AMD64 - Fix many compile-time warnings. int/ptr type mismatches, %llx, etc.
[dragonfly.git] / sys / platform / pc64 / amd64 / trap.c
CommitLineData
d7f50089 1/*-
d7f50089
YY
2 * Copyright (c) 1990, 1993
3 * The Regents of the University of California. All rights reserved.
c8fe38ae
MD
4 * Copyright (C) 1994, David Greenman
5 * Copyright (c) 2008 The DragonFly Project.
6 * Copyright (c) 2008 Jordan Gordeev.
d7f50089
YY
7 *
8 * This code is derived from software contributed to Berkeley by
9 * the University of Utah, and William Jolitz.
10 *
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
13 * are met:
14 * 1. Redistributions of source code must retain the above copyright
15 * notice, this list of conditions and the following disclaimer.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
19 * 3. All advertising materials mentioning features or use of this software
20 * must display the following acknowledgement:
21 * This product includes software developed by the University of
22 * California, Berkeley and its contributors.
23 * 4. Neither the name of the University nor the names of its contributors
24 * may be used to endorse or promote products derived from this software
25 * without specific prior written permission.
26 *
27 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
28 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
29 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
30 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
31 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
32 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
33 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
34 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
35 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
36 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
37 * SUCH DAMAGE.
38 *
c8fe38ae 39 * from: @(#)trap.c 7.4 (Berkeley) 5/13/91
d7f50089 40 * $FreeBSD: src/sys/i386/i386/trap.c,v 1.147.2.11 2003/02/27 19:09:59 luoqi Exp $
c730be20 41 * $DragonFly: src/sys/platform/pc64/amd64/trap.c,v 1.3 2008/09/09 04:06:18 dillon Exp $
d7f50089
YY
42 */
43
c8fe38ae
MD
44/*
45 * AMD64 Trap and System call handling
46 */
47
48#include "opt_ddb.h"
49#include "opt_ktrace.h"
d7f50089
YY
50
51#include <machine/frame.h>
c8fe38ae
MD
52#include <sys/param.h>
53#include <sys/systm.h>
54#include <sys/kernel.h>
55#include <sys/proc.h>
56#include <sys/pioctl.h>
57#include <sys/types.h>
58#include <sys/signal2.h>
59#include <sys/syscall.h>
60#include <sys/sysctl.h>
61#include <sys/sysent.h>
62#include <sys/systm.h>
63#ifdef KTRACE
64#include <sys/ktrace.h>
65#endif
66#include <sys/ktr.h>
67#include <sys/sysmsg.h>
68#include <sys/sysproto.h>
69#include <sys/sysunion.h>
70
71#include <vm/pmap.h>
72#include <vm/vm.h>
73#include <vm/vm_extern.h>
74#include <vm/vm_kern.h>
75#include <vm/vm_param.h>
76#include <machine/cpu.h>
77#include <machine/pcb.h>
78#include <machine/thread.h>
79#include <machine/vmparam.h>
80#include <machine/md_var.h>
81
82#include <ddb/ddb.h>
973c11b9 83#include <sys/thread2.h>
c8fe38ae
MD
84
85#ifdef SMP
86
87#define MAKEMPSAFE(have_mplock) \
88 if (have_mplock == 0) { \
89 get_mplock(); \
90 have_mplock = 1; \
91 }
92
93#else
94
95#define MAKEMPSAFE(have_mplock)
96
97#endif
98
99extern void trap(struct trapframe *frame);
c8fe38ae
MD
100
101static int trap_pfault(struct trapframe *, int);
102static void trap_fatal(struct trapframe *, vm_offset_t);
103void dblfault_handler(struct trapframe *frame);
104
105#define PCPU_GET(member) ((mycpu)->gd_##member)
106#define PCPU_INC(member) ((mycpu)->gd_##member)++
107
108#define MAX_TRAP_MSG 30
109static char *trap_msg[] = {
110 "", /* 0 unused */
111 "privileged instruction fault", /* 1 T_PRIVINFLT */
112 "", /* 2 unused */
113 "breakpoint instruction fault", /* 3 T_BPTFLT */
114 "", /* 4 unused */
115 "", /* 5 unused */
116 "arithmetic trap", /* 6 T_ARITHTRAP */
117 "system forced exception", /* 7 T_ASTFLT */
118 "", /* 8 unused */
119 "general protection fault", /* 9 T_PROTFLT */
120 "trace trap", /* 10 T_TRCTRAP */
121 "", /* 11 unused */
122 "page fault", /* 12 T_PAGEFLT */
123 "", /* 13 unused */
124 "alignment fault", /* 14 T_ALIGNFLT */
125 "", /* 15 unused */
126 "", /* 16 unused */
127 "", /* 17 unused */
128 "integer divide fault", /* 18 T_DIVIDE */
129 "non-maskable interrupt trap", /* 19 T_NMI */
130 "overflow trap", /* 20 T_OFLOW */
131 "FPU bounds check fault", /* 21 T_BOUND */
132 "FPU device not available", /* 22 T_DNA */
133 "double fault", /* 23 T_DOUBLEFLT */
134 "FPU operand fetch fault", /* 24 T_FPOPFLT */
135 "invalid TSS fault", /* 25 T_TSSFLT */
136 "segment not present fault", /* 26 T_SEGNPFLT */
137 "stack fault", /* 27 T_STKFLT */
138 "machine check trap", /* 28 T_MCHK */
139 "SIMD floating-point exception", /* 29 T_XMMFLT */
140 "reserved (unknown) fault", /* 30 T_RESERVED */
141};
142
143#ifdef DDB
144static int ddb_on_nmi = 1;
145SYSCTL_INT(_machdep, OID_AUTO, ddb_on_nmi, CTLFLAG_RW,
146 &ddb_on_nmi, 0, "Go to DDB on NMI");
147#endif
148static int panic_on_nmi = 1;
149SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RW,
150 &panic_on_nmi, 0, "Panic on NMI");
151static int fast_release;
152SYSCTL_INT(_machdep, OID_AUTO, fast_release, CTLFLAG_RW,
153 &fast_release, 0, "Passive Release was optimal");
154static int slow_release;
155SYSCTL_INT(_machdep, OID_AUTO, slow_release, CTLFLAG_RW,
156 &slow_release, 0, "Passive Release was nonoptimal");
157#ifdef SMP
158static int syscall_mpsafe = 1;
159SYSCTL_INT(_kern, OID_AUTO, syscall_mpsafe, CTLFLAG_RW,
160 &syscall_mpsafe, 0, "Allow MPSAFE marked syscalls to run without BGL");
161TUNABLE_INT("kern.syscall_mpsafe", &syscall_mpsafe);
162static int trap_mpsafe = 1;
163SYSCTL_INT(_kern, OID_AUTO, trap_mpsafe, CTLFLAG_RW,
164 &trap_mpsafe, 0, "Allow traps to mostly run without the BGL");
165TUNABLE_INT("kern.trap_mpsafe", &trap_mpsafe);
166#endif
167
c8fe38ae
MD
168/*
169 * userenter() passively intercepts the thread switch function to increase
170 * the thread priority from a user priority to a kernel priority, reducing
171 * syscall and trap overhead for the case where no switch occurs.
172 */
173
174static __inline void
175userenter(struct thread *curtd)
176{
3824f392 177 curtd->td_release = lwkt_passive_release;
c8fe38ae
MD
178}
179
180/*
181 * Handle signals, upcalls, profiling, and other AST's and/or tasks that
182 * must be completed before we can return to or try to return to userland.
183 *
184 * Note that td_sticks is a 64 bit quantity, but there's no point doing 64
185 * arithmatic on the delta calculation so the absolute tick values are
186 * truncated to an integer.
187 */
188static void
189userret(struct lwp *lp, struct trapframe *frame, int sticks)
190{
191 struct proc *p = lp->lwp_proc;
192 int sig;
193
194 /*
195 * Charge system time if profiling. Note: times are in microseconds.
196 * This may do a copyout and block, so do it first even though it
197 * means some system time will be charged as user time.
198 */
199 if (p->p_flag & P_PROFIL) {
200 addupc_task(p, frame->tf_rip,
201 (u_int)((int)lp->lwp_thread->td_sticks - sticks));
202 }
203
204recheck:
205 /*
206 * If the jungle wants us dead, so be it.
207 */
208 if (lp->lwp_flag & LWP_WEXIT) {
209 get_mplock();
210 lwp_exit(0);
211 rel_mplock(); /* NOT REACHED */
212 }
213
214 /*
215 * Block here if we are in a stopped state.
216 */
217 if (p->p_stat == SSTOP) {
218 get_mplock();
219 tstop();
220 rel_mplock();
221 goto recheck;
222 }
223
224 /*
225 * Post any pending upcalls. If running a virtual kernel be sure
226 * to restore the virtual kernel's vmspace before posting the upcall.
227 */
228 if (p->p_flag & P_UPCALLPEND) {
229 p->p_flag &= ~P_UPCALLPEND;
230 get_mplock();
231 postupcall(lp);
232 rel_mplock();
233 goto recheck;
234 }
235
236 /*
237 * Post any pending signals. If running a virtual kernel be sure
238 * to restore the virtual kernel's vmspace before posting the signal.
239 */
e473f776 240 if ((sig = CURSIG_TRACE(lp)) != 0) {
c8fe38ae
MD
241 get_mplock();
242 postsig(sig);
243 rel_mplock();
244 goto recheck;
245 }
246
247 /*
248 * block here if we are swapped out, but still process signals
249 * (such as SIGKILL). proc0 (the swapin scheduler) is already
250 * aware of our situation, we do not have to wake it up.
251 */
252 if (p->p_flag & P_SWAPPEDOUT) {
253 get_mplock();
254 p->p_flag |= P_SWAPWAIT;
255 swapin_request();
256 if (p->p_flag & P_SWAPWAIT)
257 tsleep(p, PCATCH, "SWOUT", 0);
258 p->p_flag &= ~P_SWAPWAIT;
259 rel_mplock();
260 goto recheck;
261 }
262
263 /*
264 * Make sure postsig() handled request to restore old signal mask after
265 * running signal handler.
266 */
267 KKASSERT((lp->lwp_flag & LWP_OLDMASK) == 0);
268}
269
270/*
271 * Cleanup from userenter and any passive release that might have occured.
272 * We must reclaim the current-process designation before we can return
273 * to usermode. We also handle both LWKT and USER reschedule requests.
274 */
275static __inline void
276userexit(struct lwp *lp)
277{
278 struct thread *td = lp->lwp_thread;
973c11b9 279/* globaldata_t gd = td->td_gd;*/
c8fe38ae 280
c8fe38ae 281 /*
89ffa1cf
SS
282 * Handle stop requests at kernel priority. Any requests queued
283 * after this loop will generate another AST.
c8fe38ae 284 */
89ffa1cf
SS
285 while (lp->lwp_proc->p_stat == SSTOP) {
286 get_mplock();
287 tstop();
288 rel_mplock();
c8fe38ae
MD
289 }
290
291 /*
c8fe38ae
MD
292 * Reduce our priority in preparation for a return to userland. If
293 * our passive release function was still in place, our priority was
294 * never raised and does not need to be reduced.
295 */
3824f392 296 lwkt_passive_recover(td);
89ffa1cf
SS
297
298 /*
299 * Become the current user scheduled process if we aren't already,
300 * and deal with reschedule requests and other factors.
301 */
302 lp->lwp_proc->p_usched->acquire_curproc(lp);
303 /* WARNING: we may have migrated cpu's */
304 /* gd = td->td_gd; */
c8fe38ae
MD
305}
306
0855a2af
JG
307#if !defined(KTR_KERNENTRY)
308#define KTR_KERNENTRY KTR_ALL
309#endif
310KTR_INFO_MASTER(kernentry);
311KTR_INFO(KTR_KERNENTRY, kernentry, trap, 0, "STR",
312 sizeof(long) + sizeof(long) + sizeof(long) + sizeof(vm_offset_t));
313KTR_INFO(KTR_KERNENTRY, kernentry, trap_ret, 0, "STR",
314 sizeof(long) + sizeof(long));
315KTR_INFO(KTR_KERNENTRY, kernentry, syscall, 0, "STR",
316 sizeof(long) + sizeof(long) + sizeof(long));
317KTR_INFO(KTR_KERNENTRY, kernentry, syscall_ret, 0, "STR",
318 sizeof(long) + sizeof(long) + sizeof(long));
319KTR_INFO(KTR_KERNENTRY, kernentry, fork_ret, 0, "STR",
320 sizeof(long) + sizeof(long));
321
c8fe38ae
MD
322/*
323 * Exception, fault, and trap interface to the kernel.
324 * This common code is called from assembly language IDT gate entry
325 * routines that prepare a suitable stack frame, and restore this
326 * frame after the exception has been processed.
327 *
328 * This function is also called from doreti in an interlock to handle ASTs.
329 * For example: hardwareint->INTROUTINE->(set ast)->doreti->trap
330 *
331 * NOTE! We have to retrieve the fault address prior to obtaining the
332 * MP lock because get_mplock() may switch out. YYY cr2 really ought
333 * to be retrieved by the assembly code, not here.
334 *
335 * XXX gd_trap_nesting_level currently prevents lwkt_switch() from panicing
336 * if an attempt is made to switch from a fast interrupt or IPI. This is
337 * necessary to properly take fatal kernel traps on SMP machines if
338 * get_mplock() has to block.
339 */
340
341void
342trap(struct trapframe *frame)
343{
344 struct globaldata *gd = mycpu;
345 struct thread *td = gd->gd_curthread;
346 struct lwp *lp = td->td_lwp;
347 struct proc *p;
348 int sticks = 0;
349 int i = 0, ucode = 0, type, code;
350#ifdef SMP
351 int have_mplock = 0;
352#endif
353#ifdef INVARIANTS
354 int crit_count = td->td_pri & ~TDPRI_MASK;
355#endif
356 vm_offset_t eva;
357
358 p = td->td_proc;
359
360#ifndef JG
361 kprintf0("TRAP ");
362 kprintf0("\"%s\" type=%ld\n",
363 trap_msg[frame->tf_trapno], frame->tf_trapno);
364 kprintf0(" rip=%lx rsp=%lx\n", frame->tf_rip, frame->tf_rsp);
365 kprintf0(" err=%lx addr=%lx\n", frame->tf_err, frame->tf_addr);
366 kprintf0(" cs=%lx ss=%lx rflags=%lx\n", (unsigned long)frame->tf_cs, (unsigned long)frame->tf_ss, frame->tf_rflags);
367#endif
368
369#ifdef DDB
370 if (db_active) {
371 ++gd->gd_trap_nesting_level;
372 MAKEMPSAFE(have_mplock);
373 trap_fatal(frame, frame->tf_addr);
374 --gd->gd_trap_nesting_level;
375 goto out2;
376 }
377#endif
378#ifdef DDB
379 if (db_active) {
380 eva = (frame->tf_trapno == T_PAGEFLT ? frame->tf_addr : 0);
381 ++gd->gd_trap_nesting_level;
382 MAKEMPSAFE(have_mplock);
383 trap_fatal(frame, eva);
384 --gd->gd_trap_nesting_level;
385 goto out2;
386 }
387#endif
388
389 eva = 0;
390
391#ifdef SMP
392 if (trap_mpsafe == 0) {
393 ++gd->gd_trap_nesting_level;
394 MAKEMPSAFE(have_mplock);
395 --gd->gd_trap_nesting_level;
396 }
397#endif
398
399 if ((frame->tf_rflags & PSL_I) == 0) {
400 /*
401 * Buggy application or kernel code has disabled interrupts
402 * and then trapped. Enabling interrupts now is wrong, but
403 * it is better than running with interrupts disabled until
404 * they are accidentally enabled later.
405 */
406 type = frame->tf_trapno;
407 if (ISPL(frame->tf_cs) == SEL_UPL) {
408 MAKEMPSAFE(have_mplock);
409 /* JG curproc can be NULL */
410 kprintf(
411 "pid %ld (%s): trap %d with interrupts disabled\n",
412 (long)curproc->p_pid, curproc->p_comm, type);
413 } else if (type != T_NMI && type != T_BPTFLT &&
414 type != T_TRCTRAP) {
415 /*
416 * XXX not quite right, since this may be for a
417 * multiple fault in user mode.
418 */
419 MAKEMPSAFE(have_mplock);
420 kprintf("kernel trap %d with interrupts disabled\n",
421 type);
422 }
423 cpu_enable_intr();
424 }
425
426 type = frame->tf_trapno;
427 code = frame->tf_err;
428
429 if (ISPL(frame->tf_cs) == SEL_UPL) {
430 /* user trap */
431
432 KTR_LOG(kernentry_trap, p->p_pid, lp->lwp_tid,
433 frame->tf_trapno, eva);
434
435 userenter(td);
436
437 sticks = (int)td->td_sticks;
438 lp->lwp_md.md_regs = frame;
439
440 switch (type) {
441 case T_PRIVINFLT: /* privileged instruction fault */
442 ucode = ILL_PRVOPC;
443 i = SIGILL;
444 break;
445
446 case T_BPTFLT: /* bpt instruction fault */
447 case T_TRCTRAP: /* trace trap */
448 frame->tf_rflags &= ~PSL_T;
449 i = SIGTRAP;
450 break;
451
452 case T_ARITHTRAP: /* arithmetic trap */
453 ucode = code;
454 i = SIGFPE;
455#if 0
456#if JG
457 ucode = fputrap();
458#else
459 ucode = code;
460#endif
461 i = SIGFPE;
462#endif
463 break;
464
465 case T_ASTFLT: /* Allow process switch */
466 mycpu->gd_cnt.v_soft++;
467 if (mycpu->gd_reqflags & RQF_AST_OWEUPC) {
468 atomic_clear_int_nonlocked(&mycpu->gd_reqflags,
469 RQF_AST_OWEUPC);
470 addupc_task(p, p->p_prof.pr_addr,
471 p->p_prof.pr_ticks);
472 }
473 goto out;
474
475 case T_PROTFLT: /* general protection fault */
476 case T_SEGNPFLT: /* segment not present fault */
477 case T_TSSFLT: /* invalid TSS fault */
478 case T_DOUBLEFLT: /* double fault */
479 default:
480 ucode = code + BUS_SEGM_FAULT ;
481 i = SIGBUS;
482 break;
483
484 case T_PAGEFLT: /* page fault */
485 MAKEMPSAFE(have_mplock);
486 i = trap_pfault(frame, TRUE);
c8fe38ae 487 if (frame->tf_rip == 0)
973c11b9 488 kprintf("T_PAGEFLT: Warning %%rip == 0!\n");
c8fe38ae
MD
489 if (i == -1)
490 goto out;
491 if (i == 0)
492 goto out;
493
494 ucode = T_PAGEFLT;
495 break;
496
497 case T_DIVIDE: /* integer divide fault */
498 ucode = FPE_INTDIV;
499 i = SIGFPE;
500 break;
501
502 case T_NMI:
503 MAKEMPSAFE(have_mplock);
504 /* machine/parity/power fail/"kitchen sink" faults */
505 if (isa_nmi(code) == 0) {
506#ifdef DDB
507 /*
508 * NMI can be hooked up to a pushbutton
509 * for debugging.
510 */
511 if (ddb_on_nmi) {
512 kprintf ("NMI ... going to debugger\n");
513 kdb_trap(type, 0, frame);
514 }
515#endif /* DDB */
516 goto out2;
517 } else if (panic_on_nmi)
518 panic("NMI indicates hardware failure");
519 break;
520
521 case T_OFLOW: /* integer overflow fault */
522 ucode = FPE_INTOVF;
523 i = SIGFPE;
524 break;
525
526 case T_BOUND: /* bounds check fault */
527 ucode = FPE_FLTSUB;
528 i = SIGFPE;
529 break;
530
531 case T_DNA:
532 /*
533 * Virtual kernel intercept - pass the DNA exception
534 * to the virtual kernel if it asked to handle it.
535 * This occurs when the virtual kernel is holding
536 * onto the FP context for a different emulated
537 * process then the one currently running.
538 *
539 * We must still call npxdna() since we may have
540 * saved FP state that the virtual kernel needs
541 * to hand over to a different emulated process.
542 */
543 if (lp->lwp_vkernel && lp->lwp_vkernel->ve &&
544 (td->td_pcb->pcb_flags & FP_VIRTFP)
545 ) {
546 npxdna();
547 break;
548 }
549
550 /*
551 * The kernel may have switched out the FP unit's
552 * state, causing the user process to take a fault
553 * when it tries to use the FP unit. Restore the
554 * state here
555 */
556 if (npxdna())
557 goto out;
558 i = SIGFPE;
559 ucode = FPE_FPU_NP_TRAP;
560 break;
561
562 case T_FPOPFLT: /* FPU operand fetch fault */
563 ucode = T_FPOPFLT;
564 i = SIGILL;
565 break;
566
567 case T_XMMFLT: /* SIMD floating-point exception */
568 ucode = 0; /* XXX */
569 i = SIGFPE;
570 break;
571 }
572 } else {
573 /* kernel trap */
574
575 switch (type) {
576 case T_PAGEFLT: /* page fault */
577 MAKEMPSAFE(have_mplock);
578 trap_pfault(frame, FALSE);
579 goto out2;
580
581 case T_DNA:
582 /*
583 * The kernel is apparently using fpu for copying.
584 * XXX this should be fatal unless the kernel has
585 * registered such use.
586 */
587 if (npxdna())
588 goto out2;
589 break;
590
591 case T_STKFLT: /* stack fault */
592 break;
593
594 case T_PROTFLT: /* general protection fault */
595 case T_SEGNPFLT: /* segment not present fault */
596 /*
597 * Invalid segment selectors and out of bounds
598 * %rip's and %rsp's can be set up in user mode.
599 * This causes a fault in kernel mode when the
600 * kernel tries to return to user mode. We want
601 * to get this fault so that we can fix the
602 * problem here and not have to check all the
603 * selectors and pointers when the user changes
604 * them.
605 */
606 kprintf0("trap.c line %d\n", __LINE__);
607 if (mycpu->gd_intr_nesting_level == 0) {
608 if (td->td_pcb->pcb_onfault) {
609 frame->tf_rip = (register_t)
610 td->td_pcb->pcb_onfault;
611 goto out2;
612 }
89ffa1cf
SS
613 if (frame->tf_rip == (long)doreti_iret) {
614 frame->tf_rip = (long)doreti_iret_fault;
615 goto out2;
616 }
c8fe38ae
MD
617 }
618 break;
619
620 case T_TSSFLT:
621 /*
622 * PSL_NT can be set in user mode and isn't cleared
623 * automatically when the kernel is entered. This
624 * causes a TSS fault when the kernel attempts to
625 * `iret' because the TSS link is uninitialized. We
626 * want to get this fault so that we can fix the
627 * problem here and not every time the kernel is
628 * entered.
629 */
630 if (frame->tf_rflags & PSL_NT) {
631 frame->tf_rflags &= ~PSL_NT;
632 goto out2;
633 }
634 break;
635
636 case T_TRCTRAP: /* trace trap */
637#if 0
638 if (frame->tf_rip == (int)IDTVEC(syscall)) {
639 /*
640 * We've just entered system mode via the
641 * syscall lcall. Continue single stepping
642 * silently until the syscall handler has
643 * saved the flags.
644 */
645 goto out2;
646 }
647 if (frame->tf_rip == (int)IDTVEC(syscall) + 1) {
648 /*
649 * The syscall handler has now saved the
650 * flags. Stop single stepping it.
651 */
652 frame->tf_rflags &= ~PSL_T;
653 goto out2;
654 }
655#endif
656
657 /*
658 * Ignore debug register trace traps due to
659 * accesses in the user's address space, which
660 * can happen under several conditions such as
661 * if a user sets a watchpoint on a buffer and
662 * then passes that buffer to a system call.
663 * We still want to get TRCTRAPS for addresses
664 * in kernel space because that is useful when
665 * debugging the kernel.
666 */
667#if JG
668 if (user_dbreg_trap()) {
669 /*
670 * Reset breakpoint bits because the
671 * processor doesn't
672 */
673 /* XXX check upper bits here */
674 load_dr6(rdr6() & 0xfffffff0);
675 goto out2;
676 }
677#endif
678 /*
679 * FALLTHROUGH (TRCTRAP kernel mode, kernel address)
680 */
681 case T_BPTFLT:
682 /*
683 * If DDB is enabled, let it handle the debugger trap.
684 * Otherwise, debugger traps "can't happen".
685 */
686#ifdef DDB
687 MAKEMPSAFE(have_mplock);
688 if (kdb_trap(type, 0, frame))
689 goto out2;
690#endif
691 break;
692
693 case T_NMI:
694 MAKEMPSAFE(have_mplock);
695 /* machine/parity/power fail/"kitchen sink" faults */
696#if NISA > 0
697 if (isa_nmi(code) == 0) {
698#ifdef DDB
699 /*
700 * NMI can be hooked up to a pushbutton
701 * for debugging.
702 */
703 if (ddb_on_nmi) {
704 kprintf ("NMI ... going to debugger\n");
705 kdb_trap(type, 0, frame);
706 }
707#endif /* DDB */
708 goto out2;
709 } else if (panic_on_nmi == 0)
710 goto out2;
711 /* FALL THROUGH */
712#endif /* NISA > 0 */
713 }
714 MAKEMPSAFE(have_mplock);
715 trap_fatal(frame, 0);
716 goto out2;
717 }
718
719 /*
720 * Virtual kernel intercept - if the fault is directly related to a
721 * VM context managed by a virtual kernel then let the virtual kernel
722 * handle it.
723 */
724 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
725 vkernel_trap(lp, frame);
726 goto out2;
727 }
728
729 /*
730 * Virtual kernel intercept - if the fault is directly related to a
731 * VM context managed by a virtual kernel then let the virtual kernel
732 * handle it.
733 */
734 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
735 vkernel_trap(lp, frame);
736 goto out;
737 }
738
739 /*
740 * Translate fault for emulators (e.g. Linux)
741 */
742 if (*p->p_sysent->sv_transtrap)
743 i = (*p->p_sysent->sv_transtrap)(i, type);
744
745 MAKEMPSAFE(have_mplock);
746 trapsignal(lp, i, ucode);
747
748#ifdef DEBUG
749 if (type <= MAX_TRAP_MSG) {
750 uprintf("fatal process exception: %s",
751 trap_msg[type]);
752 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
753 uprintf(", fault VA = 0x%lx", frame->tf_addr);
754 uprintf("\n");
755 }
756#endif
757
758out:
759#ifdef SMP
760 if (ISPL(frame->tf_cs) == SEL_UPL)
761 KASSERT(td->td_mpcount == have_mplock, ("badmpcount trap/end from %p", (void *)frame->tf_rip));
762#endif
763 userret(lp, frame, sticks);
764 userexit(lp);
765out2: ;
766#ifdef SMP
767 if (have_mplock)
768 rel_mplock();
769#endif
770 if (p != NULL && lp != NULL)
771 KTR_LOG(kernentry_trap_ret, p->p_pid, lp->lwp_tid);
772#ifdef INVARIANTS
773 KASSERT(crit_count == (td->td_pri & ~TDPRI_MASK),
774 ("syscall: critical section count mismatch! %d/%d",
775 crit_count / TDPRI_CRIT, td->td_pri / TDPRI_CRIT));
776#endif
777}
778
779static int
780trap_pfault(struct trapframe *frame, int usermode)
781{
782 vm_offset_t va;
783 struct vmspace *vm = NULL;
784 vm_map_t map;
785 int rv = 0;
786 vm_prot_t ftype;
787 thread_t td = curthread;
788 struct lwp *lp = td->td_lwp;
789
790 va = trunc_page(frame->tf_addr);
791 if (va >= VM_MIN_KERNEL_ADDRESS) {
792 /*
793 * Don't allow user-mode faults in kernel address space.
794 */
795 if (usermode)
796 goto nogo;
797
798 map = &kernel_map;
799 } else {
800 /*
801 * This is a fault on non-kernel virtual memory.
802 * vm is initialized above to NULL. If curproc is NULL
803 * or curproc->p_vmspace is NULL the fault is fatal.
804 */
805 if (lp != NULL)
806 vm = lp->lwp_vmspace;
807
808 if (vm == NULL)
809 goto nogo;
810
811 map = &vm->vm_map;
812 }
813
814 /*
815 * PGEX_I is defined only if the execute disable bit capability is
816 * supported and enabled.
817 */
818 if (frame->tf_err & PGEX_W)
819 ftype = VM_PROT_WRITE;
820#if JG
821 else if ((frame->tf_err & PGEX_I) && pg_nx != 0)
822 ftype = VM_PROT_EXECUTE;
823#endif
824 else
825 ftype = VM_PROT_READ;
826
827 if (map != &kernel_map) {
828 /*
829 * Keep swapout from messing with us during this
830 * critical time.
831 */
832 PHOLD(lp->lwp_proc);
833
834 /*
835 * Grow the stack if necessary
836 */
837 /* grow_stack returns false only if va falls into
838 * a growable stack region and the stack growth
839 * fails. It returns true if va was not within
840 * a growable stack region, or if the stack
841 * growth succeeded.
842 */
843 if (!grow_stack(lp->lwp_proc, va)) {
844 rv = KERN_FAILURE;
845 PRELE(lp->lwp_proc);
846 goto nogo;
847 }
848
849 /* Fault in the user page: */
850 rv = vm_fault(map, va, ftype,
851 (ftype & VM_PROT_WRITE) ? VM_FAULT_DIRTY
852 : VM_FAULT_NORMAL);
853
854 PRELE(lp->lwp_proc);
855 } else {
856 /*
857 * Don't have to worry about process locking or stacks
858 * in the kernel.
859 */
860 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
861 }
862
863 if (rv == KERN_SUCCESS)
864 return (0);
865nogo:
866 if (!usermode) {
867 if (td->td_gd->gd_intr_nesting_level == 0 &&
868 td->td_pcb->pcb_onfault) {
869 frame->tf_rip = (register_t)td->td_pcb->pcb_onfault;
870 return (0);
871 }
872 trap_fatal(frame, frame->tf_addr);
873 return (-1);
874 }
875
876 /*
877 * NOTE: on amd64 we have a tf_addr field in the trapframe, no
878 * kludge is needed to pass the fault address to signal handlers.
879 */
5b9f6cc4 880 kprintf("seg-fault accessing address %p ip=%p\n",
973c11b9 881 (void *)va, (void *)frame->tf_rip);
f741dc5a 882 /* Debugger("seg-fault"); */
c8fe38ae
MD
883
884 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
885}
886
887static void
888trap_fatal(struct trapframe *frame, vm_offset_t eva)
889{
890 int code, ss;
891 u_int type;
892 long rsp;
893 struct soft_segment_descriptor softseg;
894 char *msg;
895
896 code = frame->tf_err;
897 type = frame->tf_trapno;
898 sdtossd(&gdt[IDXSEL(frame->tf_cs & 0xffff)], &softseg);
899
900 if (type <= MAX_TRAP_MSG)
901 msg = trap_msg[type];
902 else
903 msg = "UNKNOWN";
904 kprintf("\n\nFatal trap %d: %s while in %s mode\n", type, msg,
905 ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel");
906#ifdef SMP
907 /* two separate prints in case of a trap on an unmapped page */
908 kprintf("cpuid = %d; ", PCPU_GET(cpuid));
909 kprintf("apic id = %02x\n", PCPU_GET(apic_id));
910#endif
911 if (type == T_PAGEFLT) {
912 kprintf("fault virtual address = 0x%lx\n", eva);
913 kprintf("fault code = %s %s %s, %s\n",
914 code & PGEX_U ? "user" : "supervisor",
915 code & PGEX_W ? "write" : "read",
916 code & PGEX_I ? "instruction" : "data",
917 code & PGEX_P ? "protection violation" : "page not present");
918 }
919 kprintf("instruction pointer = 0x%lx:0x%lx\n",
920 frame->tf_cs & 0xffff, frame->tf_rip);
921 if (ISPL(frame->tf_cs) == SEL_UPL) {
922 ss = frame->tf_ss & 0xffff;
923 rsp = frame->tf_rsp;
924 } else {
925 ss = GSEL(GDATA_SEL, SEL_KPL);
926 rsp = (long)&frame->tf_rsp;
927 }
928 kprintf("stack pointer = 0x%x:0x%lx\n", ss, rsp);
929 kprintf("frame pointer = 0x%x:0x%lx\n", ss, frame->tf_rbp);
930 kprintf("code segment = base 0x%lx, limit 0x%lx, type 0x%x\n",
931 softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type);
932 kprintf(" = DPL %d, pres %d, long %d, def32 %d, gran %d\n",
933 softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_long, softseg.ssd_def32,
934 softseg.ssd_gran);
935 kprintf("processor eflags = ");
936 if (frame->tf_rflags & PSL_T)
937 kprintf("trace trap, ");
938 if (frame->tf_rflags & PSL_I)
939 kprintf("interrupt enabled, ");
940 if (frame->tf_rflags & PSL_NT)
941 kprintf("nested task, ");
942 if (frame->tf_rflags & PSL_RF)
943 kprintf("resume, ");
944 kprintf("IOPL = %ld\n", (frame->tf_rflags & PSL_IOPL) >> 12);
945 kprintf("current process = ");
946 if (curproc) {
947 kprintf("%lu\n",
948 (u_long)curproc->p_pid);
949 } else {
950 kprintf("Idle\n");
951 }
952 kprintf("current thread = pri %d ", curthread->td_pri);
953 if (curthread->td_pri >= TDPRI_CRIT)
954 kprintf("(CRIT)");
955 kprintf("\n");
956
957#ifdef DDB
958 if ((debugger_on_panic || db_active) && kdb_trap(type, code, frame))
959 return;
960#endif
961 kprintf("trap number = %d\n", type);
962 if (type <= MAX_TRAP_MSG)
963 panic("%s", trap_msg[type]);
964 else
965 panic("unknown/reserved trap");
966}
967
968/*
969 * Double fault handler. Called when a fault occurs while writing
970 * a frame for a trap/exception onto the stack. This usually occurs
971 * when the stack overflows (such is the case with infinite recursion,
972 * for example).
973 */
974void
975dblfault_handler(struct trapframe *frame)
976{
977 kprintf0("DOUBLE FAULT\n");
978 kprintf("\nFatal double fault\n");
979 kprintf("rip = 0x%lx\n", frame->tf_rip);
980 kprintf("rsp = 0x%lx\n", frame->tf_rsp);
981 kprintf("rbp = 0x%lx\n", frame->tf_rbp);
982#ifdef SMP
983 /* two separate prints in case of a trap on an unmapped page */
984 kprintf("cpuid = %d; ", PCPU_GET(cpuid));
985 kprintf("apic id = %02x\n", PCPU_GET(apic_id));
986#endif
987 panic("double fault");
988}
989
990/*
991 * syscall2 - MP aware system call request C handler
992 *
993 * A system call is essentially treated as a trap except that the
994 * MP lock is not held on entry or return. We are responsible for
995 * obtaining the MP lock if necessary and for handling ASTs
996 * (e.g. a task switch) prior to return.
997 *
998 * In general, only simple access and manipulation of curproc and
999 * the current stack is allowed without having to hold MP lock.
1000 *
1001 * MPSAFE - note that large sections of this routine are run without
1002 * the MP lock.
1003 */
1004void
1005syscall2(struct trapframe *frame)
1006{
1007 struct thread *td = curthread;
1008 struct proc *p = td->td_proc;
1009 struct lwp *lp = td->td_lwp;
1010 caddr_t params;
1011 struct sysent *callp;
1012 register_t orig_tf_rflags;
1013 int sticks;
1014 int error;
1015 int narg;
1016#ifdef INVARIANTS
1017 int crit_count = td->td_pri & ~TDPRI_MASK;
1018#endif
1019#ifdef SMP
1020 int have_mplock = 0;
1021#endif
1022 register_t *argp;
1023 u_int code;
1024 int reg, regcnt;
1025 union sysunion args;
1026 register_t *argsdst;
1027 kprintf0("SYSCALL rip = %016llx\n", frame->tf_rip);
1028
1029 PCPU_INC(cnt.v_syscall);
1030
1031 kprintf0("\033[31mSYSCALL %ld\033[39m\n", frame->tf_rax);
1032#ifdef DIAGNOSTIC
1033 if (ISPL(frame->tf_cs) != SEL_UPL) {
1034 get_mplock();
1035 panic("syscall");
1036 /* NOT REACHED */
1037 }
1038#endif
1039
1040 KTR_LOG(kernentry_syscall, p->p_pid, lp->lwp_tid,
1041 frame->tf_eax);
1042
1043#ifdef SMP
1044 KASSERT(td->td_mpcount == 0, ("badmpcount syscall2 from %p", (void *)frame->tf_eip));
1045 if (syscall_mpsafe == 0)
1046 MAKEMPSAFE(have_mplock);
1047#endif
1048 userenter(td); /* lazy raise our priority */
1049
1050 reg = 0;
1051 regcnt = 6;
1052 /*
1053 * Misc
1054 */
1055 sticks = (int)td->td_sticks;
1056 orig_tf_rflags = frame->tf_rflags;
1057
1058 /*
1059 * Virtual kernel intercept - if a VM context managed by a virtual
1060 * kernel issues a system call the virtual kernel handles it, not us.
1061 * Restore the virtual kernel context and return from its system
1062 * call. The current frame is copied out to the virtual kernel.
1063 */
1064 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
1065 error = vkernel_trap(lp, frame);
1066 frame->tf_rax = error;
1067 if (error)
1068 frame->tf_rflags |= PSL_C;
1069 error = EJUSTRETURN;
1070 goto out;
1071 }
1072
1073 /*
1074 * Get the system call parameters and account for time
1075 */
1076 lp->lwp_md.md_regs = frame;
1077 params = (caddr_t)frame->tf_rsp + sizeof(register_t);
1078 code = frame->tf_rax;
1079
1080 if (p->p_sysent->sv_prepsyscall) {
1081 (*p->p_sysent->sv_prepsyscall)(
1082 frame, (int *)(&args.nosys.sysmsg + 1),
1083 &code, &params);
1084 } else {
1085 if (code == SYS_syscall || code == SYS___syscall) {
1086 code = frame->tf_rdi;
1087 reg++;
1088 regcnt--;
1089 }
1090 }
1091
1092 if (p->p_sysent->sv_mask)
1093 code &= p->p_sysent->sv_mask;
1094
1095 if (code >= p->p_sysent->sv_size)
1096 callp = &p->p_sysent->sv_table[0];
1097 else
1098 callp = &p->p_sysent->sv_table[code];
1099
1100 narg = callp->sy_narg & SYF_ARGMASK;
1101
1102 /*
1103 * On amd64 we get up to six arguments in registers. The rest are
a6a09809 1104 * on the stack. The first six members of 'struct trapframe' happen
c8fe38ae
MD
1105 * to be the registers used to pass arguments, in exactly the right
1106 * order.
1107 */
1108 argp = &frame->tf_rdi;
1109 argp += reg;
1110 argsdst = (register_t *)(&args.nosys.sysmsg + 1);
1111 /*
1112 * JG can we overflow the space pointed to by 'argsdst'
1113 * either with 'bcopy' or with 'copyin'?
1114 */
1115 bcopy(argp, argsdst, sizeof(register_t) * regcnt);
1116 /*
1117 * copyin is MP aware, but the tracing code is not
1118 */
1119 if (narg > regcnt) {
1120 KASSERT(params != NULL, ("copyin args with no params!"));
1121 error = copyin(params, &argsdst[regcnt],
1122 (narg - regcnt) * sizeof(register_t));
1123 if (error) {
1124#ifdef KTRACE
1125 if (KTRPOINT(td, KTR_SYSCALL)) {
1126 MAKEMPSAFE(have_mplock);
1127
1128 ktrsyscall(lp, code, narg,
1129 (void *)(&args.nosys.sysmsg + 1));
1130 }
1131#endif
1132 goto bad;
1133 }
1134 }
1135
1136#ifdef KTRACE
1137 if (KTRPOINT(td, KTR_SYSCALL)) {
1138 MAKEMPSAFE(have_mplock);
1139 ktrsyscall(lp, code, narg, (void *)(&args.nosys.sysmsg + 1));
1140 }
1141#endif
1142
1143 /*
1144 * Default return value is 0 (will be copied to %rax). Double-value
1145 * returns use %rax and %rdx. %rdx is left unchanged for system
1146 * calls which return only one result.
1147 */
1148 args.sysmsg_fds[0] = 0;
1149 args.sysmsg_fds[1] = frame->tf_rdx;
1150
1151 /*
1152 * The syscall might manipulate the trap frame. If it does it
1153 * will probably return EJUSTRETURN.
1154 */
1155 args.sysmsg_frame = frame;
1156
1157 STOPEVENT(p, S_SCE, narg); /* MP aware */
1158
1159#ifdef SMP
1160 /*
1161 * Try to run the syscall without the MP lock if the syscall
1162 * is MP safe. We have to obtain the MP lock no matter what if
1163 * we are ktracing
1164 */
1165 if ((callp->sy_narg & SYF_MPSAFE) == 0)
1166 MAKEMPSAFE(have_mplock);
1167#endif
1168
1169 error = (*callp->sy_call)(&args);
1170
1171out:
1172 /*
1173 * MP SAFE (we may or may not have the MP lock at this point)
1174 */
0855a2af 1175 //kprintf("SYSMSG %d ", error);
c8fe38ae
MD
1176 switch (error) {
1177 case 0:
1178 /*
1179 * Reinitialize proc pointer `p' as it may be different
1180 * if this is a child returning from fork syscall.
1181 */
1182 p = curproc;
1183 lp = curthread->td_lwp;
1184 frame->tf_rax = args.sysmsg_fds[0];
1185 frame->tf_rdx = args.sysmsg_fds[1];
1186 kprintf0("RESULT %lld %lld\n", frame->tf_rax, frame->tf_rdx);
1187 frame->tf_rflags &= ~PSL_C;
1188 break;
1189 case ERESTART:
1190 /*
1191 * Reconstruct pc, we know that 'syscall' is 2 bytes.
1192 * We have to do a full context restore so that %r10
1193 * (which was holding the value of %rcx) is restored for
1194 * the next iteration.
1195 */
1196 frame->tf_rip -= frame->tf_err;
1197 frame->tf_r10 = frame->tf_rcx;
1198 td->td_pcb->pcb_flags |= PCB_FULLCTX;
1199 break;
1200 case EJUSTRETURN:
1201 break;
1202 case EASYNC:
1203 panic("Unexpected EASYNC return value (for now)");
1204 default:
1205bad:
1206 if (p->p_sysent->sv_errsize) {
1207 if (error >= p->p_sysent->sv_errsize)
1208 error = -1; /* XXX */
1209 else
1210 error = p->p_sysent->sv_errtbl[error];
1211 }
1212 kprintf0("ERROR %d\n", error);
1213 frame->tf_rax = error;
1214 frame->tf_rflags |= PSL_C;
1215 break;
1216 }
1217
1218 /*
1219 * Traced syscall. trapsignal() is not MP aware.
1220 */
1221 if (orig_tf_rflags & PSL_T) {
1222 MAKEMPSAFE(have_mplock);
1223 frame->tf_rflags &= ~PSL_T;
1224 trapsignal(lp, SIGTRAP, 0);
1225 }
1226
1227 /*
1228 * Handle reschedule and other end-of-syscall issues
1229 */
1230 userret(lp, frame, sticks);
1231
1232#ifdef KTRACE
1233 if (KTRPOINT(td, KTR_SYSRET)) {
1234 MAKEMPSAFE(have_mplock);
1235 ktrsysret(lp, code, error, args.sysmsg_result);
1236 }
1237#endif
1238
1239 /*
1240 * This works because errno is findable through the
1241 * register set. If we ever support an emulation where this
1242 * is not the case, this code will need to be revisited.
1243 */
1244 STOPEVENT(p, S_SCX, code);
1245
1246 userexit(lp);
1247#ifdef SMP
1248 /*
1249 * Release the MP lock if we had to get it
1250 */
1251 KASSERT(td->td_mpcount == have_mplock,
1252 ("badmpcount syscall2/end from %p", (void *)frame->tf_eip));
1253 if (have_mplock)
1254 rel_mplock();
1255#endif
1256 KTR_LOG(kernentry_syscall_ret, p->p_pid, lp->lwp_tid, error);
1257#ifdef INVARIANTS
1258 KASSERT(crit_count == (td->td_pri & ~TDPRI_MASK),
1259 ("syscall: critical section count mismatch! %d/%d",
1260 crit_count / TDPRI_CRIT, td->td_pri / TDPRI_CRIT));
1261#endif
1262}
d7f50089
YY
1263
1264void
1265fork_return(struct lwp *lp, struct trapframe *frame)
1266{
c8fe38ae
MD
1267 kprintf0("fork return\n");
1268 frame->tf_rax = 0; /* Child returns zero */
1269 frame->tf_rflags &= ~PSL_C; /* success */
1270 frame->tf_rdx = 1;
1271
1272 generic_lwp_return(lp, frame);
1273 KTR_LOG(kernentry_fork_ret, lp->lwp_proc->p_pid, lp->lwp_tid);
d7f50089
YY
1274}
1275
1276/*
1277 * Simplified back end of syscall(), used when returning from fork()
c8fe38ae
MD
1278 * directly into user mode. MP lock is held on entry and should be
1279 * released on return. This code will return back into the fork
d7f50089
YY
1280 * trampoline code which then runs doreti.
1281 */
1282void
1283generic_lwp_return(struct lwp *lp, struct trapframe *frame)
1284{
c8fe38ae
MD
1285 kprintf0("generic_lwp_return\n");
1286 struct proc *p = lp->lwp_proc;
1287
1288 /*
1289 * Newly forked processes are given a kernel priority. We have to
1290 * adjust the priority to a normal user priority and fake entry
1291 * into the kernel (call userenter()) to install a passive release
1292 * function just in case userret() decides to stop the process. This
1293 * can occur when ^Z races a fork. If we do not install the passive
1294 * release function the current process designation will not be
1295 * released when the thread goes to sleep.
1296 */
1297 lwkt_setpri_self(TDPRI_USER_NORM);
1298 userenter(lp->lwp_thread);
1299 userret(lp, frame, 0);
1300#ifdef KTRACE
1301 if (KTRPOINT(lp->lwp_thread, KTR_SYSRET))
1302 ktrsysret(lp, SYS_fork, 0, 0);
1303#endif
1304 p->p_flag |= P_PASSIVE_ACQ;
1305 userexit(lp);
1306 p->p_flag &= ~P_PASSIVE_ACQ;
1307#ifdef SMP
1308 KKASSERT(lp->lwp_thread->td_mpcount == 1);
1309 rel_mplock();
1310#endif
d7f50089
YY
1311}
1312
1313/*
1314 * If PGEX_FPFAULT is set then set FP_VIRTFP in the PCB to force a T_DNA
1315 * fault (which is then passed back to the virtual kernel) if an attempt is
1316 * made to use the FP unit.
c8fe38ae 1317 *
d7f50089
YY
1318 * XXX this is a fairly big hack.
1319 */
1320void
1321set_vkernel_fp(struct trapframe *frame)
1322{
c8fe38ae 1323 /* JGXXX */
d7f50089 1324}