Ok. I didn't make 2.4.0 in 2000. Tough. I tried, but we had some
[davej-history.git] / arch / i386 / kernel / vm86.c
blob15b20d6956f7a50109cc42e53bbad3fc21a73400
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/pgalloc.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 if (!current->thread.vm86_info) {
73 printk("no vm86_info: BAD\n");
74 do_exit(SIGSEGV);
76 set_flags(regs->eflags, VEFLAGS, VIF_MASK | current->thread.v86mask);
77 tmp = copy_to_user(&current->thread.vm86_info->regs,regs, VM86_REGS_SIZE1);
78 tmp += copy_to_user(&current->thread.vm86_info->regs.VM86_REGS_PART2,
79 &regs->VM86_REGS_PART2, VM86_REGS_SIZE2);
80 tmp += put_user(current->thread.screen_bitmap,&current->thread.vm86_info->screen_bitmap);
81 if (tmp) {
82 printk("vm86: could not access userspace vm86_info\n");
83 do_exit(SIGSEGV);
85 tss = init_tss + smp_processor_id();
86 tss->esp0 = current->thread.esp0 = current->thread.saved_esp0;
87 current->thread.saved_esp0 = 0;
88 ret = KVM86->regs32;
89 return ret;
92 static void mark_screen_rdonly(struct task_struct * tsk)
94 pgd_t *pgd;
95 pmd_t *pmd;
96 pte_t *pte;
97 int i;
99 pgd = pgd_offset(tsk->mm, 0xA0000);
100 if (pgd_none(*pgd))
101 return;
102 if (pgd_bad(*pgd)) {
103 pgd_ERROR(*pgd);
104 pgd_clear(pgd);
105 return;
107 pmd = pmd_offset(pgd, 0xA0000);
108 if (pmd_none(*pmd))
109 return;
110 if (pmd_bad(*pmd)) {
111 pmd_ERROR(*pmd);
112 pmd_clear(pmd);
113 return;
115 pte = pte_offset(pmd, 0xA0000);
116 for (i = 0; i < 32; i++) {
117 if (pte_present(*pte))
118 set_pte(pte, pte_wrprotect(*pte));
119 pte++;
121 flush_tlb();
126 static int do_vm86_irq_handling(int subfunction, int irqnumber);
127 static void do_sys_vm86(struct kernel_vm86_struct *info, struct task_struct *tsk);
129 asmlinkage int sys_vm86old(struct vm86_struct * v86)
131 struct kernel_vm86_struct info; /* declare this _on top_,
132 * this avoids wasting of stack space.
133 * This remains on the stack until we
134 * return to 32 bit user space.
136 struct task_struct *tsk;
137 int tmp, ret = -EPERM;
139 tsk = current;
140 if (tsk->thread.saved_esp0)
141 goto out;
142 tmp = copy_from_user(&info, v86, VM86_REGS_SIZE1);
143 tmp += copy_from_user(&info.regs.VM86_REGS_PART2, &v86->regs.VM86_REGS_PART2,
144 (long)&info.vm86plus - (long)&info.regs.VM86_REGS_PART2);
145 ret = -EFAULT;
146 if (tmp)
147 goto out;
148 memset(&info.vm86plus, 0, (int)&info.regs32 - (int)&info.vm86plus);
149 info.regs32 = (struct pt_regs *) &v86;
150 tsk->thread.vm86_info = v86;
151 do_sys_vm86(&info, tsk);
152 ret = 0; /* we never return here */
153 out:
154 return ret;
158 asmlinkage int sys_vm86(unsigned long subfunction, struct vm86plus_struct * v86)
160 struct kernel_vm86_struct info; /* declare this _on top_,
161 * this avoids wasting of stack space.
162 * This remains on the stack until we
163 * return to 32 bit user space.
165 struct task_struct *tsk;
166 int tmp, ret;
168 tsk = current;
169 switch (subfunction) {
170 case VM86_REQUEST_IRQ:
171 case VM86_FREE_IRQ:
172 case VM86_GET_IRQ_BITS:
173 case VM86_GET_AND_RESET_IRQ:
174 ret = do_vm86_irq_handling(subfunction,(int)v86);
175 goto out;
176 case VM86_PLUS_INSTALL_CHECK:
177 /* NOTE: on old vm86 stuff this will return the error
178 from verify_area(), because the subfunction is
179 interpreted as (invalid) address to vm86_struct.
180 So the installation check works.
182 ret = 0;
183 goto out;
186 /* we come here only for functions VM86_ENTER, VM86_ENTER_NO_BYPASS */
187 ret = -EPERM;
188 if (tsk->thread.saved_esp0)
189 goto out;
190 tmp = copy_from_user(&info, v86, VM86_REGS_SIZE1);
191 tmp += copy_from_user(&info.regs.VM86_REGS_PART2, &v86->regs.VM86_REGS_PART2,
192 (long)&info.regs32 - (long)&info.regs.VM86_REGS_PART2);
193 ret = -EFAULT;
194 if (tmp)
195 goto out;
196 info.regs32 = (struct pt_regs *) &subfunction;
197 info.vm86plus.is_vm86pus = 1;
198 tsk->thread.vm86_info = (struct vm86_struct *)v86;
199 do_sys_vm86(&info, tsk);
200 ret = 0; /* we never return here */
201 out:
202 return ret;
206 static void do_sys_vm86(struct kernel_vm86_struct *info, struct task_struct *tsk)
208 struct tss_struct *tss;
210 * make sure the vm86() system call doesn't try to do anything silly
212 info->regs.__null_ds = 0;
213 info->regs.__null_es = 0;
215 /* we are clearing fs,gs later just before "jmp ret_from_sys_call",
216 * because starting with Linux 2.1.x they aren't no longer saved/restored
220 * The eflags register is also special: we cannot trust that the user
221 * has set it up safely, so this makes sure interrupt etc flags are
222 * inherited from protected mode.
224 VEFLAGS = info->regs.eflags;
225 info->regs.eflags &= SAFE_MASK;
226 info->regs.eflags |= info->regs32->eflags & ~SAFE_MASK;
227 info->regs.eflags |= VM_MASK;
229 switch (info->cpu_type) {
230 case CPU_286:
231 tsk->thread.v86mask = 0;
232 break;
233 case CPU_386:
234 tsk->thread.v86mask = NT_MASK | IOPL_MASK;
235 break;
236 case CPU_486:
237 tsk->thread.v86mask = AC_MASK | NT_MASK | IOPL_MASK;
238 break;
239 default:
240 tsk->thread.v86mask = ID_MASK | AC_MASK | NT_MASK | IOPL_MASK;
241 break;
245 * Save old state, set default return value (%eax) to 0
247 info->regs32->eax = 0;
248 tsk->thread.saved_esp0 = tsk->thread.esp0;
249 tss = init_tss + smp_processor_id();
250 tss->esp0 = tsk->thread.esp0 = (unsigned long) &info->VM86_TSS_ESP0;
252 tsk->thread.screen_bitmap = info->screen_bitmap;
253 if (info->flags & VM86_SCREEN_BITMAP)
254 mark_screen_rdonly(tsk);
255 __asm__ __volatile__(
256 "xorl %%eax,%%eax; movl %%eax,%%fs; movl %%eax,%%gs\n\t"
257 "movl %0,%%esp\n\t"
258 "jmp ret_from_sys_call"
259 : /* no outputs */
260 :"r" (&info->regs), "b" (tsk) : "ax");
261 /* we never return here */
264 static inline void return_to_32bit(struct kernel_vm86_regs * regs16, int retval)
266 struct pt_regs * regs32;
268 regs32 = save_v86_state(regs16);
269 regs32->eax = retval;
270 __asm__ __volatile__("movl %0,%%esp\n\t"
271 "jmp ret_from_sys_call"
272 : : "r" (regs32), "b" (current));
275 static inline void set_IF(struct kernel_vm86_regs * regs)
277 VEFLAGS |= VIF_MASK;
278 if (VEFLAGS & VIP_MASK)
279 return_to_32bit(regs, VM86_STI);
282 static inline void clear_IF(struct kernel_vm86_regs * regs)
284 VEFLAGS &= ~VIF_MASK;
287 static inline void clear_TF(struct kernel_vm86_regs * regs)
289 regs->eflags &= ~TF_MASK;
292 static inline void set_vflags_long(unsigned long eflags, struct kernel_vm86_regs * regs)
294 set_flags(VEFLAGS, eflags, current->thread.v86mask);
295 set_flags(regs->eflags, eflags, SAFE_MASK);
296 if (eflags & IF_MASK)
297 set_IF(regs);
300 static inline void set_vflags_short(unsigned short flags, struct kernel_vm86_regs * regs)
302 set_flags(VFLAGS, flags, current->thread.v86mask);
303 set_flags(regs->eflags, flags, SAFE_MASK);
304 if (flags & IF_MASK)
305 set_IF(regs);
308 static inline unsigned long get_vflags(struct kernel_vm86_regs * regs)
310 unsigned long flags = regs->eflags & RETURN_MASK;
312 if (VEFLAGS & VIF_MASK)
313 flags |= IF_MASK;
314 return flags | (VEFLAGS & current->thread.v86mask);
317 static inline int is_revectored(int nr, struct revectored_struct * bitmap)
319 __asm__ __volatile__("btl %2,%1\n\tsbbl %0,%0"
320 :"=r" (nr)
321 :"m" (*bitmap),"r" (nr));
322 return nr;
326 * Boy are these ugly, but we need to do the correct 16-bit arithmetic.
327 * Gcc makes a mess of it, so we do it inline and use non-obvious calling
328 * conventions..
330 #define pushb(base, ptr, val) \
331 __asm__ __volatile__( \
332 "decw %w0\n\t" \
333 "movb %2,0(%1,%0)" \
334 : "=r" (ptr) \
335 : "r" (base), "q" (val), "0" (ptr))
337 #define pushw(base, ptr, val) \
338 __asm__ __volatile__( \
339 "decw %w0\n\t" \
340 "movb %h2,0(%1,%0)\n\t" \
341 "decw %w0\n\t" \
342 "movb %b2,0(%1,%0)" \
343 : "=r" (ptr) \
344 : "r" (base), "q" (val), "0" (ptr))
346 #define pushl(base, ptr, val) \
347 __asm__ __volatile__( \
348 "decw %w0\n\t" \
349 "rorl $16,%2\n\t" \
350 "movb %h2,0(%1,%0)\n\t" \
351 "decw %w0\n\t" \
352 "movb %b2,0(%1,%0)\n\t" \
353 "decw %w0\n\t" \
354 "rorl $16,%2\n\t" \
355 "movb %h2,0(%1,%0)\n\t" \
356 "decw %w0\n\t" \
357 "movb %b2,0(%1,%0)" \
358 : "=r" (ptr) \
359 : "r" (base), "q" (val), "0" (ptr))
361 #define popb(base, ptr) \
362 ({ unsigned long __res; \
363 __asm__ __volatile__( \
364 "movb 0(%1,%0),%b2\n\t" \
365 "incw %w0" \
366 : "=r" (ptr), "=r" (base), "=q" (__res) \
367 : "0" (ptr), "1" (base), "2" (0)); \
368 __res; })
370 #define popw(base, ptr) \
371 ({ unsigned long __res; \
372 __asm__ __volatile__( \
373 "movb 0(%1,%0),%b2\n\t" \
374 "incw %w0\n\t" \
375 "movb 0(%1,%0),%h2\n\t" \
376 "incw %w0" \
377 : "=r" (ptr), "=r" (base), "=q" (__res) \
378 : "0" (ptr), "1" (base), "2" (0)); \
379 __res; })
381 #define popl(base, ptr) \
382 ({ unsigned long __res; \
383 __asm__ __volatile__( \
384 "movb 0(%1,%0),%b2\n\t" \
385 "incw %w0\n\t" \
386 "movb 0(%1,%0),%h2\n\t" \
387 "incw %w0\n\t" \
388 "rorl $16,%2\n\t" \
389 "movb 0(%1,%0),%b2\n\t" \
390 "incw %w0\n\t" \
391 "movb 0(%1,%0),%h2\n\t" \
392 "incw %w0\n\t" \
393 "rorl $16,%2" \
394 : "=r" (ptr), "=r" (base), "=q" (__res) \
395 : "0" (ptr), "1" (base)); \
396 __res; })
398 static void do_int(struct kernel_vm86_regs *regs, int i, unsigned char * ssp, unsigned long sp)
400 unsigned long *intr_ptr, segoffs;
402 if (regs->cs == BIOSSEG)
403 goto cannot_handle;
404 if (is_revectored(i, &KVM86->int_revectored))
405 goto cannot_handle;
406 if (i==0x21 && is_revectored(AH(regs),&KVM86->int21_revectored))
407 goto cannot_handle;
408 intr_ptr = (unsigned long *) (i << 2);
409 if (get_user(segoffs, intr_ptr))
410 goto cannot_handle;
411 if ((segoffs >> 16) == BIOSSEG)
412 goto cannot_handle;
413 pushw(ssp, sp, get_vflags(regs));
414 pushw(ssp, sp, regs->cs);
415 pushw(ssp, sp, IP(regs));
416 regs->cs = segoffs >> 16;
417 SP(regs) -= 6;
418 IP(regs) = segoffs & 0xffff;
419 clear_TF(regs);
420 clear_IF(regs);
421 return;
423 cannot_handle:
424 return_to_32bit(regs, VM86_INTx + (i << 8));
427 int handle_vm86_trap(struct kernel_vm86_regs * regs, long error_code, int trapno)
429 if (VMPI.is_vm86pus) {
430 if ( (trapno==3) || (trapno==1) )
431 return_to_32bit(regs, VM86_TRAP + (trapno << 8));
432 do_int(regs, trapno, (unsigned char *) (regs->ss << 4), SP(regs));
433 return 0;
435 if (trapno !=1)
436 return 1; /* we let this handle by the calling routine */
437 if (current->ptrace & PT_PTRACED) {
438 unsigned long flags;
439 spin_lock_irqsave(&current->sigmask_lock, flags);
440 sigdelset(&current->blocked, SIGTRAP);
441 recalc_sigpending(current);
442 spin_unlock_irqrestore(&current->sigmask_lock, flags);
444 send_sig(SIGTRAP, current, 1);
445 current->thread.trap_no = trapno;
446 current->thread.error_code = error_code;
447 return 0;
450 void handle_vm86_fault(struct kernel_vm86_regs * regs, long error_code)
452 unsigned char *csp, *ssp;
453 unsigned long ip, sp;
455 #define CHECK_IF_IN_TRAP \
456 if (VMPI.vm86dbg_active && VMPI.vm86dbg_TFpendig) \
457 pushw(ssp,sp,popw(ssp,sp) | TF_MASK);
458 #define VM86_FAULT_RETURN \
459 if (VMPI.force_return_for_pic && (VEFLAGS & IF_MASK)) \
460 return_to_32bit(regs, VM86_PICRETURN); \
461 return;
463 csp = (unsigned char *) (regs->cs << 4);
464 ssp = (unsigned char *) (regs->ss << 4);
465 sp = SP(regs);
466 ip = IP(regs);
468 switch (popb(csp, ip)) {
470 /* operand size override */
471 case 0x66:
472 switch (popb(csp, ip)) {
474 /* pushfd */
475 case 0x9c:
476 SP(regs) -= 4;
477 IP(regs) += 2;
478 pushl(ssp, sp, get_vflags(regs));
479 VM86_FAULT_RETURN;
481 /* popfd */
482 case 0x9d:
483 SP(regs) += 4;
484 IP(regs) += 2;
485 CHECK_IF_IN_TRAP
486 set_vflags_long(popl(ssp, sp), regs);
487 VM86_FAULT_RETURN;
489 /* iretd */
490 case 0xcf:
491 SP(regs) += 12;
492 IP(regs) = (unsigned short)popl(ssp, sp);
493 regs->cs = (unsigned short)popl(ssp, sp);
494 CHECK_IF_IN_TRAP
495 set_vflags_long(popl(ssp, sp), regs);
496 VM86_FAULT_RETURN;
497 /* need this to avoid a fallthrough */
498 default:
499 return_to_32bit(regs, VM86_UNKNOWN);
502 /* pushf */
503 case 0x9c:
504 SP(regs) -= 2;
505 IP(regs)++;
506 pushw(ssp, sp, get_vflags(regs));
507 VM86_FAULT_RETURN;
509 /* popf */
510 case 0x9d:
511 SP(regs) += 2;
512 IP(regs)++;
513 CHECK_IF_IN_TRAP
514 set_vflags_short(popw(ssp, sp), regs);
515 VM86_FAULT_RETURN;
517 /* int xx */
518 case 0xcd: {
519 int intno=popb(csp, ip);
520 IP(regs) += 2;
521 if (VMPI.vm86dbg_active) {
522 if ( (1 << (intno &7)) & VMPI.vm86dbg_intxxtab[intno >> 3] )
523 return_to_32bit(regs, VM86_INTx + (intno << 8));
525 do_int(regs, intno, ssp, sp);
526 return;
529 /* iret */
530 case 0xcf:
531 SP(regs) += 6;
532 IP(regs) = popw(ssp, sp);
533 regs->cs = popw(ssp, sp);
534 CHECK_IF_IN_TRAP
535 set_vflags_short(popw(ssp, sp), regs);
536 VM86_FAULT_RETURN;
538 /* cli */
539 case 0xfa:
540 IP(regs)++;
541 clear_IF(regs);
542 VM86_FAULT_RETURN;
544 /* sti */
546 * Damn. This is incorrect: the 'sti' instruction should actually
547 * enable interrupts after the /next/ instruction. Not good.
549 * Probably needs some horsing around with the TF flag. Aiee..
551 case 0xfb:
552 IP(regs)++;
553 set_IF(regs);
554 VM86_FAULT_RETURN;
556 default:
557 return_to_32bit(regs, VM86_UNKNOWN);
561 /* ---------------- vm86 special IRQ passing stuff ----------------- */
563 #define VM86_IRQNAME "vm86irq"
565 static struct vm86_irqs {
566 struct task_struct *tsk;
567 int sig;
568 } vm86_irqs[16];
569 static int irqbits;
571 #define ALLOWED_SIGS ( 1 /* 0 = don't send a signal */ \
572 | (1 << SIGUSR1) | (1 << SIGUSR2) | (1 << SIGIO) | (1 << SIGURG) \
573 | (1 << SIGUNUSED) )
575 static void irq_handler(int intno, void *dev_id, struct pt_regs * regs) {
576 int irq_bit;
577 unsigned long flags;
579 save_flags(flags);
580 cli();
581 irq_bit = 1 << intno;
582 if ((irqbits & irq_bit) || ! vm86_irqs[intno].tsk)
583 goto out;
584 irqbits |= irq_bit;
585 if (vm86_irqs[intno].sig)
586 send_sig(vm86_irqs[intno].sig, vm86_irqs[intno].tsk, 1);
587 /* else user will poll for IRQs */
588 out:
589 restore_flags(flags);
592 static inline void free_vm86_irq(int irqnumber)
594 free_irq(irqnumber,0);
595 vm86_irqs[irqnumber].tsk = 0;
596 irqbits &= ~(1 << irqnumber);
599 static inline int task_valid(struct task_struct *tsk)
601 struct task_struct *p;
602 int ret = 0;
604 read_lock(&tasklist_lock);
605 for_each_task(p) {
606 if ((p == tsk) && (p->sig)) {
607 ret = 1;
608 break;
611 read_unlock(&tasklist_lock);
612 return ret;
615 static inline void handle_irq_zombies(void)
617 int i;
618 for (i=3; i<16; i++) {
619 if (vm86_irqs[i].tsk) {
620 if (task_valid(vm86_irqs[i].tsk)) continue;
621 free_vm86_irq(i);
626 static inline int get_and_reset_irq(int irqnumber)
628 int bit;
629 unsigned long flags;
631 if ( (irqnumber<3) || (irqnumber>15) ) return 0;
632 if (vm86_irqs[irqnumber].tsk != current) return 0;
633 save_flags(flags);
634 cli();
635 bit = irqbits & (1 << irqnumber);
636 irqbits &= ~bit;
637 restore_flags(flags);
638 return bit;
642 static int do_vm86_irq_handling(int subfunction, int irqnumber)
644 int ret;
645 switch (subfunction) {
646 case VM86_GET_AND_RESET_IRQ: {
647 return get_and_reset_irq(irqnumber);
649 case VM86_GET_IRQ_BITS: {
650 return irqbits;
652 case VM86_REQUEST_IRQ: {
653 int sig = irqnumber >> 8;
654 int irq = irqnumber & 255;
655 handle_irq_zombies();
656 if (!capable(CAP_SYS_ADMIN)) return -EPERM;
657 if (!((1 << sig) & ALLOWED_SIGS)) return -EPERM;
658 if ( (irq<3) || (irq>15) ) return -EPERM;
659 if (vm86_irqs[irq].tsk) return -EPERM;
660 ret = request_irq(irq, &irq_handler, 0, VM86_IRQNAME, 0);
661 if (ret) return ret;
662 vm86_irqs[irq].sig = sig;
663 vm86_irqs[irq].tsk = current;
664 return irq;
666 case VM86_FREE_IRQ: {
667 handle_irq_zombies();
668 if ( (irqnumber<3) || (irqnumber>15) ) return -EPERM;
669 if (!vm86_irqs[irqnumber].tsk) return 0;
670 if (vm86_irqs[irqnumber].tsk != current) return -EPERM;
671 free_vm86_irq(irqnumber);
672 return 0;
675 return -EINVAL;