1/* $NetBSD: uvm_device.c,v 1.64 2014/12/14 23:48:58 chs 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 * from: Id: uvm_device.c,v 1.1.2.9 1998/02/06 05:11:47 chs Exp
28 */
29
30/*
31 * uvm_device.c: the device pager.
32 */
33
34#include <sys/cdefs.h>
35__KERNEL_RCSID(0, "$NetBSD: uvm_device.c,v 1.64 2014/12/14 23:48:58 chs Exp $");
36
37#include "opt_uvmhist.h"
38
39#include <sys/param.h>
40#include <sys/systm.h>
41#include <sys/conf.h>
42#include <sys/proc.h>
43#include <sys/kmem.h>
44
45#include <uvm/uvm.h>
46#include <uvm/uvm_device.h>
47#include <uvm/uvm_pmap.h>
48
49/*
50 * private global data structure
51 *
52 * we keep a list of active device objects in the system.
53 */
54
55LIST_HEAD(udv_list_struct, uvm_device);
56static struct udv_list_struct udv_list;
57static kmutex_t udv_lock;
58
59/*
60 * functions
61 */
62
63static void udv_init(void);
64static void udv_reference(struct uvm_object *);
65static void udv_detach(struct uvm_object *);
66static int udv_fault(struct uvm_faultinfo *, vaddr_t,
67 struct vm_page **, int, int, vm_prot_t,
68 int);
69
70/*
71 * master pager structure
72 */
73
74const struct uvm_pagerops uvm_deviceops = {
75 .pgo_init = udv_init,
76 .pgo_reference = udv_reference,
77 .pgo_detach = udv_detach,
78 .pgo_fault = udv_fault,
79};
80
81/*
82 * the ops!
83 */
84
85/*
86 * udv_init
87 *
88 * init pager private data structures.
89 */
90
91static void
92udv_init(void)
93{
94 LIST_INIT(&udv_list);
95 mutex_init(&udv_lock, MUTEX_DEFAULT, IPL_NONE);
96}
97
98/*
99 * udv_attach
100 *
101 * get a VM object that is associated with a device. allocate a new
102 * one if needed.
103 *
104 * => caller must _not_ already be holding the lock on the uvm_object.
105 * => in fact, nothing should be locked so that we can sleep here.
106 */
107
108struct uvm_object *
109udv_attach(dev_t device, vm_prot_t accessprot,
110 voff_t off, /* used only for access check */
111 vsize_t size /* used only for access check */)
112{
113 struct uvm_device *udv, *lcv;
114 const struct cdevsw *cdev;
115 dev_type_mmap((*mapfn));
116
117 UVMHIST_FUNC("udv_attach"); UVMHIST_CALLED(maphist);
118
119 UVMHIST_LOG(maphist, "(device=0x%x)", device,0,0,0);
120
121 /*
122 * before we do anything, ensure this device supports mmap
123 */
124
125 cdev = cdevsw_lookup(device);
126 if (cdev == NULL) {
127 return (NULL);
128 }
129 mapfn = cdev->d_mmap;
130 if (mapfn == NULL || mapfn == nommap || mapfn == nullmmap) {
131 return(NULL);
132 }
133
134 /*
135 * Negative offsets on the object are not allowed.
136 */
137
138 if ((cdev->d_flag & D_NEGOFFSAFE) == 0 &&
139 off != UVM_UNKNOWN_OFFSET && off < 0)
140 return(NULL);
141
142 /*
143 * Check that the specified range of the device allows the
144 * desired protection.
145 *
146 * XXX assumes VM_PROT_* == PROT_*
147 * XXX clobbers off and size, but nothing else here needs them.
148 */
149
150 while (size != 0) {
151 if (cdev_mmap(device, off, accessprot) == -1) {
152 return (NULL);
153 }
154 off += PAGE_SIZE; size -= PAGE_SIZE;
155 }
156
157 /*
158 * keep looping until we get it
159 */
160
161 for (;;) {
162
163 /*
164 * first, attempt to find it on the main list
165 */
166
167 mutex_enter(&udv_lock);
168 LIST_FOREACH(lcv, &udv_list, u_list) {
169 if (device == lcv->u_device)
170 break;
171 }
172
173 /*
174 * got it on main list. put a hold on it and unlock udv_lock.
175 */
176
177 if (lcv) {
178
179 /*
180 * if someone else has a hold on it, sleep and start
181 * over again.
182 */
183
184 if (lcv->u_flags & UVM_DEVICE_HOLD) {
185 lcv->u_flags |= UVM_DEVICE_WANTED;
186 UVM_UNLOCK_AND_WAIT(lcv, &udv_lock, false,
187 "udv_attach",0);
188 continue;
189 }
190
191 /* we are now holding it */
192 lcv->u_flags |= UVM_DEVICE_HOLD;
193 mutex_exit(&udv_lock);
194
195 /*
196 * bump reference count, unhold, return.
197 */
198
199 mutex_enter(lcv->u_obj.vmobjlock);
200 lcv->u_obj.uo_refs++;
201 mutex_exit(lcv->u_obj.vmobjlock);
202
203 mutex_enter(&udv_lock);
204 if (lcv->u_flags & UVM_DEVICE_WANTED)
205 wakeup(lcv);
206 lcv->u_flags &= ~(UVM_DEVICE_WANTED|UVM_DEVICE_HOLD);
207 mutex_exit(&udv_lock);
208 return(&lcv->u_obj);
209 }
210
211 /*
212 * Did not find it on main list. Need to allocate a new one.
213 */
214
215 mutex_exit(&udv_lock);
216
217 /* Note: both calls may allocate memory and sleep. */
218 udv = kmem_alloc(sizeof(*udv), KM_SLEEP);
219 uvm_obj_init(&udv->u_obj, &uvm_deviceops, true, 1);
220
221 mutex_enter(&udv_lock);
222
223 /*
224 * now we have to double check to make sure no one added it
225 * to the list while we were sleeping...
226 */
227
228 LIST_FOREACH(lcv, &udv_list, u_list) {
229 if (device == lcv->u_device)
230 break;
231 }
232
233 /*
234 * did we lose a race to someone else?
235 * free our memory and retry.
236 */
237
238 if (lcv) {
239 mutex_exit(&udv_lock);
240 uvm_obj_destroy(&udv->u_obj, true);
241 kmem_free(udv, sizeof(*udv));
242 continue;
243 }
244
245 /*
246 * we have it! init the data structures, add to list
247 * and return.
248 */
249
250 udv->u_flags = 0;
251 udv->u_device = device;
252 LIST_INSERT_HEAD(&udv_list, udv, u_list);
253 mutex_exit(&udv_lock);
254 return(&udv->u_obj);
255 }
256 /*NOTREACHED*/
257}
258
259/*
260 * udv_reference
261 *
262 * add a reference to a VM object. Note that the reference count must
263 * already be one (the passed in reference) so there is no chance of the
264 * udv being released or locked out here.
265 *
266 * => caller must call with object unlocked.
267 */
268
269static void
270udv_reference(struct uvm_object *uobj)
271{
272 UVMHIST_FUNC("udv_reference"); UVMHIST_CALLED(maphist);
273
274 mutex_enter(uobj->vmobjlock);
275 uobj->uo_refs++;
276 UVMHIST_LOG(maphist, "<- done (uobj=0x%x, ref = %d)",
277 uobj, uobj->uo_refs,0,0);
278 mutex_exit(uobj->vmobjlock);
279}
280
281/*
282 * udv_detach
283 *
284 * remove a reference to a VM object.
285 *
286 * => caller must call with object unlocked and map locked.
287 */
288
289static void
290udv_detach(struct uvm_object *uobj)
291{
292 struct uvm_device *udv = (struct uvm_device *)uobj;
293 UVMHIST_FUNC("udv_detach"); UVMHIST_CALLED(maphist);
294
295 /*
296 * loop until done
297 */
298again:
299 mutex_enter(uobj->vmobjlock);
300 if (uobj->uo_refs > 1) {
301 uobj->uo_refs--;
302 mutex_exit(uobj->vmobjlock);
303 UVMHIST_LOG(maphist," <- done, uobj=0x%x, ref=%d",
304 uobj,uobj->uo_refs,0,0);
305 return;
306 }
307
308 /*
309 * is it being held? if so, wait until others are done.
310 */
311
312 mutex_enter(&udv_lock);
313 if (udv->u_flags & UVM_DEVICE_HOLD) {
314 udv->u_flags |= UVM_DEVICE_WANTED;
315 mutex_exit(uobj->vmobjlock);
316 UVM_UNLOCK_AND_WAIT(udv, &udv_lock, false, "udv_detach",0);
317 goto again;
318 }
319
320 /*
321 * got it! nuke it now.
322 */
323
324 LIST_REMOVE(udv, u_list);
325 if (udv->u_flags & UVM_DEVICE_WANTED)
326 wakeup(udv);
327 mutex_exit(&udv_lock);
328 mutex_exit(uobj->vmobjlock);
329
330 uvm_obj_destroy(uobj, true);
331 kmem_free(udv, sizeof(*udv));
332 UVMHIST_LOG(maphist," <- done, freed uobj=0x%x", uobj,0,0,0);
333}
334
335/*
336 * udv_fault: non-standard fault routine for device "pages"
337 *
338 * => rather than having a "get" function, we have a fault routine
339 * since we don't return vm_pages we need full control over the
340 * pmap_enter map in
341 * => all the usual fault data structured are locked by the caller
342 * (i.e. maps(read), amap (if any), uobj)
343 * => on return, we unlock all fault data structures
344 * => flags: PGO_ALLPAGES: get all of the pages
345 * PGO_LOCKED: fault data structures are locked
346 * XXX: currently PGO_LOCKED is always required ... consider removing
347 * it as a flag
348 * => NOTE: vaddr is the VA of pps[0] in ufi->entry, _NOT_ pps[centeridx]
349 */
350
351static int
352udv_fault(struct uvm_faultinfo *ufi, vaddr_t vaddr, struct vm_page **pps,
353 int npages, int centeridx, vm_prot_t access_type,
354 int flags)
355{
356 struct vm_map_entry *entry = ufi->entry;
357 struct uvm_object *uobj = entry->object.uvm_obj;
358 struct uvm_device *udv = (struct uvm_device *)uobj;
359 vaddr_t curr_va;
360 off_t curr_offset;
361 paddr_t paddr, mdpgno;
362 u_int mmapflags;
363 int lcv, retval;
364 dev_t device;
365 vm_prot_t mapprot;
366 UVMHIST_FUNC("udv_fault"); UVMHIST_CALLED(maphist);
367 UVMHIST_LOG(maphist," flags=%d", flags,0,0,0);
368
369 /*
370 * we do not allow device mappings to be mapped copy-on-write
371 * so we kill any attempt to do so here.
372 */
373
374 if (UVM_ET_ISCOPYONWRITE(entry)) {
375 UVMHIST_LOG(maphist, "<- failed -- COW entry (etype=0x%x)",
376 entry->etype, 0,0,0);
377 uvmfault_unlockall(ufi, ufi->entry->aref.ar_amap, uobj);
378 return(EIO);
379 }
380
381 /*
382 * get device map function.
383 */
384
385 device = udv->u_device;
386 if (cdevsw_lookup(device) == NULL) {
387 /* XXX This should not happen */
388 uvmfault_unlockall(ufi, ufi->entry->aref.ar_amap, uobj);
389 return (EIO);
390 }
391
392 /*
393 * now we must determine the offset in udv to use and the VA to
394 * use for pmap_enter. note that we always use orig_map's pmap
395 * for pmap_enter (even if we have a submap). since virtual
396 * addresses in a submap must match the main map, this is ok.
397 */
398
399 /* udv offset = (offset from start of entry) + entry's offset */
400 curr_offset = entry->offset + (vaddr - entry->start);
401 /* pmap va = vaddr (virtual address of pps[0]) */
402 curr_va = vaddr;
403
404 /*
405 * loop over the page range entering in as needed
406 */
407
408 retval = 0;
409 for (lcv = 0 ; lcv < npages ; lcv++, curr_offset += PAGE_SIZE,
410 curr_va += PAGE_SIZE) {
411 if ((flags & PGO_ALLPAGES) == 0 && lcv != centeridx)
412 continue;
413
414 if (pps[lcv] == PGO_DONTCARE)
415 continue;
416
417 mdpgno = cdev_mmap(device, curr_offset, access_type);
418 if (mdpgno == -1) {
419 retval = EIO;
420 break;
421 }
422 paddr = pmap_phys_address(mdpgno);
423 mmapflags = pmap_mmap_flags(mdpgno);
424 mapprot = ufi->entry->protection;
425 UVMHIST_LOG(maphist,
426 " MAPPING: device: pm=0x%x, va=0x%x, pa=0x%lx, at=%d",
427 ufi->orig_map->pmap, curr_va, paddr, mapprot);
428 if (pmap_enter(ufi->orig_map->pmap, curr_va, paddr, mapprot,
429 PMAP_CANFAIL | mapprot | mmapflags) != 0) {
430 /*
431 * pmap_enter() didn't have the resource to
432 * enter this mapping. Unlock everything,
433 * wait for the pagedaemon to free up some
434 * pages, and then tell uvm_fault() to start
435 * the fault again.
436 *
437 * XXX Needs some rethinking for the PGO_ALLPAGES
438 * XXX case.
439 */
440 pmap_update(ufi->orig_map->pmap); /* sync what we have so far */
441 uvmfault_unlockall(ufi, ufi->entry->aref.ar_amap,
442 uobj);
443 uvm_wait("udv_fault");
444 return (ERESTART);
445 }
446 }
447
448 pmap_update(ufi->orig_map->pmap);
449 uvmfault_unlockall(ufi, ufi->entry->aref.ar_amap, uobj);
450 return (retval);
451}
452