1 /* $NetBSD: uvm_device.c,v 1.42 2004/03/24 07:55:01 junyoung Exp $ */ 2 3 /* 4 * 5 * Copyright (c) 1997 Charles D. Cranor and Washington University. 6 * All rights reserved. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 3. All advertising materials mentioning features or use of this software 17 * must display the following acknowledgement: 18 * This product includes software developed by Charles D. Cranor and 19 * Washington University. 20 * 4. The name of the author may not be used to endorse or promote products 21 * derived from this software without specific prior written permission. 22 * 23 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 24 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 25 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 26 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, 27 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 28 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 29 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 30 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 31 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 32 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 33 * 34 * from: Id: uvm_device.c,v 1.1.2.9 1998/02/06 05:11:47 chs Exp 35 */ 36 37 /* 38 * uvm_device.c: the device pager. 39 */ 40 41 #include <sys/cdefs.h> 42 __KERNEL_RCSID(0, "$NetBSD: uvm_device.c,v 1.42 2004/03/24 07:55:01 junyoung Exp $"); 43 44 #include "opt_uvmhist.h" 45 46 #include <sys/param.h> 47 #include <sys/systm.h> 48 #include <sys/conf.h> 49 #include <sys/proc.h> 50 #include <sys/malloc.h> 51 #include <sys/vnode.h> 52 53 #include <uvm/uvm.h> 54 #include <uvm/uvm_device.h> 55 56 /* 57 * private global data structure 58 * 59 * we keep a list of active device objects in the system. 60 */ 61 62 LIST_HEAD(udv_list_struct, uvm_device); 63 static struct udv_list_struct udv_list; 64 static struct simplelock udv_lock; 65 66 /* 67 * functions 68 */ 69 70 static void udv_init(void); 71 static void udv_reference(struct uvm_object *); 72 static void udv_detach(struct uvm_object *); 73 static int udv_fault(struct uvm_faultinfo *, vaddr_t, 74 struct vm_page **, int, int, vm_fault_t, vm_prot_t, int); 75 76 /* 77 * master pager structure 78 */ 79 80 struct uvm_pagerops uvm_deviceops = { 81 udv_init, 82 udv_reference, 83 udv_detach, 84 udv_fault, 85 }; 86 87 /* 88 * the ops! 89 */ 90 91 /* 92 * udv_init 93 * 94 * init pager private data structures. 95 */ 96 97 static void 98 udv_init(void) 99 { 100 LIST_INIT(&udv_list); 101 simple_lock_init(&udv_lock); 102 } 103 104 /* 105 * udv_attach 106 * 107 * get a VM object that is associated with a device. allocate a new 108 * one if needed. 109 * 110 * => caller must _not_ already be holding the lock on the uvm_object. 111 * => in fact, nothing should be locked so that we can sleep here. 112 */ 113 114 struct uvm_object * 115 udv_attach(arg, accessprot, off, size) 116 void *arg; 117 vm_prot_t accessprot; 118 voff_t off; /* used only for access check */ 119 vsize_t size; /* used only for access check */ 120 { 121 dev_t device = *((dev_t *)arg); 122 struct uvm_device *udv, *lcv; 123 const struct cdevsw *cdev; 124 dev_type_mmap((*mapfn)); 125 126 UVMHIST_FUNC("udv_attach"); UVMHIST_CALLED(maphist); 127 128 UVMHIST_LOG(maphist, "(device=0x%x)", device,0,0,0); 129 130 /* 131 * before we do anything, ensure this device supports mmap 132 */ 133 134 cdev = cdevsw_lookup(device); 135 if (cdev == NULL) 136 return (NULL); 137 mapfn = cdev->d_mmap; 138 if (mapfn == NULL || mapfn == nommap || mapfn == nullmmap) 139 return(NULL); 140 141 /* 142 * Negative offsets on the object are not allowed. 143 */ 144 145 if (off < 0) 146 return(NULL); 147 148 /* 149 * Check that the specified range of the device allows the 150 * desired protection. 151 * 152 * XXX assumes VM_PROT_* == PROT_* 153 * XXX clobbers off and size, but nothing else here needs them. 154 */ 155 156 while (size != 0) { 157 if ((*mapfn)(device, off, accessprot) == -1) 158 return (NULL); 159 off += PAGE_SIZE; size -= PAGE_SIZE; 160 } 161 162 /* 163 * keep looping until we get it 164 */ 165 166 for (;;) { 167 168 /* 169 * first, attempt to find it on the main list 170 */ 171 172 simple_lock(&udv_lock); 173 LIST_FOREACH(lcv, &udv_list, u_list) { 174 if (device == lcv->u_device) 175 break; 176 } 177 178 /* 179 * got it on main list. put a hold on it and unlock udv_lock. 180 */ 181 182 if (lcv) { 183 184 /* 185 * if someone else has a hold on it, sleep and start 186 * over again. 187 */ 188 189 if (lcv->u_flags & UVM_DEVICE_HOLD) { 190 lcv->u_flags |= UVM_DEVICE_WANTED; 191 UVM_UNLOCK_AND_WAIT(lcv, &udv_lock, FALSE, 192 "udv_attach",0); 193 continue; 194 } 195 196 /* we are now holding it */ 197 lcv->u_flags |= UVM_DEVICE_HOLD; 198 simple_unlock(&udv_lock); 199 200 /* 201 * bump reference count, unhold, return. 202 */ 203 204 simple_lock(&lcv->u_obj.vmobjlock); 205 lcv->u_obj.uo_refs++; 206 simple_unlock(&lcv->u_obj.vmobjlock); 207 208 simple_lock(&udv_lock); 209 if (lcv->u_flags & UVM_DEVICE_WANTED) 210 wakeup(lcv); 211 lcv->u_flags &= ~(UVM_DEVICE_WANTED|UVM_DEVICE_HOLD); 212 simple_unlock(&udv_lock); 213 return(&lcv->u_obj); 214 } 215 216 /* 217 * did not find it on main list. need to malloc a new one. 218 */ 219 220 simple_unlock(&udv_lock); 221 /* NOTE: we could sleep in the following malloc() */ 222 MALLOC(udv, struct uvm_device *, sizeof(*udv), M_TEMP, 223 M_WAITOK); 224 simple_lock(&udv_lock); 225 226 /* 227 * now we have to double check to make sure no one added it 228 * to the list while we were sleeping... 229 */ 230 231 LIST_FOREACH(lcv, &udv_list, u_list) { 232 if (device == lcv->u_device) 233 break; 234 } 235 236 /* 237 * did we lose a race to someone else? 238 * free our memory and retry. 239 */ 240 241 if (lcv) { 242 simple_unlock(&udv_lock); 243 FREE(udv, M_TEMP); 244 continue; 245 } 246 247 /* 248 * we have it! init the data structures, add to list 249 * and return. 250 */ 251 252 simple_lock_init(&udv->u_obj.vmobjlock); 253 udv->u_obj.pgops = &uvm_deviceops; 254 TAILQ_INIT(&udv->u_obj.memq); 255 udv->u_obj.uo_npages = 0; 256 udv->u_obj.uo_refs = 1; 257 udv->u_flags = 0; 258 udv->u_device = device; 259 LIST_INSERT_HEAD(&udv_list, udv, u_list); 260 simple_unlock(&udv_lock); 261 return(&udv->u_obj); 262 } 263 /*NOTREACHED*/ 264 } 265 266 /* 267 * udv_reference 268 * 269 * add a reference to a VM object. Note that the reference count must 270 * already be one (the passed in reference) so there is no chance of the 271 * udv being released or locked out here. 272 * 273 * => caller must call with object unlocked. 274 */ 275 276 static void 277 udv_reference(uobj) 278 struct uvm_object *uobj; 279 { 280 UVMHIST_FUNC("udv_reference"); UVMHIST_CALLED(maphist); 281 282 simple_lock(&uobj->vmobjlock); 283 uobj->uo_refs++; 284 UVMHIST_LOG(maphist, "<- done (uobj=0x%x, ref = %d)", 285 uobj, uobj->uo_refs,0,0); 286 simple_unlock(&uobj->vmobjlock); 287 } 288 289 /* 290 * udv_detach 291 * 292 * remove a reference to a VM object. 293 * 294 * => caller must call with object unlocked and map locked. 295 */ 296 297 static void 298 udv_detach(uobj) 299 struct uvm_object *uobj; 300 { 301 struct uvm_device *udv = (struct uvm_device *)uobj; 302 UVMHIST_FUNC("udv_detach"); UVMHIST_CALLED(maphist); 303 304 /* 305 * loop until done 306 */ 307 again: 308 simple_lock(&uobj->vmobjlock); 309 if (uobj->uo_refs > 1) { 310 uobj->uo_refs--; 311 simple_unlock(&uobj->vmobjlock); 312 UVMHIST_LOG(maphist," <- done, uobj=0x%x, ref=%d", 313 uobj,uobj->uo_refs,0,0); 314 return; 315 } 316 317 /* 318 * is it being held? if so, wait until others are done. 319 */ 320 321 simple_lock(&udv_lock); 322 if (udv->u_flags & UVM_DEVICE_HOLD) { 323 udv->u_flags |= UVM_DEVICE_WANTED; 324 simple_unlock(&uobj->vmobjlock); 325 UVM_UNLOCK_AND_WAIT(udv, &udv_lock, FALSE, "udv_detach",0); 326 goto again; 327 } 328 329 /* 330 * got it! nuke it now. 331 */ 332 333 LIST_REMOVE(udv, u_list); 334 if (udv->u_flags & UVM_DEVICE_WANTED) 335 wakeup(udv); 336 simple_unlock(&udv_lock); 337 simple_unlock(&uobj->vmobjlock); 338 FREE(udv, M_TEMP); 339 UVMHIST_LOG(maphist," <- done, freed uobj=0x%x", uobj,0,0,0); 340 } 341 342 /* 343 * udv_fault: non-standard fault routine for device "pages" 344 * 345 * => rather than having a "get" function, we have a fault routine 346 * since we don't return vm_pages we need full control over the 347 * pmap_enter map in 348 * => all the usual fault data structured are locked by the caller 349 * (i.e. maps(read), amap (if any), uobj) 350 * => on return, we unlock all fault data structures 351 * => flags: PGO_ALLPAGES: get all of the pages 352 * PGO_LOCKED: fault data structures are locked 353 * XXX: currently PGO_LOCKED is always required ... consider removing 354 * it as a flag 355 * => NOTE: vaddr is the VA of pps[0] in ufi->entry, _NOT_ pps[centeridx] 356 */ 357 358 static int 359 udv_fault(ufi, vaddr, pps, npages, centeridx, fault_type, access_type, flags) 360 struct uvm_faultinfo *ufi; 361 vaddr_t vaddr; 362 struct vm_page **pps; 363 int npages, centeridx, flags; 364 vm_fault_t fault_type; 365 vm_prot_t access_type; 366 { 367 struct vm_map_entry *entry = ufi->entry; 368 struct uvm_object *uobj = entry->object.uvm_obj; 369 struct uvm_device *udv = (struct uvm_device *)uobj; 370 const struct cdevsw *cdev; 371 vaddr_t curr_va; 372 off_t curr_offset; 373 paddr_t paddr, mdpgno; 374 int lcv, retval; 375 dev_t device; 376 paddr_t (*mapfn)(dev_t, off_t, int); 377 vm_prot_t mapprot; 378 UVMHIST_FUNC("udv_fault"); UVMHIST_CALLED(maphist); 379 UVMHIST_LOG(maphist," flags=%d", flags,0,0,0); 380 381 /* 382 * we do not allow device mappings to be mapped copy-on-write 383 * so we kill any attempt to do so here. 384 */ 385 386 if (UVM_ET_ISCOPYONWRITE(entry)) { 387 UVMHIST_LOG(maphist, "<- failed -- COW entry (etype=0x%x)", 388 entry->etype, 0,0,0); 389 uvmfault_unlockall(ufi, ufi->entry->aref.ar_amap, uobj, NULL); 390 return(EIO); 391 } 392 393 /* 394 * get device map function. 395 */ 396 397 device = udv->u_device; 398 cdev = cdevsw_lookup(device); 399 if (cdev == NULL) { 400 uvmfault_unlockall(ufi, ufi->entry->aref.ar_amap, uobj, NULL); 401 return (EIO); 402 } 403 mapfn = cdev->d_mmap; 404 405 /* 406 * now we must determine the offset in udv to use and the VA to 407 * use for pmap_enter. note that we always use orig_map's pmap 408 * for pmap_enter (even if we have a submap). since virtual 409 * addresses in a submap must match the main map, this is ok. 410 */ 411 412 /* udv offset = (offset from start of entry) + entry's offset */ 413 curr_offset = entry->offset + (vaddr - entry->start); 414 /* pmap va = vaddr (virtual address of pps[0]) */ 415 curr_va = vaddr; 416 417 /* 418 * loop over the page range entering in as needed 419 */ 420 421 retval = 0; 422 for (lcv = 0 ; lcv < npages ; lcv++, curr_offset += PAGE_SIZE, 423 curr_va += PAGE_SIZE) { 424 if ((flags & PGO_ALLPAGES) == 0 && lcv != centeridx) 425 continue; 426 427 if (pps[lcv] == PGO_DONTCARE) 428 continue; 429 430 mdpgno = (*mapfn)(device, curr_offset, access_type); 431 if (mdpgno == -1) { 432 retval = EIO; 433 break; 434 } 435 paddr = pmap_phys_address(mdpgno); 436 mapprot = ufi->entry->protection; 437 UVMHIST_LOG(maphist, 438 " MAPPING: device: pm=0x%x, va=0x%x, pa=0x%lx, at=%d", 439 ufi->orig_map->pmap, curr_va, paddr, mapprot); 440 if (pmap_enter(ufi->orig_map->pmap, curr_va, paddr, 441 mapprot, PMAP_CANFAIL | mapprot) != 0) { 442 /* 443 * pmap_enter() didn't have the resource to 444 * enter this mapping. Unlock everything, 445 * wait for the pagedaemon to free up some 446 * pages, and then tell uvm_fault() to start 447 * the fault again. 448 * 449 * XXX Needs some rethinking for the PGO_ALLPAGES 450 * XXX case. 451 */ 452 uvmfault_unlockall(ufi, ufi->entry->aref.ar_amap, 453 uobj, NULL); 454 pmap_update(ufi->orig_map->pmap); /* sync what we have so far */ 455 uvm_wait("udv_fault"); 456 return (ERESTART); 457 } 458 } 459 460 uvmfault_unlockall(ufi, ufi->entry->aref.ar_amap, uobj, NULL); 461 pmap_update(ufi->orig_map->pmap); 462 return (retval); 463 } 464