arch/x86/kernel/vm86_32.c

   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 #include <linux/capability.h>
  32 #include <linux/errno.h>
  33 #include <linux/interrupt.h>
  34 #include <linux/sched.h>
  35 #include <linux/kernel.h>
  36 #include <linux/signal.h>
  37 #include <linux/string.h>
  38 #include <linux/mm.h>
  39 #include <linux/smp.h>
  40 #include <linux/highmem.h>
  41 #include <linux/ptrace.h>
  42 #include <linux/audit.h>
  43 #include <linux/stddef.h>
  44
  45 #include <asm/uaccess.h>
  46 #include <asm/io.h>
  47 #include <asm/tlbflush.h>
  48 #include <asm/irq.h>
  49
  50 /*
  51  * Known problems:
  52  *
  53  * Interrupt handling is not guaranteed:
  54  * - a real x86 will disable all interrupts for one instruction
  55  *   after a "mov ss,xx" to make stack handling atomic even without
  56  *   the 'lss' instruction. We can't guarantee this in v86 mode,
  57  *   as the next instruction might result in a page fault or similar.
  58  * - a real x86 will have interrupts disabled for one instruction
  59  *   past the 'sti' that enables them. We don't bother with all the
  60  *   details yet.
  61  *
  62  * Let's hope these problems do not actually matter for anything.
  63  */
  64
  65
  66 #define KVM86   ((struct kernel_vm86_struct *)regs)
  67 #define VMPI    KVM86->vm86plus
  68
  69
  70 /*
  71  * 8- and 16-bit register defines..
  72  */
  73 #define AL(regs)        (((unsigned char *)&((regs)->pt.ax))[0])
  74 #define AH(regs)        (((unsigned char *)&((regs)->pt.ax))[1])
  75 #define IP(regs)        (*(unsigned short *)&((regs)->pt.ip))
  76 #define SP(regs)        (*(unsigned short *)&((regs)->pt.sp))
  77
  78 /*
  79  * virtual flags (16 and 32-bit versions)
  80  */
  81 #define VFLAGS  (*(unsigned short *)&(current->thread.v86flags))
  82 #define VEFLAGS (current->thread.v86flags)
  83
  84 #define set_flags(X,new,mask) \
  85 ((X) = ((X) & ~(mask)) | ((new) & (mask)))
  86
  87 #define SAFE_MASK       (0xDD5)
  88 #define RETURN_MASK     (0xDFF)
  89
  90 /* convert kernel_vm86_regs to vm86_regs */
  91 static int copy_vm86_regs_to_user(struct vm86_regs __user *user,
  92                                   const struct kernel_vm86_regs *regs)
  93 {
  94         int ret = 0;
  95
  96         /* kernel_vm86_regs is missing gs, so copy everything up to
  97            (but not including) orig_eax, and then rest including orig_eax. */
  98         ret += copy_to_user(user, regs, offsetof(struct kernel_vm86_regs, pt.orig_ax));
  99         ret += copy_to_user(&user->orig_eax, &regs->pt.orig_ax,
 100                             sizeof(struct kernel_vm86_regs) -
 101                             offsetof(struct kernel_vm86_regs, pt.orig_ax));
 102
 103         return ret;
 104 }
 105
 106 /* convert vm86_regs to kernel_vm86_regs */
 107 static int copy_vm86_regs_from_user(struct kernel_vm86_regs *regs,
 108                                     const struct vm86_regs __user *user,
 109                                     unsigned extra)
 110 {
 111         int ret = 0;
 112
 113         /* copy ax-fs inclusive */
 114         ret += copy_from_user(regs, user, offsetof(struct kernel_vm86_regs, pt.orig_ax));
 115         /* copy orig_ax-__gsh+extra */
 116         ret += copy_from_user(&regs->pt.orig_ax, &user->orig_eax,
 117                               sizeof(struct kernel_vm86_regs) -
 118                               offsetof(struct kernel_vm86_regs, pt.orig_ax) +
 119                               extra);
 120         return ret;
 121 }
 122
 123 struct pt_regs * save_v86_state(struct kernel_vm86_regs * regs)
 124 {
 125         struct tss_struct *tss;
 126         struct pt_regs *ret;
 127         unsigned long tmp;
 128
 129         /*
 130          * This gets called from entry.S with interrupts disabled, but
 131          * from process context. Enable interrupts here, before trying
 132          * to access user space.
 133          */
 134         local_irq_enable();
 135
 136         if (!current->thread.vm86_info) {
 137                 printk("no vm86_info: BAD\n");
 138                 do_exit(SIGSEGV);
 139         }
 140         set_flags(regs->pt.flags, VEFLAGS, VIF_MASK | current->thread.v86mask);
 141         tmp = copy_vm86_regs_to_user(&current->thread.vm86_info->regs,regs);
 142         tmp += put_user(current->thread.screen_bitmap,&current->thread.vm86_info->screen_bitmap);
 143         if (tmp) {
 144                 printk("vm86: could not access userspace vm86_info\n");
 145                 do_exit(SIGSEGV);
 146         }
 147
 148         tss = &per_cpu(init_tss, get_cpu());
 149         current->thread.sp0 = current->thread.saved_sp0;
 150         current->thread.sysenter_cs = __KERNEL_CS;
 151         load_sp0(tss, &current->thread);
 152         current->thread.saved_sp0 = 0;
 153         put_cpu();
 154
 155         ret = KVM86->regs32;
 156
 157         ret->fs = current->thread.saved_fs;
 158         loadsegment(gs, current->thread.saved_gs);
 159
 160         return ret;
 161 }
 162
 163 static void mark_screen_rdonly(struct mm_struct *mm)
 164 {
 165         pgd_t *pgd;
 166         pud_t *pud;
 167         pmd_t *pmd;
 168         pte_t *pte;
 169         spinlock_t *ptl;
 170         int i;
 171
 172         pgd = pgd_offset(mm, 0xA0000);
 173         if (pgd_none_or_clear_bad(pgd))
 174                 goto out;
 175         pud = pud_offset(pgd, 0xA0000);
 176         if (pud_none_or_clear_bad(pud))
 177                 goto out;
 178         pmd = pmd_offset(pud, 0xA0000);
 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         flush_tlb();
 190 }
 191
 192
 193
 194 static int do_vm86_irq_handling(int subfunction, int irqnumber);
 195 static void do_sys_vm86(struct kernel_vm86_struct *info, struct task_struct *tsk);
 196
 197 asmlinkage int sys_vm86old(struct pt_regs regs)
 198 {
 199         struct vm86_struct __user *v86 = (struct vm86_struct __user *)regs.bx;
 200         struct kernel_vm86_struct info; /* declare this _on top_,
 201                                          * this avoids wasting of stack space.
 202                                          * This remains on the stack until we
 203                                          * return to 32 bit user space.
 204                                          */
 205         struct task_struct *tsk;
 206         int tmp, ret = -EPERM;
 207
 208         tsk = current;
 209         if (tsk->thread.saved_sp0)
 210                 goto out;
 211         tmp = copy_vm86_regs_from_user(&info.regs, &v86->regs,
 212                                        offsetof(struct kernel_vm86_struct, vm86plus) -
 213                                        sizeof(info.regs));
 214         ret = -EFAULT;
 215         if (tmp)
 216                 goto out;
 217         memset(&info.vm86plus, 0, (int)&info.regs32 - (int)&info.vm86plus);
 218         info.regs32 = &regs;
 219         tsk->thread.vm86_info = v86;
 220         do_sys_vm86(&info, tsk);
 221         ret = 0;        /* we never return here */
 222 out:
 223         return ret;
 224 }
 225
 226
 227 asmlinkage int sys_vm86(struct pt_regs regs)
 228 {
 229         struct kernel_vm86_struct info; /* declare this _on top_,
 230                                          * this avoids wasting of stack space.
 231                                          * This remains on the stack until we
 232                                          * return to 32 bit user space.
 233                                          */
 234         struct task_struct *tsk;
 235         int tmp, ret;
 236         struct vm86plus_struct __user *v86;
 237
 238         tsk = current;
 239         switch (regs.bx) {
 240                 case VM86_REQUEST_IRQ:
 241                 case VM86_FREE_IRQ:
 242                 case VM86_GET_IRQ_BITS:
 243                 case VM86_GET_AND_RESET_IRQ:
 244                         ret = do_vm86_irq_handling(regs.bx, (int)regs.cx);
 245                         goto out;
 246                 case VM86_PLUS_INSTALL_CHECK:
 247                         /* NOTE: on old vm86 stuff this will return the error
 248                            from access_ok(), because the subfunction is
 249                            interpreted as (invalid) address to vm86_struct.
 250                            So the installation check works.
 251                          */
 252                         ret = 0;
 253                         goto out;
 254         }
 255
 256         /* we come here only for functions VM86_ENTER, VM86_ENTER_NO_BYPASS */
 257         ret = -EPERM;
 258         if (tsk->thread.saved_sp0)
 259                 goto out;
 260         v86 = (struct vm86plus_struct __user *)regs.cx;
 261         tmp = copy_vm86_regs_from_user(&info.regs, &v86->regs,
 262                                        offsetof(struct kernel_vm86_struct, regs32) -
 263                                        sizeof(info.regs));
 264         ret = -EFAULT;
 265         if (tmp)
 266                 goto out;
 267         info.regs32 = &regs;
 268         info.vm86plus.is_vm86pus = 1;
 269         tsk->thread.vm86_info = (struct vm86_struct __user *)v86;
 270         do_sys_vm86(&info, tsk);
 271         ret = 0;        /* we never return here */
 272 out:
 273         return ret;
 274 }
 275
 276
 277 static void do_sys_vm86(struct kernel_vm86_struct *info, struct task_struct *tsk)
 278 {
 279         struct tss_struct *tss;
 280 /*
 281  * make sure the vm86() system call doesn't try to do anything silly
 282  */
 283         info->regs.pt.ds = 0;
 284         info->regs.pt.es = 0;
 285         info->regs.pt.fs = 0;
 286
 287 /* we are clearing gs later just before "jmp resume_userspace",
 288  * because it is not saved/restored.
 289  */
 290
 291 /*
 292  * The flags register is also special: we cannot trust that the user
 293  * has set it up safely, so this makes sure interrupt etc flags are
 294  * inherited from protected mode.
 295  */
 296         VEFLAGS = info->regs.pt.flags;
 297         info->regs.pt.flags &= SAFE_MASK;
 298         info->regs.pt.flags |= info->regs32->flags & ~SAFE_MASK;
 299         info->regs.pt.flags |= VM_MASK;
 300
 301         switch (info->cpu_type) {
 302                 case CPU_286:
 303                         tsk->thread.v86mask = 0;
 304                         break;
 305                 case CPU_386:
 306                         tsk->thread.v86mask = NT_MASK | IOPL_MASK;
 307                         break;
 308                 case CPU_486:
 309                         tsk->thread.v86mask = AC_MASK | NT_MASK | IOPL_MASK;
 310                         break;
 311                 default:
 312                         tsk->thread.v86mask = ID_MASK | AC_MASK | NT_MASK | IOPL_MASK;
 313                         break;
 314         }
 315
 316 /*
 317  * Save old state, set default return value (%ax) to 0
 318  */
 319         info->regs32->ax = 0;
 320         tsk->thread.saved_sp0 = tsk->thread.sp0;
 321         tsk->thread.saved_fs = info->regs32->fs;
 322         savesegment(gs, tsk->thread.saved_gs);
 323
 324         tss = &per_cpu(init_tss, get_cpu());
 325         tsk->thread.sp0 = (unsigned long) &info->VM86_TSS_ESP0;
 326         if (cpu_has_sep)
 327                 tsk->thread.sysenter_cs = 0;
 328         load_sp0(tss, &tsk->thread);
 329         put_cpu();
 330
 331         tsk->thread.screen_bitmap = info->screen_bitmap;
 332         if (info->flags & VM86_SCREEN_BITMAP)
 333                 mark_screen_rdonly(tsk->mm);
 334
 335         /*call audit_syscall_exit since we do not exit via the normal paths */
 336         if (unlikely(current->audit_context))
 337                 audit_syscall_exit(AUDITSC_RESULT(0), 0);
 338
 339         __asm__ __volatile__(
 340                 "movl %0,%%esp\n\t"
 341                 "movl %1,%%ebp\n\t"
 342                 "mov  %2, %%gs\n\t"
 343                 "jmp resume_userspace"
 344                 : /* no outputs */
 345                 :"r" (&info->regs), "r" (task_thread_info(tsk)), "r" (0));
 346         /* we never return here */
 347 }
 348
 349 static inline void return_to_32bit(struct kernel_vm86_regs * regs16, int retval)
 350 {
 351         struct pt_regs * regs32;
 352
 353         regs32 = save_v86_state(regs16);
 354         regs32->ax = retval;
 355         __asm__ __volatile__("movl %0,%%esp\n\t"
 356                 "movl %1,%%ebp\n\t"
 357                 "jmp resume_userspace"
 358                 : : "r" (regs32), "r" (current_thread_info()));
 359 }
 360
 361 static inline void set_IF(struct kernel_vm86_regs * regs)
 362 {
 363         VEFLAGS |= VIF_MASK;
 364         if (VEFLAGS & VIP_MASK)
 365                 return_to_32bit(regs, VM86_STI);
 366 }
 367
 368 static inline void clear_IF(struct kernel_vm86_regs * regs)
 369 {
 370         VEFLAGS &= ~VIF_MASK;
 371 }
 372
 373 static inline void clear_TF(struct kernel_vm86_regs * regs)
 374 {
 375         regs->pt.flags &= ~TF_MASK;
 376 }
 377
 378 static inline void clear_AC(struct kernel_vm86_regs * regs)
 379 {
 380         regs->pt.flags &= ~AC_MASK;
 381 }
 382
 383 /* It is correct to call set_IF(regs) from the set_vflags_*
 384  * functions. However someone forgot to call clear_IF(regs)
 385  * in the opposite case.
 386  * After the command sequence CLI PUSHF STI POPF you should
 387  * end up with interrupts disabled, but you ended up with
 388  * interrupts enabled.
 389  *  ( I was testing my own changes, but the only bug I
 390  *    could find was in a function I had not changed. )
 391  * [KD]
 392  */
 393
 394 static inline void set_vflags_long(unsigned long flags, struct kernel_vm86_regs * regs)
 395 {
 396         set_flags(VEFLAGS, flags, current->thread.v86mask);
 397         set_flags(regs->pt.flags, flags, SAFE_MASK);
 398         if (flags & IF_MASK)
 399                 set_IF(regs);
 400         else
 401                 clear_IF(regs);
 402 }
 403
 404 static inline void set_vflags_short(unsigned short flags, struct kernel_vm86_regs * regs)
 405 {
 406         set_flags(VFLAGS, flags, current->thread.v86mask);
 407         set_flags(regs->pt.flags, flags, SAFE_MASK);
 408         if (flags & IF_MASK)
 409                 set_IF(regs);
 410         else
 411                 clear_IF(regs);
 412 }
 413
 414 static inline unsigned long get_vflags(struct kernel_vm86_regs * regs)
 415 {
 416         unsigned long flags = regs->pt.flags & RETURN_MASK;
 417
 418         if (VEFLAGS & VIF_MASK)
 419                 flags |= IF_MASK;
 420         flags |= IOPL_MASK;
 421         return flags | (VEFLAGS & current->thread.v86mask);
 422 }
 423
 424 static inline int is_revectored(int nr, struct revectored_struct * bitmap)
 425 {
 426         __asm__ __volatile__("btl %2,%1\n\tsbbl %0,%0"
 427                 :"=r" (nr)
 428                 :"m" (*bitmap),"r" (nr));
 429         return nr;
 430 }
 431
 432 #define val_byte(val, n) (((__u8 *)&val)[n])
 433
 434 #define pushb(base, ptr, val, err_label) \
 435         do { \
 436                 __u8 __val = val; \
 437                 ptr--; \
 438                 if (put_user(__val, base + ptr) < 0) \
 439                         goto err_label; \
 440         } while(0)
 441
 442 #define pushw(base, ptr, val, err_label) \
 443         do { \
 444                 __u16 __val = val; \
 445                 ptr--; \
 446                 if (put_user(val_byte(__val, 1), base + ptr) < 0) \
 447                         goto err_label; \
 448                 ptr--; \
 449                 if (put_user(val_byte(__val, 0), base + ptr) < 0) \
 450                         goto err_label; \
 451         } while(0)
 452
 453 #define pushl(base, ptr, val, err_label) \
 454         do { \
 455                 __u32 __val = val; \
 456                 ptr--; \
 457                 if (put_user(val_byte(__val, 3), base + ptr) < 0) \
 458                         goto err_label; \
 459                 ptr--; \
 460                 if (put_user(val_byte(__val, 2), base + ptr) < 0) \
 461                         goto err_label; \
 462                 ptr--; \
 463                 if (put_user(val_byte(__val, 1), base + ptr) < 0) \
 464                         goto err_label; \
 465                 ptr--; \
 466                 if (put_user(val_byte(__val, 0), base + ptr) < 0) \
 467                         goto err_label; \
 468         } while(0)
 469
 470 #define popb(base, ptr, err_label) \
 471         ({ \
 472                 __u8 __res; \
 473                 if (get_user(__res, base + ptr) < 0) \
 474                         goto err_label; \
 475                 ptr++; \
 476                 __res; \
 477         })
 478
 479 #define popw(base, ptr, err_label) \
 480         ({ \
 481                 __u16 __res; \
 482                 if (get_user(val_byte(__res, 0), base + ptr) < 0) \
 483                         goto err_label; \
 484                 ptr++; \
 485                 if (get_user(val_byte(__res, 1), base + ptr) < 0) \
 486                         goto err_label; \
 487                 ptr++; \
 488                 __res; \
 489         })
 490
 491 #define popl(base, ptr, err_label) \
 492         ({ \
 493                 __u32 __res; \
 494                 if (get_user(val_byte(__res, 0), base + ptr) < 0) \
 495                         goto err_label; \
 496                 ptr++; \
 497                 if (get_user(val_byte(__res, 1), base + ptr) < 0) \
 498                         goto err_label; \
 499                 ptr++; \
 500                 if (get_user(val_byte(__res, 2), base + ptr) < 0) \
 501                         goto err_label; \
 502                 ptr++; \
 503                 if (get_user(val_byte(__res, 3), base + ptr) < 0) \
 504                         goto err_label; \
 505                 ptr++; \
 506                 __res; \
 507         })
 508
 509 /* There are so many possible reasons for this function to return
 510  * VM86_INTx, so adding another doesn't bother me. We can expect
 511  * userspace programs to be able to handle it. (Getting a problem
 512  * in userspace is always better than an Oops anyway.) [KD]
 513  */
 514 static void do_int(struct kernel_vm86_regs *regs, int i,
 515     unsigned char __user * ssp, unsigned short sp)
 516 {
 517         unsigned long __user *intr_ptr;
 518         unsigned long segoffs;
 519
 520         if (regs->pt.cs == BIOSSEG)
 521                 goto cannot_handle;
 522         if (is_revectored(i, &KVM86->int_revectored))
 523                 goto cannot_handle;
 524         if (i==0x21 && is_revectored(AH(regs),&KVM86->int21_revectored))
 525                 goto cannot_handle;
 526         intr_ptr = (unsigned long __user *) (i << 2);
 527         if (get_user(segoffs, intr_ptr))
 528                 goto cannot_handle;
 529         if ((segoffs >> 16) == BIOSSEG)
 530                 goto cannot_handle;
 531         pushw(ssp, sp, get_vflags(regs), cannot_handle);
 532         pushw(ssp, sp, regs->pt.cs, cannot_handle);
 533         pushw(ssp, sp, IP(regs), cannot_handle);
 534         regs->pt.cs = segoffs >> 16;
 535         SP(regs) -= 6;
 536         IP(regs) = segoffs & 0xffff;
 537         clear_TF(regs);
 538         clear_IF(regs);
 539         clear_AC(regs);
 540         return;
 541
 542 cannot_handle:
 543         return_to_32bit(regs, VM86_INTx + (i << 8));
 544 }
 545
 546 int handle_vm86_trap(struct kernel_vm86_regs * regs, long error_code, int trapno)
 547 {
 548         if (VMPI.is_vm86pus) {
 549                 if ( (trapno==3) || (trapno==1) )
 550                         return_to_32bit(regs, VM86_TRAP + (trapno << 8));
 551                 do_int(regs, trapno, (unsigned char __user *) (regs->pt.ss << 4), SP(regs));
 552                 return 0;
 553         }
 554         if (trapno !=1)
 555                 return 1; /* we let this handle by the calling routine */
 556         if (current->ptrace & PT_PTRACED) {
 557                 unsigned long flags;
 558                 spin_lock_irqsave(&current->sighand->siglock, flags);
 559                 sigdelset(&current->blocked, SIGTRAP);
 560                 recalc_sigpending();
 561                 spin_unlock_irqrestore(&current->sighand->siglock, flags);
 562         }
 563         send_sig(SIGTRAP, current, 1);
 564         current->thread.trap_no = trapno;
 565         current->thread.error_code = error_code;
 566         return 0;
 567 }
 568
 569 void handle_vm86_fault(struct kernel_vm86_regs * regs, long error_code)
 570 {
 571         unsigned char opcode;
 572         unsigned char __user *csp;
 573         unsigned char __user *ssp;
 574         unsigned short ip, sp, orig_flags;
 575         int data32, pref_done;
 576
 577 #define CHECK_IF_IN_TRAP \
 578         if (VMPI.vm86dbg_active && VMPI.vm86dbg_TFpendig) \
 579                 newflags |= TF_MASK
 580 #define VM86_FAULT_RETURN do { \
 581         if (VMPI.force_return_for_pic  && (VEFLAGS & (IF_MASK | VIF_MASK))) \
 582                 return_to_32bit(regs, VM86_PICRETURN); \
 583         if (orig_flags & TF_MASK) \
 584                 handle_vm86_trap(regs, 0, 1); \
 585         return; } while (0)
 586
 587         orig_flags = *(unsigned short *)&regs->pt.flags;
 588
 589         csp = (unsigned char __user *) (regs->pt.cs << 4);
 590         ssp = (unsigned char __user *) (regs->pt.ss << 4);
 591         sp = SP(regs);
 592         ip = IP(regs);
 593
 594         data32 = 0;
 595         pref_done = 0;
 596         do {
 597                 switch (opcode = popb(csp, ip, simulate_sigsegv)) {
 598                         case 0x66:      /* 32-bit data */     data32=1; break;
 599                         case 0x67:      /* 32-bit address */  break;
 600                         case 0x2e:      /* CS */              break;
 601                         case 0x3e:      /* DS */              break;
 602                         case 0x26:      /* ES */              break;
 603                         case 0x36:      /* SS */              break;
 604                         case 0x65:      /* GS */              break;
 605                         case 0x64:      /* FS */              break;
 606                         case 0xf2:      /* repnz */       break;
 607                         case 0xf3:      /* rep */             break;
 608                         default: pref_done = 1;
 609                 }
 610         } while (!pref_done);
 611
 612         switch (opcode) {
 613
 614         /* pushf */
 615         case 0x9c:
 616                 if (data32) {
 617                         pushl(ssp, sp, get_vflags(regs), simulate_sigsegv);
 618                         SP(regs) -= 4;
 619                 } else {
 620                         pushw(ssp, sp, get_vflags(regs), simulate_sigsegv);
 621                         SP(regs) -= 2;
 622                 }
 623                 IP(regs) = ip;
 624                 VM86_FAULT_RETURN;
 625
 626         /* popf */
 627         case 0x9d:
 628                 {
 629                 unsigned long newflags;
 630                 if (data32) {
 631                         newflags=popl(ssp, sp, simulate_sigsegv);
 632                         SP(regs) += 4;
 633                 } else {
 634                         newflags = popw(ssp, sp, simulate_sigsegv);
 635                         SP(regs) += 2;
 636                 }
 637                 IP(regs) = ip;
 638                 CHECK_IF_IN_TRAP;
 639                 if (data32) {
 640                         set_vflags_long(newflags, regs);
 641                 } else {
 642                         set_vflags_short(newflags, regs);
 643                 }
 644                 VM86_FAULT_RETURN;
 645                 }
 646
 647         /* int xx */
 648         case 0xcd: {
 649                 int intno=popb(csp, ip, simulate_sigsegv);
 650                 IP(regs) = ip;
 651                 if (VMPI.vm86dbg_active) {
 652                         if ( (1 << (intno &7)) & VMPI.vm86dbg_intxxtab[intno >> 3] )
 653                                 return_to_32bit(regs, VM86_INTx + (intno << 8));
 654                 }
 655                 do_int(regs, intno, ssp, sp);
 656                 return;
 657         }
 658
 659         /* iret */
 660         case 0xcf:
 661                 {
 662                 unsigned long newip;
 663                 unsigned long newcs;
 664                 unsigned long newflags;
 665                 if (data32) {
 666                         newip=popl(ssp, sp, simulate_sigsegv);
 667                         newcs=popl(ssp, sp, simulate_sigsegv);
 668                         newflags=popl(ssp, sp, simulate_sigsegv);
 669                         SP(regs) += 12;
 670                 } else {
 671                         newip = popw(ssp, sp, simulate_sigsegv);
 672                         newcs = popw(ssp, sp, simulate_sigsegv);
 673                         newflags = popw(ssp, sp, simulate_sigsegv);
 674                         SP(regs) += 6;
 675                 }
 676                 IP(regs) = newip;
 677                 regs->pt.cs = newcs;
 678                 CHECK_IF_IN_TRAP;
 679                 if (data32) {
 680                         set_vflags_long(newflags, regs);
 681                 } else {
 682                         set_vflags_short(newflags, regs);
 683                 }
 684                 VM86_FAULT_RETURN;
 685                 }
 686
 687         /* cli */
 688         case 0xfa:
 689                 IP(regs) = ip;
 690                 clear_IF(regs);
 691                 VM86_FAULT_RETURN;
 692
 693         /* sti */
 694         /*
 695          * Damn. This is incorrect: the 'sti' instruction should actually
 696          * enable interrupts after the /next/ instruction. Not good.
 697          *
 698          * Probably needs some horsing around with the TF flag. Aiee..
 699          */
 700         case 0xfb:
 701                 IP(regs) = ip;
 702                 set_IF(regs);
 703                 VM86_FAULT_RETURN;
 704
 705         default:
 706                 return_to_32bit(regs, VM86_UNKNOWN);
 707         }
 708
 709         return;
 710
 711 simulate_sigsegv:
 712         /* FIXME: After a long discussion with Stas we finally
 713          *        agreed, that this is wrong. Here we should
 714          *        really send a SIGSEGV to the user program.
 715          *        But how do we create the correct context? We
 716          *        are inside a general protection fault handler
 717          *        and has just returned from a page fault handler.
 718          *        The correct context for the signal handler
 719          *        should be a mixture of the two, but how do we
 720          *        get the information? [KD]
 721          */
 722         return_to_32bit(regs, VM86_UNKNOWN);
 723 }
 724
 725 /* ---------------- vm86 special IRQ passing stuff ----------------- */
 726
 727 #define VM86_IRQNAME            "vm86irq"
 728
 729 static struct vm86_irqs {
 730         struct task_struct *tsk;
 731         int sig;
 732 } vm86_irqs[16];
 733
 734 static DEFINE_SPINLOCK(irqbits_lock);
 735 static int irqbits;
 736
 737 #define ALLOWED_SIGS ( 1 /* 0 = don't send a signal */ \
 738         | (1 << SIGUSR1) | (1 << SIGUSR2) | (1 << SIGIO)  | (1 << SIGURG) \
 739         | (1 << SIGUNUSED) )
 740
 741 static irqreturn_t irq_handler(int intno, void *dev_id)
 742 {
 743         int irq_bit;
 744         unsigned long flags;
 745
 746         spin_lock_irqsave(&irqbits_lock, flags);
 747         irq_bit = 1 << intno;
 748         if ((irqbits & irq_bit) || ! vm86_irqs[intno].tsk)
 749                 goto out;
 750         irqbits |= irq_bit;
 751         if (vm86_irqs[intno].sig)
 752                 send_sig(vm86_irqs[intno].sig, vm86_irqs[intno].tsk, 1);
 753         /*
 754          * IRQ will be re-enabled when user asks for the irq (whether
 755          * polling or as a result of the signal)
 756          */
 757         disable_irq_nosync(intno);
 758         spin_unlock_irqrestore(&irqbits_lock, flags);
 759         return IRQ_HANDLED;
 760
 761 out:
 762         spin_unlock_irqrestore(&irqbits_lock, flags);
 763         return IRQ_NONE;
 764 }
 765
 766 static inline void free_vm86_irq(int irqnumber)
 767 {
 768         unsigned long flags;
 769
 770         free_irq(irqnumber, NULL);
 771         vm86_irqs[irqnumber].tsk = NULL;
 772
 773         spin_lock_irqsave(&irqbits_lock, flags);
 774         irqbits &= ~(1 << irqnumber);
 775         spin_unlock_irqrestore(&irqbits_lock, flags);
 776 }
 777
 778 void release_vm86_irqs(struct task_struct *task)
 779 {
 780         int i;
 781         for (i = FIRST_VM86_IRQ ; i <= LAST_VM86_IRQ; i++)
 782             if (vm86_irqs[i].tsk == task)
 783                 free_vm86_irq(i);
 784 }
 785
 786 static inline int get_and_reset_irq(int irqnumber)
 787 {
 788         int bit;
 789         unsigned long flags;
 790         int ret = 0;
 791
 792         if (invalid_vm86_irq(irqnumber)) return 0;
 793         if (vm86_irqs[irqnumber].tsk != current) return 0;
 794         spin_lock_irqsave(&irqbits_lock, flags);
 795         bit = irqbits & (1 << irqnumber);
 796         irqbits &= ~bit;
 797         if (bit) {
 798                 enable_irq(irqnumber);
 799                 ret = 1;
 800         }
 801
 802         spin_unlock_irqrestore(&irqbits_lock, flags);
 803         return ret;
 804 }
 805
 806
 807 static int do_vm86_irq_handling(int subfunction, int irqnumber)
 808 {
 809         int ret;
 810         switch (subfunction) {
 811                 case VM86_GET_AND_RESET_IRQ: {
 812                         return get_and_reset_irq(irqnumber);
 813                 }
 814                 case VM86_GET_IRQ_BITS: {
 815                         return irqbits;
 816                 }
 817                 case VM86_REQUEST_IRQ: {
 818                         int sig = irqnumber >> 8;
 819                         int irq = irqnumber & 255;
 820                         if (!capable(CAP_SYS_ADMIN)) return -EPERM;
 821                         if (!((1 << sig) & ALLOWED_SIGS)) return -EPERM;
 822                         if (invalid_vm86_irq(irq)) return -EPERM;
 823                         if (vm86_irqs[irq].tsk) return -EPERM;
 824                         ret = request_irq(irq, &irq_handler, 0, VM86_IRQNAME, NULL);
 825                         if (ret) return ret;
 826                         vm86_irqs[irq].sig = sig;
 827                         vm86_irqs[irq].tsk = current;
 828                         return irq;
 829                 }
 830                 case  VM86_FREE_IRQ: {
 831                         if (invalid_vm86_irq(irqnumber)) return -EPERM;
 832                         if (!vm86_irqs[irqnumber].tsk) return 0;
 833                         if (vm86_irqs[irqnumber].tsk != current) return -EPERM;
 834                         free_vm86_irq(irqnumber);
 835                         return 0;
 836                 }
 837         }
 838         return -EINVAL;
 839 }
 840