This commit was manufactured by cvs2svn to create branch 'vserver'.
[linux-2.6.git] / arch / x86_64 / kernel / ldt-xen.c
diff --git a/arch/x86_64/kernel/ldt-xen.c b/arch/x86_64/kernel/ldt-xen.c
new file mode 100644 (file)
index 0000000..6efbb64
--- /dev/null
@@ -0,0 +1,282 @@
+/*
+ * linux/arch/x86_64/kernel/ldt.c
+ *
+ * Copyright (C) 1992 Krishna Balasubramanian and Linus Torvalds
+ * Copyright (C) 1999 Ingo Molnar <mingo@redhat.com>
+ * Copyright (C) 2002 Andi Kleen
+ * 
+ * This handles calls from both 32bit and 64bit mode.
+ */
+
+#include <linux/errno.h>
+#include <linux/sched.h>
+#include <linux/string.h>
+#include <linux/mm.h>
+#include <linux/smp.h>
+#include <linux/smp_lock.h>
+#include <linux/vmalloc.h>
+#include <linux/slab.h>
+
+#include <asm/uaccess.h>
+#include <asm/system.h>
+#include <asm/ldt.h>
+#include <asm/desc.h>
+#include <asm/proto.h>
+#include <asm/pgalloc.h>
+
+#ifdef CONFIG_SMP /* avoids "defined but not used" warnig */
+static void flush_ldt(void *null)
+{
+       if (current->active_mm)
+               load_LDT(&current->active_mm->context);
+}
+#endif
+
+static int alloc_ldt(mm_context_t *pc, unsigned mincount, int reload)
+{
+       void *oldldt;
+       void *newldt;
+       unsigned oldsize;
+
+       if (mincount <= (unsigned)pc->size)
+               return 0;
+       oldsize = pc->size;
+       mincount = (mincount+511)&(~511);
+       if (mincount*LDT_ENTRY_SIZE > PAGE_SIZE)
+               newldt = vmalloc(mincount*LDT_ENTRY_SIZE);
+       else
+               newldt = kmalloc(mincount*LDT_ENTRY_SIZE, GFP_KERNEL);
+
+       if (!newldt)
+               return -ENOMEM;
+
+       if (oldsize)
+               memcpy(newldt, pc->ldt, oldsize*LDT_ENTRY_SIZE);
+       oldldt = pc->ldt;
+       memset(newldt+oldsize*LDT_ENTRY_SIZE, 0, (mincount-oldsize)*LDT_ENTRY_SIZE);
+       wmb();
+       pc->ldt = newldt;
+       wmb();
+       pc->size = mincount;
+       wmb();
+       if (reload) {
+#ifdef CONFIG_SMP
+               cpumask_t mask;
+
+               preempt_disable();
+#endif
+               make_pages_readonly(
+                       pc->ldt,
+                       (pc->size * LDT_ENTRY_SIZE) / PAGE_SIZE,
+                       XENFEAT_writable_descriptor_tables);
+               load_LDT(pc);
+#ifdef CONFIG_SMP
+               mask = cpumask_of_cpu(smp_processor_id());
+               if (!cpus_equal(current->mm->cpu_vm_mask, mask))
+                       smp_call_function(flush_ldt, NULL, 1, 1);
+               preempt_enable();
+#endif
+       }
+       if (oldsize) {
+               make_pages_writable(
+                       oldldt,
+                       (oldsize * LDT_ENTRY_SIZE) / PAGE_SIZE,
+                       XENFEAT_writable_descriptor_tables);
+               if (oldsize*LDT_ENTRY_SIZE > PAGE_SIZE)
+                       vfree(oldldt);
+               else
+                       kfree(oldldt);
+       }
+       return 0;
+}
+
+static inline int copy_ldt(mm_context_t *new, mm_context_t *old)
+{
+       int err = alloc_ldt(new, old->size, 0);
+       if (err < 0)
+               return err;
+       memcpy(new->ldt, old->ldt, old->size*LDT_ENTRY_SIZE);
+       make_pages_readonly(
+               new->ldt,
+               (new->size * LDT_ENTRY_SIZE) / PAGE_SIZE,
+               XENFEAT_writable_descriptor_tables);
+       return 0;
+}
+
+/*
+ * we do not have to muck with descriptors here, that is
+ * done in switch_mm() as needed.
+ */
+int init_new_context(struct task_struct *tsk, struct mm_struct *mm)
+{
+       struct mm_struct * old_mm;
+       int retval = 0;
+
+       memset(&mm->context, 0, sizeof(mm->context));
+       init_MUTEX(&mm->context.sem);
+       old_mm = current->mm;
+       if (old_mm && old_mm->context.size > 0) {
+               down(&old_mm->context.sem);
+               retval = copy_ldt(&mm->context, &old_mm->context);
+               up(&old_mm->context.sem);
+       }
+       if (retval == 0) {
+               spin_lock(&mm_unpinned_lock);
+               list_add(&mm->context.unpinned, &mm_unpinned);
+               spin_unlock(&mm_unpinned_lock);
+       }
+       return retval;
+}
+
+/*
+ * 
+ * Don't touch the LDT register - we're already in the next thread.
+ */
+void destroy_context(struct mm_struct *mm)
+{
+       if (mm->context.size) {
+               if (mm == current->active_mm)
+                       clear_LDT();
+               make_pages_writable(
+                       mm->context.ldt,
+                       (mm->context.size * LDT_ENTRY_SIZE) / PAGE_SIZE,
+                       XENFEAT_writable_descriptor_tables);
+               if (mm->context.size*LDT_ENTRY_SIZE > PAGE_SIZE)
+                       vfree(mm->context.ldt);
+               else
+                       kfree(mm->context.ldt);
+               mm->context.size = 0;
+       }
+       if (!mm->context.pinned) {
+               spin_lock(&mm_unpinned_lock);
+               list_del(&mm->context.unpinned);
+               spin_unlock(&mm_unpinned_lock);
+       }
+}
+
+static int read_ldt(void __user * ptr, unsigned long bytecount)
+{
+       int err;
+       unsigned long size;
+       struct mm_struct * mm = current->mm;
+
+       if (!mm->context.size)
+               return 0;
+       if (bytecount > LDT_ENTRY_SIZE*LDT_ENTRIES)
+               bytecount = LDT_ENTRY_SIZE*LDT_ENTRIES;
+
+       down(&mm->context.sem);
+       size = mm->context.size*LDT_ENTRY_SIZE;
+       if (size > bytecount)
+               size = bytecount;
+
+       err = 0;
+       if (copy_to_user(ptr, mm->context.ldt, size))
+               err = -EFAULT;
+       up(&mm->context.sem);
+       if (err < 0)
+               goto error_return;
+       if (size != bytecount) {
+               /* zero-fill the rest */
+               if (clear_user(ptr+size, bytecount-size) != 0) {
+                       err = -EFAULT;
+                       goto error_return;
+               }
+       }
+       return bytecount;
+error_return:
+       return err;
+}
+
+static int read_default_ldt(void __user * ptr, unsigned long bytecount)
+{
+       /* Arbitrary number */ 
+       /* x86-64 default LDT is all zeros */
+       if (bytecount > 128) 
+               bytecount = 128;        
+       if (clear_user(ptr, bytecount))
+               return -EFAULT;
+       return bytecount; 
+}
+
+static int write_ldt(void __user * ptr, unsigned long bytecount, int oldmode)
+{
+       struct task_struct *me = current;
+       struct mm_struct * mm = me->mm;
+       __u32 entry_1, entry_2, *lp;
+       unsigned long mach_lp;
+       int error;
+       struct user_desc ldt_info;
+
+       error = -EINVAL;
+
+       if (bytecount != sizeof(ldt_info))
+               goto out;
+       error = -EFAULT;        
+       if (copy_from_user(&ldt_info, ptr, bytecount))
+               goto out;
+
+       error = -EINVAL;
+       if (ldt_info.entry_number >= LDT_ENTRIES)
+               goto out;
+       if (ldt_info.contents == 3) {
+               if (oldmode)
+                       goto out;
+               if (ldt_info.seg_not_present == 0)
+                       goto out;
+       }
+
+       down(&mm->context.sem);
+       if (ldt_info.entry_number >= (unsigned)mm->context.size) {
+               error = alloc_ldt(&current->mm->context, ldt_info.entry_number+1, 1);
+               if (error < 0)
+                       goto out_unlock;
+       }
+
+       lp = (__u32 *) ((ldt_info.entry_number << 3) + (char *) mm->context.ldt);
+       mach_lp = arbitrary_virt_to_machine(lp);
+
+       /* Allow LDTs to be cleared by the user. */
+       if (ldt_info.base_addr == 0 && ldt_info.limit == 0) {
+               if (oldmode || LDT_empty(&ldt_info)) {
+                       entry_1 = 0;
+                       entry_2 = 0;
+                       goto install;
+               }
+       }
+
+       entry_1 = LDT_entry_a(&ldt_info);
+       entry_2 = LDT_entry_b(&ldt_info);
+       if (oldmode)
+               entry_2 &= ~(1 << 20);
+
+       /* Install the new entry ...  */
+install:
+       error = HYPERVISOR_update_descriptor(mach_lp, (unsigned long)((entry_1 | (unsigned long) entry_2 << 32)));
+
+out_unlock:
+       up(&mm->context.sem);
+out:
+       return error;
+}
+
+asmlinkage int sys_modify_ldt(int func, void __user *ptr, unsigned long bytecount)
+{
+       int ret = -ENOSYS;
+
+       switch (func) {
+       case 0:
+               ret = read_ldt(ptr, bytecount);
+               break;
+       case 1:
+               ret = write_ldt(ptr, bytecount, 1);
+               break;
+       case 2:
+               ret = read_default_ldt(ptr, bytecount);
+               break;
+       case 0x11:
+               ret = write_ldt(ptr, bytecount, 0);
+               break;
+       }
+       return ret;
+}