Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
1da177e4 LT |
2 | * Copyright (C) 1994 Linus Torvalds |
3 | * | |
4 | * 29 dec 2001 - Fixed oopses caused by unchecked access to the vm86 | |
624dffcb | 5 | * stack - Manfred Spraul <manfred@colorfullife.com> |
1da177e4 LT |
6 | * |
7 | * 22 mar 2002 - Manfred detected the stackfaults, but didn't handle | |
8 | * them correctly. Now the emulation will be in a | |
9 | * consistent state after stackfaults - Kasper Dupont | |
10 | * <kasperd@daimi.au.dk> | |
11 | * | |
12 | * 22 mar 2002 - Added missing clear_IF in set_vflags_* Kasper Dupont | |
13 | * <kasperd@daimi.au.dk> | |
14 | * | |
15 | * ?? ??? 2002 - Fixed premature returns from handle_vm86_fault | |
16 | * caused by Kasper Dupont's changes - Stas Sergeev | |
17 | * | |
18 | * 4 apr 2002 - Fixed CHECK_IF_IN_TRAP broken by Stas' changes. | |
19 | * Kasper Dupont <kasperd@daimi.au.dk> | |
20 | * | |
21 | * 9 apr 2002 - Changed syntax of macros in handle_vm86_fault. | |
22 | * Kasper Dupont <kasperd@daimi.au.dk> | |
23 | * | |
24 | * 9 apr 2002 - Changed stack access macros to jump to a label | |
25 | * instead of returning to userspace. This simplifies | |
26 | * do_int, and is needed by handle_vm6_fault. Kasper | |
27 | * Dupont <kasperd@daimi.au.dk> | |
28 | * | |
29 | */ | |
30 | ||
a9415644 | 31 | #include <linux/capability.h> |
1da177e4 LT |
32 | #include <linux/errno.h> |
33 | #include <linux/interrupt.h> | |
34 | #include <linux/sched.h> | |
35 | #include <linux/kernel.h> | |
36 | #include <linux/signal.h> | |
37 | #include <linux/string.h> | |
38 | #include <linux/mm.h> | |
39 | #include <linux/smp.h> | |
1da177e4 LT |
40 | #include <linux/highmem.h> |
41 | #include <linux/ptrace.h> | |
7e7f8a03 | 42 | #include <linux/audit.h> |
49d26b6e | 43 | #include <linux/stddef.h> |
1da177e4 LT |
44 | |
45 | #include <asm/uaccess.h> | |
46 | #include <asm/io.h> | |
47 | #include <asm/tlbflush.h> | |
48 | #include <asm/irq.h> | |
49 | ||
50 | /* | |
51 | * Known problems: | |
52 | * | |
53 | * Interrupt handling is not guaranteed: | |
54 | * - a real x86 will disable all interrupts for one instruction | |
55 | * after a "mov ss,xx" to make stack handling atomic even without | |
56 | * the 'lss' instruction. We can't guarantee this in v86 mode, | |
57 | * as the next instruction might result in a page fault or similar. | |
58 | * - a real x86 will have interrupts disabled for one instruction | |
59 | * past the 'sti' that enables them. We don't bother with all the | |
60 | * details yet. | |
61 | * | |
62 | * Let's hope these problems do not actually matter for anything. | |
63 | */ | |
64 | ||
65 | ||
66 | #define KVM86 ((struct kernel_vm86_struct *)regs) | |
83e714e8 | 67 | #define VMPI KVM86->vm86plus |
1da177e4 LT |
68 | |
69 | ||
70 | /* | |
71 | * 8- and 16-bit register defines.. | |
72 | */ | |
65ea5b03 PA |
73 | #define AL(regs) (((unsigned char *)&((regs)->pt.ax))[0]) |
74 | #define AH(regs) (((unsigned char *)&((regs)->pt.ax))[1]) | |
75 | #define IP(regs) (*(unsigned short *)&((regs)->pt.ip)) | |
76 | #define SP(regs) (*(unsigned short *)&((regs)->pt.sp)) | |
1da177e4 LT |
77 | |
78 | /* | |
79 | * virtual flags (16 and 32-bit versions) | |
80 | */ | |
81 | #define VFLAGS (*(unsigned short *)&(current->thread.v86flags)) | |
82 | #define VEFLAGS (current->thread.v86flags) | |
83 | ||
83e714e8 | 84 | #define set_flags(X, new, mask) \ |
1da177e4 LT |
85 | ((X) = ((X) & ~(mask)) | ((new) & (mask))) |
86 | ||
87 | #define SAFE_MASK (0xDD5) | |
88 | #define RETURN_MASK (0xDFF) | |
89 | ||
49d26b6e JF |
90 | /* convert kernel_vm86_regs to vm86_regs */ |
91 | static int copy_vm86_regs_to_user(struct vm86_regs __user *user, | |
92 | const struct kernel_vm86_regs *regs) | |
93 | { | |
94 | int ret = 0; | |
95 | ||
83e714e8 PC |
96 | /* |
97 | * kernel_vm86_regs is missing gs, so copy everything up to | |
98 | * (but not including) orig_eax, and then rest including orig_eax. | |
99 | */ | |
65ea5b03 PA |
100 | ret += copy_to_user(user, regs, offsetof(struct kernel_vm86_regs, pt.orig_ax)); |
101 | ret += copy_to_user(&user->orig_eax, ®s->pt.orig_ax, | |
49d26b6e | 102 | sizeof(struct kernel_vm86_regs) - |
65ea5b03 | 103 | offsetof(struct kernel_vm86_regs, pt.orig_ax)); |
49d26b6e JF |
104 | |
105 | return ret; | |
106 | } | |
107 | ||
108 | /* convert vm86_regs to kernel_vm86_regs */ | |
109 | static int copy_vm86_regs_from_user(struct kernel_vm86_regs *regs, | |
110 | const struct vm86_regs __user *user, | |
111 | unsigned extra) | |
112 | { | |
113 | int ret = 0; | |
114 | ||
65ea5b03 PA |
115 | /* copy ax-fs inclusive */ |
116 | ret += copy_from_user(regs, user, offsetof(struct kernel_vm86_regs, pt.orig_ax)); | |
117 | /* copy orig_ax-__gsh+extra */ | |
118 | ret += copy_from_user(®s->pt.orig_ax, &user->orig_eax, | |
49d26b6e | 119 | sizeof(struct kernel_vm86_regs) - |
65ea5b03 | 120 | offsetof(struct kernel_vm86_regs, pt.orig_ax) + |
49d26b6e | 121 | extra); |
49d26b6e JF |
122 | return ret; |
123 | } | |
1da177e4 | 124 | |
83e714e8 | 125 | struct pt_regs *save_v86_state(struct kernel_vm86_regs *regs) |
1da177e4 LT |
126 | { |
127 | struct tss_struct *tss; | |
128 | struct pt_regs *ret; | |
129 | unsigned long tmp; | |
130 | ||
131 | /* | |
132 | * This gets called from entry.S with interrupts disabled, but | |
133 | * from process context. Enable interrupts here, before trying | |
134 | * to access user space. | |
135 | */ | |
136 | local_irq_enable(); | |
137 | ||
138 | if (!current->thread.vm86_info) { | |
139 | printk("no vm86_info: BAD\n"); | |
140 | do_exit(SIGSEGV); | |
141 | } | |
a5c15d41 | 142 | set_flags(regs->pt.flags, VEFLAGS, X86_EFLAGS_VIF | current->thread.v86mask); |
83e714e8 PC |
143 | tmp = copy_vm86_regs_to_user(¤t->thread.vm86_info->regs, regs); |
144 | tmp += put_user(current->thread.screen_bitmap, ¤t->thread.vm86_info->screen_bitmap); | |
1da177e4 LT |
145 | if (tmp) { |
146 | printk("vm86: could not access userspace vm86_info\n"); | |
147 | do_exit(SIGSEGV); | |
148 | } | |
149 | ||
150 | tss = &per_cpu(init_tss, get_cpu()); | |
faca6227 | 151 | current->thread.sp0 = current->thread.saved_sp0; |
1da177e4 | 152 | current->thread.sysenter_cs = __KERNEL_CS; |
faca6227 PA |
153 | load_sp0(tss, ¤t->thread); |
154 | current->thread.saved_sp0 = 0; | |
1da177e4 LT |
155 | put_cpu(); |
156 | ||
1da177e4 | 157 | ret = KVM86->regs32; |
49d26b6e | 158 | |
65ea5b03 | 159 | ret->fs = current->thread.saved_fs; |
464d1a78 | 160 | loadsegment(gs, current->thread.saved_gs); |
49d26b6e | 161 | |
1da177e4 LT |
162 | return ret; |
163 | } | |
164 | ||
60ec5585 | 165 | static void mark_screen_rdonly(struct mm_struct *mm) |
1da177e4 LT |
166 | { |
167 | pgd_t *pgd; | |
168 | pud_t *pud; | |
169 | pmd_t *pmd; | |
60ec5585 HD |
170 | pte_t *pte; |
171 | spinlock_t *ptl; | |
1da177e4 LT |
172 | int i; |
173 | ||
60ec5585 | 174 | pgd = pgd_offset(mm, 0xA0000); |
1da177e4 LT |
175 | if (pgd_none_or_clear_bad(pgd)) |
176 | goto out; | |
177 | pud = pud_offset(pgd, 0xA0000); | |
178 | if (pud_none_or_clear_bad(pud)) | |
179 | goto out; | |
180 | pmd = pmd_offset(pud, 0xA0000); | |
181 | if (pmd_none_or_clear_bad(pmd)) | |
182 | goto out; | |
60ec5585 | 183 | pte = pte_offset_map_lock(mm, pmd, 0xA0000, &ptl); |
1da177e4 LT |
184 | for (i = 0; i < 32; i++) { |
185 | if (pte_present(*pte)) | |
186 | set_pte(pte, pte_wrprotect(*pte)); | |
187 | pte++; | |
188 | } | |
60ec5585 | 189 | pte_unmap_unlock(pte, ptl); |
1da177e4 | 190 | out: |
1da177e4 LT |
191 | flush_tlb(); |
192 | } | |
193 | ||
194 | ||
195 | ||
196 | static int do_vm86_irq_handling(int subfunction, int irqnumber); | |
197 | static void do_sys_vm86(struct kernel_vm86_struct *info, struct task_struct *tsk); | |
198 | ||
199 | asmlinkage int sys_vm86old(struct pt_regs regs) | |
200 | { | |
65ea5b03 | 201 | struct vm86_struct __user *v86 = (struct vm86_struct __user *)regs.bx; |
1da177e4 LT |
202 | struct kernel_vm86_struct info; /* declare this _on top_, |
203 | * this avoids wasting of stack space. | |
204 | * This remains on the stack until we | |
205 | * return to 32 bit user space. | |
206 | */ | |
207 | struct task_struct *tsk; | |
208 | int tmp, ret = -EPERM; | |
209 | ||
210 | tsk = current; | |
faca6227 | 211 | if (tsk->thread.saved_sp0) |
1da177e4 | 212 | goto out; |
49d26b6e JF |
213 | tmp = copy_vm86_regs_from_user(&info.regs, &v86->regs, |
214 | offsetof(struct kernel_vm86_struct, vm86plus) - | |
215 | sizeof(info.regs)); | |
1da177e4 LT |
216 | ret = -EFAULT; |
217 | if (tmp) | |
218 | goto out; | |
219 | memset(&info.vm86plus, 0, (int)&info.regs32 - (int)&info.vm86plus); | |
220 | info.regs32 = ®s; | |
221 | tsk->thread.vm86_info = v86; | |
222 | do_sys_vm86(&info, tsk); | |
223 | ret = 0; /* we never return here */ | |
224 | out: | |
225 | return ret; | |
226 | } | |
227 | ||
228 | ||
229 | asmlinkage int sys_vm86(struct pt_regs regs) | |
230 | { | |
231 | struct kernel_vm86_struct info; /* declare this _on top_, | |
232 | * this avoids wasting of stack space. | |
233 | * This remains on the stack until we | |
234 | * return to 32 bit user space. | |
235 | */ | |
236 | struct task_struct *tsk; | |
237 | int tmp, ret; | |
238 | struct vm86plus_struct __user *v86; | |
239 | ||
240 | tsk = current; | |
65ea5b03 | 241 | switch (regs.bx) { |
83e714e8 PC |
242 | case VM86_REQUEST_IRQ: |
243 | case VM86_FREE_IRQ: | |
244 | case VM86_GET_IRQ_BITS: | |
245 | case VM86_GET_AND_RESET_IRQ: | |
246 | ret = do_vm86_irq_handling(regs.bx, (int)regs.cx); | |
247 | goto out; | |
248 | case VM86_PLUS_INSTALL_CHECK: | |
249 | /* | |
250 | * NOTE: on old vm86 stuff this will return the error | |
251 | * from access_ok(), because the subfunction is | |
252 | * interpreted as (invalid) address to vm86_struct. | |
253 | * So the installation check works. | |
254 | */ | |
255 | ret = 0; | |
256 | goto out; | |
1da177e4 LT |
257 | } |
258 | ||
259 | /* we come here only for functions VM86_ENTER, VM86_ENTER_NO_BYPASS */ | |
260 | ret = -EPERM; | |
faca6227 | 261 | if (tsk->thread.saved_sp0) |
1da177e4 | 262 | goto out; |
65ea5b03 | 263 | v86 = (struct vm86plus_struct __user *)regs.cx; |
49d26b6e JF |
264 | tmp = copy_vm86_regs_from_user(&info.regs, &v86->regs, |
265 | offsetof(struct kernel_vm86_struct, regs32) - | |
266 | sizeof(info.regs)); | |
1da177e4 LT |
267 | ret = -EFAULT; |
268 | if (tmp) | |
269 | goto out; | |
270 | info.regs32 = ®s; | |
271 | info.vm86plus.is_vm86pus = 1; | |
272 | tsk->thread.vm86_info = (struct vm86_struct __user *)v86; | |
273 | do_sys_vm86(&info, tsk); | |
274 | ret = 0; /* we never return here */ | |
275 | out: | |
276 | return ret; | |
277 | } | |
278 | ||
279 | ||
280 | static void do_sys_vm86(struct kernel_vm86_struct *info, struct task_struct *tsk) | |
281 | { | |
282 | struct tss_struct *tss; | |
283 | /* | |
284 | * make sure the vm86() system call doesn't try to do anything silly | |
285 | */ | |
65ea5b03 PA |
286 | info->regs.pt.ds = 0; |
287 | info->regs.pt.es = 0; | |
288 | info->regs.pt.fs = 0; | |
1da177e4 | 289 | |
464d1a78 | 290 | /* we are clearing gs later just before "jmp resume_userspace", |
49d26b6e | 291 | * because it is not saved/restored. |
1da177e4 LT |
292 | */ |
293 | ||
294 | /* | |
65ea5b03 | 295 | * The flags register is also special: we cannot trust that the user |
1da177e4 LT |
296 | * has set it up safely, so this makes sure interrupt etc flags are |
297 | * inherited from protected mode. | |
298 | */ | |
65ea5b03 PA |
299 | VEFLAGS = info->regs.pt.flags; |
300 | info->regs.pt.flags &= SAFE_MASK; | |
301 | info->regs.pt.flags |= info->regs32->flags & ~SAFE_MASK; | |
6b6891f9 | 302 | info->regs.pt.flags |= X86_VM_MASK; |
1da177e4 LT |
303 | |
304 | switch (info->cpu_type) { | |
83e714e8 PC |
305 | case CPU_286: |
306 | tsk->thread.v86mask = 0; | |
307 | break; | |
308 | case CPU_386: | |
a5c15d41 | 309 | tsk->thread.v86mask = X86_EFLAGS_NT | X86_EFLAGS_IOPL; |
83e714e8 PC |
310 | break; |
311 | case CPU_486: | |
a5c15d41 | 312 | tsk->thread.v86mask = X86_EFLAGS_AC | X86_EFLAGS_NT | X86_EFLAGS_IOPL; |
83e714e8 PC |
313 | break; |
314 | default: | |
a5c15d41 | 315 | tsk->thread.v86mask = X86_EFLAGS_ID | X86_EFLAGS_AC | X86_EFLAGS_NT | X86_EFLAGS_IOPL; |
83e714e8 | 316 | break; |
1da177e4 LT |
317 | } |
318 | ||
319 | /* | |
65ea5b03 | 320 | * Save old state, set default return value (%ax) to 0 |
1da177e4 | 321 | */ |
65ea5b03 | 322 | info->regs32->ax = 0; |
faca6227 | 323 | tsk->thread.saved_sp0 = tsk->thread.sp0; |
65ea5b03 | 324 | tsk->thread.saved_fs = info->regs32->fs; |
464d1a78 | 325 | savesegment(gs, tsk->thread.saved_gs); |
1da177e4 LT |
326 | |
327 | tss = &per_cpu(init_tss, get_cpu()); | |
faca6227 | 328 | tsk->thread.sp0 = (unsigned long) &info->VM86_TSS_ESP0; |
1da177e4 LT |
329 | if (cpu_has_sep) |
330 | tsk->thread.sysenter_cs = 0; | |
faca6227 | 331 | load_sp0(tss, &tsk->thread); |
1da177e4 LT |
332 | put_cpu(); |
333 | ||
334 | tsk->thread.screen_bitmap = info->screen_bitmap; | |
335 | if (info->flags & VM86_SCREEN_BITMAP) | |
60ec5585 | 336 | mark_screen_rdonly(tsk->mm); |
7e7f8a03 JB |
337 | |
338 | /*call audit_syscall_exit since we do not exit via the normal paths */ | |
339 | if (unlikely(current->audit_context)) | |
49d26b6e | 340 | audit_syscall_exit(AUDITSC_RESULT(0), 0); |
7e7f8a03 | 341 | |
1da177e4 | 342 | __asm__ __volatile__( |
1da177e4 LT |
343 | "movl %0,%%esp\n\t" |
344 | "movl %1,%%ebp\n\t" | |
464d1a78 | 345 | "mov %2, %%gs\n\t" |
1da177e4 LT |
346 | "jmp resume_userspace" |
347 | : /* no outputs */ | |
49d26b6e | 348 | :"r" (&info->regs), "r" (task_thread_info(tsk)), "r" (0)); |
1da177e4 LT |
349 | /* we never return here */ |
350 | } | |
351 | ||
83e714e8 | 352 | static inline void return_to_32bit(struct kernel_vm86_regs *regs16, int retval) |
1da177e4 | 353 | { |
83e714e8 | 354 | struct pt_regs *regs32; |
1da177e4 LT |
355 | |
356 | regs32 = save_v86_state(regs16); | |
65ea5b03 | 357 | regs32->ax = retval; |
1da177e4 LT |
358 | __asm__ __volatile__("movl %0,%%esp\n\t" |
359 | "movl %1,%%ebp\n\t" | |
360 | "jmp resume_userspace" | |
361 | : : "r" (regs32), "r" (current_thread_info())); | |
362 | } | |
363 | ||
83e714e8 | 364 | static inline void set_IF(struct kernel_vm86_regs *regs) |
1da177e4 | 365 | { |
a5c15d41 | 366 | VEFLAGS |= X86_EFLAGS_VIF; |
367 | if (VEFLAGS & X86_EFLAGS_VIP) | |
1da177e4 LT |
368 | return_to_32bit(regs, VM86_STI); |
369 | } | |
370 | ||
83e714e8 | 371 | static inline void clear_IF(struct kernel_vm86_regs *regs) |
1da177e4 | 372 | { |
a5c15d41 | 373 | VEFLAGS &= ~X86_EFLAGS_VIF; |
1da177e4 LT |
374 | } |
375 | ||
83e714e8 | 376 | static inline void clear_TF(struct kernel_vm86_regs *regs) |
1da177e4 | 377 | { |
a5c15d41 | 378 | regs->pt.flags &= ~X86_EFLAGS_TF; |
1da177e4 LT |
379 | } |
380 | ||
83e714e8 | 381 | static inline void clear_AC(struct kernel_vm86_regs *regs) |
1da177e4 | 382 | { |
a5c15d41 | 383 | regs->pt.flags &= ~X86_EFLAGS_AC; |
1da177e4 LT |
384 | } |
385 | ||
83e714e8 PC |
386 | /* |
387 | * It is correct to call set_IF(regs) from the set_vflags_* | |
1da177e4 LT |
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 | |
ab4a574e | 391 | * end up with interrupts disabled, but you ended up with |
1da177e4 LT |
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 | ||
83e714e8 | 398 | static inline void set_vflags_long(unsigned long flags, struct kernel_vm86_regs *regs) |
1da177e4 | 399 | { |
65ea5b03 PA |
400 | set_flags(VEFLAGS, flags, current->thread.v86mask); |
401 | set_flags(regs->pt.flags, flags, SAFE_MASK); | |
a5c15d41 | 402 | if (flags & X86_EFLAGS_IF) |
1da177e4 LT |
403 | set_IF(regs); |
404 | else | |
405 | clear_IF(regs); | |
406 | } | |
407 | ||
83e714e8 | 408 | static inline void set_vflags_short(unsigned short flags, struct kernel_vm86_regs *regs) |
1da177e4 LT |
409 | { |
410 | set_flags(VFLAGS, flags, current->thread.v86mask); | |
65ea5b03 | 411 | set_flags(regs->pt.flags, flags, SAFE_MASK); |
a5c15d41 | 412 | if (flags & X86_EFLAGS_IF) |
1da177e4 LT |
413 | set_IF(regs); |
414 | else | |
415 | clear_IF(regs); | |
416 | } | |
417 | ||
83e714e8 | 418 | static inline unsigned long get_vflags(struct kernel_vm86_regs *regs) |
1da177e4 | 419 | { |
65ea5b03 | 420 | unsigned long flags = regs->pt.flags & RETURN_MASK; |
1da177e4 | 421 | |
a5c15d41 | 422 | if (VEFLAGS & X86_EFLAGS_VIF) |
423 | flags |= X86_EFLAGS_IF; | |
424 | flags |= X86_EFLAGS_IOPL; | |
1da177e4 LT |
425 | return flags | (VEFLAGS & current->thread.v86mask); |
426 | } | |
427 | ||
83e714e8 | 428 | static inline int is_revectored(int nr, struct revectored_struct *bitmap) |
1da177e4 LT |
429 | { |
430 | __asm__ __volatile__("btl %2,%1\n\tsbbl %0,%0" | |
431 | :"=r" (nr) | |
83e714e8 | 432 | :"m" (*bitmap), "r" (nr)); |
1da177e4 LT |
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; \ | |
83e714e8 | 444 | } while (0) |
1da177e4 LT |
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; \ | |
83e714e8 | 455 | } while (0) |
1da177e4 LT |
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; \ | |
83e714e8 | 472 | } while (0) |
1da177e4 LT |
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, | |
83e714e8 | 519 | unsigned char __user *ssp, unsigned short sp) |
1da177e4 LT |
520 | { |
521 | unsigned long __user *intr_ptr; | |
522 | unsigned long segoffs; | |
523 | ||
65ea5b03 | 524 | if (regs->pt.cs == BIOSSEG) |
1da177e4 LT |
525 | goto cannot_handle; |
526 | if (is_revectored(i, &KVM86->int_revectored)) | |
527 | goto cannot_handle; | |
83e714e8 | 528 | if (i == 0x21 && is_revectored(AH(regs), &KVM86->int21_revectored)) |
1da177e4 LT |
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); | |
65ea5b03 | 536 | pushw(ssp, sp, regs->pt.cs, cannot_handle); |
1da177e4 | 537 | pushw(ssp, sp, IP(regs), cannot_handle); |
65ea5b03 | 538 | regs->pt.cs = 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 | ||
83e714e8 | 550 | int handle_vm86_trap(struct kernel_vm86_regs *regs, long error_code, int trapno) |
1da177e4 LT |
551 | { |
552 | if (VMPI.is_vm86pus) { | |
83e714e8 | 553 | if ((trapno == 3) || (trapno == 1)) |
1da177e4 | 554 | return_to_32bit(regs, VM86_TRAP + (trapno << 8)); |
65ea5b03 | 555 | do_int(regs, trapno, (unsigned char __user *) (regs->pt.ss << 4), SP(regs)); |
1da177e4 LT |
556 | return 0; |
557 | } | |
83e714e8 | 558 | if (trapno != 1) |
1da177e4 | 559 | return 1; /* we let this handle by the calling routine */ |
1da177e4 LT |
560 | current->thread.trap_no = trapno; |
561 | current->thread.error_code = error_code; | |
0f540910 | 562 | force_sig(SIGTRAP, current); |
1da177e4 LT |
563 | return 0; |
564 | } | |
565 | ||
83e714e8 | 566 | void handle_vm86_fault(struct kernel_vm86_regs *regs, long error_code) |
1da177e4 LT |
567 | { |
568 | unsigned char opcode; | |
569 | unsigned char __user *csp; | |
570 | unsigned char __user *ssp; | |
5fd75ebb | 571 | unsigned short ip, sp, orig_flags; |
1da177e4 LT |
572 | int data32, pref_done; |
573 | ||
574 | #define CHECK_IF_IN_TRAP \ | |
575 | if (VMPI.vm86dbg_active && VMPI.vm86dbg_TFpendig) \ | |
a5c15d41 | 576 | newflags |= X86_EFLAGS_TF |
1da177e4 | 577 | #define VM86_FAULT_RETURN do { \ |
a5c15d41 | 578 | if (VMPI.force_return_for_pic && (VEFLAGS & (X86_EFLAGS_IF | X86_EFLAGS_VIF))) \ |
1da177e4 | 579 | return_to_32bit(regs, VM86_PICRETURN); \ |
a5c15d41 | 580 | if (orig_flags & X86_EFLAGS_TF) \ |
5fd75ebb | 581 | handle_vm86_trap(regs, 0, 1); \ |
1da177e4 LT |
582 | return; } while (0) |
583 | ||
65ea5b03 | 584 | orig_flags = *(unsigned short *)®s->pt.flags; |
5fd75ebb | 585 | |
65ea5b03 PA |
586 | csp = (unsigned char __user *) (regs->pt.cs << 4); |
587 | ssp = (unsigned char __user *) (regs->pt.ss << 4); | |
1da177e4 LT |
588 | sp = SP(regs); |
589 | ip = IP(regs); | |
590 | ||
591 | data32 = 0; | |
592 | pref_done = 0; | |
593 | do { | |
594 | switch (opcode = popb(csp, ip, simulate_sigsegv)) { | |
83e714e8 PC |
595 | case 0x66: /* 32-bit data */ data32 = 1; break; |
596 | case 0x67: /* 32-bit address */ break; | |
597 | case 0x2e: /* CS */ break; | |
598 | case 0x3e: /* DS */ break; | |
599 | case 0x26: /* ES */ break; | |
600 | case 0x36: /* SS */ break; | |
601 | case 0x65: /* GS */ break; | |
602 | case 0x64: /* FS */ break; | |
603 | case 0xf2: /* repnz */ break; | |
604 | case 0xf3: /* rep */ break; | |
605 | default: pref_done = 1; | |
1da177e4 LT |
606 | } |
607 | } while (!pref_done); | |
608 | ||
609 | switch (opcode) { | |
610 | ||
611 | /* pushf */ | |
612 | case 0x9c: | |
613 | if (data32) { | |
614 | pushl(ssp, sp, get_vflags(regs), simulate_sigsegv); | |
615 | SP(regs) -= 4; | |
616 | } else { | |
617 | pushw(ssp, sp, get_vflags(regs), simulate_sigsegv); | |
618 | SP(regs) -= 2; | |
619 | } | |
620 | IP(regs) = ip; | |
621 | VM86_FAULT_RETURN; | |
622 | ||
623 | /* popf */ | |
624 | case 0x9d: | |
625 | { | |
626 | unsigned long newflags; | |
627 | if (data32) { | |
83e714e8 | 628 | newflags = popl(ssp, sp, simulate_sigsegv); |
1da177e4 LT |
629 | SP(regs) += 4; |
630 | } else { | |
631 | newflags = popw(ssp, sp, simulate_sigsegv); | |
632 | SP(regs) += 2; | |
633 | } | |
634 | IP(regs) = ip; | |
635 | CHECK_IF_IN_TRAP; | |
83e714e8 | 636 | if (data32) |
1da177e4 | 637 | set_vflags_long(newflags, regs); |
83e714e8 | 638 | else |
1da177e4 | 639 | set_vflags_short(newflags, regs); |
83e714e8 | 640 | |
1da177e4 LT |
641 | VM86_FAULT_RETURN; |
642 | } | |
643 | ||
644 | /* int xx */ | |
645 | case 0xcd: { | |
83e714e8 | 646 | int intno = popb(csp, ip, simulate_sigsegv); |
1da177e4 LT |
647 | IP(regs) = ip; |
648 | if (VMPI.vm86dbg_active) { | |
83e714e8 | 649 | if ((1 << (intno & 7)) & VMPI.vm86dbg_intxxtab[intno >> 3]) |
1da177e4 LT |
650 | return_to_32bit(regs, VM86_INTx + (intno << 8)); |
651 | } | |
652 | do_int(regs, intno, ssp, sp); | |
653 | return; | |
654 | } | |
655 | ||
656 | /* iret */ | |
657 | case 0xcf: | |
658 | { | |
659 | unsigned long newip; | |
660 | unsigned long newcs; | |
661 | unsigned long newflags; | |
662 | if (data32) { | |
83e714e8 PC |
663 | newip = popl(ssp, sp, simulate_sigsegv); |
664 | newcs = popl(ssp, sp, simulate_sigsegv); | |
665 | newflags = popl(ssp, sp, simulate_sigsegv); | |
1da177e4 LT |
666 | SP(regs) += 12; |
667 | } else { | |
668 | newip = popw(ssp, sp, simulate_sigsegv); | |
669 | newcs = popw(ssp, sp, simulate_sigsegv); | |
670 | newflags = popw(ssp, sp, simulate_sigsegv); | |
671 | SP(regs) += 6; | |
672 | } | |
673 | IP(regs) = newip; | |
65ea5b03 | 674 | regs->pt.cs = newcs; |
1da177e4 LT |
675 | CHECK_IF_IN_TRAP; |
676 | if (data32) { | |
677 | set_vflags_long(newflags, regs); | |
678 | } else { | |
679 | set_vflags_short(newflags, regs); | |
680 | } | |
681 | VM86_FAULT_RETURN; | |
682 | } | |
683 | ||
684 | /* cli */ | |
685 | case 0xfa: | |
686 | IP(regs) = ip; | |
687 | clear_IF(regs); | |
688 | VM86_FAULT_RETURN; | |
689 | ||
690 | /* sti */ | |
691 | /* | |
692 | * Damn. This is incorrect: the 'sti' instruction should actually | |
693 | * enable interrupts after the /next/ instruction. Not good. | |
694 | * | |
695 | * Probably needs some horsing around with the TF flag. Aiee.. | |
696 | */ | |
697 | case 0xfb: | |
698 | IP(regs) = ip; | |
699 | set_IF(regs); | |
700 | VM86_FAULT_RETURN; | |
701 | ||
702 | default: | |
703 | return_to_32bit(regs, VM86_UNKNOWN); | |
704 | } | |
705 | ||
706 | return; | |
707 | ||
708 | simulate_sigsegv: | |
709 | /* FIXME: After a long discussion with Stas we finally | |
710 | * agreed, that this is wrong. Here we should | |
711 | * really send a SIGSEGV to the user program. | |
712 | * But how do we create the correct context? We | |
713 | * are inside a general protection fault handler | |
714 | * and has just returned from a page fault handler. | |
715 | * The correct context for the signal handler | |
716 | * should be a mixture of the two, but how do we | |
717 | * get the information? [KD] | |
718 | */ | |
719 | return_to_32bit(regs, VM86_UNKNOWN); | |
720 | } | |
721 | ||
722 | /* ---------------- vm86 special IRQ passing stuff ----------------- */ | |
723 | ||
724 | #define VM86_IRQNAME "vm86irq" | |
725 | ||
726 | static struct vm86_irqs { | |
727 | struct task_struct *tsk; | |
728 | int sig; | |
729 | } vm86_irqs[16]; | |
730 | ||
731 | static DEFINE_SPINLOCK(irqbits_lock); | |
732 | static int irqbits; | |
733 | ||
83e714e8 | 734 | #define ALLOWED_SIGS (1 /* 0 = don't send a signal */ \ |
1da177e4 | 735 | | (1 << SIGUSR1) | (1 << SIGUSR2) | (1 << SIGIO) | (1 << SIGURG) \ |
83e714e8 PC |
736 | | (1 << SIGUNUSED)) |
737 | ||
7d12e780 | 738 | static irqreturn_t irq_handler(int intno, void *dev_id) |
1da177e4 LT |
739 | { |
740 | int irq_bit; | |
741 | unsigned long flags; | |
742 | ||
83e714e8 | 743 | spin_lock_irqsave(&irqbits_lock, flags); |
1da177e4 | 744 | irq_bit = 1 << intno; |
83e714e8 | 745 | if ((irqbits & irq_bit) || !vm86_irqs[intno].tsk) |
1da177e4 LT |
746 | goto out; |
747 | irqbits |= irq_bit; | |
748 | if (vm86_irqs[intno].sig) | |
749 | send_sig(vm86_irqs[intno].sig, vm86_irqs[intno].tsk, 1); | |
1da177e4 LT |
750 | /* |
751 | * IRQ will be re-enabled when user asks for the irq (whether | |
752 | * polling or as a result of the signal) | |
753 | */ | |
ad671423 PP |
754 | disable_irq_nosync(intno); |
755 | spin_unlock_irqrestore(&irqbits_lock, flags); | |
1da177e4 LT |
756 | return IRQ_HANDLED; |
757 | ||
758 | out: | |
83e714e8 | 759 | spin_unlock_irqrestore(&irqbits_lock, flags); |
1da177e4 LT |
760 | return IRQ_NONE; |
761 | } | |
762 | ||
763 | static inline void free_vm86_irq(int irqnumber) | |
764 | { | |
765 | unsigned long flags; | |
766 | ||
767 | free_irq(irqnumber, NULL); | |
768 | vm86_irqs[irqnumber].tsk = NULL; | |
769 | ||
83e714e8 | 770 | spin_lock_irqsave(&irqbits_lock, flags); |
1da177e4 | 771 | irqbits &= ~(1 << irqnumber); |
83e714e8 | 772 | spin_unlock_irqrestore(&irqbits_lock, flags); |
1da177e4 LT |
773 | } |
774 | ||
775 | void release_vm86_irqs(struct task_struct *task) | |
776 | { | |
777 | int i; | |
778 | for (i = FIRST_VM86_IRQ ; i <= LAST_VM86_IRQ; i++) | |
779 | if (vm86_irqs[i].tsk == task) | |
780 | free_vm86_irq(i); | |
781 | } | |
782 | ||
783 | static inline int get_and_reset_irq(int irqnumber) | |
784 | { | |
785 | int bit; | |
786 | unsigned long flags; | |
ad671423 | 787 | int ret = 0; |
83e714e8 | 788 | |
1da177e4 LT |
789 | if (invalid_vm86_irq(irqnumber)) return 0; |
790 | if (vm86_irqs[irqnumber].tsk != current) return 0; | |
83e714e8 | 791 | spin_lock_irqsave(&irqbits_lock, flags); |
1da177e4 LT |
792 | bit = irqbits & (1 << irqnumber); |
793 | irqbits &= ~bit; | |
ad671423 PP |
794 | if (bit) { |
795 | enable_irq(irqnumber); | |
796 | ret = 1; | |
797 | } | |
798 | ||
83e714e8 | 799 | spin_unlock_irqrestore(&irqbits_lock, flags); |
ad671423 | 800 | return ret; |
1da177e4 LT |
801 | } |
802 | ||
803 | ||
804 | static int do_vm86_irq_handling(int subfunction, int irqnumber) | |
805 | { | |
806 | int ret; | |
807 | switch (subfunction) { | |
808 | case VM86_GET_AND_RESET_IRQ: { | |
809 | return get_and_reset_irq(irqnumber); | |
810 | } | |
811 | case VM86_GET_IRQ_BITS: { | |
812 | return irqbits; | |
813 | } | |
814 | case VM86_REQUEST_IRQ: { | |
815 | int sig = irqnumber >> 8; | |
816 | int irq = irqnumber & 255; | |
817 | if (!capable(CAP_SYS_ADMIN)) return -EPERM; | |
818 | if (!((1 << sig) & ALLOWED_SIGS)) return -EPERM; | |
819 | if (invalid_vm86_irq(irq)) return -EPERM; | |
820 | if (vm86_irqs[irq].tsk) return -EPERM; | |
821 | ret = request_irq(irq, &irq_handler, 0, VM86_IRQNAME, NULL); | |
822 | if (ret) return ret; | |
823 | vm86_irqs[irq].sig = sig; | |
824 | vm86_irqs[irq].tsk = current; | |
825 | return irq; | |
826 | } | |
827 | case VM86_FREE_IRQ: { | |
828 | if (invalid_vm86_irq(irqnumber)) return -EPERM; | |
829 | if (!vm86_irqs[irqnumber].tsk) return 0; | |
830 | if (vm86_irqs[irqnumber].tsk != current) return -EPERM; | |
831 | free_vm86_irq(irqnumber); | |
832 | return 0; | |
833 | } | |
834 | } | |
835 | return -EINVAL; | |
836 | } | |
837 |