2 * Copyright (c) 1997 Jonathan Lemon
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
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.
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
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/i386/i386/Attic/vm86.c,v 1.12 2004/11/20 20:25:05 dillon Exp $
31 #include <sys/param.h>
32 #include <sys/systm.h>
35 #include <sys/malloc.h>
39 #include <vm/vm_map.h>
40 #include <vm/vm_page.h>
43 #include <sys/thread2.h>
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>
55 extern int i386_extend_pcb (struct proc *);
57 extern struct pcb *vm86pcb;
59 extern int vm86_bioscall(struct vm86frame *);
60 extern void vm86_biosret(struct vm86frame *);
62 #define PGTABLE_SIZE ((1024 + 64) * 1024 / PAGE_SIZE)
63 #define INTMAP_SIZE 32
64 #define IOMAP_SIZE ctob(IOPAGES)
66 (sizeof(struct pcb_ext) - sizeof(struct segment_descriptor) + \
67 INTMAP_SIZE + IOMAP_SIZE + 1)
70 pt_entry_t vml_pgtbl[PGTABLE_SIZE];
72 struct pcb_ext vml_ext;
73 char vml_intmap[INTMAP_SIZE];
74 char vml_iomap[IOMAP_SIZE];
75 char vml_iomap_trailer;
78 void vm86_prepcall(struct vm86frame);
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)
107 static void vm86_setup_timer_fault(void);
108 static void vm86_clear_timer_fault(void);
110 static int vm86_blew_up_timer;
112 static __inline caddr_t
113 MAKE_ADDR(u_short sel, u_short off)
115 return ((caddr_t)((sel << 4) + off));
119 GET_VEC(u_int vec, u_short *sel, u_short *off)
125 static __inline u_int
126 MAKE_VEC(u_short sel, u_short off)
128 return ((sel << 16) | off);
132 PUSH(u_short x, struct vm86frame *vmf)
135 susword(MAKE_ADDR(vmf->vmf_ss, vmf->vmf_sp), x);
139 PUSHL(u_int x, struct vm86frame *vmf)
142 suword(MAKE_ADDR(vmf->vmf_ss, vmf->vmf_sp), x);
145 static __inline u_short
146 POP(struct vm86frame *vmf)
148 u_short x = fusword(MAKE_ADDR(vmf->vmf_ss, vmf->vmf_sp));
154 static __inline u_int
155 POPL(struct vm86frame *vmf)
157 u_int x = fuword(MAKE_ADDR(vmf->vmf_ss, vmf->vmf_sp));
165 struct vm86frame *vmf;
167 struct vm86_kernel *vm86;
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)
179 if (curthread->td_pcb->pcb_ext == 0)
181 vm86 = &curthread->td_pcb->pcb_ext->ext_vm86;
183 if (vmf->vmf_eflags & PSL_T)
187 * Instruction emulation
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);
197 * I/O emulation (TIMER only, a big hack). Just reenable the
198 * IO bits involved, flag it, and retry the instruction.
205 vm86_blew_up_timer = 1;
211 vm86_clear_timer_fault();
216 if (vm86->vm86_has_vme) {
218 case OPERAND_SIZE_PREFIX:
219 i_byte = fubyte(++addr);
223 if (vmf->vmf_eflags & PSL_VIF)
224 PUSHL((vmf->vmf_eflags & PUSH_MASK)
225 | PSL_IOPL | PSL_I, vmf);
227 PUSHL((vmf->vmf_eflags & PUSH_MASK)
229 vmf->vmf_ip += inc_ip;
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)
242 vmf->vmf_eflags &= ~PSL_VIF;
248 /* VME faults here if VIP is set, but does not set VIF. */
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");
258 /* VME if no redirection support */
262 /* VME if trying to set PSL_TF, or PSL_I when VIP is set */
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)
273 vmf->vmf_eflags &= ~PSL_VIF;
277 /* VME if trying to set PSL_TF, or PSL_I when VIP is set */
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)
289 vmf->vmf_eflags &= ~PSL_VIF;
298 case OPERAND_SIZE_PREFIX:
299 i_byte = fubyte(++addr);
303 if (vm86->vm86_eflags & PSL_VIF)
304 PUSHL((vmf->vmf_flags & PUSH_MASK)
305 | PSL_IOPL | PSL_I, vmf);
307 PUSHL((vmf->vmf_flags & PUSH_MASK)
309 vmf->vmf_ip += inc_ip;
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)
322 vm86->vm86_eflags &= ~PSL_VIF;
329 vm86->vm86_eflags &= ~PSL_VIF;
330 vmf->vmf_ip += inc_ip;
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)
342 if (vm86->vm86_eflags & PSL_VIF)
343 PUSH((vmf->vmf_flags & PUSH_MASK)
344 | PSL_IOPL | PSL_I, vmf);
346 PUSH((vmf->vmf_flags & PUSH_MASK) | PSL_IOPL, vmf);
347 vmf->vmf_ip += inc_ip;
351 i_byte = fubyte(addr + 1);
352 if ((vm86->vm86_intmap[i_byte >> 3] & (1 << (i_byte & 7))) != 0)
354 if (vm86->vm86_eflags & PSL_VIF)
355 PUSH((vmf->vmf_flags & PUSH_MASK)
356 | PSL_IOPL | PSL_I, vmf);
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;
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)
378 vm86->vm86_eflags &= ~PSL_VIF;
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)
392 vm86->vm86_eflags &= ~PSL_VIF;
400 vm86_initialize(void)
404 struct vm86_layout *vml = (struct vm86_layout *)vm86paddr;
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 */
414 0, /* default 16 size */
419 * this should be a compile time error, but cpp doesn't grok sizeof().
421 if (sizeof(struct vm86_layout) > ctob(3))
422 panic("struct vm86_layout exceeds space allocated in locore.s");
425 * Below is the memory layout that we use for the vm86 region.
433 * +--------+ +--------+ <--------- vm86paddr
434 * | | |Page Tbl| 1M + 64K = 272 entries = 1088 bytes
436 * | | | PCB | size: ~240 bytes
437 * | page 1 | |PCB Ext | size: ~140 bytes (includes TSS)
443 * +--------+ | bitmap |
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.
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
462 #define new_ptd pcb_esi
463 #define vm86_frame pcb_ebp
464 #define pgtable_va pcb_ebx
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;
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;
483 if (cpu_feature & CPUID_VME)
484 ext->ext_vm86.vm86_has_vme = (rcr4() & CR4_VME ? 1 : 0);
486 addr = (u_int *)ext->ext_vm86.vm86_intmap;
487 for (i = 0; i < (INTMAP_SIZE + IOMAP_SIZE) / sizeof(u_int); i++)
489 vml->vml_iomap_trailer = 0xff;
491 ssd.ssd_base = (u_int)&ext->ext_tss;
492 ssd.ssd_limit = TSS_SIZE - 1;
493 ssdtosd(&ssd, &ext->ext_tssd);
499 * use whatever is leftover of the vm86 page layout as a
500 * message buffer so we can capture early output.
502 msgbufinit((vm_offset_t)vm86paddr + sizeof(struct vm86_layout),
503 ctob(3) - sizeof(struct vm86_layout));
508 vm86_getpage(struct vm86context *vmc, int pagenum)
512 for (i = 0; i < vmc->npages; i++)
513 if (vmc->pmap[i].pte_num == pagenum)
514 return (vmc->pmap[i].kva);
519 vm86_addpage(struct vm86context *vmc, int pagenum, vm_offset_t kva)
523 for (i = 0; i < vmc->npages; i++)
524 if (vmc->pmap[i].pte_num == pagenum)
527 if (vmc->npages == VM86_PMAPSIZE)
528 goto bad; /* XXX grow map? */
531 kva = (vm_offset_t)malloc(PAGE_SIZE, M_TEMP, M_WAITOK);
536 vmc->pmap[i].flags = flags;
537 vmc->pmap[i].kva = kva;
538 vmc->pmap[i].pte_num = pagenum;
541 panic("vm86_addpage: not enough room, or overlap");
545 vm86_initflags(struct vm86frame *vmf)
547 int eflags = vmf->vmf_eflags;
548 struct vm86_kernel *vm86 = &curthread->td_pcb->pcb_ext->ext_vm86;
550 if (vm86->vm86_has_vme) {
551 eflags = (vmf->vmf_eflags & ~VME_USERCHANGE) |
552 (eflags & VME_USERCHANGE) | PSL_VM;
554 vm86->vm86_eflags = eflags; /* save VIF, VIP */
555 eflags = (vmf->vmf_eflags & ~VM_USERCHANGE) |
556 (eflags & VM_USERCHANGE) | PSL_VM;
558 vmf->vmf_eflags = eflags | PSL_VM;
562 * called from vm86_bioscall, while in vm86 address space, to finalize setup.
565 vm86_prepcall(struct vm86frame vmf)
567 uintptr_t addr[] = { 0xA00, 0x1000 }; /* code, stack */
569 CLI, INTn, 0x00, STI, HLT
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];
579 vmf.vmf_sp = addr[1] - 2; /* keep aligned */
580 vmf.kernel_fs = vmf.kernel_es = vmf.kernel_ds = 0;
582 vmf.vmf_eflags = PSL_VIF | PSL_VM | PSL_USER;
583 vm86_initflags(&vmf);
587 * vm86 trap handler; determines whether routine succeeded or not.
588 * Called while in vm86 space, returns to calling process.
590 * A MP lock ref is held on entry from trap() and must be released prior
591 * to returning to the VM86 call.
594 vm86_trap(struct vm86frame *vmf)
598 /* "should not happen" */
599 if ((vmf->vmf_eflags & PSL_VM) == 0)
600 panic("vm86_trap called, but not in vm86 mode");
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;
606 vmf->vmf_trapno = vmf->vmf_trapno << 16;
613 vm86_intcall(int intnum, struct vm86frame *vmf)
617 if (intnum < 0 || intnum > 0xff)
621 ASSERT_MP_LOCK_HELD();
623 vm86_setup_timer_fault();
624 vmf->vmf_trapno = intnum;
625 error = vm86_bioscall(vmf);
628 * Yes, this happens, especially with video BIOS calls. The BIOS
629 * will sometimes eat timer 2 for lunch, and we need timer 2.
631 if (vm86_blew_up_timer) {
632 vm86_blew_up_timer = 0;
634 printf("Warning: BIOS messed around with the 8254, resetting it\n");
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.
647 vm86_datacall(intnum, vmf, vmc)
649 struct vm86frame *vmf;
650 struct vm86context *vmc;
652 pt_entry_t *pte = vm86paddr;
654 int i, entry, retval;
657 ASSERT_MP_LOCK_HELD();
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;
666 vmf->vmf_trapno = intnum;
667 retval = vm86_bioscall(vmf);
669 for (i = 0; i < vmc->npages; i++) {
670 entry = vmc->pmap[i].pte_num;
671 pte[entry] = vmc->pmap[i].old_pte;
678 vm86_getaddr(vmc, sel, off)
679 struct vm86context *vmc;
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));
695 vm86_getptr(vmc, kva, sel, off)
696 struct vm86context *vmc;
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;
711 panic("vm86_getptr: address not found");
715 vm86_sysarch(struct proc *p, char *args)
718 struct i386_vm86_args ua;
719 struct vm86_kernel *vm86;
721 if ((error = copyin(args, &ua, sizeof(struct i386_vm86_args))) != 0)
724 if (p->p_thread->td_pcb->pcb_ext == 0)
725 if ((error = i386_extend_pcb(p)) != 0)
727 vm86 = &p->p_thread->td_pcb->pcb_ext->ext_vm86;
731 struct vm86_init_args sa;
733 if ((error = copyin(ua.sub_args, &sa, sizeof(sa))) != 0)
735 if (cpu_feature & CPUID_VME)
736 vm86->vm86_has_vme = (rcr4() & CR4_VME ? 1 : 0);
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);
747 struct vm86_vme_args sa;
749 if ((cpu_feature & CPUID_VME) == 0)
752 if (error = copyin(ua.sub_args, &sa, sizeof(sa)))
755 load_cr4(rcr4() | CR4_VME);
757 load_cr4(rcr4() & ~CR4_VME);
763 struct vm86_vme_args sa;
765 sa.state = (rcr4() & CR4_VME ? 1 : 0);
766 error = copyout(&sa, ua.sub_args, sizeof(sa));
771 struct vm86_intcall_args sa;
773 if ((error = suser_cred(p->p_ucred, 0)))
775 if ((error = copyin(ua.sub_args, &sa, sizeof(sa))))
777 if ((error = vm86_intcall(sa.intnum, &sa.vmf)))
779 error = copyout(&sa, ua.sub_args, sizeof(sa));
790 * Setup the VM86 I/O map to take faults on the timer
793 vm86_setup_timer_fault(void)
795 struct vm86_layout *vml = (struct vm86_layout *)vm86paddr;
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);
804 * Setup the VM86 I/O map to not fault on the timer
807 vm86_clear_timer_fault(void)
809 struct vm86_layout *vml = (struct vm86_layout *)vm86paddr;
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));