MM: Make Page Tables Relcatable -- relocation code

diff -uprwNBb -X 2.6.25-rc8/Documentation/dontdiff /home/rossb/local/linux-2.6.25-rc8/arch/powerpc/mm/fault.c 2.6.25-rc8/arch/powerpc/mm/fault.c
@@ -299,6 +299,8 @@ good_area:
 		if (get_pteptr(mm, address, &ptep, &pmdp)) {
 			spinlock_t *ptl = pte_lockptr(mm, pmdp);
 			spin_lock(ptl);
+			delimbo_pte(&ptep, &ptl, &pmdp, mm, address);
+
 			if (pte_present(*ptep)) {
 				struct page *page = pte_page(*ptep);
 

diff -uprwNBb -X 2.6.25-rc8/Documentation/dontdiff /home/rossb/local/linux-2.6.25-rc8/arch/powerpc/mm/hugetlbpage.c 2.6.25-rc8/arch/powerpc/mm/hugetlbpage.c
@@ -73,6 +73,7 @@ static int __hugepte_alloc(struct mm_str
 		return -ENOMEM;
 
 	spin_lock(&mm->page_table_lock);
+	delimbo_hpd(&hpdp, mm, address);
 	if (!hugepd_none(*hpdp))
 		kmem_cache_free(huge_pgtable_cache, new);
 	else

diff -uprwNBb -X 2.6.25-rc8/Documentation/dontdiff /home/rossb/local/linux-2.6.25-rc8/arch/ppc/mm/fault.c 2.6.25-rc8/arch/ppc/mm/fault.c
@@ -219,6 +219,7 @@ good_area:
 		if (get_pteptr(mm, address, &ptep, &pmdp)) {
 			spinlock_t *ptl = pte_lockptr(mm, pmdp);
 			spin_lock(ptl);
+			delimbo_pte(&ptep, &ptl, &pmdp, mm, address);
 			if (pte_present(*ptep)) {
 				struct page *page = pte_page(*ptep);
 

diff -uprwNBb -X 2.6.25-rc8/Documentation/dontdiff /home/rossb/local/linux-2.6.25-rc8/arch/x86/mm/hugetlbpage.c 2.6.25-rc8/arch/x86/mm/hugetlbpage.c
@@ -88,6 +88,7 @@ static void huge_pmd_share(struct mm_str
 		goto out;
 
 	spin_lock(&mm->page_table_lock);
+	delimbo_pud(&pud, mm, addr);
 	if (pud_none(*pud))
 		pud_populate(mm, pud, (pmd_t *)((unsigned long)spte & PAGE_MASK));
 	else

diff -uprwNBb -X 2.6.25-rc8/Documentation/dontdiff /home/rossb/local/linux-2.6.25-rc8/include/asm-generic/pgalloc.h 2.6.25-rc8/include/asm-generic/pgalloc.h
@@ -0,0 +1,37 @@
+#ifndef _ASM_GENERIC_PGALLOC_H
+#define _ASM_GENERIC_PGALLOC_H
+
+
+
+/* Page Table Levels used for alloc_page_table. */
+#define PAGE_TABLE_PGD 0
+#define PAGE_TABLE_PUD 1
+#define PAGE_TABLE_PMD 2
+#define PAGE_TABLE_PTE 3
+
+static inline struct page *alloc_page_table_node(struct mm_struct *mm,
+						 unsigned long addr,
+						 int node,
+						 int page_table_level)
+{
+	switch (page_table_level) {
+	case PAGE_TABLE_PGD:
+		return virt_to_page(pgd_alloc_node(mm, node));
+
+	case PAGE_TABLE_PUD:
+		return virt_to_page(pud_alloc_one_node(mm, addr, node));
+
+	case PAGE_TABLE_PMD:
+		return virt_to_page(pmd_alloc_one_node(mm, addr, node));
+
+	case PAGE_TABLE_PTE:
+		return pte_alloc_one_node(mm, addr, node);
+
+	default:
+		BUG();
+		return NULL;
+	}
+}
+
+
+#endif /* _ASM_GENERIC_PGALLOC_H */

56 lines not on this page · raw message

205 lines not on this page · raw message

diff -uprwNBb -X 2.6.25-rc8/Documentation/dontdiff /home/rossb/local/linux-2.6.25-rc8/include/linux/migrate.h 2.6.25-rc8/include/linux/migrate.h
@@ -6,6 +6,10 @@
 #include <linux/pagemap.h>
 
 typedef struct page *new_page_t(struct page *, unsigned long private, int **);
+typedef struct page *new_page_table_t(struct mm_struct *,
+				      unsigned long addr,
+				      unsigned long private,
+				      int **, int page_table_level);
 
 #ifdef CONFIG_MIGRATION
 /* Check if a vma is migratable */

113 lines not on this page · raw message

diff -uprwNBb -X 2.6.25-rc8/Documentation/dontdiff /home/rossb/local/linux-2.6.25-rc8/include/linux/mm_types.h 2.6.25-rc8/include/linux/mm_types.h
@@ -10,6 +10,9 @@
 #include <linux/rbtree.h>
 #include <linux/rwsem.h>
 #include <linux/completion.h>
+#include <linux/rcupdate.h>
+#include <linux/semaphore.h>
+#include <linux/workqueue.h>
 #include <asm/page.h>
 #include <asm/mmu.h>
 

@@ -74,9 +77,18 @@ struct page {
 		pgoff_t index;		/* Our offset within mapping. */
 		void *freelist;		/* SLUB: freelist req. slab lock */
 	};
+
+	union {
 	struct list_head lru;		/* Pageout list, eg. active_list
 					 * protected by zone->lru_lock !
 					 */
+		struct rcu_head rcu;	/* Used by page table relocation code
+					 * to remember page for later freeing,
+					 * after we are sure anyone
+					 * poking at the page tables is no
+					 * longer looking at this page.
+					 */
+	};
 	/*
 	 * On machines where all RAM is mapped into kernel address space,
 	 * we can simply calculate the virtual address. On machines with

@@ -182,6 +194,8 @@ struct mm_struct {
 	int map_count;				/* number of VMAs */
 	int core_waiters;
 	struct rw_semaphore mmap_sem;
+	unsigned long flags; /* Must use atomic bitops to access the bits */
+
 	spinlock_t page_table_lock;		/* Protects page tables and some counters */
 
 	struct list_head mmlist;		/* List of maybe swapped mm's.	These are globally strung

@@ -222,8 +236,6 @@ struct mm_struct {
 	unsigned int token_priority;
 	unsigned int last_interval;
 
-	unsigned long flags; /* Must use atomic bitops to access the bits */
-
 	/* coredumping support */
 	struct completion *core_startup_done, core_done;
 

@@ -239,6 +251,11 @@ struct mm_struct {
 	struct file *exe_file;
 	unsigned long num_exe_file_vmas;
 #endif
+
+#ifdef CONFIG_RELOCATE_PAGE_TABLES
+	/* Page table relocation support. */
+	struct semaphore	page_table_relocation_lock;
+#endif /* CONFIG_RELOCATE_PAGE_TABLES */
 };
 
 #endif /* _LINUX_MM_TYPES_H */

diff -uprwNBb -X 2.6.25-rc8/Documentation/dontdiff /home/rossb/local/linux-2.6.25-rc8/kernel/fork.c 2.6.25-rc8/kernel/fork.c
@@ -360,6 +360,10 @@ static struct mm_struct * mm_init(struct
 	mm->cached_hole_size = ~0UL;
 	mm_init_cgroup(mm, p);
 
+#ifdef CONFIG_RELOCATE_PAGE_TABLES
+	init_MUTEX(&mm->page_table_relocation_lock);
+#endif
+
 	if (likely(!mm_alloc_pgd(mm))) {
 		mm->def_flags = 0;
 		return mm;

diff -uprwNBb -X 2.6.25-rc8/Documentation/dontdiff /home/rossb/local/linux-2.6.25-rc8/mm/hugetlb.c 2.6.25-rc8/mm/hugetlb.c
@@ -784,6 +784,8 @@ int copy_hugetlb_page_range(struct mm_st
 
 		spin_lock(&dst->page_table_lock);
 		spin_lock(&src->page_table_lock);
+		delimbo_huge_pte(&src_pte, src, addr);
+		delimbo_huge_pte(&dst_pte, dst, addr);
 		if (!huge_pte_none(huge_ptep_get(src_pte))) {
 			if (cow)
 				huge_ptep_set_wrprotect(src, addr, src_pte);

@@ -960,6 +962,7 @@ retry:
 	}
 
 	spin_lock(&mm->page_table_lock);
+	delimbo_huge_pte(&ptep, mm, address);
 	size = i_size_read(mapping->host) >> HPAGE_SHIFT;
 	if (idx >= size)
 		goto backout;

@@ -1017,6 +1020,7 @@ int hugetlb_fault(struct mm_struct *mm, 
 	ret = 0;
 
 	spin_lock(&mm->page_table_lock);
+	delimbo_huge_pte(&ptep, mm, address);
 	/* Check for a racing update before calling hugetlb_cow */
 	if (likely(pte_same(entry, huge_ptep_get(ptep))))
 		if (write_access && !pte_write(entry))

diff -uprwNBb -X 2.6.25-rc8/Documentation/dontdiff /home/rossb/local/linux-2.6.25-rc8/mm/memory.c 2.6.25-rc8/mm/memory.c
@@ -312,6 +312,7 @@ int __pte_alloc(struct mm_struct *mm, pm
 		return -ENOMEM;
 
 	spin_lock(&mm->page_table_lock);
+	delimbo_pmd(&pmd, mm, address);
 	if (!pmd_present(*pmd)) {	/* Has another populated it ? */
 		mm->nr_ptes++;
 		pmd_populate(mm, pmd, new);

@@ -330,6 +331,7 @@ int __pte_alloc_kernel(pmd_t *pmd, unsig
 		return -ENOMEM;
 
 	spin_lock(&init_mm.page_table_lock);
+	delimbo_pmd(&pmd, &init_mm, address);
 	if (!pmd_present(*pmd)) {	/* Has another populated it ? */
 		pmd_populate_kernel(&init_mm, pmd, new);
 		new = NULL;

@@ -549,6 +551,8 @@ again:
 	src_pte = pte_offset_map_nested(src_pmd, addr);
 	src_ptl = pte_lockptr(src_mm, src_pmd);
 	spin_lock_nested(src_ptl, SINGLE_DEPTH_NESTING);
+
+	delimbo_pte(&src_pte, &src_ptl, &src_pmd, src_mm, addr);
 	arch_enter_lazy_mmu_mode();
 
 	do {

@@ -1566,13 +1570,15 @@ EXPORT_SYMBOL_GPL(apply_to_page_range);
  * and do_anonymous_page and do_no_page can safely check later on).
  */
 static inline int pte_unmap_same(struct mm_struct *mm, pmd_t *pmd,
-				pte_t *page_table, pte_t orig_pte)
+				pte_t *page_table, pte_t orig_pte,
+				unsigned long address)
 {
 	int same = 1;
 #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT)
 	if (sizeof(pte_t) > sizeof(unsigned long)) {
 		spinlock_t *ptl = pte_lockptr(mm, pmd);
 		spin_lock(ptl);
+		delimbo_pte(&page_table, &ptl, &pmd, mm, address);
 		same = pte_same(*page_table, orig_pte);
 		spin_unlock(ptl);
 	}

@@ -2099,7 +2105,7 @@ static int do_swap_page(struct mm_struct
 	pte_t pte;
 	int ret = 0;
 
-	if (!pte_unmap_same(mm, pmd, page_table, orig_pte))
+	if (!pte_unmap_same(mm, pmd, page_table, orig_pte, address))
 		goto out;
 
 	entry = pte_to_swp_entry(orig_pte);

@@ -2178,6 +2184,10 @@ static int do_swap_page(struct mm_struct
 	}
 
 	/* No need to invalidate - it was non-present before */
+	/* Unless of course the cpu might be looking at an old
+	   copy of the pte. */
+	maybe_reload_tlb_mm(mm);
+
 	update_mmu_cache(vma, address, pte);
 unlock:
 	pte_unmap_unlock(page_table, ptl);

@@ -2229,6 +2239,10 @@ static int do_anonymous_page(struct mm_s
 	set_pte_at(mm, address, page_table, entry);
 
 	/* No need to invalidate - it was non-present before */
+	/* Unless of course the cpu might be looking at an old
+	   copy of the pte. */
+	maybe_reload_tlb_mm(mm);
+
 	update_mmu_cache(vma, address, entry);
 unlock:
 	pte_unmap_unlock(page_table, ptl);

@@ -2379,6 +2393,10 @@ static int __do_fault(struct mm_struct *
 		}
 
 		/* no need to invalidate: a not-present page won't be cached */
+		/* Unless of course the cpu could be looking at an old page
+		   table entry. */
+		maybe_reload_tlb_mm(mm);
+
 		update_mmu_cache(vma, address, entry);
 	} else {
 		mem_cgroup_uncharge_page(page);

@@ -2488,7 +2506,7 @@ static int do_nonlinear_fault(struct mm_
 				(write_access ? FAULT_FLAG_WRITE : 0);
 	pgoff_t pgoff;
 
-	if (!pte_unmap_same(mm, pmd, page_table, orig_pte))
+	if (!pte_unmap_same(mm, pmd, page_table, orig_pte, address))
 		return 0;
 
 	if (unlikely(!(vma->vm_flags & VM_NONLINEAR) ||

@@ -2547,6 +2565,7 @@ static inline int handle_pte_fault(struc
 
 	ptl = pte_lockptr(mm, pmd);
 	spin_lock(ptl);
+	delimbo_pte(&pte, &ptl, &pmd, mm, address);
 	if (unlikely(!pte_same(*pte, entry)))
 		goto unlock;
 	if (write_access) {

@@ -2568,6 +2587,12 @@ static inline int handle_pte_fault(struc
 		if (write_access)
 			flush_tlb_page(vma, address);
 	}
+
+	/* if the cpu could be looking at an old page table, we need to
+	   flush out everything. */
+	maybe_reload_tlb_mm(mm);
+
+
 unlock:
 	pte_unmap_unlock(pte, ptl);
 	return 0;

@@ -2617,6 +2642,7 @@ int __pud_alloc(struct mm_struct *mm, pg
 		return -ENOMEM;
 
 	spin_lock(&mm->page_table_lock);
+	delimbo_pgd(&pgd, mm, address);
 	if (pgd_present(*pgd))		/* Another has populated it */
 		pud_free(mm, new);
 	else

@@ -2638,6 +2664,7 @@ int __pmd_alloc(struct mm_struct *mm, pu
 		return -ENOMEM;
 
 	spin_lock(&mm->page_table_lock);
+	delimbo_pud(&pud, mm, address);
 #ifndef __ARCH_HAS_4LEVEL_HACK
 	if (pud_present(*pud))		/* Another has populated it */
 		pmd_free(mm, new);

diff -uprwNBb -X 2.6.25-rc8/Documentation/dontdiff /home/rossb/local/linux-2.6.25-rc8/mm/mempolicy.c 2.6.25-rc8/mm/mempolicy.c
@@ -100,6 +100,12 @@
 static struct kmem_cache *policy_cache;
 static struct kmem_cache *sn_cache;
 
+#ifdef CONFIG_RELOCATE_PAGE_TABLES
+int migrate_page_tables_mm(struct mm_struct *mm,  int source,
+			   new_page_table_t get_new_page,
+			   unsigned long private);
+#endif
+
 /* Highest zone. An specific allocation for a zone below that is not
    policied. */
 enum zone_type policy_zone = 0;

@@ -730,6 +736,20 @@ static struct page *new_node_page(struct
 	return alloc_pages_node(node, GFP_HIGHUSER_MOVABLE, 0);
 }
 
+#ifdef CONFIG_RELOCATE_PAGE_TABLES
+static struct page *new_node_page_page_tables(struct mm_struct *mm,
+					      unsigned long addr,
+					      unsigned long node,
+					      int **x,
+					      int level)
+{
+	struct page *p;
+	p = alloc_page_table_node(mm, addr, node, level);
+	return p;
+}
+
+#endif /* CONFIG_RELOCATE_PAGE_TABLES  */
+
 /*
  * Migrate pages from one node to a target node.
  * Returns error or the number of pages not migrated.

@@ -750,6 +770,12 @@ static int migrate_to_node(struct mm_str
 	if (!list_empty(&pagelist))
 		err = migrate_pages(&pagelist, new_node_page, dest);
 
+#ifdef CONFIG_RELOCATE_PAGE_TABLES
+	if (!err)
+		err = migrate_page_tables_mm(mm, source,
+					     new_node_page_page_tables, dest);
+#endif /* CONFIG_RELOCATE_PAGE_TABLES */
+
 	return err;
 }
 

380 lines not on this page · raw message

diff -uprwNBb -X 2.6.25-rc8/Documentation/dontdiff /home/rossb/local/linux-2.6.25-rc8/mm/mremap.c 2.6.25-rc8/mm/mremap.c
@@ -98,6 +98,7 @@ static void move_ptes(struct vm_area_str
 	new_ptl = pte_lockptr(mm, new_pmd);
 	if (new_ptl != old_ptl)
 		spin_lock_nested(new_ptl, SINGLE_DEPTH_NESTING);
+	delimbo_pte(&new_pte, &new_ptl, &new_pmd, mm, new_addr);
 	arch_enter_lazy_mmu_mode();
 
 	for (; old_addr < old_end; old_pte++, old_addr += PAGE_SIZE,

diff -uprwNBb -X 2.6.25-rc8/Documentation/dontdiff /home/rossb/local/linux-2.6.25-rc8/mm/rmap.c 2.6.25-rc8/mm/rmap.c
@@ -255,6 +255,7 @@ pte_t *page_check_address(struct page *p
 
 	ptl = pte_lockptr(mm, pmd);
 	spin_lock(ptl);
+	delimbo_pte(&pte, &ptl, &pmd, mm, address);
 	if (pte_present(*pte) && page_to_pfn(page) == pte_pfn(*pte)) {
 		*ptlp = ptl;
 		return pte;