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 1.26 2008/08/02 01:14:43 dillon 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/priv.h>
  36 #include <sys/lock.h>
  37 #include <sys/malloc.h>
  38 #include <sys/sysctl.h>
  39
  40 #include <vm/vm.h>
  41 #include <vm/pmap.h>
  42 #include <vm/vm_map.h>
  43 #include <vm/vm_page.h>
  44
  45 #include <sys/user.h>
  46
  47 #include <sys/thread2.h>
  48 #include <sys/mplock2.h>
  49
  50 #include <machine/md_var.h>
  51 #include <machine/pcb_ext.h>    /* pcb.h included via sys/user.h */
  52 #include <machine/psl.h>
  53 #include <machine/specialreg.h>
  54 #include <machine/sysarch.h>
  55 #include <machine/clock.h>
  56 #include <bus/isa/isa.h>
  57 #include <bus/isa/rtc.h>
  58 #include <machine_base/isa/timerreg.h>
  59
  60 extern int i386_extend_pcb      (struct lwp *);
  61 extern int vm86pa;
  62 extern struct pcb *vm86pcb;
  63
  64 extern int vm86_bioscall(struct vm86frame *);
  65 extern void vm86_biosret(struct vm86frame *);
  66
  67 #define PGTABLE_SIZE    ((1024 + 64) * 1024 / PAGE_SIZE)
  68 #define INTMAP_SIZE     32
  69 #define IOMAP_SIZE      ctob(IOPAGES)
  70 #define TSS_SIZE \
  71         (sizeof(struct pcb_ext) - sizeof(struct segment_descriptor) + \
  72          INTMAP_SIZE + IOMAP_SIZE + 1)
  73
  74 struct vm86_layout {
  75         pt_entry_t      vml_pgtbl[PGTABLE_SIZE];
  76         struct  pcb vml_pcb;
  77         struct  pcb_ext vml_ext;
  78         char    vml_intmap[INTMAP_SIZE];
  79         char    vml_iomap[IOMAP_SIZE];
  80         char    vml_iomap_trailer;
  81 };
  82
  83 void vm86_prepcall(struct vm86frame *);
  84
  85 struct system_map {
  86         int             type;
  87         vm_offset_t     start;
  88         vm_offset_t     end;
  89 };
  90
  91 #define HLT     0xf4
  92 #define CLI     0xfa
  93 #define STI     0xfb
  94 #define PUSHF   0x9c
  95 #define POPF    0x9d
  96 #define INTn    0xcd
  97 #define IRET    0xcf
  98 #define INB     0xe4
  99 #define INW     0xe5
 100 #define INBDX   0xec
 101 #define INWDX   0xed
 102 #define OUTB    0xe6
 103 #define OUTW    0xe7
 104 #define OUTBDX  0xee
 105 #define OUTWDX  0xef
 106 #define CALLm   0xff
 107 #define OPERAND_SIZE_PREFIX     0x66
 108 #define ADDRESS_SIZE_PREFIX     0x67
 109 #define PUSH_MASK       ~(PSL_VM | PSL_RF | PSL_I)
 110 #define POP_MASK        ~(PSL_VIP | PSL_VIF | PSL_VM | PSL_RF | PSL_IOPL)
 111
 112 static void vm86_setup_timer_fault(void);
 113 static void vm86_clear_timer_fault(void);
 114
 115 static int vm86_blew_up_timer;
 116
 117 static int timer_warn = 1;
 118 SYSCTL_INT(_debug, OID_AUTO, timer_warn, CTLFLAG_RW, &timer_warn, 0, "");
 119
 120 static __inline caddr_t
 121 MAKE_ADDR(u_short sel, u_short off)
 122 {
 123         return ((caddr_t)((sel << 4) + off));
 124 }
 125
 126 static __inline void
 127 GET_VEC(u_int vec, u_short *sel, u_short *off)
 128 {
 129         *sel = vec >> 16;
 130         *off = vec & 0xffff;
 131 }
 132
 133 static __inline u_int
 134 MAKE_VEC(u_short sel, u_short off)
 135 {
 136         return ((sel << 16) | off);
 137 }
 138
 139 static __inline void
 140 PUSH(u_short x, struct vm86frame *vmf)
 141 {
 142         vmf->vmf_sp -= 2;
 143         susword(MAKE_ADDR(vmf->vmf_ss, vmf->vmf_sp), x);
 144 }
 145
 146 static __inline void
 147 PUSHL(u_int x, struct vm86frame *vmf)
 148 {
 149         vmf->vmf_sp -= 4;
 150         suword(MAKE_ADDR(vmf->vmf_ss, vmf->vmf_sp), x);
 151 }
 152
 153 static __inline u_short
 154 POP(struct vm86frame *vmf)
 155 {
 156         u_short x = fusword(MAKE_ADDR(vmf->vmf_ss, vmf->vmf_sp));
 157
 158         vmf->vmf_sp += 2;
 159         return (x);
 160 }
 161
 162 static __inline u_int
 163 POPL(struct vm86frame *vmf)
 164 {
 165         u_int x = fuword(MAKE_ADDR(vmf->vmf_ss, vmf->vmf_sp));
 166
 167         vmf->vmf_sp += 4;
 168         return (x);
 169 }
 170
 171 /*
 172  * MPSAFE
 173  */
 174 int
 175 vm86_emulate(struct vm86frame *vmf)
 176 {
 177         struct vm86_kernel *vm86;
 178         caddr_t addr;
 179         u_char i_byte;
 180         u_int temp_flags;
 181         int inc_ip = 1;
 182         int retcode = 0;
 183
 184         /*
 185          * pcb_ext contains the address of the extension area, or zero if
 186          * the extension is not present.  (This check should not be needed,
 187          * as we can't enter vm86 mode until we set up an extension area)
 188          */
 189         if (curthread->td_pcb->pcb_ext == 0)
 190                 return (SIGBUS);
 191         vm86 = &curthread->td_pcb->pcb_ext->ext_vm86;
 192
 193         if (vmf->vmf_eflags & PSL_T)
 194                 retcode = SIGTRAP;
 195
 196         /*
 197          * Instruction emulation
 198          */
 199         addr = MAKE_ADDR(vmf->vmf_cs, vmf->vmf_ip);
 200         i_byte = fubyte(addr);
 201         if (i_byte == ADDRESS_SIZE_PREFIX) {
 202                 i_byte = fubyte(++addr);
 203                 inc_ip++;
 204         }
 205
 206         /*
 207          * I/O emulation (TIMER only, a big hack).  Just reenable the
 208          * IO bits involved, flag it, and retry the instruction.
 209          */
 210         switch(i_byte) {
 211         case OUTB:
 212         case OUTW:
 213         case OUTBDX:
 214         case OUTWDX:
 215                 vm86_blew_up_timer = 1;
 216                 /* fall through */
 217         case INB:
 218         case INW:
 219         case INBDX:
 220         case INWDX:
 221                 vm86_clear_timer_fault();
 222                 /* retry insn */
 223                 return(0);
 224         }
 225
 226         if (vm86->vm86_has_vme) {
 227                 switch (i_byte) {
 228                 case OPERAND_SIZE_PREFIX:
 229                         i_byte = fubyte(++addr);
 230                         inc_ip++;
 231                         switch (i_byte) {
 232                         case PUSHF:
 233                                 if (vmf->vmf_eflags & PSL_VIF)
 234                                         PUSHL((vmf->vmf_eflags & PUSH_MASK)
 235                                             | PSL_IOPL | PSL_I, vmf);
 236                                 else
 237                                         PUSHL((vmf->vmf_eflags & PUSH_MASK)
 238                                             | PSL_IOPL, vmf);
 239                                 vmf->vmf_ip += inc_ip;
 240                                 return (0);
 241
 242                         case POPF:
 243                                 temp_flags = POPL(vmf) & POP_MASK;
 244                                 vmf->vmf_eflags = (vmf->vmf_eflags & ~POP_MASK)
 245                                     | temp_flags | PSL_VM | PSL_I;
 246                                 vmf->vmf_ip += inc_ip;
 247                                 if (temp_flags & PSL_I) {
 248                                         vmf->vmf_eflags |= PSL_VIF;
 249                                         if (vmf->vmf_eflags & PSL_VIP)
 250                                                 break;
 251                                 } else {
 252                                         vmf->vmf_eflags &= ~PSL_VIF;
 253                                 }
 254                                 return (0);
 255                         }
 256                         break;
 257
 258                 /* VME faults here if VIP is set, but does not set VIF. */
 259                 case STI:
 260                         vmf->vmf_eflags |= PSL_VIF;
 261                         vmf->vmf_ip += inc_ip;
 262                         if ((vmf->vmf_eflags & PSL_VIP) == 0) {
 263                                 uprintf("fatal sti\n");
 264                                 return (SIGKILL);
 265                         }
 266                         break;
 267
 268                 /* VME if no redirection support */
 269                 case INTn:
 270                         break;
 271
 272                 /* VME if trying to set PSL_TF, or PSL_I when VIP is set */
 273                 case POPF:
 274                         temp_flags = POP(vmf) & POP_MASK;
 275                         vmf->vmf_flags = (vmf->vmf_flags & ~POP_MASK)
 276                             | temp_flags | PSL_VM | PSL_I;
 277                         vmf->vmf_ip += inc_ip;
 278                         if (temp_flags & PSL_I) {
 279                                 vmf->vmf_eflags |= PSL_VIF;
 280                                 if (vmf->vmf_eflags & PSL_VIP)
 281                                         break;
 282                         } else {
 283                                 vmf->vmf_eflags &= ~PSL_VIF;
 284                         }
 285                         return (retcode);
 286
 287                 /* VME if trying to set PSL_TF, or PSL_I when VIP is set */
 288                 case IRET:
 289                         vmf->vmf_ip = POP(vmf);
 290                         vmf->vmf_cs = POP(vmf);
 291                         temp_flags = POP(vmf) & POP_MASK;
 292                         vmf->vmf_flags = (vmf->vmf_flags & ~POP_MASK)
 293                             | temp_flags | PSL_VM | PSL_I;
 294                         if (temp_flags & PSL_I) {
 295                                 vmf->vmf_eflags |= PSL_VIF;
 296                                 if (vmf->vmf_eflags & PSL_VIP)
 297                                         break;
 298                         } else {
 299                                 vmf->vmf_eflags &= ~PSL_VIF;
 300                         }
 301                         return (retcode);
 302
 303                 }
 304                 return (SIGBUS);
 305         }
 306
 307         switch (i_byte) {
 308         case OPERAND_SIZE_PREFIX:
 309                 i_byte = fubyte(++addr);
 310                 inc_ip++;
 311                 switch (i_byte) {
 312                 case PUSHF:
 313                         if (vm86->vm86_eflags & PSL_VIF)
 314                                 PUSHL((vmf->vmf_flags & PUSH_MASK)
 315                                     | PSL_IOPL | PSL_I, vmf);
 316                         else
 317                                 PUSHL((vmf->vmf_flags & PUSH_MASK)
 318                                     | PSL_IOPL, vmf);
 319                         vmf->vmf_ip += inc_ip;
 320                         return (retcode);
 321
 322                 case POPF:
 323                         temp_flags = POPL(vmf) & POP_MASK;
 324                         vmf->vmf_eflags = (vmf->vmf_eflags & ~POP_MASK)
 325                             | temp_flags | PSL_VM | PSL_I;
 326                         vmf->vmf_ip += inc_ip;
 327                         if (temp_flags & PSL_I) {
 328                                 vm86->vm86_eflags |= PSL_VIF;
 329                                 if (vm86->vm86_eflags & PSL_VIP)
 330                                         break;
 331                         } else {
 332                                 vm86->vm86_eflags &= ~PSL_VIF;
 333                         }
 334                         return (retcode);
 335                 }
 336                 return (SIGBUS);
 337
 338         case CLI:
 339                 vm86->vm86_eflags &= ~PSL_VIF;
 340                 vmf->vmf_ip += inc_ip;
 341                 return (retcode);
 342
 343         case STI:
 344                 /* if there is a pending interrupt, go to the emulator */
 345                 vm86->vm86_eflags |= PSL_VIF;
 346                 vmf->vmf_ip += inc_ip;
 347                 if (vm86->vm86_eflags & PSL_VIP)
 348                         break;
 349                 return (retcode);
 350
 351         case PUSHF:
 352                 if (vm86->vm86_eflags & PSL_VIF)
 353                         PUSH((vmf->vmf_flags & PUSH_MASK)
 354                             | PSL_IOPL | PSL_I, vmf);
 355                 else
 356                         PUSH((vmf->vmf_flags & PUSH_MASK) | PSL_IOPL, vmf);
 357                 vmf->vmf_ip += inc_ip;
 358                 return (retcode);
 359
 360         case INTn:
 361                 i_byte = fubyte(addr + 1);
 362                 if ((vm86->vm86_intmap[i_byte >> 3] & (1 << (i_byte & 7))) != 0)
 363                         break;
 364                 if (vm86->vm86_eflags & PSL_VIF)
 365                         PUSH((vmf->vmf_flags & PUSH_MASK)
 366                             | PSL_IOPL | PSL_I, vmf);
 367                 else
 368                         PUSH((vmf->vmf_flags & PUSH_MASK) | PSL_IOPL, vmf);
 369                 PUSH(vmf->vmf_cs, vmf);
 370                 PUSH(vmf->vmf_ip + inc_ip + 1, vmf);    /* increment IP */
 371                 GET_VEC(fuword((caddr_t)(i_byte * 4)),
 372                      &vmf->vmf_cs, &vmf->vmf_ip);
 373                 vmf->vmf_flags &= ~PSL_T;
 374                 vm86->vm86_eflags &= ~PSL_VIF;
 375                 return (retcode);
 376
 377         case IRET:
 378                 vmf->vmf_ip = POP(vmf);
 379                 vmf->vmf_cs = POP(vmf);
 380                 temp_flags = POP(vmf) & POP_MASK;
 381                 vmf->vmf_flags = (vmf->vmf_flags & ~POP_MASK)
 382                     | temp_flags | PSL_VM | PSL_I;
 383                 if (temp_flags & PSL_I) {
 384                         vm86->vm86_eflags |= PSL_VIF;
 385                         if (vm86->vm86_eflags & PSL_VIP)
 386                                 break;
 387                 } else {
 388                         vm86->vm86_eflags &= ~PSL_VIF;
 389                 }
 390                 return (retcode);
 391
 392         case POPF:
 393                 temp_flags = POP(vmf) & POP_MASK;
 394                 vmf->vmf_flags = (vmf->vmf_flags & ~POP_MASK)
 395                     | temp_flags | PSL_VM | PSL_I;
 396                 vmf->vmf_ip += inc_ip;
 397                 if (temp_flags & PSL_I) {
 398                         vm86->vm86_eflags |= PSL_VIF;
 399                         if (vm86->vm86_eflags & PSL_VIP)
 400                                 break;
 401                 } else {
 402                         vm86->vm86_eflags &= ~PSL_VIF;
 403                 }
 404                 return (retcode);
 405         }
 406         return (SIGBUS);
 407 }
 408
 409 void
 410 vm86_initialize(void)
 411 {
 412         int i;
 413         u_int *addr;
 414         struct vm86_layout *vml = (struct vm86_layout *)vm86paddr;
 415         struct pcb *pcb;
 416         struct pcb_ext *ext;
 417         struct soft_segment_descriptor ssd = {
 418                 0,                      /* segment base address (overwritten) */
 419                 0,                      /* length (overwritten) */
 420                 SDT_SYS386TSS,          /* segment type */
 421                 0,                      /* priority level */
 422                 1,                      /* descriptor present */
 423                 0, 0,
 424                 0,                      /* default 16 size */
 425                 0                       /* granularity */
 426         };
 427
 428         /*
 429          * this should be a compile time error, but cpp doesn't grok sizeof().
 430          */
 431         if (sizeof(struct vm86_layout) > ctob(3))
 432                 panic("struct vm86_layout exceeds space allocated in locore.s");
 433
 434         /*
 435          * Below is the memory layout that we use for the vm86 region.
 436          *
 437          * +--------+
 438          * |        |
 439          * |        |
 440          * | page 0 |
 441          * |        | +--------+
 442          * |        | | stack  |
 443          * +--------+ +--------+ <--------- vm86paddr
 444          * |        | |Page Tbl| 1M + 64K = 272 entries = 1088 bytes
 445          * |        | +--------+
 446          * |        | |  PCB   | size: ~240 bytes
 447          * | page 1 | |PCB Ext | size: ~140 bytes (includes TSS)
 448          * |        | +--------+
 449          * |        | |int map |
 450          * |        | +--------+
 451          * +--------+ |        |
 452          * | page 2 | |  I/O   |
 453          * +--------+ | bitmap |
 454          * | page 3 | |        |
 455          * |        | +--------+
 456          * +--------+
 457          */
 458
 459         /*
 460          * A rudimentary PCB must be installed, in order to get to the
 461          * PCB extension area.  We use the PCB area as a scratchpad for
 462          * data storage, the layout of which is shown below.
 463          *
 464          * pcb_esi      = new PTD entry 0
 465          * pcb_ebp      = pointer to frame on vm86 stack
 466          * pcb_esp      =    stack frame pointer at time of switch
 467          * pcb_ebx      = va of vm86 page table
 468          * pcb_eip      =    argument pointer to initial call
 469          * pcb_spare[0] =    saved TSS descriptor, word 0
 470          * pcb_space[1] =    saved TSS descriptor, word 1
 471          */
 472 #define new_ptd         pcb_esi
 473 #define vm86_frame      pcb_ebp
 474 #define pgtable_va      pcb_ebx
 475
 476         pcb = &vml->vml_pcb;
 477         ext = &vml->vml_ext;
 478
 479         bzero(pcb, sizeof(struct pcb));
 480         pcb->new_ptd = vm86pa | PG_V | PG_RW | PG_U;
 481         pcb->vm86_frame = (pt_entry_t)vm86paddr - sizeof(struct vm86frame);
 482         pcb->pgtable_va = (vm_offset_t)vm86paddr;
 483         pcb->pcb_ext = ext;
 484
 485         bzero(ext, sizeof(struct pcb_ext));
 486         ext->ext_tss.tss_esp0 = (vm_offset_t)vm86paddr;
 487         ext->ext_tss.tss_ss0 = GSEL(GDATA_SEL, SEL_KPL);
 488         ext->ext_tss.tss_ioopt =
 489                 ((u_int)vml->vml_iomap - (u_int)&ext->ext_tss) << 16;
 490         ext->ext_iomap = vml->vml_iomap;
 491         ext->ext_vm86.vm86_intmap = vml->vml_intmap;
 492
 493         if (cpu_feature & CPUID_VME)
 494                 ext->ext_vm86.vm86_has_vme = (rcr4() & CR4_VME ? 1 : 0);
 495
 496         addr = (u_int *)ext->ext_vm86.vm86_intmap;
 497         for (i = 0; i < (INTMAP_SIZE + IOMAP_SIZE) / sizeof(u_int); i++)
 498                 *addr++ = 0;
 499         vml->vml_iomap_trailer = 0xff;
 500
 501         ssd.ssd_base = (u_int)&ext->ext_tss;
 502         ssd.ssd_limit = TSS_SIZE - 1;
 503         ssdtosd(&ssd, &ext->ext_tssd);
 504
 505         vm86pcb = pcb;
 506
 507 #if 0
 508         /*
 509          * use whatever is leftover of the vm86 page layout as a
 510          * message buffer so we can capture early output.
 511          */
 512         msgbufinit((vm_offset_t)vm86paddr + sizeof(struct vm86_layout),
 513             ctob(3) - sizeof(struct vm86_layout));
 514 #endif
 515 }
 516
 517 vm_offset_t
 518 vm86_getpage(struct vm86context *vmc, int pagenum)
 519 {
 520         int i;
 521
 522         for (i = 0; i < vmc->npages; i++)
 523                 if (vmc->pmap[i].pte_num == pagenum)
 524                         return (vmc->pmap[i].kva);
 525         return (0);
 526 }
 527
 528 vm_offset_t
 529 vm86_addpage(struct vm86context *vmc, int pagenum, vm_offset_t kva)
 530 {
 531         int i, flags = 0;
 532
 533         for (i = 0; i < vmc->npages; i++)
 534                 if (vmc->pmap[i].pte_num == pagenum)
 535                         goto bad;
 536
 537         if (vmc->npages == VM86_PMAPSIZE)
 538                 goto bad;                       /* XXX grow map? */
 539
 540         if (kva == 0) {
 541                 kva = (vm_offset_t)kmalloc(PAGE_SIZE, M_TEMP, M_WAITOK);
 542                 flags = VMAP_MALLOC;
 543         }
 544
 545         i = vmc->npages++;
 546         vmc->pmap[i].flags = flags;
 547         vmc->pmap[i].kva = kva;
 548         vmc->pmap[i].pte_num = pagenum;
 549         return (kva);
 550 bad:
 551         panic("vm86_addpage: not enough room, or overlap");
 552 }
 553
 554 static void
 555 vm86_initflags(struct vm86frame *vmf)
 556 {
 557         int eflags = vmf->vmf_eflags;
 558         struct vm86_kernel *vm86 = &curthread->td_pcb->pcb_ext->ext_vm86;
 559
 560         if (vm86->vm86_has_vme) {
 561                 eflags = (vmf->vmf_eflags & ~VME_USERCHANGE) |
 562                     (eflags & VME_USERCHANGE) | PSL_VM;
 563         } else {
 564                 vm86->vm86_eflags = eflags;     /* save VIF, VIP */
 565                 eflags = (vmf->vmf_eflags & ~VM_USERCHANGE) |
 566                     (eflags & VM_USERCHANGE) | PSL_VM;
 567         }
 568         vmf->vmf_eflags = eflags | PSL_VM;
 569 }
 570
 571 /*
 572  * called from vm86_bioscall, while in vm86 address space, to finalize setup.
 573  */
 574 void
 575 vm86_prepcall(struct vm86frame *vmf)
 576 {
 577         uintptr_t addr[] = { 0xA00, 0x1000 };   /* code, stack */
 578         u_char intcall[] = {
 579                 CLI, INTn, 0x00, STI, HLT
 580         };
 581
 582         if ((vmf->vmf_trapno & PAGE_MASK) <= 0xff) {
 583                 /* interrupt call requested */
 584                 intcall[2] = (u_char)(vmf->vmf_trapno & 0xff);
 585                 memcpy((void *)addr[0], (void *)intcall, sizeof(intcall));
 586                 vmf->vmf_ip = addr[0];
 587                 vmf->vmf_cs = 0;
 588         }
 589         vmf->vmf_sp = addr[1] - 2;              /* keep aligned */
 590         vmf->kernel_fs = vmf->kernel_es = vmf->kernel_ds = vmf->kernel_gs = 0;
 591         vmf->vmf_ss = 0;
 592         vmf->vmf_eflags = PSL_VIF | PSL_VM | PSL_USER;
 593         vm86_initflags(vmf);
 594 }
 595
 596 /*
 597  * vm86 trap handler; determines whether routine succeeded or not.
 598  * Called while in vm86 space, returns to calling process.
 599  *
 600  * A MP lock ref is held on entry from trap() and must be released prior
 601  * to returning to the VM86 call.
 602  */
 603 void
 604 vm86_trap(struct vm86frame *vmf, int have_mplock)
 605 {
 606         caddr_t addr;
 607
 608         /* "should not happen" */
 609         if ((vmf->vmf_eflags & PSL_VM) == 0)
 610                 panic("vm86_trap called, but not in vm86 mode");
 611
 612         addr = MAKE_ADDR(vmf->vmf_cs, vmf->vmf_ip);
 613         if (*(u_char *)addr == HLT)
 614                 vmf->vmf_trapno = vmf->vmf_eflags & PSL_C;
 615         else
 616                 vmf->vmf_trapno = vmf->vmf_trapno << 16;
 617
 618         if (have_mplock)
 619                 rel_mplock();
 620         vm86_biosret(vmf);
 621 }
 622
 623 int
 624 vm86_intcall(int intnum, struct vm86frame *vmf)
 625 {
 626         int error;
 627
 628         if (intnum < 0 || intnum > 0xff)
 629                 return (EINVAL);
 630
 631         crit_enter();
 632         ASSERT_MP_LOCK_HELD(curthread);
 633
 634         vm86_setup_timer_fault();
 635         vmf->vmf_trapno = intnum;
 636         error = vm86_bioscall(vmf);
 637
 638         /*
 639          * Yes, this happens, especially with video BIOS calls.  The BIOS
 640          * will sometimes eat timer 2 for lunch, and we need timer 2.
 641          */
 642         if (vm86_blew_up_timer) {
 643                 vm86_blew_up_timer = 0;
 644                 timer_restore();
 645                 if (timer_warn) {
 646                         kprintf("Warning: BIOS played with the 8254, "
 647                                 "resetting it\n");
 648                 }
 649         }
 650         crit_exit();
 651         return(error);
 652 }
 653
 654 /*
 655  * struct vm86context contains the page table to use when making
 656  * vm86 calls.  If intnum is a valid interrupt number (0-255), then
 657  * the "interrupt trampoline" will be used, otherwise we use the
 658  * caller's cs:ip routine.
 659  */
 660 int
 661 vm86_datacall(int intnum, struct vm86frame *vmf, struct vm86context *vmc)
 662 {
 663         pt_entry_t *pte = vm86paddr;
 664         u_int page;
 665         int i, entry, retval;
 666
 667         crit_enter();
 668         ASSERT_MP_LOCK_HELD(curthread);
 669
 670         for (i = 0; i < vmc->npages; i++) {
 671                 page = vtophys(vmc->pmap[i].kva & PG_FRAME);
 672                 entry = vmc->pmap[i].pte_num;
 673                 vmc->pmap[i].old_pte = pte[entry];
 674                 pte[entry] = page | PG_V | PG_RW | PG_U;
 675         }
 676
 677         vmf->vmf_trapno = intnum;
 678         retval = vm86_bioscall(vmf);
 679
 680         for (i = 0; i < vmc->npages; i++) {
 681                 entry = vmc->pmap[i].pte_num;
 682                 pte[entry] = vmc->pmap[i].old_pte;
 683         }
 684         crit_exit();
 685         return (retval);
 686 }
 687
 688 vm_offset_t
 689 vm86_getaddr(struct vm86context *vmc, u_short sel, 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(struct vm86context *vmc, vm_offset_t kva, u_short *sel,
 704             u_short *off)
 705 {
 706         int i;
 707
 708         for (i = 0; i < vmc->npages; i++)
 709                 if (kva >= vmc->pmap[i].kva &&
 710                     kva < vmc->pmap[i].kva + PAGE_SIZE) {
 711                         *off = kva - vmc->pmap[i].kva;
 712                         *sel = vmc->pmap[i].pte_num << 8;
 713                         return (1);
 714                 }
 715         return (0);
 716         panic("vm86_getptr: address not found");
 717 }
 718
 719 int
 720 vm86_sysarch(struct lwp *lp, char *args)
 721 {
 722         int error = 0;
 723         struct i386_vm86_args ua;
 724         struct vm86_kernel *vm86;
 725
 726         if ((error = copyin(args, &ua, sizeof(struct i386_vm86_args))) != 0)
 727                 return (error);
 728
 729         if (lp->lwp_thread->td_pcb->pcb_ext == 0)
 730                 if ((error = i386_extend_pcb(lp)) != 0)
 731                         return (error);
 732         vm86 = &lp->lwp_thread->td_pcb->pcb_ext->ext_vm86;
 733
 734         switch (ua.sub_op) {
 735         case VM86_INIT: {
 736                 struct vm86_init_args sa;
 737
 738                 if ((error = copyin(ua.sub_args, &sa, sizeof(sa))) != 0)
 739                         return (error);
 740                 if (cpu_feature & CPUID_VME)
 741                         vm86->vm86_has_vme = (rcr4() & CR4_VME ? 1 : 0);
 742                 else
 743                         vm86->vm86_has_vme = 0;
 744                 vm86->vm86_inited = 1;
 745                 vm86->vm86_debug = sa.debug;
 746                 bcopy(&sa.int_map, vm86->vm86_intmap, 32);
 747                 }
 748                 break;
 749
 750 #if 0
 751         case VM86_SET_VME: {
 752                 struct vm86_vme_args sa;
 753
 754                 if ((cpu_feature & CPUID_VME) == 0)
 755                         return (ENODEV);
 756
 757                 if (error = copyin(ua.sub_args, &sa, sizeof(sa)))
 758                         return (error);
 759                 if (sa.state)
 760                         load_cr4(rcr4() | CR4_VME);
 761                 else
 762                         load_cr4(rcr4() & ~CR4_VME);
 763                 }
 764                 break;
 765 #endif
 766
 767         case VM86_GET_VME: {
 768                 struct vm86_vme_args sa;
 769
 770                 sa.state = (rcr4() & CR4_VME ? 1 : 0);
 771                 error = copyout(&sa, ua.sub_args, sizeof(sa));
 772                 }
 773                 break;
 774
 775         case VM86_INTCALL: {
 776                 struct vm86_intcall_args sa;
 777
 778                 if ((error = priv_check_cred(lp->lwp_proc->p_ucred, PRIV_ROOT, 0)))
 779                         return (error);
 780                 if ((error = copyin(ua.sub_args, &sa, sizeof(sa))))
 781                         return (error);
 782                 if ((error = vm86_intcall(sa.intnum, &sa.vmf)))
 783                         return (error);
 784                 error = copyout(&sa, ua.sub_args, sizeof(sa));
 785                 }
 786                 break;
 787
 788         default:
 789                 error = EINVAL;
 790         }
 791         return (error);
 792 }
 793
 794 /*
 795  * Setup the VM86 I/O map to take faults on the timer
 796  */
 797 static void
 798 vm86_setup_timer_fault(void)
 799 {
 800         struct vm86_layout *vml = (struct vm86_layout *)vm86paddr;
 801
 802         vml->vml_iomap[TIMER_MODE >> 3] |= 1 << (TIMER_MODE & 7);
 803         vml->vml_iomap[TIMER_CNTR0 >> 3] |= 1 << (TIMER_CNTR0 & 7);
 804         vml->vml_iomap[TIMER_CNTR1 >> 3] |= 1 << (TIMER_CNTR1 & 7);
 805         vml->vml_iomap[TIMER_CNTR2 >> 3] |= 1 << (TIMER_CNTR2 & 7);
 806 }
 807
 808 /*
 809  * Setup the VM86 I/O map to not fault on the timer
 810  */
 811 static void
 812 vm86_clear_timer_fault(void)
 813 {
 814         struct vm86_layout *vml = (struct vm86_layout *)vm86paddr;
 815
 816         vml->vml_iomap[TIMER_MODE >> 3] &= ~(1 << (TIMER_MODE & 7));
 817         vml->vml_iomap[TIMER_CNTR0 >> 3] &= ~(1 << (TIMER_CNTR0 & 7));
 818         vml->vml_iomap[TIMER_CNTR1 >> 3] &= ~(1 << (TIMER_CNTR1 & 7));
 819         vml->vml_iomap[TIMER_CNTR2 >> 3] &= ~(1 << (TIMER_CNTR2 & 7));
 820 }
 821