Commit | Line | Data |
---|---|---|
15637ed4 RG |
1 | /* |
2 | * Copyright (c) 1990 University of Utah. | |
3 | * Copyright (c) 1991 The Regents of the University of California. | |
4 | * All rights reserved. | |
5 | * | |
6 | * This code is derived from software contributed to Berkeley by | |
7 | * the Systems Programming Group of the University of Utah Computer | |
8 | * Science Department. | |
9 | * | |
10 | * Redistribution and use in source and binary forms, with or without | |
11 | * modification, are permitted provided that the following conditions | |
12 | * are met: | |
13 | * 1. Redistributions of source code must retain the above copyright | |
14 | * notice, this list of conditions and the following disclaimer. | |
15 | * 2. Redistributions in binary form must reproduce the above copyright | |
16 | * notice, this list of conditions and the following disclaimer in the | |
17 | * documentation and/or other materials provided with the distribution. | |
18 | * 3. All advertising materials mentioning features or use of this software | |
19 | * must display the following acknowledgement: | |
20 | * This product includes software developed by the University of | |
21 | * California, Berkeley and its contributors. | |
22 | * 4. Neither the name of the University nor the names of its contributors | |
23 | * may be used to endorse or promote products derived from this software | |
24 | * without specific prior written permission. | |
25 | * | |
26 | * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND | |
27 | * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE | |
28 | * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE | |
29 | * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE | |
30 | * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL | |
31 | * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS | |
32 | * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) | |
33 | * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT | |
34 | * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY | |
35 | * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF | |
36 | * SUCH DAMAGE. | |
37 | * | |
1284e777 RG |
38 | * from: Utah $Hdr: swap_pager.c 1.4 91/04/30$ |
39 | * from: @(#)swap_pager.c 7.4 (Berkeley) 5/7/91 | |
bbc3f849 | 40 | * $Id: swap_pager.c,v 1.2 1993/10/16 16:20:19 rgrimes Exp $ |
15637ed4 | 41 | */ |
15637ed4 RG |
42 | |
43 | /* | |
44 | * Quick hack to page to dedicated partition(s). | |
45 | * TODO: | |
46 | * Add multiprocessor locks | |
47 | * Deal with async writes in a better fashion | |
48 | */ | |
49 | ||
50 | #include "swappager.h" | |
51 | #if NSWAPPAGER > 0 | |
52 | ||
53 | #include "param.h" | |
54 | #include "proc.h" | |
55 | #include "buf.h" | |
56 | #include "systm.h" | |
57 | #include "specdev.h" | |
58 | #include "vnode.h" | |
59 | #include "malloc.h" | |
60 | #include "queue.h" | |
61 | #include "rlist.h" | |
62 | ||
63 | #include "vm_param.h" | |
64 | #include "queue.h" | |
65 | #include "lock.h" | |
66 | #include "vm_prot.h" | |
67 | #include "vm_object.h" | |
68 | #include "vm_page.h" | |
69 | #include "vm_pageout.h" | |
70 | #include "swap_pager.h" | |
71 | ||
bbc3f849 GW |
72 | #include "dmap.h" |
73 | struct dmap zdmap; /* not used */ | |
74 | int dmmin, dmmax, dmtext; /* dmtext not used */ | |
75 | ||
15637ed4 RG |
76 | #define NSWSIZES 16 /* size of swtab */ |
77 | #define NPENDINGIO 64 /* max # of pending cleans */ | |
78 | #define MAXDADDRS 64 /* max # of disk addrs for fixed allocations */ | |
79 | ||
80 | #ifdef DEBUG | |
81 | int swpagerdebug = 0 /*0x100*/; | |
82 | #define SDB_FOLLOW 0x001 | |
83 | #define SDB_INIT 0x002 | |
84 | #define SDB_ALLOC 0x004 | |
85 | #define SDB_IO 0x008 | |
86 | #define SDB_WRITE 0x010 | |
87 | #define SDB_FAIL 0x020 | |
88 | #define SDB_ALLOCBLK 0x040 | |
89 | #define SDB_FULL 0x080 | |
90 | #define SDB_ANOM 0x100 | |
91 | #define SDB_ANOMPANIC 0x200 | |
92 | #endif | |
93 | ||
bbc3f849 GW |
94 | struct pagerops swappagerops = { |
95 | swap_pager_init, | |
96 | swap_pager_alloc, | |
97 | swap_pager_dealloc, | |
98 | swap_pager_getpage, | |
99 | swap_pager_putpage, | |
100 | swap_pager_haspage | |
101 | }; | |
102 | ||
15637ed4 RG |
103 | struct swpagerclean { |
104 | queue_head_t spc_list; | |
105 | int spc_flags; | |
106 | struct buf *spc_bp; | |
107 | sw_pager_t spc_swp; | |
108 | vm_offset_t spc_kva; | |
109 | vm_page_t spc_m; | |
110 | } swcleanlist[NPENDINGIO]; | |
111 | typedef struct swpagerclean *swp_clean_t; | |
112 | ||
113 | /* spc_flags values */ | |
114 | #define SPC_FREE 0x00 | |
115 | #define SPC_BUSY 0x01 | |
116 | #define SPC_DONE 0x02 | |
117 | #define SPC_ERROR 0x04 | |
118 | #define SPC_DIRTY 0x08 | |
119 | ||
120 | struct swtab { | |
121 | vm_size_t st_osize; /* size of object (bytes) */ | |
122 | int st_bsize; /* vs. size of swap block (DEV_BSIZE units) */ | |
123 | #ifdef DEBUG | |
124 | u_long st_inuse; /* number in this range in use */ | |
125 | u_long st_usecnt; /* total used of this size */ | |
126 | #endif | |
127 | } swtab[NSWSIZES+1]; | |
128 | ||
129 | #ifdef DEBUG | |
130 | int swap_pager_pendingio; /* max pending async "clean" ops */ | |
131 | int swap_pager_poip; /* pageouts in progress */ | |
132 | int swap_pager_piip; /* pageins in progress */ | |
133 | #endif | |
134 | ||
135 | queue_head_t swap_pager_inuse; /* list of pending page cleans */ | |
136 | queue_head_t swap_pager_free; /* list of free pager clean structs */ | |
137 | queue_head_t swap_pager_list; /* list of "named" anon regions */ | |
138 | ||
139 | void | |
140 | swap_pager_init() | |
141 | { | |
142 | register swp_clean_t spc; | |
143 | register int i, bsize; | |
15637ed4 RG |
144 | int maxbsize; |
145 | ||
146 | #ifdef DEBUG | |
147 | if (swpagerdebug & (SDB_FOLLOW|SDB_INIT)) | |
148 | printf("swpg_init()\n"); | |
149 | #endif | |
150 | dfltpagerops = &swappagerops; | |
151 | queue_init(&swap_pager_list); | |
152 | ||
153 | /* | |
154 | * Initialize clean lists | |
155 | */ | |
156 | queue_init(&swap_pager_inuse); | |
157 | queue_init(&swap_pager_free); | |
158 | for (i = 0, spc = swcleanlist; i < NPENDINGIO; i++, spc++) { | |
159 | queue_enter(&swap_pager_free, spc, swp_clean_t, spc_list); | |
160 | spc->spc_flags = SPC_FREE; | |
161 | } | |
162 | ||
163 | /* | |
164 | * Calculate the swap allocation constants. | |
165 | */ | |
166 | if (dmmin == 0) { | |
167 | dmmin = DMMIN; | |
168 | if (dmmin < CLBYTES/DEV_BSIZE) | |
169 | dmmin = CLBYTES/DEV_BSIZE; | |
170 | } | |
171 | if (dmmax == 0) | |
172 | dmmax = DMMAX; | |
173 | ||
174 | /* | |
175 | * Fill in our table of object size vs. allocation size | |
176 | */ | |
177 | bsize = btodb(PAGE_SIZE); | |
178 | if (bsize < dmmin) | |
179 | bsize = dmmin; | |
180 | maxbsize = btodb(sizeof(sw_bm_t) * NBBY * PAGE_SIZE); | |
181 | if (maxbsize > dmmax) | |
182 | maxbsize = dmmax; | |
183 | for (i = 0; i < NSWSIZES; i++) { | |
184 | swtab[i].st_osize = (vm_size_t) (MAXDADDRS * dbtob(bsize)); | |
185 | swtab[i].st_bsize = bsize; | |
186 | #ifdef DEBUG | |
187 | if (swpagerdebug & SDB_INIT) | |
188 | printf("swpg_init: ix %d, size %x, bsize %x\n", | |
189 | i, swtab[i].st_osize, swtab[i].st_bsize); | |
190 | #endif | |
191 | if (bsize >= maxbsize) | |
192 | break; | |
193 | bsize *= 2; | |
194 | } | |
195 | swtab[i].st_osize = 0; | |
196 | swtab[i].st_bsize = bsize; | |
197 | } | |
198 | ||
199 | /* | |
200 | * Allocate a pager structure and associated resources. | |
201 | * Note that if we are called from the pageout daemon (handle == NULL) | |
202 | * we should not wait for memory as it could resulting in deadlock. | |
203 | */ | |
204 | vm_pager_t | |
205 | swap_pager_alloc(handle, size, prot) | |
206 | caddr_t handle; | |
207 | register vm_size_t size; | |
208 | vm_prot_t prot; | |
209 | { | |
210 | register vm_pager_t pager; | |
211 | register sw_pager_t swp; | |
212 | struct swtab *swt; | |
213 | int waitok; | |
214 | ||
215 | #ifdef DEBUG | |
216 | if (swpagerdebug & (SDB_FOLLOW|SDB_ALLOC)) | |
217 | printf("swpg_alloc(%x, %x, %x)\n", handle, size, prot); | |
218 | #endif | |
219 | /* | |
220 | * If this is a "named" anonymous region, look it up and | |
221 | * return the appropriate pager if it exists. | |
222 | */ | |
223 | if (handle) { | |
224 | pager = vm_pager_lookup(&swap_pager_list, handle); | |
225 | if (pager != NULL) { | |
226 | /* | |
227 | * Use vm_object_lookup to gain a reference | |
228 | * to the object and also to remove from the | |
229 | * object cache. | |
230 | */ | |
231 | if (vm_object_lookup(pager) == NULL) | |
232 | panic("swap_pager_alloc: bad object"); | |
233 | return(pager); | |
234 | } | |
235 | } | |
236 | /* | |
237 | * Pager doesn't exist, allocate swap management resources | |
238 | * and initialize. | |
239 | */ | |
240 | waitok = handle ? M_WAITOK : M_NOWAIT; | |
241 | pager = (vm_pager_t)malloc(sizeof *pager, M_VMPAGER, waitok); | |
242 | if (pager == NULL) | |
243 | return(NULL); | |
244 | swp = (sw_pager_t)malloc(sizeof *swp, M_VMPGDATA, waitok); | |
245 | if (swp == NULL) { | |
246 | #ifdef DEBUG | |
247 | if (swpagerdebug & SDB_FAIL) | |
248 | printf("swpg_alloc: swpager malloc failed\n"); | |
249 | #endif | |
250 | free((caddr_t)pager, M_VMPAGER); | |
251 | return(NULL); | |
252 | } | |
253 | size = round_page(size); | |
254 | for (swt = swtab; swt->st_osize; swt++) | |
255 | if (size <= swt->st_osize) | |
256 | break; | |
257 | #ifdef DEBUG | |
258 | swt->st_inuse++; | |
259 | swt->st_usecnt++; | |
260 | #endif | |
261 | swp->sw_osize = size; | |
262 | swp->sw_bsize = swt->st_bsize; | |
263 | swp->sw_nblocks = (btodb(size) + swp->sw_bsize - 1) / swp->sw_bsize; | |
264 | swp->sw_blocks = (sw_blk_t) | |
265 | malloc(swp->sw_nblocks*sizeof(*swp->sw_blocks), | |
266 | M_VMPGDATA, M_NOWAIT); | |
267 | if (swp->sw_blocks == NULL) { | |
268 | free((caddr_t)swp, M_VMPGDATA); | |
269 | free((caddr_t)pager, M_VMPAGER); | |
270 | #ifdef DEBUG | |
271 | if (swpagerdebug & SDB_FAIL) | |
272 | printf("swpg_alloc: sw_blocks malloc failed\n"); | |
273 | swt->st_inuse--; | |
274 | swt->st_usecnt--; | |
275 | #endif | |
276 | return(FALSE); | |
277 | } | |
278 | bzero((caddr_t)swp->sw_blocks, | |
279 | swp->sw_nblocks * sizeof(*swp->sw_blocks)); | |
280 | swp->sw_poip = 0; | |
281 | if (handle) { | |
282 | vm_object_t object; | |
283 | ||
284 | swp->sw_flags = SW_NAMED; | |
285 | queue_enter(&swap_pager_list, pager, vm_pager_t, pg_list); | |
286 | /* | |
287 | * Consistant with other pagers: return with object | |
288 | * referenced. Can't do this with handle == NULL | |
289 | * since it might be the pageout daemon calling. | |
290 | */ | |
291 | object = vm_object_allocate(size); | |
292 | vm_object_enter(object, pager); | |
293 | vm_object_setpager(object, pager, 0, FALSE); | |
294 | } else { | |
295 | swp->sw_flags = 0; | |
296 | queue_init(&pager->pg_list); | |
297 | } | |
298 | pager->pg_handle = handle; | |
299 | pager->pg_ops = &swappagerops; | |
300 | pager->pg_type = PG_SWAP; | |
301 | pager->pg_data = (caddr_t)swp; | |
302 | ||
303 | #ifdef DEBUG | |
304 | if (swpagerdebug & SDB_ALLOC) | |
305 | printf("swpg_alloc: pg_data %x, %x of %x at %x\n", | |
306 | swp, swp->sw_nblocks, swp->sw_bsize, swp->sw_blocks); | |
307 | #endif | |
308 | return(pager); | |
309 | } | |
310 | ||
311 | void | |
312 | swap_pager_dealloc(pager) | |
313 | vm_pager_t pager; | |
314 | { | |
315 | register int i; | |
316 | register sw_blk_t bp; | |
317 | register sw_pager_t swp; | |
318 | struct swtab *swt; | |
319 | int s; | |
320 | ||
321 | #ifdef DEBUG | |
322 | /* save panic time state */ | |
323 | if ((swpagerdebug & SDB_ANOMPANIC) && panicstr) | |
324 | return; | |
325 | if (swpagerdebug & (SDB_FOLLOW|SDB_ALLOC)) | |
326 | printf("swpg_dealloc(%x)\n", pager); | |
327 | #endif | |
328 | /* | |
329 | * Remove from list right away so lookups will fail if we | |
330 | * block for pageout completion. | |
331 | */ | |
332 | swp = (sw_pager_t) pager->pg_data; | |
333 | if (swp->sw_flags & SW_NAMED) { | |
334 | queue_remove(&swap_pager_list, pager, vm_pager_t, pg_list); | |
335 | swp->sw_flags &= ~SW_NAMED; | |
336 | } | |
337 | #ifdef DEBUG | |
338 | for (swt = swtab; swt->st_osize; swt++) | |
339 | if (swp->sw_osize <= swt->st_osize) | |
340 | break; | |
341 | swt->st_inuse--; | |
342 | #endif | |
343 | ||
344 | /* | |
345 | * Wait for all pageouts to finish and remove | |
346 | * all entries from cleaning list. | |
347 | */ | |
348 | s = splbio(); | |
349 | while (swp->sw_poip) { | |
350 | swp->sw_flags |= SW_WANTED; | |
351 | assert_wait((int)swp); | |
352 | thread_block(); | |
353 | } | |
354 | splx(s); | |
355 | (void) swap_pager_clean(NULL, B_WRITE); | |
356 | ||
357 | /* | |
358 | * Free left over swap blocks | |
359 | */ | |
360 | s = splbio(); | |
361 | for (i = 0, bp = swp->sw_blocks; i < swp->sw_nblocks; i++, bp++) | |
362 | if (bp->swb_block) { | |
363 | #ifdef DEBUG | |
364 | if (swpagerdebug & (SDB_ALLOCBLK|SDB_FULL)) | |
365 | printf("swpg_dealloc: blk %x\n", | |
366 | bp->swb_block); | |
367 | #endif | |
368 | rlist_free(&swapmap, (unsigned)bp->swb_block, | |
369 | (unsigned)bp->swb_block + swp->sw_bsize - 1); | |
370 | } | |
371 | splx(s); | |
372 | /* | |
373 | * Free swap management resources | |
374 | */ | |
375 | free((caddr_t)swp->sw_blocks, M_VMPGDATA); | |
376 | free((caddr_t)swp, M_VMPGDATA); | |
377 | free((caddr_t)pager, M_VMPAGER); | |
378 | } | |
379 | ||
380 | swap_pager_getpage(pager, m, sync) | |
381 | vm_pager_t pager; | |
382 | vm_page_t m; | |
383 | boolean_t sync; | |
384 | { | |
385 | #ifdef DEBUG | |
386 | if (swpagerdebug & SDB_FOLLOW) | |
387 | printf("swpg_getpage(%x, %x, %d)\n", pager, m, sync); | |
388 | #endif | |
389 | return(swap_pager_io((sw_pager_t)pager->pg_data, m, B_READ)); | |
390 | } | |
391 | ||
392 | swap_pager_putpage(pager, m, sync) | |
393 | vm_pager_t pager; | |
394 | vm_page_t m; | |
395 | boolean_t sync; | |
396 | { | |
397 | int flags; | |
398 | ||
399 | #ifdef DEBUG | |
400 | if (swpagerdebug & SDB_FOLLOW) | |
401 | printf("swpg_putpage(%x, %x, %d)\n", pager, m, sync); | |
402 | #endif | |
403 | if (pager == NULL) { | |
404 | (void) swap_pager_clean(NULL, B_WRITE); | |
405 | return; | |
406 | } | |
407 | flags = B_WRITE; | |
408 | if (!sync) | |
409 | flags |= B_ASYNC; | |
410 | return(swap_pager_io((sw_pager_t)pager->pg_data, m, flags)); | |
411 | } | |
412 | ||
413 | boolean_t | |
414 | swap_pager_haspage(pager, offset) | |
415 | vm_pager_t pager; | |
416 | vm_offset_t offset; | |
417 | { | |
418 | register sw_pager_t swp; | |
419 | register sw_blk_t swb; | |
420 | int ix; | |
421 | ||
422 | #ifdef DEBUG | |
423 | if (swpagerdebug & (SDB_FOLLOW|SDB_ALLOCBLK)) | |
424 | printf("swpg_haspage(%x, %x) ", pager, offset); | |
425 | #endif | |
426 | swp = (sw_pager_t) pager->pg_data; | |
427 | ix = offset / dbtob(swp->sw_bsize); | |
428 | if (swp->sw_blocks == NULL || ix >= swp->sw_nblocks) { | |
429 | #ifdef DEBUG | |
430 | if (swpagerdebug & (SDB_FAIL|SDB_FOLLOW|SDB_ALLOCBLK)) | |
431 | printf("swpg_haspage: %x bad offset %x, ix %x\n", | |
432 | swp->sw_blocks, offset, ix); | |
433 | #endif | |
434 | return(FALSE); | |
435 | } | |
436 | swb = &swp->sw_blocks[ix]; | |
437 | if (swb->swb_block) | |
438 | ix = atop(offset % dbtob(swp->sw_bsize)); | |
439 | #ifdef DEBUG | |
440 | if (swpagerdebug & SDB_ALLOCBLK) | |
441 | printf("%x blk %x+%x ", swp->sw_blocks, swb->swb_block, ix); | |
442 | if (swpagerdebug & (SDB_FOLLOW|SDB_ALLOCBLK)) | |
443 | printf("-> %c\n", | |
444 | "FT"[swb->swb_block && (swb->swb_mask & (1 << ix))]); | |
445 | #endif | |
446 | if (swb->swb_block && (swb->swb_mask & (1 << ix))) | |
447 | return(TRUE); | |
448 | return(FALSE); | |
449 | } | |
450 | ||
451 | /* | |
452 | * Scaled down version of swap(). | |
453 | * Assumes that PAGE_SIZE < MAXPHYS; i.e. only one operation needed. | |
454 | * BOGUS: lower level IO routines expect a KVA so we have to map our | |
455 | * provided physical page into the KVA to keep them happy. | |
456 | */ | |
457 | swap_pager_io(swp, m, flags) | |
458 | register sw_pager_t swp; | |
459 | vm_page_t m; | |
460 | int flags; | |
461 | { | |
462 | register struct buf *bp; | |
463 | register sw_blk_t swb; | |
464 | register int s; | |
465 | int ix; | |
466 | boolean_t rv; | |
467 | vm_offset_t kva, off; | |
468 | swp_clean_t spc; | |
469 | ||
470 | #ifdef DEBUG | |
471 | /* save panic time state */ | |
472 | if ((swpagerdebug & SDB_ANOMPANIC) && panicstr) | |
473 | return; | |
474 | if (swpagerdebug & (SDB_FOLLOW|SDB_IO)) | |
475 | printf("swpg_io(%x, %x, %x)\n", swp, m, flags); | |
476 | #endif | |
477 | ||
478 | /* | |
479 | * For reads (pageins) and synchronous writes, we clean up | |
480 | * all completed async pageouts. | |
481 | */ | |
482 | if ((flags & B_ASYNC) == 0) { | |
483 | s = splbio(); | |
484 | #ifdef DEBUG | |
485 | /* | |
486 | * Check to see if this page is currently being cleaned. | |
487 | * If it is, we just wait til the operation is done before | |
488 | * continuing. | |
489 | */ | |
490 | while (swap_pager_clean(m, flags&B_READ)) { | |
491 | if (swpagerdebug & SDB_ANOM) | |
492 | printf("swap_pager_io: page %x cleaning\n", m); | |
493 | ||
494 | swp->sw_flags |= SW_WANTED; | |
495 | assert_wait((int)swp); | |
496 | thread_block(); | |
497 | } | |
498 | #else | |
499 | (void) swap_pager_clean(m, flags&B_READ); | |
500 | #endif | |
501 | splx(s); | |
502 | } | |
503 | /* | |
504 | * For async writes (pageouts), we cleanup completed pageouts so | |
505 | * that all available resources are freed. Also tells us if this | |
506 | * page is already being cleaned. If it is, or no resources | |
507 | * are available, we try again later. | |
508 | */ | |
509 | else if (swap_pager_clean(m, B_WRITE) || | |
510 | queue_empty(&swap_pager_free)) { | |
511 | #ifdef DEBUG | |
512 | if ((swpagerdebug & SDB_ANOM) && | |
513 | !queue_empty(&swap_pager_free)) | |
514 | printf("swap_pager_io: page %x already cleaning\n", m); | |
515 | #endif | |
516 | return(VM_PAGER_FAIL); | |
517 | } | |
518 | ||
519 | /* | |
520 | * Determine swap block and allocate as necessary. | |
521 | */ | |
522 | off = m->offset + m->object->paging_offset; | |
523 | ix = off / dbtob(swp->sw_bsize); | |
524 | if (swp->sw_blocks == NULL || ix >= swp->sw_nblocks) { | |
525 | #ifdef DEBUG | |
526 | if (swpagerdebug & SDB_FAIL) | |
527 | printf("swpg_io: bad offset %x+%x(%d) in %x\n", | |
528 | m->offset, m->object->paging_offset, | |
529 | ix, swp->sw_blocks); | |
530 | #endif | |
531 | return(VM_PAGER_FAIL); | |
532 | } | |
533 | s = splbio(); | |
534 | swb = &swp->sw_blocks[ix]; | |
535 | off = off % dbtob(swp->sw_bsize); | |
536 | if (flags & B_READ) { | |
537 | if (swb->swb_block == 0 || | |
538 | (swb->swb_mask & (1 << atop(off))) == 0) { | |
539 | #ifdef DEBUG | |
540 | if (swpagerdebug & (SDB_ALLOCBLK|SDB_FAIL)) | |
541 | printf("swpg_io: %x bad read: blk %x+%x, mask %x, off %x+%x\n", | |
542 | swp->sw_blocks, | |
543 | swb->swb_block, atop(off), | |
544 | swb->swb_mask, | |
545 | m->offset, m->object->paging_offset); | |
546 | #endif | |
547 | /* XXX: should we zero page here?? */ | |
548 | splx(s); | |
549 | return(VM_PAGER_FAIL); | |
550 | } | |
551 | } else if (swb->swb_block == 0) { | |
552 | #ifdef old | |
553 | swb->swb_block = rmalloc(swapmap, swp->sw_bsize); | |
554 | if (swb->swb_block == 0) { | |
555 | #else | |
556 | if (!rlist_alloc(&swapmap, (unsigned)swp->sw_bsize, | |
557 | (unsigned *)&swb->swb_block)) { | |
558 | #endif | |
559 | #ifdef DEBUG | |
560 | if (swpagerdebug & SDB_FAIL) | |
561 | printf("swpg_io: rmalloc of %x failed\n", | |
562 | swp->sw_bsize); | |
563 | #endif | |
564 | splx(s); | |
565 | return(VM_PAGER_FAIL); | |
566 | } | |
567 | #ifdef DEBUG | |
568 | if (swpagerdebug & (SDB_FULL|SDB_ALLOCBLK)) | |
569 | printf("swpg_io: %x alloc blk %x at ix %x\n", | |
570 | swp->sw_blocks, swb->swb_block, ix); | |
571 | #endif | |
572 | } | |
573 | splx(s); | |
574 | ||
575 | /* | |
576 | * Allocate a kernel virtual address and initialize so that PTE | |
577 | * is available for lower level IO drivers. | |
578 | */ | |
579 | kva = vm_pager_map_page(m); | |
580 | ||
581 | /* | |
582 | * Get a swap buffer header and perform the IO | |
583 | */ | |
584 | s = splbio(); | |
585 | while (bswlist.av_forw == NULL) { | |
586 | #ifdef DEBUG | |
587 | if (swpagerdebug & SDB_ANOM) | |
588 | printf("swap_pager_io: wait on swbuf for %x (%d)\n", | |
589 | m, flags); | |
590 | #endif | |
591 | bswlist.b_flags |= B_WANTED; | |
592 | sleep((caddr_t)&bswlist, PSWP+1); | |
593 | } | |
594 | bp = bswlist.av_forw; | |
595 | bswlist.av_forw = bp->av_forw; | |
596 | splx(s); | |
597 | bp->b_flags = B_BUSY | (flags & B_READ); | |
598 | bp->b_proc = &proc0; /* XXX (but without B_PHYS set this is ok) */ | |
599 | bp->b_un.b_addr = (caddr_t)kva; | |
600 | bp->b_blkno = swb->swb_block + btodb(off); | |
601 | VHOLD(swapdev_vp); | |
602 | bp->b_vp = swapdev_vp; | |
603 | if (swapdev_vp->v_type == VBLK) | |
604 | bp->b_dev = swapdev_vp->v_rdev; | |
605 | bp->b_bcount = PAGE_SIZE; | |
606 | if ((bp->b_flags & B_READ) == 0) | |
607 | swapdev_vp->v_numoutput++; | |
608 | ||
609 | /* | |
610 | * If this is an async write we set up additional buffer fields | |
611 | * and place a "cleaning" entry on the inuse queue. | |
612 | */ | |
613 | if ((flags & (B_READ|B_ASYNC)) == B_ASYNC) { | |
614 | #ifdef DEBUG | |
615 | if (queue_empty(&swap_pager_free)) | |
616 | panic("swpg_io: lost spc"); | |
617 | #endif | |
618 | queue_remove_first(&swap_pager_free, | |
619 | spc, swp_clean_t, spc_list); | |
620 | #ifdef DEBUG | |
621 | if (spc->spc_flags != SPC_FREE) | |
622 | panic("swpg_io: bad free spc"); | |
623 | #endif | |
624 | spc->spc_flags = SPC_BUSY; | |
625 | spc->spc_bp = bp; | |
626 | spc->spc_swp = swp; | |
627 | spc->spc_kva = kva; | |
628 | spc->spc_m = m; | |
629 | bp->b_flags |= B_CALL; | |
630 | bp->b_iodone = swap_pager_iodone; | |
631 | s = splbio(); | |
632 | swp->sw_poip++; | |
633 | queue_enter(&swap_pager_inuse, spc, swp_clean_t, spc_list); | |
634 | ||
635 | #ifdef DEBUG | |
636 | swap_pager_poip++; | |
637 | if (swpagerdebug & SDB_WRITE) | |
638 | printf("swpg_io: write: bp=%x swp=%x spc=%x poip=%d\n", | |
639 | bp, swp, spc, swp->sw_poip); | |
640 | if ((swpagerdebug & SDB_ALLOCBLK) && | |
641 | (swb->swb_mask & (1 << atop(off))) == 0) | |
642 | printf("swpg_io: %x write blk %x+%x\n", | |
643 | swp->sw_blocks, swb->swb_block, atop(off)); | |
644 | #endif | |
645 | swb->swb_mask |= (1 << atop(off)); | |
646 | splx(s); | |
647 | } | |
648 | #ifdef DEBUG | |
649 | if (swpagerdebug & SDB_IO) | |
650 | printf("swpg_io: IO start: bp %x, db %x, va %x, pa %x\n", | |
651 | bp, swb->swb_block+btodb(off), kva, VM_PAGE_TO_PHYS(m)); | |
652 | #endif | |
653 | VOP_STRATEGY(bp); | |
654 | if ((flags & (B_READ|B_ASYNC)) == B_ASYNC) { | |
655 | #ifdef DEBUG | |
656 | if (swpagerdebug & SDB_IO) | |
657 | printf("swpg_io: IO started: bp %x\n", bp); | |
658 | #endif | |
659 | return(VM_PAGER_PEND); | |
660 | } | |
661 | s = splbio(); | |
662 | #ifdef DEBUG | |
663 | if (flags & B_READ) | |
664 | swap_pager_piip++; | |
665 | else | |
666 | swap_pager_poip++; | |
667 | #endif | |
668 | while ((bp->b_flags & B_DONE) == 0) { | |
669 | assert_wait((int)bp); | |
670 | thread_block(); | |
671 | } | |
672 | #ifdef DEBUG | |
673 | if (flags & B_READ) | |
674 | --swap_pager_piip; | |
675 | else | |
676 | --swap_pager_poip; | |
677 | #endif | |
678 | rv = (bp->b_flags & B_ERROR) ? VM_PAGER_FAIL : VM_PAGER_OK; | |
679 | bp->b_flags &= ~(B_BUSY|B_WANTED|B_PHYS|B_DIRTY); | |
680 | bp->av_forw = bswlist.av_forw; | |
681 | bswlist.av_forw = bp; | |
682 | if (bp->b_vp) | |
683 | brelvp(bp); | |
684 | if (bswlist.b_flags & B_WANTED) { | |
685 | bswlist.b_flags &= ~B_WANTED; | |
686 | thread_wakeup((int)&bswlist); | |
687 | } | |
688 | if ((flags & B_READ) == 0 && rv == VM_PAGER_OK) { | |
689 | m->clean = TRUE; | |
690 | pmap_clear_modify(VM_PAGE_TO_PHYS(m)); | |
691 | } | |
692 | splx(s); | |
693 | #ifdef DEBUG | |
694 | if (swpagerdebug & SDB_IO) | |
695 | printf("swpg_io: IO done: bp %x, rv %d\n", bp, rv); | |
696 | if ((swpagerdebug & SDB_FAIL) && rv == VM_PAGER_FAIL) | |
697 | printf("swpg_io: IO error\n"); | |
698 | #endif | |
699 | vm_pager_unmap_page(kva); | |
700 | return(rv); | |
701 | } | |
702 | ||
703 | boolean_t | |
704 | swap_pager_clean(m, rw) | |
705 | vm_page_t m; | |
706 | int rw; | |
707 | { | |
708 | register swp_clean_t spc, tspc; | |
709 | register int s; | |
710 | ||
711 | #ifdef DEBUG | |
712 | /* save panic time state */ | |
713 | if ((swpagerdebug & SDB_ANOMPANIC) && panicstr) | |
714 | return; | |
715 | if (swpagerdebug & SDB_FOLLOW) | |
716 | printf("swpg_clean(%x, %d)\n", m, rw); | |
717 | #endif | |
718 | tspc = NULL; | |
719 | for (;;) { | |
720 | /* | |
721 | * Look up and removal from inuse list must be done | |
722 | * at splbio() to avoid conflicts with swap_pager_iodone. | |
723 | */ | |
724 | s = splbio(); | |
725 | spc = (swp_clean_t) queue_first(&swap_pager_inuse); | |
726 | while (!queue_end(&swap_pager_inuse, (queue_entry_t)spc)) { | |
727 | if ((spc->spc_flags & SPC_DONE) && | |
728 | swap_pager_finish(spc)) { | |
729 | queue_remove(&swap_pager_inuse, spc, | |
730 | swp_clean_t, spc_list); | |
731 | break; | |
732 | } | |
733 | if (m && m == spc->spc_m) { | |
734 | #ifdef DEBUG | |
735 | if (swpagerdebug & SDB_ANOM) | |
736 | printf("swap_pager_clean: page %x on list, flags %x\n", | |
737 | m, spc->spc_flags); | |
738 | #endif | |
739 | tspc = spc; | |
740 | } | |
741 | spc = (swp_clean_t) queue_next(&spc->spc_list); | |
742 | } | |
743 | ||
744 | /* | |
745 | * No operations done, thats all we can do for now. | |
746 | */ | |
747 | if (queue_end(&swap_pager_inuse, (queue_entry_t)spc)) | |
748 | break; | |
749 | splx(s); | |
750 | ||
751 | /* | |
752 | * The desired page was found to be busy earlier in | |
753 | * the scan but has since completed. | |
754 | */ | |
755 | if (tspc && tspc == spc) { | |
756 | #ifdef DEBUG | |
757 | if (swpagerdebug & SDB_ANOM) | |
758 | printf("swap_pager_clean: page %x done while looking\n", | |
759 | m); | |
760 | #endif | |
761 | tspc = NULL; | |
762 | } | |
763 | spc->spc_flags = SPC_FREE; | |
764 | vm_pager_unmap_page(spc->spc_kva); | |
765 | queue_enter(&swap_pager_free, spc, swp_clean_t, spc_list); | |
766 | #ifdef DEBUG | |
767 | if (swpagerdebug & SDB_WRITE) | |
768 | printf("swpg_clean: free spc %x\n", spc); | |
769 | #endif | |
770 | } | |
771 | #ifdef DEBUG | |
772 | /* | |
773 | * If we found that the desired page is already being cleaned | |
774 | * mark it so that swap_pager_iodone() will not set the clean | |
775 | * flag before the pageout daemon has another chance to clean it. | |
776 | */ | |
777 | if (tspc && rw == B_WRITE) { | |
778 | if (swpagerdebug & SDB_ANOM) | |
779 | printf("swap_pager_clean: page %x on clean list\n", | |
780 | tspc); | |
781 | tspc->spc_flags |= SPC_DIRTY; | |
782 | } | |
783 | #endif | |
784 | splx(s); | |
785 | ||
786 | #ifdef DEBUG | |
787 | if (swpagerdebug & SDB_WRITE) | |
788 | printf("swpg_clean: return %d\n", tspc ? TRUE : FALSE); | |
789 | if ((swpagerdebug & SDB_ANOM) && tspc) | |
790 | printf("swpg_clean: %s of cleaning page %x\n", | |
791 | rw == B_READ ? "get" : "put", m); | |
792 | #endif | |
793 | return(tspc ? TRUE : FALSE); | |
794 | } | |
795 | ||
796 | swap_pager_finish(spc) | |
797 | register swp_clean_t spc; | |
798 | { | |
799 | vm_object_t object = spc->spc_m->object; | |
800 | ||
801 | /* | |
802 | * Mark the paging operation as done. | |
803 | * (XXX) If we cannot get the lock, leave it til later. | |
804 | * (XXX) Also we are assuming that an async write is a | |
805 | * pageout operation that has incremented the counter. | |
806 | */ | |
807 | if (!vm_object_lock_try(object)) | |
808 | return(0); | |
809 | ||
810 | if (--object->paging_in_progress == 0) | |
811 | thread_wakeup((int) object); | |
812 | ||
813 | #ifdef DEBUG | |
814 | /* | |
815 | * XXX: this isn't even close to the right thing to do, | |
816 | * introduces a variety of race conditions. | |
817 | * | |
818 | * If dirty, vm_pageout() has attempted to clean the page | |
819 | * again. In this case we do not do anything as we will | |
820 | * see the page again shortly. | |
821 | */ | |
822 | if (spc->spc_flags & SPC_DIRTY) { | |
823 | if (swpagerdebug & SDB_ANOM) | |
824 | printf("swap_pager_finish: page %x dirty again\n", | |
825 | spc->spc_m); | |
826 | spc->spc_m->busy = FALSE; | |
827 | PAGE_WAKEUP(spc->spc_m); | |
828 | vm_object_unlock(object); | |
829 | return(1); | |
830 | } | |
831 | #endif | |
832 | /* | |
833 | * If no error mark as clean and inform the pmap system. | |
834 | * If error, mark as dirty so we will try again. | |
835 | * (XXX could get stuck doing this, should give up after awhile) | |
836 | */ | |
837 | if (spc->spc_flags & SPC_ERROR) { | |
838 | printf("swap_pager_finish: clean of page %x failed\n", | |
839 | VM_PAGE_TO_PHYS(spc->spc_m)); | |
840 | spc->spc_m->laundry = TRUE; | |
841 | } else { | |
842 | spc->spc_m->clean = TRUE; | |
843 | pmap_clear_modify(VM_PAGE_TO_PHYS(spc->spc_m)); | |
844 | } | |
845 | spc->spc_m->busy = FALSE; | |
846 | PAGE_WAKEUP(spc->spc_m); | |
847 | ||
848 | vm_object_unlock(object); | |
849 | return(1); | |
850 | } | |
851 | ||
852 | swap_pager_iodone(bp) | |
853 | register struct buf *bp; | |
854 | { | |
855 | register swp_clean_t spc; | |
856 | daddr_t blk; | |
857 | int s; | |
858 | ||
859 | #ifdef DEBUG | |
860 | /* save panic time state */ | |
861 | if ((swpagerdebug & SDB_ANOMPANIC) && panicstr) | |
862 | return; | |
863 | if (swpagerdebug & SDB_FOLLOW) | |
864 | printf("swpg_iodone(%x)\n", bp); | |
865 | #endif | |
866 | s = splbio(); | |
867 | spc = (swp_clean_t) queue_first(&swap_pager_inuse); | |
868 | while (!queue_end(&swap_pager_inuse, (queue_entry_t)spc)) { | |
869 | if (spc->spc_bp == bp) | |
870 | break; | |
871 | spc = (swp_clean_t) queue_next(&spc->spc_list); | |
872 | } | |
873 | #ifdef DEBUG | |
874 | if (queue_end(&swap_pager_inuse, (queue_entry_t)spc)) | |
875 | panic("swap_pager_iodone: bp not found"); | |
876 | #endif | |
877 | ||
878 | spc->spc_flags &= ~SPC_BUSY; | |
879 | spc->spc_flags |= SPC_DONE; | |
880 | if (bp->b_flags & B_ERROR) { | |
881 | spc->spc_flags |= SPC_ERROR; | |
882 | printf("error %d blkno %d sz %d ", bp->b_error, bp->b_blkno, bp->b_bcount); | |
883 | } | |
884 | spc->spc_bp = NULL; | |
885 | blk = bp->b_blkno; | |
886 | ||
887 | #ifdef DEBUG | |
888 | --swap_pager_poip; | |
889 | if (swpagerdebug & SDB_WRITE) | |
890 | printf("swpg_iodone: bp=%x swp=%x flags=%x spc=%x poip=%x\n", | |
891 | bp, spc->spc_swp, spc->spc_swp->sw_flags, | |
892 | spc, spc->spc_swp->sw_poip); | |
893 | #endif | |
894 | ||
895 | spc->spc_swp->sw_poip--; | |
896 | if (spc->spc_swp->sw_flags & SW_WANTED) { | |
897 | spc->spc_swp->sw_flags &= ~SW_WANTED; | |
898 | thread_wakeup((int)spc->spc_swp); | |
899 | } | |
900 | ||
901 | bp->b_flags &= ~(B_BUSY|B_WANTED|B_PHYS|B_DIRTY); | |
902 | bp->av_forw = bswlist.av_forw; | |
903 | bswlist.av_forw = bp; | |
904 | if (bp->b_vp) | |
905 | brelvp(bp); | |
906 | if (bswlist.b_flags & B_WANTED) { | |
907 | bswlist.b_flags &= ~B_WANTED; | |
908 | thread_wakeup((int)&bswlist); | |
909 | } | |
910 | thread_wakeup((int) &vm_pages_needed); | |
911 | splx(s); | |
912 | } | |
913 | #endif |