vm86_32.c revision bae9c19bf12bb2a914a8e530270f41d36cc87c63
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#include <asm/syscalls.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)->pt.ax))[0]) 75#define AH(regs) (((unsigned char *)&((regs)->pt.ax))[1]) 76#define IP(regs) (*(unsigned short *)&((regs)->pt.ip)) 77#define SP(regs) (*(unsigned short *)&((regs)->pt.sp)) 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/* convert kernel_vm86_regs to vm86_regs */ 92static int copy_vm86_regs_to_user(struct vm86_regs __user *user, 93 const struct kernel_vm86_regs *regs) 94{ 95 int ret = 0; 96 97 /* 98 * kernel_vm86_regs is missing gs, so copy everything up to 99 * (but not including) orig_eax, and then rest including orig_eax. 100 */ 101 ret += copy_to_user(user, regs, offsetof(struct kernel_vm86_regs, pt.orig_ax)); 102 ret += copy_to_user(&user->orig_eax, ®s->pt.orig_ax, 103 sizeof(struct kernel_vm86_regs) - 104 offsetof(struct kernel_vm86_regs, pt.orig_ax)); 105 106 return ret; 107} 108 109/* convert vm86_regs to kernel_vm86_regs */ 110static int copy_vm86_regs_from_user(struct kernel_vm86_regs *regs, 111 const struct vm86_regs __user *user, 112 unsigned extra) 113{ 114 int ret = 0; 115 116 /* copy ax-fs inclusive */ 117 ret += copy_from_user(regs, user, offsetof(struct kernel_vm86_regs, pt.orig_ax)); 118 /* copy orig_ax-__gsh+extra */ 119 ret += copy_from_user(®s->pt.orig_ax, &user->orig_eax, 120 sizeof(struct kernel_vm86_regs) - 121 offsetof(struct kernel_vm86_regs, pt.orig_ax) + 122 extra); 123 return ret; 124} 125 126struct pt_regs *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.flags, VEFLAGS, X86_EFLAGS_VIF | 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.sp0 = current->thread.saved_sp0; 153 current->thread.sysenter_cs = __KERNEL_CS; 154 load_sp0(tss, ¤t->thread); 155 current->thread.saved_sp0 = 0; 156 put_cpu(); 157 158 ret = KVM86->regs32; 159 160 ret->fs = current->thread.saved_fs; 161 set_user_gs(ret, current->thread.saved_gs); 162 163 return ret; 164} 165 166static 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 split_huge_page_pmd(mm, pmd); 183 if (pmd_none_or_clear_bad(pmd)) 184 goto out; 185 pte = pte_offset_map_lock(mm, pmd, 0xA0000, &ptl); 186 for (i = 0; i < 32; i++) { 187 if (pte_present(*pte)) 188 set_pte(pte, pte_wrprotect(*pte)); 189 pte++; 190 } 191 pte_unmap_unlock(pte, ptl); 192out: 193 flush_tlb(); 194} 195 196 197 198static int do_vm86_irq_handling(int subfunction, int irqnumber); 199static void do_sys_vm86(struct kernel_vm86_struct *info, struct task_struct *tsk); 200 201int sys_vm86old(struct vm86_struct __user *v86, struct pt_regs *regs) 202{ 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_sp0) 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 = regs; 222 tsk->thread.vm86_info = v86; 223 do_sys_vm86(&info, tsk); 224 ret = 0; /* we never return here */ 225out: 226 return ret; 227} 228 229 230int sys_vm86(unsigned long cmd, unsigned long arg, 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 (cmd) { 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(cmd, (int)arg); 248 goto out; 249 case VM86_PLUS_INSTALL_CHECK: 250 /* 251 * NOTE: on old vm86 stuff this will return the error 252 * from access_ok(), because the subfunction is 253 * interpreted as (invalid) address to vm86_struct. 254 * So the installation check works. 255 */ 256 ret = 0; 257 goto out; 258 } 259 260 /* we come here only for functions VM86_ENTER, VM86_ENTER_NO_BYPASS */ 261 ret = -EPERM; 262 if (tsk->thread.saved_sp0) 263 goto out; 264 v86 = (struct vm86plus_struct __user *)arg; 265 tmp = copy_vm86_regs_from_user(&info.regs, &v86->regs, 266 offsetof(struct kernel_vm86_struct, regs32) - 267 sizeof(info.regs)); 268 ret = -EFAULT; 269 if (tmp) 270 goto out; 271 info.regs32 = regs; 272 info.vm86plus.is_vm86pus = 1; 273 tsk->thread.vm86_info = (struct vm86_struct __user *)v86; 274 do_sys_vm86(&info, tsk); 275 ret = 0; /* we never return here */ 276out: 277 return ret; 278} 279 280 281static void do_sys_vm86(struct kernel_vm86_struct *info, struct task_struct *tsk) 282{ 283 struct tss_struct *tss; 284/* 285 * make sure the vm86() system call doesn't try to do anything silly 286 */ 287 info->regs.pt.ds = 0; 288 info->regs.pt.es = 0; 289 info->regs.pt.fs = 0; 290#ifndef CONFIG_X86_32_LAZY_GS 291 info->regs.pt.gs = 0; 292#endif 293 294/* 295 * The flags 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.flags; 300 info->regs.pt.flags &= SAFE_MASK; 301 info->regs.pt.flags |= info->regs32->flags & ~SAFE_MASK; 302 info->regs.pt.flags |= X86_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 = X86_EFLAGS_NT | X86_EFLAGS_IOPL; 310 break; 311 case CPU_486: 312 tsk->thread.v86mask = X86_EFLAGS_AC | X86_EFLAGS_NT | X86_EFLAGS_IOPL; 313 break; 314 default: 315 tsk->thread.v86mask = X86_EFLAGS_ID | X86_EFLAGS_AC | X86_EFLAGS_NT | X86_EFLAGS_IOPL; 316 break; 317 } 318 319/* 320 * Save old state, set default return value (%ax) to 0 (VM86_SIGNAL) 321 */ 322 info->regs32->ax = VM86_SIGNAL; 323 tsk->thread.saved_sp0 = tsk->thread.sp0; 324 tsk->thread.saved_fs = info->regs32->fs; 325 tsk->thread.saved_gs = get_user_gs(info->regs32); 326 327 tss = &per_cpu(init_tss, get_cpu()); 328 tsk->thread.sp0 = (unsigned long) &info->VM86_TSS_ESP0; 329 if (cpu_has_sep) 330 tsk->thread.sysenter_cs = 0; 331 load_sp0(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#ifdef CONFIG_X86_32_LAZY_GS 346 "mov %2, %%gs\n\t" 347#endif 348 "jmp resume_userspace" 349 : /* no outputs */ 350 :"r" (&info->regs), "r" (task_thread_info(tsk)), "r" (0)); 351 /* we never return here */ 352} 353 354static inline void return_to_32bit(struct kernel_vm86_regs *regs16, int retval) 355{ 356 struct pt_regs *regs32; 357 358 regs32 = save_v86_state(regs16); 359 regs32->ax = retval; 360 __asm__ __volatile__("movl %0,%%esp\n\t" 361 "movl %1,%%ebp\n\t" 362 "jmp resume_userspace" 363 : : "r" (regs32), "r" (current_thread_info())); 364} 365 366static inline void set_IF(struct kernel_vm86_regs *regs) 367{ 368 VEFLAGS |= X86_EFLAGS_VIF; 369 if (VEFLAGS & X86_EFLAGS_VIP) 370 return_to_32bit(regs, VM86_STI); 371} 372 373static inline void clear_IF(struct kernel_vm86_regs *regs) 374{ 375 VEFLAGS &= ~X86_EFLAGS_VIF; 376} 377 378static inline void clear_TF(struct kernel_vm86_regs *regs) 379{ 380 regs->pt.flags &= ~X86_EFLAGS_TF; 381} 382 383static inline void clear_AC(struct kernel_vm86_regs *regs) 384{ 385 regs->pt.flags &= ~X86_EFLAGS_AC; 386} 387 388/* 389 * It is correct to call set_IF(regs) from the set_vflags_* 390 * functions. However someone forgot to call clear_IF(regs) 391 * in the opposite case. 392 * After the command sequence CLI PUSHF STI POPF you should 393 * end up with interrupts disabled, but you ended up with 394 * interrupts enabled. 395 * ( I was testing my own changes, but the only bug I 396 * could find was in a function I had not changed. ) 397 * [KD] 398 */ 399 400static inline void set_vflags_long(unsigned long flags, struct kernel_vm86_regs *regs) 401{ 402 set_flags(VEFLAGS, flags, current->thread.v86mask); 403 set_flags(regs->pt.flags, flags, SAFE_MASK); 404 if (flags & X86_EFLAGS_IF) 405 set_IF(regs); 406 else 407 clear_IF(regs); 408} 409 410static inline void set_vflags_short(unsigned short flags, struct kernel_vm86_regs *regs) 411{ 412 set_flags(VFLAGS, flags, current->thread.v86mask); 413 set_flags(regs->pt.flags, flags, SAFE_MASK); 414 if (flags & X86_EFLAGS_IF) 415 set_IF(regs); 416 else 417 clear_IF(regs); 418} 419 420static inline unsigned long get_vflags(struct kernel_vm86_regs *regs) 421{ 422 unsigned long flags = regs->pt.flags & RETURN_MASK; 423 424 if (VEFLAGS & X86_EFLAGS_VIF) 425 flags |= X86_EFLAGS_IF; 426 flags |= X86_EFLAGS_IOPL; 427 return flags | (VEFLAGS & current->thread.v86mask); 428} 429 430static inline int is_revectored(int nr, struct revectored_struct *bitmap) 431{ 432 __asm__ __volatile__("btl %2,%1\n\tsbbl %0,%0" 433 :"=r" (nr) 434 :"m" (*bitmap), "r" (nr)); 435 return nr; 436} 437 438#define val_byte(val, n) (((__u8 *)&val)[n]) 439 440#define pushb(base, ptr, val, err_label) \ 441 do { \ 442 __u8 __val = val; \ 443 ptr--; \ 444 if (put_user(__val, base + ptr) < 0) \ 445 goto err_label; \ 446 } while (0) 447 448#define pushw(base, ptr, val, err_label) \ 449 do { \ 450 __u16 __val = val; \ 451 ptr--; \ 452 if (put_user(val_byte(__val, 1), base + ptr) < 0) \ 453 goto err_label; \ 454 ptr--; \ 455 if (put_user(val_byte(__val, 0), base + ptr) < 0) \ 456 goto err_label; \ 457 } while (0) 458 459#define pushl(base, ptr, val, err_label) \ 460 do { \ 461 __u32 __val = val; \ 462 ptr--; \ 463 if (put_user(val_byte(__val, 3), base + ptr) < 0) \ 464 goto err_label; \ 465 ptr--; \ 466 if (put_user(val_byte(__val, 2), base + ptr) < 0) \ 467 goto err_label; \ 468 ptr--; \ 469 if (put_user(val_byte(__val, 1), base + ptr) < 0) \ 470 goto err_label; \ 471 ptr--; \ 472 if (put_user(val_byte(__val, 0), base + ptr) < 0) \ 473 goto err_label; \ 474 } while (0) 475 476#define popb(base, ptr, err_label) \ 477 ({ \ 478 __u8 __res; \ 479 if (get_user(__res, base + ptr) < 0) \ 480 goto err_label; \ 481 ptr++; \ 482 __res; \ 483 }) 484 485#define popw(base, ptr, err_label) \ 486 ({ \ 487 __u16 __res; \ 488 if (get_user(val_byte(__res, 0), base + ptr) < 0) \ 489 goto err_label; \ 490 ptr++; \ 491 if (get_user(val_byte(__res, 1), base + ptr) < 0) \ 492 goto err_label; \ 493 ptr++; \ 494 __res; \ 495 }) 496 497#define popl(base, ptr, err_label) \ 498 ({ \ 499 __u32 __res; \ 500 if (get_user(val_byte(__res, 0), base + ptr) < 0) \ 501 goto err_label; \ 502 ptr++; \ 503 if (get_user(val_byte(__res, 1), base + ptr) < 0) \ 504 goto err_label; \ 505 ptr++; \ 506 if (get_user(val_byte(__res, 2), base + ptr) < 0) \ 507 goto err_label; \ 508 ptr++; \ 509 if (get_user(val_byte(__res, 3), base + ptr) < 0) \ 510 goto err_label; \ 511 ptr++; \ 512 __res; \ 513 }) 514 515/* There are so many possible reasons for this function to return 516 * VM86_INTx, so adding another doesn't bother me. We can expect 517 * userspace programs to be able to handle it. (Getting a problem 518 * in userspace is always better than an Oops anyway.) [KD] 519 */ 520static void do_int(struct kernel_vm86_regs *regs, int i, 521 unsigned char __user *ssp, unsigned short sp) 522{ 523 unsigned long __user *intr_ptr; 524 unsigned long segoffs; 525 526 if (regs->pt.cs == BIOSSEG) 527 goto cannot_handle; 528 if (is_revectored(i, &KVM86->int_revectored)) 529 goto cannot_handle; 530 if (i == 0x21 && is_revectored(AH(regs), &KVM86->int21_revectored)) 531 goto cannot_handle; 532 intr_ptr = (unsigned long __user *) (i << 2); 533 if (get_user(segoffs, intr_ptr)) 534 goto cannot_handle; 535 if ((segoffs >> 16) == BIOSSEG) 536 goto cannot_handle; 537 pushw(ssp, sp, get_vflags(regs), cannot_handle); 538 pushw(ssp, sp, regs->pt.cs, cannot_handle); 539 pushw(ssp, sp, IP(regs), cannot_handle); 540 regs->pt.cs = segoffs >> 16; 541 SP(regs) -= 6; 542 IP(regs) = segoffs & 0xffff; 543 clear_TF(regs); 544 clear_IF(regs); 545 clear_AC(regs); 546 return; 547 548cannot_handle: 549 return_to_32bit(regs, VM86_INTx + (i << 8)); 550} 551 552int handle_vm86_trap(struct kernel_vm86_regs *regs, long error_code, int trapno) 553{ 554 if (VMPI.is_vm86pus) { 555 if ((trapno == 3) || (trapno == 1)) { 556 KVM86->regs32->ax = VM86_TRAP + (trapno << 8); 557 /* setting this flag forces the code in entry_32.S to 558 call save_v86_state() and change the stack pointer 559 to KVM86->regs32 */ 560 set_thread_flag(TIF_IRET); 561 return 0; 562 } 563 do_int(regs, trapno, (unsigned char __user *) (regs->pt.ss << 4), SP(regs)); 564 return 0; 565 } 566 if (trapno != 1) 567 return 1; /* we let this handle by the calling routine */ 568 current->thread.trap_no = trapno; 569 current->thread.error_code = error_code; 570 force_sig(SIGTRAP, current); 571 return 0; 572} 573 574void handle_vm86_fault(struct kernel_vm86_regs *regs, long error_code) 575{ 576 unsigned char opcode; 577 unsigned char __user *csp; 578 unsigned char __user *ssp; 579 unsigned short ip, sp, orig_flags; 580 int data32, pref_done; 581 582#define CHECK_IF_IN_TRAP \ 583 if (VMPI.vm86dbg_active && VMPI.vm86dbg_TFpendig) \ 584 newflags |= X86_EFLAGS_TF 585#define VM86_FAULT_RETURN do { \ 586 if (VMPI.force_return_for_pic && (VEFLAGS & (X86_EFLAGS_IF | X86_EFLAGS_VIF))) \ 587 return_to_32bit(regs, VM86_PICRETURN); \ 588 if (orig_flags & X86_EFLAGS_TF) \ 589 handle_vm86_trap(regs, 0, 1); \ 590 return; } while (0) 591 592 orig_flags = *(unsigned short *)®s->pt.flags; 593 594 csp = (unsigned char __user *) (regs->pt.cs << 4); 595 ssp = (unsigned char __user *) (regs->pt.ss << 4); 596 sp = SP(regs); 597 ip = IP(regs); 598 599 data32 = 0; 600 pref_done = 0; 601 do { 602 switch (opcode = popb(csp, ip, simulate_sigsegv)) { 603 case 0x66: /* 32-bit data */ data32 = 1; break; 604 case 0x67: /* 32-bit address */ break; 605 case 0x2e: /* CS */ break; 606 case 0x3e: /* DS */ break; 607 case 0x26: /* ES */ break; 608 case 0x36: /* SS */ break; 609 case 0x65: /* GS */ break; 610 case 0x64: /* FS */ break; 611 case 0xf2: /* repnz */ break; 612 case 0xf3: /* rep */ break; 613 default: pref_done = 1; 614 } 615 } while (!pref_done); 616 617 switch (opcode) { 618 619 /* pushf */ 620 case 0x9c: 621 if (data32) { 622 pushl(ssp, sp, get_vflags(regs), simulate_sigsegv); 623 SP(regs) -= 4; 624 } else { 625 pushw(ssp, sp, get_vflags(regs), simulate_sigsegv); 626 SP(regs) -= 2; 627 } 628 IP(regs) = ip; 629 VM86_FAULT_RETURN; 630 631 /* popf */ 632 case 0x9d: 633 { 634 unsigned long newflags; 635 if (data32) { 636 newflags = popl(ssp, sp, simulate_sigsegv); 637 SP(regs) += 4; 638 } else { 639 newflags = popw(ssp, sp, simulate_sigsegv); 640 SP(regs) += 2; 641 } 642 IP(regs) = ip; 643 CHECK_IF_IN_TRAP; 644 if (data32) 645 set_vflags_long(newflags, regs); 646 else 647 set_vflags_short(newflags, regs); 648 649 VM86_FAULT_RETURN; 650 } 651 652 /* int xx */ 653 case 0xcd: { 654 int intno = popb(csp, ip, simulate_sigsegv); 655 IP(regs) = ip; 656 if (VMPI.vm86dbg_active) { 657 if ((1 << (intno & 7)) & VMPI.vm86dbg_intxxtab[intno >> 3]) 658 return_to_32bit(regs, VM86_INTx + (intno << 8)); 659 } 660 do_int(regs, intno, ssp, sp); 661 return; 662 } 663 664 /* iret */ 665 case 0xcf: 666 { 667 unsigned long newip; 668 unsigned long newcs; 669 unsigned long newflags; 670 if (data32) { 671 newip = popl(ssp, sp, simulate_sigsegv); 672 newcs = popl(ssp, sp, simulate_sigsegv); 673 newflags = popl(ssp, sp, simulate_sigsegv); 674 SP(regs) += 12; 675 } else { 676 newip = popw(ssp, sp, simulate_sigsegv); 677 newcs = popw(ssp, sp, simulate_sigsegv); 678 newflags = popw(ssp, sp, simulate_sigsegv); 679 SP(regs) += 6; 680 } 681 IP(regs) = newip; 682 regs->pt.cs = newcs; 683 CHECK_IF_IN_TRAP; 684 if (data32) { 685 set_vflags_long(newflags, regs); 686 } else { 687 set_vflags_short(newflags, regs); 688 } 689 VM86_FAULT_RETURN; 690 } 691 692 /* cli */ 693 case 0xfa: 694 IP(regs) = ip; 695 clear_IF(regs); 696 VM86_FAULT_RETURN; 697 698 /* sti */ 699 /* 700 * Damn. This is incorrect: the 'sti' instruction should actually 701 * enable interrupts after the /next/ instruction. Not good. 702 * 703 * Probably needs some horsing around with the TF flag. Aiee.. 704 */ 705 case 0xfb: 706 IP(regs) = ip; 707 set_IF(regs); 708 VM86_FAULT_RETURN; 709 710 default: 711 return_to_32bit(regs, VM86_UNKNOWN); 712 } 713 714 return; 715 716simulate_sigsegv: 717 /* FIXME: After a long discussion with Stas we finally 718 * agreed, that this is wrong. Here we should 719 * really send a SIGSEGV to the user program. 720 * But how do we create the correct context? We 721 * are inside a general protection fault handler 722 * and has just returned from a page fault handler. 723 * The correct context for the signal handler 724 * should be a mixture of the two, but how do we 725 * get the information? [KD] 726 */ 727 return_to_32bit(regs, VM86_UNKNOWN); 728} 729 730/* ---------------- vm86 special IRQ passing stuff ----------------- */ 731 732#define VM86_IRQNAME "vm86irq" 733 734static struct vm86_irqs { 735 struct task_struct *tsk; 736 int sig; 737} vm86_irqs[16]; 738 739static DEFINE_SPINLOCK(irqbits_lock); 740static int irqbits; 741 742#define ALLOWED_SIGS (1 /* 0 = don't send a signal */ \ 743 | (1 << SIGUSR1) | (1 << SIGUSR2) | (1 << SIGIO) | (1 << SIGURG) \ 744 | (1 << SIGUNUSED)) 745 746static irqreturn_t irq_handler(int intno, void *dev_id) 747{ 748 int irq_bit; 749 unsigned long flags; 750 751 spin_lock_irqsave(&irqbits_lock, flags); 752 irq_bit = 1 << intno; 753 if ((irqbits & irq_bit) || !vm86_irqs[intno].tsk) 754 goto out; 755 irqbits |= irq_bit; 756 if (vm86_irqs[intno].sig) 757 send_sig(vm86_irqs[intno].sig, vm86_irqs[intno].tsk, 1); 758 /* 759 * IRQ will be re-enabled when user asks for the irq (whether 760 * polling or as a result of the signal) 761 */ 762 disable_irq_nosync(intno); 763 spin_unlock_irqrestore(&irqbits_lock, flags); 764 return IRQ_HANDLED; 765 766out: 767 spin_unlock_irqrestore(&irqbits_lock, flags); 768 return IRQ_NONE; 769} 770 771static inline void free_vm86_irq(int irqnumber) 772{ 773 unsigned long flags; 774 775 free_irq(irqnumber, NULL); 776 vm86_irqs[irqnumber].tsk = NULL; 777 778 spin_lock_irqsave(&irqbits_lock, flags); 779 irqbits &= ~(1 << irqnumber); 780 spin_unlock_irqrestore(&irqbits_lock, flags); 781} 782 783void release_vm86_irqs(struct task_struct *task) 784{ 785 int i; 786 for (i = FIRST_VM86_IRQ ; i <= LAST_VM86_IRQ; i++) 787 if (vm86_irqs[i].tsk == task) 788 free_vm86_irq(i); 789} 790 791static inline int get_and_reset_irq(int irqnumber) 792{ 793 int bit; 794 unsigned long flags; 795 int ret = 0; 796 797 if (invalid_vm86_irq(irqnumber)) return 0; 798 if (vm86_irqs[irqnumber].tsk != current) return 0; 799 spin_lock_irqsave(&irqbits_lock, flags); 800 bit = irqbits & (1 << irqnumber); 801 irqbits &= ~bit; 802 if (bit) { 803 enable_irq(irqnumber); 804 ret = 1; 805 } 806 807 spin_unlock_irqrestore(&irqbits_lock, flags); 808 return ret; 809} 810 811 812static int do_vm86_irq_handling(int subfunction, int irqnumber) 813{ 814 int ret; 815 switch (subfunction) { 816 case VM86_GET_AND_RESET_IRQ: { 817 return get_and_reset_irq(irqnumber); 818 } 819 case VM86_GET_IRQ_BITS: { 820 return irqbits; 821 } 822 case VM86_REQUEST_IRQ: { 823 int sig = irqnumber >> 8; 824 int irq = irqnumber & 255; 825 if (!capable(CAP_SYS_ADMIN)) return -EPERM; 826 if (!((1 << sig) & ALLOWED_SIGS)) return -EPERM; 827 if (invalid_vm86_irq(irq)) return -EPERM; 828 if (vm86_irqs[irq].tsk) return -EPERM; 829 ret = request_irq(irq, &irq_handler, 0, VM86_IRQNAME, NULL); 830 if (ret) return ret; 831 vm86_irqs[irq].sig = sig; 832 vm86_irqs[irq].tsk = current; 833 return irq; 834 } 835 case VM86_FREE_IRQ: { 836 if (invalid_vm86_irq(irqnumber)) return -EPERM; 837 if (!vm86_irqs[irqnumber].tsk) return 0; 838 if (vm86_irqs[irqnumber].tsk != current) return -EPERM; 839 free_vm86_irq(irqnumber); 840 return 0; 841 } 842 } 843 return -EINVAL; 844} 845 846