4 * Copyright (C) 1994 Linus Torvalds
6 * 29 dec 2001 - Fixed oopses caused by unchecked access to the vm86
7 * stack - Manfred Spraul <manfreds@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/config.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>
46 #include <asm/uaccess.h>
47 #include <asm/pgalloc.h>
49 #include <asm/tlbflush.h>
55 * Interrupt handling is not guaranteed:
56 * - a real x86 will disable all interrupts for one instruction
57 * after a "mov ss,xx" to make stack handling atomic even without
58 * the 'lss' instruction. We can't guarantee this in v86 mode,
59 * as the next instruction might result in a page fault or similar.
60 * - a real x86 will have interrupts disabled for one instruction
61 * past the 'sti' that enables them. We don't bother with all the
64 * Let's hope these problems do not actually matter for anything.
68 #define KVM86 ((struct kernel_vm86_struct *)regs)
69 #define VMPI KVM86->vm86plus
73 * 8- and 16-bit register defines..
75 #define AL(regs) (((unsigned char *)&((regs)->eax))[0])
76 #define AH(regs) (((unsigned char *)&((regs)->eax))[1])
77 #define IP(regs) (*(unsigned short *)&((regs)->eip))
78 #define SP(regs) (*(unsigned short *)&((regs)->esp))
81 * virtual flags (16 and 32-bit versions)
83 #define VFLAGS (*(unsigned short *)&(current->thread.v86flags))
84 #define VEFLAGS (current->thread.v86flags)
86 #define set_flags(X,new,mask) \
87 ((X) = ((X) & ~(mask)) | ((new) & (mask)))
89 #define SAFE_MASK (0xDD5)
90 #define RETURN_MASK (0xDFF)
92 #define VM86_REGS_PART2 orig_eax
93 #define VM86_REGS_SIZE1 \
94 ( (unsigned)( & (((struct kernel_vm86_regs *)0)->VM86_REGS_PART2) ) )
95 #define VM86_REGS_SIZE2 (sizeof(struct kernel_vm86_regs) - VM86_REGS_SIZE1)
97 struct pt_regs * FASTCALL(save_v86_state(struct kernel_vm86_regs * regs));
98 struct pt_regs * fastcall save_v86_state(struct kernel_vm86_regs * regs)
100 struct tss_struct *tss;
105 * This gets called from entry.S with interrupts disabled, but
106 * from process context. Enable interrupts here, before trying
107 * to access user space.
111 if (!current->thread.vm86_info) {
112 printk("no vm86_info: BAD\n");
115 set_flags(regs->eflags, VEFLAGS, VIF_MASK | current->thread.v86mask);
116 tmp = copy_to_user(¤t->thread.vm86_info->regs,regs, VM86_REGS_SIZE1);
117 tmp += copy_to_user(¤t->thread.vm86_info->regs.VM86_REGS_PART2,
118 ®s->VM86_REGS_PART2, VM86_REGS_SIZE2);
119 tmp += put_user(current->thread.screen_bitmap,¤t->thread.vm86_info->screen_bitmap);
121 printk("vm86: could not access userspace vm86_info\n");
125 tss = init_tss + get_cpu();
126 current->thread.esp0 = current->thread.saved_esp0;
127 current->thread.sysenter_cs = __KERNEL_CS;
128 load_virtual_esp0(tss, current);
129 current->thread.saved_esp0 = 0;
132 loadsegment(fs, current->thread.saved_fs);
133 loadsegment(gs, current->thread.saved_gs);
138 static void mark_screen_rdonly(struct task_struct * tsk)
146 spin_lock(&tsk->mm->page_table_lock);
147 pgd = pgd_offset(tsk->mm, 0xA0000);
155 pmd = pmd_offset(pgd, 0xA0000);
163 pte = mapped = pte_offset_map(pmd, 0xA0000);
164 for (i = 0; i < 32; i++) {
165 if (pte_present(*pte))
166 set_pte(pte, pte_wrprotect(*pte));
171 spin_unlock(&tsk->mm->page_table_lock);
178 static int do_vm86_irq_handling(int subfunction, int irqnumber);
179 static void do_sys_vm86(struct kernel_vm86_struct *info, struct task_struct *tsk);
181 asmlinkage int sys_vm86old(struct vm86_struct __user * v86)
183 struct kernel_vm86_struct info; /* declare this _on top_,
184 * this avoids wasting of stack space.
185 * This remains on the stack until we
186 * return to 32 bit user space.
188 struct task_struct *tsk;
189 int tmp, ret = -EPERM;
192 if (tsk->thread.saved_esp0)
194 tmp = copy_from_user(&info, v86, VM86_REGS_SIZE1);
195 tmp += copy_from_user(&info.regs.VM86_REGS_PART2, &v86->regs.VM86_REGS_PART2,
196 (long)&info.vm86plus - (long)&info.regs.VM86_REGS_PART2);
200 memset(&info.vm86plus, 0, (int)&info.regs32 - (int)&info.vm86plus);
201 info.regs32 = (struct pt_regs *) &v86;
202 tsk->thread.vm86_info = v86;
203 do_sys_vm86(&info, tsk);
204 ret = 0; /* we never return here */
210 asmlinkage int sys_vm86(unsigned long subfunction, struct vm86plus_struct __user * v86)
212 struct kernel_vm86_struct info; /* declare this _on top_,
213 * this avoids wasting of stack space.
214 * This remains on the stack until we
215 * return to 32 bit user space.
217 struct task_struct *tsk;
221 switch (subfunction) {
222 case VM86_REQUEST_IRQ:
224 case VM86_GET_IRQ_BITS:
225 case VM86_GET_AND_RESET_IRQ:
226 ret = do_vm86_irq_handling(subfunction,(int)v86);
228 case VM86_PLUS_INSTALL_CHECK:
229 /* NOTE: on old vm86 stuff this will return the error
230 from verify_area(), because the subfunction is
231 interpreted as (invalid) address to vm86_struct.
232 So the installation check works.
238 /* we come here only for functions VM86_ENTER, VM86_ENTER_NO_BYPASS */
240 if (tsk->thread.saved_esp0)
242 tmp = copy_from_user(&info, v86, VM86_REGS_SIZE1);
243 tmp += copy_from_user(&info.regs.VM86_REGS_PART2, &v86->regs.VM86_REGS_PART2,
244 (long)&info.regs32 - (long)&info.regs.VM86_REGS_PART2);
248 info.regs32 = (struct pt_regs *) &subfunction;
249 info.vm86plus.is_vm86pus = 1;
250 tsk->thread.vm86_info = (struct vm86_struct __user *)v86;
251 do_sys_vm86(&info, tsk);
252 ret = 0; /* we never return here */
258 static void do_sys_vm86(struct kernel_vm86_struct *info, struct task_struct *tsk)
260 struct tss_struct *tss;
262 * make sure the vm86() system call doesn't try to do anything silly
264 info->regs.__null_ds = 0;
265 info->regs.__null_es = 0;
267 /* we are clearing fs,gs later just before "jmp resume_userspace",
268 * because starting with Linux 2.1.x they aren't no longer saved/restored
272 * The eflags register is also special: we cannot trust that the user
273 * has set it up safely, so this makes sure interrupt etc flags are
274 * inherited from protected mode.
276 VEFLAGS = info->regs.eflags;
277 info->regs.eflags &= SAFE_MASK;
278 info->regs.eflags |= info->regs32->eflags & ~SAFE_MASK;
279 info->regs.eflags |= VM_MASK;
281 switch (info->cpu_type) {
283 tsk->thread.v86mask = 0;
286 tsk->thread.v86mask = NT_MASK | IOPL_MASK;
289 tsk->thread.v86mask = AC_MASK | NT_MASK | IOPL_MASK;
292 tsk->thread.v86mask = ID_MASK | AC_MASK | NT_MASK | IOPL_MASK;
297 * Save old state, set default return value (%eax) to 0
299 info->regs32->eax = 0;
300 tsk->thread.saved_esp0 = tsk->thread.esp0;
301 asm volatile("movl %%fs,%0":"=m" (tsk->thread.saved_fs));
302 asm volatile("movl %%gs,%0":"=m" (tsk->thread.saved_gs));
304 tss = init_tss + get_cpu();
305 tsk->thread.esp0 = (unsigned long) &info->VM86_TSS_ESP0;
307 tsk->thread.sysenter_cs = 0;
308 load_virtual_esp0(tss, tsk);
311 tsk->thread.screen_bitmap = info->screen_bitmap;
312 if (info->flags & VM86_SCREEN_BITMAP)
313 mark_screen_rdonly(tsk);
314 __asm__ __volatile__(
315 "xorl %%eax,%%eax; movl %%eax,%%fs; movl %%eax,%%gs\n\t"
318 "jmp resume_userspace"
320 :"r" (&info->regs), "r" (tsk->thread_info) : "ax");
321 /* we never return here */
324 static inline void return_to_32bit(struct kernel_vm86_regs * regs16, int retval)
326 struct pt_regs * regs32;
328 regs32 = save_v86_state(regs16);
329 regs32->eax = retval;
330 __asm__ __volatile__("movl %0,%%esp\n\t"
332 "jmp resume_userspace"
333 : : "r" (regs32), "r" (current_thread_info()));
336 static inline void set_IF(struct kernel_vm86_regs * regs)
339 if (VEFLAGS & VIP_MASK)
340 return_to_32bit(regs, VM86_STI);
343 static inline void clear_IF(struct kernel_vm86_regs * regs)
345 VEFLAGS &= ~VIF_MASK;
348 static inline void clear_TF(struct kernel_vm86_regs * regs)
350 regs->eflags &= ~TF_MASK;
353 static inline void clear_AC(struct kernel_vm86_regs * regs)
355 regs->eflags &= ~AC_MASK;
358 /* It is correct to call set_IF(regs) from the set_vflags_*
359 * functions. However someone forgot to call clear_IF(regs)
360 * in the opposite case.
361 * After the command sequence CLI PUSHF STI POPF you should
362 * end up with interrups disabled, but you ended up with
363 * interrupts enabled.
364 * ( I was testing my own changes, but the only bug I
365 * could find was in a function I had not changed. )
369 static inline void set_vflags_long(unsigned long eflags, struct kernel_vm86_regs * regs)
371 set_flags(VEFLAGS, eflags, current->thread.v86mask);
372 set_flags(regs->eflags, eflags, SAFE_MASK);
373 if (eflags & IF_MASK)
379 static inline void set_vflags_short(unsigned short flags, struct kernel_vm86_regs * regs)
381 set_flags(VFLAGS, flags, current->thread.v86mask);
382 set_flags(regs->eflags, flags, SAFE_MASK);
389 static inline unsigned long get_vflags(struct kernel_vm86_regs * regs)
391 unsigned long flags = regs->eflags & RETURN_MASK;
393 if (VEFLAGS & VIF_MASK)
395 return flags | (VEFLAGS & current->thread.v86mask);
398 static inline int is_revectored(int nr, struct revectored_struct * bitmap)
400 __asm__ __volatile__("btl %2,%1\n\tsbbl %0,%0"
402 :"m" (*bitmap),"r" (nr));
406 #define val_byte(val, n) (((__u8 *)&val)[n])
408 #define pushb(base, ptr, val, err_label) \
412 if (put_user(__val, base + ptr) < 0) \
416 #define pushw(base, ptr, val, err_label) \
420 if (put_user(val_byte(__val, 1), base + ptr) < 0) \
423 if (put_user(val_byte(__val, 0), base + ptr) < 0) \
427 #define pushl(base, ptr, val, err_label) \
431 if (put_user(val_byte(__val, 3), base + ptr) < 0) \
434 if (put_user(val_byte(__val, 2), base + ptr) < 0) \
437 if (put_user(val_byte(__val, 1), base + ptr) < 0) \
440 if (put_user(val_byte(__val, 0), base + ptr) < 0) \
444 #define popb(base, ptr, err_label) \
447 if (get_user(__res, base + ptr) < 0) \
453 #define popw(base, ptr, err_label) \
456 if (get_user(val_byte(__res, 0), base + ptr) < 0) \
459 if (get_user(val_byte(__res, 1), base + ptr) < 0) \
465 #define popl(base, ptr, err_label) \
468 if (get_user(val_byte(__res, 0), base + ptr) < 0) \
471 if (get_user(val_byte(__res, 1), base + ptr) < 0) \
474 if (get_user(val_byte(__res, 2), base + ptr) < 0) \
477 if (get_user(val_byte(__res, 3), base + ptr) < 0) \
483 /* There are so many possible reasons for this function to return
484 * VM86_INTx, so adding another doesn't bother me. We can expect
485 * userspace programs to be able to handle it. (Getting a problem
486 * in userspace is always better than an Oops anyway.) [KD]
488 static void do_int(struct kernel_vm86_regs *regs, int i,
489 unsigned char * ssp, unsigned short sp)
491 unsigned long *intr_ptr, segoffs;
493 if (regs->cs == BIOSSEG)
495 if (is_revectored(i, &KVM86->int_revectored))
497 if (i==0x21 && is_revectored(AH(regs),&KVM86->int21_revectored))
499 intr_ptr = (unsigned long *) (i << 2);
500 if (get_user(segoffs, intr_ptr))
502 if ((segoffs >> 16) == BIOSSEG)
504 pushw(ssp, sp, get_vflags(regs), cannot_handle);
505 pushw(ssp, sp, regs->cs, cannot_handle);
506 pushw(ssp, sp, IP(regs), cannot_handle);
507 regs->cs = segoffs >> 16;
509 IP(regs) = segoffs & 0xffff;
516 return_to_32bit(regs, VM86_INTx + (i << 8));
519 int handle_vm86_trap(struct kernel_vm86_regs * regs, long error_code, int trapno)
521 if (VMPI.is_vm86pus) {
522 if ( (trapno==3) || (trapno==1) )
523 return_to_32bit(regs, VM86_TRAP + (trapno << 8));
524 do_int(regs, trapno, (unsigned char *) (regs->ss << 4), SP(regs));
528 return 1; /* we let this handle by the calling routine */
529 if (current->ptrace & PT_PTRACED) {
531 spin_lock_irqsave(¤t->sighand->siglock, flags);
532 sigdelset(¤t->blocked, SIGTRAP);
534 spin_unlock_irqrestore(¤t->sighand->siglock, flags);
536 send_sig(SIGTRAP, current, 1);
537 current->thread.trap_no = trapno;
538 current->thread.error_code = error_code;
542 void handle_vm86_fault(struct kernel_vm86_regs * regs, long error_code)
544 unsigned char *csp, *ssp, opcode;
545 unsigned short ip, sp;
546 int data32, pref_done;
548 #define CHECK_IF_IN_TRAP \
549 if (VMPI.vm86dbg_active && VMPI.vm86dbg_TFpendig) \
551 #define VM86_FAULT_RETURN do { \
552 if (VMPI.force_return_for_pic && (VEFLAGS & (IF_MASK | VIF_MASK))) \
553 return_to_32bit(regs, VM86_PICRETURN); \
556 csp = (unsigned char *) (regs->cs << 4);
557 ssp = (unsigned char *) (regs->ss << 4);
564 switch (opcode = popb(csp, ip, simulate_sigsegv)) {
565 case 0x66: /* 32-bit data */ data32=1; break;
566 case 0x67: /* 32-bit address */ break;
567 case 0x2e: /* CS */ break;
568 case 0x3e: /* DS */ break;
569 case 0x26: /* ES */ break;
570 case 0x36: /* SS */ break;
571 case 0x65: /* GS */ break;
572 case 0x64: /* FS */ break;
573 case 0xf2: /* repnz */ break;
574 case 0xf3: /* rep */ break;
575 default: pref_done = 1;
577 } while (!pref_done);
584 pushl(ssp, sp, get_vflags(regs), simulate_sigsegv);
587 pushw(ssp, sp, get_vflags(regs), simulate_sigsegv);
596 unsigned long newflags;
598 newflags=popl(ssp, sp, simulate_sigsegv);
601 newflags = popw(ssp, sp, simulate_sigsegv);
607 set_vflags_long(newflags, regs);
609 set_vflags_short(newflags, regs);
616 int intno=popb(csp, ip, simulate_sigsegv);
618 if (VMPI.vm86dbg_active) {
619 if ( (1 << (intno &7)) & VMPI.vm86dbg_intxxtab[intno >> 3] )
620 return_to_32bit(regs, VM86_INTx + (intno << 8));
622 do_int(regs, intno, ssp, sp);
631 unsigned long newflags;
633 newip=popl(ssp, sp, simulate_sigsegv);
634 newcs=popl(ssp, sp, simulate_sigsegv);
635 newflags=popl(ssp, sp, simulate_sigsegv);
638 newip = popw(ssp, sp, simulate_sigsegv);
639 newcs = popw(ssp, sp, simulate_sigsegv);
640 newflags = popw(ssp, sp, simulate_sigsegv);
647 set_vflags_long(newflags, regs);
649 set_vflags_short(newflags, regs);
662 * Damn. This is incorrect: the 'sti' instruction should actually
663 * enable interrupts after the /next/ instruction. Not good.
665 * Probably needs some horsing around with the TF flag. Aiee..
673 return_to_32bit(regs, VM86_UNKNOWN);
679 /* FIXME: After a long discussion with Stas we finally
680 * agreed, that this is wrong. Here we should
681 * really send a SIGSEGV to the user program.
682 * But how do we create the correct context? We
683 * are inside a general protection fault handler
684 * and has just returned from a page fault handler.
685 * The correct context for the signal handler
686 * should be a mixture of the two, but how do we
687 * get the information? [KD]
689 return_to_32bit(regs, VM86_UNKNOWN);
692 /* ---------------- vm86 special IRQ passing stuff ----------------- */
694 #define VM86_IRQNAME "vm86irq"
696 static struct vm86_irqs {
697 struct task_struct *tsk;
701 static spinlock_t irqbits_lock = SPIN_LOCK_UNLOCKED;
704 #define ALLOWED_SIGS ( 1 /* 0 = don't send a signal */ \
705 | (1 << SIGUSR1) | (1 << SIGUSR2) | (1 << SIGIO) | (1 << SIGURG) \
708 static irqreturn_t irq_handler(int intno, void *dev_id, struct pt_regs * regs)
713 spin_lock_irqsave(&irqbits_lock, flags);
714 irq_bit = 1 << intno;
715 if ((irqbits & irq_bit) || ! vm86_irqs[intno].tsk)
718 if (vm86_irqs[intno].sig)
719 send_sig(vm86_irqs[intno].sig, vm86_irqs[intno].tsk, 1);
720 /* else user will poll for IRQs */
722 spin_unlock_irqrestore(&irqbits_lock, flags);
726 static inline void free_vm86_irq(int irqnumber)
730 free_irq(irqnumber,0);
731 vm86_irqs[irqnumber].tsk = 0;
733 spin_lock_irqsave(&irqbits_lock, flags);
734 irqbits &= ~(1 << irqnumber);
735 spin_unlock_irqrestore(&irqbits_lock, flags);
738 void release_x86_irqs(struct task_struct *task)
741 for (i = FIRST_VM86_IRQ ; i <= LAST_VM86_IRQ; i++)
742 if (vm86_irqs[i].tsk == task)
746 static inline int get_and_reset_irq(int irqnumber)
751 if (invalid_vm86_irq(irqnumber)) return 0;
752 if (vm86_irqs[irqnumber].tsk != current) return 0;
753 spin_lock_irqsave(&irqbits_lock, flags);
754 bit = irqbits & (1 << irqnumber);
756 spin_unlock_irqrestore(&irqbits_lock, flags);
759 enable_irq(irqnumber);
764 static int do_vm86_irq_handling(int subfunction, int irqnumber)
767 switch (subfunction) {
768 case VM86_GET_AND_RESET_IRQ: {
769 return get_and_reset_irq(irqnumber);
771 case VM86_GET_IRQ_BITS: {
774 case VM86_REQUEST_IRQ: {
775 int sig = irqnumber >> 8;
776 int irq = irqnumber & 255;
777 if (!capable(CAP_SYS_ADMIN)) return -EPERM;
778 if (!((1 << sig) & ALLOWED_SIGS)) return -EPERM;
779 if (invalid_vm86_irq(irq)) return -EPERM;
780 if (vm86_irqs[irq].tsk) return -EPERM;
781 ret = request_irq(irq, &irq_handler, 0, VM86_IRQNAME, 0);
783 vm86_irqs[irq].sig = sig;
784 vm86_irqs[irq].tsk = current;
787 case VM86_FREE_IRQ: {
788 if (invalid_vm86_irq(irqnumber)) return -EPERM;
789 if (!vm86_irqs[irqnumber].tsk) return 0;
790 if (vm86_irqs[irqnumber].tsk != current) return -EPERM;
791 free_vm86_irq(irqnumber);