]>
Commit | Line | Data |
---|---|---|
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, [email protected] | |
6 | * Gerhard Wichert, Siemens AG, [email protected] | |
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 <[email protected]> | |
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> | |
1da177e4 LT |
26 | #include <linux/init.h> |
27 | #include <linux/hash.h> | |
28 | #include <linux/highmem.h> | |
eac79005 | 29 | #include <linux/kgdb.h> |
1da177e4 | 30 | #include <asm/tlbflush.h> |
186525bd | 31 | #include <linux/vmalloc.h> |
a8e23a29 | 32 | |
ef6e06b2 IW |
33 | #ifdef CONFIG_KMAP_LOCAL |
34 | static inline int kmap_local_calc_idx(int idx) | |
35 | { | |
36 | return idx + KM_MAX_IDX * smp_processor_id(); | |
37 | } | |
38 | ||
39 | #ifndef arch_kmap_local_map_idx | |
40 | #define arch_kmap_local_map_idx(idx, pfn) kmap_local_calc_idx(idx) | |
41 | #endif | |
42 | #endif /* CONFIG_KMAP_LOCAL */ | |
43 | ||
1da177e4 LT |
44 | /* |
45 | * Virtual_count is not a pure "count". | |
46 | * 0 means that it is not mapped, and has not been mapped | |
47 | * since a TLB flush - it is usable. | |
48 | * 1 means that there are no users, but it has been mapped | |
49 | * since the last TLB flush - so we can't use it. | |
50 | * n means that there are (n-1) current users of it. | |
51 | */ | |
52 | #ifdef CONFIG_HIGHMEM | |
260b2367 | 53 | |
15de36a4 MF |
54 | /* |
55 | * Architecture with aliasing data cache may define the following family of | |
56 | * helper functions in its asm/highmem.h to control cache color of virtual | |
57 | * addresses where physical memory pages are mapped by kmap. | |
58 | */ | |
59 | #ifndef get_pkmap_color | |
60 | ||
61 | /* | |
62 | * Determine color of virtual address where the page should be mapped. | |
63 | */ | |
64 | static inline unsigned int get_pkmap_color(struct page *page) | |
65 | { | |
66 | return 0; | |
67 | } | |
68 | #define get_pkmap_color get_pkmap_color | |
69 | ||
70 | /* | |
71 | * Get next index for mapping inside PKMAP region for page with given color. | |
72 | */ | |
73 | static inline unsigned int get_next_pkmap_nr(unsigned int color) | |
74 | { | |
75 | static unsigned int last_pkmap_nr; | |
76 | ||
77 | last_pkmap_nr = (last_pkmap_nr + 1) & LAST_PKMAP_MASK; | |
78 | return last_pkmap_nr; | |
79 | } | |
80 | ||
81 | /* | |
82 | * Determine if page index inside PKMAP region (pkmap_nr) of given color | |
83 | * has wrapped around PKMAP region end. When this happens an attempt to | |
84 | * flush all unused PKMAP slots is made. | |
85 | */ | |
86 | static inline int no_more_pkmaps(unsigned int pkmap_nr, unsigned int color) | |
87 | { | |
88 | return pkmap_nr == 0; | |
89 | } | |
90 | ||
91 | /* | |
92 | * Get the number of PKMAP entries of the given color. If no free slot is | |
93 | * found after checking that many entries, kmap will sleep waiting for | |
94 | * someone to call kunmap and free PKMAP slot. | |
95 | */ | |
96 | static inline int get_pkmap_entries_count(unsigned int color) | |
97 | { | |
98 | return LAST_PKMAP; | |
99 | } | |
100 | ||
101 | /* | |
102 | * Get head of a wait queue for PKMAP entries of the given color. | |
103 | * Wait queues for different mapping colors should be independent to avoid | |
104 | * unnecessary wakeups caused by freeing of slots of other colors. | |
105 | */ | |
106 | static inline wait_queue_head_t *get_pkmap_wait_queue_head(unsigned int color) | |
107 | { | |
108 | static DECLARE_WAIT_QUEUE_HEAD(pkmap_map_wait); | |
109 | ||
110 | return &pkmap_map_wait; | |
111 | } | |
112 | #endif | |
113 | ||
ca79b0c2 AK |
114 | atomic_long_t _totalhigh_pages __read_mostly; |
115 | EXPORT_SYMBOL(_totalhigh_pages); | |
3e4d3af5 | 116 | |
9727688d | 117 | unsigned int __nr_free_highpages(void) |
c1f60a5a | 118 | { |
33499bfe | 119 | struct zone *zone; |
c1f60a5a CL |
120 | unsigned int pages = 0; |
121 | ||
33499bfe JK |
122 | for_each_populated_zone(zone) { |
123 | if (is_highmem(zone)) | |
124 | pages += zone_page_state(zone, NR_FREE_PAGES); | |
2a1e274a | 125 | } |
c1f60a5a CL |
126 | |
127 | return pages; | |
128 | } | |
129 | ||
1da177e4 | 130 | static int pkmap_count[LAST_PKMAP]; |
1da177e4 LT |
131 | static __cacheline_aligned_in_smp DEFINE_SPINLOCK(kmap_lock); |
132 | ||
9727688d | 133 | pte_t *pkmap_page_table; |
1da177e4 | 134 | |
3297e760 NP |
135 | /* |
136 | * Most architectures have no use for kmap_high_get(), so let's abstract | |
137 | * the disabling of IRQ out of the locking in that case to save on a | |
138 | * potential useless overhead. | |
139 | */ | |
140 | #ifdef ARCH_NEEDS_KMAP_HIGH_GET | |
141 | #define lock_kmap() spin_lock_irq(&kmap_lock) | |
142 | #define unlock_kmap() spin_unlock_irq(&kmap_lock) | |
143 | #define lock_kmap_any(flags) spin_lock_irqsave(&kmap_lock, flags) | |
144 | #define unlock_kmap_any(flags) spin_unlock_irqrestore(&kmap_lock, flags) | |
145 | #else | |
146 | #define lock_kmap() spin_lock(&kmap_lock) | |
147 | #define unlock_kmap() spin_unlock(&kmap_lock) | |
148 | #define lock_kmap_any(flags) \ | |
149 | do { spin_lock(&kmap_lock); (void)(flags); } while (0) | |
150 | #define unlock_kmap_any(flags) \ | |
151 | do { spin_unlock(&kmap_lock); (void)(flags); } while (0) | |
152 | #endif | |
153 | ||
13f876ba | 154 | struct page *__kmap_to_page(void *vaddr) |
5a178119 | 155 | { |
ef6e06b2 IW |
156 | unsigned long base = (unsigned long) vaddr & PAGE_MASK; |
157 | struct kmap_ctrl *kctrl = ¤t->kmap_ctrl; | |
5a178119 | 158 | unsigned long addr = (unsigned long)vaddr; |
ef6e06b2 IW |
159 | int i; |
160 | ||
161 | /* kmap() mappings */ | |
162 | if (WARN_ON_ONCE(addr >= PKMAP_ADDR(0) && | |
163 | addr < PKMAP_ADDR(LAST_PKMAP))) | |
164 | return pte_page(pkmap_page_table[PKMAP_NR(addr)]); | |
5a178119 | 165 | |
ef6e06b2 IW |
166 | /* kmap_local_page() mappings */ |
167 | if (WARN_ON_ONCE(base >= __fix_to_virt(FIX_KMAP_END) && | |
168 | base < __fix_to_virt(FIX_KMAP_BEGIN))) { | |
169 | for (i = 0; i < kctrl->idx; i++) { | |
170 | unsigned long base_addr; | |
171 | int idx; | |
9727688d | 172 | |
ef6e06b2 IW |
173 | idx = arch_kmap_local_map_idx(i, pte_pfn(pteval)); |
174 | base_addr = __fix_to_virt(FIX_KMAP_BEGIN + idx); | |
175 | ||
176 | if (base_addr == base) | |
177 | return pte_page(kctrl->pteval[i]); | |
178 | } | |
5a178119 MG |
179 | } |
180 | ||
259ecb34 | 181 | return virt_to_page(vaddr); |
5a178119 | 182 | } |
13f876ba | 183 | EXPORT_SYMBOL(__kmap_to_page); |
5a178119 | 184 | |
1da177e4 LT |
185 | static void flush_all_zero_pkmaps(void) |
186 | { | |
187 | int i; | |
5843d9a4 | 188 | int need_flush = 0; |
1da177e4 LT |
189 | |
190 | flush_cache_kmaps(); | |
191 | ||
192 | for (i = 0; i < LAST_PKMAP; i++) { | |
193 | struct page *page; | |
194 | ||
195 | /* | |
196 | * zero means we don't have anything to do, | |
197 | * >1 means that it is still in use. Only | |
198 | * a count of 1 means that it is free but | |
199 | * needs to be unmapped | |
200 | */ | |
201 | if (pkmap_count[i] != 1) | |
202 | continue; | |
203 | pkmap_count[i] = 0; | |
204 | ||
205 | /* sanity check */ | |
75babcac | 206 | BUG_ON(pte_none(pkmap_page_table[i])); |
1da177e4 LT |
207 | |
208 | /* | |
209 | * Don't need an atomic fetch-and-clear op here; | |
210 | * no-one has the page mapped, and cannot get at | |
211 | * its virtual address (and hence PTE) without first | |
212 | * getting the kmap_lock (which is held here). | |
213 | * So no dangers, even with speculative execution. | |
214 | */ | |
215 | page = pte_page(pkmap_page_table[i]); | |
eb2db439 | 216 | pte_clear(&init_mm, PKMAP_ADDR(i), &pkmap_page_table[i]); |
1da177e4 LT |
217 | |
218 | set_page_address(page, NULL); | |
5843d9a4 | 219 | need_flush = 1; |
1da177e4 | 220 | } |
5843d9a4 NP |
221 | if (need_flush) |
222 | flush_tlb_kernel_range(PKMAP_ADDR(0), PKMAP_ADDR(LAST_PKMAP)); | |
1da177e4 LT |
223 | } |
224 | ||
13f876ba | 225 | void __kmap_flush_unused(void) |
ce6234b5 | 226 | { |
3297e760 | 227 | lock_kmap(); |
ce6234b5 | 228 | flush_all_zero_pkmaps(); |
3297e760 | 229 | unlock_kmap(); |
ce6234b5 JF |
230 | } |
231 | ||
1da177e4 LT |
232 | static inline unsigned long map_new_virtual(struct page *page) |
233 | { | |
234 | unsigned long vaddr; | |
235 | int count; | |
15de36a4 MF |
236 | unsigned int last_pkmap_nr; |
237 | unsigned int color = get_pkmap_color(page); | |
1da177e4 LT |
238 | |
239 | start: | |
15de36a4 | 240 | count = get_pkmap_entries_count(color); |
1da177e4 LT |
241 | /* Find an empty entry */ |
242 | for (;;) { | |
15de36a4 MF |
243 | last_pkmap_nr = get_next_pkmap_nr(color); |
244 | if (no_more_pkmaps(last_pkmap_nr, color)) { | |
1da177e4 | 245 | flush_all_zero_pkmaps(); |
15de36a4 | 246 | count = get_pkmap_entries_count(color); |
1da177e4 LT |
247 | } |
248 | if (!pkmap_count[last_pkmap_nr]) | |
249 | break; /* Found a usable entry */ | |
250 | if (--count) | |
251 | continue; | |
252 | ||
253 | /* | |
254 | * Sleep for somebody else to unmap their entries | |
255 | */ | |
256 | { | |
257 | DECLARE_WAITQUEUE(wait, current); | |
15de36a4 MF |
258 | wait_queue_head_t *pkmap_map_wait = |
259 | get_pkmap_wait_queue_head(color); | |
1da177e4 LT |
260 | |
261 | __set_current_state(TASK_UNINTERRUPTIBLE); | |
15de36a4 | 262 | add_wait_queue(pkmap_map_wait, &wait); |
3297e760 | 263 | unlock_kmap(); |
1da177e4 | 264 | schedule(); |
15de36a4 | 265 | remove_wait_queue(pkmap_map_wait, &wait); |
3297e760 | 266 | lock_kmap(); |
1da177e4 LT |
267 | |
268 | /* Somebody else might have mapped it while we slept */ | |
269 | if (page_address(page)) | |
270 | return (unsigned long)page_address(page); | |
271 | ||
272 | /* Re-start */ | |
273 | goto start; | |
274 | } | |
275 | } | |
276 | vaddr = PKMAP_ADDR(last_pkmap_nr); | |
277 | set_pte_at(&init_mm, vaddr, | |
278 | &(pkmap_page_table[last_pkmap_nr]), mk_pte(page, kmap_prot)); | |
279 | ||
280 | pkmap_count[last_pkmap_nr] = 1; | |
281 | set_page_address(page, (void *)vaddr); | |
282 | ||
283 | return vaddr; | |
284 | } | |
285 | ||
77f6078a RD |
286 | /** |
287 | * kmap_high - map a highmem page into memory | |
288 | * @page: &struct page to map | |
289 | * | |
290 | * Returns the page's virtual memory address. | |
291 | * | |
292 | * We cannot call this from interrupts, as it may block. | |
293 | */ | |
920c7a5d | 294 | void *kmap_high(struct page *page) |
1da177e4 LT |
295 | { |
296 | unsigned long vaddr; | |
297 | ||
298 | /* | |
299 | * For highmem pages, we can't trust "virtual" until | |
300 | * after we have the lock. | |
1da177e4 | 301 | */ |
3297e760 | 302 | lock_kmap(); |
1da177e4 LT |
303 | vaddr = (unsigned long)page_address(page); |
304 | if (!vaddr) | |
305 | vaddr = map_new_virtual(page); | |
306 | pkmap_count[PKMAP_NR(vaddr)]++; | |
75babcac | 307 | BUG_ON(pkmap_count[PKMAP_NR(vaddr)] < 2); |
3297e760 | 308 | unlock_kmap(); |
9727688d | 309 | return (void *) vaddr; |
1da177e4 | 310 | } |
1da177e4 LT |
311 | EXPORT_SYMBOL(kmap_high); |
312 | ||
3297e760 NP |
313 | #ifdef ARCH_NEEDS_KMAP_HIGH_GET |
314 | /** | |
315 | * kmap_high_get - pin a highmem page into memory | |
316 | * @page: &struct page to pin | |
317 | * | |
318 | * Returns the page's current virtual memory address, or NULL if no mapping | |
5e39df56 | 319 | * exists. If and only if a non null address is returned then a |
3297e760 NP |
320 | * matching call to kunmap_high() is necessary. |
321 | * | |
322 | * This can be called from any context. | |
323 | */ | |
324 | void *kmap_high_get(struct page *page) | |
325 | { | |
326 | unsigned long vaddr, flags; | |
327 | ||
328 | lock_kmap_any(flags); | |
329 | vaddr = (unsigned long)page_address(page); | |
330 | if (vaddr) { | |
331 | BUG_ON(pkmap_count[PKMAP_NR(vaddr)] < 1); | |
332 | pkmap_count[PKMAP_NR(vaddr)]++; | |
333 | } | |
334 | unlock_kmap_any(flags); | |
9727688d | 335 | return (void *) vaddr; |
3297e760 NP |
336 | } |
337 | #endif | |
338 | ||
77f6078a | 339 | /** |
4e9dc5df | 340 | * kunmap_high - unmap a highmem page into memory |
77f6078a | 341 | * @page: &struct page to unmap |
3297e760 NP |
342 | * |
343 | * If ARCH_NEEDS_KMAP_HIGH_GET is not defined then this may be called | |
344 | * only from user context. | |
77f6078a | 345 | */ |
920c7a5d | 346 | void kunmap_high(struct page *page) |
1da177e4 LT |
347 | { |
348 | unsigned long vaddr; | |
349 | unsigned long nr; | |
3297e760 | 350 | unsigned long flags; |
1da177e4 | 351 | int need_wakeup; |
15de36a4 MF |
352 | unsigned int color = get_pkmap_color(page); |
353 | wait_queue_head_t *pkmap_map_wait; | |
1da177e4 | 354 | |
3297e760 | 355 | lock_kmap_any(flags); |
1da177e4 | 356 | vaddr = (unsigned long)page_address(page); |
75babcac | 357 | BUG_ON(!vaddr); |
1da177e4 LT |
358 | nr = PKMAP_NR(vaddr); |
359 | ||
360 | /* | |
361 | * A count must never go down to zero | |
362 | * without a TLB flush! | |
363 | */ | |
364 | need_wakeup = 0; | |
365 | switch (--pkmap_count[nr]) { | |
366 | case 0: | |
367 | BUG(); | |
368 | case 1: | |
369 | /* | |
370 | * Avoid an unnecessary wake_up() function call. | |
371 | * The common case is pkmap_count[] == 1, but | |
372 | * no waiters. | |
373 | * The tasks queued in the wait-queue are guarded | |
374 | * by both the lock in the wait-queue-head and by | |
375 | * the kmap_lock. As the kmap_lock is held here, | |
376 | * no need for the wait-queue-head's lock. Simply | |
377 | * test if the queue is empty. | |
378 | */ | |
15de36a4 MF |
379 | pkmap_map_wait = get_pkmap_wait_queue_head(color); |
380 | need_wakeup = waitqueue_active(pkmap_map_wait); | |
1da177e4 | 381 | } |
3297e760 | 382 | unlock_kmap_any(flags); |
1da177e4 LT |
383 | |
384 | /* do wake-up, if needed, race-free outside of the spin lock */ | |
385 | if (need_wakeup) | |
15de36a4 | 386 | wake_up(pkmap_map_wait); |
1da177e4 | 387 | } |
1da177e4 | 388 | EXPORT_SYMBOL(kunmap_high); |
0060ef3b | 389 | |
0060ef3b MWO |
390 | void zero_user_segments(struct page *page, unsigned start1, unsigned end1, |
391 | unsigned start2, unsigned end2) | |
392 | { | |
393 | unsigned int i; | |
394 | ||
395 | BUG_ON(end1 > page_size(page) || end2 > page_size(page)); | |
396 | ||
184cee51 OH |
397 | if (start1 >= end1) |
398 | start1 = end1 = 0; | |
399 | if (start2 >= end2) | |
400 | start2 = end2 = 0; | |
401 | ||
0060ef3b MWO |
402 | for (i = 0; i < compound_nr(page); i++) { |
403 | void *kaddr = NULL; | |
404 | ||
0060ef3b MWO |
405 | if (start1 >= PAGE_SIZE) { |
406 | start1 -= PAGE_SIZE; | |
407 | end1 -= PAGE_SIZE; | |
408 | } else { | |
409 | unsigned this_end = min_t(unsigned, end1, PAGE_SIZE); | |
410 | ||
184cee51 | 411 | if (end1 > start1) { |
d2c20e51 | 412 | kaddr = kmap_local_page(page + i); |
0060ef3b | 413 | memset(kaddr + start1, 0, this_end - start1); |
184cee51 | 414 | } |
0060ef3b MWO |
415 | end1 -= this_end; |
416 | start1 = 0; | |
417 | } | |
418 | ||
419 | if (start2 >= PAGE_SIZE) { | |
420 | start2 -= PAGE_SIZE; | |
421 | end2 -= PAGE_SIZE; | |
422 | } else { | |
423 | unsigned this_end = min_t(unsigned, end2, PAGE_SIZE); | |
424 | ||
184cee51 OH |
425 | if (end2 > start2) { |
426 | if (!kaddr) | |
d2c20e51 | 427 | kaddr = kmap_local_page(page + i); |
0060ef3b | 428 | memset(kaddr + start2, 0, this_end - start2); |
184cee51 | 429 | } |
0060ef3b MWO |
430 | end2 -= this_end; |
431 | start2 = 0; | |
432 | } | |
433 | ||
434 | if (kaddr) { | |
d2c20e51 | 435 | kunmap_local(kaddr); |
0060ef3b MWO |
436 | flush_dcache_page(page + i); |
437 | } | |
438 | ||
439 | if (!end1 && !end2) | |
440 | break; | |
441 | } | |
442 | ||
443 | BUG_ON((start1 | start2 | end1 | end2) != 0); | |
444 | } | |
445 | EXPORT_SYMBOL(zero_user_segments); | |
298fa1ad TG |
446 | #endif /* CONFIG_HIGHMEM */ |
447 | ||
448 | #ifdef CONFIG_KMAP_LOCAL | |
449 | ||
450 | #include <asm/kmap_size.h> | |
451 | ||
389755c2 | 452 | /* |
6e799cb6 | 453 | * With DEBUG_KMAP_LOCAL the stack depth is doubled and every second |
389755c2 TG |
454 | * slot is unused which acts as a guard page |
455 | */ | |
6e799cb6 | 456 | #ifdef CONFIG_DEBUG_KMAP_LOCAL |
389755c2 TG |
457 | # define KM_INCR 2 |
458 | #else | |
459 | # define KM_INCR 1 | |
460 | #endif | |
461 | ||
298fa1ad TG |
462 | static inline int kmap_local_idx_push(void) |
463 | { | |
ea0eafea | 464 | WARN_ON_ONCE(in_hardirq() && !irqs_disabled()); |
5fbda3ec TG |
465 | current->kmap_ctrl.idx += KM_INCR; |
466 | BUG_ON(current->kmap_ctrl.idx >= KM_MAX_IDX); | |
467 | return current->kmap_ctrl.idx - 1; | |
298fa1ad TG |
468 | } |
469 | ||
470 | static inline int kmap_local_idx(void) | |
471 | { | |
5fbda3ec | 472 | return current->kmap_ctrl.idx - 1; |
298fa1ad TG |
473 | } |
474 | ||
475 | static inline void kmap_local_idx_pop(void) | |
476 | { | |
5fbda3ec TG |
477 | current->kmap_ctrl.idx -= KM_INCR; |
478 | BUG_ON(current->kmap_ctrl.idx < 0); | |
298fa1ad TG |
479 | } |
480 | ||
481 | #ifndef arch_kmap_local_post_map | |
482 | # define arch_kmap_local_post_map(vaddr, pteval) do { } while (0) | |
483 | #endif | |
3c1016b5 | 484 | |
298fa1ad TG |
485 | #ifndef arch_kmap_local_pre_unmap |
486 | # define arch_kmap_local_pre_unmap(vaddr) do { } while (0) | |
487 | #endif | |
488 | ||
489 | #ifndef arch_kmap_local_post_unmap | |
490 | # define arch_kmap_local_post_unmap(vaddr) do { } while (0) | |
491 | #endif | |
492 | ||
298fa1ad TG |
493 | #ifndef arch_kmap_local_unmap_idx |
494 | #define arch_kmap_local_unmap_idx(idx, vaddr) kmap_local_calc_idx(idx) | |
495 | #endif | |
496 | ||
497 | #ifndef arch_kmap_local_high_get | |
498 | static inline void *arch_kmap_local_high_get(struct page *page) | |
499 | { | |
500 | return NULL; | |
501 | } | |
502 | #endif | |
503 | ||
a1dce7fd TG |
504 | #ifndef arch_kmap_local_set_pte |
505 | #define arch_kmap_local_set_pte(mm, vaddr, ptep, ptev) \ | |
506 | set_pte_at(mm, vaddr, ptep, ptev) | |
507 | #endif | |
508 | ||
298fa1ad | 509 | /* Unmap a local mapping which was obtained by kmap_high_get() */ |
2a656cad | 510 | static inline bool kmap_high_unmap_local(unsigned long vaddr) |
298fa1ad TG |
511 | { |
512 | #ifdef ARCH_NEEDS_KMAP_HIGH_GET | |
2a656cad | 513 | if (vaddr >= PKMAP_ADDR(0) && vaddr < PKMAP_ADDR(LAST_PKMAP)) { |
298fa1ad | 514 | kunmap_high(pte_page(pkmap_page_table[PKMAP_NR(vaddr)])); |
2a656cad TG |
515 | return true; |
516 | } | |
298fa1ad | 517 | #endif |
2a656cad | 518 | return false; |
298fa1ad TG |
519 | } |
520 | ||
298fa1ad TG |
521 | static pte_t *__kmap_pte; |
522 | ||
825c43f5 | 523 | static pte_t *kmap_get_pte(unsigned long vaddr, int idx) |
298fa1ad | 524 | { |
825c43f5 AB |
525 | if (IS_ENABLED(CONFIG_KMAP_LOCAL_NON_LINEAR_PTE_ARRAY)) |
526 | /* | |
527 | * Set by the arch if __kmap_pte[-idx] does not produce | |
528 | * the correct entry. | |
529 | */ | |
530 | return virt_to_kpte(vaddr); | |
298fa1ad TG |
531 | if (!__kmap_pte) |
532 | __kmap_pte = virt_to_kpte(__fix_to_virt(FIX_KMAP_BEGIN)); | |
825c43f5 | 533 | return &__kmap_pte[-idx]; |
298fa1ad TG |
534 | } |
535 | ||
536 | void *__kmap_local_pfn_prot(unsigned long pfn, pgprot_t prot) | |
537 | { | |
825c43f5 | 538 | pte_t pteval, *kmap_pte; |
298fa1ad TG |
539 | unsigned long vaddr; |
540 | int idx; | |
541 | ||
f3ba3c71 TG |
542 | /* |
543 | * Disable migration so resulting virtual address is stable | |
f0953a1b | 544 | * across preemption. |
f3ba3c71 TG |
545 | */ |
546 | migrate_disable(); | |
298fa1ad TG |
547 | preempt_disable(); |
548 | idx = arch_kmap_local_map_idx(kmap_local_idx_push(), pfn); | |
549 | vaddr = __fix_to_virt(FIX_KMAP_BEGIN + idx); | |
825c43f5 AB |
550 | kmap_pte = kmap_get_pte(vaddr, idx); |
551 | BUG_ON(!pte_none(*kmap_pte)); | |
298fa1ad | 552 | pteval = pfn_pte(pfn, prot); |
825c43f5 | 553 | arch_kmap_local_set_pte(&init_mm, vaddr, kmap_pte, pteval); |
298fa1ad | 554 | arch_kmap_local_post_map(vaddr, pteval); |
5fbda3ec | 555 | current->kmap_ctrl.pteval[kmap_local_idx()] = pteval; |
298fa1ad TG |
556 | preempt_enable(); |
557 | ||
558 | return (void *)vaddr; | |
559 | } | |
560 | EXPORT_SYMBOL_GPL(__kmap_local_pfn_prot); | |
561 | ||
562 | void *__kmap_local_page_prot(struct page *page, pgprot_t prot) | |
563 | { | |
564 | void *kmap; | |
565 | ||
0e91a0c6 TG |
566 | /* |
567 | * To broaden the usage of the actual kmap_local() machinery always map | |
568 | * pages when debugging is enabled and the architecture has no problems | |
569 | * with alias mappings. | |
570 | */ | |
571 | if (!IS_ENABLED(CONFIG_DEBUG_KMAP_LOCAL_FORCE_MAP) && !PageHighMem(page)) | |
298fa1ad TG |
572 | return page_address(page); |
573 | ||
574 | /* Try kmap_high_get() if architecture has it enabled */ | |
575 | kmap = arch_kmap_local_high_get(page); | |
576 | if (kmap) | |
577 | return kmap; | |
578 | ||
579 | return __kmap_local_pfn_prot(page_to_pfn(page), prot); | |
580 | } | |
581 | EXPORT_SYMBOL(__kmap_local_page_prot); | |
582 | ||
39ade048 | 583 | void kunmap_local_indexed(const void *vaddr) |
298fa1ad TG |
584 | { |
585 | unsigned long addr = (unsigned long) vaddr & PAGE_MASK; | |
825c43f5 | 586 | pte_t *kmap_pte; |
298fa1ad TG |
587 | int idx; |
588 | ||
589 | if (addr < __fix_to_virt(FIX_KMAP_END) || | |
590 | addr > __fix_to_virt(FIX_KMAP_BEGIN)) { | |
0e91a0c6 TG |
591 | if (IS_ENABLED(CONFIG_DEBUG_KMAP_LOCAL_FORCE_MAP)) { |
592 | /* This _should_ never happen! See above. */ | |
593 | WARN_ON_ONCE(1); | |
594 | return; | |
595 | } | |
2a656cad TG |
596 | /* |
597 | * Handle mappings which were obtained by kmap_high_get() | |
598 | * first as the virtual address of such mappings is below | |
599 | * PAGE_OFFSET. Warn for all other addresses which are in | |
600 | * the user space part of the virtual address space. | |
601 | */ | |
602 | if (!kmap_high_unmap_local(addr)) | |
603 | WARN_ON_ONCE(addr < PAGE_OFFSET); | |
298fa1ad TG |
604 | return; |
605 | } | |
606 | ||
607 | preempt_disable(); | |
608 | idx = arch_kmap_local_unmap_idx(kmap_local_idx(), addr); | |
609 | WARN_ON_ONCE(addr != __fix_to_virt(FIX_KMAP_BEGIN + idx)); | |
610 | ||
825c43f5 | 611 | kmap_pte = kmap_get_pte(addr, idx); |
298fa1ad | 612 | arch_kmap_local_pre_unmap(addr); |
825c43f5 | 613 | pte_clear(&init_mm, addr, kmap_pte); |
298fa1ad | 614 | arch_kmap_local_post_unmap(addr); |
5fbda3ec | 615 | current->kmap_ctrl.pteval[kmap_local_idx()] = __pte(0); |
298fa1ad TG |
616 | kmap_local_idx_pop(); |
617 | preempt_enable(); | |
f3ba3c71 | 618 | migrate_enable(); |
298fa1ad TG |
619 | } |
620 | EXPORT_SYMBOL(kunmap_local_indexed); | |
5fbda3ec TG |
621 | |
622 | /* | |
623 | * Invoked before switch_to(). This is safe even when during or after | |
624 | * clearing the maps an interrupt which needs a kmap_local happens because | |
625 | * the task::kmap_ctrl.idx is not modified by the unmapping code so a | |
626 | * nested kmap_local will use the next unused index and restore the index | |
627 | * on unmap. The already cleared kmaps of the outgoing task are irrelevant | |
628 | * because the interrupt context does not know about them. The same applies | |
629 | * when scheduling back in for an interrupt which happens before the | |
630 | * restore is complete. | |
631 | */ | |
632 | void __kmap_local_sched_out(void) | |
633 | { | |
634 | struct task_struct *tsk = current; | |
825c43f5 | 635 | pte_t *kmap_pte; |
5fbda3ec TG |
636 | int i; |
637 | ||
638 | /* Clear kmaps */ | |
639 | for (i = 0; i < tsk->kmap_ctrl.idx; i++) { | |
640 | pte_t pteval = tsk->kmap_ctrl.pteval[i]; | |
641 | unsigned long addr; | |
642 | int idx; | |
643 | ||
644 | /* With debug all even slots are unmapped and act as guard */ | |
487cfade | 645 | if (IS_ENABLED(CONFIG_DEBUG_KMAP_LOCAL) && !(i & 0x01)) { |
66f133ce | 646 | WARN_ON_ONCE(pte_val(pteval) != 0); |
5fbda3ec TG |
647 | continue; |
648 | } | |
649 | if (WARN_ON_ONCE(pte_none(pteval))) | |
650 | continue; | |
651 | ||
652 | /* | |
653 | * This is a horrible hack for XTENSA to calculate the | |
654 | * coloured PTE index. Uses the PFN encoded into the pteval | |
655 | * and the map index calculation because the actual mapped | |
656 | * virtual address is not stored in task::kmap_ctrl. | |
657 | * For any sane architecture this is optimized out. | |
658 | */ | |
659 | idx = arch_kmap_local_map_idx(i, pte_pfn(pteval)); | |
660 | ||
661 | addr = __fix_to_virt(FIX_KMAP_BEGIN + idx); | |
825c43f5 | 662 | kmap_pte = kmap_get_pte(addr, idx); |
5fbda3ec | 663 | arch_kmap_local_pre_unmap(addr); |
825c43f5 | 664 | pte_clear(&init_mm, addr, kmap_pte); |
5fbda3ec TG |
665 | arch_kmap_local_post_unmap(addr); |
666 | } | |
667 | } | |
668 | ||
669 | void __kmap_local_sched_in(void) | |
670 | { | |
671 | struct task_struct *tsk = current; | |
825c43f5 | 672 | pte_t *kmap_pte; |
5fbda3ec TG |
673 | int i; |
674 | ||
675 | /* Restore kmaps */ | |
676 | for (i = 0; i < tsk->kmap_ctrl.idx; i++) { | |
677 | pte_t pteval = tsk->kmap_ctrl.pteval[i]; | |
678 | unsigned long addr; | |
679 | int idx; | |
680 | ||
681 | /* With debug all even slots are unmapped and act as guard */ | |
487cfade | 682 | if (IS_ENABLED(CONFIG_DEBUG_KMAP_LOCAL) && !(i & 0x01)) { |
66f133ce | 683 | WARN_ON_ONCE(pte_val(pteval) != 0); |
5fbda3ec TG |
684 | continue; |
685 | } | |
686 | if (WARN_ON_ONCE(pte_none(pteval))) | |
687 | continue; | |
688 | ||
689 | /* See comment in __kmap_local_sched_out() */ | |
690 | idx = arch_kmap_local_map_idx(i, pte_pfn(pteval)); | |
691 | addr = __fix_to_virt(FIX_KMAP_BEGIN + idx); | |
825c43f5 AB |
692 | kmap_pte = kmap_get_pte(addr, idx); |
693 | set_pte_at(&init_mm, addr, kmap_pte, pteval); | |
5fbda3ec TG |
694 | arch_kmap_local_post_map(addr, pteval); |
695 | } | |
696 | } | |
697 | ||
698 | void kmap_local_fork(struct task_struct *tsk) | |
699 | { | |
700 | if (WARN_ON_ONCE(tsk->kmap_ctrl.idx)) | |
701 | memset(&tsk->kmap_ctrl, 0, sizeof(tsk->kmap_ctrl)); | |
702 | } | |
703 | ||
298fa1ad | 704 | #endif |
1da177e4 | 705 | |
1da177e4 LT |
706 | #if defined(HASHED_PAGE_VIRTUAL) |
707 | ||
708 | #define PA_HASH_ORDER 7 | |
709 | ||
710 | /* | |
711 | * Describes one page->virtual association | |
712 | */ | |
713 | struct page_address_map { | |
714 | struct page *page; | |
715 | void *virtual; | |
716 | struct list_head list; | |
717 | }; | |
718 | ||
a354e2c8 | 719 | static struct page_address_map page_address_maps[LAST_PKMAP]; |
1da177e4 LT |
720 | |
721 | /* | |
722 | * Hash table bucket | |
723 | */ | |
724 | static struct page_address_slot { | |
725 | struct list_head lh; /* List of page_address_maps */ | |
726 | spinlock_t lock; /* Protect this bucket's list */ | |
727 | } ____cacheline_aligned_in_smp page_address_htable[1<<PA_HASH_ORDER]; | |
728 | ||
f9918794 | 729 | static struct page_address_slot *page_slot(const struct page *page) |
1da177e4 LT |
730 | { |
731 | return &page_address_htable[hash_ptr(page, PA_HASH_ORDER)]; | |
732 | } | |
733 | ||
77f6078a RD |
734 | /** |
735 | * page_address - get the mapped virtual address of a page | |
736 | * @page: &struct page to get the virtual address of | |
737 | * | |
738 | * Returns the page's virtual address. | |
739 | */ | |
f9918794 | 740 | void *page_address(const struct page *page) |
1da177e4 LT |
741 | { |
742 | unsigned long flags; | |
743 | void *ret; | |
744 | struct page_address_slot *pas; | |
745 | ||
746 | if (!PageHighMem(page)) | |
747 | return lowmem_page_address(page); | |
748 | ||
749 | pas = page_slot(page); | |
750 | ret = NULL; | |
751 | spin_lock_irqsave(&pas->lock, flags); | |
752 | if (!list_empty(&pas->lh)) { | |
753 | struct page_address_map *pam; | |
754 | ||
755 | list_for_each_entry(pam, &pas->lh, list) { | |
756 | if (pam->page == page) { | |
757 | ret = pam->virtual; | |
7a3f2263 | 758 | break; |
1da177e4 LT |
759 | } |
760 | } | |
761 | } | |
7a3f2263 | 762 | |
1da177e4 LT |
763 | spin_unlock_irqrestore(&pas->lock, flags); |
764 | return ret; | |
765 | } | |
1da177e4 LT |
766 | EXPORT_SYMBOL(page_address); |
767 | ||
77f6078a RD |
768 | /** |
769 | * set_page_address - set a page's virtual address | |
770 | * @page: &struct page to set | |
771 | * @virtual: virtual address to use | |
772 | */ | |
1da177e4 LT |
773 | void set_page_address(struct page *page, void *virtual) |
774 | { | |
775 | unsigned long flags; | |
776 | struct page_address_slot *pas; | |
777 | struct page_address_map *pam; | |
778 | ||
779 | BUG_ON(!PageHighMem(page)); | |
780 | ||
781 | pas = page_slot(page); | |
782 | if (virtual) { /* Add */ | |
a354e2c8 | 783 | pam = &page_address_maps[PKMAP_NR((unsigned long)virtual)]; |
1da177e4 LT |
784 | pam->page = page; |
785 | pam->virtual = virtual; | |
786 | ||
787 | spin_lock_irqsave(&pas->lock, flags); | |
788 | list_add_tail(&pam->list, &pas->lh); | |
789 | spin_unlock_irqrestore(&pas->lock, flags); | |
790 | } else { /* Remove */ | |
791 | spin_lock_irqsave(&pas->lock, flags); | |
792 | list_for_each_entry(pam, &pas->lh, list) { | |
793 | if (pam->page == page) { | |
794 | list_del(&pam->list); | |
7a3f2263 | 795 | break; |
1da177e4 LT |
796 | } |
797 | } | |
798 | spin_unlock_irqrestore(&pas->lock, flags); | |
799 | } | |
7a3f2263 | 800 | |
1da177e4 LT |
801 | return; |
802 | } | |
803 | ||
1da177e4 LT |
804 | void __init page_address_init(void) |
805 | { | |
806 | int i; | |
807 | ||
1da177e4 LT |
808 | for (i = 0; i < ARRAY_SIZE(page_address_htable); i++) { |
809 | INIT_LIST_HEAD(&page_address_htable[i].lh); | |
810 | spin_lock_init(&page_address_htable[i].lock); | |
811 | } | |
1da177e4 LT |
812 | } |
813 | ||
955cc774 | 814 | #endif /* defined(HASHED_PAGE_VIRTUAL) */ |