curpcb && curpcb->pcb_onfault) {
673 			frame->tf_eip = (int)curpcb->pcb_onfault;
674 			return (0);
675 		}
676 		trap_fatal(frame);
677 		return (-1);
678 	}
679 
680 	/* kludge to pass faulting virtual address to sendsig */
681 	frame->tf_err = eva;
682 
683 	return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
684 }
685 
686 static void
687 trap_fatal(frame)
688 	struct trapframe *frame;
689 {
690 	int code, type, eva, ss, esp;
691 	struct soft_segment_descriptor softseg;
692 
693 	code = frame->tf_err;
694 	type = frame->tf_trapno;
695 	eva = rcr2();
696 	sdtossd(&gdt[IDXSEL(frame->tf_cs & 0xffff)].sd, &softseg);
697 
698 	if (type <= MAX_TRAP_MSG)
699 		printf("\n\nFatal trap %d: %s while in %s mode\n",
700 			type, trap_msg[type],
701 			ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel");
702 #ifdef SMP
703 	printf("cpuid = %d\n", cpuid);
704 #endif
705 	if (type == T_PAGEFLT) {
706 		printf("fault virtual address	= 0x%x\n", eva);
707 		printf("fault code		= %s %s, %s\n",
708 			code & PGEX_U ? "user" : "supervisor",
709 			code & PGEX_W ? "write" : "read",
710 			code & PGEX_P ? "protection violation" : "page not present");
711 	}
712 	printf("instruction pointer	= 0x%x:0x%x\n",
713 	       frame->tf_cs & 0xffff, frame->tf_eip);
714 	if (ISPL(frame->tf_cs) == SEL_UPL) {
715 		ss = frame->tf_ss & 0xffff;
716 		esp = frame->tf_esp;
717 	} else {
718 		ss = GSEL(GDATA_SEL, SEL_KPL);
719 		esp = (int)&frame->tf_esp;
720 	}
721 	printf("stack pointer	        = 0x%x:0x%x\n", ss, esp);
722 	printf("frame pointer	        = 0x%x:0x%x\n", ss, frame->tf_ebp);
723 	printf("code segment		= base 0x%x, limit 0x%x, type 0x%x\n",
724 	       softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type);
725 	printf("			= DPL %d, pres %d, def32 %d, gran %d\n",
726 	       softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_def32,
727 	       softseg.ssd_gran);
728 	printf("processor eflags	= ");
729 	if (frame->tf_eflags & PSL_T)
730 		printf("trace trap, ");
731 	if (frame->tf_eflags & PSL_I)
732 		printf("interrupt enabled, ");
733 	if (frame->tf_eflags & PSL_NT)
734 		printf("nested task, ");
735 	if (frame->tf_eflags & PSL_RF)
736 		printf("resume, ");
737 	if (frame->tf_eflags & PSL_VM)
738 		printf("vm86, ");
739 	printf("IOPL = %d\n", (frame->tf_eflags & PSL_IOPL) >> 12);
740 	printf("current process		= ");
741 	if (curproc) {
742 		printf("%lu (%s)\n",
743 		    (u_long)curproc->p_pid, curproc->p_comm ?
744 		    curproc->p_comm : "");
745 	} else {
746 		printf("Idle\n");
747 	}
748 	printf("interrupt mask		= ");
749 	if ((cpl & net_imask) == net_imask)
750 		printf("net ");
751 	if ((cpl & tty_imask) == tty_imask)
752 		printf("tty ");
753 	if ((cpl & bio_imask) == bio_imask)
754 		printf("bio ");
755 	if (cpl == 0)
756 		printf("none");
757 	printf("\n");
758 
759 #ifdef KDB
760 	if (kdb_trap(&psl))
761 		return;
762 #endif
763 #ifdef DDB
764 	if (kdb_trap (type, 0, frame))
765 		return;
766 #endif
767 	printf("trap number		= %d\n", type);
768 	if (type <= MAX_TRAP_MSG)
769 		panic(trap_msg[type]);
770 	else
771 		panic("unknown/reserved trap");
772 }
773 
774 /*
775  * Double fault handler. Called when a fault occurs while writing
776  * a frame for a trap/exception onto the stack. This usually occurs
777  * when the stack overflows (such is the case with infinite recursion,
778  * for example).
779  *
780  * XXX Note that the current PTD gets replaced by IdlePTD when the
781  * task switch occurs. This means that the stack that was active at
782  * the time of the double fault is not available at <kstack> unless
783  * the machine was idle when the double fault occurred. The downside
784  * of this is that "trace <ebp>" in ddb won't work.
785  */
786 void
787 dblfault_handler()
788 {
789 	printf("\nFatal double fault:\n");
790 	printf("eip = 0x%x\n", common_tss.tss_eip);
791 	printf("esp = 0x%x\n", common_tss.tss_esp);
792 	printf("ebp = 0x%x\n", common_tss.tss_ebp);
793 #ifdef SMP
794 	printf("cpuid = %d\n", cpuid);
795 #endif
796 	panic("double fault");
797 }
798 
799 /*
800  * Compensate for 386 brain damage (missing URKR).
801  * This is a little simpler than the pagefault handler in trap() because
802  * it the page tables have already been faulted in and high addresses
803  * are thrown out early for other reasons.
804  */
805 int trapwrite(addr)
806 	unsigned addr;
807 {
808 	struct proc *p;
809 	vm_offset_t va;
810 	struct vmspace *vm;
811 	int rv;
812 
813 	va = trunc_page((vm_offset_t)addr);
814 	/*
815 	 * XXX - MAX is END.  Changed > to >= for temp. fix.
816 	 */
817 	if (va >= VM_MAXUSER_ADDRESS)
818 		return (1);
819 
820 	p = curproc;
821 	vm = p->p_vmspace;
822 
823 	++p->p_lock;
824 
825 	if ((caddr_t)va >= vm->vm_maxsaddr
826 	    && (caddr_t)va < (caddr_t)USRSTACK) {
827 		if (!grow(p, va)) {
828 			--p->p_lock;
829 			return (1);
830 		}
831 	}
832 
833 	/*
834 	 * fault the data page
835 	 */
836 	rv = vm_fault(&vm->vm_map, va, VM_PROT_READ|VM_PROT_WRITE, VM_FAULT_DIRTY);
837 
838 	--p->p_lock;
839 
840 	if (rv != KERN_SUCCESS)
841 		return 1;
842 
843 	return (0);
844 }
845 
846 /*
847  * System call request from POSIX system call gate interface to kernel.
848  * Like trap(), argument is call by reference.
849  */
850 void
851 syscall(frame)
852 	struct trapframe frame;
853 {
854 	caddr_t params;
855 	int i;
856 	struct sysent *callp;
857 	struct proc *p = curproc;
858 	u_quad_t sticks;
859 	int error;
860 	int args[8], rval[2];
861 	u_int code;
862 
863 	sticks = p->p_sticks;
864 	if (ISPL(frame.tf_cs) != SEL_UPL)
865 		panic("syscall");
866 
867 	p->p_md.md_regs = &frame;
868 	params = (caddr_t)frame.tf_esp + sizeof(int);
869 	code = frame.tf_eax;
870 	if (p->p_sysent->sv_prepsyscall) {
871 		(*p->p_sysent->sv_prepsyscall)(&frame, args, &code, &params);
872 	} else {
873 		/*
874 		 * Need to check if this is a 32 bit or 64 bit syscall.
875 		 */
876 		if (code == SYS_syscall) {
877 			/*
878 			 * Code is first argument, followed by actual args.
879 			 */
880 			code = fuword(params);
881 			params += sizeof(int);
882 		} else if (code == SYS___syscall) {
883 			/*
884 			 * Like syscall, but code is a quad, so as to maintain
885 			 * quad alignment for the rest of the arguments.
886 			 */
887 			code = fuword(params);
888 			params += sizeof(quad_t);
889 		}
890 	}
891 
892  	if (p->p_sysent->sv_mask)
893  		code &= p->p_sysent->sv_mask;
894 
895  	if (code >= p->p_sysent->sv_size)
896  		callp = &p->p_sysent->sv_table[0];
897   	else
898  		callp = &p->p_sysent->sv_table[code];
899 
900 	if (params && (i = callp->sy_narg * sizeof(int)) &&
901 	    (error = copyin(params, (caddr_t)args, (u_int)i))) {
902 #ifdef KTRACE
903 		if (KTRPOINT(p, KTR_SYSCALL))
904 			ktrsyscall(p->p_tracep, code, callp->sy_narg, args);
905 #endif
906 		goto bad;
907 	}
908 #ifdef KTRACE
909 	if (KTRPOINT(p, KTR_SYSCALL))
910 		ktrsyscall(p->p_tracep, code, callp->sy_narg, args);
911 #endif
912 	rval[0] = 0;
913 	rval[1] = frame.tf_edx;
914 
915 	error = (*callp->sy_call)(p, args, rval);
916 
917 	switch (error) {
918 
919 	case 0:
920 		/*
921 		 * Reinitialize proc pointer `p' as it may be different
922 		 * if this is a child returning from fork syscall.
923 		 */
924 		p = curproc;
925 		frame.tf_eax = rval[0];
926 		frame.tf_edx = rval[1];
927 		frame.tf_eflags &= ~PSL_C;
928 		break;
929 
930 	case ERESTART:
931 		/*
932 		 * Reconstruct pc, assuming lcall $X,y is 7 bytes,
933 		 * int 0x80 is 2 bytes. We saved this in tf_err.
934 		 */
935 		frame.tf_eip -= frame.tf_err;
936 		break;
937 
938 	case EJUSTRETURN:
939 		break;
940 
941 	default:
942 bad:
943  		if (p->p_sysent->sv_errsize)
944  			if (error >= p->p_sysent->sv_errsize)
945   				error = -1;	/* XXX */
946    			else
947   				error = p->p_sysent->sv_errtbl[error];
948 		frame.tf_eax = error;
949 		frame.tf_eflags |= PSL_C;
950 		break;
951 	}
952 
953 	if (frame.tf_eflags & PSL_T) {
954 		/* Traced syscall. */
955 		frame.tf_eflags &= ~PSL_T;
956 		trapsignal(p, SIGTRAP, 0);
957 	}
958 
959 	userret(p, &frame, sticks);
960 
961 #ifdef KTRACE
962 	if (KTRPOINT(p, KTR_SYSRET))
963 		ktrsysret(p->p_tracep, code, error, rval[0]);
964 #endif
965 }
966 
967 /*
968  * Simplified back end of syscall(), used when returning from fork()
969  * directly into user mode.
970  */
971 void
972 fork_return(p, frame)
973 	struct proc *p;
974 	struct trapframe frame;
975 {
976 	frame.tf_eax = 0;		/* Child returns zero */
977 	frame.tf_eflags &= ~PSL_C;	/* success */
978 	frame.tf_edx = 1;
979 
980 	userret(p, &frame, 0);
981 #ifdef KTRACE
982 	if (KTRPOINT(p, KTR_SYSRET))
983 		ktrsysret(p->p_tracep, SYS_fork, 0, 0);
984 #endif
985 }
986