subr_syscall.c revision 78962
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 78962 2001-06-29 11:10:41Z jhb $ 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/kernel.h> 59#include <sys/ktr.h> 60#include <sys/mutex.h> 61#include <sys/resourcevar.h> 62#include <sys/signalvar.h> 63#include <sys/syscall.h> 64#include <sys/sysctl.h> 65#include <sys/sysent.h> 66#include <sys/uio.h> 67#include <sys/vmmeter.h> 68#ifdef KTRACE 69#include <sys/ktrace.h> 70#endif 71 72#include <vm/vm.h> 73#include <vm/vm_param.h> 74#include <sys/lock.h> 75#include <vm/pmap.h> 76#include <vm/vm_kern.h> 77#include <vm/vm_map.h> 78#include <vm/vm_page.h> 79#include <vm/vm_extern.h> 80 81#include <machine/cpu.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> 88 89#include <i386/isa/icu.h> 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> 100 101#include <sys/sysctl.h> 102 103int (*pmath_emulate) __P((struct trapframe *)); 104 105extern void trap __P((struct trapframe frame)); 106extern int trapwrite __P((unsigned addr)); 107extern void syscall __P((struct trapframe frame)); 108extern void ast __P((struct trapframe *framep)); 109 110static int trap_pfault __P((struct trapframe *, int, vm_offset_t)); 111static void trap_fatal __P((struct trapframe *, vm_offset_t)); 112void dblfault_handler __P((void)); 113 114extern inthand_t IDTVEC(lcall_syscall); 115 116#define MAX_TRAP_MSG 28 117static char *trap_msg[] = { 118 "", /* 0 unused */ 119 "privileged instruction fault", /* 1 T_PRIVINFLT */ 120 "", /* 2 unused */ 121 "breakpoint instruction fault", /* 3 T_BPTFLT */ 122 "", /* 4 unused */ 123 "", /* 5 unused */ 124 "arithmetic trap", /* 6 T_ARITHTRAP */ 125 "", /* 7 unused */ 126 "", /* 8 unused */ 127 "general protection fault", /* 9 T_PROTFLT */ 128 "trace trap", /* 10 T_TRCTRAP */ 129 "", /* 11 unused */ 130 "page fault", /* 12 T_PAGEFLT */ 131 "", /* 13 unused */ 132 "alignment fault", /* 14 T_ALIGNFLT */ 133 "", /* 15 unused */ 134 "", /* 16 unused */ 135 "", /* 17 unused */ 136 "integer divide fault", /* 18 T_DIVIDE */ 137 "non-maskable interrupt trap", /* 19 T_NMI */ 138 "overflow trap", /* 20 T_OFLOW */ 139 "FPU bounds check fault", /* 21 T_BOUND */ 140 "FPU device not available", /* 22 T_DNA */ 141 "double fault", /* 23 T_DOUBLEFLT */ 142 "FPU operand fetch fault", /* 24 T_FPOPFLT */ 143 "invalid TSS fault", /* 25 T_TSSFLT */ 144 "segment not present fault", /* 26 T_SEGNPFLT */ 145 "stack fault", /* 27 T_STKFLT */ 146 "machine check trap", /* 28 T_MCHK */ 147}; 148 149#if defined(I586_CPU) && !defined(NO_F00F_HACK) 150extern int has_f00f_bug; 151#endif 152 153#ifdef DDB 154static int ddb_on_nmi = 1; 155SYSCTL_INT(_machdep, OID_AUTO, ddb_on_nmi, CTLFLAG_RW, 156 &ddb_on_nmi, 0, "Go to DDB on NMI"); 157#endif 158static int panic_on_nmi = 1; 159SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RW, 160 &panic_on_nmi, 0, "Panic on NMI"); 161 162#ifdef WITNESS 163extern char *syscallnames[]; 164#endif 165 166void 167userret(p, frame, oticks) 168 struct proc *p; 169 struct trapframe *frame; 170 u_quad_t oticks; 171{ 172 int sig; 173 174 PROC_LOCK(p); 175 while ((sig = CURSIG(p)) != 0) 176 postsig(sig); 177 178 mtx_lock_spin(&sched_lock); 179 PROC_UNLOCK_NOSWITCH(p); 180 p->p_pri.pri_level = p->p_pri.pri_user; 181 if (resched_wanted(p)) { 182 /* 183 * Since we are curproc, clock will normally just change 184 * our priority without moving us from one queue to another 185 * (since the running process is not on a queue.) 186 * If that happened after we setrunqueue ourselves but before we 187 * mi_switch()'ed, we might not be on the queue indicated by 188 * our priority. 189 */ 190 DROP_GIANT_NOSWITCH(); 191 setrunqueue(p); 192 p->p_stats->p_ru.ru_nivcsw++; 193 mi_switch(); 194 mtx_unlock_spin(&sched_lock); 195 PICKUP_GIANT(); 196 PROC_LOCK(p); 197 while ((sig = CURSIG(p)) != 0) 198 postsig(sig); 199 mtx_lock_spin(&sched_lock); 200 PROC_UNLOCK_NOSWITCH(p); 201 } 202 203 /* 204 * Charge system time if profiling. 205 */ 206 if (p->p_sflag & PS_PROFIL) { 207 mtx_unlock_spin(&sched_lock); 208 addupc_task(p, TRAPF_PC(frame), 209 (u_int)(p->p_sticks - oticks) * psratio); 210 } else 211 mtx_unlock_spin(&sched_lock); 212} 213 214/* 215 * Exception, fault, and trap interface to the FreeBSD kernel. 216 * This common code is called from assembly language IDT gate entry 217 * routines that prepare a suitable stack frame, and restore this 218 * frame after the exception has been processed. 219 */ 220 221void 222trap(frame) 223 struct trapframe frame; 224{ 225 struct proc *p = curproc; 226 u_quad_t sticks = 0; 227 int i = 0, ucode = 0, type, code; 228 vm_offset_t eva; 229#ifdef POWERFAIL_NMI 230 static int lastalert = 0; 231#endif 232 233 atomic_add_int(&cnt.v_trap, 1); 234 235 if ((frame.tf_eflags & PSL_I) == 0) { 236 /* 237 * Buggy application or kernel code has disabled 238 * interrupts and then trapped. Enabling interrupts 239 * now is wrong, but it is better than running with 240 * interrupts disabled until they are accidentally 241 * enabled later. XXX This is really bad if we trap 242 * while holding a spin lock. 243 */ 244 type = frame.tf_trapno; 245 if (ISPL(frame.tf_cs) == SEL_UPL || (frame.tf_eflags & PSL_VM)) 246 printf( 247 "pid %ld (%s): trap %d with interrupts disabled\n", 248 (long)curproc->p_pid, curproc->p_comm, type); 249 else if (type != T_BPTFLT && type != T_TRCTRAP) { 250 /* 251 * XXX not quite right, since this may be for a 252 * multiple fault in user mode. 253 */ 254 printf("kernel trap %d with interrupts disabled\n", 255 type); 256 /* 257 * We should walk p_heldmtx here and see if any are 258 * spin mutexes, and not do this if so. 259 */ 260 enable_intr(); 261 } 262 } 263 264 eva = 0; 265 266#if defined(I586_CPU) && !defined(NO_F00F_HACK) 267restart: 268#endif 269 270 type = frame.tf_trapno; 271 code = frame.tf_err; 272 273 if ((ISPL(frame.tf_cs) == SEL_UPL) || 274 ((frame.tf_eflags & PSL_VM) && !in_vm86call)) { 275 /* user trap */ 276 277 mtx_lock_spin(&sched_lock); 278 sticks = p->p_sticks; 279 mtx_unlock_spin(&sched_lock); 280 p->p_frame = &frame; 281 282 switch (type) { 283 case T_PRIVINFLT: /* privileged instruction fault */ 284 ucode = type; 285 i = SIGILL; 286 break; 287 288 case T_BPTFLT: /* bpt instruction fault */ 289 case T_TRCTRAP: /* trace trap */ 290 frame.tf_eflags &= ~PSL_T; 291 i = SIGTRAP; 292 break; 293 294 case T_ARITHTRAP: /* arithmetic trap */ 295#ifdef DEV_NPX 296 ucode = npxtrap(); 297 if (ucode == -1) 298 return; 299#else 300 ucode = code; 301#endif 302 i = SIGFPE; 303 break; 304 305 /* 306 * The following two traps can happen in 307 * vm86 mode, and, if so, we want to handle 308 * them specially. 309 */ 310 case T_PROTFLT: /* general protection fault */ 311 case T_STKFLT: /* stack fault */ 312 if (frame.tf_eflags & PSL_VM) { 313 mtx_lock(&Giant); 314 i = vm86_emulate((struct vm86frame *)&frame); 315 mtx_unlock(&Giant); 316 if (i == 0) 317 goto user; 318 break; 319 } 320 /* FALL THROUGH */ 321 322 case T_SEGNPFLT: /* segment not present fault */ 323 case T_TSSFLT: /* invalid TSS fault */ 324 case T_DOUBLEFLT: /* double fault */ 325 default: 326 ucode = code + BUS_SEGM_FAULT ; 327 i = SIGBUS; 328 break; 329 330 case T_PAGEFLT: /* page fault */ 331 /* 332 * For some Cyrix CPUs, %cr2 is clobbered by 333 * interrupts. This problem is worked around by using 334 * an interrupt gate for the pagefault handler. We 335 * are finally ready to read %cr2 and then must 336 * reenable interrupts. 337 */ 338 eva = rcr2(); 339 enable_intr(); 340 mtx_lock(&Giant); 341 i = trap_pfault(&frame, TRUE, eva); 342 mtx_unlock(&Giant); 343#if defined(I586_CPU) && !defined(NO_F00F_HACK) 344 if (i == -2) { 345 /* 346 * f00f hack workaround has triggered, treat 347 * as illegal instruction not page fault. 348 */ 349 frame.tf_trapno = T_PRIVINFLT; 350 goto restart; 351 } 352#endif 353 if (i == -1) 354 goto out; 355 if (i == 0) 356 goto user; 357 358 ucode = T_PAGEFLT; 359 break; 360 361 case T_DIVIDE: /* integer divide fault */ 362 ucode = FPE_INTDIV; 363 i = SIGFPE; 364 break; 365 366#ifdef DEV_ISA 367 case T_NMI: 368#ifdef POWERFAIL_NMI 369#ifndef TIMER_FREQ 370# define TIMER_FREQ 1193182 371#endif 372 mtx_lock(&Giant); 373 if (time_second - lastalert > 10) { 374 log(LOG_WARNING, "NMI: power fail\n"); 375 sysbeep(TIMER_FREQ/880, hz); 376 lastalert = time_second; 377 } 378 mtx_unlock(&Giant); 379 goto out; 380#else /* !POWERFAIL_NMI */ 381 /* machine/parity/power fail/"kitchen sink" faults */ 382 /* XXX Giant */ 383 if (isa_nmi(code) == 0) { 384#ifdef DDB 385 /* 386 * NMI can be hooked up to a pushbutton 387 * for debugging. 388 */ 389 if (ddb_on_nmi) { 390 printf ("NMI ... going to debugger\n"); 391 kdb_trap (type, 0, &frame); 392 } 393#endif /* DDB */ 394 goto out; 395 } else if (panic_on_nmi) 396 panic("NMI indicates hardware failure"); 397 break; 398#endif /* POWERFAIL_NMI */ 399#endif /* DEV_ISA */ 400 401 case T_OFLOW: /* integer overflow fault */ 402 ucode = FPE_INTOVF; 403 i = SIGFPE; 404 break; 405 406 case T_BOUND: /* bounds check fault */ 407 ucode = FPE_FLTSUB; 408 i = SIGFPE; 409 break; 410 411 case T_DNA: 412#ifdef DEV_NPX 413 /* transparent fault (due to context switch "late") */ 414 if (npxdna()) 415 goto out; 416#endif 417 if (!pmath_emulate) { 418 i = SIGFPE; 419 ucode = FPE_FPU_NP_TRAP; 420 break; 421 } 422 mtx_lock(&Giant); 423 i = (*pmath_emulate)(&frame); 424 mtx_unlock(&Giant); 425 if (i == 0) { 426 if (!(frame.tf_eflags & PSL_T)) 427 goto out; 428 frame.tf_eflags &= ~PSL_T; 429 i = SIGTRAP; 430 } 431 /* else ucode = emulator_only_knows() XXX */ 432 break; 433 434 case T_FPOPFLT: /* FPU operand fetch fault */ 435 ucode = T_FPOPFLT; 436 i = SIGILL; 437 break; 438 } 439 } else { 440 /* kernel trap */ 441 442 switch (type) { 443 case T_PAGEFLT: /* page fault */ 444 /* 445 * For some Cyrix CPUs, %cr2 is clobbered by 446 * interrupts. This problem is worked around by using 447 * an interrupt gate for the pagefault handler. We 448 * are finally ready to read %cr2 and then must 449 * reenable interrupts. 450 */ 451 eva = rcr2(); 452 enable_intr(); 453 mtx_lock(&Giant); 454 (void) trap_pfault(&frame, FALSE, eva); 455 mtx_unlock(&Giant); 456 goto out; 457 458 case T_DNA: 459#ifdef DEV_NPX 460 /* 461 * The kernel is apparently using npx for copying. 462 * XXX this should be fatal unless the kernel has 463 * registered such use. 464 */ 465 if (npxdna()) 466 goto out; 467#endif 468 break; 469 470 /* 471 * The following two traps can happen in 472 * vm86 mode, and, if so, we want to handle 473 * them specially. 474 */ 475 case T_PROTFLT: /* general protection fault */ 476 case T_STKFLT: /* stack fault */ 477 if (frame.tf_eflags & PSL_VM) { 478 mtx_lock(&Giant); 479 i = vm86_emulate((struct vm86frame *)&frame); 480 mtx_unlock(&Giant); 481 if (i != 0) 482 /* 483 * returns to original process 484 */ 485 vm86_trap((struct vm86frame *)&frame); 486 goto out; 487 } 488 if (type == T_STKFLT) 489 break; 490 491 /* FALL THROUGH */ 492 493 case T_SEGNPFLT: /* segment not present fault */ 494 if (in_vm86call) 495 break; 496 497 if (p->p_intr_nesting_level != 0) 498 break; 499 500 /* 501 * Invalid %fs's and %gs's can be created using 502 * procfs or PT_SETREGS or by invalidating the 503 * underlying LDT entry. This causes a fault 504 * in kernel mode when the kernel attempts to 505 * switch contexts. Lose the bad context 506 * (XXX) so that we can continue, and generate 507 * a signal. 508 */ 509 if (frame.tf_eip == (int)cpu_switch_load_gs) { 510 PCPU_GET(curpcb)->pcb_gs = 0; 511 PROC_LOCK(p); 512 psignal(p, SIGBUS); 513 PROC_UNLOCK(p); 514 goto out; 515 } 516 517 /* 518 * Invalid segment selectors and out of bounds 519 * %eip's and %esp's can be set up in user mode. 520 * This causes a fault in kernel mode when the 521 * kernel tries to return to user mode. We want 522 * to get this fault so that we can fix the 523 * problem here and not have to check all the 524 * selectors and pointers when the user changes 525 * them. 526 */ 527 if (frame.tf_eip == (int)doreti_iret) { 528 frame.tf_eip = (int)doreti_iret_fault; 529 goto out; 530 } 531 if (frame.tf_eip == (int)doreti_popl_ds) { 532 frame.tf_eip = (int)doreti_popl_ds_fault; 533 goto out; 534 } 535 if (frame.tf_eip == (int)doreti_popl_es) { 536 frame.tf_eip = (int)doreti_popl_es_fault; 537 goto out; 538 } 539 if (frame.tf_eip == (int)doreti_popl_fs) { 540 frame.tf_eip = (int)doreti_popl_fs_fault; 541 goto out; 542 } 543 if (PCPU_GET(curpcb) != NULL && 544 PCPU_GET(curpcb)->pcb_onfault != NULL) { 545 frame.tf_eip = 546 (int)PCPU_GET(curpcb)->pcb_onfault; 547 goto out; 548 } 549 break; 550 551 case T_TSSFLT: 552 /* 553 * PSL_NT can be set in user mode and isn't cleared 554 * automatically when the kernel is entered. This 555 * causes a TSS fault when the kernel attempts to 556 * `iret' because the TSS link is uninitialized. We 557 * want to get this fault so that we can fix the 558 * problem here and not every time the kernel is 559 * entered. 560 */ 561 if (frame.tf_eflags & PSL_NT) { 562 frame.tf_eflags &= ~PSL_NT; 563 goto out; 564 } 565 break; 566 567 case T_TRCTRAP: /* trace trap */ 568 if (frame.tf_eip == (int)IDTVEC(lcall_syscall)) { 569 /* 570 * We've just entered system mode via the 571 * syscall lcall. Continue single stepping 572 * silently until the syscall handler has 573 * saved the flags. 574 */ 575 goto out; 576 } 577 if (frame.tf_eip == (int)IDTVEC(lcall_syscall) + 1) { 578 /* 579 * The syscall handler has now saved the 580 * flags. Stop single stepping it. 581 */ 582 frame.tf_eflags &= ~PSL_T; 583 goto out; 584 } 585 /* 586 * Ignore debug register trace traps due to 587 * accesses in the user's address space, which 588 * can happen under several conditions such as 589 * if a user sets a watchpoint on a buffer and 590 * then passes that buffer to a system call. 591 * We still want to get TRCTRAPS for addresses 592 * in kernel space because that is useful when 593 * debugging the kernel. 594 */ 595 /* XXX Giant */ 596 if (user_dbreg_trap() && !in_vm86call) { 597 /* 598 * Reset breakpoint bits because the 599 * processor doesn't 600 */ 601 load_dr6(rdr6() & 0xfffffff0); 602 goto out; 603 } 604 /* 605 * Fall through (TRCTRAP kernel mode, kernel address) 606 */ 607 case T_BPTFLT: 608 /* 609 * If DDB is enabled, let it handle the debugger trap. 610 * Otherwise, debugger traps "can't happen". 611 */ 612#ifdef DDB 613 /* XXX Giant */ 614 if (kdb_trap (type, 0, &frame)) 615 goto out; 616#endif 617 break; 618 619#ifdef DEV_ISA 620 case T_NMI: 621#ifdef POWERFAIL_NMI 622 mtx_lock(&Giant); 623 if (time_second - lastalert > 10) { 624 log(LOG_WARNING, "NMI: power fail\n"); 625 sysbeep(TIMER_FREQ/880, hz); 626 lastalert = time_second; 627 } 628 mtx_unlock(&Giant); 629 goto out; 630#else /* !POWERFAIL_NMI */ 631 /* XXX Giant */ 632 /* machine/parity/power fail/"kitchen sink" faults */ 633 if (isa_nmi(code) == 0) { 634#ifdef DDB 635 /* 636 * NMI can be hooked up to a pushbutton 637 * for debugging. 638 */ 639 if (ddb_on_nmi) { 640 printf ("NMI ... going to debugger\n"); 641 kdb_trap (type, 0, &frame); 642 } 643#endif /* DDB */ 644 goto out; 645 } else if (panic_on_nmi == 0) 646 goto out; 647 /* FALL THROUGH */ 648#endif /* POWERFAIL_NMI */ 649#endif /* DEV_ISA */ 650 } 651 652 trap_fatal(&frame, eva); 653 goto out; 654 } 655 656 mtx_lock(&Giant); 657 /* Translate fault for emulators (e.g. Linux) */ 658 if (*p->p_sysent->sv_transtrap) 659 i = (*p->p_sysent->sv_transtrap)(i, type); 660 661 trapsignal(p, i, ucode); 662 663#ifdef DEBUG 664 if (type <= MAX_TRAP_MSG) { 665 uprintf("fatal process exception: %s", 666 trap_msg[type]); 667 if ((type == T_PAGEFLT) || (type == T_PROTFLT)) 668 uprintf(", fault VA = 0x%lx", (u_long)eva); 669 uprintf("\n"); 670 } 671#endif 672 mtx_unlock(&Giant); 673 674user: 675 userret(p, &frame, sticks); 676 if (mtx_owned(&Giant)) 677 mtx_unlock(&Giant); 678out: 679 return; 680} 681 682#ifdef notyet 683/* 684 * This version doesn't allow a page fault to user space while 685 * in the kernel. The rest of the kernel needs to be made "safe" 686 * before this can be used. I think the only things remaining 687 * to be made safe are the iBCS2 code and the process tracing/ 688 * debugging code. 689 */ 690static int 691trap_pfault(frame, usermode, eva) 692 struct trapframe *frame; 693 int usermode; 694 vm_offset_t eva; 695{ 696 vm_offset_t va; 697 struct vmspace *vm = NULL; 698 vm_map_t map = 0; 699 int rv = 0; 700 vm_prot_t ftype; 701 struct proc *p = curproc; 702 703 if (frame->tf_err & PGEX_W) 704 ftype = VM_PROT_WRITE; 705 else 706 ftype = VM_PROT_READ; 707 708 va = trunc_page(eva); 709 if (va < VM_MIN_KERNEL_ADDRESS) { 710 vm_offset_t v; 711 vm_page_t mpte; 712 713 if (p == NULL || 714 (!usermode && va < VM_MAXUSER_ADDRESS && 715 (p->p_intr_nesting_level != 0 || 716 PCPU_GET(curpcb) == NULL || 717 PCPU_GET(curpcb)->pcb_onfault == NULL))) { 718 trap_fatal(frame, eva); 719 return (-1); 720 } 721 722 /* 723 * This is a fault on non-kernel virtual memory. 724 * vm is initialized above to NULL. If curproc is NULL 725 * or curproc->p_vmspace is NULL the fault is fatal. 726 */ 727 vm = p->p_vmspace; 728 if (vm == NULL) 729 goto nogo; 730 731 map = &vm->vm_map; 732 733 /* 734 * Keep swapout from messing with us during this 735 * critical time. 736 */ 737 PROC_LOCK(p); 738 ++p->p_lock; 739 PROC_UNLOCK(p); 740 741 /* 742 * Grow the stack if necessary 743 */ 744 /* grow_stack returns false only if va falls into 745 * a growable stack region and the stack growth 746 * fails. It returns true if va was not within 747 * a growable stack region, or if the stack 748 * growth succeeded. 749 */ 750 if (!grow_stack (p, va)) 751 rv = KERN_FAILURE; 752 else 753 /* Fault in the user page: */ 754 rv = vm_fault(map, va, ftype, 755 (ftype & VM_PROT_WRITE) ? VM_FAULT_DIRTY 756 : VM_FAULT_NORMAL); 757 758 PROC_LOCK(p); 759 --p->p_lock; 760 PROC_UNLOCK(p); 761 } else { 762 /* 763 * Don't allow user-mode faults in kernel address space. 764 */ 765 if (usermode) 766 goto nogo; 767 768 /* 769 * Since we know that kernel virtual address addresses 770 * always have pte pages mapped, we just have to fault 771 * the page. 772 */ 773 rv = vm_fault(kernel_map, va, ftype, VM_FAULT_NORMAL); 774 } 775 776 if (rv == KERN_SUCCESS) 777 return (0); 778nogo: 779 if (!usermode) { 780 if (p->p_intr_nesting_level == 0 && 781 PCPU_GET(curpcb) != NULL && 782 PCPU_GET(curpcb)->pcb_onfault != NULL) { 783 frame->tf_eip = (int)PCPU_GET(curpcb)->pcb_onfault; 784 return (0); 785 } 786 trap_fatal(frame, eva); 787 return (-1); 788 } 789 790 /* kludge to pass faulting virtual address to sendsig */ 791 frame->tf_err = eva; 792 793 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV); 794} 795#endif 796 797int 798trap_pfault(frame, usermode, eva) 799 struct trapframe *frame; 800 int usermode; 801 vm_offset_t eva; 802{ 803 vm_offset_t va; 804 struct vmspace *vm = NULL; 805 vm_map_t map = 0; 806 int rv = 0; 807 vm_prot_t ftype; 808 struct proc *p = curproc; 809 810 va = trunc_page(eva); 811 if (va >= KERNBASE) { 812 /* 813 * Don't allow user-mode faults in kernel address space. 814 * An exception: if the faulting address is the invalid 815 * instruction entry in the IDT, then the Intel Pentium 816 * F00F bug workaround was triggered, and we need to 817 * treat it is as an illegal instruction, and not a page 818 * fault. 819 */ 820#if defined(I586_CPU) && !defined(NO_F00F_HACK) 821 if ((eva == (unsigned int)&idt[6]) && has_f00f_bug) 822 return -2; 823#endif 824 if (usermode) 825 goto nogo; 826 827 map = kernel_map; 828 } else { 829 /* 830 * This is a fault on non-kernel virtual memory. 831 * vm is initialized above to NULL. If curproc is NULL 832 * or curproc->p_vmspace is NULL the fault is fatal. 833 */ 834 if (p != NULL) 835 vm = p->p_vmspace; 836 837 if (vm == NULL) 838 goto nogo; 839 840 map = &vm->vm_map; 841 } 842 843 if (frame->tf_err & PGEX_W) 844 ftype = VM_PROT_WRITE; 845 else 846 ftype = VM_PROT_READ; 847 848 if (map != kernel_map) { 849 /* 850 * Keep swapout from messing with us during this 851 * critical time. 852 */ 853 PROC_LOCK(p); 854 ++p->p_lock; 855 PROC_UNLOCK(p); 856 857 /* 858 * Grow the stack if necessary 859 */ 860 /* grow_stack returns false only if va falls into 861 * a growable stack region and the stack growth 862 * fails. It returns true if va was not within 863 * a growable stack region, or if the stack 864 * growth succeeded. 865 */ 866 if (!grow_stack (p, va)) 867 rv = KERN_FAILURE; 868 else 869 /* Fault in the user page: */ 870 rv = vm_fault(map, va, ftype, 871 (ftype & VM_PROT_WRITE) ? VM_FAULT_DIRTY 872 : VM_FAULT_NORMAL); 873 874 PROC_LOCK(p); 875 --p->p_lock; 876 PROC_UNLOCK(p); 877 } else { 878 /* 879 * Don't have to worry about process locking or stacks in the 880 * kernel. 881 */ 882 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL); 883 } 884 885 if (rv == KERN_SUCCESS) 886 return (0); 887nogo: 888 if (!usermode) { 889 if (p->p_intr_nesting_level == 0 && 890 PCPU_GET(curpcb) != NULL && 891 PCPU_GET(curpcb)->pcb_onfault != NULL) { 892 frame->tf_eip = (int)PCPU_GET(curpcb)->pcb_onfault; 893 return (0); 894 } 895 trap_fatal(frame, eva); 896 return (-1); 897 } 898 899 /* kludge to pass faulting virtual address to sendsig */ 900 frame->tf_err = eva; 901 902 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV); 903} 904 905static void 906trap_fatal(frame, eva) 907 struct trapframe *frame; 908 vm_offset_t eva; 909{ 910 int code, type, ss, esp; 911 struct soft_segment_descriptor softseg; 912 913 code = frame->tf_err; 914 type = frame->tf_trapno; 915 sdtossd(&gdt[IDXSEL(frame->tf_cs & 0xffff)].sd, &softseg); 916 917 if (type <= MAX_TRAP_MSG) 918 printf("\n\nFatal trap %d: %s while in %s mode\n", 919 type, trap_msg[type], 920 frame->tf_eflags & PSL_VM ? "vm86" : 921 ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel"); 922#ifdef SMP 923 /* two separate prints in case of a trap on an unmapped page */ 924 printf("cpuid = %d; ", PCPU_GET(cpuid)); 925 printf("lapic.id = %08x\n", lapic.id); 926#endif 927 if (type == T_PAGEFLT) { 928 printf("fault virtual address = 0x%x\n", eva); 929 printf("fault code = %s %s, %s\n", 930 code & PGEX_U ? "user" : "supervisor", 931 code & PGEX_W ? "write" : "read", 932 code & PGEX_P ? "protection violation" : "page not present"); 933 } 934 printf("instruction pointer = 0x%x:0x%x\n", 935 frame->tf_cs & 0xffff, frame->tf_eip); 936 if ((ISPL(frame->tf_cs) == SEL_UPL) || (frame->tf_eflags & PSL_VM)) { 937 ss = frame->tf_ss & 0xffff; 938 esp = frame->tf_esp; 939 } else { 940 ss = GSEL(GDATA_SEL, SEL_KPL); 941 esp = (int)&frame->tf_esp; 942 } 943 printf("stack pointer = 0x%x:0x%x\n", ss, esp); 944 printf("frame pointer = 0x%x:0x%x\n", ss, frame->tf_ebp); 945 printf("code segment = base 0x%x, limit 0x%x, type 0x%x\n", 946 softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type); 947 printf(" = DPL %d, pres %d, def32 %d, gran %d\n", 948 softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_def32, 949 softseg.ssd_gran); 950 printf("processor eflags = "); 951 if (frame->tf_eflags & PSL_T) 952 printf("trace trap, "); 953 if (frame->tf_eflags & PSL_I) 954 printf("interrupt enabled, "); 955 if (frame->tf_eflags & PSL_NT) 956 printf("nested task, "); 957 if (frame->tf_eflags & PSL_RF) 958 printf("resume, "); 959 if (frame->tf_eflags & PSL_VM) 960 printf("vm86, "); 961 printf("IOPL = %d\n", (frame->tf_eflags & PSL_IOPL) >> 12); 962 printf("current process = "); 963 if (curproc) { 964 printf("%lu (%s)\n", 965 (u_long)curproc->p_pid, curproc->p_comm ? 966 curproc->p_comm : ""); 967 } else { 968 printf("Idle\n"); 969 } 970 971#ifdef KDB 972 if (kdb_trap(&psl)) 973 return; 974#endif 975#ifdef DDB 976 if ((debugger_on_panic || db_active) && kdb_trap(type, 0, frame)) 977 return; 978#endif 979 printf("trap number = %d\n", type); 980 if (type <= MAX_TRAP_MSG) 981 panic(trap_msg[type]); 982 else 983 panic("unknown/reserved trap"); 984} 985 986/* 987 * Double fault handler. Called when a fault occurs while writing 988 * a frame for a trap/exception onto the stack. This usually occurs 989 * when the stack overflows (such is the case with infinite recursion, 990 * for example). 991 * 992 * XXX Note that the current PTD gets replaced by IdlePTD when the 993 * task switch occurs. This means that the stack that was active at 994 * the time of the double fault is not available at <kstack> unless 995 * the machine was idle when the double fault occurred. The downside 996 * of this is that "trace <ebp>" in ddb won't work. 997 */ 998void 999dblfault_handler() 1000{ 1001 printf("\nFatal double fault:\n"); 1002 printf("eip = 0x%x\n", PCPU_GET(common_tss.tss_eip)); 1003 printf("esp = 0x%x\n", PCPU_GET(common_tss.tss_esp)); 1004 printf("ebp = 0x%x\n", PCPU_GET(common_tss.tss_ebp)); 1005#ifdef SMP 1006 /* two separate prints in case of a trap on an unmapped page */ 1007 printf("cpuid = %d; ", PCPU_GET(cpuid)); 1008 printf("lapic.id = %08x\n", lapic.id); 1009#endif 1010 panic("double fault"); 1011} 1012 1013/* 1014 * Compensate for 386 brain damage (missing URKR). 1015 * This is a little simpler than the pagefault handler in trap() because 1016 * it the page tables have already been faulted in and high addresses 1017 * are thrown out early for other reasons. 1018 */ 1019int trapwrite(addr) 1020 unsigned addr; 1021{ 1022 struct proc *p; 1023 vm_offset_t va; 1024 struct vmspace *vm; 1025 int rv; 1026 1027 va = trunc_page((vm_offset_t)addr); 1028 /* 1029 * XXX - MAX is END. Changed > to >= for temp. fix. 1030 */ 1031 if (va >= VM_MAXUSER_ADDRESS) 1032 return (1); 1033 1034 p = curproc; 1035 vm = p->p_vmspace; 1036 1037 PROC_LOCK(p); 1038 ++p->p_lock; 1039 PROC_UNLOCK(p); 1040 1041 if (!grow_stack (p, va)) 1042 rv = KERN_FAILURE; 1043 else 1044 /* 1045 * fault the data page 1046 */ 1047 rv = vm_fault(&vm->vm_map, va, VM_PROT_WRITE, VM_FAULT_DIRTY); 1048 1049 PROC_LOCK(p); 1050 --p->p_lock; 1051 PROC_UNLOCK(p); 1052 1053 if (rv != KERN_SUCCESS) 1054 return 1; 1055 1056 return (0); 1057} 1058 1059/* 1060 * syscall - MP aware system call request C handler 1061 * 1062 * A system call is essentially treated as a trap except that the 1063 * MP lock is not held on entry or return. We are responsible for 1064 * obtaining the MP lock if necessary and for handling ASTs 1065 * (e.g. a task switch) prior to return. 1066 * 1067 * In general, only simple access and manipulation of curproc and 1068 * the current stack is allowed without having to hold MP lock. 1069 */ 1070void 1071syscall(frame) 1072 struct trapframe frame; 1073{ 1074 caddr_t params; 1075 int i; 1076 struct sysent *callp; 1077 struct proc *p = curproc; 1078 u_quad_t sticks; 1079 int error; 1080 int narg; 1081 int args[8]; 1082 u_int code; 1083 1084 atomic_add_int(&cnt.v_syscall, 1); 1085 1086#ifdef DIAGNOSTIC 1087 if (ISPL(frame.tf_cs) != SEL_UPL) { 1088 mtx_lock(&Giant); 1089 panic("syscall"); 1090 /* NOT REACHED */ 1091 } 1092#endif 1093 1094 mtx_lock_spin(&sched_lock); 1095 sticks = p->p_sticks; 1096 mtx_unlock_spin(&sched_lock); 1097 1098 p->p_frame = &frame; 1099 params = (caddr_t)frame.tf_esp + sizeof(int); 1100 code = frame.tf_eax; 1101 1102 if (p->p_sysent->sv_prepsyscall) { 1103 /* 1104 * The prep code is not MP aware. 1105 */ 1106 mtx_lock(&Giant); 1107 (*p->p_sysent->sv_prepsyscall)(&frame, args, &code, ¶ms); 1108 mtx_unlock(&Giant); 1109 } else { 1110 /* 1111 * Need to check if this is a 32 bit or 64 bit syscall. 1112 * fuword is MP aware. 1113 */ 1114 if (code == SYS_syscall) { 1115 /* 1116 * Code is first argument, followed by actual args. 1117 */ 1118 code = fuword(params); 1119 params += sizeof(int); 1120 } else if (code == SYS___syscall) { 1121 /* 1122 * Like syscall, but code is a quad, so as to maintain 1123 * quad alignment for the rest of the arguments. 1124 */ 1125 code = fuword(params); 1126 params += sizeof(quad_t); 1127 } 1128 } 1129 1130 if (p->p_sysent->sv_mask) 1131 code &= p->p_sysent->sv_mask; 1132 1133 if (code >= p->p_sysent->sv_size) 1134 callp = &p->p_sysent->sv_table[0]; 1135 else 1136 callp = &p->p_sysent->sv_table[code]; 1137 1138 narg = callp->sy_narg & SYF_ARGMASK; 1139 1140 /* 1141 * copyin is MP aware, but the tracing code is not 1142 */ 1143 if (params && (i = narg * sizeof(int)) && 1144 (error = copyin(params, (caddr_t)args, (u_int)i))) { 1145 mtx_lock(&Giant); 1146#ifdef KTRACE 1147 if (KTRPOINT(p, KTR_SYSCALL)) 1148 ktrsyscall(p->p_tracep, code, narg, args); 1149#endif 1150 goto bad; 1151 } 1152 1153 /* 1154 * Try to run the syscall without the MP lock if the syscall 1155 * is MP safe. 1156 */ 1157 if ((callp->sy_narg & SYF_MPSAFE) == 0) { 1158 mtx_lock(&Giant); 1159 } 1160 1161#ifdef KTRACE 1162 /* 1163 * We have to obtain the MP lock no matter what if 1164 * we are ktracing 1165 */ 1166 if (KTRPOINT(p, KTR_SYSCALL)) { 1167 if (!mtx_owned(&Giant)) 1168 mtx_lock(&Giant); 1169 ktrsyscall(p->p_tracep, code, narg, args); 1170 } 1171#endif 1172 p->p_retval[0] = 0; 1173 p->p_retval[1] = frame.tf_edx; 1174 1175 STOPEVENT(p, S_SCE, narg); /* MP aware */ 1176 1177 error = (*callp->sy_call)(p, args); 1178 1179 /* 1180 * MP SAFE (we may or may not have the MP lock at this point) 1181 */ 1182 switch (error) { 1183 case 0: 1184 frame.tf_eax = p->p_retval[0]; 1185 frame.tf_edx = p->p_retval[1]; 1186 frame.tf_eflags &= ~PSL_C; 1187 break; 1188 1189 case ERESTART: 1190 /* 1191 * Reconstruct pc, assuming lcall $X,y is 7 bytes, 1192 * int 0x80 is 2 bytes. We saved this in tf_err. 1193 */ 1194 frame.tf_eip -= frame.tf_err; 1195 break; 1196 1197 case EJUSTRETURN: 1198 break; 1199 1200 default: 1201bad: 1202 if (p->p_sysent->sv_errsize) { 1203 if (error >= p->p_sysent->sv_errsize) 1204 error = -1; /* XXX */ 1205 else 1206 error = p->p_sysent->sv_errtbl[error]; 1207 } 1208 frame.tf_eax = error; 1209 frame.tf_eflags |= PSL_C; 1210 break; 1211 } 1212 1213 /* 1214 * Traced syscall. trapsignal() is not MP aware. 1215 */ 1216 if ((frame.tf_eflags & PSL_T) && !(frame.tf_eflags & PSL_VM)) { 1217 if (!mtx_owned(&Giant)) 1218 mtx_lock(&Giant); 1219 frame.tf_eflags &= ~PSL_T; 1220 trapsignal(p, SIGTRAP, 0); 1221 } 1222 1223 /* 1224 * Handle reschedule and other end-of-syscall issues 1225 */ 1226 userret(p, &frame, sticks); 1227 1228#ifdef KTRACE 1229 if (KTRPOINT(p, KTR_SYSRET)) { 1230 if (!mtx_owned(&Giant)) 1231 mtx_lock(&Giant); 1232 ktrsysret(p->p_tracep, code, error, p->p_retval[0]); 1233 } 1234#endif 1235 1236 /* 1237 * Release Giant if we had to get it 1238 */ 1239 if (mtx_owned(&Giant)) 1240 mtx_unlock(&Giant); 1241 1242 /* 1243 * This works because errno is findable through the 1244 * register set. If we ever support an emulation where this 1245 * is not the case, this code will need to be revisited. 1246 */ 1247 STOPEVENT(p, S_SCX, code); 1248 1249#ifdef WITNESS 1250 if (witness_list(p)) { 1251 panic("system call %s returning with mutex(s) held\n", 1252 syscallnames[code]); 1253 } 1254#endif 1255 mtx_assert(&sched_lock, MA_NOTOWNED); 1256 mtx_assert(&Giant, MA_NOTOWNED); 1257} 1258 1259void 1260ast(framep) 1261 struct trapframe *framep; 1262{ 1263 struct proc *p = CURPROC; 1264 u_quad_t sticks; 1265#if defined(DEV_NPX) && !defined(SMP) 1266 int ucode; 1267#endif 1268 1269 KASSERT(TRAPF_USERMODE(framep), ("ast in kernel mode")); 1270 1271 /* 1272 * We check for a pending AST here rather than in the assembly as 1273 * acquiring and releasing mutexes in assembly is not fun. 1274 */ 1275 mtx_lock_spin(&sched_lock); 1276 if (!(astpending(p) || resched_wanted(p))) { 1277 mtx_unlock_spin(&sched_lock); 1278 return; 1279 } 1280 1281 sticks = p->p_sticks; 1282 p->p_frame = framep; 1283 1284 astoff(p); 1285 cnt.v_soft++; 1286 mtx_intr_enable(&sched_lock); 1287 if (p->p_sflag & PS_OWEUPC) { 1288 p->p_sflag &= ~PS_OWEUPC; 1289 mtx_unlock_spin(&sched_lock); 1290 mtx_lock(&Giant); 1291 addupc_task(p, p->p_stats->p_prof.pr_addr, 1292 p->p_stats->p_prof.pr_ticks); 1293 mtx_lock_spin(&sched_lock); 1294 } 1295 if (p->p_sflag & PS_ALRMPEND) { 1296 p->p_sflag &= ~PS_ALRMPEND; 1297 mtx_unlock_spin(&sched_lock); 1298 PROC_LOCK(p); 1299 psignal(p, SIGVTALRM); 1300 PROC_UNLOCK(p); 1301 mtx_lock_spin(&sched_lock); 1302 } 1303#if defined(DEV_NPX) && !defined(SMP) 1304 if (PCPU_GET(curpcb)->pcb_flags & PCB_NPXTRAP) { 1305 PCPU_GET(curpcb)->pcb_flags &= ~PCB_NPXTRAP; 1306 mtx_unlock_spin(&sched_lock); 1307 ucode = npxtrap(); 1308 if (ucode != -1) { 1309 if (!mtx_owned(&Giant)) 1310 mtx_lock(&Giant); 1311 trapsignal(p, SIGFPE, ucode); 1312 } 1313 mtx_lock_spin(&sched_lock); 1314 } 1315#endif 1316 if (p->p_sflag & PS_PROFPEND) { 1317 p->p_sflag &= ~PS_PROFPEND; 1318 mtx_unlock_spin(&sched_lock); 1319 PROC_LOCK(p); 1320 psignal(p, SIGPROF); 1321 PROC_UNLOCK(p); 1322 } else 1323 mtx_unlock_spin(&sched_lock); 1324 1325 userret(p, framep, sticks); 1326 1327 if (mtx_owned(&Giant)) 1328 mtx_unlock(&Giant); 1329} 1330