mm: Make page tables relocatable

diff -uprwNbB -X 2.6.23/Documentation/dontdiff 2.6.23/arch/i386/mm/hugetlbpage.c 2.6.23a/arch/i386/mm/hugetlbpage.c
@@ -87,6 +87,7 @@ 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, (unsigned long) spte & PAGE_MASK);
 	else

diff -uprwNbB -X 2.6.23/Documentation/dontdiff 2.6.23/arch/powerpc/mm/fault.c 2.6.23a/arch/powerpc/mm/fault.c
@@ -301,6 +301,8 @@ 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 2.6.23/Documentation/dontdiff 2.6.23/arch/powerpc/mm/hugetlbpage.c 2.6.23a/arch/powerpc/mm/hugetlbpage.c
@@ -77,6 +77,7 @@ 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 2.6.23/Documentation/dontdiff 2.6.23/arch/ppc/mm/fault.c 2.6.23a/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 2.6.23/Documentation/dontdiff 2.6.23/arch/x86_64/kernel/smp.c 2.6.23a/arch/x86_64/kernel/smp.c
@@ -56,6 +56,7 @@ union smp_flush_state {
 		struct mm_struct *flush_mm;
 		unsigned long flush_va;
 #define FLUSH_ALL	-1ULL
+#define RELOAD_ALL	-2ULL
 		spinlock_t tlbstate_lock;
 	};
 	char pad[SMP_CACHE_BYTES];

@@ -155,6 +156,8 @@ asmlinkage void smp_invalidate_interrupt
 		if (read_pda(mmu_state) == TLBSTATE_OK) {
 			if (f->flush_va == FLUSH_ALL)
 				local_flush_tlb();
+			else if (f->flush_va == RELOAD_ALL)
+				local_reload_tlb_mm(f->flush_mm);
 			else
 				__flush_tlb_one(f->flush_va);
 		} else

@@ -225,10 +228,36 @@ void flush_tlb_current_task(void)
 }
 EXPORT_SYMBOL(flush_tlb_current_task);
 
+void reload_tlb_mm(struct mm_struct *mm)
+{
+	cpumask_t cpu_mask;
+
+	clear_bit(MMF_NEED_RELOAD, &mm->flags);
+	clear_bit(MMF_NEED_FLUSH, &mm->flags);
+
+	preempt_disable();
+	cpu_mask = mm->cpu_vm_mask;
+	cpu_clear(smp_processor_id(), cpu_mask);
+
+	if (current->active_mm == mm) {
+		if (current->mm)
+			local_reload_tlb_mm(mm);
+		else
+			leave_mm(smp_processor_id());
+	}
+	if (!cpus_empty(cpu_mask))
+		flush_tlb_others(cpu_mask, mm, RELOAD_ALL);
+
+	preempt_enable();
+
+}
+
 void flush_tlb_mm (struct mm_struct * mm)
 {
 	cpumask_t cpu_mask;
 
+	clear_bit(MMF_NEED_FLUSH, &mm->flags);
+
 	preempt_disable();
 	cpu_mask = mm->cpu_vm_mask;
 	cpu_clear(smp_processor_id(), cpu_mask);

57 lines not on this page · raw message

diff -uprwNbB -X 2.6.23/Documentation/dontdiff 2.6.23/include/asm-x86_64/tlbflush.h 2.6.23a/include/asm-x86_64/tlbflush.h
@@ -6,6 +6,13 @@
 #include <asm/processor.h>
 #include <asm/system.h>
 
+#define ARCH_HAS_RELOAD_TLB
+static inline void load_cr3(pgd_t *pgd);
+static inline void __reload_tlb_mm(struct mm_struct *mm)
+{
+	load_cr3(mm->pgd);
+}
+
 static inline void __flush_tlb(void)
 {
 	write_cr3(read_cr3());

@@ -44,6 +50,12 @@ static inline void __flush_tlb_all(void)
 #define flush_tlb_all() __flush_tlb_all()
 #define local_flush_tlb() __flush_tlb()
 
+static inline void reload_tlb_mm(struct mm_struct *mm)
+{
+	if (mm == current->active_mm)
+		__reload_tlb_mm(mm);
+}
+
 static inline void flush_tlb_mm(struct mm_struct *mm)
 {
 	if (mm == current->active_mm)

@@ -71,6 +83,10 @@ static inline void flush_tlb_range(struc
 #define local_flush_tlb() \
 	__flush_tlb()
 
+#define local_reload_tlb_mm(mm) \
+	__reload_tlb_mm(mm)
+
+extern void reload_tlb_mm(struct mm_struct *mm);
 extern void flush_tlb_all(void);
 extern void flush_tlb_current_task(void);
 extern void flush_tlb_mm(struct mm_struct *);

@@ -106,4 +122,6 @@ static inline void flush_tlb_pgtables(st
 	   by the normal TLB flushing algorithms. */
 }
 
+#include <asm-generic/tlbflush.h>
+
 #endif /* _X8664_TLBFLUSH_H */

76 lines not on this page · raw message

diff -uprwNbB -X 2.6.23/Documentation/dontdiff 2.6.23/include/linux/mm_types.h 2.6.23a/include/linux/mm_types.h
@@ -5,6 +5,7 @@
 #include <linux/threads.h>
 #include <linux/list.h>
 #include <linux/spinlock.h>
+#include <linux/rcupdate.h>
 
 struct address_space;
 

@@ -61,9 +62,18 @@ 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

diff -uprwNbB -X 2.6.23/Documentation/dontdiff 2.6.23/mm/hugetlb.c 2.6.23a/mm/hugetlb.c
@@ -379,6 +379,8 @@ int copy_hugetlb_page_range(struct mm_st
 			goto nomem;
 		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 (!pte_none(*src_pte)) {
 			if (cow)
 				ptep_set_wrprotect(src, addr, src_pte);

@@ -551,6 +553,7 @@ 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;

@@ -609,6 +612,7 @@ 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, *ptep)))
 		if (write_access && !pte_write(entry))

diff -uprwNbB -X 2.6.23/Documentation/dontdiff 2.6.23/mm/memory.c 2.6.23a/mm/memory.c
@@ -306,6 +306,7 @@ int __pte_alloc(struct mm_struct *mm, pm
 
 	pte_lock_init(new);
 	spin_lock(&mm->page_table_lock);
+	delimbo_pmd(&pmd, mm, address);
 	if (pmd_present(*pmd)) {	/* Another has populated it */
 		pte_lock_deinit(new);
 		pte_free(new);

@@ -325,6 +326,7 @@ 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))		/* Another has populated it */
 		pte_free_kernel(new);
 	else

@@ -504,6 +506,8 @@ 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 {

@@ -1558,13 +1562,15 @@ 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);
 	}

@@ -2153,7 +2159,7 @@ 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);

@@ -2227,6 +2233,10 @@ 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);

@@ -2279,6 +2289,7 @@ static int do_anonymous_page(struct mm_s
 
 		ptl = pte_lockptr(mm, pmd);
 		spin_lock(ptl);
+		delimbo_pte(&page_table, &ptl, &pmd, mm, address);
 		if (!pte_none(*page_table))
 			goto release;
 		inc_mm_counter(mm, file_rss);

@@ -2288,6 +2299,10 @@ 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);
 	lazy_mmu_prot_update(entry);
 unlock:

@@ -2441,6 +2456,10 @@ 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);
 		lazy_mmu_prot_update(entry);
 	} else {

@@ -2544,7 +2563,7 @@ 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) ||

@@ -2603,6 +2622,7 @@ 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) {

@@ -2625,6 +2645,12 @@ 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;

@@ -2674,6 +2700,7 @@ 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(new);
 	else

@@ -2695,6 +2722,7 @@ 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(new);

diff -uprwNbB -X 2.6.23/Documentation/dontdiff 2.6.23/mm/mempolicy.c 2.6.23a/mm/mempolicy.c
@@ -101,6 +101,11 @@
 static struct kmem_cache *policy_cache;
 static struct kmem_cache *sn_cache;
 
+
+int migrate_page_tables_mm(struct mm_struct *mm,  int source,
+			   new_page_t get_new_page, unsigned long private);
+
+
 /* Highest zone. An specific allocation for a zone below that is not
    policied. */
 enum zone_type policy_zone = 0;

@@ -597,6 +602,11 @@ static struct page *new_node_page(struct
 	return alloc_pages_node(node, GFP_HIGHUSER_MOVABLE, 0);
 }
 
+static struct page *new_node_page_page_tables(struct page *page,
+					      unsigned long node, int **x) {
+	return alloc_pages_node(node, GFP_KERNEL, 0);
+}
+
 /*
  * Migrate pages from one node to a target node.
  * Returns error or the number of pages not migrated.

@@ -616,6 +626,10 @@ int migrate_to_node(struct mm_struct *mm
 	if (!list_empty(&pagelist))
 		err = migrate_pages(&pagelist, new_node_page, dest);
 
+	if (!err)
+		err = migrate_page_tables_mm(mm, source,
+					     new_node_page_page_tables, dest);
+
 	return err;
 }
 

346 lines not on this page · raw message

diff -uprwNbB -X 2.6.23/Documentation/dontdiff 2.6.23/mm/mremap.c 2.6.23a/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 2.6.23/Documentation/dontdiff 2.6.23/mm/rmap.c 2.6.23a/mm/rmap.c
@@ -254,6 +254,7 @@ 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;