blob: 4d49bb2ca3d4fca7dd2278dc294fd7b747dd94ec [file] [log] [blame]
Thomas Gleixner457c8992019-05-19 13:08:55 +01001// SPDX-License-Identifier: GPL-2.0-only
Linus Torvalds1da177e2005-04-16 15:20:36 -07002/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003 * Copyright (C) 1993 Linus Torvalds
4 * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
5 * SMP-safe vmalloc/vfree/ioremap, Tigran Aivazian <tigran@veritas.com>, May 2000
6 * Major rework to support vmap/vunmap, Christoph Hellwig, SGI, August 2002
Christoph Lameter930fc452005-10-29 18:15:41 -07007 * Numa awareness, Christoph Lameter, SGI, June 2005
Uladzislau Rezki (Sony)d758ffe2020-08-06 23:24:18 -07008 * Improving global KVA allocator, Uladzislau Rezki, Sony, May 2019
Linus Torvalds1da177e2005-04-16 15:20:36 -07009 */
10
Nick Piggindb64fe02008-10-18 20:27:03 -070011#include <linux/vmalloc.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070012#include <linux/mm.h>
13#include <linux/module.h>
14#include <linux/highmem.h>
Ingo Molnarc3edc402017-02-02 08:35:14 +010015#include <linux/sched/signal.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070016#include <linux/slab.h>
17#include <linux/spinlock.h>
18#include <linux/interrupt.h>
Alexey Dobriyan5f6a6a92008-10-06 03:50:47 +040019#include <linux/proc_fs.h>
Christoph Lametera10aa572008-04-28 02:12:40 -070020#include <linux/seq_file.h>
Rick Edgecombe868b1042019-04-25 17:11:36 -070021#include <linux/set_memory.h>
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -070022#include <linux/debugobjects.h>
Christoph Lameter23016962008-04-28 02:12:42 -070023#include <linux/kallsyms.h>
Nick Piggindb64fe02008-10-18 20:27:03 -070024#include <linux/list.h>
Chris Wilson4da56b92016-04-04 14:46:42 +010025#include <linux/notifier.h>
Nick Piggindb64fe02008-10-18 20:27:03 -070026#include <linux/rbtree.h>
Matthew Wilcox (Oracle)0f145992020-08-06 23:24:05 -070027#include <linux/xarray.h>
Nick Piggindb64fe02008-10-18 20:27:03 -070028#include <linux/rcupdate.h>
Tejun Heof0aa6612009-02-20 16:29:08 +090029#include <linux/pfn.h>
Catalin Marinas89219d32009-06-11 13:23:19 +010030#include <linux/kmemleak.h>
Arun Sharma600634972011-07-26 16:09:06 -070031#include <linux/atomic.h>
Gideon Israel Dsouza3b321232014-04-07 15:37:26 -070032#include <linux/compiler.h>
Al Viro32fcfd42013-03-10 20:14:08 -040033#include <linux/llist.h>
Toshi Kani0f616be2015-04-14 15:47:17 -070034#include <linux/bitops.h>
Uladzislau Rezki (Sony)68ad4a32019-05-17 14:31:31 -070035#include <linux/rbtree_augmented.h>
Jann Hornbdebd6a22020-04-20 18:14:11 -070036#include <linux/overflow.h>
Gideon Israel Dsouza3b321232014-04-07 15:37:26 -070037
Linus Torvalds7c0f6ba2016-12-24 11:46:01 -080038#include <linux/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070039#include <asm/tlbflush.h>
David Miller2dca6992009-09-21 12:22:34 -070040#include <asm/shmparam.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070041
Mel Gormandd56b042015-11-06 16:28:43 -080042#include "internal.h"
Joerg Roedel2a681cf2020-08-06 23:22:55 -070043#include "pgalloc-track.h"
Mel Gormandd56b042015-11-06 16:28:43 -080044
Ingo Molnar186525b2019-11-29 08:17:25 +010045bool is_vmalloc_addr(const void *x)
46{
47 unsigned long addr = (unsigned long)x;
48
49 return addr >= VMALLOC_START && addr < VMALLOC_END;
50}
51EXPORT_SYMBOL(is_vmalloc_addr);
52
Al Viro32fcfd42013-03-10 20:14:08 -040053struct vfree_deferred {
54 struct llist_head list;
55 struct work_struct wq;
56};
57static DEFINE_PER_CPU(struct vfree_deferred, vfree_deferred);
58
59static void __vunmap(const void *, int);
60
61static void free_work(struct work_struct *w)
62{
63 struct vfree_deferred *p = container_of(w, struct vfree_deferred, wq);
Byungchul Park894e58c2017-09-06 16:24:26 -070064 struct llist_node *t, *llnode;
65
66 llist_for_each_safe(llnode, t, llist_del_all(&p->list))
67 __vunmap((void *)llnode, 1);
Al Viro32fcfd42013-03-10 20:14:08 -040068}
69
Nick Piggindb64fe02008-10-18 20:27:03 -070070/*** Page table manipulation functions ***/
Adrian Bunkb2213852006-09-25 23:31:02 -070071
Joerg Roedel2ba3e692020-06-01 21:52:22 -070072static void vunmap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
73 pgtbl_mod_mask *mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -070074{
75 pte_t *pte;
76
77 pte = pte_offset_kernel(pmd, addr);
78 do {
79 pte_t ptent = ptep_get_and_clear(&init_mm, addr, pte);
80 WARN_ON(!pte_none(ptent) && !pte_present(ptent));
81 } while (pte++, addr += PAGE_SIZE, addr != end);
Joerg Roedel2ba3e692020-06-01 21:52:22 -070082 *mask |= PGTBL_PTE_MODIFIED;
Linus Torvalds1da177e2005-04-16 15:20:36 -070083}
84
Joerg Roedel2ba3e692020-06-01 21:52:22 -070085static void vunmap_pmd_range(pud_t *pud, unsigned long addr, unsigned long end,
86 pgtbl_mod_mask *mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -070087{
88 pmd_t *pmd;
89 unsigned long next;
Joerg Roedel2ba3e692020-06-01 21:52:22 -070090 int cleared;
Linus Torvalds1da177e2005-04-16 15:20:36 -070091
92 pmd = pmd_offset(pud, addr);
93 do {
94 next = pmd_addr_end(addr, end);
Joerg Roedel2ba3e692020-06-01 21:52:22 -070095
96 cleared = pmd_clear_huge(pmd);
97 if (cleared || pmd_bad(*pmd))
98 *mask |= PGTBL_PMD_MODIFIED;
99
100 if (cleared)
Toshi Kanib9820d82015-04-14 15:47:26 -0700101 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700102 if (pmd_none_or_clear_bad(pmd))
103 continue;
Joerg Roedel2ba3e692020-06-01 21:52:22 -0700104 vunmap_pte_range(pmd, addr, next, mask);
Aneesh Kumar K.Ve47110e2020-08-20 17:42:05 -0700105
106 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107 } while (pmd++, addr = next, addr != end);
108}
109
Joerg Roedel2ba3e692020-06-01 21:52:22 -0700110static void vunmap_pud_range(p4d_t *p4d, unsigned long addr, unsigned long end,
111 pgtbl_mod_mask *mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112{
113 pud_t *pud;
114 unsigned long next;
Joerg Roedel2ba3e692020-06-01 21:52:22 -0700115 int cleared;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +0300117 pud = pud_offset(p4d, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118 do {
119 next = pud_addr_end(addr, end);
Joerg Roedel2ba3e692020-06-01 21:52:22 -0700120
121 cleared = pud_clear_huge(pud);
122 if (cleared || pud_bad(*pud))
123 *mask |= PGTBL_PUD_MODIFIED;
124
125 if (cleared)
Toshi Kanib9820d82015-04-14 15:47:26 -0700126 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700127 if (pud_none_or_clear_bad(pud))
128 continue;
Joerg Roedel2ba3e692020-06-01 21:52:22 -0700129 vunmap_pmd_range(pud, addr, next, mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700130 } while (pud++, addr = next, addr != end);
131}
132
Joerg Roedel2ba3e692020-06-01 21:52:22 -0700133static void vunmap_p4d_range(pgd_t *pgd, unsigned long addr, unsigned long end,
134 pgtbl_mod_mask *mask)
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +0300135{
136 p4d_t *p4d;
137 unsigned long next;
Joerg Roedel2ba3e692020-06-01 21:52:22 -0700138 int cleared;
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +0300139
140 p4d = p4d_offset(pgd, addr);
141 do {
142 next = p4d_addr_end(addr, end);
Joerg Roedel2ba3e692020-06-01 21:52:22 -0700143
144 cleared = p4d_clear_huge(p4d);
145 if (cleared || p4d_bad(*p4d))
146 *mask |= PGTBL_P4D_MODIFIED;
147
148 if (cleared)
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +0300149 continue;
150 if (p4d_none_or_clear_bad(p4d))
151 continue;
Joerg Roedel2ba3e692020-06-01 21:52:22 -0700152 vunmap_pud_range(p4d, addr, next, mask);
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +0300153 } while (p4d++, addr = next, addr != end);
154}
155
Christoph Hellwigb521c432020-06-01 21:51:07 -0700156/**
157 * unmap_kernel_range_noflush - unmap kernel VM area
Joerg Roedel2ba3e692020-06-01 21:52:22 -0700158 * @start: start of the VM area to unmap
Christoph Hellwigb521c432020-06-01 21:51:07 -0700159 * @size: size of the VM area to unmap
160 *
161 * Unmap PFN_UP(@size) pages at @addr. The VM area @addr and @size specify
162 * should have been allocated using get_vm_area() and its friends.
163 *
164 * NOTE:
165 * This function does NOT do any cache flushing. The caller is responsible
166 * for calling flush_cache_vunmap() on to-be-mapped areas before calling this
167 * function and flush_tlb_kernel_range() after.
168 */
Joerg Roedel2ba3e692020-06-01 21:52:22 -0700169void unmap_kernel_range_noflush(unsigned long start, unsigned long size)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170{
Joerg Roedel2ba3e692020-06-01 21:52:22 -0700171 unsigned long end = start + size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172 unsigned long next;
Christoph Hellwigb521c432020-06-01 21:51:07 -0700173 pgd_t *pgd;
Joerg Roedel2ba3e692020-06-01 21:52:22 -0700174 unsigned long addr = start;
175 pgtbl_mod_mask mask = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176
177 BUG_ON(addr >= end);
178 pgd = pgd_offset_k(addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179 do {
180 next = pgd_addr_end(addr, end);
Joerg Roedel2ba3e692020-06-01 21:52:22 -0700181 if (pgd_bad(*pgd))
182 mask |= PGTBL_PGD_MODIFIED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183 if (pgd_none_or_clear_bad(pgd))
184 continue;
Joerg Roedel2ba3e692020-06-01 21:52:22 -0700185 vunmap_p4d_range(pgd, addr, next, &mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186 } while (pgd++, addr = next, addr != end);
Joerg Roedel2ba3e692020-06-01 21:52:22 -0700187
188 if (mask & ARCH_PAGE_TABLE_SYNC_MASK)
189 arch_sync_kernel_mappings(start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190}
191
192static int vmap_pte_range(pmd_t *pmd, unsigned long addr,
Joerg Roedel2ba3e692020-06-01 21:52:22 -0700193 unsigned long end, pgprot_t prot, struct page **pages, int *nr,
194 pgtbl_mod_mask *mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195{
196 pte_t *pte;
197
Nick Piggindb64fe02008-10-18 20:27:03 -0700198 /*
199 * nr is a running index into the array which helps higher level
200 * callers keep track of where we're up to.
201 */
202
Joerg Roedel2ba3e692020-06-01 21:52:22 -0700203 pte = pte_alloc_kernel_track(pmd, addr, mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204 if (!pte)
205 return -ENOMEM;
206 do {
Nick Piggindb64fe02008-10-18 20:27:03 -0700207 struct page *page = pages[*nr];
208
209 if (WARN_ON(!pte_none(*pte)))
210 return -EBUSY;
211 if (WARN_ON(!page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212 return -ENOMEM;
213 set_pte_at(&init_mm, addr, pte, mk_pte(page, prot));
Nick Piggindb64fe02008-10-18 20:27:03 -0700214 (*nr)++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215 } while (pte++, addr += PAGE_SIZE, addr != end);
Joerg Roedel2ba3e692020-06-01 21:52:22 -0700216 *mask |= PGTBL_PTE_MODIFIED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217 return 0;
218}
219
Nick Piggindb64fe02008-10-18 20:27:03 -0700220static int vmap_pmd_range(pud_t *pud, unsigned long addr,
Joerg Roedel2ba3e692020-06-01 21:52:22 -0700221 unsigned long end, pgprot_t prot, struct page **pages, int *nr,
222 pgtbl_mod_mask *mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223{
224 pmd_t *pmd;
225 unsigned long next;
226
Joerg Roedel2ba3e692020-06-01 21:52:22 -0700227 pmd = pmd_alloc_track(&init_mm, pud, addr, mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228 if (!pmd)
229 return -ENOMEM;
230 do {
231 next = pmd_addr_end(addr, end);
Joerg Roedel2ba3e692020-06-01 21:52:22 -0700232 if (vmap_pte_range(pmd, addr, next, prot, pages, nr, mask))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233 return -ENOMEM;
234 } while (pmd++, addr = next, addr != end);
235 return 0;
236}
237
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +0300238static int vmap_pud_range(p4d_t *p4d, unsigned long addr,
Joerg Roedel2ba3e692020-06-01 21:52:22 -0700239 unsigned long end, pgprot_t prot, struct page **pages, int *nr,
240 pgtbl_mod_mask *mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241{
242 pud_t *pud;
243 unsigned long next;
244
Joerg Roedel2ba3e692020-06-01 21:52:22 -0700245 pud = pud_alloc_track(&init_mm, p4d, addr, mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246 if (!pud)
247 return -ENOMEM;
248 do {
249 next = pud_addr_end(addr, end);
Joerg Roedel2ba3e692020-06-01 21:52:22 -0700250 if (vmap_pmd_range(pud, addr, next, prot, pages, nr, mask))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251 return -ENOMEM;
252 } while (pud++, addr = next, addr != end);
253 return 0;
254}
255
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +0300256static int vmap_p4d_range(pgd_t *pgd, unsigned long addr,
Joerg Roedel2ba3e692020-06-01 21:52:22 -0700257 unsigned long end, pgprot_t prot, struct page **pages, int *nr,
258 pgtbl_mod_mask *mask)
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +0300259{
260 p4d_t *p4d;
261 unsigned long next;
262
Joerg Roedel2ba3e692020-06-01 21:52:22 -0700263 p4d = p4d_alloc_track(&init_mm, pgd, addr, mask);
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +0300264 if (!p4d)
265 return -ENOMEM;
266 do {
267 next = p4d_addr_end(addr, end);
Joerg Roedel2ba3e692020-06-01 21:52:22 -0700268 if (vmap_pud_range(p4d, addr, next, prot, pages, nr, mask))
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +0300269 return -ENOMEM;
270 } while (p4d++, addr = next, addr != end);
271 return 0;
272}
273
Christoph Hellwigb521c432020-06-01 21:51:07 -0700274/**
275 * map_kernel_range_noflush - map kernel VM area with the specified pages
276 * @addr: start of the VM area to map
277 * @size: size of the VM area to map
278 * @prot: page protection flags to use
279 * @pages: pages to map
Nick Piggindb64fe02008-10-18 20:27:03 -0700280 *
Christoph Hellwigb521c432020-06-01 21:51:07 -0700281 * Map PFN_UP(@size) pages at @addr. The VM area @addr and @size specify should
282 * have been allocated using get_vm_area() and its friends.
283 *
284 * NOTE:
285 * This function does NOT do any cache flushing. The caller is responsible for
286 * calling flush_cache_vmap() on to-be-mapped areas before calling this
287 * function.
288 *
289 * RETURNS:
Christoph Hellwig60bb44652020-06-01 21:51:15 -0700290 * 0 on success, -errno on failure.
Nick Piggindb64fe02008-10-18 20:27:03 -0700291 */
Christoph Hellwigb521c432020-06-01 21:51:07 -0700292int map_kernel_range_noflush(unsigned long addr, unsigned long size,
293 pgprot_t prot, struct page **pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294{
Joerg Roedel2ba3e692020-06-01 21:52:22 -0700295 unsigned long start = addr;
Christoph Hellwigb521c432020-06-01 21:51:07 -0700296 unsigned long end = addr + size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297 unsigned long next;
Christoph Hellwigb521c432020-06-01 21:51:07 -0700298 pgd_t *pgd;
Nick Piggindb64fe02008-10-18 20:27:03 -0700299 int err = 0;
300 int nr = 0;
Joerg Roedel2ba3e692020-06-01 21:52:22 -0700301 pgtbl_mod_mask mask = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302
303 BUG_ON(addr >= end);
304 pgd = pgd_offset_k(addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700305 do {
306 next = pgd_addr_end(addr, end);
Joerg Roedel2ba3e692020-06-01 21:52:22 -0700307 if (pgd_bad(*pgd))
308 mask |= PGTBL_PGD_MODIFIED;
309 err = vmap_p4d_range(pgd, addr, next, prot, pages, &nr, &mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310 if (err)
Figo.zhangbf88c8c2009-09-21 17:01:47 -0700311 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312 } while (pgd++, addr = next, addr != end);
Nick Piggindb64fe02008-10-18 20:27:03 -0700313
Joerg Roedel2ba3e692020-06-01 21:52:22 -0700314 if (mask & ARCH_PAGE_TABLE_SYNC_MASK)
315 arch_sync_kernel_mappings(start, end);
316
Christoph Hellwig60bb44652020-06-01 21:51:15 -0700317 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318}
319
Christoph Hellwiged1f3242020-06-01 21:51:19 -0700320int map_kernel_range(unsigned long start, unsigned long size, pgprot_t prot,
321 struct page **pages)
Tejun Heo8fc48982009-02-20 16:29:08 +0900322{
323 int ret;
324
Christoph Hellwiga29adb62020-06-01 21:51:11 -0700325 ret = map_kernel_range_noflush(start, size, prot, pages);
326 flush_cache_vmap(start, start + size);
Tejun Heo8fc48982009-02-20 16:29:08 +0900327 return ret;
328}
329
KAMEZAWA Hiroyuki81ac3ad2009-09-22 16:45:49 -0700330int is_vmalloc_or_module_addr(const void *x)
Linus Torvalds73bdf0a2008-10-15 08:35:12 -0700331{
332 /*
Russell Kingab4f2ee2008-11-06 17:11:07 +0000333 * ARM, x86-64 and sparc64 put modules in a special place,
Linus Torvalds73bdf0a2008-10-15 08:35:12 -0700334 * and fall back on vmalloc() if that fails. Others
335 * just put it in the vmalloc space.
336 */
337#if defined(CONFIG_MODULES) && defined(MODULES_VADDR)
338 unsigned long addr = (unsigned long)x;
339 if (addr >= MODULES_VADDR && addr < MODULES_END)
340 return 1;
341#endif
342 return is_vmalloc_addr(x);
343}
344
Christoph Lameter48667e72008-02-04 22:28:31 -0800345/*
malcadd688f2014-01-27 17:06:53 -0800346 * Walk a vmap address to the struct page it maps.
Christoph Lameter48667e72008-02-04 22:28:31 -0800347 */
malcadd688f2014-01-27 17:06:53 -0800348struct page *vmalloc_to_page(const void *vmalloc_addr)
Christoph Lameter48667e72008-02-04 22:28:31 -0800349{
350 unsigned long addr = (unsigned long) vmalloc_addr;
malcadd688f2014-01-27 17:06:53 -0800351 struct page *page = NULL;
Christoph Lameter48667e72008-02-04 22:28:31 -0800352 pgd_t *pgd = pgd_offset_k(addr);
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +0300353 p4d_t *p4d;
354 pud_t *pud;
355 pmd_t *pmd;
356 pte_t *ptep, pte;
Christoph Lameter48667e72008-02-04 22:28:31 -0800357
Ingo Molnar7aa413d2008-06-19 13:28:11 +0200358 /*
359 * XXX we might need to change this if we add VIRTUAL_BUG_ON for
360 * architectures that do not vmalloc module space
361 */
Linus Torvalds73bdf0a2008-10-15 08:35:12 -0700362 VIRTUAL_BUG_ON(!is_vmalloc_or_module_addr(vmalloc_addr));
Jiri Slaby59ea7462008-06-12 13:56:40 +0200363
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +0300364 if (pgd_none(*pgd))
365 return NULL;
366 p4d = p4d_offset(pgd, addr);
367 if (p4d_none(*p4d))
368 return NULL;
369 pud = pud_offset(p4d, addr);
Ard Biesheuvel029c54b2017-06-23 15:08:41 -0700370
371 /*
372 * Don't dereference bad PUD or PMD (below) entries. This will also
373 * identify huge mappings, which we may encounter on architectures
374 * that define CONFIG_HAVE_ARCH_HUGE_VMAP=y. Such regions will be
375 * identified as vmalloc addresses by is_vmalloc_addr(), but are
376 * not [unambiguously] associated with a struct page, so there is
377 * no correct value to return for them.
378 */
379 WARN_ON_ONCE(pud_bad(*pud));
380 if (pud_none(*pud) || pud_bad(*pud))
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +0300381 return NULL;
382 pmd = pmd_offset(pud, addr);
Ard Biesheuvel029c54b2017-06-23 15:08:41 -0700383 WARN_ON_ONCE(pmd_bad(*pmd));
384 if (pmd_none(*pmd) || pmd_bad(*pmd))
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +0300385 return NULL;
Nick Piggindb64fe02008-10-18 20:27:03 -0700386
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +0300387 ptep = pte_offset_map(pmd, addr);
388 pte = *ptep;
389 if (pte_present(pte))
390 page = pte_page(pte);
391 pte_unmap(ptep);
malcadd688f2014-01-27 17:06:53 -0800392 return page;
Jianyu Zhanece86e222014-01-21 15:49:12 -0800393}
394EXPORT_SYMBOL(vmalloc_to_page);
395
malcadd688f2014-01-27 17:06:53 -0800396/*
397 * Map a vmalloc()-space virtual address to the physical page frame number.
398 */
399unsigned long vmalloc_to_pfn(const void *vmalloc_addr)
400{
401 return page_to_pfn(vmalloc_to_page(vmalloc_addr));
402}
403EXPORT_SYMBOL(vmalloc_to_pfn);
404
Nick Piggindb64fe02008-10-18 20:27:03 -0700405
406/*** Global kva allocator ***/
407
Uladzislau Rezki (Sony)bb850f42019-05-17 14:31:34 -0700408#define DEBUG_AUGMENT_PROPAGATE_CHECK 0
Uladzislau Rezki (Sony)a6cf4e02019-05-17 14:31:37 -0700409#define DEBUG_AUGMENT_LOWEST_MATCH_CHECK 0
Uladzislau Rezki (Sony)bb850f42019-05-17 14:31:34 -0700410
Nick Piggindb64fe02008-10-18 20:27:03 -0700411
Nick Piggindb64fe02008-10-18 20:27:03 -0700412static DEFINE_SPINLOCK(vmap_area_lock);
Uladzislau Rezki (Sony)e36176b2019-11-30 17:54:47 -0800413static DEFINE_SPINLOCK(free_vmap_area_lock);
Joonsoo Kimf1c40692013-04-29 15:07:37 -0700414/* Export for kexec only */
415LIST_HEAD(vmap_area_list);
Chris Wilson80c4bd72016-05-20 16:57:38 -0700416static LLIST_HEAD(vmap_purge_list);
Nick Piggin89699602011-03-22 16:30:36 -0700417static struct rb_root vmap_area_root = RB_ROOT;
Uladzislau Rezki (Sony)68ad4a32019-05-17 14:31:31 -0700418static bool vmap_initialized __read_mostly;
Nick Piggin89699602011-03-22 16:30:36 -0700419
Uladzislau Rezki (Sony)68ad4a32019-05-17 14:31:31 -0700420/*
421 * This kmem_cache is used for vmap_area objects. Instead of
422 * allocating from slab we reuse an object from this cache to
423 * make things faster. Especially in "no edge" splitting of
424 * free block.
425 */
426static struct kmem_cache *vmap_area_cachep;
Nick Piggin89699602011-03-22 16:30:36 -0700427
Uladzislau Rezki (Sony)68ad4a32019-05-17 14:31:31 -0700428/*
429 * This linked list is used in pair with free_vmap_area_root.
430 * It gives O(1) access to prev/next to perform fast coalescing.
431 */
432static LIST_HEAD(free_vmap_area_list);
433
434/*
435 * This augment red-black tree represents the free vmap space.
436 * All vmap_area objects in this tree are sorted by va->va_start
437 * address. It is used for allocation and merging when a vmap
438 * object is released.
439 *
440 * Each vmap_area node contains a maximum available free block
441 * of its sub-tree, right or left. Therefore it is possible to
442 * find a lowest match of free area.
443 */
444static struct rb_root free_vmap_area_root = RB_ROOT;
445
Uladzislau Rezki (Sony)82dd23e2019-07-11 20:58:57 -0700446/*
447 * Preload a CPU with one object for "no edge" split case. The
448 * aim is to get rid of allocations from the atomic context, thus
449 * to use more permissive allocation masks.
450 */
451static DEFINE_PER_CPU(struct vmap_area *, ne_fit_preload_node);
452
Uladzislau Rezki (Sony)68ad4a32019-05-17 14:31:31 -0700453static __always_inline unsigned long
454va_size(struct vmap_area *va)
455{
456 return (va->va_end - va->va_start);
457}
458
459static __always_inline unsigned long
460get_subtree_max_size(struct rb_node *node)
461{
462 struct vmap_area *va;
463
464 va = rb_entry_safe(node, struct vmap_area, rb_node);
465 return va ? va->subtree_max_size : 0;
466}
467
468/*
469 * Gets called when remove the node and rotate.
470 */
471static __always_inline unsigned long
472compute_subtree_max_size(struct vmap_area *va)
473{
474 return max3(va_size(va),
475 get_subtree_max_size(va->rb_node.rb_left),
476 get_subtree_max_size(va->rb_node.rb_right));
477}
478
Michel Lespinasse315cc062019-09-25 16:46:07 -0700479RB_DECLARE_CALLBACKS_MAX(static, free_vmap_area_rb_augment_cb,
480 struct vmap_area, rb_node, unsigned long, subtree_max_size, va_size)
Uladzislau Rezki (Sony)68ad4a32019-05-17 14:31:31 -0700481
482static void purge_vmap_area_lazy(void);
483static BLOCKING_NOTIFIER_HEAD(vmap_notify_list);
484static unsigned long lazy_max_pages(void);
Nick Piggindb64fe02008-10-18 20:27:03 -0700485
Roman Gushchin97105f02019-07-11 21:00:13 -0700486static atomic_long_t nr_vmalloc_pages;
487
488unsigned long vmalloc_nr_pages(void)
489{
490 return atomic_long_read(&nr_vmalloc_pages);
491}
Vijayanand Jittae46a1a52020-12-29 15:55:19 +0530492EXPORT_SYMBOL_GPL(vmalloc_nr_pages);
Roman Gushchin97105f02019-07-11 21:00:13 -0700493
Nick Piggindb64fe02008-10-18 20:27:03 -0700494static struct vmap_area *__find_vmap_area(unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495{
Nick Piggindb64fe02008-10-18 20:27:03 -0700496 struct rb_node *n = vmap_area_root.rb_node;
497
498 while (n) {
499 struct vmap_area *va;
500
501 va = rb_entry(n, struct vmap_area, rb_node);
502 if (addr < va->va_start)
503 n = n->rb_left;
HATAYAMA Daisukecef2ac32013-07-03 15:02:17 -0700504 else if (addr >= va->va_end)
Nick Piggindb64fe02008-10-18 20:27:03 -0700505 n = n->rb_right;
506 else
507 return va;
508 }
509
510 return NULL;
511}
512
Uladzislau Rezki (Sony)68ad4a32019-05-17 14:31:31 -0700513/*
514 * This function returns back addresses of parent node
515 * and its left or right link for further processing.
Uladzislau Rezki (Sony)9c801f62020-08-06 23:24:24 -0700516 *
517 * Otherwise NULL is returned. In that case all further
518 * steps regarding inserting of conflicting overlap range
519 * have to be declined and actually considered as a bug.
Uladzislau Rezki (Sony)68ad4a32019-05-17 14:31:31 -0700520 */
521static __always_inline struct rb_node **
522find_va_links(struct vmap_area *va,
523 struct rb_root *root, struct rb_node *from,
524 struct rb_node **parent)
Nick Piggindb64fe02008-10-18 20:27:03 -0700525{
Uladzislau Rezki (Sony)68ad4a32019-05-17 14:31:31 -0700526 struct vmap_area *tmp_va;
527 struct rb_node **link;
Nick Piggindb64fe02008-10-18 20:27:03 -0700528
Uladzislau Rezki (Sony)68ad4a32019-05-17 14:31:31 -0700529 if (root) {
530 link = &root->rb_node;
531 if (unlikely(!*link)) {
532 *parent = NULL;
533 return link;
534 }
535 } else {
536 link = &from;
Nick Piggindb64fe02008-10-18 20:27:03 -0700537 }
538
Uladzislau Rezki (Sony)68ad4a32019-05-17 14:31:31 -0700539 /*
540 * Go to the bottom of the tree. When we hit the last point
541 * we end up with parent rb_node and correct direction, i name
542 * it link, where the new va->rb_node will be attached to.
543 */
544 do {
545 tmp_va = rb_entry(*link, struct vmap_area, rb_node);
Nick Piggindb64fe02008-10-18 20:27:03 -0700546
Uladzislau Rezki (Sony)68ad4a32019-05-17 14:31:31 -0700547 /*
548 * During the traversal we also do some sanity check.
549 * Trigger the BUG() if there are sides(left/right)
550 * or full overlaps.
551 */
552 if (va->va_start < tmp_va->va_end &&
553 va->va_end <= tmp_va->va_start)
554 link = &(*link)->rb_left;
555 else if (va->va_end > tmp_va->va_start &&
556 va->va_start >= tmp_va->va_end)
557 link = &(*link)->rb_right;
Uladzislau Rezki (Sony)9c801f62020-08-06 23:24:24 -0700558 else {
559 WARN(1, "vmalloc bug: 0x%lx-0x%lx overlaps with 0x%lx-0x%lx\n",
560 va->va_start, va->va_end, tmp_va->va_start, tmp_va->va_end);
561
562 return NULL;
563 }
Uladzislau Rezki (Sony)68ad4a32019-05-17 14:31:31 -0700564 } while (*link);
565
566 *parent = &tmp_va->rb_node;
567 return link;
Nick Piggindb64fe02008-10-18 20:27:03 -0700568}
569
Uladzislau Rezki (Sony)68ad4a32019-05-17 14:31:31 -0700570static __always_inline struct list_head *
571get_va_next_sibling(struct rb_node *parent, struct rb_node **link)
572{
573 struct list_head *list;
Nick Piggindb64fe02008-10-18 20:27:03 -0700574
Uladzislau Rezki (Sony)68ad4a32019-05-17 14:31:31 -0700575 if (unlikely(!parent))
576 /*
577 * The red-black tree where we try to find VA neighbors
578 * before merging or inserting is empty, i.e. it means
579 * there is no free vmap space. Normally it does not
580 * happen but we handle this case anyway.
581 */
582 return NULL;
583
584 list = &rb_entry(parent, struct vmap_area, rb_node)->list;
585 return (&parent->rb_right == link ? list->next : list);
586}
587
588static __always_inline void
589link_va(struct vmap_area *va, struct rb_root *root,
590 struct rb_node *parent, struct rb_node **link, struct list_head *head)
591{
592 /*
593 * VA is still not in the list, but we can
594 * identify its future previous list_head node.
595 */
596 if (likely(parent)) {
597 head = &rb_entry(parent, struct vmap_area, rb_node)->list;
598 if (&parent->rb_right != link)
599 head = head->prev;
600 }
601
602 /* Insert to the rb-tree */
603 rb_link_node(&va->rb_node, parent, link);
604 if (root == &free_vmap_area_root) {
605 /*
606 * Some explanation here. Just perform simple insertion
607 * to the tree. We do not set va->subtree_max_size to
608 * its current size before calling rb_insert_augmented().
609 * It is because of we populate the tree from the bottom
610 * to parent levels when the node _is_ in the tree.
611 *
612 * Therefore we set subtree_max_size to zero after insertion,
613 * to let __augment_tree_propagate_from() puts everything to
614 * the correct order later on.
615 */
616 rb_insert_augmented(&va->rb_node,
617 root, &free_vmap_area_rb_augment_cb);
618 va->subtree_max_size = 0;
619 } else {
620 rb_insert_color(&va->rb_node, root);
621 }
622
623 /* Address-sort this list */
624 list_add(&va->list, head);
625}
626
627static __always_inline void
628unlink_va(struct vmap_area *va, struct rb_root *root)
629{
Uladzislau Rezki (Sony)460e42d2019-07-11 20:59:03 -0700630 if (WARN_ON(RB_EMPTY_NODE(&va->rb_node)))
631 return;
Uladzislau Rezki (Sony)68ad4a32019-05-17 14:31:31 -0700632
Uladzislau Rezki (Sony)460e42d2019-07-11 20:59:03 -0700633 if (root == &free_vmap_area_root)
634 rb_erase_augmented(&va->rb_node,
635 root, &free_vmap_area_rb_augment_cb);
636 else
637 rb_erase(&va->rb_node, root);
638
639 list_del(&va->list);
640 RB_CLEAR_NODE(&va->rb_node);
Uladzislau Rezki (Sony)68ad4a32019-05-17 14:31:31 -0700641}
642
Uladzislau Rezki (Sony)bb850f42019-05-17 14:31:34 -0700643#if DEBUG_AUGMENT_PROPAGATE_CHECK
644static void
Uladzislau Rezki (Sony)da27c9e2020-08-06 23:24:12 -0700645augment_tree_propagate_check(void)
Uladzislau Rezki (Sony)bb850f42019-05-17 14:31:34 -0700646{
647 struct vmap_area *va;
Uladzislau Rezki (Sony)da27c9e2020-08-06 23:24:12 -0700648 unsigned long computed_size;
Uladzislau Rezki (Sony)bb850f42019-05-17 14:31:34 -0700649
Uladzislau Rezki (Sony)da27c9e2020-08-06 23:24:12 -0700650 list_for_each_entry(va, &free_vmap_area_list, list) {
651 computed_size = compute_subtree_max_size(va);
652 if (computed_size != va->subtree_max_size)
653 pr_emerg("tree is corrupted: %lu, %lu\n",
654 va_size(va), va->subtree_max_size);
Uladzislau Rezki (Sony)bb850f42019-05-17 14:31:34 -0700655 }
Uladzislau Rezki (Sony)bb850f42019-05-17 14:31:34 -0700656}
657#endif
658
Uladzislau Rezki (Sony)68ad4a32019-05-17 14:31:31 -0700659/*
660 * This function populates subtree_max_size from bottom to upper
661 * levels starting from VA point. The propagation must be done
662 * when VA size is modified by changing its va_start/va_end. Or
663 * in case of newly inserting of VA to the tree.
664 *
665 * It means that __augment_tree_propagate_from() must be called:
666 * - After VA has been inserted to the tree(free path);
667 * - After VA has been shrunk(allocation path);
668 * - After VA has been increased(merging path).
669 *
670 * Please note that, it does not mean that upper parent nodes
671 * and their subtree_max_size are recalculated all the time up
672 * to the root node.
673 *
674 * 4--8
675 * /\
676 * / \
677 * / \
678 * 2--2 8--8
679 *
680 * For example if we modify the node 4, shrinking it to 2, then
681 * no any modification is required. If we shrink the node 2 to 1
682 * its subtree_max_size is updated only, and set to 1. If we shrink
683 * the node 8 to 6, then its subtree_max_size is set to 6 and parent
684 * node becomes 4--6.
685 */
686static __always_inline void
687augment_tree_propagate_from(struct vmap_area *va)
688{
Uladzislau Rezki (Sony)15ae1442020-08-06 23:24:15 -0700689 /*
690 * Populate the tree from bottom towards the root until
691 * the calculated maximum available size of checked node
692 * is equal to its current one.
693 */
694 free_vmap_area_rb_augment_cb_propagate(&va->rb_node, NULL);
Uladzislau Rezki (Sony)bb850f42019-05-17 14:31:34 -0700695
696#if DEBUG_AUGMENT_PROPAGATE_CHECK
Uladzislau Rezki (Sony)da27c9e2020-08-06 23:24:12 -0700697 augment_tree_propagate_check();
Uladzislau Rezki (Sony)bb850f42019-05-17 14:31:34 -0700698#endif
Uladzislau Rezki (Sony)68ad4a32019-05-17 14:31:31 -0700699}
700
701static void
702insert_vmap_area(struct vmap_area *va,
703 struct rb_root *root, struct list_head *head)
704{
705 struct rb_node **link;
706 struct rb_node *parent;
707
708 link = find_va_links(va, root, NULL, &parent);
Uladzislau Rezki (Sony)9c801f62020-08-06 23:24:24 -0700709 if (link)
710 link_va(va, root, parent, link, head);
Uladzislau Rezki (Sony)68ad4a32019-05-17 14:31:31 -0700711}
712
713static void
714insert_vmap_area_augment(struct vmap_area *va,
715 struct rb_node *from, struct rb_root *root,
716 struct list_head *head)
717{
718 struct rb_node **link;
719 struct rb_node *parent;
720
721 if (from)
722 link = find_va_links(va, NULL, from, &parent);
723 else
724 link = find_va_links(va, root, NULL, &parent);
725
Uladzislau Rezki (Sony)9c801f62020-08-06 23:24:24 -0700726 if (link) {
727 link_va(va, root, parent, link, head);
728 augment_tree_propagate_from(va);
729 }
Uladzislau Rezki (Sony)68ad4a32019-05-17 14:31:31 -0700730}
731
732/*
733 * Merge de-allocated chunk of VA memory with previous
734 * and next free blocks. If coalesce is not done a new
735 * free area is inserted. If VA has been merged, it is
736 * freed.
Uladzislau Rezki (Sony)9c801f62020-08-06 23:24:24 -0700737 *
738 * Please note, it can return NULL in case of overlap
739 * ranges, followed by WARN() report. Despite it is a
740 * buggy behaviour, a system can be alive and keep
741 * ongoing.
Uladzislau Rezki (Sony)68ad4a32019-05-17 14:31:31 -0700742 */
Daniel Axtens3c5c3cf2019-11-30 17:54:50 -0800743static __always_inline struct vmap_area *
Uladzislau Rezki (Sony)68ad4a32019-05-17 14:31:31 -0700744merge_or_add_vmap_area(struct vmap_area *va,
745 struct rb_root *root, struct list_head *head)
746{
747 struct vmap_area *sibling;
748 struct list_head *next;
749 struct rb_node **link;
750 struct rb_node *parent;
751 bool merged = false;
752
753 /*
754 * Find a place in the tree where VA potentially will be
755 * inserted, unless it is merged with its sibling/siblings.
756 */
757 link = find_va_links(va, root, NULL, &parent);
Uladzislau Rezki (Sony)9c801f62020-08-06 23:24:24 -0700758 if (!link)
759 return NULL;
Uladzislau Rezki (Sony)68ad4a32019-05-17 14:31:31 -0700760
761 /*
762 * Get next node of VA to check if merging can be done.
763 */
764 next = get_va_next_sibling(parent, link);
765 if (unlikely(next == NULL))
766 goto insert;
767
768 /*
769 * start end
770 * | |
771 * |<------VA------>|<-----Next----->|
772 * | |
773 * start end
774 */
775 if (next != head) {
776 sibling = list_entry(next, struct vmap_area, list);
777 if (sibling->va_start == va->va_end) {
778 sibling->va_start = va->va_start;
779
Uladzislau Rezki (Sony)68ad4a32019-05-17 14:31:31 -0700780 /* Free vmap_area object. */
781 kmem_cache_free(vmap_area_cachep, va);
782
783 /* Point to the new merged area. */
784 va = sibling;
785 merged = true;
786 }
787 }
788
789 /*
790 * start end
791 * | |
792 * |<-----Prev----->|<------VA------>|
793 * | |
794 * start end
795 */
796 if (next->prev != head) {
797 sibling = list_entry(next->prev, struct vmap_area, list);
798 if (sibling->va_end == va->va_start) {
Uladzislau Rezki (Sony)5dd78642020-08-06 23:24:09 -0700799 /*
800 * If both neighbors are coalesced, it is important
801 * to unlink the "next" node first, followed by merging
802 * with "previous" one. Otherwise the tree might not be
803 * fully populated if a sibling's augmented value is
804 * "normalized" because of rotation operations.
805 */
Uladzislau Rezki (Sony)54f63d92019-07-11 20:59:00 -0700806 if (merged)
807 unlink_va(va, root);
Uladzislau Rezki (Sony)68ad4a32019-05-17 14:31:31 -0700808
Uladzislau Rezki (Sony)5dd78642020-08-06 23:24:09 -0700809 sibling->va_end = va->va_end;
810
Uladzislau Rezki (Sony)68ad4a32019-05-17 14:31:31 -0700811 /* Free vmap_area object. */
812 kmem_cache_free(vmap_area_cachep, va);
Daniel Axtens3c5c3cf2019-11-30 17:54:50 -0800813
814 /* Point to the new merged area. */
815 va = sibling;
816 merged = true;
Uladzislau Rezki (Sony)68ad4a32019-05-17 14:31:31 -0700817 }
818 }
819
820insert:
Uladzislau Rezki (Sony)5dd78642020-08-06 23:24:09 -0700821 if (!merged)
Uladzislau Rezki (Sony)68ad4a32019-05-17 14:31:31 -0700822 link_va(va, root, parent, link, head);
Daniel Axtens3c5c3cf2019-11-30 17:54:50 -0800823
Uladzislau Rezki (Sony)5dd78642020-08-06 23:24:09 -0700824 /*
825 * Last step is to check and update the tree.
826 */
827 augment_tree_propagate_from(va);
Daniel Axtens3c5c3cf2019-11-30 17:54:50 -0800828 return va;
Uladzislau Rezki (Sony)68ad4a32019-05-17 14:31:31 -0700829}
830
831static __always_inline bool
832is_within_this_va(struct vmap_area *va, unsigned long size,
833 unsigned long align, unsigned long vstart)
834{
835 unsigned long nva_start_addr;
836
837 if (va->va_start > vstart)
838 nva_start_addr = ALIGN(va->va_start, align);
839 else
840 nva_start_addr = ALIGN(vstart, align);
841
842 /* Can be overflowed due to big size or alignment. */
843 if (nva_start_addr + size < nva_start_addr ||
844 nva_start_addr < vstart)
845 return false;
846
847 return (nva_start_addr + size <= va->va_end);
848}
849
850/*
851 * Find the first free block(lowest start address) in the tree,
852 * that will accomplish the request corresponding to passing
853 * parameters.
854 */
855static __always_inline struct vmap_area *
856find_vmap_lowest_match(unsigned long size,
857 unsigned long align, unsigned long vstart)
858{
859 struct vmap_area *va;
860 struct rb_node *node;
861 unsigned long length;
862
863 /* Start from the root. */
864 node = free_vmap_area_root.rb_node;
865
866 /* Adjust the search size for alignment overhead. */
867 length = size + align - 1;
868
869 while (node) {
870 va = rb_entry(node, struct vmap_area, rb_node);
871
872 if (get_subtree_max_size(node->rb_left) >= length &&
873 vstart < va->va_start) {
874 node = node->rb_left;
875 } else {
876 if (is_within_this_va(va, size, align, vstart))
877 return va;
878
879 /*
880 * Does not make sense to go deeper towards the right
881 * sub-tree if it does not have a free block that is
882 * equal or bigger to the requested search length.
883 */
884 if (get_subtree_max_size(node->rb_right) >= length) {
885 node = node->rb_right;
886 continue;
887 }
888
889 /*
Andrew Morton3806b042019-05-31 22:30:03 -0700890 * OK. We roll back and find the first right sub-tree,
Uladzislau Rezki (Sony)68ad4a32019-05-17 14:31:31 -0700891 * that will satisfy the search criteria. It can happen
892 * only once due to "vstart" restriction.
893 */
894 while ((node = rb_parent(node))) {
895 va = rb_entry(node, struct vmap_area, rb_node);
896 if (is_within_this_va(va, size, align, vstart))
897 return va;
898
899 if (get_subtree_max_size(node->rb_right) >= length &&
900 vstart <= va->va_start) {
901 node = node->rb_right;
902 break;
903 }
904 }
905 }
906 }
907
908 return NULL;
909}
910
Uladzislau Rezki (Sony)a6cf4e02019-05-17 14:31:37 -0700911#if DEBUG_AUGMENT_LOWEST_MATCH_CHECK
912#include <linux/random.h>
913
914static struct vmap_area *
915find_vmap_lowest_linear_match(unsigned long size,
916 unsigned long align, unsigned long vstart)
917{
918 struct vmap_area *va;
919
920 list_for_each_entry(va, &free_vmap_area_list, list) {
921 if (!is_within_this_va(va, size, align, vstart))
922 continue;
923
924 return va;
925 }
926
927 return NULL;
928}
929
930static void
931find_vmap_lowest_match_check(unsigned long size)
932{
933 struct vmap_area *va_1, *va_2;
934 unsigned long vstart;
935 unsigned int rnd;
936
937 get_random_bytes(&rnd, sizeof(rnd));
938 vstart = VMALLOC_START + rnd;
939
940 va_1 = find_vmap_lowest_match(size, 1, vstart);
941 va_2 = find_vmap_lowest_linear_match(size, 1, vstart);
942
943 if (va_1 != va_2)
944 pr_emerg("not lowest: t: 0x%p, l: 0x%p, v: 0x%lx\n",
945 va_1, va_2, vstart);
946}
947#endif
948
Uladzislau Rezki (Sony)68ad4a32019-05-17 14:31:31 -0700949enum fit_type {
950 NOTHING_FIT = 0,
951 FL_FIT_TYPE = 1, /* full fit */
952 LE_FIT_TYPE = 2, /* left edge fit */
953 RE_FIT_TYPE = 3, /* right edge fit */
954 NE_FIT_TYPE = 4 /* no edge fit */
955};
956
957static __always_inline enum fit_type
958classify_va_fit_type(struct vmap_area *va,
959 unsigned long nva_start_addr, unsigned long size)
960{
961 enum fit_type type;
962
963 /* Check if it is within VA. */
964 if (nva_start_addr < va->va_start ||
965 nva_start_addr + size > va->va_end)
966 return NOTHING_FIT;
967
968 /* Now classify. */
969 if (va->va_start == nva_start_addr) {
970 if (va->va_end == nva_start_addr + size)
971 type = FL_FIT_TYPE;
972 else
973 type = LE_FIT_TYPE;
974 } else if (va->va_end == nva_start_addr + size) {
975 type = RE_FIT_TYPE;
976 } else {
977 type = NE_FIT_TYPE;
978 }
979
980 return type;
981}
982
983static __always_inline int
984adjust_va_to_fit_type(struct vmap_area *va,
985 unsigned long nva_start_addr, unsigned long size,
986 enum fit_type type)
987{
Arnd Bergmann2c929232019-06-28 12:07:09 -0700988 struct vmap_area *lva = NULL;
Uladzislau Rezki (Sony)68ad4a32019-05-17 14:31:31 -0700989
990 if (type == FL_FIT_TYPE) {
991 /*
992 * No need to split VA, it fully fits.
993 *
994 * | |
995 * V NVA V
996 * |---------------|
997 */
998 unlink_va(va, &free_vmap_area_root);
999 kmem_cache_free(vmap_area_cachep, va);
1000 } else if (type == LE_FIT_TYPE) {
1001 /*
1002 * Split left edge of fit VA.
1003 *
1004 * | |
1005 * V NVA V R
1006 * |-------|-------|
1007 */
1008 va->va_start += size;
1009 } else if (type == RE_FIT_TYPE) {
1010 /*
1011 * Split right edge of fit VA.
1012 *
1013 * | |
1014 * L V NVA V
1015 * |-------|-------|
1016 */
1017 va->va_end = nva_start_addr;
1018 } else if (type == NE_FIT_TYPE) {
1019 /*
1020 * Split no edge of fit VA.
1021 *
1022 * | |
1023 * L V NVA V R
1024 * |---|-------|---|
1025 */
Uladzislau Rezki (Sony)82dd23e2019-07-11 20:58:57 -07001026 lva = __this_cpu_xchg(ne_fit_preload_node, NULL);
1027 if (unlikely(!lva)) {
1028 /*
1029 * For percpu allocator we do not do any pre-allocation
1030 * and leave it as it is. The reason is it most likely
1031 * never ends up with NE_FIT_TYPE splitting. In case of
1032 * percpu allocations offsets and sizes are aligned to
1033 * fixed align request, i.e. RE_FIT_TYPE and FL_FIT_TYPE
1034 * are its main fitting cases.
1035 *
1036 * There are a few exceptions though, as an example it is
1037 * a first allocation (early boot up) when we have "one"
1038 * big free space that has to be split.
Uladzislau Rezki (Sony)060650a2019-11-30 17:54:40 -08001039 *
1040 * Also we can hit this path in case of regular "vmap"
1041 * allocations, if "this" current CPU was not preloaded.
1042 * See the comment in alloc_vmap_area() why. If so, then
1043 * GFP_NOWAIT is used instead to get an extra object for
1044 * split purpose. That is rare and most time does not
1045 * occur.
1046 *
1047 * What happens if an allocation gets failed. Basically,
1048 * an "overflow" path is triggered to purge lazily freed
1049 * areas to free some memory, then, the "retry" path is
1050 * triggered to repeat one more time. See more details
1051 * in alloc_vmap_area() function.
Uladzislau Rezki (Sony)82dd23e2019-07-11 20:58:57 -07001052 */
1053 lva = kmem_cache_alloc(vmap_area_cachep, GFP_NOWAIT);
1054 if (!lva)
1055 return -1;
1056 }
Uladzislau Rezki (Sony)68ad4a32019-05-17 14:31:31 -07001057
1058 /*
1059 * Build the remainder.
1060 */
1061 lva->va_start = va->va_start;
1062 lva->va_end = nva_start_addr;
1063
1064 /*
1065 * Shrink this VA to remaining size.
1066 */
1067 va->va_start = nva_start_addr + size;
1068 } else {
1069 return -1;
1070 }
1071
1072 if (type != FL_FIT_TYPE) {
1073 augment_tree_propagate_from(va);
1074
Arnd Bergmann2c929232019-06-28 12:07:09 -07001075 if (lva) /* type == NE_FIT_TYPE */
Uladzislau Rezki (Sony)68ad4a32019-05-17 14:31:31 -07001076 insert_vmap_area_augment(lva, &va->rb_node,
1077 &free_vmap_area_root, &free_vmap_area_list);
1078 }
1079
1080 return 0;
1081}
1082
1083/*
1084 * Returns a start address of the newly allocated area, if success.
1085 * Otherwise a vend is returned that indicates failure.
1086 */
1087static __always_inline unsigned long
1088__alloc_vmap_area(unsigned long size, unsigned long align,
Uladzislau Rezki (Sony)cacca6b2019-07-11 20:58:53 -07001089 unsigned long vstart, unsigned long vend)
Uladzislau Rezki (Sony)68ad4a32019-05-17 14:31:31 -07001090{
1091 unsigned long nva_start_addr;
1092 struct vmap_area *va;
1093 enum fit_type type;
1094 int ret;
1095
1096 va = find_vmap_lowest_match(size, align, vstart);
1097 if (unlikely(!va))
1098 return vend;
1099
1100 if (va->va_start > vstart)
1101 nva_start_addr = ALIGN(va->va_start, align);
1102 else
1103 nva_start_addr = ALIGN(vstart, align);
1104
1105 /* Check the "vend" restriction. */
1106 if (nva_start_addr + size > vend)
1107 return vend;
1108
1109 /* Classify what we have found. */
1110 type = classify_va_fit_type(va, nva_start_addr, size);
1111 if (WARN_ON_ONCE(type == NOTHING_FIT))
1112 return vend;
1113
1114 /* Update the free vmap_area. */
1115 ret = adjust_va_to_fit_type(va, nva_start_addr, size, type);
1116 if (ret)
1117 return vend;
1118
Uladzislau Rezki (Sony)a6cf4e02019-05-17 14:31:37 -07001119#if DEBUG_AUGMENT_LOWEST_MATCH_CHECK
1120 find_vmap_lowest_match_check(size);
1121#endif
1122
Uladzislau Rezki (Sony)68ad4a32019-05-17 14:31:31 -07001123 return nva_start_addr;
1124}
Chris Wilson4da56b92016-04-04 14:46:42 +01001125
Nick Piggindb64fe02008-10-18 20:27:03 -07001126/*
Andrey Ryabinind98c9e82019-12-17 20:51:38 -08001127 * Free a region of KVA allocated by alloc_vmap_area
1128 */
1129static void free_vmap_area(struct vmap_area *va)
1130{
1131 /*
1132 * Remove from the busy tree/list.
1133 */
1134 spin_lock(&vmap_area_lock);
1135 unlink_va(va, &vmap_area_root);
1136 spin_unlock(&vmap_area_lock);
1137
1138 /*
1139 * Insert/Merge it back to the free tree/list.
1140 */
1141 spin_lock(&free_vmap_area_lock);
1142 merge_or_add_vmap_area(va, &free_vmap_area_root, &free_vmap_area_list);
1143 spin_unlock(&free_vmap_area_lock);
1144}
1145
1146/*
Nick Piggindb64fe02008-10-18 20:27:03 -07001147 * Allocate a region of KVA of the specified size and alignment, within the
1148 * vstart and vend.
1149 */
1150static struct vmap_area *alloc_vmap_area(unsigned long size,
1151 unsigned long align,
1152 unsigned long vstart, unsigned long vend,
1153 int node, gfp_t gfp_mask)
1154{
Uladzislau Rezki (Sony)82dd23e2019-07-11 20:58:57 -07001155 struct vmap_area *va, *pva;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156 unsigned long addr;
Nick Piggindb64fe02008-10-18 20:27:03 -07001157 int purged = 0;
Andrey Ryabinind98c9e82019-12-17 20:51:38 -08001158 int ret;
Nick Piggindb64fe02008-10-18 20:27:03 -07001159
Nick Piggin77669702009-02-27 14:03:03 -08001160 BUG_ON(!size);
Alexander Kuleshov891c49a2015-11-05 18:46:51 -08001161 BUG_ON(offset_in_page(size));
Nick Piggin89699602011-03-22 16:30:36 -07001162 BUG_ON(!is_power_of_2(align));
Nick Piggindb64fe02008-10-18 20:27:03 -07001163
Uladzislau Rezki (Sony)68ad4a32019-05-17 14:31:31 -07001164 if (unlikely(!vmap_initialized))
1165 return ERR_PTR(-EBUSY);
1166
Christoph Hellwig5803ed22016-12-12 16:44:20 -08001167 might_sleep();
Uladzislau Rezki (Sony)f07116d2019-11-30 17:54:37 -08001168 gfp_mask = gfp_mask & GFP_RECLAIM_MASK;
Chris Wilson4da56b92016-04-04 14:46:42 +01001169
Uladzislau Rezki (Sony)f07116d2019-11-30 17:54:37 -08001170 va = kmem_cache_alloc_node(vmap_area_cachep, gfp_mask, node);
Nick Piggindb64fe02008-10-18 20:27:03 -07001171 if (unlikely(!va))
1172 return ERR_PTR(-ENOMEM);
1173
Catalin Marinas7f88f882013-11-12 15:07:45 -08001174 /*
1175 * Only scan the relevant parts containing pointers to other objects
1176 * to avoid false negatives.
1177 */
Uladzislau Rezki (Sony)f07116d2019-11-30 17:54:37 -08001178 kmemleak_scan_area(&va->rb_node, SIZE_MAX, gfp_mask);
Catalin Marinas7f88f882013-11-12 15:07:45 -08001179
Nick Piggindb64fe02008-10-18 20:27:03 -07001180retry:
Uladzislau Rezki (Sony)82dd23e2019-07-11 20:58:57 -07001181 /*
Uladzislau Rezki (Sony)81f1ba52019-11-30 17:54:33 -08001182 * Preload this CPU with one extra vmap_area object. It is used
1183 * when fit type of free area is NE_FIT_TYPE. Please note, it
1184 * does not guarantee that an allocation occurs on a CPU that
1185 * is preloaded, instead we minimize the case when it is not.
1186 * It can happen because of cpu migration, because there is a
1187 * race until the below spinlock is taken.
Uladzislau Rezki (Sony)82dd23e2019-07-11 20:58:57 -07001188 *
1189 * The preload is done in non-atomic context, thus it allows us
1190 * to use more permissive allocation masks to be more stable under
Uladzislau Rezki (Sony)81f1ba52019-11-30 17:54:33 -08001191 * low memory condition and high memory pressure. In rare case,
1192 * if not preloaded, GFP_NOWAIT is used.
Uladzislau Rezki (Sony)82dd23e2019-07-11 20:58:57 -07001193 *
Uladzislau Rezki (Sony)81f1ba52019-11-30 17:54:33 -08001194 * Set "pva" to NULL here, because of "retry" path.
Uladzislau Rezki (Sony)82dd23e2019-07-11 20:58:57 -07001195 */
Uladzislau Rezki (Sony)81f1ba52019-11-30 17:54:33 -08001196 pva = NULL;
Uladzislau Rezki (Sony)82dd23e2019-07-11 20:58:57 -07001197
Uladzislau Rezki (Sony)81f1ba52019-11-30 17:54:33 -08001198 if (!this_cpu_read(ne_fit_preload_node))
1199 /*
1200 * Even if it fails we do not really care about that.
1201 * Just proceed as it is. If needed "overflow" path
1202 * will refill the cache we allocate from.
1203 */
Uladzislau Rezki (Sony)f07116d2019-11-30 17:54:37 -08001204 pva = kmem_cache_alloc_node(vmap_area_cachep, gfp_mask, node);
Uladzislau Rezki (Sony)82dd23e2019-07-11 20:58:57 -07001205
Uladzislau Rezki (Sony)e36176b2019-11-30 17:54:47 -08001206 spin_lock(&free_vmap_area_lock);
Uladzislau Rezki (Sony)81f1ba52019-11-30 17:54:33 -08001207
1208 if (pva && __this_cpu_cmpxchg(ne_fit_preload_node, NULL, pva))
1209 kmem_cache_free(vmap_area_cachep, pva);
Uladzislau Rezki (Sony)68ad4a32019-05-17 14:31:31 -07001210
Nick Piggin89699602011-03-22 16:30:36 -07001211 /*
Uladzislau Rezki (Sony)68ad4a32019-05-17 14:31:31 -07001212 * If an allocation fails, the "vend" address is
1213 * returned. Therefore trigger the overflow path.
Nick Piggin89699602011-03-22 16:30:36 -07001214 */
Uladzislau Rezki (Sony)cacca6b2019-07-11 20:58:53 -07001215 addr = __alloc_vmap_area(size, align, vstart, vend);
Uladzislau Rezki (Sony)e36176b2019-11-30 17:54:47 -08001216 spin_unlock(&free_vmap_area_lock);
1217
Uladzislau Rezki (Sony)68ad4a32019-05-17 14:31:31 -07001218 if (unlikely(addr == vend))
Nick Piggin89699602011-03-22 16:30:36 -07001219 goto overflow;
Nick Piggindb64fe02008-10-18 20:27:03 -07001220
1221 va->va_start = addr;
1222 va->va_end = addr + size;
Pengfei Li688fcbf2019-09-23 15:36:39 -07001223 va->vm = NULL;
Uladzislau Rezki (Sony)68ad4a32019-05-17 14:31:31 -07001224
Andrey Ryabinind98c9e82019-12-17 20:51:38 -08001225
Uladzislau Rezki (Sony)e36176b2019-11-30 17:54:47 -08001226 spin_lock(&vmap_area_lock);
1227 insert_vmap_area(va, &vmap_area_root, &vmap_area_list);
Nick Piggindb64fe02008-10-18 20:27:03 -07001228 spin_unlock(&vmap_area_lock);
1229
Wang Xiaoqiang61e16552016-01-15 16:57:19 -08001230 BUG_ON(!IS_ALIGNED(va->va_start, align));
Nick Piggin89699602011-03-22 16:30:36 -07001231 BUG_ON(va->va_start < vstart);
1232 BUG_ON(va->va_end > vend);
1233
Andrey Ryabinind98c9e82019-12-17 20:51:38 -08001234 ret = kasan_populate_vmalloc(addr, size);
1235 if (ret) {
1236 free_vmap_area(va);
1237 return ERR_PTR(ret);
1238 }
1239
Nick Piggindb64fe02008-10-18 20:27:03 -07001240 return va;
Nick Piggin89699602011-03-22 16:30:36 -07001241
1242overflow:
Nick Piggin89699602011-03-22 16:30:36 -07001243 if (!purged) {
1244 purge_vmap_area_lazy();
1245 purged = 1;
1246 goto retry;
1247 }
Chris Wilson4da56b92016-04-04 14:46:42 +01001248
1249 if (gfpflags_allow_blocking(gfp_mask)) {
1250 unsigned long freed = 0;
1251 blocking_notifier_call_chain(&vmap_notify_list, 0, &freed);
1252 if (freed > 0) {
1253 purged = 0;
1254 goto retry;
1255 }
1256 }
1257
Florian Fainelli03497d72017-04-27 11:19:00 -07001258 if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit())
Joe Perches756a0252016-03-17 14:19:47 -07001259 pr_warn("vmap allocation for size %lu failed: use vmalloc=<size> to increase size\n",
1260 size);
Uladzislau Rezki (Sony)68ad4a32019-05-17 14:31:31 -07001261
1262 kmem_cache_free(vmap_area_cachep, va);
Nick Piggin89699602011-03-22 16:30:36 -07001263 return ERR_PTR(-EBUSY);
Nick Piggindb64fe02008-10-18 20:27:03 -07001264}
1265
Chris Wilson4da56b92016-04-04 14:46:42 +01001266int register_vmap_purge_notifier(struct notifier_block *nb)
1267{
1268 return blocking_notifier_chain_register(&vmap_notify_list, nb);
1269}
1270EXPORT_SYMBOL_GPL(register_vmap_purge_notifier);
1271
1272int unregister_vmap_purge_notifier(struct notifier_block *nb)
1273{
1274 return blocking_notifier_chain_unregister(&vmap_notify_list, nb);
1275}
1276EXPORT_SYMBOL_GPL(unregister_vmap_purge_notifier);
1277
Nick Piggindb64fe02008-10-18 20:27:03 -07001278/*
Nick Piggindb64fe02008-10-18 20:27:03 -07001279 * lazy_max_pages is the maximum amount of virtual address space we gather up
1280 * before attempting to purge with a TLB flush.
1281 *
1282 * There is a tradeoff here: a larger number will cover more kernel page tables
1283 * and take slightly longer to purge, but it will linearly reduce the number of
1284 * global TLB flushes that must be performed. It would seem natural to scale
1285 * this number up linearly with the number of CPUs (because vmapping activity
1286 * could also scale linearly with the number of CPUs), however it is likely
1287 * that in practice, workloads might be constrained in other ways that mean
1288 * vmap activity will not scale linearly with CPUs. Also, I want to be
1289 * conservative and not introduce a big latency on huge systems, so go with
1290 * a less aggressive log scale. It will still be an improvement over the old
1291 * code, and it will be simple to change the scale factor if we find that it
1292 * becomes a problem on bigger systems.
1293 */
1294static unsigned long lazy_max_pages(void)
1295{
1296 unsigned int log;
1297
1298 log = fls(num_online_cpus());
1299
1300 return log * (32UL * 1024 * 1024 / PAGE_SIZE);
1301}
1302
Uladzislau Rezki (Sony)4d36e6f2019-05-14 15:41:25 -07001303static atomic_long_t vmap_lazy_nr = ATOMIC_LONG_INIT(0);
Nick Piggindb64fe02008-10-18 20:27:03 -07001304
Christoph Hellwig0574ecd2016-12-12 16:44:07 -08001305/*
1306 * Serialize vmap purging. There is no actual criticial section protected
1307 * by this look, but we want to avoid concurrent calls for performance
1308 * reasons and to make the pcpu_get_vm_areas more deterministic.
1309 */
Christoph Hellwigf9e09972016-12-12 16:44:23 -08001310static DEFINE_MUTEX(vmap_purge_lock);
Christoph Hellwig0574ecd2016-12-12 16:44:07 -08001311
Nick Piggin02b709d2010-02-01 22:25:57 +11001312/* for per-CPU blocks */
1313static void purge_fragmented_blocks_allcpus(void);
1314
Nick Piggindb64fe02008-10-18 20:27:03 -07001315/*
Cliff Wickman3ee48b62010-09-16 11:44:02 -05001316 * called before a call to iounmap() if the caller wants vm_area_struct's
1317 * immediately freed.
1318 */
1319void set_iounmap_nonlazy(void)
1320{
Uladzislau Rezki (Sony)4d36e6f2019-05-14 15:41:25 -07001321 atomic_long_set(&vmap_lazy_nr, lazy_max_pages()+1);
Cliff Wickman3ee48b62010-09-16 11:44:02 -05001322}
1323
1324/*
Nick Piggindb64fe02008-10-18 20:27:03 -07001325 * Purges all lazily-freed vmap areas.
Nick Piggindb64fe02008-10-18 20:27:03 -07001326 */
Christoph Hellwig0574ecd2016-12-12 16:44:07 -08001327static bool __purge_vmap_area_lazy(unsigned long start, unsigned long end)
Nick Piggindb64fe02008-10-18 20:27:03 -07001328{
Uladzislau Rezki (Sony)4d36e6f2019-05-14 15:41:25 -07001329 unsigned long resched_threshold;
Chris Wilson80c4bd72016-05-20 16:57:38 -07001330 struct llist_node *valist;
Nick Piggindb64fe02008-10-18 20:27:03 -07001331 struct vmap_area *va;
Vegard Nossumcbb76672009-02-27 14:03:04 -08001332 struct vmap_area *n_va;
Nick Piggindb64fe02008-10-18 20:27:03 -07001333
Christoph Hellwig0574ecd2016-12-12 16:44:07 -08001334 lockdep_assert_held(&vmap_purge_lock);
Nick Piggin02b709d2010-02-01 22:25:57 +11001335
Chris Wilson80c4bd72016-05-20 16:57:38 -07001336 valist = llist_del_all(&vmap_purge_list);
Uladzislau Rezki (Sony)68571be92019-05-14 15:41:22 -07001337 if (unlikely(valist == NULL))
1338 return false;
1339
1340 /*
1341 * TODO: to calculate a flush range without looping.
1342 * The list can be up to lazy_max_pages() elements.
1343 */
Chris Wilson80c4bd72016-05-20 16:57:38 -07001344 llist_for_each_entry(va, valist, purge_list) {
Christoph Hellwig0574ecd2016-12-12 16:44:07 -08001345 if (va->va_start < start)
1346 start = va->va_start;
1347 if (va->va_end > end)
1348 end = va->va_end;
Nick Piggindb64fe02008-10-18 20:27:03 -07001349 }
Nick Piggindb64fe02008-10-18 20:27:03 -07001350
Christoph Hellwig0574ecd2016-12-12 16:44:07 -08001351 flush_tlb_kernel_range(start, end);
Uladzislau Rezki (Sony)4d36e6f2019-05-14 15:41:25 -07001352 resched_threshold = lazy_max_pages() << 1;
Nick Piggindb64fe02008-10-18 20:27:03 -07001353
Uladzislau Rezki (Sony)e36176b2019-11-30 17:54:47 -08001354 spin_lock(&free_vmap_area_lock);
Joel Fernandes763b2182016-12-12 16:44:26 -08001355 llist_for_each_entry_safe(va, n_va, valist, purge_list) {
Uladzislau Rezki (Sony)4d36e6f2019-05-14 15:41:25 -07001356 unsigned long nr = (va->va_end - va->va_start) >> PAGE_SHIFT;
Daniel Axtens3c5c3cf2019-11-30 17:54:50 -08001357 unsigned long orig_start = va->va_start;
1358 unsigned long orig_end = va->va_end;
Joel Fernandes763b2182016-12-12 16:44:26 -08001359
Uladzislau Rezki (Sony)dd3b8352019-09-23 15:36:36 -07001360 /*
1361 * Finally insert or merge lazily-freed area. It is
1362 * detached and there is no need to "unlink" it from
1363 * anything.
1364 */
Daniel Axtens3c5c3cf2019-11-30 17:54:50 -08001365 va = merge_or_add_vmap_area(va, &free_vmap_area_root,
1366 &free_vmap_area_list);
1367
Uladzislau Rezki (Sony)9c801f62020-08-06 23:24:24 -07001368 if (!va)
1369 continue;
1370
Daniel Axtens3c5c3cf2019-11-30 17:54:50 -08001371 if (is_vmalloc_or_module_addr((void *)orig_start))
1372 kasan_release_vmalloc(orig_start, orig_end,
1373 va->va_start, va->va_end);
Uladzislau Rezki (Sony)dd3b8352019-09-23 15:36:36 -07001374
Uladzislau Rezki (Sony)4d36e6f2019-05-14 15:41:25 -07001375 atomic_long_sub(nr, &vmap_lazy_nr);
Uladzislau Rezki (Sony)68571be92019-05-14 15:41:22 -07001376
Uladzislau Rezki (Sony)4d36e6f2019-05-14 15:41:25 -07001377 if (atomic_long_read(&vmap_lazy_nr) < resched_threshold)
Uladzislau Rezki (Sony)e36176b2019-11-30 17:54:47 -08001378 cond_resched_lock(&free_vmap_area_lock);
Joel Fernandes763b2182016-12-12 16:44:26 -08001379 }
Uladzislau Rezki (Sony)e36176b2019-11-30 17:54:47 -08001380 spin_unlock(&free_vmap_area_lock);
Christoph Hellwig0574ecd2016-12-12 16:44:07 -08001381 return true;
Nick Piggindb64fe02008-10-18 20:27:03 -07001382}
1383
1384/*
Nick Piggin496850e2008-11-19 15:36:33 -08001385 * Kick off a purge of the outstanding lazy areas. Don't bother if somebody
1386 * is already purging.
1387 */
1388static void try_purge_vmap_area_lazy(void)
1389{
Christoph Hellwigf9e09972016-12-12 16:44:23 -08001390 if (mutex_trylock(&vmap_purge_lock)) {
Christoph Hellwig0574ecd2016-12-12 16:44:07 -08001391 __purge_vmap_area_lazy(ULONG_MAX, 0);
Christoph Hellwigf9e09972016-12-12 16:44:23 -08001392 mutex_unlock(&vmap_purge_lock);
Christoph Hellwig0574ecd2016-12-12 16:44:07 -08001393 }
Nick Piggin496850e2008-11-19 15:36:33 -08001394}
1395
1396/*
Nick Piggindb64fe02008-10-18 20:27:03 -07001397 * Kick off a purge of the outstanding lazy areas.
1398 */
1399static void purge_vmap_area_lazy(void)
1400{
Christoph Hellwigf9e09972016-12-12 16:44:23 -08001401 mutex_lock(&vmap_purge_lock);
Christoph Hellwig0574ecd2016-12-12 16:44:07 -08001402 purge_fragmented_blocks_allcpus();
1403 __purge_vmap_area_lazy(ULONG_MAX, 0);
Christoph Hellwigf9e09972016-12-12 16:44:23 -08001404 mutex_unlock(&vmap_purge_lock);
Nick Piggindb64fe02008-10-18 20:27:03 -07001405}
1406
1407/*
Jeremy Fitzhardinge64141da2010-12-02 14:31:18 -08001408 * Free a vmap area, caller ensuring that the area has been unmapped
1409 * and flush_cache_vunmap had been called for the correct range
1410 * previously.
Nick Piggindb64fe02008-10-18 20:27:03 -07001411 */
Jeremy Fitzhardinge64141da2010-12-02 14:31:18 -08001412static void free_vmap_area_noflush(struct vmap_area *va)
Nick Piggindb64fe02008-10-18 20:27:03 -07001413{
Uladzislau Rezki (Sony)4d36e6f2019-05-14 15:41:25 -07001414 unsigned long nr_lazy;
Chris Wilson80c4bd72016-05-20 16:57:38 -07001415
Uladzislau Rezki (Sony)dd3b8352019-09-23 15:36:36 -07001416 spin_lock(&vmap_area_lock);
1417 unlink_va(va, &vmap_area_root);
1418 spin_unlock(&vmap_area_lock);
1419
Uladzislau Rezki (Sony)4d36e6f2019-05-14 15:41:25 -07001420 nr_lazy = atomic_long_add_return((va->va_end - va->va_start) >>
1421 PAGE_SHIFT, &vmap_lazy_nr);
Chris Wilson80c4bd72016-05-20 16:57:38 -07001422
1423 /* After this point, we may free va at any time */
1424 llist_add(&va->purge_list, &vmap_purge_list);
1425
1426 if (unlikely(nr_lazy > lazy_max_pages()))
Nick Piggin496850e2008-11-19 15:36:33 -08001427 try_purge_vmap_area_lazy();
Nick Piggindb64fe02008-10-18 20:27:03 -07001428}
1429
Nick Pigginb29acbd2008-12-01 13:13:47 -08001430/*
1431 * Free and unmap a vmap area
1432 */
1433static void free_unmap_vmap_area(struct vmap_area *va)
1434{
1435 flush_cache_vunmap(va->va_start, va->va_end);
Christoph Hellwig855e57a2020-06-01 21:51:23 -07001436 unmap_kernel_range_noflush(va->va_start, va->va_end - va->va_start);
Vlastimil Babka8e57f8a2020-01-13 16:29:20 -08001437 if (debug_pagealloc_enabled_static())
Chintan Pandya82a2e922018-06-07 17:06:46 -07001438 flush_tlb_kernel_range(va->va_start, va->va_end);
1439
Christoph Hellwigc8eef012016-12-12 16:44:01 -08001440 free_vmap_area_noflush(va);
Nick Pigginb29acbd2008-12-01 13:13:47 -08001441}
1442
Nick Piggindb64fe02008-10-18 20:27:03 -07001443static struct vmap_area *find_vmap_area(unsigned long addr)
1444{
1445 struct vmap_area *va;
1446
1447 spin_lock(&vmap_area_lock);
1448 va = __find_vmap_area(addr);
1449 spin_unlock(&vmap_area_lock);
1450
1451 return va;
1452}
1453
Nick Piggindb64fe02008-10-18 20:27:03 -07001454/*** Per cpu kva allocator ***/
1455
1456/*
1457 * vmap space is limited especially on 32 bit architectures. Ensure there is
1458 * room for at least 16 percpu vmap blocks per CPU.
1459 */
1460/*
1461 * If we had a constant VMALLOC_START and VMALLOC_END, we'd like to be able
1462 * to #define VMALLOC_SPACE (VMALLOC_END-VMALLOC_START). Guess
1463 * instead (we just need a rough idea)
1464 */
1465#if BITS_PER_LONG == 32
1466#define VMALLOC_SPACE (128UL*1024*1024)
1467#else
1468#define VMALLOC_SPACE (128UL*1024*1024*1024)
1469#endif
1470
1471#define VMALLOC_PAGES (VMALLOC_SPACE / PAGE_SIZE)
1472#define VMAP_MAX_ALLOC BITS_PER_LONG /* 256K with 4K pages */
1473#define VMAP_BBMAP_BITS_MAX 1024 /* 4MB with 4K pages */
1474#define VMAP_BBMAP_BITS_MIN (VMAP_MAX_ALLOC*2)
1475#define VMAP_MIN(x, y) ((x) < (y) ? (x) : (y)) /* can't use min() */
1476#define VMAP_MAX(x, y) ((x) > (y) ? (x) : (y)) /* can't use max() */
Clemens Ladischf982f9152011-06-21 22:09:50 +02001477#define VMAP_BBMAP_BITS \
1478 VMAP_MIN(VMAP_BBMAP_BITS_MAX, \
1479 VMAP_MAX(VMAP_BBMAP_BITS_MIN, \
1480 VMALLOC_PAGES / roundup_pow_of_two(NR_CPUS) / 16))
Nick Piggindb64fe02008-10-18 20:27:03 -07001481
1482#define VMAP_BLOCK_SIZE (VMAP_BBMAP_BITS * PAGE_SIZE)
1483
1484struct vmap_block_queue {
1485 spinlock_t lock;
1486 struct list_head free;
Nick Piggindb64fe02008-10-18 20:27:03 -07001487};
1488
1489struct vmap_block {
1490 spinlock_t lock;
1491 struct vmap_area *va;
Nick Piggindb64fe02008-10-18 20:27:03 -07001492 unsigned long free, dirty;
Roman Pen7d61bfe2015-04-15 16:13:55 -07001493 unsigned long dirty_min, dirty_max; /*< dirty range */
Nick Pigginde560422010-02-01 22:24:18 +11001494 struct list_head free_list;
1495 struct rcu_head rcu_head;
Nick Piggin02b709d2010-02-01 22:25:57 +11001496 struct list_head purge;
Nick Piggindb64fe02008-10-18 20:27:03 -07001497};
1498
1499/* Queue of free and dirty vmap blocks, for allocation and flushing purposes */
1500static DEFINE_PER_CPU(struct vmap_block_queue, vmap_block_queue);
1501
1502/*
Matthew Wilcox (Oracle)0f145992020-08-06 23:24:05 -07001503 * XArray of vmap blocks, indexed by address, to quickly find a vmap block
Nick Piggindb64fe02008-10-18 20:27:03 -07001504 * in the free path. Could get rid of this if we change the API to return a
1505 * "cookie" from alloc, to be passed to free. But no big deal yet.
1506 */
Matthew Wilcox (Oracle)0f145992020-08-06 23:24:05 -07001507static DEFINE_XARRAY(vmap_blocks);
Nick Piggindb64fe02008-10-18 20:27:03 -07001508
1509/*
1510 * We should probably have a fallback mechanism to allocate virtual memory
1511 * out of partially filled vmap blocks. However vmap block sizing should be
1512 * fairly reasonable according to the vmalloc size, so it shouldn't be a
1513 * big problem.
1514 */
1515
1516static unsigned long addr_to_vb_idx(unsigned long addr)
1517{
1518 addr -= VMALLOC_START & ~(VMAP_BLOCK_SIZE-1);
1519 addr /= VMAP_BLOCK_SIZE;
1520 return addr;
1521}
1522
Roman Pencf725ce2015-04-15 16:13:52 -07001523static void *vmap_block_vaddr(unsigned long va_start, unsigned long pages_off)
1524{
1525 unsigned long addr;
1526
1527 addr = va_start + (pages_off << PAGE_SHIFT);
1528 BUG_ON(addr_to_vb_idx(addr) != addr_to_vb_idx(va_start));
1529 return (void *)addr;
1530}
1531
1532/**
1533 * new_vmap_block - allocates new vmap_block and occupies 2^order pages in this
1534 * block. Of course pages number can't exceed VMAP_BBMAP_BITS
1535 * @order: how many 2^order pages should be occupied in newly allocated block
1536 * @gfp_mask: flags for the page level allocator
1537 *
Mike Rapoporta862f682019-03-05 15:48:42 -08001538 * Return: virtual address in a newly allocated block or ERR_PTR(-errno)
Roman Pencf725ce2015-04-15 16:13:52 -07001539 */
1540static void *new_vmap_block(unsigned int order, gfp_t gfp_mask)
Nick Piggindb64fe02008-10-18 20:27:03 -07001541{
1542 struct vmap_block_queue *vbq;
1543 struct vmap_block *vb;
1544 struct vmap_area *va;
1545 unsigned long vb_idx;
1546 int node, err;
Roman Pencf725ce2015-04-15 16:13:52 -07001547 void *vaddr;
Nick Piggindb64fe02008-10-18 20:27:03 -07001548
1549 node = numa_node_id();
1550
1551 vb = kmalloc_node(sizeof(struct vmap_block),
1552 gfp_mask & GFP_RECLAIM_MASK, node);
1553 if (unlikely(!vb))
1554 return ERR_PTR(-ENOMEM);
1555
1556 va = alloc_vmap_area(VMAP_BLOCK_SIZE, VMAP_BLOCK_SIZE,
1557 VMALLOC_START, VMALLOC_END,
1558 node, gfp_mask);
Tobias Klauserddf9c6d42011-01-13 15:46:15 -08001559 if (IS_ERR(va)) {
Nick Piggindb64fe02008-10-18 20:27:03 -07001560 kfree(vb);
Julia Lawalle7d86342010-08-09 17:18:28 -07001561 return ERR_CAST(va);
Nick Piggindb64fe02008-10-18 20:27:03 -07001562 }
1563
Roman Pencf725ce2015-04-15 16:13:52 -07001564 vaddr = vmap_block_vaddr(va->va_start, 0);
Nick Piggindb64fe02008-10-18 20:27:03 -07001565 spin_lock_init(&vb->lock);
1566 vb->va = va;
Roman Pencf725ce2015-04-15 16:13:52 -07001567 /* At least something should be left free */
1568 BUG_ON(VMAP_BBMAP_BITS <= (1UL << order));
1569 vb->free = VMAP_BBMAP_BITS - (1UL << order);
Nick Piggindb64fe02008-10-18 20:27:03 -07001570 vb->dirty = 0;
Roman Pen7d61bfe2015-04-15 16:13:55 -07001571 vb->dirty_min = VMAP_BBMAP_BITS;
1572 vb->dirty_max = 0;
Nick Piggindb64fe02008-10-18 20:27:03 -07001573 INIT_LIST_HEAD(&vb->free_list);
Nick Piggindb64fe02008-10-18 20:27:03 -07001574
1575 vb_idx = addr_to_vb_idx(va->va_start);
Matthew Wilcox (Oracle)0f145992020-08-06 23:24:05 -07001576 err = xa_insert(&vmap_blocks, vb_idx, vb, gfp_mask);
1577 if (err) {
1578 kfree(vb);
1579 free_vmap_area(va);
1580 return ERR_PTR(err);
1581 }
Nick Piggindb64fe02008-10-18 20:27:03 -07001582
1583 vbq = &get_cpu_var(vmap_block_queue);
Nick Piggindb64fe02008-10-18 20:27:03 -07001584 spin_lock(&vbq->lock);
Roman Pen68ac5462015-04-15 16:13:48 -07001585 list_add_tail_rcu(&vb->free_list, &vbq->free);
Nick Piggindb64fe02008-10-18 20:27:03 -07001586 spin_unlock(&vbq->lock);
Tejun Heo3f04ba82009-10-29 22:34:12 +09001587 put_cpu_var(vmap_block_queue);
Nick Piggindb64fe02008-10-18 20:27:03 -07001588
Roman Pencf725ce2015-04-15 16:13:52 -07001589 return vaddr;
Nick Piggindb64fe02008-10-18 20:27:03 -07001590}
1591
Nick Piggindb64fe02008-10-18 20:27:03 -07001592static void free_vmap_block(struct vmap_block *vb)
1593{
1594 struct vmap_block *tmp;
Nick Piggindb64fe02008-10-18 20:27:03 -07001595
Matthew Wilcox (Oracle)0f145992020-08-06 23:24:05 -07001596 tmp = xa_erase(&vmap_blocks, addr_to_vb_idx(vb->va->va_start));
Nick Piggindb64fe02008-10-18 20:27:03 -07001597 BUG_ON(tmp != vb);
1598
Jeremy Fitzhardinge64141da2010-12-02 14:31:18 -08001599 free_vmap_area_noflush(vb->va);
Lai Jiangshan22a3c7d2011-03-18 12:13:08 +08001600 kfree_rcu(vb, rcu_head);
Nick Piggindb64fe02008-10-18 20:27:03 -07001601}
1602
Nick Piggin02b709d2010-02-01 22:25:57 +11001603static void purge_fragmented_blocks(int cpu)
1604{
1605 LIST_HEAD(purge);
1606 struct vmap_block *vb;
1607 struct vmap_block *n_vb;
1608 struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu);
1609
1610 rcu_read_lock();
1611 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
1612
1613 if (!(vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS))
1614 continue;
1615
1616 spin_lock(&vb->lock);
1617 if (vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS) {
1618 vb->free = 0; /* prevent further allocs after releasing lock */
1619 vb->dirty = VMAP_BBMAP_BITS; /* prevent purging it again */
Roman Pen7d61bfe2015-04-15 16:13:55 -07001620 vb->dirty_min = 0;
1621 vb->dirty_max = VMAP_BBMAP_BITS;
Nick Piggin02b709d2010-02-01 22:25:57 +11001622 spin_lock(&vbq->lock);
1623 list_del_rcu(&vb->free_list);
1624 spin_unlock(&vbq->lock);
1625 spin_unlock(&vb->lock);
1626 list_add_tail(&vb->purge, &purge);
1627 } else
1628 spin_unlock(&vb->lock);
1629 }
1630 rcu_read_unlock();
1631
1632 list_for_each_entry_safe(vb, n_vb, &purge, purge) {
1633 list_del(&vb->purge);
1634 free_vmap_block(vb);
1635 }
1636}
1637
Nick Piggin02b709d2010-02-01 22:25:57 +11001638static void purge_fragmented_blocks_allcpus(void)
1639{
1640 int cpu;
1641
1642 for_each_possible_cpu(cpu)
1643 purge_fragmented_blocks(cpu);
1644}
1645
Nick Piggindb64fe02008-10-18 20:27:03 -07001646static void *vb_alloc(unsigned long size, gfp_t gfp_mask)
1647{
1648 struct vmap_block_queue *vbq;
1649 struct vmap_block *vb;
Roman Pencf725ce2015-04-15 16:13:52 -07001650 void *vaddr = NULL;
Nick Piggindb64fe02008-10-18 20:27:03 -07001651 unsigned int order;
1652
Alexander Kuleshov891c49a2015-11-05 18:46:51 -08001653 BUG_ON(offset_in_page(size));
Nick Piggindb64fe02008-10-18 20:27:03 -07001654 BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
Jan Karaaa91c4d2012-07-31 16:41:37 -07001655 if (WARN_ON(size == 0)) {
1656 /*
1657 * Allocating 0 bytes isn't what caller wants since
1658 * get_order(0) returns funny result. Just warn and terminate
1659 * early.
1660 */
1661 return NULL;
1662 }
Nick Piggindb64fe02008-10-18 20:27:03 -07001663 order = get_order(size);
1664
Nick Piggindb64fe02008-10-18 20:27:03 -07001665 rcu_read_lock();
1666 vbq = &get_cpu_var(vmap_block_queue);
1667 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
Roman Pencf725ce2015-04-15 16:13:52 -07001668 unsigned long pages_off;
Nick Piggindb64fe02008-10-18 20:27:03 -07001669
1670 spin_lock(&vb->lock);
Roman Pencf725ce2015-04-15 16:13:52 -07001671 if (vb->free < (1UL << order)) {
1672 spin_unlock(&vb->lock);
1673 continue;
1674 }
Nick Piggin02b709d2010-02-01 22:25:57 +11001675
Roman Pencf725ce2015-04-15 16:13:52 -07001676 pages_off = VMAP_BBMAP_BITS - vb->free;
1677 vaddr = vmap_block_vaddr(vb->va->va_start, pages_off);
Nick Piggin02b709d2010-02-01 22:25:57 +11001678 vb->free -= 1UL << order;
1679 if (vb->free == 0) {
1680 spin_lock(&vbq->lock);
1681 list_del_rcu(&vb->free_list);
1682 spin_unlock(&vbq->lock);
Nick Piggindb64fe02008-10-18 20:27:03 -07001683 }
Roman Pencf725ce2015-04-15 16:13:52 -07001684
Nick Piggindb64fe02008-10-18 20:27:03 -07001685 spin_unlock(&vb->lock);
Nick Piggin02b709d2010-02-01 22:25:57 +11001686 break;
Nick Piggindb64fe02008-10-18 20:27:03 -07001687 }
Nick Piggin02b709d2010-02-01 22:25:57 +11001688
Tejun Heo3f04ba82009-10-29 22:34:12 +09001689 put_cpu_var(vmap_block_queue);
Nick Piggindb64fe02008-10-18 20:27:03 -07001690 rcu_read_unlock();
1691
Roman Pencf725ce2015-04-15 16:13:52 -07001692 /* Allocate new block if nothing was found */
1693 if (!vaddr)
1694 vaddr = new_vmap_block(order, gfp_mask);
Nick Piggindb64fe02008-10-18 20:27:03 -07001695
Roman Pencf725ce2015-04-15 16:13:52 -07001696 return vaddr;
Nick Piggindb64fe02008-10-18 20:27:03 -07001697}
1698
Christoph Hellwig78a0e8c2020-06-01 21:51:02 -07001699static void vb_free(unsigned long addr, unsigned long size)
Nick Piggindb64fe02008-10-18 20:27:03 -07001700{
1701 unsigned long offset;
Nick Piggindb64fe02008-10-18 20:27:03 -07001702 unsigned int order;
1703 struct vmap_block *vb;
1704
Alexander Kuleshov891c49a2015-11-05 18:46:51 -08001705 BUG_ON(offset_in_page(size));
Nick Piggindb64fe02008-10-18 20:27:03 -07001706 BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
Nick Pigginb29acbd2008-12-01 13:13:47 -08001707
Christoph Hellwig78a0e8c2020-06-01 21:51:02 -07001708 flush_cache_vunmap(addr, addr + size);
Nick Pigginb29acbd2008-12-01 13:13:47 -08001709
Nick Piggindb64fe02008-10-18 20:27:03 -07001710 order = get_order(size);
Christoph Hellwig78a0e8c2020-06-01 21:51:02 -07001711 offset = (addr & (VMAP_BLOCK_SIZE - 1)) >> PAGE_SHIFT;
Matthew Wilcox (Oracle)0f145992020-08-06 23:24:05 -07001712 vb = xa_load(&vmap_blocks, addr_to_vb_idx(addr));
Nick Piggindb64fe02008-10-18 20:27:03 -07001713
Christoph Hellwigb521c432020-06-01 21:51:07 -07001714 unmap_kernel_range_noflush(addr, size);
Jeremy Fitzhardinge64141da2010-12-02 14:31:18 -08001715
Vlastimil Babka8e57f8a2020-01-13 16:29:20 -08001716 if (debug_pagealloc_enabled_static())
Christoph Hellwig78a0e8c2020-06-01 21:51:02 -07001717 flush_tlb_kernel_range(addr, addr + size);
Chintan Pandya82a2e922018-06-07 17:06:46 -07001718
Nick Piggindb64fe02008-10-18 20:27:03 -07001719 spin_lock(&vb->lock);
Roman Pen7d61bfe2015-04-15 16:13:55 -07001720
1721 /* Expand dirty range */
1722 vb->dirty_min = min(vb->dirty_min, offset);
1723 vb->dirty_max = max(vb->dirty_max, offset + (1UL << order));
MinChan Kimd0868172009-03-31 15:19:26 -07001724
Nick Piggindb64fe02008-10-18 20:27:03 -07001725 vb->dirty += 1UL << order;
1726 if (vb->dirty == VMAP_BBMAP_BITS) {
Nick Pigginde560422010-02-01 22:24:18 +11001727 BUG_ON(vb->free);
Nick Piggindb64fe02008-10-18 20:27:03 -07001728 spin_unlock(&vb->lock);
1729 free_vmap_block(vb);
1730 } else
1731 spin_unlock(&vb->lock);
1732}
1733
Rick Edgecombe868b1042019-04-25 17:11:36 -07001734static void _vm_unmap_aliases(unsigned long start, unsigned long end, int flush)
Nick Piggindb64fe02008-10-18 20:27:03 -07001735{
Nick Piggindb64fe02008-10-18 20:27:03 -07001736 int cpu;
Nick Piggindb64fe02008-10-18 20:27:03 -07001737
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +11001738 if (unlikely(!vmap_initialized))
1739 return;
1740
Christoph Hellwig5803ed22016-12-12 16:44:20 -08001741 might_sleep();
1742
Nick Piggindb64fe02008-10-18 20:27:03 -07001743 for_each_possible_cpu(cpu) {
1744 struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu);
1745 struct vmap_block *vb;
1746
1747 rcu_read_lock();
1748 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
Nick Piggindb64fe02008-10-18 20:27:03 -07001749 spin_lock(&vb->lock);
Roman Pen7d61bfe2015-04-15 16:13:55 -07001750 if (vb->dirty) {
1751 unsigned long va_start = vb->va->va_start;
Nick Piggindb64fe02008-10-18 20:27:03 -07001752 unsigned long s, e;
Joonsoo Kimb136be5e2013-09-11 14:21:40 -07001753
Roman Pen7d61bfe2015-04-15 16:13:55 -07001754 s = va_start + (vb->dirty_min << PAGE_SHIFT);
1755 e = va_start + (vb->dirty_max << PAGE_SHIFT);
Nick Piggindb64fe02008-10-18 20:27:03 -07001756
Roman Pen7d61bfe2015-04-15 16:13:55 -07001757 start = min(s, start);
1758 end = max(e, end);
1759
Nick Piggindb64fe02008-10-18 20:27:03 -07001760 flush = 1;
Nick Piggindb64fe02008-10-18 20:27:03 -07001761 }
1762 spin_unlock(&vb->lock);
1763 }
1764 rcu_read_unlock();
1765 }
1766
Christoph Hellwigf9e09972016-12-12 16:44:23 -08001767 mutex_lock(&vmap_purge_lock);
Christoph Hellwig0574ecd2016-12-12 16:44:07 -08001768 purge_fragmented_blocks_allcpus();
1769 if (!__purge_vmap_area_lazy(start, end) && flush)
1770 flush_tlb_kernel_range(start, end);
Christoph Hellwigf9e09972016-12-12 16:44:23 -08001771 mutex_unlock(&vmap_purge_lock);
Nick Piggindb64fe02008-10-18 20:27:03 -07001772}
Rick Edgecombe868b1042019-04-25 17:11:36 -07001773
1774/**
1775 * vm_unmap_aliases - unmap outstanding lazy aliases in the vmap layer
1776 *
1777 * The vmap/vmalloc layer lazily flushes kernel virtual mappings primarily
1778 * to amortize TLB flushing overheads. What this means is that any page you
1779 * have now, may, in a former life, have been mapped into kernel virtual
1780 * address by the vmap layer and so there might be some CPUs with TLB entries
1781 * still referencing that page (additional to the regular 1:1 kernel mapping).
1782 *
1783 * vm_unmap_aliases flushes all such lazy mappings. After it returns, we can
1784 * be sure that none of the pages we have control over will have any aliases
1785 * from the vmap layer.
1786 */
1787void vm_unmap_aliases(void)
1788{
1789 unsigned long start = ULONG_MAX, end = 0;
1790 int flush = 0;
1791
1792 _vm_unmap_aliases(start, end, flush);
1793}
Nick Piggindb64fe02008-10-18 20:27:03 -07001794EXPORT_SYMBOL_GPL(vm_unmap_aliases);
1795
1796/**
1797 * vm_unmap_ram - unmap linear kernel address space set up by vm_map_ram
1798 * @mem: the pointer returned by vm_map_ram
1799 * @count: the count passed to that vm_map_ram call (cannot unmap partial)
1800 */
1801void vm_unmap_ram(const void *mem, unsigned int count)
1802{
Guillermo Julián Moreno65ee03c2016-06-03 14:55:33 -07001803 unsigned long size = (unsigned long)count << PAGE_SHIFT;
Nick Piggindb64fe02008-10-18 20:27:03 -07001804 unsigned long addr = (unsigned long)mem;
Christoph Hellwig9c3acf62016-12-12 16:44:04 -08001805 struct vmap_area *va;
Nick Piggindb64fe02008-10-18 20:27:03 -07001806
Christoph Hellwig5803ed22016-12-12 16:44:20 -08001807 might_sleep();
Nick Piggindb64fe02008-10-18 20:27:03 -07001808 BUG_ON(!addr);
1809 BUG_ON(addr < VMALLOC_START);
1810 BUG_ON(addr > VMALLOC_END);
Shawn Lina1c0b1a2016-03-17 14:20:37 -07001811 BUG_ON(!PAGE_ALIGNED(addr));
Nick Piggindb64fe02008-10-18 20:27:03 -07001812
Andrey Ryabinind98c9e82019-12-17 20:51:38 -08001813 kasan_poison_vmalloc(mem, size);
1814
Christoph Hellwig9c3acf62016-12-12 16:44:04 -08001815 if (likely(count <= VMAP_MAX_ALLOC)) {
Chintan Pandya05e3ff92018-06-07 17:06:53 -07001816 debug_check_no_locks_freed(mem, size);
Christoph Hellwig78a0e8c2020-06-01 21:51:02 -07001817 vb_free(addr, size);
Christoph Hellwig9c3acf62016-12-12 16:44:04 -08001818 return;
1819 }
1820
1821 va = find_vmap_area(addr);
1822 BUG_ON(!va);
Chintan Pandya05e3ff92018-06-07 17:06:53 -07001823 debug_check_no_locks_freed((void *)va->va_start,
1824 (va->va_end - va->va_start));
Christoph Hellwig9c3acf62016-12-12 16:44:04 -08001825 free_unmap_vmap_area(va);
Nick Piggindb64fe02008-10-18 20:27:03 -07001826}
1827EXPORT_SYMBOL(vm_unmap_ram);
1828
1829/**
1830 * vm_map_ram - map pages linearly into kernel virtual address (vmalloc space)
1831 * @pages: an array of pointers to the pages to be mapped
1832 * @count: number of pages
1833 * @node: prefer to allocate data structures on this node
Randy Dunlape99c97a2008-10-29 14:01:09 -07001834 *
Gioh Kim36437632014-04-07 15:37:37 -07001835 * If you use this function for less than VMAP_MAX_ALLOC pages, it could be
1836 * faster than vmap so it's good. But if you mix long-life and short-life
1837 * objects with vm_map_ram(), it could consume lots of address space through
1838 * fragmentation (especially on a 32bit machine). You could see failures in
1839 * the end. Please use this function for short-lived objects.
1840 *
Randy Dunlape99c97a2008-10-29 14:01:09 -07001841 * Returns: a pointer to the address that has been mapped, or %NULL on failure
Nick Piggindb64fe02008-10-18 20:27:03 -07001842 */
Christoph Hellwigd4efd792020-06-01 21:51:27 -07001843void *vm_map_ram(struct page **pages, unsigned int count, int node)
Nick Piggindb64fe02008-10-18 20:27:03 -07001844{
Guillermo Julián Moreno65ee03c2016-06-03 14:55:33 -07001845 unsigned long size = (unsigned long)count << PAGE_SHIFT;
Nick Piggindb64fe02008-10-18 20:27:03 -07001846 unsigned long addr;
1847 void *mem;
1848
1849 if (likely(count <= VMAP_MAX_ALLOC)) {
1850 mem = vb_alloc(size, GFP_KERNEL);
1851 if (IS_ERR(mem))
1852 return NULL;
1853 addr = (unsigned long)mem;
1854 } else {
1855 struct vmap_area *va;
1856 va = alloc_vmap_area(size, PAGE_SIZE,
1857 VMALLOC_START, VMALLOC_END, node, GFP_KERNEL);
1858 if (IS_ERR(va))
1859 return NULL;
1860
1861 addr = va->va_start;
1862 mem = (void *)addr;
1863 }
Andrey Ryabinind98c9e82019-12-17 20:51:38 -08001864
1865 kasan_unpoison_vmalloc(mem, size);
1866
Christoph Hellwigd4efd792020-06-01 21:51:27 -07001867 if (map_kernel_range(addr, size, PAGE_KERNEL, pages) < 0) {
Nick Piggindb64fe02008-10-18 20:27:03 -07001868 vm_unmap_ram(mem, count);
1869 return NULL;
1870 }
1871 return mem;
1872}
1873EXPORT_SYMBOL(vm_map_ram);
1874
Joonsoo Kim4341fa42013-04-29 15:07:39 -07001875static struct vm_struct *vmlist __initdata;
Mike Rapoport92eac162019-03-05 15:48:36 -08001876
Tejun Heof0aa6612009-02-20 16:29:08 +09001877/**
Nicolas Pitrebe9b7332011-08-25 00:24:21 -04001878 * vm_area_add_early - add vmap area early during boot
1879 * @vm: vm_struct to add
1880 *
1881 * This function is used to add fixed kernel vm area to vmlist before
1882 * vmalloc_init() is called. @vm->addr, @vm->size, and @vm->flags
1883 * should contain proper values and the other fields should be zero.
1884 *
1885 * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1886 */
1887void __init vm_area_add_early(struct vm_struct *vm)
1888{
1889 struct vm_struct *tmp, **p;
1890
1891 BUG_ON(vmap_initialized);
1892 for (p = &vmlist; (tmp = *p) != NULL; p = &tmp->next) {
1893 if (tmp->addr >= vm->addr) {
1894 BUG_ON(tmp->addr < vm->addr + vm->size);
1895 break;
1896 } else
1897 BUG_ON(tmp->addr + tmp->size > vm->addr);
1898 }
1899 vm->next = *p;
1900 *p = vm;
1901}
1902
1903/**
Tejun Heof0aa6612009-02-20 16:29:08 +09001904 * vm_area_register_early - register vmap area early during boot
1905 * @vm: vm_struct to register
Tejun Heoc0c0a292009-02-24 11:57:21 +09001906 * @align: requested alignment
Tejun Heof0aa6612009-02-20 16:29:08 +09001907 *
1908 * This function is used to register kernel vm area before
1909 * vmalloc_init() is called. @vm->size and @vm->flags should contain
1910 * proper values on entry and other fields should be zero. On return,
1911 * vm->addr contains the allocated address.
1912 *
1913 * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1914 */
Tejun Heoc0c0a292009-02-24 11:57:21 +09001915void __init vm_area_register_early(struct vm_struct *vm, size_t align)
Tejun Heof0aa6612009-02-20 16:29:08 +09001916{
1917 static size_t vm_init_off __initdata;
Tejun Heoc0c0a292009-02-24 11:57:21 +09001918 unsigned long addr;
Tejun Heof0aa6612009-02-20 16:29:08 +09001919
Tejun Heoc0c0a292009-02-24 11:57:21 +09001920 addr = ALIGN(VMALLOC_START + vm_init_off, align);
1921 vm_init_off = PFN_ALIGN(addr + vm->size) - VMALLOC_START;
1922
1923 vm->addr = (void *)addr;
Tejun Heof0aa6612009-02-20 16:29:08 +09001924
Nicolas Pitrebe9b7332011-08-25 00:24:21 -04001925 vm_area_add_early(vm);
Tejun Heof0aa6612009-02-20 16:29:08 +09001926}
1927
Uladzislau Rezki (Sony)68ad4a32019-05-17 14:31:31 -07001928static void vmap_init_free_space(void)
1929{
1930 unsigned long vmap_start = 1;
1931 const unsigned long vmap_end = ULONG_MAX;
1932 struct vmap_area *busy, *free;
1933
1934 /*
1935 * B F B B B F
1936 * -|-----|.....|-----|-----|-----|.....|-
1937 * | The KVA space |
1938 * |<--------------------------------->|
1939 */
1940 list_for_each_entry(busy, &vmap_area_list, list) {
1941 if (busy->va_start - vmap_start > 0) {
1942 free = kmem_cache_zalloc(vmap_area_cachep, GFP_NOWAIT);
1943 if (!WARN_ON_ONCE(!free)) {
1944 free->va_start = vmap_start;
1945 free->va_end = busy->va_start;
1946
1947 insert_vmap_area_augment(free, NULL,
1948 &free_vmap_area_root,
1949 &free_vmap_area_list);
1950 }
1951 }
1952
1953 vmap_start = busy->va_end;
1954 }
1955
1956 if (vmap_end - vmap_start > 0) {
1957 free = kmem_cache_zalloc(vmap_area_cachep, GFP_NOWAIT);
1958 if (!WARN_ON_ONCE(!free)) {
1959 free->va_start = vmap_start;
1960 free->va_end = vmap_end;
1961
1962 insert_vmap_area_augment(free, NULL,
1963 &free_vmap_area_root,
1964 &free_vmap_area_list);
1965 }
1966 }
1967}
1968
Nick Piggindb64fe02008-10-18 20:27:03 -07001969void __init vmalloc_init(void)
1970{
Ivan Kokshaysky822c18f2009-01-15 13:50:48 -08001971 struct vmap_area *va;
1972 struct vm_struct *tmp;
Nick Piggindb64fe02008-10-18 20:27:03 -07001973 int i;
1974
Uladzislau Rezki (Sony)68ad4a32019-05-17 14:31:31 -07001975 /*
1976 * Create the cache for vmap_area objects.
1977 */
1978 vmap_area_cachep = KMEM_CACHE(vmap_area, SLAB_PANIC);
1979
Nick Piggindb64fe02008-10-18 20:27:03 -07001980 for_each_possible_cpu(i) {
1981 struct vmap_block_queue *vbq;
Al Viro32fcfd42013-03-10 20:14:08 -04001982 struct vfree_deferred *p;
Nick Piggindb64fe02008-10-18 20:27:03 -07001983
1984 vbq = &per_cpu(vmap_block_queue, i);
1985 spin_lock_init(&vbq->lock);
1986 INIT_LIST_HEAD(&vbq->free);
Al Viro32fcfd42013-03-10 20:14:08 -04001987 p = &per_cpu(vfree_deferred, i);
1988 init_llist_head(&p->list);
1989 INIT_WORK(&p->wq, free_work);
Nick Piggindb64fe02008-10-18 20:27:03 -07001990 }
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +11001991
Ivan Kokshaysky822c18f2009-01-15 13:50:48 -08001992 /* Import existing vmlist entries. */
1993 for (tmp = vmlist; tmp; tmp = tmp->next) {
Uladzislau Rezki (Sony)68ad4a32019-05-17 14:31:31 -07001994 va = kmem_cache_zalloc(vmap_area_cachep, GFP_NOWAIT);
1995 if (WARN_ON_ONCE(!va))
1996 continue;
1997
Ivan Kokshaysky822c18f2009-01-15 13:50:48 -08001998 va->va_start = (unsigned long)tmp->addr;
1999 va->va_end = va->va_start + tmp->size;
KyongHodbda5912012-05-29 15:06:49 -07002000 va->vm = tmp;
Uladzislau Rezki (Sony)68ad4a32019-05-17 14:31:31 -07002001 insert_vmap_area(va, &vmap_area_root, &vmap_area_list);
Ivan Kokshaysky822c18f2009-01-15 13:50:48 -08002002 }
Tejun Heoca23e402009-08-14 15:00:52 +09002003
Uladzislau Rezki (Sony)68ad4a32019-05-17 14:31:31 -07002004 /*
2005 * Now we can initialize a free vmap space.
2006 */
2007 vmap_init_free_space();
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +11002008 vmap_initialized = true;
Nick Piggindb64fe02008-10-18 20:27:03 -07002009}
2010
Tejun Heo8fc48982009-02-20 16:29:08 +09002011/**
Tejun Heo8fc48982009-02-20 16:29:08 +09002012 * unmap_kernel_range - unmap kernel VM area and flush cache and TLB
2013 * @addr: start of the VM area to unmap
2014 * @size: size of the VM area to unmap
2015 *
2016 * Similar to unmap_kernel_range_noflush() but flushes vcache before
2017 * the unmapping and tlb after.
2018 */
Nick Piggindb64fe02008-10-18 20:27:03 -07002019void unmap_kernel_range(unsigned long addr, unsigned long size)
2020{
2021 unsigned long end = addr + size;
Tejun Heof6fcba72009-02-20 15:38:48 -08002022
2023 flush_cache_vunmap(addr, end);
Christoph Hellwigb521c432020-06-01 21:51:07 -07002024 unmap_kernel_range_noflush(addr, size);
Nick Piggindb64fe02008-10-18 20:27:03 -07002025 flush_tlb_kernel_range(addr, end);
2026}
2027
Uladzislau Rezki (Sony)e36176b2019-11-30 17:54:47 -08002028static inline void setup_vmalloc_vm_locked(struct vm_struct *vm,
2029 struct vmap_area *va, unsigned long flags, const void *caller)
Tejun Heocf88c792009-08-14 15:00:52 +09002030{
Tejun Heocf88c792009-08-14 15:00:52 +09002031 vm->flags = flags;
2032 vm->addr = (void *)va->va_start;
2033 vm->size = va->va_end - va->va_start;
2034 vm->caller = caller;
Minchan Kimdb1aeca2012-01-10 15:08:39 -08002035 va->vm = vm;
Uladzislau Rezki (Sony)e36176b2019-11-30 17:54:47 -08002036}
2037
2038static void setup_vmalloc_vm(struct vm_struct *vm, struct vmap_area *va,
2039 unsigned long flags, const void *caller)
2040{
2041 spin_lock(&vmap_area_lock);
2042 setup_vmalloc_vm_locked(vm, va, flags, caller);
Joonsoo Kimc69480a2013-04-29 15:07:30 -07002043 spin_unlock(&vmap_area_lock);
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07002044}
Tejun Heocf88c792009-08-14 15:00:52 +09002045
Zhang Yanfei20fc02b2013-07-08 15:59:58 -07002046static void clear_vm_uninitialized_flag(struct vm_struct *vm)
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07002047{
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002048 /*
Zhang Yanfei20fc02b2013-07-08 15:59:58 -07002049 * Before removing VM_UNINITIALIZED,
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002050 * we should make sure that vm has proper values.
2051 * Pair with smp_rmb() in show_numa_info().
2052 */
2053 smp_wmb();
Zhang Yanfei20fc02b2013-07-08 15:59:58 -07002054 vm->flags &= ~VM_UNINITIALIZED;
Tejun Heocf88c792009-08-14 15:00:52 +09002055}
2056
Nick Piggindb64fe02008-10-18 20:27:03 -07002057static struct vm_struct *__get_vm_area_node(unsigned long size,
David Miller2dca6992009-09-21 12:22:34 -07002058 unsigned long align, unsigned long flags, unsigned long start,
Marek Szyprowski5e6cafc2012-04-13 12:32:09 +02002059 unsigned long end, int node, gfp_t gfp_mask, const void *caller)
Nick Piggindb64fe02008-10-18 20:27:03 -07002060{
Kautuk Consul00065262011-12-19 17:12:04 -08002061 struct vmap_area *va;
Nick Piggindb64fe02008-10-18 20:27:03 -07002062 struct vm_struct *area;
Andrey Ryabinind98c9e82019-12-17 20:51:38 -08002063 unsigned long requested_size = size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002064
Giridhar Pemmasani52fd24c2006-10-28 10:38:34 -07002065 BUG_ON(in_interrupt());
Linus Torvalds1da177e2005-04-16 15:20:36 -07002066 size = PAGE_ALIGN(size);
OGAWA Hirofumi31be8302006-11-16 01:19:29 -08002067 if (unlikely(!size))
2068 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002069
zijun_hu252e5c62016-10-07 16:57:26 -07002070 if (flags & VM_IOREMAP)
2071 align = 1ul << clamp_t(int, get_count_order_long(size),
2072 PAGE_SHIFT, IOREMAP_MAX_ORDER);
2073
Tejun Heocf88c792009-08-14 15:00:52 +09002074 area = kzalloc_node(sizeof(*area), gfp_mask & GFP_RECLAIM_MASK, node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002075 if (unlikely(!area))
2076 return NULL;
2077
Andrey Ryabinin71394fe2015-02-13 14:40:03 -08002078 if (!(flags & VM_NO_GUARD))
2079 size += PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002080
Nick Piggindb64fe02008-10-18 20:27:03 -07002081 va = alloc_vmap_area(size, align, start, end, node, gfp_mask);
2082 if (IS_ERR(va)) {
2083 kfree(area);
2084 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002085 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002086
Andrey Ryabinind98c9e82019-12-17 20:51:38 -08002087 kasan_unpoison_vmalloc((void *)va->va_start, requested_size);
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07002088
Andrey Ryabinind98c9e82019-12-17 20:51:38 -08002089 setup_vmalloc_vm(area, va, flags, caller);
Daniel Axtens3c5c3cf2019-11-30 17:54:50 -08002090
Linus Torvalds1da177e2005-04-16 15:20:36 -07002091 return area;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002092}
2093
Benjamin Herrenschmidtc2968612009-02-18 14:48:12 -08002094struct vm_struct *__get_vm_area_caller(unsigned long size, unsigned long flags,
2095 unsigned long start, unsigned long end,
Marek Szyprowski5e6cafc2012-04-13 12:32:09 +02002096 const void *caller)
Benjamin Herrenschmidtc2968612009-02-18 14:48:12 -08002097{
David Rientjes00ef2d22013-02-22 16:35:36 -08002098 return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE,
2099 GFP_KERNEL, caller);
Benjamin Herrenschmidtc2968612009-02-18 14:48:12 -08002100}
2101
Linus Torvalds1da177e2005-04-16 15:20:36 -07002102/**
Mike Rapoport92eac162019-03-05 15:48:36 -08002103 * get_vm_area - reserve a contiguous kernel virtual area
2104 * @size: size of the area
2105 * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC
Linus Torvalds1da177e2005-04-16 15:20:36 -07002106 *
Mike Rapoport92eac162019-03-05 15:48:36 -08002107 * Search an area of @size in the kernel virtual mapping area,
2108 * and reserved it for out purposes. Returns the area descriptor
2109 * on success or %NULL on failure.
Mike Rapoporta862f682019-03-05 15:48:42 -08002110 *
2111 * Return: the area descriptor on success or %NULL on failure.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002112 */
2113struct vm_struct *get_vm_area(unsigned long size, unsigned long flags)
2114{
David Miller2dca6992009-09-21 12:22:34 -07002115 return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END,
David Rientjes00ef2d22013-02-22 16:35:36 -08002116 NUMA_NO_NODE, GFP_KERNEL,
2117 __builtin_return_address(0));
Christoph Lameter23016962008-04-28 02:12:42 -07002118}
2119
2120struct vm_struct *get_vm_area_caller(unsigned long size, unsigned long flags,
Marek Szyprowski5e6cafc2012-04-13 12:32:09 +02002121 const void *caller)
Christoph Lameter23016962008-04-28 02:12:42 -07002122{
David Miller2dca6992009-09-21 12:22:34 -07002123 return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END,
David Rientjes00ef2d22013-02-22 16:35:36 -08002124 NUMA_NO_NODE, GFP_KERNEL, caller);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002125}
2126
Marek Szyprowskie9da6e92012-07-30 09:11:33 +02002127/**
Mike Rapoport92eac162019-03-05 15:48:36 -08002128 * find_vm_area - find a continuous kernel virtual area
2129 * @addr: base address
Marek Szyprowskie9da6e92012-07-30 09:11:33 +02002130 *
Mike Rapoport92eac162019-03-05 15:48:36 -08002131 * Search for the kernel VM area starting at @addr, and return it.
2132 * It is up to the caller to do all required locking to keep the returned
2133 * pointer valid.
Mike Rapoporta862f682019-03-05 15:48:42 -08002134 *
Hui Su74640612020-10-13 16:54:51 -07002135 * Return: the area descriptor on success or %NULL on failure.
Marek Szyprowskie9da6e92012-07-30 09:11:33 +02002136 */
2137struct vm_struct *find_vm_area(const void *addr)
Nick Piggin83342312006-06-23 02:03:20 -07002138{
Nick Piggindb64fe02008-10-18 20:27:03 -07002139 struct vmap_area *va;
Nick Piggin83342312006-06-23 02:03:20 -07002140
Nick Piggindb64fe02008-10-18 20:27:03 -07002141 va = find_vmap_area((unsigned long)addr);
Pengfei Li688fcbf2019-09-23 15:36:39 -07002142 if (!va)
2143 return NULL;
Nick Piggin83342312006-06-23 02:03:20 -07002144
Pengfei Li688fcbf2019-09-23 15:36:39 -07002145 return va->vm;
Andi Kleen7856dfe2005-05-20 14:27:57 -07002146}
2147
Linus Torvalds1da177e2005-04-16 15:20:36 -07002148/**
Mike Rapoport92eac162019-03-05 15:48:36 -08002149 * remove_vm_area - find and remove a continuous kernel virtual area
2150 * @addr: base address
Linus Torvalds1da177e2005-04-16 15:20:36 -07002151 *
Mike Rapoport92eac162019-03-05 15:48:36 -08002152 * Search for the kernel VM area starting at @addr, and remove it.
2153 * This function returns the found VM area, but using it is NOT safe
2154 * on SMP machines, except for its size or flags.
Mike Rapoporta862f682019-03-05 15:48:42 -08002155 *
Hui Su74640612020-10-13 16:54:51 -07002156 * Return: the area descriptor on success or %NULL on failure.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002157 */
Christoph Lameterb3bdda02008-02-04 22:28:32 -08002158struct vm_struct *remove_vm_area(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002159{
Nick Piggindb64fe02008-10-18 20:27:03 -07002160 struct vmap_area *va;
2161
Christoph Hellwig5803ed22016-12-12 16:44:20 -08002162 might_sleep();
2163
Uladzislau Rezki (Sony)dd3b8352019-09-23 15:36:36 -07002164 spin_lock(&vmap_area_lock);
2165 va = __find_vmap_area((unsigned long)addr);
Pengfei Li688fcbf2019-09-23 15:36:39 -07002166 if (va && va->vm) {
Minchan Kimdb1aeca2012-01-10 15:08:39 -08002167 struct vm_struct *vm = va->vm;
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07002168
Joonsoo Kimc69480a2013-04-29 15:07:30 -07002169 va->vm = NULL;
Joonsoo Kimc69480a2013-04-29 15:07:30 -07002170 spin_unlock(&vmap_area_lock);
2171
Andrey Ryabinina5af5aa2015-03-12 16:26:11 -07002172 kasan_free_shadow(vm);
KAMEZAWA Hiroyukidd32c272009-09-21 17:02:32 -07002173 free_unmap_vmap_area(va);
KAMEZAWA Hiroyukidd32c272009-09-21 17:02:32 -07002174
Nick Piggindb64fe02008-10-18 20:27:03 -07002175 return vm;
2176 }
Uladzislau Rezki (Sony)dd3b8352019-09-23 15:36:36 -07002177
2178 spin_unlock(&vmap_area_lock);
Nick Piggindb64fe02008-10-18 20:27:03 -07002179 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002180}
2181
Rick Edgecombe868b1042019-04-25 17:11:36 -07002182static inline void set_area_direct_map(const struct vm_struct *area,
2183 int (*set_direct_map)(struct page *page))
2184{
2185 int i;
2186
2187 for (i = 0; i < area->nr_pages; i++)
2188 if (page_address(area->pages[i]))
2189 set_direct_map(area->pages[i]);
2190}
2191
2192/* Handle removing and resetting vm mappings related to the vm_struct. */
2193static void vm_remove_mappings(struct vm_struct *area, int deallocate_pages)
2194{
Rick Edgecombe868b1042019-04-25 17:11:36 -07002195 unsigned long start = ULONG_MAX, end = 0;
2196 int flush_reset = area->flags & VM_FLUSH_RESET_PERMS;
Rick Edgecombe31e67342019-05-27 14:10:58 -07002197 int flush_dmap = 0;
Rick Edgecombe868b1042019-04-25 17:11:36 -07002198 int i;
2199
Rick Edgecombe868b1042019-04-25 17:11:36 -07002200 remove_vm_area(area->addr);
2201
2202 /* If this is not VM_FLUSH_RESET_PERMS memory, no need for the below. */
2203 if (!flush_reset)
2204 return;
2205
2206 /*
2207 * If not deallocating pages, just do the flush of the VM area and
2208 * return.
2209 */
2210 if (!deallocate_pages) {
2211 vm_unmap_aliases();
2212 return;
2213 }
2214
2215 /*
2216 * If execution gets here, flush the vm mapping and reset the direct
2217 * map. Find the start and end range of the direct mappings to make sure
2218 * the vm_unmap_aliases() flush includes the direct map.
2219 */
2220 for (i = 0; i < area->nr_pages; i++) {
Rick Edgecombe8e41f872019-05-27 14:10:57 -07002221 unsigned long addr = (unsigned long)page_address(area->pages[i]);
2222 if (addr) {
Rick Edgecombe868b1042019-04-25 17:11:36 -07002223 start = min(addr, start);
Rick Edgecombe8e41f872019-05-27 14:10:57 -07002224 end = max(addr + PAGE_SIZE, end);
Rick Edgecombe31e67342019-05-27 14:10:58 -07002225 flush_dmap = 1;
Rick Edgecombe868b1042019-04-25 17:11:36 -07002226 }
2227 }
2228
2229 /*
2230 * Set direct map to something invalid so that it won't be cached if
2231 * there are any accesses after the TLB flush, then flush the TLB and
2232 * reset the direct map permissions to the default.
2233 */
2234 set_area_direct_map(area, set_direct_map_invalid_noflush);
Rick Edgecombe31e67342019-05-27 14:10:58 -07002235 _vm_unmap_aliases(start, end, flush_dmap);
Rick Edgecombe868b1042019-04-25 17:11:36 -07002236 set_area_direct_map(area, set_direct_map_default_noflush);
2237}
2238
Christoph Lameterb3bdda02008-02-04 22:28:32 -08002239static void __vunmap(const void *addr, int deallocate_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002240{
2241 struct vm_struct *area;
2242
2243 if (!addr)
2244 return;
2245
HATAYAMA Daisukee69e9d4a2013-07-03 15:02:18 -07002246 if (WARN(!PAGE_ALIGNED(addr), "Trying to vfree() bad address (%p)\n",
Dan Carpenterab15d9b2013-07-08 15:59:53 -07002247 addr))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002248 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002249
Liviu Dudau6ade2032019-03-05 15:42:54 -08002250 area = find_vm_area(addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002251 if (unlikely(!area)) {
Arjan van de Ven4c8573e2008-07-25 19:45:37 -07002252 WARN(1, KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002253 addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002254 return;
2255 }
2256
Chintan Pandya05e3ff92018-06-07 17:06:53 -07002257 debug_check_no_locks_freed(area->addr, get_vm_area_size(area));
2258 debug_check_no_obj_freed(area->addr, get_vm_area_size(area));
Ingo Molnar9a11b49a2006-07-03 00:24:33 -07002259
Vincenzo Frascino23713b42020-12-14 19:09:06 -08002260 kasan_poison_vmalloc(area->addr, get_vm_area_size(area));
Daniel Axtens3c5c3cf2019-11-30 17:54:50 -08002261
Rick Edgecombe868b1042019-04-25 17:11:36 -07002262 vm_remove_mappings(area, deallocate_pages);
2263
Linus Torvalds1da177e2005-04-16 15:20:36 -07002264 if (deallocate_pages) {
2265 int i;
2266
2267 for (i = 0; i < area->nr_pages; i++) {
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08002268 struct page *page = area->pages[i];
2269
2270 BUG_ON(!page);
Vladimir Davydov49491482016-07-26 15:24:24 -07002271 __free_pages(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002272 }
Roman Gushchin97105f02019-07-11 21:00:13 -07002273 atomic_long_sub(area->nr_pages, &nr_vmalloc_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002274
David Rientjes244d63e2016-01-14 15:19:35 -08002275 kvfree(area->pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002276 }
2277
2278 kfree(area);
2279 return;
2280}
Andrey Ryabininbf22e372016-12-12 16:44:10 -08002281
2282static inline void __vfree_deferred(const void *addr)
2283{
2284 /*
2285 * Use raw_cpu_ptr() because this can be called from preemptible
2286 * context. Preemption is absolutely fine here, because the llist_add()
2287 * implementation is lockless, so it works even if we are adding to
Jeongtae Park73221d82020-06-04 16:47:19 -07002288 * another cpu's list. schedule_work() should be fine with this too.
Andrey Ryabininbf22e372016-12-12 16:44:10 -08002289 */
2290 struct vfree_deferred *p = raw_cpu_ptr(&vfree_deferred);
2291
2292 if (llist_add((struct llist_node *)addr, &p->list))
2293 schedule_work(&p->wq);
2294}
2295
2296/**
Mike Rapoport92eac162019-03-05 15:48:36 -08002297 * vfree_atomic - release memory allocated by vmalloc()
2298 * @addr: memory base address
Andrey Ryabininbf22e372016-12-12 16:44:10 -08002299 *
Mike Rapoport92eac162019-03-05 15:48:36 -08002300 * This one is just like vfree() but can be called in any atomic context
2301 * except NMIs.
Andrey Ryabininbf22e372016-12-12 16:44:10 -08002302 */
2303void vfree_atomic(const void *addr)
2304{
2305 BUG_ON(in_nmi());
2306
2307 kmemleak_free(addr);
2308
2309 if (!addr)
2310 return;
2311 __vfree_deferred(addr);
2312}
2313
Roman Penyaevc67dc622019-03-05 15:43:24 -08002314static void __vfree(const void *addr)
2315{
2316 if (unlikely(in_interrupt()))
2317 __vfree_deferred(addr);
2318 else
2319 __vunmap(addr, 1);
2320}
2321
Linus Torvalds1da177e2005-04-16 15:20:36 -07002322/**
Matthew Wilcox (Oracle)fa307472020-10-17 16:15:03 -07002323 * vfree - Release memory allocated by vmalloc()
2324 * @addr: Memory base address
Linus Torvalds1da177e2005-04-16 15:20:36 -07002325 *
Matthew Wilcox (Oracle)fa307472020-10-17 16:15:03 -07002326 * Free the virtually continuous memory area starting at @addr, as obtained
2327 * from one of the vmalloc() family of APIs. This will usually also free the
2328 * physical memory underlying the virtual allocation, but that memory is
2329 * reference counted, so it will not be freed until the last user goes away.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002330 *
Matthew Wilcox (Oracle)fa307472020-10-17 16:15:03 -07002331 * If @addr is NULL, no operation is performed.
Andrew Mortonc9fcee52013-05-07 16:18:18 -07002332 *
Matthew Wilcox (Oracle)fa307472020-10-17 16:15:03 -07002333 * Context:
Mike Rapoport92eac162019-03-05 15:48:36 -08002334 * May sleep if called *not* from interrupt context.
Matthew Wilcox (Oracle)fa307472020-10-17 16:15:03 -07002335 * Must not be called in NMI context (strictly speaking, it could be
2336 * if we have CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG, but making the calling
2337 * conventions for vfree() arch-depenedent would be a really bad idea).
Linus Torvalds1da177e2005-04-16 15:20:36 -07002338 */
Christoph Lameterb3bdda02008-02-04 22:28:32 -08002339void vfree(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002340{
Al Viro32fcfd42013-03-10 20:14:08 -04002341 BUG_ON(in_nmi());
Catalin Marinas89219d32009-06-11 13:23:19 +01002342
2343 kmemleak_free(addr);
2344
Andrey Ryabinina8dda162018-10-26 15:07:07 -07002345 might_sleep_if(!in_interrupt());
2346
Al Viro32fcfd42013-03-10 20:14:08 -04002347 if (!addr)
2348 return;
Roman Penyaevc67dc622019-03-05 15:43:24 -08002349
2350 __vfree(addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002351}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002352EXPORT_SYMBOL(vfree);
2353
2354/**
Mike Rapoport92eac162019-03-05 15:48:36 -08002355 * vunmap - release virtual mapping obtained by vmap()
2356 * @addr: memory base address
Linus Torvalds1da177e2005-04-16 15:20:36 -07002357 *
Mike Rapoport92eac162019-03-05 15:48:36 -08002358 * Free the virtually contiguous memory area starting at @addr,
2359 * which was created from the page array passed to vmap().
Linus Torvalds1da177e2005-04-16 15:20:36 -07002360 *
Mike Rapoport92eac162019-03-05 15:48:36 -08002361 * Must not be called in interrupt context.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002362 */
Christoph Lameterb3bdda02008-02-04 22:28:32 -08002363void vunmap(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002364{
2365 BUG_ON(in_interrupt());
Peter Zijlstra34754b62009-02-25 16:04:03 +01002366 might_sleep();
Al Viro32fcfd42013-03-10 20:14:08 -04002367 if (addr)
2368 __vunmap(addr, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002369}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002370EXPORT_SYMBOL(vunmap);
2371
2372/**
Mike Rapoport92eac162019-03-05 15:48:36 -08002373 * vmap - map an array of pages into virtually contiguous space
2374 * @pages: array of page pointers
2375 * @count: number of pages to map
2376 * @flags: vm_area->flags
2377 * @prot: page protection for the mapping
Linus Torvalds1da177e2005-04-16 15:20:36 -07002378 *
Christoph Hellwigb944afc2020-10-17 16:15:06 -07002379 * Maps @count pages from @pages into contiguous kernel virtual space.
2380 * If @flags contains %VM_MAP_PUT_PAGES the ownership of the pages array itself
2381 * (which must be kmalloc or vmalloc memory) and one reference per pages in it
2382 * are transferred from the caller to vmap(), and will be freed / dropped when
2383 * vfree() is called on the return value.
Mike Rapoporta862f682019-03-05 15:48:42 -08002384 *
2385 * Return: the address of the area or %NULL on failure
Linus Torvalds1da177e2005-04-16 15:20:36 -07002386 */
2387void *vmap(struct page **pages, unsigned int count,
Mike Rapoport92eac162019-03-05 15:48:36 -08002388 unsigned long flags, pgprot_t prot)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002389{
2390 struct vm_struct *area;
Guillermo Julián Moreno65ee03c2016-06-03 14:55:33 -07002391 unsigned long size; /* In bytes */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002392
Peter Zijlstra34754b62009-02-25 16:04:03 +01002393 might_sleep();
2394
Arun KSca79b0c2018-12-28 00:34:29 -08002395 if (count > totalram_pages())
Linus Torvalds1da177e2005-04-16 15:20:36 -07002396 return NULL;
2397
Guillermo Julián Moreno65ee03c2016-06-03 14:55:33 -07002398 size = (unsigned long)count << PAGE_SHIFT;
2399 area = get_vm_area_caller(size, flags, __builtin_return_address(0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002400 if (!area)
2401 return NULL;
Christoph Lameter23016962008-04-28 02:12:42 -07002402
Christoph Hellwigcca98e92020-06-01 21:51:32 -07002403 if (map_kernel_range((unsigned long)area->addr, size, pgprot_nx(prot),
Christoph Hellwiged1f3242020-06-01 21:51:19 -07002404 pages) < 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002405 vunmap(area->addr);
2406 return NULL;
2407 }
2408
Christoph Hellwigb944afc2020-10-17 16:15:06 -07002409 if (flags & VM_MAP_PUT_PAGES)
2410 area->pages = pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002411 return area->addr;
2412}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002413EXPORT_SYMBOL(vmap);
2414
Christoph Hellwig3e9a9e22020-10-17 16:15:10 -07002415#ifdef CONFIG_VMAP_PFN
2416struct vmap_pfn_data {
2417 unsigned long *pfns;
2418 pgprot_t prot;
2419 unsigned int idx;
2420};
2421
2422static int vmap_pfn_apply(pte_t *pte, unsigned long addr, void *private)
2423{
2424 struct vmap_pfn_data *data = private;
2425
2426 if (WARN_ON_ONCE(pfn_valid(data->pfns[data->idx])))
2427 return -EINVAL;
2428 *pte = pte_mkspecial(pfn_pte(data->pfns[data->idx++], data->prot));
2429 return 0;
2430}
2431
2432/**
2433 * vmap_pfn - map an array of PFNs into virtually contiguous space
2434 * @pfns: array of PFNs
2435 * @count: number of pages to map
2436 * @prot: page protection for the mapping
2437 *
2438 * Maps @count PFNs from @pfns into contiguous kernel virtual space and returns
2439 * the start address of the mapping.
2440 */
2441void *vmap_pfn(unsigned long *pfns, unsigned int count, pgprot_t prot)
2442{
2443 struct vmap_pfn_data data = { .pfns = pfns, .prot = pgprot_nx(prot) };
2444 struct vm_struct *area;
2445
2446 area = get_vm_area_caller(count * PAGE_SIZE, VM_IOREMAP,
2447 __builtin_return_address(0));
2448 if (!area)
2449 return NULL;
2450 if (apply_to_page_range(&init_mm, (unsigned long)area->addr,
2451 count * PAGE_SIZE, vmap_pfn_apply, &data)) {
2452 free_vm_area(area);
2453 return NULL;
2454 }
2455 return area->addr;
2456}
2457EXPORT_SYMBOL_GPL(vmap_pfn);
2458#endif /* CONFIG_VMAP_PFN */
2459
Adrian Bunke31d9eb2008-02-04 22:29:09 -08002460static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask,
Wanpeng Li3722e132013-11-12 15:07:29 -08002461 pgprot_t prot, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002462{
David Rientjes930f0362014-08-06 16:06:28 -07002463 const gfp_t nested_gfp = (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO;
Christoph Hellwigf2559352020-10-17 16:15:43 -07002464 unsigned int nr_pages = get_vm_area_size(area) >> PAGE_SHIFT;
2465 unsigned int array_size = nr_pages * sizeof(struct page *), i;
2466 struct page **pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002467
Christoph Hellwigf2559352020-10-17 16:15:43 -07002468 gfp_mask |= __GFP_NOWARN;
2469 if (!(gfp_mask & (GFP_DMA | GFP_DMA32)))
2470 gfp_mask |= __GFP_HIGHMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002471
Linus Torvalds1da177e2005-04-16 15:20:36 -07002472 /* Please note that the recursion is strictly bounded. */
Jan Kiszka8757d5f2006-07-14 00:23:56 -07002473 if (array_size > PAGE_SIZE) {
Christoph Hellwigf2559352020-10-17 16:15:43 -07002474 pages = __vmalloc_node(array_size, 1, nested_gfp, node,
2475 area->caller);
Andrew Morton286e1ea2006-10-17 00:09:57 -07002476 } else {
Jan Beulich976d6df2009-12-14 17:58:39 -08002477 pages = kmalloc_node(array_size, nested_gfp, node);
Andrew Morton286e1ea2006-10-17 00:09:57 -07002478 }
Austin Kim7ea362422019-09-23 15:36:42 -07002479
2480 if (!pages) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002481 remove_vm_area(area->addr);
2482 kfree(area);
2483 return NULL;
2484 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002485
Austin Kim7ea362422019-09-23 15:36:42 -07002486 area->pages = pages;
2487 area->nr_pages = nr_pages;
2488
Linus Torvalds1da177e2005-04-16 15:20:36 -07002489 for (i = 0; i < area->nr_pages; i++) {
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08002490 struct page *page;
2491
Jianguo Wu4b909512013-11-12 15:07:11 -08002492 if (node == NUMA_NO_NODE)
Christoph Hellwigf2559352020-10-17 16:15:43 -07002493 page = alloc_page(gfp_mask);
Christoph Lameter930fc452005-10-29 18:15:41 -07002494 else
Christoph Hellwigf2559352020-10-17 16:15:43 -07002495 page = alloc_pages_node(node, gfp_mask, 0);
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08002496
2497 if (unlikely(!page)) {
Hui Su82afbc32020-10-13 16:54:48 -07002498 /* Successfully allocated i pages, free them in __vfree() */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002499 area->nr_pages = i;
Roman Gushchin97105f02019-07-11 21:00:13 -07002500 atomic_long_add(area->nr_pages, &nr_vmalloc_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002501 goto fail;
2502 }
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08002503 area->pages[i] = page;
Liu Xiangdcf61ff2019-11-30 17:54:30 -08002504 if (gfpflags_allow_blocking(gfp_mask))
Eric Dumazet660654f2014-08-06 16:06:25 -07002505 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002506 }
Roman Gushchin97105f02019-07-11 21:00:13 -07002507 atomic_long_add(area->nr_pages, &nr_vmalloc_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002508
Christoph Hellwiged1f3242020-06-01 21:51:19 -07002509 if (map_kernel_range((unsigned long)area->addr, get_vm_area_size(area),
2510 prot, pages) < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002511 goto fail;
Christoph Hellwiged1f3242020-06-01 21:51:19 -07002512
Linus Torvalds1da177e2005-04-16 15:20:36 -07002513 return area->addr;
2514
2515fail:
Michal Hockoa8e99252017-02-22 15:46:10 -08002516 warn_alloc(gfp_mask, NULL,
Michal Hocko7877cdc2016-10-07 17:01:55 -07002517 "vmalloc: allocation failure, allocated %ld of %ld bytes",
Dave Hansen22943ab2011-05-24 17:12:18 -07002518 (area->nr_pages*PAGE_SIZE), area->size);
Roman Penyaevc67dc622019-03-05 15:43:24 -08002519 __vfree(area->addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002520 return NULL;
2521}
2522
David Rientjesd0a21262011-01-13 15:46:02 -08002523/**
Mike Rapoport92eac162019-03-05 15:48:36 -08002524 * __vmalloc_node_range - allocate virtually contiguous memory
2525 * @size: allocation size
2526 * @align: desired alignment
2527 * @start: vm area range start
2528 * @end: vm area range end
2529 * @gfp_mask: flags for the page level allocator
2530 * @prot: protection mask for the allocated pages
2531 * @vm_flags: additional vm area flags (e.g. %VM_NO_GUARD)
2532 * @node: node to use for allocation or NUMA_NO_NODE
2533 * @caller: caller's return address
David Rientjesd0a21262011-01-13 15:46:02 -08002534 *
Mike Rapoport92eac162019-03-05 15:48:36 -08002535 * Allocate enough pages to cover @size from the page level
2536 * allocator with @gfp_mask flags. Map them into contiguous
2537 * kernel virtual space, using a pagetable protection of @prot.
Mike Rapoporta862f682019-03-05 15:48:42 -08002538 *
2539 * Return: the address of the area or %NULL on failure
David Rientjesd0a21262011-01-13 15:46:02 -08002540 */
2541void *__vmalloc_node_range(unsigned long size, unsigned long align,
2542 unsigned long start, unsigned long end, gfp_t gfp_mask,
Andrey Ryabinincb9e3c22015-02-13 14:40:07 -08002543 pgprot_t prot, unsigned long vm_flags, int node,
2544 const void *caller)
Christoph Lameter930fc452005-10-29 18:15:41 -07002545{
David Rientjesd0a21262011-01-13 15:46:02 -08002546 struct vm_struct *area;
2547 void *addr;
2548 unsigned long real_size = size;
2549
2550 size = PAGE_ALIGN(size);
Arun KSca79b0c2018-12-28 00:34:29 -08002551 if (!size || (size >> PAGE_SHIFT) > totalram_pages())
Joe Perchesde7d2b52011-10-31 17:08:48 -07002552 goto fail;
David Rientjesd0a21262011-01-13 15:46:02 -08002553
Andrey Ryabinind98c9e82019-12-17 20:51:38 -08002554 area = __get_vm_area_node(real_size, align, VM_ALLOC | VM_UNINITIALIZED |
Andrey Ryabinincb9e3c22015-02-13 14:40:07 -08002555 vm_flags, start, end, node, gfp_mask, caller);
David Rientjesd0a21262011-01-13 15:46:02 -08002556 if (!area)
Joe Perchesde7d2b52011-10-31 17:08:48 -07002557 goto fail;
David Rientjesd0a21262011-01-13 15:46:02 -08002558
Wanpeng Li3722e132013-11-12 15:07:29 -08002559 addr = __vmalloc_area_node(area, gfp_mask, prot, node);
Mel Gorman1368edf2011-12-08 14:34:30 -08002560 if (!addr)
Wanpeng Lib82225f32013-11-12 15:07:33 -08002561 return NULL;
Catalin Marinas89219d32009-06-11 13:23:19 +01002562
2563 /*
Zhang Yanfei20fc02b2013-07-08 15:59:58 -07002564 * In this function, newly allocated vm_struct has VM_UNINITIALIZED
2565 * flag. It means that vm_struct is not fully initialized.
Joonsoo Kim4341fa42013-04-29 15:07:39 -07002566 * Now, it is fully initialized, so remove this flag here.
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07002567 */
Zhang Yanfei20fc02b2013-07-08 15:59:58 -07002568 clear_vm_uninitialized_flag(area);
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07002569
Catalin Marinas94f4a162017-07-06 15:40:22 -07002570 kmemleak_vmalloc(area, size, gfp_mask);
Catalin Marinas89219d32009-06-11 13:23:19 +01002571
2572 return addr;
Joe Perchesde7d2b52011-10-31 17:08:48 -07002573
2574fail:
Michal Hockoa8e99252017-02-22 15:46:10 -08002575 warn_alloc(gfp_mask, NULL,
Michal Hocko7877cdc2016-10-07 17:01:55 -07002576 "vmalloc: allocation failure: %lu bytes", real_size);
Joe Perchesde7d2b52011-10-31 17:08:48 -07002577 return NULL;
Christoph Lameter930fc452005-10-29 18:15:41 -07002578}
2579
Linus Torvalds1da177e2005-04-16 15:20:36 -07002580/**
Mike Rapoport92eac162019-03-05 15:48:36 -08002581 * __vmalloc_node - allocate virtually contiguous memory
2582 * @size: allocation size
2583 * @align: desired alignment
2584 * @gfp_mask: flags for the page level allocator
Mike Rapoport92eac162019-03-05 15:48:36 -08002585 * @node: node to use for allocation or NUMA_NO_NODE
2586 * @caller: caller's return address
Linus Torvalds1da177e2005-04-16 15:20:36 -07002587 *
Christoph Hellwigf38fcb92020-06-01 21:51:45 -07002588 * Allocate enough pages to cover @size from the page level allocator with
2589 * @gfp_mask flags. Map them into contiguous kernel virtual space.
Michal Hockoa7c3e902017-05-08 15:57:09 -07002590 *
Mike Rapoport92eac162019-03-05 15:48:36 -08002591 * Reclaim modifiers in @gfp_mask - __GFP_NORETRY, __GFP_RETRY_MAYFAIL
2592 * and __GFP_NOFAIL are not supported
Michal Hockoa7c3e902017-05-08 15:57:09 -07002593 *
Mike Rapoport92eac162019-03-05 15:48:36 -08002594 * Any use of gfp flags outside of GFP_KERNEL should be consulted
2595 * with mm people.
Mike Rapoporta862f682019-03-05 15:48:42 -08002596 *
2597 * Return: pointer to the allocated memory or %NULL on error
Linus Torvalds1da177e2005-04-16 15:20:36 -07002598 */
Christoph Hellwig2b905942020-06-01 21:51:53 -07002599void *__vmalloc_node(unsigned long size, unsigned long align,
Christoph Hellwigf38fcb92020-06-01 21:51:45 -07002600 gfp_t gfp_mask, int node, const void *caller)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002601{
David Rientjesd0a21262011-01-13 15:46:02 -08002602 return __vmalloc_node_range(size, align, VMALLOC_START, VMALLOC_END,
Christoph Hellwigf38fcb92020-06-01 21:51:45 -07002603 gfp_mask, PAGE_KERNEL, 0, node, caller);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002604}
Christoph Hellwigc3f896d2020-06-01 21:51:57 -07002605/*
2606 * This is only for performance analysis of vmalloc and stress purpose.
2607 * It is required by vmalloc test module, therefore do not use it other
2608 * than that.
2609 */
2610#ifdef CONFIG_TEST_VMALLOC_MODULE
2611EXPORT_SYMBOL_GPL(__vmalloc_node);
2612#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002613
Christoph Hellwig88dca4c2020-06-01 21:51:40 -07002614void *__vmalloc(unsigned long size, gfp_t gfp_mask)
Christoph Lameter930fc452005-10-29 18:15:41 -07002615{
Christoph Hellwigf38fcb92020-06-01 21:51:45 -07002616 return __vmalloc_node(size, 1, gfp_mask, NUMA_NO_NODE,
Christoph Lameter23016962008-04-28 02:12:42 -07002617 __builtin_return_address(0));
Christoph Lameter930fc452005-10-29 18:15:41 -07002618}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002619EXPORT_SYMBOL(__vmalloc);
2620
2621/**
Mike Rapoport92eac162019-03-05 15:48:36 -08002622 * vmalloc - allocate virtually contiguous memory
2623 * @size: allocation size
Linus Torvalds1da177e2005-04-16 15:20:36 -07002624 *
Mike Rapoport92eac162019-03-05 15:48:36 -08002625 * Allocate enough pages to cover @size from the page level
2626 * allocator and map them into contiguous kernel virtual space.
2627 *
2628 * For tight control over page level allocator and protection flags
2629 * use __vmalloc() instead.
Mike Rapoporta862f682019-03-05 15:48:42 -08002630 *
2631 * Return: pointer to the allocated memory or %NULL on error
Linus Torvalds1da177e2005-04-16 15:20:36 -07002632 */
2633void *vmalloc(unsigned long size)
2634{
Christoph Hellwig4d39d722020-06-01 21:51:49 -07002635 return __vmalloc_node(size, 1, GFP_KERNEL, NUMA_NO_NODE,
2636 __builtin_return_address(0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002637}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002638EXPORT_SYMBOL(vmalloc);
2639
Christoph Lameter930fc452005-10-29 18:15:41 -07002640/**
Mike Rapoport92eac162019-03-05 15:48:36 -08002641 * vzalloc - allocate virtually contiguous memory with zero fill
2642 * @size: allocation size
Dave Younge1ca7782010-10-26 14:22:06 -07002643 *
Mike Rapoport92eac162019-03-05 15:48:36 -08002644 * Allocate enough pages to cover @size from the page level
2645 * allocator and map them into contiguous kernel virtual space.
2646 * The memory allocated is set to zero.
2647 *
2648 * For tight control over page level allocator and protection flags
2649 * use __vmalloc() instead.
Mike Rapoporta862f682019-03-05 15:48:42 -08002650 *
2651 * Return: pointer to the allocated memory or %NULL on error
Dave Younge1ca7782010-10-26 14:22:06 -07002652 */
2653void *vzalloc(unsigned long size)
2654{
Christoph Hellwig4d39d722020-06-01 21:51:49 -07002655 return __vmalloc_node(size, 1, GFP_KERNEL | __GFP_ZERO, NUMA_NO_NODE,
2656 __builtin_return_address(0));
Dave Younge1ca7782010-10-26 14:22:06 -07002657}
2658EXPORT_SYMBOL(vzalloc);
2659
2660/**
Rolf Eike Beeread04082006-09-27 01:50:13 -07002661 * vmalloc_user - allocate zeroed virtually contiguous memory for userspace
2662 * @size: allocation size
Nick Piggin83342312006-06-23 02:03:20 -07002663 *
Rolf Eike Beeread04082006-09-27 01:50:13 -07002664 * The resulting memory area is zeroed so it can be mapped to userspace
2665 * without leaking data.
Mike Rapoporta862f682019-03-05 15:48:42 -08002666 *
2667 * Return: pointer to the allocated memory or %NULL on error
Nick Piggin83342312006-06-23 02:03:20 -07002668 */
2669void *vmalloc_user(unsigned long size)
2670{
Roman Penyaevbc84c532019-03-05 15:43:27 -08002671 return __vmalloc_node_range(size, SHMLBA, VMALLOC_START, VMALLOC_END,
2672 GFP_KERNEL | __GFP_ZERO, PAGE_KERNEL,
2673 VM_USERMAP, NUMA_NO_NODE,
2674 __builtin_return_address(0));
Nick Piggin83342312006-06-23 02:03:20 -07002675}
2676EXPORT_SYMBOL(vmalloc_user);
2677
2678/**
Mike Rapoport92eac162019-03-05 15:48:36 -08002679 * vmalloc_node - allocate memory on a specific node
2680 * @size: allocation size
2681 * @node: numa node
Christoph Lameter930fc452005-10-29 18:15:41 -07002682 *
Mike Rapoport92eac162019-03-05 15:48:36 -08002683 * Allocate enough pages to cover @size from the page level
2684 * allocator and map them into contiguous kernel virtual space.
Christoph Lameter930fc452005-10-29 18:15:41 -07002685 *
Mike Rapoport92eac162019-03-05 15:48:36 -08002686 * For tight control over page level allocator and protection flags
2687 * use __vmalloc() instead.
Mike Rapoporta862f682019-03-05 15:48:42 -08002688 *
2689 * Return: pointer to the allocated memory or %NULL on error
Christoph Lameter930fc452005-10-29 18:15:41 -07002690 */
2691void *vmalloc_node(unsigned long size, int node)
2692{
Christoph Hellwigf38fcb92020-06-01 21:51:45 -07002693 return __vmalloc_node(size, 1, GFP_KERNEL, node,
2694 __builtin_return_address(0));
Christoph Lameter930fc452005-10-29 18:15:41 -07002695}
2696EXPORT_SYMBOL(vmalloc_node);
2697
Dave Younge1ca7782010-10-26 14:22:06 -07002698/**
2699 * vzalloc_node - allocate memory on a specific node with zero fill
2700 * @size: allocation size
2701 * @node: numa node
2702 *
2703 * Allocate enough pages to cover @size from the page level
2704 * allocator and map them into contiguous kernel virtual space.
2705 * The memory allocated is set to zero.
2706 *
Mike Rapoporta862f682019-03-05 15:48:42 -08002707 * Return: pointer to the allocated memory or %NULL on error
Dave Younge1ca7782010-10-26 14:22:06 -07002708 */
2709void *vzalloc_node(unsigned long size, int node)
2710{
Christoph Hellwig4d39d722020-06-01 21:51:49 -07002711 return __vmalloc_node(size, 1, GFP_KERNEL | __GFP_ZERO, node,
2712 __builtin_return_address(0));
Dave Younge1ca7782010-10-26 14:22:06 -07002713}
2714EXPORT_SYMBOL(vzalloc_node);
2715
Andi Kleen0d08e0d2007-05-02 19:27:12 +02002716#if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32)
Michal Hocko698d0832018-02-21 14:46:01 -08002717#define GFP_VMALLOC32 (GFP_DMA32 | GFP_KERNEL)
Andi Kleen0d08e0d2007-05-02 19:27:12 +02002718#elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA)
Michal Hocko698d0832018-02-21 14:46:01 -08002719#define GFP_VMALLOC32 (GFP_DMA | GFP_KERNEL)
Andi Kleen0d08e0d2007-05-02 19:27:12 +02002720#else
Michal Hocko698d0832018-02-21 14:46:01 -08002721/*
2722 * 64b systems should always have either DMA or DMA32 zones. For others
2723 * GFP_DMA32 should do the right thing and use the normal zone.
2724 */
2725#define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL
Andi Kleen0d08e0d2007-05-02 19:27:12 +02002726#endif
2727
Linus Torvalds1da177e2005-04-16 15:20:36 -07002728/**
Mike Rapoport92eac162019-03-05 15:48:36 -08002729 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
2730 * @size: allocation size
Linus Torvalds1da177e2005-04-16 15:20:36 -07002731 *
Mike Rapoport92eac162019-03-05 15:48:36 -08002732 * Allocate enough 32bit PA addressable pages to cover @size from the
2733 * page level allocator and map them into contiguous kernel virtual space.
Mike Rapoporta862f682019-03-05 15:48:42 -08002734 *
2735 * Return: pointer to the allocated memory or %NULL on error
Linus Torvalds1da177e2005-04-16 15:20:36 -07002736 */
2737void *vmalloc_32(unsigned long size)
2738{
Christoph Hellwigf38fcb92020-06-01 21:51:45 -07002739 return __vmalloc_node(size, 1, GFP_VMALLOC32, NUMA_NO_NODE,
2740 __builtin_return_address(0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002741}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002742EXPORT_SYMBOL(vmalloc_32);
2743
Nick Piggin83342312006-06-23 02:03:20 -07002744/**
Rolf Eike Beeread04082006-09-27 01:50:13 -07002745 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
Mike Rapoport92eac162019-03-05 15:48:36 -08002746 * @size: allocation size
Rolf Eike Beeread04082006-09-27 01:50:13 -07002747 *
2748 * The resulting memory area is 32bit addressable and zeroed so it can be
2749 * mapped to userspace without leaking data.
Mike Rapoporta862f682019-03-05 15:48:42 -08002750 *
2751 * Return: pointer to the allocated memory or %NULL on error
Nick Piggin83342312006-06-23 02:03:20 -07002752 */
2753void *vmalloc_32_user(unsigned long size)
2754{
Roman Penyaevbc84c532019-03-05 15:43:27 -08002755 return __vmalloc_node_range(size, SHMLBA, VMALLOC_START, VMALLOC_END,
2756 GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL,
2757 VM_USERMAP, NUMA_NO_NODE,
2758 __builtin_return_address(0));
Nick Piggin83342312006-06-23 02:03:20 -07002759}
2760EXPORT_SYMBOL(vmalloc_32_user);
2761
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002762/*
2763 * small helper routine , copy contents to buf from addr.
2764 * If the page is not present, fill zero.
2765 */
2766
2767static int aligned_vread(char *buf, char *addr, unsigned long count)
2768{
2769 struct page *p;
2770 int copied = 0;
2771
2772 while (count) {
2773 unsigned long offset, length;
2774
Alexander Kuleshov891c49a2015-11-05 18:46:51 -08002775 offset = offset_in_page(addr);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002776 length = PAGE_SIZE - offset;
2777 if (length > count)
2778 length = count;
2779 p = vmalloc_to_page(addr);
2780 /*
2781 * To do safe access to this _mapped_ area, we need
2782 * lock. But adding lock here means that we need to add
2783 * overhead of vmalloc()/vfree() calles for this _debug_
2784 * interface, rarely used. Instead of that, we'll use
2785 * kmap() and get small overhead in this access function.
2786 */
2787 if (p) {
2788 /*
2789 * we can expect USER0 is not used (see vread/vwrite's
2790 * function description)
2791 */
Cong Wang9b04c5f2011-11-25 23:14:39 +08002792 void *map = kmap_atomic(p);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002793 memcpy(buf, map + offset, length);
Cong Wang9b04c5f2011-11-25 23:14:39 +08002794 kunmap_atomic(map);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002795 } else
2796 memset(buf, 0, length);
2797
2798 addr += length;
2799 buf += length;
2800 copied += length;
2801 count -= length;
2802 }
2803 return copied;
2804}
2805
2806static int aligned_vwrite(char *buf, char *addr, unsigned long count)
2807{
2808 struct page *p;
2809 int copied = 0;
2810
2811 while (count) {
2812 unsigned long offset, length;
2813
Alexander Kuleshov891c49a2015-11-05 18:46:51 -08002814 offset = offset_in_page(addr);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002815 length = PAGE_SIZE - offset;
2816 if (length > count)
2817 length = count;
2818 p = vmalloc_to_page(addr);
2819 /*
2820 * To do safe access to this _mapped_ area, we need
2821 * lock. But adding lock here means that we need to add
2822 * overhead of vmalloc()/vfree() calles for this _debug_
2823 * interface, rarely used. Instead of that, we'll use
2824 * kmap() and get small overhead in this access function.
2825 */
2826 if (p) {
2827 /*
2828 * we can expect USER0 is not used (see vread/vwrite's
2829 * function description)
2830 */
Cong Wang9b04c5f2011-11-25 23:14:39 +08002831 void *map = kmap_atomic(p);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002832 memcpy(map + offset, buf, length);
Cong Wang9b04c5f2011-11-25 23:14:39 +08002833 kunmap_atomic(map);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002834 }
2835 addr += length;
2836 buf += length;
2837 copied += length;
2838 count -= length;
2839 }
2840 return copied;
2841}
2842
2843/**
Mike Rapoport92eac162019-03-05 15:48:36 -08002844 * vread() - read vmalloc area in a safe way.
2845 * @buf: buffer for reading data
2846 * @addr: vm address.
2847 * @count: number of bytes to be read.
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002848 *
Mike Rapoport92eac162019-03-05 15:48:36 -08002849 * This function checks that addr is a valid vmalloc'ed area, and
2850 * copy data from that area to a given buffer. If the given memory range
2851 * of [addr...addr+count) includes some valid address, data is copied to
2852 * proper area of @buf. If there are memory holes, they'll be zero-filled.
2853 * IOREMAP area is treated as memory hole and no copy is done.
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002854 *
Mike Rapoport92eac162019-03-05 15:48:36 -08002855 * If [addr...addr+count) doesn't includes any intersects with alive
2856 * vm_struct area, returns 0. @buf should be kernel's buffer.
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002857 *
Mike Rapoport92eac162019-03-05 15:48:36 -08002858 * Note: In usual ops, vread() is never necessary because the caller
2859 * should know vmalloc() area is valid and can use memcpy().
2860 * This is for routines which have to access vmalloc area without
Geert Uytterhoevend9009d62019-07-11 20:59:06 -07002861 * any information, as /dev/kmem.
Mike Rapoporta862f682019-03-05 15:48:42 -08002862 *
2863 * Return: number of bytes for which addr and buf should be increased
2864 * (same number as @count) or %0 if [addr...addr+count) doesn't
2865 * include any intersection with valid vmalloc area
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002866 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002867long vread(char *buf, char *addr, unsigned long count)
2868{
Joonsoo Kime81ce852013-04-29 15:07:32 -07002869 struct vmap_area *va;
2870 struct vm_struct *vm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002871 char *vaddr, *buf_start = buf;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002872 unsigned long buflen = count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002873 unsigned long n;
2874
2875 /* Don't allow overflow */
2876 if ((unsigned long) addr + count < count)
2877 count = -(unsigned long) addr;
2878
Joonsoo Kime81ce852013-04-29 15:07:32 -07002879 spin_lock(&vmap_area_lock);
2880 list_for_each_entry(va, &vmap_area_list, list) {
2881 if (!count)
2882 break;
2883
Pengfei Li688fcbf2019-09-23 15:36:39 -07002884 if (!va->vm)
Joonsoo Kime81ce852013-04-29 15:07:32 -07002885 continue;
2886
2887 vm = va->vm;
2888 vaddr = (char *) vm->addr;
Wanpeng Li762216a2013-09-11 14:22:42 -07002889 if (addr >= vaddr + get_vm_area_size(vm))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002890 continue;
2891 while (addr < vaddr) {
2892 if (count == 0)
2893 goto finished;
2894 *buf = '\0';
2895 buf++;
2896 addr++;
2897 count--;
2898 }
Wanpeng Li762216a2013-09-11 14:22:42 -07002899 n = vaddr + get_vm_area_size(vm) - addr;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002900 if (n > count)
2901 n = count;
Joonsoo Kime81ce852013-04-29 15:07:32 -07002902 if (!(vm->flags & VM_IOREMAP))
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002903 aligned_vread(buf, addr, n);
2904 else /* IOREMAP area is treated as memory hole */
2905 memset(buf, 0, n);
2906 buf += n;
2907 addr += n;
2908 count -= n;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002909 }
2910finished:
Joonsoo Kime81ce852013-04-29 15:07:32 -07002911 spin_unlock(&vmap_area_lock);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002912
2913 if (buf == buf_start)
2914 return 0;
2915 /* zero-fill memory holes */
2916 if (buf != buf_start + buflen)
2917 memset(buf, 0, buflen - (buf - buf_start));
2918
2919 return buflen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002920}
2921
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002922/**
Mike Rapoport92eac162019-03-05 15:48:36 -08002923 * vwrite() - write vmalloc area in a safe way.
2924 * @buf: buffer for source data
2925 * @addr: vm address.
2926 * @count: number of bytes to be read.
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002927 *
Mike Rapoport92eac162019-03-05 15:48:36 -08002928 * This function checks that addr is a valid vmalloc'ed area, and
2929 * copy data from a buffer to the given addr. If specified range of
2930 * [addr...addr+count) includes some valid address, data is copied from
2931 * proper area of @buf. If there are memory holes, no copy to hole.
2932 * IOREMAP area is treated as memory hole and no copy is done.
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002933 *
Mike Rapoport92eac162019-03-05 15:48:36 -08002934 * If [addr...addr+count) doesn't includes any intersects with alive
2935 * vm_struct area, returns 0. @buf should be kernel's buffer.
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002936 *
Mike Rapoport92eac162019-03-05 15:48:36 -08002937 * Note: In usual ops, vwrite() is never necessary because the caller
2938 * should know vmalloc() area is valid and can use memcpy().
2939 * This is for routines which have to access vmalloc area without
Geert Uytterhoevend9009d62019-07-11 20:59:06 -07002940 * any information, as /dev/kmem.
Mike Rapoporta862f682019-03-05 15:48:42 -08002941 *
2942 * Return: number of bytes for which addr and buf should be
2943 * increased (same number as @count) or %0 if [addr...addr+count)
2944 * doesn't include any intersection with valid vmalloc area
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002945 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002946long vwrite(char *buf, char *addr, unsigned long count)
2947{
Joonsoo Kime81ce852013-04-29 15:07:32 -07002948 struct vmap_area *va;
2949 struct vm_struct *vm;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002950 char *vaddr;
2951 unsigned long n, buflen;
2952 int copied = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002953
2954 /* Don't allow overflow */
2955 if ((unsigned long) addr + count < count)
2956 count = -(unsigned long) addr;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002957 buflen = count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002958
Joonsoo Kime81ce852013-04-29 15:07:32 -07002959 spin_lock(&vmap_area_lock);
2960 list_for_each_entry(va, &vmap_area_list, list) {
2961 if (!count)
2962 break;
2963
Pengfei Li688fcbf2019-09-23 15:36:39 -07002964 if (!va->vm)
Joonsoo Kime81ce852013-04-29 15:07:32 -07002965 continue;
2966
2967 vm = va->vm;
2968 vaddr = (char *) vm->addr;
Wanpeng Li762216a2013-09-11 14:22:42 -07002969 if (addr >= vaddr + get_vm_area_size(vm))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002970 continue;
2971 while (addr < vaddr) {
2972 if (count == 0)
2973 goto finished;
2974 buf++;
2975 addr++;
2976 count--;
2977 }
Wanpeng Li762216a2013-09-11 14:22:42 -07002978 n = vaddr + get_vm_area_size(vm) - addr;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002979 if (n > count)
2980 n = count;
Joonsoo Kime81ce852013-04-29 15:07:32 -07002981 if (!(vm->flags & VM_IOREMAP)) {
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002982 aligned_vwrite(buf, addr, n);
2983 copied++;
2984 }
2985 buf += n;
2986 addr += n;
2987 count -= n;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002988 }
2989finished:
Joonsoo Kime81ce852013-04-29 15:07:32 -07002990 spin_unlock(&vmap_area_lock);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002991 if (!copied)
2992 return 0;
2993 return buflen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002994}
Nick Piggin83342312006-06-23 02:03:20 -07002995
2996/**
Mike Rapoport92eac162019-03-05 15:48:36 -08002997 * remap_vmalloc_range_partial - map vmalloc pages to userspace
2998 * @vma: vma to cover
2999 * @uaddr: target user address to start at
3000 * @kaddr: virtual address of vmalloc kernel memory
Jann Hornbdebd6a22020-04-20 18:14:11 -07003001 * @pgoff: offset from @kaddr to start at
Mike Rapoport92eac162019-03-05 15:48:36 -08003002 * @size: size of map area
HATAYAMA Daisukee69e9d4a2013-07-03 15:02:18 -07003003 *
Mike Rapoport92eac162019-03-05 15:48:36 -08003004 * Returns: 0 for success, -Exxx on failure
HATAYAMA Daisukee69e9d4a2013-07-03 15:02:18 -07003005 *
Mike Rapoport92eac162019-03-05 15:48:36 -08003006 * This function checks that @kaddr is a valid vmalloc'ed area,
3007 * and that it is big enough to cover the range starting at
3008 * @uaddr in @vma. Will return failure if that criteria isn't
3009 * met.
HATAYAMA Daisukee69e9d4a2013-07-03 15:02:18 -07003010 *
Mike Rapoport92eac162019-03-05 15:48:36 -08003011 * Similar to remap_pfn_range() (see mm/memory.c)
HATAYAMA Daisukee69e9d4a2013-07-03 15:02:18 -07003012 */
3013int remap_vmalloc_range_partial(struct vm_area_struct *vma, unsigned long uaddr,
Jann Hornbdebd6a22020-04-20 18:14:11 -07003014 void *kaddr, unsigned long pgoff,
3015 unsigned long size)
HATAYAMA Daisukee69e9d4a2013-07-03 15:02:18 -07003016{
3017 struct vm_struct *area;
Jann Hornbdebd6a22020-04-20 18:14:11 -07003018 unsigned long off;
3019 unsigned long end_index;
3020
3021 if (check_shl_overflow(pgoff, PAGE_SHIFT, &off))
3022 return -EINVAL;
HATAYAMA Daisukee69e9d4a2013-07-03 15:02:18 -07003023
3024 size = PAGE_ALIGN(size);
3025
3026 if (!PAGE_ALIGNED(uaddr) || !PAGE_ALIGNED(kaddr))
3027 return -EINVAL;
3028
3029 area = find_vm_area(kaddr);
3030 if (!area)
3031 return -EINVAL;
3032
Christoph Hellwigfe9041c2019-06-03 08:55:13 +02003033 if (!(area->flags & (VM_USERMAP | VM_DMA_COHERENT)))
HATAYAMA Daisukee69e9d4a2013-07-03 15:02:18 -07003034 return -EINVAL;
3035
Jann Hornbdebd6a22020-04-20 18:14:11 -07003036 if (check_add_overflow(size, off, &end_index) ||
3037 end_index > get_vm_area_size(area))
HATAYAMA Daisukee69e9d4a2013-07-03 15:02:18 -07003038 return -EINVAL;
Jann Hornbdebd6a22020-04-20 18:14:11 -07003039 kaddr += off;
HATAYAMA Daisukee69e9d4a2013-07-03 15:02:18 -07003040
3041 do {
3042 struct page *page = vmalloc_to_page(kaddr);
3043 int ret;
3044
3045 ret = vm_insert_page(vma, uaddr, page);
3046 if (ret)
3047 return ret;
3048
3049 uaddr += PAGE_SIZE;
3050 kaddr += PAGE_SIZE;
3051 size -= PAGE_SIZE;
3052 } while (size > 0);
3053
3054 vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
3055
3056 return 0;
3057}
3058EXPORT_SYMBOL(remap_vmalloc_range_partial);
3059
3060/**
Mike Rapoport92eac162019-03-05 15:48:36 -08003061 * remap_vmalloc_range - map vmalloc pages to userspace
3062 * @vma: vma to cover (map full range of vma)
3063 * @addr: vmalloc memory
3064 * @pgoff: number of pages into addr before first page to map
Randy Dunlap76824862008-03-19 17:00:40 -07003065 *
Mike Rapoport92eac162019-03-05 15:48:36 -08003066 * Returns: 0 for success, -Exxx on failure
Nick Piggin83342312006-06-23 02:03:20 -07003067 *
Mike Rapoport92eac162019-03-05 15:48:36 -08003068 * This function checks that addr is a valid vmalloc'ed area, and
3069 * that it is big enough to cover the vma. Will return failure if
3070 * that criteria isn't met.
Nick Piggin83342312006-06-23 02:03:20 -07003071 *
Mike Rapoport92eac162019-03-05 15:48:36 -08003072 * Similar to remap_pfn_range() (see mm/memory.c)
Nick Piggin83342312006-06-23 02:03:20 -07003073 */
3074int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
3075 unsigned long pgoff)
3076{
HATAYAMA Daisukee69e9d4a2013-07-03 15:02:18 -07003077 return remap_vmalloc_range_partial(vma, vma->vm_start,
Jann Hornbdebd6a22020-04-20 18:14:11 -07003078 addr, pgoff,
HATAYAMA Daisukee69e9d4a2013-07-03 15:02:18 -07003079 vma->vm_end - vma->vm_start);
Nick Piggin83342312006-06-23 02:03:20 -07003080}
3081EXPORT_SYMBOL(remap_vmalloc_range);
3082
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07003083void free_vm_area(struct vm_struct *area)
3084{
3085 struct vm_struct *ret;
3086 ret = remove_vm_area(area->addr);
3087 BUG_ON(ret != area);
3088 kfree(area);
3089}
3090EXPORT_SYMBOL_GPL(free_vm_area);
Christoph Lametera10aa572008-04-28 02:12:40 -07003091
Tejun Heo4f8b02b2010-09-03 18:22:47 +02003092#ifdef CONFIG_SMP
Tejun Heoca23e402009-08-14 15:00:52 +09003093static struct vmap_area *node_to_va(struct rb_node *n)
3094{
Geliang Tang4583e772017-02-22 15:41:54 -08003095 return rb_entry_safe(n, struct vmap_area, rb_node);
Tejun Heoca23e402009-08-14 15:00:52 +09003096}
3097
3098/**
Uladzislau Rezki (Sony)68ad4a32019-05-17 14:31:31 -07003099 * pvm_find_va_enclose_addr - find the vmap_area @addr belongs to
3100 * @addr: target address
Tejun Heoca23e402009-08-14 15:00:52 +09003101 *
Uladzislau Rezki (Sony)68ad4a32019-05-17 14:31:31 -07003102 * Returns: vmap_area if it is found. If there is no such area
3103 * the first highest(reverse order) vmap_area is returned
3104 * i.e. va->va_start < addr && va->va_end < addr or NULL
3105 * if there are no any areas before @addr.
Tejun Heoca23e402009-08-14 15:00:52 +09003106 */
Uladzislau Rezki (Sony)68ad4a32019-05-17 14:31:31 -07003107static struct vmap_area *
3108pvm_find_va_enclose_addr(unsigned long addr)
Tejun Heoca23e402009-08-14 15:00:52 +09003109{
Uladzislau Rezki (Sony)68ad4a32019-05-17 14:31:31 -07003110 struct vmap_area *va, *tmp;
3111 struct rb_node *n;
3112
3113 n = free_vmap_area_root.rb_node;
3114 va = NULL;
Tejun Heoca23e402009-08-14 15:00:52 +09003115
3116 while (n) {
Uladzislau Rezki (Sony)68ad4a32019-05-17 14:31:31 -07003117 tmp = rb_entry(n, struct vmap_area, rb_node);
3118 if (tmp->va_start <= addr) {
3119 va = tmp;
3120 if (tmp->va_end >= addr)
3121 break;
3122
Tejun Heoca23e402009-08-14 15:00:52 +09003123 n = n->rb_right;
Uladzislau Rezki (Sony)68ad4a32019-05-17 14:31:31 -07003124 } else {
3125 n = n->rb_left;
3126 }
Tejun Heoca23e402009-08-14 15:00:52 +09003127 }
3128
Uladzislau Rezki (Sony)68ad4a32019-05-17 14:31:31 -07003129 return va;
Tejun Heoca23e402009-08-14 15:00:52 +09003130}
3131
3132/**
Uladzislau Rezki (Sony)68ad4a32019-05-17 14:31:31 -07003133 * pvm_determine_end_from_reverse - find the highest aligned address
3134 * of free block below VMALLOC_END
3135 * @va:
3136 * in - the VA we start the search(reverse order);
3137 * out - the VA with the highest aligned end address.
Tejun Heoca23e402009-08-14 15:00:52 +09003138 *
Uladzislau Rezki (Sony)68ad4a32019-05-17 14:31:31 -07003139 * Returns: determined end address within vmap_area
Tejun Heoca23e402009-08-14 15:00:52 +09003140 */
Uladzislau Rezki (Sony)68ad4a32019-05-17 14:31:31 -07003141static unsigned long
3142pvm_determine_end_from_reverse(struct vmap_area **va, unsigned long align)
Tejun Heoca23e402009-08-14 15:00:52 +09003143{
Uladzislau Rezki (Sony)68ad4a32019-05-17 14:31:31 -07003144 unsigned long vmalloc_end = VMALLOC_END & ~(align - 1);
Tejun Heoca23e402009-08-14 15:00:52 +09003145 unsigned long addr;
3146
Uladzislau Rezki (Sony)68ad4a32019-05-17 14:31:31 -07003147 if (likely(*va)) {
3148 list_for_each_entry_from_reverse((*va),
3149 &free_vmap_area_list, list) {
3150 addr = min((*va)->va_end & ~(align - 1), vmalloc_end);
3151 if ((*va)->va_start < addr)
3152 return addr;
3153 }
Tejun Heoca23e402009-08-14 15:00:52 +09003154 }
3155
Uladzislau Rezki (Sony)68ad4a32019-05-17 14:31:31 -07003156 return 0;
Tejun Heoca23e402009-08-14 15:00:52 +09003157}
3158
3159/**
3160 * pcpu_get_vm_areas - allocate vmalloc areas for percpu allocator
3161 * @offsets: array containing offset of each area
3162 * @sizes: array containing size of each area
3163 * @nr_vms: the number of areas to allocate
3164 * @align: alignment, all entries in @offsets and @sizes must be aligned to this
Tejun Heoca23e402009-08-14 15:00:52 +09003165 *
3166 * Returns: kmalloc'd vm_struct pointer array pointing to allocated
3167 * vm_structs on success, %NULL on failure
3168 *
3169 * Percpu allocator wants to use congruent vm areas so that it can
3170 * maintain the offsets among percpu areas. This function allocates
David Rientjesec3f64f2011-01-13 15:46:01 -08003171 * congruent vmalloc areas for it with GFP_KERNEL. These areas tend to
3172 * be scattered pretty far, distance between two areas easily going up
3173 * to gigabytes. To avoid interacting with regular vmallocs, these
3174 * areas are allocated from top.
Tejun Heoca23e402009-08-14 15:00:52 +09003175 *
Uladzislau Rezki (Sony)68ad4a32019-05-17 14:31:31 -07003176 * Despite its complicated look, this allocator is rather simple. It
3177 * does everything top-down and scans free blocks from the end looking
3178 * for matching base. While scanning, if any of the areas do not fit the
3179 * base address is pulled down to fit the area. Scanning is repeated till
3180 * all the areas fit and then all necessary data structures are inserted
3181 * and the result is returned.
Tejun Heoca23e402009-08-14 15:00:52 +09003182 */
3183struct vm_struct **pcpu_get_vm_areas(const unsigned long *offsets,
3184 const size_t *sizes, int nr_vms,
David Rientjesec3f64f2011-01-13 15:46:01 -08003185 size_t align)
Tejun Heoca23e402009-08-14 15:00:52 +09003186{
3187 const unsigned long vmalloc_start = ALIGN(VMALLOC_START, align);
3188 const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1);
Uladzislau Rezki (Sony)68ad4a32019-05-17 14:31:31 -07003189 struct vmap_area **vas, *va;
Tejun Heoca23e402009-08-14 15:00:52 +09003190 struct vm_struct **vms;
3191 int area, area2, last_area, term_area;
Daniel Axtens253a4962019-12-17 20:51:49 -08003192 unsigned long base, start, size, end, last_end, orig_start, orig_end;
Tejun Heoca23e402009-08-14 15:00:52 +09003193 bool purged = false;
Uladzislau Rezki (Sony)68ad4a32019-05-17 14:31:31 -07003194 enum fit_type type;
Tejun Heoca23e402009-08-14 15:00:52 +09003195
Tejun Heoca23e402009-08-14 15:00:52 +09003196 /* verify parameters and allocate data structures */
Alexander Kuleshov891c49a2015-11-05 18:46:51 -08003197 BUG_ON(offset_in_page(align) || !is_power_of_2(align));
Tejun Heoca23e402009-08-14 15:00:52 +09003198 for (last_area = 0, area = 0; area < nr_vms; area++) {
3199 start = offsets[area];
3200 end = start + sizes[area];
3201
3202 /* is everything aligned properly? */
3203 BUG_ON(!IS_ALIGNED(offsets[area], align));
3204 BUG_ON(!IS_ALIGNED(sizes[area], align));
3205
3206 /* detect the area with the highest address */
3207 if (start > offsets[last_area])
3208 last_area = area;
3209
Wei Yangc568da22017-09-06 16:24:09 -07003210 for (area2 = area + 1; area2 < nr_vms; area2++) {
Tejun Heoca23e402009-08-14 15:00:52 +09003211 unsigned long start2 = offsets[area2];
3212 unsigned long end2 = start2 + sizes[area2];
3213
Wei Yangc568da22017-09-06 16:24:09 -07003214 BUG_ON(start2 < end && start < end2);
Tejun Heoca23e402009-08-14 15:00:52 +09003215 }
3216 }
3217 last_end = offsets[last_area] + sizes[last_area];
3218
3219 if (vmalloc_end - vmalloc_start < last_end) {
3220 WARN_ON(true);
3221 return NULL;
3222 }
3223
Thomas Meyer4d67d862012-05-29 15:06:21 -07003224 vms = kcalloc(nr_vms, sizeof(vms[0]), GFP_KERNEL);
3225 vas = kcalloc(nr_vms, sizeof(vas[0]), GFP_KERNEL);
Tejun Heoca23e402009-08-14 15:00:52 +09003226 if (!vas || !vms)
Kautuk Consulf1db7af2012-01-12 17:20:08 -08003227 goto err_free2;
Tejun Heoca23e402009-08-14 15:00:52 +09003228
3229 for (area = 0; area < nr_vms; area++) {
Uladzislau Rezki (Sony)68ad4a32019-05-17 14:31:31 -07003230 vas[area] = kmem_cache_zalloc(vmap_area_cachep, GFP_KERNEL);
David Rientjesec3f64f2011-01-13 15:46:01 -08003231 vms[area] = kzalloc(sizeof(struct vm_struct), GFP_KERNEL);
Tejun Heoca23e402009-08-14 15:00:52 +09003232 if (!vas[area] || !vms[area])
3233 goto err_free;
3234 }
3235retry:
Uladzislau Rezki (Sony)e36176b2019-11-30 17:54:47 -08003236 spin_lock(&free_vmap_area_lock);
Tejun Heoca23e402009-08-14 15:00:52 +09003237
3238 /* start scanning - we scan from the top, begin with the last area */
3239 area = term_area = last_area;
3240 start = offsets[area];
3241 end = start + sizes[area];
3242
Uladzislau Rezki (Sony)68ad4a32019-05-17 14:31:31 -07003243 va = pvm_find_va_enclose_addr(vmalloc_end);
3244 base = pvm_determine_end_from_reverse(&va, align) - end;
Tejun Heoca23e402009-08-14 15:00:52 +09003245
3246 while (true) {
Tejun Heoca23e402009-08-14 15:00:52 +09003247 /*
3248 * base might have underflowed, add last_end before
3249 * comparing.
3250 */
Uladzislau Rezki (Sony)68ad4a32019-05-17 14:31:31 -07003251 if (base + last_end < vmalloc_start + last_end)
3252 goto overflow;
Tejun Heoca23e402009-08-14 15:00:52 +09003253
3254 /*
Uladzislau Rezki (Sony)68ad4a32019-05-17 14:31:31 -07003255 * Fitting base has not been found.
Tejun Heoca23e402009-08-14 15:00:52 +09003256 */
Uladzislau Rezki (Sony)68ad4a32019-05-17 14:31:31 -07003257 if (va == NULL)
3258 goto overflow;
Tejun Heoca23e402009-08-14 15:00:52 +09003259
3260 /*
Qiujun Huangd8cc3232020-04-06 20:04:02 -07003261 * If required width exceeds current VA block, move
Kuppuswamy Sathyanarayanan5336e522019-08-13 15:37:31 -07003262 * base downwards and then recheck.
3263 */
3264 if (base + end > va->va_end) {
3265 base = pvm_determine_end_from_reverse(&va, align) - end;
3266 term_area = area;
3267 continue;
3268 }
3269
3270 /*
Uladzislau Rezki (Sony)68ad4a32019-05-17 14:31:31 -07003271 * If this VA does not fit, move base downwards and recheck.
Tejun Heoca23e402009-08-14 15:00:52 +09003272 */
Kuppuswamy Sathyanarayanan5336e522019-08-13 15:37:31 -07003273 if (base + start < va->va_start) {
Uladzislau Rezki (Sony)68ad4a32019-05-17 14:31:31 -07003274 va = node_to_va(rb_prev(&va->rb_node));
3275 base = pvm_determine_end_from_reverse(&va, align) - end;
Tejun Heoca23e402009-08-14 15:00:52 +09003276 term_area = area;
3277 continue;
3278 }
3279
3280 /*
3281 * This area fits, move on to the previous one. If
3282 * the previous one is the terminal one, we're done.
3283 */
3284 area = (area + nr_vms - 1) % nr_vms;
3285 if (area == term_area)
3286 break;
Uladzislau Rezki (Sony)68ad4a32019-05-17 14:31:31 -07003287
Tejun Heoca23e402009-08-14 15:00:52 +09003288 start = offsets[area];
3289 end = start + sizes[area];
Uladzislau Rezki (Sony)68ad4a32019-05-17 14:31:31 -07003290 va = pvm_find_va_enclose_addr(base + end);
Tejun Heoca23e402009-08-14 15:00:52 +09003291 }
Uladzislau Rezki (Sony)68ad4a32019-05-17 14:31:31 -07003292
Tejun Heoca23e402009-08-14 15:00:52 +09003293 /* we've found a fitting base, insert all va's */
3294 for (area = 0; area < nr_vms; area++) {
Uladzislau Rezki (Sony)68ad4a32019-05-17 14:31:31 -07003295 int ret;
Tejun Heoca23e402009-08-14 15:00:52 +09003296
Uladzislau Rezki (Sony)68ad4a32019-05-17 14:31:31 -07003297 start = base + offsets[area];
3298 size = sizes[area];
3299
3300 va = pvm_find_va_enclose_addr(start);
3301 if (WARN_ON_ONCE(va == NULL))
3302 /* It is a BUG(), but trigger recovery instead. */
3303 goto recovery;
3304
3305 type = classify_va_fit_type(va, start, size);
3306 if (WARN_ON_ONCE(type == NOTHING_FIT))
3307 /* It is a BUG(), but trigger recovery instead. */
3308 goto recovery;
3309
3310 ret = adjust_va_to_fit_type(va, start, size, type);
3311 if (unlikely(ret))
3312 goto recovery;
3313
3314 /* Allocated area. */
3315 va = vas[area];
3316 va->va_start = start;
3317 va->va_end = start + size;
Tejun Heoca23e402009-08-14 15:00:52 +09003318 }
3319
Uladzislau Rezki (Sony)e36176b2019-11-30 17:54:47 -08003320 spin_unlock(&free_vmap_area_lock);
Tejun Heoca23e402009-08-14 15:00:52 +09003321
Daniel Axtens253a4962019-12-17 20:51:49 -08003322 /* populate the kasan shadow space */
3323 for (area = 0; area < nr_vms; area++) {
3324 if (kasan_populate_vmalloc(vas[area]->va_start, sizes[area]))
3325 goto err_free_shadow;
3326
3327 kasan_unpoison_vmalloc((void *)vas[area]->va_start,
3328 sizes[area]);
3329 }
3330
Tejun Heoca23e402009-08-14 15:00:52 +09003331 /* insert all vm's */
Uladzislau Rezki (Sony)e36176b2019-11-30 17:54:47 -08003332 spin_lock(&vmap_area_lock);
3333 for (area = 0; area < nr_vms; area++) {
3334 insert_vmap_area(vas[area], &vmap_area_root, &vmap_area_list);
3335
3336 setup_vmalloc_vm_locked(vms[area], vas[area], VM_ALLOC,
Zhang Yanfei3645cb42013-07-03 15:04:48 -07003337 pcpu_get_vm_areas);
Uladzislau Rezki (Sony)e36176b2019-11-30 17:54:47 -08003338 }
3339 spin_unlock(&vmap_area_lock);
Tejun Heoca23e402009-08-14 15:00:52 +09003340
3341 kfree(vas);
3342 return vms;
3343
Uladzislau Rezki (Sony)68ad4a32019-05-17 14:31:31 -07003344recovery:
Uladzislau Rezki (Sony)e36176b2019-11-30 17:54:47 -08003345 /*
3346 * Remove previously allocated areas. There is no
3347 * need in removing these areas from the busy tree,
3348 * because they are inserted only on the final step
3349 * and when pcpu_get_vm_areas() is success.
3350 */
Uladzislau Rezki (Sony)68ad4a32019-05-17 14:31:31 -07003351 while (area--) {
Daniel Axtens253a4962019-12-17 20:51:49 -08003352 orig_start = vas[area]->va_start;
3353 orig_end = vas[area]->va_end;
3354 va = merge_or_add_vmap_area(vas[area], &free_vmap_area_root,
3355 &free_vmap_area_list);
Uladzislau Rezki (Sony)9c801f62020-08-06 23:24:24 -07003356 if (va)
3357 kasan_release_vmalloc(orig_start, orig_end,
3358 va->va_start, va->va_end);
Uladzislau Rezki (Sony)68ad4a32019-05-17 14:31:31 -07003359 vas[area] = NULL;
3360 }
3361
3362overflow:
Uladzislau Rezki (Sony)e36176b2019-11-30 17:54:47 -08003363 spin_unlock(&free_vmap_area_lock);
Uladzislau Rezki (Sony)68ad4a32019-05-17 14:31:31 -07003364 if (!purged) {
3365 purge_vmap_area_lazy();
3366 purged = true;
3367
3368 /* Before "retry", check if we recover. */
3369 for (area = 0; area < nr_vms; area++) {
3370 if (vas[area])
3371 continue;
3372
3373 vas[area] = kmem_cache_zalloc(
3374 vmap_area_cachep, GFP_KERNEL);
3375 if (!vas[area])
3376 goto err_free;
3377 }
3378
3379 goto retry;
3380 }
3381
Tejun Heoca23e402009-08-14 15:00:52 +09003382err_free:
3383 for (area = 0; area < nr_vms; area++) {
Uladzislau Rezki (Sony)68ad4a32019-05-17 14:31:31 -07003384 if (vas[area])
3385 kmem_cache_free(vmap_area_cachep, vas[area]);
3386
Kautuk Consulf1db7af2012-01-12 17:20:08 -08003387 kfree(vms[area]);
Tejun Heoca23e402009-08-14 15:00:52 +09003388 }
Kautuk Consulf1db7af2012-01-12 17:20:08 -08003389err_free2:
Tejun Heoca23e402009-08-14 15:00:52 +09003390 kfree(vas);
3391 kfree(vms);
3392 return NULL;
Daniel Axtens253a4962019-12-17 20:51:49 -08003393
3394err_free_shadow:
3395 spin_lock(&free_vmap_area_lock);
3396 /*
3397 * We release all the vmalloc shadows, even the ones for regions that
3398 * hadn't been successfully added. This relies on kasan_release_vmalloc
3399 * being able to tolerate this case.
3400 */
3401 for (area = 0; area < nr_vms; area++) {
3402 orig_start = vas[area]->va_start;
3403 orig_end = vas[area]->va_end;
3404 va = merge_or_add_vmap_area(vas[area], &free_vmap_area_root,
3405 &free_vmap_area_list);
Uladzislau Rezki (Sony)9c801f62020-08-06 23:24:24 -07003406 if (va)
3407 kasan_release_vmalloc(orig_start, orig_end,
3408 va->va_start, va->va_end);
Daniel Axtens253a4962019-12-17 20:51:49 -08003409 vas[area] = NULL;
3410 kfree(vms[area]);
3411 }
3412 spin_unlock(&free_vmap_area_lock);
3413 kfree(vas);
3414 kfree(vms);
3415 return NULL;
Tejun Heoca23e402009-08-14 15:00:52 +09003416}
3417
3418/**
3419 * pcpu_free_vm_areas - free vmalloc areas for percpu allocator
3420 * @vms: vm_struct pointer array returned by pcpu_get_vm_areas()
3421 * @nr_vms: the number of allocated areas
3422 *
3423 * Free vm_structs and the array allocated by pcpu_get_vm_areas().
3424 */
3425void pcpu_free_vm_areas(struct vm_struct **vms, int nr_vms)
3426{
3427 int i;
3428
3429 for (i = 0; i < nr_vms; i++)
3430 free_vm_area(vms[i]);
3431 kfree(vms);
3432}
Tejun Heo4f8b02b2010-09-03 18:22:47 +02003433#endif /* CONFIG_SMP */
Christoph Lametera10aa572008-04-28 02:12:40 -07003434
3435#ifdef CONFIG_PROC_FS
3436static void *s_start(struct seq_file *m, loff_t *pos)
Uladzislau Rezki (Sony)e36176b2019-11-30 17:54:47 -08003437 __acquires(&vmap_purge_lock)
Joonsoo Kimd4033af2013-04-29 15:07:35 -07003438 __acquires(&vmap_area_lock)
Christoph Lametera10aa572008-04-28 02:12:40 -07003439{
Uladzislau Rezki (Sony)e36176b2019-11-30 17:54:47 -08003440 mutex_lock(&vmap_purge_lock);
Joonsoo Kimd4033af2013-04-29 15:07:35 -07003441 spin_lock(&vmap_area_lock);
Uladzislau Rezki (Sony)e36176b2019-11-30 17:54:47 -08003442
zijun_hu3f500062016-12-12 16:42:17 -08003443 return seq_list_start(&vmap_area_list, *pos);
Christoph Lametera10aa572008-04-28 02:12:40 -07003444}
3445
3446static void *s_next(struct seq_file *m, void *p, loff_t *pos)
3447{
zijun_hu3f500062016-12-12 16:42:17 -08003448 return seq_list_next(p, &vmap_area_list, pos);
Christoph Lametera10aa572008-04-28 02:12:40 -07003449}
3450
3451static void s_stop(struct seq_file *m, void *p)
Joonsoo Kimd4033af2013-04-29 15:07:35 -07003452 __releases(&vmap_area_lock)
Waiman Long4a9d8b02020-12-14 19:08:59 -08003453 __releases(&vmap_purge_lock)
Christoph Lametera10aa572008-04-28 02:12:40 -07003454{
Joonsoo Kimd4033af2013-04-29 15:07:35 -07003455 spin_unlock(&vmap_area_lock);
Waiman Long4a9d8b02020-12-14 19:08:59 -08003456 mutex_unlock(&vmap_purge_lock);
Christoph Lametera10aa572008-04-28 02:12:40 -07003457}
3458
Eric Dumazeta47a1262008-07-23 21:27:38 -07003459static void show_numa_info(struct seq_file *m, struct vm_struct *v)
3460{
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08003461 if (IS_ENABLED(CONFIG_NUMA)) {
Eric Dumazeta47a1262008-07-23 21:27:38 -07003462 unsigned int nr, *counters = m->private;
3463
3464 if (!counters)
3465 return;
3466
Wanpeng Liaf123462013-11-12 15:07:32 -08003467 if (v->flags & VM_UNINITIALIZED)
3468 return;
Dmitry Vyukov7e5b5282014-12-12 16:56:30 -08003469 /* Pair with smp_wmb() in clear_vm_uninitialized_flag() */
3470 smp_rmb();
Wanpeng Liaf123462013-11-12 15:07:32 -08003471
Eric Dumazeta47a1262008-07-23 21:27:38 -07003472 memset(counters, 0, nr_node_ids * sizeof(unsigned int));
3473
3474 for (nr = 0; nr < v->nr_pages; nr++)
3475 counters[page_to_nid(v->pages[nr])]++;
3476
3477 for_each_node_state(nr, N_HIGH_MEMORY)
3478 if (counters[nr])
3479 seq_printf(m, " N%u=%u", nr, counters[nr]);
3480 }
3481}
3482
Uladzislau Rezki (Sony)dd3b8352019-09-23 15:36:36 -07003483static void show_purge_info(struct seq_file *m)
3484{
3485 struct llist_node *head;
3486 struct vmap_area *va;
3487
3488 head = READ_ONCE(vmap_purge_list.first);
3489 if (head == NULL)
3490 return;
3491
3492 llist_for_each_entry(va, head, purge_list) {
3493 seq_printf(m, "0x%pK-0x%pK %7ld unpurged vm_area\n",
3494 (void *)va->va_start, (void *)va->va_end,
3495 va->va_end - va->va_start);
3496 }
3497}
3498
Christoph Lametera10aa572008-04-28 02:12:40 -07003499static int s_show(struct seq_file *m, void *p)
3500{
zijun_hu3f500062016-12-12 16:42:17 -08003501 struct vmap_area *va;
Joonsoo Kimd4033af2013-04-29 15:07:35 -07003502 struct vm_struct *v;
3503
zijun_hu3f500062016-12-12 16:42:17 -08003504 va = list_entry(p, struct vmap_area, list);
3505
Wanpeng Lic2ce8c12013-11-12 15:07:31 -08003506 /*
Pengfei Li688fcbf2019-09-23 15:36:39 -07003507 * s_show can encounter race with remove_vm_area, !vm on behalf
3508 * of vmap area is being tear down or vm_map_ram allocation.
Wanpeng Lic2ce8c12013-11-12 15:07:31 -08003509 */
Pengfei Li688fcbf2019-09-23 15:36:39 -07003510 if (!va->vm) {
Uladzislau Rezki (Sony)dd3b8352019-09-23 15:36:36 -07003511 seq_printf(m, "0x%pK-0x%pK %7ld vm_map_ram\n",
Yisheng Xie78c72742017-07-10 15:48:09 -07003512 (void *)va->va_start, (void *)va->va_end,
Uladzislau Rezki (Sony)dd3b8352019-09-23 15:36:36 -07003513 va->va_end - va->va_start);
Yisheng Xie78c72742017-07-10 15:48:09 -07003514
Joonsoo Kimd4033af2013-04-29 15:07:35 -07003515 return 0;
Yisheng Xie78c72742017-07-10 15:48:09 -07003516 }
Joonsoo Kimd4033af2013-04-29 15:07:35 -07003517
Joonsoo Kimd4033af2013-04-29 15:07:35 -07003518 v = va->vm;
Christoph Lametera10aa572008-04-28 02:12:40 -07003519
Kees Cook45ec1692012-10-08 16:34:09 -07003520 seq_printf(m, "0x%pK-0x%pK %7ld",
Christoph Lametera10aa572008-04-28 02:12:40 -07003521 v->addr, v->addr + v->size, v->size);
3522
Joe Perches62c70bc2011-01-13 15:45:52 -08003523 if (v->caller)
3524 seq_printf(m, " %pS", v->caller);
Christoph Lameter23016962008-04-28 02:12:42 -07003525
Christoph Lametera10aa572008-04-28 02:12:40 -07003526 if (v->nr_pages)
3527 seq_printf(m, " pages=%d", v->nr_pages);
3528
3529 if (v->phys_addr)
Miles Chen199eaa02017-02-24 14:59:51 -08003530 seq_printf(m, " phys=%pa", &v->phys_addr);
Christoph Lametera10aa572008-04-28 02:12:40 -07003531
3532 if (v->flags & VM_IOREMAP)
Fabian Frederickf4527c92014-06-04 16:08:09 -07003533 seq_puts(m, " ioremap");
Christoph Lametera10aa572008-04-28 02:12:40 -07003534
3535 if (v->flags & VM_ALLOC)
Fabian Frederickf4527c92014-06-04 16:08:09 -07003536 seq_puts(m, " vmalloc");
Christoph Lametera10aa572008-04-28 02:12:40 -07003537
3538 if (v->flags & VM_MAP)
Fabian Frederickf4527c92014-06-04 16:08:09 -07003539 seq_puts(m, " vmap");
Christoph Lametera10aa572008-04-28 02:12:40 -07003540
3541 if (v->flags & VM_USERMAP)
Fabian Frederickf4527c92014-06-04 16:08:09 -07003542 seq_puts(m, " user");
Christoph Lametera10aa572008-04-28 02:12:40 -07003543
Christoph Hellwigfe9041c2019-06-03 08:55:13 +02003544 if (v->flags & VM_DMA_COHERENT)
3545 seq_puts(m, " dma-coherent");
3546
David Rientjes244d63e2016-01-14 15:19:35 -08003547 if (is_vmalloc_addr(v->pages))
Fabian Frederickf4527c92014-06-04 16:08:09 -07003548 seq_puts(m, " vpages");
Christoph Lametera10aa572008-04-28 02:12:40 -07003549
Eric Dumazeta47a1262008-07-23 21:27:38 -07003550 show_numa_info(m, v);
Christoph Lametera10aa572008-04-28 02:12:40 -07003551 seq_putc(m, '\n');
Uladzislau Rezki (Sony)dd3b8352019-09-23 15:36:36 -07003552
3553 /*
3554 * As a final step, dump "unpurged" areas. Note,
3555 * that entire "/proc/vmallocinfo" output will not
3556 * be address sorted, because the purge list is not
3557 * sorted.
3558 */
3559 if (list_is_last(&va->list, &vmap_area_list))
3560 show_purge_info(m);
3561
Christoph Lametera10aa572008-04-28 02:12:40 -07003562 return 0;
3563}
3564
Alexey Dobriyan5f6a6a92008-10-06 03:50:47 +04003565static const struct seq_operations vmalloc_op = {
Christoph Lametera10aa572008-04-28 02:12:40 -07003566 .start = s_start,
3567 .next = s_next,
3568 .stop = s_stop,
3569 .show = s_show,
3570};
Alexey Dobriyan5f6a6a92008-10-06 03:50:47 +04003571
Alexey Dobriyan5f6a6a92008-10-06 03:50:47 +04003572static int __init proc_vmalloc_init(void)
3573{
Christoph Hellwigfddda2b2018-04-13 19:44:18 +02003574 if (IS_ENABLED(CONFIG_NUMA))
Joe Perches0825a6f2018-06-14 15:27:58 -07003575 proc_create_seq_private("vmallocinfo", 0400, NULL,
Christoph Hellwig44414d82018-04-24 17:05:17 +02003576 &vmalloc_op,
3577 nr_node_ids * sizeof(unsigned int), NULL);
Christoph Hellwigfddda2b2018-04-13 19:44:18 +02003578 else
Joe Perches0825a6f2018-06-14 15:27:58 -07003579 proc_create_seq("vmallocinfo", 0400, NULL, &vmalloc_op);
Alexey Dobriyan5f6a6a92008-10-06 03:50:47 +04003580 return 0;
3581}
3582module_init(proc_vmalloc_init);
Joonsoo Kimdb3808c2013-04-29 15:07:28 -07003583
Christoph Lametera10aa572008-04-28 02:12:40 -07003584#endif