1.1 --- /dev/null	Thu Jan 01 00:00:00 1970 +0000
     1.2 +++ b/linux-2.6-xen-sparse/arch/i386/kernel/asm-offsets.c	Mon Feb 20 14:37:13 2006 +0000
     1.3 @@ -0,0 +1,72 @@
     1.4 +/*
     1.5 + * Generate definitions needed by assembly language modules.
     1.6 + * This code generates raw asm output which is post-processed
     1.7 + * to extract and format the required data.
     1.8 + */
     1.9 +
    1.10 +#include <linux/sched.h>
    1.11 +#include <linux/signal.h>
    1.12 +#include <linux/personality.h>
    1.13 +#include <linux/suspend.h>
    1.14 +#include <asm/ucontext.h>
    1.15 +#include "sigframe.h"
    1.16 +#include <asm/fixmap.h>
    1.17 +#include <asm/processor.h>
    1.18 +#include <asm/thread_info.h>
    1.19 +
    1.20 +#define DEFINE(sym, val) \
    1.21 +        asm volatile("\n->" #sym " %0 " #val : : "i" (val))
    1.22 +
    1.23 +#define BLANK() asm volatile("\n->" : : )
    1.24 +
    1.25 +#define OFFSET(sym, str, mem) \
    1.26 +	DEFINE(sym, offsetof(struct str, mem));
    1.27 +
    1.28 +void foo(void)
    1.29 +{
    1.30 +	OFFSET(SIGCONTEXT_eax, sigcontext, eax);
    1.31 +	OFFSET(SIGCONTEXT_ebx, sigcontext, ebx);
    1.32 +	OFFSET(SIGCONTEXT_ecx, sigcontext, ecx);
    1.33 +	OFFSET(SIGCONTEXT_edx, sigcontext, edx);
    1.34 +	OFFSET(SIGCONTEXT_esi, sigcontext, esi);
    1.35 +	OFFSET(SIGCONTEXT_edi, sigcontext, edi);
    1.36 +	OFFSET(SIGCONTEXT_ebp, sigcontext, ebp);
    1.37 +	OFFSET(SIGCONTEXT_esp, sigcontext, esp);
    1.38 +	OFFSET(SIGCONTEXT_eip, sigcontext, eip);
    1.39 +	BLANK();
    1.40 +
    1.41 +	OFFSET(CPUINFO_x86, cpuinfo_x86, x86);
    1.42 +	OFFSET(CPUINFO_x86_vendor, cpuinfo_x86, x86_vendor);
    1.43 +	OFFSET(CPUINFO_x86_model, cpuinfo_x86, x86_model);
    1.44 +	OFFSET(CPUINFO_x86_mask, cpuinfo_x86, x86_mask);
    1.45 +	OFFSET(CPUINFO_hard_math, cpuinfo_x86, hard_math);
    1.46 +	OFFSET(CPUINFO_cpuid_level, cpuinfo_x86, cpuid_level);
    1.47 +	OFFSET(CPUINFO_x86_capability, cpuinfo_x86, x86_capability);
    1.48 +	OFFSET(CPUINFO_x86_vendor_id, cpuinfo_x86, x86_vendor_id);
    1.49 +	BLANK();
    1.50 +
    1.51 +	OFFSET(TI_task, thread_info, task);
    1.52 +	OFFSET(TI_exec_domain, thread_info, exec_domain);
    1.53 +	OFFSET(TI_flags, thread_info, flags);
    1.54 +	OFFSET(TI_status, thread_info, status);
    1.55 +	OFFSET(TI_cpu, thread_info, cpu);
    1.56 +	OFFSET(TI_preempt_count, thread_info, preempt_count);
    1.57 +	OFFSET(TI_addr_limit, thread_info, addr_limit);
    1.58 +	OFFSET(TI_restart_block, thread_info, restart_block);
    1.59 +	BLANK();
    1.60 +
    1.61 +	OFFSET(EXEC_DOMAIN_handler, exec_domain, handler);
    1.62 +	OFFSET(RT_SIGFRAME_sigcontext, rt_sigframe, uc.uc_mcontext);
    1.63 +	BLANK();
    1.64 +
    1.65 +	OFFSET(pbe_address, pbe, address);
    1.66 +	OFFSET(pbe_orig_address, pbe, orig_address);
    1.67 +	OFFSET(pbe_next, pbe, next);
    1.68 +
    1.69 +	/* Offset from the sysenter stack to tss.esp0 */
    1.70 +	DEFINE(TSS_sysenter_esp0, offsetof(struct tss_struct, esp0) -
    1.71 +		 sizeof(struct tss_struct));
    1.72 +
    1.73 +	DEFINE(PAGE_SIZE_asm, PAGE_SIZE);
    1.74 +	DEFINE(VSYSCALL_BASE, __fix_to_virt(FIX_VSYSCALL));
    1.75 +}

     2.1 --- /dev/null	Thu Jan 01 00:00:00 1970 +0000
     2.2 +++ b/linux-2.6-xen-sparse/arch/i386/kernel/sysenter.c	Mon Feb 20 14:37:13 2006 +0000
     2.3 @@ -0,0 +1,67 @@
     2.4 +/*
     2.5 + * linux/arch/i386/kernel/sysenter.c
     2.6 + *
     2.7 + * (C) Copyright 2002 Linus Torvalds
     2.8 + *
     2.9 + * This file contains the needed initializations to support sysenter.
    2.10 + */
    2.11 +
    2.12 +#include <linux/init.h>
    2.13 +#include <linux/smp.h>
    2.14 +#include <linux/thread_info.h>
    2.15 +#include <linux/sched.h>
    2.16 +#include <linux/gfp.h>
    2.17 +#include <linux/string.h>
    2.18 +#include <linux/elf.h>
    2.19 +
    2.20 +#include <asm/cpufeature.h>
    2.21 +#include <asm/msr.h>
    2.22 +#include <asm/pgtable.h>
    2.23 +#include <asm/unistd.h>
    2.24 +
    2.25 +extern asmlinkage void sysenter_entry(void);
    2.26 +
    2.27 +void enable_sep_cpu(void)
    2.28 +{
    2.29 +	int cpu = get_cpu();
    2.30 +	struct tss_struct *tss = &per_cpu(init_tss, cpu);
    2.31 +
    2.32 +	if (!boot_cpu_has(X86_FEATURE_SEP)) {
    2.33 +		put_cpu();
    2.34 +		return;
    2.35 +	}
    2.36 +
    2.37 +	tss->ss1 = __KERNEL_CS;
    2.38 +	tss->esp1 = sizeof(struct tss_struct) + (unsigned long) tss;
    2.39 +	wrmsr(MSR_IA32_SYSENTER_CS, __KERNEL_CS, 0);
    2.40 +	wrmsr(MSR_IA32_SYSENTER_ESP, tss->esp1, 0);
    2.41 +	wrmsr(MSR_IA32_SYSENTER_EIP, (unsigned long) sysenter_entry, 0);
    2.42 +	put_cpu();	
    2.43 +}
    2.44 +
    2.45 +/*
    2.46 + * These symbols are defined by vsyscall.o to mark the bounds
    2.47 + * of the ELF DSO images included therein.
    2.48 + */
    2.49 +extern const char vsyscall_int80_start, vsyscall_int80_end;
    2.50 +extern const char vsyscall_sysenter_start, vsyscall_sysenter_end;
    2.51 +
    2.52 +int __init sysenter_setup(void)
    2.53 +{
    2.54 +	void *page = (void *)get_zeroed_page(GFP_ATOMIC);
    2.55 +
    2.56 +	__set_fixmap(FIX_VSYSCALL, __pa(page), PAGE_READONLY_EXEC);
    2.57 +
    2.58 +	if (!boot_cpu_has(X86_FEATURE_SEP)) {
    2.59 +		memcpy(page,
    2.60 +		       &vsyscall_int80_start,
    2.61 +		       &vsyscall_int80_end - &vsyscall_int80_start);
    2.62 +		return 0;
    2.63 +	}
    2.64 +
    2.65 +	memcpy(page,
    2.66 +	       &vsyscall_sysenter_start,
    2.67 +	       &vsyscall_sysenter_end - &vsyscall_sysenter_start);
    2.68 +
    2.69 +	return 0;
    2.70 +}

     3.1 --- /dev/null	Thu Jan 01 00:00:00 1970 +0000
     3.2 +++ b/linux-2.6-xen-sparse/arch/i386/kernel/vm86.c	Mon Feb 20 14:37:13 2006 +0000
     3.3 @@ -0,0 +1,809 @@
     3.4 +/*
     3.5 + *  linux/kernel/vm86.c
     3.6 + *
     3.7 + *  Copyright (C) 1994  Linus Torvalds
     3.8 + *
     3.9 + *  29 dec 2001 - Fixed oopses caused by unchecked access to the vm86
    3.10 + *                stack - Manfred Spraul <manfred@colorfullife.com>
    3.11 + *
    3.12 + *  22 mar 2002 - Manfred detected the stackfaults, but didn't handle
    3.13 + *                them correctly. Now the emulation will be in a
    3.14 + *                consistent state after stackfaults - Kasper Dupont
    3.15 + *                <kasperd@daimi.au.dk>
    3.16 + *
    3.17 + *  22 mar 2002 - Added missing clear_IF in set_vflags_* Kasper Dupont
    3.18 + *                <kasperd@daimi.au.dk>
    3.19 + *
    3.20 + *  ?? ??? 2002 - Fixed premature returns from handle_vm86_fault
    3.21 + *                caused by Kasper Dupont's changes - Stas Sergeev
    3.22 + *
    3.23 + *   4 apr 2002 - Fixed CHECK_IF_IN_TRAP broken by Stas' changes.
    3.24 + *                Kasper Dupont <kasperd@daimi.au.dk>
    3.25 + *
    3.26 + *   9 apr 2002 - Changed syntax of macros in handle_vm86_fault.
    3.27 + *                Kasper Dupont <kasperd@daimi.au.dk>
    3.28 + *
    3.29 + *   9 apr 2002 - Changed stack access macros to jump to a label
    3.30 + *                instead of returning to userspace. This simplifies
    3.31 + *                do_int, and is needed by handle_vm6_fault. Kasper
    3.32 + *                Dupont <kasperd@daimi.au.dk>
    3.33 + *
    3.34 + */
    3.35 +
    3.36 +#include <linux/capability.h>
    3.37 +#include <linux/config.h>
    3.38 +#include <linux/errno.h>
    3.39 +#include <linux/interrupt.h>
    3.40 +#include <linux/sched.h>
    3.41 +#include <linux/kernel.h>
    3.42 +#include <linux/signal.h>
    3.43 +#include <linux/string.h>
    3.44 +#include <linux/mm.h>
    3.45 +#include <linux/smp.h>
    3.46 +#include <linux/smp_lock.h>
    3.47 +#include <linux/highmem.h>
    3.48 +#include <linux/ptrace.h>
    3.49 +
    3.50 +#include <asm/uaccess.h>
    3.51 +#include <asm/io.h>
    3.52 +#include <asm/tlbflush.h>
    3.53 +#include <asm/irq.h>
    3.54 +
    3.55 +/*
    3.56 + * Known problems:
    3.57 + *
    3.58 + * Interrupt handling is not guaranteed:
    3.59 + * - a real x86 will disable all interrupts for one instruction
    3.60 + *   after a "mov ss,xx" to make stack handling atomic even without
    3.61 + *   the 'lss' instruction. We can't guarantee this in v86 mode,
    3.62 + *   as the next instruction might result in a page fault or similar.
    3.63 + * - a real x86 will have interrupts disabled for one instruction
    3.64 + *   past the 'sti' that enables them. We don't bother with all the
    3.65 + *   details yet.
    3.66 + *
    3.67 + * Let's hope these problems do not actually matter for anything.
    3.68 + */
    3.69 +
    3.70 +
    3.71 +#define KVM86	((struct kernel_vm86_struct *)regs)
    3.72 +#define VMPI 	KVM86->vm86plus
    3.73 +
    3.74 +
    3.75 +/*
    3.76 + * 8- and 16-bit register defines..
    3.77 + */
    3.78 +#define AL(regs)	(((unsigned char *)&((regs)->eax))[0])
    3.79 +#define AH(regs)	(((unsigned char *)&((regs)->eax))[1])
    3.80 +#define IP(regs)	(*(unsigned short *)&((regs)->eip))
    3.81 +#define SP(regs)	(*(unsigned short *)&((regs)->esp))
    3.82 +
    3.83 +/*
    3.84 + * virtual flags (16 and 32-bit versions)
    3.85 + */
    3.86 +#define VFLAGS	(*(unsigned short *)&(current->thread.v86flags))
    3.87 +#define VEFLAGS	(current->thread.v86flags)
    3.88 +
    3.89 +#define set_flags(X,new,mask) \
    3.90 +((X) = ((X) & ~(mask)) | ((new) & (mask)))
    3.91 +
    3.92 +#define SAFE_MASK	(0xDD5)
    3.93 +#define RETURN_MASK	(0xDFF)
    3.94 +
    3.95 +#define VM86_REGS_PART2 orig_eax
    3.96 +#define VM86_REGS_SIZE1 \
    3.97 +        ( (unsigned)( & (((struct kernel_vm86_regs *)0)->VM86_REGS_PART2) ) )
    3.98 +#define VM86_REGS_SIZE2 (sizeof(struct kernel_vm86_regs) - VM86_REGS_SIZE1)
    3.99 +
   3.100 +struct pt_regs * FASTCALL(save_v86_state(struct kernel_vm86_regs * regs));
   3.101 +struct pt_regs * fastcall save_v86_state(struct kernel_vm86_regs * regs)
   3.102 +{
   3.103 +	struct tss_struct *tss;
   3.104 +	struct pt_regs *ret;
   3.105 +	unsigned long tmp;
   3.106 +
   3.107 +	/*
   3.108 +	 * This gets called from entry.S with interrupts disabled, but
   3.109 +	 * from process context. Enable interrupts here, before trying
   3.110 +	 * to access user space.
   3.111 +	 */
   3.112 +	local_irq_enable();
   3.113 +
   3.114 +	if (!current->thread.vm86_info) {
   3.115 +		printk("no vm86_info: BAD\n");
   3.116 +		do_exit(SIGSEGV);
   3.117 +	}
   3.118 +	set_flags(regs->eflags, VEFLAGS, VIF_MASK | current->thread.v86mask);
   3.119 +	tmp = copy_to_user(&current->thread.vm86_info->regs,regs, VM86_REGS_SIZE1);
   3.120 +	tmp += copy_to_user(&current->thread.vm86_info->regs.VM86_REGS_PART2,
   3.121 +		&regs->VM86_REGS_PART2, VM86_REGS_SIZE2);
   3.122 +	tmp += put_user(current->thread.screen_bitmap,&current->thread.vm86_info->screen_bitmap);
   3.123 +	if (tmp) {
   3.124 +		printk("vm86: could not access userspace vm86_info\n");
   3.125 +		do_exit(SIGSEGV);
   3.126 +	}
   3.127 +
   3.128 +	tss = &per_cpu(init_tss, get_cpu());
   3.129 +	current->thread.esp0 = current->thread.saved_esp0;
   3.130 +	current->thread.sysenter_cs = __KERNEL_CS;
   3.131 +	load_esp0(tss, &current->thread);
   3.132 +	current->thread.saved_esp0 = 0;
   3.133 +	put_cpu();
   3.134 +
   3.135 +	loadsegment(fs, current->thread.saved_fs);
   3.136 +	loadsegment(gs, current->thread.saved_gs);
   3.137 +	ret = KVM86->regs32;
   3.138 +	return ret;
   3.139 +}
   3.140 +
   3.141 +static void mark_screen_rdonly(struct mm_struct *mm)
   3.142 +{
   3.143 +	pgd_t *pgd;
   3.144 +	pud_t *pud;
   3.145 +	pmd_t *pmd;
   3.146 +	pte_t *pte;
   3.147 +	spinlock_t *ptl;
   3.148 +	int i;
   3.149 +
   3.150 +	pgd = pgd_offset(mm, 0xA0000);
   3.151 +	if (pgd_none_or_clear_bad(pgd))
   3.152 +		goto out;
   3.153 +	pud = pud_offset(pgd, 0xA0000);
   3.154 +	if (pud_none_or_clear_bad(pud))
   3.155 +		goto out;
   3.156 +	pmd = pmd_offset(pud, 0xA0000);
   3.157 +	if (pmd_none_or_clear_bad(pmd))
   3.158 +		goto out;
   3.159 +	pte = pte_offset_map_lock(mm, pmd, 0xA0000, &ptl);
   3.160 +	for (i = 0; i < 32; i++) {
   3.161 +		if (pte_present(*pte))
   3.162 +			set_pte(pte, pte_wrprotect(*pte));
   3.163 +		pte++;
   3.164 +	}
   3.165 +	pte_unmap_unlock(pte, ptl);
   3.166 +out:
   3.167 +	flush_tlb();
   3.168 +}
   3.169 +
   3.170 +
   3.171 +
   3.172 +static int do_vm86_irq_handling(int subfunction, int irqnumber);
   3.173 +static void do_sys_vm86(struct kernel_vm86_struct *info, struct task_struct *tsk);
   3.174 +
   3.175 +asmlinkage int sys_vm86old(struct pt_regs regs)
   3.176 +{
   3.177 +	struct vm86_struct __user *v86 = (struct vm86_struct __user *)regs.ebx;
   3.178 +	struct kernel_vm86_struct info; /* declare this _on top_,
   3.179 +					 * this avoids wasting of stack space.
   3.180 +					 * This remains on the stack until we
   3.181 +					 * return to 32 bit user space.
   3.182 +					 */
   3.183 +	struct task_struct *tsk;
   3.184 +	int tmp, ret = -EPERM;
   3.185 +
   3.186 +	tsk = current;
   3.187 +	if (tsk->thread.saved_esp0)
   3.188 +		goto out;
   3.189 +	tmp  = copy_from_user(&info, v86, VM86_REGS_SIZE1);
   3.190 +	tmp += copy_from_user(&info.regs.VM86_REGS_PART2, &v86->regs.VM86_REGS_PART2,
   3.191 +		(long)&info.vm86plus - (long)&info.regs.VM86_REGS_PART2);
   3.192 +	ret = -EFAULT;
   3.193 +	if (tmp)
   3.194 +		goto out;
   3.195 +	memset(&info.vm86plus, 0, (int)&info.regs32 - (int)&info.vm86plus);
   3.196 +	info.regs32 = &regs;
   3.197 +	tsk->thread.vm86_info = v86;
   3.198 +	do_sys_vm86(&info, tsk);
   3.199 +	ret = 0;	/* we never return here */
   3.200 +out:
   3.201 +	return ret;
   3.202 +}
   3.203 +
   3.204 +
   3.205 +asmlinkage int sys_vm86(struct pt_regs regs)
   3.206 +{
   3.207 +	struct kernel_vm86_struct info; /* declare this _on top_,
   3.208 +					 * this avoids wasting of stack space.
   3.209 +					 * This remains on the stack until we
   3.210 +					 * return to 32 bit user space.
   3.211 +					 */
   3.212 +	struct task_struct *tsk;
   3.213 +	int tmp, ret;
   3.214 +	struct vm86plus_struct __user *v86;
   3.215 +
   3.216 +	tsk = current;
   3.217 +	switch (regs.ebx) {
   3.218 +		case VM86_REQUEST_IRQ:
   3.219 +		case VM86_FREE_IRQ:
   3.220 +		case VM86_GET_IRQ_BITS:
   3.221 +		case VM86_GET_AND_RESET_IRQ:
   3.222 +			ret = do_vm86_irq_handling(regs.ebx, (int)regs.ecx);
   3.223 +			goto out;
   3.224 +		case VM86_PLUS_INSTALL_CHECK:
   3.225 +			/* NOTE: on old vm86 stuff this will return the error
   3.226 +			   from access_ok(), because the subfunction is
   3.227 +			   interpreted as (invalid) address to vm86_struct.
   3.228 +			   So the installation check works.
   3.229 +			 */
   3.230 +			ret = 0;
   3.231 +			goto out;
   3.232 +	}
   3.233 +
   3.234 +	/* we come here only for functions VM86_ENTER, VM86_ENTER_NO_BYPASS */
   3.235 +	ret = -EPERM;
   3.236 +	if (tsk->thread.saved_esp0)
   3.237 +		goto out;
   3.238 +	v86 = (struct vm86plus_struct __user *)regs.ecx;
   3.239 +	tmp  = copy_from_user(&info, v86, VM86_REGS_SIZE1);
   3.240 +	tmp += copy_from_user(&info.regs.VM86_REGS_PART2, &v86->regs.VM86_REGS_PART2,
   3.241 +		(long)&info.regs32 - (long)&info.regs.VM86_REGS_PART2);
   3.242 +	ret = -EFAULT;
   3.243 +	if (tmp)
   3.244 +		goto out;
   3.245 +	info.regs32 = &regs;
   3.246 +	info.vm86plus.is_vm86pus = 1;
   3.247 +	tsk->thread.vm86_info = (struct vm86_struct __user *)v86;
   3.248 +	do_sys_vm86(&info, tsk);
   3.249 +	ret = 0;	/* we never return here */
   3.250 +out:
   3.251 +	return ret;
   3.252 +}
   3.253 +
   3.254 +
   3.255 +static void do_sys_vm86(struct kernel_vm86_struct *info, struct task_struct *tsk)
   3.256 +{
   3.257 +	struct tss_struct *tss;
   3.258 +/*
   3.259 + * make sure the vm86() system call doesn't try to do anything silly
   3.260 + */
   3.261 +	info->regs.__null_ds = 0;
   3.262 +	info->regs.__null_es = 0;
   3.263 +
   3.264 +/* we are clearing fs,gs later just before "jmp resume_userspace",
   3.265 + * because starting with Linux 2.1.x they aren't no longer saved/restored
   3.266 + */
   3.267 +
   3.268 +/*
   3.269 + * The eflags register is also special: we cannot trust that the user
   3.270 + * has set it up safely, so this makes sure interrupt etc flags are
   3.271 + * inherited from protected mode.
   3.272 + */
   3.273 + 	VEFLAGS = info->regs.eflags;
   3.274 +	info->regs.eflags &= SAFE_MASK;
   3.275 +	info->regs.eflags |= info->regs32->eflags & ~SAFE_MASK;
   3.276 +	info->regs.eflags |= VM_MASK;
   3.277 +
   3.278 +	switch (info->cpu_type) {
   3.279 +		case CPU_286:
   3.280 +			tsk->thread.v86mask = 0;
   3.281 +			break;
   3.282 +		case CPU_386:
   3.283 +			tsk->thread.v86mask = NT_MASK | IOPL_MASK;
   3.284 +			break;
   3.285 +		case CPU_486:
   3.286 +			tsk->thread.v86mask = AC_MASK | NT_MASK | IOPL_MASK;
   3.287 +			break;
   3.288 +		default:
   3.289 +			tsk->thread.v86mask = ID_MASK | AC_MASK | NT_MASK | IOPL_MASK;
   3.290 +			break;
   3.291 +	}
   3.292 +
   3.293 +/*
   3.294 + * Save old state, set default return value (%eax) to 0
   3.295 + */
   3.296 +	info->regs32->eax = 0;
   3.297 +	tsk->thread.saved_esp0 = tsk->thread.esp0;
   3.298 +	savesegment(fs, tsk->thread.saved_fs);
   3.299 +	savesegment(gs, tsk->thread.saved_gs);
   3.300 +
   3.301 +	tss = &per_cpu(init_tss, get_cpu());
   3.302 +	tsk->thread.esp0 = (unsigned long) &info->VM86_TSS_ESP0;
   3.303 +	if (cpu_has_sep)
   3.304 +		tsk->thread.sysenter_cs = 0;
   3.305 +	load_esp0(tss, &tsk->thread);
   3.306 +	put_cpu();
   3.307 +
   3.308 +	tsk->thread.screen_bitmap = info->screen_bitmap;
   3.309 +	if (info->flags & VM86_SCREEN_BITMAP)
   3.310 +		mark_screen_rdonly(tsk->mm);
   3.311 +	__asm__ __volatile__(
   3.312 +		"xorl %%eax,%%eax; movl %%eax,%%fs; movl %%eax,%%gs\n\t"
   3.313 +		"movl %0,%%esp\n\t"
   3.314 +		"movl %1,%%ebp\n\t"
   3.315 +		"jmp resume_userspace"
   3.316 +		: /* no outputs */
   3.317 +		:"r" (&info->regs), "r" (task_thread_info(tsk)) : "ax");
   3.318 +	/* we never return here */
   3.319 +}
   3.320 +
   3.321 +static inline void return_to_32bit(struct kernel_vm86_regs * regs16, int retval)
   3.322 +{
   3.323 +	struct pt_regs * regs32;
   3.324 +
   3.325 +	regs32 = save_v86_state(regs16);
   3.326 +	regs32->eax = retval;
   3.327 +	__asm__ __volatile__("movl %0,%%esp\n\t"
   3.328 +		"movl %1,%%ebp\n\t"
   3.329 +		"jmp resume_userspace"
   3.330 +		: : "r" (regs32), "r" (current_thread_info()));
   3.331 +}
   3.332 +
   3.333 +static inline void set_IF(struct kernel_vm86_regs * regs)
   3.334 +{
   3.335 +	VEFLAGS |= VIF_MASK;
   3.336 +	if (VEFLAGS & VIP_MASK)
   3.337 +		return_to_32bit(regs, VM86_STI);
   3.338 +}
   3.339 +
   3.340 +static inline void clear_IF(struct kernel_vm86_regs * regs)
   3.341 +{
   3.342 +	VEFLAGS &= ~VIF_MASK;
   3.343 +}
   3.344 +
   3.345 +static inline void clear_TF(struct kernel_vm86_regs * regs)
   3.346 +{
   3.347 +	regs->eflags &= ~TF_MASK;
   3.348 +}
   3.349 +
   3.350 +static inline void clear_AC(struct kernel_vm86_regs * regs)
   3.351 +{
   3.352 +	regs->eflags &= ~AC_MASK;
   3.353 +}
   3.354 +
   3.355 +/* It is correct to call set_IF(regs) from the set_vflags_*
   3.356 + * functions. However someone forgot to call clear_IF(regs)
   3.357 + * in the opposite case.
   3.358 + * After the command sequence CLI PUSHF STI POPF you should
   3.359 + * end up with interrups disabled, but you ended up with
   3.360 + * interrupts enabled.
   3.361 + *  ( I was testing my own changes, but the only bug I
   3.362 + *    could find was in a function I had not changed. )
   3.363 + * [KD]
   3.364 + */
   3.365 +
   3.366 +static inline void set_vflags_long(unsigned long eflags, struct kernel_vm86_regs * regs)
   3.367 +{
   3.368 +	set_flags(VEFLAGS, eflags, current->thread.v86mask);
   3.369 +	set_flags(regs->eflags, eflags, SAFE_MASK);
   3.370 +	if (eflags & IF_MASK)
   3.371 +		set_IF(regs);
   3.372 +	else
   3.373 +		clear_IF(regs);
   3.374 +}
   3.375 +
   3.376 +static inline void set_vflags_short(unsigned short flags, struct kernel_vm86_regs * regs)
   3.377 +{
   3.378 +	set_flags(VFLAGS, flags, current->thread.v86mask);
   3.379 +	set_flags(regs->eflags, flags, SAFE_MASK);
   3.380 +	if (flags & IF_MASK)
   3.381 +		set_IF(regs);
   3.382 +	else
   3.383 +		clear_IF(regs);
   3.384 +}
   3.385 +
   3.386 +static inline unsigned long get_vflags(struct kernel_vm86_regs * regs)
   3.387 +{
   3.388 +	unsigned long flags = regs->eflags & RETURN_MASK;
   3.389 +
   3.390 +	if (VEFLAGS & VIF_MASK)
   3.391 +		flags |= IF_MASK;
   3.392 +	flags |= IOPL_MASK;
   3.393 +	return flags | (VEFLAGS & current->thread.v86mask);
   3.394 +}
   3.395 +
   3.396 +static inline int is_revectored(int nr, struct revectored_struct * bitmap)
   3.397 +{
   3.398 +	__asm__ __volatile__("btl %2,%1\n\tsbbl %0,%0"
   3.399 +		:"=r" (nr)
   3.400 +		:"m" (*bitmap),"r" (nr));
   3.401 +	return nr;
   3.402 +}
   3.403 +
   3.404 +#define val_byte(val, n) (((__u8 *)&val)[n])
   3.405 +
   3.406 +#define pushb(base, ptr, val, err_label) \
   3.407 +	do { \
   3.408 +		__u8 __val = val; \
   3.409 +		ptr--; \
   3.410 +		if (put_user(__val, base + ptr) < 0) \
   3.411 +			goto err_label; \
   3.412 +	} while(0)
   3.413 +
   3.414 +#define pushw(base, ptr, val, err_label) \
   3.415 +	do { \
   3.416 +		__u16 __val = val; \
   3.417 +		ptr--; \
   3.418 +		if (put_user(val_byte(__val, 1), base + ptr) < 0) \
   3.419 +			goto err_label; \
   3.420 +		ptr--; \
   3.421 +		if (put_user(val_byte(__val, 0), base + ptr) < 0) \
   3.422 +			goto err_label; \
   3.423 +	} while(0)
   3.424 +
   3.425 +#define pushl(base, ptr, val, err_label) \
   3.426 +	do { \
   3.427 +		__u32 __val = val; \
   3.428 +		ptr--; \
   3.429 +		if (put_user(val_byte(__val, 3), base + ptr) < 0) \
   3.430 +			goto err_label; \
   3.431 +		ptr--; \
   3.432 +		if (put_user(val_byte(__val, 2), base + ptr) < 0) \
   3.433 +			goto err_label; \
   3.434 +		ptr--; \
   3.435 +		if (put_user(val_byte(__val, 1), base + ptr) < 0) \
   3.436 +			goto err_label; \
   3.437 +		ptr--; \
   3.438 +		if (put_user(val_byte(__val, 0), base + ptr) < 0) \
   3.439 +			goto err_label; \
   3.440 +	} while(0)
   3.441 +
   3.442 +#define popb(base, ptr, err_label) \
   3.443 +	({ \
   3.444 +		__u8 __res; \
   3.445 +		if (get_user(__res, base + ptr) < 0) \
   3.446 +			goto err_label; \
   3.447 +		ptr++; \
   3.448 +		__res; \
   3.449 +	})
   3.450 +
   3.451 +#define popw(base, ptr, err_label) \
   3.452 +	({ \
   3.453 +		__u16 __res; \
   3.454 +		if (get_user(val_byte(__res, 0), base + ptr) < 0) \
   3.455 +			goto err_label; \
   3.456 +		ptr++; \
   3.457 +		if (get_user(val_byte(__res, 1), base + ptr) < 0) \
   3.458 +			goto err_label; \
   3.459 +		ptr++; \
   3.460 +		__res; \
   3.461 +	})
   3.462 +
   3.463 +#define popl(base, ptr, err_label) \
   3.464 +	({ \
   3.465 +		__u32 __res; \
   3.466 +		if (get_user(val_byte(__res, 0), base + ptr) < 0) \
   3.467 +			goto err_label; \
   3.468 +		ptr++; \
   3.469 +		if (get_user(val_byte(__res, 1), base + ptr) < 0) \
   3.470 +			goto err_label; \
   3.471 +		ptr++; \
   3.472 +		if (get_user(val_byte(__res, 2), base + ptr) < 0) \
   3.473 +			goto err_label; \
   3.474 +		ptr++; \
   3.475 +		if (get_user(val_byte(__res, 3), base + ptr) < 0) \
   3.476 +			goto err_label; \
   3.477 +		ptr++; \
   3.478 +		__res; \
   3.479 +	})
   3.480 +
   3.481 +/* There are so many possible reasons for this function to return
   3.482 + * VM86_INTx, so adding another doesn't bother me. We can expect
   3.483 + * userspace programs to be able to handle it. (Getting a problem
   3.484 + * in userspace is always better than an Oops anyway.) [KD]
   3.485 + */
   3.486 +static void do_int(struct kernel_vm86_regs *regs, int i,
   3.487 +    unsigned char __user * ssp, unsigned short sp)
   3.488 +{
   3.489 +	unsigned long __user *intr_ptr;
   3.490 +	unsigned long segoffs;
   3.491 +
   3.492 +	if (regs->cs == BIOSSEG)
   3.493 +		goto cannot_handle;
   3.494 +	if (is_revectored(i, &KVM86->int_revectored))
   3.495 +		goto cannot_handle;
   3.496 +	if (i==0x21 && is_revectored(AH(regs),&KVM86->int21_revectored))
   3.497 +		goto cannot_handle;
   3.498 +	intr_ptr = (unsigned long __user *) (i << 2);
   3.499 +	if (get_user(segoffs, intr_ptr))
   3.500 +		goto cannot_handle;
   3.501 +	if ((segoffs >> 16) == BIOSSEG)
   3.502 +		goto cannot_handle;
   3.503 +	pushw(ssp, sp, get_vflags(regs), cannot_handle);
   3.504 +	pushw(ssp, sp, regs->cs, cannot_handle);
   3.505 +	pushw(ssp, sp, IP(regs), cannot_handle);
   3.506 +	regs->cs = segoffs >> 16;
   3.507 +	SP(regs) -= 6;
   3.508 +	IP(regs) = segoffs & 0xffff;
   3.509 +	clear_TF(regs);
   3.510 +	clear_IF(regs);
   3.511 +	clear_AC(regs);
   3.512 +	return;
   3.513 +
   3.514 +cannot_handle:
   3.515 +	return_to_32bit(regs, VM86_INTx + (i << 8));
   3.516 +}
   3.517 +
   3.518 +int handle_vm86_trap(struct kernel_vm86_regs * regs, long error_code, int trapno)
   3.519 +{
   3.520 +	if (VMPI.is_vm86pus) {
   3.521 +		if ( (trapno==3) || (trapno==1) )
   3.522 +			return_to_32bit(regs, VM86_TRAP + (trapno << 8));
   3.523 +		do_int(regs, trapno, (unsigned char __user *) (regs->ss << 4), SP(regs));
   3.524 +		return 0;
   3.525 +	}
   3.526 +	if (trapno !=1)
   3.527 +		return 1; /* we let this handle by the calling routine */
   3.528 +	if (current->ptrace & PT_PTRACED) {
   3.529 +		unsigned long flags;
   3.530 +		spin_lock_irqsave(&current->sighand->siglock, flags);
   3.531 +		sigdelset(&current->blocked, SIGTRAP);
   3.532 +		recalc_sigpending();
   3.533 +		spin_unlock_irqrestore(&current->sighand->siglock, flags);
   3.534 +	}
   3.535 +	send_sig(SIGTRAP, current, 1);
   3.536 +	current->thread.trap_no = trapno;
   3.537 +	current->thread.error_code = error_code;
   3.538 +	return 0;
   3.539 +}
   3.540 +
   3.541 +void handle_vm86_fault(struct kernel_vm86_regs * regs, long error_code)
   3.542 +{
   3.543 +	unsigned char opcode;
   3.544 +	unsigned char __user *csp;
   3.545 +	unsigned char __user *ssp;
   3.546 +	unsigned short ip, sp, orig_flags;
   3.547 +	int data32, pref_done;
   3.548 +
   3.549 +#define CHECK_IF_IN_TRAP \
   3.550 +	if (VMPI.vm86dbg_active && VMPI.vm86dbg_TFpendig) \
   3.551 +		newflags |= TF_MASK
   3.552 +#define VM86_FAULT_RETURN do { \
   3.553 +	if (VMPI.force_return_for_pic  && (VEFLAGS & (IF_MASK | VIF_MASK))) \
   3.554 +		return_to_32bit(regs, VM86_PICRETURN); \
   3.555 +	if (orig_flags & TF_MASK) \
   3.556 +		handle_vm86_trap(regs, 0, 1); \
   3.557 +	return; } while (0)
   3.558 +
   3.559 +	orig_flags = *(unsigned short *)&regs->eflags;
   3.560 +
   3.561 +	csp = (unsigned char __user *) (regs->cs << 4);
   3.562 +	ssp = (unsigned char __user *) (regs->ss << 4);
   3.563 +	sp = SP(regs);
   3.564 +	ip = IP(regs);
   3.565 +
   3.566 +	data32 = 0;
   3.567 +	pref_done = 0;
   3.568 +	do {
   3.569 +		switch (opcode = popb(csp, ip, simulate_sigsegv)) {
   3.570 +			case 0x66:      /* 32-bit data */     data32=1; break;
   3.571 +			case 0x67:      /* 32-bit address */  break;
   3.572 +			case 0x2e:      /* CS */              break;
   3.573 +			case 0x3e:      /* DS */              break;
   3.574 +			case 0x26:      /* ES */              break;
   3.575 +			case 0x36:      /* SS */              break;
   3.576 +			case 0x65:      /* GS */              break;
   3.577 +			case 0x64:      /* FS */              break;
   3.578 +			case 0xf2:      /* repnz */       break;
   3.579 +			case 0xf3:      /* rep */             break;
   3.580 +			default: pref_done = 1;
   3.581 +		}
   3.582 +	} while (!pref_done);
   3.583 +
   3.584 +	switch (opcode) {
   3.585 +
   3.586 +	/* pushf */
   3.587 +	case 0x9c:
   3.588 +		if (data32) {
   3.589 +			pushl(ssp, sp, get_vflags(regs), simulate_sigsegv);
   3.590 +			SP(regs) -= 4;
   3.591 +		} else {
   3.592 +			pushw(ssp, sp, get_vflags(regs), simulate_sigsegv);
   3.593 +			SP(regs) -= 2;
   3.594 +		}
   3.595 +		IP(regs) = ip;
   3.596 +		VM86_FAULT_RETURN;
   3.597 +
   3.598 +	/* popf */
   3.599 +	case 0x9d:
   3.600 +		{
   3.601 +		unsigned long newflags;
   3.602 +		if (data32) {
   3.603 +			newflags=popl(ssp, sp, simulate_sigsegv);
   3.604 +			SP(regs) += 4;
   3.605 +		} else {
   3.606 +			newflags = popw(ssp, sp, simulate_sigsegv);
   3.607 +			SP(regs) += 2;
   3.608 +		}
   3.609 +		IP(regs) = ip;
   3.610 +		CHECK_IF_IN_TRAP;
   3.611 +		if (data32) {
   3.612 +			set_vflags_long(newflags, regs);
   3.613 +		} else {
   3.614 +			set_vflags_short(newflags, regs);
   3.615 +		}
   3.616 +		VM86_FAULT_RETURN;
   3.617 +		}
   3.618 +
   3.619 +	/* int xx */
   3.620 +	case 0xcd: {
   3.621 +		int intno=popb(csp, ip, simulate_sigsegv);
   3.622 +		IP(regs) = ip;
   3.623 +		if (VMPI.vm86dbg_active) {
   3.624 +			if ( (1 << (intno &7)) & VMPI.vm86dbg_intxxtab[intno >> 3] )
   3.625 +				return_to_32bit(regs, VM86_INTx + (intno << 8));
   3.626 +		}
   3.627 +		do_int(regs, intno, ssp, sp);
   3.628 +		return;
   3.629 +	}
   3.630 +
   3.631 +	/* iret */
   3.632 +	case 0xcf:
   3.633 +		{
   3.634 +		unsigned long newip;
   3.635 +		unsigned long newcs;
   3.636 +		unsigned long newflags;
   3.637 +		if (data32) {
   3.638 +			newip=popl(ssp, sp, simulate_sigsegv);
   3.639 +			newcs=popl(ssp, sp, simulate_sigsegv);
   3.640 +			newflags=popl(ssp, sp, simulate_sigsegv);
   3.641 +			SP(regs) += 12;
   3.642 +		} else {
   3.643 +			newip = popw(ssp, sp, simulate_sigsegv);
   3.644 +			newcs = popw(ssp, sp, simulate_sigsegv);
   3.645 +			newflags = popw(ssp, sp, simulate_sigsegv);
   3.646 +			SP(regs) += 6;
   3.647 +		}
   3.648 +		IP(regs) = newip;
   3.649 +		regs->cs = newcs;
   3.650 +		CHECK_IF_IN_TRAP;
   3.651 +		if (data32) {
   3.652 +			set_vflags_long(newflags, regs);
   3.653 +		} else {
   3.654 +			set_vflags_short(newflags, regs);
   3.655 +		}
   3.656 +		VM86_FAULT_RETURN;
   3.657 +		}
   3.658 +
   3.659 +	/* cli */
   3.660 +	case 0xfa:
   3.661 +		IP(regs) = ip;
   3.662 +		clear_IF(regs);
   3.663 +		VM86_FAULT_RETURN;
   3.664 +
   3.665 +	/* sti */
   3.666 +	/*
   3.667 +	 * Damn. This is incorrect: the 'sti' instruction should actually
   3.668 +	 * enable interrupts after the /next/ instruction. Not good.
   3.669 +	 *
   3.670 +	 * Probably needs some horsing around with the TF flag. Aiee..
   3.671 +	 */
   3.672 +	case 0xfb:
   3.673 +		IP(regs) = ip;
   3.674 +		set_IF(regs);
   3.675 +		VM86_FAULT_RETURN;
   3.676 +
   3.677 +	default:
   3.678 +		return_to_32bit(regs, VM86_UNKNOWN);
   3.679 +	}
   3.680 +
   3.681 +	return;
   3.682 +
   3.683 +simulate_sigsegv:
   3.684 +	/* FIXME: After a long discussion with Stas we finally
   3.685 +	 *        agreed, that this is wrong. Here we should
   3.686 +	 *        really send a SIGSEGV to the user program.
   3.687 +	 *        But how do we create the correct context? We
   3.688 +	 *        are inside a general protection fault handler
   3.689 +	 *        and has just returned from a page fault handler.
   3.690 +	 *        The correct context for the signal handler
   3.691 +	 *        should be a mixture of the two, but how do we
   3.692 +	 *        get the information? [KD]
   3.693 +	 */
   3.694 +	return_to_32bit(regs, VM86_UNKNOWN);
   3.695 +}
   3.696 +
   3.697 +/* ---------------- vm86 special IRQ passing stuff ----------------- */
   3.698 +
   3.699 +#define VM86_IRQNAME		"vm86irq"
   3.700 +
   3.701 +static struct vm86_irqs {
   3.702 +	struct task_struct *tsk;
   3.703 +	int sig;
   3.704 +} vm86_irqs[16];
   3.705 +
   3.706 +static DEFINE_SPINLOCK(irqbits_lock);
   3.707 +static int irqbits;
   3.708 +
   3.709 +#define ALLOWED_SIGS ( 1 /* 0 = don't send a signal */ \
   3.710 +	| (1 << SIGUSR1) | (1 << SIGUSR2) | (1 << SIGIO)  | (1 << SIGURG) \
   3.711 +	| (1 << SIGUNUSED) )
   3.712 +	
   3.713 +static irqreturn_t irq_handler(int intno, void *dev_id, struct pt_regs * regs)
   3.714 +{
   3.715 +	int irq_bit;
   3.716 +	unsigned long flags;
   3.717 +
   3.718 +	spin_lock_irqsave(&irqbits_lock, flags);	
   3.719 +	irq_bit = 1 << intno;
   3.720 +	if ((irqbits & irq_bit) || ! vm86_irqs[intno].tsk)
   3.721 +		goto out;
   3.722 +	irqbits |= irq_bit;
   3.723 +	if (vm86_irqs[intno].sig)
   3.724 +		send_sig(vm86_irqs[intno].sig, vm86_irqs[intno].tsk, 1);
   3.725 +	/*
   3.726 +	 * IRQ will be re-enabled when user asks for the irq (whether
   3.727 +	 * polling or as a result of the signal)
   3.728 +	 */
   3.729 +	disable_irq_nosync(intno);
   3.730 +	spin_unlock_irqrestore(&irqbits_lock, flags);
   3.731 +	return IRQ_HANDLED;
   3.732 +
   3.733 +out:
   3.734 +	spin_unlock_irqrestore(&irqbits_lock, flags);	
   3.735 +	return IRQ_NONE;
   3.736 +}
   3.737 +
   3.738 +static inline void free_vm86_irq(int irqnumber)
   3.739 +{
   3.740 +	unsigned long flags;
   3.741 +
   3.742 +	free_irq(irqnumber, NULL);
   3.743 +	vm86_irqs[irqnumber].tsk = NULL;
   3.744 +
   3.745 +	spin_lock_irqsave(&irqbits_lock, flags);	
   3.746 +	irqbits &= ~(1 << irqnumber);
   3.747 +	spin_unlock_irqrestore(&irqbits_lock, flags);	
   3.748 +}
   3.749 +
   3.750 +void release_vm86_irqs(struct task_struct *task)
   3.751 +{
   3.752 +	int i;
   3.753 +	for (i = FIRST_VM86_IRQ ; i <= LAST_VM86_IRQ; i++)
   3.754 +	    if (vm86_irqs[i].tsk == task)
   3.755 +		free_vm86_irq(i);
   3.756 +}
   3.757 +
   3.758 +static inline int get_and_reset_irq(int irqnumber)
   3.759 +{
   3.760 +	int bit;
   3.761 +	unsigned long flags;
   3.762 +	int ret = 0;
   3.763 +	
   3.764 +	if (invalid_vm86_irq(irqnumber)) return 0;
   3.765 +	if (vm86_irqs[irqnumber].tsk != current) return 0;
   3.766 +	spin_lock_irqsave(&irqbits_lock, flags);	
   3.767 +	bit = irqbits & (1 << irqnumber);
   3.768 +	irqbits &= ~bit;
   3.769 +	if (bit) {
   3.770 +		enable_irq(irqnumber);
   3.771 +		ret = 1;
   3.772 +	}
   3.773 +
   3.774 +	spin_unlock_irqrestore(&irqbits_lock, flags);	
   3.775 +	return ret;
   3.776 +}
   3.777 +
   3.778 +
   3.779 +static int do_vm86_irq_handling(int subfunction, int irqnumber)
   3.780 +{
   3.781 +	int ret;
   3.782 +	switch (subfunction) {
   3.783 +		case VM86_GET_AND_RESET_IRQ: {
   3.784 +			return get_and_reset_irq(irqnumber);
   3.785 +		}
   3.786 +		case VM86_GET_IRQ_BITS: {
   3.787 +			return irqbits;
   3.788 +		}
   3.789 +		case VM86_REQUEST_IRQ: {
   3.790 +			int sig = irqnumber >> 8;
   3.791 +			int irq = irqnumber & 255;
   3.792 +			if (!capable(CAP_SYS_ADMIN)) return -EPERM;
   3.793 +			if (!((1 << sig) & ALLOWED_SIGS)) return -EPERM;
   3.794 +			if (invalid_vm86_irq(irq)) return -EPERM;
   3.795 +			if (vm86_irqs[irq].tsk) return -EPERM;
   3.796 +			ret = request_irq(irq, &irq_handler, 0, VM86_IRQNAME, NULL);
   3.797 +			if (ret) return ret;
   3.798 +			vm86_irqs[irq].sig = sig;
   3.799 +			vm86_irqs[irq].tsk = current;
   3.800 +			return irq;
   3.801 +		}
   3.802 +		case  VM86_FREE_IRQ: {
   3.803 +			if (invalid_vm86_irq(irqnumber)) return -EPERM;
   3.804 +			if (!vm86_irqs[irqnumber].tsk) return 0;
   3.805 +			if (vm86_irqs[irqnumber].tsk != current) return -EPERM;
   3.806 +			free_vm86_irq(irqnumber);
   3.807 +			return 0;
   3.808 +		}
   3.809 +	}
   3.810 +	return -EINVAL;
   3.811 +}
   3.812 +

     4.1 --- /dev/null	Thu Jan 01 00:00:00 1970 +0000
     4.2 +++ b/linux-2.6-xen-sparse/arch/i386/kernel/vsyscall.S	Mon Feb 20 14:37:13 2006 +0000
     4.3 @@ -0,0 +1,15 @@
     4.4 +#include <linux/init.h>
     4.5 +
     4.6 +__INITDATA
     4.7 +
     4.8 +	.globl vsyscall_int80_start, vsyscall_int80_end
     4.9 +vsyscall_int80_start:
    4.10 +	.incbin "arch/i386/kernel/vsyscall-int80.so"
    4.11 +vsyscall_int80_end:
    4.12 +
    4.13 +	.globl vsyscall_sysenter_start, vsyscall_sysenter_end
    4.14 +vsyscall_sysenter_start:
    4.15 +	.incbin "arch/i386/kernel/vsyscall-sysenter.so"
    4.16 +vsyscall_sysenter_end:
    4.17 +
    4.18 +__FINIT