blob: 5edb27f1a2c407ff8173161fb87cbd9130b76e8e [file] [log] [blame]
Greg Kroah-Hartmanb2441312017-11-01 15:07:57 +01001// SPDX-License-Identifier: GPL-2.0
Linus Torvalds1da177e2005-04-16 15:20:36 -07002/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003 * Copyright (C) 1994 Linus Torvalds
4 *
5 * 29 dec 2001 - Fixed oopses caused by unchecked access to the vm86
Christian Kujau624dffc2006-01-15 02:43:54 +01006 * stack - Manfred Spraul <manfred@colorfullife.com>
Linus Torvalds1da177e2005-04-16 15:20:36 -07007 *
8 * 22 mar 2002 - Manfred detected the stackfaults, but didn't handle
9 * them correctly. Now the emulation will be in a
10 * consistent state after stackfaults - Kasper Dupont
11 * <kasperd@daimi.au.dk>
12 *
13 * 22 mar 2002 - Added missing clear_IF in set_vflags_* Kasper Dupont
14 * <kasperd@daimi.au.dk>
15 *
16 * ?? ??? 2002 - Fixed premature returns from handle_vm86_fault
17 * caused by Kasper Dupont's changes - Stas Sergeev
18 *
19 * 4 apr 2002 - Fixed CHECK_IF_IN_TRAP broken by Stas' changes.
20 * Kasper Dupont <kasperd@daimi.au.dk>
21 *
22 * 9 apr 2002 - Changed syntax of macros in handle_vm86_fault.
23 * Kasper Dupont <kasperd@daimi.au.dk>
24 *
25 * 9 apr 2002 - Changed stack access macros to jump to a label
26 * instead of returning to userspace. This simplifies
27 * do_int, and is needed by handle_vm6_fault. Kasper
28 * Dupont <kasperd@daimi.au.dk>
29 *
30 */
31
Joe Perchesc767a542012-05-21 19:50:07 -070032#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
33
Randy Dunlapa9415642006-01-11 12:17:48 -080034#include <linux/capability.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070035#include <linux/errno.h>
36#include <linux/interrupt.h>
Alexander van Heukelum5522ddb2013-03-27 22:18:05 +010037#include <linux/syscalls.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070038#include <linux/sched.h>
Ingo Molnar68db0cf2017-02-08 18:51:37 +010039#include <linux/sched/task_stack.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070040#include <linux/kernel.h>
41#include <linux/signal.h>
42#include <linux/string.h>
43#include <linux/mm.h>
44#include <linux/smp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070045#include <linux/highmem.h>
46#include <linux/ptrace.h>
Jason Baron7e7f8a02006-01-31 16:56:28 -050047#include <linux/audit.h>
Jeremy Fitzhardinge49d26b62006-12-07 02:14:03 +010048#include <linux/stddef.h>
Brian Gerst9fda6a02015-07-29 01:41:16 -040049#include <linux/slab.h>
Andy Lutomirski76fc5e72015-09-04 17:00:43 -070050#include <linux/security.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070051
Linus Torvalds7c0f6ba2016-12-24 11:46:01 -080052#include <linux/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070053#include <asm/io.h>
54#include <asm/tlbflush.h>
55#include <asm/irq.h>
Brian Gerst5ed92a82015-07-29 01:41:19 -040056#include <asm/traps.h>
Brian Gerstba3e1272015-07-29 01:41:21 -040057#include <asm/vm86.h>
Andy Lutomirskibd7dc5a2017-11-02 00:59:09 -070058#include <asm/switch_to.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070059
60/*
61 * Known problems:
62 *
63 * Interrupt handling is not guaranteed:
64 * - a real x86 will disable all interrupts for one instruction
65 * after a "mov ss,xx" to make stack handling atomic even without
66 * the 'lss' instruction. We can't guarantee this in v86 mode,
67 * as the next instruction might result in a page fault or similar.
68 * - a real x86 will have interrupts disabled for one instruction
69 * past the 'sti' that enables them. We don't bother with all the
70 * details yet.
71 *
72 * Let's hope these problems do not actually matter for anything.
73 */
74
75
Linus Torvalds1da177e2005-04-16 15:20:36 -070076/*
77 * 8- and 16-bit register defines..
78 */
H. Peter Anvin65ea5b02008-01-30 13:30:56 +010079#define AL(regs) (((unsigned char *)&((regs)->pt.ax))[0])
80#define AH(regs) (((unsigned char *)&((regs)->pt.ax))[1])
81#define IP(regs) (*(unsigned short *)&((regs)->pt.ip))
82#define SP(regs) (*(unsigned short *)&((regs)->pt.sp))
Linus Torvalds1da177e2005-04-16 15:20:36 -070083
84/*
85 * virtual flags (16 and 32-bit versions)
86 */
Brian Gerstdecd2752015-07-29 01:41:23 -040087#define VFLAGS (*(unsigned short *)&(current->thread.vm86->veflags))
88#define VEFLAGS (current->thread.vm86->veflags)
Linus Torvalds1da177e2005-04-16 15:20:36 -070089
Paolo Ciarrocchi83e714e2008-02-22 23:10:40 +010090#define set_flags(X, new, mask) \
Linus Torvalds1da177e2005-04-16 15:20:36 -070091((X) = ((X) & ~(mask)) | ((new) & (mask)))
92
93#define SAFE_MASK (0xDD5)
94#define RETURN_MASK (0xDFF)
95
Brian Gerst5ed92a82015-07-29 01:41:19 -040096void save_v86_state(struct kernel_vm86_regs *regs, int retval)
Linus Torvalds1da177e2005-04-16 15:20:36 -070097{
Brian Gersted0b2ed2015-07-19 21:09:06 -040098 struct task_struct *tsk = current;
99 struct vm86plus_struct __user *user;
Brian Gerst9fda6a02015-07-29 01:41:16 -0400100 struct vm86 *vm86 = current->thread.vm86;
Brian Gersted0b2ed2015-07-19 21:09:06 -0400101 long err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700102
103 /*
104 * This gets called from entry.S with interrupts disabled, but
105 * from process context. Enable interrupts here, before trying
106 * to access user space.
107 */
108 local_irq_enable();
109
Brian Gerst13426352015-07-29 01:41:22 -0400110 if (!vm86 || !vm86->user_vm86) {
111 pr_alert("no user_vm86: BAD\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112 do_exit(SIGSEGV);
113 }
Brian Gerstdecd2752015-07-29 01:41:23 -0400114 set_flags(regs->pt.flags, VEFLAGS, X86_EFLAGS_VIF | vm86->veflags_mask);
Brian Gerst13426352015-07-29 01:41:22 -0400115 user = vm86->user_vm86;
Brian Gersted0b2ed2015-07-19 21:09:06 -0400116
Brian Gerstd4ce0f22015-07-29 01:41:17 -0400117 if (!access_ok(VERIFY_WRITE, user, vm86->vm86plus.is_vm86pus ?
Brian Gersted0b2ed2015-07-19 21:09:06 -0400118 sizeof(struct vm86plus_struct) :
119 sizeof(struct vm86_struct))) {
Brian Gerst13426352015-07-29 01:41:22 -0400120 pr_alert("could not access userspace vm86 info\n");
Brian Gersted0b2ed2015-07-19 21:09:06 -0400121 do_exit(SIGSEGV);
122 }
123
124 put_user_try {
125 put_user_ex(regs->pt.bx, &user->regs.ebx);
126 put_user_ex(regs->pt.cx, &user->regs.ecx);
127 put_user_ex(regs->pt.dx, &user->regs.edx);
128 put_user_ex(regs->pt.si, &user->regs.esi);
129 put_user_ex(regs->pt.di, &user->regs.edi);
130 put_user_ex(regs->pt.bp, &user->regs.ebp);
131 put_user_ex(regs->pt.ax, &user->regs.eax);
132 put_user_ex(regs->pt.ip, &user->regs.eip);
133 put_user_ex(regs->pt.cs, &user->regs.cs);
134 put_user_ex(regs->pt.flags, &user->regs.eflags);
135 put_user_ex(regs->pt.sp, &user->regs.esp);
136 put_user_ex(regs->pt.ss, &user->regs.ss);
137 put_user_ex(regs->es, &user->regs.es);
138 put_user_ex(regs->ds, &user->regs.ds);
139 put_user_ex(regs->fs, &user->regs.fs);
140 put_user_ex(regs->gs, &user->regs.gs);
141
Brian Gerst9fda6a02015-07-29 01:41:16 -0400142 put_user_ex(vm86->screen_bitmap, &user->screen_bitmap);
Brian Gersted0b2ed2015-07-19 21:09:06 -0400143 } put_user_catch(err);
144 if (err) {
Brian Gerst13426352015-07-29 01:41:22 -0400145 pr_alert("could not access userspace vm86 info\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146 do_exit(SIGSEGV);
147 }
148
Andy Lutomirskida51da12017-11-02 00:59:10 -0700149 preempt_disable();
Brian Gerst9fda6a02015-07-29 01:41:16 -0400150 tsk->thread.sp0 = vm86->saved_sp0;
Brian Gersted0b2ed2015-07-19 21:09:06 -0400151 tsk->thread.sysenter_cs = __KERNEL_CS;
Andy Lutomirski46f5a102017-11-02 00:59:14 -0700152 update_sp0(tsk);
Andy Lutomirskibd7dc5a2017-11-02 00:59:09 -0700153 refresh_sysenter_cs(&tsk->thread);
Brian Gerst9fda6a02015-07-29 01:41:16 -0400154 vm86->saved_sp0 = 0;
Andy Lutomirskida51da12017-11-02 00:59:10 -0700155 preempt_enable();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156
Brian Gerst5ed92a82015-07-29 01:41:19 -0400157 memcpy(&regs->pt, &vm86->regs32, sizeof(struct pt_regs));
Jeremy Fitzhardinge49d26b62006-12-07 02:14:03 +0100158
Brian Gerst5ed92a82015-07-29 01:41:19 -0400159 lazy_load_gs(vm86->regs32.gs);
Jeremy Fitzhardinge49d26b62006-12-07 02:14:03 +0100160
Brian Gerst5ed92a82015-07-29 01:41:19 -0400161 regs->pt.ax = retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162}
163
Hugh Dickins60ec5582005-10-29 18:16:34 -0700164static void mark_screen_rdonly(struct mm_struct *mm)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700165{
Kirill A. Shutemov3ba5b5e2017-02-13 15:52:28 +0300166 struct vm_area_struct *vma;
167 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700168 pgd_t *pgd;
Kirill A. Shutemove0c4f672017-03-13 17:33:05 +0300169 p4d_t *p4d;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170 pud_t *pud;
171 pmd_t *pmd;
Hugh Dickins60ec5582005-10-29 18:16:34 -0700172 pte_t *pte;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173 int i;
174
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -0700175 down_write(&mm->mmap_sem);
Hugh Dickins60ec5582005-10-29 18:16:34 -0700176 pgd = pgd_offset(mm, 0xA0000);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177 if (pgd_none_or_clear_bad(pgd))
178 goto out;
Kirill A. Shutemove0c4f672017-03-13 17:33:05 +0300179 p4d = p4d_offset(pgd, 0xA0000);
180 if (p4d_none_or_clear_bad(p4d))
181 goto out;
182 pud = pud_offset(p4d, 0xA0000);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183 if (pud_none_or_clear_bad(pud))
184 goto out;
185 pmd = pmd_offset(pud, 0xA0000);
Kirill A. Shutemov78ddc532016-01-15 16:52:42 -0800186
187 if (pmd_trans_huge(*pmd)) {
Kirill A. Shutemov3ba5b5e2017-02-13 15:52:28 +0300188 vma = find_vma(mm, 0xA0000);
Kirill A. Shutemov78ddc532016-01-15 16:52:42 -0800189 split_huge_pmd(vma, pmd, 0xA0000);
190 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191 if (pmd_none_or_clear_bad(pmd))
192 goto out;
Hugh Dickins60ec5582005-10-29 18:16:34 -0700193 pte = pte_offset_map_lock(mm, pmd, 0xA0000, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194 for (i = 0; i < 32; i++) {
195 if (pte_present(*pte))
196 set_pte(pte, pte_wrprotect(*pte));
197 pte++;
198 }
Hugh Dickins60ec5582005-10-29 18:16:34 -0700199 pte_unmap_unlock(pte, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200out:
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -0700201 up_write(&mm->mmap_sem);
Andy Lutomirski9ccee232017-04-22 00:01:19 -0700202 flush_tlb_mm_range(mm, 0xA0000, 0xA0000 + 32*PAGE_SIZE, 0UL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203}
204
205
206
207static int do_vm86_irq_handling(int subfunction, int irqnumber);
Brian Gerst13426352015-07-29 01:41:22 -0400208static long do_sys_vm86(struct vm86plus_struct __user *user_vm86, bool plus);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209
Brian Gerst13426352015-07-29 01:41:22 -0400210SYSCALL_DEFINE1(vm86old, struct vm86_struct __user *, user_vm86)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211{
Brian Gerst13426352015-07-29 01:41:22 -0400212 return do_sys_vm86((struct vm86plus_struct __user *) user_vm86, false);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213}
214
215
Alexander van Heukelum5522ddb2013-03-27 22:18:05 +0100216SYSCALL_DEFINE2(vm86, unsigned long, cmd, unsigned long, arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217{
Brian Gerstf1382f12009-12-09 19:01:55 -0500218 switch (cmd) {
Paolo Ciarrocchi83e714e2008-02-22 23:10:40 +0100219 case VM86_REQUEST_IRQ:
220 case VM86_FREE_IRQ:
221 case VM86_GET_IRQ_BITS:
222 case VM86_GET_AND_RESET_IRQ:
Alexander van Heukelum5522ddb2013-03-27 22:18:05 +0100223 return do_vm86_irq_handling(cmd, (int)arg);
Paolo Ciarrocchi83e714e2008-02-22 23:10:40 +0100224 case VM86_PLUS_INSTALL_CHECK:
225 /*
226 * NOTE: on old vm86 stuff this will return the error
227 * from access_ok(), because the subfunction is
228 * interpreted as (invalid) address to vm86_struct.
229 * So the installation check works.
230 */
Alexander van Heukelum5522ddb2013-03-27 22:18:05 +0100231 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232 }
233
234 /* we come here only for functions VM86_ENTER, VM86_ENTER_NO_BYPASS */
Brian Gerst90c60852015-07-29 01:41:18 -0400235 return do_sys_vm86((struct vm86plus_struct __user *) arg, true);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236}
237
238
Brian Gerst13426352015-07-29 01:41:22 -0400239static long do_sys_vm86(struct vm86plus_struct __user *user_vm86, bool plus)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240{
Brian Gersted0b2ed2015-07-19 21:09:06 -0400241 struct task_struct *tsk = current;
Brian Gerst9fda6a02015-07-29 01:41:16 -0400242 struct vm86 *vm86 = tsk->thread.vm86;
Brian Gerst90c60852015-07-29 01:41:18 -0400243 struct kernel_vm86_regs vm86regs;
Brian Gerst5ed92a82015-07-29 01:41:19 -0400244 struct pt_regs *regs = current_pt_regs();
Brian Gersted0b2ed2015-07-19 21:09:06 -0400245 unsigned long err = 0;
246
Andy Lutomirski76fc5e72015-09-04 17:00:43 -0700247 err = security_mmap_addr(0);
248 if (err) {
249 /*
250 * vm86 cannot virtualize the address space, so vm86 users
251 * need to manage the low 1MB themselves using mmap. Given
252 * that BIOS places important data in the first page, vm86
253 * is essentially useless if mmap_min_addr != 0. DOSEMU,
254 * for example, won't even bother trying to use vm86 if it
255 * can't map a page at virtual address 0.
256 *
257 * To reduce the available kernel attack surface, simply
258 * disallow vm86(old) for users who cannot mmap at va 0.
259 *
260 * The implementation of security_mmap_addr will allow
261 * suitably privileged users to map va 0 even if
262 * vm.mmap_min_addr is set above 0, and we want this
263 * behavior for vm86 as well, as it ensures that legacy
264 * tools like vbetool will not fail just because of
265 * vm.mmap_min_addr.
266 */
267 pr_info_once("Denied a call to vm86(old) from %s[%d] (uid: %d). Set the vm.mmap_min_addr sysctl to 0 and/or adjust LSM mmap_min_addr policy to enable vm86 if you are using a vm86-based DOS emulator.\n",
268 current->comm, task_pid_nr(current),
269 from_kuid_munged(&init_user_ns, current_uid()));
270 return -EPERM;
271 }
272
Brian Gerst9fda6a02015-07-29 01:41:16 -0400273 if (!vm86) {
274 if (!(vm86 = kzalloc(sizeof(*vm86), GFP_KERNEL)))
275 return -ENOMEM;
276 tsk->thread.vm86 = vm86;
277 }
278 if (vm86->saved_sp0)
Brian Gersted0b2ed2015-07-19 21:09:06 -0400279 return -EPERM;
280
Brian Gerst13426352015-07-29 01:41:22 -0400281 if (!access_ok(VERIFY_READ, user_vm86, plus ?
Brian Gersted0b2ed2015-07-19 21:09:06 -0400282 sizeof(struct vm86_struct) :
283 sizeof(struct vm86plus_struct)))
284 return -EFAULT;
285
Brian Gerst90c60852015-07-29 01:41:18 -0400286 memset(&vm86regs, 0, sizeof(vm86regs));
Brian Gersted0b2ed2015-07-19 21:09:06 -0400287 get_user_try {
288 unsigned short seg;
Brian Gerst13426352015-07-29 01:41:22 -0400289 get_user_ex(vm86regs.pt.bx, &user_vm86->regs.ebx);
290 get_user_ex(vm86regs.pt.cx, &user_vm86->regs.ecx);
291 get_user_ex(vm86regs.pt.dx, &user_vm86->regs.edx);
292 get_user_ex(vm86regs.pt.si, &user_vm86->regs.esi);
293 get_user_ex(vm86regs.pt.di, &user_vm86->regs.edi);
294 get_user_ex(vm86regs.pt.bp, &user_vm86->regs.ebp);
295 get_user_ex(vm86regs.pt.ax, &user_vm86->regs.eax);
296 get_user_ex(vm86regs.pt.ip, &user_vm86->regs.eip);
297 get_user_ex(seg, &user_vm86->regs.cs);
Brian Gerst90c60852015-07-29 01:41:18 -0400298 vm86regs.pt.cs = seg;
Brian Gerst13426352015-07-29 01:41:22 -0400299 get_user_ex(vm86regs.pt.flags, &user_vm86->regs.eflags);
300 get_user_ex(vm86regs.pt.sp, &user_vm86->regs.esp);
301 get_user_ex(seg, &user_vm86->regs.ss);
Brian Gerst90c60852015-07-29 01:41:18 -0400302 vm86regs.pt.ss = seg;
Brian Gerst13426352015-07-29 01:41:22 -0400303 get_user_ex(vm86regs.es, &user_vm86->regs.es);
304 get_user_ex(vm86regs.ds, &user_vm86->regs.ds);
305 get_user_ex(vm86regs.fs, &user_vm86->regs.fs);
306 get_user_ex(vm86regs.gs, &user_vm86->regs.gs);
Brian Gersted0b2ed2015-07-19 21:09:06 -0400307
Brian Gerst13426352015-07-29 01:41:22 -0400308 get_user_ex(vm86->flags, &user_vm86->flags);
309 get_user_ex(vm86->screen_bitmap, &user_vm86->screen_bitmap);
310 get_user_ex(vm86->cpu_type, &user_vm86->cpu_type);
Brian Gersted0b2ed2015-07-19 21:09:06 -0400311 } get_user_catch(err);
312 if (err)
313 return err;
314
Brian Gerst13426352015-07-29 01:41:22 -0400315 if (copy_from_user(&vm86->int_revectored,
316 &user_vm86->int_revectored,
Brian Gersted0b2ed2015-07-19 21:09:06 -0400317 sizeof(struct revectored_struct)))
318 return -EFAULT;
Brian Gerst13426352015-07-29 01:41:22 -0400319 if (copy_from_user(&vm86->int21_revectored,
320 &user_vm86->int21_revectored,
Brian Gersted0b2ed2015-07-19 21:09:06 -0400321 sizeof(struct revectored_struct)))
322 return -EFAULT;
323 if (plus) {
Brian Gerst13426352015-07-29 01:41:22 -0400324 if (copy_from_user(&vm86->vm86plus, &user_vm86->vm86plus,
Brian Gersted0b2ed2015-07-19 21:09:06 -0400325 sizeof(struct vm86plus_info_struct)))
326 return -EFAULT;
Brian Gerstd4ce0f22015-07-29 01:41:17 -0400327 vm86->vm86plus.is_vm86pus = 1;
328 } else
329 memset(&vm86->vm86plus, 0,
330 sizeof(struct vm86plus_info_struct));
Brian Gerst5ed92a82015-07-29 01:41:19 -0400331
332 memcpy(&vm86->regs32, regs, sizeof(struct pt_regs));
Brian Gerst13426352015-07-29 01:41:22 -0400333 vm86->user_vm86 = user_vm86;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334
335/*
H. Peter Anvin65ea5b02008-01-30 13:30:56 +0100336 * The flags register is also special: we cannot trust that the user
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337 * has set it up safely, so this makes sure interrupt etc flags are
338 * inherited from protected mode.
339 */
Brian Gerst90c60852015-07-29 01:41:18 -0400340 VEFLAGS = vm86regs.pt.flags;
341 vm86regs.pt.flags &= SAFE_MASK;
Brian Gerst5ed92a82015-07-29 01:41:19 -0400342 vm86regs.pt.flags |= regs->flags & ~SAFE_MASK;
Brian Gerst90c60852015-07-29 01:41:18 -0400343 vm86regs.pt.flags |= X86_VM_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700344
Brian Gerst5ed92a82015-07-29 01:41:19 -0400345 vm86regs.pt.orig_ax = regs->orig_ax;
Brian Gerstdf1ae9a2015-07-19 21:09:05 -0400346
Brian Gerstd4ce0f22015-07-29 01:41:17 -0400347 switch (vm86->cpu_type) {
Paolo Ciarrocchi83e714e2008-02-22 23:10:40 +0100348 case CPU_286:
Brian Gerstdecd2752015-07-29 01:41:23 -0400349 vm86->veflags_mask = 0;
Paolo Ciarrocchi83e714e2008-02-22 23:10:40 +0100350 break;
351 case CPU_386:
Brian Gerstdecd2752015-07-29 01:41:23 -0400352 vm86->veflags_mask = X86_EFLAGS_NT | X86_EFLAGS_IOPL;
Paolo Ciarrocchi83e714e2008-02-22 23:10:40 +0100353 break;
354 case CPU_486:
Brian Gerstdecd2752015-07-29 01:41:23 -0400355 vm86->veflags_mask = X86_EFLAGS_AC | X86_EFLAGS_NT | X86_EFLAGS_IOPL;
Paolo Ciarrocchi83e714e2008-02-22 23:10:40 +0100356 break;
357 default:
Brian Gerstdecd2752015-07-29 01:41:23 -0400358 vm86->veflags_mask = X86_EFLAGS_ID | X86_EFLAGS_AC | X86_EFLAGS_NT | X86_EFLAGS_IOPL;
Paolo Ciarrocchi83e714e2008-02-22 23:10:40 +0100359 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700360 }
361
362/*
Brian Gerst5ed92a82015-07-29 01:41:19 -0400363 * Save old state
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364 */
Brian Gerst9fda6a02015-07-29 01:41:16 -0400365 vm86->saved_sp0 = tsk->thread.sp0;
Brian Gerst5ed92a82015-07-29 01:41:19 -0400366 lazy_save_gs(vm86->regs32.gs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367
Brian Gerst5ed92a82015-07-29 01:41:19 -0400368 /* make room for real-mode segments */
Andy Lutomirskida51da12017-11-02 00:59:10 -0700369 preempt_disable();
Brian Gerst5ed92a82015-07-29 01:41:19 -0400370 tsk->thread.sp0 += 16;
Borislav Petkov362f9242015-12-07 10:39:41 +0100371
Andy Lutomirskibd7dc5a2017-11-02 00:59:09 -0700372 if (static_cpu_has(X86_FEATURE_SEP)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373 tsk->thread.sysenter_cs = 0;
Andy Lutomirskibd7dc5a2017-11-02 00:59:09 -0700374 refresh_sysenter_cs(&tsk->thread);
375 }
Borislav Petkov362f9242015-12-07 10:39:41 +0100376
Andy Lutomirski46f5a102017-11-02 00:59:14 -0700377 update_sp0(tsk);
Andy Lutomirskida51da12017-11-02 00:59:10 -0700378 preempt_enable();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379
Brian Gerstd4ce0f22015-07-29 01:41:17 -0400380 if (vm86->flags & VM86_SCREEN_BITMAP)
Hugh Dickins60ec5582005-10-29 18:16:34 -0700381 mark_screen_rdonly(tsk->mm);
Jason Baron7e7f8a02006-01-31 16:56:28 -0500382
Brian Gerst5ed92a82015-07-29 01:41:19 -0400383 memcpy((struct kernel_vm86_regs *)regs, &vm86regs, sizeof(vm86regs));
384 force_iret();
385 return regs->ax;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386}
387
Paolo Ciarrocchi83e714e2008-02-22 23:10:40 +0100388static inline void set_IF(struct kernel_vm86_regs *regs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389{
gorcunov@gmail.coma5c15d42008-03-28 17:56:56 +0300390 VEFLAGS |= X86_EFLAGS_VIF;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391}
392
Paolo Ciarrocchi83e714e2008-02-22 23:10:40 +0100393static inline void clear_IF(struct kernel_vm86_regs *regs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394{
gorcunov@gmail.coma5c15d42008-03-28 17:56:56 +0300395 VEFLAGS &= ~X86_EFLAGS_VIF;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396}
397
Paolo Ciarrocchi83e714e2008-02-22 23:10:40 +0100398static inline void clear_TF(struct kernel_vm86_regs *regs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399{
gorcunov@gmail.coma5c15d42008-03-28 17:56:56 +0300400 regs->pt.flags &= ~X86_EFLAGS_TF;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700401}
402
Paolo Ciarrocchi83e714e2008-02-22 23:10:40 +0100403static inline void clear_AC(struct kernel_vm86_regs *regs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404{
gorcunov@gmail.coma5c15d42008-03-28 17:56:56 +0300405 regs->pt.flags &= ~X86_EFLAGS_AC;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406}
407
Paolo Ciarrocchi83e714e2008-02-22 23:10:40 +0100408/*
409 * It is correct to call set_IF(regs) from the set_vflags_*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410 * functions. However someone forgot to call clear_IF(regs)
411 * in the opposite case.
412 * After the command sequence CLI PUSHF STI POPF you should
Joe Perchesab4a5742008-01-30 13:31:42 +0100413 * end up with interrupts disabled, but you ended up with
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414 * interrupts enabled.
415 * ( I was testing my own changes, but the only bug I
416 * could find was in a function I had not changed. )
417 * [KD]
418 */
419
Paolo Ciarrocchi83e714e2008-02-22 23:10:40 +0100420static inline void set_vflags_long(unsigned long flags, struct kernel_vm86_regs *regs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421{
Brian Gerstdecd2752015-07-29 01:41:23 -0400422 set_flags(VEFLAGS, flags, current->thread.vm86->veflags_mask);
H. Peter Anvin65ea5b02008-01-30 13:30:56 +0100423 set_flags(regs->pt.flags, flags, SAFE_MASK);
gorcunov@gmail.coma5c15d42008-03-28 17:56:56 +0300424 if (flags & X86_EFLAGS_IF)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425 set_IF(regs);
426 else
427 clear_IF(regs);
428}
429
Paolo Ciarrocchi83e714e2008-02-22 23:10:40 +0100430static inline void set_vflags_short(unsigned short flags, struct kernel_vm86_regs *regs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431{
Brian Gerstdecd2752015-07-29 01:41:23 -0400432 set_flags(VFLAGS, flags, current->thread.vm86->veflags_mask);
H. Peter Anvin65ea5b02008-01-30 13:30:56 +0100433 set_flags(regs->pt.flags, flags, SAFE_MASK);
gorcunov@gmail.coma5c15d42008-03-28 17:56:56 +0300434 if (flags & X86_EFLAGS_IF)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435 set_IF(regs);
436 else
437 clear_IF(regs);
438}
439
Paolo Ciarrocchi83e714e2008-02-22 23:10:40 +0100440static inline unsigned long get_vflags(struct kernel_vm86_regs *regs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441{
H. Peter Anvin65ea5b02008-01-30 13:30:56 +0100442 unsigned long flags = regs->pt.flags & RETURN_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443
gorcunov@gmail.coma5c15d42008-03-28 17:56:56 +0300444 if (VEFLAGS & X86_EFLAGS_VIF)
445 flags |= X86_EFLAGS_IF;
446 flags |= X86_EFLAGS_IOPL;
Brian Gerstdecd2752015-07-29 01:41:23 -0400447 return flags | (VEFLAGS & current->thread.vm86->veflags_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448}
449
Paolo Ciarrocchi83e714e2008-02-22 23:10:40 +0100450static inline int is_revectored(int nr, struct revectored_struct *bitmap)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451{
H. Peter Anvin2823d4d2016-06-08 12:38:37 -0700452 return test_bit(nr, bitmap->__map);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453}
454
455#define val_byte(val, n) (((__u8 *)&val)[n])
456
457#define pushb(base, ptr, val, err_label) \
458 do { \
459 __u8 __val = val; \
460 ptr--; \
461 if (put_user(__val, base + ptr) < 0) \
462 goto err_label; \
Paolo Ciarrocchi83e714e2008-02-22 23:10:40 +0100463 } while (0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464
465#define pushw(base, ptr, val, err_label) \
466 do { \
467 __u16 __val = val; \
468 ptr--; \
469 if (put_user(val_byte(__val, 1), base + ptr) < 0) \
470 goto err_label; \
471 ptr--; \
472 if (put_user(val_byte(__val, 0), base + ptr) < 0) \
473 goto err_label; \
Paolo Ciarrocchi83e714e2008-02-22 23:10:40 +0100474 } while (0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475
476#define pushl(base, ptr, val, err_label) \
477 do { \
478 __u32 __val = val; \
479 ptr--; \
480 if (put_user(val_byte(__val, 3), base + ptr) < 0) \
481 goto err_label; \
482 ptr--; \
483 if (put_user(val_byte(__val, 2), base + ptr) < 0) \
484 goto err_label; \
485 ptr--; \
486 if (put_user(val_byte(__val, 1), base + ptr) < 0) \
487 goto err_label; \
488 ptr--; \
489 if (put_user(val_byte(__val, 0), base + ptr) < 0) \
490 goto err_label; \
Paolo Ciarrocchi83e714e2008-02-22 23:10:40 +0100491 } while (0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492
493#define popb(base, ptr, err_label) \
494 ({ \
495 __u8 __res; \
496 if (get_user(__res, base + ptr) < 0) \
497 goto err_label; \
498 ptr++; \
499 __res; \
500 })
501
502#define popw(base, ptr, err_label) \
503 ({ \
504 __u16 __res; \
505 if (get_user(val_byte(__res, 0), base + ptr) < 0) \
506 goto err_label; \
507 ptr++; \
508 if (get_user(val_byte(__res, 1), base + ptr) < 0) \
509 goto err_label; \
510 ptr++; \
511 __res; \
512 })
513
514#define popl(base, ptr, err_label) \
515 ({ \
516 __u32 __res; \
517 if (get_user(val_byte(__res, 0), base + ptr) < 0) \
518 goto err_label; \
519 ptr++; \
520 if (get_user(val_byte(__res, 1), base + ptr) < 0) \
521 goto err_label; \
522 ptr++; \
523 if (get_user(val_byte(__res, 2), base + ptr) < 0) \
524 goto err_label; \
525 ptr++; \
526 if (get_user(val_byte(__res, 3), base + ptr) < 0) \
527 goto err_label; \
528 ptr++; \
529 __res; \
530 })
531
532/* There are so many possible reasons for this function to return
533 * VM86_INTx, so adding another doesn't bother me. We can expect
534 * userspace programs to be able to handle it. (Getting a problem
535 * in userspace is always better than an Oops anyway.) [KD]
536 */
537static void do_int(struct kernel_vm86_regs *regs, int i,
Paolo Ciarrocchi83e714e2008-02-22 23:10:40 +0100538 unsigned char __user *ssp, unsigned short sp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539{
540 unsigned long __user *intr_ptr;
541 unsigned long segoffs;
Brian Gerst90c60852015-07-29 01:41:18 -0400542 struct vm86 *vm86 = current->thread.vm86;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543
H. Peter Anvin65ea5b02008-01-30 13:30:56 +0100544 if (regs->pt.cs == BIOSSEG)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700545 goto cannot_handle;
Brian Gerstd4ce0f22015-07-29 01:41:17 -0400546 if (is_revectored(i, &vm86->int_revectored))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547 goto cannot_handle;
Brian Gerstd4ce0f22015-07-29 01:41:17 -0400548 if (i == 0x21 && is_revectored(AH(regs), &vm86->int21_revectored))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549 goto cannot_handle;
550 intr_ptr = (unsigned long __user *) (i << 2);
551 if (get_user(segoffs, intr_ptr))
552 goto cannot_handle;
553 if ((segoffs >> 16) == BIOSSEG)
554 goto cannot_handle;
555 pushw(ssp, sp, get_vflags(regs), cannot_handle);
H. Peter Anvin65ea5b02008-01-30 13:30:56 +0100556 pushw(ssp, sp, regs->pt.cs, cannot_handle);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557 pushw(ssp, sp, IP(regs), cannot_handle);
H. Peter Anvin65ea5b02008-01-30 13:30:56 +0100558 regs->pt.cs = segoffs >> 16;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559 SP(regs) -= 6;
560 IP(regs) = segoffs & 0xffff;
561 clear_TF(regs);
562 clear_IF(regs);
563 clear_AC(regs);
564 return;
565
566cannot_handle:
Brian Gerst5ed92a82015-07-29 01:41:19 -0400567 save_v86_state(regs, VM86_INTx + (i << 8));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568}
569
Paolo Ciarrocchi83e714e2008-02-22 23:10:40 +0100570int handle_vm86_trap(struct kernel_vm86_regs *regs, long error_code, int trapno)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571{
Brian Gerst90c60852015-07-29 01:41:18 -0400572 struct vm86 *vm86 = current->thread.vm86;
573
574 if (vm86->vm86plus.is_vm86pus) {
Bart Oldeman65542872010-09-23 13:16:58 -0400575 if ((trapno == 3) || (trapno == 1)) {
Brian Gerst5ed92a82015-07-29 01:41:19 -0400576 save_v86_state(regs, VM86_TRAP + (trapno << 8));
Bart Oldeman65542872010-09-23 13:16:58 -0400577 return 0;
578 }
H. Peter Anvin65ea5b02008-01-30 13:30:56 +0100579 do_int(regs, trapno, (unsigned char __user *) (regs->pt.ss << 4), SP(regs));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580 return 0;
581 }
Paolo Ciarrocchi83e714e2008-02-22 23:10:40 +0100582 if (trapno != 1)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583 return 1; /* we let this handle by the calling routine */
Srikar Dronamraju51e7dc72012-03-12 14:55:55 +0530584 current->thread.trap_nr = trapno;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585 current->thread.error_code = error_code;
Roland McGrath0f540912008-03-17 02:21:08 -0700586 force_sig(SIGTRAP, current);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587 return 0;
588}
589
Paolo Ciarrocchi83e714e2008-02-22 23:10:40 +0100590void handle_vm86_fault(struct kernel_vm86_regs *regs, long error_code)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591{
592 unsigned char opcode;
593 unsigned char __user *csp;
594 unsigned char __user *ssp;
Petr Tesarik5fd75eb2005-09-03 15:56:28 -0700595 unsigned short ip, sp, orig_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596 int data32, pref_done;
Brian Gerstd4ce0f22015-07-29 01:41:17 -0400597 struct vm86plus_info_struct *vmpi = &current->thread.vm86->vm86plus;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598
599#define CHECK_IF_IN_TRAP \
Brian Gerstd4ce0f22015-07-29 01:41:17 -0400600 if (vmpi->vm86dbg_active && vmpi->vm86dbg_TFpendig) \
gorcunov@gmail.coma5c15d42008-03-28 17:56:56 +0300601 newflags |= X86_EFLAGS_TF
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602
H. Peter Anvin65ea5b02008-01-30 13:30:56 +0100603 orig_flags = *(unsigned short *)&regs->pt.flags;
Petr Tesarik5fd75eb2005-09-03 15:56:28 -0700604
H. Peter Anvin65ea5b02008-01-30 13:30:56 +0100605 csp = (unsigned char __user *) (regs->pt.cs << 4);
606 ssp = (unsigned char __user *) (regs->pt.ss << 4);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607 sp = SP(regs);
608 ip = IP(regs);
609
610 data32 = 0;
611 pref_done = 0;
612 do {
613 switch (opcode = popb(csp, ip, simulate_sigsegv)) {
Paolo Ciarrocchi83e714e2008-02-22 23:10:40 +0100614 case 0x66: /* 32-bit data */ data32 = 1; break;
615 case 0x67: /* 32-bit address */ break;
616 case 0x2e: /* CS */ break;
617 case 0x3e: /* DS */ break;
618 case 0x26: /* ES */ break;
619 case 0x36: /* SS */ break;
620 case 0x65: /* GS */ break;
621 case 0x64: /* FS */ break;
622 case 0xf2: /* repnz */ break;
623 case 0xf3: /* rep */ break;
624 default: pref_done = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625 }
626 } while (!pref_done);
627
628 switch (opcode) {
629
630 /* pushf */
631 case 0x9c:
632 if (data32) {
633 pushl(ssp, sp, get_vflags(regs), simulate_sigsegv);
634 SP(regs) -= 4;
635 } else {
636 pushw(ssp, sp, get_vflags(regs), simulate_sigsegv);
637 SP(regs) -= 2;
638 }
639 IP(regs) = ip;
Brian Gerst5ed92a82015-07-29 01:41:19 -0400640 goto vm86_fault_return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641
642 /* popf */
643 case 0x9d:
644 {
645 unsigned long newflags;
646 if (data32) {
Paolo Ciarrocchi83e714e2008-02-22 23:10:40 +0100647 newflags = popl(ssp, sp, simulate_sigsegv);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648 SP(regs) += 4;
649 } else {
650 newflags = popw(ssp, sp, simulate_sigsegv);
651 SP(regs) += 2;
652 }
653 IP(regs) = ip;
654 CHECK_IF_IN_TRAP;
Paolo Ciarrocchi83e714e2008-02-22 23:10:40 +0100655 if (data32)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656 set_vflags_long(newflags, regs);
Paolo Ciarrocchi83e714e2008-02-22 23:10:40 +0100657 else
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658 set_vflags_short(newflags, regs);
Paolo Ciarrocchi83e714e2008-02-22 23:10:40 +0100659
Brian Gerst5ed92a82015-07-29 01:41:19 -0400660 goto check_vip;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661 }
662
663 /* int xx */
664 case 0xcd: {
Paolo Ciarrocchi83e714e2008-02-22 23:10:40 +0100665 int intno = popb(csp, ip, simulate_sigsegv);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666 IP(regs) = ip;
Brian Gerstd4ce0f22015-07-29 01:41:17 -0400667 if (vmpi->vm86dbg_active) {
Brian Gerst5ed92a82015-07-29 01:41:19 -0400668 if ((1 << (intno & 7)) & vmpi->vm86dbg_intxxtab[intno >> 3]) {
669 save_v86_state(regs, VM86_INTx + (intno << 8));
670 return;
671 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672 }
673 do_int(regs, intno, ssp, sp);
674 return;
675 }
676
677 /* iret */
678 case 0xcf:
679 {
680 unsigned long newip;
681 unsigned long newcs;
682 unsigned long newflags;
683 if (data32) {
Paolo Ciarrocchi83e714e2008-02-22 23:10:40 +0100684 newip = popl(ssp, sp, simulate_sigsegv);
685 newcs = popl(ssp, sp, simulate_sigsegv);
686 newflags = popl(ssp, sp, simulate_sigsegv);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687 SP(regs) += 12;
688 } else {
689 newip = popw(ssp, sp, simulate_sigsegv);
690 newcs = popw(ssp, sp, simulate_sigsegv);
691 newflags = popw(ssp, sp, simulate_sigsegv);
692 SP(regs) += 6;
693 }
694 IP(regs) = newip;
H. Peter Anvin65ea5b02008-01-30 13:30:56 +0100695 regs->pt.cs = newcs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696 CHECK_IF_IN_TRAP;
697 if (data32) {
698 set_vflags_long(newflags, regs);
699 } else {
700 set_vflags_short(newflags, regs);
701 }
Brian Gerst5ed92a82015-07-29 01:41:19 -0400702 goto check_vip;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700703 }
704
705 /* cli */
706 case 0xfa:
707 IP(regs) = ip;
708 clear_IF(regs);
Brian Gerst5ed92a82015-07-29 01:41:19 -0400709 goto vm86_fault_return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710
711 /* sti */
712 /*
713 * Damn. This is incorrect: the 'sti' instruction should actually
714 * enable interrupts after the /next/ instruction. Not good.
715 *
716 * Probably needs some horsing around with the TF flag. Aiee..
717 */
718 case 0xfb:
719 IP(regs) = ip;
720 set_IF(regs);
Brian Gerst5ed92a82015-07-29 01:41:19 -0400721 goto check_vip;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722
723 default:
Brian Gerst5ed92a82015-07-29 01:41:19 -0400724 save_v86_state(regs, VM86_UNKNOWN);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725 }
726
727 return;
728
Brian Gerst5ed92a82015-07-29 01:41:19 -0400729check_vip:
730 if (VEFLAGS & X86_EFLAGS_VIP) {
731 save_v86_state(regs, VM86_STI);
732 return;
733 }
734
735vm86_fault_return:
736 if (vmpi->force_return_for_pic && (VEFLAGS & (X86_EFLAGS_IF | X86_EFLAGS_VIF))) {
737 save_v86_state(regs, VM86_PICRETURN);
738 return;
739 }
740 if (orig_flags & X86_EFLAGS_TF)
741 handle_vm86_trap(regs, 0, X86_TRAP_DB);
742 return;
743
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744simulate_sigsegv:
745 /* FIXME: After a long discussion with Stas we finally
746 * agreed, that this is wrong. Here we should
747 * really send a SIGSEGV to the user program.
748 * But how do we create the correct context? We
749 * are inside a general protection fault handler
750 * and has just returned from a page fault handler.
751 * The correct context for the signal handler
752 * should be a mixture of the two, but how do we
753 * get the information? [KD]
754 */
Brian Gerst5ed92a82015-07-29 01:41:19 -0400755 save_v86_state(regs, VM86_UNKNOWN);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756}
757
758/* ---------------- vm86 special IRQ passing stuff ----------------- */
759
760#define VM86_IRQNAME "vm86irq"
761
762static struct vm86_irqs {
763 struct task_struct *tsk;
764 int sig;
765} vm86_irqs[16];
766
767static DEFINE_SPINLOCK(irqbits_lock);
768static int irqbits;
769
Paolo Ciarrocchi83e714e2008-02-22 23:10:40 +0100770#define ALLOWED_SIGS (1 /* 0 = don't send a signal */ \
Linus Torvalds1da177e2005-04-16 15:20:36 -0700771 | (1 << SIGUSR1) | (1 << SIGUSR2) | (1 << SIGIO) | (1 << SIGURG) \
Paolo Ciarrocchi83e714e2008-02-22 23:10:40 +0100772 | (1 << SIGUNUSED))
773
David Howells7d12e782006-10-05 14:55:46 +0100774static irqreturn_t irq_handler(int intno, void *dev_id)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775{
776 int irq_bit;
777 unsigned long flags;
778
Paolo Ciarrocchi83e714e2008-02-22 23:10:40 +0100779 spin_lock_irqsave(&irqbits_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780 irq_bit = 1 << intno;
Paolo Ciarrocchi83e714e2008-02-22 23:10:40 +0100781 if ((irqbits & irq_bit) || !vm86_irqs[intno].tsk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782 goto out;
783 irqbits |= irq_bit;
784 if (vm86_irqs[intno].sig)
785 send_sig(vm86_irqs[intno].sig, vm86_irqs[intno].tsk, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786 /*
787 * IRQ will be re-enabled when user asks for the irq (whether
788 * polling or as a result of the signal)
789 */
Pavel Pisaad671422005-05-01 08:58:52 -0700790 disable_irq_nosync(intno);
791 spin_unlock_irqrestore(&irqbits_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792 return IRQ_HANDLED;
793
794out:
Paolo Ciarrocchi83e714e2008-02-22 23:10:40 +0100795 spin_unlock_irqrestore(&irqbits_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796 return IRQ_NONE;
797}
798
799static inline void free_vm86_irq(int irqnumber)
800{
801 unsigned long flags;
802
803 free_irq(irqnumber, NULL);
804 vm86_irqs[irqnumber].tsk = NULL;
805
Paolo Ciarrocchi83e714e2008-02-22 23:10:40 +0100806 spin_lock_irqsave(&irqbits_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807 irqbits &= ~(1 << irqnumber);
Paolo Ciarrocchi83e714e2008-02-22 23:10:40 +0100808 spin_unlock_irqrestore(&irqbits_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700809}
810
811void release_vm86_irqs(struct task_struct *task)
812{
813 int i;
814 for (i = FIRST_VM86_IRQ ; i <= LAST_VM86_IRQ; i++)
815 if (vm86_irqs[i].tsk == task)
816 free_vm86_irq(i);
817}
818
819static inline int get_and_reset_irq(int irqnumber)
820{
821 int bit;
822 unsigned long flags;
Pavel Pisaad671422005-05-01 08:58:52 -0700823 int ret = 0;
Paolo Ciarrocchi83e714e2008-02-22 23:10:40 +0100824
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825 if (invalid_vm86_irq(irqnumber)) return 0;
826 if (vm86_irqs[irqnumber].tsk != current) return 0;
Paolo Ciarrocchi83e714e2008-02-22 23:10:40 +0100827 spin_lock_irqsave(&irqbits_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828 bit = irqbits & (1 << irqnumber);
829 irqbits &= ~bit;
Pavel Pisaad671422005-05-01 08:58:52 -0700830 if (bit) {
831 enable_irq(irqnumber);
832 ret = 1;
833 }
834
Paolo Ciarrocchi83e714e2008-02-22 23:10:40 +0100835 spin_unlock_irqrestore(&irqbits_lock, flags);
Pavel Pisaad671422005-05-01 08:58:52 -0700836 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837}
838
839
840static int do_vm86_irq_handling(int subfunction, int irqnumber)
841{
842 int ret;
843 switch (subfunction) {
844 case VM86_GET_AND_RESET_IRQ: {
845 return get_and_reset_irq(irqnumber);
846 }
847 case VM86_GET_IRQ_BITS: {
848 return irqbits;
849 }
850 case VM86_REQUEST_IRQ: {
851 int sig = irqnumber >> 8;
852 int irq = irqnumber & 255;
853 if (!capable(CAP_SYS_ADMIN)) return -EPERM;
854 if (!((1 << sig) & ALLOWED_SIGS)) return -EPERM;
855 if (invalid_vm86_irq(irq)) return -EPERM;
856 if (vm86_irqs[irq].tsk) return -EPERM;
857 ret = request_irq(irq, &irq_handler, 0, VM86_IRQNAME, NULL);
858 if (ret) return ret;
859 vm86_irqs[irq].sig = sig;
860 vm86_irqs[irq].tsk = current;
861 return irq;
862 }
863 case VM86_FREE_IRQ: {
864 if (invalid_vm86_irq(irqnumber)) return -EPERM;
865 if (!vm86_irqs[irqnumber].tsk) return 0;
866 if (vm86_irqs[irqnumber].tsk != current) return -EPERM;
867 free_vm86_irq(irqnumber);
868 return 0;
869 }
870 }
871 return -EINVAL;
872}
873