kernel - Fix signal delivery races
[dragonfly.git] / sys / platform / vkernel / i386 / trap.c
CommitLineData
6f7b98e0
MD
1/*-
2 * Copyright (C) 1994, David Greenman
3 * Copyright (c) 1990, 1993
4 * The Regents of the University of California. All rights reserved.
5 *
6 * This code is derived from software contributed to Berkeley by
7 * the University of Utah, and William Jolitz.
8 *
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
11 * are met:
12 * 1. Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in the
16 * documentation and/or other materials provided with the distribution.
17 * 3. All advertising materials mentioning features or use of this software
18 * must display the following acknowledgement:
19 * This product includes software developed by the University of
20 * California, Berkeley and its contributors.
21 * 4. Neither the name of the University nor the names of its contributors
22 * may be used to endorse or promote products derived from this software
23 * without specific prior written permission.
24 *
25 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
26 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
27 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
28 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
29 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
30 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
31 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
32 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
33 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
34 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35 * SUCH DAMAGE.
36 *
37 * from: @(#)trap.c 7.4 (Berkeley) 5/13/91
38 * $FreeBSD: src/sys/i386/i386/trap.c,v 1.147.2.11 2003/02/27 19:09:59 luoqi Exp $
c730be20 39 * $DragonFly: src/sys/platform/vkernel/i386/trap.c,v 1.35 2008/09/09 04:06:19 dillon Exp $
6f7b98e0
MD
40 */
41
42/*
43 * 386 Trap and System call handling
44 */
45
46#include "use_isa.h"
47#include "use_npx.h"
48
49#include "opt_ddb.h"
50#include "opt_ktrace.h"
51
52#include <sys/param.h>
53#include <sys/systm.h>
54#include <sys/proc.h>
55#include <sys/pioctl.h>
56#include <sys/kernel.h>
57#include <sys/resourcevar.h>
58#include <sys/signalvar.h>
b1b4e5a6 59#include <sys/signal2.h>
6f7b98e0
MD
60#include <sys/syscall.h>
61#include <sys/sysctl.h>
62#include <sys/sysent.h>
63#include <sys/uio.h>
64#include <sys/vmmeter.h>
65#include <sys/malloc.h>
66#ifdef KTRACE
67#include <sys/ktrace.h>
68#endif
80bbc910 69#include <sys/ktr.h>
6f7b98e0
MD
70#include <sys/upcall.h>
71#include <sys/vkernel.h>
72#include <sys/sysproto.h>
73#include <sys/sysunion.h>
135d7199 74#include <sys/vmspace.h>
6f7b98e0
MD
75
76#include <vm/vm.h>
77#include <vm/vm_param.h>
78#include <sys/lock.h>
79#include <vm/pmap.h>
80#include <vm/vm_kern.h>
81#include <vm/vm_map.h>
82#include <vm/vm_page.h>
83#include <vm/vm_extern.h>
84
85#include <machine/cpu.h>
86#include <machine/md_var.h>
87#include <machine/pcb.h>
88#include <machine/smp.h>
89#include <machine/tss.h>
90#include <machine/globaldata.h>
91
92#include <machine/vm86.h>
93
94#include <ddb/ddb.h>
684a93c4 95
6f7b98e0
MD
96#include <sys/msgport2.h>
97#include <sys/thread2.h>
684a93c4 98#include <sys/mplock2.h>
6f7b98e0
MD
99
100#ifdef SMP
101
102#define MAKEMPSAFE(have_mplock) \
103 if (have_mplock == 0) { \
104 get_mplock(); \
105 have_mplock = 1; \
106 }
107
108#else
109
110#define MAKEMPSAFE(have_mplock)
111
112#endif
113
114int (*pmath_emulate) (struct trapframe *);
115
6f7b98e0 116static int trap_pfault (struct trapframe *, int, vm_offset_t);
6092278a 117static void trap_fatal (struct trapframe *, int, vm_offset_t);
6f7b98e0
MD
118void dblfault_handler (void);
119
120#if 0
121extern inthand_t IDTVEC(syscall);
122#endif
123
124#define MAX_TRAP_MSG 28
125static char *trap_msg[] = {
126 "", /* 0 unused */
127 "privileged instruction fault", /* 1 T_PRIVINFLT */
128 "", /* 2 unused */
129 "breakpoint instruction fault", /* 3 T_BPTFLT */
130 "", /* 4 unused */
131 "", /* 5 unused */
132 "arithmetic trap", /* 6 T_ARITHTRAP */
133 "system forced exception", /* 7 T_ASTFLT */
134 "", /* 8 unused */
135 "general protection fault", /* 9 T_PROTFLT */
136 "trace trap", /* 10 T_TRCTRAP */
137 "", /* 11 unused */
138 "page fault", /* 12 T_PAGEFLT */
139 "", /* 13 unused */
140 "alignment fault", /* 14 T_ALIGNFLT */
141 "", /* 15 unused */
142 "", /* 16 unused */
143 "", /* 17 unused */
144 "integer divide fault", /* 18 T_DIVIDE */
145 "non-maskable interrupt trap", /* 19 T_NMI */
146 "overflow trap", /* 20 T_OFLOW */
147 "FPU bounds check fault", /* 21 T_BOUND */
148 "FPU device not available", /* 22 T_DNA */
149 "double fault", /* 23 T_DOUBLEFLT */
150 "FPU operand fetch fault", /* 24 T_FPOPFLT */
151 "invalid TSS fault", /* 25 T_TSSFLT */
152 "segment not present fault", /* 26 T_SEGNPFLT */
153 "stack fault", /* 27 T_STKFLT */
154 "machine check trap", /* 28 T_MCHK */
155};
156
157#ifdef DDB
158static int ddb_on_nmi = 1;
159SYSCTL_INT(_machdep, OID_AUTO, ddb_on_nmi, CTLFLAG_RW,
160 &ddb_on_nmi, 0, "Go to DDB on NMI");
161#endif
162static int panic_on_nmi = 1;
163SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RW,
164 &panic_on_nmi, 0, "Panic on NMI");
165static int fast_release;
166SYSCTL_INT(_machdep, OID_AUTO, fast_release, CTLFLAG_RW,
167 &fast_release, 0, "Passive Release was optimal");
168static int slow_release;
169SYSCTL_INT(_machdep, OID_AUTO, slow_release, CTLFLAG_RW,
170 &slow_release, 0, "Passive Release was nonoptimal");
6f7b98e0
MD
171
172MALLOC_DEFINE(M_SYSMSG, "sysmsg", "sysmsg structure");
173extern int max_sysmsg;
174
6f7b98e0 175/*
7bd858e5 176 * Passively intercepts the thread switch function to increase
6f7b98e0
MD
177 * the thread priority from a user priority to a kernel priority, reducing
178 * syscall and trap overhead for the case where no switch occurs.
7bd858e5 179 *
d86a23e0 180 * Synchronizes td_ucred with p_ucred. This is used by system calls,
7bd858e5
MD
181 * signal handling, faults, AST traps, and anything else that enters the
182 * kernel from userland and provides the kernel with a stable read-only
183 * copy of the process ucred.
6f7b98e0 184 */
6f7b98e0 185static __inline void
d86a23e0 186userenter(struct thread *curtd, struct proc *curp)
6f7b98e0 187{
7bd858e5
MD
188 struct ucred *ocred;
189 struct ucred *ncred;
190
3824f392 191 curtd->td_release = lwkt_passive_release;
7bd858e5 192
d86a23e0
MD
193 if (curtd->td_ucred != curp->p_ucred) {
194 ncred = crhold(curp->p_ucred);
195 ocred = curtd->td_ucred;
196 curtd->td_ucred = ncred;
7bd858e5
MD
197 if (ocred)
198 crfree(ocred);
199 }
6f7b98e0
MD
200}
201
202/*
203 * Handle signals, upcalls, profiling, and other AST's and/or tasks that
204 * must be completed before we can return to or try to return to userland.
205 *
206 * Note that td_sticks is a 64 bit quantity, but there's no point doing 64
207 * arithmatic on the delta calculation so the absolute tick values are
208 * truncated to an integer.
209 */
210static void
211userret(struct lwp *lp, struct trapframe *frame, int sticks)
212{
213 struct proc *p = lp->lwp_proc;
214 int sig;
215
216 /*
217 * Charge system time if profiling. Note: times are in microseconds.
218 * This may do a copyout and block, so do it first even though it
219 * means some system time will be charged as user time.
220 */
221 if (p->p_flag & P_PROFIL) {
222 addupc_task(p, frame->tf_eip,
223 (u_int)((int)lp->lwp_thread->td_sticks - sticks));
224 }
225
226recheck:
70d3d461
SS
227 /*
228 * If the jungle wants us dead, so be it.
229 */
e207d1fd 230 if (lp->lwp_flag & LWP_WEXIT) {
b5c4d81f 231 lwkt_gettoken(&p->p_token);
e3161323 232 lwp_exit(0);
b5c4d81f 233 lwkt_reltoken(&p->p_token); /* NOT REACHED */
e207d1fd 234 }
70d3d461 235
6f7b98e0
MD
236 /*
237 * Block here if we are in a stopped state.
238 */
164b8401 239 if (p->p_stat == SSTOP) {
6f7b98e0 240 get_mplock();
9a379a4a 241 tstop();
6f7b98e0
MD
242 rel_mplock();
243 goto recheck;
244 }
245
246 /*
247 * Post any pending upcalls
248 */
249 if (p->p_flag & P_UPCALLPEND) {
025d695a 250 lwkt_gettoken(&p->p_token);
4e7c41c5 251 p->p_flag &= ~P_UPCALLPEND;
6f7b98e0 252 postupcall(lp);
025d695a 253 lwkt_reltoken(&p->p_token);
6f7b98e0
MD
254 goto recheck;
255 }
256
257 /*
258 * Post any pending signals
2883d2d8
MD
259 *
260 * WARNING! postsig() can exit and not return.
6f7b98e0 261 */
e473f776 262 if ((sig = CURSIG_TRACE(lp)) != 0) {
025d695a 263 lwkt_gettoken(&p->p_token);
6f7b98e0 264 postsig(sig);
025d695a 265 lwkt_reltoken(&p->p_token);
6f7b98e0
MD
266 goto recheck;
267 }
268
269 /*
270 * block here if we are swapped out, but still process signals
271 * (such as SIGKILL). proc0 (the swapin scheduler) is already
272 * aware of our situation, we do not have to wake it up.
273 */
274 if (p->p_flag & P_SWAPPEDOUT) {
275 get_mplock();
276 p->p_flag |= P_SWAPWAIT;
277 swapin_request();
278 if (p->p_flag & P_SWAPWAIT)
279 tsleep(p, PCATCH, "SWOUT", 0);
280 p->p_flag &= ~P_SWAPWAIT;
281 rel_mplock();
282 goto recheck;
283 }
ae7cb1b5
NT
284
285 /*
286 * Make sure postsig() handled request to restore old signal mask after
287 * running signal handler.
288 */
289 KKASSERT((lp->lwp_flag & LWP_OLDMASK) == 0);
6f7b98e0
MD
290}
291
292/*
293 * Cleanup from userenter and any passive release that might have occured.
294 * We must reclaim the current-process designation before we can return
295 * to usermode. We also handle both LWKT and USER reschedule requests.
296 */
297static __inline void
298userexit(struct lwp *lp)
299{
300 struct thread *td = lp->lwp_thread;
f5df782c 301 /* globaldata_t gd = td->td_gd; */
6f7b98e0 302
6f7b98e0 303 /*
b9eb1c19
MD
304 * Handle stop requests at kernel priority. Any requests queued
305 * after this loop will generate another AST.
6f7b98e0 306 */
b9eb1c19
MD
307 while (lp->lwp_proc->p_stat == SSTOP) {
308 get_mplock();
309 tstop();
310 rel_mplock();
fede0c7f
NT
311 }
312
6f7b98e0
MD
313 /*
314 * Reduce our priority in preparation for a return to userland. If
315 * our passive release function was still in place, our priority was
316 * never raised and does not need to be reduced.
317 */
3824f392 318 lwkt_passive_recover(td);
6f7b98e0 319
b9eb1c19
MD
320 /*
321 * Become the current user scheduled process if we aren't already,
322 * and deal with reschedule requests and other factors.
323 */
324 lp->lwp_proc->p_usched->acquire_curproc(lp);
325 /* WARNING: we may have migrated cpu's */
326 /* gd = td->td_gd; */
327}
80bbc910
SS
328
329#if !defined(KTR_KERNENTRY)
330#define KTR_KERNENTRY KTR_ALL
331#endif
332KTR_INFO_MASTER(kernentry);
333KTR_INFO(KTR_KERNENTRY, kernentry, trap, 0, "pid=%d, tid=%d, trapno=%d, eva=%p",
334 sizeof(int) + sizeof(int) + sizeof(int) + sizeof(vm_offset_t));
335KTR_INFO(KTR_KERNENTRY, kernentry, trap_ret, 0, "pid=%d, tid=%d",
336 sizeof(int) + sizeof(int));
337KTR_INFO(KTR_KERNENTRY, kernentry, syscall, 0, "pid=%d, tid=%d, call=%d",
338 sizeof(int) + sizeof(int) + sizeof(int));
339KTR_INFO(KTR_KERNENTRY, kernentry, syscall_ret, 0, "pid=%d, tid=%d, err=%d",
340 sizeof(int) + sizeof(int) + sizeof(int));
341KTR_INFO(KTR_KERNENTRY, kernentry, fork_ret, 0, "pid=%d, tid=%d",
342 sizeof(int) + sizeof(int));
343
6f7b98e0
MD
344/*
345 * Exception, fault, and trap interface to the kernel.
346 * This common code is called from assembly language IDT gate entry
347 * routines that prepare a suitable stack frame, and restore this
348 * frame after the exception has been processed.
349 *
350 * This function is also called from doreti in an interlock to handle ASTs.
351 * For example: hardwareint->INTROUTINE->(set ast)->doreti->trap
352 *
353 * NOTE! We have to retrieve the fault address prior to obtaining the
354 * MP lock because get_mplock() may switch out. YYY cr2 really ought
355 * to be retrieved by the assembly code, not here.
356 *
357 * XXX gd_trap_nesting_level currently prevents lwkt_switch() from panicing
358 * if an attempt is made to switch from a fast interrupt or IPI. This is
359 * necessary to properly take fatal kernel traps on SMP machines if
360 * get_mplock() has to block.
361 */
362
363void
6092278a 364user_trap(struct trapframe *frame)
6f7b98e0
MD
365{
366 struct globaldata *gd = mycpu;
367 struct thread *td = gd->gd_curthread;
368 struct lwp *lp = td->td_lwp;
369 struct proc *p;
370 int sticks = 0;
371 int i = 0, ucode = 0, type, code;
372#ifdef SMP
373 int have_mplock = 0;
374#endif
375#ifdef INVARIANTS
f9235b6d 376 int crit_count = td->td_critcount;
3933a3ab 377 lwkt_tokref_t curstop = td->td_toks_stop;
6f7b98e0
MD
378#endif
379 vm_offset_t eva;
380
381 p = td->td_proc;
135d7199
MD
382
383 /*
384 * This is a bad kludge to avoid changing the various trapframe
385 * structures. Because we are enabled as a virtual kernel,
386 * the original tf_err field will be passed to us shifted 16
387 * over in the tf_trapno field for T_PAGEFLT.
388 */
4e7c41c5 389 if (frame->tf_trapno == T_PAGEFLT)
135d7199 390 eva = frame->tf_err;
4e7c41c5 391 else
135d7199 392 eva = 0;
4e7c41c5
MD
393#if 0
394 kprintf("USER_TRAP AT %08x xflags %d trapno %d eva %08x\n",
395 frame->tf_eip, frame->tf_xflags, frame->tf_trapno, eva);
396#endif
135d7199
MD
397
398 /*
399 * Everything coming from user mode runs through user_trap,
400 * including system calls.
401 */
402 if (frame->tf_trapno == T_SYSCALL80) {
403 syscall2(frame);
404 return;
405 }
406
80bbc910
SS
407 KTR_LOG(kernentry_trap, lp->lwp_proc->p_pid, lp->lwp_tid,
408 frame->tf_trapno, eva);
409
6f7b98e0
MD
410#ifdef DDB
411 if (db_active) {
6092278a 412 eva = (frame->tf_trapno == T_PAGEFLT ? rcr2() : 0);
6f7b98e0
MD
413 ++gd->gd_trap_nesting_level;
414 MAKEMPSAFE(have_mplock);
6092278a 415 trap_fatal(frame, TRUE, eva);
6f7b98e0
MD
416 --gd->gd_trap_nesting_level;
417 goto out2;
418 }
419#endif
420
6f7b98e0
MD
421#if defined(I586_CPU) && !defined(NO_F00F_HACK)
422restart:
423#endif
6092278a
MD
424 type = frame->tf_trapno;
425 code = frame->tf_err;
6f7b98e0 426
d86a23e0 427 userenter(td, p);
6f7b98e0 428
6092278a
MD
429 sticks = (int)td->td_sticks;
430 lp->lwp_md.md_regs = frame;
6f7b98e0 431
6092278a
MD
432 switch (type) {
433 case T_PRIVINFLT: /* privileged instruction fault */
f2642d5a 434 ucode = ILL_PRVOPC;
6092278a
MD
435 i = SIGILL;
436 break;
6f7b98e0 437
6092278a
MD
438 case T_BPTFLT: /* bpt instruction fault */
439 case T_TRCTRAP: /* trace trap */
440 frame->tf_eflags &= ~PSL_T;
f2642d5a 441 ucode = TRAP_TRACE;
6092278a
MD
442 i = SIGTRAP;
443 break;
6f7b98e0 444
6092278a
MD
445 case T_ARITHTRAP: /* arithmetic trap */
446 ucode = code;
447 i = SIGFPE;
448 break;
6f7b98e0 449
6092278a
MD
450 case T_ASTFLT: /* Allow process switch */
451 mycpu->gd_cnt.v_soft++;
452 if (mycpu->gd_reqflags & RQF_AST_OWEUPC) {
2a418930 453 atomic_clear_int(&mycpu->gd_reqflags,
6092278a
MD
454 RQF_AST_OWEUPC);
455 addupc_task(p, p->p_prof.pr_addr,
456 p->p_prof.pr_ticks);
457 }
458 goto out;
6f7b98e0 459
6092278a
MD
460 /*
461 * The following two traps can happen in
462 * vm86 mode, and, if so, we want to handle
463 * them specially.
464 */
465 case T_PROTFLT: /* general protection fault */
466 case T_STKFLT: /* stack fault */
6f7b98e0 467#if 0
6092278a
MD
468 if (frame->tf_eflags & PSL_VM) {
469 i = vm86_emulate((struct vm86frame *)frame);
470 if (i == 0)
471 goto out;
6f7b98e0 472 break;
6092278a
MD
473 }
474#endif
f2642d5a
AH
475 i = SIGBUS;
476 ucode = (type == T_PROTFLT) ? BUS_OBJERR : BUS_ADRERR;
477 break;
6092278a 478 case T_SEGNPFLT: /* segment not present fault */
f2642d5a
AH
479 i = SIGBUS;
480 ucode = BUS_ADRERR;
481 break;
6092278a
MD
482 case T_TSSFLT: /* invalid TSS fault */
483 case T_DOUBLEFLT: /* double fault */
f2642d5a
AH
484 i = SIGBUS;
485 ucode = BUS_OBJERR;
6092278a 486 default:
f2642d5a
AH
487#if 0
488 ucode = code + BUS_SEGM_FAULT ; /* XXX: ???*/
489#endif
490 ucode = BUS_OBJERR;
6092278a
MD
491 i = SIGBUS;
492 break;
493
494 case T_PAGEFLT: /* page fault */
495 MAKEMPSAFE(have_mplock);
496 i = trap_pfault(frame, TRUE, eva);
497 if (i == -1)
498 goto out;
6f7b98e0 499#if defined(I586_CPU) && !defined(NO_F00F_HACK)
6092278a
MD
500 if (i == -2)
501 goto restart;
6f7b98e0 502#endif
6092278a
MD
503 if (i == 0)
504 goto out;
6f7b98e0 505
f2642d5a 506#if 0
6092278a 507 ucode = T_PAGEFLT;
f2642d5a
AH
508#endif
509 if (i == SIGSEGV)
510 ucode = SEGV_MAPERR;
511 else
512 ucode = BUS_ADRERR;
6092278a 513 break;
6f7b98e0 514
6092278a
MD
515 case T_DIVIDE: /* integer divide fault */
516 ucode = FPE_INTDIV;
517 i = SIGFPE;
518 break;
6f7b98e0
MD
519
520#if NISA > 0
6092278a
MD
521 case T_NMI:
522 MAKEMPSAFE(have_mplock);
523 /* machine/parity/power fail/"kitchen sink" faults */
524 if (isa_nmi(code) == 0) {
6f7b98e0 525#ifdef DDB
6092278a
MD
526 /*
527 * NMI can be hooked up to a pushbutton
528 * for debugging.
529 */
530 if (ddb_on_nmi) {
531 kprintf ("NMI ... going to debugger\n");
532 kdb_trap (type, 0, frame);
533 }
6f7b98e0 534#endif /* DDB */
6092278a
MD
535 goto out2;
536 } else if (panic_on_nmi)
537 panic("NMI indicates hardware failure");
538 break;
6f7b98e0
MD
539#endif /* NISA > 0 */
540
6092278a
MD
541 case T_OFLOW: /* integer overflow fault */
542 ucode = FPE_INTOVF;
543 i = SIGFPE;
544 break;
6f7b98e0 545
6092278a
MD
546 case T_BOUND: /* bounds check fault */
547 ucode = FPE_FLTSUB;
548 i = SIGFPE;
549 break;
6f7b98e0 550
6092278a 551 case T_DNA:
431d0fef
MD
552 /*
553 * Virtual kernel intercept - pass the DNA exception
554 * to the (emulated) virtual kernel if it asked to handle
555 * it. This occurs when the virtual kernel is holding
556 * onto the FP context for a different emulated
557 * process then the one currently running.
558 *
559 * We must still call npxdna() since we may have
560 * saved FP state that the (emulated) virtual kernel
561 * needs to hand over to a different emulated process.
562 */
39005e16 563 if (lp->lwp_vkernel && lp->lwp_vkernel->ve &&
431d0fef
MD
564 (td->td_pcb->pcb_flags & FP_VIRTFP)
565 ) {
566 npxdna(frame);
567 break;
568 }
6f7b98e0 569#if NNPX > 0
6092278a
MD
570 /*
571 * The kernel may have switched out the FP unit's
572 * state, causing the user process to take a fault
573 * when it tries to use the FP unit. Restore the
574 * state here
575 */
431d0fef 576 if (npxdna(frame))
6092278a 577 goto out;
6f7b98e0 578#endif
6092278a 579 if (!pmath_emulate) {
6f7b98e0 580 i = SIGFPE;
6092278a 581 ucode = FPE_FPU_NP_TRAP;
6f7b98e0
MD
582 break;
583 }
6092278a
MD
584 i = (*pmath_emulate)(frame);
585 if (i == 0) {
586 if (!(frame->tf_eflags & PSL_T))
6f7b98e0 587 goto out2;
6092278a
MD
588 frame->tf_eflags &= ~PSL_T;
589 i = SIGTRAP;
6f7b98e0 590 }
6092278a
MD
591 /* else ucode = emulator_only_knows() XXX */
592 break;
6f7b98e0 593
6092278a 594 case T_FPOPFLT: /* FPU operand fetch fault */
f2642d5a 595 ucode = ILL_COPROC;
6092278a
MD
596 i = SIGILL;
597 break;
598
599 case T_XMMFLT: /* SIMD floating-point exception */
600 ucode = 0; /* XXX */
601 i = SIGFPE;
602 break;
6f7b98e0
MD
603 }
604
605 /*
606 * Virtual kernel intercept - if the fault is directly related to a
607 * VM context managed by a virtual kernel then let the virtual kernel
608 * handle it.
609 */
39005e16 610 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
287ebb09 611 vkernel_trap(lp, frame);
6f7b98e0
MD
612 goto out;
613 }
614
615 /*
616 * Translate fault for emulators (e.g. Linux)
617 */
618 if (*p->p_sysent->sv_transtrap)
619 i = (*p->p_sysent->sv_transtrap)(i, type);
620
621 MAKEMPSAFE(have_mplock);
08f2f1bb 622 trapsignal(lp, i, ucode);
6f7b98e0
MD
623
624#ifdef DEBUG
625 if (type <= MAX_TRAP_MSG) {
626 uprintf("fatal process exception: %s",
627 trap_msg[type]);
628 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
629 uprintf(", fault VA = 0x%lx", (u_long)eva);
630 uprintf("\n");
631 }
632#endif
633
634out:
6092278a 635 userret(lp, frame, sticks);
6f7b98e0
MD
636 userexit(lp);
637out2: ;
638#ifdef SMP
639 if (have_mplock)
640 rel_mplock();
641#endif
80bbc910 642 KTR_LOG(kernentry_trap_ret, lp->lwp_proc->p_pid, lp->lwp_tid);
6f7b98e0 643#ifdef INVARIANTS
f9235b6d 644 KASSERT(crit_count == td->td_critcount,
3933a3ab 645 ("trap: critical section count mismatch! %d/%d",
f9235b6d 646 crit_count, td->td_pri));
3933a3ab
MD
647 KASSERT(curstop == td->td_toks_stop,
648 ("trap: extra tokens held after trap! %zd/%zd",
649 curstop - &td->td_toks_base,
650 td->td_toks_stop - &td->td_toks_base));
6f7b98e0
MD
651#endif
652}
653
6092278a
MD
654void
655kern_trap(struct trapframe *frame)
6f7b98e0 656{
6092278a
MD
657 struct globaldata *gd = mycpu;
658 struct thread *td = gd->gd_curthread;
08f2f1bb 659 struct lwp *lp;
6092278a
MD
660 struct proc *p;
661 int i = 0, ucode = 0, type, code;
662#ifdef SMP
663 int have_mplock = 0;
664#endif
665#ifdef INVARIANTS
f9235b6d 666 int crit_count = td->td_critcount;
3933a3ab 667 lwkt_tokref_t curstop = td->td_toks_stop;
6092278a
MD
668#endif
669 vm_offset_t eva;
6f7b98e0 670
08f2f1bb 671 lp = td->td_lwp;
6092278a 672 p = td->td_proc;
6f7b98e0 673
4e7c41c5 674 if (frame->tf_trapno == T_PAGEFLT)
6092278a 675 eva = frame->tf_err;
4e7c41c5 676 else
135d7199 677 eva = 0;
135d7199
MD
678
679#ifdef DDB
680 if (db_active) {
681 ++gd->gd_trap_nesting_level;
682 MAKEMPSAFE(have_mplock);
683 trap_fatal(frame, FALSE, eva);
684 --gd->gd_trap_nesting_level;
685 goto out2;
686 }
687#endif
6092278a
MD
688 type = frame->tf_trapno;
689 code = frame->tf_err;
690
691#if 0
692kernel_trap:
693#endif
694 /* kernel trap */
695
696 switch (type) {
697 case T_PAGEFLT: /* page fault */
698 MAKEMPSAFE(have_mplock);
699 trap_pfault(frame, FALSE, eva);
700 goto out2;
6f7b98e0 701
6092278a
MD
702 case T_DNA:
703#if NNPX > 0
6f7b98e0 704 /*
6092278a
MD
705 * The kernel may be using npx for copying or other
706 * purposes.
6f7b98e0 707 */
431d0fef
MD
708 panic("kernel NPX should not happen");
709 if (npxdna(frame))
6092278a
MD
710 goto out2;
711#endif
712 break;
6f7b98e0 713
6092278a
MD
714 case T_PROTFLT: /* general protection fault */
715 case T_SEGNPFLT: /* segment not present fault */
6f7b98e0 716 /*
6092278a
MD
717 * Invalid segment selectors and out of bounds
718 * %eip's and %esp's can be set up in user mode.
719 * This causes a fault in kernel mode when the
720 * kernel tries to return to user mode. We want
721 * to get this fault so that we can fix the
722 * problem here and not have to check all the
723 * selectors and pointers when the user changes
724 * them.
6f7b98e0 725 */
6092278a
MD
726 if (mycpu->gd_intr_nesting_level == 0) {
727 if (td->td_pcb->pcb_onfault) {
728 frame->tf_eip =
729 (register_t)td->td_pcb->pcb_onfault;
730 goto out2;
731 }
6f7b98e0 732 }
6092278a 733 break;
6f7b98e0 734
6092278a 735 case T_TSSFLT:
6f7b98e0 736 /*
6092278a
MD
737 * PSL_NT can be set in user mode and isn't cleared
738 * automatically when the kernel is entered. This
739 * causes a TSS fault when the kernel attempts to
740 * `iret' because the TSS link is uninitialized. We
741 * want to get this fault so that we can fix the
742 * problem here and not every time the kernel is
743 * entered.
6f7b98e0 744 */
6092278a
MD
745 if (frame->tf_eflags & PSL_NT) {
746 frame->tf_eflags &= ~PSL_NT;
747 goto out2;
748 }
749 break;
6f7b98e0 750
6092278a
MD
751 case T_TRCTRAP: /* trace trap */
752#if 0
753 if (frame->tf_eip == (int)IDTVEC(syscall)) {
754 /*
755 * We've just entered system mode via the
756 * syscall lcall. Continue single stepping
757 * silently until the syscall handler has
758 * saved the flags.
759 */
760 goto out2;
761 }
762 if (frame->tf_eip == (int)IDTVEC(syscall) + 1) {
763 /*
764 * The syscall handler has now saved the
765 * flags. Stop single stepping it.
766 */
767 frame->tf_eflags &= ~PSL_T;
768 goto out2;
769 }
770#endif
771#if 0
6f7b98e0 772 /*
6092278a
MD
773 * Ignore debug register trace traps due to
774 * accesses in the user's address space, which
775 * can happen under several conditions such as
776 * if a user sets a watchpoint on a buffer and
777 * then passes that buffer to a system call.
778 * We still want to get TRCTRAPS for addresses
779 * in kernel space because that is useful when
780 * debugging the kernel.
6f7b98e0 781 */
6092278a
MD
782 if (user_dbreg_trap()) {
783 /*
784 * Reset breakpoint bits because the
785 * processor doesn't
786 */
787 load_dr6(rdr6() & 0xfffffff0);
788 goto out2;
6f7b98e0 789 }
6092278a
MD
790#endif
791 /*
792 * Fall through (TRCTRAP kernel mode, kernel address)
793 */
794 case T_BPTFLT:
795 /*
796 * If DDB is enabled, let it handle the debugger trap.
797 * Otherwise, debugger traps "can't happen".
798 */
799#ifdef DDB
800 MAKEMPSAFE(have_mplock);
801 if (kdb_trap (type, 0, frame))
802 goto out2;
803#endif
804 break;
58e0cae5
MD
805 case T_DIVIDE:
806 MAKEMPSAFE(have_mplock);
807 trap_fatal(frame, FALSE, eva);
808 goto out2;
6092278a
MD
809 case T_NMI:
810 MAKEMPSAFE(have_mplock);
811 trap_fatal(frame, FALSE, eva);
812 goto out2;
2612135b
TS
813 case T_SYSCALL80:
814 /*
815 * Ignore this trap generated from a spurious SIGTRAP.
816 *
817 * single stepping in / syscalls leads to spurious / SIGTRAP
818 * so ignore
819 *
820 * Haiku (c) 2007 Simon 'corecode' Schubert
821 */
822 goto out2;
6f7b98e0
MD
823 }
824
6092278a
MD
825 /*
826 * Translate fault for emulators (e.g. Linux)
827 */
828 if (*p->p_sysent->sv_transtrap)
829 i = (*p->p_sysent->sv_transtrap)(i, type);
6f7b98e0 830
6092278a 831 MAKEMPSAFE(have_mplock);
08f2f1bb 832 trapsignal(lp, i, ucode);
6092278a
MD
833
834#ifdef DEBUG
835 if (type <= MAX_TRAP_MSG) {
836 uprintf("fatal process exception: %s",
837 trap_msg[type]);
838 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
839 uprintf(", fault VA = 0x%lx", (u_long)eva);
840 uprintf("\n");
841 }
842#endif
843
844out2:
845 ;
846#ifdef SMP
847 if (have_mplock)
848 rel_mplock();
849#endif
850#ifdef INVARIANTS
f9235b6d 851 KASSERT(crit_count == td->td_critcount,
3933a3ab 852 ("trap: critical section count mismatch! %d/%d",
f9235b6d 853 crit_count, td->td_pri));
3933a3ab
MD
854 KASSERT(curstop == td->td_toks_stop,
855 ("trap: extra tokens held after trap! %zd/%zd",
856 curstop - &td->td_toks_base,
857 td->td_toks_stop - &td->td_toks_base));
6f7b98e0 858#endif
6092278a 859}
6f7b98e0
MD
860
861int
862trap_pfault(struct trapframe *frame, int usermode, vm_offset_t eva)
863{
864 vm_offset_t va;
865 struct vmspace *vm = NULL;
866 vm_map_t map = 0;
867 int rv = 0;
1b9d3514 868 int fault_flags;
6f7b98e0
MD
869 vm_prot_t ftype;
870 thread_t td = curthread;
287ebb09 871 struct lwp *lp = td->td_lwp;
6f7b98e0
MD
872
873 va = trunc_page(eva);
6092278a 874 if (usermode == FALSE) {
6f7b98e0 875 /*
6092278a 876 * This is a fault on kernel virtual memory.
6f7b98e0 877 */
6f7b98e0
MD
878 map = &kernel_map;
879 } else {
880 /*
881 * This is a fault on non-kernel virtual memory.
882 * vm is initialized above to NULL. If curproc is NULL
883 * or curproc->p_vmspace is NULL the fault is fatal.
884 */
287ebb09
MD
885 if (lp != NULL)
886 vm = lp->lwp_vmspace;
6f7b98e0
MD
887
888 if (vm == NULL)
889 goto nogo;
890
891 map = &vm->vm_map;
892 }
893
4e7c41c5 894 if (frame->tf_xflags & PGEX_W)
0035dca9 895 ftype = VM_PROT_READ | VM_PROT_WRITE;
6f7b98e0
MD
896 else
897 ftype = VM_PROT_READ;
898
899 if (map != &kernel_map) {
900 /*
901 * Keep swapout from messing with us during this
902 * critical time.
903 */
287ebb09 904 PHOLD(lp->lwp_proc);
6f7b98e0 905
1b9d3514
MD
906 /*
907 * Issue fault
908 */
909 fault_flags = 0;
910 if (usermode)
911 fault_flags |= VM_FAULT_BURST;
912 if (ftype & VM_PROT_WRITE)
913 fault_flags |= VM_FAULT_DIRTY;
914 else
915 fault_flags |= VM_FAULT_NORMAL;
916 rv = vm_fault(map, va, ftype, fault_flags);
6f7b98e0 917
287ebb09 918 PRELE(lp->lwp_proc);
6f7b98e0
MD
919 } else {
920 /*
921 * Don't have to worry about process locking or stacks in the kernel.
922 */
923 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
924 }
925
926 if (rv == KERN_SUCCESS)
927 return (0);
928nogo:
929 if (!usermode) {
930 if (td->td_gd->gd_intr_nesting_level == 0 &&
931 td->td_pcb->pcb_onfault) {
932 frame->tf_eip = (register_t)td->td_pcb->pcb_onfault;
933 return (0);
934 }
6092278a 935 trap_fatal(frame, usermode, eva);
6f7b98e0
MD
936 return (-1);
937 }
6f7b98e0
MD
938 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
939}
940
941static void
6092278a 942trap_fatal(struct trapframe *frame, int usermode, vm_offset_t eva)
6f7b98e0
MD
943{
944 int code, type, ss, esp;
945
4e7c41c5 946 code = frame->tf_xflags;
6f7b98e0
MD
947 type = frame->tf_trapno;
948
9097944f 949 if (type <= MAX_TRAP_MSG) {
6f7b98e0
MD
950 kprintf("\n\nFatal trap %d: %s while in %s mode\n",
951 type, trap_msg[type],
9097944f
MD
952 (usermode ? "user" : "kernel"));
953 }
6f7b98e0 954#ifdef SMP
6a8aa90e 955 /* two separate prints in case of a trap on an unmapped page */
39005e16 956 kprintf("cpuid = %d\n", mycpu->gd_cpuid);
6f7b98e0
MD
957#endif
958 if (type == T_PAGEFLT) {
d557216f 959 kprintf("fault virtual address = %p\n", (void *)eva);
6f7b98e0 960 kprintf("fault code = %s %s, %s\n",
6092278a 961 usermode ? "user" : "supervisor",
6f7b98e0
MD
962 code & PGEX_W ? "write" : "read",
963 code & PGEX_P ? "protection violation" : "page not present");
964 }
965 kprintf("instruction pointer = 0x%x:0x%x\n",
966 frame->tf_cs & 0xffff, frame->tf_eip);
9097944f 967 if (usermode) {
6f7b98e0
MD
968 ss = frame->tf_ss & 0xffff;
969 esp = frame->tf_esp;
970 } else {
971 ss = GSEL(GDATA_SEL, SEL_KPL);
972 esp = (int)&frame->tf_esp;
973 }
974 kprintf("stack pointer = 0x%x:0x%x\n", ss, esp);
975 kprintf("frame pointer = 0x%x:0x%x\n", ss, frame->tf_ebp);
976 kprintf("processor eflags = ");
977 if (frame->tf_eflags & PSL_T)
978 kprintf("trace trap, ");
979 if (frame->tf_eflags & PSL_I)
980 kprintf("interrupt enabled, ");
981 if (frame->tf_eflags & PSL_NT)
982 kprintf("nested task, ");
983 if (frame->tf_eflags & PSL_RF)
984 kprintf("resume, ");
985#if 0
986 if (frame->tf_eflags & PSL_VM)
987 kprintf("vm86, ");
988#endif
989 kprintf("IOPL = %d\n", (frame->tf_eflags & PSL_IOPL) >> 12);
990 kprintf("current process = ");
991 if (curproc) {
992 kprintf("%lu (%s)\n",
993 (u_long)curproc->p_pid, curproc->p_comm ?
994 curproc->p_comm : "");
995 } else {
996 kprintf("Idle\n");
997 }
998 kprintf("current thread = pri %d ", curthread->td_pri);
f9235b6d 999 if (curthread->td_critcount)
6f7b98e0
MD
1000 kprintf("(CRIT)");
1001 kprintf("\n");
1002#ifdef SMP
1003/**
1004 * XXX FIXME:
1005 * we probably SHOULD have stopped the other CPUs before now!
1006 * another CPU COULD have been touching cpl at this moment...
1007 */
1008 kprintf(" <- SMP: XXX");
1009#endif
1010 kprintf("\n");
1011
1012#ifdef KDB
1013 if (kdb_trap(&psl))
1014 return;
1015#endif
1016#ifdef DDB
1017 if ((debugger_on_panic || db_active) && kdb_trap(type, code, frame))
1018 return;
1019#endif
1020 kprintf("trap number = %d\n", type);
1021 if (type <= MAX_TRAP_MSG)
1022 panic("%s", trap_msg[type]);
1023 else
1024 panic("unknown/reserved trap");
1025}
1026
1027/*
1028 * Double fault handler. Called when a fault occurs while writing
1029 * a frame for a trap/exception onto the stack. This usually occurs
1030 * when the stack overflows (such is the case with infinite recursion,
1031 * for example).
1032 *
1033 * XXX Note that the current PTD gets replaced by IdlePTD when the
1034 * task switch occurs. This means that the stack that was active at
1035 * the time of the double fault is not available at <kstack> unless
1036 * the machine was idle when the double fault occurred. The downside
1037 * of this is that "trace <ebp>" in ddb won't work.
1038 */
1039void
1040dblfault_handler(void)
1041{
1042 struct mdglobaldata *gd = mdcpu;
1043
1044 kprintf("\nFatal double fault:\n");
1045 kprintf("eip = 0x%x\n", gd->gd_common_tss.tss_eip);
1046 kprintf("esp = 0x%x\n", gd->gd_common_tss.tss_esp);
1047 kprintf("ebp = 0x%x\n", gd->gd_common_tss.tss_ebp);
1048#ifdef SMP
6a8aa90e 1049 /* two separate prints in case of a trap on an unmapped page */
39005e16 1050 kprintf("cpuid = %d\n", mycpu->gd_cpuid);
6f7b98e0
MD
1051#endif
1052 panic("double fault");
1053}
1054
6f7b98e0 1055/*
7bd858e5 1056 * syscall2 - MP aware system call request C handler
6f7b98e0 1057 *
7bd858e5
MD
1058 * A system call is essentially treated as a trap except that the
1059 * MP lock is not held on entry or return. We are responsible for
1060 * obtaining the MP lock if necessary and for handling ASTs
1061 * (e.g. a task switch) prior to return.
6f7b98e0 1062 *
7bd858e5 1063 * MPSAFE
6f7b98e0 1064 */
6f7b98e0 1065void
6092278a 1066syscall2(struct trapframe *frame)
6f7b98e0
MD
1067{
1068 struct thread *td = curthread;
1069 struct proc *p = td->td_proc;
1070 struct lwp *lp = td->td_lwp;
1071 caddr_t params;
1072 struct sysent *callp;
1073 register_t orig_tf_eflags;
1074 int sticks;
1075 int error;
1076 int narg;
1077#ifdef INVARIANTS
f9235b6d 1078 int crit_count = td->td_critcount;
6f7b98e0
MD
1079#endif
1080#ifdef SMP
1081 int have_mplock = 0;
1082#endif
1083 u_int code;
1084 union sysunion args;
1085
80bbc910
SS
1086 KTR_LOG(kernentry_syscall, lp->lwp_proc->p_pid, lp->lwp_tid,
1087 frame->tf_eax);
1088
d86a23e0 1089 userenter(td, p); /* lazy raise our priority */
6f7b98e0
MD
1090
1091 /*
1092 * Misc
1093 */
1094 sticks = (int)td->td_sticks;
6092278a 1095 orig_tf_eflags = frame->tf_eflags;
6f7b98e0
MD
1096
1097 /*
1098 * Virtual kernel intercept - if a VM context managed by a virtual
1099 * kernel issues a system call the virtual kernel handles it, not us.
1100 * Restore the virtual kernel context and return from its system
1101 * call. The current frame is copied out to the virtual kernel.
1102 */
39005e16 1103 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
bb47c072 1104 vkernel_trap(lp, frame);
6f7b98e0
MD
1105 error = EJUSTRETURN;
1106 goto out;
1107 }
1108
1109 /*
1110 * Get the system call parameters and account for time
1111 */
6092278a
MD
1112 lp->lwp_md.md_regs = frame;
1113 params = (caddr_t)frame->tf_esp + sizeof(int);
1114 code = frame->tf_eax;
6f7b98e0
MD
1115
1116 if (p->p_sysent->sv_prepsyscall) {
1117 (*p->p_sysent->sv_prepsyscall)(
6092278a 1118 frame, (int *)(&args.nosys.sysmsg + 1),
6f7b98e0
MD
1119 &code, &params);
1120 } else {
1121 /*
1122 * Need to check if this is a 32 bit or 64 bit syscall.
1123 * fuword is MP aware.
1124 */
1125 if (code == SYS_syscall) {
1126 /*
1127 * Code is first argument, followed by actual args.
1128 */
1129 code = fuword(params);
1130 params += sizeof(int);
1131 } else if (code == SYS___syscall) {
1132 /*
1133 * Like syscall, but code is a quad, so as to maintain
1134 * quad alignment for the rest of the arguments.
1135 */
1136 code = fuword(params);
1137 params += sizeof(quad_t);
1138 }
1139 }
1140
1141 code &= p->p_sysent->sv_mask;
1142 if (code >= p->p_sysent->sv_size)
1143 callp = &p->p_sysent->sv_table[0];
1144 else
1145 callp = &p->p_sysent->sv_table[code];
1146
1147 narg = callp->sy_narg & SYF_ARGMASK;
1148
1149 /*
1150 * copyin is MP aware, but the tracing code is not
1151 */
1152 if (narg && params) {
1153 error = copyin(params, (caddr_t)(&args.nosys.sysmsg + 1),
1154 narg * sizeof(register_t));
1155 if (error) {
1156#ifdef KTRACE
1157 if (KTRPOINT(td, KTR_SYSCALL)) {
1158 MAKEMPSAFE(have_mplock);
1159
9fb04d14 1160 ktrsyscall(lp, code, narg,
6f7b98e0
MD
1161 (void *)(&args.nosys.sysmsg + 1));
1162 }
1163#endif
1164 goto bad;
1165 }
1166 }
1167
1168#ifdef KTRACE
1169 if (KTRPOINT(td, KTR_SYSCALL)) {
1170 MAKEMPSAFE(have_mplock);
9fb04d14 1171 ktrsyscall(lp, code, narg, (void *)(&args.nosys.sysmsg + 1));
6f7b98e0
MD
1172 }
1173#endif
1174
1175 /*
1176 * For traditional syscall code edx is left untouched when 32 bit
1177 * results are returned. Since edx is loaded from fds[1] when the
1178 * system call returns we pre-set it here.
1179 */
1180 args.sysmsg_fds[0] = 0;
6092278a 1181 args.sysmsg_fds[1] = frame->tf_edx;
6f7b98e0
MD
1182
1183 /*
1184 * The syscall might manipulate the trap frame. If it does it
1185 * will probably return EJUSTRETURN.
1186 */
6092278a 1187 args.sysmsg_frame = frame;
6f7b98e0
MD
1188
1189 STOPEVENT(p, S_SCE, narg); /* MP aware */
1190
6f7b98e0 1191 /*
3919ced0
MD
1192 * NOTE: All system calls run MPSAFE now. The system call itself
1193 * is responsible for getting the MP lock.
6f7b98e0 1194 */
6f7b98e0
MD
1195 error = (*callp->sy_call)(&args);
1196
aaf8b91f
MD
1197#if 0
1198 kprintf("system call %d returned %d\n", code, error);
1199#endif
1200
6f7b98e0
MD
1201out:
1202 /*
1203 * MP SAFE (we may or may not have the MP lock at this point)
1204 */
1205 switch (error) {
1206 case 0:
1207 /*
1208 * Reinitialize proc pointer `p' as it may be different
1209 * if this is a child returning from fork syscall.
1210 */
1211 p = curproc;
1212 lp = curthread->td_lwp;
6092278a
MD
1213 frame->tf_eax = args.sysmsg_fds[0];
1214 frame->tf_edx = args.sysmsg_fds[1];
1215 frame->tf_eflags &= ~PSL_C;
6f7b98e0
MD
1216 break;
1217 case ERESTART:
1218 /*
1219 * Reconstruct pc, assuming lcall $X,y is 7 bytes,
1220 * int 0x80 is 2 bytes. We saved this in tf_err.
1221 */
6092278a 1222 frame->tf_eip -= frame->tf_err;
6f7b98e0
MD
1223 break;
1224 case EJUSTRETURN:
1225 break;
1226 case EASYNC:
1227 panic("Unexpected EASYNC return value (for now)");
1228 default:
1229bad:
1230 if (p->p_sysent->sv_errsize) {
1231 if (error >= p->p_sysent->sv_errsize)
1232 error = -1; /* XXX */
1233 else
1234 error = p->p_sysent->sv_errtbl[error];
1235 }
6092278a
MD
1236 frame->tf_eax = error;
1237 frame->tf_eflags |= PSL_C;
6f7b98e0
MD
1238 break;
1239 }
1240
1241 /*
1242 * Traced syscall. trapsignal() is not MP aware.
1243 */
1244 if ((orig_tf_eflags & PSL_T) /*&& !(orig_tf_eflags & PSL_VM)*/) {
1245 MAKEMPSAFE(have_mplock);
6092278a 1246 frame->tf_eflags &= ~PSL_T;
f2642d5a 1247 trapsignal(lp, SIGTRAP, TRAP_TRACE);
6f7b98e0
MD
1248 }
1249
1250 /*
1251 * Handle reschedule and other end-of-syscall issues
1252 */
6092278a 1253 userret(lp, frame, sticks);
6f7b98e0
MD
1254
1255#ifdef KTRACE
1256 if (KTRPOINT(td, KTR_SYSRET)) {
1257 MAKEMPSAFE(have_mplock);
9fb04d14 1258 ktrsysret(lp, code, error, args.sysmsg_result);
6f7b98e0
MD
1259 }
1260#endif
1261
1262 /*
1263 * This works because errno is findable through the
1264 * register set. If we ever support an emulation where this
1265 * is not the case, this code will need to be revisited.
1266 */
1267 STOPEVENT(p, S_SCX, code);
1268
1269 userexit(lp);
1270#ifdef SMP
1271 /*
1272 * Release the MP lock if we had to get it
1273 */
6f7b98e0
MD
1274 if (have_mplock)
1275 rel_mplock();
1276#endif
80bbc910 1277 KTR_LOG(kernentry_syscall_ret, lp->lwp_proc->p_pid, lp->lwp_tid, error);
6f7b98e0 1278#ifdef INVARIANTS
f9235b6d 1279 KASSERT(crit_count == td->td_critcount,
6f7b98e0 1280 ("syscall: critical section count mismatch! %d/%d",
f9235b6d 1281 crit_count, td->td_pri));
3933a3ab
MD
1282 KASSERT(&td->td_toks_base == td->td_toks_stop,
1283 ("syscall: extra tokens held after trap! %zd",
1284 td->td_toks_stop - &td->td_toks_base));
6f7b98e0
MD
1285#endif
1286}
1287
2b0bd8aa
MD
1288/*
1289 * NOTE: mplock not held at any point
1290 */
91bd9c1e
SS
1291void
1292fork_return(struct lwp *lp, struct trapframe *frame)
1293{
1294 frame->tf_eax = 0; /* Child returns zero */
1295 frame->tf_eflags &= ~PSL_C; /* success */
1296 frame->tf_edx = 1;
1297
1298 generic_lwp_return(lp, frame);
80bbc910 1299 KTR_LOG(kernentry_fork_ret, lp->lwp_proc->p_pid, lp->lwp_tid);
91bd9c1e
SS
1300}
1301
6f7b98e0
MD
1302/*
1303 * Simplified back end of syscall(), used when returning from fork()
2b0bd8aa
MD
1304 * directly into user mode.
1305 *
1306 * This code will return back into the fork trampoline code which then
1307 * runs doreti.
1308 *
1309 * NOTE: The mplock is not held at any point.
6f7b98e0
MD
1310 */
1311void
91bd9c1e 1312generic_lwp_return(struct lwp *lp, struct trapframe *frame)
6f7b98e0
MD
1313{
1314 struct proc *p = lp->lwp_proc;
1315
6f7b98e0
MD
1316 /*
1317 * Newly forked processes are given a kernel priority. We have to
1318 * adjust the priority to a normal user priority and fake entry
1319 * into the kernel (call userenter()) to install a passive release
1320 * function just in case userret() decides to stop the process. This
1321 * can occur when ^Z races a fork. If we do not install the passive
1322 * release function the current process designation will not be
1323 * released when the thread goes to sleep.
1324 */
1325 lwkt_setpri_self(TDPRI_USER_NORM);
d86a23e0 1326 userenter(lp->lwp_thread, p);
c7eb0589 1327 userret(lp, frame, 0);
6f7b98e0
MD
1328#ifdef KTRACE
1329 if (KTRPOINT(lp->lwp_thread, KTR_SYSRET))
9fb04d14 1330 ktrsysret(lp, SYS_fork, 0, 0);
6f7b98e0
MD
1331#endif
1332 p->p_flag |= P_PASSIVE_ACQ;
1333 userexit(lp);
1334 p->p_flag &= ~P_PASSIVE_ACQ;
6f7b98e0 1335}
9ad680a3 1336
4e7c41c5
MD
1337/*
1338 * doreti has turned into this. The frame is directly on the stack. We
1339 * pull everything else we need (fpu and tls context) from the current
1340 * thread.
1341 *
1342 * Note on fpu interactions: In a virtual kernel, the fpu context for
1343 * an emulated user mode process is not shared with the virtual kernel's
1344 * fpu context, so we only have to 'stack' fpu contexts within the virtual
1345 * kernel itself, and not even then since the signal() contexts that we care
1346 * about save and restore the FPU state (I think anyhow).
1347 *
1348 * vmspace_ctl() returns an error only if it had problems instaling the
1349 * context we supplied or problems copying data to/from our VM space.
1350 */
9ad680a3 1351void
c7eb0589 1352go_user(struct intrframe *frame)
9ad680a3 1353{
c7eb0589 1354 struct trapframe *tf = (void *)&frame->if_gs;
4e7c41c5
MD
1355 int r;
1356
1fdc022f
MD
1357 /*
1358 * Interrupts may be disabled on entry, make sure all signals
1359 * can be received before beginning our loop.
1360 */
1361 sigsetmask(0);
1362
1363 /*
1364 * Switch to the current simulated user process, then call
1365 * user_trap() when we break out of it (usually due to a signal).
1366 */
135d7199 1367 for (;;) {
431d0fef
MD
1368 /*
1369 * Tell the real kernel whether it is ok to use the FP
1370 * unit or not.
eaaa6d60
MD
1371 *
1372 * The critical section is required to prevent an interrupt
1373 * from causing a preemptive task switch and changing
1374 * the FP state.
431d0fef 1375 */
eaaa6d60 1376 crit_enter();
431d0fef 1377 if (mdcpu->gd_npxthread == curthread) {
4b486183 1378 tf->tf_xflags &= ~PGEX_FPFAULT;
431d0fef 1379 } else {
4b486183 1380 tf->tf_xflags |= PGEX_FPFAULT;
431d0fef
MD
1381 }
1382
7c1212ec
MD
1383 /*
1384 * Run emulated user process context. This call interlocks
1385 * with new mailbox signals.
4b486183
MD
1386 *
1387 * Set PGEX_U unconditionally, indicating a user frame (the
1388 * bit is normally set only by T_PAGEFLT).
7c1212ec
MD
1389 */
1390 r = vmspace_ctl(&curproc->p_vmspace->vm_pmap, VMSPACE_CTL_RUN,
1391 tf, &curthread->td_savevext);
eaaa6d60 1392 crit_exit();
c7eb0589 1393 frame->if_xflags |= PGEX_U;
4b486183
MD
1394#if 0
1395 kprintf("GO USER %d trap %d EVA %08x EIP %08x ESP %08x XFLAGS %02x/%02x\n",
1396 r, tf->tf_trapno, tf->tf_err, tf->tf_eip, tf->tf_esp,
c7eb0589 1397 tf->tf_xflags, frame->if_xflags);
4b486183 1398#endif
7c1212ec 1399 if (r < 0) {
24eb47e0 1400 if (errno != EINTR)
bb47c072 1401 panic("vmspace_ctl failed error %d", errno);
aaf8b91f 1402 } else {
7c1212ec
MD
1403 if (tf->tf_trapno) {
1404 user_trap(tf);
7c1212ec 1405 }
aaf8b91f 1406 }
6179d4f3
MD
1407 if (mycpu->gd_reqflags & RQF_AST_MASK) {
1408 tf->tf_trapno = T_ASTFLT;
1409 user_trap(tf);
1410 }
1411 tf->tf_trapno = 0;
135d7199 1412 }
9ad680a3 1413}
135d7199 1414
431d0fef 1415/*
4b486183 1416 * If PGEX_FPFAULT is set then set FP_VIRTFP in the PCB to force a T_DNA
431d0fef
MD
1417 * fault (which is then passed back to the virtual kernel) if an attempt is
1418 * made to use the FP unit.
1419 *
1420 * XXX this is a fairly big hack.
1421 */
1422void
1423set_vkernel_fp(struct trapframe *frame)
1424{
1425 struct thread *td = curthread;
1426
4b486183 1427 if (frame->tf_xflags & PGEX_FPFAULT) {
431d0fef
MD
1428 td->td_pcb->pcb_flags |= FP_VIRTFP;
1429 if (mdcpu->gd_npxthread == td)
1430 npxexit();
1431 } else {
1432 td->td_pcb->pcb_flags &= ~FP_VIRTFP;
1433 }
1434}
1435
bb47c072
MD
1436/*
1437 * Called from vkernel_trap() to fixup the vkernel's syscall
1438 * frame for vmspace_ctl() return.
1439 */
1440void
1441cpu_vkernel_trap(struct trapframe *frame, int error)
1442{
1443 frame->tf_eax = error;
1444 if (error)
1445 frame->tf_eflags |= PSL_C;
1446 else
1447 frame->tf_eflags &= ~PSL_C;
1448}