kernel - Major signal path adjustments to fix races, tsleep race fixes, +more
[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
JG
95
96#ifdef SMP
97
98#define MAKEMPSAFE(have_mplock) \
99 if (have_mplock == 0) { \
100 get_mplock(); \
101 have_mplock = 1; \
102 }
103
104#else
105
106#define MAKEMPSAFE(have_mplock)
107
108#endif
109
110int (*pmath_emulate) (struct trapframe *);
111
112extern int trapwrite (unsigned addr);
113
114static int trap_pfault (struct trapframe *, int, vm_offset_t);
115static void trap_fatal (struct trapframe *, int, vm_offset_t);
116void dblfault_handler (void);
117
118#if 0
119extern inthand_t IDTVEC(syscall);
120#endif
121
122#define MAX_TRAP_MSG 30
123static char *trap_msg[] = {
124 "", /* 0 unused */
125 "privileged instruction fault", /* 1 T_PRIVINFLT */
126 "", /* 2 unused */
127 "breakpoint instruction fault", /* 3 T_BPTFLT */
128 "", /* 4 unused */
129 "", /* 5 unused */
130 "arithmetic trap", /* 6 T_ARITHTRAP */
131 "system forced exception", /* 7 T_ASTFLT */
132 "", /* 8 unused */
133 "general protection fault", /* 9 T_PROTFLT */
134 "trace trap", /* 10 T_TRCTRAP */
135 "", /* 11 unused */
136 "page fault", /* 12 T_PAGEFLT */
137 "", /* 13 unused */
138 "alignment fault", /* 14 T_ALIGNFLT */
139 "", /* 15 unused */
140 "", /* 16 unused */
141 "", /* 17 unused */
142 "integer divide fault", /* 18 T_DIVIDE */
143 "non-maskable interrupt trap", /* 19 T_NMI */
144 "overflow trap", /* 20 T_OFLOW */
145 "FPU bounds check fault", /* 21 T_BOUND */
146 "FPU device not available", /* 22 T_DNA */
147 "double fault", /* 23 T_DOUBLEFLT */
148 "FPU operand fetch fault", /* 24 T_FPOPFLT */
149 "invalid TSS fault", /* 25 T_TSSFLT */
150 "segment not present fault", /* 26 T_SEGNPFLT */
151 "stack fault", /* 27 T_STKFLT */
152 "machine check trap", /* 28 T_MCHK */
153 "SIMD floating-point exception", /* 29 T_XMMFLT */
154 "reserved (unknown) fault", /* 30 T_RESERVED */
155};
156
157#ifdef DDB
158static int ddb_on_nmi = 1;
159SYSCTL_INT(_machdep, OID_AUTO, ddb_on_nmi, CTLFLAG_RW,
160 &ddb_on_nmi, 0, "Go to DDB on NMI");
161#endif
162static int panic_on_nmi = 1;
163SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RW,
164 &panic_on_nmi, 0, "Panic on NMI");
165static int fast_release;
166SYSCTL_INT(_machdep, OID_AUTO, fast_release, CTLFLAG_RW,
167 &fast_release, 0, "Passive Release was optimal");
168static int slow_release;
169SYSCTL_INT(_machdep, OID_AUTO, slow_release, CTLFLAG_RW,
170 &slow_release, 0, "Passive Release was nonoptimal");
da673940
JG
171
172MALLOC_DEFINE(M_SYSMSG, "sysmsg", "sysmsg structure");
173extern int max_sysmsg;
174
175/*
73e24181
MD
176 * Passively intercepts the thread switch function to increase the thread
177 * priority from a user priority to a kernel priority, reducing
da673940 178 * syscall and trap overhead for the case where no switch occurs.
73e24181
MD
179 *
180 * Synchronizes td_ucred with p_ucred. This is used by system calls,
181 * signal handling, faults, AST traps, and anything else that enters the
182 * kernel from userland and provides the kernel with a stable read-only
183 * copy of the process ucred.
da673940 184 */
da673940 185static __inline void
73e24181 186userenter(struct thread *curtd, struct proc *curp)
da673940 187{
73e24181
MD
188 struct ucred *ocred;
189 struct ucred *ncred;
190
da673940 191 curtd->td_release = lwkt_passive_release;
73e24181
MD
192
193 if (curtd->td_ucred != curp->p_ucred) {
194 ncred = crhold(curp->p_ucred);
195 ocred = curtd->td_ucred;
196 curtd->td_ucred = ncred;
197 if (ocred)
198 crfree(ocred);
199 }
da673940
JG
200}
201
202/*
203 * Handle signals, upcalls, profiling, and other AST's and/or tasks that
204 * must be completed before we can return to or try to return to userland.
205 *
206 * Note that td_sticks is a 64 bit quantity, but there's no point doing 64
207 * arithmatic on the delta calculation so the absolute tick values are
208 * truncated to an integer.
209 */
210static void
211userret(struct lwp *lp, struct trapframe *frame, int sticks)
212{
213 struct proc *p = lp->lwp_proc;
214 int sig;
215
216 /*
217 * Charge system time if profiling. Note: times are in microseconds.
218 * This may do a copyout and block, so do it first even though it
219 * means some system time will be charged as user time.
220 */
4643740a 221 if (p->p_flags & P_PROFIL) {
da673940
JG
222 addupc_task(p, frame->tf_rip,
223 (u_int)((int)lp->lwp_thread->td_sticks - sticks));
224 }
225
226recheck:
227 /*
228 * If the jungle wants us dead, so be it.
229 */
4643740a 230 if (lp->lwp_mpflags & LWP_MP_WEXIT) {
b5c4d81f 231 lwkt_gettoken(&p->p_token);
da673940 232 lwp_exit(0);
b5c4d81f 233 lwkt_reltoken(&p->p_token); /* NOT REACHED */
da673940
JG
234 }
235
236 /*
237 * Block here if we are in a stopped state.
238 */
239 if (p->p_stat == SSTOP) {
4643740a 240 lwkt_gettoken(&p->p_token);
da673940 241 tstop();
4643740a 242 lwkt_reltoken(&p->p_token);
da673940
JG
243 goto recheck;
244 }
245
246 /*
898e34b3
MD
247 * Post any pending upcalls. If running a virtual kernel be sure
248 * to restore the virtual kernel's vmspace before posting the upcall.
da673940 249 */
4643740a 250 if (p->p_flags & (P_SIGVTALRM | P_SIGPROF | P_UPCALLPEND)) {
fc509460 251 lwkt_gettoken(&p->p_token);
4643740a
MD
252 if (p->p_flags & P_SIGVTALRM) {
253 p->p_flags &= ~P_SIGVTALRM;
898e34b3
MD
254 ksignal(p, SIGVTALRM);
255 }
4643740a
MD
256 if (p->p_flags & P_SIGPROF) {
257 p->p_flags &= ~P_SIGPROF;
898e34b3
MD
258 ksignal(p, SIGPROF);
259 }
4643740a
MD
260 if (p->p_flags & P_UPCALLPEND) {
261 p->p_flags &= ~P_UPCALLPEND;
898e34b3
MD
262 postupcall(lp);
263 }
fc509460 264 lwkt_reltoken(&p->p_token);
da673940
JG
265 goto recheck;
266 }
267
268 /*
269 * Post any pending signals
2883d2d8
MD
270 *
271 * WARNING! postsig() can exit and not return.
da673940
JG
272 */
273 if ((sig = CURSIG_TRACE(lp)) != 0) {
fc509460 274 lwkt_gettoken(&p->p_token);
da673940 275 postsig(sig);
fc509460 276 lwkt_reltoken(&p->p_token);
da673940
JG
277 goto recheck;
278 }
279
280 /*
281 * block here if we are swapped out, but still process signals
282 * (such as SIGKILL). proc0 (the swapin scheduler) is already
283 * aware of our situation, we do not have to wake it up.
284 */
4643740a 285 if (p->p_flags & P_SWAPPEDOUT) {
616516c8 286 lwkt_gettoken(&p->p_token);
da673940 287 get_mplock();
4643740a 288 p->p_flags |= P_SWAPWAIT;
da673940 289 swapin_request();
4643740a 290 if (p->p_flags & P_SWAPWAIT)
da673940 291 tsleep(p, PCATCH, "SWOUT", 0);
4643740a 292 p->p_flags &= ~P_SWAPWAIT;
da673940 293 rel_mplock();
616516c8 294 lwkt_reltoken(&p->p_token);
da673940
JG
295 goto recheck;
296 }
297
298 /*
299 * Make sure postsig() handled request to restore old signal mask after
300 * running signal handler.
301 */
4643740a 302 KKASSERT((lp->lwp_flags & LWP_OLDMASK) == 0);
da673940
JG
303}
304
305/*
306 * Cleanup from userenter and any passive release that might have occured.
307 * We must reclaim the current-process designation before we can return
308 * to usermode. We also handle both LWKT and USER reschedule requests.
309 */
310static __inline void
311userexit(struct lwp *lp)
312{
313 struct thread *td = lp->lwp_thread;
314 /* globaldata_t gd = td->td_gd; */
315
316 /*
317 * Handle stop requests at kernel priority. Any requests queued
318 * after this loop will generate another AST.
319 */
320 while (lp->lwp_proc->p_stat == SSTOP) {
4643740a 321 lwkt_gettoken(&lp->lwp_proc->p_token);
da673940 322 tstop();
4643740a 323 lwkt_reltoken(&lp->lwp_proc->p_token);
da673940
JG
324 }
325
326 /*
327 * Reduce our priority in preparation for a return to userland. If
328 * our passive release function was still in place, our priority was
329 * never raised and does not need to be reduced.
330 */
331 lwkt_passive_recover(td);
332
333 /*
334 * Become the current user scheduled process if we aren't already,
335 * and deal with reschedule requests and other factors.
336 */
337 lp->lwp_proc->p_usched->acquire_curproc(lp);
338 /* WARNING: we may have migrated cpu's */
339 /* gd = td->td_gd; */
340}
341
342#if !defined(KTR_KERNENTRY)
343#define KTR_KERNENTRY KTR_ALL
344#endif
345KTR_INFO_MASTER(kernentry);
346KTR_INFO(KTR_KERNENTRY, kernentry, trap, 0, "pid=%d, tid=%d, trapno=%d, eva=%p",
347 sizeof(int) + sizeof(int) + sizeof(int) + sizeof(vm_offset_t));
348KTR_INFO(KTR_KERNENTRY, kernentry, trap_ret, 0, "pid=%d, tid=%d",
349 sizeof(int) + sizeof(int));
350KTR_INFO(KTR_KERNENTRY, kernentry, syscall, 0, "pid=%d, tid=%d, call=%d",
351 sizeof(int) + sizeof(int) + sizeof(int));
352KTR_INFO(KTR_KERNENTRY, kernentry, syscall_ret, 0, "pid=%d, tid=%d, err=%d",
353 sizeof(int) + sizeof(int) + sizeof(int));
354KTR_INFO(KTR_KERNENTRY, kernentry, fork_ret, 0, "pid=%d, tid=%d",
355 sizeof(int) + sizeof(int));
356
357/*
358 * Exception, fault, and trap interface to the kernel.
359 * This common code is called from assembly language IDT gate entry
360 * routines that prepare a suitable stack frame, and restore this
361 * frame after the exception has been processed.
362 *
363 * This function is also called from doreti in an interlock to handle ASTs.
364 * For example: hardwareint->INTROUTINE->(set ast)->doreti->trap
365 *
366 * NOTE! We have to retrieve the fault address prior to obtaining the
367 * MP lock because get_mplock() may switch out. YYY cr2 really ought
368 * to be retrieved by the assembly code, not here.
369 *
370 * XXX gd_trap_nesting_level currently prevents lwkt_switch() from panicing
371 * if an attempt is made to switch from a fast interrupt or IPI. This is
372 * necessary to properly take fatal kernel traps on SMP machines if
373 * get_mplock() has to block.
374 */
375
376void
377user_trap(struct trapframe *frame)
378{
379 struct globaldata *gd = mycpu;
380 struct thread *td = gd->gd_curthread;
381 struct lwp *lp = td->td_lwp;
382 struct proc *p;
383 int sticks = 0;
384 int i = 0, ucode = 0, type, code;
385#ifdef SMP
386 int have_mplock = 0;
387#endif
388#ifdef INVARIANTS
f9235b6d 389 int crit_count = td->td_critcount;
3933a3ab 390 lwkt_tokref_t curstop = td->td_toks_stop;
da673940
JG
391#endif
392 vm_offset_t eva;
393
394 p = td->td_proc;
395
396 if (frame->tf_trapno == T_PAGEFLT)
397 eva = frame->tf_addr;
398 else
399 eva = 0;
400#if 0
401 kprintf("USER_TRAP AT %08lx xflags %ld trapno %ld eva %08lx\n",
402 frame->tf_rip, frame->tf_xflags, frame->tf_trapno, eva);
403#endif
404
405 /*
406 * Everything coming from user mode runs through user_trap,
407 * including system calls.
408 */
571cdd83 409 if (frame->tf_trapno == T_FAST_SYSCALL) {
da673940
JG
410 syscall2(frame);
411 return;
412 }
413
414 KTR_LOG(kernentry_trap, lp->lwp_proc->p_pid, lp->lwp_tid,
415 frame->tf_trapno, eva);
416
417#ifdef DDB
418 if (db_active) {
419 eva = (frame->tf_trapno == T_PAGEFLT ? rcr2() : 0);
420 ++gd->gd_trap_nesting_level;
421 MAKEMPSAFE(have_mplock);
422 trap_fatal(frame, TRUE, eva);
423 --gd->gd_trap_nesting_level;
424 goto out2;
425 }
426#endif
427
da673940
JG
428 type = frame->tf_trapno;
429 code = frame->tf_err;
430
73e24181 431 userenter(td, p);
da673940
JG
432
433 sticks = (int)td->td_sticks;
434 lp->lwp_md.md_regs = frame;
435
436 switch (type) {
437 case T_PRIVINFLT: /* privileged instruction fault */
438 ucode = type;
439 i = SIGILL;
440 break;
441
442 case T_BPTFLT: /* bpt instruction fault */
443 case T_TRCTRAP: /* trace trap */
444 frame->tf_rflags &= ~PSL_T;
445 i = SIGTRAP;
446 break;
447
448 case T_ARITHTRAP: /* arithmetic trap */
449 ucode = code;
450 i = SIGFPE;
451 break;
452
453 case T_ASTFLT: /* Allow process switch */
454 mycpu->gd_cnt.v_soft++;
455 if (mycpu->gd_reqflags & RQF_AST_OWEUPC) {
2a418930
MD
456 atomic_clear_int(&mycpu->gd_reqflags, RQF_AST_OWEUPC);
457 addupc_task(p, p->p_prof.pr_addr, p->p_prof.pr_ticks);
da673940
JG
458 }
459 goto out;
460
461 /*
462 * The following two traps can happen in
463 * vm86 mode, and, if so, we want to handle
464 * them specially.
465 */
466 case T_PROTFLT: /* general protection fault */
467 case T_STKFLT: /* stack fault */
468#if 0
469 if (frame->tf_eflags & PSL_VM) {
470 i = vm86_emulate((struct vm86frame *)frame);
471 if (i == 0)
472 goto out;
473 break;
474 }
475#endif
476 /* FALL THROUGH */
477
478 case T_SEGNPFLT: /* segment not present fault */
479 case T_TSSFLT: /* invalid TSS fault */
480 case T_DOUBLEFLT: /* double fault */
481 default:
482 ucode = code + BUS_SEGM_FAULT ;
483 i = SIGBUS;
484 break;
485
486 case T_PAGEFLT: /* page fault */
487 MAKEMPSAFE(have_mplock);
488 i = trap_pfault(frame, TRUE, eva);
fbff57d1 489 if (i == -1 || i == 0)
da673940
JG
490 goto out;
491
492 ucode = T_PAGEFLT;
493 break;
494
495 case T_DIVIDE: /* integer divide fault */
496 ucode = FPE_INTDIV;
497 i = SIGFPE;
498 break;
499
500#if NISA > 0
501 case T_NMI:
502 MAKEMPSAFE(have_mplock);
503 /* machine/parity/power fail/"kitchen sink" faults */
504 if (isa_nmi(code) == 0) {
505#ifdef DDB
506 /*
507 * NMI can be hooked up to a pushbutton
508 * for debugging.
509 */
510 if (ddb_on_nmi) {
511 kprintf ("NMI ... going to debugger\n");
512 kdb_trap (type, 0, frame);
513 }
514#endif /* DDB */
515 goto out2;
516 } else if (panic_on_nmi)
517 panic("NMI indicates hardware failure");
518 break;
519#endif /* NISA > 0 */
520
521 case T_OFLOW: /* integer overflow fault */
522 ucode = FPE_INTOVF;
523 i = SIGFPE;
524 break;
525
526 case T_BOUND: /* bounds check fault */
527 ucode = FPE_FLTSUB;
528 i = SIGFPE;
529 break;
530
531 case T_DNA:
532 /*
533 * Virtual kernel intercept - pass the DNA exception
534 * to the (emulated) virtual kernel if it asked to handle
535 * it. This occurs when the virtual kernel is holding
536 * onto the FP context for a different emulated
537 * process then the one currently running.
538 *
539 * We must still call npxdna() since we may have
540 * saved FP state that the (emulated) virtual kernel
541 * needs to hand over to a different emulated process.
542 */
543 if (lp->lwp_vkernel && lp->lwp_vkernel->ve &&
544 (td->td_pcb->pcb_flags & FP_VIRTFP)
545 ) {
546 npxdna(frame);
547 break;
548 }
549 /*
550 * The kernel may have switched out the FP unit's
551 * state, causing the user process to take a fault
552 * when it tries to use the FP unit. Restore the
553 * state here
554 */
555 if (npxdna(frame))
556 goto out;
557 if (!pmath_emulate) {
558 i = SIGFPE;
559 ucode = FPE_FPU_NP_TRAP;
560 break;
561 }
562 i = (*pmath_emulate)(frame);
563 if (i == 0) {
564 if (!(frame->tf_rflags & PSL_T))
565 goto out2;
566 frame->tf_rflags &= ~PSL_T;
567 i = SIGTRAP;
568 }
569 /* else ucode = emulator_only_knows() XXX */
570 break;
571
572 case T_FPOPFLT: /* FPU operand fetch fault */
573 ucode = T_FPOPFLT;
574 i = SIGILL;
575 break;
576
577 case T_XMMFLT: /* SIMD floating-point exception */
578 ucode = 0; /* XXX */
579 i = SIGFPE;
580 break;
581 }
582
583 /*
584 * Virtual kernel intercept - if the fault is directly related to a
585 * VM context managed by a virtual kernel then let the virtual kernel
586 * handle it.
587 */
588 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
589 vkernel_trap(lp, frame);
590 goto out;
591 }
592
593 /*
594 * Translate fault for emulators (e.g. Linux)
595 */
596 if (*p->p_sysent->sv_transtrap)
597 i = (*p->p_sysent->sv_transtrap)(i, type);
598
599 MAKEMPSAFE(have_mplock);
600 trapsignal(lp, i, ucode);
601
602#ifdef DEBUG
603 if (type <= MAX_TRAP_MSG) {
604 uprintf("fatal process exception: %s",
605 trap_msg[type]);
606 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
607 uprintf(", fault VA = 0x%lx", (u_long)eva);
608 uprintf("\n");
609 }
610#endif
611
612out:
da673940
JG
613 userret(lp, frame, sticks);
614 userexit(lp);
615out2: ;
616#ifdef SMP
617 if (have_mplock)
618 rel_mplock();
619#endif
620 KTR_LOG(kernentry_trap_ret, lp->lwp_proc->p_pid, lp->lwp_tid);
621#ifdef INVARIANTS
f9235b6d 622 KASSERT(crit_count == td->td_critcount,
3933a3ab 623 ("trap: critical section count mismatch! %d/%d",
f9235b6d 624 crit_count, td->td_pri));
3933a3ab
MD
625 KASSERT(curstop == td->td_toks_stop,
626 ("trap: extra tokens held after trap! %ld/%ld",
627 curstop - &td->td_toks_base,
628 td->td_toks_stop - &td->td_toks_base));
da673940
JG
629#endif
630}
631
632void
633kern_trap(struct trapframe *frame)
634{
635 struct globaldata *gd = mycpu;
636 struct thread *td = gd->gd_curthread;
637 struct lwp *lp;
638 struct proc *p;
639 int i = 0, ucode = 0, type, code;
640#ifdef SMP
641 int have_mplock = 0;
642#endif
643#ifdef INVARIANTS
f9235b6d 644 int crit_count = td->td_critcount;
3933a3ab 645 lwkt_tokref_t curstop = td->td_toks_stop;
da673940
JG
646#endif
647 vm_offset_t eva;
648
649 lp = td->td_lwp;
650 p = td->td_proc;
651
652 if (frame->tf_trapno == T_PAGEFLT)
653 eva = frame->tf_addr;
654 else
655 eva = 0;
656
657#ifdef DDB
658 if (db_active) {
659 ++gd->gd_trap_nesting_level;
660 MAKEMPSAFE(have_mplock);
661 trap_fatal(frame, FALSE, eva);
662 --gd->gd_trap_nesting_level;
663 goto out2;
664 }
665#endif
666
da673940
JG
667 type = frame->tf_trapno;
668 code = frame->tf_err;
669
670#if 0
671kernel_trap:
672#endif
673 /* kernel trap */
674
675 switch (type) {
676 case T_PAGEFLT: /* page fault */
677 MAKEMPSAFE(have_mplock);
678 trap_pfault(frame, FALSE, eva);
679 goto out2;
680
681 case T_DNA:
682 /*
683 * The kernel may be using npx for copying or other
684 * purposes.
685 */
686 panic("kernel NPX should not happen");
687 if (npxdna(frame))
688 goto out2;
689 break;
690
691 case T_PROTFLT: /* general protection fault */
692 case T_SEGNPFLT: /* segment not present fault */
693 /*
694 * Invalid segment selectors and out of bounds
695 * %eip's and %esp's can be set up in user mode.
696 * This causes a fault in kernel mode when the
697 * kernel tries to return to user mode. We want
698 * to get this fault so that we can fix the
699 * problem here and not have to check all the
700 * selectors and pointers when the user changes
701 * them.
702 */
703 if (mycpu->gd_intr_nesting_level == 0) {
704 if (td->td_pcb->pcb_onfault) {
705 frame->tf_rip =
706 (register_t)td->td_pcb->pcb_onfault;
707 goto out2;
708 }
709 }
710 break;
711
712 case T_TSSFLT:
713 /*
714 * PSL_NT can be set in user mode and isn't cleared
715 * automatically when the kernel is entered. This
716 * causes a TSS fault when the kernel attempts to
717 * `iret' because the TSS link is uninitialized. We
718 * want to get this fault so that we can fix the
719 * problem here and not every time the kernel is
720 * entered.
721 */
722 if (frame->tf_rflags & PSL_NT) {
723 frame->tf_rflags &= ~PSL_NT;
724 goto out2;
725 }
726 break;
727
728 case T_TRCTRAP: /* trace trap */
729#if 0
730 if (frame->tf_eip == (int)IDTVEC(syscall)) {
731 /*
732 * We've just entered system mode via the
733 * syscall lcall. Continue single stepping
734 * silently until the syscall handler has
735 * saved the flags.
736 */
737 goto out2;
738 }
739 if (frame->tf_eip == (int)IDTVEC(syscall) + 1) {
740 /*
741 * The syscall handler has now saved the
742 * flags. Stop single stepping it.
743 */
744 frame->tf_eflags &= ~PSL_T;
745 goto out2;
746 }
747#endif
748#if 0
749 /*
750 * Ignore debug register trace traps due to
751 * accesses in the user's address space, which
752 * can happen under several conditions such as
753 * if a user sets a watchpoint on a buffer and
754 * then passes that buffer to a system call.
755 * We still want to get TRCTRAPS for addresses
756 * in kernel space because that is useful when
757 * debugging the kernel.
758 */
759 if (user_dbreg_trap()) {
760 /*
761 * Reset breakpoint bits because the
762 * processor doesn't
763 */
764 load_dr6(rdr6() & 0xfffffff0);
765 goto out2;
766 }
767#endif
768 /*
769 * Fall through (TRCTRAP kernel mode, kernel address)
770 */
771 case T_BPTFLT:
772 /*
773 * If DDB is enabled, let it handle the debugger trap.
774 * Otherwise, debugger traps "can't happen".
775 */
776#ifdef DDB
777 MAKEMPSAFE(have_mplock);
778 if (kdb_trap (type, 0, frame))
779 goto out2;
780#endif
781 break;
782 case T_DIVIDE:
783 MAKEMPSAFE(have_mplock);
784 trap_fatal(frame, FALSE, eva);
785 goto out2;
786 case T_NMI:
787 MAKEMPSAFE(have_mplock);
788 trap_fatal(frame, FALSE, eva);
789 goto out2;
b9d01986
MD
790 case T_SYSCALL80:
791 case T_FAST_SYSCALL:
da673940
JG
792 /*
793 * Ignore this trap generated from a spurious SIGTRAP.
794 *
795 * single stepping in / syscalls leads to spurious / SIGTRAP
796 * so ignore
797 *
798 * Haiku (c) 2007 Simon 'corecode' Schubert
799 */
800 goto out2;
801 }
802
803 /*
804 * Translate fault for emulators (e.g. Linux)
805 */
806 if (*p->p_sysent->sv_transtrap)
807 i = (*p->p_sysent->sv_transtrap)(i, type);
808
809 MAKEMPSAFE(have_mplock);
810 trapsignal(lp, i, ucode);
811
812#ifdef DEBUG
813 if (type <= MAX_TRAP_MSG) {
814 uprintf("fatal process exception: %s",
815 trap_msg[type]);
816 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
817 uprintf(", fault VA = 0x%lx", (u_long)eva);
818 uprintf("\n");
819 }
820#endif
821
822out2:
823 ;
824#ifdef SMP
825 if (have_mplock)
826 rel_mplock();
827#endif
828#ifdef INVARIANTS
f9235b6d 829 KASSERT(crit_count == td->td_critcount,
3933a3ab 830 ("trap: critical section count mismatch! %d/%d",
f9235b6d 831 crit_count, td->td_pri));
3933a3ab
MD
832 KASSERT(curstop == td->td_toks_stop,
833 ("trap: extra tokens held after trap! %ld/%ld",
834 curstop - &td->td_toks_base,
835 td->td_toks_stop - &td->td_toks_base));
da673940
JG
836#endif
837}
838
839int
840trap_pfault(struct trapframe *frame, int usermode, vm_offset_t eva)
841{
842 vm_offset_t va;
843 struct vmspace *vm = NULL;
844 vm_map_t map = 0;
845 int rv = 0;
846 vm_prot_t ftype;
847 thread_t td = curthread;
848 struct lwp *lp = td->td_lwp;
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
898 /* Fault in the user page: */
899 rv = vm_fault(map, va, ftype,
900 (ftype & VM_PROT_WRITE) ? VM_FAULT_DIRTY
901 : VM_FAULT_NORMAL);
902
903 PRELE(lp->lwp_proc);
904 } else {
905 /*
906 * Don't have to worry about process locking or stacks in the kernel.
907 */
908 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
909 }
910
911 if (rv == KERN_SUCCESS)
912 return (0);
913nogo:
914 if (!usermode) {
915 if (td->td_gd->gd_intr_nesting_level == 0 &&
916 td->td_pcb->pcb_onfault) {
917 frame->tf_rip = (register_t)td->td_pcb->pcb_onfault;
918 return (0);
919 }
920 trap_fatal(frame, usermode, eva);
921 return (-1);
922 }
923
924 /*
a76ca9b9 925 * NOTE: on x86_64 we have a tf_addr field in the trapframe, no
da673940
JG
926 * kludge is needed to pass the fault address to signal handlers.
927 */
928 struct proc *p = td->td_proc;
929 kprintf("seg-fault accessing address %p rip=%p pid=%d p_comm=%s\n",
930 (void *)va, (void *)frame->tf_rip, p->p_pid, p->p_comm);
931 /* Debugger("seg-fault"); */
932
933 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
934}
935
936static void
937trap_fatal(struct trapframe *frame, int usermode, vm_offset_t eva)
938{
939 int code, type, ss;
940 long rsp;
941
942 code = frame->tf_xflags;
943 type = frame->tf_trapno;
944
945 if (type <= MAX_TRAP_MSG) {
946 kprintf("\n\nFatal trap %d: %s while in %s mode\n",
947 type, trap_msg[type],
948 (usermode ? "user" : "kernel"));
949 }
950#ifdef SMP
951 /* two separate prints in case of a trap on an unmapped page */
da673940
JG
952 kprintf("cpuid = %d\n", mycpu->gd_cpuid);
953#endif
954 if (type == T_PAGEFLT) {
0e6594a8 955 kprintf("fault virtual address = %p\n", (void *)eva);
da673940
JG
956 kprintf("fault code = %s %s, %s\n",
957 usermode ? "user" : "supervisor",
958 code & PGEX_W ? "write" : "read",
959 code & PGEX_P ? "protection violation" : "page not present");
960 }
961 kprintf("instruction pointer = 0x%lx:0x%lx\n",
962 frame->tf_cs & 0xffff, frame->tf_rip);
963 if (usermode) {
964 ss = frame->tf_ss & 0xffff;
965 rsp = frame->tf_rsp;
966 } else {
967 ss = GSEL(GDATA_SEL, SEL_KPL);
968 rsp = (long)&frame->tf_rsp;
969 }
970 kprintf("stack pointer = 0x%x:0x%lx\n", ss, rsp);
971 kprintf("frame pointer = 0x%x:0x%lx\n", ss, frame->tf_rbp);
972 kprintf("processor eflags = ");
973 if (frame->tf_rflags & PSL_T)
974 kprintf("trace trap, ");
975 if (frame->tf_rflags & PSL_I)
976 kprintf("interrupt enabled, ");
977 if (frame->tf_rflags & PSL_NT)
978 kprintf("nested task, ");
979 if (frame->tf_rflags & PSL_RF)
980 kprintf("resume, ");
981#if 0
982 if (frame->tf_eflags & PSL_VM)
983 kprintf("vm86, ");
984#endif
0e6594a8 985 kprintf("IOPL = %jd\n", (intmax_t)((frame->tf_rflags & PSL_IOPL) >> 12));
da673940
JG
986 kprintf("current process = ");
987 if (curproc) {
988 kprintf("%lu (%s)\n",
989 (u_long)curproc->p_pid, curproc->p_comm ?
990 curproc->p_comm : "");
991 } else {
992 kprintf("Idle\n");
993 }
994 kprintf("current thread = pri %d ", curthread->td_pri);
f9235b6d 995 if (curthread->td_critcount)
da673940
JG
996 kprintf("(CRIT)");
997 kprintf("\n");
998#ifdef SMP
999/**
1000 * XXX FIXME:
1001 * we probably SHOULD have stopped the other CPUs before now!
1002 * another CPU COULD have been touching cpl at this moment...
1003 */
1004 kprintf(" <- SMP: XXX");
1005#endif
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
1048#ifdef SMP
1049 /* two separate prints in case of a trap on an unmapped page */
da673940
JG
1050 kprintf("cpuid = %d\n", mycpu->gd_cpuid);
1051#endif
1052 panic("double fault");
1053}
1054
1055/*
1056 * Compensate for 386 brain damage (missing URKR).
1057 * This is a little simpler than the pagefault handler in trap() because
1058 * it the page tables have already been faulted in and high addresses
1059 * are thrown out early for other reasons.
1060 */
1061int
1062trapwrite(unsigned addr)
1063{
1064 struct lwp *lp;
1065 vm_offset_t va;
1066 struct vmspace *vm;
1067 int rv;
1068
1069 va = trunc_page((vm_offset_t)addr);
1070 /*
1071 * XXX - MAX is END. Changed > to >= for temp. fix.
1072 */
1073 if (va >= VM_MAX_USER_ADDRESS)
1074 return (1);
1075
1076 lp = curthread->td_lwp;
1077 vm = lp->lwp_vmspace;
1078
1079 PHOLD(lp->lwp_proc);
1080
1081 if (!grow_stack (lp->lwp_proc, va)) {
1082 PRELE(lp->lwp_proc);
1083 return (1);
1084 }
1085
1086 /*
1087 * fault the data page
1088 */
1089 rv = vm_fault(&vm->vm_map, va, VM_PROT_WRITE, VM_FAULT_DIRTY);
1090
1091 PRELE(lp->lwp_proc);
1092
1093 if (rv != KERN_SUCCESS)
1094 return 1;
1095
1096 return (0);
1097}
1098
1099/*
1100 * syscall2 - MP aware system call request C handler
1101 *
1102 * A system call is essentially treated as a trap except that the
1103 * MP lock is not held on entry or return. We are responsible for
1104 * obtaining the MP lock if necessary and for handling ASTs
1105 * (e.g. a task switch) prior to return.
1106 *
1107 * In general, only simple access and manipulation of curproc and
1108 * the current stack is allowed without having to hold MP lock.
1109 *
1110 * MPSAFE - note that large sections of this routine are run without
1111 * the MP lock.
1112 */
1113void
1114syscall2(struct trapframe *frame)
1115{
1116 struct thread *td = curthread;
1117 struct proc *p = td->td_proc;
1118 struct lwp *lp = td->td_lwp;
1119 caddr_t params;
1120 struct sysent *callp;
1121 register_t orig_tf_rflags;
1122 int sticks;
1123 int error;
1124 int narg;
1125#ifdef INVARIANTS
f9235b6d 1126 int crit_count = td->td_critcount;
aba00258 1127 lwkt_tokref_t curstop = td->td_toks_stop;
da673940
JG
1128#endif
1129#ifdef SMP
1130 int have_mplock = 0;
1131#endif
1132 register_t *argp;
1133 u_int code;
1134 int reg, regcnt;
1135 union sysunion args;
1136 register_t *argsdst;
1137
1138 mycpu->gd_cnt.v_syscall++;
1139
1140 KTR_LOG(kernentry_syscall, lp->lwp_proc->p_pid, lp->lwp_tid,
1141 frame->tf_eax);
1142
73e24181 1143 userenter(td, p); /* lazy raise our priority */
da673940
JG
1144
1145 reg = 0;
1146 regcnt = 6;
1147 /*
1148 * Misc
1149 */
1150 sticks = (int)td->td_sticks;
1151 orig_tf_rflags = frame->tf_rflags;
1152
1153 /*
1154 * Virtual kernel intercept - if a VM context managed by a virtual
1155 * kernel issues a system call the virtual kernel handles it, not us.
1156 * Restore the virtual kernel context and return from its system
1157 * call. The current frame is copied out to the virtual kernel.
1158 */
1159 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
0e6594a8 1160 vkernel_trap(lp, frame);
da673940
JG
1161 error = EJUSTRETURN;
1162 goto out;
1163 }
1164
1165 /*
1166 * Get the system call parameters and account for time
1167 */
1168 lp->lwp_md.md_regs = frame;
1169 params = (caddr_t)frame->tf_rsp + sizeof(register_t);
1170 code = frame->tf_rax;
1171
1172 if (p->p_sysent->sv_prepsyscall) {
1173 (*p->p_sysent->sv_prepsyscall)(
1174 frame, (int *)(&args.nosys.sysmsg + 1),
1175 &code, &params);
1176 } else {
1177 if (code == SYS_syscall || code == SYS___syscall) {
1178 code = frame->tf_rdi;
1179 reg++;
1180 regcnt--;
1181 }
1182 }
1183
1184 if (p->p_sysent->sv_mask)
1185 code &= p->p_sysent->sv_mask;
1186
1187 if (code >= p->p_sysent->sv_size)
1188 callp = &p->p_sysent->sv_table[0];
1189 else
1190 callp = &p->p_sysent->sv_table[code];
1191
1192 narg = callp->sy_narg & SYF_ARGMASK;
1193
1194 /*
a76ca9b9 1195 * On x86_64 we get up to six arguments in registers. The rest are
da673940
JG
1196 * on the stack. The first six members of 'struct trapframe' happen
1197 * to be the registers used to pass arguments, in exactly the right
1198 * order.
1199 */
1200 argp = &frame->tf_rdi;
1201 argp += reg;
1202 argsdst = (register_t *)(&args.nosys.sysmsg + 1);
1203 /*
1204 * JG can we overflow the space pointed to by 'argsdst'
1205 * either with 'bcopy' or with 'copyin'?
1206 */
1207 bcopy(argp, argsdst, sizeof(register_t) * regcnt);
1208 /*
1209 * copyin is MP aware, but the tracing code is not
1210 */
1211 if (narg > regcnt) {
1212 KASSERT(params != NULL, ("copyin args with no params!"));
1213 error = copyin(params, &argsdst[regcnt],
1214 (narg - regcnt) * sizeof(register_t));
1215 if (error) {
1216#ifdef KTRACE
1217 if (KTRPOINT(td, KTR_SYSCALL)) {
1218 MAKEMPSAFE(have_mplock);
1219
1220 ktrsyscall(lp, code, narg,
1221 (void *)(&args.nosys.sysmsg + 1));
1222 }
1223#endif
1224 goto bad;
1225 }
1226 }
1227
1228#ifdef KTRACE
1229 if (KTRPOINT(td, KTR_SYSCALL)) {
1230 MAKEMPSAFE(have_mplock);
1231 ktrsyscall(lp, code, narg, (void *)(&args.nosys.sysmsg + 1));
1232 }
1233#endif
1234
1235 /*
1236 * Default return value is 0 (will be copied to %rax). Double-value
1237 * returns use %rax and %rdx. %rdx is left unchanged for system
1238 * calls which return only one result.
1239 */
1240 args.sysmsg_fds[0] = 0;
1241 args.sysmsg_fds[1] = frame->tf_rdx;
1242
1243 /*
1244 * The syscall might manipulate the trap frame. If it does it
1245 * will probably return EJUSTRETURN.
1246 */
1247 args.sysmsg_frame = frame;
1248
1249 STOPEVENT(p, S_SCE, narg); /* MP aware */
1250
da673940 1251 /*
0e6594a8
SW
1252 * NOTE: All system calls run MPSAFE now. The system call itself
1253 * is responsible for getting the MP lock.
da673940 1254 */
da673940
JG
1255 error = (*callp->sy_call)(&args);
1256
1257#if 0
1258 kprintf("system call %d returned %d\n", code, error);
1259#endif
1260
1261out:
1262 /*
1263 * MP SAFE (we may or may not have the MP lock at this point)
1264 */
1265 switch (error) {
1266 case 0:
1267 /*
1268 * Reinitialize proc pointer `p' as it may be different
1269 * if this is a child returning from fork syscall.
1270 */
1271 p = curproc;
1272 lp = curthread->td_lwp;
1273 frame->tf_rax = args.sysmsg_fds[0];
1274 frame->tf_rdx = args.sysmsg_fds[1];
1275 frame->tf_rflags &= ~PSL_C;
1276 break;
1277 case ERESTART:
1278 /*
1279 * Reconstruct pc, we know that 'syscall' is 2 bytes.
1280 * We have to do a full context restore so that %r10
1281 * (which was holding the value of %rcx) is restored for
1282 * the next iteration.
1283 */
1284 frame->tf_rip -= frame->tf_err;
1285 frame->tf_r10 = frame->tf_rcx;
1286 break;
1287 case EJUSTRETURN:
1288 break;
1289 case EASYNC:
1290 panic("Unexpected EASYNC return value (for now)");
1291 default:
1292bad:
1293 if (p->p_sysent->sv_errsize) {
1294 if (error >= p->p_sysent->sv_errsize)
1295 error = -1; /* XXX */
1296 else
1297 error = p->p_sysent->sv_errtbl[error];
1298 }
1299 frame->tf_rax = error;
1300 frame->tf_rflags |= PSL_C;
1301 break;
1302 }
1303
1304 /*
1305 * Traced syscall. trapsignal() is not MP aware.
1306 */
1307 if (orig_tf_rflags & PSL_T) {
1308 MAKEMPSAFE(have_mplock);
1309 frame->tf_rflags &= ~PSL_T;
1310 trapsignal(lp, SIGTRAP, 0);
1311 }
1312
1313 /*
1314 * Handle reschedule and other end-of-syscall issues
1315 */
1316 userret(lp, frame, sticks);
1317
1318#ifdef KTRACE
1319 if (KTRPOINT(td, KTR_SYSRET)) {
1320 MAKEMPSAFE(have_mplock);
1321 ktrsysret(lp, code, error, args.sysmsg_result);
1322 }
1323#endif
1324
1325 /*
1326 * This works because errno is findable through the
1327 * register set. If we ever support an emulation where this
1328 * is not the case, this code will need to be revisited.
1329 */
1330 STOPEVENT(p, S_SCX, code);
1331
1332 userexit(lp);
1333#ifdef SMP
1334 /*
1335 * Release the MP lock if we had to get it
1336 */
da673940
JG
1337 if (have_mplock)
1338 rel_mplock();
1339#endif
1340 KTR_LOG(kernentry_syscall_ret, lp->lwp_proc->p_pid, lp->lwp_tid, error);
1341#ifdef INVARIANTS
3933a3ab 1342 KASSERT(&td->td_toks_base == td->td_toks_stop,
da673940 1343 ("syscall: critical section count mismatch! %d/%d",
f9235b6d 1344 crit_count, td->td_pri));
3933a3ab
MD
1345 KASSERT(curstop == td->td_toks_stop,
1346 ("syscall: extra tokens held after trap! %ld",
1347 td->td_toks_stop - &td->td_toks_base));
da673940
JG
1348#endif
1349}
1350
2b0bd8aa
MD
1351/*
1352 * NOTE: mplock not held at any point
1353 */
da673940
JG
1354void
1355fork_return(struct lwp *lp, struct trapframe *frame)
1356{
1357 frame->tf_rax = 0; /* Child returns zero */
1358 frame->tf_rflags &= ~PSL_C; /* success */
1359 frame->tf_rdx = 1;
1360
1361 generic_lwp_return(lp, frame);
1362 KTR_LOG(kernentry_fork_ret, lp->lwp_proc->p_pid, lp->lwp_tid);
1363}
1364
1365/*
1366 * Simplified back end of syscall(), used when returning from fork()
2b0bd8aa
MD
1367 * directly into user mode.
1368 *
1369 * This code will return back into the fork trampoline code which then
1370 * runs doreti.
1371 *
1372 * NOTE: The mplock is not held at any point.
da673940
JG
1373 */
1374void
1375generic_lwp_return(struct lwp *lp, struct trapframe *frame)
1376{
1377 struct proc *p = lp->lwp_proc;
1378
1379 /*
1380 * Newly forked processes are given a kernel priority. We have to
1381 * adjust the priority to a normal user priority and fake entry
1382 * into the kernel (call userenter()) to install a passive release
1383 * function just in case userret() decides to stop the process. This
1384 * can occur when ^Z races a fork. If we do not install the passive
1385 * release function the current process designation will not be
1386 * released when the thread goes to sleep.
1387 */
1388 lwkt_setpri_self(TDPRI_USER_NORM);
73e24181 1389 userenter(lp->lwp_thread, p);
da673940
JG
1390 userret(lp, frame, 0);
1391#ifdef KTRACE
1392 if (KTRPOINT(lp->lwp_thread, KTR_SYSRET))
1393 ktrsysret(lp, SYS_fork, 0, 0);
1394#endif
4643740a 1395 lp->lwp_flags |= LWP_PASSIVE_ACQ;
da673940 1396 userexit(lp);
4643740a 1397 lp->lwp_flags &= ~LWP_PASSIVE_ACQ;
da673940
JG
1398}
1399
1400/*
1401 * doreti has turned into this. The frame is directly on the stack. We
1402 * pull everything else we need (fpu and tls context) from the current
1403 * thread.
1404 *
1405 * Note on fpu interactions: In a virtual kernel, the fpu context for
1406 * an emulated user mode process is not shared with the virtual kernel's
1407 * fpu context, so we only have to 'stack' fpu contexts within the virtual
1408 * kernel itself, and not even then since the signal() contexts that we care
1409 * about save and restore the FPU state (I think anyhow).
1410 *
1411 * vmspace_ctl() returns an error only if it had problems instaling the
1412 * context we supplied or problems copying data to/from our VM space.
1413 */
1414void
1415go_user(struct intrframe *frame)
1416{
1417 struct trapframe *tf = (void *)&frame->if_rdi;
1418 int r;
1419
1420 /*
1421 * Interrupts may be disabled on entry, make sure all signals
1422 * can be received before beginning our loop.
1423 */
1424 sigsetmask(0);
1425
1426 /*
1427 * Switch to the current simulated user process, then call
1428 * user_trap() when we break out of it (usually due to a signal).
1429 */
1430 for (;;) {
1431 /*
1432 * Tell the real kernel whether it is ok to use the FP
1433 * unit or not.
1434 */
1435 if (mdcpu->gd_npxthread == curthread) {
1436 tf->tf_xflags &= ~PGEX_FPFAULT;
1437 } else {
1438 tf->tf_xflags |= PGEX_FPFAULT;
1439 }
1440
1441 /*
1442 * Run emulated user process context. This call interlocks
1443 * with new mailbox signals.
1444 *
1445 * Set PGEX_U unconditionally, indicating a user frame (the
1446 * bit is normally set only by T_PAGEFLT).
1447 */
1448 r = vmspace_ctl(&curproc->p_vmspace->vm_pmap, VMSPACE_CTL_RUN,
1449 tf, &curthread->td_savevext);
1450 frame->if_xflags |= PGEX_U;
1451#if 0
1452 kprintf("GO USER %d trap %ld EVA %08lx RIP %08lx RSP %08lx XFLAGS %02lx/%02lx\n",
1453 r, tf->tf_trapno, tf->tf_addr, tf->tf_rip, tf->tf_rsp,
1454 tf->tf_xflags, frame->if_xflags);
1455#endif
1456 if (r < 0) {
1457 if (errno != EINTR)
0e6594a8 1458 panic("vmspace_ctl failed error %d", errno);
da673940
JG
1459 } else {
1460 if (tf->tf_trapno) {
1461 user_trap(tf);
1462 }
1463 }
1464 if (mycpu->gd_reqflags & RQF_AST_MASK) {
1465 tf->tf_trapno = T_ASTFLT;
1466 user_trap(tf);
1467 }
1468 tf->tf_trapno = 0;
1469 }
1470}
1471
1472/*
1473 * If PGEX_FPFAULT is set then set FP_VIRTFP in the PCB to force a T_DNA
1474 * fault (which is then passed back to the virtual kernel) if an attempt is
1475 * made to use the FP unit.
1476 *
1477 * XXX this is a fairly big hack.
1478 */
1479void
1480set_vkernel_fp(struct trapframe *frame)
1481{
1482 struct thread *td = curthread;
1483
1484 if (frame->tf_xflags & PGEX_FPFAULT) {
1485 td->td_pcb->pcb_flags |= FP_VIRTFP;
1486 if (mdcpu->gd_npxthread == td)
1487 npxexit();
1488 } else {
1489 td->td_pcb->pcb_flags &= ~FP_VIRTFP;
1490 }
1491}
0e6594a8
SW
1492
1493/*
1494 * Called from vkernel_trap() to fixup the vkernel's syscall
1495 * frame for vmspace_ctl() return.
1496 */
1497void
1498cpu_vkernel_trap(struct trapframe *frame, int error)
1499{
1500 frame->tf_rax = error;
1501 if (error)
1502 frame->tf_rflags |= PSL_C;
1503 else
1504 frame->tf_rflags &= ~PSL_C;
1505}