11da177e4SLinus Torvalds /* 21da177e4SLinus Torvalds * Generic hugetlb support. 31da177e4SLinus Torvalds * (C) William Irwin, April 2004 41da177e4SLinus Torvalds */ 51da177e4SLinus Torvalds #include <linux/list.h> 61da177e4SLinus Torvalds #include <linux/init.h> 71da177e4SLinus Torvalds #include <linux/module.h> 81da177e4SLinus Torvalds #include <linux/mm.h> 9e1759c21SAlexey Dobriyan #include <linux/seq_file.h> 101da177e4SLinus Torvalds #include <linux/sysctl.h> 111da177e4SLinus Torvalds #include <linux/highmem.h> 12cddb8a5cSAndrea Arcangeli #include <linux/mmu_notifier.h> 131da177e4SLinus Torvalds #include <linux/nodemask.h> 1463551ae0SDavid Gibson #include <linux/pagemap.h> 155da7ca86SChristoph Lameter #include <linux/mempolicy.h> 16aea47ff3SChristoph Lameter #include <linux/cpuset.h> 173935baa9SDavid Gibson #include <linux/mutex.h> 18aa888a74SAndi Kleen #include <linux/bootmem.h> 19a3437870SNishanth Aravamudan #include <linux/sysfs.h> 205a0e3ad6STejun Heo #include <linux/slab.h> 210fe6e20bSNaoya Horiguchi #include <linux/rmap.h> 22fd6a03edSNaoya Horiguchi #include <linux/swap.h> 23fd6a03edSNaoya Horiguchi #include <linux/swapops.h> 24d6606683SLinus Torvalds 2563551ae0SDavid Gibson #include <asm/page.h> 2663551ae0SDavid Gibson #include <asm/pgtable.h> 2732f84528SChris Forbes #include <linux/io.h> 2863551ae0SDavid Gibson 2963551ae0SDavid Gibson #include <linux/hugetlb.h> 309a305230SLee Schermerhorn #include <linux/node.h> 317835e98bSNick Piggin #include "internal.h" 321da177e4SLinus Torvalds 331da177e4SLinus Torvalds const unsigned long hugetlb_zero = 0, hugetlb_infinity = ~0UL; 34396faf03SMel Gorman static gfp_t htlb_alloc_mask = GFP_HIGHUSER; 35396faf03SMel Gorman unsigned long hugepages_treat_as_movable; 36a5516438SAndi Kleen 37e5ff2159SAndi Kleen static int max_hstate; 38e5ff2159SAndi Kleen unsigned int default_hstate_idx; 39e5ff2159SAndi Kleen struct hstate hstates[HUGE_MAX_HSTATE]; 40e5ff2159SAndi Kleen 4153ba51d2SJon Tollefson __initdata LIST_HEAD(huge_boot_pages); 4253ba51d2SJon Tollefson 43e5ff2159SAndi Kleen /* for command line parsing */ 44e5ff2159SAndi Kleen static struct hstate * __initdata parsed_hstate; 45e5ff2159SAndi Kleen static unsigned long __initdata default_hstate_max_huge_pages; 46e11bfbfcSNick Piggin static unsigned long __initdata default_hstate_size; 47e5ff2159SAndi Kleen 48e5ff2159SAndi Kleen #define for_each_hstate(h) \ 49e5ff2159SAndi Kleen for ((h) = hstates; (h) < &hstates[max_hstate]; (h)++) 50396faf03SMel Gorman 513935baa9SDavid Gibson /* 523935baa9SDavid Gibson * Protects updates to hugepage_freelists, nr_huge_pages, and free_huge_pages 533935baa9SDavid Gibson */ 543935baa9SDavid Gibson static DEFINE_SPINLOCK(hugetlb_lock); 550bd0f9fbSEric Paris 56e7c4b0bfSAndy Whitcroft /* 5796822904SAndy Whitcroft * Region tracking -- allows tracking of reservations and instantiated pages 5896822904SAndy Whitcroft * across the pages in a mapping. 5984afd99bSAndy Whitcroft * 6084afd99bSAndy Whitcroft * The region data structures are protected by a combination of the mmap_sem 6184afd99bSAndy Whitcroft * and the hugetlb_instantion_mutex. To access or modify a region the caller 6284afd99bSAndy Whitcroft * must either hold the mmap_sem for write, or the mmap_sem for read and 6384afd99bSAndy Whitcroft * the hugetlb_instantiation mutex: 6484afd99bSAndy Whitcroft * 6584afd99bSAndy Whitcroft * down_write(&mm->mmap_sem); 6684afd99bSAndy Whitcroft * or 6784afd99bSAndy Whitcroft * down_read(&mm->mmap_sem); 6884afd99bSAndy Whitcroft * mutex_lock(&hugetlb_instantiation_mutex); 6996822904SAndy Whitcroft */ 7096822904SAndy Whitcroft struct file_region { 7196822904SAndy Whitcroft struct list_head link; 7296822904SAndy Whitcroft long from; 7396822904SAndy Whitcroft long to; 7496822904SAndy Whitcroft }; 7596822904SAndy Whitcroft 7696822904SAndy Whitcroft static long region_add(struct list_head *head, long f, long t) 7796822904SAndy Whitcroft { 7896822904SAndy Whitcroft struct file_region *rg, *nrg, *trg; 7996822904SAndy Whitcroft 8096822904SAndy Whitcroft /* Locate the region we are either in or before. */ 8196822904SAndy Whitcroft list_for_each_entry(rg, head, link) 8296822904SAndy Whitcroft if (f <= rg->to) 8396822904SAndy Whitcroft break; 8496822904SAndy Whitcroft 8596822904SAndy Whitcroft /* Round our left edge to the current segment if it encloses us. */ 8696822904SAndy Whitcroft if (f > rg->from) 8796822904SAndy Whitcroft f = rg->from; 8896822904SAndy Whitcroft 8996822904SAndy Whitcroft /* Check for and consume any regions we now overlap with. */ 9096822904SAndy Whitcroft nrg = rg; 9196822904SAndy Whitcroft list_for_each_entry_safe(rg, trg, rg->link.prev, link) { 9296822904SAndy Whitcroft if (&rg->link == head) 9396822904SAndy Whitcroft break; 9496822904SAndy Whitcroft if (rg->from > t) 9596822904SAndy Whitcroft break; 9696822904SAndy Whitcroft 9796822904SAndy Whitcroft /* If this area reaches higher then extend our area to 9896822904SAndy Whitcroft * include it completely. If this is not the first area 9996822904SAndy Whitcroft * which we intend to reuse, free it. */ 10096822904SAndy Whitcroft if (rg->to > t) 10196822904SAndy Whitcroft t = rg->to; 10296822904SAndy Whitcroft if (rg != nrg) { 10396822904SAndy Whitcroft list_del(&rg->link); 10496822904SAndy Whitcroft kfree(rg); 10596822904SAndy Whitcroft } 10696822904SAndy Whitcroft } 10796822904SAndy Whitcroft nrg->from = f; 10896822904SAndy Whitcroft nrg->to = t; 10996822904SAndy Whitcroft return 0; 11096822904SAndy Whitcroft } 11196822904SAndy Whitcroft 11296822904SAndy Whitcroft static long region_chg(struct list_head *head, long f, long t) 11396822904SAndy Whitcroft { 11496822904SAndy Whitcroft struct file_region *rg, *nrg; 11596822904SAndy Whitcroft long chg = 0; 11696822904SAndy Whitcroft 11796822904SAndy Whitcroft /* Locate the region we are before or in. */ 11896822904SAndy Whitcroft list_for_each_entry(rg, head, link) 11996822904SAndy Whitcroft if (f <= rg->to) 12096822904SAndy Whitcroft break; 12196822904SAndy Whitcroft 12296822904SAndy Whitcroft /* If we are below the current region then a new region is required. 12396822904SAndy Whitcroft * Subtle, allocate a new region at the position but make it zero 12496822904SAndy Whitcroft * size such that we can guarantee to record the reservation. */ 12596822904SAndy Whitcroft if (&rg->link == head || t < rg->from) { 12696822904SAndy Whitcroft nrg = kmalloc(sizeof(*nrg), GFP_KERNEL); 12796822904SAndy Whitcroft if (!nrg) 12896822904SAndy Whitcroft return -ENOMEM; 12996822904SAndy Whitcroft nrg->from = f; 13096822904SAndy Whitcroft nrg->to = f; 13196822904SAndy Whitcroft INIT_LIST_HEAD(&nrg->link); 13296822904SAndy Whitcroft list_add(&nrg->link, rg->link.prev); 13396822904SAndy Whitcroft 13496822904SAndy Whitcroft return t - f; 13596822904SAndy Whitcroft } 13696822904SAndy Whitcroft 13796822904SAndy Whitcroft /* Round our left edge to the current segment if it encloses us. */ 13896822904SAndy Whitcroft if (f > rg->from) 13996822904SAndy Whitcroft f = rg->from; 14096822904SAndy Whitcroft chg = t - f; 14196822904SAndy Whitcroft 14296822904SAndy Whitcroft /* Check for and consume any regions we now overlap with. */ 14396822904SAndy Whitcroft list_for_each_entry(rg, rg->link.prev, link) { 14496822904SAndy Whitcroft if (&rg->link == head) 14596822904SAndy Whitcroft break; 14696822904SAndy Whitcroft if (rg->from > t) 14796822904SAndy Whitcroft return chg; 14896822904SAndy Whitcroft 14925985edcSLucas De Marchi /* We overlap with this area, if it extends further than 15096822904SAndy Whitcroft * us then we must extend ourselves. Account for its 15196822904SAndy Whitcroft * existing reservation. */ 15296822904SAndy Whitcroft if (rg->to > t) { 15396822904SAndy Whitcroft chg += rg->to - t; 15496822904SAndy Whitcroft t = rg->to; 15596822904SAndy Whitcroft } 15696822904SAndy Whitcroft chg -= rg->to - rg->from; 15796822904SAndy Whitcroft } 15896822904SAndy Whitcroft return chg; 15996822904SAndy Whitcroft } 16096822904SAndy Whitcroft 16196822904SAndy Whitcroft static long region_truncate(struct list_head *head, long end) 16296822904SAndy Whitcroft { 16396822904SAndy Whitcroft struct file_region *rg, *trg; 16496822904SAndy Whitcroft long chg = 0; 16596822904SAndy Whitcroft 16696822904SAndy Whitcroft /* Locate the region we are either in or before. */ 16796822904SAndy Whitcroft list_for_each_entry(rg, head, link) 16896822904SAndy Whitcroft if (end <= rg->to) 16996822904SAndy Whitcroft break; 17096822904SAndy Whitcroft if (&rg->link == head) 17196822904SAndy Whitcroft return 0; 17296822904SAndy Whitcroft 17396822904SAndy Whitcroft /* If we are in the middle of a region then adjust it. */ 17496822904SAndy Whitcroft if (end > rg->from) { 17596822904SAndy Whitcroft chg = rg->to - end; 17696822904SAndy Whitcroft rg->to = end; 17796822904SAndy Whitcroft rg = list_entry(rg->link.next, typeof(*rg), link); 17896822904SAndy Whitcroft } 17996822904SAndy Whitcroft 18096822904SAndy Whitcroft /* Drop any remaining regions. */ 18196822904SAndy Whitcroft list_for_each_entry_safe(rg, trg, rg->link.prev, link) { 18296822904SAndy Whitcroft if (&rg->link == head) 18396822904SAndy Whitcroft break; 18496822904SAndy Whitcroft chg += rg->to - rg->from; 18596822904SAndy Whitcroft list_del(&rg->link); 18696822904SAndy Whitcroft kfree(rg); 18796822904SAndy Whitcroft } 18896822904SAndy Whitcroft return chg; 18996822904SAndy Whitcroft } 19096822904SAndy Whitcroft 19184afd99bSAndy Whitcroft static long region_count(struct list_head *head, long f, long t) 19284afd99bSAndy Whitcroft { 19384afd99bSAndy Whitcroft struct file_region *rg; 19484afd99bSAndy Whitcroft long chg = 0; 19584afd99bSAndy Whitcroft 19684afd99bSAndy Whitcroft /* Locate each segment we overlap with, and count that overlap. */ 19784afd99bSAndy Whitcroft list_for_each_entry(rg, head, link) { 19884afd99bSAndy Whitcroft int seg_from; 19984afd99bSAndy Whitcroft int seg_to; 20084afd99bSAndy Whitcroft 20184afd99bSAndy Whitcroft if (rg->to <= f) 20284afd99bSAndy Whitcroft continue; 20384afd99bSAndy Whitcroft if (rg->from >= t) 20484afd99bSAndy Whitcroft break; 20584afd99bSAndy Whitcroft 20684afd99bSAndy Whitcroft seg_from = max(rg->from, f); 20784afd99bSAndy Whitcroft seg_to = min(rg->to, t); 20884afd99bSAndy Whitcroft 20984afd99bSAndy Whitcroft chg += seg_to - seg_from; 21084afd99bSAndy Whitcroft } 21184afd99bSAndy Whitcroft 21284afd99bSAndy Whitcroft return chg; 21384afd99bSAndy Whitcroft } 21484afd99bSAndy Whitcroft 21596822904SAndy Whitcroft /* 216e7c4b0bfSAndy Whitcroft * Convert the address within this vma to the page offset within 217e7c4b0bfSAndy Whitcroft * the mapping, in pagecache page units; huge pages here. 218e7c4b0bfSAndy Whitcroft */ 219a5516438SAndi Kleen static pgoff_t vma_hugecache_offset(struct hstate *h, 220a5516438SAndi Kleen struct vm_area_struct *vma, unsigned long address) 221e7c4b0bfSAndy Whitcroft { 222a5516438SAndi Kleen return ((address - vma->vm_start) >> huge_page_shift(h)) + 223a5516438SAndi Kleen (vma->vm_pgoff >> huge_page_order(h)); 224e7c4b0bfSAndy Whitcroft } 225e7c4b0bfSAndy Whitcroft 2260fe6e20bSNaoya Horiguchi pgoff_t linear_hugepage_index(struct vm_area_struct *vma, 2270fe6e20bSNaoya Horiguchi unsigned long address) 2280fe6e20bSNaoya Horiguchi { 2290fe6e20bSNaoya Horiguchi return vma_hugecache_offset(hstate_vma(vma), vma, address); 2300fe6e20bSNaoya Horiguchi } 2310fe6e20bSNaoya Horiguchi 23284afd99bSAndy Whitcroft /* 23308fba699SMel Gorman * Return the size of the pages allocated when backing a VMA. In the majority 23408fba699SMel Gorman * cases this will be same size as used by the page table entries. 23508fba699SMel Gorman */ 23608fba699SMel Gorman unsigned long vma_kernel_pagesize(struct vm_area_struct *vma) 23708fba699SMel Gorman { 23808fba699SMel Gorman struct hstate *hstate; 23908fba699SMel Gorman 24008fba699SMel Gorman if (!is_vm_hugetlb_page(vma)) 24108fba699SMel Gorman return PAGE_SIZE; 24208fba699SMel Gorman 24308fba699SMel Gorman hstate = hstate_vma(vma); 24408fba699SMel Gorman 24508fba699SMel Gorman return 1UL << (hstate->order + PAGE_SHIFT); 24608fba699SMel Gorman } 247f340ca0fSJoerg Roedel EXPORT_SYMBOL_GPL(vma_kernel_pagesize); 24808fba699SMel Gorman 24908fba699SMel Gorman /* 2503340289dSMel Gorman * Return the page size being used by the MMU to back a VMA. In the majority 2513340289dSMel Gorman * of cases, the page size used by the kernel matches the MMU size. On 2523340289dSMel Gorman * architectures where it differs, an architecture-specific version of this 2533340289dSMel Gorman * function is required. 2543340289dSMel Gorman */ 2553340289dSMel Gorman #ifndef vma_mmu_pagesize 2563340289dSMel Gorman unsigned long vma_mmu_pagesize(struct vm_area_struct *vma) 2573340289dSMel Gorman { 2583340289dSMel Gorman return vma_kernel_pagesize(vma); 2593340289dSMel Gorman } 2603340289dSMel Gorman #endif 2613340289dSMel Gorman 2623340289dSMel Gorman /* 26384afd99bSAndy Whitcroft * Flags for MAP_PRIVATE reservations. These are stored in the bottom 26484afd99bSAndy Whitcroft * bits of the reservation map pointer, which are always clear due to 26584afd99bSAndy Whitcroft * alignment. 26684afd99bSAndy Whitcroft */ 26784afd99bSAndy Whitcroft #define HPAGE_RESV_OWNER (1UL << 0) 26884afd99bSAndy Whitcroft #define HPAGE_RESV_UNMAPPED (1UL << 1) 26904f2cbe3SMel Gorman #define HPAGE_RESV_MASK (HPAGE_RESV_OWNER | HPAGE_RESV_UNMAPPED) 27084afd99bSAndy Whitcroft 271a1e78772SMel Gorman /* 272a1e78772SMel Gorman * These helpers are used to track how many pages are reserved for 273a1e78772SMel Gorman * faults in a MAP_PRIVATE mapping. Only the process that called mmap() 274a1e78772SMel Gorman * is guaranteed to have their future faults succeed. 275a1e78772SMel Gorman * 276a1e78772SMel Gorman * With the exception of reset_vma_resv_huge_pages() which is called at fork(), 277a1e78772SMel Gorman * the reserve counters are updated with the hugetlb_lock held. It is safe 278a1e78772SMel Gorman * to reset the VMA at fork() time as it is not in use yet and there is no 279a1e78772SMel Gorman * chance of the global counters getting corrupted as a result of the values. 28084afd99bSAndy Whitcroft * 28184afd99bSAndy Whitcroft * The private mapping reservation is represented in a subtly different 28284afd99bSAndy Whitcroft * manner to a shared mapping. A shared mapping has a region map associated 28384afd99bSAndy Whitcroft * with the underlying file, this region map represents the backing file 28484afd99bSAndy Whitcroft * pages which have ever had a reservation assigned which this persists even 28584afd99bSAndy Whitcroft * after the page is instantiated. A private mapping has a region map 28684afd99bSAndy Whitcroft * associated with the original mmap which is attached to all VMAs which 28784afd99bSAndy Whitcroft * reference it, this region map represents those offsets which have consumed 28884afd99bSAndy Whitcroft * reservation ie. where pages have been instantiated. 289a1e78772SMel Gorman */ 290e7c4b0bfSAndy Whitcroft static unsigned long get_vma_private_data(struct vm_area_struct *vma) 291e7c4b0bfSAndy Whitcroft { 292e7c4b0bfSAndy Whitcroft return (unsigned long)vma->vm_private_data; 293e7c4b0bfSAndy Whitcroft } 294e7c4b0bfSAndy Whitcroft 295e7c4b0bfSAndy Whitcroft static void set_vma_private_data(struct vm_area_struct *vma, 296e7c4b0bfSAndy Whitcroft unsigned long value) 297e7c4b0bfSAndy Whitcroft { 298e7c4b0bfSAndy Whitcroft vma->vm_private_data = (void *)value; 299e7c4b0bfSAndy Whitcroft } 300e7c4b0bfSAndy Whitcroft 30184afd99bSAndy Whitcroft struct resv_map { 30284afd99bSAndy Whitcroft struct kref refs; 30384afd99bSAndy Whitcroft struct list_head regions; 30484afd99bSAndy Whitcroft }; 30584afd99bSAndy Whitcroft 3062a4b3dedSHarvey Harrison static struct resv_map *resv_map_alloc(void) 30784afd99bSAndy Whitcroft { 30884afd99bSAndy Whitcroft struct resv_map *resv_map = kmalloc(sizeof(*resv_map), GFP_KERNEL); 30984afd99bSAndy Whitcroft if (!resv_map) 31084afd99bSAndy Whitcroft return NULL; 31184afd99bSAndy Whitcroft 31284afd99bSAndy Whitcroft kref_init(&resv_map->refs); 31384afd99bSAndy Whitcroft INIT_LIST_HEAD(&resv_map->regions); 31484afd99bSAndy Whitcroft 31584afd99bSAndy Whitcroft return resv_map; 31684afd99bSAndy Whitcroft } 31784afd99bSAndy Whitcroft 3182a4b3dedSHarvey Harrison static void resv_map_release(struct kref *ref) 31984afd99bSAndy Whitcroft { 32084afd99bSAndy Whitcroft struct resv_map *resv_map = container_of(ref, struct resv_map, refs); 32184afd99bSAndy Whitcroft 32284afd99bSAndy Whitcroft /* Clear out any active regions before we release the map. */ 32384afd99bSAndy Whitcroft region_truncate(&resv_map->regions, 0); 32484afd99bSAndy Whitcroft kfree(resv_map); 32584afd99bSAndy Whitcroft } 32684afd99bSAndy Whitcroft 32784afd99bSAndy Whitcroft static struct resv_map *vma_resv_map(struct vm_area_struct *vma) 328a1e78772SMel Gorman { 329a1e78772SMel Gorman VM_BUG_ON(!is_vm_hugetlb_page(vma)); 330f83a275dSMel Gorman if (!(vma->vm_flags & VM_MAYSHARE)) 33184afd99bSAndy Whitcroft return (struct resv_map *)(get_vma_private_data(vma) & 33284afd99bSAndy Whitcroft ~HPAGE_RESV_MASK); 3332a4b3dedSHarvey Harrison return NULL; 334a1e78772SMel Gorman } 335a1e78772SMel Gorman 33684afd99bSAndy Whitcroft static void set_vma_resv_map(struct vm_area_struct *vma, struct resv_map *map) 337a1e78772SMel Gorman { 338a1e78772SMel Gorman VM_BUG_ON(!is_vm_hugetlb_page(vma)); 339f83a275dSMel Gorman VM_BUG_ON(vma->vm_flags & VM_MAYSHARE); 340a1e78772SMel Gorman 34184afd99bSAndy Whitcroft set_vma_private_data(vma, (get_vma_private_data(vma) & 34284afd99bSAndy Whitcroft HPAGE_RESV_MASK) | (unsigned long)map); 34304f2cbe3SMel Gorman } 34404f2cbe3SMel Gorman 34504f2cbe3SMel Gorman static void set_vma_resv_flags(struct vm_area_struct *vma, unsigned long flags) 34604f2cbe3SMel Gorman { 34704f2cbe3SMel Gorman VM_BUG_ON(!is_vm_hugetlb_page(vma)); 348f83a275dSMel Gorman VM_BUG_ON(vma->vm_flags & VM_MAYSHARE); 349e7c4b0bfSAndy Whitcroft 350e7c4b0bfSAndy Whitcroft set_vma_private_data(vma, get_vma_private_data(vma) | flags); 35104f2cbe3SMel Gorman } 35204f2cbe3SMel Gorman 35304f2cbe3SMel Gorman static int is_vma_resv_set(struct vm_area_struct *vma, unsigned long flag) 35404f2cbe3SMel Gorman { 35504f2cbe3SMel Gorman VM_BUG_ON(!is_vm_hugetlb_page(vma)); 356e7c4b0bfSAndy Whitcroft 357e7c4b0bfSAndy Whitcroft return (get_vma_private_data(vma) & flag) != 0; 358a1e78772SMel Gorman } 359a1e78772SMel Gorman 360a1e78772SMel Gorman /* Decrement the reserved pages in the hugepage pool by one */ 361a5516438SAndi Kleen static void decrement_hugepage_resv_vma(struct hstate *h, 362a5516438SAndi Kleen struct vm_area_struct *vma) 363a1e78772SMel Gorman { 364c37f9fb1SAndy Whitcroft if (vma->vm_flags & VM_NORESERVE) 365c37f9fb1SAndy Whitcroft return; 366c37f9fb1SAndy Whitcroft 367f83a275dSMel Gorman if (vma->vm_flags & VM_MAYSHARE) { 368a1e78772SMel Gorman /* Shared mappings always use reserves */ 369a5516438SAndi Kleen h->resv_huge_pages--; 37084afd99bSAndy Whitcroft } else if (is_vma_resv_set(vma, HPAGE_RESV_OWNER)) { 371a1e78772SMel Gorman /* 372a1e78772SMel Gorman * Only the process that called mmap() has reserves for 373a1e78772SMel Gorman * private mappings. 374a1e78772SMel Gorman */ 375a5516438SAndi Kleen h->resv_huge_pages--; 376a1e78772SMel Gorman } 377a1e78772SMel Gorman } 378a1e78772SMel Gorman 37904f2cbe3SMel Gorman /* Reset counters to 0 and clear all HPAGE_RESV_* flags */ 380a1e78772SMel Gorman void reset_vma_resv_huge_pages(struct vm_area_struct *vma) 381a1e78772SMel Gorman { 382a1e78772SMel Gorman VM_BUG_ON(!is_vm_hugetlb_page(vma)); 383f83a275dSMel Gorman if (!(vma->vm_flags & VM_MAYSHARE)) 384a1e78772SMel Gorman vma->vm_private_data = (void *)0; 385a1e78772SMel Gorman } 386a1e78772SMel Gorman 387a1e78772SMel Gorman /* Returns true if the VMA has associated reserve pages */ 3887f09ca51SMel Gorman static int vma_has_reserves(struct vm_area_struct *vma) 389a1e78772SMel Gorman { 390f83a275dSMel Gorman if (vma->vm_flags & VM_MAYSHARE) 391a1e78772SMel Gorman return 1; 3927f09ca51SMel Gorman if (is_vma_resv_set(vma, HPAGE_RESV_OWNER)) 3937f09ca51SMel Gorman return 1; 3947f09ca51SMel Gorman return 0; 395a1e78772SMel Gorman } 396a1e78772SMel Gorman 3970ebabb41SNaoya Horiguchi static void copy_gigantic_page(struct page *dst, struct page *src) 3980ebabb41SNaoya Horiguchi { 3990ebabb41SNaoya Horiguchi int i; 4000ebabb41SNaoya Horiguchi struct hstate *h = page_hstate(src); 4010ebabb41SNaoya Horiguchi struct page *dst_base = dst; 4020ebabb41SNaoya Horiguchi struct page *src_base = src; 4030ebabb41SNaoya Horiguchi 4040ebabb41SNaoya Horiguchi for (i = 0; i < pages_per_huge_page(h); ) { 4050ebabb41SNaoya Horiguchi cond_resched(); 4060ebabb41SNaoya Horiguchi copy_highpage(dst, src); 4070ebabb41SNaoya Horiguchi 4080ebabb41SNaoya Horiguchi i++; 4090ebabb41SNaoya Horiguchi dst = mem_map_next(dst, dst_base, i); 4100ebabb41SNaoya Horiguchi src = mem_map_next(src, src_base, i); 4110ebabb41SNaoya Horiguchi } 4120ebabb41SNaoya Horiguchi } 4130ebabb41SNaoya Horiguchi 4140ebabb41SNaoya Horiguchi void copy_huge_page(struct page *dst, struct page *src) 4150ebabb41SNaoya Horiguchi { 4160ebabb41SNaoya Horiguchi int i; 4170ebabb41SNaoya Horiguchi struct hstate *h = page_hstate(src); 4180ebabb41SNaoya Horiguchi 4190ebabb41SNaoya Horiguchi if (unlikely(pages_per_huge_page(h) > MAX_ORDER_NR_PAGES)) { 4200ebabb41SNaoya Horiguchi copy_gigantic_page(dst, src); 4210ebabb41SNaoya Horiguchi return; 4220ebabb41SNaoya Horiguchi } 4230ebabb41SNaoya Horiguchi 4240ebabb41SNaoya Horiguchi might_sleep(); 4250ebabb41SNaoya Horiguchi for (i = 0; i < pages_per_huge_page(h); i++) { 4260ebabb41SNaoya Horiguchi cond_resched(); 4270ebabb41SNaoya Horiguchi copy_highpage(dst + i, src + i); 4280ebabb41SNaoya Horiguchi } 4290ebabb41SNaoya Horiguchi } 4300ebabb41SNaoya Horiguchi 431a5516438SAndi Kleen static void enqueue_huge_page(struct hstate *h, struct page *page) 4321da177e4SLinus Torvalds { 4331da177e4SLinus Torvalds int nid = page_to_nid(page); 434a5516438SAndi Kleen list_add(&page->lru, &h->hugepage_freelists[nid]); 435a5516438SAndi Kleen h->free_huge_pages++; 436a5516438SAndi Kleen h->free_huge_pages_node[nid]++; 4371da177e4SLinus Torvalds } 4381da177e4SLinus Torvalds 439bf50bab2SNaoya Horiguchi static struct page *dequeue_huge_page_node(struct hstate *h, int nid) 440bf50bab2SNaoya Horiguchi { 441bf50bab2SNaoya Horiguchi struct page *page; 442bf50bab2SNaoya Horiguchi 443bf50bab2SNaoya Horiguchi if (list_empty(&h->hugepage_freelists[nid])) 444bf50bab2SNaoya Horiguchi return NULL; 445bf50bab2SNaoya Horiguchi page = list_entry(h->hugepage_freelists[nid].next, struct page, lru); 446bf50bab2SNaoya Horiguchi list_del(&page->lru); 447a9869b83SNaoya Horiguchi set_page_refcounted(page); 448bf50bab2SNaoya Horiguchi h->free_huge_pages--; 449bf50bab2SNaoya Horiguchi h->free_huge_pages_node[nid]--; 450bf50bab2SNaoya Horiguchi return page; 451bf50bab2SNaoya Horiguchi } 452bf50bab2SNaoya Horiguchi 453a5516438SAndi Kleen static struct page *dequeue_huge_page_vma(struct hstate *h, 454a5516438SAndi Kleen struct vm_area_struct *vma, 45504f2cbe3SMel Gorman unsigned long address, int avoid_reserve) 4561da177e4SLinus Torvalds { 4571da177e4SLinus Torvalds struct page *page = NULL; 458480eccf9SLee Schermerhorn struct mempolicy *mpol; 45919770b32SMel Gorman nodemask_t *nodemask; 460c0ff7453SMiao Xie struct zonelist *zonelist; 461dd1a239fSMel Gorman struct zone *zone; 462dd1a239fSMel Gorman struct zoneref *z; 4631da177e4SLinus Torvalds 464c0ff7453SMiao Xie get_mems_allowed(); 465c0ff7453SMiao Xie zonelist = huge_zonelist(vma, address, 466c0ff7453SMiao Xie htlb_alloc_mask, &mpol, &nodemask); 467a1e78772SMel Gorman /* 468a1e78772SMel Gorman * A child process with MAP_PRIVATE mappings created by their parent 469a1e78772SMel Gorman * have no page reserves. This check ensures that reservations are 470a1e78772SMel Gorman * not "stolen". The child may still get SIGKILLed 471a1e78772SMel Gorman */ 4727f09ca51SMel Gorman if (!vma_has_reserves(vma) && 473a5516438SAndi Kleen h->free_huge_pages - h->resv_huge_pages == 0) 474c0ff7453SMiao Xie goto err; 475a1e78772SMel Gorman 47604f2cbe3SMel Gorman /* If reserves cannot be used, ensure enough pages are in the pool */ 477a5516438SAndi Kleen if (avoid_reserve && h->free_huge_pages - h->resv_huge_pages == 0) 4786eab04a8SJustin P. Mattock goto err; 47904f2cbe3SMel Gorman 48019770b32SMel Gorman for_each_zone_zonelist_nodemask(zone, z, zonelist, 48119770b32SMel Gorman MAX_NR_ZONES - 1, nodemask) { 482bf50bab2SNaoya Horiguchi if (cpuset_zone_allowed_softwall(zone, htlb_alloc_mask)) { 483bf50bab2SNaoya Horiguchi page = dequeue_huge_page_node(h, zone_to_nid(zone)); 484bf50bab2SNaoya Horiguchi if (page) { 48504f2cbe3SMel Gorman if (!avoid_reserve) 486a5516438SAndi Kleen decrement_hugepage_resv_vma(h, vma); 4875ab3ee7bSKen Chen break; 4881da177e4SLinus Torvalds } 4893abf7afdSAndrew Morton } 490bf50bab2SNaoya Horiguchi } 491c0ff7453SMiao Xie err: 49252cd3b07SLee Schermerhorn mpol_cond_put(mpol); 493c0ff7453SMiao Xie put_mems_allowed(); 4941da177e4SLinus Torvalds return page; 4951da177e4SLinus Torvalds } 4961da177e4SLinus Torvalds 497a5516438SAndi Kleen static void update_and_free_page(struct hstate *h, struct page *page) 4986af2acb6SAdam Litke { 4996af2acb6SAdam Litke int i; 500a5516438SAndi Kleen 50118229df5SAndy Whitcroft VM_BUG_ON(h->order >= MAX_ORDER); 50218229df5SAndy Whitcroft 503a5516438SAndi Kleen h->nr_huge_pages--; 504a5516438SAndi Kleen h->nr_huge_pages_node[page_to_nid(page)]--; 505a5516438SAndi Kleen for (i = 0; i < pages_per_huge_page(h); i++) { 50632f84528SChris Forbes page[i].flags &= ~(1 << PG_locked | 1 << PG_error | 50732f84528SChris Forbes 1 << PG_referenced | 1 << PG_dirty | 50832f84528SChris Forbes 1 << PG_active | 1 << PG_reserved | 5096af2acb6SAdam Litke 1 << PG_private | 1 << PG_writeback); 5106af2acb6SAdam Litke } 5116af2acb6SAdam Litke set_compound_page_dtor(page, NULL); 5126af2acb6SAdam Litke set_page_refcounted(page); 5137f2e9525SGerald Schaefer arch_release_hugepage(page); 514a5516438SAndi Kleen __free_pages(page, huge_page_order(h)); 5156af2acb6SAdam Litke } 5166af2acb6SAdam Litke 517e5ff2159SAndi Kleen struct hstate *size_to_hstate(unsigned long size) 518e5ff2159SAndi Kleen { 519e5ff2159SAndi Kleen struct hstate *h; 520e5ff2159SAndi Kleen 521e5ff2159SAndi Kleen for_each_hstate(h) { 522e5ff2159SAndi Kleen if (huge_page_size(h) == size) 523e5ff2159SAndi Kleen return h; 524e5ff2159SAndi Kleen } 525e5ff2159SAndi Kleen return NULL; 526e5ff2159SAndi Kleen } 527e5ff2159SAndi Kleen 52827a85ef1SDavid Gibson static void free_huge_page(struct page *page) 52927a85ef1SDavid Gibson { 530a5516438SAndi Kleen /* 531a5516438SAndi Kleen * Can't pass hstate in here because it is called from the 532a5516438SAndi Kleen * compound page destructor. 533a5516438SAndi Kleen */ 534e5ff2159SAndi Kleen struct hstate *h = page_hstate(page); 5357893d1d5SAdam Litke int nid = page_to_nid(page); 536c79fb75eSAdam Litke struct address_space *mapping; 53727a85ef1SDavid Gibson 538c79fb75eSAdam Litke mapping = (struct address_space *) page_private(page); 539e5df70abSAndy Whitcroft set_page_private(page, 0); 54023be7468SMel Gorman page->mapping = NULL; 5417893d1d5SAdam Litke BUG_ON(page_count(page)); 5420fe6e20bSNaoya Horiguchi BUG_ON(page_mapcount(page)); 54327a85ef1SDavid Gibson INIT_LIST_HEAD(&page->lru); 54427a85ef1SDavid Gibson 54527a85ef1SDavid Gibson spin_lock(&hugetlb_lock); 546aa888a74SAndi Kleen if (h->surplus_huge_pages_node[nid] && huge_page_order(h) < MAX_ORDER) { 547a5516438SAndi Kleen update_and_free_page(h, page); 548a5516438SAndi Kleen h->surplus_huge_pages--; 549a5516438SAndi Kleen h->surplus_huge_pages_node[nid]--; 5507893d1d5SAdam Litke } else { 551a5516438SAndi Kleen enqueue_huge_page(h, page); 5527893d1d5SAdam Litke } 55327a85ef1SDavid Gibson spin_unlock(&hugetlb_lock); 554c79fb75eSAdam Litke if (mapping) 5559a119c05SAdam Litke hugetlb_put_quota(mapping, 1); 55627a85ef1SDavid Gibson } 55727a85ef1SDavid Gibson 558a5516438SAndi Kleen static void prep_new_huge_page(struct hstate *h, struct page *page, int nid) 559b7ba30c6SAndi Kleen { 560b7ba30c6SAndi Kleen set_compound_page_dtor(page, free_huge_page); 561b7ba30c6SAndi Kleen spin_lock(&hugetlb_lock); 562a5516438SAndi Kleen h->nr_huge_pages++; 563a5516438SAndi Kleen h->nr_huge_pages_node[nid]++; 564b7ba30c6SAndi Kleen spin_unlock(&hugetlb_lock); 565b7ba30c6SAndi Kleen put_page(page); /* free it into the hugepage allocator */ 566b7ba30c6SAndi Kleen } 567b7ba30c6SAndi Kleen 56820a0307cSWu Fengguang static void prep_compound_gigantic_page(struct page *page, unsigned long order) 56920a0307cSWu Fengguang { 57020a0307cSWu Fengguang int i; 57120a0307cSWu Fengguang int nr_pages = 1 << order; 57220a0307cSWu Fengguang struct page *p = page + 1; 57320a0307cSWu Fengguang 57420a0307cSWu Fengguang /* we rely on prep_new_huge_page to set the destructor */ 57520a0307cSWu Fengguang set_compound_order(page, order); 57620a0307cSWu Fengguang __SetPageHead(page); 57720a0307cSWu Fengguang for (i = 1; i < nr_pages; i++, p = mem_map_next(p, page, i)) { 57820a0307cSWu Fengguang __SetPageTail(p); 57958a84aa9SYouquan Song set_page_count(p, 0); 58020a0307cSWu Fengguang p->first_page = page; 58120a0307cSWu Fengguang } 58220a0307cSWu Fengguang } 58320a0307cSWu Fengguang 58420a0307cSWu Fengguang int PageHuge(struct page *page) 58520a0307cSWu Fengguang { 58620a0307cSWu Fengguang compound_page_dtor *dtor; 58720a0307cSWu Fengguang 58820a0307cSWu Fengguang if (!PageCompound(page)) 58920a0307cSWu Fengguang return 0; 59020a0307cSWu Fengguang 59120a0307cSWu Fengguang page = compound_head(page); 59220a0307cSWu Fengguang dtor = get_compound_page_dtor(page); 59320a0307cSWu Fengguang 59420a0307cSWu Fengguang return dtor == free_huge_page; 59520a0307cSWu Fengguang } 59643131e14SNaoya Horiguchi EXPORT_SYMBOL_GPL(PageHuge); 59743131e14SNaoya Horiguchi 598a5516438SAndi Kleen static struct page *alloc_fresh_huge_page_node(struct hstate *h, int nid) 5991da177e4SLinus Torvalds { 6001da177e4SLinus Torvalds struct page *page; 601f96efd58SJoe Jin 602aa888a74SAndi Kleen if (h->order >= MAX_ORDER) 603aa888a74SAndi Kleen return NULL; 604aa888a74SAndi Kleen 6056484eb3eSMel Gorman page = alloc_pages_exact_node(nid, 606551883aeSNishanth Aravamudan htlb_alloc_mask|__GFP_COMP|__GFP_THISNODE| 607551883aeSNishanth Aravamudan __GFP_REPEAT|__GFP_NOWARN, 608a5516438SAndi Kleen huge_page_order(h)); 6091da177e4SLinus Torvalds if (page) { 6107f2e9525SGerald Schaefer if (arch_prepare_hugepage(page)) { 611caff3a2cSGerald Schaefer __free_pages(page, huge_page_order(h)); 6127b8ee84dSHarvey Harrison return NULL; 6137f2e9525SGerald Schaefer } 614a5516438SAndi Kleen prep_new_huge_page(h, page, nid); 6151da177e4SLinus Torvalds } 61663b4613cSNishanth Aravamudan 61763b4613cSNishanth Aravamudan return page; 61863b4613cSNishanth Aravamudan } 61963b4613cSNishanth Aravamudan 6205ced66c9SAndi Kleen /* 6216ae11b27SLee Schermerhorn * common helper functions for hstate_next_node_to_{alloc|free}. 6226ae11b27SLee Schermerhorn * We may have allocated or freed a huge page based on a different 6236ae11b27SLee Schermerhorn * nodes_allowed previously, so h->next_node_to_{alloc|free} might 6246ae11b27SLee Schermerhorn * be outside of *nodes_allowed. Ensure that we use an allowed 6256ae11b27SLee Schermerhorn * node for alloc or free. 6269a76db09SLee Schermerhorn */ 6276ae11b27SLee Schermerhorn static int next_node_allowed(int nid, nodemask_t *nodes_allowed) 6289a76db09SLee Schermerhorn { 6296ae11b27SLee Schermerhorn nid = next_node(nid, *nodes_allowed); 6309a76db09SLee Schermerhorn if (nid == MAX_NUMNODES) 6316ae11b27SLee Schermerhorn nid = first_node(*nodes_allowed); 6329a76db09SLee Schermerhorn VM_BUG_ON(nid >= MAX_NUMNODES); 6339a76db09SLee Schermerhorn 6349a76db09SLee Schermerhorn return nid; 6359a76db09SLee Schermerhorn } 6369a76db09SLee Schermerhorn 6376ae11b27SLee Schermerhorn static int get_valid_node_allowed(int nid, nodemask_t *nodes_allowed) 6385ced66c9SAndi Kleen { 6396ae11b27SLee Schermerhorn if (!node_isset(nid, *nodes_allowed)) 6406ae11b27SLee Schermerhorn nid = next_node_allowed(nid, nodes_allowed); 6419a76db09SLee Schermerhorn return nid; 6425ced66c9SAndi Kleen } 6435ced66c9SAndi Kleen 6446ae11b27SLee Schermerhorn /* 6456ae11b27SLee Schermerhorn * returns the previously saved node ["this node"] from which to 6466ae11b27SLee Schermerhorn * allocate a persistent huge page for the pool and advance the 6476ae11b27SLee Schermerhorn * next node from which to allocate, handling wrap at end of node 6486ae11b27SLee Schermerhorn * mask. 6496ae11b27SLee Schermerhorn */ 6506ae11b27SLee Schermerhorn static int hstate_next_node_to_alloc(struct hstate *h, 6516ae11b27SLee Schermerhorn nodemask_t *nodes_allowed) 6526ae11b27SLee Schermerhorn { 6536ae11b27SLee Schermerhorn int nid; 6546ae11b27SLee Schermerhorn 6556ae11b27SLee Schermerhorn VM_BUG_ON(!nodes_allowed); 6566ae11b27SLee Schermerhorn 6576ae11b27SLee Schermerhorn nid = get_valid_node_allowed(h->next_nid_to_alloc, nodes_allowed); 6586ae11b27SLee Schermerhorn h->next_nid_to_alloc = next_node_allowed(nid, nodes_allowed); 6596ae11b27SLee Schermerhorn 6606ae11b27SLee Schermerhorn return nid; 6616ae11b27SLee Schermerhorn } 6626ae11b27SLee Schermerhorn 6636ae11b27SLee Schermerhorn static int alloc_fresh_huge_page(struct hstate *h, nodemask_t *nodes_allowed) 66463b4613cSNishanth Aravamudan { 66563b4613cSNishanth Aravamudan struct page *page; 66663b4613cSNishanth Aravamudan int start_nid; 66763b4613cSNishanth Aravamudan int next_nid; 66863b4613cSNishanth Aravamudan int ret = 0; 66963b4613cSNishanth Aravamudan 6706ae11b27SLee Schermerhorn start_nid = hstate_next_node_to_alloc(h, nodes_allowed); 671e8c5c824SLee Schermerhorn next_nid = start_nid; 67263b4613cSNishanth Aravamudan 67363b4613cSNishanth Aravamudan do { 674e8c5c824SLee Schermerhorn page = alloc_fresh_huge_page_node(h, next_nid); 6759a76db09SLee Schermerhorn if (page) { 67663b4613cSNishanth Aravamudan ret = 1; 6779a76db09SLee Schermerhorn break; 6789a76db09SLee Schermerhorn } 6796ae11b27SLee Schermerhorn next_nid = hstate_next_node_to_alloc(h, nodes_allowed); 6809a76db09SLee Schermerhorn } while (next_nid != start_nid); 68163b4613cSNishanth Aravamudan 6823b116300SAdam Litke if (ret) 6833b116300SAdam Litke count_vm_event(HTLB_BUDDY_PGALLOC); 6843b116300SAdam Litke else 6853b116300SAdam Litke count_vm_event(HTLB_BUDDY_PGALLOC_FAIL); 6863b116300SAdam Litke 68763b4613cSNishanth Aravamudan return ret; 6881da177e4SLinus Torvalds } 6891da177e4SLinus Torvalds 690e8c5c824SLee Schermerhorn /* 6916ae11b27SLee Schermerhorn * helper for free_pool_huge_page() - return the previously saved 6926ae11b27SLee Schermerhorn * node ["this node"] from which to free a huge page. Advance the 6936ae11b27SLee Schermerhorn * next node id whether or not we find a free huge page to free so 6946ae11b27SLee Schermerhorn * that the next attempt to free addresses the next node. 695e8c5c824SLee Schermerhorn */ 6966ae11b27SLee Schermerhorn static int hstate_next_node_to_free(struct hstate *h, nodemask_t *nodes_allowed) 697e8c5c824SLee Schermerhorn { 6986ae11b27SLee Schermerhorn int nid; 6999a76db09SLee Schermerhorn 7006ae11b27SLee Schermerhorn VM_BUG_ON(!nodes_allowed); 7016ae11b27SLee Schermerhorn 7026ae11b27SLee Schermerhorn nid = get_valid_node_allowed(h->next_nid_to_free, nodes_allowed); 7036ae11b27SLee Schermerhorn h->next_nid_to_free = next_node_allowed(nid, nodes_allowed); 7046ae11b27SLee Schermerhorn 7059a76db09SLee Schermerhorn return nid; 706e8c5c824SLee Schermerhorn } 707e8c5c824SLee Schermerhorn 708e8c5c824SLee Schermerhorn /* 709e8c5c824SLee Schermerhorn * Free huge page from pool from next node to free. 710e8c5c824SLee Schermerhorn * Attempt to keep persistent huge pages more or less 711e8c5c824SLee Schermerhorn * balanced over allowed nodes. 712e8c5c824SLee Schermerhorn * Called with hugetlb_lock locked. 713e8c5c824SLee Schermerhorn */ 7146ae11b27SLee Schermerhorn static int free_pool_huge_page(struct hstate *h, nodemask_t *nodes_allowed, 7156ae11b27SLee Schermerhorn bool acct_surplus) 716e8c5c824SLee Schermerhorn { 717e8c5c824SLee Schermerhorn int start_nid; 718e8c5c824SLee Schermerhorn int next_nid; 719e8c5c824SLee Schermerhorn int ret = 0; 720e8c5c824SLee Schermerhorn 7216ae11b27SLee Schermerhorn start_nid = hstate_next_node_to_free(h, nodes_allowed); 722e8c5c824SLee Schermerhorn next_nid = start_nid; 723e8c5c824SLee Schermerhorn 724e8c5c824SLee Schermerhorn do { 725685f3457SLee Schermerhorn /* 726685f3457SLee Schermerhorn * If we're returning unused surplus pages, only examine 727685f3457SLee Schermerhorn * nodes with surplus pages. 728685f3457SLee Schermerhorn */ 729685f3457SLee Schermerhorn if ((!acct_surplus || h->surplus_huge_pages_node[next_nid]) && 730685f3457SLee Schermerhorn !list_empty(&h->hugepage_freelists[next_nid])) { 731e8c5c824SLee Schermerhorn struct page *page = 732e8c5c824SLee Schermerhorn list_entry(h->hugepage_freelists[next_nid].next, 733e8c5c824SLee Schermerhorn struct page, lru); 734e8c5c824SLee Schermerhorn list_del(&page->lru); 735e8c5c824SLee Schermerhorn h->free_huge_pages--; 736e8c5c824SLee Schermerhorn h->free_huge_pages_node[next_nid]--; 737685f3457SLee Schermerhorn if (acct_surplus) { 738685f3457SLee Schermerhorn h->surplus_huge_pages--; 739685f3457SLee Schermerhorn h->surplus_huge_pages_node[next_nid]--; 740685f3457SLee Schermerhorn } 741e8c5c824SLee Schermerhorn update_and_free_page(h, page); 742e8c5c824SLee Schermerhorn ret = 1; 7439a76db09SLee Schermerhorn break; 744e8c5c824SLee Schermerhorn } 7456ae11b27SLee Schermerhorn next_nid = hstate_next_node_to_free(h, nodes_allowed); 7469a76db09SLee Schermerhorn } while (next_nid != start_nid); 747e8c5c824SLee Schermerhorn 748e8c5c824SLee Schermerhorn return ret; 749e8c5c824SLee Schermerhorn } 750e8c5c824SLee Schermerhorn 751bf50bab2SNaoya Horiguchi static struct page *alloc_buddy_huge_page(struct hstate *h, int nid) 7527893d1d5SAdam Litke { 7537893d1d5SAdam Litke struct page *page; 754bf50bab2SNaoya Horiguchi unsigned int r_nid; 7557893d1d5SAdam Litke 756aa888a74SAndi Kleen if (h->order >= MAX_ORDER) 757aa888a74SAndi Kleen return NULL; 758aa888a74SAndi Kleen 759d1c3fb1fSNishanth Aravamudan /* 760d1c3fb1fSNishanth Aravamudan * Assume we will successfully allocate the surplus page to 761d1c3fb1fSNishanth Aravamudan * prevent racing processes from causing the surplus to exceed 762d1c3fb1fSNishanth Aravamudan * overcommit 763d1c3fb1fSNishanth Aravamudan * 764d1c3fb1fSNishanth Aravamudan * This however introduces a different race, where a process B 765d1c3fb1fSNishanth Aravamudan * tries to grow the static hugepage pool while alloc_pages() is 766d1c3fb1fSNishanth Aravamudan * called by process A. B will only examine the per-node 767d1c3fb1fSNishanth Aravamudan * counters in determining if surplus huge pages can be 768d1c3fb1fSNishanth Aravamudan * converted to normal huge pages in adjust_pool_surplus(). A 769d1c3fb1fSNishanth Aravamudan * won't be able to increment the per-node counter, until the 770d1c3fb1fSNishanth Aravamudan * lock is dropped by B, but B doesn't drop hugetlb_lock until 771d1c3fb1fSNishanth Aravamudan * no more huge pages can be converted from surplus to normal 772d1c3fb1fSNishanth Aravamudan * state (and doesn't try to convert again). Thus, we have a 773d1c3fb1fSNishanth Aravamudan * case where a surplus huge page exists, the pool is grown, and 774d1c3fb1fSNishanth Aravamudan * the surplus huge page still exists after, even though it 775d1c3fb1fSNishanth Aravamudan * should just have been converted to a normal huge page. This 776d1c3fb1fSNishanth Aravamudan * does not leak memory, though, as the hugepage will be freed 777d1c3fb1fSNishanth Aravamudan * once it is out of use. It also does not allow the counters to 778d1c3fb1fSNishanth Aravamudan * go out of whack in adjust_pool_surplus() as we don't modify 779d1c3fb1fSNishanth Aravamudan * the node values until we've gotten the hugepage and only the 780d1c3fb1fSNishanth Aravamudan * per-node value is checked there. 781d1c3fb1fSNishanth Aravamudan */ 782d1c3fb1fSNishanth Aravamudan spin_lock(&hugetlb_lock); 783a5516438SAndi Kleen if (h->surplus_huge_pages >= h->nr_overcommit_huge_pages) { 784d1c3fb1fSNishanth Aravamudan spin_unlock(&hugetlb_lock); 785d1c3fb1fSNishanth Aravamudan return NULL; 786d1c3fb1fSNishanth Aravamudan } else { 787a5516438SAndi Kleen h->nr_huge_pages++; 788a5516438SAndi Kleen h->surplus_huge_pages++; 789d1c3fb1fSNishanth Aravamudan } 790d1c3fb1fSNishanth Aravamudan spin_unlock(&hugetlb_lock); 791d1c3fb1fSNishanth Aravamudan 792bf50bab2SNaoya Horiguchi if (nid == NUMA_NO_NODE) 793551883aeSNishanth Aravamudan page = alloc_pages(htlb_alloc_mask|__GFP_COMP| 794551883aeSNishanth Aravamudan __GFP_REPEAT|__GFP_NOWARN, 795a5516438SAndi Kleen huge_page_order(h)); 796bf50bab2SNaoya Horiguchi else 797bf50bab2SNaoya Horiguchi page = alloc_pages_exact_node(nid, 798bf50bab2SNaoya Horiguchi htlb_alloc_mask|__GFP_COMP|__GFP_THISNODE| 799bf50bab2SNaoya Horiguchi __GFP_REPEAT|__GFP_NOWARN, huge_page_order(h)); 800d1c3fb1fSNishanth Aravamudan 801caff3a2cSGerald Schaefer if (page && arch_prepare_hugepage(page)) { 802caff3a2cSGerald Schaefer __free_pages(page, huge_page_order(h)); 803caff3a2cSGerald Schaefer return NULL; 804caff3a2cSGerald Schaefer } 805caff3a2cSGerald Schaefer 8067893d1d5SAdam Litke spin_lock(&hugetlb_lock); 807d1c3fb1fSNishanth Aravamudan if (page) { 808bf50bab2SNaoya Horiguchi r_nid = page_to_nid(page); 809d1c3fb1fSNishanth Aravamudan set_compound_page_dtor(page, free_huge_page); 810d1c3fb1fSNishanth Aravamudan /* 811d1c3fb1fSNishanth Aravamudan * We incremented the global counters already 812d1c3fb1fSNishanth Aravamudan */ 813bf50bab2SNaoya Horiguchi h->nr_huge_pages_node[r_nid]++; 814bf50bab2SNaoya Horiguchi h->surplus_huge_pages_node[r_nid]++; 8153b116300SAdam Litke __count_vm_event(HTLB_BUDDY_PGALLOC); 816d1c3fb1fSNishanth Aravamudan } else { 817a5516438SAndi Kleen h->nr_huge_pages--; 818a5516438SAndi Kleen h->surplus_huge_pages--; 8193b116300SAdam Litke __count_vm_event(HTLB_BUDDY_PGALLOC_FAIL); 8207893d1d5SAdam Litke } 821d1c3fb1fSNishanth Aravamudan spin_unlock(&hugetlb_lock); 8227893d1d5SAdam Litke 8237893d1d5SAdam Litke return page; 8247893d1d5SAdam Litke } 8257893d1d5SAdam Litke 826e4e574b7SAdam Litke /* 827bf50bab2SNaoya Horiguchi * This allocation function is useful in the context where vma is irrelevant. 828bf50bab2SNaoya Horiguchi * E.g. soft-offlining uses this function because it only cares physical 829bf50bab2SNaoya Horiguchi * address of error page. 830bf50bab2SNaoya Horiguchi */ 831bf50bab2SNaoya Horiguchi struct page *alloc_huge_page_node(struct hstate *h, int nid) 832bf50bab2SNaoya Horiguchi { 833bf50bab2SNaoya Horiguchi struct page *page; 834bf50bab2SNaoya Horiguchi 835bf50bab2SNaoya Horiguchi spin_lock(&hugetlb_lock); 836bf50bab2SNaoya Horiguchi page = dequeue_huge_page_node(h, nid); 837bf50bab2SNaoya Horiguchi spin_unlock(&hugetlb_lock); 838bf50bab2SNaoya Horiguchi 839bf50bab2SNaoya Horiguchi if (!page) 840bf50bab2SNaoya Horiguchi page = alloc_buddy_huge_page(h, nid); 841bf50bab2SNaoya Horiguchi 842bf50bab2SNaoya Horiguchi return page; 843bf50bab2SNaoya Horiguchi } 844bf50bab2SNaoya Horiguchi 845bf50bab2SNaoya Horiguchi /* 84625985edcSLucas De Marchi * Increase the hugetlb pool such that it can accommodate a reservation 847e4e574b7SAdam Litke * of size 'delta'. 848e4e574b7SAdam Litke */ 849a5516438SAndi Kleen static int gather_surplus_pages(struct hstate *h, int delta) 850e4e574b7SAdam Litke { 851e4e574b7SAdam Litke struct list_head surplus_list; 852e4e574b7SAdam Litke struct page *page, *tmp; 853e4e574b7SAdam Litke int ret, i; 854e4e574b7SAdam Litke int needed, allocated; 855e4e574b7SAdam Litke 856a5516438SAndi Kleen needed = (h->resv_huge_pages + delta) - h->free_huge_pages; 857ac09b3a1SAdam Litke if (needed <= 0) { 858a5516438SAndi Kleen h->resv_huge_pages += delta; 859e4e574b7SAdam Litke return 0; 860ac09b3a1SAdam Litke } 861e4e574b7SAdam Litke 862e4e574b7SAdam Litke allocated = 0; 863e4e574b7SAdam Litke INIT_LIST_HEAD(&surplus_list); 864e4e574b7SAdam Litke 865e4e574b7SAdam Litke ret = -ENOMEM; 866e4e574b7SAdam Litke retry: 867e4e574b7SAdam Litke spin_unlock(&hugetlb_lock); 868e4e574b7SAdam Litke for (i = 0; i < needed; i++) { 869bf50bab2SNaoya Horiguchi page = alloc_buddy_huge_page(h, NUMA_NO_NODE); 870a9869b83SNaoya Horiguchi if (!page) 871e4e574b7SAdam Litke /* 872e4e574b7SAdam Litke * We were not able to allocate enough pages to 873e4e574b7SAdam Litke * satisfy the entire reservation so we free what 874e4e574b7SAdam Litke * we've allocated so far. 875e4e574b7SAdam Litke */ 876e4e574b7SAdam Litke goto free; 877e4e574b7SAdam Litke 878e4e574b7SAdam Litke list_add(&page->lru, &surplus_list); 879e4e574b7SAdam Litke } 880e4e574b7SAdam Litke allocated += needed; 881e4e574b7SAdam Litke 882e4e574b7SAdam Litke /* 883e4e574b7SAdam Litke * After retaking hugetlb_lock, we need to recalculate 'needed' 884e4e574b7SAdam Litke * because either resv_huge_pages or free_huge_pages may have changed. 885e4e574b7SAdam Litke */ 886e4e574b7SAdam Litke spin_lock(&hugetlb_lock); 887a5516438SAndi Kleen needed = (h->resv_huge_pages + delta) - 888a5516438SAndi Kleen (h->free_huge_pages + allocated); 889e4e574b7SAdam Litke if (needed > 0) 890e4e574b7SAdam Litke goto retry; 891e4e574b7SAdam Litke 892e4e574b7SAdam Litke /* 893e4e574b7SAdam Litke * The surplus_list now contains _at_least_ the number of extra pages 89425985edcSLucas De Marchi * needed to accommodate the reservation. Add the appropriate number 895e4e574b7SAdam Litke * of pages to the hugetlb pool and free the extras back to the buddy 896ac09b3a1SAdam Litke * allocator. Commit the entire reservation here to prevent another 897ac09b3a1SAdam Litke * process from stealing the pages as they are added to the pool but 898ac09b3a1SAdam Litke * before they are reserved. 899e4e574b7SAdam Litke */ 900e4e574b7SAdam Litke needed += allocated; 901a5516438SAndi Kleen h->resv_huge_pages += delta; 902e4e574b7SAdam Litke ret = 0; 903a9869b83SNaoya Horiguchi 90419fc3f0aSAdam Litke /* Free the needed pages to the hugetlb pool */ 90519fc3f0aSAdam Litke list_for_each_entry_safe(page, tmp, &surplus_list, lru) { 90619fc3f0aSAdam Litke if ((--needed) < 0) 90719fc3f0aSAdam Litke break; 90819fc3f0aSAdam Litke list_del(&page->lru); 909a9869b83SNaoya Horiguchi /* 910a9869b83SNaoya Horiguchi * This page is now managed by the hugetlb allocator and has 911a9869b83SNaoya Horiguchi * no users -- drop the buddy allocator's reference. 912a9869b83SNaoya Horiguchi */ 913a9869b83SNaoya Horiguchi put_page_testzero(page); 914a9869b83SNaoya Horiguchi VM_BUG_ON(page_count(page)); 915a5516438SAndi Kleen enqueue_huge_page(h, page); 91619fc3f0aSAdam Litke } 917b0365c8dSHillf Danton spin_unlock(&hugetlb_lock); 91819fc3f0aSAdam Litke 91919fc3f0aSAdam Litke /* Free unnecessary surplus pages to the buddy allocator */ 920a9869b83SNaoya Horiguchi free: 92119fc3f0aSAdam Litke if (!list_empty(&surplus_list)) { 922e4e574b7SAdam Litke list_for_each_entry_safe(page, tmp, &surplus_list, lru) { 923e4e574b7SAdam Litke list_del(&page->lru); 924a9869b83SNaoya Horiguchi put_page(page); 925a9869b83SNaoya Horiguchi } 926af767cbdSAdam Litke } 92719fc3f0aSAdam Litke spin_lock(&hugetlb_lock); 928e4e574b7SAdam Litke 929e4e574b7SAdam Litke return ret; 930e4e574b7SAdam Litke } 931e4e574b7SAdam Litke 932e4e574b7SAdam Litke /* 933e4e574b7SAdam Litke * When releasing a hugetlb pool reservation, any surplus pages that were 934e4e574b7SAdam Litke * allocated to satisfy the reservation must be explicitly freed if they were 935e4e574b7SAdam Litke * never used. 936685f3457SLee Schermerhorn * Called with hugetlb_lock held. 937e4e574b7SAdam Litke */ 938a5516438SAndi Kleen static void return_unused_surplus_pages(struct hstate *h, 939a5516438SAndi Kleen unsigned long unused_resv_pages) 940e4e574b7SAdam Litke { 941e4e574b7SAdam Litke unsigned long nr_pages; 942e4e574b7SAdam Litke 943ac09b3a1SAdam Litke /* Uncommit the reservation */ 944a5516438SAndi Kleen h->resv_huge_pages -= unused_resv_pages; 945ac09b3a1SAdam Litke 946aa888a74SAndi Kleen /* Cannot return gigantic pages currently */ 947aa888a74SAndi Kleen if (h->order >= MAX_ORDER) 948aa888a74SAndi Kleen return; 949aa888a74SAndi Kleen 950a5516438SAndi Kleen nr_pages = min(unused_resv_pages, h->surplus_huge_pages); 951e4e574b7SAdam Litke 952685f3457SLee Schermerhorn /* 953685f3457SLee Schermerhorn * We want to release as many surplus pages as possible, spread 9549b5e5d0fSLee Schermerhorn * evenly across all nodes with memory. Iterate across these nodes 9559b5e5d0fSLee Schermerhorn * until we can no longer free unreserved surplus pages. This occurs 9569b5e5d0fSLee Schermerhorn * when the nodes with surplus pages have no free pages. 9579b5e5d0fSLee Schermerhorn * free_pool_huge_page() will balance the the freed pages across the 9589b5e5d0fSLee Schermerhorn * on-line nodes with memory and will handle the hstate accounting. 959685f3457SLee Schermerhorn */ 960685f3457SLee Schermerhorn while (nr_pages--) { 9619b5e5d0fSLee Schermerhorn if (!free_pool_huge_page(h, &node_states[N_HIGH_MEMORY], 1)) 962685f3457SLee Schermerhorn break; 963e4e574b7SAdam Litke } 964e4e574b7SAdam Litke } 965e4e574b7SAdam Litke 966c37f9fb1SAndy Whitcroft /* 967c37f9fb1SAndy Whitcroft * Determine if the huge page at addr within the vma has an associated 968c37f9fb1SAndy Whitcroft * reservation. Where it does not we will need to logically increase 969c37f9fb1SAndy Whitcroft * reservation and actually increase quota before an allocation can occur. 970c37f9fb1SAndy Whitcroft * Where any new reservation would be required the reservation change is 971c37f9fb1SAndy Whitcroft * prepared, but not committed. Once the page has been quota'd allocated 972c37f9fb1SAndy Whitcroft * an instantiated the change should be committed via vma_commit_reservation. 973c37f9fb1SAndy Whitcroft * No action is required on failure. 974c37f9fb1SAndy Whitcroft */ 975e2f17d94SRoel Kluin static long vma_needs_reservation(struct hstate *h, 976a5516438SAndi Kleen struct vm_area_struct *vma, unsigned long addr) 977c37f9fb1SAndy Whitcroft { 978c37f9fb1SAndy Whitcroft struct address_space *mapping = vma->vm_file->f_mapping; 979c37f9fb1SAndy Whitcroft struct inode *inode = mapping->host; 980c37f9fb1SAndy Whitcroft 981f83a275dSMel Gorman if (vma->vm_flags & VM_MAYSHARE) { 982a5516438SAndi Kleen pgoff_t idx = vma_hugecache_offset(h, vma, addr); 983c37f9fb1SAndy Whitcroft return region_chg(&inode->i_mapping->private_list, 984c37f9fb1SAndy Whitcroft idx, idx + 1); 985c37f9fb1SAndy Whitcroft 98684afd99bSAndy Whitcroft } else if (!is_vma_resv_set(vma, HPAGE_RESV_OWNER)) { 987c37f9fb1SAndy Whitcroft return 1; 988c37f9fb1SAndy Whitcroft 98984afd99bSAndy Whitcroft } else { 990e2f17d94SRoel Kluin long err; 991a5516438SAndi Kleen pgoff_t idx = vma_hugecache_offset(h, vma, addr); 99284afd99bSAndy Whitcroft struct resv_map *reservations = vma_resv_map(vma); 99384afd99bSAndy Whitcroft 99484afd99bSAndy Whitcroft err = region_chg(&reservations->regions, idx, idx + 1); 99584afd99bSAndy Whitcroft if (err < 0) 99684afd99bSAndy Whitcroft return err; 997c37f9fb1SAndy Whitcroft return 0; 998c37f9fb1SAndy Whitcroft } 99984afd99bSAndy Whitcroft } 1000a5516438SAndi Kleen static void vma_commit_reservation(struct hstate *h, 1001a5516438SAndi Kleen struct vm_area_struct *vma, unsigned long addr) 1002c37f9fb1SAndy Whitcroft { 1003c37f9fb1SAndy Whitcroft struct address_space *mapping = vma->vm_file->f_mapping; 1004c37f9fb1SAndy Whitcroft struct inode *inode = mapping->host; 1005c37f9fb1SAndy Whitcroft 1006f83a275dSMel Gorman if (vma->vm_flags & VM_MAYSHARE) { 1007a5516438SAndi Kleen pgoff_t idx = vma_hugecache_offset(h, vma, addr); 1008c37f9fb1SAndy Whitcroft region_add(&inode->i_mapping->private_list, idx, idx + 1); 100984afd99bSAndy Whitcroft 101084afd99bSAndy Whitcroft } else if (is_vma_resv_set(vma, HPAGE_RESV_OWNER)) { 1011a5516438SAndi Kleen pgoff_t idx = vma_hugecache_offset(h, vma, addr); 101284afd99bSAndy Whitcroft struct resv_map *reservations = vma_resv_map(vma); 101384afd99bSAndy Whitcroft 101484afd99bSAndy Whitcroft /* Mark this page used in the map. */ 101584afd99bSAndy Whitcroft region_add(&reservations->regions, idx, idx + 1); 1016c37f9fb1SAndy Whitcroft } 1017c37f9fb1SAndy Whitcroft } 1018c37f9fb1SAndy Whitcroft 1019348ea204SAdam Litke static struct page *alloc_huge_page(struct vm_area_struct *vma, 102004f2cbe3SMel Gorman unsigned long addr, int avoid_reserve) 1021348ea204SAdam Litke { 1022a5516438SAndi Kleen struct hstate *h = hstate_vma(vma); 1023348ea204SAdam Litke struct page *page; 10242fc39cecSAdam Litke struct address_space *mapping = vma->vm_file->f_mapping; 1025a1e78772SMel Gorman struct inode *inode = mapping->host; 1026e2f17d94SRoel Kluin long chg; 10272fc39cecSAdam Litke 1028a1e78772SMel Gorman /* 1029a1e78772SMel Gorman * Processes that did not create the mapping will have no reserves and 1030a1e78772SMel Gorman * will not have accounted against quota. Check that the quota can be 1031a1e78772SMel Gorman * made before satisfying the allocation 1032c37f9fb1SAndy Whitcroft * MAP_NORESERVE mappings may also need pages and quota allocated 1033c37f9fb1SAndy Whitcroft * if no reserve mapping overlaps. 1034a1e78772SMel Gorman */ 1035a5516438SAndi Kleen chg = vma_needs_reservation(h, vma, addr); 1036c37f9fb1SAndy Whitcroft if (chg < 0) 1037e0dcd8a0SHugh Dickins return ERR_PTR(-VM_FAULT_OOM); 1038c37f9fb1SAndy Whitcroft if (chg) 1039a1e78772SMel Gorman if (hugetlb_get_quota(inode->i_mapping, chg)) 1040e0dcd8a0SHugh Dickins return ERR_PTR(-VM_FAULT_SIGBUS); 104190d8b7e6SAdam Litke 1042a1e78772SMel Gorman spin_lock(&hugetlb_lock); 1043a5516438SAndi Kleen page = dequeue_huge_page_vma(h, vma, addr, avoid_reserve); 1044a1e78772SMel Gorman spin_unlock(&hugetlb_lock); 1045a1e78772SMel Gorman 1046a1e78772SMel Gorman if (!page) { 1047bf50bab2SNaoya Horiguchi page = alloc_buddy_huge_page(h, NUMA_NO_NODE); 1048a1e78772SMel Gorman if (!page) { 1049a1e78772SMel Gorman hugetlb_put_quota(inode->i_mapping, chg); 10504a6018f7SMel Gorman return ERR_PTR(-VM_FAULT_SIGBUS); 1051a1e78772SMel Gorman } 1052a1e78772SMel Gorman } 1053a1e78772SMel Gorman 10542fc39cecSAdam Litke set_page_private(page, (unsigned long) mapping); 1055a1e78772SMel Gorman 1056a5516438SAndi Kleen vma_commit_reservation(h, vma, addr); 1057c37f9fb1SAndy Whitcroft 10587893d1d5SAdam Litke return page; 1059b45b5bd6SDavid Gibson } 1060b45b5bd6SDavid Gibson 106191f47662SCyrill Gorcunov int __weak alloc_bootmem_huge_page(struct hstate *h) 1062aa888a74SAndi Kleen { 1063aa888a74SAndi Kleen struct huge_bootmem_page *m; 10649b5e5d0fSLee Schermerhorn int nr_nodes = nodes_weight(node_states[N_HIGH_MEMORY]); 1065aa888a74SAndi Kleen 1066aa888a74SAndi Kleen while (nr_nodes) { 1067aa888a74SAndi Kleen void *addr; 1068aa888a74SAndi Kleen 1069aa888a74SAndi Kleen addr = __alloc_bootmem_node_nopanic( 10706ae11b27SLee Schermerhorn NODE_DATA(hstate_next_node_to_alloc(h, 10719b5e5d0fSLee Schermerhorn &node_states[N_HIGH_MEMORY])), 1072aa888a74SAndi Kleen huge_page_size(h), huge_page_size(h), 0); 1073aa888a74SAndi Kleen 1074aa888a74SAndi Kleen if (addr) { 1075aa888a74SAndi Kleen /* 1076aa888a74SAndi Kleen * Use the beginning of the huge page to store the 1077aa888a74SAndi Kleen * huge_bootmem_page struct (until gather_bootmem 1078aa888a74SAndi Kleen * puts them into the mem_map). 1079aa888a74SAndi Kleen */ 1080aa888a74SAndi Kleen m = addr; 1081aa888a74SAndi Kleen goto found; 1082aa888a74SAndi Kleen } 1083aa888a74SAndi Kleen nr_nodes--; 1084aa888a74SAndi Kleen } 1085aa888a74SAndi Kleen return 0; 1086aa888a74SAndi Kleen 1087aa888a74SAndi Kleen found: 1088aa888a74SAndi Kleen BUG_ON((unsigned long)virt_to_phys(m) & (huge_page_size(h) - 1)); 1089aa888a74SAndi Kleen /* Put them into a private list first because mem_map is not up yet */ 1090aa888a74SAndi Kleen list_add(&m->list, &huge_boot_pages); 1091aa888a74SAndi Kleen m->hstate = h; 1092aa888a74SAndi Kleen return 1; 1093aa888a74SAndi Kleen } 1094aa888a74SAndi Kleen 109518229df5SAndy Whitcroft static void prep_compound_huge_page(struct page *page, int order) 109618229df5SAndy Whitcroft { 109718229df5SAndy Whitcroft if (unlikely(order > (MAX_ORDER - 1))) 109818229df5SAndy Whitcroft prep_compound_gigantic_page(page, order); 109918229df5SAndy Whitcroft else 110018229df5SAndy Whitcroft prep_compound_page(page, order); 110118229df5SAndy Whitcroft } 110218229df5SAndy Whitcroft 1103aa888a74SAndi Kleen /* Put bootmem huge pages into the standard lists after mem_map is up */ 1104aa888a74SAndi Kleen static void __init gather_bootmem_prealloc(void) 1105aa888a74SAndi Kleen { 1106aa888a74SAndi Kleen struct huge_bootmem_page *m; 1107aa888a74SAndi Kleen 1108aa888a74SAndi Kleen list_for_each_entry(m, &huge_boot_pages, list) { 1109aa888a74SAndi Kleen struct hstate *h = m->hstate; 1110ee8f248dSBecky Bruce struct page *page; 1111ee8f248dSBecky Bruce 1112ee8f248dSBecky Bruce #ifdef CONFIG_HIGHMEM 1113ee8f248dSBecky Bruce page = pfn_to_page(m->phys >> PAGE_SHIFT); 1114ee8f248dSBecky Bruce free_bootmem_late((unsigned long)m, 1115ee8f248dSBecky Bruce sizeof(struct huge_bootmem_page)); 1116ee8f248dSBecky Bruce #else 1117ee8f248dSBecky Bruce page = virt_to_page(m); 1118ee8f248dSBecky Bruce #endif 1119aa888a74SAndi Kleen __ClearPageReserved(page); 1120aa888a74SAndi Kleen WARN_ON(page_count(page) != 1); 112118229df5SAndy Whitcroft prep_compound_huge_page(page, h->order); 1122aa888a74SAndi Kleen prep_new_huge_page(h, page, page_to_nid(page)); 1123b0320c7bSRafael Aquini /* 1124b0320c7bSRafael Aquini * If we had gigantic hugepages allocated at boot time, we need 1125b0320c7bSRafael Aquini * to restore the 'stolen' pages to totalram_pages in order to 1126b0320c7bSRafael Aquini * fix confusing memory reports from free(1) and another 1127b0320c7bSRafael Aquini * side-effects, like CommitLimit going negative. 1128b0320c7bSRafael Aquini */ 1129b0320c7bSRafael Aquini if (h->order > (MAX_ORDER - 1)) 1130b0320c7bSRafael Aquini totalram_pages += 1 << h->order; 1131aa888a74SAndi Kleen } 1132aa888a74SAndi Kleen } 1133aa888a74SAndi Kleen 11348faa8b07SAndi Kleen static void __init hugetlb_hstate_alloc_pages(struct hstate *h) 11351da177e4SLinus Torvalds { 11361da177e4SLinus Torvalds unsigned long i; 11371da177e4SLinus Torvalds 1138e5ff2159SAndi Kleen for (i = 0; i < h->max_huge_pages; ++i) { 1139aa888a74SAndi Kleen if (h->order >= MAX_ORDER) { 1140aa888a74SAndi Kleen if (!alloc_bootmem_huge_page(h)) 1141aa888a74SAndi Kleen break; 11429b5e5d0fSLee Schermerhorn } else if (!alloc_fresh_huge_page(h, 11439b5e5d0fSLee Schermerhorn &node_states[N_HIGH_MEMORY])) 11441da177e4SLinus Torvalds break; 11451da177e4SLinus Torvalds } 11468faa8b07SAndi Kleen h->max_huge_pages = i; 1147e5ff2159SAndi Kleen } 1148e5ff2159SAndi Kleen 1149e5ff2159SAndi Kleen static void __init hugetlb_init_hstates(void) 1150e5ff2159SAndi Kleen { 1151e5ff2159SAndi Kleen struct hstate *h; 1152e5ff2159SAndi Kleen 1153e5ff2159SAndi Kleen for_each_hstate(h) { 11548faa8b07SAndi Kleen /* oversize hugepages were init'ed in early boot */ 11558faa8b07SAndi Kleen if (h->order < MAX_ORDER) 11568faa8b07SAndi Kleen hugetlb_hstate_alloc_pages(h); 1157e5ff2159SAndi Kleen } 1158e5ff2159SAndi Kleen } 1159e5ff2159SAndi Kleen 11604abd32dbSAndi Kleen static char * __init memfmt(char *buf, unsigned long n) 11614abd32dbSAndi Kleen { 11624abd32dbSAndi Kleen if (n >= (1UL << 30)) 11634abd32dbSAndi Kleen sprintf(buf, "%lu GB", n >> 30); 11644abd32dbSAndi Kleen else if (n >= (1UL << 20)) 11654abd32dbSAndi Kleen sprintf(buf, "%lu MB", n >> 20); 11664abd32dbSAndi Kleen else 11674abd32dbSAndi Kleen sprintf(buf, "%lu KB", n >> 10); 11684abd32dbSAndi Kleen return buf; 11694abd32dbSAndi Kleen } 11704abd32dbSAndi Kleen 1171e5ff2159SAndi Kleen static void __init report_hugepages(void) 1172e5ff2159SAndi Kleen { 1173e5ff2159SAndi Kleen struct hstate *h; 1174e5ff2159SAndi Kleen 1175e5ff2159SAndi Kleen for_each_hstate(h) { 11764abd32dbSAndi Kleen char buf[32]; 11774abd32dbSAndi Kleen printk(KERN_INFO "HugeTLB registered %s page size, " 11784abd32dbSAndi Kleen "pre-allocated %ld pages\n", 11794abd32dbSAndi Kleen memfmt(buf, huge_page_size(h)), 11804abd32dbSAndi Kleen h->free_huge_pages); 1181e5ff2159SAndi Kleen } 1182e5ff2159SAndi Kleen } 1183e5ff2159SAndi Kleen 11841da177e4SLinus Torvalds #ifdef CONFIG_HIGHMEM 11856ae11b27SLee Schermerhorn static void try_to_free_low(struct hstate *h, unsigned long count, 11866ae11b27SLee Schermerhorn nodemask_t *nodes_allowed) 11871da177e4SLinus Torvalds { 11884415cc8dSChristoph Lameter int i; 11894415cc8dSChristoph Lameter 1190aa888a74SAndi Kleen if (h->order >= MAX_ORDER) 1191aa888a74SAndi Kleen return; 1192aa888a74SAndi Kleen 11936ae11b27SLee Schermerhorn for_each_node_mask(i, *nodes_allowed) { 11941da177e4SLinus Torvalds struct page *page, *next; 1195a5516438SAndi Kleen struct list_head *freel = &h->hugepage_freelists[i]; 1196a5516438SAndi Kleen list_for_each_entry_safe(page, next, freel, lru) { 1197a5516438SAndi Kleen if (count >= h->nr_huge_pages) 11986b0c880dSAdam Litke return; 11991da177e4SLinus Torvalds if (PageHighMem(page)) 12001da177e4SLinus Torvalds continue; 12011da177e4SLinus Torvalds list_del(&page->lru); 1202e5ff2159SAndi Kleen update_and_free_page(h, page); 1203a5516438SAndi Kleen h->free_huge_pages--; 1204a5516438SAndi Kleen h->free_huge_pages_node[page_to_nid(page)]--; 12051da177e4SLinus Torvalds } 12061da177e4SLinus Torvalds } 12071da177e4SLinus Torvalds } 12081da177e4SLinus Torvalds #else 12096ae11b27SLee Schermerhorn static inline void try_to_free_low(struct hstate *h, unsigned long count, 12106ae11b27SLee Schermerhorn nodemask_t *nodes_allowed) 12111da177e4SLinus Torvalds { 12121da177e4SLinus Torvalds } 12131da177e4SLinus Torvalds #endif 12141da177e4SLinus Torvalds 121520a0307cSWu Fengguang /* 121620a0307cSWu Fengguang * Increment or decrement surplus_huge_pages. Keep node-specific counters 121720a0307cSWu Fengguang * balanced by operating on them in a round-robin fashion. 121820a0307cSWu Fengguang * Returns 1 if an adjustment was made. 121920a0307cSWu Fengguang */ 12206ae11b27SLee Schermerhorn static int adjust_pool_surplus(struct hstate *h, nodemask_t *nodes_allowed, 12216ae11b27SLee Schermerhorn int delta) 122220a0307cSWu Fengguang { 1223e8c5c824SLee Schermerhorn int start_nid, next_nid; 122420a0307cSWu Fengguang int ret = 0; 122520a0307cSWu Fengguang 122620a0307cSWu Fengguang VM_BUG_ON(delta != -1 && delta != 1); 122720a0307cSWu Fengguang 1228e8c5c824SLee Schermerhorn if (delta < 0) 12296ae11b27SLee Schermerhorn start_nid = hstate_next_node_to_alloc(h, nodes_allowed); 1230e8c5c824SLee Schermerhorn else 12316ae11b27SLee Schermerhorn start_nid = hstate_next_node_to_free(h, nodes_allowed); 1232e8c5c824SLee Schermerhorn next_nid = start_nid; 1233e8c5c824SLee Schermerhorn 1234e8c5c824SLee Schermerhorn do { 1235e8c5c824SLee Schermerhorn int nid = next_nid; 1236e8c5c824SLee Schermerhorn if (delta < 0) { 1237e8c5c824SLee Schermerhorn /* 1238e8c5c824SLee Schermerhorn * To shrink on this node, there must be a surplus page 1239e8c5c824SLee Schermerhorn */ 12409a76db09SLee Schermerhorn if (!h->surplus_huge_pages_node[nid]) { 12416ae11b27SLee Schermerhorn next_nid = hstate_next_node_to_alloc(h, 12426ae11b27SLee Schermerhorn nodes_allowed); 124320a0307cSWu Fengguang continue; 1244e8c5c824SLee Schermerhorn } 12459a76db09SLee Schermerhorn } 1246e8c5c824SLee Schermerhorn if (delta > 0) { 1247e8c5c824SLee Schermerhorn /* 1248e8c5c824SLee Schermerhorn * Surplus cannot exceed the total number of pages 1249e8c5c824SLee Schermerhorn */ 1250e8c5c824SLee Schermerhorn if (h->surplus_huge_pages_node[nid] >= 12519a76db09SLee Schermerhorn h->nr_huge_pages_node[nid]) { 12526ae11b27SLee Schermerhorn next_nid = hstate_next_node_to_free(h, 12536ae11b27SLee Schermerhorn nodes_allowed); 125420a0307cSWu Fengguang continue; 1255e8c5c824SLee Schermerhorn } 12569a76db09SLee Schermerhorn } 125720a0307cSWu Fengguang 125820a0307cSWu Fengguang h->surplus_huge_pages += delta; 125920a0307cSWu Fengguang h->surplus_huge_pages_node[nid] += delta; 126020a0307cSWu Fengguang ret = 1; 126120a0307cSWu Fengguang break; 1262e8c5c824SLee Schermerhorn } while (next_nid != start_nid); 126320a0307cSWu Fengguang 126420a0307cSWu Fengguang return ret; 126520a0307cSWu Fengguang } 126620a0307cSWu Fengguang 1267a5516438SAndi Kleen #define persistent_huge_pages(h) (h->nr_huge_pages - h->surplus_huge_pages) 12686ae11b27SLee Schermerhorn static unsigned long set_max_huge_pages(struct hstate *h, unsigned long count, 12696ae11b27SLee Schermerhorn nodemask_t *nodes_allowed) 12701da177e4SLinus Torvalds { 12717893d1d5SAdam Litke unsigned long min_count, ret; 12721da177e4SLinus Torvalds 1273aa888a74SAndi Kleen if (h->order >= MAX_ORDER) 1274aa888a74SAndi Kleen return h->max_huge_pages; 1275aa888a74SAndi Kleen 12767893d1d5SAdam Litke /* 12777893d1d5SAdam Litke * Increase the pool size 12787893d1d5SAdam Litke * First take pages out of surplus state. Then make up the 12797893d1d5SAdam Litke * remaining difference by allocating fresh huge pages. 1280d1c3fb1fSNishanth Aravamudan * 1281d1c3fb1fSNishanth Aravamudan * We might race with alloc_buddy_huge_page() here and be unable 1282d1c3fb1fSNishanth Aravamudan * to convert a surplus huge page to a normal huge page. That is 1283d1c3fb1fSNishanth Aravamudan * not critical, though, it just means the overall size of the 1284d1c3fb1fSNishanth Aravamudan * pool might be one hugepage larger than it needs to be, but 1285d1c3fb1fSNishanth Aravamudan * within all the constraints specified by the sysctls. 12867893d1d5SAdam Litke */ 12871da177e4SLinus Torvalds spin_lock(&hugetlb_lock); 1288a5516438SAndi Kleen while (h->surplus_huge_pages && count > persistent_huge_pages(h)) { 12896ae11b27SLee Schermerhorn if (!adjust_pool_surplus(h, nodes_allowed, -1)) 12907893d1d5SAdam Litke break; 12917893d1d5SAdam Litke } 12927893d1d5SAdam Litke 1293a5516438SAndi Kleen while (count > persistent_huge_pages(h)) { 12947893d1d5SAdam Litke /* 12957893d1d5SAdam Litke * If this allocation races such that we no longer need the 12967893d1d5SAdam Litke * page, free_huge_page will handle it by freeing the page 12977893d1d5SAdam Litke * and reducing the surplus. 12987893d1d5SAdam Litke */ 12997893d1d5SAdam Litke spin_unlock(&hugetlb_lock); 13006ae11b27SLee Schermerhorn ret = alloc_fresh_huge_page(h, nodes_allowed); 13017893d1d5SAdam Litke spin_lock(&hugetlb_lock); 13027893d1d5SAdam Litke if (!ret) 13037893d1d5SAdam Litke goto out; 13047893d1d5SAdam Litke 1305536240f2SMel Gorman /* Bail for signals. Probably ctrl-c from user */ 1306536240f2SMel Gorman if (signal_pending(current)) 1307536240f2SMel Gorman goto out; 13087893d1d5SAdam Litke } 13097893d1d5SAdam Litke 13107893d1d5SAdam Litke /* 13117893d1d5SAdam Litke * Decrease the pool size 13127893d1d5SAdam Litke * First return free pages to the buddy allocator (being careful 13137893d1d5SAdam Litke * to keep enough around to satisfy reservations). Then place 13147893d1d5SAdam Litke * pages into surplus state as needed so the pool will shrink 13157893d1d5SAdam Litke * to the desired size as pages become free. 1316d1c3fb1fSNishanth Aravamudan * 1317d1c3fb1fSNishanth Aravamudan * By placing pages into the surplus state independent of the 1318d1c3fb1fSNishanth Aravamudan * overcommit value, we are allowing the surplus pool size to 1319d1c3fb1fSNishanth Aravamudan * exceed overcommit. There are few sane options here. Since 1320d1c3fb1fSNishanth Aravamudan * alloc_buddy_huge_page() is checking the global counter, 1321d1c3fb1fSNishanth Aravamudan * though, we'll note that we're not allowed to exceed surplus 1322d1c3fb1fSNishanth Aravamudan * and won't grow the pool anywhere else. Not until one of the 1323d1c3fb1fSNishanth Aravamudan * sysctls are changed, or the surplus pages go out of use. 13247893d1d5SAdam Litke */ 1325a5516438SAndi Kleen min_count = h->resv_huge_pages + h->nr_huge_pages - h->free_huge_pages; 13266b0c880dSAdam Litke min_count = max(count, min_count); 13276ae11b27SLee Schermerhorn try_to_free_low(h, min_count, nodes_allowed); 1328a5516438SAndi Kleen while (min_count < persistent_huge_pages(h)) { 13296ae11b27SLee Schermerhorn if (!free_pool_huge_page(h, nodes_allowed, 0)) 13301da177e4SLinus Torvalds break; 13311da177e4SLinus Torvalds } 1332a5516438SAndi Kleen while (count < persistent_huge_pages(h)) { 13336ae11b27SLee Schermerhorn if (!adjust_pool_surplus(h, nodes_allowed, 1)) 13347893d1d5SAdam Litke break; 13357893d1d5SAdam Litke } 13367893d1d5SAdam Litke out: 1337a5516438SAndi Kleen ret = persistent_huge_pages(h); 13381da177e4SLinus Torvalds spin_unlock(&hugetlb_lock); 13397893d1d5SAdam Litke return ret; 13401da177e4SLinus Torvalds } 13411da177e4SLinus Torvalds 1342a3437870SNishanth Aravamudan #define HSTATE_ATTR_RO(_name) \ 1343a3437870SNishanth Aravamudan static struct kobj_attribute _name##_attr = __ATTR_RO(_name) 1344a3437870SNishanth Aravamudan 1345a3437870SNishanth Aravamudan #define HSTATE_ATTR(_name) \ 1346a3437870SNishanth Aravamudan static struct kobj_attribute _name##_attr = \ 1347a3437870SNishanth Aravamudan __ATTR(_name, 0644, _name##_show, _name##_store) 1348a3437870SNishanth Aravamudan 1349a3437870SNishanth Aravamudan static struct kobject *hugepages_kobj; 1350a3437870SNishanth Aravamudan static struct kobject *hstate_kobjs[HUGE_MAX_HSTATE]; 1351a3437870SNishanth Aravamudan 13529a305230SLee Schermerhorn static struct hstate *kobj_to_node_hstate(struct kobject *kobj, int *nidp); 13539a305230SLee Schermerhorn 13549a305230SLee Schermerhorn static struct hstate *kobj_to_hstate(struct kobject *kobj, int *nidp) 1355a3437870SNishanth Aravamudan { 1356a3437870SNishanth Aravamudan int i; 13579a305230SLee Schermerhorn 1358a3437870SNishanth Aravamudan for (i = 0; i < HUGE_MAX_HSTATE; i++) 13599a305230SLee Schermerhorn if (hstate_kobjs[i] == kobj) { 13609a305230SLee Schermerhorn if (nidp) 13619a305230SLee Schermerhorn *nidp = NUMA_NO_NODE; 1362a3437870SNishanth Aravamudan return &hstates[i]; 13639a305230SLee Schermerhorn } 13649a305230SLee Schermerhorn 13659a305230SLee Schermerhorn return kobj_to_node_hstate(kobj, nidp); 1366a3437870SNishanth Aravamudan } 1367a3437870SNishanth Aravamudan 136806808b08SLee Schermerhorn static ssize_t nr_hugepages_show_common(struct kobject *kobj, 1369a3437870SNishanth Aravamudan struct kobj_attribute *attr, char *buf) 1370a3437870SNishanth Aravamudan { 13719a305230SLee Schermerhorn struct hstate *h; 13729a305230SLee Schermerhorn unsigned long nr_huge_pages; 13739a305230SLee Schermerhorn int nid; 13749a305230SLee Schermerhorn 13759a305230SLee Schermerhorn h = kobj_to_hstate(kobj, &nid); 13769a305230SLee Schermerhorn if (nid == NUMA_NO_NODE) 13779a305230SLee Schermerhorn nr_huge_pages = h->nr_huge_pages; 13789a305230SLee Schermerhorn else 13799a305230SLee Schermerhorn nr_huge_pages = h->nr_huge_pages_node[nid]; 13809a305230SLee Schermerhorn 13819a305230SLee Schermerhorn return sprintf(buf, "%lu\n", nr_huge_pages); 1382a3437870SNishanth Aravamudan } 1383adbe8726SEric B Munson 138406808b08SLee Schermerhorn static ssize_t nr_hugepages_store_common(bool obey_mempolicy, 138506808b08SLee Schermerhorn struct kobject *kobj, struct kobj_attribute *attr, 138606808b08SLee Schermerhorn const char *buf, size_t len) 1387a3437870SNishanth Aravamudan { 1388a3437870SNishanth Aravamudan int err; 13899a305230SLee Schermerhorn int nid; 139006808b08SLee Schermerhorn unsigned long count; 13919a305230SLee Schermerhorn struct hstate *h; 1392bad44b5bSDavid Rientjes NODEMASK_ALLOC(nodemask_t, nodes_allowed, GFP_KERNEL | __GFP_NORETRY); 1393a3437870SNishanth Aravamudan 139406808b08SLee Schermerhorn err = strict_strtoul(buf, 10, &count); 139573ae31e5SEric B Munson if (err) 1396adbe8726SEric B Munson goto out; 1397a3437870SNishanth Aravamudan 13989a305230SLee Schermerhorn h = kobj_to_hstate(kobj, &nid); 1399adbe8726SEric B Munson if (h->order >= MAX_ORDER) { 1400adbe8726SEric B Munson err = -EINVAL; 1401adbe8726SEric B Munson goto out; 1402adbe8726SEric B Munson } 1403adbe8726SEric B Munson 14049a305230SLee Schermerhorn if (nid == NUMA_NO_NODE) { 14059a305230SLee Schermerhorn /* 14069a305230SLee Schermerhorn * global hstate attribute 14079a305230SLee Schermerhorn */ 14089a305230SLee Schermerhorn if (!(obey_mempolicy && 14099a305230SLee Schermerhorn init_nodemask_of_mempolicy(nodes_allowed))) { 141006808b08SLee Schermerhorn NODEMASK_FREE(nodes_allowed); 14119a305230SLee Schermerhorn nodes_allowed = &node_states[N_HIGH_MEMORY]; 141206808b08SLee Schermerhorn } 14139a305230SLee Schermerhorn } else if (nodes_allowed) { 14149a305230SLee Schermerhorn /* 14159a305230SLee Schermerhorn * per node hstate attribute: adjust count to global, 14169a305230SLee Schermerhorn * but restrict alloc/free to the specified node. 14179a305230SLee Schermerhorn */ 14189a305230SLee Schermerhorn count += h->nr_huge_pages - h->nr_huge_pages_node[nid]; 14199a305230SLee Schermerhorn init_nodemask_of_node(nodes_allowed, nid); 14209a305230SLee Schermerhorn } else 14219a305230SLee Schermerhorn nodes_allowed = &node_states[N_HIGH_MEMORY]; 14229a305230SLee Schermerhorn 142306808b08SLee Schermerhorn h->max_huge_pages = set_max_huge_pages(h, count, nodes_allowed); 1424a3437870SNishanth Aravamudan 14259b5e5d0fSLee Schermerhorn if (nodes_allowed != &node_states[N_HIGH_MEMORY]) 142606808b08SLee Schermerhorn NODEMASK_FREE(nodes_allowed); 142706808b08SLee Schermerhorn 142806808b08SLee Schermerhorn return len; 1429adbe8726SEric B Munson out: 1430adbe8726SEric B Munson NODEMASK_FREE(nodes_allowed); 1431adbe8726SEric B Munson return err; 143206808b08SLee Schermerhorn } 143306808b08SLee Schermerhorn 143406808b08SLee Schermerhorn static ssize_t nr_hugepages_show(struct kobject *kobj, 143506808b08SLee Schermerhorn struct kobj_attribute *attr, char *buf) 143606808b08SLee Schermerhorn { 143706808b08SLee Schermerhorn return nr_hugepages_show_common(kobj, attr, buf); 143806808b08SLee Schermerhorn } 143906808b08SLee Schermerhorn 144006808b08SLee Schermerhorn static ssize_t nr_hugepages_store(struct kobject *kobj, 144106808b08SLee Schermerhorn struct kobj_attribute *attr, const char *buf, size_t len) 144206808b08SLee Schermerhorn { 144306808b08SLee Schermerhorn return nr_hugepages_store_common(false, kobj, attr, buf, len); 1444a3437870SNishanth Aravamudan } 1445a3437870SNishanth Aravamudan HSTATE_ATTR(nr_hugepages); 1446a3437870SNishanth Aravamudan 144706808b08SLee Schermerhorn #ifdef CONFIG_NUMA 144806808b08SLee Schermerhorn 144906808b08SLee Schermerhorn /* 145006808b08SLee Schermerhorn * hstate attribute for optionally mempolicy-based constraint on persistent 145106808b08SLee Schermerhorn * huge page alloc/free. 145206808b08SLee Schermerhorn */ 145306808b08SLee Schermerhorn static ssize_t nr_hugepages_mempolicy_show(struct kobject *kobj, 145406808b08SLee Schermerhorn struct kobj_attribute *attr, char *buf) 145506808b08SLee Schermerhorn { 145606808b08SLee Schermerhorn return nr_hugepages_show_common(kobj, attr, buf); 145706808b08SLee Schermerhorn } 145806808b08SLee Schermerhorn 145906808b08SLee Schermerhorn static ssize_t nr_hugepages_mempolicy_store(struct kobject *kobj, 146006808b08SLee Schermerhorn struct kobj_attribute *attr, const char *buf, size_t len) 146106808b08SLee Schermerhorn { 146206808b08SLee Schermerhorn return nr_hugepages_store_common(true, kobj, attr, buf, len); 146306808b08SLee Schermerhorn } 146406808b08SLee Schermerhorn HSTATE_ATTR(nr_hugepages_mempolicy); 146506808b08SLee Schermerhorn #endif 146606808b08SLee Schermerhorn 146706808b08SLee Schermerhorn 1468a3437870SNishanth Aravamudan static ssize_t nr_overcommit_hugepages_show(struct kobject *kobj, 1469a3437870SNishanth Aravamudan struct kobj_attribute *attr, char *buf) 1470a3437870SNishanth Aravamudan { 14719a305230SLee Schermerhorn struct hstate *h = kobj_to_hstate(kobj, NULL); 1472a3437870SNishanth Aravamudan return sprintf(buf, "%lu\n", h->nr_overcommit_huge_pages); 1473a3437870SNishanth Aravamudan } 1474adbe8726SEric B Munson 1475a3437870SNishanth Aravamudan static ssize_t nr_overcommit_hugepages_store(struct kobject *kobj, 1476a3437870SNishanth Aravamudan struct kobj_attribute *attr, const char *buf, size_t count) 1477a3437870SNishanth Aravamudan { 1478a3437870SNishanth Aravamudan int err; 1479a3437870SNishanth Aravamudan unsigned long input; 14809a305230SLee Schermerhorn struct hstate *h = kobj_to_hstate(kobj, NULL); 1481a3437870SNishanth Aravamudan 1482adbe8726SEric B Munson if (h->order >= MAX_ORDER) 1483adbe8726SEric B Munson return -EINVAL; 1484adbe8726SEric B Munson 1485a3437870SNishanth Aravamudan err = strict_strtoul(buf, 10, &input); 1486a3437870SNishanth Aravamudan if (err) 148773ae31e5SEric B Munson return err; 1488a3437870SNishanth Aravamudan 1489a3437870SNishanth Aravamudan spin_lock(&hugetlb_lock); 1490a3437870SNishanth Aravamudan h->nr_overcommit_huge_pages = input; 1491a3437870SNishanth Aravamudan spin_unlock(&hugetlb_lock); 1492a3437870SNishanth Aravamudan 1493a3437870SNishanth Aravamudan return count; 1494a3437870SNishanth Aravamudan } 1495a3437870SNishanth Aravamudan HSTATE_ATTR(nr_overcommit_hugepages); 1496a3437870SNishanth Aravamudan 1497a3437870SNishanth Aravamudan static ssize_t free_hugepages_show(struct kobject *kobj, 1498a3437870SNishanth Aravamudan struct kobj_attribute *attr, char *buf) 1499a3437870SNishanth Aravamudan { 15009a305230SLee Schermerhorn struct hstate *h; 15019a305230SLee Schermerhorn unsigned long free_huge_pages; 15029a305230SLee Schermerhorn int nid; 15039a305230SLee Schermerhorn 15049a305230SLee Schermerhorn h = kobj_to_hstate(kobj, &nid); 15059a305230SLee Schermerhorn if (nid == NUMA_NO_NODE) 15069a305230SLee Schermerhorn free_huge_pages = h->free_huge_pages; 15079a305230SLee Schermerhorn else 15089a305230SLee Schermerhorn free_huge_pages = h->free_huge_pages_node[nid]; 15099a305230SLee Schermerhorn 15109a305230SLee Schermerhorn return sprintf(buf, "%lu\n", free_huge_pages); 1511a3437870SNishanth Aravamudan } 1512a3437870SNishanth Aravamudan HSTATE_ATTR_RO(free_hugepages); 1513a3437870SNishanth Aravamudan 1514a3437870SNishanth Aravamudan static ssize_t resv_hugepages_show(struct kobject *kobj, 1515a3437870SNishanth Aravamudan struct kobj_attribute *attr, char *buf) 1516a3437870SNishanth Aravamudan { 15179a305230SLee Schermerhorn struct hstate *h = kobj_to_hstate(kobj, NULL); 1518a3437870SNishanth Aravamudan return sprintf(buf, "%lu\n", h->resv_huge_pages); 1519a3437870SNishanth Aravamudan } 1520a3437870SNishanth Aravamudan HSTATE_ATTR_RO(resv_hugepages); 1521a3437870SNishanth Aravamudan 1522a3437870SNishanth Aravamudan static ssize_t surplus_hugepages_show(struct kobject *kobj, 1523a3437870SNishanth Aravamudan struct kobj_attribute *attr, char *buf) 1524a3437870SNishanth Aravamudan { 15259a305230SLee Schermerhorn struct hstate *h; 15269a305230SLee Schermerhorn unsigned long surplus_huge_pages; 15279a305230SLee Schermerhorn int nid; 15289a305230SLee Schermerhorn 15299a305230SLee Schermerhorn h = kobj_to_hstate(kobj, &nid); 15309a305230SLee Schermerhorn if (nid == NUMA_NO_NODE) 15319a305230SLee Schermerhorn surplus_huge_pages = h->surplus_huge_pages; 15329a305230SLee Schermerhorn else 15339a305230SLee Schermerhorn surplus_huge_pages = h->surplus_huge_pages_node[nid]; 15349a305230SLee Schermerhorn 15359a305230SLee Schermerhorn return sprintf(buf, "%lu\n", surplus_huge_pages); 1536a3437870SNishanth Aravamudan } 1537a3437870SNishanth Aravamudan HSTATE_ATTR_RO(surplus_hugepages); 1538a3437870SNishanth Aravamudan 1539a3437870SNishanth Aravamudan static struct attribute *hstate_attrs[] = { 1540a3437870SNishanth Aravamudan &nr_hugepages_attr.attr, 1541a3437870SNishanth Aravamudan &nr_overcommit_hugepages_attr.attr, 1542a3437870SNishanth Aravamudan &free_hugepages_attr.attr, 1543a3437870SNishanth Aravamudan &resv_hugepages_attr.attr, 1544a3437870SNishanth Aravamudan &surplus_hugepages_attr.attr, 154506808b08SLee Schermerhorn #ifdef CONFIG_NUMA 154606808b08SLee Schermerhorn &nr_hugepages_mempolicy_attr.attr, 154706808b08SLee Schermerhorn #endif 1548a3437870SNishanth Aravamudan NULL, 1549a3437870SNishanth Aravamudan }; 1550a3437870SNishanth Aravamudan 1551a3437870SNishanth Aravamudan static struct attribute_group hstate_attr_group = { 1552a3437870SNishanth Aravamudan .attrs = hstate_attrs, 1553a3437870SNishanth Aravamudan }; 1554a3437870SNishanth Aravamudan 1555094e9539SJeff Mahoney static int hugetlb_sysfs_add_hstate(struct hstate *h, struct kobject *parent, 15569a305230SLee Schermerhorn struct kobject **hstate_kobjs, 15579a305230SLee Schermerhorn struct attribute_group *hstate_attr_group) 1558a3437870SNishanth Aravamudan { 1559a3437870SNishanth Aravamudan int retval; 15609a305230SLee Schermerhorn int hi = h - hstates; 1561a3437870SNishanth Aravamudan 15629a305230SLee Schermerhorn hstate_kobjs[hi] = kobject_create_and_add(h->name, parent); 15639a305230SLee Schermerhorn if (!hstate_kobjs[hi]) 1564a3437870SNishanth Aravamudan return -ENOMEM; 1565a3437870SNishanth Aravamudan 15669a305230SLee Schermerhorn retval = sysfs_create_group(hstate_kobjs[hi], hstate_attr_group); 1567a3437870SNishanth Aravamudan if (retval) 15689a305230SLee Schermerhorn kobject_put(hstate_kobjs[hi]); 1569a3437870SNishanth Aravamudan 1570a3437870SNishanth Aravamudan return retval; 1571a3437870SNishanth Aravamudan } 1572a3437870SNishanth Aravamudan 1573a3437870SNishanth Aravamudan static void __init hugetlb_sysfs_init(void) 1574a3437870SNishanth Aravamudan { 1575a3437870SNishanth Aravamudan struct hstate *h; 1576a3437870SNishanth Aravamudan int err; 1577a3437870SNishanth Aravamudan 1578a3437870SNishanth Aravamudan hugepages_kobj = kobject_create_and_add("hugepages", mm_kobj); 1579a3437870SNishanth Aravamudan if (!hugepages_kobj) 1580a3437870SNishanth Aravamudan return; 1581a3437870SNishanth Aravamudan 1582a3437870SNishanth Aravamudan for_each_hstate(h) { 15839a305230SLee Schermerhorn err = hugetlb_sysfs_add_hstate(h, hugepages_kobj, 15849a305230SLee Schermerhorn hstate_kobjs, &hstate_attr_group); 1585a3437870SNishanth Aravamudan if (err) 1586a3437870SNishanth Aravamudan printk(KERN_ERR "Hugetlb: Unable to add hstate %s", 1587a3437870SNishanth Aravamudan h->name); 1588a3437870SNishanth Aravamudan } 1589a3437870SNishanth Aravamudan } 1590a3437870SNishanth Aravamudan 15919a305230SLee Schermerhorn #ifdef CONFIG_NUMA 15929a305230SLee Schermerhorn 15939a305230SLee Schermerhorn /* 15949a305230SLee Schermerhorn * node_hstate/s - associate per node hstate attributes, via their kobjects, 159510fbcf4cSKay Sievers * with node devices in node_devices[] using a parallel array. The array 159610fbcf4cSKay Sievers * index of a node device or _hstate == node id. 159710fbcf4cSKay Sievers * This is here to avoid any static dependency of the node device driver, in 15989a305230SLee Schermerhorn * the base kernel, on the hugetlb module. 15999a305230SLee Schermerhorn */ 16009a305230SLee Schermerhorn struct node_hstate { 16019a305230SLee Schermerhorn struct kobject *hugepages_kobj; 16029a305230SLee Schermerhorn struct kobject *hstate_kobjs[HUGE_MAX_HSTATE]; 16039a305230SLee Schermerhorn }; 16049a305230SLee Schermerhorn struct node_hstate node_hstates[MAX_NUMNODES]; 16059a305230SLee Schermerhorn 16069a305230SLee Schermerhorn /* 160710fbcf4cSKay Sievers * A subset of global hstate attributes for node devices 16089a305230SLee Schermerhorn */ 16099a305230SLee Schermerhorn static struct attribute *per_node_hstate_attrs[] = { 16109a305230SLee Schermerhorn &nr_hugepages_attr.attr, 16119a305230SLee Schermerhorn &free_hugepages_attr.attr, 16129a305230SLee Schermerhorn &surplus_hugepages_attr.attr, 16139a305230SLee Schermerhorn NULL, 16149a305230SLee Schermerhorn }; 16159a305230SLee Schermerhorn 16169a305230SLee Schermerhorn static struct attribute_group per_node_hstate_attr_group = { 16179a305230SLee Schermerhorn .attrs = per_node_hstate_attrs, 16189a305230SLee Schermerhorn }; 16199a305230SLee Schermerhorn 16209a305230SLee Schermerhorn /* 162110fbcf4cSKay Sievers * kobj_to_node_hstate - lookup global hstate for node device hstate attr kobj. 16229a305230SLee Schermerhorn * Returns node id via non-NULL nidp. 16239a305230SLee Schermerhorn */ 16249a305230SLee Schermerhorn static struct hstate *kobj_to_node_hstate(struct kobject *kobj, int *nidp) 16259a305230SLee Schermerhorn { 16269a305230SLee Schermerhorn int nid; 16279a305230SLee Schermerhorn 16289a305230SLee Schermerhorn for (nid = 0; nid < nr_node_ids; nid++) { 16299a305230SLee Schermerhorn struct node_hstate *nhs = &node_hstates[nid]; 16309a305230SLee Schermerhorn int i; 16319a305230SLee Schermerhorn for (i = 0; i < HUGE_MAX_HSTATE; i++) 16329a305230SLee Schermerhorn if (nhs->hstate_kobjs[i] == kobj) { 16339a305230SLee Schermerhorn if (nidp) 16349a305230SLee Schermerhorn *nidp = nid; 16359a305230SLee Schermerhorn return &hstates[i]; 16369a305230SLee Schermerhorn } 16379a305230SLee Schermerhorn } 16389a305230SLee Schermerhorn 16399a305230SLee Schermerhorn BUG(); 16409a305230SLee Schermerhorn return NULL; 16419a305230SLee Schermerhorn } 16429a305230SLee Schermerhorn 16439a305230SLee Schermerhorn /* 164410fbcf4cSKay Sievers * Unregister hstate attributes from a single node device. 16459a305230SLee Schermerhorn * No-op if no hstate attributes attached. 16469a305230SLee Schermerhorn */ 16479a305230SLee Schermerhorn void hugetlb_unregister_node(struct node *node) 16489a305230SLee Schermerhorn { 16499a305230SLee Schermerhorn struct hstate *h; 165010fbcf4cSKay Sievers struct node_hstate *nhs = &node_hstates[node->dev.id]; 16519a305230SLee Schermerhorn 16529a305230SLee Schermerhorn if (!nhs->hugepages_kobj) 16539b5e5d0fSLee Schermerhorn return; /* no hstate attributes */ 16549a305230SLee Schermerhorn 16559a305230SLee Schermerhorn for_each_hstate(h) 16569a305230SLee Schermerhorn if (nhs->hstate_kobjs[h - hstates]) { 16579a305230SLee Schermerhorn kobject_put(nhs->hstate_kobjs[h - hstates]); 16589a305230SLee Schermerhorn nhs->hstate_kobjs[h - hstates] = NULL; 16599a305230SLee Schermerhorn } 16609a305230SLee Schermerhorn 16619a305230SLee Schermerhorn kobject_put(nhs->hugepages_kobj); 16629a305230SLee Schermerhorn nhs->hugepages_kobj = NULL; 16639a305230SLee Schermerhorn } 16649a305230SLee Schermerhorn 16659a305230SLee Schermerhorn /* 166610fbcf4cSKay Sievers * hugetlb module exit: unregister hstate attributes from node devices 16679a305230SLee Schermerhorn * that have them. 16689a305230SLee Schermerhorn */ 16699a305230SLee Schermerhorn static void hugetlb_unregister_all_nodes(void) 16709a305230SLee Schermerhorn { 16719a305230SLee Schermerhorn int nid; 16729a305230SLee Schermerhorn 16739a305230SLee Schermerhorn /* 167410fbcf4cSKay Sievers * disable node device registrations. 16759a305230SLee Schermerhorn */ 16769a305230SLee Schermerhorn register_hugetlbfs_with_node(NULL, NULL); 16779a305230SLee Schermerhorn 16789a305230SLee Schermerhorn /* 16799a305230SLee Schermerhorn * remove hstate attributes from any nodes that have them. 16809a305230SLee Schermerhorn */ 16819a305230SLee Schermerhorn for (nid = 0; nid < nr_node_ids; nid++) 16829a305230SLee Schermerhorn hugetlb_unregister_node(&node_devices[nid]); 16839a305230SLee Schermerhorn } 16849a305230SLee Schermerhorn 16859a305230SLee Schermerhorn /* 168610fbcf4cSKay Sievers * Register hstate attributes for a single node device. 16879a305230SLee Schermerhorn * No-op if attributes already registered. 16889a305230SLee Schermerhorn */ 16899a305230SLee Schermerhorn void hugetlb_register_node(struct node *node) 16909a305230SLee Schermerhorn { 16919a305230SLee Schermerhorn struct hstate *h; 169210fbcf4cSKay Sievers struct node_hstate *nhs = &node_hstates[node->dev.id]; 16939a305230SLee Schermerhorn int err; 16949a305230SLee Schermerhorn 16959a305230SLee Schermerhorn if (nhs->hugepages_kobj) 16969a305230SLee Schermerhorn return; /* already allocated */ 16979a305230SLee Schermerhorn 16989a305230SLee Schermerhorn nhs->hugepages_kobj = kobject_create_and_add("hugepages", 169910fbcf4cSKay Sievers &node->dev.kobj); 17009a305230SLee Schermerhorn if (!nhs->hugepages_kobj) 17019a305230SLee Schermerhorn return; 17029a305230SLee Schermerhorn 17039a305230SLee Schermerhorn for_each_hstate(h) { 17049a305230SLee Schermerhorn err = hugetlb_sysfs_add_hstate(h, nhs->hugepages_kobj, 17059a305230SLee Schermerhorn nhs->hstate_kobjs, 17069a305230SLee Schermerhorn &per_node_hstate_attr_group); 17079a305230SLee Schermerhorn if (err) { 17089a305230SLee Schermerhorn printk(KERN_ERR "Hugetlb: Unable to add hstate %s" 17099a305230SLee Schermerhorn " for node %d\n", 171010fbcf4cSKay Sievers h->name, node->dev.id); 17119a305230SLee Schermerhorn hugetlb_unregister_node(node); 17129a305230SLee Schermerhorn break; 17139a305230SLee Schermerhorn } 17149a305230SLee Schermerhorn } 17159a305230SLee Schermerhorn } 17169a305230SLee Schermerhorn 17179a305230SLee Schermerhorn /* 17189b5e5d0fSLee Schermerhorn * hugetlb init time: register hstate attributes for all registered node 171910fbcf4cSKay Sievers * devices of nodes that have memory. All on-line nodes should have 172010fbcf4cSKay Sievers * registered their associated device by this time. 17219a305230SLee Schermerhorn */ 17229a305230SLee Schermerhorn static void hugetlb_register_all_nodes(void) 17239a305230SLee Schermerhorn { 17249a305230SLee Schermerhorn int nid; 17259a305230SLee Schermerhorn 17269b5e5d0fSLee Schermerhorn for_each_node_state(nid, N_HIGH_MEMORY) { 17279a305230SLee Schermerhorn struct node *node = &node_devices[nid]; 172810fbcf4cSKay Sievers if (node->dev.id == nid) 17299a305230SLee Schermerhorn hugetlb_register_node(node); 17309a305230SLee Schermerhorn } 17319a305230SLee Schermerhorn 17329a305230SLee Schermerhorn /* 173310fbcf4cSKay Sievers * Let the node device driver know we're here so it can 17349a305230SLee Schermerhorn * [un]register hstate attributes on node hotplug. 17359a305230SLee Schermerhorn */ 17369a305230SLee Schermerhorn register_hugetlbfs_with_node(hugetlb_register_node, 17379a305230SLee Schermerhorn hugetlb_unregister_node); 17389a305230SLee Schermerhorn } 17399a305230SLee Schermerhorn #else /* !CONFIG_NUMA */ 17409a305230SLee Schermerhorn 17419a305230SLee Schermerhorn static struct hstate *kobj_to_node_hstate(struct kobject *kobj, int *nidp) 17429a305230SLee Schermerhorn { 17439a305230SLee Schermerhorn BUG(); 17449a305230SLee Schermerhorn if (nidp) 17459a305230SLee Schermerhorn *nidp = -1; 17469a305230SLee Schermerhorn return NULL; 17479a305230SLee Schermerhorn } 17489a305230SLee Schermerhorn 17499a305230SLee Schermerhorn static void hugetlb_unregister_all_nodes(void) { } 17509a305230SLee Schermerhorn 17519a305230SLee Schermerhorn static void hugetlb_register_all_nodes(void) { } 17529a305230SLee Schermerhorn 17539a305230SLee Schermerhorn #endif 17549a305230SLee Schermerhorn 1755a3437870SNishanth Aravamudan static void __exit hugetlb_exit(void) 1756a3437870SNishanth Aravamudan { 1757a3437870SNishanth Aravamudan struct hstate *h; 1758a3437870SNishanth Aravamudan 17599a305230SLee Schermerhorn hugetlb_unregister_all_nodes(); 17609a305230SLee Schermerhorn 1761a3437870SNishanth Aravamudan for_each_hstate(h) { 1762a3437870SNishanth Aravamudan kobject_put(hstate_kobjs[h - hstates]); 1763a3437870SNishanth Aravamudan } 1764a3437870SNishanth Aravamudan 1765a3437870SNishanth Aravamudan kobject_put(hugepages_kobj); 1766a3437870SNishanth Aravamudan } 1767a3437870SNishanth Aravamudan module_exit(hugetlb_exit); 1768a3437870SNishanth Aravamudan 1769a3437870SNishanth Aravamudan static int __init hugetlb_init(void) 1770a3437870SNishanth Aravamudan { 17710ef89d25SBenjamin Herrenschmidt /* Some platform decide whether they support huge pages at boot 17720ef89d25SBenjamin Herrenschmidt * time. On these, such as powerpc, HPAGE_SHIFT is set to 0 when 17730ef89d25SBenjamin Herrenschmidt * there is no such support 17740ef89d25SBenjamin Herrenschmidt */ 17750ef89d25SBenjamin Herrenschmidt if (HPAGE_SHIFT == 0) 17760ef89d25SBenjamin Herrenschmidt return 0; 1777a3437870SNishanth Aravamudan 1778e11bfbfcSNick Piggin if (!size_to_hstate(default_hstate_size)) { 1779e11bfbfcSNick Piggin default_hstate_size = HPAGE_SIZE; 1780e11bfbfcSNick Piggin if (!size_to_hstate(default_hstate_size)) 1781a3437870SNishanth Aravamudan hugetlb_add_hstate(HUGETLB_PAGE_ORDER); 1782a3437870SNishanth Aravamudan } 1783e11bfbfcSNick Piggin default_hstate_idx = size_to_hstate(default_hstate_size) - hstates; 1784e11bfbfcSNick Piggin if (default_hstate_max_huge_pages) 1785e11bfbfcSNick Piggin default_hstate.max_huge_pages = default_hstate_max_huge_pages; 1786a3437870SNishanth Aravamudan 1787a3437870SNishanth Aravamudan hugetlb_init_hstates(); 1788a3437870SNishanth Aravamudan 1789aa888a74SAndi Kleen gather_bootmem_prealloc(); 1790aa888a74SAndi Kleen 1791a3437870SNishanth Aravamudan report_hugepages(); 1792a3437870SNishanth Aravamudan 1793a3437870SNishanth Aravamudan hugetlb_sysfs_init(); 1794a3437870SNishanth Aravamudan 17959a305230SLee Schermerhorn hugetlb_register_all_nodes(); 17969a305230SLee Schermerhorn 1797a3437870SNishanth Aravamudan return 0; 1798a3437870SNishanth Aravamudan } 1799a3437870SNishanth Aravamudan module_init(hugetlb_init); 1800a3437870SNishanth Aravamudan 1801a3437870SNishanth Aravamudan /* Should be called on processing a hugepagesz=... option */ 1802a3437870SNishanth Aravamudan void __init hugetlb_add_hstate(unsigned order) 1803a3437870SNishanth Aravamudan { 1804a3437870SNishanth Aravamudan struct hstate *h; 18058faa8b07SAndi Kleen unsigned long i; 18068faa8b07SAndi Kleen 1807a3437870SNishanth Aravamudan if (size_to_hstate(PAGE_SIZE << order)) { 1808a3437870SNishanth Aravamudan printk(KERN_WARNING "hugepagesz= specified twice, ignoring\n"); 1809a3437870SNishanth Aravamudan return; 1810a3437870SNishanth Aravamudan } 1811a3437870SNishanth Aravamudan BUG_ON(max_hstate >= HUGE_MAX_HSTATE); 1812a3437870SNishanth Aravamudan BUG_ON(order == 0); 1813a3437870SNishanth Aravamudan h = &hstates[max_hstate++]; 1814a3437870SNishanth Aravamudan h->order = order; 1815a3437870SNishanth Aravamudan h->mask = ~((1ULL << (order + PAGE_SHIFT)) - 1); 18168faa8b07SAndi Kleen h->nr_huge_pages = 0; 18178faa8b07SAndi Kleen h->free_huge_pages = 0; 18188faa8b07SAndi Kleen for (i = 0; i < MAX_NUMNODES; ++i) 18198faa8b07SAndi Kleen INIT_LIST_HEAD(&h->hugepage_freelists[i]); 18209b5e5d0fSLee Schermerhorn h->next_nid_to_alloc = first_node(node_states[N_HIGH_MEMORY]); 18219b5e5d0fSLee Schermerhorn h->next_nid_to_free = first_node(node_states[N_HIGH_MEMORY]); 1822a3437870SNishanth Aravamudan snprintf(h->name, HSTATE_NAME_LEN, "hugepages-%lukB", 1823a3437870SNishanth Aravamudan huge_page_size(h)/1024); 18248faa8b07SAndi Kleen 1825a3437870SNishanth Aravamudan parsed_hstate = h; 1826a3437870SNishanth Aravamudan } 1827a3437870SNishanth Aravamudan 1828e11bfbfcSNick Piggin static int __init hugetlb_nrpages_setup(char *s) 1829a3437870SNishanth Aravamudan { 1830a3437870SNishanth Aravamudan unsigned long *mhp; 18318faa8b07SAndi Kleen static unsigned long *last_mhp; 1832a3437870SNishanth Aravamudan 1833a3437870SNishanth Aravamudan /* 1834a3437870SNishanth Aravamudan * !max_hstate means we haven't parsed a hugepagesz= parameter yet, 1835a3437870SNishanth Aravamudan * so this hugepages= parameter goes to the "default hstate". 1836a3437870SNishanth Aravamudan */ 1837a3437870SNishanth Aravamudan if (!max_hstate) 1838a3437870SNishanth Aravamudan mhp = &default_hstate_max_huge_pages; 1839a3437870SNishanth Aravamudan else 1840a3437870SNishanth Aravamudan mhp = &parsed_hstate->max_huge_pages; 1841a3437870SNishanth Aravamudan 18428faa8b07SAndi Kleen if (mhp == last_mhp) { 18438faa8b07SAndi Kleen printk(KERN_WARNING "hugepages= specified twice without " 18448faa8b07SAndi Kleen "interleaving hugepagesz=, ignoring\n"); 18458faa8b07SAndi Kleen return 1; 18468faa8b07SAndi Kleen } 18478faa8b07SAndi Kleen 1848a3437870SNishanth Aravamudan if (sscanf(s, "%lu", mhp) <= 0) 1849a3437870SNishanth Aravamudan *mhp = 0; 1850a3437870SNishanth Aravamudan 18518faa8b07SAndi Kleen /* 18528faa8b07SAndi Kleen * Global state is always initialized later in hugetlb_init. 18538faa8b07SAndi Kleen * But we need to allocate >= MAX_ORDER hstates here early to still 18548faa8b07SAndi Kleen * use the bootmem allocator. 18558faa8b07SAndi Kleen */ 18568faa8b07SAndi Kleen if (max_hstate && parsed_hstate->order >= MAX_ORDER) 18578faa8b07SAndi Kleen hugetlb_hstate_alloc_pages(parsed_hstate); 18588faa8b07SAndi Kleen 18598faa8b07SAndi Kleen last_mhp = mhp; 18608faa8b07SAndi Kleen 1861a3437870SNishanth Aravamudan return 1; 1862a3437870SNishanth Aravamudan } 1863e11bfbfcSNick Piggin __setup("hugepages=", hugetlb_nrpages_setup); 1864e11bfbfcSNick Piggin 1865e11bfbfcSNick Piggin static int __init hugetlb_default_setup(char *s) 1866e11bfbfcSNick Piggin { 1867e11bfbfcSNick Piggin default_hstate_size = memparse(s, &s); 1868e11bfbfcSNick Piggin return 1; 1869e11bfbfcSNick Piggin } 1870e11bfbfcSNick Piggin __setup("default_hugepagesz=", hugetlb_default_setup); 1871a3437870SNishanth Aravamudan 18728a213460SNishanth Aravamudan static unsigned int cpuset_mems_nr(unsigned int *array) 18738a213460SNishanth Aravamudan { 18748a213460SNishanth Aravamudan int node; 18758a213460SNishanth Aravamudan unsigned int nr = 0; 18768a213460SNishanth Aravamudan 18778a213460SNishanth Aravamudan for_each_node_mask(node, cpuset_current_mems_allowed) 18788a213460SNishanth Aravamudan nr += array[node]; 18798a213460SNishanth Aravamudan 18808a213460SNishanth Aravamudan return nr; 18818a213460SNishanth Aravamudan } 18828a213460SNishanth Aravamudan 18838a213460SNishanth Aravamudan #ifdef CONFIG_SYSCTL 188406808b08SLee Schermerhorn static int hugetlb_sysctl_handler_common(bool obey_mempolicy, 188506808b08SLee Schermerhorn struct ctl_table *table, int write, 188606808b08SLee Schermerhorn void __user *buffer, size_t *length, loff_t *ppos) 18871da177e4SLinus Torvalds { 1888e5ff2159SAndi Kleen struct hstate *h = &default_hstate; 1889e5ff2159SAndi Kleen unsigned long tmp; 189008d4a246SMichal Hocko int ret; 1891e5ff2159SAndi Kleen 1892e5ff2159SAndi Kleen tmp = h->max_huge_pages; 1893e5ff2159SAndi Kleen 1894adbe8726SEric B Munson if (write && h->order >= MAX_ORDER) 1895adbe8726SEric B Munson return -EINVAL; 1896adbe8726SEric B Munson 1897e5ff2159SAndi Kleen table->data = &tmp; 1898e5ff2159SAndi Kleen table->maxlen = sizeof(unsigned long); 189908d4a246SMichal Hocko ret = proc_doulongvec_minmax(table, write, buffer, length, ppos); 190008d4a246SMichal Hocko if (ret) 190108d4a246SMichal Hocko goto out; 1902e5ff2159SAndi Kleen 190306808b08SLee Schermerhorn if (write) { 1904bad44b5bSDavid Rientjes NODEMASK_ALLOC(nodemask_t, nodes_allowed, 1905bad44b5bSDavid Rientjes GFP_KERNEL | __GFP_NORETRY); 190606808b08SLee Schermerhorn if (!(obey_mempolicy && 190706808b08SLee Schermerhorn init_nodemask_of_mempolicy(nodes_allowed))) { 190806808b08SLee Schermerhorn NODEMASK_FREE(nodes_allowed); 190906808b08SLee Schermerhorn nodes_allowed = &node_states[N_HIGH_MEMORY]; 191006808b08SLee Schermerhorn } 191106808b08SLee Schermerhorn h->max_huge_pages = set_max_huge_pages(h, tmp, nodes_allowed); 191206808b08SLee Schermerhorn 191306808b08SLee Schermerhorn if (nodes_allowed != &node_states[N_HIGH_MEMORY]) 191406808b08SLee Schermerhorn NODEMASK_FREE(nodes_allowed); 191506808b08SLee Schermerhorn } 191608d4a246SMichal Hocko out: 191708d4a246SMichal Hocko return ret; 19181da177e4SLinus Torvalds } 1919396faf03SMel Gorman 192006808b08SLee Schermerhorn int hugetlb_sysctl_handler(struct ctl_table *table, int write, 192106808b08SLee Schermerhorn void __user *buffer, size_t *length, loff_t *ppos) 192206808b08SLee Schermerhorn { 192306808b08SLee Schermerhorn 192406808b08SLee Schermerhorn return hugetlb_sysctl_handler_common(false, table, write, 192506808b08SLee Schermerhorn buffer, length, ppos); 192606808b08SLee Schermerhorn } 192706808b08SLee Schermerhorn 192806808b08SLee Schermerhorn #ifdef CONFIG_NUMA 192906808b08SLee Schermerhorn int hugetlb_mempolicy_sysctl_handler(struct ctl_table *table, int write, 193006808b08SLee Schermerhorn void __user *buffer, size_t *length, loff_t *ppos) 193106808b08SLee Schermerhorn { 193206808b08SLee Schermerhorn return hugetlb_sysctl_handler_common(true, table, write, 193306808b08SLee Schermerhorn buffer, length, ppos); 193406808b08SLee Schermerhorn } 193506808b08SLee Schermerhorn #endif /* CONFIG_NUMA */ 193606808b08SLee Schermerhorn 1937396faf03SMel Gorman int hugetlb_treat_movable_handler(struct ctl_table *table, int write, 19388d65af78SAlexey Dobriyan void __user *buffer, 1939396faf03SMel Gorman size_t *length, loff_t *ppos) 1940396faf03SMel Gorman { 19418d65af78SAlexey Dobriyan proc_dointvec(table, write, buffer, length, ppos); 1942396faf03SMel Gorman if (hugepages_treat_as_movable) 1943396faf03SMel Gorman htlb_alloc_mask = GFP_HIGHUSER_MOVABLE; 1944396faf03SMel Gorman else 1945396faf03SMel Gorman htlb_alloc_mask = GFP_HIGHUSER; 1946396faf03SMel Gorman return 0; 1947396faf03SMel Gorman } 1948396faf03SMel Gorman 1949a3d0c6aaSNishanth Aravamudan int hugetlb_overcommit_handler(struct ctl_table *table, int write, 19508d65af78SAlexey Dobriyan void __user *buffer, 1951a3d0c6aaSNishanth Aravamudan size_t *length, loff_t *ppos) 1952a3d0c6aaSNishanth Aravamudan { 1953a5516438SAndi Kleen struct hstate *h = &default_hstate; 1954e5ff2159SAndi Kleen unsigned long tmp; 195508d4a246SMichal Hocko int ret; 1956e5ff2159SAndi Kleen 1957e5ff2159SAndi Kleen tmp = h->nr_overcommit_huge_pages; 1958e5ff2159SAndi Kleen 1959adbe8726SEric B Munson if (write && h->order >= MAX_ORDER) 1960adbe8726SEric B Munson return -EINVAL; 1961adbe8726SEric B Munson 1962e5ff2159SAndi Kleen table->data = &tmp; 1963e5ff2159SAndi Kleen table->maxlen = sizeof(unsigned long); 196408d4a246SMichal Hocko ret = proc_doulongvec_minmax(table, write, buffer, length, ppos); 196508d4a246SMichal Hocko if (ret) 196608d4a246SMichal Hocko goto out; 1967e5ff2159SAndi Kleen 1968e5ff2159SAndi Kleen if (write) { 1969064d9efeSNishanth Aravamudan spin_lock(&hugetlb_lock); 1970e5ff2159SAndi Kleen h->nr_overcommit_huge_pages = tmp; 1971a3d0c6aaSNishanth Aravamudan spin_unlock(&hugetlb_lock); 1972e5ff2159SAndi Kleen } 197308d4a246SMichal Hocko out: 197408d4a246SMichal Hocko return ret; 1975a3d0c6aaSNishanth Aravamudan } 1976a3d0c6aaSNishanth Aravamudan 19771da177e4SLinus Torvalds #endif /* CONFIG_SYSCTL */ 19781da177e4SLinus Torvalds 1979e1759c21SAlexey Dobriyan void hugetlb_report_meminfo(struct seq_file *m) 19801da177e4SLinus Torvalds { 1981a5516438SAndi Kleen struct hstate *h = &default_hstate; 1982e1759c21SAlexey Dobriyan seq_printf(m, 19831da177e4SLinus Torvalds "HugePages_Total: %5lu\n" 19841da177e4SLinus Torvalds "HugePages_Free: %5lu\n" 1985b45b5bd6SDavid Gibson "HugePages_Rsvd: %5lu\n" 19867893d1d5SAdam Litke "HugePages_Surp: %5lu\n" 19874f98a2feSRik van Riel "Hugepagesize: %8lu kB\n", 1988a5516438SAndi Kleen h->nr_huge_pages, 1989a5516438SAndi Kleen h->free_huge_pages, 1990a5516438SAndi Kleen h->resv_huge_pages, 1991a5516438SAndi Kleen h->surplus_huge_pages, 1992a5516438SAndi Kleen 1UL << (huge_page_order(h) + PAGE_SHIFT - 10)); 19931da177e4SLinus Torvalds } 19941da177e4SLinus Torvalds 19951da177e4SLinus Torvalds int hugetlb_report_node_meminfo(int nid, char *buf) 19961da177e4SLinus Torvalds { 1997a5516438SAndi Kleen struct hstate *h = &default_hstate; 19981da177e4SLinus Torvalds return sprintf(buf, 19991da177e4SLinus Torvalds "Node %d HugePages_Total: %5u\n" 2000a1de0919SNishanth Aravamudan "Node %d HugePages_Free: %5u\n" 2001a1de0919SNishanth Aravamudan "Node %d HugePages_Surp: %5u\n", 2002a5516438SAndi Kleen nid, h->nr_huge_pages_node[nid], 2003a5516438SAndi Kleen nid, h->free_huge_pages_node[nid], 2004a5516438SAndi Kleen nid, h->surplus_huge_pages_node[nid]); 20051da177e4SLinus Torvalds } 20061da177e4SLinus Torvalds 20071da177e4SLinus Torvalds /* Return the number pages of memory we physically have, in PAGE_SIZE units. */ 20081da177e4SLinus Torvalds unsigned long hugetlb_total_pages(void) 20091da177e4SLinus Torvalds { 2010a5516438SAndi Kleen struct hstate *h = &default_hstate; 2011a5516438SAndi Kleen return h->nr_huge_pages * pages_per_huge_page(h); 20121da177e4SLinus Torvalds } 20131da177e4SLinus Torvalds 2014a5516438SAndi Kleen static int hugetlb_acct_memory(struct hstate *h, long delta) 2015fc1b8a73SMel Gorman { 2016fc1b8a73SMel Gorman int ret = -ENOMEM; 2017fc1b8a73SMel Gorman 2018fc1b8a73SMel Gorman spin_lock(&hugetlb_lock); 2019fc1b8a73SMel Gorman /* 2020fc1b8a73SMel Gorman * When cpuset is configured, it breaks the strict hugetlb page 2021fc1b8a73SMel Gorman * reservation as the accounting is done on a global variable. Such 2022fc1b8a73SMel Gorman * reservation is completely rubbish in the presence of cpuset because 2023fc1b8a73SMel Gorman * the reservation is not checked against page availability for the 2024fc1b8a73SMel Gorman * current cpuset. Application can still potentially OOM'ed by kernel 2025fc1b8a73SMel Gorman * with lack of free htlb page in cpuset that the task is in. 2026fc1b8a73SMel Gorman * Attempt to enforce strict accounting with cpuset is almost 2027fc1b8a73SMel Gorman * impossible (or too ugly) because cpuset is too fluid that 2028fc1b8a73SMel Gorman * task or memory node can be dynamically moved between cpusets. 2029fc1b8a73SMel Gorman * 2030fc1b8a73SMel Gorman * The change of semantics for shared hugetlb mapping with cpuset is 2031fc1b8a73SMel Gorman * undesirable. However, in order to preserve some of the semantics, 2032fc1b8a73SMel Gorman * we fall back to check against current free page availability as 2033fc1b8a73SMel Gorman * a best attempt and hopefully to minimize the impact of changing 2034fc1b8a73SMel Gorman * semantics that cpuset has. 2035fc1b8a73SMel Gorman */ 2036fc1b8a73SMel Gorman if (delta > 0) { 2037a5516438SAndi Kleen if (gather_surplus_pages(h, delta) < 0) 2038fc1b8a73SMel Gorman goto out; 2039fc1b8a73SMel Gorman 2040a5516438SAndi Kleen if (delta > cpuset_mems_nr(h->free_huge_pages_node)) { 2041a5516438SAndi Kleen return_unused_surplus_pages(h, delta); 2042fc1b8a73SMel Gorman goto out; 2043fc1b8a73SMel Gorman } 2044fc1b8a73SMel Gorman } 2045fc1b8a73SMel Gorman 2046fc1b8a73SMel Gorman ret = 0; 2047fc1b8a73SMel Gorman if (delta < 0) 2048a5516438SAndi Kleen return_unused_surplus_pages(h, (unsigned long) -delta); 2049fc1b8a73SMel Gorman 2050fc1b8a73SMel Gorman out: 2051fc1b8a73SMel Gorman spin_unlock(&hugetlb_lock); 2052fc1b8a73SMel Gorman return ret; 2053fc1b8a73SMel Gorman } 2054fc1b8a73SMel Gorman 205584afd99bSAndy Whitcroft static void hugetlb_vm_op_open(struct vm_area_struct *vma) 205684afd99bSAndy Whitcroft { 205784afd99bSAndy Whitcroft struct resv_map *reservations = vma_resv_map(vma); 205884afd99bSAndy Whitcroft 205984afd99bSAndy Whitcroft /* 206084afd99bSAndy Whitcroft * This new VMA should share its siblings reservation map if present. 206184afd99bSAndy Whitcroft * The VMA will only ever have a valid reservation map pointer where 206284afd99bSAndy Whitcroft * it is being copied for another still existing VMA. As that VMA 206325985edcSLucas De Marchi * has a reference to the reservation map it cannot disappear until 206484afd99bSAndy Whitcroft * after this open call completes. It is therefore safe to take a 206584afd99bSAndy Whitcroft * new reference here without additional locking. 206684afd99bSAndy Whitcroft */ 206784afd99bSAndy Whitcroft if (reservations) 206884afd99bSAndy Whitcroft kref_get(&reservations->refs); 206984afd99bSAndy Whitcroft } 207084afd99bSAndy Whitcroft 2071a1e78772SMel Gorman static void hugetlb_vm_op_close(struct vm_area_struct *vma) 2072a1e78772SMel Gorman { 2073a5516438SAndi Kleen struct hstate *h = hstate_vma(vma); 207484afd99bSAndy Whitcroft struct resv_map *reservations = vma_resv_map(vma); 207584afd99bSAndy Whitcroft unsigned long reserve; 207684afd99bSAndy Whitcroft unsigned long start; 207784afd99bSAndy Whitcroft unsigned long end; 207884afd99bSAndy Whitcroft 207984afd99bSAndy Whitcroft if (reservations) { 2080a5516438SAndi Kleen start = vma_hugecache_offset(h, vma, vma->vm_start); 2081a5516438SAndi Kleen end = vma_hugecache_offset(h, vma, vma->vm_end); 208284afd99bSAndy Whitcroft 208384afd99bSAndy Whitcroft reserve = (end - start) - 208484afd99bSAndy Whitcroft region_count(&reservations->regions, start, end); 208584afd99bSAndy Whitcroft 208684afd99bSAndy Whitcroft kref_put(&reservations->refs, resv_map_release); 208784afd99bSAndy Whitcroft 20887251ff78SAdam Litke if (reserve) { 2089a5516438SAndi Kleen hugetlb_acct_memory(h, -reserve); 20907251ff78SAdam Litke hugetlb_put_quota(vma->vm_file->f_mapping, reserve); 20917251ff78SAdam Litke } 2092a1e78772SMel Gorman } 209384afd99bSAndy Whitcroft } 2094a1e78772SMel Gorman 20951da177e4SLinus Torvalds /* 20961da177e4SLinus Torvalds * We cannot handle pagefaults against hugetlb pages at all. They cause 20971da177e4SLinus Torvalds * handle_mm_fault() to try to instantiate regular-sized pages in the 20981da177e4SLinus Torvalds * hugegpage VMA. do_page_fault() is supposed to trap this, so BUG is we get 20991da177e4SLinus Torvalds * this far. 21001da177e4SLinus Torvalds */ 2101d0217ac0SNick Piggin static int hugetlb_vm_op_fault(struct vm_area_struct *vma, struct vm_fault *vmf) 21021da177e4SLinus Torvalds { 21031da177e4SLinus Torvalds BUG(); 2104d0217ac0SNick Piggin return 0; 21051da177e4SLinus Torvalds } 21061da177e4SLinus Torvalds 2107f0f37e2fSAlexey Dobriyan const struct vm_operations_struct hugetlb_vm_ops = { 2108d0217ac0SNick Piggin .fault = hugetlb_vm_op_fault, 210984afd99bSAndy Whitcroft .open = hugetlb_vm_op_open, 2110a1e78772SMel Gorman .close = hugetlb_vm_op_close, 21111da177e4SLinus Torvalds }; 21121da177e4SLinus Torvalds 21131e8f889bSDavid Gibson static pte_t make_huge_pte(struct vm_area_struct *vma, struct page *page, 21141e8f889bSDavid Gibson int writable) 211563551ae0SDavid Gibson { 211663551ae0SDavid Gibson pte_t entry; 211763551ae0SDavid Gibson 21181e8f889bSDavid Gibson if (writable) { 211963551ae0SDavid Gibson entry = 212063551ae0SDavid Gibson pte_mkwrite(pte_mkdirty(mk_pte(page, vma->vm_page_prot))); 212163551ae0SDavid Gibson } else { 21227f2e9525SGerald Schaefer entry = huge_pte_wrprotect(mk_pte(page, vma->vm_page_prot)); 212363551ae0SDavid Gibson } 212463551ae0SDavid Gibson entry = pte_mkyoung(entry); 212563551ae0SDavid Gibson entry = pte_mkhuge(entry); 212663551ae0SDavid Gibson 212763551ae0SDavid Gibson return entry; 212863551ae0SDavid Gibson } 212963551ae0SDavid Gibson 21301e8f889bSDavid Gibson static void set_huge_ptep_writable(struct vm_area_struct *vma, 21311e8f889bSDavid Gibson unsigned long address, pte_t *ptep) 21321e8f889bSDavid Gibson { 21331e8f889bSDavid Gibson pte_t entry; 21341e8f889bSDavid Gibson 21357f2e9525SGerald Schaefer entry = pte_mkwrite(pte_mkdirty(huge_ptep_get(ptep))); 213632f84528SChris Forbes if (huge_ptep_set_access_flags(vma, address, ptep, entry, 1)) 21374b3073e1SRussell King update_mmu_cache(vma, address, ptep); 21381e8f889bSDavid Gibson } 21391e8f889bSDavid Gibson 21401e8f889bSDavid Gibson 214163551ae0SDavid Gibson int copy_hugetlb_page_range(struct mm_struct *dst, struct mm_struct *src, 214263551ae0SDavid Gibson struct vm_area_struct *vma) 214363551ae0SDavid Gibson { 214463551ae0SDavid Gibson pte_t *src_pte, *dst_pte, entry; 214563551ae0SDavid Gibson struct page *ptepage; 21461c59827dSHugh Dickins unsigned long addr; 21471e8f889bSDavid Gibson int cow; 2148a5516438SAndi Kleen struct hstate *h = hstate_vma(vma); 2149a5516438SAndi Kleen unsigned long sz = huge_page_size(h); 21501e8f889bSDavid Gibson 21511e8f889bSDavid Gibson cow = (vma->vm_flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE; 215263551ae0SDavid Gibson 2153a5516438SAndi Kleen for (addr = vma->vm_start; addr < vma->vm_end; addr += sz) { 2154c74df32cSHugh Dickins src_pte = huge_pte_offset(src, addr); 2155c74df32cSHugh Dickins if (!src_pte) 2156c74df32cSHugh Dickins continue; 2157a5516438SAndi Kleen dst_pte = huge_pte_alloc(dst, addr, sz); 215863551ae0SDavid Gibson if (!dst_pte) 215963551ae0SDavid Gibson goto nomem; 2160c5c99429SLarry Woodman 2161c5c99429SLarry Woodman /* If the pagetables are shared don't copy or take references */ 2162c5c99429SLarry Woodman if (dst_pte == src_pte) 2163c5c99429SLarry Woodman continue; 2164c5c99429SLarry Woodman 2165c74df32cSHugh Dickins spin_lock(&dst->page_table_lock); 216646478758SNick Piggin spin_lock_nested(&src->page_table_lock, SINGLE_DEPTH_NESTING); 21677f2e9525SGerald Schaefer if (!huge_pte_none(huge_ptep_get(src_pte))) { 21681e8f889bSDavid Gibson if (cow) 21697f2e9525SGerald Schaefer huge_ptep_set_wrprotect(src, addr, src_pte); 21707f2e9525SGerald Schaefer entry = huge_ptep_get(src_pte); 217163551ae0SDavid Gibson ptepage = pte_page(entry); 217263551ae0SDavid Gibson get_page(ptepage); 21730fe6e20bSNaoya Horiguchi page_dup_rmap(ptepage); 217463551ae0SDavid Gibson set_huge_pte_at(dst, addr, dst_pte, entry); 21751c59827dSHugh Dickins } 21761c59827dSHugh Dickins spin_unlock(&src->page_table_lock); 2177c74df32cSHugh Dickins spin_unlock(&dst->page_table_lock); 217863551ae0SDavid Gibson } 217963551ae0SDavid Gibson return 0; 218063551ae0SDavid Gibson 218163551ae0SDavid Gibson nomem: 218263551ae0SDavid Gibson return -ENOMEM; 218363551ae0SDavid Gibson } 218463551ae0SDavid Gibson 2185290408d4SNaoya Horiguchi static int is_hugetlb_entry_migration(pte_t pte) 2186290408d4SNaoya Horiguchi { 2187290408d4SNaoya Horiguchi swp_entry_t swp; 2188290408d4SNaoya Horiguchi 2189290408d4SNaoya Horiguchi if (huge_pte_none(pte) || pte_present(pte)) 2190290408d4SNaoya Horiguchi return 0; 2191290408d4SNaoya Horiguchi swp = pte_to_swp_entry(pte); 219232f84528SChris Forbes if (non_swap_entry(swp) && is_migration_entry(swp)) 2193290408d4SNaoya Horiguchi return 1; 219432f84528SChris Forbes else 2195290408d4SNaoya Horiguchi return 0; 2196290408d4SNaoya Horiguchi } 2197290408d4SNaoya Horiguchi 2198fd6a03edSNaoya Horiguchi static int is_hugetlb_entry_hwpoisoned(pte_t pte) 2199fd6a03edSNaoya Horiguchi { 2200fd6a03edSNaoya Horiguchi swp_entry_t swp; 2201fd6a03edSNaoya Horiguchi 2202fd6a03edSNaoya Horiguchi if (huge_pte_none(pte) || pte_present(pte)) 2203fd6a03edSNaoya Horiguchi return 0; 2204fd6a03edSNaoya Horiguchi swp = pte_to_swp_entry(pte); 220532f84528SChris Forbes if (non_swap_entry(swp) && is_hwpoison_entry(swp)) 2206fd6a03edSNaoya Horiguchi return 1; 220732f84528SChris Forbes else 2208fd6a03edSNaoya Horiguchi return 0; 2209fd6a03edSNaoya Horiguchi } 2210fd6a03edSNaoya Horiguchi 2211502717f4SChen, Kenneth W void __unmap_hugepage_range(struct vm_area_struct *vma, unsigned long start, 221204f2cbe3SMel Gorman unsigned long end, struct page *ref_page) 221363551ae0SDavid Gibson { 221463551ae0SDavid Gibson struct mm_struct *mm = vma->vm_mm; 221563551ae0SDavid Gibson unsigned long address; 2216c7546f8fSDavid Gibson pte_t *ptep; 221763551ae0SDavid Gibson pte_t pte; 221863551ae0SDavid Gibson struct page *page; 2219fe1668aeSChen, Kenneth W struct page *tmp; 2220a5516438SAndi Kleen struct hstate *h = hstate_vma(vma); 2221a5516438SAndi Kleen unsigned long sz = huge_page_size(h); 2222a5516438SAndi Kleen 2223c0a499c2SChen, Kenneth W /* 22243d48ae45SPeter Zijlstra * A page gathering list, protected by per file i_mmap_mutex. The 2225c0a499c2SChen, Kenneth W * lock is used to avoid list corruption from multiple unmapping 2226c0a499c2SChen, Kenneth W * of the same page since we are using page->lru. 2227c0a499c2SChen, Kenneth W */ 2228fe1668aeSChen, Kenneth W LIST_HEAD(page_list); 222963551ae0SDavid Gibson 223063551ae0SDavid Gibson WARN_ON(!is_vm_hugetlb_page(vma)); 2231a5516438SAndi Kleen BUG_ON(start & ~huge_page_mask(h)); 2232a5516438SAndi Kleen BUG_ON(end & ~huge_page_mask(h)); 223363551ae0SDavid Gibson 2234cddb8a5cSAndrea Arcangeli mmu_notifier_invalidate_range_start(mm, start, end); 2235508034a3SHugh Dickins spin_lock(&mm->page_table_lock); 2236a5516438SAndi Kleen for (address = start; address < end; address += sz) { 2237c7546f8fSDavid Gibson ptep = huge_pte_offset(mm, address); 2238c7546f8fSDavid Gibson if (!ptep) 2239c7546f8fSDavid Gibson continue; 2240c7546f8fSDavid Gibson 224139dde65cSChen, Kenneth W if (huge_pmd_unshare(mm, &address, ptep)) 224239dde65cSChen, Kenneth W continue; 224339dde65cSChen, Kenneth W 224404f2cbe3SMel Gorman /* 224504f2cbe3SMel Gorman * If a reference page is supplied, it is because a specific 224604f2cbe3SMel Gorman * page is being unmapped, not a range. Ensure the page we 224704f2cbe3SMel Gorman * are about to unmap is the actual page of interest. 224804f2cbe3SMel Gorman */ 224904f2cbe3SMel Gorman if (ref_page) { 225004f2cbe3SMel Gorman pte = huge_ptep_get(ptep); 225104f2cbe3SMel Gorman if (huge_pte_none(pte)) 225204f2cbe3SMel Gorman continue; 225304f2cbe3SMel Gorman page = pte_page(pte); 225404f2cbe3SMel Gorman if (page != ref_page) 225504f2cbe3SMel Gorman continue; 225604f2cbe3SMel Gorman 225704f2cbe3SMel Gorman /* 225804f2cbe3SMel Gorman * Mark the VMA as having unmapped its page so that 225904f2cbe3SMel Gorman * future faults in this VMA will fail rather than 226004f2cbe3SMel Gorman * looking like data was lost 226104f2cbe3SMel Gorman */ 226204f2cbe3SMel Gorman set_vma_resv_flags(vma, HPAGE_RESV_UNMAPPED); 226304f2cbe3SMel Gorman } 226404f2cbe3SMel Gorman 2265c7546f8fSDavid Gibson pte = huge_ptep_get_and_clear(mm, address, ptep); 22667f2e9525SGerald Schaefer if (huge_pte_none(pte)) 226763551ae0SDavid Gibson continue; 2268c7546f8fSDavid Gibson 2269fd6a03edSNaoya Horiguchi /* 2270fd6a03edSNaoya Horiguchi * HWPoisoned hugepage is already unmapped and dropped reference 2271fd6a03edSNaoya Horiguchi */ 2272fd6a03edSNaoya Horiguchi if (unlikely(is_hugetlb_entry_hwpoisoned(pte))) 2273fd6a03edSNaoya Horiguchi continue; 2274fd6a03edSNaoya Horiguchi 227563551ae0SDavid Gibson page = pte_page(pte); 22766649a386SKen Chen if (pte_dirty(pte)) 22776649a386SKen Chen set_page_dirty(page); 2278fe1668aeSChen, Kenneth W list_add(&page->lru, &page_list); 227963551ae0SDavid Gibson } 22801da177e4SLinus Torvalds spin_unlock(&mm->page_table_lock); 2281508034a3SHugh Dickins flush_tlb_range(vma, start, end); 2282cddb8a5cSAndrea Arcangeli mmu_notifier_invalidate_range_end(mm, start, end); 2283fe1668aeSChen, Kenneth W list_for_each_entry_safe(page, tmp, &page_list, lru) { 22840fe6e20bSNaoya Horiguchi page_remove_rmap(page); 2285fe1668aeSChen, Kenneth W list_del(&page->lru); 2286fe1668aeSChen, Kenneth W put_page(page); 2287fe1668aeSChen, Kenneth W } 22881da177e4SLinus Torvalds } 228963551ae0SDavid Gibson 2290502717f4SChen, Kenneth W void unmap_hugepage_range(struct vm_area_struct *vma, unsigned long start, 229104f2cbe3SMel Gorman unsigned long end, struct page *ref_page) 2292502717f4SChen, Kenneth W { 22933d48ae45SPeter Zijlstra mutex_lock(&vma->vm_file->f_mapping->i_mmap_mutex); 229404f2cbe3SMel Gorman __unmap_hugepage_range(vma, start, end, ref_page); 22953d48ae45SPeter Zijlstra mutex_unlock(&vma->vm_file->f_mapping->i_mmap_mutex); 2296502717f4SChen, Kenneth W } 2297502717f4SChen, Kenneth W 229804f2cbe3SMel Gorman /* 229904f2cbe3SMel Gorman * This is called when the original mapper is failing to COW a MAP_PRIVATE 230004f2cbe3SMel Gorman * mappping it owns the reserve page for. The intention is to unmap the page 230104f2cbe3SMel Gorman * from other VMAs and let the children be SIGKILLed if they are faulting the 230204f2cbe3SMel Gorman * same region. 230304f2cbe3SMel Gorman */ 23042a4b3dedSHarvey Harrison static int unmap_ref_private(struct mm_struct *mm, struct vm_area_struct *vma, 23052a4b3dedSHarvey Harrison struct page *page, unsigned long address) 230604f2cbe3SMel Gorman { 23077526674dSAdam Litke struct hstate *h = hstate_vma(vma); 230804f2cbe3SMel Gorman struct vm_area_struct *iter_vma; 230904f2cbe3SMel Gorman struct address_space *mapping; 231004f2cbe3SMel Gorman struct prio_tree_iter iter; 231104f2cbe3SMel Gorman pgoff_t pgoff; 231204f2cbe3SMel Gorman 231304f2cbe3SMel Gorman /* 231404f2cbe3SMel Gorman * vm_pgoff is in PAGE_SIZE units, hence the different calculation 231504f2cbe3SMel Gorman * from page cache lookup which is in HPAGE_SIZE units. 231604f2cbe3SMel Gorman */ 23177526674dSAdam Litke address = address & huge_page_mask(h); 231804f2cbe3SMel Gorman pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) 231904f2cbe3SMel Gorman + (vma->vm_pgoff >> PAGE_SHIFT); 232004f2cbe3SMel Gorman mapping = (struct address_space *)page_private(page); 232104f2cbe3SMel Gorman 23224eb2b1dcSMel Gorman /* 23234eb2b1dcSMel Gorman * Take the mapping lock for the duration of the table walk. As 23244eb2b1dcSMel Gorman * this mapping should be shared between all the VMAs, 23254eb2b1dcSMel Gorman * __unmap_hugepage_range() is called as the lock is already held 23264eb2b1dcSMel Gorman */ 23273d48ae45SPeter Zijlstra mutex_lock(&mapping->i_mmap_mutex); 232804f2cbe3SMel Gorman vma_prio_tree_foreach(iter_vma, &iter, &mapping->i_mmap, pgoff, pgoff) { 232904f2cbe3SMel Gorman /* Do not unmap the current VMA */ 233004f2cbe3SMel Gorman if (iter_vma == vma) 233104f2cbe3SMel Gorman continue; 233204f2cbe3SMel Gorman 233304f2cbe3SMel Gorman /* 233404f2cbe3SMel Gorman * Unmap the page from other VMAs without their own reserves. 233504f2cbe3SMel Gorman * They get marked to be SIGKILLed if they fault in these 233604f2cbe3SMel Gorman * areas. This is because a future no-page fault on this VMA 233704f2cbe3SMel Gorman * could insert a zeroed page instead of the data existing 233804f2cbe3SMel Gorman * from the time of fork. This would look like data corruption 233904f2cbe3SMel Gorman */ 234004f2cbe3SMel Gorman if (!is_vma_resv_set(iter_vma, HPAGE_RESV_OWNER)) 23414eb2b1dcSMel Gorman __unmap_hugepage_range(iter_vma, 23427526674dSAdam Litke address, address + huge_page_size(h), 234304f2cbe3SMel Gorman page); 234404f2cbe3SMel Gorman } 23453d48ae45SPeter Zijlstra mutex_unlock(&mapping->i_mmap_mutex); 234604f2cbe3SMel Gorman 234704f2cbe3SMel Gorman return 1; 234804f2cbe3SMel Gorman } 234904f2cbe3SMel Gorman 23500fe6e20bSNaoya Horiguchi /* 23510fe6e20bSNaoya Horiguchi * Hugetlb_cow() should be called with page lock of the original hugepage held. 2352ef009b25SMichal Hocko * Called with hugetlb_instantiation_mutex held and pte_page locked so we 2353ef009b25SMichal Hocko * cannot race with other handlers or page migration. 2354ef009b25SMichal Hocko * Keep the pte_same checks anyway to make transition from the mutex easier. 23550fe6e20bSNaoya Horiguchi */ 23561e8f889bSDavid Gibson static int hugetlb_cow(struct mm_struct *mm, struct vm_area_struct *vma, 235704f2cbe3SMel Gorman unsigned long address, pte_t *ptep, pte_t pte, 235804f2cbe3SMel Gorman struct page *pagecache_page) 23591e8f889bSDavid Gibson { 2360a5516438SAndi Kleen struct hstate *h = hstate_vma(vma); 23611e8f889bSDavid Gibson struct page *old_page, *new_page; 236279ac6ba4SDavid Gibson int avoidcopy; 236304f2cbe3SMel Gorman int outside_reserve = 0; 23641e8f889bSDavid Gibson 23651e8f889bSDavid Gibson old_page = pte_page(pte); 23661e8f889bSDavid Gibson 236704f2cbe3SMel Gorman retry_avoidcopy: 23681e8f889bSDavid Gibson /* If no-one else is actually using this page, avoid the copy 23691e8f889bSDavid Gibson * and just make the page writable */ 23700fe6e20bSNaoya Horiguchi avoidcopy = (page_mapcount(old_page) == 1); 23711e8f889bSDavid Gibson if (avoidcopy) { 23720fe6e20bSNaoya Horiguchi if (PageAnon(old_page)) 23730fe6e20bSNaoya Horiguchi page_move_anon_rmap(old_page, vma, address); 23741e8f889bSDavid Gibson set_huge_ptep_writable(vma, address, ptep); 237583c54070SNick Piggin return 0; 23761e8f889bSDavid Gibson } 23771e8f889bSDavid Gibson 237804f2cbe3SMel Gorman /* 237904f2cbe3SMel Gorman * If the process that created a MAP_PRIVATE mapping is about to 238004f2cbe3SMel Gorman * perform a COW due to a shared page count, attempt to satisfy 238104f2cbe3SMel Gorman * the allocation without using the existing reserves. The pagecache 238204f2cbe3SMel Gorman * page is used to determine if the reserve at this address was 238304f2cbe3SMel Gorman * consumed or not. If reserves were used, a partial faulted mapping 238404f2cbe3SMel Gorman * at the time of fork() could consume its reserves on COW instead 238504f2cbe3SMel Gorman * of the full address range. 238604f2cbe3SMel Gorman */ 2387f83a275dSMel Gorman if (!(vma->vm_flags & VM_MAYSHARE) && 238804f2cbe3SMel Gorman is_vma_resv_set(vma, HPAGE_RESV_OWNER) && 238904f2cbe3SMel Gorman old_page != pagecache_page) 239004f2cbe3SMel Gorman outside_reserve = 1; 239104f2cbe3SMel Gorman 23921e8f889bSDavid Gibson page_cache_get(old_page); 2393b76c8cfbSLarry Woodman 2394b76c8cfbSLarry Woodman /* Drop page_table_lock as buddy allocator may be called */ 2395b76c8cfbSLarry Woodman spin_unlock(&mm->page_table_lock); 239604f2cbe3SMel Gorman new_page = alloc_huge_page(vma, address, outside_reserve); 23971e8f889bSDavid Gibson 23982fc39cecSAdam Litke if (IS_ERR(new_page)) { 23991e8f889bSDavid Gibson page_cache_release(old_page); 240004f2cbe3SMel Gorman 240104f2cbe3SMel Gorman /* 240204f2cbe3SMel Gorman * If a process owning a MAP_PRIVATE mapping fails to COW, 240304f2cbe3SMel Gorman * it is due to references held by a child and an insufficient 240404f2cbe3SMel Gorman * huge page pool. To guarantee the original mappers 240504f2cbe3SMel Gorman * reliability, unmap the page from child processes. The child 240604f2cbe3SMel Gorman * may get SIGKILLed if it later faults. 240704f2cbe3SMel Gorman */ 240804f2cbe3SMel Gorman if (outside_reserve) { 240904f2cbe3SMel Gorman BUG_ON(huge_pte_none(pte)); 241004f2cbe3SMel Gorman if (unmap_ref_private(mm, vma, old_page, address)) { 241104f2cbe3SMel Gorman BUG_ON(page_count(old_page) != 1); 241204f2cbe3SMel Gorman BUG_ON(huge_pte_none(pte)); 2413b76c8cfbSLarry Woodman spin_lock(&mm->page_table_lock); 2414a734bcc8SHillf Danton ptep = huge_pte_offset(mm, address & huge_page_mask(h)); 2415a734bcc8SHillf Danton if (likely(pte_same(huge_ptep_get(ptep), pte))) 241604f2cbe3SMel Gorman goto retry_avoidcopy; 2417a734bcc8SHillf Danton /* 2418a734bcc8SHillf Danton * race occurs while re-acquiring page_table_lock, and 2419a734bcc8SHillf Danton * our job is done. 2420a734bcc8SHillf Danton */ 2421a734bcc8SHillf Danton return 0; 242204f2cbe3SMel Gorman } 242304f2cbe3SMel Gorman WARN_ON_ONCE(1); 242404f2cbe3SMel Gorman } 242504f2cbe3SMel Gorman 2426b76c8cfbSLarry Woodman /* Caller expects lock to be held */ 2427b76c8cfbSLarry Woodman spin_lock(&mm->page_table_lock); 24282fc39cecSAdam Litke return -PTR_ERR(new_page); 24291e8f889bSDavid Gibson } 24301e8f889bSDavid Gibson 24310fe6e20bSNaoya Horiguchi /* 24320fe6e20bSNaoya Horiguchi * When the original hugepage is shared one, it does not have 24330fe6e20bSNaoya Horiguchi * anon_vma prepared. 24340fe6e20bSNaoya Horiguchi */ 243544e2aa93SDean Nelson if (unlikely(anon_vma_prepare(vma))) { 2436ea4039a3SHillf Danton page_cache_release(new_page); 2437ea4039a3SHillf Danton page_cache_release(old_page); 243844e2aa93SDean Nelson /* Caller expects lock to be held */ 243944e2aa93SDean Nelson spin_lock(&mm->page_table_lock); 24400fe6e20bSNaoya Horiguchi return VM_FAULT_OOM; 244144e2aa93SDean Nelson } 24420fe6e20bSNaoya Horiguchi 244347ad8475SAndrea Arcangeli copy_user_huge_page(new_page, old_page, address, vma, 244447ad8475SAndrea Arcangeli pages_per_huge_page(h)); 24450ed361deSNick Piggin __SetPageUptodate(new_page); 24461e8f889bSDavid Gibson 2447b76c8cfbSLarry Woodman /* 2448b76c8cfbSLarry Woodman * Retake the page_table_lock to check for racing updates 2449b76c8cfbSLarry Woodman * before the page tables are altered 2450b76c8cfbSLarry Woodman */ 2451b76c8cfbSLarry Woodman spin_lock(&mm->page_table_lock); 2452a5516438SAndi Kleen ptep = huge_pte_offset(mm, address & huge_page_mask(h)); 24537f2e9525SGerald Schaefer if (likely(pte_same(huge_ptep_get(ptep), pte))) { 24541e8f889bSDavid Gibson /* Break COW */ 24553edd4fc9SDoug Doan mmu_notifier_invalidate_range_start(mm, 24563edd4fc9SDoug Doan address & huge_page_mask(h), 24573edd4fc9SDoug Doan (address & huge_page_mask(h)) + huge_page_size(h)); 24588fe627ecSGerald Schaefer huge_ptep_clear_flush(vma, address, ptep); 24591e8f889bSDavid Gibson set_huge_pte_at(mm, address, ptep, 24601e8f889bSDavid Gibson make_huge_pte(vma, new_page, 1)); 24610fe6e20bSNaoya Horiguchi page_remove_rmap(old_page); 2462cd67f0d2SNaoya Horiguchi hugepage_add_new_anon_rmap(new_page, vma, address); 24631e8f889bSDavid Gibson /* Make the old page be freed below */ 24641e8f889bSDavid Gibson new_page = old_page; 24653edd4fc9SDoug Doan mmu_notifier_invalidate_range_end(mm, 24663edd4fc9SDoug Doan address & huge_page_mask(h), 24673edd4fc9SDoug Doan (address & huge_page_mask(h)) + huge_page_size(h)); 24681e8f889bSDavid Gibson } 24691e8f889bSDavid Gibson page_cache_release(new_page); 24701e8f889bSDavid Gibson page_cache_release(old_page); 247183c54070SNick Piggin return 0; 24721e8f889bSDavid Gibson } 24731e8f889bSDavid Gibson 247404f2cbe3SMel Gorman /* Return the pagecache page at a given address within a VMA */ 2475a5516438SAndi Kleen static struct page *hugetlbfs_pagecache_page(struct hstate *h, 2476a5516438SAndi Kleen struct vm_area_struct *vma, unsigned long address) 247704f2cbe3SMel Gorman { 247804f2cbe3SMel Gorman struct address_space *mapping; 2479e7c4b0bfSAndy Whitcroft pgoff_t idx; 248004f2cbe3SMel Gorman 248104f2cbe3SMel Gorman mapping = vma->vm_file->f_mapping; 2482a5516438SAndi Kleen idx = vma_hugecache_offset(h, vma, address); 248304f2cbe3SMel Gorman 248404f2cbe3SMel Gorman return find_lock_page(mapping, idx); 248504f2cbe3SMel Gorman } 248604f2cbe3SMel Gorman 24873ae77f43SHugh Dickins /* 24883ae77f43SHugh Dickins * Return whether there is a pagecache page to back given address within VMA. 24893ae77f43SHugh Dickins * Caller follow_hugetlb_page() holds page_table_lock so we cannot lock_page. 24903ae77f43SHugh Dickins */ 24913ae77f43SHugh Dickins static bool hugetlbfs_pagecache_present(struct hstate *h, 24922a15efc9SHugh Dickins struct vm_area_struct *vma, unsigned long address) 24932a15efc9SHugh Dickins { 24942a15efc9SHugh Dickins struct address_space *mapping; 24952a15efc9SHugh Dickins pgoff_t idx; 24962a15efc9SHugh Dickins struct page *page; 24972a15efc9SHugh Dickins 24982a15efc9SHugh Dickins mapping = vma->vm_file->f_mapping; 24992a15efc9SHugh Dickins idx = vma_hugecache_offset(h, vma, address); 25002a15efc9SHugh Dickins 25012a15efc9SHugh Dickins page = find_get_page(mapping, idx); 25022a15efc9SHugh Dickins if (page) 25032a15efc9SHugh Dickins put_page(page); 25042a15efc9SHugh Dickins return page != NULL; 25052a15efc9SHugh Dickins } 25062a15efc9SHugh Dickins 2507a1ed3ddaSRobert P. J. Day static int hugetlb_no_page(struct mm_struct *mm, struct vm_area_struct *vma, 2508788c7df4SHugh Dickins unsigned long address, pte_t *ptep, unsigned int flags) 2509ac9b9c66SHugh Dickins { 2510a5516438SAndi Kleen struct hstate *h = hstate_vma(vma); 2511ac9b9c66SHugh Dickins int ret = VM_FAULT_SIGBUS; 2512e7c4b0bfSAndy Whitcroft pgoff_t idx; 25134c887265SAdam Litke unsigned long size; 25144c887265SAdam Litke struct page *page; 25154c887265SAdam Litke struct address_space *mapping; 25161e8f889bSDavid Gibson pte_t new_pte; 25174c887265SAdam Litke 251804f2cbe3SMel Gorman /* 251904f2cbe3SMel Gorman * Currently, we are forced to kill the process in the event the 252004f2cbe3SMel Gorman * original mapper has unmapped pages from the child due to a failed 252125985edcSLucas De Marchi * COW. Warn that such a situation has occurred as it may not be obvious 252204f2cbe3SMel Gorman */ 252304f2cbe3SMel Gorman if (is_vma_resv_set(vma, HPAGE_RESV_UNMAPPED)) { 252404f2cbe3SMel Gorman printk(KERN_WARNING 252504f2cbe3SMel Gorman "PID %d killed due to inadequate hugepage pool\n", 252604f2cbe3SMel Gorman current->pid); 252704f2cbe3SMel Gorman return ret; 252804f2cbe3SMel Gorman } 252904f2cbe3SMel Gorman 25304c887265SAdam Litke mapping = vma->vm_file->f_mapping; 2531a5516438SAndi Kleen idx = vma_hugecache_offset(h, vma, address); 25324c887265SAdam Litke 25334c887265SAdam Litke /* 25344c887265SAdam Litke * Use page lock to guard against racing truncation 25354c887265SAdam Litke * before we get page_table_lock. 25364c887265SAdam Litke */ 25376bda666aSChristoph Lameter retry: 25386bda666aSChristoph Lameter page = find_lock_page(mapping, idx); 25396bda666aSChristoph Lameter if (!page) { 2540a5516438SAndi Kleen size = i_size_read(mapping->host) >> huge_page_shift(h); 2541ebed4bfcSHugh Dickins if (idx >= size) 2542ebed4bfcSHugh Dickins goto out; 254304f2cbe3SMel Gorman page = alloc_huge_page(vma, address, 0); 25442fc39cecSAdam Litke if (IS_ERR(page)) { 25452fc39cecSAdam Litke ret = -PTR_ERR(page); 25466bda666aSChristoph Lameter goto out; 25476bda666aSChristoph Lameter } 254847ad8475SAndrea Arcangeli clear_huge_page(page, address, pages_per_huge_page(h)); 25490ed361deSNick Piggin __SetPageUptodate(page); 2550ac9b9c66SHugh Dickins 2551f83a275dSMel Gorman if (vma->vm_flags & VM_MAYSHARE) { 25526bda666aSChristoph Lameter int err; 255345c682a6SKen Chen struct inode *inode = mapping->host; 25546bda666aSChristoph Lameter 25556bda666aSChristoph Lameter err = add_to_page_cache(page, mapping, idx, GFP_KERNEL); 25566bda666aSChristoph Lameter if (err) { 25576bda666aSChristoph Lameter put_page(page); 25586bda666aSChristoph Lameter if (err == -EEXIST) 25596bda666aSChristoph Lameter goto retry; 25606bda666aSChristoph Lameter goto out; 25616bda666aSChristoph Lameter } 256245c682a6SKen Chen 256345c682a6SKen Chen spin_lock(&inode->i_lock); 2564a5516438SAndi Kleen inode->i_blocks += blocks_per_huge_page(h); 256545c682a6SKen Chen spin_unlock(&inode->i_lock); 25660fe6e20bSNaoya Horiguchi page_dup_rmap(page); 256723be7468SMel Gorman } else { 25686bda666aSChristoph Lameter lock_page(page); 25690fe6e20bSNaoya Horiguchi if (unlikely(anon_vma_prepare(vma))) { 25700fe6e20bSNaoya Horiguchi ret = VM_FAULT_OOM; 25710fe6e20bSNaoya Horiguchi goto backout_unlocked; 257223be7468SMel Gorman } 25730fe6e20bSNaoya Horiguchi hugepage_add_new_anon_rmap(page, vma, address); 25740fe6e20bSNaoya Horiguchi } 25750fe6e20bSNaoya Horiguchi } else { 257657303d80SAndy Whitcroft /* 2577998b4382SNaoya Horiguchi * If memory error occurs between mmap() and fault, some process 2578998b4382SNaoya Horiguchi * don't have hwpoisoned swap entry for errored virtual address. 2579998b4382SNaoya Horiguchi * So we need to block hugepage fault by PG_hwpoison bit check. 2580fd6a03edSNaoya Horiguchi */ 2581fd6a03edSNaoya Horiguchi if (unlikely(PageHWPoison(page))) { 2582aa50d3a7SAndi Kleen ret = VM_FAULT_HWPOISON | 2583aa50d3a7SAndi Kleen VM_FAULT_SET_HINDEX(h - hstates); 2584fd6a03edSNaoya Horiguchi goto backout_unlocked; 25856bda666aSChristoph Lameter } 2586998b4382SNaoya Horiguchi page_dup_rmap(page); 2587998b4382SNaoya Horiguchi } 25881e8f889bSDavid Gibson 258957303d80SAndy Whitcroft /* 259057303d80SAndy Whitcroft * If we are going to COW a private mapping later, we examine the 259157303d80SAndy Whitcroft * pending reservations for this page now. This will ensure that 259257303d80SAndy Whitcroft * any allocations necessary to record that reservation occur outside 259357303d80SAndy Whitcroft * the spinlock. 259457303d80SAndy Whitcroft */ 2595788c7df4SHugh Dickins if ((flags & FAULT_FLAG_WRITE) && !(vma->vm_flags & VM_SHARED)) 25962b26736cSAndy Whitcroft if (vma_needs_reservation(h, vma, address) < 0) { 25972b26736cSAndy Whitcroft ret = VM_FAULT_OOM; 25982b26736cSAndy Whitcroft goto backout_unlocked; 25992b26736cSAndy Whitcroft } 260057303d80SAndy Whitcroft 2601ac9b9c66SHugh Dickins spin_lock(&mm->page_table_lock); 2602a5516438SAndi Kleen size = i_size_read(mapping->host) >> huge_page_shift(h); 26034c887265SAdam Litke if (idx >= size) 26044c887265SAdam Litke goto backout; 26054c887265SAdam Litke 260683c54070SNick Piggin ret = 0; 26077f2e9525SGerald Schaefer if (!huge_pte_none(huge_ptep_get(ptep))) 26084c887265SAdam Litke goto backout; 26094c887265SAdam Litke 26101e8f889bSDavid Gibson new_pte = make_huge_pte(vma, page, ((vma->vm_flags & VM_WRITE) 26111e8f889bSDavid Gibson && (vma->vm_flags & VM_SHARED))); 26121e8f889bSDavid Gibson set_huge_pte_at(mm, address, ptep, new_pte); 26131e8f889bSDavid Gibson 2614788c7df4SHugh Dickins if ((flags & FAULT_FLAG_WRITE) && !(vma->vm_flags & VM_SHARED)) { 26151e8f889bSDavid Gibson /* Optimization, do the COW without a second fault */ 261604f2cbe3SMel Gorman ret = hugetlb_cow(mm, vma, address, ptep, new_pte, page); 26171e8f889bSDavid Gibson } 26181e8f889bSDavid Gibson 2619ac9b9c66SHugh Dickins spin_unlock(&mm->page_table_lock); 26204c887265SAdam Litke unlock_page(page); 26214c887265SAdam Litke out: 2622ac9b9c66SHugh Dickins return ret; 26234c887265SAdam Litke 26244c887265SAdam Litke backout: 26254c887265SAdam Litke spin_unlock(&mm->page_table_lock); 26262b26736cSAndy Whitcroft backout_unlocked: 26274c887265SAdam Litke unlock_page(page); 26284c887265SAdam Litke put_page(page); 26294c887265SAdam Litke goto out; 2630ac9b9c66SHugh Dickins } 2631ac9b9c66SHugh Dickins 263286e5216fSAdam Litke int hugetlb_fault(struct mm_struct *mm, struct vm_area_struct *vma, 2633788c7df4SHugh Dickins unsigned long address, unsigned int flags) 263486e5216fSAdam Litke { 263586e5216fSAdam Litke pte_t *ptep; 263686e5216fSAdam Litke pte_t entry; 26371e8f889bSDavid Gibson int ret; 26380fe6e20bSNaoya Horiguchi struct page *page = NULL; 263957303d80SAndy Whitcroft struct page *pagecache_page = NULL; 26403935baa9SDavid Gibson static DEFINE_MUTEX(hugetlb_instantiation_mutex); 2641a5516438SAndi Kleen struct hstate *h = hstate_vma(vma); 264286e5216fSAdam Litke 2643fd6a03edSNaoya Horiguchi ptep = huge_pte_offset(mm, address); 2644fd6a03edSNaoya Horiguchi if (ptep) { 2645fd6a03edSNaoya Horiguchi entry = huge_ptep_get(ptep); 2646290408d4SNaoya Horiguchi if (unlikely(is_hugetlb_entry_migration(entry))) { 2647290408d4SNaoya Horiguchi migration_entry_wait(mm, (pmd_t *)ptep, address); 2648290408d4SNaoya Horiguchi return 0; 2649290408d4SNaoya Horiguchi } else if (unlikely(is_hugetlb_entry_hwpoisoned(entry))) 2650aa50d3a7SAndi Kleen return VM_FAULT_HWPOISON_LARGE | 2651aa50d3a7SAndi Kleen VM_FAULT_SET_HINDEX(h - hstates); 2652fd6a03edSNaoya Horiguchi } 2653fd6a03edSNaoya Horiguchi 2654a5516438SAndi Kleen ptep = huge_pte_alloc(mm, address, huge_page_size(h)); 265586e5216fSAdam Litke if (!ptep) 265686e5216fSAdam Litke return VM_FAULT_OOM; 265786e5216fSAdam Litke 26583935baa9SDavid Gibson /* 26593935baa9SDavid Gibson * Serialize hugepage allocation and instantiation, so that we don't 26603935baa9SDavid Gibson * get spurious allocation failures if two CPUs race to instantiate 26613935baa9SDavid Gibson * the same page in the page cache. 26623935baa9SDavid Gibson */ 26633935baa9SDavid Gibson mutex_lock(&hugetlb_instantiation_mutex); 26647f2e9525SGerald Schaefer entry = huge_ptep_get(ptep); 26657f2e9525SGerald Schaefer if (huge_pte_none(entry)) { 2666788c7df4SHugh Dickins ret = hugetlb_no_page(mm, vma, address, ptep, flags); 2667b4d1d99fSDavid Gibson goto out_mutex; 26683935baa9SDavid Gibson } 266986e5216fSAdam Litke 267083c54070SNick Piggin ret = 0; 26711e8f889bSDavid Gibson 267257303d80SAndy Whitcroft /* 267357303d80SAndy Whitcroft * If we are going to COW the mapping later, we examine the pending 267457303d80SAndy Whitcroft * reservations for this page now. This will ensure that any 267557303d80SAndy Whitcroft * allocations necessary to record that reservation occur outside the 267657303d80SAndy Whitcroft * spinlock. For private mappings, we also lookup the pagecache 267757303d80SAndy Whitcroft * page now as it is used to determine if a reservation has been 267857303d80SAndy Whitcroft * consumed. 267957303d80SAndy Whitcroft */ 2680788c7df4SHugh Dickins if ((flags & FAULT_FLAG_WRITE) && !pte_write(entry)) { 26812b26736cSAndy Whitcroft if (vma_needs_reservation(h, vma, address) < 0) { 26822b26736cSAndy Whitcroft ret = VM_FAULT_OOM; 2683b4d1d99fSDavid Gibson goto out_mutex; 26842b26736cSAndy Whitcroft } 268557303d80SAndy Whitcroft 2686f83a275dSMel Gorman if (!(vma->vm_flags & VM_MAYSHARE)) 268757303d80SAndy Whitcroft pagecache_page = hugetlbfs_pagecache_page(h, 268857303d80SAndy Whitcroft vma, address); 268957303d80SAndy Whitcroft } 269057303d80SAndy Whitcroft 269156c9cfb1SNaoya Horiguchi /* 269256c9cfb1SNaoya Horiguchi * hugetlb_cow() requires page locks of pte_page(entry) and 269356c9cfb1SNaoya Horiguchi * pagecache_page, so here we need take the former one 269456c9cfb1SNaoya Horiguchi * when page != pagecache_page or !pagecache_page. 269556c9cfb1SNaoya Horiguchi * Note that locking order is always pagecache_page -> page, 269656c9cfb1SNaoya Horiguchi * so no worry about deadlock. 269756c9cfb1SNaoya Horiguchi */ 26980fe6e20bSNaoya Horiguchi page = pte_page(entry); 269956c9cfb1SNaoya Horiguchi if (page != pagecache_page) 27000fe6e20bSNaoya Horiguchi lock_page(page); 27010fe6e20bSNaoya Horiguchi 27021e8f889bSDavid Gibson spin_lock(&mm->page_table_lock); 27031e8f889bSDavid Gibson /* Check for a racing update before calling hugetlb_cow */ 2704b4d1d99fSDavid Gibson if (unlikely(!pte_same(entry, huge_ptep_get(ptep)))) 2705b4d1d99fSDavid Gibson goto out_page_table_lock; 2706b4d1d99fSDavid Gibson 2707b4d1d99fSDavid Gibson 2708788c7df4SHugh Dickins if (flags & FAULT_FLAG_WRITE) { 2709b4d1d99fSDavid Gibson if (!pte_write(entry)) { 271057303d80SAndy Whitcroft ret = hugetlb_cow(mm, vma, address, ptep, entry, 271157303d80SAndy Whitcroft pagecache_page); 2712b4d1d99fSDavid Gibson goto out_page_table_lock; 2713b4d1d99fSDavid Gibson } 2714b4d1d99fSDavid Gibson entry = pte_mkdirty(entry); 2715b4d1d99fSDavid Gibson } 2716b4d1d99fSDavid Gibson entry = pte_mkyoung(entry); 2717788c7df4SHugh Dickins if (huge_ptep_set_access_flags(vma, address, ptep, entry, 2718788c7df4SHugh Dickins flags & FAULT_FLAG_WRITE)) 27194b3073e1SRussell King update_mmu_cache(vma, address, ptep); 2720b4d1d99fSDavid Gibson 2721b4d1d99fSDavid Gibson out_page_table_lock: 27221e8f889bSDavid Gibson spin_unlock(&mm->page_table_lock); 272357303d80SAndy Whitcroft 272457303d80SAndy Whitcroft if (pagecache_page) { 272557303d80SAndy Whitcroft unlock_page(pagecache_page); 272657303d80SAndy Whitcroft put_page(pagecache_page); 272757303d80SAndy Whitcroft } 27281f64d69cSDean Nelson if (page != pagecache_page) 272956c9cfb1SNaoya Horiguchi unlock_page(page); 273057303d80SAndy Whitcroft 2731b4d1d99fSDavid Gibson out_mutex: 27323935baa9SDavid Gibson mutex_unlock(&hugetlb_instantiation_mutex); 27331e8f889bSDavid Gibson 27341e8f889bSDavid Gibson return ret; 273586e5216fSAdam Litke } 273686e5216fSAdam Litke 2737ceb86879SAndi Kleen /* Can be overriden by architectures */ 2738ceb86879SAndi Kleen __attribute__((weak)) struct page * 2739ceb86879SAndi Kleen follow_huge_pud(struct mm_struct *mm, unsigned long address, 2740ceb86879SAndi Kleen pud_t *pud, int write) 2741ceb86879SAndi Kleen { 2742ceb86879SAndi Kleen BUG(); 2743ceb86879SAndi Kleen return NULL; 2744ceb86879SAndi Kleen } 2745ceb86879SAndi Kleen 274663551ae0SDavid Gibson int follow_hugetlb_page(struct mm_struct *mm, struct vm_area_struct *vma, 274763551ae0SDavid Gibson struct page **pages, struct vm_area_struct **vmas, 27485b23dbe8SAdam Litke unsigned long *position, int *length, int i, 27492a15efc9SHugh Dickins unsigned int flags) 275063551ae0SDavid Gibson { 2751d5d4b0aaSChen, Kenneth W unsigned long pfn_offset; 2752d5d4b0aaSChen, Kenneth W unsigned long vaddr = *position; 275363551ae0SDavid Gibson int remainder = *length; 2754a5516438SAndi Kleen struct hstate *h = hstate_vma(vma); 275563551ae0SDavid Gibson 27561c59827dSHugh Dickins spin_lock(&mm->page_table_lock); 275763551ae0SDavid Gibson while (vaddr < vma->vm_end && remainder) { 275863551ae0SDavid Gibson pte_t *pte; 27592a15efc9SHugh Dickins int absent; 276063551ae0SDavid Gibson struct page *page; 276163551ae0SDavid Gibson 27624c887265SAdam Litke /* 27634c887265SAdam Litke * Some archs (sparc64, sh*) have multiple pte_ts to 27642a15efc9SHugh Dickins * each hugepage. We have to make sure we get the 27654c887265SAdam Litke * first, for the page indexing below to work. 27664c887265SAdam Litke */ 2767a5516438SAndi Kleen pte = huge_pte_offset(mm, vaddr & huge_page_mask(h)); 27682a15efc9SHugh Dickins absent = !pte || huge_pte_none(huge_ptep_get(pte)); 276963551ae0SDavid Gibson 27702a15efc9SHugh Dickins /* 27712a15efc9SHugh Dickins * When coredumping, it suits get_dump_page if we just return 27723ae77f43SHugh Dickins * an error where there's an empty slot with no huge pagecache 27733ae77f43SHugh Dickins * to back it. This way, we avoid allocating a hugepage, and 27743ae77f43SHugh Dickins * the sparse dumpfile avoids allocating disk blocks, but its 27753ae77f43SHugh Dickins * huge holes still show up with zeroes where they need to be. 27762a15efc9SHugh Dickins */ 27773ae77f43SHugh Dickins if (absent && (flags & FOLL_DUMP) && 27783ae77f43SHugh Dickins !hugetlbfs_pagecache_present(h, vma, vaddr)) { 27792a15efc9SHugh Dickins remainder = 0; 27802a15efc9SHugh Dickins break; 27812a15efc9SHugh Dickins } 27822a15efc9SHugh Dickins 27832a15efc9SHugh Dickins if (absent || 27842a15efc9SHugh Dickins ((flags & FOLL_WRITE) && !pte_write(huge_ptep_get(pte)))) { 27854c887265SAdam Litke int ret; 27864c887265SAdam Litke 27874c887265SAdam Litke spin_unlock(&mm->page_table_lock); 27882a15efc9SHugh Dickins ret = hugetlb_fault(mm, vma, vaddr, 27892a15efc9SHugh Dickins (flags & FOLL_WRITE) ? FAULT_FLAG_WRITE : 0); 27904c887265SAdam Litke spin_lock(&mm->page_table_lock); 2791a89182c7SAdam Litke if (!(ret & VM_FAULT_ERROR)) 27924c887265SAdam Litke continue; 27934c887265SAdam Litke 27941c59827dSHugh Dickins remainder = 0; 27951c59827dSHugh Dickins break; 27961c59827dSHugh Dickins } 279763551ae0SDavid Gibson 2798a5516438SAndi Kleen pfn_offset = (vaddr & ~huge_page_mask(h)) >> PAGE_SHIFT; 27997f2e9525SGerald Schaefer page = pte_page(huge_ptep_get(pte)); 2800d5d4b0aaSChen, Kenneth W same_page: 2801d6692183SChen, Kenneth W if (pages) { 280269d177c2SAndy Whitcroft pages[i] = mem_map_offset(page, pfn_offset); 28034b2e38adSKOSAKI Motohiro get_page(pages[i]); 2804d6692183SChen, Kenneth W } 280563551ae0SDavid Gibson 280663551ae0SDavid Gibson if (vmas) 280763551ae0SDavid Gibson vmas[i] = vma; 280863551ae0SDavid Gibson 280963551ae0SDavid Gibson vaddr += PAGE_SIZE; 2810d5d4b0aaSChen, Kenneth W ++pfn_offset; 281163551ae0SDavid Gibson --remainder; 281263551ae0SDavid Gibson ++i; 2813d5d4b0aaSChen, Kenneth W if (vaddr < vma->vm_end && remainder && 2814a5516438SAndi Kleen pfn_offset < pages_per_huge_page(h)) { 2815d5d4b0aaSChen, Kenneth W /* 2816d5d4b0aaSChen, Kenneth W * We use pfn_offset to avoid touching the pageframes 2817d5d4b0aaSChen, Kenneth W * of this compound page. 2818d5d4b0aaSChen, Kenneth W */ 2819d5d4b0aaSChen, Kenneth W goto same_page; 2820d5d4b0aaSChen, Kenneth W } 282163551ae0SDavid Gibson } 28221c59827dSHugh Dickins spin_unlock(&mm->page_table_lock); 282363551ae0SDavid Gibson *length = remainder; 282463551ae0SDavid Gibson *position = vaddr; 282563551ae0SDavid Gibson 28262a15efc9SHugh Dickins return i ? i : -EFAULT; 282763551ae0SDavid Gibson } 28288f860591SZhang, Yanmin 28298f860591SZhang, Yanmin void hugetlb_change_protection(struct vm_area_struct *vma, 28308f860591SZhang, Yanmin unsigned long address, unsigned long end, pgprot_t newprot) 28318f860591SZhang, Yanmin { 28328f860591SZhang, Yanmin struct mm_struct *mm = vma->vm_mm; 28338f860591SZhang, Yanmin unsigned long start = address; 28348f860591SZhang, Yanmin pte_t *ptep; 28358f860591SZhang, Yanmin pte_t pte; 2836a5516438SAndi Kleen struct hstate *h = hstate_vma(vma); 28378f860591SZhang, Yanmin 28388f860591SZhang, Yanmin BUG_ON(address >= end); 28398f860591SZhang, Yanmin flush_cache_range(vma, address, end); 28408f860591SZhang, Yanmin 28413d48ae45SPeter Zijlstra mutex_lock(&vma->vm_file->f_mapping->i_mmap_mutex); 28428f860591SZhang, Yanmin spin_lock(&mm->page_table_lock); 2843a5516438SAndi Kleen for (; address < end; address += huge_page_size(h)) { 28448f860591SZhang, Yanmin ptep = huge_pte_offset(mm, address); 28458f860591SZhang, Yanmin if (!ptep) 28468f860591SZhang, Yanmin continue; 284739dde65cSChen, Kenneth W if (huge_pmd_unshare(mm, &address, ptep)) 284839dde65cSChen, Kenneth W continue; 28497f2e9525SGerald Schaefer if (!huge_pte_none(huge_ptep_get(ptep))) { 28508f860591SZhang, Yanmin pte = huge_ptep_get_and_clear(mm, address, ptep); 28518f860591SZhang, Yanmin pte = pte_mkhuge(pte_modify(pte, newprot)); 28528f860591SZhang, Yanmin set_huge_pte_at(mm, address, ptep, pte); 28538f860591SZhang, Yanmin } 28548f860591SZhang, Yanmin } 28558f860591SZhang, Yanmin spin_unlock(&mm->page_table_lock); 28563d48ae45SPeter Zijlstra mutex_unlock(&vma->vm_file->f_mapping->i_mmap_mutex); 28578f860591SZhang, Yanmin 28588f860591SZhang, Yanmin flush_tlb_range(vma, start, end); 28598f860591SZhang, Yanmin } 28608f860591SZhang, Yanmin 2861a1e78772SMel Gorman int hugetlb_reserve_pages(struct inode *inode, 2862a1e78772SMel Gorman long from, long to, 28635a6fe125SMel Gorman struct vm_area_struct *vma, 2864ca16d140SKOSAKI Motohiro vm_flags_t vm_flags) 2865e4e574b7SAdam Litke { 286617c9d12eSMel Gorman long ret, chg; 2867a5516438SAndi Kleen struct hstate *h = hstate_inode(inode); 2868e4e574b7SAdam Litke 2869a1e78772SMel Gorman /* 287017c9d12eSMel Gorman * Only apply hugepage reservation if asked. At fault time, an 287117c9d12eSMel Gorman * attempt will be made for VM_NORESERVE to allocate a page 287217c9d12eSMel Gorman * and filesystem quota without using reserves 287317c9d12eSMel Gorman */ 2874ca16d140SKOSAKI Motohiro if (vm_flags & VM_NORESERVE) 287517c9d12eSMel Gorman return 0; 287617c9d12eSMel Gorman 287717c9d12eSMel Gorman /* 2878a1e78772SMel Gorman * Shared mappings base their reservation on the number of pages that 2879a1e78772SMel Gorman * are already allocated on behalf of the file. Private mappings need 2880a1e78772SMel Gorman * to reserve the full area even if read-only as mprotect() may be 2881a1e78772SMel Gorman * called to make the mapping read-write. Assume !vma is a shm mapping 2882a1e78772SMel Gorman */ 2883f83a275dSMel Gorman if (!vma || vma->vm_flags & VM_MAYSHARE) 2884e4e574b7SAdam Litke chg = region_chg(&inode->i_mapping->private_list, from, to); 28855a6fe125SMel Gorman else { 28865a6fe125SMel Gorman struct resv_map *resv_map = resv_map_alloc(); 28875a6fe125SMel Gorman if (!resv_map) 28885a6fe125SMel Gorman return -ENOMEM; 28895a6fe125SMel Gorman 289017c9d12eSMel Gorman chg = to - from; 289117c9d12eSMel Gorman 28925a6fe125SMel Gorman set_vma_resv_map(vma, resv_map); 28935a6fe125SMel Gorman set_vma_resv_flags(vma, HPAGE_RESV_OWNER); 28945a6fe125SMel Gorman } 28955a6fe125SMel Gorman 289617c9d12eSMel Gorman if (chg < 0) 289717c9d12eSMel Gorman return chg; 289817c9d12eSMel Gorman 289917c9d12eSMel Gorman /* There must be enough filesystem quota for the mapping */ 290017c9d12eSMel Gorman if (hugetlb_get_quota(inode->i_mapping, chg)) 290117c9d12eSMel Gorman return -ENOSPC; 290217c9d12eSMel Gorman 290317c9d12eSMel Gorman /* 290417c9d12eSMel Gorman * Check enough hugepages are available for the reservation. 290517c9d12eSMel Gorman * Hand back the quota if there are not 290617c9d12eSMel Gorman */ 290717c9d12eSMel Gorman ret = hugetlb_acct_memory(h, chg); 290817c9d12eSMel Gorman if (ret < 0) { 290917c9d12eSMel Gorman hugetlb_put_quota(inode->i_mapping, chg); 291017c9d12eSMel Gorman return ret; 291117c9d12eSMel Gorman } 291217c9d12eSMel Gorman 291317c9d12eSMel Gorman /* 291417c9d12eSMel Gorman * Account for the reservations made. Shared mappings record regions 291517c9d12eSMel Gorman * that have reservations as they are shared by multiple VMAs. 291617c9d12eSMel Gorman * When the last VMA disappears, the region map says how much 291717c9d12eSMel Gorman * the reservation was and the page cache tells how much of 291817c9d12eSMel Gorman * the reservation was consumed. Private mappings are per-VMA and 291917c9d12eSMel Gorman * only the consumed reservations are tracked. When the VMA 292017c9d12eSMel Gorman * disappears, the original reservation is the VMA size and the 292117c9d12eSMel Gorman * consumed reservations are stored in the map. Hence, nothing 292217c9d12eSMel Gorman * else has to be done for private mappings here 292317c9d12eSMel Gorman */ 2924f83a275dSMel Gorman if (!vma || vma->vm_flags & VM_MAYSHARE) 292517c9d12eSMel Gorman region_add(&inode->i_mapping->private_list, from, to); 2926a43a8c39SChen, Kenneth W return 0; 2927a43a8c39SChen, Kenneth W } 2928a43a8c39SChen, Kenneth W 2929a43a8c39SChen, Kenneth W void hugetlb_unreserve_pages(struct inode *inode, long offset, long freed) 2930a43a8c39SChen, Kenneth W { 2931a5516438SAndi Kleen struct hstate *h = hstate_inode(inode); 2932a43a8c39SChen, Kenneth W long chg = region_truncate(&inode->i_mapping->private_list, offset); 293345c682a6SKen Chen 293445c682a6SKen Chen spin_lock(&inode->i_lock); 2935e4c6f8beSEric Sandeen inode->i_blocks -= (blocks_per_huge_page(h) * freed); 293645c682a6SKen Chen spin_unlock(&inode->i_lock); 293745c682a6SKen Chen 293890d8b7e6SAdam Litke hugetlb_put_quota(inode->i_mapping, (chg - freed)); 2939a5516438SAndi Kleen hugetlb_acct_memory(h, -(chg - freed)); 2940a43a8c39SChen, Kenneth W } 294193f70f90SNaoya Horiguchi 2942d5bd9106SAndi Kleen #ifdef CONFIG_MEMORY_FAILURE 2943d5bd9106SAndi Kleen 29446de2b1aaSNaoya Horiguchi /* Should be called in hugetlb_lock */ 29456de2b1aaSNaoya Horiguchi static int is_hugepage_on_freelist(struct page *hpage) 29466de2b1aaSNaoya Horiguchi { 29476de2b1aaSNaoya Horiguchi struct page *page; 29486de2b1aaSNaoya Horiguchi struct page *tmp; 29496de2b1aaSNaoya Horiguchi struct hstate *h = page_hstate(hpage); 29506de2b1aaSNaoya Horiguchi int nid = page_to_nid(hpage); 29516de2b1aaSNaoya Horiguchi 29526de2b1aaSNaoya Horiguchi list_for_each_entry_safe(page, tmp, &h->hugepage_freelists[nid], lru) 29536de2b1aaSNaoya Horiguchi if (page == hpage) 29546de2b1aaSNaoya Horiguchi return 1; 29556de2b1aaSNaoya Horiguchi return 0; 29566de2b1aaSNaoya Horiguchi } 29576de2b1aaSNaoya Horiguchi 295893f70f90SNaoya Horiguchi /* 295993f70f90SNaoya Horiguchi * This function is called from memory failure code. 296093f70f90SNaoya Horiguchi * Assume the caller holds page lock of the head page. 296193f70f90SNaoya Horiguchi */ 29626de2b1aaSNaoya Horiguchi int dequeue_hwpoisoned_huge_page(struct page *hpage) 296393f70f90SNaoya Horiguchi { 296493f70f90SNaoya Horiguchi struct hstate *h = page_hstate(hpage); 296593f70f90SNaoya Horiguchi int nid = page_to_nid(hpage); 29666de2b1aaSNaoya Horiguchi int ret = -EBUSY; 296793f70f90SNaoya Horiguchi 296893f70f90SNaoya Horiguchi spin_lock(&hugetlb_lock); 29696de2b1aaSNaoya Horiguchi if (is_hugepage_on_freelist(hpage)) { 297093f70f90SNaoya Horiguchi list_del(&hpage->lru); 29718c6c2ecbSNaoya Horiguchi set_page_refcounted(hpage); 297293f70f90SNaoya Horiguchi h->free_huge_pages--; 297393f70f90SNaoya Horiguchi h->free_huge_pages_node[nid]--; 29746de2b1aaSNaoya Horiguchi ret = 0; 297593f70f90SNaoya Horiguchi } 29766de2b1aaSNaoya Horiguchi spin_unlock(&hugetlb_lock); 29776de2b1aaSNaoya Horiguchi return ret; 29786de2b1aaSNaoya Horiguchi } 29796de2b1aaSNaoya Horiguchi #endif 2980