1 | /* |
---|
2 | * linux/kernel/vm86.c |
---|
3 | * |
---|
4 | * Copyright (C) 1994 Linus Torvalds |
---|
5 | * |
---|
6 | * 29 dec 2001 - Fixed oopses caused by unchecked access to the vm86 |
---|
7 | * stack - Manfred Spraul <manfred@colorfullife.com> |
---|
8 | * |
---|
9 | * 22 mar 2002 - Manfred detected the stackfaults, but didn't handle |
---|
10 | * them correctly. Now the emulation will be in a |
---|
11 | * consistent state after stackfaults - Kasper Dupont |
---|
12 | * <kasperd@daimi.au.dk> |
---|
13 | * |
---|
14 | * 22 mar 2002 - Added missing clear_IF in set_vflags_* Kasper Dupont |
---|
15 | * <kasperd@daimi.au.dk> |
---|
16 | * |
---|
17 | * ?? ??? 2002 - Fixed premature returns from handle_vm86_fault |
---|
18 | * caused by Kasper Dupont's changes - Stas Sergeev |
---|
19 | * |
---|
20 | * 4 apr 2002 - Fixed CHECK_IF_IN_TRAP broken by Stas' changes. |
---|
21 | * Kasper Dupont <kasperd@daimi.au.dk> |
---|
22 | * |
---|
23 | * 9 apr 2002 - Changed syntax of macros in handle_vm86_fault. |
---|
24 | * Kasper Dupont <kasperd@daimi.au.dk> |
---|
25 | * |
---|
26 | * 9 apr 2002 - Changed stack access macros to jump to a label |
---|
27 | * instead of returning to userspace. This simplifies |
---|
28 | * do_int, and is needed by handle_vm6_fault. Kasper |
---|
29 | * Dupont <kasperd@daimi.au.dk> |
---|
30 | * |
---|
31 | */ |
---|
32 | |
---|
33 | #include <linux/capability.h> |
---|
34 | #include <linux/errno.h> |
---|
35 | #include <linux/interrupt.h> |
---|
36 | #include <linux/sched.h> |
---|
37 | #include <linux/kernel.h> |
---|
38 | #include <linux/signal.h> |
---|
39 | #include <linux/string.h> |
---|
40 | #include <linux/mm.h> |
---|
41 | #include <linux/smp.h> |
---|
42 | #include <linux/smp_lock.h> |
---|
43 | #include <linux/highmem.h> |
---|
44 | #include <linux/ptrace.h> |
---|
45 | #include <linux/audit.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 | |
---|
97 | struct pt_regs * FASTCALL(save_v86_state(struct kernel_vm86_regs * regs)); |
---|
98 | struct pt_regs * fastcall save_v86_state(struct kernel_vm86_regs * regs) |
---|
99 | { |
---|
100 | #ifndef CONFIG_X86_NO_TSS |
---|
101 | struct tss_struct *tss; |
---|
102 | #endif |
---|
103 | struct pt_regs *ret; |
---|
104 | unsigned long tmp; |
---|
105 | |
---|
106 | /* |
---|
107 | * This gets called from entry.S with interrupts disabled, but |
---|
108 | * from process context. Enable interrupts here, before trying |
---|
109 | * to access user space. |
---|
110 | */ |
---|
111 | local_irq_enable(); |
---|
112 | |
---|
113 | if (!current->thread.vm86_info) { |
---|
114 | printk("no vm86_info: BAD\n"); |
---|
115 | do_exit(SIGSEGV); |
---|
116 | } |
---|
117 | set_flags(regs->eflags, VEFLAGS, VIF_MASK | current->thread.v86mask); |
---|
118 | tmp = copy_to_user(¤t->thread.vm86_info->regs,regs, VM86_REGS_SIZE1); |
---|
119 | tmp += copy_to_user(¤t->thread.vm86_info->regs.VM86_REGS_PART2, |
---|
120 | ®s->VM86_REGS_PART2, VM86_REGS_SIZE2); |
---|
121 | tmp += put_user(current->thread.screen_bitmap,¤t->thread.vm86_info->screen_bitmap); |
---|
122 | if (tmp) { |
---|
123 | printk("vm86: could not access userspace vm86_info\n"); |
---|
124 | do_exit(SIGSEGV); |
---|
125 | } |
---|
126 | |
---|
127 | #ifndef CONFIG_X86_NO_TSS |
---|
128 | tss = &per_cpu(init_tss, get_cpu()); |
---|
129 | #endif |
---|
130 | current->thread.esp0 = current->thread.saved_esp0; |
---|
131 | current->thread.sysenter_cs = __KERNEL_CS; |
---|
132 | load_esp0(tss, ¤t->thread); |
---|
133 | current->thread.saved_esp0 = 0; |
---|
134 | #ifndef CONFIG_X86_NO_TSS |
---|
135 | put_cpu(); |
---|
136 | #endif |
---|
137 | |
---|
138 | loadsegment(fs, current->thread.saved_fs); |
---|
139 | loadsegment(gs, current->thread.saved_gs); |
---|
140 | ret = KVM86->regs32; |
---|
141 | return ret; |
---|
142 | } |
---|
143 | |
---|
144 | static void mark_screen_rdonly(struct mm_struct *mm) |
---|
145 | { |
---|
146 | pgd_t *pgd; |
---|
147 | pud_t *pud; |
---|
148 | pmd_t *pmd; |
---|
149 | pte_t *pte; |
---|
150 | spinlock_t *ptl; |
---|
151 | int i; |
---|
152 | |
---|
153 | pgd = pgd_offset(mm, 0xA0000); |
---|
154 | if (pgd_none_or_clear_bad(pgd)) |
---|
155 | goto out; |
---|
156 | pud = pud_offset(pgd, 0xA0000); |
---|
157 | if (pud_none_or_clear_bad(pud)) |
---|
158 | goto out; |
---|
159 | pmd = pmd_offset(pud, 0xA0000); |
---|
160 | if (pmd_none_or_clear_bad(pmd)) |
---|
161 | goto out; |
---|
162 | pte = pte_offset_map_lock(mm, pmd, 0xA0000, &ptl); |
---|
163 | for (i = 0; i < 32; i++) { |
---|
164 | if (pte_present(*pte)) |
---|
165 | set_pte(pte, pte_wrprotect(*pte)); |
---|
166 | pte++; |
---|
167 | } |
---|
168 | pte_unmap_unlock(pte, ptl); |
---|
169 | out: |
---|
170 | flush_tlb(); |
---|
171 | } |
---|
172 | |
---|
173 | |
---|
174 | |
---|
175 | static int do_vm86_irq_handling(int subfunction, int irqnumber); |
---|
176 | static void do_sys_vm86(struct kernel_vm86_struct *info, struct task_struct *tsk); |
---|
177 | |
---|
178 | asmlinkage int sys_vm86old(struct pt_regs regs) |
---|
179 | { |
---|
180 | struct vm86_struct __user *v86 = (struct vm86_struct __user *)regs.ebx; |
---|
181 | struct kernel_vm86_struct info; /* declare this _on top_, |
---|
182 | * this avoids wasting of stack space. |
---|
183 | * This remains on the stack until we |
---|
184 | * return to 32 bit user space. |
---|
185 | */ |
---|
186 | struct task_struct *tsk; |
---|
187 | int tmp, ret = -EPERM; |
---|
188 | |
---|
189 | tsk = current; |
---|
190 | if (tsk->thread.saved_esp0) |
---|
191 | goto out; |
---|
192 | tmp = copy_from_user(&info, v86, VM86_REGS_SIZE1); |
---|
193 | tmp += copy_from_user(&info.regs.VM86_REGS_PART2, &v86->regs.VM86_REGS_PART2, |
---|
194 | (long)&info.vm86plus - (long)&info.regs.VM86_REGS_PART2); |
---|
195 | ret = -EFAULT; |
---|
196 | if (tmp) |
---|
197 | goto out; |
---|
198 | memset(&info.vm86plus, 0, (int)&info.regs32 - (int)&info.vm86plus); |
---|
199 | info.regs32 = ®s; |
---|
200 | tsk->thread.vm86_info = v86; |
---|
201 | do_sys_vm86(&info, tsk); |
---|
202 | ret = 0; /* we never return here */ |
---|
203 | out: |
---|
204 | return ret; |
---|
205 | } |
---|
206 | |
---|
207 | |
---|
208 | asmlinkage int sys_vm86(struct pt_regs regs) |
---|
209 | { |
---|
210 | struct kernel_vm86_struct info; /* declare this _on top_, |
---|
211 | * this avoids wasting of stack space. |
---|
212 | * This remains on the stack until we |
---|
213 | * return to 32 bit user space. |
---|
214 | */ |
---|
215 | struct task_struct *tsk; |
---|
216 | int tmp, ret; |
---|
217 | struct vm86plus_struct __user *v86; |
---|
218 | |
---|
219 | tsk = current; |
---|
220 | switch (regs.ebx) { |
---|
221 | case VM86_REQUEST_IRQ: |
---|
222 | case VM86_FREE_IRQ: |
---|
223 | case VM86_GET_IRQ_BITS: |
---|
224 | case VM86_GET_AND_RESET_IRQ: |
---|
225 | ret = do_vm86_irq_handling(regs.ebx, (int)regs.ecx); |
---|
226 | goto out; |
---|
227 | case VM86_PLUS_INSTALL_CHECK: |
---|
228 | /* NOTE: on old vm86 stuff this will return the error |
---|
229 | from access_ok(), because the subfunction is |
---|
230 | interpreted as (invalid) address to vm86_struct. |
---|
231 | So the installation check works. |
---|
232 | */ |
---|
233 | ret = 0; |
---|
234 | goto out; |
---|
235 | } |
---|
236 | |
---|
237 | /* we come here only for functions VM86_ENTER, VM86_ENTER_NO_BYPASS */ |
---|
238 | ret = -EPERM; |
---|
239 | if (tsk->thread.saved_esp0) |
---|
240 | goto out; |
---|
241 | v86 = (struct vm86plus_struct __user *)regs.ecx; |
---|
242 | tmp = copy_from_user(&info, v86, VM86_REGS_SIZE1); |
---|
243 | tmp += copy_from_user(&info.regs.VM86_REGS_PART2, &v86->regs.VM86_REGS_PART2, |
---|
244 | (long)&info.regs32 - (long)&info.regs.VM86_REGS_PART2); |
---|
245 | ret = -EFAULT; |
---|
246 | if (tmp) |
---|
247 | goto out; |
---|
248 | info.regs32 = ®s; |
---|
249 | info.vm86plus.is_vm86pus = 1; |
---|
250 | tsk->thread.vm86_info = (struct vm86_struct __user *)v86; |
---|
251 | do_sys_vm86(&info, tsk); |
---|
252 | ret = 0; /* we never return here */ |
---|
253 | out: |
---|
254 | return ret; |
---|
255 | } |
---|
256 | |
---|
257 | |
---|
258 | static void do_sys_vm86(struct kernel_vm86_struct *info, struct task_struct *tsk) |
---|
259 | { |
---|
260 | #ifndef CONFIG_X86_NO_TSS |
---|
261 | struct tss_struct *tss; |
---|
262 | #endif |
---|
263 | long eax; |
---|
264 | /* |
---|
265 | * make sure the vm86() system call doesn't try to do anything silly |
---|
266 | */ |
---|
267 | info->regs.__null_ds = 0; |
---|
268 | info->regs.__null_es = 0; |
---|
269 | |
---|
270 | /* we are clearing fs,gs later just before "jmp resume_userspace", |
---|
271 | * because starting with Linux 2.1.x they aren't no longer saved/restored |
---|
272 | */ |
---|
273 | |
---|
274 | /* |
---|
275 | * The eflags register is also special: we cannot trust that the user |
---|
276 | * has set it up safely, so this makes sure interrupt etc flags are |
---|
277 | * inherited from protected mode. |
---|
278 | */ |
---|
279 | VEFLAGS = info->regs.eflags; |
---|
280 | info->regs.eflags &= SAFE_MASK; |
---|
281 | info->regs.eflags |= info->regs32->eflags & ~SAFE_MASK; |
---|
282 | info->regs.eflags |= VM_MASK; |
---|
283 | |
---|
284 | switch (info->cpu_type) { |
---|
285 | case CPU_286: |
---|
286 | tsk->thread.v86mask = 0; |
---|
287 | break; |
---|
288 | case CPU_386: |
---|
289 | tsk->thread.v86mask = NT_MASK | IOPL_MASK; |
---|
290 | break; |
---|
291 | case CPU_486: |
---|
292 | tsk->thread.v86mask = AC_MASK | NT_MASK | IOPL_MASK; |
---|
293 | break; |
---|
294 | default: |
---|
295 | tsk->thread.v86mask = ID_MASK | AC_MASK | NT_MASK | IOPL_MASK; |
---|
296 | break; |
---|
297 | } |
---|
298 | |
---|
299 | /* |
---|
300 | * Save old state, set default return value (%eax) to 0 |
---|
301 | */ |
---|
302 | info->regs32->eax = 0; |
---|
303 | tsk->thread.saved_esp0 = tsk->thread.esp0; |
---|
304 | savesegment(fs, tsk->thread.saved_fs); |
---|
305 | savesegment(gs, tsk->thread.saved_gs); |
---|
306 | |
---|
307 | #ifndef CONFIG_X86_NO_TSS |
---|
308 | tss = &per_cpu(init_tss, get_cpu()); |
---|
309 | #endif |
---|
310 | tsk->thread.esp0 = (unsigned long) &info->VM86_TSS_ESP0; |
---|
311 | if (cpu_has_sep) |
---|
312 | tsk->thread.sysenter_cs = 0; |
---|
313 | load_esp0(tss, &tsk->thread); |
---|
314 | #ifndef CONFIG_X86_NO_TSS |
---|
315 | put_cpu(); |
---|
316 | #endif |
---|
317 | |
---|
318 | tsk->thread.screen_bitmap = info->screen_bitmap; |
---|
319 | if (info->flags & VM86_SCREEN_BITMAP) |
---|
320 | mark_screen_rdonly(tsk->mm); |
---|
321 | __asm__ __volatile__("xorl %eax,%eax; movl %eax,%fs; movl %eax,%gs\n\t"); |
---|
322 | __asm__ __volatile__("movl %%eax, %0\n" :"=r"(eax)); |
---|
323 | |
---|
324 | /*call audit_syscall_exit since we do not exit via the normal paths */ |
---|
325 | if (unlikely(current->audit_context)) |
---|
326 | audit_syscall_exit(AUDITSC_RESULT(eax), eax); |
---|
327 | |
---|
328 | __asm__ __volatile__( |
---|
329 | "movl %0,%%esp\n\t" |
---|
330 | "movl %1,%%ebp\n\t" |
---|
331 | "jmp resume_userspace" |
---|
332 | : /* no outputs */ |
---|
333 | :"r" (&info->regs), "r" (task_thread_info(tsk))); |
---|
334 | /* we never return here */ |
---|
335 | } |
---|
336 | |
---|
337 | static inline void return_to_32bit(struct kernel_vm86_regs * regs16, int retval) |
---|
338 | { |
---|
339 | struct pt_regs * regs32; |
---|
340 | |
---|
341 | regs32 = save_v86_state(regs16); |
---|
342 | regs32->eax = retval; |
---|
343 | __asm__ __volatile__("movl %0,%%esp\n\t" |
---|
344 | "movl %1,%%ebp\n\t" |
---|
345 | "jmp resume_userspace" |
---|
346 | : : "r" (regs32), "r" (current_thread_info())); |
---|
347 | } |
---|
348 | |
---|
349 | static inline void set_IF(struct kernel_vm86_regs * regs) |
---|
350 | { |
---|
351 | VEFLAGS |= VIF_MASK; |
---|
352 | if (VEFLAGS & VIP_MASK) |
---|
353 | return_to_32bit(regs, VM86_STI); |
---|
354 | } |
---|
355 | |
---|
356 | static inline void clear_IF(struct kernel_vm86_regs * regs) |
---|
357 | { |
---|
358 | VEFLAGS &= ~VIF_MASK; |
---|
359 | } |
---|
360 | |
---|
361 | static inline void clear_TF(struct kernel_vm86_regs * regs) |
---|
362 | { |
---|
363 | regs->eflags &= ~TF_MASK; |
---|
364 | } |
---|
365 | |
---|
366 | static inline void clear_AC(struct kernel_vm86_regs * regs) |
---|
367 | { |
---|
368 | regs->eflags &= ~AC_MASK; |
---|
369 | } |
---|
370 | |
---|
371 | /* It is correct to call set_IF(regs) from the set_vflags_* |
---|
372 | * functions. However someone forgot to call clear_IF(regs) |
---|
373 | * in the opposite case. |
---|
374 | * After the command sequence CLI PUSHF STI POPF you should |
---|
375 | * end up with interrups disabled, but you ended up with |
---|
376 | * interrupts enabled. |
---|
377 | * ( I was testing my own changes, but the only bug I |
---|
378 | * could find was in a function I had not changed. ) |
---|
379 | * [KD] |
---|
380 | */ |
---|
381 | |
---|
382 | static inline void set_vflags_long(unsigned long eflags, struct kernel_vm86_regs * regs) |
---|
383 | { |
---|
384 | set_flags(VEFLAGS, eflags, current->thread.v86mask); |
---|
385 | set_flags(regs->eflags, eflags, SAFE_MASK); |
---|
386 | if (eflags & IF_MASK) |
---|
387 | set_IF(regs); |
---|
388 | else |
---|
389 | clear_IF(regs); |
---|
390 | } |
---|
391 | |
---|
392 | static inline void set_vflags_short(unsigned short flags, struct kernel_vm86_regs * regs) |
---|
393 | { |
---|
394 | set_flags(VFLAGS, flags, current->thread.v86mask); |
---|
395 | set_flags(regs->eflags, flags, SAFE_MASK); |
---|
396 | if (flags & IF_MASK) |
---|
397 | set_IF(regs); |
---|
398 | else |
---|
399 | clear_IF(regs); |
---|
400 | } |
---|
401 | |
---|
402 | static inline unsigned long get_vflags(struct kernel_vm86_regs * regs) |
---|
403 | { |
---|
404 | unsigned long flags = regs->eflags & RETURN_MASK; |
---|
405 | |
---|
406 | if (VEFLAGS & VIF_MASK) |
---|
407 | flags |= IF_MASK; |
---|
408 | flags |= IOPL_MASK; |
---|
409 | return flags | (VEFLAGS & current->thread.v86mask); |
---|
410 | } |
---|
411 | |
---|
412 | static inline int is_revectored(int nr, struct revectored_struct * bitmap) |
---|
413 | { |
---|
414 | __asm__ __volatile__("btl %2,%1\n\tsbbl %0,%0" |
---|
415 | :"=r" (nr) |
---|
416 | :"m" (*bitmap),"r" (nr)); |
---|
417 | return nr; |
---|
418 | } |
---|
419 | |
---|
420 | #define val_byte(val, n) (((__u8 *)&val)[n]) |
---|
421 | |
---|
422 | #define pushb(base, ptr, val, err_label) \ |
---|
423 | do { \ |
---|
424 | __u8 __val = val; \ |
---|
425 | ptr--; \ |
---|
426 | if (put_user(__val, base + ptr) < 0) \ |
---|
427 | goto err_label; \ |
---|
428 | } while(0) |
---|
429 | |
---|
430 | #define pushw(base, ptr, val, err_label) \ |
---|
431 | do { \ |
---|
432 | __u16 __val = val; \ |
---|
433 | ptr--; \ |
---|
434 | if (put_user(val_byte(__val, 1), base + ptr) < 0) \ |
---|
435 | goto err_label; \ |
---|
436 | ptr--; \ |
---|
437 | if (put_user(val_byte(__val, 0), base + ptr) < 0) \ |
---|
438 | goto err_label; \ |
---|
439 | } while(0) |
---|
440 | |
---|
441 | #define pushl(base, ptr, val, err_label) \ |
---|
442 | do { \ |
---|
443 | __u32 __val = val; \ |
---|
444 | ptr--; \ |
---|
445 | if (put_user(val_byte(__val, 3), base + ptr) < 0) \ |
---|
446 | goto err_label; \ |
---|
447 | ptr--; \ |
---|
448 | if (put_user(val_byte(__val, 2), base + ptr) < 0) \ |
---|
449 | goto err_label; \ |
---|
450 | ptr--; \ |
---|
451 | if (put_user(val_byte(__val, 1), base + ptr) < 0) \ |
---|
452 | goto err_label; \ |
---|
453 | ptr--; \ |
---|
454 | if (put_user(val_byte(__val, 0), base + ptr) < 0) \ |
---|
455 | goto err_label; \ |
---|
456 | } while(0) |
---|
457 | |
---|
458 | #define popb(base, ptr, err_label) \ |
---|
459 | ({ \ |
---|
460 | __u8 __res; \ |
---|
461 | if (get_user(__res, base + ptr) < 0) \ |
---|
462 | goto err_label; \ |
---|
463 | ptr++; \ |
---|
464 | __res; \ |
---|
465 | }) |
---|
466 | |
---|
467 | #define popw(base, ptr, err_label) \ |
---|
468 | ({ \ |
---|
469 | __u16 __res; \ |
---|
470 | if (get_user(val_byte(__res, 0), base + ptr) < 0) \ |
---|
471 | goto err_label; \ |
---|
472 | ptr++; \ |
---|
473 | if (get_user(val_byte(__res, 1), base + ptr) < 0) \ |
---|
474 | goto err_label; \ |
---|
475 | ptr++; \ |
---|
476 | __res; \ |
---|
477 | }) |
---|
478 | |
---|
479 | #define popl(base, ptr, err_label) \ |
---|
480 | ({ \ |
---|
481 | __u32 __res; \ |
---|
482 | if (get_user(val_byte(__res, 0), base + ptr) < 0) \ |
---|
483 | goto err_label; \ |
---|
484 | ptr++; \ |
---|
485 | if (get_user(val_byte(__res, 1), base + ptr) < 0) \ |
---|
486 | goto err_label; \ |
---|
487 | ptr++; \ |
---|
488 | if (get_user(val_byte(__res, 2), base + ptr) < 0) \ |
---|
489 | goto err_label; \ |
---|
490 | ptr++; \ |
---|
491 | if (get_user(val_byte(__res, 3), base + ptr) < 0) \ |
---|
492 | goto err_label; \ |
---|
493 | ptr++; \ |
---|
494 | __res; \ |
---|
495 | }) |
---|
496 | |
---|
497 | /* There are so many possible reasons for this function to return |
---|
498 | * VM86_INTx, so adding another doesn't bother me. We can expect |
---|
499 | * userspace programs to be able to handle it. (Getting a problem |
---|
500 | * in userspace is always better than an Oops anyway.) [KD] |
---|
501 | */ |
---|
502 | static void do_int(struct kernel_vm86_regs *regs, int i, |
---|
503 | unsigned char __user * ssp, unsigned short sp) |
---|
504 | { |
---|
505 | unsigned long __user *intr_ptr; |
---|
506 | unsigned long segoffs; |
---|
507 | |
---|
508 | if (regs->cs == BIOSSEG) |
---|
509 | goto cannot_handle; |
---|
510 | if (is_revectored(i, &KVM86->int_revectored)) |
---|
511 | goto cannot_handle; |
---|
512 | if (i==0x21 && is_revectored(AH(regs),&KVM86->int21_revectored)) |
---|
513 | goto cannot_handle; |
---|
514 | intr_ptr = (unsigned long __user *) (i << 2); |
---|
515 | if (get_user(segoffs, intr_ptr)) |
---|
516 | goto cannot_handle; |
---|
517 | if ((segoffs >> 16) == BIOSSEG) |
---|
518 | goto cannot_handle; |
---|
519 | pushw(ssp, sp, get_vflags(regs), cannot_handle); |
---|
520 | pushw(ssp, sp, regs->cs, cannot_handle); |
---|
521 | pushw(ssp, sp, IP(regs), cannot_handle); |
---|
522 | regs->cs = segoffs >> 16; |
---|
523 | SP(regs) -= 6; |
---|
524 | IP(regs) = segoffs & 0xffff; |
---|
525 | clear_TF(regs); |
---|
526 | clear_IF(regs); |
---|
527 | clear_AC(regs); |
---|
528 | return; |
---|
529 | |
---|
530 | cannot_handle: |
---|
531 | return_to_32bit(regs, VM86_INTx + (i << 8)); |
---|
532 | } |
---|
533 | |
---|
534 | int handle_vm86_trap(struct kernel_vm86_regs * regs, long error_code, int trapno) |
---|
535 | { |
---|
536 | if (VMPI.is_vm86pus) { |
---|
537 | if ( (trapno==3) || (trapno==1) ) |
---|
538 | return_to_32bit(regs, VM86_TRAP + (trapno << 8)); |
---|
539 | do_int(regs, trapno, (unsigned char __user *) (regs->ss << 4), SP(regs)); |
---|
540 | return 0; |
---|
541 | } |
---|
542 | if (trapno !=1) |
---|
543 | return 1; /* we let this handle by the calling routine */ |
---|
544 | if (current->ptrace & PT_PTRACED) { |
---|
545 | unsigned long flags; |
---|
546 | spin_lock_irqsave(¤t->sighand->siglock, flags); |
---|
547 | sigdelset(¤t->blocked, SIGTRAP); |
---|
548 | recalc_sigpending(); |
---|
549 | spin_unlock_irqrestore(¤t->sighand->siglock, flags); |
---|
550 | } |
---|
551 | send_sig(SIGTRAP, current, 1); |
---|
552 | current->thread.trap_no = trapno; |
---|
553 | current->thread.error_code = error_code; |
---|
554 | return 0; |
---|
555 | } |
---|
556 | |
---|
557 | void handle_vm86_fault(struct kernel_vm86_regs * regs, long error_code) |
---|
558 | { |
---|
559 | unsigned char opcode; |
---|
560 | unsigned char __user *csp; |
---|
561 | unsigned char __user *ssp; |
---|
562 | unsigned short ip, sp, orig_flags; |
---|
563 | int data32, pref_done; |
---|
564 | |
---|
565 | #define CHECK_IF_IN_TRAP \ |
---|
566 | if (VMPI.vm86dbg_active && VMPI.vm86dbg_TFpendig) \ |
---|
567 | newflags |= TF_MASK |
---|
568 | #define VM86_FAULT_RETURN do { \ |
---|
569 | if (VMPI.force_return_for_pic && (VEFLAGS & (IF_MASK | VIF_MASK))) \ |
---|
570 | return_to_32bit(regs, VM86_PICRETURN); \ |
---|
571 | if (orig_flags & TF_MASK) \ |
---|
572 | handle_vm86_trap(regs, 0, 1); \ |
---|
573 | return; } while (0) |
---|
574 | |
---|
575 | orig_flags = *(unsigned short *)®s->eflags; |
---|
576 | |
---|
577 | csp = (unsigned char __user *) (regs->cs << 4); |
---|
578 | ssp = (unsigned char __user *) (regs->ss << 4); |
---|
579 | sp = SP(regs); |
---|
580 | ip = IP(regs); |
---|
581 | |
---|
582 | data32 = 0; |
---|
583 | pref_done = 0; |
---|
584 | do { |
---|
585 | switch (opcode = popb(csp, ip, simulate_sigsegv)) { |
---|
586 | case 0x66: /* 32-bit data */ data32=1; break; |
---|
587 | case 0x67: /* 32-bit address */ break; |
---|
588 | case 0x2e: /* CS */ break; |
---|
589 | case 0x3e: /* DS */ break; |
---|
590 | case 0x26: /* ES */ break; |
---|
591 | case 0x36: /* SS */ break; |
---|
592 | case 0x65: /* GS */ break; |
---|
593 | case 0x64: /* FS */ break; |
---|
594 | case 0xf2: /* repnz */ break; |
---|
595 | case 0xf3: /* rep */ break; |
---|
596 | default: pref_done = 1; |
---|
597 | } |
---|
598 | } while (!pref_done); |
---|
599 | |
---|
600 | switch (opcode) { |
---|
601 | |
---|
602 | /* pushf */ |
---|
603 | case 0x9c: |
---|
604 | if (data32) { |
---|
605 | pushl(ssp, sp, get_vflags(regs), simulate_sigsegv); |
---|
606 | SP(regs) -= 4; |
---|
607 | } else { |
---|
608 | pushw(ssp, sp, get_vflags(regs), simulate_sigsegv); |
---|
609 | SP(regs) -= 2; |
---|
610 | } |
---|
611 | IP(regs) = ip; |
---|
612 | VM86_FAULT_RETURN; |
---|
613 | |
---|
614 | /* popf */ |
---|
615 | case 0x9d: |
---|
616 | { |
---|
617 | unsigned long newflags; |
---|
618 | if (data32) { |
---|
619 | newflags=popl(ssp, sp, simulate_sigsegv); |
---|
620 | SP(regs) += 4; |
---|
621 | } else { |
---|
622 | newflags = popw(ssp, sp, simulate_sigsegv); |
---|
623 | SP(regs) += 2; |
---|
624 | } |
---|
625 | IP(regs) = ip; |
---|
626 | CHECK_IF_IN_TRAP; |
---|
627 | if (data32) { |
---|
628 | set_vflags_long(newflags, regs); |
---|
629 | } else { |
---|
630 | set_vflags_short(newflags, regs); |
---|
631 | } |
---|
632 | VM86_FAULT_RETURN; |
---|
633 | } |
---|
634 | |
---|
635 | /* int xx */ |
---|
636 | case 0xcd: { |
---|
637 | int intno=popb(csp, ip, simulate_sigsegv); |
---|
638 | IP(regs) = ip; |
---|
639 | if (VMPI.vm86dbg_active) { |
---|
640 | if ( (1 << (intno &7)) & VMPI.vm86dbg_intxxtab[intno >> 3] ) |
---|
641 | return_to_32bit(regs, VM86_INTx + (intno << 8)); |
---|
642 | } |
---|
643 | do_int(regs, intno, ssp, sp); |
---|
644 | return; |
---|
645 | } |
---|
646 | |
---|
647 | /* iret */ |
---|
648 | case 0xcf: |
---|
649 | { |
---|
650 | unsigned long newip; |
---|
651 | unsigned long newcs; |
---|
652 | unsigned long newflags; |
---|
653 | if (data32) { |
---|
654 | newip=popl(ssp, sp, simulate_sigsegv); |
---|
655 | newcs=popl(ssp, sp, simulate_sigsegv); |
---|
656 | newflags=popl(ssp, sp, simulate_sigsegv); |
---|
657 | SP(regs) += 12; |
---|
658 | } else { |
---|
659 | newip = popw(ssp, sp, simulate_sigsegv); |
---|
660 | newcs = popw(ssp, sp, simulate_sigsegv); |
---|
661 | newflags = popw(ssp, sp, simulate_sigsegv); |
---|
662 | SP(regs) += 6; |
---|
663 | } |
---|
664 | IP(regs) = newip; |
---|
665 | regs->cs = newcs; |
---|
666 | CHECK_IF_IN_TRAP; |
---|
667 | if (data32) { |
---|
668 | set_vflags_long(newflags, regs); |
---|
669 | } else { |
---|
670 | set_vflags_short(newflags, regs); |
---|
671 | } |
---|
672 | VM86_FAULT_RETURN; |
---|
673 | } |
---|
674 | |
---|
675 | /* cli */ |
---|
676 | case 0xfa: |
---|
677 | IP(regs) = ip; |
---|
678 | clear_IF(regs); |
---|
679 | VM86_FAULT_RETURN; |
---|
680 | |
---|
681 | /* sti */ |
---|
682 | /* |
---|
683 | * Damn. This is incorrect: the 'sti' instruction should actually |
---|
684 | * enable interrupts after the /next/ instruction. Not good. |
---|
685 | * |
---|
686 | * Probably needs some horsing around with the TF flag. Aiee.. |
---|
687 | */ |
---|
688 | case 0xfb: |
---|
689 | IP(regs) = ip; |
---|
690 | set_IF(regs); |
---|
691 | VM86_FAULT_RETURN; |
---|
692 | |
---|
693 | default: |
---|
694 | return_to_32bit(regs, VM86_UNKNOWN); |
---|
695 | } |
---|
696 | |
---|
697 | return; |
---|
698 | |
---|
699 | simulate_sigsegv: |
---|
700 | /* FIXME: After a long discussion with Stas we finally |
---|
701 | * agreed, that this is wrong. Here we should |
---|
702 | * really send a SIGSEGV to the user program. |
---|
703 | * But how do we create the correct context? We |
---|
704 | * are inside a general protection fault handler |
---|
705 | * and has just returned from a page fault handler. |
---|
706 | * The correct context for the signal handler |
---|
707 | * should be a mixture of the two, but how do we |
---|
708 | * get the information? [KD] |
---|
709 | */ |
---|
710 | return_to_32bit(regs, VM86_UNKNOWN); |
---|
711 | } |
---|
712 | |
---|
713 | /* ---------------- vm86 special IRQ passing stuff ----------------- */ |
---|
714 | |
---|
715 | #define VM86_IRQNAME "vm86irq" |
---|
716 | |
---|
717 | static struct vm86_irqs { |
---|
718 | struct task_struct *tsk; |
---|
719 | int sig; |
---|
720 | } vm86_irqs[16]; |
---|
721 | |
---|
722 | static DEFINE_SPINLOCK(irqbits_lock); |
---|
723 | static int irqbits; |
---|
724 | |
---|
725 | #define ALLOWED_SIGS ( 1 /* 0 = don't send a signal */ \ |
---|
726 | | (1 << SIGUSR1) | (1 << SIGUSR2) | (1 << SIGIO) | (1 << SIGURG) \ |
---|
727 | | (1 << SIGUNUSED) ) |
---|
728 | |
---|
729 | static irqreturn_t irq_handler(int intno, void *dev_id, struct pt_regs * regs) |
---|
730 | { |
---|
731 | int irq_bit; |
---|
732 | unsigned long flags; |
---|
733 | |
---|
734 | spin_lock_irqsave(&irqbits_lock, flags); |
---|
735 | irq_bit = 1 << intno; |
---|
736 | if ((irqbits & irq_bit) || ! vm86_irqs[intno].tsk) |
---|
737 | goto out; |
---|
738 | irqbits |= irq_bit; |
---|
739 | if (vm86_irqs[intno].sig) |
---|
740 | send_sig(vm86_irqs[intno].sig, vm86_irqs[intno].tsk, 1); |
---|
741 | /* |
---|
742 | * IRQ will be re-enabled when user asks for the irq (whether |
---|
743 | * polling or as a result of the signal) |
---|
744 | */ |
---|
745 | disable_irq_nosync(intno); |
---|
746 | spin_unlock_irqrestore(&irqbits_lock, flags); |
---|
747 | return IRQ_HANDLED; |
---|
748 | |
---|
749 | out: |
---|
750 | spin_unlock_irqrestore(&irqbits_lock, flags); |
---|
751 | return IRQ_NONE; |
---|
752 | } |
---|
753 | |
---|
754 | static inline void free_vm86_irq(int irqnumber) |
---|
755 | { |
---|
756 | unsigned long flags; |
---|
757 | |
---|
758 | free_irq(irqnumber, NULL); |
---|
759 | vm86_irqs[irqnumber].tsk = NULL; |
---|
760 | |
---|
761 | spin_lock_irqsave(&irqbits_lock, flags); |
---|
762 | irqbits &= ~(1 << irqnumber); |
---|
763 | spin_unlock_irqrestore(&irqbits_lock, flags); |
---|
764 | } |
---|
765 | |
---|
766 | void release_vm86_irqs(struct task_struct *task) |
---|
767 | { |
---|
768 | int i; |
---|
769 | for (i = FIRST_VM86_IRQ ; i <= LAST_VM86_IRQ; i++) |
---|
770 | if (vm86_irqs[i].tsk == task) |
---|
771 | free_vm86_irq(i); |
---|
772 | } |
---|
773 | |
---|
774 | static inline int get_and_reset_irq(int irqnumber) |
---|
775 | { |
---|
776 | int bit; |
---|
777 | unsigned long flags; |
---|
778 | int ret = 0; |
---|
779 | |
---|
780 | if (invalid_vm86_irq(irqnumber)) return 0; |
---|
781 | if (vm86_irqs[irqnumber].tsk != current) return 0; |
---|
782 | spin_lock_irqsave(&irqbits_lock, flags); |
---|
783 | bit = irqbits & (1 << irqnumber); |
---|
784 | irqbits &= ~bit; |
---|
785 | if (bit) { |
---|
786 | enable_irq(irqnumber); |
---|
787 | ret = 1; |
---|
788 | } |
---|
789 | |
---|
790 | spin_unlock_irqrestore(&irqbits_lock, flags); |
---|
791 | return ret; |
---|
792 | } |
---|
793 | |
---|
794 | |
---|
795 | static int do_vm86_irq_handling(int subfunction, int irqnumber) |
---|
796 | { |
---|
797 | int ret; |
---|
798 | switch (subfunction) { |
---|
799 | case VM86_GET_AND_RESET_IRQ: { |
---|
800 | return get_and_reset_irq(irqnumber); |
---|
801 | } |
---|
802 | case VM86_GET_IRQ_BITS: { |
---|
803 | return irqbits; |
---|
804 | } |
---|
805 | case VM86_REQUEST_IRQ: { |
---|
806 | int sig = irqnumber >> 8; |
---|
807 | int irq = irqnumber & 255; |
---|
808 | if (!capable(CAP_SYS_ADMIN)) return -EPERM; |
---|
809 | if (!((1 << sig) & ALLOWED_SIGS)) return -EPERM; |
---|
810 | if (invalid_vm86_irq(irq)) return -EPERM; |
---|
811 | if (vm86_irqs[irq].tsk) return -EPERM; |
---|
812 | ret = request_irq(irq, &irq_handler, 0, VM86_IRQNAME, NULL); |
---|
813 | if (ret) return ret; |
---|
814 | vm86_irqs[irq].sig = sig; |
---|
815 | vm86_irqs[irq].tsk = current; |
---|
816 | return irq; |
---|
817 | } |
---|
818 | case VM86_FREE_IRQ: { |
---|
819 | if (invalid_vm86_irq(irqnumber)) return -EPERM; |
---|
820 | if (!vm86_irqs[irqnumber].tsk) return 0; |
---|
821 | if (vm86_irqs[irqnumber].tsk != current) return -EPERM; |
---|
822 | free_vm86_irq(irqnumber); |
---|
823 | return 0; |
---|
824 | } |
---|
825 | } |
---|
826 | return -EINVAL; |
---|
827 | } |
---|
828 | |
---|