diff -purN -X /home/mbligh/.diff.exclude 850-mbind_part1/arch/i386/kernel/entry.S 851-mbind_part2/arch/i386/kernel/entry.S
--- 850-mbind_part1/arch/i386/kernel/entry.S 2003-10-01 11:40:40.000000000 -0700
+++ 851-mbind_part2/arch/i386/kernel/entry.S 2003-10-02 16:41:14.000000000 -0700
@@ -829,7 +829,7 @@ ENTRY(sys_call_table)
.long sys_getdents64 /* 220 */
.long sys_fcntl64
.long sys_ni_syscall /* reserved for TUX */
- .long sys_ni_syscall
+ .long sys_mbind
.long sys_gettid
.long sys_readahead /* 225 */
.long sys_setxattr
diff -purN -X /home/mbligh/.diff.exclude 850-mbind_part1/fs/inode.c 851-mbind_part2/fs/inode.c
--- 850-mbind_part1/fs/inode.c 2003-10-01 11:47:01.000000000 -0700
+++ 851-mbind_part2/fs/inode.c 2003-10-02 16:41:14.000000000 -0700
@@ -147,6 +147,9 @@ static struct inode *alloc_inode(struct
mapping->dirtied_when = 0;
mapping->assoc_mapping = NULL;
mapping->backing_dev_info = &default_backing_dev_info;
+#ifdef CONFIG_NUMA
+ mapping->binding = NULL;
+#endif
if (sb->s_bdev)
mapping->backing_dev_info = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
memset(&inode->u, 0, sizeof(inode->u));
diff -purN -X /home/mbligh/.diff.exclude 850-mbind_part1/include/asm-i386/unistd.h 851-mbind_part2/include/asm-i386/unistd.h
--- 850-mbind_part1/include/asm-i386/unistd.h 2003-10-01 11:41:15.000000000 -0700
+++ 851-mbind_part2/include/asm-i386/unistd.h 2003-10-02 16:41:14.000000000 -0700
@@ -228,7 +228,7 @@
#define __NR_madvise1 219 /* delete when C lib stub is removed */
#define __NR_getdents64 220
#define __NR_fcntl64 221
-/* 223 is unused */
+#define __NR_mbind 223
#define __NR_gettid 224
#define __NR_readahead 225
#define __NR_setxattr 226
diff -purN -X /home/mbligh/.diff.exclude 850-mbind_part1/include/linux/fs.h 851-mbind_part2/include/linux/fs.h
--- 850-mbind_part1/include/linux/fs.h 2003-10-01 11:48:25.000000000 -0700
+++ 851-mbind_part2/include/linux/fs.h 2003-10-02 16:41:14.000000000 -0700
@@ -336,6 +336,9 @@ struct address_space {
spinlock_t private_lock; /* for use by the address_space */
struct list_head private_list; /* ditto */
struct address_space *assoc_mapping; /* ditto */
+#ifdef CONFIG_NUMA
+ struct binding *binding; /* for memory bindings */
+#endif
};
struct block_device {
diff -purN -X /home/mbligh/.diff.exclude 850-mbind_part1/include/linux/mmzone.h 851-mbind_part2/include/linux/mmzone.h
--- 850-mbind_part1/include/linux/mmzone.h 2003-10-02 16:41:10.000000000 -0700
+++ 851-mbind_part2/include/linux/mmzone.h 2003-10-02 16:41:14.000000000 -0700
@@ -377,6 +377,11 @@ static inline struct zonelist *get_node_
#define get_zonelist(gfp_mask) get_node_zonelist(numa_node_id(), gfp_mask)
+/* Structure to keep track of memory segment (VMA) bindings */
+struct binding {
+ struct zonelist zonelist;
+};
+
#endif /* !__ASSEMBLY__ */
#endif /* __KERNEL__ */
#endif /* _LINUX_MMZONE_H */
diff -purN -X /home/mbligh/.diff.exclude 850-mbind_part1/include/linux/pagemap.h 851-mbind_part2/include/linux/pagemap.h
--- 850-mbind_part1/include/linux/pagemap.h 2003-10-02 16:41:10.000000000 -0700
+++ 851-mbind_part2/include/linux/pagemap.h 2003-10-02 16:41:14.000000000 -0700
@@ -50,11 +50,29 @@ static inline void mapping_set_gfp_mask(
#define page_cache_release(page) put_page(page)
void release_pages(struct page **pages, int nr, int cold);
+#ifndef CONFIG_NUMA
+
static inline struct page *__page_cache_alloc(struct address_space *x, int gfp_mask)
{
return alloc_pages(gfp_mask, 0);
}
+#else /* CONFIG_NUMA */
+
+static inline struct page *__page_cache_alloc(struct address_space *x, int gfp_mask)
+{
+ struct zonelist *zonelist;
+
+ if (!x->binding)
+ zonelist = get_zonelist(gfp_mask);
+ else
+ zonelist = &x->binding->zonelist;
+
+ return __alloc_pages(gfp_mask, 0, zonelist);
+}
+
+#endif /* !CONFIG_NUMA */
+
static inline struct page *page_cache_alloc(struct address_space *x)
{
return __page_cache_alloc(x, mapping_gfp_mask(x));
diff -purN -X /home/mbligh/.diff.exclude 850-mbind_part1/kernel/sys.c 851-mbind_part2/kernel/sys.c
--- 850-mbind_part1/kernel/sys.c 2003-10-01 11:48:28.000000000 -0700
+++ 851-mbind_part2/kernel/sys.c 2003-10-02 16:41:14.000000000 -0700
@@ -235,6 +235,7 @@ cond_syscall(sys_epoll_ctl)
cond_syscall(sys_epoll_wait)
cond_syscall(sys_pciconfig_read)
cond_syscall(sys_pciconfig_write)
+cond_syscall(sys_mbind)
static int set_one_prio(struct task_struct *p, int niceval, int error)
{
diff -purN -X /home/mbligh/.diff.exclude 850-mbind_part1/mm/Makefile 851-mbind_part2/mm/Makefile
--- 850-mbind_part1/mm/Makefile 2003-10-01 11:47:15.000000000 -0700
+++ 851-mbind_part2/mm/Makefile 2003-10-02 16:41:14.000000000 -0700
@@ -7,8 +7,10 @@ mmu-$(CONFIG_MMU) := fremap.o highmem.o
mlock.o mmap.o mprotect.o mremap.o msync.o rmap.o \
shmem.o vmalloc.o
-obj-y := bootmem.o filemap.o mempool.o oom_kill.o fadvise.o \
+obj-y := bootmem.o fadvise.o filemap.o mempool.o oom_kill.o \
page_alloc.o page-writeback.o pdflush.o readahead.o \
slab.o swap.o truncate.o vmscan.o $(mmu-y)
obj-$(CONFIG_SWAP) += page_io.o swap_state.o swapfile.o
+
+obj-$(CONFIG_NUMA) += mbind.o
diff -purN -X /home/mbligh/.diff.exclude 850-mbind_part1/mm/mbind.c 851-mbind_part2/mm/mbind.c
--- 850-mbind_part1/mm/mbind.c 1969-12-31 16:00:00.000000000 -0800
+++ 851-mbind_part2/mm/mbind.c 2003-10-02 16:41:14.000000000 -0700
@@ -0,0 +1,147 @@
+/*
+ * mm/mbind.c
+ *
+ * Written by: Matthew Dobson, IBM Corporation
+ *
+ * Copyright (C) 2003, IBM Corp.
+ *
+ * All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
+ * NON INFRINGEMENT. See the GNU General Public License for more
+ * details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+ *
+ * Send feedback to <colpatch@us.ibm.com>
+ */
+#include <linux/errno.h>
+#include <linux/mm.h>
+#include <linux/string.h>
+#include <asm/topology.h>
+#include <asm/uaccess.h>
+
+/* Translate a cpumask to a nodemask */
+static inline void cpumask_to_nodemask(unsigned long * cpumask, unsigned long * nodemask)
+{
+ int i;
+
+ for (i = 0; i < NR_CPUS; i++)
+ if (test_bit(i, cpumask))
+ set_bit(cpu_to_node(i), nodemask);
+}
+
+/*
+ * Adds the zones belonging to @pgdat to @zonelist. Returns the next
+ * index in @zonelist.
+ */
+static inline int add_node(pg_data_t *pgdat, struct zonelist *zonelist, int zone_num)
+{
+ int i;
+ struct zone *zone;
+
+ for (i = MAX_NR_ZONES-1; i >=0 ; i--) {
+ zone = pgdat->node_zones + i;
+ if (zone->present_pages)
+ zonelist->zones[zone_num++] = zone;
+ }
+ return zone_num;
+}
+
+/* Builds a binding for a region of memory, based on a bitmask of nodes. */
+static inline int build_binding(unsigned long * nodemask, struct binding *binding)
+{
+ int node, zone_num;
+
+ memset(binding, 0, sizeof(struct binding));
+
+ /* Build binding zonelist */
+ for (node = 0, zone_num = 0; node < MAX_NUMNODES; node++)
+ if (test_bit(node, nodemask) && node_online(node))
+ zone_num = add_node(NODE_DATA(node),
+ &binding->zonelist, zone_num);
+ binding->zonelist.zones[zone_num] = NULL;
+
+ if (zone_num == 0)
+ /* No zones were added to the zonelist. Let the caller know. */
+ return -EINVAL;
+
+ return 0;
+}
+
+
+/*
+ * mbind - Bind a range of a process' VM space to a set of memory blocks according to
+ * a predefined policy.
+ * @start: beginning address of memory region to bind
+ * @len: length of memory region to bind
+ * @mask_ptr: pointer to bitmask of cpus
+ * @mask_len: length of the bitmask
+ * @policy: flag specifying the policy to use for the segment
+ */
+asmlinkage unsigned long sys_mbind(unsigned long start, unsigned long len,
+ unsigned long *mask_ptr, unsigned int mask_len, unsigned long policy)
+{
+ DECLARE_BITMAP(cpu_mask, NR_CPUS);
+ DECLARE_BITMAP(node_mask, MAX_NUMNODES);
+ struct vm_area_struct *vma = NULL;
+ struct address_space *mapping;
+ int copy_len, error = 0;
+
+ /* Deal with getting cpu_mask from userspace & translating to node_mask */
+ CLEAR_BITMAP(cpu_mask, NR_CPUS);
+ CLEAR_BITMAP(node_mask, MAX_NUMNODES);
+ copy_len = min(mask_len, (unsigned int)NR_CPUS);
+ if (copy_from_user(cpu_mask, mask_ptr, (copy_len+7)/8)) {
+ error = -EFAULT;
+ goto out;
+ }
+ cpumask_to_nodemask(cpu_mask, node_mask);
+
+ down_read(¤t->mm->mmap_sem);
+ vma = find_vma(current->mm, start);
+ up_read(¤t->mm->mmap_sem);
+ /* This is an ugly, gross hack. This is purely because I've hurt my
+ * brain trying to come up with a brilliant way of implementing this
+ * for VMA's in general. Shared Memory VMA's lend themselves to binding
+ * both because of how they're implemented, and their actual uses.
+ * If anyone has a great place to squirrel-away some data about the
+ * requested binding, and a way to easily force the allocator to respect
+ * these bindings, then send a patch, or let me know. Otherwise, this
+ * will have to wait for a stroke of insight.
+ */
+ if (!(vma && vma->vm_file && vma->vm_ops &&
+ vma->vm_ops->nopage == shmem_nopage)) {
+ /* This isn't a shm segment. For now, we bail. */
+ error = -EINVAL;
+ goto out;
+ }
+
+ mapping = vma->vm_file->f_dentry->d_inode->i_mapping;
+ if (mapping->binding) {
+ kfree(mapping->binding);
+ mapping->binding = NULL;
+ }
+ mapping->binding = kmalloc(sizeof(struct binding), GFP_KERNEL);
+ if (!mapping->binding) {
+ error = -ENOMEM;
+ goto out;
+ }
+ error = build_binding(node_mask, mapping->binding);
+ if (error) {
+ kfree(mapping->binding);
+ mapping->binding = NULL;
+ }
+
+out:
+ return error;
+}