xref: /linux/arch/x86/kernel/vm86_32.c (revision b85d45947951d23cb22d90caecf4c1eb81342c96)
1 /*
2  *  Copyright (C) 1994  Linus Torvalds
3  *
4  *  29 dec 2001 - Fixed oopses caused by unchecked access to the vm86
5  *                stack - Manfred Spraul <manfred@colorfullife.com>
6  *
7  *  22 mar 2002 - Manfred detected the stackfaults, but didn't handle
8  *                them correctly. Now the emulation will be in a
9  *                consistent state after stackfaults - Kasper Dupont
10  *                <kasperd@daimi.au.dk>
11  *
12  *  22 mar 2002 - Added missing clear_IF in set_vflags_* Kasper Dupont
13  *                <kasperd@daimi.au.dk>
14  *
15  *  ?? ??? 2002 - Fixed premature returns from handle_vm86_fault
16  *                caused by Kasper Dupont's changes - Stas Sergeev
17  *
18  *   4 apr 2002 - Fixed CHECK_IF_IN_TRAP broken by Stas' changes.
19  *                Kasper Dupont <kasperd@daimi.au.dk>
20  *
21  *   9 apr 2002 - Changed syntax of macros in handle_vm86_fault.
22  *                Kasper Dupont <kasperd@daimi.au.dk>
23  *
24  *   9 apr 2002 - Changed stack access macros to jump to a label
25  *                instead of returning to userspace. This simplifies
26  *                do_int, and is needed by handle_vm6_fault. Kasper
27  *                Dupont <kasperd@daimi.au.dk>
28  *
29  */
30 
31 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
32 
33 #include <linux/capability.h>
34 #include <linux/errno.h>
35 #include <linux/interrupt.h>
36 #include <linux/syscalls.h>
37 #include <linux/sched.h>
38 #include <linux/kernel.h>
39 #include <linux/signal.h>
40 #include <linux/string.h>
41 #include <linux/mm.h>
42 #include <linux/smp.h>
43 #include <linux/highmem.h>
44 #include <linux/ptrace.h>
45 #include <linux/audit.h>
46 #include <linux/stddef.h>
47 #include <linux/slab.h>
48 #include <linux/security.h>
49 
50 #include <asm/uaccess.h>
51 #include <asm/io.h>
52 #include <asm/tlbflush.h>
53 #include <asm/irq.h>
54 #include <asm/traps.h>
55 #include <asm/vm86.h>
56 
57 /*
58  * Known problems:
59  *
60  * Interrupt handling is not guaranteed:
61  * - a real x86 will disable all interrupts for one instruction
62  *   after a "mov ss,xx" to make stack handling atomic even without
63  *   the 'lss' instruction. We can't guarantee this in v86 mode,
64  *   as the next instruction might result in a page fault or similar.
65  * - a real x86 will have interrupts disabled for one instruction
66  *   past the 'sti' that enables them. We don't bother with all the
67  *   details yet.
68  *
69  * Let's hope these problems do not actually matter for anything.
70  */
71 
72 
73 /*
74  * 8- and 16-bit register defines..
75  */
76 #define AL(regs)	(((unsigned char *)&((regs)->pt.ax))[0])
77 #define AH(regs)	(((unsigned char *)&((regs)->pt.ax))[1])
78 #define IP(regs)	(*(unsigned short *)&((regs)->pt.ip))
79 #define SP(regs)	(*(unsigned short *)&((regs)->pt.sp))
80 
81 /*
82  * virtual flags (16 and 32-bit versions)
83  */
84 #define VFLAGS	(*(unsigned short *)&(current->thread.vm86->veflags))
85 #define VEFLAGS	(current->thread.vm86->veflags)
86 
87 #define set_flags(X, new, mask) \
88 ((X) = ((X) & ~(mask)) | ((new) & (mask)))
89 
90 #define SAFE_MASK	(0xDD5)
91 #define RETURN_MASK	(0xDFF)
92 
93 void save_v86_state(struct kernel_vm86_regs *regs, int retval)
94 {
95 	struct tss_struct *tss;
96 	struct task_struct *tsk = current;
97 	struct vm86plus_struct __user *user;
98 	struct vm86 *vm86 = current->thread.vm86;
99 	long err = 0;
100 
101 	/*
102 	 * This gets called from entry.S with interrupts disabled, but
103 	 * from process context. Enable interrupts here, before trying
104 	 * to access user space.
105 	 */
106 	local_irq_enable();
107 
108 	if (!vm86 || !vm86->user_vm86) {
109 		pr_alert("no user_vm86: BAD\n");
110 		do_exit(SIGSEGV);
111 	}
112 	set_flags(regs->pt.flags, VEFLAGS, X86_EFLAGS_VIF | vm86->veflags_mask);
113 	user = vm86->user_vm86;
114 
115 	if (!access_ok(VERIFY_WRITE, user, vm86->vm86plus.is_vm86pus ?
116 		       sizeof(struct vm86plus_struct) :
117 		       sizeof(struct vm86_struct))) {
118 		pr_alert("could not access userspace vm86 info\n");
119 		do_exit(SIGSEGV);
120 	}
121 
122 	put_user_try {
123 		put_user_ex(regs->pt.bx, &user->regs.ebx);
124 		put_user_ex(regs->pt.cx, &user->regs.ecx);
125 		put_user_ex(regs->pt.dx, &user->regs.edx);
126 		put_user_ex(regs->pt.si, &user->regs.esi);
127 		put_user_ex(regs->pt.di, &user->regs.edi);
128 		put_user_ex(regs->pt.bp, &user->regs.ebp);
129 		put_user_ex(regs->pt.ax, &user->regs.eax);
130 		put_user_ex(regs->pt.ip, &user->regs.eip);
131 		put_user_ex(regs->pt.cs, &user->regs.cs);
132 		put_user_ex(regs->pt.flags, &user->regs.eflags);
133 		put_user_ex(regs->pt.sp, &user->regs.esp);
134 		put_user_ex(regs->pt.ss, &user->regs.ss);
135 		put_user_ex(regs->es, &user->regs.es);
136 		put_user_ex(regs->ds, &user->regs.ds);
137 		put_user_ex(regs->fs, &user->regs.fs);
138 		put_user_ex(regs->gs, &user->regs.gs);
139 
140 		put_user_ex(vm86->screen_bitmap, &user->screen_bitmap);
141 	} put_user_catch(err);
142 	if (err) {
143 		pr_alert("could not access userspace vm86 info\n");
144 		do_exit(SIGSEGV);
145 	}
146 
147 	tss = &per_cpu(cpu_tss, get_cpu());
148 	tsk->thread.sp0 = vm86->saved_sp0;
149 	tsk->thread.sysenter_cs = __KERNEL_CS;
150 	load_sp0(tss, &tsk->thread);
151 	vm86->saved_sp0 = 0;
152 	put_cpu();
153 
154 	memcpy(&regs->pt, &vm86->regs32, sizeof(struct pt_regs));
155 
156 	lazy_load_gs(vm86->regs32.gs);
157 
158 	regs->pt.ax = retval;
159 }
160 
161 static void mark_screen_rdonly(struct mm_struct *mm)
162 {
163 	pgd_t *pgd;
164 	pud_t *pud;
165 	pmd_t *pmd;
166 	pte_t *pte;
167 	spinlock_t *ptl;
168 	int i;
169 
170 	down_write(&mm->mmap_sem);
171 	pgd = pgd_offset(mm, 0xA0000);
172 	if (pgd_none_or_clear_bad(pgd))
173 		goto out;
174 	pud = pud_offset(pgd, 0xA0000);
175 	if (pud_none_or_clear_bad(pud))
176 		goto out;
177 	pmd = pmd_offset(pud, 0xA0000);
178 	split_huge_page_pmd_mm(mm, 0xA0000, pmd);
179 	if (pmd_none_or_clear_bad(pmd))
180 		goto out;
181 	pte = pte_offset_map_lock(mm, pmd, 0xA0000, &ptl);
182 	for (i = 0; i < 32; i++) {
183 		if (pte_present(*pte))
184 			set_pte(pte, pte_wrprotect(*pte));
185 		pte++;
186 	}
187 	pte_unmap_unlock(pte, ptl);
188 out:
189 	up_write(&mm->mmap_sem);
190 	flush_tlb();
191 }
192 
193 
194 
195 static int do_vm86_irq_handling(int subfunction, int irqnumber);
196 static long do_sys_vm86(struct vm86plus_struct __user *user_vm86, bool plus);
197 
198 SYSCALL_DEFINE1(vm86old, struct vm86_struct __user *, user_vm86)
199 {
200 	return do_sys_vm86((struct vm86plus_struct __user *) user_vm86, false);
201 }
202 
203 
204 SYSCALL_DEFINE2(vm86, unsigned long, cmd, unsigned long, arg)
205 {
206 	switch (cmd) {
207 	case VM86_REQUEST_IRQ:
208 	case VM86_FREE_IRQ:
209 	case VM86_GET_IRQ_BITS:
210 	case VM86_GET_AND_RESET_IRQ:
211 		return do_vm86_irq_handling(cmd, (int)arg);
212 	case VM86_PLUS_INSTALL_CHECK:
213 		/*
214 		 * NOTE: on old vm86 stuff this will return the error
215 		 *  from access_ok(), because the subfunction is
216 		 *  interpreted as (invalid) address to vm86_struct.
217 		 *  So the installation check works.
218 		 */
219 		return 0;
220 	}
221 
222 	/* we come here only for functions VM86_ENTER, VM86_ENTER_NO_BYPASS */
223 	return do_sys_vm86((struct vm86plus_struct __user *) arg, true);
224 }
225 
226 
227 static long do_sys_vm86(struct vm86plus_struct __user *user_vm86, bool plus)
228 {
229 	struct tss_struct *tss;
230 	struct task_struct *tsk = current;
231 	struct vm86 *vm86 = tsk->thread.vm86;
232 	struct kernel_vm86_regs vm86regs;
233 	struct pt_regs *regs = current_pt_regs();
234 	unsigned long err = 0;
235 
236 	err = security_mmap_addr(0);
237 	if (err) {
238 		/*
239 		 * vm86 cannot virtualize the address space, so vm86 users
240 		 * need to manage the low 1MB themselves using mmap.  Given
241 		 * that BIOS places important data in the first page, vm86
242 		 * is essentially useless if mmap_min_addr != 0.  DOSEMU,
243 		 * for example, won't even bother trying to use vm86 if it
244 		 * can't map a page at virtual address 0.
245 		 *
246 		 * To reduce the available kernel attack surface, simply
247 		 * disallow vm86(old) for users who cannot mmap at va 0.
248 		 *
249 		 * The implementation of security_mmap_addr will allow
250 		 * suitably privileged users to map va 0 even if
251 		 * vm.mmap_min_addr is set above 0, and we want this
252 		 * behavior for vm86 as well, as it ensures that legacy
253 		 * tools like vbetool will not fail just because of
254 		 * vm.mmap_min_addr.
255 		 */
256 		pr_info_once("Denied a call to vm86(old) from %s[%d] (uid: %d).  Set the vm.mmap_min_addr sysctl to 0 and/or adjust LSM mmap_min_addr policy to enable vm86 if you are using a vm86-based DOS emulator.\n",
257 			     current->comm, task_pid_nr(current),
258 			     from_kuid_munged(&init_user_ns, current_uid()));
259 		return -EPERM;
260 	}
261 
262 	if (!vm86) {
263 		if (!(vm86 = kzalloc(sizeof(*vm86), GFP_KERNEL)))
264 			return -ENOMEM;
265 		tsk->thread.vm86 = vm86;
266 	}
267 	if (vm86->saved_sp0)
268 		return -EPERM;
269 
270 	if (!access_ok(VERIFY_READ, user_vm86, plus ?
271 		       sizeof(struct vm86_struct) :
272 		       sizeof(struct vm86plus_struct)))
273 		return -EFAULT;
274 
275 	memset(&vm86regs, 0, sizeof(vm86regs));
276 	get_user_try {
277 		unsigned short seg;
278 		get_user_ex(vm86regs.pt.bx, &user_vm86->regs.ebx);
279 		get_user_ex(vm86regs.pt.cx, &user_vm86->regs.ecx);
280 		get_user_ex(vm86regs.pt.dx, &user_vm86->regs.edx);
281 		get_user_ex(vm86regs.pt.si, &user_vm86->regs.esi);
282 		get_user_ex(vm86regs.pt.di, &user_vm86->regs.edi);
283 		get_user_ex(vm86regs.pt.bp, &user_vm86->regs.ebp);
284 		get_user_ex(vm86regs.pt.ax, &user_vm86->regs.eax);
285 		get_user_ex(vm86regs.pt.ip, &user_vm86->regs.eip);
286 		get_user_ex(seg, &user_vm86->regs.cs);
287 		vm86regs.pt.cs = seg;
288 		get_user_ex(vm86regs.pt.flags, &user_vm86->regs.eflags);
289 		get_user_ex(vm86regs.pt.sp, &user_vm86->regs.esp);
290 		get_user_ex(seg, &user_vm86->regs.ss);
291 		vm86regs.pt.ss = seg;
292 		get_user_ex(vm86regs.es, &user_vm86->regs.es);
293 		get_user_ex(vm86regs.ds, &user_vm86->regs.ds);
294 		get_user_ex(vm86regs.fs, &user_vm86->regs.fs);
295 		get_user_ex(vm86regs.gs, &user_vm86->regs.gs);
296 
297 		get_user_ex(vm86->flags, &user_vm86->flags);
298 		get_user_ex(vm86->screen_bitmap, &user_vm86->screen_bitmap);
299 		get_user_ex(vm86->cpu_type, &user_vm86->cpu_type);
300 	} get_user_catch(err);
301 	if (err)
302 		return err;
303 
304 	if (copy_from_user(&vm86->int_revectored,
305 			   &user_vm86->int_revectored,
306 			   sizeof(struct revectored_struct)))
307 		return -EFAULT;
308 	if (copy_from_user(&vm86->int21_revectored,
309 			   &user_vm86->int21_revectored,
310 			   sizeof(struct revectored_struct)))
311 		return -EFAULT;
312 	if (plus) {
313 		if (copy_from_user(&vm86->vm86plus, &user_vm86->vm86plus,
314 				   sizeof(struct vm86plus_info_struct)))
315 			return -EFAULT;
316 		vm86->vm86plus.is_vm86pus = 1;
317 	} else
318 		memset(&vm86->vm86plus, 0,
319 		       sizeof(struct vm86plus_info_struct));
320 
321 	memcpy(&vm86->regs32, regs, sizeof(struct pt_regs));
322 	vm86->user_vm86 = user_vm86;
323 
324 /*
325  * The flags register is also special: we cannot trust that the user
326  * has set it up safely, so this makes sure interrupt etc flags are
327  * inherited from protected mode.
328  */
329 	VEFLAGS = vm86regs.pt.flags;
330 	vm86regs.pt.flags &= SAFE_MASK;
331 	vm86regs.pt.flags |= regs->flags & ~SAFE_MASK;
332 	vm86regs.pt.flags |= X86_VM_MASK;
333 
334 	vm86regs.pt.orig_ax = regs->orig_ax;
335 
336 	switch (vm86->cpu_type) {
337 	case CPU_286:
338 		vm86->veflags_mask = 0;
339 		break;
340 	case CPU_386:
341 		vm86->veflags_mask = X86_EFLAGS_NT | X86_EFLAGS_IOPL;
342 		break;
343 	case CPU_486:
344 		vm86->veflags_mask = X86_EFLAGS_AC | X86_EFLAGS_NT | X86_EFLAGS_IOPL;
345 		break;
346 	default:
347 		vm86->veflags_mask = X86_EFLAGS_ID | X86_EFLAGS_AC | X86_EFLAGS_NT | X86_EFLAGS_IOPL;
348 		break;
349 	}
350 
351 /*
352  * Save old state
353  */
354 	vm86->saved_sp0 = tsk->thread.sp0;
355 	lazy_save_gs(vm86->regs32.gs);
356 
357 	tss = &per_cpu(cpu_tss, get_cpu());
358 	/* make room for real-mode segments */
359 	tsk->thread.sp0 += 16;
360 	if (cpu_has_sep)
361 		tsk->thread.sysenter_cs = 0;
362 	load_sp0(tss, &tsk->thread);
363 	put_cpu();
364 
365 	if (vm86->flags & VM86_SCREEN_BITMAP)
366 		mark_screen_rdonly(tsk->mm);
367 
368 	memcpy((struct kernel_vm86_regs *)regs, &vm86regs, sizeof(vm86regs));
369 	force_iret();
370 	return regs->ax;
371 }
372 
373 static inline void set_IF(struct kernel_vm86_regs *regs)
374 {
375 	VEFLAGS |= X86_EFLAGS_VIF;
376 }
377 
378 static inline void clear_IF(struct kernel_vm86_regs *regs)
379 {
380 	VEFLAGS &= ~X86_EFLAGS_VIF;
381 }
382 
383 static inline void clear_TF(struct kernel_vm86_regs *regs)
384 {
385 	regs->pt.flags &= ~X86_EFLAGS_TF;
386 }
387 
388 static inline void clear_AC(struct kernel_vm86_regs *regs)
389 {
390 	regs->pt.flags &= ~X86_EFLAGS_AC;
391 }
392 
393 /*
394  * It is correct to call set_IF(regs) from the set_vflags_*
395  * functions. However someone forgot to call clear_IF(regs)
396  * in the opposite case.
397  * After the command sequence CLI PUSHF STI POPF you should
398  * end up with interrupts disabled, but you ended up with
399  * interrupts enabled.
400  *  ( I was testing my own changes, but the only bug I
401  *    could find was in a function I had not changed. )
402  * [KD]
403  */
404 
405 static inline void set_vflags_long(unsigned long flags, struct kernel_vm86_regs *regs)
406 {
407 	set_flags(VEFLAGS, flags, current->thread.vm86->veflags_mask);
408 	set_flags(regs->pt.flags, flags, SAFE_MASK);
409 	if (flags & X86_EFLAGS_IF)
410 		set_IF(regs);
411 	else
412 		clear_IF(regs);
413 }
414 
415 static inline void set_vflags_short(unsigned short flags, struct kernel_vm86_regs *regs)
416 {
417 	set_flags(VFLAGS, flags, current->thread.vm86->veflags_mask);
418 	set_flags(regs->pt.flags, flags, SAFE_MASK);
419 	if (flags & X86_EFLAGS_IF)
420 		set_IF(regs);
421 	else
422 		clear_IF(regs);
423 }
424 
425 static inline unsigned long get_vflags(struct kernel_vm86_regs *regs)
426 {
427 	unsigned long flags = regs->pt.flags & RETURN_MASK;
428 
429 	if (VEFLAGS & X86_EFLAGS_VIF)
430 		flags |= X86_EFLAGS_IF;
431 	flags |= X86_EFLAGS_IOPL;
432 	return flags | (VEFLAGS & current->thread.vm86->veflags_mask);
433 }
434 
435 static inline int is_revectored(int nr, struct revectored_struct *bitmap)
436 {
437 	__asm__ __volatile__("btl %2,%1\n\tsbbl %0,%0"
438 		:"=r" (nr)
439 		:"m" (*bitmap), "r" (nr));
440 	return nr;
441 }
442 
443 #define val_byte(val, n) (((__u8 *)&val)[n])
444 
445 #define pushb(base, ptr, val, err_label) \
446 	do { \
447 		__u8 __val = val; \
448 		ptr--; \
449 		if (put_user(__val, base + ptr) < 0) \
450 			goto err_label; \
451 	} while (0)
452 
453 #define pushw(base, ptr, val, err_label) \
454 	do { \
455 		__u16 __val = val; \
456 		ptr--; \
457 		if (put_user(val_byte(__val, 1), base + ptr) < 0) \
458 			goto err_label; \
459 		ptr--; \
460 		if (put_user(val_byte(__val, 0), base + ptr) < 0) \
461 			goto err_label; \
462 	} while (0)
463 
464 #define pushl(base, ptr, val, err_label) \
465 	do { \
466 		__u32 __val = val; \
467 		ptr--; \
468 		if (put_user(val_byte(__val, 3), base + ptr) < 0) \
469 			goto err_label; \
470 		ptr--; \
471 		if (put_user(val_byte(__val, 2), base + ptr) < 0) \
472 			goto err_label; \
473 		ptr--; \
474 		if (put_user(val_byte(__val, 1), base + ptr) < 0) \
475 			goto err_label; \
476 		ptr--; \
477 		if (put_user(val_byte(__val, 0), base + ptr) < 0) \
478 			goto err_label; \
479 	} while (0)
480 
481 #define popb(base, ptr, err_label) \
482 	({ \
483 		__u8 __res; \
484 		if (get_user(__res, base + ptr) < 0) \
485 			goto err_label; \
486 		ptr++; \
487 		__res; \
488 	})
489 
490 #define popw(base, ptr, err_label) \
491 	({ \
492 		__u16 __res; \
493 		if (get_user(val_byte(__res, 0), base + ptr) < 0) \
494 			goto err_label; \
495 		ptr++; \
496 		if (get_user(val_byte(__res, 1), base + ptr) < 0) \
497 			goto err_label; \
498 		ptr++; \
499 		__res; \
500 	})
501 
502 #define popl(base, ptr, err_label) \
503 	({ \
504 		__u32 __res; \
505 		if (get_user(val_byte(__res, 0), base + ptr) < 0) \
506 			goto err_label; \
507 		ptr++; \
508 		if (get_user(val_byte(__res, 1), base + ptr) < 0) \
509 			goto err_label; \
510 		ptr++; \
511 		if (get_user(val_byte(__res, 2), base + ptr) < 0) \
512 			goto err_label; \
513 		ptr++; \
514 		if (get_user(val_byte(__res, 3), base + ptr) < 0) \
515 			goto err_label; \
516 		ptr++; \
517 		__res; \
518 	})
519 
520 /* There are so many possible reasons for this function to return
521  * VM86_INTx, so adding another doesn't bother me. We can expect
522  * userspace programs to be able to handle it. (Getting a problem
523  * in userspace is always better than an Oops anyway.) [KD]
524  */
525 static void do_int(struct kernel_vm86_regs *regs, int i,
526     unsigned char __user *ssp, unsigned short sp)
527 {
528 	unsigned long __user *intr_ptr;
529 	unsigned long segoffs;
530 	struct vm86 *vm86 = current->thread.vm86;
531 
532 	if (regs->pt.cs == BIOSSEG)
533 		goto cannot_handle;
534 	if (is_revectored(i, &vm86->int_revectored))
535 		goto cannot_handle;
536 	if (i == 0x21 && is_revectored(AH(regs), &vm86->int21_revectored))
537 		goto cannot_handle;
538 	intr_ptr = (unsigned long __user *) (i << 2);
539 	if (get_user(segoffs, intr_ptr))
540 		goto cannot_handle;
541 	if ((segoffs >> 16) == BIOSSEG)
542 		goto cannot_handle;
543 	pushw(ssp, sp, get_vflags(regs), cannot_handle);
544 	pushw(ssp, sp, regs->pt.cs, cannot_handle);
545 	pushw(ssp, sp, IP(regs), cannot_handle);
546 	regs->pt.cs = segoffs >> 16;
547 	SP(regs) -= 6;
548 	IP(regs) = segoffs & 0xffff;
549 	clear_TF(regs);
550 	clear_IF(regs);
551 	clear_AC(regs);
552 	return;
553 
554 cannot_handle:
555 	save_v86_state(regs, VM86_INTx + (i << 8));
556 }
557 
558 int handle_vm86_trap(struct kernel_vm86_regs *regs, long error_code, int trapno)
559 {
560 	struct vm86 *vm86 = current->thread.vm86;
561 
562 	if (vm86->vm86plus.is_vm86pus) {
563 		if ((trapno == 3) || (trapno == 1)) {
564 			save_v86_state(regs, VM86_TRAP + (trapno << 8));
565 			return 0;
566 		}
567 		do_int(regs, trapno, (unsigned char __user *) (regs->pt.ss << 4), SP(regs));
568 		return 0;
569 	}
570 	if (trapno != 1)
571 		return 1; /* we let this handle by the calling routine */
572 	current->thread.trap_nr = trapno;
573 	current->thread.error_code = error_code;
574 	force_sig(SIGTRAP, current);
575 	return 0;
576 }
577 
578 void handle_vm86_fault(struct kernel_vm86_regs *regs, long error_code)
579 {
580 	unsigned char opcode;
581 	unsigned char __user *csp;
582 	unsigned char __user *ssp;
583 	unsigned short ip, sp, orig_flags;
584 	int data32, pref_done;
585 	struct vm86plus_info_struct *vmpi = &current->thread.vm86->vm86plus;
586 
587 #define CHECK_IF_IN_TRAP \
588 	if (vmpi->vm86dbg_active && vmpi->vm86dbg_TFpendig) \
589 		newflags |= X86_EFLAGS_TF
590 
591 	orig_flags = *(unsigned short *)&regs->pt.flags;
592 
593 	csp = (unsigned char __user *) (regs->pt.cs << 4);
594 	ssp = (unsigned char __user *) (regs->pt.ss << 4);
595 	sp = SP(regs);
596 	ip = IP(regs);
597 
598 	data32 = 0;
599 	pref_done = 0;
600 	do {
601 		switch (opcode = popb(csp, ip, simulate_sigsegv)) {
602 		case 0x66:      /* 32-bit data */     data32 = 1; break;
603 		case 0x67:      /* 32-bit address */  break;
604 		case 0x2e:      /* CS */              break;
605 		case 0x3e:      /* DS */              break;
606 		case 0x26:      /* ES */              break;
607 		case 0x36:      /* SS */              break;
608 		case 0x65:      /* GS */              break;
609 		case 0x64:      /* FS */              break;
610 		case 0xf2:      /* repnz */       break;
611 		case 0xf3:      /* rep */             break;
612 		default: pref_done = 1;
613 		}
614 	} while (!pref_done);
615 
616 	switch (opcode) {
617 
618 	/* pushf */
619 	case 0x9c:
620 		if (data32) {
621 			pushl(ssp, sp, get_vflags(regs), simulate_sigsegv);
622 			SP(regs) -= 4;
623 		} else {
624 			pushw(ssp, sp, get_vflags(regs), simulate_sigsegv);
625 			SP(regs) -= 2;
626 		}
627 		IP(regs) = ip;
628 		goto vm86_fault_return;
629 
630 	/* popf */
631 	case 0x9d:
632 		{
633 		unsigned long newflags;
634 		if (data32) {
635 			newflags = popl(ssp, sp, simulate_sigsegv);
636 			SP(regs) += 4;
637 		} else {
638 			newflags = popw(ssp, sp, simulate_sigsegv);
639 			SP(regs) += 2;
640 		}
641 		IP(regs) = ip;
642 		CHECK_IF_IN_TRAP;
643 		if (data32)
644 			set_vflags_long(newflags, regs);
645 		else
646 			set_vflags_short(newflags, regs);
647 
648 		goto check_vip;
649 		}
650 
651 	/* int xx */
652 	case 0xcd: {
653 		int intno = popb(csp, ip, simulate_sigsegv);
654 		IP(regs) = ip;
655 		if (vmpi->vm86dbg_active) {
656 			if ((1 << (intno & 7)) & vmpi->vm86dbg_intxxtab[intno >> 3]) {
657 				save_v86_state(regs, VM86_INTx + (intno << 8));
658 				return;
659 			}
660 		}
661 		do_int(regs, intno, ssp, sp);
662 		return;
663 	}
664 
665 	/* iret */
666 	case 0xcf:
667 		{
668 		unsigned long newip;
669 		unsigned long newcs;
670 		unsigned long newflags;
671 		if (data32) {
672 			newip = popl(ssp, sp, simulate_sigsegv);
673 			newcs = popl(ssp, sp, simulate_sigsegv);
674 			newflags = popl(ssp, sp, simulate_sigsegv);
675 			SP(regs) += 12;
676 		} else {
677 			newip = popw(ssp, sp, simulate_sigsegv);
678 			newcs = popw(ssp, sp, simulate_sigsegv);
679 			newflags = popw(ssp, sp, simulate_sigsegv);
680 			SP(regs) += 6;
681 		}
682 		IP(regs) = newip;
683 		regs->pt.cs = newcs;
684 		CHECK_IF_IN_TRAP;
685 		if (data32) {
686 			set_vflags_long(newflags, regs);
687 		} else {
688 			set_vflags_short(newflags, regs);
689 		}
690 		goto check_vip;
691 		}
692 
693 	/* cli */
694 	case 0xfa:
695 		IP(regs) = ip;
696 		clear_IF(regs);
697 		goto vm86_fault_return;
698 
699 	/* sti */
700 	/*
701 	 * Damn. This is incorrect: the 'sti' instruction should actually
702 	 * enable interrupts after the /next/ instruction. Not good.
703 	 *
704 	 * Probably needs some horsing around with the TF flag. Aiee..
705 	 */
706 	case 0xfb:
707 		IP(regs) = ip;
708 		set_IF(regs);
709 		goto check_vip;
710 
711 	default:
712 		save_v86_state(regs, VM86_UNKNOWN);
713 	}
714 
715 	return;
716 
717 check_vip:
718 	if (VEFLAGS & X86_EFLAGS_VIP) {
719 		save_v86_state(regs, VM86_STI);
720 		return;
721 	}
722 
723 vm86_fault_return:
724 	if (vmpi->force_return_for_pic  && (VEFLAGS & (X86_EFLAGS_IF | X86_EFLAGS_VIF))) {
725 		save_v86_state(regs, VM86_PICRETURN);
726 		return;
727 	}
728 	if (orig_flags & X86_EFLAGS_TF)
729 		handle_vm86_trap(regs, 0, X86_TRAP_DB);
730 	return;
731 
732 simulate_sigsegv:
733 	/* FIXME: After a long discussion with Stas we finally
734 	 *        agreed, that this is wrong. Here we should
735 	 *        really send a SIGSEGV to the user program.
736 	 *        But how do we create the correct context? We
737 	 *        are inside a general protection fault handler
738 	 *        and has just returned from a page fault handler.
739 	 *        The correct context for the signal handler
740 	 *        should be a mixture of the two, but how do we
741 	 *        get the information? [KD]
742 	 */
743 	save_v86_state(regs, VM86_UNKNOWN);
744 }
745 
746 /* ---------------- vm86 special IRQ passing stuff ----------------- */
747 
748 #define VM86_IRQNAME		"vm86irq"
749 
750 static struct vm86_irqs {
751 	struct task_struct *tsk;
752 	int sig;
753 } vm86_irqs[16];
754 
755 static DEFINE_SPINLOCK(irqbits_lock);
756 static int irqbits;
757 
758 #define ALLOWED_SIGS (1 /* 0 = don't send a signal */ \
759 	| (1 << SIGUSR1) | (1 << SIGUSR2) | (1 << SIGIO)  | (1 << SIGURG) \
760 	| (1 << SIGUNUSED))
761 
762 static irqreturn_t irq_handler(int intno, void *dev_id)
763 {
764 	int irq_bit;
765 	unsigned long flags;
766 
767 	spin_lock_irqsave(&irqbits_lock, flags);
768 	irq_bit = 1 << intno;
769 	if ((irqbits & irq_bit) || !vm86_irqs[intno].tsk)
770 		goto out;
771 	irqbits |= irq_bit;
772 	if (vm86_irqs[intno].sig)
773 		send_sig(vm86_irqs[intno].sig, vm86_irqs[intno].tsk, 1);
774 	/*
775 	 * IRQ will be re-enabled when user asks for the irq (whether
776 	 * polling or as a result of the signal)
777 	 */
778 	disable_irq_nosync(intno);
779 	spin_unlock_irqrestore(&irqbits_lock, flags);
780 	return IRQ_HANDLED;
781 
782 out:
783 	spin_unlock_irqrestore(&irqbits_lock, flags);
784 	return IRQ_NONE;
785 }
786 
787 static inline void free_vm86_irq(int irqnumber)
788 {
789 	unsigned long flags;
790 
791 	free_irq(irqnumber, NULL);
792 	vm86_irqs[irqnumber].tsk = NULL;
793 
794 	spin_lock_irqsave(&irqbits_lock, flags);
795 	irqbits &= ~(1 << irqnumber);
796 	spin_unlock_irqrestore(&irqbits_lock, flags);
797 }
798 
799 void release_vm86_irqs(struct task_struct *task)
800 {
801 	int i;
802 	for (i = FIRST_VM86_IRQ ; i <= LAST_VM86_IRQ; i++)
803 	    if (vm86_irqs[i].tsk == task)
804 		free_vm86_irq(i);
805 }
806 
807 static inline int get_and_reset_irq(int irqnumber)
808 {
809 	int bit;
810 	unsigned long flags;
811 	int ret = 0;
812 
813 	if (invalid_vm86_irq(irqnumber)) return 0;
814 	if (vm86_irqs[irqnumber].tsk != current) return 0;
815 	spin_lock_irqsave(&irqbits_lock, flags);
816 	bit = irqbits & (1 << irqnumber);
817 	irqbits &= ~bit;
818 	if (bit) {
819 		enable_irq(irqnumber);
820 		ret = 1;
821 	}
822 
823 	spin_unlock_irqrestore(&irqbits_lock, flags);
824 	return ret;
825 }
826 
827 
828 static int do_vm86_irq_handling(int subfunction, int irqnumber)
829 {
830 	int ret;
831 	switch (subfunction) {
832 		case VM86_GET_AND_RESET_IRQ: {
833 			return get_and_reset_irq(irqnumber);
834 		}
835 		case VM86_GET_IRQ_BITS: {
836 			return irqbits;
837 		}
838 		case VM86_REQUEST_IRQ: {
839 			int sig = irqnumber >> 8;
840 			int irq = irqnumber & 255;
841 			if (!capable(CAP_SYS_ADMIN)) return -EPERM;
842 			if (!((1 << sig) & ALLOWED_SIGS)) return -EPERM;
843 			if (invalid_vm86_irq(irq)) return -EPERM;
844 			if (vm86_irqs[irq].tsk) return -EPERM;
845 			ret = request_irq(irq, &irq_handler, 0, VM86_IRQNAME, NULL);
846 			if (ret) return ret;
847 			vm86_irqs[irq].sig = sig;
848 			vm86_irqs[irq].tsk = current;
849 			return irq;
850 		}
851 		case  VM86_FREE_IRQ: {
852 			if (invalid_vm86_irq(irqnumber)) return -EPERM;
853 			if (!vm86_irqs[irqnumber].tsk) return 0;
854 			if (vm86_irqs[irqnumber].tsk != current) return -EPERM;
855 			free_vm86_irq(irqnumber);
856 			return 0;
857 		}
858 	}
859 	return -EINVAL;
860 }
861 
862