syscall messaging 3: Expand the 'header' that goes in front of the syscall
[dragonfly.git] / sys / i386 / 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 $
c7114eea 39 * $DragonFly: src/sys/i386/i386/Attic/trap.c,v 1.30 2003/07/30 00:19:13 dillon Exp $
984263bc
MD
40 */
41
42/*
43 * 386 Trap and System call handling
44 */
45
46#include "opt_cpu.h"
47#include "opt_ddb.h"
48#include "opt_ktrace.h"
49#include "opt_clock.h"
50#include "opt_trap.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>
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>
4fd10eb6 64#include <sys/malloc.h>
984263bc
MD
65#ifdef KTRACE
66#include <sys/ktrace.h>
67#endif
a64ba182
MD
68#include <sys/sysproto.h>
69#include <sys/sysunion.h>
984263bc
MD
70
71#include <vm/vm.h>
72#include <vm/vm_param.h>
73#include <sys/lock.h>
74#include <vm/pmap.h>
75#include <vm/vm_kern.h>
76#include <vm/vm_map.h>
77#include <vm/vm_page.h>
78#include <vm/vm_extern.h>
79
80#include <machine/cpu.h>
81#include <machine/ipl.h>
82#include <machine/md_var.h>
83#include <machine/pcb.h>
84#ifdef SMP
85#include <machine/smp.h>
86#endif
87#include <machine/tss.h>
85100692 88#include <machine/globaldata.h>
984263bc
MD
89
90#include <i386/isa/intr_machdep.h>
91
92#ifdef POWERFAIL_NMI
93#include <sys/syslog.h>
94#include <machine/clock.h>
95#endif
96
97#include <machine/vm86.h>
98
99#include <ddb/ddb.h>
41c20dac 100#include <sys/thread2.h>
984263bc
MD
101
102#include "isa.h"
103#include "npx.h"
104
105int (*pmath_emulate) __P((struct trapframe *));
106
107extern void trap __P((struct trapframe frame));
108extern int trapwrite __P((unsigned addr));
109extern void syscall2 __P((struct trapframe frame));
a64ba182 110extern void sendsys2 __P((struct trapframe frame));
984263bc
MD
111
112static int trap_pfault __P((struct trapframe *, int, vm_offset_t));
113static void trap_fatal __P((struct trapframe *, vm_offset_t));
114void dblfault_handler __P((void));
115
116extern inthand_t IDTVEC(syscall);
117
118#define MAX_TRAP_MSG 28
119static char *trap_msg[] = {
120 "", /* 0 unused */
121 "privileged instruction fault", /* 1 T_PRIVINFLT */
122 "", /* 2 unused */
123 "breakpoint instruction fault", /* 3 T_BPTFLT */
124 "", /* 4 unused */
125 "", /* 5 unused */
126 "arithmetic trap", /* 6 T_ARITHTRAP */
127 "system forced exception", /* 7 T_ASTFLT */
128 "", /* 8 unused */
129 "general protection fault", /* 9 T_PROTFLT */
130 "trace trap", /* 10 T_TRCTRAP */
131 "", /* 11 unused */
132 "page fault", /* 12 T_PAGEFLT */
133 "", /* 13 unused */
134 "alignment fault", /* 14 T_ALIGNFLT */
135 "", /* 15 unused */
136 "", /* 16 unused */
137 "", /* 17 unused */
138 "integer divide fault", /* 18 T_DIVIDE */
139 "non-maskable interrupt trap", /* 19 T_NMI */
140 "overflow trap", /* 20 T_OFLOW */
141 "FPU bounds check fault", /* 21 T_BOUND */
142 "FPU device not available", /* 22 T_DNA */
143 "double fault", /* 23 T_DOUBLEFLT */
144 "FPU operand fetch fault", /* 24 T_FPOPFLT */
145 "invalid TSS fault", /* 25 T_TSSFLT */
146 "segment not present fault", /* 26 T_SEGNPFLT */
147 "stack fault", /* 27 T_STKFLT */
148 "machine check trap", /* 28 T_MCHK */
149};
150
984263bc
MD
151#if defined(I586_CPU) && !defined(NO_F00F_HACK)
152extern int has_f00f_bug;
153#endif
154
155#ifdef DDB
156static int ddb_on_nmi = 1;
157SYSCTL_INT(_machdep, OID_AUTO, ddb_on_nmi, CTLFLAG_RW,
158 &ddb_on_nmi, 0, "Go to DDB on NMI");
159#endif
160static int panic_on_nmi = 1;
161SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RW,
162 &panic_on_nmi, 0, "Panic on NMI");
d9eea1a5
MD
163static int fast_release;
164SYSCTL_INT(_machdep, OID_AUTO, fast_release, CTLFLAG_RW,
165 &fast_release, 0, "Passive Release was optimal");
166static int slow_release;
167SYSCTL_INT(_machdep, OID_AUTO, slow_release, CTLFLAG_RW,
168 &slow_release, 0, "Passive Release was nonoptimal");
984263bc 169
4fd10eb6
MD
170MALLOC_DEFINE(M_SYSMSG, "sysmsg", "sysmsg structure");
171
26a0694b
MD
172/*
173 * USER->KERNEL transition. Do not transition us out of userland from the
174 * point of view of the userland scheduler unless we actually have to
175 * switch.
176 *
a2a5ad0d
MD
177 * usertdsw is called from within a critical section and the BGL will still
178 * be held. This function is NOT called for preemptions, only for switchouts.
26a0694b
MD
179 */
180static void
a2a5ad0d 181passive_release(struct thread *td)
26a0694b 182{
a2a5ad0d 183 struct proc *p = td->td_proc;
26a0694b 184
a2a5ad0d 185 td->td_release = NULL;
26a0694b 186 lwkt_setpri_self(TDPRI_KERN_USER);
26a0694b 187 if (p->p_flag & P_CURPROC) {
a2a5ad0d 188 release_curproc(p);
26a0694b 189 }
26a0694b
MD
190}
191
192/*
d81ccc3e
MD
193 * userenter() passively intercepts the thread switch function to increase
194 * the thread priority from a user priority to a kernel priority, reducing
195 * syscall and trap overhead for the case where no switch occurs.
26a0694b 196 */
efd3c4c3 197
26a0694b 198static __inline void
7966cb69 199userenter(struct thread *curtd)
26a0694b 200{
7966cb69 201 curtd->td_release = passive_release;
26a0694b
MD
202}
203
a2a5ad0d
MD
204static __inline void
205userexit(struct proc *p)
984263bc 206{
a2a5ad0d 207 struct thread *td = p->p_thread;
984263bc 208
26a0694b 209 /*
a2a5ad0d
MD
210 * If we did not have to release we should already be P_CURPROC. If
211 * we did have to release we must acquire P_CURPROC again and then
212 * restore our priority for user return.
cb973d15
MD
213 *
214 * Lowering our priority may make other higher priority threads
215 * runnable. lwkt_setpri_self() does not switch away, so call
216 * lwkt_maybe_switch() to deal with it.
26a0694b 217 */
a2a5ad0d 218 if (td->td_release) {
d9eea1a5 219 ++fast_release;
a2a5ad0d
MD
220 td->td_release = NULL;
221 KKASSERT(p->p_flag & P_CURPROC);
222 } else {
d9eea1a5 223 ++slow_release;
a2a5ad0d 224 acquire_curproc(p);
26a0694b
MD
225 switch(p->p_rtprio.type) {
226 case RTP_PRIO_IDLE:
227 lwkt_setpri_self(TDPRI_USER_IDLE);
228 break;
229 case RTP_PRIO_REALTIME:
230 case RTP_PRIO_FIFO:
231 lwkt_setpri_self(TDPRI_USER_REAL);
232 break;
233 default:
234 lwkt_setpri_self(TDPRI_USER_NORM);
235 break;
236 }
cb973d15 237 lwkt_maybe_switch();
26a0694b 238 }
a2a5ad0d
MD
239}
240
241
242static void
243userret(struct proc *p, struct trapframe *frame, u_quad_t oticks)
244{
245 int sig;
246
247 /*
248 * Post any pending signals
249 */
250 while ((sig = CURSIG(p)) != 0) {
251 postsig(sig);
252 }
26a0694b
MD
253
254 /*
cb973d15 255 * If a reschedule has been requested then the easiest solution
5c323556
MD
256 * is to run our passive release function which will possibly
257 * shift our P_CURPROC designation to another user process.
258 * We don't actually switch here because that would be a waste
259 * of cycles (the newly scheduled user process would just switch
260 * back to us since we might be running at a kernel priority).
261 * Instead we fall through and will switch away when we attempt
262 * to reacquire our P_CURPROC designation.
26a0694b 263 */
984263bc 264 if (resched_wanted()) {
cb973d15
MD
265 if (curthread->td_release)
266 passive_release(curthread);
984263bc 267 }
26a0694b 268
984263bc 269 /*
6ad39cae 270 * Charge system time if profiling. Note: times are in microseconds.
984263bc
MD
271 */
272 if (p->p_flag & P_PROFIL) {
d16a8831 273 addupc_task(p, frame->tf_eip,
6ad39cae 274 (u_int)(curthread->td_sticks - oticks));
984263bc 275 }
26a0694b
MD
276
277 /*
a2a5ad0d 278 * Post any pending signals XXX
26a0694b 279 */
a2a5ad0d
MD
280 while ((sig = CURSIG(p)) != 0)
281 postsig(sig);
984263bc
MD
282}
283
284#ifdef DEVICE_POLLING
285extern u_int32_t poll_in_trap;
286extern int ether_poll __P((int count));
287#endif /* DEVICE_POLLING */
288
289/*
290 * Exception, fault, and trap interface to the FreeBSD kernel.
291 * This common code is called from assembly language IDT gate entry
292 * routines that prepare a suitable stack frame, and restore this
293 * frame after the exception has been processed.
a2a5ad0d
MD
294 *
295 * This function is also called from doreti in an interlock to handle ASTs.
296 * For example: hardwareint->INTROUTINE->(set ast)->doreti->trap
297 *
298 * NOTE! We have to retrieve the fault address prior to obtaining the
299 * MP lock because get_mplock() may switch out. YYY cr2 really ought
300 * to be retrieved by the assembly code, not here.
984263bc 301 */
984263bc
MD
302void
303trap(frame)
304 struct trapframe frame;
305{
7966cb69
MD
306 struct thread *td = curthread;
307 struct proc *p;
984263bc
MD
308 u_quad_t sticks = 0;
309 int i = 0, ucode = 0, type, code;
310 vm_offset_t eva;
311
7966cb69 312 p = td->td_proc;
984263bc
MD
313#ifdef DDB
314 if (db_active) {
315 eva = (frame.tf_trapno == T_PAGEFLT ? rcr2() : 0);
a2a5ad0d 316 get_mplock();
984263bc 317 trap_fatal(&frame, eva);
8a8d5d85 318 goto out2;
984263bc
MD
319 }
320#endif
321
a2a5ad0d
MD
322 eva = 0;
323 if (frame.tf_trapno == T_PAGEFLT) {
324 /*
325 * For some Cyrix CPUs, %cr2 is clobbered by interrupts.
326 * This problem is worked around by using an interrupt
327 * gate for the pagefault handler. We are finally ready
328 * to read %cr2 and then must reenable interrupts.
329 *
330 * XXX this should be in the switch statement, but the
331 * NO_FOOF_HACK and VM86 goto and ifdefs obfuscate the
332 * flow of control too much for this to be obviously
333 * correct.
334 */
335 eva = rcr2();
336 get_mplock();
337 cpu_enable_intr();
338 } else {
339 get_mplock();
340 }
341 /*
342 * MP lock is held at this point
343 */
344
984263bc
MD
345 if (!(frame.tf_eflags & PSL_I)) {
346 /*
347 * Buggy application or kernel code has disabled interrupts
348 * and then trapped. Enabling interrupts now is wrong, but
349 * it is better than running with interrupts disabled until
350 * they are accidentally enabled later.
351 */
352 type = frame.tf_trapno;
a2a5ad0d 353 if (ISPL(frame.tf_cs)==SEL_UPL || (frame.tf_eflags & PSL_VM)) {
984263bc
MD
354 printf(
355 "pid %ld (%s): trap %d with interrupts disabled\n",
356 (long)curproc->p_pid, curproc->p_comm, type);
a2a5ad0d 357 } else if (type != T_BPTFLT && type != T_TRCTRAP) {
984263bc
MD
358 /*
359 * XXX not quite right, since this may be for a
360 * multiple fault in user mode.
361 */
362 printf("kernel trap %d with interrupts disabled\n",
363 type);
a2a5ad0d 364 }
8a8d5d85 365 cpu_enable_intr();
984263bc
MD
366 }
367
984263bc
MD
368
369#ifdef DEVICE_POLLING
370 if (poll_in_trap)
371 ether_poll(poll_in_trap);
372#endif /* DEVICE_POLLING */
373
374#if defined(I586_CPU) && !defined(NO_F00F_HACK)
375restart:
376#endif
377 type = frame.tf_trapno;
378 code = frame.tf_err;
379
380 if (in_vm86call) {
381 if (frame.tf_eflags & PSL_VM &&
382 (type == T_PROTFLT || type == T_STKFLT)) {
96728c05
MD
383#ifdef SMP
384 KKASSERT(curthread->td_mpcount > 0);
385#endif
984263bc 386 i = vm86_emulate((struct vm86frame *)&frame);
96728c05
MD
387#ifdef SMP
388 KKASSERT(curthread->td_mpcount > 0);
389#endif
8a8d5d85 390 if (i != 0) {
984263bc
MD
391 /*
392 * returns to original process
393 */
394 vm86_trap((struct vm86frame *)&frame);
96728c05 395 KKASSERT(0);
8a8d5d85
MD
396 }
397 goto out2;
984263bc
MD
398 }
399 switch (type) {
400 /*
401 * these traps want either a process context, or
402 * assume a normal userspace trap.
403 */
404 case T_PROTFLT:
405 case T_SEGNPFLT:
406 trap_fatal(&frame, eva);
8a8d5d85 407 goto out2;
984263bc
MD
408 case T_TRCTRAP:
409 type = T_BPTFLT; /* kernel breakpoint */
410 /* FALL THROUGH */
411 }
412 goto kernel_trap; /* normal kernel trap handling */
413 }
414
415 if ((ISPL(frame.tf_cs) == SEL_UPL) || (frame.tf_eflags & PSL_VM)) {
416 /* user trap */
417
7966cb69 418 userenter(td);
26a0694b 419
d16a8831 420 sticks = curthread->td_sticks;
984263bc
MD
421 p->p_md.md_regs = &frame;
422
423 switch (type) {
424 case T_PRIVINFLT: /* privileged instruction fault */
425 ucode = type;
426 i = SIGILL;
427 break;
428
429 case T_BPTFLT: /* bpt instruction fault */
430 case T_TRCTRAP: /* trace trap */
431 frame.tf_eflags &= ~PSL_T;
432 i = SIGTRAP;
433 break;
434
435 case T_ARITHTRAP: /* arithmetic trap */
436 ucode = code;
437 i = SIGFPE;
438 break;
439
440 case T_ASTFLT: /* Allow process switch */
12e4aaff 441 mycpu->gd_cnt.v_soft++;
235957ed
MD
442 if (mycpu->gd_reqflags & RQF_AST_OWEUPC) {
443 atomic_clear_int_nonlocked(&mycpu->gd_reqflags,
444 RQF_AST_OWEUPC);
984263bc
MD
445 addupc_task(p, p->p_stats->p_prof.pr_addr,
446 p->p_stats->p_prof.pr_ticks);
447 }
448 goto out;
449
450 /*
451 * The following two traps can happen in
452 * vm86 mode, and, if so, we want to handle
453 * them specially.
454 */
455 case T_PROTFLT: /* general protection fault */
456 case T_STKFLT: /* stack fault */
457 if (frame.tf_eflags & PSL_VM) {
458 i = vm86_emulate((struct vm86frame *)&frame);
459 if (i == 0)
460 goto out;
461 break;
462 }
463 /* FALL THROUGH */
464
465 case T_SEGNPFLT: /* segment not present fault */
466 case T_TSSFLT: /* invalid TSS fault */
467 case T_DOUBLEFLT: /* double fault */
468 default:
469 ucode = code + BUS_SEGM_FAULT ;
470 i = SIGBUS;
471 break;
472
473 case T_PAGEFLT: /* page fault */
474 i = trap_pfault(&frame, TRUE, eva);
475 if (i == -1)
d81ccc3e 476 goto out;
984263bc
MD
477#if defined(I586_CPU) && !defined(NO_F00F_HACK)
478 if (i == -2)
479 goto restart;
480#endif
481 if (i == 0)
482 goto out;
483
484 ucode = T_PAGEFLT;
485 break;
486
487 case T_DIVIDE: /* integer divide fault */
488 ucode = FPE_INTDIV;
489 i = SIGFPE;
490 break;
491
492#if NISA > 0
493 case T_NMI:
494#ifdef POWERFAIL_NMI
495 goto handle_powerfail;
496#else /* !POWERFAIL_NMI */
497 /* machine/parity/power fail/"kitchen sink" faults */
498 if (isa_nmi(code) == 0) {
499#ifdef DDB
500 /*
501 * NMI can be hooked up to a pushbutton
502 * for debugging.
503 */
504 if (ddb_on_nmi) {
505 printf ("NMI ... going to debugger\n");
506 kdb_trap (type, 0, &frame);
507 }
508#endif /* DDB */
8a8d5d85 509 goto out2;
984263bc
MD
510 } else if (panic_on_nmi)
511 panic("NMI indicates hardware failure");
512 break;
513#endif /* POWERFAIL_NMI */
514#endif /* NISA > 0 */
515
516 case T_OFLOW: /* integer overflow fault */
517 ucode = FPE_INTOVF;
518 i = SIGFPE;
519 break;
520
521 case T_BOUND: /* bounds check fault */
522 ucode = FPE_FLTSUB;
523 i = SIGFPE;
524 break;
525
526 case T_DNA:
527#if NNPX > 0
528 /* if a transparent fault (due to context switch "late") */
529 if (npxdna())
d81ccc3e 530 goto out;
984263bc
MD
531#endif
532 if (!pmath_emulate) {
533 i = SIGFPE;
534 ucode = FPE_FPU_NP_TRAP;
535 break;
536 }
537 i = (*pmath_emulate)(&frame);
538 if (i == 0) {
539 if (!(frame.tf_eflags & PSL_T))
8a8d5d85 540 goto out2;
984263bc
MD
541 frame.tf_eflags &= ~PSL_T;
542 i = SIGTRAP;
543 }
544 /* else ucode = emulator_only_knows() XXX */
545 break;
546
547 case T_FPOPFLT: /* FPU operand fetch fault */
548 ucode = T_FPOPFLT;
549 i = SIGILL;
550 break;
551
552 case T_XMMFLT: /* SIMD floating-point exception */
553 ucode = 0; /* XXX */
554 i = SIGFPE;
555 break;
556 }
557 } else {
558kernel_trap:
559 /* kernel trap */
560
561 switch (type) {
562 case T_PAGEFLT: /* page fault */
563 (void) trap_pfault(&frame, FALSE, eva);
8a8d5d85 564 goto out2;
984263bc
MD
565
566 case T_DNA:
567#if NNPX > 0
568 /*
569 * The kernel is apparently using npx for copying.
570 * XXX this should be fatal unless the kernel has
571 * registered such use.
572 */
573 if (npxdna())
8a8d5d85 574 goto out2;
984263bc
MD
575#endif
576 break;
577
578 case T_PROTFLT: /* general protection fault */
579 case T_SEGNPFLT: /* segment not present fault */
580 /*
581 * Invalid segment selectors and out of bounds
582 * %eip's and %esp's can be set up in user mode.
583 * This causes a fault in kernel mode when the
584 * kernel tries to return to user mode. We want
585 * to get this fault so that we can fix the
586 * problem here and not have to check all the
587 * selectors and pointers when the user changes
588 * them.
589 */
590#define MAYBE_DORETI_FAULT(where, whereto) \
591 do { \
592 if (frame.tf_eip == (int)where) { \
593 frame.tf_eip = (int)whereto; \
8a8d5d85 594 goto out2; \
984263bc
MD
595 } \
596 } while (0)
597
ef0fdad1 598 if (mycpu->gd_intr_nesting_level == 0) {
984263bc
MD
599 /*
600 * Invalid %fs's and %gs's can be created using
601 * procfs or PT_SETREGS or by invalidating the
602 * underlying LDT entry. This causes a fault
603 * in kernel mode when the kernel attempts to
604 * switch contexts. Lose the bad context
605 * (XXX) so that we can continue, and generate
606 * a signal.
607 */
608 if (frame.tf_eip == (int)cpu_switch_load_gs) {
b7c628e4 609 curthread->td_pcb->pcb_gs = 0;
984263bc 610 psignal(p, SIGBUS);
8a8d5d85 611 goto out2;
984263bc
MD
612 }
613 MAYBE_DORETI_FAULT(doreti_iret,
614 doreti_iret_fault);
615 MAYBE_DORETI_FAULT(doreti_popl_ds,
616 doreti_popl_ds_fault);
617 MAYBE_DORETI_FAULT(doreti_popl_es,
618 doreti_popl_es_fault);
619 MAYBE_DORETI_FAULT(doreti_popl_fs,
620 doreti_popl_fs_fault);
b7c628e4
MD
621 if (curthread->td_pcb->pcb_onfault) {
622 frame.tf_eip = (int)curthread->td_pcb->pcb_onfault;
8a8d5d85 623 goto out2;
984263bc
MD
624 }
625 }
626 break;
627
628 case T_TSSFLT:
629 /*
630 * PSL_NT can be set in user mode and isn't cleared
631 * automatically when the kernel is entered. This
632 * causes a TSS fault when the kernel attempts to
633 * `iret' because the TSS link is uninitialized. We
634 * want to get this fault so that we can fix the
635 * problem here and not every time the kernel is
636 * entered.
637 */
638 if (frame.tf_eflags & PSL_NT) {
639 frame.tf_eflags &= ~PSL_NT;
8a8d5d85 640 goto out2;
984263bc
MD
641 }
642 break;
643
644 case T_TRCTRAP: /* trace trap */
645 if (frame.tf_eip == (int)IDTVEC(syscall)) {
646 /*
647 * We've just entered system mode via the
648 * syscall lcall. Continue single stepping
649 * silently until the syscall handler has
650 * saved the flags.
651 */
8a8d5d85 652 goto out2;
984263bc
MD
653 }
654 if (frame.tf_eip == (int)IDTVEC(syscall) + 1) {
655 /*
656 * The syscall handler has now saved the
657 * flags. Stop single stepping it.
658 */
659 frame.tf_eflags &= ~PSL_T;
8a8d5d85 660 goto out2;
984263bc
MD
661 }
662 /*
663 * Ignore debug register trace traps due to
664 * accesses in the user's address space, which
665 * can happen under several conditions such as
666 * if a user sets a watchpoint on a buffer and
667 * then passes that buffer to a system call.
668 * We still want to get TRCTRAPS for addresses
669 * in kernel space because that is useful when
670 * debugging the kernel.
671 */
672 if (user_dbreg_trap()) {
673 /*
674 * Reset breakpoint bits because the
675 * processor doesn't
676 */
677 load_dr6(rdr6() & 0xfffffff0);
8a8d5d85 678 goto out2;
984263bc
MD
679 }
680 /*
681 * Fall through (TRCTRAP kernel mode, kernel address)
682 */
683 case T_BPTFLT:
684 /*
685 * If DDB is enabled, let it handle the debugger trap.
686 * Otherwise, debugger traps "can't happen".
687 */
688#ifdef DDB
689 if (kdb_trap (type, 0, &frame))
8a8d5d85 690 goto out2;
984263bc
MD
691#endif
692 break;
693
694#if NISA > 0
695 case T_NMI:
696#ifdef POWERFAIL_NMI
697#ifndef TIMER_FREQ
698# define TIMER_FREQ 1193182
699#endif
700 handle_powerfail:
701 {
702 static unsigned lastalert = 0;
703
704 if(time_second - lastalert > 10)
705 {
706 log(LOG_WARNING, "NMI: power fail\n");
707 sysbeep(TIMER_FREQ/880, hz);
708 lastalert = time_second;
709 }
8a8d5d85
MD
710 /* YYY mp count */
711 goto out2;
984263bc
MD
712 }
713#else /* !POWERFAIL_NMI */
714 /* machine/parity/power fail/"kitchen sink" faults */
715 if (isa_nmi(code) == 0) {
716#ifdef DDB
717 /*
718 * NMI can be hooked up to a pushbutton
719 * for debugging.
720 */
721 if (ddb_on_nmi) {
722 printf ("NMI ... going to debugger\n");
723 kdb_trap (type, 0, &frame);
724 }
725#endif /* DDB */
8a8d5d85 726 goto out2;
984263bc 727 } else if (panic_on_nmi == 0)
8a8d5d85 728 goto out2;
984263bc
MD
729 /* FALL THROUGH */
730#endif /* POWERFAIL_NMI */
731#endif /* NISA > 0 */
732 }
733
734 trap_fatal(&frame, eva);
8a8d5d85 735 goto out2;
984263bc
MD
736 }
737
738 /* Translate fault for emulators (e.g. Linux) */
739 if (*p->p_sysent->sv_transtrap)
740 i = (*p->p_sysent->sv_transtrap)(i, type);
741
742 trapsignal(p, i, ucode);
743
744#ifdef DEBUG
745 if (type <= MAX_TRAP_MSG) {
746 uprintf("fatal process exception: %s",
747 trap_msg[type]);
748 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
749 uprintf(", fault VA = 0x%lx", (u_long)eva);
750 uprintf("\n");
751 }
752#endif
753
754out:
8a8d5d85
MD
755#ifdef SMP
756 if (ISPL(frame.tf_cs) == SEL_UPL)
757 KASSERT(curthread->td_mpcount == 1, ("badmpcount trap from %p", (void *)frame.tf_eip));
758#endif
759 userret(p, &frame, sticks);
a2a5ad0d 760 userexit(p);
8a8d5d85 761out2:
96728c05
MD
762#ifdef SMP
763 KKASSERT(curthread->td_mpcount > 0);
764#endif
8a8d5d85 765 rel_mplock();
984263bc
MD
766}
767
768#ifdef notyet
769/*
770 * This version doesn't allow a page fault to user space while
771 * in the kernel. The rest of the kernel needs to be made "safe"
772 * before this can be used. I think the only things remaining
773 * to be made safe are the iBCS2 code and the process tracing/
774 * debugging code.
775 */
776static int
777trap_pfault(frame, usermode, eva)
778 struct trapframe *frame;
779 int usermode;
780 vm_offset_t eva;
781{
782 vm_offset_t va;
783 struct vmspace *vm = NULL;
784 vm_map_t map = 0;
785 int rv = 0;
786 vm_prot_t ftype;
787 struct proc *p = curproc;
788
789 if (frame->tf_err & PGEX_W)
790 ftype = VM_PROT_WRITE;
791 else
792 ftype = VM_PROT_READ;
793
794 va = trunc_page(eva);
795 if (va < VM_MIN_KERNEL_ADDRESS) {
796 vm_offset_t v;
797 vm_page_t mpte;
798
799 if (p == NULL ||
800 (!usermode && va < VM_MAXUSER_ADDRESS &&
ef0fdad1 801 (mycpu->gd_intr_nesting_level != 0 ||
b7c628e4 802 curthread->td_pcb->pcb_onfault == NULL))) {
984263bc
MD
803 trap_fatal(frame, eva);
804 return (-1);
805 }
806
807 /*
808 * This is a fault on non-kernel virtual memory.
809 * vm is initialized above to NULL. If curproc is NULL
810 * or curproc->p_vmspace is NULL the fault is fatal.
811 */
812 vm = p->p_vmspace;
813 if (vm == NULL)
814 goto nogo;
815
816 map = &vm->vm_map;
817
818 /*
819 * Keep swapout from messing with us during this
820 * critical time.
821 */
822 ++p->p_lock;
823
824 /*
825 * Grow the stack if necessary
826 */
827 /* grow_stack returns false only if va falls into
828 * a growable stack region and the stack growth
829 * fails. It returns true if va was not within
830 * a growable stack region, or if the stack
831 * growth succeeded.
832 */
833 if (!grow_stack (p, va)) {
834 rv = KERN_FAILURE;
835 --p->p_lock;
836 goto nogo;
837 }
838
839 /* Fault in the user page: */
840 rv = vm_fault(map, va, ftype,
841 (ftype & VM_PROT_WRITE) ? VM_FAULT_DIRTY
842 : VM_FAULT_NORMAL);
843
844 --p->p_lock;
845 } else {
846 /*
847 * Don't allow user-mode faults in kernel address space.
848 */
849 if (usermode)
850 goto nogo;
851
852 /*
853 * Since we know that kernel virtual address addresses
854 * always have pte pages mapped, we just have to fault
855 * the page.
856 */
857 rv = vm_fault(kernel_map, va, ftype, VM_FAULT_NORMAL);
858 }
859
860 if (rv == KERN_SUCCESS)
861 return (0);
862nogo:
863 if (!usermode) {
ef0fdad1 864 if (mycpu->gd_intr_nesting_level == 0 && curthread->td_pcb->pcb_onfault) {
b7c628e4 865 frame->tf_eip = (int)curthread->td_pcb->pcb_onfault;
984263bc
MD
866 return (0);
867 }
868 trap_fatal(frame, eva);
869 return (-1);
870 }
871
872 /* kludge to pass faulting virtual address to sendsig */
873 frame->tf_err = eva;
874
875 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
876}
877#endif
878
879int
880trap_pfault(frame, usermode, eva)
881 struct trapframe *frame;
882 int usermode;
883 vm_offset_t eva;
884{
885 vm_offset_t va;
886 struct vmspace *vm = NULL;
887 vm_map_t map = 0;
888 int rv = 0;
889 vm_prot_t ftype;
890 struct proc *p = curproc;
891
892 va = trunc_page(eva);
893 if (va >= KERNBASE) {
894 /*
895 * Don't allow user-mode faults in kernel address space.
896 * An exception: if the faulting address is the invalid
897 * instruction entry in the IDT, then the Intel Pentium
898 * F00F bug workaround was triggered, and we need to
899 * treat it is as an illegal instruction, and not a page
900 * fault.
901 */
902#if defined(I586_CPU) && !defined(NO_F00F_HACK)
903 if ((eva == (unsigned int)&idt[6]) && has_f00f_bug) {
904 frame->tf_trapno = T_PRIVINFLT;
905 return -2;
906 }
907#endif
908 if (usermode)
909 goto nogo;
910
911 map = kernel_map;
912 } else {
913 /*
914 * This is a fault on non-kernel virtual memory.
915 * vm is initialized above to NULL. If curproc is NULL
916 * or curproc->p_vmspace is NULL the fault is fatal.
917 */
918 if (p != NULL)
919 vm = p->p_vmspace;
920
921 if (vm == NULL)
922 goto nogo;
923
924 map = &vm->vm_map;
925 }
926
927 if (frame->tf_err & PGEX_W)
928 ftype = VM_PROT_WRITE;
929 else
930 ftype = VM_PROT_READ;
931
932 if (map != kernel_map) {
933 /*
934 * Keep swapout from messing with us during this
935 * critical time.
936 */
937 ++p->p_lock;
938
939 /*
940 * Grow the stack if necessary
941 */
942 /* grow_stack returns false only if va falls into
943 * a growable stack region and the stack growth
944 * fails. It returns true if va was not within
945 * a growable stack region, or if the stack
946 * growth succeeded.
947 */
948 if (!grow_stack (p, va)) {
949 rv = KERN_FAILURE;
950 --p->p_lock;
951 goto nogo;
952 }
953
954 /* Fault in the user page: */
955 rv = vm_fault(map, va, ftype,
956 (ftype & VM_PROT_WRITE) ? VM_FAULT_DIRTY
957 : VM_FAULT_NORMAL);
958
959 --p->p_lock;
960 } else {
961 /*
962 * Don't have to worry about process locking or stacks in the kernel.
963 */
964 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
965 }
966
967 if (rv == KERN_SUCCESS)
968 return (0);
969nogo:
970 if (!usermode) {
ef0fdad1 971 if (mycpu->gd_intr_nesting_level == 0 && curthread->td_pcb->pcb_onfault) {
b7c628e4 972 frame->tf_eip = (int)curthread->td_pcb->pcb_onfault;
984263bc
MD
973 return (0);
974 }
975 trap_fatal(frame, eva);
976 return (-1);
977 }
978
979 /* kludge to pass faulting virtual address to sendsig */
980 frame->tf_err = eva;
981
982 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
983}
984
985static void
986trap_fatal(frame, eva)
987 struct trapframe *frame;
988 vm_offset_t eva;
989{
990 int code, type, ss, esp;
991 struct soft_segment_descriptor softseg;
992
993 code = frame->tf_err;
994 type = frame->tf_trapno;
995 sdtossd(&gdt[IDXSEL(frame->tf_cs & 0xffff)].sd, &softseg);
996
997 if (type <= MAX_TRAP_MSG)
998 printf("\n\nFatal trap %d: %s while in %s mode\n",
999 type, trap_msg[type],
1000 frame->tf_eflags & PSL_VM ? "vm86" :
1001 ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel");
1002#ifdef SMP
1003 /* three seperate prints in case of a trap on an unmapped page */
1004 printf("mp_lock = %08x; ", mp_lock);
8a8d5d85 1005 printf("cpuid = %d; ", mycpu->gd_cpuid);
984263bc
MD
1006 printf("lapic.id = %08x\n", lapic.id);
1007#endif
1008 if (type == T_PAGEFLT) {
1009 printf("fault virtual address = 0x%x\n", eva);
1010 printf("fault code = %s %s, %s\n",
1011 code & PGEX_U ? "user" : "supervisor",
1012 code & PGEX_W ? "write" : "read",
1013 code & PGEX_P ? "protection violation" : "page not present");
1014 }
1015 printf("instruction pointer = 0x%x:0x%x\n",
1016 frame->tf_cs & 0xffff, frame->tf_eip);
1017 if ((ISPL(frame->tf_cs) == SEL_UPL) || (frame->tf_eflags & PSL_VM)) {
1018 ss = frame->tf_ss & 0xffff;
1019 esp = frame->tf_esp;
1020 } else {
1021 ss = GSEL(GDATA_SEL, SEL_KPL);
1022 esp = (int)&frame->tf_esp;
1023 }
1024 printf("stack pointer = 0x%x:0x%x\n", ss, esp);
1025 printf("frame pointer = 0x%x:0x%x\n", ss, frame->tf_ebp);
1026 printf("code segment = base 0x%x, limit 0x%x, type 0x%x\n",
1027 softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type);
1028 printf(" = DPL %d, pres %d, def32 %d, gran %d\n",
1029 softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_def32,
1030 softseg.ssd_gran);
1031 printf("processor eflags = ");
1032 if (frame->tf_eflags & PSL_T)
1033 printf("trace trap, ");
1034 if (frame->tf_eflags & PSL_I)
1035 printf("interrupt enabled, ");
1036 if (frame->tf_eflags & PSL_NT)
1037 printf("nested task, ");
1038 if (frame->tf_eflags & PSL_RF)
1039 printf("resume, ");
1040 if (frame->tf_eflags & PSL_VM)
1041 printf("vm86, ");
1042 printf("IOPL = %d\n", (frame->tf_eflags & PSL_IOPL) >> 12);
1043 printf("current process = ");
1044 if (curproc) {
1045 printf("%lu (%s)\n",
1046 (u_long)curproc->p_pid, curproc->p_comm ?
1047 curproc->p_comm : "");
1048 } else {
1049 printf("Idle\n");
1050 }
f1d1c3fa
MD
1051 printf("current thread = pri %d ", curthread->td_pri);
1052 if (curthread->td_pri >= TDPRI_CRIT)
1053 printf("(CRIT)");
1054 printf("\n");
984263bc 1055 printf("interrupt mask = ");
8f41e33b 1056 if ((curthread->td_cpl & net_imask) == net_imask)
984263bc 1057 printf("net ");
8f41e33b 1058 if ((curthread->td_cpl & tty_imask) == tty_imask)
984263bc 1059 printf("tty ");
8f41e33b 1060 if ((curthread->td_cpl & bio_imask) == bio_imask)
984263bc 1061 printf("bio ");
8f41e33b 1062 if ((curthread->td_cpl & cam_imask) == cam_imask)
984263bc 1063 printf("cam ");
8f41e33b 1064 if (curthread->td_cpl == 0)
984263bc
MD
1065 printf("none");
1066#ifdef SMP
1067/**
1068 * XXX FIXME:
1069 * we probably SHOULD have stopped the other CPUs before now!
1070 * another CPU COULD have been touching cpl at this moment...
1071 */
1072 printf(" <- SMP: XXX");
1073#endif
1074 printf("\n");
1075
1076#ifdef KDB
1077 if (kdb_trap(&psl))
1078 return;
1079#endif
1080#ifdef DDB
1081 if ((debugger_on_panic || db_active) && kdb_trap(type, 0, frame))
1082 return;
1083#endif
1084 printf("trap number = %d\n", type);
1085 if (type <= MAX_TRAP_MSG)
1086 panic("%s", trap_msg[type]);
1087 else
1088 panic("unknown/reserved trap");
1089}
1090
1091/*
1092 * Double fault handler. Called when a fault occurs while writing
1093 * a frame for a trap/exception onto the stack. This usually occurs
1094 * when the stack overflows (such is the case with infinite recursion,
1095 * for example).
1096 *
1097 * XXX Note that the current PTD gets replaced by IdlePTD when the
1098 * task switch occurs. This means that the stack that was active at
1099 * the time of the double fault is not available at <kstack> unless
1100 * the machine was idle when the double fault occurred. The downside
1101 * of this is that "trace <ebp>" in ddb won't work.
1102 */
1103void
1104dblfault_handler()
1105{
85100692 1106 struct mdglobaldata *gd = mdcpu;
17a9f566 1107
984263bc 1108 printf("\nFatal double fault:\n");
17a9f566
MD
1109 printf("eip = 0x%x\n", gd->gd_common_tss.tss_eip);
1110 printf("esp = 0x%x\n", gd->gd_common_tss.tss_esp);
1111 printf("ebp = 0x%x\n", gd->gd_common_tss.tss_ebp);
984263bc
MD
1112#ifdef SMP
1113 /* three seperate prints in case of a trap on an unmapped page */
1114 printf("mp_lock = %08x; ", mp_lock);
8a8d5d85 1115 printf("cpuid = %d; ", mycpu->gd_cpuid);
984263bc
MD
1116 printf("lapic.id = %08x\n", lapic.id);
1117#endif
1118 panic("double fault");
1119}
1120
1121/*
1122 * Compensate for 386 brain damage (missing URKR).
1123 * This is a little simpler than the pagefault handler in trap() because
1124 * it the page tables have already been faulted in and high addresses
1125 * are thrown out early for other reasons.
1126 */
1127int trapwrite(addr)
1128 unsigned addr;
1129{
1130 struct proc *p;
1131 vm_offset_t va;
1132 struct vmspace *vm;
1133 int rv;
1134
1135 va = trunc_page((vm_offset_t)addr);
1136 /*
1137 * XXX - MAX is END. Changed > to >= for temp. fix.
1138 */
1139 if (va >= VM_MAXUSER_ADDRESS)
1140 return (1);
1141
1142 p = curproc;
1143 vm = p->p_vmspace;
1144
1145 ++p->p_lock;
1146
1147 if (!grow_stack (p, va)) {
1148 --p->p_lock;
1149 return (1);
1150 }
1151
1152 /*
1153 * fault the data page
1154 */
1155 rv = vm_fault(&vm->vm_map, va, VM_PROT_WRITE, VM_FAULT_DIRTY);
1156
1157 --p->p_lock;
1158
1159 if (rv != KERN_SUCCESS)
1160 return 1;
1161
1162 return (0);
1163}
1164
1165/*
1166 * syscall2 - MP aware system call request C handler
1167 *
1168 * A system call is essentially treated as a trap except that the
1169 * MP lock is not held on entry or return. We are responsible for
1170 * obtaining the MP lock if necessary and for handling ASTs
1171 * (e.g. a task switch) prior to return.
1172 *
1173 * In general, only simple access and manipulation of curproc and
1174 * the current stack is allowed without having to hold MP lock.
1175 */
1176void
a64ba182 1177syscall2(struct trapframe frame)
984263bc 1178{
dadab5e9
MD
1179 struct thread *td = curthread;
1180 struct proc *p = td->td_proc;
984263bc
MD
1181 caddr_t params;
1182 int i;
1183 struct sysent *callp;
984263bc
MD
1184 register_t orig_tf_eflags;
1185 u_quad_t sticks;
1186 int error;
1187 int narg;
984263bc 1188 u_int code;
a64ba182 1189 union sysunion args;
984263bc
MD
1190
1191#ifdef DIAGNOSTIC
1192 if (ISPL(frame.tf_cs) != SEL_UPL) {
1193 get_mplock();
1194 panic("syscall");
1195 /* NOT REACHED */
1196 }
1197#endif
1198
8a8d5d85
MD
1199#ifdef SMP
1200 KASSERT(curthread->td_mpcount == 0, ("badmpcount syscall from %p", (void *)frame.tf_eip));
1201 get_mplock();
1202#endif
984263bc 1203 /*
41c20dac 1204 * access non-atomic field from critical section. p_sticks is
26a0694b 1205 * updated by the clock interrupt. Also use this opportunity
8a8d5d85 1206 * to lazy-raise our LWKT priority.
984263bc 1207 */
7966cb69
MD
1208 userenter(td);
1209 crit_enter_quick(td);
d16a8831 1210 sticks = curthread->td_sticks;
7966cb69 1211 crit_exit_quick(td);
984263bc
MD
1212
1213 p->p_md.md_regs = &frame;
1214 params = (caddr_t)frame.tf_esp + sizeof(int);
1215 code = frame.tf_eax;
1216 orig_tf_eflags = frame.tf_eflags;
1217
1218 if (p->p_sysent->sv_prepsyscall) {
1219 /*
1220 * The prep code is not MP aware.
1221 */
c7114eea 1222 (*p->p_sysent->sv_prepsyscall)(&frame, (int *)(&args.sysmsg + 1), &code, &params);
984263bc
MD
1223 } else {
1224 /*
1225 * Need to check if this is a 32 bit or 64 bit syscall.
1226 * fuword is MP aware.
1227 */
1228 if (code == SYS_syscall) {
1229 /*
1230 * Code is first argument, followed by actual args.
1231 */
1232 code = fuword(params);
1233 params += sizeof(int);
1234 } else if (code == SYS___syscall) {
1235 /*
1236 * Like syscall, but code is a quad, so as to maintain
1237 * quad alignment for the rest of the arguments.
1238 */
1239 code = fuword(params);
1240 params += sizeof(quad_t);
1241 }
1242 }
1243
1244 if (p->p_sysent->sv_mask)
1245 code &= p->p_sysent->sv_mask;
1246
1247 if (code >= p->p_sysent->sv_size)
1248 callp = &p->p_sysent->sv_table[0];
1249 else
1250 callp = &p->p_sysent->sv_table[code];
1251
1252 narg = callp->sy_narg & SYF_ARGMASK;
1253
1254 /*
1255 * copyin is MP aware, but the tracing code is not
1256 */
1257 if (params && (i = narg * sizeof(int)) &&
c7114eea 1258 (error = copyin(params, (caddr_t)(&args.sysmsg + 1), (u_int)i))) {
984263bc 1259#ifdef KTRACE
dadab5e9 1260 if (KTRPOINT(td, KTR_SYSCALL))
c7114eea 1261 ktrsyscall(p->p_tracep, code, narg, (void *)(&args.sysmsg + 1));
984263bc
MD
1262#endif
1263 goto bad;
1264 }
1265
8a8d5d85 1266#if 0
984263bc
MD
1267 /*
1268 * Try to run the syscall without the MP lock if the syscall
1269 * is MP safe. We have to obtain the MP lock no matter what if
1270 * we are ktracing
1271 */
1272 if ((callp->sy_narg & SYF_MPSAFE) == 0) {
1273 get_mplock();
1274 have_mplock = 1;
1275 }
8a8d5d85 1276#endif
984263bc
MD
1277
1278#ifdef KTRACE
dadab5e9 1279 if (KTRPOINT(td, KTR_SYSCALL)) {
c7114eea 1280 ktrsyscall(p->p_tracep, code, narg, (void *)(&args.sysmsg + 1));
984263bc
MD
1281 }
1282#endif
c7114eea
MD
1283 args.sysmsg_fds[0] = 0;
1284 args.sysmsg_fds[1] = frame.tf_edx;
984263bc
MD
1285
1286 STOPEVENT(p, S_SCE, narg); /* MP aware */
1287
a64ba182 1288 error = (*callp->sy_call)(&args);
984263bc
MD
1289
1290 /*
1291 * MP SAFE (we may or may not have the MP lock at this point)
1292 */
1293 switch (error) {
1294 case 0:
1295 /*
1296 * Reinitialize proc pointer `p' as it may be different
1297 * if this is a child returning from fork syscall.
1298 */
1299 p = curproc;
c7114eea
MD
1300 frame.tf_eax = args.sysmsg_fds[0];
1301 frame.tf_edx = args.sysmsg_fds[1];
984263bc
MD
1302 frame.tf_eflags &= ~PSL_C;
1303 break;
1304
1305 case ERESTART:
1306 /*
1307 * Reconstruct pc, assuming lcall $X,y is 7 bytes,
1308 * int 0x80 is 2 bytes. We saved this in tf_err.
1309 */
1310 frame.tf_eip -= frame.tf_err;
1311 break;
1312
1313 case EJUSTRETURN:
1314 break;
1315
1316 default:
1317bad:
1318 if (p->p_sysent->sv_errsize) {
1319 if (error >= p->p_sysent->sv_errsize)
1320 error = -1; /* XXX */
1321 else
1322 error = p->p_sysent->sv_errtbl[error];
1323 }
1324 frame.tf_eax = error;
1325 frame.tf_eflags |= PSL_C;
1326 break;
1327 }
1328
1329 /*
1330 * Traced syscall. trapsignal() is not MP aware.
1331 */
1332 if ((orig_tf_eflags & PSL_T) && !(orig_tf_eflags & PSL_VM)) {
984263bc
MD
1333 frame.tf_eflags &= ~PSL_T;
1334 trapsignal(p, SIGTRAP, 0);
1335 }
1336
1337 /*
1338 * Handle reschedule and other end-of-syscall issues
1339 */
8a8d5d85 1340 userret(p, &frame, sticks);
984263bc
MD
1341
1342#ifdef KTRACE
dadab5e9 1343 if (KTRPOINT(td, KTR_SYSRET)) {
c7114eea 1344 ktrsysret(p->p_tracep, code, error, args.sysmsg_result);
984263bc
MD
1345 }
1346#endif
1347
1348 /*
1349 * This works because errno is findable through the
1350 * register set. If we ever support an emulation where this
1351 * is not the case, this code will need to be revisited.
1352 */
1353 STOPEVENT(p, S_SCX, code);
1354
a2a5ad0d 1355 userexit(p);
8a8d5d85 1356#ifdef SMP
984263bc
MD
1357 /*
1358 * Release the MP lock if we had to get it
1359 */
8a8d5d85
MD
1360 KASSERT(curthread->td_mpcount == 1, ("badmpcount syscall from %p", (void *)frame.tf_eip));
1361 rel_mplock();
1362#endif
984263bc
MD
1363}
1364
a64ba182
MD
1365/*
1366 * sendsys2 - MP aware system message request C handler
1367 */
1368void
1369sendsys2(struct trapframe frame)
1370{
7966cb69 1371 struct globaldata *gd;
a64ba182
MD
1372 struct thread *td = curthread;
1373 struct proc *p = td->td_proc;
4fd10eb6 1374 register_t orig_tf_eflags;
a64ba182 1375 struct sysent *callp;
4fd10eb6
MD
1376 sysmsg_t sysmsg;
1377 lwkt_msg_t umsg;
a64ba182
MD
1378 u_quad_t sticks;
1379 int error;
1380 int narg;
4fd10eb6 1381 u_int code = 0;
a64ba182 1382 int msgsize;
90b9818c 1383 int result;
a64ba182
MD
1384
1385#ifdef DIAGNOSTIC
1386 if (ISPL(frame.tf_cs) != SEL_UPL) {
1387 get_mplock();
1388 panic("syscall");
1389 /* NOT REACHED */
1390 }
1391#endif
1392
1393#ifdef SMP
1394 KASSERT(curthread->td_mpcount == 0, ("badmpcount syscall from %p", (void *)frame.tf_eip));
1395 get_mplock();
1396#endif
1397 /*
1398 * access non-atomic field from critical section. p_sticks is
1399 * updated by the clock interrupt. Also use this opportunity
1400 * to lazy-raise our LWKT priority.
1401 */
7966cb69
MD
1402 userenter(td);
1403 crit_enter_quick(td);
a64ba182 1404 sticks = curthread->td_sticks;
7966cb69 1405 crit_exit_quick(td);
a64ba182
MD
1406
1407 p->p_md.md_regs = &frame;
4fd10eb6 1408 orig_tf_eflags = frame.tf_eflags;
90b9818c 1409 result = 0;
a64ba182
MD
1410
1411 /*
1412 * Extract the system call message. If msgsize is zero we are
4fd10eb6 1413 * blocking on a message and/or message port. YYY
c7114eea
MD
1414 *
1415 * The userland system call message size includes the size of the
1416 * userland lwkt_msg plus arguments.
a64ba182
MD
1417 */
1418 if ((msgsize = frame.tf_edx) == 0) {
4fd10eb6
MD
1419 printf("waitport %08x msg %08x\n", frame.tf_eax, frame.tf_ecx);
1420 error = ENOSYS;
1421 goto bad2;
a64ba182
MD
1422 }
1423
1424 /*
1425 * Bad message size
1426 */
c7114eea
MD
1427 if (msgsize < sizeof(struct lwkt_msg) ||
1428 msgsize > sizeof(struct lwkt_msg) + sizeof(union sysunion) - sizeof(union sysmsg)) {
a64ba182 1429 error = ENOSYS;
4fd10eb6 1430 goto bad2;
a64ba182
MD
1431 }
1432
1433 /*
4fd10eb6
MD
1434 * Obtain a sysmsg from our per-cpu cache or allocate a new one. Use
1435 * the opaque field to store the original (user) message pointer.
1436 * A critical section is necessary to interlock against interrupts
1437 * returning system messages to the thread cache.
c7114eea
MD
1438 *
1439 * The sysmsg is actually larger (i.e. sizeof(union sysunion)), in
1440 * order to hold the syscall arguments.
a64ba182 1441 */
7966cb69
MD
1442 gd = td->td_gd;
1443 crit_enter_quick(td);
1444 if ((sysmsg = gd->gd_freesysmsg) != NULL) {
c7114eea
MD
1445 gd->gd_freesysmsg = sysmsg->sm_msg.opaque.ms_sysnext;
1446 crit_exit_quick(td);
a64ba182 1447 } else {
c7114eea
MD
1448 crit_exit_quick(td);
1449 sysmsg = malloc(sizeof(union sysunion), M_SYSMSG, M_WAITOK);
a64ba182 1450 }
4fd10eb6
MD
1451
1452 /*
1453 * Copy the user request in. YYY if the userland lwkt_msg is
1454 * different from the kernel lwkt_msg, this is where we deal with
1455 * it.
1456 */
a64ba182 1457 umsg = (void *)frame.tf_ecx;
c7114eea
MD
1458 error = copyin(umsg, &sysmsg->sm_msg, sizeof(struct lwkt_msg));
1459 if (error)
4fd10eb6 1460 goto bad1;
c7114eea
MD
1461 if (msgsize > sizeof(struct lwkt_msg)) {
1462 int rsize = msgsize - sizeof(struct lwkt_msg);
1463 error = copyin(umsg + 1, sysmsg + 1, rsize);
1464 if (error)
1465 goto bad1;
1466 }
a64ba182 1467
4fd10eb6
MD
1468 /*
1469 * Initialize the parts of the message required for kernel sanity.
1470 */
c7114eea
MD
1471 sysmsg->sm_msg.opaque.ms_umsg = umsg;
1472 sysmsg->sm_msg.ms_reply_port = &td->td_msgport;
1473 sysmsg->sm_msg.ms_flags &= MSGF_ASYNC;
a64ba182 1474
4fd10eb6
MD
1475 /*
1476 * Extract the system call number, lookup the system call, and
1477 * set the default return value.
1478 */
c7114eea 1479 code = (u_int)sysmsg->sm_msg.ms_cmd;
a64ba182
MD
1480 if (code >= p->p_sysent->sv_size) {
1481 error = ENOSYS;
4fd10eb6 1482 goto bad1;
a64ba182
MD
1483 }
1484
1485 callp = &p->p_sysent->sv_table[code];
1486
c7114eea 1487 narg = (msgsize - sizeof(struct lwkt_msg)) / sizeof(register_t);
4fd10eb6 1488
a64ba182
MD
1489#ifdef KTRACE
1490 if (KTRPOINT(td, KTR_SYSCALL)) {
c7114eea 1491 ktrsyscall(p->p_tracep, code, narg, (void *)(&sysmsg + 1));
a64ba182
MD
1492 }
1493#endif
c7114eea
MD
1494 sysmsg->sm_msg.u.ms_fds[0] = 0;
1495 sysmsg->sm_msg.u.ms_fds[1] = 0;
a64ba182
MD
1496
1497 STOPEVENT(p, S_SCE, narg); /* MP aware */
1498
1499 /*
1500 * Make the system call. An error code is always returned, results
4fd10eb6
MD
1501 * are copied back via ms_result32 or ms_result64. YYY temporary
1502 * stage copy p_retval[] into ms_result32/64
a64ba182
MD
1503 *
1504 * NOTE! XXX if this is a child returning from a fork curproc
4fd10eb6
MD
1505 * might be different. YYY huh? a child returning from a fork
1506 * should never 'return' from this call, it should go right to the
1507 * fork_trampoline function.
a64ba182
MD
1508 */
1509 error = (*callp->sy_call)(sysmsg);
7966cb69 1510 gd = td->td_gd; /* RELOAD, might have switched cpus */
a64ba182 1511
4fd10eb6 1512bad1:
a64ba182 1513 /*
4fd10eb6
MD
1514 * If a synchronous return copy p_retval to ms_result64 and return
1515 * the sysmsg to the free pool.
90b9818c
MD
1516 *
1517 * YYY Don't writeback message if execve() YYY
a64ba182
MD
1518 */
1519 if (error != EASYNC) {
c7114eea 1520 result = sysmsg->sm_msg.u.ms_fds[0];
90b9818c 1521 if (error == 0 && code != SYS_execve) {
c7114eea
MD
1522 error = suword(&umsg->u.ms_result32 + 0, sysmsg->sm_msg.u.ms_fds[0]);
1523 error = suword(&umsg->u.ms_result32 + 1, sysmsg->sm_msg.u.ms_fds[1]);
90b9818c 1524 }
7966cb69 1525 crit_enter_quick(td);
c7114eea 1526 sysmsg->sm_msg.opaque.ms_sysnext = gd->gd_freesysmsg;
7966cb69
MD
1527 gd->gd_freesysmsg = sysmsg;
1528 crit_exit_quick(td);
a64ba182 1529 }
4fd10eb6 1530bad2:
a64ba182
MD
1531 frame.tf_eax = error;
1532
1533 /*
1534 * Traced syscall. trapsignal() is not MP aware.
1535 */
1536 if ((orig_tf_eflags & PSL_T) && !(orig_tf_eflags & PSL_VM)) {
1537 frame.tf_eflags &= ~PSL_T;
1538 trapsignal(p, SIGTRAP, 0);
1539 }
1540
1541 /*
1542 * Handle reschedule and other end-of-syscall issues
1543 */
1544 userret(p, &frame, sticks);
1545
1546#ifdef KTRACE
1547 if (KTRPOINT(td, KTR_SYSRET)) {
90b9818c 1548 ktrsysret(p->p_tracep, code, error, result);
a64ba182
MD
1549 }
1550#endif
1551
1552 /*
1553 * This works because errno is findable through the
1554 * register set. If we ever support an emulation where this
1555 * is not the case, this code will need to be revisited.
1556 */
1557 STOPEVENT(p, S_SCX, code);
1558
1559 userexit(p);
1560#ifdef SMP
1561 /*
1562 * Release the MP lock if we had to get it
1563 */
1564 KASSERT(curthread->td_mpcount == 1, ("badmpcount syscall from %p", (void *)frame.tf_eip));
1565 rel_mplock();
1566#endif
1567}
1568
984263bc
MD
1569/*
1570 * Simplified back end of syscall(), used when returning from fork()
8a8d5d85
MD
1571 * directly into user mode. MP lock is held on entry and should be
1572 * released on return. This code will return back into the fork
1573 * trampoline code which then runs doreti.
984263bc
MD
1574 */
1575void
1576fork_return(p, frame)
1577 struct proc *p;
1578 struct trapframe frame;
1579{
1580 frame.tf_eax = 0; /* Child returns zero */
1581 frame.tf_eflags &= ~PSL_C; /* success */
1582 frame.tf_edx = 1;
1583
8a8d5d85 1584 userret(p, &frame, 0);
984263bc 1585#ifdef KTRACE
dadab5e9 1586 if (KTRPOINT(p->p_thread, KTR_SYSRET))
984263bc
MD
1587 ktrsysret(p->p_tracep, SYS_fork, 0, 0);
1588#endif
d9eea1a5 1589 p->p_flag |= P_PASSIVE_ACQ;
a2a5ad0d 1590 userexit(p);
d9eea1a5 1591 p->p_flag &= ~P_PASSIVE_ACQ;
8a8d5d85
MD
1592#ifdef SMP
1593 KKASSERT(curthread->td_mpcount == 1);
1594 rel_mplock();
1595#endif
984263bc 1596}
8a8d5d85 1597