NOMMU: Make mmap allocation page trimming behaviour configurable.
NOMMU mmap allocates a piece of memory for an mmap that's rounded up in size to
the nearest power-of-2 number of pages. Currently it then discards the excess
pages back to the page allocator, making that memory available for use by other
things. This can, however, cause greater amount of fragmentation.
To counter this, a sysctl is added in order to fine-tune the trimming
behaviour. The default behaviour remains to trim pages aggressively, while
this can either be disabled completely or set to a higher page-granular
watermark in order to have finer-grained control.
vm region vm_top bits taken from an earlier patch by David Howells.
Signed-off-by: Paul Mundt <lethal@linux-sh.org>
Signed-off-by: David Howells <dhowells@redhat.com>
Tested-by: Mike Frysinger <vapier.adi@gmail.com>
diff --git a/Documentation/nommu-mmap.txt b/Documentation/nommu-mmap.txt
index 02b89dc..b565e82 100644
--- a/Documentation/nommu-mmap.txt
+++ b/Documentation/nommu-mmap.txt
@@ -248,3 +248,18 @@
Provision of shared mappings on block device files is exactly the same as for
character devices. If there isn't a real device underneath, then the driver
should allocate sufficient contiguous memory to honour any supported mapping.
+
+
+=================================
+ADJUSTING PAGE TRIMMING BEHAVIOUR
+=================================
+
+NOMMU mmap automatically rounds up to the nearest power-of-2 number of pages
+when performing an allocation. This can have adverse effects on memory
+fragmentation, and as such, is left configurable. The default behaviour is to
+aggressively trim allocations and discard any excess pages back in to the page
+allocator. In order to retain finer-grained control over fragmentation, this
+behaviour can either be disabled completely, or bumped up to a higher page
+watermark where trimming begins.
+
+Page trimming behaviour is configurable via the sysctl `vm.nr_trim_pages'.
diff --git a/Documentation/sysctl/vm.txt b/Documentation/sysctl/vm.txt
index cd05994..a341507 100644
--- a/Documentation/sysctl/vm.txt
+++ b/Documentation/sysctl/vm.txt
@@ -38,6 +38,7 @@
- numa_zonelist_order
- nr_hugepages
- nr_overcommit_hugepages
+- nr_trim_pages (only if CONFIG_MMU=n)
==============================================================
@@ -348,3 +349,20 @@
nr_hugepages + nr_overcommit_hugepages.
See Documentation/vm/hugetlbpage.txt
+
+==============================================================
+
+nr_trim_pages
+
+This is available only on NOMMU kernels.
+
+This value adjusts the excess page trimming behaviour of power-of-2 aligned
+NOMMU mmap allocations.
+
+A value of 0 disables trimming of allocations entirely, while a value of 1
+trims excess pages aggressively. Any value >= 1 acts as the watermark where
+trimming of allocations is initiated.
+
+The default value is 1.
+
+See Documentation/nommu-mmap.txt for more information.
diff --git a/include/linux/mm_types.h b/include/linux/mm_types.h
index 1c1e0d3..92915e8 100644
--- a/include/linux/mm_types.h
+++ b/include/linux/mm_types.h
@@ -106,6 +106,7 @@
unsigned long vm_flags; /* VMA vm_flags */
unsigned long vm_start; /* start address of region */
unsigned long vm_end; /* region initialised to here */
+ unsigned long vm_top; /* region allocated to here */
unsigned long vm_pgoff; /* the offset in vm_file corresponding to vm_start */
struct file *vm_file; /* the backing file or NULL */
diff --git a/kernel/sysctl.c b/kernel/sysctl.c
index 92f6e5b..89d7443 100644
--- a/kernel/sysctl.c
+++ b/kernel/sysctl.c
@@ -82,6 +82,9 @@
extern int compat_log;
extern int latencytop_enabled;
extern int sysctl_nr_open_min, sysctl_nr_open_max;
+#ifndef CONFIG_MMU
+extern int sysctl_nr_trim_pages;
+#endif
#ifdef CONFIG_RCU_TORTURE_TEST
extern int rcutorture_runnable;
#endif /* #ifdef CONFIG_RCU_TORTURE_TEST */
@@ -1102,6 +1105,17 @@
.mode = 0644,
.proc_handler = &proc_dointvec
},
+#else
+ {
+ .ctl_name = CTL_UNNUMBERED,
+ .procname = "nr_trim_pages",
+ .data = &sysctl_nr_trim_pages,
+ .maxlen = sizeof(sysctl_nr_trim_pages),
+ .mode = 0644,
+ .proc_handler = &proc_dointvec_minmax,
+ .strategy = &sysctl_intvec,
+ .extra1 = &zero,
+ },
#endif
{
.ctl_name = VM_LAPTOP_MODE,
diff --git a/mm/nommu.c b/mm/nommu.c
index 0d363df..a6e8ccf 100644
--- a/mm/nommu.c
+++ b/mm/nommu.c
@@ -10,7 +10,7 @@
* Copyright (c) 2000-2003 David McCullough <davidm@snapgear.com>
* Copyright (c) 2000-2001 D Jeff Dionne <jeff@uClinux.org>
* Copyright (c) 2002 Greg Ungerer <gerg@snapgear.com>
- * Copyright (c) 2007 Paul Mundt <lethal@linux-sh.org>
+ * Copyright (c) 2007-2008 Paul Mundt <lethal@linux-sh.org>
*/
#include <linux/module.h>
@@ -66,6 +66,7 @@
int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */
int sysctl_overcommit_ratio = 50; /* default is 50% */
int sysctl_max_map_count = DEFAULT_MAX_MAP_COUNT;
+int sysctl_nr_trim_pages = 1; /* page trimming behaviour */
int heap_stack_gap = 0;
atomic_t mmap_pages_allocated;
@@ -455,6 +456,8 @@
last = rb_entry(lastp, struct vm_region, vm_rb);
if (unlikely(last->vm_end <= last->vm_start))
BUG();
+ if (unlikely(last->vm_top < last->vm_end))
+ BUG();
while ((p = rb_next(lastp))) {
region = rb_entry(p, struct vm_region, vm_rb);
@@ -462,7 +465,9 @@
if (unlikely(region->vm_end <= region->vm_start))
BUG();
- if (unlikely(region->vm_start < last->vm_end))
+ if (unlikely(region->vm_top < region->vm_end))
+ BUG();
+ if (unlikely(region->vm_start < last->vm_top))
BUG();
lastp = p;
@@ -536,7 +541,7 @@
/*
* release a reference to a region
* - the caller must hold the region semaphore, which this releases
- * - the region may not have been added to the tree yet, in which case vm_end
+ * - the region may not have been added to the tree yet, in which case vm_top
* will equal vm_start
*/
static void __put_nommu_region(struct vm_region *region)
@@ -547,7 +552,7 @@
BUG_ON(!nommu_region_tree.rb_node);
if (atomic_dec_and_test(®ion->vm_usage)) {
- if (region->vm_end > region->vm_start)
+ if (region->vm_top > region->vm_start)
delete_nommu_region(region);
up_write(&nommu_region_sem);
@@ -558,7 +563,7 @@
* from ramfs/tmpfs mustn't be released here */
if (region->vm_flags & VM_MAPPED_COPY) {
kdebug("free series");
- free_page_series(region->vm_start, region->vm_end);
+ free_page_series(region->vm_start, region->vm_top);
}
kmem_cache_free(vm_region_jar, region);
} else {
@@ -999,6 +1004,10 @@
int ret;
ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
+ if (ret == 0) {
+ vma->vm_region->vm_top = vma->vm_region->vm_end;
+ return ret;
+ }
if (ret != -ENOSYS)
return ret;
@@ -1027,11 +1036,14 @@
*/
if (vma->vm_file) {
ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
- if (ret != -ENOSYS) {
+ if (ret == 0) {
/* shouldn't return success if we're not sharing */
- BUG_ON(ret == 0 && !(vma->vm_flags & VM_MAYSHARE));
- return ret; /* success or a real error */
+ BUG_ON(!(vma->vm_flags & VM_MAYSHARE));
+ vma->vm_region->vm_top = vma->vm_region->vm_end;
+ return ret;
}
+ if (ret != -ENOSYS)
+ return ret;
/* getting an ENOSYS error indicates that direct mmap isn't
* possible (as opposed to tried but failed) so we'll try to
@@ -1051,23 +1063,25 @@
if (!pages)
goto enomem;
- /* we allocated a power-of-2 sized page set, so we need to trim off the
- * excess */
total = 1 << order;
atomic_add(total, &mmap_pages_allocated);
point = rlen >> PAGE_SHIFT;
- while (total > point) {
- order = ilog2(total - point);
- n = 1 << order;
- kdebug("shave %lu/%lu @%lu", n, total - point, total);
- atomic_sub(n, &mmap_pages_allocated);
- total -= n;
- set_page_refcounted(pages + total);
- __free_pages(pages + total, order);
+
+ /* we allocated a power-of-2 sized page set, so we may want to trim off
+ * the excess */
+ if (sysctl_nr_trim_pages && total - point >= sysctl_nr_trim_pages) {
+ while (total > point) {
+ order = ilog2(total - point);
+ n = 1 << order;
+ kdebug("shave %lu/%lu @%lu", n, total - point, total);
+ atomic_sub(n, &mmap_pages_allocated);
+ total -= n;
+ set_page_refcounted(pages + total);
+ __free_pages(pages + total, order);
+ }
}
- total = rlen >> PAGE_SHIFT;
for (point = 1; point < total; point++)
set_page_refcounted(&pages[point]);
@@ -1075,6 +1089,7 @@
region->vm_flags = vma->vm_flags |= VM_MAPPED_COPY;
region->vm_start = (unsigned long) base;
region->vm_end = region->vm_start + rlen;
+ region->vm_top = region->vm_start + (total << PAGE_SHIFT);
vma->vm_start = region->vm_start;
vma->vm_end = region->vm_start + len;
@@ -1110,6 +1125,7 @@
free_page_series(region->vm_start, region->vm_end);
region->vm_start = vma->vm_start = 0;
region->vm_end = vma->vm_end = 0;
+ region->vm_top = 0;
return ret;
enomem:
@@ -1401,7 +1417,7 @@
npages = (addr - vma->vm_start) >> PAGE_SHIFT;
if (new_below) {
- region->vm_end = new->vm_end = addr;
+ region->vm_top = region->vm_end = new->vm_end = addr;
} else {
region->vm_start = new->vm_start = addr;
region->vm_pgoff = new->vm_pgoff += npages;
@@ -1418,6 +1434,7 @@
vma->vm_region->vm_pgoff = vma->vm_pgoff += npages;
} else {
vma->vm_region->vm_end = vma->vm_end = addr;
+ vma->vm_region->vm_top = addr;
}
add_nommu_region(vma->vm_region);
add_nommu_region(new->vm_region);
@@ -1454,10 +1471,12 @@
down_write(&nommu_region_sem);
delete_nommu_region(region);
- if (from > region->vm_start)
- region->vm_end = from;
- else
+ if (from > region->vm_start) {
+ to = region->vm_top;
+ region->vm_top = region->vm_end = from;
+ } else {
region->vm_start = to;
+ }
add_nommu_region(region);
up_write(&nommu_region_sem);