The Design and Implementation of the FreeBSD Operating System, Second Edition
Now available: The Design and Implementation of the FreeBSD Operating System (Second Edition)


[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]

FreeBSD/Linux Kernel Cross Reference
sys/mm/mempolicy.c

Version: -  FREEBSD  -  FREEBSD-13-STABLE  -  FREEBSD-13-0  -  FREEBSD-12-STABLE  -  FREEBSD-12-0  -  FREEBSD-11-STABLE  -  FREEBSD-11-0  -  FREEBSD-10-STABLE  -  FREEBSD-10-0  -  FREEBSD-9-STABLE  -  FREEBSD-9-0  -  FREEBSD-8-STABLE  -  FREEBSD-8-0  -  FREEBSD-7-STABLE  -  FREEBSD-7-0  -  FREEBSD-6-STABLE  -  FREEBSD-6-0  -  FREEBSD-5-STABLE  -  FREEBSD-5-0  -  FREEBSD-4-STABLE  -  FREEBSD-3-STABLE  -  FREEBSD22  -  l41  -  OPENBSD  -  linux-2.6  -  MK84  -  PLAN9  -  xnu-8792 
SearchContext: -  none  -  3  -  10 

    1 /*
    2  * Simple NUMA memory policy for the Linux kernel.
    3  *
    4  * Copyright 2003,2004 Andi Kleen, SuSE Labs.
    5  * (C) Copyright 2005 Christoph Lameter, Silicon Graphics, Inc.
    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.
   21  *
   22  * bind           Only allocate memory on a specific set of nodes,
   23  *                no fallback.
   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  *
   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.
   33  *
   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 */
   67 
   68 #include <linux/mempolicy.h>
   69 #include <linux/mm.h>
   70 #include <linux/highmem.h>
   71 #include <linux/hugetlb.h>
   72 #include <linux/kernel.h>
   73 #include <linux/sched.h>
   74 #include <linux/nodemask.h>
   75 #include <linux/cpuset.h>
   76 #include <linux/slab.h>
   77 #include <linux/string.h>
   78 #include <linux/export.h>
   79 #include <linux/nsproxy.h>
   80 #include <linux/interrupt.h>
   81 #include <linux/init.h>
   82 #include <linux/compat.h>
   83 #include <linux/swap.h>
   84 #include <linux/seq_file.h>
   85 #include <linux/proc_fs.h>
   86 #include <linux/migrate.h>
   87 #include <linux/ksm.h>
   88 #include <linux/rmap.h>
   89 #include <linux/security.h>
   90 #include <linux/syscalls.h>
   91 #include <linux/ctype.h>
   92 #include <linux/mm_inline.h>
   93 #include <linux/mmu_notifier.h>
   94 
   95 #include <asm/tlbflush.h>
   96 #include <asm/uaccess.h>
   97 #include <linux/random.h>
   98 
   99 #include "internal.h"
  100 
  101 /* Internal flags */
  102 #define MPOL_MF_DISCONTIG_OK (MPOL_MF_INTERNAL << 0)    /* Skip checks for continuous vmas */
  103 #define MPOL_MF_INVERT (MPOL_MF_INTERNAL << 1)          /* Invert check for nodemask */
  104 
  105 static struct kmem_cache *policy_cache;
  106 static struct kmem_cache *sn_cache;
  107 
  108 /* Highest zone. An specific allocation for a zone below that is not
  109    policied. */
  110 enum zone_type policy_zone = 0;
  111 
  112 /*
  113  * run-time system-wide default policy => local allocation
  114  */
  115 static struct mempolicy default_policy = {
  116         .refcnt = ATOMIC_INIT(1), /* never free it */
  117         .mode = MPOL_PREFERRED,
  118         .flags = MPOL_F_LOCAL,
  119 };
  120 
  121 static struct mempolicy preferred_node_policy[MAX_NUMNODES];
  122 
  123 static struct mempolicy *get_task_policy(struct task_struct *p)
  124 {
  125         struct mempolicy *pol = p->mempolicy;
  126         int node;
  127 
  128         if (!pol) {
  129                 node = numa_node_id();
  130                 if (node != -1)
  131                         pol = &preferred_node_policy[node];
  132 
  133                 /* preferred_node_policy is not initialised early in boot */
  134                 if (!pol->mode)
  135                         pol = NULL;
  136         }
  137 
  138         return pol;
  139 }
  140 
  141 static const struct mempolicy_operations {
  142         int (*create)(struct mempolicy *pol, const nodemask_t *nodes);
  143         /*
  144          * If read-side task has no lock to protect task->mempolicy, write-side
  145          * task will rebind the task->mempolicy by two step. The first step is
  146          * setting all the newly nodes, and the second step is cleaning all the
  147          * disallowed nodes. In this way, we can avoid finding no node to alloc
  148          * page.
  149          * If we have a lock to protect task->mempolicy in read-side, we do
  150          * rebind directly.
  151          *
  152          * step:
  153          *      MPOL_REBIND_ONCE - do rebind work at once
  154          *      MPOL_REBIND_STEP1 - set all the newly nodes
  155          *      MPOL_REBIND_STEP2 - clean all the disallowed nodes
  156          */
  157         void (*rebind)(struct mempolicy *pol, const nodemask_t *nodes,
  158                         enum mpol_rebind_step step);
  159 } mpol_ops[MPOL_MAX];
  160 
  161 /* Check that the nodemask contains at least one populated zone */
  162 static int is_valid_nodemask(const nodemask_t *nodemask)
  163 {
  164         int nd, k;
  165 
  166         for_each_node_mask(nd, *nodemask) {
  167                 struct zone *z;
  168 
  169                 for (k = 0; k <= policy_zone; k++) {
  170                         z = &NODE_DATA(nd)->node_zones[k];
  171                         if (z->present_pages > 0)
  172                                 return 1;
  173                 }
  174         }
  175 
  176         return 0;
  177 }
  178 
  179 static inline int mpol_store_user_nodemask(const struct mempolicy *pol)
  180 {
  181         return pol->flags & MPOL_MODE_FLAGS;
  182 }
  183 
  184 static void mpol_relative_nodemask(nodemask_t *ret, const nodemask_t *orig,
  185                                    const nodemask_t *rel)
  186 {
  187         nodemask_t tmp;
  188         nodes_fold(tmp, *orig, nodes_weight(*rel));
  189         nodes_onto(*ret, tmp, *rel);
  190 }
  191 
  192 static int mpol_new_interleave(struct mempolicy *pol, const nodemask_t *nodes)
  193 {
  194         if (nodes_empty(*nodes))
  195                 return -EINVAL;
  196         pol->v.nodes = *nodes;
  197         return 0;
  198 }
  199 
  200 static int mpol_new_preferred(struct mempolicy *pol, const nodemask_t *nodes)
  201 {
  202         if (!nodes)
  203                 pol->flags |= MPOL_F_LOCAL;     /* local allocation */
  204         else if (nodes_empty(*nodes))
  205                 return -EINVAL;                 /*  no allowed nodes */
  206         else
  207                 pol->v.preferred_node = first_node(*nodes);
  208         return 0;
  209 }
  210 
  211 static int mpol_new_bind(struct mempolicy *pol, const nodemask_t *nodes)
  212 {
  213         if (!is_valid_nodemask(nodes))
  214                 return -EINVAL;
  215         pol->v.nodes = *nodes;
  216         return 0;
  217 }
  218 
  219 /*
  220  * mpol_set_nodemask is called after mpol_new() to set up the nodemask, if
  221  * any, for the new policy.  mpol_new() has already validated the nodes
  222  * parameter with respect to the policy mode and flags.  But, we need to
  223  * handle an empty nodemask with MPOL_PREFERRED here.
  224  *
  225  * Must be called holding task's alloc_lock to protect task's mems_allowed
  226  * and mempolicy.  May also be called holding the mmap_semaphore for write.
  227  */
  228 static int mpol_set_nodemask(struct mempolicy *pol,
  229                      const nodemask_t *nodes, struct nodemask_scratch *nsc)
  230 {
  231         int ret;
  232 
  233         /* if mode is MPOL_DEFAULT, pol is NULL. This is right. */
  234         if (pol == NULL)
  235                 return 0;
  236         /* Check N_MEMORY */
  237         nodes_and(nsc->mask1,
  238                   cpuset_current_mems_allowed, node_states[N_MEMORY]);
  239 
  240         VM_BUG_ON(!nodes);
  241         if (pol->mode == MPOL_PREFERRED && nodes_empty(*nodes))
  242                 nodes = NULL;   /* explicit local allocation */
  243         else {
  244                 if (pol->flags & MPOL_F_RELATIVE_NODES)
  245                         mpol_relative_nodemask(&nsc->mask2, nodes,&nsc->mask1);
  246                 else
  247                         nodes_and(nsc->mask2, *nodes, nsc->mask1);
  248 
  249                 if (mpol_store_user_nodemask(pol))
  250                         pol->w.user_nodemask = *nodes;
  251                 else
  252                         pol->w.cpuset_mems_allowed =
  253                                                 cpuset_current_mems_allowed;
  254         }
  255 
  256         if (nodes)
  257                 ret = mpol_ops[pol->mode].create(pol, &nsc->mask2);
  258         else
  259                 ret = mpol_ops[pol->mode].create(pol, NULL);
  260         return ret;
  261 }
  262 
  263 /*
  264  * This function just creates a new policy, does some check and simple
  265  * initialization. You must invoke mpol_set_nodemask() to set nodes.
  266  */
  267 static struct mempolicy *mpol_new(unsigned short mode, unsigned short flags,
  268                                   nodemask_t *nodes)
  269 {
  270         struct mempolicy *policy;
  271 
  272         pr_debug("setting mode %d flags %d nodes[0] %lx\n",
  273                  mode, flags, nodes ? nodes_addr(*nodes)[0] : -1);
  274 
  275         if (mode == MPOL_DEFAULT) {
  276                 if (nodes && !nodes_empty(*nodes))
  277                         return ERR_PTR(-EINVAL);
  278                 return NULL;
  279         }
  280         VM_BUG_ON(!nodes);
  281 
  282         /*
  283          * MPOL_PREFERRED cannot be used with MPOL_F_STATIC_NODES or
  284          * MPOL_F_RELATIVE_NODES if the nodemask is empty (local allocation).
  285          * All other modes require a valid pointer to a non-empty nodemask.
  286          */
  287         if (mode == MPOL_PREFERRED) {
  288                 if (nodes_empty(*nodes)) {
  289                         if (((flags & MPOL_F_STATIC_NODES) ||
  290                              (flags & MPOL_F_RELATIVE_NODES)))
  291                                 return ERR_PTR(-EINVAL);
  292                 }
  293         } else if (mode == MPOL_LOCAL) {
  294                 if (!nodes_empty(*nodes))
  295                         return ERR_PTR(-EINVAL);
  296                 mode = MPOL_PREFERRED;
  297         } else if (nodes_empty(*nodes))
  298                 return ERR_PTR(-EINVAL);
  299         policy = kmem_cache_alloc(policy_cache, GFP_KERNEL);
  300         if (!policy)
  301                 return ERR_PTR(-ENOMEM);
  302         atomic_set(&policy->refcnt, 1);
  303         policy->mode = mode;
  304         policy->flags = flags;
  305 
  306         return policy;
  307 }
  308 
  309 /* Slow path of a mpol destructor. */
  310 void __mpol_put(struct mempolicy *p)
  311 {
  312         if (!atomic_dec_and_test(&p->refcnt))
  313                 return;
  314         kmem_cache_free(policy_cache, p);
  315 }
  316 
  317 static void mpol_rebind_default(struct mempolicy *pol, const nodemask_t *nodes,
  318                                 enum mpol_rebind_step step)
  319 {
  320 }
  321 
  322 /*
  323  * step:
  324  *      MPOL_REBIND_ONCE  - do rebind work at once
  325  *      MPOL_REBIND_STEP1 - set all the newly nodes
  326  *      MPOL_REBIND_STEP2 - clean all the disallowed nodes
  327  */
  328 static void mpol_rebind_nodemask(struct mempolicy *pol, const nodemask_t *nodes,
  329                                  enum mpol_rebind_step step)
  330 {
  331         nodemask_t tmp;
  332 
  333         if (pol->flags & MPOL_F_STATIC_NODES)
  334                 nodes_and(tmp, pol->w.user_nodemask, *nodes);
  335         else if (pol->flags & MPOL_F_RELATIVE_NODES)
  336                 mpol_relative_nodemask(&tmp, &pol->w.user_nodemask, nodes);
  337         else {
  338                 /*
  339                  * if step == 1, we use ->w.cpuset_mems_allowed to cache the
  340                  * result
  341                  */
  342                 if (step == MPOL_REBIND_ONCE || step == MPOL_REBIND_STEP1) {
  343                         nodes_remap(tmp, pol->v.nodes,
  344                                         pol->w.cpuset_mems_allowed, *nodes);
  345                         pol->w.cpuset_mems_allowed = step ? tmp : *nodes;
  346                 } else if (step == MPOL_REBIND_STEP2) {
  347                         tmp = pol->w.cpuset_mems_allowed;
  348                         pol->w.cpuset_mems_allowed = *nodes;
  349                 } else
  350                         BUG();
  351         }
  352 
  353         if (nodes_empty(tmp))
  354                 tmp = *nodes;
  355 
  356         if (step == MPOL_REBIND_STEP1)
  357                 nodes_or(pol->v.nodes, pol->v.nodes, tmp);
  358         else if (step == MPOL_REBIND_ONCE || step == MPOL_REBIND_STEP2)
  359                 pol->v.nodes = tmp;
  360         else
  361                 BUG();
  362 
  363         if (!node_isset(current->il_next, tmp)) {
  364                 current->il_next = next_node(current->il_next, tmp);
  365                 if (current->il_next >= MAX_NUMNODES)
  366                         current->il_next = first_node(tmp);
  367                 if (current->il_next >= MAX_NUMNODES)
  368                         current->il_next = numa_node_id();
  369         }
  370 }
  371 
  372 static void mpol_rebind_preferred(struct mempolicy *pol,
  373                                   const nodemask_t *nodes,
  374                                   enum mpol_rebind_step step)
  375 {
  376         nodemask_t tmp;
  377 
  378         if (pol->flags & MPOL_F_STATIC_NODES) {
  379                 int node = first_node(pol->w.user_nodemask);
  380 
  381                 if (node_isset(node, *nodes)) {
  382                         pol->v.preferred_node = node;
  383                         pol->flags &= ~MPOL_F_LOCAL;
  384                 } else
  385                         pol->flags |= MPOL_F_LOCAL;
  386         } else if (pol->flags & MPOL_F_RELATIVE_NODES) {
  387                 mpol_relative_nodemask(&tmp, &pol->w.user_nodemask, nodes);
  388                 pol->v.preferred_node = first_node(tmp);
  389         } else if (!(pol->flags & MPOL_F_LOCAL)) {
  390                 pol->v.preferred_node = node_remap(pol->v.preferred_node,
  391                                                    pol->w.cpuset_mems_allowed,
  392                                                    *nodes);
  393                 pol->w.cpuset_mems_allowed = *nodes;
  394         }
  395 }
  396 
  397 /*
  398  * mpol_rebind_policy - Migrate a policy to a different set of nodes
  399  *
  400  * If read-side task has no lock to protect task->mempolicy, write-side
  401  * task will rebind the task->mempolicy by two step. The first step is
  402  * setting all the newly nodes, and the second step is cleaning all the
  403  * disallowed nodes. In this way, we can avoid finding no node to alloc
  404  * page.
  405  * If we have a lock to protect task->mempolicy in read-side, we do
  406  * rebind directly.
  407  *
  408  * step:
  409  *      MPOL_REBIND_ONCE  - do rebind work at once
  410  *      MPOL_REBIND_STEP1 - set all the newly nodes
  411  *      MPOL_REBIND_STEP2 - clean all the disallowed nodes
  412  */
  413 static void mpol_rebind_policy(struct mempolicy *pol, const nodemask_t *newmask,
  414                                 enum mpol_rebind_step step)
  415 {
  416         if (!pol)
  417                 return;
  418         if (!mpol_store_user_nodemask(pol) && step == MPOL_REBIND_ONCE &&
  419             nodes_equal(pol->w.cpuset_mems_allowed, *newmask))
  420                 return;
  421 
  422         if (step == MPOL_REBIND_STEP1 && (pol->flags & MPOL_F_REBINDING))
  423                 return;
  424 
  425         if (step == MPOL_REBIND_STEP2 && !(pol->flags & MPOL_F_REBINDING))
  426                 BUG();
  427 
  428         if (step == MPOL_REBIND_STEP1)
  429                 pol->flags |= MPOL_F_REBINDING;
  430         else if (step == MPOL_REBIND_STEP2)
  431                 pol->flags &= ~MPOL_F_REBINDING;
  432         else if (step >= MPOL_REBIND_NSTEP)
  433                 BUG();
  434 
  435         mpol_ops[pol->mode].rebind(pol, newmask, step);
  436 }
  437 
  438 /*
  439  * Wrapper for mpol_rebind_policy() that just requires task
  440  * pointer, and updates task mempolicy.
  441  *
  442  * Called with task's alloc_lock held.
  443  */
  444 
  445 void mpol_rebind_task(struct task_struct *tsk, const nodemask_t *new,
  446                         enum mpol_rebind_step step)
  447 {
  448         mpol_rebind_policy(tsk->mempolicy, new, step);
  449 }
  450 
  451 /*
  452  * Rebind each vma in mm to new nodemask.
  453  *
  454  * Call holding a reference to mm.  Takes mm->mmap_sem during call.
  455  */
  456 
  457 void mpol_rebind_mm(struct mm_struct *mm, nodemask_t *new)
  458 {
  459         struct vm_area_struct *vma;
  460 
  461         down_write(&mm->mmap_sem);
  462         for (vma = mm->mmap; vma; vma = vma->vm_next)
  463                 mpol_rebind_policy(vma->vm_policy, new, MPOL_REBIND_ONCE);
  464         up_write(&mm->mmap_sem);
  465 }
  466 
  467 static const struct mempolicy_operations mpol_ops[MPOL_MAX] = {
  468         [MPOL_DEFAULT] = {
  469                 .rebind = mpol_rebind_default,
  470         },
  471         [MPOL_INTERLEAVE] = {
  472                 .create = mpol_new_interleave,
  473                 .rebind = mpol_rebind_nodemask,
  474         },
  475         [MPOL_PREFERRED] = {
  476                 .create = mpol_new_preferred,
  477                 .rebind = mpol_rebind_preferred,
  478         },
  479         [MPOL_BIND] = {
  480                 .create = mpol_new_bind,
  481                 .rebind = mpol_rebind_nodemask,
  482         },
  483 };
  484 
  485 static void migrate_page_add(struct page *page, struct list_head *pagelist,
  486                                 unsigned long flags);
  487 
  488 /* Scan through pages checking if pages follow certain conditions. */
  489 static int check_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
  490                 unsigned long addr, unsigned long end,
  491                 const nodemask_t *nodes, unsigned long flags,
  492                 void *private)
  493 {
  494         pte_t *orig_pte;
  495         pte_t *pte;
  496         spinlock_t *ptl;
  497 
  498         orig_pte = pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
  499         do {
  500                 struct page *page;
  501                 int nid;
  502 
  503                 if (!pte_present(*pte))
  504                         continue;
  505                 page = vm_normal_page(vma, addr, *pte);
  506                 if (!page)
  507                         continue;
  508                 /*
  509                  * vm_normal_page() filters out zero pages, but there might
  510                  * still be PageReserved pages to skip, perhaps in a VDSO.
  511                  * And we cannot move PageKsm pages sensibly or safely yet.
  512                  */
  513                 if (PageReserved(page) || PageKsm(page))
  514                         continue;
  515                 nid = page_to_nid(page);
  516                 if (node_isset(nid, *nodes) == !!(flags & MPOL_MF_INVERT))
  517                         continue;
  518 
  519                 if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL))
  520                         migrate_page_add(page, private, flags);
  521                 else
  522                         break;
  523         } while (pte++, addr += PAGE_SIZE, addr != end);
  524         pte_unmap_unlock(orig_pte, ptl);
  525         return addr != end;
  526 }
  527 
  528 static inline int check_pmd_range(struct vm_area_struct *vma, pud_t *pud,
  529                 unsigned long addr, unsigned long end,
  530                 const nodemask_t *nodes, unsigned long flags,
  531                 void *private)
  532 {
  533         pmd_t *pmd;
  534         unsigned long next;
  535 
  536         pmd = pmd_offset(pud, addr);
  537         do {
  538                 next = pmd_addr_end(addr, end);
  539                 split_huge_page_pmd(vma, addr, pmd);
  540                 if (pmd_none_or_trans_huge_or_clear_bad(pmd))
  541                         continue;
  542                 if (check_pte_range(vma, pmd, addr, next, nodes,
  543                                     flags, private))
  544                         return -EIO;
  545         } while (pmd++, addr = next, addr != end);
  546         return 0;
  547 }
  548 
  549 static inline int check_pud_range(struct vm_area_struct *vma, pgd_t *pgd,
  550                 unsigned long addr, unsigned long end,
  551                 const nodemask_t *nodes, unsigned long flags,
  552                 void *private)
  553 {
  554         pud_t *pud;
  555         unsigned long next;
  556 
  557         pud = pud_offset(pgd, addr);
  558         do {
  559                 next = pud_addr_end(addr, end);
  560                 if (pud_none_or_clear_bad(pud))
  561                         continue;
  562                 if (check_pmd_range(vma, pud, addr, next, nodes,
  563                                     flags, private))
  564                         return -EIO;
  565         } while (pud++, addr = next, addr != end);
  566         return 0;
  567 }
  568 
  569 static inline int check_pgd_range(struct vm_area_struct *vma,
  570                 unsigned long addr, unsigned long end,
  571                 const nodemask_t *nodes, unsigned long flags,
  572                 void *private)
  573 {
  574         pgd_t *pgd;
  575         unsigned long next;
  576 
  577         pgd = pgd_offset(vma->vm_mm, addr);
  578         do {
  579                 next = pgd_addr_end(addr, end);
  580                 if (pgd_none_or_clear_bad(pgd))
  581                         continue;
  582                 if (check_pud_range(vma, pgd, addr, next, nodes,
  583                                     flags, private))
  584                         return -EIO;
  585         } while (pgd++, addr = next, addr != end);
  586         return 0;
  587 }
  588 
  589 #ifdef CONFIG_ARCH_USES_NUMA_PROT_NONE
  590 /*
  591  * This is used to mark a range of virtual addresses to be inaccessible.
  592  * These are later cleared by a NUMA hinting fault. Depending on these
  593  * faults, pages may be migrated for better NUMA placement.
  594  *
  595  * This is assuming that NUMA faults are handled using PROT_NONE. If
  596  * an architecture makes a different choice, it will need further
  597  * changes to the core.
  598  */
  599 unsigned long change_prot_numa(struct vm_area_struct *vma,
  600                         unsigned long addr, unsigned long end)
  601 {
  602         int nr_updated;
  603         BUILD_BUG_ON(_PAGE_NUMA != _PAGE_PROTNONE);
  604 
  605         nr_updated = change_protection(vma, addr, end, vma->vm_page_prot, 0, 1);
  606         if (nr_updated)
  607                 count_vm_numa_events(NUMA_PTE_UPDATES, nr_updated);
  608 
  609         return nr_updated;
  610 }
  611 #else
  612 static unsigned long change_prot_numa(struct vm_area_struct *vma,
  613                         unsigned long addr, unsigned long end)
  614 {
  615         return 0;
  616 }
  617 #endif /* CONFIG_ARCH_USES_NUMA_PROT_NONE */
  618 
  619 /*
  620  * Check if all pages in a range are on a set of nodes.
  621  * If pagelist != NULL then isolate pages from the LRU and
  622  * put them on the pagelist.
  623  */
  624 static struct vm_area_struct *
  625 check_range(struct mm_struct *mm, unsigned long start, unsigned long end,
  626                 const nodemask_t *nodes, unsigned long flags, void *private)
  627 {
  628         int err;
  629         struct vm_area_struct *first, *vma, *prev;
  630 
  631 
  632         first = find_vma(mm, start);
  633         if (!first)
  634                 return ERR_PTR(-EFAULT);
  635         prev = NULL;
  636         for (vma = first; vma && vma->vm_start < end; vma = vma->vm_next) {
  637                 unsigned long endvma = vma->vm_end;
  638 
  639                 if (endvma > end)
  640                         endvma = end;
  641                 if (vma->vm_start > start)
  642                         start = vma->vm_start;
  643 
  644                 if (!(flags & MPOL_MF_DISCONTIG_OK)) {
  645                         if (!vma->vm_next && vma->vm_end < end)
  646                                 return ERR_PTR(-EFAULT);
  647                         if (prev && prev->vm_end < vma->vm_start)
  648                                 return ERR_PTR(-EFAULT);
  649                 }
  650 
  651                 if (is_vm_hugetlb_page(vma))
  652                         goto next;
  653 
  654                 if (flags & MPOL_MF_LAZY) {
  655                         change_prot_numa(vma, start, endvma);
  656                         goto next;
  657                 }
  658 
  659                 if ((flags & MPOL_MF_STRICT) ||
  660                      ((flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) &&
  661                       vma_migratable(vma))) {
  662 
  663                         err = check_pgd_range(vma, start, endvma, nodes,
  664                                                 flags, private);
  665                         if (err) {
  666                                 first = ERR_PTR(err);
  667                                 break;
  668                         }
  669                 }
  670 next:
  671                 prev = vma;
  672         }
  673         return first;
  674 }
  675 
  676 /*
  677  * Apply policy to a single VMA
  678  * This must be called with the mmap_sem held for writing.
  679  */
  680 static int vma_replace_policy(struct vm_area_struct *vma,
  681                                                 struct mempolicy *pol)
  682 {
  683         int err;
  684         struct mempolicy *old;
  685         struct mempolicy *new;
  686 
  687         pr_debug("vma %lx-%lx/%lx vm_ops %p vm_file %p set_policy %p\n",
  688                  vma->vm_start, vma->vm_end, vma->vm_pgoff,
  689                  vma->vm_ops, vma->vm_file,
  690                  vma->vm_ops ? vma->vm_ops->set_policy : NULL);
  691 
  692         new = mpol_dup(pol);
  693         if (IS_ERR(new))
  694                 return PTR_ERR(new);
  695 
  696         if (vma->vm_ops && vma->vm_ops->set_policy) {
  697                 err = vma->vm_ops->set_policy(vma, new);
  698                 if (err)
  699                         goto err_out;
  700         }
  701 
  702         old = vma->vm_policy;
  703         vma->vm_policy = new; /* protected by mmap_sem */
  704         mpol_put(old);
  705 
  706         return 0;
  707  err_out:
  708         mpol_put(new);
  709         return err;
  710 }
  711 
  712 /* Step 2: apply policy to a range and do splits. */
  713 static int mbind_range(struct mm_struct *mm, unsigned long start,
  714                        unsigned long end, struct mempolicy *new_pol)
  715 {
  716         struct vm_area_struct *next;
  717         struct vm_area_struct *prev;
  718         struct vm_area_struct *vma;
  719         int err = 0;
  720         pgoff_t pgoff;
  721         unsigned long vmstart;
  722         unsigned long vmend;
  723 
  724         vma = find_vma(mm, start);
  725         if (!vma || vma->vm_start > start)
  726                 return -EFAULT;
  727 
  728         prev = vma->vm_prev;
  729         if (start > vma->vm_start)
  730                 prev = vma;
  731 
  732         for (; vma && vma->vm_start < end; prev = vma, vma = next) {
  733                 next = vma->vm_next;
  734                 vmstart = max(start, vma->vm_start);
  735                 vmend   = min(end, vma->vm_end);
  736 
  737                 if (mpol_equal(vma_policy(vma), new_pol))
  738                         continue;
  739 
  740                 pgoff = vma->vm_pgoff +
  741                         ((vmstart - vma->vm_start) >> PAGE_SHIFT);
  742                 prev = vma_merge(mm, prev, vmstart, vmend, vma->vm_flags,
  743                                   vma->anon_vma, vma->vm_file, pgoff,
  744                                   new_pol);
  745                 if (prev) {
  746                         vma = prev;
  747                         next = vma->vm_next;
  748                         continue;
  749                 }
  750                 if (vma->vm_start != vmstart) {
  751                         err = split_vma(vma->vm_mm, vma, vmstart, 1);
  752                         if (err)
  753                                 goto out;
  754                 }
  755                 if (vma->vm_end != vmend) {
  756                         err = split_vma(vma->vm_mm, vma, vmend, 0);
  757                         if (err)
  758                                 goto out;
  759                 }
  760                 err = vma_replace_policy(vma, new_pol);
  761                 if (err)
  762                         goto out;
  763         }
  764 
  765  out:
  766         return err;
  767 }
  768 
  769 /*
  770  * Update task->flags PF_MEMPOLICY bit: set iff non-default
  771  * mempolicy.  Allows more rapid checking of this (combined perhaps
  772  * with other PF_* flag bits) on memory allocation hot code paths.
  773  *
  774  * If called from outside this file, the task 'p' should -only- be
  775  * a newly forked child not yet visible on the task list, because
  776  * manipulating the task flags of a visible task is not safe.
  777  *
  778  * The above limitation is why this routine has the funny name
  779  * mpol_fix_fork_child_flag().
  780  *
  781  * It is also safe to call this with a task pointer of current,
  782  * which the static wrapper mpol_set_task_struct_flag() does,
  783  * for use within this file.
  784  */
  785 
  786 void mpol_fix_fork_child_flag(struct task_struct *p)
  787 {
  788         if (p->mempolicy)
  789                 p->flags |= PF_MEMPOLICY;
  790         else
  791                 p->flags &= ~PF_MEMPOLICY;
  792 }
  793 
  794 static void mpol_set_task_struct_flag(void)
  795 {
  796         mpol_fix_fork_child_flag(current);
  797 }
  798 
  799 /* Set the process memory policy */
  800 static long do_set_mempolicy(unsigned short mode, unsigned short flags,
  801                              nodemask_t *nodes)
  802 {
  803         struct mempolicy *new, *old;
  804         struct mm_struct *mm = current->mm;
  805         NODEMASK_SCRATCH(scratch);
  806         int ret;
  807 
  808         if (!scratch)
  809                 return -ENOMEM;
  810 
  811         new = mpol_new(mode, flags, nodes);
  812         if (IS_ERR(new)) {
  813                 ret = PTR_ERR(new);
  814                 goto out;
  815         }
  816         /*
  817          * prevent changing our mempolicy while show_numa_maps()
  818          * is using it.
  819          * Note:  do_set_mempolicy() can be called at init time
  820          * with no 'mm'.
  821          */
  822         if (mm)
  823                 down_write(&mm->mmap_sem);
  824         task_lock(current);
  825         ret = mpol_set_nodemask(new, nodes, scratch);
  826         if (ret) {
  827                 task_unlock(current);
  828                 if (mm)
  829                         up_write(&mm->mmap_sem);
  830                 mpol_put(new);
  831                 goto out;
  832         }
  833         old = current->mempolicy;
  834         current->mempolicy = new;
  835         mpol_set_task_struct_flag();
  836         if (new && new->mode == MPOL_INTERLEAVE &&
  837             nodes_weight(new->v.nodes))
  838                 current->il_next = first_node(new->v.nodes);
  839         task_unlock(current);
  840         if (mm)
  841                 up_write(&mm->mmap_sem);
  842 
  843         mpol_put(old);
  844         ret = 0;
  845 out:
  846         NODEMASK_SCRATCH_FREE(scratch);
  847         return ret;
  848 }
  849 
  850 /*
  851  * Return nodemask for policy for get_mempolicy() query
  852  *
  853  * Called with task's alloc_lock held
  854  */
  855 static void get_policy_nodemask(struct mempolicy *p, nodemask_t *nodes)
  856 {
  857         nodes_clear(*nodes);
  858         if (p == &default_policy)
  859                 return;
  860 
  861         switch (p->mode) {
  862         case MPOL_BIND:
  863                 /* Fall through */
  864         case MPOL_INTERLEAVE:
  865                 *nodes = p->v.nodes;
  866                 break;
  867         case MPOL_PREFERRED:
  868                 if (!(p->flags & MPOL_F_LOCAL))
  869                         node_set(p->v.preferred_node, *nodes);
  870                 /* else return empty node mask for local allocation */
  871                 break;
  872         default:
  873                 BUG();
  874         }
  875 }
  876 
  877 static int lookup_node(struct mm_struct *mm, unsigned long addr)
  878 {
  879         struct page *p;
  880         int err;
  881 
  882         err = get_user_pages(current, mm, addr & PAGE_MASK, 1, 0, 0, &p, NULL);
  883         if (err >= 0) {
  884                 err = page_to_nid(p);
  885                 put_page(p);
  886         }
  887         return err;
  888 }
  889 
  890 /* Retrieve NUMA policy */
  891 static long do_get_mempolicy(int *policy, nodemask_t *nmask,
  892                              unsigned long addr, unsigned long flags)
  893 {
  894         int err;
  895         struct mm_struct *mm = current->mm;
  896         struct vm_area_struct *vma = NULL;
  897         struct mempolicy *pol = current->mempolicy;
  898 
  899         if (flags &
  900                 ~(unsigned long)(MPOL_F_NODE|MPOL_F_ADDR|MPOL_F_MEMS_ALLOWED))
  901                 return -EINVAL;
  902 
  903         if (flags & MPOL_F_MEMS_ALLOWED) {
  904                 if (flags & (MPOL_F_NODE|MPOL_F_ADDR))
  905                         return -EINVAL;
  906                 *policy = 0;    /* just so it's initialized */
  907                 task_lock(current);
  908                 *nmask  = cpuset_current_mems_allowed;
  909                 task_unlock(current);
  910                 return 0;
  911         }
  912 
  913         if (flags & MPOL_F_ADDR) {
  914                 /*
  915                  * Do NOT fall back to task policy if the
  916                  * vma/shared policy at addr is NULL.  We
  917                  * want to return MPOL_DEFAULT in this case.
  918                  */
  919                 down_read(&mm->mmap_sem);
  920                 vma = find_vma_intersection(mm, addr, addr+1);
  921                 if (!vma) {
  922                         up_read(&mm->mmap_sem);
  923                         return -EFAULT;
  924                 }
  925                 if (vma->vm_ops && vma->vm_ops->get_policy)
  926                         pol = vma->vm_ops->get_policy(vma, addr);
  927                 else
  928                         pol = vma->vm_policy;
  929         } else if (addr)
  930                 return -EINVAL;
  931 
  932         if (!pol)
  933                 pol = &default_policy;  /* indicates default behavior */
  934 
  935         if (flags & MPOL_F_NODE) {
  936                 if (flags & MPOL_F_ADDR) {
  937                         err = lookup_node(mm, addr);
  938                         if (err < 0)
  939                                 goto out;
  940                         *policy = err;
  941                 } else if (pol == current->mempolicy &&
  942                                 pol->mode == MPOL_INTERLEAVE) {
  943                         *policy = current->il_next;
  944                 } else {
  945                         err = -EINVAL;
  946                         goto out;
  947                 }
  948         } else {
  949                 *policy = pol == &default_policy ? MPOL_DEFAULT :
  950                                                 pol->mode;
  951                 /*
  952                  * Internal mempolicy flags must be masked off before exposing
  953                  * the policy to userspace.
  954                  */
  955                 *policy |= (pol->flags & MPOL_MODE_FLAGS);
  956         }
  957 
  958         if (vma) {
  959                 up_read(&current->mm->mmap_sem);
  960                 vma = NULL;
  961         }
  962 
  963         err = 0;
  964         if (nmask) {
  965                 if (mpol_store_user_nodemask(pol)) {
  966                         *nmask = pol->w.user_nodemask;
  967                 } else {
  968                         task_lock(current);
  969                         get_policy_nodemask(pol, nmask);
  970                         task_unlock(current);
  971                 }
  972         }
  973 
  974  out:
  975         mpol_cond_put(pol);
  976         if (vma)
  977                 up_read(&current->mm->mmap_sem);
  978         return err;
  979 }
  980 
  981 #ifdef CONFIG_MIGRATION
  982 /*
  983  * page migration
  984  */
  985 static void migrate_page_add(struct page *page, struct list_head *pagelist,
  986                                 unsigned long flags)
  987 {
  988         /*
  989          * Avoid migrating a page that is shared with others.
  990          */
  991         if ((flags & MPOL_MF_MOVE_ALL) || page_mapcount(page) == 1) {
  992                 if (!isolate_lru_page(page)) {
  993                         list_add_tail(&page->lru, pagelist);
  994                         inc_zone_page_state(page, NR_ISOLATED_ANON +
  995                                             page_is_file_cache(page));
  996                 }
  997         }
  998 }
  999 
 1000 static struct page *new_node_page(struct page *page, unsigned long node, int **x)
 1001 {
 1002         return alloc_pages_exact_node(node, GFP_HIGHUSER_MOVABLE, 0);
 1003 }
 1004 
 1005 /*
 1006  * Migrate pages from one node to a target node.
 1007  * Returns error or the number of pages not migrated.
 1008  */
 1009 static int migrate_to_node(struct mm_struct *mm, int source, int dest,
 1010                            int flags)
 1011 {
 1012         nodemask_t nmask;
 1013         LIST_HEAD(pagelist);
 1014         int err = 0;
 1015 
 1016         nodes_clear(nmask);
 1017         node_set(source, nmask);
 1018 
 1019         /*
 1020          * This does not "check" the range but isolates all pages that
 1021          * need migration.  Between passing in the full user address
 1022          * space range and MPOL_MF_DISCONTIG_OK, this call can not fail.
 1023          */
 1024         VM_BUG_ON(!(flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)));
 1025         check_range(mm, mm->mmap->vm_start, mm->task_size, &nmask,
 1026                         flags | MPOL_MF_DISCONTIG_OK, &pagelist);
 1027 
 1028         if (!list_empty(&pagelist)) {
 1029                 err = migrate_pages(&pagelist, new_node_page, dest,
 1030                                                         false, MIGRATE_SYNC,
 1031                                                         MR_SYSCALL);
 1032                 if (err)
 1033                         putback_lru_pages(&pagelist);
 1034         }
 1035 
 1036         return err;
 1037 }
 1038 
 1039 /*
 1040  * Move pages between the two nodesets so as to preserve the physical
 1041  * layout as much as possible.
 1042  *
 1043  * Returns the number of page that could not be moved.
 1044  */
 1045 int do_migrate_pages(struct mm_struct *mm, const nodemask_t *from,
 1046                      const nodemask_t *to, int flags)
 1047 {
 1048         int busy = 0;
 1049         int err;
 1050         nodemask_t tmp;
 1051 
 1052         err = migrate_prep();
 1053         if (err)
 1054                 return err;
 1055 
 1056         down_read(&mm->mmap_sem);
 1057 
 1058         err = migrate_vmas(mm, from, to, flags);
 1059         if (err)
 1060                 goto out;
 1061 
 1062         /*
 1063          * Find a 'source' bit set in 'tmp' whose corresponding 'dest'
 1064          * bit in 'to' is not also set in 'tmp'.  Clear the found 'source'
 1065          * bit in 'tmp', and return that <source, dest> pair for migration.
 1066          * The pair of nodemasks 'to' and 'from' define the map.
 1067          *
 1068          * If no pair of bits is found that way, fallback to picking some
 1069          * pair of 'source' and 'dest' bits that are not the same.  If the
 1070          * 'source' and 'dest' bits are the same, this represents a node
 1071          * that will be migrating to itself, so no pages need move.
 1072          *
 1073          * If no bits are left in 'tmp', or if all remaining bits left
 1074          * in 'tmp' correspond to the same bit in 'to', return false
 1075          * (nothing left to migrate).
 1076          *
 1077          * This lets us pick a pair of nodes to migrate between, such that
 1078          * if possible the dest node is not already occupied by some other
 1079          * source node, minimizing the risk of overloading the memory on a
 1080          * node that would happen if we migrated incoming memory to a node
 1081          * before migrating outgoing memory source that same node.
 1082          *
 1083          * A single scan of tmp is sufficient.  As we go, we remember the
 1084          * most recent <s, d> pair that moved (s != d).  If we find a pair
 1085          * that not only moved, but what's better, moved to an empty slot
 1086          * (d is not set in tmp), then we break out then, with that pair.
 1087          * Otherwise when we finish scanning from_tmp, we at least have the
 1088          * most recent <s, d> pair that moved.  If we get all the way through
 1089          * the scan of tmp without finding any node that moved, much less
 1090          * moved to an empty node, then there is nothing left worth migrating.
 1091          */
 1092 
 1093         tmp = *from;
 1094         while (!nodes_empty(tmp)) {
 1095                 int s,d;
 1096                 int source = -1;
 1097                 int dest = 0;
 1098 
 1099                 for_each_node_mask(s, tmp) {
 1100 
 1101                         /*
 1102                          * do_migrate_pages() tries to maintain the relative
 1103                          * node relationship of the pages established between
 1104                          * threads and memory areas.
 1105                          *
 1106                          * However if the number of source nodes is not equal to
 1107                          * the number of destination nodes we can not preserve
 1108                          * this node relative relationship.  In that case, skip
 1109                          * copying memory from a node that is in the destination
 1110                          * mask.
 1111                          *
 1112                          * Example: [2,3,4] -> [3,4,5] moves everything.
 1113                          *          [0-7] - > [3,4,5] moves only 0,1,2,6,7.
 1114                          */
 1115 
 1116                         if ((nodes_weight(*from) != nodes_weight(*to)) &&
 1117                                                 (node_isset(s, *to)))
 1118                                 continue;
 1119 
 1120                         d = node_remap(s, *from, *to);
 1121                         if (s == d)
 1122                                 continue;
 1123 
 1124                         source = s;     /* Node moved. Memorize */
 1125                         dest = d;
 1126 
 1127                         /* dest not in remaining from nodes? */
 1128                         if (!node_isset(dest, tmp))
 1129                                 break;
 1130                 }
 1131                 if (source == -1)
 1132                         break;
 1133 
 1134                 node_clear(source, tmp);
 1135                 err = migrate_to_node(mm, source, dest, flags);
 1136                 if (err > 0)
 1137                         busy += err;
 1138                 if (err < 0)
 1139                         break;
 1140         }
 1141 out:
 1142         up_read(&mm->mmap_sem);
 1143         if (err < 0)
 1144                 return err;
 1145         return busy;
 1146 
 1147 }
 1148 
 1149 /*
 1150  * Allocate a new page for page migration based on vma policy.
 1151  * Start assuming that page is mapped by vma pointed to by @private.
 1152  * Search forward from there, if not.  N.B., this assumes that the
 1153  * list of pages handed to migrate_pages()--which is how we get here--
 1154  * is in virtual address order.
 1155  */
 1156 static struct page *new_vma_page(struct page *page, unsigned long private, int **x)
 1157 {
 1158         struct vm_area_struct *vma = (struct vm_area_struct *)private;
 1159         unsigned long uninitialized_var(address);
 1160 
 1161         while (vma) {
 1162                 address = page_address_in_vma(page, vma);
 1163                 if (address != -EFAULT)
 1164                         break;
 1165                 vma = vma->vm_next;
 1166         }
 1167 
 1168         /*
 1169          * if !vma, alloc_page_vma() will use task or system default policy
 1170          */
 1171         return alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address);
 1172 }
 1173 #else
 1174 
 1175 static void migrate_page_add(struct page *page, struct list_head *pagelist,
 1176                                 unsigned long flags)
 1177 {
 1178 }
 1179 
 1180 int do_migrate_pages(struct mm_struct *mm, const nodemask_t *from,
 1181                      const nodemask_t *to, int flags)
 1182 {
 1183         return -ENOSYS;
 1184 }
 1185 
 1186 static struct page *new_vma_page(struct page *page, unsigned long private, int **x)
 1187 {
 1188         return NULL;
 1189 }
 1190 #endif
 1191 
 1192 static long do_mbind(unsigned long start, unsigned long len,
 1193                      unsigned short mode, unsigned short mode_flags,
 1194                      nodemask_t *nmask, unsigned long flags)
 1195 {
 1196         struct vm_area_struct *vma;
 1197         struct mm_struct *mm = current->mm;
 1198         struct mempolicy *new;
 1199         unsigned long end;
 1200         int err;
 1201         LIST_HEAD(pagelist);
 1202 
 1203         if (flags & ~(unsigned long)MPOL_MF_VALID)
 1204                 return -EINVAL;
 1205         if ((flags & MPOL_MF_MOVE_ALL) && !capable(CAP_SYS_NICE))
 1206                 return -EPERM;
 1207 
 1208         if (start & ~PAGE_MASK)
 1209                 return -EINVAL;
 1210 
 1211         if (mode == MPOL_DEFAULT)
 1212                 flags &= ~MPOL_MF_STRICT;
 1213 
 1214         len = (len + PAGE_SIZE - 1) & PAGE_MASK;
 1215         end = start + len;
 1216 
 1217         if (end < start)
 1218                 return -EINVAL;
 1219         if (end == start)
 1220                 return 0;
 1221 
 1222         new = mpol_new(mode, mode_flags, nmask);
 1223         if (IS_ERR(new))
 1224                 return PTR_ERR(new);
 1225 
 1226         if (flags & MPOL_MF_LAZY)
 1227                 new->flags |= MPOL_F_MOF;
 1228 
 1229         /*
 1230          * If we are using the default policy then operation
 1231          * on discontinuous address spaces is okay after all
 1232          */
 1233         if (!new)
 1234                 flags |= MPOL_MF_DISCONTIG_OK;
 1235 
 1236         pr_debug("mbind %lx-%lx mode:%d flags:%d nodes:%lx\n",
 1237                  start, start + len, mode, mode_flags,
 1238                  nmask ? nodes_addr(*nmask)[0] : -1);
 1239 
 1240         if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) {
 1241 
 1242                 err = migrate_prep();
 1243                 if (err)
 1244                         goto mpol_out;
 1245         }
 1246         {
 1247                 NODEMASK_SCRATCH(scratch);
 1248                 if (scratch) {
 1249                         down_write(&mm->mmap_sem);
 1250                         task_lock(current);
 1251                         err = mpol_set_nodemask(new, nmask, scratch);
 1252                         task_unlock(current);
 1253                         if (err)
 1254                                 up_write(&mm->mmap_sem);
 1255                 } else
 1256                         err = -ENOMEM;
 1257                 NODEMASK_SCRATCH_FREE(scratch);
 1258         }
 1259         if (err)
 1260                 goto mpol_out;
 1261 
 1262         vma = check_range(mm, start, end, nmask,
 1263                           flags | MPOL_MF_INVERT, &pagelist);
 1264 
 1265         err = PTR_ERR(vma);     /* maybe ... */
 1266         if (!IS_ERR(vma))
 1267                 err = mbind_range(mm, start, end, new);
 1268 
 1269         if (!err) {
 1270                 int nr_failed = 0;
 1271 
 1272                 if (!list_empty(&pagelist)) {
 1273                         WARN_ON_ONCE(flags & MPOL_MF_LAZY);
 1274                         nr_failed = migrate_pages(&pagelist, new_vma_page,
 1275                                                 (unsigned long)vma,
 1276                                                 false, MIGRATE_SYNC,
 1277                                                 MR_MEMPOLICY_MBIND);
 1278                         if (nr_failed)
 1279                                 putback_lru_pages(&pagelist);
 1280                 }
 1281 
 1282                 if (nr_failed && (flags & MPOL_MF_STRICT))
 1283                         err = -EIO;
 1284         } else
 1285                 putback_lru_pages(&pagelist);
 1286 
 1287         up_write(&mm->mmap_sem);
 1288  mpol_out:
 1289         mpol_put(new);
 1290         return err;
 1291 }
 1292 
 1293 /*
 1294  * User space interface with variable sized bitmaps for nodelists.
 1295  */
 1296 
 1297 /* Copy a node mask from user space. */
 1298 static int get_nodes(nodemask_t *nodes, const unsigned long __user *nmask,
 1299                      unsigned long maxnode)
 1300 {
 1301         unsigned long k;
 1302         unsigned long nlongs;
 1303         unsigned long endmask;
 1304 
 1305         --maxnode;
 1306         nodes_clear(*nodes);
 1307         if (maxnode == 0 || !nmask)
 1308                 return 0;
 1309         if (maxnode > PAGE_SIZE*BITS_PER_BYTE)
 1310                 return -EINVAL;
 1311 
 1312         nlongs = BITS_TO_LONGS(maxnode);
 1313         if ((maxnode % BITS_PER_LONG) == 0)
 1314                 endmask = ~0UL;
 1315         else
 1316                 endmask = (1UL << (maxnode % BITS_PER_LONG)) - 1;
 1317 
 1318         /* When the user specified more nodes than supported just check
 1319            if the non supported part is all zero. */
 1320         if (nlongs > BITS_TO_LONGS(MAX_NUMNODES)) {
 1321                 if (nlongs > PAGE_SIZE/sizeof(long))
 1322                         return -EINVAL;
 1323                 for (k = BITS_TO_LONGS(MAX_NUMNODES); k < nlongs; k++) {
 1324                         unsigned long t;
 1325                         if (get_user(t, nmask + k))
 1326                                 return -EFAULT;
 1327                         if (k == nlongs - 1) {
 1328                                 if (t & endmask)
 1329                                         return -EINVAL;
 1330                         } else if (t)
 1331                                 return -EINVAL;
 1332                 }
 1333                 nlongs = BITS_TO_LONGS(MAX_NUMNODES);
 1334                 endmask = ~0UL;
 1335         }
 1336 
 1337         if (copy_from_user(nodes_addr(*nodes), nmask, nlongs*sizeof(unsigned long)))
 1338                 return -EFAULT;
 1339         nodes_addr(*nodes)[nlongs-1] &= endmask;
 1340         return 0;
 1341 }
 1342 
 1343 /* Copy a kernel node mask to user space */
 1344 static int copy_nodes_to_user(unsigned long __user *mask, unsigned long maxnode,
 1345                               nodemask_t *nodes)
 1346 {
 1347         unsigned long copy = ALIGN(maxnode-1, 64) / 8;
 1348         const int nbytes = BITS_TO_LONGS(MAX_NUMNODES) * sizeof(long);
 1349 
 1350         if (copy > nbytes) {
 1351                 if (copy > PAGE_SIZE)
 1352                         return -EINVAL;
 1353                 if (clear_user((char __user *)mask + nbytes, copy - nbytes))
 1354                         return -EFAULT;
 1355                 copy = nbytes;
 1356         }
 1357         return copy_to_user(mask, nodes_addr(*nodes), copy) ? -EFAULT : 0;
 1358 }
 1359 
 1360 SYSCALL_DEFINE6(mbind, unsigned long, start, unsigned long, len,
 1361                 unsigned long, mode, unsigned long __user *, nmask,
 1362                 unsigned long, maxnode, unsigned, flags)
 1363 {
 1364         nodemask_t nodes;
 1365         int err;
 1366         unsigned short mode_flags;
 1367 
 1368         mode_flags = mode & MPOL_MODE_FLAGS;
 1369         mode &= ~MPOL_MODE_FLAGS;
 1370         if (mode >= MPOL_MAX)
 1371                 return -EINVAL;
 1372         if ((mode_flags & MPOL_F_STATIC_NODES) &&
 1373             (mode_flags & MPOL_F_RELATIVE_NODES))
 1374                 return -EINVAL;
 1375         err = get_nodes(&nodes, nmask, maxnode);
 1376         if (err)
 1377                 return err;
 1378         return do_mbind(start, len, mode, mode_flags, &nodes, flags);
 1379 }
 1380 
 1381 /* Set the process memory policy */
 1382 SYSCALL_DEFINE3(set_mempolicy, int, mode, unsigned long __user *, nmask,
 1383                 unsigned long, maxnode)
 1384 {
 1385         int err;
 1386         nodemask_t nodes;
 1387         unsigned short flags;
 1388 
 1389         flags = mode & MPOL_MODE_FLAGS;
 1390         mode &= ~MPOL_MODE_FLAGS;
 1391         if ((unsigned int)mode >= MPOL_MAX)
 1392                 return -EINVAL;
 1393         if ((flags & MPOL_F_STATIC_NODES) && (flags & MPOL_F_RELATIVE_NODES))
 1394                 return -EINVAL;
 1395         err = get_nodes(&nodes, nmask, maxnode);
 1396         if (err)
 1397                 return err;
 1398         return do_set_mempolicy(mode, flags, &nodes);
 1399 }
 1400 
 1401 SYSCALL_DEFINE4(migrate_pages, pid_t, pid, unsigned long, maxnode,
 1402                 const unsigned long __user *, old_nodes,
 1403                 const unsigned long __user *, new_nodes)
 1404 {
 1405         const struct cred *cred = current_cred(), *tcred;
 1406         struct mm_struct *mm = NULL;
 1407         struct task_struct *task;
 1408         nodemask_t task_nodes;
 1409         int err;
 1410         nodemask_t *old;
 1411         nodemask_t *new;
 1412         NODEMASK_SCRATCH(scratch);
 1413 
 1414         if (!scratch)
 1415                 return -ENOMEM;
 1416 
 1417         old = &scratch->mask1;
 1418         new = &scratch->mask2;
 1419 
 1420         err = get_nodes(old, old_nodes, maxnode);
 1421         if (err)
 1422                 goto out;
 1423 
 1424         err = get_nodes(new, new_nodes, maxnode);
 1425         if (err)
 1426                 goto out;
 1427 
 1428         /* Find the mm_struct */
 1429         rcu_read_lock();
 1430         task = pid ? find_task_by_vpid(pid) : current;
 1431         if (!task) {
 1432                 rcu_read_unlock();
 1433                 err = -ESRCH;
 1434                 goto out;
 1435         }
 1436         get_task_struct(task);
 1437 
 1438         err = -EINVAL;
 1439 
 1440         /*
 1441          * Check if this process has the right to modify the specified
 1442          * process. The right exists if the process has administrative
 1443          * capabilities, superuser privileges or the same
 1444          * userid as the target process.
 1445          */
 1446         tcred = __task_cred(task);
 1447         if (!uid_eq(cred->euid, tcred->suid) && !uid_eq(cred->euid, tcred->uid) &&
 1448             !uid_eq(cred->uid,  tcred->suid) && !uid_eq(cred->uid,  tcred->uid) &&
 1449             !capable(CAP_SYS_NICE)) {
 1450                 rcu_read_unlock();
 1451                 err = -EPERM;
 1452                 goto out_put;
 1453         }
 1454         rcu_read_unlock();
 1455 
 1456         task_nodes = cpuset_mems_allowed(task);
 1457         /* Is the user allowed to access the target nodes? */
 1458         if (!nodes_subset(*new, task_nodes) && !capable(CAP_SYS_NICE)) {
 1459                 err = -EPERM;
 1460                 goto out_put;
 1461         }
 1462 
 1463         if (!nodes_subset(*new, node_states[N_MEMORY])) {
 1464                 err = -EINVAL;
 1465                 goto out_put;
 1466         }
 1467 
 1468         err = security_task_movememory(task);
 1469         if (err)
 1470                 goto out_put;
 1471 
 1472         mm = get_task_mm(task);
 1473         put_task_struct(task);
 1474 
 1475         if (!mm) {
 1476                 err = -EINVAL;
 1477                 goto out;
 1478         }
 1479 
 1480         err = do_migrate_pages(mm, old, new,
 1481                 capable(CAP_SYS_NICE) ? MPOL_MF_MOVE_ALL : MPOL_MF_MOVE);
 1482 
 1483         mmput(mm);
 1484 out:
 1485         NODEMASK_SCRATCH_FREE(scratch);
 1486 
 1487         return err;
 1488 
 1489 out_put:
 1490         put_task_struct(task);
 1491         goto out;
 1492 
 1493 }
 1494 
 1495 
 1496 /* Retrieve NUMA policy */
 1497 SYSCALL_DEFINE5(get_mempolicy, int __user *, policy,
 1498                 unsigned long __user *, nmask, unsigned long, maxnode,
 1499                 unsigned long, addr, unsigned long, flags)
 1500 {
 1501         int err;
 1502         int uninitialized_var(pval);
 1503         nodemask_t nodes;
 1504 
 1505         if (nmask != NULL && maxnode < MAX_NUMNODES)
 1506                 return -EINVAL;
 1507 
 1508         err = do_get_mempolicy(&pval, &nodes, addr, flags);
 1509 
 1510         if (err)
 1511                 return err;
 1512 
 1513         if (policy && put_user(pval, policy))
 1514                 return -EFAULT;
 1515 
 1516         if (nmask)
 1517                 err = copy_nodes_to_user(nmask, maxnode, &nodes);
 1518 
 1519         return err;
 1520 }
 1521 
 1522 #ifdef CONFIG_COMPAT
 1523 
 1524 asmlinkage long compat_sys_get_mempolicy(int __user *policy,
 1525                                      compat_ulong_t __user *nmask,
 1526                                      compat_ulong_t maxnode,
 1527                                      compat_ulong_t addr, compat_ulong_t flags)
 1528 {
 1529         long err;
 1530         unsigned long __user *nm = NULL;
 1531         unsigned long nr_bits, alloc_size;
 1532         DECLARE_BITMAP(bm, MAX_NUMNODES);
 1533 
 1534         nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES);
 1535         alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8;
 1536 
 1537         if (nmask)
 1538                 nm = compat_alloc_user_space(alloc_size);
 1539 
 1540         err = sys_get_mempolicy(policy, nm, nr_bits+1, addr, flags);
 1541 
 1542         if (!err && nmask) {
 1543                 unsigned long copy_size;
 1544                 copy_size = min_t(unsigned long, sizeof(bm), alloc_size);
 1545                 err = copy_from_user(bm, nm, copy_size);
 1546                 /* ensure entire bitmap is zeroed */
 1547                 err |= clear_user(nmask, ALIGN(maxnode-1, 8) / 8);
 1548                 err |= compat_put_bitmap(nmask, bm, nr_bits);
 1549         }
 1550 
 1551         return err;
 1552 }
 1553 
 1554 asmlinkage long compat_sys_set_mempolicy(int mode, compat_ulong_t __user *nmask,
 1555                                      compat_ulong_t maxnode)
 1556 {
 1557         long err = 0;
 1558         unsigned long __user *nm = NULL;
 1559         unsigned long nr_bits, alloc_size;
 1560         DECLARE_BITMAP(bm, MAX_NUMNODES);
 1561 
 1562         nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES);
 1563         alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8;
 1564 
 1565         if (nmask) {
 1566                 err = compat_get_bitmap(bm, nmask, nr_bits);
 1567                 nm = compat_alloc_user_space(alloc_size);
 1568                 err |= copy_to_user(nm, bm, alloc_size);
 1569         }
 1570 
 1571         if (err)
 1572                 return -EFAULT;
 1573 
 1574         return sys_set_mempolicy(mode, nm, nr_bits+1);
 1575 }
 1576 
 1577 asmlinkage long compat_sys_mbind(compat_ulong_t start, compat_ulong_t len,
 1578                              compat_ulong_t mode, compat_ulong_t __user *nmask,
 1579                              compat_ulong_t maxnode, compat_ulong_t flags)
 1580 {
 1581         long err = 0;
 1582         unsigned long __user *nm = NULL;
 1583         unsigned long nr_bits, alloc_size;
 1584         nodemask_t bm;
 1585 
 1586         nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES);
 1587         alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8;
 1588 
 1589         if (nmask) {
 1590                 err = compat_get_bitmap(nodes_addr(bm), nmask, nr_bits);
 1591                 nm = compat_alloc_user_space(alloc_size);
 1592                 err |= copy_to_user(nm, nodes_addr(bm), alloc_size);
 1593         }
 1594 
 1595         if (err)
 1596                 return -EFAULT;
 1597 
 1598         return sys_mbind(start, len, mode, nm, nr_bits+1, flags);
 1599 }
 1600 
 1601 #endif
 1602 
 1603 /*
 1604  * get_vma_policy(@task, @vma, @addr)
 1605  * @task - task for fallback if vma policy == default
 1606  * @vma   - virtual memory area whose policy is sought
 1607  * @addr  - address in @vma for shared policy lookup
 1608  *
 1609  * Returns effective policy for a VMA at specified address.
 1610  * Falls back to @task or system default policy, as necessary.
 1611  * Current or other task's task mempolicy and non-shared vma policies must be
 1612  * protected by task_lock(task) by the caller.
 1613  * Shared policies [those marked as MPOL_F_SHARED] require an extra reference
 1614  * count--added by the get_policy() vm_op, as appropriate--to protect against
 1615  * freeing by another task.  It is the caller's responsibility to free the
 1616  * extra reference for shared policies.
 1617  */
 1618 struct mempolicy *get_vma_policy(struct task_struct *task,
 1619                 struct vm_area_struct *vma, unsigned long addr)
 1620 {
 1621         struct mempolicy *pol = get_task_policy(task);
 1622 
 1623         if (vma) {
 1624                 if (vma->vm_ops && vma->vm_ops->get_policy) {
 1625                         struct mempolicy *vpol = vma->vm_ops->get_policy(vma,
 1626                                                                         addr);
 1627                         if (vpol)
 1628                                 pol = vpol;
 1629                 } else if (vma->vm_policy) {
 1630                         pol = vma->vm_policy;
 1631 
 1632                         /*
 1633                          * shmem_alloc_page() passes MPOL_F_SHARED policy with
 1634                          * a pseudo vma whose vma->vm_ops=NULL. Take a reference
 1635                          * count on these policies which will be dropped by
 1636                          * mpol_cond_put() later
 1637                          */
 1638                         if (mpol_needs_cond_ref(pol))
 1639                                 mpol_get(pol);
 1640                 }
 1641         }
 1642         if (!pol)
 1643                 pol = &default_policy;
 1644         return pol;
 1645 }
 1646 
 1647 /*
 1648  * Return a nodemask representing a mempolicy for filtering nodes for
 1649  * page allocation
 1650  */
 1651 static nodemask_t *policy_nodemask(gfp_t gfp, struct mempolicy *policy)
 1652 {
 1653         /* Lower zones don't get a nodemask applied for MPOL_BIND */
 1654         if (unlikely(policy->mode == MPOL_BIND) &&
 1655                         gfp_zone(gfp) >= policy_zone &&
 1656                         cpuset_nodemask_valid_mems_allowed(&policy->v.nodes))
 1657                 return &policy->v.nodes;
 1658 
 1659         return NULL;
 1660 }
 1661 
 1662 /* Return a zonelist indicated by gfp for node representing a mempolicy */
 1663 static struct zonelist *policy_zonelist(gfp_t gfp, struct mempolicy *policy,
 1664         int nd)
 1665 {
 1666         switch (policy->mode) {
 1667         case MPOL_PREFERRED:
 1668                 if (!(policy->flags & MPOL_F_LOCAL))
 1669                         nd = policy->v.preferred_node;
 1670                 break;
 1671         case MPOL_BIND:
 1672                 /*
 1673                  * Normally, MPOL_BIND allocations are node-local within the
 1674                  * allowed nodemask.  However, if __GFP_THISNODE is set and the
 1675                  * current node isn't part of the mask, we use the zonelist for
 1676                  * the first node in the mask instead.
 1677                  */
 1678                 if (unlikely(gfp & __GFP_THISNODE) &&
 1679                                 unlikely(!node_isset(nd, policy->v.nodes)))
 1680                         nd = first_node(policy->v.nodes);
 1681                 break;
 1682         default:
 1683                 BUG();
 1684         }
 1685         return node_zonelist(nd, gfp);
 1686 }
 1687 
 1688 /* Do dynamic interleaving for a process */
 1689 static unsigned interleave_nodes(struct mempolicy *policy)
 1690 {
 1691         unsigned nid, next;
 1692         struct task_struct *me = current;
 1693 
 1694         nid = me->il_next;
 1695         next = next_node(nid, policy->v.nodes);
 1696         if (next >= MAX_NUMNODES)
 1697                 next = first_node(policy->v.nodes);
 1698         if (next < MAX_NUMNODES)
 1699                 me->il_next = next;
 1700         return nid;
 1701 }
 1702 
 1703 /*
 1704  * Depending on the memory policy provide a node from which to allocate the
 1705  * next slab entry.
 1706  * @policy must be protected by freeing by the caller.  If @policy is
 1707  * the current task's mempolicy, this protection is implicit, as only the
 1708  * task can change it's policy.  The system default policy requires no
 1709  * such protection.
 1710  */
 1711 unsigned slab_node(void)
 1712 {
 1713         struct mempolicy *policy;
 1714 
 1715         if (in_interrupt())
 1716                 return numa_node_id();
 1717 
 1718         policy = current->mempolicy;
 1719         if (!policy || policy->flags & MPOL_F_LOCAL)
 1720                 return numa_node_id();
 1721 
 1722         switch (policy->mode) {
 1723         case MPOL_PREFERRED:
 1724                 /*
 1725                  * handled MPOL_F_LOCAL above
 1726                  */
 1727                 return policy->v.preferred_node;
 1728 
 1729         case MPOL_INTERLEAVE:
 1730                 return interleave_nodes(policy);
 1731 
 1732         case MPOL_BIND: {
 1733                 /*
 1734                  * Follow bind policy behavior and start allocation at the
 1735                  * first node.
 1736                  */
 1737                 struct zonelist *zonelist;
 1738                 struct zone *zone;
 1739                 enum zone_type highest_zoneidx = gfp_zone(GFP_KERNEL);
 1740                 zonelist = &NODE_DATA(numa_node_id())->node_zonelists[0];
 1741                 (void)first_zones_zonelist(zonelist, highest_zoneidx,
 1742                                                         &policy->v.nodes,
 1743                                                         &zone);
 1744                 return zone ? zone->node : numa_node_id();
 1745         }
 1746 
 1747         default:
 1748                 BUG();
 1749         }
 1750 }
 1751 
 1752 /* Do static interleaving for a VMA with known offset. */
 1753 static unsigned offset_il_node(struct mempolicy *pol,
 1754                 struct vm_area_struct *vma, unsigned long off)
 1755 {
 1756         unsigned nnodes = nodes_weight(pol->v.nodes);
 1757         unsigned target;
 1758         int c;
 1759         int nid = -1;
 1760 
 1761         if (!nnodes)
 1762                 return numa_node_id();
 1763         target = (unsigned int)off % nnodes;
 1764         c = 0;
 1765         do {
 1766                 nid = next_node(nid, pol->v.nodes);
 1767                 c++;
 1768         } while (c <= target);
 1769         return nid;
 1770 }
 1771 
 1772 /* Determine a node number for interleave */
 1773 static inline unsigned interleave_nid(struct mempolicy *pol,
 1774                  struct vm_area_struct *vma, unsigned long addr, int shift)
 1775 {
 1776         if (vma) {
 1777                 unsigned long off;
 1778 
 1779                 /*
 1780                  * for small pages, there is no difference between
 1781                  * shift and PAGE_SHIFT, so the bit-shift is safe.
 1782                  * for huge pages, since vm_pgoff is in units of small
 1783                  * pages, we need to shift off the always 0 bits to get
 1784                  * a useful offset.
 1785                  */
 1786                 BUG_ON(shift < PAGE_SHIFT);
 1787                 off = vma->vm_pgoff >> (shift - PAGE_SHIFT);
 1788                 off += (addr - vma->vm_start) >> shift;
 1789                 return offset_il_node(pol, vma, off);
 1790         } else
 1791                 return interleave_nodes(pol);
 1792 }
 1793 
 1794 /*
 1795  * Return the bit number of a random bit set in the nodemask.
 1796  * (returns -1 if nodemask is empty)
 1797  */
 1798 int node_random(const nodemask_t *maskp)
 1799 {
 1800         int w, bit = -1;
 1801 
 1802         w = nodes_weight(*maskp);
 1803         if (w)
 1804                 bit = bitmap_ord_to_pos(maskp->bits,
 1805                         get_random_int() % w, MAX_NUMNODES);
 1806         return bit;
 1807 }
 1808 
 1809 #ifdef CONFIG_HUGETLBFS
 1810 /*
 1811  * huge_zonelist(@vma, @addr, @gfp_flags, @mpol)
 1812  * @vma = virtual memory area whose policy is sought
 1813  * @addr = address in @vma for shared policy lookup and interleave policy
 1814  * @gfp_flags = for requested zone
 1815  * @mpol = pointer to mempolicy pointer for reference counted mempolicy
 1816  * @nodemask = pointer to nodemask pointer for MPOL_BIND nodemask
 1817  *
 1818  * Returns a zonelist suitable for a huge page allocation and a pointer
 1819  * to the struct mempolicy for conditional unref after allocation.
 1820  * If the effective policy is 'BIND, returns a pointer to the mempolicy's
 1821  * @nodemask for filtering the zonelist.
 1822  *
 1823  * Must be protected by get_mems_allowed()
 1824  */
 1825 struct zonelist *huge_zonelist(struct vm_area_struct *vma, unsigned long addr,
 1826                                 gfp_t gfp_flags, struct mempolicy **mpol,
 1827                                 nodemask_t **nodemask)
 1828 {
 1829         struct zonelist *zl;
 1830 
 1831         *mpol = get_vma_policy(current, vma, addr);
 1832         *nodemask = NULL;       /* assume !MPOL_BIND */
 1833 
 1834         if (unlikely((*mpol)->mode == MPOL_INTERLEAVE)) {
 1835                 zl = node_zonelist(interleave_nid(*mpol, vma, addr,
 1836                                 huge_page_shift(hstate_vma(vma))), gfp_flags);
 1837         } else {
 1838                 zl = policy_zonelist(gfp_flags, *mpol, numa_node_id());
 1839                 if ((*mpol)->mode == MPOL_BIND)
 1840                         *nodemask = &(*mpol)->v.nodes;
 1841         }
 1842         return zl;
 1843 }
 1844 
 1845 /*
 1846  * init_nodemask_of_mempolicy
 1847  *
 1848  * If the current task's mempolicy is "default" [NULL], return 'false'
 1849  * to indicate default policy.  Otherwise, extract the policy nodemask
 1850  * for 'bind' or 'interleave' policy into the argument nodemask, or
 1851  * initialize the argument nodemask to contain the single node for
 1852  * 'preferred' or 'local' policy and return 'true' to indicate presence
 1853  * of non-default mempolicy.
 1854  *
 1855  * We don't bother with reference counting the mempolicy [mpol_get/put]
 1856  * because the current task is examining it's own mempolicy and a task's
 1857  * mempolicy is only ever changed by the task itself.
 1858  *
 1859  * N.B., it is the caller's responsibility to free a returned nodemask.
 1860  */
 1861 bool init_nodemask_of_mempolicy(nodemask_t *mask)
 1862 {
 1863         struct mempolicy *mempolicy;
 1864         int nid;
 1865 
 1866         if (!(mask && current->mempolicy))
 1867                 return false;
 1868 
 1869         task_lock(current);
 1870         mempolicy = current->mempolicy;
 1871         switch (mempolicy->mode) {
 1872         case MPOL_PREFERRED:
 1873                 if (mempolicy->flags & MPOL_F_LOCAL)
 1874                         nid = numa_node_id();
 1875                 else
 1876                         nid = mempolicy->v.preferred_node;
 1877                 init_nodemask_of_node(mask, nid);
 1878                 break;
 1879 
 1880         case MPOL_BIND:
 1881                 /* Fall through */
 1882         case MPOL_INTERLEAVE:
 1883                 *mask =  mempolicy->v.nodes;
 1884                 break;
 1885 
 1886         default:
 1887                 BUG();
 1888         }
 1889         task_unlock(current);
 1890 
 1891         return true;
 1892 }
 1893 #endif
 1894 
 1895 /*
 1896  * mempolicy_nodemask_intersects
 1897  *
 1898  * If tsk's mempolicy is "default" [NULL], return 'true' to indicate default
 1899  * policy.  Otherwise, check for intersection between mask and the policy
 1900  * nodemask for 'bind' or 'interleave' policy.  For 'perferred' or 'local'
 1901  * policy, always return true since it may allocate elsewhere on fallback.
 1902  *
 1903  * Takes task_lock(tsk) to prevent freeing of its mempolicy.
 1904  */
 1905 bool mempolicy_nodemask_intersects(struct task_struct *tsk,
 1906                                         const nodemask_t *mask)
 1907 {
 1908         struct mempolicy *mempolicy;
 1909         bool ret = true;
 1910 
 1911         if (!mask)
 1912                 return ret;
 1913         task_lock(tsk);
 1914         mempolicy = tsk->mempolicy;
 1915         if (!mempolicy)
 1916                 goto out;
 1917 
 1918         switch (mempolicy->mode) {
 1919         case MPOL_PREFERRED:
 1920                 /*
 1921                  * MPOL_PREFERRED and MPOL_F_LOCAL are only preferred nodes to
 1922                  * allocate from, they may fallback to other nodes when oom.
 1923                  * Thus, it's possible for tsk to have allocated memory from
 1924                  * nodes in mask.
 1925                  */
 1926                 break;
 1927         case MPOL_BIND:
 1928         case MPOL_INTERLEAVE:
 1929                 ret = nodes_intersects(mempolicy->v.nodes, *mask);
 1930                 break;
 1931         default:
 1932                 BUG();
 1933         }
 1934 out:
 1935         task_unlock(tsk);
 1936         return ret;
 1937 }
 1938 
 1939 /* Allocate a page in interleaved policy.
 1940    Own path because it needs to do special accounting. */
 1941 static struct page *alloc_page_interleave(gfp_t gfp, unsigned order,
 1942                                         unsigned nid)
 1943 {
 1944         struct zonelist *zl;
 1945         struct page *page;
 1946 
 1947         zl = node_zonelist(nid, gfp);
 1948         page = __alloc_pages(gfp, order, zl);
 1949         if (page && page_zone(page) == zonelist_zone(&zl->_zonerefs[0]))
 1950                 inc_zone_page_state(page, NUMA_INTERLEAVE_HIT);
 1951         return page;
 1952 }
 1953 
 1954 /**
 1955  *      alloc_pages_vma - Allocate a page for a VMA.
 1956  *
 1957  *      @gfp:
 1958  *      %GFP_USER    user allocation.
 1959  *      %GFP_KERNEL  kernel allocations,
 1960  *      %GFP_HIGHMEM highmem/user allocations,
 1961  *      %GFP_FS      allocation should not call back into a file system.
 1962  *      %GFP_ATOMIC  don't sleep.
 1963  *
 1964  *      @order:Order of the GFP allocation.
 1965  *      @vma:  Pointer to VMA or NULL if not available.
 1966  *      @addr: Virtual Address of the allocation. Must be inside the VMA.
 1967  *
 1968  *      This function allocates a page from the kernel page pool and applies
 1969  *      a NUMA policy associated with the VMA or the current process.
 1970  *      When VMA is not NULL caller must hold down_read on the mmap_sem of the
 1971  *      mm_struct of the VMA to prevent it from going away. Should be used for
 1972  *      all allocations for pages that will be mapped into
 1973  *      user space. Returns NULL when no page can be allocated.
 1974  *
 1975  *      Should be called with the mm_sem of the vma hold.
 1976  */
 1977 struct page *
 1978 alloc_pages_vma(gfp_t gfp, int order, struct vm_area_struct *vma,
 1979                 unsigned long addr, int node)
 1980 {
 1981         struct mempolicy *pol;
 1982         struct page *page;
 1983         unsigned int cpuset_mems_cookie;
 1984 
 1985 retry_cpuset:
 1986         pol = get_vma_policy(current, vma, addr);
 1987         cpuset_mems_cookie = get_mems_allowed();
 1988 
 1989         if (unlikely(pol->mode == MPOL_INTERLEAVE)) {
 1990                 unsigned nid;
 1991 
 1992                 nid = interleave_nid(pol, vma, addr, PAGE_SHIFT + order);
 1993                 mpol_cond_put(pol);
 1994                 page = alloc_page_interleave(gfp, order, nid);
 1995                 if (unlikely(!put_mems_allowed(cpuset_mems_cookie) && !page))
 1996                         goto retry_cpuset;
 1997 
 1998                 return page;
 1999         }
 2000         page = __alloc_pages_nodemask(gfp, order,
 2001                                       policy_zonelist(gfp, pol, node),
 2002                                       policy_nodemask(gfp, pol));
 2003         if (unlikely(mpol_needs_cond_ref(pol)))
 2004                 __mpol_put(pol);
 2005         if (unlikely(!put_mems_allowed(cpuset_mems_cookie) && !page))
 2006                 goto retry_cpuset;
 2007         return page;
 2008 }
 2009 
 2010 /**
 2011  *      alloc_pages_current - Allocate pages.
 2012  *
 2013  *      @gfp:
 2014  *              %GFP_USER   user allocation,
 2015  *              %GFP_KERNEL kernel allocation,
 2016  *              %GFP_HIGHMEM highmem allocation,
 2017  *              %GFP_FS     don't call back into a file system.
 2018  *              %GFP_ATOMIC don't sleep.
 2019  *      @order: Power of two of allocation size in pages. 0 is a single page.
 2020  *
 2021  *      Allocate a page from the kernel page pool.  When not in
 2022  *      interrupt context and apply the current process NUMA policy.
 2023  *      Returns NULL when no page can be allocated.
 2024  *
 2025  *      Don't call cpuset_update_task_memory_state() unless
 2026  *      1) it's ok to take cpuset_sem (can WAIT), and
 2027  *      2) allocating for current task (not interrupt).
 2028  */
 2029 struct page *alloc_pages_current(gfp_t gfp, unsigned order)
 2030 {
 2031         struct mempolicy *pol = get_task_policy(current);
 2032         struct page *page;
 2033         unsigned int cpuset_mems_cookie;
 2034 
 2035         if (!pol || in_interrupt() || (gfp & __GFP_THISNODE))
 2036                 pol = &default_policy;
 2037 
 2038 retry_cpuset:
 2039         cpuset_mems_cookie = get_mems_allowed();
 2040 
 2041         /*
 2042          * No reference counting needed for current->mempolicy
 2043          * nor system default_policy
 2044          */
 2045         if (pol->mode == MPOL_INTERLEAVE)
 2046                 page = alloc_page_interleave(gfp, order, interleave_nodes(pol));
 2047         else
 2048                 page = __alloc_pages_nodemask(gfp, order,
 2049                                 policy_zonelist(gfp, pol, numa_node_id()),
 2050                                 policy_nodemask(gfp, pol));
 2051 
 2052         if (unlikely(!put_mems_allowed(cpuset_mems_cookie) && !page))
 2053                 goto retry_cpuset;
 2054 
 2055         return page;
 2056 }
 2057 EXPORT_SYMBOL(alloc_pages_current);
 2058 
 2059 /*
 2060  * If mpol_dup() sees current->cpuset == cpuset_being_rebound, then it
 2061  * rebinds the mempolicy its copying by calling mpol_rebind_policy()
 2062  * with the mems_allowed returned by cpuset_mems_allowed().  This
 2063  * keeps mempolicies cpuset relative after its cpuset moves.  See
 2064  * further kernel/cpuset.c update_nodemask().
 2065  *
 2066  * current's mempolicy may be rebinded by the other task(the task that changes
 2067  * cpuset's mems), so we needn't do rebind work for current task.
 2068  */
 2069 
 2070 /* Slow path of a mempolicy duplicate */
 2071 struct mempolicy *__mpol_dup(struct mempolicy *old)
 2072 {
 2073         struct mempolicy *new = kmem_cache_alloc(policy_cache, GFP_KERNEL);
 2074 
 2075         if (!new)
 2076                 return ERR_PTR(-ENOMEM);
 2077 
 2078         /* task's mempolicy is protected by alloc_lock */
 2079         if (old == current->mempolicy) {
 2080                 task_lock(current);
 2081                 *new = *old;
 2082                 task_unlock(current);
 2083         } else
 2084                 *new = *old;
 2085 
 2086         rcu_read_lock();
 2087         if (current_cpuset_is_being_rebound()) {
 2088                 nodemask_t mems = cpuset_mems_allowed(current);
 2089                 if (new->flags & MPOL_F_REBINDING)
 2090                         mpol_rebind_policy(new, &mems, MPOL_REBIND_STEP2);
 2091                 else
 2092                         mpol_rebind_policy(new, &mems, MPOL_REBIND_ONCE);
 2093         }
 2094         rcu_read_unlock();
 2095         atomic_set(&new->refcnt, 1);
 2096         return new;
 2097 }
 2098 
 2099 /* Slow path of a mempolicy comparison */
 2100 bool __mpol_equal(struct mempolicy *a, struct mempolicy *b)
 2101 {
 2102         if (!a || !b)
 2103                 return false;
 2104         if (a->mode != b->mode)
 2105                 return false;
 2106         if (a->flags != b->flags)
 2107                 return false;
 2108         if (mpol_store_user_nodemask(a))
 2109                 if (!nodes_equal(a->w.user_nodemask, b->w.user_nodemask))
 2110                         return false;
 2111 
 2112         switch (a->mode) {
 2113         case MPOL_BIND:
 2114                 /* Fall through */
 2115         case MPOL_INTERLEAVE:
 2116                 return !!nodes_equal(a->v.nodes, b->v.nodes);
 2117         case MPOL_PREFERRED:
 2118                 return a->v.preferred_node == b->v.preferred_node;
 2119         default:
 2120                 BUG();
 2121                 return false;
 2122         }
 2123 }
 2124 
 2125 /*
 2126  * Shared memory backing store policy support.
 2127  *
 2128  * Remember policies even when nobody has shared memory mapped.
 2129  * The policies are kept in Red-Black tree linked from the inode.
 2130  * They are protected by the sp->lock spinlock, which should be held
 2131  * for any accesses to the tree.
 2132  */
 2133 
 2134 /* lookup first element intersecting start-end */
 2135 /* Caller holds sp->lock */
 2136 static struct sp_node *
 2137 sp_lookup(struct shared_policy *sp, unsigned long start, unsigned long end)
 2138 {
 2139         struct rb_node *n = sp->root.rb_node;
 2140 
 2141         while (n) {
 2142                 struct sp_node *p = rb_entry(n, struct sp_node, nd);
 2143 
 2144                 if (start >= p->end)
 2145                         n = n->rb_right;
 2146                 else if (end <= p->start)
 2147                         n = n->rb_left;
 2148                 else
 2149                         break;
 2150         }
 2151         if (!n)
 2152                 return NULL;
 2153         for (;;) {
 2154                 struct sp_node *w = NULL;
 2155                 struct rb_node *prev = rb_prev(n);
 2156                 if (!prev)
 2157                         break;
 2158                 w = rb_entry(prev, struct sp_node, nd);
 2159                 if (w->end <= start)
 2160                         break;
 2161                 n = prev;
 2162         }
 2163         return rb_entry(n, struct sp_node, nd);
 2164 }
 2165 
 2166 /* Insert a new shared policy into the list. */
 2167 /* Caller holds sp->lock */
 2168 static void sp_insert(struct shared_policy *sp, struct sp_node *new)
 2169 {
 2170         struct rb_node **p = &sp->root.rb_node;
 2171         struct rb_node *parent = NULL;
 2172         struct sp_node *nd;
 2173 
 2174         while (*p) {
 2175                 parent = *p;
 2176                 nd = rb_entry(parent, struct sp_node, nd);
 2177                 if (new->start < nd->start)
 2178                         p = &(*p)->rb_left;
 2179                 else if (new->end > nd->end)
 2180                         p = &(*p)->rb_right;
 2181                 else
 2182                         BUG();
 2183         }
 2184         rb_link_node(&new->nd, parent, p);
 2185         rb_insert_color(&new->nd, &sp->root);
 2186         pr_debug("inserting %lx-%lx: %d\n", new->start, new->end,
 2187                  new->policy ? new->policy->mode : 0);
 2188 }
 2189 
 2190 /* Find shared policy intersecting idx */
 2191 struct mempolicy *
 2192 mpol_shared_policy_lookup(struct shared_policy *sp, unsigned long idx)
 2193 {
 2194         struct mempolicy *pol = NULL;
 2195         struct sp_node *sn;
 2196 
 2197         if (!sp->root.rb_node)
 2198                 return NULL;
 2199         spin_lock(&sp->lock);
 2200         sn = sp_lookup(sp, idx, idx+1);
 2201         if (sn) {
 2202                 mpol_get(sn->policy);
 2203                 pol = sn->policy;
 2204         }
 2205         spin_unlock(&sp->lock);
 2206         return pol;
 2207 }
 2208 
 2209 static void sp_free(struct sp_node *n)
 2210 {
 2211         mpol_put(n->policy);
 2212         kmem_cache_free(sn_cache, n);
 2213 }
 2214 
 2215 /**
 2216  * mpol_misplaced - check whether current page node is valid in policy
 2217  *
 2218  * @page   - page to be checked
 2219  * @vma    - vm area where page mapped
 2220  * @addr   - virtual address where page mapped
 2221  *
 2222  * Lookup current policy node id for vma,addr and "compare to" page's
 2223  * node id.
 2224  *
 2225  * Returns:
 2226  *      -1      - not misplaced, page is in the right node
 2227  *      node    - node id where the page should be
 2228  *
 2229  * Policy determination "mimics" alloc_page_vma().
 2230  * Called from fault path where we know the vma and faulting address.
 2231  */
 2232 int mpol_misplaced(struct page *page, struct vm_area_struct *vma, unsigned long addr)
 2233 {
 2234         struct mempolicy *pol;
 2235         struct zone *zone;
 2236         int curnid = page_to_nid(page);
 2237         unsigned long pgoff;
 2238         int polnid = -1;
 2239         int ret = -1;
 2240 
 2241         BUG_ON(!vma);
 2242 
 2243         pol = get_vma_policy(current, vma, addr);
 2244         if (!(pol->flags & MPOL_F_MOF))
 2245                 goto out;
 2246 
 2247         switch (pol->mode) {
 2248         case MPOL_INTERLEAVE:
 2249                 BUG_ON(addr >= vma->vm_end);
 2250                 BUG_ON(addr < vma->vm_start);
 2251 
 2252                 pgoff = vma->vm_pgoff;
 2253                 pgoff += (addr - vma->vm_start) >> PAGE_SHIFT;
 2254                 polnid = offset_il_node(pol, vma, pgoff);
 2255                 break;
 2256 
 2257         case MPOL_PREFERRED:
 2258                 if (pol->flags & MPOL_F_LOCAL)
 2259                         polnid = numa_node_id();
 2260                 else
 2261                         polnid = pol->v.preferred_node;
 2262                 break;
 2263 
 2264         case MPOL_BIND:
 2265                 /*
 2266                  * allows binding to multiple nodes.
 2267                  * use current page if in policy nodemask,
 2268                  * else select nearest allowed node, if any.
 2269                  * If no allowed nodes, use current [!misplaced].
 2270                  */
 2271                 if (node_isset(curnid, pol->v.nodes))
 2272                         goto out;
 2273                 (void)first_zones_zonelist(
 2274                                 node_zonelist(numa_node_id(), GFP_HIGHUSER),
 2275                                 gfp_zone(GFP_HIGHUSER),
 2276                                 &pol->v.nodes, &zone);
 2277                 polnid = zone->node;
 2278                 break;
 2279 
 2280         default:
 2281                 BUG();
 2282         }
 2283 
 2284         /* Migrate the page towards the node whose CPU is referencing it */
 2285         if (pol->flags & MPOL_F_MORON) {
 2286                 int last_nid;
 2287 
 2288                 polnid = numa_node_id();
 2289 
 2290                 /*
 2291                  * Multi-stage node selection is used in conjunction
 2292                  * with a periodic migration fault to build a temporal
 2293                  * task<->page relation. By using a two-stage filter we
 2294                  * remove short/unlikely relations.
 2295                  *
 2296                  * Using P(p) ~ n_p / n_t as per frequentist
 2297                  * probability, we can equate a task's usage of a
 2298                  * particular page (n_p) per total usage of this
 2299                  * page (n_t) (in a given time-span) to a probability.
 2300                  *
 2301                  * Our periodic faults will sample this probability and
 2302                  * getting the same result twice in a row, given these
 2303                  * samples are fully independent, is then given by
 2304                  * P(n)^2, provided our sample period is sufficiently
 2305                  * short compared to the usage pattern.
 2306                  *
 2307                  * This quadric squishes small probabilities, making
 2308                  * it less likely we act on an unlikely task<->page
 2309                  * relation.
 2310                  */
 2311                 last_nid = page_xchg_last_nid(page, polnid);
 2312                 if (last_nid != polnid)
 2313                         goto out;
 2314         }
 2315 
 2316         if (curnid != polnid)
 2317                 ret = polnid;
 2318 out:
 2319         mpol_cond_put(pol);
 2320 
 2321         return ret;
 2322 }
 2323 
 2324 static void sp_delete(struct shared_policy *sp, struct sp_node *n)
 2325 {
 2326         pr_debug("deleting %lx-l%lx\n", n->start, n->end);
 2327         rb_erase(&n->nd, &sp->root);
 2328         sp_free(n);
 2329 }
 2330 
 2331 static void sp_node_init(struct sp_node *node, unsigned long start,
 2332                         unsigned long end, struct mempolicy *pol)
 2333 {
 2334         node->start = start;
 2335         node->end = end;
 2336         node->policy = pol;
 2337 }
 2338 
 2339 static struct sp_node *sp_alloc(unsigned long start, unsigned long end,
 2340                                 struct mempolicy *pol)
 2341 {
 2342         struct sp_node *n;
 2343         struct mempolicy *newpol;
 2344 
 2345         n = kmem_cache_alloc(sn_cache, GFP_KERNEL);
 2346         if (!n)
 2347                 return NULL;
 2348 
 2349         newpol = mpol_dup(pol);
 2350         if (IS_ERR(newpol)) {
 2351                 kmem_cache_free(sn_cache, n);
 2352                 return NULL;
 2353         }
 2354         newpol->flags |= MPOL_F_SHARED;
 2355         sp_node_init(n, start, end, newpol);
 2356 
 2357         return n;
 2358 }
 2359 
 2360 /* Replace a policy range. */
 2361 static int shared_policy_replace(struct shared_policy *sp, unsigned long start,
 2362                                  unsigned long end, struct sp_node *new)
 2363 {
 2364         struct sp_node *n;
 2365         struct sp_node *n_new = NULL;
 2366         struct mempolicy *mpol_new = NULL;
 2367         int ret = 0;
 2368 
 2369 restart:
 2370         spin_lock(&sp->lock);
 2371         n = sp_lookup(sp, start, end);
 2372         /* Take care of old policies in the same range. */
 2373         while (n && n->start < end) {
 2374                 struct rb_node *next = rb_next(&n->nd);
 2375                 if (n->start >= start) {
 2376                         if (n->end <= end)
 2377                                 sp_delete(sp, n);
 2378                         else
 2379                                 n->start = end;
 2380                 } else {
 2381                         /* Old policy spanning whole new range. */
 2382                         if (n->end > end) {
 2383                                 if (!n_new)
 2384                                         goto alloc_new;
 2385 
 2386                                 *mpol_new = *n->policy;
 2387                                 atomic_set(&mpol_new->refcnt, 1);
 2388                                 sp_node_init(n_new, n->end, end, mpol_new);
 2389                                 sp_insert(sp, n_new);
 2390                                 n->end = start;
 2391                                 n_new = NULL;
 2392                                 mpol_new = NULL;
 2393                                 break;
 2394                         } else
 2395                                 n->end = start;
 2396                 }
 2397                 if (!next)
 2398                         break;
 2399                 n = rb_entry(next, struct sp_node, nd);
 2400         }
 2401         if (new)
 2402                 sp_insert(sp, new);
 2403         spin_unlock(&sp->lock);
 2404         ret = 0;
 2405 
 2406 err_out:
 2407         if (mpol_new)
 2408                 mpol_put(mpol_new);
 2409         if (n_new)
 2410                 kmem_cache_free(sn_cache, n_new);
 2411 
 2412         return ret;
 2413 
 2414 alloc_new:
 2415         spin_unlock(&sp->lock);
 2416         ret = -ENOMEM;
 2417         n_new = kmem_cache_alloc(sn_cache, GFP_KERNEL);
 2418         if (!n_new)
 2419                 goto err_out;
 2420         mpol_new = kmem_cache_alloc(policy_cache, GFP_KERNEL);
 2421         if (!mpol_new)
 2422                 goto err_out;
 2423         goto restart;
 2424 }
 2425 
 2426 /**
 2427  * mpol_shared_policy_init - initialize shared policy for inode
 2428  * @sp: pointer to inode shared policy
 2429  * @mpol:  struct mempolicy to install
 2430  *
 2431  * Install non-NULL @mpol in inode's shared policy rb-tree.
 2432  * On entry, the current task has a reference on a non-NULL @mpol.
 2433  * This must be released on exit.
 2434  * This is called at get_inode() calls and we can use GFP_KERNEL.
 2435  */
 2436 void mpol_shared_policy_init(struct shared_policy *sp, struct mempolicy *mpol)
 2437 {
 2438         int ret;
 2439 
 2440         sp->root = RB_ROOT;             /* empty tree == default mempolicy */
 2441         spin_lock_init(&sp->lock);
 2442 
 2443         if (mpol) {
 2444                 struct vm_area_struct pvma;
 2445                 struct mempolicy *new;
 2446                 NODEMASK_SCRATCH(scratch);
 2447 
 2448                 if (!scratch)
 2449                         goto put_mpol;
 2450                 /* contextualize the tmpfs mount point mempolicy */
 2451                 new = mpol_new(mpol->mode, mpol->flags, &mpol->w.user_nodemask);
 2452                 if (IS_ERR(new))
 2453                         goto free_scratch; /* no valid nodemask intersection */
 2454 
 2455                 task_lock(current);
 2456                 ret = mpol_set_nodemask(new, &mpol->w.user_nodemask, scratch);
 2457                 task_unlock(current);
 2458                 if (ret)
 2459                         goto put_new;
 2460 
 2461                 /* Create pseudo-vma that contains just the policy */
 2462                 memset(&pvma, 0, sizeof(struct vm_area_struct));
 2463                 pvma.vm_end = TASK_SIZE;        /* policy covers entire file */
 2464                 mpol_set_shared_policy(sp, &pvma, new); /* adds ref */
 2465 
 2466 put_new:
 2467                 mpol_put(new);                  /* drop initial ref */
 2468 free_scratch:
 2469                 NODEMASK_SCRATCH_FREE(scratch);
 2470 put_mpol:
 2471                 mpol_put(mpol); /* drop our incoming ref on sb mpol */
 2472         }
 2473 }
 2474 
 2475 int mpol_set_shared_policy(struct shared_policy *info,
 2476                         struct vm_area_struct *vma, struct mempolicy *npol)
 2477 {
 2478         int err;
 2479         struct sp_node *new = NULL;
 2480         unsigned long sz = vma_pages(vma);
 2481 
 2482         pr_debug("set_shared_policy %lx sz %lu %d %d %lx\n",
 2483                  vma->vm_pgoff,
 2484                  sz, npol ? npol->mode : -1,
 2485                  npol ? npol->flags : -1,
 2486                  npol ? nodes_addr(npol->v.nodes)[0] : -1);
 2487 
 2488         if (npol) {
 2489                 new = sp_alloc(vma->vm_pgoff, vma->vm_pgoff + sz, npol);
 2490                 if (!new)
 2491                         return -ENOMEM;
 2492         }
 2493         err = shared_policy_replace(info, vma->vm_pgoff, vma->vm_pgoff+sz, new);
 2494         if (err && new)
 2495                 sp_free(new);
 2496         return err;
 2497 }
 2498 
 2499 /* Free a backing policy store on inode delete. */
 2500 void mpol_free_shared_policy(struct shared_policy *p)
 2501 {
 2502         struct sp_node *n;
 2503         struct rb_node *next;
 2504 
 2505         if (!p->root.rb_node)
 2506                 return;
 2507         spin_lock(&p->lock);
 2508         next = rb_first(&p->root);
 2509         while (next) {
 2510                 n = rb_entry(next, struct sp_node, nd);
 2511                 next = rb_next(&n->nd);
 2512                 sp_delete(p, n);
 2513         }
 2514         spin_unlock(&p->lock);
 2515 }
 2516 
 2517 #ifdef CONFIG_NUMA_BALANCING
 2518 static bool __initdata numabalancing_override;
 2519 
 2520 static void __init check_numabalancing_enable(void)
 2521 {
 2522         bool numabalancing_default = false;
 2523 
 2524         if (IS_ENABLED(CONFIG_NUMA_BALANCING_DEFAULT_ENABLED))
 2525                 numabalancing_default = true;
 2526 
 2527         if (nr_node_ids > 1 && !numabalancing_override) {
 2528                 printk(KERN_INFO "Enabling automatic NUMA balancing. "
 2529                         "Configure with numa_balancing= or sysctl");
 2530                 set_numabalancing_state(numabalancing_default);
 2531         }
 2532 }
 2533 
 2534 static int __init setup_numabalancing(char *str)
 2535 {
 2536         int ret = 0;
 2537         if (!str)
 2538                 goto out;
 2539         numabalancing_override = true;
 2540 
 2541         if (!strcmp(str, "enable")) {
 2542                 set_numabalancing_state(true);
 2543                 ret = 1;
 2544         } else if (!strcmp(str, "disable")) {
 2545                 set_numabalancing_state(false);
 2546                 ret = 1;
 2547         }
 2548 out:
 2549         if (!ret)
 2550                 printk(KERN_WARNING "Unable to parse numa_balancing=\n");
 2551 
 2552         return ret;
 2553 }
 2554 __setup("numa_balancing=", setup_numabalancing);
 2555 #else
 2556 static inline void __init check_numabalancing_enable(void)
 2557 {
 2558 }
 2559 #endif /* CONFIG_NUMA_BALANCING */
 2560 
 2561 /* assumes fs == KERNEL_DS */
 2562 void __init numa_policy_init(void)
 2563 {
 2564         nodemask_t interleave_nodes;
 2565         unsigned long largest = 0;
 2566         int nid, prefer = 0;
 2567 
 2568         policy_cache = kmem_cache_create("numa_policy",
 2569                                          sizeof(struct mempolicy),
 2570                                          0, SLAB_PANIC, NULL);
 2571 
 2572         sn_cache = kmem_cache_create("shared_policy_node",
 2573                                      sizeof(struct sp_node),
 2574                                      0, SLAB_PANIC, NULL);
 2575 
 2576         for_each_node(nid) {
 2577                 preferred_node_policy[nid] = (struct mempolicy) {
 2578                         .refcnt = ATOMIC_INIT(1),
 2579                         .mode = MPOL_PREFERRED,
 2580                         .flags = MPOL_F_MOF | MPOL_F_MORON,
 2581                         .v = { .preferred_node = nid, },
 2582                 };
 2583         }
 2584 
 2585         /*
 2586          * Set interleaving policy for system init. Interleaving is only
 2587          * enabled across suitably sized nodes (default is >= 16MB), or
 2588          * fall back to the largest node if they're all smaller.
 2589          */
 2590         nodes_clear(interleave_nodes);
 2591         for_each_node_state(nid, N_MEMORY) {
 2592                 unsigned long total_pages = node_present_pages(nid);
 2593 
 2594                 /* Preserve the largest node */
 2595                 if (largest < total_pages) {
 2596                         largest = total_pages;
 2597                         prefer = nid;
 2598                 }
 2599 
 2600                 /* Interleave this node? */
 2601                 if ((total_pages << PAGE_SHIFT) >= (16 << 20))
 2602                         node_set(nid, interleave_nodes);
 2603         }
 2604 
 2605         /* All too small, use the largest */
 2606         if (unlikely(nodes_empty(interleave_nodes)))
 2607                 node_set(prefer, interleave_nodes);
 2608 
 2609         if (do_set_mempolicy(MPOL_INTERLEAVE, 0, &interleave_nodes))
 2610                 printk("numa_policy_init: interleaving failed\n");
 2611 
 2612         check_numabalancing_enable();
 2613 }
 2614 
 2615 /* Reset policy of current process to default */
 2616 void numa_default_policy(void)
 2617 {
 2618         do_set_mempolicy(MPOL_DEFAULT, 0, NULL);
 2619 }
 2620 
 2621 /*
 2622  * Parse and format mempolicy from/to strings
 2623  */
 2624 
 2625 /*
 2626  * "local" is implemented internally by MPOL_PREFERRED with MPOL_F_LOCAL flag.
 2627  */
 2628 static const char * const policy_modes[] =
 2629 {
 2630         [MPOL_DEFAULT]    = "default",
 2631         [MPOL_PREFERRED]  = "prefer",
 2632         [MPOL_BIND]       = "bind",
 2633         [MPOL_INTERLEAVE] = "interleave",
 2634         [MPOL_LOCAL]      = "local",
 2635 };
 2636 
 2637 
 2638 #ifdef CONFIG_TMPFS
 2639 /**
 2640  * mpol_parse_str - parse string to mempolicy, for tmpfs mpol mount option.
 2641  * @str:  string containing mempolicy to parse
 2642  * @mpol:  pointer to struct mempolicy pointer, returned on success.
 2643  *
 2644  * Format of input:
 2645  *      <mode>[=<flags>][:<nodelist>]
 2646  *
 2647  * On success, returns 0, else 1
 2648  */
 2649 int mpol_parse_str(char *str, struct mempolicy **mpol)
 2650 {
 2651         struct mempolicy *new = NULL;
 2652         unsigned short mode;
 2653         unsigned short mode_flags;
 2654         nodemask_t nodes;
 2655         char *nodelist = strchr(str, ':');
 2656         char *flags = strchr(str, '=');
 2657         int err = 1;
 2658 
 2659         if (nodelist) {
 2660                 /* NUL-terminate mode or flags string */
 2661                 *nodelist++ = '\0';
 2662                 if (nodelist_parse(nodelist, nodes))
 2663                         goto out;
 2664                 if (!nodes_subset(nodes, node_states[N_MEMORY]))
 2665                         goto out;
 2666         } else
 2667                 nodes_clear(nodes);
 2668 
 2669         if (flags)
 2670                 *flags++ = '\0';        /* terminate mode string */
 2671 
 2672         for (mode = 0; mode < MPOL_MAX; mode++) {
 2673                 if (!strcmp(str, policy_modes[mode])) {
 2674                         break;
 2675                 }
 2676         }
 2677         if (mode >= MPOL_MAX)
 2678                 goto out;
 2679 
 2680         switch (mode) {
 2681         case MPOL_PREFERRED:
 2682                 /*
 2683                  * Insist on a nodelist of one node only
 2684                  */
 2685                 if (nodelist) {
 2686                         char *rest = nodelist;
 2687                         while (isdigit(*rest))
 2688                                 rest++;
 2689                         if (*rest)
 2690                                 goto out;
 2691                 }
 2692                 break;
 2693         case MPOL_INTERLEAVE:
 2694                 /*
 2695                  * Default to online nodes with memory if no nodelist
 2696                  */
 2697                 if (!nodelist)
 2698                         nodes = node_states[N_MEMORY];
 2699                 break;
 2700         case MPOL_LOCAL:
 2701                 /*
 2702                  * Don't allow a nodelist;  mpol_new() checks flags
 2703                  */
 2704                 if (nodelist)
 2705                         goto out;
 2706                 mode = MPOL_PREFERRED;
 2707                 break;
 2708         case MPOL_DEFAULT:
 2709                 /*
 2710                  * Insist on a empty nodelist
 2711                  */
 2712                 if (!nodelist)
 2713                         err = 0;
 2714                 goto out;
 2715         case MPOL_BIND:
 2716                 /*
 2717                  * Insist on a nodelist
 2718                  */
 2719                 if (!nodelist)
 2720                         goto out;
 2721         }
 2722 
 2723         mode_flags = 0;
 2724         if (flags) {
 2725                 /*
 2726                  * Currently, we only support two mutually exclusive
 2727                  * mode flags.
 2728                  */
 2729                 if (!strcmp(flags, "static"))
 2730                         mode_flags |= MPOL_F_STATIC_NODES;
 2731                 else if (!strcmp(flags, "relative"))
 2732                         mode_flags |= MPOL_F_RELATIVE_NODES;
 2733                 else
 2734                         goto out;
 2735         }
 2736 
 2737         new = mpol_new(mode, mode_flags, &nodes);
 2738         if (IS_ERR(new))
 2739                 goto out;
 2740 
 2741         /*
 2742          * Save nodes for mpol_to_str() to show the tmpfs mount options
 2743          * for /proc/mounts, /proc/pid/mounts and /proc/pid/mountinfo.
 2744          */
 2745         if (mode != MPOL_PREFERRED)
 2746                 new->v.nodes = nodes;
 2747         else if (nodelist)
 2748                 new->v.preferred_node = first_node(nodes);
 2749         else
 2750                 new->flags |= MPOL_F_LOCAL;
 2751 
 2752         /*
 2753          * Save nodes for contextualization: this will be used to "clone"
 2754          * the mempolicy in a specific context [cpuset] at a later time.
 2755          */
 2756         new->w.user_nodemask = nodes;
 2757 
 2758         err = 0;
 2759 
 2760 out:
 2761         /* Restore string for error message */
 2762         if (nodelist)
 2763                 *--nodelist = ':';
 2764         if (flags)
 2765                 *--flags = '=';
 2766         if (!err)
 2767                 *mpol = new;
 2768         return err;
 2769 }
 2770 #endif /* CONFIG_TMPFS */
 2771 
 2772 /**
 2773  * mpol_to_str - format a mempolicy structure for printing
 2774  * @buffer:  to contain formatted mempolicy string
 2775  * @maxlen:  length of @buffer
 2776  * @pol:  pointer to mempolicy to be formatted
 2777  *
 2778  * Convert a mempolicy into a string.
 2779  * Returns the number of characters in buffer (if positive)
 2780  * or an error (negative)
 2781  */
 2782 int mpol_to_str(char *buffer, int maxlen, struct mempolicy *pol)
 2783 {
 2784         char *p = buffer;
 2785         int l;
 2786         nodemask_t nodes;
 2787         unsigned short mode;
 2788         unsigned short flags = pol ? pol->flags : 0;
 2789 
 2790         /*
 2791          * Sanity check:  room for longest mode, flag and some nodes
 2792          */
 2793         VM_BUG_ON(maxlen < strlen("interleave") + strlen("relative") + 16);
 2794 
 2795         if (!pol || pol == &default_policy)
 2796                 mode = MPOL_DEFAULT;
 2797         else
 2798                 mode = pol->mode;
 2799 
 2800         switch (mode) {
 2801         case MPOL_DEFAULT:
 2802                 nodes_clear(nodes);
 2803                 break;
 2804 
 2805         case MPOL_PREFERRED:
 2806                 nodes_clear(nodes);
 2807                 if (flags & MPOL_F_LOCAL)
 2808                         mode = MPOL_LOCAL;
 2809                 else
 2810                         node_set(pol->v.preferred_node, nodes);
 2811                 break;
 2812 
 2813         case MPOL_BIND:
 2814                 /* Fall through */
 2815         case MPOL_INTERLEAVE:
 2816                 nodes = pol->v.nodes;
 2817                 break;
 2818 
 2819         default:
 2820                 return -EINVAL;
 2821         }
 2822 
 2823         l = strlen(policy_modes[mode]);
 2824         if (buffer + maxlen < p + l + 1)
 2825                 return -ENOSPC;
 2826 
 2827         strcpy(p, policy_modes[mode]);
 2828         p += l;
 2829 
 2830         if (flags & MPOL_MODE_FLAGS) {
 2831                 if (buffer + maxlen < p + 2)
 2832                         return -ENOSPC;
 2833                 *p++ = '=';
 2834 
 2835                 /*
 2836                  * Currently, the only defined flags are mutually exclusive
 2837                  */
 2838                 if (flags & MPOL_F_STATIC_NODES)
 2839                         p += snprintf(p, buffer + maxlen - p, "static");
 2840                 else if (flags & MPOL_F_RELATIVE_NODES)
 2841                         p += snprintf(p, buffer + maxlen - p, "relative");
 2842         }
 2843 
 2844         if (!nodes_empty(nodes)) {
 2845                 if (buffer + maxlen < p + 2)
 2846                         return -ENOSPC;
 2847                 *p++ = ':';
 2848                 p += nodelist_scnprintf(p, buffer + maxlen - p, nodes);
 2849         }
 2850         return p - buffer;
 2851 }

Cache object: d0b5bdeb21166de516713f15adc37628


[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]


This page is part of the FreeBSD/Linux Linux Kernel Cross-Reference, and was automatically generated using a modified version of the LXR engine.