Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
| 2 | * Simple NUMA memory policy for the Linux kernel. |
| 3 | * |
| 4 | * Copyright 2003,2004 Andi Kleen, SuSE Labs. |
Christoph Lameter | 8bccd85 | 2005-10-29 18:16:59 -0700 | [diff] [blame] | 5 | * (C) Copyright 2005 Christoph Lameter, Silicon Graphics, Inc. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6 | * Subject to the GNU Public License, version 2. |
| 7 | * |
| 8 | * NUMA policy allows the user to give hints in which node(s) memory should |
| 9 | * be allocated. |
| 10 | * |
| 11 | * Support four policies per VMA and per process: |
| 12 | * |
| 13 | * The VMA policy has priority over the process policy for a page fault. |
| 14 | * |
| 15 | * interleave Allocate memory interleaved over a set of nodes, |
| 16 | * with normal fallback if it fails. |
| 17 | * For VMA based allocations this interleaves based on the |
| 18 | * offset into the backing object or offset into the mapping |
| 19 | * for anonymous memory. For process policy an process counter |
| 20 | * is used. |
Christoph Lameter | 8bccd85 | 2005-10-29 18:16:59 -0700 | [diff] [blame] | 21 | * |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 22 | * bind Only allocate memory on a specific set of nodes, |
| 23 | * no fallback. |
Christoph Lameter | 8bccd85 | 2005-10-29 18:16:59 -0700 | [diff] [blame] | 24 | * FIXME: memory is allocated starting with the first node |
| 25 | * to the last. It would be better if bind would truly restrict |
| 26 | * the allocation to memory nodes instead |
| 27 | * |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 28 | * preferred Try a specific node first before normal fallback. |
| 29 | * As a special case node -1 here means do the allocation |
| 30 | * on the local CPU. This is normally identical to default, |
| 31 | * but useful to set in a VMA when you have a non default |
| 32 | * process policy. |
Christoph Lameter | 8bccd85 | 2005-10-29 18:16:59 -0700 | [diff] [blame] | 33 | * |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 34 | * default Allocate on the local node first, or when on a VMA |
| 35 | * use the process policy. This is what Linux always did |
| 36 | * in a NUMA aware kernel and still does by, ahem, default. |
| 37 | * |
| 38 | * The process policy is applied for most non interrupt memory allocations |
| 39 | * in that process' context. Interrupts ignore the policies and always |
| 40 | * try to allocate on the local CPU. The VMA policy is only applied for memory |
| 41 | * allocations for a VMA in the VM. |
| 42 | * |
| 43 | * Currently there are a few corner cases in swapping where the policy |
| 44 | * is not applied, but the majority should be handled. When process policy |
| 45 | * is used it is not remembered over swap outs/swap ins. |
| 46 | * |
| 47 | * Only the highest zone in the zone hierarchy gets policied. Allocations |
| 48 | * requesting a lower zone just use default policy. This implies that |
| 49 | * on systems with highmem kernel lowmem allocation don't get policied. |
| 50 | * Same with GFP_DMA allocations. |
| 51 | * |
| 52 | * For shmfs/tmpfs/hugetlbfs shared memory the policy is shared between |
| 53 | * all users and remembered even when nobody has memory mapped. |
| 54 | */ |
| 55 | |
| 56 | /* Notebook: |
| 57 | fix mmap readahead to honour policy and enable policy for any page cache |
| 58 | object |
| 59 | statistics for bigpages |
| 60 | global policy for page cache? currently it uses process policy. Requires |
| 61 | first item above. |
| 62 | handle mremap for shared memory (currently ignored for the policy) |
| 63 | grows down? |
| 64 | make bind policy root only? It can trigger oom much faster and the |
| 65 | kernel is not always grateful with that. |
| 66 | could replace all the switch()es with a mempolicy_ops structure. |
| 67 | */ |
| 68 | |
| 69 | #include <linux/mempolicy.h> |
| 70 | #include <linux/mm.h> |
| 71 | #include <linux/highmem.h> |
| 72 | #include <linux/hugetlb.h> |
| 73 | #include <linux/kernel.h> |
| 74 | #include <linux/sched.h> |
| 75 | #include <linux/mm.h> |
| 76 | #include <linux/nodemask.h> |
| 77 | #include <linux/cpuset.h> |
| 78 | #include <linux/gfp.h> |
| 79 | #include <linux/slab.h> |
| 80 | #include <linux/string.h> |
| 81 | #include <linux/module.h> |
| 82 | #include <linux/interrupt.h> |
| 83 | #include <linux/init.h> |
| 84 | #include <linux/compat.h> |
| 85 | #include <linux/mempolicy.h> |
Christoph Lameter | dc9aa5b | 2006-01-08 01:00:50 -0800 | [diff] [blame] | 86 | #include <linux/swap.h> |
Christoph Lameter | 1a75a6c | 2006-01-08 01:01:02 -0800 | [diff] [blame] | 87 | #include <linux/seq_file.h> |
| 88 | #include <linux/proc_fs.h> |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 89 | #include <linux/migrate.h> |
Christoph Lameter | 95a402c | 2006-06-23 02:03:53 -0700 | [diff] [blame] | 90 | #include <linux/rmap.h> |
David Quigley | 86c3a76 | 2006-06-23 02:04:02 -0700 | [diff] [blame] | 91 | #include <linux/security.h> |
Christoph Lameter | dc9aa5b | 2006-01-08 01:00:50 -0800 | [diff] [blame] | 92 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 93 | #include <asm/tlbflush.h> |
| 94 | #include <asm/uaccess.h> |
| 95 | |
Christoph Lameter | 38e3586 | 2006-01-08 01:01:01 -0800 | [diff] [blame] | 96 | /* Internal flags */ |
Christoph Lameter | dc9aa5b | 2006-01-08 01:00:50 -0800 | [diff] [blame] | 97 | #define MPOL_MF_DISCONTIG_OK (MPOL_MF_INTERNAL << 0) /* Skip checks for continuous vmas */ |
Christoph Lameter | 38e3586 | 2006-01-08 01:01:01 -0800 | [diff] [blame] | 98 | #define MPOL_MF_INVERT (MPOL_MF_INTERNAL << 1) /* Invert check for nodemask */ |
Christoph Lameter | 1a75a6c | 2006-01-08 01:01:02 -0800 | [diff] [blame] | 99 | #define MPOL_MF_STATS (MPOL_MF_INTERNAL << 2) /* Gather statistics */ |
Christoph Lameter | dc9aa5b | 2006-01-08 01:00:50 -0800 | [diff] [blame] | 100 | |
Pekka Enberg | fcc234f | 2006-03-22 00:08:13 -0800 | [diff] [blame] | 101 | static struct kmem_cache *policy_cache; |
| 102 | static struct kmem_cache *sn_cache; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 103 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 104 | /* Highest zone. An specific allocation for a zone below that is not |
| 105 | policied. */ |
Christoph Lameter | 6267276 | 2007-02-10 01:43:07 -0800 | [diff] [blame] | 106 | enum zone_type policy_zone = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 107 | |
Andi Kleen | d42c699 | 2005-07-06 19:56:03 +0200 | [diff] [blame] | 108 | struct mempolicy default_policy = { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 109 | .refcnt = ATOMIC_INIT(1), /* never free it */ |
| 110 | .policy = MPOL_DEFAULT, |
| 111 | }; |
| 112 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 113 | /* Do sanity checking on a policy */ |
Andi Kleen | dfcd3c0 | 2005-10-29 18:15:48 -0700 | [diff] [blame] | 114 | static int mpol_check_policy(int mode, nodemask_t *nodes) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 115 | { |
Andi Kleen | dfcd3c0 | 2005-10-29 18:15:48 -0700 | [diff] [blame] | 116 | int empty = nodes_empty(*nodes); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 117 | |
| 118 | switch (mode) { |
| 119 | case MPOL_DEFAULT: |
| 120 | if (!empty) |
| 121 | return -EINVAL; |
| 122 | break; |
| 123 | case MPOL_BIND: |
| 124 | case MPOL_INTERLEAVE: |
| 125 | /* Preferred will only use the first bit, but allow |
| 126 | more for now. */ |
| 127 | if (empty) |
| 128 | return -EINVAL; |
| 129 | break; |
| 130 | } |
Andi Kleen | dfcd3c0 | 2005-10-29 18:15:48 -0700 | [diff] [blame] | 131 | return nodes_subset(*nodes, node_online_map) ? 0 : -EINVAL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 132 | } |
Andi Kleen | dd942ae | 2006-02-17 01:39:16 +0100 | [diff] [blame] | 133 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 134 | /* Generate a custom zonelist for the BIND policy. */ |
Andi Kleen | dfcd3c0 | 2005-10-29 18:15:48 -0700 | [diff] [blame] | 135 | static struct zonelist *bind_zonelist(nodemask_t *nodes) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 136 | { |
| 137 | struct zonelist *zl; |
Christoph Lameter | 2f6726e | 2006-09-25 23:31:18 -0700 | [diff] [blame] | 138 | int num, max, nd; |
| 139 | enum zone_type k; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 140 | |
Andi Kleen | dfcd3c0 | 2005-10-29 18:15:48 -0700 | [diff] [blame] | 141 | max = 1 + MAX_NR_ZONES * nodes_weight(*nodes); |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 142 | max++; /* space for zlcache_ptr (see mmzone.h) */ |
Andi Kleen | dd942ae | 2006-02-17 01:39:16 +0100 | [diff] [blame] | 143 | zl = kmalloc(sizeof(struct zone *) * max, GFP_KERNEL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 144 | if (!zl) |
KAMEZAWA Hiroyuki | 8af5e2e | 2007-02-20 13:57:49 -0800 | [diff] [blame] | 145 | return ERR_PTR(-ENOMEM); |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 146 | zl->zlcache_ptr = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 147 | num = 0; |
Andi Kleen | dd942ae | 2006-02-17 01:39:16 +0100 | [diff] [blame] | 148 | /* First put in the highest zones from all nodes, then all the next |
| 149 | lower zones etc. Avoid empty zones because the memory allocator |
| 150 | doesn't like them. If you implement node hot removal you |
| 151 | have to fix that. */ |
Christoph Lameter | 2f6726e | 2006-09-25 23:31:18 -0700 | [diff] [blame] | 152 | k = policy_zone; |
| 153 | while (1) { |
Andi Kleen | dd942ae | 2006-02-17 01:39:16 +0100 | [diff] [blame] | 154 | for_each_node_mask(nd, *nodes) { |
| 155 | struct zone *z = &NODE_DATA(nd)->node_zones[k]; |
| 156 | if (z->present_pages > 0) |
| 157 | zl->zones[num++] = z; |
| 158 | } |
Christoph Lameter | 2f6726e | 2006-09-25 23:31:18 -0700 | [diff] [blame] | 159 | if (k == 0) |
| 160 | break; |
| 161 | k--; |
Andi Kleen | dd942ae | 2006-02-17 01:39:16 +0100 | [diff] [blame] | 162 | } |
KAMEZAWA Hiroyuki | 8af5e2e | 2007-02-20 13:57:49 -0800 | [diff] [blame] | 163 | if (num == 0) { |
| 164 | kfree(zl); |
| 165 | return ERR_PTR(-EINVAL); |
| 166 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 167 | zl->zones[num] = NULL; |
| 168 | return zl; |
| 169 | } |
| 170 | |
| 171 | /* Create a new policy */ |
Andi Kleen | dfcd3c0 | 2005-10-29 18:15:48 -0700 | [diff] [blame] | 172 | static struct mempolicy *mpol_new(int mode, nodemask_t *nodes) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 173 | { |
| 174 | struct mempolicy *policy; |
| 175 | |
Paul Mundt | 140d5a4 | 2007-07-15 23:38:16 -0700 | [diff] [blame] | 176 | pr_debug("setting mode %d nodes[0] %lx\n", |
| 177 | mode, nodes ? nodes_addr(*nodes)[0] : -1); |
| 178 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 179 | if (mode == MPOL_DEFAULT) |
| 180 | return NULL; |
| 181 | policy = kmem_cache_alloc(policy_cache, GFP_KERNEL); |
| 182 | if (!policy) |
| 183 | return ERR_PTR(-ENOMEM); |
| 184 | atomic_set(&policy->refcnt, 1); |
| 185 | switch (mode) { |
| 186 | case MPOL_INTERLEAVE: |
Andi Kleen | dfcd3c0 | 2005-10-29 18:15:48 -0700 | [diff] [blame] | 187 | policy->v.nodes = *nodes; |
Andi Kleen | 8f493d7 | 2006-01-03 00:07:28 +0100 | [diff] [blame] | 188 | if (nodes_weight(*nodes) == 0) { |
| 189 | kmem_cache_free(policy_cache, policy); |
| 190 | return ERR_PTR(-EINVAL); |
| 191 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 192 | break; |
| 193 | case MPOL_PREFERRED: |
Andi Kleen | dfcd3c0 | 2005-10-29 18:15:48 -0700 | [diff] [blame] | 194 | policy->v.preferred_node = first_node(*nodes); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 195 | if (policy->v.preferred_node >= MAX_NUMNODES) |
| 196 | policy->v.preferred_node = -1; |
| 197 | break; |
| 198 | case MPOL_BIND: |
| 199 | policy->v.zonelist = bind_zonelist(nodes); |
KAMEZAWA Hiroyuki | 8af5e2e | 2007-02-20 13:57:49 -0800 | [diff] [blame] | 200 | if (IS_ERR(policy->v.zonelist)) { |
| 201 | void *error_code = policy->v.zonelist; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 202 | kmem_cache_free(policy_cache, policy); |
KAMEZAWA Hiroyuki | 8af5e2e | 2007-02-20 13:57:49 -0800 | [diff] [blame] | 203 | return error_code; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 204 | } |
| 205 | break; |
| 206 | } |
| 207 | policy->policy = mode; |
Paul Jackson | 74cb215 | 2006-01-08 01:01:56 -0800 | [diff] [blame] | 208 | policy->cpuset_mems_allowed = cpuset_mems_allowed(current); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 209 | return policy; |
| 210 | } |
| 211 | |
Christoph Lameter | 397874d | 2006-03-06 15:42:53 -0800 | [diff] [blame] | 212 | static void gather_stats(struct page *, void *, int pte_dirty); |
Christoph Lameter | fc30128 | 2006-01-18 17:42:29 -0800 | [diff] [blame] | 213 | static void migrate_page_add(struct page *page, struct list_head *pagelist, |
| 214 | unsigned long flags); |
Christoph Lameter | 1a75a6c | 2006-01-08 01:01:02 -0800 | [diff] [blame] | 215 | |
Christoph Lameter | 38e3586 | 2006-01-08 01:01:01 -0800 | [diff] [blame] | 216 | /* Scan through pages checking if pages follow certain conditions. */ |
Nick Piggin | b581003 | 2005-10-29 18:16:12 -0700 | [diff] [blame] | 217 | static int check_pte_range(struct vm_area_struct *vma, pmd_t *pmd, |
Christoph Lameter | dc9aa5b | 2006-01-08 01:00:50 -0800 | [diff] [blame] | 218 | unsigned long addr, unsigned long end, |
| 219 | const nodemask_t *nodes, unsigned long flags, |
Christoph Lameter | 38e3586 | 2006-01-08 01:01:01 -0800 | [diff] [blame] | 220 | void *private) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 221 | { |
Hugh Dickins | 91612e0 | 2005-06-21 17:15:07 -0700 | [diff] [blame] | 222 | pte_t *orig_pte; |
| 223 | pte_t *pte; |
Hugh Dickins | 705e87c | 2005-10-29 18:16:27 -0700 | [diff] [blame] | 224 | spinlock_t *ptl; |
Hugh Dickins | 941150a | 2005-06-21 17:15:06 -0700 | [diff] [blame] | 225 | |
Hugh Dickins | 705e87c | 2005-10-29 18:16:27 -0700 | [diff] [blame] | 226 | orig_pte = pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
Hugh Dickins | 91612e0 | 2005-06-21 17:15:07 -0700 | [diff] [blame] | 227 | do { |
Linus Torvalds | 6aab341 | 2005-11-28 14:34:23 -0800 | [diff] [blame] | 228 | struct page *page; |
Andy Whitcroft | 25ba77c | 2006-12-06 20:33:03 -0800 | [diff] [blame] | 229 | int nid; |
Hugh Dickins | 91612e0 | 2005-06-21 17:15:07 -0700 | [diff] [blame] | 230 | |
| 231 | if (!pte_present(*pte)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 232 | continue; |
Linus Torvalds | 6aab341 | 2005-11-28 14:34:23 -0800 | [diff] [blame] | 233 | page = vm_normal_page(vma, addr, *pte); |
| 234 | if (!page) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 235 | continue; |
Nick Piggin | 053837f | 2006-01-18 17:42:27 -0800 | [diff] [blame] | 236 | /* |
| 237 | * The check for PageReserved here is important to avoid |
| 238 | * handling zero pages and other pages that may have been |
| 239 | * marked special by the system. |
| 240 | * |
| 241 | * If the PageReserved would not be checked here then f.e. |
| 242 | * the location of the zero page could have an influence |
| 243 | * on MPOL_MF_STRICT, zero pages would be counted for |
| 244 | * the per node stats, and there would be useless attempts |
| 245 | * to put zero pages on the migration list. |
| 246 | */ |
Christoph Lameter | f4598c8 | 2006-01-12 01:05:20 -0800 | [diff] [blame] | 247 | if (PageReserved(page)) |
| 248 | continue; |
Linus Torvalds | 6aab341 | 2005-11-28 14:34:23 -0800 | [diff] [blame] | 249 | nid = page_to_nid(page); |
Christoph Lameter | 38e3586 | 2006-01-08 01:01:01 -0800 | [diff] [blame] | 250 | if (node_isset(nid, *nodes) == !!(flags & MPOL_MF_INVERT)) |
| 251 | continue; |
| 252 | |
Christoph Lameter | 1a75a6c | 2006-01-08 01:01:02 -0800 | [diff] [blame] | 253 | if (flags & MPOL_MF_STATS) |
Christoph Lameter | 397874d | 2006-03-06 15:42:53 -0800 | [diff] [blame] | 254 | gather_stats(page, private, pte_dirty(*pte)); |
Nick Piggin | 053837f | 2006-01-18 17:42:27 -0800 | [diff] [blame] | 255 | else if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) |
Christoph Lameter | fc30128 | 2006-01-18 17:42:29 -0800 | [diff] [blame] | 256 | migrate_page_add(page, private, flags); |
Christoph Lameter | 38e3586 | 2006-01-08 01:01:01 -0800 | [diff] [blame] | 257 | else |
| 258 | break; |
Hugh Dickins | 91612e0 | 2005-06-21 17:15:07 -0700 | [diff] [blame] | 259 | } while (pte++, addr += PAGE_SIZE, addr != end); |
Hugh Dickins | 705e87c | 2005-10-29 18:16:27 -0700 | [diff] [blame] | 260 | pte_unmap_unlock(orig_pte, ptl); |
Hugh Dickins | 91612e0 | 2005-06-21 17:15:07 -0700 | [diff] [blame] | 261 | return addr != end; |
| 262 | } |
| 263 | |
Nick Piggin | b581003 | 2005-10-29 18:16:12 -0700 | [diff] [blame] | 264 | static inline int check_pmd_range(struct vm_area_struct *vma, pud_t *pud, |
Christoph Lameter | dc9aa5b | 2006-01-08 01:00:50 -0800 | [diff] [blame] | 265 | unsigned long addr, unsigned long end, |
| 266 | const nodemask_t *nodes, unsigned long flags, |
Christoph Lameter | 38e3586 | 2006-01-08 01:01:01 -0800 | [diff] [blame] | 267 | void *private) |
Hugh Dickins | 91612e0 | 2005-06-21 17:15:07 -0700 | [diff] [blame] | 268 | { |
| 269 | pmd_t *pmd; |
| 270 | unsigned long next; |
| 271 | |
| 272 | pmd = pmd_offset(pud, addr); |
| 273 | do { |
| 274 | next = pmd_addr_end(addr, end); |
| 275 | if (pmd_none_or_clear_bad(pmd)) |
| 276 | continue; |
Christoph Lameter | dc9aa5b | 2006-01-08 01:00:50 -0800 | [diff] [blame] | 277 | if (check_pte_range(vma, pmd, addr, next, nodes, |
Christoph Lameter | 38e3586 | 2006-01-08 01:01:01 -0800 | [diff] [blame] | 278 | flags, private)) |
Hugh Dickins | 91612e0 | 2005-06-21 17:15:07 -0700 | [diff] [blame] | 279 | return -EIO; |
| 280 | } while (pmd++, addr = next, addr != end); |
| 281 | return 0; |
| 282 | } |
| 283 | |
Nick Piggin | b581003 | 2005-10-29 18:16:12 -0700 | [diff] [blame] | 284 | static inline int check_pud_range(struct vm_area_struct *vma, pgd_t *pgd, |
Christoph Lameter | dc9aa5b | 2006-01-08 01:00:50 -0800 | [diff] [blame] | 285 | unsigned long addr, unsigned long end, |
| 286 | const nodemask_t *nodes, unsigned long flags, |
Christoph Lameter | 38e3586 | 2006-01-08 01:01:01 -0800 | [diff] [blame] | 287 | void *private) |
Hugh Dickins | 91612e0 | 2005-06-21 17:15:07 -0700 | [diff] [blame] | 288 | { |
| 289 | pud_t *pud; |
| 290 | unsigned long next; |
| 291 | |
| 292 | pud = pud_offset(pgd, addr); |
| 293 | do { |
| 294 | next = pud_addr_end(addr, end); |
| 295 | if (pud_none_or_clear_bad(pud)) |
| 296 | continue; |
Christoph Lameter | dc9aa5b | 2006-01-08 01:00:50 -0800 | [diff] [blame] | 297 | if (check_pmd_range(vma, pud, addr, next, nodes, |
Christoph Lameter | 38e3586 | 2006-01-08 01:01:01 -0800 | [diff] [blame] | 298 | flags, private)) |
Hugh Dickins | 91612e0 | 2005-06-21 17:15:07 -0700 | [diff] [blame] | 299 | return -EIO; |
| 300 | } while (pud++, addr = next, addr != end); |
| 301 | return 0; |
| 302 | } |
| 303 | |
Nick Piggin | b581003 | 2005-10-29 18:16:12 -0700 | [diff] [blame] | 304 | static inline int check_pgd_range(struct vm_area_struct *vma, |
Christoph Lameter | dc9aa5b | 2006-01-08 01:00:50 -0800 | [diff] [blame] | 305 | unsigned long addr, unsigned long end, |
| 306 | const nodemask_t *nodes, unsigned long flags, |
Christoph Lameter | 38e3586 | 2006-01-08 01:01:01 -0800 | [diff] [blame] | 307 | void *private) |
Hugh Dickins | 91612e0 | 2005-06-21 17:15:07 -0700 | [diff] [blame] | 308 | { |
| 309 | pgd_t *pgd; |
| 310 | unsigned long next; |
| 311 | |
Nick Piggin | b581003 | 2005-10-29 18:16:12 -0700 | [diff] [blame] | 312 | pgd = pgd_offset(vma->vm_mm, addr); |
Hugh Dickins | 91612e0 | 2005-06-21 17:15:07 -0700 | [diff] [blame] | 313 | do { |
| 314 | next = pgd_addr_end(addr, end); |
| 315 | if (pgd_none_or_clear_bad(pgd)) |
| 316 | continue; |
Christoph Lameter | dc9aa5b | 2006-01-08 01:00:50 -0800 | [diff] [blame] | 317 | if (check_pud_range(vma, pgd, addr, next, nodes, |
Christoph Lameter | 38e3586 | 2006-01-08 01:01:01 -0800 | [diff] [blame] | 318 | flags, private)) |
Hugh Dickins | 91612e0 | 2005-06-21 17:15:07 -0700 | [diff] [blame] | 319 | return -EIO; |
| 320 | } while (pgd++, addr = next, addr != end); |
| 321 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 322 | } |
| 323 | |
Christoph Lameter | dc9aa5b | 2006-01-08 01:00:50 -0800 | [diff] [blame] | 324 | /* |
| 325 | * Check if all pages in a range are on a set of nodes. |
| 326 | * If pagelist != NULL then isolate pages from the LRU and |
| 327 | * put them on the pagelist. |
| 328 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 329 | static struct vm_area_struct * |
| 330 | check_range(struct mm_struct *mm, unsigned long start, unsigned long end, |
Christoph Lameter | 38e3586 | 2006-01-08 01:01:01 -0800 | [diff] [blame] | 331 | const nodemask_t *nodes, unsigned long flags, void *private) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 332 | { |
| 333 | int err; |
| 334 | struct vm_area_struct *first, *vma, *prev; |
| 335 | |
Christoph Lameter | 90036ee | 2006-03-16 23:03:59 -0800 | [diff] [blame] | 336 | if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) { |
Christoph Lameter | 90036ee | 2006-03-16 23:03:59 -0800 | [diff] [blame] | 337 | |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 338 | err = migrate_prep(); |
| 339 | if (err) |
| 340 | return ERR_PTR(err); |
Christoph Lameter | 90036ee | 2006-03-16 23:03:59 -0800 | [diff] [blame] | 341 | } |
Nick Piggin | 053837f | 2006-01-18 17:42:27 -0800 | [diff] [blame] | 342 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 343 | first = find_vma(mm, start); |
| 344 | if (!first) |
| 345 | return ERR_PTR(-EFAULT); |
| 346 | prev = NULL; |
| 347 | for (vma = first; vma && vma->vm_start < end; vma = vma->vm_next) { |
Christoph Lameter | dc9aa5b | 2006-01-08 01:00:50 -0800 | [diff] [blame] | 348 | if (!(flags & MPOL_MF_DISCONTIG_OK)) { |
| 349 | if (!vma->vm_next && vma->vm_end < end) |
| 350 | return ERR_PTR(-EFAULT); |
| 351 | if (prev && prev->vm_end < vma->vm_start) |
| 352 | return ERR_PTR(-EFAULT); |
| 353 | } |
| 354 | if (!is_vm_hugetlb_page(vma) && |
| 355 | ((flags & MPOL_MF_STRICT) || |
| 356 | ((flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) && |
| 357 | vma_migratable(vma)))) { |
Andi Kleen | 5b952b3 | 2005-09-13 01:25:08 -0700 | [diff] [blame] | 358 | unsigned long endvma = vma->vm_end; |
Christoph Lameter | dc9aa5b | 2006-01-08 01:00:50 -0800 | [diff] [blame] | 359 | |
Andi Kleen | 5b952b3 | 2005-09-13 01:25:08 -0700 | [diff] [blame] | 360 | if (endvma > end) |
| 361 | endvma = end; |
| 362 | if (vma->vm_start > start) |
| 363 | start = vma->vm_start; |
Christoph Lameter | dc9aa5b | 2006-01-08 01:00:50 -0800 | [diff] [blame] | 364 | err = check_pgd_range(vma, start, endvma, nodes, |
Christoph Lameter | 38e3586 | 2006-01-08 01:01:01 -0800 | [diff] [blame] | 365 | flags, private); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 366 | if (err) { |
| 367 | first = ERR_PTR(err); |
| 368 | break; |
| 369 | } |
| 370 | } |
| 371 | prev = vma; |
| 372 | } |
| 373 | return first; |
| 374 | } |
| 375 | |
| 376 | /* Apply policy to a single VMA */ |
| 377 | static int policy_vma(struct vm_area_struct *vma, struct mempolicy *new) |
| 378 | { |
| 379 | int err = 0; |
| 380 | struct mempolicy *old = vma->vm_policy; |
| 381 | |
Paul Mundt | 140d5a4 | 2007-07-15 23:38:16 -0700 | [diff] [blame] | 382 | pr_debug("vma %lx-%lx/%lx vm_ops %p vm_file %p set_policy %p\n", |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 383 | vma->vm_start, vma->vm_end, vma->vm_pgoff, |
| 384 | vma->vm_ops, vma->vm_file, |
| 385 | vma->vm_ops ? vma->vm_ops->set_policy : NULL); |
| 386 | |
| 387 | if (vma->vm_ops && vma->vm_ops->set_policy) |
| 388 | err = vma->vm_ops->set_policy(vma, new); |
| 389 | if (!err) { |
| 390 | mpol_get(new); |
| 391 | vma->vm_policy = new; |
| 392 | mpol_free(old); |
| 393 | } |
| 394 | return err; |
| 395 | } |
| 396 | |
| 397 | /* Step 2: apply policy to a range and do splits. */ |
| 398 | static int mbind_range(struct vm_area_struct *vma, unsigned long start, |
| 399 | unsigned long end, struct mempolicy *new) |
| 400 | { |
| 401 | struct vm_area_struct *next; |
| 402 | int err; |
| 403 | |
| 404 | err = 0; |
| 405 | for (; vma && vma->vm_start < end; vma = next) { |
| 406 | next = vma->vm_next; |
| 407 | if (vma->vm_start < start) |
| 408 | err = split_vma(vma->vm_mm, vma, start, 1); |
| 409 | if (!err && vma->vm_end > end) |
| 410 | err = split_vma(vma->vm_mm, vma, end, 0); |
| 411 | if (!err) |
| 412 | err = policy_vma(vma, new); |
| 413 | if (err) |
| 414 | break; |
| 415 | } |
| 416 | return err; |
| 417 | } |
| 418 | |
Christoph Lameter | 8bccd85 | 2005-10-29 18:16:59 -0700 | [diff] [blame] | 419 | static int contextualize_policy(int mode, nodemask_t *nodes) |
| 420 | { |
| 421 | if (!nodes) |
| 422 | return 0; |
| 423 | |
Paul Jackson | cf2a473 | 2006-01-08 01:01:54 -0800 | [diff] [blame] | 424 | cpuset_update_task_memory_state(); |
Paul Jackson | 5966514 | 2006-01-08 01:01:47 -0800 | [diff] [blame] | 425 | if (!cpuset_nodes_subset_current_mems_allowed(*nodes)) |
| 426 | return -EINVAL; |
Christoph Lameter | 8bccd85 | 2005-10-29 18:16:59 -0700 | [diff] [blame] | 427 | return mpol_check_policy(mode, nodes); |
| 428 | } |
| 429 | |
Paul Jackson | c61afb1 | 2006-03-24 03:16:08 -0800 | [diff] [blame] | 430 | |
| 431 | /* |
| 432 | * Update task->flags PF_MEMPOLICY bit: set iff non-default |
| 433 | * mempolicy. Allows more rapid checking of this (combined perhaps |
| 434 | * with other PF_* flag bits) on memory allocation hot code paths. |
| 435 | * |
| 436 | * If called from outside this file, the task 'p' should -only- be |
| 437 | * a newly forked child not yet visible on the task list, because |
| 438 | * manipulating the task flags of a visible task is not safe. |
| 439 | * |
| 440 | * The above limitation is why this routine has the funny name |
| 441 | * mpol_fix_fork_child_flag(). |
| 442 | * |
| 443 | * It is also safe to call this with a task pointer of current, |
| 444 | * which the static wrapper mpol_set_task_struct_flag() does, |
| 445 | * for use within this file. |
| 446 | */ |
| 447 | |
| 448 | void mpol_fix_fork_child_flag(struct task_struct *p) |
| 449 | { |
| 450 | if (p->mempolicy) |
| 451 | p->flags |= PF_MEMPOLICY; |
| 452 | else |
| 453 | p->flags &= ~PF_MEMPOLICY; |
| 454 | } |
| 455 | |
| 456 | static void mpol_set_task_struct_flag(void) |
| 457 | { |
| 458 | mpol_fix_fork_child_flag(current); |
| 459 | } |
| 460 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 461 | /* Set the process memory policy */ |
Christoph Lameter | 8bccd85 | 2005-10-29 18:16:59 -0700 | [diff] [blame] | 462 | long do_set_mempolicy(int mode, nodemask_t *nodes) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 463 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 464 | struct mempolicy *new; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 465 | |
Christoph Lameter | 8bccd85 | 2005-10-29 18:16:59 -0700 | [diff] [blame] | 466 | if (contextualize_policy(mode, nodes)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 467 | return -EINVAL; |
Christoph Lameter | 8bccd85 | 2005-10-29 18:16:59 -0700 | [diff] [blame] | 468 | new = mpol_new(mode, nodes); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 469 | if (IS_ERR(new)) |
| 470 | return PTR_ERR(new); |
| 471 | mpol_free(current->mempolicy); |
| 472 | current->mempolicy = new; |
Paul Jackson | c61afb1 | 2006-03-24 03:16:08 -0800 | [diff] [blame] | 473 | mpol_set_task_struct_flag(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 474 | if (new && new->policy == MPOL_INTERLEAVE) |
Andi Kleen | dfcd3c0 | 2005-10-29 18:15:48 -0700 | [diff] [blame] | 475 | current->il_next = first_node(new->v.nodes); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 476 | return 0; |
| 477 | } |
| 478 | |
| 479 | /* Fill a zone bitmap for a policy */ |
Andi Kleen | dfcd3c0 | 2005-10-29 18:15:48 -0700 | [diff] [blame] | 480 | static void get_zonemask(struct mempolicy *p, nodemask_t *nodes) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 481 | { |
| 482 | int i; |
| 483 | |
Andi Kleen | dfcd3c0 | 2005-10-29 18:15:48 -0700 | [diff] [blame] | 484 | nodes_clear(*nodes); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 485 | switch (p->policy) { |
| 486 | case MPOL_BIND: |
| 487 | for (i = 0; p->v.zonelist->zones[i]; i++) |
Christoph Lameter | 89fa302 | 2006-09-25 23:31:55 -0700 | [diff] [blame] | 488 | node_set(zone_to_nid(p->v.zonelist->zones[i]), |
Christoph Lameter | 8bccd85 | 2005-10-29 18:16:59 -0700 | [diff] [blame] | 489 | *nodes); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 490 | break; |
| 491 | case MPOL_DEFAULT: |
| 492 | break; |
| 493 | case MPOL_INTERLEAVE: |
Andi Kleen | dfcd3c0 | 2005-10-29 18:15:48 -0700 | [diff] [blame] | 494 | *nodes = p->v.nodes; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 495 | break; |
| 496 | case MPOL_PREFERRED: |
| 497 | /* or use current node instead of online map? */ |
| 498 | if (p->v.preferred_node < 0) |
Andi Kleen | dfcd3c0 | 2005-10-29 18:15:48 -0700 | [diff] [blame] | 499 | *nodes = node_online_map; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 500 | else |
Andi Kleen | dfcd3c0 | 2005-10-29 18:15:48 -0700 | [diff] [blame] | 501 | node_set(p->v.preferred_node, *nodes); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 502 | break; |
| 503 | default: |
| 504 | BUG(); |
| 505 | } |
| 506 | } |
| 507 | |
| 508 | static int lookup_node(struct mm_struct *mm, unsigned long addr) |
| 509 | { |
| 510 | struct page *p; |
| 511 | int err; |
| 512 | |
| 513 | err = get_user_pages(current, mm, addr & PAGE_MASK, 1, 0, 0, &p, NULL); |
| 514 | if (err >= 0) { |
| 515 | err = page_to_nid(p); |
| 516 | put_page(p); |
| 517 | } |
| 518 | return err; |
| 519 | } |
| 520 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 521 | /* Retrieve NUMA policy */ |
Christoph Lameter | 8bccd85 | 2005-10-29 18:16:59 -0700 | [diff] [blame] | 522 | long do_get_mempolicy(int *policy, nodemask_t *nmask, |
| 523 | unsigned long addr, unsigned long flags) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 524 | { |
Christoph Lameter | 8bccd85 | 2005-10-29 18:16:59 -0700 | [diff] [blame] | 525 | int err; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 526 | struct mm_struct *mm = current->mm; |
| 527 | struct vm_area_struct *vma = NULL; |
| 528 | struct mempolicy *pol = current->mempolicy; |
| 529 | |
Paul Jackson | cf2a473 | 2006-01-08 01:01:54 -0800 | [diff] [blame] | 530 | cpuset_update_task_memory_state(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 531 | if (flags & ~(unsigned long)(MPOL_F_NODE|MPOL_F_ADDR)) |
| 532 | return -EINVAL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 533 | if (flags & MPOL_F_ADDR) { |
| 534 | down_read(&mm->mmap_sem); |
| 535 | vma = find_vma_intersection(mm, addr, addr+1); |
| 536 | if (!vma) { |
| 537 | up_read(&mm->mmap_sem); |
| 538 | return -EFAULT; |
| 539 | } |
| 540 | if (vma->vm_ops && vma->vm_ops->get_policy) |
| 541 | pol = vma->vm_ops->get_policy(vma, addr); |
| 542 | else |
| 543 | pol = vma->vm_policy; |
| 544 | } else if (addr) |
| 545 | return -EINVAL; |
| 546 | |
| 547 | if (!pol) |
| 548 | pol = &default_policy; |
| 549 | |
| 550 | if (flags & MPOL_F_NODE) { |
| 551 | if (flags & MPOL_F_ADDR) { |
| 552 | err = lookup_node(mm, addr); |
| 553 | if (err < 0) |
| 554 | goto out; |
Christoph Lameter | 8bccd85 | 2005-10-29 18:16:59 -0700 | [diff] [blame] | 555 | *policy = err; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 556 | } else if (pol == current->mempolicy && |
| 557 | pol->policy == MPOL_INTERLEAVE) { |
Christoph Lameter | 8bccd85 | 2005-10-29 18:16:59 -0700 | [diff] [blame] | 558 | *policy = current->il_next; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 559 | } else { |
| 560 | err = -EINVAL; |
| 561 | goto out; |
| 562 | } |
| 563 | } else |
Christoph Lameter | 8bccd85 | 2005-10-29 18:16:59 -0700 | [diff] [blame] | 564 | *policy = pol->policy; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 565 | |
| 566 | if (vma) { |
| 567 | up_read(¤t->mm->mmap_sem); |
| 568 | vma = NULL; |
| 569 | } |
| 570 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 571 | err = 0; |
Christoph Lameter | 8bccd85 | 2005-10-29 18:16:59 -0700 | [diff] [blame] | 572 | if (nmask) |
| 573 | get_zonemask(pol, nmask); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 574 | |
| 575 | out: |
| 576 | if (vma) |
| 577 | up_read(¤t->mm->mmap_sem); |
| 578 | return err; |
| 579 | } |
| 580 | |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 581 | #ifdef CONFIG_MIGRATION |
Christoph Lameter | 8bccd85 | 2005-10-29 18:16:59 -0700 | [diff] [blame] | 582 | /* |
Christoph Lameter | 6ce3c4c | 2006-01-08 01:01:04 -0800 | [diff] [blame] | 583 | * page migration |
| 584 | */ |
Christoph Lameter | fc30128 | 2006-01-18 17:42:29 -0800 | [diff] [blame] | 585 | static void migrate_page_add(struct page *page, struct list_head *pagelist, |
| 586 | unsigned long flags) |
Christoph Lameter | 6ce3c4c | 2006-01-08 01:01:04 -0800 | [diff] [blame] | 587 | { |
| 588 | /* |
Christoph Lameter | fc30128 | 2006-01-18 17:42:29 -0800 | [diff] [blame] | 589 | * Avoid migrating a page that is shared with others. |
Christoph Lameter | 6ce3c4c | 2006-01-08 01:01:04 -0800 | [diff] [blame] | 590 | */ |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 591 | if ((flags & MPOL_MF_MOVE_ALL) || page_mapcount(page) == 1) |
| 592 | isolate_lru_page(page, pagelist); |
Christoph Lameter | 6ce3c4c | 2006-01-08 01:01:04 -0800 | [diff] [blame] | 593 | } |
| 594 | |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 595 | static struct page *new_node_page(struct page *page, unsigned long node, int **x) |
Christoph Lameter | 95a402c | 2006-06-23 02:03:53 -0700 | [diff] [blame] | 596 | { |
Mel Gorman | 769848c | 2007-07-17 04:03:05 -0700 | [diff] [blame^] | 597 | return alloc_pages_node(node, GFP_HIGHUSER_MOVABLE, 0); |
Christoph Lameter | 95a402c | 2006-06-23 02:03:53 -0700 | [diff] [blame] | 598 | } |
| 599 | |
Christoph Lameter | 6ce3c4c | 2006-01-08 01:01:04 -0800 | [diff] [blame] | 600 | /* |
Christoph Lameter | 7e2ab15 | 2006-02-01 03:05:40 -0800 | [diff] [blame] | 601 | * Migrate pages from one node to a target node. |
| 602 | * Returns error or the number of pages not migrated. |
| 603 | */ |
| 604 | int migrate_to_node(struct mm_struct *mm, int source, int dest, int flags) |
| 605 | { |
| 606 | nodemask_t nmask; |
| 607 | LIST_HEAD(pagelist); |
| 608 | int err = 0; |
| 609 | |
| 610 | nodes_clear(nmask); |
| 611 | node_set(source, nmask); |
| 612 | |
| 613 | check_range(mm, mm->mmap->vm_start, TASK_SIZE, &nmask, |
| 614 | flags | MPOL_MF_DISCONTIG_OK, &pagelist); |
| 615 | |
Christoph Lameter | aaa994b | 2006-06-23 02:03:52 -0700 | [diff] [blame] | 616 | if (!list_empty(&pagelist)) |
Christoph Lameter | 95a402c | 2006-06-23 02:03:53 -0700 | [diff] [blame] | 617 | err = migrate_pages(&pagelist, new_node_page, dest); |
| 618 | |
Christoph Lameter | 7e2ab15 | 2006-02-01 03:05:40 -0800 | [diff] [blame] | 619 | return err; |
| 620 | } |
| 621 | |
| 622 | /* |
| 623 | * Move pages between the two nodesets so as to preserve the physical |
| 624 | * layout as much as possible. |
Christoph Lameter | 3974388 | 2006-01-08 01:00:51 -0800 | [diff] [blame] | 625 | * |
| 626 | * Returns the number of page that could not be moved. |
| 627 | */ |
| 628 | int do_migrate_pages(struct mm_struct *mm, |
| 629 | const nodemask_t *from_nodes, const nodemask_t *to_nodes, int flags) |
| 630 | { |
| 631 | LIST_HEAD(pagelist); |
Christoph Lameter | 7e2ab15 | 2006-02-01 03:05:40 -0800 | [diff] [blame] | 632 | int busy = 0; |
| 633 | int err = 0; |
| 634 | nodemask_t tmp; |
Christoph Lameter | 3974388 | 2006-01-08 01:00:51 -0800 | [diff] [blame] | 635 | |
Christoph Lameter | 7e2ab15 | 2006-02-01 03:05:40 -0800 | [diff] [blame] | 636 | down_read(&mm->mmap_sem); |
Christoph Lameter | 3974388 | 2006-01-08 01:00:51 -0800 | [diff] [blame] | 637 | |
Christoph Lameter | 7b2259b | 2006-06-25 05:46:48 -0700 | [diff] [blame] | 638 | err = migrate_vmas(mm, from_nodes, to_nodes, flags); |
| 639 | if (err) |
| 640 | goto out; |
| 641 | |
Christoph Lameter | 7e2ab15 | 2006-02-01 03:05:40 -0800 | [diff] [blame] | 642 | /* |
| 643 | * Find a 'source' bit set in 'tmp' whose corresponding 'dest' |
| 644 | * bit in 'to' is not also set in 'tmp'. Clear the found 'source' |
| 645 | * bit in 'tmp', and return that <source, dest> pair for migration. |
| 646 | * The pair of nodemasks 'to' and 'from' define the map. |
| 647 | * |
| 648 | * If no pair of bits is found that way, fallback to picking some |
| 649 | * pair of 'source' and 'dest' bits that are not the same. If the |
| 650 | * 'source' and 'dest' bits are the same, this represents a node |
| 651 | * that will be migrating to itself, so no pages need move. |
| 652 | * |
| 653 | * If no bits are left in 'tmp', or if all remaining bits left |
| 654 | * in 'tmp' correspond to the same bit in 'to', return false |
| 655 | * (nothing left to migrate). |
| 656 | * |
| 657 | * This lets us pick a pair of nodes to migrate between, such that |
| 658 | * if possible the dest node is not already occupied by some other |
| 659 | * source node, minimizing the risk of overloading the memory on a |
| 660 | * node that would happen if we migrated incoming memory to a node |
| 661 | * before migrating outgoing memory source that same node. |
| 662 | * |
| 663 | * A single scan of tmp is sufficient. As we go, we remember the |
| 664 | * most recent <s, d> pair that moved (s != d). If we find a pair |
| 665 | * that not only moved, but what's better, moved to an empty slot |
| 666 | * (d is not set in tmp), then we break out then, with that pair. |
| 667 | * Otherwise when we finish scannng from_tmp, we at least have the |
| 668 | * most recent <s, d> pair that moved. If we get all the way through |
| 669 | * the scan of tmp without finding any node that moved, much less |
| 670 | * moved to an empty node, then there is nothing left worth migrating. |
| 671 | */ |
Christoph Lameter | d498471 | 2006-01-08 01:00:55 -0800 | [diff] [blame] | 672 | |
Christoph Lameter | 7e2ab15 | 2006-02-01 03:05:40 -0800 | [diff] [blame] | 673 | tmp = *from_nodes; |
| 674 | while (!nodes_empty(tmp)) { |
| 675 | int s,d; |
| 676 | int source = -1; |
| 677 | int dest = 0; |
| 678 | |
| 679 | for_each_node_mask(s, tmp) { |
| 680 | d = node_remap(s, *from_nodes, *to_nodes); |
| 681 | if (s == d) |
| 682 | continue; |
| 683 | |
| 684 | source = s; /* Node moved. Memorize */ |
| 685 | dest = d; |
| 686 | |
| 687 | /* dest not in remaining from nodes? */ |
| 688 | if (!node_isset(dest, tmp)) |
| 689 | break; |
| 690 | } |
| 691 | if (source == -1) |
| 692 | break; |
| 693 | |
| 694 | node_clear(source, tmp); |
| 695 | err = migrate_to_node(mm, source, dest, flags); |
| 696 | if (err > 0) |
| 697 | busy += err; |
| 698 | if (err < 0) |
| 699 | break; |
Christoph Lameter | 3974388 | 2006-01-08 01:00:51 -0800 | [diff] [blame] | 700 | } |
Christoph Lameter | 7b2259b | 2006-06-25 05:46:48 -0700 | [diff] [blame] | 701 | out: |
Christoph Lameter | 3974388 | 2006-01-08 01:00:51 -0800 | [diff] [blame] | 702 | up_read(&mm->mmap_sem); |
Christoph Lameter | 7e2ab15 | 2006-02-01 03:05:40 -0800 | [diff] [blame] | 703 | if (err < 0) |
| 704 | return err; |
| 705 | return busy; |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 706 | |
Christoph Lameter | 3974388 | 2006-01-08 01:00:51 -0800 | [diff] [blame] | 707 | } |
| 708 | |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 709 | static struct page *new_vma_page(struct page *page, unsigned long private, int **x) |
Christoph Lameter | 95a402c | 2006-06-23 02:03:53 -0700 | [diff] [blame] | 710 | { |
| 711 | struct vm_area_struct *vma = (struct vm_area_struct *)private; |
| 712 | |
Mel Gorman | 769848c | 2007-07-17 04:03:05 -0700 | [diff] [blame^] | 713 | return alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, |
| 714 | page_address_in_vma(page, vma)); |
Christoph Lameter | 95a402c | 2006-06-23 02:03:53 -0700 | [diff] [blame] | 715 | } |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 716 | #else |
| 717 | |
| 718 | static void migrate_page_add(struct page *page, struct list_head *pagelist, |
| 719 | unsigned long flags) |
| 720 | { |
| 721 | } |
| 722 | |
| 723 | int do_migrate_pages(struct mm_struct *mm, |
| 724 | const nodemask_t *from_nodes, const nodemask_t *to_nodes, int flags) |
| 725 | { |
| 726 | return -ENOSYS; |
| 727 | } |
Christoph Lameter | 95a402c | 2006-06-23 02:03:53 -0700 | [diff] [blame] | 728 | |
Keith Owens | 6993974 | 2006-10-11 01:21:28 -0700 | [diff] [blame] | 729 | static struct page *new_vma_page(struct page *page, unsigned long private, int **x) |
Christoph Lameter | 95a402c | 2006-06-23 02:03:53 -0700 | [diff] [blame] | 730 | { |
| 731 | return NULL; |
| 732 | } |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 733 | #endif |
| 734 | |
Christoph Lameter | 6ce3c4c | 2006-01-08 01:01:04 -0800 | [diff] [blame] | 735 | long do_mbind(unsigned long start, unsigned long len, |
| 736 | unsigned long mode, nodemask_t *nmask, unsigned long flags) |
| 737 | { |
| 738 | struct vm_area_struct *vma; |
| 739 | struct mm_struct *mm = current->mm; |
| 740 | struct mempolicy *new; |
| 741 | unsigned long end; |
| 742 | int err; |
| 743 | LIST_HEAD(pagelist); |
| 744 | |
| 745 | if ((flags & ~(unsigned long)(MPOL_MF_STRICT | |
| 746 | MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) |
| 747 | || mode > MPOL_MAX) |
| 748 | return -EINVAL; |
Christoph Lameter | 74c0024 | 2006-03-14 19:50:21 -0800 | [diff] [blame] | 749 | if ((flags & MPOL_MF_MOVE_ALL) && !capable(CAP_SYS_NICE)) |
Christoph Lameter | 6ce3c4c | 2006-01-08 01:01:04 -0800 | [diff] [blame] | 750 | return -EPERM; |
| 751 | |
| 752 | if (start & ~PAGE_MASK) |
| 753 | return -EINVAL; |
| 754 | |
| 755 | if (mode == MPOL_DEFAULT) |
| 756 | flags &= ~MPOL_MF_STRICT; |
| 757 | |
| 758 | len = (len + PAGE_SIZE - 1) & PAGE_MASK; |
| 759 | end = start + len; |
| 760 | |
| 761 | if (end < start) |
| 762 | return -EINVAL; |
| 763 | if (end == start) |
| 764 | return 0; |
| 765 | |
| 766 | if (mpol_check_policy(mode, nmask)) |
| 767 | return -EINVAL; |
| 768 | |
| 769 | new = mpol_new(mode, nmask); |
| 770 | if (IS_ERR(new)) |
| 771 | return PTR_ERR(new); |
| 772 | |
| 773 | /* |
| 774 | * If we are using the default policy then operation |
| 775 | * on discontinuous address spaces is okay after all |
| 776 | */ |
| 777 | if (!new) |
| 778 | flags |= MPOL_MF_DISCONTIG_OK; |
| 779 | |
Paul Mundt | 140d5a4 | 2007-07-15 23:38:16 -0700 | [diff] [blame] | 780 | pr_debug("mbind %lx-%lx mode:%ld nodes:%lx\n",start,start+len, |
| 781 | mode, nmask ? nodes_addr(*nmask)[0] : -1); |
Christoph Lameter | 6ce3c4c | 2006-01-08 01:01:04 -0800 | [diff] [blame] | 782 | |
| 783 | down_write(&mm->mmap_sem); |
| 784 | vma = check_range(mm, start, end, nmask, |
| 785 | flags | MPOL_MF_INVERT, &pagelist); |
| 786 | |
| 787 | err = PTR_ERR(vma); |
| 788 | if (!IS_ERR(vma)) { |
| 789 | int nr_failed = 0; |
| 790 | |
| 791 | err = mbind_range(vma, start, end, new); |
Christoph Lameter | 7e2ab15 | 2006-02-01 03:05:40 -0800 | [diff] [blame] | 792 | |
Christoph Lameter | 6ce3c4c | 2006-01-08 01:01:04 -0800 | [diff] [blame] | 793 | if (!list_empty(&pagelist)) |
Christoph Lameter | 95a402c | 2006-06-23 02:03:53 -0700 | [diff] [blame] | 794 | nr_failed = migrate_pages(&pagelist, new_vma_page, |
| 795 | (unsigned long)vma); |
Christoph Lameter | 6ce3c4c | 2006-01-08 01:01:04 -0800 | [diff] [blame] | 796 | |
| 797 | if (!err && nr_failed && (flags & MPOL_MF_STRICT)) |
| 798 | err = -EIO; |
| 799 | } |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 800 | |
Christoph Lameter | 6ce3c4c | 2006-01-08 01:01:04 -0800 | [diff] [blame] | 801 | up_write(&mm->mmap_sem); |
| 802 | mpol_free(new); |
| 803 | return err; |
| 804 | } |
| 805 | |
Christoph Lameter | 3974388 | 2006-01-08 01:00:51 -0800 | [diff] [blame] | 806 | /* |
Christoph Lameter | 8bccd85 | 2005-10-29 18:16:59 -0700 | [diff] [blame] | 807 | * User space interface with variable sized bitmaps for nodelists. |
| 808 | */ |
| 809 | |
| 810 | /* Copy a node mask from user space. */ |
Christoph Lameter | 3974388 | 2006-01-08 01:00:51 -0800 | [diff] [blame] | 811 | static int get_nodes(nodemask_t *nodes, const unsigned long __user *nmask, |
Christoph Lameter | 8bccd85 | 2005-10-29 18:16:59 -0700 | [diff] [blame] | 812 | unsigned long maxnode) |
| 813 | { |
| 814 | unsigned long k; |
| 815 | unsigned long nlongs; |
| 816 | unsigned long endmask; |
| 817 | |
| 818 | --maxnode; |
| 819 | nodes_clear(*nodes); |
| 820 | if (maxnode == 0 || !nmask) |
| 821 | return 0; |
Andi Kleen | a9c930b | 2006-02-20 18:27:59 -0800 | [diff] [blame] | 822 | if (maxnode > PAGE_SIZE*BITS_PER_BYTE) |
Chris Wright | 636f13c | 2006-02-17 13:59:36 -0800 | [diff] [blame] | 823 | return -EINVAL; |
Christoph Lameter | 8bccd85 | 2005-10-29 18:16:59 -0700 | [diff] [blame] | 824 | |
| 825 | nlongs = BITS_TO_LONGS(maxnode); |
| 826 | if ((maxnode % BITS_PER_LONG) == 0) |
| 827 | endmask = ~0UL; |
| 828 | else |
| 829 | endmask = (1UL << (maxnode % BITS_PER_LONG)) - 1; |
| 830 | |
| 831 | /* When the user specified more nodes than supported just check |
| 832 | if the non supported part is all zero. */ |
| 833 | if (nlongs > BITS_TO_LONGS(MAX_NUMNODES)) { |
| 834 | if (nlongs > PAGE_SIZE/sizeof(long)) |
| 835 | return -EINVAL; |
| 836 | for (k = BITS_TO_LONGS(MAX_NUMNODES); k < nlongs; k++) { |
| 837 | unsigned long t; |
| 838 | if (get_user(t, nmask + k)) |
| 839 | return -EFAULT; |
| 840 | if (k == nlongs - 1) { |
| 841 | if (t & endmask) |
| 842 | return -EINVAL; |
| 843 | } else if (t) |
| 844 | return -EINVAL; |
| 845 | } |
| 846 | nlongs = BITS_TO_LONGS(MAX_NUMNODES); |
| 847 | endmask = ~0UL; |
| 848 | } |
| 849 | |
| 850 | if (copy_from_user(nodes_addr(*nodes), nmask, nlongs*sizeof(unsigned long))) |
| 851 | return -EFAULT; |
| 852 | nodes_addr(*nodes)[nlongs-1] &= endmask; |
| 853 | return 0; |
| 854 | } |
| 855 | |
| 856 | /* Copy a kernel node mask to user space */ |
| 857 | static int copy_nodes_to_user(unsigned long __user *mask, unsigned long maxnode, |
| 858 | nodemask_t *nodes) |
| 859 | { |
| 860 | unsigned long copy = ALIGN(maxnode-1, 64) / 8; |
| 861 | const int nbytes = BITS_TO_LONGS(MAX_NUMNODES) * sizeof(long); |
| 862 | |
| 863 | if (copy > nbytes) { |
| 864 | if (copy > PAGE_SIZE) |
| 865 | return -EINVAL; |
| 866 | if (clear_user((char __user *)mask + nbytes, copy - nbytes)) |
| 867 | return -EFAULT; |
| 868 | copy = nbytes; |
| 869 | } |
| 870 | return copy_to_user(mask, nodes_addr(*nodes), copy) ? -EFAULT : 0; |
| 871 | } |
| 872 | |
| 873 | asmlinkage long sys_mbind(unsigned long start, unsigned long len, |
| 874 | unsigned long mode, |
| 875 | unsigned long __user *nmask, unsigned long maxnode, |
| 876 | unsigned flags) |
| 877 | { |
| 878 | nodemask_t nodes; |
| 879 | int err; |
| 880 | |
| 881 | err = get_nodes(&nodes, nmask, maxnode); |
| 882 | if (err) |
| 883 | return err; |
Christoph Lameter | 30150f8 | 2007-01-22 20:40:45 -0800 | [diff] [blame] | 884 | #ifdef CONFIG_CPUSETS |
| 885 | /* Restrict the nodes to the allowed nodes in the cpuset */ |
| 886 | nodes_and(nodes, nodes, current->mems_allowed); |
| 887 | #endif |
Christoph Lameter | 8bccd85 | 2005-10-29 18:16:59 -0700 | [diff] [blame] | 888 | return do_mbind(start, len, mode, &nodes, flags); |
| 889 | } |
| 890 | |
| 891 | /* Set the process memory policy */ |
| 892 | asmlinkage long sys_set_mempolicy(int mode, unsigned long __user *nmask, |
| 893 | unsigned long maxnode) |
| 894 | { |
| 895 | int err; |
| 896 | nodemask_t nodes; |
| 897 | |
| 898 | if (mode < 0 || mode > MPOL_MAX) |
| 899 | return -EINVAL; |
| 900 | err = get_nodes(&nodes, nmask, maxnode); |
| 901 | if (err) |
| 902 | return err; |
| 903 | return do_set_mempolicy(mode, &nodes); |
| 904 | } |
| 905 | |
Christoph Lameter | 3974388 | 2006-01-08 01:00:51 -0800 | [diff] [blame] | 906 | asmlinkage long sys_migrate_pages(pid_t pid, unsigned long maxnode, |
| 907 | const unsigned long __user *old_nodes, |
| 908 | const unsigned long __user *new_nodes) |
| 909 | { |
| 910 | struct mm_struct *mm; |
| 911 | struct task_struct *task; |
| 912 | nodemask_t old; |
| 913 | nodemask_t new; |
| 914 | nodemask_t task_nodes; |
| 915 | int err; |
| 916 | |
| 917 | err = get_nodes(&old, old_nodes, maxnode); |
| 918 | if (err) |
| 919 | return err; |
| 920 | |
| 921 | err = get_nodes(&new, new_nodes, maxnode); |
| 922 | if (err) |
| 923 | return err; |
| 924 | |
| 925 | /* Find the mm_struct */ |
| 926 | read_lock(&tasklist_lock); |
| 927 | task = pid ? find_task_by_pid(pid) : current; |
| 928 | if (!task) { |
| 929 | read_unlock(&tasklist_lock); |
| 930 | return -ESRCH; |
| 931 | } |
| 932 | mm = get_task_mm(task); |
| 933 | read_unlock(&tasklist_lock); |
| 934 | |
| 935 | if (!mm) |
| 936 | return -EINVAL; |
| 937 | |
| 938 | /* |
| 939 | * Check if this process has the right to modify the specified |
| 940 | * process. The right exists if the process has administrative |
Alexey Dobriyan | 7f927fc | 2006-03-28 01:56:53 -0800 | [diff] [blame] | 941 | * capabilities, superuser privileges or the same |
Christoph Lameter | 3974388 | 2006-01-08 01:00:51 -0800 | [diff] [blame] | 942 | * userid as the target process. |
| 943 | */ |
| 944 | if ((current->euid != task->suid) && (current->euid != task->uid) && |
| 945 | (current->uid != task->suid) && (current->uid != task->uid) && |
Christoph Lameter | 74c0024 | 2006-03-14 19:50:21 -0800 | [diff] [blame] | 946 | !capable(CAP_SYS_NICE)) { |
Christoph Lameter | 3974388 | 2006-01-08 01:00:51 -0800 | [diff] [blame] | 947 | err = -EPERM; |
| 948 | goto out; |
| 949 | } |
| 950 | |
| 951 | task_nodes = cpuset_mems_allowed(task); |
| 952 | /* Is the user allowed to access the target nodes? */ |
Christoph Lameter | 74c0024 | 2006-03-14 19:50:21 -0800 | [diff] [blame] | 953 | if (!nodes_subset(new, task_nodes) && !capable(CAP_SYS_NICE)) { |
Christoph Lameter | 3974388 | 2006-01-08 01:00:51 -0800 | [diff] [blame] | 954 | err = -EPERM; |
| 955 | goto out; |
| 956 | } |
| 957 | |
David Quigley | 86c3a76 | 2006-06-23 02:04:02 -0700 | [diff] [blame] | 958 | err = security_task_movememory(task); |
| 959 | if (err) |
| 960 | goto out; |
| 961 | |
Christoph Lameter | 511030b | 2006-02-28 16:58:57 -0800 | [diff] [blame] | 962 | err = do_migrate_pages(mm, &old, &new, |
Christoph Lameter | 74c0024 | 2006-03-14 19:50:21 -0800 | [diff] [blame] | 963 | capable(CAP_SYS_NICE) ? MPOL_MF_MOVE_ALL : MPOL_MF_MOVE); |
Christoph Lameter | 3974388 | 2006-01-08 01:00:51 -0800 | [diff] [blame] | 964 | out: |
| 965 | mmput(mm); |
| 966 | return err; |
| 967 | } |
| 968 | |
| 969 | |
Christoph Lameter | 8bccd85 | 2005-10-29 18:16:59 -0700 | [diff] [blame] | 970 | /* Retrieve NUMA policy */ |
| 971 | asmlinkage long sys_get_mempolicy(int __user *policy, |
| 972 | unsigned long __user *nmask, |
| 973 | unsigned long maxnode, |
| 974 | unsigned long addr, unsigned long flags) |
| 975 | { |
| 976 | int err, pval; |
| 977 | nodemask_t nodes; |
| 978 | |
| 979 | if (nmask != NULL && maxnode < MAX_NUMNODES) |
| 980 | return -EINVAL; |
| 981 | |
| 982 | err = do_get_mempolicy(&pval, &nodes, addr, flags); |
| 983 | |
| 984 | if (err) |
| 985 | return err; |
| 986 | |
| 987 | if (policy && put_user(pval, policy)) |
| 988 | return -EFAULT; |
| 989 | |
| 990 | if (nmask) |
| 991 | err = copy_nodes_to_user(nmask, maxnode, &nodes); |
| 992 | |
| 993 | return err; |
| 994 | } |
| 995 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 996 | #ifdef CONFIG_COMPAT |
| 997 | |
| 998 | asmlinkage long compat_sys_get_mempolicy(int __user *policy, |
| 999 | compat_ulong_t __user *nmask, |
| 1000 | compat_ulong_t maxnode, |
| 1001 | compat_ulong_t addr, compat_ulong_t flags) |
| 1002 | { |
| 1003 | long err; |
| 1004 | unsigned long __user *nm = NULL; |
| 1005 | unsigned long nr_bits, alloc_size; |
| 1006 | DECLARE_BITMAP(bm, MAX_NUMNODES); |
| 1007 | |
| 1008 | nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES); |
| 1009 | alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8; |
| 1010 | |
| 1011 | if (nmask) |
| 1012 | nm = compat_alloc_user_space(alloc_size); |
| 1013 | |
| 1014 | err = sys_get_mempolicy(policy, nm, nr_bits+1, addr, flags); |
| 1015 | |
| 1016 | if (!err && nmask) { |
| 1017 | err = copy_from_user(bm, nm, alloc_size); |
| 1018 | /* ensure entire bitmap is zeroed */ |
| 1019 | err |= clear_user(nmask, ALIGN(maxnode-1, 8) / 8); |
| 1020 | err |= compat_put_bitmap(nmask, bm, nr_bits); |
| 1021 | } |
| 1022 | |
| 1023 | return err; |
| 1024 | } |
| 1025 | |
| 1026 | asmlinkage long compat_sys_set_mempolicy(int mode, compat_ulong_t __user *nmask, |
| 1027 | compat_ulong_t maxnode) |
| 1028 | { |
| 1029 | long err = 0; |
| 1030 | unsigned long __user *nm = NULL; |
| 1031 | unsigned long nr_bits, alloc_size; |
| 1032 | DECLARE_BITMAP(bm, MAX_NUMNODES); |
| 1033 | |
| 1034 | nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES); |
| 1035 | alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8; |
| 1036 | |
| 1037 | if (nmask) { |
| 1038 | err = compat_get_bitmap(bm, nmask, nr_bits); |
| 1039 | nm = compat_alloc_user_space(alloc_size); |
| 1040 | err |= copy_to_user(nm, bm, alloc_size); |
| 1041 | } |
| 1042 | |
| 1043 | if (err) |
| 1044 | return -EFAULT; |
| 1045 | |
| 1046 | return sys_set_mempolicy(mode, nm, nr_bits+1); |
| 1047 | } |
| 1048 | |
| 1049 | asmlinkage long compat_sys_mbind(compat_ulong_t start, compat_ulong_t len, |
| 1050 | compat_ulong_t mode, compat_ulong_t __user *nmask, |
| 1051 | compat_ulong_t maxnode, compat_ulong_t flags) |
| 1052 | { |
| 1053 | long err = 0; |
| 1054 | unsigned long __user *nm = NULL; |
| 1055 | unsigned long nr_bits, alloc_size; |
Andi Kleen | dfcd3c0 | 2005-10-29 18:15:48 -0700 | [diff] [blame] | 1056 | nodemask_t bm; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1057 | |
| 1058 | nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES); |
| 1059 | alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8; |
| 1060 | |
| 1061 | if (nmask) { |
Andi Kleen | dfcd3c0 | 2005-10-29 18:15:48 -0700 | [diff] [blame] | 1062 | err = compat_get_bitmap(nodes_addr(bm), nmask, nr_bits); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1063 | nm = compat_alloc_user_space(alloc_size); |
Andi Kleen | dfcd3c0 | 2005-10-29 18:15:48 -0700 | [diff] [blame] | 1064 | err |= copy_to_user(nm, nodes_addr(bm), alloc_size); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1065 | } |
| 1066 | |
| 1067 | if (err) |
| 1068 | return -EFAULT; |
| 1069 | |
| 1070 | return sys_mbind(start, len, mode, nm, nr_bits+1, flags); |
| 1071 | } |
| 1072 | |
| 1073 | #endif |
| 1074 | |
| 1075 | /* Return effective policy for a VMA */ |
Christoph Lameter | 48fce34 | 2006-01-08 01:01:03 -0800 | [diff] [blame] | 1076 | static struct mempolicy * get_vma_policy(struct task_struct *task, |
| 1077 | struct vm_area_struct *vma, unsigned long addr) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1078 | { |
Christoph Lameter | 6e21c8f | 2005-09-03 15:54:45 -0700 | [diff] [blame] | 1079 | struct mempolicy *pol = task->mempolicy; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1080 | |
| 1081 | if (vma) { |
| 1082 | if (vma->vm_ops && vma->vm_ops->get_policy) |
Christoph Lameter | 8bccd85 | 2005-10-29 18:16:59 -0700 | [diff] [blame] | 1083 | pol = vma->vm_ops->get_policy(vma, addr); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1084 | else if (vma->vm_policy && |
| 1085 | vma->vm_policy->policy != MPOL_DEFAULT) |
| 1086 | pol = vma->vm_policy; |
| 1087 | } |
| 1088 | if (!pol) |
| 1089 | pol = &default_policy; |
| 1090 | return pol; |
| 1091 | } |
| 1092 | |
| 1093 | /* Return a zonelist representing a mempolicy */ |
Al Viro | dd0fc66 | 2005-10-07 07:46:04 +0100 | [diff] [blame] | 1094 | static struct zonelist *zonelist_policy(gfp_t gfp, struct mempolicy *policy) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1095 | { |
| 1096 | int nd; |
| 1097 | |
| 1098 | switch (policy->policy) { |
| 1099 | case MPOL_PREFERRED: |
| 1100 | nd = policy->v.preferred_node; |
| 1101 | if (nd < 0) |
| 1102 | nd = numa_node_id(); |
| 1103 | break; |
| 1104 | case MPOL_BIND: |
| 1105 | /* Lower zones don't get a policy applied */ |
| 1106 | /* Careful: current->mems_allowed might have moved */ |
Christoph Lameter | 19655d3 | 2006-09-25 23:31:19 -0700 | [diff] [blame] | 1107 | if (gfp_zone(gfp) >= policy_zone) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1108 | if (cpuset_zonelist_valid_mems_allowed(policy->v.zonelist)) |
| 1109 | return policy->v.zonelist; |
| 1110 | /*FALL THROUGH*/ |
| 1111 | case MPOL_INTERLEAVE: /* should not happen */ |
| 1112 | case MPOL_DEFAULT: |
| 1113 | nd = numa_node_id(); |
| 1114 | break; |
| 1115 | default: |
| 1116 | nd = 0; |
| 1117 | BUG(); |
| 1118 | } |
Al Viro | af4ca45 | 2005-10-21 02:55:38 -0400 | [diff] [blame] | 1119 | return NODE_DATA(nd)->node_zonelists + gfp_zone(gfp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1120 | } |
| 1121 | |
| 1122 | /* Do dynamic interleaving for a process */ |
| 1123 | static unsigned interleave_nodes(struct mempolicy *policy) |
| 1124 | { |
| 1125 | unsigned nid, next; |
| 1126 | struct task_struct *me = current; |
| 1127 | |
| 1128 | nid = me->il_next; |
Andi Kleen | dfcd3c0 | 2005-10-29 18:15:48 -0700 | [diff] [blame] | 1129 | next = next_node(nid, policy->v.nodes); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1130 | if (next >= MAX_NUMNODES) |
Andi Kleen | dfcd3c0 | 2005-10-29 18:15:48 -0700 | [diff] [blame] | 1131 | next = first_node(policy->v.nodes); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1132 | me->il_next = next; |
| 1133 | return nid; |
| 1134 | } |
| 1135 | |
Christoph Lameter | dc85da1 | 2006-01-18 17:42:36 -0800 | [diff] [blame] | 1136 | /* |
| 1137 | * Depending on the memory policy provide a node from which to allocate the |
| 1138 | * next slab entry. |
| 1139 | */ |
| 1140 | unsigned slab_node(struct mempolicy *policy) |
| 1141 | { |
Christoph Lameter | 765c450 | 2006-09-27 01:50:08 -0700 | [diff] [blame] | 1142 | int pol = policy ? policy->policy : MPOL_DEFAULT; |
| 1143 | |
| 1144 | switch (pol) { |
Christoph Lameter | dc85da1 | 2006-01-18 17:42:36 -0800 | [diff] [blame] | 1145 | case MPOL_INTERLEAVE: |
| 1146 | return interleave_nodes(policy); |
| 1147 | |
| 1148 | case MPOL_BIND: |
| 1149 | /* |
| 1150 | * Follow bind policy behavior and start allocation at the |
| 1151 | * first node. |
| 1152 | */ |
Christoph Lameter | 89fa302 | 2006-09-25 23:31:55 -0700 | [diff] [blame] | 1153 | return zone_to_nid(policy->v.zonelist->zones[0]); |
Christoph Lameter | dc85da1 | 2006-01-18 17:42:36 -0800 | [diff] [blame] | 1154 | |
| 1155 | case MPOL_PREFERRED: |
| 1156 | if (policy->v.preferred_node >= 0) |
| 1157 | return policy->v.preferred_node; |
| 1158 | /* Fall through */ |
| 1159 | |
| 1160 | default: |
| 1161 | return numa_node_id(); |
| 1162 | } |
| 1163 | } |
| 1164 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1165 | /* Do static interleaving for a VMA with known offset. */ |
| 1166 | static unsigned offset_il_node(struct mempolicy *pol, |
| 1167 | struct vm_area_struct *vma, unsigned long off) |
| 1168 | { |
Andi Kleen | dfcd3c0 | 2005-10-29 18:15:48 -0700 | [diff] [blame] | 1169 | unsigned nnodes = nodes_weight(pol->v.nodes); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1170 | unsigned target = (unsigned)off % nnodes; |
| 1171 | int c; |
| 1172 | int nid = -1; |
| 1173 | |
| 1174 | c = 0; |
| 1175 | do { |
Andi Kleen | dfcd3c0 | 2005-10-29 18:15:48 -0700 | [diff] [blame] | 1176 | nid = next_node(nid, pol->v.nodes); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1177 | c++; |
| 1178 | } while (c <= target); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1179 | return nid; |
| 1180 | } |
| 1181 | |
Christoph Lameter | 5da7ca8 | 2006-01-06 00:10:46 -0800 | [diff] [blame] | 1182 | /* Determine a node number for interleave */ |
| 1183 | static inline unsigned interleave_nid(struct mempolicy *pol, |
| 1184 | struct vm_area_struct *vma, unsigned long addr, int shift) |
| 1185 | { |
| 1186 | if (vma) { |
| 1187 | unsigned long off; |
| 1188 | |
Nishanth Aravamudan | 3b98b08 | 2006-08-31 21:27:53 -0700 | [diff] [blame] | 1189 | /* |
| 1190 | * for small pages, there is no difference between |
| 1191 | * shift and PAGE_SHIFT, so the bit-shift is safe. |
| 1192 | * for huge pages, since vm_pgoff is in units of small |
| 1193 | * pages, we need to shift off the always 0 bits to get |
| 1194 | * a useful offset. |
| 1195 | */ |
| 1196 | BUG_ON(shift < PAGE_SHIFT); |
| 1197 | off = vma->vm_pgoff >> (shift - PAGE_SHIFT); |
Christoph Lameter | 5da7ca8 | 2006-01-06 00:10:46 -0800 | [diff] [blame] | 1198 | off += (addr - vma->vm_start) >> shift; |
| 1199 | return offset_il_node(pol, vma, off); |
| 1200 | } else |
| 1201 | return interleave_nodes(pol); |
| 1202 | } |
| 1203 | |
Chen, Kenneth W | 00ac59a | 2006-02-03 21:51:14 +0100 | [diff] [blame] | 1204 | #ifdef CONFIG_HUGETLBFS |
Christoph Lameter | 5da7ca8 | 2006-01-06 00:10:46 -0800 | [diff] [blame] | 1205 | /* Return a zonelist suitable for a huge page allocation. */ |
| 1206 | struct zonelist *huge_zonelist(struct vm_area_struct *vma, unsigned long addr) |
| 1207 | { |
| 1208 | struct mempolicy *pol = get_vma_policy(current, vma, addr); |
| 1209 | |
| 1210 | if (pol->policy == MPOL_INTERLEAVE) { |
| 1211 | unsigned nid; |
| 1212 | |
| 1213 | nid = interleave_nid(pol, vma, addr, HPAGE_SHIFT); |
| 1214 | return NODE_DATA(nid)->node_zonelists + gfp_zone(GFP_HIGHUSER); |
| 1215 | } |
| 1216 | return zonelist_policy(GFP_HIGHUSER, pol); |
| 1217 | } |
Chen, Kenneth W | 00ac59a | 2006-02-03 21:51:14 +0100 | [diff] [blame] | 1218 | #endif |
Christoph Lameter | 5da7ca8 | 2006-01-06 00:10:46 -0800 | [diff] [blame] | 1219 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1220 | /* Allocate a page in interleaved policy. |
| 1221 | Own path because it needs to do special accounting. */ |
Andi Kleen | 662f3a0 | 2005-10-29 18:15:49 -0700 | [diff] [blame] | 1222 | static struct page *alloc_page_interleave(gfp_t gfp, unsigned order, |
| 1223 | unsigned nid) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1224 | { |
| 1225 | struct zonelist *zl; |
| 1226 | struct page *page; |
| 1227 | |
Al Viro | af4ca45 | 2005-10-21 02:55:38 -0400 | [diff] [blame] | 1228 | zl = NODE_DATA(nid)->node_zonelists + gfp_zone(gfp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1229 | page = __alloc_pages(gfp, order, zl); |
Christoph Lameter | ca889e6 | 2006-06-30 01:55:44 -0700 | [diff] [blame] | 1230 | if (page && page_zone(page) == zl->zones[0]) |
| 1231 | inc_zone_page_state(page, NUMA_INTERLEAVE_HIT); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1232 | return page; |
| 1233 | } |
| 1234 | |
| 1235 | /** |
| 1236 | * alloc_page_vma - Allocate a page for a VMA. |
| 1237 | * |
| 1238 | * @gfp: |
| 1239 | * %GFP_USER user allocation. |
| 1240 | * %GFP_KERNEL kernel allocations, |
| 1241 | * %GFP_HIGHMEM highmem/user allocations, |
| 1242 | * %GFP_FS allocation should not call back into a file system. |
| 1243 | * %GFP_ATOMIC don't sleep. |
| 1244 | * |
| 1245 | * @vma: Pointer to VMA or NULL if not available. |
| 1246 | * @addr: Virtual Address of the allocation. Must be inside the VMA. |
| 1247 | * |
| 1248 | * This function allocates a page from the kernel page pool and applies |
| 1249 | * a NUMA policy associated with the VMA or the current process. |
| 1250 | * When VMA is not NULL caller must hold down_read on the mmap_sem of the |
| 1251 | * mm_struct of the VMA to prevent it from going away. Should be used for |
| 1252 | * all allocations for pages that will be mapped into |
| 1253 | * user space. Returns NULL when no page can be allocated. |
| 1254 | * |
| 1255 | * Should be called with the mm_sem of the vma hold. |
| 1256 | */ |
| 1257 | struct page * |
Al Viro | dd0fc66 | 2005-10-07 07:46:04 +0100 | [diff] [blame] | 1258 | alloc_page_vma(gfp_t gfp, struct vm_area_struct *vma, unsigned long addr) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1259 | { |
Christoph Lameter | 6e21c8f | 2005-09-03 15:54:45 -0700 | [diff] [blame] | 1260 | struct mempolicy *pol = get_vma_policy(current, vma, addr); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1261 | |
Paul Jackson | cf2a473 | 2006-01-08 01:01:54 -0800 | [diff] [blame] | 1262 | cpuset_update_task_memory_state(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1263 | |
| 1264 | if (unlikely(pol->policy == MPOL_INTERLEAVE)) { |
| 1265 | unsigned nid; |
Christoph Lameter | 5da7ca8 | 2006-01-06 00:10:46 -0800 | [diff] [blame] | 1266 | |
| 1267 | nid = interleave_nid(pol, vma, addr, PAGE_SHIFT); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1268 | return alloc_page_interleave(gfp, 0, nid); |
| 1269 | } |
| 1270 | return __alloc_pages(gfp, 0, zonelist_policy(gfp, pol)); |
| 1271 | } |
| 1272 | |
| 1273 | /** |
| 1274 | * alloc_pages_current - Allocate pages. |
| 1275 | * |
| 1276 | * @gfp: |
| 1277 | * %GFP_USER user allocation, |
| 1278 | * %GFP_KERNEL kernel allocation, |
| 1279 | * %GFP_HIGHMEM highmem allocation, |
| 1280 | * %GFP_FS don't call back into a file system. |
| 1281 | * %GFP_ATOMIC don't sleep. |
| 1282 | * @order: Power of two of allocation size in pages. 0 is a single page. |
| 1283 | * |
| 1284 | * Allocate a page from the kernel page pool. When not in |
| 1285 | * interrupt context and apply the current process NUMA policy. |
| 1286 | * Returns NULL when no page can be allocated. |
| 1287 | * |
Paul Jackson | cf2a473 | 2006-01-08 01:01:54 -0800 | [diff] [blame] | 1288 | * Don't call cpuset_update_task_memory_state() unless |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1289 | * 1) it's ok to take cpuset_sem (can WAIT), and |
| 1290 | * 2) allocating for current task (not interrupt). |
| 1291 | */ |
Al Viro | dd0fc66 | 2005-10-07 07:46:04 +0100 | [diff] [blame] | 1292 | struct page *alloc_pages_current(gfp_t gfp, unsigned order) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1293 | { |
| 1294 | struct mempolicy *pol = current->mempolicy; |
| 1295 | |
| 1296 | if ((gfp & __GFP_WAIT) && !in_interrupt()) |
Paul Jackson | cf2a473 | 2006-01-08 01:01:54 -0800 | [diff] [blame] | 1297 | cpuset_update_task_memory_state(); |
Christoph Lameter | 9b819d2 | 2006-09-25 23:31:40 -0700 | [diff] [blame] | 1298 | if (!pol || in_interrupt() || (gfp & __GFP_THISNODE)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1299 | pol = &default_policy; |
| 1300 | if (pol->policy == MPOL_INTERLEAVE) |
| 1301 | return alloc_page_interleave(gfp, order, interleave_nodes(pol)); |
| 1302 | return __alloc_pages(gfp, order, zonelist_policy(gfp, pol)); |
| 1303 | } |
| 1304 | EXPORT_SYMBOL(alloc_pages_current); |
| 1305 | |
Paul Jackson | 4225399 | 2006-01-08 01:01:59 -0800 | [diff] [blame] | 1306 | /* |
| 1307 | * If mpol_copy() sees current->cpuset == cpuset_being_rebound, then it |
| 1308 | * rebinds the mempolicy its copying by calling mpol_rebind_policy() |
| 1309 | * with the mems_allowed returned by cpuset_mems_allowed(). This |
| 1310 | * keeps mempolicies cpuset relative after its cpuset moves. See |
| 1311 | * further kernel/cpuset.c update_nodemask(). |
| 1312 | */ |
| 1313 | void *cpuset_being_rebound; |
| 1314 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1315 | /* Slow path of a mempolicy copy */ |
| 1316 | struct mempolicy *__mpol_copy(struct mempolicy *old) |
| 1317 | { |
| 1318 | struct mempolicy *new = kmem_cache_alloc(policy_cache, GFP_KERNEL); |
| 1319 | |
| 1320 | if (!new) |
| 1321 | return ERR_PTR(-ENOMEM); |
Paul Jackson | 4225399 | 2006-01-08 01:01:59 -0800 | [diff] [blame] | 1322 | if (current_cpuset_is_being_rebound()) { |
| 1323 | nodemask_t mems = cpuset_mems_allowed(current); |
| 1324 | mpol_rebind_policy(old, &mems); |
| 1325 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1326 | *new = *old; |
| 1327 | atomic_set(&new->refcnt, 1); |
| 1328 | if (new->policy == MPOL_BIND) { |
| 1329 | int sz = ksize(old->v.zonelist); |
Christoph Lameter | e94b176 | 2006-12-06 20:33:17 -0800 | [diff] [blame] | 1330 | new->v.zonelist = kmemdup(old->v.zonelist, sz, GFP_KERNEL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1331 | if (!new->v.zonelist) { |
| 1332 | kmem_cache_free(policy_cache, new); |
| 1333 | return ERR_PTR(-ENOMEM); |
| 1334 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1335 | } |
| 1336 | return new; |
| 1337 | } |
| 1338 | |
| 1339 | /* Slow path of a mempolicy comparison */ |
| 1340 | int __mpol_equal(struct mempolicy *a, struct mempolicy *b) |
| 1341 | { |
| 1342 | if (!a || !b) |
| 1343 | return 0; |
| 1344 | if (a->policy != b->policy) |
| 1345 | return 0; |
| 1346 | switch (a->policy) { |
| 1347 | case MPOL_DEFAULT: |
| 1348 | return 1; |
| 1349 | case MPOL_INTERLEAVE: |
Andi Kleen | dfcd3c0 | 2005-10-29 18:15:48 -0700 | [diff] [blame] | 1350 | return nodes_equal(a->v.nodes, b->v.nodes); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1351 | case MPOL_PREFERRED: |
| 1352 | return a->v.preferred_node == b->v.preferred_node; |
| 1353 | case MPOL_BIND: { |
| 1354 | int i; |
| 1355 | for (i = 0; a->v.zonelist->zones[i]; i++) |
| 1356 | if (a->v.zonelist->zones[i] != b->v.zonelist->zones[i]) |
| 1357 | return 0; |
| 1358 | return b->v.zonelist->zones[i] == NULL; |
| 1359 | } |
| 1360 | default: |
| 1361 | BUG(); |
| 1362 | return 0; |
| 1363 | } |
| 1364 | } |
| 1365 | |
| 1366 | /* Slow path of a mpol destructor. */ |
| 1367 | void __mpol_free(struct mempolicy *p) |
| 1368 | { |
| 1369 | if (!atomic_dec_and_test(&p->refcnt)) |
| 1370 | return; |
| 1371 | if (p->policy == MPOL_BIND) |
| 1372 | kfree(p->v.zonelist); |
| 1373 | p->policy = MPOL_DEFAULT; |
| 1374 | kmem_cache_free(policy_cache, p); |
| 1375 | } |
| 1376 | |
| 1377 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1378 | * Shared memory backing store policy support. |
| 1379 | * |
| 1380 | * Remember policies even when nobody has shared memory mapped. |
| 1381 | * The policies are kept in Red-Black tree linked from the inode. |
| 1382 | * They are protected by the sp->lock spinlock, which should be held |
| 1383 | * for any accesses to the tree. |
| 1384 | */ |
| 1385 | |
| 1386 | /* lookup first element intersecting start-end */ |
| 1387 | /* Caller holds sp->lock */ |
| 1388 | static struct sp_node * |
| 1389 | sp_lookup(struct shared_policy *sp, unsigned long start, unsigned long end) |
| 1390 | { |
| 1391 | struct rb_node *n = sp->root.rb_node; |
| 1392 | |
| 1393 | while (n) { |
| 1394 | struct sp_node *p = rb_entry(n, struct sp_node, nd); |
| 1395 | |
| 1396 | if (start >= p->end) |
| 1397 | n = n->rb_right; |
| 1398 | else if (end <= p->start) |
| 1399 | n = n->rb_left; |
| 1400 | else |
| 1401 | break; |
| 1402 | } |
| 1403 | if (!n) |
| 1404 | return NULL; |
| 1405 | for (;;) { |
| 1406 | struct sp_node *w = NULL; |
| 1407 | struct rb_node *prev = rb_prev(n); |
| 1408 | if (!prev) |
| 1409 | break; |
| 1410 | w = rb_entry(prev, struct sp_node, nd); |
| 1411 | if (w->end <= start) |
| 1412 | break; |
| 1413 | n = prev; |
| 1414 | } |
| 1415 | return rb_entry(n, struct sp_node, nd); |
| 1416 | } |
| 1417 | |
| 1418 | /* Insert a new shared policy into the list. */ |
| 1419 | /* Caller holds sp->lock */ |
| 1420 | static void sp_insert(struct shared_policy *sp, struct sp_node *new) |
| 1421 | { |
| 1422 | struct rb_node **p = &sp->root.rb_node; |
| 1423 | struct rb_node *parent = NULL; |
| 1424 | struct sp_node *nd; |
| 1425 | |
| 1426 | while (*p) { |
| 1427 | parent = *p; |
| 1428 | nd = rb_entry(parent, struct sp_node, nd); |
| 1429 | if (new->start < nd->start) |
| 1430 | p = &(*p)->rb_left; |
| 1431 | else if (new->end > nd->end) |
| 1432 | p = &(*p)->rb_right; |
| 1433 | else |
| 1434 | BUG(); |
| 1435 | } |
| 1436 | rb_link_node(&new->nd, parent, p); |
| 1437 | rb_insert_color(&new->nd, &sp->root); |
Paul Mundt | 140d5a4 | 2007-07-15 23:38:16 -0700 | [diff] [blame] | 1438 | pr_debug("inserting %lx-%lx: %d\n", new->start, new->end, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1439 | new->policy ? new->policy->policy : 0); |
| 1440 | } |
| 1441 | |
| 1442 | /* Find shared policy intersecting idx */ |
| 1443 | struct mempolicy * |
| 1444 | mpol_shared_policy_lookup(struct shared_policy *sp, unsigned long idx) |
| 1445 | { |
| 1446 | struct mempolicy *pol = NULL; |
| 1447 | struct sp_node *sn; |
| 1448 | |
| 1449 | if (!sp->root.rb_node) |
| 1450 | return NULL; |
| 1451 | spin_lock(&sp->lock); |
| 1452 | sn = sp_lookup(sp, idx, idx+1); |
| 1453 | if (sn) { |
| 1454 | mpol_get(sn->policy); |
| 1455 | pol = sn->policy; |
| 1456 | } |
| 1457 | spin_unlock(&sp->lock); |
| 1458 | return pol; |
| 1459 | } |
| 1460 | |
| 1461 | static void sp_delete(struct shared_policy *sp, struct sp_node *n) |
| 1462 | { |
Paul Mundt | 140d5a4 | 2007-07-15 23:38:16 -0700 | [diff] [blame] | 1463 | pr_debug("deleting %lx-l%lx\n", n->start, n->end); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1464 | rb_erase(&n->nd, &sp->root); |
| 1465 | mpol_free(n->policy); |
| 1466 | kmem_cache_free(sn_cache, n); |
| 1467 | } |
| 1468 | |
| 1469 | struct sp_node * |
| 1470 | sp_alloc(unsigned long start, unsigned long end, struct mempolicy *pol) |
| 1471 | { |
| 1472 | struct sp_node *n = kmem_cache_alloc(sn_cache, GFP_KERNEL); |
| 1473 | |
| 1474 | if (!n) |
| 1475 | return NULL; |
| 1476 | n->start = start; |
| 1477 | n->end = end; |
| 1478 | mpol_get(pol); |
| 1479 | n->policy = pol; |
| 1480 | return n; |
| 1481 | } |
| 1482 | |
| 1483 | /* Replace a policy range. */ |
| 1484 | static int shared_policy_replace(struct shared_policy *sp, unsigned long start, |
| 1485 | unsigned long end, struct sp_node *new) |
| 1486 | { |
| 1487 | struct sp_node *n, *new2 = NULL; |
| 1488 | |
| 1489 | restart: |
| 1490 | spin_lock(&sp->lock); |
| 1491 | n = sp_lookup(sp, start, end); |
| 1492 | /* Take care of old policies in the same range. */ |
| 1493 | while (n && n->start < end) { |
| 1494 | struct rb_node *next = rb_next(&n->nd); |
| 1495 | if (n->start >= start) { |
| 1496 | if (n->end <= end) |
| 1497 | sp_delete(sp, n); |
| 1498 | else |
| 1499 | n->start = end; |
| 1500 | } else { |
| 1501 | /* Old policy spanning whole new range. */ |
| 1502 | if (n->end > end) { |
| 1503 | if (!new2) { |
| 1504 | spin_unlock(&sp->lock); |
| 1505 | new2 = sp_alloc(end, n->end, n->policy); |
| 1506 | if (!new2) |
| 1507 | return -ENOMEM; |
| 1508 | goto restart; |
| 1509 | } |
| 1510 | n->end = start; |
| 1511 | sp_insert(sp, new2); |
| 1512 | new2 = NULL; |
| 1513 | break; |
| 1514 | } else |
| 1515 | n->end = start; |
| 1516 | } |
| 1517 | if (!next) |
| 1518 | break; |
| 1519 | n = rb_entry(next, struct sp_node, nd); |
| 1520 | } |
| 1521 | if (new) |
| 1522 | sp_insert(sp, new); |
| 1523 | spin_unlock(&sp->lock); |
| 1524 | if (new2) { |
| 1525 | mpol_free(new2->policy); |
| 1526 | kmem_cache_free(sn_cache, new2); |
| 1527 | } |
| 1528 | return 0; |
| 1529 | } |
| 1530 | |
Robin Holt | 7339ff8 | 2006-01-14 13:20:48 -0800 | [diff] [blame] | 1531 | void mpol_shared_policy_init(struct shared_policy *info, int policy, |
| 1532 | nodemask_t *policy_nodes) |
| 1533 | { |
| 1534 | info->root = RB_ROOT; |
| 1535 | spin_lock_init(&info->lock); |
| 1536 | |
| 1537 | if (policy != MPOL_DEFAULT) { |
| 1538 | struct mempolicy *newpol; |
| 1539 | |
| 1540 | /* Falls back to MPOL_DEFAULT on any error */ |
| 1541 | newpol = mpol_new(policy, policy_nodes); |
| 1542 | if (!IS_ERR(newpol)) { |
| 1543 | /* Create pseudo-vma that contains just the policy */ |
| 1544 | struct vm_area_struct pvma; |
| 1545 | |
| 1546 | memset(&pvma, 0, sizeof(struct vm_area_struct)); |
| 1547 | /* Policy covers entire file */ |
| 1548 | pvma.vm_end = TASK_SIZE; |
| 1549 | mpol_set_shared_policy(info, &pvma, newpol); |
| 1550 | mpol_free(newpol); |
| 1551 | } |
| 1552 | } |
| 1553 | } |
| 1554 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1555 | int mpol_set_shared_policy(struct shared_policy *info, |
| 1556 | struct vm_area_struct *vma, struct mempolicy *npol) |
| 1557 | { |
| 1558 | int err; |
| 1559 | struct sp_node *new = NULL; |
| 1560 | unsigned long sz = vma_pages(vma); |
| 1561 | |
Paul Mundt | 140d5a4 | 2007-07-15 23:38:16 -0700 | [diff] [blame] | 1562 | pr_debug("set_shared_policy %lx sz %lu %d %lx\n", |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1563 | vma->vm_pgoff, |
| 1564 | sz, npol? npol->policy : -1, |
Paul Mundt | 140d5a4 | 2007-07-15 23:38:16 -0700 | [diff] [blame] | 1565 | npol ? nodes_addr(npol->v.nodes)[0] : -1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1566 | |
| 1567 | if (npol) { |
| 1568 | new = sp_alloc(vma->vm_pgoff, vma->vm_pgoff + sz, npol); |
| 1569 | if (!new) |
| 1570 | return -ENOMEM; |
| 1571 | } |
| 1572 | err = shared_policy_replace(info, vma->vm_pgoff, vma->vm_pgoff+sz, new); |
| 1573 | if (err && new) |
| 1574 | kmem_cache_free(sn_cache, new); |
| 1575 | return err; |
| 1576 | } |
| 1577 | |
| 1578 | /* Free a backing policy store on inode delete. */ |
| 1579 | void mpol_free_shared_policy(struct shared_policy *p) |
| 1580 | { |
| 1581 | struct sp_node *n; |
| 1582 | struct rb_node *next; |
| 1583 | |
| 1584 | if (!p->root.rb_node) |
| 1585 | return; |
| 1586 | spin_lock(&p->lock); |
| 1587 | next = rb_first(&p->root); |
| 1588 | while (next) { |
| 1589 | n = rb_entry(next, struct sp_node, nd); |
| 1590 | next = rb_next(&n->nd); |
Andi Kleen | 90c5029 | 2005-07-27 11:43:50 -0700 | [diff] [blame] | 1591 | rb_erase(&n->nd, &p->root); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1592 | mpol_free(n->policy); |
| 1593 | kmem_cache_free(sn_cache, n); |
| 1594 | } |
| 1595 | spin_unlock(&p->lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1596 | } |
| 1597 | |
| 1598 | /* assumes fs == KERNEL_DS */ |
| 1599 | void __init numa_policy_init(void) |
| 1600 | { |
Paul Mundt | b71636e | 2007-07-15 23:38:15 -0700 | [diff] [blame] | 1601 | nodemask_t interleave_nodes; |
| 1602 | unsigned long largest = 0; |
| 1603 | int nid, prefer = 0; |
| 1604 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1605 | policy_cache = kmem_cache_create("numa_policy", |
| 1606 | sizeof(struct mempolicy), |
| 1607 | 0, SLAB_PANIC, NULL, NULL); |
| 1608 | |
| 1609 | sn_cache = kmem_cache_create("shared_policy_node", |
| 1610 | sizeof(struct sp_node), |
| 1611 | 0, SLAB_PANIC, NULL, NULL); |
| 1612 | |
Paul Mundt | b71636e | 2007-07-15 23:38:15 -0700 | [diff] [blame] | 1613 | /* |
| 1614 | * Set interleaving policy for system init. Interleaving is only |
| 1615 | * enabled across suitably sized nodes (default is >= 16MB), or |
| 1616 | * fall back to the largest node if they're all smaller. |
| 1617 | */ |
| 1618 | nodes_clear(interleave_nodes); |
| 1619 | for_each_online_node(nid) { |
| 1620 | unsigned long total_pages = node_present_pages(nid); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1621 | |
Paul Mundt | b71636e | 2007-07-15 23:38:15 -0700 | [diff] [blame] | 1622 | /* Preserve the largest node */ |
| 1623 | if (largest < total_pages) { |
| 1624 | largest = total_pages; |
| 1625 | prefer = nid; |
| 1626 | } |
| 1627 | |
| 1628 | /* Interleave this node? */ |
| 1629 | if ((total_pages << PAGE_SHIFT) >= (16 << 20)) |
| 1630 | node_set(nid, interleave_nodes); |
| 1631 | } |
| 1632 | |
| 1633 | /* All too small, use the largest */ |
| 1634 | if (unlikely(nodes_empty(interleave_nodes))) |
| 1635 | node_set(prefer, interleave_nodes); |
| 1636 | |
| 1637 | if (do_set_mempolicy(MPOL_INTERLEAVE, &interleave_nodes)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1638 | printk("numa_policy_init: interleaving failed\n"); |
| 1639 | } |
| 1640 | |
Christoph Lameter | 8bccd85 | 2005-10-29 18:16:59 -0700 | [diff] [blame] | 1641 | /* Reset policy of current process to default */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1642 | void numa_default_policy(void) |
| 1643 | { |
Christoph Lameter | 8bccd85 | 2005-10-29 18:16:59 -0700 | [diff] [blame] | 1644 | do_set_mempolicy(MPOL_DEFAULT, NULL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1645 | } |
Paul Jackson | 68860ec | 2005-10-30 15:02:36 -0800 | [diff] [blame] | 1646 | |
| 1647 | /* Migrate a policy to a different set of nodes */ |
Paul Jackson | 74cb215 | 2006-01-08 01:01:56 -0800 | [diff] [blame] | 1648 | void mpol_rebind_policy(struct mempolicy *pol, const nodemask_t *newmask) |
Paul Jackson | 68860ec | 2005-10-30 15:02:36 -0800 | [diff] [blame] | 1649 | { |
Paul Jackson | 74cb215 | 2006-01-08 01:01:56 -0800 | [diff] [blame] | 1650 | nodemask_t *mpolmask; |
Paul Jackson | 68860ec | 2005-10-30 15:02:36 -0800 | [diff] [blame] | 1651 | nodemask_t tmp; |
| 1652 | |
| 1653 | if (!pol) |
| 1654 | return; |
Paul Jackson | 74cb215 | 2006-01-08 01:01:56 -0800 | [diff] [blame] | 1655 | mpolmask = &pol->cpuset_mems_allowed; |
| 1656 | if (nodes_equal(*mpolmask, *newmask)) |
| 1657 | return; |
Paul Jackson | 68860ec | 2005-10-30 15:02:36 -0800 | [diff] [blame] | 1658 | |
| 1659 | switch (pol->policy) { |
| 1660 | case MPOL_DEFAULT: |
| 1661 | break; |
| 1662 | case MPOL_INTERLEAVE: |
Paul Jackson | 74cb215 | 2006-01-08 01:01:56 -0800 | [diff] [blame] | 1663 | nodes_remap(tmp, pol->v.nodes, *mpolmask, *newmask); |
Paul Jackson | 68860ec | 2005-10-30 15:02:36 -0800 | [diff] [blame] | 1664 | pol->v.nodes = tmp; |
Paul Jackson | 74cb215 | 2006-01-08 01:01:56 -0800 | [diff] [blame] | 1665 | *mpolmask = *newmask; |
| 1666 | current->il_next = node_remap(current->il_next, |
| 1667 | *mpolmask, *newmask); |
Paul Jackson | 68860ec | 2005-10-30 15:02:36 -0800 | [diff] [blame] | 1668 | break; |
| 1669 | case MPOL_PREFERRED: |
| 1670 | pol->v.preferred_node = node_remap(pol->v.preferred_node, |
Paul Jackson | 74cb215 | 2006-01-08 01:01:56 -0800 | [diff] [blame] | 1671 | *mpolmask, *newmask); |
| 1672 | *mpolmask = *newmask; |
Paul Jackson | 68860ec | 2005-10-30 15:02:36 -0800 | [diff] [blame] | 1673 | break; |
| 1674 | case MPOL_BIND: { |
| 1675 | nodemask_t nodes; |
| 1676 | struct zone **z; |
| 1677 | struct zonelist *zonelist; |
| 1678 | |
| 1679 | nodes_clear(nodes); |
| 1680 | for (z = pol->v.zonelist->zones; *z; z++) |
Christoph Lameter | 89fa302 | 2006-09-25 23:31:55 -0700 | [diff] [blame] | 1681 | node_set(zone_to_nid(*z), nodes); |
Paul Jackson | 74cb215 | 2006-01-08 01:01:56 -0800 | [diff] [blame] | 1682 | nodes_remap(tmp, nodes, *mpolmask, *newmask); |
Paul Jackson | 68860ec | 2005-10-30 15:02:36 -0800 | [diff] [blame] | 1683 | nodes = tmp; |
| 1684 | |
| 1685 | zonelist = bind_zonelist(&nodes); |
| 1686 | |
| 1687 | /* If no mem, then zonelist is NULL and we keep old zonelist. |
| 1688 | * If that old zonelist has no remaining mems_allowed nodes, |
| 1689 | * then zonelist_policy() will "FALL THROUGH" to MPOL_DEFAULT. |
| 1690 | */ |
| 1691 | |
KAMEZAWA Hiroyuki | 8af5e2e | 2007-02-20 13:57:49 -0800 | [diff] [blame] | 1692 | if (!IS_ERR(zonelist)) { |
Paul Jackson | 68860ec | 2005-10-30 15:02:36 -0800 | [diff] [blame] | 1693 | /* Good - got mem - substitute new zonelist */ |
| 1694 | kfree(pol->v.zonelist); |
| 1695 | pol->v.zonelist = zonelist; |
| 1696 | } |
Paul Jackson | 74cb215 | 2006-01-08 01:01:56 -0800 | [diff] [blame] | 1697 | *mpolmask = *newmask; |
Paul Jackson | 68860ec | 2005-10-30 15:02:36 -0800 | [diff] [blame] | 1698 | break; |
| 1699 | } |
| 1700 | default: |
| 1701 | BUG(); |
| 1702 | break; |
| 1703 | } |
| 1704 | } |
| 1705 | |
| 1706 | /* |
Paul Jackson | 74cb215 | 2006-01-08 01:01:56 -0800 | [diff] [blame] | 1707 | * Wrapper for mpol_rebind_policy() that just requires task |
| 1708 | * pointer, and updates task mempolicy. |
Paul Jackson | 68860ec | 2005-10-30 15:02:36 -0800 | [diff] [blame] | 1709 | */ |
Paul Jackson | 74cb215 | 2006-01-08 01:01:56 -0800 | [diff] [blame] | 1710 | |
| 1711 | void mpol_rebind_task(struct task_struct *tsk, const nodemask_t *new) |
Paul Jackson | 68860ec | 2005-10-30 15:02:36 -0800 | [diff] [blame] | 1712 | { |
Paul Jackson | 74cb215 | 2006-01-08 01:01:56 -0800 | [diff] [blame] | 1713 | mpol_rebind_policy(tsk->mempolicy, new); |
Paul Jackson | 68860ec | 2005-10-30 15:02:36 -0800 | [diff] [blame] | 1714 | } |
Christoph Lameter | 1a75a6c | 2006-01-08 01:01:02 -0800 | [diff] [blame] | 1715 | |
| 1716 | /* |
Paul Jackson | 4225399 | 2006-01-08 01:01:59 -0800 | [diff] [blame] | 1717 | * Rebind each vma in mm to new nodemask. |
| 1718 | * |
| 1719 | * Call holding a reference to mm. Takes mm->mmap_sem during call. |
| 1720 | */ |
| 1721 | |
| 1722 | void mpol_rebind_mm(struct mm_struct *mm, nodemask_t *new) |
| 1723 | { |
| 1724 | struct vm_area_struct *vma; |
| 1725 | |
| 1726 | down_write(&mm->mmap_sem); |
| 1727 | for (vma = mm->mmap; vma; vma = vma->vm_next) |
| 1728 | mpol_rebind_policy(vma->vm_policy, new); |
| 1729 | up_write(&mm->mmap_sem); |
| 1730 | } |
| 1731 | |
| 1732 | /* |
Christoph Lameter | 1a75a6c | 2006-01-08 01:01:02 -0800 | [diff] [blame] | 1733 | * Display pages allocated per node and memory policy via /proc. |
| 1734 | */ |
| 1735 | |
Helge Deller | 15ad7cd | 2006-12-06 20:40:36 -0800 | [diff] [blame] | 1736 | static const char * const policy_types[] = |
| 1737 | { "default", "prefer", "bind", "interleave" }; |
Christoph Lameter | 1a75a6c | 2006-01-08 01:01:02 -0800 | [diff] [blame] | 1738 | |
| 1739 | /* |
| 1740 | * Convert a mempolicy into a string. |
| 1741 | * Returns the number of characters in buffer (if positive) |
| 1742 | * or an error (negative) |
| 1743 | */ |
| 1744 | static inline int mpol_to_str(char *buffer, int maxlen, struct mempolicy *pol) |
| 1745 | { |
| 1746 | char *p = buffer; |
| 1747 | int l; |
| 1748 | nodemask_t nodes; |
| 1749 | int mode = pol ? pol->policy : MPOL_DEFAULT; |
| 1750 | |
| 1751 | switch (mode) { |
| 1752 | case MPOL_DEFAULT: |
| 1753 | nodes_clear(nodes); |
| 1754 | break; |
| 1755 | |
| 1756 | case MPOL_PREFERRED: |
| 1757 | nodes_clear(nodes); |
| 1758 | node_set(pol->v.preferred_node, nodes); |
| 1759 | break; |
| 1760 | |
| 1761 | case MPOL_BIND: |
| 1762 | get_zonemask(pol, &nodes); |
| 1763 | break; |
| 1764 | |
| 1765 | case MPOL_INTERLEAVE: |
| 1766 | nodes = pol->v.nodes; |
| 1767 | break; |
| 1768 | |
| 1769 | default: |
| 1770 | BUG(); |
| 1771 | return -EFAULT; |
| 1772 | } |
| 1773 | |
| 1774 | l = strlen(policy_types[mode]); |
| 1775 | if (buffer + maxlen < p + l + 1) |
| 1776 | return -ENOSPC; |
| 1777 | |
| 1778 | strcpy(p, policy_types[mode]); |
| 1779 | p += l; |
| 1780 | |
| 1781 | if (!nodes_empty(nodes)) { |
| 1782 | if (buffer + maxlen < p + 2) |
| 1783 | return -ENOSPC; |
| 1784 | *p++ = '='; |
| 1785 | p += nodelist_scnprintf(p, buffer + maxlen - p, nodes); |
| 1786 | } |
| 1787 | return p - buffer; |
| 1788 | } |
| 1789 | |
| 1790 | struct numa_maps { |
| 1791 | unsigned long pages; |
| 1792 | unsigned long anon; |
Christoph Lameter | 397874d | 2006-03-06 15:42:53 -0800 | [diff] [blame] | 1793 | unsigned long active; |
| 1794 | unsigned long writeback; |
Christoph Lameter | 1a75a6c | 2006-01-08 01:01:02 -0800 | [diff] [blame] | 1795 | unsigned long mapcount_max; |
Christoph Lameter | 397874d | 2006-03-06 15:42:53 -0800 | [diff] [blame] | 1796 | unsigned long dirty; |
| 1797 | unsigned long swapcache; |
Christoph Lameter | 1a75a6c | 2006-01-08 01:01:02 -0800 | [diff] [blame] | 1798 | unsigned long node[MAX_NUMNODES]; |
| 1799 | }; |
| 1800 | |
Christoph Lameter | 397874d | 2006-03-06 15:42:53 -0800 | [diff] [blame] | 1801 | static void gather_stats(struct page *page, void *private, int pte_dirty) |
Christoph Lameter | 1a75a6c | 2006-01-08 01:01:02 -0800 | [diff] [blame] | 1802 | { |
| 1803 | struct numa_maps *md = private; |
| 1804 | int count = page_mapcount(page); |
| 1805 | |
Christoph Lameter | 1a75a6c | 2006-01-08 01:01:02 -0800 | [diff] [blame] | 1806 | md->pages++; |
Christoph Lameter | 397874d | 2006-03-06 15:42:53 -0800 | [diff] [blame] | 1807 | if (pte_dirty || PageDirty(page)) |
| 1808 | md->dirty++; |
| 1809 | |
| 1810 | if (PageSwapCache(page)) |
| 1811 | md->swapcache++; |
| 1812 | |
| 1813 | if (PageActive(page)) |
| 1814 | md->active++; |
| 1815 | |
| 1816 | if (PageWriteback(page)) |
| 1817 | md->writeback++; |
Christoph Lameter | 1a75a6c | 2006-01-08 01:01:02 -0800 | [diff] [blame] | 1818 | |
| 1819 | if (PageAnon(page)) |
| 1820 | md->anon++; |
| 1821 | |
Christoph Lameter | 397874d | 2006-03-06 15:42:53 -0800 | [diff] [blame] | 1822 | if (count > md->mapcount_max) |
| 1823 | md->mapcount_max = count; |
| 1824 | |
Christoph Lameter | 1a75a6c | 2006-01-08 01:01:02 -0800 | [diff] [blame] | 1825 | md->node[page_to_nid(page)]++; |
Christoph Lameter | 1a75a6c | 2006-01-08 01:01:02 -0800 | [diff] [blame] | 1826 | } |
| 1827 | |
Andrew Morton | 7f709ed | 2006-03-07 21:55:22 -0800 | [diff] [blame] | 1828 | #ifdef CONFIG_HUGETLB_PAGE |
Christoph Lameter | 397874d | 2006-03-06 15:42:53 -0800 | [diff] [blame] | 1829 | static void check_huge_range(struct vm_area_struct *vma, |
| 1830 | unsigned long start, unsigned long end, |
| 1831 | struct numa_maps *md) |
| 1832 | { |
| 1833 | unsigned long addr; |
| 1834 | struct page *page; |
| 1835 | |
| 1836 | for (addr = start; addr < end; addr += HPAGE_SIZE) { |
| 1837 | pte_t *ptep = huge_pte_offset(vma->vm_mm, addr & HPAGE_MASK); |
| 1838 | pte_t pte; |
| 1839 | |
| 1840 | if (!ptep) |
| 1841 | continue; |
| 1842 | |
| 1843 | pte = *ptep; |
| 1844 | if (pte_none(pte)) |
| 1845 | continue; |
| 1846 | |
| 1847 | page = pte_page(pte); |
| 1848 | if (!page) |
| 1849 | continue; |
| 1850 | |
| 1851 | gather_stats(page, md, pte_dirty(*ptep)); |
| 1852 | } |
| 1853 | } |
Andrew Morton | 7f709ed | 2006-03-07 21:55:22 -0800 | [diff] [blame] | 1854 | #else |
| 1855 | static inline void check_huge_range(struct vm_area_struct *vma, |
| 1856 | unsigned long start, unsigned long end, |
| 1857 | struct numa_maps *md) |
| 1858 | { |
| 1859 | } |
| 1860 | #endif |
Christoph Lameter | 397874d | 2006-03-06 15:42:53 -0800 | [diff] [blame] | 1861 | |
Christoph Lameter | 1a75a6c | 2006-01-08 01:01:02 -0800 | [diff] [blame] | 1862 | int show_numa_map(struct seq_file *m, void *v) |
| 1863 | { |
Eric W. Biederman | 99f8955 | 2006-06-26 00:25:55 -0700 | [diff] [blame] | 1864 | struct proc_maps_private *priv = m->private; |
Christoph Lameter | 1a75a6c | 2006-01-08 01:01:02 -0800 | [diff] [blame] | 1865 | struct vm_area_struct *vma = v; |
| 1866 | struct numa_maps *md; |
Christoph Lameter | 397874d | 2006-03-06 15:42:53 -0800 | [diff] [blame] | 1867 | struct file *file = vma->vm_file; |
| 1868 | struct mm_struct *mm = vma->vm_mm; |
Christoph Lameter | 1a75a6c | 2006-01-08 01:01:02 -0800 | [diff] [blame] | 1869 | int n; |
| 1870 | char buffer[50]; |
| 1871 | |
Christoph Lameter | 397874d | 2006-03-06 15:42:53 -0800 | [diff] [blame] | 1872 | if (!mm) |
Christoph Lameter | 1a75a6c | 2006-01-08 01:01:02 -0800 | [diff] [blame] | 1873 | return 0; |
| 1874 | |
| 1875 | md = kzalloc(sizeof(struct numa_maps), GFP_KERNEL); |
| 1876 | if (!md) |
| 1877 | return 0; |
| 1878 | |
Christoph Lameter | 397874d | 2006-03-06 15:42:53 -0800 | [diff] [blame] | 1879 | mpol_to_str(buffer, sizeof(buffer), |
Eric W. Biederman | 99f8955 | 2006-06-26 00:25:55 -0700 | [diff] [blame] | 1880 | get_vma_policy(priv->task, vma, vma->vm_start)); |
Christoph Lameter | 1a75a6c | 2006-01-08 01:01:02 -0800 | [diff] [blame] | 1881 | |
Christoph Lameter | 397874d | 2006-03-06 15:42:53 -0800 | [diff] [blame] | 1882 | seq_printf(m, "%08lx %s", vma->vm_start, buffer); |
Christoph Lameter | 1a75a6c | 2006-01-08 01:01:02 -0800 | [diff] [blame] | 1883 | |
Christoph Lameter | 397874d | 2006-03-06 15:42:53 -0800 | [diff] [blame] | 1884 | if (file) { |
| 1885 | seq_printf(m, " file="); |
Josef Sipek | e9536ae | 2006-12-08 02:37:21 -0800 | [diff] [blame] | 1886 | seq_path(m, file->f_path.mnt, file->f_path.dentry, "\n\t= "); |
Christoph Lameter | 397874d | 2006-03-06 15:42:53 -0800 | [diff] [blame] | 1887 | } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) { |
| 1888 | seq_printf(m, " heap"); |
| 1889 | } else if (vma->vm_start <= mm->start_stack && |
| 1890 | vma->vm_end >= mm->start_stack) { |
| 1891 | seq_printf(m, " stack"); |
Christoph Lameter | 1a75a6c | 2006-01-08 01:01:02 -0800 | [diff] [blame] | 1892 | } |
Christoph Lameter | 397874d | 2006-03-06 15:42:53 -0800 | [diff] [blame] | 1893 | |
| 1894 | if (is_vm_hugetlb_page(vma)) { |
| 1895 | check_huge_range(vma, vma->vm_start, vma->vm_end, md); |
| 1896 | seq_printf(m, " huge"); |
| 1897 | } else { |
| 1898 | check_pgd_range(vma, vma->vm_start, vma->vm_end, |
| 1899 | &node_online_map, MPOL_MF_STATS, md); |
| 1900 | } |
| 1901 | |
| 1902 | if (!md->pages) |
| 1903 | goto out; |
| 1904 | |
| 1905 | if (md->anon) |
| 1906 | seq_printf(m," anon=%lu",md->anon); |
| 1907 | |
| 1908 | if (md->dirty) |
| 1909 | seq_printf(m," dirty=%lu",md->dirty); |
| 1910 | |
| 1911 | if (md->pages != md->anon && md->pages != md->dirty) |
| 1912 | seq_printf(m, " mapped=%lu", md->pages); |
| 1913 | |
| 1914 | if (md->mapcount_max > 1) |
| 1915 | seq_printf(m, " mapmax=%lu", md->mapcount_max); |
| 1916 | |
| 1917 | if (md->swapcache) |
| 1918 | seq_printf(m," swapcache=%lu", md->swapcache); |
| 1919 | |
| 1920 | if (md->active < md->pages && !is_vm_hugetlb_page(vma)) |
| 1921 | seq_printf(m," active=%lu", md->active); |
| 1922 | |
| 1923 | if (md->writeback) |
| 1924 | seq_printf(m," writeback=%lu", md->writeback); |
| 1925 | |
| 1926 | for_each_online_node(n) |
| 1927 | if (md->node[n]) |
| 1928 | seq_printf(m, " N%d=%lu", n, md->node[n]); |
| 1929 | out: |
| 1930 | seq_putc(m, '\n'); |
Christoph Lameter | 1a75a6c | 2006-01-08 01:01:02 -0800 | [diff] [blame] | 1931 | kfree(md); |
| 1932 | |
| 1933 | if (m->count < m->size) |
Eric W. Biederman | 99f8955 | 2006-06-26 00:25:55 -0700 | [diff] [blame] | 1934 | m->version = (vma != priv->tail_vma) ? vma->vm_start : 0; |
Christoph Lameter | 1a75a6c | 2006-01-08 01:01:02 -0800 | [diff] [blame] | 1935 | return 0; |
| 1936 | } |
| 1937 | |