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 1.12 2004/11/20 20:25:05 dillon Exp $
  29  */
  30
  31 #include <sys/param.h>
  32 #include <sys/systm.h>
  33 #include <sys/proc.h>
  34 #include <sys/lock.h>
  35 #include <sys/malloc.h>
  36
  37 #include <vm/vm.h>
  38 #include <vm/pmap.h>
  39 #include <vm/vm_map.h>
  40 #include <vm/vm_page.h>
  41
  42 #include <sys/user.h>
  43 #include <sys/thread2.h>
  44
  45 #include <machine/md_var.h>
  46 #include <machine/pcb_ext.h>    /* pcb.h included via sys/user.h */
  47 #include <machine/psl.h>
  48 #include <machine/specialreg.h>
  49 #include <machine/sysarch.h>
  50 #include <machine/clock.h>
  51 #include <bus/isa/i386/isa.h>
  52 #include <bus/isa/rtc.h>
  53 #include <i386/isa/timerreg.h>
  54
  55 extern int i386_extend_pcb      (struct proc *);
  56 extern int vm86pa;
  57 extern struct pcb *vm86pcb;
  58
  59 extern int vm86_bioscall(struct vm86frame *);
  60 extern void vm86_biosret(struct vm86frame *);
  61
  62 #define PGTABLE_SIZE    ((1024 + 64) * 1024 / PAGE_SIZE)
  63 #define INTMAP_SIZE     32
  64 #define IOMAP_SIZE      ctob(IOPAGES)
  65 #define TSS_SIZE \
  66         (sizeof(struct pcb_ext) - sizeof(struct segment_descriptor) + \
  67          INTMAP_SIZE + IOMAP_SIZE + 1)
  68
  69 struct vm86_layout {
  70         pt_entry_t      vml_pgtbl[PGTABLE_SIZE];
  71         struct  pcb vml_pcb;
  72         struct  pcb_ext vml_ext;
  73         char    vml_intmap[INTMAP_SIZE];
  74         char    vml_iomap[IOMAP_SIZE];
  75         char    vml_iomap_trailer;
  76 };
  77
  78 void vm86_prepcall(struct vm86frame);
  79
  80 struct system_map {
  81         int             type;
  82         vm_offset_t     start;
  83         vm_offset_t     end;
  84 };
  85
  86 #define HLT     0xf4
  87 #define CLI     0xfa
  88 #define STI     0xfb
  89 #define PUSHF   0x9c
  90 #define POPF    0x9d
  91 #define INTn    0xcd
  92 #define IRET    0xcf
  93 #define INB     0xe4
  94 #define INW     0xe5
  95 #define INBDX   0xec
  96 #define INWDX   0xed
  97 #define OUTB    0xe6
  98 #define OUTW    0xe7
  99 #define OUTBDX  0xee
 100 #define OUTWDX  0xef
 101 #define CALLm   0xff
 102 #define OPERAND_SIZE_PREFIX     0x66
 103 #define ADDRESS_SIZE_PREFIX     0x67
 104 #define PUSH_MASK       ~(PSL_VM | PSL_RF | PSL_I)
 105 #define POP_MASK        ~(PSL_VIP | PSL_VIF | PSL_VM | PSL_RF | PSL_IOPL)
 106
 107 static void vm86_setup_timer_fault(void);
 108 static void vm86_clear_timer_fault(void);
 109
 110 static int vm86_blew_up_timer;
 111
 112 static __inline caddr_t
 113 MAKE_ADDR(u_short sel, u_short off)
 114 {
 115         return ((caddr_t)((sel << 4) + off));
 116 }
 117
 118 static __inline void
 119 GET_VEC(u_int vec, u_short *sel, u_short *off)
 120 {
 121         *sel = vec >> 16;
 122         *off = vec & 0xffff;
 123 }
 124
 125 static __inline u_int
 126 MAKE_VEC(u_short sel, u_short off)
 127 {
 128         return ((sel << 16) | off);
 129 }
 130
 131 static __inline void
 132 PUSH(u_short x, struct vm86frame *vmf)
 133 {
 134         vmf->vmf_sp -= 2;
 135         susword(MAKE_ADDR(vmf->vmf_ss, vmf->vmf_sp), x);
 136 }
 137
 138 static __inline void
 139 PUSHL(u_int x, struct vm86frame *vmf)
 140 {
 141         vmf->vmf_sp -= 4;
 142         suword(MAKE_ADDR(vmf->vmf_ss, vmf->vmf_sp), x);
 143 }
 144
 145 static __inline u_short
 146 POP(struct vm86frame *vmf)
 147 {
 148         u_short x = fusword(MAKE_ADDR(vmf->vmf_ss, vmf->vmf_sp));
 149
 150         vmf->vmf_sp += 2;
 151         return (x);
 152 }
 153
 154 static __inline u_int
 155 POPL(struct vm86frame *vmf)
 156 {
 157         u_int x = fuword(MAKE_ADDR(vmf->vmf_ss, vmf->vmf_sp));
 158
 159         vmf->vmf_sp += 4;
 160         return (x);
 161 }
 162
 163 int
 164 vm86_emulate(vmf)
 165         struct vm86frame *vmf;
 166 {
 167         struct vm86_kernel *vm86;
 168         caddr_t addr;
 169         u_char i_byte;
 170         u_int temp_flags;
 171         int inc_ip = 1;
 172         int retcode = 0;
 173
 174         /*
 175          * pcb_ext contains the address of the extension area, or zero if
 176          * the extension is not present.  (This check should not be needed,
 177          * as we can't enter vm86 mode until we set up an extension area)
 178          */
 179         if (curthread->td_pcb->pcb_ext == 0)
 180                 return (SIGBUS);
 181         vm86 = &curthread->td_pcb->pcb_ext->ext_vm86;
 182
 183         if (vmf->vmf_eflags & PSL_T)
 184                 retcode = SIGTRAP;
 185
 186         /*
 187          * Instruction emulation
 188          */
 189         addr = MAKE_ADDR(vmf->vmf_cs, vmf->vmf_ip);
 190         i_byte = fubyte(addr);
 191         if (i_byte == ADDRESS_SIZE_PREFIX) {
 192                 i_byte = fubyte(++addr);
 193                 inc_ip++;
 194         }
 195
 196         /*
 197          * I/O emulation (TIMER only, a big hack).  Just reenable the
 198          * IO bits involved, flag it, and retry the instruction.
 199          */
 200         switch(i_byte) {
 201         case OUTB:
 202         case OUTW:
 203         case OUTBDX:
 204         case OUTWDX:
 205                 vm86_blew_up_timer = 1;
 206                 /* fall through */
 207         case INB:
 208         case INW:
 209         case INBDX:
 210         case INWDX:
 211                 vm86_clear_timer_fault();
 212                 /* retry insn */
 213                 return(0);
 214         }
 215
 216         if (vm86->vm86_has_vme) {
 217                 switch (i_byte) {
 218                 case OPERAND_SIZE_PREFIX:
 219                         i_byte = fubyte(++addr);
 220                         inc_ip++;
 221                         switch (i_byte) {
 222                         case PUSHF:
 223                                 if (vmf->vmf_eflags & PSL_VIF)
 224                                         PUSHL((vmf->vmf_eflags & PUSH_MASK)
 225                                             | PSL_IOPL | PSL_I, vmf);
 226                                 else
 227                                         PUSHL((vmf->vmf_eflags & PUSH_MASK)
 228                                             | PSL_IOPL, vmf);
 229                                 vmf->vmf_ip += inc_ip;
 230                                 return (0);
 231
 232                         case POPF:
 233                                 temp_flags = POPL(vmf) & POP_MASK;
 234                                 vmf->vmf_eflags = (vmf->vmf_eflags & ~POP_MASK)
 235                                     | temp_flags | PSL_VM | PSL_I;
 236                                 vmf->vmf_ip += inc_ip;
 237                                 if (temp_flags & PSL_I) {
 238                                         vmf->vmf_eflags |= PSL_VIF;
 239                                         if (vmf->vmf_eflags & PSL_VIP)
 240                                                 break;
 241                                 } else {
 242                                         vmf->vmf_eflags &= ~PSL_VIF;
 243                                 }
 244                                 return (0);
 245                         }
 246                         break;
 247
 248                 /* VME faults here if VIP is set, but does not set VIF. */
 249                 case STI:
 250                         vmf->vmf_eflags |= PSL_VIF;
 251                         vmf->vmf_ip += inc_ip;
 252                         if ((vmf->vmf_eflags & PSL_VIP) == 0) {
 253                                 uprintf("fatal sti\n");
 254                                 return (SIGKILL);
 255                         }
 256                         break;
 257
 258                 /* VME if no redirection support */
 259                 case INTn:
 260                         break;
 261
 262                 /* VME if trying to set PSL_TF, or PSL_I when VIP is set */
 263                 case POPF:
 264                         temp_flags = POP(vmf) & POP_MASK;
 265                         vmf->vmf_flags = (vmf->vmf_flags & ~POP_MASK)
 266                             | temp_flags | PSL_VM | PSL_I;
 267                         vmf->vmf_ip += inc_ip;
 268                         if (temp_flags & PSL_I) {
 269                                 vmf->vmf_eflags |= PSL_VIF;
 270                                 if (vmf->vmf_eflags & PSL_VIP)
 271                                         break;
 272                         } else {
 273                                 vmf->vmf_eflags &= ~PSL_VIF;
 274                         }
 275                         return (retcode);
 276
 277                 /* VME if trying to set PSL_TF, or PSL_I when VIP is set */
 278                 case IRET:
 279                         vmf->vmf_ip = POP(vmf);
 280                         vmf->vmf_cs = POP(vmf);
 281                         temp_flags = POP(vmf) & POP_MASK;
 282                         vmf->vmf_flags = (vmf->vmf_flags & ~POP_MASK)
 283                             | temp_flags | PSL_VM | PSL_I;
 284                         if (temp_flags & PSL_I) {
 285                                 vmf->vmf_eflags |= PSL_VIF;
 286                                 if (vmf->vmf_eflags & PSL_VIP)
 287                                         break;
 288                         } else {
 289                                 vmf->vmf_eflags &= ~PSL_VIF;
 290                         }
 291                         return (retcode);
 292
 293                 }
 294                 return (SIGBUS);
 295         }
 296
 297         switch (i_byte) {
 298         case OPERAND_SIZE_PREFIX:
 299                 i_byte = fubyte(++addr);
 300                 inc_ip++;
 301                 switch (i_byte) {
 302                 case PUSHF:
 303                         if (vm86->vm86_eflags & PSL_VIF)
 304                                 PUSHL((vmf->vmf_flags & PUSH_MASK)
 305                                     | PSL_IOPL | PSL_I, vmf);
 306                         else
 307                                 PUSHL((vmf->vmf_flags & PUSH_MASK)
 308                                     | PSL_IOPL, vmf);
 309                         vmf->vmf_ip += inc_ip;
 310                         return (retcode);
 311
 312                 case POPF:
 313                         temp_flags = POPL(vmf) & POP_MASK;
 314                         vmf->vmf_eflags = (vmf->vmf_eflags & ~POP_MASK)
 315                             | temp_flags | PSL_VM | PSL_I;
 316                         vmf->vmf_ip += inc_ip;
 317                         if (temp_flags & PSL_I) {
 318                                 vm86->vm86_eflags |= PSL_VIF;
 319                                 if (vm86->vm86_eflags & PSL_VIP)
 320                                         break;
 321                         } else {
 322                                 vm86->vm86_eflags &= ~PSL_VIF;
 323                         }
 324                         return (retcode);
 325                 }
 326                 return (SIGBUS);
 327
 328         case CLI:
 329                 vm86->vm86_eflags &= ~PSL_VIF;
 330                 vmf->vmf_ip += inc_ip;
 331                 return (retcode);
 332
 333         case STI:
 334                 /* if there is a pending interrupt, go to the emulator */
 335                 vm86->vm86_eflags |= PSL_VIF;
 336                 vmf->vmf_ip += inc_ip;
 337                 if (vm86->vm86_eflags & PSL_VIP)
 338                         break;
 339                 return (retcode);
 340
 341         case PUSHF:
 342                 if (vm86->vm86_eflags & PSL_VIF)
 343                         PUSH((vmf->vmf_flags & PUSH_MASK)
 344                             | PSL_IOPL | PSL_I, vmf);
 345                 else
 346                         PUSH((vmf->vmf_flags & PUSH_MASK) | PSL_IOPL, vmf);
 347                 vmf->vmf_ip += inc_ip;
 348                 return (retcode);
 349
 350         case INTn:
 351                 i_byte = fubyte(addr + 1);
 352                 if ((vm86->vm86_intmap[i_byte >> 3] & (1 << (i_byte & 7))) != 0)
 353                         break;
 354                 if (vm86->vm86_eflags & PSL_VIF)
 355                         PUSH((vmf->vmf_flags & PUSH_MASK)
 356                             | PSL_IOPL | PSL_I, vmf);
 357                 else
 358                         PUSH((vmf->vmf_flags & PUSH_MASK) | PSL_IOPL, vmf);
 359                 PUSH(vmf->vmf_cs, vmf);
 360                 PUSH(vmf->vmf_ip + inc_ip + 1, vmf);    /* increment IP */
 361                 GET_VEC(fuword((caddr_t)(i_byte * 4)),
 362                      &vmf->vmf_cs, &vmf->vmf_ip);
 363                 vmf->vmf_flags &= ~PSL_T;
 364                 vm86->vm86_eflags &= ~PSL_VIF;
 365                 return (retcode);
 366
 367         case IRET:
 368                 vmf->vmf_ip = POP(vmf);
 369                 vmf->vmf_cs = POP(vmf);
 370                 temp_flags = POP(vmf) & POP_MASK;
 371                 vmf->vmf_flags = (vmf->vmf_flags & ~POP_MASK)
 372                     | temp_flags | PSL_VM | PSL_I;
 373                 if (temp_flags & PSL_I) {
 374                         vm86->vm86_eflags |= PSL_VIF;
 375                         if (vm86->vm86_eflags & PSL_VIP)
 376                                 break;
 377                 } else {
 378                         vm86->vm86_eflags &= ~PSL_VIF;
 379                 }
 380                 return (retcode);
 381
 382         case POPF:
 383                 temp_flags = POP(vmf) & POP_MASK;
 384                 vmf->vmf_flags = (vmf->vmf_flags & ~POP_MASK)
 385                     | temp_flags | PSL_VM | PSL_I;
 386                 vmf->vmf_ip += inc_ip;
 387                 if (temp_flags & PSL_I) {
 388                         vm86->vm86_eflags |= PSL_VIF;
 389                         if (vm86->vm86_eflags & PSL_VIP)
 390                                 break;
 391                 } else {
 392                         vm86->vm86_eflags &= ~PSL_VIF;
 393                 }
 394                 return (retcode);
 395         }
 396         return (SIGBUS);
 397 }
 398
 399 void
 400 vm86_initialize(void)
 401 {
 402         int i;
 403         u_int *addr;
 404         struct vm86_layout *vml = (struct vm86_layout *)vm86paddr;
 405         struct pcb *pcb;
 406         struct pcb_ext *ext;
 407         struct soft_segment_descriptor ssd = {
 408                 0,                      /* segment base address (overwritten) */
 409                 0,                      /* length (overwritten) */
 410                 SDT_SYS386TSS,          /* segment type */
 411                 0,                      /* priority level */
 412                 1,                      /* descriptor present */
 413                 0, 0,
 414                 0,                      /* default 16 size */
 415                 0                       /* granularity */
 416         };
 417
 418         /*
 419          * this should be a compile time error, but cpp doesn't grok sizeof().
 420          */
 421         if (sizeof(struct vm86_layout) > ctob(3))
 422                 panic("struct vm86_layout exceeds space allocated in locore.s");
 423
 424         /*
 425          * Below is the memory layout that we use for the vm86 region.
 426          *
 427          * +--------+
 428          * |        |
 429          * |        |
 430          * | page 0 |
 431          * |        | +--------+
 432          * |        | | stack  |
 433          * +--------+ +--------+ <--------- vm86paddr
 434          * |        | |Page Tbl| 1M + 64K = 272 entries = 1088 bytes
 435          * |        | +--------+
 436          * |        | |  PCB   | size: ~240 bytes
 437          * | page 1 | |PCB Ext | size: ~140 bytes (includes TSS)
 438          * |        | +--------+
 439          * |        | |int map |
 440          * |        | +--------+
 441          * +--------+ |        |
 442          * | page 2 | |  I/O   |
 443          * +--------+ | bitmap |
 444          * | page 3 | |        |
 445          * |        | +--------+
 446          * +--------+
 447          */
 448
 449         /*
 450          * A rudimentary PCB must be installed, in order to get to the
 451          * PCB extension area.  We use the PCB area as a scratchpad for
 452          * data storage, the layout of which is shown below.
 453          *
 454          * pcb_esi      = new PTD entry 0
 455          * pcb_ebp      = pointer to frame on vm86 stack
 456          * pcb_esp      =    stack frame pointer at time of switch
 457          * pcb_ebx      = va of vm86 page table
 458          * pcb_eip      =    argument pointer to initial call
 459          * pcb_spare[0] =    saved TSS descriptor, word 0
 460          * pcb_space[1] =    saved TSS descriptor, word 1
 461          */
 462 #define new_ptd         pcb_esi
 463 #define vm86_frame      pcb_ebp
 464 #define pgtable_va      pcb_ebx
 465
 466         pcb = &vml->vml_pcb;
 467         ext = &vml->vml_ext;
 468
 469         bzero(pcb, sizeof(struct pcb));
 470         pcb->new_ptd = vm86pa | PG_V | PG_RW | PG_U;
 471         pcb->vm86_frame = (pt_entry_t)vm86paddr - sizeof(struct vm86frame);
 472         pcb->pgtable_va = (vm_offset_t)vm86paddr;
 473         pcb->pcb_ext = ext;
 474
 475         bzero(ext, sizeof(struct pcb_ext));
 476         ext->ext_tss.tss_esp0 = (vm_offset_t)vm86paddr;
 477         ext->ext_tss.tss_ss0 = GSEL(GDATA_SEL, SEL_KPL);
 478         ext->ext_tss.tss_ioopt =
 479                 ((u_int)vml->vml_iomap - (u_int)&ext->ext_tss) << 16;
 480         ext->ext_iomap = vml->vml_iomap;
 481         ext->ext_vm86.vm86_intmap = vml->vml_intmap;
 482
 483         if (cpu_feature & CPUID_VME)
 484                 ext->ext_vm86.vm86_has_vme = (rcr4() & CR4_VME ? 1 : 0);
 485
 486         addr = (u_int *)ext->ext_vm86.vm86_intmap;
 487         for (i = 0; i < (INTMAP_SIZE + IOMAP_SIZE) / sizeof(u_int); i++)
 488                 *addr++ = 0;
 489         vml->vml_iomap_trailer = 0xff;
 490
 491         ssd.ssd_base = (u_int)&ext->ext_tss;
 492         ssd.ssd_limit = TSS_SIZE - 1;
 493         ssdtosd(&ssd, &ext->ext_tssd);
 494
 495         vm86pcb = pcb;
 496
 497 #if 0
 498         /*
 499          * use whatever is leftover of the vm86 page layout as a
 500          * message buffer so we can capture early output.
 501          */
 502         msgbufinit((vm_offset_t)vm86paddr + sizeof(struct vm86_layout),
 503             ctob(3) - sizeof(struct vm86_layout));
 504 #endif
 505 }
 506
 507 vm_offset_t
 508 vm86_getpage(struct vm86context *vmc, int pagenum)
 509 {
 510         int i;
 511
 512         for (i = 0; i < vmc->npages; i++)
 513                 if (vmc->pmap[i].pte_num == pagenum)
 514                         return (vmc->pmap[i].kva);
 515         return (0);
 516 }
 517
 518 vm_offset_t
 519 vm86_addpage(struct vm86context *vmc, int pagenum, vm_offset_t kva)
 520 {
 521         int i, flags = 0;
 522
 523         for (i = 0; i < vmc->npages; i++)
 524                 if (vmc->pmap[i].pte_num == pagenum)
 525                         goto bad;
 526
 527         if (vmc->npages == VM86_PMAPSIZE)
 528                 goto bad;                       /* XXX grow map? */
 529
 530         if (kva == 0) {
 531                 kva = (vm_offset_t)malloc(PAGE_SIZE, M_TEMP, M_WAITOK);
 532                 flags = VMAP_MALLOC;
 533         }
 534
 535         i = vmc->npages++;
 536         vmc->pmap[i].flags = flags;
 537         vmc->pmap[i].kva = kva;
 538         vmc->pmap[i].pte_num = pagenum;
 539         return (kva);
 540 bad:
 541         panic("vm86_addpage: not enough room, or overlap");
 542 }
 543
 544 static void
 545 vm86_initflags(struct vm86frame *vmf)
 546 {
 547         int eflags = vmf->vmf_eflags;
 548         struct vm86_kernel *vm86 = &curthread->td_pcb->pcb_ext->ext_vm86;
 549
 550         if (vm86->vm86_has_vme) {
 551                 eflags = (vmf->vmf_eflags & ~VME_USERCHANGE) |
 552                     (eflags & VME_USERCHANGE) | PSL_VM;
 553         } else {
 554                 vm86->vm86_eflags = eflags;     /* save VIF, VIP */
 555                 eflags = (vmf->vmf_eflags & ~VM_USERCHANGE) |
 556                     (eflags & VM_USERCHANGE) | PSL_VM;
 557         }
 558         vmf->vmf_eflags = eflags | PSL_VM;
 559 }
 560
 561 /*
 562  * called from vm86_bioscall, while in vm86 address space, to finalize setup.
 563  */
 564 void
 565 vm86_prepcall(struct vm86frame vmf)
 566 {
 567         uintptr_t addr[] = { 0xA00, 0x1000 };   /* code, stack */
 568         u_char intcall[] = {
 569                 CLI, INTn, 0x00, STI, HLT
 570         };
 571
 572         if ((vmf.vmf_trapno & PAGE_MASK) <= 0xff) {
 573                 /* interrupt call requested */
 574                 intcall[2] = (u_char)(vmf.vmf_trapno & 0xff);
 575                 memcpy((void *)addr[0], (void *)intcall, sizeof(intcall));
 576                 vmf.vmf_ip = addr[0];
 577                 vmf.vmf_cs = 0;
 578         }
 579         vmf.vmf_sp = addr[1] - 2;              /* keep aligned */
 580         vmf.kernel_fs = vmf.kernel_es = vmf.kernel_ds = 0;
 581         vmf.vmf_ss = 0;
 582         vmf.vmf_eflags = PSL_VIF | PSL_VM | PSL_USER;
 583         vm86_initflags(&vmf);
 584 }
 585
 586 /*
 587  * vm86 trap handler; determines whether routine succeeded or not.
 588  * Called while in vm86 space, returns to calling process.
 589  *
 590  * A MP lock ref is held on entry from trap() and must be released prior
 591  * to returning to the VM86 call.
 592  */
 593 void
 594 vm86_trap(struct vm86frame *vmf)
 595 {
 596         caddr_t addr;
 597
 598         /* "should not happen" */
 599         if ((vmf->vmf_eflags & PSL_VM) == 0)
 600                 panic("vm86_trap called, but not in vm86 mode");
 601
 602         addr = MAKE_ADDR(vmf->vmf_cs, vmf->vmf_ip);
 603         if (*(u_char *)addr == HLT)
 604                 vmf->vmf_trapno = vmf->vmf_eflags & PSL_C;
 605         else
 606                 vmf->vmf_trapno = vmf->vmf_trapno << 16;
 607
 608         rel_mplock();
 609         vm86_biosret(vmf);
 610 }
 611
 612 int
 613 vm86_intcall(int intnum, struct vm86frame *vmf)
 614 {
 615         int error;
 616
 617         if (intnum < 0 || intnum > 0xff)
 618                 return (EINVAL);
 619
 620         crit_enter();
 621         ASSERT_MP_LOCK_HELD();
 622
 623         vm86_setup_timer_fault();
 624         vmf->vmf_trapno = intnum;
 625         error = vm86_bioscall(vmf);
 626
 627         /*
 628          * Yes, this happens, especially with video BIOS calls.  The BIOS
 629          * will sometimes eat timer 2 for lunch, and we need timer 2.
 630          */
 631         if (vm86_blew_up_timer) {
 632                 vm86_blew_up_timer = 0;
 633                 timer_restore();
 634                 printf("Warning: BIOS messed around with the 8254, resetting it\n");
 635         }
 636         crit_exit();
 637         return(error);
 638 }
 639
 640 /*
 641  * struct vm86context contains the page table to use when making
 642  * vm86 calls.  If intnum is a valid interrupt number (0-255), then
 643  * the "interrupt trampoline" will be used, otherwise we use the
 644  * caller's cs:ip routine.
 645  */
 646 int
 647 vm86_datacall(intnum, vmf, vmc)
 648         int intnum;
 649         struct vm86frame *vmf;
 650         struct vm86context *vmc;
 651 {
 652         pt_entry_t *pte = vm86paddr;
 653         u_int page;
 654         int i, entry, retval;
 655
 656         crit_enter();
 657         ASSERT_MP_LOCK_HELD();
 658
 659         for (i = 0; i < vmc->npages; i++) {
 660                 page = vtophys(vmc->pmap[i].kva & PG_FRAME);
 661                 entry = vmc->pmap[i].pte_num;
 662                 vmc->pmap[i].old_pte = pte[entry];
 663                 pte[entry] = page | PG_V | PG_RW | PG_U;
 664         }
 665
 666         vmf->vmf_trapno = intnum;
 667         retval = vm86_bioscall(vmf);
 668
 669         for (i = 0; i < vmc->npages; i++) {
 670                 entry = vmc->pmap[i].pte_num;
 671                 pte[entry] = vmc->pmap[i].old_pte;
 672         }
 673         crit_exit();
 674         return (retval);
 675 }
 676
 677 vm_offset_t
 678 vm86_getaddr(vmc, sel, off)
 679         struct vm86context *vmc;
 680         u_short sel;
 681         u_short off;
 682 {
 683         int i, page;
 684         vm_offset_t addr;
 685
 686         addr = (vm_offset_t)MAKE_ADDR(sel, off);
 687         page = addr >> PAGE_SHIFT;
 688         for (i = 0; i < vmc->npages; i++)
 689                 if (page == vmc->pmap[i].pte_num)
 690                         return (vmc->pmap[i].kva + (addr & PAGE_MASK));
 691         return (0);
 692 }
 693
 694 int
 695 vm86_getptr(vmc, kva, sel, off)
 696         struct vm86context *vmc;
 697         vm_offset_t kva;
 698         u_short *sel;
 699         u_short *off;
 700 {
 701         int i;
 702
 703         for (i = 0; i < vmc->npages; i++)
 704                 if (kva >= vmc->pmap[i].kva &&
 705                     kva < vmc->pmap[i].kva + PAGE_SIZE) {
 706                         *off = kva - vmc->pmap[i].kva;
 707                         *sel = vmc->pmap[i].pte_num << 8;
 708                         return (1);
 709                 }
 710         return (0);
 711         panic("vm86_getptr: address not found");
 712 }
 713
 714 int
 715 vm86_sysarch(struct proc *p, char *args)
 716 {
 717         int error = 0;
 718         struct i386_vm86_args ua;
 719         struct vm86_kernel *vm86;
 720
 721         if ((error = copyin(args, &ua, sizeof(struct i386_vm86_args))) != 0)
 722                 return (error);
 723
 724         if (p->p_thread->td_pcb->pcb_ext == 0)
 725                 if ((error = i386_extend_pcb(p)) != 0)
 726                         return (error);
 727         vm86 = &p->p_thread->td_pcb->pcb_ext->ext_vm86;
 728
 729         switch (ua.sub_op) {
 730         case VM86_INIT: {
 731                 struct vm86_init_args sa;
 732
 733                 if ((error = copyin(ua.sub_args, &sa, sizeof(sa))) != 0)
 734                         return (error);
 735                 if (cpu_feature & CPUID_VME)
 736                         vm86->vm86_has_vme = (rcr4() & CR4_VME ? 1 : 0);
 737                 else
 738                         vm86->vm86_has_vme = 0;
 739                 vm86->vm86_inited = 1;
 740                 vm86->vm86_debug = sa.debug;
 741                 bcopy(&sa.int_map, vm86->vm86_intmap, 32);
 742                 }
 743                 break;
 744
 745 #if 0
 746         case VM86_SET_VME: {
 747                 struct vm86_vme_args sa;
 748
 749                 if ((cpu_feature & CPUID_VME) == 0)
 750                         return (ENODEV);
 751
 752                 if (error = copyin(ua.sub_args, &sa, sizeof(sa)))
 753                         return (error);
 754                 if (sa.state)
 755                         load_cr4(rcr4() | CR4_VME);
 756                 else
 757                         load_cr4(rcr4() & ~CR4_VME);
 758                 }
 759                 break;
 760 #endif
 761
 762         case VM86_GET_VME: {
 763                 struct vm86_vme_args sa;
 764
 765                 sa.state = (rcr4() & CR4_VME ? 1 : 0);
 766                 error = copyout(&sa, ua.sub_args, sizeof(sa));
 767                 }
 768                 break;
 769
 770         case VM86_INTCALL: {
 771                 struct vm86_intcall_args sa;
 772
 773                 if ((error = suser_cred(p->p_ucred, 0)))
 774                         return (error);
 775                 if ((error = copyin(ua.sub_args, &sa, sizeof(sa))))
 776                         return (error);
 777                 if ((error = vm86_intcall(sa.intnum, &sa.vmf)))
 778                         return (error);
 779                 error = copyout(&sa, ua.sub_args, sizeof(sa));
 780                 }
 781                 break;
 782
 783         default:
 784                 error = EINVAL;
 785         }
 786         return (error);
 787 }
 788
 789 /*
 790  * Setup the VM86 I/O map to take faults on the timer
 791  */
 792 static void
 793 vm86_setup_timer_fault(void)
 794 {
 795         struct vm86_layout *vml = (struct vm86_layout *)vm86paddr;
 796
 797         vml->vml_iomap[TIMER_MODE >> 3] |= 1 << (TIMER_MODE & 7);
 798         vml->vml_iomap[TIMER_CNTR0 >> 3] |= 1 << (TIMER_CNTR0 & 7);
 799         vml->vml_iomap[TIMER_CNTR1 >> 3] |= 1 << (TIMER_CNTR1 & 7);
 800         vml->vml_iomap[TIMER_CNTR2 >> 3] |= 1 << (TIMER_CNTR2 & 7);
 801 }
 802
 803 /*
 804  * Setup the VM86 I/O map to not fault on the timer
 805  */
 806 static void
 807 vm86_clear_timer_fault(void)
 808 {
 809         struct vm86_layout *vml = (struct vm86_layout *)vm86paddr;
 810
 811         vml->vml_iomap[TIMER_MODE >> 3] &= ~(1 << (TIMER_MODE & 7));
 812         vml->vml_iomap[TIMER_CNTR0 >> 3] &= ~(1 << (TIMER_CNTR0 & 7));
 813         vml->vml_iomap[TIMER_CNTR1 >> 3] &= ~(1 << (TIMER_CNTR1 & 7));
 814         vml->vml_iomap[TIMER_CNTR2 >> 3] &= ~(1 << (TIMER_CNTR2 & 7));
 815 }
 816