kernel - simplify vm pager ops, add pre-faulting for zero-fill pages.
[dragonfly.git] / sys / platform / vkernel / i386 / trap.c
CommitLineData
6f7b98e0
MD
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 $
c730be20 39 * $DragonFly: src/sys/platform/vkernel/i386/trap.c,v 1.35 2008/09/09 04:06:19 dillon Exp $
6f7b98e0
MD
40 */
41
42/*
43 * 386 Trap and System call handling
44 */
45
46#include "use_isa.h"
47#include "use_npx.h"
48
49#include "opt_ddb.h"
50#include "opt_ktrace.h"
51
52#include <sys/param.h>
53#include <sys/systm.h>
54#include <sys/proc.h>
55#include <sys/pioctl.h>
56#include <sys/kernel.h>
57#include <sys/resourcevar.h>
58#include <sys/signalvar.h>
b1b4e5a6 59#include <sys/signal2.h>
6f7b98e0
MD
60#include <sys/syscall.h>
61#include <sys/sysctl.h>
62#include <sys/sysent.h>
63#include <sys/uio.h>
64#include <sys/vmmeter.h>
65#include <sys/malloc.h>
66#ifdef KTRACE
67#include <sys/ktrace.h>
68#endif
80bbc910 69#include <sys/ktr.h>
6f7b98e0
MD
70#include <sys/upcall.h>
71#include <sys/vkernel.h>
72#include <sys/sysproto.h>
73#include <sys/sysunion.h>
135d7199 74#include <sys/vmspace.h>
6f7b98e0
MD
75
76#include <vm/vm.h>
77#include <vm/vm_param.h>
78#include <sys/lock.h>
79#include <vm/pmap.h>
80#include <vm/vm_kern.h>
81#include <vm/vm_map.h>
82#include <vm/vm_page.h>
83#include <vm/vm_extern.h>
84
85#include <machine/cpu.h>
86#include <machine/md_var.h>
87#include <machine/pcb.h>
88#include <machine/smp.h>
89#include <machine/tss.h>
90#include <machine/globaldata.h>
91
92#include <machine/vm86.h>
93
94#include <ddb/ddb.h>
684a93c4 95
6f7b98e0
MD
96#include <sys/msgport2.h>
97#include <sys/thread2.h>
684a93c4 98#include <sys/mplock2.h>
6f7b98e0
MD
99
100#ifdef SMP
101
102#define MAKEMPSAFE(have_mplock) \
103 if (have_mplock == 0) { \
104 get_mplock(); \
105 have_mplock = 1; \
106 }
107
108#else
109
110#define MAKEMPSAFE(have_mplock)
111
112#endif
113
114int (*pmath_emulate) (struct trapframe *);
115
6f7b98e0 116extern int trapwrite (unsigned addr);
6f7b98e0
MD
117
118static int trap_pfault (struct trapframe *, int, vm_offset_t);
6092278a 119static void trap_fatal (struct trapframe *, int, vm_offset_t);
6f7b98e0
MD
120void dblfault_handler (void);
121
122#if 0
123extern inthand_t IDTVEC(syscall);
124#endif
125
126#define MAX_TRAP_MSG 28
127static char *trap_msg[] = {
128 "", /* 0 unused */
129 "privileged instruction fault", /* 1 T_PRIVINFLT */
130 "", /* 2 unused */
131 "breakpoint instruction fault", /* 3 T_BPTFLT */
132 "", /* 4 unused */
133 "", /* 5 unused */
134 "arithmetic trap", /* 6 T_ARITHTRAP */
135 "system forced exception", /* 7 T_ASTFLT */
136 "", /* 8 unused */
137 "general protection fault", /* 9 T_PROTFLT */
138 "trace trap", /* 10 T_TRCTRAP */
139 "", /* 11 unused */
140 "page fault", /* 12 T_PAGEFLT */
141 "", /* 13 unused */
142 "alignment fault", /* 14 T_ALIGNFLT */
143 "", /* 15 unused */
144 "", /* 16 unused */
145 "", /* 17 unused */
146 "integer divide fault", /* 18 T_DIVIDE */
147 "non-maskable interrupt trap", /* 19 T_NMI */
148 "overflow trap", /* 20 T_OFLOW */
149 "FPU bounds check fault", /* 21 T_BOUND */
150 "FPU device not available", /* 22 T_DNA */
151 "double fault", /* 23 T_DOUBLEFLT */
152 "FPU operand fetch fault", /* 24 T_FPOPFLT */
153 "invalid TSS fault", /* 25 T_TSSFLT */
154 "segment not present fault", /* 26 T_SEGNPFLT */
155 "stack fault", /* 27 T_STKFLT */
156 "machine check trap", /* 28 T_MCHK */
157};
158
159#ifdef DDB
160static int ddb_on_nmi = 1;
161SYSCTL_INT(_machdep, OID_AUTO, ddb_on_nmi, CTLFLAG_RW,
162 &ddb_on_nmi, 0, "Go to DDB on NMI");
163#endif
164static int panic_on_nmi = 1;
165SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RW,
166 &panic_on_nmi, 0, "Panic on NMI");
167static int fast_release;
168SYSCTL_INT(_machdep, OID_AUTO, fast_release, CTLFLAG_RW,
169 &fast_release, 0, "Passive Release was optimal");
170static int slow_release;
171SYSCTL_INT(_machdep, OID_AUTO, slow_release, CTLFLAG_RW,
172 &slow_release, 0, "Passive Release was nonoptimal");
173#ifdef SMP
225caa2d 174static int syscall_mpsafe = 1;
6f7b98e0
MD
175SYSCTL_INT(_kern, OID_AUTO, syscall_mpsafe, CTLFLAG_RW,
176 &syscall_mpsafe, 0, "Allow MPSAFE marked syscalls to run without BGL");
177TUNABLE_INT("kern.syscall_mpsafe", &syscall_mpsafe);
225caa2d 178static int trap_mpsafe = 1;
6f7b98e0
MD
179SYSCTL_INT(_kern, OID_AUTO, trap_mpsafe, CTLFLAG_RW,
180 &trap_mpsafe, 0, "Allow traps to mostly run without the BGL");
181TUNABLE_INT("kern.trap_mpsafe", &trap_mpsafe);
182#endif
183
184MALLOC_DEFINE(M_SYSMSG, "sysmsg", "sysmsg structure");
185extern int max_sysmsg;
186
6f7b98e0 187/*
7bd858e5 188 * Passively intercepts the thread switch function to increase
6f7b98e0
MD
189 * the thread priority from a user priority to a kernel priority, reducing
190 * syscall and trap overhead for the case where no switch occurs.
7bd858e5 191 *
d86a23e0 192 * Synchronizes td_ucred with p_ucred. This is used by system calls,
7bd858e5
MD
193 * signal handling, faults, AST traps, and anything else that enters the
194 * kernel from userland and provides the kernel with a stable read-only
195 * copy of the process ucred.
6f7b98e0 196 */
6f7b98e0 197static __inline void
d86a23e0 198userenter(struct thread *curtd, struct proc *curp)
6f7b98e0 199{
7bd858e5
MD
200 struct ucred *ocred;
201 struct ucred *ncred;
202
3824f392 203 curtd->td_release = lwkt_passive_release;
7bd858e5 204
d86a23e0
MD
205 if (curtd->td_ucred != curp->p_ucred) {
206 ncred = crhold(curp->p_ucred);
207 ocred = curtd->td_ucred;
208 curtd->td_ucred = ncred;
7bd858e5
MD
209 if (ocred)
210 crfree(ocred);
211 }
6f7b98e0
MD
212}
213
214/*
215 * Handle signals, upcalls, profiling, and other AST's and/or tasks that
216 * must be completed before we can return to or try to return to userland.
217 *
218 * Note that td_sticks is a 64 bit quantity, but there's no point doing 64
219 * arithmatic on the delta calculation so the absolute tick values are
220 * truncated to an integer.
221 */
222static void
223userret(struct lwp *lp, struct trapframe *frame, int sticks)
224{
225 struct proc *p = lp->lwp_proc;
226 int sig;
227
228 /*
229 * Charge system time if profiling. Note: times are in microseconds.
230 * This may do a copyout and block, so do it first even though it
231 * means some system time will be charged as user time.
232 */
233 if (p->p_flag & P_PROFIL) {
234 addupc_task(p, frame->tf_eip,
235 (u_int)((int)lp->lwp_thread->td_sticks - sticks));
236 }
237
238recheck:
70d3d461
SS
239 /*
240 * If the jungle wants us dead, so be it.
241 */
e207d1fd
MD
242 if (lp->lwp_flag & LWP_WEXIT) {
243 get_mplock();
e3161323 244 lwp_exit(0);
e207d1fd
MD
245 rel_mplock(); /* NOT REACHED */
246 }
70d3d461 247
6f7b98e0
MD
248 /*
249 * Block here if we are in a stopped state.
250 */
164b8401 251 if (p->p_stat == SSTOP) {
6f7b98e0 252 get_mplock();
9a379a4a 253 tstop();
6f7b98e0
MD
254 rel_mplock();
255 goto recheck;
256 }
257
258 /*
259 * Post any pending upcalls
260 */
261 if (p->p_flag & P_UPCALLPEND) {
6f7b98e0 262 get_mplock();
4e7c41c5 263 p->p_flag &= ~P_UPCALLPEND;
6f7b98e0
MD
264 postupcall(lp);
265 rel_mplock();
266 goto recheck;
267 }
268
269 /*
270 * Post any pending signals
271 */
e473f776 272 if ((sig = CURSIG_TRACE(lp)) != 0) {
6f7b98e0
MD
273 get_mplock();
274 postsig(sig);
275 rel_mplock();
276 goto recheck;
277 }
278
279 /*
280 * block here if we are swapped out, but still process signals
281 * (such as SIGKILL). proc0 (the swapin scheduler) is already
282 * aware of our situation, we do not have to wake it up.
283 */
284 if (p->p_flag & P_SWAPPEDOUT) {
285 get_mplock();
286 p->p_flag |= P_SWAPWAIT;
287 swapin_request();
288 if (p->p_flag & P_SWAPWAIT)
289 tsleep(p, PCATCH, "SWOUT", 0);
290 p->p_flag &= ~P_SWAPWAIT;
291 rel_mplock();
292 goto recheck;
293 }
ae7cb1b5
NT
294
295 /*
296 * Make sure postsig() handled request to restore old signal mask after
297 * running signal handler.
298 */
299 KKASSERT((lp->lwp_flag & LWP_OLDMASK) == 0);
6f7b98e0
MD
300}
301
302/*
303 * Cleanup from userenter and any passive release that might have occured.
304 * We must reclaim the current-process designation before we can return
305 * to usermode. We also handle both LWKT and USER reschedule requests.
306 */
307static __inline void
308userexit(struct lwp *lp)
309{
310 struct thread *td = lp->lwp_thread;
f5df782c 311 /* globaldata_t gd = td->td_gd; */
6f7b98e0 312
6f7b98e0 313 /*
b9eb1c19
MD
314 * Handle stop requests at kernel priority. Any requests queued
315 * after this loop will generate another AST.
6f7b98e0 316 */
b9eb1c19
MD
317 while (lp->lwp_proc->p_stat == SSTOP) {
318 get_mplock();
319 tstop();
320 rel_mplock();
fede0c7f
NT
321 }
322
6f7b98e0
MD
323 /*
324 * Reduce our priority in preparation for a return to userland. If
325 * our passive release function was still in place, our priority was
326 * never raised and does not need to be reduced.
327 */
3824f392 328 lwkt_passive_recover(td);
6f7b98e0 329
b9eb1c19
MD
330 /*
331 * Become the current user scheduled process if we aren't already,
332 * and deal with reschedule requests and other factors.
333 */
334 lp->lwp_proc->p_usched->acquire_curproc(lp);
335 /* WARNING: we may have migrated cpu's */
336 /* gd = td->td_gd; */
337}
80bbc910
SS
338
339#if !defined(KTR_KERNENTRY)
340#define KTR_KERNENTRY KTR_ALL
341#endif
342KTR_INFO_MASTER(kernentry);
343KTR_INFO(KTR_KERNENTRY, kernentry, trap, 0, "pid=%d, tid=%d, trapno=%d, eva=%p",
344 sizeof(int) + sizeof(int) + sizeof(int) + sizeof(vm_offset_t));
345KTR_INFO(KTR_KERNENTRY, kernentry, trap_ret, 0, "pid=%d, tid=%d",
346 sizeof(int) + sizeof(int));
347KTR_INFO(KTR_KERNENTRY, kernentry, syscall, 0, "pid=%d, tid=%d, call=%d",
348 sizeof(int) + sizeof(int) + sizeof(int));
349KTR_INFO(KTR_KERNENTRY, kernentry, syscall_ret, 0, "pid=%d, tid=%d, err=%d",
350 sizeof(int) + sizeof(int) + sizeof(int));
351KTR_INFO(KTR_KERNENTRY, kernentry, fork_ret, 0, "pid=%d, tid=%d",
352 sizeof(int) + sizeof(int));
353
6f7b98e0
MD
354/*
355 * Exception, fault, and trap interface to the kernel.
356 * This common code is called from assembly language IDT gate entry
357 * routines that prepare a suitable stack frame, and restore this
358 * frame after the exception has been processed.
359 *
360 * This function is also called from doreti in an interlock to handle ASTs.
361 * For example: hardwareint->INTROUTINE->(set ast)->doreti->trap
362 *
363 * NOTE! We have to retrieve the fault address prior to obtaining the
364 * MP lock because get_mplock() may switch out. YYY cr2 really ought
365 * to be retrieved by the assembly code, not here.
366 *
367 * XXX gd_trap_nesting_level currently prevents lwkt_switch() from panicing
368 * if an attempt is made to switch from a fast interrupt or IPI. This is
369 * necessary to properly take fatal kernel traps on SMP machines if
370 * get_mplock() has to block.
371 */
372
373void
6092278a 374user_trap(struct trapframe *frame)
6f7b98e0
MD
375{
376 struct globaldata *gd = mycpu;
377 struct thread *td = gd->gd_curthread;
378 struct lwp *lp = td->td_lwp;
379 struct proc *p;
380 int sticks = 0;
381 int i = 0, ucode = 0, type, code;
382#ifdef SMP
383 int have_mplock = 0;
384#endif
385#ifdef INVARIANTS
386 int crit_count = td->td_pri & ~TDPRI_MASK;
387#endif
388 vm_offset_t eva;
389
390 p = td->td_proc;
135d7199
MD
391
392 /*
393 * This is a bad kludge to avoid changing the various trapframe
394 * structures. Because we are enabled as a virtual kernel,
395 * the original tf_err field will be passed to us shifted 16
396 * over in the tf_trapno field for T_PAGEFLT.
397 */
4e7c41c5 398 if (frame->tf_trapno == T_PAGEFLT)
135d7199 399 eva = frame->tf_err;
4e7c41c5 400 else
135d7199 401 eva = 0;
4e7c41c5
MD
402#if 0
403 kprintf("USER_TRAP AT %08x xflags %d trapno %d eva %08x\n",
404 frame->tf_eip, frame->tf_xflags, frame->tf_trapno, eva);
405#endif
135d7199
MD
406
407 /*
408 * Everything coming from user mode runs through user_trap,
409 * including system calls.
410 */
411 if (frame->tf_trapno == T_SYSCALL80) {
412 syscall2(frame);
413 return;
414 }
415
80bbc910
SS
416 KTR_LOG(kernentry_trap, lp->lwp_proc->p_pid, lp->lwp_tid,
417 frame->tf_trapno, eva);
418
6f7b98e0
MD
419#ifdef DDB
420 if (db_active) {
6092278a 421 eva = (frame->tf_trapno == T_PAGEFLT ? rcr2() : 0);
6f7b98e0
MD
422 ++gd->gd_trap_nesting_level;
423 MAKEMPSAFE(have_mplock);
6092278a 424 trap_fatal(frame, TRUE, eva);
6f7b98e0
MD
425 --gd->gd_trap_nesting_level;
426 goto out2;
427 }
428#endif
429
6f7b98e0 430 ++gd->gd_trap_nesting_level;
6f7b98e0
MD
431#ifdef SMP
432 if (trap_mpsafe == 0)
433 MAKEMPSAFE(have_mplock);
434#endif
435
436 --gd->gd_trap_nesting_level;
437
6f7b98e0
MD
438#if defined(I586_CPU) && !defined(NO_F00F_HACK)
439restart:
440#endif
6092278a
MD
441 type = frame->tf_trapno;
442 code = frame->tf_err;
6f7b98e0 443
d86a23e0 444 userenter(td, p);
6f7b98e0 445
6092278a
MD
446 sticks = (int)td->td_sticks;
447 lp->lwp_md.md_regs = frame;
6f7b98e0 448
6092278a
MD
449 switch (type) {
450 case T_PRIVINFLT: /* privileged instruction fault */
f2642d5a 451 ucode = ILL_PRVOPC;
6092278a
MD
452 i = SIGILL;
453 break;
6f7b98e0 454
6092278a
MD
455 case T_BPTFLT: /* bpt instruction fault */
456 case T_TRCTRAP: /* trace trap */
457 frame->tf_eflags &= ~PSL_T;
f2642d5a 458 ucode = TRAP_TRACE;
6092278a
MD
459 i = SIGTRAP;
460 break;
6f7b98e0 461
6092278a
MD
462 case T_ARITHTRAP: /* arithmetic trap */
463 ucode = code;
464 i = SIGFPE;
465 break;
6f7b98e0 466
6092278a
MD
467 case T_ASTFLT: /* Allow process switch */
468 mycpu->gd_cnt.v_soft++;
469 if (mycpu->gd_reqflags & RQF_AST_OWEUPC) {
470 atomic_clear_int_nonlocked(&mycpu->gd_reqflags,
471 RQF_AST_OWEUPC);
472 addupc_task(p, p->p_prof.pr_addr,
473 p->p_prof.pr_ticks);
474 }
475 goto out;
6f7b98e0 476
6092278a
MD
477 /*
478 * The following two traps can happen in
479 * vm86 mode, and, if so, we want to handle
480 * them specially.
481 */
482 case T_PROTFLT: /* general protection fault */
483 case T_STKFLT: /* stack fault */
6f7b98e0 484#if 0
6092278a
MD
485 if (frame->tf_eflags & PSL_VM) {
486 i = vm86_emulate((struct vm86frame *)frame);
487 if (i == 0)
488 goto out;
6f7b98e0 489 break;
6092278a
MD
490 }
491#endif
f2642d5a
AH
492 i = SIGBUS;
493 ucode = (type == T_PROTFLT) ? BUS_OBJERR : BUS_ADRERR;
494 break;
6092278a 495 case T_SEGNPFLT: /* segment not present fault */
f2642d5a
AH
496 i = SIGBUS;
497 ucode = BUS_ADRERR;
498 break;
6092278a
MD
499 case T_TSSFLT: /* invalid TSS fault */
500 case T_DOUBLEFLT: /* double fault */
f2642d5a
AH
501 i = SIGBUS;
502 ucode = BUS_OBJERR;
6092278a 503 default:
f2642d5a
AH
504#if 0
505 ucode = code + BUS_SEGM_FAULT ; /* XXX: ???*/
506#endif
507 ucode = BUS_OBJERR;
6092278a
MD
508 i = SIGBUS;
509 break;
510
511 case T_PAGEFLT: /* page fault */
512 MAKEMPSAFE(have_mplock);
513 i = trap_pfault(frame, TRUE, eva);
514 if (i == -1)
515 goto out;
6f7b98e0 516#if defined(I586_CPU) && !defined(NO_F00F_HACK)
6092278a
MD
517 if (i == -2)
518 goto restart;
6f7b98e0 519#endif
6092278a
MD
520 if (i == 0)
521 goto out;
6f7b98e0 522
f2642d5a 523#if 0
6092278a 524 ucode = T_PAGEFLT;
f2642d5a
AH
525#endif
526 if (i == SIGSEGV)
527 ucode = SEGV_MAPERR;
528 else
529 ucode = BUS_ADRERR;
6092278a 530 break;
6f7b98e0 531
6092278a
MD
532 case T_DIVIDE: /* integer divide fault */
533 ucode = FPE_INTDIV;
534 i = SIGFPE;
535 break;
6f7b98e0
MD
536
537#if NISA > 0
6092278a
MD
538 case T_NMI:
539 MAKEMPSAFE(have_mplock);
540 /* machine/parity/power fail/"kitchen sink" faults */
541 if (isa_nmi(code) == 0) {
6f7b98e0 542#ifdef DDB
6092278a
MD
543 /*
544 * NMI can be hooked up to a pushbutton
545 * for debugging.
546 */
547 if (ddb_on_nmi) {
548 kprintf ("NMI ... going to debugger\n");
549 kdb_trap (type, 0, frame);
550 }
6f7b98e0 551#endif /* DDB */
6092278a
MD
552 goto out2;
553 } else if (panic_on_nmi)
554 panic("NMI indicates hardware failure");
555 break;
6f7b98e0
MD
556#endif /* NISA > 0 */
557
6092278a
MD
558 case T_OFLOW: /* integer overflow fault */
559 ucode = FPE_INTOVF;
560 i = SIGFPE;
561 break;
6f7b98e0 562
6092278a
MD
563 case T_BOUND: /* bounds check fault */
564 ucode = FPE_FLTSUB;
565 i = SIGFPE;
566 break;
6f7b98e0 567
6092278a 568 case T_DNA:
431d0fef
MD
569 /*
570 * Virtual kernel intercept - pass the DNA exception
571 * to the (emulated) virtual kernel if it asked to handle
572 * it. This occurs when the virtual kernel is holding
573 * onto the FP context for a different emulated
574 * process then the one currently running.
575 *
576 * We must still call npxdna() since we may have
577 * saved FP state that the (emulated) virtual kernel
578 * needs to hand over to a different emulated process.
579 */
39005e16 580 if (lp->lwp_vkernel && lp->lwp_vkernel->ve &&
431d0fef
MD
581 (td->td_pcb->pcb_flags & FP_VIRTFP)
582 ) {
583 npxdna(frame);
584 break;
585 }
6f7b98e0 586#if NNPX > 0
6092278a
MD
587 /*
588 * The kernel may have switched out the FP unit's
589 * state, causing the user process to take a fault
590 * when it tries to use the FP unit. Restore the
591 * state here
592 */
431d0fef 593 if (npxdna(frame))
6092278a 594 goto out;
6f7b98e0 595#endif
6092278a 596 if (!pmath_emulate) {
6f7b98e0 597 i = SIGFPE;
6092278a 598 ucode = FPE_FPU_NP_TRAP;
6f7b98e0
MD
599 break;
600 }
6092278a
MD
601 i = (*pmath_emulate)(frame);
602 if (i == 0) {
603 if (!(frame->tf_eflags & PSL_T))
6f7b98e0 604 goto out2;
6092278a
MD
605 frame->tf_eflags &= ~PSL_T;
606 i = SIGTRAP;
6f7b98e0 607 }
6092278a
MD
608 /* else ucode = emulator_only_knows() XXX */
609 break;
6f7b98e0 610
6092278a 611 case T_FPOPFLT: /* FPU operand fetch fault */
f2642d5a 612 ucode = ILL_COPROC;
6092278a
MD
613 i = SIGILL;
614 break;
615
616 case T_XMMFLT: /* SIMD floating-point exception */
617 ucode = 0; /* XXX */
618 i = SIGFPE;
619 break;
6f7b98e0
MD
620 }
621
622 /*
623 * Virtual kernel intercept - if the fault is directly related to a
624 * VM context managed by a virtual kernel then let the virtual kernel
625 * handle it.
626 */
39005e16 627 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
287ebb09 628 vkernel_trap(lp, frame);
6f7b98e0
MD
629 goto out;
630 }
631
632 /*
633 * Translate fault for emulators (e.g. Linux)
634 */
635 if (*p->p_sysent->sv_transtrap)
636 i = (*p->p_sysent->sv_transtrap)(i, type);
637
638 MAKEMPSAFE(have_mplock);
08f2f1bb 639 trapsignal(lp, i, ucode);
6f7b98e0
MD
640
641#ifdef DEBUG
642 if (type <= MAX_TRAP_MSG) {
643 uprintf("fatal process exception: %s",
644 trap_msg[type]);
645 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
646 uprintf(", fault VA = 0x%lx", (u_long)eva);
647 uprintf("\n");
648 }
649#endif
650
651out:
652#ifdef SMP
6092278a 653 KASSERT(td->td_mpcount == have_mplock, ("badmpcount trap/end from %p", (void *)frame->tf_eip));
6f7b98e0 654#endif
6092278a 655 userret(lp, frame, sticks);
6f7b98e0
MD
656 userexit(lp);
657out2: ;
658#ifdef SMP
659 if (have_mplock)
660 rel_mplock();
661#endif
80bbc910 662 KTR_LOG(kernentry_trap_ret, lp->lwp_proc->p_pid, lp->lwp_tid);
6f7b98e0
MD
663#ifdef INVARIANTS
664 KASSERT(crit_count == (td->td_pri & ~TDPRI_MASK),
665 ("syscall: critical section count mismatch! %d/%d",
666 crit_count / TDPRI_CRIT, td->td_pri / TDPRI_CRIT));
667#endif
668}
669
6092278a
MD
670void
671kern_trap(struct trapframe *frame)
6f7b98e0 672{
6092278a
MD
673 struct globaldata *gd = mycpu;
674 struct thread *td = gd->gd_curthread;
08f2f1bb 675 struct lwp *lp;
6092278a
MD
676 struct proc *p;
677 int i = 0, ucode = 0, type, code;
678#ifdef SMP
679 int have_mplock = 0;
680#endif
681#ifdef INVARIANTS
682 int crit_count = td->td_pri & ~TDPRI_MASK;
683#endif
684 vm_offset_t eva;
6f7b98e0 685
08f2f1bb 686 lp = td->td_lwp;
6092278a 687 p = td->td_proc;
6f7b98e0 688
4e7c41c5 689 if (frame->tf_trapno == T_PAGEFLT)
6092278a 690 eva = frame->tf_err;
4e7c41c5 691 else
135d7199 692 eva = 0;
135d7199
MD
693
694#ifdef DDB
695 if (db_active) {
696 ++gd->gd_trap_nesting_level;
697 MAKEMPSAFE(have_mplock);
698 trap_fatal(frame, FALSE, eva);
699 --gd->gd_trap_nesting_level;
700 goto out2;
701 }
702#endif
703
704 ++gd->gd_trap_nesting_level;
705
6092278a
MD
706#ifdef SMP
707 if (trap_mpsafe == 0)
708 MAKEMPSAFE(have_mplock);
709#endif
6f7b98e0 710
6092278a
MD
711 --gd->gd_trap_nesting_level;
712
713 type = frame->tf_trapno;
714 code = frame->tf_err;
715
716#if 0
717kernel_trap:
718#endif
719 /* kernel trap */
720
721 switch (type) {
722 case T_PAGEFLT: /* page fault */
723 MAKEMPSAFE(have_mplock);
724 trap_pfault(frame, FALSE, eva);
725 goto out2;
6f7b98e0 726
6092278a
MD
727 case T_DNA:
728#if NNPX > 0
6f7b98e0 729 /*
6092278a
MD
730 * The kernel may be using npx for copying or other
731 * purposes.
6f7b98e0 732 */
431d0fef
MD
733 panic("kernel NPX should not happen");
734 if (npxdna(frame))
6092278a
MD
735 goto out2;
736#endif
737 break;
6f7b98e0 738
6092278a
MD
739 case T_PROTFLT: /* general protection fault */
740 case T_SEGNPFLT: /* segment not present fault */
6f7b98e0 741 /*
6092278a
MD
742 * Invalid segment selectors and out of bounds
743 * %eip's and %esp's can be set up in user mode.
744 * This causes a fault in kernel mode when the
745 * kernel tries to return to user mode. We want
746 * to get this fault so that we can fix the
747 * problem here and not have to check all the
748 * selectors and pointers when the user changes
749 * them.
6f7b98e0 750 */
6092278a
MD
751 if (mycpu->gd_intr_nesting_level == 0) {
752 if (td->td_pcb->pcb_onfault) {
753 frame->tf_eip =
754 (register_t)td->td_pcb->pcb_onfault;
755 goto out2;
756 }
6f7b98e0 757 }
6092278a 758 break;
6f7b98e0 759
6092278a 760 case T_TSSFLT:
6f7b98e0 761 /*
6092278a
MD
762 * PSL_NT can be set in user mode and isn't cleared
763 * automatically when the kernel is entered. This
764 * causes a TSS fault when the kernel attempts to
765 * `iret' because the TSS link is uninitialized. We
766 * want to get this fault so that we can fix the
767 * problem here and not every time the kernel is
768 * entered.
6f7b98e0 769 */
6092278a
MD
770 if (frame->tf_eflags & PSL_NT) {
771 frame->tf_eflags &= ~PSL_NT;
772 goto out2;
773 }
774 break;
6f7b98e0 775
6092278a
MD
776 case T_TRCTRAP: /* trace trap */
777#if 0
778 if (frame->tf_eip == (int)IDTVEC(syscall)) {
779 /*
780 * We've just entered system mode via the
781 * syscall lcall. Continue single stepping
782 * silently until the syscall handler has
783 * saved the flags.
784 */
785 goto out2;
786 }
787 if (frame->tf_eip == (int)IDTVEC(syscall) + 1) {
788 /*
789 * The syscall handler has now saved the
790 * flags. Stop single stepping it.
791 */
792 frame->tf_eflags &= ~PSL_T;
793 goto out2;
794 }
795#endif
796#if 0
6f7b98e0 797 /*
6092278a
MD
798 * Ignore debug register trace traps due to
799 * accesses in the user's address space, which
800 * can happen under several conditions such as
801 * if a user sets a watchpoint on a buffer and
802 * then passes that buffer to a system call.
803 * We still want to get TRCTRAPS for addresses
804 * in kernel space because that is useful when
805 * debugging the kernel.
6f7b98e0 806 */
6092278a
MD
807 if (user_dbreg_trap()) {
808 /*
809 * Reset breakpoint bits because the
810 * processor doesn't
811 */
812 load_dr6(rdr6() & 0xfffffff0);
813 goto out2;
6f7b98e0 814 }
6092278a
MD
815#endif
816 /*
817 * Fall through (TRCTRAP kernel mode, kernel address)
818 */
819 case T_BPTFLT:
820 /*
821 * If DDB is enabled, let it handle the debugger trap.
822 * Otherwise, debugger traps "can't happen".
823 */
824#ifdef DDB
825 MAKEMPSAFE(have_mplock);
826 if (kdb_trap (type, 0, frame))
827 goto out2;
828#endif
829 break;
58e0cae5
MD
830 case T_DIVIDE:
831 MAKEMPSAFE(have_mplock);
832 trap_fatal(frame, FALSE, eva);
833 goto out2;
6092278a
MD
834 case T_NMI:
835 MAKEMPSAFE(have_mplock);
836 trap_fatal(frame, FALSE, eva);
837 goto out2;
2612135b
TS
838 case T_SYSCALL80:
839 /*
840 * Ignore this trap generated from a spurious SIGTRAP.
841 *
842 * single stepping in / syscalls leads to spurious / SIGTRAP
843 * so ignore
844 *
845 * Haiku (c) 2007 Simon 'corecode' Schubert
846 */
847 goto out2;
6f7b98e0
MD
848 }
849
6092278a
MD
850 /*
851 * Translate fault for emulators (e.g. Linux)
852 */
853 if (*p->p_sysent->sv_transtrap)
854 i = (*p->p_sysent->sv_transtrap)(i, type);
6f7b98e0 855
6092278a 856 MAKEMPSAFE(have_mplock);
08f2f1bb 857 trapsignal(lp, i, ucode);
6092278a
MD
858
859#ifdef DEBUG
860 if (type <= MAX_TRAP_MSG) {
861 uprintf("fatal process exception: %s",
862 trap_msg[type]);
863 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
864 uprintf(", fault VA = 0x%lx", (u_long)eva);
865 uprintf("\n");
866 }
867#endif
868
869out2:
870 ;
871#ifdef SMP
872 if (have_mplock)
873 rel_mplock();
874#endif
875#ifdef INVARIANTS
876 KASSERT(crit_count == (td->td_pri & ~TDPRI_MASK),
877 ("syscall: critical section count mismatch! %d/%d",
878 crit_count / TDPRI_CRIT, td->td_pri / TDPRI_CRIT));
6f7b98e0 879#endif
6092278a 880}
6f7b98e0
MD
881
882int
883trap_pfault(struct trapframe *frame, int usermode, vm_offset_t eva)
884{
885 vm_offset_t va;
886 struct vmspace *vm = NULL;
887 vm_map_t map = 0;
888 int rv = 0;
1b9d3514 889 int fault_flags;
6f7b98e0
MD
890 vm_prot_t ftype;
891 thread_t td = curthread;
287ebb09 892 struct lwp *lp = td->td_lwp;
6f7b98e0
MD
893
894 va = trunc_page(eva);
6092278a 895 if (usermode == FALSE) {
6f7b98e0 896 /*
6092278a 897 * This is a fault on kernel virtual memory.
6f7b98e0 898 */
6f7b98e0
MD
899 map = &kernel_map;
900 } else {
901 /*
902 * This is a fault on non-kernel virtual memory.
903 * vm is initialized above to NULL. If curproc is NULL
904 * or curproc->p_vmspace is NULL the fault is fatal.
905 */
287ebb09
MD
906 if (lp != NULL)
907 vm = lp->lwp_vmspace;
6f7b98e0
MD
908
909 if (vm == NULL)
910 goto nogo;
911
912 map = &vm->vm_map;
913 }
914
4e7c41c5 915 if (frame->tf_xflags & PGEX_W)
0035dca9 916 ftype = VM_PROT_READ | VM_PROT_WRITE;
6f7b98e0
MD
917 else
918 ftype = VM_PROT_READ;
919
920 if (map != &kernel_map) {
921 /*
922 * Keep swapout from messing with us during this
923 * critical time.
924 */
287ebb09 925 PHOLD(lp->lwp_proc);
6f7b98e0
MD
926
927 /*
928 * Grow the stack if necessary
929 */
930 /* grow_stack returns false only if va falls into
931 * a growable stack region and the stack growth
932 * fails. It returns true if va was not within
933 * a growable stack region, or if the stack
934 * growth succeeded.
935 */
287ebb09 936 if (!grow_stack (lp->lwp_proc, va)) {
6f7b98e0 937 rv = KERN_FAILURE;
287ebb09 938 PRELE(lp->lwp_proc);
6f7b98e0
MD
939 goto nogo;
940 }
941
1b9d3514
MD
942 /*
943 * Issue fault
944 */
945 fault_flags = 0;
946 if (usermode)
947 fault_flags |= VM_FAULT_BURST;
948 if (ftype & VM_PROT_WRITE)
949 fault_flags |= VM_FAULT_DIRTY;
950 else
951 fault_flags |= VM_FAULT_NORMAL;
952 rv = vm_fault(map, va, ftype, fault_flags);
6f7b98e0 953
287ebb09 954 PRELE(lp->lwp_proc);
6f7b98e0
MD
955 } else {
956 /*
957 * Don't have to worry about process locking or stacks in the kernel.
958 */
959 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
960 }
961
962 if (rv == KERN_SUCCESS)
963 return (0);
964nogo:
965 if (!usermode) {
966 if (td->td_gd->gd_intr_nesting_level == 0 &&
967 td->td_pcb->pcb_onfault) {
968 frame->tf_eip = (register_t)td->td_pcb->pcb_onfault;
969 return (0);
970 }
6092278a 971 trap_fatal(frame, usermode, eva);
6f7b98e0
MD
972 return (-1);
973 }
6f7b98e0
MD
974 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
975}
976
977static void
6092278a 978trap_fatal(struct trapframe *frame, int usermode, vm_offset_t eva)
6f7b98e0
MD
979{
980 int code, type, ss, esp;
981
4e7c41c5 982 code = frame->tf_xflags;
6f7b98e0
MD
983 type = frame->tf_trapno;
984
9097944f 985 if (type <= MAX_TRAP_MSG) {
6f7b98e0
MD
986 kprintf("\n\nFatal trap %d: %s while in %s mode\n",
987 type, trap_msg[type],
9097944f
MD
988 (usermode ? "user" : "kernel"));
989 }
6f7b98e0 990#ifdef SMP
6a8aa90e 991 /* two separate prints in case of a trap on an unmapped page */
6f7b98e0 992 kprintf("mp_lock = %08x; ", mp_lock);
39005e16 993 kprintf("cpuid = %d\n", mycpu->gd_cpuid);
6f7b98e0
MD
994#endif
995 if (type == T_PAGEFLT) {
d557216f 996 kprintf("fault virtual address = %p\n", (void *)eva);
6f7b98e0 997 kprintf("fault code = %s %s, %s\n",
6092278a 998 usermode ? "user" : "supervisor",
6f7b98e0
MD
999 code & PGEX_W ? "write" : "read",
1000 code & PGEX_P ? "protection violation" : "page not present");
1001 }
1002 kprintf("instruction pointer = 0x%x:0x%x\n",
1003 frame->tf_cs & 0xffff, frame->tf_eip);
9097944f 1004 if (usermode) {
6f7b98e0
MD
1005 ss = frame->tf_ss & 0xffff;
1006 esp = frame->tf_esp;
1007 } else {
1008 ss = GSEL(GDATA_SEL, SEL_KPL);
1009 esp = (int)&frame->tf_esp;
1010 }
1011 kprintf("stack pointer = 0x%x:0x%x\n", ss, esp);
1012 kprintf("frame pointer = 0x%x:0x%x\n", ss, frame->tf_ebp);
1013 kprintf("processor eflags = ");
1014 if (frame->tf_eflags & PSL_T)
1015 kprintf("trace trap, ");
1016 if (frame->tf_eflags & PSL_I)
1017 kprintf("interrupt enabled, ");
1018 if (frame->tf_eflags & PSL_NT)
1019 kprintf("nested task, ");
1020 if (frame->tf_eflags & PSL_RF)
1021 kprintf("resume, ");
1022#if 0
1023 if (frame->tf_eflags & PSL_VM)
1024 kprintf("vm86, ");
1025#endif
1026 kprintf("IOPL = %d\n", (frame->tf_eflags & PSL_IOPL) >> 12);
1027 kprintf("current process = ");
1028 if (curproc) {
1029 kprintf("%lu (%s)\n",
1030 (u_long)curproc->p_pid, curproc->p_comm ?
1031 curproc->p_comm : "");
1032 } else {
1033 kprintf("Idle\n");
1034 }
1035 kprintf("current thread = pri %d ", curthread->td_pri);
1036 if (curthread->td_pri >= TDPRI_CRIT)
1037 kprintf("(CRIT)");
1038 kprintf("\n");
1039#ifdef SMP
1040/**
1041 * XXX FIXME:
1042 * we probably SHOULD have stopped the other CPUs before now!
1043 * another CPU COULD have been touching cpl at this moment...
1044 */
1045 kprintf(" <- SMP: XXX");
1046#endif
1047 kprintf("\n");
1048
1049#ifdef KDB
1050 if (kdb_trap(&psl))
1051 return;
1052#endif
1053#ifdef DDB
1054 if ((debugger_on_panic || db_active) && kdb_trap(type, code, frame))
1055 return;
1056#endif
1057 kprintf("trap number = %d\n", type);
1058 if (type <= MAX_TRAP_MSG)
1059 panic("%s", trap_msg[type]);
1060 else
1061 panic("unknown/reserved trap");
1062}
1063
1064/*
1065 * Double fault handler. Called when a fault occurs while writing
1066 * a frame for a trap/exception onto the stack. This usually occurs
1067 * when the stack overflows (such is the case with infinite recursion,
1068 * for example).
1069 *
1070 * XXX Note that the current PTD gets replaced by IdlePTD when the
1071 * task switch occurs. This means that the stack that was active at
1072 * the time of the double fault is not available at <kstack> unless
1073 * the machine was idle when the double fault occurred. The downside
1074 * of this is that "trace <ebp>" in ddb won't work.
1075 */
1076void
1077dblfault_handler(void)
1078{
1079 struct mdglobaldata *gd = mdcpu;
1080
1081 kprintf("\nFatal double fault:\n");
1082 kprintf("eip = 0x%x\n", gd->gd_common_tss.tss_eip);
1083 kprintf("esp = 0x%x\n", gd->gd_common_tss.tss_esp);
1084 kprintf("ebp = 0x%x\n", gd->gd_common_tss.tss_ebp);
1085#ifdef SMP
6a8aa90e 1086 /* two separate prints in case of a trap on an unmapped page */
6f7b98e0 1087 kprintf("mp_lock = %08x; ", mp_lock);
39005e16 1088 kprintf("cpuid = %d\n", mycpu->gd_cpuid);
6f7b98e0
MD
1089#endif
1090 panic("double fault");
1091}
1092
1093/*
1094 * Compensate for 386 brain damage (missing URKR).
1095 * This is a little simpler than the pagefault handler in trap() because
1096 * it the page tables have already been faulted in and high addresses
1097 * are thrown out early for other reasons.
1098 */
1099int
1100trapwrite(unsigned addr)
1101{
287ebb09 1102 struct lwp *lp;
6f7b98e0
MD
1103 vm_offset_t va;
1104 struct vmspace *vm;
1105 int rv;
1106
1107 va = trunc_page((vm_offset_t)addr);
1108 /*
1109 * XXX - MAX is END. Changed > to >= for temp. fix.
1110 */
1111 if (va >= VM_MAX_USER_ADDRESS)
1112 return (1);
1113
287ebb09
MD
1114 lp = curthread->td_lwp;
1115 vm = lp->lwp_vmspace;
6f7b98e0 1116
287ebb09 1117 PHOLD(lp->lwp_proc);
6f7b98e0 1118
287ebb09
MD
1119 if (!grow_stack (lp->lwp_proc, va)) {
1120 PRELE(lp->lwp_proc);
6f7b98e0
MD
1121 return (1);
1122 }
1123
1124 /*
1125 * fault the data page
1126 */
1127 rv = vm_fault(&vm->vm_map, va, VM_PROT_WRITE, VM_FAULT_DIRTY);
1128
287ebb09 1129 PRELE(lp->lwp_proc);
6f7b98e0
MD
1130
1131 if (rv != KERN_SUCCESS)
1132 return 1;
1133
1134 return (0);
1135}
1136
1137/*
7bd858e5 1138 * syscall2 - MP aware system call request C handler
6f7b98e0 1139 *
7bd858e5
MD
1140 * A system call is essentially treated as a trap except that the
1141 * MP lock is not held on entry or return. We are responsible for
1142 * obtaining the MP lock if necessary and for handling ASTs
1143 * (e.g. a task switch) prior to return.
6f7b98e0 1144 *
7bd858e5 1145 * MPSAFE
6f7b98e0 1146 */
6f7b98e0 1147void
6092278a 1148syscall2(struct trapframe *frame)
6f7b98e0
MD
1149{
1150 struct thread *td = curthread;
1151 struct proc *p = td->td_proc;
1152 struct lwp *lp = td->td_lwp;
1153 caddr_t params;
1154 struct sysent *callp;
1155 register_t orig_tf_eflags;
1156 int sticks;
1157 int error;
1158 int narg;
1159#ifdef INVARIANTS
1160 int crit_count = td->td_pri & ~TDPRI_MASK;
1161#endif
1162#ifdef SMP
1163 int have_mplock = 0;
1164#endif
1165 u_int code;
1166 union sysunion args;
1167
80bbc910
SS
1168 KTR_LOG(kernentry_syscall, lp->lwp_proc->p_pid, lp->lwp_tid,
1169 frame->tf_eax);
1170
6f7b98e0 1171#ifdef SMP
6092278a 1172 KASSERT(td->td_mpcount == 0, ("badmpcount syscall2 from %p", (void *)frame->tf_eip));
6f7b98e0
MD
1173 if (syscall_mpsafe == 0)
1174 MAKEMPSAFE(have_mplock);
1175#endif
d86a23e0 1176 userenter(td, p); /* lazy raise our priority */
6f7b98e0
MD
1177
1178 /*
1179 * Misc
1180 */
1181 sticks = (int)td->td_sticks;
6092278a 1182 orig_tf_eflags = frame->tf_eflags;
6f7b98e0
MD
1183
1184 /*
1185 * Virtual kernel intercept - if a VM context managed by a virtual
1186 * kernel issues a system call the virtual kernel handles it, not us.
1187 * Restore the virtual kernel context and return from its system
1188 * call. The current frame is copied out to the virtual kernel.
1189 */
39005e16 1190 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
287ebb09 1191 error = vkernel_trap(lp, frame);
6092278a 1192 frame->tf_eax = error;
6f7b98e0 1193 if (error)
6092278a 1194 frame->tf_eflags |= PSL_C;
6f7b98e0
MD
1195 error = EJUSTRETURN;
1196 goto out;
1197 }
1198
1199 /*
1200 * Get the system call parameters and account for time
1201 */
6092278a
MD
1202 lp->lwp_md.md_regs = frame;
1203 params = (caddr_t)frame->tf_esp + sizeof(int);
1204 code = frame->tf_eax;
6f7b98e0
MD
1205
1206 if (p->p_sysent->sv_prepsyscall) {
1207 (*p->p_sysent->sv_prepsyscall)(
6092278a 1208 frame, (int *)(&args.nosys.sysmsg + 1),
6f7b98e0
MD
1209 &code, &params);
1210 } else {
1211 /*
1212 * Need to check if this is a 32 bit or 64 bit syscall.
1213 * fuword is MP aware.
1214 */
1215 if (code == SYS_syscall) {
1216 /*
1217 * Code is first argument, followed by actual args.
1218 */
1219 code = fuword(params);
1220 params += sizeof(int);
1221 } else if (code == SYS___syscall) {
1222 /*
1223 * Like syscall, but code is a quad, so as to maintain
1224 * quad alignment for the rest of the arguments.
1225 */
1226 code = fuword(params);
1227 params += sizeof(quad_t);
1228 }
1229 }
1230
1231 code &= p->p_sysent->sv_mask;
1232 if (code >= p->p_sysent->sv_size)
1233 callp = &p->p_sysent->sv_table[0];
1234 else
1235 callp = &p->p_sysent->sv_table[code];
1236
1237 narg = callp->sy_narg & SYF_ARGMASK;
1238
1239 /*
1240 * copyin is MP aware, but the tracing code is not
1241 */
1242 if (narg && params) {
1243 error = copyin(params, (caddr_t)(&args.nosys.sysmsg + 1),
1244 narg * sizeof(register_t));
1245 if (error) {
1246#ifdef KTRACE
1247 if (KTRPOINT(td, KTR_SYSCALL)) {
1248 MAKEMPSAFE(have_mplock);
1249
9fb04d14 1250 ktrsyscall(lp, code, narg,
6f7b98e0
MD
1251 (void *)(&args.nosys.sysmsg + 1));
1252 }
1253#endif
1254 goto bad;
1255 }
1256 }
1257
1258#ifdef KTRACE
1259 if (KTRPOINT(td, KTR_SYSCALL)) {
1260 MAKEMPSAFE(have_mplock);
9fb04d14 1261 ktrsyscall(lp, code, narg, (void *)(&args.nosys.sysmsg + 1));
6f7b98e0
MD
1262 }
1263#endif
1264
1265 /*
1266 * For traditional syscall code edx is left untouched when 32 bit
1267 * results are returned. Since edx is loaded from fds[1] when the
1268 * system call returns we pre-set it here.
1269 */
1270 args.sysmsg_fds[0] = 0;
6092278a 1271 args.sysmsg_fds[1] = frame->tf_edx;
6f7b98e0
MD
1272
1273 /*
1274 * The syscall might manipulate the trap frame. If it does it
1275 * will probably return EJUSTRETURN.
1276 */
6092278a 1277 args.sysmsg_frame = frame;
6f7b98e0
MD
1278
1279 STOPEVENT(p, S_SCE, narg); /* MP aware */
1280
6f7b98e0 1281 /*
3919ced0
MD
1282 * NOTE: All system calls run MPSAFE now. The system call itself
1283 * is responsible for getting the MP lock.
6f7b98e0 1284 */
6f7b98e0
MD
1285 error = (*callp->sy_call)(&args);
1286
aaf8b91f
MD
1287#if 0
1288 kprintf("system call %d returned %d\n", code, error);
1289#endif
1290
6f7b98e0
MD
1291out:
1292 /*
1293 * MP SAFE (we may or may not have the MP lock at this point)
1294 */
1295 switch (error) {
1296 case 0:
1297 /*
1298 * Reinitialize proc pointer `p' as it may be different
1299 * if this is a child returning from fork syscall.
1300 */
1301 p = curproc;
1302 lp = curthread->td_lwp;
6092278a
MD
1303 frame->tf_eax = args.sysmsg_fds[0];
1304 frame->tf_edx = args.sysmsg_fds[1];
1305 frame->tf_eflags &= ~PSL_C;
6f7b98e0
MD
1306 break;
1307 case ERESTART:
1308 /*
1309 * Reconstruct pc, assuming lcall $X,y is 7 bytes,
1310 * int 0x80 is 2 bytes. We saved this in tf_err.
1311 */
6092278a 1312 frame->tf_eip -= frame->tf_err;
6f7b98e0
MD
1313 break;
1314 case EJUSTRETURN:
1315 break;
1316 case EASYNC:
1317 panic("Unexpected EASYNC return value (for now)");
1318 default:
1319bad:
1320 if (p->p_sysent->sv_errsize) {
1321 if (error >= p->p_sysent->sv_errsize)
1322 error = -1; /* XXX */
1323 else
1324 error = p->p_sysent->sv_errtbl[error];
1325 }
6092278a
MD
1326 frame->tf_eax = error;
1327 frame->tf_eflags |= PSL_C;
6f7b98e0
MD
1328 break;
1329 }
1330
1331 /*
1332 * Traced syscall. trapsignal() is not MP aware.
1333 */
1334 if ((orig_tf_eflags & PSL_T) /*&& !(orig_tf_eflags & PSL_VM)*/) {
1335 MAKEMPSAFE(have_mplock);
6092278a 1336 frame->tf_eflags &= ~PSL_T;
f2642d5a 1337 trapsignal(lp, SIGTRAP, TRAP_TRACE);
6f7b98e0
MD
1338 }
1339
1340 /*
1341 * Handle reschedule and other end-of-syscall issues
1342 */
6092278a 1343 userret(lp, frame, sticks);
6f7b98e0
MD
1344
1345#ifdef KTRACE
1346 if (KTRPOINT(td, KTR_SYSRET)) {
1347 MAKEMPSAFE(have_mplock);
9fb04d14 1348 ktrsysret(lp, code, error, args.sysmsg_result);
6f7b98e0
MD
1349 }
1350#endif
1351
1352 /*
1353 * This works because errno is findable through the
1354 * register set. If we ever support an emulation where this
1355 * is not the case, this code will need to be revisited.
1356 */
1357 STOPEVENT(p, S_SCX, code);
1358
1359 userexit(lp);
1360#ifdef SMP
1361 /*
1362 * Release the MP lock if we had to get it
1363 */
1364 KASSERT(td->td_mpcount == have_mplock,
6092278a 1365 ("badmpcount syscall2/end from %p", (void *)frame->tf_eip));
6f7b98e0
MD
1366 if (have_mplock)
1367 rel_mplock();
1368#endif
80bbc910 1369 KTR_LOG(kernentry_syscall_ret, lp->lwp_proc->p_pid, lp->lwp_tid, error);
6f7b98e0
MD
1370#ifdef INVARIANTS
1371 KASSERT(crit_count == (td->td_pri & ~TDPRI_MASK),
1372 ("syscall: critical section count mismatch! %d/%d",
1373 crit_count / TDPRI_CRIT, td->td_pri / TDPRI_CRIT));
1374#endif
1375}
1376
91bd9c1e
SS
1377void
1378fork_return(struct lwp *lp, struct trapframe *frame)
1379{
1380 frame->tf_eax = 0; /* Child returns zero */
1381 frame->tf_eflags &= ~PSL_C; /* success */
1382 frame->tf_edx = 1;
1383
1384 generic_lwp_return(lp, frame);
80bbc910 1385 KTR_LOG(kernentry_fork_ret, lp->lwp_proc->p_pid, lp->lwp_tid);
91bd9c1e
SS
1386}
1387
6f7b98e0
MD
1388/*
1389 * Simplified back end of syscall(), used when returning from fork()
91bd9c1e
SS
1390 * or lwp_create() directly into user mode. MP lock is held on entry and
1391 * should be released on return. This code will return back into the fork
6f7b98e0
MD
1392 * trampoline code which then runs doreti.
1393 */
1394void
91bd9c1e 1395generic_lwp_return(struct lwp *lp, struct trapframe *frame)
6f7b98e0
MD
1396{
1397 struct proc *p = lp->lwp_proc;
1398
6f7b98e0
MD
1399 /*
1400 * Newly forked processes are given a kernel priority. We have to
1401 * adjust the priority to a normal user priority and fake entry
1402 * into the kernel (call userenter()) to install a passive release
1403 * function just in case userret() decides to stop the process. This
1404 * can occur when ^Z races a fork. If we do not install the passive
1405 * release function the current process designation will not be
1406 * released when the thread goes to sleep.
1407 */
1408 lwkt_setpri_self(TDPRI_USER_NORM);
d86a23e0 1409 userenter(lp->lwp_thread, p);
c7eb0589 1410 userret(lp, frame, 0);
6f7b98e0
MD
1411#ifdef KTRACE
1412 if (KTRPOINT(lp->lwp_thread, KTR_SYSRET))
9fb04d14 1413 ktrsysret(lp, SYS_fork, 0, 0);
6f7b98e0
MD
1414#endif
1415 p->p_flag |= P_PASSIVE_ACQ;
1416 userexit(lp);
1417 p->p_flag &= ~P_PASSIVE_ACQ;
1418#ifdef SMP
1419 KKASSERT(lp->lwp_thread->td_mpcount == 1);
1420 rel_mplock();
1421#endif
1422}
9ad680a3 1423
4e7c41c5
MD
1424/*
1425 * doreti has turned into this. The frame is directly on the stack. We
1426 * pull everything else we need (fpu and tls context) from the current
1427 * thread.
1428 *
1429 * Note on fpu interactions: In a virtual kernel, the fpu context for
1430 * an emulated user mode process is not shared with the virtual kernel's
1431 * fpu context, so we only have to 'stack' fpu contexts within the virtual
1432 * kernel itself, and not even then since the signal() contexts that we care
1433 * about save and restore the FPU state (I think anyhow).
1434 *
1435 * vmspace_ctl() returns an error only if it had problems instaling the
1436 * context we supplied or problems copying data to/from our VM space.
1437 */
9ad680a3 1438void
c7eb0589 1439go_user(struct intrframe *frame)
9ad680a3 1440{
c7eb0589 1441 struct trapframe *tf = (void *)&frame->if_gs;
4e7c41c5
MD
1442 int r;
1443
1fdc022f
MD
1444 /*
1445 * Interrupts may be disabled on entry, make sure all signals
1446 * can be received before beginning our loop.
1447 */
1448 sigsetmask(0);
1449
1450 /*
1451 * Switch to the current simulated user process, then call
1452 * user_trap() when we break out of it (usually due to a signal).
1453 */
135d7199 1454 for (;;) {
431d0fef
MD
1455 /*
1456 * Tell the real kernel whether it is ok to use the FP
1457 * unit or not.
eaaa6d60
MD
1458 *
1459 * The critical section is required to prevent an interrupt
1460 * from causing a preemptive task switch and changing
1461 * the FP state.
431d0fef 1462 */
eaaa6d60 1463 crit_enter();
431d0fef 1464 if (mdcpu->gd_npxthread == curthread) {
4b486183 1465 tf->tf_xflags &= ~PGEX_FPFAULT;
431d0fef 1466 } else {
4b486183 1467 tf->tf_xflags |= PGEX_FPFAULT;
431d0fef
MD
1468 }
1469
7c1212ec
MD
1470 /*
1471 * Run emulated user process context. This call interlocks
1472 * with new mailbox signals.
4b486183
MD
1473 *
1474 * Set PGEX_U unconditionally, indicating a user frame (the
1475 * bit is normally set only by T_PAGEFLT).
7c1212ec
MD
1476 */
1477 r = vmspace_ctl(&curproc->p_vmspace->vm_pmap, VMSPACE_CTL_RUN,
1478 tf, &curthread->td_savevext);
eaaa6d60 1479 crit_exit();
c7eb0589 1480 frame->if_xflags |= PGEX_U;
4b486183
MD
1481#if 0
1482 kprintf("GO USER %d trap %d EVA %08x EIP %08x ESP %08x XFLAGS %02x/%02x\n",
1483 r, tf->tf_trapno, tf->tf_err, tf->tf_eip, tf->tf_esp,
c7eb0589 1484 tf->tf_xflags, frame->if_xflags);
4b486183 1485#endif
7c1212ec 1486 if (r < 0) {
24eb47e0 1487 if (errno != EINTR)
7c1212ec 1488 panic("vmspace_ctl failed");
aaf8b91f 1489 } else {
7c1212ec
MD
1490 if (tf->tf_trapno) {
1491 user_trap(tf);
7c1212ec 1492 }
aaf8b91f 1493 }
6179d4f3
MD
1494 if (mycpu->gd_reqflags & RQF_AST_MASK) {
1495 tf->tf_trapno = T_ASTFLT;
1496 user_trap(tf);
1497 }
1498 tf->tf_trapno = 0;
135d7199 1499 }
9ad680a3 1500}
135d7199 1501
431d0fef 1502/*
4b486183 1503 * If PGEX_FPFAULT is set then set FP_VIRTFP in the PCB to force a T_DNA
431d0fef
MD
1504 * fault (which is then passed back to the virtual kernel) if an attempt is
1505 * made to use the FP unit.
1506 *
1507 * XXX this is a fairly big hack.
1508 */
1509void
1510set_vkernel_fp(struct trapframe *frame)
1511{
1512 struct thread *td = curthread;
1513
4b486183 1514 if (frame->tf_xflags & PGEX_FPFAULT) {
431d0fef
MD
1515 td->td_pcb->pcb_flags |= FP_VIRTFP;
1516 if (mdcpu->gd_npxthread == td)
1517 npxexit();
1518 } else {
1519 td->td_pcb->pcb_flags &= ~FP_VIRTFP;
1520 }
1521}
1522