4 * Copyright (C) 1994 Linus Torvalds
6 * 29 dec 2001 - Fixed oopses caused by unchecked access to the vm86
7 * stack - Manfred Spraul <manfred@colorfullife.com>
9 * 22 mar 2002 - Manfred detected the stackfaults, but didn't handle
10 * them correctly. Now the emulation will be in a
11 * consistent state after stackfaults - Kasper Dupont
12 * <kasperd@daimi.au.dk>
14 * 22 mar 2002 - Added missing clear_IF in set_vflags_* Kasper Dupont
15 * <kasperd@daimi.au.dk>
17 * ?? ??? 2002 - Fixed premature returns from handle_vm86_fault
18 * caused by Kasper Dupont's changes - Stas Sergeev
20 * 4 apr 2002 - Fixed CHECK_IF_IN_TRAP broken by Stas' changes.
21 * Kasper Dupont <kasperd@daimi.au.dk>
23 * 9 apr 2002 - Changed syntax of macros in handle_vm86_fault.
24 * Kasper Dupont <kasperd@daimi.au.dk>
26 * 9 apr 2002 - Changed stack access macros to jump to a label
27 * instead of returning to userspace. This simplifies
28 * do_int, and is needed by handle_vm6_fault. Kasper
29 * Dupont <kasperd@daimi.au.dk>
33 #include <linux/capability.h>
34 #include <linux/errno.h>
35 #include <linux/interrupt.h>
36 #include <linux/sched.h>
37 #include <linux/kernel.h>
38 #include <linux/signal.h>
39 #include <linux/string.h>
41 #include <linux/smp.h>
42 #include <linux/smp_lock.h>
43 #include <linux/highmem.h>
44 #include <linux/ptrace.h>
45 #include <linux/audit.h>
46 #include <linux/stddef.h>
48 #include <asm/uaccess.h>
50 #include <asm/tlbflush.h>
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
65 * Let's hope these problems do not actually matter for anything.
69 #define KVM86 ((struct kernel_vm86_struct *)regs)
70 #define VMPI KVM86->vm86plus
74 * 8- and 16-bit register defines..
76 #define AL(regs) (((unsigned char *)&((regs)->pt.eax))[0])
77 #define AH(regs) (((unsigned char *)&((regs)->pt.eax))[1])
78 #define IP(regs) (*(unsigned short *)&((regs)->pt.eip))
79 #define SP(regs) (*(unsigned short *)&((regs)->pt.esp))
82 * virtual flags (16 and 32-bit versions)
84 #define VFLAGS (*(unsigned short *)&(current->thread.v86flags))
85 #define VEFLAGS (current->thread.v86flags)
87 #define set_flags(X,new,mask) \
88 ((X) = ((X) & ~(mask)) | ((new) & (mask)))
90 #define SAFE_MASK (0xDD5)
91 #define RETURN_MASK (0xDFF)
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)
99 /* kernel_vm86_regs is missing xgs, so copy everything up to
100 (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_eax));
102 ret += copy_to_user(&user->orig_eax, ®s->pt.orig_eax,
103 sizeof(struct kernel_vm86_regs) -
104 offsetof(struct kernel_vm86_regs, pt.orig_eax));
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 eax-xfs inclusive */
117 ret += copy_from_user(regs, user, offsetof(struct kernel_vm86_regs, pt.orig_eax));
118 /* copy orig_eax-__gsh+extra */
119 ret += copy_from_user(®s->pt.orig_eax, &user->orig_eax,
120 sizeof(struct kernel_vm86_regs) -
121 offsetof(struct kernel_vm86_regs, pt.orig_eax) +
126 struct pt_regs * FASTCALL(save_v86_state(struct kernel_vm86_regs * regs));
127 struct pt_regs * fastcall save_v86_state(struct kernel_vm86_regs * regs)
129 struct tss_struct *tss;
134 * This gets called from entry.S with interrupts disabled, but
135 * from process context. Enable interrupts here, before trying
136 * to access user space.
140 if (!current->thread.vm86_info) {
141 printk("no vm86_info: BAD\n");
144 set_flags(regs->pt.eflags, VEFLAGS, VIF_MASK | current->thread.v86mask);
145 tmp = copy_vm86_regs_to_user(¤t->thread.vm86_info->regs,regs);
146 tmp += put_user(current->thread.screen_bitmap,¤t->thread.vm86_info->screen_bitmap);
148 printk("vm86: could not access userspace vm86_info\n");
152 tss = &per_cpu(init_tss, get_cpu());
153 current->thread.esp0 = current->thread.saved_esp0;
154 current->thread.sysenter_cs = __KERNEL_CS;
155 load_esp0(tss, ¤t->thread);
156 current->thread.saved_esp0 = 0;
161 ret->xfs = current->thread.saved_fs;
162 loadsegment(gs, current->thread.saved_gs);
167 static void mark_screen_rdonly(struct mm_struct *mm)
176 pgd = pgd_offset(mm, 0xA0000);
177 if (pgd_none_or_clear_bad(pgd))
179 pud = pud_offset(pgd, 0xA0000);
180 if (pud_none_or_clear_bad(pud))
182 pmd = pmd_offset(pud, 0xA0000);
183 if (pmd_none_or_clear_bad(pmd))
185 pte = pte_offset_map_lock(mm, pmd, 0xA0000, &ptl);
186 for (i = 0; i < 32; i++) {
187 if (pte_present(*pte))
188 set_pte(pte, pte_wrprotect(*pte));
191 pte_unmap_unlock(pte, ptl);
198 static int do_vm86_irq_handling(int subfunction, int irqnumber);
199 static void do_sys_vm86(struct kernel_vm86_struct *info, struct task_struct *tsk);
201 asmlinkage int sys_vm86old(struct pt_regs regs)
203 struct vm86_struct __user *v86 = (struct vm86_struct __user *)regs.ebx;
204 struct kernel_vm86_struct info; /* declare this _on top_,
205 * this avoids wasting of stack space.
206 * This remains on the stack until we
207 * return to 32 bit user space.
209 struct task_struct *tsk;
210 int tmp, ret = -EPERM;
213 if (tsk->thread.saved_esp0)
215 tmp = copy_vm86_regs_from_user(&info.regs, &v86->regs,
216 offsetof(struct kernel_vm86_struct, vm86plus) -
221 memset(&info.vm86plus, 0, (int)&info.regs32 - (int)&info.vm86plus);
223 tsk->thread.vm86_info = v86;
224 do_sys_vm86(&info, tsk);
225 ret = 0; /* we never return here */
231 asmlinkage int sys_vm86(struct pt_regs regs)
233 struct kernel_vm86_struct info; /* declare this _on top_,
234 * this avoids wasting of stack space.
235 * This remains on the stack until we
236 * return to 32 bit user space.
238 struct task_struct *tsk;
240 struct vm86plus_struct __user *v86;
244 case VM86_REQUEST_IRQ:
246 case VM86_GET_IRQ_BITS:
247 case VM86_GET_AND_RESET_IRQ:
248 ret = do_vm86_irq_handling(regs.ebx, (int)regs.ecx);
250 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_esp0)
264 v86 = (struct vm86plus_struct __user *)regs.ecx;
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.xds = 0;
288 info->regs.pt.xes = 0;
289 info->regs.pt.xfs = 0;
291 /* we are clearing gs later just before "jmp resume_userspace",
292 * because it is not saved/restored.
296 * The eflags register is also special: we cannot trust that the user
297 * has set it up safely, so this makes sure interrupt etc flags are
298 * inherited from protected mode.
300 VEFLAGS = info->regs.pt.eflags;
301 info->regs.pt.eflags &= SAFE_MASK;
302 info->regs.pt.eflags |= info->regs32->eflags & ~SAFE_MASK;
303 info->regs.pt.eflags |= VM_MASK;
305 switch (info->cpu_type) {
307 tsk->thread.v86mask = 0;
310 tsk->thread.v86mask = NT_MASK | IOPL_MASK;
313 tsk->thread.v86mask = AC_MASK | NT_MASK | IOPL_MASK;
316 tsk->thread.v86mask = ID_MASK | AC_MASK | NT_MASK | IOPL_MASK;
321 * Save old state, set default return value (%eax) to 0
323 info->regs32->eax = 0;
324 tsk->thread.saved_esp0 = tsk->thread.esp0;
325 tsk->thread.saved_fs = info->regs32->xfs;
326 savesegment(gs, tsk->thread.saved_gs);
328 tss = &per_cpu(init_tss, get_cpu());
329 tsk->thread.esp0 = (unsigned long) &info->VM86_TSS_ESP0;
331 tsk->thread.sysenter_cs = 0;
332 load_esp0(tss, &tsk->thread);
335 tsk->thread.screen_bitmap = info->screen_bitmap;
336 if (info->flags & VM86_SCREEN_BITMAP)
337 mark_screen_rdonly(tsk->mm);
339 /*call audit_syscall_exit since we do not exit via the normal paths */
340 if (unlikely(current->audit_context))
341 audit_syscall_exit(AUDITSC_RESULT(0), 0);
343 __asm__ __volatile__(
347 "jmp resume_userspace"
349 :"r" (&info->regs), "r" (task_thread_info(tsk)), "r" (0));
350 /* we never return here */
353 static inline void return_to_32bit(struct kernel_vm86_regs * regs16, int retval)
355 struct pt_regs * regs32;
357 regs32 = save_v86_state(regs16);
358 regs32->eax = retval;
359 __asm__ __volatile__("movl %0,%%esp\n\t"
361 "jmp resume_userspace"
362 : : "r" (regs32), "r" (current_thread_info()));
365 static inline void set_IF(struct kernel_vm86_regs * regs)
368 if (VEFLAGS & VIP_MASK)
369 return_to_32bit(regs, VM86_STI);
372 static inline void clear_IF(struct kernel_vm86_regs * regs)
374 VEFLAGS &= ~VIF_MASK;
377 static inline void clear_TF(struct kernel_vm86_regs * regs)
379 regs->pt.eflags &= ~TF_MASK;
382 static inline void clear_AC(struct kernel_vm86_regs * regs)
384 regs->pt.eflags &= ~AC_MASK;
387 /* It is correct to call set_IF(regs) from the set_vflags_*
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
391 * end up with interrups disabled, but you ended up with
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. )
398 static inline void set_vflags_long(unsigned long eflags, struct kernel_vm86_regs * regs)
400 set_flags(VEFLAGS, eflags, current->thread.v86mask);
401 set_flags(regs->pt.eflags, eflags, SAFE_MASK);
402 if (eflags & IF_MASK)
408 static inline void set_vflags_short(unsigned short flags, struct kernel_vm86_regs * regs)
410 set_flags(VFLAGS, flags, current->thread.v86mask);
411 set_flags(regs->pt.eflags, flags, SAFE_MASK);
418 static inline unsigned long get_vflags(struct kernel_vm86_regs * regs)
420 unsigned long flags = regs->pt.eflags & RETURN_MASK;
422 if (VEFLAGS & VIF_MASK)
425 return flags | (VEFLAGS & current->thread.v86mask);
428 static inline int is_revectored(int nr, struct revectored_struct * bitmap)
430 __asm__ __volatile__("btl %2,%1\n\tsbbl %0,%0"
432 :"m" (*bitmap),"r" (nr));
436 #define val_byte(val, n) (((__u8 *)&val)[n])
438 #define pushb(base, ptr, val, err_label) \
442 if (put_user(__val, base + ptr) < 0) \
446 #define pushw(base, ptr, val, err_label) \
450 if (put_user(val_byte(__val, 1), base + ptr) < 0) \
453 if (put_user(val_byte(__val, 0), base + ptr) < 0) \
457 #define pushl(base, ptr, val, err_label) \
461 if (put_user(val_byte(__val, 3), base + ptr) < 0) \
464 if (put_user(val_byte(__val, 2), base + ptr) < 0) \
467 if (put_user(val_byte(__val, 1), base + ptr) < 0) \
470 if (put_user(val_byte(__val, 0), base + ptr) < 0) \
474 #define popb(base, ptr, err_label) \
477 if (get_user(__res, base + ptr) < 0) \
483 #define popw(base, ptr, err_label) \
486 if (get_user(val_byte(__res, 0), base + ptr) < 0) \
489 if (get_user(val_byte(__res, 1), base + ptr) < 0) \
495 #define popl(base, ptr, err_label) \
498 if (get_user(val_byte(__res, 0), base + ptr) < 0) \
501 if (get_user(val_byte(__res, 1), base + ptr) < 0) \
504 if (get_user(val_byte(__res, 2), base + ptr) < 0) \
507 if (get_user(val_byte(__res, 3), base + ptr) < 0) \
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]
518 static void do_int(struct kernel_vm86_regs *regs, int i,
519 unsigned char __user * ssp, unsigned short sp)
521 unsigned long __user *intr_ptr;
522 unsigned long segoffs;
524 if (regs->pt.xcs == BIOSSEG)
526 if (is_revectored(i, &KVM86->int_revectored))
528 if (i==0x21 && is_revectored(AH(regs),&KVM86->int21_revectored))
530 intr_ptr = (unsigned long __user *) (i << 2);
531 if (get_user(segoffs, intr_ptr))
533 if ((segoffs >> 16) == BIOSSEG)
535 pushw(ssp, sp, get_vflags(regs), cannot_handle);
536 pushw(ssp, sp, regs->pt.xcs, cannot_handle);
537 pushw(ssp, sp, IP(regs), cannot_handle);
538 regs->pt.xcs = segoffs >> 16;
540 IP(regs) = segoffs & 0xffff;
547 return_to_32bit(regs, VM86_INTx + (i << 8));
550 int handle_vm86_trap(struct kernel_vm86_regs * regs, long error_code, int trapno)
552 if (VMPI.is_vm86pus) {
553 if ( (trapno==3) || (trapno==1) )
554 return_to_32bit(regs, VM86_TRAP + (trapno << 8));
555 do_int(regs, trapno, (unsigned char __user *) (regs->pt.xss << 4), SP(regs));
559 return 1; /* we let this handle by the calling routine */
560 if (current->ptrace & PT_PTRACED) {
562 spin_lock_irqsave(¤t->sighand->siglock, flags);
563 sigdelset(¤t->blocked, SIGTRAP);
565 spin_unlock_irqrestore(¤t->sighand->siglock, flags);
567 send_sig(SIGTRAP, current, 1);
568 current->thread.trap_no = trapno;
569 current->thread.error_code = error_code;
573 void handle_vm86_fault(struct kernel_vm86_regs * regs, long error_code)
575 unsigned char opcode;
576 unsigned char __user *csp;
577 unsigned char __user *ssp;
578 unsigned short ip, sp, orig_flags;
579 int data32, pref_done;
581 #define CHECK_IF_IN_TRAP \
582 if (VMPI.vm86dbg_active && VMPI.vm86dbg_TFpendig) \
584 #define VM86_FAULT_RETURN do { \
585 if (VMPI.force_return_for_pic && (VEFLAGS & (IF_MASK | VIF_MASK))) \
586 return_to_32bit(regs, VM86_PICRETURN); \
587 if (orig_flags & TF_MASK) \
588 handle_vm86_trap(regs, 0, 1); \
591 orig_flags = *(unsigned short *)®s->pt.eflags;
593 csp = (unsigned char __user *) (regs->pt.xcs << 4);
594 ssp = (unsigned char __user *) (regs->pt.xss << 4);
601 switch (opcode = popb(csp, ip, simulate_sigsegv)) {
602 case 0x66: /* 32-bit data */ data32=1; break;
603 case 0x67: /* 32-bit address */ break;
604 case 0x2e: /* CS */ break;
605 case 0x3e: /* DS */ break;
606 case 0x26: /* ES */ break;
607 case 0x36: /* SS */ break;
608 case 0x65: /* GS */ break;
609 case 0x64: /* FS */ break;
610 case 0xf2: /* repnz */ break;
611 case 0xf3: /* rep */ break;
612 default: pref_done = 1;
614 } while (!pref_done);
621 pushl(ssp, sp, get_vflags(regs), simulate_sigsegv);
624 pushw(ssp, sp, get_vflags(regs), simulate_sigsegv);
633 unsigned long newflags;
635 newflags=popl(ssp, sp, simulate_sigsegv);
638 newflags = popw(ssp, sp, simulate_sigsegv);
644 set_vflags_long(newflags, regs);
646 set_vflags_short(newflags, regs);
653 int intno=popb(csp, ip, simulate_sigsegv);
655 if (VMPI.vm86dbg_active) {
656 if ( (1 << (intno &7)) & VMPI.vm86dbg_intxxtab[intno >> 3] )
657 return_to_32bit(regs, VM86_INTx + (intno << 8));
659 do_int(regs, intno, ssp, sp);
668 unsigned long newflags;
670 newip=popl(ssp, sp, simulate_sigsegv);
671 newcs=popl(ssp, sp, simulate_sigsegv);
672 newflags=popl(ssp, sp, simulate_sigsegv);
675 newip = popw(ssp, sp, simulate_sigsegv);
676 newcs = popw(ssp, sp, simulate_sigsegv);
677 newflags = popw(ssp, sp, simulate_sigsegv);
681 regs->pt.xcs = newcs;
684 set_vflags_long(newflags, regs);
686 set_vflags_short(newflags, regs);
699 * Damn. This is incorrect: the 'sti' instruction should actually
700 * enable interrupts after the /next/ instruction. Not good.
702 * Probably needs some horsing around with the TF flag. Aiee..
710 return_to_32bit(regs, VM86_UNKNOWN);
716 /* FIXME: After a long discussion with Stas we finally
717 * agreed, that this is wrong. Here we should
718 * really send a SIGSEGV to the user program.
719 * But how do we create the correct context? We
720 * are inside a general protection fault handler
721 * and has just returned from a page fault handler.
722 * The correct context for the signal handler
723 * should be a mixture of the two, but how do we
724 * get the information? [KD]
726 return_to_32bit(regs, VM86_UNKNOWN);
729 /* ---------------- vm86 special IRQ passing stuff ----------------- */
731 #define VM86_IRQNAME "vm86irq"
733 static struct vm86_irqs {
734 struct task_struct *tsk;
738 static DEFINE_SPINLOCK(irqbits_lock);
741 #define ALLOWED_SIGS ( 1 /* 0 = don't send a signal */ \
742 | (1 << SIGUSR1) | (1 << SIGUSR2) | (1 << SIGIO) | (1 << SIGURG) \
745 static irqreturn_t irq_handler(int intno, void *dev_id)
750 spin_lock_irqsave(&irqbits_lock, flags);
751 irq_bit = 1 << intno;
752 if ((irqbits & irq_bit) || ! vm86_irqs[intno].tsk)
755 if (vm86_irqs[intno].sig)
756 send_sig(vm86_irqs[intno].sig, vm86_irqs[intno].tsk, 1);
758 * IRQ will be re-enabled when user asks for the irq (whether
759 * polling or as a result of the signal)
761 disable_irq_nosync(intno);
762 spin_unlock_irqrestore(&irqbits_lock, flags);
766 spin_unlock_irqrestore(&irqbits_lock, flags);
770 static inline void free_vm86_irq(int irqnumber)
774 free_irq(irqnumber, NULL);
775 vm86_irqs[irqnumber].tsk = NULL;
777 spin_lock_irqsave(&irqbits_lock, flags);
778 irqbits &= ~(1 << irqnumber);
779 spin_unlock_irqrestore(&irqbits_lock, flags);
782 void release_vm86_irqs(struct task_struct *task)
785 for (i = FIRST_VM86_IRQ ; i <= LAST_VM86_IRQ; i++)
786 if (vm86_irqs[i].tsk == task)
790 static inline int get_and_reset_irq(int irqnumber)
796 if (invalid_vm86_irq(irqnumber)) return 0;
797 if (vm86_irqs[irqnumber].tsk != current) return 0;
798 spin_lock_irqsave(&irqbits_lock, flags);
799 bit = irqbits & (1 << irqnumber);
802 enable_irq(irqnumber);
806 spin_unlock_irqrestore(&irqbits_lock, flags);
811 static int do_vm86_irq_handling(int subfunction, int irqnumber)
814 switch (subfunction) {
815 case VM86_GET_AND_RESET_IRQ: {
816 return get_and_reset_irq(irqnumber);
818 case VM86_GET_IRQ_BITS: {
821 case VM86_REQUEST_IRQ: {
822 int sig = irqnumber >> 8;
823 int irq = irqnumber & 255;
824 if (!capable(CAP_SYS_ADMIN)) return -EPERM;
825 if (!((1 << sig) & ALLOWED_SIGS)) return -EPERM;
826 if (invalid_vm86_irq(irq)) return -EPERM;
827 if (vm86_irqs[irq].tsk) return -EPERM;
828 ret = request_irq(irq, &irq_handler, 0, VM86_IRQNAME, NULL);
830 vm86_irqs[irq].sig = sig;
831 vm86_irqs[irq].tsk = current;
834 case VM86_FREE_IRQ: {
835 if (invalid_vm86_irq(irqnumber)) return -EPERM;
836 if (!vm86_irqs[irqnumber].tsk) return 0;
837 if (vm86_irqs[irqnumber].tsk != current) return -EPERM;
838 free_vm86_irq(irqnumber);