xref: /netbsd-src/sys/uvm/uvm_pager.c (revision 1ca5c1b28139779176bd5c13ad7c5f25c0bcd5f8)
1 /*	$NetBSD: uvm_pager.c,v 1.55 2001/12/31 19:21:38 chs Exp $	*/
2 
3 /*
4  *
5  * Copyright (c) 1997 Charles D. Cranor and Washington University.
6  * All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
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 the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. All advertising materials mentioning features or use of this software
17  *    must display the following acknowledgement:
18  *      This product includes software developed by Charles D. Cranor and
19  *      Washington University.
20  * 4. The name of the author may not be used to endorse or promote products
21  *    derived from this software without specific prior written permission.
22  *
23  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
24  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
25  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
26  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
27  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
28  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
29  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
30  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
31  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
32  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
33  *
34  * from: Id: uvm_pager.c,v 1.1.2.23 1998/02/02 20:38:06 chuck Exp
35  */
36 
37 /*
38  * uvm_pager.c: generic functions used to assist the pagers.
39  */
40 
41 #include <sys/cdefs.h>
42 __KERNEL_RCSID(0, "$NetBSD: uvm_pager.c,v 1.55 2001/12/31 19:21:38 chs Exp $");
43 
44 #include "opt_uvmhist.h"
45 
46 #include <sys/param.h>
47 #include <sys/systm.h>
48 #include <sys/proc.h>
49 #include <sys/malloc.h>
50 #include <sys/pool.h>
51 #include <sys/vnode.h>
52 
53 #define UVM_PAGER
54 #include <uvm/uvm.h>
55 
56 struct pool *uvm_aiobuf_pool;
57 
58 /*
59  * list of uvm pagers in the system
60  */
61 
62 extern struct uvm_pagerops uvm_deviceops;
63 extern struct uvm_pagerops uvm_vnodeops;
64 extern struct uvm_pagerops ubc_pager;
65 
66 struct uvm_pagerops *uvmpagerops[] = {
67 	&aobj_pager,
68 	&uvm_deviceops,
69 	&uvm_vnodeops,
70 	&ubc_pager,
71 };
72 
73 /*
74  * the pager map: provides KVA for I/O
75  */
76 
77 struct vm_map *pager_map;		/* XXX */
78 struct simplelock pager_map_wanted_lock;
79 boolean_t pager_map_wanted;	/* locked by pager map */
80 static vaddr_t emergva;
81 static boolean_t emerginuse;
82 
83 /*
84  * uvm_pager_init: init pagers (at boot time)
85  */
86 
87 void
88 uvm_pager_init()
89 {
90 	int lcv;
91 	vaddr_t sva, eva;
92 
93 	/*
94 	 * init pager map
95 	 */
96 
97 	sva = 0;
98 	pager_map = uvm_km_suballoc(kernel_map, &sva, &eva, PAGER_MAP_SIZE, 0,
99 	    FALSE, NULL);
100 	simple_lock_init(&pager_map_wanted_lock);
101 	pager_map_wanted = FALSE;
102 	emergva = uvm_km_valloc(kernel_map, MAXBSIZE);
103 	emerginuse = FALSE;
104 
105 	/*
106 	 * init ASYNC I/O queue
107 	 */
108 
109 	TAILQ_INIT(&uvm.aio_done);
110 
111 	/*
112 	 * call pager init functions
113 	 */
114 	for (lcv = 0 ; lcv < sizeof(uvmpagerops)/sizeof(struct uvm_pagerops *);
115 	    lcv++) {
116 		if (uvmpagerops[lcv]->pgo_init)
117 			uvmpagerops[lcv]->pgo_init();
118 	}
119 }
120 
121 /*
122  * uvm_pagermapin: map pages into KVA (pager_map) for I/O that needs mappings
123  *
124  * we basically just map in a blank map entry to reserve the space in the
125  * map and then use pmap_enter() to put the mappings in by hand.
126  */
127 
128 vaddr_t
129 uvm_pagermapin(pps, npages, flags)
130 	struct vm_page **pps;
131 	int npages;
132 	int flags;
133 {
134 	vsize_t size;
135 	vaddr_t kva;
136 	vaddr_t cva;
137 	struct vm_page *pp;
138 	vm_prot_t prot;
139 	UVMHIST_FUNC("uvm_pagermapin"); UVMHIST_CALLED(maphist);
140 
141 	UVMHIST_LOG(maphist,"(pps=0x%x, npages=%d)", pps, npages,0,0);
142 
143 	/*
144 	 * compute protection.  outgoing I/O only needs read
145 	 * access to the page, whereas incoming needs read/write.
146 	 */
147 
148 	prot = VM_PROT_READ;
149 	if (flags & UVMPAGER_MAPIN_READ)
150 		prot |= VM_PROT_WRITE;
151 
152 ReStart:
153 	size = npages << PAGE_SHIFT;
154 	kva = 0;			/* let system choose VA */
155 
156 	if (uvm_map(pager_map, &kva, size, NULL,
157 	      UVM_UNKNOWN_OFFSET, 0, UVM_FLAG_NOMERGE) != 0) {
158 		if (curproc == uvm.pagedaemon_proc) {
159 			simple_lock(&pager_map_wanted_lock);
160 			if (emerginuse) {
161 				UVM_UNLOCK_AND_WAIT(&emergva,
162 				    &pager_map_wanted_lock, FALSE,
163 				    "emergva", 0);
164 				goto ReStart;
165 			}
166 			emerginuse = TRUE;
167 			simple_unlock(&pager_map_wanted_lock);
168 			kva = emergva;
169 			KASSERT(npages <= MAXBSIZE >> PAGE_SHIFT);
170 			goto enter;
171 		}
172 		if ((flags & UVMPAGER_MAPIN_WAITOK) == 0) {
173 			UVMHIST_LOG(maphist,"<- NOWAIT failed", 0,0,0,0);
174 			return(0);
175 		}
176 		simple_lock(&pager_map_wanted_lock);
177 		pager_map_wanted = TRUE;
178 		UVMHIST_LOG(maphist, "  SLEEPING on pager_map",0,0,0,0);
179 		UVM_UNLOCK_AND_WAIT(pager_map, &pager_map_wanted_lock, FALSE,
180 		    "pager_map", 0);
181 		goto ReStart;
182 	}
183 
184 enter:
185 	/* got it */
186 	for (cva = kva ; size != 0 ; size -= PAGE_SIZE, cva += PAGE_SIZE) {
187 		pp = *pps++;
188 		KASSERT(pp);
189 		KASSERT(pp->flags & PG_BUSY);
190 		pmap_kenter_pa(cva, VM_PAGE_TO_PHYS(pp), prot);
191 	}
192 	pmap_update(vm_map_pmap(pager_map));
193 
194 	UVMHIST_LOG(maphist, "<- done (KVA=0x%x)", kva,0,0,0);
195 	return(kva);
196 }
197 
198 /*
199  * uvm_pagermapout: remove pager_map mapping
200  *
201  * we remove our mappings by hand and then remove the mapping (waking
202  * up anyone wanting space).
203  */
204 
205 void
206 uvm_pagermapout(kva, npages)
207 	vaddr_t kva;
208 	int npages;
209 {
210 	vsize_t size = npages << PAGE_SHIFT;
211 	struct vm_map_entry *entries;
212 	UVMHIST_FUNC("uvm_pagermapout"); UVMHIST_CALLED(maphist);
213 
214 	UVMHIST_LOG(maphist, " (kva=0x%x, npages=%d)", kva, npages,0,0);
215 
216 	/*
217 	 * duplicate uvm_unmap, but add in pager_map_wanted handling.
218 	 */
219 
220 	pmap_kremove(kva, npages << PAGE_SHIFT);
221 	if (kva == emergva) {
222 		simple_lock(&pager_map_wanted_lock);
223 		emerginuse = FALSE;
224 		wakeup(&emergva);
225 		simple_unlock(&pager_map_wanted_lock);
226 		return;
227 	}
228 
229 	vm_map_lock(pager_map);
230 	uvm_unmap_remove(pager_map, kva, kva + size, &entries);
231 	simple_lock(&pager_map_wanted_lock);
232 	if (pager_map_wanted) {
233 		pager_map_wanted = FALSE;
234 		wakeup(pager_map);
235 	}
236 	simple_unlock(&pager_map_wanted_lock);
237 	vm_map_unlock(pager_map);
238 	if (entries)
239 		uvm_unmap_detach(entries, 0);
240 	pmap_update(pmap_kernel());
241 	UVMHIST_LOG(maphist,"<- done",0,0,0,0);
242 }
243 
244 /*
245  * interrupt-context iodone handler for nested i/o bufs.
246  *
247  * => must be at splbio().
248  */
249 
250 void
251 uvm_aio_biodone1(bp)
252 	struct buf *bp;
253 {
254 	struct buf *mbp = bp->b_private;
255 
256 	KASSERT(mbp != bp);
257 	if (bp->b_flags & B_ERROR) {
258 		mbp->b_flags |= B_ERROR;
259 		mbp->b_error = bp->b_error;
260 	}
261 	mbp->b_resid -= bp->b_bcount;
262 	pool_put(&bufpool, bp);
263 	if (mbp->b_resid == 0) {
264 		biodone(mbp);
265 	}
266 }
267 
268 /*
269  * interrupt-context iodone handler for single-buf i/os
270  * or the top-level buf of a nested-buf i/o.
271  *
272  * => must be at splbio().
273  */
274 
275 void
276 uvm_aio_biodone(bp)
277 	struct buf *bp;
278 {
279 	/* reset b_iodone for when this is a single-buf i/o. */
280 	bp->b_iodone = uvm_aio_aiodone;
281 
282 	simple_lock(&uvm.aiodoned_lock);	/* locks uvm.aio_done */
283 	TAILQ_INSERT_TAIL(&uvm.aio_done, bp, b_freelist);
284 	wakeup(&uvm.aiodoned);
285 	simple_unlock(&uvm.aiodoned_lock);
286 }
287 
288 /*
289  * uvm_aio_aiodone: do iodone processing for async i/os.
290  * this should be called in thread context, not interrupt context.
291  */
292 
293 void
294 uvm_aio_aiodone(bp)
295 	struct buf *bp;
296 {
297 	int npages = bp->b_bufsize >> PAGE_SHIFT;
298 	struct vm_page *pg, *pgs[npages];
299 	struct uvm_object *uobj;
300 	struct simplelock *slock;
301 	int s, i, error, swslot;
302 	boolean_t write, swap, pageout;
303 	UVMHIST_FUNC("uvm_aio_aiodone"); UVMHIST_CALLED(ubchist);
304 	UVMHIST_LOG(ubchist, "bp %p", bp, 0,0,0);
305 
306 	error = (bp->b_flags & B_ERROR) ? (bp->b_error ? bp->b_error : EIO) : 0;
307 	write = (bp->b_flags & B_READ) == 0;
308 	/* XXXUBC B_NOCACHE is for swap pager, should be done differently */
309 	if (write && !(bp->b_flags & B_NOCACHE) && bioops.io_pageiodone) {
310 		(*bioops.io_pageiodone)(bp);
311 	}
312 
313 	uobj = NULL;
314 	for (i = 0; i < npages; i++) {
315 		pgs[i] = uvm_pageratop((vaddr_t)bp->b_data + (i << PAGE_SHIFT));
316 		UVMHIST_LOG(ubchist, "pgs[%d] = %p", i, pgs[i],0,0);
317 	}
318 	uvm_pagermapout((vaddr_t)bp->b_data, npages);
319 
320 	swslot = 0;
321 	slock = NULL;
322 	pg = pgs[0];
323 	swap = (pg->uanon != NULL && pg->uobject == NULL) ||
324 		(pg->pqflags & PQ_AOBJ) != 0;
325 	pageout = (pg->flags & PG_PAGEOUT) != 0;
326 	if (!swap) {
327 		uobj = pg->uobject;
328 		slock = &uobj->vmobjlock;
329 		simple_lock(slock);
330 		uvm_lock_pageq();
331 	} else if (error) {
332 		if (pg->uobject != NULL) {
333 			swslot = uao_find_swslot(pg->uobject, pg->offset);
334 		} else {
335 			swslot = pg->uanon->an_swslot;
336 		}
337 		KASSERT(swslot);
338 	}
339 	for (i = 0; i < npages; i++) {
340 		pg = pgs[i];
341 		KASSERT(swap || pg->uobject == uobj);
342 		KASSERT(pageout ^ ((pg->flags & PG_PAGEOUT) == 0));
343 		UVMHIST_LOG(ubchist, "pg %p", pg, 0,0,0);
344 
345 		/*
346 		 * for swap i/os, lock each page's object (or anon)
347 		 * individually since each page may need a different lock.
348 		 */
349 
350 		if (swap) {
351 			if (pg->uobject != NULL) {
352 				slock = &pg->uobject->vmobjlock;
353 			} else {
354 				slock = &pg->uanon->an_lock;
355 			}
356 			simple_lock(slock);
357 			uvm_lock_pageq();
358 		}
359 
360 		/*
361 		 * process errors.  for reads, just mark the page to be freed.
362 		 * for writes, if the error was ENOMEM, we assume this was
363 		 * a transient failure so we mark the page dirty so that
364 		 * we'll try to write it again later.  for all other write
365 		 * errors, we assume the error is permanent, thus the data
366 		 * in the page is lost.  bummer.
367 		 */
368 
369 		if (error) {
370 			if (!write) {
371 				pg->flags |= PG_RELEASED;
372 				continue;
373 			} else if (error == ENOMEM) {
374 				if (pg->flags & PG_PAGEOUT) {
375 					pg->flags &= ~PG_PAGEOUT;
376 					uvmexp.paging--;
377 				}
378 				pg->flags &= ~PG_CLEAN;
379 				uvm_pageactivate(pg);
380 			}
381 		}
382 
383 		/*
384 		 * if the page is PG_FAKE, this must have been a read to
385 		 * initialize the page.  clear PG_FAKE and activate the page.
386 		 * we must also clear the pmap "modified" flag since it may
387 		 * still be set from the page's previous identity.
388 		 */
389 
390 		if (pg->flags & PG_FAKE) {
391 			KASSERT(!write);
392 			pg->flags &= ~PG_FAKE;
393 			uvm_pageactivate(pg);
394 			pmap_clear_modify(pg);
395 		}
396 
397 		/*
398 		 * do accounting for pagedaemon i/o and arrange to free
399 		 * the pages instead of just unbusying them.
400 		 */
401 
402 		if (pg->flags & PG_PAGEOUT) {
403 			pg->flags &= ~PG_PAGEOUT;
404 			uvmexp.paging--;
405 			pg->flags |= PG_RELEASED;
406 		}
407 
408 		/*
409 		 * for swap pages, unlock everything for this page now.
410 		 */
411 
412 		if (swap) {
413 			uvm_page_unbusy(&pg, 1);
414 			uvm_unlock_pageq();
415 			simple_unlock(slock);
416 		}
417 	}
418 	if (!swap) {
419 		uvm_page_unbusy(pgs, npages);
420 		uvm_unlock_pageq();
421 		simple_unlock(slock);
422 	} else {
423 		KASSERT(write);
424 		KASSERT(pageout);
425 
426 		/* these pages are now only in swap. */
427 		simple_lock(&uvm.swap_data_lock);
428 		KASSERT(uvmexp.swpgonly + npages <= uvmexp.swpginuse);
429 		uvmexp.swpgonly += npages;
430 		simple_unlock(&uvm.swap_data_lock);
431 		if (error) {
432 			uvm_swap_markbad(swslot, npages);
433 		}
434 	}
435 	s = splbio();
436 	if (write && (bp->b_flags & B_AGE) != 0) {
437 		vwakeup(bp);
438 	}
439 	pool_put(&bufpool, bp);
440 	splx(s);
441 }
442