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/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 52 /* 53 * Known problems: 54 * 55 * Interrupt handling is not guaranteed: 56 * - a real x86 will disable all interrupts for one instruction 57 * after a "mov ss,xx" to make stack handling atomic even without 58 * the 'lss' instruction. We can't guarantee this in v86 mode, 59 * as the next instruction might result in a page fault or similar. 60 * - a real x86 will have interrupts disabled for one instruction 61 * past the 'sti' that enables them. We don't bother with all the 62 * details yet. 63 * 64 * Let's hope these problems do not actually matter for anything. 65 */ 66 67 68 #define KVM86 ((struct kernel_vm86_struct *)regs) 69 #define VMPI KVM86->vm86plus 70 71 72 /* 73 * 8- and 16-bit register defines.. 74 */ 75 #define AL(regs) (((unsigned char *)&((regs)->pt.eax))[0]) 76 #define AH(regs) (((unsigned char *)&((regs)->pt.eax))[1]) 77 #define IP(regs) (*(unsigned short *)&((regs)->pt.eip)) 78 #define SP(regs) (*(unsigned short *)&((regs)->pt.esp)) 79 80 /* 81 * virtual flags (16 and 32-bit versions) 82 */ 83 #define VFLAGS (*(unsigned short *)&(current->thread.v86flags)) 84 #define VEFLAGS (current->thread.v86flags) 85 86 #define set_flags(X,new,mask) \ 87 ((X) = ((X) & ~(mask)) | ((new) & (mask))) 88 89 #define SAFE_MASK (0xDD5) 90 #define RETURN_MASK (0xDFF) 91 92 /* convert kernel_vm86_regs to vm86_regs */ 93 static int copy_vm86_regs_to_user(struct vm86_regs __user *user, 94 const struct kernel_vm86_regs *regs) 95 { 96 int ret = 0; 97 98 /* kernel_vm86_regs is missing xgs, so copy everything up to 99 (but not including) orig_eax, and then rest including orig_eax. */ 100 ret += copy_to_user(user, regs, offsetof(struct kernel_vm86_regs, pt.orig_eax)); 101 ret += copy_to_user(&user->orig_eax, ®s->pt.orig_eax, 102 sizeof(struct kernel_vm86_regs) - 103 offsetof(struct kernel_vm86_regs, pt.orig_eax)); 104 105 return ret; 106 } 107 108 /* convert vm86_regs to kernel_vm86_regs */ 109 static int copy_vm86_regs_from_user(struct kernel_vm86_regs *regs, 110 const struct vm86_regs __user *user, 111 unsigned extra) 112 { 113 int ret = 0; 114 115 /* copy eax-xfs inclusive */ 116 ret += copy_from_user(regs, user, offsetof(struct kernel_vm86_regs, pt.orig_eax)); 117 /* copy orig_eax-__gsh+extra */ 118 ret += copy_from_user(®s->pt.orig_eax, &user->orig_eax, 119 sizeof(struct kernel_vm86_regs) - 120 offsetof(struct kernel_vm86_regs, pt.orig_eax) + 121 extra); 122 return ret; 123 } 124 125 struct pt_regs * FASTCALL(save_v86_state(struct kernel_vm86_regs * regs)); 126 struct pt_regs * fastcall save_v86_state(struct kernel_vm86_regs * regs) 127 { 128 struct tss_struct *tss; 129 struct pt_regs *ret; 130 unsigned long tmp; 131 132 /* 133 * This gets called from entry.S with interrupts disabled, but 134 * from process context. Enable interrupts here, before trying 135 * to access user space. 136 */ 137 local_irq_enable(); 138 139 if (!current->thread.vm86_info) { 140 printk("no vm86_info: BAD\n"); 141 do_exit(SIGSEGV); 142 } 143 set_flags(regs->pt.eflags, VEFLAGS, VIF_MASK | current->thread.v86mask); 144 tmp = copy_vm86_regs_to_user(¤t->thread.vm86_info->regs,regs); 145 tmp += put_user(current->thread.screen_bitmap,¤t->thread.vm86_info->screen_bitmap); 146 if (tmp) { 147 printk("vm86: could not access userspace vm86_info\n"); 148 do_exit(SIGSEGV); 149 } 150 151 tss = &per_cpu(init_tss, get_cpu()); 152 current->thread.esp0 = current->thread.saved_esp0; 153 current->thread.sysenter_cs = __KERNEL_CS; 154 load_esp0(tss, ¤t->thread); 155 current->thread.saved_esp0 = 0; 156 put_cpu(); 157 158 ret = KVM86->regs32; 159 160 ret->xfs = current->thread.saved_fs; 161 loadsegment(gs, current->thread.saved_gs); 162 163 return ret; 164 } 165 166 static void mark_screen_rdonly(struct mm_struct *mm) 167 { 168 pgd_t *pgd; 169 pud_t *pud; 170 pmd_t *pmd; 171 pte_t *pte; 172 spinlock_t *ptl; 173 int i; 174 175 pgd = pgd_offset(mm, 0xA0000); 176 if (pgd_none_or_clear_bad(pgd)) 177 goto out; 178 pud = pud_offset(pgd, 0xA0000); 179 if (pud_none_or_clear_bad(pud)) 180 goto out; 181 pmd = pmd_offset(pud, 0xA0000); 182 if (pmd_none_or_clear_bad(pmd)) 183 goto out; 184 pte = pte_offset_map_lock(mm, pmd, 0xA0000, &ptl); 185 for (i = 0; i < 32; i++) { 186 if (pte_present(*pte)) 187 set_pte(pte, pte_wrprotect(*pte)); 188 pte++; 189 } 190 pte_unmap_unlock(pte, ptl); 191 out: 192 flush_tlb(); 193 } 194 195 196 197 static int do_vm86_irq_handling(int subfunction, int irqnumber); 198 static void do_sys_vm86(struct kernel_vm86_struct *info, struct task_struct *tsk); 199 200 asmlinkage int sys_vm86old(struct pt_regs regs) 201 { 202 struct vm86_struct __user *v86 = (struct vm86_struct __user *)regs.ebx; 203 struct kernel_vm86_struct info; /* declare this _on top_, 204 * this avoids wasting of stack space. 205 * This remains on the stack until we 206 * return to 32 bit user space. 207 */ 208 struct task_struct *tsk; 209 int tmp, ret = -EPERM; 210 211 tsk = current; 212 if (tsk->thread.saved_esp0) 213 goto out; 214 tmp = copy_vm86_regs_from_user(&info.regs, &v86->regs, 215 offsetof(struct kernel_vm86_struct, vm86plus) - 216 sizeof(info.regs)); 217 ret = -EFAULT; 218 if (tmp) 219 goto out; 220 memset(&info.vm86plus, 0, (int)&info.regs32 - (int)&info.vm86plus); 221 info.regs32 = ®s; 222 tsk->thread.vm86_info = v86; 223 do_sys_vm86(&info, tsk); 224 ret = 0; /* we never return here */ 225 out: 226 return ret; 227 } 228 229 230 asmlinkage int sys_vm86(struct pt_regs regs) 231 { 232 struct kernel_vm86_struct info; /* declare this _on top_, 233 * this avoids wasting of stack space. 234 * This remains on the stack until we 235 * return to 32 bit user space. 236 */ 237 struct task_struct *tsk; 238 int tmp, ret; 239 struct vm86plus_struct __user *v86; 240 241 tsk = current; 242 switch (regs.ebx) { 243 case VM86_REQUEST_IRQ: 244 case VM86_FREE_IRQ: 245 case VM86_GET_IRQ_BITS: 246 case VM86_GET_AND_RESET_IRQ: 247 ret = do_vm86_irq_handling(regs.ebx, (int)regs.ecx); 248 goto out; 249 case VM86_PLUS_INSTALL_CHECK: 250 /* NOTE: on old vm86 stuff this will return the error 251 from access_ok(), because the subfunction is 252 interpreted as (invalid) address to vm86_struct. 253 So the installation check works. 254 */ 255 ret = 0; 256 goto out; 257 } 258 259 /* we come here only for functions VM86_ENTER, VM86_ENTER_NO_BYPASS */ 260 ret = -EPERM; 261 if (tsk->thread.saved_esp0) 262 goto out; 263 v86 = (struct vm86plus_struct __user *)regs.ecx; 264 tmp = copy_vm86_regs_from_user(&info.regs, &v86->regs, 265 offsetof(struct kernel_vm86_struct, regs32) - 266 sizeof(info.regs)); 267 ret = -EFAULT; 268 if (tmp) 269 goto out; 270 info.regs32 = ®s; 271 info.vm86plus.is_vm86pus = 1; 272 tsk->thread.vm86_info = (struct vm86_struct __user *)v86; 273 do_sys_vm86(&info, tsk); 274 ret = 0; /* we never return here */ 275 out: 276 return ret; 277 } 278 279 280 static void do_sys_vm86(struct kernel_vm86_struct *info, struct task_struct *tsk) 281 { 282 struct tss_struct *tss; 283 /* 284 * make sure the vm86() system call doesn't try to do anything silly 285 */ 286 info->regs.pt.xds = 0; 287 info->regs.pt.xes = 0; 288 info->regs.pt.xfs = 0; 289 290 /* we are clearing gs later just before "jmp resume_userspace", 291 * because it is not saved/restored. 292 */ 293 294 /* 295 * The eflags register is also special: we cannot trust that the user 296 * has set it up safely, so this makes sure interrupt etc flags are 297 * inherited from protected mode. 298 */ 299 VEFLAGS = info->regs.pt.eflags; 300 info->regs.pt.eflags &= SAFE_MASK; 301 info->regs.pt.eflags |= info->regs32->eflags & ~SAFE_MASK; 302 info->regs.pt.eflags |= VM_MASK; 303 304 switch (info->cpu_type) { 305 case CPU_286: 306 tsk->thread.v86mask = 0; 307 break; 308 case CPU_386: 309 tsk->thread.v86mask = NT_MASK | IOPL_MASK; 310 break; 311 case CPU_486: 312 tsk->thread.v86mask = AC_MASK | NT_MASK | IOPL_MASK; 313 break; 314 default: 315 tsk->thread.v86mask = ID_MASK | AC_MASK | NT_MASK | IOPL_MASK; 316 break; 317 } 318 319 /* 320 * Save old state, set default return value (%eax) to 0 321 */ 322 info->regs32->eax = 0; 323 tsk->thread.saved_esp0 = tsk->thread.esp0; 324 tsk->thread.saved_fs = info->regs32->xfs; 325 savesegment(gs, tsk->thread.saved_gs); 326 327 tss = &per_cpu(init_tss, get_cpu()); 328 tsk->thread.esp0 = (unsigned long) &info->VM86_TSS_ESP0; 329 if (cpu_has_sep) 330 tsk->thread.sysenter_cs = 0; 331 load_esp0(tss, &tsk->thread); 332 put_cpu(); 333 334 tsk->thread.screen_bitmap = info->screen_bitmap; 335 if (info->flags & VM86_SCREEN_BITMAP) 336 mark_screen_rdonly(tsk->mm); 337 338 /*call audit_syscall_exit since we do not exit via the normal paths */ 339 if (unlikely(current->audit_context)) 340 audit_syscall_exit(AUDITSC_RESULT(0), 0); 341 342 __asm__ __volatile__( 343 "movl %0,%%esp\n\t" 344 "movl %1,%%ebp\n\t" 345 "mov %2, %%gs\n\t" 346 "jmp resume_userspace" 347 : /* no outputs */ 348 :"r" (&info->regs), "r" (task_thread_info(tsk)), "r" (0)); 349 /* we never return here */ 350 } 351 352 static inline void return_to_32bit(struct kernel_vm86_regs * regs16, int retval) 353 { 354 struct pt_regs * regs32; 355 356 regs32 = save_v86_state(regs16); 357 regs32->eax = retval; 358 __asm__ __volatile__("movl %0,%%esp\n\t" 359 "movl %1,%%ebp\n\t" 360 "jmp resume_userspace" 361 : : "r" (regs32), "r" (current_thread_info())); 362 } 363 364 static inline void set_IF(struct kernel_vm86_regs * regs) 365 { 366 VEFLAGS |= VIF_MASK; 367 if (VEFLAGS & VIP_MASK) 368 return_to_32bit(regs, VM86_STI); 369 } 370 371 static inline void clear_IF(struct kernel_vm86_regs * regs) 372 { 373 VEFLAGS &= ~VIF_MASK; 374 } 375 376 static inline void clear_TF(struct kernel_vm86_regs * regs) 377 { 378 regs->pt.eflags &= ~TF_MASK; 379 } 380 381 static inline void clear_AC(struct kernel_vm86_regs * regs) 382 { 383 regs->pt.eflags &= ~AC_MASK; 384 } 385 386 /* It is correct to call set_IF(regs) from the set_vflags_* 387 * functions. However someone forgot to call clear_IF(regs) 388 * in the opposite case. 389 * After the command sequence CLI PUSHF STI POPF you should 390 * end up with interrups disabled, but you ended up with 391 * interrupts enabled. 392 * ( I was testing my own changes, but the only bug I 393 * could find was in a function I had not changed. ) 394 * [KD] 395 */ 396 397 static inline void set_vflags_long(unsigned long eflags, struct kernel_vm86_regs * regs) 398 { 399 set_flags(VEFLAGS, eflags, current->thread.v86mask); 400 set_flags(regs->pt.eflags, eflags, SAFE_MASK); 401 if (eflags & IF_MASK) 402 set_IF(regs); 403 else 404 clear_IF(regs); 405 } 406 407 static inline void set_vflags_short(unsigned short flags, struct kernel_vm86_regs * regs) 408 { 409 set_flags(VFLAGS, flags, current->thread.v86mask); 410 set_flags(regs->pt.eflags, flags, SAFE_MASK); 411 if (flags & IF_MASK) 412 set_IF(regs); 413 else 414 clear_IF(regs); 415 } 416 417 static inline unsigned long get_vflags(struct kernel_vm86_regs * regs) 418 { 419 unsigned long flags = regs->pt.eflags & RETURN_MASK; 420 421 if (VEFLAGS & VIF_MASK) 422 flags |= IF_MASK; 423 flags |= IOPL_MASK; 424 return flags | (VEFLAGS & current->thread.v86mask); 425 } 426 427 static inline int is_revectored(int nr, struct revectored_struct * bitmap) 428 { 429 __asm__ __volatile__("btl %2,%1\n\tsbbl %0,%0" 430 :"=r" (nr) 431 :"m" (*bitmap),"r" (nr)); 432 return nr; 433 } 434 435 #define val_byte(val, n) (((__u8 *)&val)[n]) 436 437 #define pushb(base, ptr, val, err_label) \ 438 do { \ 439 __u8 __val = val; \ 440 ptr--; \ 441 if (put_user(__val, base + ptr) < 0) \ 442 goto err_label; \ 443 } while(0) 444 445 #define pushw(base, ptr, val, err_label) \ 446 do { \ 447 __u16 __val = val; \ 448 ptr--; \ 449 if (put_user(val_byte(__val, 1), base + ptr) < 0) \ 450 goto err_label; \ 451 ptr--; \ 452 if (put_user(val_byte(__val, 0), base + ptr) < 0) \ 453 goto err_label; \ 454 } while(0) 455 456 #define pushl(base, ptr, val, err_label) \ 457 do { \ 458 __u32 __val = val; \ 459 ptr--; \ 460 if (put_user(val_byte(__val, 3), base + ptr) < 0) \ 461 goto err_label; \ 462 ptr--; \ 463 if (put_user(val_byte(__val, 2), base + ptr) < 0) \ 464 goto err_label; \ 465 ptr--; \ 466 if (put_user(val_byte(__val, 1), base + ptr) < 0) \ 467 goto err_label; \ 468 ptr--; \ 469 if (put_user(val_byte(__val, 0), base + ptr) < 0) \ 470 goto err_label; \ 471 } while(0) 472 473 #define popb(base, ptr, err_label) \ 474 ({ \ 475 __u8 __res; \ 476 if (get_user(__res, base + ptr) < 0) \ 477 goto err_label; \ 478 ptr++; \ 479 __res; \ 480 }) 481 482 #define popw(base, ptr, err_label) \ 483 ({ \ 484 __u16 __res; \ 485 if (get_user(val_byte(__res, 0), base + ptr) < 0) \ 486 goto err_label; \ 487 ptr++; \ 488 if (get_user(val_byte(__res, 1), base + ptr) < 0) \ 489 goto err_label; \ 490 ptr++; \ 491 __res; \ 492 }) 493 494 #define popl(base, ptr, err_label) \ 495 ({ \ 496 __u32 __res; \ 497 if (get_user(val_byte(__res, 0), base + ptr) < 0) \ 498 goto err_label; \ 499 ptr++; \ 500 if (get_user(val_byte(__res, 1), base + ptr) < 0) \ 501 goto err_label; \ 502 ptr++; \ 503 if (get_user(val_byte(__res, 2), base + ptr) < 0) \ 504 goto err_label; \ 505 ptr++; \ 506 if (get_user(val_byte(__res, 3), base + ptr) < 0) \ 507 goto err_label; \ 508 ptr++; \ 509 __res; \ 510 }) 511 512 /* There are so many possible reasons for this function to return 513 * VM86_INTx, so adding another doesn't bother me. We can expect 514 * userspace programs to be able to handle it. (Getting a problem 515 * in userspace is always better than an Oops anyway.) [KD] 516 */ 517 static void do_int(struct kernel_vm86_regs *regs, int i, 518 unsigned char __user * ssp, unsigned short sp) 519 { 520 unsigned long __user *intr_ptr; 521 unsigned long segoffs; 522 523 if (regs->pt.xcs == BIOSSEG) 524 goto cannot_handle; 525 if (is_revectored(i, &KVM86->int_revectored)) 526 goto cannot_handle; 527 if (i==0x21 && is_revectored(AH(regs),&KVM86->int21_revectored)) 528 goto cannot_handle; 529 intr_ptr = (unsigned long __user *) (i << 2); 530 if (get_user(segoffs, intr_ptr)) 531 goto cannot_handle; 532 if ((segoffs >> 16) == BIOSSEG) 533 goto cannot_handle; 534 pushw(ssp, sp, get_vflags(regs), cannot_handle); 535 pushw(ssp, sp, regs->pt.xcs, cannot_handle); 536 pushw(ssp, sp, IP(regs), cannot_handle); 537 regs->pt.xcs = segoffs >> 16; 538 SP(regs) -= 6; 539 IP(regs) = segoffs & 0xffff; 540 clear_TF(regs); 541 clear_IF(regs); 542 clear_AC(regs); 543 return; 544 545 cannot_handle: 546 return_to_32bit(regs, VM86_INTx + (i << 8)); 547 } 548 549 int handle_vm86_trap(struct kernel_vm86_regs * regs, long error_code, int trapno) 550 { 551 if (VMPI.is_vm86pus) { 552 if ( (trapno==3) || (trapno==1) ) 553 return_to_32bit(regs, VM86_TRAP + (trapno << 8)); 554 do_int(regs, trapno, (unsigned char __user *) (regs->pt.xss << 4), SP(regs)); 555 return 0; 556 } 557 if (trapno !=1) 558 return 1; /* we let this handle by the calling routine */ 559 if (current->ptrace & PT_PTRACED) { 560 unsigned long flags; 561 spin_lock_irqsave(¤t->sighand->siglock, flags); 562 sigdelset(¤t->blocked, SIGTRAP); 563 recalc_sigpending(); 564 spin_unlock_irqrestore(¤t->sighand->siglock, flags); 565 } 566 send_sig(SIGTRAP, current, 1); 567 current->thread.trap_no = trapno; 568 current->thread.error_code = error_code; 569 return 0; 570 } 571 572 void handle_vm86_fault(struct kernel_vm86_regs * regs, long error_code) 573 { 574 unsigned char opcode; 575 unsigned char __user *csp; 576 unsigned char __user *ssp; 577 unsigned short ip, sp, orig_flags; 578 int data32, pref_done; 579 580 #define CHECK_IF_IN_TRAP \ 581 if (VMPI.vm86dbg_active && VMPI.vm86dbg_TFpendig) \ 582 newflags |= TF_MASK 583 #define VM86_FAULT_RETURN do { \ 584 if (VMPI.force_return_for_pic && (VEFLAGS & (IF_MASK | VIF_MASK))) \ 585 return_to_32bit(regs, VM86_PICRETURN); \ 586 if (orig_flags & TF_MASK) \ 587 handle_vm86_trap(regs, 0, 1); \ 588 return; } while (0) 589 590 orig_flags = *(unsigned short *)®s->pt.eflags; 591 592 csp = (unsigned char __user *) (regs->pt.xcs << 4); 593 ssp = (unsigned char __user *) (regs->pt.xss << 4); 594 sp = SP(regs); 595 ip = IP(regs); 596 597 data32 = 0; 598 pref_done = 0; 599 do { 600 switch (opcode = popb(csp, ip, simulate_sigsegv)) { 601 case 0x66: /* 32-bit data */ data32=1; break; 602 case 0x67: /* 32-bit address */ break; 603 case 0x2e: /* CS */ break; 604 case 0x3e: /* DS */ break; 605 case 0x26: /* ES */ break; 606 case 0x36: /* SS */ break; 607 case 0x65: /* GS */ break; 608 case 0x64: /* FS */ break; 609 case 0xf2: /* repnz */ break; 610 case 0xf3: /* rep */ break; 611 default: pref_done = 1; 612 } 613 } while (!pref_done); 614 615 switch (opcode) { 616 617 /* pushf */ 618 case 0x9c: 619 if (data32) { 620 pushl(ssp, sp, get_vflags(regs), simulate_sigsegv); 621 SP(regs) -= 4; 622 } else { 623 pushw(ssp, sp, get_vflags(regs), simulate_sigsegv); 624 SP(regs) -= 2; 625 } 626 IP(regs) = ip; 627 VM86_FAULT_RETURN; 628 629 /* popf */ 630 case 0x9d: 631 { 632 unsigned long newflags; 633 if (data32) { 634 newflags=popl(ssp, sp, simulate_sigsegv); 635 SP(regs) += 4; 636 } else { 637 newflags = popw(ssp, sp, simulate_sigsegv); 638 SP(regs) += 2; 639 } 640 IP(regs) = ip; 641 CHECK_IF_IN_TRAP; 642 if (data32) { 643 set_vflags_long(newflags, regs); 644 } else { 645 set_vflags_short(newflags, regs); 646 } 647 VM86_FAULT_RETURN; 648 } 649 650 /* int xx */ 651 case 0xcd: { 652 int intno=popb(csp, ip, simulate_sigsegv); 653 IP(regs) = ip; 654 if (VMPI.vm86dbg_active) { 655 if ( (1 << (intno &7)) & VMPI.vm86dbg_intxxtab[intno >> 3] ) 656 return_to_32bit(regs, VM86_INTx + (intno << 8)); 657 } 658 do_int(regs, intno, ssp, sp); 659 return; 660 } 661 662 /* iret */ 663 case 0xcf: 664 { 665 unsigned long newip; 666 unsigned long newcs; 667 unsigned long newflags; 668 if (data32) { 669 newip=popl(ssp, sp, simulate_sigsegv); 670 newcs=popl(ssp, sp, simulate_sigsegv); 671 newflags=popl(ssp, sp, simulate_sigsegv); 672 SP(regs) += 12; 673 } else { 674 newip = popw(ssp, sp, simulate_sigsegv); 675 newcs = popw(ssp, sp, simulate_sigsegv); 676 newflags = popw(ssp, sp, simulate_sigsegv); 677 SP(regs) += 6; 678 } 679 IP(regs) = newip; 680 regs->pt.xcs = newcs; 681 CHECK_IF_IN_TRAP; 682 if (data32) { 683 set_vflags_long(newflags, regs); 684 } else { 685 set_vflags_short(newflags, regs); 686 } 687 VM86_FAULT_RETURN; 688 } 689 690 /* cli */ 691 case 0xfa: 692 IP(regs) = ip; 693 clear_IF(regs); 694 VM86_FAULT_RETURN; 695 696 /* sti */ 697 /* 698 * Damn. This is incorrect: the 'sti' instruction should actually 699 * enable interrupts after the /next/ instruction. Not good. 700 * 701 * Probably needs some horsing around with the TF flag. Aiee.. 702 */ 703 case 0xfb: 704 IP(regs) = ip; 705 set_IF(regs); 706 VM86_FAULT_RETURN; 707 708 default: 709 return_to_32bit(regs, VM86_UNKNOWN); 710 } 711 712 return; 713 714 simulate_sigsegv: 715 /* FIXME: After a long discussion with Stas we finally 716 * agreed, that this is wrong. Here we should 717 * really send a SIGSEGV to the user program. 718 * But how do we create the correct context? We 719 * are inside a general protection fault handler 720 * and has just returned from a page fault handler. 721 * The correct context for the signal handler 722 * should be a mixture of the two, but how do we 723 * get the information? [KD] 724 */ 725 return_to_32bit(regs, VM86_UNKNOWN); 726 } 727 728 /* ---------------- vm86 special IRQ passing stuff ----------------- */ 729 730 #define VM86_IRQNAME "vm86irq" 731 732 static struct vm86_irqs { 733 struct task_struct *tsk; 734 int sig; 735 } vm86_irqs[16]; 736 737 static DEFINE_SPINLOCK(irqbits_lock); 738 static int irqbits; 739 740 #define ALLOWED_SIGS ( 1 /* 0 = don't send a signal */ \ 741 | (1 << SIGUSR1) | (1 << SIGUSR2) | (1 << SIGIO) | (1 << SIGURG) \ 742 | (1 << SIGUNUSED) ) 743 744 static irqreturn_t irq_handler(int intno, void *dev_id) 745 { 746 int irq_bit; 747 unsigned long flags; 748 749 spin_lock_irqsave(&irqbits_lock, flags); 750 irq_bit = 1 << intno; 751 if ((irqbits & irq_bit) || ! vm86_irqs[intno].tsk) 752 goto out; 753 irqbits |= irq_bit; 754 if (vm86_irqs[intno].sig) 755 send_sig(vm86_irqs[intno].sig, vm86_irqs[intno].tsk, 1); 756 /* 757 * IRQ will be re-enabled when user asks for the irq (whether 758 * polling or as a result of the signal) 759 */ 760 disable_irq_nosync(intno); 761 spin_unlock_irqrestore(&irqbits_lock, flags); 762 return IRQ_HANDLED; 763 764 out: 765 spin_unlock_irqrestore(&irqbits_lock, flags); 766 return IRQ_NONE; 767 } 768 769 static inline void free_vm86_irq(int irqnumber) 770 { 771 unsigned long flags; 772 773 free_irq(irqnumber, NULL); 774 vm86_irqs[irqnumber].tsk = NULL; 775 776 spin_lock_irqsave(&irqbits_lock, flags); 777 irqbits &= ~(1 << irqnumber); 778 spin_unlock_irqrestore(&irqbits_lock, flags); 779 } 780 781 void release_vm86_irqs(struct task_struct *task) 782 { 783 int i; 784 for (i = FIRST_VM86_IRQ ; i <= LAST_VM86_IRQ; i++) 785 if (vm86_irqs[i].tsk == task) 786 free_vm86_irq(i); 787 } 788 789 static inline int get_and_reset_irq(int irqnumber) 790 { 791 int bit; 792 unsigned long flags; 793 int ret = 0; 794 795 if (invalid_vm86_irq(irqnumber)) return 0; 796 if (vm86_irqs[irqnumber].tsk != current) return 0; 797 spin_lock_irqsave(&irqbits_lock, flags); 798 bit = irqbits & (1 << irqnumber); 799 irqbits &= ~bit; 800 if (bit) { 801 enable_irq(irqnumber); 802 ret = 1; 803 } 804 805 spin_unlock_irqrestore(&irqbits_lock, flags); 806 return ret; 807 } 808 809 810 static int do_vm86_irq_handling(int subfunction, int irqnumber) 811 { 812 int ret; 813 switch (subfunction) { 814 case VM86_GET_AND_RESET_IRQ: { 815 return get_and_reset_irq(irqnumber); 816 } 817 case VM86_GET_IRQ_BITS: { 818 return irqbits; 819 } 820 case VM86_REQUEST_IRQ: { 821 int sig = irqnumber >> 8; 822 int irq = irqnumber & 255; 823 if (!capable(CAP_SYS_ADMIN)) return -EPERM; 824 if (!((1 << sig) & ALLOWED_SIGS)) return -EPERM; 825 if (invalid_vm86_irq(irq)) return -EPERM; 826 if (vm86_irqs[irq].tsk) return -EPERM; 827 ret = request_irq(irq, &irq_handler, 0, VM86_IRQNAME, NULL); 828 if (ret) return ret; 829 vm86_irqs[irq].sig = sig; 830 vm86_irqs[irq].tsk = current; 831 return irq; 832 } 833 case VM86_FREE_IRQ: { 834 if (invalid_vm86_irq(irqnumber)) return -EPERM; 835 if (!vm86_irqs[irqnumber].tsk) return 0; 836 if (vm86_irqs[irqnumber].tsk != current) return -EPERM; 837 free_vm86_irq(irqnumber); 838 return 0; 839 } 840 } 841 return -EINVAL; 842 } 843 844