Remove upc_{control,register} syscalls and everything that has to do with it.
[dragonfly.git] / sys / platform / vkernel / i386 / trap.c
CommitLineData
6f7b98e0
MD
1/*-
2 * Copyright (C) 1994, David Greenman
3 * Copyright (c) 1990, 1993
4 * The Regents of the University of California. All rights reserved.
5 *
6 * This code is derived from software contributed to Berkeley by
7 * the University of Utah, and William Jolitz.
8 *
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
11 * are met:
12 * 1. Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in the
16 * documentation and/or other materials provided with the distribution.
17 * 3. All advertising materials mentioning features or use of this software
18 * must display the following acknowledgement:
19 * This product includes software developed by the University of
20 * California, Berkeley and its contributors.
21 * 4. Neither the name of the University nor the names of its contributors
22 * may be used to endorse or promote products derived from this software
23 * without specific prior written permission.
24 *
25 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
26 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
27 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
28 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
29 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
30 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
31 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
32 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
33 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
34 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35 * SUCH DAMAGE.
36 *
37 * from: @(#)trap.c 7.4 (Berkeley) 5/13/91
38 * $FreeBSD: src/sys/i386/i386/trap.c,v 1.147.2.11 2003/02/27 19:09:59 luoqi Exp $
6f7b98e0
MD
39 */
40
41/*
42 * 386 Trap and System call handling
43 */
44
45#include "use_isa.h"
46#include "use_npx.h"
47
48#include "opt_ddb.h"
49#include "opt_ktrace.h"
50
51#include <sys/param.h>
52#include <sys/systm.h>
53#include <sys/proc.h>
54#include <sys/pioctl.h>
55#include <sys/kernel.h>
56#include <sys/resourcevar.h>
57#include <sys/signalvar.h>
b1b4e5a6 58#include <sys/signal2.h>
6f7b98e0
MD
59#include <sys/syscall.h>
60#include <sys/sysctl.h>
61#include <sys/sysent.h>
62#include <sys/uio.h>
63#include <sys/vmmeter.h>
64#include <sys/malloc.h>
65#ifdef KTRACE
66#include <sys/ktrace.h>
67#endif
80bbc910 68#include <sys/ktr.h>
6f7b98e0
MD
69#include <sys/vkernel.h>
70#include <sys/sysproto.h>
71#include <sys/sysunion.h>
135d7199 72#include <sys/vmspace.h>
6f7b98e0
MD
73
74#include <vm/vm.h>
75#include <vm/vm_param.h>
76#include <sys/lock.h>
77#include <vm/pmap.h>
78#include <vm/vm_kern.h>
79#include <vm/vm_map.h>
80#include <vm/vm_page.h>
81#include <vm/vm_extern.h>
82
83#include <machine/cpu.h>
84#include <machine/md_var.h>
85#include <machine/pcb.h>
86#include <machine/smp.h>
87#include <machine/tss.h>
88#include <machine/globaldata.h>
89
90#include <machine/vm86.h>
91
92#include <ddb/ddb.h>
684a93c4 93
6f7b98e0
MD
94#include <sys/msgport2.h>
95#include <sys/thread2.h>
684a93c4 96#include <sys/mplock2.h>
6f7b98e0 97
6f7b98e0
MD
98#define MAKEMPSAFE(have_mplock) \
99 if (have_mplock == 0) { \
100 get_mplock(); \
101 have_mplock = 1; \
102 }
103
6f7b98e0
MD
104int (*pmath_emulate) (struct trapframe *);
105
6f7b98e0 106static int trap_pfault (struct trapframe *, int, vm_offset_t);
6092278a 107static void trap_fatal (struct trapframe *, int, vm_offset_t);
6f7b98e0
MD
108void dblfault_handler (void);
109
110#if 0
111extern inthand_t IDTVEC(syscall);
112#endif
113
114#define MAX_TRAP_MSG 28
115static char *trap_msg[] = {
116 "", /* 0 unused */
117 "privileged instruction fault", /* 1 T_PRIVINFLT */
118 "", /* 2 unused */
119 "breakpoint instruction fault", /* 3 T_BPTFLT */
120 "", /* 4 unused */
121 "", /* 5 unused */
122 "arithmetic trap", /* 6 T_ARITHTRAP */
123 "system forced exception", /* 7 T_ASTFLT */
124 "", /* 8 unused */
125 "general protection fault", /* 9 T_PROTFLT */
126 "trace trap", /* 10 T_TRCTRAP */
127 "", /* 11 unused */
128 "page fault", /* 12 T_PAGEFLT */
129 "", /* 13 unused */
130 "alignment fault", /* 14 T_ALIGNFLT */
131 "", /* 15 unused */
132 "", /* 16 unused */
133 "", /* 17 unused */
134 "integer divide fault", /* 18 T_DIVIDE */
135 "non-maskable interrupt trap", /* 19 T_NMI */
136 "overflow trap", /* 20 T_OFLOW */
137 "FPU bounds check fault", /* 21 T_BOUND */
138 "FPU device not available", /* 22 T_DNA */
139 "double fault", /* 23 T_DOUBLEFLT */
140 "FPU operand fetch fault", /* 24 T_FPOPFLT */
141 "invalid TSS fault", /* 25 T_TSSFLT */
142 "segment not present fault", /* 26 T_SEGNPFLT */
143 "stack fault", /* 27 T_STKFLT */
144 "machine check trap", /* 28 T_MCHK */
145};
146
147#ifdef DDB
148static int ddb_on_nmi = 1;
149SYSCTL_INT(_machdep, OID_AUTO, ddb_on_nmi, CTLFLAG_RW,
150 &ddb_on_nmi, 0, "Go to DDB on NMI");
151#endif
152static int panic_on_nmi = 1;
153SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RW,
154 &panic_on_nmi, 0, "Panic on NMI");
155static int fast_release;
156SYSCTL_INT(_machdep, OID_AUTO, fast_release, CTLFLAG_RW,
157 &fast_release, 0, "Passive Release was optimal");
158static int slow_release;
159SYSCTL_INT(_machdep, OID_AUTO, slow_release, CTLFLAG_RW,
160 &slow_release, 0, "Passive Release was nonoptimal");
6f7b98e0
MD
161
162MALLOC_DEFINE(M_SYSMSG, "sysmsg", "sysmsg structure");
163extern int max_sysmsg;
164
165/*
bab69519
JM
166 * Passively intercepts the thread switch function to increase the thread
167 * priority from a user priority to a kernel priority, reducing
6f7b98e0 168 * syscall and trap overhead for the case where no switch occurs.
7bd858e5 169 *
d86a23e0 170 * Synchronizes td_ucred with p_ucred. This is used by system calls,
7bd858e5
MD
171 * signal handling, faults, AST traps, and anything else that enters the
172 * kernel from userland and provides the kernel with a stable read-only
173 * copy of the process ucred.
6f7b98e0 174 */
6f7b98e0 175static __inline void
d86a23e0 176userenter(struct thread *curtd, struct proc *curp)
6f7b98e0 177{
7bd858e5
MD
178 struct ucred *ocred;
179 struct ucred *ncred;
180
3824f392 181 curtd->td_release = lwkt_passive_release;
7bd858e5 182
d86a23e0
MD
183 if (curtd->td_ucred != curp->p_ucred) {
184 ncred = crhold(curp->p_ucred);
185 ocred = curtd->td_ucred;
186 curtd->td_ucred = ncred;
7bd858e5
MD
187 if (ocred)
188 crfree(ocred);
189 }
bab69519 190
6f7b98e0
MD
191}
192
193/*
7adb15b6 194 * Handle signals, profiling, and other AST's and/or tasks that
6f7b98e0
MD
195 * must be completed before we can return to or try to return to userland.
196 *
197 * Note that td_sticks is a 64 bit quantity, but there's no point doing 64
198 * arithmatic on the delta calculation so the absolute tick values are
199 * truncated to an integer.
200 */
201static void
202userret(struct lwp *lp, struct trapframe *frame, int sticks)
203{
204 struct proc *p = lp->lwp_proc;
205 int sig;
206
207 /*
208 * Charge system time if profiling. Note: times are in microseconds.
209 * This may do a copyout and block, so do it first even though it
210 * means some system time will be charged as user time.
211 */
4643740a 212 if (p->p_flags & P_PROFIL) {
6f7b98e0
MD
213 addupc_task(p, frame->tf_eip,
214 (u_int)((int)lp->lwp_thread->td_sticks - sticks));
215 }
216
217recheck:
218 /*
62ae46c9
MD
219 * Specific on-return-to-usermode checks (LWP_MP_WEXIT,
220 * LWP_MP_VNLRU, etc).
70d3d461 221 */
62ae46c9
MD
222 if (lp->lwp_mpflags & LWP_MP_URETMASK)
223 lwpuserret(lp);
70d3d461
SS
224
225 /*
6f7b98e0
MD
226 * Block here if we are in a stopped state.
227 */
164b8401 228 if (p->p_stat == SSTOP) {
4643740a 229 lwkt_gettoken(&p->p_token);
9a379a4a 230 tstop();
4643740a 231 lwkt_reltoken(&p->p_token);
6f7b98e0
MD
232 goto recheck;
233 }
234
235 /*
898e34b3
MD
236 * Post any pending upcalls. If running a virtual kernel be sure
237 * to restore the virtual kernel's vmspace before posting the upcall.
6f7b98e0 238 */
7adb15b6 239 if (p->p_flags & (P_SIGVTALRM | P_SIGPROF)) {
fc509460 240 lwkt_gettoken(&p->p_token);
4643740a
MD
241 if (p->p_flags & P_SIGVTALRM) {
242 p->p_flags &= ~P_SIGVTALRM;
898e34b3
MD
243 ksignal(p, SIGVTALRM);
244 }
4643740a
MD
245 if (p->p_flags & P_SIGPROF) {
246 p->p_flags &= ~P_SIGPROF;
898e34b3
MD
247 ksignal(p, SIGPROF);
248 }
fc509460 249 lwkt_reltoken(&p->p_token);
6f7b98e0
MD
250 goto recheck;
251 }
252
253 /*
254 * Post any pending signals
2883d2d8
MD
255 *
256 * WARNING! postsig() can exit and not return.
6f7b98e0 257 */
e473f776 258 if ((sig = CURSIG_TRACE(lp)) != 0) {
fc509460 259 lwkt_gettoken(&p->p_token);
6f7b98e0 260 postsig(sig);
fc509460 261 lwkt_reltoken(&p->p_token);
6f7b98e0
MD
262 goto recheck;
263 }
264
265 /*
266 * block here if we are swapped out, but still process signals
267 * (such as SIGKILL). proc0 (the swapin scheduler) is already
268 * aware of our situation, we do not have to wake it up.
269 */
4643740a 270 if (p->p_flags & P_SWAPPEDOUT) {
616516c8 271 lwkt_gettoken(&p->p_token);
6f7b98e0 272 get_mplock();
4643740a 273 p->p_flags |= P_SWAPWAIT;
6f7b98e0 274 swapin_request();
4643740a 275 if (p->p_flags & P_SWAPWAIT)
6f7b98e0 276 tsleep(p, PCATCH, "SWOUT", 0);
4643740a 277 p->p_flags &= ~P_SWAPWAIT;
6f7b98e0 278 rel_mplock();
616516c8 279 lwkt_reltoken(&p->p_token);
6f7b98e0
MD
280 goto recheck;
281 }
ae7cb1b5
NT
282
283 /*
6562e2d8
MD
284 * In a multi-threaded program it is possible for a thread to change
285 * signal state during a system call which temporarily changes the
286 * signal mask. In this case postsig() might not be run and we
287 * have to restore the mask ourselves.
ae7cb1b5 288 */
6562e2d8
MD
289 if (lp->lwp_flags & LWP_OLDMASK) {
290 lp->lwp_flags &= ~LWP_OLDMASK;
291 lp->lwp_sigmask = lp->lwp_oldsigmask;
292 goto recheck;
293 }
6f7b98e0
MD
294}
295
296/*
297 * Cleanup from userenter and any passive release that might have occured.
298 * We must reclaim the current-process designation before we can return
299 * to usermode. We also handle both LWKT and USER reschedule requests.
300 */
301static __inline void
302userexit(struct lwp *lp)
303{
304 struct thread *td = lp->lwp_thread;
f5df782c 305 /* globaldata_t gd = td->td_gd; */
6f7b98e0 306
6f7b98e0 307 /*
b9eb1c19
MD
308 * Handle stop requests at kernel priority. Any requests queued
309 * after this loop will generate another AST.
6f7b98e0 310 */
b9eb1c19 311 while (lp->lwp_proc->p_stat == SSTOP) {
4643740a 312 lwkt_gettoken(&lp->lwp_proc->p_token);
b9eb1c19 313 tstop();
4643740a 314 lwkt_reltoken(&lp->lwp_proc->p_token);
fede0c7f
NT
315 }
316
6f7b98e0 317 /*
b9eb1c19
MD
318 * Become the current user scheduled process if we aren't already,
319 * and deal with reschedule requests and other factors.
320 */
321 lp->lwp_proc->p_usched->acquire_curproc(lp);
322 /* WARNING: we may have migrated cpu's */
323 /* gd = td->td_gd; */
e3e6be1f
MD
324
325 /*
326 * Reduce our priority in preparation for a return to userland. If
327 * our passive release function was still in place, our priority was
328 * never raised and does not need to be reduced.
329 */
330 lwkt_passive_recover(td);
b9eb1c19 331}
80bbc910
SS
332
333#if !defined(KTR_KERNENTRY)
334#define KTR_KERNENTRY KTR_ALL
335#endif
336KTR_INFO_MASTER(kernentry);
cacf1050
VS
337KTR_INFO(KTR_KERNENTRY, kernentry, trap, 0,
338 "TRAP(pid %d, tid %d, trapno %d, eva %lu)",
339 pid_t pid, lwpid_t tid, register_t trapno, vm_offset_t eva);
340KTR_INFO(KTR_KERNENTRY, kernentry, trap_ret, 0, "TRAP_RET(pid %d, tid %d)",
341 pid_t pid, lwpid_t tid);
342KTR_INFO(KTR_KERNENTRY, kernentry, syscall, 0, "SYSC(pid %d, tid %d, nr %d)",
343 pid_t pid, lwpid_t tid, register_t trapno);
344KTR_INFO(KTR_KERNENTRY, kernentry, syscall_ret, 0, "SYSRET(pid %d, tid %d, err %d)",
345 pid_t pid, lwpid_t tid, int err);
346KTR_INFO(KTR_KERNENTRY, kernentry, fork_ret, 0, "FORKRET(pid %d, tid %d)",
347 pid_t pid, lwpid_t tid);
80bbc910 348
6f7b98e0
MD
349/*
350 * Exception, fault, and trap interface to the kernel.
351 * This common code is called from assembly language IDT gate entry
352 * routines that prepare a suitable stack frame, and restore this
353 * frame after the exception has been processed.
354 *
355 * This function is also called from doreti in an interlock to handle ASTs.
356 * For example: hardwareint->INTROUTINE->(set ast)->doreti->trap
357 *
358 * NOTE! We have to retrieve the fault address prior to obtaining the
359 * MP lock because get_mplock() may switch out. YYY cr2 really ought
360 * to be retrieved by the assembly code, not here.
361 *
362 * XXX gd_trap_nesting_level currently prevents lwkt_switch() from panicing
363 * if an attempt is made to switch from a fast interrupt or IPI. This is
364 * necessary to properly take fatal kernel traps on SMP machines if
365 * get_mplock() has to block.
366 */
367
368void
6092278a 369user_trap(struct trapframe *frame)
6f7b98e0
MD
370{
371 struct globaldata *gd = mycpu;
372 struct thread *td = gd->gd_curthread;
373 struct lwp *lp = td->td_lwp;
374 struct proc *p;
375 int sticks = 0;
376 int i = 0, ucode = 0, type, code;
6f7b98e0 377 int have_mplock = 0;
6f7b98e0 378#ifdef INVARIANTS
f9235b6d 379 int crit_count = td->td_critcount;
3933a3ab 380 lwkt_tokref_t curstop = td->td_toks_stop;
6f7b98e0
MD
381#endif
382 vm_offset_t eva;
383
384 p = td->td_proc;
135d7199
MD
385
386 /*
387 * This is a bad kludge to avoid changing the various trapframe
388 * structures. Because we are enabled as a virtual kernel,
389 * the original tf_err field will be passed to us shifted 16
390 * over in the tf_trapno field for T_PAGEFLT.
391 */
4e7c41c5 392 if (frame->tf_trapno == T_PAGEFLT)
135d7199 393 eva = frame->tf_err;
4e7c41c5 394 else
135d7199 395 eva = 0;
4e7c41c5
MD
396#if 0
397 kprintf("USER_TRAP AT %08x xflags %d trapno %d eva %08x\n",
398 frame->tf_eip, frame->tf_xflags, frame->tf_trapno, eva);
399#endif
135d7199
MD
400
401 /*
402 * Everything coming from user mode runs through user_trap,
403 * including system calls.
404 */
405 if (frame->tf_trapno == T_SYSCALL80) {
406 syscall2(frame);
407 return;
408 }
409
80bbc910
SS
410 KTR_LOG(kernentry_trap, lp->lwp_proc->p_pid, lp->lwp_tid,
411 frame->tf_trapno, eva);
412
6f7b98e0
MD
413#ifdef DDB
414 if (db_active) {
6092278a 415 eva = (frame->tf_trapno == T_PAGEFLT ? rcr2() : 0);
6f7b98e0
MD
416 ++gd->gd_trap_nesting_level;
417 MAKEMPSAFE(have_mplock);
6092278a 418 trap_fatal(frame, TRUE, eva);
6f7b98e0
MD
419 --gd->gd_trap_nesting_level;
420 goto out2;
421 }
422#endif
423
6f7b98e0
MD
424#if defined(I586_CPU) && !defined(NO_F00F_HACK)
425restart:
426#endif
6092278a
MD
427 type = frame->tf_trapno;
428 code = frame->tf_err;
6f7b98e0 429
d86a23e0 430 userenter(td, p);
6f7b98e0 431
6092278a
MD
432 sticks = (int)td->td_sticks;
433 lp->lwp_md.md_regs = frame;
6f7b98e0 434
6092278a
MD
435 switch (type) {
436 case T_PRIVINFLT: /* privileged instruction fault */
6092278a 437 i = SIGILL;
bab69519 438 ucode = ILL_PRVOPC;
6092278a 439 break;
6f7b98e0 440
6092278a
MD
441 case T_BPTFLT: /* bpt instruction fault */
442 case T_TRCTRAP: /* trace trap */
443 frame->tf_eflags &= ~PSL_T;
444 i = SIGTRAP;
bab69519 445 ucode = (type == T_TRCTRAP ? TRAP_TRACE : TRAP_BRKPT);
6092278a 446 break;
6f7b98e0 447
6092278a
MD
448 case T_ARITHTRAP: /* arithmetic trap */
449 ucode = code;
450 i = SIGFPE;
451 break;
6f7b98e0 452
6092278a
MD
453 case T_ASTFLT: /* Allow process switch */
454 mycpu->gd_cnt.v_soft++;
455 if (mycpu->gd_reqflags & RQF_AST_OWEUPC) {
2a418930 456 atomic_clear_int(&mycpu->gd_reqflags,
6092278a
MD
457 RQF_AST_OWEUPC);
458 addupc_task(p, p->p_prof.pr_addr,
459 p->p_prof.pr_ticks);
460 }
461 goto out;
6f7b98e0 462
6092278a
MD
463 /*
464 * The following two traps can happen in
465 * vm86 mode, and, if so, we want to handle
466 * them specially.
467 */
468 case T_PROTFLT: /* general protection fault */
469 case T_STKFLT: /* stack fault */
6f7b98e0 470#if 0
6092278a
MD
471 if (frame->tf_eflags & PSL_VM) {
472 i = vm86_emulate((struct vm86frame *)frame);
473 if (i == 0)
474 goto out;
6f7b98e0 475 break;
6092278a
MD
476 }
477#endif
f2642d5a
AH
478 i = SIGBUS;
479 ucode = (type == T_PROTFLT) ? BUS_OBJERR : BUS_ADRERR;
480 break;
6092278a 481 case T_SEGNPFLT: /* segment not present fault */
f2642d5a
AH
482 i = SIGBUS;
483 ucode = BUS_ADRERR;
484 break;
6092278a
MD
485 case T_TSSFLT: /* invalid TSS fault */
486 case T_DOUBLEFLT: /* double fault */
487 default:
6092278a 488 i = SIGBUS;
bab69519 489 ucode = BUS_OBJERR;
6092278a
MD
490 break;
491
492 case T_PAGEFLT: /* page fault */
493 MAKEMPSAFE(have_mplock);
494 i = trap_pfault(frame, TRUE, eva);
495 if (i == -1)
496 goto out;
6f7b98e0 497#if defined(I586_CPU) && !defined(NO_F00F_HACK)
6092278a
MD
498 if (i == -2)
499 goto restart;
6f7b98e0 500#endif
6092278a
MD
501 if (i == 0)
502 goto out;
6f7b98e0 503
f2642d5a
AH
504 if (i == SIGSEGV)
505 ucode = SEGV_MAPERR;
bab69519
JM
506 else {
507 i = SIGSEGV;
508 ucode = SEGV_ACCERR;
509 }
6092278a 510 break;
6f7b98e0 511
6092278a
MD
512 case T_DIVIDE: /* integer divide fault */
513 ucode = FPE_INTDIV;
514 i = SIGFPE;
515 break;
6f7b98e0
MD
516
517#if NISA > 0
6092278a
MD
518 case T_NMI:
519 MAKEMPSAFE(have_mplock);
520 /* machine/parity/power fail/"kitchen sink" faults */
521 if (isa_nmi(code) == 0) {
6f7b98e0 522#ifdef DDB
6092278a
MD
523 /*
524 * NMI can be hooked up to a pushbutton
525 * for debugging.
526 */
527 if (ddb_on_nmi) {
528 kprintf ("NMI ... going to debugger\n");
529 kdb_trap (type, 0, frame);
530 }
6f7b98e0 531#endif /* DDB */
6092278a
MD
532 goto out2;
533 } else if (panic_on_nmi)
534 panic("NMI indicates hardware failure");
535 break;
6f7b98e0
MD
536#endif /* NISA > 0 */
537
6092278a
MD
538 case T_OFLOW: /* integer overflow fault */
539 ucode = FPE_INTOVF;
540 i = SIGFPE;
541 break;
6f7b98e0 542
6092278a
MD
543 case T_BOUND: /* bounds check fault */
544 ucode = FPE_FLTSUB;
545 i = SIGFPE;
546 break;
6f7b98e0 547
6092278a 548 case T_DNA:
431d0fef
MD
549 /*
550 * Virtual kernel intercept - pass the DNA exception
551 * to the (emulated) virtual kernel if it asked to handle
552 * it. This occurs when the virtual kernel is holding
553 * onto the FP context for a different emulated
554 * process then the one currently running.
555 *
556 * We must still call npxdna() since we may have
557 * saved FP state that the (emulated) virtual kernel
558 * needs to hand over to a different emulated process.
559 */
39005e16 560 if (lp->lwp_vkernel && lp->lwp_vkernel->ve &&
431d0fef
MD
561 (td->td_pcb->pcb_flags & FP_VIRTFP)
562 ) {
563 npxdna(frame);
564 break;
565 }
bab69519 566
6f7b98e0 567#if NNPX > 0
6092278a
MD
568 /*
569 * The kernel may have switched out the FP unit's
570 * state, causing the user process to take a fault
571 * when it tries to use the FP unit. Restore the
572 * state here
573 */
431d0fef 574 if (npxdna(frame))
6092278a 575 goto out;
6f7b98e0 576#endif
6092278a 577 if (!pmath_emulate) {
6f7b98e0 578 i = SIGFPE;
6092278a 579 ucode = FPE_FPU_NP_TRAP;
6f7b98e0
MD
580 break;
581 }
6092278a
MD
582 i = (*pmath_emulate)(frame);
583 if (i == 0) {
584 if (!(frame->tf_eflags & PSL_T))
6f7b98e0 585 goto out2;
6092278a
MD
586 frame->tf_eflags &= ~PSL_T;
587 i = SIGTRAP;
6f7b98e0 588 }
6092278a
MD
589 /* else ucode = emulator_only_knows() XXX */
590 break;
6f7b98e0 591
6092278a 592 case T_FPOPFLT: /* FPU operand fetch fault */
f2642d5a 593 ucode = ILL_COPROC;
6092278a
MD
594 i = SIGILL;
595 break;
596
597 case T_XMMFLT: /* SIMD floating-point exception */
598 ucode = 0; /* XXX */
599 i = SIGFPE;
600 break;
6f7b98e0
MD
601 }
602
603 /*
604 * Virtual kernel intercept - if the fault is directly related to a
605 * VM context managed by a virtual kernel then let the virtual kernel
606 * handle it.
607 */
39005e16 608 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
287ebb09 609 vkernel_trap(lp, frame);
6f7b98e0
MD
610 goto out;
611 }
612
613 /*
614 * Translate fault for emulators (e.g. Linux)
615 */
616 if (*p->p_sysent->sv_transtrap)
617 i = (*p->p_sysent->sv_transtrap)(i, type);
618
619 MAKEMPSAFE(have_mplock);
08f2f1bb 620 trapsignal(lp, i, ucode);
6f7b98e0
MD
621
622#ifdef DEBUG
623 if (type <= MAX_TRAP_MSG) {
624 uprintf("fatal process exception: %s",
625 trap_msg[type]);
626 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
627 uprintf(", fault VA = 0x%lx", (u_long)eva);
628 uprintf("\n");
629 }
630#endif
631
632out:
6092278a 633 userret(lp, frame, sticks);
6f7b98e0
MD
634 userexit(lp);
635out2: ;
6f7b98e0
MD
636 if (have_mplock)
637 rel_mplock();
80bbc910 638 KTR_LOG(kernentry_trap_ret, lp->lwp_proc->p_pid, lp->lwp_tid);
6f7b98e0 639#ifdef INVARIANTS
f9235b6d 640 KASSERT(crit_count == td->td_critcount,
3933a3ab 641 ("trap: critical section count mismatch! %d/%d",
f9235b6d 642 crit_count, td->td_pri));
3933a3ab
MD
643 KASSERT(curstop == td->td_toks_stop,
644 ("trap: extra tokens held after trap! %zd/%zd",
645 curstop - &td->td_toks_base,
646 td->td_toks_stop - &td->td_toks_base));
6f7b98e0
MD
647#endif
648}
649
6092278a
MD
650void
651kern_trap(struct trapframe *frame)
6f7b98e0 652{
6092278a
MD
653 struct globaldata *gd = mycpu;
654 struct thread *td = gd->gd_curthread;
08f2f1bb 655 struct lwp *lp;
6092278a
MD
656 struct proc *p;
657 int i = 0, ucode = 0, type, code;
6092278a 658 int have_mplock = 0;
6092278a 659#ifdef INVARIANTS
f9235b6d 660 int crit_count = td->td_critcount;
3933a3ab 661 lwkt_tokref_t curstop = td->td_toks_stop;
6092278a
MD
662#endif
663 vm_offset_t eva;
6f7b98e0 664
08f2f1bb 665 lp = td->td_lwp;
6092278a 666 p = td->td_proc;
6f7b98e0 667
4e7c41c5 668 if (frame->tf_trapno == T_PAGEFLT)
6092278a 669 eva = frame->tf_err;
4e7c41c5 670 else
135d7199 671 eva = 0;
135d7199
MD
672
673#ifdef DDB
674 if (db_active) {
675 ++gd->gd_trap_nesting_level;
676 MAKEMPSAFE(have_mplock);
677 trap_fatal(frame, FALSE, eva);
678 --gd->gd_trap_nesting_level;
679 goto out2;
680 }
681#endif
6092278a
MD
682 type = frame->tf_trapno;
683 code = frame->tf_err;
684
685#if 0
686kernel_trap:
687#endif
688 /* kernel trap */
689
690 switch (type) {
691 case T_PAGEFLT: /* page fault */
692 MAKEMPSAFE(have_mplock);
693 trap_pfault(frame, FALSE, eva);
694 goto out2;
6f7b98e0 695
6092278a
MD
696 case T_DNA:
697#if NNPX > 0
6f7b98e0 698 /*
6092278a
MD
699 * The kernel may be using npx for copying or other
700 * purposes.
6f7b98e0 701 */
431d0fef
MD
702 panic("kernel NPX should not happen");
703 if (npxdna(frame))
6092278a
MD
704 goto out2;
705#endif
706 break;
6f7b98e0 707
6092278a
MD
708 case T_PROTFLT: /* general protection fault */
709 case T_SEGNPFLT: /* segment not present fault */
6f7b98e0 710 /*
6092278a
MD
711 * Invalid segment selectors and out of bounds
712 * %eip's and %esp's can be set up in user mode.
713 * This causes a fault in kernel mode when the
714 * kernel tries to return to user mode. We want
715 * to get this fault so that we can fix the
716 * problem here and not have to check all the
717 * selectors and pointers when the user changes
718 * them.
6f7b98e0 719 */
6092278a
MD
720 if (mycpu->gd_intr_nesting_level == 0) {
721 if (td->td_pcb->pcb_onfault) {
722 frame->tf_eip =
723 (register_t)td->td_pcb->pcb_onfault;
724 goto out2;
725 }
6f7b98e0 726 }
6092278a 727 break;
6f7b98e0 728
6092278a 729 case T_TSSFLT:
6f7b98e0 730 /*
6092278a
MD
731 * PSL_NT can be set in user mode and isn't cleared
732 * automatically when the kernel is entered. This
733 * causes a TSS fault when the kernel attempts to
734 * `iret' because the TSS link is uninitialized. We
735 * want to get this fault so that we can fix the
736 * problem here and not every time the kernel is
737 * entered.
6f7b98e0 738 */
6092278a
MD
739 if (frame->tf_eflags & PSL_NT) {
740 frame->tf_eflags &= ~PSL_NT;
741 goto out2;
742 }
743 break;
6f7b98e0 744
6092278a
MD
745 case T_TRCTRAP: /* trace trap */
746#if 0
747 if (frame->tf_eip == (int)IDTVEC(syscall)) {
748 /*
749 * We've just entered system mode via the
750 * syscall lcall. Continue single stepping
751 * silently until the syscall handler has
752 * saved the flags.
753 */
754 goto out2;
755 }
756 if (frame->tf_eip == (int)IDTVEC(syscall) + 1) {
757 /*
758 * The syscall handler has now saved the
759 * flags. Stop single stepping it.
760 */
761 frame->tf_eflags &= ~PSL_T;
762 goto out2;
763 }
764#endif
765#if 0
6f7b98e0 766 /*
6092278a
MD
767 * Ignore debug register trace traps due to
768 * accesses in the user's address space, which
769 * can happen under several conditions such as
770 * if a user sets a watchpoint on a buffer and
771 * then passes that buffer to a system call.
772 * We still want to get TRCTRAPS for addresses
773 * in kernel space because that is useful when
774 * debugging the kernel.
6f7b98e0 775 */
6092278a
MD
776 if (user_dbreg_trap()) {
777 /*
778 * Reset breakpoint bits because the
779 * processor doesn't
780 */
781 load_dr6(rdr6() & 0xfffffff0);
782 goto out2;
6f7b98e0 783 }
6092278a
MD
784#endif
785 /*
bab69519 786 * FALLTHROUGH (TRCTRAP kernel mode, kernel address)
6092278a
MD
787 */
788 case T_BPTFLT:
789 /*
790 * If DDB is enabled, let it handle the debugger trap.
791 * Otherwise, debugger traps "can't happen".
792 */
793#ifdef DDB
794 MAKEMPSAFE(have_mplock);
795 if (kdb_trap (type, 0, frame))
796 goto out2;
797#endif
798 break;
58e0cae5
MD
799 case T_DIVIDE:
800 MAKEMPSAFE(have_mplock);
801 trap_fatal(frame, FALSE, eva);
802 goto out2;
6092278a
MD
803 case T_NMI:
804 MAKEMPSAFE(have_mplock);
805 trap_fatal(frame, FALSE, eva);
806 goto out2;
2612135b
TS
807 case T_SYSCALL80:
808 /*
809 * Ignore this trap generated from a spurious SIGTRAP.
810 *
811 * single stepping in / syscalls leads to spurious / SIGTRAP
812 * so ignore
813 *
814 * Haiku (c) 2007 Simon 'corecode' Schubert
815 */
816 goto out2;
6f7b98e0
MD
817 }
818
6092278a
MD
819 /*
820 * Translate fault for emulators (e.g. Linux)
821 */
822 if (*p->p_sysent->sv_transtrap)
823 i = (*p->p_sysent->sv_transtrap)(i, type);
6f7b98e0 824
6092278a 825 MAKEMPSAFE(have_mplock);
08f2f1bb 826 trapsignal(lp, i, ucode);
6092278a
MD
827
828#ifdef DEBUG
829 if (type <= MAX_TRAP_MSG) {
830 uprintf("fatal process exception: %s",
831 trap_msg[type]);
832 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
833 uprintf(", fault VA = 0x%lx", (u_long)eva);
834 uprintf("\n");
835 }
836#endif
837
838out2:
839 ;
6092278a
MD
840 if (have_mplock)
841 rel_mplock();
6092278a 842#ifdef INVARIANTS
f9235b6d 843 KASSERT(crit_count == td->td_critcount,
3933a3ab 844 ("trap: critical section count mismatch! %d/%d",
f9235b6d 845 crit_count, td->td_pri));
3933a3ab
MD
846 KASSERT(curstop == td->td_toks_stop,
847 ("trap: extra tokens held after trap! %zd/%zd",
848 curstop - &td->td_toks_base,
849 td->td_toks_stop - &td->td_toks_base));
6f7b98e0 850#endif
6092278a 851}
6f7b98e0
MD
852
853int
854trap_pfault(struct trapframe *frame, int usermode, vm_offset_t eva)
855{
856 vm_offset_t va;
857 struct vmspace *vm = NULL;
858 vm_map_t map = 0;
859 int rv = 0;
1b9d3514 860 int fault_flags;
6f7b98e0
MD
861 vm_prot_t ftype;
862 thread_t td = curthread;
287ebb09 863 struct lwp *lp = td->td_lwp;
6f7b98e0
MD
864
865 va = trunc_page(eva);
6092278a 866 if (usermode == FALSE) {
6f7b98e0 867 /*
6092278a 868 * This is a fault on kernel virtual memory.
6f7b98e0 869 */
6f7b98e0
MD
870 map = &kernel_map;
871 } else {
872 /*
873 * This is a fault on non-kernel virtual memory.
874 * vm is initialized above to NULL. If curproc is NULL
875 * or curproc->p_vmspace is NULL the fault is fatal.
876 */
287ebb09
MD
877 if (lp != NULL)
878 vm = lp->lwp_vmspace;
6f7b98e0
MD
879
880 if (vm == NULL)
881 goto nogo;
882
883 map = &vm->vm_map;
884 }
885
4e7c41c5 886 if (frame->tf_xflags & PGEX_W)
0035dca9 887 ftype = VM_PROT_READ | VM_PROT_WRITE;
6f7b98e0
MD
888 else
889 ftype = VM_PROT_READ;
890
891 if (map != &kernel_map) {
892 /*
893 * Keep swapout from messing with us during this
894 * critical time.
895 */
287ebb09 896 PHOLD(lp->lwp_proc);
6f7b98e0
MD
897
898 /*
1b9d3514
MD
899 * Issue fault
900 */
901 fault_flags = 0;
902 if (usermode)
903 fault_flags |= VM_FAULT_BURST;
904 if (ftype & VM_PROT_WRITE)
905 fault_flags |= VM_FAULT_DIRTY;
906 else
907 fault_flags |= VM_FAULT_NORMAL;
908 rv = vm_fault(map, va, ftype, fault_flags);
287ebb09 909 PRELE(lp->lwp_proc);
6f7b98e0
MD
910 } else {
911 /*
912 * Don't have to worry about process locking or stacks in the kernel.
913 */
914 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
915 }
6f7b98e0
MD
916 if (rv == KERN_SUCCESS)
917 return (0);
918nogo:
919 if (!usermode) {
920 if (td->td_gd->gd_intr_nesting_level == 0 &&
921 td->td_pcb->pcb_onfault) {
922 frame->tf_eip = (register_t)td->td_pcb->pcb_onfault;
923 return (0);
924 }
6092278a 925 trap_fatal(frame, usermode, eva);
6f7b98e0
MD
926 return (-1);
927 }
6f7b98e0
MD
928 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
929}
930
931static void
6092278a 932trap_fatal(struct trapframe *frame, int usermode, vm_offset_t eva)
6f7b98e0
MD
933{
934 int code, type, ss, esp;
935
4e7c41c5 936 code = frame->tf_xflags;
6f7b98e0
MD
937 type = frame->tf_trapno;
938
9097944f 939 if (type <= MAX_TRAP_MSG) {
6f7b98e0
MD
940 kprintf("\n\nFatal trap %d: %s while in %s mode\n",
941 type, trap_msg[type],
9097944f
MD
942 (usermode ? "user" : "kernel"));
943 }
6a8aa90e 944 /* two separate prints in case of a trap on an unmapped page */
39005e16 945 kprintf("cpuid = %d\n", mycpu->gd_cpuid);
6f7b98e0 946 if (type == T_PAGEFLT) {
d557216f 947 kprintf("fault virtual address = %p\n", (void *)eva);
6f7b98e0 948 kprintf("fault code = %s %s, %s\n",
6092278a 949 usermode ? "user" : "supervisor",
6f7b98e0
MD
950 code & PGEX_W ? "write" : "read",
951 code & PGEX_P ? "protection violation" : "page not present");
952 }
953 kprintf("instruction pointer = 0x%x:0x%x\n",
954 frame->tf_cs & 0xffff, frame->tf_eip);
9097944f 955 if (usermode) {
6f7b98e0
MD
956 ss = frame->tf_ss & 0xffff;
957 esp = frame->tf_esp;
958 } else {
959 ss = GSEL(GDATA_SEL, SEL_KPL);
960 esp = (int)&frame->tf_esp;
961 }
962 kprintf("stack pointer = 0x%x:0x%x\n", ss, esp);
963 kprintf("frame pointer = 0x%x:0x%x\n", ss, frame->tf_ebp);
964 kprintf("processor eflags = ");
965 if (frame->tf_eflags & PSL_T)
966 kprintf("trace trap, ");
967 if (frame->tf_eflags & PSL_I)
968 kprintf("interrupt enabled, ");
969 if (frame->tf_eflags & PSL_NT)
970 kprintf("nested task, ");
971 if (frame->tf_eflags & PSL_RF)
972 kprintf("resume, ");
973#if 0
974 if (frame->tf_eflags & PSL_VM)
975 kprintf("vm86, ");
976#endif
977 kprintf("IOPL = %d\n", (frame->tf_eflags & PSL_IOPL) >> 12);
978 kprintf("current process = ");
979 if (curproc) {
980 kprintf("%lu (%s)\n",
981 (u_long)curproc->p_pid, curproc->p_comm ?
982 curproc->p_comm : "");
983 } else {
984 kprintf("Idle\n");
985 }
986 kprintf("current thread = pri %d ", curthread->td_pri);
f9235b6d 987 if (curthread->td_critcount)
6f7b98e0
MD
988 kprintf("(CRIT)");
989 kprintf("\n");
6f7b98e0
MD
990/**
991 * XXX FIXME:
992 * we probably SHOULD have stopped the other CPUs before now!
993 * another CPU COULD have been touching cpl at this moment...
994 */
995 kprintf(" <- SMP: XXX");
6f7b98e0
MD
996 kprintf("\n");
997
998#ifdef KDB
999 if (kdb_trap(&psl))
1000 return;
1001#endif
1002#ifdef DDB
1003 if ((debugger_on_panic || db_active) && kdb_trap(type, code, frame))
1004 return;
1005#endif
1006 kprintf("trap number = %d\n", type);
1007 if (type <= MAX_TRAP_MSG)
1008 panic("%s", trap_msg[type]);
1009 else
1010 panic("unknown/reserved trap");
1011}
1012
1013/*
1014 * Double fault handler. Called when a fault occurs while writing
1015 * a frame for a trap/exception onto the stack. This usually occurs
1016 * when the stack overflows (such is the case with infinite recursion,
1017 * for example).
1018 *
1019 * XXX Note that the current PTD gets replaced by IdlePTD when the
1020 * task switch occurs. This means that the stack that was active at
1021 * the time of the double fault is not available at <kstack> unless
1022 * the machine was idle when the double fault occurred. The downside
1023 * of this is that "trace <ebp>" in ddb won't work.
1024 */
1025void
1026dblfault_handler(void)
1027{
1028 struct mdglobaldata *gd = mdcpu;
1029
1030 kprintf("\nFatal double fault:\n");
1031 kprintf("eip = 0x%x\n", gd->gd_common_tss.tss_eip);
1032 kprintf("esp = 0x%x\n", gd->gd_common_tss.tss_esp);
1033 kprintf("ebp = 0x%x\n", gd->gd_common_tss.tss_ebp);
6a8aa90e 1034 /* two separate prints in case of a trap on an unmapped page */
39005e16 1035 kprintf("cpuid = %d\n", mycpu->gd_cpuid);
6f7b98e0
MD
1036 panic("double fault");
1037}
1038
1039/*
7bd858e5 1040 * syscall2 - MP aware system call request C handler
6f7b98e0 1041 *
7bd858e5
MD
1042 * A system call is essentially treated as a trap except that the
1043 * MP lock is not held on entry or return. We are responsible for
1044 * obtaining the MP lock if necessary and for handling ASTs
1045 * (e.g. a task switch) prior to return.
6f7b98e0 1046 *
7bd858e5 1047 * MPSAFE
6f7b98e0 1048 */
6f7b98e0 1049void
6092278a 1050syscall2(struct trapframe *frame)
6f7b98e0
MD
1051{
1052 struct thread *td = curthread;
1053 struct proc *p = td->td_proc;
1054 struct lwp *lp = td->td_lwp;
1055 caddr_t params;
1056 struct sysent *callp;
1057 register_t orig_tf_eflags;
1058 int sticks;
1059 int error;
1060 int narg;
1061#ifdef INVARIANTS
f9235b6d 1062 int crit_count = td->td_critcount;
6f7b98e0 1063#endif
6f7b98e0 1064 int have_mplock = 0;
6f7b98e0
MD
1065 u_int code;
1066 union sysunion args;
1067
80bbc910
SS
1068 KTR_LOG(kernentry_syscall, lp->lwp_proc->p_pid, lp->lwp_tid,
1069 frame->tf_eax);
1070
d86a23e0 1071 userenter(td, p); /* lazy raise our priority */
6f7b98e0
MD
1072
1073 /*
1074 * Misc
1075 */
1076 sticks = (int)td->td_sticks;
6092278a 1077 orig_tf_eflags = frame->tf_eflags;
6f7b98e0
MD
1078
1079 /*
1080 * Virtual kernel intercept - if a VM context managed by a virtual
1081 * kernel issues a system call the virtual kernel handles it, not us.
1082 * Restore the virtual kernel context and return from its system
1083 * call. The current frame is copied out to the virtual kernel.
1084 */
39005e16 1085 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
bb47c072 1086 vkernel_trap(lp, frame);
6f7b98e0
MD
1087 error = EJUSTRETURN;
1088 goto out;
1089 }
1090
1091 /*
1092 * Get the system call parameters and account for time
1093 */
6092278a
MD
1094 lp->lwp_md.md_regs = frame;
1095 params = (caddr_t)frame->tf_esp + sizeof(int);
1096 code = frame->tf_eax;
6f7b98e0
MD
1097
1098 if (p->p_sysent->sv_prepsyscall) {
1099 (*p->p_sysent->sv_prepsyscall)(
6092278a 1100 frame, (int *)(&args.nosys.sysmsg + 1),
6f7b98e0
MD
1101 &code, &params);
1102 } else {
1103 /*
1104 * Need to check if this is a 32 bit or 64 bit syscall.
1105 * fuword is MP aware.
1106 */
1107 if (code == SYS_syscall) {
1108 /*
1109 * Code is first argument, followed by actual args.
1110 */
1111 code = fuword(params);
1112 params += sizeof(int);
1113 } else if (code == SYS___syscall) {
1114 /*
1115 * Like syscall, but code is a quad, so as to maintain
1116 * quad alignment for the rest of the arguments.
1117 */
1118 code = fuword(params);
1119 params += sizeof(quad_t);
1120 }
1121 }
1122
1123 code &= p->p_sysent->sv_mask;
1124 if (code >= p->p_sysent->sv_size)
1125 callp = &p->p_sysent->sv_table[0];
1126 else
1127 callp = &p->p_sysent->sv_table[code];
1128
1129 narg = callp->sy_narg & SYF_ARGMASK;
1130
1131 /*
1132 * copyin is MP aware, but the tracing code is not
1133 */
1134 if (narg && params) {
1135 error = copyin(params, (caddr_t)(&args.nosys.sysmsg + 1),
1136 narg * sizeof(register_t));
1137 if (error) {
1138#ifdef KTRACE
1139 if (KTRPOINT(td, KTR_SYSCALL)) {
1140 MAKEMPSAFE(have_mplock);
1141
9fb04d14 1142 ktrsyscall(lp, code, narg,
6f7b98e0
MD
1143 (void *)(&args.nosys.sysmsg + 1));
1144 }
1145#endif
1146 goto bad;
1147 }
1148 }
1149
1150#ifdef KTRACE
1151 if (KTRPOINT(td, KTR_SYSCALL)) {
1152 MAKEMPSAFE(have_mplock);
9fb04d14 1153 ktrsyscall(lp, code, narg, (void *)(&args.nosys.sysmsg + 1));
6f7b98e0
MD
1154 }
1155#endif
1156
1157 /*
1158 * For traditional syscall code edx is left untouched when 32 bit
1159 * results are returned. Since edx is loaded from fds[1] when the
1160 * system call returns we pre-set it here.
1161 */
1162 args.sysmsg_fds[0] = 0;
6092278a 1163 args.sysmsg_fds[1] = frame->tf_edx;
6f7b98e0
MD
1164
1165 /*
1166 * The syscall might manipulate the trap frame. If it does it
1167 * will probably return EJUSTRETURN.
1168 */
6092278a 1169 args.sysmsg_frame = frame;
6f7b98e0
MD
1170
1171 STOPEVENT(p, S_SCE, narg); /* MP aware */
1172
6f7b98e0 1173 /*
3919ced0
MD
1174 * NOTE: All system calls run MPSAFE now. The system call itself
1175 * is responsible for getting the MP lock.
6f7b98e0 1176 */
6f7b98e0
MD
1177 error = (*callp->sy_call)(&args);
1178
aaf8b91f
MD
1179#if 0
1180 kprintf("system call %d returned %d\n", code, error);
1181#endif
1182
6f7b98e0
MD
1183out:
1184 /*
1185 * MP SAFE (we may or may not have the MP lock at this point)
1186 */
1187 switch (error) {
1188 case 0:
1189 /*
1190 * Reinitialize proc pointer `p' as it may be different
1191 * if this is a child returning from fork syscall.
1192 */
1193 p = curproc;
1194 lp = curthread->td_lwp;
6092278a
MD
1195 frame->tf_eax = args.sysmsg_fds[0];
1196 frame->tf_edx = args.sysmsg_fds[1];
1197 frame->tf_eflags &= ~PSL_C;
6f7b98e0
MD
1198 break;
1199 case ERESTART:
1200 /*
1201 * Reconstruct pc, assuming lcall $X,y is 7 bytes,
1202 * int 0x80 is 2 bytes. We saved this in tf_err.
1203 */
6092278a 1204 frame->tf_eip -= frame->tf_err;
6f7b98e0
MD
1205 break;
1206 case EJUSTRETURN:
1207 break;
1208 case EASYNC:
1209 panic("Unexpected EASYNC return value (for now)");
1210 default:
1211bad:
1212 if (p->p_sysent->sv_errsize) {
1213 if (error >= p->p_sysent->sv_errsize)
1214 error = -1; /* XXX */
1215 else
1216 error = p->p_sysent->sv_errtbl[error];
1217 }
6092278a
MD
1218 frame->tf_eax = error;
1219 frame->tf_eflags |= PSL_C;
6f7b98e0
MD
1220 break;
1221 }
1222
1223 /*
1224 * Traced syscall. trapsignal() is not MP aware.
1225 */
1226 if ((orig_tf_eflags & PSL_T) /*&& !(orig_tf_eflags & PSL_VM)*/) {
1227 MAKEMPSAFE(have_mplock);
6092278a 1228 frame->tf_eflags &= ~PSL_T;
f2642d5a 1229 trapsignal(lp, SIGTRAP, TRAP_TRACE);
6f7b98e0
MD
1230 }
1231
1232 /*
1233 * Handle reschedule and other end-of-syscall issues
1234 */
6092278a 1235 userret(lp, frame, sticks);
6f7b98e0
MD
1236
1237#ifdef KTRACE
1238 if (KTRPOINT(td, KTR_SYSRET)) {
1239 MAKEMPSAFE(have_mplock);
9fb04d14 1240 ktrsysret(lp, code, error, args.sysmsg_result);
6f7b98e0
MD
1241 }
1242#endif
1243
1244 /*
1245 * This works because errno is findable through the
1246 * register set. If we ever support an emulation where this
1247 * is not the case, this code will need to be revisited.
1248 */
1249 STOPEVENT(p, S_SCX, code);
1250
1251 userexit(lp);
6f7b98e0
MD
1252 /*
1253 * Release the MP lock if we had to get it
1254 */
6f7b98e0
MD
1255 if (have_mplock)
1256 rel_mplock();
80bbc910 1257 KTR_LOG(kernentry_syscall_ret, lp->lwp_proc->p_pid, lp->lwp_tid, error);
6f7b98e0 1258#ifdef INVARIANTS
f9235b6d 1259 KASSERT(crit_count == td->td_critcount,
6f7b98e0 1260 ("syscall: critical section count mismatch! %d/%d",
f9235b6d 1261 crit_count, td->td_pri));
3933a3ab
MD
1262 KASSERT(&td->td_toks_base == td->td_toks_stop,
1263 ("syscall: extra tokens held after trap! %zd",
1264 td->td_toks_stop - &td->td_toks_base));
6f7b98e0
MD
1265#endif
1266}
1267
2b0bd8aa
MD
1268/*
1269 * NOTE: mplock not held at any point
1270 */
91bd9c1e
SS
1271void
1272fork_return(struct lwp *lp, struct trapframe *frame)
1273{
1274 frame->tf_eax = 0; /* Child returns zero */
1275 frame->tf_eflags &= ~PSL_C; /* success */
1276 frame->tf_edx = 1;
1277
1278 generic_lwp_return(lp, frame);
80bbc910 1279 KTR_LOG(kernentry_fork_ret, lp->lwp_proc->p_pid, lp->lwp_tid);
91bd9c1e
SS
1280}
1281
6f7b98e0
MD
1282/*
1283 * Simplified back end of syscall(), used when returning from fork()
2b0bd8aa
MD
1284 * directly into user mode.
1285 *
1286 * This code will return back into the fork trampoline code which then
1287 * runs doreti.
1288 *
1289 * NOTE: The mplock is not held at any point.
6f7b98e0
MD
1290 */
1291void
91bd9c1e 1292generic_lwp_return(struct lwp *lp, struct trapframe *frame)
6f7b98e0
MD
1293{
1294 struct proc *p = lp->lwp_proc;
1295
6f7b98e0
MD
1296 /*
1297 * Newly forked processes are given a kernel priority. We have to
1298 * adjust the priority to a normal user priority and fake entry
1299 * into the kernel (call userenter()) to install a passive release
1300 * function just in case userret() decides to stop the process. This
1301 * can occur when ^Z races a fork. If we do not install the passive
1302 * release function the current process designation will not be
1303 * released when the thread goes to sleep.
1304 */
1305 lwkt_setpri_self(TDPRI_USER_NORM);
d86a23e0 1306 userenter(lp->lwp_thread, p);
c7eb0589 1307 userret(lp, frame, 0);
6f7b98e0
MD
1308#ifdef KTRACE
1309 if (KTRPOINT(lp->lwp_thread, KTR_SYSRET))
9fb04d14 1310 ktrsysret(lp, SYS_fork, 0, 0);
6f7b98e0 1311#endif
4643740a 1312 lp->lwp_flags |= LWP_PASSIVE_ACQ;
6f7b98e0 1313 userexit(lp);
4643740a 1314 lp->lwp_flags &= ~LWP_PASSIVE_ACQ;
6f7b98e0 1315}
9ad680a3 1316
4e7c41c5
MD
1317/*
1318 * doreti has turned into this. The frame is directly on the stack. We
1319 * pull everything else we need (fpu and tls context) from the current
1320 * thread.
1321 *
1322 * Note on fpu interactions: In a virtual kernel, the fpu context for
1323 * an emulated user mode process is not shared with the virtual kernel's
1324 * fpu context, so we only have to 'stack' fpu contexts within the virtual
1325 * kernel itself, and not even then since the signal() contexts that we care
1326 * about save and restore the FPU state (I think anyhow).
1327 *
1328 * vmspace_ctl() returns an error only if it had problems instaling the
1329 * context we supplied or problems copying data to/from our VM space.
1330 */
9ad680a3 1331void
c7eb0589 1332go_user(struct intrframe *frame)
9ad680a3 1333{
c7eb0589 1334 struct trapframe *tf = (void *)&frame->if_gs;
4e7c41c5
MD
1335 int r;
1336
1fdc022f
MD
1337 /*
1338 * Interrupts may be disabled on entry, make sure all signals
1339 * can be received before beginning our loop.
1340 */
1341 sigsetmask(0);
1342
1343 /*
1344 * Switch to the current simulated user process, then call
1345 * user_trap() when we break out of it (usually due to a signal).
1346 */
135d7199 1347 for (;;) {
431d0fef
MD
1348 /*
1349 * Tell the real kernel whether it is ok to use the FP
1350 * unit or not.
eaaa6d60
MD
1351 *
1352 * The critical section is required to prevent an interrupt
1353 * from causing a preemptive task switch and changing
1354 * the FP state.
431d0fef 1355 */
eaaa6d60 1356 crit_enter();
431d0fef 1357 if (mdcpu->gd_npxthread == curthread) {
4b486183 1358 tf->tf_xflags &= ~PGEX_FPFAULT;
431d0fef 1359 } else {
4b486183 1360 tf->tf_xflags |= PGEX_FPFAULT;
431d0fef
MD
1361 }
1362
1363 /*
7c1212ec
MD
1364 * Run emulated user process context. This call interlocks
1365 * with new mailbox signals.
4b486183
MD
1366 *
1367 * Set PGEX_U unconditionally, indicating a user frame (the
1368 * bit is normally set only by T_PAGEFLT).
7c1212ec
MD
1369 */
1370 r = vmspace_ctl(&curproc->p_vmspace->vm_pmap, VMSPACE_CTL_RUN,
1371 tf, &curthread->td_savevext);
eaaa6d60 1372 crit_exit();
c7eb0589 1373 frame->if_xflags |= PGEX_U;
4b486183
MD
1374#if 0
1375 kprintf("GO USER %d trap %d EVA %08x EIP %08x ESP %08x XFLAGS %02x/%02x\n",
1376 r, tf->tf_trapno, tf->tf_err, tf->tf_eip, tf->tf_esp,
c7eb0589 1377 tf->tf_xflags, frame->if_xflags);
4b486183 1378#endif
7c1212ec 1379 if (r < 0) {
24eb47e0 1380 if (errno != EINTR)
bb47c072 1381 panic("vmspace_ctl failed error %d", errno);
aaf8b91f 1382 } else {
7c1212ec
MD
1383 if (tf->tf_trapno) {
1384 user_trap(tf);
7c1212ec 1385 }
aaf8b91f 1386 }
6179d4f3
MD
1387 if (mycpu->gd_reqflags & RQF_AST_MASK) {
1388 tf->tf_trapno = T_ASTFLT;
1389 user_trap(tf);
1390 }
1391 tf->tf_trapno = 0;
135d7199 1392 }
9ad680a3 1393}
135d7199 1394
431d0fef 1395/*
4b486183 1396 * If PGEX_FPFAULT is set then set FP_VIRTFP in the PCB to force a T_DNA
431d0fef
MD
1397 * fault (which is then passed back to the virtual kernel) if an attempt is
1398 * made to use the FP unit.
bab69519 1399 *
431d0fef
MD
1400 * XXX this is a fairly big hack.
1401 */
1402void
1403set_vkernel_fp(struct trapframe *frame)
1404{
1405 struct thread *td = curthread;
1406
4b486183 1407 if (frame->tf_xflags & PGEX_FPFAULT) {
431d0fef
MD
1408 td->td_pcb->pcb_flags |= FP_VIRTFP;
1409 if (mdcpu->gd_npxthread == td)
1410 npxexit();
1411 } else {
1412 td->td_pcb->pcb_flags &= ~FP_VIRTFP;
1413 }
1414}
1415
bb47c072
MD
1416/*
1417 * Called from vkernel_trap() to fixup the vkernel's syscall
1418 * frame for vmspace_ctl() return.
1419 */
1420void
1421cpu_vkernel_trap(struct trapframe *frame, int error)
1422{
1423 frame->tf_eax = error;
1424 if (error)
1425 frame->tf_eflags |= PSL_C;
1426 else
1427 frame->tf_eflags &= ~PSL_C;
1428}