xref: /freebsd/sys/kern/subr_trap.c (revision 5129159789cc9d7bc514e4546b88e3427695002d)
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/systm.h>
53 #include <sys/proc.h>
54 #include <sys/pioctl.h>
55 #include <sys/kernel.h>
56 #include <sys/resourcevar.h>
57 #include <sys/signalvar.h>
58 #include <sys/syscall.h>
59 #include <sys/sysent.h>
60 #include <sys/uio.h>
61 #include <sys/vmmeter.h>
62 #ifdef KTRACE
63 #include <sys/ktrace.h>
64 #endif
65 
66 #include <vm/vm.h>
67 #include <vm/vm_param.h>
68 #include <sys/lock.h>
69 #include <vm/pmap.h>
70 #include <vm/vm_kern.h>
71 #include <vm/vm_map.h>
72 #include <vm/vm_page.h>
73 #include <vm/vm_extern.h>
74 
75 #include <machine/cpu.h>
76 #include <machine/ipl.h>
77 #include <machine/md_var.h>
78 #include <machine/pcb.h>
79 #ifdef SMP
80 #include <machine/smp.h>
81 #endif
82 #include <machine/tss.h>
83 
84 #include <i386/isa/intr_machdep.h>
85 
86 #ifdef POWERFAIL_NMI
87 #include <sys/syslog.h>
88 #include <machine/clock.h>
89 #endif
90 
91 #include <machine/vm86.h>
92 
93 #ifdef DDB
94 	extern int in_Debugger, debugger_on_panic;
95 #endif
96 
97 #include "isa.h"
98 #include "npx.h"
99 
100 int (*pmath_emulate) __P((struct trapframe *));
101 
102 extern void trap __P((struct trapframe frame));
103 extern int trapwrite __P((unsigned addr));
104 extern void syscall __P((struct trapframe frame));
105 
106 static int trap_pfault __P((struct trapframe *, int, vm_offset_t));
107 static void trap_fatal __P((struct trapframe *, vm_offset_t));
108 void dblfault_handler __P((void));
109 
110 extern inthand_t IDTVEC(syscall);
111 
112 #define MAX_TRAP_MSG		28
113 static char *trap_msg[] = {
114 	"",					/*  0 unused */
115 	"privileged instruction fault",		/*  1 T_PRIVINFLT */
116 	"",					/*  2 unused */
117 	"breakpoint instruction fault",		/*  3 T_BPTFLT */
118 	"",					/*  4 unused */
119 	"",					/*  5 unused */
120 	"arithmetic trap",			/*  6 T_ARITHTRAP */
121 	"system forced exception",		/*  7 T_ASTFLT */
122 	"",					/*  8 unused */
123 	"general protection fault",		/*  9 T_PROTFLT */
124 	"trace trap",				/* 10 T_TRCTRAP */
125 	"",					/* 11 unused */
126 	"page fault",				/* 12 T_PAGEFLT */
127 	"",					/* 13 unused */
128 	"alignment fault",			/* 14 T_ALIGNFLT */
129 	"",					/* 15 unused */
130 	"",					/* 16 unused */
131 	"",					/* 17 unused */
132 	"integer divide fault",			/* 18 T_DIVIDE */
133 	"non-maskable interrupt trap",		/* 19 T_NMI */
134 	"overflow trap",			/* 20 T_OFLOW */
135 	"FPU bounds check fault",		/* 21 T_BOUND */
136 	"FPU device not available",		/* 22 T_DNA */
137 	"double fault",				/* 23 T_DOUBLEFLT */
138 	"FPU operand fetch fault",		/* 24 T_FPOPFLT */
139 	"invalid TSS fault",			/* 25 T_TSSFLT */
140 	"segment not present fault",		/* 26 T_SEGNPFLT */
141 	"stack fault",				/* 27 T_STKFLT */
142 	"machine check trap",			/* 28 T_MCHK */
143 };
144 
145 static __inline void userret __P((struct proc *p, struct trapframe *frame,
146 				  u_quad_t oticks));
147 
148 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
149 extern int has_f00f_bug;
150 #endif
151 
152 static __inline void
153 userret(p, frame, oticks)
154 	struct proc *p;
155 	struct trapframe *frame;
156 	u_quad_t oticks;
157 {
158 	int sig, s;
159 
160 	while ((sig = CURSIG(p)) != 0)
161 		postsig(sig);
162 
163 #if 0
164 	if (!want_resched &&
165 		(p->p_priority <= p->p_usrpri) &&
166 		(p->p_rtprio.type == RTP_PRIO_NORMAL)) {
167 		 int newpriority;
168 		 p->p_estcpu += 1;
169 		 newpriority = PUSER + p->p_estcpu / 4 + 2 * p->p_nice;
170 		 newpriority = min(newpriority, MAXPRI);
171 		 p->p_usrpri = newpriority;
172 	}
173 #endif
174 
175 	p->p_priority = p->p_usrpri;
176 	if (want_resched) {
177 		/*
178 		 * Since we are curproc, clock will normally just change
179 		 * our priority without moving us from one queue to another
180 		 * (since the running process is not on a queue.)
181 		 * If that happened after we setrunqueue ourselves but before we
182 		 * mi_switch()'ed, we might not be on the queue indicated by
183 		 * our priority.
184 		 */
185 		s = splhigh();
186 		setrunqueue(p);
187 		p->p_stats->p_ru.ru_nivcsw++;
188 		mi_switch();
189 		splx(s);
190 		while ((sig = CURSIG(p)) != 0)
191 			postsig(sig);
192 	}
193 	/*
194 	 * Charge system time if profiling.
195 	 */
196 	if (p->p_flag & P_PROFIL)
197 		addupc_task(p, frame->tf_eip,
198 			    (u_int)(p->p_sticks - oticks) * psratio);
199 
200 	curpriority = p->p_priority;
201 }
202 
203 /*
204  * Exception, fault, and trap interface to the FreeBSD kernel.
205  * This common code is called from assembly language IDT gate entry
206  * routines that prepare a suitable stack frame, and restore this
207  * frame after the exception has been processed.
208  */
209 
210 void
211 trap(frame)
212 	struct trapframe frame;
213 {
214 	struct proc *p = curproc;
215 	u_quad_t sticks = 0;
216 	int i = 0, ucode = 0, type, code;
217 	vm_offset_t eva;
218 
219 	if (!(frame.tf_eflags & PSL_I)) {
220 		/*
221 		 * Buggy application or kernel code has disabled interrupts
222 		 * and then trapped.  Enabling interrupts now is wrong, but
223 		 * it is better than running with interrupts disabled until
224 		 * they are accidentally enabled later.
225 		 */
226 		type = frame.tf_trapno;
227 		if (ISPL(frame.tf_cs) == SEL_UPL || (frame.tf_eflags & PSL_VM))
228 			printf(
229 			    "pid %ld (%s): trap %d with interrupts disabled\n",
230 			    (long)curproc->p_pid, curproc->p_comm, type);
231 		else if (type != T_BPTFLT && type != T_TRCTRAP)
232 			/*
233 			 * XXX not quite right, since this may be for a
234 			 * multiple fault in user mode.
235 			 */
236 			printf("kernel trap %d with interrupts disabled\n",
237 			    type);
238 		enable_intr();
239 	}
240 
241 	eva = 0;
242 	if (frame.tf_trapno == T_PAGEFLT) {
243 		/*
244 		 * For some Cyrix CPUs, %cr2 is clobbered by interrupts.
245 		 * This problem is worked around by using an interrupt
246 		 * gate for the pagefault handler.  We are finally ready
247 		 * to read %cr2 and then must reenable interrupts.
248 		 *
249 		 * XXX this should be in the switch statement, but the
250 		 * NO_FOOF_HACK and VM86 goto and ifdefs obfuscate the
251 		 * flow of control too much for this to be obviously
252 		 * correct.
253 		 */
254 		eva = rcr2();
255 		enable_intr();
256 	}
257 
258 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
259 restart:
260 #endif
261 	type = frame.tf_trapno;
262 	code = frame.tf_err;
263 
264 	if (in_vm86call) {
265 		if (frame.tf_eflags & PSL_VM &&
266 		    (type == T_PROTFLT || type == T_STKFLT)) {
267 			i = vm86_emulate((struct vm86frame *)&frame);
268 			if (i != 0)
269 				/*
270 				 * returns to original process
271 				 */
272 				vm86_trap((struct vm86frame *)&frame);
273 			return;
274 		}
275 		switch (type) {
276 			/*
277 			 * these traps want either a process context, or
278 			 * assume a normal userspace trap.
279 			 */
280 		case T_PROTFLT:
281 		case T_SEGNPFLT:
282 			trap_fatal(&frame, eva);
283 			return;
284 		case T_TRCTRAP:
285 			type = T_BPTFLT;	/* kernel breakpoint */
286 			/* FALL THROUGH */
287 		}
288 		goto kernel_trap;	/* normal kernel trap handling */
289 	}
290 
291         if ((ISPL(frame.tf_cs) == SEL_UPL) || (frame.tf_eflags & PSL_VM)) {
292 		/* user trap */
293 
294 		sticks = p->p_sticks;
295 		p->p_md.md_regs = &frame;
296 
297 		switch (type) {
298 		case T_PRIVINFLT:	/* privileged instruction fault */
299 			ucode = type;
300 			i = SIGILL;
301 			break;
302 
303 		case T_BPTFLT:		/* bpt instruction fault */
304 		case T_TRCTRAP:		/* trace trap */
305 			frame.tf_eflags &= ~PSL_T;
306 			i = SIGTRAP;
307 			break;
308 
309 		case T_ARITHTRAP:	/* arithmetic trap */
310 			ucode = code;
311 			i = SIGFPE;
312 			break;
313 
314 		case T_ASTFLT:		/* Allow process switch */
315 			astoff();
316 			cnt.v_soft++;
317 			if (p->p_flag & P_OWEUPC) {
318 				p->p_flag &= ~P_OWEUPC;
319 				addupc_task(p, p->p_stats->p_prof.pr_addr,
320 					    p->p_stats->p_prof.pr_ticks);
321 			}
322 			goto out;
323 
324 			/*
325 			 * The following two traps can happen in
326 			 * vm86 mode, and, if so, we want to handle
327 			 * them specially.
328 			 */
329 		case T_PROTFLT:		/* general protection fault */
330 		case T_STKFLT:		/* stack fault */
331 			if (frame.tf_eflags & PSL_VM) {
332 				i = vm86_emulate((struct vm86frame *)&frame);
333 				if (i == 0)
334 					goto out;
335 				break;
336 			}
337 			/* FALL THROUGH */
338 
339 		case T_SEGNPFLT:	/* segment not present fault */
340 		case T_TSSFLT:		/* invalid TSS fault */
341 		case T_DOUBLEFLT:	/* double fault */
342 		default:
343 			ucode = code + BUS_SEGM_FAULT ;
344 			i = SIGBUS;
345 			break;
346 
347 		case T_PAGEFLT:		/* page fault */
348 			i = trap_pfault(&frame, TRUE, eva);
349 			if (i == -1)
350 				return;
351 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
352 			if (i == -2)
353 				goto restart;
354 #endif
355 			if (i == 0)
356 				goto out;
357 
358 			ucode = T_PAGEFLT;
359 			break;
360 
361 		case T_DIVIDE:		/* integer divide fault */
362 			ucode = FPE_INTDIV;
363 			i = SIGFPE;
364 			break;
365 
366 #if NISA > 0
367 		case T_NMI:
368 #ifdef POWERFAIL_NMI
369 			goto handle_powerfail;
370 #else /* !POWERFAIL_NMI */
371 #ifdef DDB
372 			/* NMI can be hooked up to a pushbutton for debugging */
373 			printf ("NMI ... going to debugger\n");
374 			if (kdb_trap (type, 0, &frame))
375 				return;
376 #endif /* DDB */
377 			/* machine/parity/power fail/"kitchen sink" faults */
378 			if (isa_nmi(code) == 0) return;
379 			panic("NMI indicates hardware failure");
380 #endif /* POWERFAIL_NMI */
381 #endif /* NISA > 0 */
382 
383 		case T_OFLOW:		/* integer overflow fault */
384 			ucode = FPE_INTOVF;
385 			i = SIGFPE;
386 			break;
387 
388 		case T_BOUND:		/* bounds check fault */
389 			ucode = FPE_FLTSUB;
390 			i = SIGFPE;
391 			break;
392 
393 		case T_DNA:
394 #if NNPX > 0
395 			/* if a transparent fault (due to context switch "late") */
396 			if (npxdna())
397 				return;
398 #endif
399 			if (!pmath_emulate) {
400 				i = SIGFPE;
401 				ucode = FPE_FPU_NP_TRAP;
402 				break;
403 			}
404 			i = (*pmath_emulate)(&frame);
405 			if (i == 0) {
406 				if (!(frame.tf_eflags & PSL_T))
407 					return;
408 				frame.tf_eflags &= ~PSL_T;
409 				i = SIGTRAP;
410 			}
411 			/* else ucode = emulator_only_knows() XXX */
412 			break;
413 
414 		case T_FPOPFLT:		/* FPU operand fetch fault */
415 			ucode = T_FPOPFLT;
416 			i = SIGILL;
417 			break;
418 		}
419 	} else {
420 kernel_trap:
421 		/* kernel trap */
422 
423 		switch (type) {
424 		case T_PAGEFLT:			/* page fault */
425 			(void) trap_pfault(&frame, FALSE, eva);
426 			return;
427 
428 		case T_DNA:
429 #if NNPX > 0
430 			/*
431 			 * The kernel is apparently using npx for copying.
432 			 * XXX this should be fatal unless the kernel has
433 			 * registered such use.
434 			 */
435 			if (npxdna())
436 				return;
437 #endif
438 			break;
439 
440 		case T_PROTFLT:		/* general protection fault */
441 		case T_SEGNPFLT:	/* segment not present fault */
442 			/*
443 			 * Invalid segment selectors and out of bounds
444 			 * %eip's and %esp's can be set up in user mode.
445 			 * This causes a fault in kernel mode when the
446 			 * kernel tries to return to user mode.  We want
447 			 * to get this fault so that we can fix the
448 			 * problem here and not have to check all the
449 			 * selectors and pointers when the user changes
450 			 * them.
451 			 */
452 #define	MAYBE_DORETI_FAULT(where, whereto)				\
453 	do {								\
454 		if (frame.tf_eip == (int)where) {			\
455 			frame.tf_eip = (int)whereto;			\
456 			return;						\
457 		}							\
458 	} while (0)
459 
460 			if (intr_nesting_level == 0) {
461 				/*
462 				 * Invalid %fs's and %gs's can be created using
463 				 * procfs or PT_SETREGS or by invalidating the
464 				 * underlying LDT entry.  This causes a fault
465 				 * in kernel mode when the kernel attempts to
466 				 * switch contexts.  Lose the bad context
467 				 * (XXX) so that we can continue, and generate
468 				 * a signal.
469 				 */
470 				if (frame.tf_eip == (int)cpu_switch_load_gs) {
471 					curpcb->pcb_gs = 0;
472 					psignal(p, SIGBUS);
473 					return;
474 				}
475 				MAYBE_DORETI_FAULT(doreti_iret,
476 						   doreti_iret_fault);
477 				MAYBE_DORETI_FAULT(doreti_popl_ds,
478 						   doreti_popl_ds_fault);
479 				MAYBE_DORETI_FAULT(doreti_popl_es,
480 						   doreti_popl_es_fault);
481 				MAYBE_DORETI_FAULT(doreti_popl_fs,
482 						   doreti_popl_fs_fault);
483 				if (curpcb && curpcb->pcb_onfault) {
484 					frame.tf_eip = (int)curpcb->pcb_onfault;
485 					return;
486 				}
487 			}
488 			break;
489 
490 		case T_TSSFLT:
491 			/*
492 			 * PSL_NT can be set in user mode and isn't cleared
493 			 * automatically when the kernel is entered.  This
494 			 * causes a TSS fault when the kernel attempts to
495 			 * `iret' because the TSS link is uninitialized.  We
496 			 * want to get this fault so that we can fix the
497 			 * problem here and not every time the kernel is
498 			 * entered.
499 			 */
500 			if (frame.tf_eflags & PSL_NT) {
501 				frame.tf_eflags &= ~PSL_NT;
502 				return;
503 			}
504 			break;
505 
506 		case T_TRCTRAP:	 /* trace trap */
507 			if (frame.tf_eip == (int)IDTVEC(syscall)) {
508 				/*
509 				 * We've just entered system mode via the
510 				 * syscall lcall.  Continue single stepping
511 				 * silently until the syscall handler has
512 				 * saved the flags.
513 				 */
514 				return;
515 			}
516 			if (frame.tf_eip == (int)IDTVEC(syscall) + 1) {
517 				/*
518 				 * The syscall handler has now saved the
519 				 * flags.  Stop single stepping it.
520 				 */
521 				frame.tf_eflags &= ~PSL_T;
522 				return;
523 			}
524 			/*
525 			 * Fall through.
526 			 */
527 		case T_BPTFLT:
528 			/*
529 			 * If DDB is enabled, let it handle the debugger trap.
530 			 * Otherwise, debugger traps "can't happen".
531 			 */
532 #ifdef DDB
533 			if (kdb_trap (type, 0, &frame))
534 				return;
535 #endif
536 			break;
537 
538 #if NISA > 0
539 		case T_NMI:
540 #ifdef POWERFAIL_NMI
541 #ifndef TIMER_FREQ
542 #  define TIMER_FREQ 1193182
543 #endif
544 	handle_powerfail:
545 		{
546 		  static unsigned lastalert = 0;
547 
548 		  if(time_second - lastalert > 10)
549 		    {
550 		      log(LOG_WARNING, "NMI: power fail\n");
551 		      sysbeep(TIMER_FREQ/880, hz);
552 		      lastalert = time_second;
553 		    }
554 		  return;
555 		}
556 #else /* !POWERFAIL_NMI */
557 #ifdef DDB
558 			/* NMI can be hooked up to a pushbutton for debugging */
559 			printf ("NMI ... going to debugger\n");
560 			if (kdb_trap (type, 0, &frame))
561 				return;
562 #endif /* DDB */
563 			/* machine/parity/power fail/"kitchen sink" faults */
564 			if (isa_nmi(code) == 0) return;
565 			/* FALL THROUGH */
566 #endif /* POWERFAIL_NMI */
567 #endif /* NISA > 0 */
568 		}
569 
570 		trap_fatal(&frame, eva);
571 		return;
572 	}
573 
574 	/* Translate fault for emulators (e.g. Linux) */
575 	if (*p->p_sysent->sv_transtrap)
576 		i = (*p->p_sysent->sv_transtrap)(i, type);
577 
578 	trapsignal(p, i, ucode);
579 
580 #ifdef DEBUG
581 	if (type <= MAX_TRAP_MSG) {
582 		uprintf("fatal process exception: %s",
583 			trap_msg[type]);
584 		if ((type == T_PAGEFLT) || (type == T_PROTFLT))
585 			uprintf(", fault VA = 0x%lx", (u_long)eva);
586 		uprintf("\n");
587 	}
588 #endif
589 
590 out:
591 	userret(p, &frame, sticks);
592 }
593 
594 #ifdef notyet
595 /*
596  * This version doesn't allow a page fault to user space while
597  * in the kernel. The rest of the kernel needs to be made "safe"
598  * before this can be used. I think the only things remaining
599  * to be made safe are the iBCS2 code and the process tracing/
600  * debugging code.
601  */
602 static int
603 trap_pfault(frame, usermode, eva)
604 	struct trapframe *frame;
605 	int usermode;
606 	vm_offset_t eva;
607 {
608 	vm_offset_t va;
609 	struct vmspace *vm = NULL;
610 	vm_map_t map = 0;
611 	int rv = 0;
612 	vm_prot_t ftype;
613 	struct proc *p = curproc;
614 
615 	if (frame->tf_err & PGEX_W)
616 		ftype = VM_PROT_READ | VM_PROT_WRITE;
617 	else
618 		ftype = VM_PROT_READ;
619 
620 	va = trunc_page(eva);
621 	if (va < VM_MIN_KERNEL_ADDRESS) {
622 		vm_offset_t v;
623 		vm_page_t mpte;
624 
625 		if (p == NULL ||
626 		    (!usermode && va < VM_MAXUSER_ADDRESS &&
627 		     (intr_nesting_level != 0 || curpcb == NULL ||
628 		      curpcb->pcb_onfault == NULL))) {
629 			trap_fatal(frame, eva);
630 			return (-1);
631 		}
632 
633 		/*
634 		 * This is a fault on non-kernel virtual memory.
635 		 * vm is initialized above to NULL. If curproc is NULL
636 		 * or curproc->p_vmspace is NULL the fault is fatal.
637 		 */
638 		vm = p->p_vmspace;
639 		if (vm == NULL)
640 			goto nogo;
641 
642 		map = &vm->vm_map;
643 
644 		/*
645 		 * Keep swapout from messing with us during this
646 		 *	critical time.
647 		 */
648 		++p->p_lock;
649 
650 		/*
651 		 * Grow the stack if necessary
652 		 */
653 		/* grow_stack returns false only if va falls into
654 		 * a growable stack region and the stack growth
655 		 * fails.  It returns true if va was not within
656 		 * a growable stack region, or if the stack
657 		 * growth succeeded.
658 		 */
659 		if (!grow_stack (p, va)) {
660 			rv = KERN_FAILURE;
661 			--p->p_lock;
662 			goto nogo;
663 		}
664 
665 		/* Fault in the user page: */
666 		rv = vm_fault(map, va, ftype,
667 			      (ftype & VM_PROT_WRITE) ? VM_FAULT_DIRTY
668 						      : VM_FAULT_NORMAL);
669 
670 		--p->p_lock;
671 	} else {
672 		/*
673 		 * Don't allow user-mode faults in kernel address space.
674 		 */
675 		if (usermode)
676 			goto nogo;
677 
678 		/*
679 		 * Since we know that kernel virtual address addresses
680 		 * always have pte pages mapped, we just have to fault
681 		 * the page.
682 		 */
683 		rv = vm_fault(kernel_map, va, ftype, VM_FAULT_NORMAL);
684 	}
685 
686 	if (rv == KERN_SUCCESS)
687 		return (0);
688 nogo:
689 	if (!usermode) {
690 		if (intr_nesting_level == 0 && curpcb && curpcb->pcb_onfault) {
691 			frame->tf_eip = (int)curpcb->pcb_onfault;
692 			return (0);
693 		}
694 		trap_fatal(frame, eva);
695 		return (-1);
696 	}
697 
698 	/* kludge to pass faulting virtual address to sendsig */
699 	frame->tf_err = eva;
700 
701 	return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
702 }
703 #endif
704 
705 int
706 trap_pfault(frame, usermode, eva)
707 	struct trapframe *frame;
708 	int usermode;
709 	vm_offset_t eva;
710 {
711 	vm_offset_t va;
712 	struct vmspace *vm = NULL;
713 	vm_map_t map = 0;
714 	int rv = 0;
715 	vm_prot_t ftype;
716 	struct proc *p = curproc;
717 
718 	va = trunc_page(eva);
719 	if (va >= KERNBASE) {
720 		/*
721 		 * Don't allow user-mode faults in kernel address space.
722 		 * An exception:  if the faulting address is the invalid
723 		 * instruction entry in the IDT, then the Intel Pentium
724 		 * F00F bug workaround was triggered, and we need to
725 		 * treat it is as an illegal instruction, and not a page
726 		 * fault.
727 		 */
728 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
729 		if ((eva == (unsigned int)&idt[6]) && has_f00f_bug) {
730 			frame->tf_trapno = T_PRIVINFLT;
731 			return -2;
732 		}
733 #endif
734 		if (usermode)
735 			goto nogo;
736 
737 		map = kernel_map;
738 	} else {
739 		/*
740 		 * This is a fault on non-kernel virtual memory.
741 		 * vm is initialized above to NULL. If curproc is NULL
742 		 * or curproc->p_vmspace is NULL the fault is fatal.
743 		 */
744 		if (p != NULL)
745 			vm = p->p_vmspace;
746 
747 		if (vm == NULL)
748 			goto nogo;
749 
750 		map = &vm->vm_map;
751 	}
752 
753 	if (frame->tf_err & PGEX_W)
754 		ftype = VM_PROT_READ | VM_PROT_WRITE;
755 	else
756 		ftype = VM_PROT_READ;
757 
758 	if (map != kernel_map) {
759 		/*
760 		 * Keep swapout from messing with us during this
761 		 *	critical time.
762 		 */
763 		++p->p_lock;
764 
765 		/*
766 		 * Grow the stack if necessary
767 		 */
768 		/* grow_stack returns false only if va falls into
769 		 * a growable stack region and the stack growth
770 		 * fails.  It returns true if va was not within
771 		 * a growable stack region, or if the stack
772 		 * growth succeeded.
773 		 */
774 		if (!grow_stack (p, va)) {
775 			rv = KERN_FAILURE;
776 			--p->p_lock;
777 			goto nogo;
778 		}
779 
780 		/* Fault in the user page: */
781 		rv = vm_fault(map, va, ftype,
782 			      (ftype & VM_PROT_WRITE) ? VM_FAULT_DIRTY
783 						      : VM_FAULT_NORMAL);
784 
785 		--p->p_lock;
786 	} else {
787 		/*
788 		 * Don't have to worry about process locking or stacks in the kernel.
789 		 */
790 		rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
791 	}
792 
793 	if (rv == KERN_SUCCESS)
794 		return (0);
795 nogo:
796 	if (!usermode) {
797 		if (intr_nesting_level == 0 && curpcb && curpcb->pcb_onfault) {
798 			frame->tf_eip = (int)curpcb->pcb_onfault;
799 			return (0);
800 		}
801 		trap_fatal(frame, eva);
802 		return (-1);
803 	}
804 
805 	/* kludge to pass faulting virtual address to sendsig */
806 	frame->tf_err = eva;
807 
808 	return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
809 }
810 
811 static void
812 trap_fatal(frame, eva)
813 	struct trapframe *frame;
814 	vm_offset_t eva;
815 {
816 	int code, type, ss, esp;
817 	struct soft_segment_descriptor softseg;
818 
819 	code = frame->tf_err;
820 	type = frame->tf_trapno;
821 	sdtossd(&gdt[IDXSEL(frame->tf_cs & 0xffff)].sd, &softseg);
822 
823 	if (type <= MAX_TRAP_MSG)
824 		printf("\n\nFatal trap %d: %s while in %s mode\n",
825 			type, trap_msg[type],
826         		frame->tf_eflags & PSL_VM ? "vm86" :
827 			ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel");
828 #ifdef SMP
829 	/* three seperate prints in case of a trap on an unmapped page */
830 	printf("mp_lock = %08x; ", mp_lock);
831 	printf("cpuid = %d; ", cpuid);
832 	printf("lapic.id = %08x\n", lapic.id);
833 #endif
834 	if (type == T_PAGEFLT) {
835 		printf("fault virtual address	= 0x%x\n", eva);
836 		printf("fault code		= %s %s, %s\n",
837 			code & PGEX_U ? "user" : "supervisor",
838 			code & PGEX_W ? "write" : "read",
839 			code & PGEX_P ? "protection violation" : "page not present");
840 	}
841 	printf("instruction pointer	= 0x%x:0x%x\n",
842 	       frame->tf_cs & 0xffff, frame->tf_eip);
843         if ((ISPL(frame->tf_cs) == SEL_UPL) || (frame->tf_eflags & PSL_VM)) {
844 		ss = frame->tf_ss & 0xffff;
845 		esp = frame->tf_esp;
846 	} else {
847 		ss = GSEL(GDATA_SEL, SEL_KPL);
848 		esp = (int)&frame->tf_esp;
849 	}
850 	printf("stack pointer	        = 0x%x:0x%x\n", ss, esp);
851 	printf("frame pointer	        = 0x%x:0x%x\n", ss, frame->tf_ebp);
852 	printf("code segment		= base 0x%x, limit 0x%x, type 0x%x\n",
853 	       softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type);
854 	printf("			= DPL %d, pres %d, def32 %d, gran %d\n",
855 	       softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_def32,
856 	       softseg.ssd_gran);
857 	printf("processor eflags	= ");
858 	if (frame->tf_eflags & PSL_T)
859 		printf("trace trap, ");
860 	if (frame->tf_eflags & PSL_I)
861 		printf("interrupt enabled, ");
862 	if (frame->tf_eflags & PSL_NT)
863 		printf("nested task, ");
864 	if (frame->tf_eflags & PSL_RF)
865 		printf("resume, ");
866 	if (frame->tf_eflags & PSL_VM)
867 		printf("vm86, ");
868 	printf("IOPL = %d\n", (frame->tf_eflags & PSL_IOPL) >> 12);
869 	printf("current process		= ");
870 	if (curproc) {
871 		printf("%lu (%s)\n",
872 		    (u_long)curproc->p_pid, curproc->p_comm ?
873 		    curproc->p_comm : "");
874 	} else {
875 		printf("Idle\n");
876 	}
877 	printf("interrupt mask		= ");
878 	if ((cpl & net_imask) == net_imask)
879 		printf("net ");
880 	if ((cpl & tty_imask) == tty_imask)
881 		printf("tty ");
882 	if ((cpl & bio_imask) == bio_imask)
883 		printf("bio ");
884 	if ((cpl & cam_imask) == cam_imask)
885 		printf("cam ");
886 	if (cpl == 0)
887 		printf("none");
888 #ifdef SMP
889 /**
890  *  XXX FIXME:
891  *	we probably SHOULD have stopped the other CPUs before now!
892  *	another CPU COULD have been touching cpl at this moment...
893  */
894 	printf(" <- SMP: XXX");
895 #endif
896 	printf("\n");
897 
898 #ifdef KDB
899 	if (kdb_trap(&psl))
900 		return;
901 #endif
902 #ifdef DDB
903 	if ((debugger_on_panic || in_Debugger) && kdb_trap(type, 0, frame))
904 		return;
905 #endif
906 	printf("trap number		= %d\n", type);
907 	if (type <= MAX_TRAP_MSG)
908 		panic(trap_msg[type]);
909 	else
910 		panic("unknown/reserved trap");
911 }
912 
913 /*
914  * Double fault handler. Called when a fault occurs while writing
915  * a frame for a trap/exception onto the stack. This usually occurs
916  * when the stack overflows (such is the case with infinite recursion,
917  * for example).
918  *
919  * XXX Note that the current PTD gets replaced by IdlePTD when the
920  * task switch occurs. This means that the stack that was active at
921  * the time of the double fault is not available at <kstack> unless
922  * the machine was idle when the double fault occurred. The downside
923  * of this is that "trace <ebp>" in ddb won't work.
924  */
925 void
926 dblfault_handler()
927 {
928 	printf("\nFatal double fault:\n");
929 	printf("eip = 0x%x\n", common_tss.tss_eip);
930 	printf("esp = 0x%x\n", common_tss.tss_esp);
931 	printf("ebp = 0x%x\n", common_tss.tss_ebp);
932 #ifdef SMP
933 	/* three seperate prints in case of a trap on an unmapped page */
934 	printf("mp_lock = %08x; ", mp_lock);
935 	printf("cpuid = %d; ", cpuid);
936 	printf("lapic.id = %08x\n", lapic.id);
937 #endif
938 	panic("double fault");
939 }
940 
941 /*
942  * Compensate for 386 brain damage (missing URKR).
943  * This is a little simpler than the pagefault handler in trap() because
944  * it the page tables have already been faulted in and high addresses
945  * are thrown out early for other reasons.
946  */
947 int trapwrite(addr)
948 	unsigned addr;
949 {
950 	struct proc *p;
951 	vm_offset_t va;
952 	struct vmspace *vm;
953 	int rv;
954 
955 	va = trunc_page((vm_offset_t)addr);
956 	/*
957 	 * XXX - MAX is END.  Changed > to >= for temp. fix.
958 	 */
959 	if (va >= VM_MAXUSER_ADDRESS)
960 		return (1);
961 
962 	p = curproc;
963 	vm = p->p_vmspace;
964 
965 	++p->p_lock;
966 
967 	if (!grow_stack (p, va)) {
968 		--p->p_lock;
969 		return (1);
970 	}
971 
972 	/*
973 	 * fault the data page
974 	 */
975 	rv = vm_fault(&vm->vm_map, va, VM_PROT_READ|VM_PROT_WRITE, VM_FAULT_DIRTY);
976 
977 	--p->p_lock;
978 
979 	if (rv != KERN_SUCCESS)
980 		return 1;
981 
982 	return (0);
983 }
984 
985 /*
986  * System call request from POSIX system call gate interface to kernel.
987  * Like trap(), argument is call by reference.
988  */
989 void
990 syscall(frame)
991 	struct trapframe frame;
992 {
993 	caddr_t params;
994 	int i;
995 	struct sysent *callp;
996 	struct proc *p = curproc;
997 	u_quad_t sticks;
998 	int error;
999 	int args[8];
1000 	u_int code;
1001 
1002 #ifdef DIAGNOSTIC
1003 	if (ISPL(frame.tf_cs) != SEL_UPL)
1004 		panic("syscall");
1005 #endif
1006 	sticks = p->p_sticks;
1007 	p->p_md.md_regs = &frame;
1008 	params = (caddr_t)frame.tf_esp + sizeof(int);
1009 	code = frame.tf_eax;
1010 	if (p->p_sysent->sv_prepsyscall) {
1011 		(*p->p_sysent->sv_prepsyscall)(&frame, args, &code, &params);
1012 	} else {
1013 		/*
1014 		 * Need to check if this is a 32 bit or 64 bit syscall.
1015 		 */
1016 		if (code == SYS_syscall) {
1017 			/*
1018 			 * Code is first argument, followed by actual args.
1019 			 */
1020 			code = fuword(params);
1021 			params += sizeof(int);
1022 		} else if (code == SYS___syscall) {
1023 			/*
1024 			 * Like syscall, but code is a quad, so as to maintain
1025 			 * quad alignment for the rest of the arguments.
1026 			 */
1027 			code = fuword(params);
1028 			params += sizeof(quad_t);
1029 		}
1030 	}
1031 
1032  	if (p->p_sysent->sv_mask)
1033  		code &= p->p_sysent->sv_mask;
1034 
1035  	if (code >= p->p_sysent->sv_size)
1036  		callp = &p->p_sysent->sv_table[0];
1037   	else
1038  		callp = &p->p_sysent->sv_table[code];
1039 
1040 	if (params && (i = callp->sy_narg * sizeof(int)) &&
1041 	    (error = copyin(params, (caddr_t)args, (u_int)i))) {
1042 #ifdef KTRACE
1043 		if (KTRPOINT(p, KTR_SYSCALL))
1044 			ktrsyscall(p->p_tracep, code, callp->sy_narg, args);
1045 #endif
1046 		goto bad;
1047 	}
1048 #ifdef KTRACE
1049 	if (KTRPOINT(p, KTR_SYSCALL))
1050 		ktrsyscall(p->p_tracep, code, callp->sy_narg, args);
1051 #endif
1052 	p->p_retval[0] = 0;
1053 	p->p_retval[1] = frame.tf_edx;
1054 
1055 	STOPEVENT(p, S_SCE, callp->sy_narg);
1056 
1057 	error = (*callp->sy_call)(p, args);
1058 
1059 	switch (error) {
1060 
1061 	case 0:
1062 		/*
1063 		 * Reinitialize proc pointer `p' as it may be different
1064 		 * if this is a child returning from fork syscall.
1065 		 */
1066 		p = curproc;
1067 		frame.tf_eax = p->p_retval[0];
1068 		frame.tf_edx = p->p_retval[1];
1069 		frame.tf_eflags &= ~PSL_C;
1070 		break;
1071 
1072 	case ERESTART:
1073 		/*
1074 		 * Reconstruct pc, assuming lcall $X,y is 7 bytes,
1075 		 * int 0x80 is 2 bytes. We saved this in tf_err.
1076 		 */
1077 		frame.tf_eip -= frame.tf_err;
1078 		break;
1079 
1080 	case EJUSTRETURN:
1081 		break;
1082 
1083 	default:
1084 bad:
1085  		if (p->p_sysent->sv_errsize) {
1086  			if (error >= p->p_sysent->sv_errsize)
1087   				error = -1;	/* XXX */
1088    			else
1089   				error = p->p_sysent->sv_errtbl[error];
1090 		}
1091 		frame.tf_eax = error;
1092 		frame.tf_eflags |= PSL_C;
1093 		break;
1094 	}
1095 
1096 	if ((frame.tf_eflags & PSL_T) && !(frame.tf_eflags & PSL_VM)) {
1097 		/* Traced syscall. */
1098 		frame.tf_eflags &= ~PSL_T;
1099 		trapsignal(p, SIGTRAP, 0);
1100 	}
1101 
1102 	userret(p, &frame, sticks);
1103 
1104 #ifdef KTRACE
1105 	if (KTRPOINT(p, KTR_SYSRET))
1106 		ktrsysret(p->p_tracep, code, error, p->p_retval[0]);
1107 #endif
1108 
1109 	/*
1110 	 * This works because errno is findable through the
1111 	 * register set.  If we ever support an emulation where this
1112 	 * is not the case, this code will need to be revisited.
1113 	 */
1114 	STOPEVENT(p, S_SCX, code);
1115 
1116 }
1117 
1118 /*
1119  * Simplified back end of syscall(), used when returning from fork()
1120  * directly into user mode.
1121  */
1122 void
1123 fork_return(p, frame)
1124 	struct proc *p;
1125 	struct trapframe frame;
1126 {
1127 	frame.tf_eax = 0;		/* Child returns zero */
1128 	frame.tf_eflags &= ~PSL_C;	/* success */
1129 	frame.tf_edx = 1;
1130 
1131 	userret(p, &frame, 0);
1132 #ifdef KTRACE
1133 	if (KTRPOINT(p, KTR_SYSRET))
1134 		ktrsysret(p->p_tracep, SYS_fork, 0, 0);
1135 #endif
1136 }
1137