subr_syscall.c revision 55823
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 55823 2000-01-11 14:54:01Z yokota $ 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} 200 201/* 202 * Exception, fault, and trap interface to the FreeBSD kernel. 203 * This common code is called from assembly language IDT gate entry 204 * routines that prepare a suitable stack frame, and restore this 205 * frame after the exception has been processed. 206 */ 207 208void 209trap(frame) 210 struct trapframe frame; 211{ 212 struct proc *p = curproc; 213 u_quad_t sticks = 0; 214 int i = 0, ucode = 0, type, code; 215 vm_offset_t eva; 216 217 if (!(frame.tf_eflags & PSL_I)) { 218 /* 219 * Buggy application or kernel code has disabled interrupts 220 * and then trapped. Enabling interrupts now is wrong, but 221 * it is better than running with interrupts disabled until 222 * they are accidentally enabled later. 223 */ 224 type = frame.tf_trapno; 225 if (ISPL(frame.tf_cs) == SEL_UPL || (frame.tf_eflags & PSL_VM)) 226 printf( 227 "pid %ld (%s): trap %d with interrupts disabled\n", 228 (long)curproc->p_pid, curproc->p_comm, type); 229 else if (type != T_BPTFLT && type != T_TRCTRAP) 230 /* 231 * XXX not quite right, since this may be for a 232 * multiple fault in user mode. 233 */ 234 printf("kernel trap %d with interrupts disabled\n", 235 type); 236 enable_intr(); 237 } 238 239 eva = 0; 240 if (frame.tf_trapno == T_PAGEFLT) { 241 /* 242 * For some Cyrix CPUs, %cr2 is clobbered by interrupts. 243 * This problem is worked around by using an interrupt 244 * gate for the pagefault handler. We are finally ready 245 * to read %cr2 and then must reenable interrupts. 246 * 247 * XXX this should be in the switch statement, but the 248 * NO_FOOF_HACK and VM86 goto and ifdefs obfuscate the 249 * flow of control too much for this to be obviously 250 * correct. 251 */ 252 eva = rcr2(); 253 enable_intr(); 254 } 255 256#if defined(I586_CPU) && !defined(NO_F00F_HACK) 257restart: 258#endif 259 type = frame.tf_trapno; 260 code = frame.tf_err; 261 262 if (in_vm86call) { 263 if (frame.tf_eflags & PSL_VM && 264 (type == T_PROTFLT || type == T_STKFLT)) { 265 i = vm86_emulate((struct vm86frame *)&frame); 266 if (i != 0) 267 /* 268 * returns to original process 269 */ 270 vm86_trap((struct vm86frame *)&frame); 271 return; 272 } 273 switch (type) { 274 /* 275 * these traps want either a process context, or 276 * assume a normal userspace trap. 277 */ 278 case T_PROTFLT: 279 case T_SEGNPFLT: 280 trap_fatal(&frame, eva); 281 return; 282 case T_TRCTRAP: 283 type = T_BPTFLT; /* kernel breakpoint */ 284 /* FALL THROUGH */ 285 } 286 goto kernel_trap; /* normal kernel trap handling */ 287 } 288 289 if ((ISPL(frame.tf_cs) == SEL_UPL) || (frame.tf_eflags & PSL_VM)) { 290 /* user trap */ 291 292 sticks = p->p_sticks; 293 p->p_md.md_regs = &frame; 294 295 switch (type) { 296 case T_PRIVINFLT: /* privileged instruction fault */ 297 ucode = type; 298 i = SIGILL; 299 break; 300 301 case T_BPTFLT: /* bpt instruction fault */ 302 case T_TRCTRAP: /* trace trap */ 303 frame.tf_eflags &= ~PSL_T; 304 i = SIGTRAP; 305 break; 306 307 case T_ARITHTRAP: /* arithmetic trap */ 308 ucode = code; 309 i = SIGFPE; 310 break; 311 312 case T_ASTFLT: /* Allow process switch */ 313 astoff(); 314 cnt.v_soft++; 315 if (p->p_flag & P_OWEUPC) { 316 p->p_flag &= ~P_OWEUPC; 317 addupc_task(p, p->p_stats->p_prof.pr_addr, 318 p->p_stats->p_prof.pr_ticks); 319 } 320 goto out; 321 322 /* 323 * The following two traps can happen in 324 * vm86 mode, and, if so, we want to handle 325 * them specially. 326 */ 327 case T_PROTFLT: /* general protection fault */ 328 case T_STKFLT: /* stack fault */ 329 if (frame.tf_eflags & PSL_VM) { 330 i = vm86_emulate((struct vm86frame *)&frame); 331 if (i == 0) 332 goto out; 333 break; 334 } 335 /* FALL THROUGH */ 336 337 case T_SEGNPFLT: /* segment not present fault */ 338 case T_TSSFLT: /* invalid TSS fault */ 339 case T_DOUBLEFLT: /* double fault */ 340 default: 341 ucode = code + BUS_SEGM_FAULT ; 342 i = SIGBUS; 343 break; 344 345 case T_PAGEFLT: /* page fault */ 346 i = trap_pfault(&frame, TRUE, eva); 347 if (i == -1) 348 return; 349#if defined(I586_CPU) && !defined(NO_F00F_HACK) 350 if (i == -2) 351 goto restart; 352#endif 353 if (i == 0) 354 goto out; 355 356 ucode = T_PAGEFLT; 357 break; 358 359 case T_DIVIDE: /* integer divide fault */ 360 ucode = FPE_INTDIV; 361 i = SIGFPE; 362 break; 363 364#if NISA > 0 365 case T_NMI: 366#ifdef POWERFAIL_NMI 367 goto handle_powerfail; 368#else /* !POWERFAIL_NMI */ 369#ifdef DDB 370 /* NMI can be hooked up to a pushbutton for debugging */ 371 printf ("NMI ... going to debugger\n"); 372 if (kdb_trap (type, 0, &frame)) 373 return; 374#endif /* DDB */ 375 /* machine/parity/power fail/"kitchen sink" faults */ 376 if (isa_nmi(code) == 0) return; 377 panic("NMI indicates hardware failure"); 378#endif /* POWERFAIL_NMI */ 379#endif /* NISA > 0 */ 380 381 case T_OFLOW: /* integer overflow fault */ 382 ucode = FPE_INTOVF; 383 i = SIGFPE; 384 break; 385 386 case T_BOUND: /* bounds check fault */ 387 ucode = FPE_FLTSUB; 388 i = SIGFPE; 389 break; 390 391 case T_DNA: 392#if NNPX > 0 393 /* if a transparent fault (due to context switch "late") */ 394 if (npxdna()) 395 return; 396#endif 397 if (!pmath_emulate) { 398 i = SIGFPE; 399 ucode = FPE_FPU_NP_TRAP; 400 break; 401 } 402 i = (*pmath_emulate)(&frame); 403 if (i == 0) { 404 if (!(frame.tf_eflags & PSL_T)) 405 return; 406 frame.tf_eflags &= ~PSL_T; 407 i = SIGTRAP; 408 } 409 /* else ucode = emulator_only_knows() XXX */ 410 break; 411 412 case T_FPOPFLT: /* FPU operand fetch fault */ 413 ucode = T_FPOPFLT; 414 i = SIGILL; 415 break; 416 } 417 } else { 418kernel_trap: 419 /* kernel trap */ 420 421 switch (type) { 422 case T_PAGEFLT: /* page fault */ 423 (void) trap_pfault(&frame, FALSE, eva); 424 return; 425 426 case T_DNA: 427#if NNPX > 0 428 /* 429 * The kernel is apparently using npx for copying. 430 * XXX this should be fatal unless the kernel has 431 * registered such use. 432 */ 433 if (npxdna()) 434 return; 435#endif 436 break; 437 438 case T_PROTFLT: /* general protection fault */ 439 case T_SEGNPFLT: /* segment not present fault */ 440 /* 441 * Invalid segment selectors and out of bounds 442 * %eip's and %esp's can be set up in user mode. 443 * This causes a fault in kernel mode when the 444 * kernel tries to return to user mode. We want 445 * to get this fault so that we can fix the 446 * problem here and not have to check all the 447 * selectors and pointers when the user changes 448 * them. 449 */ 450#define MAYBE_DORETI_FAULT(where, whereto) \ 451 do { \ 452 if (frame.tf_eip == (int)where) { \ 453 frame.tf_eip = (int)whereto; \ 454 return; \ 455 } \ 456 } while (0) 457 458 if (intr_nesting_level == 0) { 459 /* 460 * Invalid %fs's and %gs's can be created using 461 * procfs or PT_SETREGS or by invalidating the 462 * underlying LDT entry. This causes a fault 463 * in kernel mode when the kernel attempts to 464 * switch contexts. Lose the bad context 465 * (XXX) so that we can continue, and generate 466 * a signal. 467 */ 468 if (frame.tf_eip == (int)cpu_switch_load_gs) { 469 curpcb->pcb_gs = 0; 470 psignal(p, SIGBUS); 471 return; 472 } 473 MAYBE_DORETI_FAULT(doreti_iret, 474 doreti_iret_fault); 475 MAYBE_DORETI_FAULT(doreti_popl_ds, 476 doreti_popl_ds_fault); 477 MAYBE_DORETI_FAULT(doreti_popl_es, 478 doreti_popl_es_fault); 479 MAYBE_DORETI_FAULT(doreti_popl_fs, 480 doreti_popl_fs_fault); 481 if (curpcb && curpcb->pcb_onfault) { 482 frame.tf_eip = (int)curpcb->pcb_onfault; 483 return; 484 } 485 } 486 break; 487 488 case T_TSSFLT: 489 /* 490 * PSL_NT can be set in user mode and isn't cleared 491 * automatically when the kernel is entered. This 492 * causes a TSS fault when the kernel attempts to 493 * `iret' because the TSS link is uninitialized. We 494 * want to get this fault so that we can fix the 495 * problem here and not every time the kernel is 496 * entered. 497 */ 498 if (frame.tf_eflags & PSL_NT) { 499 frame.tf_eflags &= ~PSL_NT; 500 return; 501 } 502 break; 503 504 case T_TRCTRAP: /* trace trap */ 505 if (frame.tf_eip == (int)IDTVEC(syscall)) { 506 /* 507 * We've just entered system mode via the 508 * syscall lcall. Continue single stepping 509 * silently until the syscall handler has 510 * saved the flags. 511 */ 512 return; 513 } 514 if (frame.tf_eip == (int)IDTVEC(syscall) + 1) { 515 /* 516 * The syscall handler has now saved the 517 * flags. Stop single stepping it. 518 */ 519 frame.tf_eflags &= ~PSL_T; 520 return; 521 } 522 /* 523 * Fall through. 524 */ 525 case T_BPTFLT: 526 /* 527 * If DDB is enabled, let it handle the debugger trap. 528 * Otherwise, debugger traps "can't happen". 529 */ 530#ifdef DDB 531 if (kdb_trap (type, 0, &frame)) 532 return; 533#endif 534 break; 535 536#if NISA > 0 537 case T_NMI: 538#ifdef POWERFAIL_NMI 539#ifndef TIMER_FREQ 540# define TIMER_FREQ 1193182 541#endif 542 handle_powerfail: 543 { 544 static unsigned lastalert = 0; 545 546 if(time_second - lastalert > 10) 547 { 548 log(LOG_WARNING, "NMI: power fail\n"); 549 sysbeep(TIMER_FREQ/880, hz); 550 lastalert = time_second; 551 } 552 return; 553 } 554#else /* !POWERFAIL_NMI */ 555#ifdef DDB 556 /* NMI can be hooked up to a pushbutton for debugging */ 557 printf ("NMI ... going to debugger\n"); 558 if (kdb_trap (type, 0, &frame)) 559 return; 560#endif /* DDB */ 561 /* machine/parity/power fail/"kitchen sink" faults */ 562 if (isa_nmi(code) == 0) return; 563 /* FALL THROUGH */ 564#endif /* POWERFAIL_NMI */ 565#endif /* NISA > 0 */ 566 } 567 568 trap_fatal(&frame, eva); 569 return; 570 } 571 572 /* Translate fault for emulators (e.g. Linux) */ 573 if (*p->p_sysent->sv_transtrap) 574 i = (*p->p_sysent->sv_transtrap)(i, type); 575 576 trapsignal(p, i, ucode); 577 578#ifdef DEBUG 579 if (type <= MAX_TRAP_MSG) { 580 uprintf("fatal process exception: %s", 581 trap_msg[type]); 582 if ((type == T_PAGEFLT) || (type == T_PROTFLT)) 583 uprintf(", fault VA = 0x%lx", (u_long)eva); 584 uprintf("\n"); 585 } 586#endif 587 588out: 589 userret(p, &frame, sticks); 590} 591 592#ifdef notyet 593/* 594 * This version doesn't allow a page fault to user space while 595 * in the kernel. The rest of the kernel needs to be made "safe" 596 * before this can be used. I think the only things remaining 597 * to be made safe are the iBCS2 code and the process tracing/ 598 * debugging code. 599 */ 600static int 601trap_pfault(frame, usermode, eva) 602 struct trapframe *frame; 603 int usermode; 604 vm_offset_t eva; 605{ 606 vm_offset_t va; 607 struct vmspace *vm = NULL; 608 vm_map_t map = 0; 609 int rv = 0; 610 vm_prot_t ftype; 611 struct proc *p = curproc; 612 613 if (frame->tf_err & PGEX_W) 614 ftype = VM_PROT_READ | VM_PROT_WRITE; 615 else 616 ftype = VM_PROT_READ; 617 618 va = trunc_page(eva); 619 if (va < VM_MIN_KERNEL_ADDRESS) { 620 vm_offset_t v; 621 vm_page_t mpte; 622 623 if (p == NULL || 624 (!usermode && va < VM_MAXUSER_ADDRESS && 625 (intr_nesting_level != 0 || curpcb == NULL || 626 curpcb->pcb_onfault == NULL))) { 627 trap_fatal(frame, eva); 628 return (-1); 629 } 630 631 /* 632 * This is a fault on non-kernel virtual memory. 633 * vm is initialized above to NULL. If curproc is NULL 634 * or curproc->p_vmspace is NULL the fault is fatal. 635 */ 636 vm = p->p_vmspace; 637 if (vm == NULL) 638 goto nogo; 639 640 map = &vm->vm_map; 641 642 /* 643 * Keep swapout from messing with us during this 644 * critical time. 645 */ 646 ++p->p_lock; 647 648 /* 649 * Grow the stack if necessary 650 */ 651 /* grow_stack returns false only if va falls into 652 * a growable stack region and the stack growth 653 * fails. It returns true if va was not within 654 * a growable stack region, or if the stack 655 * growth succeeded. 656 */ 657 if (!grow_stack (p, va)) { 658 rv = KERN_FAILURE; 659 --p->p_lock; 660 goto nogo; 661 } 662 663 /* Fault in the user page: */ 664 rv = vm_fault(map, va, ftype, 665 (ftype & VM_PROT_WRITE) ? VM_FAULT_DIRTY 666 : VM_FAULT_NORMAL); 667 668 --p->p_lock; 669 } else { 670 /* 671 * Don't allow user-mode faults in kernel address space. 672 */ 673 if (usermode) 674 goto nogo; 675 676 /* 677 * Since we know that kernel virtual address addresses 678 * always have pte pages mapped, we just have to fault 679 * the page. 680 */ 681 rv = vm_fault(kernel_map, va, ftype, VM_FAULT_NORMAL); 682 } 683 684 if (rv == KERN_SUCCESS) 685 return (0); 686nogo: 687 if (!usermode) { 688 if (intr_nesting_level == 0 && curpcb && curpcb->pcb_onfault) { 689 frame->tf_eip = (int)curpcb->pcb_onfault; 690 return (0); 691 } 692 trap_fatal(frame, eva); 693 return (-1); 694 } 695 696 /* kludge to pass faulting virtual address to sendsig */ 697 frame->tf_err = eva; 698 699 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV); 700} 701#endif 702 703int 704trap_pfault(frame, usermode, eva) 705 struct trapframe *frame; 706 int usermode; 707 vm_offset_t eva; 708{ 709 vm_offset_t va; 710 struct vmspace *vm = NULL; 711 vm_map_t map = 0; 712 int rv = 0; 713 vm_prot_t ftype; 714 struct proc *p = curproc; 715 716 va = trunc_page(eva); 717 if (va >= KERNBASE) { 718 /* 719 * Don't allow user-mode faults in kernel address space. 720 * An exception: if the faulting address is the invalid 721 * instruction entry in the IDT, then the Intel Pentium 722 * F00F bug workaround was triggered, and we need to 723 * treat it is as an illegal instruction, and not a page 724 * fault. 725 */ 726#if defined(I586_CPU) && !defined(NO_F00F_HACK) 727 if ((eva == (unsigned int)&idt[6]) && has_f00f_bug) { 728 frame->tf_trapno = T_PRIVINFLT; 729 return -2; 730 } 731#endif 732 if (usermode) 733 goto nogo; 734 735 map = kernel_map; 736 } else { 737 /* 738 * This is a fault on non-kernel virtual memory. 739 * vm is initialized above to NULL. If curproc is NULL 740 * or curproc->p_vmspace is NULL the fault is fatal. 741 */ 742 if (p != NULL) 743 vm = p->p_vmspace; 744 745 if (vm == NULL) 746 goto nogo; 747 748 map = &vm->vm_map; 749 } 750 751 if (frame->tf_err & PGEX_W) 752 ftype = VM_PROT_READ | VM_PROT_WRITE; 753 else 754 ftype = VM_PROT_READ; 755 756 if (map != kernel_map) { 757 /* 758 * Keep swapout from messing with us during this 759 * critical time. 760 */ 761 ++p->p_lock; 762 763 /* 764 * Grow the stack if necessary 765 */ 766 /* grow_stack returns false only if va falls into 767 * a growable stack region and the stack growth 768 * fails. It returns true if va was not within 769 * a growable stack region, or if the stack 770 * growth succeeded. 771 */ 772 if (!grow_stack (p, va)) { 773 rv = KERN_FAILURE; 774 --p->p_lock; 775 goto nogo; 776 } 777 778 /* Fault in the user page: */ 779 rv = vm_fault(map, va, ftype, 780 (ftype & VM_PROT_WRITE) ? VM_FAULT_DIRTY 781 : VM_FAULT_NORMAL); 782 783 --p->p_lock; 784 } else { 785 /* 786 * Don't have to worry about process locking or stacks in the kernel. 787 */ 788 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL); 789 } 790 791 if (rv == KERN_SUCCESS) 792 return (0); 793nogo: 794 if (!usermode) { 795 if (intr_nesting_level == 0 && curpcb && curpcb->pcb_onfault) { 796 frame->tf_eip = (int)curpcb->pcb_onfault; 797 return (0); 798 } 799 trap_fatal(frame, eva); 800 return (-1); 801 } 802 803 /* kludge to pass faulting virtual address to sendsig */ 804 frame->tf_err = eva; 805 806 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV); 807} 808 809static void 810trap_fatal(frame, eva) 811 struct trapframe *frame; 812 vm_offset_t eva; 813{ 814 int code, type, ss, esp; 815 struct soft_segment_descriptor softseg; 816 817 code = frame->tf_err; 818 type = frame->tf_trapno; 819 sdtossd(&gdt[IDXSEL(frame->tf_cs & 0xffff)].sd, &softseg); 820 821 if (type <= MAX_TRAP_MSG) 822 printf("\n\nFatal trap %d: %s while in %s mode\n", 823 type, trap_msg[type], 824 frame->tf_eflags & PSL_VM ? "vm86" : 825 ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel"); 826#ifdef SMP 827 /* three seperate prints in case of a trap on an unmapped page */ 828 printf("mp_lock = %08x; ", mp_lock); 829 printf("cpuid = %d; ", cpuid); 830 printf("lapic.id = %08x\n", lapic.id); 831#endif 832 if (type == T_PAGEFLT) { 833 printf("fault virtual address = 0x%x\n", eva); 834 printf("fault code = %s %s, %s\n", 835 code & PGEX_U ? "user" : "supervisor", 836 code & PGEX_W ? "write" : "read", 837 code & PGEX_P ? "protection violation" : "page not present"); 838 } 839 printf("instruction pointer = 0x%x:0x%x\n", 840 frame->tf_cs & 0xffff, frame->tf_eip); 841 if ((ISPL(frame->tf_cs) == SEL_UPL) || (frame->tf_eflags & PSL_VM)) { 842 ss = frame->tf_ss & 0xffff; 843 esp = frame->tf_esp; 844 } else { 845 ss = GSEL(GDATA_SEL, SEL_KPL); 846 esp = (int)&frame->tf_esp; 847 } 848 printf("stack pointer = 0x%x:0x%x\n", ss, esp); 849 printf("frame pointer = 0x%x:0x%x\n", ss, frame->tf_ebp); 850 printf("code segment = base 0x%x, limit 0x%x, type 0x%x\n", 851 softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type); 852 printf(" = DPL %d, pres %d, def32 %d, gran %d\n", 853 softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_def32, 854 softseg.ssd_gran); 855 printf("processor eflags = "); 856 if (frame->tf_eflags & PSL_T) 857 printf("trace trap, "); 858 if (frame->tf_eflags & PSL_I) 859 printf("interrupt enabled, "); 860 if (frame->tf_eflags & PSL_NT) 861 printf("nested task, "); 862 if (frame->tf_eflags & PSL_RF) 863 printf("resume, "); 864 if (frame->tf_eflags & PSL_VM) 865 printf("vm86, "); 866 printf("IOPL = %d\n", (frame->tf_eflags & PSL_IOPL) >> 12); 867 printf("current process = "); 868 if (curproc) { 869 printf("%lu (%s)\n", 870 (u_long)curproc->p_pid, curproc->p_comm ? 871 curproc->p_comm : ""); 872 } else { 873 printf("Idle\n"); 874 } 875 printf("interrupt mask = "); 876 if ((cpl & net_imask) == net_imask) 877 printf("net "); 878 if ((cpl & tty_imask) == tty_imask) 879 printf("tty "); 880 if ((cpl & bio_imask) == bio_imask) 881 printf("bio "); 882 if ((cpl & cam_imask) == cam_imask) 883 printf("cam "); 884 if (cpl == 0) 885 printf("none"); 886#ifdef SMP 887/** 888 * XXX FIXME: 889 * we probably SHOULD have stopped the other CPUs before now! 890 * another CPU COULD have been touching cpl at this moment... 891 */ 892 printf(" <- SMP: XXX"); 893#endif 894 printf("\n"); 895 896#ifdef KDB 897 if (kdb_trap(&psl)) 898 return; 899#endif 900#ifdef DDB 901 if ((debugger_on_panic || db_active) && kdb_trap(type, 0, frame)) 902 return; 903#endif 904 printf("trap number = %d\n", type); 905 if (type <= MAX_TRAP_MSG) 906 panic(trap_msg[type]); 907 else 908 panic("unknown/reserved trap"); 909} 910 911/* 912 * Double fault handler. Called when a fault occurs while writing 913 * a frame for a trap/exception onto the stack. This usually occurs 914 * when the stack overflows (such is the case with infinite recursion, 915 * for example). 916 * 917 * XXX Note that the current PTD gets replaced by IdlePTD when the 918 * task switch occurs. This means that the stack that was active at 919 * the time of the double fault is not available at <kstack> unless 920 * the machine was idle when the double fault occurred. The downside 921 * of this is that "trace <ebp>" in ddb won't work. 922 */ 923void 924dblfault_handler() 925{ 926 printf("\nFatal double fault:\n"); 927 printf("eip = 0x%x\n", common_tss.tss_eip); 928 printf("esp = 0x%x\n", common_tss.tss_esp); 929 printf("ebp = 0x%x\n", common_tss.tss_ebp); 930#ifdef SMP 931 /* three seperate prints in case of a trap on an unmapped page */ 932 printf("mp_lock = %08x; ", mp_lock); 933 printf("cpuid = %d; ", cpuid); 934 printf("lapic.id = %08x\n", lapic.id); 935#endif 936 panic("double fault"); 937} 938 939/* 940 * Compensate for 386 brain damage (missing URKR). 941 * This is a little simpler than the pagefault handler in trap() because 942 * it the page tables have already been faulted in and high addresses 943 * are thrown out early for other reasons. 944 */ 945int trapwrite(addr) 946 unsigned addr; 947{ 948 struct proc *p; 949 vm_offset_t va; 950 struct vmspace *vm; 951 int rv; 952 953 va = trunc_page((vm_offset_t)addr); 954 /* 955 * XXX - MAX is END. Changed > to >= for temp. fix. 956 */ 957 if (va >= VM_MAXUSER_ADDRESS) 958 return (1); 959 960 p = curproc; 961 vm = p->p_vmspace; 962 963 ++p->p_lock; 964 965 if (!grow_stack (p, va)) { 966 --p->p_lock; 967 return (1); 968 } 969 970 /* 971 * fault the data page 972 */ 973 rv = vm_fault(&vm->vm_map, va, VM_PROT_READ|VM_PROT_WRITE, VM_FAULT_DIRTY); 974 975 --p->p_lock; 976 977 if (rv != KERN_SUCCESS) 978 return 1; 979 980 return (0); 981} 982 983/* 984 * System call request from POSIX system call gate interface to kernel. 985 * Like trap(), argument is call by reference. 986 */ 987void 988syscall(frame) 989 struct trapframe frame; 990{ 991 caddr_t params; 992 int i; 993 struct sysent *callp; 994 struct proc *p = curproc; 995 u_quad_t sticks; 996 int error; 997 int args[8]; 998 u_int code; 999 1000#ifdef DIAGNOSTIC 1001 if (ISPL(frame.tf_cs) != SEL_UPL) 1002 panic("syscall"); 1003#endif 1004 sticks = p->p_sticks; 1005 p->p_md.md_regs = &frame; 1006 params = (caddr_t)frame.tf_esp + sizeof(int); 1007 code = frame.tf_eax; 1008 if (p->p_sysent->sv_prepsyscall) { 1009 (*p->p_sysent->sv_prepsyscall)(&frame, args, &code, ¶ms); 1010 } else { 1011 /* 1012 * Need to check if this is a 32 bit or 64 bit syscall. 1013 */ 1014 if (code == SYS_syscall) { 1015 /* 1016 * Code is first argument, followed by actual args. 1017 */ 1018 code = fuword(params); 1019 params += sizeof(int); 1020 } else if (code == SYS___syscall) { 1021 /* 1022 * Like syscall, but code is a quad, so as to maintain 1023 * quad alignment for the rest of the arguments. 1024 */ 1025 code = fuword(params); 1026 params += sizeof(quad_t); 1027 } 1028 } 1029 1030 if (p->p_sysent->sv_mask) 1031 code &= p->p_sysent->sv_mask; 1032 1033 if (code >= p->p_sysent->sv_size) 1034 callp = &p->p_sysent->sv_table[0]; 1035 else 1036 callp = &p->p_sysent->sv_table[code]; 1037 1038 if (params && (i = callp->sy_narg * sizeof(int)) && 1039 (error = copyin(params, (caddr_t)args, (u_int)i))) { 1040#ifdef KTRACE 1041 if (KTRPOINT(p, KTR_SYSCALL)) 1042 ktrsyscall(p->p_tracep, code, callp->sy_narg, args); 1043#endif 1044 goto bad; 1045 } 1046#ifdef KTRACE 1047 if (KTRPOINT(p, KTR_SYSCALL)) 1048 ktrsyscall(p->p_tracep, code, callp->sy_narg, args); 1049#endif 1050 p->p_retval[0] = 0; 1051 p->p_retval[1] = frame.tf_edx; 1052 1053 STOPEVENT(p, S_SCE, callp->sy_narg); 1054 1055 error = (*callp->sy_call)(p, args); 1056 1057 switch (error) { 1058 1059 case 0: 1060 /* 1061 * Reinitialize proc pointer `p' as it may be different 1062 * if this is a child returning from fork syscall. 1063 */ 1064 p = curproc; 1065 frame.tf_eax = p->p_retval[0]; 1066 frame.tf_edx = p->p_retval[1]; 1067 frame.tf_eflags &= ~PSL_C; 1068 break; 1069 1070 case ERESTART: 1071 /* 1072 * Reconstruct pc, assuming lcall $X,y is 7 bytes, 1073 * int 0x80 is 2 bytes. We saved this in tf_err. 1074 */ 1075 frame.tf_eip -= frame.tf_err; 1076 break; 1077 1078 case EJUSTRETURN: 1079 break; 1080 1081 default: 1082bad: 1083 if (p->p_sysent->sv_errsize) { 1084 if (error >= p->p_sysent->sv_errsize) 1085 error = -1; /* XXX */ 1086 else 1087 error = p->p_sysent->sv_errtbl[error]; 1088 } 1089 frame.tf_eax = error; 1090 frame.tf_eflags |= PSL_C; 1091 break; 1092 } 1093 1094 if ((frame.tf_eflags & PSL_T) && !(frame.tf_eflags & PSL_VM)) { 1095 /* Traced syscall. */ 1096 frame.tf_eflags &= ~PSL_T; 1097 trapsignal(p, SIGTRAP, 0); 1098 } 1099 1100 userret(p, &frame, sticks); 1101 1102#ifdef KTRACE 1103 if (KTRPOINT(p, KTR_SYSRET)) 1104 ktrsysret(p->p_tracep, code, error, p->p_retval[0]); 1105#endif 1106 1107 /* 1108 * This works because errno is findable through the 1109 * register set. If we ever support an emulation where this 1110 * is not the case, this code will need to be revisited. 1111 */ 1112 STOPEVENT(p, S_SCX, code); 1113 1114} 1115 1116/* 1117 * Simplified back end of syscall(), used when returning from fork() 1118 * directly into user mode. 1119 */ 1120void 1121fork_return(p, frame) 1122 struct proc *p; 1123 struct trapframe frame; 1124{ 1125 frame.tf_eax = 0; /* Child returns zero */ 1126 frame.tf_eflags &= ~PSL_C; /* success */ 1127 frame.tf_edx = 1; 1128 1129 userret(p, &frame, 0); 1130#ifdef KTRACE 1131 if (KTRPOINT(p, KTR_SYSRET)) 1132 ktrsysret(p->p_tracep, SYS_fork, 0, 0); 1133#endif 1134} 1135