xen64: set up userspace syscall patch
authorJeremy Fitzhardinge <jeremy@goop.org>
Tue, 8 Jul 2008 22:07:15 +0000 (15:07 -0700)
committerIngo Molnar <mingo@elte.hu>
Wed, 16 Jul 2008 09:06:06 +0000 (11:06 +0200)
64-bit userspace expects the vdso to be mapped at a specific fixed
address, which happens to be in the middle of the kernel address
space.  Because we have split user and kernel pagetables, we need to
make special arrangements for the vsyscall mapping to appear in the
kernel part of the user pagetable.

Signed-off-by: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
Cc: Stephen Tweedie <sct@redhat.com>
Cc: Eduardo Habkost <ehabkost@redhat.com>
Cc: Mark McLoughlin <markmc@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
arch/x86/xen/enlighten.c

index 87d36044054d7f90976bb6c53ef0d45adc75dacf..f64b8729cd0719891f4c8c3e6c4f01e77c385943 100644 (file)
@@ -56,6 +56,18 @@ EXPORT_SYMBOL_GPL(hypercall_page);
 DEFINE_PER_CPU(struct vcpu_info *, xen_vcpu);
 DEFINE_PER_CPU(struct vcpu_info, xen_vcpu_info);
 
+/*
+ * Identity map, in addition to plain kernel map.  This needs to be
+ * large enough to allocate page table pages to allocate the rest.
+ * Each page can map 2MB.
+ */
+static pte_t level1_ident_pgt[PTRS_PER_PTE * 4] __page_aligned_bss;
+
+#ifdef CONFIG_X86_64
+/* l3 pud for userspace vsyscall mapping */
+static pud_t level3_user_vsyscall[PTRS_PER_PUD] __page_aligned_bss;
+#endif /* CONFIG_X86_64 */
+
 /*
  * Note about cr3 (pagetable base) values:
  *
@@ -831,12 +843,20 @@ static int xen_pgd_alloc(struct mm_struct *mm)
 #ifdef CONFIG_X86_64
        {
                struct page *page = virt_to_page(pgd);
+               pgd_t *user_pgd;
 
                BUG_ON(page->private != 0);
 
-               page->private = __get_free_page(GFP_KERNEL | __GFP_ZERO);
-               if (page->private == 0)
-                       ret = -ENOMEM;
+               ret = -ENOMEM;
+
+               user_pgd = (pgd_t *)__get_free_page(GFP_KERNEL | __GFP_ZERO);
+               page->private = (unsigned long)user_pgd;
+
+               if (user_pgd != NULL) {
+                       user_pgd[pgd_index(VSYSCALL_START)] =
+                               __pgd(__pa(level3_user_vsyscall) | _PAGE_TABLE);
+                       ret = 0;
+               }
 
                BUG_ON(PagePinned(virt_to_page(xen_get_user_pgd(pgd))));
        }
@@ -977,6 +997,9 @@ static __init void xen_post_allocator_init(void)
        pv_mmu_ops.release_pud = xen_release_pud;
 #endif
 
+#ifdef CONFIG_X86_64
+       SetPagePinned(virt_to_page(level3_user_vsyscall));
+#endif
        xen_mark_init_mm_pinned();
 }
 
@@ -1088,6 +1111,15 @@ static void xen_set_fixmap(unsigned idx, unsigned long phys, pgprot_t prot)
        }
 
        __native_set_fixmap(idx, pte);
+
+#ifdef CONFIG_X86_64
+       /* Replicate changes to map the vsyscall page into the user
+          pagetable vsyscall mapping. */
+       if (idx >= VSYSCALL_LAST_PAGE && idx <= VSYSCALL_FIRST_PAGE) {
+               unsigned long vaddr = __fix_to_virt(idx);
+               set_pte_vaddr_pud(level3_user_vsyscall, vaddr, pte);
+       }
+#endif
 }
 
 static const struct pv_info xen_info __initdata = {
@@ -1427,13 +1459,6 @@ static void set_page_prot(void *addr, pgprot_t prot)
                BUG();
 }
 
-/*
- * Identity map, in addition to plain kernel map.  This needs to be
- * large enough to allocate page table pages to allocate the rest.
- * Each page can map 2MB.
- */
-static pte_t level1_ident_pgt[PTRS_PER_PTE * 4] __page_aligned_bss;
-
 static __init void xen_map_identity_early(pmd_t *pmd, unsigned long max_pfn)
 {
        unsigned pmdidx, pteidx;
@@ -1533,6 +1558,7 @@ static __init pgd_t *xen_setup_kernel_pagetable(pgd_t *pgd, unsigned long max_pf
        set_page_prot(init_level4_pgt, PAGE_KERNEL_RO);
        set_page_prot(level3_ident_pgt, PAGE_KERNEL_RO);
        set_page_prot(level3_kernel_pgt, PAGE_KERNEL_RO);
+       set_page_prot(level3_user_vsyscall, PAGE_KERNEL_RO);
        set_page_prot(level2_kernel_pgt, PAGE_KERNEL_RO);
        set_page_prot(level2_fixmap_pgt, PAGE_KERNEL_RO);