Commit | Line | Data |
---|---|---|
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/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> | |
7e7f8a03 | 45 | #include <linux/audit.h> |
49d26b6e | 46 | #include <linux/stddef.h> |
1da177e4 LT |
47 | |
48 | #include <asm/uaccess.h> | |
49 | #include <asm/io.h> | |
50 | #include <asm/tlbflush.h> | |
51 | #include <asm/irq.h> | |
52 | ||
53 | /* | |
54 | * Known problems: | |
55 | * | |
56 | * Interrupt handling is not guaranteed: | |
57 | * - a real x86 will disable all interrupts for one instruction | |
58 | * after a "mov ss,xx" to make stack handling atomic even without | |
59 | * the 'lss' instruction. We can't guarantee this in v86 mode, | |
60 | * as the next instruction might result in a page fault or similar. | |
61 | * - a real x86 will have interrupts disabled for one instruction | |
62 | * past the 'sti' that enables them. We don't bother with all the | |
63 | * details yet. | |
64 | * | |
65 | * Let's hope these problems do not actually matter for anything. | |
66 | */ | |
67 | ||
68 | ||
69 | #define KVM86 ((struct kernel_vm86_struct *)regs) | |
70 | #define VMPI KVM86->vm86plus | |
71 | ||
72 | ||
73 | /* | |
74 | * 8- and 16-bit register defines.. | |
75 | */ | |
49d26b6e JF |
76 | #define AL(regs) (((unsigned char *)&((regs)->pt.eax))[0]) |
77 | #define AH(regs) (((unsigned char *)&((regs)->pt.eax))[1]) | |
78 | #define IP(regs) (*(unsigned short *)&((regs)->pt.eip)) | |
79 | #define SP(regs) (*(unsigned short *)&((regs)->pt.esp)) | |
1da177e4 LT |
80 | |
81 | /* | |
82 | * virtual flags (16 and 32-bit versions) | |
83 | */ | |
84 | #define VFLAGS (*(unsigned short *)&(current->thread.v86flags)) | |
85 | #define VEFLAGS (current->thread.v86flags) | |
86 | ||
87 | #define set_flags(X,new,mask) \ | |
88 | ((X) = ((X) & ~(mask)) | ((new) & (mask))) | |
89 | ||
90 | #define SAFE_MASK (0xDD5) | |
91 | #define RETURN_MASK (0xDFF) | |
92 | ||
49d26b6e JF |
93 | /* convert kernel_vm86_regs to vm86_regs */ |
94 | static int copy_vm86_regs_to_user(struct vm86_regs __user *user, | |
95 | const struct kernel_vm86_regs *regs) | |
96 | { | |
97 | int ret = 0; | |
98 | ||
464d1a78 JF |
99 | /* kernel_vm86_regs is missing xgs, so copy everything up to |
100 | (but not including) orig_eax, and then rest including orig_eax. */ | |
101 | ret += copy_to_user(user, regs, offsetof(struct kernel_vm86_regs, pt.orig_eax)); | |
102 | ret += copy_to_user(&user->orig_eax, ®s->pt.orig_eax, | |
49d26b6e | 103 | sizeof(struct kernel_vm86_regs) - |
464d1a78 | 104 | offsetof(struct kernel_vm86_regs, pt.orig_eax)); |
49d26b6e JF |
105 | |
106 | return ret; | |
107 | } | |
108 | ||
109 | /* convert vm86_regs to kernel_vm86_regs */ | |
110 | static 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 | ||
464d1a78 JF |
116 | /* copy eax-xfs inclusive */ |
117 | ret += copy_from_user(regs, user, offsetof(struct kernel_vm86_regs, pt.orig_eax)); | |
118 | /* copy orig_eax-__gsh+extra */ | |
119 | ret += copy_from_user(®s->pt.orig_eax, &user->orig_eax, | |
49d26b6e | 120 | sizeof(struct kernel_vm86_regs) - |
464d1a78 | 121 | offsetof(struct kernel_vm86_regs, pt.orig_eax) + |
49d26b6e | 122 | extra); |
49d26b6e JF |
123 | return ret; |
124 | } | |
1da177e4 LT |
125 | |
126 | struct pt_regs * FASTCALL(save_v86_state(struct kernel_vm86_regs * regs)); | |
127 | struct pt_regs * fastcall save_v86_state(struct kernel_vm86_regs * regs) | |
128 | { | |
129 | struct tss_struct *tss; | |
130 | struct pt_regs *ret; | |
131 | unsigned long tmp; | |
132 | ||
133 | /* | |
134 | * This gets called from entry.S with interrupts disabled, but | |
135 | * from process context. Enable interrupts here, before trying | |
136 | * to access user space. | |
137 | */ | |
138 | local_irq_enable(); | |
139 | ||
140 | if (!current->thread.vm86_info) { | |
141 | printk("no vm86_info: BAD\n"); | |
142 | do_exit(SIGSEGV); | |
143 | } | |
49d26b6e JF |
144 | set_flags(regs->pt.eflags, VEFLAGS, VIF_MASK | current->thread.v86mask); |
145 | tmp = copy_vm86_regs_to_user(¤t->thread.vm86_info->regs,regs); | |
1da177e4 LT |
146 | tmp += put_user(current->thread.screen_bitmap,¤t->thread.vm86_info->screen_bitmap); |
147 | if (tmp) { | |
148 | printk("vm86: could not access userspace vm86_info\n"); | |
149 | do_exit(SIGSEGV); | |
150 | } | |
151 | ||
152 | tss = &per_cpu(init_tss, get_cpu()); | |
153 | current->thread.esp0 = current->thread.saved_esp0; | |
154 | current->thread.sysenter_cs = __KERNEL_CS; | |
155 | load_esp0(tss, ¤t->thread); | |
156 | current->thread.saved_esp0 = 0; | |
157 | put_cpu(); | |
158 | ||
1da177e4 | 159 | ret = KVM86->regs32; |
49d26b6e | 160 | |
464d1a78 JF |
161 | ret->xfs = current->thread.saved_fs; |
162 | loadsegment(gs, current->thread.saved_gs); | |
49d26b6e | 163 | |
1da177e4 LT |
164 | return ret; |
165 | } | |
166 | ||
60ec5585 | 167 | static void mark_screen_rdonly(struct mm_struct *mm) |
1da177e4 LT |
168 | { |
169 | pgd_t *pgd; | |
170 | pud_t *pud; | |
171 | pmd_t *pmd; | |
60ec5585 HD |
172 | pte_t *pte; |
173 | spinlock_t *ptl; | |
1da177e4 LT |
174 | int i; |
175 | ||
60ec5585 | 176 | pgd = pgd_offset(mm, 0xA0000); |
1da177e4 LT |
177 | if (pgd_none_or_clear_bad(pgd)) |
178 | goto out; | |
179 | pud = pud_offset(pgd, 0xA0000); | |
180 | if (pud_none_or_clear_bad(pud)) | |
181 | goto out; | |
182 | pmd = pmd_offset(pud, 0xA0000); | |
183 | if (pmd_none_or_clear_bad(pmd)) | |
184 | goto out; | |
60ec5585 | 185 | pte = pte_offset_map_lock(mm, pmd, 0xA0000, &ptl); |
1da177e4 LT |
186 | for (i = 0; i < 32; i++) { |
187 | if (pte_present(*pte)) | |
188 | set_pte(pte, pte_wrprotect(*pte)); | |
189 | pte++; | |
190 | } | |
60ec5585 | 191 | pte_unmap_unlock(pte, ptl); |
1da177e4 | 192 | out: |
1da177e4 LT |
193 | flush_tlb(); |
194 | } | |
195 | ||
196 | ||
197 | ||
198 | static int do_vm86_irq_handling(int subfunction, int irqnumber); | |
199 | static void do_sys_vm86(struct kernel_vm86_struct *info, struct task_struct *tsk); | |
200 | ||
201 | asmlinkage int sys_vm86old(struct pt_regs regs) | |
202 | { | |
203 | struct vm86_struct __user *v86 = (struct vm86_struct __user *)regs.ebx; | |
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 = -EPERM; | |
211 | ||
212 | tsk = current; | |
213 | if (tsk->thread.saved_esp0) | |
214 | goto out; | |
49d26b6e JF |
215 | tmp = copy_vm86_regs_from_user(&info.regs, &v86->regs, |
216 | offsetof(struct kernel_vm86_struct, vm86plus) - | |
217 | sizeof(info.regs)); | |
1da177e4 LT |
218 | ret = -EFAULT; |
219 | if (tmp) | |
220 | goto out; | |
221 | memset(&info.vm86plus, 0, (int)&info.regs32 - (int)&info.vm86plus); | |
222 | info.regs32 = ®s; | |
223 | tsk->thread.vm86_info = v86; | |
224 | do_sys_vm86(&info, tsk); | |
225 | ret = 0; /* we never return here */ | |
226 | out: | |
227 | return ret; | |
228 | } | |
229 | ||
230 | ||
231 | asmlinkage int sys_vm86(struct pt_regs regs) | |
232 | { | |
233 | struct kernel_vm86_struct info; /* declare this _on top_, | |
234 | * this avoids wasting of stack space. | |
235 | * This remains on the stack until we | |
236 | * return to 32 bit user space. | |
237 | */ | |
238 | struct task_struct *tsk; | |
239 | int tmp, ret; | |
240 | struct vm86plus_struct __user *v86; | |
241 | ||
242 | tsk = current; | |
243 | switch (regs.ebx) { | |
244 | case VM86_REQUEST_IRQ: | |
245 | case VM86_FREE_IRQ: | |
246 | case VM86_GET_IRQ_BITS: | |
247 | case VM86_GET_AND_RESET_IRQ: | |
248 | ret = do_vm86_irq_handling(regs.ebx, (int)regs.ecx); | |
249 | goto out; | |
250 | case VM86_PLUS_INSTALL_CHECK: | |
251 | /* NOTE: on old vm86 stuff this will return the error | |
e49332bd | 252 | from access_ok(), because the subfunction is |
1da177e4 LT |
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_esp0) | |
263 | goto out; | |
264 | v86 = (struct vm86plus_struct __user *)regs.ecx; | |
49d26b6e JF |
265 | tmp = copy_vm86_regs_from_user(&info.regs, &v86->regs, |
266 | offsetof(struct kernel_vm86_struct, regs32) - | |
267 | sizeof(info.regs)); | |
1da177e4 LT |
268 | ret = -EFAULT; |
269 | if (tmp) | |
270 | goto out; | |
271 | info.regs32 = ®s; | |
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 */ | |
276 | out: | |
277 | return ret; | |
278 | } | |
279 | ||
280 | ||
281 | static 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 | */ | |
49d26b6e JF |
287 | info->regs.pt.xds = 0; |
288 | info->regs.pt.xes = 0; | |
464d1a78 | 289 | info->regs.pt.xfs = 0; |
1da177e4 | 290 | |
464d1a78 | 291 | /* we are clearing gs later just before "jmp resume_userspace", |
49d26b6e | 292 | * because it is not saved/restored. |
1da177e4 LT |
293 | */ |
294 | ||
295 | /* | |
296 | * The eflags register is also special: we cannot trust that the user | |
297 | * has set it up safely, so this makes sure interrupt etc flags are | |
298 | * inherited from protected mode. | |
299 | */ | |
49d26b6e JF |
300 | VEFLAGS = info->regs.pt.eflags; |
301 | info->regs.pt.eflags &= SAFE_MASK; | |
302 | info->regs.pt.eflags |= info->regs32->eflags & ~SAFE_MASK; | |
303 | info->regs.pt.eflags |= VM_MASK; | |
1da177e4 LT |
304 | |
305 | switch (info->cpu_type) { | |
306 | case CPU_286: | |
307 | tsk->thread.v86mask = 0; | |
308 | break; | |
309 | case CPU_386: | |
310 | tsk->thread.v86mask = NT_MASK | IOPL_MASK; | |
311 | break; | |
312 | case CPU_486: | |
313 | tsk->thread.v86mask = AC_MASK | NT_MASK | IOPL_MASK; | |
314 | break; | |
315 | default: | |
316 | tsk->thread.v86mask = ID_MASK | AC_MASK | NT_MASK | IOPL_MASK; | |
317 | break; | |
318 | } | |
319 | ||
320 | /* | |
321 | * Save old state, set default return value (%eax) to 0 | |
322 | */ | |
323 | info->regs32->eax = 0; | |
324 | tsk->thread.saved_esp0 = tsk->thread.esp0; | |
464d1a78 JF |
325 | tsk->thread.saved_fs = info->regs32->xfs; |
326 | savesegment(gs, tsk->thread.saved_gs); | |
1da177e4 LT |
327 | |
328 | tss = &per_cpu(init_tss, get_cpu()); | |
329 | tsk->thread.esp0 = (unsigned long) &info->VM86_TSS_ESP0; | |
330 | if (cpu_has_sep) | |
331 | tsk->thread.sysenter_cs = 0; | |
332 | load_esp0(tss, &tsk->thread); | |
333 | put_cpu(); | |
334 | ||
335 | tsk->thread.screen_bitmap = info->screen_bitmap; | |
336 | if (info->flags & VM86_SCREEN_BITMAP) | |
60ec5585 | 337 | mark_screen_rdonly(tsk->mm); |
7e7f8a03 JB |
338 | |
339 | /*call audit_syscall_exit since we do not exit via the normal paths */ | |
340 | if (unlikely(current->audit_context)) | |
49d26b6e | 341 | audit_syscall_exit(AUDITSC_RESULT(0), 0); |
7e7f8a03 | 342 | |
1da177e4 | 343 | __asm__ __volatile__( |
1da177e4 LT |
344 | "movl %0,%%esp\n\t" |
345 | "movl %1,%%ebp\n\t" | |
464d1a78 | 346 | "mov %2, %%gs\n\t" |
1da177e4 LT |
347 | "jmp resume_userspace" |
348 | : /* no outputs */ | |
49d26b6e | 349 | :"r" (&info->regs), "r" (task_thread_info(tsk)), "r" (0)); |
1da177e4 LT |
350 | /* we never return here */ |
351 | } | |
352 | ||
353 | static inline void return_to_32bit(struct kernel_vm86_regs * regs16, int retval) | |
354 | { | |
355 | struct pt_regs * regs32; | |
356 | ||
357 | regs32 = save_v86_state(regs16); | |
358 | regs32->eax = retval; | |
359 | __asm__ __volatile__("movl %0,%%esp\n\t" | |
360 | "movl %1,%%ebp\n\t" | |
361 | "jmp resume_userspace" | |
362 | : : "r" (regs32), "r" (current_thread_info())); | |
363 | } | |
364 | ||
365 | static inline void set_IF(struct kernel_vm86_regs * regs) | |
366 | { | |
367 | VEFLAGS |= VIF_MASK; | |
368 | if (VEFLAGS & VIP_MASK) | |
369 | return_to_32bit(regs, VM86_STI); | |
370 | } | |
371 | ||
372 | static inline void clear_IF(struct kernel_vm86_regs * regs) | |
373 | { | |
374 | VEFLAGS &= ~VIF_MASK; | |
375 | } | |
376 | ||
377 | static inline void clear_TF(struct kernel_vm86_regs * regs) | |
378 | { | |
49d26b6e | 379 | regs->pt.eflags &= ~TF_MASK; |
1da177e4 LT |
380 | } |
381 | ||
382 | static inline void clear_AC(struct kernel_vm86_regs * regs) | |
383 | { | |
49d26b6e | 384 | regs->pt.eflags &= ~AC_MASK; |
1da177e4 LT |
385 | } |
386 | ||
387 | /* It is correct to call set_IF(regs) from the set_vflags_* | |
388 | * functions. However someone forgot to call clear_IF(regs) | |
389 | * in the opposite case. | |
390 | * After the command sequence CLI PUSHF STI POPF you should | |
391 | * end up with interrups disabled, but you ended up with | |
392 | * interrupts enabled. | |
393 | * ( I was testing my own changes, but the only bug I | |
394 | * could find was in a function I had not changed. ) | |
395 | * [KD] | |
396 | */ | |
397 | ||
398 | static inline void set_vflags_long(unsigned long eflags, struct kernel_vm86_regs * regs) | |
399 | { | |
400 | set_flags(VEFLAGS, eflags, current->thread.v86mask); | |
49d26b6e | 401 | set_flags(regs->pt.eflags, eflags, SAFE_MASK); |
1da177e4 LT |
402 | if (eflags & IF_MASK) |
403 | set_IF(regs); | |
404 | else | |
405 | clear_IF(regs); | |
406 | } | |
407 | ||
408 | static inline void set_vflags_short(unsigned short flags, struct kernel_vm86_regs * regs) | |
409 | { | |
410 | set_flags(VFLAGS, flags, current->thread.v86mask); | |
49d26b6e | 411 | set_flags(regs->pt.eflags, flags, SAFE_MASK); |
1da177e4 LT |
412 | if (flags & IF_MASK) |
413 | set_IF(regs); | |
414 | else | |
415 | clear_IF(regs); | |
416 | } | |
417 | ||
418 | static inline unsigned long get_vflags(struct kernel_vm86_regs * regs) | |
419 | { | |
49d26b6e | 420 | unsigned long flags = regs->pt.eflags & RETURN_MASK; |
1da177e4 LT |
421 | |
422 | if (VEFLAGS & VIF_MASK) | |
423 | flags |= IF_MASK; | |
424 | flags |= IOPL_MASK; | |
425 | return flags | (VEFLAGS & current->thread.v86mask); | |
426 | } | |
427 | ||
428 | static inline int is_revectored(int nr, struct revectored_struct * bitmap) | |
429 | { | |
430 | __asm__ __volatile__("btl %2,%1\n\tsbbl %0,%0" | |
431 | :"=r" (nr) | |
432 | :"m" (*bitmap),"r" (nr)); | |
433 | return nr; | |
434 | } | |
435 | ||
436 | #define val_byte(val, n) (((__u8 *)&val)[n]) | |
437 | ||
438 | #define pushb(base, ptr, val, err_label) \ | |
439 | do { \ | |
440 | __u8 __val = val; \ | |
441 | ptr--; \ | |
442 | if (put_user(__val, base + ptr) < 0) \ | |
443 | goto err_label; \ | |
444 | } while(0) | |
445 | ||
446 | #define pushw(base, ptr, val, err_label) \ | |
447 | do { \ | |
448 | __u16 __val = val; \ | |
449 | ptr--; \ | |
450 | if (put_user(val_byte(__val, 1), base + ptr) < 0) \ | |
451 | goto err_label; \ | |
452 | ptr--; \ | |
453 | if (put_user(val_byte(__val, 0), base + ptr) < 0) \ | |
454 | goto err_label; \ | |
455 | } while(0) | |
456 | ||
457 | #define pushl(base, ptr, val, err_label) \ | |
458 | do { \ | |
459 | __u32 __val = val; \ | |
460 | ptr--; \ | |
461 | if (put_user(val_byte(__val, 3), base + ptr) < 0) \ | |
462 | goto err_label; \ | |
463 | ptr--; \ | |
464 | if (put_user(val_byte(__val, 2), base + ptr) < 0) \ | |
465 | goto err_label; \ | |
466 | ptr--; \ | |
467 | if (put_user(val_byte(__val, 1), base + ptr) < 0) \ | |
468 | goto err_label; \ | |
469 | ptr--; \ | |
470 | if (put_user(val_byte(__val, 0), base + ptr) < 0) \ | |
471 | goto err_label; \ | |
472 | } while(0) | |
473 | ||
474 | #define popb(base, ptr, err_label) \ | |
475 | ({ \ | |
476 | __u8 __res; \ | |
477 | if (get_user(__res, base + ptr) < 0) \ | |
478 | goto err_label; \ | |
479 | ptr++; \ | |
480 | __res; \ | |
481 | }) | |
482 | ||
483 | #define popw(base, ptr, err_label) \ | |
484 | ({ \ | |
485 | __u16 __res; \ | |
486 | if (get_user(val_byte(__res, 0), base + ptr) < 0) \ | |
487 | goto err_label; \ | |
488 | ptr++; \ | |
489 | if (get_user(val_byte(__res, 1), base + ptr) < 0) \ | |
490 | goto err_label; \ | |
491 | ptr++; \ | |
492 | __res; \ | |
493 | }) | |
494 | ||
495 | #define popl(base, ptr, err_label) \ | |
496 | ({ \ | |
497 | __u32 __res; \ | |
498 | if (get_user(val_byte(__res, 0), base + ptr) < 0) \ | |
499 | goto err_label; \ | |
500 | ptr++; \ | |
501 | if (get_user(val_byte(__res, 1), base + ptr) < 0) \ | |
502 | goto err_label; \ | |
503 | ptr++; \ | |
504 | if (get_user(val_byte(__res, 2), base + ptr) < 0) \ | |
505 | goto err_label; \ | |
506 | ptr++; \ | |
507 | if (get_user(val_byte(__res, 3), base + ptr) < 0) \ | |
508 | goto err_label; \ | |
509 | ptr++; \ | |
510 | __res; \ | |
511 | }) | |
512 | ||
513 | /* There are so many possible reasons for this function to return | |
514 | * VM86_INTx, so adding another doesn't bother me. We can expect | |
515 | * userspace programs to be able to handle it. (Getting a problem | |
516 | * in userspace is always better than an Oops anyway.) [KD] | |
517 | */ | |
518 | static void do_int(struct kernel_vm86_regs *regs, int i, | |
519 | unsigned char __user * ssp, unsigned short sp) | |
520 | { | |
521 | unsigned long __user *intr_ptr; | |
522 | unsigned long segoffs; | |
523 | ||
49d26b6e | 524 | if (regs->pt.xcs == BIOSSEG) |
1da177e4 LT |
525 | goto cannot_handle; |
526 | if (is_revectored(i, &KVM86->int_revectored)) | |
527 | goto cannot_handle; | |
528 | if (i==0x21 && is_revectored(AH(regs),&KVM86->int21_revectored)) | |
529 | goto cannot_handle; | |
530 | intr_ptr = (unsigned long __user *) (i << 2); | |
531 | if (get_user(segoffs, intr_ptr)) | |
532 | goto cannot_handle; | |
533 | if ((segoffs >> 16) == BIOSSEG) | |
534 | goto cannot_handle; | |
535 | pushw(ssp, sp, get_vflags(regs), cannot_handle); | |
49d26b6e | 536 | pushw(ssp, sp, regs->pt.xcs, cannot_handle); |
1da177e4 | 537 | pushw(ssp, sp, IP(regs), cannot_handle); |
49d26b6e | 538 | regs->pt.xcs = segoffs >> 16; |
1da177e4 LT |
539 | SP(regs) -= 6; |
540 | IP(regs) = segoffs & 0xffff; | |
541 | clear_TF(regs); | |
542 | clear_IF(regs); | |
543 | clear_AC(regs); | |
544 | return; | |
545 | ||
546 | cannot_handle: | |
547 | return_to_32bit(regs, VM86_INTx + (i << 8)); | |
548 | } | |
549 | ||
550 | int handle_vm86_trap(struct kernel_vm86_regs * regs, long error_code, int trapno) | |
551 | { | |
552 | if (VMPI.is_vm86pus) { | |
553 | if ( (trapno==3) || (trapno==1) ) | |
554 | return_to_32bit(regs, VM86_TRAP + (trapno << 8)); | |
49d26b6e | 555 | do_int(regs, trapno, (unsigned char __user *) (regs->pt.xss << 4), SP(regs)); |
1da177e4 LT |
556 | return 0; |
557 | } | |
558 | if (trapno !=1) | |
559 | return 1; /* we let this handle by the calling routine */ | |
560 | if (current->ptrace & PT_PTRACED) { | |
561 | unsigned long flags; | |
562 | spin_lock_irqsave(¤t->sighand->siglock, flags); | |
563 | sigdelset(¤t->blocked, SIGTRAP); | |
564 | recalc_sigpending(); | |
565 | spin_unlock_irqrestore(¤t->sighand->siglock, flags); | |
566 | } | |
567 | send_sig(SIGTRAP, current, 1); | |
568 | current->thread.trap_no = trapno; | |
569 | current->thread.error_code = error_code; | |
570 | return 0; | |
571 | } | |
572 | ||
573 | void handle_vm86_fault(struct kernel_vm86_regs * regs, long error_code) | |
574 | { | |
575 | unsigned char opcode; | |
576 | unsigned char __user *csp; | |
577 | unsigned char __user *ssp; | |
5fd75ebb | 578 | unsigned short ip, sp, orig_flags; |
1da177e4 LT |
579 | int data32, pref_done; |
580 | ||
581 | #define CHECK_IF_IN_TRAP \ | |
582 | if (VMPI.vm86dbg_active && VMPI.vm86dbg_TFpendig) \ | |
583 | newflags |= TF_MASK | |
584 | #define VM86_FAULT_RETURN do { \ | |
585 | if (VMPI.force_return_for_pic && (VEFLAGS & (IF_MASK | VIF_MASK))) \ | |
586 | return_to_32bit(regs, VM86_PICRETURN); \ | |
5fd75ebb PT |
587 | if (orig_flags & TF_MASK) \ |
588 | handle_vm86_trap(regs, 0, 1); \ | |
1da177e4 LT |
589 | return; } while (0) |
590 | ||
49d26b6e | 591 | orig_flags = *(unsigned short *)®s->pt.eflags; |
5fd75ebb | 592 | |
49d26b6e JF |
593 | csp = (unsigned char __user *) (regs->pt.xcs << 4); |
594 | ssp = (unsigned char __user *) (regs->pt.xss << 4); | |
1da177e4 LT |
595 | sp = SP(regs); |
596 | ip = IP(regs); | |
597 | ||
598 | data32 = 0; | |
599 | pref_done = 0; | |
600 | do { | |
601 | switch (opcode = popb(csp, ip, simulate_sigsegv)) { | |
602 | case 0x66: /* 32-bit data */ data32=1; break; | |
603 | case 0x67: /* 32-bit address */ break; | |
604 | case 0x2e: /* CS */ break; | |
605 | case 0x3e: /* DS */ break; | |
606 | case 0x26: /* ES */ break; | |
607 | case 0x36: /* SS */ break; | |
608 | case 0x65: /* GS */ break; | |
609 | case 0x64: /* FS */ break; | |
610 | case 0xf2: /* repnz */ break; | |
611 | case 0xf3: /* rep */ break; | |
612 | default: pref_done = 1; | |
613 | } | |
614 | } while (!pref_done); | |
615 | ||
616 | switch (opcode) { | |
617 | ||
618 | /* pushf */ | |
619 | case 0x9c: | |
620 | if (data32) { | |
621 | pushl(ssp, sp, get_vflags(regs), simulate_sigsegv); | |
622 | SP(regs) -= 4; | |
623 | } else { | |
624 | pushw(ssp, sp, get_vflags(regs), simulate_sigsegv); | |
625 | SP(regs) -= 2; | |
626 | } | |
627 | IP(regs) = ip; | |
628 | VM86_FAULT_RETURN; | |
629 | ||
630 | /* popf */ | |
631 | case 0x9d: | |
632 | { | |
633 | unsigned long newflags; | |
634 | if (data32) { | |
635 | newflags=popl(ssp, sp, simulate_sigsegv); | |
636 | SP(regs) += 4; | |
637 | } else { | |
638 | newflags = popw(ssp, sp, simulate_sigsegv); | |
639 | SP(regs) += 2; | |
640 | } | |
641 | IP(regs) = ip; | |
642 | CHECK_IF_IN_TRAP; | |
643 | if (data32) { | |
644 | set_vflags_long(newflags, regs); | |
645 | } else { | |
646 | set_vflags_short(newflags, regs); | |
647 | } | |
648 | VM86_FAULT_RETURN; | |
649 | } | |
650 | ||
651 | /* int xx */ | |
652 | case 0xcd: { | |
653 | int intno=popb(csp, ip, simulate_sigsegv); | |
654 | IP(regs) = ip; | |
655 | if (VMPI.vm86dbg_active) { | |
656 | if ( (1 << (intno &7)) & VMPI.vm86dbg_intxxtab[intno >> 3] ) | |
657 | return_to_32bit(regs, VM86_INTx + (intno << 8)); | |
658 | } | |
659 | do_int(regs, intno, ssp, sp); | |
660 | return; | |
661 | } | |
662 | ||
663 | /* iret */ | |
664 | case 0xcf: | |
665 | { | |
666 | unsigned long newip; | |
667 | unsigned long newcs; | |
668 | unsigned long newflags; | |
669 | if (data32) { | |
670 | newip=popl(ssp, sp, simulate_sigsegv); | |
671 | newcs=popl(ssp, sp, simulate_sigsegv); | |
672 | newflags=popl(ssp, sp, simulate_sigsegv); | |
673 | SP(regs) += 12; | |
674 | } else { | |
675 | newip = popw(ssp, sp, simulate_sigsegv); | |
676 | newcs = popw(ssp, sp, simulate_sigsegv); | |
677 | newflags = popw(ssp, sp, simulate_sigsegv); | |
678 | SP(regs) += 6; | |
679 | } | |
680 | IP(regs) = newip; | |
49d26b6e | 681 | regs->pt.xcs = newcs; |
1da177e4 LT |
682 | CHECK_IF_IN_TRAP; |
683 | if (data32) { | |
684 | set_vflags_long(newflags, regs); | |
685 | } else { | |
686 | set_vflags_short(newflags, regs); | |
687 | } | |
688 | VM86_FAULT_RETURN; | |
689 | } | |
690 | ||
691 | /* cli */ | |
692 | case 0xfa: | |
693 | IP(regs) = ip; | |
694 | clear_IF(regs); | |
695 | VM86_FAULT_RETURN; | |
696 | ||
697 | /* sti */ | |
698 | /* | |
699 | * Damn. This is incorrect: the 'sti' instruction should actually | |
700 | * enable interrupts after the /next/ instruction. Not good. | |
701 | * | |
702 | * Probably needs some horsing around with the TF flag. Aiee.. | |
703 | */ | |
704 | case 0xfb: | |
705 | IP(regs) = ip; | |
706 | set_IF(regs); | |
707 | VM86_FAULT_RETURN; | |
708 | ||
709 | default: | |
710 | return_to_32bit(regs, VM86_UNKNOWN); | |
711 | } | |
712 | ||
713 | return; | |
714 | ||
715 | simulate_sigsegv: | |
716 | /* FIXME: After a long discussion with Stas we finally | |
717 | * agreed, that this is wrong. Here we should | |
718 | * really send a SIGSEGV to the user program. | |
719 | * But how do we create the correct context? We | |
720 | * are inside a general protection fault handler | |
721 | * and has just returned from a page fault handler. | |
722 | * The correct context for the signal handler | |
723 | * should be a mixture of the two, but how do we | |
724 | * get the information? [KD] | |
725 | */ | |
726 | return_to_32bit(regs, VM86_UNKNOWN); | |
727 | } | |
728 | ||
729 | /* ---------------- vm86 special IRQ passing stuff ----------------- */ | |
730 | ||
731 | #define VM86_IRQNAME "vm86irq" | |
732 | ||
733 | static struct vm86_irqs { | |
734 | struct task_struct *tsk; | |
735 | int sig; | |
736 | } vm86_irqs[16]; | |
737 | ||
738 | static DEFINE_SPINLOCK(irqbits_lock); | |
739 | static int irqbits; | |
740 | ||
741 | #define ALLOWED_SIGS ( 1 /* 0 = don't send a signal */ \ | |
742 | | (1 << SIGUSR1) | (1 << SIGUSR2) | (1 << SIGIO) | (1 << SIGURG) \ | |
743 | | (1 << SIGUNUSED) ) | |
744 | ||
7d12e780 | 745 | static irqreturn_t irq_handler(int intno, void *dev_id) |
1da177e4 LT |
746 | { |
747 | int irq_bit; | |
748 | unsigned long flags; | |
749 | ||
750 | spin_lock_irqsave(&irqbits_lock, flags); | |
751 | irq_bit = 1 << intno; | |
752 | if ((irqbits & irq_bit) || ! vm86_irqs[intno].tsk) | |
753 | goto out; | |
754 | irqbits |= irq_bit; | |
755 | if (vm86_irqs[intno].sig) | |
756 | send_sig(vm86_irqs[intno].sig, vm86_irqs[intno].tsk, 1); | |
1da177e4 LT |
757 | /* |
758 | * IRQ will be re-enabled when user asks for the irq (whether | |
759 | * polling or as a result of the signal) | |
760 | */ | |
ad671423 PP |
761 | disable_irq_nosync(intno); |
762 | spin_unlock_irqrestore(&irqbits_lock, flags); | |
1da177e4 LT |
763 | return IRQ_HANDLED; |
764 | ||
765 | out: | |
766 | spin_unlock_irqrestore(&irqbits_lock, flags); | |
767 | return IRQ_NONE; | |
768 | } | |
769 | ||
770 | static inline void free_vm86_irq(int irqnumber) | |
771 | { | |
772 | unsigned long flags; | |
773 | ||
774 | free_irq(irqnumber, NULL); | |
775 | vm86_irqs[irqnumber].tsk = NULL; | |
776 | ||
777 | spin_lock_irqsave(&irqbits_lock, flags); | |
778 | irqbits &= ~(1 << irqnumber); | |
779 | spin_unlock_irqrestore(&irqbits_lock, flags); | |
780 | } | |
781 | ||
782 | void release_vm86_irqs(struct task_struct *task) | |
783 | { | |
784 | int i; | |
785 | for (i = FIRST_VM86_IRQ ; i <= LAST_VM86_IRQ; i++) | |
786 | if (vm86_irqs[i].tsk == task) | |
787 | free_vm86_irq(i); | |
788 | } | |
789 | ||
790 | static inline int get_and_reset_irq(int irqnumber) | |
791 | { | |
792 | int bit; | |
793 | unsigned long flags; | |
ad671423 | 794 | int ret = 0; |
1da177e4 LT |
795 | |
796 | if (invalid_vm86_irq(irqnumber)) return 0; | |
797 | if (vm86_irqs[irqnumber].tsk != current) return 0; | |
798 | spin_lock_irqsave(&irqbits_lock, flags); | |
799 | bit = irqbits & (1 << irqnumber); | |
800 | irqbits &= ~bit; | |
ad671423 PP |
801 | if (bit) { |
802 | enable_irq(irqnumber); | |
803 | ret = 1; | |
804 | } | |
805 | ||
1da177e4 | 806 | spin_unlock_irqrestore(&irqbits_lock, flags); |
ad671423 | 807 | return ret; |
1da177e4 LT |
808 | } |
809 | ||
810 | ||
811 | static int do_vm86_irq_handling(int subfunction, int irqnumber) | |
812 | { | |
813 | int ret; | |
814 | switch (subfunction) { | |
815 | case VM86_GET_AND_RESET_IRQ: { | |
816 | return get_and_reset_irq(irqnumber); | |
817 | } | |
818 | case VM86_GET_IRQ_BITS: { | |
819 | return irqbits; | |
820 | } | |
821 | case VM86_REQUEST_IRQ: { | |
822 | int sig = irqnumber >> 8; | |
823 | int irq = irqnumber & 255; | |
824 | if (!capable(CAP_SYS_ADMIN)) return -EPERM; | |
825 | if (!((1 << sig) & ALLOWED_SIGS)) return -EPERM; | |
826 | if (invalid_vm86_irq(irq)) return -EPERM; | |
827 | if (vm86_irqs[irq].tsk) return -EPERM; | |
828 | ret = request_irq(irq, &irq_handler, 0, VM86_IRQNAME, NULL); | |
829 | if (ret) return ret; | |
830 | vm86_irqs[irq].sig = sig; | |
831 | vm86_irqs[irq].tsk = current; | |
832 | return irq; | |
833 | } | |
834 | case VM86_FREE_IRQ: { | |
835 | if (invalid_vm86_irq(irqnumber)) return -EPERM; | |
836 | if (!vm86_irqs[irqnumber].tsk) return 0; | |
837 | if (vm86_irqs[irqnumber].tsk != current) return -EPERM; | |
838 | free_vm86_irq(irqnumber); | |
839 | return 0; | |
840 | } | |
841 | } | |
842 | return -EINVAL; | |
843 | } | |
844 |