kernel - simplify vm pager ops, add pre-faulting for zero-fill pages.
[dragonfly.git] / sys / platform / pc32 / i386 / trap.c
CommitLineData
984263bc
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/pc32/i386/trap.c,v 1.115 2008/09/09 04:06:17 dillon Exp $
984263bc
MD
40 */
41
42/*
43 * 386 Trap and System call handling
44 */
45
1f2de5d4
MD
46#include "use_isa.h"
47#include "use_npx.h"
48
984263bc
MD
49#include "opt_cpu.h"
50#include "opt_ddb.h"
51#include "opt_ktrace.h"
52#include "opt_clock.h"
53#include "opt_trap.h"
54
55#include <sys/param.h>
56#include <sys/systm.h>
57#include <sys/proc.h>
58#include <sys/pioctl.h>
59#include <sys/kernel.h>
60#include <sys/resourcevar.h>
61#include <sys/signalvar.h>
b1b4e5a6 62#include <sys/signal2.h>
984263bc
MD
63#include <sys/syscall.h>
64#include <sys/sysctl.h>
65#include <sys/sysent.h>
66#include <sys/uio.h>
67#include <sys/vmmeter.h>
4fd10eb6 68#include <sys/malloc.h>
984263bc
MD
69#ifdef KTRACE
70#include <sys/ktrace.h>
71#endif
80bbc910 72#include <sys/ktr.h>
a722be49 73#include <sys/upcall.h>
4a22e893 74#include <sys/vkernel.h>
a64ba182
MD
75#include <sys/sysproto.h>
76#include <sys/sysunion.h>
984263bc
MD
77
78#include <vm/vm.h>
79#include <vm/vm_param.h>
80#include <sys/lock.h>
81#include <vm/pmap.h>
82#include <vm/vm_kern.h>
83#include <vm/vm_map.h>
84#include <vm/vm_page.h>
85#include <vm/vm_extern.h>
86
87#include <machine/cpu.h>
984263bc
MD
88#include <machine/md_var.h>
89#include <machine/pcb.h>
984263bc 90#include <machine/smp.h>
984263bc 91#include <machine/tss.h>
431d0fef 92#include <machine/specialreg.h>
85100692 93#include <machine/globaldata.h>
984263bc 94
a9295349 95#include <machine_base/isa/intr_machdep.h>
984263bc
MD
96
97#ifdef POWERFAIL_NMI
98#include <sys/syslog.h>
99#include <machine/clock.h>
100#endif
101
102#include <machine/vm86.h>
103
104#include <ddb/ddb.h>
684a93c4 105
245e4f17 106#include <sys/msgport2.h>
41c20dac 107#include <sys/thread2.h>
684a93c4 108#include <sys/mplock2.h>
984263bc 109
9753ab6c
MD
110#ifdef SMP
111
112#define MAKEMPSAFE(have_mplock) \
113 if (have_mplock == 0) { \
114 get_mplock(); \
115 have_mplock = 1; \
116 }
117
118#else
119
120#define MAKEMPSAFE(have_mplock)
121
122#endif
123
3ae0cd58 124int (*pmath_emulate) (struct trapframe *);
984263bc 125
c7eb0589 126extern void trap (struct trapframe *frame);
3ae0cd58 127extern int trapwrite (unsigned addr);
c7eb0589 128extern void syscall2 (struct trapframe *frame);
984263bc 129
3ae0cd58
RG
130static int trap_pfault (struct trapframe *, int, vm_offset_t);
131static void trap_fatal (struct trapframe *, vm_offset_t);
132void dblfault_handler (void);
984263bc
MD
133
134extern inthand_t IDTVEC(syscall);
135
136#define MAX_TRAP_MSG 28
137static char *trap_msg[] = {
138 "", /* 0 unused */
139 "privileged instruction fault", /* 1 T_PRIVINFLT */
140 "", /* 2 unused */
141 "breakpoint instruction fault", /* 3 T_BPTFLT */
142 "", /* 4 unused */
143 "", /* 5 unused */
144 "arithmetic trap", /* 6 T_ARITHTRAP */
145 "system forced exception", /* 7 T_ASTFLT */
146 "", /* 8 unused */
147 "general protection fault", /* 9 T_PROTFLT */
148 "trace trap", /* 10 T_TRCTRAP */
149 "", /* 11 unused */
150 "page fault", /* 12 T_PAGEFLT */
151 "", /* 13 unused */
152 "alignment fault", /* 14 T_ALIGNFLT */
153 "", /* 15 unused */
154 "", /* 16 unused */
155 "", /* 17 unused */
156 "integer divide fault", /* 18 T_DIVIDE */
157 "non-maskable interrupt trap", /* 19 T_NMI */
158 "overflow trap", /* 20 T_OFLOW */
159 "FPU bounds check fault", /* 21 T_BOUND */
160 "FPU device not available", /* 22 T_DNA */
161 "double fault", /* 23 T_DOUBLEFLT */
162 "FPU operand fetch fault", /* 24 T_FPOPFLT */
163 "invalid TSS fault", /* 25 T_TSSFLT */
164 "segment not present fault", /* 26 T_SEGNPFLT */
165 "stack fault", /* 27 T_STKFLT */
166 "machine check trap", /* 28 T_MCHK */
167};
168
984263bc
MD
169#if defined(I586_CPU) && !defined(NO_F00F_HACK)
170extern int has_f00f_bug;
171#endif
172
173#ifdef DDB
174static int ddb_on_nmi = 1;
175SYSCTL_INT(_machdep, OID_AUTO, ddb_on_nmi, CTLFLAG_RW,
176 &ddb_on_nmi, 0, "Go to DDB on NMI");
177#endif
178static int panic_on_nmi = 1;
179SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RW,
180 &panic_on_nmi, 0, "Panic on NMI");
d9eea1a5
MD
181static int fast_release;
182SYSCTL_INT(_machdep, OID_AUTO, fast_release, CTLFLAG_RW,
183 &fast_release, 0, "Passive Release was optimal");
184static int slow_release;
185SYSCTL_INT(_machdep, OID_AUTO, slow_release, CTLFLAG_RW,
186 &slow_release, 0, "Passive Release was nonoptimal");
0e6beaa3 187#ifdef SMP
74742609 188static int syscall_mpsafe = 1;
270ac911
MD
189SYSCTL_INT(_kern, OID_AUTO, syscall_mpsafe, CTLFLAG_RW,
190 &syscall_mpsafe, 0, "Allow MPSAFE marked syscalls to run without BGL");
191TUNABLE_INT("kern.syscall_mpsafe", &syscall_mpsafe);
74742609 192static int trap_mpsafe = 1;
9753ab6c
MD
193SYSCTL_INT(_kern, OID_AUTO, trap_mpsafe, CTLFLAG_RW,
194 &trap_mpsafe, 0, "Allow traps to mostly run without the BGL");
195TUNABLE_INT("kern.trap_mpsafe", &trap_mpsafe);
0e6beaa3 196#endif
984263bc 197
4fd10eb6 198MALLOC_DEFINE(M_SYSMSG, "sysmsg", "sysmsg structure");
7062f5b4 199extern int max_sysmsg;
4fd10eb6 200
26a0694b 201/*
7bd858e5
MD
202 * Passively intercepts the thread switch function to increase the thread
203 * priority from a user priority to a kernel priority, reducing
d81ccc3e 204 * syscall and trap overhead for the case where no switch occurs.
7bd858e5 205 *
d86a23e0 206 * Synchronizes td_ucred with p_ucred. This is used by system calls,
7bd858e5
MD
207 * signal handling, faults, AST traps, and anything else that enters the
208 * kernel from userland and provides the kernel with a stable read-only
209 * copy of the process ucred.
26a0694b
MD
210 */
211static __inline void
d86a23e0 212userenter(struct thread *curtd, struct proc *curp)
26a0694b 213{
7bd858e5
MD
214 struct ucred *ocred;
215 struct ucred *ncred;
216
3824f392 217 curtd->td_release = lwkt_passive_release;
7bd858e5 218
d86a23e0
MD
219 if (curtd->td_ucred != curp->p_ucred) {
220 ncred = crhold(curp->p_ucred);
221 ocred = curtd->td_ucred;
222 curtd->td_ucred = ncred;
7bd858e5
MD
223 if (ocred)
224 crfree(ocred);
225 }
226
26a0694b
MD
227}
228
0a3f9b47 229/*
8ec60c3f
MD
230 * Handle signals, upcalls, profiling, and other AST's and/or tasks that
231 * must be completed before we can return to or try to return to userland.
0a3f9b47 232 *
8ec60c3f
MD
233 * Note that td_sticks is a 64 bit quantity, but there's no point doing 64
234 * arithmatic on the delta calculation so the absolute tick values are
235 * truncated to an integer.
37af14fe 236 */
a2a5ad0d 237static void
81eea9f4 238userret(struct lwp *lp, struct trapframe *frame, int sticks)
a2a5ad0d 239{
81eea9f4 240 struct proc *p = lp->lwp_proc;
a2a5ad0d
MD
241 int sig;
242
344ad853
MD
243 /*
244 * Charge system time if profiling. Note: times are in microseconds.
245 * This may do a copyout and block, so do it first even though it
246 * means some system time will be charged as user time.
247 */
248 if (p->p_flag & P_PROFIL) {
249 addupc_task(p, frame->tf_eip,
bb3cd951 250 (u_int)((int)lp->lwp_thread->td_sticks - sticks));
344ad853
MD
251 }
252
253recheck:
70d3d461
SS
254 /*
255 * If the jungle wants us dead, so be it.
256 */
e207d1fd
MD
257 if (lp->lwp_flag & LWP_WEXIT) {
258 get_mplock();
e3161323 259 lwp_exit(0);
e207d1fd
MD
260 rel_mplock(); /* NOT REACHED */
261 }
70d3d461 262
344ad853
MD
263 /*
264 * Block here if we are in a stopped state.
265 */
164b8401 266 if (p->p_stat == SSTOP) {
144ce500 267 get_mplock();
9a379a4a 268 tstop();
144ce500 269 rel_mplock();
344ad853
MD
270 goto recheck;
271 }
272
a722be49 273 /*
4e7c41c5
MD
274 * Post any pending upcalls. If running a virtual kernel be sure
275 * to restore the virtual kernel's vmspace before posting the upcall.
a722be49
MD
276 */
277 if (p->p_flag & P_UPCALLPEND) {
278 p->p_flag &= ~P_UPCALLPEND;
144ce500 279 get_mplock();
81eea9f4 280 postupcall(lp);
144ce500 281 rel_mplock();
344ad853 282 goto recheck;
a722be49
MD
283 }
284
a2a5ad0d 285 /*
4e7c41c5
MD
286 * Post any pending signals. If running a virtual kernel be sure
287 * to restore the virtual kernel's vmspace before posting the signal.
a2a5ad0d 288 */
e473f776 289 if ((sig = CURSIG_TRACE(lp)) != 0) {
144ce500 290 get_mplock();
a2a5ad0d 291 postsig(sig);
144ce500 292 rel_mplock();
344ad853 293 goto recheck;
a2a5ad0d 294 }
26a0694b 295
984263bc 296 /*
344ad853
MD
297 * block here if we are swapped out, but still process signals
298 * (such as SIGKILL). proc0 (the swapin scheduler) is already
299 * aware of our situation, we do not have to wake it up.
984263bc 300 */
344ad853 301 if (p->p_flag & P_SWAPPEDOUT) {
144ce500 302 get_mplock();
344ad853
MD
303 p->p_flag |= P_SWAPWAIT;
304 swapin_request();
305 if (p->p_flag & P_SWAPWAIT)
306 tsleep(p, PCATCH, "SWOUT", 0);
307 p->p_flag &= ~P_SWAPWAIT;
144ce500 308 rel_mplock();
344ad853 309 goto recheck;
984263bc 310 }
ae7cb1b5
NT
311
312 /*
313 * Make sure postsig() handled request to restore old signal mask after
314 * running signal handler.
315 */
316 KKASSERT((lp->lwp_flag & LWP_OLDMASK) == 0);
984263bc
MD
317}
318
8ec60c3f
MD
319/*
320 * Cleanup from userenter and any passive release that might have occured.
321 * We must reclaim the current-process designation before we can return
322 * to usermode. We also handle both LWKT and USER reschedule requests.
323 */
324static __inline void
81eea9f4 325userexit(struct lwp *lp)
8ec60c3f 326{
81eea9f4 327 struct thread *td = lp->lwp_thread;
8397969c 328 /* globaldata_t gd = td->td_gd; */
8ec60c3f 329
8ec60c3f 330 /*
b9eb1c19
MD
331 * Handle stop requests at kernel priority. Any requests queued
332 * after this loop will generate another AST.
8ec60c3f 333 */
b9eb1c19
MD
334 while (lp->lwp_proc->p_stat == SSTOP) {
335 get_mplock();
336 tstop();
337 rel_mplock();
fede0c7f
NT
338 }
339
8ec60c3f
MD
340 /*
341 * Reduce our priority in preparation for a return to userland. If
342 * our passive release function was still in place, our priority was
343 * never raised and does not need to be reduced.
344 */
3824f392 345 lwkt_passive_recover(td);
b9eb1c19
MD
346
347 /*
348 * Become the current user scheduled process if we aren't already,
349 * and deal with reschedule requests and other factors.
350 */
351 lp->lwp_proc->p_usched->acquire_curproc(lp);
352 /* WARNING: we may have migrated cpu's */
353 /* gd = td->td_gd; */
8ec60c3f
MD
354}
355
80bbc910
SS
356#if !defined(KTR_KERNENTRY)
357#define KTR_KERNENTRY KTR_ALL
358#endif
359KTR_INFO_MASTER(kernentry);
360KTR_INFO(KTR_KERNENTRY, kernentry, trap, 0, "pid=%d, tid=%d, trapno=%d, eva=%p",
361 sizeof(int) + sizeof(int) + sizeof(int) + sizeof(vm_offset_t));
362KTR_INFO(KTR_KERNENTRY, kernentry, trap_ret, 0, "pid=%d, tid=%d",
363 sizeof(int) + sizeof(int));
364KTR_INFO(KTR_KERNENTRY, kernentry, syscall, 0, "pid=%d, tid=%d, call=%d",
365 sizeof(int) + sizeof(int) + sizeof(int));
366KTR_INFO(KTR_KERNENTRY, kernentry, syscall_ret, 0, "pid=%d, tid=%d, err=%d",
367 sizeof(int) + sizeof(int) + sizeof(int));
368KTR_INFO(KTR_KERNENTRY, kernentry, fork_ret, 0, "pid=%d, tid=%d",
369 sizeof(int) + sizeof(int));
370
984263bc 371/*
7517cb03 372 * Exception, fault, and trap interface to the kernel.
984263bc
MD
373 * This common code is called from assembly language IDT gate entry
374 * routines that prepare a suitable stack frame, and restore this
375 * frame after the exception has been processed.
a2a5ad0d
MD
376 *
377 * This function is also called from doreti in an interlock to handle ASTs.
378 * For example: hardwareint->INTROUTINE->(set ast)->doreti->trap
379 *
380 * NOTE! We have to retrieve the fault address prior to obtaining the
381 * MP lock because get_mplock() may switch out. YYY cr2 really ought
382 * to be retrieved by the assembly code, not here.
27e88a6e
MD
383 *
384 * XXX gd_trap_nesting_level currently prevents lwkt_switch() from panicing
385 * if an attempt is made to switch from a fast interrupt or IPI. This is
386 * necessary to properly take fatal kernel traps on SMP machines if
387 * get_mplock() has to block.
984263bc 388 */
9753ab6c 389
984263bc 390void
c7eb0589 391trap(struct trapframe *frame)
984263bc 392{
27e88a6e
MD
393 struct globaldata *gd = mycpu;
394 struct thread *td = gd->gd_curthread;
81eea9f4 395 struct lwp *lp = td->td_lwp;
7966cb69 396 struct proc *p;
37af14fe 397 int sticks = 0;
984263bc 398 int i = 0, ucode = 0, type, code;
9753ab6c
MD
399#ifdef SMP
400 int have_mplock = 0;
7527bc81
MD
401#endif
402#ifdef INVARIANTS
403 int crit_count = td->td_pri & ~TDPRI_MASK;
9753ab6c 404#endif
984263bc
MD
405 vm_offset_t eva;
406
7966cb69 407 p = td->td_proc;
984263bc
MD
408#ifdef DDB
409 if (db_active) {
c7eb0589 410 eva = (frame->tf_trapno == T_PAGEFLT ? rcr2() : 0);
27e88a6e 411 ++gd->gd_trap_nesting_level;
9753ab6c 412 MAKEMPSAFE(have_mplock);
c7eb0589 413 trap_fatal(frame, eva);
27e88a6e 414 --gd->gd_trap_nesting_level;
8a8d5d85 415 goto out2;
984263bc
MD
416 }
417#endif
418
a2a5ad0d 419 eva = 0;
27e88a6e 420 ++gd->gd_trap_nesting_level;
c7eb0589 421 if (frame->tf_trapno == T_PAGEFLT) {
a2a5ad0d
MD
422 /*
423 * For some Cyrix CPUs, %cr2 is clobbered by interrupts.
424 * This problem is worked around by using an interrupt
425 * gate for the pagefault handler. We are finally ready
426 * to read %cr2 and then must reenable interrupts.
427 *
428 * XXX this should be in the switch statement, but the
429 * NO_FOOF_HACK and VM86 goto and ifdefs obfuscate the
430 * flow of control too much for this to be obviously
431 * correct.
432 */
433 eva = rcr2();
a2a5ad0d 434 cpu_enable_intr();
a2a5ad0d 435 }
80bbc910 436
9753ab6c
MD
437#ifdef SMP
438 if (trap_mpsafe == 0)
439 MAKEMPSAFE(have_mplock);
440#endif
441
27e88a6e 442 --gd->gd_trap_nesting_level;
a2a5ad0d 443
c7eb0589 444 if (!(frame->tf_eflags & PSL_I)) {
984263bc
MD
445 /*
446 * Buggy application or kernel code has disabled interrupts
447 * and then trapped. Enabling interrupts now is wrong, but
448 * it is better than running with interrupts disabled until
449 * they are accidentally enabled later.
450 */
c7eb0589
SS
451 type = frame->tf_trapno;
452 if (ISPL(frame->tf_cs)==SEL_UPL || (frame->tf_eflags & PSL_VM)) {
9753ab6c 453 MAKEMPSAFE(have_mplock);
26be20a0 454 kprintf(
984263bc
MD
455 "pid %ld (%s): trap %d with interrupts disabled\n",
456 (long)curproc->p_pid, curproc->p_comm, type);
a2a5ad0d 457 } else if (type != T_BPTFLT && type != T_TRCTRAP) {
984263bc
MD
458 /*
459 * XXX not quite right, since this may be for a
460 * multiple fault in user mode.
461 */
9753ab6c 462 MAKEMPSAFE(have_mplock);
26be20a0 463 kprintf("kernel trap %d with interrupts disabled\n",
984263bc 464 type);
a2a5ad0d 465 }
8a8d5d85 466 cpu_enable_intr();
984263bc
MD
467 }
468
984263bc
MD
469#if defined(I586_CPU) && !defined(NO_F00F_HACK)
470restart:
471#endif
c7eb0589
SS
472 type = frame->tf_trapno;
473 code = frame->tf_err;
984263bc
MD
474
475 if (in_vm86call) {
c7eb0589 476 if (frame->tf_eflags & PSL_VM &&
984263bc 477 (type == T_PROTFLT || type == T_STKFLT)) {
96728c05 478#ifdef SMP
37af14fe 479 KKASSERT(td->td_mpcount > 0);
96728c05 480#endif
c7eb0589 481 i = vm86_emulate((struct vm86frame *)frame);
96728c05 482#ifdef SMP
37af14fe 483 KKASSERT(td->td_mpcount > 0);
96728c05 484#endif
8a8d5d85 485 if (i != 0) {
984263bc
MD
486 /*
487 * returns to original process
488 */
d3d32139 489#ifdef SMP
c7eb0589 490 vm86_trap((struct vm86frame *)frame,
d3d32139
MD
491 have_mplock);
492#else
c7eb0589 493 vm86_trap((struct vm86frame *)frame, 0);
d3d32139
MD
494#endif
495 KKASSERT(0); /* NOT REACHED */
8a8d5d85
MD
496 }
497 goto out2;
984263bc
MD
498 }
499 switch (type) {
500 /*
501 * these traps want either a process context, or
502 * assume a normal userspace trap.
503 */
504 case T_PROTFLT:
505 case T_SEGNPFLT:
c7eb0589 506 trap_fatal(frame, eva);
8a8d5d85 507 goto out2;
984263bc
MD
508 case T_TRCTRAP:
509 type = T_BPTFLT; /* kernel breakpoint */
510 /* FALL THROUGH */
511 }
512 goto kernel_trap; /* normal kernel trap handling */
513 }
514
c7eb0589 515 if ((ISPL(frame->tf_cs) == SEL_UPL) || (frame->tf_eflags & PSL_VM)) {
984263bc
MD
516 /* user trap */
517
80bbc910
SS
518 KTR_LOG(kernentry_trap, p->p_pid, lp->lwp_tid,
519 frame->tf_trapno, eva);
520
d86a23e0 521 userenter(td, p);
26a0694b 522
37af14fe 523 sticks = (int)td->td_sticks;
c7eb0589 524 lp->lwp_md.md_regs = frame;
984263bc
MD
525
526 switch (type) {
527 case T_PRIVINFLT: /* privileged instruction fault */
f2642d5a 528 ucode = ILL_COPROC;
984263bc
MD
529 i = SIGILL;
530 break;
531
532 case T_BPTFLT: /* bpt instruction fault */
533 case T_TRCTRAP: /* trace trap */
c7eb0589 534 frame->tf_eflags &= ~PSL_T;
f2642d5a 535 ucode = TRAP_TRACE;
984263bc
MD
536 i = SIGTRAP;
537 break;
538
539 case T_ARITHTRAP: /* arithmetic trap */
540 ucode = code;
541 i = SIGFPE;
542 break;
543
544 case T_ASTFLT: /* Allow process switch */
12e4aaff 545 mycpu->gd_cnt.v_soft++;
235957ed
MD
546 if (mycpu->gd_reqflags & RQF_AST_OWEUPC) {
547 atomic_clear_int_nonlocked(&mycpu->gd_reqflags,
548 RQF_AST_OWEUPC);
93328593
SS
549 addupc_task(p, p->p_prof.pr_addr,
550 p->p_prof.pr_ticks);
984263bc
MD
551 }
552 goto out;
553
554 /*
555 * The following two traps can happen in
556 * vm86 mode, and, if so, we want to handle
557 * them specially.
558 */
559 case T_PROTFLT: /* general protection fault */
560 case T_STKFLT: /* stack fault */
c7eb0589
SS
561 if (frame->tf_eflags & PSL_VM) {
562 i = vm86_emulate((struct vm86frame *)frame);
984263bc
MD
563 if (i == 0)
564 goto out;
565 break;
566 }
f2642d5a
AH
567 i = SIGBUS;
568 ucode = (type == T_PROTFLT) ? BUS_OBJERR : BUS_ADRERR;
569 break;
984263bc 570 case T_SEGNPFLT: /* segment not present fault */
f2642d5a
AH
571 i = SIGBUS;
572 ucode = BUS_ADRERR;
573 break;
984263bc
MD
574 case T_TSSFLT: /* invalid TSS fault */
575 case T_DOUBLEFLT: /* double fault */
f2642d5a
AH
576 i = SIGBUS;
577 ucode = BUS_OBJERR;
984263bc 578 default:
f2642d5a
AH
579#if 0
580 ucode = code + BUS_SEGM_FAULT ; /* XXX: ???*/
581#endif
582 ucode = BUS_OBJERR;
984263bc
MD
583 i = SIGBUS;
584 break;
585
586 case T_PAGEFLT: /* page fault */
9753ab6c 587 MAKEMPSAFE(have_mplock);
c7eb0589 588 i = trap_pfault(frame, TRUE, eva);
984263bc 589 if (i == -1)
d81ccc3e 590 goto out;
984263bc
MD
591#if defined(I586_CPU) && !defined(NO_F00F_HACK)
592 if (i == -2)
593 goto restart;
594#endif
595 if (i == 0)
596 goto out;
f2642d5a 597#if 0
984263bc 598 ucode = T_PAGEFLT;
f2642d5a
AH
599#endif
600 if (i == SIGSEGV)
601 ucode = SEGV_MAPERR;
602 else
603 ucode = BUS_ADRERR; /* XXX */
984263bc
MD
604 break;
605
606 case T_DIVIDE: /* integer divide fault */
607 ucode = FPE_INTDIV;
608 i = SIGFPE;
609 break;
610
611#if NISA > 0
612 case T_NMI:
9753ab6c 613 MAKEMPSAFE(have_mplock);
984263bc
MD
614#ifdef POWERFAIL_NMI
615 goto handle_powerfail;
616#else /* !POWERFAIL_NMI */
617 /* machine/parity/power fail/"kitchen sink" faults */
618 if (isa_nmi(code) == 0) {
619#ifdef DDB
620 /*
621 * NMI can be hooked up to a pushbutton
622 * for debugging.
623 */
624 if (ddb_on_nmi) {
26be20a0 625 kprintf ("NMI ... going to debugger\n");
c7eb0589 626 kdb_trap (type, 0, frame);
984263bc
MD
627 }
628#endif /* DDB */
8a8d5d85 629 goto out2;
984263bc
MD
630 } else if (panic_on_nmi)
631 panic("NMI indicates hardware failure");
632 break;
633#endif /* POWERFAIL_NMI */
634#endif /* NISA > 0 */
635
636 case T_OFLOW: /* integer overflow fault */
637 ucode = FPE_INTOVF;
638 i = SIGFPE;
639 break;
640
641 case T_BOUND: /* bounds check fault */
642 ucode = FPE_FLTSUB;
643 i = SIGFPE;
644 break;
645
646 case T_DNA:
431d0fef
MD
647 /*
648 * Virtual kernel intercept - pass the DNA exception
649 * to the virtual kernel if it asked to handle it.
650 * This occurs when the virtual kernel is holding
651 * onto the FP context for a different emulated
652 * process then the one currently running.
653 *
654 * We must still call npxdna() since we may have
655 * saved FP state that the virtual kernel needs
656 * to hand over to a different emulated process.
657 */
39005e16 658 if (lp->lwp_vkernel && lp->lwp_vkernel->ve &&
431d0fef
MD
659 (td->td_pcb->pcb_flags & FP_VIRTFP)
660 ) {
661 npxdna();
662 break;
663 }
664
984263bc 665#if NNPX > 0
a02705a9
MD
666 /*
667 * The kernel may have switched out the FP unit's
668 * state, causing the user process to take a fault
669 * when it tries to use the FP unit. Restore the
670 * state here
671 */
984263bc 672 if (npxdna())
d81ccc3e 673 goto out;
984263bc
MD
674#endif
675 if (!pmath_emulate) {
676 i = SIGFPE;
677 ucode = FPE_FPU_NP_TRAP;
678 break;
679 }
c7eb0589 680 i = (*pmath_emulate)(frame);
984263bc 681 if (i == 0) {
c7eb0589 682 if (!(frame->tf_eflags & PSL_T))
8a8d5d85 683 goto out2;
c7eb0589 684 frame->tf_eflags &= ~PSL_T;
984263bc
MD
685 i = SIGTRAP;
686 }
687 /* else ucode = emulator_only_knows() XXX */
688 break;
689
690 case T_FPOPFLT: /* FPU operand fetch fault */
f2642d5a 691 ucode = ILL_ILLOPN;
984263bc
MD
692 i = SIGILL;
693 break;
694
695 case T_XMMFLT: /* SIMD floating-point exception */
696 ucode = 0; /* XXX */
697 i = SIGFPE;
698 break;
699 }
700 } else {
701kernel_trap:
702 /* kernel trap */
703
704 switch (type) {
705 case T_PAGEFLT: /* page fault */
9753ab6c 706 MAKEMPSAFE(have_mplock);
c7eb0589 707 trap_pfault(frame, FALSE, eva);
8a8d5d85 708 goto out2;
984263bc
MD
709
710 case T_DNA:
711#if NNPX > 0
712 /*
a02705a9
MD
713 * The kernel may be using npx for copying or other
714 * purposes.
984263bc
MD
715 */
716 if (npxdna())
8a8d5d85 717 goto out2;
984263bc
MD
718#endif
719 break;
720
721 case T_PROTFLT: /* general protection fault */
722 case T_SEGNPFLT: /* segment not present fault */
723 /*
724 * Invalid segment selectors and out of bounds
725 * %eip's and %esp's can be set up in user mode.
726 * This causes a fault in kernel mode when the
727 * kernel tries to return to user mode. We want
728 * to get this fault so that we can fix the
729 * problem here and not have to check all the
730 * selectors and pointers when the user changes
731 * them.
732 */
733#define MAYBE_DORETI_FAULT(where, whereto) \
734 do { \
c7eb0589
SS
735 if (frame->tf_eip == (int)where) { \
736 frame->tf_eip = (int)whereto; \
8a8d5d85 737 goto out2; \
984263bc
MD
738 } \
739 } while (0)
ef0fdad1 740 if (mycpu->gd_intr_nesting_level == 0) {
984263bc
MD
741 /*
742 * Invalid %fs's and %gs's can be created using
743 * procfs or PT_SETREGS or by invalidating the
744 * underlying LDT entry. This causes a fault
745 * in kernel mode when the kernel attempts to
746 * switch contexts. Lose the bad context
747 * (XXX) so that we can continue, and generate
748 * a signal.
749 */
984263bc
MD
750 MAYBE_DORETI_FAULT(doreti_iret,
751 doreti_iret_fault);
752 MAYBE_DORETI_FAULT(doreti_popl_ds,
753 doreti_popl_ds_fault);
754 MAYBE_DORETI_FAULT(doreti_popl_es,
755 doreti_popl_es_fault);
756 MAYBE_DORETI_FAULT(doreti_popl_fs,
757 doreti_popl_fs_fault);
4e7c41c5
MD
758 MAYBE_DORETI_FAULT(doreti_popl_gs,
759 doreti_popl_gs_fault);
37af14fe 760 if (td->td_pcb->pcb_onfault) {
c7eb0589 761 frame->tf_eip =
37af14fe 762 (register_t)td->td_pcb->pcb_onfault;
8a8d5d85 763 goto out2;
984263bc
MD
764 }
765 }
766 break;
767
768 case T_TSSFLT:
769 /*
770 * PSL_NT can be set in user mode and isn't cleared
771 * automatically when the kernel is entered. This
772 * causes a TSS fault when the kernel attempts to
773 * `iret' because the TSS link is uninitialized. We
774 * want to get this fault so that we can fix the
775 * problem here and not every time the kernel is
776 * entered.
777 */
c7eb0589
SS
778 if (frame->tf_eflags & PSL_NT) {
779 frame->tf_eflags &= ~PSL_NT;
8a8d5d85 780 goto out2;
984263bc
MD
781 }
782 break;
783
784 case T_TRCTRAP: /* trace trap */
c7eb0589 785 if (frame->tf_eip == (int)IDTVEC(syscall)) {
984263bc
MD
786 /*
787 * We've just entered system mode via the
788 * syscall lcall. Continue single stepping
789 * silently until the syscall handler has
790 * saved the flags.
791 */
8a8d5d85 792 goto out2;
984263bc 793 }
c7eb0589 794 if (frame->tf_eip == (int)IDTVEC(syscall) + 1) {
984263bc
MD
795 /*
796 * The syscall handler has now saved the
797 * flags. Stop single stepping it.
798 */
c7eb0589 799 frame->tf_eflags &= ~PSL_T;
8a8d5d85 800 goto out2;
984263bc
MD
801 }
802 /*
803 * Ignore debug register trace traps due to
804 * accesses in the user's address space, which
805 * can happen under several conditions such as
806 * if a user sets a watchpoint on a buffer and
807 * then passes that buffer to a system call.
808 * We still want to get TRCTRAPS for addresses
809 * in kernel space because that is useful when
810 * debugging the kernel.
811 */
812 if (user_dbreg_trap()) {
813 /*
814 * Reset breakpoint bits because the
815 * processor doesn't
816 */
817 load_dr6(rdr6() & 0xfffffff0);
8a8d5d85 818 goto out2;
984263bc
MD
819 }
820 /*
821 * Fall through (TRCTRAP kernel mode, kernel address)
822 */
823 case T_BPTFLT:
824 /*
825 * If DDB is enabled, let it handle the debugger trap.
826 * Otherwise, debugger traps "can't happen".
827 */
f2642d5a 828 ucode = TRAP_BRKPT;
984263bc 829#ifdef DDB
9753ab6c 830 MAKEMPSAFE(have_mplock);
c7eb0589 831 if (kdb_trap (type, 0, frame))
8a8d5d85 832 goto out2;
984263bc
MD
833#endif
834 break;
835
836#if NISA > 0
837 case T_NMI:
9753ab6c 838 MAKEMPSAFE(have_mplock);
984263bc
MD
839#ifdef POWERFAIL_NMI
840#ifndef TIMER_FREQ
841# define TIMER_FREQ 1193182
842#endif
843 handle_powerfail:
844 {
845 static unsigned lastalert = 0;
846
847 if(time_second - lastalert > 10)
848 {
849 log(LOG_WARNING, "NMI: power fail\n");
850 sysbeep(TIMER_FREQ/880, hz);
851 lastalert = time_second;
852 }
8a8d5d85
MD
853 /* YYY mp count */
854 goto out2;
984263bc
MD
855 }
856#else /* !POWERFAIL_NMI */
857 /* machine/parity/power fail/"kitchen sink" faults */
858 if (isa_nmi(code) == 0) {
859#ifdef DDB
860 /*
861 * NMI can be hooked up to a pushbutton
862 * for debugging.
863 */
864 if (ddb_on_nmi) {
26be20a0 865 kprintf ("NMI ... going to debugger\n");
c7eb0589 866 kdb_trap (type, 0, frame);
984263bc
MD
867 }
868#endif /* DDB */
8a8d5d85 869 goto out2;
984263bc 870 } else if (panic_on_nmi == 0)
8a8d5d85 871 goto out2;
984263bc
MD
872 /* FALL THROUGH */
873#endif /* POWERFAIL_NMI */
874#endif /* NISA > 0 */
875 }
876
9753ab6c 877 MAKEMPSAFE(have_mplock);
c7eb0589 878 trap_fatal(frame, eva);
8a8d5d85 879 goto out2;
984263bc
MD
880 }
881
4a22e893
MD
882 /*
883 * Virtual kernel intercept - if the fault is directly related to a
884 * VM context managed by a virtual kernel then let the virtual kernel
885 * handle it.
886 */
39005e16 887 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
287ebb09 888 vkernel_trap(lp, frame);
4a22e893
MD
889 goto out;
890 }
891
892 /*
893 * Translate fault for emulators (e.g. Linux)
894 */
984263bc
MD
895 if (*p->p_sysent->sv_transtrap)
896 i = (*p->p_sysent->sv_transtrap)(i, type);
897
9753ab6c 898 MAKEMPSAFE(have_mplock);
08f2f1bb 899 trapsignal(lp, i, ucode);
984263bc
MD
900
901#ifdef DEBUG
902 if (type <= MAX_TRAP_MSG) {
903 uprintf("fatal process exception: %s",
904 trap_msg[type]);
905 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
906 uprintf(", fault VA = 0x%lx", (u_long)eva);
907 uprintf("\n");
908 }
909#endif
910
911out:
8a8d5d85 912#ifdef SMP
c7eb0589
SS
913 if (ISPL(frame->tf_cs) == SEL_UPL)
914 KASSERT(td->td_mpcount == have_mplock, ("badmpcount trap/end from %p", (void *)frame->tf_eip));
8a8d5d85 915#endif
c7eb0589 916 userret(lp, frame, sticks);
81eea9f4 917 userexit(lp);
9753ab6c 918out2: ;
96728c05 919#ifdef SMP
9753ab6c
MD
920 if (have_mplock)
921 rel_mplock();
96728c05 922#endif
80bbc910
SS
923 if (p != NULL && lp != NULL)
924 KTR_LOG(kernentry_trap_ret, p->p_pid, lp->lwp_tid);
7527bc81
MD
925#ifdef INVARIANTS
926 KASSERT(crit_count == (td->td_pri & ~TDPRI_MASK),
927 ("syscall: critical section count mismatch! %d/%d",
928 crit_count / TDPRI_CRIT, td->td_pri / TDPRI_CRIT));
929#endif
984263bc
MD
930}
931
984263bc 932int
f123d5a1 933trap_pfault(struct trapframe *frame, int usermode, vm_offset_t eva)
984263bc
MD
934{
935 vm_offset_t va;
936 struct vmspace *vm = NULL;
937 vm_map_t map = 0;
938 int rv = 0;
1b9d3514 939 int fault_flags;
984263bc 940 vm_prot_t ftype;
37af14fe 941 thread_t td = curthread;
287ebb09 942 struct lwp *lp = td->td_lwp;
984263bc
MD
943
944 va = trunc_page(eva);
945 if (va >= KERNBASE) {
946 /*
947 * Don't allow user-mode faults in kernel address space.
948 * An exception: if the faulting address is the invalid
949 * instruction entry in the IDT, then the Intel Pentium
950 * F00F bug workaround was triggered, and we need to
951 * treat it is as an illegal instruction, and not a page
952 * fault.
953 */
954#if defined(I586_CPU) && !defined(NO_F00F_HACK)
955 if ((eva == (unsigned int)&idt[6]) && has_f00f_bug) {
956 frame->tf_trapno = T_PRIVINFLT;
957 return -2;
958 }
959#endif
960 if (usermode)
961 goto nogo;
962
e4846942 963 map = &kernel_map;
984263bc
MD
964 } else {
965 /*
966 * This is a fault on non-kernel virtual memory.
967 * vm is initialized above to NULL. If curproc is NULL
968 * or curproc->p_vmspace is NULL the fault is fatal.
969 */
287ebb09
MD
970 if (lp != NULL)
971 vm = lp->lwp_vmspace;
984263bc
MD
972
973 if (vm == NULL)
974 goto nogo;
975
976 map = &vm->vm_map;
977 }
978
979 if (frame->tf_err & PGEX_W)
980 ftype = VM_PROT_WRITE;
981 else
982 ftype = VM_PROT_READ;
983
e4846942 984 if (map != &kernel_map) {
984263bc
MD
985 /*
986 * Keep swapout from messing with us during this
987 * critical time.
988 */
287ebb09 989 PHOLD(lp->lwp_proc);
984263bc
MD
990
991 /*
992 * Grow the stack if necessary
993 */
994 /* grow_stack returns false only if va falls into
995 * a growable stack region and the stack growth
996 * fails. It returns true if va was not within
997 * a growable stack region, or if the stack
998 * growth succeeded.
999 */
287ebb09 1000 if (!grow_stack(lp->lwp_proc, va)) {
984263bc 1001 rv = KERN_FAILURE;
287ebb09 1002 PRELE(lp->lwp_proc);
984263bc
MD
1003 goto nogo;
1004 }
1005
1b9d3514
MD
1006 /*
1007 * Issue fault
1008 */
1009 fault_flags = 0;
1010 if (usermode)
1011 fault_flags |= VM_FAULT_BURST;
1012 if (ftype & VM_PROT_WRITE)
1013 fault_flags |= VM_FAULT_DIRTY;
1014 else
1015 fault_flags |= VM_FAULT_NORMAL;
1016 rv = vm_fault(map, va, ftype, fault_flags);
287ebb09 1017 PRELE(lp->lwp_proc);
984263bc
MD
1018 } else {
1019 /*
17cde63e
MD
1020 * Don't have to worry about process locking or stacks
1021 * in the kernel.
984263bc
MD
1022 */
1023 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
1024 }
1025
1026 if (rv == KERN_SUCCESS)
1027 return (0);
1028nogo:
1029 if (!usermode) {
37af14fe
MD
1030 if (td->td_gd->gd_intr_nesting_level == 0 &&
1031 td->td_pcb->pcb_onfault) {
1032 frame->tf_eip = (register_t)td->td_pcb->pcb_onfault;
984263bc
MD
1033 return (0);
1034 }
1035 trap_fatal(frame, eva);
1036 return (-1);
1037 }
1038
1039 /* kludge to pass faulting virtual address to sendsig */
4e7c41c5 1040 frame->tf_xflags = frame->tf_err;
984263bc
MD
1041 frame->tf_err = eva;
1042
1043 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
1044}
1045
1046static void
f123d5a1 1047trap_fatal(struct trapframe *frame, vm_offset_t eva)
984263bc
MD
1048{
1049 int code, type, ss, esp;
1050 struct soft_segment_descriptor softseg;
1051
1052 code = frame->tf_err;
1053 type = frame->tf_trapno;
3951a45f 1054 sdtossd(&gdt[mycpu->gd_cpuid * NGDT + IDXSEL(frame->tf_cs & 0xffff)].sd, &softseg);
984263bc
MD
1055
1056 if (type <= MAX_TRAP_MSG)
26be20a0 1057 kprintf("\n\nFatal trap %d: %s while in %s mode\n",
984263bc
MD
1058 type, trap_msg[type],
1059 frame->tf_eflags & PSL_VM ? "vm86" :
1060 ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel");
1061#ifdef SMP
7a44d1cb 1062 /* three separate prints in case of a trap on an unmapped page */
26be20a0
SW
1063 kprintf("mp_lock = %08x; ", mp_lock);
1064 kprintf("cpuid = %d; ", mycpu->gd_cpuid);
1065 kprintf("lapic.id = %08x\n", lapic.id);
984263bc
MD
1066#endif
1067 if (type == T_PAGEFLT) {
d557216f 1068 kprintf("fault virtual address = %p\n", (void *)eva);
26be20a0 1069 kprintf("fault code = %s %s, %s\n",
984263bc
MD
1070 code & PGEX_U ? "user" : "supervisor",
1071 code & PGEX_W ? "write" : "read",
1072 code & PGEX_P ? "protection violation" : "page not present");
1073 }
26be20a0 1074 kprintf("instruction pointer = 0x%x:0x%x\n",
984263bc
MD
1075 frame->tf_cs & 0xffff, frame->tf_eip);
1076 if ((ISPL(frame->tf_cs) == SEL_UPL) || (frame->tf_eflags & PSL_VM)) {
1077 ss = frame->tf_ss & 0xffff;
1078 esp = frame->tf_esp;
1079 } else {
1080 ss = GSEL(GDATA_SEL, SEL_KPL);
1081 esp = (int)&frame->tf_esp;
1082 }
26be20a0
SW
1083 kprintf("stack pointer = 0x%x:0x%x\n", ss, esp);
1084 kprintf("frame pointer = 0x%x:0x%x\n", ss, frame->tf_ebp);
1085 kprintf("code segment = base 0x%x, limit 0x%x, type 0x%x\n",
984263bc 1086 softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type);
26be20a0 1087 kprintf(" = DPL %d, pres %d, def32 %d, gran %d\n",
984263bc
MD
1088 softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_def32,
1089 softseg.ssd_gran);
26be20a0 1090 kprintf("processor eflags = ");
984263bc 1091 if (frame->tf_eflags & PSL_T)
26be20a0 1092 kprintf("trace trap, ");
984263bc 1093 if (frame->tf_eflags & PSL_I)
26be20a0 1094 kprintf("interrupt enabled, ");
984263bc 1095 if (frame->tf_eflags & PSL_NT)
26be20a0 1096 kprintf("nested task, ");
984263bc 1097 if (frame->tf_eflags & PSL_RF)
26be20a0 1098 kprintf("resume, ");
984263bc 1099 if (frame->tf_eflags & PSL_VM)
26be20a0
SW
1100 kprintf("vm86, ");
1101 kprintf("IOPL = %d\n", (frame->tf_eflags & PSL_IOPL) >> 12);
1102 kprintf("current process = ");
984263bc 1103 if (curproc) {
26be20a0 1104 kprintf("%lu (%s)\n",
984263bc
MD
1105 (u_long)curproc->p_pid, curproc->p_comm ?
1106 curproc->p_comm : "");
1107 } else {
26be20a0 1108 kprintf("Idle\n");
984263bc 1109 }
26be20a0 1110 kprintf("current thread = pri %d ", curthread->td_pri);
f1d1c3fa 1111 if (curthread->td_pri >= TDPRI_CRIT)
26be20a0
SW
1112 kprintf("(CRIT)");
1113 kprintf("\n");
984263bc
MD
1114#ifdef SMP
1115/**
1116 * XXX FIXME:
1117 * we probably SHOULD have stopped the other CPUs before now!
1118 * another CPU COULD have been touching cpl at this moment...
1119 */
26be20a0 1120 kprintf(" <- SMP: XXX");
984263bc 1121#endif
26be20a0 1122 kprintf("\n");
984263bc
MD
1123
1124#ifdef KDB
1125 if (kdb_trap(&psl))
1126 return;
1127#endif
1128#ifdef DDB
f7bc9806 1129 if ((debugger_on_panic || db_active) && kdb_trap(type, code, frame))
984263bc
MD
1130 return;
1131#endif
26be20a0 1132 kprintf("trap number = %d\n", type);
984263bc
MD
1133 if (type <= MAX_TRAP_MSG)
1134 panic("%s", trap_msg[type]);
1135 else
1136 panic("unknown/reserved trap");
1137}
1138
1139/*
1140 * Double fault handler. Called when a fault occurs while writing
1141 * a frame for a trap/exception onto the stack. This usually occurs
1142 * when the stack overflows (such is the case with infinite recursion,
1143 * for example).
1144 *
1145 * XXX Note that the current PTD gets replaced by IdlePTD when the
1146 * task switch occurs. This means that the stack that was active at
1147 * the time of the double fault is not available at <kstack> unless
1148 * the machine was idle when the double fault occurred. The downside
1149 * of this is that "trace <ebp>" in ddb won't work.
1150 */
1151void
f123d5a1 1152dblfault_handler(void)
984263bc 1153{
85100692 1154 struct mdglobaldata *gd = mdcpu;
17a9f566 1155
26be20a0
SW
1156 kprintf("\nFatal double fault:\n");
1157 kprintf("eip = 0x%x\n", gd->gd_common_tss.tss_eip);
1158 kprintf("esp = 0x%x\n", gd->gd_common_tss.tss_esp);
1159 kprintf("ebp = 0x%x\n", gd->gd_common_tss.tss_ebp);
984263bc 1160#ifdef SMP
7a44d1cb 1161 /* three separate prints in case of a trap on an unmapped page */
26be20a0
SW
1162 kprintf("mp_lock = %08x; ", mp_lock);
1163 kprintf("cpuid = %d; ", mycpu->gd_cpuid);
1164 kprintf("lapic.id = %08x\n", lapic.id);
984263bc
MD
1165#endif
1166 panic("double fault");
1167}
1168
1169/*
1170 * Compensate for 386 brain damage (missing URKR).
1171 * This is a little simpler than the pagefault handler in trap() because
1172 * it the page tables have already been faulted in and high addresses
1173 * are thrown out early for other reasons.
1174 */
f123d5a1
SW
1175int
1176trapwrite(unsigned addr)
984263bc 1177{
287ebb09 1178 struct lwp *lp;
984263bc
MD
1179 vm_offset_t va;
1180 struct vmspace *vm;
1181 int rv;
1182
1183 va = trunc_page((vm_offset_t)addr);
1184 /*
1185 * XXX - MAX is END. Changed > to >= for temp. fix.
1186 */
88181b08 1187 if (va >= VM_MAX_USER_ADDRESS)
984263bc
MD
1188 return (1);
1189
287ebb09
MD
1190 lp = curthread->td_lwp;
1191 vm = lp->lwp_vmspace;
984263bc 1192
287ebb09 1193 PHOLD(lp->lwp_proc);
984263bc 1194
287ebb09
MD
1195 if (!grow_stack(lp->lwp_proc, va)) {
1196 PRELE(lp->lwp_proc);
984263bc
MD
1197 return (1);
1198 }
1199
1200 /*
1201 * fault the data page
1202 */
1203 rv = vm_fault(&vm->vm_map, va, VM_PROT_WRITE, VM_FAULT_DIRTY);
1204
287ebb09 1205 PRELE(lp->lwp_proc);
984263bc
MD
1206
1207 if (rv != KERN_SUCCESS)
1208 return 1;
1209
1210 return (0);
1211}
1212
1213/*
0d355d3b 1214 * syscall2 - MP aware system call request C handler
984263bc 1215 *
0d355d3b
MD
1216 * A system call is essentially treated as a trap. The MP lock is not
1217 * held on entry or return. We are responsible for handling ASTs
1218 * (e.g. a task switch) prior to return.
984263bc 1219 *
0d355d3b 1220 * MPSAFE
984263bc
MD
1221 */
1222void
c7eb0589 1223syscall2(struct trapframe *frame)
984263bc 1224{
dadab5e9
MD
1225 struct thread *td = curthread;
1226 struct proc *p = td->td_proc;
81eea9f4 1227 struct lwp *lp = td->td_lwp;
984263bc 1228 caddr_t params;
984263bc 1229 struct sysent *callp;
984263bc 1230 register_t orig_tf_eflags;
37af14fe 1231 int sticks;
984263bc
MD
1232 int error;
1233 int narg;
7527bc81
MD
1234#ifdef INVARIANTS
1235 int crit_count = td->td_pri & ~TDPRI_MASK;
1236#endif
270ac911
MD
1237#ifdef SMP
1238 int have_mplock = 0;
1239#endif
984263bc 1240 u_int code;
a64ba182 1241 union sysunion args;
984263bc
MD
1242
1243#ifdef DIAGNOSTIC
c7eb0589 1244 if (ISPL(frame->tf_cs) != SEL_UPL) {
984263bc
MD
1245 get_mplock();
1246 panic("syscall");
1247 /* NOT REACHED */
1248 }
1249#endif
1250
80bbc910
SS
1251 KTR_LOG(kernentry_syscall, p->p_pid, lp->lwp_tid,
1252 frame->tf_eax);
1253
8a8d5d85 1254#ifdef SMP
c7eb0589 1255 KASSERT(td->td_mpcount == 0, ("badmpcount syscall2 from %p", (void *)frame->tf_eip));
270ac911
MD
1256 if (syscall_mpsafe == 0)
1257 MAKEMPSAFE(have_mplock);
8a8d5d85 1258#endif
d86a23e0 1259 userenter(td, p); /* lazy raise our priority */
37af14fe 1260
4a22e893
MD
1261 /*
1262 * Misc
1263 */
37af14fe 1264 sticks = (int)td->td_sticks;
c7eb0589 1265 orig_tf_eflags = frame->tf_eflags;
4a22e893
MD
1266
1267 /*
1268 * Virtual kernel intercept - if a VM context managed by a virtual
1269 * kernel issues a system call the virtual kernel handles it, not us.
1270 * Restore the virtual kernel context and return from its system
1271 * call. The current frame is copied out to the virtual kernel.
1272 */
39005e16 1273 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
287ebb09 1274 error = vkernel_trap(lp, frame);
c7eb0589 1275 frame->tf_eax = error;
4a22e893 1276 if (error)
c7eb0589 1277 frame->tf_eflags |= PSL_C;
4a22e893 1278 error = EJUSTRETURN;
e7912058 1279 callp = NULL;
4a22e893
MD
1280 goto out;
1281 }
984263bc 1282
4a22e893
MD
1283 /*
1284 * Get the system call parameters and account for time
1285 */
c7eb0589
SS
1286 lp->lwp_md.md_regs = frame;
1287 params = (caddr_t)frame->tf_esp + sizeof(int);
1288 code = frame->tf_eax;
984263bc
MD
1289
1290 if (p->p_sysent->sv_prepsyscall) {
270ac911 1291 (*p->p_sysent->sv_prepsyscall)(
c7eb0589 1292 frame, (int *)(&args.nosys.sysmsg + 1),
270ac911 1293 &code, &params);
984263bc
MD
1294 } else {
1295 /*
1296 * Need to check if this is a 32 bit or 64 bit syscall.
1297 * fuword is MP aware.
1298 */
1299 if (code == SYS_syscall) {
1300 /*
1301 * Code is first argument, followed by actual args.
1302 */
1303 code = fuword(params);
1304 params += sizeof(int);
1305 } else if (code == SYS___syscall) {
1306 /*
1307 * Like syscall, but code is a quad, so as to maintain
1308 * quad alignment for the rest of the arguments.
1309 */
1310 code = fuword(params);
1311 params += sizeof(quad_t);
1312 }
1313 }
1314
8ec60c3f 1315 code &= p->p_sysent->sv_mask;
7062f5b4
EN
1316 if (code >= p->p_sysent->sv_size)
1317 callp = &p->p_sysent->sv_table[0];
1318 else
1319 callp = &p->p_sysent->sv_table[code];
984263bc
MD
1320
1321 narg = callp->sy_narg & SYF_ARGMASK;
1322
1323 /*
1324 * copyin is MP aware, but the tracing code is not
1325 */
8ec60c3f 1326 if (narg && params) {
f9a13fc4 1327 error = copyin(params, (caddr_t)(&args.nosys.sysmsg + 1),
8ec60c3f
MD
1328 narg * sizeof(register_t));
1329 if (error) {
984263bc 1330#ifdef KTRACE
270ac911
MD
1331 if (KTRPOINT(td, KTR_SYSCALL)) {
1332 MAKEMPSAFE(have_mplock);
1333
9fb04d14 1334 ktrsyscall(lp, code, narg,
f9a13fc4 1335 (void *)(&args.nosys.sysmsg + 1));
270ac911 1336 }
984263bc 1337#endif
8ec60c3f
MD
1338 goto bad;
1339 }
984263bc
MD
1340 }
1341
984263bc 1342#ifdef KTRACE
dadab5e9 1343 if (KTRPOINT(td, KTR_SYSCALL)) {
270ac911 1344 MAKEMPSAFE(have_mplock);
9fb04d14 1345 ktrsyscall(lp, code, narg, (void *)(&args.nosys.sysmsg + 1));
984263bc
MD
1346 }
1347#endif
01dce7bb
MD
1348
1349 /*
1350 * For traditional syscall code edx is left untouched when 32 bit
1351 * results are returned. Since edx is loaded from fds[1] when the
1352 * system call returns we pre-set it here.
1353 */
c7114eea 1354 args.sysmsg_fds[0] = 0;
c7eb0589 1355 args.sysmsg_fds[1] = frame->tf_edx;
984263bc 1356
4a22e893
MD
1357 /*
1358 * The syscall might manipulate the trap frame. If it does it
1359 * will probably return EJUSTRETURN.
1360 */
c7eb0589 1361 args.sysmsg_frame = frame;
4a22e893 1362
984263bc
MD
1363 STOPEVENT(p, S_SCE, narg); /* MP aware */
1364
270ac911 1365 /*
3919ced0
MD
1366 * NOTE: All system calls run MPSAFE now. The system call itself
1367 * is responsible for getting the MP lock.
270ac911 1368 */
a64ba182 1369 error = (*callp->sy_call)(&args);
984263bc 1370
4a22e893 1371out:
984263bc
MD
1372 /*
1373 * MP SAFE (we may or may not have the MP lock at this point)
1374 */
1375 switch (error) {
1376 case 0:
1377 /*
1378 * Reinitialize proc pointer `p' as it may be different
1379 * if this is a child returning from fork syscall.
1380 */
1381 p = curproc;
81eea9f4 1382 lp = curthread->td_lwp;
c7eb0589
SS
1383 frame->tf_eax = args.sysmsg_fds[0];
1384 frame->tf_edx = args.sysmsg_fds[1];
1385 frame->tf_eflags &= ~PSL_C;
984263bc 1386 break;
984263bc
MD
1387 case ERESTART:
1388 /*
1389 * Reconstruct pc, assuming lcall $X,y is 7 bytes,
1390 * int 0x80 is 2 bytes. We saved this in tf_err.
1391 */
c7eb0589 1392 frame->tf_eip -= frame->tf_err;
984263bc 1393 break;
984263bc
MD
1394 case EJUSTRETURN:
1395 break;
245e4f17
MD
1396 case EASYNC:
1397 panic("Unexpected EASYNC return value (for now)");
984263bc
MD
1398 default:
1399bad:
7062f5b4
EN
1400 if (p->p_sysent->sv_errsize) {
1401 if (error >= p->p_sysent->sv_errsize)
1402 error = -1; /* XXX */
1403 else
1404 error = p->p_sysent->sv_errtbl[error];
984263bc 1405 }
c7eb0589
SS
1406 frame->tf_eax = error;
1407 frame->tf_eflags |= PSL_C;
984263bc
MD
1408 break;
1409 }
1410
1411 /*
1412 * Traced syscall. trapsignal() is not MP aware.
1413 */
1414 if ((orig_tf_eflags & PSL_T) && !(orig_tf_eflags & PSL_VM)) {
270ac911 1415 MAKEMPSAFE(have_mplock);
c7eb0589 1416 frame->tf_eflags &= ~PSL_T;
f2642d5a 1417 trapsignal(lp, SIGTRAP, TRAP_TRACE);
984263bc
MD
1418 }
1419
1420 /*
1421 * Handle reschedule and other end-of-syscall issues
1422 */
c7eb0589 1423 userret(lp, frame, sticks);
984263bc
MD
1424
1425#ifdef KTRACE
dadab5e9 1426 if (KTRPOINT(td, KTR_SYSRET)) {
270ac911 1427 MAKEMPSAFE(have_mplock);
9fb04d14 1428 ktrsysret(lp, code, error, args.sysmsg_result);
984263bc
MD
1429 }
1430#endif
1431
1432 /*
1433 * This works because errno is findable through the
1434 * register set. If we ever support an emulation where this
1435 * is not the case, this code will need to be revisited.
1436 */
1437 STOPEVENT(p, S_SCX, code);
1438
81eea9f4 1439 userexit(lp);
8a8d5d85 1440#ifdef SMP
984263bc
MD
1441 /*
1442 * Release the MP lock if we had to get it
1443 */
270ac911 1444 KASSERT(td->td_mpcount == have_mplock,
e7912058
MD
1445 ("badmpcount syscall2/end from %p callp %p",
1446 (void *)frame->tf_eip, callp));
270ac911
MD
1447 if (have_mplock)
1448 rel_mplock();
8a8d5d85 1449#endif
80bbc910 1450 KTR_LOG(kernentry_syscall_ret, p->p_pid, lp->lwp_tid, error);
7527bc81
MD
1451#ifdef INVARIANTS
1452 KASSERT(crit_count == (td->td_pri & ~TDPRI_MASK),
1453 ("syscall: critical section count mismatch! %d/%d",
1454 crit_count / TDPRI_CRIT, td->td_pri / TDPRI_CRIT));
1455#endif
984263bc
MD
1456}
1457
91bd9c1e
SS
1458void
1459fork_return(struct lwp *lp, struct trapframe *frame)
1460{
1461 frame->tf_eax = 0; /* Child returns zero */
1462 frame->tf_eflags &= ~PSL_C; /* success */
1463 frame->tf_edx = 1;
1464
1465 generic_lwp_return(lp, frame);
80bbc910 1466 KTR_LOG(kernentry_fork_ret, lp->lwp_proc->p_pid, lp->lwp_tid);
91bd9c1e
SS
1467}
1468
984263bc
MD
1469/*
1470 * Simplified back end of syscall(), used when returning from fork()
8a8d5d85
MD
1471 * directly into user mode. MP lock is held on entry and should be
1472 * released on return. This code will return back into the fork
1473 * trampoline code which then runs doreti.
984263bc
MD
1474 */
1475void
91bd9c1e 1476generic_lwp_return(struct lwp *lp, struct trapframe *frame)
984263bc 1477{
bb3cd951 1478 struct proc *p = lp->lwp_proc;
81eea9f4 1479
3a57cd9c
MD
1480 /*
1481 * Newly forked processes are given a kernel priority. We have to
1482 * adjust the priority to a normal user priority and fake entry
1483 * into the kernel (call userenter()) to install a passive release
1484 * function just in case userret() decides to stop the process. This
1485 * can occur when ^Z races a fork. If we do not install the passive
1486 * release function the current process designation will not be
1487 * released when the thread goes to sleep.
1488 */
1489 lwkt_setpri_self(TDPRI_USER_NORM);
d86a23e0 1490 userenter(lp->lwp_thread, p);
c7eb0589 1491 userret(lp, frame, 0);
984263bc 1492#ifdef KTRACE
81eea9f4 1493 if (KTRPOINT(lp->lwp_thread, KTR_SYSRET))
9fb04d14 1494 ktrsysret(lp, SYS_fork, 0, 0);
984263bc 1495#endif
d9eea1a5 1496 p->p_flag |= P_PASSIVE_ACQ;
81eea9f4 1497 userexit(lp);
d9eea1a5 1498 p->p_flag &= ~P_PASSIVE_ACQ;
8a8d5d85 1499#ifdef SMP
81eea9f4 1500 KKASSERT(lp->lwp_thread->td_mpcount == 1);
8a8d5d85
MD
1501 rel_mplock();
1502#endif
984263bc 1503}
431d0fef
MD
1504
1505/*
4b486183 1506 * If PGEX_FPFAULT is set then set FP_VIRTFP in the PCB to force a T_DNA
431d0fef
MD
1507 * fault (which is then passed back to the virtual kernel) if an attempt is
1508 * made to use the FP unit.
1509 *
1510 * XXX this is a fairly big hack.
1511 */
1512void
1513set_vkernel_fp(struct trapframe *frame)
1514{
1515 struct thread *td = curthread;
1516
4b486183 1517 if (frame->tf_xflags & PGEX_FPFAULT) {
431d0fef
MD
1518 td->td_pcb->pcb_flags |= FP_VIRTFP;
1519 if (mdcpu->gd_npxthread == td)
1520 npxexit();
1521 } else {
1522 td->td_pcb->pcb_flags &= ~FP_VIRTFP;
1523 }
1524}
1525