]>
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> | |
bbc1f698 | 49 | #include <asm/syscalls.h> |
1da177e4 LT |
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) | |
83e714e8 | 68 | #define VMPI KVM86->vm86plus |
1da177e4 LT |
69 | |
70 | ||
71 | /* | |
72 | * 8- and 16-bit register defines.. | |
73 | */ | |
65ea5b03 PA |
74 | #define AL(regs) (((unsigned char *)&((regs)->pt.ax))[0]) |
75 | #define AH(regs) (((unsigned char *)&((regs)->pt.ax))[1]) | |
76 | #define IP(regs) (*(unsigned short *)&((regs)->pt.ip)) | |
77 | #define SP(regs) (*(unsigned short *)&((regs)->pt.sp)) | |
1da177e4 LT |
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 | ||
83e714e8 | 85 | #define set_flags(X, new, mask) \ |
1da177e4 LT |
86 | ((X) = ((X) & ~(mask)) | ((new) & (mask))) |
87 | ||
88 | #define SAFE_MASK (0xDD5) | |
89 | #define RETURN_MASK (0xDFF) | |
90 | ||
49d26b6e JF |
91 | /* convert kernel_vm86_regs to vm86_regs */ |
92 | static int copy_vm86_regs_to_user(struct vm86_regs __user *user, | |
93 | const struct kernel_vm86_regs *regs) | |
94 | { | |
95 | int ret = 0; | |
96 | ||
83e714e8 PC |
97 | /* |
98 | * kernel_vm86_regs is missing gs, so copy everything up to | |
99 | * (but not including) orig_eax, and then rest including orig_eax. | |
100 | */ | |
65ea5b03 PA |
101 | ret += copy_to_user(user, regs, offsetof(struct kernel_vm86_regs, pt.orig_ax)); |
102 | ret += copy_to_user(&user->orig_eax, ®s->pt.orig_ax, | |
49d26b6e | 103 | sizeof(struct kernel_vm86_regs) - |
65ea5b03 | 104 | offsetof(struct kernel_vm86_regs, pt.orig_ax)); |
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 | ||
65ea5b03 PA |
116 | /* copy ax-fs inclusive */ |
117 | ret += copy_from_user(regs, user, offsetof(struct kernel_vm86_regs, pt.orig_ax)); | |
118 | /* copy orig_ax-__gsh+extra */ | |
119 | ret += copy_from_user(®s->pt.orig_ax, &user->orig_eax, | |
49d26b6e | 120 | sizeof(struct kernel_vm86_regs) - |
65ea5b03 | 121 | offsetof(struct kernel_vm86_regs, pt.orig_ax) + |
49d26b6e | 122 | extra); |
49d26b6e JF |
123 | return ret; |
124 | } | |
1da177e4 | 125 | |
83e714e8 | 126 | struct pt_regs *save_v86_state(struct kernel_vm86_regs *regs) |
1da177e4 LT |
127 | { |
128 | struct tss_struct *tss; | |
129 | struct pt_regs *ret; | |
130 | unsigned long tmp; | |
131 | ||
132 | /* | |
133 | * This gets called from entry.S with interrupts disabled, but | |
134 | * from process context. Enable interrupts here, before trying | |
135 | * to access user space. | |
136 | */ | |
137 | local_irq_enable(); | |
138 | ||
139 | if (!current->thread.vm86_info) { | |
140 | printk("no vm86_info: BAD\n"); | |
141 | do_exit(SIGSEGV); | |
142 | } | |
a5c15d41 | 143 | set_flags(regs->pt.flags, VEFLAGS, X86_EFLAGS_VIF | current->thread.v86mask); |
83e714e8 PC |
144 | tmp = copy_vm86_regs_to_user(¤t->thread.vm86_info->regs, regs); |
145 | tmp += put_user(current->thread.screen_bitmap, ¤t->thread.vm86_info->screen_bitmap); | |
1da177e4 LT |
146 | if (tmp) { |
147 | printk("vm86: could not access userspace vm86_info\n"); | |
148 | do_exit(SIGSEGV); | |
149 | } | |
150 | ||
151 | tss = &per_cpu(init_tss, get_cpu()); | |
faca6227 | 152 | current->thread.sp0 = current->thread.saved_sp0; |
1da177e4 | 153 | current->thread.sysenter_cs = __KERNEL_CS; |
faca6227 PA |
154 | load_sp0(tss, ¤t->thread); |
155 | current->thread.saved_sp0 = 0; | |
1da177e4 LT |
156 | put_cpu(); |
157 | ||
1da177e4 | 158 | ret = KVM86->regs32; |
49d26b6e | 159 | |
65ea5b03 | 160 | ret->fs = current->thread.saved_fs; |
d9a89a26 | 161 | set_user_gs(ret, current->thread.saved_gs); |
49d26b6e | 162 | |
1da177e4 LT |
163 | return ret; |
164 | } | |
165 | ||
60ec5585 | 166 | static void mark_screen_rdonly(struct mm_struct *mm) |
1da177e4 LT |
167 | { |
168 | pgd_t *pgd; | |
169 | pud_t *pud; | |
170 | pmd_t *pmd; | |
60ec5585 HD |
171 | pte_t *pte; |
172 | spinlock_t *ptl; | |
1da177e4 LT |
173 | int i; |
174 | ||
60ec5585 | 175 | pgd = pgd_offset(mm, 0xA0000); |
1da177e4 LT |
176 | if (pgd_none_or_clear_bad(pgd)) |
177 | goto out; | |
178 | pud = pud_offset(pgd, 0xA0000); | |
179 | if (pud_none_or_clear_bad(pud)) | |
180 | goto out; | |
181 | pmd = pmd_offset(pud, 0xA0000); | |
182 | if (pmd_none_or_clear_bad(pmd)) | |
183 | goto out; | |
60ec5585 | 184 | pte = pte_offset_map_lock(mm, pmd, 0xA0000, &ptl); |
1da177e4 LT |
185 | for (i = 0; i < 32; i++) { |
186 | if (pte_present(*pte)) | |
187 | set_pte(pte, pte_wrprotect(*pte)); | |
188 | pte++; | |
189 | } | |
60ec5585 | 190 | pte_unmap_unlock(pte, ptl); |
1da177e4 | 191 | out: |
1da177e4 LT |
192 | flush_tlb(); |
193 | } | |
194 | ||
195 | ||
196 | ||
197 | static int do_vm86_irq_handling(int subfunction, int irqnumber); | |
198 | static void do_sys_vm86(struct kernel_vm86_struct *info, struct task_struct *tsk); | |
199 | ||
b12bdaf1 | 200 | int sys_vm86old(struct pt_regs *regs) |
1da177e4 | 201 | { |
b12bdaf1 | 202 | struct vm86_struct __user *v86 = (struct vm86_struct __user *)regs->bx; |
1da177e4 LT |
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 = -EPERM; | |
210 | ||
211 | tsk = current; | |
faca6227 | 212 | if (tsk->thread.saved_sp0) |
1da177e4 | 213 | goto out; |
49d26b6e JF |
214 | tmp = copy_vm86_regs_from_user(&info.regs, &v86->regs, |
215 | offsetof(struct kernel_vm86_struct, vm86plus) - | |
216 | sizeof(info.regs)); | |
1da177e4 LT |
217 | ret = -EFAULT; |
218 | if (tmp) | |
219 | goto out; | |
220 | memset(&info.vm86plus, 0, (int)&info.regs32 - (int)&info.vm86plus); | |
253f29a4 | 221 | info.regs32 = regs; |
1da177e4 LT |
222 | tsk->thread.vm86_info = v86; |
223 | do_sys_vm86(&info, tsk); | |
224 | ret = 0; /* we never return here */ | |
225 | out: | |
226 | return ret; | |
227 | } | |
228 | ||
229 | ||
b12bdaf1 | 230 | int sys_vm86(struct pt_regs *regs) |
1da177e4 LT |
231 | { |
232 | struct kernel_vm86_struct info; /* declare this _on top_, | |
233 | * this avoids wasting of stack space. | |
234 | * This remains on the stack until we | |
235 | * return to 32 bit user space. | |
236 | */ | |
237 | struct task_struct *tsk; | |
238 | int tmp, ret; | |
239 | struct vm86plus_struct __user *v86; | |
240 | ||
241 | tsk = current; | |
b12bdaf1 | 242 | switch (regs->bx) { |
83e714e8 PC |
243 | case VM86_REQUEST_IRQ: |
244 | case VM86_FREE_IRQ: | |
245 | case VM86_GET_IRQ_BITS: | |
246 | case VM86_GET_AND_RESET_IRQ: | |
b12bdaf1 | 247 | ret = do_vm86_irq_handling(regs->bx, (int)regs->cx); |
83e714e8 PC |
248 | goto out; |
249 | case VM86_PLUS_INSTALL_CHECK: | |
250 | /* | |
251 | * NOTE: on old vm86 stuff this will return the error | |
252 | * from access_ok(), because the subfunction is | |
253 | * interpreted as (invalid) address to vm86_struct. | |
254 | * So the installation check works. | |
255 | */ | |
256 | ret = 0; | |
257 | goto out; | |
1da177e4 LT |
258 | } |
259 | ||
260 | /* we come here only for functions VM86_ENTER, VM86_ENTER_NO_BYPASS */ | |
261 | ret = -EPERM; | |
faca6227 | 262 | if (tsk->thread.saved_sp0) |
1da177e4 | 263 | goto out; |
b12bdaf1 | 264 | v86 = (struct vm86plus_struct __user *)regs->cx; |
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; | |
253f29a4 | 271 | info.regs32 = regs; |
1da177e4 LT |
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 | */ | |
65ea5b03 PA |
287 | info->regs.pt.ds = 0; |
288 | info->regs.pt.es = 0; | |
289 | info->regs.pt.fs = 0; | |
3aa6b186 LR |
290 | #ifndef CONFIG_X86_32_LAZY_GS |
291 | info->regs.pt.gs = 0; | |
292 | #endif | |
1da177e4 LT |
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 | /* | |
975e5f45 | 320 | * Save old state, set default return value (%ax) to 0 (VM86_SIGNAL) |
1da177e4 | 321 | */ |
975e5f45 | 322 | info->regs32->ax = VM86_SIGNAL; |
faca6227 | 323 | tsk->thread.saved_sp0 = tsk->thread.sp0; |
65ea5b03 | 324 | tsk->thread.saved_fs = info->regs32->fs; |
d9a89a26 | 325 | tsk->thread.saved_gs = get_user_gs(info->regs32); |
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" | |
3aa6b186 | 345 | #ifdef CONFIG_X86_32_LAZY_GS |
464d1a78 | 346 | "mov %2, %%gs\n\t" |
3aa6b186 | 347 | #endif |
1da177e4 LT |
348 | "jmp resume_userspace" |
349 | : /* no outputs */ | |
49d26b6e | 350 | :"r" (&info->regs), "r" (task_thread_info(tsk)), "r" (0)); |
1da177e4 LT |
351 | /* we never return here */ |
352 | } | |
353 | ||
83e714e8 | 354 | static inline void return_to_32bit(struct kernel_vm86_regs *regs16, int retval) |
1da177e4 | 355 | { |
83e714e8 | 356 | struct pt_regs *regs32; |
1da177e4 LT |
357 | |
358 | regs32 = save_v86_state(regs16); | |
65ea5b03 | 359 | regs32->ax = retval; |
1da177e4 LT |
360 | __asm__ __volatile__("movl %0,%%esp\n\t" |
361 | "movl %1,%%ebp\n\t" | |
362 | "jmp resume_userspace" | |
363 | : : "r" (regs32), "r" (current_thread_info())); | |
364 | } | |
365 | ||
83e714e8 | 366 | static inline void set_IF(struct kernel_vm86_regs *regs) |
1da177e4 | 367 | { |
a5c15d41 | 368 | VEFLAGS |= X86_EFLAGS_VIF; |
369 | if (VEFLAGS & X86_EFLAGS_VIP) | |
1da177e4 LT |
370 | return_to_32bit(regs, VM86_STI); |
371 | } | |
372 | ||
83e714e8 | 373 | static inline void clear_IF(struct kernel_vm86_regs *regs) |
1da177e4 | 374 | { |
a5c15d41 | 375 | VEFLAGS &= ~X86_EFLAGS_VIF; |
1da177e4 LT |
376 | } |
377 | ||
83e714e8 | 378 | static inline void clear_TF(struct kernel_vm86_regs *regs) |
1da177e4 | 379 | { |
a5c15d41 | 380 | regs->pt.flags &= ~X86_EFLAGS_TF; |
1da177e4 LT |
381 | } |
382 | ||
83e714e8 | 383 | static inline void clear_AC(struct kernel_vm86_regs *regs) |
1da177e4 | 384 | { |
a5c15d41 | 385 | regs->pt.flags &= ~X86_EFLAGS_AC; |
1da177e4 LT |
386 | } |
387 | ||
83e714e8 PC |
388 | /* |
389 | * It is correct to call set_IF(regs) from the set_vflags_* | |
1da177e4 LT |
390 | * functions. However someone forgot to call clear_IF(regs) |
391 | * in the opposite case. | |
392 | * After the command sequence CLI PUSHF STI POPF you should | |
ab4a574e | 393 | * end up with interrupts disabled, but you ended up with |
1da177e4 LT |
394 | * interrupts enabled. |
395 | * ( I was testing my own changes, but the only bug I | |
396 | * could find was in a function I had not changed. ) | |
397 | * [KD] | |
398 | */ | |
399 | ||
83e714e8 | 400 | static inline void set_vflags_long(unsigned long flags, struct kernel_vm86_regs *regs) |
1da177e4 | 401 | { |
65ea5b03 PA |
402 | set_flags(VEFLAGS, flags, current->thread.v86mask); |
403 | set_flags(regs->pt.flags, flags, SAFE_MASK); | |
a5c15d41 | 404 | if (flags & X86_EFLAGS_IF) |
1da177e4 LT |
405 | set_IF(regs); |
406 | else | |
407 | clear_IF(regs); | |
408 | } | |
409 | ||
83e714e8 | 410 | static inline void set_vflags_short(unsigned short flags, struct kernel_vm86_regs *regs) |
1da177e4 LT |
411 | { |
412 | set_flags(VFLAGS, flags, current->thread.v86mask); | |
65ea5b03 | 413 | set_flags(regs->pt.flags, flags, SAFE_MASK); |
a5c15d41 | 414 | if (flags & X86_EFLAGS_IF) |
1da177e4 LT |
415 | set_IF(regs); |
416 | else | |
417 | clear_IF(regs); | |
418 | } | |
419 | ||
83e714e8 | 420 | static inline unsigned long get_vflags(struct kernel_vm86_regs *regs) |
1da177e4 | 421 | { |
65ea5b03 | 422 | unsigned long flags = regs->pt.flags & RETURN_MASK; |
1da177e4 | 423 | |
a5c15d41 | 424 | if (VEFLAGS & X86_EFLAGS_VIF) |
425 | flags |= X86_EFLAGS_IF; | |
426 | flags |= X86_EFLAGS_IOPL; | |
1da177e4 LT |
427 | return flags | (VEFLAGS & current->thread.v86mask); |
428 | } | |
429 | ||
83e714e8 | 430 | static inline int is_revectored(int nr, struct revectored_struct *bitmap) |
1da177e4 LT |
431 | { |
432 | __asm__ __volatile__("btl %2,%1\n\tsbbl %0,%0" | |
433 | :"=r" (nr) | |
83e714e8 | 434 | :"m" (*bitmap), "r" (nr)); |
1da177e4 LT |
435 | return nr; |
436 | } | |
437 | ||
438 | #define val_byte(val, n) (((__u8 *)&val)[n]) | |
439 | ||
440 | #define pushb(base, ptr, val, err_label) \ | |
441 | do { \ | |
442 | __u8 __val = val; \ | |
443 | ptr--; \ | |
444 | if (put_user(__val, base + ptr) < 0) \ | |
445 | goto err_label; \ | |
83e714e8 | 446 | } while (0) |
1da177e4 LT |
447 | |
448 | #define pushw(base, ptr, val, err_label) \ | |
449 | do { \ | |
450 | __u16 __val = val; \ | |
451 | ptr--; \ | |
452 | if (put_user(val_byte(__val, 1), base + ptr) < 0) \ | |
453 | goto err_label; \ | |
454 | ptr--; \ | |
455 | if (put_user(val_byte(__val, 0), base + ptr) < 0) \ | |
456 | goto err_label; \ | |
83e714e8 | 457 | } while (0) |
1da177e4 LT |
458 | |
459 | #define pushl(base, ptr, val, err_label) \ | |
460 | do { \ | |
461 | __u32 __val = val; \ | |
462 | ptr--; \ | |
463 | if (put_user(val_byte(__val, 3), base + ptr) < 0) \ | |
464 | goto err_label; \ | |
465 | ptr--; \ | |
466 | if (put_user(val_byte(__val, 2), base + ptr) < 0) \ | |
467 | goto err_label; \ | |
468 | ptr--; \ | |
469 | if (put_user(val_byte(__val, 1), base + ptr) < 0) \ | |
470 | goto err_label; \ | |
471 | ptr--; \ | |
472 | if (put_user(val_byte(__val, 0), base + ptr) < 0) \ | |
473 | goto err_label; \ | |
83e714e8 | 474 | } while (0) |
1da177e4 LT |
475 | |
476 | #define popb(base, ptr, err_label) \ | |
477 | ({ \ | |
478 | __u8 __res; \ | |
479 | if (get_user(__res, base + ptr) < 0) \ | |
480 | goto err_label; \ | |
481 | ptr++; \ | |
482 | __res; \ | |
483 | }) | |
484 | ||
485 | #define popw(base, ptr, err_label) \ | |
486 | ({ \ | |
487 | __u16 __res; \ | |
488 | if (get_user(val_byte(__res, 0), base + ptr) < 0) \ | |
489 | goto err_label; \ | |
490 | ptr++; \ | |
491 | if (get_user(val_byte(__res, 1), base + ptr) < 0) \ | |
492 | goto err_label; \ | |
493 | ptr++; \ | |
494 | __res; \ | |
495 | }) | |
496 | ||
497 | #define popl(base, ptr, err_label) \ | |
498 | ({ \ | |
499 | __u32 __res; \ | |
500 | if (get_user(val_byte(__res, 0), base + ptr) < 0) \ | |
501 | goto err_label; \ | |
502 | ptr++; \ | |
503 | if (get_user(val_byte(__res, 1), base + ptr) < 0) \ | |
504 | goto err_label; \ | |
505 | ptr++; \ | |
506 | if (get_user(val_byte(__res, 2), base + ptr) < 0) \ | |
507 | goto err_label; \ | |
508 | ptr++; \ | |
509 | if (get_user(val_byte(__res, 3), base + ptr) < 0) \ | |
510 | goto err_label; \ | |
511 | ptr++; \ | |
512 | __res; \ | |
513 | }) | |
514 | ||
515 | /* There are so many possible reasons for this function to return | |
516 | * VM86_INTx, so adding another doesn't bother me. We can expect | |
517 | * userspace programs to be able to handle it. (Getting a problem | |
518 | * in userspace is always better than an Oops anyway.) [KD] | |
519 | */ | |
520 | static void do_int(struct kernel_vm86_regs *regs, int i, | |
83e714e8 | 521 | unsigned char __user *ssp, unsigned short sp) |
1da177e4 LT |
522 | { |
523 | unsigned long __user *intr_ptr; | |
524 | unsigned long segoffs; | |
525 | ||
65ea5b03 | 526 | if (regs->pt.cs == BIOSSEG) |
1da177e4 LT |
527 | goto cannot_handle; |
528 | if (is_revectored(i, &KVM86->int_revectored)) | |
529 | goto cannot_handle; | |
83e714e8 | 530 | if (i == 0x21 && is_revectored(AH(regs), &KVM86->int21_revectored)) |
1da177e4 LT |
531 | goto cannot_handle; |
532 | intr_ptr = (unsigned long __user *) (i << 2); | |
533 | if (get_user(segoffs, intr_ptr)) | |
534 | goto cannot_handle; | |
535 | if ((segoffs >> 16) == BIOSSEG) | |
536 | goto cannot_handle; | |
537 | pushw(ssp, sp, get_vflags(regs), cannot_handle); | |
65ea5b03 | 538 | pushw(ssp, sp, regs->pt.cs, cannot_handle); |
1da177e4 | 539 | pushw(ssp, sp, IP(regs), cannot_handle); |
65ea5b03 | 540 | regs->pt.cs = segoffs >> 16; |
1da177e4 LT |
541 | SP(regs) -= 6; |
542 | IP(regs) = segoffs & 0xffff; | |
543 | clear_TF(regs); | |
544 | clear_IF(regs); | |
545 | clear_AC(regs); | |
546 | return; | |
547 | ||
548 | cannot_handle: | |
549 | return_to_32bit(regs, VM86_INTx + (i << 8)); | |
550 | } | |
551 | ||
83e714e8 | 552 | int handle_vm86_trap(struct kernel_vm86_regs *regs, long error_code, int trapno) |
1da177e4 LT |
553 | { |
554 | if (VMPI.is_vm86pus) { | |
83e714e8 | 555 | if ((trapno == 3) || (trapno == 1)) |
1da177e4 | 556 | return_to_32bit(regs, VM86_TRAP + (trapno << 8)); |
65ea5b03 | 557 | do_int(regs, trapno, (unsigned char __user *) (regs->pt.ss << 4), SP(regs)); |
1da177e4 LT |
558 | return 0; |
559 | } | |
83e714e8 | 560 | if (trapno != 1) |
1da177e4 | 561 | return 1; /* we let this handle by the calling routine */ |
1da177e4 LT |
562 | current->thread.trap_no = trapno; |
563 | current->thread.error_code = error_code; | |
0f540910 | 564 | force_sig(SIGTRAP, current); |
1da177e4 LT |
565 | return 0; |
566 | } | |
567 | ||
83e714e8 | 568 | void handle_vm86_fault(struct kernel_vm86_regs *regs, long error_code) |
1da177e4 LT |
569 | { |
570 | unsigned char opcode; | |
571 | unsigned char __user *csp; | |
572 | unsigned char __user *ssp; | |
5fd75ebb | 573 | unsigned short ip, sp, orig_flags; |
1da177e4 LT |
574 | int data32, pref_done; |
575 | ||
576 | #define CHECK_IF_IN_TRAP \ | |
577 | if (VMPI.vm86dbg_active && VMPI.vm86dbg_TFpendig) \ | |
a5c15d41 | 578 | newflags |= X86_EFLAGS_TF |
1da177e4 | 579 | #define VM86_FAULT_RETURN do { \ |
a5c15d41 | 580 | if (VMPI.force_return_for_pic && (VEFLAGS & (X86_EFLAGS_IF | X86_EFLAGS_VIF))) \ |
1da177e4 | 581 | return_to_32bit(regs, VM86_PICRETURN); \ |
a5c15d41 | 582 | if (orig_flags & X86_EFLAGS_TF) \ |
5fd75ebb | 583 | handle_vm86_trap(regs, 0, 1); \ |
1da177e4 LT |
584 | return; } while (0) |
585 | ||
65ea5b03 | 586 | orig_flags = *(unsigned short *)®s->pt.flags; |
5fd75ebb | 587 | |
65ea5b03 PA |
588 | csp = (unsigned char __user *) (regs->pt.cs << 4); |
589 | ssp = (unsigned char __user *) (regs->pt.ss << 4); | |
1da177e4 LT |
590 | sp = SP(regs); |
591 | ip = IP(regs); | |
592 | ||
593 | data32 = 0; | |
594 | pref_done = 0; | |
595 | do { | |
596 | switch (opcode = popb(csp, ip, simulate_sigsegv)) { | |
83e714e8 PC |
597 | case 0x66: /* 32-bit data */ data32 = 1; break; |
598 | case 0x67: /* 32-bit address */ break; | |
599 | case 0x2e: /* CS */ break; | |
600 | case 0x3e: /* DS */ break; | |
601 | case 0x26: /* ES */ break; | |
602 | case 0x36: /* SS */ break; | |
603 | case 0x65: /* GS */ break; | |
604 | case 0x64: /* FS */ break; | |
605 | case 0xf2: /* repnz */ break; | |
606 | case 0xf3: /* rep */ break; | |
607 | default: pref_done = 1; | |
1da177e4 LT |
608 | } |
609 | } while (!pref_done); | |
610 | ||
611 | switch (opcode) { | |
612 | ||
613 | /* pushf */ | |
614 | case 0x9c: | |
615 | if (data32) { | |
616 | pushl(ssp, sp, get_vflags(regs), simulate_sigsegv); | |
617 | SP(regs) -= 4; | |
618 | } else { | |
619 | pushw(ssp, sp, get_vflags(regs), simulate_sigsegv); | |
620 | SP(regs) -= 2; | |
621 | } | |
622 | IP(regs) = ip; | |
623 | VM86_FAULT_RETURN; | |
624 | ||
625 | /* popf */ | |
626 | case 0x9d: | |
627 | { | |
628 | unsigned long newflags; | |
629 | if (data32) { | |
83e714e8 | 630 | newflags = popl(ssp, sp, simulate_sigsegv); |
1da177e4 LT |
631 | SP(regs) += 4; |
632 | } else { | |
633 | newflags = popw(ssp, sp, simulate_sigsegv); | |
634 | SP(regs) += 2; | |
635 | } | |
636 | IP(regs) = ip; | |
637 | CHECK_IF_IN_TRAP; | |
83e714e8 | 638 | if (data32) |
1da177e4 | 639 | set_vflags_long(newflags, regs); |
83e714e8 | 640 | else |
1da177e4 | 641 | set_vflags_short(newflags, regs); |
83e714e8 | 642 | |
1da177e4 LT |
643 | VM86_FAULT_RETURN; |
644 | } | |
645 | ||
646 | /* int xx */ | |
647 | case 0xcd: { | |
83e714e8 | 648 | int intno = popb(csp, ip, simulate_sigsegv); |
1da177e4 LT |
649 | IP(regs) = ip; |
650 | if (VMPI.vm86dbg_active) { | |
83e714e8 | 651 | if ((1 << (intno & 7)) & VMPI.vm86dbg_intxxtab[intno >> 3]) |
1da177e4 LT |
652 | return_to_32bit(regs, VM86_INTx + (intno << 8)); |
653 | } | |
654 | do_int(regs, intno, ssp, sp); | |
655 | return; | |
656 | } | |
657 | ||
658 | /* iret */ | |
659 | case 0xcf: | |
660 | { | |
661 | unsigned long newip; | |
662 | unsigned long newcs; | |
663 | unsigned long newflags; | |
664 | if (data32) { | |
83e714e8 PC |
665 | newip = popl(ssp, sp, simulate_sigsegv); |
666 | newcs = popl(ssp, sp, simulate_sigsegv); | |
667 | newflags = popl(ssp, sp, simulate_sigsegv); | |
1da177e4 LT |
668 | SP(regs) += 12; |
669 | } else { | |
670 | newip = popw(ssp, sp, simulate_sigsegv); | |
671 | newcs = popw(ssp, sp, simulate_sigsegv); | |
672 | newflags = popw(ssp, sp, simulate_sigsegv); | |
673 | SP(regs) += 6; | |
674 | } | |
675 | IP(regs) = newip; | |
65ea5b03 | 676 | regs->pt.cs = newcs; |
1da177e4 LT |
677 | CHECK_IF_IN_TRAP; |
678 | if (data32) { | |
679 | set_vflags_long(newflags, regs); | |
680 | } else { | |
681 | set_vflags_short(newflags, regs); | |
682 | } | |
683 | VM86_FAULT_RETURN; | |
684 | } | |
685 | ||
686 | /* cli */ | |
687 | case 0xfa: | |
688 | IP(regs) = ip; | |
689 | clear_IF(regs); | |
690 | VM86_FAULT_RETURN; | |
691 | ||
692 | /* sti */ | |
693 | /* | |
694 | * Damn. This is incorrect: the 'sti' instruction should actually | |
695 | * enable interrupts after the /next/ instruction. Not good. | |
696 | * | |
697 | * Probably needs some horsing around with the TF flag. Aiee.. | |
698 | */ | |
699 | case 0xfb: | |
700 | IP(regs) = ip; | |
701 | set_IF(regs); | |
702 | VM86_FAULT_RETURN; | |
703 | ||
704 | default: | |
705 | return_to_32bit(regs, VM86_UNKNOWN); | |
706 | } | |
707 | ||
708 | return; | |
709 | ||
710 | simulate_sigsegv: | |
711 | /* FIXME: After a long discussion with Stas we finally | |
712 | * agreed, that this is wrong. Here we should | |
713 | * really send a SIGSEGV to the user program. | |
714 | * But how do we create the correct context? We | |
715 | * are inside a general protection fault handler | |
716 | * and has just returned from a page fault handler. | |
717 | * The correct context for the signal handler | |
718 | * should be a mixture of the two, but how do we | |
719 | * get the information? [KD] | |
720 | */ | |
721 | return_to_32bit(regs, VM86_UNKNOWN); | |
722 | } | |
723 | ||
724 | /* ---------------- vm86 special IRQ passing stuff ----------------- */ | |
725 | ||
726 | #define VM86_IRQNAME "vm86irq" | |
727 | ||
728 | static struct vm86_irqs { | |
729 | struct task_struct *tsk; | |
730 | int sig; | |
731 | } vm86_irqs[16]; | |
732 | ||
733 | static DEFINE_SPINLOCK(irqbits_lock); | |
734 | static int irqbits; | |
735 | ||
83e714e8 | 736 | #define ALLOWED_SIGS (1 /* 0 = don't send a signal */ \ |
1da177e4 | 737 | | (1 << SIGUSR1) | (1 << SIGUSR2) | (1 << SIGIO) | (1 << SIGURG) \ |
83e714e8 PC |
738 | | (1 << SIGUNUSED)) |
739 | ||
7d12e780 | 740 | static irqreturn_t irq_handler(int intno, void *dev_id) |
1da177e4 LT |
741 | { |
742 | int irq_bit; | |
743 | unsigned long flags; | |
744 | ||
83e714e8 | 745 | spin_lock_irqsave(&irqbits_lock, flags); |
1da177e4 | 746 | irq_bit = 1 << intno; |
83e714e8 | 747 | if ((irqbits & irq_bit) || !vm86_irqs[intno].tsk) |
1da177e4 LT |
748 | goto out; |
749 | irqbits |= irq_bit; | |
750 | if (vm86_irqs[intno].sig) | |
751 | send_sig(vm86_irqs[intno].sig, vm86_irqs[intno].tsk, 1); | |
1da177e4 LT |
752 | /* |
753 | * IRQ will be re-enabled when user asks for the irq (whether | |
754 | * polling or as a result of the signal) | |
755 | */ | |
ad671423 PP |
756 | disable_irq_nosync(intno); |
757 | spin_unlock_irqrestore(&irqbits_lock, flags); | |
1da177e4 LT |
758 | return IRQ_HANDLED; |
759 | ||
760 | out: | |
83e714e8 | 761 | spin_unlock_irqrestore(&irqbits_lock, flags); |
1da177e4 LT |
762 | return IRQ_NONE; |
763 | } | |
764 | ||
765 | static inline void free_vm86_irq(int irqnumber) | |
766 | { | |
767 | unsigned long flags; | |
768 | ||
769 | free_irq(irqnumber, NULL); | |
770 | vm86_irqs[irqnumber].tsk = NULL; | |
771 | ||
83e714e8 | 772 | spin_lock_irqsave(&irqbits_lock, flags); |
1da177e4 | 773 | irqbits &= ~(1 << irqnumber); |
83e714e8 | 774 | spin_unlock_irqrestore(&irqbits_lock, flags); |
1da177e4 LT |
775 | } |
776 | ||
777 | void release_vm86_irqs(struct task_struct *task) | |
778 | { | |
779 | int i; | |
780 | for (i = FIRST_VM86_IRQ ; i <= LAST_VM86_IRQ; i++) | |
781 | if (vm86_irqs[i].tsk == task) | |
782 | free_vm86_irq(i); | |
783 | } | |
784 | ||
785 | static inline int get_and_reset_irq(int irqnumber) | |
786 | { | |
787 | int bit; | |
788 | unsigned long flags; | |
ad671423 | 789 | int ret = 0; |
83e714e8 | 790 | |
1da177e4 LT |
791 | if (invalid_vm86_irq(irqnumber)) return 0; |
792 | if (vm86_irqs[irqnumber].tsk != current) return 0; | |
83e714e8 | 793 | spin_lock_irqsave(&irqbits_lock, flags); |
1da177e4 LT |
794 | bit = irqbits & (1 << irqnumber); |
795 | irqbits &= ~bit; | |
ad671423 PP |
796 | if (bit) { |
797 | enable_irq(irqnumber); | |
798 | ret = 1; | |
799 | } | |
800 | ||
83e714e8 | 801 | spin_unlock_irqrestore(&irqbits_lock, flags); |
ad671423 | 802 | return ret; |
1da177e4 LT |
803 | } |
804 | ||
805 | ||
806 | static int do_vm86_irq_handling(int subfunction, int irqnumber) | |
807 | { | |
808 | int ret; | |
809 | switch (subfunction) { | |
810 | case VM86_GET_AND_RESET_IRQ: { | |
811 | return get_and_reset_irq(irqnumber); | |
812 | } | |
813 | case VM86_GET_IRQ_BITS: { | |
814 | return irqbits; | |
815 | } | |
816 | case VM86_REQUEST_IRQ: { | |
817 | int sig = irqnumber >> 8; | |
818 | int irq = irqnumber & 255; | |
819 | if (!capable(CAP_SYS_ADMIN)) return -EPERM; | |
820 | if (!((1 << sig) & ALLOWED_SIGS)) return -EPERM; | |
821 | if (invalid_vm86_irq(irq)) return -EPERM; | |
822 | if (vm86_irqs[irq].tsk) return -EPERM; | |
823 | ret = request_irq(irq, &irq_handler, 0, VM86_IRQNAME, NULL); | |
824 | if (ret) return ret; | |
825 | vm86_irqs[irq].sig = sig; | |
826 | vm86_irqs[irq].tsk = current; | |
827 | return irq; | |
828 | } | |
829 | case VM86_FREE_IRQ: { | |
830 | if (invalid_vm86_irq(irqnumber)) return -EPERM; | |
831 | if (!vm86_irqs[irqnumber].tsk) return 0; | |
832 | if (vm86_irqs[irqnumber].tsk != current) return -EPERM; | |
833 | free_vm86_irq(irqnumber); | |
834 | return 0; | |
835 | } | |
836 | } | |
837 | return -EINVAL; | |
838 | } | |
839 |