arch/i386/kernel/vm86.c

   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 <manfred@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/capability.h>
  34 #include <linux/config.h>
  35 #include <linux/errno.h>
  36 #include <linux/interrupt.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/smp_lock.h>
  44 #include <linux/highmem.h>
  45 #include <linux/ptrace.h>
  46 #include <linux/audit.h>
  47
  48 #include <asm/uaccess.h>
  49 #include <asm/io.h>
  50 #include <asm/tlbflush.h>
  51 #include <asm/irq.h>
  52
  53 /*
  54  * Known problems:
  55  *
  56  * Interrupt handling is not guaranteed:
  57  * - a real x86 will disable all interrupts for one instruction
  58  *   after a "mov ss,xx" to make stack handling atomic even without
  59  *   the 'lss' instruction. We can't guarantee this in v86 mode,
  60  *   as the next instruction might result in a page fault or similar.
  61  * - a real x86 will have interrupts disabled for one instruction
  62  *   past the 'sti' that enables them. We don't bother with all the
  63  *   details yet.
  64  *
  65  * Let's hope these problems do not actually matter for anything.
  66  */
  67
  68
  69 #define KVM86   ((struct kernel_vm86_struct *)regs)
  70 #define VMPI    KVM86->vm86plus
  71
  72
  73 /*
  74  * 8- and 16-bit register defines..
  75  */
  76 #define AL(regs)        (((unsigned char *)&((regs)->eax))[0])
  77 #define AH(regs)        (((unsigned char *)&((regs)->eax))[1])
  78 #define IP(regs)        (*(unsigned short *)&((regs)->eip))
  79 #define SP(regs)        (*(unsigned short *)&((regs)->esp))
  80
  81 /*
  82  * virtual flags (16 and 32-bit versions)
  83  */
  84 #define VFLAGS  (*(unsigned short *)&(current->thread.v86flags))
  85 #define VEFLAGS (current->thread.v86flags)
  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 #define VM86_REGS_PART2 orig_eax
  94 #define VM86_REGS_SIZE1 \
  95         ( (unsigned)( & (((struct kernel_vm86_regs *)0)->VM86_REGS_PART2) ) )
  96 #define VM86_REGS_SIZE2 (sizeof(struct kernel_vm86_regs) - VM86_REGS_SIZE1)
  97
  98 struct pt_regs * FASTCALL(save_v86_state(struct kernel_vm86_regs * regs));
  99 struct pt_regs * fastcall save_v86_state(struct kernel_vm86_regs * regs)
 100 {
 101         struct tss_struct *tss;
 102         struct pt_regs *ret;
 103         unsigned long tmp;
 104
 105         /*
 106          * This gets called from entry.S with interrupts disabled, but
 107          * from process context. Enable interrupts here, before trying
 108          * to access user space.
 109          */
 110         local_irq_enable();
 111
 112         if (!current->thread.vm86_info) {
 113                 printk("no vm86_info: BAD\n");
 114                 do_exit(SIGSEGV);
 115         }
 116         set_flags(regs->eflags, VEFLAGS, VIF_MASK | current->thread.v86mask);
 117         tmp = copy_to_user(&current->thread.vm86_info->regs,regs, VM86_REGS_SIZE1);
 118         tmp += copy_to_user(&current->thread.vm86_info->regs.VM86_REGS_PART2,
 119                 &regs->VM86_REGS_PART2, VM86_REGS_SIZE2);
 120         tmp += put_user(current->thread.screen_bitmap,&current->thread.vm86_info->screen_bitmap);
 121         if (tmp) {
 122                 printk("vm86: could not access userspace vm86_info\n");
 123                 do_exit(SIGSEGV);
 124         }
 125
 126         tss = &per_cpu(init_tss, get_cpu());
 127         current->thread.esp0 = current->thread.saved_esp0;
 128         current->thread.sysenter_cs = __KERNEL_CS;
 129         load_esp0(tss, &current->thread);
 130         current->thread.saved_esp0 = 0;
 131         put_cpu();
 132
 133         loadsegment(fs, current->thread.saved_fs);
 134         loadsegment(gs, current->thread.saved_gs);
 135         ret = KVM86->regs32;
 136         return ret;
 137 }
 138
 139 static void mark_screen_rdonly(struct mm_struct *mm)
 140 {
 141         pgd_t *pgd;
 142         pud_t *pud;
 143         pmd_t *pmd;
 144         pte_t *pte;
 145         spinlock_t *ptl;
 146         int i;
 147
 148         pgd = pgd_offset(mm, 0xA0000);
 149         if (pgd_none_or_clear_bad(pgd))
 150                 goto out;
 151         pud = pud_offset(pgd, 0xA0000);
 152         if (pud_none_or_clear_bad(pud))
 153                 goto out;
 154         pmd = pmd_offset(pud, 0xA0000);
 155         if (pmd_none_or_clear_bad(pmd))
 156                 goto out;
 157         pte = pte_offset_map_lock(mm, pmd, 0xA0000, &ptl);
 158         for (i = 0; i < 32; i++) {
 159                 if (pte_present(*pte))
 160                         set_pte(pte, pte_wrprotect(*pte));
 161                 pte++;
 162         }
 163         pte_unmap_unlock(pte, ptl);
 164 out:
 165         flush_tlb();
 166 }
 167
 168
 169
 170 static int do_vm86_irq_handling(int subfunction, int irqnumber);
 171 static void do_sys_vm86(struct kernel_vm86_struct *info, struct task_struct *tsk);
 172
 173 asmlinkage int sys_vm86old(struct pt_regs regs)
 174 {
 175         struct vm86_struct __user *v86 = (struct vm86_struct __user *)regs.ebx;
 176         struct kernel_vm86_struct info; /* declare this _on top_,
 177                                          * this avoids wasting of stack space.
 178                                          * This remains on the stack until we
 179                                          * return to 32 bit user space.
 180                                          */
 181         struct task_struct *tsk;
 182         int tmp, ret = -EPERM;
 183
 184         tsk = current;
 185         if (tsk->thread.saved_esp0)
 186                 goto out;
 187         tmp  = copy_from_user(&info, v86, VM86_REGS_SIZE1);
 188         tmp += copy_from_user(&info.regs.VM86_REGS_PART2, &v86->regs.VM86_REGS_PART2,
 189                 (long)&info.vm86plus - (long)&info.regs.VM86_REGS_PART2);
 190         ret = -EFAULT;
 191         if (tmp)
 192                 goto out;
 193         memset(&info.vm86plus, 0, (int)&info.regs32 - (int)&info.vm86plus);
 194         info.regs32 = &regs;
 195         tsk->thread.vm86_info = v86;
 196         do_sys_vm86(&info, tsk);
 197         ret = 0;        /* we never return here */
 198 out:
 199         return ret;
 200 }
 201
 202
 203 asmlinkage int sys_vm86(struct pt_regs regs)
 204 {
 205         struct kernel_vm86_struct info; /* declare this _on top_,
 206                                          * this avoids wasting of stack space.
 207                                          * This remains on the stack until we
 208                                          * return to 32 bit user space.
 209                                          */
 210         struct task_struct *tsk;
 211         int tmp, ret;
 212         struct vm86plus_struct __user *v86;
 213
 214         tsk = current;
 215         switch (regs.ebx) {
 216                 case VM86_REQUEST_IRQ:
 217                 case VM86_FREE_IRQ:
 218                 case VM86_GET_IRQ_BITS:
 219                 case VM86_GET_AND_RESET_IRQ:
 220                         ret = do_vm86_irq_handling(regs.ebx, (int)regs.ecx);
 221                         goto out;
 222                 case VM86_PLUS_INSTALL_CHECK:
 223                         /* NOTE: on old vm86 stuff this will return the error
 224                            from access_ok(), because the subfunction is
 225                            interpreted as (invalid) address to vm86_struct.
 226                            So the installation check works.
 227                          */
 228                         ret = 0;
 229                         goto out;
 230         }
 231
 232         /* we come here only for functions VM86_ENTER, VM86_ENTER_NO_BYPASS */
 233         ret = -EPERM;
 234         if (tsk->thread.saved_esp0)
 235                 goto out;
 236         v86 = (struct vm86plus_struct __user *)regs.ecx;
 237         tmp  = copy_from_user(&info, v86, VM86_REGS_SIZE1);
 238         tmp += copy_from_user(&info.regs.VM86_REGS_PART2, &v86->regs.VM86_REGS_PART2,
 239                 (long)&info.regs32 - (long)&info.regs.VM86_REGS_PART2);
 240         ret = -EFAULT;
 241         if (tmp)
 242                 goto out;
 243         info.regs32 = &regs;
 244         info.vm86plus.is_vm86pus = 1;
 245         tsk->thread.vm86_info = (struct vm86_struct __user *)v86;
 246         do_sys_vm86(&info, tsk);
 247         ret = 0;        /* we never return here */
 248 out:
 249         return ret;
 250 }
 251
 252
 253 static void do_sys_vm86(struct kernel_vm86_struct *info, struct task_struct *tsk)
 254 {
 255         struct tss_struct *tss;
 256         long eax;
 257 /*
 258  * make sure the vm86() system call doesn't try to do anything silly
 259  */
 260         info->regs.__null_ds = 0;
 261         info->regs.__null_es = 0;
 262
 263 /* we are clearing fs,gs later just before "jmp resume_userspace",
 264  * because starting with Linux 2.1.x they aren't no longer saved/restored
 265  */
 266
 267 /*
 268  * The eflags register is also special: we cannot trust that the user
 269  * has set it up safely, so this makes sure interrupt etc flags are
 270  * inherited from protected mode.
 271  */
 272         VEFLAGS = info->regs.eflags;
 273         info->regs.eflags &= SAFE_MASK;
 274         info->regs.eflags |= info->regs32->eflags & ~SAFE_MASK;
 275         info->regs.eflags |= VM_MASK;
 276
 277         switch (info->cpu_type) {
 278                 case CPU_286:
 279                         tsk->thread.v86mask = 0;
 280                         break;
 281                 case CPU_386:
 282                         tsk->thread.v86mask = NT_MASK | IOPL_MASK;
 283                         break;
 284                 case CPU_486:
 285                         tsk->thread.v86mask = AC_MASK | NT_MASK | IOPL_MASK;
 286                         break;
 287                 default:
 288                         tsk->thread.v86mask = ID_MASK | AC_MASK | NT_MASK | IOPL_MASK;
 289                         break;
 290         }
 291
 292 /*
 293  * Save old state, set default return value (%eax) to 0
 294  */
 295         info->regs32->eax = 0;
 296         tsk->thread.saved_esp0 = tsk->thread.esp0;
 297         savesegment(fs, tsk->thread.saved_fs);
 298         savesegment(gs, tsk->thread.saved_gs);
 299
 300         tss = &per_cpu(init_tss, get_cpu());
 301         tsk->thread.esp0 = (unsigned long) &info->VM86_TSS_ESP0;
 302         if (cpu_has_sep)
 303                 tsk->thread.sysenter_cs = 0;
 304         load_esp0(tss, &tsk->thread);
 305         put_cpu();
 306
 307         tsk->thread.screen_bitmap = info->screen_bitmap;
 308         if (info->flags & VM86_SCREEN_BITMAP)
 309                 mark_screen_rdonly(tsk->mm);
 310         __asm__ __volatile__("xorl %eax,%eax; movl %eax,%fs; movl %eax,%gs\n\t");
 311         __asm__ __volatile__("movl %%eax, %0\n" :"=r"(eax));
 312
 313         /*call audit_syscall_exit since we do not exit via the normal paths */
 314         if (unlikely(current->audit_context))
 315                 audit_syscall_exit(AUDITSC_RESULT(eax), eax);
 316
 317         __asm__ __volatile__(
 318                 "movl %0,%%esp\n\t"
 319                 "movl %1,%%ebp\n\t"
 320                 "jmp resume_userspace"
 321                 : /* no outputs */
 322                 :"r" (&info->regs), "r" (task_thread_info(tsk)));
 323         /* we never return here */
 324 }
 325
 326 static inline void return_to_32bit(struct kernel_vm86_regs * regs16, int retval)
 327 {
 328         struct pt_regs * regs32;
 329
 330         regs32 = save_v86_state(regs16);
 331         regs32->eax = retval;
 332         __asm__ __volatile__("movl %0,%%esp\n\t"
 333                 "movl %1,%%ebp\n\t"
 334                 "jmp resume_userspace"
 335                 : : "r" (regs32), "r" (current_thread_info()));
 336 }
 337
 338 static inline void set_IF(struct kernel_vm86_regs * regs)
 339 {
 340         VEFLAGS |= VIF_MASK;
 341         if (VEFLAGS & VIP_MASK)
 342                 return_to_32bit(regs, VM86_STI);
 343 }
 344
 345 static inline void clear_IF(struct kernel_vm86_regs * regs)
 346 {
 347         VEFLAGS &= ~VIF_MASK;
 348 }
 349
 350 static inline void clear_TF(struct kernel_vm86_regs * regs)
 351 {
 352         regs->eflags &= ~TF_MASK;
 353 }
 354
 355 static inline void clear_AC(struct kernel_vm86_regs * regs)
 356 {
 357         regs->eflags &= ~AC_MASK;
 358 }
 359
 360 /* It is correct to call set_IF(regs) from the set_vflags_*
 361  * functions. However someone forgot to call clear_IF(regs)
 362  * in the opposite case.
 363  * After the command sequence CLI PUSHF STI POPF you should
 364  * end up with interrups disabled, but you ended up with
 365  * interrupts enabled.
 366  *  ( I was testing my own changes, but the only bug I
 367  *    could find was in a function I had not changed. )
 368  * [KD]
 369  */
 370
 371 static inline void set_vflags_long(unsigned long eflags, struct kernel_vm86_regs * regs)
 372 {
 373         set_flags(VEFLAGS, eflags, current->thread.v86mask);
 374         set_flags(regs->eflags, eflags, SAFE_MASK);
 375         if (eflags & IF_MASK)
 376                 set_IF(regs);
 377         else
 378                 clear_IF(regs);
 379 }
 380
 381 static inline void set_vflags_short(unsigned short flags, struct kernel_vm86_regs * regs)
 382 {
 383         set_flags(VFLAGS, flags, current->thread.v86mask);
 384         set_flags(regs->eflags, flags, SAFE_MASK);
 385         if (flags & IF_MASK)
 386                 set_IF(regs);
 387         else
 388                 clear_IF(regs);
 389 }
 390
 391 static inline unsigned long get_vflags(struct kernel_vm86_regs * regs)
 392 {
 393         unsigned long flags = regs->eflags & RETURN_MASK;
 394
 395         if (VEFLAGS & VIF_MASK)
 396                 flags |= IF_MASK;
 397         flags |= IOPL_MASK;
 398         return flags | (VEFLAGS & current->thread.v86mask);
 399 }
 400
 401 static inline int is_revectored(int nr, struct revectored_struct * bitmap)
 402 {
 403         __asm__ __volatile__("btl %2,%1\n\tsbbl %0,%0"
 404                 :"=r" (nr)
 405                 :"m" (*bitmap),"r" (nr));
 406         return nr;
 407 }
 408
 409 #define val_byte(val, n) (((__u8 *)&val)[n])
 410
 411 #define pushb(base, ptr, val, err_label) \
 412         do { \
 413                 __u8 __val = val; \
 414                 ptr--; \
 415                 if (put_user(__val, base + ptr) < 0) \
 416                         goto err_label; \
 417         } while(0)
 418
 419 #define pushw(base, ptr, val, err_label) \
 420         do { \
 421                 __u16 __val = val; \
 422                 ptr--; \
 423                 if (put_user(val_byte(__val, 1), base + ptr) < 0) \
 424                         goto err_label; \
 425                 ptr--; \
 426                 if (put_user(val_byte(__val, 0), base + ptr) < 0) \
 427                         goto err_label; \
 428         } while(0)
 429
 430 #define pushl(base, ptr, val, err_label) \
 431         do { \
 432                 __u32 __val = val; \
 433                 ptr--; \
 434                 if (put_user(val_byte(__val, 3), base + ptr) < 0) \
 435                         goto err_label; \
 436                 ptr--; \
 437                 if (put_user(val_byte(__val, 2), base + ptr) < 0) \
 438                         goto err_label; \
 439                 ptr--; \
 440                 if (put_user(val_byte(__val, 1), base + ptr) < 0) \
 441                         goto err_label; \
 442                 ptr--; \
 443                 if (put_user(val_byte(__val, 0), base + ptr) < 0) \
 444                         goto err_label; \
 445         } while(0)
 446
 447 #define popb(base, ptr, err_label) \
 448         ({ \
 449                 __u8 __res; \
 450                 if (get_user(__res, base + ptr) < 0) \
 451                         goto err_label; \
 452                 ptr++; \
 453                 __res; \
 454         })
 455
 456 #define popw(base, ptr, err_label) \
 457         ({ \
 458                 __u16 __res; \
 459                 if (get_user(val_byte(__res, 0), base + ptr) < 0) \
 460                         goto err_label; \
 461                 ptr++; \
 462                 if (get_user(val_byte(__res, 1), base + ptr) < 0) \
 463                         goto err_label; \
 464                 ptr++; \
 465                 __res; \
 466         })
 467
 468 #define popl(base, ptr, err_label) \
 469         ({ \
 470                 __u32 __res; \
 471                 if (get_user(val_byte(__res, 0), base + ptr) < 0) \
 472                         goto err_label; \
 473                 ptr++; \
 474                 if (get_user(val_byte(__res, 1), base + ptr) < 0) \
 475                         goto err_label; \
 476                 ptr++; \
 477                 if (get_user(val_byte(__res, 2), base + ptr) < 0) \
 478                         goto err_label; \
 479                 ptr++; \
 480                 if (get_user(val_byte(__res, 3), base + ptr) < 0) \
 481                         goto err_label; \
 482                 ptr++; \
 483                 __res; \
 484         })
 485
 486 /* There are so many possible reasons for this function to return
 487  * VM86_INTx, so adding another doesn't bother me. We can expect
 488  * userspace programs to be able to handle it. (Getting a problem
 489  * in userspace is always better than an Oops anyway.) [KD]
 490  */
 491 static void do_int(struct kernel_vm86_regs *regs, int i,
 492     unsigned char __user * ssp, unsigned short sp)
 493 {
 494         unsigned long __user *intr_ptr;
 495         unsigned long segoffs;
 496
 497         if (regs->cs == BIOSSEG)
 498                 goto cannot_handle;
 499         if (is_revectored(i, &KVM86->int_revectored))
 500                 goto cannot_handle;
 501         if (i==0x21 && is_revectored(AH(regs),&KVM86->int21_revectored))
 502                 goto cannot_handle;
 503         intr_ptr = (unsigned long __user *) (i << 2);
 504         if (get_user(segoffs, intr_ptr))
 505                 goto cannot_handle;
 506         if ((segoffs >> 16) == BIOSSEG)
 507                 goto cannot_handle;
 508         pushw(ssp, sp, get_vflags(regs), cannot_handle);
 509         pushw(ssp, sp, regs->cs, cannot_handle);
 510         pushw(ssp, sp, IP(regs), cannot_handle);
 511         regs->cs = segoffs >> 16;
 512         SP(regs) -= 6;
 513         IP(regs) = segoffs & 0xffff;
 514         clear_TF(regs);
 515         clear_IF(regs);
 516         clear_AC(regs);
 517         return;
 518
 519 cannot_handle:
 520         return_to_32bit(regs, VM86_INTx + (i << 8));
 521 }
 522
 523 int handle_vm86_trap(struct kernel_vm86_regs * regs, long error_code, int trapno)
 524 {
 525         if (VMPI.is_vm86pus) {
 526                 if ( (trapno==3) || (trapno==1) )
 527                         return_to_32bit(regs, VM86_TRAP + (trapno << 8));
 528                 do_int(regs, trapno, (unsigned char __user *) (regs->ss << 4), SP(regs));
 529                 return 0;
 530         }
 531         if (trapno !=1)
 532                 return 1; /* we let this handle by the calling routine */
 533         if (current->ptrace & PT_PTRACED) {
 534                 unsigned long flags;
 535                 spin_lock_irqsave(&current->sighand->siglock, flags);
 536                 sigdelset(&current->blocked, SIGTRAP);
 537                 recalc_sigpending();
 538                 spin_unlock_irqrestore(&current->sighand->siglock, flags);
 539         }
 540         send_sig(SIGTRAP, current, 1);
 541         current->thread.trap_no = trapno;
 542         current->thread.error_code = error_code;
 543         return 0;
 544 }
 545
 546 void handle_vm86_fault(struct kernel_vm86_regs * regs, long error_code)
 547 {
 548         unsigned char opcode;
 549         unsigned char __user *csp;
 550         unsigned char __user *ssp;
 551         unsigned short ip, sp, orig_flags;
 552         int data32, pref_done;
 553
 554 #define CHECK_IF_IN_TRAP \
 555         if (VMPI.vm86dbg_active && VMPI.vm86dbg_TFpendig) \
 556                 newflags |= TF_MASK
 557 #define VM86_FAULT_RETURN do { \
 558         if (VMPI.force_return_for_pic  && (VEFLAGS & (IF_MASK | VIF_MASK))) \
 559                 return_to_32bit(regs, VM86_PICRETURN); \
 560         if (orig_flags & TF_MASK) \
 561                 handle_vm86_trap(regs, 0, 1); \
 562         return; } while (0)
 563
 564         orig_flags = *(unsigned short *)&regs->eflags;
 565
 566         csp = (unsigned char __user *) (regs->cs << 4);
 567         ssp = (unsigned char __user *) (regs->ss << 4);
 568         sp = SP(regs);
 569         ip = IP(regs);
 570
 571         data32 = 0;
 572         pref_done = 0;
 573         do {
 574                 switch (opcode = popb(csp, ip, simulate_sigsegv)) {
 575                         case 0x66:      /* 32-bit data */     data32=1; break;
 576                         case 0x67:      /* 32-bit address */  break;
 577                         case 0x2e:      /* CS */              break;
 578                         case 0x3e:      /* DS */              break;
 579                         case 0x26:      /* ES */              break;
 580                         case 0x36:      /* SS */              break;
 581                         case 0x65:      /* GS */              break;
 582                         case 0x64:      /* FS */              break;
 583                         case 0xf2:      /* repnz */       break;
 584                         case 0xf3:      /* rep */             break;
 585                         default: pref_done = 1;
 586                 }
 587         } while (!pref_done);
 588
 589         switch (opcode) {
 590
 591         /* pushf */
 592         case 0x9c:
 593                 if (data32) {
 594                         pushl(ssp, sp, get_vflags(regs), simulate_sigsegv);
 595                         SP(regs) -= 4;
 596                 } else {
 597                         pushw(ssp, sp, get_vflags(regs), simulate_sigsegv);
 598                         SP(regs) -= 2;
 599                 }
 600                 IP(regs) = ip;
 601                 VM86_FAULT_RETURN;
 602
 603         /* popf */
 604         case 0x9d:
 605                 {
 606                 unsigned long newflags;
 607                 if (data32) {
 608                         newflags=popl(ssp, sp, simulate_sigsegv);
 609                         SP(regs) += 4;
 610                 } else {
 611                         newflags = popw(ssp, sp, simulate_sigsegv);
 612                         SP(regs) += 2;
 613                 }
 614                 IP(regs) = ip;
 615                 CHECK_IF_IN_TRAP;
 616                 if (data32) {
 617                         set_vflags_long(newflags, regs);
 618                 } else {
 619                         set_vflags_short(newflags, regs);
 620                 }
 621                 VM86_FAULT_RETURN;
 622                 }
 623
 624         /* int xx */
 625         case 0xcd: {
 626                 int intno=popb(csp, ip, simulate_sigsegv);
 627                 IP(regs) = ip;
 628                 if (VMPI.vm86dbg_active) {
 629                         if ( (1 << (intno &7)) & VMPI.vm86dbg_intxxtab[intno >> 3] )
 630                                 return_to_32bit(regs, VM86_INTx + (intno << 8));
 631                 }
 632                 do_int(regs, intno, ssp, sp);
 633                 return;
 634         }
 635
 636         /* iret */
 637         case 0xcf:
 638                 {
 639                 unsigned long newip;
 640                 unsigned long newcs;
 641                 unsigned long newflags;
 642                 if (data32) {
 643                         newip=popl(ssp, sp, simulate_sigsegv);
 644                         newcs=popl(ssp, sp, simulate_sigsegv);
 645                         newflags=popl(ssp, sp, simulate_sigsegv);
 646                         SP(regs) += 12;
 647                 } else {
 648                         newip = popw(ssp, sp, simulate_sigsegv);
 649                         newcs = popw(ssp, sp, simulate_sigsegv);
 650                         newflags = popw(ssp, sp, simulate_sigsegv);
 651                         SP(regs) += 6;
 652                 }
 653                 IP(regs) = newip;
 654                 regs->cs = newcs;
 655                 CHECK_IF_IN_TRAP;
 656                 if (data32) {
 657                         set_vflags_long(newflags, regs);
 658                 } else {
 659                         set_vflags_short(newflags, regs);
 660                 }
 661                 VM86_FAULT_RETURN;
 662                 }
 663
 664         /* cli */
 665         case 0xfa:
 666                 IP(regs) = ip;
 667                 clear_IF(regs);
 668                 VM86_FAULT_RETURN;
 669
 670         /* sti */
 671         /*
 672          * Damn. This is incorrect: the 'sti' instruction should actually
 673          * enable interrupts after the /next/ instruction. Not good.
 674          *
 675          * Probably needs some horsing around with the TF flag. Aiee..
 676          */
 677         case 0xfb:
 678                 IP(regs) = ip;
 679                 set_IF(regs);
 680                 VM86_FAULT_RETURN;
 681
 682         default:
 683                 return_to_32bit(regs, VM86_UNKNOWN);
 684         }
 685
 686         return;
 687
 688 simulate_sigsegv:
 689         /* FIXME: After a long discussion with Stas we finally
 690          *        agreed, that this is wrong. Here we should
 691          *        really send a SIGSEGV to the user program.
 692          *        But how do we create the correct context? We
 693          *        are inside a general protection fault handler
 694          *        and has just returned from a page fault handler.
 695          *        The correct context for the signal handler
 696          *        should be a mixture of the two, but how do we
 697          *        get the information? [KD]
 698          */
 699         return_to_32bit(regs, VM86_UNKNOWN);
 700 }
 701
 702 /* ---------------- vm86 special IRQ passing stuff ----------------- */
 703
 704 #define VM86_IRQNAME            "vm86irq"
 705
 706 static struct vm86_irqs {
 707         struct task_struct *tsk;
 708         int sig;
 709 } vm86_irqs[16];
 710
 711 static DEFINE_SPINLOCK(irqbits_lock);
 712 static int irqbits;
 713
 714 #define ALLOWED_SIGS ( 1 /* 0 = don't send a signal */ \
 715         | (1 << SIGUSR1) | (1 << SIGUSR2) | (1 << SIGIO)  | (1 << SIGURG) \
 716         | (1 << SIGUNUSED) )
 717
 718 static irqreturn_t irq_handler(int intno, void *dev_id, struct pt_regs * regs)
 719 {
 720         int irq_bit;
 721         unsigned long flags;
 722
 723         spin_lock_irqsave(&irqbits_lock, flags);
 724         irq_bit = 1 << intno;
 725         if ((irqbits & irq_bit) || ! vm86_irqs[intno].tsk)
 726                 goto out;
 727         irqbits |= irq_bit;
 728         if (vm86_irqs[intno].sig)
 729                 send_sig(vm86_irqs[intno].sig, vm86_irqs[intno].tsk, 1);
 730         /*
 731          * IRQ will be re-enabled when user asks for the irq (whether
 732          * polling or as a result of the signal)
 733          */
 734         disable_irq_nosync(intno);
 735         spin_unlock_irqrestore(&irqbits_lock, flags);
 736         return IRQ_HANDLED;
 737
 738 out:
 739         spin_unlock_irqrestore(&irqbits_lock, flags);
 740         return IRQ_NONE;
 741 }
 742
 743 static inline void free_vm86_irq(int irqnumber)
 744 {
 745         unsigned long flags;
 746
 747         free_irq(irqnumber, NULL);
 748         vm86_irqs[irqnumber].tsk = NULL;
 749
 750         spin_lock_irqsave(&irqbits_lock, flags);
 751         irqbits &= ~(1 << irqnumber);
 752         spin_unlock_irqrestore(&irqbits_lock, flags);
 753 }
 754
 755 void release_vm86_irqs(struct task_struct *task)
 756 {
 757         int i;
 758         for (i = FIRST_VM86_IRQ ; i <= LAST_VM86_IRQ; i++)
 759             if (vm86_irqs[i].tsk == task)
 760                 free_vm86_irq(i);
 761 }
 762
 763 static inline int get_and_reset_irq(int irqnumber)
 764 {
 765         int bit;
 766         unsigned long flags;
 767         int ret = 0;
 768
 769         if (invalid_vm86_irq(irqnumber)) return 0;
 770         if (vm86_irqs[irqnumber].tsk != current) return 0;
 771         spin_lock_irqsave(&irqbits_lock, flags);
 772         bit = irqbits & (1 << irqnumber);
 773         irqbits &= ~bit;
 774         if (bit) {
 775                 enable_irq(irqnumber);
 776                 ret = 1;
 777         }
 778
 779         spin_unlock_irqrestore(&irqbits_lock, flags);
 780         return ret;
 781 }
 782
 783
 784 static int do_vm86_irq_handling(int subfunction, int irqnumber)
 785 {
 786         int ret;
 787         switch (subfunction) {
 788                 case VM86_GET_AND_RESET_IRQ: {
 789                         return get_and_reset_irq(irqnumber);
 790                 }
 791                 case VM86_GET_IRQ_BITS: {
 792                         return irqbits;
 793                 }
 794                 case VM86_REQUEST_IRQ: {
 795                         int sig = irqnumber >> 8;
 796                         int irq = irqnumber & 255;
 797                         if (!capable(CAP_SYS_ADMIN)) return -EPERM;
 798                         if (!((1 << sig) & ALLOWED_SIGS)) return -EPERM;
 799                         if (invalid_vm86_irq(irq)) return -EPERM;
 800                         if (vm86_irqs[irq].tsk) return -EPERM;
 801                         ret = request_irq(irq, &irq_handler, 0, VM86_IRQNAME, NULL);
 802                         if (ret) return ret;
 803                         vm86_irqs[irq].sig = sig;
 804                         vm86_irqs[irq].tsk = current;
 805                         return irq;
 806                 }
 807                 case  VM86_FREE_IRQ: {
 808                         if (invalid_vm86_irq(irqnumber)) return -EPERM;
 809                         if (!vm86_irqs[irqnumber].tsk) return 0;
 810                         if (vm86_irqs[irqnumber].tsk != current) return -EPERM;
 811                         free_vm86_irq(irqnumber);
 812                         return 0;
 813                 }
 814         }
 815         return -EINVAL;
 816 }
 817