Modify ktr(4) to be typesafe
[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>
5ea440eb 60#include <sys/kerneldump.h>
984263bc
MD
61#include <sys/resourcevar.h>
62#include <sys/signalvar.h>
b1b4e5a6 63#include <sys/signal2.h>
984263bc
MD
64#include <sys/syscall.h>
65#include <sys/sysctl.h>
66#include <sys/sysent.h>
67#include <sys/uio.h>
68#include <sys/vmmeter.h>
4fd10eb6 69#include <sys/malloc.h>
984263bc
MD
70#ifdef KTRACE
71#include <sys/ktrace.h>
72#endif
80bbc910 73#include <sys/ktr.h>
a722be49 74#include <sys/upcall.h>
4a22e893 75#include <sys/vkernel.h>
a64ba182
MD
76#include <sys/sysproto.h>
77#include <sys/sysunion.h>
984263bc
MD
78
79#include <vm/vm.h>
80#include <vm/vm_param.h>
81#include <sys/lock.h>
82#include <vm/pmap.h>
83#include <vm/vm_kern.h>
84#include <vm/vm_map.h>
85#include <vm/vm_page.h>
86#include <vm/vm_extern.h>
87
88#include <machine/cpu.h>
984263bc
MD
89#include <machine/md_var.h>
90#include <machine/pcb.h>
984263bc 91#include <machine/smp.h>
984263bc 92#include <machine/tss.h>
431d0fef 93#include <machine/specialreg.h>
85100692 94#include <machine/globaldata.h>
87cf6827 95#include <machine/intr_machdep.h>
984263bc 96
87cf6827 97#include <machine_base/isa/isa_intr.h>
cb7d6921 98#include <machine_base/apic/lapic.h>
984263bc
MD
99
100#ifdef POWERFAIL_NMI
101#include <sys/syslog.h>
102#include <machine/clock.h>
103#endif
104
105#include <machine/vm86.h>
106
107#include <ddb/ddb.h>
684a93c4 108
245e4f17 109#include <sys/msgport2.h>
41c20dac 110#include <sys/thread2.h>
684a93c4 111#include <sys/mplock2.h>
984263bc 112
9753ab6c
MD
113#ifdef SMP
114
115#define MAKEMPSAFE(have_mplock) \
116 if (have_mplock == 0) { \
117 get_mplock(); \
118 have_mplock = 1; \
119 }
120
121#else
122
123#define MAKEMPSAFE(have_mplock)
124
125#endif
126
3ae0cd58 127int (*pmath_emulate) (struct trapframe *);
984263bc 128
c7eb0589 129extern void trap (struct trapframe *frame);
c7eb0589 130extern void syscall2 (struct trapframe *frame);
984263bc 131
3ae0cd58
RG
132static int trap_pfault (struct trapframe *, int, vm_offset_t);
133static void trap_fatal (struct trapframe *, vm_offset_t);
134void dblfault_handler (void);
984263bc
MD
135
136extern inthand_t IDTVEC(syscall);
137
138#define MAX_TRAP_MSG 28
139static char *trap_msg[] = {
140 "", /* 0 unused */
141 "privileged instruction fault", /* 1 T_PRIVINFLT */
142 "", /* 2 unused */
143 "breakpoint instruction fault", /* 3 T_BPTFLT */
144 "", /* 4 unused */
145 "", /* 5 unused */
146 "arithmetic trap", /* 6 T_ARITHTRAP */
147 "system forced exception", /* 7 T_ASTFLT */
148 "", /* 8 unused */
149 "general protection fault", /* 9 T_PROTFLT */
150 "trace trap", /* 10 T_TRCTRAP */
151 "", /* 11 unused */
152 "page fault", /* 12 T_PAGEFLT */
153 "", /* 13 unused */
154 "alignment fault", /* 14 T_ALIGNFLT */
155 "", /* 15 unused */
156 "", /* 16 unused */
157 "", /* 17 unused */
158 "integer divide fault", /* 18 T_DIVIDE */
159 "non-maskable interrupt trap", /* 19 T_NMI */
160 "overflow trap", /* 20 T_OFLOW */
161 "FPU bounds check fault", /* 21 T_BOUND */
162 "FPU device not available", /* 22 T_DNA */
163 "double fault", /* 23 T_DOUBLEFLT */
164 "FPU operand fetch fault", /* 24 T_FPOPFLT */
165 "invalid TSS fault", /* 25 T_TSSFLT */
166 "segment not present fault", /* 26 T_SEGNPFLT */
167 "stack fault", /* 27 T_STKFLT */
168 "machine check trap", /* 28 T_MCHK */
169};
170
984263bc
MD
171#if defined(I586_CPU) && !defined(NO_F00F_HACK)
172extern int has_f00f_bug;
173#endif
174
175#ifdef DDB
176static int ddb_on_nmi = 1;
177SYSCTL_INT(_machdep, OID_AUTO, ddb_on_nmi, CTLFLAG_RW,
178 &ddb_on_nmi, 0, "Go to DDB on NMI");
179#endif
180static int panic_on_nmi = 1;
181SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RW,
182 &panic_on_nmi, 0, "Panic on NMI");
d9eea1a5
MD
183static int fast_release;
184SYSCTL_INT(_machdep, OID_AUTO, fast_release, CTLFLAG_RW,
185 &fast_release, 0, "Passive Release was optimal");
186static int slow_release;
187SYSCTL_INT(_machdep, OID_AUTO, slow_release, CTLFLAG_RW,
188 &slow_release, 0, "Passive Release was nonoptimal");
984263bc 189
4fd10eb6 190MALLOC_DEFINE(M_SYSMSG, "sysmsg", "sysmsg structure");
7062f5b4 191extern int max_sysmsg;
4fd10eb6 192
26a0694b 193/*
7bd858e5
MD
194 * Passively intercepts the thread switch function to increase the thread
195 * priority from a user priority to a kernel priority, reducing
d81ccc3e 196 * syscall and trap overhead for the case where no switch occurs.
7bd858e5 197 *
d86a23e0 198 * Synchronizes td_ucred with p_ucred. This is used by system calls,
7bd858e5
MD
199 * signal handling, faults, AST traps, and anything else that enters the
200 * kernel from userland and provides the kernel with a stable read-only
201 * copy of the process ucred.
26a0694b
MD
202 */
203static __inline void
d86a23e0 204userenter(struct thread *curtd, struct proc *curp)
26a0694b 205{
7bd858e5
MD
206 struct ucred *ocred;
207 struct ucred *ncred;
208
3824f392 209 curtd->td_release = lwkt_passive_release;
7bd858e5 210
d86a23e0
MD
211 if (curtd->td_ucred != curp->p_ucred) {
212 ncred = crhold(curp->p_ucred);
213 ocred = curtd->td_ucred;
214 curtd->td_ucred = ncred;
7bd858e5
MD
215 if (ocred)
216 crfree(ocred);
217 }
218
26a0694b
MD
219}
220
0a3f9b47 221/*
8ec60c3f
MD
222 * Handle signals, upcalls, profiling, and other AST's and/or tasks that
223 * must be completed before we can return to or try to return to userland.
0a3f9b47 224 *
8ec60c3f
MD
225 * Note that td_sticks is a 64 bit quantity, but there's no point doing 64
226 * arithmatic on the delta calculation so the absolute tick values are
227 * truncated to an integer.
37af14fe 228 */
a2a5ad0d 229static void
81eea9f4 230userret(struct lwp *lp, struct trapframe *frame, int sticks)
a2a5ad0d 231{
81eea9f4 232 struct proc *p = lp->lwp_proc;
8ba5f7ef 233 void (*hook)(void);
a2a5ad0d
MD
234 int sig;
235
8ba5f7ef
AH
236 if (p->p_userret != NULL) {
237 hook = p->p_userret;
238 p->p_userret = NULL;
239 (*hook)();
240 }
241
a2a5ad0d 242 /*
344ad853
MD
243 * Charge system time if profiling. Note: times are in microseconds.
244 * This may do a copyout and block, so do it first even though it
245 * means some system time will be charged as user time.
246 */
4643740a 247 if (p->p_flags & P_PROFIL) {
344ad853 248 addupc_task(p, frame->tf_eip,
bb3cd951 249 (u_int)((int)lp->lwp_thread->td_sticks - sticks));
344ad853
MD
250 }
251
252recheck:
253 /*
70d3d461
SS
254 * If the jungle wants us dead, so be it.
255 */
4643740a 256 if (lp->lwp_mpflags & LWP_MP_WEXIT) {
b5c4d81f 257 lwkt_gettoken(&p->p_token);
e3161323 258 lwp_exit(0);
b5c4d81f 259 lwkt_reltoken(&p->p_token); /* NOT REACHED */
e207d1fd 260 }
70d3d461
SS
261
262 /*
344ad853
MD
263 * Block here if we are in a stopped state.
264 */
5ea440eb 265 if (p->p_stat == SSTOP || dump_stop_usertds) {
4643740a 266 lwkt_gettoken(&p->p_token);
9a379a4a 267 tstop();
4643740a 268 lwkt_reltoken(&p->p_token);
344ad853
MD
269 goto recheck;
270 }
271
272 /*
4e7c41c5
MD
273 * Post any pending upcalls. If running a virtual kernel be sure
274 * to restore the virtual kernel's vmspace before posting the upcall.
a722be49 275 */
4643740a 276 if (p->p_flags & (P_SIGVTALRM | P_SIGPROF | P_UPCALLPEND)) {
fc509460 277 lwkt_gettoken(&p->p_token);
4643740a
MD
278 if (p->p_flags & P_SIGVTALRM) {
279 p->p_flags &= ~P_SIGVTALRM;
898e34b3
MD
280 ksignal(p, SIGVTALRM);
281 }
4643740a
MD
282 if (p->p_flags & P_SIGPROF) {
283 p->p_flags &= ~P_SIGPROF;
898e34b3
MD
284 ksignal(p, SIGPROF);
285 }
4643740a
MD
286 if (p->p_flags & P_UPCALLPEND) {
287 p->p_flags &= ~P_UPCALLPEND;
898e34b3
MD
288 postupcall(lp);
289 }
fc509460 290 lwkt_reltoken(&p->p_token);
344ad853 291 goto recheck;
a722be49
MD
292 }
293
294 /*
4e7c41c5
MD
295 * Post any pending signals. If running a virtual kernel be sure
296 * to restore the virtual kernel's vmspace before posting the signal.
2883d2d8
MD
297 *
298 * WARNING! postsig() can exit and not return.
a2a5ad0d 299 */
e473f776 300 if ((sig = CURSIG_TRACE(lp)) != 0) {
fc509460 301 lwkt_gettoken(&p->p_token);
a2a5ad0d 302 postsig(sig);
fc509460 303 lwkt_reltoken(&p->p_token);
344ad853 304 goto recheck;
a2a5ad0d 305 }
26a0694b
MD
306
307 /*
344ad853
MD
308 * block here if we are swapped out, but still process signals
309 * (such as SIGKILL). proc0 (the swapin scheduler) is already
310 * aware of our situation, we do not have to wake it up.
984263bc 311 */
4643740a 312 if (p->p_flags & P_SWAPPEDOUT) {
616516c8 313 lwkt_gettoken(&p->p_token);
144ce500 314 get_mplock();
4643740a 315 p->p_flags |= P_SWAPWAIT;
344ad853 316 swapin_request();
4643740a 317 if (p->p_flags & P_SWAPWAIT)
344ad853 318 tsleep(p, PCATCH, "SWOUT", 0);
4643740a 319 p->p_flags &= ~P_SWAPWAIT;
144ce500 320 rel_mplock();
616516c8 321 lwkt_reltoken(&p->p_token);
344ad853 322 goto recheck;
984263bc 323 }
ae7cb1b5
NT
324
325 /*
326 * Make sure postsig() handled request to restore old signal mask after
327 * running signal handler.
328 */
4643740a 329 KKASSERT((lp->lwp_flags & LWP_OLDMASK) == 0);
984263bc
MD
330}
331
8ec60c3f
MD
332/*
333 * Cleanup from userenter and any passive release that might have occured.
334 * We must reclaim the current-process designation before we can return
335 * to usermode. We also handle both LWKT and USER reschedule requests.
336 */
337static __inline void
81eea9f4 338userexit(struct lwp *lp)
8ec60c3f 339{
81eea9f4 340 struct thread *td = lp->lwp_thread;
8397969c 341 /* globaldata_t gd = td->td_gd; */
8ec60c3f 342
8ec60c3f 343 /*
b9eb1c19
MD
344 * Handle stop requests at kernel priority. Any requests queued
345 * after this loop will generate another AST.
8ec60c3f 346 */
b9eb1c19 347 while (lp->lwp_proc->p_stat == SSTOP) {
4643740a 348 lwkt_gettoken(&lp->lwp_proc->p_token);
b9eb1c19 349 tstop();
4643740a 350 lwkt_reltoken(&lp->lwp_proc->p_token);
fede0c7f
NT
351 }
352
8ec60c3f 353 /*
8ec60c3f
MD
354 * Reduce our priority in preparation for a return to userland. If
355 * our passive release function was still in place, our priority was
356 * never raised and does not need to be reduced.
357 */
3824f392 358 lwkt_passive_recover(td);
b9eb1c19
MD
359
360 /*
361 * Become the current user scheduled process if we aren't already,
362 * and deal with reschedule requests and other factors.
363 */
364 lp->lwp_proc->p_usched->acquire_curproc(lp);
365 /* WARNING: we may have migrated cpu's */
366 /* gd = td->td_gd; */
8ec60c3f
MD
367}
368
80bbc910
SS
369#if !defined(KTR_KERNENTRY)
370#define KTR_KERNENTRY KTR_ALL
371#endif
372KTR_INFO_MASTER(kernentry);
5bf48697
AE
373KTR_INFO(KTR_KERNENTRY, kernentry, trap, 0,
374 "TRAP(pid %d, tid %d, trapno %d, eva %lu)",
375 pid_t pid, lwpid_t tid, register_t trapno, vm_offset_t eva);
376KTR_INFO(KTR_KERNENTRY, kernentry, trap_ret, 0, "TRAP_RET(pid %d, tid %d)",
377 pid_t pid, lwpid_t tid);
378KTR_INFO(KTR_KERNENTRY, kernentry, syscall, 0, "SYSC(pid %d, tid %d, nr %d)",
379 pid_t pid, lwpid_t tid, register_t trapno);
380KTR_INFO(KTR_KERNENTRY, kernentry, syscall_ret, 0, "SYSRET(pid %d, tid %d, err %d)",
381 pid_t pid, lwpid_t tid, int err);
382KTR_INFO(KTR_KERNENTRY, kernentry, fork_ret, 0, "FORKRET(pid %d, tid %d)",
383 pid_t pid, lwpid_t tid);
80bbc910 384
984263bc 385/*
7517cb03 386 * Exception, fault, and trap interface to the kernel.
984263bc
MD
387 * This common code is called from assembly language IDT gate entry
388 * routines that prepare a suitable stack frame, and restore this
389 * frame after the exception has been processed.
a2a5ad0d
MD
390 *
391 * This function is also called from doreti in an interlock to handle ASTs.
392 * For example: hardwareint->INTROUTINE->(set ast)->doreti->trap
393 *
394 * NOTE! We have to retrieve the fault address prior to obtaining the
395 * MP lock because get_mplock() may switch out. YYY cr2 really ought
396 * to be retrieved by the assembly code, not here.
27e88a6e
MD
397 *
398 * XXX gd_trap_nesting_level currently prevents lwkt_switch() from panicing
399 * if an attempt is made to switch from a fast interrupt or IPI. This is
400 * necessary to properly take fatal kernel traps on SMP machines if
401 * get_mplock() has to block.
984263bc 402 */
9753ab6c 403
984263bc 404void
c7eb0589 405trap(struct trapframe *frame)
984263bc 406{
27e88a6e
MD
407 struct globaldata *gd = mycpu;
408 struct thread *td = gd->gd_curthread;
81eea9f4 409 struct lwp *lp = td->td_lwp;
7966cb69 410 struct proc *p;
37af14fe 411 int sticks = 0;
984263bc 412 int i = 0, ucode = 0, type, code;
9753ab6c
MD
413#ifdef SMP
414 int have_mplock = 0;
415#endif
7527bc81 416#ifdef INVARIANTS
f9235b6d 417 int crit_count = td->td_critcount;
3933a3ab 418 lwkt_tokref_t curstop = td->td_toks_stop;
7527bc81 419#endif
984263bc
MD
420 vm_offset_t eva;
421
7966cb69 422 p = td->td_proc;
984263bc 423#ifdef DDB
bfb6d329
AH
424 /*
425 * We need to allow T_DNA faults when the debugger is active since
2921c824 426 * some dumping paths do large bcopy() which use the floating
bfb6d329
AH
427 * point registers for faster copying.
428 */
5ea440eb 429 if (db_active && frame->tf_trapno != T_DNA) {
c7eb0589 430 eva = (frame->tf_trapno == T_PAGEFLT ? rcr2() : 0);
27e88a6e 431 ++gd->gd_trap_nesting_level;
9753ab6c 432 MAKEMPSAFE(have_mplock);
c7eb0589 433 trap_fatal(frame, eva);
27e88a6e 434 --gd->gd_trap_nesting_level;
8a8d5d85 435 goto out2;
984263bc
MD
436 }
437#endif
438
a2a5ad0d 439 eva = 0;
27e88a6e 440 ++gd->gd_trap_nesting_level;
c7eb0589 441 if (frame->tf_trapno == T_PAGEFLT) {
a2a5ad0d
MD
442 /*
443 * For some Cyrix CPUs, %cr2 is clobbered by interrupts.
444 * This problem is worked around by using an interrupt
445 * gate for the pagefault handler. We are finally ready
446 * to read %cr2 and then must reenable interrupts.
447 *
448 * XXX this should be in the switch statement, but the
449 * NO_FOOF_HACK and VM86 goto and ifdefs obfuscate the
450 * flow of control too much for this to be obviously
451 * correct.
452 */
453 eva = rcr2();
a2a5ad0d 454 cpu_enable_intr();
a2a5ad0d 455 }
80bbc910 456
27e88a6e 457 --gd->gd_trap_nesting_level;
a2a5ad0d 458
c7eb0589 459 if (!(frame->tf_eflags & PSL_I)) {
984263bc
MD
460 /*
461 * Buggy application or kernel code has disabled interrupts
462 * and then trapped. Enabling interrupts now is wrong, but
463 * it is better than running with interrupts disabled until
464 * they are accidentally enabled later.
465 */
c7eb0589
SS
466 type = frame->tf_trapno;
467 if (ISPL(frame->tf_cs)==SEL_UPL || (frame->tf_eflags & PSL_VM)) {
9753ab6c 468 MAKEMPSAFE(have_mplock);
26be20a0 469 kprintf(
984263bc
MD
470 "pid %ld (%s): trap %d with interrupts disabled\n",
471 (long)curproc->p_pid, curproc->p_comm, type);
a2a5ad0d 472 } else if (type != T_BPTFLT && type != T_TRCTRAP) {
984263bc
MD
473 /*
474 * XXX not quite right, since this may be for a
475 * multiple fault in user mode.
476 */
9753ab6c 477 MAKEMPSAFE(have_mplock);
26be20a0 478 kprintf("kernel trap %d with interrupts disabled\n",
984263bc 479 type);
a2a5ad0d 480 }
8a8d5d85 481 cpu_enable_intr();
984263bc
MD
482 }
483
984263bc
MD
484#if defined(I586_CPU) && !defined(NO_F00F_HACK)
485restart:
486#endif
c7eb0589
SS
487 type = frame->tf_trapno;
488 code = frame->tf_err;
984263bc
MD
489
490 if (in_vm86call) {
c7eb0589 491 if (frame->tf_eflags & PSL_VM &&
984263bc 492 (type == T_PROTFLT || type == T_STKFLT)) {
96728c05 493#ifdef SMP
b5d16701 494 KKASSERT(get_mplock_count(curthread) > 0);
96728c05 495#endif
c7eb0589 496 i = vm86_emulate((struct vm86frame *)frame);
96728c05 497#ifdef SMP
b5d16701 498 KKASSERT(get_mplock_count(curthread) > 0);
96728c05 499#endif
8a8d5d85 500 if (i != 0) {
984263bc
MD
501 /*
502 * returns to original process
503 */
d3d32139 504#ifdef SMP
c7eb0589 505 vm86_trap((struct vm86frame *)frame,
d3d32139
MD
506 have_mplock);
507#else
c7eb0589 508 vm86_trap((struct vm86frame *)frame, 0);
d3d32139
MD
509#endif
510 KKASSERT(0); /* NOT REACHED */
8a8d5d85
MD
511 }
512 goto out2;
984263bc
MD
513 }
514 switch (type) {
515 /*
516 * these traps want either a process context, or
517 * assume a normal userspace trap.
518 */
519 case T_PROTFLT:
520 case T_SEGNPFLT:
c7eb0589 521 trap_fatal(frame, eva);
8a8d5d85 522 goto out2;
984263bc
MD
523 case T_TRCTRAP:
524 type = T_BPTFLT; /* kernel breakpoint */
525 /* FALL THROUGH */
526 }
527 goto kernel_trap; /* normal kernel trap handling */
528 }
529
c7eb0589 530 if ((ISPL(frame->tf_cs) == SEL_UPL) || (frame->tf_eflags & PSL_VM)) {
984263bc
MD
531 /* user trap */
532
80bbc910
SS
533 KTR_LOG(kernentry_trap, p->p_pid, lp->lwp_tid,
534 frame->tf_trapno, eva);
535
d86a23e0 536 userenter(td, p);
26a0694b 537
37af14fe 538 sticks = (int)td->td_sticks;
c7eb0589 539 lp->lwp_md.md_regs = frame;
984263bc
MD
540
541 switch (type) {
542 case T_PRIVINFLT: /* privileged instruction fault */
f2642d5a 543 ucode = ILL_COPROC;
984263bc
MD
544 i = SIGILL;
545 break;
546
547 case T_BPTFLT: /* bpt instruction fault */
548 case T_TRCTRAP: /* trace trap */
c7eb0589 549 frame->tf_eflags &= ~PSL_T;
f2642d5a 550 ucode = TRAP_TRACE;
984263bc
MD
551 i = SIGTRAP;
552 break;
553
554 case T_ARITHTRAP: /* arithmetic trap */
555 ucode = code;
556 i = SIGFPE;
557 break;
558
559 case T_ASTFLT: /* Allow process switch */
12e4aaff 560 mycpu->gd_cnt.v_soft++;
235957ed 561 if (mycpu->gd_reqflags & RQF_AST_OWEUPC) {
2a418930
MD
562 atomic_clear_int(&mycpu->gd_reqflags,
563 RQF_AST_OWEUPC);
93328593
SS
564 addupc_task(p, p->p_prof.pr_addr,
565 p->p_prof.pr_ticks);
984263bc
MD
566 }
567 goto out;
568
569 /*
570 * The following two traps can happen in
571 * vm86 mode, and, if so, we want to handle
572 * them specially.
573 */
574 case T_PROTFLT: /* general protection fault */
575 case T_STKFLT: /* stack fault */
c7eb0589
SS
576 if (frame->tf_eflags & PSL_VM) {
577 i = vm86_emulate((struct vm86frame *)frame);
984263bc
MD
578 if (i == 0)
579 goto out;
580 break;
581 }
f2642d5a
AH
582 i = SIGBUS;
583 ucode = (type == T_PROTFLT) ? BUS_OBJERR : BUS_ADRERR;
584 break;
984263bc 585 case T_SEGNPFLT: /* segment not present fault */
f2642d5a
AH
586 i = SIGBUS;
587 ucode = BUS_ADRERR;
588 break;
984263bc
MD
589 case T_TSSFLT: /* invalid TSS fault */
590 case T_DOUBLEFLT: /* double fault */
f2642d5a
AH
591 i = SIGBUS;
592 ucode = BUS_OBJERR;
984263bc 593 default:
f2642d5a
AH
594#if 0
595 ucode = code + BUS_SEGM_FAULT ; /* XXX: ???*/
596#endif
597 ucode = BUS_OBJERR;
984263bc
MD
598 i = SIGBUS;
599 break;
600
601 case T_PAGEFLT: /* page fault */
c7eb0589 602 i = trap_pfault(frame, TRUE, eva);
984263bc 603 if (i == -1)
d81ccc3e 604 goto out;
984263bc
MD
605#if defined(I586_CPU) && !defined(NO_F00F_HACK)
606 if (i == -2)
607 goto restart;
608#endif
609 if (i == 0)
610 goto out;
f2642d5a 611#if 0
984263bc 612 ucode = T_PAGEFLT;
f2642d5a
AH
613#endif
614 if (i == SIGSEGV)
615 ucode = SEGV_MAPERR;
616 else
617 ucode = BUS_ADRERR; /* XXX */
984263bc
MD
618 break;
619
620 case T_DIVIDE: /* integer divide fault */
621 ucode = FPE_INTDIV;
622 i = SIGFPE;
623 break;
624
625#if NISA > 0
626 case T_NMI:
9753ab6c 627 MAKEMPSAFE(have_mplock);
984263bc
MD
628#ifdef POWERFAIL_NMI
629 goto handle_powerfail;
630#else /* !POWERFAIL_NMI */
631 /* machine/parity/power fail/"kitchen sink" faults */
632 if (isa_nmi(code) == 0) {
633#ifdef DDB
634 /*
635 * NMI can be hooked up to a pushbutton
636 * for debugging.
637 */
638 if (ddb_on_nmi) {
26be20a0 639 kprintf ("NMI ... going to debugger\n");
c7eb0589 640 kdb_trap (type, 0, frame);
984263bc
MD
641 }
642#endif /* DDB */
8a8d5d85 643 goto out2;
984263bc
MD
644 } else if (panic_on_nmi)
645 panic("NMI indicates hardware failure");
646 break;
647#endif /* POWERFAIL_NMI */
648#endif /* NISA > 0 */
649
650 case T_OFLOW: /* integer overflow fault */
651 ucode = FPE_INTOVF;
652 i = SIGFPE;
653 break;
654
655 case T_BOUND: /* bounds check fault */
656 ucode = FPE_FLTSUB;
657 i = SIGFPE;
658 break;
659
660 case T_DNA:
431d0fef
MD
661 /*
662 * Virtual kernel intercept - pass the DNA exception
663 * to the virtual kernel if it asked to handle it.
664 * This occurs when the virtual kernel is holding
665 * onto the FP context for a different emulated
666 * process then the one currently running.
667 *
668 * We must still call npxdna() since we may have
669 * saved FP state that the virtual kernel needs
670 * to hand over to a different emulated process.
671 */
39005e16 672 if (lp->lwp_vkernel && lp->lwp_vkernel->ve &&
431d0fef
MD
673 (td->td_pcb->pcb_flags & FP_VIRTFP)
674 ) {
675 npxdna();
676 break;
677 }
678
984263bc 679#if NNPX > 0
a02705a9
MD
680 /*
681 * The kernel may have switched out the FP unit's
682 * state, causing the user process to take a fault
683 * when it tries to use the FP unit. Restore the
684 * state here
685 */
984263bc 686 if (npxdna())
d81ccc3e 687 goto out;
984263bc
MD
688#endif
689 if (!pmath_emulate) {
690 i = SIGFPE;
691 ucode = FPE_FPU_NP_TRAP;
692 break;
693 }
c7eb0589 694 i = (*pmath_emulate)(frame);
984263bc 695 if (i == 0) {
c7eb0589 696 if (!(frame->tf_eflags & PSL_T))
8a8d5d85 697 goto out2;
c7eb0589 698 frame->tf_eflags &= ~PSL_T;
984263bc
MD
699 i = SIGTRAP;
700 }
701 /* else ucode = emulator_only_knows() XXX */
702 break;
703
704 case T_FPOPFLT: /* FPU operand fetch fault */
f2642d5a 705 ucode = ILL_ILLOPN;
984263bc
MD
706 i = SIGILL;
707 break;
708
709 case T_XMMFLT: /* SIMD floating-point exception */
710 ucode = 0; /* XXX */
711 i = SIGFPE;
712 break;
713 }
714 } else {
715kernel_trap:
716 /* kernel trap */
717
718 switch (type) {
719 case T_PAGEFLT: /* page fault */
c7eb0589 720 trap_pfault(frame, FALSE, eva);
8a8d5d85 721 goto out2;
984263bc
MD
722
723 case T_DNA:
724#if NNPX > 0
725 /*
a02705a9
MD
726 * The kernel may be using npx for copying or other
727 * purposes.
984263bc
MD
728 */
729 if (npxdna())
8a8d5d85 730 goto out2;
984263bc
MD
731#endif
732 break;
733
734 case T_PROTFLT: /* general protection fault */
735 case T_SEGNPFLT: /* segment not present fault */
736 /*
737 * Invalid segment selectors and out of bounds
738 * %eip's and %esp's can be set up in user mode.
739 * This causes a fault in kernel mode when the
740 * kernel tries to return to user mode. We want
741 * to get this fault so that we can fix the
742 * problem here and not have to check all the
743 * selectors and pointers when the user changes
744 * them.
745 */
746#define MAYBE_DORETI_FAULT(where, whereto) \
747 do { \
c7eb0589
SS
748 if (frame->tf_eip == (int)where) { \
749 frame->tf_eip = (int)whereto; \
8a8d5d85 750 goto out2; \
984263bc
MD
751 } \
752 } while (0)
ef0fdad1 753 if (mycpu->gd_intr_nesting_level == 0) {
984263bc
MD
754 /*
755 * Invalid %fs's and %gs's can be created using
756 * procfs or PT_SETREGS or by invalidating the
757 * underlying LDT entry. This causes a fault
758 * in kernel mode when the kernel attempts to
759 * switch contexts. Lose the bad context
760 * (XXX) so that we can continue, and generate
761 * a signal.
762 */
984263bc
MD
763 MAYBE_DORETI_FAULT(doreti_iret,
764 doreti_iret_fault);
765 MAYBE_DORETI_FAULT(doreti_popl_ds,
766 doreti_popl_ds_fault);
767 MAYBE_DORETI_FAULT(doreti_popl_es,
768 doreti_popl_es_fault);
769 MAYBE_DORETI_FAULT(doreti_popl_fs,
770 doreti_popl_fs_fault);
4e7c41c5
MD
771 MAYBE_DORETI_FAULT(doreti_popl_gs,
772 doreti_popl_gs_fault);
37af14fe 773 if (td->td_pcb->pcb_onfault) {
c7eb0589 774 frame->tf_eip =
37af14fe 775 (register_t)td->td_pcb->pcb_onfault;
8a8d5d85 776 goto out2;
984263bc
MD
777 }
778 }
779 break;
780
781 case T_TSSFLT:
782 /*
783 * PSL_NT can be set in user mode and isn't cleared
784 * automatically when the kernel is entered. This
785 * causes a TSS fault when the kernel attempts to
786 * `iret' because the TSS link is uninitialized. We
787 * want to get this fault so that we can fix the
788 * problem here and not every time the kernel is
789 * entered.
790 */
c7eb0589
SS
791 if (frame->tf_eflags & PSL_NT) {
792 frame->tf_eflags &= ~PSL_NT;
8a8d5d85 793 goto out2;
984263bc
MD
794 }
795 break;
796
797 case T_TRCTRAP: /* trace trap */
c7eb0589 798 if (frame->tf_eip == (int)IDTVEC(syscall)) {
984263bc
MD
799 /*
800 * We've just entered system mode via the
801 * syscall lcall. Continue single stepping
802 * silently until the syscall handler has
803 * saved the flags.
804 */
8a8d5d85 805 goto out2;
984263bc 806 }
c7eb0589 807 if (frame->tf_eip == (int)IDTVEC(syscall) + 1) {
984263bc
MD
808 /*
809 * The syscall handler has now saved the
810 * flags. Stop single stepping it.
811 */
c7eb0589 812 frame->tf_eflags &= ~PSL_T;
8a8d5d85 813 goto out2;
984263bc
MD
814 }
815 /*
816 * Ignore debug register trace traps due to
817 * accesses in the user's address space, which
818 * can happen under several conditions such as
819 * if a user sets a watchpoint on a buffer and
820 * then passes that buffer to a system call.
821 * We still want to get TRCTRAPS for addresses
822 * in kernel space because that is useful when
823 * debugging the kernel.
824 */
825 if (user_dbreg_trap()) {
826 /*
827 * Reset breakpoint bits because the
828 * processor doesn't
829 */
830 load_dr6(rdr6() & 0xfffffff0);
8a8d5d85 831 goto out2;
984263bc
MD
832 }
833 /*
834 * Fall through (TRCTRAP kernel mode, kernel address)
835 */
836 case T_BPTFLT:
837 /*
838 * If DDB is enabled, let it handle the debugger trap.
839 * Otherwise, debugger traps "can't happen".
840 */
f2642d5a 841 ucode = TRAP_BRKPT;
984263bc 842#ifdef DDB
9753ab6c 843 MAKEMPSAFE(have_mplock);
c7eb0589 844 if (kdb_trap (type, 0, frame))
8a8d5d85 845 goto out2;
984263bc
MD
846#endif
847 break;
848
849#if NISA > 0
850 case T_NMI:
9753ab6c 851 MAKEMPSAFE(have_mplock);
984263bc
MD
852#ifdef POWERFAIL_NMI
853#ifndef TIMER_FREQ
854# define TIMER_FREQ 1193182
855#endif
856 handle_powerfail:
857 {
858 static unsigned lastalert = 0;
859
860 if(time_second - lastalert > 10)
861 {
862 log(LOG_WARNING, "NMI: power fail\n");
863 sysbeep(TIMER_FREQ/880, hz);
864 lastalert = time_second;
865 }
8a8d5d85
MD
866 /* YYY mp count */
867 goto out2;
984263bc
MD
868 }
869#else /* !POWERFAIL_NMI */
870 /* machine/parity/power fail/"kitchen sink" faults */
871 if (isa_nmi(code) == 0) {
872#ifdef DDB
873 /*
874 * NMI can be hooked up to a pushbutton
875 * for debugging.
876 */
877 if (ddb_on_nmi) {
26be20a0 878 kprintf ("NMI ... going to debugger\n");
c7eb0589 879 kdb_trap (type, 0, frame);
984263bc
MD
880 }
881#endif /* DDB */
8a8d5d85 882 goto out2;
984263bc 883 } else if (panic_on_nmi == 0)
8a8d5d85 884 goto out2;
984263bc
MD
885 /* FALL THROUGH */
886#endif /* POWERFAIL_NMI */
887#endif /* NISA > 0 */
888 }
889
9753ab6c 890 MAKEMPSAFE(have_mplock);
c7eb0589 891 trap_fatal(frame, eva);
8a8d5d85 892 goto out2;
984263bc
MD
893 }
894
4a22e893
MD
895 /*
896 * Virtual kernel intercept - if the fault is directly related to a
897 * VM context managed by a virtual kernel then let the virtual kernel
898 * handle it.
899 */
39005e16 900 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
287ebb09 901 vkernel_trap(lp, frame);
4a22e893
MD
902 goto out;
903 }
904
905 /*
906 * Translate fault for emulators (e.g. Linux)
907 */
984263bc
MD
908 if (*p->p_sysent->sv_transtrap)
909 i = (*p->p_sysent->sv_transtrap)(i, type);
910
9753ab6c 911 MAKEMPSAFE(have_mplock);
08f2f1bb 912 trapsignal(lp, i, ucode);
984263bc
MD
913
914#ifdef DEBUG
915 if (type <= MAX_TRAP_MSG) {
916 uprintf("fatal process exception: %s",
917 trap_msg[type]);
918 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
919 uprintf(", fault VA = 0x%lx", (u_long)eva);
920 uprintf("\n");
921 }
922#endif
923
924out:
c7eb0589 925 userret(lp, frame, sticks);
81eea9f4 926 userexit(lp);
9753ab6c 927out2: ;
96728c05 928#ifdef SMP
9753ab6c
MD
929 if (have_mplock)
930 rel_mplock();
96728c05 931#endif
80bbc910
SS
932 if (p != NULL && lp != NULL)
933 KTR_LOG(kernentry_trap_ret, p->p_pid, lp->lwp_tid);
7527bc81 934#ifdef INVARIANTS
f9235b6d 935 KASSERT(crit_count == td->td_critcount,
3933a3ab 936 ("trap: critical section count mismatch! %d/%d",
f9235b6d 937 crit_count, td->td_pri));
3933a3ab
MD
938 KASSERT(curstop == td->td_toks_stop,
939 ("trap: extra tokens held after trap! %zd/%zd",
940 curstop - &td->td_toks_base,
941 td->td_toks_stop - &td->td_toks_base));
7527bc81 942#endif
984263bc
MD
943}
944
984263bc 945int
f123d5a1 946trap_pfault(struct trapframe *frame, int usermode, vm_offset_t eva)
984263bc
MD
947{
948 vm_offset_t va;
949 struct vmspace *vm = NULL;
950 vm_map_t map = 0;
951 int rv = 0;
1b9d3514 952 int fault_flags;
984263bc 953 vm_prot_t ftype;
37af14fe 954 thread_t td = curthread;
287ebb09 955 struct lwp *lp = td->td_lwp;
984263bc
MD
956
957 va = trunc_page(eva);
958 if (va >= KERNBASE) {
959 /*
960 * Don't allow user-mode faults in kernel address space.
961 * An exception: if the faulting address is the invalid
962 * instruction entry in the IDT, then the Intel Pentium
963 * F00F bug workaround was triggered, and we need to
964 * treat it is as an illegal instruction, and not a page
965 * fault.
966 */
967#if defined(I586_CPU) && !defined(NO_F00F_HACK)
968 if ((eva == (unsigned int)&idt[6]) && has_f00f_bug) {
969 frame->tf_trapno = T_PRIVINFLT;
970 return -2;
971 }
972#endif
973 if (usermode)
974 goto nogo;
975
e4846942 976 map = &kernel_map;
984263bc
MD
977 } else {
978 /*
979 * This is a fault on non-kernel virtual memory.
980 * vm is initialized above to NULL. If curproc is NULL
981 * or curproc->p_vmspace is NULL the fault is fatal.
982 */
287ebb09
MD
983 if (lp != NULL)
984 vm = lp->lwp_vmspace;
984263bc
MD
985
986 if (vm == NULL)
987 goto nogo;
988
989 map = &vm->vm_map;
990 }
991
992 if (frame->tf_err & PGEX_W)
993 ftype = VM_PROT_WRITE;
994 else
995 ftype = VM_PROT_READ;
996
e4846942 997 if (map != &kernel_map) {
984263bc
MD
998 /*
999 * Keep swapout from messing with us during this
1000 * critical time.
1001 */
287ebb09 1002 PHOLD(lp->lwp_proc);
984263bc
MD
1003
1004 /*
1b9d3514
MD
1005 * Issue fault
1006 */
1007 fault_flags = 0;
1008 if (usermode)
1009 fault_flags |= VM_FAULT_BURST;
1010 if (ftype & VM_PROT_WRITE)
1011 fault_flags |= VM_FAULT_DIRTY;
1012 else
1013 fault_flags |= VM_FAULT_NORMAL;
1014 rv = vm_fault(map, va, ftype, fault_flags);
287ebb09 1015 PRELE(lp->lwp_proc);
984263bc
MD
1016 } else {
1017 /*
17cde63e
MD
1018 * Don't have to worry about process locking or stacks
1019 * in the kernel.
984263bc
MD
1020 */
1021 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
1022 }
1023
1024 if (rv == KERN_SUCCESS)
1025 return (0);
1026nogo:
1027 if (!usermode) {
37af14fe
MD
1028 if (td->td_gd->gd_intr_nesting_level == 0 &&
1029 td->td_pcb->pcb_onfault) {
1030 frame->tf_eip = (register_t)td->td_pcb->pcb_onfault;
984263bc
MD
1031 return (0);
1032 }
1033 trap_fatal(frame, eva);
1034 return (-1);
1035 }
1036
1037 /* kludge to pass faulting virtual address to sendsig */
4e7c41c5 1038 frame->tf_xflags = frame->tf_err;
984263bc
MD
1039 frame->tf_err = eva;
1040
1041 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
1042}
1043
1044static void
f123d5a1 1045trap_fatal(struct trapframe *frame, vm_offset_t eva)
984263bc
MD
1046{
1047 int code, type, ss, esp;
1048 struct soft_segment_descriptor softseg;
1049
1050 code = frame->tf_err;
1051 type = frame->tf_trapno;
3951a45f 1052 sdtossd(&gdt[mycpu->gd_cpuid * NGDT + IDXSEL(frame->tf_cs & 0xffff)].sd, &softseg);
984263bc
MD
1053
1054 if (type <= MAX_TRAP_MSG)
26be20a0 1055 kprintf("\n\nFatal trap %d: %s while in %s mode\n",
984263bc
MD
1056 type, trap_msg[type],
1057 frame->tf_eflags & PSL_VM ? "vm86" :
1058 ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel");
1059#ifdef SMP
7a44d1cb 1060 /* three separate prints in case of a trap on an unmapped page */
26be20a0 1061 kprintf("cpuid = %d; ", mycpu->gd_cpuid);
cb7d6921 1062 kprintf("lapic.id = %08x\n", lapic->id);
984263bc
MD
1063#endif
1064 if (type == T_PAGEFLT) {
d557216f 1065 kprintf("fault virtual address = %p\n", (void *)eva);
26be20a0 1066 kprintf("fault code = %s %s, %s\n",
984263bc
MD
1067 code & PGEX_U ? "user" : "supervisor",
1068 code & PGEX_W ? "write" : "read",
1069 code & PGEX_P ? "protection violation" : "page not present");
1070 }
26be20a0 1071 kprintf("instruction pointer = 0x%x:0x%x\n",
984263bc
MD
1072 frame->tf_cs & 0xffff, frame->tf_eip);
1073 if ((ISPL(frame->tf_cs) == SEL_UPL) || (frame->tf_eflags & PSL_VM)) {
1074 ss = frame->tf_ss & 0xffff;
1075 esp = frame->tf_esp;
1076 } else {
1077 ss = GSEL(GDATA_SEL, SEL_KPL);
1078 esp = (int)&frame->tf_esp;
1079 }
26be20a0
SW
1080 kprintf("stack pointer = 0x%x:0x%x\n", ss, esp);
1081 kprintf("frame pointer = 0x%x:0x%x\n", ss, frame->tf_ebp);
1082 kprintf("code segment = base 0x%x, limit 0x%x, type 0x%x\n",
984263bc 1083 softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type);
26be20a0 1084 kprintf(" = DPL %d, pres %d, def32 %d, gran %d\n",
984263bc
MD
1085 softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_def32,
1086 softseg.ssd_gran);
26be20a0 1087 kprintf("processor eflags = ");
984263bc 1088 if (frame->tf_eflags & PSL_T)
26be20a0 1089 kprintf("trace trap, ");
984263bc 1090 if (frame->tf_eflags & PSL_I)
26be20a0 1091 kprintf("interrupt enabled, ");
984263bc 1092 if (frame->tf_eflags & PSL_NT)
26be20a0 1093 kprintf("nested task, ");
984263bc 1094 if (frame->tf_eflags & PSL_RF)
26be20a0 1095 kprintf("resume, ");
984263bc 1096 if (frame->tf_eflags & PSL_VM)
26be20a0
SW
1097 kprintf("vm86, ");
1098 kprintf("IOPL = %d\n", (frame->tf_eflags & PSL_IOPL) >> 12);
1099 kprintf("current process = ");
984263bc 1100 if (curproc) {
26be20a0 1101 kprintf("%lu (%s)\n",
984263bc
MD
1102 (u_long)curproc->p_pid, curproc->p_comm ?
1103 curproc->p_comm : "");
1104 } else {
26be20a0 1105 kprintf("Idle\n");
984263bc 1106 }
26be20a0 1107 kprintf("current thread = pri %d ", curthread->td_pri);
f9235b6d 1108 if (curthread->td_critcount)
26be20a0
SW
1109 kprintf("(CRIT)");
1110 kprintf("\n");
984263bc
MD
1111#ifdef SMP
1112/**
1113 * XXX FIXME:
1114 * we probably SHOULD have stopped the other CPUs before now!
1115 * another CPU COULD have been touching cpl at this moment...
1116 */
26be20a0 1117 kprintf(" <- SMP: XXX");
984263bc 1118#endif
26be20a0 1119 kprintf("\n");
984263bc
MD
1120
1121#ifdef KDB
1122 if (kdb_trap(&psl))
1123 return;
1124#endif
1125#ifdef DDB
f7bc9806 1126 if ((debugger_on_panic || db_active) && kdb_trap(type, code, frame))
984263bc
MD
1127 return;
1128#endif
26be20a0 1129 kprintf("trap number = %d\n", type);
984263bc
MD
1130 if (type <= MAX_TRAP_MSG)
1131 panic("%s", trap_msg[type]);
1132 else
1133 panic("unknown/reserved trap");
1134}
1135
1136/*
1137 * Double fault handler. Called when a fault occurs while writing
1138 * a frame for a trap/exception onto the stack. This usually occurs
1139 * when the stack overflows (such is the case with infinite recursion,
1140 * for example).
1141 *
1142 * XXX Note that the current PTD gets replaced by IdlePTD when the
1143 * task switch occurs. This means that the stack that was active at
1144 * the time of the double fault is not available at <kstack> unless
1145 * the machine was idle when the double fault occurred. The downside
1146 * of this is that "trace <ebp>" in ddb won't work.
1147 */
093565f2
MD
1148static __inline
1149int
1150in_kstack_guard(register_t rptr)
1151{
1152 thread_t td = curthread;
1153
1154 if ((char *)rptr >= td->td_kstack &&
1155 (char *)rptr < td->td_kstack + PAGE_SIZE) {
1156 return 1;
1157 }
1158 return 0;
1159}
1160
984263bc 1161void
f123d5a1 1162dblfault_handler(void)
984263bc 1163{
85100692 1164 struct mdglobaldata *gd = mdcpu;
17a9f566 1165
093565f2
MD
1166 if (in_kstack_guard(gd->gd_common_tss.tss_esp) ||
1167 in_kstack_guard(gd->gd_common_tss.tss_ebp)) {
0174149e 1168 kprintf("DOUBLE FAULT - KERNEL STACK GUARD HIT!\n");
093565f2 1169 } else {
0174149e 1170 kprintf("DOUBLE FAULT:\n");
093565f2 1171 }
26be20a0
SW
1172 kprintf("eip = 0x%x\n", gd->gd_common_tss.tss_eip);
1173 kprintf("esp = 0x%x\n", gd->gd_common_tss.tss_esp);
1174 kprintf("ebp = 0x%x\n", gd->gd_common_tss.tss_ebp);
984263bc 1175#ifdef SMP
7a44d1cb 1176 /* three separate prints in case of a trap on an unmapped page */
093565f2 1177 kprintf("cpuid = %d; ", gd->mi.gd_cpuid);
cb7d6921 1178 kprintf("lapic.id = %08x\n", lapic->id);
984263bc
MD
1179#endif
1180 panic("double fault");
1181}
1182
1183/*
0d355d3b 1184 * syscall2 - MP aware system call request C handler
984263bc 1185 *
0d355d3b
MD
1186 * A system call is essentially treated as a trap. The MP lock is not
1187 * held on entry or return. We are responsible for handling ASTs
1188 * (e.g. a task switch) prior to return.
984263bc 1189 *
0d355d3b 1190 * MPSAFE
984263bc
MD
1191 */
1192void
c7eb0589 1193syscall2(struct trapframe *frame)
984263bc 1194{
dadab5e9
MD
1195 struct thread *td = curthread;
1196 struct proc *p = td->td_proc;
81eea9f4 1197 struct lwp *lp = td->td_lwp;
984263bc 1198 caddr_t params;
984263bc 1199 struct sysent *callp;
984263bc 1200 register_t orig_tf_eflags;
37af14fe 1201 int sticks;
984263bc
MD
1202 int error;
1203 int narg;
7527bc81 1204#ifdef INVARIANTS
f9235b6d 1205 int crit_count = td->td_critcount;
7527bc81 1206#endif
270ac911
MD
1207#ifdef SMP
1208 int have_mplock = 0;
1209#endif
984263bc 1210 u_int code;
a64ba182 1211 union sysunion args;
984263bc
MD
1212
1213#ifdef DIAGNOSTIC
c7eb0589 1214 if (ISPL(frame->tf_cs) != SEL_UPL) {
984263bc
MD
1215 get_mplock();
1216 panic("syscall");
1217 /* NOT REACHED */
1218 }
1219#endif
1220
80bbc910
SS
1221 KTR_LOG(kernentry_syscall, p->p_pid, lp->lwp_tid,
1222 frame->tf_eax);
1223
d86a23e0 1224 userenter(td, p); /* lazy raise our priority */
37af14fe 1225
4a22e893
MD
1226 /*
1227 * Misc
1228 */
37af14fe 1229 sticks = (int)td->td_sticks;
c7eb0589 1230 orig_tf_eflags = frame->tf_eflags;
4a22e893
MD
1231
1232 /*
1233 * Virtual kernel intercept - if a VM context managed by a virtual
1234 * kernel issues a system call the virtual kernel handles it, not us.
1235 * Restore the virtual kernel context and return from its system
1236 * call. The current frame is copied out to the virtual kernel.
1237 */
39005e16 1238 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
bb47c072 1239 vkernel_trap(lp, frame);
4a22e893 1240 error = EJUSTRETURN;
e7912058 1241 callp = NULL;
4a22e893
MD
1242 goto out;
1243 }
984263bc 1244
4a22e893
MD
1245 /*
1246 * Get the system call parameters and account for time
1247 */
c7eb0589
SS
1248 lp->lwp_md.md_regs = frame;
1249 params = (caddr_t)frame->tf_esp + sizeof(int);
1250 code = frame->tf_eax;
984263bc
MD
1251
1252 if (p->p_sysent->sv_prepsyscall) {
270ac911 1253 (*p->p_sysent->sv_prepsyscall)(
c7eb0589 1254 frame, (int *)(&args.nosys.sysmsg + 1),
270ac911 1255 &code, &params);
984263bc
MD
1256 } else {
1257 /*
1258 * Need to check if this is a 32 bit or 64 bit syscall.
1259 * fuword is MP aware.
1260 */
1261 if (code == SYS_syscall) {
1262 /*
1263 * Code is first argument, followed by actual args.
1264 */
1265 code = fuword(params);
1266 params += sizeof(int);
1267 } else if (code == SYS___syscall) {
1268 /*
1269 * Like syscall, but code is a quad, so as to maintain
1270 * quad alignment for the rest of the arguments.
1271 */
1272 code = fuword(params);
1273 params += sizeof(quad_t);
1274 }
1275 }
1276
8ec60c3f 1277 code &= p->p_sysent->sv_mask;
8ba5f7ef 1278
7062f5b4
EN
1279 if (code >= p->p_sysent->sv_size)
1280 callp = &p->p_sysent->sv_table[0];
1281 else
1282 callp = &p->p_sysent->sv_table[code];
984263bc
MD
1283
1284 narg = callp->sy_narg & SYF_ARGMASK;
1285
8ba5f7ef
AH
1286#if 0
1287 if (p->p_sysent->sv_name[0] == 'L')
1288 kprintf("Linux syscall, code = %d\n", code);
1289#endif
1290
984263bc
MD
1291 /*
1292 * copyin is MP aware, but the tracing code is not
1293 */
8ec60c3f 1294 if (narg && params) {
f9a13fc4 1295 error = copyin(params, (caddr_t)(&args.nosys.sysmsg + 1),
8ec60c3f
MD
1296 narg * sizeof(register_t));
1297 if (error) {
984263bc 1298#ifdef KTRACE
270ac911
MD
1299 if (KTRPOINT(td, KTR_SYSCALL)) {
1300 MAKEMPSAFE(have_mplock);
1301
9fb04d14 1302 ktrsyscall(lp, code, narg,
f9a13fc4 1303 (void *)(&args.nosys.sysmsg + 1));
270ac911 1304 }
984263bc 1305#endif
8ec60c3f
MD
1306 goto bad;
1307 }
984263bc
MD
1308 }
1309
984263bc 1310#ifdef KTRACE
dadab5e9 1311 if (KTRPOINT(td, KTR_SYSCALL)) {
270ac911 1312 MAKEMPSAFE(have_mplock);
9fb04d14 1313 ktrsyscall(lp, code, narg, (void *)(&args.nosys.sysmsg + 1));
984263bc
MD
1314 }
1315#endif
01dce7bb
MD
1316
1317 /*
1318 * For traditional syscall code edx is left untouched when 32 bit
1319 * results are returned. Since edx is loaded from fds[1] when the
1320 * system call returns we pre-set it here.
1321 */
c7114eea 1322 args.sysmsg_fds[0] = 0;
c7eb0589 1323 args.sysmsg_fds[1] = frame->tf_edx;
984263bc 1324
4a22e893
MD
1325 /*
1326 * The syscall might manipulate the trap frame. If it does it
1327 * will probably return EJUSTRETURN.
1328 */
c7eb0589 1329 args.sysmsg_frame = frame;
4a22e893 1330
984263bc
MD
1331 STOPEVENT(p, S_SCE, narg); /* MP aware */
1332
270ac911 1333 /*
3919ced0
MD
1334 * NOTE: All system calls run MPSAFE now. The system call itself
1335 * is responsible for getting the MP lock.
270ac911 1336 */
a64ba182 1337 error = (*callp->sy_call)(&args);
984263bc 1338
4a22e893 1339out:
984263bc
MD
1340 /*
1341 * MP SAFE (we may or may not have the MP lock at this point)
1342 */
1343 switch (error) {
1344 case 0:
1345 /*
1346 * Reinitialize proc pointer `p' as it may be different
1347 * if this is a child returning from fork syscall.
1348 */
1349 p = curproc;
81eea9f4 1350 lp = curthread->td_lwp;
c7eb0589
SS
1351 frame->tf_eax = args.sysmsg_fds[0];
1352 frame->tf_edx = args.sysmsg_fds[1];
1353 frame->tf_eflags &= ~PSL_C;
984263bc 1354 break;
984263bc
MD
1355 case ERESTART:
1356 /*
1357 * Reconstruct pc, assuming lcall $X,y is 7 bytes,
1358 * int 0x80 is 2 bytes. We saved this in tf_err.
1359 */
c7eb0589 1360 frame->tf_eip -= frame->tf_err;
984263bc 1361 break;
984263bc
MD
1362 case EJUSTRETURN:
1363 break;
245e4f17
MD
1364 case EASYNC:
1365 panic("Unexpected EASYNC return value (for now)");
984263bc
MD
1366 default:
1367bad:
7062f5b4
EN
1368 if (p->p_sysent->sv_errsize) {
1369 if (error >= p->p_sysent->sv_errsize)
1370 error = -1; /* XXX */
1371 else
1372 error = p->p_sysent->sv_errtbl[error];
984263bc 1373 }
c7eb0589
SS
1374 frame->tf_eax = error;
1375 frame->tf_eflags |= PSL_C;
984263bc
MD
1376 break;
1377 }
1378
1379 /*
1380 * Traced syscall. trapsignal() is not MP aware.
1381 */
1382 if ((orig_tf_eflags & PSL_T) && !(orig_tf_eflags & PSL_VM)) {
270ac911 1383 MAKEMPSAFE(have_mplock);
c7eb0589 1384 frame->tf_eflags &= ~PSL_T;
f2642d5a 1385 trapsignal(lp, SIGTRAP, TRAP_TRACE);
984263bc
MD
1386 }
1387
1388 /*
1389 * Handle reschedule and other end-of-syscall issues
1390 */
c7eb0589 1391 userret(lp, frame, sticks);
984263bc
MD
1392
1393#ifdef KTRACE
dadab5e9 1394 if (KTRPOINT(td, KTR_SYSRET)) {
270ac911 1395 MAKEMPSAFE(have_mplock);
9fb04d14 1396 ktrsysret(lp, code, error, args.sysmsg_result);
984263bc
MD
1397 }
1398#endif
1399
1400 /*
1401 * This works because errno is findable through the
1402 * register set. If we ever support an emulation where this
1403 * is not the case, this code will need to be revisited.
1404 */
1405 STOPEVENT(p, S_SCX, code);
1406
81eea9f4 1407 userexit(lp);
8a8d5d85 1408#ifdef SMP
984263bc
MD
1409 /*
1410 * Release the MP lock if we had to get it
1411 */
270ac911
MD
1412 if (have_mplock)
1413 rel_mplock();
8a8d5d85 1414#endif
80bbc910 1415 KTR_LOG(kernentry_syscall_ret, p->p_pid, lp->lwp_tid, error);
7527bc81 1416#ifdef INVARIANTS
f9235b6d 1417 KASSERT(crit_count == td->td_critcount,
7527bc81 1418 ("syscall: critical section count mismatch! %d/%d",
f9235b6d 1419 crit_count, td->td_pri));
3933a3ab
MD
1420 KASSERT(&td->td_toks_base == td->td_toks_stop,
1421 ("syscall: extra tokens held after trap! %zd",
1422 td->td_toks_stop - &td->td_toks_base));
7527bc81 1423#endif
984263bc
MD
1424}
1425
2b0bd8aa
MD
1426/*
1427 * NOTE: MP lock not held at any point.
1428 */
91bd9c1e
SS
1429void
1430fork_return(struct lwp *lp, struct trapframe *frame)
1431{
1432 frame->tf_eax = 0; /* Child returns zero */
1433 frame->tf_eflags &= ~PSL_C; /* success */
1434 frame->tf_edx = 1;
1435
1436 generic_lwp_return(lp, frame);
80bbc910 1437 KTR_LOG(kernentry_fork_ret, lp->lwp_proc->p_pid, lp->lwp_tid);
91bd9c1e
SS
1438}
1439
a64ba182 1440/*
984263bc 1441 * Simplified back end of syscall(), used when returning from fork()
2b0bd8aa
MD
1442 * directly into user mode.
1443 *
1444 * This code will return back into the fork trampoline code which then
1445 * runs doreti.
1446 *
1447 * NOTE: The mplock is not held at any point.
984263bc
MD
1448 */
1449void
91bd9c1e 1450generic_lwp_return(struct lwp *lp, struct trapframe *frame)
984263bc 1451{
bb3cd951 1452 struct proc *p = lp->lwp_proc;
81eea9f4 1453
3a57cd9c
MD
1454 /*
1455 * Newly forked processes are given a kernel priority. We have to
1456 * adjust the priority to a normal user priority and fake entry
1457 * into the kernel (call userenter()) to install a passive release
1458 * function just in case userret() decides to stop the process. This
1459 * can occur when ^Z races a fork. If we do not install the passive
1460 * release function the current process designation will not be
1461 * released when the thread goes to sleep.
1462 */
1463 lwkt_setpri_self(TDPRI_USER_NORM);
d86a23e0 1464 userenter(lp->lwp_thread, p);
c7eb0589 1465 userret(lp, frame, 0);
984263bc 1466#ifdef KTRACE
81eea9f4 1467 if (KTRPOINT(lp->lwp_thread, KTR_SYSRET))
9fb04d14 1468 ktrsysret(lp, SYS_fork, 0, 0);
984263bc 1469#endif
4643740a 1470 lp->lwp_flags |= LWP_PASSIVE_ACQ;
81eea9f4 1471 userexit(lp);
4643740a 1472 lp->lwp_flags &= ~LWP_PASSIVE_ACQ;
984263bc 1473}
431d0fef
MD
1474
1475/*
4b486183 1476 * If PGEX_FPFAULT is set then set FP_VIRTFP in the PCB to force a T_DNA
431d0fef
MD
1477 * fault (which is then passed back to the virtual kernel) if an attempt is
1478 * made to use the FP unit.
1479 *
1480 * XXX this is a fairly big hack.
1481 */
1482void
1483set_vkernel_fp(struct trapframe *frame)
1484{
1485 struct thread *td = curthread;
1486
4b486183 1487 if (frame->tf_xflags & PGEX_FPFAULT) {
431d0fef
MD
1488 td->td_pcb->pcb_flags |= FP_VIRTFP;
1489 if (mdcpu->gd_npxthread == td)
1490 npxexit();
1491 } else {
1492 td->td_pcb->pcb_flags &= ~FP_VIRTFP;
1493 }
1494}
1495
bb47c072
MD
1496/*
1497 * Called from vkernel_trap() to fixup the vkernel's syscall
1498 * frame for vmspace_ctl() return.
1499 */
1500void
1501cpu_vkernel_trap(struct trapframe *frame, int error)
1502{
1503 frame->tf_eax = error;
1504 if (error)
1505 frame->tf_eflags |= PSL_C;
1506 else
1507 frame->tf_eflags &= ~PSL_C;
1508}