1 /* $NetBSD: uvm_anon.c,v 1.79 2020/07/09 05:57:15 skrll Exp $ */ 2 3 /* 4 * Copyright (c) 1997 Charles D. Cranor and Washington University. 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 16 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 17 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 18 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 19 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, 20 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 21 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 22 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 23 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 24 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 25 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 26 */ 27 28 /* 29 * uvm_anon.c: uvm anon ops 30 */ 31 32 #include <sys/cdefs.h> 33 __KERNEL_RCSID(0, "$NetBSD: uvm_anon.c,v 1.79 2020/07/09 05:57:15 skrll Exp $"); 34 35 #include "opt_uvmhist.h" 36 37 #include <sys/param.h> 38 #include <sys/systm.h> 39 #include <sys/pool.h> 40 #include <sys/kernel.h> 41 #include <sys/atomic.h> 42 43 #include <uvm/uvm.h> 44 #include <uvm/uvm_swap.h> 45 #include <uvm/uvm_pdpolicy.h> 46 47 static struct pool_cache uvm_anon_cache; 48 49 static int uvm_anon_ctor(void *, void *, int); 50 51 void 52 uvm_anon_init(void) 53 { 54 55 pool_cache_bootstrap(&uvm_anon_cache, sizeof(struct vm_anon), 0, 0, 56 PR_LARGECACHE, "anonpl", NULL, IPL_NONE, uvm_anon_ctor, 57 NULL, NULL); 58 } 59 60 static int 61 uvm_anon_ctor(void *arg, void *object, int flags) 62 { 63 struct vm_anon *anon = object; 64 65 anon->an_ref = 0; 66 anon->an_lock = NULL; 67 anon->an_page = NULL; 68 #if defined(VMSWAP) 69 anon->an_swslot = 0; 70 #endif 71 return 0; 72 } 73 74 /* 75 * uvm_analloc: allocate a new anon. 76 * 77 * => anon will have no lock associated. 78 */ 79 struct vm_anon * 80 uvm_analloc(void) 81 { 82 struct vm_anon *anon; 83 84 anon = pool_cache_get(&uvm_anon_cache, PR_NOWAIT); 85 if (anon) { 86 KASSERT(anon->an_ref == 0); 87 KASSERT(anon->an_lock == NULL); 88 KASSERT(anon->an_page == NULL); 89 #if defined(VMSWAP) 90 KASSERT(anon->an_swslot == 0); 91 #endif 92 anon->an_ref = 1; 93 } 94 return anon; 95 } 96 97 /* 98 * uvm_anfree: free a single anon structure 99 * 100 * => anon must be removed from the amap (if anon was in an amap). 101 * => amap must be locked, if anon was owned by amap. 102 * => we may drop and re-acquire the lock here (to break loans). 103 */ 104 void 105 uvm_anfree(struct vm_anon *anon) 106 { 107 struct vm_page *pg = anon->an_page, *pg2 __diagused; 108 109 UVMHIST_FUNC(__func__); 110 UVMHIST_CALLARGS(maphist,"(anon=%#jx)", (uintptr_t)anon, 0,0,0); 111 112 KASSERT(anon->an_lock == NULL || rw_write_held(anon->an_lock)); 113 KASSERT(anon->an_ref == 0); 114 115 /* 116 * Dispose of the page, if it is resident. 117 */ 118 119 if (__predict_true(pg != NULL)) { 120 KASSERT(anon->an_lock != NULL); 121 122 /* 123 * If there is a resident page and it is loaned, then anon 124 * may not own it. Call out to uvm_anon_lockloanpg() to 125 * identify and lock the real owner of the page. 126 */ 127 128 if (__predict_false(pg->loan_count != 0)) { 129 pg2 = uvm_anon_lockloanpg(anon); 130 KASSERT(pg2 == pg); 131 } 132 133 /* 134 * If the page is owned by a UVM object (now locked), 135 * then kill the loan on the page rather than free it, 136 * and release the object lock. 137 */ 138 139 if (__predict_false(pg->uobject != NULL)) { 140 mutex_enter(&pg->interlock); 141 KASSERT(pg->loan_count > 0); 142 pg->loan_count--; 143 pg->uanon = NULL; 144 mutex_exit(&pg->interlock); 145 rw_exit(pg->uobject->vmobjlock); 146 } else { 147 148 /* 149 * If page has no UVM object, then anon is the owner, 150 * and it is already locked. 151 */ 152 153 KASSERT((pg->flags & PG_RELEASED) == 0); 154 pmap_page_protect(pg, VM_PROT_NONE); 155 156 /* 157 * If the page is busy, mark it as PG_RELEASED, so 158 * that uvm_anon_release(9) would release it later. 159 */ 160 161 if (__predict_false((pg->flags & PG_BUSY) != 0)) { 162 pg->flags |= PG_RELEASED; 163 rw_obj_hold(anon->an_lock); 164 return; 165 } 166 uvm_pagefree(pg); 167 UVMHIST_LOG(maphist, "anon %#jx, page %#jx: " 168 "freed now!", (uintptr_t)anon, (uintptr_t)pg, 169 0, 0); 170 } 171 } else { 172 #if defined(VMSWAP) 173 if (anon->an_swslot > 0) { 174 /* This page is no longer only in swap. */ 175 KASSERT(uvmexp.swpgonly > 0); 176 atomic_dec_uint(&uvmexp.swpgonly); 177 } 178 #endif 179 } 180 anon->an_lock = NULL; 181 182 /* 183 * Free any swap resources, leave a page replacement hint. 184 */ 185 186 uvm_anon_dropswap(anon); 187 uvmpdpol_anfree(anon); 188 UVMHIST_LOG(maphist,"<- done!",0,0,0,0); 189 pool_cache_put(&uvm_anon_cache, anon); 190 } 191 192 /* 193 * uvm_anon_lockloanpg: given a locked anon, lock its resident page owner. 194 * 195 * => anon is locked by caller 196 * => on return: anon is locked 197 * if there is a resident page: 198 * if it has a uobject, it is locked by us 199 * if it is ownerless, we take over as owner 200 * we return the resident page (it can change during 201 * this function) 202 * => note that the only time an anon has an ownerless resident page 203 * is if the page was loaned from a uvm_object and the uvm_object 204 * disowned it 205 * => this only needs to be called when you want to do an operation 206 * on an anon's resident page and that page has a non-zero loan 207 * count. 208 */ 209 struct vm_page * 210 uvm_anon_lockloanpg(struct vm_anon *anon) 211 { 212 struct vm_page *pg; 213 krw_t op; 214 215 KASSERT(rw_lock_held(anon->an_lock)); 216 217 /* 218 * loop while we have a resident page that has a non-zero loan count. 219 * if we successfully get our lock, we will "break" the loop. 220 * note that the test for pg->loan_count is not protected -- this 221 * may produce false positive results. note that a false positive 222 * result may cause us to do more work than we need to, but it will 223 * not produce an incorrect result. 224 */ 225 226 while (((pg = anon->an_page) != NULL) && pg->loan_count != 0) { 227 mutex_enter(&pg->interlock); 228 if (pg->uobject) { 229 /* 230 * if we didn't get a lock (try lock failed), then we 231 * toggle our anon lock and try again 232 */ 233 234 if (!rw_tryenter(pg->uobject->vmobjlock, RW_WRITER)) { 235 /* 236 * someone locking the object has a chance to 237 * lock us right now 238 * 239 * XXX Better than yielding but inadequate. 240 */ 241 mutex_exit(&pg->interlock); 242 op = rw_lock_op(anon->an_lock); 243 rw_exit(anon->an_lock); 244 kpause("lkloanpg", false, 1, NULL); 245 rw_enter(anon->an_lock, op); 246 continue; 247 } 248 } 249 250 /* 251 * If page is un-owned i.e. the object dropped its ownership, 252 * then we have to take the ownership. 253 */ 254 255 if (pg->uobject == NULL && (pg->flags & PG_ANON) == 0) { 256 pg->flags |= PG_ANON; 257 pg->loan_count--; 258 } 259 mutex_exit(&pg->interlock); 260 break; 261 } 262 return pg; 263 } 264 265 #if defined(VMSWAP) 266 267 /* 268 * uvm_anon_pagein: fetch an anon's page. 269 * 270 * => anon must be locked, and is unlocked upon return. 271 * => returns true if pagein was aborted due to lack of memory. 272 */ 273 274 bool 275 uvm_anon_pagein(struct vm_amap *amap, struct vm_anon *anon) 276 { 277 struct vm_page *pg; 278 struct uvm_object *uobj; 279 280 KASSERT(rw_write_held(anon->an_lock)); 281 KASSERT(anon->an_lock == amap->am_lock); 282 283 /* 284 * Get the page of the anon. 285 */ 286 287 switch (uvmfault_anonget(NULL, amap, anon)) { 288 case 0: 289 /* Success - we have the page. */ 290 KASSERT(rw_write_held(anon->an_lock)); 291 break; 292 case EIO: 293 case ERESTART: 294 /* 295 * Nothing more to do on errors. ERESTART means that the 296 * anon was freed. 297 */ 298 return false; 299 case ENOLCK: 300 panic("uvm_anon_pagein"); 301 default: 302 return true; 303 } 304 305 /* 306 * Mark the page as dirty and clear its swslot. 307 */ 308 309 pg = anon->an_page; 310 uobj = pg->uobject; 311 if (anon->an_swslot > 0) { 312 uvm_swap_free(anon->an_swslot, 1); 313 } 314 anon->an_swslot = 0; 315 uvm_pagemarkdirty(pg, UVM_PAGE_STATUS_DIRTY); 316 317 /* 318 * Deactivate the page (to put it on a page queue). 319 */ 320 321 uvm_pagelock(pg); 322 uvm_pagedeactivate(pg); 323 uvm_pageunlock(pg); 324 rw_exit(anon->an_lock); 325 if (uobj) { 326 rw_exit(uobj->vmobjlock); 327 } 328 return false; 329 } 330 331 /* 332 * uvm_anon_dropswap: release any swap resources from this anon. 333 * 334 * => anon must be locked or have a reference count of 0. 335 */ 336 void 337 uvm_anon_dropswap(struct vm_anon *anon) 338 { 339 UVMHIST_FUNC(__func__); UVMHIST_CALLED(maphist); 340 341 if (anon->an_swslot == 0) 342 return; 343 344 UVMHIST_LOG(maphist,"freeing swap for anon %#jx, paged to swslot %#jx", 345 (uintptr_t)anon, anon->an_swslot, 0, 0); 346 uvm_swap_free(anon->an_swslot, 1); 347 anon->an_swslot = 0; 348 } 349 350 #endif 351 352 /* 353 * uvm_anon_release: release an anon and its page. 354 * 355 * => anon should not have any references. 356 * => anon must be locked. 357 */ 358 359 void 360 uvm_anon_release(struct vm_anon *anon) 361 { 362 struct vm_page *pg = anon->an_page; 363 krwlock_t *lock; 364 365 KASSERT(rw_write_held(anon->an_lock)); 366 KASSERT(pg != NULL); 367 KASSERT((pg->flags & PG_RELEASED) != 0); 368 KASSERT((pg->flags & PG_BUSY) != 0); 369 KASSERT(pg->uobject == NULL); 370 KASSERT(pg->uanon == anon); 371 KASSERT(pg->loan_count == 0); 372 KASSERT(anon->an_ref == 0); 373 374 uvm_pagefree(pg); 375 KASSERT(anon->an_page == NULL); 376 lock = anon->an_lock; 377 uvm_anfree(anon); 378 rw_exit(lock); 379 /* Note: extra reference is held for PG_RELEASED case. */ 380 rw_obj_free(lock); 381 } 382