subr_syscall.c revision 71785
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: head/sys/kern/subr_trap.c 71785 2001-01-29 09:38:39Z peter $ 39 */ 40 41/* 42 * 386 Trap and System call handling 43 */ 44 45#include "opt_clock.h" 46#include "opt_cpu.h" 47#include "opt_ddb.h" 48#include "opt_isa.h" 49#include "opt_ktrace.h" 50#include "opt_npx.h" 51#include "opt_trap.h" 52 53#include <sys/param.h> 54#include <sys/bus.h> 55#include <sys/systm.h> 56#include <sys/proc.h> 57#include <sys/pioctl.h> 58#include <sys/ipl.h> 59#include <sys/kernel.h> 60#include <sys/ktr.h> 61#include <sys/mutex.h> 62#include <sys/resourcevar.h> 63#include <sys/signalvar.h> 64#include <sys/syscall.h> 65#include <sys/sysctl.h> 66#include <sys/sysent.h> 67#include <sys/uio.h> 68#include <sys/vmmeter.h> 69#ifdef KTRACE 70#include <sys/ktrace.h> 71#endif 72 73#include <vm/vm.h> 74#include <vm/vm_param.h> 75#include <sys/lock.h> 76#include <vm/pmap.h> 77#include <vm/vm_kern.h> 78#include <vm/vm_map.h> 79#include <vm/vm_page.h> 80#include <vm/vm_extern.h> 81 82#include <machine/cpu.h> 83#include <machine/md_var.h> 84#include <machine/pcb.h> 85#ifdef SMP 86#include <machine/smp.h> 87#endif 88#include <machine/tss.h> 89 90#include <i386/isa/icu.h> 91#include <i386/isa/intr_machdep.h> 92 93#ifdef POWERFAIL_NMI 94#include <sys/syslog.h> 95#include <machine/clock.h> 96#endif 97 98#include <machine/vm86.h> 99 100#include <ddb/ddb.h> 101 102#include <sys/sysctl.h> 103 104int (*pmath_emulate) __P((struct trapframe *)); 105 106extern void trap __P((struct trapframe frame)); 107extern int trapwrite __P((unsigned addr)); 108extern void syscall2 __P((struct trapframe frame)); 109extern void ast __P((struct trapframe frame)); 110 111static int trap_pfault __P((struct trapframe *, int, vm_offset_t)); 112static void trap_fatal __P((struct trapframe *, vm_offset_t)); 113void dblfault_handler __P((void)); 114 115extern inthand_t IDTVEC(syscall); 116 117#define MAX_TRAP_MSG 28 118static char *trap_msg[] = { 119 "", /* 0 unused */ 120 "privileged instruction fault", /* 1 T_PRIVINFLT */ 121 "", /* 2 unused */ 122 "breakpoint instruction fault", /* 3 T_BPTFLT */ 123 "", /* 4 unused */ 124 "", /* 5 unused */ 125 "arithmetic trap", /* 6 T_ARITHTRAP */ 126 "system forced exception", /* 7 T_ASTFLT */ 127 "", /* 8 unused */ 128 "general protection fault", /* 9 T_PROTFLT */ 129 "trace trap", /* 10 T_TRCTRAP */ 130 "", /* 11 unused */ 131 "page fault", /* 12 T_PAGEFLT */ 132 "", /* 13 unused */ 133 "alignment fault", /* 14 T_ALIGNFLT */ 134 "", /* 15 unused */ 135 "", /* 16 unused */ 136 "", /* 17 unused */ 137 "integer divide fault", /* 18 T_DIVIDE */ 138 "non-maskable interrupt trap", /* 19 T_NMI */ 139 "overflow trap", /* 20 T_OFLOW */ 140 "FPU bounds check fault", /* 21 T_BOUND */ 141 "FPU device not available", /* 22 T_DNA */ 142 "double fault", /* 23 T_DOUBLEFLT */ 143 "FPU operand fetch fault", /* 24 T_FPOPFLT */ 144 "invalid TSS fault", /* 25 T_TSSFLT */ 145 "segment not present fault", /* 26 T_SEGNPFLT */ 146 "stack fault", /* 27 T_STKFLT */ 147 "machine check trap", /* 28 T_MCHK */ 148}; 149 150#if defined(I586_CPU) && !defined(NO_F00F_HACK) 151extern int has_f00f_bug; 152#endif 153 154#ifdef DDB 155static int ddb_on_nmi = 1; 156SYSCTL_INT(_machdep, OID_AUTO, ddb_on_nmi, CTLFLAG_RW, 157 &ddb_on_nmi, 0, "Go to DDB on NMI"); 158#endif 159static int panic_on_nmi = 1; 160SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RW, 161 &panic_on_nmi, 0, "Panic on NMI"); 162 163#ifdef WITNESS 164extern char *syscallnames[]; 165#endif 166 167void 168userret(p, frame, oticks) 169 struct proc *p; 170 struct trapframe *frame; 171 u_quad_t oticks; 172{ 173 int sig; 174 175 while ((sig = CURSIG(p)) != 0) { 176 if (!mtx_owned(&Giant)) 177 mtx_enter(&Giant, MTX_DEF); 178 postsig(sig); 179 } 180 181 mtx_enter(&sched_lock, MTX_SPIN); 182 p->p_priority = p->p_usrpri; 183 if (resched_wanted()) { 184 /* 185 * Since we are curproc, clock will normally just change 186 * our priority without moving us from one queue to another 187 * (since the running process is not on a queue.) 188 * If that happened after we setrunqueue ourselves but before we 189 * mi_switch()'ed, we might not be on the queue indicated by 190 * our priority. 191 */ 192 DROP_GIANT_NOSWITCH(); 193 setrunqueue(p); 194 p->p_stats->p_ru.ru_nivcsw++; 195 mi_switch(); 196 mtx_exit(&sched_lock, MTX_SPIN); 197 PICKUP_GIANT(); 198 while ((sig = CURSIG(p)) != 0) { 199 if (!mtx_owned(&Giant)) 200 mtx_enter(&Giant, MTX_DEF); 201 postsig(sig); 202 } 203 mtx_enter(&sched_lock, MTX_SPIN); 204 } 205 206 /* 207 * Charge system time if profiling. 208 */ 209 if (p->p_sflag & PS_PROFIL) { 210 mtx_exit(&sched_lock, MTX_SPIN); 211 /* XXX - do we need Giant? */ 212 if (!mtx_owned(&Giant)) 213 mtx_enter(&Giant, MTX_DEF); 214 mtx_enter(&sched_lock, MTX_SPIN); 215 addupc_task(p, frame->tf_eip, 216 (u_int)(p->p_sticks - oticks) * psratio); 217 } 218 curpriority = p->p_priority; 219 mtx_exit(&sched_lock, MTX_SPIN); 220} 221 222/* 223 * Exception, fault, and trap interface to the FreeBSD kernel. 224 * This common code is called from assembly language IDT gate entry 225 * routines that prepare a suitable stack frame, and restore this 226 * frame after the exception has been processed. 227 */ 228 229void 230trap(frame) 231 struct trapframe frame; 232{ 233 struct proc *p = curproc; 234 u_quad_t sticks = 0; 235 int i = 0, ucode = 0, type, code; 236 vm_offset_t eva; 237#ifdef POWERFAIL_NMI 238 static int lastalert = 0; 239#endif 240 241 atomic_add_int(&cnt.v_trap, 1); 242 243 if ((frame.tf_eflags & PSL_I) == 0) { 244 /* 245 * Buggy application or kernel code has disabled 246 * interrupts and then trapped. Enabling interrupts 247 * now is wrong, but it is better than running with 248 * interrupts disabled until they are accidentally 249 * enabled later. XXX This is really bad if we trap 250 * while holding a spin lock. 251 */ 252 type = frame.tf_trapno; 253 if (ISPL(frame.tf_cs) == SEL_UPL || (frame.tf_eflags & PSL_VM)) 254 printf( 255 "pid %ld (%s): trap %d with interrupts disabled\n", 256 (long)curproc->p_pid, curproc->p_comm, type); 257 else if (type != T_BPTFLT && type != T_TRCTRAP) 258 /* 259 * XXX not quite right, since this may be for a 260 * multiple fault in user mode. 261 */ 262 printf("kernel trap %d with interrupts disabled\n", 263 type); 264 /* 265 * We should walk p_heldmtx here and see if any are 266 * spin mutexes, and not do this if so. 267 */ 268 enable_intr(); 269 } 270 271 eva = 0; 272 273#if defined(I586_CPU) && !defined(NO_F00F_HACK) 274restart: 275#endif 276 277 type = frame.tf_trapno; 278 code = frame.tf_err; 279 280 if ((ISPL(frame.tf_cs) == SEL_UPL) || 281 ((frame.tf_eflags & PSL_VM) && !in_vm86call)) { 282 /* user trap */ 283 284 mtx_enter(&sched_lock, MTX_SPIN); 285 sticks = p->p_sticks; 286 mtx_exit(&sched_lock, MTX_SPIN); 287 p->p_md.md_regs = &frame; 288 289 switch (type) { 290 case T_PRIVINFLT: /* privileged instruction fault */ 291 ucode = type; 292 i = SIGILL; 293 break; 294 295 case T_BPTFLT: /* bpt instruction fault */ 296 case T_TRCTRAP: /* trace trap */ 297 frame.tf_eflags &= ~PSL_T; 298 i = SIGTRAP; 299 break; 300 301 case T_ARITHTRAP: /* arithmetic trap */ 302 ucode = code; 303 i = SIGFPE; 304 break; 305 306 /* 307 * The following two traps can happen in 308 * vm86 mode, and, if so, we want to handle 309 * them specially. 310 */ 311 case T_PROTFLT: /* general protection fault */ 312 case T_STKFLT: /* stack fault */ 313 if (frame.tf_eflags & PSL_VM) { 314 mtx_enter(&Giant, MTX_DEF); 315 i = vm86_emulate((struct vm86frame *)&frame); 316 mtx_exit(&Giant, MTX_DEF); 317 if (i == 0) 318 goto user; 319 break; 320 } 321 /* FALL THROUGH */ 322 323 case T_SEGNPFLT: /* segment not present fault */ 324 case T_TSSFLT: /* invalid TSS fault */ 325 case T_DOUBLEFLT: /* double fault */ 326 default: 327 ucode = code + BUS_SEGM_FAULT ; 328 i = SIGBUS; 329 break; 330 331 case T_PAGEFLT: /* page fault */ 332 /* 333 * For some Cyrix CPUs, %cr2 is clobbered by 334 * interrupts. This problem is worked around by using 335 * an interrupt gate for the pagefault handler. We 336 * are finally ready to read %cr2 and then must 337 * reenable interrupts. 338 */ 339 eva = rcr2(); 340 enable_intr(); 341 mtx_enter(&Giant, MTX_DEF); 342 i = trap_pfault(&frame, TRUE, eva); 343 mtx_exit(&Giant, MTX_DEF); 344#if defined(I586_CPU) && !defined(NO_F00F_HACK) 345 if (i == -2) { 346 /* 347 * f00f hack workaround has triggered, treat 348 * as illegal instruction not page fault. 349 */ 350 frame.tf_trapno = T_PRIVINFLT; 351 goto restart; 352 } 353#endif 354 if (i == -1) 355 goto out; 356 if (i == 0) 357 goto user; 358 359 ucode = T_PAGEFLT; 360 break; 361 362 case T_DIVIDE: /* integer divide fault */ 363 ucode = FPE_INTDIV; 364 i = SIGFPE; 365 break; 366 367#ifdef DEV_ISA 368 case T_NMI: 369#ifdef POWERFAIL_NMI 370#ifndef TIMER_FREQ 371# define TIMER_FREQ 1193182 372#endif 373 mtx_enter(&Giant, MTX_DEF); 374 if (time_second - lastalert > 10) { 375 log(LOG_WARNING, "NMI: power fail\n"); 376 sysbeep(TIMER_FREQ/880, hz); 377 lastalert = time_second; 378 } 379 mtx_exit(&Giant, MTX_DEF); 380 goto out; 381#else /* !POWERFAIL_NMI */ 382 /* machine/parity/power fail/"kitchen sink" faults */ 383 /* XXX Giant */ 384 if (isa_nmi(code) == 0) { 385#ifdef DDB 386 /* 387 * NMI can be hooked up to a pushbutton 388 * for debugging. 389 */ 390 if (ddb_on_nmi) { 391 printf ("NMI ... going to debugger\n"); 392 kdb_trap (type, 0, &frame); 393 } 394#endif /* DDB */ 395 goto out; 396 } else if (panic_on_nmi) 397 panic("NMI indicates hardware failure"); 398 break; 399#endif /* POWERFAIL_NMI */ 400#endif /* DEV_ISA */ 401 402 case T_OFLOW: /* integer overflow fault */ 403 ucode = FPE_INTOVF; 404 i = SIGFPE; 405 break; 406 407 case T_BOUND: /* bounds check fault */ 408 ucode = FPE_FLTSUB; 409 i = SIGFPE; 410 break; 411 412 case T_DNA: 413#ifdef DEV_NPX 414 /* transparent fault (due to context switch "late") */ 415 if (npxdna()) 416 goto out; 417#endif 418 if (!pmath_emulate) { 419 i = SIGFPE; 420 ucode = FPE_FPU_NP_TRAP; 421 break; 422 } 423 mtx_enter(&Giant, MTX_DEF); 424 i = (*pmath_emulate)(&frame); 425 mtx_exit(&Giant, MTX_DEF); 426 if (i == 0) { 427 if (!(frame.tf_eflags & PSL_T)) 428 goto out; 429 frame.tf_eflags &= ~PSL_T; 430 i = SIGTRAP; 431 } 432 /* else ucode = emulator_only_knows() XXX */ 433 break; 434 435 case T_FPOPFLT: /* FPU operand fetch fault */ 436 ucode = T_FPOPFLT; 437 i = SIGILL; 438 break; 439 } 440 } else { 441 /* kernel trap */ 442 443 switch (type) { 444 case T_PAGEFLT: /* page fault */ 445 /* 446 * For some Cyrix CPUs, %cr2 is clobbered by 447 * interrupts. This problem is worked around by using 448 * an interrupt gate for the pagefault handler. We 449 * are finally ready to read %cr2 and then must 450 * reenable interrupts. 451 */ 452 eva = rcr2(); 453 enable_intr(); 454 mtx_enter(&Giant, MTX_DEF); 455 (void) trap_pfault(&frame, FALSE, eva); 456 mtx_exit(&Giant, MTX_DEF); 457 goto out; 458 459 case T_DNA: 460#ifdef DEV_NPX 461 /* 462 * The kernel is apparently using npx for copying. 463 * XXX this should be fatal unless the kernel has 464 * registered such use. 465 */ 466 if (npxdna()) 467 goto out; 468#endif 469 break; 470 471 /* 472 * The following two traps can happen in 473 * vm86 mode, and, if so, we want to handle 474 * them specially. 475 */ 476 case T_PROTFLT: /* general protection fault */ 477 case T_STKFLT: /* stack fault */ 478 if (frame.tf_eflags & PSL_VM) { 479 mtx_enter(&Giant, MTX_DEF); 480 i = vm86_emulate((struct vm86frame *)&frame); 481 mtx_exit(&Giant, MTX_DEF); 482 if (i != 0) 483 /* 484 * returns to original process 485 */ 486 vm86_trap((struct vm86frame *)&frame); 487 goto out; 488 } 489 if (type == T_STKFLT) 490 break; 491 492 /* FALL THROUGH */ 493 494 case T_SEGNPFLT: /* segment not present fault */ 495 if (in_vm86call) 496 break; 497 498 if (p->p_intr_nesting_level != 0) 499 break; 500 501 /* 502 * Invalid %fs's and %gs's can be created using 503 * procfs or PT_SETREGS or by invalidating the 504 * underlying LDT entry. This causes a fault 505 * in kernel mode when the kernel attempts to 506 * switch contexts. Lose the bad context 507 * (XXX) so that we can continue, and generate 508 * a signal. 509 */ 510 if (frame.tf_eip == (int)cpu_switch_load_gs) { 511 PCPU_GET(curpcb)->pcb_gs = 0; 512 mtx_enter(&Giant, MTX_DEF); 513 psignal(p, SIGBUS); 514 mtx_exit(&Giant, MTX_DEF); 515 goto out; 516 } 517 518 /* 519 * Invalid segment selectors and out of bounds 520 * %eip's and %esp's can be set up in user mode. 521 * This causes a fault in kernel mode when the 522 * kernel tries to return to user mode. We want 523 * to get this fault so that we can fix the 524 * problem here and not have to check all the 525 * selectors and pointers when the user changes 526 * them. 527 */ 528 if (frame.tf_eip == (int)doreti_iret) { 529 frame.tf_eip = (int)doreti_iret_fault; 530 goto out; 531 } 532 if (frame.tf_eip == (int)doreti_popl_ds) { 533 frame.tf_eip = (int)doreti_popl_ds_fault; 534 goto out; 535 } 536 if (frame.tf_eip == (int)doreti_popl_es) { 537 frame.tf_eip = (int)doreti_popl_es_fault; 538 goto out; 539 } 540 if (frame.tf_eip == (int)doreti_popl_fs) { 541 frame.tf_eip = (int)doreti_popl_fs_fault; 542 goto out; 543 } 544 if (PCPU_GET(curpcb) != NULL && 545 PCPU_GET(curpcb)->pcb_onfault != NULL) { 546 frame.tf_eip = 547 (int)PCPU_GET(curpcb)->pcb_onfault; 548 goto out; 549 } 550 break; 551 552 case T_TSSFLT: 553 /* 554 * PSL_NT can be set in user mode and isn't cleared 555 * automatically when the kernel is entered. This 556 * causes a TSS fault when the kernel attempts to 557 * `iret' because the TSS link is uninitialized. We 558 * want to get this fault so that we can fix the 559 * problem here and not every time the kernel is 560 * entered. 561 */ 562 if (frame.tf_eflags & PSL_NT) { 563 frame.tf_eflags &= ~PSL_NT; 564 goto out; 565 } 566 break; 567 568 case T_TRCTRAP: /* trace trap */ 569 if (frame.tf_eip == (int)IDTVEC(syscall)) { 570 /* 571 * We've just entered system mode via the 572 * syscall lcall. Continue single stepping 573 * silently until the syscall handler has 574 * saved the flags. 575 */ 576 goto out; 577 } 578 if (frame.tf_eip == (int)IDTVEC(syscall) + 1) { 579 /* 580 * The syscall handler has now saved the 581 * flags. Stop single stepping it. 582 */ 583 frame.tf_eflags &= ~PSL_T; 584 goto out; 585 } 586 /* 587 * Ignore debug register trace traps due to 588 * accesses in the user's address space, which 589 * can happen under several conditions such as 590 * if a user sets a watchpoint on a buffer and 591 * then passes that buffer to a system call. 592 * We still want to get TRCTRAPS for addresses 593 * in kernel space because that is useful when 594 * debugging the kernel. 595 */ 596 /* XXX Giant */ 597 if (user_dbreg_trap() && !in_vm86call) { 598 /* 599 * Reset breakpoint bits because the 600 * processor doesn't 601 */ 602 load_dr6(rdr6() & 0xfffffff0); 603 goto out; 604 } 605 /* 606 * Fall through (TRCTRAP kernel mode, kernel address) 607 */ 608 case T_BPTFLT: 609 /* 610 * If DDB is enabled, let it handle the debugger trap. 611 * Otherwise, debugger traps "can't happen". 612 */ 613#ifdef DDB 614 /* XXX Giant */ 615 if (kdb_trap (type, 0, &frame)) 616 goto out; 617#endif 618 break; 619 620#ifdef DEV_ISA 621 case T_NMI: 622#ifdef POWERFAIL_NMI 623 mtx_enter(&Giant, MTX_DEF); 624 if (time_second - lastalert > 10) { 625 log(LOG_WARNING, "NMI: power fail\n"); 626 sysbeep(TIMER_FREQ/880, hz); 627 lastalert = time_second; 628 } 629 mtx_exit(&Giant, MTX_DEF); 630 goto out; 631#else /* !POWERFAIL_NMI */ 632 /* XXX Giant */ 633 /* machine/parity/power fail/"kitchen sink" faults */ 634 if (isa_nmi(code) == 0) { 635#ifdef DDB 636 /* 637 * NMI can be hooked up to a pushbutton 638 * for debugging. 639 */ 640 if (ddb_on_nmi) { 641 printf ("NMI ... going to debugger\n"); 642 kdb_trap (type, 0, &frame); 643 } 644#endif /* DDB */ 645 goto out; 646 } else if (panic_on_nmi == 0) 647 goto out; 648 /* FALL THROUGH */ 649#endif /* POWERFAIL_NMI */ 650#endif /* DEV_ISA */ 651 } 652 653 mtx_enter(&Giant, MTX_DEF); 654 trap_fatal(&frame, eva); 655 mtx_exit(&Giant, MTX_DEF); 656 goto out; 657 } 658 659 mtx_enter(&Giant, MTX_DEF); 660 /* Translate fault for emulators (e.g. Linux) */ 661 if (*p->p_sysent->sv_transtrap) 662 i = (*p->p_sysent->sv_transtrap)(i, type); 663 664 trapsignal(p, i, ucode); 665 666#ifdef DEBUG 667 if (type <= MAX_TRAP_MSG) { 668 uprintf("fatal process exception: %s", 669 trap_msg[type]); 670 if ((type == T_PAGEFLT) || (type == T_PROTFLT)) 671 uprintf(", fault VA = 0x%lx", (u_long)eva); 672 uprintf("\n"); 673 } 674#endif 675 mtx_exit(&Giant, MTX_DEF); 676 677user: 678 userret(p, &frame, sticks); 679 if (mtx_owned(&Giant)) 680 mtx_exit(&Giant, MTX_DEF); 681out: 682 return; 683} 684 685#ifdef notyet 686/* 687 * This version doesn't allow a page fault to user space while 688 * in the kernel. The rest of the kernel needs to be made "safe" 689 * before this can be used. I think the only things remaining 690 * to be made safe are the iBCS2 code and the process tracing/ 691 * debugging code. 692 */ 693static int 694trap_pfault(frame, usermode, eva) 695 struct trapframe *frame; 696 int usermode; 697 vm_offset_t eva; 698{ 699 vm_offset_t va; 700 struct vmspace *vm = NULL; 701 vm_map_t map = 0; 702 int rv = 0; 703 vm_prot_t ftype; 704 struct proc *p = curproc; 705 706 if (frame->tf_err & PGEX_W) 707 ftype = VM_PROT_WRITE; 708 else 709 ftype = VM_PROT_READ; 710 711 va = trunc_page(eva); 712 if (va < VM_MIN_KERNEL_ADDRESS) { 713 vm_offset_t v; 714 vm_page_t mpte; 715 716 if (p == NULL || 717 (!usermode && va < VM_MAXUSER_ADDRESS && 718 (p->p_intr_nesting_level != 0 || 719 PCPU_GET(curpcb) == NULL || 720 PCPU_GET(curpcb)->pcb_onfault == NULL))) { 721 trap_fatal(frame, eva); 722 return (-1); 723 } 724 725 /* 726 * This is a fault on non-kernel virtual memory. 727 * vm is initialized above to NULL. If curproc is NULL 728 * or curproc->p_vmspace is NULL the fault is fatal. 729 */ 730 vm = p->p_vmspace; 731 if (vm == NULL) 732 goto nogo; 733 734 map = &vm->vm_map; 735 736 /* 737 * Keep swapout from messing with us during this 738 * critical time. 739 */ 740 PROC_LOCK(p); 741 ++p->p_lock; 742 PROC_UNLOCK(p); 743 744 /* 745 * Grow the stack if necessary 746 */ 747 /* grow_stack returns false only if va falls into 748 * a growable stack region and the stack growth 749 * fails. It returns true if va was not within 750 * a growable stack region, or if the stack 751 * growth succeeded. 752 */ 753 if (!grow_stack (p, va)) { 754 rv = KERN_FAILURE; 755 PROC_LOCK(p); 756 --p->p_lock; 757 PROC_UNLOCK(p); 758 goto nogo; 759 } 760 761 /* Fault in the user page: */ 762 rv = vm_fault(map, va, ftype, 763 (ftype & VM_PROT_WRITE) ? VM_FAULT_DIRTY 764 : VM_FAULT_NORMAL); 765 766 PROC_LOCK(p); 767 --p->p_lock; 768 PROC_UNLOCK(p); 769 } else { 770 /* 771 * Don't allow user-mode faults in kernel address space. 772 */ 773 if (usermode) 774 goto nogo; 775 776 /* 777 * Since we know that kernel virtual address addresses 778 * always have pte pages mapped, we just have to fault 779 * the page. 780 */ 781 rv = vm_fault(kernel_map, va, ftype, VM_FAULT_NORMAL); 782 } 783 784 if (rv == KERN_SUCCESS) 785 return (0); 786nogo: 787 if (!usermode) { 788 if (p->p_intr_nesting_level == 0 && 789 PCPU_GET(curpcb) != NULL && 790 PCPU_GET(curpcb)->pcb_onfault != NULL) { 791 frame->tf_eip = (int)PCPU_GET(curpcb)->pcb_onfault; 792 return (0); 793 } 794 trap_fatal(frame, eva); 795 return (-1); 796 } 797 798 /* kludge to pass faulting virtual address to sendsig */ 799 frame->tf_err = eva; 800 801 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV); 802} 803#endif 804 805int 806trap_pfault(frame, usermode, eva) 807 struct trapframe *frame; 808 int usermode; 809 vm_offset_t eva; 810{ 811 vm_offset_t va; 812 struct vmspace *vm = NULL; 813 vm_map_t map = 0; 814 int rv = 0; 815 vm_prot_t ftype; 816 struct proc *p = curproc; 817 818 va = trunc_page(eva); 819 if (va >= KERNBASE) { 820 /* 821 * Don't allow user-mode faults in kernel address space. 822 * An exception: if the faulting address is the invalid 823 * instruction entry in the IDT, then the Intel Pentium 824 * F00F bug workaround was triggered, and we need to 825 * treat it is as an illegal instruction, and not a page 826 * fault. 827 */ 828#if defined(I586_CPU) && !defined(NO_F00F_HACK) 829 if ((eva == (unsigned int)&idt[6]) && has_f00f_bug) 830 return -2; 831#endif 832 if (usermode) 833 goto nogo; 834 835 map = kernel_map; 836 } else { 837 /* 838 * This is a fault on non-kernel virtual memory. 839 * vm is initialized above to NULL. If curproc is NULL 840 * or curproc->p_vmspace is NULL the fault is fatal. 841 */ 842 if (p != NULL) 843 vm = p->p_vmspace; 844 845 if (vm == NULL) 846 goto nogo; 847 848 map = &vm->vm_map; 849 } 850 851 if (frame->tf_err & PGEX_W) 852 ftype = VM_PROT_WRITE; 853 else 854 ftype = VM_PROT_READ; 855 856 if (map != kernel_map) { 857 /* 858 * Keep swapout from messing with us during this 859 * critical time. 860 */ 861 PROC_LOCK(p); 862 ++p->p_lock; 863 PROC_UNLOCK(p); 864 865 /* 866 * Grow the stack if necessary 867 */ 868 /* grow_stack returns false only if va falls into 869 * a growable stack region and the stack growth 870 * fails. It returns true if va was not within 871 * a growable stack region, or if the stack 872 * growth succeeded. 873 */ 874 if (!grow_stack (p, va)) { 875 rv = KERN_FAILURE; 876 PROC_LOCK(p); 877 --p->p_lock; 878 PROC_UNLOCK(p); 879 goto nogo; 880 } 881 882 /* Fault in the user page: */ 883 rv = vm_fault(map, va, ftype, 884 (ftype & VM_PROT_WRITE) ? VM_FAULT_DIRTY 885 : VM_FAULT_NORMAL); 886 887 PROC_LOCK(p); 888 --p->p_lock; 889 PROC_UNLOCK(p); 890 } else { 891 /* 892 * Don't have to worry about process locking or stacks in the 893 * kernel. 894 */ 895 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL); 896 } 897 898 if (rv == KERN_SUCCESS) 899 return (0); 900nogo: 901 if (!usermode) { 902 if (p->p_intr_nesting_level == 0 && 903 PCPU_GET(curpcb) != NULL && 904 PCPU_GET(curpcb)->pcb_onfault != NULL) { 905 frame->tf_eip = (int)PCPU_GET(curpcb)->pcb_onfault; 906 return (0); 907 } 908 trap_fatal(frame, eva); 909 return (-1); 910 } 911 912 /* kludge to pass faulting virtual address to sendsig */ 913 frame->tf_err = eva; 914 915 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV); 916} 917 918static void 919trap_fatal(frame, eva) 920 struct trapframe *frame; 921 vm_offset_t eva; 922{ 923 int code, type, ss, esp; 924 struct soft_segment_descriptor softseg; 925 926 code = frame->tf_err; 927 type = frame->tf_trapno; 928 sdtossd(&gdt[IDXSEL(frame->tf_cs & 0xffff)].sd, &softseg); 929 930 if (type <= MAX_TRAP_MSG) 931 printf("\n\nFatal trap %d: %s while in %s mode\n", 932 type, trap_msg[type], 933 frame->tf_eflags & PSL_VM ? "vm86" : 934 ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel"); 935#ifdef SMP 936 /* two seperate prints in case of a trap on an unmapped page */ 937 printf("cpuid = %d; ", PCPU_GET(cpuid)); 938 printf("lapic.id = %08x\n", lapic.id); 939#endif 940 if (type == T_PAGEFLT) { 941 printf("fault virtual address = 0x%x\n", eva); 942 printf("fault code = %s %s, %s\n", 943 code & PGEX_U ? "user" : "supervisor", 944 code & PGEX_W ? "write" : "read", 945 code & PGEX_P ? "protection violation" : "page not present"); 946 } 947 printf("instruction pointer = 0x%x:0x%x\n", 948 frame->tf_cs & 0xffff, frame->tf_eip); 949 if ((ISPL(frame->tf_cs) == SEL_UPL) || (frame->tf_eflags & PSL_VM)) { 950 ss = frame->tf_ss & 0xffff; 951 esp = frame->tf_esp; 952 } else { 953 ss = GSEL(GDATA_SEL, SEL_KPL); 954 esp = (int)&frame->tf_esp; 955 } 956 printf("stack pointer = 0x%x:0x%x\n", ss, esp); 957 printf("frame pointer = 0x%x:0x%x\n", ss, frame->tf_ebp); 958 printf("code segment = base 0x%x, limit 0x%x, type 0x%x\n", 959 softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type); 960 printf(" = DPL %d, pres %d, def32 %d, gran %d\n", 961 softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_def32, 962 softseg.ssd_gran); 963 printf("processor eflags = "); 964 if (frame->tf_eflags & PSL_T) 965 printf("trace trap, "); 966 if (frame->tf_eflags & PSL_I) 967 printf("interrupt enabled, "); 968 if (frame->tf_eflags & PSL_NT) 969 printf("nested task, "); 970 if (frame->tf_eflags & PSL_RF) 971 printf("resume, "); 972 if (frame->tf_eflags & PSL_VM) 973 printf("vm86, "); 974 printf("IOPL = %d\n", (frame->tf_eflags & PSL_IOPL) >> 12); 975 printf("current process = "); 976 if (curproc) { 977 printf("%lu (%s)\n", 978 (u_long)curproc->p_pid, curproc->p_comm ? 979 curproc->p_comm : ""); 980 } else { 981 printf("Idle\n"); 982 } 983 984#ifdef KDB 985 if (kdb_trap(&psl)) 986 return; 987#endif 988#ifdef DDB 989 if ((debugger_on_panic || db_active) && kdb_trap(type, 0, frame)) 990 return; 991#endif 992 printf("trap number = %d\n", type); 993 if (type <= MAX_TRAP_MSG) 994 panic(trap_msg[type]); 995 else 996 panic("unknown/reserved trap"); 997} 998 999/* 1000 * Double fault handler. Called when a fault occurs while writing 1001 * a frame for a trap/exception onto the stack. This usually occurs 1002 * when the stack overflows (such is the case with infinite recursion, 1003 * for example). 1004 * 1005 * XXX Note that the current PTD gets replaced by IdlePTD when the 1006 * task switch occurs. This means that the stack that was active at 1007 * the time of the double fault is not available at <kstack> unless 1008 * the machine was idle when the double fault occurred. The downside 1009 * of this is that "trace <ebp>" in ddb won't work. 1010 */ 1011void 1012dblfault_handler() 1013{ 1014 printf("\nFatal double fault:\n"); 1015 printf("eip = 0x%x\n", PCPU_GET(common_tss.tss_eip)); 1016 printf("esp = 0x%x\n", PCPU_GET(common_tss.tss_esp)); 1017 printf("ebp = 0x%x\n", PCPU_GET(common_tss.tss_ebp)); 1018#ifdef SMP 1019 /* two seperate prints in case of a trap on an unmapped page */ 1020 printf("cpuid = %d; ", PCPU_GET(cpuid)); 1021 printf("lapic.id = %08x\n", lapic.id); 1022#endif 1023 panic("double fault"); 1024} 1025 1026/* 1027 * Compensate for 386 brain damage (missing URKR). 1028 * This is a little simpler than the pagefault handler in trap() because 1029 * it the page tables have already been faulted in and high addresses 1030 * are thrown out early for other reasons. 1031 */ 1032int trapwrite(addr) 1033 unsigned addr; 1034{ 1035 struct proc *p; 1036 vm_offset_t va; 1037 struct vmspace *vm; 1038 int rv; 1039 1040 va = trunc_page((vm_offset_t)addr); 1041 /* 1042 * XXX - MAX is END. Changed > to >= for temp. fix. 1043 */ 1044 if (va >= VM_MAXUSER_ADDRESS) 1045 return (1); 1046 1047 p = curproc; 1048 vm = p->p_vmspace; 1049 1050 PROC_LOCK(p); 1051 ++p->p_lock; 1052 PROC_UNLOCK(p); 1053 1054 if (!grow_stack (p, va)) { 1055 PROC_LOCK(p); 1056 --p->p_lock; 1057 PROC_UNLOCK(p); 1058 return (1); 1059 } 1060 1061 /* 1062 * fault the data page 1063 */ 1064 rv = vm_fault(&vm->vm_map, va, VM_PROT_WRITE, VM_FAULT_DIRTY); 1065 1066 PROC_LOCK(p); 1067 --p->p_lock; 1068 PROC_UNLOCK(p); 1069 1070 if (rv != KERN_SUCCESS) 1071 return 1; 1072 1073 return (0); 1074} 1075 1076/* 1077 * syscall2 - MP aware system call request C handler 1078 * 1079 * A system call is essentially treated as a trap except that the 1080 * MP lock is not held on entry or return. We are responsible for 1081 * obtaining the MP lock if necessary and for handling ASTs 1082 * (e.g. a task switch) prior to return. 1083 * 1084 * In general, only simple access and manipulation of curproc and 1085 * the current stack is allowed without having to hold MP lock. 1086 */ 1087void 1088syscall2(frame) 1089 struct trapframe frame; 1090{ 1091 caddr_t params; 1092 int i; 1093 struct sysent *callp; 1094 struct proc *p = curproc; 1095 u_quad_t sticks; 1096 int error; 1097 int narg; 1098 int args[8]; 1099 u_int code; 1100 1101 atomic_add_int(&cnt.v_syscall, 1); 1102 1103#ifdef DIAGNOSTIC 1104 if (ISPL(frame.tf_cs) != SEL_UPL) { 1105 mtx_enter(&Giant, MTX_DEF); 1106 panic("syscall"); 1107 /* NOT REACHED */ 1108 } 1109#endif 1110 1111 mtx_enter(&sched_lock, MTX_SPIN); 1112 sticks = p->p_sticks; 1113 mtx_exit(&sched_lock, MTX_SPIN); 1114 1115 p->p_md.md_regs = &frame; 1116 params = (caddr_t)frame.tf_esp + sizeof(int); 1117 code = frame.tf_eax; 1118 1119 if (p->p_sysent->sv_prepsyscall) { 1120 /* 1121 * The prep code is not MP aware. 1122 */ 1123 mtx_enter(&Giant, MTX_DEF); 1124 (*p->p_sysent->sv_prepsyscall)(&frame, args, &code, ¶ms); 1125 mtx_exit(&Giant, MTX_DEF); 1126 } else { 1127 /* 1128 * Need to check if this is a 32 bit or 64 bit syscall. 1129 * fuword is MP aware. 1130 */ 1131 if (code == SYS_syscall) { 1132 /* 1133 * Code is first argument, followed by actual args. 1134 */ 1135 code = fuword(params); 1136 params += sizeof(int); 1137 } else if (code == SYS___syscall) { 1138 /* 1139 * Like syscall, but code is a quad, so as to maintain 1140 * quad alignment for the rest of the arguments. 1141 */ 1142 code = fuword(params); 1143 params += sizeof(quad_t); 1144 } 1145 } 1146 1147 if (p->p_sysent->sv_mask) 1148 code &= p->p_sysent->sv_mask; 1149 1150 if (code >= p->p_sysent->sv_size) 1151 callp = &p->p_sysent->sv_table[0]; 1152 else 1153 callp = &p->p_sysent->sv_table[code]; 1154 1155 narg = callp->sy_narg & SYF_ARGMASK; 1156 1157 /* 1158 * copyin is MP aware, but the tracing code is not 1159 */ 1160 if (params && (i = narg * sizeof(int)) && 1161 (error = copyin(params, (caddr_t)args, (u_int)i))) { 1162 mtx_enter(&Giant, MTX_DEF); 1163#ifdef KTRACE 1164 if (KTRPOINT(p, KTR_SYSCALL)) 1165 ktrsyscall(p->p_tracep, code, narg, args); 1166#endif 1167 goto bad; 1168 } 1169 1170 /* 1171 * Try to run the syscall without the MP lock if the syscall 1172 * is MP safe. We have to obtain the MP lock no matter what if 1173 * we are ktracing 1174 */ 1175 if ((callp->sy_narg & SYF_MPSAFE) == 0) { 1176 mtx_enter(&Giant, MTX_DEF); 1177 } 1178 1179#ifdef KTRACE 1180 if (KTRPOINT(p, KTR_SYSCALL)) { 1181 if (!mtx_owned(&Giant)) 1182 mtx_enter(&Giant, MTX_DEF); 1183 ktrsyscall(p->p_tracep, code, narg, args); 1184 } 1185#endif 1186 p->p_retval[0] = 0; 1187 p->p_retval[1] = frame.tf_edx; 1188 1189 STOPEVENT(p, S_SCE, narg); /* MP aware */ 1190 1191 error = (*callp->sy_call)(p, args); 1192 1193 /* 1194 * MP SAFE (we may or may not have the MP lock at this point) 1195 */ 1196 switch (error) { 1197 case 0: 1198 frame.tf_eax = p->p_retval[0]; 1199 frame.tf_edx = p->p_retval[1]; 1200 frame.tf_eflags &= ~PSL_C; 1201 break; 1202 1203 case ERESTART: 1204 /* 1205 * Reconstruct pc, assuming lcall $X,y is 7 bytes, 1206 * int 0x80 is 2 bytes. We saved this in tf_err. 1207 */ 1208 frame.tf_eip -= frame.tf_err; 1209 break; 1210 1211 case EJUSTRETURN: 1212 break; 1213 1214 default: 1215bad: 1216 if (p->p_sysent->sv_errsize) { 1217 if (error >= p->p_sysent->sv_errsize) 1218 error = -1; /* XXX */ 1219 else 1220 error = p->p_sysent->sv_errtbl[error]; 1221 } 1222 frame.tf_eax = error; 1223 frame.tf_eflags |= PSL_C; 1224 break; 1225 } 1226 1227 /* 1228 * Traced syscall. trapsignal() is not MP aware. 1229 */ 1230 if ((frame.tf_eflags & PSL_T) && !(frame.tf_eflags & PSL_VM)) { 1231 if (!mtx_owned(&Giant)) 1232 mtx_enter(&Giant, MTX_DEF); 1233 frame.tf_eflags &= ~PSL_T; 1234 trapsignal(p, SIGTRAP, 0); 1235 } 1236 1237 /* 1238 * Handle reschedule and other end-of-syscall issues 1239 */ 1240 userret(p, &frame, sticks); 1241 1242#ifdef KTRACE 1243 if (KTRPOINT(p, KTR_SYSRET)) { 1244 if (!mtx_owned(&Giant)) 1245 mtx_enter(&Giant, MTX_DEF); 1246 ktrsysret(p->p_tracep, code, error, p->p_retval[0]); 1247 } 1248#endif 1249 1250 /* 1251 * This works because errno is findable through the 1252 * register set. If we ever support an emulation where this 1253 * is not the case, this code will need to be revisited. 1254 */ 1255 STOPEVENT(p, S_SCX, code); 1256 1257 /* 1258 * Release Giant if we had to get it 1259 */ 1260 if (mtx_owned(&Giant)) 1261 mtx_exit(&Giant, MTX_DEF); 1262 1263#ifdef WITNESS 1264 if (witness_list(p)) { 1265 panic("system call %s returning with mutex(s) held\n", 1266 syscallnames[code]); 1267 } 1268#endif 1269 mtx_assert(&sched_lock, MA_NOTOWNED); 1270 mtx_assert(&Giant, MA_NOTOWNED); 1271} 1272 1273void 1274ast(frame) 1275 struct trapframe frame; 1276{ 1277 struct proc *p = CURPROC; 1278 u_quad_t sticks; 1279 1280 mtx_enter(&sched_lock, MTX_SPIN); 1281 sticks = p->p_sticks; 1282 1283 astoff(); 1284 atomic_add_int(&cnt.v_soft, 1); 1285 if (p->p_sflag & PS_OWEUPC) { 1286 p->p_sflag &= ~PS_OWEUPC; 1287 mtx_exit(&sched_lock, MTX_SPIN); 1288 mtx_enter(&Giant, MTX_DEF); 1289 mtx_enter(&sched_lock, MTX_SPIN); 1290 addupc_task(p, p->p_stats->p_prof.pr_addr, 1291 p->p_stats->p_prof.pr_ticks); 1292 } 1293 if (p->p_sflag & PS_ALRMPEND) { 1294 p->p_sflag &= ~PS_ALRMPEND; 1295 mtx_exit(&sched_lock, MTX_SPIN); 1296 if (!mtx_owned(&Giant)) 1297 mtx_enter(&Giant, MTX_DEF); 1298 psignal(p, SIGVTALRM); 1299 mtx_enter(&sched_lock, MTX_SPIN); 1300 } 1301 if (p->p_sflag & PS_PROFPEND) { 1302 p->p_sflag &= ~PS_PROFPEND; 1303 mtx_exit(&sched_lock, MTX_SPIN); 1304 if (!mtx_owned(&Giant)) 1305 mtx_enter(&Giant, MTX_DEF); 1306 psignal(p, SIGPROF); 1307 } else 1308 mtx_exit(&sched_lock, MTX_SPIN); 1309 1310 userret(p, &frame, sticks); 1311 1312 if (mtx_owned(&Giant)) 1313 mtx_exit(&Giant, MTX_DEF); 1314} 1315