Greg Kroah-Hartman | b244131 | 2017-11-01 15:07:57 +0100 | [diff] [blame] | 1 | // SPDX-License-Identifier: GPL-2.0 |
Matt Mackall | 10cef60 | 2006-01-08 01:01:45 -0800 | [diff] [blame] | 2 | /* |
| 3 | * SLOB Allocator: Simple List Of Blocks |
| 4 | * |
| 5 | * Matt Mackall <mpm@selenic.com> 12/30/03 |
| 6 | * |
Paul Mundt | 6193a2f | 2007-07-15 23:38:22 -0700 | [diff] [blame] | 7 | * NUMA support by Paul Mundt, 2007. |
| 8 | * |
Matt Mackall | 10cef60 | 2006-01-08 01:01:45 -0800 | [diff] [blame] | 9 | * How SLOB works: |
| 10 | * |
| 11 | * The core of SLOB is a traditional K&R style heap allocator, with |
| 12 | * support for returning aligned objects. The granularity of this |
Nick Piggin | 5539484 | 2007-07-15 23:38:09 -0700 | [diff] [blame] | 13 | * allocator is as little as 2 bytes, however typically most architectures |
| 14 | * will require 4 bytes on 32-bit and 8 bytes on 64-bit. |
Nick Piggin | 95b3512 | 2007-07-15 23:38:07 -0700 | [diff] [blame] | 15 | * |
Matt Mackall | 20cecba | 2008-02-04 22:29:37 -0800 | [diff] [blame] | 16 | * The slob heap is a set of linked list of pages from alloc_pages(), |
| 17 | * and within each page, there is a singly-linked list of free blocks |
| 18 | * (slob_t). The heap is grown on demand. To reduce fragmentation, |
| 19 | * heap pages are segregated into three lists, with objects less than |
| 20 | * 256 bytes, objects less than 1024 bytes, and all other objects. |
| 21 | * |
| 22 | * Allocation from heap involves first searching for a page with |
| 23 | * sufficient free blocks (using a next-fit-like approach) followed by |
| 24 | * a first-fit scan of the page. Deallocation inserts objects back |
| 25 | * into the free list in address order, so this is effectively an |
| 26 | * address-ordered first fit. |
Matt Mackall | 10cef60 | 2006-01-08 01:01:45 -0800 | [diff] [blame] | 27 | * |
| 28 | * Above this is an implementation of kmalloc/kfree. Blocks returned |
Nick Piggin | 5539484 | 2007-07-15 23:38:09 -0700 | [diff] [blame] | 29 | * from kmalloc are prepended with a 4-byte header with the kmalloc size. |
Matt Mackall | 10cef60 | 2006-01-08 01:01:45 -0800 | [diff] [blame] | 30 | * If kmalloc is asked for objects of PAGE_SIZE or larger, it calls |
Paul Mundt | 6193a2f | 2007-07-15 23:38:22 -0700 | [diff] [blame] | 31 | * alloc_pages() directly, allocating compound pages so the page order |
Ezequiel Garcia | 999d879 | 2012-10-19 09:33:10 -0300 | [diff] [blame] | 32 | * does not have to be separately tracked. |
| 33 | * These objects are detected in kfree() because PageSlab() |
Nick Piggin | d87a133 | 2007-07-15 23:38:08 -0700 | [diff] [blame] | 34 | * is false for them. |
Matt Mackall | 10cef60 | 2006-01-08 01:01:45 -0800 | [diff] [blame] | 35 | * |
| 36 | * SLAB is emulated on top of SLOB by simply calling constructors and |
Nick Piggin | 95b3512 | 2007-07-15 23:38:07 -0700 | [diff] [blame] | 37 | * destructors for every SLAB allocation. Objects are returned with the |
| 38 | * 4-byte alignment unless the SLAB_HWCACHE_ALIGN flag is set, in which |
| 39 | * case the low-level allocator will fragment blocks to create the proper |
| 40 | * alignment. Again, objects of page-size or greater are allocated by |
Paul Mundt | 6193a2f | 2007-07-15 23:38:22 -0700 | [diff] [blame] | 41 | * calling alloc_pages(). As SLAB objects know their size, no separate |
Nick Piggin | 95b3512 | 2007-07-15 23:38:07 -0700 | [diff] [blame] | 42 | * size bookkeeping is necessary and there is essentially no allocation |
Nick Piggin | d87a133 | 2007-07-15 23:38:08 -0700 | [diff] [blame] | 43 | * space overhead, and compound pages aren't needed for multi-page |
| 44 | * allocations. |
Paul Mundt | 6193a2f | 2007-07-15 23:38:22 -0700 | [diff] [blame] | 45 | * |
| 46 | * NUMA support in SLOB is fairly simplistic, pushing most of the real |
| 47 | * logic down to the page allocator, and simply doing the node accounting |
| 48 | * on the upper levels. In the event that a node id is explicitly |
Vlastimil Babka | 96db800 | 2015-09-08 15:03:50 -0700 | [diff] [blame] | 49 | * provided, __alloc_pages_node() with the specified node id is used |
Paul Mundt | 6193a2f | 2007-07-15 23:38:22 -0700 | [diff] [blame] | 50 | * instead. The common case (or when the node id isn't explicitly provided) |
| 51 | * will default to the current node, as per numa_node_id(). |
| 52 | * |
| 53 | * Node aware pages are still inserted in to the global freelist, and |
| 54 | * these are scanned for by matching against the node id encoded in the |
| 55 | * page flags. As a result, block allocations that can be satisfied from |
| 56 | * the freelist will only be done so on pages residing on the same node, |
| 57 | * in order to prevent random node placement. |
Matt Mackall | 10cef60 | 2006-01-08 01:01:45 -0800 | [diff] [blame] | 58 | */ |
| 59 | |
Nick Piggin | 95b3512 | 2007-07-15 23:38:07 -0700 | [diff] [blame] | 60 | #include <linux/kernel.h> |
Matt Mackall | 10cef60 | 2006-01-08 01:01:45 -0800 | [diff] [blame] | 61 | #include <linux/slab.h> |
Christoph Lameter | 97d0660 | 2012-07-06 15:25:11 -0500 | [diff] [blame] | 62 | |
Matt Mackall | 10cef60 | 2006-01-08 01:01:45 -0800 | [diff] [blame] | 63 | #include <linux/mm.h> |
Nick Piggin | 1f0532e | 2009-05-05 19:13:45 +1000 | [diff] [blame] | 64 | #include <linux/swap.h> /* struct reclaim_state */ |
Matt Mackall | 10cef60 | 2006-01-08 01:01:45 -0800 | [diff] [blame] | 65 | #include <linux/cache.h> |
| 66 | #include <linux/init.h> |
Paul Gortmaker | b95f1b31 | 2011-10-16 02:01:52 -0400 | [diff] [blame] | 67 | #include <linux/export.h> |
Nick Piggin | afc0ced | 2007-05-16 22:10:49 -0700 | [diff] [blame] | 68 | #include <linux/rcupdate.h> |
Nick Piggin | 95b3512 | 2007-07-15 23:38:07 -0700 | [diff] [blame] | 69 | #include <linux/list.h> |
Catalin Marinas | 4374e61 | 2009-06-11 13:23:17 +0100 | [diff] [blame] | 70 | #include <linux/kmemleak.h> |
Li Zefan | 039ca4e | 2010-05-26 17:22:17 +0800 | [diff] [blame] | 71 | |
| 72 | #include <trace/events/kmem.h> |
| 73 | |
Arun Sharma | 60063497 | 2011-07-26 16:09:06 -0700 | [diff] [blame] | 74 | #include <linux/atomic.h> |
Matt Mackall | 10cef60 | 2006-01-08 01:01:45 -0800 | [diff] [blame] | 75 | |
Glauber Costa | b9ce5ef | 2012-12-18 14:22:46 -0800 | [diff] [blame] | 76 | #include "slab.h" |
Nick Piggin | 95b3512 | 2007-07-15 23:38:07 -0700 | [diff] [blame] | 77 | /* |
| 78 | * slob_block has a field 'units', which indicates size of block if +ve, |
| 79 | * or offset of next block if -ve (in SLOB_UNITs). |
| 80 | * |
| 81 | * Free blocks of size 1 unit simply contain the offset of the next block. |
| 82 | * Those with larger size contain their size in the first SLOB_UNIT of |
| 83 | * memory, and the offset of the next free block in the second SLOB_UNIT. |
| 84 | */ |
Nick Piggin | 5539484 | 2007-07-15 23:38:09 -0700 | [diff] [blame] | 85 | #if PAGE_SIZE <= (32767 * 2) |
Nick Piggin | 95b3512 | 2007-07-15 23:38:07 -0700 | [diff] [blame] | 86 | typedef s16 slobidx_t; |
| 87 | #else |
| 88 | typedef s32 slobidx_t; |
| 89 | #endif |
| 90 | |
Matt Mackall | 10cef60 | 2006-01-08 01:01:45 -0800 | [diff] [blame] | 91 | struct slob_block { |
Nick Piggin | 95b3512 | 2007-07-15 23:38:07 -0700 | [diff] [blame] | 92 | slobidx_t units; |
Nick Piggin | 5539484 | 2007-07-15 23:38:09 -0700 | [diff] [blame] | 93 | }; |
Matt Mackall | 10cef60 | 2006-01-08 01:01:45 -0800 | [diff] [blame] | 94 | typedef struct slob_block slob_t; |
| 95 | |
Nick Piggin | 95b3512 | 2007-07-15 23:38:07 -0700 | [diff] [blame] | 96 | /* |
Matt Mackall | 20cecba | 2008-02-04 22:29:37 -0800 | [diff] [blame] | 97 | * All partially free slob pages go on these lists. |
Nick Piggin | 95b3512 | 2007-07-15 23:38:07 -0700 | [diff] [blame] | 98 | */ |
Matt Mackall | 20cecba | 2008-02-04 22:29:37 -0800 | [diff] [blame] | 99 | #define SLOB_BREAK1 256 |
| 100 | #define SLOB_BREAK2 1024 |
| 101 | static LIST_HEAD(free_slob_small); |
| 102 | static LIST_HEAD(free_slob_medium); |
| 103 | static LIST_HEAD(free_slob_large); |
Nick Piggin | 95b3512 | 2007-07-15 23:38:07 -0700 | [diff] [blame] | 104 | |
| 105 | /* |
Nick Piggin | 95b3512 | 2007-07-15 23:38:07 -0700 | [diff] [blame] | 106 | * slob_page_free: true for pages on free_slob_pages list. |
| 107 | */ |
Christoph Lameter | b8c24c4 | 2012-06-13 10:24:52 -0500 | [diff] [blame] | 108 | static inline int slob_page_free(struct page *sp) |
Nick Piggin | 95b3512 | 2007-07-15 23:38:07 -0700 | [diff] [blame] | 109 | { |
Christoph Lameter | b8c24c4 | 2012-06-13 10:24:52 -0500 | [diff] [blame] | 110 | return PageSlobFree(sp); |
Nick Piggin | 95b3512 | 2007-07-15 23:38:07 -0700 | [diff] [blame] | 111 | } |
| 112 | |
Christoph Lameter | b8c24c4 | 2012-06-13 10:24:52 -0500 | [diff] [blame] | 113 | static void set_slob_page_free(struct page *sp, struct list_head *list) |
Nick Piggin | 95b3512 | 2007-07-15 23:38:07 -0700 | [diff] [blame] | 114 | { |
Tobin C. Harding | adab7b6 | 2019-05-13 17:16:06 -0700 | [diff] [blame] | 115 | list_add(&sp->slab_list, list); |
Christoph Lameter | b8c24c4 | 2012-06-13 10:24:52 -0500 | [diff] [blame] | 116 | __SetPageSlobFree(sp); |
Nick Piggin | 95b3512 | 2007-07-15 23:38:07 -0700 | [diff] [blame] | 117 | } |
| 118 | |
Christoph Lameter | b8c24c4 | 2012-06-13 10:24:52 -0500 | [diff] [blame] | 119 | static inline void clear_slob_page_free(struct page *sp) |
Nick Piggin | 95b3512 | 2007-07-15 23:38:07 -0700 | [diff] [blame] | 120 | { |
Tobin C. Harding | adab7b6 | 2019-05-13 17:16:06 -0700 | [diff] [blame] | 121 | list_del(&sp->slab_list); |
Christoph Lameter | b8c24c4 | 2012-06-13 10:24:52 -0500 | [diff] [blame] | 122 | __ClearPageSlobFree(sp); |
Nick Piggin | 95b3512 | 2007-07-15 23:38:07 -0700 | [diff] [blame] | 123 | } |
| 124 | |
Matt Mackall | 10cef60 | 2006-01-08 01:01:45 -0800 | [diff] [blame] | 125 | #define SLOB_UNIT sizeof(slob_t) |
Sasha Levin | a6d7815 | 2012-12-20 14:11:39 -0500 | [diff] [blame] | 126 | #define SLOB_UNITS(size) DIV_ROUND_UP(size, SLOB_UNIT) |
Matt Mackall | 10cef60 | 2006-01-08 01:01:45 -0800 | [diff] [blame] | 127 | |
Nick Piggin | afc0ced | 2007-05-16 22:10:49 -0700 | [diff] [blame] | 128 | /* |
| 129 | * struct slob_rcu is inserted at the tail of allocated slob blocks, which |
Paul E. McKenney | 5f0d5a3 | 2017-01-18 02:53:44 -0800 | [diff] [blame] | 130 | * were created with a SLAB_TYPESAFE_BY_RCU slab. slob_rcu is used to free |
Nick Piggin | afc0ced | 2007-05-16 22:10:49 -0700 | [diff] [blame] | 131 | * the block using call_rcu. |
| 132 | */ |
| 133 | struct slob_rcu { |
| 134 | struct rcu_head head; |
| 135 | int size; |
| 136 | }; |
| 137 | |
Nick Piggin | 95b3512 | 2007-07-15 23:38:07 -0700 | [diff] [blame] | 138 | /* |
| 139 | * slob_lock protects all slob allocator structures. |
| 140 | */ |
Matt Mackall | 10cef60 | 2006-01-08 01:01:45 -0800 | [diff] [blame] | 141 | static DEFINE_SPINLOCK(slob_lock); |
Matt Mackall | 10cef60 | 2006-01-08 01:01:45 -0800 | [diff] [blame] | 142 | |
Nick Piggin | 95b3512 | 2007-07-15 23:38:07 -0700 | [diff] [blame] | 143 | /* |
| 144 | * Encode the given size and next info into a free slob block s. |
| 145 | */ |
| 146 | static void set_slob(slob_t *s, slobidx_t size, slob_t *next) |
| 147 | { |
| 148 | slob_t *base = (slob_t *)((unsigned long)s & PAGE_MASK); |
| 149 | slobidx_t offset = next - base; |
Dimitri Gorokhovik | bcb4ddb | 2006-12-29 16:48:28 -0800 | [diff] [blame] | 150 | |
Nick Piggin | 95b3512 | 2007-07-15 23:38:07 -0700 | [diff] [blame] | 151 | if (size > 1) { |
| 152 | s[0].units = size; |
| 153 | s[1].units = offset; |
| 154 | } else |
| 155 | s[0].units = -offset; |
| 156 | } |
Matt Mackall | 10cef60 | 2006-01-08 01:01:45 -0800 | [diff] [blame] | 157 | |
Nick Piggin | 95b3512 | 2007-07-15 23:38:07 -0700 | [diff] [blame] | 158 | /* |
| 159 | * Return the size of a slob block. |
| 160 | */ |
| 161 | static slobidx_t slob_units(slob_t *s) |
| 162 | { |
| 163 | if (s->units > 0) |
| 164 | return s->units; |
| 165 | return 1; |
| 166 | } |
| 167 | |
| 168 | /* |
| 169 | * Return the next free slob block pointer after this one. |
| 170 | */ |
| 171 | static slob_t *slob_next(slob_t *s) |
| 172 | { |
| 173 | slob_t *base = (slob_t *)((unsigned long)s & PAGE_MASK); |
| 174 | slobidx_t next; |
| 175 | |
| 176 | if (s[0].units < 0) |
| 177 | next = -s[0].units; |
| 178 | else |
| 179 | next = s[1].units; |
| 180 | return base+next; |
| 181 | } |
| 182 | |
| 183 | /* |
| 184 | * Returns true if s is the last free block in its page. |
| 185 | */ |
| 186 | static int slob_last(slob_t *s) |
| 187 | { |
| 188 | return !((unsigned long)slob_next(s) & ~PAGE_MASK); |
| 189 | } |
| 190 | |
Américo Wang | 6e9ed0c | 2009-01-19 02:00:38 +0800 | [diff] [blame] | 191 | static void *slob_new_pages(gfp_t gfp, int order, int node) |
Paul Mundt | 6193a2f | 2007-07-15 23:38:22 -0700 | [diff] [blame] | 192 | { |
Vlastimil Babka | 6a486c0 | 2019-10-06 17:58:42 -0700 | [diff] [blame] | 193 | struct page *page; |
Paul Mundt | 6193a2f | 2007-07-15 23:38:22 -0700 | [diff] [blame] | 194 | |
| 195 | #ifdef CONFIG_NUMA |
Ezequiel Garcia | 90f2cbb | 2012-09-08 17:47:51 -0300 | [diff] [blame] | 196 | if (node != NUMA_NO_NODE) |
Vlastimil Babka | 96db800 | 2015-09-08 15:03:50 -0700 | [diff] [blame] | 197 | page = __alloc_pages_node(node, gfp, order); |
Paul Mundt | 6193a2f | 2007-07-15 23:38:22 -0700 | [diff] [blame] | 198 | else |
| 199 | #endif |
| 200 | page = alloc_pages(gfp, order); |
| 201 | |
| 202 | if (!page) |
| 203 | return NULL; |
| 204 | |
Vlastimil Babka | 6a486c0 | 2019-10-06 17:58:42 -0700 | [diff] [blame] | 205 | mod_node_page_state(page_pgdat(page), NR_SLAB_UNRECLAIMABLE, |
| 206 | 1 << order); |
Paul Mundt | 6193a2f | 2007-07-15 23:38:22 -0700 | [diff] [blame] | 207 | return page_address(page); |
| 208 | } |
| 209 | |
Américo Wang | 6e9ed0c | 2009-01-19 02:00:38 +0800 | [diff] [blame] | 210 | static void slob_free_pages(void *b, int order) |
| 211 | { |
Vlastimil Babka | 6a486c0 | 2019-10-06 17:58:42 -0700 | [diff] [blame] | 212 | struct page *sp = virt_to_page(b); |
| 213 | |
Nick Piggin | 1f0532e | 2009-05-05 19:13:45 +1000 | [diff] [blame] | 214 | if (current->reclaim_state) |
| 215 | current->reclaim_state->reclaimed_slab += 1 << order; |
Vlastimil Babka | 6a486c0 | 2019-10-06 17:58:42 -0700 | [diff] [blame] | 216 | |
| 217 | mod_node_page_state(page_pgdat(sp), NR_SLAB_UNRECLAIMABLE, |
| 218 | -(1 << order)); |
| 219 | __free_pages(sp, order); |
Américo Wang | 6e9ed0c | 2009-01-19 02:00:38 +0800 | [diff] [blame] | 220 | } |
| 221 | |
Nick Piggin | 95b3512 | 2007-07-15 23:38:07 -0700 | [diff] [blame] | 222 | /* |
Tobin C. Harding | 130e8e0 | 2019-05-13 17:16:03 -0700 | [diff] [blame] | 223 | * slob_page_alloc() - Allocate a slob block within a given slob_page sp. |
| 224 | * @sp: Page to look in. |
| 225 | * @size: Size of the allocation. |
| 226 | * @align: Allocation alignment. |
Vlastimil Babka | 59bb479 | 2019-10-06 17:58:45 -0700 | [diff] [blame] | 227 | * @align_offset: Offset in the allocated block that will be aligned. |
Tobin C. Harding | 130e8e0 | 2019-05-13 17:16:03 -0700 | [diff] [blame] | 228 | * @page_removed_from_list: Return parameter. |
| 229 | * |
| 230 | * Tries to find a chunk of memory at least @size bytes big within @page. |
| 231 | * |
| 232 | * Return: Pointer to memory if allocated, %NULL otherwise. If the |
| 233 | * allocation fills up @page then the page is removed from the |
| 234 | * freelist, in this case @page_removed_from_list will be set to |
| 235 | * true (set to false otherwise). |
Nick Piggin | 95b3512 | 2007-07-15 23:38:07 -0700 | [diff] [blame] | 236 | */ |
Tobin C. Harding | 130e8e0 | 2019-05-13 17:16:03 -0700 | [diff] [blame] | 237 | static void *slob_page_alloc(struct page *sp, size_t size, int align, |
Vlastimil Babka | 59bb479 | 2019-10-06 17:58:45 -0700 | [diff] [blame] | 238 | int align_offset, bool *page_removed_from_list) |
Matt Mackall | 10cef60 | 2006-01-08 01:01:45 -0800 | [diff] [blame] | 239 | { |
Américo Wang | 6e9ed0c | 2009-01-19 02:00:38 +0800 | [diff] [blame] | 240 | slob_t *prev, *cur, *aligned = NULL; |
Matt Mackall | 10cef60 | 2006-01-08 01:01:45 -0800 | [diff] [blame] | 241 | int delta = 0, units = SLOB_UNITS(size); |
Matt Mackall | 10cef60 | 2006-01-08 01:01:45 -0800 | [diff] [blame] | 242 | |
Tobin C. Harding | 130e8e0 | 2019-05-13 17:16:03 -0700 | [diff] [blame] | 243 | *page_removed_from_list = false; |
Christoph Lameter | b8c24c4 | 2012-06-13 10:24:52 -0500 | [diff] [blame] | 244 | for (prev = NULL, cur = sp->freelist; ; prev = cur, cur = slob_next(cur)) { |
Nick Piggin | 95b3512 | 2007-07-15 23:38:07 -0700 | [diff] [blame] | 245 | slobidx_t avail = slob_units(cur); |
| 246 | |
Vlastimil Babka | 59bb479 | 2019-10-06 17:58:45 -0700 | [diff] [blame] | 247 | /* |
| 248 | * 'aligned' will hold the address of the slob block so that the |
| 249 | * address 'aligned'+'align_offset' is aligned according to the |
| 250 | * 'align' parameter. This is for kmalloc() which prepends the |
| 251 | * allocated block with its size, so that the block itself is |
| 252 | * aligned when needed. |
| 253 | */ |
Matt Mackall | 10cef60 | 2006-01-08 01:01:45 -0800 | [diff] [blame] | 254 | if (align) { |
Vlastimil Babka | 59bb479 | 2019-10-06 17:58:45 -0700 | [diff] [blame] | 255 | aligned = (slob_t *) |
| 256 | (ALIGN((unsigned long)cur + align_offset, align) |
| 257 | - align_offset); |
Matt Mackall | 10cef60 | 2006-01-08 01:01:45 -0800 | [diff] [blame] | 258 | delta = aligned - cur; |
| 259 | } |
Nick Piggin | 95b3512 | 2007-07-15 23:38:07 -0700 | [diff] [blame] | 260 | if (avail >= units + delta) { /* room enough? */ |
| 261 | slob_t *next; |
| 262 | |
Matt Mackall | 10cef60 | 2006-01-08 01:01:45 -0800 | [diff] [blame] | 263 | if (delta) { /* need to fragment head to align? */ |
Nick Piggin | 95b3512 | 2007-07-15 23:38:07 -0700 | [diff] [blame] | 264 | next = slob_next(cur); |
| 265 | set_slob(aligned, avail - delta, next); |
| 266 | set_slob(cur, delta, aligned); |
Matt Mackall | 10cef60 | 2006-01-08 01:01:45 -0800 | [diff] [blame] | 267 | prev = cur; |
| 268 | cur = aligned; |
Nick Piggin | 95b3512 | 2007-07-15 23:38:07 -0700 | [diff] [blame] | 269 | avail = slob_units(cur); |
Matt Mackall | 10cef60 | 2006-01-08 01:01:45 -0800 | [diff] [blame] | 270 | } |
| 271 | |
Nick Piggin | 95b3512 | 2007-07-15 23:38:07 -0700 | [diff] [blame] | 272 | next = slob_next(cur); |
| 273 | if (avail == units) { /* exact fit? unlink. */ |
| 274 | if (prev) |
| 275 | set_slob(prev, slob_units(prev), next); |
| 276 | else |
Christoph Lameter | b8c24c4 | 2012-06-13 10:24:52 -0500 | [diff] [blame] | 277 | sp->freelist = next; |
Nick Piggin | 95b3512 | 2007-07-15 23:38:07 -0700 | [diff] [blame] | 278 | } else { /* fragment */ |
| 279 | if (prev) |
| 280 | set_slob(prev, slob_units(prev), cur + units); |
| 281 | else |
Christoph Lameter | b8c24c4 | 2012-06-13 10:24:52 -0500 | [diff] [blame] | 282 | sp->freelist = cur + units; |
Nick Piggin | 95b3512 | 2007-07-15 23:38:07 -0700 | [diff] [blame] | 283 | set_slob(cur + units, avail - units, next); |
Matt Mackall | 10cef60 | 2006-01-08 01:01:45 -0800 | [diff] [blame] | 284 | } |
| 285 | |
Nick Piggin | 95b3512 | 2007-07-15 23:38:07 -0700 | [diff] [blame] | 286 | sp->units -= units; |
Tobin C. Harding | 130e8e0 | 2019-05-13 17:16:03 -0700 | [diff] [blame] | 287 | if (!sp->units) { |
Nick Piggin | 95b3512 | 2007-07-15 23:38:07 -0700 | [diff] [blame] | 288 | clear_slob_page_free(sp); |
Tobin C. Harding | 130e8e0 | 2019-05-13 17:16:03 -0700 | [diff] [blame] | 289 | *page_removed_from_list = true; |
| 290 | } |
Matt Mackall | 10cef60 | 2006-01-08 01:01:45 -0800 | [diff] [blame] | 291 | return cur; |
| 292 | } |
Nick Piggin | 95b3512 | 2007-07-15 23:38:07 -0700 | [diff] [blame] | 293 | if (slob_last(cur)) |
| 294 | return NULL; |
Matt Mackall | 10cef60 | 2006-01-08 01:01:45 -0800 | [diff] [blame] | 295 | } |
| 296 | } |
| 297 | |
Nick Piggin | 95b3512 | 2007-07-15 23:38:07 -0700 | [diff] [blame] | 298 | /* |
| 299 | * slob_alloc: entry point into the slob allocator. |
| 300 | */ |
Vlastimil Babka | 59bb479 | 2019-10-06 17:58:45 -0700 | [diff] [blame] | 301 | static void *slob_alloc(size_t size, gfp_t gfp, int align, int node, |
| 302 | int align_offset) |
Nick Piggin | 95b3512 | 2007-07-15 23:38:07 -0700 | [diff] [blame] | 303 | { |
Christoph Lameter | b8c24c4 | 2012-06-13 10:24:52 -0500 | [diff] [blame] | 304 | struct page *sp; |
Matt Mackall | 20cecba | 2008-02-04 22:29:37 -0800 | [diff] [blame] | 305 | struct list_head *slob_list; |
Nick Piggin | 95b3512 | 2007-07-15 23:38:07 -0700 | [diff] [blame] | 306 | slob_t *b = NULL; |
| 307 | unsigned long flags; |
Tobin C. Harding | 130e8e0 | 2019-05-13 17:16:03 -0700 | [diff] [blame] | 308 | bool _unused; |
Nick Piggin | 95b3512 | 2007-07-15 23:38:07 -0700 | [diff] [blame] | 309 | |
Matt Mackall | 20cecba | 2008-02-04 22:29:37 -0800 | [diff] [blame] | 310 | if (size < SLOB_BREAK1) |
| 311 | slob_list = &free_slob_small; |
| 312 | else if (size < SLOB_BREAK2) |
| 313 | slob_list = &free_slob_medium; |
| 314 | else |
| 315 | slob_list = &free_slob_large; |
| 316 | |
Nick Piggin | 95b3512 | 2007-07-15 23:38:07 -0700 | [diff] [blame] | 317 | spin_lock_irqsave(&slob_lock, flags); |
| 318 | /* Iterate through each partially free page, try to find room */ |
Tobin C. Harding | adab7b6 | 2019-05-13 17:16:06 -0700 | [diff] [blame] | 319 | list_for_each_entry(sp, slob_list, slab_list) { |
Tobin C. Harding | 130e8e0 | 2019-05-13 17:16:03 -0700 | [diff] [blame] | 320 | bool page_removed_from_list = false; |
Paul Mundt | 6193a2f | 2007-07-15 23:38:22 -0700 | [diff] [blame] | 321 | #ifdef CONFIG_NUMA |
| 322 | /* |
| 323 | * If there's a node specification, search for a partial |
| 324 | * page with a matching node id in the freelist. |
| 325 | */ |
Ezequiel Garcia | 90f2cbb | 2012-09-08 17:47:51 -0300 | [diff] [blame] | 326 | if (node != NUMA_NO_NODE && page_to_nid(sp) != node) |
Paul Mundt | 6193a2f | 2007-07-15 23:38:22 -0700 | [diff] [blame] | 327 | continue; |
| 328 | #endif |
Matt Mackall | d626954 | 2007-07-21 04:37:40 -0700 | [diff] [blame] | 329 | /* Enough room on this page? */ |
| 330 | if (sp->units < SLOB_UNITS(size)) |
| 331 | continue; |
Paul Mundt | 6193a2f | 2007-07-15 23:38:22 -0700 | [diff] [blame] | 332 | |
Vlastimil Babka | 59bb479 | 2019-10-06 17:58:45 -0700 | [diff] [blame] | 333 | b = slob_page_alloc(sp, size, align, align_offset, &page_removed_from_list); |
Matt Mackall | d626954 | 2007-07-21 04:37:40 -0700 | [diff] [blame] | 334 | if (!b) |
| 335 | continue; |
| 336 | |
Tobin C. Harding | 130e8e0 | 2019-05-13 17:16:03 -0700 | [diff] [blame] | 337 | /* |
| 338 | * If slob_page_alloc() removed sp from the list then we |
| 339 | * cannot call list functions on sp. If so allocation |
| 340 | * did not fragment the page anyway so optimisation is |
| 341 | * unnecessary. |
| 342 | */ |
| 343 | if (!page_removed_from_list) { |
| 344 | /* |
| 345 | * Improve fragment distribution and reduce our average |
| 346 | * search time by starting our next search here. (see |
| 347 | * Knuth vol 1, sec 2.5, pg 449) |
| 348 | */ |
Tobin C. Harding | adab7b6 | 2019-05-13 17:16:06 -0700 | [diff] [blame] | 349 | if (!list_is_first(&sp->slab_list, slob_list)) |
| 350 | list_rotate_to_front(&sp->slab_list, slob_list); |
Tobin C. Harding | 130e8e0 | 2019-05-13 17:16:03 -0700 | [diff] [blame] | 351 | } |
Matt Mackall | d626954 | 2007-07-21 04:37:40 -0700 | [diff] [blame] | 352 | break; |
Nick Piggin | 95b3512 | 2007-07-15 23:38:07 -0700 | [diff] [blame] | 353 | } |
| 354 | spin_unlock_irqrestore(&slob_lock, flags); |
| 355 | |
| 356 | /* Not enough space: must allocate a new page */ |
| 357 | if (!b) { |
Américo Wang | 6e9ed0c | 2009-01-19 02:00:38 +0800 | [diff] [blame] | 358 | b = slob_new_pages(gfp & ~__GFP_ZERO, 0, node); |
Nick Piggin | 95b3512 | 2007-07-15 23:38:07 -0700 | [diff] [blame] | 359 | if (!b) |
Américo Wang | 6e9ed0c | 2009-01-19 02:00:38 +0800 | [diff] [blame] | 360 | return NULL; |
Christoph Lameter | b556828 | 2012-06-13 10:24:54 -0500 | [diff] [blame] | 361 | sp = virt_to_page(b); |
| 362 | __SetPageSlab(sp); |
Nick Piggin | 95b3512 | 2007-07-15 23:38:07 -0700 | [diff] [blame] | 363 | |
| 364 | spin_lock_irqsave(&slob_lock, flags); |
| 365 | sp->units = SLOB_UNITS(PAGE_SIZE); |
Christoph Lameter | b8c24c4 | 2012-06-13 10:24:52 -0500 | [diff] [blame] | 366 | sp->freelist = b; |
Tobin C. Harding | adab7b6 | 2019-05-13 17:16:06 -0700 | [diff] [blame] | 367 | INIT_LIST_HEAD(&sp->slab_list); |
Nick Piggin | 95b3512 | 2007-07-15 23:38:07 -0700 | [diff] [blame] | 368 | set_slob(b, SLOB_UNITS(PAGE_SIZE), b + SLOB_UNITS(PAGE_SIZE)); |
Matt Mackall | 20cecba | 2008-02-04 22:29:37 -0800 | [diff] [blame] | 369 | set_slob_page_free(sp, slob_list); |
Vlastimil Babka | 59bb479 | 2019-10-06 17:58:45 -0700 | [diff] [blame] | 370 | b = slob_page_alloc(sp, size, align, align_offset, &_unused); |
Nick Piggin | 95b3512 | 2007-07-15 23:38:07 -0700 | [diff] [blame] | 371 | BUG_ON(!b); |
| 372 | spin_unlock_irqrestore(&slob_lock, flags); |
| 373 | } |
Miles Chen | 9f88fae | 2017-11-15 17:32:10 -0800 | [diff] [blame] | 374 | if (unlikely(gfp & __GFP_ZERO)) |
Christoph Lameter | d07dbea | 2007-07-17 04:03:23 -0700 | [diff] [blame] | 375 | memset(b, 0, size); |
Nick Piggin | 95b3512 | 2007-07-15 23:38:07 -0700 | [diff] [blame] | 376 | return b; |
| 377 | } |
| 378 | |
| 379 | /* |
| 380 | * slob_free: entry point into the slob allocator. |
| 381 | */ |
Matt Mackall | 10cef60 | 2006-01-08 01:01:45 -0800 | [diff] [blame] | 382 | static void slob_free(void *block, int size) |
| 383 | { |
Christoph Lameter | b8c24c4 | 2012-06-13 10:24:52 -0500 | [diff] [blame] | 384 | struct page *sp; |
Nick Piggin | 95b3512 | 2007-07-15 23:38:07 -0700 | [diff] [blame] | 385 | slob_t *prev, *next, *b = (slob_t *)block; |
| 386 | slobidx_t units; |
Matt Mackall | 10cef60 | 2006-01-08 01:01:45 -0800 | [diff] [blame] | 387 | unsigned long flags; |
Bob Liu | d602dab | 2010-07-10 18:05:33 +0800 | [diff] [blame] | 388 | struct list_head *slob_list; |
Matt Mackall | 10cef60 | 2006-01-08 01:01:45 -0800 | [diff] [blame] | 389 | |
Satyam Sharma | 2408c55 | 2007-10-16 01:24:44 -0700 | [diff] [blame] | 390 | if (unlikely(ZERO_OR_NULL_PTR(block))) |
Matt Mackall | 10cef60 | 2006-01-08 01:01:45 -0800 | [diff] [blame] | 391 | return; |
Nick Piggin | 95b3512 | 2007-07-15 23:38:07 -0700 | [diff] [blame] | 392 | BUG_ON(!size); |
Matt Mackall | 10cef60 | 2006-01-08 01:01:45 -0800 | [diff] [blame] | 393 | |
Christoph Lameter | b556828 | 2012-06-13 10:24:54 -0500 | [diff] [blame] | 394 | sp = virt_to_page(block); |
Nick Piggin | 95b3512 | 2007-07-15 23:38:07 -0700 | [diff] [blame] | 395 | units = SLOB_UNITS(size); |
Matt Mackall | 10cef60 | 2006-01-08 01:01:45 -0800 | [diff] [blame] | 396 | |
Matt Mackall | 10cef60 | 2006-01-08 01:01:45 -0800 | [diff] [blame] | 397 | spin_lock_irqsave(&slob_lock, flags); |
Matt Mackall | 10cef60 | 2006-01-08 01:01:45 -0800 | [diff] [blame] | 398 | |
Nick Piggin | 95b3512 | 2007-07-15 23:38:07 -0700 | [diff] [blame] | 399 | if (sp->units + units == SLOB_UNITS(PAGE_SIZE)) { |
| 400 | /* Go directly to page allocator. Do not pass slob allocator */ |
| 401 | if (slob_page_free(sp)) |
| 402 | clear_slob_page_free(sp); |
Nick Piggin | 6fb8f42 | 2009-03-16 21:00:28 +1100 | [diff] [blame] | 403 | spin_unlock_irqrestore(&slob_lock, flags); |
Christoph Lameter | b556828 | 2012-06-13 10:24:54 -0500 | [diff] [blame] | 404 | __ClearPageSlab(sp); |
Mel Gorman | 22b751c | 2013-02-22 16:34:59 -0800 | [diff] [blame] | 405 | page_mapcount_reset(sp); |
Nick Piggin | 1f0532e | 2009-05-05 19:13:45 +1000 | [diff] [blame] | 406 | slob_free_pages(b, 0); |
Nick Piggin | 6fb8f42 | 2009-03-16 21:00:28 +1100 | [diff] [blame] | 407 | return; |
Nick Piggin | 95b3512 | 2007-07-15 23:38:07 -0700 | [diff] [blame] | 408 | } |
Matt Mackall | 10cef60 | 2006-01-08 01:01:45 -0800 | [diff] [blame] | 409 | |
Nick Piggin | 95b3512 | 2007-07-15 23:38:07 -0700 | [diff] [blame] | 410 | if (!slob_page_free(sp)) { |
| 411 | /* This slob page is about to become partially free. Easy! */ |
| 412 | sp->units = units; |
Christoph Lameter | b8c24c4 | 2012-06-13 10:24:52 -0500 | [diff] [blame] | 413 | sp->freelist = b; |
Nick Piggin | 95b3512 | 2007-07-15 23:38:07 -0700 | [diff] [blame] | 414 | set_slob(b, units, |
| 415 | (void *)((unsigned long)(b + |
| 416 | SLOB_UNITS(PAGE_SIZE)) & PAGE_MASK)); |
Bob Liu | d602dab | 2010-07-10 18:05:33 +0800 | [diff] [blame] | 417 | if (size < SLOB_BREAK1) |
| 418 | slob_list = &free_slob_small; |
| 419 | else if (size < SLOB_BREAK2) |
| 420 | slob_list = &free_slob_medium; |
| 421 | else |
| 422 | slob_list = &free_slob_large; |
| 423 | set_slob_page_free(sp, slob_list); |
Nick Piggin | 95b3512 | 2007-07-15 23:38:07 -0700 | [diff] [blame] | 424 | goto out; |
| 425 | } |
Matt Mackall | 10cef60 | 2006-01-08 01:01:45 -0800 | [diff] [blame] | 426 | |
Nick Piggin | 95b3512 | 2007-07-15 23:38:07 -0700 | [diff] [blame] | 427 | /* |
| 428 | * Otherwise the page is already partially free, so find reinsertion |
| 429 | * point. |
| 430 | */ |
| 431 | sp->units += units; |
Matt Mackall | 10cef60 | 2006-01-08 01:01:45 -0800 | [diff] [blame] | 432 | |
Christoph Lameter | b8c24c4 | 2012-06-13 10:24:52 -0500 | [diff] [blame] | 433 | if (b < (slob_t *)sp->freelist) { |
| 434 | if (b + units == sp->freelist) { |
| 435 | units += slob_units(sp->freelist); |
| 436 | sp->freelist = slob_next(sp->freelist); |
Matt Mackall | 679299b | 2008-02-04 22:29:37 -0800 | [diff] [blame] | 437 | } |
Christoph Lameter | b8c24c4 | 2012-06-13 10:24:52 -0500 | [diff] [blame] | 438 | set_slob(b, units, sp->freelist); |
| 439 | sp->freelist = b; |
Nick Piggin | 95b3512 | 2007-07-15 23:38:07 -0700 | [diff] [blame] | 440 | } else { |
Christoph Lameter | b8c24c4 | 2012-06-13 10:24:52 -0500 | [diff] [blame] | 441 | prev = sp->freelist; |
Nick Piggin | 95b3512 | 2007-07-15 23:38:07 -0700 | [diff] [blame] | 442 | next = slob_next(prev); |
| 443 | while (b > next) { |
| 444 | prev = next; |
| 445 | next = slob_next(prev); |
| 446 | } |
| 447 | |
| 448 | if (!slob_last(prev) && b + units == next) { |
| 449 | units += slob_units(next); |
| 450 | set_slob(b, units, slob_next(next)); |
| 451 | } else |
| 452 | set_slob(b, units, next); |
| 453 | |
| 454 | if (prev + slob_units(prev) == b) { |
| 455 | units = slob_units(b) + slob_units(prev); |
| 456 | set_slob(prev, units, slob_next(b)); |
| 457 | } else |
| 458 | set_slob(prev, slob_units(prev), b); |
| 459 | } |
| 460 | out: |
Matt Mackall | 10cef60 | 2006-01-08 01:01:45 -0800 | [diff] [blame] | 461 | spin_unlock_irqrestore(&slob_lock, flags); |
| 462 | } |
| 463 | |
Nick Piggin | 95b3512 | 2007-07-15 23:38:07 -0700 | [diff] [blame] | 464 | /* |
| 465 | * End of slob allocator proper. Begin kmem_cache_alloc and kmalloc frontend. |
| 466 | */ |
| 467 | |
Ezequiel Garcia | f3f7410 | 2012-09-08 17:47:53 -0300 | [diff] [blame] | 468 | static __always_inline void * |
| 469 | __do_kmalloc_node(size_t size, gfp_t gfp, int node, unsigned long caller) |
Matt Mackall | 10cef60 | 2006-01-08 01:01:45 -0800 | [diff] [blame] | 470 | { |
Christoph Lameter | 6cb8f91 | 2007-07-17 04:03:22 -0700 | [diff] [blame] | 471 | unsigned int *m; |
Vlastimil Babka | 59bb479 | 2019-10-06 17:58:45 -0700 | [diff] [blame] | 472 | int minalign = max_t(size_t, ARCH_KMALLOC_MINALIGN, ARCH_SLAB_MINALIGN); |
Eduard - Gabriel Munteanu | 3eae2cb2 | 2008-08-10 20:14:07 +0300 | [diff] [blame] | 473 | void *ret; |
Nick Piggin | 5539484 | 2007-07-15 23:38:09 -0700 | [diff] [blame] | 474 | |
Steven Rostedt | bd50cfa | 2011-06-07 07:18:45 -0400 | [diff] [blame] | 475 | gfp &= gfp_allowed_mask; |
| 476 | |
Peter Zijlstra | d92a8cf | 2017-03-03 10:13:38 +0100 | [diff] [blame] | 477 | fs_reclaim_acquire(gfp); |
| 478 | fs_reclaim_release(gfp); |
Nick Piggin | cf40bd1 | 2009-01-21 08:12:39 +0100 | [diff] [blame] | 479 | |
Vlastimil Babka | 59bb479 | 2019-10-06 17:58:45 -0700 | [diff] [blame] | 480 | if (size < PAGE_SIZE - minalign) { |
| 481 | int align = minalign; |
| 482 | |
| 483 | /* |
| 484 | * For power of two sizes, guarantee natural alignment for |
| 485 | * kmalloc()'d objects. |
| 486 | */ |
| 487 | if (is_power_of_2(size)) |
| 488 | align = max(minalign, (int) size); |
| 489 | |
Christoph Lameter | 6cb8f91 | 2007-07-17 04:03:22 -0700 | [diff] [blame] | 490 | if (!size) |
| 491 | return ZERO_SIZE_PTR; |
| 492 | |
Vlastimil Babka | 59bb479 | 2019-10-06 17:58:45 -0700 | [diff] [blame] | 493 | m = slob_alloc(size + minalign, gfp, align, node, minalign); |
Eduard - Gabriel Munteanu | 3eae2cb2 | 2008-08-10 20:14:07 +0300 | [diff] [blame] | 494 | |
MinChan Kim | 239f49c | 2008-05-19 22:12:08 +0900 | [diff] [blame] | 495 | if (!m) |
| 496 | return NULL; |
| 497 | *m = size; |
Vlastimil Babka | 59bb479 | 2019-10-06 17:58:45 -0700 | [diff] [blame] | 498 | ret = (void *)m + minalign; |
Nick Piggin | d87a133 | 2007-07-15 23:38:08 -0700 | [diff] [blame] | 499 | |
Ezequiel Garcia | f3f7410 | 2012-09-08 17:47:53 -0300 | [diff] [blame] | 500 | trace_kmalloc_node(caller, ret, |
Vlastimil Babka | 59bb479 | 2019-10-06 17:58:45 -0700 | [diff] [blame] | 501 | size, size + minalign, gfp, node); |
Nick Piggin | d87a133 | 2007-07-15 23:38:08 -0700 | [diff] [blame] | 502 | } else { |
Eduard - Gabriel Munteanu | 3eae2cb2 | 2008-08-10 20:14:07 +0300 | [diff] [blame] | 503 | unsigned int order = get_order(size); |
Nick Piggin | d87a133 | 2007-07-15 23:38:08 -0700 | [diff] [blame] | 504 | |
David Rientjes | 8df275a | 2010-08-22 16:16:06 -0700 | [diff] [blame] | 505 | if (likely(order)) |
| 506 | gfp |= __GFP_COMP; |
| 507 | ret = slob_new_pages(gfp, order, node); |
Eduard - Gabriel Munteanu | 3eae2cb2 | 2008-08-10 20:14:07 +0300 | [diff] [blame] | 508 | |
Ezequiel Garcia | f3f7410 | 2012-09-08 17:47:53 -0300 | [diff] [blame] | 509 | trace_kmalloc_node(caller, ret, |
Eduard - Gabriel Munteanu | ca2b84cb | 2009-03-23 15:12:24 +0200 | [diff] [blame] | 510 | size, PAGE_SIZE << order, gfp, node); |
Matt Mackall | 10cef60 | 2006-01-08 01:01:45 -0800 | [diff] [blame] | 511 | } |
Eduard - Gabriel Munteanu | 3eae2cb2 | 2008-08-10 20:14:07 +0300 | [diff] [blame] | 512 | |
Catalin Marinas | 4374e61 | 2009-06-11 13:23:17 +0100 | [diff] [blame] | 513 | kmemleak_alloc(ret, size, 1, gfp); |
Eduard - Gabriel Munteanu | 3eae2cb2 | 2008-08-10 20:14:07 +0300 | [diff] [blame] | 514 | return ret; |
Matt Mackall | 10cef60 | 2006-01-08 01:01:45 -0800 | [diff] [blame] | 515 | } |
Ezequiel Garcia | f3f7410 | 2012-09-08 17:47:53 -0300 | [diff] [blame] | 516 | |
Christoph Lameter | f1b6eb6 | 2013-09-04 16:35:34 +0000 | [diff] [blame] | 517 | void *__kmalloc(size_t size, gfp_t gfp) |
Ezequiel Garcia | f3f7410 | 2012-09-08 17:47:53 -0300 | [diff] [blame] | 518 | { |
Christoph Lameter | f1b6eb6 | 2013-09-04 16:35:34 +0000 | [diff] [blame] | 519 | return __do_kmalloc_node(size, gfp, NUMA_NO_NODE, _RET_IP_); |
Ezequiel Garcia | f3f7410 | 2012-09-08 17:47:53 -0300 | [diff] [blame] | 520 | } |
Christoph Lameter | f1b6eb6 | 2013-09-04 16:35:34 +0000 | [diff] [blame] | 521 | EXPORT_SYMBOL(__kmalloc); |
Matt Mackall | 10cef60 | 2006-01-08 01:01:45 -0800 | [diff] [blame] | 522 | |
Ezequiel Garcia | f3f7410 | 2012-09-08 17:47:53 -0300 | [diff] [blame] | 523 | void *__kmalloc_track_caller(size_t size, gfp_t gfp, unsigned long caller) |
| 524 | { |
| 525 | return __do_kmalloc_node(size, gfp, NUMA_NO_NODE, caller); |
| 526 | } |
| 527 | |
| 528 | #ifdef CONFIG_NUMA |
David Rientjes | 82bd550 | 2012-09-25 12:53:51 -0700 | [diff] [blame] | 529 | void *__kmalloc_node_track_caller(size_t size, gfp_t gfp, |
Ezequiel Garcia | f3f7410 | 2012-09-08 17:47:53 -0300 | [diff] [blame] | 530 | int node, unsigned long caller) |
| 531 | { |
| 532 | return __do_kmalloc_node(size, gfp, node, caller); |
| 533 | } |
| 534 | #endif |
Ezequiel Garcia | f3f7410 | 2012-09-08 17:47:53 -0300 | [diff] [blame] | 535 | |
Matt Mackall | 10cef60 | 2006-01-08 01:01:45 -0800 | [diff] [blame] | 536 | void kfree(const void *block) |
| 537 | { |
Christoph Lameter | b8c24c4 | 2012-06-13 10:24:52 -0500 | [diff] [blame] | 538 | struct page *sp; |
Matt Mackall | 10cef60 | 2006-01-08 01:01:45 -0800 | [diff] [blame] | 539 | |
Pekka Enberg | 2121db7 | 2009-03-25 11:05:57 +0200 | [diff] [blame] | 540 | trace_kfree(_RET_IP_, block); |
| 541 | |
Satyam Sharma | 2408c55 | 2007-10-16 01:24:44 -0700 | [diff] [blame] | 542 | if (unlikely(ZERO_OR_NULL_PTR(block))) |
Matt Mackall | 10cef60 | 2006-01-08 01:01:45 -0800 | [diff] [blame] | 543 | return; |
Catalin Marinas | 4374e61 | 2009-06-11 13:23:17 +0100 | [diff] [blame] | 544 | kmemleak_free(block); |
Matt Mackall | 10cef60 | 2006-01-08 01:01:45 -0800 | [diff] [blame] | 545 | |
Christoph Lameter | b556828 | 2012-06-13 10:24:54 -0500 | [diff] [blame] | 546 | sp = virt_to_page(block); |
| 547 | if (PageSlab(sp)) { |
Arnd Bergmann | 789306e | 2012-10-05 16:55:20 +0200 | [diff] [blame] | 548 | int align = max_t(size_t, ARCH_KMALLOC_MINALIGN, ARCH_SLAB_MINALIGN); |
Nick Piggin | 5539484 | 2007-07-15 23:38:09 -0700 | [diff] [blame] | 549 | unsigned int *m = (unsigned int *)(block - align); |
| 550 | slob_free(m, *m + align); |
Vlastimil Babka | 6a486c0 | 2019-10-06 17:58:42 -0700 | [diff] [blame] | 551 | } else { |
| 552 | unsigned int order = compound_order(sp); |
| 553 | mod_node_page_state(page_pgdat(sp), NR_SLAB_UNRECLAIMABLE, |
| 554 | -(1 << order)); |
| 555 | __free_pages(sp, order); |
| 556 | |
| 557 | } |
Matt Mackall | 10cef60 | 2006-01-08 01:01:45 -0800 | [diff] [blame] | 558 | } |
Matt Mackall | 10cef60 | 2006-01-08 01:01:45 -0800 | [diff] [blame] | 559 | EXPORT_SYMBOL(kfree); |
| 560 | |
Nick Piggin | d87a133 | 2007-07-15 23:38:08 -0700 | [diff] [blame] | 561 | /* can't use ksize for kmem_cache_alloc memory, only kmalloc */ |
Marco Elver | 10d1f8c | 2019-07-11 20:54:14 -0700 | [diff] [blame] | 562 | size_t __ksize(const void *block) |
Matt Mackall | 10cef60 | 2006-01-08 01:01:45 -0800 | [diff] [blame] | 563 | { |
Christoph Lameter | b8c24c4 | 2012-06-13 10:24:52 -0500 | [diff] [blame] | 564 | struct page *sp; |
Ezequiel Garcia | 999d879 | 2012-10-19 09:33:10 -0300 | [diff] [blame] | 565 | int align; |
| 566 | unsigned int *m; |
Matt Mackall | 10cef60 | 2006-01-08 01:01:45 -0800 | [diff] [blame] | 567 | |
Christoph Lameter | ef8b452 | 2007-10-16 01:24:46 -0700 | [diff] [blame] | 568 | BUG_ON(!block); |
| 569 | if (unlikely(block == ZERO_SIZE_PTR)) |
Matt Mackall | 10cef60 | 2006-01-08 01:01:45 -0800 | [diff] [blame] | 570 | return 0; |
| 571 | |
Christoph Lameter | b556828 | 2012-06-13 10:24:54 -0500 | [diff] [blame] | 572 | sp = virt_to_page(block); |
Ezequiel Garcia | 999d879 | 2012-10-19 09:33:10 -0300 | [diff] [blame] | 573 | if (unlikely(!PageSlab(sp))) |
Matthew Wilcox (Oracle) | a50b854 | 2019-09-23 15:34:25 -0700 | [diff] [blame] | 574 | return page_size(sp); |
Ezequiel Garcia | 999d879 | 2012-10-19 09:33:10 -0300 | [diff] [blame] | 575 | |
Arnd Bergmann | 789306e | 2012-10-05 16:55:20 +0200 | [diff] [blame] | 576 | align = max_t(size_t, ARCH_KMALLOC_MINALIGN, ARCH_SLAB_MINALIGN); |
Ezequiel Garcia | 999d879 | 2012-10-19 09:33:10 -0300 | [diff] [blame] | 577 | m = (unsigned int *)(block - align); |
| 578 | return SLOB_UNITS(*m) * SLOB_UNIT; |
Matt Mackall | 10cef60 | 2006-01-08 01:01:45 -0800 | [diff] [blame] | 579 | } |
Marco Elver | 10d1f8c | 2019-07-11 20:54:14 -0700 | [diff] [blame] | 580 | EXPORT_SYMBOL(__ksize); |
Matt Mackall | 10cef60 | 2006-01-08 01:01:45 -0800 | [diff] [blame] | 581 | |
Alexey Dobriyan | d50112e | 2017-11-15 17:32:18 -0800 | [diff] [blame] | 582 | int __kmem_cache_create(struct kmem_cache *c, slab_flags_t flags) |
Matt Mackall | 10cef60 | 2006-01-08 01:01:45 -0800 | [diff] [blame] | 583 | { |
Paul E. McKenney | 5f0d5a3 | 2017-01-18 02:53:44 -0800 | [diff] [blame] | 584 | if (flags & SLAB_TYPESAFE_BY_RCU) { |
Christoph Lameter | 278b1bb | 2012-09-05 00:20:34 +0000 | [diff] [blame] | 585 | /* leave room for rcu footer at the end of object */ |
| 586 | c->size += sizeof(struct slob_rcu); |
Christoph Lameter | 039363f | 2012-07-06 15:25:10 -0500 | [diff] [blame] | 587 | } |
Christoph Lameter | 278b1bb | 2012-09-05 00:20:34 +0000 | [diff] [blame] | 588 | c->flags = flags; |
Christoph Lameter | 278b1bb | 2012-09-05 00:20:34 +0000 | [diff] [blame] | 589 | return 0; |
Matt Mackall | 10cef60 | 2006-01-08 01:01:45 -0800 | [diff] [blame] | 590 | } |
Matt Mackall | 10cef60 | 2006-01-08 01:01:45 -0800 | [diff] [blame] | 591 | |
Fabian Frederick | c21a6da | 2015-04-14 15:44:34 -0700 | [diff] [blame] | 592 | static void *slob_alloc_node(struct kmem_cache *c, gfp_t flags, int node) |
Matt Mackall | 10cef60 | 2006-01-08 01:01:45 -0800 | [diff] [blame] | 593 | { |
| 594 | void *b; |
| 595 | |
Steven Rostedt | bd50cfa | 2011-06-07 07:18:45 -0400 | [diff] [blame] | 596 | flags &= gfp_allowed_mask; |
| 597 | |
Peter Zijlstra | d92a8cf | 2017-03-03 10:13:38 +0100 | [diff] [blame] | 598 | fs_reclaim_acquire(flags); |
| 599 | fs_reclaim_release(flags); |
Steven Rostedt | bd50cfa | 2011-06-07 07:18:45 -0400 | [diff] [blame] | 600 | |
Eduard - Gabriel Munteanu | 3eae2cb2 | 2008-08-10 20:14:07 +0300 | [diff] [blame] | 601 | if (c->size < PAGE_SIZE) { |
Vlastimil Babka | 59bb479 | 2019-10-06 17:58:45 -0700 | [diff] [blame] | 602 | b = slob_alloc(c->size, flags, c->align, node, 0); |
Ezequiel Garcia | fe74fe2 | 2012-10-19 09:33:11 -0300 | [diff] [blame] | 603 | trace_kmem_cache_alloc_node(_RET_IP_, b, c->object_size, |
Eduard - Gabriel Munteanu | ca2b84cb | 2009-03-23 15:12:24 +0200 | [diff] [blame] | 604 | SLOB_UNITS(c->size) * SLOB_UNIT, |
| 605 | flags, node); |
Eduard - Gabriel Munteanu | 3eae2cb2 | 2008-08-10 20:14:07 +0300 | [diff] [blame] | 606 | } else { |
Américo Wang | 6e9ed0c | 2009-01-19 02:00:38 +0800 | [diff] [blame] | 607 | b = slob_new_pages(flags, get_order(c->size), node); |
Ezequiel Garcia | fe74fe2 | 2012-10-19 09:33:11 -0300 | [diff] [blame] | 608 | trace_kmem_cache_alloc_node(_RET_IP_, b, c->object_size, |
Eduard - Gabriel Munteanu | ca2b84cb | 2009-03-23 15:12:24 +0200 | [diff] [blame] | 609 | PAGE_SIZE << get_order(c->size), |
| 610 | flags, node); |
Eduard - Gabriel Munteanu | 3eae2cb2 | 2008-08-10 20:14:07 +0300 | [diff] [blame] | 611 | } |
Matt Mackall | 10cef60 | 2006-01-08 01:01:45 -0800 | [diff] [blame] | 612 | |
Matthew Wilcox | 128227e | 2018-06-07 17:05:13 -0700 | [diff] [blame] | 613 | if (b && c->ctor) { |
| 614 | WARN_ON_ONCE(flags & __GFP_ZERO); |
Alexey Dobriyan | 51cc506 | 2008-07-25 19:45:34 -0700 | [diff] [blame] | 615 | c->ctor(b); |
Matthew Wilcox | 128227e | 2018-06-07 17:05:13 -0700 | [diff] [blame] | 616 | } |
Matt Mackall | 10cef60 | 2006-01-08 01:01:45 -0800 | [diff] [blame] | 617 | |
Catalin Marinas | 4374e61 | 2009-06-11 13:23:17 +0100 | [diff] [blame] | 618 | kmemleak_alloc_recursive(b, c->size, 1, c->flags, flags); |
Matt Mackall | 10cef60 | 2006-01-08 01:01:45 -0800 | [diff] [blame] | 619 | return b; |
| 620 | } |
Christoph Lameter | f1b6eb6 | 2013-09-04 16:35:34 +0000 | [diff] [blame] | 621 | |
| 622 | void *kmem_cache_alloc(struct kmem_cache *cachep, gfp_t flags) |
| 623 | { |
| 624 | return slob_alloc_node(cachep, flags, NUMA_NO_NODE); |
| 625 | } |
| 626 | EXPORT_SYMBOL(kmem_cache_alloc); |
| 627 | |
| 628 | #ifdef CONFIG_NUMA |
| 629 | void *__kmalloc_node(size_t size, gfp_t gfp, int node) |
| 630 | { |
| 631 | return __do_kmalloc_node(size, gfp, node, _RET_IP_); |
| 632 | } |
| 633 | EXPORT_SYMBOL(__kmalloc_node); |
| 634 | |
| 635 | void *kmem_cache_alloc_node(struct kmem_cache *cachep, gfp_t gfp, int node) |
| 636 | { |
| 637 | return slob_alloc_node(cachep, gfp, node); |
| 638 | } |
Paul Mundt | 6193a2f | 2007-07-15 23:38:22 -0700 | [diff] [blame] | 639 | EXPORT_SYMBOL(kmem_cache_alloc_node); |
Christoph Lameter | f1b6eb6 | 2013-09-04 16:35:34 +0000 | [diff] [blame] | 640 | #endif |
Matt Mackall | 10cef60 | 2006-01-08 01:01:45 -0800 | [diff] [blame] | 641 | |
Nick Piggin | afc0ced | 2007-05-16 22:10:49 -0700 | [diff] [blame] | 642 | static void __kmem_cache_free(void *b, int size) |
| 643 | { |
| 644 | if (size < PAGE_SIZE) |
| 645 | slob_free(b, size); |
| 646 | else |
Américo Wang | 6e9ed0c | 2009-01-19 02:00:38 +0800 | [diff] [blame] | 647 | slob_free_pages(b, get_order(size)); |
Nick Piggin | afc0ced | 2007-05-16 22:10:49 -0700 | [diff] [blame] | 648 | } |
| 649 | |
| 650 | static void kmem_rcu_free(struct rcu_head *head) |
| 651 | { |
| 652 | struct slob_rcu *slob_rcu = (struct slob_rcu *)head; |
| 653 | void *b = (void *)slob_rcu - (slob_rcu->size - sizeof(struct slob_rcu)); |
| 654 | |
| 655 | __kmem_cache_free(b, slob_rcu->size); |
| 656 | } |
| 657 | |
Matt Mackall | 10cef60 | 2006-01-08 01:01:45 -0800 | [diff] [blame] | 658 | void kmem_cache_free(struct kmem_cache *c, void *b) |
| 659 | { |
Catalin Marinas | 4374e61 | 2009-06-11 13:23:17 +0100 | [diff] [blame] | 660 | kmemleak_free_recursive(b, c->flags); |
Paul E. McKenney | 5f0d5a3 | 2017-01-18 02:53:44 -0800 | [diff] [blame] | 661 | if (unlikely(c->flags & SLAB_TYPESAFE_BY_RCU)) { |
Nick Piggin | afc0ced | 2007-05-16 22:10:49 -0700 | [diff] [blame] | 662 | struct slob_rcu *slob_rcu; |
| 663 | slob_rcu = b + (c->size - sizeof(struct slob_rcu)); |
Nick Piggin | afc0ced | 2007-05-16 22:10:49 -0700 | [diff] [blame] | 664 | slob_rcu->size = c->size; |
| 665 | call_rcu(&slob_rcu->head, kmem_rcu_free); |
| 666 | } else { |
Nick Piggin | afc0ced | 2007-05-16 22:10:49 -0700 | [diff] [blame] | 667 | __kmem_cache_free(b, c->size); |
| 668 | } |
Eduard - Gabriel Munteanu | 3eae2cb2 | 2008-08-10 20:14:07 +0300 | [diff] [blame] | 669 | |
Eduard - Gabriel Munteanu | ca2b84cb | 2009-03-23 15:12:24 +0200 | [diff] [blame] | 670 | trace_kmem_cache_free(_RET_IP_, b); |
Matt Mackall | 10cef60 | 2006-01-08 01:01:45 -0800 | [diff] [blame] | 671 | } |
| 672 | EXPORT_SYMBOL(kmem_cache_free); |
| 673 | |
Christoph Lameter | 484748f | 2015-09-04 15:45:34 -0700 | [diff] [blame] | 674 | void kmem_cache_free_bulk(struct kmem_cache *s, size_t size, void **p) |
| 675 | { |
| 676 | __kmem_cache_free_bulk(s, size, p); |
| 677 | } |
| 678 | EXPORT_SYMBOL(kmem_cache_free_bulk); |
| 679 | |
Jesper Dangaard Brouer | 865762a | 2015-11-20 15:57:58 -0800 | [diff] [blame] | 680 | int kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t size, |
Christoph Lameter | 484748f | 2015-09-04 15:45:34 -0700 | [diff] [blame] | 681 | void **p) |
| 682 | { |
| 683 | return __kmem_cache_alloc_bulk(s, flags, size, p); |
| 684 | } |
| 685 | EXPORT_SYMBOL(kmem_cache_alloc_bulk); |
| 686 | |
Christoph Lameter | 945cf2b | 2012-09-04 23:18:33 +0000 | [diff] [blame] | 687 | int __kmem_cache_shutdown(struct kmem_cache *c) |
| 688 | { |
| 689 | /* No way to check for remaining objects */ |
| 690 | return 0; |
| 691 | } |
| 692 | |
Dmitry Safonov | 52b4b95 | 2016-02-17 13:11:37 -0800 | [diff] [blame] | 693 | void __kmem_cache_release(struct kmem_cache *c) |
| 694 | { |
| 695 | } |
| 696 | |
Vladimir Davydov | 89e364d | 2016-12-12 16:41:32 -0800 | [diff] [blame] | 697 | int __kmem_cache_shrink(struct kmem_cache *d) |
Christoph Lameter | 2e892f4 | 2006-12-13 00:34:23 -0800 | [diff] [blame] | 698 | { |
| 699 | return 0; |
| 700 | } |
Christoph Lameter | 2e892f4 | 2006-12-13 00:34:23 -0800 | [diff] [blame] | 701 | |
Christoph Lameter | 9b030cb | 2012-09-05 00:20:33 +0000 | [diff] [blame] | 702 | struct kmem_cache kmem_cache_boot = { |
| 703 | .name = "kmem_cache", |
| 704 | .size = sizeof(struct kmem_cache), |
| 705 | .flags = SLAB_PANIC, |
| 706 | .align = ARCH_KMALLOC_MINALIGN, |
| 707 | }; |
| 708 | |
Dimitri Gorokhovik | bcb4ddb | 2006-12-29 16:48:28 -0800 | [diff] [blame] | 709 | void __init kmem_cache_init(void) |
| 710 | { |
Christoph Lameter | 9b030cb | 2012-09-05 00:20:33 +0000 | [diff] [blame] | 711 | kmem_cache = &kmem_cache_boot; |
Christoph Lameter | 97d0660 | 2012-07-06 15:25:11 -0500 | [diff] [blame] | 712 | slab_state = UP; |
Matt Mackall | 10cef60 | 2006-01-08 01:01:45 -0800 | [diff] [blame] | 713 | } |
Wu Fengguang | bbff2e4 | 2009-08-06 11:36:25 +0300 | [diff] [blame] | 714 | |
| 715 | void __init kmem_cache_init_late(void) |
| 716 | { |
Christoph Lameter | 97d0660 | 2012-07-06 15:25:11 -0500 | [diff] [blame] | 717 | slab_state = FULL; |
Wu Fengguang | bbff2e4 | 2009-08-06 11:36:25 +0300 | [diff] [blame] | 718 | } |