blob: 6ef8f5e05e7e5db835337728cd9f57e1f3d98f77 [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>
26#include <linux/blkdev.h>
27#include <linux/init.h>
28#include <linux/hash.h>
29#include <linux/highmem.h>
Jason Wesseleac790052010-08-05 09:22:24 -050030#include <linux/kgdb.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070031#include <asm/tlbflush.h>
Ingo Molnar186525b2019-11-29 08:17:25 +010032#include <linux/vmalloc.h>
Peter Zijlstraa8e23a22010-10-27 15:32:57 -070033
Linus Torvalds1da177e2005-04-16 15:20:36 -070034/*
35 * Virtual_count is not a pure "count".
36 * 0 means that it is not mapped, and has not been mapped
37 * since a TLB flush - it is usable.
38 * 1 means that there are no users, but it has been mapped
39 * since the last TLB flush - so we can't use it.
40 * n means that there are (n-1) current users of it.
41 */
42#ifdef CONFIG_HIGHMEM
Al Viro260b2362005-10-21 03:22:44 -040043
Max Filippov15de36a2014-08-06 16:08:23 -070044/*
45 * Architecture with aliasing data cache may define the following family of
46 * helper functions in its asm/highmem.h to control cache color of virtual
47 * addresses where physical memory pages are mapped by kmap.
48 */
49#ifndef get_pkmap_color
50
51/*
52 * Determine color of virtual address where the page should be mapped.
53 */
54static inline unsigned int get_pkmap_color(struct page *page)
55{
56 return 0;
57}
58#define get_pkmap_color get_pkmap_color
59
60/*
61 * Get next index for mapping inside PKMAP region for page with given color.
62 */
63static inline unsigned int get_next_pkmap_nr(unsigned int color)
64{
65 static unsigned int last_pkmap_nr;
66
67 last_pkmap_nr = (last_pkmap_nr + 1) & LAST_PKMAP_MASK;
68 return last_pkmap_nr;
69}
70
71/*
72 * Determine if page index inside PKMAP region (pkmap_nr) of given color
73 * has wrapped around PKMAP region end. When this happens an attempt to
74 * flush all unused PKMAP slots is made.
75 */
76static inline int no_more_pkmaps(unsigned int pkmap_nr, unsigned int color)
77{
78 return pkmap_nr == 0;
79}
80
81/*
82 * Get the number of PKMAP entries of the given color. If no free slot is
83 * found after checking that many entries, kmap will sleep waiting for
84 * someone to call kunmap and free PKMAP slot.
85 */
86static inline int get_pkmap_entries_count(unsigned int color)
87{
88 return LAST_PKMAP;
89}
90
91/*
92 * Get head of a wait queue for PKMAP entries of the given color.
93 * Wait queues for different mapping colors should be independent to avoid
94 * unnecessary wakeups caused by freeing of slots of other colors.
95 */
96static inline wait_queue_head_t *get_pkmap_wait_queue_head(unsigned int color)
97{
98 static DECLARE_WAIT_QUEUE_HEAD(pkmap_map_wait);
99
100 return &pkmap_map_wait;
101}
102#endif
103
Arun KSca79b0c2018-12-28 00:34:29 -0800104atomic_long_t _totalhigh_pages __read_mostly;
105EXPORT_SYMBOL(_totalhigh_pages);
Peter Zijlstra3e4d3af2010-10-26 14:21:51 -0700106
Thomas Gleixner13f876b2020-11-03 10:27:34 +0100107unsigned int __nr_free_highpages (void)
Christoph Lameterc1f60a52006-09-25 23:31:11 -0700108{
Joonsoo Kim33499bf2016-05-19 17:12:26 -0700109 struct zone *zone;
Christoph Lameterc1f60a52006-09-25 23:31:11 -0700110 unsigned int pages = 0;
111
Joonsoo Kim33499bf2016-05-19 17:12:26 -0700112 for_each_populated_zone(zone) {
113 if (is_highmem(zone))
114 pages += zone_page_state(zone, NR_FREE_PAGES);
Mel Gorman2a1e2742007-07-17 04:03:12 -0700115 }
Christoph Lameterc1f60a52006-09-25 23:31:11 -0700116
117 return pages;
118}
119
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120static int pkmap_count[LAST_PKMAP];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121static __cacheline_aligned_in_smp DEFINE_SPINLOCK(kmap_lock);
122
123pte_t * pkmap_page_table;
124
Nicolas Pitre3297e762009-03-04 22:49:41 -0500125/*
126 * Most architectures have no use for kmap_high_get(), so let's abstract
127 * the disabling of IRQ out of the locking in that case to save on a
128 * potential useless overhead.
129 */
130#ifdef ARCH_NEEDS_KMAP_HIGH_GET
131#define lock_kmap() spin_lock_irq(&kmap_lock)
132#define unlock_kmap() spin_unlock_irq(&kmap_lock)
133#define lock_kmap_any(flags) spin_lock_irqsave(&kmap_lock, flags)
134#define unlock_kmap_any(flags) spin_unlock_irqrestore(&kmap_lock, flags)
135#else
136#define lock_kmap() spin_lock(&kmap_lock)
137#define unlock_kmap() spin_unlock(&kmap_lock)
138#define lock_kmap_any(flags) \
139 do { spin_lock(&kmap_lock); (void)(flags); } while (0)
140#define unlock_kmap_any(flags) \
141 do { spin_unlock(&kmap_lock); (void)(flags); } while (0)
142#endif
143
Thomas Gleixner13f876b2020-11-03 10:27:34 +0100144struct page *__kmap_to_page(void *vaddr)
Mel Gorman5a178112012-07-31 16:45:02 -0700145{
146 unsigned long addr = (unsigned long)vaddr;
147
Will Deacon498c2282012-11-16 14:15:00 -0800148 if (addr >= PKMAP_ADDR(0) && addr < PKMAP_ADDR(LAST_PKMAP)) {
Joonsoo Kim4de22c02012-12-11 16:01:17 -0800149 int i = PKMAP_NR(addr);
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();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281 return (void*) vaddr;
282}
283
284EXPORT_SYMBOL(kmap_high);
285
Nicolas Pitre3297e762009-03-04 22:49:41 -0500286#ifdef ARCH_NEEDS_KMAP_HIGH_GET
287/**
288 * kmap_high_get - pin a highmem page into memory
289 * @page: &struct page to pin
290 *
291 * Returns the page's current virtual memory address, or NULL if no mapping
Uwe Kleine-König5e39df52010-01-25 21:38:09 +0100292 * exists. If and only if a non null address is returned then a
Nicolas Pitre3297e762009-03-04 22:49:41 -0500293 * matching call to kunmap_high() is necessary.
294 *
295 * This can be called from any context.
296 */
297void *kmap_high_get(struct page *page)
298{
299 unsigned long vaddr, flags;
300
301 lock_kmap_any(flags);
302 vaddr = (unsigned long)page_address(page);
303 if (vaddr) {
304 BUG_ON(pkmap_count[PKMAP_NR(vaddr)] < 1);
305 pkmap_count[PKMAP_NR(vaddr)]++;
306 }
307 unlock_kmap_any(flags);
308 return (void*) vaddr;
309}
310#endif
311
Randy Dunlap77f60782008-03-19 17:00:42 -0700312/**
Li Haifeng4e9dc5d2011-10-31 17:09:09 -0700313 * kunmap_high - unmap a highmem page into memory
Randy Dunlap77f60782008-03-19 17:00:42 -0700314 * @page: &struct page to unmap
Nicolas Pitre3297e762009-03-04 22:49:41 -0500315 *
316 * If ARCH_NEEDS_KMAP_HIGH_GET is not defined then this may be called
317 * only from user context.
Randy Dunlap77f60782008-03-19 17:00:42 -0700318 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800319void kunmap_high(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320{
321 unsigned long vaddr;
322 unsigned long nr;
Nicolas Pitre3297e762009-03-04 22:49:41 -0500323 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324 int need_wakeup;
Max Filippov15de36a2014-08-06 16:08:23 -0700325 unsigned int color = get_pkmap_color(page);
326 wait_queue_head_t *pkmap_map_wait;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327
Nicolas Pitre3297e762009-03-04 22:49:41 -0500328 lock_kmap_any(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329 vaddr = (unsigned long)page_address(page);
Eric Sesterhenn75babca2006-04-02 13:47:35 +0200330 BUG_ON(!vaddr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331 nr = PKMAP_NR(vaddr);
332
333 /*
334 * A count must never go down to zero
335 * without a TLB flush!
336 */
337 need_wakeup = 0;
338 switch (--pkmap_count[nr]) {
339 case 0:
340 BUG();
341 case 1:
342 /*
343 * Avoid an unnecessary wake_up() function call.
344 * The common case is pkmap_count[] == 1, but
345 * no waiters.
346 * The tasks queued in the wait-queue are guarded
347 * by both the lock in the wait-queue-head and by
348 * the kmap_lock. As the kmap_lock is held here,
349 * no need for the wait-queue-head's lock. Simply
350 * test if the queue is empty.
351 */
Max Filippov15de36a2014-08-06 16:08:23 -0700352 pkmap_map_wait = get_pkmap_wait_queue_head(color);
353 need_wakeup = waitqueue_active(pkmap_map_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354 }
Nicolas Pitre3297e762009-03-04 22:49:41 -0500355 unlock_kmap_any(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356
357 /* do wake-up, if needed, race-free outside of the spin lock */
358 if (need_wakeup)
Max Filippov15de36a2014-08-06 16:08:23 -0700359 wake_up(pkmap_map_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700360}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361EXPORT_SYMBOL(kunmap_high);
Matthew Wilcox (Oracle)0060ef32020-12-14 19:12:59 -0800362
363#ifdef CONFIG_TRANSPARENT_HUGEPAGE
364void zero_user_segments(struct page *page, unsigned start1, unsigned end1,
365 unsigned start2, unsigned end2)
366{
367 unsigned int i;
368
369 BUG_ON(end1 > page_size(page) || end2 > page_size(page));
370
OGAWA Hirofumi184cee52021-03-12 21:07:37 -0800371 if (start1 >= end1)
372 start1 = end1 = 0;
373 if (start2 >= end2)
374 start2 = end2 = 0;
375
Matthew Wilcox (Oracle)0060ef32020-12-14 19:12:59 -0800376 for (i = 0; i < compound_nr(page); i++) {
377 void *kaddr = NULL;
378
Matthew Wilcox (Oracle)0060ef32020-12-14 19:12:59 -0800379 if (start1 >= PAGE_SIZE) {
380 start1 -= PAGE_SIZE;
381 end1 -= PAGE_SIZE;
382 } else {
383 unsigned this_end = min_t(unsigned, end1, PAGE_SIZE);
384
OGAWA Hirofumi184cee52021-03-12 21:07:37 -0800385 if (end1 > start1) {
386 kaddr = kmap_atomic(page + i);
Matthew Wilcox (Oracle)0060ef32020-12-14 19:12:59 -0800387 memset(kaddr + start1, 0, this_end - start1);
OGAWA Hirofumi184cee52021-03-12 21:07:37 -0800388 }
Matthew Wilcox (Oracle)0060ef32020-12-14 19:12:59 -0800389 end1 -= this_end;
390 start1 = 0;
391 }
392
393 if (start2 >= PAGE_SIZE) {
394 start2 -= PAGE_SIZE;
395 end2 -= PAGE_SIZE;
396 } else {
397 unsigned this_end = min_t(unsigned, end2, PAGE_SIZE);
398
OGAWA Hirofumi184cee52021-03-12 21:07:37 -0800399 if (end2 > start2) {
400 if (!kaddr)
401 kaddr = kmap_atomic(page + i);
Matthew Wilcox (Oracle)0060ef32020-12-14 19:12:59 -0800402 memset(kaddr + start2, 0, this_end - start2);
OGAWA Hirofumi184cee52021-03-12 21:07:37 -0800403 }
Matthew Wilcox (Oracle)0060ef32020-12-14 19:12:59 -0800404 end2 -= this_end;
405 start2 = 0;
406 }
407
408 if (kaddr) {
409 kunmap_atomic(kaddr);
410 flush_dcache_page(page + i);
411 }
412
413 if (!end1 && !end2)
414 break;
415 }
416
417 BUG_ON((start1 | start2 | end1 | end2) != 0);
418}
419EXPORT_SYMBOL(zero_user_segments);
420#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
Thomas Gleixner298fa1a2020-11-03 10:27:18 +0100421#endif /* CONFIG_HIGHMEM */
422
423#ifdef CONFIG_KMAP_LOCAL
424
425#include <asm/kmap_size.h>
426
Thomas Gleixner389755c2020-11-03 10:27:19 +0100427/*
Thomas Gleixner6e799cb2020-11-18 20:48:39 +0100428 * With DEBUG_KMAP_LOCAL the stack depth is doubled and every second
Thomas Gleixner389755c2020-11-03 10:27:19 +0100429 * slot is unused which acts as a guard page
430 */
Thomas Gleixner6e799cb2020-11-18 20:48:39 +0100431#ifdef CONFIG_DEBUG_KMAP_LOCAL
Thomas Gleixner389755c2020-11-03 10:27:19 +0100432# define KM_INCR 2
433#else
434# define KM_INCR 1
435#endif
436
Thomas Gleixner298fa1a2020-11-03 10:27:18 +0100437static inline int kmap_local_idx_push(void)
438{
Thomas Gleixner298fa1a2020-11-03 10:27:18 +0100439 WARN_ON_ONCE(in_irq() && !irqs_disabled());
Thomas Gleixner5fbda3e2020-11-18 20:48:43 +0100440 current->kmap_ctrl.idx += KM_INCR;
441 BUG_ON(current->kmap_ctrl.idx >= KM_MAX_IDX);
442 return current->kmap_ctrl.idx - 1;
Thomas Gleixner298fa1a2020-11-03 10:27:18 +0100443}
444
445static inline int kmap_local_idx(void)
446{
Thomas Gleixner5fbda3e2020-11-18 20:48:43 +0100447 return current->kmap_ctrl.idx - 1;
Thomas Gleixner298fa1a2020-11-03 10:27:18 +0100448}
449
450static inline void kmap_local_idx_pop(void)
451{
Thomas Gleixner5fbda3e2020-11-18 20:48:43 +0100452 current->kmap_ctrl.idx -= KM_INCR;
453 BUG_ON(current->kmap_ctrl.idx < 0);
Thomas Gleixner298fa1a2020-11-03 10:27:18 +0100454}
455
456#ifndef arch_kmap_local_post_map
457# define arch_kmap_local_post_map(vaddr, pteval) do { } while (0)
458#endif
Thomas Gleixner3c1016b2020-11-03 10:27:31 +0100459
Thomas Gleixner298fa1a2020-11-03 10:27:18 +0100460#ifndef arch_kmap_local_pre_unmap
461# define arch_kmap_local_pre_unmap(vaddr) do { } while (0)
462#endif
463
464#ifndef arch_kmap_local_post_unmap
465# define arch_kmap_local_post_unmap(vaddr) do { } while (0)
466#endif
467
468#ifndef arch_kmap_local_map_idx
469#define arch_kmap_local_map_idx(idx, pfn) kmap_local_calc_idx(idx)
470#endif
471
472#ifndef arch_kmap_local_unmap_idx
473#define arch_kmap_local_unmap_idx(idx, vaddr) kmap_local_calc_idx(idx)
474#endif
475
476#ifndef arch_kmap_local_high_get
477static inline void *arch_kmap_local_high_get(struct page *page)
478{
479 return NULL;
480}
481#endif
482
Thomas Gleixnera1dce7f2021-01-23 21:02:02 -0800483#ifndef arch_kmap_local_set_pte
484#define arch_kmap_local_set_pte(mm, vaddr, ptep, ptev) \
485 set_pte_at(mm, vaddr, ptep, ptev)
486#endif
487
Thomas Gleixner298fa1a2020-11-03 10:27:18 +0100488/* Unmap a local mapping which was obtained by kmap_high_get() */
Thomas Gleixner2a656ca2020-11-12 11:59:32 +0100489static inline bool kmap_high_unmap_local(unsigned long vaddr)
Thomas Gleixner298fa1a2020-11-03 10:27:18 +0100490{
491#ifdef ARCH_NEEDS_KMAP_HIGH_GET
Thomas Gleixner2a656ca2020-11-12 11:59:32 +0100492 if (vaddr >= PKMAP_ADDR(0) && vaddr < PKMAP_ADDR(LAST_PKMAP)) {
Thomas Gleixner298fa1a2020-11-03 10:27:18 +0100493 kunmap_high(pte_page(pkmap_page_table[PKMAP_NR(vaddr)]));
Thomas Gleixner2a656ca2020-11-12 11:59:32 +0100494 return true;
495 }
Thomas Gleixner298fa1a2020-11-03 10:27:18 +0100496#endif
Thomas Gleixner2a656ca2020-11-12 11:59:32 +0100497 return false;
Thomas Gleixner298fa1a2020-11-03 10:27:18 +0100498}
499
500static inline int kmap_local_calc_idx(int idx)
501{
502 return idx + KM_MAX_IDX * smp_processor_id();
503}
504
505static pte_t *__kmap_pte;
506
507static pte_t *kmap_get_pte(void)
508{
509 if (!__kmap_pte)
510 __kmap_pte = virt_to_kpte(__fix_to_virt(FIX_KMAP_BEGIN));
511 return __kmap_pte;
512}
513
514void *__kmap_local_pfn_prot(unsigned long pfn, pgprot_t prot)
515{
516 pte_t pteval, *kmap_pte = kmap_get_pte();
517 unsigned long vaddr;
518 int idx;
519
Thomas Gleixnerf3ba3c72020-11-18 20:48:44 +0100520 /*
521 * Disable migration so resulting virtual address is stable
522 * accross preemption.
523 */
524 migrate_disable();
Thomas Gleixner298fa1a2020-11-03 10:27:18 +0100525 preempt_disable();
526 idx = arch_kmap_local_map_idx(kmap_local_idx_push(), pfn);
527 vaddr = __fix_to_virt(FIX_KMAP_BEGIN + idx);
528 BUG_ON(!pte_none(*(kmap_pte - idx)));
529 pteval = pfn_pte(pfn, prot);
Thomas Gleixnera1dce7f2021-01-23 21:02:02 -0800530 arch_kmap_local_set_pte(&init_mm, vaddr, kmap_pte - idx, pteval);
Thomas Gleixner298fa1a2020-11-03 10:27:18 +0100531 arch_kmap_local_post_map(vaddr, pteval);
Thomas Gleixner5fbda3e2020-11-18 20:48:43 +0100532 current->kmap_ctrl.pteval[kmap_local_idx()] = pteval;
Thomas Gleixner298fa1a2020-11-03 10:27:18 +0100533 preempt_enable();
534
535 return (void *)vaddr;
536}
537EXPORT_SYMBOL_GPL(__kmap_local_pfn_prot);
538
539void *__kmap_local_page_prot(struct page *page, pgprot_t prot)
540{
541 void *kmap;
542
Thomas Gleixner0e91a0c2020-11-18 20:48:40 +0100543 /*
544 * To broaden the usage of the actual kmap_local() machinery always map
545 * pages when debugging is enabled and the architecture has no problems
546 * with alias mappings.
547 */
548 if (!IS_ENABLED(CONFIG_DEBUG_KMAP_LOCAL_FORCE_MAP) && !PageHighMem(page))
Thomas Gleixner298fa1a2020-11-03 10:27:18 +0100549 return page_address(page);
550
551 /* Try kmap_high_get() if architecture has it enabled */
552 kmap = arch_kmap_local_high_get(page);
553 if (kmap)
554 return kmap;
555
556 return __kmap_local_pfn_prot(page_to_pfn(page), prot);
557}
558EXPORT_SYMBOL(__kmap_local_page_prot);
559
560void kunmap_local_indexed(void *vaddr)
561{
562 unsigned long addr = (unsigned long) vaddr & PAGE_MASK;
563 pte_t *kmap_pte = kmap_get_pte();
564 int idx;
565
566 if (addr < __fix_to_virt(FIX_KMAP_END) ||
567 addr > __fix_to_virt(FIX_KMAP_BEGIN)) {
Thomas Gleixner0e91a0c2020-11-18 20:48:40 +0100568 if (IS_ENABLED(CONFIG_DEBUG_KMAP_LOCAL_FORCE_MAP)) {
569 /* This _should_ never happen! See above. */
570 WARN_ON_ONCE(1);
571 return;
572 }
Thomas Gleixner2a656ca2020-11-12 11:59:32 +0100573 /*
574 * Handle mappings which were obtained by kmap_high_get()
575 * first as the virtual address of such mappings is below
576 * PAGE_OFFSET. Warn for all other addresses which are in
577 * the user space part of the virtual address space.
578 */
579 if (!kmap_high_unmap_local(addr))
580 WARN_ON_ONCE(addr < PAGE_OFFSET);
Thomas Gleixner298fa1a2020-11-03 10:27:18 +0100581 return;
582 }
583
584 preempt_disable();
585 idx = arch_kmap_local_unmap_idx(kmap_local_idx(), addr);
586 WARN_ON_ONCE(addr != __fix_to_virt(FIX_KMAP_BEGIN + idx));
587
588 arch_kmap_local_pre_unmap(addr);
589 pte_clear(&init_mm, addr, kmap_pte - idx);
590 arch_kmap_local_post_unmap(addr);
Thomas Gleixner5fbda3e2020-11-18 20:48:43 +0100591 current->kmap_ctrl.pteval[kmap_local_idx()] = __pte(0);
Thomas Gleixner298fa1a2020-11-03 10:27:18 +0100592 kmap_local_idx_pop();
593 preempt_enable();
Thomas Gleixnerf3ba3c72020-11-18 20:48:44 +0100594 migrate_enable();
Thomas Gleixner298fa1a2020-11-03 10:27:18 +0100595}
596EXPORT_SYMBOL(kunmap_local_indexed);
Thomas Gleixner5fbda3e2020-11-18 20:48:43 +0100597
598/*
599 * Invoked before switch_to(). This is safe even when during or after
600 * clearing the maps an interrupt which needs a kmap_local happens because
601 * the task::kmap_ctrl.idx is not modified by the unmapping code so a
602 * nested kmap_local will use the next unused index and restore the index
603 * on unmap. The already cleared kmaps of the outgoing task are irrelevant
604 * because the interrupt context does not know about them. The same applies
605 * when scheduling back in for an interrupt which happens before the
606 * restore is complete.
607 */
608void __kmap_local_sched_out(void)
609{
610 struct task_struct *tsk = current;
611 pte_t *kmap_pte = kmap_get_pte();
612 int i;
613
614 /* Clear kmaps */
615 for (i = 0; i < tsk->kmap_ctrl.idx; i++) {
616 pte_t pteval = tsk->kmap_ctrl.pteval[i];
617 unsigned long addr;
618 int idx;
619
620 /* With debug all even slots are unmapped and act as guard */
Ira Weiny487cfad2021-03-24 21:37:53 -0700621 if (IS_ENABLED(CONFIG_DEBUG_KMAP_LOCAL) && !(i & 0x01)) {
Thomas Gleixner5fbda3e2020-11-18 20:48:43 +0100622 WARN_ON_ONCE(!pte_none(pteval));
623 continue;
624 }
625 if (WARN_ON_ONCE(pte_none(pteval)))
626 continue;
627
628 /*
629 * This is a horrible hack for XTENSA to calculate the
630 * coloured PTE index. Uses the PFN encoded into the pteval
631 * and the map index calculation because the actual mapped
632 * virtual address is not stored in task::kmap_ctrl.
633 * For any sane architecture this is optimized out.
634 */
635 idx = arch_kmap_local_map_idx(i, pte_pfn(pteval));
636
637 addr = __fix_to_virt(FIX_KMAP_BEGIN + idx);
638 arch_kmap_local_pre_unmap(addr);
639 pte_clear(&init_mm, addr, kmap_pte - idx);
640 arch_kmap_local_post_unmap(addr);
641 }
642}
643
644void __kmap_local_sched_in(void)
645{
646 struct task_struct *tsk = current;
647 pte_t *kmap_pte = kmap_get_pte();
648 int i;
649
650 /* Restore kmaps */
651 for (i = 0; i < tsk->kmap_ctrl.idx; i++) {
652 pte_t pteval = tsk->kmap_ctrl.pteval[i];
653 unsigned long addr;
654 int idx;
655
656 /* With debug all even slots are unmapped and act as guard */
Ira Weiny487cfad2021-03-24 21:37:53 -0700657 if (IS_ENABLED(CONFIG_DEBUG_KMAP_LOCAL) && !(i & 0x01)) {
Thomas Gleixner5fbda3e2020-11-18 20:48:43 +0100658 WARN_ON_ONCE(!pte_none(pteval));
659 continue;
660 }
661 if (WARN_ON_ONCE(pte_none(pteval)))
662 continue;
663
664 /* See comment in __kmap_local_sched_out() */
665 idx = arch_kmap_local_map_idx(i, pte_pfn(pteval));
666 addr = __fix_to_virt(FIX_KMAP_BEGIN + idx);
667 set_pte_at(&init_mm, addr, kmap_pte - idx, pteval);
668 arch_kmap_local_post_map(addr, pteval);
669 }
670}
671
672void kmap_local_fork(struct task_struct *tsk)
673{
674 if (WARN_ON_ONCE(tsk->kmap_ctrl.idx))
675 memset(&tsk->kmap_ctrl, 0, sizeof(tsk->kmap_ctrl));
676}
677
Thomas Gleixner298fa1a2020-11-03 10:27:18 +0100678#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680#if defined(HASHED_PAGE_VIRTUAL)
681
682#define PA_HASH_ORDER 7
683
684/*
685 * Describes one page->virtual association
686 */
687struct page_address_map {
688 struct page *page;
689 void *virtual;
690 struct list_head list;
691};
692
Joonsoo Kima354e2c2012-12-11 16:01:23 -0800693static struct page_address_map page_address_maps[LAST_PKMAP];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694
695/*
696 * Hash table bucket
697 */
698static struct page_address_slot {
699 struct list_head lh; /* List of page_address_maps */
700 spinlock_t lock; /* Protect this bucket's list */
701} ____cacheline_aligned_in_smp page_address_htable[1<<PA_HASH_ORDER];
702
Ian Campbellf9918792011-08-17 13:45:09 +0100703static struct page_address_slot *page_slot(const struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704{
705 return &page_address_htable[hash_ptr(page, PA_HASH_ORDER)];
706}
707
Randy Dunlap77f60782008-03-19 17:00:42 -0700708/**
709 * page_address - get the mapped virtual address of a page
710 * @page: &struct page to get the virtual address of
711 *
712 * Returns the page's virtual address.
713 */
Ian Campbellf9918792011-08-17 13:45:09 +0100714void *page_address(const struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715{
716 unsigned long flags;
717 void *ret;
718 struct page_address_slot *pas;
719
720 if (!PageHighMem(page))
721 return lowmem_page_address(page);
722
723 pas = page_slot(page);
724 ret = NULL;
725 spin_lock_irqsave(&pas->lock, flags);
726 if (!list_empty(&pas->lh)) {
727 struct page_address_map *pam;
728
729 list_for_each_entry(pam, &pas->lh, list) {
730 if (pam->page == page) {
731 ret = pam->virtual;
732 goto done;
733 }
734 }
735 }
736done:
737 spin_unlock_irqrestore(&pas->lock, flags);
738 return ret;
739}
740
741EXPORT_SYMBOL(page_address);
742
Randy Dunlap77f60782008-03-19 17:00:42 -0700743/**
744 * set_page_address - set a page's virtual address
745 * @page: &struct page to set
746 * @virtual: virtual address to use
747 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748void set_page_address(struct page *page, void *virtual)
749{
750 unsigned long flags;
751 struct page_address_slot *pas;
752 struct page_address_map *pam;
753
754 BUG_ON(!PageHighMem(page));
755
756 pas = page_slot(page);
757 if (virtual) { /* Add */
Joonsoo Kima354e2c2012-12-11 16:01:23 -0800758 pam = &page_address_maps[PKMAP_NR((unsigned long)virtual)];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759 pam->page = page;
760 pam->virtual = virtual;
761
762 spin_lock_irqsave(&pas->lock, flags);
763 list_add_tail(&pam->list, &pas->lh);
764 spin_unlock_irqrestore(&pas->lock, flags);
765 } else { /* Remove */
766 spin_lock_irqsave(&pas->lock, flags);
767 list_for_each_entry(pam, &pas->lh, list) {
768 if (pam->page == page) {
769 list_del(&pam->list);
770 spin_unlock_irqrestore(&pas->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700771 goto done;
772 }
773 }
774 spin_unlock_irqrestore(&pas->lock, flags);
775 }
776done:
777 return;
778}
779
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780void __init page_address_init(void)
781{
782 int i;
783
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784 for (i = 0; i < ARRAY_SIZE(page_address_htable); i++) {
785 INIT_LIST_HEAD(&page_address_htable[i].lh);
786 spin_lock_init(&page_address_htable[i].lock);
787 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788}
789
Ira Weiny955cc772020-10-15 20:09:52 -0700790#endif /* defined(HASHED_PAGE_VIRTUAL) */