xref: /freebsd/sys/kern/subr_trap.c (revision 6e8394b8baa7d5d9153ab90de6824bcd19b3b4e1)
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 1.138 1999/06/01 18:19:47 jlemon Exp $
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 <vm/vm_prot.h>
69 #include <sys/lock.h>
70 #include <vm/pmap.h>
71 #include <vm/vm_kern.h>
72 #include <vm/vm_map.h>
73 #include <vm/vm_page.h>
74 #include <vm/vm_extern.h>
75 
76 #include <machine/cpu.h>
77 #include <machine/ipl.h>
78 #include <machine/md_var.h>
79 #include <machine/pcb.h>
80 #ifdef SMP
81 #include <machine/smp.h>
82 #endif
83 #include <machine/tss.h>
84 
85 #include <i386/isa/intr_machdep.h>
86 
87 #ifdef POWERFAIL_NMI
88 #include <sys/syslog.h>
89 #include <machine/clock.h>
90 #endif
91 
92 #include <machine/vm86.h>
93 
94 #ifdef DDB
95 	extern int in_Debugger, debugger_on_panic;
96 #endif
97 
98 #include "isa.h"
99 #include "npx.h"
100 
101 int (*pmath_emulate) __P((struct trapframe *));
102 
103 extern void trap __P((struct trapframe frame));
104 extern int trapwrite __P((unsigned addr));
105 extern void syscall __P((struct trapframe frame));
106 
107 static int trap_pfault __P((struct trapframe *, int, vm_offset_t));
108 static void trap_fatal __P((struct trapframe *, vm_offset_t));
109 void dblfault_handler __P((void));
110 
111 extern inthand_t IDTVEC(syscall);
112 
113 #define MAX_TRAP_MSG		28
114 static char *trap_msg[] = {
115 	"",					/*  0 unused */
116 	"privileged instruction fault",		/*  1 T_PRIVINFLT */
117 	"",					/*  2 unused */
118 	"breakpoint instruction fault",		/*  3 T_BPTFLT */
119 	"",					/*  4 unused */
120 	"",					/*  5 unused */
121 	"arithmetic trap",			/*  6 T_ARITHTRAP */
122 	"system forced exception",		/*  7 T_ASTFLT */
123 	"",					/*  8 unused */
124 	"general protection fault",		/*  9 T_PROTFLT */
125 	"trace trap",				/* 10 T_TRCTRAP */
126 	"",					/* 11 unused */
127 	"page fault",				/* 12 T_PAGEFLT */
128 	"",					/* 13 unused */
129 	"alignment fault",			/* 14 T_ALIGNFLT */
130 	"",					/* 15 unused */
131 	"",					/* 16 unused */
132 	"",					/* 17 unused */
133 	"integer divide fault",			/* 18 T_DIVIDE */
134 	"non-maskable interrupt trap",		/* 19 T_NMI */
135 	"overflow trap",			/* 20 T_OFLOW */
136 	"FPU bounds check fault",		/* 21 T_BOUND */
137 	"FPU device not available",		/* 22 T_DNA */
138 	"double fault",				/* 23 T_DOUBLEFLT */
139 	"FPU operand fetch fault",		/* 24 T_FPOPFLT */
140 	"invalid TSS fault",			/* 25 T_TSSFLT */
141 	"segment not present fault",		/* 26 T_SEGNPFLT */
142 	"stack fault",				/* 27 T_STKFLT */
143 	"machine check trap",			/* 28 T_MCHK */
144 };
145 
146 static __inline void userret __P((struct proc *p, struct trapframe *frame,
147 				  u_quad_t oticks));
148 
149 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
150 extern int has_f00f_bug;
151 #endif
152 
153 static __inline void
154 userret(p, frame, oticks)
155 	struct proc *p;
156 	struct trapframe *frame;
157 	u_quad_t oticks;
158 {
159 	int sig, s;
160 
161 	while ((sig = CURSIG(p)) != 0)
162 		postsig(sig);
163 
164 #if 0
165 	if (!want_resched &&
166 		(p->p_priority <= p->p_usrpri) &&
167 		(p->p_rtprio.type == RTP_PRIO_NORMAL)) {
168 		 int newpriority;
169 		 p->p_estcpu += 1;
170 		 newpriority = PUSER + p->p_estcpu / 4 + 2 * p->p_nice;
171 		 newpriority = min(newpriority, MAXPRI);
172 		 p->p_usrpri = newpriority;
173 	}
174 #endif
175 
176 	p->p_priority = p->p_usrpri;
177 	if (want_resched) {
178 		/*
179 		 * Since we are curproc, clock will normally just change
180 		 * our priority without moving us from one queue to another
181 		 * (since the running process is not on a queue.)
182 		 * If that happened after we setrunqueue ourselves but before we
183 		 * mi_switch()'ed, we might not be on the queue indicated by
184 		 * our priority.
185 		 */
186 		s = splhigh();
187 		setrunqueue(p);
188 		p->p_stats->p_ru.ru_nivcsw++;
189 		mi_switch();
190 		splx(s);
191 		while ((sig = CURSIG(p)) != 0)
192 			postsig(sig);
193 	}
194 	/*
195 	 * Charge system time if profiling.
196 	 */
197 	if (p->p_flag & P_PROFIL)
198 		addupc_task(p, frame->tf_eip,
199 			    (u_int)(p->p_sticks - oticks) * psratio);
200 
201 	curpriority = p->p_priority;
202 }
203 
204 /*
205  * Exception, fault, and trap interface to the FreeBSD kernel.
206  * This common code is called from assembly language IDT gate entry
207  * routines that prepare a suitable stack frame, and restore this
208  * frame after the exception has been processed.
209  */
210 
211 void
212 trap(frame)
213 	struct trapframe frame;
214 {
215 	struct proc *p = curproc;
216 	u_quad_t sticks = 0;
217 	int i = 0, ucode = 0, type, code;
218 	vm_offset_t eva;
219 
220 	if (!(frame.tf_eflags & PSL_I)) {
221 		/*
222 		 * Buggy application or kernel code has disabled interrupts
223 		 * and then trapped.  Enabling interrupts now is wrong, but
224 		 * it is better than running with interrupts disabled until
225 		 * they are accidentally enabled later.
226 		 */
227 		type = frame.tf_trapno;
228 		if (ISPL(frame.tf_cs) == SEL_UPL || (frame.tf_eflags & PSL_VM))
229 			printf(
230 			    "pid %ld (%s): trap %d with interrupts disabled\n",
231 			    (long)curproc->p_pid, curproc->p_comm, type);
232 		else if (type != T_BPTFLT && type != T_TRCTRAP)
233 			/*
234 			 * XXX not quite right, since this may be for a
235 			 * multiple fault in user mode.
236 			 */
237 			printf("kernel trap %d with interrupts disabled\n",
238 			    type);
239 		enable_intr();
240 	}
241 
242 	eva = 0;
243 	if (frame.tf_trapno == T_PAGEFLT) {
244 		/*
245 		 * For some Cyrix CPUs, %cr2 is clobbered by interrupts.
246 		 * This problem is worked around by using an interrupt
247 		 * gate for the pagefault handler.  We are finally ready
248 		 * to read %cr2 and then must reenable interrupts.
249 		 *
250 		 * XXX this should be in the switch statement, but the
251 		 * NO_FOOF_HACK and VM86 goto and ifdefs obfuscate the
252 		 * flow of control too much for this to be obviously
253 		 * correct.
254 		 */
255 		eva = rcr2();
256 		enable_intr();
257 	}
258 
259 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
260 restart:
261 #endif
262 	type = frame.tf_trapno;
263 	code = frame.tf_err;
264 
265 	if (in_vm86call) {
266 		if (frame.tf_eflags & PSL_VM &&
267 		    (type == T_PROTFLT || type == T_STKFLT)) {
268 			i = vm86_emulate((struct vm86frame *)&frame);
269 			if (i != 0)
270 				/*
271 				 * returns to original process
272 				 */
273 				vm86_trap((struct vm86frame *)&frame);
274 			return;
275 		}
276 		switch (type) {
277 			/*
278 			 * these traps want either a process context, or
279 			 * assume a normal userspace trap.
280 			 */
281 		case T_PROTFLT:
282 		case T_SEGNPFLT:
283 			trap_fatal(&frame, eva);
284 			return;
285 		case T_TRCTRAP:
286 			type = T_BPTFLT;	/* kernel breakpoint */
287 			/* FALL THROUGH */
288 		}
289 		goto kernel_trap;	/* normal kernel trap handling */
290 	}
291 
292         if ((ISPL(frame.tf_cs) == SEL_UPL) || (frame.tf_eflags & PSL_VM)) {
293 		/* user trap */
294 
295 		sticks = p->p_sticks;
296 		p->p_md.md_regs = &frame;
297 
298 		switch (type) {
299 		case T_PRIVINFLT:	/* privileged instruction fault */
300 			ucode = type;
301 			i = SIGILL;
302 			break;
303 
304 		case T_BPTFLT:		/* bpt instruction fault */
305 		case T_TRCTRAP:		/* trace trap */
306 			frame.tf_eflags &= ~PSL_T;
307 			i = SIGTRAP;
308 			break;
309 
310 		case T_ARITHTRAP:	/* arithmetic trap */
311 			ucode = code;
312 			i = SIGFPE;
313 			break;
314 
315 		case T_ASTFLT:		/* Allow process switch */
316 			astoff();
317 			cnt.v_soft++;
318 			if (p->p_flag & P_OWEUPC) {
319 				p->p_flag &= ~P_OWEUPC;
320 				addupc_task(p, p->p_stats->p_prof.pr_addr,
321 					    p->p_stats->p_prof.pr_ticks);
322 			}
323 			goto out;
324 
325 			/*
326 			 * The following two traps can happen in
327 			 * vm86 mode, and, if so, we want to handle
328 			 * them specially.
329 			 */
330 		case T_PROTFLT:		/* general protection fault */
331 		case T_STKFLT:		/* stack fault */
332 			if (frame.tf_eflags & PSL_VM) {
333 				i = vm86_emulate((struct vm86frame *)&frame);
334 				if (i == 0)
335 					goto out;
336 				break;
337 			}
338 			/* FALL THROUGH */
339 
340 		case T_SEGNPFLT:	/* segment not present fault */
341 		case T_TSSFLT:		/* invalid TSS fault */
342 		case T_DOUBLEFLT:	/* double fault */
343 		default:
344 			ucode = code + BUS_SEGM_FAULT ;
345 			i = SIGBUS;
346 			break;
347 
348 		case T_PAGEFLT:		/* page fault */
349 			i = trap_pfault(&frame, TRUE, eva);
350 			if (i == -1)
351 				return;
352 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
353 			if (i == -2)
354 				goto restart;
355 #endif
356 			if (i == 0)
357 				goto out;
358 
359 			ucode = T_PAGEFLT;
360 			break;
361 
362 		case T_DIVIDE:		/* integer divide fault */
363 			ucode = FPE_INTDIV_TRAP;
364 			i = SIGFPE;
365 			break;
366 
367 #if NISA > 0
368 		case T_NMI:
369 #ifdef POWERFAIL_NMI
370 			goto handle_powerfail;
371 #else /* !POWERFAIL_NMI */
372 #ifdef DDB
373 			/* NMI can be hooked up to a pushbutton for debugging */
374 			printf ("NMI ... going to debugger\n");
375 			if (kdb_trap (type, 0, &frame))
376 				return;
377 #endif /* DDB */
378 			/* machine/parity/power fail/"kitchen sink" faults */
379 			if (isa_nmi(code) == 0) return;
380 			panic("NMI indicates hardware failure");
381 #endif /* POWERFAIL_NMI */
382 #endif /* NISA > 0 */
383 
384 		case T_OFLOW:		/* integer overflow fault */
385 			ucode = FPE_INTOVF_TRAP;
386 			i = SIGFPE;
387 			break;
388 
389 		case T_BOUND:		/* bounds check fault */
390 			ucode = FPE_SUBRNG_TRAP;
391 			i = SIGFPE;
392 			break;
393 
394 		case T_DNA:
395 #if NNPX > 0
396 			/* if a transparent fault (due to context switch "late") */
397 			if (npxdna())
398 				return;
399 #endif
400 			if (!pmath_emulate) {
401 				i = SIGFPE;
402 				ucode = FPE_FPU_NP_TRAP;
403 				break;
404 			}
405 			i = (*pmath_emulate)(&frame);
406 			if (i == 0) {
407 				if (!(frame.tf_eflags & PSL_T))
408 					return;
409 				frame.tf_eflags &= ~PSL_T;
410 				i = SIGTRAP;
411 			}
412 			/* else ucode = emulator_only_knows() XXX */
413 			break;
414 
415 		case T_FPOPFLT:		/* FPU operand fetch fault */
416 			ucode = T_FPOPFLT;
417 			i = SIGILL;
418 			break;
419 		}
420 	} else {
421 kernel_trap:
422 		/* kernel trap */
423 
424 		switch (type) {
425 		case T_PAGEFLT:			/* page fault */
426 			(void) trap_pfault(&frame, FALSE, eva);
427 			return;
428 
429 		case T_DNA:
430 #if NNPX > 0
431 			/*
432 			 * The kernel is apparently using npx for copying.
433 			 * XXX this should be fatal unless the kernel has
434 			 * registered such use.
435 			 */
436 			if (npxdna())
437 				return;
438 #endif
439 			break;
440 
441 		case T_PROTFLT:		/* general protection fault */
442 		case T_SEGNPFLT:	/* segment not present fault */
443 			/*
444 			 * Invalid segment selectors and out of bounds
445 			 * %eip's and %esp's can be set up in user mode.
446 			 * This causes a fault in kernel mode when the
447 			 * kernel tries to return to user mode.  We want
448 			 * to get this fault so that we can fix the
449 			 * problem here and not have to check all the
450 			 * selectors and pointers when the user changes
451 			 * them.
452 			 */
453 #define	MAYBE_DORETI_FAULT(where, whereto)				\
454 	do {								\
455 		if (frame.tf_eip == (int)where) {			\
456 			frame.tf_eip = (int)whereto;			\
457 			return;						\
458 		}							\
459 	} while (0)
460 
461 			if (intr_nesting_level == 0) {
462 				/*
463 				 * Invalid %fs's and %gs's can be created using
464 				 * procfs or PT_SETREGS or by invalidating the
465 				 * underlying LDT entry.  This causes a fault
466 				 * in kernel mode when the kernel attempts to
467 				 * switch contexts.  Lose the bad context
468 				 * (XXX) so that we can continue, and generate
469 				 * a signal.
470 				 */
471 				if (frame.tf_eip == (int)cpu_switch_load_gs) {
472 					curpcb->pcb_gs = 0;
473 					psignal(p, SIGBUS);
474 					return;
475 				}
476 				MAYBE_DORETI_FAULT(doreti_iret,
477 						   doreti_iret_fault);
478 				MAYBE_DORETI_FAULT(doreti_popl_ds,
479 						   doreti_popl_ds_fault);
480 				MAYBE_DORETI_FAULT(doreti_popl_es,
481 						   doreti_popl_es_fault);
482 				MAYBE_DORETI_FAULT(doreti_popl_fs,
483 						   doreti_popl_fs_fault);
484 				if (curpcb && curpcb->pcb_onfault) {
485 					frame.tf_eip = (int)curpcb->pcb_onfault;
486 					return;
487 				}
488 			}
489 			break;
490 
491 		case T_TSSFLT:
492 			/*
493 			 * PSL_NT can be set in user mode and isn't cleared
494 			 * automatically when the kernel is entered.  This
495 			 * causes a TSS fault when the kernel attempts to
496 			 * `iret' because the TSS link is uninitialized.  We
497 			 * want to get this fault so that we can fix the
498 			 * problem here and not every time the kernel is
499 			 * entered.
500 			 */
501 			if (frame.tf_eflags & PSL_NT) {
502 				frame.tf_eflags &= ~PSL_NT;
503 				return;
504 			}
505 			break;
506 
507 		case T_TRCTRAP:	 /* trace trap */
508 			if (frame.tf_eip == (int)IDTVEC(syscall)) {
509 				/*
510 				 * We've just entered system mode via the
511 				 * syscall lcall.  Continue single stepping
512 				 * silently until the syscall handler has
513 				 * saved the flags.
514 				 */
515 				return;
516 			}
517 			if (frame.tf_eip == (int)IDTVEC(syscall) + 1) {
518 				/*
519 				 * The syscall handler has now saved the
520 				 * flags.  Stop single stepping it.
521 				 */
522 				frame.tf_eflags &= ~PSL_T;
523 				return;
524 			}
525 			/*
526 			 * Fall through.
527 			 */
528 		case T_BPTFLT:
529 			/*
530 			 * If DDB is enabled, let it handle the debugger trap.
531 			 * Otherwise, debugger traps "can't happen".
532 			 */
533 #ifdef DDB
534 			if (kdb_trap (type, 0, &frame))
535 				return;
536 #endif
537 			break;
538 
539 #if NISA > 0
540 		case T_NMI:
541 #ifdef POWERFAIL_NMI
542 #ifndef TIMER_FREQ
543 #  define TIMER_FREQ 1193182
544 #endif
545 	handle_powerfail:
546 		{
547 		  static unsigned lastalert = 0;
548 
549 		  if(time_second - lastalert > 10)
550 		    {
551 		      log(LOG_WARNING, "NMI: power fail\n");
552 		      sysbeep(TIMER_FREQ/880, hz);
553 		      lastalert = time_second;
554 		    }
555 		  return;
556 		}
557 #else /* !POWERFAIL_NMI */
558 #ifdef DDB
559 			/* NMI can be hooked up to a pushbutton for debugging */
560 			printf ("NMI ... going to debugger\n");
561 			if (kdb_trap (type, 0, &frame))
562 				return;
563 #endif /* DDB */
564 			/* machine/parity/power fail/"kitchen sink" faults */
565 			if (isa_nmi(code) == 0) return;
566 			/* FALL THROUGH */
567 #endif /* POWERFAIL_NMI */
568 #endif /* NISA > 0 */
569 		}
570 
571 		trap_fatal(&frame, eva);
572 		return;
573 	}
574 
575 	/* Translate fault for emulators (e.g. Linux) */
576 	if (*p->p_sysent->sv_transtrap)
577 		i = (*p->p_sysent->sv_transtrap)(i, type);
578 
579 	trapsignal(p, i, ucode);
580 
581 #ifdef DEBUG
582 	if (type <= MAX_TRAP_MSG) {
583 		uprintf("fatal process exception: %s",
584 			trap_msg[type]);
585 		if ((type == T_PAGEFLT) || (type == T_PROTFLT))
586 			uprintf(", fault VA = 0x%lx", (u_long)eva);
587 		uprintf("\n");
588 	}
589 #endif
590 
591 out:
592 	userret(p, &frame, sticks);
593 }
594 
595 #ifdef notyet
596 /*
597  * This version doesn't allow a page fault to user space while
598  * in the kernel. The rest of the kernel needs to be made "safe"
599  * before this can be used. I think the only things remaining
600  * to be made safe are the iBCS2 code and the process tracing/
601  * debugging code.
602  */
603 static int
604 trap_pfault(frame, usermode, eva)
605 	struct trapframe *frame;
606 	int usermode;
607 	vm_offset_t eva;
608 {
609 	vm_offset_t va;
610 	struct vmspace *vm = NULL;
611 	vm_map_t map = 0;
612 	int rv = 0;
613 	vm_prot_t ftype;
614 	struct proc *p = curproc;
615 
616 	if (frame->tf_err & PGEX_W)
617 		ftype = VM_PROT_READ | VM_PROT_WRITE;
618 	else
619 		ftype = VM_PROT_READ;
620 
621 	va = trunc_page(eva);
622 	if (va < VM_MIN_KERNEL_ADDRESS) {
623 		vm_offset_t v;
624 		vm_page_t mpte;
625 
626 		if (p == NULL ||
627 		    (!usermode && va < VM_MAXUSER_ADDRESS &&
628 		     (intr_nesting_level != 0 || curpcb == NULL ||
629 		      curpcb->pcb_onfault == NULL))) {
630 			trap_fatal(frame, eva);
631 			return (-1);
632 		}
633 
634 		/*
635 		 * This is a fault on non-kernel virtual memory.
636 		 * vm is initialized above to NULL. If curproc is NULL
637 		 * or curproc->p_vmspace is NULL the fault is fatal.
638 		 */
639 		vm = p->p_vmspace;
640 		if (vm == NULL)
641 			goto nogo;
642 
643 		map = &vm->vm_map;
644 
645 		/*
646 		 * Keep swapout from messing with us during this
647 		 *	critical time.
648 		 */
649 		++p->p_lock;
650 
651 		/*
652 		 * Grow the stack if necessary
653 		 */
654 		/* grow_stack returns false only if va falls into
655 		 * a growable stack region and the stack growth
656 		 * fails.  It returns true if va was not within
657 		 * a growable stack region, or if the stack
658 		 * growth succeeded.
659 		 */
660 		if (!grow_stack (p, va)) {
661 			rv = KERN_FAILURE;
662 			--p->p_lock;
663 			goto nogo;
664 		}
665 
666 		/* Fault in the user page: */
667 		rv = vm_fault(map, va, ftype,
668 			(ftype & VM_PROT_WRITE) ? VM_FAULT_DIRTY : 0);
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, FALSE);
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 : 0);
783 
784 		--p->p_lock;
785 	} else {
786 		/*
787 		 * Don't have to worry about process locking or stacks in the kernel.
788 		 */
789 		rv = vm_fault(map, va, ftype, FALSE);
790 	}
791 
792 	if (rv == KERN_SUCCESS)
793 		return (0);
794 nogo:
795 	if (!usermode) {
796 		if (intr_nesting_level == 0 && curpcb && curpcb->pcb_onfault) {
797 			frame->tf_eip = (int)curpcb->pcb_onfault;
798 			return (0);
799 		}
800 		trap_fatal(frame, eva);
801 		return (-1);
802 	}
803 
804 	/* kludge to pass faulting virtual address to sendsig */
805 	frame->tf_err = eva;
806 
807 	return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
808 }
809 
810 static void
811 trap_fatal(frame, eva)
812 	struct trapframe *frame;
813 	vm_offset_t eva;
814 {
815 	int code, type, ss, esp;
816 	struct soft_segment_descriptor softseg;
817 
818 	code = frame->tf_err;
819 	type = frame->tf_trapno;
820 	sdtossd(&gdt[IDXSEL(frame->tf_cs & 0xffff)].sd, &softseg);
821 
822 	if (type <= MAX_TRAP_MSG)
823 		printf("\n\nFatal trap %d: %s while in %s mode\n",
824 			type, trap_msg[type],
825         		frame->tf_eflags & PSL_VM ? "vm86" :
826 			ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel");
827 #ifdef SMP
828 	/* three seperate prints in case of a trap on an unmapped page */
829 	printf("mp_lock = %08x; ", mp_lock);
830 	printf("cpuid = %d; ", cpuid);
831 	printf("lapic.id = %08x\n", lapic.id);
832 #endif
833 	if (type == T_PAGEFLT) {
834 		printf("fault virtual address	= 0x%x\n", eva);
835 		printf("fault code		= %s %s, %s\n",
836 			code & PGEX_U ? "user" : "supervisor",
837 			code & PGEX_W ? "write" : "read",
838 			code & PGEX_P ? "protection violation" : "page not present");
839 	}
840 	printf("instruction pointer	= 0x%x:0x%x\n",
841 	       frame->tf_cs & 0xffff, frame->tf_eip);
842         if ((ISPL(frame->tf_cs) == SEL_UPL) || (frame->tf_eflags & PSL_VM)) {
843 		ss = frame->tf_ss & 0xffff;
844 		esp = frame->tf_esp;
845 	} else {
846 		ss = GSEL(GDATA_SEL, SEL_KPL);
847 		esp = (int)&frame->tf_esp;
848 	}
849 	printf("stack pointer	        = 0x%x:0x%x\n", ss, esp);
850 	printf("frame pointer	        = 0x%x:0x%x\n", ss, frame->tf_ebp);
851 	printf("code segment		= base 0x%x, limit 0x%x, type 0x%x\n",
852 	       softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type);
853 	printf("			= DPL %d, pres %d, def32 %d, gran %d\n",
854 	       softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_def32,
855 	       softseg.ssd_gran);
856 	printf("processor eflags	= ");
857 	if (frame->tf_eflags & PSL_T)
858 		printf("trace trap, ");
859 	if (frame->tf_eflags & PSL_I)
860 		printf("interrupt enabled, ");
861 	if (frame->tf_eflags & PSL_NT)
862 		printf("nested task, ");
863 	if (frame->tf_eflags & PSL_RF)
864 		printf("resume, ");
865 	if (frame->tf_eflags & PSL_VM)
866 		printf("vm86, ");
867 	printf("IOPL = %d\n", (frame->tf_eflags & PSL_IOPL) >> 12);
868 	printf("current process		= ");
869 	if (curproc) {
870 		printf("%lu (%s)\n",
871 		    (u_long)curproc->p_pid, curproc->p_comm ?
872 		    curproc->p_comm : "");
873 	} else {
874 		printf("Idle\n");
875 	}
876 	printf("interrupt mask		= ");
877 	if ((cpl & net_imask) == net_imask)
878 		printf("net ");
879 	if ((cpl & tty_imask) == tty_imask)
880 		printf("tty ");
881 	if ((cpl & bio_imask) == bio_imask)
882 		printf("bio ");
883 	if ((cpl & cam_imask) == cam_imask)
884 		printf("cam ");
885 	if (cpl == 0)
886 		printf("none");
887 #ifdef SMP
888 /**
889  *  XXX FIXME:
890  *	we probably SHOULD have stopped the other CPUs before now!
891  *	another CPU COULD have been touching cpl at this moment...
892  */
893 	printf(" <- SMP: XXX");
894 #endif
895 	printf("\n");
896 
897 #ifdef KDB
898 	if (kdb_trap(&psl))
899 		return;
900 #endif
901 #ifdef DDB
902 	if ((debugger_on_panic || in_Debugger) && kdb_trap(type, 0, frame))
903 		return;
904 #endif
905 	printf("trap number		= %d\n", type);
906 	if (type <= MAX_TRAP_MSG)
907 		panic(trap_msg[type]);
908 	else
909 		panic("unknown/reserved trap");
910 }
911 
912 /*
913  * Double fault handler. Called when a fault occurs while writing
914  * a frame for a trap/exception onto the stack. This usually occurs
915  * when the stack overflows (such is the case with infinite recursion,
916  * for example).
917  *
918  * XXX Note that the current PTD gets replaced by IdlePTD when the
919  * task switch occurs. This means that the stack that was active at
920  * the time of the double fault is not available at <kstack> unless
921  * the machine was idle when the double fault occurred. The downside
922  * of this is that "trace <ebp>" in ddb won't work.
923  */
924 void
925 dblfault_handler()
926 {
927 	printf("\nFatal double fault:\n");
928 	printf("eip = 0x%x\n", common_tss.tss_eip);
929 	printf("esp = 0x%x\n", common_tss.tss_esp);
930 	printf("ebp = 0x%x\n", common_tss.tss_ebp);
931 #ifdef SMP
932 	/* three seperate prints in case of a trap on an unmapped page */
933 	printf("mp_lock = %08x; ", mp_lock);
934 	printf("cpuid = %d; ", cpuid);
935 	printf("lapic.id = %08x\n", lapic.id);
936 #endif
937 	panic("double fault");
938 }
939 
940 /*
941  * Compensate for 386 brain damage (missing URKR).
942  * This is a little simpler than the pagefault handler in trap() because
943  * it the page tables have already been faulted in and high addresses
944  * are thrown out early for other reasons.
945  */
946 int trapwrite(addr)
947 	unsigned addr;
948 {
949 	struct proc *p;
950 	vm_offset_t va;
951 	struct vmspace *vm;
952 	int rv;
953 
954 	va = trunc_page((vm_offset_t)addr);
955 	/*
956 	 * XXX - MAX is END.  Changed > to >= for temp. fix.
957 	 */
958 	if (va >= VM_MAXUSER_ADDRESS)
959 		return (1);
960 
961 	p = curproc;
962 	vm = p->p_vmspace;
963 
964 	++p->p_lock;
965 
966 	if (!grow_stack (p, va)) {
967 		--p->p_lock;
968 		return (1);
969 	}
970 
971 	/*
972 	 * fault the data page
973 	 */
974 	rv = vm_fault(&vm->vm_map, va, VM_PROT_READ|VM_PROT_WRITE, VM_FAULT_DIRTY);
975 
976 	--p->p_lock;
977 
978 	if (rv != KERN_SUCCESS)
979 		return 1;
980 
981 	return (0);
982 }
983 
984 /*
985  * System call request from POSIX system call gate interface to kernel.
986  * Like trap(), argument is call by reference.
987  */
988 void
989 syscall(frame)
990 	struct trapframe frame;
991 {
992 	caddr_t params;
993 	int i;
994 	struct sysent *callp;
995 	struct proc *p = curproc;
996 	u_quad_t sticks;
997 	int error;
998 	int args[8];
999 	u_int code;
1000 
1001 #ifdef DIAGNOSTIC
1002 	if (ISPL(frame.tf_cs) != SEL_UPL)
1003 		panic("syscall");
1004 #endif
1005 	sticks = p->p_sticks;
1006 	p->p_md.md_regs = &frame;
1007 	params = (caddr_t)frame.tf_esp + sizeof(int);
1008 	code = frame.tf_eax;
1009 	if (p->p_sysent->sv_prepsyscall) {
1010 		(*p->p_sysent->sv_prepsyscall)(&frame, args, &code, &params);
1011 	} else {
1012 		/*
1013 		 * Need to check if this is a 32 bit or 64 bit syscall.
1014 		 */
1015 		if (code == SYS_syscall) {
1016 			/*
1017 			 * Code is first argument, followed by actual args.
1018 			 */
1019 			code = fuword(params);
1020 			params += sizeof(int);
1021 		} else if (code == SYS___syscall) {
1022 			/*
1023 			 * Like syscall, but code is a quad, so as to maintain
1024 			 * quad alignment for the rest of the arguments.
1025 			 */
1026 			code = fuword(params);
1027 			params += sizeof(quad_t);
1028 		}
1029 	}
1030 
1031  	if (p->p_sysent->sv_mask)
1032  		code &= p->p_sysent->sv_mask;
1033 
1034  	if (code >= p->p_sysent->sv_size)
1035  		callp = &p->p_sysent->sv_table[0];
1036   	else
1037  		callp = &p->p_sysent->sv_table[code];
1038 
1039 	if (params && (i = callp->sy_narg * sizeof(int)) &&
1040 	    (error = copyin(params, (caddr_t)args, (u_int)i))) {
1041 #ifdef KTRACE
1042 		if (KTRPOINT(p, KTR_SYSCALL))
1043 			ktrsyscall(p->p_tracep, code, callp->sy_narg, args);
1044 #endif
1045 		goto bad;
1046 	}
1047 #ifdef KTRACE
1048 	if (KTRPOINT(p, KTR_SYSCALL))
1049 		ktrsyscall(p->p_tracep, code, callp->sy_narg, args);
1050 #endif
1051 	p->p_retval[0] = 0;
1052 	p->p_retval[1] = frame.tf_edx;
1053 
1054 	STOPEVENT(p, S_SCE, callp->sy_narg);
1055 
1056 	error = (*callp->sy_call)(p, args);
1057 
1058 	switch (error) {
1059 
1060 	case 0:
1061 		/*
1062 		 * Reinitialize proc pointer `p' as it may be different
1063 		 * if this is a child returning from fork syscall.
1064 		 */
1065 		p = curproc;
1066 		frame.tf_eax = p->p_retval[0];
1067 		frame.tf_edx = p->p_retval[1];
1068 		frame.tf_eflags &= ~PSL_C;
1069 		break;
1070 
1071 	case ERESTART:
1072 		/*
1073 		 * Reconstruct pc, assuming lcall $X,y is 7 bytes,
1074 		 * int 0x80 is 2 bytes. We saved this in tf_err.
1075 		 */
1076 		frame.tf_eip -= frame.tf_err;
1077 		break;
1078 
1079 	case EJUSTRETURN:
1080 		break;
1081 
1082 	default:
1083 bad:
1084  		if (p->p_sysent->sv_errsize) {
1085  			if (error >= p->p_sysent->sv_errsize)
1086   				error = -1;	/* XXX */
1087    			else
1088   				error = p->p_sysent->sv_errtbl[error];
1089 		}
1090 		frame.tf_eax = error;
1091 		frame.tf_eflags |= PSL_C;
1092 		break;
1093 	}
1094 
1095 	if ((frame.tf_eflags & PSL_T) && !(frame.tf_eflags & PSL_VM)) {
1096 		/* Traced syscall. */
1097 		frame.tf_eflags &= ~PSL_T;
1098 		trapsignal(p, SIGTRAP, 0);
1099 	}
1100 
1101 	userret(p, &frame, sticks);
1102 
1103 #ifdef KTRACE
1104 	if (KTRPOINT(p, KTR_SYSRET))
1105 		ktrsysret(p->p_tracep, code, error, p->p_retval[0]);
1106 #endif
1107 
1108 	/*
1109 	 * This works because errno is findable through the
1110 	 * register set.  If we ever support an emulation where this
1111 	 * is not the case, this code will need to be revisited.
1112 	 */
1113 	STOPEVENT(p, S_SCX, code);
1114 
1115 }
1116 
1117 /*
1118  * Simplified back end of syscall(), used when returning from fork()
1119  * directly into user mode.
1120  */
1121 void
1122 fork_return(p, frame)
1123 	struct proc *p;
1124 	struct trapframe frame;
1125 {
1126 	frame.tf_eax = 0;		/* Child returns zero */
1127 	frame.tf_eflags &= ~PSL_C;	/* success */
1128 	frame.tf_edx = 1;
1129 
1130 	userret(p, &frame, 0);
1131 #ifdef KTRACE
1132 	if (KTRPOINT(p, KTR_SYSRET))
1133 		ktrsysret(p->p_tracep, SYS_fork, 0, 0);
1134 #endif
1135 }
1136