x86, 64bit: #PF handler set page to cover only 2M per #PF

We only map a single 2 MiB page per #PF, even though we should be able
to do this a full gigabyte at a time with no additional memory cost.
This is a workaround for a broken AMD reference BIOS (and its
derivatives in shipping system) which maps a large chunk of memory as
WB in the MTRR system but will #MC if the processor wanders off and
tries to prefetch that memory, which can happen any time the memory is
mapped in the TLB.

Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Link: http://lkml.kernel.org/r/1359058816-7615-13-git-send-email-yinghai@kernel.org
Cc: Alexander Duyck <alexander.h.duyck@intel.com>
[ hpa: rewrote the patch description ]
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>

diff --git a/arch/x86/kernel/head64.c b/arch/x86/kernel/head64.c
index f57df05..816fc85 100644 (file)
--- a/arch/x86/kernel/head64.c
+++ b/arch/x86/kernel/head64.c
@@ -53,15 +53,15 @@ int __init early_make_pgtable(unsigned long address)
        unsigned long physaddr = address - __PAGE_OFFSET;
        unsigned long i;
        pgdval_t pgd, *pgd_p;
-       pudval_t *pud_p;
+       pudval_t pud, *pud_p;
        pmdval_t pmd, *pmd_p;
 
        /* Invalid address or early pgt is done ?  */
        if (physaddr >= MAXMEM || read_cr3() != __pa(early_level4_pgt))
                return -1;
 
-       i = (address >> PGDIR_SHIFT) & (PTRS_PER_PGD - 1);
-       pgd_p = &early_level4_pgt[i].pgd;
+again:
+       pgd_p = &early_level4_pgt[pgd_index(address)].pgd;
        pgd = *pgd_p;
 
        /*
@@ -69,29 +69,37 @@ int __init early_make_pgtable(unsigned long address)
         * critical -- __PAGE_OFFSET would point us back into the dynamic
         * range and we might end up looping forever...
         */
-       if (pgd && next_early_pgt < EARLY_DYNAMIC_PAGE_TABLES) {
+       if (pgd)
                pud_p = (pudval_t *)((pgd & PTE_PFN_MASK) + __START_KERNEL_map - phys_base);
-       } else {
-               if (next_early_pgt >= EARLY_DYNAMIC_PAGE_TABLES-1)
+       else {
+               if (next_early_pgt >= EARLY_DYNAMIC_PAGE_TABLES) {
                        reset_early_page_tables();
+                       goto again;
+               }
 
                pud_p = (pudval_t *)early_dynamic_pgts[next_early_pgt++];
                for (i = 0; i < PTRS_PER_PUD; i++)
                        pud_p[i] = 0;
-
                *pgd_p = (pgdval_t)pud_p - __START_KERNEL_map + phys_base + _KERNPG_TABLE;
        }
-       i = (address >> PUD_SHIFT) & (PTRS_PER_PUD - 1);
-       pud_p += i;
-
-       pmd_p = (pmdval_t *)early_dynamic_pgts[next_early_pgt++];
-       pmd = (physaddr & PUD_MASK) + (__PAGE_KERNEL_LARGE & ~_PAGE_GLOBAL);
-       for (i = 0; i < PTRS_PER_PMD; i++) {
-               pmd_p[i] = pmd;
-               pmd += PMD_SIZE;
-       }
+       pud_p += pud_index(address);
+       pud = *pud_p;
 
-       *pud_p = (pudval_t)pmd_p - __START_KERNEL_map + phys_base + _KERNPG_TABLE;
+       if (pud)
+               pmd_p = (pmdval_t *)((pud & PTE_PFN_MASK) + __START_KERNEL_map - phys_base);
+       else {
+               if (next_early_pgt >= EARLY_DYNAMIC_PAGE_TABLES) {
+                       reset_early_page_tables();
+                       goto again;
+               }
+
+               pmd_p = (pmdval_t *)early_dynamic_pgts[next_early_pgt++];
+               for (i = 0; i < PTRS_PER_PMD; i++)
+                       pmd_p[i] = 0;
+               *pud_p = (pudval_t)pmd_p - __START_KERNEL_map + phys_base + _KERNPG_TABLE;
+       }
+       pmd = (physaddr & PMD_MASK) + (__PAGE_KERNEL_LARGE & ~_PAGE_GLOBAL);
+       pmd_p[pmd_index(address)] = pmd;
 
        return 0;
 }