xref: /netbsd-src/sys/uvm/uvm_pager.c (revision 27578b9aac214cc7796ead81dcc5427e79d5f2a0)
1 /*	$NetBSD: uvm_pager.c,v 1.50 2001/09/15 20:36:47 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 #include "opt_uvmhist.h"
38 
39 /*
40  * uvm_pager.c: generic functions used to assist the pagers.
41  */
42 
43 #include <sys/param.h>
44 #include <sys/systm.h>
45 #include <sys/proc.h>
46 #include <sys/malloc.h>
47 #include <sys/pool.h>
48 #include <sys/vnode.h>
49 
50 #define UVM_PAGER
51 #include <uvm/uvm.h>
52 
53 struct pool *uvm_aiobuf_pool;
54 
55 /*
56  * list of uvm pagers in the system
57  */
58 
59 extern struct uvm_pagerops uvm_deviceops;
60 extern struct uvm_pagerops uvm_vnodeops;
61 extern struct uvm_pagerops ubc_pager;
62 
63 struct uvm_pagerops *uvmpagerops[] = {
64 	&aobj_pager,
65 	&uvm_deviceops,
66 	&uvm_vnodeops,
67 	&ubc_pager,
68 };
69 
70 /*
71  * the pager map: provides KVA for I/O
72  */
73 
74 struct vm_map *pager_map;		/* XXX */
75 struct simplelock pager_map_wanted_lock;
76 boolean_t pager_map_wanted;	/* locked by pager map */
77 static vaddr_t emergva;
78 static boolean_t emerginuse;
79 
80 /*
81  * uvm_pager_init: init pagers (at boot time)
82  */
83 
84 void
85 uvm_pager_init()
86 {
87 	int lcv;
88 	vaddr_t sva, eva;
89 
90 	/*
91 	 * init pager map
92 	 */
93 
94 	sva = 0;
95 	pager_map = uvm_km_suballoc(kernel_map, &sva, &eva, PAGER_MAP_SIZE, 0,
96 	    FALSE, NULL);
97 	simple_lock_init(&pager_map_wanted_lock);
98 	pager_map_wanted = FALSE;
99 	emergva = uvm_km_valloc(kernel_map, MAXBSIZE);
100 	emerginuse = FALSE;
101 
102 	/*
103 	 * init ASYNC I/O queue
104 	 */
105 
106 	TAILQ_INIT(&uvm.aio_done);
107 
108 	/*
109 	 * call pager init functions
110 	 */
111 	for (lcv = 0 ; lcv < sizeof(uvmpagerops)/sizeof(struct uvm_pagerops *);
112 	    lcv++) {
113 		if (uvmpagerops[lcv]->pgo_init)
114 			uvmpagerops[lcv]->pgo_init();
115 	}
116 }
117 
118 /*
119  * uvm_pagermapin: map pages into KVA (pager_map) for I/O that needs mappings
120  *
121  * we basically just map in a blank map entry to reserve the space in the
122  * map and then use pmap_enter() to put the mappings in by hand.
123  */
124 
125 vaddr_t
126 uvm_pagermapin(pps, npages, flags)
127 	struct vm_page **pps;
128 	int npages;
129 	int flags;
130 {
131 	vsize_t size;
132 	vaddr_t kva;
133 	vaddr_t cva;
134 	struct vm_page *pp;
135 	vm_prot_t prot;
136 	UVMHIST_FUNC("uvm_pagermapin"); UVMHIST_CALLED(maphist);
137 
138 	UVMHIST_LOG(maphist,"(pps=0x%x, npages=%d)", pps, npages,0,0);
139 
140 	/*
141 	 * compute protection.  outgoing I/O only needs read
142 	 * access to the page, whereas incoming needs read/write.
143 	 */
144 
145 	prot = VM_PROT_READ;
146 	if (flags & UVMPAGER_MAPIN_READ)
147 		prot |= VM_PROT_WRITE;
148 
149 ReStart:
150 	size = npages << PAGE_SHIFT;
151 	kva = 0;			/* let system choose VA */
152 
153 	if (uvm_map(pager_map, &kva, size, NULL,
154 	      UVM_UNKNOWN_OFFSET, 0, UVM_FLAG_NOMERGE) != 0) {
155 		if (curproc == uvm.pagedaemon_proc) {
156 			simple_lock(&pager_map_wanted_lock);
157 			if (emerginuse) {
158 				UVM_UNLOCK_AND_WAIT(&emergva,
159 				    &pager_map_wanted_lock, FALSE,
160 				    "emergva", 0);
161 				goto ReStart;
162 			}
163 			emerginuse = TRUE;
164 			simple_unlock(&pager_map_wanted_lock);
165 			kva = emergva;
166 			KASSERT(npages <= MAXBSIZE >> PAGE_SHIFT);
167 			goto enter;
168 		}
169 		if ((flags & UVMPAGER_MAPIN_WAITOK) == 0) {
170 			UVMHIST_LOG(maphist,"<- NOWAIT failed", 0,0,0,0);
171 			return(0);
172 		}
173 		simple_lock(&pager_map_wanted_lock);
174 		pager_map_wanted = TRUE;
175 		UVMHIST_LOG(maphist, "  SLEEPING on pager_map",0,0,0,0);
176 		UVM_UNLOCK_AND_WAIT(pager_map, &pager_map_wanted_lock, FALSE,
177 		    "pager_map", 0);
178 		goto ReStart;
179 	}
180 
181 enter:
182 	/* got it */
183 	for (cva = kva ; size != 0 ; size -= PAGE_SIZE, cva += PAGE_SIZE) {
184 		pp = *pps++;
185 		KASSERT(pp);
186 		KASSERT(pp->flags & PG_BUSY);
187 		pmap_kenter_pa(cva, VM_PAGE_TO_PHYS(pp), prot);
188 	}
189 	pmap_update(vm_map_pmap(pager_map));
190 
191 	UVMHIST_LOG(maphist, "<- done (KVA=0x%x)", kva,0,0,0);
192 	return(kva);
193 }
194 
195 /*
196  * uvm_pagermapout: remove pager_map mapping
197  *
198  * we remove our mappings by hand and then remove the mapping (waking
199  * up anyone wanting space).
200  */
201 
202 void
203 uvm_pagermapout(kva, npages)
204 	vaddr_t kva;
205 	int npages;
206 {
207 	vsize_t size = npages << PAGE_SHIFT;
208 	struct vm_map_entry *entries;
209 	UVMHIST_FUNC("uvm_pagermapout"); UVMHIST_CALLED(maphist);
210 
211 	UVMHIST_LOG(maphist, " (kva=0x%x, npages=%d)", kva, npages,0,0);
212 
213 	/*
214 	 * duplicate uvm_unmap, but add in pager_map_wanted handling.
215 	 */
216 
217 	pmap_kremove(kva, npages << PAGE_SHIFT);
218 	if (kva == emergva) {
219 		simple_lock(&pager_map_wanted_lock);
220 		emerginuse = FALSE;
221 		wakeup(&emergva);
222 		simple_unlock(&pager_map_wanted_lock);
223 		return;
224 	}
225 
226 	vm_map_lock(pager_map);
227 	uvm_unmap_remove(pager_map, kva, kva + size, &entries);
228 	simple_lock(&pager_map_wanted_lock);
229 	if (pager_map_wanted) {
230 		pager_map_wanted = FALSE;
231 		wakeup(pager_map);
232 	}
233 	simple_unlock(&pager_map_wanted_lock);
234 	vm_map_unlock(pager_map);
235 	if (entries)
236 		uvm_unmap_detach(entries, 0);
237 	pmap_update(pmap_kernel());
238 	UVMHIST_LOG(maphist,"<- done",0,0,0,0);
239 }
240 
241 /*
242  * interrupt-context iodone handler for nested i/o bufs.
243  *
244  * => must be at splbio().
245  */
246 
247 void
248 uvm_aio_biodone1(bp)
249 	struct buf *bp;
250 {
251 	struct buf *mbp = bp->b_private;
252 
253 	KASSERT(mbp != bp);
254 	if (bp->b_flags & B_ERROR) {
255 		mbp->b_flags |= B_ERROR;
256 		mbp->b_error = bp->b_error;
257 	}
258 	mbp->b_resid -= bp->b_bcount;
259 	pool_put(&bufpool, bp);
260 	if (mbp->b_resid == 0) {
261 		biodone(mbp);
262 	}
263 }
264 
265 /*
266  * interrupt-context iodone handler for single-buf i/os
267  * or the top-level buf of a nested-buf i/o.
268  *
269  * => must be at splbio().
270  */
271 
272 void
273 uvm_aio_biodone(bp)
274 	struct buf *bp;
275 {
276 	/* reset b_iodone for when this is a single-buf i/o. */
277 	bp->b_iodone = uvm_aio_aiodone;
278 
279 	simple_lock(&uvm.aiodoned_lock);	/* locks uvm.aio_done */
280 	TAILQ_INSERT_TAIL(&uvm.aio_done, bp, b_freelist);
281 	wakeup(&uvm.aiodoned);
282 	simple_unlock(&uvm.aiodoned_lock);
283 }
284 
285 /*
286  * uvm_aio_aiodone: do iodone processing for async i/os.
287  * this should be called in thread context, not interrupt context.
288  */
289 
290 void
291 uvm_aio_aiodone(bp)
292 	struct buf *bp;
293 {
294 	int npages = bp->b_bufsize >> PAGE_SHIFT;
295 	struct vm_page *pg, *pgs[npages];
296 	struct uvm_object *uobj;
297 	struct simplelock *slock;
298 	int s, i, error, swslot;
299 	boolean_t write, swap, pageout, async;
300 	UVMHIST_FUNC("uvm_aio_aiodone"); UVMHIST_CALLED(ubchist);
301 	UVMHIST_LOG(ubchist, "bp %p", bp, 0,0,0);
302 
303 	error = (bp->b_flags & B_ERROR) ? (bp->b_error ? bp->b_error : EIO) : 0;
304 	write = (bp->b_flags & B_READ) == 0;
305 	/* XXXUBC B_NOCACHE is for swap pager, should be done differently */
306 	if (write && !(bp->b_flags & B_NOCACHE) && bioops.io_pageiodone) {
307 		(*bioops.io_pageiodone)(bp);
308 	}
309 
310 	uobj = NULL;
311 	for (i = 0; i < npages; i++) {
312 		pgs[i] = uvm_pageratop((vaddr_t)bp->b_data + (i << PAGE_SHIFT));
313 		UVMHIST_LOG(ubchist, "pgs[%d] = %p", i, pgs[i],0,0);
314 	}
315 	uvm_pagermapout((vaddr_t)bp->b_data, npages);
316 
317 	swap = (pgs[0]->pqflags & PQ_SWAPBACKED) != 0;
318 	swslot = 0;
319 	slock = NULL;
320 	pageout = (pgs[0]->flags & PG_PAGEOUT) != 0;
321 	async = (bp->b_flags & B_ASYNC) != 0;
322 	if (!swap) {
323 		uobj = pgs[0]->uobject;
324 		slock = &uobj->vmobjlock;
325 		simple_lock(slock);
326 		uvm_lock_pageq();
327 	} else if (error) {
328 		if (pgs[0]->pqflags & PQ_ANON) {
329 			swslot = pg->uanon->an_swslot;
330 		} else {
331 			swslot = uao_find_swslot(pg->uobject, pg->offset);
332 		}
333 		KASSERT(swslot);
334 	}
335 	for (i = 0; i < npages; i++) {
336 		pg = pgs[i];
337 		KASSERT(swap || pg->uobject == uobj);
338 		KASSERT(pageout ^ ((pg->flags & PG_PAGEOUT) == 0));
339 		UVMHIST_LOG(ubchist, "pg %p", pg, 0,0,0);
340 
341 		/*
342 		 * for swap i/os, lock each page's object (or anon)
343 		 * individually since each page may need a different lock.
344 		 */
345 
346 		if (swap) {
347 			if (pg->pqflags & PQ_ANON) {
348 				slock = &pg->uanon->an_lock;
349 			} else {
350 				slock = &pg->uobject->vmobjlock;
351 			}
352 			simple_lock(slock);
353 			uvm_lock_pageq();
354 		}
355 
356 		/*
357 		 * process errors.  for reads, just mark the page to be freed.
358 		 * for writes, if the error was ENOMEM, we assume this was
359 		 * a transient failure so we mark the page dirty so that
360 		 * we'll try to write it again later.  for all other write
361 		 * errors, we assume the error is permanent, thus the data
362 		 * in the page is lost.  bummer.
363 		 */
364 
365 		if (error) {
366 			if (!write) {
367 				KASSERT(!swap);
368 				pg->flags |= PG_RELEASED;
369 				continue;
370 			} else if (error == ENOMEM) {
371 				if (pg->flags & PG_PAGEOUT) {
372 					pg->flags &= ~PG_PAGEOUT;
373 					uvmexp.paging--;
374 				}
375 				pg->flags &= ~PG_CLEAN;
376 				uvm_pageactivate(pg);
377 			}
378 		}
379 
380 		/*
381 		 * if the page is PG_FAKE, this must have been a read to
382 		 * initialize the page.  clear PG_FAKE and activate the page.
383 		 */
384 
385 		if (pg->flags & PG_FAKE) {
386 			KASSERT(!write);
387 			pg->flags &= ~PG_FAKE;
388 			uvm_pageactivate(pg);
389 			pmap_clear_modify(pg);
390 		}
391 
392 		/*
393 		 * for async reads, this may be the first time the page
394 		 * is unlocked after being created, so we need to be sure
395 		 * the page is on a paging queue.
396 		 */
397 
398 		if (!write) {
399 			uvm_pageactivate(pg);
400 		} else if (pg->flags & PG_PAGEOUT) {
401 			pg->flags &= ~PG_PAGEOUT;
402 			uvmexp.paging--;
403 			pg->flags |= PG_RELEASED;
404 		}
405 
406 		/*
407 		 * for swap pages, unlock everything for this page now.
408 		 */
409 
410 		if (swap) {
411 			uvm_page_unbusy(&pg, 1);
412 			uvm_unlock_pageq();
413 			simple_unlock(slock);
414 		}
415 	}
416 	if (!swap) {
417 		uvm_page_unbusy(pgs, npages);
418 		uvm_unlock_pageq();
419 		simple_unlock(slock);
420 	} else {
421 		KASSERT(write && pageout);
422 		if (error) {
423 			uvm_swap_markbad(swslot, npages);
424 		}
425 	}
426 	s = splbio();
427 	if (write && (bp->b_flags & B_AGE) != 0) {
428 		vwakeup(bp);
429 	}
430 	pool_put(&bufpool, bp);
431 	splx(s);
432 }
433