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 | * | |
112457b3 | 14 | * @(#)swap_pager.c 7.5 (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 | ||
27 | #include "param.h" | |
619edcce KM |
28 | #include "proc.h" |
29 | #include "buf.h" | |
30 | #include "map.h" | |
31 | #include "systm.h" | |
32 | #include "specdev.h" | |
33 | #include "vnode.h" | |
34 | #include "malloc.h" | |
35 | #include "queue.h" | |
36 | ||
112457b3 | 37 | #include "vm.h" |
ffe0d082 MK |
38 | #include "vm_page.h" |
39 | #include "vm_pageout.h" | |
40 | #include "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]; | |
68 | typedef struct swpagerclean *swp_clean_t; | |
69 | ||
619edcce KM |
70 | /* spc_flags values */ |
71 | #define SPC_FREE 0x00 | |
72 | #define SPC_BUSY 0x01 | |
73 | #define SPC_DONE 0x02 | |
74 | #define SPC_ERROR 0x04 | |
75 | #define SPC_DIRTY 0x08 | |
76 | ||
77 | struct swtab { | |
78 | vm_size_t st_osize; /* size of object (bytes) */ | |
79 | int st_bsize; /* vs. size of swap block (DEV_BSIZE units) */ | |
80 | #ifdef DEBUG | |
81 | u_long st_inuse; /* number in this range in use */ | |
82 | u_long st_usecnt; /* total used of this size */ | |
83 | #endif | |
84 | } swtab[NSWSIZES+1]; | |
85 | ||
86 | #ifdef DEBUG | |
87 | int swap_pager_pendingio; /* max pending async "clean" ops */ | |
88 | int swap_pager_poip; /* pageouts in progress */ | |
89 | int swap_pager_piip; /* pageins in progress */ | |
90 | #endif | |
91 | ||
92 | queue_head_t swap_pager_inuse; /* list of pending page cleans */ | |
93 | queue_head_t swap_pager_free; /* list of free pager clean structs */ | |
94 | queue_head_t swap_pager_list; /* list of "named" anon regions */ | |
95 | ||
96 | void | |
97 | swap_pager_init() | |
98 | { | |
99 | register swp_clean_t spc; | |
100 | register int i, bsize; | |
101 | extern int dmmin, dmmax; | |
102 | int maxbsize; | |
103 | ||
104 | #ifdef DEBUG | |
105 | if (swpagerdebug & (SDB_FOLLOW|SDB_INIT)) | |
106 | printf("swpg_init()\n"); | |
107 | #endif | |
108 | dfltpagerops = &swappagerops; | |
109 | queue_init(&swap_pager_list); | |
110 | ||
111 | /* | |
112 | * Initialize clean lists | |
113 | */ | |
114 | queue_init(&swap_pager_inuse); | |
115 | queue_init(&swap_pager_free); | |
116 | for (i = 0, spc = swcleanlist; i < NPENDINGIO; i++, spc++) { | |
117 | queue_enter(&swap_pager_free, spc, swp_clean_t, spc_list); | |
118 | spc->spc_flags = SPC_FREE; | |
119 | } | |
120 | ||
121 | /* | |
122 | * Calculate the swap allocation constants. | |
123 | */ | |
124 | if (dmmin == 0) { | |
125 | dmmin = DMMIN; | |
126 | if (dmmin < CLBYTES/DEV_BSIZE) | |
127 | dmmin = CLBYTES/DEV_BSIZE; | |
128 | } | |
129 | if (dmmax == 0) | |
130 | dmmax = DMMAX; | |
131 | ||
132 | /* | |
133 | * Fill in our table of object size vs. allocation size | |
134 | */ | |
135 | bsize = btodb(PAGE_SIZE); | |
136 | if (bsize < dmmin) | |
137 | bsize = dmmin; | |
138 | maxbsize = btodb(sizeof(sw_bm_t) * NBBY * PAGE_SIZE); | |
139 | if (maxbsize > dmmax) | |
140 | maxbsize = dmmax; | |
141 | for (i = 0; i < NSWSIZES; i++) { | |
142 | swtab[i].st_osize = (vm_size_t) (MAXDADDRS * dbtob(bsize)); | |
143 | swtab[i].st_bsize = bsize; | |
144 | #ifdef DEBUG | |
145 | if (swpagerdebug & SDB_INIT) | |
146 | printf("swpg_init: ix %d, size %x, bsize %x\n", | |
147 | i, swtab[i].st_osize, swtab[i].st_bsize); | |
148 | #endif | |
149 | if (bsize >= maxbsize) | |
150 | break; | |
151 | bsize *= 2; | |
152 | } | |
153 | swtab[i].st_osize = 0; | |
154 | swtab[i].st_bsize = bsize; | |
155 | } | |
156 | ||
157 | /* | |
158 | * Allocate a pager structure and associated resources. | |
159 | * Note that if we are called from the pageout daemon (handle == NULL) | |
160 | * we should not wait for memory as it could resulting in deadlock. | |
161 | */ | |
162 | vm_pager_t | |
163 | swap_pager_alloc(handle, size, prot) | |
164 | caddr_t handle; | |
165 | register vm_size_t size; | |
166 | vm_prot_t prot; | |
167 | { | |
168 | register vm_pager_t pager; | |
169 | register sw_pager_t swp; | |
170 | struct swtab *swt; | |
171 | int waitok; | |
172 | ||
173 | #ifdef DEBUG | |
174 | if (swpagerdebug & (SDB_FOLLOW|SDB_ALLOC)) | |
175 | printf("swpg_alloc(%x, %x, %x)\n", handle, size, prot); | |
176 | #endif | |
177 | /* | |
178 | * If this is a "named" anonymous region, look it up and | |
179 | * return the appropriate pager if it exists. | |
180 | */ | |
181 | if (handle) { | |
182 | pager = vm_pager_lookup(&swap_pager_list, handle); | |
ffe0d082 | 183 | if (pager != NULL) { |
619edcce KM |
184 | /* |
185 | * Use vm_object_lookup to gain a reference | |
186 | * to the object and also to remove from the | |
187 | * object cache. | |
188 | */ | |
ffe0d082 | 189 | if (vm_object_lookup(pager) == NULL) |
619edcce KM |
190 | panic("swap_pager_alloc: bad object"); |
191 | return(pager); | |
192 | } | |
193 | } | |
194 | /* | |
195 | * Pager doesn't exist, allocate swap management resources | |
196 | * and initialize. | |
197 | */ | |
198 | waitok = handle ? M_WAITOK : M_NOWAIT; | |
199 | pager = (vm_pager_t)malloc(sizeof *pager, M_VMPAGER, waitok); | |
ffe0d082 MK |
200 | if (pager == NULL) |
201 | return(NULL); | |
619edcce KM |
202 | swp = (sw_pager_t)malloc(sizeof *swp, M_VMPGDATA, waitok); |
203 | if (swp == NULL) { | |
204 | #ifdef DEBUG | |
205 | if (swpagerdebug & SDB_FAIL) | |
206 | printf("swpg_alloc: swpager malloc failed\n"); | |
207 | #endif | |
208 | free((caddr_t)pager, M_VMPAGER); | |
ffe0d082 | 209 | return(NULL); |
619edcce KM |
210 | } |
211 | size = round_page(size); | |
212 | for (swt = swtab; swt->st_osize; swt++) | |
213 | if (size <= swt->st_osize) | |
214 | break; | |
215 | #ifdef DEBUG | |
216 | swt->st_inuse++; | |
217 | swt->st_usecnt++; | |
218 | #endif | |
219 | swp->sw_osize = size; | |
220 | swp->sw_bsize = swt->st_bsize; | |
221 | swp->sw_nblocks = (btodb(size) + swp->sw_bsize - 1) / swp->sw_bsize; | |
222 | swp->sw_blocks = (sw_blk_t) | |
223 | malloc(swp->sw_nblocks*sizeof(*swp->sw_blocks), | |
224 | M_VMPGDATA, M_NOWAIT); | |
225 | if (swp->sw_blocks == NULL) { | |
226 | free((caddr_t)swp, M_VMPGDATA); | |
227 | free((caddr_t)pager, M_VMPAGER); | |
228 | #ifdef DEBUG | |
229 | if (swpagerdebug & SDB_FAIL) | |
230 | printf("swpg_alloc: sw_blocks malloc failed\n"); | |
231 | swt->st_inuse--; | |
232 | swt->st_usecnt--; | |
233 | #endif | |
234 | return(FALSE); | |
235 | } | |
236 | bzero((caddr_t)swp->sw_blocks, | |
237 | swp->sw_nblocks * sizeof(*swp->sw_blocks)); | |
238 | swp->sw_poip = 0; | |
239 | if (handle) { | |
240 | vm_object_t object; | |
241 | ||
242 | swp->sw_flags = SW_NAMED; | |
243 | queue_enter(&swap_pager_list, pager, vm_pager_t, pg_list); | |
244 | /* | |
245 | * Consistant with other pagers: return with object | |
246 | * referenced. Can't do this with handle == NULL | |
247 | * since it might be the pageout daemon calling. | |
248 | */ | |
249 | object = vm_object_allocate(size); | |
250 | vm_object_enter(object, pager); | |
251 | vm_object_setpager(object, pager, 0, FALSE); | |
252 | } else { | |
253 | swp->sw_flags = 0; | |
254 | queue_init(&pager->pg_list); | |
255 | } | |
256 | pager->pg_handle = handle; | |
257 | pager->pg_ops = &swappagerops; | |
258 | pager->pg_type = PG_SWAP; | |
259 | pager->pg_data = (caddr_t)swp; | |
260 | ||
261 | #ifdef DEBUG | |
262 | if (swpagerdebug & SDB_ALLOC) | |
263 | printf("swpg_alloc: pg_data %x, %x of %x at %x\n", | |
264 | swp, swp->sw_nblocks, swp->sw_bsize, swp->sw_blocks); | |
265 | #endif | |
266 | return(pager); | |
267 | } | |
268 | ||
269 | void | |
270 | swap_pager_dealloc(pager) | |
271 | vm_pager_t pager; | |
272 | { | |
273 | register int i; | |
274 | register sw_blk_t bp; | |
275 | register sw_pager_t swp; | |
276 | struct swtab *swt; | |
277 | int s; | |
278 | ||
279 | #ifdef DEBUG | |
280 | /* save panic time state */ | |
281 | if ((swpagerdebug & SDB_ANOMPANIC) && panicstr) | |
282 | return; | |
283 | if (swpagerdebug & (SDB_FOLLOW|SDB_ALLOC)) | |
284 | printf("swpg_dealloc(%x)\n", pager); | |
285 | #endif | |
286 | /* | |
287 | * Remove from list right away so lookups will fail if we | |
288 | * block for pageout completion. | |
289 | */ | |
290 | swp = (sw_pager_t) pager->pg_data; | |
291 | if (swp->sw_flags & SW_NAMED) { | |
292 | queue_remove(&swap_pager_list, pager, vm_pager_t, pg_list); | |
293 | swp->sw_flags &= ~SW_NAMED; | |
294 | } | |
295 | #ifdef DEBUG | |
296 | for (swt = swtab; swt->st_osize; swt++) | |
297 | if (swp->sw_osize <= swt->st_osize) | |
298 | break; | |
299 | swt->st_inuse--; | |
300 | #endif | |
301 | ||
302 | /* | |
303 | * Wait for all pageouts to finish and remove | |
304 | * all entries from cleaning list. | |
305 | */ | |
306 | s = splbio(); | |
307 | while (swp->sw_poip) { | |
308 | swp->sw_flags |= SW_WANTED; | |
309 | assert_wait((int)swp); | |
310 | thread_block(); | |
311 | } | |
312 | splx(s); | |
ffe0d082 | 313 | (void) swap_pager_clean(NULL, B_WRITE); |
619edcce KM |
314 | |
315 | /* | |
316 | * Free left over swap blocks | |
317 | */ | |
318 | for (i = 0, bp = swp->sw_blocks; i < swp->sw_nblocks; i++, bp++) | |
319 | if (bp->swb_block) { | |
320 | #ifdef DEBUG | |
321 | if (swpagerdebug & (SDB_ALLOCBLK|SDB_FULL)) | |
322 | printf("swpg_dealloc: blk %x\n", | |
323 | bp->swb_block); | |
324 | #endif | |
325 | rmfree(swapmap, swp->sw_bsize, bp->swb_block); | |
326 | } | |
327 | /* | |
328 | * Free swap management resources | |
329 | */ | |
330 | free((caddr_t)swp->sw_blocks, M_VMPGDATA); | |
331 | free((caddr_t)swp, M_VMPGDATA); | |
332 | free((caddr_t)pager, M_VMPAGER); | |
333 | } | |
334 | ||
335 | swap_pager_getpage(pager, m, sync) | |
336 | vm_pager_t pager; | |
337 | vm_page_t m; | |
338 | boolean_t sync; | |
339 | { | |
340 | #ifdef DEBUG | |
341 | if (swpagerdebug & SDB_FOLLOW) | |
342 | printf("swpg_getpage(%x, %x, %d)\n", pager, m, sync); | |
343 | #endif | |
344 | return(swap_pager_io((sw_pager_t)pager->pg_data, m, B_READ)); | |
345 | } | |
346 | ||
347 | swap_pager_putpage(pager, m, sync) | |
348 | vm_pager_t pager; | |
349 | vm_page_t m; | |
350 | boolean_t sync; | |
351 | { | |
352 | int flags; | |
353 | ||
354 | #ifdef DEBUG | |
355 | if (swpagerdebug & SDB_FOLLOW) | |
356 | printf("swpg_putpage(%x, %x, %d)\n", pager, m, sync); | |
357 | #endif | |
ffe0d082 MK |
358 | if (pager == NULL) { |
359 | (void) swap_pager_clean(NULL, B_WRITE); | |
619edcce KM |
360 | return; |
361 | } | |
362 | flags = B_WRITE; | |
363 | if (!sync) | |
364 | flags |= B_ASYNC; | |
365 | return(swap_pager_io((sw_pager_t)pager->pg_data, m, flags)); | |
366 | } | |
367 | ||
368 | boolean_t | |
369 | swap_pager_haspage(pager, offset) | |
370 | vm_pager_t pager; | |
371 | vm_offset_t offset; | |
372 | { | |
373 | register sw_pager_t swp; | |
374 | register sw_blk_t swb; | |
375 | int ix; | |
376 | ||
377 | #ifdef DEBUG | |
378 | if (swpagerdebug & (SDB_FOLLOW|SDB_ALLOCBLK)) | |
379 | printf("swpg_haspage(%x, %x) ", pager, offset); | |
380 | #endif | |
381 | swp = (sw_pager_t) pager->pg_data; | |
382 | ix = offset / dbtob(swp->sw_bsize); | |
383 | if (swp->sw_blocks == NULL || ix >= swp->sw_nblocks) { | |
384 | #ifdef DEBUG | |
385 | if (swpagerdebug & (SDB_FAIL|SDB_FOLLOW|SDB_ALLOCBLK)) | |
386 | printf("swpg_haspage: %x bad offset %x, ix %x\n", | |
387 | swp->sw_blocks, offset, ix); | |
388 | #endif | |
389 | return(FALSE); | |
390 | } | |
391 | swb = &swp->sw_blocks[ix]; | |
392 | if (swb->swb_block) | |
393 | ix = atop(offset % dbtob(swp->sw_bsize)); | |
394 | #ifdef DEBUG | |
395 | if (swpagerdebug & SDB_ALLOCBLK) | |
396 | printf("%x blk %x+%x ", swp->sw_blocks, swb->swb_block, ix); | |
397 | if (swpagerdebug & (SDB_FOLLOW|SDB_ALLOCBLK)) | |
398 | printf("-> %c\n", | |
399 | "FT"[swb->swb_block && (swb->swb_mask & (1 << ix))]); | |
400 | #endif | |
401 | if (swb->swb_block && (swb->swb_mask & (1 << ix))) | |
402 | return(TRUE); | |
403 | return(FALSE); | |
404 | } | |
405 | ||
406 | /* | |
407 | * Scaled down version of swap(). | |
408 | * Assumes that PAGE_SIZE < MAXPHYS; i.e. only one operation needed. | |
409 | * BOGUS: lower level IO routines expect a KVA so we have to map our | |
410 | * provided physical page into the KVA to keep them happy. | |
411 | */ | |
412 | swap_pager_io(swp, m, flags) | |
413 | register sw_pager_t swp; | |
414 | vm_page_t m; | |
415 | int flags; | |
416 | { | |
417 | register struct buf *bp; | |
418 | register sw_blk_t swb; | |
419 | register int s; | |
420 | int ix; | |
421 | boolean_t rv; | |
422 | vm_offset_t kva, off; | |
423 | swp_clean_t spc; | |
424 | ||
425 | #ifdef DEBUG | |
426 | /* save panic time state */ | |
427 | if ((swpagerdebug & SDB_ANOMPANIC) && panicstr) | |
428 | return; | |
429 | if (swpagerdebug & (SDB_FOLLOW|SDB_IO)) | |
430 | printf("swpg_io(%x, %x, %x)\n", swp, m, flags); | |
431 | #endif | |
432 | ||
433 | /* | |
434 | * For reads (pageins) and synchronous writes, we clean up | |
07a039d6 | 435 | * all completed async pageouts. |
619edcce KM |
436 | */ |
437 | if ((flags & B_ASYNC) == 0) { | |
438 | s = splbio(); | |
07a039d6 MH |
439 | #ifdef DEBUG |
440 | /* | |
441 | * Check to see if this page is currently being cleaned. | |
442 | * If it is, we just wait til the operation is done before | |
443 | * continuing. | |
444 | */ | |
619edcce | 445 | while (swap_pager_clean(m, flags&B_READ)) { |
07a039d6 MH |
446 | if (swpagerdebug & SDB_ANOM) |
447 | printf("swap_pager_io: page %x cleaning\n", m); | |
448 | ||
619edcce KM |
449 | swp->sw_flags |= SW_WANTED; |
450 | assert_wait((int)swp); | |
451 | thread_block(); | |
452 | } | |
07a039d6 MH |
453 | #else |
454 | (void) swap_pager_clean(m, flags&B_READ); | |
455 | #endif | |
619edcce KM |
456 | splx(s); |
457 | } | |
458 | /* | |
459 | * For async writes (pageouts), we cleanup completed pageouts so | |
460 | * that all available resources are freed. Also tells us if this | |
461 | * page is already being cleaned. If it is, or no resources | |
462 | * are available, we try again later. | |
463 | */ | |
07a039d6 MH |
464 | else if (swap_pager_clean(m, B_WRITE) || |
465 | queue_empty(&swap_pager_free)) { | |
466 | #ifdef DEBUG | |
467 | if ((swpagerdebug & SDB_ANOM) && | |
468 | !queue_empty(&swap_pager_free)) | |
469 | printf("swap_pager_io: page %x already cleaning\n", m); | |
470 | #endif | |
619edcce | 471 | return(VM_PAGER_FAIL); |
07a039d6 | 472 | } |
619edcce KM |
473 | |
474 | /* | |
475 | * Determine swap block and allocate as necessary. | |
476 | */ | |
477 | off = m->offset + m->object->paging_offset; | |
478 | ix = off / dbtob(swp->sw_bsize); | |
479 | if (swp->sw_blocks == NULL || ix >= swp->sw_nblocks) { | |
480 | #ifdef DEBUG | |
481 | if (swpagerdebug & SDB_FAIL) | |
482 | printf("swpg_io: bad offset %x+%x(%d) in %x\n", | |
483 | m->offset, m->object->paging_offset, | |
484 | ix, swp->sw_blocks); | |
485 | #endif | |
486 | return(VM_PAGER_FAIL); | |
487 | } | |
488 | swb = &swp->sw_blocks[ix]; | |
489 | off = off % dbtob(swp->sw_bsize); | |
490 | if (flags & B_READ) { | |
491 | if (swb->swb_block == 0 || | |
492 | (swb->swb_mask & (1 << atop(off))) == 0) { | |
493 | #ifdef DEBUG | |
494 | if (swpagerdebug & (SDB_ALLOCBLK|SDB_FAIL)) | |
495 | printf("swpg_io: %x bad read: blk %x+%x, mask %x, off %x+%x\n", | |
496 | swp->sw_blocks, | |
497 | swb->swb_block, atop(off), | |
498 | swb->swb_mask, | |
499 | m->offset, m->object->paging_offset); | |
500 | #endif | |
501 | /* XXX: should we zero page here?? */ | |
502 | return(VM_PAGER_FAIL); | |
503 | } | |
504 | } else if (swb->swb_block == 0) { | |
505 | swb->swb_block = rmalloc(swapmap, swp->sw_bsize); | |
506 | if (swb->swb_block == 0) { | |
507 | #ifdef DEBUG | |
508 | if (swpagerdebug & SDB_FAIL) | |
509 | printf("swpg_io: rmalloc of %x failed\n", | |
510 | swp->sw_bsize); | |
511 | #endif | |
512 | return(VM_PAGER_FAIL); | |
513 | } | |
514 | #ifdef DEBUG | |
515 | if (swpagerdebug & (SDB_FULL|SDB_ALLOCBLK)) | |
516 | printf("swpg_io: %x alloc blk %x at ix %x\n", | |
517 | swp->sw_blocks, swb->swb_block, ix); | |
518 | #endif | |
519 | } | |
520 | ||
521 | /* | |
522 | * Allocate a kernel virtual address and initialize so that PTE | |
523 | * is available for lower level IO drivers. | |
524 | */ | |
525 | kva = vm_pager_map_page(m); | |
526 | ||
527 | /* | |
528 | * Get a swap buffer header and perform the IO | |
529 | */ | |
530 | s = splbio(); | |
531 | while (bswlist.av_forw == NULL) { | |
532 | #ifdef DEBUG | |
533 | if (swpagerdebug & SDB_ANOM) | |
07a039d6 | 534 | printf("swap_pager_io: wait on swbuf for %x (%d)\n", |
619edcce KM |
535 | m, flags); |
536 | #endif | |
537 | bswlist.b_flags |= B_WANTED; | |
538 | sleep((caddr_t)&bswlist, PSWP+1); | |
539 | } | |
540 | bp = bswlist.av_forw; | |
541 | bswlist.av_forw = bp->av_forw; | |
542 | splx(s); | |
543 | bp->b_flags = B_BUSY | (flags & B_READ); | |
ffe0d082 | 544 | bp->b_proc = &proc0; /* XXX (but without B_PHYS set this is ok) */ |
619edcce KM |
545 | bp->b_un.b_addr = (caddr_t)kva; |
546 | bp->b_blkno = swb->swb_block + btodb(off); | |
547 | VHOLD(swapdev_vp); | |
548 | bp->b_vp = swapdev_vp; | |
e950b33e KM |
549 | if (swapdev_vp->v_type == VBLK) |
550 | bp->b_dev = swapdev_vp->v_rdev; | |
619edcce KM |
551 | bp->b_bcount = PAGE_SIZE; |
552 | if ((bp->b_flags & B_READ) == 0) | |
553 | swapdev_vp->v_numoutput++; | |
554 | ||
555 | /* | |
556 | * If this is an async write we set up additional buffer fields | |
557 | * and place a "cleaning" entry on the inuse queue. | |
558 | */ | |
559 | if ((flags & (B_READ|B_ASYNC)) == B_ASYNC) { | |
560 | #ifdef DEBUG | |
561 | if (queue_empty(&swap_pager_free)) | |
562 | panic("swpg_io: lost spc"); | |
563 | #endif | |
564 | queue_remove_first(&swap_pager_free, | |
565 | spc, swp_clean_t, spc_list); | |
566 | #ifdef DEBUG | |
567 | if (spc->spc_flags != SPC_FREE) | |
568 | panic("swpg_io: bad free spc"); | |
569 | #endif | |
570 | spc->spc_flags = SPC_BUSY; | |
571 | spc->spc_bp = bp; | |
572 | spc->spc_swp = swp; | |
573 | spc->spc_kva = kva; | |
574 | spc->spc_m = m; | |
619edcce KM |
575 | bp->b_flags |= B_CALL; |
576 | bp->b_iodone = swap_pager_iodone; | |
577 | s = splbio(); | |
578 | swp->sw_poip++; | |
579 | queue_enter(&swap_pager_inuse, spc, swp_clean_t, spc_list); | |
580 | ||
581 | #ifdef DEBUG | |
582 | swap_pager_poip++; | |
583 | if (swpagerdebug & SDB_WRITE) | |
584 | printf("swpg_io: write: bp=%x swp=%x spc=%x poip=%d\n", | |
585 | bp, swp, spc, swp->sw_poip); | |
586 | if ((swpagerdebug & SDB_ALLOCBLK) && | |
587 | (swb->swb_mask & (1 << atop(off))) == 0) | |
588 | printf("swpg_io: %x write blk %x+%x\n", | |
589 | swp->sw_blocks, swb->swb_block, atop(off)); | |
590 | #endif | |
591 | swb->swb_mask |= (1 << atop(off)); | |
619edcce KM |
592 | splx(s); |
593 | } | |
594 | #ifdef DEBUG | |
595 | if (swpagerdebug & SDB_IO) | |
596 | printf("swpg_io: IO start: bp %x, db %x, va %x, pa %x\n", | |
597 | bp, swb->swb_block+btodb(off), kva, VM_PAGE_TO_PHYS(m)); | |
598 | #endif | |
599 | VOP_STRATEGY(bp); | |
600 | if ((flags & (B_READ|B_ASYNC)) == B_ASYNC) { | |
601 | #ifdef DEBUG | |
602 | if (swpagerdebug & SDB_IO) | |
603 | printf("swpg_io: IO started: bp %x\n", bp); | |
604 | #endif | |
605 | return(VM_PAGER_PEND); | |
606 | } | |
607 | s = splbio(); | |
608 | #ifdef DEBUG | |
609 | if (flags & B_READ) | |
610 | swap_pager_piip++; | |
611 | else | |
612 | swap_pager_poip++; | |
613 | #endif | |
614 | while ((bp->b_flags & B_DONE) == 0) { | |
615 | assert_wait((int)bp); | |
616 | thread_block(); | |
617 | } | |
618 | #ifdef DEBUG | |
619 | if (flags & B_READ) | |
620 | --swap_pager_piip; | |
621 | else | |
622 | --swap_pager_poip; | |
623 | #endif | |
624 | rv = (bp->b_flags & B_ERROR) ? VM_PAGER_FAIL : VM_PAGER_OK; | |
625 | bp->b_flags &= ~(B_BUSY|B_WANTED|B_PHYS|B_PAGET|B_UAREA|B_DIRTY); | |
626 | bp->av_forw = bswlist.av_forw; | |
627 | bswlist.av_forw = bp; | |
628 | if (bp->b_vp) | |
629 | brelvp(bp); | |
630 | if (bswlist.b_flags & B_WANTED) { | |
631 | bswlist.b_flags &= ~B_WANTED; | |
632 | thread_wakeup((int)&bswlist); | |
633 | } | |
634 | if ((flags & B_READ) == 0 && rv == VM_PAGER_OK) { | |
07a039d6 | 635 | m->clean = TRUE; |
619edcce KM |
636 | pmap_clear_modify(VM_PAGE_TO_PHYS(m)); |
637 | } | |
638 | splx(s); | |
639 | #ifdef DEBUG | |
640 | if (swpagerdebug & SDB_IO) | |
641 | printf("swpg_io: IO done: bp %x, rv %d\n", bp, rv); | |
642 | if ((swpagerdebug & SDB_FAIL) && rv == VM_PAGER_FAIL) | |
643 | printf("swpg_io: IO error\n"); | |
644 | #endif | |
645 | vm_pager_unmap_page(kva); | |
646 | return(rv); | |
647 | } | |
648 | ||
649 | boolean_t | |
650 | swap_pager_clean(m, rw) | |
651 | vm_page_t m; | |
652 | int rw; | |
653 | { | |
654 | register swp_clean_t spc, tspc; | |
655 | register int s; | |
656 | ||
657 | #ifdef DEBUG | |
658 | /* save panic time state */ | |
659 | if ((swpagerdebug & SDB_ANOMPANIC) && panicstr) | |
660 | return; | |
661 | if (swpagerdebug & SDB_FOLLOW) | |
662 | printf("swpg_clean(%x, %d)\n", m, rw); | |
663 | #endif | |
ffe0d082 | 664 | tspc = NULL; |
619edcce KM |
665 | for (;;) { |
666 | /* | |
667 | * Look up and removal from inuse list must be done | |
668 | * at splbio() to avoid conflicts with swap_pager_iodone. | |
669 | */ | |
670 | s = splbio(); | |
671 | spc = (swp_clean_t) queue_first(&swap_pager_inuse); | |
672 | while (!queue_end(&swap_pager_inuse, (queue_entry_t)spc)) { | |
673 | if ((spc->spc_flags & SPC_DONE) && | |
674 | swap_pager_finish(spc)) { | |
675 | queue_remove(&swap_pager_inuse, spc, | |
676 | swp_clean_t, spc_list); | |
677 | break; | |
678 | } | |
679 | if (m && m == spc->spc_m) { | |
680 | #ifdef DEBUG | |
681 | if (swpagerdebug & SDB_ANOM) | |
07a039d6 | 682 | printf("swap_pager_clean: page %x on list, flags %x\n", |
619edcce KM |
683 | m, spc->spc_flags); |
684 | #endif | |
685 | tspc = spc; | |
686 | } | |
687 | spc = (swp_clean_t) queue_next(&spc->spc_list); | |
688 | } | |
689 | ||
690 | /* | |
691 | * No operations done, thats all we can do for now. | |
692 | */ | |
693 | if (queue_end(&swap_pager_inuse, (queue_entry_t)spc)) | |
694 | break; | |
695 | splx(s); | |
696 | ||
697 | /* | |
698 | * The desired page was found to be busy earlier in | |
699 | * the scan but has since completed. | |
700 | */ | |
701 | if (tspc && tspc == spc) { | |
702 | #ifdef DEBUG | |
703 | if (swpagerdebug & SDB_ANOM) | |
07a039d6 | 704 | printf("swap_pager_clean: page %x done while looking\n", |
619edcce KM |
705 | m); |
706 | #endif | |
ffe0d082 | 707 | tspc = NULL; |
619edcce KM |
708 | } |
709 | spc->spc_flags = SPC_FREE; | |
710 | vm_pager_unmap_page(spc->spc_kva); | |
711 | queue_enter(&swap_pager_free, spc, swp_clean_t, spc_list); | |
712 | #ifdef DEBUG | |
713 | if (swpagerdebug & SDB_WRITE) | |
714 | printf("swpg_clean: free spc %x\n", spc); | |
715 | #endif | |
716 | } | |
07a039d6 | 717 | #ifdef DEBUG |
619edcce KM |
718 | /* |
719 | * If we found that the desired page is already being cleaned | |
720 | * mark it so that swap_pager_iodone() will not set the clean | |
721 | * flag before the pageout daemon has another chance to clean it. | |
722 | */ | |
723 | if (tspc && rw == B_WRITE) { | |
619edcce | 724 | if (swpagerdebug & SDB_ANOM) |
07a039d6 MH |
725 | printf("swap_pager_clean: page %x on clean list\n", |
726 | tspc); | |
619edcce KM |
727 | tspc->spc_flags |= SPC_DIRTY; |
728 | } | |
07a039d6 | 729 | #endif |
619edcce KM |
730 | splx(s); |
731 | ||
732 | #ifdef DEBUG | |
733 | if (swpagerdebug & SDB_WRITE) | |
734 | printf("swpg_clean: return %d\n", tspc ? TRUE : FALSE); | |
735 | if ((swpagerdebug & SDB_ANOM) && tspc) | |
736 | printf("swpg_clean: %s of cleaning page %x\n", | |
737 | rw == B_READ ? "get" : "put", m); | |
738 | #endif | |
739 | return(tspc ? TRUE : FALSE); | |
740 | } | |
741 | ||
742 | swap_pager_finish(spc) | |
743 | register swp_clean_t spc; | |
744 | { | |
745 | vm_object_t object = spc->spc_m->object; | |
746 | ||
747 | /* | |
748 | * Mark the paging operation as done. | |
749 | * (XXX) If we cannot get the lock, leave it til later. | |
750 | * (XXX) Also we are assuming that an async write is a | |
751 | * pageout operation that has incremented the counter. | |
752 | */ | |
753 | if (!vm_object_lock_try(object)) | |
754 | return(0); | |
755 | ||
619edcce KM |
756 | if (--object->paging_in_progress == 0) |
757 | thread_wakeup((int) object); | |
758 | ||
07a039d6 | 759 | #ifdef DEBUG |
619edcce KM |
760 | /* |
761 | * XXX: this isn't even close to the right thing to do, | |
762 | * introduces a variety of race conditions. | |
763 | * | |
764 | * If dirty, vm_pageout() has attempted to clean the page | |
765 | * again. In this case we do not do anything as we will | |
07a039d6 | 766 | * see the page again shortly. |
619edcce | 767 | */ |
07a039d6 MH |
768 | if (spc->spc_flags & SPC_DIRTY) { |
769 | if (swpagerdebug & SDB_ANOM) | |
770 | printf("swap_pager_finish: page %x dirty again\n", | |
771 | spc->spc_m); | |
772 | spc->spc_m->busy = FALSE; | |
773 | PAGE_WAKEUP(spc->spc_m); | |
774 | vm_object_unlock(object); | |
775 | return(1); | |
619edcce | 776 | } |
07a039d6 | 777 | #endif |
619edcce | 778 | /* |
07a039d6 MH |
779 | * If no error mark as clean and inform the pmap system. |
780 | * If error, mark as dirty so we will try again. | |
781 | * (XXX could get stuck doing this, should give up after awhile) | |
619edcce | 782 | */ |
07a039d6 MH |
783 | if (spc->spc_flags & SPC_ERROR) { |
784 | printf("swap_pager_finish: clean of page %x failed\n", | |
785 | VM_PAGE_TO_PHYS(spc->spc_m)); | |
786 | spc->spc_m->laundry = TRUE; | |
787 | } else { | |
788 | spc->spc_m->clean = TRUE; | |
789 | pmap_clear_modify(VM_PAGE_TO_PHYS(spc->spc_m)); | |
790 | } | |
791 | spc->spc_m->busy = FALSE; | |
619edcce KM |
792 | PAGE_WAKEUP(spc->spc_m); |
793 | ||
794 | vm_object_unlock(object); | |
795 | return(1); | |
796 | } | |
797 | ||
798 | swap_pager_iodone(bp) | |
799 | register struct buf *bp; | |
800 | { | |
801 | register swp_clean_t spc; | |
802 | daddr_t blk; | |
803 | int s; | |
804 | ||
805 | #ifdef DEBUG | |
806 | /* save panic time state */ | |
807 | if ((swpagerdebug & SDB_ANOMPANIC) && panicstr) | |
808 | return; | |
809 | if (swpagerdebug & SDB_FOLLOW) | |
810 | printf("swpg_iodone(%x)\n", bp); | |
811 | #endif | |
812 | s = splbio(); | |
813 | spc = (swp_clean_t) queue_first(&swap_pager_inuse); | |
814 | while (!queue_end(&swap_pager_inuse, (queue_entry_t)spc)) { | |
815 | if (spc->spc_bp == bp) | |
816 | break; | |
817 | spc = (swp_clean_t) queue_next(&spc->spc_list); | |
818 | } | |
819 | #ifdef DEBUG | |
820 | if (queue_end(&swap_pager_inuse, (queue_entry_t)spc)) | |
07a039d6 | 821 | panic("swap_pager_iodone: bp not found"); |
619edcce KM |
822 | #endif |
823 | ||
824 | spc->spc_flags &= ~SPC_BUSY; | |
825 | spc->spc_flags |= SPC_DONE; | |
826 | if (bp->b_flags & B_ERROR) | |
827 | spc->spc_flags |= SPC_ERROR; | |
828 | spc->spc_bp = NULL; | |
829 | blk = bp->b_blkno; | |
830 | ||
831 | #ifdef DEBUG | |
832 | --swap_pager_poip; | |
833 | if (swpagerdebug & SDB_WRITE) | |
834 | printf("swpg_iodone: bp=%x swp=%x flags=%x spc=%x poip=%x\n", | |
835 | bp, spc->spc_swp, spc->spc_swp->sw_flags, | |
836 | spc, spc->spc_swp->sw_poip); | |
837 | #endif | |
838 | ||
839 | spc->spc_swp->sw_poip--; | |
840 | if (spc->spc_swp->sw_flags & SW_WANTED) { | |
841 | spc->spc_swp->sw_flags &= ~SW_WANTED; | |
842 | thread_wakeup((int)spc->spc_swp); | |
843 | } | |
844 | ||
845 | bp->b_flags &= ~(B_BUSY|B_WANTED|B_PHYS|B_PAGET|B_UAREA|B_DIRTY); | |
846 | bp->av_forw = bswlist.av_forw; | |
847 | bswlist.av_forw = bp; | |
848 | if (bp->b_vp) | |
849 | brelvp(bp); | |
850 | if (bswlist.b_flags & B_WANTED) { | |
851 | bswlist.b_flags &= ~B_WANTED; | |
852 | thread_wakeup((int)&bswlist); | |
853 | } | |
619edcce KM |
854 | thread_wakeup((int) &vm_pages_needed); |
855 | splx(s); | |
856 | } | |
857 | #endif |