2 * High memory handling common code and variables.
4 * (C) 1999 Andrea Arcangeli, SuSE GmbH, andrea@suse.de
5 * Gerhard Wichert, Siemens AG, Gerhard.Wichert@pdb.siemens.de
8 * Redesigned the x86 32-bit VM architecture to deal with
9 * 64-bit physical space. With current x86 CPUs this
10 * means up to 64 Gigabytes physical RAM.
12 * Rewrote high memory support to move the page cache into
13 * high memory. Implemented permanent (schedulable) kmaps
14 * based on Linus' idea.
16 * Copyright (C) 1999 Ingo Molnar <mingo@redhat.com>
20 #include <linux/module.h>
21 #include <linux/swap.h>
22 #include <linux/bio.h>
23 #include <linux/pagemap.h>
24 #include <linux/mempool.h>
25 #include <linux/blkdev.h>
26 #include <linux/init.h>
27 #include <linux/hash.h>
28 #include <linux/highmem.h>
29 #include <linux/blktrace_api.h>
30 #include <asm/tlbflush.h>
33 * Virtual_count is not a pure "count".
34 * 0 means that it is not mapped, and has not been mapped
35 * since a TLB flush - it is usable.
36 * 1 means that there are no users, but it has been mapped
37 * since the last TLB flush - so we can't use it.
38 * n means that there are (n-1) current users of it.
42 unsigned long totalhigh_pages __read_mostly
;
43 EXPORT_SYMBOL(totalhigh_pages
);
45 unsigned int nr_free_highpages (void)
48 unsigned int pages
= 0;
50 for_each_online_pgdat(pgdat
) {
51 pages
+= zone_page_state(&pgdat
->node_zones
[ZONE_HIGHMEM
],
53 if (zone_movable_is_highmem())
54 pages
+= zone_page_state(
55 &pgdat
->node_zones
[ZONE_MOVABLE
],
62 static int pkmap_count
[LAST_PKMAP
];
63 static unsigned int last_pkmap_nr
;
64 static __cacheline_aligned_in_smp
DEFINE_SPINLOCK(kmap_lock
);
66 pte_t
* pkmap_page_table
;
68 static DECLARE_WAIT_QUEUE_HEAD(pkmap_map_wait
);
70 static void flush_all_zero_pkmaps(void)
77 for (i
= 0; i
< LAST_PKMAP
; i
++) {
81 * zero means we don't have anything to do,
82 * >1 means that it is still in use. Only
83 * a count of 1 means that it is free but
84 * needs to be unmapped
86 if (pkmap_count
[i
] != 1)
91 BUG_ON(pte_none(pkmap_page_table
[i
]));
94 * Don't need an atomic fetch-and-clear op here;
95 * no-one has the page mapped, and cannot get at
96 * its virtual address (and hence PTE) without first
97 * getting the kmap_lock (which is held here).
98 * So no dangers, even with speculative execution.
100 page
= pte_page(pkmap_page_table
[i
]);
101 pte_clear(&init_mm
, (unsigned long)page_address(page
),
102 &pkmap_page_table
[i
]);
104 set_page_address(page
, NULL
);
108 flush_tlb_kernel_range(PKMAP_ADDR(0), PKMAP_ADDR(LAST_PKMAP
));
112 * kmap_flush_unused - flush all unused kmap mappings in order to remove stray mappings
114 void kmap_flush_unused(void)
116 spin_lock(&kmap_lock
);
117 flush_all_zero_pkmaps();
118 spin_unlock(&kmap_lock
);
121 static inline unsigned long map_new_virtual(struct page
*page
)
128 /* Find an empty entry */
130 last_pkmap_nr
= (last_pkmap_nr
+ 1) & LAST_PKMAP_MASK
;
131 if (!last_pkmap_nr
) {
132 flush_all_zero_pkmaps();
135 if (!pkmap_count
[last_pkmap_nr
])
136 break; /* Found a usable entry */
141 * Sleep for somebody else to unmap their entries
144 DECLARE_WAITQUEUE(wait
, current
);
146 __set_current_state(TASK_UNINTERRUPTIBLE
);
147 add_wait_queue(&pkmap_map_wait
, &wait
);
148 spin_unlock(&kmap_lock
);
150 remove_wait_queue(&pkmap_map_wait
, &wait
);
151 spin_lock(&kmap_lock
);
153 /* Somebody else might have mapped it while we slept */
154 if (page_address(page
))
155 return (unsigned long)page_address(page
);
161 vaddr
= PKMAP_ADDR(last_pkmap_nr
);
162 set_pte_at(&init_mm
, vaddr
,
163 &(pkmap_page_table
[last_pkmap_nr
]), mk_pte(page
, kmap_prot
));
165 pkmap_count
[last_pkmap_nr
] = 1;
166 set_page_address(page
, (void *)vaddr
);
172 * kmap_high - map a highmem page into memory
173 * @page: &struct page to map
175 * Returns the page's virtual memory address.
177 * We cannot call this from interrupts, as it may block.
179 void *kmap_high(struct page
*page
)
184 * For highmem pages, we can't trust "virtual" until
185 * after we have the lock.
187 spin_lock(&kmap_lock
);
188 vaddr
= (unsigned long)page_address(page
);
190 vaddr
= map_new_virtual(page
);
191 pkmap_count
[PKMAP_NR(vaddr
)]++;
192 BUG_ON(pkmap_count
[PKMAP_NR(vaddr
)] < 2);
193 spin_unlock(&kmap_lock
);
194 return (void*) vaddr
;
197 EXPORT_SYMBOL(kmap_high
);
200 * kunmap_high - map a highmem page into memory
201 * @page: &struct page to unmap
203 void kunmap_high(struct page
*page
)
209 spin_lock(&kmap_lock
);
210 vaddr
= (unsigned long)page_address(page
);
212 nr
= PKMAP_NR(vaddr
);
215 * A count must never go down to zero
216 * without a TLB flush!
219 switch (--pkmap_count
[nr
]) {
224 * Avoid an unnecessary wake_up() function call.
225 * The common case is pkmap_count[] == 1, but
227 * The tasks queued in the wait-queue are guarded
228 * by both the lock in the wait-queue-head and by
229 * the kmap_lock. As the kmap_lock is held here,
230 * no need for the wait-queue-head's lock. Simply
231 * test if the queue is empty.
233 need_wakeup
= waitqueue_active(&pkmap_map_wait
);
235 spin_unlock(&kmap_lock
);
237 /* do wake-up, if needed, race-free outside of the spin lock */
239 wake_up(&pkmap_map_wait
);
242 EXPORT_SYMBOL(kunmap_high
);
245 #if defined(HASHED_PAGE_VIRTUAL)
247 #define PA_HASH_ORDER 7
250 * Describes one page->virtual association
252 struct page_address_map
{
255 struct list_head list
;
259 * page_address_map freelist, allocated from page_address_maps.
261 static struct list_head page_address_pool
; /* freelist */
262 static spinlock_t pool_lock
; /* protects page_address_pool */
267 static struct page_address_slot
{
268 struct list_head lh
; /* List of page_address_maps */
269 spinlock_t lock
; /* Protect this bucket's list */
270 } ____cacheline_aligned_in_smp page_address_htable
[1<<PA_HASH_ORDER
];
272 static struct page_address_slot
*page_slot(struct page
*page
)
274 return &page_address_htable
[hash_ptr(page
, PA_HASH_ORDER
)];
278 * page_address - get the mapped virtual address of a page
279 * @page: &struct page to get the virtual address of
281 * Returns the page's virtual address.
283 void *page_address(struct page
*page
)
287 struct page_address_slot
*pas
;
289 if (!PageHighMem(page
))
290 return lowmem_page_address(page
);
292 pas
= page_slot(page
);
294 spin_lock_irqsave(&pas
->lock
, flags
);
295 if (!list_empty(&pas
->lh
)) {
296 struct page_address_map
*pam
;
298 list_for_each_entry(pam
, &pas
->lh
, list
) {
299 if (pam
->page
== page
) {
306 spin_unlock_irqrestore(&pas
->lock
, flags
);
310 EXPORT_SYMBOL(page_address
);
313 * set_page_address - set a page's virtual address
314 * @page: &struct page to set
315 * @virtual: virtual address to use
317 void set_page_address(struct page
*page
, void *virtual)
320 struct page_address_slot
*pas
;
321 struct page_address_map
*pam
;
323 BUG_ON(!PageHighMem(page
));
325 pas
= page_slot(page
);
326 if (virtual) { /* Add */
327 BUG_ON(list_empty(&page_address_pool
));
329 spin_lock_irqsave(&pool_lock
, flags
);
330 pam
= list_entry(page_address_pool
.next
,
331 struct page_address_map
, list
);
332 list_del(&pam
->list
);
333 spin_unlock_irqrestore(&pool_lock
, flags
);
336 pam
->virtual = virtual;
338 spin_lock_irqsave(&pas
->lock
, flags
);
339 list_add_tail(&pam
->list
, &pas
->lh
);
340 spin_unlock_irqrestore(&pas
->lock
, flags
);
341 } else { /* Remove */
342 spin_lock_irqsave(&pas
->lock
, flags
);
343 list_for_each_entry(pam
, &pas
->lh
, list
) {
344 if (pam
->page
== page
) {
345 list_del(&pam
->list
);
346 spin_unlock_irqrestore(&pas
->lock
, flags
);
347 spin_lock_irqsave(&pool_lock
, flags
);
348 list_add_tail(&pam
->list
, &page_address_pool
);
349 spin_unlock_irqrestore(&pool_lock
, flags
);
353 spin_unlock_irqrestore(&pas
->lock
, flags
);
359 static struct page_address_map page_address_maps
[LAST_PKMAP
];
361 void __init
page_address_init(void)
365 INIT_LIST_HEAD(&page_address_pool
);
366 for (i
= 0; i
< ARRAY_SIZE(page_address_maps
); i
++)
367 list_add(&page_address_maps
[i
].list
, &page_address_pool
);
368 for (i
= 0; i
< ARRAY_SIZE(page_address_htable
); i
++) {
369 INIT_LIST_HEAD(&page_address_htable
[i
].lh
);
370 spin_lock_init(&page_address_htable
[i
].lock
);
372 spin_lock_init(&pool_lock
);
375 #endif /* defined(CONFIG_HIGHMEM) && !defined(WANT_PAGE_VIRTUAL) */