kernel: Make SMP support default (and non-optional).
[dragonfly.git] / sys / platform / vkernel64 / x86_64 / trap.c
CommitLineData
da673940
JG
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: src/sys/i386/i386/trap.c,v 1.147.2.11 2003/02/27 19:09:59 luoqi Exp $
da673940
JG
39 */
40
41/*
a76ca9b9 42 * x86_64 Trap and System call handling
da673940
JG
43 */
44
45#include "use_isa.h"
46
47#include "opt_ddb.h"
48#include "opt_ktrace.h"
49
50#include <sys/param.h>
51#include <sys/systm.h>
52#include <sys/proc.h>
53#include <sys/pioctl.h>
54#include <sys/kernel.h>
55#include <sys/resourcevar.h>
56#include <sys/signalvar.h>
57#include <sys/signal2.h>
58#include <sys/syscall.h>
59#include <sys/sysctl.h>
60#include <sys/sysent.h>
61#include <sys/uio.h>
62#include <sys/vmmeter.h>
63#include <sys/malloc.h>
64#ifdef KTRACE
65#include <sys/ktrace.h>
66#endif
67#include <sys/ktr.h>
68#include <sys/upcall.h>
69#include <sys/vkernel.h>
70#include <sys/sysproto.h>
71#include <sys/sysunion.h>
72#include <sys/vmspace.h>
73
74#include <vm/vm.h>
75#include <vm/vm_param.h>
76#include <sys/lock.h>
77#include <vm/pmap.h>
78#include <vm/vm_kern.h>
79#include <vm/vm_map.h>
80#include <vm/vm_page.h>
81#include <vm/vm_extern.h>
82
83#include <machine/cpu.h>
84#include <machine/md_var.h>
85#include <machine/pcb.h>
86#include <machine/smp.h>
87#include <machine/tss.h>
88#include <machine/globaldata.h>
89
da673940 90#include <ddb/ddb.h>
0e6594a8 91
da673940
JG
92#include <sys/msgport2.h>
93#include <sys/thread2.h>
0e6594a8 94#include <sys/mplock2.h>
da673940 95
da673940
JG
96#define MAKEMPSAFE(have_mplock) \
97 if (have_mplock == 0) { \
98 get_mplock(); \
99 have_mplock = 1; \
100 }
101
da673940
JG
102int (*pmath_emulate) (struct trapframe *);
103
104extern int trapwrite (unsigned addr);
105
106static int trap_pfault (struct trapframe *, int, vm_offset_t);
107static void trap_fatal (struct trapframe *, int, vm_offset_t);
108void dblfault_handler (void);
109
110#if 0
111extern inthand_t IDTVEC(syscall);
112#endif
113
114#define MAX_TRAP_MSG 30
115static char *trap_msg[] = {
116 "", /* 0 unused */
117 "privileged instruction fault", /* 1 T_PRIVINFLT */
118 "", /* 2 unused */
119 "breakpoint instruction fault", /* 3 T_BPTFLT */
120 "", /* 4 unused */
121 "", /* 5 unused */
122 "arithmetic trap", /* 6 T_ARITHTRAP */
123 "system forced exception", /* 7 T_ASTFLT */
124 "", /* 8 unused */
125 "general protection fault", /* 9 T_PROTFLT */
126 "trace trap", /* 10 T_TRCTRAP */
127 "", /* 11 unused */
128 "page fault", /* 12 T_PAGEFLT */
129 "", /* 13 unused */
130 "alignment fault", /* 14 T_ALIGNFLT */
131 "", /* 15 unused */
132 "", /* 16 unused */
133 "", /* 17 unused */
134 "integer divide fault", /* 18 T_DIVIDE */
135 "non-maskable interrupt trap", /* 19 T_NMI */
136 "overflow trap", /* 20 T_OFLOW */
137 "FPU bounds check fault", /* 21 T_BOUND */
138 "FPU device not available", /* 22 T_DNA */
139 "double fault", /* 23 T_DOUBLEFLT */
140 "FPU operand fetch fault", /* 24 T_FPOPFLT */
141 "invalid TSS fault", /* 25 T_TSSFLT */
142 "segment not present fault", /* 26 T_SEGNPFLT */
143 "stack fault", /* 27 T_STKFLT */
144 "machine check trap", /* 28 T_MCHK */
145 "SIMD floating-point exception", /* 29 T_XMMFLT */
146 "reserved (unknown) fault", /* 30 T_RESERVED */
147};
148
149#ifdef DDB
150static int ddb_on_nmi = 1;
151SYSCTL_INT(_machdep, OID_AUTO, ddb_on_nmi, CTLFLAG_RW,
152 &ddb_on_nmi, 0, "Go to DDB on NMI");
153#endif
154static int panic_on_nmi = 1;
155SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RW,
156 &panic_on_nmi, 0, "Panic on NMI");
157static int fast_release;
158SYSCTL_INT(_machdep, OID_AUTO, fast_release, CTLFLAG_RW,
159 &fast_release, 0, "Passive Release was optimal");
160static int slow_release;
161SYSCTL_INT(_machdep, OID_AUTO, slow_release, CTLFLAG_RW,
162 &slow_release, 0, "Passive Release was nonoptimal");
da673940
JG
163
164MALLOC_DEFINE(M_SYSMSG, "sysmsg", "sysmsg structure");
165extern int max_sysmsg;
166
167/*
bab69519
JM
168 * Passively intercepts the thread switch function to increase
169 * the thread priority from a user priority to a kernel priority, reducing
da673940 170 * syscall and trap overhead for the case where no switch occurs.
73e24181
MD
171 *
172 * Synchronizes td_ucred with p_ucred. This is used by system calls,
173 * signal handling, faults, AST traps, and anything else that enters the
174 * kernel from userland and provides the kernel with a stable read-only
175 * copy of the process ucred.
da673940 176 */
da673940 177static __inline void
73e24181 178userenter(struct thread *curtd, struct proc *curp)
da673940 179{
73e24181
MD
180 struct ucred *ocred;
181 struct ucred *ncred;
182
da673940 183 curtd->td_release = lwkt_passive_release;
73e24181
MD
184
185 if (curtd->td_ucred != curp->p_ucred) {
186 ncred = crhold(curp->p_ucred);
187 ocred = curtd->td_ucred;
188 curtd->td_ucred = ncred;
189 if (ocred)
190 crfree(ocred);
191 }
da673940
JG
192}
193
194/*
195 * Handle signals, upcalls, profiling, and other AST's and/or tasks that
196 * must be completed before we can return to or try to return to userland.
197 *
198 * Note that td_sticks is a 64 bit quantity, but there's no point doing 64
199 * arithmatic on the delta calculation so the absolute tick values are
200 * truncated to an integer.
201 */
202static void
203userret(struct lwp *lp, struct trapframe *frame, int sticks)
204{
205 struct proc *p = lp->lwp_proc;
206 int sig;
207
208 /*
209 * Charge system time if profiling. Note: times are in microseconds.
210 * This may do a copyout and block, so do it first even though it
211 * means some system time will be charged as user time.
212 */
4643740a 213 if (p->p_flags & P_PROFIL) {
da673940
JG
214 addupc_task(p, frame->tf_rip,
215 (u_int)((int)lp->lwp_thread->td_sticks - sticks));
216 }
217
218recheck:
219 /*
220 * If the jungle wants us dead, so be it.
221 */
4643740a 222 if (lp->lwp_mpflags & LWP_MP_WEXIT) {
b5c4d81f 223 lwkt_gettoken(&p->p_token);
da673940 224 lwp_exit(0);
b5c4d81f 225 lwkt_reltoken(&p->p_token); /* NOT REACHED */
da673940
JG
226 }
227
228 /*
229 * Block here if we are in a stopped state.
230 */
231 if (p->p_stat == SSTOP) {
4643740a 232 lwkt_gettoken(&p->p_token);
da673940 233 tstop();
4643740a 234 lwkt_reltoken(&p->p_token);
da673940
JG
235 goto recheck;
236 }
237
238 /*
898e34b3
MD
239 * Post any pending upcalls. If running a virtual kernel be sure
240 * to restore the virtual kernel's vmspace before posting the upcall.
da673940 241 */
4643740a 242 if (p->p_flags & (P_SIGVTALRM | P_SIGPROF | P_UPCALLPEND)) {
fc509460 243 lwkt_gettoken(&p->p_token);
4643740a
MD
244 if (p->p_flags & P_SIGVTALRM) {
245 p->p_flags &= ~P_SIGVTALRM;
898e34b3
MD
246 ksignal(p, SIGVTALRM);
247 }
4643740a
MD
248 if (p->p_flags & P_SIGPROF) {
249 p->p_flags &= ~P_SIGPROF;
898e34b3
MD
250 ksignal(p, SIGPROF);
251 }
4643740a
MD
252 if (p->p_flags & P_UPCALLPEND) {
253 p->p_flags &= ~P_UPCALLPEND;
898e34b3
MD
254 postupcall(lp);
255 }
fc509460 256 lwkt_reltoken(&p->p_token);
da673940
JG
257 goto recheck;
258 }
259
260 /*
261 * Post any pending signals
2883d2d8
MD
262 *
263 * WARNING! postsig() can exit and not return.
da673940
JG
264 */
265 if ((sig = CURSIG_TRACE(lp)) != 0) {
fc509460 266 lwkt_gettoken(&p->p_token);
da673940 267 postsig(sig);
fc509460 268 lwkt_reltoken(&p->p_token);
da673940
JG
269 goto recheck;
270 }
271
272 /*
273 * block here if we are swapped out, but still process signals
274 * (such as SIGKILL). proc0 (the swapin scheduler) is already
275 * aware of our situation, we do not have to wake it up.
276 */
4643740a 277 if (p->p_flags & P_SWAPPEDOUT) {
616516c8 278 lwkt_gettoken(&p->p_token);
da673940 279 get_mplock();
4643740a 280 p->p_flags |= P_SWAPWAIT;
da673940 281 swapin_request();
4643740a 282 if (p->p_flags & P_SWAPWAIT)
da673940 283 tsleep(p, PCATCH, "SWOUT", 0);
4643740a 284 p->p_flags &= ~P_SWAPWAIT;
da673940 285 rel_mplock();
616516c8 286 lwkt_reltoken(&p->p_token);
da673940
JG
287 goto recheck;
288 }
289
290 /*
6562e2d8
MD
291 * In a multi-threaded program it is possible for a thread to change
292 * signal state during a system call which temporarily changes the
293 * signal mask. In this case postsig() might not be run and we
294 * have to restore the mask ourselves.
da673940 295 */
6562e2d8
MD
296 if (lp->lwp_flags & LWP_OLDMASK) {
297 lp->lwp_flags &= ~LWP_OLDMASK;
298 lp->lwp_sigmask = lp->lwp_oldsigmask;
299 goto recheck;
300 }
da673940
JG
301}
302
303/*
304 * Cleanup from userenter and any passive release that might have occured.
305 * We must reclaim the current-process designation before we can return
306 * to usermode. We also handle both LWKT and USER reschedule requests.
307 */
308static __inline void
309userexit(struct lwp *lp)
310{
311 struct thread *td = lp->lwp_thread;
312 /* globaldata_t gd = td->td_gd; */
313
314 /*
315 * Handle stop requests at kernel priority. Any requests queued
316 * after this loop will generate another AST.
317 */
318 while (lp->lwp_proc->p_stat == SSTOP) {
4643740a 319 lwkt_gettoken(&lp->lwp_proc->p_token);
da673940 320 tstop();
4643740a 321 lwkt_reltoken(&lp->lwp_proc->p_token);
da673940
JG
322 }
323
324 /*
325 * Reduce our priority in preparation for a return to userland. If
326 * our passive release function was still in place, our priority was
327 * never raised and does not need to be reduced.
328 */
329 lwkt_passive_recover(td);
330
331 /*
332 * Become the current user scheduled process if we aren't already,
333 * and deal with reschedule requests and other factors.
334 */
335 lp->lwp_proc->p_usched->acquire_curproc(lp);
336 /* WARNING: we may have migrated cpu's */
337 /* gd = td->td_gd; */
338}
339
340#if !defined(KTR_KERNENTRY)
341#define KTR_KERNENTRY KTR_ALL
342#endif
343KTR_INFO_MASTER(kernentry);
b19d1a6b
SW
344KTR_INFO(KTR_KERNENTRY, kernentry, trap, 0,
345 "TRAP(pid %hd, tid %hd, trapno %ld, eva %lu)",
346 pid_t pid, lwpid_t tid, register_t trapno, vm_offset_t eva);
347KTR_INFO(KTR_KERNENTRY, kernentry, trap_ret, 0, "TRAP_RET(pid %hd, tid %hd)",
348 pid_t pid, lwpid_t tid);
349KTR_INFO(KTR_KERNENTRY, kernentry, syscall, 0, "SYSC(pid %hd, tid %hd, nr %ld)",
350 pid_t pid, lwpid_t tid, register_t trapno);
351KTR_INFO(KTR_KERNENTRY, kernentry, syscall_ret, 0, "SYSRET(pid %hd, tid %hd, err %d)",
352 pid_t pid, lwpid_t tid, int err);
353KTR_INFO(KTR_KERNENTRY, kernentry, fork_ret, 0, "FORKRET(pid %hd, tid %hd)",
354 pid_t pid, lwpid_t tid);
da673940
JG
355
356/*
357 * Exception, fault, and trap interface to the kernel.
358 * This common code is called from assembly language IDT gate entry
359 * routines that prepare a suitable stack frame, and restore this
360 * frame after the exception has been processed.
361 *
362 * This function is also called from doreti in an interlock to handle ASTs.
363 * For example: hardwareint->INTROUTINE->(set ast)->doreti->trap
364 *
365 * NOTE! We have to retrieve the fault address prior to obtaining the
366 * MP lock because get_mplock() may switch out. YYY cr2 really ought
367 * to be retrieved by the assembly code, not here.
368 *
369 * XXX gd_trap_nesting_level currently prevents lwkt_switch() from panicing
370 * if an attempt is made to switch from a fast interrupt or IPI. This is
371 * necessary to properly take fatal kernel traps on SMP machines if
372 * get_mplock() has to block.
373 */
374
375void
376user_trap(struct trapframe *frame)
377{
378 struct globaldata *gd = mycpu;
379 struct thread *td = gd->gd_curthread;
380 struct lwp *lp = td->td_lwp;
381 struct proc *p;
382 int sticks = 0;
383 int i = 0, ucode = 0, type, code;
da673940 384 int have_mplock = 0;
da673940 385#ifdef INVARIANTS
f9235b6d 386 int crit_count = td->td_critcount;
3933a3ab 387 lwkt_tokref_t curstop = td->td_toks_stop;
da673940
JG
388#endif
389 vm_offset_t eva;
390
391 p = td->td_proc;
392
393 if (frame->tf_trapno == T_PAGEFLT)
394 eva = frame->tf_addr;
395 else
396 eva = 0;
397#if 0
398 kprintf("USER_TRAP AT %08lx xflags %ld trapno %ld eva %08lx\n",
399 frame->tf_rip, frame->tf_xflags, frame->tf_trapno, eva);
400#endif
401
402 /*
403 * Everything coming from user mode runs through user_trap,
404 * including system calls.
405 */
571cdd83 406 if (frame->tf_trapno == T_FAST_SYSCALL) {
da673940
JG
407 syscall2(frame);
408 return;
409 }
410
411 KTR_LOG(kernentry_trap, lp->lwp_proc->p_pid, lp->lwp_tid,
412 frame->tf_trapno, eva);
413
414#ifdef DDB
415 if (db_active) {
416 eva = (frame->tf_trapno == T_PAGEFLT ? rcr2() : 0);
417 ++gd->gd_trap_nesting_level;
418 MAKEMPSAFE(have_mplock);
419 trap_fatal(frame, TRUE, eva);
420 --gd->gd_trap_nesting_level;
421 goto out2;
422 }
423#endif
424
da673940
JG
425 type = frame->tf_trapno;
426 code = frame->tf_err;
427
73e24181 428 userenter(td, p);
da673940
JG
429
430 sticks = (int)td->td_sticks;
431 lp->lwp_md.md_regs = frame;
432
433 switch (type) {
434 case T_PRIVINFLT: /* privileged instruction fault */
da673940 435 i = SIGILL;
bab69519 436 ucode = ILL_PRVOPC;
da673940
JG
437 break;
438
439 case T_BPTFLT: /* bpt instruction fault */
440 case T_TRCTRAP: /* trace trap */
441 frame->tf_rflags &= ~PSL_T;
442 i = SIGTRAP;
bab69519 443 ucode = (type == T_TRCTRAP ? TRAP_TRACE : TRAP_BRKPT);
da673940
JG
444 break;
445
446 case T_ARITHTRAP: /* arithmetic trap */
447 ucode = code;
448 i = SIGFPE;
449 break;
450
451 case T_ASTFLT: /* Allow process switch */
452 mycpu->gd_cnt.v_soft++;
453 if (mycpu->gd_reqflags & RQF_AST_OWEUPC) {
2a418930
MD
454 atomic_clear_int(&mycpu->gd_reqflags, RQF_AST_OWEUPC);
455 addupc_task(p, p->p_prof.pr_addr, p->p_prof.pr_ticks);
da673940
JG
456 }
457 goto out;
458
459 /*
460 * The following two traps can happen in
461 * vm86 mode, and, if so, we want to handle
462 * them specially.
463 */
464 case T_PROTFLT: /* general protection fault */
465 case T_STKFLT: /* stack fault */
466#if 0
467 if (frame->tf_eflags & PSL_VM) {
468 i = vm86_emulate((struct vm86frame *)frame);
469 if (i == 0)
470 goto out;
471 break;
472 }
473#endif
474 /* FALL THROUGH */
475
476 case T_SEGNPFLT: /* segment not present fault */
477 case T_TSSFLT: /* invalid TSS fault */
478 case T_DOUBLEFLT: /* double fault */
479 default:
da673940 480 i = SIGBUS;
bab69519 481 ucode = code + BUS_SEGM_FAULT ;
da673940
JG
482 break;
483
484 case T_PAGEFLT: /* page fault */
485 MAKEMPSAFE(have_mplock);
486 i = trap_pfault(frame, TRUE, eva);
fbff57d1 487 if (i == -1 || i == 0)
da673940
JG
488 goto out;
489
bab69519
JM
490
491 if (i == SIGSEGV)
492 ucode = SEGV_MAPERR;
493 else {
494 i = SIGSEGV;
495 ucode = SEGV_ACCERR;
496 }
da673940
JG
497 break;
498
499 case T_DIVIDE: /* integer divide fault */
500 ucode = FPE_INTDIV;
501 i = SIGFPE;
502 break;
503
504#if NISA > 0
505 case T_NMI:
506 MAKEMPSAFE(have_mplock);
507 /* machine/parity/power fail/"kitchen sink" faults */
508 if (isa_nmi(code) == 0) {
509#ifdef DDB
510 /*
511 * NMI can be hooked up to a pushbutton
512 * for debugging.
513 */
514 if (ddb_on_nmi) {
515 kprintf ("NMI ... going to debugger\n");
bab69519 516 kdb_trap(type, 0, frame);
da673940
JG
517 }
518#endif /* DDB */
519 goto out2;
520 } else if (panic_on_nmi)
521 panic("NMI indicates hardware failure");
522 break;
523#endif /* NISA > 0 */
524
525 case T_OFLOW: /* integer overflow fault */
526 ucode = FPE_INTOVF;
527 i = SIGFPE;
528 break;
529
530 case T_BOUND: /* bounds check fault */
531 ucode = FPE_FLTSUB;
532 i = SIGFPE;
533 break;
534
535 case T_DNA:
536 /*
537 * Virtual kernel intercept - pass the DNA exception
538 * to the (emulated) virtual kernel if it asked to handle
539 * it. This occurs when the virtual kernel is holding
540 * onto the FP context for a different emulated
541 * process then the one currently running.
542 *
543 * We must still call npxdna() since we may have
544 * saved FP state that the (emulated) virtual kernel
545 * needs to hand over to a different emulated process.
546 */
547 if (lp->lwp_vkernel && lp->lwp_vkernel->ve &&
548 (td->td_pcb->pcb_flags & FP_VIRTFP)
549 ) {
550 npxdna(frame);
551 break;
552 }
bab69519 553
da673940
JG
554 /*
555 * The kernel may have switched out the FP unit's
556 * state, causing the user process to take a fault
557 * when it tries to use the FP unit. Restore the
558 * state here
559 */
560 if (npxdna(frame))
561 goto out;
562 if (!pmath_emulate) {
563 i = SIGFPE;
564 ucode = FPE_FPU_NP_TRAP;
565 break;
566 }
567 i = (*pmath_emulate)(frame);
568 if (i == 0) {
569 if (!(frame->tf_rflags & PSL_T))
570 goto out2;
571 frame->tf_rflags &= ~PSL_T;
572 i = SIGTRAP;
573 }
574 /* else ucode = emulator_only_knows() XXX */
575 break;
576
577 case T_FPOPFLT: /* FPU operand fetch fault */
578 ucode = T_FPOPFLT;
579 i = SIGILL;
580 break;
581
582 case T_XMMFLT: /* SIMD floating-point exception */
583 ucode = 0; /* XXX */
584 i = SIGFPE;
585 break;
586 }
587
588 /*
589 * Virtual kernel intercept - if the fault is directly related to a
590 * VM context managed by a virtual kernel then let the virtual kernel
591 * handle it.
592 */
593 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
594 vkernel_trap(lp, frame);
595 goto out;
596 }
597
598 /*
599 * Translate fault for emulators (e.g. Linux)
600 */
601 if (*p->p_sysent->sv_transtrap)
602 i = (*p->p_sysent->sv_transtrap)(i, type);
603
604 MAKEMPSAFE(have_mplock);
605 trapsignal(lp, i, ucode);
606
607#ifdef DEBUG
608 if (type <= MAX_TRAP_MSG) {
609 uprintf("fatal process exception: %s",
610 trap_msg[type]);
611 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
612 uprintf(", fault VA = 0x%lx", (u_long)eva);
613 uprintf("\n");
614 }
615#endif
616
617out:
da673940
JG
618 userret(lp, frame, sticks);
619 userexit(lp);
620out2: ;
da673940
JG
621 if (have_mplock)
622 rel_mplock();
da673940
JG
623 KTR_LOG(kernentry_trap_ret, lp->lwp_proc->p_pid, lp->lwp_tid);
624#ifdef INVARIANTS
f9235b6d 625 KASSERT(crit_count == td->td_critcount,
3933a3ab 626 ("trap: critical section count mismatch! %d/%d",
f9235b6d 627 crit_count, td->td_pri));
3933a3ab
MD
628 KASSERT(curstop == td->td_toks_stop,
629 ("trap: extra tokens held after trap! %ld/%ld",
630 curstop - &td->td_toks_base,
631 td->td_toks_stop - &td->td_toks_base));
da673940
JG
632#endif
633}
634
635void
636kern_trap(struct trapframe *frame)
637{
638 struct globaldata *gd = mycpu;
639 struct thread *td = gd->gd_curthread;
640 struct lwp *lp;
641 struct proc *p;
642 int i = 0, ucode = 0, type, code;
da673940 643 int have_mplock = 0;
da673940 644#ifdef INVARIANTS
f9235b6d 645 int crit_count = td->td_critcount;
3933a3ab 646 lwkt_tokref_t curstop = td->td_toks_stop;
da673940
JG
647#endif
648 vm_offset_t eva;
649
650 lp = td->td_lwp;
651 p = td->td_proc;
652
653 if (frame->tf_trapno == T_PAGEFLT)
654 eva = frame->tf_addr;
655 else
656 eva = 0;
657
658#ifdef DDB
659 if (db_active) {
660 ++gd->gd_trap_nesting_level;
661 MAKEMPSAFE(have_mplock);
662 trap_fatal(frame, FALSE, eva);
663 --gd->gd_trap_nesting_level;
664 goto out2;
665 }
666#endif
667
da673940
JG
668 type = frame->tf_trapno;
669 code = frame->tf_err;
670
671#if 0
672kernel_trap:
673#endif
674 /* kernel trap */
675
676 switch (type) {
677 case T_PAGEFLT: /* page fault */
678 MAKEMPSAFE(have_mplock);
679 trap_pfault(frame, FALSE, eva);
680 goto out2;
681
682 case T_DNA:
683 /*
684 * The kernel may be using npx for copying or other
685 * purposes.
686 */
687 panic("kernel NPX should not happen");
688 if (npxdna(frame))
689 goto out2;
690 break;
691
692 case T_PROTFLT: /* general protection fault */
693 case T_SEGNPFLT: /* segment not present fault */
694 /*
695 * Invalid segment selectors and out of bounds
696 * %eip's and %esp's can be set up in user mode.
697 * This causes a fault in kernel mode when the
698 * kernel tries to return to user mode. We want
699 * to get this fault so that we can fix the
700 * problem here and not have to check all the
701 * selectors and pointers when the user changes
702 * them.
703 */
704 if (mycpu->gd_intr_nesting_level == 0) {
705 if (td->td_pcb->pcb_onfault) {
706 frame->tf_rip =
707 (register_t)td->td_pcb->pcb_onfault;
708 goto out2;
709 }
710 }
711 break;
712
713 case T_TSSFLT:
714 /*
715 * PSL_NT can be set in user mode and isn't cleared
716 * automatically when the kernel is entered. This
717 * causes a TSS fault when the kernel attempts to
718 * `iret' because the TSS link is uninitialized. We
719 * want to get this fault so that we can fix the
720 * problem here and not every time the kernel is
721 * entered.
722 */
723 if (frame->tf_rflags & PSL_NT) {
724 frame->tf_rflags &= ~PSL_NT;
725 goto out2;
726 }
727 break;
728
729 case T_TRCTRAP: /* trace trap */
730#if 0
731 if (frame->tf_eip == (int)IDTVEC(syscall)) {
732 /*
733 * We've just entered system mode via the
734 * syscall lcall. Continue single stepping
735 * silently until the syscall handler has
736 * saved the flags.
737 */
738 goto out2;
739 }
740 if (frame->tf_eip == (int)IDTVEC(syscall) + 1) {
741 /*
742 * The syscall handler has now saved the
743 * flags. Stop single stepping it.
744 */
745 frame->tf_eflags &= ~PSL_T;
746 goto out2;
747 }
748#endif
749#if 0
750 /*
751 * Ignore debug register trace traps due to
752 * accesses in the user's address space, which
753 * can happen under several conditions such as
754 * if a user sets a watchpoint on a buffer and
755 * then passes that buffer to a system call.
756 * We still want to get TRCTRAPS for addresses
757 * in kernel space because that is useful when
758 * debugging the kernel.
759 */
760 if (user_dbreg_trap()) {
761 /*
762 * Reset breakpoint bits because the
763 * processor doesn't
764 */
765 load_dr6(rdr6() & 0xfffffff0);
766 goto out2;
767 }
768#endif
769 /*
770 * Fall through (TRCTRAP kernel mode, kernel address)
771 */
772 case T_BPTFLT:
773 /*
774 * If DDB is enabled, let it handle the debugger trap.
775 * Otherwise, debugger traps "can't happen".
776 */
777#ifdef DDB
778 MAKEMPSAFE(have_mplock);
779 if (kdb_trap (type, 0, frame))
780 goto out2;
781#endif
782 break;
783 case T_DIVIDE:
784 MAKEMPSAFE(have_mplock);
785 trap_fatal(frame, FALSE, eva);
786 goto out2;
787 case T_NMI:
788 MAKEMPSAFE(have_mplock);
789 trap_fatal(frame, FALSE, eva);
790 goto out2;
b9d01986
MD
791 case T_SYSCALL80:
792 case T_FAST_SYSCALL:
da673940
JG
793 /*
794 * Ignore this trap generated from a spurious SIGTRAP.
795 *
796 * single stepping in / syscalls leads to spurious / SIGTRAP
797 * so ignore
798 *
799 * Haiku (c) 2007 Simon 'corecode' Schubert
800 */
801 goto out2;
802 }
803
804 /*
805 * Translate fault for emulators (e.g. Linux)
806 */
807 if (*p->p_sysent->sv_transtrap)
808 i = (*p->p_sysent->sv_transtrap)(i, type);
809
810 MAKEMPSAFE(have_mplock);
811 trapsignal(lp, i, ucode);
812
813#ifdef DEBUG
814 if (type <= MAX_TRAP_MSG) {
815 uprintf("fatal process exception: %s",
816 trap_msg[type]);
817 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
818 uprintf(", fault VA = 0x%lx", (u_long)eva);
819 uprintf("\n");
820 }
821#endif
822
823out2:
824 ;
da673940
JG
825 if (have_mplock)
826 rel_mplock();
da673940 827#ifdef INVARIANTS
f9235b6d 828 KASSERT(crit_count == td->td_critcount,
3933a3ab 829 ("trap: critical section count mismatch! %d/%d",
f9235b6d 830 crit_count, td->td_pri));
3933a3ab
MD
831 KASSERT(curstop == td->td_toks_stop,
832 ("trap: extra tokens held after trap! %ld/%ld",
833 curstop - &td->td_toks_base,
834 td->td_toks_stop - &td->td_toks_base));
da673940
JG
835#endif
836}
837
838int
839trap_pfault(struct trapframe *frame, int usermode, vm_offset_t eva)
840{
841 vm_offset_t va;
842 struct vmspace *vm = NULL;
843 vm_map_t map = 0;
844 int rv = 0;
845 vm_prot_t ftype;
846 thread_t td = curthread;
847 struct lwp *lp = td->td_lwp;
f8a5cc8f 848 int fault_flags;
da673940
JG
849
850 va = trunc_page(eva);
851 if (usermode == FALSE) {
852 /*
853 * This is a fault on kernel virtual memory.
854 */
855 map = &kernel_map;
856 } else {
857 /*
858 * This is a fault on non-kernel virtual memory.
859 * vm is initialized above to NULL. If curproc is NULL
860 * or curproc->p_vmspace is NULL the fault is fatal.
861 */
862 if (lp != NULL)
863 vm = lp->lwp_vmspace;
864
865 if (vm == NULL)
866 goto nogo;
867
868 map = &vm->vm_map;
869 }
870
871 if (frame->tf_err & PGEX_W)
872 ftype = VM_PROT_READ | VM_PROT_WRITE;
873 else
874 ftype = VM_PROT_READ;
875
876 if (map != &kernel_map) {
877 /*
878 * Keep swapout from messing with us during this
879 * critical time.
880 */
881 PHOLD(lp->lwp_proc);
882
883 /*
884 * Grow the stack if necessary
885 */
886 /* grow_stack returns false only if va falls into
887 * a growable stack region and the stack growth
888 * fails. It returns true if va was not within
889 * a growable stack region, or if the stack
890 * growth succeeded.
891 */
892 if (!grow_stack (lp->lwp_proc, va)) {
893 rv = KERN_FAILURE;
894 PRELE(lp->lwp_proc);
895 goto nogo;
896 }
897
f8a5cc8f
VS
898 fault_flags = 0;
899 if (usermode)
900 fault_flags |= VM_FAULT_BURST;
901 if (ftype & VM_PROT_WRITE)
902 fault_flags |= VM_FAULT_DIRTY;
903 else
904 fault_flags |= VM_FAULT_NORMAL;
f8a5cc8f 905 rv = vm_fault(map, va, ftype, fault_flags);
bab69519 906
da673940
JG
907 PRELE(lp->lwp_proc);
908 } else {
909 /*
910 * Don't have to worry about process locking or stacks in the kernel.
911 */
912 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
913 }
914
915 if (rv == KERN_SUCCESS)
916 return (0);
917nogo:
918 if (!usermode) {
919 if (td->td_gd->gd_intr_nesting_level == 0 &&
920 td->td_pcb->pcb_onfault) {
921 frame->tf_rip = (register_t)td->td_pcb->pcb_onfault;
922 return (0);
923 }
924 trap_fatal(frame, usermode, eva);
925 return (-1);
926 }
927
928 /*
a76ca9b9 929 * NOTE: on x86_64 we have a tf_addr field in the trapframe, no
da673940
JG
930 * kludge is needed to pass the fault address to signal handlers.
931 */
932 struct proc *p = td->td_proc;
933 kprintf("seg-fault accessing address %p rip=%p pid=%d p_comm=%s\n",
934 (void *)va, (void *)frame->tf_rip, p->p_pid, p->p_comm);
935 /* Debugger("seg-fault"); */
936
937 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
938}
939
940static void
941trap_fatal(struct trapframe *frame, int usermode, vm_offset_t eva)
942{
943 int code, type, ss;
944 long rsp;
945
946 code = frame->tf_xflags;
947 type = frame->tf_trapno;
948
949 if (type <= MAX_TRAP_MSG) {
950 kprintf("\n\nFatal trap %d: %s while in %s mode\n",
951 type, trap_msg[type],
952 (usermode ? "user" : "kernel"));
953 }
da673940 954 /* two separate prints in case of a trap on an unmapped page */
da673940 955 kprintf("cpuid = %d\n", mycpu->gd_cpuid);
da673940 956 if (type == T_PAGEFLT) {
0e6594a8 957 kprintf("fault virtual address = %p\n", (void *)eva);
da673940
JG
958 kprintf("fault code = %s %s, %s\n",
959 usermode ? "user" : "supervisor",
960 code & PGEX_W ? "write" : "read",
961 code & PGEX_P ? "protection violation" : "page not present");
962 }
963 kprintf("instruction pointer = 0x%lx:0x%lx\n",
964 frame->tf_cs & 0xffff, frame->tf_rip);
965 if (usermode) {
966 ss = frame->tf_ss & 0xffff;
967 rsp = frame->tf_rsp;
968 } else {
969 ss = GSEL(GDATA_SEL, SEL_KPL);
970 rsp = (long)&frame->tf_rsp;
971 }
972 kprintf("stack pointer = 0x%x:0x%lx\n", ss, rsp);
973 kprintf("frame pointer = 0x%x:0x%lx\n", ss, frame->tf_rbp);
974 kprintf("processor eflags = ");
975 if (frame->tf_rflags & PSL_T)
976 kprintf("trace trap, ");
977 if (frame->tf_rflags & PSL_I)
978 kprintf("interrupt enabled, ");
979 if (frame->tf_rflags & PSL_NT)
980 kprintf("nested task, ");
981 if (frame->tf_rflags & PSL_RF)
982 kprintf("resume, ");
983#if 0
984 if (frame->tf_eflags & PSL_VM)
985 kprintf("vm86, ");
986#endif
0e6594a8 987 kprintf("IOPL = %jd\n", (intmax_t)((frame->tf_rflags & PSL_IOPL) >> 12));
da673940
JG
988 kprintf("current process = ");
989 if (curproc) {
990 kprintf("%lu (%s)\n",
991 (u_long)curproc->p_pid, curproc->p_comm ?
992 curproc->p_comm : "");
993 } else {
994 kprintf("Idle\n");
995 }
996 kprintf("current thread = pri %d ", curthread->td_pri);
f9235b6d 997 if (curthread->td_critcount)
da673940
JG
998 kprintf("(CRIT)");
999 kprintf("\n");
da673940
JG
1000/**
1001 * XXX FIXME:
1002 * we probably SHOULD have stopped the other CPUs before now!
1003 * another CPU COULD have been touching cpl at this moment...
1004 */
1005 kprintf(" <- SMP: XXX");
da673940
JG
1006 kprintf("\n");
1007
1008#ifdef KDB
1009 if (kdb_trap(&psl))
1010 return;
1011#endif
1012#ifdef DDB
1013 if ((debugger_on_panic || db_active) && kdb_trap(type, code, frame))
1014 return;
1015#endif
1016 kprintf("trap number = %d\n", type);
1017 if (type <= MAX_TRAP_MSG)
1018 panic("%s", trap_msg[type]);
1019 else
1020 panic("unknown/reserved trap");
1021}
1022
1023/*
1024 * Double fault handler. Called when a fault occurs while writing
1025 * a frame for a trap/exception onto the stack. This usually occurs
1026 * when the stack overflows (such is the case with infinite recursion,
1027 * for example).
1028 *
1029 * XXX Note that the current PTD gets replaced by IdlePTD when the
1030 * task switch occurs. This means that the stack that was active at
1031 * the time of the double fault is not available at <kstack> unless
1032 * the machine was idle when the double fault occurred. The downside
1033 * of this is that "trace <ebp>" in ddb won't work.
1034 */
1035void
1036dblfault_handler(void)
1037{
0e6594a8 1038#if JG
da673940 1039 struct mdglobaldata *gd = mdcpu;
0e6594a8 1040#endif
da673940
JG
1041
1042 kprintf("\nFatal double fault:\n");
1043#if JG
1044 kprintf("rip = 0x%lx\n", gd->gd_common_tss.tss_rip);
1045 kprintf("rsp = 0x%lx\n", gd->gd_common_tss.tss_rsp);
1046 kprintf("rbp = 0x%lx\n", gd->gd_common_tss.tss_rbp);
1047#endif
da673940 1048 /* two separate prints in case of a trap on an unmapped page */
da673940 1049 kprintf("cpuid = %d\n", mycpu->gd_cpuid);
da673940
JG
1050 panic("double fault");
1051}
1052
1053/*
1054 * Compensate for 386 brain damage (missing URKR).
1055 * This is a little simpler than the pagefault handler in trap() because
1056 * it the page tables have already been faulted in and high addresses
1057 * are thrown out early for other reasons.
1058 */
1059int
1060trapwrite(unsigned addr)
1061{
1062 struct lwp *lp;
1063 vm_offset_t va;
1064 struct vmspace *vm;
1065 int rv;
1066
1067 va = trunc_page((vm_offset_t)addr);
1068 /*
1069 * XXX - MAX is END. Changed > to >= for temp. fix.
1070 */
1071 if (va >= VM_MAX_USER_ADDRESS)
1072 return (1);
1073
1074 lp = curthread->td_lwp;
1075 vm = lp->lwp_vmspace;
1076
1077 PHOLD(lp->lwp_proc);
1078
1079 if (!grow_stack (lp->lwp_proc, va)) {
1080 PRELE(lp->lwp_proc);
1081 return (1);
1082 }
1083
1084 /*
1085 * fault the data page
1086 */
1087 rv = vm_fault(&vm->vm_map, va, VM_PROT_WRITE, VM_FAULT_DIRTY);
1088
1089 PRELE(lp->lwp_proc);
1090
1091 if (rv != KERN_SUCCESS)
1092 return 1;
1093
1094 return (0);
1095}
1096
1097/*
1098 * syscall2 - MP aware system call request C handler
1099 *
1100 * A system call is essentially treated as a trap except that the
1101 * MP lock is not held on entry or return. We are responsible for
1102 * obtaining the MP lock if necessary and for handling ASTs
1103 * (e.g. a task switch) prior to return.
1104 *
1105 * In general, only simple access and manipulation of curproc and
1106 * the current stack is allowed without having to hold MP lock.
1107 *
1108 * MPSAFE - note that large sections of this routine are run without
1109 * the MP lock.
1110 */
1111void
1112syscall2(struct trapframe *frame)
1113{
1114 struct thread *td = curthread;
1115 struct proc *p = td->td_proc;
1116 struct lwp *lp = td->td_lwp;
1117 caddr_t params;
1118 struct sysent *callp;
1119 register_t orig_tf_rflags;
1120 int sticks;
1121 int error;
1122 int narg;
1123#ifdef INVARIANTS
f9235b6d 1124 int crit_count = td->td_critcount;
aba00258 1125 lwkt_tokref_t curstop = td->td_toks_stop;
da673940 1126#endif
da673940 1127 int have_mplock = 0;
da673940
JG
1128 register_t *argp;
1129 u_int code;
1130 int reg, regcnt;
1131 union sysunion args;
1132 register_t *argsdst;
1133
1134 mycpu->gd_cnt.v_syscall++;
1135
1136 KTR_LOG(kernentry_syscall, lp->lwp_proc->p_pid, lp->lwp_tid,
b19d1a6b 1137 frame->tf_rax);
da673940 1138
73e24181 1139 userenter(td, p); /* lazy raise our priority */
da673940
JG
1140
1141 reg = 0;
1142 regcnt = 6;
1143 /*
1144 * Misc
1145 */
1146 sticks = (int)td->td_sticks;
1147 orig_tf_rflags = frame->tf_rflags;
1148
1149 /*
1150 * Virtual kernel intercept - if a VM context managed by a virtual
1151 * kernel issues a system call the virtual kernel handles it, not us.
1152 * Restore the virtual kernel context and return from its system
1153 * call. The current frame is copied out to the virtual kernel.
1154 */
1155 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
0e6594a8 1156 vkernel_trap(lp, frame);
da673940
JG
1157 error = EJUSTRETURN;
1158 goto out;
1159 }
1160
1161 /*
1162 * Get the system call parameters and account for time
1163 */
1164 lp->lwp_md.md_regs = frame;
1165 params = (caddr_t)frame->tf_rsp + sizeof(register_t);
1166 code = frame->tf_rax;
1167
1168 if (p->p_sysent->sv_prepsyscall) {
1169 (*p->p_sysent->sv_prepsyscall)(
1170 frame, (int *)(&args.nosys.sysmsg + 1),
1171 &code, &params);
1172 } else {
1173 if (code == SYS_syscall || code == SYS___syscall) {
1174 code = frame->tf_rdi;
1175 reg++;
1176 regcnt--;
1177 }
1178 }
1179
1180 if (p->p_sysent->sv_mask)
1181 code &= p->p_sysent->sv_mask;
1182
1183 if (code >= p->p_sysent->sv_size)
1184 callp = &p->p_sysent->sv_table[0];
1185 else
1186 callp = &p->p_sysent->sv_table[code];
1187
1188 narg = callp->sy_narg & SYF_ARGMASK;
1189
1190 /*
a76ca9b9 1191 * On x86_64 we get up to six arguments in registers. The rest are
da673940
JG
1192 * on the stack. The first six members of 'struct trapframe' happen
1193 * to be the registers used to pass arguments, in exactly the right
1194 * order.
1195 */
1196 argp = &frame->tf_rdi;
1197 argp += reg;
1198 argsdst = (register_t *)(&args.nosys.sysmsg + 1);
1199 /*
1200 * JG can we overflow the space pointed to by 'argsdst'
1201 * either with 'bcopy' or with 'copyin'?
1202 */
1203 bcopy(argp, argsdst, sizeof(register_t) * regcnt);
1204 /*
1205 * copyin is MP aware, but the tracing code is not
1206 */
1207 if (narg > regcnt) {
1208 KASSERT(params != NULL, ("copyin args with no params!"));
1209 error = copyin(params, &argsdst[regcnt],
1210 (narg - regcnt) * sizeof(register_t));
1211 if (error) {
1212#ifdef KTRACE
1213 if (KTRPOINT(td, KTR_SYSCALL)) {
1214 MAKEMPSAFE(have_mplock);
1215
1216 ktrsyscall(lp, code, narg,
1217 (void *)(&args.nosys.sysmsg + 1));
1218 }
1219#endif
1220 goto bad;
1221 }
1222 }
1223
1224#ifdef KTRACE
1225 if (KTRPOINT(td, KTR_SYSCALL)) {
1226 MAKEMPSAFE(have_mplock);
1227 ktrsyscall(lp, code, narg, (void *)(&args.nosys.sysmsg + 1));
1228 }
1229#endif
1230
1231 /*
1232 * Default return value is 0 (will be copied to %rax). Double-value
1233 * returns use %rax and %rdx. %rdx is left unchanged for system
1234 * calls which return only one result.
1235 */
1236 args.sysmsg_fds[0] = 0;
1237 args.sysmsg_fds[1] = frame->tf_rdx;
1238
1239 /*
1240 * The syscall might manipulate the trap frame. If it does it
1241 * will probably return EJUSTRETURN.
1242 */
1243 args.sysmsg_frame = frame;
1244
1245 STOPEVENT(p, S_SCE, narg); /* MP aware */
1246
da673940 1247 /*
0e6594a8
SW
1248 * NOTE: All system calls run MPSAFE now. The system call itself
1249 * is responsible for getting the MP lock.
da673940 1250 */
da673940
JG
1251 error = (*callp->sy_call)(&args);
1252
1253#if 0
1254 kprintf("system call %d returned %d\n", code, error);
1255#endif
1256
1257out:
1258 /*
1259 * MP SAFE (we may or may not have the MP lock at this point)
1260 */
1261 switch (error) {
1262 case 0:
1263 /*
1264 * Reinitialize proc pointer `p' as it may be different
1265 * if this is a child returning from fork syscall.
1266 */
1267 p = curproc;
1268 lp = curthread->td_lwp;
1269 frame->tf_rax = args.sysmsg_fds[0];
1270 frame->tf_rdx = args.sysmsg_fds[1];
1271 frame->tf_rflags &= ~PSL_C;
1272 break;
1273 case ERESTART:
1274 /*
1275 * Reconstruct pc, we know that 'syscall' is 2 bytes.
1276 * We have to do a full context restore so that %r10
1277 * (which was holding the value of %rcx) is restored for
1278 * the next iteration.
1279 */
1280 frame->tf_rip -= frame->tf_err;
1281 frame->tf_r10 = frame->tf_rcx;
1282 break;
1283 case EJUSTRETURN:
1284 break;
1285 case EASYNC:
1286 panic("Unexpected EASYNC return value (for now)");
1287 default:
1288bad:
1289 if (p->p_sysent->sv_errsize) {
1290 if (error >= p->p_sysent->sv_errsize)
1291 error = -1; /* XXX */
1292 else
1293 error = p->p_sysent->sv_errtbl[error];
1294 }
1295 frame->tf_rax = error;
1296 frame->tf_rflags |= PSL_C;
1297 break;
1298 }
1299
1300 /*
1301 * Traced syscall. trapsignal() is not MP aware.
1302 */
1303 if (orig_tf_rflags & PSL_T) {
1304 MAKEMPSAFE(have_mplock);
1305 frame->tf_rflags &= ~PSL_T;
1306 trapsignal(lp, SIGTRAP, 0);
1307 }
1308
1309 /*
1310 * Handle reschedule and other end-of-syscall issues
1311 */
1312 userret(lp, frame, sticks);
1313
1314#ifdef KTRACE
1315 if (KTRPOINT(td, KTR_SYSRET)) {
1316 MAKEMPSAFE(have_mplock);
1317 ktrsysret(lp, code, error, args.sysmsg_result);
1318 }
1319#endif
1320
1321 /*
1322 * This works because errno is findable through the
1323 * register set. If we ever support an emulation where this
1324 * is not the case, this code will need to be revisited.
1325 */
1326 STOPEVENT(p, S_SCX, code);
1327
1328 userexit(lp);
da673940
JG
1329 /*
1330 * Release the MP lock if we had to get it
1331 */
da673940
JG
1332 if (have_mplock)
1333 rel_mplock();
da673940
JG
1334 KTR_LOG(kernentry_syscall_ret, lp->lwp_proc->p_pid, lp->lwp_tid, error);
1335#ifdef INVARIANTS
3933a3ab 1336 KASSERT(&td->td_toks_base == td->td_toks_stop,
da673940 1337 ("syscall: critical section count mismatch! %d/%d",
f9235b6d 1338 crit_count, td->td_pri));
3933a3ab
MD
1339 KASSERT(curstop == td->td_toks_stop,
1340 ("syscall: extra tokens held after trap! %ld",
1341 td->td_toks_stop - &td->td_toks_base));
da673940
JG
1342#endif
1343}
1344
2b0bd8aa
MD
1345/*
1346 * NOTE: mplock not held at any point
1347 */
da673940
JG
1348void
1349fork_return(struct lwp *lp, struct trapframe *frame)
1350{
1351 frame->tf_rax = 0; /* Child returns zero */
1352 frame->tf_rflags &= ~PSL_C; /* success */
1353 frame->tf_rdx = 1;
1354
1355 generic_lwp_return(lp, frame);
1356 KTR_LOG(kernentry_fork_ret, lp->lwp_proc->p_pid, lp->lwp_tid);
1357}
1358
1359/*
1360 * Simplified back end of syscall(), used when returning from fork()
2b0bd8aa
MD
1361 * directly into user mode.
1362 *
1363 * This code will return back into the fork trampoline code which then
1364 * runs doreti.
1365 *
1366 * NOTE: The mplock is not held at any point.
da673940
JG
1367 */
1368void
1369generic_lwp_return(struct lwp *lp, struct trapframe *frame)
1370{
1371 struct proc *p = lp->lwp_proc;
1372
1373 /*
1374 * Newly forked processes are given a kernel priority. We have to
1375 * adjust the priority to a normal user priority and fake entry
1376 * into the kernel (call userenter()) to install a passive release
1377 * function just in case userret() decides to stop the process. This
1378 * can occur when ^Z races a fork. If we do not install the passive
1379 * release function the current process designation will not be
1380 * released when the thread goes to sleep.
1381 */
1382 lwkt_setpri_self(TDPRI_USER_NORM);
73e24181 1383 userenter(lp->lwp_thread, p);
da673940
JG
1384 userret(lp, frame, 0);
1385#ifdef KTRACE
1386 if (KTRPOINT(lp->lwp_thread, KTR_SYSRET))
1387 ktrsysret(lp, SYS_fork, 0, 0);
1388#endif
4643740a 1389 lp->lwp_flags |= LWP_PASSIVE_ACQ;
da673940 1390 userexit(lp);
4643740a 1391 lp->lwp_flags &= ~LWP_PASSIVE_ACQ;
da673940
JG
1392}
1393
1394/*
1395 * doreti has turned into this. The frame is directly on the stack. We
1396 * pull everything else we need (fpu and tls context) from the current
1397 * thread.
1398 *
1399 * Note on fpu interactions: In a virtual kernel, the fpu context for
1400 * an emulated user mode process is not shared with the virtual kernel's
1401 * fpu context, so we only have to 'stack' fpu contexts within the virtual
1402 * kernel itself, and not even then since the signal() contexts that we care
1403 * about save and restore the FPU state (I think anyhow).
1404 *
1405 * vmspace_ctl() returns an error only if it had problems instaling the
1406 * context we supplied or problems copying data to/from our VM space.
1407 */
1408void
1409go_user(struct intrframe *frame)
1410{
1411 struct trapframe *tf = (void *)&frame->if_rdi;
1412 int r;
1413
1414 /*
1415 * Interrupts may be disabled on entry, make sure all signals
1416 * can be received before beginning our loop.
1417 */
1418 sigsetmask(0);
1419
1420 /*
1421 * Switch to the current simulated user process, then call
1422 * user_trap() when we break out of it (usually due to a signal).
1423 */
1424 for (;;) {
1425 /*
1426 * Tell the real kernel whether it is ok to use the FP
1427 * unit or not.
1428 */
1429 if (mdcpu->gd_npxthread == curthread) {
1430 tf->tf_xflags &= ~PGEX_FPFAULT;
1431 } else {
1432 tf->tf_xflags |= PGEX_FPFAULT;
1433 }
1434
1435 /*
1436 * Run emulated user process context. This call interlocks
1437 * with new mailbox signals.
1438 *
1439 * Set PGEX_U unconditionally, indicating a user frame (the
1440 * bit is normally set only by T_PAGEFLT).
1441 */
1442 r = vmspace_ctl(&curproc->p_vmspace->vm_pmap, VMSPACE_CTL_RUN,
1443 tf, &curthread->td_savevext);
1444 frame->if_xflags |= PGEX_U;
1445#if 0
1446 kprintf("GO USER %d trap %ld EVA %08lx RIP %08lx RSP %08lx XFLAGS %02lx/%02lx\n",
1447 r, tf->tf_trapno, tf->tf_addr, tf->tf_rip, tf->tf_rsp,
1448 tf->tf_xflags, frame->if_xflags);
1449#endif
1450 if (r < 0) {
1451 if (errno != EINTR)
0e6594a8 1452 panic("vmspace_ctl failed error %d", errno);
da673940
JG
1453 } else {
1454 if (tf->tf_trapno) {
1455 user_trap(tf);
1456 }
1457 }
1458 if (mycpu->gd_reqflags & RQF_AST_MASK) {
1459 tf->tf_trapno = T_ASTFLT;
1460 user_trap(tf);
1461 }
1462 tf->tf_trapno = 0;
1463 }
1464}
1465
1466/*
1467 * If PGEX_FPFAULT is set then set FP_VIRTFP in the PCB to force a T_DNA
1468 * fault (which is then passed back to the virtual kernel) if an attempt is
1469 * made to use the FP unit.
1470 *
1471 * XXX this is a fairly big hack.
1472 */
1473void
1474set_vkernel_fp(struct trapframe *frame)
1475{
1476 struct thread *td = curthread;
1477
1478 if (frame->tf_xflags & PGEX_FPFAULT) {
1479 td->td_pcb->pcb_flags |= FP_VIRTFP;
1480 if (mdcpu->gd_npxthread == td)
1481 npxexit();
1482 } else {
1483 td->td_pcb->pcb_flags &= ~FP_VIRTFP;
1484 }
1485}
0e6594a8
SW
1486
1487/*
1488 * Called from vkernel_trap() to fixup the vkernel's syscall
1489 * frame for vmspace_ctl() return.
1490 */
1491void
1492cpu_vkernel_trap(struct trapframe *frame, int error)
1493{
1494 frame->tf_rax = error;
1495 if (error)
1496 frame->tf_rflags |= PSL_C;
1497 else
1498 frame->tf_rflags &= ~PSL_C;
1499}