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/platform/pc32/i386/vm86.c,v 1.25 2007/01/22 19:37:04 corecode Exp $
31 #include <sys/param.h>
32 #include <sys/systm.h>
33 #include <sys/kernel.h>
36 #include <sys/malloc.h>
37 #include <sys/sysctl.h>
41 #include <vm/vm_map.h>
42 #include <vm/vm_page.h>
45 #include <sys/thread2.h>
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>
57 extern int i386_extend_pcb (struct lwp *);
59 extern struct pcb *vm86pcb;
61 extern int vm86_bioscall(struct vm86frame *);
62 extern void vm86_biosret(struct vm86frame *);
64 #define PGTABLE_SIZE ((1024 + 64) * 1024 / PAGE_SIZE)
65 #define INTMAP_SIZE 32
66 #define IOMAP_SIZE ctob(IOPAGES)
68 (sizeof(struct pcb_ext) - sizeof(struct segment_descriptor) + \
69 INTMAP_SIZE + IOMAP_SIZE + 1)
72 pt_entry_t vml_pgtbl[PGTABLE_SIZE];
74 struct pcb_ext vml_ext;
75 char vml_intmap[INTMAP_SIZE];
76 char vml_iomap[IOMAP_SIZE];
77 char vml_iomap_trailer;
80 void vm86_prepcall(struct vm86frame *);
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)
109 static void vm86_setup_timer_fault(void);
110 static void vm86_clear_timer_fault(void);
112 static int vm86_blew_up_timer;
114 static int timer_warn = 1;
115 SYSCTL_INT(_debug, OID_AUTO, timer_warn, CTLFLAG_RW, &timer_warn, 0, "");
117 static __inline caddr_t
118 MAKE_ADDR(u_short sel, u_short off)
120 return ((caddr_t)((sel << 4) + off));
124 GET_VEC(u_int vec, u_short *sel, u_short *off)
130 static __inline u_int
131 MAKE_VEC(u_short sel, u_short off)
133 return ((sel << 16) | off);
137 PUSH(u_short x, struct vm86frame *vmf)
140 susword(MAKE_ADDR(vmf->vmf_ss, vmf->vmf_sp), x);
144 PUSHL(u_int x, struct vm86frame *vmf)
147 suword(MAKE_ADDR(vmf->vmf_ss, vmf->vmf_sp), x);
150 static __inline u_short
151 POP(struct vm86frame *vmf)
153 u_short x = fusword(MAKE_ADDR(vmf->vmf_ss, vmf->vmf_sp));
159 static __inline u_int
160 POPL(struct vm86frame *vmf)
162 u_int x = fuword(MAKE_ADDR(vmf->vmf_ss, vmf->vmf_sp));
172 vm86_emulate(struct vm86frame *vmf)
174 struct vm86_kernel *vm86;
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)
186 if (curthread->td_pcb->pcb_ext == 0)
188 vm86 = &curthread->td_pcb->pcb_ext->ext_vm86;
190 if (vmf->vmf_eflags & PSL_T)
194 * Instruction emulation
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);
204 * I/O emulation (TIMER only, a big hack). Just reenable the
205 * IO bits involved, flag it, and retry the instruction.
212 vm86_blew_up_timer = 1;
218 vm86_clear_timer_fault();
223 if (vm86->vm86_has_vme) {
225 case OPERAND_SIZE_PREFIX:
226 i_byte = fubyte(++addr);
230 if (vmf->vmf_eflags & PSL_VIF)
231 PUSHL((vmf->vmf_eflags & PUSH_MASK)
232 | PSL_IOPL | PSL_I, vmf);
234 PUSHL((vmf->vmf_eflags & PUSH_MASK)
236 vmf->vmf_ip += inc_ip;
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)
249 vmf->vmf_eflags &= ~PSL_VIF;
255 /* VME faults here if VIP is set, but does not set VIF. */
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");
265 /* VME if no redirection support */
269 /* VME if trying to set PSL_TF, or PSL_I when VIP is set */
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)
280 vmf->vmf_eflags &= ~PSL_VIF;
284 /* VME if trying to set PSL_TF, or PSL_I when VIP is set */
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)
296 vmf->vmf_eflags &= ~PSL_VIF;
305 case OPERAND_SIZE_PREFIX:
306 i_byte = fubyte(++addr);
310 if (vm86->vm86_eflags & PSL_VIF)
311 PUSHL((vmf->vmf_flags & PUSH_MASK)
312 | PSL_IOPL | PSL_I, vmf);
314 PUSHL((vmf->vmf_flags & PUSH_MASK)
316 vmf->vmf_ip += inc_ip;
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)
329 vm86->vm86_eflags &= ~PSL_VIF;
336 vm86->vm86_eflags &= ~PSL_VIF;
337 vmf->vmf_ip += inc_ip;
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)
349 if (vm86->vm86_eflags & PSL_VIF)
350 PUSH((vmf->vmf_flags & PUSH_MASK)
351 | PSL_IOPL | PSL_I, vmf);
353 PUSH((vmf->vmf_flags & PUSH_MASK) | PSL_IOPL, vmf);
354 vmf->vmf_ip += inc_ip;
358 i_byte = fubyte(addr + 1);
359 if ((vm86->vm86_intmap[i_byte >> 3] & (1 << (i_byte & 7))) != 0)
361 if (vm86->vm86_eflags & PSL_VIF)
362 PUSH((vmf->vmf_flags & PUSH_MASK)
363 | PSL_IOPL | PSL_I, vmf);
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;
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)
385 vm86->vm86_eflags &= ~PSL_VIF;
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)
399 vm86->vm86_eflags &= ~PSL_VIF;
407 vm86_initialize(void)
411 struct vm86_layout *vml = (struct vm86_layout *)vm86paddr;
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 */
421 0, /* default 16 size */
426 * this should be a compile time error, but cpp doesn't grok sizeof().
428 if (sizeof(struct vm86_layout) > ctob(3))
429 panic("struct vm86_layout exceeds space allocated in locore.s");
432 * Below is the memory layout that we use for the vm86 region.
440 * +--------+ +--------+ <--------- vm86paddr
441 * | | |Page Tbl| 1M + 64K = 272 entries = 1088 bytes
443 * | | | PCB | size: ~240 bytes
444 * | page 1 | |PCB Ext | size: ~140 bytes (includes TSS)
450 * +--------+ | bitmap |
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.
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
469 #define new_ptd pcb_esi
470 #define vm86_frame pcb_ebp
471 #define pgtable_va pcb_ebx
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;
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;
490 if (cpu_feature & CPUID_VME)
491 ext->ext_vm86.vm86_has_vme = (rcr4() & CR4_VME ? 1 : 0);
493 addr = (u_int *)ext->ext_vm86.vm86_intmap;
494 for (i = 0; i < (INTMAP_SIZE + IOMAP_SIZE) / sizeof(u_int); i++)
496 vml->vml_iomap_trailer = 0xff;
498 ssd.ssd_base = (u_int)&ext->ext_tss;
499 ssd.ssd_limit = TSS_SIZE - 1;
500 ssdtosd(&ssd, &ext->ext_tssd);
506 * use whatever is leftover of the vm86 page layout as a
507 * message buffer so we can capture early output.
509 msgbufinit((vm_offset_t)vm86paddr + sizeof(struct vm86_layout),
510 ctob(3) - sizeof(struct vm86_layout));
515 vm86_getpage(struct vm86context *vmc, int pagenum)
519 for (i = 0; i < vmc->npages; i++)
520 if (vmc->pmap[i].pte_num == pagenum)
521 return (vmc->pmap[i].kva);
526 vm86_addpage(struct vm86context *vmc, int pagenum, vm_offset_t kva)
530 for (i = 0; i < vmc->npages; i++)
531 if (vmc->pmap[i].pte_num == pagenum)
534 if (vmc->npages == VM86_PMAPSIZE)
535 goto bad; /* XXX grow map? */
538 kva = (vm_offset_t)kmalloc(PAGE_SIZE, M_TEMP, M_WAITOK);
543 vmc->pmap[i].flags = flags;
544 vmc->pmap[i].kva = kva;
545 vmc->pmap[i].pte_num = pagenum;
548 panic("vm86_addpage: not enough room, or overlap");
552 vm86_initflags(struct vm86frame *vmf)
554 int eflags = vmf->vmf_eflags;
555 struct vm86_kernel *vm86 = &curthread->td_pcb->pcb_ext->ext_vm86;
557 if (vm86->vm86_has_vme) {
558 eflags = (vmf->vmf_eflags & ~VME_USERCHANGE) |
559 (eflags & VME_USERCHANGE) | PSL_VM;
561 vm86->vm86_eflags = eflags; /* save VIF, VIP */
562 eflags = (vmf->vmf_eflags & ~VM_USERCHANGE) |
563 (eflags & VM_USERCHANGE) | PSL_VM;
565 vmf->vmf_eflags = eflags | PSL_VM;
569 * called from vm86_bioscall, while in vm86 address space, to finalize setup.
572 vm86_prepcall(struct vm86frame *vmf)
574 uintptr_t addr[] = { 0xA00, 0x1000 }; /* code, stack */
576 CLI, INTn, 0x00, STI, HLT
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];
586 vmf->vmf_sp = addr[1] - 2; /* keep aligned */
587 vmf->kernel_fs = vmf->kernel_es = vmf->kernel_ds = vmf->kernel_gs = 0;
589 vmf->vmf_eflags = PSL_VIF | PSL_VM | PSL_USER;
594 * vm86 trap handler; determines whether routine succeeded or not.
595 * Called while in vm86 space, returns to calling process.
597 * A MP lock ref is held on entry from trap() and must be released prior
598 * to returning to the VM86 call.
601 vm86_trap(struct vm86frame *vmf, int have_mplock)
605 /* "should not happen" */
606 if ((vmf->vmf_eflags & PSL_VM) == 0)
607 panic("vm86_trap called, but not in vm86 mode");
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;
613 vmf->vmf_trapno = vmf->vmf_trapno << 16;
621 vm86_intcall(int intnum, struct vm86frame *vmf)
625 if (intnum < 0 || intnum > 0xff)
629 ASSERT_MP_LOCK_HELD(curthread);
631 vm86_setup_timer_fault();
632 vmf->vmf_trapno = intnum;
633 error = vm86_bioscall(vmf);
636 * Yes, this happens, especially with video BIOS calls. The BIOS
637 * will sometimes eat timer 2 for lunch, and we need timer 2.
639 if (vm86_blew_up_timer) {
640 vm86_blew_up_timer = 0;
643 kprintf("Warning: BIOS played with the 8254, "
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.
658 vm86_datacall(int intnum, struct vm86frame *vmf, struct vm86context *vmc)
660 pt_entry_t *pte = vm86paddr;
662 int i, entry, retval;
665 ASSERT_MP_LOCK_HELD(curthread);
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;
674 vmf->vmf_trapno = intnum;
675 retval = vm86_bioscall(vmf);
677 for (i = 0; i < vmc->npages; i++) {
678 entry = vmc->pmap[i].pte_num;
679 pte[entry] = vmc->pmap[i].old_pte;
686 vm86_getaddr(struct vm86context *vmc, u_short sel, u_short off)
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));
700 vm86_getptr(struct vm86context *vmc, vm_offset_t kva, u_short *sel,
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;
713 panic("vm86_getptr: address not found");
717 vm86_sysarch(struct lwp *lp, char *args)
720 struct i386_vm86_args ua;
721 struct vm86_kernel *vm86;
723 if ((error = copyin(args, &ua, sizeof(struct i386_vm86_args))) != 0)
726 if (lp->lwp_thread->td_pcb->pcb_ext == 0)
727 if ((error = i386_extend_pcb(lp)) != 0)
729 vm86 = &lp->lwp_thread->td_pcb->pcb_ext->ext_vm86;
733 struct vm86_init_args sa;
735 if ((error = copyin(ua.sub_args, &sa, sizeof(sa))) != 0)
737 if (cpu_feature & CPUID_VME)
738 vm86->vm86_has_vme = (rcr4() & CR4_VME ? 1 : 0);
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);
749 struct vm86_vme_args sa;
751 if ((cpu_feature & CPUID_VME) == 0)
754 if (error = copyin(ua.sub_args, &sa, sizeof(sa)))
757 load_cr4(rcr4() | CR4_VME);
759 load_cr4(rcr4() & ~CR4_VME);
765 struct vm86_vme_args sa;
767 sa.state = (rcr4() & CR4_VME ? 1 : 0);
768 error = copyout(&sa, ua.sub_args, sizeof(sa));
773 struct vm86_intcall_args sa;
775 if ((error = suser_cred(lp->lwp_proc->p_ucred, 0)))
777 if ((error = copyin(ua.sub_args, &sa, sizeof(sa))))
779 if ((error = vm86_intcall(sa.intnum, &sa.vmf)))
781 error = copyout(&sa, ua.sub_args, sizeof(sa));
792 * Setup the VM86 I/O map to take faults on the timer
795 vm86_setup_timer_fault(void)
797 struct vm86_layout *vml = (struct vm86_layout *)vm86paddr;
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);
806 * Setup the VM86 I/O map to not fault on the timer
809 vm86_clear_timer_fault(void)
811 struct vm86_layout *vml = (struct vm86_layout *)vm86paddr;
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));