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 | ||
c767a54b JP |
31 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
32 | ||
a9415644 | 33 | #include <linux/capability.h> |
1da177e4 LT |
34 | #include <linux/errno.h> |
35 | #include <linux/interrupt.h> | |
36 | #include <linux/sched.h> | |
37 | #include <linux/kernel.h> | |
38 | #include <linux/signal.h> | |
39 | #include <linux/string.h> | |
40 | #include <linux/mm.h> | |
41 | #include <linux/smp.h> | |
1da177e4 LT |
42 | #include <linux/highmem.h> |
43 | #include <linux/ptrace.h> | |
7e7f8a03 | 44 | #include <linux/audit.h> |
49d26b6e | 45 | #include <linux/stddef.h> |
1da177e4 LT |
46 | |
47 | #include <asm/uaccess.h> | |
48 | #include <asm/io.h> | |
49 | #include <asm/tlbflush.h> | |
50 | #include <asm/irq.h> | |
bbc1f698 | 51 | #include <asm/syscalls.h> |
1da177e4 LT |
52 | |
53 | /* | |
54 | * Known problems: | |
55 | * | |
56 | * Interrupt handling is not guaranteed: | |
57 | * - a real x86 will disable all interrupts for one instruction | |
58 | * after a "mov ss,xx" to make stack handling atomic even without | |
59 | * the 'lss' instruction. We can't guarantee this in v86 mode, | |
60 | * as the next instruction might result in a page fault or similar. | |
61 | * - a real x86 will have interrupts disabled for one instruction | |
62 | * past the 'sti' that enables them. We don't bother with all the | |
63 | * details yet. | |
64 | * | |
65 | * Let's hope these problems do not actually matter for anything. | |
66 | */ | |
67 | ||
68 | ||
69 | #define KVM86 ((struct kernel_vm86_struct *)regs) | |
83e714e8 | 70 | #define VMPI KVM86->vm86plus |
1da177e4 LT |
71 | |
72 | ||
73 | /* | |
74 | * 8- and 16-bit register defines.. | |
75 | */ | |
65ea5b03 PA |
76 | #define AL(regs) (((unsigned char *)&((regs)->pt.ax))[0]) |
77 | #define AH(regs) (((unsigned char *)&((regs)->pt.ax))[1]) | |
78 | #define IP(regs) (*(unsigned short *)&((regs)->pt.ip)) | |
79 | #define SP(regs) (*(unsigned short *)&((regs)->pt.sp)) | |
1da177e4 LT |
80 | |
81 | /* | |
82 | * virtual flags (16 and 32-bit versions) | |
83 | */ | |
84 | #define VFLAGS (*(unsigned short *)&(current->thread.v86flags)) | |
85 | #define VEFLAGS (current->thread.v86flags) | |
86 | ||
83e714e8 | 87 | #define set_flags(X, new, mask) \ |
1da177e4 LT |
88 | ((X) = ((X) & ~(mask)) | ((new) & (mask))) |
89 | ||
90 | #define SAFE_MASK (0xDD5) | |
91 | #define RETURN_MASK (0xDFF) | |
92 | ||
49d26b6e JF |
93 | /* convert kernel_vm86_regs to vm86_regs */ |
94 | static int copy_vm86_regs_to_user(struct vm86_regs __user *user, | |
95 | const struct kernel_vm86_regs *regs) | |
96 | { | |
97 | int ret = 0; | |
98 | ||
83e714e8 PC |
99 | /* |
100 | * kernel_vm86_regs is missing gs, so copy everything up to | |
101 | * (but not including) orig_eax, and then rest including orig_eax. | |
102 | */ | |
65ea5b03 PA |
103 | ret += copy_to_user(user, regs, offsetof(struct kernel_vm86_regs, pt.orig_ax)); |
104 | ret += copy_to_user(&user->orig_eax, ®s->pt.orig_ax, | |
49d26b6e | 105 | sizeof(struct kernel_vm86_regs) - |
65ea5b03 | 106 | offsetof(struct kernel_vm86_regs, pt.orig_ax)); |
49d26b6e JF |
107 | |
108 | return ret; | |
109 | } | |
110 | ||
111 | /* convert vm86_regs to kernel_vm86_regs */ | |
112 | static int copy_vm86_regs_from_user(struct kernel_vm86_regs *regs, | |
113 | const struct vm86_regs __user *user, | |
114 | unsigned extra) | |
115 | { | |
116 | int ret = 0; | |
117 | ||
65ea5b03 PA |
118 | /* copy ax-fs inclusive */ |
119 | ret += copy_from_user(regs, user, offsetof(struct kernel_vm86_regs, pt.orig_ax)); | |
120 | /* copy orig_ax-__gsh+extra */ | |
121 | ret += copy_from_user(®s->pt.orig_ax, &user->orig_eax, | |
49d26b6e | 122 | sizeof(struct kernel_vm86_regs) - |
65ea5b03 | 123 | offsetof(struct kernel_vm86_regs, pt.orig_ax) + |
49d26b6e | 124 | extra); |
49d26b6e JF |
125 | return ret; |
126 | } | |
1da177e4 | 127 | |
83e714e8 | 128 | struct pt_regs *save_v86_state(struct kernel_vm86_regs *regs) |
1da177e4 LT |
129 | { |
130 | struct tss_struct *tss; | |
131 | struct pt_regs *ret; | |
132 | unsigned long tmp; | |
133 | ||
134 | /* | |
135 | * This gets called from entry.S with interrupts disabled, but | |
136 | * from process context. Enable interrupts here, before trying | |
137 | * to access user space. | |
138 | */ | |
139 | local_irq_enable(); | |
140 | ||
141 | if (!current->thread.vm86_info) { | |
c767a54b | 142 | pr_alert("no vm86_info: BAD\n"); |
1da177e4 LT |
143 | do_exit(SIGSEGV); |
144 | } | |
a5c15d41 | 145 | set_flags(regs->pt.flags, VEFLAGS, X86_EFLAGS_VIF | current->thread.v86mask); |
83e714e8 PC |
146 | tmp = copy_vm86_regs_to_user(¤t->thread.vm86_info->regs, regs); |
147 | tmp += put_user(current->thread.screen_bitmap, ¤t->thread.vm86_info->screen_bitmap); | |
1da177e4 | 148 | if (tmp) { |
c767a54b | 149 | pr_alert("could not access userspace vm86_info\n"); |
1da177e4 LT |
150 | do_exit(SIGSEGV); |
151 | } | |
152 | ||
153 | tss = &per_cpu(init_tss, get_cpu()); | |
faca6227 | 154 | current->thread.sp0 = current->thread.saved_sp0; |
1da177e4 | 155 | current->thread.sysenter_cs = __KERNEL_CS; |
faca6227 PA |
156 | load_sp0(tss, ¤t->thread); |
157 | current->thread.saved_sp0 = 0; | |
1da177e4 LT |
158 | put_cpu(); |
159 | ||
1da177e4 | 160 | ret = KVM86->regs32; |
49d26b6e | 161 | |
65ea5b03 | 162 | ret->fs = current->thread.saved_fs; |
d9a89a26 | 163 | set_user_gs(ret, current->thread.saved_gs); |
49d26b6e | 164 | |
1da177e4 LT |
165 | return ret; |
166 | } | |
167 | ||
60ec5585 | 168 | static void mark_screen_rdonly(struct mm_struct *mm) |
1da177e4 LT |
169 | { |
170 | pgd_t *pgd; | |
171 | pud_t *pud; | |
172 | pmd_t *pmd; | |
60ec5585 HD |
173 | pte_t *pte; |
174 | spinlock_t *ptl; | |
1da177e4 LT |
175 | int i; |
176 | ||
1a5a9906 | 177 | down_write(&mm->mmap_sem); |
60ec5585 | 178 | pgd = pgd_offset(mm, 0xA0000); |
1da177e4 LT |
179 | if (pgd_none_or_clear_bad(pgd)) |
180 | goto out; | |
181 | pud = pud_offset(pgd, 0xA0000); | |
182 | if (pud_none_or_clear_bad(pud)) | |
183 | goto out; | |
184 | pmd = pmd_offset(pud, 0xA0000); | |
e180377f | 185 | split_huge_page_pmd_mm(mm, 0xA0000, pmd); |
1da177e4 LT |
186 | if (pmd_none_or_clear_bad(pmd)) |
187 | goto out; | |
60ec5585 | 188 | pte = pte_offset_map_lock(mm, pmd, 0xA0000, &ptl); |
1da177e4 LT |
189 | for (i = 0; i < 32; i++) { |
190 | if (pte_present(*pte)) | |
191 | set_pte(pte, pte_wrprotect(*pte)); | |
192 | pte++; | |
193 | } | |
60ec5585 | 194 | pte_unmap_unlock(pte, ptl); |
1da177e4 | 195 | out: |
1a5a9906 | 196 | up_write(&mm->mmap_sem); |
1da177e4 LT |
197 | flush_tlb(); |
198 | } | |
199 | ||
200 | ||
201 | ||
202 | static int do_vm86_irq_handling(int subfunction, int irqnumber); | |
203 | static void do_sys_vm86(struct kernel_vm86_struct *info, struct task_struct *tsk); | |
204 | ||
f1382f15 | 205 | int sys_vm86old(struct vm86_struct __user *v86, struct pt_regs *regs) |
1da177e4 | 206 | { |
1da177e4 LT |
207 | struct kernel_vm86_struct info; /* declare this _on top_, |
208 | * this avoids wasting of stack space. | |
209 | * This remains on the stack until we | |
210 | * return to 32 bit user space. | |
211 | */ | |
212 | struct task_struct *tsk; | |
213 | int tmp, ret = -EPERM; | |
214 | ||
215 | tsk = current; | |
faca6227 | 216 | if (tsk->thread.saved_sp0) |
1da177e4 | 217 | goto out; |
49d26b6e JF |
218 | tmp = copy_vm86_regs_from_user(&info.regs, &v86->regs, |
219 | offsetof(struct kernel_vm86_struct, vm86plus) - | |
220 | sizeof(info.regs)); | |
1da177e4 LT |
221 | ret = -EFAULT; |
222 | if (tmp) | |
223 | goto out; | |
224 | memset(&info.vm86plus, 0, (int)&info.regs32 - (int)&info.vm86plus); | |
253f29a4 | 225 | info.regs32 = regs; |
1da177e4 LT |
226 | tsk->thread.vm86_info = v86; |
227 | do_sys_vm86(&info, tsk); | |
228 | ret = 0; /* we never return here */ | |
229 | out: | |
230 | return ret; | |
231 | } | |
232 | ||
233 | ||
f1382f15 | 234 | int sys_vm86(unsigned long cmd, unsigned long arg, struct pt_regs *regs) |
1da177e4 LT |
235 | { |
236 | struct kernel_vm86_struct info; /* declare this _on top_, | |
237 | * this avoids wasting of stack space. | |
238 | * This remains on the stack until we | |
239 | * return to 32 bit user space. | |
240 | */ | |
241 | struct task_struct *tsk; | |
242 | int tmp, ret; | |
243 | struct vm86plus_struct __user *v86; | |
244 | ||
245 | tsk = current; | |
f1382f15 | 246 | switch (cmd) { |
83e714e8 PC |
247 | case VM86_REQUEST_IRQ: |
248 | case VM86_FREE_IRQ: | |
249 | case VM86_GET_IRQ_BITS: | |
250 | case VM86_GET_AND_RESET_IRQ: | |
f1382f15 | 251 | ret = do_vm86_irq_handling(cmd, (int)arg); |
83e714e8 PC |
252 | goto out; |
253 | case VM86_PLUS_INSTALL_CHECK: | |
254 | /* | |
255 | * NOTE: on old vm86 stuff this will return the error | |
256 | * from access_ok(), because the subfunction is | |
257 | * interpreted as (invalid) address to vm86_struct. | |
258 | * So the installation check works. | |
259 | */ | |
260 | ret = 0; | |
261 | goto out; | |
1da177e4 LT |
262 | } |
263 | ||
264 | /* we come here only for functions VM86_ENTER, VM86_ENTER_NO_BYPASS */ | |
265 | ret = -EPERM; | |
faca6227 | 266 | if (tsk->thread.saved_sp0) |
1da177e4 | 267 | goto out; |
f1382f15 | 268 | v86 = (struct vm86plus_struct __user *)arg; |
49d26b6e JF |
269 | tmp = copy_vm86_regs_from_user(&info.regs, &v86->regs, |
270 | offsetof(struct kernel_vm86_struct, regs32) - | |
271 | sizeof(info.regs)); | |
1da177e4 LT |
272 | ret = -EFAULT; |
273 | if (tmp) | |
274 | goto out; | |
253f29a4 | 275 | info.regs32 = regs; |
1da177e4 LT |
276 | info.vm86plus.is_vm86pus = 1; |
277 | tsk->thread.vm86_info = (struct vm86_struct __user *)v86; | |
278 | do_sys_vm86(&info, tsk); | |
279 | ret = 0; /* we never return here */ | |
280 | out: | |
281 | return ret; | |
282 | } | |
283 | ||
284 | ||
285 | static void do_sys_vm86(struct kernel_vm86_struct *info, struct task_struct *tsk) | |
286 | { | |
287 | struct tss_struct *tss; | |
288 | /* | |
289 | * make sure the vm86() system call doesn't try to do anything silly | |
290 | */ | |
65ea5b03 PA |
291 | info->regs.pt.ds = 0; |
292 | info->regs.pt.es = 0; | |
293 | info->regs.pt.fs = 0; | |
3aa6b186 LR |
294 | #ifndef CONFIG_X86_32_LAZY_GS |
295 | info->regs.pt.gs = 0; | |
296 | #endif | |
1da177e4 LT |
297 | |
298 | /* | |
65ea5b03 | 299 | * The flags register is also special: we cannot trust that the user |
1da177e4 LT |
300 | * has set it up safely, so this makes sure interrupt etc flags are |
301 | * inherited from protected mode. | |
302 | */ | |
65ea5b03 PA |
303 | VEFLAGS = info->regs.pt.flags; |
304 | info->regs.pt.flags &= SAFE_MASK; | |
305 | info->regs.pt.flags |= info->regs32->flags & ~SAFE_MASK; | |
6b6891f9 | 306 | info->regs.pt.flags |= X86_VM_MASK; |
1da177e4 LT |
307 | |
308 | switch (info->cpu_type) { | |
83e714e8 PC |
309 | case CPU_286: |
310 | tsk->thread.v86mask = 0; | |
311 | break; | |
312 | case CPU_386: | |
a5c15d41 | 313 | tsk->thread.v86mask = X86_EFLAGS_NT | X86_EFLAGS_IOPL; |
83e714e8 PC |
314 | break; |
315 | case CPU_486: | |
a5c15d41 | 316 | tsk->thread.v86mask = X86_EFLAGS_AC | X86_EFLAGS_NT | X86_EFLAGS_IOPL; |
83e714e8 PC |
317 | break; |
318 | default: | |
a5c15d41 | 319 | tsk->thread.v86mask = X86_EFLAGS_ID | X86_EFLAGS_AC | X86_EFLAGS_NT | X86_EFLAGS_IOPL; |
83e714e8 | 320 | break; |
1da177e4 LT |
321 | } |
322 | ||
323 | /* | |
975e5f45 | 324 | * Save old state, set default return value (%ax) to 0 (VM86_SIGNAL) |
1da177e4 | 325 | */ |
975e5f45 | 326 | info->regs32->ax = VM86_SIGNAL; |
faca6227 | 327 | tsk->thread.saved_sp0 = tsk->thread.sp0; |
65ea5b03 | 328 | tsk->thread.saved_fs = info->regs32->fs; |
d9a89a26 | 329 | tsk->thread.saved_gs = get_user_gs(info->regs32); |
1da177e4 LT |
330 | |
331 | tss = &per_cpu(init_tss, get_cpu()); | |
faca6227 | 332 | tsk->thread.sp0 = (unsigned long) &info->VM86_TSS_ESP0; |
1da177e4 LT |
333 | if (cpu_has_sep) |
334 | tsk->thread.sysenter_cs = 0; | |
faca6227 | 335 | load_sp0(tss, &tsk->thread); |
1da177e4 LT |
336 | put_cpu(); |
337 | ||
338 | tsk->thread.screen_bitmap = info->screen_bitmap; | |
339 | if (info->flags & VM86_SCREEN_BITMAP) | |
60ec5585 | 340 | mark_screen_rdonly(tsk->mm); |
7e7f8a03 | 341 | |
d7e7528b | 342 | /*call __audit_syscall_exit since we do not exit via the normal paths */ |
6015ff10 | 343 | #ifdef CONFIG_AUDITSYSCALL |
7e7f8a03 | 344 | if (unlikely(current->audit_context)) |
d7e7528b | 345 | __audit_syscall_exit(1, 0); |
6015ff10 | 346 | #endif |
7e7f8a03 | 347 | |
1da177e4 | 348 | __asm__ __volatile__( |
1da177e4 LT |
349 | "movl %0,%%esp\n\t" |
350 | "movl %1,%%ebp\n\t" | |
3aa6b186 | 351 | #ifdef CONFIG_X86_32_LAZY_GS |
464d1a78 | 352 | "mov %2, %%gs\n\t" |
3aa6b186 | 353 | #endif |
1da177e4 LT |
354 | "jmp resume_userspace" |
355 | : /* no outputs */ | |
49d26b6e | 356 | :"r" (&info->regs), "r" (task_thread_info(tsk)), "r" (0)); |
1da177e4 LT |
357 | /* we never return here */ |
358 | } | |
359 | ||
83e714e8 | 360 | static inline void return_to_32bit(struct kernel_vm86_regs *regs16, int retval) |
1da177e4 | 361 | { |
83e714e8 | 362 | struct pt_regs *regs32; |
1da177e4 LT |
363 | |
364 | regs32 = save_v86_state(regs16); | |
65ea5b03 | 365 | regs32->ax = retval; |
1da177e4 LT |
366 | __asm__ __volatile__("movl %0,%%esp\n\t" |
367 | "movl %1,%%ebp\n\t" | |
368 | "jmp resume_userspace" | |
369 | : : "r" (regs32), "r" (current_thread_info())); | |
370 | } | |
371 | ||
83e714e8 | 372 | static inline void set_IF(struct kernel_vm86_regs *regs) |
1da177e4 | 373 | { |
a5c15d41 | 374 | VEFLAGS |= X86_EFLAGS_VIF; |
375 | if (VEFLAGS & X86_EFLAGS_VIP) | |
1da177e4 LT |
376 | return_to_32bit(regs, VM86_STI); |
377 | } | |
378 | ||
83e714e8 | 379 | static inline void clear_IF(struct kernel_vm86_regs *regs) |
1da177e4 | 380 | { |
a5c15d41 | 381 | VEFLAGS &= ~X86_EFLAGS_VIF; |
1da177e4 LT |
382 | } |
383 | ||
83e714e8 | 384 | static inline void clear_TF(struct kernel_vm86_regs *regs) |
1da177e4 | 385 | { |
a5c15d41 | 386 | regs->pt.flags &= ~X86_EFLAGS_TF; |
1da177e4 LT |
387 | } |
388 | ||
83e714e8 | 389 | static inline void clear_AC(struct kernel_vm86_regs *regs) |
1da177e4 | 390 | { |
a5c15d41 | 391 | regs->pt.flags &= ~X86_EFLAGS_AC; |
1da177e4 LT |
392 | } |
393 | ||
83e714e8 PC |
394 | /* |
395 | * It is correct to call set_IF(regs) from the set_vflags_* | |
1da177e4 LT |
396 | * functions. However someone forgot to call clear_IF(regs) |
397 | * in the opposite case. | |
398 | * After the command sequence CLI PUSHF STI POPF you should | |
ab4a574e | 399 | * end up with interrupts disabled, but you ended up with |
1da177e4 LT |
400 | * interrupts enabled. |
401 | * ( I was testing my own changes, but the only bug I | |
402 | * could find was in a function I had not changed. ) | |
403 | * [KD] | |
404 | */ | |
405 | ||
83e714e8 | 406 | static inline void set_vflags_long(unsigned long flags, struct kernel_vm86_regs *regs) |
1da177e4 | 407 | { |
65ea5b03 PA |
408 | set_flags(VEFLAGS, flags, current->thread.v86mask); |
409 | set_flags(regs->pt.flags, flags, SAFE_MASK); | |
a5c15d41 | 410 | if (flags & X86_EFLAGS_IF) |
1da177e4 LT |
411 | set_IF(regs); |
412 | else | |
413 | clear_IF(regs); | |
414 | } | |
415 | ||
83e714e8 | 416 | static inline void set_vflags_short(unsigned short flags, struct kernel_vm86_regs *regs) |
1da177e4 LT |
417 | { |
418 | set_flags(VFLAGS, flags, current->thread.v86mask); | |
65ea5b03 | 419 | set_flags(regs->pt.flags, flags, SAFE_MASK); |
a5c15d41 | 420 | if (flags & X86_EFLAGS_IF) |
1da177e4 LT |
421 | set_IF(regs); |
422 | else | |
423 | clear_IF(regs); | |
424 | } | |
425 | ||
83e714e8 | 426 | static inline unsigned long get_vflags(struct kernel_vm86_regs *regs) |
1da177e4 | 427 | { |
65ea5b03 | 428 | unsigned long flags = regs->pt.flags & RETURN_MASK; |
1da177e4 | 429 | |
a5c15d41 | 430 | if (VEFLAGS & X86_EFLAGS_VIF) |
431 | flags |= X86_EFLAGS_IF; | |
432 | flags |= X86_EFLAGS_IOPL; | |
1da177e4 LT |
433 | return flags | (VEFLAGS & current->thread.v86mask); |
434 | } | |
435 | ||
83e714e8 | 436 | static inline int is_revectored(int nr, struct revectored_struct *bitmap) |
1da177e4 LT |
437 | { |
438 | __asm__ __volatile__("btl %2,%1\n\tsbbl %0,%0" | |
439 | :"=r" (nr) | |
83e714e8 | 440 | :"m" (*bitmap), "r" (nr)); |
1da177e4 LT |
441 | return nr; |
442 | } | |
443 | ||
444 | #define val_byte(val, n) (((__u8 *)&val)[n]) | |
445 | ||
446 | #define pushb(base, ptr, val, err_label) \ | |
447 | do { \ | |
448 | __u8 __val = val; \ | |
449 | ptr--; \ | |
450 | if (put_user(__val, base + ptr) < 0) \ | |
451 | goto err_label; \ | |
83e714e8 | 452 | } while (0) |
1da177e4 LT |
453 | |
454 | #define pushw(base, ptr, val, err_label) \ | |
455 | do { \ | |
456 | __u16 __val = val; \ | |
457 | ptr--; \ | |
458 | if (put_user(val_byte(__val, 1), base + ptr) < 0) \ | |
459 | goto err_label; \ | |
460 | ptr--; \ | |
461 | if (put_user(val_byte(__val, 0), base + ptr) < 0) \ | |
462 | goto err_label; \ | |
83e714e8 | 463 | } while (0) |
1da177e4 LT |
464 | |
465 | #define pushl(base, ptr, val, err_label) \ | |
466 | do { \ | |
467 | __u32 __val = val; \ | |
468 | ptr--; \ | |
469 | if (put_user(val_byte(__val, 3), base + ptr) < 0) \ | |
470 | goto err_label; \ | |
471 | ptr--; \ | |
472 | if (put_user(val_byte(__val, 2), base + ptr) < 0) \ | |
473 | goto err_label; \ | |
474 | ptr--; \ | |
475 | if (put_user(val_byte(__val, 1), base + ptr) < 0) \ | |
476 | goto err_label; \ | |
477 | ptr--; \ | |
478 | if (put_user(val_byte(__val, 0), base + ptr) < 0) \ | |
479 | goto err_label; \ | |
83e714e8 | 480 | } while (0) |
1da177e4 LT |
481 | |
482 | #define popb(base, ptr, err_label) \ | |
483 | ({ \ | |
484 | __u8 __res; \ | |
485 | if (get_user(__res, base + ptr) < 0) \ | |
486 | goto err_label; \ | |
487 | ptr++; \ | |
488 | __res; \ | |
489 | }) | |
490 | ||
491 | #define popw(base, ptr, err_label) \ | |
492 | ({ \ | |
493 | __u16 __res; \ | |
494 | if (get_user(val_byte(__res, 0), base + ptr) < 0) \ | |
495 | goto err_label; \ | |
496 | ptr++; \ | |
497 | if (get_user(val_byte(__res, 1), base + ptr) < 0) \ | |
498 | goto err_label; \ | |
499 | ptr++; \ | |
500 | __res; \ | |
501 | }) | |
502 | ||
503 | #define popl(base, ptr, err_label) \ | |
504 | ({ \ | |
505 | __u32 __res; \ | |
506 | if (get_user(val_byte(__res, 0), base + ptr) < 0) \ | |
507 | goto err_label; \ | |
508 | ptr++; \ | |
509 | if (get_user(val_byte(__res, 1), base + ptr) < 0) \ | |
510 | goto err_label; \ | |
511 | ptr++; \ | |
512 | if (get_user(val_byte(__res, 2), base + ptr) < 0) \ | |
513 | goto err_label; \ | |
514 | ptr++; \ | |
515 | if (get_user(val_byte(__res, 3), base + ptr) < 0) \ | |
516 | goto err_label; \ | |
517 | ptr++; \ | |
518 | __res; \ | |
519 | }) | |
520 | ||
521 | /* There are so many possible reasons for this function to return | |
522 | * VM86_INTx, so adding another doesn't bother me. We can expect | |
523 | * userspace programs to be able to handle it. (Getting a problem | |
524 | * in userspace is always better than an Oops anyway.) [KD] | |
525 | */ | |
526 | static void do_int(struct kernel_vm86_regs *regs, int i, | |
83e714e8 | 527 | unsigned char __user *ssp, unsigned short sp) |
1da177e4 LT |
528 | { |
529 | unsigned long __user *intr_ptr; | |
530 | unsigned long segoffs; | |
531 | ||
65ea5b03 | 532 | if (regs->pt.cs == BIOSSEG) |
1da177e4 LT |
533 | goto cannot_handle; |
534 | if (is_revectored(i, &KVM86->int_revectored)) | |
535 | goto cannot_handle; | |
83e714e8 | 536 | if (i == 0x21 && is_revectored(AH(regs), &KVM86->int21_revectored)) |
1da177e4 LT |
537 | goto cannot_handle; |
538 | intr_ptr = (unsigned long __user *) (i << 2); | |
539 | if (get_user(segoffs, intr_ptr)) | |
540 | goto cannot_handle; | |
541 | if ((segoffs >> 16) == BIOSSEG) | |
542 | goto cannot_handle; | |
543 | pushw(ssp, sp, get_vflags(regs), cannot_handle); | |
65ea5b03 | 544 | pushw(ssp, sp, regs->pt.cs, cannot_handle); |
1da177e4 | 545 | pushw(ssp, sp, IP(regs), cannot_handle); |
65ea5b03 | 546 | regs->pt.cs = segoffs >> 16; |
1da177e4 LT |
547 | SP(regs) -= 6; |
548 | IP(regs) = segoffs & 0xffff; | |
549 | clear_TF(regs); | |
550 | clear_IF(regs); | |
551 | clear_AC(regs); | |
552 | return; | |
553 | ||
554 | cannot_handle: | |
555 | return_to_32bit(regs, VM86_INTx + (i << 8)); | |
556 | } | |
557 | ||
83e714e8 | 558 | int handle_vm86_trap(struct kernel_vm86_regs *regs, long error_code, int trapno) |
1da177e4 LT |
559 | { |
560 | if (VMPI.is_vm86pus) { | |
6554287b BO |
561 | if ((trapno == 3) || (trapno == 1)) { |
562 | KVM86->regs32->ax = VM86_TRAP + (trapno << 8); | |
563 | /* setting this flag forces the code in entry_32.S to | |
e76623d6 AV |
564 | the path where we call save_v86_state() and change |
565 | the stack pointer to KVM86->regs32 */ | |
566 | set_thread_flag(TIF_NOTIFY_RESUME); | |
6554287b BO |
567 | return 0; |
568 | } | |
65ea5b03 | 569 | do_int(regs, trapno, (unsigned char __user *) (regs->pt.ss << 4), SP(regs)); |
1da177e4 LT |
570 | return 0; |
571 | } | |
83e714e8 | 572 | if (trapno != 1) |
1da177e4 | 573 | return 1; /* we let this handle by the calling routine */ |
51e7dc70 | 574 | current->thread.trap_nr = trapno; |
1da177e4 | 575 | current->thread.error_code = error_code; |
0f540910 | 576 | force_sig(SIGTRAP, current); |
1da177e4 LT |
577 | return 0; |
578 | } | |
579 | ||
83e714e8 | 580 | void handle_vm86_fault(struct kernel_vm86_regs *regs, long error_code) |
1da177e4 LT |
581 | { |
582 | unsigned char opcode; | |
583 | unsigned char __user *csp; | |
584 | unsigned char __user *ssp; | |
5fd75ebb | 585 | unsigned short ip, sp, orig_flags; |
1da177e4 LT |
586 | int data32, pref_done; |
587 | ||
588 | #define CHECK_IF_IN_TRAP \ | |
589 | if (VMPI.vm86dbg_active && VMPI.vm86dbg_TFpendig) \ | |
a5c15d41 | 590 | newflags |= X86_EFLAGS_TF |
1da177e4 | 591 | #define VM86_FAULT_RETURN do { \ |
a5c15d41 | 592 | if (VMPI.force_return_for_pic && (VEFLAGS & (X86_EFLAGS_IF | X86_EFLAGS_VIF))) \ |
1da177e4 | 593 | return_to_32bit(regs, VM86_PICRETURN); \ |
a5c15d41 | 594 | if (orig_flags & X86_EFLAGS_TF) \ |
5fd75ebb | 595 | handle_vm86_trap(regs, 0, 1); \ |
1da177e4 LT |
596 | return; } while (0) |
597 | ||
65ea5b03 | 598 | orig_flags = *(unsigned short *)®s->pt.flags; |
5fd75ebb | 599 | |
65ea5b03 PA |
600 | csp = (unsigned char __user *) (regs->pt.cs << 4); |
601 | ssp = (unsigned char __user *) (regs->pt.ss << 4); | |
1da177e4 LT |
602 | sp = SP(regs); |
603 | ip = IP(regs); | |
604 | ||
605 | data32 = 0; | |
606 | pref_done = 0; | |
607 | do { | |
608 | switch (opcode = popb(csp, ip, simulate_sigsegv)) { | |
83e714e8 PC |
609 | case 0x66: /* 32-bit data */ data32 = 1; break; |
610 | case 0x67: /* 32-bit address */ break; | |
611 | case 0x2e: /* CS */ break; | |
612 | case 0x3e: /* DS */ break; | |
613 | case 0x26: /* ES */ break; | |
614 | case 0x36: /* SS */ break; | |
615 | case 0x65: /* GS */ break; | |
616 | case 0x64: /* FS */ break; | |
617 | case 0xf2: /* repnz */ break; | |
618 | case 0xf3: /* rep */ break; | |
619 | default: pref_done = 1; | |
1da177e4 LT |
620 | } |
621 | } while (!pref_done); | |
622 | ||
623 | switch (opcode) { | |
624 | ||
625 | /* pushf */ | |
626 | case 0x9c: | |
627 | if (data32) { | |
628 | pushl(ssp, sp, get_vflags(regs), simulate_sigsegv); | |
629 | SP(regs) -= 4; | |
630 | } else { | |
631 | pushw(ssp, sp, get_vflags(regs), simulate_sigsegv); | |
632 | SP(regs) -= 2; | |
633 | } | |
634 | IP(regs) = ip; | |
635 | VM86_FAULT_RETURN; | |
636 | ||
637 | /* popf */ | |
638 | case 0x9d: | |
639 | { | |
640 | unsigned long newflags; | |
641 | if (data32) { | |
83e714e8 | 642 | newflags = popl(ssp, sp, simulate_sigsegv); |
1da177e4 LT |
643 | SP(regs) += 4; |
644 | } else { | |
645 | newflags = popw(ssp, sp, simulate_sigsegv); | |
646 | SP(regs) += 2; | |
647 | } | |
648 | IP(regs) = ip; | |
649 | CHECK_IF_IN_TRAP; | |
83e714e8 | 650 | if (data32) |
1da177e4 | 651 | set_vflags_long(newflags, regs); |
83e714e8 | 652 | else |
1da177e4 | 653 | set_vflags_short(newflags, regs); |
83e714e8 | 654 | |
1da177e4 LT |
655 | VM86_FAULT_RETURN; |
656 | } | |
657 | ||
658 | /* int xx */ | |
659 | case 0xcd: { | |
83e714e8 | 660 | int intno = popb(csp, ip, simulate_sigsegv); |
1da177e4 LT |
661 | IP(regs) = ip; |
662 | if (VMPI.vm86dbg_active) { | |
83e714e8 | 663 | if ((1 << (intno & 7)) & VMPI.vm86dbg_intxxtab[intno >> 3]) |
1da177e4 LT |
664 | return_to_32bit(regs, VM86_INTx + (intno << 8)); |
665 | } | |
666 | do_int(regs, intno, ssp, sp); | |
667 | return; | |
668 | } | |
669 | ||
670 | /* iret */ | |
671 | case 0xcf: | |
672 | { | |
673 | unsigned long newip; | |
674 | unsigned long newcs; | |
675 | unsigned long newflags; | |
676 | if (data32) { | |
83e714e8 PC |
677 | newip = popl(ssp, sp, simulate_sigsegv); |
678 | newcs = popl(ssp, sp, simulate_sigsegv); | |
679 | newflags = popl(ssp, sp, simulate_sigsegv); | |
1da177e4 LT |
680 | SP(regs) += 12; |
681 | } else { | |
682 | newip = popw(ssp, sp, simulate_sigsegv); | |
683 | newcs = popw(ssp, sp, simulate_sigsegv); | |
684 | newflags = popw(ssp, sp, simulate_sigsegv); | |
685 | SP(regs) += 6; | |
686 | } | |
687 | IP(regs) = newip; | |
65ea5b03 | 688 | regs->pt.cs = newcs; |
1da177e4 LT |
689 | CHECK_IF_IN_TRAP; |
690 | if (data32) { | |
691 | set_vflags_long(newflags, regs); | |
692 | } else { | |
693 | set_vflags_short(newflags, regs); | |
694 | } | |
695 | VM86_FAULT_RETURN; | |
696 | } | |
697 | ||
698 | /* cli */ | |
699 | case 0xfa: | |
700 | IP(regs) = ip; | |
701 | clear_IF(regs); | |
702 | VM86_FAULT_RETURN; | |
703 | ||
704 | /* sti */ | |
705 | /* | |
706 | * Damn. This is incorrect: the 'sti' instruction should actually | |
707 | * enable interrupts after the /next/ instruction. Not good. | |
708 | * | |
709 | * Probably needs some horsing around with the TF flag. Aiee.. | |
710 | */ | |
711 | case 0xfb: | |
712 | IP(regs) = ip; | |
713 | set_IF(regs); | |
714 | VM86_FAULT_RETURN; | |
715 | ||
716 | default: | |
717 | return_to_32bit(regs, VM86_UNKNOWN); | |
718 | } | |
719 | ||
720 | return; | |
721 | ||
722 | simulate_sigsegv: | |
723 | /* FIXME: After a long discussion with Stas we finally | |
724 | * agreed, that this is wrong. Here we should | |
725 | * really send a SIGSEGV to the user program. | |
726 | * But how do we create the correct context? We | |
727 | * are inside a general protection fault handler | |
728 | * and has just returned from a page fault handler. | |
729 | * The correct context for the signal handler | |
730 | * should be a mixture of the two, but how do we | |
731 | * get the information? [KD] | |
732 | */ | |
733 | return_to_32bit(regs, VM86_UNKNOWN); | |
734 | } | |
735 | ||
736 | /* ---------------- vm86 special IRQ passing stuff ----------------- */ | |
737 | ||
738 | #define VM86_IRQNAME "vm86irq" | |
739 | ||
740 | static struct vm86_irqs { | |
741 | struct task_struct *tsk; | |
742 | int sig; | |
743 | } vm86_irqs[16]; | |
744 | ||
745 | static DEFINE_SPINLOCK(irqbits_lock); | |
746 | static int irqbits; | |
747 | ||
83e714e8 | 748 | #define ALLOWED_SIGS (1 /* 0 = don't send a signal */ \ |
1da177e4 | 749 | | (1 << SIGUSR1) | (1 << SIGUSR2) | (1 << SIGIO) | (1 << SIGURG) \ |
83e714e8 PC |
750 | | (1 << SIGUNUSED)) |
751 | ||
7d12e780 | 752 | static irqreturn_t irq_handler(int intno, void *dev_id) |
1da177e4 LT |
753 | { |
754 | int irq_bit; | |
755 | unsigned long flags; | |
756 | ||
83e714e8 | 757 | spin_lock_irqsave(&irqbits_lock, flags); |
1da177e4 | 758 | irq_bit = 1 << intno; |
83e714e8 | 759 | if ((irqbits & irq_bit) || !vm86_irqs[intno].tsk) |
1da177e4 LT |
760 | goto out; |
761 | irqbits |= irq_bit; | |
762 | if (vm86_irqs[intno].sig) | |
763 | send_sig(vm86_irqs[intno].sig, vm86_irqs[intno].tsk, 1); | |
1da177e4 LT |
764 | /* |
765 | * IRQ will be re-enabled when user asks for the irq (whether | |
766 | * polling or as a result of the signal) | |
767 | */ | |
ad671423 PP |
768 | disable_irq_nosync(intno); |
769 | spin_unlock_irqrestore(&irqbits_lock, flags); | |
1da177e4 LT |
770 | return IRQ_HANDLED; |
771 | ||
772 | out: | |
83e714e8 | 773 | spin_unlock_irqrestore(&irqbits_lock, flags); |
1da177e4 LT |
774 | return IRQ_NONE; |
775 | } | |
776 | ||
777 | static inline void free_vm86_irq(int irqnumber) | |
778 | { | |
779 | unsigned long flags; | |
780 | ||
781 | free_irq(irqnumber, NULL); | |
782 | vm86_irqs[irqnumber].tsk = NULL; | |
783 | ||
83e714e8 | 784 | spin_lock_irqsave(&irqbits_lock, flags); |
1da177e4 | 785 | irqbits &= ~(1 << irqnumber); |
83e714e8 | 786 | spin_unlock_irqrestore(&irqbits_lock, flags); |
1da177e4 LT |
787 | } |
788 | ||
789 | void release_vm86_irqs(struct task_struct *task) | |
790 | { | |
791 | int i; | |
792 | for (i = FIRST_VM86_IRQ ; i <= LAST_VM86_IRQ; i++) | |
793 | if (vm86_irqs[i].tsk == task) | |
794 | free_vm86_irq(i); | |
795 | } | |
796 | ||
797 | static inline int get_and_reset_irq(int irqnumber) | |
798 | { | |
799 | int bit; | |
800 | unsigned long flags; | |
ad671423 | 801 | int ret = 0; |
83e714e8 | 802 | |
1da177e4 LT |
803 | if (invalid_vm86_irq(irqnumber)) return 0; |
804 | if (vm86_irqs[irqnumber].tsk != current) return 0; | |
83e714e8 | 805 | spin_lock_irqsave(&irqbits_lock, flags); |
1da177e4 LT |
806 | bit = irqbits & (1 << irqnumber); |
807 | irqbits &= ~bit; | |
ad671423 PP |
808 | if (bit) { |
809 | enable_irq(irqnumber); | |
810 | ret = 1; | |
811 | } | |
812 | ||
83e714e8 | 813 | spin_unlock_irqrestore(&irqbits_lock, flags); |
ad671423 | 814 | return ret; |
1da177e4 LT |
815 | } |
816 | ||
817 | ||
818 | static int do_vm86_irq_handling(int subfunction, int irqnumber) | |
819 | { | |
820 | int ret; | |
821 | switch (subfunction) { | |
822 | case VM86_GET_AND_RESET_IRQ: { | |
823 | return get_and_reset_irq(irqnumber); | |
824 | } | |
825 | case VM86_GET_IRQ_BITS: { | |
826 | return irqbits; | |
827 | } | |
828 | case VM86_REQUEST_IRQ: { | |
829 | int sig = irqnumber >> 8; | |
830 | int irq = irqnumber & 255; | |
831 | if (!capable(CAP_SYS_ADMIN)) return -EPERM; | |
832 | if (!((1 << sig) & ALLOWED_SIGS)) return -EPERM; | |
833 | if (invalid_vm86_irq(irq)) return -EPERM; | |
834 | if (vm86_irqs[irq].tsk) return -EPERM; | |
835 | ret = request_irq(irq, &irq_handler, 0, VM86_IRQNAME, NULL); | |
836 | if (ret) return ret; | |
837 | vm86_irqs[irq].sig = sig; | |
838 | vm86_irqs[irq].tsk = current; | |
839 | return irq; | |
840 | } | |
841 | case VM86_FREE_IRQ: { | |
842 | if (invalid_vm86_irq(irqnumber)) return -EPERM; | |
843 | if (!vm86_irqs[irqnumber].tsk) return 0; | |
844 | if (vm86_irqs[irqnumber].tsk != current) return -EPERM; | |
845 | free_vm86_irq(irqnumber); | |
846 | return 0; | |
847 | } | |
848 | } | |
849 | return -EINVAL; | |
850 | } | |
851 |