1 files changed, 42 insertions, 3 deletions

diff --git a/arch/powerpc/mm/pgtable_64.c b/arch/powerpc/mm/pgtable_64.c
index 3124a20d0fab..347106080bb1 100644
--- a/arch/powerpc/mm/pgtable_64.c
+++ b/arch/powerpc/mm/pgtable_64.c
@@ -88,7 +88,7 @@ static __ref void *early_alloc_pgtable(unsigned long size)
  * map_kernel_page adds an entry to the ioremap page table
  * and adds an entry to the HPT, possibly bolting it
  */
-int map_kernel_page(unsigned long ea, unsigned long pa, int flags)
+int map_kernel_page(unsigned long ea, unsigned long pa, unsigned long flags)
 {
 	pgd_t *pgdp;
 	pud_t *pudp;
@@ -403,7 +403,7 @@ static pte_t *__alloc_for_cache(struct mm_struct *mm, int kernel)
 	 * count.
 	 */
 	if (likely(!mm->context.pte_frag)) {
-		atomic_set(&page->_count, PTE_FRAG_NR);
+		set_page_count(page, PTE_FRAG_NR);
 		mm->context.pte_frag = ret + PTE_FRAG_SIZE;
 	}
 	spin_unlock(&mm->page_table_lock);
@@ -646,6 +646,28 @@ pgtable_t pgtable_trans_huge_withdraw(struct mm_struct *mm, pmd_t *pmdp)
 	return pgtable;
 }
 
+void pmdp_huge_split_prepare(struct vm_area_struct *vma,
+			     unsigned long address, pmd_t *pmdp)
+{
+	VM_BUG_ON(address & ~HPAGE_PMD_MASK);
+	VM_BUG_ON(REGION_ID(address) != USER_REGION_ID);
+
+	/*
+	 * We can't mark the pmd none here, because that will cause a race
+	 * against exit_mmap. We need to continue mark pmd TRANS HUGE, while
+	 * we spilt, but at the same time we wan't rest of the ppc64 code
+	 * not to insert hash pte on this, because we will be modifying
+	 * the deposited pgtable in the caller of this function. Hence
+	 * clear the _PAGE_USER so that we move the fault handling to
+	 * higher level function and that will serialize against ptl.
+	 * We need to flush existing hash pte entries here even though,
+	 * the translation is still valid, because we will withdraw
+	 * pgtable_t after this.
+	 */
+	pmd_hugepage_update(vma->vm_mm, address, pmdp, _PAGE_USER, 0);
+}
+
+
 /*
  * set a new huge pmd. We should not be called for updating
  * an existing pmd entry. That should go via pmd_hugepage_update.
@@ -663,10 +685,20 @@ void set_pmd_at(struct mm_struct *mm, unsigned long addr,
 	return set_pte_at(mm, addr, pmdp_ptep(pmdp), pmd_pte(pmd));
 }
 
+/*
+ * We use this to invalidate a pmdp entry before switching from a
+ * hugepte to regular pmd entry.
+ */
 void pmdp_invalidate(struct vm_area_struct *vma, unsigned long address,
 		     pmd_t *pmdp)
 {
 	pmd_hugepage_update(vma->vm_mm, address, pmdp, _PAGE_PRESENT, 0);
+
+	/*
+	 * This ensures that generic code that rely on IRQ disabling
+	 * to prevent a parallel THP split work as expected.
+	 */
+	kick_all_cpus_sync();
 }
 
 /*
@@ -717,7 +749,7 @@ pmd_t pfn_pmd(unsigned long pfn, pgprot_t pgprot)
 {
 	unsigned long pmdv;
 
-	pmdv = pfn << PTE_RPN_SHIFT;
+	pmdv = (pfn << PTE_RPN_SHIFT) & PTE_RPN_MASK;
 	return pmd_set_protbits(__pmd(pmdv), pgprot);
 }
 
@@ -785,6 +817,13 @@ pmd_t pmdp_huge_get_and_clear(struct mm_struct *mm,
 
 int has_transparent_hugepage(void)
 {
+
+	BUILD_BUG_ON_MSG((PMD_SHIFT - PAGE_SHIFT) >= MAX_ORDER,
+		"hugepages can't be allocated by the buddy allocator");
+
+	BUILD_BUG_ON_MSG((PMD_SHIFT - PAGE_SHIFT) < 2,
+			 "We need more than 2 pages to do deferred thp split");
+
 	if (!mmu_has_feature(MMU_FTR_16M_PAGE))
 		return 0;
 	/*