blob: 762679050c9a033151a0d982a426a7675fd2a251 [file] [log] [blame]
Greg Kroah-Hartmanb2441312017-11-01 15:07:57 +01001// SPDX-License-Identifier: GPL-2.0
Linus Torvalds1da177e2005-04-16 15:20:36 -07002/*
3 * High memory handling common code and variables.
4 *
5 * (C) 1999 Andrea Arcangeli, SuSE GmbH, andrea@suse.de
6 * Gerhard Wichert, Siemens AG, Gerhard.Wichert@pdb.siemens.de
7 *
8 *
9 * Redesigned the x86 32-bit VM architecture to deal with
10 * 64-bit physical space. With current x86 CPUs this
11 * means up to 64 Gigabytes physical RAM.
12 *
13 * Rewrote high memory support to move the page cache into
14 * high memory. Implemented permanent (schedulable) kmaps
15 * based on Linus' idea.
16 *
17 * Copyright (C) 1999 Ingo Molnar <mingo@redhat.com>
18 */
19
20#include <linux/mm.h>
Paul Gortmakerb95f1b312011-10-16 02:01:52 -040021#include <linux/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include <linux/swap.h>
23#include <linux/bio.h>
24#include <linux/pagemap.h>
25#include <linux/mempool.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070026#include <linux/init.h>
27#include <linux/hash.h>
28#include <linux/highmem.h>
Jason Wesseleac790052010-08-05 09:22:24 -050029#include <linux/kgdb.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070030#include <asm/tlbflush.h>
Ingo Molnar186525b2019-11-29 08:17:25 +010031#include <linux/vmalloc.h>
Peter Zijlstraa8e23a22010-10-27 15:32:57 -070032
Linus Torvalds1da177e2005-04-16 15:20:36 -070033/*
34 * Virtual_count is not a pure "count".
35 * 0 means that it is not mapped, and has not been mapped
36 * since a TLB flush - it is usable.
37 * 1 means that there are no users, but it has been mapped
38 * since the last TLB flush - so we can't use it.
39 * n means that there are (n-1) current users of it.
40 */
41#ifdef CONFIG_HIGHMEM
Al Viro260b2362005-10-21 03:22:44 -040042
Max Filippov15de36a2014-08-06 16:08:23 -070043/*
44 * Architecture with aliasing data cache may define the following family of
45 * helper functions in its asm/highmem.h to control cache color of virtual
46 * addresses where physical memory pages are mapped by kmap.
47 */
48#ifndef get_pkmap_color
49
50/*
51 * Determine color of virtual address where the page should be mapped.
52 */
53static inline unsigned int get_pkmap_color(struct page *page)
54{
55 return 0;
56}
57#define get_pkmap_color get_pkmap_color
58
59/*
60 * Get next index for mapping inside PKMAP region for page with given color.
61 */
62static inline unsigned int get_next_pkmap_nr(unsigned int color)
63{
64 static unsigned int last_pkmap_nr;
65
66 last_pkmap_nr = (last_pkmap_nr + 1) & LAST_PKMAP_MASK;
67 return last_pkmap_nr;
68}
69
70/*
71 * Determine if page index inside PKMAP region (pkmap_nr) of given color
72 * has wrapped around PKMAP region end. When this happens an attempt to
73 * flush all unused PKMAP slots is made.
74 */
75static inline int no_more_pkmaps(unsigned int pkmap_nr, unsigned int color)
76{
77 return pkmap_nr == 0;
78}
79
80/*
81 * Get the number of PKMAP entries of the given color. If no free slot is
82 * found after checking that many entries, kmap will sleep waiting for
83 * someone to call kunmap and free PKMAP slot.
84 */
85static inline int get_pkmap_entries_count(unsigned int color)
86{
87 return LAST_PKMAP;
88}
89
90/*
91 * Get head of a wait queue for PKMAP entries of the given color.
92 * Wait queues for different mapping colors should be independent to avoid
93 * unnecessary wakeups caused by freeing of slots of other colors.
94 */
95static inline wait_queue_head_t *get_pkmap_wait_queue_head(unsigned int color)
96{
97 static DECLARE_WAIT_QUEUE_HEAD(pkmap_map_wait);
98
99 return &pkmap_map_wait;
100}
101#endif
102
Arun KSca79b0c2018-12-28 00:34:29 -0800103atomic_long_t _totalhigh_pages __read_mostly;
104EXPORT_SYMBOL(_totalhigh_pages);
Peter Zijlstra3e4d3af2010-10-26 14:21:51 -0700105
songqiang97276882021-05-04 18:40:09 -0700106unsigned int __nr_free_highpages(void)
Christoph Lameterc1f60a52006-09-25 23:31:11 -0700107{
Joonsoo Kim33499bf2016-05-19 17:12:26 -0700108 struct zone *zone;
Christoph Lameterc1f60a52006-09-25 23:31:11 -0700109 unsigned int pages = 0;
110
Joonsoo Kim33499bf2016-05-19 17:12:26 -0700111 for_each_populated_zone(zone) {
112 if (is_highmem(zone))
113 pages += zone_page_state(zone, NR_FREE_PAGES);
Mel Gorman2a1e2742007-07-17 04:03:12 -0700114 }
Christoph Lameterc1f60a52006-09-25 23:31:11 -0700115
116 return pages;
117}
118
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119static int pkmap_count[LAST_PKMAP];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120static __cacheline_aligned_in_smp DEFINE_SPINLOCK(kmap_lock);
121
songqiang97276882021-05-04 18:40:09 -0700122pte_t *pkmap_page_table;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700123
Nicolas Pitre3297e762009-03-04 22:49:41 -0500124/*
125 * Most architectures have no use for kmap_high_get(), so let's abstract
126 * the disabling of IRQ out of the locking in that case to save on a
127 * potential useless overhead.
128 */
129#ifdef ARCH_NEEDS_KMAP_HIGH_GET
130#define lock_kmap() spin_lock_irq(&kmap_lock)
131#define unlock_kmap() spin_unlock_irq(&kmap_lock)
132#define lock_kmap_any(flags) spin_lock_irqsave(&kmap_lock, flags)
133#define unlock_kmap_any(flags) spin_unlock_irqrestore(&kmap_lock, flags)
134#else
135#define lock_kmap() spin_lock(&kmap_lock)
136#define unlock_kmap() spin_unlock(&kmap_lock)
137#define lock_kmap_any(flags) \
138 do { spin_lock(&kmap_lock); (void)(flags); } while (0)
139#define unlock_kmap_any(flags) \
140 do { spin_unlock(&kmap_lock); (void)(flags); } while (0)
141#endif
142
Thomas Gleixner13f876b2020-11-03 10:27:34 +0100143struct page *__kmap_to_page(void *vaddr)
Mel Gorman5a178112012-07-31 16:45:02 -0700144{
145 unsigned long addr = (unsigned long)vaddr;
146
Will Deacon498c2282012-11-16 14:15:00 -0800147 if (addr >= PKMAP_ADDR(0) && addr < PKMAP_ADDR(LAST_PKMAP)) {
Joonsoo Kim4de22c02012-12-11 16:01:17 -0800148 int i = PKMAP_NR(addr);
songqiang97276882021-05-04 18:40:09 -0700149
Mel Gorman5a178112012-07-31 16:45:02 -0700150 return pte_page(pkmap_page_table[i]);
151 }
152
153 return virt_to_page(addr);
154}
Thomas Gleixner13f876b2020-11-03 10:27:34 +0100155EXPORT_SYMBOL(__kmap_to_page);
Mel Gorman5a178112012-07-31 16:45:02 -0700156
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157static void flush_all_zero_pkmaps(void)
158{
159 int i;
Nick Piggin5843d9a2008-08-01 03:15:21 +0200160 int need_flush = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161
162 flush_cache_kmaps();
163
164 for (i = 0; i < LAST_PKMAP; i++) {
165 struct page *page;
166
167 /*
168 * zero means we don't have anything to do,
169 * >1 means that it is still in use. Only
170 * a count of 1 means that it is free but
171 * needs to be unmapped
172 */
173 if (pkmap_count[i] != 1)
174 continue;
175 pkmap_count[i] = 0;
176
177 /* sanity check */
Eric Sesterhenn75babca2006-04-02 13:47:35 +0200178 BUG_ON(pte_none(pkmap_page_table[i]));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179
180 /*
181 * Don't need an atomic fetch-and-clear op here;
182 * no-one has the page mapped, and cannot get at
183 * its virtual address (and hence PTE) without first
184 * getting the kmap_lock (which is held here).
185 * So no dangers, even with speculative execution.
186 */
187 page = pte_page(pkmap_page_table[i]);
Joonsoo Kimeb2db432012-12-11 16:01:24 -0800188 pte_clear(&init_mm, PKMAP_ADDR(i), &pkmap_page_table[i]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189
190 set_page_address(page, NULL);
Nick Piggin5843d9a2008-08-01 03:15:21 +0200191 need_flush = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192 }
Nick Piggin5843d9a2008-08-01 03:15:21 +0200193 if (need_flush)
194 flush_tlb_kernel_range(PKMAP_ADDR(0), PKMAP_ADDR(LAST_PKMAP));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195}
196
Thomas Gleixner13f876b2020-11-03 10:27:34 +0100197void __kmap_flush_unused(void)
Jeremy Fitzhardingece6234b2007-05-02 19:27:15 +0200198{
Nicolas Pitre3297e762009-03-04 22:49:41 -0500199 lock_kmap();
Jeremy Fitzhardingece6234b2007-05-02 19:27:15 +0200200 flush_all_zero_pkmaps();
Nicolas Pitre3297e762009-03-04 22:49:41 -0500201 unlock_kmap();
Jeremy Fitzhardingece6234b2007-05-02 19:27:15 +0200202}
203
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204static inline unsigned long map_new_virtual(struct page *page)
205{
206 unsigned long vaddr;
207 int count;
Max Filippov15de36a2014-08-06 16:08:23 -0700208 unsigned int last_pkmap_nr;
209 unsigned int color = get_pkmap_color(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210
211start:
Max Filippov15de36a2014-08-06 16:08:23 -0700212 count = get_pkmap_entries_count(color);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213 /* Find an empty entry */
214 for (;;) {
Max Filippov15de36a2014-08-06 16:08:23 -0700215 last_pkmap_nr = get_next_pkmap_nr(color);
216 if (no_more_pkmaps(last_pkmap_nr, color)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217 flush_all_zero_pkmaps();
Max Filippov15de36a2014-08-06 16:08:23 -0700218 count = get_pkmap_entries_count(color);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219 }
220 if (!pkmap_count[last_pkmap_nr])
221 break; /* Found a usable entry */
222 if (--count)
223 continue;
224
225 /*
226 * Sleep for somebody else to unmap their entries
227 */
228 {
229 DECLARE_WAITQUEUE(wait, current);
Max Filippov15de36a2014-08-06 16:08:23 -0700230 wait_queue_head_t *pkmap_map_wait =
231 get_pkmap_wait_queue_head(color);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232
233 __set_current_state(TASK_UNINTERRUPTIBLE);
Max Filippov15de36a2014-08-06 16:08:23 -0700234 add_wait_queue(pkmap_map_wait, &wait);
Nicolas Pitre3297e762009-03-04 22:49:41 -0500235 unlock_kmap();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236 schedule();
Max Filippov15de36a2014-08-06 16:08:23 -0700237 remove_wait_queue(pkmap_map_wait, &wait);
Nicolas Pitre3297e762009-03-04 22:49:41 -0500238 lock_kmap();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239
240 /* Somebody else might have mapped it while we slept */
241 if (page_address(page))
242 return (unsigned long)page_address(page);
243
244 /* Re-start */
245 goto start;
246 }
247 }
248 vaddr = PKMAP_ADDR(last_pkmap_nr);
249 set_pte_at(&init_mm, vaddr,
250 &(pkmap_page_table[last_pkmap_nr]), mk_pte(page, kmap_prot));
251
252 pkmap_count[last_pkmap_nr] = 1;
253 set_page_address(page, (void *)vaddr);
254
255 return vaddr;
256}
257
Randy Dunlap77f60782008-03-19 17:00:42 -0700258/**
259 * kmap_high - map a highmem page into memory
260 * @page: &struct page to map
261 *
262 * Returns the page's virtual memory address.
263 *
264 * We cannot call this from interrupts, as it may block.
265 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800266void *kmap_high(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267{
268 unsigned long vaddr;
269
270 /*
271 * For highmem pages, we can't trust "virtual" until
272 * after we have the lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273 */
Nicolas Pitre3297e762009-03-04 22:49:41 -0500274 lock_kmap();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275 vaddr = (unsigned long)page_address(page);
276 if (!vaddr)
277 vaddr = map_new_virtual(page);
278 pkmap_count[PKMAP_NR(vaddr)]++;
Eric Sesterhenn75babca2006-04-02 13:47:35 +0200279 BUG_ON(pkmap_count[PKMAP_NR(vaddr)] < 2);
Nicolas Pitre3297e762009-03-04 22:49:41 -0500280 unlock_kmap();
songqiang97276882021-05-04 18:40:09 -0700281 return (void *) vaddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700283EXPORT_SYMBOL(kmap_high);
284
Nicolas Pitre3297e762009-03-04 22:49:41 -0500285#ifdef ARCH_NEEDS_KMAP_HIGH_GET
286/**
287 * kmap_high_get - pin a highmem page into memory
288 * @page: &struct page to pin
289 *
290 * Returns the page's current virtual memory address, or NULL if no mapping
Uwe Kleine-König5e39df52010-01-25 21:38:09 +0100291 * exists. If and only if a non null address is returned then a
Nicolas Pitre3297e762009-03-04 22:49:41 -0500292 * matching call to kunmap_high() is necessary.
293 *
294 * This can be called from any context.
295 */
296void *kmap_high_get(struct page *page)
297{
298 unsigned long vaddr, flags;
299
300 lock_kmap_any(flags);
301 vaddr = (unsigned long)page_address(page);
302 if (vaddr) {
303 BUG_ON(pkmap_count[PKMAP_NR(vaddr)] < 1);
304 pkmap_count[PKMAP_NR(vaddr)]++;
305 }
306 unlock_kmap_any(flags);
songqiang97276882021-05-04 18:40:09 -0700307 return (void *) vaddr;
Nicolas Pitre3297e762009-03-04 22:49:41 -0500308}
309#endif
310
Randy Dunlap77f60782008-03-19 17:00:42 -0700311/**
Li Haifeng4e9dc5d2011-10-31 17:09:09 -0700312 * kunmap_high - unmap a highmem page into memory
Randy Dunlap77f60782008-03-19 17:00:42 -0700313 * @page: &struct page to unmap
Nicolas Pitre3297e762009-03-04 22:49:41 -0500314 *
315 * If ARCH_NEEDS_KMAP_HIGH_GET is not defined then this may be called
316 * only from user context.
Randy Dunlap77f60782008-03-19 17:00:42 -0700317 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800318void kunmap_high(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700319{
320 unsigned long vaddr;
321 unsigned long nr;
Nicolas Pitre3297e762009-03-04 22:49:41 -0500322 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323 int need_wakeup;
Max Filippov15de36a2014-08-06 16:08:23 -0700324 unsigned int color = get_pkmap_color(page);
325 wait_queue_head_t *pkmap_map_wait;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326
Nicolas Pitre3297e762009-03-04 22:49:41 -0500327 lock_kmap_any(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328 vaddr = (unsigned long)page_address(page);
Eric Sesterhenn75babca2006-04-02 13:47:35 +0200329 BUG_ON(!vaddr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330 nr = PKMAP_NR(vaddr);
331
332 /*
333 * A count must never go down to zero
334 * without a TLB flush!
335 */
336 need_wakeup = 0;
337 switch (--pkmap_count[nr]) {
338 case 0:
339 BUG();
340 case 1:
341 /*
342 * Avoid an unnecessary wake_up() function call.
343 * The common case is pkmap_count[] == 1, but
344 * no waiters.
345 * The tasks queued in the wait-queue are guarded
346 * by both the lock in the wait-queue-head and by
347 * the kmap_lock. As the kmap_lock is held here,
348 * no need for the wait-queue-head's lock. Simply
349 * test if the queue is empty.
350 */
Max Filippov15de36a2014-08-06 16:08:23 -0700351 pkmap_map_wait = get_pkmap_wait_queue_head(color);
352 need_wakeup = waitqueue_active(pkmap_map_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353 }
Nicolas Pitre3297e762009-03-04 22:49:41 -0500354 unlock_kmap_any(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355
356 /* do wake-up, if needed, race-free outside of the spin lock */
357 if (need_wakeup)
Max Filippov15de36a2014-08-06 16:08:23 -0700358 wake_up(pkmap_map_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700360EXPORT_SYMBOL(kunmap_high);
Matthew Wilcox (Oracle)0060ef32020-12-14 19:12:59 -0800361
Matthew Wilcox (Oracle)0060ef32020-12-14 19:12:59 -0800362void zero_user_segments(struct page *page, unsigned start1, unsigned end1,
363 unsigned start2, unsigned end2)
364{
365 unsigned int i;
366
367 BUG_ON(end1 > page_size(page) || end2 > page_size(page));
368
OGAWA Hirofumi184cee52021-03-12 21:07:37 -0800369 if (start1 >= end1)
370 start1 = end1 = 0;
371 if (start2 >= end2)
372 start2 = end2 = 0;
373
Matthew Wilcox (Oracle)0060ef32020-12-14 19:12:59 -0800374 for (i = 0; i < compound_nr(page); i++) {
375 void *kaddr = NULL;
376
Matthew Wilcox (Oracle)0060ef32020-12-14 19:12:59 -0800377 if (start1 >= PAGE_SIZE) {
378 start1 -= PAGE_SIZE;
379 end1 -= PAGE_SIZE;
380 } else {
381 unsigned this_end = min_t(unsigned, end1, PAGE_SIZE);
382
OGAWA Hirofumi184cee52021-03-12 21:07:37 -0800383 if (end1 > start1) {
Ira Weinyd2c20e52021-11-05 13:45:06 -0700384 kaddr = kmap_local_page(page + i);
Matthew Wilcox (Oracle)0060ef32020-12-14 19:12:59 -0800385 memset(kaddr + start1, 0, this_end - start1);
OGAWA Hirofumi184cee52021-03-12 21:07:37 -0800386 }
Matthew Wilcox (Oracle)0060ef32020-12-14 19:12:59 -0800387 end1 -= this_end;
388 start1 = 0;
389 }
390
391 if (start2 >= PAGE_SIZE) {
392 start2 -= PAGE_SIZE;
393 end2 -= PAGE_SIZE;
394 } else {
395 unsigned this_end = min_t(unsigned, end2, PAGE_SIZE);
396
OGAWA Hirofumi184cee52021-03-12 21:07:37 -0800397 if (end2 > start2) {
398 if (!kaddr)
Ira Weinyd2c20e52021-11-05 13:45:06 -0700399 kaddr = kmap_local_page(page + i);
Matthew Wilcox (Oracle)0060ef32020-12-14 19:12:59 -0800400 memset(kaddr + start2, 0, this_end - start2);
OGAWA Hirofumi184cee52021-03-12 21:07:37 -0800401 }
Matthew Wilcox (Oracle)0060ef32020-12-14 19:12:59 -0800402 end2 -= this_end;
403 start2 = 0;
404 }
405
406 if (kaddr) {
Ira Weinyd2c20e52021-11-05 13:45:06 -0700407 kunmap_local(kaddr);
Matthew Wilcox (Oracle)0060ef32020-12-14 19:12:59 -0800408 flush_dcache_page(page + i);
409 }
410
411 if (!end1 && !end2)
412 break;
413 }
414
415 BUG_ON((start1 | start2 | end1 | end2) != 0);
416}
417EXPORT_SYMBOL(zero_user_segments);
Thomas Gleixner298fa1a2020-11-03 10:27:18 +0100418#endif /* CONFIG_HIGHMEM */
419
420#ifdef CONFIG_KMAP_LOCAL
421
422#include <asm/kmap_size.h>
423
Thomas Gleixner389755c2020-11-03 10:27:19 +0100424/*
Thomas Gleixner6e799cb2020-11-18 20:48:39 +0100425 * With DEBUG_KMAP_LOCAL the stack depth is doubled and every second
Thomas Gleixner389755c2020-11-03 10:27:19 +0100426 * slot is unused which acts as a guard page
427 */
Thomas Gleixner6e799cb2020-11-18 20:48:39 +0100428#ifdef CONFIG_DEBUG_KMAP_LOCAL
Thomas Gleixner389755c2020-11-03 10:27:19 +0100429# define KM_INCR 2
430#else
431# define KM_INCR 1
432#endif
433
Thomas Gleixner298fa1a2020-11-03 10:27:18 +0100434static inline int kmap_local_idx_push(void)
435{
Changbin Duea0eafe2021-09-07 19:56:12 -0700436 WARN_ON_ONCE(in_hardirq() && !irqs_disabled());
Thomas Gleixner5fbda3e2020-11-18 20:48:43 +0100437 current->kmap_ctrl.idx += KM_INCR;
438 BUG_ON(current->kmap_ctrl.idx >= KM_MAX_IDX);
439 return current->kmap_ctrl.idx - 1;
Thomas Gleixner298fa1a2020-11-03 10:27:18 +0100440}
441
442static inline int kmap_local_idx(void)
443{
Thomas Gleixner5fbda3e2020-11-18 20:48:43 +0100444 return current->kmap_ctrl.idx - 1;
Thomas Gleixner298fa1a2020-11-03 10:27:18 +0100445}
446
447static inline void kmap_local_idx_pop(void)
448{
Thomas Gleixner5fbda3e2020-11-18 20:48:43 +0100449 current->kmap_ctrl.idx -= KM_INCR;
450 BUG_ON(current->kmap_ctrl.idx < 0);
Thomas Gleixner298fa1a2020-11-03 10:27:18 +0100451}
452
453#ifndef arch_kmap_local_post_map
454# define arch_kmap_local_post_map(vaddr, pteval) do { } while (0)
455#endif
Thomas Gleixner3c1016b2020-11-03 10:27:31 +0100456
Thomas Gleixner298fa1a2020-11-03 10:27:18 +0100457#ifndef arch_kmap_local_pre_unmap
458# define arch_kmap_local_pre_unmap(vaddr) do { } while (0)
459#endif
460
461#ifndef arch_kmap_local_post_unmap
462# define arch_kmap_local_post_unmap(vaddr) do { } while (0)
463#endif
464
465#ifndef arch_kmap_local_map_idx
466#define arch_kmap_local_map_idx(idx, pfn) kmap_local_calc_idx(idx)
467#endif
468
469#ifndef arch_kmap_local_unmap_idx
470#define arch_kmap_local_unmap_idx(idx, vaddr) kmap_local_calc_idx(idx)
471#endif
472
473#ifndef arch_kmap_local_high_get
474static inline void *arch_kmap_local_high_get(struct page *page)
475{
476 return NULL;
477}
478#endif
479
Thomas Gleixnera1dce7f2021-01-23 21:02:02 -0800480#ifndef arch_kmap_local_set_pte
481#define arch_kmap_local_set_pte(mm, vaddr, ptep, ptev) \
482 set_pte_at(mm, vaddr, ptep, ptev)
483#endif
484
Thomas Gleixner298fa1a2020-11-03 10:27:18 +0100485/* Unmap a local mapping which was obtained by kmap_high_get() */
Thomas Gleixner2a656ca2020-11-12 11:59:32 +0100486static inline bool kmap_high_unmap_local(unsigned long vaddr)
Thomas Gleixner298fa1a2020-11-03 10:27:18 +0100487{
488#ifdef ARCH_NEEDS_KMAP_HIGH_GET
Thomas Gleixner2a656ca2020-11-12 11:59:32 +0100489 if (vaddr >= PKMAP_ADDR(0) && vaddr < PKMAP_ADDR(LAST_PKMAP)) {
Thomas Gleixner298fa1a2020-11-03 10:27:18 +0100490 kunmap_high(pte_page(pkmap_page_table[PKMAP_NR(vaddr)]));
Thomas Gleixner2a656ca2020-11-12 11:59:32 +0100491 return true;
492 }
Thomas Gleixner298fa1a2020-11-03 10:27:18 +0100493#endif
Thomas Gleixner2a656ca2020-11-12 11:59:32 +0100494 return false;
Thomas Gleixner298fa1a2020-11-03 10:27:18 +0100495}
496
497static inline int kmap_local_calc_idx(int idx)
498{
499 return idx + KM_MAX_IDX * smp_processor_id();
500}
501
502static pte_t *__kmap_pte;
503
Ard Biesheuvel825c43f2021-11-19 16:43:55 -0800504static pte_t *kmap_get_pte(unsigned long vaddr, int idx)
Thomas Gleixner298fa1a2020-11-03 10:27:18 +0100505{
Ard Biesheuvel825c43f2021-11-19 16:43:55 -0800506 if (IS_ENABLED(CONFIG_KMAP_LOCAL_NON_LINEAR_PTE_ARRAY))
507 /*
508 * Set by the arch if __kmap_pte[-idx] does not produce
509 * the correct entry.
510 */
511 return virt_to_kpte(vaddr);
Thomas Gleixner298fa1a2020-11-03 10:27:18 +0100512 if (!__kmap_pte)
513 __kmap_pte = virt_to_kpte(__fix_to_virt(FIX_KMAP_BEGIN));
Ard Biesheuvel825c43f2021-11-19 16:43:55 -0800514 return &__kmap_pte[-idx];
Thomas Gleixner298fa1a2020-11-03 10:27:18 +0100515}
516
517void *__kmap_local_pfn_prot(unsigned long pfn, pgprot_t prot)
518{
Ard Biesheuvel825c43f2021-11-19 16:43:55 -0800519 pte_t pteval, *kmap_pte;
Thomas Gleixner298fa1a2020-11-03 10:27:18 +0100520 unsigned long vaddr;
521 int idx;
522
Thomas Gleixnerf3ba3c72020-11-18 20:48:44 +0100523 /*
524 * Disable migration so resulting virtual address is stable
Ingo Molnarf0953a12021-05-06 18:06:47 -0700525 * across preemption.
Thomas Gleixnerf3ba3c72020-11-18 20:48:44 +0100526 */
527 migrate_disable();
Thomas Gleixner298fa1a2020-11-03 10:27:18 +0100528 preempt_disable();
529 idx = arch_kmap_local_map_idx(kmap_local_idx_push(), pfn);
530 vaddr = __fix_to_virt(FIX_KMAP_BEGIN + idx);
Ard Biesheuvel825c43f2021-11-19 16:43:55 -0800531 kmap_pte = kmap_get_pte(vaddr, idx);
532 BUG_ON(!pte_none(*kmap_pte));
Thomas Gleixner298fa1a2020-11-03 10:27:18 +0100533 pteval = pfn_pte(pfn, prot);
Ard Biesheuvel825c43f2021-11-19 16:43:55 -0800534 arch_kmap_local_set_pte(&init_mm, vaddr, kmap_pte, pteval);
Thomas Gleixner298fa1a2020-11-03 10:27:18 +0100535 arch_kmap_local_post_map(vaddr, pteval);
Thomas Gleixner5fbda3e2020-11-18 20:48:43 +0100536 current->kmap_ctrl.pteval[kmap_local_idx()] = pteval;
Thomas Gleixner298fa1a2020-11-03 10:27:18 +0100537 preempt_enable();
538
539 return (void *)vaddr;
540}
541EXPORT_SYMBOL_GPL(__kmap_local_pfn_prot);
542
543void *__kmap_local_page_prot(struct page *page, pgprot_t prot)
544{
545 void *kmap;
546
Thomas Gleixner0e91a0c2020-11-18 20:48:40 +0100547 /*
548 * To broaden the usage of the actual kmap_local() machinery always map
549 * pages when debugging is enabled and the architecture has no problems
550 * with alias mappings.
551 */
552 if (!IS_ENABLED(CONFIG_DEBUG_KMAP_LOCAL_FORCE_MAP) && !PageHighMem(page))
Thomas Gleixner298fa1a2020-11-03 10:27:18 +0100553 return page_address(page);
554
555 /* Try kmap_high_get() if architecture has it enabled */
556 kmap = arch_kmap_local_high_get(page);
557 if (kmap)
558 return kmap;
559
560 return __kmap_local_pfn_prot(page_to_pfn(page), prot);
561}
562EXPORT_SYMBOL(__kmap_local_page_prot);
563
564void kunmap_local_indexed(void *vaddr)
565{
566 unsigned long addr = (unsigned long) vaddr & PAGE_MASK;
Ard Biesheuvel825c43f2021-11-19 16:43:55 -0800567 pte_t *kmap_pte;
Thomas Gleixner298fa1a2020-11-03 10:27:18 +0100568 int idx;
569
570 if (addr < __fix_to_virt(FIX_KMAP_END) ||
571 addr > __fix_to_virt(FIX_KMAP_BEGIN)) {
Thomas Gleixner0e91a0c2020-11-18 20:48:40 +0100572 if (IS_ENABLED(CONFIG_DEBUG_KMAP_LOCAL_FORCE_MAP)) {
573 /* This _should_ never happen! See above. */
574 WARN_ON_ONCE(1);
575 return;
576 }
Thomas Gleixner2a656ca2020-11-12 11:59:32 +0100577 /*
578 * Handle mappings which were obtained by kmap_high_get()
579 * first as the virtual address of such mappings is below
580 * PAGE_OFFSET. Warn for all other addresses which are in
581 * the user space part of the virtual address space.
582 */
583 if (!kmap_high_unmap_local(addr))
584 WARN_ON_ONCE(addr < PAGE_OFFSET);
Thomas Gleixner298fa1a2020-11-03 10:27:18 +0100585 return;
586 }
587
588 preempt_disable();
589 idx = arch_kmap_local_unmap_idx(kmap_local_idx(), addr);
590 WARN_ON_ONCE(addr != __fix_to_virt(FIX_KMAP_BEGIN + idx));
591
Ard Biesheuvel825c43f2021-11-19 16:43:55 -0800592 kmap_pte = kmap_get_pte(addr, idx);
Thomas Gleixner298fa1a2020-11-03 10:27:18 +0100593 arch_kmap_local_pre_unmap(addr);
Ard Biesheuvel825c43f2021-11-19 16:43:55 -0800594 pte_clear(&init_mm, addr, kmap_pte);
Thomas Gleixner298fa1a2020-11-03 10:27:18 +0100595 arch_kmap_local_post_unmap(addr);
Thomas Gleixner5fbda3e2020-11-18 20:48:43 +0100596 current->kmap_ctrl.pteval[kmap_local_idx()] = __pte(0);
Thomas Gleixner298fa1a2020-11-03 10:27:18 +0100597 kmap_local_idx_pop();
598 preempt_enable();
Thomas Gleixnerf3ba3c72020-11-18 20:48:44 +0100599 migrate_enable();
Thomas Gleixner298fa1a2020-11-03 10:27:18 +0100600}
601EXPORT_SYMBOL(kunmap_local_indexed);
Thomas Gleixner5fbda3e2020-11-18 20:48:43 +0100602
603/*
604 * Invoked before switch_to(). This is safe even when during or after
605 * clearing the maps an interrupt which needs a kmap_local happens because
606 * the task::kmap_ctrl.idx is not modified by the unmapping code so a
607 * nested kmap_local will use the next unused index and restore the index
608 * on unmap. The already cleared kmaps of the outgoing task are irrelevant
609 * because the interrupt context does not know about them. The same applies
610 * when scheduling back in for an interrupt which happens before the
611 * restore is complete.
612 */
613void __kmap_local_sched_out(void)
614{
615 struct task_struct *tsk = current;
Ard Biesheuvel825c43f2021-11-19 16:43:55 -0800616 pte_t *kmap_pte;
Thomas Gleixner5fbda3e2020-11-18 20:48:43 +0100617 int i;
618
619 /* Clear kmaps */
620 for (i = 0; i < tsk->kmap_ctrl.idx; i++) {
621 pte_t pteval = tsk->kmap_ctrl.pteval[i];
622 unsigned long addr;
623 int idx;
624
625 /* With debug all even slots are unmapped and act as guard */
Ira Weiny487cfad2021-03-24 21:37:53 -0700626 if (IS_ENABLED(CONFIG_DEBUG_KMAP_LOCAL) && !(i & 0x01)) {
Thomas Gleixner5fbda3e2020-11-18 20:48:43 +0100627 WARN_ON_ONCE(!pte_none(pteval));
628 continue;
629 }
630 if (WARN_ON_ONCE(pte_none(pteval)))
631 continue;
632
633 /*
634 * This is a horrible hack for XTENSA to calculate the
635 * coloured PTE index. Uses the PFN encoded into the pteval
636 * and the map index calculation because the actual mapped
637 * virtual address is not stored in task::kmap_ctrl.
638 * For any sane architecture this is optimized out.
639 */
640 idx = arch_kmap_local_map_idx(i, pte_pfn(pteval));
641
642 addr = __fix_to_virt(FIX_KMAP_BEGIN + idx);
Ard Biesheuvel825c43f2021-11-19 16:43:55 -0800643 kmap_pte = kmap_get_pte(addr, idx);
Thomas Gleixner5fbda3e2020-11-18 20:48:43 +0100644 arch_kmap_local_pre_unmap(addr);
Ard Biesheuvel825c43f2021-11-19 16:43:55 -0800645 pte_clear(&init_mm, addr, kmap_pte);
Thomas Gleixner5fbda3e2020-11-18 20:48:43 +0100646 arch_kmap_local_post_unmap(addr);
647 }
648}
649
650void __kmap_local_sched_in(void)
651{
652 struct task_struct *tsk = current;
Ard Biesheuvel825c43f2021-11-19 16:43:55 -0800653 pte_t *kmap_pte;
Thomas Gleixner5fbda3e2020-11-18 20:48:43 +0100654 int i;
655
656 /* Restore kmaps */
657 for (i = 0; i < tsk->kmap_ctrl.idx; i++) {
658 pte_t pteval = tsk->kmap_ctrl.pteval[i];
659 unsigned long addr;
660 int idx;
661
662 /* With debug all even slots are unmapped and act as guard */
Ira Weiny487cfad2021-03-24 21:37:53 -0700663 if (IS_ENABLED(CONFIG_DEBUG_KMAP_LOCAL) && !(i & 0x01)) {
Thomas Gleixner5fbda3e2020-11-18 20:48:43 +0100664 WARN_ON_ONCE(!pte_none(pteval));
665 continue;
666 }
667 if (WARN_ON_ONCE(pte_none(pteval)))
668 continue;
669
670 /* See comment in __kmap_local_sched_out() */
671 idx = arch_kmap_local_map_idx(i, pte_pfn(pteval));
672 addr = __fix_to_virt(FIX_KMAP_BEGIN + idx);
Ard Biesheuvel825c43f2021-11-19 16:43:55 -0800673 kmap_pte = kmap_get_pte(addr, idx);
674 set_pte_at(&init_mm, addr, kmap_pte, pteval);
Thomas Gleixner5fbda3e2020-11-18 20:48:43 +0100675 arch_kmap_local_post_map(addr, pteval);
676 }
677}
678
679void kmap_local_fork(struct task_struct *tsk)
680{
681 if (WARN_ON_ONCE(tsk->kmap_ctrl.idx))
682 memset(&tsk->kmap_ctrl, 0, sizeof(tsk->kmap_ctrl));
683}
684
Thomas Gleixner298fa1a2020-11-03 10:27:18 +0100685#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687#if defined(HASHED_PAGE_VIRTUAL)
688
689#define PA_HASH_ORDER 7
690
691/*
692 * Describes one page->virtual association
693 */
694struct page_address_map {
695 struct page *page;
696 void *virtual;
697 struct list_head list;
698};
699
Joonsoo Kima354e2c2012-12-11 16:01:23 -0800700static struct page_address_map page_address_maps[LAST_PKMAP];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701
702/*
703 * Hash table bucket
704 */
705static struct page_address_slot {
706 struct list_head lh; /* List of page_address_maps */
707 spinlock_t lock; /* Protect this bucket's list */
708} ____cacheline_aligned_in_smp page_address_htable[1<<PA_HASH_ORDER];
709
Ian Campbellf9918792011-08-17 13:45:09 +0100710static struct page_address_slot *page_slot(const struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711{
712 return &page_address_htable[hash_ptr(page, PA_HASH_ORDER)];
713}
714
Randy Dunlap77f60782008-03-19 17:00:42 -0700715/**
716 * page_address - get the mapped virtual address of a page
717 * @page: &struct page to get the virtual address of
718 *
719 * Returns the page's virtual address.
720 */
Ian Campbellf9918792011-08-17 13:45:09 +0100721void *page_address(const struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722{
723 unsigned long flags;
724 void *ret;
725 struct page_address_slot *pas;
726
727 if (!PageHighMem(page))
728 return lowmem_page_address(page);
729
730 pas = page_slot(page);
731 ret = NULL;
732 spin_lock_irqsave(&pas->lock, flags);
733 if (!list_empty(&pas->lh)) {
734 struct page_address_map *pam;
735
736 list_for_each_entry(pam, &pas->lh, list) {
737 if (pam->page == page) {
738 ret = pam->virtual;
739 goto done;
740 }
741 }
742 }
743done:
744 spin_unlock_irqrestore(&pas->lock, flags);
745 return ret;
746}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747EXPORT_SYMBOL(page_address);
748
Randy Dunlap77f60782008-03-19 17:00:42 -0700749/**
750 * set_page_address - set a page's virtual address
751 * @page: &struct page to set
752 * @virtual: virtual address to use
753 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754void set_page_address(struct page *page, void *virtual)
755{
756 unsigned long flags;
757 struct page_address_slot *pas;
758 struct page_address_map *pam;
759
760 BUG_ON(!PageHighMem(page));
761
762 pas = page_slot(page);
763 if (virtual) { /* Add */
Joonsoo Kima354e2c2012-12-11 16:01:23 -0800764 pam = &page_address_maps[PKMAP_NR((unsigned long)virtual)];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765 pam->page = page;
766 pam->virtual = virtual;
767
768 spin_lock_irqsave(&pas->lock, flags);
769 list_add_tail(&pam->list, &pas->lh);
770 spin_unlock_irqrestore(&pas->lock, flags);
771 } else { /* Remove */
772 spin_lock_irqsave(&pas->lock, flags);
773 list_for_each_entry(pam, &pas->lh, list) {
774 if (pam->page == page) {
775 list_del(&pam->list);
776 spin_unlock_irqrestore(&pas->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777 goto done;
778 }
779 }
780 spin_unlock_irqrestore(&pas->lock, flags);
781 }
782done:
783 return;
784}
785
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786void __init page_address_init(void)
787{
788 int i;
789
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790 for (i = 0; i < ARRAY_SIZE(page_address_htable); i++) {
791 INIT_LIST_HEAD(&page_address_htable[i].lh);
792 spin_lock_init(&page_address_htable[i].lock);
793 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700794}
795
Ira Weiny955cc772020-10-15 20:09:52 -0700796#endif /* defined(HASHED_PAGE_VIRTUAL) */