pc32: Split out isa_intr.h and move isa/intr_machdep.h to include/
[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>
984263bc
MD
98
99#ifdef POWERFAIL_NMI
100#include <sys/syslog.h>
101#include <machine/clock.h>
102#endif
103
104#include <machine/vm86.h>
105
106#include <ddb/ddb.h>
684a93c4 107
245e4f17 108#include <sys/msgport2.h>
41c20dac 109#include <sys/thread2.h>
684a93c4 110#include <sys/mplock2.h>
984263bc 111
9753ab6c
MD
112#ifdef SMP
113
114#define MAKEMPSAFE(have_mplock) \
115 if (have_mplock == 0) { \
116 get_mplock(); \
117 have_mplock = 1; \
118 }
119
120#else
121
122#define MAKEMPSAFE(have_mplock)
123
124#endif
125
3ae0cd58 126int (*pmath_emulate) (struct trapframe *);
984263bc 127
c7eb0589 128extern void trap (struct trapframe *frame);
c7eb0589 129extern void syscall2 (struct trapframe *frame);
984263bc 130
3ae0cd58
RG
131static int trap_pfault (struct trapframe *, int, vm_offset_t);
132static void trap_fatal (struct trapframe *, vm_offset_t);
133void dblfault_handler (void);
984263bc
MD
134
135extern inthand_t IDTVEC(syscall);
136
137#define MAX_TRAP_MSG 28
138static char *trap_msg[] = {
139 "", /* 0 unused */
140 "privileged instruction fault", /* 1 T_PRIVINFLT */
141 "", /* 2 unused */
142 "breakpoint instruction fault", /* 3 T_BPTFLT */
143 "", /* 4 unused */
144 "", /* 5 unused */
145 "arithmetic trap", /* 6 T_ARITHTRAP */
146 "system forced exception", /* 7 T_ASTFLT */
147 "", /* 8 unused */
148 "general protection fault", /* 9 T_PROTFLT */
149 "trace trap", /* 10 T_TRCTRAP */
150 "", /* 11 unused */
151 "page fault", /* 12 T_PAGEFLT */
152 "", /* 13 unused */
153 "alignment fault", /* 14 T_ALIGNFLT */
154 "", /* 15 unused */
155 "", /* 16 unused */
156 "", /* 17 unused */
157 "integer divide fault", /* 18 T_DIVIDE */
158 "non-maskable interrupt trap", /* 19 T_NMI */
159 "overflow trap", /* 20 T_OFLOW */
160 "FPU bounds check fault", /* 21 T_BOUND */
161 "FPU device not available", /* 22 T_DNA */
162 "double fault", /* 23 T_DOUBLEFLT */
163 "FPU operand fetch fault", /* 24 T_FPOPFLT */
164 "invalid TSS fault", /* 25 T_TSSFLT */
165 "segment not present fault", /* 26 T_SEGNPFLT */
166 "stack fault", /* 27 T_STKFLT */
167 "machine check trap", /* 28 T_MCHK */
168};
169
984263bc
MD
170#if defined(I586_CPU) && !defined(NO_F00F_HACK)
171extern int has_f00f_bug;
172#endif
173
174#ifdef DDB
175static int ddb_on_nmi = 1;
176SYSCTL_INT(_machdep, OID_AUTO, ddb_on_nmi, CTLFLAG_RW,
177 &ddb_on_nmi, 0, "Go to DDB on NMI");
178#endif
179static int panic_on_nmi = 1;
180SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RW,
181 &panic_on_nmi, 0, "Panic on NMI");
d9eea1a5
MD
182static int fast_release;
183SYSCTL_INT(_machdep, OID_AUTO, fast_release, CTLFLAG_RW,
184 &fast_release, 0, "Passive Release was optimal");
185static int slow_release;
186SYSCTL_INT(_machdep, OID_AUTO, slow_release, CTLFLAG_RW,
187 &slow_release, 0, "Passive Release was nonoptimal");
984263bc 188
4fd10eb6 189MALLOC_DEFINE(M_SYSMSG, "sysmsg", "sysmsg structure");
7062f5b4 190extern int max_sysmsg;
4fd10eb6 191
26a0694b 192/*
7bd858e5
MD
193 * Passively intercepts the thread switch function to increase the thread
194 * priority from a user priority to a kernel priority, reducing
d81ccc3e 195 * syscall and trap overhead for the case where no switch occurs.
7bd858e5 196 *
d86a23e0 197 * Synchronizes td_ucred with p_ucred. This is used by system calls,
7bd858e5
MD
198 * signal handling, faults, AST traps, and anything else that enters the
199 * kernel from userland and provides the kernel with a stable read-only
200 * copy of the process ucred.
26a0694b
MD
201 */
202static __inline void
d86a23e0 203userenter(struct thread *curtd, struct proc *curp)
26a0694b 204{
7bd858e5
MD
205 struct ucred *ocred;
206 struct ucred *ncred;
207
3824f392 208 curtd->td_release = lwkt_passive_release;
7bd858e5 209
d86a23e0
MD
210 if (curtd->td_ucred != curp->p_ucred) {
211 ncred = crhold(curp->p_ucred);
212 ocred = curtd->td_ucred;
213 curtd->td_ucred = ncred;
7bd858e5
MD
214 if (ocred)
215 crfree(ocred);
216 }
217
26a0694b
MD
218}
219
0a3f9b47 220/*
8ec60c3f
MD
221 * Handle signals, upcalls, profiling, and other AST's and/or tasks that
222 * must be completed before we can return to or try to return to userland.
0a3f9b47 223 *
8ec60c3f
MD
224 * Note that td_sticks is a 64 bit quantity, but there's no point doing 64
225 * arithmatic on the delta calculation so the absolute tick values are
226 * truncated to an integer.
37af14fe 227 */
a2a5ad0d 228static void
81eea9f4 229userret(struct lwp *lp, struct trapframe *frame, int sticks)
a2a5ad0d 230{
81eea9f4 231 struct proc *p = lp->lwp_proc;
8ba5f7ef 232 void (*hook)(void);
a2a5ad0d
MD
233 int sig;
234
8ba5f7ef
AH
235 if (p->p_userret != NULL) {
236 hook = p->p_userret;
237 p->p_userret = NULL;
238 (*hook)();
239 }
240
a2a5ad0d 241 /*
344ad853
MD
242 * Charge system time if profiling. Note: times are in microseconds.
243 * This may do a copyout and block, so do it first even though it
244 * means some system time will be charged as user time.
245 */
246 if (p->p_flag & P_PROFIL) {
247 addupc_task(p, frame->tf_eip,
bb3cd951 248 (u_int)((int)lp->lwp_thread->td_sticks - sticks));
344ad853
MD
249 }
250
251recheck:
252 /*
70d3d461
SS
253 * If the jungle wants us dead, so be it.
254 */
e207d1fd
MD
255 if (lp->lwp_flag & LWP_WEXIT) {
256 get_mplock();
e3161323 257 lwp_exit(0);
e207d1fd
MD
258 rel_mplock(); /* NOT REACHED */
259 }
70d3d461
SS
260
261 /*
344ad853
MD
262 * Block here if we are in a stopped state.
263 */
5ea440eb 264 if (p->p_stat == SSTOP || dump_stop_usertds) {
144ce500 265 get_mplock();
9a379a4a 266 tstop();
144ce500 267 rel_mplock();
344ad853
MD
268 goto recheck;
269 }
270
271 /*
4e7c41c5
MD
272 * Post any pending upcalls. If running a virtual kernel be sure
273 * to restore the virtual kernel's vmspace before posting the upcall.
a722be49
MD
274 */
275 if (p->p_flag & P_UPCALLPEND) {
276 p->p_flag &= ~P_UPCALLPEND;
144ce500 277 get_mplock();
81eea9f4 278 postupcall(lp);
144ce500 279 rel_mplock();
344ad853 280 goto recheck;
a722be49
MD
281 }
282
283 /*
4e7c41c5
MD
284 * Post any pending signals. If running a virtual kernel be sure
285 * to restore the virtual kernel's vmspace before posting the signal.
2883d2d8
MD
286 *
287 * WARNING! postsig() can exit and not return.
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
MD
295
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 340 /*
8ec60c3f
MD
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;
401#endif
7527bc81 402#ifdef INVARIANTS
f9235b6d 403 int crit_count = td->td_critcount;
3933a3ab 404 lwkt_tokref_t curstop = td->td_toks_stop;
7527bc81 405#endif
984263bc
MD
406 vm_offset_t eva;
407
7966cb69 408 p = td->td_proc;
984263bc 409#ifdef DDB
bfb6d329
AH
410 /*
411 * We need to allow T_DNA faults when the debugger is active since
2921c824 412 * some dumping paths do large bcopy() which use the floating
bfb6d329
AH
413 * point registers for faster copying.
414 */
5ea440eb 415 if (db_active && frame->tf_trapno != T_DNA) {
c7eb0589 416 eva = (frame->tf_trapno == T_PAGEFLT ? rcr2() : 0);
27e88a6e 417 ++gd->gd_trap_nesting_level;
9753ab6c 418 MAKEMPSAFE(have_mplock);
c7eb0589 419 trap_fatal(frame, eva);
27e88a6e 420 --gd->gd_trap_nesting_level;
8a8d5d85 421 goto out2;
984263bc
MD
422 }
423#endif
424
a2a5ad0d 425 eva = 0;
27e88a6e 426 ++gd->gd_trap_nesting_level;
c7eb0589 427 if (frame->tf_trapno == T_PAGEFLT) {
a2a5ad0d
MD
428 /*
429 * For some Cyrix CPUs, %cr2 is clobbered by interrupts.
430 * This problem is worked around by using an interrupt
431 * gate for the pagefault handler. We are finally ready
432 * to read %cr2 and then must reenable interrupts.
433 *
434 * XXX this should be in the switch statement, but the
435 * NO_FOOF_HACK and VM86 goto and ifdefs obfuscate the
436 * flow of control too much for this to be obviously
437 * correct.
438 */
439 eva = rcr2();
a2a5ad0d 440 cpu_enable_intr();
a2a5ad0d 441 }
80bbc910 442
27e88a6e 443 --gd->gd_trap_nesting_level;
a2a5ad0d 444
c7eb0589 445 if (!(frame->tf_eflags & PSL_I)) {
984263bc
MD
446 /*
447 * Buggy application or kernel code has disabled interrupts
448 * and then trapped. Enabling interrupts now is wrong, but
449 * it is better than running with interrupts disabled until
450 * they are accidentally enabled later.
451 */
c7eb0589
SS
452 type = frame->tf_trapno;
453 if (ISPL(frame->tf_cs)==SEL_UPL || (frame->tf_eflags & PSL_VM)) {
9753ab6c 454 MAKEMPSAFE(have_mplock);
26be20a0 455 kprintf(
984263bc
MD
456 "pid %ld (%s): trap %d with interrupts disabled\n",
457 (long)curproc->p_pid, curproc->p_comm, type);
a2a5ad0d 458 } else if (type != T_BPTFLT && type != T_TRCTRAP) {
984263bc
MD
459 /*
460 * XXX not quite right, since this may be for a
461 * multiple fault in user mode.
462 */
9753ab6c 463 MAKEMPSAFE(have_mplock);
26be20a0 464 kprintf("kernel trap %d with interrupts disabled\n",
984263bc 465 type);
a2a5ad0d 466 }
8a8d5d85 467 cpu_enable_intr();
984263bc
MD
468 }
469
984263bc
MD
470#if defined(I586_CPU) && !defined(NO_F00F_HACK)
471restart:
472#endif
c7eb0589
SS
473 type = frame->tf_trapno;
474 code = frame->tf_err;
984263bc
MD
475
476 if (in_vm86call) {
c7eb0589 477 if (frame->tf_eflags & PSL_VM &&
984263bc 478 (type == T_PROTFLT || type == T_STKFLT)) {
96728c05 479#ifdef SMP
b5d16701 480 KKASSERT(get_mplock_count(curthread) > 0);
96728c05 481#endif
c7eb0589 482 i = vm86_emulate((struct vm86frame *)frame);
96728c05 483#ifdef SMP
b5d16701 484 KKASSERT(get_mplock_count(curthread) > 0);
96728c05 485#endif
8a8d5d85 486 if (i != 0) {
984263bc
MD
487 /*
488 * returns to original process
489 */
d3d32139 490#ifdef SMP
c7eb0589 491 vm86_trap((struct vm86frame *)frame,
d3d32139
MD
492 have_mplock);
493#else
c7eb0589 494 vm86_trap((struct vm86frame *)frame, 0);
d3d32139
MD
495#endif
496 KKASSERT(0); /* NOT REACHED */
8a8d5d85
MD
497 }
498 goto out2;
984263bc
MD
499 }
500 switch (type) {
501 /*
502 * these traps want either a process context, or
503 * assume a normal userspace trap.
504 */
505 case T_PROTFLT:
506 case T_SEGNPFLT:
c7eb0589 507 trap_fatal(frame, eva);
8a8d5d85 508 goto out2;
984263bc
MD
509 case T_TRCTRAP:
510 type = T_BPTFLT; /* kernel breakpoint */
511 /* FALL THROUGH */
512 }
513 goto kernel_trap; /* normal kernel trap handling */
514 }
515
c7eb0589 516 if ((ISPL(frame->tf_cs) == SEL_UPL) || (frame->tf_eflags & PSL_VM)) {
984263bc
MD
517 /* user trap */
518
80bbc910
SS
519 KTR_LOG(kernentry_trap, p->p_pid, lp->lwp_tid,
520 frame->tf_trapno, eva);
521
d86a23e0 522 userenter(td, p);
26a0694b 523
37af14fe 524 sticks = (int)td->td_sticks;
c7eb0589 525 lp->lwp_md.md_regs = frame;
984263bc
MD
526
527 switch (type) {
528 case T_PRIVINFLT: /* privileged instruction fault */
f2642d5a 529 ucode = ILL_COPROC;
984263bc
MD
530 i = SIGILL;
531 break;
532
533 case T_BPTFLT: /* bpt instruction fault */
534 case T_TRCTRAP: /* trace trap */
c7eb0589 535 frame->tf_eflags &= ~PSL_T;
f2642d5a 536 ucode = TRAP_TRACE;
984263bc
MD
537 i = SIGTRAP;
538 break;
539
540 case T_ARITHTRAP: /* arithmetic trap */
541 ucode = code;
542 i = SIGFPE;
543 break;
544
545 case T_ASTFLT: /* Allow process switch */
12e4aaff 546 mycpu->gd_cnt.v_soft++;
235957ed 547 if (mycpu->gd_reqflags & RQF_AST_OWEUPC) {
2a418930
MD
548 atomic_clear_int(&mycpu->gd_reqflags,
549 RQF_AST_OWEUPC);
93328593
SS
550 addupc_task(p, p->p_prof.pr_addr,
551 p->p_prof.pr_ticks);
984263bc
MD
552 }
553 goto out;
554
555 /*
556 * The following two traps can happen in
557 * vm86 mode, and, if so, we want to handle
558 * them specially.
559 */
560 case T_PROTFLT: /* general protection fault */
561 case T_STKFLT: /* stack fault */
c7eb0589
SS
562 if (frame->tf_eflags & PSL_VM) {
563 i = vm86_emulate((struct vm86frame *)frame);
984263bc
MD
564 if (i == 0)
565 goto out;
566 break;
567 }
f2642d5a
AH
568 i = SIGBUS;
569 ucode = (type == T_PROTFLT) ? BUS_OBJERR : BUS_ADRERR;
570 break;
984263bc 571 case T_SEGNPFLT: /* segment not present fault */
f2642d5a
AH
572 i = SIGBUS;
573 ucode = BUS_ADRERR;
574 break;
984263bc
MD
575 case T_TSSFLT: /* invalid TSS fault */
576 case T_DOUBLEFLT: /* double fault */
f2642d5a
AH
577 i = SIGBUS;
578 ucode = BUS_OBJERR;
984263bc 579 default:
f2642d5a
AH
580#if 0
581 ucode = code + BUS_SEGM_FAULT ; /* XXX: ???*/
582#endif
583 ucode = BUS_OBJERR;
984263bc
MD
584 i = SIGBUS;
585 break;
586
587 case T_PAGEFLT: /* page fault */
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 */
c7eb0589 706 trap_pfault(frame, FALSE, eva);
8a8d5d85 707 goto out2;
984263bc
MD
708
709 case T_DNA:
710#if NNPX > 0
711 /*
a02705a9
MD
712 * The kernel may be using npx for copying or other
713 * purposes.
984263bc
MD
714 */
715 if (npxdna())
8a8d5d85 716 goto out2;
984263bc
MD
717#endif
718 break;
719
720 case T_PROTFLT: /* general protection fault */
721 case T_SEGNPFLT: /* segment not present fault */
722 /*
723 * Invalid segment selectors and out of bounds
724 * %eip's and %esp's can be set up in user mode.
725 * This causes a fault in kernel mode when the
726 * kernel tries to return to user mode. We want
727 * to get this fault so that we can fix the
728 * problem here and not have to check all the
729 * selectors and pointers when the user changes
730 * them.
731 */
732#define MAYBE_DORETI_FAULT(where, whereto) \
733 do { \
c7eb0589
SS
734 if (frame->tf_eip == (int)where) { \
735 frame->tf_eip = (int)whereto; \
8a8d5d85 736 goto out2; \
984263bc
MD
737 } \
738 } while (0)
ef0fdad1 739 if (mycpu->gd_intr_nesting_level == 0) {
984263bc
MD
740 /*
741 * Invalid %fs's and %gs's can be created using
742 * procfs or PT_SETREGS or by invalidating the
743 * underlying LDT entry. This causes a fault
744 * in kernel mode when the kernel attempts to
745 * switch contexts. Lose the bad context
746 * (XXX) so that we can continue, and generate
747 * a signal.
748 */
984263bc
MD
749 MAYBE_DORETI_FAULT(doreti_iret,
750 doreti_iret_fault);
751 MAYBE_DORETI_FAULT(doreti_popl_ds,
752 doreti_popl_ds_fault);
753 MAYBE_DORETI_FAULT(doreti_popl_es,
754 doreti_popl_es_fault);
755 MAYBE_DORETI_FAULT(doreti_popl_fs,
756 doreti_popl_fs_fault);
4e7c41c5
MD
757 MAYBE_DORETI_FAULT(doreti_popl_gs,
758 doreti_popl_gs_fault);
37af14fe 759 if (td->td_pcb->pcb_onfault) {
c7eb0589 760 frame->tf_eip =
37af14fe 761 (register_t)td->td_pcb->pcb_onfault;
8a8d5d85 762 goto out2;
984263bc
MD
763 }
764 }
765 break;
766
767 case T_TSSFLT:
768 /*
769 * PSL_NT can be set in user mode and isn't cleared
770 * automatically when the kernel is entered. This
771 * causes a TSS fault when the kernel attempts to
772 * `iret' because the TSS link is uninitialized. We
773 * want to get this fault so that we can fix the
774 * problem here and not every time the kernel is
775 * entered.
776 */
c7eb0589
SS
777 if (frame->tf_eflags & PSL_NT) {
778 frame->tf_eflags &= ~PSL_NT;
8a8d5d85 779 goto out2;
984263bc
MD
780 }
781 break;
782
783 case T_TRCTRAP: /* trace trap */
c7eb0589 784 if (frame->tf_eip == (int)IDTVEC(syscall)) {
984263bc
MD
785 /*
786 * We've just entered system mode via the
787 * syscall lcall. Continue single stepping
788 * silently until the syscall handler has
789 * saved the flags.
790 */
8a8d5d85 791 goto out2;
984263bc 792 }
c7eb0589 793 if (frame->tf_eip == (int)IDTVEC(syscall) + 1) {
984263bc
MD
794 /*
795 * The syscall handler has now saved the
796 * flags. Stop single stepping it.
797 */
c7eb0589 798 frame->tf_eflags &= ~PSL_T;
8a8d5d85 799 goto out2;
984263bc
MD
800 }
801 /*
802 * Ignore debug register trace traps due to
803 * accesses in the user's address space, which
804 * can happen under several conditions such as
805 * if a user sets a watchpoint on a buffer and
806 * then passes that buffer to a system call.
807 * We still want to get TRCTRAPS for addresses
808 * in kernel space because that is useful when
809 * debugging the kernel.
810 */
811 if (user_dbreg_trap()) {
812 /*
813 * Reset breakpoint bits because the
814 * processor doesn't
815 */
816 load_dr6(rdr6() & 0xfffffff0);
8a8d5d85 817 goto out2;
984263bc
MD
818 }
819 /*
820 * Fall through (TRCTRAP kernel mode, kernel address)
821 */
822 case T_BPTFLT:
823 /*
824 * If DDB is enabled, let it handle the debugger trap.
825 * Otherwise, debugger traps "can't happen".
826 */
f2642d5a 827 ucode = TRAP_BRKPT;
984263bc 828#ifdef DDB
9753ab6c 829 MAKEMPSAFE(have_mplock);
c7eb0589 830 if (kdb_trap (type, 0, frame))
8a8d5d85 831 goto out2;
984263bc
MD
832#endif
833 break;
834
835#if NISA > 0
836 case T_NMI:
9753ab6c 837 MAKEMPSAFE(have_mplock);
984263bc
MD
838#ifdef POWERFAIL_NMI
839#ifndef TIMER_FREQ
840# define TIMER_FREQ 1193182
841#endif
842 handle_powerfail:
843 {
844 static unsigned lastalert = 0;
845
846 if(time_second - lastalert > 10)
847 {
848 log(LOG_WARNING, "NMI: power fail\n");
849 sysbeep(TIMER_FREQ/880, hz);
850 lastalert = time_second;
851 }
8a8d5d85
MD
852 /* YYY mp count */
853 goto out2;
984263bc
MD
854 }
855#else /* !POWERFAIL_NMI */
856 /* machine/parity/power fail/"kitchen sink" faults */
857 if (isa_nmi(code) == 0) {
858#ifdef DDB
859 /*
860 * NMI can be hooked up to a pushbutton
861 * for debugging.
862 */
863 if (ddb_on_nmi) {
26be20a0 864 kprintf ("NMI ... going to debugger\n");
c7eb0589 865 kdb_trap (type, 0, frame);
984263bc
MD
866 }
867#endif /* DDB */
8a8d5d85 868 goto out2;
984263bc 869 } else if (panic_on_nmi == 0)
8a8d5d85 870 goto out2;
984263bc
MD
871 /* FALL THROUGH */
872#endif /* POWERFAIL_NMI */
873#endif /* NISA > 0 */
874 }
875
9753ab6c 876 MAKEMPSAFE(have_mplock);
c7eb0589 877 trap_fatal(frame, eva);
8a8d5d85 878 goto out2;
984263bc
MD
879 }
880
4a22e893
MD
881 /*
882 * Virtual kernel intercept - if the fault is directly related to a
883 * VM context managed by a virtual kernel then let the virtual kernel
884 * handle it.
885 */
39005e16 886 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
287ebb09 887 vkernel_trap(lp, frame);
4a22e893
MD
888 goto out;
889 }
890
891 /*
892 * Translate fault for emulators (e.g. Linux)
893 */
984263bc
MD
894 if (*p->p_sysent->sv_transtrap)
895 i = (*p->p_sysent->sv_transtrap)(i, type);
896
9753ab6c 897 MAKEMPSAFE(have_mplock);
08f2f1bb 898 trapsignal(lp, i, ucode);
984263bc
MD
899
900#ifdef DEBUG
901 if (type <= MAX_TRAP_MSG) {
902 uprintf("fatal process exception: %s",
903 trap_msg[type]);
904 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
905 uprintf(", fault VA = 0x%lx", (u_long)eva);
906 uprintf("\n");
907 }
908#endif
909
910out:
c7eb0589 911 userret(lp, frame, sticks);
81eea9f4 912 userexit(lp);
9753ab6c 913out2: ;
96728c05 914#ifdef SMP
9753ab6c
MD
915 if (have_mplock)
916 rel_mplock();
96728c05 917#endif
80bbc910
SS
918 if (p != NULL && lp != NULL)
919 KTR_LOG(kernentry_trap_ret, p->p_pid, lp->lwp_tid);
7527bc81 920#ifdef INVARIANTS
f9235b6d 921 KASSERT(crit_count == td->td_critcount,
3933a3ab 922 ("trap: critical section count mismatch! %d/%d",
f9235b6d 923 crit_count, td->td_pri));
3933a3ab
MD
924 KASSERT(curstop == td->td_toks_stop,
925 ("trap: extra tokens held after trap! %zd/%zd",
926 curstop - &td->td_toks_base,
927 td->td_toks_stop - &td->td_toks_base));
7527bc81 928#endif
984263bc
MD
929}
930
984263bc 931int
f123d5a1 932trap_pfault(struct trapframe *frame, int usermode, vm_offset_t eva)
984263bc
MD
933{
934 vm_offset_t va;
935 struct vmspace *vm = NULL;
936 vm_map_t map = 0;
937 int rv = 0;
1b9d3514 938 int fault_flags;
984263bc 939 vm_prot_t ftype;
37af14fe 940 thread_t td = curthread;
287ebb09 941 struct lwp *lp = td->td_lwp;
984263bc
MD
942
943 va = trunc_page(eva);
944 if (va >= KERNBASE) {
945 /*
946 * Don't allow user-mode faults in kernel address space.
947 * An exception: if the faulting address is the invalid
948 * instruction entry in the IDT, then the Intel Pentium
949 * F00F bug workaround was triggered, and we need to
950 * treat it is as an illegal instruction, and not a page
951 * fault.
952 */
953#if defined(I586_CPU) && !defined(NO_F00F_HACK)
954 if ((eva == (unsigned int)&idt[6]) && has_f00f_bug) {
955 frame->tf_trapno = T_PRIVINFLT;
956 return -2;
957 }
958#endif
959 if (usermode)
960 goto nogo;
961
e4846942 962 map = &kernel_map;
984263bc
MD
963 } else {
964 /*
965 * This is a fault on non-kernel virtual memory.
966 * vm is initialized above to NULL. If curproc is NULL
967 * or curproc->p_vmspace is NULL the fault is fatal.
968 */
287ebb09
MD
969 if (lp != NULL)
970 vm = lp->lwp_vmspace;
984263bc
MD
971
972 if (vm == NULL)
973 goto nogo;
974
975 map = &vm->vm_map;
976 }
977
978 if (frame->tf_err & PGEX_W)
979 ftype = VM_PROT_WRITE;
980 else
981 ftype = VM_PROT_READ;
982
e4846942 983 if (map != &kernel_map) {
984263bc
MD
984 /*
985 * Keep swapout from messing with us during this
986 * critical time.
987 */
287ebb09 988 PHOLD(lp->lwp_proc);
984263bc
MD
989
990 /*
1b9d3514
MD
991 * Issue fault
992 */
993 fault_flags = 0;
994 if (usermode)
995 fault_flags |= VM_FAULT_BURST;
996 if (ftype & VM_PROT_WRITE)
997 fault_flags |= VM_FAULT_DIRTY;
998 else
999 fault_flags |= VM_FAULT_NORMAL;
1000 rv = vm_fault(map, va, ftype, fault_flags);
287ebb09 1001 PRELE(lp->lwp_proc);
984263bc
MD
1002 } else {
1003 /*
17cde63e
MD
1004 * Don't have to worry about process locking or stacks
1005 * in the kernel.
984263bc
MD
1006 */
1007 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
1008 }
1009
1010 if (rv == KERN_SUCCESS)
1011 return (0);
1012nogo:
1013 if (!usermode) {
37af14fe
MD
1014 if (td->td_gd->gd_intr_nesting_level == 0 &&
1015 td->td_pcb->pcb_onfault) {
1016 frame->tf_eip = (register_t)td->td_pcb->pcb_onfault;
984263bc
MD
1017 return (0);
1018 }
1019 trap_fatal(frame, eva);
1020 return (-1);
1021 }
1022
1023 /* kludge to pass faulting virtual address to sendsig */
4e7c41c5 1024 frame->tf_xflags = frame->tf_err;
984263bc
MD
1025 frame->tf_err = eva;
1026
1027 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
1028}
1029
1030static void
f123d5a1 1031trap_fatal(struct trapframe *frame, vm_offset_t eva)
984263bc
MD
1032{
1033 int code, type, ss, esp;
1034 struct soft_segment_descriptor softseg;
1035
1036 code = frame->tf_err;
1037 type = frame->tf_trapno;
3951a45f 1038 sdtossd(&gdt[mycpu->gd_cpuid * NGDT + IDXSEL(frame->tf_cs & 0xffff)].sd, &softseg);
984263bc
MD
1039
1040 if (type <= MAX_TRAP_MSG)
26be20a0 1041 kprintf("\n\nFatal trap %d: %s while in %s mode\n",
984263bc
MD
1042 type, trap_msg[type],
1043 frame->tf_eflags & PSL_VM ? "vm86" :
1044 ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel");
1045#ifdef SMP
7a44d1cb 1046 /* three separate prints in case of a trap on an unmapped page */
26be20a0
SW
1047 kprintf("cpuid = %d; ", mycpu->gd_cpuid);
1048 kprintf("lapic.id = %08x\n", lapic.id);
984263bc
MD
1049#endif
1050 if (type == T_PAGEFLT) {
d557216f 1051 kprintf("fault virtual address = %p\n", (void *)eva);
26be20a0 1052 kprintf("fault code = %s %s, %s\n",
984263bc
MD
1053 code & PGEX_U ? "user" : "supervisor",
1054 code & PGEX_W ? "write" : "read",
1055 code & PGEX_P ? "protection violation" : "page not present");
1056 }
26be20a0 1057 kprintf("instruction pointer = 0x%x:0x%x\n",
984263bc
MD
1058 frame->tf_cs & 0xffff, frame->tf_eip);
1059 if ((ISPL(frame->tf_cs) == SEL_UPL) || (frame->tf_eflags & PSL_VM)) {
1060 ss = frame->tf_ss & 0xffff;
1061 esp = frame->tf_esp;
1062 } else {
1063 ss = GSEL(GDATA_SEL, SEL_KPL);
1064 esp = (int)&frame->tf_esp;
1065 }
26be20a0
SW
1066 kprintf("stack pointer = 0x%x:0x%x\n", ss, esp);
1067 kprintf("frame pointer = 0x%x:0x%x\n", ss, frame->tf_ebp);
1068 kprintf("code segment = base 0x%x, limit 0x%x, type 0x%x\n",
984263bc 1069 softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type);
26be20a0 1070 kprintf(" = DPL %d, pres %d, def32 %d, gran %d\n",
984263bc
MD
1071 softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_def32,
1072 softseg.ssd_gran);
26be20a0 1073 kprintf("processor eflags = ");
984263bc 1074 if (frame->tf_eflags & PSL_T)
26be20a0 1075 kprintf("trace trap, ");
984263bc 1076 if (frame->tf_eflags & PSL_I)
26be20a0 1077 kprintf("interrupt enabled, ");
984263bc 1078 if (frame->tf_eflags & PSL_NT)
26be20a0 1079 kprintf("nested task, ");
984263bc 1080 if (frame->tf_eflags & PSL_RF)
26be20a0 1081 kprintf("resume, ");
984263bc 1082 if (frame->tf_eflags & PSL_VM)
26be20a0
SW
1083 kprintf("vm86, ");
1084 kprintf("IOPL = %d\n", (frame->tf_eflags & PSL_IOPL) >> 12);
1085 kprintf("current process = ");
984263bc 1086 if (curproc) {
26be20a0 1087 kprintf("%lu (%s)\n",
984263bc
MD
1088 (u_long)curproc->p_pid, curproc->p_comm ?
1089 curproc->p_comm : "");
1090 } else {
26be20a0 1091 kprintf("Idle\n");
984263bc 1092 }
26be20a0 1093 kprintf("current thread = pri %d ", curthread->td_pri);
f9235b6d 1094 if (curthread->td_critcount)
26be20a0
SW
1095 kprintf("(CRIT)");
1096 kprintf("\n");
984263bc
MD
1097#ifdef SMP
1098/**
1099 * XXX FIXME:
1100 * we probably SHOULD have stopped the other CPUs before now!
1101 * another CPU COULD have been touching cpl at this moment...
1102 */
26be20a0 1103 kprintf(" <- SMP: XXX");
984263bc 1104#endif
26be20a0 1105 kprintf("\n");
984263bc
MD
1106
1107#ifdef KDB
1108 if (kdb_trap(&psl))
1109 return;
1110#endif
1111#ifdef DDB
f7bc9806 1112 if ((debugger_on_panic || db_active) && kdb_trap(type, code, frame))
984263bc
MD
1113 return;
1114#endif
26be20a0 1115 kprintf("trap number = %d\n", type);
984263bc
MD
1116 if (type <= MAX_TRAP_MSG)
1117 panic("%s", trap_msg[type]);
1118 else
1119 panic("unknown/reserved trap");
1120}
1121
1122/*
1123 * Double fault handler. Called when a fault occurs while writing
1124 * a frame for a trap/exception onto the stack. This usually occurs
1125 * when the stack overflows (such is the case with infinite recursion,
1126 * for example).
1127 *
1128 * XXX Note that the current PTD gets replaced by IdlePTD when the
1129 * task switch occurs. This means that the stack that was active at
1130 * the time of the double fault is not available at <kstack> unless
1131 * the machine was idle when the double fault occurred. The downside
1132 * of this is that "trace <ebp>" in ddb won't work.
1133 */
093565f2
MD
1134static __inline
1135int
1136in_kstack_guard(register_t rptr)
1137{
1138 thread_t td = curthread;
1139
1140 if ((char *)rptr >= td->td_kstack &&
1141 (char *)rptr < td->td_kstack + PAGE_SIZE) {
1142 return 1;
1143 }
1144 return 0;
1145}
1146
984263bc 1147void
f123d5a1 1148dblfault_handler(void)
984263bc 1149{
85100692 1150 struct mdglobaldata *gd = mdcpu;
17a9f566 1151
093565f2
MD
1152 if (in_kstack_guard(gd->gd_common_tss.tss_esp) ||
1153 in_kstack_guard(gd->gd_common_tss.tss_ebp)) {
0174149e 1154 kprintf("DOUBLE FAULT - KERNEL STACK GUARD HIT!\n");
093565f2 1155 } else {
0174149e 1156 kprintf("DOUBLE FAULT:\n");
093565f2 1157 }
26be20a0
SW
1158 kprintf("eip = 0x%x\n", gd->gd_common_tss.tss_eip);
1159 kprintf("esp = 0x%x\n", gd->gd_common_tss.tss_esp);
1160 kprintf("ebp = 0x%x\n", gd->gd_common_tss.tss_ebp);
984263bc 1161#ifdef SMP
7a44d1cb 1162 /* three separate prints in case of a trap on an unmapped page */
093565f2 1163 kprintf("cpuid = %d; ", gd->mi.gd_cpuid);
26be20a0 1164 kprintf("lapic.id = %08x\n", lapic.id);
984263bc
MD
1165#endif
1166 panic("double fault");
1167}
1168
1169/*
0d355d3b 1170 * syscall2 - MP aware system call request C handler
984263bc 1171 *
0d355d3b
MD
1172 * A system call is essentially treated as a trap. The MP lock is not
1173 * held on entry or return. We are responsible for handling ASTs
1174 * (e.g. a task switch) prior to return.
984263bc 1175 *
0d355d3b 1176 * MPSAFE
984263bc
MD
1177 */
1178void
c7eb0589 1179syscall2(struct trapframe *frame)
984263bc 1180{
dadab5e9
MD
1181 struct thread *td = curthread;
1182 struct proc *p = td->td_proc;
81eea9f4 1183 struct lwp *lp = td->td_lwp;
984263bc 1184 caddr_t params;
984263bc 1185 struct sysent *callp;
984263bc 1186 register_t orig_tf_eflags;
37af14fe 1187 int sticks;
984263bc
MD
1188 int error;
1189 int narg;
7527bc81 1190#ifdef INVARIANTS
f9235b6d 1191 int crit_count = td->td_critcount;
7527bc81 1192#endif
270ac911
MD
1193#ifdef SMP
1194 int have_mplock = 0;
1195#endif
984263bc 1196 u_int code;
a64ba182 1197 union sysunion args;
984263bc
MD
1198
1199#ifdef DIAGNOSTIC
c7eb0589 1200 if (ISPL(frame->tf_cs) != SEL_UPL) {
984263bc
MD
1201 get_mplock();
1202 panic("syscall");
1203 /* NOT REACHED */
1204 }
1205#endif
1206
80bbc910
SS
1207 KTR_LOG(kernentry_syscall, p->p_pid, lp->lwp_tid,
1208 frame->tf_eax);
1209
d86a23e0 1210 userenter(td, p); /* lazy raise our priority */
37af14fe 1211
4a22e893
MD
1212 /*
1213 * Misc
1214 */
37af14fe 1215 sticks = (int)td->td_sticks;
c7eb0589 1216 orig_tf_eflags = frame->tf_eflags;
4a22e893
MD
1217
1218 /*
1219 * Virtual kernel intercept - if a VM context managed by a virtual
1220 * kernel issues a system call the virtual kernel handles it, not us.
1221 * Restore the virtual kernel context and return from its system
1222 * call. The current frame is copied out to the virtual kernel.
1223 */
39005e16 1224 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
bb47c072 1225 vkernel_trap(lp, frame);
4a22e893 1226 error = EJUSTRETURN;
e7912058 1227 callp = NULL;
4a22e893
MD
1228 goto out;
1229 }
984263bc 1230
4a22e893
MD
1231 /*
1232 * Get the system call parameters and account for time
1233 */
c7eb0589
SS
1234 lp->lwp_md.md_regs = frame;
1235 params = (caddr_t)frame->tf_esp + sizeof(int);
1236 code = frame->tf_eax;
984263bc
MD
1237
1238 if (p->p_sysent->sv_prepsyscall) {
270ac911 1239 (*p->p_sysent->sv_prepsyscall)(
c7eb0589 1240 frame, (int *)(&args.nosys.sysmsg + 1),
270ac911 1241 &code, &params);
984263bc
MD
1242 } else {
1243 /*
1244 * Need to check if this is a 32 bit or 64 bit syscall.
1245 * fuword is MP aware.
1246 */
1247 if (code == SYS_syscall) {
1248 /*
1249 * Code is first argument, followed by actual args.
1250 */
1251 code = fuword(params);
1252 params += sizeof(int);
1253 } else if (code == SYS___syscall) {
1254 /*
1255 * Like syscall, but code is a quad, so as to maintain
1256 * quad alignment for the rest of the arguments.
1257 */
1258 code = fuword(params);
1259 params += sizeof(quad_t);
1260 }
1261 }
1262
8ec60c3f 1263 code &= p->p_sysent->sv_mask;
8ba5f7ef 1264
7062f5b4
EN
1265 if (code >= p->p_sysent->sv_size)
1266 callp = &p->p_sysent->sv_table[0];
1267 else
1268 callp = &p->p_sysent->sv_table[code];
984263bc
MD
1269
1270 narg = callp->sy_narg & SYF_ARGMASK;
1271
8ba5f7ef
AH
1272#if 0
1273 if (p->p_sysent->sv_name[0] == 'L')
1274 kprintf("Linux syscall, code = %d\n", code);
1275#endif
1276
984263bc
MD
1277 /*
1278 * copyin is MP aware, but the tracing code is not
1279 */
8ec60c3f 1280 if (narg && params) {
f9a13fc4 1281 error = copyin(params, (caddr_t)(&args.nosys.sysmsg + 1),
8ec60c3f
MD
1282 narg * sizeof(register_t));
1283 if (error) {
984263bc 1284#ifdef KTRACE
270ac911
MD
1285 if (KTRPOINT(td, KTR_SYSCALL)) {
1286 MAKEMPSAFE(have_mplock);
1287
9fb04d14 1288 ktrsyscall(lp, code, narg,
f9a13fc4 1289 (void *)(&args.nosys.sysmsg + 1));
270ac911 1290 }
984263bc 1291#endif
8ec60c3f
MD
1292 goto bad;
1293 }
984263bc
MD
1294 }
1295
984263bc 1296#ifdef KTRACE
dadab5e9 1297 if (KTRPOINT(td, KTR_SYSCALL)) {
270ac911 1298 MAKEMPSAFE(have_mplock);
9fb04d14 1299 ktrsyscall(lp, code, narg, (void *)(&args.nosys.sysmsg + 1));
984263bc
MD
1300 }
1301#endif
01dce7bb
MD
1302
1303 /*
1304 * For traditional syscall code edx is left untouched when 32 bit
1305 * results are returned. Since edx is loaded from fds[1] when the
1306 * system call returns we pre-set it here.
1307 */
c7114eea 1308 args.sysmsg_fds[0] = 0;
c7eb0589 1309 args.sysmsg_fds[1] = frame->tf_edx;
984263bc 1310
4a22e893
MD
1311 /*
1312 * The syscall might manipulate the trap frame. If it does it
1313 * will probably return EJUSTRETURN.
1314 */
c7eb0589 1315 args.sysmsg_frame = frame;
4a22e893 1316
984263bc
MD
1317 STOPEVENT(p, S_SCE, narg); /* MP aware */
1318
270ac911 1319 /*
3919ced0
MD
1320 * NOTE: All system calls run MPSAFE now. The system call itself
1321 * is responsible for getting the MP lock.
270ac911 1322 */
a64ba182 1323 error = (*callp->sy_call)(&args);
984263bc 1324
4a22e893 1325out:
984263bc
MD
1326 /*
1327 * MP SAFE (we may or may not have the MP lock at this point)
1328 */
1329 switch (error) {
1330 case 0:
1331 /*
1332 * Reinitialize proc pointer `p' as it may be different
1333 * if this is a child returning from fork syscall.
1334 */
1335 p = curproc;
81eea9f4 1336 lp = curthread->td_lwp;
c7eb0589
SS
1337 frame->tf_eax = args.sysmsg_fds[0];
1338 frame->tf_edx = args.sysmsg_fds[1];
1339 frame->tf_eflags &= ~PSL_C;
984263bc 1340 break;
984263bc
MD
1341 case ERESTART:
1342 /*
1343 * Reconstruct pc, assuming lcall $X,y is 7 bytes,
1344 * int 0x80 is 2 bytes. We saved this in tf_err.
1345 */
c7eb0589 1346 frame->tf_eip -= frame->tf_err;
984263bc 1347 break;
984263bc
MD
1348 case EJUSTRETURN:
1349 break;
245e4f17
MD
1350 case EASYNC:
1351 panic("Unexpected EASYNC return value (for now)");
984263bc
MD
1352 default:
1353bad:
7062f5b4
EN
1354 if (p->p_sysent->sv_errsize) {
1355 if (error >= p->p_sysent->sv_errsize)
1356 error = -1; /* XXX */
1357 else
1358 error = p->p_sysent->sv_errtbl[error];
984263bc 1359 }
c7eb0589
SS
1360 frame->tf_eax = error;
1361 frame->tf_eflags |= PSL_C;
984263bc
MD
1362 break;
1363 }
1364
1365 /*
1366 * Traced syscall. trapsignal() is not MP aware.
1367 */
1368 if ((orig_tf_eflags & PSL_T) && !(orig_tf_eflags & PSL_VM)) {
270ac911 1369 MAKEMPSAFE(have_mplock);
c7eb0589 1370 frame->tf_eflags &= ~PSL_T;
f2642d5a 1371 trapsignal(lp, SIGTRAP, TRAP_TRACE);
984263bc
MD
1372 }
1373
1374 /*
1375 * Handle reschedule and other end-of-syscall issues
1376 */
c7eb0589 1377 userret(lp, frame, sticks);
984263bc
MD
1378
1379#ifdef KTRACE
dadab5e9 1380 if (KTRPOINT(td, KTR_SYSRET)) {
270ac911 1381 MAKEMPSAFE(have_mplock);
9fb04d14 1382 ktrsysret(lp, code, error, args.sysmsg_result);
984263bc
MD
1383 }
1384#endif
1385
1386 /*
1387 * This works because errno is findable through the
1388 * register set. If we ever support an emulation where this
1389 * is not the case, this code will need to be revisited.
1390 */
1391 STOPEVENT(p, S_SCX, code);
1392
81eea9f4 1393 userexit(lp);
8a8d5d85 1394#ifdef SMP
984263bc
MD
1395 /*
1396 * Release the MP lock if we had to get it
1397 */
270ac911
MD
1398 if (have_mplock)
1399 rel_mplock();
8a8d5d85 1400#endif
80bbc910 1401 KTR_LOG(kernentry_syscall_ret, p->p_pid, lp->lwp_tid, error);
7527bc81 1402#ifdef INVARIANTS
f9235b6d 1403 KASSERT(crit_count == td->td_critcount,
7527bc81 1404 ("syscall: critical section count mismatch! %d/%d",
f9235b6d 1405 crit_count, td->td_pri));
3933a3ab
MD
1406 KASSERT(&td->td_toks_base == td->td_toks_stop,
1407 ("syscall: extra tokens held after trap! %zd",
1408 td->td_toks_stop - &td->td_toks_base));
7527bc81 1409#endif
984263bc
MD
1410}
1411
2b0bd8aa
MD
1412/*
1413 * NOTE: MP lock not held at any point.
1414 */
91bd9c1e
SS
1415void
1416fork_return(struct lwp *lp, struct trapframe *frame)
1417{
1418 frame->tf_eax = 0; /* Child returns zero */
1419 frame->tf_eflags &= ~PSL_C; /* success */
1420 frame->tf_edx = 1;
1421
1422 generic_lwp_return(lp, frame);
80bbc910 1423 KTR_LOG(kernentry_fork_ret, lp->lwp_proc->p_pid, lp->lwp_tid);
91bd9c1e
SS
1424}
1425
a64ba182 1426/*
984263bc 1427 * Simplified back end of syscall(), used when returning from fork()
2b0bd8aa
MD
1428 * directly into user mode.
1429 *
1430 * This code will return back into the fork trampoline code which then
1431 * runs doreti.
1432 *
1433 * NOTE: The mplock is not held at any point.
984263bc
MD
1434 */
1435void
91bd9c1e 1436generic_lwp_return(struct lwp *lp, struct trapframe *frame)
984263bc 1437{
bb3cd951 1438 struct proc *p = lp->lwp_proc;
81eea9f4 1439
3a57cd9c
MD
1440 /*
1441 * Newly forked processes are given a kernel priority. We have to
1442 * adjust the priority to a normal user priority and fake entry
1443 * into the kernel (call userenter()) to install a passive release
1444 * function just in case userret() decides to stop the process. This
1445 * can occur when ^Z races a fork. If we do not install the passive
1446 * release function the current process designation will not be
1447 * released when the thread goes to sleep.
1448 */
1449 lwkt_setpri_self(TDPRI_USER_NORM);
d86a23e0 1450 userenter(lp->lwp_thread, p);
c7eb0589 1451 userret(lp, frame, 0);
984263bc 1452#ifdef KTRACE
81eea9f4 1453 if (KTRPOINT(lp->lwp_thread, KTR_SYSRET))
9fb04d14 1454 ktrsysret(lp, SYS_fork, 0, 0);
984263bc 1455#endif
d9eea1a5 1456 p->p_flag |= P_PASSIVE_ACQ;
81eea9f4 1457 userexit(lp);
d9eea1a5 1458 p->p_flag &= ~P_PASSIVE_ACQ;
984263bc 1459}
431d0fef
MD
1460
1461/*
4b486183 1462 * If PGEX_FPFAULT is set then set FP_VIRTFP in the PCB to force a T_DNA
431d0fef
MD
1463 * fault (which is then passed back to the virtual kernel) if an attempt is
1464 * made to use the FP unit.
1465 *
1466 * XXX this is a fairly big hack.
1467 */
1468void
1469set_vkernel_fp(struct trapframe *frame)
1470{
1471 struct thread *td = curthread;
1472
4b486183 1473 if (frame->tf_xflags & PGEX_FPFAULT) {
431d0fef
MD
1474 td->td_pcb->pcb_flags |= FP_VIRTFP;
1475 if (mdcpu->gd_npxthread == td)
1476 npxexit();
1477 } else {
1478 td->td_pcb->pcb_flags &= ~FP_VIRTFP;
1479 }
1480}
1481
bb47c072
MD
1482/*
1483 * Called from vkernel_trap() to fixup the vkernel's syscall
1484 * frame for vmspace_ctl() return.
1485 */
1486void
1487cpu_vkernel_trap(struct trapframe *frame, int error)
1488{
1489 frame->tf_eax = error;
1490 if (error)
1491 frame->tf_eflags |= PSL_C;
1492 else
1493 frame->tf_eflags &= ~PSL_C;
1494}