1 /*
2 * Copyright (c) 2003, 2004 The DragonFly Project. All rights reserved.
3 *
4 * This code is derived from software contributed to The DragonFly Project
5 * by Hiten Pandya <hmp@backplane.com>.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 *
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in
15 * the documentation and/or other materials provided with the
16 * distribution.
17 * 3. Neither the name of The DragonFly Project nor the names of its
18 * contributors may be used to endorse or promote products derived
19 * from this software without specific, prior written permission.
20 *
21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
24 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
25 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
26 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
27 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
28 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
29 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
30 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
31 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * SUCH DAMAGE.
33 *
34 */
35 /*
36 * Copyright (c) 1991 Regents of the University of California.
37 * All rights reserved.
38 *
39 * This code is derived from software contributed to Berkeley by
40 * The Mach Operating System project at Carnegie-Mellon University.
41 *
42 * Redistribution and use in source and binary forms, with or without
43 * modification, are permitted provided that the following conditions
44 * are met:
45 * 1. Redistributions of source code must retain the above copyright
46 * notice, this list of conditions and the following disclaimer.
47 * 2. Redistributions in binary form must reproduce the above copyright
48 * notice, this list of conditions and the following disclaimer in the
49 * documentation and/or other materials provided with the distribution.
50 * 3. Neither the name of the University nor the names of its contributors
51 * may be used to endorse or promote products derived from this software
52 * without specific prior written permission.
53 *
54 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
55 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
56 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
57 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
58 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
59 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
60 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
61 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
62 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
63 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
64 * SUCH DAMAGE.
65 *
66 * from: @(#)vm_page.c 7.4 (Berkeley) 5/7/91
67 */
68
69 /*
70 * Copyright (c) 1987, 1990 Carnegie-Mellon University.
71 * All rights reserved.
72 *
73 * Authors: Avadis Tevanian, Jr., Michael Wayne Young
74 *
75 * Permission to use, copy, modify and distribute this software and
76 * its documentation is hereby granted, provided that both the copyright
77 * notice and this permission notice appear in all copies of the
78 * software, derivative works or modified versions, and any portions
79 * thereof, and that both notices appear in supporting documentation.
80 *
81 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
82 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
83 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
84 *
85 * Carnegie Mellon requests users of this software to return to
86 *
87 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU
88 * School of Computer Science
89 * Carnegie Mellon University
90 * Pittsburgh PA 15213-3890
91 *
92 * any improvements or extensions that they make and grant Carnegie the
93 * rights to redistribute these changes.
94 */
95
96 /*
97 * Contiguous memory allocation API.
98 */
99
100 #include <sys/param.h>
101 #include <sys/systm.h>
102 #include <sys/malloc.h>
103 #include <sys/proc.h>
104 #include <sys/lock.h>
105 #include <sys/vmmeter.h>
106 #include <sys/vnode.h>
107
108 #include <vm/vm.h>
109 #include <vm/vm_param.h>
110 #include <vm/vm_kern.h>
111 #include <vm/pmap.h>
112 #include <vm/vm_map.h>
113 #include <vm/vm_object.h>
114 #include <vm/vm_page.h>
115 #include <vm/vm_pageout.h>
116 #include <vm/vm_pager.h>
117 #include <vm/vm_extern.h>
118
119 #include <sys/spinlock2.h>
120 #include <vm/vm_page2.h>
121
122 #include <machine/bus_dma.h>
123
124 static void vm_contig_pg_free(vm_pindex_t start, u_long size);
125
126 /*
127 * vm_contig_pg_clean:
128 *
129 * Do a thorough cleanup of the specified 'queue', which can be either
130 * PQ_ACTIVE or PQ_INACTIVE by doing a walkthrough. If the page is not
131 * marked dirty, it is shoved into the page cache, provided no one has
132 * currently aqcuired it, otherwise localized action per object type
133 * is taken for cleanup:
134 *
135 * In the OBJT_VNODE case, the whole page range is cleaned up
136 * using the vm_object_page_clean() routine, by specyfing a
137 * start and end of '0'.
138 *
139 * Otherwise if the object is of any other type, the generic
140 * pageout (daemon) flush routine is invoked.
141 */
142 static void
vm_contig_pg_clean(int queue,vm_pindex_t count)143 vm_contig_pg_clean(int queue, vm_pindex_t count)
144 {
145 vm_object_t object;
146 vm_page_t m, m_tmp;
147 struct vm_page marker;
148 struct vpgqueues *pq = &vm_page_queues[queue];
149
150 /*
151 * Setup a local marker
152 */
153 bzero(&marker, sizeof(marker));
154 marker.flags = PG_FICTITIOUS | PG_MARKER;
155 marker.busy_count = PBUSY_LOCKED;
156 marker.queue = queue;
157 marker.wire_count = 1;
158
159 vm_page_queues_spin_lock(queue);
160 TAILQ_INSERT_HEAD(&pq->pl, &marker, pageq);
161 vm_page_queues_spin_unlock(queue);
162
163 /*
164 * Iterate the queue. Note that the vm_page spinlock must be
165 * acquired before the pageq spinlock so it's easiest to simply
166 * not hold it in the loop iteration.
167 */
168 while ((long)count-- > 0 &&
169 (m = TAILQ_NEXT(&marker, pageq)) != NULL) {
170 vm_page_and_queue_spin_lock(m);
171 if (m != TAILQ_NEXT(&marker, pageq)) {
172 vm_page_and_queue_spin_unlock(m);
173 ++count;
174 continue;
175 }
176 KKASSERT(m->queue == queue);
177
178 TAILQ_REMOVE(&pq->pl, &marker, pageq);
179 TAILQ_INSERT_AFTER(&pq->pl, m, &marker, pageq);
180
181 if (m->flags & PG_MARKER) {
182 vm_page_and_queue_spin_unlock(m);
183 continue;
184 }
185 if (vm_page_busy_try(m, TRUE)) {
186 vm_page_and_queue_spin_unlock(m);
187 continue;
188 }
189 vm_page_and_queue_spin_unlock(m);
190
191 /*
192 * We've successfully busied the page
193 */
194 if (m->queue - m->pc != queue) {
195 vm_page_wakeup(m);
196 continue;
197 }
198 if (m->wire_count || m->hold_count) {
199 vm_page_wakeup(m);
200 continue;
201 }
202 if ((object = m->object) == NULL) {
203 vm_page_wakeup(m);
204 continue;
205 }
206 vm_page_test_dirty(m);
207 if (m->dirty || (m->flags & PG_NEED_COMMIT)) {
208 vm_object_hold(object);
209 KKASSERT(m->object == object);
210
211 if (object->type == OBJT_VNODE) {
212 vm_page_wakeup(m);
213 vn_lock(object->handle, LK_EXCLUSIVE|LK_RETRY);
214 vm_object_page_clean(object, 0, 0, OBJPC_SYNC);
215 vn_unlock(((struct vnode *)object->handle));
216 } else if (object->type == OBJT_SWAP ||
217 object->type == OBJT_DEFAULT) {
218 m_tmp = m;
219 vm_pageout_flush(&m_tmp, 1, 0);
220 } else {
221 vm_page_wakeup(m);
222 }
223 vm_object_drop(object);
224 } else if (m->hold_count == 0) {
225 vm_page_cache(m);
226 } else {
227 vm_page_wakeup(m);
228 }
229 }
230
231 /*
232 * Scrap our local marker
233 */
234 vm_page_queues_spin_lock(queue);
235 TAILQ_REMOVE(&pq->pl, &marker, pageq);
236 vm_page_queues_spin_unlock(queue);
237 }
238
239 /*
240 * vm_contig_pg_alloc:
241 *
242 * Allocate contiguous pages from the VM. This function does not
243 * map the allocated pages into the kernel map, otherwise it is
244 * impossible to make large allocations (i.e. >2G).
245 *
246 * Malloc()'s data structures have been used for collection of
247 * statistics and for allocations of less than a page.
248 */
249 static vm_pindex_t
vm_contig_pg_alloc(unsigned long size,vm_paddr_t low,vm_paddr_t high,unsigned long alignment,unsigned long boundary,int mflags)250 vm_contig_pg_alloc(unsigned long size, vm_paddr_t low, vm_paddr_t high,
251 unsigned long alignment, unsigned long boundary, int mflags)
252 {
253 vm_pindex_t i, q, start;
254 vm_offset_t phys;
255 vm_page_t pga = vm_page_array;
256 vm_page_t m;
257 int pass;
258 int pqtype;
259
260 size = round_page(size);
261 if (size == 0)
262 panic("vm_contig_pg_alloc: size must not be 0");
263 if ((alignment & (alignment - 1)) != 0)
264 panic("vm_contig_pg_alloc: alignment must be a power of 2");
265 if ((boundary & (boundary - 1)) != 0)
266 panic("vm_contig_pg_alloc: boundary must be a power of 2");
267
268 /*
269 * See if we can get the pages from the contiguous page reserve
270 * alist. The returned pages will be allocated and wired but not
271 * busied.
272 *
273 * If high is not set to BUS_SPACE_MAXADDR we try using our
274 * free memory reserve first, otherwise we try it last.
275 *
276 * XXX Always use the dma reserve first for performance, until
277 * we find a better way to differentiate the DRM API.
278 */
279 #if 0
280 if (high != BUS_SPACE_MAXADDR)
281 #endif
282 {
283 m = vm_page_alloc_contig(
284 low, high, alignment, boundary,
285 size, VM_MEMATTR_DEFAULT);
286 if (m)
287 return (m - &pga[0]);
288 }
289
290 /*
291 * Three passes (0, 1, 2). Each pass scans the VM page list for
292 * free or cached pages. After each pass if the entire scan failed
293 * we attempt to flush inactive pages and reset the start index back
294 * to 0. For passes 1 and 2 we also attempt to flush active pages.
295 */
296 start = 0;
297 for (pass = 0; pass < 3; pass++) {
298 /*
299 * Find first page in array that is free, within range,
300 * aligned, and such that the boundary won't be crossed.
301 */
302 again:
303 for (i = start; i < vmstats.v_page_count; i++) {
304 m = &pga[i];
305 phys = VM_PAGE_TO_PHYS(m);
306 pqtype = m->queue - m->pc;
307 if (((pqtype == PQ_FREE) || (pqtype == PQ_CACHE)) &&
308 (phys >= low) && (phys < high) &&
309 ((phys & (alignment - 1)) == 0) &&
310 ((rounddown2(phys ^ (phys + size - 1), boundary)) == 0) &&
311 m->wire_count == 0 && m->hold_count == 0 &&
312 (m->busy_count &
313 (PBUSY_LOCKED | PBUSY_MASK)) == 0 &&
314 (m->flags & PG_NEED_COMMIT) == 0)
315 {
316 break;
317 }
318 }
319
320 /*
321 * If we cannot find the page in the given range, or we have
322 * crossed the boundary, call the vm_contig_pg_clean() function
323 * for flushing out the queues, and returning it back to
324 * normal state.
325 */
326 if ((i == vmstats.v_page_count) ||
327 ((VM_PAGE_TO_PHYS(&pga[i]) + size) > high)) {
328
329 /*
330 * Best effort flush of all inactive pages.
331 * This is quite quick, for now stall all
332 * callers, even if they've specified M_NOWAIT.
333 */
334 for (q = 0; q < PQ_L2_SIZE; ++q) {
335 vm_contig_pg_clean(PQ_INACTIVE + q,
336 vmstats.v_inactive_count);
337 lwkt_yield();
338 }
339
340 /*
341 * Best effort flush of active pages.
342 *
343 * This is very, very slow.
344 * Only do this if the caller has agreed to M_WAITOK.
345 *
346 * If enough pages are flushed, we may succeed on
347 * next (final) pass, if not the caller, contigmalloc(),
348 * will fail in the index < 0 case.
349 */
350 if (pass > 0 && (mflags & M_WAITOK)) {
351 for (q = 0; q < PQ_L2_SIZE; ++q) {
352 vm_contig_pg_clean(PQ_ACTIVE + q,
353 vmstats.v_active_count);
354 }
355 lwkt_yield();
356 }
357
358 /*
359 * We're already too high in the address space
360 * to succeed, reset to 0 for the next iteration.
361 */
362 start = 0;
363 continue; /* next pass */
364 }
365 start = i;
366
367 /*
368 * Check successive pages for contiguous and free.
369 *
370 * (still in critical section)
371 */
372 for (i = start + 1; i < (start + size / PAGE_SIZE); i++) {
373 m = &pga[i];
374 pqtype = m->queue - m->pc;
375 if ((VM_PAGE_TO_PHYS(&m[0]) !=
376 (VM_PAGE_TO_PHYS(&m[-1]) + PAGE_SIZE)) ||
377 ((pqtype != PQ_FREE) && (pqtype != PQ_CACHE)) ||
378 m->wire_count ||
379 m->hold_count ||
380 (m->busy_count & (PBUSY_LOCKED | PBUSY_MASK)) ||
381 (m->flags & PG_NEED_COMMIT))
382 {
383 start++;
384 goto again;
385 }
386 }
387
388 /*
389 * Try to allocate the pages, wiring them as we go.
390 *
391 * (still in critical section)
392 */
393 for (i = start; i < (start + size / PAGE_SIZE); i++) {
394 m = &pga[i];
395
396 if (vm_page_busy_try(m, TRUE)) {
397 vm_contig_pg_free(start,
398 (i - start) * PAGE_SIZE);
399 start++;
400 goto again;
401 }
402 pqtype = m->queue - m->pc;
403 if (pqtype == PQ_CACHE &&
404 m->hold_count == 0 &&
405 m->wire_count == 0 &&
406 (m->flags & PG_NEED_COMMIT) == 0) {
407 vm_page_protect(m, VM_PROT_NONE);
408 KKASSERT((m->flags &
409 (PG_MAPPED | PG_UNQUEUED)) == 0);
410 KKASSERT(m->dirty == 0);
411 vm_page_free(m);
412 --i;
413 continue; /* retry the page */
414 }
415 if (pqtype != PQ_FREE || m->hold_count) {
416 vm_page_wakeup(m);
417 vm_contig_pg_free(start,
418 (i - start) * PAGE_SIZE);
419 start++;
420 goto again;
421 }
422 KKASSERT((m->valid & m->dirty) == 0);
423 KKASSERT(m->wire_count == 0);
424 KKASSERT(m->object == NULL);
425 vm_page_unqueue_nowakeup(m);
426 m->valid = VM_PAGE_BITS_ALL;
427 KASSERT(m->dirty == 0,
428 ("vm_contig_pg_alloc: page %p was dirty", m));
429 KKASSERT(m->wire_count == 0);
430 KKASSERT((m->busy_count & PBUSY_MASK) == 0);
431
432 /*
433 * Clear all flags, set FICTITIOUS and UNQUEUED to
434 * indicate the the pages are special, then unbusy
435 * the now allocated page.
436 *
437 * XXX setting FICTITIOUS and UNQUEUED in the future.
438 * (also pair up with vm_contig_pg_free)
439 */
440 vm_page_flag_clear(m, ~PG_KEEP_NEWPAGE_MASK);
441 /* vm_page_flag_set(m, PG_FICTITIOUS | PG_UNQUEUED);*/
442 vm_page_wire(m);
443 vm_page_wakeup(m);
444 }
445
446 /*
447 * Our job is done, return the index page of vm_page_array.
448 */
449 return (start); /* aka &pga[start] */
450 }
451
452 #if 0
453 /*
454 * Failed, if we haven't already tried, allocate from our reserved
455 * dma memory.
456 *
457 * XXX (see conditionalized code above)
458 */
459 if (high == BUS_SPACE_MAXADDR) {
460 m = vm_page_alloc_contig(
461 low, high, alignment, boundary,
462 size, VM_MEMATTR_DEFAULT);
463 if (m)
464 return (m - &pga[0]);
465 }
466 #endif
467
468 /*
469 * Failed.
470 */
471 return ((vm_pindex_t)-1);
472 }
473
474 /*
475 * vm_contig_pg_free:
476 *
477 * Remove pages previously allocated by vm_contig_pg_alloc, and
478 * assume all references to the pages have been removed, and that
479 * it is OK to add them back to the free list.
480 *
481 * Caller must ensure no races on the page range in question.
482 * No other requirements.
483 */
484 static void
vm_contig_pg_free(vm_pindex_t start,u_long size)485 vm_contig_pg_free(vm_pindex_t start, u_long size)
486 {
487 vm_page_t pga = vm_page_array;
488
489 size = round_page(size);
490 if (size == 0)
491 panic("vm_contig_pg_free: size must not be 0");
492
493 /*
494 * The pages are wired, vm_page_free_contig() determines whether they
495 * belong to the contig space or not and either frees them to that
496 * space (leaving them wired), or unwires the page and frees it to the
497 * normal PQ_FREE queue.
498 */
499 vm_page_free_contig(&pga[start], size);
500 }
501
502 /*
503 * vm_contig_pg_kmap:
504 *
505 * Map previously allocated (vm_contig_pg_alloc) range of pages from
506 * vm_page_array[] into the KVA. Once mapped, the pages are part of
507 * the Kernel, and are to free'ed with kmem_free(kernel_map, addr, size).
508 *
509 * No requirements.
510 */
511 static vm_offset_t
vm_contig_pg_kmap(vm_pindex_t start,u_long size,vm_map_t map,int flags)512 vm_contig_pg_kmap(vm_pindex_t start, u_long size, vm_map_t map, int flags)
513 {
514 vm_offset_t addr;
515 vm_paddr_t pa;
516 vm_page_t pga = vm_page_array;
517 u_long offset;
518
519 if (size == 0)
520 panic("vm_contig_pg_kmap: size must not be 0");
521 size = round_page(size);
522 addr = kmem_alloc_pageable(kernel_map, size, VM_SUBSYS_CONTIG);
523 if (addr) {
524 pa = VM_PAGE_TO_PHYS(&pga[start]);
525 for (offset = 0; offset < size; offset += PAGE_SIZE)
526 pmap_kenter_noinval(addr + offset, pa + offset);
527 pmap_invalidate_range(kernel_pmap, addr, addr + size);
528 if (flags & M_ZERO)
529 bzero((void *)addr, size);
530 }
531 return(addr);
532 }
533
534 /*
535 * No requirements.
536 */
537 void *
contigmalloc(unsigned long size,struct malloc_type * type,int flags,vm_paddr_t low,vm_paddr_t high,unsigned long alignment,unsigned long boundary)538 contigmalloc(
539 unsigned long size, /* should be size_t here and for malloc() */
540 struct malloc_type *type,
541 int flags,
542 vm_paddr_t low,
543 vm_paddr_t high,
544 unsigned long alignment,
545 unsigned long boundary)
546 {
547 return contigmalloc_map(size, type, flags, low, high, alignment,
548 boundary, kernel_map);
549 }
550
551 /*
552 * No requirements.
553 */
554 void *
contigmalloc_map(unsigned long size,struct malloc_type * type,int flags,vm_paddr_t low,vm_paddr_t high,unsigned long alignment,unsigned long boundary,vm_map_t map)555 contigmalloc_map(unsigned long size, struct malloc_type *type,
556 int flags, vm_paddr_t low, vm_paddr_t high,
557 unsigned long alignment, unsigned long boundary,
558 vm_map_t map)
559 {
560 vm_pindex_t index;
561 void *rv;
562
563 index = vm_contig_pg_alloc(size, low, high, alignment, boundary, flags);
564 if (index == (vm_pindex_t)-1) {
565 kprintf("contigmalloc_map: failed size %lu low=%llx "
566 "high=%llx align=%lu boundary=%lu flags=%08x\n",
567 size, (long long)low, (long long)high,
568 alignment, boundary, flags);
569 return NULL;
570 }
571
572 rv = (void *)vm_contig_pg_kmap(index, size, map, flags);
573 if (rv == NULL)
574 vm_contig_pg_free(index, size);
575
576 return rv;
577 }
578
579 /*
580 * No requirements.
581 */
582 void
contigfree(void * addr,unsigned long size,struct malloc_type * type)583 contigfree(void *addr, unsigned long size, struct malloc_type *type)
584 {
585 vm_paddr_t pa;
586 vm_page_t m;
587
588 if (size == 0)
589 panic("vm_contig_pg_kmap: size must not be 0");
590 size = round_page(size);
591
592 pa = pmap_kextract((vm_offset_t)addr);
593 pmap_qremove((vm_offset_t)addr, size / PAGE_SIZE);
594 kmem_free(kernel_map, (vm_offset_t)addr, size);
595
596 m = PHYS_TO_VM_PAGE(pa);
597 vm_page_free_contig(m, size);
598 }
599
600 /*
601 * No requirements.
602 */
603 vm_offset_t
kmem_alloc_contig(vm_offset_t size,vm_paddr_t low,vm_paddr_t high,vm_offset_t alignment)604 kmem_alloc_contig(vm_offset_t size, vm_paddr_t low, vm_paddr_t high,
605 vm_offset_t alignment)
606 {
607 return ((vm_offset_t)contigmalloc_map(size, M_DEVBUF, M_NOWAIT, low,
608 high, alignment, 0ul, kernel_map));
609 }
610