x86/pgtable/pae: Use separate kernel PMDs for user page-table
authorJoerg Roedel <jroedel@suse.de>
Wed, 18 Jul 2018 09:41:09 +0000 (11:41 +0200)
committerThomas Gleixner <tglx@linutronix.de>
Thu, 19 Jul 2018 23:11:46 +0000 (01:11 +0200)
When PTI is enabled, separate kernel PMDs in the user page-table are
required to map the per-process LDT for user-space.

Signed-off-by: Joerg Roedel <jroedel@suse.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Tested-by: Pavel Machek <pavel@ucw.cz>
Cc: "H . Peter Anvin" <hpa@zytor.com>
Cc: linux-mm@kvack.org
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Dave Hansen <dave.hansen@intel.com>
Cc: Josh Poimboeuf <jpoimboe@redhat.com>
Cc: Juergen Gross <jgross@suse.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Jiri Kosina <jkosina@suse.cz>
Cc: Boris Ostrovsky <boris.ostrovsky@oracle.com>
Cc: Brian Gerst <brgerst@gmail.com>
Cc: David Laight <David.Laight@aculab.com>
Cc: Denys Vlasenko <dvlasenk@redhat.com>
Cc: Eduardo Valentin <eduval@amazon.com>
Cc: Greg KH <gregkh@linuxfoundation.org>
Cc: Will Deacon <will.deacon@arm.com>
Cc: aliguori@amazon.com
Cc: daniel.gruss@iaik.tugraz.at
Cc: hughd@google.com
Cc: keescook@google.com
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Waiman Long <llong@redhat.com>
Cc: "David H . Gutteridge" <dhgutteridge@sympatico.ca>
Cc: joro@8bytes.org
Link: https://lkml.kernel.org/r/1531906876-13451-33-git-send-email-joro@8bytes.org
arch/x86/mm/pgtable.c

index db6fb77..8e4e63d 100644 (file)
@@ -182,6 +182,14 @@ static void pgd_dtor(pgd_t *pgd)
  */
 #define PREALLOCATED_PMDS      UNSHARED_PTRS_PER_PGD
 
+/*
+ * We allocate separate PMDs for the kernel part of the user page-table
+ * when PTI is enabled. We need them to map the per-process LDT into the
+ * user-space page-table.
+ */
+#define PREALLOCATED_USER_PMDS  (static_cpu_has(X86_FEATURE_PTI) ? \
+                                       KERNEL_PGD_PTRS : 0)
+
 void pud_populate(struct mm_struct *mm, pud_t *pudp, pmd_t *pmd)
 {
        paravirt_alloc_pmd(mm, __pa(pmd) >> PAGE_SHIFT);
@@ -202,14 +210,14 @@ void pud_populate(struct mm_struct *mm, pud_t *pudp, pmd_t *pmd)
 
 /* No need to prepopulate any pagetable entries in non-PAE modes. */
 #define PREALLOCATED_PMDS      0
-
+#define PREALLOCATED_USER_PMDS  0
 #endif /* CONFIG_X86_PAE */
 
-static void free_pmds(struct mm_struct *mm, pmd_t *pmds[])
+static void free_pmds(struct mm_struct *mm, pmd_t *pmds[], int count)
 {
        int i;
 
-       for(i = 0; i < PREALLOCATED_PMDS; i++)
+       for (i = 0; i < count; i++)
                if (pmds[i]) {
                        pgtable_pmd_page_dtor(virt_to_page(pmds[i]));
                        free_page((unsigned long)pmds[i]);
@@ -217,7 +225,7 @@ static void free_pmds(struct mm_struct *mm, pmd_t *pmds[])
                }
 }
 
-static int preallocate_pmds(struct mm_struct *mm, pmd_t *pmds[])
+static int preallocate_pmds(struct mm_struct *mm, pmd_t *pmds[], int count)
 {
        int i;
        bool failed = false;
@@ -226,7 +234,7 @@ static int preallocate_pmds(struct mm_struct *mm, pmd_t *pmds[])
        if (mm == &init_mm)
                gfp &= ~__GFP_ACCOUNT;
 
-       for(i = 0; i < PREALLOCATED_PMDS; i++) {
+       for (i = 0; i < count; i++) {
                pmd_t *pmd = (pmd_t *)__get_free_page(gfp);
                if (!pmd)
                        failed = true;
@@ -241,7 +249,7 @@ static int preallocate_pmds(struct mm_struct *mm, pmd_t *pmds[])
        }
 
        if (failed) {
-               free_pmds(mm, pmds);
+               free_pmds(mm, pmds, count);
                return -ENOMEM;
        }
 
@@ -254,23 +262,38 @@ static int preallocate_pmds(struct mm_struct *mm, pmd_t *pmds[])
  * preallocate which never got a corresponding vma will need to be
  * freed manually.
  */
+static void mop_up_one_pmd(struct mm_struct *mm, pgd_t *pgdp)
+{
+       pgd_t pgd = *pgdp;
+
+       if (pgd_val(pgd) != 0) {
+               pmd_t *pmd = (pmd_t *)pgd_page_vaddr(pgd);
+
+               *pgdp = native_make_pgd(0);
+
+               paravirt_release_pmd(pgd_val(pgd) >> PAGE_SHIFT);
+               pmd_free(mm, pmd);
+               mm_dec_nr_pmds(mm);
+       }
+}
+
 static void pgd_mop_up_pmds(struct mm_struct *mm, pgd_t *pgdp)
 {
        int i;
 
-       for(i = 0; i < PREALLOCATED_PMDS; i++) {
-               pgd_t pgd = pgdp[i];
+       for (i = 0; i < PREALLOCATED_PMDS; i++)
+               mop_up_one_pmd(mm, &pgdp[i]);
 
-               if (pgd_val(pgd) != 0) {
-                       pmd_t *pmd = (pmd_t *)pgd_page_vaddr(pgd);
+#ifdef CONFIG_PAGE_TABLE_ISOLATION
 
-                       pgdp[i] = native_make_pgd(0);
+       if (!static_cpu_has(X86_FEATURE_PTI))
+               return;
 
-                       paravirt_release_pmd(pgd_val(pgd) >> PAGE_SHIFT);
-                       pmd_free(mm, pmd);
-                       mm_dec_nr_pmds(mm);
-               }
-       }
+       pgdp = kernel_to_user_pgdp(pgdp);
+
+       for (i = 0; i < PREALLOCATED_USER_PMDS; i++)
+               mop_up_one_pmd(mm, &pgdp[i + KERNEL_PGD_BOUNDARY]);
+#endif
 }
 
 static void pgd_prepopulate_pmd(struct mm_struct *mm, pgd_t *pgd, pmd_t *pmds[])
@@ -296,6 +319,38 @@ static void pgd_prepopulate_pmd(struct mm_struct *mm, pgd_t *pgd, pmd_t *pmds[])
        }
 }
 
+#ifdef CONFIG_PAGE_TABLE_ISOLATION
+static void pgd_prepopulate_user_pmd(struct mm_struct *mm,
+                                    pgd_t *k_pgd, pmd_t *pmds[])
+{
+       pgd_t *s_pgd = kernel_to_user_pgdp(swapper_pg_dir);
+       pgd_t *u_pgd = kernel_to_user_pgdp(k_pgd);
+       p4d_t *u_p4d;
+       pud_t *u_pud;
+       int i;
+
+       u_p4d = p4d_offset(u_pgd, 0);
+       u_pud = pud_offset(u_p4d, 0);
+
+       s_pgd += KERNEL_PGD_BOUNDARY;
+       u_pud += KERNEL_PGD_BOUNDARY;
+
+       for (i = 0; i < PREALLOCATED_USER_PMDS; i++, u_pud++, s_pgd++) {
+               pmd_t *pmd = pmds[i];
+
+               memcpy(pmd, (pmd_t *)pgd_page_vaddr(*s_pgd),
+                      sizeof(pmd_t) * PTRS_PER_PMD);
+
+               pud_populate(mm, u_pud, pmd);
+       }
+
+}
+#else
+static void pgd_prepopulate_user_pmd(struct mm_struct *mm,
+                                    pgd_t *k_pgd, pmd_t *pmds[])
+{
+}
+#endif
 /*
  * Xen paravirt assumes pgd table should be in one page. 64 bit kernel also
  * assumes that pgd should be in one page.
@@ -376,6 +431,7 @@ static inline void _pgd_free(pgd_t *pgd)
 pgd_t *pgd_alloc(struct mm_struct *mm)
 {
        pgd_t *pgd;
+       pmd_t *u_pmds[PREALLOCATED_USER_PMDS];
        pmd_t *pmds[PREALLOCATED_PMDS];
 
        pgd = _pgd_alloc();
@@ -385,12 +441,15 @@ pgd_t *pgd_alloc(struct mm_struct *mm)
 
        mm->pgd = pgd;
 
-       if (preallocate_pmds(mm, pmds) != 0)
+       if (preallocate_pmds(mm, pmds, PREALLOCATED_PMDS) != 0)
                goto out_free_pgd;
 
-       if (paravirt_pgd_alloc(mm) != 0)
+       if (preallocate_pmds(mm, u_pmds, PREALLOCATED_USER_PMDS) != 0)
                goto out_free_pmds;
 
+       if (paravirt_pgd_alloc(mm) != 0)
+               goto out_free_user_pmds;
+
        /*
         * Make sure that pre-populating the pmds is atomic with
         * respect to anything walking the pgd_list, so that they
@@ -400,13 +459,16 @@ pgd_t *pgd_alloc(struct mm_struct *mm)
 
        pgd_ctor(mm, pgd);
        pgd_prepopulate_pmd(mm, pgd, pmds);
+       pgd_prepopulate_user_pmd(mm, pgd, u_pmds);
 
        spin_unlock(&pgd_lock);
 
        return pgd;
 
+out_free_user_pmds:
+       free_pmds(mm, u_pmds, PREALLOCATED_USER_PMDS);
 out_free_pmds:
-       free_pmds(mm, pmds);
+       free_pmds(mm, pmds, PREALLOCATED_PMDS);
 out_free_pgd:
        _pgd_free(pgd);
 out: