xref: /dflybsd-src/sys/vm/vm_contig.c (revision dae741e33c840b92a8a53bf9f01157ede145e256)
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  * $DragonFly: src/sys/vm/vm_contig.c,v 1.21 2006/12/28 21:24:02 dillon Exp $
68  */
69 
70 /*
71  * Copyright (c) 1987, 1990 Carnegie-Mellon University.
72  * All rights reserved.
73  *
74  * Authors: Avadis Tevanian, Jr., Michael Wayne Young
75  *
76  * Permission to use, copy, modify and distribute this software and
77  * its documentation is hereby granted, provided that both the copyright
78  * notice and this permission notice appear in all copies of the
79  * software, derivative works or modified versions, and any portions
80  * thereof, and that both notices appear in supporting documentation.
81  *
82  * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
83  * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
84  * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
85  *
86  * Carnegie Mellon requests users of this software to return to
87  *
88  *  Software Distribution Coordinator  or  Software.Distribution@CS.CMU.EDU
89  *  School of Computer Science
90  *  Carnegie Mellon University
91  *  Pittsburgh PA 15213-3890
92  *
93  * any improvements or extensions that they make and grant Carnegie the
94  * rights to redistribute these changes.
95  */
96 
97 /*
98  * Contiguous memory allocation API.
99  */
100 
101 #include <sys/param.h>
102 #include <sys/systm.h>
103 #include <sys/malloc.h>
104 #include <sys/proc.h>
105 #include <sys/lock.h>
106 #include <sys/vmmeter.h>
107 #include <sys/vnode.h>
108 
109 #include <vm/vm.h>
110 #include <vm/vm_param.h>
111 #include <vm/vm_kern.h>
112 #include <vm/pmap.h>
113 #include <vm/vm_map.h>
114 #include <vm/vm_object.h>
115 #include <vm/vm_page.h>
116 #include <vm/vm_pageout.h>
117 #include <vm/vm_pager.h>
118 #include <vm/vm_extern.h>
119 
120 #include <sys/thread2.h>
121 #include <sys/spinlock2.h>
122 #include <vm/vm_page2.h>
123 
124 static void vm_contig_pg_free(int 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
143 vm_contig_pg_clean(int queue, int 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_BUSY | PG_FICTITIOUS | PG_MARKER;
155 	marker.queue = queue;
156 	marker.wire_count = 1;
157 
158 	vm_page_queues_spin_lock(queue);
159 	TAILQ_INSERT_HEAD(&pq->pl, &marker, pageq);
160 	vm_page_queues_spin_unlock(queue);
161 
162 	/*
163 	 * Iterate the queue.  Note that the vm_page spinlock must be
164 	 * acquired before the pageq spinlock so it's easiest to simply
165 	 * not hold it in the loop iteration.
166 	 */
167 	while (count-- > 0 && (m = TAILQ_NEXT(&marker, pageq)) != NULL) {
168 		vm_page_and_queue_spin_lock(m);
169 		if (m != TAILQ_NEXT(&marker, pageq)) {
170 			vm_page_and_queue_spin_unlock(m);
171 			++count;
172 			continue;
173 		}
174 		KKASSERT(m->queue == queue);
175 
176 		TAILQ_REMOVE(&pq->pl, &marker, pageq);
177 		TAILQ_INSERT_AFTER(&pq->pl, m, &marker, pageq);
178 
179 		if (m->flags & PG_MARKER) {
180 			vm_page_and_queue_spin_unlock(m);
181 			continue;
182 		}
183 		if (vm_page_busy_try(m, TRUE)) {
184 			vm_page_and_queue_spin_unlock(m);
185 			continue;
186 		}
187 		vm_page_and_queue_spin_unlock(m);
188 
189 		/*
190 		 * We've successfully busied the page
191 		 */
192 		if (m->queue - m->pc != queue) {
193 			vm_page_wakeup(m);
194 			continue;
195 		}
196 		if ((object = m->object) == NULL) {
197 			vm_page_wakeup(m);
198 			continue;
199 		}
200 		vm_page_test_dirty(m);
201 		if (m->dirty) {
202 			vm_object_hold(object);
203 			KKASSERT(m->object == object);
204 
205 			if (object->type == OBJT_VNODE) {
206 				vm_page_wakeup(m);
207 				vn_lock(object->handle, LK_EXCLUSIVE|LK_RETRY);
208 				vm_object_page_clean(object, 0, 0, OBJPC_SYNC);
209 				vn_unlock(((struct vnode *)object->handle));
210 			} else if (object->type == OBJT_SWAP ||
211 					object->type == OBJT_DEFAULT) {
212 				m_tmp = m;
213 				vm_pageout_flush(&m_tmp, 1, 0);
214 			} else {
215 				vm_page_wakeup(m);
216 			}
217 			vm_object_drop(object);
218 		} else if (m->hold_count == 0) {
219 			vm_page_cache(m);
220 		} else {
221 			vm_page_wakeup(m);
222 		}
223 	}
224 
225 	/*
226 	 * Scrap our local marker
227 	 */
228 	vm_page_queues_spin_lock(queue);
229 	TAILQ_REMOVE(&pq->pl, &marker, pageq);
230 	vm_page_queues_spin_unlock(queue);
231 }
232 
233 /*
234  * vm_contig_pg_alloc:
235  *
236  * Allocate contiguous pages from the VM.  This function does not
237  * map the allocated pages into the kernel map, otherwise it is
238  * impossible to make large allocations (i.e. >2G).
239  *
240  * Malloc()'s data structures have been used for collection of
241  * statistics and for allocations of less than a page.
242  */
243 static int
244 vm_contig_pg_alloc(unsigned long size, vm_paddr_t low, vm_paddr_t high,
245 		   unsigned long alignment, unsigned long boundary, int mflags)
246 {
247 	int i, q, start, pass;
248 	vm_offset_t phys;
249 	vm_page_t pga = vm_page_array;
250 	vm_page_t m;
251 	int pqtype;
252 
253 	size = round_page(size);
254 	if (size == 0)
255 		panic("vm_contig_pg_alloc: size must not be 0");
256 	if ((alignment & (alignment - 1)) != 0)
257 		panic("vm_contig_pg_alloc: alignment must be a power of 2");
258 	if ((boundary & (boundary - 1)) != 0)
259 		panic("vm_contig_pg_alloc: boundary must be a power of 2");
260 
261 	start = 0;
262 
263 	/*
264 	 * Three passes (0, 1, 2).  Each pass scans the VM page list for
265 	 * free or cached pages.  After each pass if the entire scan failed
266 	 * we attempt to flush inactive pages and reset the start index back
267 	 * to 0.  For passes 1 and 2 we also attempt to flush active pages.
268 	 */
269 	for (pass = 0; pass < 3; pass++) {
270 		/*
271 		 * Find first page in array that is free, within range,
272 		 * aligned, and such that the boundary won't be crossed.
273 		 */
274 again:
275 		for (i = start; i < vmstats.v_page_count; i++) {
276 			m = &pga[i];
277 			phys = VM_PAGE_TO_PHYS(m);
278 			pqtype = m->queue - m->pc;
279 			if (((pqtype == PQ_FREE) || (pqtype == PQ_CACHE)) &&
280 			    (phys >= low) && (phys < high) &&
281 			    ((phys & (alignment - 1)) == 0) &&
282 			    (((phys ^ (phys + size - 1)) & ~(boundary - 1)) == 0) &&
283 			    m->busy == 0 && m->wire_count == 0 &&
284 			    m->hold_count == 0 && (m->flags & PG_BUSY) == 0
285 
286 			) {
287 				break;
288 			}
289 		}
290 
291 		/*
292 		 * If we cannot find the page in the given range, or we have
293 		 * crossed the boundary, call the vm_contig_pg_clean() function
294 		 * for flushing out the queues, and returning it back to
295 		 * normal state.
296 		 */
297 		if ((i == vmstats.v_page_count) ||
298 		    ((VM_PAGE_TO_PHYS(&pga[i]) + size) > high)) {
299 
300 			/*
301 			 * Best effort flush of all inactive pages.
302 			 * This is quite quick, for now stall all
303 			 * callers, even if they've specified M_NOWAIT.
304 			 */
305 			for (q = 0; q < PQ_L2_SIZE; ++q) {
306 				vm_contig_pg_clean(PQ_INACTIVE + q,
307 						   vmstats.v_inactive_count);
308 				lwkt_yield();
309 			}
310 
311 			/*
312 			 * Best effort flush of active pages.
313 			 *
314 			 * This is very, very slow.
315 			 * Only do this if the caller has agreed to M_WAITOK.
316 			 *
317 			 * If enough pages are flushed, we may succeed on
318 			 * next (final) pass, if not the caller, contigmalloc(),
319 			 * will fail in the index < 0 case.
320 			 */
321 			if (pass > 0 && (mflags & M_WAITOK)) {
322 				for (q = 0; q < PQ_L2_SIZE; ++q) {
323 					vm_contig_pg_clean(PQ_ACTIVE + q,
324 						       vmstats.v_active_count);
325 				}
326 				lwkt_yield();
327 			}
328 
329 			/*
330 			 * We're already too high in the address space
331 			 * to succeed, reset to 0 for the next iteration.
332 			 */
333 			start = 0;
334 			continue;	/* next pass */
335 		}
336 		start = i;
337 
338 		/*
339 		 * Check successive pages for contiguous and free.
340 		 *
341 		 * (still in critical section)
342 		 */
343 		for (i = start + 1; i < (start + size / PAGE_SIZE); i++) {
344 			m = &pga[i];
345 			pqtype = m->queue - m->pc;
346 			if ((VM_PAGE_TO_PHYS(&m[0]) !=
347 			    (VM_PAGE_TO_PHYS(&m[-1]) + PAGE_SIZE)) ||
348 			    ((pqtype != PQ_FREE) && (pqtype != PQ_CACHE)) ||
349 			    m->busy || m->wire_count ||
350 			    m->hold_count || (m->flags & PG_BUSY)
351 			) {
352 				start++;
353 				goto again;
354 			}
355 		}
356 
357 		/*
358 		 * Try to allocate the pages.
359 		 *
360 		 * (still in critical section)
361 		 */
362 		for (i = start; i < (start + size / PAGE_SIZE); i++) {
363 			m = &pga[i];
364 
365 			if (vm_page_busy_try(m, TRUE)) {
366 				vm_contig_pg_free(start,
367 						  (i - start) * PAGE_SIZE);
368 				start++;
369 				goto again;
370 			}
371 			pqtype = m->queue - m->pc;
372 			if (pqtype == PQ_CACHE) {
373 				vm_page_free(m);
374 				--i;
375 				continue;	/* retry the page */
376 			}
377 			if (pqtype != PQ_FREE) {
378 				vm_page_wakeup(m);
379 				vm_contig_pg_free(start,
380 						  (i - start) * PAGE_SIZE);
381 				start++;
382 				goto again;
383 			}
384 			KKASSERT(m->object == NULL);
385 			vm_page_unqueue_nowakeup(m);
386 			m->valid = VM_PAGE_BITS_ALL;
387 			if (m->flags & PG_ZERO)
388 				vm_page_zero_count--;
389 			KASSERT(m->dirty == 0,
390 				("vm_contig_pg_alloc: page %p was dirty", m));
391 			KKASSERT(m->wire_count == 0);
392 			KKASSERT(m->busy == 0);
393 
394 			/*
395 			 * Clear all flags except PG_BUSY, PG_ZERO, and
396 			 * PG_WANTED, then unbusy the now allocated page.
397 			 */
398 			vm_page_flag_clear(m, ~(PG_BUSY|PG_ZERO|PG_WANTED));
399 			vm_page_wakeup(m);
400 		}
401 
402 		/*
403 		 * Our job is done, return the index page of vm_page_array.
404 		 */
405 		return (start); /* aka &pga[start] */
406 	}
407 
408 	/*
409 	 * Failed.
410 	 */
411 	return (-1);
412 }
413 
414 /*
415  * vm_contig_pg_free:
416  *
417  * Remove pages previously allocated by vm_contig_pg_alloc, and
418  * assume all references to the pages have been removed, and that
419  * it is OK to add them back to the free list.
420  *
421  * Caller must ensure no races on the page range in question.
422  * No other requirements.
423  */
424 static void
425 vm_contig_pg_free(int start, u_long size)
426 {
427 	vm_page_t pga = vm_page_array;
428 	vm_page_t m;
429 	int i;
430 
431 	size = round_page(size);
432 	if (size == 0)
433 		panic("vm_contig_pg_free: size must not be 0");
434 
435 	for (i = start; i < (start + size / PAGE_SIZE); i++) {
436 		m = &pga[i];
437 		vm_page_busy_wait(m, FALSE, "cpgfr");
438 		vm_page_free(m);
439 	}
440 }
441 
442 /*
443  * vm_contig_pg_kmap:
444  *
445  * Map previously allocated (vm_contig_pg_alloc) range of pages from
446  * vm_page_array[] into the KVA.  Once mapped, the pages are part of
447  * the Kernel, and are to free'ed with kmem_free(&kernel_map, addr, size).
448  *
449  * No requirements.
450  */
451 static vm_offset_t
452 vm_contig_pg_kmap(int start, u_long size, vm_map_t map, int flags)
453 {
454 	vm_offset_t addr, tmp_addr;
455 	vm_page_t pga = vm_page_array;
456 	int i, count;
457 
458 	size = round_page(size);
459 	if (size == 0)
460 		panic("vm_contig_pg_kmap: size must not be 0");
461 
462 	/*
463 	 * We've found a contiguous chunk that meets our requirements.
464 	 * Allocate KVM, and assign phys pages and return a kernel VM
465 	 * pointer.
466 	 */
467 	count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
468 	vm_map_lock(map);
469 	if (vm_map_findspace(map, vm_map_min(map), size, PAGE_SIZE, 0, &addr) !=
470 	    KERN_SUCCESS) {
471 		/*
472 		 * XXX We almost never run out of kernel virtual
473 		 * space, so we don't make the allocated memory
474 		 * above available.
475 		 */
476 		vm_map_unlock(map);
477 		vm_map_entry_release(count);
478 		return (0);
479 	}
480 
481 	/*
482 	 * kernel_object maps 1:1 to kernel_map.
483 	 */
484 	vm_object_hold(&kernel_object);
485 	vm_object_reference_locked(&kernel_object);
486 	vm_map_insert(map, &count,
487 		      &kernel_object, addr,
488 		      addr, addr + size,
489 		      VM_MAPTYPE_NORMAL,
490 		      VM_PROT_ALL, VM_PROT_ALL,
491 		      0);
492 	vm_map_unlock(map);
493 	vm_map_entry_release(count);
494 
495 	tmp_addr = addr;
496 	for (i = start; i < (start + size / PAGE_SIZE); i++) {
497 		vm_page_t m = &pga[i];
498 		if (vm_page_insert(m, &kernel_object, OFF_TO_IDX(tmp_addr)) ==
499 		    FALSE) {
500 			panic("vm_contig_pg_kmap: page already exists @%p",
501 			      (void *)(intptr_t)tmp_addr);
502 		}
503 		if ((flags & M_ZERO) && !(m->flags & PG_ZERO))
504 			pmap_zero_page(VM_PAGE_TO_PHYS(m));
505 		m->flags = 0;
506 		tmp_addr += PAGE_SIZE;
507  	}
508 	vm_map_wire(map, addr, addr + size, 0);
509 
510 	vm_object_drop(&kernel_object);
511 
512 	return (addr);
513 }
514 
515 /*
516  * No requirements.
517  */
518 void *
519 contigmalloc(
520 	unsigned long size,	/* should be size_t here and for malloc() */
521 	struct malloc_type *type,
522 	int flags,
523 	vm_paddr_t low,
524 	vm_paddr_t high,
525 	unsigned long alignment,
526 	unsigned long boundary)
527 {
528 	return contigmalloc_map(size, type, flags, low, high, alignment,
529 			boundary, &kernel_map);
530 }
531 
532 /*
533  * No requirements.
534  */
535 void *
536 contigmalloc_map(
537 	unsigned long size,	/* should be size_t here and for malloc() */
538 	struct malloc_type *type,
539 	int flags,
540 	vm_paddr_t low,
541 	vm_paddr_t high,
542 	unsigned long alignment,
543 	unsigned long boundary,
544 	vm_map_t map)
545 {
546 	int index;
547 	void *rv;
548 
549 	index = vm_contig_pg_alloc(size, low, high, alignment, boundary, flags);
550 	if (index < 0) {
551 		kprintf("contigmalloc_map: failed size %lu low=%llx "
552 			"high=%llx align=%lu boundary=%lu flags=%08x\n",
553 			size, (long long)low, (long long)high,
554 			alignment, boundary, flags);
555 		return NULL;
556 	}
557 
558 	rv = (void *)vm_contig_pg_kmap(index, size, map, flags);
559 	if (rv == NULL)
560 		vm_contig_pg_free(index, size);
561 
562 	return rv;
563 }
564 
565 /*
566  * No requirements.
567  */
568 void
569 contigfree(void *addr, unsigned long size, struct malloc_type *type)
570 {
571 	kmem_free(&kernel_map, (vm_offset_t)addr, size);
572 }
573 
574 /*
575  * No requirements.
576  */
577 vm_offset_t
578 vm_page_alloc_contig(
579 	vm_offset_t size,
580 	vm_paddr_t low,
581 	vm_paddr_t high,
582 	vm_offset_t alignment)
583 {
584 	return ((vm_offset_t)contigmalloc_map(size, M_DEVBUF, M_NOWAIT, low,
585 				high, alignment, 0ul, &kernel_map));
586 }
587