xref: /openbmc/linux/arch/x86/kernel/vm86_32.c (revision aac5987a)
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/sched/task_stack.h>
39 #include <linux/kernel.h>
40 #include <linux/signal.h>
41 #include <linux/string.h>
42 #include <linux/mm.h>
43 #include <linux/smp.h>
44 #include <linux/highmem.h>
45 #include <linux/ptrace.h>
46 #include <linux/audit.h>
47 #include <linux/stddef.h>
48 #include <linux/slab.h>
49 #include <linux/security.h>
50 
51 #include <linux/uaccess.h>
52 #include <asm/io.h>
53 #include <asm/tlbflush.h>
54 #include <asm/irq.h>
55 #include <asm/traps.h>
56 #include <asm/vm86.h>
57 
58 /*
59  * Known problems:
60  *
61  * Interrupt handling is not guaranteed:
62  * - a real x86 will disable all interrupts for one instruction
63  *   after a "mov ss,xx" to make stack handling atomic even without
64  *   the 'lss' instruction. We can't guarantee this in v86 mode,
65  *   as the next instruction might result in a page fault or similar.
66  * - a real x86 will have interrupts disabled for one instruction
67  *   past the 'sti' that enables them. We don't bother with all the
68  *   details yet.
69  *
70  * Let's hope these problems do not actually matter for anything.
71  */
72 
73 
74 /*
75  * 8- and 16-bit register defines..
76  */
77 #define AL(regs)	(((unsigned char *)&((regs)->pt.ax))[0])
78 #define AH(regs)	(((unsigned char *)&((regs)->pt.ax))[1])
79 #define IP(regs)	(*(unsigned short *)&((regs)->pt.ip))
80 #define SP(regs)	(*(unsigned short *)&((regs)->pt.sp))
81 
82 /*
83  * virtual flags (16 and 32-bit versions)
84  */
85 #define VFLAGS	(*(unsigned short *)&(current->thread.vm86->veflags))
86 #define VEFLAGS	(current->thread.vm86->veflags)
87 
88 #define set_flags(X, new, mask) \
89 ((X) = ((X) & ~(mask)) | ((new) & (mask)))
90 
91 #define SAFE_MASK	(0xDD5)
92 #define RETURN_MASK	(0xDFF)
93 
94 void save_v86_state(struct kernel_vm86_regs *regs, int retval)
95 {
96 	struct tss_struct *tss;
97 	struct task_struct *tsk = current;
98 	struct vm86plus_struct __user *user;
99 	struct vm86 *vm86 = current->thread.vm86;
100 	long err = 0;
101 
102 	/*
103 	 * This gets called from entry.S with interrupts disabled, but
104 	 * from process context. Enable interrupts here, before trying
105 	 * to access user space.
106 	 */
107 	local_irq_enable();
108 
109 	if (!vm86 || !vm86->user_vm86) {
110 		pr_alert("no user_vm86: BAD\n");
111 		do_exit(SIGSEGV);
112 	}
113 	set_flags(regs->pt.flags, VEFLAGS, X86_EFLAGS_VIF | vm86->veflags_mask);
114 	user = vm86->user_vm86;
115 
116 	if (!access_ok(VERIFY_WRITE, user, vm86->vm86plus.is_vm86pus ?
117 		       sizeof(struct vm86plus_struct) :
118 		       sizeof(struct vm86_struct))) {
119 		pr_alert("could not access userspace vm86 info\n");
120 		do_exit(SIGSEGV);
121 	}
122 
123 	put_user_try {
124 		put_user_ex(regs->pt.bx, &user->regs.ebx);
125 		put_user_ex(regs->pt.cx, &user->regs.ecx);
126 		put_user_ex(regs->pt.dx, &user->regs.edx);
127 		put_user_ex(regs->pt.si, &user->regs.esi);
128 		put_user_ex(regs->pt.di, &user->regs.edi);
129 		put_user_ex(regs->pt.bp, &user->regs.ebp);
130 		put_user_ex(regs->pt.ax, &user->regs.eax);
131 		put_user_ex(regs->pt.ip, &user->regs.eip);
132 		put_user_ex(regs->pt.cs, &user->regs.cs);
133 		put_user_ex(regs->pt.flags, &user->regs.eflags);
134 		put_user_ex(regs->pt.sp, &user->regs.esp);
135 		put_user_ex(regs->pt.ss, &user->regs.ss);
136 		put_user_ex(regs->es, &user->regs.es);
137 		put_user_ex(regs->ds, &user->regs.ds);
138 		put_user_ex(regs->fs, &user->regs.fs);
139 		put_user_ex(regs->gs, &user->regs.gs);
140 
141 		put_user_ex(vm86->screen_bitmap, &user->screen_bitmap);
142 	} put_user_catch(err);
143 	if (err) {
144 		pr_alert("could not access userspace vm86 info\n");
145 		do_exit(SIGSEGV);
146 	}
147 
148 	tss = &per_cpu(cpu_tss, get_cpu());
149 	tsk->thread.sp0 = vm86->saved_sp0;
150 	tsk->thread.sysenter_cs = __KERNEL_CS;
151 	load_sp0(tss, &tsk->thread);
152 	vm86->saved_sp0 = 0;
153 	put_cpu();
154 
155 	memcpy(&regs->pt, &vm86->regs32, sizeof(struct pt_regs));
156 
157 	lazy_load_gs(vm86->regs32.gs);
158 
159 	regs->pt.ax = retval;
160 }
161 
162 static void mark_screen_rdonly(struct mm_struct *mm)
163 {
164 	struct vm_area_struct *vma;
165 	spinlock_t *ptl;
166 	pgd_t *pgd;
167 	pud_t *pud;
168 	pmd_t *pmd;
169 	pte_t *pte;
170 	int i;
171 
172 	down_write(&mm->mmap_sem);
173 	pgd = pgd_offset(mm, 0xA0000);
174 	if (pgd_none_or_clear_bad(pgd))
175 		goto out;
176 	pud = pud_offset(pgd, 0xA0000);
177 	if (pud_none_or_clear_bad(pud))
178 		goto out;
179 	pmd = pmd_offset(pud, 0xA0000);
180 
181 	if (pmd_trans_huge(*pmd)) {
182 		vma = find_vma(mm, 0xA0000);
183 		split_huge_pmd(vma, pmd, 0xA0000);
184 	}
185 	if (pmd_none_or_clear_bad(pmd))
186 		goto out;
187 	pte = pte_offset_map_lock(mm, pmd, 0xA0000, &ptl);
188 	for (i = 0; i < 32; i++) {
189 		if (pte_present(*pte))
190 			set_pte(pte, pte_wrprotect(*pte));
191 		pte++;
192 	}
193 	pte_unmap_unlock(pte, ptl);
194 out:
195 	up_write(&mm->mmap_sem);
196 	flush_tlb();
197 }
198 
199 
200 
201 static int do_vm86_irq_handling(int subfunction, int irqnumber);
202 static long do_sys_vm86(struct vm86plus_struct __user *user_vm86, bool plus);
203 
204 SYSCALL_DEFINE1(vm86old, struct vm86_struct __user *, user_vm86)
205 {
206 	return do_sys_vm86((struct vm86plus_struct __user *) user_vm86, false);
207 }
208 
209 
210 SYSCALL_DEFINE2(vm86, unsigned long, cmd, unsigned long, arg)
211 {
212 	switch (cmd) {
213 	case VM86_REQUEST_IRQ:
214 	case VM86_FREE_IRQ:
215 	case VM86_GET_IRQ_BITS:
216 	case VM86_GET_AND_RESET_IRQ:
217 		return do_vm86_irq_handling(cmd, (int)arg);
218 	case VM86_PLUS_INSTALL_CHECK:
219 		/*
220 		 * NOTE: on old vm86 stuff this will return the error
221 		 *  from access_ok(), because the subfunction is
222 		 *  interpreted as (invalid) address to vm86_struct.
223 		 *  So the installation check works.
224 		 */
225 		return 0;
226 	}
227 
228 	/* we come here only for functions VM86_ENTER, VM86_ENTER_NO_BYPASS */
229 	return do_sys_vm86((struct vm86plus_struct __user *) arg, true);
230 }
231 
232 
233 static long do_sys_vm86(struct vm86plus_struct __user *user_vm86, bool plus)
234 {
235 	struct tss_struct *tss;
236 	struct task_struct *tsk = current;
237 	struct vm86 *vm86 = tsk->thread.vm86;
238 	struct kernel_vm86_regs vm86regs;
239 	struct pt_regs *regs = current_pt_regs();
240 	unsigned long err = 0;
241 
242 	err = security_mmap_addr(0);
243 	if (err) {
244 		/*
245 		 * vm86 cannot virtualize the address space, so vm86 users
246 		 * need to manage the low 1MB themselves using mmap.  Given
247 		 * that BIOS places important data in the first page, vm86
248 		 * is essentially useless if mmap_min_addr != 0.  DOSEMU,
249 		 * for example, won't even bother trying to use vm86 if it
250 		 * can't map a page at virtual address 0.
251 		 *
252 		 * To reduce the available kernel attack surface, simply
253 		 * disallow vm86(old) for users who cannot mmap at va 0.
254 		 *
255 		 * The implementation of security_mmap_addr will allow
256 		 * suitably privileged users to map va 0 even if
257 		 * vm.mmap_min_addr is set above 0, and we want this
258 		 * behavior for vm86 as well, as it ensures that legacy
259 		 * tools like vbetool will not fail just because of
260 		 * vm.mmap_min_addr.
261 		 */
262 		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",
263 			     current->comm, task_pid_nr(current),
264 			     from_kuid_munged(&init_user_ns, current_uid()));
265 		return -EPERM;
266 	}
267 
268 	if (!vm86) {
269 		if (!(vm86 = kzalloc(sizeof(*vm86), GFP_KERNEL)))
270 			return -ENOMEM;
271 		tsk->thread.vm86 = vm86;
272 	}
273 	if (vm86->saved_sp0)
274 		return -EPERM;
275 
276 	if (!access_ok(VERIFY_READ, user_vm86, plus ?
277 		       sizeof(struct vm86_struct) :
278 		       sizeof(struct vm86plus_struct)))
279 		return -EFAULT;
280 
281 	memset(&vm86regs, 0, sizeof(vm86regs));
282 	get_user_try {
283 		unsigned short seg;
284 		get_user_ex(vm86regs.pt.bx, &user_vm86->regs.ebx);
285 		get_user_ex(vm86regs.pt.cx, &user_vm86->regs.ecx);
286 		get_user_ex(vm86regs.pt.dx, &user_vm86->regs.edx);
287 		get_user_ex(vm86regs.pt.si, &user_vm86->regs.esi);
288 		get_user_ex(vm86regs.pt.di, &user_vm86->regs.edi);
289 		get_user_ex(vm86regs.pt.bp, &user_vm86->regs.ebp);
290 		get_user_ex(vm86regs.pt.ax, &user_vm86->regs.eax);
291 		get_user_ex(vm86regs.pt.ip, &user_vm86->regs.eip);
292 		get_user_ex(seg, &user_vm86->regs.cs);
293 		vm86regs.pt.cs = seg;
294 		get_user_ex(vm86regs.pt.flags, &user_vm86->regs.eflags);
295 		get_user_ex(vm86regs.pt.sp, &user_vm86->regs.esp);
296 		get_user_ex(seg, &user_vm86->regs.ss);
297 		vm86regs.pt.ss = seg;
298 		get_user_ex(vm86regs.es, &user_vm86->regs.es);
299 		get_user_ex(vm86regs.ds, &user_vm86->regs.ds);
300 		get_user_ex(vm86regs.fs, &user_vm86->regs.fs);
301 		get_user_ex(vm86regs.gs, &user_vm86->regs.gs);
302 
303 		get_user_ex(vm86->flags, &user_vm86->flags);
304 		get_user_ex(vm86->screen_bitmap, &user_vm86->screen_bitmap);
305 		get_user_ex(vm86->cpu_type, &user_vm86->cpu_type);
306 	} get_user_catch(err);
307 	if (err)
308 		return err;
309 
310 	if (copy_from_user(&vm86->int_revectored,
311 			   &user_vm86->int_revectored,
312 			   sizeof(struct revectored_struct)))
313 		return -EFAULT;
314 	if (copy_from_user(&vm86->int21_revectored,
315 			   &user_vm86->int21_revectored,
316 			   sizeof(struct revectored_struct)))
317 		return -EFAULT;
318 	if (plus) {
319 		if (copy_from_user(&vm86->vm86plus, &user_vm86->vm86plus,
320 				   sizeof(struct vm86plus_info_struct)))
321 			return -EFAULT;
322 		vm86->vm86plus.is_vm86pus = 1;
323 	} else
324 		memset(&vm86->vm86plus, 0,
325 		       sizeof(struct vm86plus_info_struct));
326 
327 	memcpy(&vm86->regs32, regs, sizeof(struct pt_regs));
328 	vm86->user_vm86 = user_vm86;
329 
330 /*
331  * The flags register is also special: we cannot trust that the user
332  * has set it up safely, so this makes sure interrupt etc flags are
333  * inherited from protected mode.
334  */
335 	VEFLAGS = vm86regs.pt.flags;
336 	vm86regs.pt.flags &= SAFE_MASK;
337 	vm86regs.pt.flags |= regs->flags & ~SAFE_MASK;
338 	vm86regs.pt.flags |= X86_VM_MASK;
339 
340 	vm86regs.pt.orig_ax = regs->orig_ax;
341 
342 	switch (vm86->cpu_type) {
343 	case CPU_286:
344 		vm86->veflags_mask = 0;
345 		break;
346 	case CPU_386:
347 		vm86->veflags_mask = X86_EFLAGS_NT | X86_EFLAGS_IOPL;
348 		break;
349 	case CPU_486:
350 		vm86->veflags_mask = X86_EFLAGS_AC | X86_EFLAGS_NT | X86_EFLAGS_IOPL;
351 		break;
352 	default:
353 		vm86->veflags_mask = X86_EFLAGS_ID | X86_EFLAGS_AC | X86_EFLAGS_NT | X86_EFLAGS_IOPL;
354 		break;
355 	}
356 
357 /*
358  * Save old state
359  */
360 	vm86->saved_sp0 = tsk->thread.sp0;
361 	lazy_save_gs(vm86->regs32.gs);
362 
363 	tss = &per_cpu(cpu_tss, get_cpu());
364 	/* make room for real-mode segments */
365 	tsk->thread.sp0 += 16;
366 
367 	if (static_cpu_has(X86_FEATURE_SEP))
368 		tsk->thread.sysenter_cs = 0;
369 
370 	load_sp0(tss, &tsk->thread);
371 	put_cpu();
372 
373 	if (vm86->flags & VM86_SCREEN_BITMAP)
374 		mark_screen_rdonly(tsk->mm);
375 
376 	memcpy((struct kernel_vm86_regs *)regs, &vm86regs, sizeof(vm86regs));
377 	force_iret();
378 	return regs->ax;
379 }
380 
381 static inline void set_IF(struct kernel_vm86_regs *regs)
382 {
383 	VEFLAGS |= X86_EFLAGS_VIF;
384 }
385 
386 static inline void clear_IF(struct kernel_vm86_regs *regs)
387 {
388 	VEFLAGS &= ~X86_EFLAGS_VIF;
389 }
390 
391 static inline void clear_TF(struct kernel_vm86_regs *regs)
392 {
393 	regs->pt.flags &= ~X86_EFLAGS_TF;
394 }
395 
396 static inline void clear_AC(struct kernel_vm86_regs *regs)
397 {
398 	regs->pt.flags &= ~X86_EFLAGS_AC;
399 }
400 
401 /*
402  * It is correct to call set_IF(regs) from the set_vflags_*
403  * functions. However someone forgot to call clear_IF(regs)
404  * in the opposite case.
405  * After the command sequence CLI PUSHF STI POPF you should
406  * end up with interrupts disabled, but you ended up with
407  * interrupts enabled.
408  *  ( I was testing my own changes, but the only bug I
409  *    could find was in a function I had not changed. )
410  * [KD]
411  */
412 
413 static inline void set_vflags_long(unsigned long flags, struct kernel_vm86_regs *regs)
414 {
415 	set_flags(VEFLAGS, flags, current->thread.vm86->veflags_mask);
416 	set_flags(regs->pt.flags, flags, SAFE_MASK);
417 	if (flags & X86_EFLAGS_IF)
418 		set_IF(regs);
419 	else
420 		clear_IF(regs);
421 }
422 
423 static inline void set_vflags_short(unsigned short flags, struct kernel_vm86_regs *regs)
424 {
425 	set_flags(VFLAGS, flags, current->thread.vm86->veflags_mask);
426 	set_flags(regs->pt.flags, flags, SAFE_MASK);
427 	if (flags & X86_EFLAGS_IF)
428 		set_IF(regs);
429 	else
430 		clear_IF(regs);
431 }
432 
433 static inline unsigned long get_vflags(struct kernel_vm86_regs *regs)
434 {
435 	unsigned long flags = regs->pt.flags & RETURN_MASK;
436 
437 	if (VEFLAGS & X86_EFLAGS_VIF)
438 		flags |= X86_EFLAGS_IF;
439 	flags |= X86_EFLAGS_IOPL;
440 	return flags | (VEFLAGS & current->thread.vm86->veflags_mask);
441 }
442 
443 static inline int is_revectored(int nr, struct revectored_struct *bitmap)
444 {
445 	return test_bit(nr, bitmap->__map);
446 }
447 
448 #define val_byte(val, n) (((__u8 *)&val)[n])
449 
450 #define pushb(base, ptr, val, err_label) \
451 	do { \
452 		__u8 __val = val; \
453 		ptr--; \
454 		if (put_user(__val, base + ptr) < 0) \
455 			goto err_label; \
456 	} while (0)
457 
458 #define pushw(base, ptr, val, err_label) \
459 	do { \
460 		__u16 __val = val; \
461 		ptr--; \
462 		if (put_user(val_byte(__val, 1), base + ptr) < 0) \
463 			goto err_label; \
464 		ptr--; \
465 		if (put_user(val_byte(__val, 0), base + ptr) < 0) \
466 			goto err_label; \
467 	} while (0)
468 
469 #define pushl(base, ptr, val, err_label) \
470 	do { \
471 		__u32 __val = val; \
472 		ptr--; \
473 		if (put_user(val_byte(__val, 3), base + ptr) < 0) \
474 			goto err_label; \
475 		ptr--; \
476 		if (put_user(val_byte(__val, 2), base + ptr) < 0) \
477 			goto err_label; \
478 		ptr--; \
479 		if (put_user(val_byte(__val, 1), base + ptr) < 0) \
480 			goto err_label; \
481 		ptr--; \
482 		if (put_user(val_byte(__val, 0), base + ptr) < 0) \
483 			goto err_label; \
484 	} while (0)
485 
486 #define popb(base, ptr, err_label) \
487 	({ \
488 		__u8 __res; \
489 		if (get_user(__res, base + ptr) < 0) \
490 			goto err_label; \
491 		ptr++; \
492 		__res; \
493 	})
494 
495 #define popw(base, ptr, err_label) \
496 	({ \
497 		__u16 __res; \
498 		if (get_user(val_byte(__res, 0), base + ptr) < 0) \
499 			goto err_label; \
500 		ptr++; \
501 		if (get_user(val_byte(__res, 1), base + ptr) < 0) \
502 			goto err_label; \
503 		ptr++; \
504 		__res; \
505 	})
506 
507 #define popl(base, ptr, err_label) \
508 	({ \
509 		__u32 __res; \
510 		if (get_user(val_byte(__res, 0), base + ptr) < 0) \
511 			goto err_label; \
512 		ptr++; \
513 		if (get_user(val_byte(__res, 1), base + ptr) < 0) \
514 			goto err_label; \
515 		ptr++; \
516 		if (get_user(val_byte(__res, 2), base + ptr) < 0) \
517 			goto err_label; \
518 		ptr++; \
519 		if (get_user(val_byte(__res, 3), base + ptr) < 0) \
520 			goto err_label; \
521 		ptr++; \
522 		__res; \
523 	})
524 
525 /* There are so many possible reasons for this function to return
526  * VM86_INTx, so adding another doesn't bother me. We can expect
527  * userspace programs to be able to handle it. (Getting a problem
528  * in userspace is always better than an Oops anyway.) [KD]
529  */
530 static void do_int(struct kernel_vm86_regs *regs, int i,
531     unsigned char __user *ssp, unsigned short sp)
532 {
533 	unsigned long __user *intr_ptr;
534 	unsigned long segoffs;
535 	struct vm86 *vm86 = current->thread.vm86;
536 
537 	if (regs->pt.cs == BIOSSEG)
538 		goto cannot_handle;
539 	if (is_revectored(i, &vm86->int_revectored))
540 		goto cannot_handle;
541 	if (i == 0x21 && is_revectored(AH(regs), &vm86->int21_revectored))
542 		goto cannot_handle;
543 	intr_ptr = (unsigned long __user *) (i << 2);
544 	if (get_user(segoffs, intr_ptr))
545 		goto cannot_handle;
546 	if ((segoffs >> 16) == BIOSSEG)
547 		goto cannot_handle;
548 	pushw(ssp, sp, get_vflags(regs), cannot_handle);
549 	pushw(ssp, sp, regs->pt.cs, cannot_handle);
550 	pushw(ssp, sp, IP(regs), cannot_handle);
551 	regs->pt.cs = segoffs >> 16;
552 	SP(regs) -= 6;
553 	IP(regs) = segoffs & 0xffff;
554 	clear_TF(regs);
555 	clear_IF(regs);
556 	clear_AC(regs);
557 	return;
558 
559 cannot_handle:
560 	save_v86_state(regs, VM86_INTx + (i << 8));
561 }
562 
563 int handle_vm86_trap(struct kernel_vm86_regs *regs, long error_code, int trapno)
564 {
565 	struct vm86 *vm86 = current->thread.vm86;
566 
567 	if (vm86->vm86plus.is_vm86pus) {
568 		if ((trapno == 3) || (trapno == 1)) {
569 			save_v86_state(regs, VM86_TRAP + (trapno << 8));
570 			return 0;
571 		}
572 		do_int(regs, trapno, (unsigned char __user *) (regs->pt.ss << 4), SP(regs));
573 		return 0;
574 	}
575 	if (trapno != 1)
576 		return 1; /* we let this handle by the calling routine */
577 	current->thread.trap_nr = trapno;
578 	current->thread.error_code = error_code;
579 	force_sig(SIGTRAP, current);
580 	return 0;
581 }
582 
583 void handle_vm86_fault(struct kernel_vm86_regs *regs, long error_code)
584 {
585 	unsigned char opcode;
586 	unsigned char __user *csp;
587 	unsigned char __user *ssp;
588 	unsigned short ip, sp, orig_flags;
589 	int data32, pref_done;
590 	struct vm86plus_info_struct *vmpi = &current->thread.vm86->vm86plus;
591 
592 #define CHECK_IF_IN_TRAP \
593 	if (vmpi->vm86dbg_active && vmpi->vm86dbg_TFpendig) \
594 		newflags |= X86_EFLAGS_TF
595 
596 	orig_flags = *(unsigned short *)&regs->pt.flags;
597 
598 	csp = (unsigned char __user *) (regs->pt.cs << 4);
599 	ssp = (unsigned char __user *) (regs->pt.ss << 4);
600 	sp = SP(regs);
601 	ip = IP(regs);
602 
603 	data32 = 0;
604 	pref_done = 0;
605 	do {
606 		switch (opcode = popb(csp, ip, simulate_sigsegv)) {
607 		case 0x66:      /* 32-bit data */     data32 = 1; break;
608 		case 0x67:      /* 32-bit address */  break;
609 		case 0x2e:      /* CS */              break;
610 		case 0x3e:      /* DS */              break;
611 		case 0x26:      /* ES */              break;
612 		case 0x36:      /* SS */              break;
613 		case 0x65:      /* GS */              break;
614 		case 0x64:      /* FS */              break;
615 		case 0xf2:      /* repnz */       break;
616 		case 0xf3:      /* rep */             break;
617 		default: pref_done = 1;
618 		}
619 	} while (!pref_done);
620 
621 	switch (opcode) {
622 
623 	/* pushf */
624 	case 0x9c:
625 		if (data32) {
626 			pushl(ssp, sp, get_vflags(regs), simulate_sigsegv);
627 			SP(regs) -= 4;
628 		} else {
629 			pushw(ssp, sp, get_vflags(regs), simulate_sigsegv);
630 			SP(regs) -= 2;
631 		}
632 		IP(regs) = ip;
633 		goto vm86_fault_return;
634 
635 	/* popf */
636 	case 0x9d:
637 		{
638 		unsigned long newflags;
639 		if (data32) {
640 			newflags = popl(ssp, sp, simulate_sigsegv);
641 			SP(regs) += 4;
642 		} else {
643 			newflags = popw(ssp, sp, simulate_sigsegv);
644 			SP(regs) += 2;
645 		}
646 		IP(regs) = ip;
647 		CHECK_IF_IN_TRAP;
648 		if (data32)
649 			set_vflags_long(newflags, regs);
650 		else
651 			set_vflags_short(newflags, regs);
652 
653 		goto check_vip;
654 		}
655 
656 	/* int xx */
657 	case 0xcd: {
658 		int intno = popb(csp, ip, simulate_sigsegv);
659 		IP(regs) = ip;
660 		if (vmpi->vm86dbg_active) {
661 			if ((1 << (intno & 7)) & vmpi->vm86dbg_intxxtab[intno >> 3]) {
662 				save_v86_state(regs, VM86_INTx + (intno << 8));
663 				return;
664 			}
665 		}
666 		do_int(regs, intno, ssp, sp);
667 		return;
668 	}
669 
670 	/* iret */
671 	case 0xcf:
672 		{
673 		unsigned long newip;
674 		unsigned long newcs;
675 		unsigned long newflags;
676 		if (data32) {
677 			newip = popl(ssp, sp, simulate_sigsegv);
678 			newcs = popl(ssp, sp, simulate_sigsegv);
679 			newflags = popl(ssp, sp, simulate_sigsegv);
680 			SP(regs) += 12;
681 		} else {
682 			newip = popw(ssp, sp, simulate_sigsegv);
683 			newcs = popw(ssp, sp, simulate_sigsegv);
684 			newflags = popw(ssp, sp, simulate_sigsegv);
685 			SP(regs) += 6;
686 		}
687 		IP(regs) = newip;
688 		regs->pt.cs = newcs;
689 		CHECK_IF_IN_TRAP;
690 		if (data32) {
691 			set_vflags_long(newflags, regs);
692 		} else {
693 			set_vflags_short(newflags, regs);
694 		}
695 		goto check_vip;
696 		}
697 
698 	/* cli */
699 	case 0xfa:
700 		IP(regs) = ip;
701 		clear_IF(regs);
702 		goto vm86_fault_return;
703 
704 	/* sti */
705 	/*
706 	 * Damn. This is incorrect: the 'sti' instruction should actually
707 	 * enable interrupts after the /next/ instruction. Not good.
708 	 *
709 	 * Probably needs some horsing around with the TF flag. Aiee..
710 	 */
711 	case 0xfb:
712 		IP(regs) = ip;
713 		set_IF(regs);
714 		goto check_vip;
715 
716 	default:
717 		save_v86_state(regs, VM86_UNKNOWN);
718 	}
719 
720 	return;
721 
722 check_vip:
723 	if (VEFLAGS & X86_EFLAGS_VIP) {
724 		save_v86_state(regs, VM86_STI);
725 		return;
726 	}
727 
728 vm86_fault_return:
729 	if (vmpi->force_return_for_pic  && (VEFLAGS & (X86_EFLAGS_IF | X86_EFLAGS_VIF))) {
730 		save_v86_state(regs, VM86_PICRETURN);
731 		return;
732 	}
733 	if (orig_flags & X86_EFLAGS_TF)
734 		handle_vm86_trap(regs, 0, X86_TRAP_DB);
735 	return;
736 
737 simulate_sigsegv:
738 	/* FIXME: After a long discussion with Stas we finally
739 	 *        agreed, that this is wrong. Here we should
740 	 *        really send a SIGSEGV to the user program.
741 	 *        But how do we create the correct context? We
742 	 *        are inside a general protection fault handler
743 	 *        and has just returned from a page fault handler.
744 	 *        The correct context for the signal handler
745 	 *        should be a mixture of the two, but how do we
746 	 *        get the information? [KD]
747 	 */
748 	save_v86_state(regs, VM86_UNKNOWN);
749 }
750 
751 /* ---------------- vm86 special IRQ passing stuff ----------------- */
752 
753 #define VM86_IRQNAME		"vm86irq"
754 
755 static struct vm86_irqs {
756 	struct task_struct *tsk;
757 	int sig;
758 } vm86_irqs[16];
759 
760 static DEFINE_SPINLOCK(irqbits_lock);
761 static int irqbits;
762 
763 #define ALLOWED_SIGS (1 /* 0 = don't send a signal */ \
764 	| (1 << SIGUSR1) | (1 << SIGUSR2) | (1 << SIGIO)  | (1 << SIGURG) \
765 	| (1 << SIGUNUSED))
766 
767 static irqreturn_t irq_handler(int intno, void *dev_id)
768 {
769 	int irq_bit;
770 	unsigned long flags;
771 
772 	spin_lock_irqsave(&irqbits_lock, flags);
773 	irq_bit = 1 << intno;
774 	if ((irqbits & irq_bit) || !vm86_irqs[intno].tsk)
775 		goto out;
776 	irqbits |= irq_bit;
777 	if (vm86_irqs[intno].sig)
778 		send_sig(vm86_irqs[intno].sig, vm86_irqs[intno].tsk, 1);
779 	/*
780 	 * IRQ will be re-enabled when user asks for the irq (whether
781 	 * polling or as a result of the signal)
782 	 */
783 	disable_irq_nosync(intno);
784 	spin_unlock_irqrestore(&irqbits_lock, flags);
785 	return IRQ_HANDLED;
786 
787 out:
788 	spin_unlock_irqrestore(&irqbits_lock, flags);
789 	return IRQ_NONE;
790 }
791 
792 static inline void free_vm86_irq(int irqnumber)
793 {
794 	unsigned long flags;
795 
796 	free_irq(irqnumber, NULL);
797 	vm86_irqs[irqnumber].tsk = NULL;
798 
799 	spin_lock_irqsave(&irqbits_lock, flags);
800 	irqbits &= ~(1 << irqnumber);
801 	spin_unlock_irqrestore(&irqbits_lock, flags);
802 }
803 
804 void release_vm86_irqs(struct task_struct *task)
805 {
806 	int i;
807 	for (i = FIRST_VM86_IRQ ; i <= LAST_VM86_IRQ; i++)
808 	    if (vm86_irqs[i].tsk == task)
809 		free_vm86_irq(i);
810 }
811 
812 static inline int get_and_reset_irq(int irqnumber)
813 {
814 	int bit;
815 	unsigned long flags;
816 	int ret = 0;
817 
818 	if (invalid_vm86_irq(irqnumber)) return 0;
819 	if (vm86_irqs[irqnumber].tsk != current) return 0;
820 	spin_lock_irqsave(&irqbits_lock, flags);
821 	bit = irqbits & (1 << irqnumber);
822 	irqbits &= ~bit;
823 	if (bit) {
824 		enable_irq(irqnumber);
825 		ret = 1;
826 	}
827 
828 	spin_unlock_irqrestore(&irqbits_lock, flags);
829 	return ret;
830 }
831 
832 
833 static int do_vm86_irq_handling(int subfunction, int irqnumber)
834 {
835 	int ret;
836 	switch (subfunction) {
837 		case VM86_GET_AND_RESET_IRQ: {
838 			return get_and_reset_irq(irqnumber);
839 		}
840 		case VM86_GET_IRQ_BITS: {
841 			return irqbits;
842 		}
843 		case VM86_REQUEST_IRQ: {
844 			int sig = irqnumber >> 8;
845 			int irq = irqnumber & 255;
846 			if (!capable(CAP_SYS_ADMIN)) return -EPERM;
847 			if (!((1 << sig) & ALLOWED_SIGS)) return -EPERM;
848 			if (invalid_vm86_irq(irq)) return -EPERM;
849 			if (vm86_irqs[irq].tsk) return -EPERM;
850 			ret = request_irq(irq, &irq_handler, 0, VM86_IRQNAME, NULL);
851 			if (ret) return ret;
852 			vm86_irqs[irq].sig = sig;
853 			vm86_irqs[irq].tsk = current;
854 			return irq;
855 		}
856 		case  VM86_FREE_IRQ: {
857 			if (invalid_vm86_irq(irqnumber)) return -EPERM;
858 			if (!vm86_irqs[irqnumber].tsk) return 0;
859 			if (vm86_irqs[irqnumber].tsk != current) return -EPERM;
860 			free_vm86_irq(irqnumber);
861 			return 0;
862 		}
863 	}
864 	return -EINVAL;
865 }
866 
867