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 <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/config.h>
  34 #include <linux/errno.h>
  35 #include <linux/interrupt.h>
  36 #include <linux/sched.h>
  37 #include <linux/kernel.h>
  38 #include <linux/signal.h>
  39 #include <linux/string.h>
  40 #include <linux/mm.h>
  41 #include <linux/smp.h>
  42 #include <linux/smp_lock.h>
  43 #include <linux/highmem.h>
  44 #include <linux/ptrace.h>
  45
  46 #include <asm/uaccess.h>
  47 #include <asm/io.h>
  48 #include <asm/tlbflush.h>
  49 #include <asm/irq.h>
  50
  51 /*
  52  * Known problems:
  53  *
  54  * Interrupt handling is not guaranteed:
  55  * - a real x86 will disable all interrupts for one instruction
  56  *   after a "mov ss,xx" to make stack handling atomic even without
  57  *   the 'lss' instruction. We can't guarantee this in v86 mode,
  58  *   as the next instruction might result in a page fault or similar.
  59  * - a real x86 will have interrupts disabled for one instruction
  60  *   past the 'sti' that enables them. We don't bother with all the
  61  *   details yet.
  62  *
  63  * Let's hope these problems do not actually matter for anything.
  64  */
  65
  66
  67 #define KVM86   ((struct kernel_vm86_struct *)regs)
  68 #define VMPI    KVM86->vm86plus
  69
  70
  71 /*
  72  * 8- and 16-bit register defines..
  73  */
  74 #define AL(regs)        (((unsigned char *)&((regs)->eax))[0])
  75 #define AH(regs)        (((unsigned char *)&((regs)->eax))[1])
  76 #define IP(regs)        (*(unsigned short *)&((regs)->eip))
  77 #define SP(regs)        (*(unsigned short *)&((regs)->esp))
  78
  79 /*
  80  * virtual flags (16 and 32-bit versions)
  81  */
  82 #define VFLAGS  (*(unsigned short *)&(current->thread.v86flags))
  83 #define VEFLAGS (current->thread.v86flags)
  84
  85 #define set_flags(X,new,mask) \
  86 ((X) = ((X) & ~(mask)) | ((new) & (mask)))
  87
  88 #define SAFE_MASK       (0xDD5)
  89 #define RETURN_MASK     (0xDFF)
  90
  91 #define VM86_REGS_PART2 orig_eax
  92 #define VM86_REGS_SIZE1 \
  93         ( (unsigned)( & (((struct kernel_vm86_regs *)0)->VM86_REGS_PART2) ) )
  94 #define VM86_REGS_SIZE2 (sizeof(struct kernel_vm86_regs) - VM86_REGS_SIZE1)
  95
  96 struct pt_regs * FASTCALL(save_v86_state(struct kernel_vm86_regs * regs));
  97 struct pt_regs * fastcall save_v86_state(struct kernel_vm86_regs * regs)
  98 {
  99         struct tss_struct *tss;
 100         struct pt_regs *ret;
 101         unsigned long tmp;
 102
 103         /*
 104          * This gets called from entry.S with interrupts disabled, but
 105          * from process context. Enable interrupts here, before trying
 106          * to access user space.
 107          */
 108         local_irq_enable();
 109
 110         if (!current->thread.vm86_info) {
 111                 printk("no vm86_info: BAD\n");
 112                 do_exit(SIGSEGV);
 113         }
 114         set_flags(regs->eflags, VEFLAGS, VIF_MASK | current->thread.v86mask);
 115         tmp = copy_to_user(&current->thread.vm86_info->regs,regs, VM86_REGS_SIZE1);
 116         tmp += copy_to_user(&current->thread.vm86_info->regs.VM86_REGS_PART2,
 117                 &regs->VM86_REGS_PART2, VM86_REGS_SIZE2);
 118         tmp += put_user(current->thread.screen_bitmap,&current->thread.vm86_info->screen_bitmap);
 119         if (tmp) {
 120                 printk("vm86: could not access userspace vm86_info\n");
 121                 do_exit(SIGSEGV);
 122         }
 123
 124         tss = &per_cpu(init_tss, get_cpu());
 125         current->thread.esp0 = current->thread.saved_esp0;
 126         current->thread.sysenter_cs = __KERNEL_CS;
 127         load_esp0(tss, &current->thread);
 128         current->thread.saved_esp0 = 0;
 129         put_cpu();
 130
 131         loadsegment(fs, current->thread.saved_fs);
 132         loadsegment(gs, current->thread.saved_gs);
 133         ret = KVM86->regs32;
 134         return ret;
 135 }
 136
 137 static void mark_screen_rdonly(struct task_struct * tsk)
 138 {
 139         pgd_t *pgd;
 140         pud_t *pud;
 141         pmd_t *pmd;
 142         pte_t *pte, *mapped;
 143         int i;
 144
 145         preempt_disable();
 146         spin_lock(&tsk->mm->page_table_lock);
 147         pgd = pgd_offset(tsk->mm, 0xA0000);
 148         if (pgd_none_or_clear_bad(pgd))
 149                 goto out;
 150         pud = pud_offset(pgd, 0xA0000);
 151         if (pud_none_or_clear_bad(pud))
 152                 goto out;
 153         pmd = pmd_offset(pud, 0xA0000);
 154         if (pmd_none_or_clear_bad(pmd))
 155                 goto out;
 156         pte = mapped = pte_offset_map(pmd, 0xA0000);
 157         for (i = 0; i < 32; i++) {
 158                 if (pte_present(*pte))
 159                         set_pte(pte, pte_wrprotect(*pte));
 160                 pte++;
 161         }
 162         pte_unmap(mapped);
 163 out:
 164         spin_unlock(&tsk->mm->page_table_lock);
 165         preempt_enable();
 166         flush_tlb();
 167 }
 168
 169
 170
 171 static int do_vm86_irq_handling(int subfunction, int irqnumber);
 172 static void do_sys_vm86(struct kernel_vm86_struct *info, struct task_struct *tsk);
 173
 174 asmlinkage int sys_vm86old(struct pt_regs regs)
 175 {
 176         struct vm86_struct __user *v86 = (struct vm86_struct __user *)regs.ebx;
 177         struct kernel_vm86_struct info; /* declare this _on top_,
 178                                          * this avoids wasting of stack space.
 179                                          * This remains on the stack until we
 180                                          * return to 32 bit user space.
 181                                          */
 182         struct task_struct *tsk;
 183         int tmp, ret = -EPERM;
 184
 185         tsk = current;
 186         if (tsk->thread.saved_esp0)
 187                 goto out;
 188         tmp  = copy_from_user(&info, v86, VM86_REGS_SIZE1);
 189         tmp += copy_from_user(&info.regs.VM86_REGS_PART2, &v86->regs.VM86_REGS_PART2,
 190                 (long)&info.vm86plus - (long)&info.regs.VM86_REGS_PART2);
 191         ret = -EFAULT;
 192         if (tmp)
 193                 goto out;
 194         memset(&info.vm86plus, 0, (int)&info.regs32 - (int)&info.vm86plus);
 195         info.regs32 = &regs;
 196         tsk->thread.vm86_info = v86;
 197         do_sys_vm86(&info, tsk);
 198         ret = 0;        /* we never return here */
 199 out:
 200         return ret;
 201 }
 202
 203
 204 asmlinkage int sys_vm86(struct pt_regs regs)
 205 {
 206         struct kernel_vm86_struct info; /* declare this _on top_,
 207                                          * this avoids wasting of stack space.
 208                                          * This remains on the stack until we
 209                                          * return to 32 bit user space.
 210                                          */
 211         struct task_struct *tsk;
 212         int tmp, ret;
 213         struct vm86plus_struct __user *v86;
 214
 215         tsk = current;
 216         switch (regs.ebx) {
 217                 case VM86_REQUEST_IRQ:
 218                 case VM86_FREE_IRQ:
 219                 case VM86_GET_IRQ_BITS:
 220                 case VM86_GET_AND_RESET_IRQ:
 221                         ret = do_vm86_irq_handling(regs.ebx, (int)regs.ecx);
 222                         goto out;
 223                 case VM86_PLUS_INSTALL_CHECK:
 224                         /* NOTE: on old vm86 stuff this will return the error
 225                            from verify_area(), because the subfunction is
 226                            interpreted as (invalid) address to vm86_struct.
 227                            So the installation check works.
 228                          */
 229                         ret = 0;
 230                         goto out;
 231         }
 232
 233         /* we come here only for functions VM86_ENTER, VM86_ENTER_NO_BYPASS */
 234         ret = -EPERM;
 235         if (tsk->thread.saved_esp0)
 236                 goto out;
 237         v86 = (struct vm86plus_struct __user *)regs.ecx;
 238         tmp  = copy_from_user(&info, v86, VM86_REGS_SIZE1);
 239         tmp += copy_from_user(&info.regs.VM86_REGS_PART2, &v86->regs.VM86_REGS_PART2,
 240                 (long)&info.regs32 - (long)&info.regs.VM86_REGS_PART2);
 241         ret = -EFAULT;
 242         if (tmp)
 243                 goto out;
 244         info.regs32 = &regs;
 245         info.vm86plus.is_vm86pus = 1;
 246         tsk->thread.vm86_info = (struct vm86_struct __user *)v86;
 247         do_sys_vm86(&info, tsk);
 248         ret = 0;        /* we never return here */
 249 out:
 250         return ret;
 251 }
 252
 253
 254 static void do_sys_vm86(struct kernel_vm86_struct *info, struct task_struct *tsk)
 255 {
 256         struct tss_struct *tss;
 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         asm volatile("movl %%fs,%0":"=m" (tsk->thread.saved_fs));
 298         asm volatile("movl %%gs,%0":"=m" (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);
 310         __asm__ __volatile__(
 311                 "xorl %%eax,%%eax; movl %%eax,%%fs; movl %%eax,%%gs\n\t"
 312                 "movl %0,%%esp\n\t"
 313                 "movl %1,%%ebp\n\t"
 314                 "jmp resume_userspace"
 315                 : /* no outputs */
 316                 :"r" (&info->regs), "r" (tsk->thread_info) : "ax");
 317         /* we never return here */
 318 }
 319
 320 static inline void return_to_32bit(struct kernel_vm86_regs * regs16, int retval)
 321 {
 322         struct pt_regs * regs32;
 323
 324         regs32 = save_v86_state(regs16);
 325         regs32->eax = retval;
 326         __asm__ __volatile__("movl %0,%%esp\n\t"
 327                 "movl %1,%%ebp\n\t"
 328                 "jmp resume_userspace"
 329                 : : "r" (regs32), "r" (current_thread_info()));
 330 }
 331
 332 static inline void set_IF(struct kernel_vm86_regs * regs)
 333 {
 334         VEFLAGS |= VIF_MASK;
 335         if (VEFLAGS & VIP_MASK)
 336                 return_to_32bit(regs, VM86_STI);
 337 }
 338
 339 static inline void clear_IF(struct kernel_vm86_regs * regs)
 340 {
 341         VEFLAGS &= ~VIF_MASK;
 342 }
 343
 344 static inline void clear_TF(struct kernel_vm86_regs * regs)
 345 {
 346         regs->eflags &= ~TF_MASK;
 347 }
 348
 349 static inline void clear_AC(struct kernel_vm86_regs * regs)
 350 {
 351         regs->eflags &= ~AC_MASK;
 352 }
 353
 354 /* It is correct to call set_IF(regs) from the set_vflags_*
 355  * functions. However someone forgot to call clear_IF(regs)
 356  * in the opposite case.
 357  * After the command sequence CLI PUSHF STI POPF you should
 358  * end up with interrups disabled, but you ended up with
 359  * interrupts enabled.
 360  *  ( I was testing my own changes, but the only bug I
 361  *    could find was in a function I had not changed. )
 362  * [KD]
 363  */
 364
 365 static inline void set_vflags_long(unsigned long eflags, struct kernel_vm86_regs * regs)
 366 {
 367         set_flags(VEFLAGS, eflags, current->thread.v86mask);
 368         set_flags(regs->eflags, eflags, SAFE_MASK);
 369         if (eflags & IF_MASK)
 370                 set_IF(regs);
 371         else
 372                 clear_IF(regs);
 373 }
 374
 375 static inline void set_vflags_short(unsigned short flags, struct kernel_vm86_regs * regs)
 376 {
 377         set_flags(VFLAGS, flags, current->thread.v86mask);
 378         set_flags(regs->eflags, flags, SAFE_MASK);
 379         if (flags & IF_MASK)
 380                 set_IF(regs);
 381         else
 382                 clear_IF(regs);
 383 }
 384
 385 static inline unsigned long get_vflags(struct kernel_vm86_regs * regs)
 386 {
 387         unsigned long flags = regs->eflags & RETURN_MASK;
 388
 389         if (VEFLAGS & VIF_MASK)
 390                 flags |= IF_MASK;
 391         flags |= IOPL_MASK;
 392         return flags | (VEFLAGS & current->thread.v86mask);
 393 }
 394
 395 static inline int is_revectored(int nr, struct revectored_struct * bitmap)
 396 {
 397         __asm__ __volatile__("btl %2,%1\n\tsbbl %0,%0"
 398                 :"=r" (nr)
 399                 :"m" (*bitmap),"r" (nr));
 400         return nr;
 401 }
 402
 403 #define val_byte(val, n) (((__u8 *)&val)[n])
 404
 405 #define pushb(base, ptr, val, err_label) \
 406         do { \
 407                 __u8 __val = val; \
 408                 ptr--; \
 409                 if (put_user(__val, base + ptr) < 0) \
 410                         goto err_label; \
 411         } while(0)
 412
 413 #define pushw(base, ptr, val, err_label) \
 414         do { \
 415                 __u16 __val = val; \
 416                 ptr--; \
 417                 if (put_user(val_byte(__val, 1), base + ptr) < 0) \
 418                         goto err_label; \
 419                 ptr--; \
 420                 if (put_user(val_byte(__val, 0), base + ptr) < 0) \
 421                         goto err_label; \
 422         } while(0)
 423
 424 #define pushl(base, ptr, val, err_label) \
 425         do { \
 426                 __u32 __val = val; \
 427                 ptr--; \
 428                 if (put_user(val_byte(__val, 3), base + ptr) < 0) \
 429                         goto err_label; \
 430                 ptr--; \
 431                 if (put_user(val_byte(__val, 2), base + ptr) < 0) \
 432                         goto err_label; \
 433                 ptr--; \
 434                 if (put_user(val_byte(__val, 1), base + ptr) < 0) \
 435                         goto err_label; \
 436                 ptr--; \
 437                 if (put_user(val_byte(__val, 0), base + ptr) < 0) \
 438                         goto err_label; \
 439         } while(0)
 440
 441 #define popb(base, ptr, err_label) \
 442         ({ \
 443                 __u8 __res; \
 444                 if (get_user(__res, base + ptr) < 0) \
 445                         goto err_label; \
 446                 ptr++; \
 447                 __res; \
 448         })
 449
 450 #define popw(base, ptr, err_label) \
 451         ({ \
 452                 __u16 __res; \
 453                 if (get_user(val_byte(__res, 0), base + ptr) < 0) \
 454                         goto err_label; \
 455                 ptr++; \
 456                 if (get_user(val_byte(__res, 1), base + ptr) < 0) \
 457                         goto err_label; \
 458                 ptr++; \
 459                 __res; \
 460         })
 461
 462 #define popl(base, ptr, err_label) \
 463         ({ \
 464                 __u32 __res; \
 465                 if (get_user(val_byte(__res, 0), base + ptr) < 0) \
 466                         goto err_label; \
 467                 ptr++; \
 468                 if (get_user(val_byte(__res, 1), base + ptr) < 0) \
 469                         goto err_label; \
 470                 ptr++; \
 471                 if (get_user(val_byte(__res, 2), base + ptr) < 0) \
 472                         goto err_label; \
 473                 ptr++; \
 474                 if (get_user(val_byte(__res, 3), base + ptr) < 0) \
 475                         goto err_label; \
 476                 ptr++; \
 477                 __res; \
 478         })
 479
 480 /* There are so many possible reasons for this function to return
 481  * VM86_INTx, so adding another doesn't bother me. We can expect
 482  * userspace programs to be able to handle it. (Getting a problem
 483  * in userspace is always better than an Oops anyway.) [KD]
 484  */
 485 static void do_int(struct kernel_vm86_regs *regs, int i,
 486     unsigned char __user * ssp, unsigned short sp)
 487 {
 488         unsigned long __user *intr_ptr;
 489         unsigned long segoffs;
 490
 491         if (regs->cs == BIOSSEG)
 492                 goto cannot_handle;
 493         if (is_revectored(i, &KVM86->int_revectored))
 494                 goto cannot_handle;
 495         if (i==0x21 && is_revectored(AH(regs),&KVM86->int21_revectored))
 496                 goto cannot_handle;
 497         intr_ptr = (unsigned long __user *) (i << 2);
 498         if (get_user(segoffs, intr_ptr))
 499                 goto cannot_handle;
 500         if ((segoffs >> 16) == BIOSSEG)
 501                 goto cannot_handle;
 502         pushw(ssp, sp, get_vflags(regs), cannot_handle);
 503         pushw(ssp, sp, regs->cs, cannot_handle);
 504         pushw(ssp, sp, IP(regs), cannot_handle);
 505         regs->cs = segoffs >> 16;
 506         SP(regs) -= 6;
 507         IP(regs) = segoffs & 0xffff;
 508         clear_TF(regs);
 509         clear_IF(regs);
 510         clear_AC(regs);
 511         return;
 512
 513 cannot_handle:
 514         return_to_32bit(regs, VM86_INTx + (i << 8));
 515 }
 516
 517 int handle_vm86_trap(struct kernel_vm86_regs * regs, long error_code, int trapno)
 518 {
 519         if (VMPI.is_vm86pus) {
 520                 if ( (trapno==3) || (trapno==1) )
 521                         return_to_32bit(regs, VM86_TRAP + (trapno << 8));
 522                 do_int(regs, trapno, (unsigned char __user *) (regs->ss << 4), SP(regs));
 523                 return 0;
 524         }
 525         if (trapno !=1)
 526                 return 1; /* we let this handle by the calling routine */
 527         if (current->ptrace & PT_PTRACED) {
 528                 unsigned long flags;
 529                 spin_lock_irqsave(&current->sighand->siglock, flags);
 530                 sigdelset(&current->blocked, SIGTRAP);
 531                 recalc_sigpending();
 532                 spin_unlock_irqrestore(&current->sighand->siglock, flags);
 533         }
 534         send_sig(SIGTRAP, current, 1);
 535         current->thread.trap_no = trapno;
 536         current->thread.error_code = error_code;
 537         return 0;
 538 }
 539
 540 void handle_vm86_fault(struct kernel_vm86_regs * regs, long error_code)
 541 {
 542         unsigned char opcode;
 543         unsigned char __user *csp;
 544         unsigned char __user *ssp;
 545         unsigned short ip, sp;
 546         int data32, pref_done;
 547
 548 #define CHECK_IF_IN_TRAP \
 549         if (VMPI.vm86dbg_active && VMPI.vm86dbg_TFpendig) \
 550                 newflags |= TF_MASK
 551 #define VM86_FAULT_RETURN do { \
 552         if (VMPI.force_return_for_pic  && (VEFLAGS & (IF_MASK | VIF_MASK))) \
 553                 return_to_32bit(regs, VM86_PICRETURN); \
 554         return; } while (0)
 555
 556         csp = (unsigned char __user *) (regs->cs << 4);
 557         ssp = (unsigned char __user *) (regs->ss << 4);
 558         sp = SP(regs);
 559         ip = IP(regs);
 560
 561         data32 = 0;
 562         pref_done = 0;
 563         do {
 564                 switch (opcode = popb(csp, ip, simulate_sigsegv)) {
 565                         case 0x66:      /* 32-bit data */     data32=1; break;
 566                         case 0x67:      /* 32-bit address */  break;
 567                         case 0x2e:      /* CS */              break;
 568                         case 0x3e:      /* DS */              break;
 569                         case 0x26:      /* ES */              break;
 570                         case 0x36:      /* SS */              break;
 571                         case 0x65:      /* GS */              break;
 572                         case 0x64:      /* FS */              break;
 573                         case 0xf2:      /* repnz */       break;
 574                         case 0xf3:      /* rep */             break;
 575                         default: pref_done = 1;
 576                 }
 577         } while (!pref_done);
 578
 579         switch (opcode) {
 580
 581         /* pushf */
 582         case 0x9c:
 583                 if (data32) {
 584                         pushl(ssp, sp, get_vflags(regs), simulate_sigsegv);
 585                         SP(regs) -= 4;
 586                 } else {
 587                         pushw(ssp, sp, get_vflags(regs), simulate_sigsegv);
 588                         SP(regs) -= 2;
 589                 }
 590                 IP(regs) = ip;
 591                 VM86_FAULT_RETURN;
 592
 593         /* popf */
 594         case 0x9d:
 595                 {
 596                 unsigned long newflags;
 597                 if (data32) {
 598                         newflags=popl(ssp, sp, simulate_sigsegv);
 599                         SP(regs) += 4;
 600                 } else {
 601                         newflags = popw(ssp, sp, simulate_sigsegv);
 602                         SP(regs) += 2;
 603                 }
 604                 IP(regs) = ip;
 605                 CHECK_IF_IN_TRAP;
 606                 if (data32) {
 607                         set_vflags_long(newflags, regs);
 608                 } else {
 609                         set_vflags_short(newflags, regs);
 610                 }
 611                 VM86_FAULT_RETURN;
 612                 }
 613
 614         /* int xx */
 615         case 0xcd: {
 616                 int intno=popb(csp, ip, simulate_sigsegv);
 617                 IP(regs) = ip;
 618                 if (VMPI.vm86dbg_active) {
 619                         if ( (1 << (intno &7)) & VMPI.vm86dbg_intxxtab[intno >> 3] )
 620                                 return_to_32bit(regs, VM86_INTx + (intno << 8));
 621                 }
 622                 do_int(regs, intno, ssp, sp);
 623                 return;
 624         }
 625
 626         /* iret */
 627         case 0xcf:
 628                 {
 629                 unsigned long newip;
 630                 unsigned long newcs;
 631                 unsigned long newflags;
 632                 if (data32) {
 633                         newip=popl(ssp, sp, simulate_sigsegv);
 634                         newcs=popl(ssp, sp, simulate_sigsegv);
 635                         newflags=popl(ssp, sp, simulate_sigsegv);
 636                         SP(regs) += 12;
 637                 } else {
 638                         newip = popw(ssp, sp, simulate_sigsegv);
 639                         newcs = popw(ssp, sp, simulate_sigsegv);
 640                         newflags = popw(ssp, sp, simulate_sigsegv);
 641                         SP(regs) += 6;
 642                 }
 643                 IP(regs) = newip;
 644                 regs->cs = newcs;
 645                 CHECK_IF_IN_TRAP;
 646                 if (data32) {
 647                         set_vflags_long(newflags, regs);
 648                 } else {
 649                         set_vflags_short(newflags, regs);
 650                 }
 651                 VM86_FAULT_RETURN;
 652                 }
 653
 654         /* cli */
 655         case 0xfa:
 656                 IP(regs) = ip;
 657                 clear_IF(regs);
 658                 VM86_FAULT_RETURN;
 659
 660         /* sti */
 661         /*
 662          * Damn. This is incorrect: the 'sti' instruction should actually
 663          * enable interrupts after the /next/ instruction. Not good.
 664          *
 665          * Probably needs some horsing around with the TF flag. Aiee..
 666          */
 667         case 0xfb:
 668                 IP(regs) = ip;
 669                 set_IF(regs);
 670                 VM86_FAULT_RETURN;
 671
 672         default:
 673                 return_to_32bit(regs, VM86_UNKNOWN);
 674         }
 675
 676         return;
 677
 678 simulate_sigsegv:
 679         /* FIXME: After a long discussion with Stas we finally
 680          *        agreed, that this is wrong. Here we should
 681          *        really send a SIGSEGV to the user program.
 682          *        But how do we create the correct context? We
 683          *        are inside a general protection fault handler
 684          *        and has just returned from a page fault handler.
 685          *        The correct context for the signal handler
 686          *        should be a mixture of the two, but how do we
 687          *        get the information? [KD]
 688          */
 689         return_to_32bit(regs, VM86_UNKNOWN);
 690 }
 691
 692 /* ---------------- vm86 special IRQ passing stuff ----------------- */
 693
 694 #define VM86_IRQNAME            "vm86irq"
 695
 696 static struct vm86_irqs {
 697         struct task_struct *tsk;
 698         int sig;
 699 } vm86_irqs[16];
 700
 701 static DEFINE_SPINLOCK(irqbits_lock);
 702 static int irqbits;
 703
 704 #define ALLOWED_SIGS ( 1 /* 0 = don't send a signal */ \
 705         | (1 << SIGUSR1) | (1 << SIGUSR2) | (1 << SIGIO)  | (1 << SIGURG) \
 706         | (1 << SIGUNUSED) )
 707
 708 static irqreturn_t irq_handler(int intno, void *dev_id, struct pt_regs * regs)
 709 {
 710         int irq_bit;
 711         unsigned long flags;
 712
 713         spin_lock_irqsave(&irqbits_lock, flags);
 714         irq_bit = 1 << intno;
 715         if ((irqbits & irq_bit) || ! vm86_irqs[intno].tsk)
 716                 goto out;
 717         irqbits |= irq_bit;
 718         if (vm86_irqs[intno].sig)
 719                 send_sig(vm86_irqs[intno].sig, vm86_irqs[intno].tsk, 1);
 720         spin_unlock_irqrestore(&irqbits_lock, flags);
 721         /*
 722          * IRQ will be re-enabled when user asks for the irq (whether
 723          * polling or as a result of the signal)
 724          */
 725         disable_irq(intno);
 726         return IRQ_HANDLED;
 727
 728 out:
 729         spin_unlock_irqrestore(&irqbits_lock, flags);
 730         return IRQ_NONE;
 731 }
 732
 733 static inline void free_vm86_irq(int irqnumber)
 734 {
 735         unsigned long flags;
 736
 737         free_irq(irqnumber, NULL);
 738         vm86_irqs[irqnumber].tsk = NULL;
 739
 740         spin_lock_irqsave(&irqbits_lock, flags);
 741         irqbits &= ~(1 << irqnumber);
 742         spin_unlock_irqrestore(&irqbits_lock, flags);
 743 }
 744
 745 void release_vm86_irqs(struct task_struct *task)
 746 {
 747         int i;
 748         for (i = FIRST_VM86_IRQ ; i <= LAST_VM86_IRQ; i++)
 749             if (vm86_irqs[i].tsk == task)
 750                 free_vm86_irq(i);
 751 }
 752
 753 static inline int get_and_reset_irq(int irqnumber)
 754 {
 755         int bit;
 756         unsigned long flags;
 757
 758         if (invalid_vm86_irq(irqnumber)) return 0;
 759         if (vm86_irqs[irqnumber].tsk != current) return 0;
 760         spin_lock_irqsave(&irqbits_lock, flags);
 761         bit = irqbits & (1 << irqnumber);
 762         irqbits &= ~bit;
 763         spin_unlock_irqrestore(&irqbits_lock, flags);
 764         if (!bit)
 765                 return 0;
 766         enable_irq(irqnumber);
 767         return 1;
 768 }
 769
 770
 771 static int do_vm86_irq_handling(int subfunction, int irqnumber)
 772 {
 773         int ret;
 774         switch (subfunction) {
 775                 case VM86_GET_AND_RESET_IRQ: {
 776                         return get_and_reset_irq(irqnumber);
 777                 }
 778                 case VM86_GET_IRQ_BITS: {
 779                         return irqbits;
 780                 }
 781                 case VM86_REQUEST_IRQ: {
 782                         int sig = irqnumber >> 8;
 783                         int irq = irqnumber & 255;
 784                         if (!capable(CAP_SYS_ADMIN)) return -EPERM;
 785                         if (!((1 << sig) & ALLOWED_SIGS)) return -EPERM;
 786                         if (invalid_vm86_irq(irq)) return -EPERM;
 787                         if (vm86_irqs[irq].tsk) return -EPERM;
 788                         ret = request_irq(irq, &irq_handler, 0, VM86_IRQNAME, NULL);
 789                         if (ret) return ret;
 790                         vm86_irqs[irq].sig = sig;
 791                         vm86_irqs[irq].tsk = current;
 792                         return irq;
 793                 }
 794                 case  VM86_FREE_IRQ: {
 795                         if (invalid_vm86_irq(irqnumber)) return -EPERM;
 796                         if (!vm86_irqs[irqnumber].tsk) return 0;
 797                         if (vm86_irqs[irqnumber].tsk != current) return -EPERM;
 798                         free_vm86_irq(irqnumber);
 799                         return 0;
 800                 }
 801         }
 802         return -EINVAL;
 803 }
 804