xref: /dflybsd-src/sys/platform/pc64/x86_64/trap.c (revision ff10d9548f6b395b5e9c0485daa45a3b236c1572)
1 /*-
2  * Copyright (c) 1990, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  * Copyright (C) 1994, David Greenman
5  * Copyright (c) 2008 The DragonFly Project.
6  * Copyright (c) 2008 Jordan Gordeev.
7  *
8  * This code is derived from software contributed to Berkeley by
9  * the University of Utah, and William Jolitz.
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  * 3. All advertising materials mentioning features or use of this software
20  *    must display the following acknowledgement:
21  *	This product includes software developed by the University of
22  *	California, Berkeley and its contributors.
23  * 4. Neither the name of the University nor the names of its contributors
24  *    may be used to endorse or promote products derived from this software
25  *    without specific prior written permission.
26  *
27  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
28  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
29  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
30  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
31  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
32  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
33  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
34  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
35  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
36  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
37  * SUCH DAMAGE.
38  *
39  * from: @(#)trap.c	7.4 (Berkeley) 5/13/91
40  * $FreeBSD: src/sys/i386/i386/trap.c,v 1.147.2.11 2003/02/27 19:09:59 luoqi Exp $
41  */
42 
43 /*
44  * x86_64 Trap and System call handling
45  */
46 
47 #include "use_isa.h"
48 
49 #include "opt_ddb.h"
50 #include "opt_ktrace.h"
51 
52 #include <machine/frame.h>
53 #include <sys/param.h>
54 #include <sys/systm.h>
55 #include <sys/kernel.h>
56 #include <sys/kerneldump.h>
57 #include <sys/proc.h>
58 #include <sys/pioctl.h>
59 #include <sys/types.h>
60 #include <sys/signal2.h>
61 #include <sys/syscall.h>
62 #include <sys/sysctl.h>
63 #include <sys/sysent.h>
64 #include <sys/systm.h>
65 #ifdef KTRACE
66 #include <sys/ktrace.h>
67 #endif
68 #include <sys/ktr.h>
69 #include <sys/sysmsg.h>
70 #include <sys/sysproto.h>
71 #include <sys/sysunion.h>
72 
73 #include <vm/pmap.h>
74 #include <vm/vm.h>
75 #include <vm/vm_extern.h>
76 #include <vm/vm_kern.h>
77 #include <vm/vm_param.h>
78 #include <machine/cpu.h>
79 #include <machine/pcb.h>
80 #include <machine/smp.h>
81 #include <machine/thread.h>
82 #include <machine/clock.h>
83 #include <machine/vmparam.h>
84 #include <machine/md_var.h>
85 #include <machine_base/isa/isa_intr.h>
86 #include <machine_base/apic/lapic.h>
87 
88 #include <ddb/ddb.h>
89 
90 #include <sys/thread2.h>
91 #include <sys/mplock2.h>
92 
93 #ifdef SMP
94 
95 #define MAKEMPSAFE(have_mplock)			\
96 	if (have_mplock == 0) {			\
97 		get_mplock();			\
98 		have_mplock = 1;		\
99 	}
100 
101 #else
102 
103 #define MAKEMPSAFE(have_mplock)
104 
105 #endif
106 
107 extern void trap(struct trapframe *frame);
108 
109 static int trap_pfault(struct trapframe *, int);
110 static void trap_fatal(struct trapframe *, vm_offset_t);
111 void dblfault_handler(struct trapframe *frame);
112 
113 #define MAX_TRAP_MSG		30
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 	"SIMD floating-point exception",	/* 29 T_XMMFLT */
145 	"reserved (unknown) fault",		/* 30 T_RESERVED */
146 };
147 
148 #ifdef DDB
149 static int ddb_on_nmi = 1;
150 SYSCTL_INT(_machdep, OID_AUTO, ddb_on_nmi, CTLFLAG_RW,
151 	&ddb_on_nmi, 0, "Go to DDB on NMI");
152 static int ddb_on_seg_fault = 0;
153 SYSCTL_INT(_machdep, OID_AUTO, ddb_on_seg_fault, CTLFLAG_RW,
154 	&ddb_on_seg_fault, 0, "Go to DDB on user seg-fault");
155 static int freeze_on_seg_fault = 0;
156 SYSCTL_INT(_machdep, OID_AUTO, freeze_on_seg_fault, CTLFLAG_RW,
157 	&freeze_on_seg_fault, 0, "Go to DDB on user seg-fault");
158 #endif
159 static int panic_on_nmi = 1;
160 SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RW,
161 	&panic_on_nmi, 0, "Panic on NMI");
162 static int fast_release;
163 SYSCTL_INT(_machdep, OID_AUTO, fast_release, CTLFLAG_RW,
164 	&fast_release, 0, "Passive Release was optimal");
165 static int slow_release;
166 SYSCTL_INT(_machdep, OID_AUTO, slow_release, CTLFLAG_RW,
167 	&slow_release, 0, "Passive Release was nonoptimal");
168 
169 /*
170  * System call debugging records the worst-case system call
171  * overhead (inclusive of blocking), but may be inaccurate.
172  */
173 /*#define SYSCALL_DEBUG*/
174 #ifdef SYSCALL_DEBUG
175 uint64_t SysCallsWorstCase[SYS_MAXSYSCALL];
176 #endif
177 
178 /*
179  * Passively intercepts the thread switch function to increase
180  * the thread priority from a user priority to a kernel priority, reducing
181  * syscall and trap overhead for the case where no switch occurs.
182  *
183  * Synchronizes td_ucred with p_ucred.  This is used by system calls,
184  * signal handling, faults, AST traps, and anything else that enters the
185  * kernel from userland and provides the kernel with a stable read-only
186  * copy of the process ucred.
187  */
188 static __inline void
189 userenter(struct thread *curtd, struct proc *curp)
190 {
191 	struct ucred *ocred;
192 	struct ucred *ncred;
193 
194 	curtd->td_release = lwkt_passive_release;
195 
196 	if (curtd->td_ucred != curp->p_ucred) {
197 		ncred = crhold(curp->p_ucred);
198 		ocred = curtd->td_ucred;
199 		curtd->td_ucred = ncred;
200 		if (ocred)
201 			crfree(ocred);
202 	}
203 }
204 
205 /*
206  * Handle signals, upcalls, profiling, and other AST's and/or tasks that
207  * must be completed before we can return to or try to return to userland.
208  *
209  * Note that td_sticks is a 64 bit quantity, but there's no point doing 64
210  * arithmatic on the delta calculation so the absolute tick values are
211  * truncated to an integer.
212  */
213 static void
214 userret(struct lwp *lp, struct trapframe *frame, int sticks)
215 {
216 	struct proc *p = lp->lwp_proc;
217 	int sig;
218 
219 	/*
220 	 * Charge system time if profiling.  Note: times are in microseconds.
221 	 * This may do a copyout and block, so do it first even though it
222 	 * means some system time will be charged as user time.
223 	 */
224 	if (p->p_flag & P_PROFIL) {
225 		addupc_task(p, frame->tf_rip,
226 			(u_int)((int)lp->lwp_thread->td_sticks - sticks));
227 	}
228 
229 recheck:
230 	/*
231 	 * If the jungle wants us dead, so be it.
232 	 */
233 	if (lp->lwp_flag & LWP_WEXIT) {
234 		lwkt_gettoken(&p->p_token);
235 		lwp_exit(0);
236 		lwkt_reltoken(&p->p_token);	/* NOT REACHED */
237 	}
238 
239 	/*
240 	 * Block here if we are in a stopped state.
241 	 */
242 	if (p->p_stat == SSTOP || dump_stop_usertds) {
243 		get_mplock();
244 		tstop();
245 		rel_mplock();
246 		goto recheck;
247 	}
248 
249 	/*
250 	 * Post any pending upcalls.  If running a virtual kernel be sure
251 	 * to restore the virtual kernel's vmspace before posting the upcall.
252 	 */
253 	if (p->p_flag & (P_SIGVTALRM | P_SIGPROF | P_UPCALLPEND)) {
254 		lwkt_gettoken(&p->p_token);
255 		if (p->p_flag & P_SIGVTALRM) {
256 			p->p_flag &= ~P_SIGVTALRM;
257 			ksignal(p, SIGVTALRM);
258 		}
259 		if (p->p_flag & P_SIGPROF) {
260 			p->p_flag &= ~P_SIGPROF;
261 			ksignal(p, SIGPROF);
262 		}
263 		if (p->p_flag & P_UPCALLPEND) {
264 			p->p_flag &= ~P_UPCALLPEND;
265 			postupcall(lp);
266 		}
267 		lwkt_reltoken(&p->p_token);
268 		goto recheck;
269 	}
270 
271 	/*
272 	 * Post any pending signals.  If running a virtual kernel be sure
273 	 * to restore the virtual kernel's vmspace before posting the signal.
274 	 *
275 	 * WARNING!  postsig() can exit and not return.
276 	 */
277 	if ((sig = CURSIG_TRACE(lp)) != 0) {
278 		lwkt_gettoken(&p->p_token);
279 		postsig(sig);
280 		lwkt_reltoken(&p->p_token);
281 		goto recheck;
282 	}
283 
284 	/*
285 	 * block here if we are swapped out, but still process signals
286 	 * (such as SIGKILL).  proc0 (the swapin scheduler) is already
287 	 * aware of our situation, we do not have to wake it up.
288 	 */
289 	if (p->p_flag & P_SWAPPEDOUT) {
290 		lwkt_gettoken(&p->p_token);
291 		get_mplock();
292 		p->p_flag |= P_SWAPWAIT;
293 		swapin_request();
294 		if (p->p_flag & P_SWAPWAIT)
295 			tsleep(p, PCATCH, "SWOUT", 0);
296 		p->p_flag &= ~P_SWAPWAIT;
297 		rel_mplock();
298 		lwkt_reltoken(&p->p_token);
299 		goto recheck;
300 	}
301 
302 	/*
303 	 * Make sure postsig() handled request to restore old signal mask after
304 	 * running signal handler.
305 	 */
306 	KKASSERT((lp->lwp_flag & LWP_OLDMASK) == 0);
307 }
308 
309 /*
310  * Cleanup from userenter and any passive release that might have occured.
311  * We must reclaim the current-process designation before we can return
312  * to usermode.  We also handle both LWKT and USER reschedule requests.
313  */
314 static __inline void
315 userexit(struct lwp *lp)
316 {
317 	struct thread *td = lp->lwp_thread;
318 /*	globaldata_t gd = td->td_gd;*/
319 
320 	/*
321 	 * Handle stop requests at kernel priority.  Any requests queued
322 	 * after this loop will generate another AST.
323 	 */
324 	while (lp->lwp_proc->p_stat == SSTOP) {
325 		get_mplock();
326 		tstop();
327 		rel_mplock();
328 	}
329 
330 	/*
331 	 * Reduce our priority in preparation for a return to userland.  If
332 	 * our passive release function was still in place, our priority was
333 	 * never raised and does not need to be reduced.
334 	 */
335 	lwkt_passive_recover(td);
336 
337 	/*
338 	 * Become the current user scheduled process if we aren't already,
339 	 * and deal with reschedule requests and other factors.
340 	 */
341 	lp->lwp_proc->p_usched->acquire_curproc(lp);
342 	/* WARNING: we may have migrated cpu's */
343 	/* gd = td->td_gd; */
344 }
345 
346 #if !defined(KTR_KERNENTRY)
347 #define	KTR_KERNENTRY	KTR_ALL
348 #endif
349 KTR_INFO_MASTER(kernentry);
350 KTR_INFO(KTR_KERNENTRY, kernentry, trap, 0, "STR",
351 	 sizeof(long) + sizeof(long) + sizeof(long) + sizeof(vm_offset_t));
352 KTR_INFO(KTR_KERNENTRY, kernentry, trap_ret, 0, "STR",
353 	 sizeof(long) + sizeof(long));
354 KTR_INFO(KTR_KERNENTRY, kernentry, syscall, 0, "STR",
355 	 sizeof(long) + sizeof(long) + sizeof(long));
356 KTR_INFO(KTR_KERNENTRY, kernentry, syscall_ret, 0, "STR",
357 	 sizeof(long) + sizeof(long) + sizeof(long));
358 KTR_INFO(KTR_KERNENTRY, kernentry, fork_ret, 0, "STR",
359 	 sizeof(long) + sizeof(long));
360 
361 /*
362  * Exception, fault, and trap interface to the kernel.
363  * This common code is called from assembly language IDT gate entry
364  * routines that prepare a suitable stack frame, and restore this
365  * frame after the exception has been processed.
366  *
367  * This function is also called from doreti in an interlock to handle ASTs.
368  * For example:  hardwareint->INTROUTINE->(set ast)->doreti->trap
369  *
370  * NOTE!  We have to retrieve the fault address prior to obtaining the
371  * MP lock because get_mplock() may switch out.  YYY cr2 really ought
372  * to be retrieved by the assembly code, not here.
373  *
374  * XXX gd_trap_nesting_level currently prevents lwkt_switch() from panicing
375  * if an attempt is made to switch from a fast interrupt or IPI.  This is
376  * necessary to properly take fatal kernel traps on SMP machines if
377  * get_mplock() has to block.
378  */
379 
380 void
381 trap(struct trapframe *frame)
382 {
383 	struct globaldata *gd = mycpu;
384 	struct thread *td = gd->gd_curthread;
385 	struct lwp *lp = td->td_lwp;
386 	struct proc *p;
387 	int sticks = 0;
388 	int i = 0, ucode = 0, type, code;
389 #ifdef SMP
390 	int have_mplock = 0;
391 #endif
392 #ifdef INVARIANTS
393 	int crit_count = td->td_critcount;
394 	lwkt_tokref_t curstop = td->td_toks_stop;
395 #endif
396 	vm_offset_t eva;
397 
398 	p = td->td_proc;
399 	clear_quickret();
400 
401 #ifdef DDB
402         /*
403 	 * We need to allow T_DNA faults when the debugger is active since
404 	 * some dumping paths do large bcopy() which use the floating
405 	 * point registers for faster copying.
406 	 */
407 	if (db_active && frame->tf_trapno != T_DNA) {
408 		eva = (frame->tf_trapno == T_PAGEFLT ? frame->tf_addr : 0);
409 		++gd->gd_trap_nesting_level;
410 		MAKEMPSAFE(have_mplock);
411 		trap_fatal(frame, eva);
412 		--gd->gd_trap_nesting_level;
413 		goto out2;
414 	}
415 #endif
416 
417 	eva = 0;
418 
419 	if ((frame->tf_rflags & PSL_I) == 0) {
420 		/*
421 		 * Buggy application or kernel code has disabled interrupts
422 		 * and then trapped.  Enabling interrupts now is wrong, but
423 		 * it is better than running with interrupts disabled until
424 		 * they are accidentally enabled later.
425 		 */
426 		type = frame->tf_trapno;
427 		if (ISPL(frame->tf_cs) == SEL_UPL) {
428 			MAKEMPSAFE(have_mplock);
429 			/* JG curproc can be NULL */
430 			kprintf(
431 			    "pid %ld (%s): trap %d with interrupts disabled\n",
432 			    (long)curproc->p_pid, curproc->p_comm, type);
433 		} else if (type != T_NMI && type != T_BPTFLT &&
434 		    type != T_TRCTRAP) {
435 			/*
436 			 * XXX not quite right, since this may be for a
437 			 * multiple fault in user mode.
438 			 */
439 			MAKEMPSAFE(have_mplock);
440 			kprintf("kernel trap %d with interrupts disabled\n",
441 			    type);
442 		}
443 		cpu_enable_intr();
444 	}
445 
446 	type = frame->tf_trapno;
447 	code = frame->tf_err;
448 
449 	if (ISPL(frame->tf_cs) == SEL_UPL) {
450 		/* user trap */
451 
452 		KTR_LOG(kernentry_trap, p->p_pid, lp->lwp_tid,
453 			frame->tf_trapno, eva);
454 
455 		userenter(td, p);
456 
457 		sticks = (int)td->td_sticks;
458 		KASSERT(lp->lwp_md.md_regs == frame,
459 			("Frame mismatch %p %p", lp->lwp_md.md_regs, frame));
460 
461 		switch (type) {
462 		case T_PRIVINFLT:	/* privileged instruction fault */
463 			ucode = ILL_PRVOPC;
464 			i = SIGILL;
465 			break;
466 
467 		case T_BPTFLT:		/* bpt instruction fault */
468 		case T_TRCTRAP:		/* trace trap */
469 			frame->tf_rflags &= ~PSL_T;
470 			ucode = TRAP_TRACE;
471 			i = SIGTRAP;
472 			break;
473 
474 		case T_ARITHTRAP:	/* arithmetic trap */
475 			ucode = code;
476 			i = SIGFPE;
477 #if 0
478 #if JG
479 			ucode = fputrap();
480 #else
481 			ucode = code;
482 #endif
483 			i = SIGFPE;
484 #endif
485 			break;
486 
487 		case T_ASTFLT:		/* Allow process switch */
488 			mycpu->gd_cnt.v_soft++;
489 			if (mycpu->gd_reqflags & RQF_AST_OWEUPC) {
490 				atomic_clear_int(&mycpu->gd_reqflags,
491 						 RQF_AST_OWEUPC);
492 				addupc_task(p, p->p_prof.pr_addr,
493 					    p->p_prof.pr_ticks);
494 			}
495 			goto out;
496 
497 		case T_PROTFLT:		/* general protection fault */
498 			i = SIGBUS;
499 			ucode = BUS_OBJERR;
500 			break;
501 		case T_SEGNPFLT:	/* segment not present fault */
502 			i = SIGBUS;
503 			ucode = BUS_ADRERR;
504 			break;
505 		case T_TSSFLT:		/* invalid TSS fault */
506 		case T_DOUBLEFLT:	/* double fault */
507 			i = SIGBUS;
508 			ucode = BUS_OBJERR;
509 		default:
510 #if 0
511 			ucode = code + BUS_SEGM_FAULT ; /* XXX: ???*/
512 #endif
513 			ucode = BUS_OBJERR;
514 			i = SIGBUS;
515 			break;
516 
517 		case T_PAGEFLT:		/* page fault */
518 			i = trap_pfault(frame, TRUE);
519 			if (frame->tf_rip == 0) {
520 				kprintf("T_PAGEFLT: Warning %%rip == 0!\n");
521 				while (freeze_on_seg_fault) {
522 					tsleep(p, 0, "freeze", hz * 20);
523 				}
524 			}
525 			if (i == -1)
526 				goto out;
527 			if (i == 0)
528 				goto out;
529 
530 #if 0
531 			ucode = T_PAGEFLT;
532 #endif
533 			if (i == SIGSEGV)
534 				ucode = SEGV_MAPERR;
535 			else
536 				ucode = BUS_ADRERR;
537 			break;
538 
539 		case T_DIVIDE:		/* integer divide fault */
540 			ucode = FPE_INTDIV;
541 			i = SIGFPE;
542 			break;
543 
544 #if NISA > 0
545 		case T_NMI:
546 			MAKEMPSAFE(have_mplock);
547 			/* machine/parity/power fail/"kitchen sink" faults */
548 			if (isa_nmi(code) == 0) {
549 #ifdef DDB
550 				/*
551 				 * NMI can be hooked up to a pushbutton
552 				 * for debugging.
553 				 */
554 				if (ddb_on_nmi) {
555 					kprintf ("NMI ... going to debugger\n");
556 					kdb_trap(type, 0, frame);
557 				}
558 #endif /* DDB */
559 				goto out2;
560 			} else if (panic_on_nmi)
561 				panic("NMI indicates hardware failure");
562 			break;
563 #endif /* NISA > 0 */
564 
565 		case T_OFLOW:		/* integer overflow fault */
566 			ucode = FPE_INTOVF;
567 			i = SIGFPE;
568 			break;
569 
570 		case T_BOUND:		/* bounds check fault */
571 			ucode = FPE_FLTSUB;
572 			i = SIGFPE;
573 			break;
574 
575 		case T_DNA:
576 			/*
577 			 * Virtual kernel intercept - pass the DNA exception
578 			 * to the virtual kernel if it asked to handle it.
579 			 * This occurs when the virtual kernel is holding
580 			 * onto the FP context for a different emulated
581 			 * process then the one currently running.
582 			 *
583 			 * We must still call npxdna() since we may have
584 			 * saved FP state that the virtual kernel needs
585 			 * to hand over to a different emulated process.
586 			 */
587 			if (lp->lwp_vkernel && lp->lwp_vkernel->ve &&
588 			    (td->td_pcb->pcb_flags & FP_VIRTFP)
589 			) {
590 				npxdna();
591 				break;
592 			}
593 
594 			/*
595 			 * The kernel may have switched out the FP unit's
596 			 * state, causing the user process to take a fault
597 			 * when it tries to use the FP unit.  Restore the
598 			 * state here
599 			 */
600 			if (npxdna())
601 				goto out;
602 			i = SIGFPE;
603 			ucode = FPE_FPU_NP_TRAP;
604 			break;
605 
606 		case T_FPOPFLT:		/* FPU operand fetch fault */
607 			ucode = ILL_COPROC;
608 			i = SIGILL;
609 			break;
610 
611 		case T_XMMFLT:		/* SIMD floating-point exception */
612 			ucode = 0; /* XXX */
613 			i = SIGFPE;
614 			break;
615 		}
616 	} else {
617 		/* kernel trap */
618 
619 		switch (type) {
620 		case T_PAGEFLT:			/* page fault */
621 			trap_pfault(frame, FALSE);
622 			goto out2;
623 
624 		case T_DNA:
625 			/*
626 			 * The kernel is apparently using fpu for copying.
627 			 * XXX this should be fatal unless the kernel has
628 			 * registered such use.
629 			 */
630 			if (npxdna())
631 				goto out2;
632 			break;
633 
634 		case T_STKFLT:		/* stack fault */
635 			break;
636 
637 		case T_PROTFLT:		/* general protection fault */
638 		case T_SEGNPFLT:	/* segment not present fault */
639 			/*
640 			 * Invalid segment selectors and out of bounds
641 			 * %rip's and %rsp's can be set up in user mode.
642 			 * This causes a fault in kernel mode when the
643 			 * kernel tries to return to user mode.  We want
644 			 * to get this fault so that we can fix the
645 			 * problem here and not have to check all the
646 			 * selectors and pointers when the user changes
647 			 * them.
648 			 */
649 			if (mycpu->gd_intr_nesting_level == 0) {
650 				if (td->td_pcb->pcb_onfault) {
651 					frame->tf_rip = (register_t)
652 						td->td_pcb->pcb_onfault;
653 					goto out2;
654 				}
655 				if (frame->tf_rip == (long)doreti_iret) {
656 					frame->tf_rip = (long)doreti_iret_fault;
657 					goto out2;
658 				}
659 			}
660 			break;
661 
662 		case T_TSSFLT:
663 			/*
664 			 * PSL_NT can be set in user mode and isn't cleared
665 			 * automatically when the kernel is entered.  This
666 			 * causes a TSS fault when the kernel attempts to
667 			 * `iret' because the TSS link is uninitialized.  We
668 			 * want to get this fault so that we can fix the
669 			 * problem here and not every time the kernel is
670 			 * entered.
671 			 */
672 			if (frame->tf_rflags & PSL_NT) {
673 				frame->tf_rflags &= ~PSL_NT;
674 				goto out2;
675 			}
676 			break;
677 
678 		case T_TRCTRAP:	 /* trace trap */
679 #if 0
680 			if (frame->tf_rip == (int)IDTVEC(syscall)) {
681 				/*
682 				 * We've just entered system mode via the
683 				 * syscall lcall.  Continue single stepping
684 				 * silently until the syscall handler has
685 				 * saved the flags.
686 				 */
687 				goto out2;
688 			}
689 			if (frame->tf_rip == (int)IDTVEC(syscall) + 1) {
690 				/*
691 				 * The syscall handler has now saved the
692 				 * flags.  Stop single stepping it.
693 				 */
694 				frame->tf_rflags &= ~PSL_T;
695 				goto out2;
696 			}
697 #endif
698 
699 			/*
700 			 * Ignore debug register trace traps due to
701 			 * accesses in the user's address space, which
702 			 * can happen under several conditions such as
703 			 * if a user sets a watchpoint on a buffer and
704 			 * then passes that buffer to a system call.
705 			 * We still want to get TRCTRAPS for addresses
706 			 * in kernel space because that is useful when
707 			 * debugging the kernel.
708 			 */
709 #if JG
710 			if (user_dbreg_trap()) {
711 				/*
712 				 * Reset breakpoint bits because the
713 				 * processor doesn't
714 				 */
715 				/* XXX check upper bits here */
716 				load_dr6(rdr6() & 0xfffffff0);
717 				goto out2;
718 			}
719 #endif
720 			/*
721 			 * FALLTHROUGH (TRCTRAP kernel mode, kernel address)
722 			 */
723 		case T_BPTFLT:
724 			/*
725 			 * If DDB is enabled, let it handle the debugger trap.
726 			 * Otherwise, debugger traps "can't happen".
727 			 */
728 			ucode = TRAP_BRKPT;
729 #ifdef DDB
730 			MAKEMPSAFE(have_mplock);
731 			if (kdb_trap(type, 0, frame))
732 				goto out2;
733 #endif
734 			break;
735 
736 #if NISA > 0
737 		case T_NMI:
738 			MAKEMPSAFE(have_mplock);
739 			/* machine/parity/power fail/"kitchen sink" faults */
740 			if (isa_nmi(code) == 0) {
741 #ifdef DDB
742 				/*
743 				 * NMI can be hooked up to a pushbutton
744 				 * for debugging.
745 				 */
746 				if (ddb_on_nmi) {
747 					kprintf ("NMI ... going to debugger\n");
748 					kdb_trap(type, 0, frame);
749 				}
750 #endif /* DDB */
751 				goto out2;
752 			} else if (panic_on_nmi == 0)
753 				goto out2;
754 			/* FALL THROUGH */
755 #endif /* NISA > 0 */
756 		}
757 		MAKEMPSAFE(have_mplock);
758 		trap_fatal(frame, 0);
759 		goto out2;
760 	}
761 
762 	/*
763 	 * Virtual kernel intercept - if the fault is directly related to a
764 	 * VM context managed by a virtual kernel then let the virtual kernel
765 	 * handle it.
766 	 */
767 	if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
768 		vkernel_trap(lp, frame);
769 		goto out;
770 	}
771 
772 	/*
773 	 * Translate fault for emulators (e.g. Linux)
774 	 */
775 	if (*p->p_sysent->sv_transtrap)
776 		i = (*p->p_sysent->sv_transtrap)(i, type);
777 
778 	MAKEMPSAFE(have_mplock);
779 	trapsignal(lp, i, ucode);
780 
781 #ifdef DEBUG
782 	if (type <= MAX_TRAP_MSG) {
783 		uprintf("fatal process exception: %s",
784 			trap_msg[type]);
785 		if ((type == T_PAGEFLT) || (type == T_PROTFLT))
786 			uprintf(", fault VA = 0x%lx", frame->tf_addr);
787 		uprintf("\n");
788 	}
789 #endif
790 
791 out:
792 	userret(lp, frame, sticks);
793 	userexit(lp);
794 out2:	;
795 #ifdef SMP
796 	if (have_mplock)
797 		rel_mplock();
798 #endif
799 	if (p != NULL && lp != NULL)
800 		KTR_LOG(kernentry_trap_ret, p->p_pid, lp->lwp_tid);
801 #ifdef INVARIANTS
802 	KASSERT(crit_count == td->td_critcount,
803 		("trap: critical section count mismatch! %d/%d",
804 		crit_count, td->td_pri));
805 	KASSERT(curstop == td->td_toks_stop,
806 		("trap: extra tokens held after trap! %ld/%ld",
807 		curstop - &td->td_toks_base,
808 		td->td_toks_stop - &td->td_toks_base));
809 #endif
810 }
811 
812 static int
813 trap_pfault(struct trapframe *frame, int usermode)
814 {
815 	vm_offset_t va;
816 	struct vmspace *vm = NULL;
817 	vm_map_t map;
818 	int rv = 0;
819 	int fault_flags;
820 	vm_prot_t ftype;
821 	thread_t td = curthread;
822 	struct lwp *lp = td->td_lwp;
823 	struct proc *p;
824 
825 	va = trunc_page(frame->tf_addr);
826 	if (va >= VM_MIN_KERNEL_ADDRESS) {
827 		/*
828 		 * Don't allow user-mode faults in kernel address space.
829 		 */
830 		if (usermode) {
831 			fault_flags = -1;
832 			ftype = -1;
833 			goto nogo;
834 		}
835 
836 		map = &kernel_map;
837 	} else {
838 		/*
839 		 * This is a fault on non-kernel virtual memory.
840 		 * vm is initialized above to NULL. If curproc is NULL
841 		 * or curproc->p_vmspace is NULL the fault is fatal.
842 		 */
843 		if (lp != NULL)
844 			vm = lp->lwp_vmspace;
845 
846 		if (vm == NULL) {
847 			fault_flags = -1;
848 			ftype = -1;
849 			goto nogo;
850 		}
851 
852 		map = &vm->vm_map;
853 	}
854 
855 	/*
856 	 * PGEX_I is defined only if the execute disable bit capability is
857 	 * supported and enabled.
858 	 */
859 	if (frame->tf_err & PGEX_W)
860 		ftype = VM_PROT_WRITE;
861 #if JG
862 	else if ((frame->tf_err & PGEX_I) && pg_nx != 0)
863 		ftype = VM_PROT_EXECUTE;
864 #endif
865 	else
866 		ftype = VM_PROT_READ;
867 
868 	if (map != &kernel_map) {
869 		/*
870 		 * Keep swapout from messing with us during this
871 		 *	critical time.
872 		 */
873 		PHOLD(lp->lwp_proc);
874 
875 		/*
876 		 * Issue fault
877 		 */
878 		fault_flags = 0;
879 		if (usermode)
880 			fault_flags |= VM_FAULT_BURST;
881 		if (ftype & VM_PROT_WRITE)
882 			fault_flags |= VM_FAULT_DIRTY;
883 		else
884 			fault_flags |= VM_FAULT_NORMAL;
885 		rv = vm_fault(map, va, ftype, fault_flags);
886 
887 		PRELE(lp->lwp_proc);
888 	} else {
889 		/*
890 		 * Don't have to worry about process locking or stacks
891 		 * in the kernel.
892 		 */
893 		fault_flags = VM_FAULT_NORMAL;
894 		rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
895 	}
896 
897 	if (rv == KERN_SUCCESS)
898 		return (0);
899 nogo:
900 	if (!usermode) {
901 		if (td->td_gd->gd_intr_nesting_level == 0 &&
902 		    td->td_pcb->pcb_onfault) {
903 			frame->tf_rip = (register_t)td->td_pcb->pcb_onfault;
904 			return (0);
905 		}
906 		trap_fatal(frame, frame->tf_addr);
907 		return (-1);
908 	}
909 
910 	/*
911 	 * NOTE: on x86_64 we have a tf_addr field in the trapframe, no
912 	 * kludge is needed to pass the fault address to signal handlers.
913 	 */
914 	p = td->td_proc;
915 	if (td->td_lwp->lwp_vkernel == NULL) {
916 		if (bootverbose || freeze_on_seg_fault || ddb_on_seg_fault) {
917 			kprintf("seg-fault ft=%04x ff=%04x addr=%p rip=%p "
918 			    "pid=%d p_comm=%s\n",
919 			    ftype, fault_flags,
920 			    (void *)frame->tf_addr,
921 			    (void *)frame->tf_rip,
922 			    p->p_pid, p->p_comm);
923 		}
924 #ifdef DDB
925 		while (freeze_on_seg_fault) {
926 			tsleep(p, 0, "freeze", hz * 20);
927 		}
928 		if (ddb_on_seg_fault)
929 			Debugger("ddb_on_seg_fault");
930 #endif
931 	}
932 
933 	return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
934 }
935 
936 static void
937 trap_fatal(struct trapframe *frame, vm_offset_t eva)
938 {
939 	int code, ss;
940 	u_int type;
941 	long rsp;
942 	struct soft_segment_descriptor softseg;
943 	char *msg;
944 
945 	code = frame->tf_err;
946 	type = frame->tf_trapno;
947 	sdtossd(&gdt[IDXSEL(frame->tf_cs & 0xffff)], &softseg);
948 
949 	if (type <= MAX_TRAP_MSG)
950 		msg = trap_msg[type];
951 	else
952 		msg = "UNKNOWN";
953 	kprintf("\n\nFatal trap %d: %s while in %s mode\n", type, msg,
954 	    ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel");
955 #ifdef SMP
956 	/* three separate prints in case of a trap on an unmapped page */
957 	kprintf("cpuid = %d; ", mycpu->gd_cpuid);
958 	kprintf("lapic->id = %08x\n", lapic->id);
959 #endif
960 	if (type == T_PAGEFLT) {
961 		kprintf("fault virtual address	= 0x%lx\n", eva);
962 		kprintf("fault code		= %s %s %s, %s\n",
963 			code & PGEX_U ? "user" : "supervisor",
964 			code & PGEX_W ? "write" : "read",
965 			code & PGEX_I ? "instruction" : "data",
966 			code & PGEX_P ? "protection violation" : "page not present");
967 	}
968 	kprintf("instruction pointer	= 0x%lx:0x%lx\n",
969 	       frame->tf_cs & 0xffff, frame->tf_rip);
970         if (ISPL(frame->tf_cs) == SEL_UPL) {
971 		ss = frame->tf_ss & 0xffff;
972 		rsp = frame->tf_rsp;
973 	} else {
974 		ss = GSEL(GDATA_SEL, SEL_KPL);
975 		rsp = (long)&frame->tf_rsp;
976 	}
977 	kprintf("stack pointer	        = 0x%x:0x%lx\n", ss, rsp);
978 	kprintf("frame pointer	        = 0x%x:0x%lx\n", ss, frame->tf_rbp);
979 	kprintf("code segment		= base 0x%lx, limit 0x%lx, type 0x%x\n",
980 	       softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type);
981 	kprintf("			= DPL %d, pres %d, long %d, def32 %d, gran %d\n",
982 	       softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_long, softseg.ssd_def32,
983 	       softseg.ssd_gran);
984 	kprintf("processor eflags	= ");
985 	if (frame->tf_rflags & PSL_T)
986 		kprintf("trace trap, ");
987 	if (frame->tf_rflags & PSL_I)
988 		kprintf("interrupt enabled, ");
989 	if (frame->tf_rflags & PSL_NT)
990 		kprintf("nested task, ");
991 	if (frame->tf_rflags & PSL_RF)
992 		kprintf("resume, ");
993 	kprintf("IOPL = %ld\n", (frame->tf_rflags & PSL_IOPL) >> 12);
994 	kprintf("current process		= ");
995 	if (curproc) {
996 		kprintf("%lu\n",
997 		    (u_long)curproc->p_pid);
998 	} else {
999 		kprintf("Idle\n");
1000 	}
1001 	kprintf("current thread          = pri %d ", curthread->td_pri);
1002 	if (curthread->td_critcount)
1003 		kprintf("(CRIT)");
1004 	kprintf("\n");
1005 
1006 #ifdef DDB
1007 	if ((debugger_on_panic || db_active) && kdb_trap(type, code, frame))
1008 		return;
1009 #endif
1010 	kprintf("trap number		= %d\n", type);
1011 	if (type <= MAX_TRAP_MSG)
1012 		panic("%s", trap_msg[type]);
1013 	else
1014 		panic("unknown/reserved trap");
1015 }
1016 
1017 /*
1018  * Double fault handler. Called when a fault occurs while writing
1019  * a frame for a trap/exception onto the stack. This usually occurs
1020  * when the stack overflows (such is the case with infinite recursion,
1021  * for example).
1022  */
1023 static __inline
1024 int
1025 in_kstack_guard(register_t rptr)
1026 {
1027 	thread_t td = curthread;
1028 
1029 	if ((char *)rptr >= td->td_kstack &&
1030 	    (char *)rptr < td->td_kstack + PAGE_SIZE) {
1031 		return 1;
1032 	}
1033 	return 0;
1034 }
1035 
1036 void
1037 dblfault_handler(struct trapframe *frame)
1038 {
1039 	thread_t td = curthread;
1040 
1041 	if (in_kstack_guard(frame->tf_rsp) || in_kstack_guard(frame->tf_rbp)) {
1042 		kprintf("DOUBLE FAULT - KERNEL STACK GUARD HIT!\n");
1043 		if (in_kstack_guard(frame->tf_rsp))
1044 			frame->tf_rsp = (register_t)(td->td_kstack + PAGE_SIZE);
1045 		if (in_kstack_guard(frame->tf_rbp))
1046 			frame->tf_rbp = (register_t)(td->td_kstack + PAGE_SIZE);
1047 	} else {
1048 		kprintf("DOUBLE FAULT\n");
1049 	}
1050 	kprintf("\nFatal double fault\n");
1051 	kprintf("rip = 0x%lx\n", frame->tf_rip);
1052 	kprintf("rsp = 0x%lx\n", frame->tf_rsp);
1053 	kprintf("rbp = 0x%lx\n", frame->tf_rbp);
1054 #ifdef SMP
1055 	/* three separate prints in case of a trap on an unmapped page */
1056 	kprintf("cpuid = %d; ", mycpu->gd_cpuid);
1057 	kprintf("lapic->id = %08x\n", lapic->id);
1058 #endif
1059 	panic("double fault");
1060 }
1061 
1062 /*
1063  * syscall2 -	MP aware system call request C handler
1064  *
1065  * A system call is essentially treated as a trap except that the
1066  * MP lock is not held on entry or return.  We are responsible for
1067  * obtaining the MP lock if necessary and for handling ASTs
1068  * (e.g. a task switch) prior to return.
1069  *
1070  * MPSAFE
1071  */
1072 void
1073 syscall2(struct trapframe *frame)
1074 {
1075 	struct thread *td = curthread;
1076 	struct proc *p = td->td_proc;
1077 	struct lwp *lp = td->td_lwp;
1078 	caddr_t params;
1079 	struct sysent *callp;
1080 	register_t orig_tf_rflags;
1081 	int sticks;
1082 	int error;
1083 	int narg;
1084 #ifdef INVARIANTS
1085 	int crit_count = td->td_critcount;
1086 #endif
1087 #ifdef SMP
1088 	int have_mplock = 0;
1089 #endif
1090 	register_t *argp;
1091 	u_int code;
1092 	int reg, regcnt;
1093 	union sysunion args;
1094 	register_t *argsdst;
1095 
1096 	mycpu->gd_cnt.v_syscall++;
1097 
1098 #ifdef DIAGNOSTIC
1099 	if (ISPL(frame->tf_cs) != SEL_UPL) {
1100 		get_mplock();
1101 		panic("syscall");
1102 		/* NOT REACHED */
1103 	}
1104 #endif
1105 
1106 	KTR_LOG(kernentry_syscall, p->p_pid, lp->lwp_tid,
1107 		frame->tf_rax);
1108 
1109 	userenter(td, p);	/* lazy raise our priority */
1110 
1111 	reg = 0;
1112 	regcnt = 6;
1113 	/*
1114 	 * Misc
1115 	 */
1116 	sticks = (int)td->td_sticks;
1117 	orig_tf_rflags = frame->tf_rflags;
1118 
1119 	/*
1120 	 * Virtual kernel intercept - if a VM context managed by a virtual
1121 	 * kernel issues a system call the virtual kernel handles it, not us.
1122 	 * Restore the virtual kernel context and return from its system
1123 	 * call.  The current frame is copied out to the virtual kernel.
1124 	 */
1125 	if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
1126 		vkernel_trap(lp, frame);
1127 		error = EJUSTRETURN;
1128 		goto out;
1129 	}
1130 
1131 	/*
1132 	 * Get the system call parameters and account for time
1133 	 */
1134 	KASSERT(lp->lwp_md.md_regs == frame,
1135 		("Frame mismatch %p %p", lp->lwp_md.md_regs, frame));
1136 	params = (caddr_t)frame->tf_rsp + sizeof(register_t);
1137 	code = frame->tf_rax;
1138 
1139 	if (p->p_sysent->sv_prepsyscall) {
1140 		(*p->p_sysent->sv_prepsyscall)(
1141 			frame, (int *)(&args.nosys.sysmsg + 1),
1142 			&code, &params);
1143 	} else {
1144 		if (code == SYS_syscall || code == SYS___syscall) {
1145 			code = frame->tf_rdi;
1146 			reg++;
1147 			regcnt--;
1148 		}
1149 	}
1150 
1151  	if (p->p_sysent->sv_mask)
1152  		code &= p->p_sysent->sv_mask;
1153 
1154 	if (code >= p->p_sysent->sv_size)
1155 		callp = &p->p_sysent->sv_table[0];
1156 	else
1157 		callp = &p->p_sysent->sv_table[code];
1158 
1159 	narg = callp->sy_narg & SYF_ARGMASK;
1160 
1161 	/*
1162 	 * On x86_64 we get up to six arguments in registers. The rest are
1163 	 * on the stack. The first six members of 'struct trapframe' happen
1164 	 * to be the registers used to pass arguments, in exactly the right
1165 	 * order.
1166 	 */
1167 	argp = &frame->tf_rdi;
1168 	argp += reg;
1169 	argsdst = (register_t *)(&args.nosys.sysmsg + 1);
1170 	/*
1171 	 * JG can we overflow the space pointed to by 'argsdst'
1172 	 * either with 'bcopy' or with 'copyin'?
1173 	 */
1174 	bcopy(argp, argsdst, sizeof(register_t) * regcnt);
1175 	/*
1176 	 * copyin is MP aware, but the tracing code is not
1177 	 */
1178 	if (narg > regcnt) {
1179 		KASSERT(params != NULL, ("copyin args with no params!"));
1180 		error = copyin(params, &argsdst[regcnt],
1181 	    		(narg - regcnt) * sizeof(register_t));
1182 		if (error) {
1183 #ifdef KTRACE
1184 			if (KTRPOINT(td, KTR_SYSCALL)) {
1185 				MAKEMPSAFE(have_mplock);
1186 
1187 				ktrsyscall(lp, code, narg,
1188 					(void *)(&args.nosys.sysmsg + 1));
1189 			}
1190 #endif
1191 			goto bad;
1192 		}
1193 	}
1194 
1195 #ifdef KTRACE
1196 	if (KTRPOINT(td, KTR_SYSCALL)) {
1197 		MAKEMPSAFE(have_mplock);
1198 		ktrsyscall(lp, code, narg, (void *)(&args.nosys.sysmsg + 1));
1199 	}
1200 #endif
1201 
1202 	/*
1203 	 * Default return value is 0 (will be copied to %rax).  Double-value
1204 	 * returns use %rax and %rdx.  %rdx is left unchanged for system
1205 	 * calls which return only one result.
1206 	 */
1207 	args.sysmsg_fds[0] = 0;
1208 	args.sysmsg_fds[1] = frame->tf_rdx;
1209 
1210 	/*
1211 	 * The syscall might manipulate the trap frame. If it does it
1212 	 * will probably return EJUSTRETURN.
1213 	 */
1214 	args.sysmsg_frame = frame;
1215 
1216 	STOPEVENT(p, S_SCE, narg);	/* MP aware */
1217 
1218 	/*
1219 	 * NOTE: All system calls run MPSAFE now.  The system call itself
1220 	 *	 is responsible for getting the MP lock.
1221 	 */
1222 #ifdef SYSCALL_DEBUG
1223 	uint64_t tscval = rdtsc();
1224 #endif
1225 	error = (*callp->sy_call)(&args);
1226 #ifdef SYSCALL_DEBUG
1227 	tscval = rdtsc() - tscval;
1228 	tscval = tscval * 1000000 / tsc_frequency;
1229 	if (SysCallsWorstCase[code] < tscval)
1230 		SysCallsWorstCase[code] = tscval;
1231 #endif
1232 
1233 out:
1234 	/*
1235 	 * MP SAFE (we may or may not have the MP lock at this point)
1236 	 */
1237 	//kprintf("SYSMSG %d ", error);
1238 	switch (error) {
1239 	case 0:
1240 		/*
1241 		 * Reinitialize proc pointer `p' as it may be different
1242 		 * if this is a child returning from fork syscall.
1243 		 */
1244 		p = curproc;
1245 		lp = curthread->td_lwp;
1246 		frame->tf_rax = args.sysmsg_fds[0];
1247 		frame->tf_rdx = args.sysmsg_fds[1];
1248 		frame->tf_rflags &= ~PSL_C;
1249 		break;
1250 	case ERESTART:
1251 		/*
1252 		 * Reconstruct pc, we know that 'syscall' is 2 bytes.
1253 		 * We have to do a full context restore so that %r10
1254 		 * (which was holding the value of %rcx) is restored for
1255 		 * the next iteration.
1256 		 */
1257 		frame->tf_rip -= frame->tf_err;
1258 		frame->tf_r10 = frame->tf_rcx;
1259 		break;
1260 	case EJUSTRETURN:
1261 		break;
1262 	case EASYNC:
1263 		panic("Unexpected EASYNC return value (for now)");
1264 	default:
1265 bad:
1266 		if (p->p_sysent->sv_errsize) {
1267 			if (error >= p->p_sysent->sv_errsize)
1268 				error = -1;	/* XXX */
1269 			else
1270 				error = p->p_sysent->sv_errtbl[error];
1271 		}
1272 		frame->tf_rax = error;
1273 		frame->tf_rflags |= PSL_C;
1274 		break;
1275 	}
1276 
1277 	/*
1278 	 * Traced syscall.  trapsignal() is not MP aware.
1279 	 */
1280 	if (orig_tf_rflags & PSL_T) {
1281 		MAKEMPSAFE(have_mplock);
1282 		frame->tf_rflags &= ~PSL_T;
1283 		trapsignal(lp, SIGTRAP, TRAP_TRACE);
1284 	}
1285 
1286 	/*
1287 	 * Handle reschedule and other end-of-syscall issues
1288 	 */
1289 	userret(lp, frame, sticks);
1290 
1291 #ifdef KTRACE
1292 	if (KTRPOINT(td, KTR_SYSRET)) {
1293 		MAKEMPSAFE(have_mplock);
1294 		ktrsysret(lp, code, error, args.sysmsg_result);
1295 	}
1296 #endif
1297 
1298 	/*
1299 	 * This works because errno is findable through the
1300 	 * register set.  If we ever support an emulation where this
1301 	 * is not the case, this code will need to be revisited.
1302 	 */
1303 	STOPEVENT(p, S_SCX, code);
1304 
1305 	userexit(lp);
1306 #ifdef SMP
1307 	/*
1308 	 * Release the MP lock if we had to get it
1309 	 */
1310 	if (have_mplock)
1311 		rel_mplock();
1312 #endif
1313 	KTR_LOG(kernentry_syscall_ret, p->p_pid, lp->lwp_tid, error);
1314 #ifdef INVARIANTS
1315 	KASSERT(crit_count == td->td_critcount,
1316 		("syscall: critical section count mismatch! %d/%d",
1317 		crit_count, td->td_pri));
1318 	KASSERT(&td->td_toks_base == td->td_toks_stop,
1319 		("syscall: extra tokens held after trap! %ld",
1320 		td->td_toks_stop - &td->td_toks_base));
1321 #endif
1322 }
1323 
1324 /*
1325  * NOTE: mplock not held at any point
1326  */
1327 void
1328 fork_return(struct lwp *lp, struct trapframe *frame)
1329 {
1330 	frame->tf_rax = 0;		/* Child returns zero */
1331 	frame->tf_rflags &= ~PSL_C;	/* success */
1332 	frame->tf_rdx = 1;
1333 
1334 	generic_lwp_return(lp, frame);
1335 	KTR_LOG(kernentry_fork_ret, lp->lwp_proc->p_pid, lp->lwp_tid);
1336 }
1337 
1338 /*
1339  * Simplified back end of syscall(), used when returning from fork()
1340  * directly into user mode.
1341  *
1342  * This code will return back into the fork trampoline code which then
1343  * runs doreti.
1344  *
1345  * NOTE: The mplock is not held at any point.
1346  */
1347 void
1348 generic_lwp_return(struct lwp *lp, struct trapframe *frame)
1349 {
1350 	struct proc *p = lp->lwp_proc;
1351 
1352 	/*
1353 	 * Newly forked processes are given a kernel priority.  We have to
1354 	 * adjust the priority to a normal user priority and fake entry
1355 	 * into the kernel (call userenter()) to install a passive release
1356 	 * function just in case userret() decides to stop the process.  This
1357 	 * can occur when ^Z races a fork.  If we do not install the passive
1358 	 * release function the current process designation will not be
1359 	 * released when the thread goes to sleep.
1360 	 */
1361 	lwkt_setpri_self(TDPRI_USER_NORM);
1362 	userenter(lp->lwp_thread, p);
1363 	userret(lp, frame, 0);
1364 #ifdef KTRACE
1365 	if (KTRPOINT(lp->lwp_thread, KTR_SYSRET))
1366 		ktrsysret(lp, SYS_fork, 0, 0);
1367 #endif
1368 	lp->lwp_flag |= LWP_PASSIVE_ACQ;
1369 	userexit(lp);
1370 	lp->lwp_flag &= ~LWP_PASSIVE_ACQ;
1371 }
1372 
1373 /*
1374  * If PGEX_FPFAULT is set then set FP_VIRTFP in the PCB to force a T_DNA
1375  * fault (which is then passed back to the virtual kernel) if an attempt is
1376  * made to use the FP unit.
1377  *
1378  * XXX this is a fairly big hack.
1379  */
1380 void
1381 set_vkernel_fp(struct trapframe *frame)
1382 {
1383 	struct thread *td = curthread;
1384 
1385 	if (frame->tf_xflags & PGEX_FPFAULT) {
1386 		td->td_pcb->pcb_flags |= FP_VIRTFP;
1387 		if (mdcpu->gd_npxthread == td)
1388 			npxexit();
1389 	} else {
1390 		td->td_pcb->pcb_flags &= ~FP_VIRTFP;
1391 	}
1392 }
1393 
1394 /*
1395  * Called from vkernel_trap() to fixup the vkernel's syscall
1396  * frame for vmspace_ctl() return.
1397  */
1398 void
1399 cpu_vkernel_trap(struct trapframe *frame, int error)
1400 {
1401 	frame->tf_rax = error;
1402 	if (error)
1403 		frame->tf_rflags |= PSL_C;
1404 	else
1405 		frame->tf_rflags &= ~PSL_C;
1406 }
1407