Commit | Line | Data |
---|---|---|
619edcce KM |
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 | * %sccs.include.redist.c% | |
11 | * | |
07a039d6 MH |
12 | * from: Utah $Hdr: swap_pager.c 1.4 91/04/30$ |
13 | * | |
2601dc74 | 14 | * @(#)swap_pager.c 7.11 (Berkeley) %G% |
619edcce KM |
15 | */ |
16 | ||
17 | /* | |
18 | * Quick hack to page to dedicated partition(s). | |
19 | * TODO: | |
20 | * Add multiprocessor locks | |
21 | * Deal with async writes in a better fashion | |
22 | */ | |
23 | ||
24 | #include "swappager.h" | |
25 | #if NSWAPPAGER > 0 | |
26 | ||
719508e0 KB |
27 | #include <sys/param.h> |
28 | #include <sys/systm.h> | |
29 | #include <sys/proc.h> | |
30 | #include <sys/buf.h> | |
31 | #include <sys/map.h> | |
719508e0 | 32 | #include <sys/vnode.h> |
7e6cbfc6 | 33 | #include <sys/specdev.h> |
719508e0 KB |
34 | #include <sys/malloc.h> |
35 | ||
36 | #include <vm/vm.h> | |
37 | #include <vm/queue.h> | |
38 | #include <vm/vm_page.h> | |
39 | #include <vm/vm_pageout.h> | |
40 | #include <vm/swap_pager.h> | |
619edcce KM |
41 | |
42 | #define NSWSIZES 16 /* size of swtab */ | |
43 | #define NPENDINGIO 64 /* max # of pending cleans */ | |
44 | #define MAXDADDRS 64 /* max # of disk addrs for fixed allocations */ | |
45 | ||
46 | #ifdef DEBUG | |
47 | int swpagerdebug = 0x100; | |
48 | #define SDB_FOLLOW 0x001 | |
49 | #define SDB_INIT 0x002 | |
50 | #define SDB_ALLOC 0x004 | |
51 | #define SDB_IO 0x008 | |
52 | #define SDB_WRITE 0x010 | |
53 | #define SDB_FAIL 0x020 | |
54 | #define SDB_ALLOCBLK 0x040 | |
55 | #define SDB_FULL 0x080 | |
56 | #define SDB_ANOM 0x100 | |
57 | #define SDB_ANOMPANIC 0x200 | |
58 | #endif | |
59 | ||
60 | struct swpagerclean { | |
61 | queue_head_t spc_list; | |
62 | int spc_flags; | |
63 | struct buf *spc_bp; | |
64 | sw_pager_t spc_swp; | |
65 | vm_offset_t spc_kva; | |
66 | vm_page_t spc_m; | |
67 | } swcleanlist[NPENDINGIO]; | |
719508e0 KB |
68 | typedef struct swpagerclean *swp_clean_t; |
69 | ||
619edcce | 70 | |
619edcce KM |
71 | /* spc_flags values */ |
72 | #define SPC_FREE 0x00 | |
73 | #define SPC_BUSY 0x01 | |
74 | #define SPC_DONE 0x02 | |
75 | #define SPC_ERROR 0x04 | |
76 | #define SPC_DIRTY 0x08 | |
77 | ||
78 | struct swtab { | |
79 | vm_size_t st_osize; /* size of object (bytes) */ | |
80 | int st_bsize; /* vs. size of swap block (DEV_BSIZE units) */ | |
81 | #ifdef DEBUG | |
82 | u_long st_inuse; /* number in this range in use */ | |
83 | u_long st_usecnt; /* total used of this size */ | |
84 | #endif | |
85 | } swtab[NSWSIZES+1]; | |
86 | ||
87 | #ifdef DEBUG | |
88 | int swap_pager_pendingio; /* max pending async "clean" ops */ | |
89 | int swap_pager_poip; /* pageouts in progress */ | |
90 | int swap_pager_piip; /* pageins in progress */ | |
91 | #endif | |
92 | ||
93 | queue_head_t swap_pager_inuse; /* list of pending page cleans */ | |
94 | queue_head_t swap_pager_free; /* list of free pager clean structs */ | |
95 | queue_head_t swap_pager_list; /* list of "named" anon regions */ | |
96 | ||
719508e0 KB |
97 | static int swap_pager_finish __P((swp_clean_t)); |
98 | static void swap_pager_init __P((void)); | |
99 | static vm_pager_t swap_pager_alloc __P((caddr_t, vm_size_t, vm_prot_t)); | |
100 | static boolean_t swap_pager_clean __P((vm_page_t, int)); | |
101 | static void swap_pager_dealloc __P((vm_pager_t)); | |
102 | static int swap_pager_getpage | |
103 | __P((vm_pager_t, vm_page_t, boolean_t)); | |
104 | static boolean_t swap_pager_haspage __P((vm_pager_t, vm_offset_t)); | |
105 | static int swap_pager_io __P((sw_pager_t, vm_page_t, int)); | |
106 | static void swap_pager_iodone __P((struct buf *)); | |
107 | static int swap_pager_putpage | |
108 | __P((vm_pager_t, vm_page_t, boolean_t)); | |
109 | ||
110 | struct pagerops swappagerops = { | |
111 | swap_pager_init, | |
112 | swap_pager_alloc, | |
113 | swap_pager_dealloc, | |
114 | swap_pager_getpage, | |
115 | swap_pager_putpage, | |
116 | swap_pager_haspage | |
117 | }; | |
118 | ||
119 | static void | |
619edcce KM |
120 | swap_pager_init() |
121 | { | |
122 | register swp_clean_t spc; | |
123 | register int i, bsize; | |
124 | extern int dmmin, dmmax; | |
125 | int maxbsize; | |
126 | ||
127 | #ifdef DEBUG | |
128 | if (swpagerdebug & (SDB_FOLLOW|SDB_INIT)) | |
129 | printf("swpg_init()\n"); | |
130 | #endif | |
131 | dfltpagerops = &swappagerops; | |
132 | queue_init(&swap_pager_list); | |
133 | ||
134 | /* | |
135 | * Initialize clean lists | |
136 | */ | |
137 | queue_init(&swap_pager_inuse); | |
138 | queue_init(&swap_pager_free); | |
139 | for (i = 0, spc = swcleanlist; i < NPENDINGIO; i++, spc++) { | |
140 | queue_enter(&swap_pager_free, spc, swp_clean_t, spc_list); | |
141 | spc->spc_flags = SPC_FREE; | |
142 | } | |
143 | ||
144 | /* | |
145 | * Calculate the swap allocation constants. | |
146 | */ | |
147 | if (dmmin == 0) { | |
148 | dmmin = DMMIN; | |
149 | if (dmmin < CLBYTES/DEV_BSIZE) | |
150 | dmmin = CLBYTES/DEV_BSIZE; | |
151 | } | |
152 | if (dmmax == 0) | |
153 | dmmax = DMMAX; | |
154 | ||
155 | /* | |
156 | * Fill in our table of object size vs. allocation size | |
157 | */ | |
158 | bsize = btodb(PAGE_SIZE); | |
159 | if (bsize < dmmin) | |
160 | bsize = dmmin; | |
161 | maxbsize = btodb(sizeof(sw_bm_t) * NBBY * PAGE_SIZE); | |
162 | if (maxbsize > dmmax) | |
163 | maxbsize = dmmax; | |
164 | for (i = 0; i < NSWSIZES; i++) { | |
165 | swtab[i].st_osize = (vm_size_t) (MAXDADDRS * dbtob(bsize)); | |
166 | swtab[i].st_bsize = bsize; | |
167 | #ifdef DEBUG | |
168 | if (swpagerdebug & SDB_INIT) | |
169 | printf("swpg_init: ix %d, size %x, bsize %x\n", | |
170 | i, swtab[i].st_osize, swtab[i].st_bsize); | |
171 | #endif | |
172 | if (bsize >= maxbsize) | |
173 | break; | |
174 | bsize *= 2; | |
175 | } | |
176 | swtab[i].st_osize = 0; | |
177 | swtab[i].st_bsize = bsize; | |
178 | } | |
179 | ||
180 | /* | |
181 | * Allocate a pager structure and associated resources. | |
182 | * Note that if we are called from the pageout daemon (handle == NULL) | |
183 | * we should not wait for memory as it could resulting in deadlock. | |
184 | */ | |
719508e0 | 185 | static vm_pager_t |
619edcce KM |
186 | swap_pager_alloc(handle, size, prot) |
187 | caddr_t handle; | |
188 | register vm_size_t size; | |
189 | vm_prot_t prot; | |
190 | { | |
191 | register vm_pager_t pager; | |
192 | register sw_pager_t swp; | |
193 | struct swtab *swt; | |
194 | int waitok; | |
195 | ||
196 | #ifdef DEBUG | |
197 | if (swpagerdebug & (SDB_FOLLOW|SDB_ALLOC)) | |
198 | printf("swpg_alloc(%x, %x, %x)\n", handle, size, prot); | |
199 | #endif | |
200 | /* | |
201 | * If this is a "named" anonymous region, look it up and | |
202 | * return the appropriate pager if it exists. | |
203 | */ | |
204 | if (handle) { | |
205 | pager = vm_pager_lookup(&swap_pager_list, handle); | |
ffe0d082 | 206 | if (pager != NULL) { |
619edcce KM |
207 | /* |
208 | * Use vm_object_lookup to gain a reference | |
209 | * to the object and also to remove from the | |
210 | * object cache. | |
211 | */ | |
ffe0d082 | 212 | if (vm_object_lookup(pager) == NULL) |
619edcce KM |
213 | panic("swap_pager_alloc: bad object"); |
214 | return(pager); | |
215 | } | |
216 | } | |
217 | /* | |
218 | * Pager doesn't exist, allocate swap management resources | |
219 | * and initialize. | |
220 | */ | |
221 | waitok = handle ? M_WAITOK : M_NOWAIT; | |
222 | pager = (vm_pager_t)malloc(sizeof *pager, M_VMPAGER, waitok); | |
ffe0d082 MK |
223 | if (pager == NULL) |
224 | return(NULL); | |
619edcce KM |
225 | swp = (sw_pager_t)malloc(sizeof *swp, M_VMPGDATA, waitok); |
226 | if (swp == NULL) { | |
227 | #ifdef DEBUG | |
228 | if (swpagerdebug & SDB_FAIL) | |
229 | printf("swpg_alloc: swpager malloc failed\n"); | |
230 | #endif | |
231 | free((caddr_t)pager, M_VMPAGER); | |
ffe0d082 | 232 | return(NULL); |
619edcce KM |
233 | } |
234 | size = round_page(size); | |
235 | for (swt = swtab; swt->st_osize; swt++) | |
236 | if (size <= swt->st_osize) | |
237 | break; | |
238 | #ifdef DEBUG | |
239 | swt->st_inuse++; | |
240 | swt->st_usecnt++; | |
241 | #endif | |
242 | swp->sw_osize = size; | |
243 | swp->sw_bsize = swt->st_bsize; | |
244 | swp->sw_nblocks = (btodb(size) + swp->sw_bsize - 1) / swp->sw_bsize; | |
245 | swp->sw_blocks = (sw_blk_t) | |
246 | malloc(swp->sw_nblocks*sizeof(*swp->sw_blocks), | |
247 | M_VMPGDATA, M_NOWAIT); | |
248 | if (swp->sw_blocks == NULL) { | |
249 | free((caddr_t)swp, M_VMPGDATA); | |
250 | free((caddr_t)pager, M_VMPAGER); | |
251 | #ifdef DEBUG | |
252 | if (swpagerdebug & SDB_FAIL) | |
253 | printf("swpg_alloc: sw_blocks malloc failed\n"); | |
254 | swt->st_inuse--; | |
255 | swt->st_usecnt--; | |
256 | #endif | |
257 | return(FALSE); | |
258 | } | |
259 | bzero((caddr_t)swp->sw_blocks, | |
260 | swp->sw_nblocks * sizeof(*swp->sw_blocks)); | |
261 | swp->sw_poip = 0; | |
262 | if (handle) { | |
263 | vm_object_t object; | |
264 | ||
265 | swp->sw_flags = SW_NAMED; | |
266 | queue_enter(&swap_pager_list, pager, vm_pager_t, pg_list); | |
267 | /* | |
268 | * Consistant with other pagers: return with object | |
269 | * referenced. Can't do this with handle == NULL | |
270 | * since it might be the pageout daemon calling. | |
271 | */ | |
272 | object = vm_object_allocate(size); | |
273 | vm_object_enter(object, pager); | |
274 | vm_object_setpager(object, pager, 0, FALSE); | |
275 | } else { | |
276 | swp->sw_flags = 0; | |
277 | queue_init(&pager->pg_list); | |
278 | } | |
279 | pager->pg_handle = handle; | |
280 | pager->pg_ops = &swappagerops; | |
281 | pager->pg_type = PG_SWAP; | |
282 | pager->pg_data = (caddr_t)swp; | |
283 | ||
284 | #ifdef DEBUG | |
285 | if (swpagerdebug & SDB_ALLOC) | |
286 | printf("swpg_alloc: pg_data %x, %x of %x at %x\n", | |
287 | swp, swp->sw_nblocks, swp->sw_bsize, swp->sw_blocks); | |
288 | #endif | |
289 | return(pager); | |
290 | } | |
291 | ||
719508e0 | 292 | static void |
619edcce KM |
293 | swap_pager_dealloc(pager) |
294 | vm_pager_t pager; | |
295 | { | |
296 | register int i; | |
297 | register sw_blk_t bp; | |
298 | register sw_pager_t swp; | |
299 | struct swtab *swt; | |
300 | int s; | |
301 | ||
302 | #ifdef DEBUG | |
303 | /* save panic time state */ | |
304 | if ((swpagerdebug & SDB_ANOMPANIC) && panicstr) | |
305 | return; | |
306 | if (swpagerdebug & (SDB_FOLLOW|SDB_ALLOC)) | |
307 | printf("swpg_dealloc(%x)\n", pager); | |
308 | #endif | |
309 | /* | |
310 | * Remove from list right away so lookups will fail if we | |
311 | * block for pageout completion. | |
312 | */ | |
313 | swp = (sw_pager_t) pager->pg_data; | |
314 | if (swp->sw_flags & SW_NAMED) { | |
315 | queue_remove(&swap_pager_list, pager, vm_pager_t, pg_list); | |
316 | swp->sw_flags &= ~SW_NAMED; | |
317 | } | |
318 | #ifdef DEBUG | |
319 | for (swt = swtab; swt->st_osize; swt++) | |
320 | if (swp->sw_osize <= swt->st_osize) | |
321 | break; | |
322 | swt->st_inuse--; | |
323 | #endif | |
324 | ||
325 | /* | |
326 | * Wait for all pageouts to finish and remove | |
327 | * all entries from cleaning list. | |
328 | */ | |
329 | s = splbio(); | |
330 | while (swp->sw_poip) { | |
331 | swp->sw_flags |= SW_WANTED; | |
719508e0 | 332 | assert_wait((int)swp, 0); |
619edcce KM |
333 | thread_block(); |
334 | } | |
335 | splx(s); | |
ffe0d082 | 336 | (void) swap_pager_clean(NULL, B_WRITE); |
619edcce KM |
337 | |
338 | /* | |
339 | * Free left over swap blocks | |
340 | */ | |
341 | for (i = 0, bp = swp->sw_blocks; i < swp->sw_nblocks; i++, bp++) | |
342 | if (bp->swb_block) { | |
343 | #ifdef DEBUG | |
344 | if (swpagerdebug & (SDB_ALLOCBLK|SDB_FULL)) | |
345 | printf("swpg_dealloc: blk %x\n", | |
346 | bp->swb_block); | |
347 | #endif | |
348 | rmfree(swapmap, swp->sw_bsize, bp->swb_block); | |
349 | } | |
350 | /* | |
351 | * Free swap management resources | |
352 | */ | |
353 | free((caddr_t)swp->sw_blocks, M_VMPGDATA); | |
354 | free((caddr_t)swp, M_VMPGDATA); | |
355 | free((caddr_t)pager, M_VMPAGER); | |
356 | } | |
357 | ||
719508e0 | 358 | static int |
619edcce KM |
359 | swap_pager_getpage(pager, m, sync) |
360 | vm_pager_t pager; | |
361 | vm_page_t m; | |
362 | boolean_t sync; | |
363 | { | |
364 | #ifdef DEBUG | |
365 | if (swpagerdebug & SDB_FOLLOW) | |
366 | printf("swpg_getpage(%x, %x, %d)\n", pager, m, sync); | |
367 | #endif | |
368 | return(swap_pager_io((sw_pager_t)pager->pg_data, m, B_READ)); | |
369 | } | |
370 | ||
719508e0 | 371 | static int |
619edcce KM |
372 | swap_pager_putpage(pager, m, sync) |
373 | vm_pager_t pager; | |
374 | vm_page_t m; | |
375 | boolean_t sync; | |
376 | { | |
377 | int flags; | |
378 | ||
379 | #ifdef DEBUG | |
380 | if (swpagerdebug & SDB_FOLLOW) | |
381 | printf("swpg_putpage(%x, %x, %d)\n", pager, m, sync); | |
382 | #endif | |
ffe0d082 MK |
383 | if (pager == NULL) { |
384 | (void) swap_pager_clean(NULL, B_WRITE); | |
2601dc74 | 385 | return (VM_PAGER_OK); /* ??? */ |
619edcce KM |
386 | } |
387 | flags = B_WRITE; | |
388 | if (!sync) | |
389 | flags |= B_ASYNC; | |
390 | return(swap_pager_io((sw_pager_t)pager->pg_data, m, flags)); | |
391 | } | |
392 | ||
719508e0 | 393 | static boolean_t |
619edcce KM |
394 | swap_pager_haspage(pager, offset) |
395 | vm_pager_t pager; | |
396 | vm_offset_t offset; | |
397 | { | |
398 | register sw_pager_t swp; | |
399 | register sw_blk_t swb; | |
400 | int ix; | |
401 | ||
402 | #ifdef DEBUG | |
403 | if (swpagerdebug & (SDB_FOLLOW|SDB_ALLOCBLK)) | |
404 | printf("swpg_haspage(%x, %x) ", pager, offset); | |
405 | #endif | |
406 | swp = (sw_pager_t) pager->pg_data; | |
407 | ix = offset / dbtob(swp->sw_bsize); | |
408 | if (swp->sw_blocks == NULL || ix >= swp->sw_nblocks) { | |
409 | #ifdef DEBUG | |
410 | if (swpagerdebug & (SDB_FAIL|SDB_FOLLOW|SDB_ALLOCBLK)) | |
411 | printf("swpg_haspage: %x bad offset %x, ix %x\n", | |
412 | swp->sw_blocks, offset, ix); | |
413 | #endif | |
414 | return(FALSE); | |
415 | } | |
416 | swb = &swp->sw_blocks[ix]; | |
417 | if (swb->swb_block) | |
418 | ix = atop(offset % dbtob(swp->sw_bsize)); | |
419 | #ifdef DEBUG | |
420 | if (swpagerdebug & SDB_ALLOCBLK) | |
421 | printf("%x blk %x+%x ", swp->sw_blocks, swb->swb_block, ix); | |
422 | if (swpagerdebug & (SDB_FOLLOW|SDB_ALLOCBLK)) | |
423 | printf("-> %c\n", | |
424 | "FT"[swb->swb_block && (swb->swb_mask & (1 << ix))]); | |
425 | #endif | |
426 | if (swb->swb_block && (swb->swb_mask & (1 << ix))) | |
427 | return(TRUE); | |
428 | return(FALSE); | |
429 | } | |
430 | ||
431 | /* | |
432 | * Scaled down version of swap(). | |
433 | * Assumes that PAGE_SIZE < MAXPHYS; i.e. only one operation needed. | |
434 | * BOGUS: lower level IO routines expect a KVA so we have to map our | |
435 | * provided physical page into the KVA to keep them happy. | |
436 | */ | |
719508e0 | 437 | static int |
619edcce KM |
438 | swap_pager_io(swp, m, flags) |
439 | register sw_pager_t swp; | |
440 | vm_page_t m; | |
441 | int flags; | |
442 | { | |
443 | register struct buf *bp; | |
444 | register sw_blk_t swb; | |
445 | register int s; | |
446 | int ix; | |
447 | boolean_t rv; | |
448 | vm_offset_t kva, off; | |
449 | swp_clean_t spc; | |
450 | ||
451 | #ifdef DEBUG | |
452 | /* save panic time state */ | |
453 | if ((swpagerdebug & SDB_ANOMPANIC) && panicstr) | |
719508e0 | 454 | return (VM_PAGER_FAIL); /* XXX: correct return? */ |
619edcce KM |
455 | if (swpagerdebug & (SDB_FOLLOW|SDB_IO)) |
456 | printf("swpg_io(%x, %x, %x)\n", swp, m, flags); | |
457 | #endif | |
458 | ||
459 | /* | |
460 | * For reads (pageins) and synchronous writes, we clean up | |
07a039d6 | 461 | * all completed async pageouts. |
619edcce KM |
462 | */ |
463 | if ((flags & B_ASYNC) == 0) { | |
464 | s = splbio(); | |
07a039d6 MH |
465 | #ifdef DEBUG |
466 | /* | |
467 | * Check to see if this page is currently being cleaned. | |
468 | * If it is, we just wait til the operation is done before | |
469 | * continuing. | |
470 | */ | |
619edcce | 471 | while (swap_pager_clean(m, flags&B_READ)) { |
07a039d6 MH |
472 | if (swpagerdebug & SDB_ANOM) |
473 | printf("swap_pager_io: page %x cleaning\n", m); | |
474 | ||
619edcce | 475 | swp->sw_flags |= SW_WANTED; |
719508e0 | 476 | assert_wait((int)swp, 0); |
619edcce KM |
477 | thread_block(); |
478 | } | |
07a039d6 MH |
479 | #else |
480 | (void) swap_pager_clean(m, flags&B_READ); | |
481 | #endif | |
619edcce KM |
482 | splx(s); |
483 | } | |
484 | /* | |
485 | * For async writes (pageouts), we cleanup completed pageouts so | |
486 | * that all available resources are freed. Also tells us if this | |
487 | * page is already being cleaned. If it is, or no resources | |
488 | * are available, we try again later. | |
489 | */ | |
07a039d6 MH |
490 | else if (swap_pager_clean(m, B_WRITE) || |
491 | queue_empty(&swap_pager_free)) { | |
492 | #ifdef DEBUG | |
493 | if ((swpagerdebug & SDB_ANOM) && | |
494 | !queue_empty(&swap_pager_free)) | |
495 | printf("swap_pager_io: page %x already cleaning\n", m); | |
496 | #endif | |
619edcce | 497 | return(VM_PAGER_FAIL); |
07a039d6 | 498 | } |
619edcce KM |
499 | |
500 | /* | |
501 | * Determine swap block and allocate as necessary. | |
502 | */ | |
503 | off = m->offset + m->object->paging_offset; | |
504 | ix = off / dbtob(swp->sw_bsize); | |
505 | if (swp->sw_blocks == NULL || ix >= swp->sw_nblocks) { | |
506 | #ifdef DEBUG | |
507 | if (swpagerdebug & SDB_FAIL) | |
508 | printf("swpg_io: bad offset %x+%x(%d) in %x\n", | |
509 | m->offset, m->object->paging_offset, | |
510 | ix, swp->sw_blocks); | |
511 | #endif | |
512 | return(VM_PAGER_FAIL); | |
513 | } | |
514 | swb = &swp->sw_blocks[ix]; | |
515 | off = off % dbtob(swp->sw_bsize); | |
516 | if (flags & B_READ) { | |
517 | if (swb->swb_block == 0 || | |
518 | (swb->swb_mask & (1 << atop(off))) == 0) { | |
519 | #ifdef DEBUG | |
520 | if (swpagerdebug & (SDB_ALLOCBLK|SDB_FAIL)) | |
521 | printf("swpg_io: %x bad read: blk %x+%x, mask %x, off %x+%x\n", | |
522 | swp->sw_blocks, | |
523 | swb->swb_block, atop(off), | |
524 | swb->swb_mask, | |
525 | m->offset, m->object->paging_offset); | |
526 | #endif | |
527 | /* XXX: should we zero page here?? */ | |
528 | return(VM_PAGER_FAIL); | |
529 | } | |
530 | } else if (swb->swb_block == 0) { | |
531 | swb->swb_block = rmalloc(swapmap, swp->sw_bsize); | |
532 | if (swb->swb_block == 0) { | |
533 | #ifdef DEBUG | |
534 | if (swpagerdebug & SDB_FAIL) | |
535 | printf("swpg_io: rmalloc of %x failed\n", | |
536 | swp->sw_bsize); | |
537 | #endif | |
538 | return(VM_PAGER_FAIL); | |
539 | } | |
540 | #ifdef DEBUG | |
541 | if (swpagerdebug & (SDB_FULL|SDB_ALLOCBLK)) | |
542 | printf("swpg_io: %x alloc blk %x at ix %x\n", | |
543 | swp->sw_blocks, swb->swb_block, ix); | |
544 | #endif | |
545 | } | |
546 | ||
547 | /* | |
548 | * Allocate a kernel virtual address and initialize so that PTE | |
549 | * is available for lower level IO drivers. | |
550 | */ | |
551 | kva = vm_pager_map_page(m); | |
552 | ||
553 | /* | |
554 | * Get a swap buffer header and perform the IO | |
555 | */ | |
556 | s = splbio(); | |
557 | while (bswlist.av_forw == NULL) { | |
558 | #ifdef DEBUG | |
559 | if (swpagerdebug & SDB_ANOM) | |
07a039d6 | 560 | printf("swap_pager_io: wait on swbuf for %x (%d)\n", |
619edcce KM |
561 | m, flags); |
562 | #endif | |
563 | bswlist.b_flags |= B_WANTED; | |
564 | sleep((caddr_t)&bswlist, PSWP+1); | |
565 | } | |
566 | bp = bswlist.av_forw; | |
567 | bswlist.av_forw = bp->av_forw; | |
568 | splx(s); | |
569 | bp->b_flags = B_BUSY | (flags & B_READ); | |
ffe0d082 | 570 | bp->b_proc = &proc0; /* XXX (but without B_PHYS set this is ok) */ |
619edcce KM |
571 | bp->b_un.b_addr = (caddr_t)kva; |
572 | bp->b_blkno = swb->swb_block + btodb(off); | |
573 | VHOLD(swapdev_vp); | |
574 | bp->b_vp = swapdev_vp; | |
e950b33e KM |
575 | if (swapdev_vp->v_type == VBLK) |
576 | bp->b_dev = swapdev_vp->v_rdev; | |
619edcce | 577 | bp->b_bcount = PAGE_SIZE; |
3c334a40 KM |
578 | if ((bp->b_flags & B_READ) == 0) { |
579 | bp->b_dirtyoff = 0; | |
580 | bp->b_dirtyend = PAGE_SIZE; | |
619edcce | 581 | swapdev_vp->v_numoutput++; |
3c334a40 | 582 | } |
619edcce KM |
583 | |
584 | /* | |
585 | * If this is an async write we set up additional buffer fields | |
586 | * and place a "cleaning" entry on the inuse queue. | |
587 | */ | |
588 | if ((flags & (B_READ|B_ASYNC)) == B_ASYNC) { | |
589 | #ifdef DEBUG | |
590 | if (queue_empty(&swap_pager_free)) | |
591 | panic("swpg_io: lost spc"); | |
592 | #endif | |
593 | queue_remove_first(&swap_pager_free, | |
594 | spc, swp_clean_t, spc_list); | |
595 | #ifdef DEBUG | |
596 | if (spc->spc_flags != SPC_FREE) | |
597 | panic("swpg_io: bad free spc"); | |
598 | #endif | |
599 | spc->spc_flags = SPC_BUSY; | |
600 | spc->spc_bp = bp; | |
601 | spc->spc_swp = swp; | |
602 | spc->spc_kva = kva; | |
603 | spc->spc_m = m; | |
619edcce KM |
604 | bp->b_flags |= B_CALL; |
605 | bp->b_iodone = swap_pager_iodone; | |
606 | s = splbio(); | |
607 | swp->sw_poip++; | |
608 | queue_enter(&swap_pager_inuse, spc, swp_clean_t, spc_list); | |
609 | ||
610 | #ifdef DEBUG | |
611 | swap_pager_poip++; | |
612 | if (swpagerdebug & SDB_WRITE) | |
613 | printf("swpg_io: write: bp=%x swp=%x spc=%x poip=%d\n", | |
614 | bp, swp, spc, swp->sw_poip); | |
615 | if ((swpagerdebug & SDB_ALLOCBLK) && | |
616 | (swb->swb_mask & (1 << atop(off))) == 0) | |
617 | printf("swpg_io: %x write blk %x+%x\n", | |
618 | swp->sw_blocks, swb->swb_block, atop(off)); | |
619 | #endif | |
620 | swb->swb_mask |= (1 << atop(off)); | |
619edcce KM |
621 | splx(s); |
622 | } | |
623 | #ifdef DEBUG | |
624 | if (swpagerdebug & SDB_IO) | |
625 | printf("swpg_io: IO start: bp %x, db %x, va %x, pa %x\n", | |
626 | bp, swb->swb_block+btodb(off), kva, VM_PAGE_TO_PHYS(m)); | |
627 | #endif | |
628 | VOP_STRATEGY(bp); | |
629 | if ((flags & (B_READ|B_ASYNC)) == B_ASYNC) { | |
630 | #ifdef DEBUG | |
631 | if (swpagerdebug & SDB_IO) | |
632 | printf("swpg_io: IO started: bp %x\n", bp); | |
633 | #endif | |
634 | return(VM_PAGER_PEND); | |
635 | } | |
636 | s = splbio(); | |
637 | #ifdef DEBUG | |
638 | if (flags & B_READ) | |
639 | swap_pager_piip++; | |
640 | else | |
641 | swap_pager_poip++; | |
642 | #endif | |
643 | while ((bp->b_flags & B_DONE) == 0) { | |
719508e0 | 644 | assert_wait((int)bp, 0); |
619edcce KM |
645 | thread_block(); |
646 | } | |
647 | #ifdef DEBUG | |
648 | if (flags & B_READ) | |
649 | --swap_pager_piip; | |
650 | else | |
651 | --swap_pager_poip; | |
652 | #endif | |
653 | rv = (bp->b_flags & B_ERROR) ? VM_PAGER_FAIL : VM_PAGER_OK; | |
654 | bp->b_flags &= ~(B_BUSY|B_WANTED|B_PHYS|B_PAGET|B_UAREA|B_DIRTY); | |
655 | bp->av_forw = bswlist.av_forw; | |
656 | bswlist.av_forw = bp; | |
657 | if (bp->b_vp) | |
658 | brelvp(bp); | |
659 | if (bswlist.b_flags & B_WANTED) { | |
660 | bswlist.b_flags &= ~B_WANTED; | |
661 | thread_wakeup((int)&bswlist); | |
662 | } | |
663 | if ((flags & B_READ) == 0 && rv == VM_PAGER_OK) { | |
07a039d6 | 664 | m->clean = TRUE; |
619edcce KM |
665 | pmap_clear_modify(VM_PAGE_TO_PHYS(m)); |
666 | } | |
667 | splx(s); | |
668 | #ifdef DEBUG | |
669 | if (swpagerdebug & SDB_IO) | |
670 | printf("swpg_io: IO done: bp %x, rv %d\n", bp, rv); | |
671 | if ((swpagerdebug & SDB_FAIL) && rv == VM_PAGER_FAIL) | |
672 | printf("swpg_io: IO error\n"); | |
673 | #endif | |
674 | vm_pager_unmap_page(kva); | |
675 | return(rv); | |
676 | } | |
677 | ||
719508e0 | 678 | static boolean_t |
619edcce KM |
679 | swap_pager_clean(m, rw) |
680 | vm_page_t m; | |
681 | int rw; | |
682 | { | |
683 | register swp_clean_t spc, tspc; | |
684 | register int s; | |
685 | ||
686 | #ifdef DEBUG | |
687 | /* save panic time state */ | |
688 | if ((swpagerdebug & SDB_ANOMPANIC) && panicstr) | |
2601dc74 | 689 | return (FALSE); /* ??? */ |
619edcce KM |
690 | if (swpagerdebug & SDB_FOLLOW) |
691 | printf("swpg_clean(%x, %d)\n", m, rw); | |
692 | #endif | |
ffe0d082 | 693 | tspc = NULL; |
619edcce KM |
694 | for (;;) { |
695 | /* | |
696 | * Look up and removal from inuse list must be done | |
697 | * at splbio() to avoid conflicts with swap_pager_iodone. | |
698 | */ | |
699 | s = splbio(); | |
700 | spc = (swp_clean_t) queue_first(&swap_pager_inuse); | |
701 | while (!queue_end(&swap_pager_inuse, (queue_entry_t)spc)) { | |
702 | if ((spc->spc_flags & SPC_DONE) && | |
703 | swap_pager_finish(spc)) { | |
704 | queue_remove(&swap_pager_inuse, spc, | |
705 | swp_clean_t, spc_list); | |
706 | break; | |
707 | } | |
708 | if (m && m == spc->spc_m) { | |
709 | #ifdef DEBUG | |
710 | if (swpagerdebug & SDB_ANOM) | |
07a039d6 | 711 | printf("swap_pager_clean: page %x on list, flags %x\n", |
619edcce KM |
712 | m, spc->spc_flags); |
713 | #endif | |
714 | tspc = spc; | |
715 | } | |
716 | spc = (swp_clean_t) queue_next(&spc->spc_list); | |
717 | } | |
718 | ||
719 | /* | |
720 | * No operations done, thats all we can do for now. | |
721 | */ | |
722 | if (queue_end(&swap_pager_inuse, (queue_entry_t)spc)) | |
723 | break; | |
724 | splx(s); | |
725 | ||
726 | /* | |
727 | * The desired page was found to be busy earlier in | |
728 | * the scan but has since completed. | |
729 | */ | |
730 | if (tspc && tspc == spc) { | |
731 | #ifdef DEBUG | |
732 | if (swpagerdebug & SDB_ANOM) | |
07a039d6 | 733 | printf("swap_pager_clean: page %x done while looking\n", |
619edcce KM |
734 | m); |
735 | #endif | |
ffe0d082 | 736 | tspc = NULL; |
619edcce KM |
737 | } |
738 | spc->spc_flags = SPC_FREE; | |
739 | vm_pager_unmap_page(spc->spc_kva); | |
740 | queue_enter(&swap_pager_free, spc, swp_clean_t, spc_list); | |
741 | #ifdef DEBUG | |
742 | if (swpagerdebug & SDB_WRITE) | |
743 | printf("swpg_clean: free spc %x\n", spc); | |
744 | #endif | |
745 | } | |
07a039d6 | 746 | #ifdef DEBUG |
619edcce KM |
747 | /* |
748 | * If we found that the desired page is already being cleaned | |
749 | * mark it so that swap_pager_iodone() will not set the clean | |
750 | * flag before the pageout daemon has another chance to clean it. | |
751 | */ | |
752 | if (tspc && rw == B_WRITE) { | |
619edcce | 753 | if (swpagerdebug & SDB_ANOM) |
07a039d6 MH |
754 | printf("swap_pager_clean: page %x on clean list\n", |
755 | tspc); | |
619edcce KM |
756 | tspc->spc_flags |= SPC_DIRTY; |
757 | } | |
07a039d6 | 758 | #endif |
619edcce KM |
759 | splx(s); |
760 | ||
761 | #ifdef DEBUG | |
762 | if (swpagerdebug & SDB_WRITE) | |
763 | printf("swpg_clean: return %d\n", tspc ? TRUE : FALSE); | |
764 | if ((swpagerdebug & SDB_ANOM) && tspc) | |
765 | printf("swpg_clean: %s of cleaning page %x\n", | |
766 | rw == B_READ ? "get" : "put", m); | |
767 | #endif | |
768 | return(tspc ? TRUE : FALSE); | |
769 | } | |
770 | ||
719508e0 | 771 | static int |
619edcce KM |
772 | swap_pager_finish(spc) |
773 | register swp_clean_t spc; | |
774 | { | |
775 | vm_object_t object = spc->spc_m->object; | |
776 | ||
777 | /* | |
778 | * Mark the paging operation as done. | |
779 | * (XXX) If we cannot get the lock, leave it til later. | |
780 | * (XXX) Also we are assuming that an async write is a | |
781 | * pageout operation that has incremented the counter. | |
782 | */ | |
783 | if (!vm_object_lock_try(object)) | |
784 | return(0); | |
785 | ||
619edcce KM |
786 | if (--object->paging_in_progress == 0) |
787 | thread_wakeup((int) object); | |
788 | ||
07a039d6 | 789 | #ifdef DEBUG |
619edcce KM |
790 | /* |
791 | * XXX: this isn't even close to the right thing to do, | |
792 | * introduces a variety of race conditions. | |
793 | * | |
794 | * If dirty, vm_pageout() has attempted to clean the page | |
795 | * again. In this case we do not do anything as we will | |
07a039d6 | 796 | * see the page again shortly. |
619edcce | 797 | */ |
07a039d6 MH |
798 | if (spc->spc_flags & SPC_DIRTY) { |
799 | if (swpagerdebug & SDB_ANOM) | |
800 | printf("swap_pager_finish: page %x dirty again\n", | |
801 | spc->spc_m); | |
802 | spc->spc_m->busy = FALSE; | |
803 | PAGE_WAKEUP(spc->spc_m); | |
804 | vm_object_unlock(object); | |
805 | return(1); | |
619edcce | 806 | } |
07a039d6 | 807 | #endif |
619edcce | 808 | /* |
07a039d6 MH |
809 | * If no error mark as clean and inform the pmap system. |
810 | * If error, mark as dirty so we will try again. | |
811 | * (XXX could get stuck doing this, should give up after awhile) | |
619edcce | 812 | */ |
07a039d6 MH |
813 | if (spc->spc_flags & SPC_ERROR) { |
814 | printf("swap_pager_finish: clean of page %x failed\n", | |
815 | VM_PAGE_TO_PHYS(spc->spc_m)); | |
816 | spc->spc_m->laundry = TRUE; | |
817 | } else { | |
818 | spc->spc_m->clean = TRUE; | |
819 | pmap_clear_modify(VM_PAGE_TO_PHYS(spc->spc_m)); | |
820 | } | |
821 | spc->spc_m->busy = FALSE; | |
619edcce KM |
822 | PAGE_WAKEUP(spc->spc_m); |
823 | ||
824 | vm_object_unlock(object); | |
825 | return(1); | |
826 | } | |
827 | ||
719508e0 | 828 | static void |
619edcce KM |
829 | swap_pager_iodone(bp) |
830 | register struct buf *bp; | |
831 | { | |
832 | register swp_clean_t spc; | |
833 | daddr_t blk; | |
834 | int s; | |
835 | ||
836 | #ifdef DEBUG | |
837 | /* save panic time state */ | |
838 | if ((swpagerdebug & SDB_ANOMPANIC) && panicstr) | |
839 | return; | |
840 | if (swpagerdebug & SDB_FOLLOW) | |
841 | printf("swpg_iodone(%x)\n", bp); | |
842 | #endif | |
843 | s = splbio(); | |
844 | spc = (swp_clean_t) queue_first(&swap_pager_inuse); | |
845 | while (!queue_end(&swap_pager_inuse, (queue_entry_t)spc)) { | |
846 | if (spc->spc_bp == bp) | |
847 | break; | |
848 | spc = (swp_clean_t) queue_next(&spc->spc_list); | |
849 | } | |
850 | #ifdef DEBUG | |
851 | if (queue_end(&swap_pager_inuse, (queue_entry_t)spc)) | |
07a039d6 | 852 | panic("swap_pager_iodone: bp not found"); |
619edcce KM |
853 | #endif |
854 | ||
855 | spc->spc_flags &= ~SPC_BUSY; | |
856 | spc->spc_flags |= SPC_DONE; | |
857 | if (bp->b_flags & B_ERROR) | |
858 | spc->spc_flags |= SPC_ERROR; | |
859 | spc->spc_bp = NULL; | |
860 | blk = bp->b_blkno; | |
861 | ||
862 | #ifdef DEBUG | |
863 | --swap_pager_poip; | |
864 | if (swpagerdebug & SDB_WRITE) | |
865 | printf("swpg_iodone: bp=%x swp=%x flags=%x spc=%x poip=%x\n", | |
866 | bp, spc->spc_swp, spc->spc_swp->sw_flags, | |
867 | spc, spc->spc_swp->sw_poip); | |
868 | #endif | |
869 | ||
870 | spc->spc_swp->sw_poip--; | |
871 | if (spc->spc_swp->sw_flags & SW_WANTED) { | |
872 | spc->spc_swp->sw_flags &= ~SW_WANTED; | |
873 | thread_wakeup((int)spc->spc_swp); | |
874 | } | |
875 | ||
876 | bp->b_flags &= ~(B_BUSY|B_WANTED|B_PHYS|B_PAGET|B_UAREA|B_DIRTY); | |
877 | bp->av_forw = bswlist.av_forw; | |
878 | bswlist.av_forw = bp; | |
879 | if (bp->b_vp) | |
880 | brelvp(bp); | |
881 | if (bswlist.b_flags & B_WANTED) { | |
882 | bswlist.b_flags &= ~B_WANTED; | |
883 | thread_wakeup((int)&bswlist); | |
884 | } | |
619edcce KM |
885 | thread_wakeup((int) &vm_pages_needed); |
886 | splx(s); | |
887 | } | |
888 | #endif |