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