sys/platform/pc32/i386/vm86.c

   1 /*-
   2  * Copyright (c) 1997 Jonathan Lemon
   3  * All rights reserved.
   4  *
   5  * Redistribution and use in source and binary forms, with or without
   6  * modification, are permitted provided that the following conditions
   7  * are met:
   8  * 1. Redistributions of source code must retain the above copyright
   9  *    notice, this list of conditions and the following disclaimer.
  10  * 2. Redistributions in binary form must reproduce the above copyright
  11  * modification, are permitted provided that the following conditions
  12  *    notice, this list of conditions and the following disclaimer in the
  13  *    documentation and/or other materials provided with the distribution.
  14  *
  15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
  16  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  17  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  18  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
  19  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  20  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  21  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  22  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  23  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  24  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  25  * SUCH DAMAGE.
  26  *
  27  * $FreeBSD: src/sys/i386/i386/vm86.c,v 1.31.2.2 2001/10/05 06:18:55 peter Exp $
  28  * $DragonFly: src/sys/platform/pc32/i386/vm86.c,v 1.16 2005/10/11 09:59:56 corecode Exp $
  29  */
  30
  31 #include <sys/param.h>
  32 #include <sys/systm.h>
  33 #include <sys/kernel.h>
  34 #include <sys/proc.h>
  35 #include <sys/lock.h>
  36 #include <sys/malloc.h>
  37 #include <sys/sysctl.h>
  38
  39 #include <vm/vm.h>
  40 #include <vm/pmap.h>
  41 #include <vm/vm_map.h>
  42 #include <vm/vm_page.h>
  43
  44 #include <sys/user.h>
  45 #include <sys/thread2.h>
  46
  47 #include <machine/md_var.h>
  48 #include <machine/pcb_ext.h>    /* pcb.h included via sys/user.h */
  49 #include <machine/psl.h>
  50 #include <machine/specialreg.h>
  51 #include <machine/sysarch.h>
  52 #include <machine/clock.h>
  53 #include <bus/isa/i386/isa.h>
  54 #include <bus/isa/rtc.h>
  55 #include <i386/isa/timerreg.h>
  56
  57 extern int i386_extend_pcb      (struct lwp *);
  58 extern int vm86pa;
  59 extern struct pcb *vm86pcb;
  60
  61 extern int vm86_bioscall(struct vm86frame *);
  62 extern void vm86_biosret(struct vm86frame *);
  63
  64 #define PGTABLE_SIZE    ((1024 + 64) * 1024 / PAGE_SIZE)
  65 #define INTMAP_SIZE     32
  66 #define IOMAP_SIZE      ctob(IOPAGES)
  67 #define TSS_SIZE \
  68         (sizeof(struct pcb_ext) - sizeof(struct segment_descriptor) + \
  69          INTMAP_SIZE + IOMAP_SIZE + 1)
  70
  71 struct vm86_layout {
  72         pt_entry_t      vml_pgtbl[PGTABLE_SIZE];
  73         struct  pcb vml_pcb;
  74         struct  pcb_ext vml_ext;
  75         char    vml_intmap[INTMAP_SIZE];
  76         char    vml_iomap[IOMAP_SIZE];
  77         char    vml_iomap_trailer;
  78 };
  79
  80 void vm86_prepcall(struct vm86frame);
  81
  82 struct system_map {
  83         int             type;
  84         vm_offset_t     start;
  85         vm_offset_t     end;
  86 };
  87
  88 #define HLT     0xf4
  89 #define CLI     0xfa
  90 #define STI     0xfb
  91 #define PUSHF   0x9c
  92 #define POPF    0x9d
  93 #define INTn    0xcd
  94 #define IRET    0xcf
  95 #define INB     0xe4
  96 #define INW     0xe5
  97 #define INBDX   0xec
  98 #define INWDX   0xed
  99 #define OUTB    0xe6
 100 #define OUTW    0xe7
 101 #define OUTBDX  0xee
 102 #define OUTWDX  0xef
 103 #define CALLm   0xff
 104 #define OPERAND_SIZE_PREFIX     0x66
 105 #define ADDRESS_SIZE_PREFIX     0x67
 106 #define PUSH_MASK       ~(PSL_VM | PSL_RF | PSL_I)
 107 #define POP_MASK        ~(PSL_VIP | PSL_VIF | PSL_VM | PSL_RF | PSL_IOPL)
 108
 109 static void vm86_setup_timer_fault(void);
 110 static void vm86_clear_timer_fault(void);
 111
 112 static int vm86_blew_up_timer;
 113
 114 static int timer_warn = 1;
 115 SYSCTL_INT(_debug, OID_AUTO, timer_warn, CTLFLAG_RW, &timer_warn, 0, "");
 116
 117 static __inline caddr_t
 118 MAKE_ADDR(u_short sel, u_short off)
 119 {
 120         return ((caddr_t)((sel << 4) + off));
 121 }
 122
 123 static __inline void
 124 GET_VEC(u_int vec, u_short *sel, u_short *off)
 125 {
 126         *sel = vec >> 16;
 127         *off = vec & 0xffff;
 128 }
 129
 130 static __inline u_int
 131 MAKE_VEC(u_short sel, u_short off)
 132 {
 133         return ((sel << 16) | off);
 134 }
 135
 136 static __inline void
 137 PUSH(u_short x, struct vm86frame *vmf)
 138 {
 139         vmf->vmf_sp -= 2;
 140         susword(MAKE_ADDR(vmf->vmf_ss, vmf->vmf_sp), x);
 141 }
 142
 143 static __inline void
 144 PUSHL(u_int x, struct vm86frame *vmf)
 145 {
 146         vmf->vmf_sp -= 4;
 147         suword(MAKE_ADDR(vmf->vmf_ss, vmf->vmf_sp), x);
 148 }
 149
 150 static __inline u_short
 151 POP(struct vm86frame *vmf)
 152 {
 153         u_short x = fusword(MAKE_ADDR(vmf->vmf_ss, vmf->vmf_sp));
 154
 155         vmf->vmf_sp += 2;
 156         return (x);
 157 }
 158
 159 static __inline u_int
 160 POPL(struct vm86frame *vmf)
 161 {
 162         u_int x = fuword(MAKE_ADDR(vmf->vmf_ss, vmf->vmf_sp));
 163
 164         vmf->vmf_sp += 4;
 165         return (x);
 166 }
 167
 168 int
 169 vm86_emulate(vmf)
 170         struct vm86frame *vmf;
 171 {
 172         struct vm86_kernel *vm86;
 173         caddr_t addr;
 174         u_char i_byte;
 175         u_int temp_flags;
 176         int inc_ip = 1;
 177         int retcode = 0;
 178
 179         /*
 180          * pcb_ext contains the address of the extension area, or zero if
 181          * the extension is not present.  (This check should not be needed,
 182          * as we can't enter vm86 mode until we set up an extension area)
 183          */
 184         if (curthread->td_pcb->pcb_ext == 0)
 185                 return (SIGBUS);
 186         vm86 = &curthread->td_pcb->pcb_ext->ext_vm86;
 187
 188         if (vmf->vmf_eflags & PSL_T)
 189                 retcode = SIGTRAP;
 190
 191         /*
 192          * Instruction emulation
 193          */
 194         addr = MAKE_ADDR(vmf->vmf_cs, vmf->vmf_ip);
 195         i_byte = fubyte(addr);
 196         if (i_byte == ADDRESS_SIZE_PREFIX) {
 197                 i_byte = fubyte(++addr);
 198                 inc_ip++;
 199         }
 200
 201         /*
 202          * I/O emulation (TIMER only, a big hack).  Just reenable the
 203          * IO bits involved, flag it, and retry the instruction.
 204          */
 205         switch(i_byte) {
 206         case OUTB:
 207         case OUTW:
 208         case OUTBDX:
 209         case OUTWDX:
 210                 vm86_blew_up_timer = 1;
 211                 /* fall through */
 212         case INB:
 213         case INW:
 214         case INBDX:
 215         case INWDX:
 216                 vm86_clear_timer_fault();
 217                 /* retry insn */
 218                 return(0);
 219         }
 220
 221         if (vm86->vm86_has_vme) {
 222                 switch (i_byte) {
 223                 case OPERAND_SIZE_PREFIX:
 224                         i_byte = fubyte(++addr);
 225                         inc_ip++;
 226                         switch (i_byte) {
 227                         case PUSHF:
 228                                 if (vmf->vmf_eflags & PSL_VIF)
 229                                         PUSHL((vmf->vmf_eflags & PUSH_MASK)
 230                                             | PSL_IOPL | PSL_I, vmf);
 231                                 else
 232                                         PUSHL((vmf->vmf_eflags & PUSH_MASK)
 233                                             | PSL_IOPL, vmf);
 234                                 vmf->vmf_ip += inc_ip;
 235                                 return (0);
 236
 237                         case POPF:
 238                                 temp_flags = POPL(vmf) & POP_MASK;
 239                                 vmf->vmf_eflags = (vmf->vmf_eflags & ~POP_MASK)
 240                                     | temp_flags | PSL_VM | PSL_I;
 241                                 vmf->vmf_ip += inc_ip;
 242                                 if (temp_flags & PSL_I) {
 243                                         vmf->vmf_eflags |= PSL_VIF;
 244                                         if (vmf->vmf_eflags & PSL_VIP)
 245                                                 break;
 246                                 } else {
 247                                         vmf->vmf_eflags &= ~PSL_VIF;
 248                                 }
 249                                 return (0);
 250                         }
 251                         break;
 252
 253                 /* VME faults here if VIP is set, but does not set VIF. */
 254                 case STI:
 255                         vmf->vmf_eflags |= PSL_VIF;
 256                         vmf->vmf_ip += inc_ip;
 257                         if ((vmf->vmf_eflags & PSL_VIP) == 0) {
 258                                 uprintf("fatal sti\n");
 259                                 return (SIGKILL);
 260                         }
 261                         break;
 262
 263                 /* VME if no redirection support */
 264                 case INTn:
 265                         break;
 266
 267                 /* VME if trying to set PSL_TF, or PSL_I when VIP is set */
 268                 case POPF:
 269                         temp_flags = POP(vmf) & POP_MASK;
 270                         vmf->vmf_flags = (vmf->vmf_flags & ~POP_MASK)
 271                             | temp_flags | PSL_VM | PSL_I;
 272                         vmf->vmf_ip += inc_ip;
 273                         if (temp_flags & PSL_I) {
 274                                 vmf->vmf_eflags |= PSL_VIF;
 275                                 if (vmf->vmf_eflags & PSL_VIP)
 276                                         break;
 277                         } else {
 278                                 vmf->vmf_eflags &= ~PSL_VIF;
 279                         }
 280                         return (retcode);
 281
 282                 /* VME if trying to set PSL_TF, or PSL_I when VIP is set */
 283                 case IRET:
 284                         vmf->vmf_ip = POP(vmf);
 285                         vmf->vmf_cs = POP(vmf);
 286                         temp_flags = POP(vmf) & POP_MASK;
 287                         vmf->vmf_flags = (vmf->vmf_flags & ~POP_MASK)
 288                             | temp_flags | PSL_VM | PSL_I;
 289                         if (temp_flags & PSL_I) {
 290                                 vmf->vmf_eflags |= PSL_VIF;
 291                                 if (vmf->vmf_eflags & PSL_VIP)
 292                                         break;
 293                         } else {
 294                                 vmf->vmf_eflags &= ~PSL_VIF;
 295                         }
 296                         return (retcode);
 297
 298                 }
 299                 return (SIGBUS);
 300         }
 301
 302         switch (i_byte) {
 303         case OPERAND_SIZE_PREFIX:
 304                 i_byte = fubyte(++addr);
 305                 inc_ip++;
 306                 switch (i_byte) {
 307                 case PUSHF:
 308                         if (vm86->vm86_eflags & PSL_VIF)
 309                                 PUSHL((vmf->vmf_flags & PUSH_MASK)
 310                                     | PSL_IOPL | PSL_I, vmf);
 311                         else
 312                                 PUSHL((vmf->vmf_flags & PUSH_MASK)
 313                                     | PSL_IOPL, vmf);
 314                         vmf->vmf_ip += inc_ip;
 315                         return (retcode);
 316
 317                 case POPF:
 318                         temp_flags = POPL(vmf) & POP_MASK;
 319                         vmf->vmf_eflags = (vmf->vmf_eflags & ~POP_MASK)
 320                             | temp_flags | PSL_VM | PSL_I;
 321                         vmf->vmf_ip += inc_ip;
 322                         if (temp_flags & PSL_I) {
 323                                 vm86->vm86_eflags |= PSL_VIF;
 324                                 if (vm86->vm86_eflags & PSL_VIP)
 325                                         break;
 326                         } else {
 327                                 vm86->vm86_eflags &= ~PSL_VIF;
 328                         }
 329                         return (retcode);
 330                 }
 331                 return (SIGBUS);
 332
 333         case CLI:
 334                 vm86->vm86_eflags &= ~PSL_VIF;
 335                 vmf->vmf_ip += inc_ip;
 336                 return (retcode);
 337
 338         case STI:
 339                 /* if there is a pending interrupt, go to the emulator */
 340                 vm86->vm86_eflags |= PSL_VIF;
 341                 vmf->vmf_ip += inc_ip;
 342                 if (vm86->vm86_eflags & PSL_VIP)
 343                         break;
 344                 return (retcode);
 345
 346         case PUSHF:
 347                 if (vm86->vm86_eflags & PSL_VIF)
 348                         PUSH((vmf->vmf_flags & PUSH_MASK)
 349                             | PSL_IOPL | PSL_I, vmf);
 350                 else
 351                         PUSH((vmf->vmf_flags & PUSH_MASK) | PSL_IOPL, vmf);
 352                 vmf->vmf_ip += inc_ip;
 353                 return (retcode);
 354
 355         case INTn:
 356                 i_byte = fubyte(addr + 1);
 357                 if ((vm86->vm86_intmap[i_byte >> 3] & (1 << (i_byte & 7))) != 0)
 358                         break;
 359                 if (vm86->vm86_eflags & PSL_VIF)
 360                         PUSH((vmf->vmf_flags & PUSH_MASK)
 361                             | PSL_IOPL | PSL_I, vmf);
 362                 else
 363                         PUSH((vmf->vmf_flags & PUSH_MASK) | PSL_IOPL, vmf);
 364                 PUSH(vmf->vmf_cs, vmf);
 365                 PUSH(vmf->vmf_ip + inc_ip + 1, vmf);    /* increment IP */
 366                 GET_VEC(fuword((caddr_t)(i_byte * 4)),
 367                      &vmf->vmf_cs, &vmf->vmf_ip);
 368                 vmf->vmf_flags &= ~PSL_T;
 369                 vm86->vm86_eflags &= ~PSL_VIF;
 370                 return (retcode);
 371
 372         case IRET:
 373                 vmf->vmf_ip = POP(vmf);
 374                 vmf->vmf_cs = POP(vmf);
 375                 temp_flags = POP(vmf) & POP_MASK;
 376                 vmf->vmf_flags = (vmf->vmf_flags & ~POP_MASK)
 377                     | temp_flags | PSL_VM | PSL_I;
 378                 if (temp_flags & PSL_I) {
 379                         vm86->vm86_eflags |= PSL_VIF;
 380                         if (vm86->vm86_eflags & PSL_VIP)
 381                                 break;
 382                 } else {
 383                         vm86->vm86_eflags &= ~PSL_VIF;
 384                 }
 385                 return (retcode);
 386
 387         case POPF:
 388                 temp_flags = POP(vmf) & POP_MASK;
 389                 vmf->vmf_flags = (vmf->vmf_flags & ~POP_MASK)
 390                     | temp_flags | PSL_VM | PSL_I;
 391                 vmf->vmf_ip += inc_ip;
 392                 if (temp_flags & PSL_I) {
 393                         vm86->vm86_eflags |= PSL_VIF;
 394                         if (vm86->vm86_eflags & PSL_VIP)
 395                                 break;
 396                 } else {
 397                         vm86->vm86_eflags &= ~PSL_VIF;
 398                 }
 399                 return (retcode);
 400         }
 401         return (SIGBUS);
 402 }
 403
 404 void
 405 vm86_initialize(void)
 406 {
 407         int i;
 408         u_int *addr;
 409         struct vm86_layout *vml = (struct vm86_layout *)vm86paddr;
 410         struct pcb *pcb;
 411         struct pcb_ext *ext;
 412         struct soft_segment_descriptor ssd = {
 413                 0,                      /* segment base address (overwritten) */
 414                 0,                      /* length (overwritten) */
 415                 SDT_SYS386TSS,          /* segment type */
 416                 0,                      /* priority level */
 417                 1,                      /* descriptor present */
 418                 0, 0,
 419                 0,                      /* default 16 size */
 420                 0                       /* granularity */
 421         };
 422
 423         /*
 424          * this should be a compile time error, but cpp doesn't grok sizeof().
 425          */
 426         if (sizeof(struct vm86_layout) > ctob(3))
 427                 panic("struct vm86_layout exceeds space allocated in locore.s");
 428
 429         /*
 430          * Below is the memory layout that we use for the vm86 region.
 431          *
 432          * +--------+
 433          * |        |
 434          * |        |
 435          * | page 0 |
 436          * |        | +--------+
 437          * |        | | stack  |
 438          * +--------+ +--------+ <--------- vm86paddr
 439          * |        | |Page Tbl| 1M + 64K = 272 entries = 1088 bytes
 440          * |        | +--------+
 441          * |        | |  PCB   | size: ~240 bytes
 442          * | page 1 | |PCB Ext | size: ~140 bytes (includes TSS)
 443          * |        | +--------+
 444          * |        | |int map |
 445          * |        | +--------+
 446          * +--------+ |        |
 447          * | page 2 | |  I/O   |
 448          * +--------+ | bitmap |
 449          * | page 3 | |        |
 450          * |        | +--------+
 451          * +--------+
 452          */
 453
 454         /*
 455          * A rudimentary PCB must be installed, in order to get to the
 456          * PCB extension area.  We use the PCB area as a scratchpad for
 457          * data storage, the layout of which is shown below.
 458          *
 459          * pcb_esi      = new PTD entry 0
 460          * pcb_ebp      = pointer to frame on vm86 stack
 461          * pcb_esp      =    stack frame pointer at time of switch
 462          * pcb_ebx      = va of vm86 page table
 463          * pcb_eip      =    argument pointer to initial call
 464          * pcb_spare[0] =    saved TSS descriptor, word 0
 465          * pcb_space[1] =    saved TSS descriptor, word 1
 466          */
 467 #define new_ptd         pcb_esi
 468 #define vm86_frame      pcb_ebp
 469 #define pgtable_va      pcb_ebx
 470
 471         pcb = &vml->vml_pcb;
 472         ext = &vml->vml_ext;
 473
 474         bzero(pcb, sizeof(struct pcb));
 475         pcb->new_ptd = vm86pa | PG_V | PG_RW | PG_U;
 476         pcb->vm86_frame = (pt_entry_t)vm86paddr - sizeof(struct vm86frame);
 477         pcb->pgtable_va = (vm_offset_t)vm86paddr;
 478         pcb->pcb_ext = ext;
 479
 480         bzero(ext, sizeof(struct pcb_ext));
 481         ext->ext_tss.tss_esp0 = (vm_offset_t)vm86paddr;
 482         ext->ext_tss.tss_ss0 = GSEL(GDATA_SEL, SEL_KPL);
 483         ext->ext_tss.tss_ioopt =
 484                 ((u_int)vml->vml_iomap - (u_int)&ext->ext_tss) << 16;
 485         ext->ext_iomap = vml->vml_iomap;
 486         ext->ext_vm86.vm86_intmap = vml->vml_intmap;
 487
 488         if (cpu_feature & CPUID_VME)
 489                 ext->ext_vm86.vm86_has_vme = (rcr4() & CR4_VME ? 1 : 0);
 490
 491         addr = (u_int *)ext->ext_vm86.vm86_intmap;
 492         for (i = 0; i < (INTMAP_SIZE + IOMAP_SIZE) / sizeof(u_int); i++)
 493                 *addr++ = 0;
 494         vml->vml_iomap_trailer = 0xff;
 495
 496         ssd.ssd_base = (u_int)&ext->ext_tss;
 497         ssd.ssd_limit = TSS_SIZE - 1;
 498         ssdtosd(&ssd, &ext->ext_tssd);
 499
 500         vm86pcb = pcb;
 501
 502 #if 0
 503         /*
 504          * use whatever is leftover of the vm86 page layout as a
 505          * message buffer so we can capture early output.
 506          */
 507         msgbufinit((vm_offset_t)vm86paddr + sizeof(struct vm86_layout),
 508             ctob(3) - sizeof(struct vm86_layout));
 509 #endif
 510 }
 511
 512 vm_offset_t
 513 vm86_getpage(struct vm86context *vmc, int pagenum)
 514 {
 515         int i;
 516
 517         for (i = 0; i < vmc->npages; i++)
 518                 if (vmc->pmap[i].pte_num == pagenum)
 519                         return (vmc->pmap[i].kva);
 520         return (0);
 521 }
 522
 523 vm_offset_t
 524 vm86_addpage(struct vm86context *vmc, int pagenum, vm_offset_t kva)
 525 {
 526         int i, flags = 0;
 527
 528         for (i = 0; i < vmc->npages; i++)
 529                 if (vmc->pmap[i].pte_num == pagenum)
 530                         goto bad;
 531
 532         if (vmc->npages == VM86_PMAPSIZE)
 533                 goto bad;                       /* XXX grow map? */
 534
 535         if (kva == 0) {
 536                 kva = (vm_offset_t)malloc(PAGE_SIZE, M_TEMP, M_WAITOK);
 537                 flags = VMAP_MALLOC;
 538         }
 539
 540         i = vmc->npages++;
 541         vmc->pmap[i].flags = flags;
 542         vmc->pmap[i].kva = kva;
 543         vmc->pmap[i].pte_num = pagenum;
 544         return (kva);
 545 bad:
 546         panic("vm86_addpage: not enough room, or overlap");
 547 }
 548
 549 static void
 550 vm86_initflags(struct vm86frame *vmf)
 551 {
 552         int eflags = vmf->vmf_eflags;
 553         struct vm86_kernel *vm86 = &curthread->td_pcb->pcb_ext->ext_vm86;
 554
 555         if (vm86->vm86_has_vme) {
 556                 eflags = (vmf->vmf_eflags & ~VME_USERCHANGE) |
 557                     (eflags & VME_USERCHANGE) | PSL_VM;
 558         } else {
 559                 vm86->vm86_eflags = eflags;     /* save VIF, VIP */
 560                 eflags = (vmf->vmf_eflags & ~VM_USERCHANGE) |
 561                     (eflags & VM_USERCHANGE) | PSL_VM;
 562         }
 563         vmf->vmf_eflags = eflags | PSL_VM;
 564 }
 565
 566 /*
 567  * called from vm86_bioscall, while in vm86 address space, to finalize setup.
 568  */
 569 void
 570 vm86_prepcall(struct vm86frame vmf)
 571 {
 572         uintptr_t addr[] = { 0xA00, 0x1000 };   /* code, stack */
 573         u_char intcall[] = {
 574                 CLI, INTn, 0x00, STI, HLT
 575         };
 576
 577         if ((vmf.vmf_trapno & PAGE_MASK) <= 0xff) {
 578                 /* interrupt call requested */
 579                 intcall[2] = (u_char)(vmf.vmf_trapno & 0xff);
 580                 memcpy((void *)addr[0], (void *)intcall, sizeof(intcall));
 581                 vmf.vmf_ip = addr[0];
 582                 vmf.vmf_cs = 0;
 583         }
 584         vmf.vmf_sp = addr[1] - 2;              /* keep aligned */
 585         vmf.kernel_fs = vmf.kernel_es = vmf.kernel_ds = 0;
 586         vmf.vmf_ss = 0;
 587         vmf.vmf_eflags = PSL_VIF | PSL_VM | PSL_USER;
 588         vm86_initflags(&vmf);
 589 }
 590
 591 /*
 592  * vm86 trap handler; determines whether routine succeeded or not.
 593  * Called while in vm86 space, returns to calling process.
 594  *
 595  * A MP lock ref is held on entry from trap() and must be released prior
 596  * to returning to the VM86 call.
 597  */
 598 void
 599 vm86_trap(struct vm86frame *vmf)
 600 {
 601         caddr_t addr;
 602
 603         /* "should not happen" */
 604         if ((vmf->vmf_eflags & PSL_VM) == 0)
 605                 panic("vm86_trap called, but not in vm86 mode");
 606
 607         addr = MAKE_ADDR(vmf->vmf_cs, vmf->vmf_ip);
 608         if (*(u_char *)addr == HLT)
 609                 vmf->vmf_trapno = vmf->vmf_eflags & PSL_C;
 610         else
 611                 vmf->vmf_trapno = vmf->vmf_trapno << 16;
 612
 613         rel_mplock();
 614         vm86_biosret(vmf);
 615 }
 616
 617 int
 618 vm86_intcall(int intnum, struct vm86frame *vmf)
 619 {
 620         int error;
 621
 622         if (intnum < 0 || intnum > 0xff)
 623                 return (EINVAL);
 624
 625         crit_enter();
 626         ASSERT_MP_LOCK_HELD(curthread);
 627
 628         vm86_setup_timer_fault();
 629         vmf->vmf_trapno = intnum;
 630         error = vm86_bioscall(vmf);
 631
 632         /*
 633          * Yes, this happens, especially with video BIOS calls.  The BIOS
 634          * will sometimes eat timer 2 for lunch, and we need timer 2.
 635          */
 636         if (vm86_blew_up_timer) {
 637                 vm86_blew_up_timer = 0;
 638                 timer_restore();
 639                 if (timer_warn) {
 640                         printf("Warning: BIOS played with the 8254, "
 641                                 "resetting it\n");
 642                 }
 643         }
 644         crit_exit();
 645         return(error);
 646 }
 647
 648 /*
 649  * struct vm86context contains the page table to use when making
 650  * vm86 calls.  If intnum is a valid interrupt number (0-255), then
 651  * the "interrupt trampoline" will be used, otherwise we use the
 652  * caller's cs:ip routine.
 653  */
 654 int
 655 vm86_datacall(intnum, vmf, vmc)
 656         int intnum;
 657         struct vm86frame *vmf;
 658         struct vm86context *vmc;
 659 {
 660         pt_entry_t *pte = vm86paddr;
 661         u_int page;
 662         int i, entry, retval;
 663
 664         crit_enter();
 665         ASSERT_MP_LOCK_HELD(curthread);
 666
 667         for (i = 0; i < vmc->npages; i++) {
 668                 page = vtophys(vmc->pmap[i].kva & PG_FRAME);
 669                 entry = vmc->pmap[i].pte_num;
 670                 vmc->pmap[i].old_pte = pte[entry];
 671                 pte[entry] = page | PG_V | PG_RW | PG_U;
 672         }
 673
 674         vmf->vmf_trapno = intnum;
 675         retval = vm86_bioscall(vmf);
 676
 677         for (i = 0; i < vmc->npages; i++) {
 678                 entry = vmc->pmap[i].pte_num;
 679                 pte[entry] = vmc->pmap[i].old_pte;
 680         }
 681         crit_exit();
 682         return (retval);
 683 }
 684
 685 vm_offset_t
 686 vm86_getaddr(vmc, sel, off)
 687         struct vm86context *vmc;
 688         u_short sel;
 689         u_short off;
 690 {
 691         int i, page;
 692         vm_offset_t addr;
 693
 694         addr = (vm_offset_t)MAKE_ADDR(sel, off);
 695         page = addr >> PAGE_SHIFT;
 696         for (i = 0; i < vmc->npages; i++)
 697                 if (page == vmc->pmap[i].pte_num)
 698                         return (vmc->pmap[i].kva + (addr & PAGE_MASK));
 699         return (0);
 700 }
 701
 702 int
 703 vm86_getptr(vmc, kva, sel, off)
 704         struct vm86context *vmc;
 705         vm_offset_t kva;
 706         u_short *sel;
 707         u_short *off;
 708 {
 709         int i;
 710
 711         for (i = 0; i < vmc->npages; i++)
 712                 if (kva >= vmc->pmap[i].kva &&
 713                     kva < vmc->pmap[i].kva + PAGE_SIZE) {
 714                         *off = kva - vmc->pmap[i].kva;
 715                         *sel = vmc->pmap[i].pte_num << 8;
 716                         return (1);
 717                 }
 718         return (0);
 719         panic("vm86_getptr: address not found");
 720 }
 721
 722 int
 723 vm86_sysarch(struct lwp *lp, char *args)
 724 {
 725         int error = 0;
 726         struct i386_vm86_args ua;
 727         struct vm86_kernel *vm86;
 728
 729         if ((error = copyin(args, &ua, sizeof(struct i386_vm86_args))) != 0)
 730                 return (error);
 731
 732         if (lp->lwp_thread->td_pcb->pcb_ext == 0)
 733                 if ((error = i386_extend_pcb(lp)) != 0)
 734                         return (error);
 735         vm86 = &lp->lwp_thread->td_pcb->pcb_ext->ext_vm86;
 736
 737         switch (ua.sub_op) {
 738         case VM86_INIT: {
 739                 struct vm86_init_args sa;
 740
 741                 if ((error = copyin(ua.sub_args, &sa, sizeof(sa))) != 0)
 742                         return (error);
 743                 if (cpu_feature & CPUID_VME)
 744                         vm86->vm86_has_vme = (rcr4() & CR4_VME ? 1 : 0);
 745                 else
 746                         vm86->vm86_has_vme = 0;
 747                 vm86->vm86_inited = 1;
 748                 vm86->vm86_debug = sa.debug;
 749                 bcopy(&sa.int_map, vm86->vm86_intmap, 32);
 750                 }
 751                 break;
 752
 753 #if 0
 754         case VM86_SET_VME: {
 755                 struct vm86_vme_args sa;
 756
 757                 if ((cpu_feature & CPUID_VME) == 0)
 758                         return (ENODEV);
 759
 760                 if (error = copyin(ua.sub_args, &sa, sizeof(sa)))
 761                         return (error);
 762                 if (sa.state)
 763                         load_cr4(rcr4() | CR4_VME);
 764                 else
 765                         load_cr4(rcr4() & ~CR4_VME);
 766                 }
 767                 break;
 768 #endif
 769
 770         case VM86_GET_VME: {
 771                 struct vm86_vme_args sa;
 772
 773                 sa.state = (rcr4() & CR4_VME ? 1 : 0);
 774                 error = copyout(&sa, ua.sub_args, sizeof(sa));
 775                 }
 776                 break;
 777
 778         case VM86_INTCALL: {
 779                 struct vm86_intcall_args sa;
 780
 781                 if ((error = suser_cred(lp->lwp_proc->p_ucred, 0)))
 782                         return (error);
 783                 if ((error = copyin(ua.sub_args, &sa, sizeof(sa))))
 784                         return (error);
 785                 if ((error = vm86_intcall(sa.intnum, &sa.vmf)))
 786                         return (error);
 787                 error = copyout(&sa, ua.sub_args, sizeof(sa));
 788                 }
 789                 break;
 790
 791         default:
 792                 error = EINVAL;
 793         }
 794         return (error);
 795 }
 796
 797 /*
 798  * Setup the VM86 I/O map to take faults on the timer
 799  */
 800 static void
 801 vm86_setup_timer_fault(void)
 802 {
 803         struct vm86_layout *vml = (struct vm86_layout *)vm86paddr;
 804
 805         vml->vml_iomap[TIMER_MODE >> 3] |= 1 << (TIMER_MODE & 7);
 806         vml->vml_iomap[TIMER_CNTR0 >> 3] |= 1 << (TIMER_CNTR0 & 7);
 807         vml->vml_iomap[TIMER_CNTR1 >> 3] |= 1 << (TIMER_CNTR1 & 7);
 808         vml->vml_iomap[TIMER_CNTR2 >> 3] |= 1 << (TIMER_CNTR2 & 7);
 809 }
 810
 811 /*
 812  * Setup the VM86 I/O map to not fault on the timer
 813  */
 814 static void
 815 vm86_clear_timer_fault(void)
 816 {
 817         struct vm86_layout *vml = (struct vm86_layout *)vm86paddr;
 818
 819         vml->vml_iomap[TIMER_MODE >> 3] &= ~(1 << (TIMER_MODE & 7));
 820         vml->vml_iomap[TIMER_CNTR0 >> 3] &= ~(1 << (TIMER_CNTR0 & 7));
 821         vml->vml_iomap[TIMER_CNTR1 >> 3] &= ~(1 << (TIMER_CNTR1 & 7));
 822         vml->vml_iomap[TIMER_CNTR2 >> 3] &= ~(1 << (TIMER_CNTR2 & 7));
 823 }
 824