1 /* 2 * Copyright (C) 1994 Linus Torvalds 3 * 4 * 29 dec 2001 - Fixed oopses caused by unchecked access to the vm86 5 * stack - Manfred Spraul <manfred@colorfullife.com> 6 * 7 * 22 mar 2002 - Manfred detected the stackfaults, but didn't handle 8 * them correctly. Now the emulation will be in a 9 * consistent state after stackfaults - Kasper Dupont 10 * <kasperd@daimi.au.dk> 11 * 12 * 22 mar 2002 - Added missing clear_IF in set_vflags_* Kasper Dupont 13 * <kasperd@daimi.au.dk> 14 * 15 * ?? ??? 2002 - Fixed premature returns from handle_vm86_fault 16 * caused by Kasper Dupont's changes - Stas Sergeev 17 * 18 * 4 apr 2002 - Fixed CHECK_IF_IN_TRAP broken by Stas' changes. 19 * Kasper Dupont <kasperd@daimi.au.dk> 20 * 21 * 9 apr 2002 - Changed syntax of macros in handle_vm86_fault. 22 * Kasper Dupont <kasperd@daimi.au.dk> 23 * 24 * 9 apr 2002 - Changed stack access macros to jump to a label 25 * instead of returning to userspace. This simplifies 26 * do_int, and is needed by handle_vm6_fault. Kasper 27 * Dupont <kasperd@daimi.au.dk> 28 * 29 */ 30 31 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 32 33 #include <linux/capability.h> 34 #include <linux/errno.h> 35 #include <linux/interrupt.h> 36 #include <linux/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/highmem.h> 43 #include <linux/ptrace.h> 44 #include <linux/audit.h> 45 #include <linux/stddef.h> 46 47 #include <asm/uaccess.h> 48 #include <asm/io.h> 49 #include <asm/tlbflush.h> 50 #include <asm/irq.h> 51 #include <asm/syscalls.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)->pt.ax))[0]) 77 #define AH(regs) (((unsigned char *)&((regs)->pt.ax))[1]) 78 #define IP(regs) (*(unsigned short *)&((regs)->pt.ip)) 79 #define SP(regs) (*(unsigned short *)&((regs)->pt.sp)) 80 81 /* 82 * virtual flags (16 and 32-bit versions) 83 */ 84 #define VFLAGS (*(unsigned short *)&(current->thread.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 /* convert kernel_vm86_regs to vm86_regs */ 94 static int copy_vm86_regs_to_user(struct vm86_regs __user *user, 95 const struct kernel_vm86_regs *regs) 96 { 97 int ret = 0; 98 99 /* 100 * kernel_vm86_regs is missing gs, so copy everything up to 101 * (but not including) orig_eax, and then rest including orig_eax. 102 */ 103 ret += copy_to_user(user, regs, offsetof(struct kernel_vm86_regs, pt.orig_ax)); 104 ret += copy_to_user(&user->orig_eax, ®s->pt.orig_ax, 105 sizeof(struct kernel_vm86_regs) - 106 offsetof(struct kernel_vm86_regs, pt.orig_ax)); 107 108 return ret; 109 } 110 111 /* convert vm86_regs to kernel_vm86_regs */ 112 static int copy_vm86_regs_from_user(struct kernel_vm86_regs *regs, 113 const struct vm86_regs __user *user, 114 unsigned extra) 115 { 116 int ret = 0; 117 118 /* copy ax-fs inclusive */ 119 ret += copy_from_user(regs, user, offsetof(struct kernel_vm86_regs, pt.orig_ax)); 120 /* copy orig_ax-__gsh+extra */ 121 ret += copy_from_user(®s->pt.orig_ax, &user->orig_eax, 122 sizeof(struct kernel_vm86_regs) - 123 offsetof(struct kernel_vm86_regs, pt.orig_ax) + 124 extra); 125 return ret; 126 } 127 128 struct pt_regs *save_v86_state(struct kernel_vm86_regs *regs) 129 { 130 struct tss_struct *tss; 131 struct pt_regs *ret; 132 unsigned long tmp; 133 134 /* 135 * This gets called from entry.S with interrupts disabled, but 136 * from process context. Enable interrupts here, before trying 137 * to access user space. 138 */ 139 local_irq_enable(); 140 141 if (!current->thread.vm86_info) { 142 pr_alert("no vm86_info: BAD\n"); 143 do_exit(SIGSEGV); 144 } 145 set_flags(regs->pt.flags, VEFLAGS, X86_EFLAGS_VIF | current->thread.v86mask); 146 tmp = copy_vm86_regs_to_user(¤t->thread.vm86_info->regs, regs); 147 tmp += put_user(current->thread.screen_bitmap, ¤t->thread.vm86_info->screen_bitmap); 148 if (tmp) { 149 pr_alert("could not access userspace vm86_info\n"); 150 do_exit(SIGSEGV); 151 } 152 153 tss = &per_cpu(init_tss, get_cpu()); 154 current->thread.sp0 = current->thread.saved_sp0; 155 current->thread.sysenter_cs = __KERNEL_CS; 156 load_sp0(tss, ¤t->thread); 157 current->thread.saved_sp0 = 0; 158 put_cpu(); 159 160 ret = KVM86->regs32; 161 162 ret->fs = current->thread.saved_fs; 163 set_user_gs(ret, current->thread.saved_gs); 164 165 return ret; 166 } 167 168 static void mark_screen_rdonly(struct mm_struct *mm) 169 { 170 pgd_t *pgd; 171 pud_t *pud; 172 pmd_t *pmd; 173 pte_t *pte; 174 spinlock_t *ptl; 175 int i; 176 177 down_write(&mm->mmap_sem); 178 pgd = pgd_offset(mm, 0xA0000); 179 if (pgd_none_or_clear_bad(pgd)) 180 goto out; 181 pud = pud_offset(pgd, 0xA0000); 182 if (pud_none_or_clear_bad(pud)) 183 goto out; 184 pmd = pmd_offset(pud, 0xA0000); 185 split_huge_page_pmd_mm(mm, 0xA0000, pmd); 186 if (pmd_none_or_clear_bad(pmd)) 187 goto out; 188 pte = pte_offset_map_lock(mm, pmd, 0xA0000, &ptl); 189 for (i = 0; i < 32; i++) { 190 if (pte_present(*pte)) 191 set_pte(pte, pte_wrprotect(*pte)); 192 pte++; 193 } 194 pte_unmap_unlock(pte, ptl); 195 out: 196 up_write(&mm->mmap_sem); 197 flush_tlb(); 198 } 199 200 201 202 static int do_vm86_irq_handling(int subfunction, int irqnumber); 203 static void do_sys_vm86(struct kernel_vm86_struct *info, struct task_struct *tsk); 204 205 int sys_vm86old(struct vm86_struct __user *v86, struct pt_regs *regs) 206 { 207 struct kernel_vm86_struct info; /* declare this _on top_, 208 * this avoids wasting of stack space. 209 * This remains on the stack until we 210 * return to 32 bit user space. 211 */ 212 struct task_struct *tsk; 213 int tmp, ret = -EPERM; 214 215 tsk = current; 216 if (tsk->thread.saved_sp0) 217 goto out; 218 tmp = copy_vm86_regs_from_user(&info.regs, &v86->regs, 219 offsetof(struct kernel_vm86_struct, vm86plus) - 220 sizeof(info.regs)); 221 ret = -EFAULT; 222 if (tmp) 223 goto out; 224 memset(&info.vm86plus, 0, (int)&info.regs32 - (int)&info.vm86plus); 225 info.regs32 = regs; 226 tsk->thread.vm86_info = v86; 227 do_sys_vm86(&info, tsk); 228 ret = 0; /* we never return here */ 229 out: 230 return ret; 231 } 232 233 234 int sys_vm86(unsigned long cmd, unsigned long arg, struct pt_regs *regs) 235 { 236 struct kernel_vm86_struct info; /* declare this _on top_, 237 * this avoids wasting of stack space. 238 * This remains on the stack until we 239 * return to 32 bit user space. 240 */ 241 struct task_struct *tsk; 242 int tmp, ret; 243 struct vm86plus_struct __user *v86; 244 245 tsk = current; 246 switch (cmd) { 247 case VM86_REQUEST_IRQ: 248 case VM86_FREE_IRQ: 249 case VM86_GET_IRQ_BITS: 250 case VM86_GET_AND_RESET_IRQ: 251 ret = do_vm86_irq_handling(cmd, (int)arg); 252 goto out; 253 case VM86_PLUS_INSTALL_CHECK: 254 /* 255 * NOTE: on old vm86 stuff this will return the error 256 * from access_ok(), because the subfunction is 257 * interpreted as (invalid) address to vm86_struct. 258 * So the installation check works. 259 */ 260 ret = 0; 261 goto out; 262 } 263 264 /* we come here only for functions VM86_ENTER, VM86_ENTER_NO_BYPASS */ 265 ret = -EPERM; 266 if (tsk->thread.saved_sp0) 267 goto out; 268 v86 = (struct vm86plus_struct __user *)arg; 269 tmp = copy_vm86_regs_from_user(&info.regs, &v86->regs, 270 offsetof(struct kernel_vm86_struct, regs32) - 271 sizeof(info.regs)); 272 ret = -EFAULT; 273 if (tmp) 274 goto out; 275 info.regs32 = regs; 276 info.vm86plus.is_vm86pus = 1; 277 tsk->thread.vm86_info = (struct vm86_struct __user *)v86; 278 do_sys_vm86(&info, tsk); 279 ret = 0; /* we never return here */ 280 out: 281 return ret; 282 } 283 284 285 static void do_sys_vm86(struct kernel_vm86_struct *info, struct task_struct *tsk) 286 { 287 struct tss_struct *tss; 288 /* 289 * make sure the vm86() system call doesn't try to do anything silly 290 */ 291 info->regs.pt.ds = 0; 292 info->regs.pt.es = 0; 293 info->regs.pt.fs = 0; 294 #ifndef CONFIG_X86_32_LAZY_GS 295 info->regs.pt.gs = 0; 296 #endif 297 298 /* 299 * The flags register is also special: we cannot trust that the user 300 * has set it up safely, so this makes sure interrupt etc flags are 301 * inherited from protected mode. 302 */ 303 VEFLAGS = info->regs.pt.flags; 304 info->regs.pt.flags &= SAFE_MASK; 305 info->regs.pt.flags |= info->regs32->flags & ~SAFE_MASK; 306 info->regs.pt.flags |= X86_VM_MASK; 307 308 switch (info->cpu_type) { 309 case CPU_286: 310 tsk->thread.v86mask = 0; 311 break; 312 case CPU_386: 313 tsk->thread.v86mask = X86_EFLAGS_NT | X86_EFLAGS_IOPL; 314 break; 315 case CPU_486: 316 tsk->thread.v86mask = X86_EFLAGS_AC | X86_EFLAGS_NT | X86_EFLAGS_IOPL; 317 break; 318 default: 319 tsk->thread.v86mask = X86_EFLAGS_ID | X86_EFLAGS_AC | X86_EFLAGS_NT | X86_EFLAGS_IOPL; 320 break; 321 } 322 323 /* 324 * Save old state, set default return value (%ax) to 0 (VM86_SIGNAL) 325 */ 326 info->regs32->ax = VM86_SIGNAL; 327 tsk->thread.saved_sp0 = tsk->thread.sp0; 328 tsk->thread.saved_fs = info->regs32->fs; 329 tsk->thread.saved_gs = get_user_gs(info->regs32); 330 331 tss = &per_cpu(init_tss, get_cpu()); 332 tsk->thread.sp0 = (unsigned long) &info->VM86_TSS_ESP0; 333 if (cpu_has_sep) 334 tsk->thread.sysenter_cs = 0; 335 load_sp0(tss, &tsk->thread); 336 put_cpu(); 337 338 tsk->thread.screen_bitmap = info->screen_bitmap; 339 if (info->flags & VM86_SCREEN_BITMAP) 340 mark_screen_rdonly(tsk->mm); 341 342 /*call __audit_syscall_exit since we do not exit via the normal paths */ 343 #ifdef CONFIG_AUDITSYSCALL 344 if (unlikely(current->audit_context)) 345 __audit_syscall_exit(1, 0); 346 #endif 347 348 __asm__ __volatile__( 349 "movl %0,%%esp\n\t" 350 "movl %1,%%ebp\n\t" 351 #ifdef CONFIG_X86_32_LAZY_GS 352 "mov %2, %%gs\n\t" 353 #endif 354 "jmp resume_userspace" 355 : /* no outputs */ 356 :"r" (&info->regs), "r" (task_thread_info(tsk)), "r" (0)); 357 /* we never return here */ 358 } 359 360 static inline void return_to_32bit(struct kernel_vm86_regs *regs16, int retval) 361 { 362 struct pt_regs *regs32; 363 364 regs32 = save_v86_state(regs16); 365 regs32->ax = retval; 366 __asm__ __volatile__("movl %0,%%esp\n\t" 367 "movl %1,%%ebp\n\t" 368 "jmp resume_userspace" 369 : : "r" (regs32), "r" (current_thread_info())); 370 } 371 372 static inline void set_IF(struct kernel_vm86_regs *regs) 373 { 374 VEFLAGS |= X86_EFLAGS_VIF; 375 if (VEFLAGS & X86_EFLAGS_VIP) 376 return_to_32bit(regs, VM86_STI); 377 } 378 379 static inline void clear_IF(struct kernel_vm86_regs *regs) 380 { 381 VEFLAGS &= ~X86_EFLAGS_VIF; 382 } 383 384 static inline void clear_TF(struct kernel_vm86_regs *regs) 385 { 386 regs->pt.flags &= ~X86_EFLAGS_TF; 387 } 388 389 static inline void clear_AC(struct kernel_vm86_regs *regs) 390 { 391 regs->pt.flags &= ~X86_EFLAGS_AC; 392 } 393 394 /* 395 * It is correct to call set_IF(regs) from the set_vflags_* 396 * functions. However someone forgot to call clear_IF(regs) 397 * in the opposite case. 398 * After the command sequence CLI PUSHF STI POPF you should 399 * end up with interrupts disabled, but you ended up with 400 * interrupts enabled. 401 * ( I was testing my own changes, but the only bug I 402 * could find was in a function I had not changed. ) 403 * [KD] 404 */ 405 406 static inline void set_vflags_long(unsigned long flags, struct kernel_vm86_regs *regs) 407 { 408 set_flags(VEFLAGS, flags, current->thread.v86mask); 409 set_flags(regs->pt.flags, flags, SAFE_MASK); 410 if (flags & X86_EFLAGS_IF) 411 set_IF(regs); 412 else 413 clear_IF(regs); 414 } 415 416 static inline void set_vflags_short(unsigned short flags, struct kernel_vm86_regs *regs) 417 { 418 set_flags(VFLAGS, flags, current->thread.v86mask); 419 set_flags(regs->pt.flags, flags, SAFE_MASK); 420 if (flags & X86_EFLAGS_IF) 421 set_IF(regs); 422 else 423 clear_IF(regs); 424 } 425 426 static inline unsigned long get_vflags(struct kernel_vm86_regs *regs) 427 { 428 unsigned long flags = regs->pt.flags & RETURN_MASK; 429 430 if (VEFLAGS & X86_EFLAGS_VIF) 431 flags |= X86_EFLAGS_IF; 432 flags |= X86_EFLAGS_IOPL; 433 return flags | (VEFLAGS & current->thread.v86mask); 434 } 435 436 static inline int is_revectored(int nr, struct revectored_struct *bitmap) 437 { 438 __asm__ __volatile__("btl %2,%1\n\tsbbl %0,%0" 439 :"=r" (nr) 440 :"m" (*bitmap), "r" (nr)); 441 return nr; 442 } 443 444 #define val_byte(val, n) (((__u8 *)&val)[n]) 445 446 #define pushb(base, ptr, val, err_label) \ 447 do { \ 448 __u8 __val = val; \ 449 ptr--; \ 450 if (put_user(__val, base + ptr) < 0) \ 451 goto err_label; \ 452 } while (0) 453 454 #define pushw(base, ptr, val, err_label) \ 455 do { \ 456 __u16 __val = val; \ 457 ptr--; \ 458 if (put_user(val_byte(__val, 1), base + ptr) < 0) \ 459 goto err_label; \ 460 ptr--; \ 461 if (put_user(val_byte(__val, 0), base + ptr) < 0) \ 462 goto err_label; \ 463 } while (0) 464 465 #define pushl(base, ptr, val, err_label) \ 466 do { \ 467 __u32 __val = val; \ 468 ptr--; \ 469 if (put_user(val_byte(__val, 3), base + ptr) < 0) \ 470 goto err_label; \ 471 ptr--; \ 472 if (put_user(val_byte(__val, 2), base + ptr) < 0) \ 473 goto err_label; \ 474 ptr--; \ 475 if (put_user(val_byte(__val, 1), base + ptr) < 0) \ 476 goto err_label; \ 477 ptr--; \ 478 if (put_user(val_byte(__val, 0), base + ptr) < 0) \ 479 goto err_label; \ 480 } while (0) 481 482 #define popb(base, ptr, err_label) \ 483 ({ \ 484 __u8 __res; \ 485 if (get_user(__res, base + ptr) < 0) \ 486 goto err_label; \ 487 ptr++; \ 488 __res; \ 489 }) 490 491 #define popw(base, ptr, err_label) \ 492 ({ \ 493 __u16 __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 __res; \ 501 }) 502 503 #define popl(base, ptr, err_label) \ 504 ({ \ 505 __u32 __res; \ 506 if (get_user(val_byte(__res, 0), base + ptr) < 0) \ 507 goto err_label; \ 508 ptr++; \ 509 if (get_user(val_byte(__res, 1), base + ptr) < 0) \ 510 goto err_label; \ 511 ptr++; \ 512 if (get_user(val_byte(__res, 2), base + ptr) < 0) \ 513 goto err_label; \ 514 ptr++; \ 515 if (get_user(val_byte(__res, 3), base + ptr) < 0) \ 516 goto err_label; \ 517 ptr++; \ 518 __res; \ 519 }) 520 521 /* There are so many possible reasons for this function to return 522 * VM86_INTx, so adding another doesn't bother me. We can expect 523 * userspace programs to be able to handle it. (Getting a problem 524 * in userspace is always better than an Oops anyway.) [KD] 525 */ 526 static void do_int(struct kernel_vm86_regs *regs, int i, 527 unsigned char __user *ssp, unsigned short sp) 528 { 529 unsigned long __user *intr_ptr; 530 unsigned long segoffs; 531 532 if (regs->pt.cs == BIOSSEG) 533 goto cannot_handle; 534 if (is_revectored(i, &KVM86->int_revectored)) 535 goto cannot_handle; 536 if (i == 0x21 && is_revectored(AH(regs), &KVM86->int21_revectored)) 537 goto cannot_handle; 538 intr_ptr = (unsigned long __user *) (i << 2); 539 if (get_user(segoffs, intr_ptr)) 540 goto cannot_handle; 541 if ((segoffs >> 16) == BIOSSEG) 542 goto cannot_handle; 543 pushw(ssp, sp, get_vflags(regs), cannot_handle); 544 pushw(ssp, sp, regs->pt.cs, cannot_handle); 545 pushw(ssp, sp, IP(regs), cannot_handle); 546 regs->pt.cs = segoffs >> 16; 547 SP(regs) -= 6; 548 IP(regs) = segoffs & 0xffff; 549 clear_TF(regs); 550 clear_IF(regs); 551 clear_AC(regs); 552 return; 553 554 cannot_handle: 555 return_to_32bit(regs, VM86_INTx + (i << 8)); 556 } 557 558 int handle_vm86_trap(struct kernel_vm86_regs *regs, long error_code, int trapno) 559 { 560 if (VMPI.is_vm86pus) { 561 if ((trapno == 3) || (trapno == 1)) { 562 KVM86->regs32->ax = VM86_TRAP + (trapno << 8); 563 /* setting this flag forces the code in entry_32.S to 564 the path where we call save_v86_state() and change 565 the stack pointer to KVM86->regs32 */ 566 set_thread_flag(TIF_NOTIFY_RESUME); 567 return 0; 568 } 569 do_int(regs, trapno, (unsigned char __user *) (regs->pt.ss << 4), SP(regs)); 570 return 0; 571 } 572 if (trapno != 1) 573 return 1; /* we let this handle by the calling routine */ 574 current->thread.trap_nr = trapno; 575 current->thread.error_code = error_code; 576 force_sig(SIGTRAP, current); 577 return 0; 578 } 579 580 void handle_vm86_fault(struct kernel_vm86_regs *regs, long error_code) 581 { 582 unsigned char opcode; 583 unsigned char __user *csp; 584 unsigned char __user *ssp; 585 unsigned short ip, sp, orig_flags; 586 int data32, pref_done; 587 588 #define CHECK_IF_IN_TRAP \ 589 if (VMPI.vm86dbg_active && VMPI.vm86dbg_TFpendig) \ 590 newflags |= X86_EFLAGS_TF 591 #define VM86_FAULT_RETURN do { \ 592 if (VMPI.force_return_for_pic && (VEFLAGS & (X86_EFLAGS_IF | X86_EFLAGS_VIF))) \ 593 return_to_32bit(regs, VM86_PICRETURN); \ 594 if (orig_flags & X86_EFLAGS_TF) \ 595 handle_vm86_trap(regs, 0, 1); \ 596 return; } while (0) 597 598 orig_flags = *(unsigned short *)®s->pt.flags; 599 600 csp = (unsigned char __user *) (regs->pt.cs << 4); 601 ssp = (unsigned char __user *) (regs->pt.ss << 4); 602 sp = SP(regs); 603 ip = IP(regs); 604 605 data32 = 0; 606 pref_done = 0; 607 do { 608 switch (opcode = popb(csp, ip, simulate_sigsegv)) { 609 case 0x66: /* 32-bit data */ data32 = 1; break; 610 case 0x67: /* 32-bit address */ break; 611 case 0x2e: /* CS */ break; 612 case 0x3e: /* DS */ break; 613 case 0x26: /* ES */ break; 614 case 0x36: /* SS */ break; 615 case 0x65: /* GS */ break; 616 case 0x64: /* FS */ break; 617 case 0xf2: /* repnz */ break; 618 case 0xf3: /* rep */ break; 619 default: pref_done = 1; 620 } 621 } while (!pref_done); 622 623 switch (opcode) { 624 625 /* pushf */ 626 case 0x9c: 627 if (data32) { 628 pushl(ssp, sp, get_vflags(regs), simulate_sigsegv); 629 SP(regs) -= 4; 630 } else { 631 pushw(ssp, sp, get_vflags(regs), simulate_sigsegv); 632 SP(regs) -= 2; 633 } 634 IP(regs) = ip; 635 VM86_FAULT_RETURN; 636 637 /* popf */ 638 case 0x9d: 639 { 640 unsigned long newflags; 641 if (data32) { 642 newflags = popl(ssp, sp, simulate_sigsegv); 643 SP(regs) += 4; 644 } else { 645 newflags = popw(ssp, sp, simulate_sigsegv); 646 SP(regs) += 2; 647 } 648 IP(regs) = ip; 649 CHECK_IF_IN_TRAP; 650 if (data32) 651 set_vflags_long(newflags, regs); 652 else 653 set_vflags_short(newflags, regs); 654 655 VM86_FAULT_RETURN; 656 } 657 658 /* int xx */ 659 case 0xcd: { 660 int intno = popb(csp, ip, simulate_sigsegv); 661 IP(regs) = ip; 662 if (VMPI.vm86dbg_active) { 663 if ((1 << (intno & 7)) & VMPI.vm86dbg_intxxtab[intno >> 3]) 664 return_to_32bit(regs, VM86_INTx + (intno << 8)); 665 } 666 do_int(regs, intno, ssp, sp); 667 return; 668 } 669 670 /* iret */ 671 case 0xcf: 672 { 673 unsigned long newip; 674 unsigned long newcs; 675 unsigned long newflags; 676 if (data32) { 677 newip = popl(ssp, sp, simulate_sigsegv); 678 newcs = popl(ssp, sp, simulate_sigsegv); 679 newflags = popl(ssp, sp, simulate_sigsegv); 680 SP(regs) += 12; 681 } else { 682 newip = popw(ssp, sp, simulate_sigsegv); 683 newcs = popw(ssp, sp, simulate_sigsegv); 684 newflags = popw(ssp, sp, simulate_sigsegv); 685 SP(regs) += 6; 686 } 687 IP(regs) = newip; 688 regs->pt.cs = newcs; 689 CHECK_IF_IN_TRAP; 690 if (data32) { 691 set_vflags_long(newflags, regs); 692 } else { 693 set_vflags_short(newflags, regs); 694 } 695 VM86_FAULT_RETURN; 696 } 697 698 /* cli */ 699 case 0xfa: 700 IP(regs) = ip; 701 clear_IF(regs); 702 VM86_FAULT_RETURN; 703 704 /* sti */ 705 /* 706 * Damn. This is incorrect: the 'sti' instruction should actually 707 * enable interrupts after the /next/ instruction. Not good. 708 * 709 * Probably needs some horsing around with the TF flag. Aiee.. 710 */ 711 case 0xfb: 712 IP(regs) = ip; 713 set_IF(regs); 714 VM86_FAULT_RETURN; 715 716 default: 717 return_to_32bit(regs, VM86_UNKNOWN); 718 } 719 720 return; 721 722 simulate_sigsegv: 723 /* FIXME: After a long discussion with Stas we finally 724 * agreed, that this is wrong. Here we should 725 * really send a SIGSEGV to the user program. 726 * But how do we create the correct context? We 727 * are inside a general protection fault handler 728 * and has just returned from a page fault handler. 729 * The correct context for the signal handler 730 * should be a mixture of the two, but how do we 731 * get the information? [KD] 732 */ 733 return_to_32bit(regs, VM86_UNKNOWN); 734 } 735 736 /* ---------------- vm86 special IRQ passing stuff ----------------- */ 737 738 #define VM86_IRQNAME "vm86irq" 739 740 static struct vm86_irqs { 741 struct task_struct *tsk; 742 int sig; 743 } vm86_irqs[16]; 744 745 static DEFINE_SPINLOCK(irqbits_lock); 746 static int irqbits; 747 748 #define ALLOWED_SIGS (1 /* 0 = don't send a signal */ \ 749 | (1 << SIGUSR1) | (1 << SIGUSR2) | (1 << SIGIO) | (1 << SIGURG) \ 750 | (1 << SIGUNUSED)) 751 752 static irqreturn_t irq_handler(int intno, void *dev_id) 753 { 754 int irq_bit; 755 unsigned long flags; 756 757 spin_lock_irqsave(&irqbits_lock, flags); 758 irq_bit = 1 << intno; 759 if ((irqbits & irq_bit) || !vm86_irqs[intno].tsk) 760 goto out; 761 irqbits |= irq_bit; 762 if (vm86_irqs[intno].sig) 763 send_sig(vm86_irqs[intno].sig, vm86_irqs[intno].tsk, 1); 764 /* 765 * IRQ will be re-enabled when user asks for the irq (whether 766 * polling or as a result of the signal) 767 */ 768 disable_irq_nosync(intno); 769 spin_unlock_irqrestore(&irqbits_lock, flags); 770 return IRQ_HANDLED; 771 772 out: 773 spin_unlock_irqrestore(&irqbits_lock, flags); 774 return IRQ_NONE; 775 } 776 777 static inline void free_vm86_irq(int irqnumber) 778 { 779 unsigned long flags; 780 781 free_irq(irqnumber, NULL); 782 vm86_irqs[irqnumber].tsk = NULL; 783 784 spin_lock_irqsave(&irqbits_lock, flags); 785 irqbits &= ~(1 << irqnumber); 786 spin_unlock_irqrestore(&irqbits_lock, flags); 787 } 788 789 void release_vm86_irqs(struct task_struct *task) 790 { 791 int i; 792 for (i = FIRST_VM86_IRQ ; i <= LAST_VM86_IRQ; i++) 793 if (vm86_irqs[i].tsk == task) 794 free_vm86_irq(i); 795 } 796 797 static inline int get_and_reset_irq(int irqnumber) 798 { 799 int bit; 800 unsigned long flags; 801 int ret = 0; 802 803 if (invalid_vm86_irq(irqnumber)) return 0; 804 if (vm86_irqs[irqnumber].tsk != current) return 0; 805 spin_lock_irqsave(&irqbits_lock, flags); 806 bit = irqbits & (1 << irqnumber); 807 irqbits &= ~bit; 808 if (bit) { 809 enable_irq(irqnumber); 810 ret = 1; 811 } 812 813 spin_unlock_irqrestore(&irqbits_lock, flags); 814 return ret; 815 } 816 817 818 static int do_vm86_irq_handling(int subfunction, int irqnumber) 819 { 820 int ret; 821 switch (subfunction) { 822 case VM86_GET_AND_RESET_IRQ: { 823 return get_and_reset_irq(irqnumber); 824 } 825 case VM86_GET_IRQ_BITS: { 826 return irqbits; 827 } 828 case VM86_REQUEST_IRQ: { 829 int sig = irqnumber >> 8; 830 int irq = irqnumber & 255; 831 if (!capable(CAP_SYS_ADMIN)) return -EPERM; 832 if (!((1 << sig) & ALLOWED_SIGS)) return -EPERM; 833 if (invalid_vm86_irq(irq)) return -EPERM; 834 if (vm86_irqs[irq].tsk) return -EPERM; 835 ret = request_irq(irq, &irq_handler, 0, VM86_IRQNAME, NULL); 836 if (ret) return ret; 837 vm86_irqs[irq].sig = sig; 838 vm86_irqs[irq].tsk = current; 839 return irq; 840 } 841 case VM86_FREE_IRQ: { 842 if (invalid_vm86_irq(irqnumber)) return -EPERM; 843 if (!vm86_irqs[irqnumber].tsk) return 0; 844 if (vm86_irqs[irqnumber].tsk != current) return -EPERM; 845 free_vm86_irq(irqnumber); 846 return 0; 847 } 848 } 849 return -EINVAL; 850 } 851 852