1 /*
2  *  linux/kernel/vm86.c
3  *
4  *  Copyright (C) 1994  Linus Torvalds
5  *
6  *  29 dec 2001 - Fixed oopses caused by unchecked access to the vm86
7  *                stack - Manfred Spraul <manfreds@colorfullife.com>
8  *
9  *  22 mar 2002 - Manfred detected the stackfaults, but didn't handle
10  *                them correctly. Now the emulation will be in a
11  *                consistent state after stackfaults - Kasper Dupont
12  *                <kasperd@daimi.au.dk>
13  *
14  *  22 mar 2002 - Added missing clear_IF in set_vflags_* Kasper Dupont
15  *                <kasperd@daimi.au.dk>
16  *
17  *  ?? ??? 2002 - Fixed premature returns from handle_vm86_fault
18  *                caused by Kasper Dupont's changes - Stas Sergeev
19  *
20  *   4 apr 2002 - Fixed CHECK_IF_IN_TRAP broken by Stas' changes.
21  *                Kasper Dupont <kasperd@daimi.au.dk>
22  *
23  *   9 apr 2002 - Changed syntax of macros in handle_vm86_fault.
24  *                Kasper Dupont <kasperd@daimi.au.dk>
25  *
26  *   9 apr 2002 - Changed stack access macros to jump to a label
27  *                instead of returning to userspace. This simplifies
28  *                do_int, and is needed by handle_vm6_fault. Kasper
29  *                Dupont <kasperd@daimi.au.dk>
30  *
31  */
32 
33 #include <linux/errno.h>
34 #include <linux/sched.h>
35 #include <linux/kernel.h>
36 #include <linux/signal.h>
37 #include <linux/string.h>
38 #include <linux/ptrace.h>
39 #include <linux/mm.h>
40 #include <linux/smp.h>
41 #include <linux/smp_lock.h>
42 
43 #include <asm/uaccess.h>
44 #include <asm/pgalloc.h>
45 #include <asm/io.h>
46 #include <asm/irq.h>
47 
48 /*
49  * Known problems:
50  *
51  * Interrupt handling is not guaranteed:
52  * - a real x86 will disable all interrupts for one instruction
53  *   after a "mov ss,xx" to make stack handling atomic even without
54  *   the 'lss' instruction. We can't guarantee this in v86 mode,
55  *   as the next instruction might result in a page fault or similar.
56  * - a real x86 will have interrupts disabled for one instruction
57  *   past the 'sti' that enables them. We don't bother with all the
58  *   details yet.
59  *
60  * Let's hope these problems do not actually matter for anything.
61  */
62 
63 
64 #define KVM86	((struct kernel_vm86_struct *)regs)
65 #define VMPI 	KVM86->vm86plus
66 
67 
68 /*
69  * 8- and 16-bit register defines..
70  */
71 #define AL(regs)	(((unsigned char *)&((regs)->eax))[0])
72 #define AH(regs)	(((unsigned char *)&((regs)->eax))[1])
73 #define IP(regs)	(*(unsigned short *)&((regs)->eip))
74 #define SP(regs)	(*(unsigned short *)&((regs)->esp))
75 
76 /*
77  * virtual flags (16 and 32-bit versions)
78  */
79 #define VFLAGS	(*(unsigned short *)&(current->thread.v86flags))
80 #define VEFLAGS	(current->thread.v86flags)
81 
82 #define set_flags(X,new,mask) \
83 ((X) = ((X) & ~(mask)) | ((new) & (mask)))
84 
85 #define SAFE_MASK	(0xDD5)
86 #define RETURN_MASK	(0xDFF)
87 
88 #define VM86_REGS_PART2 orig_eax
89 #define VM86_REGS_SIZE1 \
90         ( (unsigned)( & (((struct kernel_vm86_regs *)0)->VM86_REGS_PART2) ) )
91 #define VM86_REGS_SIZE2 (sizeof(struct kernel_vm86_regs) - VM86_REGS_SIZE1)
92 
93 struct pt_regs * FASTCALL(save_v86_state(struct kernel_vm86_regs * regs));
save_v86_state(struct kernel_vm86_regs * regs)94 struct pt_regs * fastcall save_v86_state(struct kernel_vm86_regs * regs)
95 {
96 	struct tss_struct *tss;
97 	struct pt_regs *ret;
98 	unsigned long tmp;
99 
100 	if (!current->thread.vm86_info) {
101 		printk("no vm86_info: BAD\n");
102 		do_exit(SIGSEGV);
103 	}
104 	set_flags(regs->eflags, VEFLAGS, VIF_MASK | current->thread.v86mask);
105 	tmp = copy_to_user(&current->thread.vm86_info->regs,regs, VM86_REGS_SIZE1);
106 	tmp += copy_to_user(&current->thread.vm86_info->regs.VM86_REGS_PART2,
107 		&regs->VM86_REGS_PART2, VM86_REGS_SIZE2);
108 	tmp += put_user(current->thread.screen_bitmap,&current->thread.vm86_info->screen_bitmap);
109 	if (tmp) {
110 		printk("vm86: could not access userspace vm86_info\n");
111 		do_exit(SIGSEGV);
112 	}
113 	tss = init_tss + smp_processor_id();
114 	tss->esp0 = current->thread.esp0 = current->thread.saved_esp0;
115 	current->thread.saved_esp0 = 0;
116 	ret = KVM86->regs32;
117 	return ret;
118 }
119 
mark_screen_rdonly(struct task_struct * tsk)120 static void mark_screen_rdonly(struct task_struct * tsk)
121 {
122 	pgd_t *pgd;
123 	pmd_t *pmd;
124 	pte_t *pte;
125 	int i;
126 
127 	spin_lock(&tsk->mm->page_table_lock);
128 	pgd = pgd_offset(tsk->mm, 0xA0000);
129 	if (pgd_none(*pgd))
130 		goto out;
131 	if (pgd_bad(*pgd)) {
132 		pgd_ERROR(*pgd);
133 		pgd_clear(pgd);
134 		goto out;
135 	}
136 	pmd = pmd_offset(pgd, 0xA0000);
137 	if (pmd_none(*pmd))
138 		goto out;
139 	if (pmd_bad(*pmd)) {
140 		pmd_ERROR(*pmd);
141 		pmd_clear(pmd);
142 		goto out;
143 	}
144 	pte = pte_offset(pmd, 0xA0000);
145 	for (i = 0; i < 32; i++) {
146 		if (pte_present(*pte))
147 			set_pte(pte, pte_wrprotect(*pte));
148 		pte++;
149 	}
150 out:
151 	spin_unlock(&tsk->mm->page_table_lock);
152 	flush_tlb();
153 }
154 
155 
156 
157 static int do_vm86_irq_handling(int subfunction, int irqnumber);
158 static void do_sys_vm86(struct kernel_vm86_struct *info, struct task_struct *tsk);
159 
sys_vm86old(struct vm86_struct * v86)160 asmlinkage int sys_vm86old(struct vm86_struct * v86)
161 {
162 	struct kernel_vm86_struct info; /* declare this _on top_,
163 					 * this avoids wasting of stack space.
164 					 * This remains on the stack until we
165 					 * return to 32 bit user space.
166 					 */
167 	struct task_struct *tsk;
168 	int tmp, ret = -EPERM;
169 
170 	tsk = current;
171 	if (tsk->thread.saved_esp0)
172 		goto out;
173 	tmp  = copy_from_user(&info, v86, VM86_REGS_SIZE1);
174 	tmp += copy_from_user(&info.regs.VM86_REGS_PART2, &v86->regs.VM86_REGS_PART2,
175 		(long)&info.vm86plus - (long)&info.regs.VM86_REGS_PART2);
176 	ret = -EFAULT;
177 	if (tmp)
178 		goto out;
179 	memset(&info.vm86plus, 0, (int)&info.regs32 - (int)&info.vm86plus);
180 	info.regs32 = (struct pt_regs *) &v86;
181 	tsk->thread.vm86_info = v86;
182 	do_sys_vm86(&info, tsk);
183 	ret = 0;	/* we never return here */
184 out:
185 	return ret;
186 }
187 
188 
sys_vm86(unsigned long subfunction,struct vm86plus_struct * v86)189 asmlinkage int sys_vm86(unsigned long subfunction, struct vm86plus_struct * v86)
190 {
191 	struct kernel_vm86_struct info; /* declare this _on top_,
192 					 * this avoids wasting of stack space.
193 					 * This remains on the stack until we
194 					 * return to 32 bit user space.
195 					 */
196 	struct task_struct *tsk;
197 	int tmp, ret;
198 
199 	tsk = current;
200 	switch (subfunction) {
201 		case VM86_REQUEST_IRQ:
202 		case VM86_FREE_IRQ:
203 		case VM86_GET_IRQ_BITS:
204 		case VM86_GET_AND_RESET_IRQ:
205 			ret = do_vm86_irq_handling(subfunction,(int)v86);
206 			goto out;
207 		case VM86_PLUS_INSTALL_CHECK:
208 			/* NOTE: on old vm86 stuff this will return the error
209 			   from verify_area(), because the subfunction is
210 			   interpreted as (invalid) address to vm86_struct.
211 			   So the installation check works.
212 			 */
213 			ret = 0;
214 			goto out;
215 	}
216 
217 	/* we come here only for functions VM86_ENTER, VM86_ENTER_NO_BYPASS */
218 	ret = -EPERM;
219 	if (tsk->thread.saved_esp0)
220 		goto out;
221 	tmp  = copy_from_user(&info, v86, VM86_REGS_SIZE1);
222 	tmp += copy_from_user(&info.regs.VM86_REGS_PART2, &v86->regs.VM86_REGS_PART2,
223 		(long)&info.regs32 - (long)&info.regs.VM86_REGS_PART2);
224 	ret = -EFAULT;
225 	if (tmp)
226 		goto out;
227 	info.regs32 = (struct pt_regs *) &subfunction;
228 	info.vm86plus.is_vm86pus = 1;
229 	tsk->thread.vm86_info = (struct vm86_struct *)v86;
230 	do_sys_vm86(&info, tsk);
231 	ret = 0;	/* we never return here */
232 out:
233 	return ret;
234 }
235 
236 
do_sys_vm86(struct kernel_vm86_struct * info,struct task_struct * tsk)237 static void do_sys_vm86(struct kernel_vm86_struct *info, struct task_struct *tsk)
238 {
239 	struct tss_struct *tss;
240 /*
241  * make sure the vm86() system call doesn't try to do anything silly
242  */
243 	info->regs.__null_ds = 0;
244 	info->regs.__null_es = 0;
245 
246 /* we are clearing fs,gs later just before "jmp ret_from_sys_call",
247  * because starting with Linux 2.1.x they aren't no longer saved/restored
248  */
249 
250 /*
251  * The eflags register is also special: we cannot trust that the user
252  * has set it up safely, so this makes sure interrupt etc flags are
253  * inherited from protected mode.
254  */
255  	VEFLAGS = info->regs.eflags;
256 	info->regs.eflags &= SAFE_MASK;
257 	info->regs.eflags |= info->regs32->eflags & ~SAFE_MASK;
258 	info->regs.eflags |= VM_MASK;
259 
260 	switch (info->cpu_type) {
261 		case CPU_286:
262 			tsk->thread.v86mask = 0;
263 			break;
264 		case CPU_386:
265 			tsk->thread.v86mask = NT_MASK | IOPL_MASK;
266 			break;
267 		case CPU_486:
268 			tsk->thread.v86mask = AC_MASK | NT_MASK | IOPL_MASK;
269 			break;
270 		default:
271 			tsk->thread.v86mask = ID_MASK | AC_MASK | NT_MASK | IOPL_MASK;
272 			break;
273 	}
274 
275 /*
276  * Save old state, set default return value (%eax) to 0
277  */
278 	info->regs32->eax = 0;
279 	tsk->thread.saved_esp0 = tsk->thread.esp0;
280 	tss = init_tss + smp_processor_id();
281 	tss->esp0 = tsk->thread.esp0 = (unsigned long) &info->VM86_TSS_ESP0;
282 
283 	tsk->thread.screen_bitmap = info->screen_bitmap;
284 	if (info->flags & VM86_SCREEN_BITMAP)
285 		mark_screen_rdonly(tsk);
286 	__asm__ __volatile__(
287 		"xorl %%eax,%%eax; movl %%eax,%%fs; movl %%eax,%%gs\n\t"
288 		"movl %0,%%esp\n\t"
289 		"jmp ret_from_sys_call"
290 		: /* no outputs */
291 		:"r" (&info->regs), "b" (tsk) : "ax");
292 	/* we never return here */
293 }
294 
return_to_32bit(struct kernel_vm86_regs * regs16,int retval)295 static inline void return_to_32bit(struct kernel_vm86_regs * regs16, int retval)
296 {
297 	struct pt_regs * regs32;
298 
299 	regs32 = save_v86_state(regs16);
300 	regs32->eax = retval;
301 	__asm__ __volatile__("movl %0,%%esp\n\t"
302 		"jmp ret_from_sys_call"
303 		: : "r" (regs32), "b" (current));
304 }
305 
set_IF(struct kernel_vm86_regs * regs)306 static inline void set_IF(struct kernel_vm86_regs * regs)
307 {
308 	VEFLAGS |= VIF_MASK;
309 	if (VEFLAGS & VIP_MASK)
310 		return_to_32bit(regs, VM86_STI);
311 }
312 
clear_IF(struct kernel_vm86_regs * regs)313 static inline void clear_IF(struct kernel_vm86_regs * regs)
314 {
315 	VEFLAGS &= ~VIF_MASK;
316 }
317 
clear_TF(struct kernel_vm86_regs * regs)318 static inline void clear_TF(struct kernel_vm86_regs * regs)
319 {
320 	regs->eflags &= ~TF_MASK;
321 }
322 
clear_AC(struct kernel_vm86_regs * regs)323 static inline void clear_AC(struct kernel_vm86_regs * regs)
324 {
325 	regs->eflags &= ~AC_MASK;
326 }
327 
328 /* It is correct to call set_IF(regs) from the set_vflags_*
329  * functions. However someone forgot to call clear_IF(regs)
330  * in the opposite case.
331  * After the command sequence CLI PUSHF STI POPF you should
332  * end up with interrups disabled, but you ended up with
333  * interrupts enabled.
334  *  ( I was testing my own changes, but the only bug I
335  *    could find was in a function I had not changed. )
336  * [KD]
337  */
338 
set_vflags_long(unsigned long eflags,struct kernel_vm86_regs * regs)339 static inline void set_vflags_long(unsigned long eflags, struct kernel_vm86_regs * regs)
340 {
341 	set_flags(VEFLAGS, eflags, current->thread.v86mask);
342 	set_flags(regs->eflags, eflags, SAFE_MASK);
343 	if (eflags & IF_MASK)
344 		set_IF(regs);
345 	else
346 		clear_IF(regs);
347 }
348 
set_vflags_short(unsigned short flags,struct kernel_vm86_regs * regs)349 static inline void set_vflags_short(unsigned short flags, struct kernel_vm86_regs * regs)
350 {
351 	set_flags(VFLAGS, flags, current->thread.v86mask);
352 	set_flags(regs->eflags, flags, SAFE_MASK);
353 	if (flags & IF_MASK)
354 		set_IF(regs);
355 	else
356 		clear_IF(regs);
357 }
358 
get_vflags(struct kernel_vm86_regs * regs)359 static inline unsigned long get_vflags(struct kernel_vm86_regs * regs)
360 {
361 	unsigned long flags = regs->eflags & RETURN_MASK;
362 
363 	if (VEFLAGS & VIF_MASK)
364 		flags |= IF_MASK;
365 	else
366 		flags &= ~IF_MASK;
367 	flags |= IOPL_MASK;
368 	return flags | (VEFLAGS & current->thread.v86mask);
369 }
370 
is_revectored(int nr,struct revectored_struct * bitmap)371 static inline int is_revectored(int nr, struct revectored_struct * bitmap)
372 {
373 	__asm__ __volatile__("btl %2,%1\n\tsbbl %0,%0"
374 		:"=r" (nr)
375 		:"m" (*bitmap),"r" (nr));
376 	return nr;
377 }
378 
379 #define val_byte(val, n) (((__u8 *)&val)[n])
380 
381 #define pushb(base, ptr, val, err_label) \
382 	do { \
383 		__u8 __val = val; \
384 		ptr--; \
385 		if (put_user(__val, base + ptr) < 0) \
386 			goto err_label; \
387 	} while(0)
388 
389 #define pushw(base, ptr, val, err_label) \
390 	do { \
391 		__u16 __val = val; \
392 		ptr--; \
393 		if (put_user(val_byte(__val, 1), base + ptr) < 0) \
394 			goto err_label; \
395 		ptr--; \
396 		if (put_user(val_byte(__val, 0), base + ptr) < 0) \
397 			goto err_label; \
398 	} while(0)
399 
400 #define pushl(base, ptr, val, err_label) \
401 	do { \
402 		__u32 __val = val; \
403 		ptr--; \
404 		if (put_user(val_byte(__val, 3), base + ptr) < 0) \
405 			goto err_label; \
406 		ptr--; \
407 		if (put_user(val_byte(__val, 2), base + ptr) < 0) \
408 			goto err_label; \
409 		ptr--; \
410 		if (put_user(val_byte(__val, 1), base + ptr) < 0) \
411 			goto err_label; \
412 		ptr--; \
413 		if (put_user(val_byte(__val, 0), base + ptr) < 0) \
414 			goto err_label; \
415 	} while(0)
416 
417 #define popb(base, ptr, err_label) \
418 	({ \
419 	 	__u8 __res; \
420 		if (get_user(__res, base + ptr) < 0) \
421 			goto err_label; \
422 		ptr++; \
423 		__res; \
424 	})
425 
426 #define popw(base, ptr, err_label) \
427 	({ \
428 	 	__u16 __res; \
429 		if (get_user(val_byte(__res, 0), base + ptr) < 0) \
430 			goto err_label; \
431 		ptr++; \
432 		if (get_user(val_byte(__res, 1), base + ptr) < 0) \
433 			goto err_label; \
434 		ptr++; \
435 		__res; \
436 	})
437 
438 #define popl(base, ptr, err_label) \
439 	({ \
440 	 	__u32 __res; \
441 		if (get_user(val_byte(__res, 0), base + ptr) < 0) \
442 			goto err_label; \
443 		ptr++; \
444 		if (get_user(val_byte(__res, 1), base + ptr) < 0) \
445 			goto err_label; \
446 		ptr++; \
447 		if (get_user(val_byte(__res, 2), base + ptr) < 0) \
448 			goto err_label; \
449 		ptr++; \
450 		if (get_user(val_byte(__res, 3), base + ptr) < 0) \
451 			goto err_label; \
452 		ptr++; \
453 		__res; \
454 	})
455 
456 /* There are so many possible reasons for this function to return
457  * VM86_INTx, so adding another doesn't bother me. We can expect
458  * userspace programs to be able to handle it. (Getting a problem
459  * in userspace is always better than an Oops anyway.) [KD]
460  */
do_int(struct kernel_vm86_regs * regs,int i,unsigned char * ssp,unsigned short sp)461 static void do_int(struct kernel_vm86_regs *regs, int i,
462     unsigned char * ssp, unsigned short sp)
463 {
464 	unsigned long *intr_ptr, segoffs;
465 
466 	if (regs->cs == BIOSSEG)
467 		goto cannot_handle;
468 	if (is_revectored(i, &KVM86->int_revectored))
469 		goto cannot_handle;
470 	if (i==0x21 && is_revectored(AH(regs),&KVM86->int21_revectored))
471 		goto cannot_handle;
472 	intr_ptr = (unsigned long *) (i << 2);
473 	if (get_user(segoffs, intr_ptr))
474 		goto cannot_handle;
475 	if ((segoffs >> 16) == BIOSSEG)
476 		goto cannot_handle;
477 	pushw(ssp, sp, get_vflags(regs), cannot_handle);
478 	pushw(ssp, sp, regs->cs, cannot_handle);
479 	pushw(ssp, sp, IP(regs), cannot_handle);
480 	regs->cs = segoffs >> 16;
481 	SP(regs) -= 6;
482 	IP(regs) = segoffs & 0xffff;
483 	clear_TF(regs);
484 	clear_IF(regs);
485 	clear_AC(regs);
486 	return;
487 
488 cannot_handle:
489 	return_to_32bit(regs, VM86_INTx + (i << 8));
490 }
491 
handle_vm86_trap(struct kernel_vm86_regs * regs,long error_code,int trapno)492 int handle_vm86_trap(struct kernel_vm86_regs * regs, long error_code, int trapno)
493 {
494 	if (VMPI.is_vm86pus) {
495 		if ( (trapno==3) || (trapno==1) )
496 			return_to_32bit(regs, VM86_TRAP + (trapno << 8));
497 		do_int(regs, trapno, (unsigned char *) (regs->ss << 4), SP(regs));
498 		return 0;
499 	}
500 	if (trapno !=1)
501 		return 1; /* we let this handle by the calling routine */
502 	if (current->ptrace & PT_PTRACED) {
503 		unsigned long flags;
504 		spin_lock_irqsave(&current->sigmask_lock, flags);
505 		sigdelset(&current->blocked, SIGTRAP);
506 		recalc_sigpending(current);
507 		spin_unlock_irqrestore(&current->sigmask_lock, flags);
508 	}
509 	send_sig(SIGTRAP, current, 1);
510 	current->thread.trap_no = trapno;
511 	current->thread.error_code = error_code;
512 	return 0;
513 }
514 
handle_vm86_fault(struct kernel_vm86_regs * regs,long error_code)515 void handle_vm86_fault(struct kernel_vm86_regs * regs, long error_code)
516 {
517 	unsigned char *csp, *ssp, opcode;
518 	unsigned short ip, sp;
519 	int data32, pref_done;
520 
521 #define CHECK_IF_IN_TRAP \
522 	if (VMPI.vm86dbg_active && VMPI.vm86dbg_TFpendig) \
523 		newflags |= TF_MASK
524 #define VM86_FAULT_RETURN do { \
525 	if (VMPI.force_return_for_pic  && (VEFLAGS & (IF_MASK | VIF_MASK))) \
526 		return_to_32bit(regs, VM86_PICRETURN); \
527 	return; } while (0)
528 
529 	csp = (unsigned char *) (regs->cs << 4);
530 	ssp = (unsigned char *) (regs->ss << 4);
531 	sp = SP(regs);
532 	ip = IP(regs);
533 
534 	data32 = 0;
535 	pref_done = 0;
536 	do {
537 		switch (opcode = popb(csp, ip, simulate_sigsegv)) {
538 			case 0x66:      /* 32-bit data */     data32=1; break;
539 			case 0x67:      /* 32-bit address */  break;
540 			case 0x2e:      /* CS */              break;
541 			case 0x3e:      /* DS */              break;
542 			case 0x26:      /* ES */              break;
543 			case 0x36:      /* SS */              break;
544 			case 0x65:      /* GS */              break;
545 			case 0x64:      /* FS */              break;
546 			case 0xf2:      /* repnz */	      break;
547 			case 0xf3:      /* rep */             break;
548 			default: pref_done = 1;
549 		}
550 	} while (!pref_done);
551 
552 	switch (opcode) {
553 
554 	/* pushf */
555 	case 0x9c:
556 		if (data32) {
557 			pushl(ssp, sp, get_vflags(regs), simulate_sigsegv);
558 			SP(regs) -= 4;
559 		} else {
560 			pushw(ssp, sp, get_vflags(regs), simulate_sigsegv);
561 			SP(regs) -= 2;
562 		}
563 		IP(regs) = ip;
564 		VM86_FAULT_RETURN;
565 
566 	/* popf */
567 	case 0x9d:
568 		{
569 		unsigned long newflags;
570 		if (data32) {
571 			newflags=popl(ssp, sp, simulate_sigsegv);
572 			SP(regs) += 4;
573 		} else {
574 			newflags = popw(ssp, sp, simulate_sigsegv);
575 			SP(regs) += 2;
576 		}
577 		IP(regs) = ip;
578 		CHECK_IF_IN_TRAP;
579 		if (data32) {
580 			set_vflags_long(newflags, regs);
581 		} else {
582 			set_vflags_short(newflags, regs);
583 		}
584 		VM86_FAULT_RETURN;
585 		}
586 
587 	/* int xx */
588 	case 0xcd: {
589 	        int intno=popb(csp, ip, simulate_sigsegv);
590 		IP(regs) = ip;
591 		if (VMPI.vm86dbg_active) {
592 			if ( (1 << (intno &7)) & VMPI.vm86dbg_intxxtab[intno >> 3] )
593 				return_to_32bit(regs, VM86_INTx + (intno << 8));
594 		}
595 		do_int(regs, intno, ssp, sp);
596 		return;
597 	}
598 
599 	/* iret */
600 	case 0xcf:
601 		{
602 		unsigned long newip;
603 		unsigned long newcs;
604 		unsigned long newflags;
605 		if (data32) {
606 			newip=popl(ssp, sp, simulate_sigsegv);
607 			newcs=popl(ssp, sp, simulate_sigsegv);
608 			newflags=popl(ssp, sp, simulate_sigsegv);
609 			SP(regs) += 12;
610 		} else {
611 			newip = popw(ssp, sp, simulate_sigsegv);
612 			newcs = popw(ssp, sp, simulate_sigsegv);
613 			newflags = popw(ssp, sp, simulate_sigsegv);
614 			SP(regs) += 6;
615 		}
616 		IP(regs) = newip;
617 		regs->cs = newcs;
618 		CHECK_IF_IN_TRAP;
619 		if (data32) {
620 			set_vflags_long(newflags, regs);
621 		} else {
622 			set_vflags_short(newflags, regs);
623 		}
624 		VM86_FAULT_RETURN;
625 		}
626 
627 	/* cli */
628 	case 0xfa:
629 		IP(regs) = ip;
630 		clear_IF(regs);
631 		VM86_FAULT_RETURN;
632 
633 	/* sti */
634 	/*
635 	 * Damn. This is incorrect: the 'sti' instruction should actually
636 	 * enable interrupts after the /next/ instruction. Not good.
637 	 *
638 	 * Probably needs some horsing around with the TF flag. Aiee..
639 	 */
640 	case 0xfb:
641 		IP(regs) = ip;
642 		set_IF(regs);
643 		VM86_FAULT_RETURN;
644 
645 	default:
646 		return_to_32bit(regs, VM86_UNKNOWN);
647 	}
648 
649 	return;
650 
651 simulate_sigsegv:
652 	/* FIXME: After a long discussion with Stas we finally
653 	 *        agreed, that this is wrong. Here we should
654 	 *        really send a SIGSEGV to the user program.
655 	 *        But how do we create the correct context? We
656 	 *        are inside a general protection fault handler
657 	 *        and has just returned from a page fault handler.
658 	 *        The correct context for the signal handler
659 	 *        should be a mixture of the two, but how do we
660 	 *        get the information? [KD]
661 	 */
662 	return_to_32bit(regs, VM86_UNKNOWN);
663 }
664 
665 /* ---------------- vm86 special IRQ passing stuff ----------------- */
666 
667 #define VM86_IRQNAME		"vm86irq"
668 
669 static struct vm86_irqs {
670 	struct task_struct *tsk;
671 	int sig;
672 } vm86_irqs[16];
673 static int irqbits;
674 
675 #define ALLOWED_SIGS ( 1 /* 0 = don't send a signal */ \
676 	| (1 << SIGUSR1) | (1 << SIGUSR2) | (1 << SIGIO)  | (1 << SIGURG) \
677 	| (1 << SIGUNUSED) )
678 
irq_handler(int intno,void * dev_id,struct pt_regs * regs)679 static void irq_handler(int intno, void *dev_id, struct pt_regs * regs) {
680 	int irq_bit;
681 	unsigned long flags;
682 
683 	save_flags(flags);
684 	cli();
685 	irq_bit = 1 << intno;
686 	if ((irqbits & irq_bit) || ! vm86_irqs[intno].tsk)
687 		goto out;
688 	irqbits |= irq_bit;
689 	if (vm86_irqs[intno].sig)
690 		send_sig(vm86_irqs[intno].sig, vm86_irqs[intno].tsk, 1);
691 	/* else user will poll for IRQs */
692 out:
693 	restore_flags(flags);
694 }
695 
free_vm86_irq(int irqnumber)696 static inline void free_vm86_irq(int irqnumber)
697 {
698 	free_irq(irqnumber,0);
699 	vm86_irqs[irqnumber].tsk = 0;
700 	irqbits &= ~(1 << irqnumber);
701 }
702 
task_valid(struct task_struct * tsk)703 static inline int task_valid(struct task_struct *tsk)
704 {
705 	struct task_struct *p;
706 	int ret = 0;
707 
708 	read_lock(&tasklist_lock);
709 	for_each_task(p) {
710 		if ((p == tsk) && (p->sig)) {
711 			ret = 1;
712 			break;
713 		}
714 	}
715 	read_unlock(&tasklist_lock);
716 	return ret;
717 }
718 
release_x86_irqs(struct task_struct * task)719 void release_x86_irqs(struct task_struct *task)
720 {
721 	int i;
722 	for (i=3; i<16; i++)
723 	    if (vm86_irqs[i].tsk == task)
724 		free_vm86_irq(i);
725 }
726 
handle_irq_zombies(void)727 static inline void handle_irq_zombies(void)
728 {
729 	int i;
730 	for (i=3; i<16; i++) {
731 		if (vm86_irqs[i].tsk) {
732 			if (task_valid(vm86_irqs[i].tsk)) continue;
733 			free_vm86_irq(i);
734 		}
735 	}
736 }
737 
get_and_reset_irq(int irqnumber)738 static inline int get_and_reset_irq(int irqnumber)
739 {
740 	int bit;
741 	unsigned long flags;
742 
743 	if ( (irqnumber<3) || (irqnumber>15) ) return 0;
744 	if (vm86_irqs[irqnumber].tsk != current) return 0;
745 	save_flags(flags);
746 	cli();
747 	bit = irqbits & (1 << irqnumber);
748 	irqbits &= ~bit;
749 	restore_flags(flags);
750 	return bit;
751 }
752 
753 
do_vm86_irq_handling(int subfunction,int irqnumber)754 static int do_vm86_irq_handling(int subfunction, int irqnumber)
755 {
756 	int ret;
757 	switch (subfunction) {
758 		case VM86_GET_AND_RESET_IRQ: {
759 			return get_and_reset_irq(irqnumber);
760 		}
761 		case VM86_GET_IRQ_BITS: {
762 			return irqbits;
763 		}
764 		case VM86_REQUEST_IRQ: {
765 			int sig = irqnumber >> 8;
766 			int irq = irqnumber & 255;
767 			handle_irq_zombies();
768 			if (!capable(CAP_SYS_ADMIN)) return -EPERM;
769 			if (!((1 << sig) & ALLOWED_SIGS)) return -EPERM;
770 			if ( (irq<3) || (irq>15) ) return -EPERM;
771 			if (vm86_irqs[irq].tsk) return -EPERM;
772 			ret = request_irq(irq, &irq_handler, 0, VM86_IRQNAME, 0);
773 			if (ret) return ret;
774 			vm86_irqs[irq].sig = sig;
775 			vm86_irqs[irq].tsk = current;
776 			return irq;
777 		}
778 		case  VM86_FREE_IRQ: {
779 			handle_irq_zombies();
780 			if ( (irqnumber<3) || (irqnumber>15) ) return -EPERM;
781 			if (!vm86_irqs[irqnumber].tsk) return 0;
782 			if (vm86_irqs[irqnumber].tsk != current) return -EPERM;
783 			free_vm86_irq(irqnumber);
784 			return 0;
785 		}
786 	}
787 	return -EINVAL;
788 }
789 
790