]> git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/blame - mm/highmem.c
x86/speculation/mmio: Enable CPU Fill buffer clearing on idle
[mirror_ubuntu-jammy-kernel.git] / mm / highmem.c
CommitLineData
b2441318 1// SPDX-License-Identifier: GPL-2.0
1da177e4
LT
2/*
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>
b95f1b31 21#include <linux/export.h>
1da177e4
LT
22#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>
eac79005 30#include <linux/kgdb.h>
1da177e4 31#include <asm/tlbflush.h>
186525bd 32#include <linux/vmalloc.h>
a8e23a29 33
1da177e4
LT
34/*
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
260b2367 43
15de36a4
MF
44/*
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
ca79b0c2
AK
104atomic_long_t _totalhigh_pages __read_mostly;
105EXPORT_SYMBOL(_totalhigh_pages);
3e4d3af5 106
9727688d 107unsigned int __nr_free_highpages(void)
c1f60a5a 108{
33499bfe 109 struct zone *zone;
c1f60a5a
CL
110 unsigned int pages = 0;
111
33499bfe
JK
112 for_each_populated_zone(zone) {
113 if (is_highmem(zone))
114 pages += zone_page_state(zone, NR_FREE_PAGES);
2a1e274a 115 }
c1f60a5a
CL
116
117 return pages;
118}
119
1da177e4 120static int pkmap_count[LAST_PKMAP];
1da177e4
LT
121static __cacheline_aligned_in_smp DEFINE_SPINLOCK(kmap_lock);
122
9727688d 123pte_t *pkmap_page_table;
1da177e4 124
3297e760
NP
125/*
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
13f876ba 144struct page *__kmap_to_page(void *vaddr)
5a178119
MG
145{
146 unsigned long addr = (unsigned long)vaddr;
147
498c2280 148 if (addr >= PKMAP_ADDR(0) && addr < PKMAP_ADDR(LAST_PKMAP)) {
4de22c05 149 int i = PKMAP_NR(addr);
9727688d 150
5a178119
MG
151 return pte_page(pkmap_page_table[i]);
152 }
153
154 return virt_to_page(addr);
155}
13f876ba 156EXPORT_SYMBOL(__kmap_to_page);
5a178119 157
1da177e4
LT
158static void flush_all_zero_pkmaps(void)
159{
160 int i;
5843d9a4 161 int need_flush = 0;
1da177e4
LT
162
163 flush_cache_kmaps();
164
165 for (i = 0; i < LAST_PKMAP; i++) {
166 struct page *page;
167
168 /*
169 * zero means we don't have anything to do,
170 * >1 means that it is still in use. Only
171 * a count of 1 means that it is free but
172 * needs to be unmapped
173 */
174 if (pkmap_count[i] != 1)
175 continue;
176 pkmap_count[i] = 0;
177
178 /* sanity check */
75babcac 179 BUG_ON(pte_none(pkmap_page_table[i]));
1da177e4
LT
180
181 /*
182 * Don't need an atomic fetch-and-clear op here;
183 * no-one has the page mapped, and cannot get at
184 * its virtual address (and hence PTE) without first
185 * getting the kmap_lock (which is held here).
186 * So no dangers, even with speculative execution.
187 */
188 page = pte_page(pkmap_page_table[i]);
eb2db439 189 pte_clear(&init_mm, PKMAP_ADDR(i), &pkmap_page_table[i]);
1da177e4
LT
190
191 set_page_address(page, NULL);
5843d9a4 192 need_flush = 1;
1da177e4 193 }
5843d9a4
NP
194 if (need_flush)
195 flush_tlb_kernel_range(PKMAP_ADDR(0), PKMAP_ADDR(LAST_PKMAP));
1da177e4
LT
196}
197
13f876ba 198void __kmap_flush_unused(void)
ce6234b5 199{
3297e760 200 lock_kmap();
ce6234b5 201 flush_all_zero_pkmaps();
3297e760 202 unlock_kmap();
ce6234b5
JF
203}
204
1da177e4
LT
205static inline unsigned long map_new_virtual(struct page *page)
206{
207 unsigned long vaddr;
208 int count;
15de36a4
MF
209 unsigned int last_pkmap_nr;
210 unsigned int color = get_pkmap_color(page);
1da177e4
LT
211
212start:
15de36a4 213 count = get_pkmap_entries_count(color);
1da177e4
LT
214 /* Find an empty entry */
215 for (;;) {
15de36a4
MF
216 last_pkmap_nr = get_next_pkmap_nr(color);
217 if (no_more_pkmaps(last_pkmap_nr, color)) {
1da177e4 218 flush_all_zero_pkmaps();
15de36a4 219 count = get_pkmap_entries_count(color);
1da177e4
LT
220 }
221 if (!pkmap_count[last_pkmap_nr])
222 break; /* Found a usable entry */
223 if (--count)
224 continue;
225
226 /*
227 * Sleep for somebody else to unmap their entries
228 */
229 {
230 DECLARE_WAITQUEUE(wait, current);
15de36a4
MF
231 wait_queue_head_t *pkmap_map_wait =
232 get_pkmap_wait_queue_head(color);
1da177e4
LT
233
234 __set_current_state(TASK_UNINTERRUPTIBLE);
15de36a4 235 add_wait_queue(pkmap_map_wait, &wait);
3297e760 236 unlock_kmap();
1da177e4 237 schedule();
15de36a4 238 remove_wait_queue(pkmap_map_wait, &wait);
3297e760 239 lock_kmap();
1da177e4
LT
240
241 /* Somebody else might have mapped it while we slept */
242 if (page_address(page))
243 return (unsigned long)page_address(page);
244
245 /* Re-start */
246 goto start;
247 }
248 }
249 vaddr = PKMAP_ADDR(last_pkmap_nr);
250 set_pte_at(&init_mm, vaddr,
251 &(pkmap_page_table[last_pkmap_nr]), mk_pte(page, kmap_prot));
252
253 pkmap_count[last_pkmap_nr] = 1;
254 set_page_address(page, (void *)vaddr);
255
256 return vaddr;
257}
258
77f6078a
RD
259/**
260 * kmap_high - map a highmem page into memory
261 * @page: &struct page to map
262 *
263 * Returns the page's virtual memory address.
264 *
265 * We cannot call this from interrupts, as it may block.
266 */
920c7a5d 267void *kmap_high(struct page *page)
1da177e4
LT
268{
269 unsigned long vaddr;
270
271 /*
272 * For highmem pages, we can't trust "virtual" until
273 * after we have the lock.
1da177e4 274 */
3297e760 275 lock_kmap();
1da177e4
LT
276 vaddr = (unsigned long)page_address(page);
277 if (!vaddr)
278 vaddr = map_new_virtual(page);
279 pkmap_count[PKMAP_NR(vaddr)]++;
75babcac 280 BUG_ON(pkmap_count[PKMAP_NR(vaddr)] < 2);
3297e760 281 unlock_kmap();
9727688d 282 return (void *) vaddr;
1da177e4 283}
1da177e4
LT
284EXPORT_SYMBOL(kmap_high);
285
3297e760
NP
286#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
5e39df56 292 * exists. If and only if a non null address is returned then a
3297e760
NP
293 * 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);
9727688d 308 return (void *) vaddr;
3297e760
NP
309}
310#endif
311
77f6078a 312/**
4e9dc5df 313 * kunmap_high - unmap a highmem page into memory
77f6078a 314 * @page: &struct page to unmap
3297e760
NP
315 *
316 * If ARCH_NEEDS_KMAP_HIGH_GET is not defined then this may be called
317 * only from user context.
77f6078a 318 */
920c7a5d 319void kunmap_high(struct page *page)
1da177e4
LT
320{
321 unsigned long vaddr;
322 unsigned long nr;
3297e760 323 unsigned long flags;
1da177e4 324 int need_wakeup;
15de36a4
MF
325 unsigned int color = get_pkmap_color(page);
326 wait_queue_head_t *pkmap_map_wait;
1da177e4 327
3297e760 328 lock_kmap_any(flags);
1da177e4 329 vaddr = (unsigned long)page_address(page);
75babcac 330 BUG_ON(!vaddr);
1da177e4
LT
331 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 */
15de36a4
MF
352 pkmap_map_wait = get_pkmap_wait_queue_head(color);
353 need_wakeup = waitqueue_active(pkmap_map_wait);
1da177e4 354 }
3297e760 355 unlock_kmap_any(flags);
1da177e4
LT
356
357 /* do wake-up, if needed, race-free outside of the spin lock */
358 if (need_wakeup)
15de36a4 359 wake_up(pkmap_map_wait);
1da177e4 360}
1da177e4 361EXPORT_SYMBOL(kunmap_high);
0060ef3b
MWO
362
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
184cee51
OH
371 if (start1 >= end1)
372 start1 = end1 = 0;
373 if (start2 >= end2)
374 start2 = end2 = 0;
375
0060ef3b
MWO
376 for (i = 0; i < compound_nr(page); i++) {
377 void *kaddr = NULL;
378
0060ef3b
MWO
379 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
184cee51
OH
385 if (end1 > start1) {
386 kaddr = kmap_atomic(page + i);
0060ef3b 387 memset(kaddr + start1, 0, this_end - start1);
184cee51 388 }
0060ef3b
MWO
389 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
184cee51
OH
399 if (end2 > start2) {
400 if (!kaddr)
401 kaddr = kmap_atomic(page + i);
0060ef3b 402 memset(kaddr + start2, 0, this_end - start2);
184cee51 403 }
0060ef3b
MWO
404 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 */
298fa1ad
TG
421#endif /* CONFIG_HIGHMEM */
422
423#ifdef CONFIG_KMAP_LOCAL
424
425#include <asm/kmap_size.h>
426
389755c2 427/*
6e799cb6 428 * With DEBUG_KMAP_LOCAL the stack depth is doubled and every second
389755c2
TG
429 * slot is unused which acts as a guard page
430 */
6e799cb6 431#ifdef CONFIG_DEBUG_KMAP_LOCAL
389755c2
TG
432# define KM_INCR 2
433#else
434# define KM_INCR 1
435#endif
436
298fa1ad
TG
437static inline int kmap_local_idx_push(void)
438{
ea0eafea 439 WARN_ON_ONCE(in_hardirq() && !irqs_disabled());
5fbda3ec
TG
440 current->kmap_ctrl.idx += KM_INCR;
441 BUG_ON(current->kmap_ctrl.idx >= KM_MAX_IDX);
442 return current->kmap_ctrl.idx - 1;
298fa1ad
TG
443}
444
445static inline int kmap_local_idx(void)
446{
5fbda3ec 447 return current->kmap_ctrl.idx - 1;
298fa1ad
TG
448}
449
450static inline void kmap_local_idx_pop(void)
451{
5fbda3ec
TG
452 current->kmap_ctrl.idx -= KM_INCR;
453 BUG_ON(current->kmap_ctrl.idx < 0);
298fa1ad
TG
454}
455
456#ifndef arch_kmap_local_post_map
457# define arch_kmap_local_post_map(vaddr, pteval) do { } while (0)
458#endif
3c1016b5 459
298fa1ad
TG
460#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
a1dce7fd
TG
483#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
298fa1ad 488/* Unmap a local mapping which was obtained by kmap_high_get() */
2a656cad 489static inline bool kmap_high_unmap_local(unsigned long vaddr)
298fa1ad
TG
490{
491#ifdef ARCH_NEEDS_KMAP_HIGH_GET
2a656cad 492 if (vaddr >= PKMAP_ADDR(0) && vaddr < PKMAP_ADDR(LAST_PKMAP)) {
298fa1ad 493 kunmap_high(pte_page(pkmap_page_table[PKMAP_NR(vaddr)]));
2a656cad
TG
494 return true;
495 }
298fa1ad 496#endif
2a656cad 497 return false;
298fa1ad
TG
498}
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
1fac7473 507static pte_t *kmap_get_pte(unsigned long vaddr, int idx)
298fa1ad 508{
1fac7473
AB
509 if (IS_ENABLED(CONFIG_KMAP_LOCAL_NON_LINEAR_PTE_ARRAY))
510 /*
511 * Set by the arch if __kmap_pte[-idx] does not produce
512 * the correct entry.
513 */
514 return virt_to_kpte(vaddr);
298fa1ad
TG
515 if (!__kmap_pte)
516 __kmap_pte = virt_to_kpte(__fix_to_virt(FIX_KMAP_BEGIN));
1fac7473 517 return &__kmap_pte[-idx];
298fa1ad
TG
518}
519
520void *__kmap_local_pfn_prot(unsigned long pfn, pgprot_t prot)
521{
1fac7473 522 pte_t pteval, *kmap_pte;
298fa1ad
TG
523 unsigned long vaddr;
524 int idx;
525
f3ba3c71
TG
526 /*
527 * Disable migration so resulting virtual address is stable
f0953a1b 528 * across preemption.
f3ba3c71
TG
529 */
530 migrate_disable();
298fa1ad
TG
531 preempt_disable();
532 idx = arch_kmap_local_map_idx(kmap_local_idx_push(), pfn);
533 vaddr = __fix_to_virt(FIX_KMAP_BEGIN + idx);
1fac7473
AB
534 kmap_pte = kmap_get_pte(vaddr, idx);
535 BUG_ON(!pte_none(*kmap_pte));
298fa1ad 536 pteval = pfn_pte(pfn, prot);
1fac7473 537 arch_kmap_local_set_pte(&init_mm, vaddr, kmap_pte, pteval);
298fa1ad 538 arch_kmap_local_post_map(vaddr, pteval);
5fbda3ec 539 current->kmap_ctrl.pteval[kmap_local_idx()] = pteval;
298fa1ad
TG
540 preempt_enable();
541
542 return (void *)vaddr;
543}
544EXPORT_SYMBOL_GPL(__kmap_local_pfn_prot);
545
546void *__kmap_local_page_prot(struct page *page, pgprot_t prot)
547{
548 void *kmap;
549
0e91a0c6
TG
550 /*
551 * To broaden the usage of the actual kmap_local() machinery always map
552 * pages when debugging is enabled and the architecture has no problems
553 * with alias mappings.
554 */
555 if (!IS_ENABLED(CONFIG_DEBUG_KMAP_LOCAL_FORCE_MAP) && !PageHighMem(page))
298fa1ad
TG
556 return page_address(page);
557
558 /* Try kmap_high_get() if architecture has it enabled */
559 kmap = arch_kmap_local_high_get(page);
560 if (kmap)
561 return kmap;
562
563 return __kmap_local_pfn_prot(page_to_pfn(page), prot);
564}
565EXPORT_SYMBOL(__kmap_local_page_prot);
566
567void kunmap_local_indexed(void *vaddr)
568{
569 unsigned long addr = (unsigned long) vaddr & PAGE_MASK;
1fac7473 570 pte_t *kmap_pte;
298fa1ad
TG
571 int idx;
572
573 if (addr < __fix_to_virt(FIX_KMAP_END) ||
574 addr > __fix_to_virt(FIX_KMAP_BEGIN)) {
0e91a0c6
TG
575 if (IS_ENABLED(CONFIG_DEBUG_KMAP_LOCAL_FORCE_MAP)) {
576 /* This _should_ never happen! See above. */
577 WARN_ON_ONCE(1);
578 return;
579 }
2a656cad
TG
580 /*
581 * Handle mappings which were obtained by kmap_high_get()
582 * first as the virtual address of such mappings is below
583 * PAGE_OFFSET. Warn for all other addresses which are in
584 * the user space part of the virtual address space.
585 */
586 if (!kmap_high_unmap_local(addr))
587 WARN_ON_ONCE(addr < PAGE_OFFSET);
298fa1ad
TG
588 return;
589 }
590
591 preempt_disable();
592 idx = arch_kmap_local_unmap_idx(kmap_local_idx(), addr);
593 WARN_ON_ONCE(addr != __fix_to_virt(FIX_KMAP_BEGIN + idx));
594
1fac7473 595 kmap_pte = kmap_get_pte(addr, idx);
298fa1ad 596 arch_kmap_local_pre_unmap(addr);
1fac7473 597 pte_clear(&init_mm, addr, kmap_pte);
298fa1ad 598 arch_kmap_local_post_unmap(addr);
5fbda3ec 599 current->kmap_ctrl.pteval[kmap_local_idx()] = __pte(0);
298fa1ad
TG
600 kmap_local_idx_pop();
601 preempt_enable();
f3ba3c71 602 migrate_enable();
298fa1ad
TG
603}
604EXPORT_SYMBOL(kunmap_local_indexed);
5fbda3ec
TG
605
606/*
607 * Invoked before switch_to(). This is safe even when during or after
608 * clearing the maps an interrupt which needs a kmap_local happens because
609 * the task::kmap_ctrl.idx is not modified by the unmapping code so a
610 * nested kmap_local will use the next unused index and restore the index
611 * on unmap. The already cleared kmaps of the outgoing task are irrelevant
612 * because the interrupt context does not know about them. The same applies
613 * when scheduling back in for an interrupt which happens before the
614 * restore is complete.
615 */
616void __kmap_local_sched_out(void)
617{
618 struct task_struct *tsk = current;
1fac7473 619 pte_t *kmap_pte;
5fbda3ec
TG
620 int i;
621
622 /* Clear kmaps */
623 for (i = 0; i < tsk->kmap_ctrl.idx; i++) {
624 pte_t pteval = tsk->kmap_ctrl.pteval[i];
625 unsigned long addr;
626 int idx;
627
628 /* With debug all even slots are unmapped and act as guard */
487cfade 629 if (IS_ENABLED(CONFIG_DEBUG_KMAP_LOCAL) && !(i & 0x01)) {
074db37b 630 WARN_ON_ONCE(pte_val(pteval) != 0);
5fbda3ec
TG
631 continue;
632 }
633 if (WARN_ON_ONCE(pte_none(pteval)))
634 continue;
635
636 /*
637 * This is a horrible hack for XTENSA to calculate the
638 * coloured PTE index. Uses the PFN encoded into the pteval
639 * and the map index calculation because the actual mapped
640 * virtual address is not stored in task::kmap_ctrl.
641 * For any sane architecture this is optimized out.
642 */
643 idx = arch_kmap_local_map_idx(i, pte_pfn(pteval));
644
645 addr = __fix_to_virt(FIX_KMAP_BEGIN + idx);
1fac7473 646 kmap_pte = kmap_get_pte(addr, idx);
5fbda3ec 647 arch_kmap_local_pre_unmap(addr);
1fac7473 648 pte_clear(&init_mm, addr, kmap_pte);
5fbda3ec
TG
649 arch_kmap_local_post_unmap(addr);
650 }
651}
652
653void __kmap_local_sched_in(void)
654{
655 struct task_struct *tsk = current;
1fac7473 656 pte_t *kmap_pte;
5fbda3ec
TG
657 int i;
658
659 /* Restore kmaps */
660 for (i = 0; i < tsk->kmap_ctrl.idx; i++) {
661 pte_t pteval = tsk->kmap_ctrl.pteval[i];
662 unsigned long addr;
663 int idx;
664
665 /* With debug all even slots are unmapped and act as guard */
487cfade 666 if (IS_ENABLED(CONFIG_DEBUG_KMAP_LOCAL) && !(i & 0x01)) {
074db37b 667 WARN_ON_ONCE(pte_val(pteval) != 0);
5fbda3ec
TG
668 continue;
669 }
670 if (WARN_ON_ONCE(pte_none(pteval)))
671 continue;
672
673 /* See comment in __kmap_local_sched_out() */
674 idx = arch_kmap_local_map_idx(i, pte_pfn(pteval));
675 addr = __fix_to_virt(FIX_KMAP_BEGIN + idx);
1fac7473
AB
676 kmap_pte = kmap_get_pte(addr, idx);
677 set_pte_at(&init_mm, addr, kmap_pte, pteval);
5fbda3ec
TG
678 arch_kmap_local_post_map(addr, pteval);
679 }
680}
681
682void kmap_local_fork(struct task_struct *tsk)
683{
684 if (WARN_ON_ONCE(tsk->kmap_ctrl.idx))
685 memset(&tsk->kmap_ctrl, 0, sizeof(tsk->kmap_ctrl));
686}
687
298fa1ad 688#endif
1da177e4 689
1da177e4
LT
690#if defined(HASHED_PAGE_VIRTUAL)
691
692#define PA_HASH_ORDER 7
693
694/*
695 * Describes one page->virtual association
696 */
697struct page_address_map {
698 struct page *page;
699 void *virtual;
700 struct list_head list;
701};
702
a354e2c8 703static struct page_address_map page_address_maps[LAST_PKMAP];
1da177e4
LT
704
705/*
706 * Hash table bucket
707 */
708static struct page_address_slot {
709 struct list_head lh; /* List of page_address_maps */
710 spinlock_t lock; /* Protect this bucket's list */
711} ____cacheline_aligned_in_smp page_address_htable[1<<PA_HASH_ORDER];
712
f9918794 713static struct page_address_slot *page_slot(const struct page *page)
1da177e4
LT
714{
715 return &page_address_htable[hash_ptr(page, PA_HASH_ORDER)];
716}
717
77f6078a
RD
718/**
719 * page_address - get the mapped virtual address of a page
720 * @page: &struct page to get the virtual address of
721 *
722 * Returns the page's virtual address.
723 */
f9918794 724void *page_address(const struct page *page)
1da177e4
LT
725{
726 unsigned long flags;
727 void *ret;
728 struct page_address_slot *pas;
729
730 if (!PageHighMem(page))
731 return lowmem_page_address(page);
732
733 pas = page_slot(page);
734 ret = NULL;
735 spin_lock_irqsave(&pas->lock, flags);
736 if (!list_empty(&pas->lh)) {
737 struct page_address_map *pam;
738
739 list_for_each_entry(pam, &pas->lh, list) {
740 if (pam->page == page) {
741 ret = pam->virtual;
742 goto done;
743 }
744 }
745 }
746done:
747 spin_unlock_irqrestore(&pas->lock, flags);
748 return ret;
749}
1da177e4
LT
750EXPORT_SYMBOL(page_address);
751
77f6078a
RD
752/**
753 * set_page_address - set a page's virtual address
754 * @page: &struct page to set
755 * @virtual: virtual address to use
756 */
1da177e4
LT
757void set_page_address(struct page *page, void *virtual)
758{
759 unsigned long flags;
760 struct page_address_slot *pas;
761 struct page_address_map *pam;
762
763 BUG_ON(!PageHighMem(page));
764
765 pas = page_slot(page);
766 if (virtual) { /* Add */
a354e2c8 767 pam = &page_address_maps[PKMAP_NR((unsigned long)virtual)];
1da177e4
LT
768 pam->page = page;
769 pam->virtual = virtual;
770
771 spin_lock_irqsave(&pas->lock, flags);
772 list_add_tail(&pam->list, &pas->lh);
773 spin_unlock_irqrestore(&pas->lock, flags);
774 } else { /* Remove */
775 spin_lock_irqsave(&pas->lock, flags);
776 list_for_each_entry(pam, &pas->lh, list) {
777 if (pam->page == page) {
778 list_del(&pam->list);
779 spin_unlock_irqrestore(&pas->lock, flags);
1da177e4
LT
780 goto done;
781 }
782 }
783 spin_unlock_irqrestore(&pas->lock, flags);
784 }
785done:
786 return;
787}
788
1da177e4
LT
789void __init page_address_init(void)
790{
791 int i;
792
1da177e4
LT
793 for (i = 0; i < ARRAY_SIZE(page_address_htable); i++) {
794 INIT_LIST_HEAD(&page_address_htable[i].lh);
795 spin_lock_init(&page_address_htable[i].lock);
796 }
1da177e4
LT
797}
798
955cc774 799#endif /* defined(HASHED_PAGE_VIRTUAL) */