2 * Copyright (C) 1994 Linus Torvalds
4 * 29 dec 2001 - Fixed oopses caused by unchecked access to the vm86
5 * stack - Manfred Spraul <manfred@colorfullife.com>
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>
12 * 22 mar 2002 - Added missing clear_IF in set_vflags_* Kasper Dupont
13 * <kasperd@daimi.au.dk>
15 * ?? ??? 2002 - Fixed premature returns from handle_vm86_fault
16 * caused by Kasper Dupont's changes - Stas Sergeev
18 * 4 apr 2002 - Fixed CHECK_IF_IN_TRAP broken by Stas' changes.
19 * Kasper Dupont <kasperd@daimi.au.dk>
21 * 9 apr 2002 - Changed syntax of macros in handle_vm86_fault.
22 * Kasper Dupont <kasperd@daimi.au.dk>
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>
31 #include <linux/capability.h>
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>
39 #include <linux/smp.h>
40 #include <linux/highmem.h>
41 #include <linux/ptrace.h>
42 #include <linux/audit.h>
43 #include <linux/stddef.h>
45 #include <asm/uaccess.h>
47 #include <asm/tlbflush.h>
49 #include <asm/syscalls.h>
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
63 * Let's hope these problems do not actually matter for anything.
67 #define KVM86 ((struct kernel_vm86_struct *)regs)
68 #define VMPI KVM86->vm86plus
72 * 8- and 16-bit register defines..
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))
80 * virtual flags (16 and 32-bit versions)
82 #define VFLAGS (*(unsigned short *)&(current->thread.v86flags))
83 #define VEFLAGS (current->thread.v86flags)
85 #define set_flags(X, new, mask) \
86 ((X) = ((X) & ~(mask)) | ((new) & (mask)))
88 #define SAFE_MASK (0xDD5)
89 #define RETURN_MASK (0xDFF)
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)
98 * kernel_vm86_regs is missing gs, so copy everything up to
99 * (but not including) orig_eax, and then rest including orig_eax.
101 ret += copy_to_user(user, regs, offsetof(struct kernel_vm86_regs, pt.orig_ax));
102 ret += copy_to_user(&user->orig_eax, ®s->pt.orig_ax,
103 sizeof(struct kernel_vm86_regs) -
104 offsetof(struct kernel_vm86_regs, pt.orig_ax));
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,
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,
120 sizeof(struct kernel_vm86_regs) -
121 offsetof(struct kernel_vm86_regs, pt.orig_ax) +
126 struct pt_regs *save_v86_state(struct kernel_vm86_regs *regs)
128 struct tss_struct *tss;
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.
139 if (!current->thread.vm86_info) {
140 printk("no vm86_info: BAD\n");
143 set_flags(regs->pt.flags, VEFLAGS, X86_EFLAGS_VIF | current->thread.v86mask);
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);
147 printk("vm86: could not access userspace vm86_info\n");
151 tss = &per_cpu(init_tss, get_cpu());
152 current->thread.sp0 = current->thread.saved_sp0;
153 current->thread.sysenter_cs = __KERNEL_CS;
154 load_sp0(tss, ¤t->thread);
155 current->thread.saved_sp0 = 0;
160 ret->fs = current->thread.saved_fs;
161 set_user_gs(ret, current->thread.saved_gs);
166 static void mark_screen_rdonly(struct mm_struct *mm)
175 pgd = pgd_offset(mm, 0xA0000);
176 if (pgd_none_or_clear_bad(pgd))
178 pud = pud_offset(pgd, 0xA0000);
179 if (pud_none_or_clear_bad(pud))
181 pmd = pmd_offset(pud, 0xA0000);
182 if (pmd_none_or_clear_bad(pmd))
184 pte = pte_offset_map_lock(mm, pmd, 0xA0000, &ptl);
185 for (i = 0; i < 32; i++) {
186 if (pte_present(*pte))
187 set_pte(pte, pte_wrprotect(*pte));
190 pte_unmap_unlock(pte, ptl);
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);
200 int sys_vm86old(struct pt_regs *regs)
202 struct vm86_struct __user *v86 = (struct vm86_struct __user *)regs->bx;
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.
208 struct task_struct *tsk;
209 int tmp, ret = -EPERM;
212 if (tsk->thread.saved_sp0)
214 tmp = copy_vm86_regs_from_user(&info.regs, &v86->regs,
215 offsetof(struct kernel_vm86_struct, vm86plus) -
220 memset(&info.vm86plus, 0, (int)&info.regs32 - (int)&info.vm86plus);
222 tsk->thread.vm86_info = v86;
223 do_sys_vm86(&info, tsk);
224 ret = 0; /* we never return here */
230 int sys_vm86(struct pt_regs *regs)
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.
237 struct task_struct *tsk;
239 struct vm86plus_struct __user *v86;
243 case VM86_REQUEST_IRQ:
245 case VM86_GET_IRQ_BITS:
246 case VM86_GET_AND_RESET_IRQ:
247 ret = do_vm86_irq_handling(regs->bx, (int)regs->cx);
249 case VM86_PLUS_INSTALL_CHECK:
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.
260 /* we come here only for functions VM86_ENTER, VM86_ENTER_NO_BYPASS */
262 if (tsk->thread.saved_sp0)
264 v86 = (struct vm86plus_struct __user *)regs->cx;
265 tmp = copy_vm86_regs_from_user(&info.regs, &v86->regs,
266 offsetof(struct kernel_vm86_struct, regs32) -
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 */
281 static void do_sys_vm86(struct kernel_vm86_struct *info, struct task_struct *tsk)
283 struct tss_struct *tss;
285 * make sure the vm86() system call doesn't try to do anything silly
287 info->regs.pt.ds = 0;
288 info->regs.pt.es = 0;
289 info->regs.pt.fs = 0;
290 #ifndef CONFIG_X86_32_LAZY_GS
291 info->regs.pt.gs = 0;
295 * The flags register is also special: we cannot trust that the user
296 * has set it up safely, so this makes sure interrupt etc flags are
297 * inherited from protected mode.
299 VEFLAGS = info->regs.pt.flags;
300 info->regs.pt.flags &= SAFE_MASK;
301 info->regs.pt.flags |= info->regs32->flags & ~SAFE_MASK;
302 info->regs.pt.flags |= X86_VM_MASK;
304 switch (info->cpu_type) {
306 tsk->thread.v86mask = 0;
309 tsk->thread.v86mask = X86_EFLAGS_NT | X86_EFLAGS_IOPL;
312 tsk->thread.v86mask = X86_EFLAGS_AC | X86_EFLAGS_NT | X86_EFLAGS_IOPL;
315 tsk->thread.v86mask = X86_EFLAGS_ID | X86_EFLAGS_AC | X86_EFLAGS_NT | X86_EFLAGS_IOPL;
320 * Save old state, set default return value (%ax) to 0
322 info->regs32->ax = 0;
323 tsk->thread.saved_sp0 = tsk->thread.sp0;
324 tsk->thread.saved_fs = info->regs32->fs;
325 tsk->thread.saved_gs = get_user_gs(info->regs32);
327 tss = &per_cpu(init_tss, get_cpu());
328 tsk->thread.sp0 = (unsigned long) &info->VM86_TSS_ESP0;
330 tsk->thread.sysenter_cs = 0;
331 load_sp0(tss, &tsk->thread);
334 tsk->thread.screen_bitmap = info->screen_bitmap;
335 if (info->flags & VM86_SCREEN_BITMAP)
336 mark_screen_rdonly(tsk->mm);
338 /*call audit_syscall_exit since we do not exit via the normal paths */
339 if (unlikely(current->audit_context))
340 audit_syscall_exit(AUDITSC_RESULT(0), 0);
342 __asm__ __volatile__(
345 #ifdef CONFIG_X86_32_LAZY_GS
348 "jmp resume_userspace"
350 :"r" (&info->regs), "r" (task_thread_info(tsk)), "r" (0));
351 /* we never return here */
354 static inline void return_to_32bit(struct kernel_vm86_regs *regs16, int retval)
356 struct pt_regs *regs32;
358 regs32 = save_v86_state(regs16);
360 __asm__ __volatile__("movl %0,%%esp\n\t"
362 "jmp resume_userspace"
363 : : "r" (regs32), "r" (current_thread_info()));
366 static inline void set_IF(struct kernel_vm86_regs *regs)
368 VEFLAGS |= X86_EFLAGS_VIF;
369 if (VEFLAGS & X86_EFLAGS_VIP)
370 return_to_32bit(regs, VM86_STI);
373 static inline void clear_IF(struct kernel_vm86_regs *regs)
375 VEFLAGS &= ~X86_EFLAGS_VIF;
378 static inline void clear_TF(struct kernel_vm86_regs *regs)
380 regs->pt.flags &= ~X86_EFLAGS_TF;
383 static inline void clear_AC(struct kernel_vm86_regs *regs)
385 regs->pt.flags &= ~X86_EFLAGS_AC;
389 * It is correct to call set_IF(regs) from the set_vflags_*
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
393 * end up with interrupts disabled, but you ended up with
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. )
400 static inline void set_vflags_long(unsigned long flags, struct kernel_vm86_regs *regs)
402 set_flags(VEFLAGS, flags, current->thread.v86mask);
403 set_flags(regs->pt.flags, flags, SAFE_MASK);
404 if (flags & X86_EFLAGS_IF)
410 static inline void set_vflags_short(unsigned short flags, struct kernel_vm86_regs *regs)
412 set_flags(VFLAGS, flags, current->thread.v86mask);
413 set_flags(regs->pt.flags, flags, SAFE_MASK);
414 if (flags & X86_EFLAGS_IF)
420 static inline unsigned long get_vflags(struct kernel_vm86_regs *regs)
422 unsigned long flags = regs->pt.flags & RETURN_MASK;
424 if (VEFLAGS & X86_EFLAGS_VIF)
425 flags |= X86_EFLAGS_IF;
426 flags |= X86_EFLAGS_IOPL;
427 return flags | (VEFLAGS & current->thread.v86mask);
430 static inline int is_revectored(int nr, struct revectored_struct *bitmap)
432 __asm__ __volatile__("btl %2,%1\n\tsbbl %0,%0"
434 :"m" (*bitmap), "r" (nr));
438 #define val_byte(val, n) (((__u8 *)&val)[n])
440 #define pushb(base, ptr, val, err_label) \
444 if (put_user(__val, base + ptr) < 0) \
448 #define pushw(base, ptr, val, err_label) \
452 if (put_user(val_byte(__val, 1), base + ptr) < 0) \
455 if (put_user(val_byte(__val, 0), base + ptr) < 0) \
459 #define pushl(base, ptr, val, err_label) \
463 if (put_user(val_byte(__val, 3), base + ptr) < 0) \
466 if (put_user(val_byte(__val, 2), base + ptr) < 0) \
469 if (put_user(val_byte(__val, 1), base + ptr) < 0) \
472 if (put_user(val_byte(__val, 0), base + ptr) < 0) \
476 #define popb(base, ptr, err_label) \
479 if (get_user(__res, base + ptr) < 0) \
485 #define popw(base, ptr, err_label) \
488 if (get_user(val_byte(__res, 0), base + ptr) < 0) \
491 if (get_user(val_byte(__res, 1), base + ptr) < 0) \
497 #define popl(base, ptr, err_label) \
500 if (get_user(val_byte(__res, 0), base + ptr) < 0) \
503 if (get_user(val_byte(__res, 1), base + ptr) < 0) \
506 if (get_user(val_byte(__res, 2), base + ptr) < 0) \
509 if (get_user(val_byte(__res, 3), base + ptr) < 0) \
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]
520 static void do_int(struct kernel_vm86_regs *regs, int i,
521 unsigned char __user *ssp, unsigned short sp)
523 unsigned long __user *intr_ptr;
524 unsigned long segoffs;
526 if (regs->pt.cs == BIOSSEG)
528 if (is_revectored(i, &KVM86->int_revectored))
530 if (i == 0x21 && is_revectored(AH(regs), &KVM86->int21_revectored))
532 intr_ptr = (unsigned long __user *) (i << 2);
533 if (get_user(segoffs, intr_ptr))
535 if ((segoffs >> 16) == BIOSSEG)
537 pushw(ssp, sp, get_vflags(regs), cannot_handle);
538 pushw(ssp, sp, regs->pt.cs, cannot_handle);
539 pushw(ssp, sp, IP(regs), cannot_handle);
540 regs->pt.cs = segoffs >> 16;
542 IP(regs) = segoffs & 0xffff;
549 return_to_32bit(regs, VM86_INTx + (i << 8));
552 int handle_vm86_trap(struct kernel_vm86_regs *regs, long error_code, int trapno)
554 if (VMPI.is_vm86pus) {
555 if ((trapno == 3) || (trapno == 1))
556 return_to_32bit(regs, VM86_TRAP + (trapno << 8));
557 do_int(regs, trapno, (unsigned char __user *) (regs->pt.ss << 4), SP(regs));
561 return 1; /* we let this handle by the calling routine */
562 current->thread.trap_no = trapno;
563 current->thread.error_code = error_code;
564 force_sig(SIGTRAP, current);
568 void handle_vm86_fault(struct kernel_vm86_regs *regs, long error_code)
570 unsigned char opcode;
571 unsigned char __user *csp;
572 unsigned char __user *ssp;
573 unsigned short ip, sp, orig_flags;
574 int data32, pref_done;
576 #define CHECK_IF_IN_TRAP \
577 if (VMPI.vm86dbg_active && VMPI.vm86dbg_TFpendig) \
578 newflags |= X86_EFLAGS_TF
579 #define VM86_FAULT_RETURN do { \
580 if (VMPI.force_return_for_pic && (VEFLAGS & (X86_EFLAGS_IF | X86_EFLAGS_VIF))) \
581 return_to_32bit(regs, VM86_PICRETURN); \
582 if (orig_flags & X86_EFLAGS_TF) \
583 handle_vm86_trap(regs, 0, 1); \
586 orig_flags = *(unsigned short *)®s->pt.flags;
588 csp = (unsigned char __user *) (regs->pt.cs << 4);
589 ssp = (unsigned char __user *) (regs->pt.ss << 4);
596 switch (opcode = popb(csp, ip, simulate_sigsegv)) {
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;
609 } while (!pref_done);
616 pushl(ssp, sp, get_vflags(regs), simulate_sigsegv);
619 pushw(ssp, sp, get_vflags(regs), simulate_sigsegv);
628 unsigned long newflags;
630 newflags = popl(ssp, sp, simulate_sigsegv);
633 newflags = popw(ssp, sp, simulate_sigsegv);
639 set_vflags_long(newflags, regs);
641 set_vflags_short(newflags, regs);
648 int intno = popb(csp, ip, simulate_sigsegv);
650 if (VMPI.vm86dbg_active) {
651 if ((1 << (intno & 7)) & VMPI.vm86dbg_intxxtab[intno >> 3])
652 return_to_32bit(regs, VM86_INTx + (intno << 8));
654 do_int(regs, intno, ssp, sp);
663 unsigned long newflags;
665 newip = popl(ssp, sp, simulate_sigsegv);
666 newcs = popl(ssp, sp, simulate_sigsegv);
667 newflags = popl(ssp, sp, simulate_sigsegv);
670 newip = popw(ssp, sp, simulate_sigsegv);
671 newcs = popw(ssp, sp, simulate_sigsegv);
672 newflags = popw(ssp, sp, simulate_sigsegv);
679 set_vflags_long(newflags, regs);
681 set_vflags_short(newflags, regs);
694 * Damn. This is incorrect: the 'sti' instruction should actually
695 * enable interrupts after the /next/ instruction. Not good.
697 * Probably needs some horsing around with the TF flag. Aiee..
705 return_to_32bit(regs, VM86_UNKNOWN);
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]
721 return_to_32bit(regs, VM86_UNKNOWN);
724 /* ---------------- vm86 special IRQ passing stuff ----------------- */
726 #define VM86_IRQNAME "vm86irq"
728 static struct vm86_irqs {
729 struct task_struct *tsk;
733 static DEFINE_SPINLOCK(irqbits_lock);
736 #define ALLOWED_SIGS (1 /* 0 = don't send a signal */ \
737 | (1 << SIGUSR1) | (1 << SIGUSR2) | (1 << SIGIO) | (1 << SIGURG) \
740 static irqreturn_t irq_handler(int intno, void *dev_id)
745 spin_lock_irqsave(&irqbits_lock, flags);
746 irq_bit = 1 << intno;
747 if ((irqbits & irq_bit) || !vm86_irqs[intno].tsk)
750 if (vm86_irqs[intno].sig)
751 send_sig(vm86_irqs[intno].sig, vm86_irqs[intno].tsk, 1);
753 * IRQ will be re-enabled when user asks for the irq (whether
754 * polling or as a result of the signal)
756 disable_irq_nosync(intno);
757 spin_unlock_irqrestore(&irqbits_lock, flags);
761 spin_unlock_irqrestore(&irqbits_lock, flags);
765 static inline void free_vm86_irq(int irqnumber)
769 free_irq(irqnumber, NULL);
770 vm86_irqs[irqnumber].tsk = NULL;
772 spin_lock_irqsave(&irqbits_lock, flags);
773 irqbits &= ~(1 << irqnumber);
774 spin_unlock_irqrestore(&irqbits_lock, flags);
777 void release_vm86_irqs(struct task_struct *task)
780 for (i = FIRST_VM86_IRQ ; i <= LAST_VM86_IRQ; i++)
781 if (vm86_irqs[i].tsk == task)
785 static inline int get_and_reset_irq(int irqnumber)
791 if (invalid_vm86_irq(irqnumber)) return 0;
792 if (vm86_irqs[irqnumber].tsk != current) return 0;
793 spin_lock_irqsave(&irqbits_lock, flags);
794 bit = irqbits & (1 << irqnumber);
797 enable_irq(irqnumber);
801 spin_unlock_irqrestore(&irqbits_lock, flags);
806 static int do_vm86_irq_handling(int subfunction, int irqnumber)
809 switch (subfunction) {
810 case VM86_GET_AND_RESET_IRQ: {
811 return get_and_reset_irq(irqnumber);
813 case VM86_GET_IRQ_BITS: {
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);
825 vm86_irqs[irq].sig = sig;
826 vm86_irqs[irq].tsk = current;
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);