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/errno.h>
34 #include <linux/sched.h>
35 #include <linux/kernel.h>
36 #include <linux/signal.h>
37 #include <linux/string.h>
38 #include <linux/ptrace.h>
40 #include <linux/smp.h>
41 #include <linux/smp_lock.h>
43 #include <asm/uaccess.h>
44 #include <asm/pgalloc.h>
51 * Interrupt handling is not guaranteed:
52 * - a real x86 will disable all interrupts for one instruction
53 * after a "mov ss,xx" to make stack handling atomic even without
54 * the 'lss' instruction. We can't guarantee this in v86 mode,
55 * as the next instruction might result in a page fault or similar.
56 * - a real x86 will have interrupts disabled for one instruction
57 * past the 'sti' that enables them. We don't bother with all the
60 * Let's hope these problems do not actually matter for anything.
64 #define KVM86 ((struct kernel_vm86_struct *)regs)
65 #define VMPI KVM86->vm86plus
69 * 8- and 16-bit register defines..
71 #define AL(regs) (((unsigned char *)&((regs)->eax))[0])
72 #define AH(regs) (((unsigned char *)&((regs)->eax))[1])
73 #define IP(regs) (*(unsigned short *)&((regs)->eip))
74 #define SP(regs) (*(unsigned short *)&((regs)->esp))
77 * virtual flags (16 and 32-bit versions)
79 #define VFLAGS (*(unsigned short *)&(current->thread.v86flags))
80 #define VEFLAGS (current->thread.v86flags)
82 #define set_flags(X,new,mask) \
83 ((X) = ((X) & ~(mask)) | ((new) & (mask)))
85 #define SAFE_MASK (0xDD5)
86 #define RETURN_MASK (0xDFF)
88 #define VM86_REGS_PART2 orig_eax
89 #define VM86_REGS_SIZE1 \
90 ( (unsigned)( & (((struct kernel_vm86_regs *)0)->VM86_REGS_PART2) ) )
91 #define VM86_REGS_SIZE2 (sizeof(struct kernel_vm86_regs) - VM86_REGS_SIZE1)
93 struct pt_regs * FASTCALL(save_v86_state(struct kernel_vm86_regs * regs));
94 struct pt_regs * fastcall save_v86_state(struct kernel_vm86_regs * regs)
96 struct tss_struct *tss;
100 if (!current->thread.vm86_info) {
101 printk("no vm86_info: BAD\n");
104 set_flags(regs->eflags, VEFLAGS, VIF_MASK | current->thread.v86mask);
105 tmp = copy_to_user(¤t->thread.vm86_info->regs,regs, VM86_REGS_SIZE1);
106 tmp += copy_to_user(¤t->thread.vm86_info->regs.VM86_REGS_PART2,
107 ®s->VM86_REGS_PART2, VM86_REGS_SIZE2);
108 tmp += put_user(current->thread.screen_bitmap,¤t->thread.vm86_info->screen_bitmap);
110 printk("vm86: could not access userspace vm86_info\n");
113 tss = init_tss + smp_processor_id();
114 tss->esp0 = current->thread.esp0 = current->thread.saved_esp0;
115 current->thread.saved_esp0 = 0;
120 static void mark_screen_rdonly(struct task_struct * tsk)
127 spin_lock(&tsk->mm->page_table_lock);
128 pgd = pgd_offset(tsk->mm, 0xA0000);
136 pmd = pmd_offset(pgd, 0xA0000);
144 pte = pte_offset(pmd, 0xA0000);
145 for (i = 0; i < 32; i++) {
146 if (pte_present(*pte))
147 set_pte(pte, pte_wrprotect(*pte));
151 spin_unlock(&tsk->mm->page_table_lock);
157 static int do_vm86_irq_handling(int subfunction, int irqnumber);
158 static void do_sys_vm86(struct kernel_vm86_struct *info, struct task_struct *tsk);
160 asmlinkage int sys_vm86old(struct vm86_struct * v86)
162 struct kernel_vm86_struct info; /* declare this _on top_,
163 * this avoids wasting of stack space.
164 * This remains on the stack until we
165 * return to 32 bit user space.
167 struct task_struct *tsk;
168 int tmp, ret = -EPERM;
171 if (tsk->thread.saved_esp0)
173 tmp = copy_from_user(&info, v86, VM86_REGS_SIZE1);
174 tmp += copy_from_user(&info.regs.VM86_REGS_PART2, &v86->regs.VM86_REGS_PART2,
175 (long)&info.vm86plus - (long)&info.regs.VM86_REGS_PART2);
179 memset(&info.vm86plus, 0, (int)&info.regs32 - (int)&info.vm86plus);
180 info.regs32 = (struct pt_regs *) &v86;
181 tsk->thread.vm86_info = v86;
182 do_sys_vm86(&info, tsk);
183 ret = 0; /* we never return here */
189 asmlinkage int sys_vm86(unsigned long subfunction, struct vm86plus_struct * v86)
191 struct kernel_vm86_struct info; /* declare this _on top_,
192 * this avoids wasting of stack space.
193 * This remains on the stack until we
194 * return to 32 bit user space.
196 struct task_struct *tsk;
200 switch (subfunction) {
201 case VM86_REQUEST_IRQ:
203 case VM86_GET_IRQ_BITS:
204 case VM86_GET_AND_RESET_IRQ:
205 ret = do_vm86_irq_handling(subfunction,(int)v86);
207 case VM86_PLUS_INSTALL_CHECK:
208 /* NOTE: on old vm86 stuff this will return the error
209 from verify_area(), because the subfunction is
210 interpreted as (invalid) address to vm86_struct.
211 So the installation check works.
217 /* we come here only for functions VM86_ENTER, VM86_ENTER_NO_BYPASS */
219 if (tsk->thread.saved_esp0)
221 tmp = copy_from_user(&info, v86, VM86_REGS_SIZE1);
222 tmp += copy_from_user(&info.regs.VM86_REGS_PART2, &v86->regs.VM86_REGS_PART2,
223 (long)&info.regs32 - (long)&info.regs.VM86_REGS_PART2);
227 info.regs32 = (struct pt_regs *) &subfunction;
228 info.vm86plus.is_vm86pus = 1;
229 tsk->thread.vm86_info = (struct vm86_struct *)v86;
230 do_sys_vm86(&info, tsk);
231 ret = 0; /* we never return here */
237 static void do_sys_vm86(struct kernel_vm86_struct *info, struct task_struct *tsk)
239 struct tss_struct *tss;
241 * make sure the vm86() system call doesn't try to do anything silly
243 info->regs.__null_ds = 0;
244 info->regs.__null_es = 0;
246 /* we are clearing fs,gs later just before "jmp ret_from_sys_call",
247 * because starting with Linux 2.1.x they aren't no longer saved/restored
251 * The eflags register is also special: we cannot trust that the user
252 * has set it up safely, so this makes sure interrupt etc flags are
253 * inherited from protected mode.
255 VEFLAGS = info->regs.eflags;
256 info->regs.eflags &= SAFE_MASK;
257 info->regs.eflags |= info->regs32->eflags & ~SAFE_MASK;
258 info->regs.eflags |= VM_MASK;
260 switch (info->cpu_type) {
262 tsk->thread.v86mask = 0;
265 tsk->thread.v86mask = NT_MASK | IOPL_MASK;
268 tsk->thread.v86mask = AC_MASK | NT_MASK | IOPL_MASK;
271 tsk->thread.v86mask = ID_MASK | AC_MASK | NT_MASK | IOPL_MASK;
276 * Save old state, set default return value (%eax) to 0
278 info->regs32->eax = 0;
279 tsk->thread.saved_esp0 = tsk->thread.esp0;
280 tss = init_tss + smp_processor_id();
281 tss->esp0 = tsk->thread.esp0 = (unsigned long) &info->VM86_TSS_ESP0;
283 tsk->thread.screen_bitmap = info->screen_bitmap;
284 if (info->flags & VM86_SCREEN_BITMAP)
285 mark_screen_rdonly(tsk);
286 __asm__ __volatile__(
287 "xorl %%eax,%%eax; movl %%eax,%%fs; movl %%eax,%%gs\n\t"
289 "jmp ret_from_sys_call"
291 :"r" (&info->regs), "b" (tsk) : "ax");
292 /* we never return here */
295 static inline void return_to_32bit(struct kernel_vm86_regs * regs16, int retval)
297 struct pt_regs * regs32;
299 regs32 = save_v86_state(regs16);
300 regs32->eax = retval;
301 __asm__ __volatile__("movl %0,%%esp\n\t"
302 "jmp ret_from_sys_call"
303 : : "r" (regs32), "b" (current));
306 static inline void set_IF(struct kernel_vm86_regs * regs)
309 if (VEFLAGS & VIP_MASK)
310 return_to_32bit(regs, VM86_STI);
313 static inline void clear_IF(struct kernel_vm86_regs * regs)
315 VEFLAGS &= ~VIF_MASK;
318 static inline void clear_TF(struct kernel_vm86_regs * regs)
320 regs->eflags &= ~TF_MASK;
323 static inline void clear_AC(struct kernel_vm86_regs * regs)
325 regs->eflags &= ~AC_MASK;
328 /* It is correct to call set_IF(regs) from the set_vflags_*
329 * functions. However someone forgot to call clear_IF(regs)
330 * in the opposite case.
331 * After the command sequence CLI PUSHF STI POPF you should
332 * end up with interrups disabled, but you ended up with
333 * interrupts enabled.
334 * ( I was testing my own changes, but the only bug I
335 * could find was in a function I had not changed. )
339 static inline void set_vflags_long(unsigned long eflags, struct kernel_vm86_regs * regs)
341 set_flags(VEFLAGS, eflags, current->thread.v86mask);
342 set_flags(regs->eflags, eflags, SAFE_MASK);
343 if (eflags & IF_MASK)
349 static inline void set_vflags_short(unsigned short flags, struct kernel_vm86_regs * regs)
351 set_flags(VFLAGS, flags, current->thread.v86mask);
352 set_flags(regs->eflags, flags, SAFE_MASK);
359 static inline unsigned long get_vflags(struct kernel_vm86_regs * regs)
361 unsigned long flags = regs->eflags & RETURN_MASK;
363 if (VEFLAGS & VIF_MASK)
368 return flags | (VEFLAGS & current->thread.v86mask);
371 static inline int is_revectored(int nr, struct revectored_struct * bitmap)
373 __asm__ __volatile__("btl %2,%1\n\tsbbl %0,%0"
375 :"m" (*bitmap),"r" (nr));
379 #define val_byte(val, n) (((__u8 *)&val)[n])
381 #define pushb(base, ptr, val, err_label) \
385 if (put_user(__val, base + ptr) < 0) \
389 #define pushw(base, ptr, val, err_label) \
393 if (put_user(val_byte(__val, 1), base + ptr) < 0) \
396 if (put_user(val_byte(__val, 0), base + ptr) < 0) \
400 #define pushl(base, ptr, val, err_label) \
404 if (put_user(val_byte(__val, 3), base + ptr) < 0) \
407 if (put_user(val_byte(__val, 2), base + ptr) < 0) \
410 if (put_user(val_byte(__val, 1), base + ptr) < 0) \
413 if (put_user(val_byte(__val, 0), base + ptr) < 0) \
417 #define popb(base, ptr, err_label) \
420 if (get_user(__res, base + ptr) < 0) \
426 #define popw(base, ptr, err_label) \
429 if (get_user(val_byte(__res, 0), base + ptr) < 0) \
432 if (get_user(val_byte(__res, 1), base + ptr) < 0) \
438 #define popl(base, ptr, err_label) \
441 if (get_user(val_byte(__res, 0), base + ptr) < 0) \
444 if (get_user(val_byte(__res, 1), base + ptr) < 0) \
447 if (get_user(val_byte(__res, 2), base + ptr) < 0) \
450 if (get_user(val_byte(__res, 3), base + ptr) < 0) \
456 /* There are so many possible reasons for this function to return
457 * VM86_INTx, so adding another doesn't bother me. We can expect
458 * userspace programs to be able to handle it. (Getting a problem
459 * in userspace is always better than an Oops anyway.) [KD]
461 static void do_int(struct kernel_vm86_regs *regs, int i,
462 unsigned char * ssp, unsigned short sp)
464 unsigned long *intr_ptr, segoffs;
466 if (regs->cs == BIOSSEG)
468 if (is_revectored(i, &KVM86->int_revectored))
470 if (i==0x21 && is_revectored(AH(regs),&KVM86->int21_revectored))
472 intr_ptr = (unsigned long *) (i << 2);
473 if (get_user(segoffs, intr_ptr))
475 if ((segoffs >> 16) == BIOSSEG)
477 pushw(ssp, sp, get_vflags(regs), cannot_handle);
478 pushw(ssp, sp, regs->cs, cannot_handle);
479 pushw(ssp, sp, IP(regs), cannot_handle);
480 regs->cs = segoffs >> 16;
482 IP(regs) = segoffs & 0xffff;
489 return_to_32bit(regs, VM86_INTx + (i << 8));
492 int handle_vm86_trap(struct kernel_vm86_regs * regs, long error_code, int trapno)
494 if (VMPI.is_vm86pus) {
495 if ( (trapno==3) || (trapno==1) )
496 return_to_32bit(regs, VM86_TRAP + (trapno << 8));
497 do_int(regs, trapno, (unsigned char *) (regs->ss << 4), SP(regs));
501 return 1; /* we let this handle by the calling routine */
502 if (current->ptrace & PT_PTRACED) {
504 spin_lock_irqsave(¤t->sigmask_lock, flags);
505 sigdelset(¤t->blocked, SIGTRAP);
506 recalc_sigpending(current);
507 spin_unlock_irqrestore(¤t->sigmask_lock, flags);
509 send_sig(SIGTRAP, current, 1);
510 current->thread.trap_no = trapno;
511 current->thread.error_code = error_code;
515 void handle_vm86_fault(struct kernel_vm86_regs * regs, long error_code)
517 unsigned char *csp, *ssp, opcode;
518 unsigned short ip, sp;
519 int data32, pref_done;
521 #define CHECK_IF_IN_TRAP \
522 if (VMPI.vm86dbg_active && VMPI.vm86dbg_TFpendig) \
524 #define VM86_FAULT_RETURN do { \
525 if (VMPI.force_return_for_pic && (VEFLAGS & (IF_MASK | VIF_MASK))) \
526 return_to_32bit(regs, VM86_PICRETURN); \
529 csp = (unsigned char *) (regs->cs << 4);
530 ssp = (unsigned char *) (regs->ss << 4);
537 switch (opcode = popb(csp, ip, simulate_sigsegv)) {
538 case 0x66: /* 32-bit data */ data32=1; break;
539 case 0x67: /* 32-bit address */ break;
540 case 0x2e: /* CS */ break;
541 case 0x3e: /* DS */ break;
542 case 0x26: /* ES */ break;
543 case 0x36: /* SS */ break;
544 case 0x65: /* GS */ break;
545 case 0x64: /* FS */ break;
546 case 0xf2: /* repnz */ break;
547 case 0xf3: /* rep */ break;
548 default: pref_done = 1;
550 } while (!pref_done);
557 pushl(ssp, sp, get_vflags(regs), simulate_sigsegv);
560 pushw(ssp, sp, get_vflags(regs), simulate_sigsegv);
569 unsigned long newflags;
571 newflags=popl(ssp, sp, simulate_sigsegv);
574 newflags = popw(ssp, sp, simulate_sigsegv);
580 set_vflags_long(newflags, regs);
582 set_vflags_short(newflags, regs);
589 int intno=popb(csp, ip, simulate_sigsegv);
591 if (VMPI.vm86dbg_active) {
592 if ( (1 << (intno &7)) & VMPI.vm86dbg_intxxtab[intno >> 3] )
593 return_to_32bit(regs, VM86_INTx + (intno << 8));
595 do_int(regs, intno, ssp, sp);
604 unsigned long newflags;
606 newip=popl(ssp, sp, simulate_sigsegv);
607 newcs=popl(ssp, sp, simulate_sigsegv);
608 newflags=popl(ssp, sp, simulate_sigsegv);
611 newip = popw(ssp, sp, simulate_sigsegv);
612 newcs = popw(ssp, sp, simulate_sigsegv);
613 newflags = popw(ssp, sp, simulate_sigsegv);
620 set_vflags_long(newflags, regs);
622 set_vflags_short(newflags, regs);
635 * Damn. This is incorrect: the 'sti' instruction should actually
636 * enable interrupts after the /next/ instruction. Not good.
638 * Probably needs some horsing around with the TF flag. Aiee..
646 return_to_32bit(regs, VM86_UNKNOWN);
652 /* FIXME: After a long discussion with Stas we finally
653 * agreed, that this is wrong. Here we should
654 * really send a SIGSEGV to the user program.
655 * But how do we create the correct context? We
656 * are inside a general protection fault handler
657 * and has just returned from a page fault handler.
658 * The correct context for the signal handler
659 * should be a mixture of the two, but how do we
660 * get the information? [KD]
662 return_to_32bit(regs, VM86_UNKNOWN);
665 /* ---------------- vm86 special IRQ passing stuff ----------------- */
667 #define VM86_IRQNAME "vm86irq"
669 static struct vm86_irqs {
670 struct task_struct *tsk;
675 #define ALLOWED_SIGS ( 1 /* 0 = don't send a signal */ \
676 | (1 << SIGUSR1) | (1 << SIGUSR2) | (1 << SIGIO) | (1 << SIGURG) \
679 static void irq_handler(int intno, void *dev_id, struct pt_regs * regs) {
685 irq_bit = 1 << intno;
686 if ((irqbits & irq_bit) || ! vm86_irqs[intno].tsk)
689 if (vm86_irqs[intno].sig)
690 send_sig(vm86_irqs[intno].sig, vm86_irqs[intno].tsk, 1);
691 /* else user will poll for IRQs */
693 restore_flags(flags);
696 static inline void free_vm86_irq(int irqnumber)
698 free_irq(irqnumber,0);
699 vm86_irqs[irqnumber].tsk = 0;
700 irqbits &= ~(1 << irqnumber);
703 static inline int task_valid(struct task_struct *tsk)
705 struct task_struct *p;
708 read_lock(&tasklist_lock);
710 if ((p == tsk) && (p->sig)) {
715 read_unlock(&tasklist_lock);
719 void release_x86_irqs(struct task_struct *task)
723 if (vm86_irqs[i].tsk == task)
727 static inline void handle_irq_zombies(void)
730 for (i=3; i<16; i++) {
731 if (vm86_irqs[i].tsk) {
732 if (task_valid(vm86_irqs[i].tsk)) continue;
738 static inline int get_and_reset_irq(int irqnumber)
743 if ( (irqnumber<3) || (irqnumber>15) ) return 0;
744 if (vm86_irqs[irqnumber].tsk != current) return 0;
747 bit = irqbits & (1 << irqnumber);
749 restore_flags(flags);
754 static int do_vm86_irq_handling(int subfunction, int irqnumber)
757 switch (subfunction) {
758 case VM86_GET_AND_RESET_IRQ: {
759 return get_and_reset_irq(irqnumber);
761 case VM86_GET_IRQ_BITS: {
764 case VM86_REQUEST_IRQ: {
765 int sig = irqnumber >> 8;
766 int irq = irqnumber & 255;
767 handle_irq_zombies();
768 if (!capable(CAP_SYS_ADMIN)) return -EPERM;
769 if (!((1 << sig) & ALLOWED_SIGS)) return -EPERM;
770 if ( (irq<3) || (irq>15) ) return -EPERM;
771 if (vm86_irqs[irq].tsk) return -EPERM;
772 ret = request_irq(irq, &irq_handler, 0, VM86_IRQNAME, 0);
774 vm86_irqs[irq].sig = sig;
775 vm86_irqs[irq].tsk = current;
778 case VM86_FREE_IRQ: {
779 handle_irq_zombies();
780 if ( (irqnumber<3) || (irqnumber>15) ) return -EPERM;
781 if (!vm86_irqs[irqnumber].tsk) return 0;
782 if (vm86_irqs[irqnumber].tsk != current) return -EPERM;
783 free_vm86_irq(irqnumber);