linux/mm/mmap.c

2090 lines
54 KiB
C
Raw Normal View History

/*
* mm/mmap.c
*
* Written by obz.
*
* Address space accounting code <alan@redhat.com>
*/
#include <linux/slab.h>
#include <linux/mm.h>
#include <linux/shm.h>
#include <linux/mman.h>
#include <linux/pagemap.h>
#include <linux/swap.h>
#include <linux/syscalls.h>
#include <linux/capability.h>
#include <linux/init.h>
#include <linux/file.h>
#include <linux/fs.h>
#include <linux/personality.h>
#include <linux/security.h>
#include <linux/hugetlb.h>
#include <linux/profile.h>
#include <linux/module.h>
#include <linux/mount.h>
#include <linux/mempolicy.h>
#include <linux/rmap.h>
#include <asm/uaccess.h>
#include <asm/cacheflush.h>
#include <asm/tlb.h>
#ifndef arch_mmap_check
#define arch_mmap_check(addr, len, flags) (0)
#endif
[PATCH] freepgt: free_pgtables use vma list Recent woes with some arches needing their own pgd_addr_end macro; and 4-level clear_page_range regression since 2.6.10's clear_page_tables; and its long-standing well-known inefficiency in searching throughout the higher-level page tables for those few entries to clear and free: all can be blamed on ignoring the list of vmas when we free page tables. Replace exit_mmap's clear_page_range of the total user address space by free_pgtables operating on the mm's vma list; unmap_region use it in the same way, giving floor and ceiling beyond which it may not free tables. This brings lmbench fork/exec/sh numbers back to 2.6.10 (unless preempt is enabled, in which case latency fixes spoil unmap_vmas throughput). Beware: the do_mmap_pgoff driver failure case must now use unmap_region instead of zap_page_range, since a page table might have been allocated, and can only be freed while it is touched by some vma. Move free_pgtables from mmap.c to memory.c, where its lower levels are adapted from the clear_page_range levels. (Most of free_pgtables' old code was actually for a non-existent case, prev not properly set up, dating from before hch gave us split_vma.) Pass mmu_gather** in the public interfaces, since we might want to add latency lockdrops later; but no attempt to do so yet, going by vma should itself reduce latency. But what if is_hugepage_only_range? Those ia64 and ppc64 cases need careful examination: put that off until a later patch of the series. What of x86_64's 32bit vdso page __map_syscall32 maps outside any vma? And the range to sparc64's flush_tlb_pgtables? It's less clear to me now that we need to do more than is done here - every PMD_SIZE ever occupied will be flushed, do we really have to flush every PGDIR_SIZE ever partially occupied? A shame to complicate it unnecessarily. Special thanks to David Miller for time spent repairing my ceilings. Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-04-19 20:29:15 +00:00
static void unmap_region(struct mm_struct *mm,
struct vm_area_struct *vma, struct vm_area_struct *prev,
unsigned long start, unsigned long end);
/*
* WARNING: the debugging will use recursive algorithms so never enable this
* unless you know what you are doing.
*/
#undef DEBUG_MM_RB
/* description of effects of mapping type and prot in current implementation.
* this is due to the limited x86 page protection hardware. The expected
* behavior is in parens:
*
* map_type prot
* PROT_NONE PROT_READ PROT_WRITE PROT_EXEC
* MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes
* w: (no) no w: (no) no w: (yes) yes w: (no) no
* x: (no) no x: (no) yes x: (no) yes x: (yes) yes
*
* MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes
* w: (no) no w: (no) no w: (copy) copy w: (no) no
* x: (no) no x: (no) yes x: (no) yes x: (yes) yes
*
*/
pgprot_t protection_map[16] = {
__P000, __P001, __P010, __P011, __P100, __P101, __P110, __P111,
__S000, __S001, __S010, __S011, __S100, __S101, __S110, __S111
};
int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */
int sysctl_overcommit_ratio = 50; /* default is 50% */
int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT;
atomic_t vm_committed_space = ATOMIC_INIT(0);
/*
* Check that a process has enough memory to allocate a new virtual
* mapping. 0 means there is enough memory for the allocation to
* succeed and -ENOMEM implies there is not.
*
* We currently support three overcommit policies, which are set via the
* vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
*
* Strict overcommit modes added 2002 Feb 26 by Alan Cox.
* Additional code 2002 Jul 20 by Robert Love.
*
* cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
*
* Note this is a helper function intended to be used by LSMs which
* wish to use this logic.
*/
int __vm_enough_memory(long pages, int cap_sys_admin)
{
unsigned long free, allowed;
vm_acct_memory(pages);
/*
* Sometimes we want to use more memory than we have
*/
if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
return 0;
if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
unsigned long n;
free = global_page_state(NR_FILE_PAGES);
free += nr_swap_pages;
/*
* Any slabs which are created with the
* SLAB_RECLAIM_ACCOUNT flag claim to have contents
* which are reclaimable, under pressure. The dentry
* cache and most inode caches should fall into this
*/
free += atomic_read(&slab_reclaim_pages);
/*
* Leave the last 3% for root
*/
if (!cap_sys_admin)
free -= free / 32;
if (free > pages)
return 0;
/*
* nr_free_pages() is very expensive on large systems,
* only call if we're about to fail.
*/
n = nr_free_pages();
/*
* Leave reserved pages. The pages are not for anonymous pages.
*/
if (n <= totalreserve_pages)
goto error;
else
n -= totalreserve_pages;
/*
* Leave the last 3% for root
*/
if (!cap_sys_admin)
n -= n / 32;
free += n;
if (free > pages)
return 0;
goto error;
}
allowed = (totalram_pages - hugetlb_total_pages())
* sysctl_overcommit_ratio / 100;
/*
* Leave the last 3% for root
*/
if (!cap_sys_admin)
allowed -= allowed / 32;
allowed += total_swap_pages;
/* Don't let a single process grow too big:
leave 3% of the size of this process for other processes */
allowed -= current->mm->total_vm / 32;
/*
* cast `allowed' as a signed long because vm_committed_space
* sometimes has a negative value
*/
if (atomic_read(&vm_committed_space) < (long)allowed)
return 0;
error:
vm_unacct_memory(pages);
return -ENOMEM;
}
EXPORT_SYMBOL(__vm_enough_memory);
/*
* Requires inode->i_mapping->i_mmap_lock
*/
static void __remove_shared_vm_struct(struct vm_area_struct *vma,
struct file *file, struct address_space *mapping)
{
if (vma->vm_flags & VM_DENYWRITE)
atomic_inc(&file->f_dentry->d_inode->i_writecount);
if (vma->vm_flags & VM_SHARED)
mapping->i_mmap_writable--;
flush_dcache_mmap_lock(mapping);
if (unlikely(vma->vm_flags & VM_NONLINEAR))
list_del_init(&vma->shared.vm_set.list);
else
vma_prio_tree_remove(vma, &mapping->i_mmap);
flush_dcache_mmap_unlock(mapping);
}
/*
* Unlink a file-based vm structure from its prio_tree, to hide
* vma from rmap and vmtruncate before freeing its page tables.
*/
void unlink_file_vma(struct vm_area_struct *vma)
{
struct file *file = vma->vm_file;
if (file) {
struct address_space *mapping = file->f_mapping;
spin_lock(&mapping->i_mmap_lock);
__remove_shared_vm_struct(vma, file, mapping);
spin_unlock(&mapping->i_mmap_lock);
}
}
/*
* Close a vm structure and free it, returning the next.
*/
static struct vm_area_struct *remove_vma(struct vm_area_struct *vma)
{
struct vm_area_struct *next = vma->vm_next;
might_sleep();
if (vma->vm_ops && vma->vm_ops->close)
vma->vm_ops->close(vma);
if (vma->vm_file)
fput(vma->vm_file);
mpol_free(vma_policy(vma));
kmem_cache_free(vm_area_cachep, vma);
return next;
}
asmlinkage unsigned long sys_brk(unsigned long brk)
{
unsigned long rlim, retval;
unsigned long newbrk, oldbrk;
struct mm_struct *mm = current->mm;
down_write(&mm->mmap_sem);
if (brk < mm->end_code)
goto out;
/*
* Check against rlimit here. If this check is done later after the test
* of oldbrk with newbrk then it can escape the test and let the data
* segment grow beyond its set limit the in case where the limit is
* not page aligned -Ram Gupta
*/
rlim = current->signal->rlim[RLIMIT_DATA].rlim_cur;
if (rlim < RLIM_INFINITY && brk - mm->start_data > rlim)
goto out;
newbrk = PAGE_ALIGN(brk);
oldbrk = PAGE_ALIGN(mm->brk);
if (oldbrk == newbrk)
goto set_brk;
/* Always allow shrinking brk. */
if (brk <= mm->brk) {
if (!do_munmap(mm, newbrk, oldbrk-newbrk))
goto set_brk;
goto out;
}
/* Check against existing mmap mappings. */
if (find_vma_intersection(mm, oldbrk, newbrk+PAGE_SIZE))
goto out;
/* Ok, looks good - let it rip. */
if (do_brk(oldbrk, newbrk-oldbrk) != oldbrk)
goto out;
set_brk:
mm->brk = brk;
out:
retval = mm->brk;
up_write(&mm->mmap_sem);
return retval;
}
#ifdef DEBUG_MM_RB
static int browse_rb(struct rb_root *root)
{
int i = 0, j;
struct rb_node *nd, *pn = NULL;
unsigned long prev = 0, pend = 0;
for (nd = rb_first(root); nd; nd = rb_next(nd)) {
struct vm_area_struct *vma;
vma = rb_entry(nd, struct vm_area_struct, vm_rb);
if (vma->vm_start < prev)
printk("vm_start %lx prev %lx\n", vma->vm_start, prev), i = -1;
if (vma->vm_start < pend)
printk("vm_start %lx pend %lx\n", vma->vm_start, pend);
if (vma->vm_start > vma->vm_end)
printk("vm_end %lx < vm_start %lx\n", vma->vm_end, vma->vm_start);
i++;
pn = nd;
}
j = 0;
for (nd = pn; nd; nd = rb_prev(nd)) {
j++;
}
if (i != j)
printk("backwards %d, forwards %d\n", j, i), i = 0;
return i;
}
void validate_mm(struct mm_struct *mm)
{
int bug = 0;
int i = 0;
struct vm_area_struct *tmp = mm->mmap;
while (tmp) {
tmp = tmp->vm_next;
i++;
}
if (i != mm->map_count)
printk("map_count %d vm_next %d\n", mm->map_count, i), bug = 1;
i = browse_rb(&mm->mm_rb);
if (i != mm->map_count)
printk("map_count %d rb %d\n", mm->map_count, i), bug = 1;
BUG_ON(bug);
}
#else
#define validate_mm(mm) do { } while (0)
#endif
static struct vm_area_struct *
find_vma_prepare(struct mm_struct *mm, unsigned long addr,
struct vm_area_struct **pprev, struct rb_node ***rb_link,
struct rb_node ** rb_parent)
{
struct vm_area_struct * vma;
struct rb_node ** __rb_link, * __rb_parent, * rb_prev;
__rb_link = &mm->mm_rb.rb_node;
rb_prev = __rb_parent = NULL;
vma = NULL;
while (*__rb_link) {
struct vm_area_struct *vma_tmp;
__rb_parent = *__rb_link;
vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb);
if (vma_tmp->vm_end > addr) {
vma = vma_tmp;
if (vma_tmp->vm_start <= addr)
return vma;
__rb_link = &__rb_parent->rb_left;
} else {
rb_prev = __rb_parent;
__rb_link = &__rb_parent->rb_right;
}
}
*pprev = NULL;
if (rb_prev)
*pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
*rb_link = __rb_link;
*rb_parent = __rb_parent;
return vma;
}
static inline void
__vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma,
struct vm_area_struct *prev, struct rb_node *rb_parent)
{
if (prev) {
vma->vm_next = prev->vm_next;
prev->vm_next = vma;
} else {
mm->mmap = vma;
if (rb_parent)
vma->vm_next = rb_entry(rb_parent,
struct vm_area_struct, vm_rb);
else
vma->vm_next = NULL;
}
}
void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma,
struct rb_node **rb_link, struct rb_node *rb_parent)
{
rb_link_node(&vma->vm_rb, rb_parent, rb_link);
rb_insert_color(&vma->vm_rb, &mm->mm_rb);
}
static inline void __vma_link_file(struct vm_area_struct *vma)
{
struct file * file;
file = vma->vm_file;
if (file) {
struct address_space *mapping = file->f_mapping;
if (vma->vm_flags & VM_DENYWRITE)
atomic_dec(&file->f_dentry->d_inode->i_writecount);
if (vma->vm_flags & VM_SHARED)
mapping->i_mmap_writable++;
flush_dcache_mmap_lock(mapping);
if (unlikely(vma->vm_flags & VM_NONLINEAR))
vma_nonlinear_insert(vma, &mapping->i_mmap_nonlinear);
else
vma_prio_tree_insert(vma, &mapping->i_mmap);
flush_dcache_mmap_unlock(mapping);
}
}
static void
__vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
struct vm_area_struct *prev, struct rb_node **rb_link,
struct rb_node *rb_parent)
{
__vma_link_list(mm, vma, prev, rb_parent);
__vma_link_rb(mm, vma, rb_link, rb_parent);
__anon_vma_link(vma);
}
static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
struct vm_area_struct *prev, struct rb_node **rb_link,
struct rb_node *rb_parent)
{
struct address_space *mapping = NULL;
if (vma->vm_file)
mapping = vma->vm_file->f_mapping;
if (mapping) {
spin_lock(&mapping->i_mmap_lock);
vma->vm_truncate_count = mapping->truncate_count;
}
anon_vma_lock(vma);
__vma_link(mm, vma, prev, rb_link, rb_parent);
__vma_link_file(vma);
anon_vma_unlock(vma);
if (mapping)
spin_unlock(&mapping->i_mmap_lock);
mm->map_count++;
validate_mm(mm);
}
/*
* Helper for vma_adjust in the split_vma insert case:
* insert vm structure into list and rbtree and anon_vma,
* but it has already been inserted into prio_tree earlier.
*/
static void
__insert_vm_struct(struct mm_struct * mm, struct vm_area_struct * vma)
{
struct vm_area_struct * __vma, * prev;
struct rb_node ** rb_link, * rb_parent;
__vma = find_vma_prepare(mm, vma->vm_start,&prev, &rb_link, &rb_parent);
BUG_ON(__vma && __vma->vm_start < vma->vm_end);
__vma_link(mm, vma, prev, rb_link, rb_parent);
mm->map_count++;
}
static inline void
__vma_unlink(struct mm_struct *mm, struct vm_area_struct *vma,
struct vm_area_struct *prev)
{
prev->vm_next = vma->vm_next;
rb_erase(&vma->vm_rb, &mm->mm_rb);
if (mm->mmap_cache == vma)
mm->mmap_cache = prev;
}
/*
* We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
* is already present in an i_mmap tree without adjusting the tree.
* The following helper function should be used when such adjustments
* are necessary. The "insert" vma (if any) is to be inserted
* before we drop the necessary locks.
*/
void vma_adjust(struct vm_area_struct *vma, unsigned long start,
unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert)
{
struct mm_struct *mm = vma->vm_mm;
struct vm_area_struct *next = vma->vm_next;
struct vm_area_struct *importer = NULL;
struct address_space *mapping = NULL;
struct prio_tree_root *root = NULL;
struct file *file = vma->vm_file;
struct anon_vma *anon_vma = NULL;
long adjust_next = 0;
int remove_next = 0;
if (next && !insert) {
if (end >= next->vm_end) {
/*
* vma expands, overlapping all the next, and
* perhaps the one after too (mprotect case 6).
*/
again: remove_next = 1 + (end > next->vm_end);
end = next->vm_end;
anon_vma = next->anon_vma;
importer = vma;
} else if (end > next->vm_start) {
/*
* vma expands, overlapping part of the next:
* mprotect case 5 shifting the boundary up.
*/
adjust_next = (end - next->vm_start) >> PAGE_SHIFT;
anon_vma = next->anon_vma;
importer = vma;
} else if (end < vma->vm_end) {
/*
* vma shrinks, and !insert tells it's not
* split_vma inserting another: so it must be
* mprotect case 4 shifting the boundary down.
*/
adjust_next = - ((vma->vm_end - end) >> PAGE_SHIFT);
anon_vma = next->anon_vma;
importer = next;
}
}
if (file) {
mapping = file->f_mapping;
if (!(vma->vm_flags & VM_NONLINEAR))
root = &mapping->i_mmap;
spin_lock(&mapping->i_mmap_lock);
if (importer &&
vma->vm_truncate_count != next->vm_truncate_count) {
/*
* unmap_mapping_range might be in progress:
* ensure that the expanding vma is rescanned.
*/
importer->vm_truncate_count = 0;
}
if (insert) {
insert->vm_truncate_count = vma->vm_truncate_count;
/*
* Put into prio_tree now, so instantiated pages
* are visible to arm/parisc __flush_dcache_page
* throughout; but we cannot insert into address
* space until vma start or end is updated.
*/
__vma_link_file(insert);
}
}
/*
* When changing only vma->vm_end, we don't really need
* anon_vma lock: but is that case worth optimizing out?
*/
if (vma->anon_vma)
anon_vma = vma->anon_vma;
if (anon_vma) {
spin_lock(&anon_vma->lock);
/*
* Easily overlooked: when mprotect shifts the boundary,
* make sure the expanding vma has anon_vma set if the
* shrinking vma had, to cover any anon pages imported.
*/
if (importer && !importer->anon_vma) {
importer->anon_vma = anon_vma;
__anon_vma_link(importer);
}
}
if (root) {
flush_dcache_mmap_lock(mapping);
vma_prio_tree_remove(vma, root);
if (adjust_next)
vma_prio_tree_remove(next, root);
}
vma->vm_start = start;
vma->vm_end = end;
vma->vm_pgoff = pgoff;
if (adjust_next) {
next->vm_start += adjust_next << PAGE_SHIFT;
next->vm_pgoff += adjust_next;
}
if (root) {
if (adjust_next)
vma_prio_tree_insert(next, root);
vma_prio_tree_insert(vma, root);
flush_dcache_mmap_unlock(mapping);
}
if (remove_next) {
/*
* vma_merge has merged next into vma, and needs
* us to remove next before dropping the locks.
*/
__vma_unlink(mm, next, vma);
if (file)
__remove_shared_vm_struct(next, file, mapping);
if (next->anon_vma)
__anon_vma_merge(vma, next);
} else if (insert) {
/*
* split_vma has split insert from vma, and needs
* us to insert it before dropping the locks
* (it may either follow vma or precede it).
*/
__insert_vm_struct(mm, insert);
}
if (anon_vma)
spin_unlock(&anon_vma->lock);
if (mapping)
spin_unlock(&mapping->i_mmap_lock);
if (remove_next) {
if (file)
fput(file);
mm->map_count--;
mpol_free(vma_policy(next));
kmem_cache_free(vm_area_cachep, next);
/*
* In mprotect's case 6 (see comments on vma_merge),
* we must remove another next too. It would clutter
* up the code too much to do both in one go.
*/
if (remove_next == 2) {
next = vma->vm_next;
goto again;
}
}
validate_mm(mm);
}
/*
* If the vma has a ->close operation then the driver probably needs to release
* per-vma resources, so we don't attempt to merge those.
*/
#define VM_SPECIAL (VM_IO | VM_DONTEXPAND | VM_RESERVED | VM_PFNMAP)
static inline int is_mergeable_vma(struct vm_area_struct *vma,
struct file *file, unsigned long vm_flags)
{
if (vma->vm_flags != vm_flags)
return 0;
if (vma->vm_file != file)
return 0;
if (vma->vm_ops && vma->vm_ops->close)
return 0;
return 1;
}
static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1,
struct anon_vma *anon_vma2)
{
return !anon_vma1 || !anon_vma2 || (anon_vma1 == anon_vma2);
}
/*
* Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
* in front of (at a lower virtual address and file offset than) the vma.
*
* We cannot merge two vmas if they have differently assigned (non-NULL)
* anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
*
* We don't check here for the merged mmap wrapping around the end of pagecache
* indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
* wrap, nor mmaps which cover the final page at index -1UL.
*/
static int
can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags,
struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
{
if (is_mergeable_vma(vma, file, vm_flags) &&
is_mergeable_anon_vma(anon_vma, vma->anon_vma)) {
if (vma->vm_pgoff == vm_pgoff)
return 1;
}
return 0;
}
/*
* Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
* beyond (at a higher virtual address and file offset than) the vma.
*
* We cannot merge two vmas if they have differently assigned (non-NULL)
* anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
*/
static int
can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags,
struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
{
if (is_mergeable_vma(vma, file, vm_flags) &&
is_mergeable_anon_vma(anon_vma, vma->anon_vma)) {
pgoff_t vm_pglen;
vm_pglen = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
if (vma->vm_pgoff + vm_pglen == vm_pgoff)
return 1;
}
return 0;
}
/*
* Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
* whether that can be merged with its predecessor or its successor.
* Or both (it neatly fills a hole).
*
* In most cases - when called for mmap, brk or mremap - [addr,end) is
* certain not to be mapped by the time vma_merge is called; but when
* called for mprotect, it is certain to be already mapped (either at
* an offset within prev, or at the start of next), and the flags of
* this area are about to be changed to vm_flags - and the no-change
* case has already been eliminated.
*
* The following mprotect cases have to be considered, where AAAA is
* the area passed down from mprotect_fixup, never extending beyond one
* vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
*
* AAAA AAAA AAAA AAAA
* PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
* cannot merge might become might become might become
* PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
* mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
* mremap move: PPPPNNNNNNNN 8
* AAAA
* PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
* might become case 1 below case 2 below case 3 below
*
* Odd one out? Case 8, because it extends NNNN but needs flags of XXXX:
* mprotect_fixup updates vm_flags & vm_page_prot on successful return.
*/
struct vm_area_struct *vma_merge(struct mm_struct *mm,
struct vm_area_struct *prev, unsigned long addr,
unsigned long end, unsigned long vm_flags,
struct anon_vma *anon_vma, struct file *file,
pgoff_t pgoff, struct mempolicy *policy)
{
pgoff_t pglen = (end - addr) >> PAGE_SHIFT;
struct vm_area_struct *area, *next;
/*
* We later require that vma->vm_flags == vm_flags,
* so this tests vma->vm_flags & VM_SPECIAL, too.
*/
if (vm_flags & VM_SPECIAL)
return NULL;
if (prev)
next = prev->vm_next;
else
next = mm->mmap;
area = next;
if (next && next->vm_end == end) /* cases 6, 7, 8 */
next = next->vm_next;
/*
* Can it merge with the predecessor?
*/
if (prev && prev->vm_end == addr &&
mpol_equal(vma_policy(prev), policy) &&
can_vma_merge_after(prev, vm_flags,
anon_vma, file, pgoff)) {
/*
* OK, it can. Can we now merge in the successor as well?
*/
if (next && end == next->vm_start &&
mpol_equal(policy, vma_policy(next)) &&
can_vma_merge_before(next, vm_flags,
anon_vma, file, pgoff+pglen) &&
is_mergeable_anon_vma(prev->anon_vma,
next->anon_vma)) {
/* cases 1, 6 */
vma_adjust(prev, prev->vm_start,
next->vm_end, prev->vm_pgoff, NULL);
} else /* cases 2, 5, 7 */
vma_adjust(prev, prev->vm_start,
end, prev->vm_pgoff, NULL);
return prev;
}
/*
* Can this new request be merged in front of next?
*/
if (next && end == next->vm_start &&
mpol_equal(policy, vma_policy(next)) &&
can_vma_merge_before(next, vm_flags,
anon_vma, file, pgoff+pglen)) {
if (prev && addr < prev->vm_end) /* case 4 */
vma_adjust(prev, prev->vm_start,
addr, prev->vm_pgoff, NULL);
else /* cases 3, 8 */
vma_adjust(area, addr, next->vm_end,
next->vm_pgoff - pglen, NULL);
return area;
}
return NULL;
}
/*
* find_mergeable_anon_vma is used by anon_vma_prepare, to check
* neighbouring vmas for a suitable anon_vma, before it goes off
* to allocate a new anon_vma. It checks because a repetitive
* sequence of mprotects and faults may otherwise lead to distinct
* anon_vmas being allocated, preventing vma merge in subsequent
* mprotect.
*/
struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma)
{
struct vm_area_struct *near;
unsigned long vm_flags;
near = vma->vm_next;
if (!near)
goto try_prev;
/*
* Since only mprotect tries to remerge vmas, match flags
* which might be mprotected into each other later on.
* Neither mlock nor madvise tries to remerge at present,
* so leave their flags as obstructing a merge.
*/
vm_flags = vma->vm_flags & ~(VM_READ|VM_WRITE|VM_EXEC);
vm_flags |= near->vm_flags & (VM_READ|VM_WRITE|VM_EXEC);
if (near->anon_vma && vma->vm_end == near->vm_start &&
mpol_equal(vma_policy(vma), vma_policy(near)) &&
can_vma_merge_before(near, vm_flags,
NULL, vma->vm_file, vma->vm_pgoff +
((vma->vm_end - vma->vm_start) >> PAGE_SHIFT)))
return near->anon_vma;
try_prev:
/*
* It is potentially slow to have to call find_vma_prev here.
* But it's only on the first write fault on the vma, not
* every time, and we could devise a way to avoid it later
* (e.g. stash info in next's anon_vma_node when assigning
* an anon_vma, or when trying vma_merge). Another time.
*/
BUG_ON(find_vma_prev(vma->vm_mm, vma->vm_start, &near) != vma);
if (!near)
goto none;
vm_flags = vma->vm_flags & ~(VM_READ|VM_WRITE|VM_EXEC);
vm_flags |= near->vm_flags & (VM_READ|VM_WRITE|VM_EXEC);
if (near->anon_vma && near->vm_end == vma->vm_start &&
mpol_equal(vma_policy(near), vma_policy(vma)) &&
can_vma_merge_after(near, vm_flags,
NULL, vma->vm_file, vma->vm_pgoff))
return near->anon_vma;
none:
/*
* There's no absolute need to look only at touching neighbours:
* we could search further afield for "compatible" anon_vmas.
* But it would probably just be a waste of time searching,
* or lead to too many vmas hanging off the same anon_vma.
* We're trying to allow mprotect remerging later on,
* not trying to minimize memory used for anon_vmas.
*/
return NULL;
}
#ifdef CONFIG_PROC_FS
void vm_stat_account(struct mm_struct *mm, unsigned long flags,
struct file *file, long pages)
{
const unsigned long stack_flags
= VM_STACK_FLAGS & (VM_GROWSUP|VM_GROWSDOWN);
if (file) {
mm->shared_vm += pages;
if ((flags & (VM_EXEC|VM_WRITE)) == VM_EXEC)
mm->exec_vm += pages;
} else if (flags & stack_flags)
mm->stack_vm += pages;
if (flags & (VM_RESERVED|VM_IO))
mm->reserved_vm += pages;
}
#endif /* CONFIG_PROC_FS */
/*
* The caller must hold down_write(current->mm->mmap_sem).
*/
unsigned long do_mmap_pgoff(struct file * file, unsigned long addr,
unsigned long len, unsigned long prot,
unsigned long flags, unsigned long pgoff)
{
struct mm_struct * mm = current->mm;
struct vm_area_struct * vma, * prev;
struct inode *inode;
unsigned int vm_flags;
int correct_wcount = 0;
int error;
struct rb_node ** rb_link, * rb_parent;
int accountable = 1;
unsigned long charged = 0, reqprot = prot;
if (file) {
if (is_file_hugepages(file))
accountable = 0;
if (!file->f_op || !file->f_op->mmap)
return -ENODEV;
if ((prot & PROT_EXEC) &&
(file->f_vfsmnt->mnt_flags & MNT_NOEXEC))
return -EPERM;
}
/*
* Does the application expect PROT_READ to imply PROT_EXEC?
*
* (the exception is when the underlying filesystem is noexec
* mounted, in which case we dont add PROT_EXEC.)
*/
if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
if (!(file && (file->f_vfsmnt->mnt_flags & MNT_NOEXEC)))
prot |= PROT_EXEC;
if (!len)
return -EINVAL;
error = arch_mmap_check(addr, len, flags);
if (error)
return error;
/* Careful about overflows.. */
len = PAGE_ALIGN(len);
if (!len || len > TASK_SIZE)
return -ENOMEM;
/* offset overflow? */
if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
return -EOVERFLOW;
/* Too many mappings? */
if (mm->map_count > sysctl_max_map_count)
return -ENOMEM;
/* Obtain the address to map to. we verify (or select) it and ensure
* that it represents a valid section of the address space.
*/
addr = get_unmapped_area(file, addr, len, pgoff, flags);
if (addr & ~PAGE_MASK)
return addr;
/* Do simple checking here so the lower-level routines won't have
* to. we assume access permissions have been handled by the open
* of the memory object, so we don't do any here.
*/
vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags) |
mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
if (flags & MAP_LOCKED) {
if (!can_do_mlock())
return -EPERM;
vm_flags |= VM_LOCKED;
}
/* mlock MCL_FUTURE? */
if (vm_flags & VM_LOCKED) {
unsigned long locked, lock_limit;
locked = len >> PAGE_SHIFT;
locked += mm->locked_vm;
lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur;
lock_limit >>= PAGE_SHIFT;
if (locked > lock_limit && !capable(CAP_IPC_LOCK))
return -EAGAIN;
}
inode = file ? file->f_dentry->d_inode : NULL;
if (file) {
switch (flags & MAP_TYPE) {
case MAP_SHARED:
if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE))
return -EACCES;
/*
* Make sure we don't allow writing to an append-only
* file..
*/
if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE))
return -EACCES;
/*
* Make sure there are no mandatory locks on the file.
*/
if (locks_verify_locked(inode))
return -EAGAIN;
vm_flags |= VM_SHARED | VM_MAYSHARE;
if (!(file->f_mode & FMODE_WRITE))
vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
/* fall through */
case MAP_PRIVATE:
if (!(file->f_mode & FMODE_READ))
return -EACCES;
break;
default:
return -EINVAL;
}
} else {
switch (flags & MAP_TYPE) {
case MAP_SHARED:
vm_flags |= VM_SHARED | VM_MAYSHARE;
break;
case MAP_PRIVATE:
/*
* Set pgoff according to addr for anon_vma.
*/
pgoff = addr >> PAGE_SHIFT;
break;
default:
return -EINVAL;
}
}
error = security_file_mmap(file, reqprot, prot, flags);
if (error)
return error;
/* Clear old maps */
error = -ENOMEM;
munmap_back:
vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
if (vma && vma->vm_start < addr + len) {
if (do_munmap(mm, addr, len))
return -ENOMEM;
goto munmap_back;
}
/* Check against address space limit. */
if (!may_expand_vm(mm, len >> PAGE_SHIFT))
return -ENOMEM;
if (accountable && (!(flags & MAP_NORESERVE) ||
sysctl_overcommit_memory == OVERCOMMIT_NEVER)) {
if (vm_flags & VM_SHARED) {
/* Check memory availability in shmem_file_setup? */
vm_flags |= VM_ACCOUNT;
} else if (vm_flags & VM_WRITE) {
/*
* Private writable mapping: check memory availability
*/
charged = len >> PAGE_SHIFT;
if (security_vm_enough_memory(charged))
return -ENOMEM;
vm_flags |= VM_ACCOUNT;
}
}
/*
* Can we just expand an old private anonymous mapping?
* The VM_SHARED test is necessary because shmem_zero_setup
* will create the file object for a shared anonymous map below.
*/
if (!file && !(vm_flags & VM_SHARED) &&
vma_merge(mm, prev, addr, addr + len, vm_flags,
NULL, NULL, pgoff, NULL))
goto out;
/*
* Determine the object being mapped and call the appropriate
* specific mapper. the address has already been validated, but
* not unmapped, but the maps are removed from the list.
*/
vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
if (!vma) {
error = -ENOMEM;
goto unacct_error;
}
vma->vm_mm = mm;
vma->vm_start = addr;
vma->vm_end = addr + len;
vma->vm_flags = vm_flags;
[PATCH] add page_mkwrite() vm_operations method Add a new VMA operation to notify a filesystem or other driver about the MMU generating a fault because userspace attempted to write to a page mapped through a read-only PTE. This facility permits the filesystem or driver to: (*) Implement storage allocation/reservation on attempted write, and so to deal with problems such as ENOSPC more gracefully (perhaps by generating SIGBUS). (*) Delay making the page writable until the contents have been written to a backing cache. This is useful for NFS/AFS when using FS-Cache/CacheFS. It permits the filesystem to have some guarantee about the state of the cache. (*) Account and limit number of dirty pages. This is one piece of the puzzle needed to make shared writable mapping work safely in FUSE. Needed by cachefs (Or is it cachefiles? Or fscache? <head spins>). At least four other groups have stated an interest in it or a desire to use the functionality it provides: FUSE, OCFS2, NTFS and JFFS2. Also, things like EXT3 really ought to use it to deal with the case of shared-writable mmap encountering ENOSPC before we permit the page to be dirtied. From: Peter Zijlstra <a.p.zijlstra@chello.nl> get_user_pages(.write=1, .force=1) can generate COW hits on read-only shared mappings, this patch traps those as mkpage_write candidates and fails to handle them the old way. Signed-off-by: David Howells <dhowells@redhat.com> Cc: Miklos Szeredi <miklos@szeredi.hu> Cc: Joel Becker <Joel.Becker@oracle.com> Cc: Mark Fasheh <mark.fasheh@oracle.com> Cc: Anton Altaparmakov <aia21@cantab.net> Cc: David Woodhouse <dwmw2@infradead.org> Cc: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-06-23 09:03:43 +00:00
vma->vm_page_prot = protection_map[vm_flags &
(VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)];
vma->vm_pgoff = pgoff;
if (file) {
error = -EINVAL;
if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
goto free_vma;
if (vm_flags & VM_DENYWRITE) {
error = deny_write_access(file);
if (error)
goto free_vma;
correct_wcount = 1;
}
vma->vm_file = file;
get_file(file);
error = file->f_op->mmap(file, vma);
if (error)
goto unmap_and_free_vma;
} else if (vm_flags & VM_SHARED) {
error = shmem_zero_setup(vma);
if (error)
goto free_vma;
}
[PATCH] add page_mkwrite() vm_operations method Add a new VMA operation to notify a filesystem or other driver about the MMU generating a fault because userspace attempted to write to a page mapped through a read-only PTE. This facility permits the filesystem or driver to: (*) Implement storage allocation/reservation on attempted write, and so to deal with problems such as ENOSPC more gracefully (perhaps by generating SIGBUS). (*) Delay making the page writable until the contents have been written to a backing cache. This is useful for NFS/AFS when using FS-Cache/CacheFS. It permits the filesystem to have some guarantee about the state of the cache. (*) Account and limit number of dirty pages. This is one piece of the puzzle needed to make shared writable mapping work safely in FUSE. Needed by cachefs (Or is it cachefiles? Or fscache? <head spins>). At least four other groups have stated an interest in it or a desire to use the functionality it provides: FUSE, OCFS2, NTFS and JFFS2. Also, things like EXT3 really ought to use it to deal with the case of shared-writable mmap encountering ENOSPC before we permit the page to be dirtied. From: Peter Zijlstra <a.p.zijlstra@chello.nl> get_user_pages(.write=1, .force=1) can generate COW hits on read-only shared mappings, this patch traps those as mkpage_write candidates and fails to handle them the old way. Signed-off-by: David Howells <dhowells@redhat.com> Cc: Miklos Szeredi <miklos@szeredi.hu> Cc: Joel Becker <Joel.Becker@oracle.com> Cc: Mark Fasheh <mark.fasheh@oracle.com> Cc: Anton Altaparmakov <aia21@cantab.net> Cc: David Woodhouse <dwmw2@infradead.org> Cc: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-06-23 09:03:43 +00:00
/* Don't make the VMA automatically writable if it's shared, but the
* backer wishes to know when pages are first written to */
if (vma->vm_ops && vma->vm_ops->page_mkwrite)
vma->vm_page_prot =
protection_map[vm_flags & (VM_READ|VM_WRITE|VM_EXEC)];
/* We set VM_ACCOUNT in a shared mapping's vm_flags, to inform
* shmem_zero_setup (perhaps called through /dev/zero's ->mmap)
* that memory reservation must be checked; but that reservation
* belongs to shared memory object, not to vma: so now clear it.
*/
if ((vm_flags & (VM_SHARED|VM_ACCOUNT)) == (VM_SHARED|VM_ACCOUNT))
vma->vm_flags &= ~VM_ACCOUNT;
/* Can addr have changed??
*
* Answer: Yes, several device drivers can do it in their
* f_op->mmap method. -DaveM
*/
addr = vma->vm_start;
pgoff = vma->vm_pgoff;
vm_flags = vma->vm_flags;
if (!file || !vma_merge(mm, prev, addr, vma->vm_end,
vma->vm_flags, NULL, file, pgoff, vma_policy(vma))) {
file = vma->vm_file;
vma_link(mm, vma, prev, rb_link, rb_parent);
if (correct_wcount)
atomic_inc(&inode->i_writecount);
} else {
if (file) {
if (correct_wcount)
atomic_inc(&inode->i_writecount);
fput(file);
}
mpol_free(vma_policy(vma));
kmem_cache_free(vm_area_cachep, vma);
}
out:
mm->total_vm += len >> PAGE_SHIFT;
vm_stat_account(mm, vm_flags, file, len >> PAGE_SHIFT);
if (vm_flags & VM_LOCKED) {
mm->locked_vm += len >> PAGE_SHIFT;
make_pages_present(addr, addr + len);
}
if (flags & MAP_POPULATE) {
up_write(&mm->mmap_sem);
sys_remap_file_pages(addr, len, 0,
pgoff, flags & MAP_NONBLOCK);
down_write(&mm->mmap_sem);
}
return addr;
unmap_and_free_vma:
if (correct_wcount)
atomic_inc(&inode->i_writecount);
vma->vm_file = NULL;
fput(file);
/* Undo any partial mapping done by a device driver. */
[PATCH] freepgt: free_pgtables use vma list Recent woes with some arches needing their own pgd_addr_end macro; and 4-level clear_page_range regression since 2.6.10's clear_page_tables; and its long-standing well-known inefficiency in searching throughout the higher-level page tables for those few entries to clear and free: all can be blamed on ignoring the list of vmas when we free page tables. Replace exit_mmap's clear_page_range of the total user address space by free_pgtables operating on the mm's vma list; unmap_region use it in the same way, giving floor and ceiling beyond which it may not free tables. This brings lmbench fork/exec/sh numbers back to 2.6.10 (unless preempt is enabled, in which case latency fixes spoil unmap_vmas throughput). Beware: the do_mmap_pgoff driver failure case must now use unmap_region instead of zap_page_range, since a page table might have been allocated, and can only be freed while it is touched by some vma. Move free_pgtables from mmap.c to memory.c, where its lower levels are adapted from the clear_page_range levels. (Most of free_pgtables' old code was actually for a non-existent case, prev not properly set up, dating from before hch gave us split_vma.) Pass mmu_gather** in the public interfaces, since we might want to add latency lockdrops later; but no attempt to do so yet, going by vma should itself reduce latency. But what if is_hugepage_only_range? Those ia64 and ppc64 cases need careful examination: put that off until a later patch of the series. What of x86_64's 32bit vdso page __map_syscall32 maps outside any vma? And the range to sparc64's flush_tlb_pgtables? It's less clear to me now that we need to do more than is done here - every PMD_SIZE ever occupied will be flushed, do we really have to flush every PGDIR_SIZE ever partially occupied? A shame to complicate it unnecessarily. Special thanks to David Miller for time spent repairing my ceilings. Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-04-19 20:29:15 +00:00
unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end);
charged = 0;
free_vma:
kmem_cache_free(vm_area_cachep, vma);
unacct_error:
if (charged)
vm_unacct_memory(charged);
return error;
}
EXPORT_SYMBOL(do_mmap_pgoff);
/* Get an address range which is currently unmapped.
* For shmat() with addr=0.
*
* Ugly calling convention alert:
* Return value with the low bits set means error value,
* ie
* if (ret & ~PAGE_MASK)
* error = ret;
*
* This function "knows" that -ENOMEM has the bits set.
*/
#ifndef HAVE_ARCH_UNMAPPED_AREA
unsigned long
arch_get_unmapped_area(struct file *filp, unsigned long addr,
unsigned long len, unsigned long pgoff, unsigned long flags)
{
struct mm_struct *mm = current->mm;
struct vm_area_struct *vma;
unsigned long start_addr;
if (len > TASK_SIZE)
return -ENOMEM;
if (addr) {
addr = PAGE_ALIGN(addr);
vma = find_vma(mm, addr);
if (TASK_SIZE - len >= addr &&
(!vma || addr + len <= vma->vm_start))
return addr;
}
[PATCH] Avoiding mmap fragmentation Ingo recently introduced a great speedup for allocating new mmaps using the free_area_cache pointer which boosts the specweb SSL benchmark by 4-5% and causes huge performance increases in thread creation. The downside of this patch is that it does lead to fragmentation in the mmap-ed areas (visible via /proc/self/maps), such that some applications that work fine under 2.4 kernels quickly run out of memory on any 2.6 kernel. The problem is twofold: 1) the free_area_cache is used to continue a search for memory where the last search ended. Before the change new areas were always searched from the base address on. So now new small areas are cluttering holes of all sizes throughout the whole mmap-able region whereas before small holes tended to close holes near the base leaving holes far from the base large and available for larger requests. 2) the free_area_cache also is set to the location of the last munmap-ed area so in scenarios where we allocate e.g. five regions of 1K each, then free regions 4 2 3 in this order the next request for 1K will be placed in the position of the old region 3, whereas before we appended it to the still active region 1, placing it at the location of the old region 2. Before we had 1 free region of 2K, now we only get two free regions of 1K -> fragmentation. The patch addresses thes issues by introducing yet another cache descriptor cached_hole_size that contains the largest known hole size below the current free_area_cache. If a new request comes in the size is compared against the cached_hole_size and if the request can be filled with a hole below free_area_cache the search is started from the base instead. The results look promising: Whereas 2.6.12-rc4 fragments quickly and my (earlier posted) leakme.c test program terminates after 50000+ iterations with 96 distinct and fragmented maps in /proc/self/maps it performs nicely (as expected) with thread creation, Ingo's test_str02 with 20000 threads requires 0.7s system time. Taking out Ingo's patch (un-patch available per request) by basically deleting all mentions of free_area_cache from the kernel and starting the search for new memory always at the respective bases we observe: leakme terminates successfully with 11 distinctive hardly fragmented areas in /proc/self/maps but thread creating is gringdingly slow: 30+s(!) system time for Ingo's test_str02 with 20000 threads. Now - drumroll ;-) the appended patch works fine with leakme: it ends with only 7 distinct areas in /proc/self/maps and also thread creation seems sufficiently fast with 0.71s for 20000 threads. Signed-off-by: Wolfgang Wander <wwc@rentec.com> Credit-to: "Richard Purdie" <rpurdie@rpsys.net> Signed-off-by: Ken Chen <kenneth.w.chen@intel.com> Acked-by: Ingo Molnar <mingo@elte.hu> (partly) Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-06-22 00:14:49 +00:00
if (len > mm->cached_hole_size) {
start_addr = addr = mm->free_area_cache;
} else {
start_addr = addr = TASK_UNMAPPED_BASE;
mm->cached_hole_size = 0;
}
full_search:
for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
/* At this point: (!vma || addr < vma->vm_end). */
if (TASK_SIZE - len < addr) {
/*
* Start a new search - just in case we missed
* some holes.
*/
if (start_addr != TASK_UNMAPPED_BASE) {
[PATCH] Avoiding mmap fragmentation Ingo recently introduced a great speedup for allocating new mmaps using the free_area_cache pointer which boosts the specweb SSL benchmark by 4-5% and causes huge performance increases in thread creation. The downside of this patch is that it does lead to fragmentation in the mmap-ed areas (visible via /proc/self/maps), such that some applications that work fine under 2.4 kernels quickly run out of memory on any 2.6 kernel. The problem is twofold: 1) the free_area_cache is used to continue a search for memory where the last search ended. Before the change new areas were always searched from the base address on. So now new small areas are cluttering holes of all sizes throughout the whole mmap-able region whereas before small holes tended to close holes near the base leaving holes far from the base large and available for larger requests. 2) the free_area_cache also is set to the location of the last munmap-ed area so in scenarios where we allocate e.g. five regions of 1K each, then free regions 4 2 3 in this order the next request for 1K will be placed in the position of the old region 3, whereas before we appended it to the still active region 1, placing it at the location of the old region 2. Before we had 1 free region of 2K, now we only get two free regions of 1K -> fragmentation. The patch addresses thes issues by introducing yet another cache descriptor cached_hole_size that contains the largest known hole size below the current free_area_cache. If a new request comes in the size is compared against the cached_hole_size and if the request can be filled with a hole below free_area_cache the search is started from the base instead. The results look promising: Whereas 2.6.12-rc4 fragments quickly and my (earlier posted) leakme.c test program terminates after 50000+ iterations with 96 distinct and fragmented maps in /proc/self/maps it performs nicely (as expected) with thread creation, Ingo's test_str02 with 20000 threads requires 0.7s system time. Taking out Ingo's patch (un-patch available per request) by basically deleting all mentions of free_area_cache from the kernel and starting the search for new memory always at the respective bases we observe: leakme terminates successfully with 11 distinctive hardly fragmented areas in /proc/self/maps but thread creating is gringdingly slow: 30+s(!) system time for Ingo's test_str02 with 20000 threads. Now - drumroll ;-) the appended patch works fine with leakme: it ends with only 7 distinct areas in /proc/self/maps and also thread creation seems sufficiently fast with 0.71s for 20000 threads. Signed-off-by: Wolfgang Wander <wwc@rentec.com> Credit-to: "Richard Purdie" <rpurdie@rpsys.net> Signed-off-by: Ken Chen <kenneth.w.chen@intel.com> Acked-by: Ingo Molnar <mingo@elte.hu> (partly) Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-06-22 00:14:49 +00:00
addr = TASK_UNMAPPED_BASE;
start_addr = addr;
mm->cached_hole_size = 0;
goto full_search;
}
return -ENOMEM;
}
if (!vma || addr + len <= vma->vm_start) {
/*
* Remember the place where we stopped the search:
*/
mm->free_area_cache = addr + len;
return addr;
}
[PATCH] Avoiding mmap fragmentation Ingo recently introduced a great speedup for allocating new mmaps using the free_area_cache pointer which boosts the specweb SSL benchmark by 4-5% and causes huge performance increases in thread creation. The downside of this patch is that it does lead to fragmentation in the mmap-ed areas (visible via /proc/self/maps), such that some applications that work fine under 2.4 kernels quickly run out of memory on any 2.6 kernel. The problem is twofold: 1) the free_area_cache is used to continue a search for memory where the last search ended. Before the change new areas were always searched from the base address on. So now new small areas are cluttering holes of all sizes throughout the whole mmap-able region whereas before small holes tended to close holes near the base leaving holes far from the base large and available for larger requests. 2) the free_area_cache also is set to the location of the last munmap-ed area so in scenarios where we allocate e.g. five regions of 1K each, then free regions 4 2 3 in this order the next request for 1K will be placed in the position of the old region 3, whereas before we appended it to the still active region 1, placing it at the location of the old region 2. Before we had 1 free region of 2K, now we only get two free regions of 1K -> fragmentation. The patch addresses thes issues by introducing yet another cache descriptor cached_hole_size that contains the largest known hole size below the current free_area_cache. If a new request comes in the size is compared against the cached_hole_size and if the request can be filled with a hole below free_area_cache the search is started from the base instead. The results look promising: Whereas 2.6.12-rc4 fragments quickly and my (earlier posted) leakme.c test program terminates after 50000+ iterations with 96 distinct and fragmented maps in /proc/self/maps it performs nicely (as expected) with thread creation, Ingo's test_str02 with 20000 threads requires 0.7s system time. Taking out Ingo's patch (un-patch available per request) by basically deleting all mentions of free_area_cache from the kernel and starting the search for new memory always at the respective bases we observe: leakme terminates successfully with 11 distinctive hardly fragmented areas in /proc/self/maps but thread creating is gringdingly slow: 30+s(!) system time for Ingo's test_str02 with 20000 threads. Now - drumroll ;-) the appended patch works fine with leakme: it ends with only 7 distinct areas in /proc/self/maps and also thread creation seems sufficiently fast with 0.71s for 20000 threads. Signed-off-by: Wolfgang Wander <wwc@rentec.com> Credit-to: "Richard Purdie" <rpurdie@rpsys.net> Signed-off-by: Ken Chen <kenneth.w.chen@intel.com> Acked-by: Ingo Molnar <mingo@elte.hu> (partly) Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-06-22 00:14:49 +00:00
if (addr + mm->cached_hole_size < vma->vm_start)
mm->cached_hole_size = vma->vm_start - addr;
addr = vma->vm_end;
}
}
#endif
[PATCH] Avoiding mmap fragmentation Ingo recently introduced a great speedup for allocating new mmaps using the free_area_cache pointer which boosts the specweb SSL benchmark by 4-5% and causes huge performance increases in thread creation. The downside of this patch is that it does lead to fragmentation in the mmap-ed areas (visible via /proc/self/maps), such that some applications that work fine under 2.4 kernels quickly run out of memory on any 2.6 kernel. The problem is twofold: 1) the free_area_cache is used to continue a search for memory where the last search ended. Before the change new areas were always searched from the base address on. So now new small areas are cluttering holes of all sizes throughout the whole mmap-able region whereas before small holes tended to close holes near the base leaving holes far from the base large and available for larger requests. 2) the free_area_cache also is set to the location of the last munmap-ed area so in scenarios where we allocate e.g. five regions of 1K each, then free regions 4 2 3 in this order the next request for 1K will be placed in the position of the old region 3, whereas before we appended it to the still active region 1, placing it at the location of the old region 2. Before we had 1 free region of 2K, now we only get two free regions of 1K -> fragmentation. The patch addresses thes issues by introducing yet another cache descriptor cached_hole_size that contains the largest known hole size below the current free_area_cache. If a new request comes in the size is compared against the cached_hole_size and if the request can be filled with a hole below free_area_cache the search is started from the base instead. The results look promising: Whereas 2.6.12-rc4 fragments quickly and my (earlier posted) leakme.c test program terminates after 50000+ iterations with 96 distinct and fragmented maps in /proc/self/maps it performs nicely (as expected) with thread creation, Ingo's test_str02 with 20000 threads requires 0.7s system time. Taking out Ingo's patch (un-patch available per request) by basically deleting all mentions of free_area_cache from the kernel and starting the search for new memory always at the respective bases we observe: leakme terminates successfully with 11 distinctive hardly fragmented areas in /proc/self/maps but thread creating is gringdingly slow: 30+s(!) system time for Ingo's test_str02 with 20000 threads. Now - drumroll ;-) the appended patch works fine with leakme: it ends with only 7 distinct areas in /proc/self/maps and also thread creation seems sufficiently fast with 0.71s for 20000 threads. Signed-off-by: Wolfgang Wander <wwc@rentec.com> Credit-to: "Richard Purdie" <rpurdie@rpsys.net> Signed-off-by: Ken Chen <kenneth.w.chen@intel.com> Acked-by: Ingo Molnar <mingo@elte.hu> (partly) Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-06-22 00:14:49 +00:00
void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
{
/*
* Is this a new hole at the lowest possible address?
*/
[PATCH] Avoiding mmap fragmentation Ingo recently introduced a great speedup for allocating new mmaps using the free_area_cache pointer which boosts the specweb SSL benchmark by 4-5% and causes huge performance increases in thread creation. The downside of this patch is that it does lead to fragmentation in the mmap-ed areas (visible via /proc/self/maps), such that some applications that work fine under 2.4 kernels quickly run out of memory on any 2.6 kernel. The problem is twofold: 1) the free_area_cache is used to continue a search for memory where the last search ended. Before the change new areas were always searched from the base address on. So now new small areas are cluttering holes of all sizes throughout the whole mmap-able region whereas before small holes tended to close holes near the base leaving holes far from the base large and available for larger requests. 2) the free_area_cache also is set to the location of the last munmap-ed area so in scenarios where we allocate e.g. five regions of 1K each, then free regions 4 2 3 in this order the next request for 1K will be placed in the position of the old region 3, whereas before we appended it to the still active region 1, placing it at the location of the old region 2. Before we had 1 free region of 2K, now we only get two free regions of 1K -> fragmentation. The patch addresses thes issues by introducing yet another cache descriptor cached_hole_size that contains the largest known hole size below the current free_area_cache. If a new request comes in the size is compared against the cached_hole_size and if the request can be filled with a hole below free_area_cache the search is started from the base instead. The results look promising: Whereas 2.6.12-rc4 fragments quickly and my (earlier posted) leakme.c test program terminates after 50000+ iterations with 96 distinct and fragmented maps in /proc/self/maps it performs nicely (as expected) with thread creation, Ingo's test_str02 with 20000 threads requires 0.7s system time. Taking out Ingo's patch (un-patch available per request) by basically deleting all mentions of free_area_cache from the kernel and starting the search for new memory always at the respective bases we observe: leakme terminates successfully with 11 distinctive hardly fragmented areas in /proc/self/maps but thread creating is gringdingly slow: 30+s(!) system time for Ingo's test_str02 with 20000 threads. Now - drumroll ;-) the appended patch works fine with leakme: it ends with only 7 distinct areas in /proc/self/maps and also thread creation seems sufficiently fast with 0.71s for 20000 threads. Signed-off-by: Wolfgang Wander <wwc@rentec.com> Credit-to: "Richard Purdie" <rpurdie@rpsys.net> Signed-off-by: Ken Chen <kenneth.w.chen@intel.com> Acked-by: Ingo Molnar <mingo@elte.hu> (partly) Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-06-22 00:14:49 +00:00
if (addr >= TASK_UNMAPPED_BASE && addr < mm->free_area_cache) {
mm->free_area_cache = addr;
mm->cached_hole_size = ~0UL;
}
}
/*
* This mmap-allocator allocates new areas top-down from below the
* stack's low limit (the base):
*/
#ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
unsigned long
arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
const unsigned long len, const unsigned long pgoff,
const unsigned long flags)
{
struct vm_area_struct *vma;
struct mm_struct *mm = current->mm;
unsigned long addr = addr0;
/* requested length too big for entire address space */
if (len > TASK_SIZE)
return -ENOMEM;
/* requesting a specific address */
if (addr) {
addr = PAGE_ALIGN(addr);
vma = find_vma(mm, addr);
if (TASK_SIZE - len >= addr &&
(!vma || addr + len <= vma->vm_start))
return addr;
}
[PATCH] Avoiding mmap fragmentation Ingo recently introduced a great speedup for allocating new mmaps using the free_area_cache pointer which boosts the specweb SSL benchmark by 4-5% and causes huge performance increases in thread creation. The downside of this patch is that it does lead to fragmentation in the mmap-ed areas (visible via /proc/self/maps), such that some applications that work fine under 2.4 kernels quickly run out of memory on any 2.6 kernel. The problem is twofold: 1) the free_area_cache is used to continue a search for memory where the last search ended. Before the change new areas were always searched from the base address on. So now new small areas are cluttering holes of all sizes throughout the whole mmap-able region whereas before small holes tended to close holes near the base leaving holes far from the base large and available for larger requests. 2) the free_area_cache also is set to the location of the last munmap-ed area so in scenarios where we allocate e.g. five regions of 1K each, then free regions 4 2 3 in this order the next request for 1K will be placed in the position of the old region 3, whereas before we appended it to the still active region 1, placing it at the location of the old region 2. Before we had 1 free region of 2K, now we only get two free regions of 1K -> fragmentation. The patch addresses thes issues by introducing yet another cache descriptor cached_hole_size that contains the largest known hole size below the current free_area_cache. If a new request comes in the size is compared against the cached_hole_size and if the request can be filled with a hole below free_area_cache the search is started from the base instead. The results look promising: Whereas 2.6.12-rc4 fragments quickly and my (earlier posted) leakme.c test program terminates after 50000+ iterations with 96 distinct and fragmented maps in /proc/self/maps it performs nicely (as expected) with thread creation, Ingo's test_str02 with 20000 threads requires 0.7s system time. Taking out Ingo's patch (un-patch available per request) by basically deleting all mentions of free_area_cache from the kernel and starting the search for new memory always at the respective bases we observe: leakme terminates successfully with 11 distinctive hardly fragmented areas in /proc/self/maps but thread creating is gringdingly slow: 30+s(!) system time for Ingo's test_str02 with 20000 threads. Now - drumroll ;-) the appended patch works fine with leakme: it ends with only 7 distinct areas in /proc/self/maps and also thread creation seems sufficiently fast with 0.71s for 20000 threads. Signed-off-by: Wolfgang Wander <wwc@rentec.com> Credit-to: "Richard Purdie" <rpurdie@rpsys.net> Signed-off-by: Ken Chen <kenneth.w.chen@intel.com> Acked-by: Ingo Molnar <mingo@elte.hu> (partly) Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-06-22 00:14:49 +00:00
/* check if free_area_cache is useful for us */
if (len <= mm->cached_hole_size) {
mm->cached_hole_size = 0;
mm->free_area_cache = mm->mmap_base;
}
/* either no address requested or can't fit in requested address hole */
addr = mm->free_area_cache;
/* make sure it can fit in the remaining address space */
if (addr > len) {
vma = find_vma(mm, addr-len);
if (!vma || addr <= vma->vm_start)
/* remember the address as a hint for next time */
return (mm->free_area_cache = addr-len);
}
if (mm->mmap_base < len)
goto bottomup;
addr = mm->mmap_base-len;
do {
/*
* Lookup failure means no vma is above this address,
* else if new region fits below vma->vm_start,
* return with success:
*/
vma = find_vma(mm, addr);
if (!vma || addr+len <= vma->vm_start)
/* remember the address as a hint for next time */
return (mm->free_area_cache = addr);
[PATCH] Avoiding mmap fragmentation Ingo recently introduced a great speedup for allocating new mmaps using the free_area_cache pointer which boosts the specweb SSL benchmark by 4-5% and causes huge performance increases in thread creation. The downside of this patch is that it does lead to fragmentation in the mmap-ed areas (visible via /proc/self/maps), such that some applications that work fine under 2.4 kernels quickly run out of memory on any 2.6 kernel. The problem is twofold: 1) the free_area_cache is used to continue a search for memory where the last search ended. Before the change new areas were always searched from the base address on. So now new small areas are cluttering holes of all sizes throughout the whole mmap-able region whereas before small holes tended to close holes near the base leaving holes far from the base large and available for larger requests. 2) the free_area_cache also is set to the location of the last munmap-ed area so in scenarios where we allocate e.g. five regions of 1K each, then free regions 4 2 3 in this order the next request for 1K will be placed in the position of the old region 3, whereas before we appended it to the still active region 1, placing it at the location of the old region 2. Before we had 1 free region of 2K, now we only get two free regions of 1K -> fragmentation. The patch addresses thes issues by introducing yet another cache descriptor cached_hole_size that contains the largest known hole size below the current free_area_cache. If a new request comes in the size is compared against the cached_hole_size and if the request can be filled with a hole below free_area_cache the search is started from the base instead. The results look promising: Whereas 2.6.12-rc4 fragments quickly and my (earlier posted) leakme.c test program terminates after 50000+ iterations with 96 distinct and fragmented maps in /proc/self/maps it performs nicely (as expected) with thread creation, Ingo's test_str02 with 20000 threads requires 0.7s system time. Taking out Ingo's patch (un-patch available per request) by basically deleting all mentions of free_area_cache from the kernel and starting the search for new memory always at the respective bases we observe: leakme terminates successfully with 11 distinctive hardly fragmented areas in /proc/self/maps but thread creating is gringdingly slow: 30+s(!) system time for Ingo's test_str02 with 20000 threads. Now - drumroll ;-) the appended patch works fine with leakme: it ends with only 7 distinct areas in /proc/self/maps and also thread creation seems sufficiently fast with 0.71s for 20000 threads. Signed-off-by: Wolfgang Wander <wwc@rentec.com> Credit-to: "Richard Purdie" <rpurdie@rpsys.net> Signed-off-by: Ken Chen <kenneth.w.chen@intel.com> Acked-by: Ingo Molnar <mingo@elte.hu> (partly) Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-06-22 00:14:49 +00:00
/* remember the largest hole we saw so far */
if (addr + mm->cached_hole_size < vma->vm_start)
mm->cached_hole_size = vma->vm_start - addr;
/* try just below the current vma->vm_start */
addr = vma->vm_start-len;
} while (len < vma->vm_start);
bottomup:
/*
* A failed mmap() very likely causes application failure,
* so fall back to the bottom-up function here. This scenario
* can happen with large stack limits and large mmap()
* allocations.
*/
[PATCH] Avoiding mmap fragmentation Ingo recently introduced a great speedup for allocating new mmaps using the free_area_cache pointer which boosts the specweb SSL benchmark by 4-5% and causes huge performance increases in thread creation. The downside of this patch is that it does lead to fragmentation in the mmap-ed areas (visible via /proc/self/maps), such that some applications that work fine under 2.4 kernels quickly run out of memory on any 2.6 kernel. The problem is twofold: 1) the free_area_cache is used to continue a search for memory where the last search ended. Before the change new areas were always searched from the base address on. So now new small areas are cluttering holes of all sizes throughout the whole mmap-able region whereas before small holes tended to close holes near the base leaving holes far from the base large and available for larger requests. 2) the free_area_cache also is set to the location of the last munmap-ed area so in scenarios where we allocate e.g. five regions of 1K each, then free regions 4 2 3 in this order the next request for 1K will be placed in the position of the old region 3, whereas before we appended it to the still active region 1, placing it at the location of the old region 2. Before we had 1 free region of 2K, now we only get two free regions of 1K -> fragmentation. The patch addresses thes issues by introducing yet another cache descriptor cached_hole_size that contains the largest known hole size below the current free_area_cache. If a new request comes in the size is compared against the cached_hole_size and if the request can be filled with a hole below free_area_cache the search is started from the base instead. The results look promising: Whereas 2.6.12-rc4 fragments quickly and my (earlier posted) leakme.c test program terminates after 50000+ iterations with 96 distinct and fragmented maps in /proc/self/maps it performs nicely (as expected) with thread creation, Ingo's test_str02 with 20000 threads requires 0.7s system time. Taking out Ingo's patch (un-patch available per request) by basically deleting all mentions of free_area_cache from the kernel and starting the search for new memory always at the respective bases we observe: leakme terminates successfully with 11 distinctive hardly fragmented areas in /proc/self/maps but thread creating is gringdingly slow: 30+s(!) system time for Ingo's test_str02 with 20000 threads. Now - drumroll ;-) the appended patch works fine with leakme: it ends with only 7 distinct areas in /proc/self/maps and also thread creation seems sufficiently fast with 0.71s for 20000 threads. Signed-off-by: Wolfgang Wander <wwc@rentec.com> Credit-to: "Richard Purdie" <rpurdie@rpsys.net> Signed-off-by: Ken Chen <kenneth.w.chen@intel.com> Acked-by: Ingo Molnar <mingo@elte.hu> (partly) Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-06-22 00:14:49 +00:00
mm->cached_hole_size = ~0UL;
mm->free_area_cache = TASK_UNMAPPED_BASE;
addr = arch_get_unmapped_area(filp, addr0, len, pgoff, flags);
/*
* Restore the topdown base:
*/
mm->free_area_cache = mm->mmap_base;
[PATCH] Avoiding mmap fragmentation Ingo recently introduced a great speedup for allocating new mmaps using the free_area_cache pointer which boosts the specweb SSL benchmark by 4-5% and causes huge performance increases in thread creation. The downside of this patch is that it does lead to fragmentation in the mmap-ed areas (visible via /proc/self/maps), such that some applications that work fine under 2.4 kernels quickly run out of memory on any 2.6 kernel. The problem is twofold: 1) the free_area_cache is used to continue a search for memory where the last search ended. Before the change new areas were always searched from the base address on. So now new small areas are cluttering holes of all sizes throughout the whole mmap-able region whereas before small holes tended to close holes near the base leaving holes far from the base large and available for larger requests. 2) the free_area_cache also is set to the location of the last munmap-ed area so in scenarios where we allocate e.g. five regions of 1K each, then free regions 4 2 3 in this order the next request for 1K will be placed in the position of the old region 3, whereas before we appended it to the still active region 1, placing it at the location of the old region 2. Before we had 1 free region of 2K, now we only get two free regions of 1K -> fragmentation. The patch addresses thes issues by introducing yet another cache descriptor cached_hole_size that contains the largest known hole size below the current free_area_cache. If a new request comes in the size is compared against the cached_hole_size and if the request can be filled with a hole below free_area_cache the search is started from the base instead. The results look promising: Whereas 2.6.12-rc4 fragments quickly and my (earlier posted) leakme.c test program terminates after 50000+ iterations with 96 distinct and fragmented maps in /proc/self/maps it performs nicely (as expected) with thread creation, Ingo's test_str02 with 20000 threads requires 0.7s system time. Taking out Ingo's patch (un-patch available per request) by basically deleting all mentions of free_area_cache from the kernel and starting the search for new memory always at the respective bases we observe: leakme terminates successfully with 11 distinctive hardly fragmented areas in /proc/self/maps but thread creating is gringdingly slow: 30+s(!) system time for Ingo's test_str02 with 20000 threads. Now - drumroll ;-) the appended patch works fine with leakme: it ends with only 7 distinct areas in /proc/self/maps and also thread creation seems sufficiently fast with 0.71s for 20000 threads. Signed-off-by: Wolfgang Wander <wwc@rentec.com> Credit-to: "Richard Purdie" <rpurdie@rpsys.net> Signed-off-by: Ken Chen <kenneth.w.chen@intel.com> Acked-by: Ingo Molnar <mingo@elte.hu> (partly) Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-06-22 00:14:49 +00:00
mm->cached_hole_size = ~0UL;
return addr;
}
#endif
[PATCH] Avoiding mmap fragmentation Ingo recently introduced a great speedup for allocating new mmaps using the free_area_cache pointer which boosts the specweb SSL benchmark by 4-5% and causes huge performance increases in thread creation. The downside of this patch is that it does lead to fragmentation in the mmap-ed areas (visible via /proc/self/maps), such that some applications that work fine under 2.4 kernels quickly run out of memory on any 2.6 kernel. The problem is twofold: 1) the free_area_cache is used to continue a search for memory where the last search ended. Before the change new areas were always searched from the base address on. So now new small areas are cluttering holes of all sizes throughout the whole mmap-able region whereas before small holes tended to close holes near the base leaving holes far from the base large and available for larger requests. 2) the free_area_cache also is set to the location of the last munmap-ed area so in scenarios where we allocate e.g. five regions of 1K each, then free regions 4 2 3 in this order the next request for 1K will be placed in the position of the old region 3, whereas before we appended it to the still active region 1, placing it at the location of the old region 2. Before we had 1 free region of 2K, now we only get two free regions of 1K -> fragmentation. The patch addresses thes issues by introducing yet another cache descriptor cached_hole_size that contains the largest known hole size below the current free_area_cache. If a new request comes in the size is compared against the cached_hole_size and if the request can be filled with a hole below free_area_cache the search is started from the base instead. The results look promising: Whereas 2.6.12-rc4 fragments quickly and my (earlier posted) leakme.c test program terminates after 50000+ iterations with 96 distinct and fragmented maps in /proc/self/maps it performs nicely (as expected) with thread creation, Ingo's test_str02 with 20000 threads requires 0.7s system time. Taking out Ingo's patch (un-patch available per request) by basically deleting all mentions of free_area_cache from the kernel and starting the search for new memory always at the respective bases we observe: leakme terminates successfully with 11 distinctive hardly fragmented areas in /proc/self/maps but thread creating is gringdingly slow: 30+s(!) system time for Ingo's test_str02 with 20000 threads. Now - drumroll ;-) the appended patch works fine with leakme: it ends with only 7 distinct areas in /proc/self/maps and also thread creation seems sufficiently fast with 0.71s for 20000 threads. Signed-off-by: Wolfgang Wander <wwc@rentec.com> Credit-to: "Richard Purdie" <rpurdie@rpsys.net> Signed-off-by: Ken Chen <kenneth.w.chen@intel.com> Acked-by: Ingo Molnar <mingo@elte.hu> (partly) Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-06-22 00:14:49 +00:00
void arch_unmap_area_topdown(struct mm_struct *mm, unsigned long addr)
{
/*
* Is this a new hole at the highest possible address?
*/
[PATCH] Avoiding mmap fragmentation Ingo recently introduced a great speedup for allocating new mmaps using the free_area_cache pointer which boosts the specweb SSL benchmark by 4-5% and causes huge performance increases in thread creation. The downside of this patch is that it does lead to fragmentation in the mmap-ed areas (visible via /proc/self/maps), such that some applications that work fine under 2.4 kernels quickly run out of memory on any 2.6 kernel. The problem is twofold: 1) the free_area_cache is used to continue a search for memory where the last search ended. Before the change new areas were always searched from the base address on. So now new small areas are cluttering holes of all sizes throughout the whole mmap-able region whereas before small holes tended to close holes near the base leaving holes far from the base large and available for larger requests. 2) the free_area_cache also is set to the location of the last munmap-ed area so in scenarios where we allocate e.g. five regions of 1K each, then free regions 4 2 3 in this order the next request for 1K will be placed in the position of the old region 3, whereas before we appended it to the still active region 1, placing it at the location of the old region 2. Before we had 1 free region of 2K, now we only get two free regions of 1K -> fragmentation. The patch addresses thes issues by introducing yet another cache descriptor cached_hole_size that contains the largest known hole size below the current free_area_cache. If a new request comes in the size is compared against the cached_hole_size and if the request can be filled with a hole below free_area_cache the search is started from the base instead. The results look promising: Whereas 2.6.12-rc4 fragments quickly and my (earlier posted) leakme.c test program terminates after 50000+ iterations with 96 distinct and fragmented maps in /proc/self/maps it performs nicely (as expected) with thread creation, Ingo's test_str02 with 20000 threads requires 0.7s system time. Taking out Ingo's patch (un-patch available per request) by basically deleting all mentions of free_area_cache from the kernel and starting the search for new memory always at the respective bases we observe: leakme terminates successfully with 11 distinctive hardly fragmented areas in /proc/self/maps but thread creating is gringdingly slow: 30+s(!) system time for Ingo's test_str02 with 20000 threads. Now - drumroll ;-) the appended patch works fine with leakme: it ends with only 7 distinct areas in /proc/self/maps and also thread creation seems sufficiently fast with 0.71s for 20000 threads. Signed-off-by: Wolfgang Wander <wwc@rentec.com> Credit-to: "Richard Purdie" <rpurdie@rpsys.net> Signed-off-by: Ken Chen <kenneth.w.chen@intel.com> Acked-by: Ingo Molnar <mingo@elte.hu> (partly) Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-06-22 00:14:49 +00:00
if (addr > mm->free_area_cache)
mm->free_area_cache = addr;
/* dont allow allocations above current base */
[PATCH] Avoiding mmap fragmentation Ingo recently introduced a great speedup for allocating new mmaps using the free_area_cache pointer which boosts the specweb SSL benchmark by 4-5% and causes huge performance increases in thread creation. The downside of this patch is that it does lead to fragmentation in the mmap-ed areas (visible via /proc/self/maps), such that some applications that work fine under 2.4 kernels quickly run out of memory on any 2.6 kernel. The problem is twofold: 1) the free_area_cache is used to continue a search for memory where the last search ended. Before the change new areas were always searched from the base address on. So now new small areas are cluttering holes of all sizes throughout the whole mmap-able region whereas before small holes tended to close holes near the base leaving holes far from the base large and available for larger requests. 2) the free_area_cache also is set to the location of the last munmap-ed area so in scenarios where we allocate e.g. five regions of 1K each, then free regions 4 2 3 in this order the next request for 1K will be placed in the position of the old region 3, whereas before we appended it to the still active region 1, placing it at the location of the old region 2. Before we had 1 free region of 2K, now we only get two free regions of 1K -> fragmentation. The patch addresses thes issues by introducing yet another cache descriptor cached_hole_size that contains the largest known hole size below the current free_area_cache. If a new request comes in the size is compared against the cached_hole_size and if the request can be filled with a hole below free_area_cache the search is started from the base instead. The results look promising: Whereas 2.6.12-rc4 fragments quickly and my (earlier posted) leakme.c test program terminates after 50000+ iterations with 96 distinct and fragmented maps in /proc/self/maps it performs nicely (as expected) with thread creation, Ingo's test_str02 with 20000 threads requires 0.7s system time. Taking out Ingo's patch (un-patch available per request) by basically deleting all mentions of free_area_cache from the kernel and starting the search for new memory always at the respective bases we observe: leakme terminates successfully with 11 distinctive hardly fragmented areas in /proc/self/maps but thread creating is gringdingly slow: 30+s(!) system time for Ingo's test_str02 with 20000 threads. Now - drumroll ;-) the appended patch works fine with leakme: it ends with only 7 distinct areas in /proc/self/maps and also thread creation seems sufficiently fast with 0.71s for 20000 threads. Signed-off-by: Wolfgang Wander <wwc@rentec.com> Credit-to: "Richard Purdie" <rpurdie@rpsys.net> Signed-off-by: Ken Chen <kenneth.w.chen@intel.com> Acked-by: Ingo Molnar <mingo@elte.hu> (partly) Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-06-22 00:14:49 +00:00
if (mm->free_area_cache > mm->mmap_base)
mm->free_area_cache = mm->mmap_base;
}
unsigned long
get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
unsigned long pgoff, unsigned long flags)
{
unsigned long ret;
if (!(flags & MAP_FIXED)) {
unsigned long (*get_area)(struct file *, unsigned long, unsigned long, unsigned long, unsigned long);
get_area = current->mm->get_unmapped_area;
if (file && file->f_op && file->f_op->get_unmapped_area)
get_area = file->f_op->get_unmapped_area;
addr = get_area(file, addr, len, pgoff, flags);
if (IS_ERR_VALUE(addr))
return addr;
}
if (addr > TASK_SIZE - len)
return -ENOMEM;
if (addr & ~PAGE_MASK)
return -EINVAL;
if (file && is_file_hugepages(file)) {
/*
* Check if the given range is hugepage aligned, and
* can be made suitable for hugepages.
*/
ret = prepare_hugepage_range(addr, len);
} else {
/*
* Ensure that a normal request is not falling in a
* reserved hugepage range. For some archs like IA-64,
* there is a separate region for hugepages.
*/
ret = is_hugepage_only_range(current->mm, addr, len);
}
if (ret)
return -EINVAL;
return addr;
}
EXPORT_SYMBOL(get_unmapped_area);
/* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
struct vm_area_struct * find_vma(struct mm_struct * mm, unsigned long addr)
{
struct vm_area_struct *vma = NULL;
if (mm) {
/* Check the cache first. */
/* (Cache hit rate is typically around 35%.) */
vma = mm->mmap_cache;
if (!(vma && vma->vm_end > addr && vma->vm_start <= addr)) {
struct rb_node * rb_node;
rb_node = mm->mm_rb.rb_node;
vma = NULL;
while (rb_node) {
struct vm_area_struct * vma_tmp;
vma_tmp = rb_entry(rb_node,
struct vm_area_struct, vm_rb);
if (vma_tmp->vm_end > addr) {
vma = vma_tmp;
if (vma_tmp->vm_start <= addr)
break;
rb_node = rb_node->rb_left;
} else
rb_node = rb_node->rb_right;
}
if (vma)
mm->mmap_cache = vma;
}
}
return vma;
}
EXPORT_SYMBOL(find_vma);
/* Same as find_vma, but also return a pointer to the previous VMA in *pprev. */
struct vm_area_struct *
find_vma_prev(struct mm_struct *mm, unsigned long addr,
struct vm_area_struct **pprev)
{
struct vm_area_struct *vma = NULL, *prev = NULL;
struct rb_node * rb_node;
if (!mm)
goto out;
/* Guard against addr being lower than the first VMA */
vma = mm->mmap;
/* Go through the RB tree quickly. */
rb_node = mm->mm_rb.rb_node;
while (rb_node) {
struct vm_area_struct *vma_tmp;
vma_tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb);
if (addr < vma_tmp->vm_end) {
rb_node = rb_node->rb_left;
} else {
prev = vma_tmp;
if (!prev->vm_next || (addr < prev->vm_next->vm_end))
break;
rb_node = rb_node->rb_right;
}
}
out:
*pprev = prev;
return prev ? prev->vm_next : vma;
}
/*
* Verify that the stack growth is acceptable and
* update accounting. This is shared with both the
* grow-up and grow-down cases.
*/
static int acct_stack_growth(struct vm_area_struct * vma, unsigned long size, unsigned long grow)
{
struct mm_struct *mm = vma->vm_mm;
struct rlimit *rlim = current->signal->rlim;
/* address space limit tests */
if (!may_expand_vm(mm, grow))
return -ENOMEM;
/* Stack limit test */
if (size > rlim[RLIMIT_STACK].rlim_cur)
return -ENOMEM;
/* mlock limit tests */
if (vma->vm_flags & VM_LOCKED) {
unsigned long locked;
unsigned long limit;
locked = mm->locked_vm + grow;
limit = rlim[RLIMIT_MEMLOCK].rlim_cur >> PAGE_SHIFT;
if (locked > limit && !capable(CAP_IPC_LOCK))
return -ENOMEM;
}
/*
* Overcommit.. This must be the final test, as it will
* update security statistics.
*/
if (security_vm_enough_memory(grow))
return -ENOMEM;
/* Ok, everything looks good - let it rip */
mm->total_vm += grow;
if (vma->vm_flags & VM_LOCKED)
mm->locked_vm += grow;
vm_stat_account(mm, vma->vm_flags, vma->vm_file, grow);
return 0;
}
#if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
/*
* PA-RISC uses this for its stack; IA64 for its Register Backing Store.
* vma is the last one with address > vma->vm_end. Have to extend vma.
*/
#ifndef CONFIG_IA64
static inline
#endif
int expand_upwards(struct vm_area_struct *vma, unsigned long address)
{
int error;
if (!(vma->vm_flags & VM_GROWSUP))
return -EFAULT;
/*
* We must make sure the anon_vma is allocated
* so that the anon_vma locking is not a noop.
*/
if (unlikely(anon_vma_prepare(vma)))
return -ENOMEM;
anon_vma_lock(vma);
/*
* vma->vm_start/vm_end cannot change under us because the caller
* is required to hold the mmap_sem in read mode. We need the
* anon_vma lock to serialize against concurrent expand_stacks.
*/
address += 4 + PAGE_SIZE - 1;
address &= PAGE_MASK;
error = 0;
/* Somebody else might have raced and expanded it already */
if (address > vma->vm_end) {
unsigned long size, grow;
size = address - vma->vm_start;
grow = (address - vma->vm_end) >> PAGE_SHIFT;
error = acct_stack_growth(vma, size, grow);
if (!error)
vma->vm_end = address;
}
anon_vma_unlock(vma);
return error;
}
#endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
#ifdef CONFIG_STACK_GROWSUP
int expand_stack(struct vm_area_struct *vma, unsigned long address)
{
return expand_upwards(vma, address);
}
struct vm_area_struct *
find_extend_vma(struct mm_struct *mm, unsigned long addr)
{
struct vm_area_struct *vma, *prev;
addr &= PAGE_MASK;
vma = find_vma_prev(mm, addr, &prev);
if (vma && (vma->vm_start <= addr))
return vma;
if (!prev || expand_stack(prev, addr))
return NULL;
if (prev->vm_flags & VM_LOCKED) {
make_pages_present(addr, prev->vm_end);
}
return prev;
}
#else
/*
* vma is the first one with address < vma->vm_start. Have to extend vma.
*/
int expand_stack(struct vm_area_struct *vma, unsigned long address)
{
int error;
/*
* We must make sure the anon_vma is allocated
* so that the anon_vma locking is not a noop.
*/
if (unlikely(anon_vma_prepare(vma)))
return -ENOMEM;
anon_vma_lock(vma);
/*
* vma->vm_start/vm_end cannot change under us because the caller
* is required to hold the mmap_sem in read mode. We need the
* anon_vma lock to serialize against concurrent expand_stacks.
*/
address &= PAGE_MASK;
error = 0;
/* Somebody else might have raced and expanded it already */
if (address < vma->vm_start) {
unsigned long size, grow;
size = vma->vm_end - address;
grow = (vma->vm_start - address) >> PAGE_SHIFT;
error = acct_stack_growth(vma, size, grow);
if (!error) {
vma->vm_start = address;
vma->vm_pgoff -= grow;
}
}
anon_vma_unlock(vma);
return error;
}
struct vm_area_struct *
find_extend_vma(struct mm_struct * mm, unsigned long addr)
{
struct vm_area_struct * vma;
unsigned long start;
addr &= PAGE_MASK;
vma = find_vma(mm,addr);
if (!vma)
return NULL;
if (vma->vm_start <= addr)
return vma;
if (!(vma->vm_flags & VM_GROWSDOWN))
return NULL;
start = vma->vm_start;
if (expand_stack(vma, addr))
return NULL;
if (vma->vm_flags & VM_LOCKED) {
make_pages_present(addr, start);
}
return vma;
}
#endif
/*
* Ok - we have the memory areas we should free on the vma list,
* so release them, and do the vma updates.
*
* Called with the mm semaphore held.
*/
static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma)
{
[PATCH] mm: update_hiwaters just in time update_mem_hiwater has attracted various criticisms, in particular from those concerned with mm scalability. Originally it was called whenever rss or total_vm got raised. Then many of those callsites were replaced by a timer tick call from account_system_time. Now Frank van Maarseveen reports that to be found inadequate. How about this? Works for Frank. Replace update_mem_hiwater, a poor combination of two unrelated ops, by macros update_hiwater_rss and update_hiwater_vm. Don't attempt to keep mm->hiwater_rss up to date at timer tick, nor every time we raise rss (usually by 1): those are hot paths. Do the opposite, update only when about to lower rss (usually by many), or just before final accounting in do_exit. Handle mm->hiwater_vm in the same way, though it's much less of an issue. Demand that whoever collects these hiwater statistics do the work of taking the maximum with rss or total_vm. And there has been no collector of these hiwater statistics in the tree. The new convention needs an example, so match Frank's usage by adding a VmPeak line above VmSize to /proc/<pid>/status, and also a VmHWM line above VmRSS (High-Water-Mark or High-Water-Memory). There was a particular anomaly during mremap move, that hiwater_vm might be captured too high. A fleeting such anomaly remains, but it's quickly corrected now, whereas before it would stick. What locking? None: if the app is racy then these statistics will be racy, it's not worth any overhead to make them exact. But whenever it suits, hiwater_vm is updated under exclusive mmap_sem, and hiwater_rss under page_table_lock (for now) or with preemption disabled (later on): without going to any trouble, minimize the time between reading current values and updating, to minimize those occasions when a racing thread bumps a count up and back down in between. Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-10-30 01:16:18 +00:00
/* Update high watermark before we lower total_vm */
update_hiwater_vm(mm);
do {
long nrpages = vma_pages(vma);
mm->total_vm -= nrpages;
if (vma->vm_flags & VM_LOCKED)
mm->locked_vm -= nrpages;
vm_stat_account(mm, vma->vm_flags, vma->vm_file, -nrpages);
vma = remove_vma(vma);
} while (vma);
validate_mm(mm);
}
/*
* Get rid of page table information in the indicated region.
*
* Called with the mm semaphore held.
*/
static void unmap_region(struct mm_struct *mm,
[PATCH] freepgt: free_pgtables use vma list Recent woes with some arches needing their own pgd_addr_end macro; and 4-level clear_page_range regression since 2.6.10's clear_page_tables; and its long-standing well-known inefficiency in searching throughout the higher-level page tables for those few entries to clear and free: all can be blamed on ignoring the list of vmas when we free page tables. Replace exit_mmap's clear_page_range of the total user address space by free_pgtables operating on the mm's vma list; unmap_region use it in the same way, giving floor and ceiling beyond which it may not free tables. This brings lmbench fork/exec/sh numbers back to 2.6.10 (unless preempt is enabled, in which case latency fixes spoil unmap_vmas throughput). Beware: the do_mmap_pgoff driver failure case must now use unmap_region instead of zap_page_range, since a page table might have been allocated, and can only be freed while it is touched by some vma. Move free_pgtables from mmap.c to memory.c, where its lower levels are adapted from the clear_page_range levels. (Most of free_pgtables' old code was actually for a non-existent case, prev not properly set up, dating from before hch gave us split_vma.) Pass mmu_gather** in the public interfaces, since we might want to add latency lockdrops later; but no attempt to do so yet, going by vma should itself reduce latency. But what if is_hugepage_only_range? Those ia64 and ppc64 cases need careful examination: put that off until a later patch of the series. What of x86_64's 32bit vdso page __map_syscall32 maps outside any vma? And the range to sparc64's flush_tlb_pgtables? It's less clear to me now that we need to do more than is done here - every PMD_SIZE ever occupied will be flushed, do we really have to flush every PGDIR_SIZE ever partially occupied? A shame to complicate it unnecessarily. Special thanks to David Miller for time spent repairing my ceilings. Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-04-19 20:29:15 +00:00
struct vm_area_struct *vma, struct vm_area_struct *prev,
unsigned long start, unsigned long end)
{
[PATCH] freepgt: free_pgtables use vma list Recent woes with some arches needing their own pgd_addr_end macro; and 4-level clear_page_range regression since 2.6.10's clear_page_tables; and its long-standing well-known inefficiency in searching throughout the higher-level page tables for those few entries to clear and free: all can be blamed on ignoring the list of vmas when we free page tables. Replace exit_mmap's clear_page_range of the total user address space by free_pgtables operating on the mm's vma list; unmap_region use it in the same way, giving floor and ceiling beyond which it may not free tables. This brings lmbench fork/exec/sh numbers back to 2.6.10 (unless preempt is enabled, in which case latency fixes spoil unmap_vmas throughput). Beware: the do_mmap_pgoff driver failure case must now use unmap_region instead of zap_page_range, since a page table might have been allocated, and can only be freed while it is touched by some vma. Move free_pgtables from mmap.c to memory.c, where its lower levels are adapted from the clear_page_range levels. (Most of free_pgtables' old code was actually for a non-existent case, prev not properly set up, dating from before hch gave us split_vma.) Pass mmu_gather** in the public interfaces, since we might want to add latency lockdrops later; but no attempt to do so yet, going by vma should itself reduce latency. But what if is_hugepage_only_range? Those ia64 and ppc64 cases need careful examination: put that off until a later patch of the series. What of x86_64's 32bit vdso page __map_syscall32 maps outside any vma? And the range to sparc64's flush_tlb_pgtables? It's less clear to me now that we need to do more than is done here - every PMD_SIZE ever occupied will be flushed, do we really have to flush every PGDIR_SIZE ever partially occupied? A shame to complicate it unnecessarily. Special thanks to David Miller for time spent repairing my ceilings. Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-04-19 20:29:15 +00:00
struct vm_area_struct *next = prev? prev->vm_next: mm->mmap;
struct mmu_gather *tlb;
unsigned long nr_accounted = 0;
lru_add_drain();
tlb = tlb_gather_mmu(mm, 0);
[PATCH] mm: update_hiwaters just in time update_mem_hiwater has attracted various criticisms, in particular from those concerned with mm scalability. Originally it was called whenever rss or total_vm got raised. Then many of those callsites were replaced by a timer tick call from account_system_time. Now Frank van Maarseveen reports that to be found inadequate. How about this? Works for Frank. Replace update_mem_hiwater, a poor combination of two unrelated ops, by macros update_hiwater_rss and update_hiwater_vm. Don't attempt to keep mm->hiwater_rss up to date at timer tick, nor every time we raise rss (usually by 1): those are hot paths. Do the opposite, update only when about to lower rss (usually by many), or just before final accounting in do_exit. Handle mm->hiwater_vm in the same way, though it's much less of an issue. Demand that whoever collects these hiwater statistics do the work of taking the maximum with rss or total_vm. And there has been no collector of these hiwater statistics in the tree. The new convention needs an example, so match Frank's usage by adding a VmPeak line above VmSize to /proc/<pid>/status, and also a VmHWM line above VmRSS (High-Water-Mark or High-Water-Memory). There was a particular anomaly during mremap move, that hiwater_vm might be captured too high. A fleeting such anomaly remains, but it's quickly corrected now, whereas before it would stick. What locking? None: if the app is racy then these statistics will be racy, it's not worth any overhead to make them exact. But whenever it suits, hiwater_vm is updated under exclusive mmap_sem, and hiwater_rss under page_table_lock (for now) or with preemption disabled (later on): without going to any trouble, minimize the time between reading current values and updating, to minimize those occasions when a racing thread bumps a count up and back down in between. Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-10-30 01:16:18 +00:00
update_hiwater_rss(mm);
unmap_vmas(&tlb, vma, start, end, &nr_accounted, NULL);
vm_unacct_memory(nr_accounted);
free_pgtables(&tlb, vma, prev? prev->vm_end: FIRST_USER_ADDRESS,
[PATCH] freepgt: free_pgtables use vma list Recent woes with some arches needing their own pgd_addr_end macro; and 4-level clear_page_range regression since 2.6.10's clear_page_tables; and its long-standing well-known inefficiency in searching throughout the higher-level page tables for those few entries to clear and free: all can be blamed on ignoring the list of vmas when we free page tables. Replace exit_mmap's clear_page_range of the total user address space by free_pgtables operating on the mm's vma list; unmap_region use it in the same way, giving floor and ceiling beyond which it may not free tables. This brings lmbench fork/exec/sh numbers back to 2.6.10 (unless preempt is enabled, in which case latency fixes spoil unmap_vmas throughput). Beware: the do_mmap_pgoff driver failure case must now use unmap_region instead of zap_page_range, since a page table might have been allocated, and can only be freed while it is touched by some vma. Move free_pgtables from mmap.c to memory.c, where its lower levels are adapted from the clear_page_range levels. (Most of free_pgtables' old code was actually for a non-existent case, prev not properly set up, dating from before hch gave us split_vma.) Pass mmu_gather** in the public interfaces, since we might want to add latency lockdrops later; but no attempt to do so yet, going by vma should itself reduce latency. But what if is_hugepage_only_range? Those ia64 and ppc64 cases need careful examination: put that off until a later patch of the series. What of x86_64's 32bit vdso page __map_syscall32 maps outside any vma? And the range to sparc64's flush_tlb_pgtables? It's less clear to me now that we need to do more than is done here - every PMD_SIZE ever occupied will be flushed, do we really have to flush every PGDIR_SIZE ever partially occupied? A shame to complicate it unnecessarily. Special thanks to David Miller for time spent repairing my ceilings. Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-04-19 20:29:15 +00:00
next? next->vm_start: 0);
tlb_finish_mmu(tlb, start, end);
}
/*
* Create a list of vma's touched by the unmap, removing them from the mm's
* vma list as we go..
*/
static void
detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma,
struct vm_area_struct *prev, unsigned long end)
{
struct vm_area_struct **insertion_point;
struct vm_area_struct *tail_vma = NULL;
[PATCH] Avoiding mmap fragmentation Ingo recently introduced a great speedup for allocating new mmaps using the free_area_cache pointer which boosts the specweb SSL benchmark by 4-5% and causes huge performance increases in thread creation. The downside of this patch is that it does lead to fragmentation in the mmap-ed areas (visible via /proc/self/maps), such that some applications that work fine under 2.4 kernels quickly run out of memory on any 2.6 kernel. The problem is twofold: 1) the free_area_cache is used to continue a search for memory where the last search ended. Before the change new areas were always searched from the base address on. So now new small areas are cluttering holes of all sizes throughout the whole mmap-able region whereas before small holes tended to close holes near the base leaving holes far from the base large and available for larger requests. 2) the free_area_cache also is set to the location of the last munmap-ed area so in scenarios where we allocate e.g. five regions of 1K each, then free regions 4 2 3 in this order the next request for 1K will be placed in the position of the old region 3, whereas before we appended it to the still active region 1, placing it at the location of the old region 2. Before we had 1 free region of 2K, now we only get two free regions of 1K -> fragmentation. The patch addresses thes issues by introducing yet another cache descriptor cached_hole_size that contains the largest known hole size below the current free_area_cache. If a new request comes in the size is compared against the cached_hole_size and if the request can be filled with a hole below free_area_cache the search is started from the base instead. The results look promising: Whereas 2.6.12-rc4 fragments quickly and my (earlier posted) leakme.c test program terminates after 50000+ iterations with 96 distinct and fragmented maps in /proc/self/maps it performs nicely (as expected) with thread creation, Ingo's test_str02 with 20000 threads requires 0.7s system time. Taking out Ingo's patch (un-patch available per request) by basically deleting all mentions of free_area_cache from the kernel and starting the search for new memory always at the respective bases we observe: leakme terminates successfully with 11 distinctive hardly fragmented areas in /proc/self/maps but thread creating is gringdingly slow: 30+s(!) system time for Ingo's test_str02 with 20000 threads. Now - drumroll ;-) the appended patch works fine with leakme: it ends with only 7 distinct areas in /proc/self/maps and also thread creation seems sufficiently fast with 0.71s for 20000 threads. Signed-off-by: Wolfgang Wander <wwc@rentec.com> Credit-to: "Richard Purdie" <rpurdie@rpsys.net> Signed-off-by: Ken Chen <kenneth.w.chen@intel.com> Acked-by: Ingo Molnar <mingo@elte.hu> (partly) Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-06-22 00:14:49 +00:00
unsigned long addr;
insertion_point = (prev ? &prev->vm_next : &mm->mmap);
do {
rb_erase(&vma->vm_rb, &mm->mm_rb);
mm->map_count--;
tail_vma = vma;
vma = vma->vm_next;
} while (vma && vma->vm_start < end);
*insertion_point = vma;
tail_vma->vm_next = NULL;
[PATCH] Avoiding mmap fragmentation Ingo recently introduced a great speedup for allocating new mmaps using the free_area_cache pointer which boosts the specweb SSL benchmark by 4-5% and causes huge performance increases in thread creation. The downside of this patch is that it does lead to fragmentation in the mmap-ed areas (visible via /proc/self/maps), such that some applications that work fine under 2.4 kernels quickly run out of memory on any 2.6 kernel. The problem is twofold: 1) the free_area_cache is used to continue a search for memory where the last search ended. Before the change new areas were always searched from the base address on. So now new small areas are cluttering holes of all sizes throughout the whole mmap-able region whereas before small holes tended to close holes near the base leaving holes far from the base large and available for larger requests. 2) the free_area_cache also is set to the location of the last munmap-ed area so in scenarios where we allocate e.g. five regions of 1K each, then free regions 4 2 3 in this order the next request for 1K will be placed in the position of the old region 3, whereas before we appended it to the still active region 1, placing it at the location of the old region 2. Before we had 1 free region of 2K, now we only get two free regions of 1K -> fragmentation. The patch addresses thes issues by introducing yet another cache descriptor cached_hole_size that contains the largest known hole size below the current free_area_cache. If a new request comes in the size is compared against the cached_hole_size and if the request can be filled with a hole below free_area_cache the search is started from the base instead. The results look promising: Whereas 2.6.12-rc4 fragments quickly and my (earlier posted) leakme.c test program terminates after 50000+ iterations with 96 distinct and fragmented maps in /proc/self/maps it performs nicely (as expected) with thread creation, Ingo's test_str02 with 20000 threads requires 0.7s system time. Taking out Ingo's patch (un-patch available per request) by basically deleting all mentions of free_area_cache from the kernel and starting the search for new memory always at the respective bases we observe: leakme terminates successfully with 11 distinctive hardly fragmented areas in /proc/self/maps but thread creating is gringdingly slow: 30+s(!) system time for Ingo's test_str02 with 20000 threads. Now - drumroll ;-) the appended patch works fine with leakme: it ends with only 7 distinct areas in /proc/self/maps and also thread creation seems sufficiently fast with 0.71s for 20000 threads. Signed-off-by: Wolfgang Wander <wwc@rentec.com> Credit-to: "Richard Purdie" <rpurdie@rpsys.net> Signed-off-by: Ken Chen <kenneth.w.chen@intel.com> Acked-by: Ingo Molnar <mingo@elte.hu> (partly) Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-06-22 00:14:49 +00:00
if (mm->unmap_area == arch_unmap_area)
addr = prev ? prev->vm_end : mm->mmap_base;
else
addr = vma ? vma->vm_start : mm->mmap_base;
mm->unmap_area(mm, addr);
mm->mmap_cache = NULL; /* Kill the cache. */
}
/*
* Split a vma into two pieces at address 'addr', a new vma is allocated
* either for the first part or the the tail.
*/
int split_vma(struct mm_struct * mm, struct vm_area_struct * vma,
unsigned long addr, int new_below)
{
struct mempolicy *pol;
struct vm_area_struct *new;
if (is_vm_hugetlb_page(vma) && (addr & ~HPAGE_MASK))
return -EINVAL;
if (mm->map_count >= sysctl_max_map_count)
return -ENOMEM;
new = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
if (!new)
return -ENOMEM;
/* most fields are the same, copy all, and then fixup */
*new = *vma;
if (new_below)
new->vm_end = addr;
else {
new->vm_start = addr;
new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT);
}
pol = mpol_copy(vma_policy(vma));
if (IS_ERR(pol)) {
kmem_cache_free(vm_area_cachep, new);
return PTR_ERR(pol);
}
vma_set_policy(new, pol);
if (new->vm_file)
get_file(new->vm_file);
if (new->vm_ops && new->vm_ops->open)
new->vm_ops->open(new);
if (new_below)
vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff +
((addr - new->vm_start) >> PAGE_SHIFT), new);
else
vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new);
return 0;
}
/* Munmap is split into 2 main parts -- this part which finds
* what needs doing, and the areas themselves, which do the
* work. This now handles partial unmappings.
* Jeremy Fitzhardinge <jeremy@goop.org>
*/
int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
{
unsigned long end;
struct vm_area_struct *vma, *prev, *last;
if ((start & ~PAGE_MASK) || start > TASK_SIZE || len > TASK_SIZE-start)
return -EINVAL;
if ((len = PAGE_ALIGN(len)) == 0)
return -EINVAL;
/* Find the first overlapping VMA */
vma = find_vma_prev(mm, start, &prev);
if (!vma)
return 0;
/* we have start < vma->vm_end */
/* if it doesn't overlap, we have nothing.. */
end = start + len;
if (vma->vm_start >= end)
return 0;
/*
* If we need to split any vma, do it now to save pain later.
*
* Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
* unmapped vm_area_struct will remain in use: so lower split_vma
* places tmp vma above, and higher split_vma places tmp vma below.
*/
if (start > vma->vm_start) {
int error = split_vma(mm, vma, start, 0);
if (error)
return error;
prev = vma;
}
/* Does it split the last one? */
last = find_vma(mm, end);
if (last && end > last->vm_start) {
int error = split_vma(mm, last, end, 1);
if (error)
return error;
}
vma = prev? prev->vm_next: mm->mmap;
/*
* Remove the vma's, and unmap the actual pages
*/
detach_vmas_to_be_unmapped(mm, vma, prev, end);
unmap_region(mm, vma, prev, start, end);
/* Fix up all other VM information */
remove_vma_list(mm, vma);
return 0;
}
EXPORT_SYMBOL(do_munmap);
asmlinkage long sys_munmap(unsigned long addr, size_t len)
{
int ret;
struct mm_struct *mm = current->mm;
profile_munmap(addr);
down_write(&mm->mmap_sem);
ret = do_munmap(mm, addr, len);
up_write(&mm->mmap_sem);
return ret;
}
static inline void verify_mm_writelocked(struct mm_struct *mm)
{
#ifdef CONFIG_DEBUG_VM
if (unlikely(down_read_trylock(&mm->mmap_sem))) {
WARN_ON(1);
up_read(&mm->mmap_sem);
}
#endif
}
/*
* this is really a simplified "do_mmap". it only handles
* anonymous maps. eventually we may be able to do some
* brk-specific accounting here.
*/
unsigned long do_brk(unsigned long addr, unsigned long len)
{
struct mm_struct * mm = current->mm;
struct vm_area_struct * vma, * prev;
unsigned long flags;
struct rb_node ** rb_link, * rb_parent;
pgoff_t pgoff = addr >> PAGE_SHIFT;
int error;
len = PAGE_ALIGN(len);
if (!len)
return addr;
if ((addr + len) > TASK_SIZE || (addr + len) < addr)
return -EINVAL;
flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;
error = arch_mmap_check(addr, len, flags);
if (error)
return error;
/*
* mlock MCL_FUTURE?
*/
if (mm->def_flags & VM_LOCKED) {
unsigned long locked, lock_limit;
locked = len >> PAGE_SHIFT;
locked += mm->locked_vm;
lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur;
lock_limit >>= PAGE_SHIFT;
if (locked > lock_limit && !capable(CAP_IPC_LOCK))
return -EAGAIN;
}
/*
* mm->mmap_sem is required to protect against another thread
* changing the mappings in case we sleep.
*/
verify_mm_writelocked(mm);
/*
* Clear old maps. this also does some error checking for us
*/
munmap_back:
vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
if (vma && vma->vm_start < addr + len) {
if (do_munmap(mm, addr, len))
return -ENOMEM;
goto munmap_back;
}
/* Check against address space limits *after* clearing old maps... */
if (!may_expand_vm(mm, len >> PAGE_SHIFT))
return -ENOMEM;
if (mm->map_count > sysctl_max_map_count)
return -ENOMEM;
if (security_vm_enough_memory(len >> PAGE_SHIFT))
return -ENOMEM;
/* Can we just expand an old private anonymous mapping? */
if (vma_merge(mm, prev, addr, addr + len, flags,
NULL, NULL, pgoff, NULL))
goto out;
/*
* create a vma struct for an anonymous mapping
*/
vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
if (!vma) {
vm_unacct_memory(len >> PAGE_SHIFT);
return -ENOMEM;
}
vma->vm_mm = mm;
vma->vm_start = addr;
vma->vm_end = addr + len;
vma->vm_pgoff = pgoff;
vma->vm_flags = flags;
[PATCH] add page_mkwrite() vm_operations method Add a new VMA operation to notify a filesystem or other driver about the MMU generating a fault because userspace attempted to write to a page mapped through a read-only PTE. This facility permits the filesystem or driver to: (*) Implement storage allocation/reservation on attempted write, and so to deal with problems such as ENOSPC more gracefully (perhaps by generating SIGBUS). (*) Delay making the page writable until the contents have been written to a backing cache. This is useful for NFS/AFS when using FS-Cache/CacheFS. It permits the filesystem to have some guarantee about the state of the cache. (*) Account and limit number of dirty pages. This is one piece of the puzzle needed to make shared writable mapping work safely in FUSE. Needed by cachefs (Or is it cachefiles? Or fscache? <head spins>). At least four other groups have stated an interest in it or a desire to use the functionality it provides: FUSE, OCFS2, NTFS and JFFS2. Also, things like EXT3 really ought to use it to deal with the case of shared-writable mmap encountering ENOSPC before we permit the page to be dirtied. From: Peter Zijlstra <a.p.zijlstra@chello.nl> get_user_pages(.write=1, .force=1) can generate COW hits on read-only shared mappings, this patch traps those as mkpage_write candidates and fails to handle them the old way. Signed-off-by: David Howells <dhowells@redhat.com> Cc: Miklos Szeredi <miklos@szeredi.hu> Cc: Joel Becker <Joel.Becker@oracle.com> Cc: Mark Fasheh <mark.fasheh@oracle.com> Cc: Anton Altaparmakov <aia21@cantab.net> Cc: David Woodhouse <dwmw2@infradead.org> Cc: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-06-23 09:03:43 +00:00
vma->vm_page_prot = protection_map[flags &
(VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)];
vma_link(mm, vma, prev, rb_link, rb_parent);
out:
mm->total_vm += len >> PAGE_SHIFT;
if (flags & VM_LOCKED) {
mm->locked_vm += len >> PAGE_SHIFT;
make_pages_present(addr, addr + len);
}
return addr;
}
EXPORT_SYMBOL(do_brk);
/* Release all mmaps. */
void exit_mmap(struct mm_struct *mm)
{
struct mmu_gather *tlb;
[PATCH] freepgt: free_pgtables use vma list Recent woes with some arches needing their own pgd_addr_end macro; and 4-level clear_page_range regression since 2.6.10's clear_page_tables; and its long-standing well-known inefficiency in searching throughout the higher-level page tables for those few entries to clear and free: all can be blamed on ignoring the list of vmas when we free page tables. Replace exit_mmap's clear_page_range of the total user address space by free_pgtables operating on the mm's vma list; unmap_region use it in the same way, giving floor and ceiling beyond which it may not free tables. This brings lmbench fork/exec/sh numbers back to 2.6.10 (unless preempt is enabled, in which case latency fixes spoil unmap_vmas throughput). Beware: the do_mmap_pgoff driver failure case must now use unmap_region instead of zap_page_range, since a page table might have been allocated, and can only be freed while it is touched by some vma. Move free_pgtables from mmap.c to memory.c, where its lower levels are adapted from the clear_page_range levels. (Most of free_pgtables' old code was actually for a non-existent case, prev not properly set up, dating from before hch gave us split_vma.) Pass mmu_gather** in the public interfaces, since we might want to add latency lockdrops later; but no attempt to do so yet, going by vma should itself reduce latency. But what if is_hugepage_only_range? Those ia64 and ppc64 cases need careful examination: put that off until a later patch of the series. What of x86_64's 32bit vdso page __map_syscall32 maps outside any vma? And the range to sparc64's flush_tlb_pgtables? It's less clear to me now that we need to do more than is done here - every PMD_SIZE ever occupied will be flushed, do we really have to flush every PGDIR_SIZE ever partially occupied? A shame to complicate it unnecessarily. Special thanks to David Miller for time spent repairing my ceilings. Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-04-19 20:29:15 +00:00
struct vm_area_struct *vma = mm->mmap;
unsigned long nr_accounted = 0;
unsigned long end;
lru_add_drain();
flush_cache_mm(mm);
[PATCH] freepgt: free_pgtables use vma list Recent woes with some arches needing their own pgd_addr_end macro; and 4-level clear_page_range regression since 2.6.10's clear_page_tables; and its long-standing well-known inefficiency in searching throughout the higher-level page tables for those few entries to clear and free: all can be blamed on ignoring the list of vmas when we free page tables. Replace exit_mmap's clear_page_range of the total user address space by free_pgtables operating on the mm's vma list; unmap_region use it in the same way, giving floor and ceiling beyond which it may not free tables. This brings lmbench fork/exec/sh numbers back to 2.6.10 (unless preempt is enabled, in which case latency fixes spoil unmap_vmas throughput). Beware: the do_mmap_pgoff driver failure case must now use unmap_region instead of zap_page_range, since a page table might have been allocated, and can only be freed while it is touched by some vma. Move free_pgtables from mmap.c to memory.c, where its lower levels are adapted from the clear_page_range levels. (Most of free_pgtables' old code was actually for a non-existent case, prev not properly set up, dating from before hch gave us split_vma.) Pass mmu_gather** in the public interfaces, since we might want to add latency lockdrops later; but no attempt to do so yet, going by vma should itself reduce latency. But what if is_hugepage_only_range? Those ia64 and ppc64 cases need careful examination: put that off until a later patch of the series. What of x86_64's 32bit vdso page __map_syscall32 maps outside any vma? And the range to sparc64's flush_tlb_pgtables? It's less clear to me now that we need to do more than is done here - every PMD_SIZE ever occupied will be flushed, do we really have to flush every PGDIR_SIZE ever partially occupied? A shame to complicate it unnecessarily. Special thanks to David Miller for time spent repairing my ceilings. Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-04-19 20:29:15 +00:00
tlb = tlb_gather_mmu(mm, 1);
[PATCH] mm: update_hiwaters just in time update_mem_hiwater has attracted various criticisms, in particular from those concerned with mm scalability. Originally it was called whenever rss or total_vm got raised. Then many of those callsites were replaced by a timer tick call from account_system_time. Now Frank van Maarseveen reports that to be found inadequate. How about this? Works for Frank. Replace update_mem_hiwater, a poor combination of two unrelated ops, by macros update_hiwater_rss and update_hiwater_vm. Don't attempt to keep mm->hiwater_rss up to date at timer tick, nor every time we raise rss (usually by 1): those are hot paths. Do the opposite, update only when about to lower rss (usually by many), or just before final accounting in do_exit. Handle mm->hiwater_vm in the same way, though it's much less of an issue. Demand that whoever collects these hiwater statistics do the work of taking the maximum with rss or total_vm. And there has been no collector of these hiwater statistics in the tree. The new convention needs an example, so match Frank's usage by adding a VmPeak line above VmSize to /proc/<pid>/status, and also a VmHWM line above VmRSS (High-Water-Mark or High-Water-Memory). There was a particular anomaly during mremap move, that hiwater_vm might be captured too high. A fleeting such anomaly remains, but it's quickly corrected now, whereas before it would stick. What locking? None: if the app is racy then these statistics will be racy, it's not worth any overhead to make them exact. But whenever it suits, hiwater_vm is updated under exclusive mmap_sem, and hiwater_rss under page_table_lock (for now) or with preemption disabled (later on): without going to any trouble, minimize the time between reading current values and updating, to minimize those occasions when a racing thread bumps a count up and back down in between. Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-10-30 01:16:18 +00:00
/* Don't update_hiwater_rss(mm) here, do_exit already did */
[PATCH] freepgt: free_pgtables use vma list Recent woes with some arches needing their own pgd_addr_end macro; and 4-level clear_page_range regression since 2.6.10's clear_page_tables; and its long-standing well-known inefficiency in searching throughout the higher-level page tables for those few entries to clear and free: all can be blamed on ignoring the list of vmas when we free page tables. Replace exit_mmap's clear_page_range of the total user address space by free_pgtables operating on the mm's vma list; unmap_region use it in the same way, giving floor and ceiling beyond which it may not free tables. This brings lmbench fork/exec/sh numbers back to 2.6.10 (unless preempt is enabled, in which case latency fixes spoil unmap_vmas throughput). Beware: the do_mmap_pgoff driver failure case must now use unmap_region instead of zap_page_range, since a page table might have been allocated, and can only be freed while it is touched by some vma. Move free_pgtables from mmap.c to memory.c, where its lower levels are adapted from the clear_page_range levels. (Most of free_pgtables' old code was actually for a non-existent case, prev not properly set up, dating from before hch gave us split_vma.) Pass mmu_gather** in the public interfaces, since we might want to add latency lockdrops later; but no attempt to do so yet, going by vma should itself reduce latency. But what if is_hugepage_only_range? Those ia64 and ppc64 cases need careful examination: put that off until a later patch of the series. What of x86_64's 32bit vdso page __map_syscall32 maps outside any vma? And the range to sparc64's flush_tlb_pgtables? It's less clear to me now that we need to do more than is done here - every PMD_SIZE ever occupied will be flushed, do we really have to flush every PGDIR_SIZE ever partially occupied? A shame to complicate it unnecessarily. Special thanks to David Miller for time spent repairing my ceilings. Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-04-19 20:29:15 +00:00
/* Use -1 here to ensure all VMAs in the mm are unmapped */
end = unmap_vmas(&tlb, vma, 0, -1, &nr_accounted, NULL);
vm_unacct_memory(nr_accounted);
free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, 0);
tlb_finish_mmu(tlb, 0, end);
/*
[PATCH] mm: unlink vma before pagetables In most places the descent from pgd to pud to pmd to pte holds mmap_sem (exclusively or not), which ensures that free_pgtables cannot be freeing page tables from any level at the same time. But truncation and reverse mapping descend without mmap_sem. No problem: just make sure that a vma is unlinked from its prio_tree (or nonlinear list) and from its anon_vma list, after zapping the vma, but before freeing its page tables. Then neither vmtruncate nor rmap can reach that vma whose page tables are now volatile (nor do they need to reach it, since all its page entries have been zapped by this stage). The i_mmap_lock and anon_vma->lock already serialize this correctly; but the locking hierarchy is such that we cannot take them while holding page_table_lock. Well, we're trying to push that down anyway. So in this patch, move anon_vma_unlink and unlink_file_vma into free_pgtables, at the same time as moving page_table_lock around calls to unmap_vmas. tlb_gather_mmu and tlb_finish_mmu then fall outside the page_table_lock, but we made them preempt_disable and preempt_enable earlier; and a long source audit of all the architectures has shown no problem with removing page_table_lock from them. free_pgtables doesn't need page_table_lock for itself, nor for what it calls; tlb->mm->nr_ptes is usually protected by page_table_lock, but partly by non-exclusive mmap_sem - here it's decremented with exclusive mmap_sem, or mm_users 0. update_hiwater_rss and vm_unacct_memory don't need page_table_lock either. Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-10-30 01:16:29 +00:00
* Walk the list again, actually closing and freeing it,
* with preemption enabled, without holding any MM locks.
*/
while (vma)
vma = remove_vma(vma);
[PATCH] freepgt: free_pgtables use vma list Recent woes with some arches needing their own pgd_addr_end macro; and 4-level clear_page_range regression since 2.6.10's clear_page_tables; and its long-standing well-known inefficiency in searching throughout the higher-level page tables for those few entries to clear and free: all can be blamed on ignoring the list of vmas when we free page tables. Replace exit_mmap's clear_page_range of the total user address space by free_pgtables operating on the mm's vma list; unmap_region use it in the same way, giving floor and ceiling beyond which it may not free tables. This brings lmbench fork/exec/sh numbers back to 2.6.10 (unless preempt is enabled, in which case latency fixes spoil unmap_vmas throughput). Beware: the do_mmap_pgoff driver failure case must now use unmap_region instead of zap_page_range, since a page table might have been allocated, and can only be freed while it is touched by some vma. Move free_pgtables from mmap.c to memory.c, where its lower levels are adapted from the clear_page_range levels. (Most of free_pgtables' old code was actually for a non-existent case, prev not properly set up, dating from before hch gave us split_vma.) Pass mmu_gather** in the public interfaces, since we might want to add latency lockdrops later; but no attempt to do so yet, going by vma should itself reduce latency. But what if is_hugepage_only_range? Those ia64 and ppc64 cases need careful examination: put that off until a later patch of the series. What of x86_64's 32bit vdso page __map_syscall32 maps outside any vma? And the range to sparc64's flush_tlb_pgtables? It's less clear to me now that we need to do more than is done here - every PMD_SIZE ever occupied will be flushed, do we really have to flush every PGDIR_SIZE ever partially occupied? A shame to complicate it unnecessarily. Special thanks to David Miller for time spent repairing my ceilings. Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-04-19 20:29:15 +00:00
BUG_ON(mm->nr_ptes > (FIRST_USER_ADDRESS+PMD_SIZE-1)>>PMD_SHIFT);
}
/* Insert vm structure into process list sorted by address
* and into the inode's i_mmap tree. If vm_file is non-NULL
* then i_mmap_lock is taken here.
*/
int insert_vm_struct(struct mm_struct * mm, struct vm_area_struct * vma)
{
struct vm_area_struct * __vma, * prev;
struct rb_node ** rb_link, * rb_parent;
/*
* The vm_pgoff of a purely anonymous vma should be irrelevant
* until its first write fault, when page's anon_vma and index
* are set. But now set the vm_pgoff it will almost certainly
* end up with (unless mremap moves it elsewhere before that
* first wfault), so /proc/pid/maps tells a consistent story.
*
* By setting it to reflect the virtual start address of the
* vma, merges and splits can happen in a seamless way, just
* using the existing file pgoff checks and manipulations.
* Similarly in do_mmap_pgoff and in do_brk.
*/
if (!vma->vm_file) {
BUG_ON(vma->anon_vma);
vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT;
}
__vma = find_vma_prepare(mm,vma->vm_start,&prev,&rb_link,&rb_parent);
if (__vma && __vma->vm_start < vma->vm_end)
return -ENOMEM;
if ((vma->vm_flags & VM_ACCOUNT) &&
security_vm_enough_memory(vma_pages(vma)))
return -ENOMEM;
vma_link(mm, vma, prev, rb_link, rb_parent);
return 0;
}
/*
* Copy the vma structure to a new location in the same mm,
* prior to moving page table entries, to effect an mremap move.
*/
struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
unsigned long addr, unsigned long len, pgoff_t pgoff)
{
struct vm_area_struct *vma = *vmap;
unsigned long vma_start = vma->vm_start;
struct mm_struct *mm = vma->vm_mm;
struct vm_area_struct *new_vma, *prev;
struct rb_node **rb_link, *rb_parent;
struct mempolicy *pol;
/*
* If anonymous vma has not yet been faulted, update new pgoff
* to match new location, to increase its chance of merging.
*/
if (!vma->vm_file && !vma->anon_vma)
pgoff = addr >> PAGE_SHIFT;
find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags,
vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
if (new_vma) {
/*
* Source vma may have been merged into new_vma
*/
if (vma_start >= new_vma->vm_start &&
vma_start < new_vma->vm_end)
*vmap = new_vma;
} else {
new_vma = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
if (new_vma) {
*new_vma = *vma;
pol = mpol_copy(vma_policy(vma));
if (IS_ERR(pol)) {
kmem_cache_free(vm_area_cachep, new_vma);
return NULL;
}
vma_set_policy(new_vma, pol);
new_vma->vm_start = addr;
new_vma->vm_end = addr + len;
new_vma->vm_pgoff = pgoff;
if (new_vma->vm_file)
get_file(new_vma->vm_file);
if (new_vma->vm_ops && new_vma->vm_ops->open)
new_vma->vm_ops->open(new_vma);
vma_link(mm, new_vma, prev, rb_link, rb_parent);
}
}
return new_vma;
}
/*
* Return true if the calling process may expand its vm space by the passed
* number of pages
*/
int may_expand_vm(struct mm_struct *mm, unsigned long npages)
{
unsigned long cur = mm->total_vm; /* pages */
unsigned long lim;
lim = current->signal->rlim[RLIMIT_AS].rlim_cur >> PAGE_SHIFT;
if (cur + npages > lim)
return 0;
return 1;
}