xref: /linux/mm/highmem.c (revision d2c20e51e3966bc668ef1ef21fbe90704286c8d0)
1b2441318SGreg Kroah-Hartman // SPDX-License-Identifier: GPL-2.0
21da177e4SLinus Torvalds /*
31da177e4SLinus Torvalds  * High memory handling common code and variables.
41da177e4SLinus Torvalds  *
51da177e4SLinus Torvalds  * (C) 1999 Andrea Arcangeli, SuSE GmbH, andrea@suse.de
61da177e4SLinus Torvalds  *          Gerhard Wichert, Siemens AG, Gerhard.Wichert@pdb.siemens.de
71da177e4SLinus Torvalds  *
81da177e4SLinus Torvalds  *
91da177e4SLinus Torvalds  * Redesigned the x86 32-bit VM architecture to deal with
101da177e4SLinus Torvalds  * 64-bit physical space. With current x86 CPUs this
111da177e4SLinus Torvalds  * means up to 64 Gigabytes physical RAM.
121da177e4SLinus Torvalds  *
131da177e4SLinus Torvalds  * Rewrote high memory support to move the page cache into
141da177e4SLinus Torvalds  * high memory. Implemented permanent (schedulable) kmaps
151da177e4SLinus Torvalds  * based on Linus' idea.
161da177e4SLinus Torvalds  *
171da177e4SLinus Torvalds  * Copyright (C) 1999 Ingo Molnar <mingo@redhat.com>
181da177e4SLinus Torvalds  */
191da177e4SLinus Torvalds 
201da177e4SLinus Torvalds #include <linux/mm.h>
21b95f1b31SPaul Gortmaker #include <linux/export.h>
221da177e4SLinus Torvalds #include <linux/swap.h>
231da177e4SLinus Torvalds #include <linux/bio.h>
241da177e4SLinus Torvalds #include <linux/pagemap.h>
251da177e4SLinus Torvalds #include <linux/mempool.h>
261da177e4SLinus Torvalds #include <linux/blkdev.h>
271da177e4SLinus Torvalds #include <linux/init.h>
281da177e4SLinus Torvalds #include <linux/hash.h>
291da177e4SLinus Torvalds #include <linux/highmem.h>
30eac79005SJason Wessel #include <linux/kgdb.h>
311da177e4SLinus Torvalds #include <asm/tlbflush.h>
32186525bdSIngo Molnar #include <linux/vmalloc.h>
33a8e23a29SPeter Zijlstra 
341da177e4SLinus Torvalds /*
351da177e4SLinus Torvalds  * Virtual_count is not a pure "count".
361da177e4SLinus Torvalds  *  0 means that it is not mapped, and has not been mapped
371da177e4SLinus Torvalds  *    since a TLB flush - it is usable.
381da177e4SLinus Torvalds  *  1 means that there are no users, but it has been mapped
391da177e4SLinus Torvalds  *    since the last TLB flush - so we can't use it.
401da177e4SLinus Torvalds  *  n means that there are (n-1) current users of it.
411da177e4SLinus Torvalds  */
421da177e4SLinus Torvalds #ifdef CONFIG_HIGHMEM
43260b2367SAl Viro 
4415de36a4SMax Filippov /*
4515de36a4SMax Filippov  * Architecture with aliasing data cache may define the following family of
4615de36a4SMax Filippov  * helper functions in its asm/highmem.h to control cache color of virtual
4715de36a4SMax Filippov  * addresses where physical memory pages are mapped by kmap.
4815de36a4SMax Filippov  */
4915de36a4SMax Filippov #ifndef get_pkmap_color
5015de36a4SMax Filippov 
5115de36a4SMax Filippov /*
5215de36a4SMax Filippov  * Determine color of virtual address where the page should be mapped.
5315de36a4SMax Filippov  */
5415de36a4SMax Filippov static inline unsigned int get_pkmap_color(struct page *page)
5515de36a4SMax Filippov {
5615de36a4SMax Filippov 	return 0;
5715de36a4SMax Filippov }
5815de36a4SMax Filippov #define get_pkmap_color get_pkmap_color
5915de36a4SMax Filippov 
6015de36a4SMax Filippov /*
6115de36a4SMax Filippov  * Get next index for mapping inside PKMAP region for page with given color.
6215de36a4SMax Filippov  */
6315de36a4SMax Filippov static inline unsigned int get_next_pkmap_nr(unsigned int color)
6415de36a4SMax Filippov {
6515de36a4SMax Filippov 	static unsigned int last_pkmap_nr;
6615de36a4SMax Filippov 
6715de36a4SMax Filippov 	last_pkmap_nr = (last_pkmap_nr + 1) & LAST_PKMAP_MASK;
6815de36a4SMax Filippov 	return last_pkmap_nr;
6915de36a4SMax Filippov }
7015de36a4SMax Filippov 
7115de36a4SMax Filippov /*
7215de36a4SMax Filippov  * Determine if page index inside PKMAP region (pkmap_nr) of given color
7315de36a4SMax Filippov  * has wrapped around PKMAP region end. When this happens an attempt to
7415de36a4SMax Filippov  * flush all unused PKMAP slots is made.
7515de36a4SMax Filippov  */
7615de36a4SMax Filippov static inline int no_more_pkmaps(unsigned int pkmap_nr, unsigned int color)
7715de36a4SMax Filippov {
7815de36a4SMax Filippov 	return pkmap_nr == 0;
7915de36a4SMax Filippov }
8015de36a4SMax Filippov 
8115de36a4SMax Filippov /*
8215de36a4SMax Filippov  * Get the number of PKMAP entries of the given color. If no free slot is
8315de36a4SMax Filippov  * found after checking that many entries, kmap will sleep waiting for
8415de36a4SMax Filippov  * someone to call kunmap and free PKMAP slot.
8515de36a4SMax Filippov  */
8615de36a4SMax Filippov static inline int get_pkmap_entries_count(unsigned int color)
8715de36a4SMax Filippov {
8815de36a4SMax Filippov 	return LAST_PKMAP;
8915de36a4SMax Filippov }
9015de36a4SMax Filippov 
9115de36a4SMax Filippov /*
9215de36a4SMax Filippov  * Get head of a wait queue for PKMAP entries of the given color.
9315de36a4SMax Filippov  * Wait queues for different mapping colors should be independent to avoid
9415de36a4SMax Filippov  * unnecessary wakeups caused by freeing of slots of other colors.
9515de36a4SMax Filippov  */
9615de36a4SMax Filippov static inline wait_queue_head_t *get_pkmap_wait_queue_head(unsigned int color)
9715de36a4SMax Filippov {
9815de36a4SMax Filippov 	static DECLARE_WAIT_QUEUE_HEAD(pkmap_map_wait);
9915de36a4SMax Filippov 
10015de36a4SMax Filippov 	return &pkmap_map_wait;
10115de36a4SMax Filippov }
10215de36a4SMax Filippov #endif
10315de36a4SMax Filippov 
104ca79b0c2SArun KS atomic_long_t _totalhigh_pages __read_mostly;
105ca79b0c2SArun KS EXPORT_SYMBOL(_totalhigh_pages);
1063e4d3af5SPeter Zijlstra 
10713f876baSThomas Gleixner unsigned int __nr_free_highpages(void)
108c1f60a5aSChristoph Lameter {
10933499bfeSJoonsoo Kim 	struct zone *zone;
110c1f60a5aSChristoph Lameter 	unsigned int pages = 0;
111c1f60a5aSChristoph Lameter 
11233499bfeSJoonsoo Kim 	for_each_populated_zone(zone) {
11333499bfeSJoonsoo Kim 		if (is_highmem(zone))
11433499bfeSJoonsoo Kim 			pages += zone_page_state(zone, NR_FREE_PAGES);
1152a1e274aSMel Gorman 	}
116c1f60a5aSChristoph Lameter 
117c1f60a5aSChristoph Lameter 	return pages;
118c1f60a5aSChristoph Lameter }
119c1f60a5aSChristoph Lameter 
1201da177e4SLinus Torvalds static int pkmap_count[LAST_PKMAP];
1211da177e4SLinus Torvalds static  __cacheline_aligned_in_smp DEFINE_SPINLOCK(kmap_lock);
1221da177e4SLinus Torvalds 
1231da177e4SLinus Torvalds pte_t *pkmap_page_table;
1241da177e4SLinus Torvalds 
1253297e760SNicolas Pitre /*
1263297e760SNicolas Pitre  * Most architectures have no use for kmap_high_get(), so let's abstract
1273297e760SNicolas Pitre  * the disabling of IRQ out of the locking in that case to save on a
1283297e760SNicolas Pitre  * potential useless overhead.
1293297e760SNicolas Pitre  */
1303297e760SNicolas Pitre #ifdef ARCH_NEEDS_KMAP_HIGH_GET
1313297e760SNicolas Pitre #define lock_kmap()             spin_lock_irq(&kmap_lock)
1323297e760SNicolas Pitre #define unlock_kmap()           spin_unlock_irq(&kmap_lock)
1333297e760SNicolas Pitre #define lock_kmap_any(flags)    spin_lock_irqsave(&kmap_lock, flags)
1343297e760SNicolas Pitre #define unlock_kmap_any(flags)  spin_unlock_irqrestore(&kmap_lock, flags)
1353297e760SNicolas Pitre #else
1363297e760SNicolas Pitre #define lock_kmap()             spin_lock(&kmap_lock)
1373297e760SNicolas Pitre #define unlock_kmap()           spin_unlock(&kmap_lock)
1383297e760SNicolas Pitre #define lock_kmap_any(flags)    \
1393297e760SNicolas Pitre 		do { spin_lock(&kmap_lock); (void)(flags); } while (0)
1403297e760SNicolas Pitre #define unlock_kmap_any(flags)  \
1413297e760SNicolas Pitre 		do { spin_unlock(&kmap_lock); (void)(flags); } while (0)
1423297e760SNicolas Pitre #endif
1433297e760SNicolas Pitre 
14413f876baSThomas Gleixner struct page *__kmap_to_page(void *vaddr)
1455a178119SMel Gorman {
1465a178119SMel Gorman 	unsigned long addr = (unsigned long)vaddr;
1475a178119SMel Gorman 
148498c2280SWill Deacon 	if (addr >= PKMAP_ADDR(0) && addr < PKMAP_ADDR(LAST_PKMAP)) {
1494de22c05SJoonsoo Kim 		int i = PKMAP_NR(addr);
1509727688dSsongqiang 
1515a178119SMel Gorman 		return pte_page(pkmap_page_table[i]);
1525a178119SMel Gorman 	}
1535a178119SMel Gorman 
1545a178119SMel Gorman 	return virt_to_page(addr);
1555a178119SMel Gorman }
15613f876baSThomas Gleixner EXPORT_SYMBOL(__kmap_to_page);
1575a178119SMel Gorman 
1581da177e4SLinus Torvalds static void flush_all_zero_pkmaps(void)
1591da177e4SLinus Torvalds {
1601da177e4SLinus Torvalds 	int i;
1615843d9a4SNick Piggin 	int need_flush = 0;
1621da177e4SLinus Torvalds 
1631da177e4SLinus Torvalds 	flush_cache_kmaps();
1641da177e4SLinus Torvalds 
1651da177e4SLinus Torvalds 	for (i = 0; i < LAST_PKMAP; i++) {
1661da177e4SLinus Torvalds 		struct page *page;
1671da177e4SLinus Torvalds 
1681da177e4SLinus Torvalds 		/*
1691da177e4SLinus Torvalds 		 * zero means we don't have anything to do,
1701da177e4SLinus Torvalds 		 * >1 means that it is still in use. Only
1711da177e4SLinus Torvalds 		 * a count of 1 means that it is free but
1721da177e4SLinus Torvalds 		 * needs to be unmapped
1731da177e4SLinus Torvalds 		 */
1741da177e4SLinus Torvalds 		if (pkmap_count[i] != 1)
1751da177e4SLinus Torvalds 			continue;
1761da177e4SLinus Torvalds 		pkmap_count[i] = 0;
1771da177e4SLinus Torvalds 
1781da177e4SLinus Torvalds 		/* sanity check */
17975babcacSEric Sesterhenn 		BUG_ON(pte_none(pkmap_page_table[i]));
1801da177e4SLinus Torvalds 
1811da177e4SLinus Torvalds 		/*
1821da177e4SLinus Torvalds 		 * Don't need an atomic fetch-and-clear op here;
1831da177e4SLinus Torvalds 		 * no-one has the page mapped, and cannot get at
1841da177e4SLinus Torvalds 		 * its virtual address (and hence PTE) without first
1851da177e4SLinus Torvalds 		 * getting the kmap_lock (which is held here).
1861da177e4SLinus Torvalds 		 * So no dangers, even with speculative execution.
1871da177e4SLinus Torvalds 		 */
1881da177e4SLinus Torvalds 		page = pte_page(pkmap_page_table[i]);
189eb2db439SJoonsoo Kim 		pte_clear(&init_mm, PKMAP_ADDR(i), &pkmap_page_table[i]);
1901da177e4SLinus Torvalds 
1911da177e4SLinus Torvalds 		set_page_address(page, NULL);
1925843d9a4SNick Piggin 		need_flush = 1;
1931da177e4SLinus Torvalds 	}
1945843d9a4SNick Piggin 	if (need_flush)
1951da177e4SLinus Torvalds 		flush_tlb_kernel_range(PKMAP_ADDR(0), PKMAP_ADDR(LAST_PKMAP));
1961da177e4SLinus Torvalds }
1971da177e4SLinus Torvalds 
19813f876baSThomas Gleixner void __kmap_flush_unused(void)
199ce6234b5SJeremy Fitzhardinge {
2003297e760SNicolas Pitre 	lock_kmap();
201ce6234b5SJeremy Fitzhardinge 	flush_all_zero_pkmaps();
2023297e760SNicolas Pitre 	unlock_kmap();
203ce6234b5SJeremy Fitzhardinge }
204ce6234b5SJeremy Fitzhardinge 
2051da177e4SLinus Torvalds static inline unsigned long map_new_virtual(struct page *page)
2061da177e4SLinus Torvalds {
2071da177e4SLinus Torvalds 	unsigned long vaddr;
2081da177e4SLinus Torvalds 	int count;
20915de36a4SMax Filippov 	unsigned int last_pkmap_nr;
21015de36a4SMax Filippov 	unsigned int color = get_pkmap_color(page);
2111da177e4SLinus Torvalds 
2121da177e4SLinus Torvalds start:
21315de36a4SMax Filippov 	count = get_pkmap_entries_count(color);
2141da177e4SLinus Torvalds 	/* Find an empty entry */
2151da177e4SLinus Torvalds 	for (;;) {
21615de36a4SMax Filippov 		last_pkmap_nr = get_next_pkmap_nr(color);
21715de36a4SMax Filippov 		if (no_more_pkmaps(last_pkmap_nr, color)) {
2181da177e4SLinus Torvalds 			flush_all_zero_pkmaps();
21915de36a4SMax Filippov 			count = get_pkmap_entries_count(color);
2201da177e4SLinus Torvalds 		}
2211da177e4SLinus Torvalds 		if (!pkmap_count[last_pkmap_nr])
2221da177e4SLinus Torvalds 			break;	/* Found a usable entry */
2231da177e4SLinus Torvalds 		if (--count)
2241da177e4SLinus Torvalds 			continue;
2251da177e4SLinus Torvalds 
2261da177e4SLinus Torvalds 		/*
2271da177e4SLinus Torvalds 		 * Sleep for somebody else to unmap their entries
2281da177e4SLinus Torvalds 		 */
2291da177e4SLinus Torvalds 		{
2301da177e4SLinus Torvalds 			DECLARE_WAITQUEUE(wait, current);
23115de36a4SMax Filippov 			wait_queue_head_t *pkmap_map_wait =
23215de36a4SMax Filippov 				get_pkmap_wait_queue_head(color);
2331da177e4SLinus Torvalds 
2341da177e4SLinus Torvalds 			__set_current_state(TASK_UNINTERRUPTIBLE);
23515de36a4SMax Filippov 			add_wait_queue(pkmap_map_wait, &wait);
2363297e760SNicolas Pitre 			unlock_kmap();
2371da177e4SLinus Torvalds 			schedule();
23815de36a4SMax Filippov 			remove_wait_queue(pkmap_map_wait, &wait);
2393297e760SNicolas Pitre 			lock_kmap();
2401da177e4SLinus Torvalds 
2411da177e4SLinus Torvalds 			/* Somebody else might have mapped it while we slept */
2421da177e4SLinus Torvalds 			if (page_address(page))
2431da177e4SLinus Torvalds 				return (unsigned long)page_address(page);
2441da177e4SLinus Torvalds 
2451da177e4SLinus Torvalds 			/* Re-start */
2461da177e4SLinus Torvalds 			goto start;
2471da177e4SLinus Torvalds 		}
2481da177e4SLinus Torvalds 	}
2491da177e4SLinus Torvalds 	vaddr = PKMAP_ADDR(last_pkmap_nr);
2501da177e4SLinus Torvalds 	set_pte_at(&init_mm, vaddr,
2511da177e4SLinus Torvalds 		   &(pkmap_page_table[last_pkmap_nr]), mk_pte(page, kmap_prot));
2521da177e4SLinus Torvalds 
2531da177e4SLinus Torvalds 	pkmap_count[last_pkmap_nr] = 1;
2541da177e4SLinus Torvalds 	set_page_address(page, (void *)vaddr);
2551da177e4SLinus Torvalds 
2561da177e4SLinus Torvalds 	return vaddr;
2571da177e4SLinus Torvalds }
2581da177e4SLinus Torvalds 
25977f6078aSRandy Dunlap /**
26077f6078aSRandy Dunlap  * kmap_high - map a highmem page into memory
26177f6078aSRandy Dunlap  * @page: &struct page to map
26277f6078aSRandy Dunlap  *
26377f6078aSRandy Dunlap  * Returns the page's virtual memory address.
26477f6078aSRandy Dunlap  *
26577f6078aSRandy Dunlap  * We cannot call this from interrupts, as it may block.
26677f6078aSRandy Dunlap  */
267920c7a5dSHarvey Harrison void *kmap_high(struct page *page)
2681da177e4SLinus Torvalds {
2691da177e4SLinus Torvalds 	unsigned long vaddr;
2701da177e4SLinus Torvalds 
2711da177e4SLinus Torvalds 	/*
2721da177e4SLinus Torvalds 	 * For highmem pages, we can't trust "virtual" until
2731da177e4SLinus Torvalds 	 * after we have the lock.
2741da177e4SLinus Torvalds 	 */
2753297e760SNicolas Pitre 	lock_kmap();
2761da177e4SLinus Torvalds 	vaddr = (unsigned long)page_address(page);
2771da177e4SLinus Torvalds 	if (!vaddr)
2781da177e4SLinus Torvalds 		vaddr = map_new_virtual(page);
2791da177e4SLinus Torvalds 	pkmap_count[PKMAP_NR(vaddr)]++;
28075babcacSEric Sesterhenn 	BUG_ON(pkmap_count[PKMAP_NR(vaddr)] < 2);
2813297e760SNicolas Pitre 	unlock_kmap();
2821da177e4SLinus Torvalds 	return (void *) vaddr;
2831da177e4SLinus Torvalds }
2841da177e4SLinus Torvalds EXPORT_SYMBOL(kmap_high);
2851da177e4SLinus Torvalds 
2863297e760SNicolas Pitre #ifdef ARCH_NEEDS_KMAP_HIGH_GET
2873297e760SNicolas Pitre /**
2883297e760SNicolas Pitre  * kmap_high_get - pin a highmem page into memory
2893297e760SNicolas Pitre  * @page: &struct page to pin
2903297e760SNicolas Pitre  *
2913297e760SNicolas Pitre  * Returns the page's current virtual memory address, or NULL if no mapping
2925e39df56SUwe Kleine-König  * exists.  If and only if a non null address is returned then a
2933297e760SNicolas Pitre  * matching call to kunmap_high() is necessary.
2943297e760SNicolas Pitre  *
2953297e760SNicolas Pitre  * This can be called from any context.
2963297e760SNicolas Pitre  */
2973297e760SNicolas Pitre void *kmap_high_get(struct page *page)
2983297e760SNicolas Pitre {
2993297e760SNicolas Pitre 	unsigned long vaddr, flags;
3003297e760SNicolas Pitre 
3013297e760SNicolas Pitre 	lock_kmap_any(flags);
3023297e760SNicolas Pitre 	vaddr = (unsigned long)page_address(page);
3033297e760SNicolas Pitre 	if (vaddr) {
3043297e760SNicolas Pitre 		BUG_ON(pkmap_count[PKMAP_NR(vaddr)] < 1);
3053297e760SNicolas Pitre 		pkmap_count[PKMAP_NR(vaddr)]++;
3063297e760SNicolas Pitre 	}
3073297e760SNicolas Pitre 	unlock_kmap_any(flags);
3083297e760SNicolas Pitre 	return (void *) vaddr;
3093297e760SNicolas Pitre }
3103297e760SNicolas Pitre #endif
3113297e760SNicolas Pitre 
31277f6078aSRandy Dunlap /**
3134e9dc5dfSLi Haifeng  * kunmap_high - unmap a highmem page into memory
31477f6078aSRandy Dunlap  * @page: &struct page to unmap
3153297e760SNicolas Pitre  *
3163297e760SNicolas Pitre  * If ARCH_NEEDS_KMAP_HIGH_GET is not defined then this may be called
3173297e760SNicolas Pitre  * only from user context.
31877f6078aSRandy Dunlap  */
319920c7a5dSHarvey Harrison void kunmap_high(struct page *page)
3201da177e4SLinus Torvalds {
3211da177e4SLinus Torvalds 	unsigned long vaddr;
3221da177e4SLinus Torvalds 	unsigned long nr;
3233297e760SNicolas Pitre 	unsigned long flags;
3241da177e4SLinus Torvalds 	int need_wakeup;
32515de36a4SMax Filippov 	unsigned int color = get_pkmap_color(page);
32615de36a4SMax Filippov 	wait_queue_head_t *pkmap_map_wait;
3271da177e4SLinus Torvalds 
3283297e760SNicolas Pitre 	lock_kmap_any(flags);
3291da177e4SLinus Torvalds 	vaddr = (unsigned long)page_address(page);
33075babcacSEric Sesterhenn 	BUG_ON(!vaddr);
3311da177e4SLinus Torvalds 	nr = PKMAP_NR(vaddr);
3321da177e4SLinus Torvalds 
3331da177e4SLinus Torvalds 	/*
3341da177e4SLinus Torvalds 	 * A count must never go down to zero
3351da177e4SLinus Torvalds 	 * without a TLB flush!
3361da177e4SLinus Torvalds 	 */
3371da177e4SLinus Torvalds 	need_wakeup = 0;
3381da177e4SLinus Torvalds 	switch (--pkmap_count[nr]) {
3391da177e4SLinus Torvalds 	case 0:
3401da177e4SLinus Torvalds 		BUG();
3411da177e4SLinus Torvalds 	case 1:
3421da177e4SLinus Torvalds 		/*
3431da177e4SLinus Torvalds 		 * Avoid an unnecessary wake_up() function call.
3441da177e4SLinus Torvalds 		 * The common case is pkmap_count[] == 1, but
3451da177e4SLinus Torvalds 		 * no waiters.
3461da177e4SLinus Torvalds 		 * The tasks queued in the wait-queue are guarded
3471da177e4SLinus Torvalds 		 * by both the lock in the wait-queue-head and by
3481da177e4SLinus Torvalds 		 * the kmap_lock.  As the kmap_lock is held here,
3491da177e4SLinus Torvalds 		 * no need for the wait-queue-head's lock.  Simply
3501da177e4SLinus Torvalds 		 * test if the queue is empty.
3511da177e4SLinus Torvalds 		 */
35215de36a4SMax Filippov 		pkmap_map_wait = get_pkmap_wait_queue_head(color);
35315de36a4SMax Filippov 		need_wakeup = waitqueue_active(pkmap_map_wait);
3541da177e4SLinus Torvalds 	}
3553297e760SNicolas Pitre 	unlock_kmap_any(flags);
3561da177e4SLinus Torvalds 
3571da177e4SLinus Torvalds 	/* do wake-up, if needed, race-free outside of the spin lock */
3581da177e4SLinus Torvalds 	if (need_wakeup)
35915de36a4SMax Filippov 		wake_up(pkmap_map_wait);
3601da177e4SLinus Torvalds }
3611da177e4SLinus Torvalds EXPORT_SYMBOL(kunmap_high);
3620060ef3bSMatthew Wilcox (Oracle) 
3630060ef3bSMatthew Wilcox (Oracle) #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3640060ef3bSMatthew Wilcox (Oracle) void zero_user_segments(struct page *page, unsigned start1, unsigned end1,
3650060ef3bSMatthew Wilcox (Oracle) 		unsigned start2, unsigned end2)
3660060ef3bSMatthew Wilcox (Oracle) {
3670060ef3bSMatthew Wilcox (Oracle) 	unsigned int i;
3680060ef3bSMatthew Wilcox (Oracle) 
3690060ef3bSMatthew Wilcox (Oracle) 	BUG_ON(end1 > page_size(page) || end2 > page_size(page));
3700060ef3bSMatthew Wilcox (Oracle) 
371184cee51SOGAWA Hirofumi 	if (start1 >= end1)
372184cee51SOGAWA Hirofumi 		start1 = end1 = 0;
373184cee51SOGAWA Hirofumi 	if (start2 >= end2)
374184cee51SOGAWA Hirofumi 		start2 = end2 = 0;
375184cee51SOGAWA Hirofumi 
3760060ef3bSMatthew Wilcox (Oracle) 	for (i = 0; i < compound_nr(page); i++) {
3770060ef3bSMatthew Wilcox (Oracle) 		void *kaddr = NULL;
3780060ef3bSMatthew Wilcox (Oracle) 
3790060ef3bSMatthew Wilcox (Oracle) 		if (start1 >= PAGE_SIZE) {
3800060ef3bSMatthew Wilcox (Oracle) 			start1 -= PAGE_SIZE;
3810060ef3bSMatthew Wilcox (Oracle) 			end1 -= PAGE_SIZE;
3820060ef3bSMatthew Wilcox (Oracle) 		} else {
3830060ef3bSMatthew Wilcox (Oracle) 			unsigned this_end = min_t(unsigned, end1, PAGE_SIZE);
3840060ef3bSMatthew Wilcox (Oracle) 
385184cee51SOGAWA Hirofumi 			if (end1 > start1) {
386*d2c20e51SIra Weiny 				kaddr = kmap_local_page(page + i);
3870060ef3bSMatthew Wilcox (Oracle) 				memset(kaddr + start1, 0, this_end - start1);
388184cee51SOGAWA Hirofumi 			}
3890060ef3bSMatthew Wilcox (Oracle) 			end1 -= this_end;
3900060ef3bSMatthew Wilcox (Oracle) 			start1 = 0;
3910060ef3bSMatthew Wilcox (Oracle) 		}
3920060ef3bSMatthew Wilcox (Oracle) 
3930060ef3bSMatthew Wilcox (Oracle) 		if (start2 >= PAGE_SIZE) {
3940060ef3bSMatthew Wilcox (Oracle) 			start2 -= PAGE_SIZE;
3950060ef3bSMatthew Wilcox (Oracle) 			end2 -= PAGE_SIZE;
3960060ef3bSMatthew Wilcox (Oracle) 		} else {
3970060ef3bSMatthew Wilcox (Oracle) 			unsigned this_end = min_t(unsigned, end2, PAGE_SIZE);
3980060ef3bSMatthew Wilcox (Oracle) 
399184cee51SOGAWA Hirofumi 			if (end2 > start2) {
400184cee51SOGAWA Hirofumi 				if (!kaddr)
401*d2c20e51SIra Weiny 					kaddr = kmap_local_page(page + i);
4020060ef3bSMatthew Wilcox (Oracle) 				memset(kaddr + start2, 0, this_end - start2);
403184cee51SOGAWA Hirofumi 			}
4040060ef3bSMatthew Wilcox (Oracle) 			end2 -= this_end;
4050060ef3bSMatthew Wilcox (Oracle) 			start2 = 0;
4060060ef3bSMatthew Wilcox (Oracle) 		}
4070060ef3bSMatthew Wilcox (Oracle) 
4080060ef3bSMatthew Wilcox (Oracle) 		if (kaddr) {
409*d2c20e51SIra Weiny 			kunmap_local(kaddr);
4100060ef3bSMatthew Wilcox (Oracle) 			flush_dcache_page(page + i);
4110060ef3bSMatthew Wilcox (Oracle) 		}
4120060ef3bSMatthew Wilcox (Oracle) 
4130060ef3bSMatthew Wilcox (Oracle) 		if (!end1 && !end2)
4140060ef3bSMatthew Wilcox (Oracle) 			break;
4150060ef3bSMatthew Wilcox (Oracle) 	}
4160060ef3bSMatthew Wilcox (Oracle) 
4170060ef3bSMatthew Wilcox (Oracle) 	BUG_ON((start1 | start2 | end1 | end2) != 0);
4180060ef3bSMatthew Wilcox (Oracle) }
4190060ef3bSMatthew Wilcox (Oracle) EXPORT_SYMBOL(zero_user_segments);
4200060ef3bSMatthew Wilcox (Oracle) #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
421955cc774SIra Weiny #endif /* CONFIG_HIGHMEM */
4221da177e4SLinus Torvalds 
423298fa1adSThomas Gleixner #ifdef CONFIG_KMAP_LOCAL
424298fa1adSThomas Gleixner 
425298fa1adSThomas Gleixner #include <asm/kmap_size.h>
426298fa1adSThomas Gleixner 
427389755c2SThomas Gleixner /*
4286e799cb6SThomas Gleixner  * With DEBUG_KMAP_LOCAL the stack depth is doubled and every second
429389755c2SThomas Gleixner  * slot is unused which acts as a guard page
430389755c2SThomas Gleixner  */
4316e799cb6SThomas Gleixner #ifdef CONFIG_DEBUG_KMAP_LOCAL
432389755c2SThomas Gleixner # define KM_INCR	2
433389755c2SThomas Gleixner #else
434389755c2SThomas Gleixner # define KM_INCR	1
435389755c2SThomas Gleixner #endif
436389755c2SThomas Gleixner 
437298fa1adSThomas Gleixner static inline int kmap_local_idx_push(void)
438298fa1adSThomas Gleixner {
439ea0eafeaSChangbin Du 	WARN_ON_ONCE(in_hardirq() && !irqs_disabled());
4405fbda3ecSThomas Gleixner 	current->kmap_ctrl.idx += KM_INCR;
4415fbda3ecSThomas Gleixner 	BUG_ON(current->kmap_ctrl.idx >= KM_MAX_IDX);
4425fbda3ecSThomas Gleixner 	return current->kmap_ctrl.idx - 1;
443298fa1adSThomas Gleixner }
444298fa1adSThomas Gleixner 
445298fa1adSThomas Gleixner static inline int kmap_local_idx(void)
446298fa1adSThomas Gleixner {
4475fbda3ecSThomas Gleixner 	return current->kmap_ctrl.idx - 1;
448298fa1adSThomas Gleixner }
449298fa1adSThomas Gleixner 
450298fa1adSThomas Gleixner static inline void kmap_local_idx_pop(void)
451298fa1adSThomas Gleixner {
4525fbda3ecSThomas Gleixner 	current->kmap_ctrl.idx -= KM_INCR;
4535fbda3ecSThomas Gleixner 	BUG_ON(current->kmap_ctrl.idx < 0);
454298fa1adSThomas Gleixner }
455298fa1adSThomas Gleixner 
456298fa1adSThomas Gleixner #ifndef arch_kmap_local_post_map
457298fa1adSThomas Gleixner # define arch_kmap_local_post_map(vaddr, pteval)	do { } while (0)
458298fa1adSThomas Gleixner #endif
4593c1016b5SThomas Gleixner 
460298fa1adSThomas Gleixner #ifndef arch_kmap_local_pre_unmap
461298fa1adSThomas Gleixner # define arch_kmap_local_pre_unmap(vaddr)		do { } while (0)
462298fa1adSThomas Gleixner #endif
463298fa1adSThomas Gleixner 
464298fa1adSThomas Gleixner #ifndef arch_kmap_local_post_unmap
465298fa1adSThomas Gleixner # define arch_kmap_local_post_unmap(vaddr)		do { } while (0)
466298fa1adSThomas Gleixner #endif
467298fa1adSThomas Gleixner 
468298fa1adSThomas Gleixner #ifndef arch_kmap_local_map_idx
469298fa1adSThomas Gleixner #define arch_kmap_local_map_idx(idx, pfn)	kmap_local_calc_idx(idx)
470298fa1adSThomas Gleixner #endif
471298fa1adSThomas Gleixner 
472298fa1adSThomas Gleixner #ifndef arch_kmap_local_unmap_idx
473298fa1adSThomas Gleixner #define arch_kmap_local_unmap_idx(idx, vaddr)	kmap_local_calc_idx(idx)
474298fa1adSThomas Gleixner #endif
475298fa1adSThomas Gleixner 
476298fa1adSThomas Gleixner #ifndef arch_kmap_local_high_get
477298fa1adSThomas Gleixner static inline void *arch_kmap_local_high_get(struct page *page)
478298fa1adSThomas Gleixner {
479298fa1adSThomas Gleixner 	return NULL;
480298fa1adSThomas Gleixner }
481298fa1adSThomas Gleixner #endif
482298fa1adSThomas Gleixner 
483a1dce7fdSThomas Gleixner #ifndef arch_kmap_local_set_pte
484a1dce7fdSThomas Gleixner #define arch_kmap_local_set_pte(mm, vaddr, ptep, ptev)	\
485a1dce7fdSThomas Gleixner 	set_pte_at(mm, vaddr, ptep, ptev)
486a1dce7fdSThomas Gleixner #endif
487a1dce7fdSThomas Gleixner 
488298fa1adSThomas Gleixner /* Unmap a local mapping which was obtained by kmap_high_get() */
4892a656cadSThomas Gleixner static inline bool kmap_high_unmap_local(unsigned long vaddr)
490298fa1adSThomas Gleixner {
491298fa1adSThomas Gleixner #ifdef ARCH_NEEDS_KMAP_HIGH_GET
4922a656cadSThomas Gleixner 	if (vaddr >= PKMAP_ADDR(0) && vaddr < PKMAP_ADDR(LAST_PKMAP)) {
493298fa1adSThomas Gleixner 		kunmap_high(pte_page(pkmap_page_table[PKMAP_NR(vaddr)]));
4942a656cadSThomas Gleixner 		return true;
4952a656cadSThomas Gleixner 	}
496298fa1adSThomas Gleixner #endif
4972a656cadSThomas Gleixner 	return false;
498298fa1adSThomas Gleixner }
499298fa1adSThomas Gleixner 
500298fa1adSThomas Gleixner static inline int kmap_local_calc_idx(int idx)
501298fa1adSThomas Gleixner {
502298fa1adSThomas Gleixner 	return idx + KM_MAX_IDX * smp_processor_id();
503298fa1adSThomas Gleixner }
504298fa1adSThomas Gleixner 
505298fa1adSThomas Gleixner static pte_t *__kmap_pte;
506298fa1adSThomas Gleixner 
507298fa1adSThomas Gleixner static pte_t *kmap_get_pte(void)
508298fa1adSThomas Gleixner {
509298fa1adSThomas Gleixner 	if (!__kmap_pte)
510298fa1adSThomas Gleixner 		__kmap_pte = virt_to_kpte(__fix_to_virt(FIX_KMAP_BEGIN));
511298fa1adSThomas Gleixner 	return __kmap_pte;
512298fa1adSThomas Gleixner }
513298fa1adSThomas Gleixner 
514298fa1adSThomas Gleixner void *__kmap_local_pfn_prot(unsigned long pfn, pgprot_t prot)
515298fa1adSThomas Gleixner {
516298fa1adSThomas Gleixner 	pte_t pteval, *kmap_pte = kmap_get_pte();
517298fa1adSThomas Gleixner 	unsigned long vaddr;
518298fa1adSThomas Gleixner 	int idx;
519298fa1adSThomas Gleixner 
520f3ba3c71SThomas Gleixner 	/*
521f3ba3c71SThomas Gleixner 	 * Disable migration so resulting virtual address is stable
522f0953a1bSIngo Molnar 	 * across preemption.
523f3ba3c71SThomas Gleixner 	 */
524f3ba3c71SThomas Gleixner 	migrate_disable();
525298fa1adSThomas Gleixner 	preempt_disable();
526298fa1adSThomas Gleixner 	idx = arch_kmap_local_map_idx(kmap_local_idx_push(), pfn);
527298fa1adSThomas Gleixner 	vaddr = __fix_to_virt(FIX_KMAP_BEGIN + idx);
528298fa1adSThomas Gleixner 	BUG_ON(!pte_none(*(kmap_pte - idx)));
529298fa1adSThomas Gleixner 	pteval = pfn_pte(pfn, prot);
530a1dce7fdSThomas Gleixner 	arch_kmap_local_set_pte(&init_mm, vaddr, kmap_pte - idx, pteval);
531298fa1adSThomas Gleixner 	arch_kmap_local_post_map(vaddr, pteval);
5325fbda3ecSThomas Gleixner 	current->kmap_ctrl.pteval[kmap_local_idx()] = pteval;
533298fa1adSThomas Gleixner 	preempt_enable();
534298fa1adSThomas Gleixner 
535298fa1adSThomas Gleixner 	return (void *)vaddr;
536298fa1adSThomas Gleixner }
537298fa1adSThomas Gleixner EXPORT_SYMBOL_GPL(__kmap_local_pfn_prot);
538298fa1adSThomas Gleixner 
539298fa1adSThomas Gleixner void *__kmap_local_page_prot(struct page *page, pgprot_t prot)
540298fa1adSThomas Gleixner {
541298fa1adSThomas Gleixner 	void *kmap;
542298fa1adSThomas Gleixner 
5430e91a0c6SThomas Gleixner 	/*
5440e91a0c6SThomas Gleixner 	 * To broaden the usage of the actual kmap_local() machinery always map
5450e91a0c6SThomas Gleixner 	 * pages when debugging is enabled and the architecture has no problems
5460e91a0c6SThomas Gleixner 	 * with alias mappings.
5470e91a0c6SThomas Gleixner 	 */
5480e91a0c6SThomas Gleixner 	if (!IS_ENABLED(CONFIG_DEBUG_KMAP_LOCAL_FORCE_MAP) && !PageHighMem(page))
549298fa1adSThomas Gleixner 		return page_address(page);
550298fa1adSThomas Gleixner 
551298fa1adSThomas Gleixner 	/* Try kmap_high_get() if architecture has it enabled */
552298fa1adSThomas Gleixner 	kmap = arch_kmap_local_high_get(page);
553298fa1adSThomas Gleixner 	if (kmap)
554298fa1adSThomas Gleixner 		return kmap;
555298fa1adSThomas Gleixner 
556298fa1adSThomas Gleixner 	return __kmap_local_pfn_prot(page_to_pfn(page), prot);
557298fa1adSThomas Gleixner }
558298fa1adSThomas Gleixner EXPORT_SYMBOL(__kmap_local_page_prot);
559298fa1adSThomas Gleixner 
560298fa1adSThomas Gleixner void kunmap_local_indexed(void *vaddr)
561298fa1adSThomas Gleixner {
562298fa1adSThomas Gleixner 	unsigned long addr = (unsigned long) vaddr & PAGE_MASK;
563298fa1adSThomas Gleixner 	pte_t *kmap_pte = kmap_get_pte();
564298fa1adSThomas Gleixner 	int idx;
565298fa1adSThomas Gleixner 
566298fa1adSThomas Gleixner 	if (addr < __fix_to_virt(FIX_KMAP_END) ||
567298fa1adSThomas Gleixner 	    addr > __fix_to_virt(FIX_KMAP_BEGIN)) {
5680e91a0c6SThomas Gleixner 		if (IS_ENABLED(CONFIG_DEBUG_KMAP_LOCAL_FORCE_MAP)) {
5690e91a0c6SThomas Gleixner 			/* This _should_ never happen! See above. */
5700e91a0c6SThomas Gleixner 			WARN_ON_ONCE(1);
5710e91a0c6SThomas Gleixner 			return;
5720e91a0c6SThomas Gleixner 		}
5732a656cadSThomas Gleixner 		/*
5742a656cadSThomas Gleixner 		 * Handle mappings which were obtained by kmap_high_get()
5752a656cadSThomas Gleixner 		 * first as the virtual address of such mappings is below
5762a656cadSThomas Gleixner 		 * PAGE_OFFSET. Warn for all other addresses which are in
5772a656cadSThomas Gleixner 		 * the user space part of the virtual address space.
5782a656cadSThomas Gleixner 		 */
5792a656cadSThomas Gleixner 		if (!kmap_high_unmap_local(addr))
580298fa1adSThomas Gleixner 			WARN_ON_ONCE(addr < PAGE_OFFSET);
581298fa1adSThomas Gleixner 		return;
582298fa1adSThomas Gleixner 	}
583298fa1adSThomas Gleixner 
584298fa1adSThomas Gleixner 	preempt_disable();
585298fa1adSThomas Gleixner 	idx = arch_kmap_local_unmap_idx(kmap_local_idx(), addr);
586298fa1adSThomas Gleixner 	WARN_ON_ONCE(addr != __fix_to_virt(FIX_KMAP_BEGIN + idx));
587298fa1adSThomas Gleixner 
588298fa1adSThomas Gleixner 	arch_kmap_local_pre_unmap(addr);
589298fa1adSThomas Gleixner 	pte_clear(&init_mm, addr, kmap_pte - idx);
590298fa1adSThomas Gleixner 	arch_kmap_local_post_unmap(addr);
5915fbda3ecSThomas Gleixner 	current->kmap_ctrl.pteval[kmap_local_idx()] = __pte(0);
592298fa1adSThomas Gleixner 	kmap_local_idx_pop();
593298fa1adSThomas Gleixner 	preempt_enable();
594f3ba3c71SThomas Gleixner 	migrate_enable();
595298fa1adSThomas Gleixner }
596298fa1adSThomas Gleixner EXPORT_SYMBOL(kunmap_local_indexed);
5975fbda3ecSThomas Gleixner 
5985fbda3ecSThomas Gleixner /*
5995fbda3ecSThomas Gleixner  * Invoked before switch_to(). This is safe even when during or after
6005fbda3ecSThomas Gleixner  * clearing the maps an interrupt which needs a kmap_local happens because
6015fbda3ecSThomas Gleixner  * the task::kmap_ctrl.idx is not modified by the unmapping code so a
6025fbda3ecSThomas Gleixner  * nested kmap_local will use the next unused index and restore the index
6035fbda3ecSThomas Gleixner  * on unmap. The already cleared kmaps of the outgoing task are irrelevant
6045fbda3ecSThomas Gleixner  * because the interrupt context does not know about them. The same applies
6055fbda3ecSThomas Gleixner  * when scheduling back in for an interrupt which happens before the
6065fbda3ecSThomas Gleixner  * restore is complete.
6075fbda3ecSThomas Gleixner  */
6085fbda3ecSThomas Gleixner void __kmap_local_sched_out(void)
6095fbda3ecSThomas Gleixner {
6105fbda3ecSThomas Gleixner 	struct task_struct *tsk = current;
6115fbda3ecSThomas Gleixner 	pte_t *kmap_pte = kmap_get_pte();
6125fbda3ecSThomas Gleixner 	int i;
6135fbda3ecSThomas Gleixner 
6145fbda3ecSThomas Gleixner 	/* Clear kmaps */
6155fbda3ecSThomas Gleixner 	for (i = 0; i < tsk->kmap_ctrl.idx; i++) {
6165fbda3ecSThomas Gleixner 		pte_t pteval = tsk->kmap_ctrl.pteval[i];
6175fbda3ecSThomas Gleixner 		unsigned long addr;
6185fbda3ecSThomas Gleixner 		int idx;
6195fbda3ecSThomas Gleixner 
6205fbda3ecSThomas Gleixner 		/* With debug all even slots are unmapped and act as guard */
621487cfadeSIra Weiny 		if (IS_ENABLED(CONFIG_DEBUG_KMAP_LOCAL) && !(i & 0x01)) {
6225fbda3ecSThomas Gleixner 			WARN_ON_ONCE(!pte_none(pteval));
6235fbda3ecSThomas Gleixner 			continue;
6245fbda3ecSThomas Gleixner 		}
6255fbda3ecSThomas Gleixner 		if (WARN_ON_ONCE(pte_none(pteval)))
6265fbda3ecSThomas Gleixner 			continue;
6275fbda3ecSThomas Gleixner 
6285fbda3ecSThomas Gleixner 		/*
6295fbda3ecSThomas Gleixner 		 * This is a horrible hack for XTENSA to calculate the
6305fbda3ecSThomas Gleixner 		 * coloured PTE index. Uses the PFN encoded into the pteval
6315fbda3ecSThomas Gleixner 		 * and the map index calculation because the actual mapped
6325fbda3ecSThomas Gleixner 		 * virtual address is not stored in task::kmap_ctrl.
6335fbda3ecSThomas Gleixner 		 * For any sane architecture this is optimized out.
6345fbda3ecSThomas Gleixner 		 */
6355fbda3ecSThomas Gleixner 		idx = arch_kmap_local_map_idx(i, pte_pfn(pteval));
6365fbda3ecSThomas Gleixner 
6375fbda3ecSThomas Gleixner 		addr = __fix_to_virt(FIX_KMAP_BEGIN + idx);
6385fbda3ecSThomas Gleixner 		arch_kmap_local_pre_unmap(addr);
6395fbda3ecSThomas Gleixner 		pte_clear(&init_mm, addr, kmap_pte - idx);
6405fbda3ecSThomas Gleixner 		arch_kmap_local_post_unmap(addr);
6415fbda3ecSThomas Gleixner 	}
6425fbda3ecSThomas Gleixner }
6435fbda3ecSThomas Gleixner 
6445fbda3ecSThomas Gleixner void __kmap_local_sched_in(void)
6455fbda3ecSThomas Gleixner {
6465fbda3ecSThomas Gleixner 	struct task_struct *tsk = current;
6475fbda3ecSThomas Gleixner 	pte_t *kmap_pte = kmap_get_pte();
6485fbda3ecSThomas Gleixner 	int i;
6495fbda3ecSThomas Gleixner 
6505fbda3ecSThomas Gleixner 	/* Restore kmaps */
6515fbda3ecSThomas Gleixner 	for (i = 0; i < tsk->kmap_ctrl.idx; i++) {
6525fbda3ecSThomas Gleixner 		pte_t pteval = tsk->kmap_ctrl.pteval[i];
6535fbda3ecSThomas Gleixner 		unsigned long addr;
6545fbda3ecSThomas Gleixner 		int idx;
6555fbda3ecSThomas Gleixner 
6565fbda3ecSThomas Gleixner 		/* With debug all even slots are unmapped and act as guard */
657487cfadeSIra Weiny 		if (IS_ENABLED(CONFIG_DEBUG_KMAP_LOCAL) && !(i & 0x01)) {
6585fbda3ecSThomas Gleixner 			WARN_ON_ONCE(!pte_none(pteval));
6595fbda3ecSThomas Gleixner 			continue;
6605fbda3ecSThomas Gleixner 		}
6615fbda3ecSThomas Gleixner 		if (WARN_ON_ONCE(pte_none(pteval)))
6625fbda3ecSThomas Gleixner 			continue;
6635fbda3ecSThomas Gleixner 
6645fbda3ecSThomas Gleixner 		/* See comment in __kmap_local_sched_out() */
6655fbda3ecSThomas Gleixner 		idx = arch_kmap_local_map_idx(i, pte_pfn(pteval));
6665fbda3ecSThomas Gleixner 		addr = __fix_to_virt(FIX_KMAP_BEGIN + idx);
6675fbda3ecSThomas Gleixner 		set_pte_at(&init_mm, addr, kmap_pte - idx, pteval);
6685fbda3ecSThomas Gleixner 		arch_kmap_local_post_map(addr, pteval);
6695fbda3ecSThomas Gleixner 	}
6705fbda3ecSThomas Gleixner }
6715fbda3ecSThomas Gleixner 
6725fbda3ecSThomas Gleixner void kmap_local_fork(struct task_struct *tsk)
6735fbda3ecSThomas Gleixner {
6745fbda3ecSThomas Gleixner 	if (WARN_ON_ONCE(tsk->kmap_ctrl.idx))
6755fbda3ecSThomas Gleixner 		memset(&tsk->kmap_ctrl, 0, sizeof(tsk->kmap_ctrl));
6765fbda3ecSThomas Gleixner }
6775fbda3ecSThomas Gleixner 
678298fa1adSThomas Gleixner #endif
679298fa1adSThomas Gleixner 
6801da177e4SLinus Torvalds #if defined(HASHED_PAGE_VIRTUAL)
6811da177e4SLinus Torvalds 
6821da177e4SLinus Torvalds #define PA_HASH_ORDER	7
6831da177e4SLinus Torvalds 
6841da177e4SLinus Torvalds /*
6851da177e4SLinus Torvalds  * Describes one page->virtual association
6861da177e4SLinus Torvalds  */
6871da177e4SLinus Torvalds struct page_address_map {
6881da177e4SLinus Torvalds 	struct page *page;
6891da177e4SLinus Torvalds 	void *virtual;
6901da177e4SLinus Torvalds 	struct list_head list;
6911da177e4SLinus Torvalds };
6921da177e4SLinus Torvalds 
693a354e2c8SJoonsoo Kim static struct page_address_map page_address_maps[LAST_PKMAP];
6941da177e4SLinus Torvalds 
6951da177e4SLinus Torvalds /*
6961da177e4SLinus Torvalds  * Hash table bucket
6971da177e4SLinus Torvalds  */
6981da177e4SLinus Torvalds static struct page_address_slot {
6991da177e4SLinus Torvalds 	struct list_head lh;			/* List of page_address_maps */
7001da177e4SLinus Torvalds 	spinlock_t lock;			/* Protect this bucket's list */
7011da177e4SLinus Torvalds } ____cacheline_aligned_in_smp page_address_htable[1<<PA_HASH_ORDER];
7021da177e4SLinus Torvalds 
703f9918794SIan Campbell static struct page_address_slot *page_slot(const struct page *page)
7041da177e4SLinus Torvalds {
7051da177e4SLinus Torvalds 	return &page_address_htable[hash_ptr(page, PA_HASH_ORDER)];
7061da177e4SLinus Torvalds }
7071da177e4SLinus Torvalds 
70877f6078aSRandy Dunlap /**
70977f6078aSRandy Dunlap  * page_address - get the mapped virtual address of a page
71077f6078aSRandy Dunlap  * @page: &struct page to get the virtual address of
71177f6078aSRandy Dunlap  *
71277f6078aSRandy Dunlap  * Returns the page's virtual address.
71377f6078aSRandy Dunlap  */
714f9918794SIan Campbell void *page_address(const struct page *page)
7151da177e4SLinus Torvalds {
7161da177e4SLinus Torvalds 	unsigned long flags;
7171da177e4SLinus Torvalds 	void *ret;
7181da177e4SLinus Torvalds 	struct page_address_slot *pas;
7191da177e4SLinus Torvalds 
7201da177e4SLinus Torvalds 	if (!PageHighMem(page))
7211da177e4SLinus Torvalds 		return lowmem_page_address(page);
7221da177e4SLinus Torvalds 
7231da177e4SLinus Torvalds 	pas = page_slot(page);
7241da177e4SLinus Torvalds 	ret = NULL;
7251da177e4SLinus Torvalds 	spin_lock_irqsave(&pas->lock, flags);
7261da177e4SLinus Torvalds 	if (!list_empty(&pas->lh)) {
7271da177e4SLinus Torvalds 		struct page_address_map *pam;
7281da177e4SLinus Torvalds 
7291da177e4SLinus Torvalds 		list_for_each_entry(pam, &pas->lh, list) {
7301da177e4SLinus Torvalds 			if (pam->page == page) {
7311da177e4SLinus Torvalds 				ret = pam->virtual;
7321da177e4SLinus Torvalds 				goto done;
7331da177e4SLinus Torvalds 			}
7341da177e4SLinus Torvalds 		}
7351da177e4SLinus Torvalds 	}
7361da177e4SLinus Torvalds done:
7371da177e4SLinus Torvalds 	spin_unlock_irqrestore(&pas->lock, flags);
7381da177e4SLinus Torvalds 	return ret;
7391da177e4SLinus Torvalds }
7401da177e4SLinus Torvalds EXPORT_SYMBOL(page_address);
7411da177e4SLinus Torvalds 
74277f6078aSRandy Dunlap /**
74377f6078aSRandy Dunlap  * set_page_address - set a page's virtual address
74477f6078aSRandy Dunlap  * @page: &struct page to set
74577f6078aSRandy Dunlap  * @virtual: virtual address to use
74677f6078aSRandy Dunlap  */
7471da177e4SLinus Torvalds void set_page_address(struct page *page, void *virtual)
7481da177e4SLinus Torvalds {
7491da177e4SLinus Torvalds 	unsigned long flags;
7501da177e4SLinus Torvalds 	struct page_address_slot *pas;
7511da177e4SLinus Torvalds 	struct page_address_map *pam;
7521da177e4SLinus Torvalds 
7531da177e4SLinus Torvalds 	BUG_ON(!PageHighMem(page));
7541da177e4SLinus Torvalds 
7551da177e4SLinus Torvalds 	pas = page_slot(page);
7561da177e4SLinus Torvalds 	if (virtual) {		/* Add */
757a354e2c8SJoonsoo Kim 		pam = &page_address_maps[PKMAP_NR((unsigned long)virtual)];
7581da177e4SLinus Torvalds 		pam->page = page;
7591da177e4SLinus Torvalds 		pam->virtual = virtual;
7601da177e4SLinus Torvalds 
7611da177e4SLinus Torvalds 		spin_lock_irqsave(&pas->lock, flags);
7621da177e4SLinus Torvalds 		list_add_tail(&pam->list, &pas->lh);
7631da177e4SLinus Torvalds 		spin_unlock_irqrestore(&pas->lock, flags);
7641da177e4SLinus Torvalds 	} else {		/* Remove */
7651da177e4SLinus Torvalds 		spin_lock_irqsave(&pas->lock, flags);
7661da177e4SLinus Torvalds 		list_for_each_entry(pam, &pas->lh, list) {
7671da177e4SLinus Torvalds 			if (pam->page == page) {
7681da177e4SLinus Torvalds 				list_del(&pam->list);
7691da177e4SLinus Torvalds 				spin_unlock_irqrestore(&pas->lock, flags);
7701da177e4SLinus Torvalds 				goto done;
7711da177e4SLinus Torvalds 			}
7721da177e4SLinus Torvalds 		}
7731da177e4SLinus Torvalds 		spin_unlock_irqrestore(&pas->lock, flags);
7741da177e4SLinus Torvalds 	}
7751da177e4SLinus Torvalds done:
7761da177e4SLinus Torvalds 	return;
7771da177e4SLinus Torvalds }
7781da177e4SLinus Torvalds 
7791da177e4SLinus Torvalds void __init page_address_init(void)
7801da177e4SLinus Torvalds {
7811da177e4SLinus Torvalds 	int i;
7821da177e4SLinus Torvalds 
7831da177e4SLinus Torvalds 	for (i = 0; i < ARRAY_SIZE(page_address_htable); i++) {
7841da177e4SLinus Torvalds 		INIT_LIST_HEAD(&page_address_htable[i].lh);
7851da177e4SLinus Torvalds 		spin_lock_init(&page_address_htable[i].lock);
7861da177e4SLinus Torvalds 	}
7871da177e4SLinus Torvalds }
7881da177e4SLinus Torvalds 
789955cc774SIra Weiny #endif	/* defined(HASHED_PAGE_VIRTUAL) */
790