subr_syscall.c revision 57701
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 57701 2000-03-02 16:20:07Z dufault $ 39 */ 40 41/* 42 * 386 Trap and System call handling 43 */ 44 45#include "opt_cpu.h" 46#include "opt_ddb.h" 47#include "opt_ktrace.h" 48#include "opt_clock.h" 49#include "opt_trap.h" 50 51#include <sys/param.h> 52#include <sys/systm.h> 53#include <sys/proc.h> 54#include <sys/pioctl.h> 55#include <sys/kernel.h> 56#include <sys/resourcevar.h> 57#include <sys/signalvar.h> 58#include <sys/syscall.h> 59#include <sys/sysent.h> 60#include <sys/uio.h> 61#include <sys/vmmeter.h> 62#ifdef KTRACE 63#include <sys/ktrace.h> 64#endif 65 66#include <vm/vm.h> 67#include <vm/vm_param.h> 68#include <sys/lock.h> 69#include <vm/pmap.h> 70#include <vm/vm_kern.h> 71#include <vm/vm_map.h> 72#include <vm/vm_page.h> 73#include <vm/vm_extern.h> 74 75#include <machine/cpu.h> 76#include <machine/ipl.h> 77#include <machine/md_var.h> 78#include <machine/pcb.h> 79#ifdef SMP 80#include <machine/smp.h> 81#endif 82#include <machine/tss.h> 83 84#include <i386/isa/intr_machdep.h> 85 86#ifdef POWERFAIL_NMI 87#include <sys/syslog.h> 88#include <machine/clock.h> 89#endif 90 91#include <machine/vm86.h> 92 93#include <ddb/ddb.h> 94 95#include "isa.h" 96#include "npx.h" 97 98int (*pmath_emulate) __P((struct trapframe *)); 99 100extern void trap __P((struct trapframe frame)); 101extern int trapwrite __P((unsigned addr)); 102extern void syscall __P((struct trapframe frame)); 103 104static int trap_pfault __P((struct trapframe *, int, vm_offset_t)); 105static void trap_fatal __P((struct trapframe *, vm_offset_t)); 106void dblfault_handler __P((void)); 107 108extern inthand_t IDTVEC(syscall); 109 110#define MAX_TRAP_MSG 28 111static char *trap_msg[] = { 112 "", /* 0 unused */ 113 "privileged instruction fault", /* 1 T_PRIVINFLT */ 114 "", /* 2 unused */ 115 "breakpoint instruction fault", /* 3 T_BPTFLT */ 116 "", /* 4 unused */ 117 "", /* 5 unused */ 118 "arithmetic trap", /* 6 T_ARITHTRAP */ 119 "system forced exception", /* 7 T_ASTFLT */ 120 "", /* 8 unused */ 121 "general protection fault", /* 9 T_PROTFLT */ 122 "trace trap", /* 10 T_TRCTRAP */ 123 "", /* 11 unused */ 124 "page fault", /* 12 T_PAGEFLT */ 125 "", /* 13 unused */ 126 "alignment fault", /* 14 T_ALIGNFLT */ 127 "", /* 15 unused */ 128 "", /* 16 unused */ 129 "", /* 17 unused */ 130 "integer divide fault", /* 18 T_DIVIDE */ 131 "non-maskable interrupt trap", /* 19 T_NMI */ 132 "overflow trap", /* 20 T_OFLOW */ 133 "FPU bounds check fault", /* 21 T_BOUND */ 134 "FPU device not available", /* 22 T_DNA */ 135 "double fault", /* 23 T_DOUBLEFLT */ 136 "FPU operand fetch fault", /* 24 T_FPOPFLT */ 137 "invalid TSS fault", /* 25 T_TSSFLT */ 138 "segment not present fault", /* 26 T_SEGNPFLT */ 139 "stack fault", /* 27 T_STKFLT */ 140 "machine check trap", /* 28 T_MCHK */ 141}; 142 143static __inline void userret __P((struct proc *p, struct trapframe *frame, 144 u_quad_t oticks)); 145 146#if defined(I586_CPU) && !defined(NO_F00F_HACK) 147extern int has_f00f_bug; 148#endif 149 150static __inline void 151userret(p, frame, oticks) 152 struct proc *p; 153 struct trapframe *frame; 154 u_quad_t oticks; 155{ 156 int sig, s; 157 158 while ((sig = CURSIG(p)) != 0) 159 postsig(sig); 160 161#if 0 162 if (!want_resched && 163 (p->p_priority <= p->p_usrpri) && 164 (p->p_rtprio.type == RTP_PRIO_NORMAL)) { 165 int newpriority; 166 p->p_estcpu += 1; 167 newpriority = PUSER + p->p_estcpu / 4 + 2 * p->p_nice; 168 newpriority = min(newpriority, MAXPRI); 169 p->p_usrpri = newpriority; 170 } 171#endif 172 173 p->p_priority = p->p_usrpri; 174 if (want_resched) { 175 /* 176 * Since we are curproc, clock will normally just change 177 * our priority without moving us from one queue to another 178 * (since the running process is not on a queue.) 179 * If that happened after we setrunqueue ourselves but before we 180 * mi_switch()'ed, we might not be on the queue indicated by 181 * our priority. 182 */ 183 s = splhigh(); 184 setrunqueue(p); 185 p->p_stats->p_ru.ru_nivcsw++; 186 mi_switch(); 187 splx(s); 188 while ((sig = CURSIG(p)) != 0) 189 postsig(sig); 190 } 191 /* 192 * Charge system time if profiling. 193 */ 194 if (p->p_flag & P_PROFIL) 195 addupc_task(p, frame->tf_eip, 196 (u_int)(p->p_sticks - oticks) * psratio); 197 198 curpriority = p->p_priority; 199 currtpriority = p->p_rtprio.prio; 200} 201 202/* 203 * Exception, fault, and trap interface to the FreeBSD kernel. 204 * This common code is called from assembly language IDT gate entry 205 * routines that prepare a suitable stack frame, and restore this 206 * frame after the exception has been processed. 207 */ 208 209void 210trap(frame) 211 struct trapframe frame; 212{ 213 struct proc *p = curproc; 214 u_quad_t sticks = 0; 215 int i = 0, ucode = 0, type, code; 216 vm_offset_t eva; 217 218 if (!(frame.tf_eflags & PSL_I)) { 219 /* 220 * Buggy application or kernel code has disabled interrupts 221 * and then trapped. Enabling interrupts now is wrong, but 222 * it is better than running with interrupts disabled until 223 * they are accidentally enabled later. 224 */ 225 type = frame.tf_trapno; 226 if (ISPL(frame.tf_cs) == SEL_UPL || (frame.tf_eflags & PSL_VM)) 227 printf( 228 "pid %ld (%s): trap %d with interrupts disabled\n", 229 (long)curproc->p_pid, curproc->p_comm, type); 230 else if (type != T_BPTFLT && type != T_TRCTRAP) 231 /* 232 * XXX not quite right, since this may be for a 233 * multiple fault in user mode. 234 */ 235 printf("kernel trap %d with interrupts disabled\n", 236 type); 237 enable_intr(); 238 } 239 240 eva = 0; 241 if (frame.tf_trapno == T_PAGEFLT) { 242 /* 243 * For some Cyrix CPUs, %cr2 is clobbered by interrupts. 244 * This problem is worked around by using an interrupt 245 * gate for the pagefault handler. We are finally ready 246 * to read %cr2 and then must reenable interrupts. 247 * 248 * XXX this should be in the switch statement, but the 249 * NO_FOOF_HACK and VM86 goto and ifdefs obfuscate the 250 * flow of control too much for this to be obviously 251 * correct. 252 */ 253 eva = rcr2(); 254 enable_intr(); 255 } 256 257#if defined(I586_CPU) && !defined(NO_F00F_HACK) 258restart: 259#endif 260 type = frame.tf_trapno; 261 code = frame.tf_err; 262 263 if (in_vm86call) { 264 if (frame.tf_eflags & PSL_VM && 265 (type == T_PROTFLT || type == T_STKFLT)) { 266 i = vm86_emulate((struct vm86frame *)&frame); 267 if (i != 0) 268 /* 269 * returns to original process 270 */ 271 vm86_trap((struct vm86frame *)&frame); 272 return; 273 } 274 switch (type) { 275 /* 276 * these traps want either a process context, or 277 * assume a normal userspace trap. 278 */ 279 case T_PROTFLT: 280 case T_SEGNPFLT: 281 trap_fatal(&frame, eva); 282 return; 283 case T_TRCTRAP: 284 type = T_BPTFLT; /* kernel breakpoint */ 285 /* FALL THROUGH */ 286 } 287 goto kernel_trap; /* normal kernel trap handling */ 288 } 289 290 if ((ISPL(frame.tf_cs) == SEL_UPL) || (frame.tf_eflags & PSL_VM)) { 291 /* user trap */ 292 293 sticks = p->p_sticks; 294 p->p_md.md_regs = &frame; 295 296 switch (type) { 297 case T_PRIVINFLT: /* privileged instruction fault */ 298 ucode = type; 299 i = SIGILL; 300 break; 301 302 case T_BPTFLT: /* bpt instruction fault */ 303 case T_TRCTRAP: /* trace trap */ 304 frame.tf_eflags &= ~PSL_T; 305 i = SIGTRAP; 306 break; 307 308 case T_ARITHTRAP: /* arithmetic trap */ 309 ucode = code; 310 i = SIGFPE; 311 break; 312 313 case T_ASTFLT: /* Allow process switch */ 314 astoff(); 315 cnt.v_soft++; 316 if (p->p_flag & P_OWEUPC) { 317 p->p_flag &= ~P_OWEUPC; 318 addupc_task(p, p->p_stats->p_prof.pr_addr, 319 p->p_stats->p_prof.pr_ticks); 320 } 321 goto out; 322 323 /* 324 * The following two traps can happen in 325 * vm86 mode, and, if so, we want to handle 326 * them specially. 327 */ 328 case T_PROTFLT: /* general protection fault */ 329 case T_STKFLT: /* stack fault */ 330 if (frame.tf_eflags & PSL_VM) { 331 i = vm86_emulate((struct vm86frame *)&frame); 332 if (i == 0) 333 goto out; 334 break; 335 } 336 /* FALL THROUGH */ 337 338 case T_SEGNPFLT: /* segment not present fault */ 339 case T_TSSFLT: /* invalid TSS fault */ 340 case T_DOUBLEFLT: /* double fault */ 341 default: 342 ucode = code + BUS_SEGM_FAULT ; 343 i = SIGBUS; 344 break; 345 346 case T_PAGEFLT: /* page fault */ 347 i = trap_pfault(&frame, TRUE, eva); 348 if (i == -1) 349 return; 350#if defined(I586_CPU) && !defined(NO_F00F_HACK) 351 if (i == -2) 352 goto restart; 353#endif 354 if (i == 0) 355 goto out; 356 357 ucode = T_PAGEFLT; 358 break; 359 360 case T_DIVIDE: /* integer divide fault */ 361 ucode = FPE_INTDIV; 362 i = SIGFPE; 363 break; 364 365#if NISA > 0 366 case T_NMI: 367#ifdef POWERFAIL_NMI 368 goto handle_powerfail; 369#else /* !POWERFAIL_NMI */ 370#ifdef DDB 371 /* NMI can be hooked up to a pushbutton for debugging */ 372 printf ("NMI ... going to debugger\n"); 373 if (kdb_trap (type, 0, &frame)) 374 return; 375#endif /* DDB */ 376 /* machine/parity/power fail/"kitchen sink" faults */ 377 if (isa_nmi(code) == 0) return; 378 panic("NMI indicates hardware failure"); 379#endif /* POWERFAIL_NMI */ 380#endif /* NISA > 0 */ 381 382 case T_OFLOW: /* integer overflow fault */ 383 ucode = FPE_INTOVF; 384 i = SIGFPE; 385 break; 386 387 case T_BOUND: /* bounds check fault */ 388 ucode = FPE_FLTSUB; 389 i = SIGFPE; 390 break; 391 392 case T_DNA: 393#if NNPX > 0 394 /* if a transparent fault (due to context switch "late") */ 395 if (npxdna()) 396 return; 397#endif 398 if (!pmath_emulate) { 399 i = SIGFPE; 400 ucode = FPE_FPU_NP_TRAP; 401 break; 402 } 403 i = (*pmath_emulate)(&frame); 404 if (i == 0) { 405 if (!(frame.tf_eflags & PSL_T)) 406 return; 407 frame.tf_eflags &= ~PSL_T; 408 i = SIGTRAP; 409 } 410 /* else ucode = emulator_only_knows() XXX */ 411 break; 412 413 case T_FPOPFLT: /* FPU operand fetch fault */ 414 ucode = T_FPOPFLT; 415 i = SIGILL; 416 break; 417 } 418 } else { 419kernel_trap: 420 /* kernel trap */ 421 422 switch (type) { 423 case T_PAGEFLT: /* page fault */ 424 (void) trap_pfault(&frame, FALSE, eva); 425 return; 426 427 case T_DNA: 428#if NNPX > 0 429 /* 430 * The kernel is apparently using npx for copying. 431 * XXX this should be fatal unless the kernel has 432 * registered such use. 433 */ 434 if (npxdna()) 435 return; 436#endif 437 break; 438 439 case T_PROTFLT: /* general protection fault */ 440 case T_SEGNPFLT: /* segment not present fault */ 441 /* 442 * Invalid segment selectors and out of bounds 443 * %eip's and %esp's can be set up in user mode. 444 * This causes a fault in kernel mode when the 445 * kernel tries to return to user mode. We want 446 * to get this fault so that we can fix the 447 * problem here and not have to check all the 448 * selectors and pointers when the user changes 449 * them. 450 */ 451#define MAYBE_DORETI_FAULT(where, whereto) \ 452 do { \ 453 if (frame.tf_eip == (int)where) { \ 454 frame.tf_eip = (int)whereto; \ 455 return; \ 456 } \ 457 } while (0) 458 459 if (intr_nesting_level == 0) { 460 /* 461 * Invalid %fs's and %gs's can be created using 462 * procfs or PT_SETREGS or by invalidating the 463 * underlying LDT entry. This causes a fault 464 * in kernel mode when the kernel attempts to 465 * switch contexts. Lose the bad context 466 * (XXX) so that we can continue, and generate 467 * a signal. 468 */ 469 if (frame.tf_eip == (int)cpu_switch_load_gs) { 470 curpcb->pcb_gs = 0; 471 psignal(p, SIGBUS); 472 return; 473 } 474 MAYBE_DORETI_FAULT(doreti_iret, 475 doreti_iret_fault); 476 MAYBE_DORETI_FAULT(doreti_popl_ds, 477 doreti_popl_ds_fault); 478 MAYBE_DORETI_FAULT(doreti_popl_es, 479 doreti_popl_es_fault); 480 MAYBE_DORETI_FAULT(doreti_popl_fs, 481 doreti_popl_fs_fault); 482 if (curpcb && curpcb->pcb_onfault) { 483 frame.tf_eip = (int)curpcb->pcb_onfault; 484 return; 485 } 486 } 487 break; 488 489 case T_TSSFLT: 490 /* 491 * PSL_NT can be set in user mode and isn't cleared 492 * automatically when the kernel is entered. This 493 * causes a TSS fault when the kernel attempts to 494 * `iret' because the TSS link is uninitialized. We 495 * want to get this fault so that we can fix the 496 * problem here and not every time the kernel is 497 * entered. 498 */ 499 if (frame.tf_eflags & PSL_NT) { 500 frame.tf_eflags &= ~PSL_NT; 501 return; 502 } 503 break; 504 505 case T_TRCTRAP: /* trace trap */ 506 if (frame.tf_eip == (int)IDTVEC(syscall)) { 507 /* 508 * We've just entered system mode via the 509 * syscall lcall. Continue single stepping 510 * silently until the syscall handler has 511 * saved the flags. 512 */ 513 return; 514 } 515 if (frame.tf_eip == (int)IDTVEC(syscall) + 1) { 516 /* 517 * The syscall handler has now saved the 518 * flags. Stop single stepping it. 519 */ 520 frame.tf_eflags &= ~PSL_T; 521 return; 522 } 523 /* 524 * Ignore debug register trace traps due to 525 * accesses in the user's address space, which 526 * can happen under several conditions such as 527 * if a user sets a watchpoint on a buffer and 528 * then passes that buffer to a system call. 529 * We still want to get TRCTRAPS for addresses 530 * in kernel space because that is useful when 531 * debugging the kernel. 532 */ 533 if (user_dbreg_trap()) { 534 /* 535 * Reset breakpoint bits because the 536 * processor doesn't 537 */ 538 load_dr6(rdr6() & 0xfffffff0); 539 return; 540 } 541 /* 542 * Fall through (TRCTRAP kernel mode, kernel address) 543 */ 544 case T_BPTFLT: 545 /* 546 * If DDB is enabled, let it handle the debugger trap. 547 * Otherwise, debugger traps "can't happen". 548 */ 549#ifdef DDB 550 if (kdb_trap (type, 0, &frame)) 551 return; 552#endif 553 break; 554 555#if NISA > 0 556 case T_NMI: 557#ifdef POWERFAIL_NMI 558#ifndef TIMER_FREQ 559# define TIMER_FREQ 1193182 560#endif 561 handle_powerfail: 562 { 563 static unsigned lastalert = 0; 564 565 if(time_second - lastalert > 10) 566 { 567 log(LOG_WARNING, "NMI: power fail\n"); 568 sysbeep(TIMER_FREQ/880, hz); 569 lastalert = time_second; 570 } 571 return; 572 } 573#else /* !POWERFAIL_NMI */ 574#ifdef DDB 575 /* NMI can be hooked up to a pushbutton for debugging */ 576 printf ("NMI ... going to debugger\n"); 577 if (kdb_trap (type, 0, &frame)) 578 return; 579#endif /* DDB */ 580 /* machine/parity/power fail/"kitchen sink" faults */ 581 if (isa_nmi(code) == 0) return; 582 /* FALL THROUGH */ 583#endif /* POWERFAIL_NMI */ 584#endif /* NISA > 0 */ 585 } 586 587 trap_fatal(&frame, eva); 588 return; 589 } 590 591 /* Translate fault for emulators (e.g. Linux) */ 592 if (*p->p_sysent->sv_transtrap) 593 i = (*p->p_sysent->sv_transtrap)(i, type); 594 595 trapsignal(p, i, ucode); 596 597#ifdef DEBUG 598 if (type <= MAX_TRAP_MSG) { 599 uprintf("fatal process exception: %s", 600 trap_msg[type]); 601 if ((type == T_PAGEFLT) || (type == T_PROTFLT)) 602 uprintf(", fault VA = 0x%lx", (u_long)eva); 603 uprintf("\n"); 604 } 605#endif 606 607out: 608 userret(p, &frame, sticks); 609} 610 611#ifdef notyet 612/* 613 * This version doesn't allow a page fault to user space while 614 * in the kernel. The rest of the kernel needs to be made "safe" 615 * before this can be used. I think the only things remaining 616 * to be made safe are the iBCS2 code and the process tracing/ 617 * debugging code. 618 */ 619static int 620trap_pfault(frame, usermode, eva) 621 struct trapframe *frame; 622 int usermode; 623 vm_offset_t eva; 624{ 625 vm_offset_t va; 626 struct vmspace *vm = NULL; 627 vm_map_t map = 0; 628 int rv = 0; 629 vm_prot_t ftype; 630 struct proc *p = curproc; 631 632 if (frame->tf_err & PGEX_W) 633 ftype = VM_PROT_READ | VM_PROT_WRITE; 634 else 635 ftype = VM_PROT_READ; 636 637 va = trunc_page(eva); 638 if (va < VM_MIN_KERNEL_ADDRESS) { 639 vm_offset_t v; 640 vm_page_t mpte; 641 642 if (p == NULL || 643 (!usermode && va < VM_MAXUSER_ADDRESS && 644 (intr_nesting_level != 0 || curpcb == NULL || 645 curpcb->pcb_onfault == NULL))) { 646 trap_fatal(frame, eva); 647 return (-1); 648 } 649 650 /* 651 * This is a fault on non-kernel virtual memory. 652 * vm is initialized above to NULL. If curproc is NULL 653 * or curproc->p_vmspace is NULL the fault is fatal. 654 */ 655 vm = p->p_vmspace; 656 if (vm == NULL) 657 goto nogo; 658 659 map = &vm->vm_map; 660 661 /* 662 * Keep swapout from messing with us during this 663 * critical time. 664 */ 665 ++p->p_lock; 666 667 /* 668 * Grow the stack if necessary 669 */ 670 /* grow_stack returns false only if va falls into 671 * a growable stack region and the stack growth 672 * fails. It returns true if va was not within 673 * a growable stack region, or if the stack 674 * growth succeeded. 675 */ 676 if (!grow_stack (p, va)) { 677 rv = KERN_FAILURE; 678 --p->p_lock; 679 goto nogo; 680 } 681 682 /* Fault in the user page: */ 683 rv = vm_fault(map, va, ftype, 684 (ftype & VM_PROT_WRITE) ? VM_FAULT_DIRTY 685 : VM_FAULT_NORMAL); 686 687 --p->p_lock; 688 } else { 689 /* 690 * Don't allow user-mode faults in kernel address space. 691 */ 692 if (usermode) 693 goto nogo; 694 695 /* 696 * Since we know that kernel virtual address addresses 697 * always have pte pages mapped, we just have to fault 698 * the page. 699 */ 700 rv = vm_fault(kernel_map, va, ftype, VM_FAULT_NORMAL); 701 } 702 703 if (rv == KERN_SUCCESS) 704 return (0); 705nogo: 706 if (!usermode) { 707 if (intr_nesting_level == 0 && curpcb && curpcb->pcb_onfault) { 708 frame->tf_eip = (int)curpcb->pcb_onfault; 709 return (0); 710 } 711 trap_fatal(frame, eva); 712 return (-1); 713 } 714 715 /* kludge to pass faulting virtual address to sendsig */ 716 frame->tf_err = eva; 717 718 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV); 719} 720#endif 721 722int 723trap_pfault(frame, usermode, eva) 724 struct trapframe *frame; 725 int usermode; 726 vm_offset_t eva; 727{ 728 vm_offset_t va; 729 struct vmspace *vm = NULL; 730 vm_map_t map = 0; 731 int rv = 0; 732 vm_prot_t ftype; 733 struct proc *p = curproc; 734 735 va = trunc_page(eva); 736 if (va >= KERNBASE) { 737 /* 738 * Don't allow user-mode faults in kernel address space. 739 * An exception: if the faulting address is the invalid 740 * instruction entry in the IDT, then the Intel Pentium 741 * F00F bug workaround was triggered, and we need to 742 * treat it is as an illegal instruction, and not a page 743 * fault. 744 */ 745#if defined(I586_CPU) && !defined(NO_F00F_HACK) 746 if ((eva == (unsigned int)&idt[6]) && has_f00f_bug) { 747 frame->tf_trapno = T_PRIVINFLT; 748 return -2; 749 } 750#endif 751 if (usermode) 752 goto nogo; 753 754 map = kernel_map; 755 } else { 756 /* 757 * This is a fault on non-kernel virtual memory. 758 * vm is initialized above to NULL. If curproc is NULL 759 * or curproc->p_vmspace is NULL the fault is fatal. 760 */ 761 if (p != NULL) 762 vm = p->p_vmspace; 763 764 if (vm == NULL) 765 goto nogo; 766 767 map = &vm->vm_map; 768 } 769 770 if (frame->tf_err & PGEX_W) 771 ftype = VM_PROT_READ | VM_PROT_WRITE; 772 else 773 ftype = VM_PROT_READ; 774 775 if (map != kernel_map) { 776 /* 777 * Keep swapout from messing with us during this 778 * critical time. 779 */ 780 ++p->p_lock; 781 782 /* 783 * Grow the stack if necessary 784 */ 785 /* grow_stack returns false only if va falls into 786 * a growable stack region and the stack growth 787 * fails. It returns true if va was not within 788 * a growable stack region, or if the stack 789 * growth succeeded. 790 */ 791 if (!grow_stack (p, va)) { 792 rv = KERN_FAILURE; 793 --p->p_lock; 794 goto nogo; 795 } 796 797 /* Fault in the user page: */ 798 rv = vm_fault(map, va, ftype, 799 (ftype & VM_PROT_WRITE) ? VM_FAULT_DIRTY 800 : VM_FAULT_NORMAL); 801 802 --p->p_lock; 803 } else { 804 /* 805 * Don't have to worry about process locking or stacks in the kernel. 806 */ 807 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL); 808 } 809 810 if (rv == KERN_SUCCESS) 811 return (0); 812nogo: 813 if (!usermode) { 814 if (intr_nesting_level == 0 && curpcb && curpcb->pcb_onfault) { 815 frame->tf_eip = (int)curpcb->pcb_onfault; 816 return (0); 817 } 818 trap_fatal(frame, eva); 819 return (-1); 820 } 821 822 /* kludge to pass faulting virtual address to sendsig */ 823 frame->tf_err = eva; 824 825 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV); 826} 827 828static void 829trap_fatal(frame, eva) 830 struct trapframe *frame; 831 vm_offset_t eva; 832{ 833 int code, type, ss, esp; 834 struct soft_segment_descriptor softseg; 835 836 code = frame->tf_err; 837 type = frame->tf_trapno; 838 sdtossd(&gdt[IDXSEL(frame->tf_cs & 0xffff)].sd, &softseg); 839 840 if (type <= MAX_TRAP_MSG) 841 printf("\n\nFatal trap %d: %s while in %s mode\n", 842 type, trap_msg[type], 843 frame->tf_eflags & PSL_VM ? "vm86" : 844 ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel"); 845#ifdef SMP 846 /* three seperate prints in case of a trap on an unmapped page */ 847 printf("mp_lock = %08x; ", mp_lock); 848 printf("cpuid = %d; ", cpuid); 849 printf("lapic.id = %08x\n", lapic.id); 850#endif 851 if (type == T_PAGEFLT) { 852 printf("fault virtual address = 0x%x\n", eva); 853 printf("fault code = %s %s, %s\n", 854 code & PGEX_U ? "user" : "supervisor", 855 code & PGEX_W ? "write" : "read", 856 code & PGEX_P ? "protection violation" : "page not present"); 857 } 858 printf("instruction pointer = 0x%x:0x%x\n", 859 frame->tf_cs & 0xffff, frame->tf_eip); 860 if ((ISPL(frame->tf_cs) == SEL_UPL) || (frame->tf_eflags & PSL_VM)) { 861 ss = frame->tf_ss & 0xffff; 862 esp = frame->tf_esp; 863 } else { 864 ss = GSEL(GDATA_SEL, SEL_KPL); 865 esp = (int)&frame->tf_esp; 866 } 867 printf("stack pointer = 0x%x:0x%x\n", ss, esp); 868 printf("frame pointer = 0x%x:0x%x\n", ss, frame->tf_ebp); 869 printf("code segment = base 0x%x, limit 0x%x, type 0x%x\n", 870 softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type); 871 printf(" = DPL %d, pres %d, def32 %d, gran %d\n", 872 softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_def32, 873 softseg.ssd_gran); 874 printf("processor eflags = "); 875 if (frame->tf_eflags & PSL_T) 876 printf("trace trap, "); 877 if (frame->tf_eflags & PSL_I) 878 printf("interrupt enabled, "); 879 if (frame->tf_eflags & PSL_NT) 880 printf("nested task, "); 881 if (frame->tf_eflags & PSL_RF) 882 printf("resume, "); 883 if (frame->tf_eflags & PSL_VM) 884 printf("vm86, "); 885 printf("IOPL = %d\n", (frame->tf_eflags & PSL_IOPL) >> 12); 886 printf("current process = "); 887 if (curproc) { 888 printf("%lu (%s)\n", 889 (u_long)curproc->p_pid, curproc->p_comm ? 890 curproc->p_comm : ""); 891 } else { 892 printf("Idle\n"); 893 } 894 printf("interrupt mask = "); 895 if ((cpl & net_imask) == net_imask) 896 printf("net "); 897 if ((cpl & tty_imask) == tty_imask) 898 printf("tty "); 899 if ((cpl & bio_imask) == bio_imask) 900 printf("bio "); 901 if ((cpl & cam_imask) == cam_imask) 902 printf("cam "); 903 if (cpl == 0) 904 printf("none"); 905#ifdef SMP 906/** 907 * XXX FIXME: 908 * we probably SHOULD have stopped the other CPUs before now! 909 * another CPU COULD have been touching cpl at this moment... 910 */ 911 printf(" <- SMP: XXX"); 912#endif 913 printf("\n"); 914 915#ifdef KDB 916 if (kdb_trap(&psl)) 917 return; 918#endif 919#ifdef DDB 920 if ((debugger_on_panic || db_active) && kdb_trap(type, 0, frame)) 921 return; 922#endif 923 printf("trap number = %d\n", type); 924 if (type <= MAX_TRAP_MSG) 925 panic(trap_msg[type]); 926 else 927 panic("unknown/reserved trap"); 928} 929 930/* 931 * Double fault handler. Called when a fault occurs while writing 932 * a frame for a trap/exception onto the stack. This usually occurs 933 * when the stack overflows (such is the case with infinite recursion, 934 * for example). 935 * 936 * XXX Note that the current PTD gets replaced by IdlePTD when the 937 * task switch occurs. This means that the stack that was active at 938 * the time of the double fault is not available at <kstack> unless 939 * the machine was idle when the double fault occurred. The downside 940 * of this is that "trace <ebp>" in ddb won't work. 941 */ 942void 943dblfault_handler() 944{ 945 printf("\nFatal double fault:\n"); 946 printf("eip = 0x%x\n", common_tss.tss_eip); 947 printf("esp = 0x%x\n", common_tss.tss_esp); 948 printf("ebp = 0x%x\n", common_tss.tss_ebp); 949#ifdef SMP 950 /* three seperate prints in case of a trap on an unmapped page */ 951 printf("mp_lock = %08x; ", mp_lock); 952 printf("cpuid = %d; ", cpuid); 953 printf("lapic.id = %08x\n", lapic.id); 954#endif 955 panic("double fault"); 956} 957 958/* 959 * Compensate for 386 brain damage (missing URKR). 960 * This is a little simpler than the pagefault handler in trap() because 961 * it the page tables have already been faulted in and high addresses 962 * are thrown out early for other reasons. 963 */ 964int trapwrite(addr) 965 unsigned addr; 966{ 967 struct proc *p; 968 vm_offset_t va; 969 struct vmspace *vm; 970 int rv; 971 972 va = trunc_page((vm_offset_t)addr); 973 /* 974 * XXX - MAX is END. Changed > to >= for temp. fix. 975 */ 976 if (va >= VM_MAXUSER_ADDRESS) 977 return (1); 978 979 p = curproc; 980 vm = p->p_vmspace; 981 982 ++p->p_lock; 983 984 if (!grow_stack (p, va)) { 985 --p->p_lock; 986 return (1); 987 } 988 989 /* 990 * fault the data page 991 */ 992 rv = vm_fault(&vm->vm_map, va, VM_PROT_READ|VM_PROT_WRITE, VM_FAULT_DIRTY); 993 994 --p->p_lock; 995 996 if (rv != KERN_SUCCESS) 997 return 1; 998 999 return (0); 1000} 1001 1002/* 1003 * System call request from POSIX system call gate interface to kernel. 1004 * Like trap(), argument is call by reference. 1005 */ 1006void 1007syscall(frame) 1008 struct trapframe frame; 1009{ 1010 caddr_t params; 1011 int i; 1012 struct sysent *callp; 1013 struct proc *p = curproc; 1014 u_quad_t sticks; 1015 int error; 1016 int args[8]; 1017 u_int code; 1018 1019#ifdef DIAGNOSTIC 1020 if (ISPL(frame.tf_cs) != SEL_UPL) 1021 panic("syscall"); 1022#endif 1023 sticks = p->p_sticks; 1024 p->p_md.md_regs = &frame; 1025 params = (caddr_t)frame.tf_esp + sizeof(int); 1026 code = frame.tf_eax; 1027 if (p->p_sysent->sv_prepsyscall) { 1028 (*p->p_sysent->sv_prepsyscall)(&frame, args, &code, ¶ms); 1029 } else { 1030 /* 1031 * Need to check if this is a 32 bit or 64 bit syscall. 1032 */ 1033 if (code == SYS_syscall) { 1034 /* 1035 * Code is first argument, followed by actual args. 1036 */ 1037 code = fuword(params); 1038 params += sizeof(int); 1039 } else if (code == SYS___syscall) { 1040 /* 1041 * Like syscall, but code is a quad, so as to maintain 1042 * quad alignment for the rest of the arguments. 1043 */ 1044 code = fuword(params); 1045 params += sizeof(quad_t); 1046 } 1047 } 1048 1049 if (p->p_sysent->sv_mask) 1050 code &= p->p_sysent->sv_mask; 1051 1052 if (code >= p->p_sysent->sv_size) 1053 callp = &p->p_sysent->sv_table[0]; 1054 else 1055 callp = &p->p_sysent->sv_table[code]; 1056 1057 if (params && (i = callp->sy_narg * sizeof(int)) && 1058 (error = copyin(params, (caddr_t)args, (u_int)i))) { 1059#ifdef KTRACE 1060 if (KTRPOINT(p, KTR_SYSCALL)) 1061 ktrsyscall(p->p_tracep, code, callp->sy_narg, args); 1062#endif 1063 goto bad; 1064 } 1065#ifdef KTRACE 1066 if (KTRPOINT(p, KTR_SYSCALL)) 1067 ktrsyscall(p->p_tracep, code, callp->sy_narg, args); 1068#endif 1069 p->p_retval[0] = 0; 1070 p->p_retval[1] = frame.tf_edx; 1071 1072 STOPEVENT(p, S_SCE, callp->sy_narg); 1073 1074 error = (*callp->sy_call)(p, args); 1075 1076 switch (error) { 1077 1078 case 0: 1079 /* 1080 * Reinitialize proc pointer `p' as it may be different 1081 * if this is a child returning from fork syscall. 1082 */ 1083 p = curproc; 1084 frame.tf_eax = p->p_retval[0]; 1085 frame.tf_edx = p->p_retval[1]; 1086 frame.tf_eflags &= ~PSL_C; 1087 break; 1088 1089 case ERESTART: 1090 /* 1091 * Reconstruct pc, assuming lcall $X,y is 7 bytes, 1092 * int 0x80 is 2 bytes. We saved this in tf_err. 1093 */ 1094 frame.tf_eip -= frame.tf_err; 1095 break; 1096 1097 case EJUSTRETURN: 1098 break; 1099 1100 default: 1101bad: 1102 if (p->p_sysent->sv_errsize) { 1103 if (error >= p->p_sysent->sv_errsize) 1104 error = -1; /* XXX */ 1105 else 1106 error = p->p_sysent->sv_errtbl[error]; 1107 } 1108 frame.tf_eax = error; 1109 frame.tf_eflags |= PSL_C; 1110 break; 1111 } 1112 1113 if ((frame.tf_eflags & PSL_T) && !(frame.tf_eflags & PSL_VM)) { 1114 /* Traced syscall. */ 1115 frame.tf_eflags &= ~PSL_T; 1116 trapsignal(p, SIGTRAP, 0); 1117 } 1118 1119 userret(p, &frame, sticks); 1120 1121#ifdef KTRACE 1122 if (KTRPOINT(p, KTR_SYSRET)) 1123 ktrsysret(p->p_tracep, code, error, p->p_retval[0]); 1124#endif 1125 1126 /* 1127 * This works because errno is findable through the 1128 * register set. If we ever support an emulation where this 1129 * is not the case, this code will need to be revisited. 1130 */ 1131 STOPEVENT(p, S_SCX, code); 1132 1133} 1134 1135/* 1136 * Simplified back end of syscall(), used when returning from fork() 1137 * directly into user mode. 1138 */ 1139void 1140fork_return(p, frame) 1141 struct proc *p; 1142 struct trapframe frame; 1143{ 1144 frame.tf_eax = 0; /* Child returns zero */ 1145 frame.tf_eflags &= ~PSL_C; /* success */ 1146 frame.tf_edx = 1; 1147 1148 userret(p, &frame, 0); 1149#ifdef KTRACE 1150 if (KTRPOINT(p, KTR_SYSRET)) 1151 ktrsysret(p->p_tracep, SYS_fork, 0, 0); 1152#endif 1153} 1154