powerpc: Reduce PTE table memory wastage

We allocate one page for the last level of linux page table. With THP and
large page size of 16MB, that would mean we are wasting large part
of that page. To map 16MB area, we only need a PTE space of 2K with 64K
page size. This patch reduce the space wastage by sharing the page
allocated for the last level of linux page table with multiple pmd
entries. We call these smaller chunks PTE page fragments and allocated
page, PTE page.

In order to support systems which doesn't have 64K HPTE support, we also
add another 2K to PTE page fragment. The second half of the PTE fragments
is used for storing slot and secondary bit information of an HPTE. With this
we now have a 4K PTE fragment.

We use a simple approach to share the PTE page. On allocation, we bump the
PTE page refcount to 16 and share the PTE page with the next 16 pte alloc
request. This should help in the node locality of the PTE page fragment,
assuming that the immediate pte alloc request will mostly come from the
same NUMA node. We don't try to reuse the freed PTE page fragment. Hence
we could be waisting some space.

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Acked-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>

4 files changed, 37 insertions, 57 deletions

diff --git a/arch/powerpc/include/asm/mmu-book3e.h b/arch/powerpc/include/asm/mmu-book3e.h
index 99d43e0c1e4a..8bd560c5fe84 100644
--- a/arch/powerpc/include/asm/mmu-book3e.h
+++ b/arch/powerpc/include/asm/mmu-book3e.h
@@ -231,6 +231,10 @@ typedef struct {
         u64 high_slices_psize;  /* 4 bits per slice for now */
         u16 user_psize;         /* page size index */
 #endif
+#ifdef CONFIG_PPC_64K_PAGES
+        /* for 4K PTE fragment support */
+        void *pte_frag;
+#endif
 } mm_context_t;
 
 /* Page size definitions, common between 32 and 64-bit
diff --git a/arch/powerpc/include/asm/mmu-hash64.h b/arch/powerpc/include/asm/mmu-hash64.h
index 05895cff1345..de9e577f329c 100644
--- a/arch/powerpc/include/asm/mmu-hash64.h
+++ b/arch/powerpc/include/asm/mmu-hash64.h
@@ -516,6 +516,10 @@ typedef struct {
         unsigned long acop;     /* mask of enabled coprocessor types */
         unsigned int cop_pid;   /* pid value used with coprocessors */
 #endif /* CONFIG_PPC_ICSWX */
+#ifdef CONFIG_PPC_64K_PAGES
+        /* for 4K PTE fragment support */
+        void *pte_frag;
+#endif
 } mm_context_t;
 
 
diff --git a/arch/powerpc/include/asm/page.h b/arch/powerpc/include/asm/page.h
index 711e83a79e3d..988c812aab5b 100644
--- a/arch/powerpc/include/asm/page.h
+++ b/arch/powerpc/include/asm/page.h
@@ -393,7 +393,11 @@ void arch_free_page(struct page *page, int order);
 
 struct vm_area_struct;
 
+#ifdef CONFIG_PPC_64K_PAGES
+typedef pte_t *pgtable_t;
+#else
 typedef struct page *pgtable_t;
+#endif
 
 #include <asm-generic/memory_model.h>
 #endif /* __ASSEMBLY__ */
diff --git a/arch/powerpc/include/asm/pgalloc-64.h b/arch/powerpc/include/asm/pgalloc-64.h
index d39012352f94..91acb12bac92 100644
--- a/arch/powerpc/include/asm/pgalloc-64.h
+++ b/arch/powerpc/include/asm/pgalloc-64.h
@@ -152,6 +152,23 @@ static inline void __pte_free_tlb(struct mmu_gather *tlb, pgtable_t table,
 }
 
 #else /* if CONFIG_PPC_64K_PAGES */
+/*
+ * we support 16 fragments per PTE page.
+ */
+#define PTE_FRAG_NR     16
+/*
+ * We use a 2K PTE page fragment and another 2K for storing
+ * real_pte_t hash index
+ */
+#define PTE_FRAG_SIZE_SHIFT  12
+#define PTE_FRAG_SIZE (2 * PTRS_PER_PTE * sizeof(pte_t))
+
+extern pte_t *page_table_alloc(struct mm_struct *, unsigned long, int);
+extern void page_table_free(struct mm_struct *, unsigned long *, int);
+extern void pgtable_free_tlb(struct mmu_gather *tlb, void *table, int shift);
+#ifdef CONFIG_SMP
+extern void __tlb_remove_table(void *_table);
+#endif
 
 #define pud_populate(mm, pud, pmd)      pud_set(pud, (unsigned long)pmd)
 
@@ -164,90 +181,42 @@ static inline void pmd_populate_kernel(struct mm_struct *mm, pmd_t *pmd,
 static inline void pmd_populate(struct mm_struct *mm, pmd_t *pmd,
                                 pgtable_t pte_page)
 {
-        pmd_populate_kernel(mm, pmd, page_address(pte_page));
+        pmd_set(pmd, (unsigned long)pte_page);
 }
 
 static inline pgtable_t pmd_pgtable(pmd_t pmd)
 {
-        return pmd_page(pmd);
+        return (pgtable_t)(pmd_val(pmd) & -sizeof(pte_t)*PTRS_PER_PTE);
 }
 
 static inline pte_t *pte_alloc_one_kernel(struct mm_struct *mm,
                                           unsigned long address)
 {
-        return (pte_t *)__get_free_page(GFP_KERNEL | __GFP_REPEAT | __GFP_ZERO);
+        return (pte_t *)page_table_alloc(mm, address, 1);
 }
 
 static inline pgtable_t pte_alloc_one(struct mm_struct *mm,
-                                      unsigned long address)
+                                        unsigned long address)
 {
-        struct page *page;
-        pte_t *pte;
-
-        pte = pte_alloc_one_kernel(mm, address);
-        if (!pte)
-                return NULL;
-        page = virt_to_page(pte);
-        pgtable_page_ctor(page);
-        return page;
+        return (pgtable_t)page_table_alloc(mm, address, 0);
 }
 
 static inline void pte_free_kernel(struct mm_struct *mm, pte_t *pte)
 {
-        free_page((unsigned long)pte);
+        page_table_free(mm, (unsigned long *)pte, 1);
 }
 
 static inline void pte_free(struct mm_struct *mm, pgtable_t ptepage)
 {
-        pgtable_page_dtor(ptepage);
-        __free_page(ptepage);
+        page_table_free(mm, (unsigned long *)ptepage, 0);
 }
 
-static inline void pgtable_free(void *table, unsigned index_size)
-{
-        if (!index_size)
-                free_page((unsigned long)table);
-        else {
-                BUG_ON(index_size > MAX_PGTABLE_INDEX_SIZE);
-                kmem_cache_free(PGT_CACHE(index_size), table);
-        }
-}
-
-#ifdef CONFIG_SMP
-static inline void pgtable_free_tlb(struct mmu_gather *tlb,
-                                    void *table, int shift)
-{
-        unsigned long pgf = (unsigned long)table;
-        BUG_ON(shift > MAX_PGTABLE_INDEX_SIZE);
-        pgf |= shift;
-        tlb_remove_table(tlb, (void *)pgf);
-}
-
-static inline void __tlb_remove_table(void *_table)
-{
-        void *table = (void *)((unsigned long)_table & ~MAX_PGTABLE_INDEX_SIZE);
-        unsigned shift = (unsigned long)_table & MAX_PGTABLE_INDEX_SIZE;
-
-        pgtable_free(table, shift);
-}
-#else /* !CONFIG_SMP */
-static inline void pgtable_free_tlb(struct mmu_gather *tlb,
-                                    void *table, int shift)
-{
-        pgtable_free(table, shift);
-}
-#endif /* CONFIG_SMP */
-
 static inline void __pte_free_tlb(struct mmu_gather *tlb, pgtable_t table,
                                   unsigned long address)
 {
-        struct page *page = page_address(table);
-
         tlb_flush_pgtable(tlb, address);
-        pgtable_page_dtor(page);
-        pgtable_free_tlb(tlb, page, 0);
+        pgtable_free_tlb(tlb, table, 0);
 }
-
 #endif /* CONFIG_PPC_64K_PAGES */
 
 static inline pmd_t *pmd_alloc_one(struct mm_struct *mm, unsigned long addr)
@@ -261,7 +230,6 @@ static inline void pmd_free(struct mm_struct *mm, pmd_t *pmd)
         kmem_cache_free(PGT_CACHE(PMD_INDEX_SIZE), pmd);
 }
 
-
 #define __pmd_free_tlb(tlb, pmd, addr)                \
         pgtable_free_tlb(tlb, pmd, PMD_INDEX_SIZE)
 #ifndef CONFIG_PPC_64K_PAGES