projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
correct email address of Manfred Spraul
[linux-2.6-block.git] / arch / i386 / kernel / vm86.c
CommitLineData
1da177e4
LT		 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
624dffcb 7 * stack - Manfred Spraul <manfred@colorfullife.com>
1da177e4
LT		 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
a9415644 33#include <linux/capability.h>
1da177e4
LT		 34#include <linux/config.h>
35#include <linux/errno.h>
36#include <linux/interrupt.h>
37#include <linux/sched.h>
38#include <linux/kernel.h>
39#include <linux/signal.h>
40#include <linux/string.h>
41#include <linux/mm.h>
42#include <linux/smp.h>
43#include <linux/smp_lock.h>
44#include <linux/highmem.h>
45#include <linux/ptrace.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)->eax))[0])
76#define AH(regs) (((unsigned char *)&((regs)->eax))[1])
77#define IP(regs) (*(unsigned short *)&((regs)->eip))
78#define SP(regs) (*(unsigned short *)&((regs)->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#define VM86_REGS_PART2 orig_eax
93#define VM86_REGS_SIZE1 \
94 ( (unsigned)( & (((struct kernel_vm86_regs *)0)->VM86_REGS_PART2) ) )
95#define VM86_REGS_SIZE2 (sizeof(struct kernel_vm86_regs) - VM86_REGS_SIZE1)
96
97struct pt_regs * FASTCALL(save_v86_state(struct kernel_vm86_regs * regs));
98struct pt_regs * fastcall save_v86_state(struct kernel_vm86_regs * regs)
99{
100 struct tss_struct *tss;
101 struct pt_regs *ret;
102 unsigned long tmp;
103
104 /*
105 * This gets called from entry.S with interrupts disabled, but
106 * from process context. Enable interrupts here, before trying
107 * to access user space.
108 */
109 local_irq_enable();
110
111 if (!current->thread.vm86_info) {
112 printk("no vm86_info: BAD\n");
113 do_exit(SIGSEGV);
114 }
115 set_flags(regs->eflags, VEFLAGS, VIF_MASK | current->thread.v86mask);
116 tmp = copy_to_user(&current->thread.vm86_info->regs,regs, VM86_REGS_SIZE1);
117 tmp += copy_to_user(&current->thread.vm86_info->regs.VM86_REGS_PART2,
118 &regs->VM86_REGS_PART2, VM86_REGS_SIZE2);
119 tmp += put_user(current->thread.screen_bitmap,&current->thread.vm86_info->screen_bitmap);
120 if (tmp) {
121 printk("vm86: could not access userspace vm86_info\n");
122 do_exit(SIGSEGV);
123 }
124
125 tss = &per_cpu(init_tss, get_cpu());
126 current->thread.esp0 = current->thread.saved_esp0;
127 current->thread.sysenter_cs = __KERNEL_CS;
128 load_esp0(tss, &current->thread);
129 current->thread.saved_esp0 = 0;
130 put_cpu();
131
132 loadsegment(fs, current->thread.saved_fs);
133 loadsegment(gs, current->thread.saved_gs);
134 ret = KVM86->regs32;
135 return ret;
136}
137
60ec5585 138static void mark_screen_rdonly(struct mm_struct *mm)
1da177e4
LT		 139{
140 pgd_t *pgd;
141 pud_t *pud;
142 pmd_t *pmd;
60ec5585
HD		 143 pte_t *pte;
144 spinlock_t *ptl;
1da177e4
LT		 145 int i;
146
60ec5585 147 pgd = pgd_offset(mm, 0xA0000);
1da177e4
LT		 148 if (pgd_none_or_clear_bad(pgd))
149 goto out;
150 pud = pud_offset(pgd, 0xA0000);
151 if (pud_none_or_clear_bad(pud))
152 goto out;
153 pmd = pmd_offset(pud, 0xA0000);
154 if (pmd_none_or_clear_bad(pmd))
155 goto out;
60ec5585 156 pte = pte_offset_map_lock(mm, pmd, 0xA0000, &ptl);
1da177e4
LT		 157 for (i = 0; i < 32; i++) {
158 if (pte_present(*pte))
159 set_pte(pte, pte_wrprotect(*pte));
160 pte++;
161 }
60ec5585 162 pte_unmap_unlock(pte, ptl);
1da177e4 163out:
1da177e4
LT		 164 flush_tlb();
165}
166
167
168
169static int do_vm86_irq_handling(int subfunction, int irqnumber);
170static void do_sys_vm86(struct kernel_vm86_struct *info, struct task_struct *tsk);
171
172asmlinkage int sys_vm86old(struct pt_regs regs)
173{
174 struct vm86_struct __user *v86 = (struct vm86_struct __user *)regs.ebx;
175 struct kernel_vm86_struct info; /* declare this _on top_,
176 * this avoids wasting of stack space.
177 * This remains on the stack until we
178 * return to 32 bit user space.
179 */
180 struct task_struct *tsk;
181 int tmp, ret = -EPERM;
182
183 tsk = current;
184 if (tsk->thread.saved_esp0)
185 goto out;
186 tmp = copy_from_user(&info, v86, VM86_REGS_SIZE1);
187 tmp += copy_from_user(&info.regs.VM86_REGS_PART2, &v86->regs.VM86_REGS_PART2,
188 (long)&info.vm86plus - (long)&info.regs.VM86_REGS_PART2);
189 ret = -EFAULT;
190 if (tmp)
191 goto out;
192 memset(&info.vm86plus, 0, (int)&info.regs32 - (int)&info.vm86plus);
193 info.regs32 = &regs;
194 tsk->thread.vm86_info = v86;
195 do_sys_vm86(&info, tsk);
196 ret = 0; /* we never return here */
197out:
198 return ret;
199}
200
201
202asmlinkage int sys_vm86(struct pt_regs regs)
203{
204 struct kernel_vm86_struct info; /* declare this _on top_,
205 * this avoids wasting of stack space.
206 * This remains on the stack until we
207 * return to 32 bit user space.
208 */
209 struct task_struct *tsk;
210 int tmp, ret;
211 struct vm86plus_struct __user *v86;
212
213 tsk = current;
214 switch (regs.ebx) {
215 case VM86_REQUEST_IRQ:
216 case VM86_FREE_IRQ:
217 case VM86_GET_IRQ_BITS:
218 case VM86_GET_AND_RESET_IRQ:
219 ret = do_vm86_irq_handling(regs.ebx, (int)regs.ecx);
220 goto out;
221 case VM86_PLUS_INSTALL_CHECK:
222 /* NOTE: on old vm86 stuff this will return the error
e49332bd 223 from access_ok(), because the subfunction is
1da177e4
LT		 224 interpreted as (invalid) address to vm86_struct.
225 So the installation check works.
226 */
227 ret = 0;
228 goto out;
229 }
230
231 /* we come here only for functions VM86_ENTER, VM86_ENTER_NO_BYPASS */
232 ret = -EPERM;
233 if (tsk->thread.saved_esp0)
234 goto out;
235 v86 = (struct vm86plus_struct __user *)regs.ecx;
236 tmp = copy_from_user(&info, v86, VM86_REGS_SIZE1);
237 tmp += copy_from_user(&info.regs.VM86_REGS_PART2, &v86->regs.VM86_REGS_PART2,
238 (long)&info.regs32 - (long)&info.regs.VM86_REGS_PART2);
239 ret = -EFAULT;
240 if (tmp)
241 goto out;
242 info.regs32 = &regs;
243 info.vm86plus.is_vm86pus = 1;
244 tsk->thread.vm86_info = (struct vm86_struct __user *)v86;
245 do_sys_vm86(&info, tsk);
246 ret = 0; /* we never return here */
247out:
248 return ret;
249}
250
251
252static void do_sys_vm86(struct kernel_vm86_struct *info, struct task_struct *tsk)
253{
254 struct tss_struct *tss;
255/*
256 * make sure the vm86() system call doesn't try to do anything silly
257 */
258 info->regs.__null_ds = 0;
259 info->regs.__null_es = 0;
260
261/* we are clearing fs,gs later just before "jmp resume_userspace",
262 * because starting with Linux 2.1.x they aren't no longer saved/restored
263 */
264
265/*
266 * The eflags register is also special: we cannot trust that the user
267 * has set it up safely, so this makes sure interrupt etc flags are
268 * inherited from protected mode.
269 */
270 VEFLAGS = info->regs.eflags;
271 info->regs.eflags &= SAFE_MASK;
272 info->regs.eflags |= info->regs32->eflags & ~SAFE_MASK;
273 info->regs.eflags |= VM_MASK;
274
275 switch (info->cpu_type) {
276 case CPU_286:
277 tsk->thread.v86mask = 0;
278 break;
279 case CPU_386:
280 tsk->thread.v86mask = NT_MASK | IOPL_MASK;
281 break;
282 case CPU_486:
283 tsk->thread.v86mask = AC_MASK | NT_MASK | IOPL_MASK;
284 break;
285 default:
286 tsk->thread.v86mask = ID_MASK | AC_MASK | NT_MASK | IOPL_MASK;
287 break;
288 }
289
290/*
291 * Save old state, set default return value (%eax) to 0
292 */
293 info->regs32->eax = 0;
294 tsk->thread.saved_esp0 = tsk->thread.esp0;
4d37e7e3
ZA		 295 savesegment(fs, tsk->thread.saved_fs);
296 savesegment(gs, tsk->thread.saved_gs);
1da177e4
LT		 297
298 tss = &per_cpu(init_tss, get_cpu());
299 tsk->thread.esp0 = (unsigned long) &info->VM86_TSS_ESP0;
300 if (cpu_has_sep)
301 tsk->thread.sysenter_cs = 0;
302 load_esp0(tss, &tsk->thread);
303 put_cpu();
304
305 tsk->thread.screen_bitmap = info->screen_bitmap;
306 if (info->flags & VM86_SCREEN_BITMAP)
60ec5585 307 mark_screen_rdonly(tsk->mm);
1da177e4
LT		 308 __asm__ __volatile__(
309 "xorl %%eax,%%eax; movl %%eax,%%fs; movl %%eax,%%gs\n\t"
310 "movl %0,%%esp\n\t"
311 "movl %1,%%ebp\n\t"
312 "jmp resume_userspace"
313 : /* no outputs */
06b425d8 314 :"r" (&info->regs), "r" (task_thread_info(tsk)) : "ax");
1da177e4
LT		 315 /* we never return here */
316}
317
318static inline void return_to_32bit(struct kernel_vm86_regs * regs16, int retval)
319{
320 struct pt_regs * regs32;
321
322 regs32 = save_v86_state(regs16);
323 regs32->eax = retval;
324 __asm__ __volatile__("movl %0,%%esp\n\t"
325 "movl %1,%%ebp\n\t"
326 "jmp resume_userspace"
327 : : "r" (regs32), "r" (current_thread_info()));
328}
329
330static inline void set_IF(struct kernel_vm86_regs * regs)
331{
332 VEFLAGS |= VIF_MASK;
333 if (VEFLAGS & VIP_MASK)
334 return_to_32bit(regs, VM86_STI);
335}
336
337static inline void clear_IF(struct kernel_vm86_regs * regs)
338{
339 VEFLAGS &= ~VIF_MASK;
340}
341
342static inline void clear_TF(struct kernel_vm86_regs * regs)
343{
344 regs->eflags &= ~TF_MASK;
345}
346
347static inline void clear_AC(struct kernel_vm86_regs * regs)
348{
349 regs->eflags &= ~AC_MASK;
350}
351
352/* It is correct to call set_IF(regs) from the set_vflags_*
353 * functions. However someone forgot to call clear_IF(regs)
354 * in the opposite case.
355 * After the command sequence CLI PUSHF STI POPF you should
356 * end up with interrups disabled, but you ended up with
357 * interrupts enabled.
358 * ( I was testing my own changes, but the only bug I
359 * could find was in a function I had not changed. )
360 * [KD]
361 */
362
363static inline void set_vflags_long(unsigned long eflags, struct kernel_vm86_regs * regs)
364{
365 set_flags(VEFLAGS, eflags, current->thread.v86mask);
366 set_flags(regs->eflags, eflags, SAFE_MASK);
367 if (eflags & IF_MASK)
368 set_IF(regs);
369 else
370 clear_IF(regs);
371}
372
373static inline void set_vflags_short(unsigned short flags, struct kernel_vm86_regs * regs)
374{
375 set_flags(VFLAGS, flags, current->thread.v86mask);
376 set_flags(regs->eflags, flags, SAFE_MASK);
377 if (flags & IF_MASK)
378 set_IF(regs);
379 else
380 clear_IF(regs);
381}
382
383static inline unsigned long get_vflags(struct kernel_vm86_regs * regs)
384{
385 unsigned long flags = regs->eflags & RETURN_MASK;
386
387 if (VEFLAGS & VIF_MASK)
388 flags |= IF_MASK;
389 flags |= IOPL_MASK;
390 return flags | (VEFLAGS & current->thread.v86mask);
391}
392
393static inline int is_revectored(int nr, struct revectored_struct * bitmap)
394{
395 __asm__ __volatile__("btl %2,%1\n\tsbbl %0,%0"
396 :"=r" (nr)
397 :"m" (*bitmap),"r" (nr));
398 return nr;
399}
400
401#define val_byte(val, n) (((__u8 *)&val)[n])
402
403#define pushb(base, ptr, val, err_label) \
404 do { \
405 __u8 __val = val; \
406 ptr--; \
407 if (put_user(__val, base + ptr) < 0) \
408 goto err_label; \
409 } while(0)
410
411#define pushw(base, ptr, val, err_label) \
412 do { \
413 __u16 __val = val; \
414 ptr--; \
415 if (put_user(val_byte(__val, 1), base + ptr) < 0) \
416 goto err_label; \
417 ptr--; \
418 if (put_user(val_byte(__val, 0), base + ptr) < 0) \
419 goto err_label; \
420 } while(0)
421
422#define pushl(base, ptr, val, err_label) \
423 do { \
424 __u32 __val = val; \
425 ptr--; \
426 if (put_user(val_byte(__val, 3), base + ptr) < 0) \
427 goto err_label; \
428 ptr--; \
429 if (put_user(val_byte(__val, 2), base + ptr) < 0) \
430 goto err_label; \
431 ptr--; \
432 if (put_user(val_byte(__val, 1), base + ptr) < 0) \
433 goto err_label; \
434 ptr--; \
435 if (put_user(val_byte(__val, 0), base + ptr) < 0) \
436 goto err_label; \
437 } while(0)
438
439#define popb(base, ptr, err_label) \
440 ({ \
441 __u8 __res; \
442 if (get_user(__res, base + ptr) < 0) \
443 goto err_label; \
444 ptr++; \
445 __res; \
446 })
447
448#define popw(base, ptr, err_label) \
449 ({ \
450 __u16 __res; \
451 if (get_user(val_byte(__res, 0), base + ptr) < 0) \
452 goto err_label; \
453 ptr++; \
454 if (get_user(val_byte(__res, 1), base + ptr) < 0) \
455 goto err_label; \
456 ptr++; \
457 __res; \
458 })
459
460#define popl(base, ptr, err_label) \
461 ({ \
462 __u32 __res; \
463 if (get_user(val_byte(__res, 0), base + ptr) < 0) \
464 goto err_label; \
465 ptr++; \
466 if (get_user(val_byte(__res, 1), base + ptr) < 0) \
467 goto err_label; \
468 ptr++; \
469 if (get_user(val_byte(__res, 2), base + ptr) < 0) \
470 goto err_label; \
471 ptr++; \
472 if (get_user(val_byte(__res, 3), base + ptr) < 0) \
473 goto err_label; \
474 ptr++; \
475 __res; \
476 })
477
478/* There are so many possible reasons for this function to return
479 * VM86_INTx, so adding another doesn't bother me. We can expect
480 * userspace programs to be able to handle it. (Getting a problem
481 * in userspace is always better than an Oops anyway.) [KD]
482 */
483static void do_int(struct kernel_vm86_regs *regs, int i,
484 unsigned char __user * ssp, unsigned short sp)
485{
486 unsigned long __user *intr_ptr;
487 unsigned long segoffs;
488
489 if (regs->cs == BIOSSEG)
490 goto cannot_handle;
491 if (is_revectored(i, &KVM86->int_revectored))
492 goto cannot_handle;
493 if (i==0x21 && is_revectored(AH(regs),&KVM86->int21_revectored))
494 goto cannot_handle;
495 intr_ptr = (unsigned long __user *) (i << 2);
496 if (get_user(segoffs, intr_ptr))
497 goto cannot_handle;
498 if ((segoffs >> 16) == BIOSSEG)
499 goto cannot_handle;
500 pushw(ssp, sp, get_vflags(regs), cannot_handle);
501 pushw(ssp, sp, regs->cs, cannot_handle);
502 pushw(ssp, sp, IP(regs), cannot_handle);
503 regs->cs = segoffs >> 16;
504 SP(regs) -= 6;
505 IP(regs) = segoffs & 0xffff;
506 clear_TF(regs);
507 clear_IF(regs);
508 clear_AC(regs);
509 return;
510
511cannot_handle:
512 return_to_32bit(regs, VM86_INTx + (i << 8));
513}
514
515int handle_vm86_trap(struct kernel_vm86_regs * regs, long error_code, int trapno)
516{
517 if (VMPI.is_vm86pus) {
518 if ( (trapno==3) || (trapno==1) )
519 return_to_32bit(regs, VM86_TRAP + (trapno << 8));
520 do_int(regs, trapno, (unsigned char __user *) (regs->ss << 4), SP(regs));
521 return 0;
522 }
523 if (trapno !=1)
524 return 1; /* we let this handle by the calling routine */
525 if (current->ptrace & PT_PTRACED) {
526 unsigned long flags;
527 spin_lock_irqsave(&current->sighand->siglock, flags);
528 sigdelset(&current->blocked, SIGTRAP);
529 recalc_sigpending();
530 spin_unlock_irqrestore(&current->sighand->siglock, flags);
531 }
532 send_sig(SIGTRAP, current, 1);
533 current->thread.trap_no = trapno;
534 current->thread.error_code = error_code;
535 return 0;
536}
537
538void handle_vm86_fault(struct kernel_vm86_regs * regs, long error_code)
539{
540 unsigned char opcode;
541 unsigned char __user *csp;
542 unsigned char __user *ssp;
5fd75ebb 543 unsigned short ip, sp, orig_flags;
1da177e4
LT		 544 int data32, pref_done;
545
546#define CHECK_IF_IN_TRAP \
547 if (VMPI.vm86dbg_active && VMPI.vm86dbg_TFpendig) \
548 newflags |= TF_MASK
549#define VM86_FAULT_RETURN do { \
550 if (VMPI.force_return_for_pic && (VEFLAGS & (IF_MASK | VIF_MASK))) \
551 return_to_32bit(regs, VM86_PICRETURN); \
5fd75ebb
PT		 552 if (orig_flags & TF_MASK) \
553 handle_vm86_trap(regs, 0, 1); \
1da177e4
LT		 554 return; } while (0)
555
5fd75ebb
PT		 556 orig_flags = *(unsigned short *)&regs->eflags;
557
1da177e4
LT		 558 csp = (unsigned char __user *) (regs->cs << 4);
559 ssp = (unsigned char __user *) (regs->ss << 4);
560 sp = SP(regs);
561 ip = IP(regs);
562
563 data32 = 0;
564 pref_done = 0;
565 do {
566 switch (opcode = popb(csp, ip, simulate_sigsegv)) {
567 case 0x66: /* 32-bit data */ data32=1; break;
568 case 0x67: /* 32-bit address */ break;
569 case 0x2e: /* CS */ break;
570 case 0x3e: /* DS */ break;
571 case 0x26: /* ES */ break;
572 case 0x36: /* SS */ break;
573 case 0x65: /* GS */ break;
574 case 0x64: /* FS */ break;
575 case 0xf2: /* repnz */ break;
576 case 0xf3: /* rep */ break;
577 default: pref_done = 1;
578 }
579 } while (!pref_done);
580
581 switch (opcode) {
582
583 /* pushf */
584 case 0x9c:
585 if (data32) {
586 pushl(ssp, sp, get_vflags(regs), simulate_sigsegv);
587 SP(regs) -= 4;
588 } else {
589 pushw(ssp, sp, get_vflags(regs), simulate_sigsegv);
590 SP(regs) -= 2;
591 }
592 IP(regs) = ip;
593 VM86_FAULT_RETURN;
594
595 /* popf */
596 case 0x9d:
597 {
598 unsigned long newflags;
599 if (data32) {
600 newflags=popl(ssp, sp, simulate_sigsegv);
601 SP(regs) += 4;
602 } else {
603 newflags = popw(ssp, sp, simulate_sigsegv);
604 SP(regs) += 2;
605 }
606 IP(regs) = ip;
607 CHECK_IF_IN_TRAP;
608 if (data32) {
609 set_vflags_long(newflags, regs);
610 } else {
611 set_vflags_short(newflags, regs);
612 }
613 VM86_FAULT_RETURN;
614 }
615
616 /* int xx */
617 case 0xcd: {
618 int intno=popb(csp, ip, simulate_sigsegv);
619 IP(regs) = ip;
620 if (VMPI.vm86dbg_active) {
621 if ( (1 << (intno &7)) & VMPI.vm86dbg_intxxtab[intno >> 3] )
622 return_to_32bit(regs, VM86_INTx + (intno << 8));
623 }
624 do_int(regs, intno, ssp, sp);
625 return;
626 }
627
628 /* iret */
629 case 0xcf:
630 {
631 unsigned long newip;
632 unsigned long newcs;
633 unsigned long newflags;
634 if (data32) {
635 newip=popl(ssp, sp, simulate_sigsegv);
636 newcs=popl(ssp, sp, simulate_sigsegv);
637 newflags=popl(ssp, sp, simulate_sigsegv);
638 SP(regs) += 12;
639 } else {
640 newip = popw(ssp, sp, simulate_sigsegv);
641 newcs = popw(ssp, sp, simulate_sigsegv);
642 newflags = popw(ssp, sp, simulate_sigsegv);
643 SP(regs) += 6;
644 }
645 IP(regs) = newip;
646 regs->cs = newcs;
647 CHECK_IF_IN_TRAP;
648 if (data32) {
649 set_vflags_long(newflags, regs);
650 } else {
651 set_vflags_short(newflags, regs);
652 }
653 VM86_FAULT_RETURN;
654 }
655
656 /* cli */
657 case 0xfa:
658 IP(regs) = ip;
659 clear_IF(regs);
660 VM86_FAULT_RETURN;
661
662 /* sti */
663 /*
664 * Damn. This is incorrect: the 'sti' instruction should actually
665 * enable interrupts after the /next/ instruction. Not good.
666 *
667 * Probably needs some horsing around with the TF flag. Aiee..
668 */
669 case 0xfb:
670 IP(regs) = ip;
671 set_IF(regs);
672 VM86_FAULT_RETURN;
673
674 default:
675 return_to_32bit(regs, VM86_UNKNOWN);
676 }
677
678 return;
679
680simulate_sigsegv:
681 /* FIXME: After a long discussion with Stas we finally
682 * agreed, that this is wrong. Here we should
683 * really send a SIGSEGV to the user program.
684 * But how do we create the correct context? We
685 * are inside a general protection fault handler
686 * and has just returned from a page fault handler.
687 * The correct context for the signal handler
688 * should be a mixture of the two, but how do we
689 * get the information? [KD]
690 */
691 return_to_32bit(regs, VM86_UNKNOWN);
692}
693
694/* ---------------- vm86 special IRQ passing stuff ----------------- */
695
696#define VM86_IRQNAME "vm86irq"
697
698static struct vm86_irqs {
699 struct task_struct *tsk;
700 int sig;
701} vm86_irqs[16];
702
703static DEFINE_SPINLOCK(irqbits_lock);
704static int irqbits;
705
706#define ALLOWED_SIGS ( 1 /* 0 = don't send a signal */ \
707 | (1 << SIGUSR1) | (1 << SIGUSR2) | (1 << SIGIO) | (1 << SIGURG) \
708 | (1 << SIGUNUSED) )
709
710static irqreturn_t irq_handler(int intno, void *dev_id, struct pt_regs * regs)
711{
712 int irq_bit;
713 unsigned long flags;
714
715 spin_lock_irqsave(&irqbits_lock, flags);
716 irq_bit = 1 << intno;
717 if ((irqbits & irq_bit) || ! vm86_irqs[intno].tsk)
718 goto out;
719 irqbits |= irq_bit;
720 if (vm86_irqs[intno].sig)
721 send_sig(vm86_irqs[intno].sig, vm86_irqs[intno].tsk, 1);
1da177e4
LT		 722 /*
723 * IRQ will be re-enabled when user asks for the irq (whether
724 * polling or as a result of the signal)
725 */
ad671423
PP		 726 disable_irq_nosync(intno);
727 spin_unlock_irqrestore(&irqbits_lock, flags);
1da177e4
LT		 728 return IRQ_HANDLED;
729
730out:
731 spin_unlock_irqrestore(&irqbits_lock, flags);
732 return IRQ_NONE;
733}
734
735static inline void free_vm86_irq(int irqnumber)
736{
737 unsigned long flags;
738
739 free_irq(irqnumber, NULL);
740 vm86_irqs[irqnumber].tsk = NULL;
741
742 spin_lock_irqsave(&irqbits_lock, flags);
743 irqbits &= ~(1 << irqnumber);
744 spin_unlock_irqrestore(&irqbits_lock, flags);
745}
746
747void release_vm86_irqs(struct task_struct *task)
748{
749 int i;
750 for (i = FIRST_VM86_IRQ ; i <= LAST_VM86_IRQ; i++)
751 if (vm86_irqs[i].tsk == task)
752 free_vm86_irq(i);
753}
754
755static inline int get_and_reset_irq(int irqnumber)
756{
757 int bit;
758 unsigned long flags;
ad671423 759 int ret = 0;
1da177e4
LT		 760
761 if (invalid_vm86_irq(irqnumber)) return 0;
762 if (vm86_irqs[irqnumber].tsk != current) return 0;
763 spin_lock_irqsave(&irqbits_lock, flags);
764 bit = irqbits & (1 << irqnumber);
765 irqbits &= ~bit;
ad671423
PP		 766 if (bit) {
767 enable_irq(irqnumber);
768 ret = 1;
769 }
770
1da177e4 771 spin_unlock_irqrestore(&irqbits_lock, flags);
ad671423 772 return ret;
1da177e4
LT		 773}
774
775
776static int do_vm86_irq_handling(int subfunction, int irqnumber)
777{
778 int ret;
779 switch (subfunction) {
780 case VM86_GET_AND_RESET_IRQ: {
781 return get_and_reset_irq(irqnumber);
782 }
783 case VM86_GET_IRQ_BITS: {
784 return irqbits;
785 }
786 case VM86_REQUEST_IRQ: {
787 int sig = irqnumber >> 8;
788 int irq = irqnumber & 255;
789 if (!capable(CAP_SYS_ADMIN)) return -EPERM;
790 if (!((1 << sig) & ALLOWED_SIGS)) return -EPERM;
791 if (invalid_vm86_irq(irq)) return -EPERM;
792 if (vm86_irqs[irq].tsk) return -EPERM;
793 ret = request_irq(irq, &irq_handler, 0, VM86_IRQNAME, NULL);
794 if (ret) return ret;
795 vm86_irqs[irq].sig = sig;
796 vm86_irqs[irq].tsk = current;
797 return irq;
798 }
799 case VM86_FREE_IRQ: {
800 if (invalid_vm86_irq(irqnumber)) return -EPERM;
801 if (!vm86_irqs[irqnumber].tsk) return 0;
802 if (vm86_irqs[irqnumber].tsk != current) return -EPERM;
803 free_vm86_irq(irqnumber);
804 return 0;
805 }
806 }
807 return -EINVAL;
808}
809
Linux 6.x block layer and io_uring tree(s)