Import 2.3.13pre1
[davej-history.git] / arch / i386 / kernel / vm86.c
blob65dd7e9da921484ee0a9b0895607866cd103bfb8
1 /*
2 * linux/kernel/vm86.c
4 * Copyright (C) 1994 Linus Torvalds
5 */
6 #include <linux/errno.h>
7 #include <linux/sched.h>
8 #include <linux/kernel.h>
9 #include <linux/signal.h>
10 #include <linux/string.h>
11 #include <linux/ptrace.h>
12 #include <linux/mm.h>
13 #include <linux/smp.h>
14 #include <linux/smp_lock.h>
16 #include <asm/uaccess.h>
17 #include <asm/pgtable.h>
18 #include <asm/io.h>
21 * Known problems:
23 * Interrupt handling is not guaranteed:
24 * - a real x86 will disable all interrupts for one instruction
25 * after a "mov ss,xx" to make stack handling atomic even without
26 * the 'lss' instruction. We can't guarantee this in v86 mode,
27 * as the next instruction might result in a page fault or similar.
28 * - a real x86 will have interrupts disabled for one instruction
29 * past the 'sti' that enables them. We don't bother with all the
30 * details yet.
32 * Let's hope these problems do not actually matter for anything.
36 #define KVM86 ((struct kernel_vm86_struct *)regs)
37 #define VMPI KVM86->vm86plus
41 * 8- and 16-bit register defines..
43 #define AL(regs) (((unsigned char *)&((regs)->eax))[0])
44 #define AH(regs) (((unsigned char *)&((regs)->eax))[1])
45 #define IP(regs) (*(unsigned short *)&((regs)->eip))
46 #define SP(regs) (*(unsigned short *)&((regs)->esp))
49 * virtual flags (16 and 32-bit versions)
51 #define VFLAGS (*(unsigned short *)&(current->thread.v86flags))
52 #define VEFLAGS (current->thread.v86flags)
54 #define set_flags(X,new,mask) \
55 ((X) = ((X) & ~(mask)) | ((new) & (mask)))
57 #define SAFE_MASK (0xDD5)
58 #define RETURN_MASK (0xDFF)
60 #define VM86_REGS_PART2 orig_eax
61 #define VM86_REGS_SIZE1 \
62 ( (unsigned)( & (((struct kernel_vm86_regs *)0)->VM86_REGS_PART2) ) )
63 #define VM86_REGS_SIZE2 (sizeof(struct kernel_vm86_regs) - VM86_REGS_SIZE1)
65 asmlinkage struct pt_regs * FASTCALL(save_v86_state(struct kernel_vm86_regs * regs));
66 struct pt_regs * save_v86_state(struct kernel_vm86_regs * regs)
68 struct tss_struct *tss;
69 struct pt_regs *ret;
70 unsigned long tmp;
72 lock_kernel();
73 if (!current->thread.vm86_info) {
74 printk("no vm86_info: BAD\n");
75 do_exit(SIGSEGV);
77 set_flags(regs->eflags, VEFLAGS, VIF_MASK | current->thread.v86mask);
78 tmp = copy_to_user(&current->thread.vm86_info->regs,regs, VM86_REGS_SIZE1);
79 tmp += copy_to_user(&current->thread.vm86_info->regs.VM86_REGS_PART2,
80 &regs->VM86_REGS_PART2, VM86_REGS_SIZE2);
81 tmp += put_user(current->thread.screen_bitmap,&current->thread.vm86_info->screen_bitmap);
82 if (tmp) {
83 printk("vm86: could not access userspace vm86_info\n");
84 do_exit(SIGSEGV);
86 tss = init_tss + smp_processor_id();
87 tss->esp0 = current->thread.esp0 = current->thread.saved_esp0;
88 current->thread.saved_esp0 = 0;
89 ret = KVM86->regs32;
90 unlock_kernel();
91 return ret;
94 static void mark_screen_rdonly(struct task_struct * tsk)
96 pgd_t *pgd;
97 pmd_t *pmd;
98 pte_t *pte;
99 int i;
101 pgd = pgd_offset(tsk->mm, 0xA0000);
102 if (pgd_none(*pgd))
103 return;
104 if (pgd_bad(*pgd)) {
105 printk("vm86: bad pgd entry [%p]:%08lx\n", pgd, pgd_val(*pgd));
106 pgd_clear(pgd);
107 return;
109 pmd = pmd_offset(pgd, 0xA0000);
110 if (pmd_none(*pmd))
111 return;
112 if (pmd_bad(*pmd)) {
113 printk("vm86: bad pmd entry [%p]:%08lx\n", pmd, pmd_val(*pmd));
114 pmd_clear(pmd);
115 return;
117 pte = pte_offset(pmd, 0xA0000);
118 for (i = 0; i < 32; i++) {
119 if (pte_present(*pte))
120 set_pte(pte, pte_wrprotect(*pte));
121 pte++;
123 flush_tlb();
128 static int do_vm86_irq_handling(int subfunction, int irqnumber);
129 static void do_sys_vm86(struct kernel_vm86_struct *info, struct task_struct *tsk);
131 asmlinkage int sys_vm86old(struct vm86_struct * v86)
133 struct kernel_vm86_struct info; /* declare this _on top_,
134 * this avoids wasting of stack space.
135 * This remains on the stack until we
136 * return to 32 bit user space.
138 struct task_struct *tsk;
139 int tmp, ret = -EPERM;
141 lock_kernel();
142 tsk = current;
143 if (tsk->thread.saved_esp0)
144 goto out;
145 tmp = copy_from_user(&info, v86, VM86_REGS_SIZE1);
146 tmp += copy_from_user(&info.regs.VM86_REGS_PART2, &v86->regs.VM86_REGS_PART2,
147 (long)&info.vm86plus - (long)&info.regs.VM86_REGS_PART2);
148 ret = -EFAULT;
149 if (tmp)
150 goto out;
151 memset(&info.vm86plus, 0, (int)&info.regs32 - (int)&info.vm86plus);
152 info.regs32 = (struct pt_regs *) &v86;
153 tsk->thread.vm86_info = v86;
154 do_sys_vm86(&info, tsk);
155 ret = 0; /* we never return here */
156 out:
157 unlock_kernel();
158 return ret;
162 asmlinkage int sys_vm86(unsigned long subfunction, struct vm86plus_struct * v86)
164 struct kernel_vm86_struct info; /* declare this _on top_,
165 * this avoids wasting of stack space.
166 * This remains on the stack until we
167 * return to 32 bit user space.
169 struct task_struct *tsk;
170 int tmp, ret;
172 lock_kernel();
173 tsk = current;
174 switch (subfunction) {
175 case VM86_REQUEST_IRQ:
176 case VM86_FREE_IRQ:
177 case VM86_GET_IRQ_BITS:
178 case VM86_GET_AND_RESET_IRQ:
179 ret = do_vm86_irq_handling(subfunction,(int)v86);
180 goto out;
181 case VM86_PLUS_INSTALL_CHECK:
182 /* NOTE: on old vm86 stuff this will return the error
183 from verify_area(), because the subfunction is
184 interpreted as (invalid) address to vm86_struct.
185 So the installation check works.
187 ret = 0;
188 goto out;
191 /* we come here only for functions VM86_ENTER, VM86_ENTER_NO_BYPASS */
192 ret = -EPERM;
193 if (tsk->thread.saved_esp0)
194 goto out;
195 tmp = copy_from_user(&info, v86, VM86_REGS_SIZE1);
196 tmp += copy_from_user(&info.regs.VM86_REGS_PART2, &v86->regs.VM86_REGS_PART2,
197 (long)&info.regs32 - (long)&info.regs.VM86_REGS_PART2);
198 ret = -EFAULT;
199 if (tmp)
200 goto out;
201 info.regs32 = (struct pt_regs *) &subfunction;
202 info.vm86plus.is_vm86pus = 1;
203 tsk->thread.vm86_info = (struct vm86_struct *)v86;
204 do_sys_vm86(&info, tsk);
205 ret = 0; /* we never return here */
206 out:
207 unlock_kernel();
208 return ret;
212 static void do_sys_vm86(struct kernel_vm86_struct *info, struct task_struct *tsk)
214 struct tss_struct *tss;
216 * make sure the vm86() system call doesn't try to do anything silly
218 info->regs.__null_ds = 0;
219 info->regs.__null_es = 0;
221 /* we are clearing fs,gs later just before "jmp ret_from_sys_call",
222 * because starting with Linux 2.1.x they aren't no longer saved/restored
226 * The eflags register is also special: we cannot trust that the user
227 * has set it up safely, so this makes sure interrupt etc flags are
228 * inherited from protected mode.
230 VEFLAGS = info->regs.eflags;
231 info->regs.eflags &= SAFE_MASK;
232 info->regs.eflags |= info->regs32->eflags & ~SAFE_MASK;
233 info->regs.eflags |= VM_MASK;
235 switch (info->cpu_type) {
236 case CPU_286:
237 tsk->thread.v86mask = 0;
238 break;
239 case CPU_386:
240 tsk->thread.v86mask = NT_MASK | IOPL_MASK;
241 break;
242 case CPU_486:
243 tsk->thread.v86mask = AC_MASK | NT_MASK | IOPL_MASK;
244 break;
245 default:
246 tsk->thread.v86mask = ID_MASK | AC_MASK | NT_MASK | IOPL_MASK;
247 break;
251 * Save old state, set default return value (%eax) to 0
253 info->regs32->eax = 0;
254 tsk->thread.saved_esp0 = tsk->thread.esp0;
255 tss = init_tss + smp_processor_id();
256 tss->esp0 = tsk->thread.esp0 = (unsigned long) &info->VM86_TSS_ESP0;
258 tsk->thread.screen_bitmap = info->screen_bitmap;
259 if (info->flags & VM86_SCREEN_BITMAP)
260 mark_screen_rdonly(tsk);
261 unlock_kernel();
262 __asm__ __volatile__(
263 "xorl %%eax,%%eax; movl %%ax,%%fs; movl %%ax,%%gs\n\t"
264 "movl %0,%%esp\n\t"
265 "jmp ret_from_sys_call"
266 : /* no outputs */
267 :"r" (&info->regs), "b" (tsk) : "ax");
268 /* we never return here */
271 static inline void return_to_32bit(struct kernel_vm86_regs * regs16, int retval)
273 struct pt_regs * regs32;
275 regs32 = save_v86_state(regs16);
276 regs32->eax = retval;
277 unlock_kernel();
278 __asm__ __volatile__("movl %0,%%esp\n\t"
279 "jmp ret_from_sys_call"
280 : : "r" (regs32), "b" (current));
283 static inline void set_IF(struct kernel_vm86_regs * regs)
285 VEFLAGS |= VIF_MASK;
286 if (VEFLAGS & VIP_MASK)
287 return_to_32bit(regs, VM86_STI);
290 static inline void clear_IF(struct kernel_vm86_regs * regs)
292 VEFLAGS &= ~VIF_MASK;
295 static inline void clear_TF(struct kernel_vm86_regs * regs)
297 regs->eflags &= ~TF_MASK;
300 static inline void set_vflags_long(unsigned long eflags, struct kernel_vm86_regs * regs)
302 set_flags(VEFLAGS, eflags, current->thread.v86mask);
303 set_flags(regs->eflags, eflags, SAFE_MASK);
304 if (eflags & IF_MASK)
305 set_IF(regs);
308 static inline void set_vflags_short(unsigned short flags, struct kernel_vm86_regs * regs)
310 set_flags(VFLAGS, flags, current->thread.v86mask);
311 set_flags(regs->eflags, flags, SAFE_MASK);
312 if (flags & IF_MASK)
313 set_IF(regs);
316 static inline unsigned long get_vflags(struct kernel_vm86_regs * regs)
318 unsigned long flags = regs->eflags & RETURN_MASK;
320 if (VEFLAGS & VIF_MASK)
321 flags |= IF_MASK;
322 return flags | (VEFLAGS & current->thread.v86mask);
325 static inline int is_revectored(int nr, struct revectored_struct * bitmap)
327 __asm__ __volatile__("btl %2,%1\n\tsbbl %0,%0"
328 :"=r" (nr)
329 :"m" (*bitmap),"r" (nr));
330 return nr;
334 * Boy are these ugly, but we need to do the correct 16-bit arithmetic.
335 * Gcc makes a mess of it, so we do it inline and use non-obvious calling
336 * conventions..
338 #define pushb(base, ptr, val) \
339 __asm__ __volatile__( \
340 "decw %w0\n\t" \
341 "movb %2,0(%1,%0)" \
342 : "=r" (ptr) \
343 : "r" (base), "q" (val), "0" (ptr))
345 #define pushw(base, ptr, val) \
346 __asm__ __volatile__( \
347 "decw %w0\n\t" \
348 "movb %h2,0(%1,%0)\n\t" \
349 "decw %w0\n\t" \
350 "movb %b2,0(%1,%0)" \
351 : "=r" (ptr) \
352 : "r" (base), "q" (val), "0" (ptr))
354 #define pushl(base, ptr, val) \
355 __asm__ __volatile__( \
356 "decw %w0\n\t" \
357 "rorl $16,%2\n\t" \
358 "movb %h2,0(%1,%0)\n\t" \
359 "decw %w0\n\t" \
360 "movb %b2,0(%1,%0)\n\t" \
361 "decw %w0\n\t" \
362 "rorl $16,%2\n\t" \
363 "movb %h2,0(%1,%0)\n\t" \
364 "decw %w0\n\t" \
365 "movb %b2,0(%1,%0)" \
366 : "=r" (ptr) \
367 : "r" (base), "q" (val), "0" (ptr))
369 #define popb(base, ptr) \
370 ({ unsigned long __res; \
371 __asm__ __volatile__( \
372 "movb 0(%1,%0),%b2\n\t" \
373 "incw %w0" \
374 : "=r" (ptr), "=r" (base), "=q" (__res) \
375 : "0" (ptr), "1" (base), "2" (0)); \
376 __res; })
378 #define popw(base, ptr) \
379 ({ unsigned long __res; \
380 __asm__ __volatile__( \
381 "movb 0(%1,%0),%b2\n\t" \
382 "incw %w0\n\t" \
383 "movb 0(%1,%0),%h2\n\t" \
384 "incw %w0" \
385 : "=r" (ptr), "=r" (base), "=q" (__res) \
386 : "0" (ptr), "1" (base), "2" (0)); \
387 __res; })
389 #define popl(base, ptr) \
390 ({ unsigned long __res; \
391 __asm__ __volatile__( \
392 "movb 0(%1,%0),%b2\n\t" \
393 "incw %w0\n\t" \
394 "movb 0(%1,%0),%h2\n\t" \
395 "incw %w0\n\t" \
396 "rorl $16,%2\n\t" \
397 "movb 0(%1,%0),%b2\n\t" \
398 "incw %w0\n\t" \
399 "movb 0(%1,%0),%h2\n\t" \
400 "incw %w0\n\t" \
401 "rorl $16,%2" \
402 : "=r" (ptr), "=r" (base), "=q" (__res) \
403 : "0" (ptr), "1" (base)); \
404 __res; })
406 static void do_int(struct kernel_vm86_regs *regs, int i, unsigned char * ssp, unsigned long sp)
408 unsigned long *intr_ptr, segoffs;
410 if (regs->cs == BIOSSEG)
411 goto cannot_handle;
412 if (is_revectored(i, &KVM86->int_revectored))
413 goto cannot_handle;
414 if (i==0x21 && is_revectored(AH(regs),&KVM86->int21_revectored))
415 goto cannot_handle;
416 intr_ptr = (unsigned long *) (i << 2);
417 if (get_user(segoffs, intr_ptr))
418 goto cannot_handle;
419 if ((segoffs >> 16) == BIOSSEG)
420 goto cannot_handle;
421 pushw(ssp, sp, get_vflags(regs));
422 pushw(ssp, sp, regs->cs);
423 pushw(ssp, sp, IP(regs));
424 regs->cs = segoffs >> 16;
425 SP(regs) -= 6;
426 IP(regs) = segoffs & 0xffff;
427 clear_TF(regs);
428 clear_IF(regs);
429 return;
431 cannot_handle:
432 return_to_32bit(regs, VM86_INTx + (i << 8));
435 /* This must be called with the kernel lock held. */
436 int handle_vm86_trap(struct kernel_vm86_regs * regs, long error_code, int trapno)
438 if (VMPI.is_vm86pus) {
439 if ( (trapno==3) || (trapno==1) )
440 return_to_32bit(regs, VM86_TRAP + (trapno << 8));
441 do_int(regs, trapno, (unsigned char *) (regs->ss << 4), SP(regs));
442 return 0;
444 if (trapno !=1)
445 return 1; /* we let this handle by the calling routine */
446 if (current->flags & PF_PTRACED) {
447 unsigned long flags;
448 spin_lock_irqsave(&current->sigmask_lock, flags);
449 sigdelset(&current->blocked, SIGTRAP);
450 recalc_sigpending(current);
451 spin_unlock_irqrestore(&current->sigmask_lock, flags);
453 send_sig(SIGTRAP, current, 1);
454 current->thread.trap_no = trapno;
455 current->thread.error_code = error_code;
456 return 0;
459 /* This must be called with the kernel lock held. */
460 void handle_vm86_fault(struct kernel_vm86_regs * regs, long error_code)
462 unsigned char *csp, *ssp;
463 unsigned long ip, sp;
465 #define CHECK_IF_IN_TRAP \
466 if (VMPI.vm86dbg_active && VMPI.vm86dbg_TFpendig) \
467 pushw(ssp,sp,popw(ssp,sp) | TF_MASK);
468 #define VM86_FAULT_RETURN \
469 if (VMPI.force_return_for_pic && (VEFLAGS & IF_MASK)) \
470 return_to_32bit(regs, VM86_PICRETURN); \
471 return;
473 csp = (unsigned char *) (regs->cs << 4);
474 ssp = (unsigned char *) (regs->ss << 4);
475 sp = SP(regs);
476 ip = IP(regs);
478 switch (popb(csp, ip)) {
480 /* operand size override */
481 case 0x66:
482 switch (popb(csp, ip)) {
484 /* pushfd */
485 case 0x9c:
486 SP(regs) -= 4;
487 IP(regs) += 2;
488 pushl(ssp, sp, get_vflags(regs));
489 VM86_FAULT_RETURN;
491 /* popfd */
492 case 0x9d:
493 SP(regs) += 4;
494 IP(regs) += 2;
495 CHECK_IF_IN_TRAP
496 set_vflags_long(popl(ssp, sp), regs);
497 VM86_FAULT_RETURN;
499 /* iretd */
500 case 0xcf:
501 SP(regs) += 12;
502 IP(regs) = (unsigned short)popl(ssp, sp);
503 regs->cs = (unsigned short)popl(ssp, sp);
504 CHECK_IF_IN_TRAP
505 set_vflags_long(popl(ssp, sp), regs);
506 VM86_FAULT_RETURN;
507 /* need this to avoid a fallthrough */
508 default:
509 return_to_32bit(regs, VM86_UNKNOWN);
512 /* pushf */
513 case 0x9c:
514 SP(regs) -= 2;
515 IP(regs)++;
516 pushw(ssp, sp, get_vflags(regs));
517 VM86_FAULT_RETURN;
519 /* popf */
520 case 0x9d:
521 SP(regs) += 2;
522 IP(regs)++;
523 CHECK_IF_IN_TRAP
524 set_vflags_short(popw(ssp, sp), regs);
525 VM86_FAULT_RETURN;
527 /* int xx */
528 case 0xcd: {
529 int intno=popb(csp, ip);
530 IP(regs) += 2;
531 if (VMPI.vm86dbg_active) {
532 if ( (1 << (intno &7)) & VMPI.vm86dbg_intxxtab[intno >> 3] )
533 return_to_32bit(regs, VM86_INTx + (intno << 8));
535 do_int(regs, intno, ssp, sp);
536 return;
539 /* iret */
540 case 0xcf:
541 SP(regs) += 6;
542 IP(regs) = popw(ssp, sp);
543 regs->cs = popw(ssp, sp);
544 CHECK_IF_IN_TRAP
545 set_vflags_short(popw(ssp, sp), regs);
546 VM86_FAULT_RETURN;
548 /* cli */
549 case 0xfa:
550 IP(regs)++;
551 clear_IF(regs);
552 VM86_FAULT_RETURN;
554 /* sti */
556 * Damn. This is incorrect: the 'sti' instruction should actually
557 * enable interrupts after the /next/ instruction. Not good.
559 * Probably needs some horsing around with the TF flag. Aiee..
561 case 0xfb:
562 IP(regs)++;
563 set_IF(regs);
564 VM86_FAULT_RETURN;
566 default:
567 return_to_32bit(regs, VM86_UNKNOWN);
571 /* ---------------- vm86 special IRQ passing stuff ----------------- */
573 #define VM86_IRQNAME "vm86irq"
575 static struct vm86_irqs {
576 struct task_struct *tsk;
577 int sig;
578 } vm86_irqs[16] = {{0},};
579 static int irqbits=0;
581 #define ALLOWED_SIGS ( 1 /* 0 = don't send a signal */ \
582 | (1 << SIGUSR1) | (1 << SIGUSR2) | (1 << SIGIO) | (1 << SIGURG) \
583 | (1 << SIGUNUSED) )
585 static void irq_handler(int intno, void *dev_id, struct pt_regs * regs) {
586 int irq_bit;
587 unsigned long flags;
589 lock_kernel();
590 save_flags(flags);
591 cli();
592 irq_bit = 1 << intno;
593 if ((irqbits & irq_bit) || ! vm86_irqs[intno].tsk)
594 goto out;
595 irqbits |= irq_bit;
596 if (vm86_irqs[intno].sig)
597 send_sig(vm86_irqs[intno].sig, vm86_irqs[intno].tsk, 1);
598 /* else user will poll for IRQs */
599 out:
600 restore_flags(flags);
601 unlock_kernel();
604 static inline void free_vm86_irq(int irqnumber)
606 free_irq(irqnumber,0);
607 vm86_irqs[irqnumber].tsk = 0;
608 irqbits &= ~(1 << irqnumber);
611 static inline int task_valid(struct task_struct *tsk)
613 struct task_struct *p;
614 int ret = 0;
616 read_lock(&tasklist_lock);
617 for_each_task(p) {
618 if ((p == tsk) && (p->sig)) {
619 ret = 1;
620 break;
623 read_unlock(&tasklist_lock);
624 return ret;
627 static inline void handle_irq_zombies(void)
629 int i;
630 for (i=3; i<16; i++) {
631 if (vm86_irqs[i].tsk) {
632 if (task_valid(vm86_irqs[i].tsk)) continue;
633 free_vm86_irq(i);
638 static inline int get_and_reset_irq(int irqnumber)
640 int bit;
641 unsigned long flags;
643 if ( (irqnumber<3) || (irqnumber>15) ) return 0;
644 if (vm86_irqs[irqnumber].tsk != current) return 0;
645 save_flags(flags);
646 cli();
647 bit = irqbits & (1 << irqnumber);
648 irqbits &= ~bit;
649 restore_flags(flags);
650 return bit;
654 static int do_vm86_irq_handling(int subfunction, int irqnumber)
656 int ret;
657 switch (subfunction) {
658 case VM86_GET_AND_RESET_IRQ: {
659 return get_and_reset_irq(irqnumber);
661 case VM86_GET_IRQ_BITS: {
662 return irqbits;
664 case VM86_REQUEST_IRQ: {
665 int sig = irqnumber >> 8;
666 int irq = irqnumber & 255;
667 handle_irq_zombies();
668 if (!capable(CAP_SYS_ADMIN)) return -EPERM;
669 if (!((1 << sig) & ALLOWED_SIGS)) return -EPERM;
670 if ( (irq<3) || (irq>15) ) return -EPERM;
671 if (vm86_irqs[irq].tsk) return -EPERM;
672 ret = request_irq(irq, &irq_handler, 0, VM86_IRQNAME, 0);
673 if (ret) return ret;
674 vm86_irqs[irq].sig = sig;
675 vm86_irqs[irq].tsk = current;
676 return irq;
678 case VM86_FREE_IRQ: {
679 handle_irq_zombies();
680 if ( (irqnumber<3) || (irqnumber>15) ) return -EPERM;
681 if (!vm86_irqs[irqnumber].tsk) return 0;
682 if (vm86_irqs[irqnumber].tsk != current) return -EPERM;
683 free_vm86_irq(irqnumber);
684 return 0;
687 return -EINVAL;