x86: Respect PAT bit when copying pte values between large and normal pages
authorJuergen Gross <jgross@suse.com>
Mon, 3 Nov 2014 13:02:02 +0000 (14:02 +0100)
committerThomas Gleixner <tglx@linutronix.de>
Sun, 16 Nov 2014 10:04:26 +0000 (11:04 +0100)
The PAT bit in the ptes is not moved to the correct position when
copying page protection attributes between entries of different sized
pages. Translate the ptes according to their page size.

Based-on-patch-by: Stefan Bader <stefan.bader@canonical.com>
Signed-off-by: Juergen Gross <jgross@suse.com>
Reviewed-by: Thomas Gleixner <tglx@linutronix.de>
Cc: stefan.bader@canonical.com
Cc: xen-devel@lists.xensource.com
Cc: konrad.wilk@oracle.com
Cc: ville.syrjala@linux.intel.com
Cc: david.vrabel@citrix.com
Cc: jbeulich@suse.com
Cc: toshi.kani@hp.com
Cc: plagnioj@jcrosoft.com
Cc: tomi.valkeinen@ti.com
Cc: bhelgaas@google.com
Link: http://lkml.kernel.org/r/1415019724-4317-17-git-send-email-jgross@suse.com
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
arch/x86/mm/pageattr.c

index de807c9daad1e1b0c9fc4f98bd54f962048750af..6c8e3fdaf077a1c675374996b8bef509204b9e8f 100644 (file)
@@ -485,13 +485,22 @@ try_preserve_large_page(pte_t *kpte, unsigned long address,
 
        /*
         * We are safe now. Check whether the new pgprot is the same:
+        * Convert protection attributes to 4k-format, as cpa->mask* are set
+        * up accordingly.
         */
        old_pte = *kpte;
-       old_prot = req_prot = pte_pgprot(old_pte);
+       old_prot = req_prot = pgprot_large_2_4k(pte_pgprot(old_pte));
 
        pgprot_val(req_prot) &= ~pgprot_val(cpa->mask_clr);
        pgprot_val(req_prot) |= pgprot_val(cpa->mask_set);
 
+       /*
+        * req_prot is in format of 4k pages. It must be converted to large
+        * page format: the caching mode includes the PAT bit located at
+        * different bit positions in the two formats.
+        */
+       req_prot = pgprot_4k_2_large(req_prot);
+
        /*
         * Set the PSE and GLOBAL flags only if the PRESENT flag is
         * set otherwise pmd_present/pmd_huge will return true even on
@@ -585,13 +594,10 @@ __split_large_page(struct cpa_data *cpa, pte_t *kpte, unsigned long address,
 
        paravirt_alloc_pte(&init_mm, page_to_pfn(base));
        ref_prot = pte_pgprot(pte_clrhuge(*kpte));
-       /*
-        * If we ever want to utilize the PAT bit, we need to
-        * update this function to make sure it's converted from
-        * bit 12 to bit 7 when we cross from the 2MB level to
-        * the 4K level:
-        */
-       WARN_ON_ONCE(pgprot_val(ref_prot) & _PAGE_PAT_LARGE);
+
+       /* promote PAT bit to correct position */
+       if (level == PG_LEVEL_2M)
+               ref_prot = pgprot_large_2_4k(ref_prot);
 
 #ifdef CONFIG_X86_64
        if (level == PG_LEVEL_1G) {
@@ -879,6 +885,7 @@ static int populate_pmd(struct cpa_data *cpa,
 {
        unsigned int cur_pages = 0;
        pmd_t *pmd;
+       pgprot_t pmd_pgprot;
 
        /*
         * Not on a 2M boundary?
@@ -910,6 +917,8 @@ static int populate_pmd(struct cpa_data *cpa,
        if (num_pages == cur_pages)
                return cur_pages;
 
+       pmd_pgprot = pgprot_4k_2_large(pgprot);
+
        while (end - start >= PMD_SIZE) {
 
                /*
@@ -921,7 +930,8 @@ static int populate_pmd(struct cpa_data *cpa,
 
                pmd = pmd_offset(pud, start);
 
-               set_pmd(pmd, __pmd(cpa->pfn | _PAGE_PSE | massage_pgprot(pgprot)));
+               set_pmd(pmd, __pmd(cpa->pfn | _PAGE_PSE |
+                                  massage_pgprot(pmd_pgprot)));
 
                start     += PMD_SIZE;
                cpa->pfn  += PMD_SIZE;
@@ -949,6 +959,7 @@ static int populate_pud(struct cpa_data *cpa, unsigned long start, pgd_t *pgd,
        pud_t *pud;
        unsigned long end;
        int cur_pages = 0;
+       pgprot_t pud_pgprot;
 
        end = start + (cpa->numpages << PAGE_SHIFT);
 
@@ -986,12 +997,14 @@ static int populate_pud(struct cpa_data *cpa, unsigned long start, pgd_t *pgd,
                return cur_pages;
 
        pud = pud_offset(pgd, start);
+       pud_pgprot = pgprot_4k_2_large(pgprot);
 
        /*
         * Map everything starting from the Gb boundary, possibly with 1G pages
         */
        while (end - start >= PUD_SIZE) {
-               set_pud(pud, __pud(cpa->pfn | _PAGE_PSE | massage_pgprot(pgprot)));
+               set_pud(pud, __pud(cpa->pfn | _PAGE_PSE |
+                                  massage_pgprot(pud_pgprot)));
 
                start     += PUD_SIZE;
                cpa->pfn  += PUD_SIZE;