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 1.25 2007/01/22 19:37:04 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 <machine_base/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 /*
 169  * MPSAFE
 170  */
 171 int
 172 vm86_emulate(struct vm86frame *vmf)
 173 {
 174         struct vm86_kernel *vm86;
 175         caddr_t addr;
 176         u_char i_byte;
 177         u_int temp_flags;
 178         int inc_ip = 1;
 179         int retcode = 0;
 180
 181         /*
 182          * pcb_ext contains the address of the extension area, or zero if
 183          * the extension is not present.  (This check should not be needed,
 184          * as we can't enter vm86 mode until we set up an extension area)
 185          */
 186         if (curthread->td_pcb->pcb_ext == 0)
 187                 return (SIGBUS);
 188         vm86 = &curthread->td_pcb->pcb_ext->ext_vm86;
 189
 190         if (vmf->vmf_eflags & PSL_T)
 191                 retcode = SIGTRAP;
 192
 193         /*
 194          * Instruction emulation
 195          */
 196         addr = MAKE_ADDR(vmf->vmf_cs, vmf->vmf_ip);
 197         i_byte = fubyte(addr);
 198         if (i_byte == ADDRESS_SIZE_PREFIX) {
 199                 i_byte = fubyte(++addr);
 200                 inc_ip++;
 201         }
 202
 203         /*
 204          * I/O emulation (TIMER only, a big hack).  Just reenable the
 205          * IO bits involved, flag it, and retry the instruction.
 206          */
 207         switch(i_byte) {
 208         case OUTB:
 209         case OUTW:
 210         case OUTBDX:
 211         case OUTWDX:
 212                 vm86_blew_up_timer = 1;
 213                 /* fall through */
 214         case INB:
 215         case INW:
 216         case INBDX:
 217         case INWDX:
 218                 vm86_clear_timer_fault();
 219                 /* retry insn */
 220                 return(0);
 221         }
 222
 223         if (vm86->vm86_has_vme) {
 224                 switch (i_byte) {
 225                 case OPERAND_SIZE_PREFIX:
 226                         i_byte = fubyte(++addr);
 227                         inc_ip++;
 228                         switch (i_byte) {
 229                         case PUSHF:
 230                                 if (vmf->vmf_eflags & PSL_VIF)
 231                                         PUSHL((vmf->vmf_eflags & PUSH_MASK)
 232                                             | PSL_IOPL | PSL_I, vmf);
 233                                 else
 234                                         PUSHL((vmf->vmf_eflags & PUSH_MASK)
 235                                             | PSL_IOPL, vmf);
 236                                 vmf->vmf_ip += inc_ip;
 237                                 return (0);
 238
 239                         case POPF:
 240                                 temp_flags = POPL(vmf) & POP_MASK;
 241                                 vmf->vmf_eflags = (vmf->vmf_eflags & ~POP_MASK)
 242                                     | temp_flags | PSL_VM | PSL_I;
 243                                 vmf->vmf_ip += inc_ip;
 244                                 if (temp_flags & PSL_I) {
 245                                         vmf->vmf_eflags |= PSL_VIF;
 246                                         if (vmf->vmf_eflags & PSL_VIP)
 247                                                 break;
 248                                 } else {
 249                                         vmf->vmf_eflags &= ~PSL_VIF;
 250                                 }
 251                                 return (0);
 252                         }
 253                         break;
 254
 255                 /* VME faults here if VIP is set, but does not set VIF. */
 256                 case STI:
 257                         vmf->vmf_eflags |= PSL_VIF;
 258                         vmf->vmf_ip += inc_ip;
 259                         if ((vmf->vmf_eflags & PSL_VIP) == 0) {
 260                                 uprintf("fatal sti\n");
 261                                 return (SIGKILL);
 262                         }
 263                         break;
 264
 265                 /* VME if no redirection support */
 266                 case INTn:
 267                         break;
 268
 269                 /* VME if trying to set PSL_TF, or PSL_I when VIP is set */
 270                 case POPF:
 271                         temp_flags = POP(vmf) & POP_MASK;
 272                         vmf->vmf_flags = (vmf->vmf_flags & ~POP_MASK)
 273                             | temp_flags | PSL_VM | PSL_I;
 274                         vmf->vmf_ip += inc_ip;
 275                         if (temp_flags & PSL_I) {
 276                                 vmf->vmf_eflags |= PSL_VIF;
 277                                 if (vmf->vmf_eflags & PSL_VIP)
 278                                         break;
 279                         } else {
 280                                 vmf->vmf_eflags &= ~PSL_VIF;
 281                         }
 282                         return (retcode);
 283
 284                 /* VME if trying to set PSL_TF, or PSL_I when VIP is set */
 285                 case IRET:
 286                         vmf->vmf_ip = POP(vmf);
 287                         vmf->vmf_cs = POP(vmf);
 288                         temp_flags = POP(vmf) & POP_MASK;
 289                         vmf->vmf_flags = (vmf->vmf_flags & ~POP_MASK)
 290                             | temp_flags | PSL_VM | PSL_I;
 291                         if (temp_flags & PSL_I) {
 292                                 vmf->vmf_eflags |= PSL_VIF;
 293                                 if (vmf->vmf_eflags & PSL_VIP)
 294                                         break;
 295                         } else {
 296                                 vmf->vmf_eflags &= ~PSL_VIF;
 297                         }
 298                         return (retcode);
 299
 300                 }
 301                 return (SIGBUS);
 302         }
 303
 304         switch (i_byte) {
 305         case OPERAND_SIZE_PREFIX:
 306                 i_byte = fubyte(++addr);
 307                 inc_ip++;
 308                 switch (i_byte) {
 309                 case PUSHF:
 310                         if (vm86->vm86_eflags & PSL_VIF)
 311                                 PUSHL((vmf->vmf_flags & PUSH_MASK)
 312                                     | PSL_IOPL | PSL_I, vmf);
 313                         else
 314                                 PUSHL((vmf->vmf_flags & PUSH_MASK)
 315                                     | PSL_IOPL, vmf);
 316                         vmf->vmf_ip += inc_ip;
 317                         return (retcode);
 318
 319                 case POPF:
 320                         temp_flags = POPL(vmf) & POP_MASK;
 321                         vmf->vmf_eflags = (vmf->vmf_eflags & ~POP_MASK)
 322                             | temp_flags | PSL_VM | PSL_I;
 323                         vmf->vmf_ip += inc_ip;
 324                         if (temp_flags & PSL_I) {
 325                                 vm86->vm86_eflags |= PSL_VIF;
 326                                 if (vm86->vm86_eflags & PSL_VIP)
 327                                         break;
 328                         } else {
 329                                 vm86->vm86_eflags &= ~PSL_VIF;
 330                         }
 331                         return (retcode);
 332                 }
 333                 return (SIGBUS);
 334
 335         case CLI:
 336                 vm86->vm86_eflags &= ~PSL_VIF;
 337                 vmf->vmf_ip += inc_ip;
 338                 return (retcode);
 339
 340         case STI:
 341                 /* if there is a pending interrupt, go to the emulator */
 342                 vm86->vm86_eflags |= PSL_VIF;
 343                 vmf->vmf_ip += inc_ip;
 344                 if (vm86->vm86_eflags & PSL_VIP)
 345                         break;
 346                 return (retcode);
 347
 348         case PUSHF:
 349                 if (vm86->vm86_eflags & PSL_VIF)
 350                         PUSH((vmf->vmf_flags & PUSH_MASK)
 351                             | PSL_IOPL | PSL_I, vmf);
 352                 else
 353                         PUSH((vmf->vmf_flags & PUSH_MASK) | PSL_IOPL, vmf);
 354                 vmf->vmf_ip += inc_ip;
 355                 return (retcode);
 356
 357         case INTn:
 358                 i_byte = fubyte(addr + 1);
 359                 if ((vm86->vm86_intmap[i_byte >> 3] & (1 << (i_byte & 7))) != 0)
 360                         break;
 361                 if (vm86->vm86_eflags & PSL_VIF)
 362                         PUSH((vmf->vmf_flags & PUSH_MASK)
 363                             | PSL_IOPL | PSL_I, vmf);
 364                 else
 365                         PUSH((vmf->vmf_flags & PUSH_MASK) | PSL_IOPL, vmf);
 366                 PUSH(vmf->vmf_cs, vmf);
 367                 PUSH(vmf->vmf_ip + inc_ip + 1, vmf);    /* increment IP */
 368                 GET_VEC(fuword((caddr_t)(i_byte * 4)),
 369                      &vmf->vmf_cs, &vmf->vmf_ip);
 370                 vmf->vmf_flags &= ~PSL_T;
 371                 vm86->vm86_eflags &= ~PSL_VIF;
 372                 return (retcode);
 373
 374         case IRET:
 375                 vmf->vmf_ip = POP(vmf);
 376                 vmf->vmf_cs = POP(vmf);
 377                 temp_flags = POP(vmf) & POP_MASK;
 378                 vmf->vmf_flags = (vmf->vmf_flags & ~POP_MASK)
 379                     | temp_flags | PSL_VM | PSL_I;
 380                 if (temp_flags & PSL_I) {
 381                         vm86->vm86_eflags |= PSL_VIF;
 382                         if (vm86->vm86_eflags & PSL_VIP)
 383                                 break;
 384                 } else {
 385                         vm86->vm86_eflags &= ~PSL_VIF;
 386                 }
 387                 return (retcode);
 388
 389         case POPF:
 390                 temp_flags = POP(vmf) & POP_MASK;
 391                 vmf->vmf_flags = (vmf->vmf_flags & ~POP_MASK)
 392                     | temp_flags | PSL_VM | PSL_I;
 393                 vmf->vmf_ip += inc_ip;
 394                 if (temp_flags & PSL_I) {
 395                         vm86->vm86_eflags |= PSL_VIF;
 396                         if (vm86->vm86_eflags & PSL_VIP)
 397                                 break;
 398                 } else {
 399                         vm86->vm86_eflags &= ~PSL_VIF;
 400                 }
 401                 return (retcode);
 402         }
 403         return (SIGBUS);
 404 }
 405
 406 void
 407 vm86_initialize(void)
 408 {
 409         int i;
 410         u_int *addr;
 411         struct vm86_layout *vml = (struct vm86_layout *)vm86paddr;
 412         struct pcb *pcb;
 413         struct pcb_ext *ext;
 414         struct soft_segment_descriptor ssd = {
 415                 0,                      /* segment base address (overwritten) */
 416                 0,                      /* length (overwritten) */
 417                 SDT_SYS386TSS,          /* segment type */
 418                 0,                      /* priority level */
 419                 1,                      /* descriptor present */
 420                 0, 0,
 421                 0,                      /* default 16 size */
 422                 0                       /* granularity */
 423         };
 424
 425         /*
 426          * this should be a compile time error, but cpp doesn't grok sizeof().
 427          */
 428         if (sizeof(struct vm86_layout) > ctob(3))
 429                 panic("struct vm86_layout exceeds space allocated in locore.s");
 430
 431         /*
 432          * Below is the memory layout that we use for the vm86 region.
 433          *
 434          * +--------+
 435          * |        |
 436          * |        |
 437          * | page 0 |
 438          * |        | +--------+
 439          * |        | | stack  |
 440          * +--------+ +--------+ <--------- vm86paddr
 441          * |        | |Page Tbl| 1M + 64K = 272 entries = 1088 bytes
 442          * |        | +--------+
 443          * |        | |  PCB   | size: ~240 bytes
 444          * | page 1 | |PCB Ext | size: ~140 bytes (includes TSS)
 445          * |        | +--------+
 446          * |        | |int map |
 447          * |        | +--------+
 448          * +--------+ |        |
 449          * | page 2 | |  I/O   |
 450          * +--------+ | bitmap |
 451          * | page 3 | |        |
 452          * |        | +--------+
 453          * +--------+
 454          */
 455
 456         /*
 457          * A rudimentary PCB must be installed, in order to get to the
 458          * PCB extension area.  We use the PCB area as a scratchpad for
 459          * data storage, the layout of which is shown below.
 460          *
 461          * pcb_esi      = new PTD entry 0
 462          * pcb_ebp      = pointer to frame on vm86 stack
 463          * pcb_esp      =    stack frame pointer at time of switch
 464          * pcb_ebx      = va of vm86 page table
 465          * pcb_eip      =    argument pointer to initial call
 466          * pcb_spare[0] =    saved TSS descriptor, word 0
 467          * pcb_space[1] =    saved TSS descriptor, word 1
 468          */
 469 #define new_ptd         pcb_esi
 470 #define vm86_frame      pcb_ebp
 471 #define pgtable_va      pcb_ebx
 472
 473         pcb = &vml->vml_pcb;
 474         ext = &vml->vml_ext;
 475
 476         bzero(pcb, sizeof(struct pcb));
 477         pcb->new_ptd = vm86pa | PG_V | PG_RW | PG_U;
 478         pcb->vm86_frame = (pt_entry_t)vm86paddr - sizeof(struct vm86frame);
 479         pcb->pgtable_va = (vm_offset_t)vm86paddr;
 480         pcb->pcb_ext = ext;
 481
 482         bzero(ext, sizeof(struct pcb_ext));
 483         ext->ext_tss.tss_esp0 = (vm_offset_t)vm86paddr;
 484         ext->ext_tss.tss_ss0 = GSEL(GDATA_SEL, SEL_KPL);
 485         ext->ext_tss.tss_ioopt =
 486                 ((u_int)vml->vml_iomap - (u_int)&ext->ext_tss) << 16;
 487         ext->ext_iomap = vml->vml_iomap;
 488         ext->ext_vm86.vm86_intmap = vml->vml_intmap;
 489
 490         if (cpu_feature & CPUID_VME)
 491                 ext->ext_vm86.vm86_has_vme = (rcr4() & CR4_VME ? 1 : 0);
 492
 493         addr = (u_int *)ext->ext_vm86.vm86_intmap;
 494         for (i = 0; i < (INTMAP_SIZE + IOMAP_SIZE) / sizeof(u_int); i++)
 495                 *addr++ = 0;
 496         vml->vml_iomap_trailer = 0xff;
 497
 498         ssd.ssd_base = (u_int)&ext->ext_tss;
 499         ssd.ssd_limit = TSS_SIZE - 1;
 500         ssdtosd(&ssd, &ext->ext_tssd);
 501
 502         vm86pcb = pcb;
 503
 504 #if 0
 505         /*
 506          * use whatever is leftover of the vm86 page layout as a
 507          * message buffer so we can capture early output.
 508          */
 509         msgbufinit((vm_offset_t)vm86paddr + sizeof(struct vm86_layout),
 510             ctob(3) - sizeof(struct vm86_layout));
 511 #endif
 512 }
 513
 514 vm_offset_t
 515 vm86_getpage(struct vm86context *vmc, int pagenum)
 516 {
 517         int i;
 518
 519         for (i = 0; i < vmc->npages; i++)
 520                 if (vmc->pmap[i].pte_num == pagenum)
 521                         return (vmc->pmap[i].kva);
 522         return (0);
 523 }
 524
 525 vm_offset_t
 526 vm86_addpage(struct vm86context *vmc, int pagenum, vm_offset_t kva)
 527 {
 528         int i, flags = 0;
 529
 530         for (i = 0; i < vmc->npages; i++)
 531                 if (vmc->pmap[i].pte_num == pagenum)
 532                         goto bad;
 533
 534         if (vmc->npages == VM86_PMAPSIZE)
 535                 goto bad;                       /* XXX grow map? */
 536
 537         if (kva == 0) {
 538                 kva = (vm_offset_t)kmalloc(PAGE_SIZE, M_TEMP, M_WAITOK);
 539                 flags = VMAP_MALLOC;
 540         }
 541
 542         i = vmc->npages++;
 543         vmc->pmap[i].flags = flags;
 544         vmc->pmap[i].kva = kva;
 545         vmc->pmap[i].pte_num = pagenum;
 546         return (kva);
 547 bad:
 548         panic("vm86_addpage: not enough room, or overlap");
 549 }
 550
 551 static void
 552 vm86_initflags(struct vm86frame *vmf)
 553 {
 554         int eflags = vmf->vmf_eflags;
 555         struct vm86_kernel *vm86 = &curthread->td_pcb->pcb_ext->ext_vm86;
 556
 557         if (vm86->vm86_has_vme) {
 558                 eflags = (vmf->vmf_eflags & ~VME_USERCHANGE) |
 559                     (eflags & VME_USERCHANGE) | PSL_VM;
 560         } else {
 561                 vm86->vm86_eflags = eflags;     /* save VIF, VIP */
 562                 eflags = (vmf->vmf_eflags & ~VM_USERCHANGE) |
 563                     (eflags & VM_USERCHANGE) | PSL_VM;
 564         }
 565         vmf->vmf_eflags = eflags | PSL_VM;
 566 }
 567
 568 /*
 569  * called from vm86_bioscall, while in vm86 address space, to finalize setup.
 570  */
 571 void
 572 vm86_prepcall(struct vm86frame *vmf)
 573 {
 574         uintptr_t addr[] = { 0xA00, 0x1000 };   /* code, stack */
 575         u_char intcall[] = {
 576                 CLI, INTn, 0x00, STI, HLT
 577         };
 578
 579         if ((vmf->vmf_trapno & PAGE_MASK) <= 0xff) {
 580                 /* interrupt call requested */
 581                 intcall[2] = (u_char)(vmf->vmf_trapno & 0xff);
 582                 memcpy((void *)addr[0], (void *)intcall, sizeof(intcall));
 583                 vmf->vmf_ip = addr[0];
 584                 vmf->vmf_cs = 0;
 585         }
 586         vmf->vmf_sp = addr[1] - 2;              /* keep aligned */
 587         vmf->kernel_fs = vmf->kernel_es = vmf->kernel_ds = vmf->kernel_gs = 0;
 588         vmf->vmf_ss = 0;
 589         vmf->vmf_eflags = PSL_VIF | PSL_VM | PSL_USER;
 590         vm86_initflags(vmf);
 591 }
 592
 593 /*
 594  * vm86 trap handler; determines whether routine succeeded or not.
 595  * Called while in vm86 space, returns to calling process.
 596  *
 597  * A MP lock ref is held on entry from trap() and must be released prior
 598  * to returning to the VM86 call.
 599  */
 600 void
 601 vm86_trap(struct vm86frame *vmf, int have_mplock)
 602 {
 603         caddr_t addr;
 604
 605         /* "should not happen" */
 606         if ((vmf->vmf_eflags & PSL_VM) == 0)
 607                 panic("vm86_trap called, but not in vm86 mode");
 608
 609         addr = MAKE_ADDR(vmf->vmf_cs, vmf->vmf_ip);
 610         if (*(u_char *)addr == HLT)
 611                 vmf->vmf_trapno = vmf->vmf_eflags & PSL_C;
 612         else
 613                 vmf->vmf_trapno = vmf->vmf_trapno << 16;
 614
 615         if (have_mplock)
 616                 rel_mplock();
 617         vm86_biosret(vmf);
 618 }
 619
 620 int
 621 vm86_intcall(int intnum, struct vm86frame *vmf)
 622 {
 623         int error;
 624
 625         if (intnum < 0 || intnum > 0xff)
 626                 return (EINVAL);
 627
 628         crit_enter();
 629         ASSERT_MP_LOCK_HELD(curthread);
 630
 631         vm86_setup_timer_fault();
 632         vmf->vmf_trapno = intnum;
 633         error = vm86_bioscall(vmf);
 634
 635         /*
 636          * Yes, this happens, especially with video BIOS calls.  The BIOS
 637          * will sometimes eat timer 2 for lunch, and we need timer 2.
 638          */
 639         if (vm86_blew_up_timer) {
 640                 vm86_blew_up_timer = 0;
 641                 timer_restore();
 642                 if (timer_warn) {
 643                         kprintf("Warning: BIOS played with the 8254, "
 644                                 "resetting it\n");
 645                 }
 646         }
 647         crit_exit();
 648         return(error);
 649 }
 650
 651 /*
 652  * struct vm86context contains the page table to use when making
 653  * vm86 calls.  If intnum is a valid interrupt number (0-255), then
 654  * the "interrupt trampoline" will be used, otherwise we use the
 655  * caller's cs:ip routine.
 656  */
 657 int
 658 vm86_datacall(int intnum, struct vm86frame *vmf, 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(struct vm86context *vmc, u_short sel, u_short off)
 687 {
 688         int i, page;
 689         vm_offset_t addr;
 690
 691         addr = (vm_offset_t)MAKE_ADDR(sel, off);
 692         page = addr >> PAGE_SHIFT;
 693         for (i = 0; i < vmc->npages; i++)
 694                 if (page == vmc->pmap[i].pte_num)
 695                         return (vmc->pmap[i].kva + (addr & PAGE_MASK));
 696         return (0);
 697 }
 698
 699 int
 700 vm86_getptr(struct vm86context *vmc, vm_offset_t kva, u_short *sel,
 701             u_short *off)
 702 {
 703         int i;
 704
 705         for (i = 0; i < vmc->npages; i++)
 706                 if (kva >= vmc->pmap[i].kva &&
 707                     kva < vmc->pmap[i].kva + PAGE_SIZE) {
 708                         *off = kva - vmc->pmap[i].kva;
 709                         *sel = vmc->pmap[i].pte_num << 8;
 710                         return (1);
 711                 }
 712         return (0);
 713         panic("vm86_getptr: address not found");
 714 }
 715
 716 int
 717 vm86_sysarch(struct lwp *lp, char *args)
 718 {
 719         int error = 0;
 720         struct i386_vm86_args ua;
 721         struct vm86_kernel *vm86;
 722
 723         if ((error = copyin(args, &ua, sizeof(struct i386_vm86_args))) != 0)
 724                 return (error);
 725
 726         if (lp->lwp_thread->td_pcb->pcb_ext == 0)
 727                 if ((error = i386_extend_pcb(lp)) != 0)
 728                         return (error);
 729         vm86 = &lp->lwp_thread->td_pcb->pcb_ext->ext_vm86;
 730
 731         switch (ua.sub_op) {
 732         case VM86_INIT: {
 733                 struct vm86_init_args sa;
 734
 735                 if ((error = copyin(ua.sub_args, &sa, sizeof(sa))) != 0)
 736                         return (error);
 737                 if (cpu_feature & CPUID_VME)
 738                         vm86->vm86_has_vme = (rcr4() & CR4_VME ? 1 : 0);
 739                 else
 740                         vm86->vm86_has_vme = 0;
 741                 vm86->vm86_inited = 1;
 742                 vm86->vm86_debug = sa.debug;
 743                 bcopy(&sa.int_map, vm86->vm86_intmap, 32);
 744                 }
 745                 break;
 746
 747 #if 0
 748         case VM86_SET_VME: {
 749                 struct vm86_vme_args sa;
 750
 751                 if ((cpu_feature & CPUID_VME) == 0)
 752                         return (ENODEV);
 753
 754                 if (error = copyin(ua.sub_args, &sa, sizeof(sa)))
 755                         return (error);
 756                 if (sa.state)
 757                         load_cr4(rcr4() | CR4_VME);
 758                 else
 759                         load_cr4(rcr4() & ~CR4_VME);
 760                 }
 761                 break;
 762 #endif
 763
 764         case VM86_GET_VME: {
 765                 struct vm86_vme_args sa;
 766
 767                 sa.state = (rcr4() & CR4_VME ? 1 : 0);
 768                 error = copyout(&sa, ua.sub_args, sizeof(sa));
 769                 }
 770                 break;
 771
 772         case VM86_INTCALL: {
 773                 struct vm86_intcall_args sa;
 774
 775                 if ((error = suser_cred(lp->lwp_proc->p_ucred, 0)))
 776                         return (error);
 777                 if ((error = copyin(ua.sub_args, &sa, sizeof(sa))))
 778                         return (error);
 779                 if ((error = vm86_intcall(sa.intnum, &sa.vmf)))
 780                         return (error);
 781                 error = copyout(&sa, ua.sub_args, sizeof(sa));
 782                 }
 783                 break;
 784
 785         default:
 786                 error = EINVAL;
 787         }
 788         return (error);
 789 }
 790
 791 /*
 792  * Setup the VM86 I/O map to take faults on the timer
 793  */
 794 static void
 795 vm86_setup_timer_fault(void)
 796 {
 797         struct vm86_layout *vml = (struct vm86_layout *)vm86paddr;
 798
 799         vml->vml_iomap[TIMER_MODE >> 3] |= 1 << (TIMER_MODE & 7);
 800         vml->vml_iomap[TIMER_CNTR0 >> 3] |= 1 << (TIMER_CNTR0 & 7);
 801         vml->vml_iomap[TIMER_CNTR1 >> 3] |= 1 << (TIMER_CNTR1 & 7);
 802         vml->vml_iomap[TIMER_CNTR2 >> 3] |= 1 << (TIMER_CNTR2 & 7);
 803 }
 804
 805 /*
 806  * Setup the VM86 I/O map to not fault on the timer
 807  */
 808 static void
 809 vm86_clear_timer_fault(void)
 810 {
 811         struct vm86_layout *vml = (struct vm86_layout *)vm86paddr;
 812
 813         vml->vml_iomap[TIMER_MODE >> 3] &= ~(1 << (TIMER_MODE & 7));
 814         vml->vml_iomap[TIMER_CNTR0 >> 3] &= ~(1 << (TIMER_CNTR0 & 7));
 815         vml->vml_iomap[TIMER_CNTR1 >> 3] &= ~(1 << (TIMER_CNTR1 & 7));
 816         vml->vml_iomap[TIMER_CNTR2 >> 3] &= ~(1 << (TIMER_CNTR2 & 7));
 817 }
 818