xref: /netbsd-src/sys/uvm/pmap/pmap.c (revision 82d56013d7b633d116a93943de88e08335357a7c)
1 /*	$NetBSD: pmap.c,v 1.62 2021/04/17 01:53:58 mrg Exp $	*/
2 
3 /*-
4  * Copyright (c) 1998, 2001 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Jason R. Thorpe of the Numerical Aerospace Simulation Facility,
9  * NASA Ames Research Center and by Chris G. Demetriou.
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
21  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
22  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
23  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
24  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
25  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
26  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
27  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
28  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
29  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
30  * POSSIBILITY OF SUCH DAMAGE.
31  */
32 
33 /*
34  * Copyright (c) 1992, 1993
35  *	The Regents of the University of California.  All rights reserved.
36  *
37  * This code is derived from software contributed to Berkeley by
38  * the Systems Programming Group of the University of Utah Computer
39  * Science Department and Ralph Campbell.
40  *
41  * Redistribution and use in source and binary forms, with or without
42  * modification, are permitted provided that the following conditions
43  * are met:
44  * 1. Redistributions of source code must retain the above copyright
45  *    notice, this list of conditions and the following disclaimer.
46  * 2. Redistributions in binary form must reproduce the above copyright
47  *    notice, this list of conditions and the following disclaimer in the
48  *    documentation and/or other materials provided with the distribution.
49  * 3. Neither the name of the University nor the names of its contributors
50  *    may be used to endorse or promote products derived from this software
51  *    without specific prior written permission.
52  *
53  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
54  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
55  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
56  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
57  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
58  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
59  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
60  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
61  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
62  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
63  * SUCH DAMAGE.
64  *
65  *	@(#)pmap.c	8.4 (Berkeley) 1/26/94
66  */
67 
68 #include <sys/cdefs.h>
69 
70 __KERNEL_RCSID(0, "$NetBSD: pmap.c,v 1.62 2021/04/17 01:53:58 mrg Exp $");
71 
72 /*
73  *	Manages physical address maps.
74  *
75  *	In addition to hardware address maps, this
76  *	module is called upon to provide software-use-only
77  *	maps which may or may not be stored in the same
78  *	form as hardware maps.  These pseudo-maps are
79  *	used to store intermediate results from copy
80  *	operations to and from address spaces.
81  *
82  *	Since the information managed by this module is
83  *	also stored by the logical address mapping module,
84  *	this module may throw away valid virtual-to-physical
85  *	mappings at almost any time.  However, invalidations
86  *	of virtual-to-physical mappings must be done as
87  *	requested.
88  *
89  *	In order to cope with hardware architectures which
90  *	make virtual-to-physical map invalidates expensive,
91  *	this module may delay invalidate or reduced protection
92  *	operations until such time as they are actually
93  *	necessary.  This module is given full information as
94  *	to which processors are currently using which maps,
95  *	and to when physical maps must be made correct.
96  */
97 
98 #include "opt_modular.h"
99 #include "opt_multiprocessor.h"
100 #include "opt_sysv.h"
101 
102 #define __PMAP_PRIVATE
103 
104 #include <sys/param.h>
105 
106 #include <sys/asan.h>
107 #include <sys/atomic.h>
108 #include <sys/buf.h>
109 #include <sys/cpu.h>
110 #include <sys/mutex.h>
111 #include <sys/pool.h>
112 
113 #include <uvm/uvm.h>
114 #include <uvm/uvm_physseg.h>
115 #include <uvm/pmap/pmap_pvt.h>
116 
117 #if defined(MULTIPROCESSOR) && defined(PMAP_VIRTUAL_CACHE_ALIASES) \
118     && !defined(PMAP_NO_PV_UNCACHED)
119 #error PMAP_VIRTUAL_CACHE_ALIASES with MULTIPROCESSOR requires \
120  PMAP_NO_PV_UNCACHED to be defined
121 #endif
122 
123 PMAP_COUNTER(remove_kernel_calls, "remove kernel calls");
124 PMAP_COUNTER(remove_kernel_pages, "kernel pages unmapped");
125 PMAP_COUNTER(remove_user_calls, "remove user calls");
126 PMAP_COUNTER(remove_user_pages, "user pages unmapped");
127 PMAP_COUNTER(remove_flushes, "remove cache flushes");
128 PMAP_COUNTER(remove_tlb_ops, "remove tlb ops");
129 PMAP_COUNTER(remove_pvfirst, "remove pv first");
130 PMAP_COUNTER(remove_pvsearch, "remove pv search");
131 
132 PMAP_COUNTER(prefer_requests, "prefer requests");
133 PMAP_COUNTER(prefer_adjustments, "prefer adjustments");
134 
135 PMAP_COUNTER(idlezeroed_pages, "pages idle zeroed");
136 
137 PMAP_COUNTER(kenter_pa, "kernel fast mapped pages");
138 PMAP_COUNTER(kenter_pa_bad, "kernel fast mapped pages (bad color)");
139 PMAP_COUNTER(kenter_pa_unmanaged, "kernel fast mapped unmanaged pages");
140 PMAP_COUNTER(kremove_pages, "kernel fast unmapped pages");
141 
142 PMAP_COUNTER(page_cache_evictions, "pages changed to uncacheable");
143 PMAP_COUNTER(page_cache_restorations, "pages changed to cacheable");
144 
145 PMAP_COUNTER(kernel_mappings_bad, "kernel pages mapped (bad color)");
146 PMAP_COUNTER(user_mappings_bad, "user pages mapped (bad color)");
147 PMAP_COUNTER(kernel_mappings, "kernel pages mapped");
148 PMAP_COUNTER(user_mappings, "user pages mapped");
149 PMAP_COUNTER(user_mappings_changed, "user mapping changed");
150 PMAP_COUNTER(kernel_mappings_changed, "kernel mapping changed");
151 PMAP_COUNTER(uncached_mappings, "uncached pages mapped");
152 PMAP_COUNTER(unmanaged_mappings, "unmanaged pages mapped");
153 PMAP_COUNTER(pvtracked_mappings, "pv-tracked unmanaged pages mapped");
154 PMAP_COUNTER(managed_mappings, "managed pages mapped");
155 PMAP_COUNTER(mappings, "pages mapped");
156 PMAP_COUNTER(remappings, "pages remapped");
157 PMAP_COUNTER(unmappings, "pages unmapped");
158 PMAP_COUNTER(primary_mappings, "page initial mappings");
159 PMAP_COUNTER(primary_unmappings, "page final unmappings");
160 PMAP_COUNTER(tlb_hit, "page mapping");
161 
162 PMAP_COUNTER(exec_mappings, "exec pages mapped");
163 PMAP_COUNTER(exec_synced_mappings, "exec pages synced");
164 PMAP_COUNTER(exec_synced_remove, "exec pages synced (PR)");
165 PMAP_COUNTER(exec_synced_clear_modify, "exec pages synced (CM)");
166 PMAP_COUNTER(exec_synced_page_protect, "exec pages synced (PP)");
167 PMAP_COUNTER(exec_synced_protect, "exec pages synced (P)");
168 PMAP_COUNTER(exec_uncached_page_protect, "exec pages uncached (PP)");
169 PMAP_COUNTER(exec_uncached_clear_modify, "exec pages uncached (CM)");
170 PMAP_COUNTER(exec_uncached_zero_page, "exec pages uncached (ZP)");
171 PMAP_COUNTER(exec_uncached_copy_page, "exec pages uncached (CP)");
172 PMAP_COUNTER(exec_uncached_remove, "exec pages uncached (PR)");
173 
174 PMAP_COUNTER(create, "creates");
175 PMAP_COUNTER(reference, "references");
176 PMAP_COUNTER(dereference, "dereferences");
177 PMAP_COUNTER(destroy, "destroyed");
178 PMAP_COUNTER(activate, "activations");
179 PMAP_COUNTER(deactivate, "deactivations");
180 PMAP_COUNTER(update, "updates");
181 #ifdef MULTIPROCESSOR
182 PMAP_COUNTER(shootdown_ipis, "shootdown IPIs");
183 #endif
184 PMAP_COUNTER(unwire, "unwires");
185 PMAP_COUNTER(copy, "copies");
186 PMAP_COUNTER(clear_modify, "clear_modifies");
187 PMAP_COUNTER(protect, "protects");
188 PMAP_COUNTER(page_protect, "page_protects");
189 
190 #define PMAP_ASID_RESERVED 0
191 CTASSERT(PMAP_ASID_RESERVED == 0);
192 
193 #ifndef PMAP_SEGTAB_ALIGN
194 #define PMAP_SEGTAB_ALIGN	/* nothing */
195 #endif
196 #ifdef _LP64
197 pmap_segtab_t	pmap_kstart_segtab PMAP_SEGTAB_ALIGN; /* first mid-level segtab for kernel */
198 #endif
199 pmap_segtab_t	pmap_kern_segtab PMAP_SEGTAB_ALIGN = { /* top level segtab for kernel */
200 #ifdef _LP64
201 	.seg_seg[(VM_MIN_KERNEL_ADDRESS & XSEGOFSET) >> SEGSHIFT] = &pmap_kstart_segtab,
202 #endif
203 };
204 
205 struct pmap_kernel kernel_pmap_store = {
206 	.kernel_pmap = {
207 		.pm_count = 1,
208 		.pm_segtab = &pmap_kern_segtab,
209 		.pm_minaddr = VM_MIN_KERNEL_ADDRESS,
210 		.pm_maxaddr = VM_MAX_KERNEL_ADDRESS,
211 	},
212 };
213 
214 struct pmap * const kernel_pmap_ptr = &kernel_pmap_store.kernel_pmap;
215 
216 /* The current top of kernel VM - gets updated by pmap_growkernel */
217 vaddr_t pmap_curmaxkvaddr;
218 
219 struct pmap_limits pmap_limits = {	/* VA and PA limits */
220 	.virtual_start = VM_MIN_KERNEL_ADDRESS,
221 	.virtual_end = VM_MAX_KERNEL_ADDRESS,
222 };
223 
224 #ifdef UVMHIST
225 static struct kern_history_ent pmapexechistbuf[10000];
226 static struct kern_history_ent pmaphistbuf[10000];
227 static struct kern_history_ent pmapsegtabhistbuf[1000];
228 UVMHIST_DEFINE(pmapexechist) = UVMHIST_INITIALIZER(pmapexechist, pmapexechistbuf);
229 UVMHIST_DEFINE(pmaphist) = UVMHIST_INITIALIZER(pmaphist, pmaphistbuf);
230 UVMHIST_DEFINE(pmapsegtabhist) = UVMHIST_INITIALIZER(pmapsegtabhist, pmapsegtabhistbuf);
231 #endif
232 
233 /*
234  * The pools from which pmap structures and sub-structures are allocated.
235  */
236 struct pool pmap_pmap_pool;
237 struct pool pmap_pv_pool;
238 
239 #ifndef PMAP_PV_LOWAT
240 #define	PMAP_PV_LOWAT	16
241 #endif
242 int	pmap_pv_lowat = PMAP_PV_LOWAT;
243 
244 bool	pmap_initialized = false;
245 #define	PMAP_PAGE_COLOROK_P(a, b) \
246 		((((int)(a) ^ (int)(b)) & pmap_page_colormask) == 0)
247 u_int	pmap_page_colormask;
248 
249 #define PAGE_IS_MANAGED(pa)	(pmap_initialized && uvm_pageismanaged(pa))
250 
251 #define PMAP_IS_ACTIVE(pm)						\
252 	((pm) == pmap_kernel() || 					\
253 	 (pm) == curlwp->l_proc->p_vmspace->vm_map.pmap)
254 
255 /* Forward function declarations */
256 void pmap_page_remove(struct vm_page_md *);
257 static void pmap_pvlist_check(struct vm_page_md *);
258 void pmap_remove_pv(pmap_t, vaddr_t, struct vm_page *, bool);
259 void pmap_enter_pv(pmap_t, vaddr_t, paddr_t, struct vm_page_md *, pt_entry_t *, u_int);
260 
261 /*
262  * PV table management functions.
263  */
264 void	*pmap_pv_page_alloc(struct pool *, int);
265 void	pmap_pv_page_free(struct pool *, void *);
266 
267 struct pool_allocator pmap_pv_page_allocator = {
268 	pmap_pv_page_alloc, pmap_pv_page_free, 0,
269 };
270 
271 #define	pmap_pv_alloc()		pool_get(&pmap_pv_pool, PR_NOWAIT)
272 #define	pmap_pv_free(pv)	pool_put(&pmap_pv_pool, (pv))
273 
274 #ifndef PMAP_NEED_TLB_MISS_LOCK
275 
276 #if defined(PMAP_MD_NEED_TLB_MISS_LOCK) || defined(DEBUG)
277 #define	PMAP_NEED_TLB_MISS_LOCK
278 #endif /* PMAP_MD_NEED_TLB_MISS_LOCK || DEBUG */
279 
280 #endif /* PMAP_NEED_TLB_MISS_LOCK */
281 
282 #ifdef PMAP_NEED_TLB_MISS_LOCK
283 
284 #ifdef PMAP_MD_NEED_TLB_MISS_LOCK
285 #define	pmap_tlb_miss_lock_init()	__nothing /* MD code deals with this */
286 #define	pmap_tlb_miss_lock_enter()	pmap_md_tlb_miss_lock_enter()
287 #define	pmap_tlb_miss_lock_exit()	pmap_md_tlb_miss_lock_exit()
288 #else
289 kmutex_t pmap_tlb_miss_lock 		__cacheline_aligned;
290 
291 static void
292 pmap_tlb_miss_lock_init(void)
293 {
294 	mutex_init(&pmap_tlb_miss_lock, MUTEX_SPIN, IPL_HIGH);
295 }
296 
297 static inline void
298 pmap_tlb_miss_lock_enter(void)
299 {
300 	mutex_spin_enter(&pmap_tlb_miss_lock);
301 }
302 
303 static inline void
304 pmap_tlb_miss_lock_exit(void)
305 {
306 	mutex_spin_exit(&pmap_tlb_miss_lock);
307 }
308 #endif /* PMAP_MD_NEED_TLB_MISS_LOCK */
309 
310 #else
311 
312 #define	pmap_tlb_miss_lock_init()	__nothing
313 #define	pmap_tlb_miss_lock_enter()	__nothing
314 #define	pmap_tlb_miss_lock_exit()	__nothing
315 
316 #endif /* PMAP_NEED_TLB_MISS_LOCK */
317 
318 #ifndef MULTIPROCESSOR
319 kmutex_t pmap_pvlist_mutex	__cacheline_aligned;
320 #endif
321 
322 /*
323  * Debug functions.
324  */
325 
326 #ifdef DEBUG
327 static inline void
328 pmap_asid_check(pmap_t pm, const char *func)
329 {
330 	if (!PMAP_IS_ACTIVE(pm))
331 		return;
332 
333 	struct pmap_asid_info * const pai = PMAP_PAI(pm, cpu_tlb_info(curcpu()));
334 	tlb_asid_t asid = tlb_get_asid();
335 	if (asid != pai->pai_asid)
336 		panic("%s: inconsistency for active TLB update: %u <-> %u",
337 		    func, asid, pai->pai_asid);
338 }
339 #endif
340 
341 static void
342 pmap_addr_range_check(pmap_t pmap, vaddr_t sva, vaddr_t eva, const char *func)
343 {
344 #ifdef DEBUG
345 	if (pmap == pmap_kernel()) {
346 		if (sva < VM_MIN_KERNEL_ADDRESS)
347 			panic("%s: kva %#"PRIxVADDR" not in range",
348 			    func, sva);
349 		if (eva >= pmap_limits.virtual_end)
350 			panic("%s: kva %#"PRIxVADDR" not in range",
351 			    func, eva);
352 	} else {
353 		if (eva > VM_MAXUSER_ADDRESS)
354 			panic("%s: uva %#"PRIxVADDR" not in range",
355 			    func, eva);
356 		pmap_asid_check(pmap, func);
357 	}
358 #endif
359 }
360 
361 /*
362  * Misc. functions.
363  */
364 
365 bool
366 pmap_page_clear_attributes(struct vm_page_md *mdpg, u_int clear_attributes)
367 {
368 	volatile unsigned long * const attrp = &mdpg->mdpg_attrs;
369 #ifdef MULTIPROCESSOR
370 	for (;;) {
371 		u_int old_attr = *attrp;
372 		if ((old_attr & clear_attributes) == 0)
373 			return false;
374 		u_int new_attr = old_attr & ~clear_attributes;
375 		if (old_attr == atomic_cas_ulong(attrp, old_attr, new_attr))
376 			return true;
377 	}
378 #else
379 	unsigned long old_attr = *attrp;
380 	if ((old_attr & clear_attributes) == 0)
381 		return false;
382 	*attrp &= ~clear_attributes;
383 	return true;
384 #endif
385 }
386 
387 void
388 pmap_page_set_attributes(struct vm_page_md *mdpg, u_int set_attributes)
389 {
390 #ifdef MULTIPROCESSOR
391 	atomic_or_ulong(&mdpg->mdpg_attrs, set_attributes);
392 #else
393 	mdpg->mdpg_attrs |= set_attributes;
394 #endif
395 }
396 
397 static void
398 pmap_page_syncicache(struct vm_page *pg)
399 {
400 	UVMHIST_FUNC(__func__);
401 	UVMHIST_CALLED(pmaphist);
402 #ifndef MULTIPROCESSOR
403 	struct pmap * const curpmap = curlwp->l_proc->p_vmspace->vm_map.pmap;
404 #endif
405 	struct vm_page_md * const mdpg = VM_PAGE_TO_MD(pg);
406 	pv_entry_t pv = &mdpg->mdpg_first;
407 	kcpuset_t *onproc;
408 #ifdef MULTIPROCESSOR
409 	kcpuset_create(&onproc, true);
410 	KASSERT(onproc != NULL);
411 #else
412 	onproc = NULL;
413 #endif
414 	VM_PAGEMD_PVLIST_READLOCK(mdpg);
415 	pmap_pvlist_check(mdpg);
416 
417 	UVMHIST_LOG(pmaphist, "pv %#jx pv_pmap %#jx", (uintptr_t)pv,
418 	    (uintptr_t)pv->pv_pmap, 0, 0);
419 
420 	if (pv->pv_pmap != NULL) {
421 		for (; pv != NULL; pv = pv->pv_next) {
422 #ifdef MULTIPROCESSOR
423 			UVMHIST_LOG(pmaphist, "pv %#jx pv_pmap %#jx",
424 			    (uintptr_t)pv, (uintptr_t)pv->pv_pmap, 0, 0);
425 			kcpuset_merge(onproc, pv->pv_pmap->pm_onproc);
426 			if (kcpuset_match(onproc, kcpuset_running)) {
427 				break;
428 			}
429 #else
430 			if (pv->pv_pmap == curpmap) {
431 				onproc = curcpu()->ci_data.cpu_kcpuset;
432 				break;
433 			}
434 #endif
435 		}
436 	}
437 	pmap_pvlist_check(mdpg);
438 	VM_PAGEMD_PVLIST_UNLOCK(mdpg);
439 	kpreempt_disable();
440 	pmap_md_page_syncicache(mdpg, onproc);
441 	kpreempt_enable();
442 #ifdef MULTIPROCESSOR
443 	kcpuset_destroy(onproc);
444 #endif
445 }
446 
447 /*
448  * Define the initial bounds of the kernel virtual address space.
449  */
450 void
451 pmap_virtual_space(vaddr_t *vstartp, vaddr_t *vendp)
452 {
453 
454 	*vstartp = pmap_limits.virtual_start;
455 	*vendp = pmap_limits.virtual_end;
456 }
457 
458 vaddr_t
459 pmap_growkernel(vaddr_t maxkvaddr)
460 {
461 	UVMHIST_FUNC(__func__);
462 	UVMHIST_CALLARGS(pmaphist, "maxkvaddr=%#jx (%#jx)", maxkvaddr,
463 	    pmap_curmaxkvaddr, 0, 0);
464 
465 	vaddr_t virtual_end = pmap_curmaxkvaddr;
466 	maxkvaddr = pmap_round_seg(maxkvaddr) - 1;
467 
468 	/*
469 	 * Don't exceed VM_MAX_KERNEL_ADDRESS!
470 	 */
471 	if (maxkvaddr == 0 || maxkvaddr > VM_MAX_KERNEL_ADDRESS)
472 		maxkvaddr = VM_MAX_KERNEL_ADDRESS;
473 
474 	/*
475 	 * Reserve PTEs for the new KVA space.
476 	 */
477 	for (; virtual_end < maxkvaddr; virtual_end += NBSEG) {
478 		pmap_pte_reserve(pmap_kernel(), virtual_end, 0);
479 	}
480 
481 	kasan_shadow_map((void *)pmap_curmaxkvaddr,
482 	    (size_t)(virtual_end - pmap_curmaxkvaddr));
483 
484 	/*
485 	 * Update new end.
486 	 */
487 	pmap_curmaxkvaddr = virtual_end;
488 
489 	UVMHIST_LOG(pmaphist, " <-- done", 0, 0, 0, 0);
490 
491 	return virtual_end;
492 }
493 
494 /*
495  * Bootstrap memory allocator (alternative to vm_bootstrap_steal_memory()).
496  * This function allows for early dynamic memory allocation until the virtual
497  * memory system has been bootstrapped.  After that point, either kmem_alloc
498  * or malloc should be used.  This function works by stealing pages from the
499  * (to be) managed page pool, then implicitly mapping the pages (by using
500  * their direct mapped addresses) and zeroing them.
501  *
502  * It may be used once the physical memory segments have been pre-loaded
503  * into the vm_physmem[] array.  Early memory allocation MUST use this
504  * interface!  This cannot be used after vm_page_startup(), and will
505  * generate a panic if tried.
506  *
507  * Note that this memory will never be freed, and in essence it is wired
508  * down.
509  *
510  * We must adjust *vstartp and/or *vendp iff we use address space
511  * from the kernel virtual address range defined by pmap_virtual_space().
512  */
513 vaddr_t
514 pmap_steal_memory(vsize_t size, vaddr_t *vstartp, vaddr_t *vendp)
515 {
516 	size_t npgs;
517 	paddr_t pa;
518 	vaddr_t va;
519 
520 	uvm_physseg_t maybe_bank = UVM_PHYSSEG_TYPE_INVALID;
521 
522 	size = round_page(size);
523 	npgs = atop(size);
524 
525 	aprint_debug("%s: need %zu pages\n", __func__, npgs);
526 
527 	for (uvm_physseg_t bank = uvm_physseg_get_first();
528 	     uvm_physseg_valid_p(bank);
529 	     bank = uvm_physseg_get_next(bank)) {
530 
531 		if (uvm.page_init_done == true)
532 			panic("pmap_steal_memory: called _after_ bootstrap");
533 
534 		aprint_debug("%s: seg %"PRIxPHYSSEG": %#"PRIxPADDR" %#"PRIxPADDR" %#"PRIxPADDR" %#"PRIxPADDR"\n",
535 		    __func__, bank,
536 		    uvm_physseg_get_avail_start(bank), uvm_physseg_get_start(bank),
537 		    uvm_physseg_get_avail_end(bank), uvm_physseg_get_end(bank));
538 
539 		if (uvm_physseg_get_avail_start(bank) != uvm_physseg_get_start(bank)
540 		    || uvm_physseg_get_avail_start(bank) >= uvm_physseg_get_avail_end(bank)) {
541 			aprint_debug("%s: seg %"PRIxPHYSSEG": bad start\n", __func__, bank);
542 			continue;
543 		}
544 
545 		if (uvm_physseg_get_avail_end(bank) - uvm_physseg_get_avail_start(bank) < npgs) {
546 			aprint_debug("%s: seg %"PRIxPHYSSEG": too small for %zu pages\n",
547 			    __func__, bank, npgs);
548 			continue;
549 		}
550 
551 		if (!pmap_md_ok_to_steal_p(bank, npgs)) {
552 			continue;
553 		}
554 
555 		/*
556 		 * Always try to allocate from the segment with the least
557 		 * amount of space left.
558 		 */
559 #define VM_PHYSMEM_SPACE(b)	((uvm_physseg_get_avail_end(b)) - (uvm_physseg_get_avail_start(b)))
560 		if (uvm_physseg_valid_p(maybe_bank) == false
561 		    || VM_PHYSMEM_SPACE(bank) < VM_PHYSMEM_SPACE(maybe_bank)) {
562 			maybe_bank = bank;
563 		}
564 	}
565 
566 	if (uvm_physseg_valid_p(maybe_bank)) {
567 		const uvm_physseg_t bank = maybe_bank;
568 
569 		/*
570 		 * There are enough pages here; steal them!
571 		 */
572 		pa = ptoa(uvm_physseg_get_start(bank));
573 		uvm_physseg_unplug(atop(pa), npgs);
574 
575 		aprint_debug("%s: seg %"PRIxPHYSSEG": %zu pages stolen (%#"PRIxPADDR" left)\n",
576 		    __func__, bank, npgs, VM_PHYSMEM_SPACE(bank));
577 
578 		va = pmap_md_map_poolpage(pa, size);
579 		memset((void *)va, 0, size);
580 		return va;
581 	}
582 
583 	/*
584 	 * If we got here, there was no memory left.
585 	 */
586 	panic("pmap_steal_memory: no memory to steal %zu pages", npgs);
587 }
588 
589 /*
590  *	Bootstrap the system enough to run with virtual memory.
591  *	(Common routine called by machine-dependent bootstrap code.)
592  */
593 void
594 pmap_bootstrap_common(void)
595 {
596 	pmap_tlb_miss_lock_init();
597 }
598 
599 /*
600  *	Initialize the pmap module.
601  *	Called by vm_init, to initialize any structures that the pmap
602  *	system needs to map virtual memory.
603  */
604 void
605 pmap_init(void)
606 {
607 	UVMHIST_LINK_STATIC(pmapexechist);
608 	UVMHIST_LINK_STATIC(pmaphist);
609 	UVMHIST_LINK_STATIC(pmapsegtabhist);
610 
611 	UVMHIST_FUNC(__func__);
612 	UVMHIST_CALLED(pmaphist);
613 
614 	/*
615 	 * Initialize the segtab lock.
616 	 */
617 	mutex_init(&pmap_segtab_lock, MUTEX_DEFAULT, IPL_HIGH);
618 
619 	/*
620 	 * Set a low water mark on the pv_entry pool, so that we are
621 	 * more likely to have these around even in extreme memory
622 	 * starvation.
623 	 */
624 	pool_setlowat(&pmap_pv_pool, pmap_pv_lowat);
625 
626 	/*
627 	 * Set the page colormask but allow pmap_md_init to override it.
628 	 */
629 	pmap_page_colormask = ptoa(uvmexp.colormask);
630 
631 	pmap_md_init();
632 
633 	/*
634 	 * Now it is safe to enable pv entry recording.
635 	 */
636 	pmap_initialized = true;
637 }
638 
639 /*
640  *	Create and return a physical map.
641  *
642  *	If the size specified for the map
643  *	is zero, the map is an actual physical
644  *	map, and may be referenced by the
645  *	hardware.
646  *
647  *	If the size specified is non-zero,
648  *	the map will be used in software only, and
649  *	is bounded by that size.
650  */
651 pmap_t
652 pmap_create(void)
653 {
654 	UVMHIST_FUNC(__func__);
655 	UVMHIST_CALLED(pmaphist);
656 	PMAP_COUNT(create);
657 
658 	pmap_t pmap = pool_get(&pmap_pmap_pool, PR_WAITOK);
659 	memset(pmap, 0, PMAP_SIZE);
660 
661 	KASSERT(pmap->pm_pai[0].pai_link.le_prev == NULL);
662 
663 	pmap->pm_count = 1;
664 	pmap->pm_minaddr = VM_MIN_ADDRESS;
665 	pmap->pm_maxaddr = VM_MAXUSER_ADDRESS;
666 
667 	pmap_segtab_init(pmap);
668 
669 #ifdef MULTIPROCESSOR
670 	kcpuset_create(&pmap->pm_active, true);
671 	kcpuset_create(&pmap->pm_onproc, true);
672 	KASSERT(pmap->pm_active != NULL);
673 	KASSERT(pmap->pm_onproc != NULL);
674 #endif
675 
676 	UVMHIST_LOG(pmaphist, " <-- done (pmap=%#jx)", (uintptr_t)pmap,
677 	    0, 0, 0);
678 
679 	return pmap;
680 }
681 
682 /*
683  *	Retire the given physical map from service.
684  *	Should only be called if the map contains
685  *	no valid mappings.
686  */
687 void
688 pmap_destroy(pmap_t pmap)
689 {
690 	UVMHIST_FUNC(__func__);
691 	UVMHIST_CALLARGS(pmaphist, "(pmap=%#jx)", (uintptr_t)pmap, 0, 0, 0);
692 
693 	if (atomic_dec_uint_nv(&pmap->pm_count) > 0) {
694 		PMAP_COUNT(dereference);
695 		UVMHIST_LOG(pmaphist, " <-- done (deref)", 0, 0, 0, 0);
696 		return;
697 	}
698 
699 	PMAP_COUNT(destroy);
700 	KASSERT(pmap->pm_count == 0);
701 	kpreempt_disable();
702 	pmap_tlb_miss_lock_enter();
703 	pmap_tlb_asid_release_all(pmap);
704 	pmap_segtab_destroy(pmap, NULL, 0);
705 	pmap_tlb_miss_lock_exit();
706 
707 #ifdef MULTIPROCESSOR
708 	kcpuset_destroy(pmap->pm_active);
709 	kcpuset_destroy(pmap->pm_onproc);
710 	pmap->pm_active = NULL;
711 	pmap->pm_onproc = NULL;
712 #endif
713 
714 	pool_put(&pmap_pmap_pool, pmap);
715 	kpreempt_enable();
716 
717 	UVMHIST_LOG(pmaphist, " <-- done (freed)", 0, 0, 0, 0);
718 }
719 
720 /*
721  *	Add a reference to the specified pmap.
722  */
723 void
724 pmap_reference(pmap_t pmap)
725 {
726 	UVMHIST_FUNC(__func__);
727 	UVMHIST_CALLARGS(pmaphist, "(pmap=%#jx)", (uintptr_t)pmap, 0, 0, 0);
728 	PMAP_COUNT(reference);
729 
730 	if (pmap != NULL) {
731 		atomic_inc_uint(&pmap->pm_count);
732 	}
733 
734 	UVMHIST_LOG(pmaphist, " <-- done", 0, 0, 0, 0);
735 }
736 
737 /*
738  *	Make a new pmap (vmspace) active for the given process.
739  */
740 void
741 pmap_activate(struct lwp *l)
742 {
743 	pmap_t pmap = l->l_proc->p_vmspace->vm_map.pmap;
744 
745 	UVMHIST_FUNC(__func__);
746 	UVMHIST_CALLARGS(pmaphist, "(l=%#jx pmap=%#jx)", (uintptr_t)l,
747 	    (uintptr_t)pmap, 0, 0);
748 	PMAP_COUNT(activate);
749 
750 	kpreempt_disable();
751 	pmap_tlb_miss_lock_enter();
752 	pmap_tlb_asid_acquire(pmap, l);
753 	pmap_segtab_activate(pmap, l);
754 	pmap_tlb_miss_lock_exit();
755 	kpreempt_enable();
756 
757 	UVMHIST_LOG(pmaphist, " <-- done (%ju:%ju)", l->l_proc->p_pid,
758 	    l->l_lid, 0, 0);
759 }
760 
761 /*
762  * Remove this page from all physical maps in which it resides.
763  * Reflects back modify bits to the pager.
764  */
765 void
766 pmap_page_remove(struct vm_page_md *mdpg)
767 {
768 	kpreempt_disable();
769 	VM_PAGEMD_PVLIST_LOCK(mdpg);
770 	pmap_pvlist_check(mdpg);
771 
772 	struct vm_page * const pg =
773 	    VM_PAGEMD_VMPAGE_P(mdpg) ? VM_MD_TO_PAGE(mdpg) : NULL;
774 
775 	UVMHIST_FUNC(__func__);
776 	if (pg) {
777 		UVMHIST_CALLARGS(pmaphist, "mdpg %#jx pg %#jx (pa %#jx): "
778 		    "execpage cleared", (uintptr_t)mdpg, (uintptr_t)pg,
779 		    VM_PAGE_TO_PHYS(pg), 0);
780 	} else {
781 		UVMHIST_CALLARGS(pmaphist, "mdpg %#jx", (uintptr_t)mdpg, 0,
782 		    0, 0);
783 	}
784 
785 #ifdef PMAP_VIRTUAL_CACHE_ALIASES
786 	pmap_page_clear_attributes(mdpg, VM_PAGEMD_EXECPAGE|VM_PAGEMD_UNCACHED);
787 #else
788 	pmap_page_clear_attributes(mdpg, VM_PAGEMD_EXECPAGE);
789 #endif
790 	PMAP_COUNT(exec_uncached_remove);
791 
792 	pv_entry_t pv = &mdpg->mdpg_first;
793 	if (pv->pv_pmap == NULL) {
794 		VM_PAGEMD_PVLIST_UNLOCK(mdpg);
795 		kpreempt_enable();
796 		UVMHIST_LOG(pmaphist, " <-- done (empty)", 0, 0, 0, 0);
797 		return;
798 	}
799 
800 	pv_entry_t npv;
801 	pv_entry_t pvp = NULL;
802 
803 	for (; pv != NULL; pv = npv) {
804 		npv = pv->pv_next;
805 #ifdef PMAP_VIRTUAL_CACHE_ALIASES
806 		if (PV_ISKENTER_P(pv)) {
807 			UVMHIST_LOG(pmaphist, " pv %#jx pmap %#jx va %#jx"
808 			    " skip", (uintptr_t)pv, (uintptr_t)pv->pv_pmap,
809 			    pv->pv_va, 0);
810 
811 			KASSERT(pv->pv_pmap == pmap_kernel());
812 
813 			/* Assume no more - it'll get fixed if there are */
814 			pv->pv_next = NULL;
815 
816 			/*
817 			 * pvp is non-null when we already have a PV_KENTER
818 			 * pv in pvh_first; otherwise we haven't seen a
819 			 * PV_KENTER pv and we need to copy this one to
820 			 * pvh_first
821 			 */
822 			if (pvp) {
823 				/*
824 				 * The previous PV_KENTER pv needs to point to
825 				 * this PV_KENTER pv
826 				 */
827 				pvp->pv_next = pv;
828 			} else {
829 				pv_entry_t fpv = &mdpg->mdpg_first;
830 				*fpv = *pv;
831 				KASSERT(fpv->pv_pmap == pmap_kernel());
832 			}
833 			pvp = pv;
834 			continue;
835 		}
836 #endif
837 		const pmap_t pmap = pv->pv_pmap;
838 		vaddr_t va = trunc_page(pv->pv_va);
839 		pt_entry_t * const ptep = pmap_pte_lookup(pmap, va);
840 		KASSERTMSG(ptep != NULL, "%#"PRIxVADDR " %#"PRIxVADDR, va,
841 		    pmap_limits.virtual_end);
842 		pt_entry_t pte = *ptep;
843 		UVMHIST_LOG(pmaphist, " pv %#jx pmap %#jx va %#jx"
844 		    " pte %#jx", (uintptr_t)pv, (uintptr_t)pmap, va,
845 		    pte_value(pte));
846 		if (!pte_valid_p(pte))
847 			continue;
848 		const bool is_kernel_pmap_p = (pmap == pmap_kernel());
849 		if (is_kernel_pmap_p) {
850 			PMAP_COUNT(remove_kernel_pages);
851 		} else {
852 			PMAP_COUNT(remove_user_pages);
853 		}
854 		if (pte_wired_p(pte))
855 			pmap->pm_stats.wired_count--;
856 		pmap->pm_stats.resident_count--;
857 
858 		pmap_tlb_miss_lock_enter();
859 		const pt_entry_t npte = pte_nv_entry(is_kernel_pmap_p);
860 		pte_set(ptep, npte);
861 		if (__predict_true(!(pmap->pm_flags & PMAP_DEFERRED_ACTIVATE))) {
862 			/*
863 			 * Flush the TLB for the given address.
864 			 */
865 			pmap_tlb_invalidate_addr(pmap, va);
866 		}
867 		pmap_tlb_miss_lock_exit();
868 
869 		/*
870 		 * non-null means this is a non-pvh_first pv, so we should
871 		 * free it.
872 		 */
873 		if (pvp) {
874 			KASSERT(pvp->pv_pmap == pmap_kernel());
875 			KASSERT(pvp->pv_next == NULL);
876 			pmap_pv_free(pv);
877 		} else {
878 			pv->pv_pmap = NULL;
879 			pv->pv_next = NULL;
880 		}
881 	}
882 
883 	pmap_pvlist_check(mdpg);
884 	VM_PAGEMD_PVLIST_UNLOCK(mdpg);
885 	kpreempt_enable();
886 
887 	UVMHIST_LOG(pmaphist, " <-- done", 0, 0, 0, 0);
888 }
889 
890 #ifdef __HAVE_PMAP_PV_TRACK
891 /*
892  * pmap_pv_protect: change protection of an unmanaged pv-tracked page from
893  * all pmaps that map it
894  */
895 void
896 pmap_pv_protect(paddr_t pa, vm_prot_t prot)
897 {
898 
899 	/* the only case is remove at the moment */
900 	KASSERT(prot == VM_PROT_NONE);
901 	struct pmap_page *pp;
902 
903 	pp = pmap_pv_tracked(pa);
904 	if (pp == NULL)
905 		panic("pmap_pv_protect: page not pv-tracked: 0x%"PRIxPADDR,
906 		    pa);
907 
908 	struct vm_page_md *mdpg = PMAP_PAGE_TO_MD(pp);
909 	pmap_page_remove(mdpg);
910 }
911 #endif
912 
913 /*
914  *	Make a previously active pmap (vmspace) inactive.
915  */
916 void
917 pmap_deactivate(struct lwp *l)
918 {
919 	pmap_t pmap = l->l_proc->p_vmspace->vm_map.pmap;
920 
921 	UVMHIST_FUNC(__func__);
922 	UVMHIST_CALLARGS(pmaphist, "(l=%#jx pmap=%#jx)", (uintptr_t)l,
923 	    (uintptr_t)pmap, 0, 0);
924 	PMAP_COUNT(deactivate);
925 
926 	kpreempt_disable();
927 	KASSERT(l == curlwp || l->l_cpu == curlwp->l_cpu);
928 	pmap_tlb_miss_lock_enter();
929 	pmap_tlb_asid_deactivate(pmap);
930 	pmap_segtab_deactivate(pmap);
931 	pmap_tlb_miss_lock_exit();
932 	kpreempt_enable();
933 
934 	UVMHIST_LOG(pmaphist, " <-- done (%ju:%ju)", l->l_proc->p_pid,
935 	    l->l_lid, 0, 0);
936 }
937 
938 void
939 pmap_update(struct pmap *pmap)
940 {
941 	UVMHIST_FUNC(__func__);
942 	UVMHIST_CALLARGS(pmaphist, "(pmap=%#jx)", (uintptr_t)pmap, 0, 0, 0);
943 	PMAP_COUNT(update);
944 
945 	kpreempt_disable();
946 #if defined(MULTIPROCESSOR) && defined(PMAP_TLB_NEED_SHOOTDOWN)
947 	u_int pending = atomic_swap_uint(&pmap->pm_shootdown_pending, 0);
948 	if (pending && pmap_tlb_shootdown_bystanders(pmap))
949 		PMAP_COUNT(shootdown_ipis);
950 #endif
951 	pmap_tlb_miss_lock_enter();
952 #if defined(DEBUG) && !defined(MULTIPROCESSOR)
953 	pmap_tlb_check(pmap, pmap_md_tlb_check_entry);
954 #endif /* DEBUG */
955 
956 	/*
957 	 * If pmap_remove_all was called, we deactivated ourselves and nuked
958 	 * our ASID.  Now we have to reactivate ourselves.
959 	 */
960 	if (__predict_false(pmap->pm_flags & PMAP_DEFERRED_ACTIVATE)) {
961 		pmap->pm_flags ^= PMAP_DEFERRED_ACTIVATE;
962 		pmap_tlb_asid_acquire(pmap, curlwp);
963 		pmap_segtab_activate(pmap, curlwp);
964 	}
965 	pmap_tlb_miss_lock_exit();
966 	kpreempt_enable();
967 
968 	UVMHIST_LOG(pmaphist, " <-- done (kernel=%jd)",
969 		    (pmap == pmap_kernel() ? 1 : 0), 0, 0, 0);
970 }
971 
972 /*
973  *	Remove the given range of addresses from the specified map.
974  *
975  *	It is assumed that the start and end are properly
976  *	rounded to the page size.
977  */
978 
979 static bool
980 pmap_pte_remove(pmap_t pmap, vaddr_t sva, vaddr_t eva, pt_entry_t *ptep,
981 	uintptr_t flags)
982 {
983 	const pt_entry_t npte = flags;
984 	const bool is_kernel_pmap_p = (pmap == pmap_kernel());
985 
986 	UVMHIST_FUNC(__func__);
987 	UVMHIST_CALLARGS(pmaphist, "(pmap=%#jx kernel=%jd va=%#jx..%#jx)",
988 	    (uintptr_t)pmap, (pmap == pmap_kernel() ? 1 : 0), sva, eva);
989 	UVMHIST_LOG(pmaphist, "ptep=%#jx, flags(npte)=%#jx)",
990 	    (uintptr_t)ptep, flags, 0, 0);
991 
992 	KASSERT(kpreempt_disabled());
993 
994 	for (; sva < eva; sva += NBPG, ptep++) {
995 		const pt_entry_t pte = *ptep;
996 		if (!pte_valid_p(pte))
997 			continue;
998 		if (is_kernel_pmap_p) {
999 			PMAP_COUNT(remove_kernel_pages);
1000 		} else {
1001 			PMAP_COUNT(remove_user_pages);
1002 		}
1003 		if (pte_wired_p(pte))
1004 			pmap->pm_stats.wired_count--;
1005 		pmap->pm_stats.resident_count--;
1006 		struct vm_page * const pg = PHYS_TO_VM_PAGE(pte_to_paddr(pte));
1007 		if (__predict_true(pg != NULL)) {
1008 			pmap_remove_pv(pmap, sva, pg, pte_modified_p(pte));
1009 		}
1010 		pmap_tlb_miss_lock_enter();
1011 		pte_set(ptep, npte);
1012 		if (__predict_true(!(pmap->pm_flags & PMAP_DEFERRED_ACTIVATE))) {
1013 
1014 			/*
1015 			 * Flush the TLB for the given address.
1016 			 */
1017 			pmap_tlb_invalidate_addr(pmap, sva);
1018 		}
1019 		pmap_tlb_miss_lock_exit();
1020 	}
1021 
1022 	UVMHIST_LOG(pmaphist, " <-- done", 0, 0, 0, 0);
1023 
1024 	return false;
1025 }
1026 
1027 void
1028 pmap_remove(pmap_t pmap, vaddr_t sva, vaddr_t eva)
1029 {
1030 	const bool is_kernel_pmap_p = (pmap == pmap_kernel());
1031 	const pt_entry_t npte = pte_nv_entry(is_kernel_pmap_p);
1032 
1033 	UVMHIST_FUNC(__func__);
1034 	UVMHIST_CALLARGS(pmaphist, "(pmap=%#jx, va=%#jx..%#jx)",
1035 	    (uintptr_t)pmap, sva, eva, 0);
1036 
1037 	if (is_kernel_pmap_p) {
1038 		PMAP_COUNT(remove_kernel_calls);
1039 	} else {
1040 		PMAP_COUNT(remove_user_calls);
1041 	}
1042 #ifdef PMAP_FAULTINFO
1043 	curpcb->pcb_faultinfo.pfi_faultaddr = 0;
1044 	curpcb->pcb_faultinfo.pfi_repeats = 0;
1045 	curpcb->pcb_faultinfo.pfi_faultptep = NULL;
1046 #endif
1047 	kpreempt_disable();
1048 	pmap_addr_range_check(pmap, sva, eva, __func__);
1049 	pmap_pte_process(pmap, sva, eva, pmap_pte_remove, npte);
1050 	kpreempt_enable();
1051 
1052 	UVMHIST_LOG(pmaphist, " <-- done", 0, 0, 0, 0);
1053 }
1054 
1055 /*
1056  *	pmap_page_protect:
1057  *
1058  *	Lower the permission for all mappings to a given page.
1059  */
1060 void
1061 pmap_page_protect(struct vm_page *pg, vm_prot_t prot)
1062 {
1063 	struct vm_page_md * const mdpg = VM_PAGE_TO_MD(pg);
1064 	pv_entry_t pv;
1065 	vaddr_t va;
1066 
1067 	UVMHIST_FUNC(__func__);
1068 	UVMHIST_CALLARGS(pmaphist, "(pg=%#jx (pa %#jx) prot=%#jx)",
1069 	    (uintptr_t)pg, VM_PAGE_TO_PHYS(pg), prot, 0);
1070 	PMAP_COUNT(page_protect);
1071 
1072 	switch (prot) {
1073 	case VM_PROT_READ|VM_PROT_WRITE:
1074 	case VM_PROT_ALL:
1075 		break;
1076 
1077 	/* copy_on_write */
1078 	case VM_PROT_READ:
1079 	case VM_PROT_READ|VM_PROT_EXECUTE:
1080 		pv = &mdpg->mdpg_first;
1081 		kpreempt_disable();
1082 		VM_PAGEMD_PVLIST_READLOCK(mdpg);
1083 		pmap_pvlist_check(mdpg);
1084 		/*
1085 		 * Loop over all current mappings setting/clearing as
1086 		 * appropriate.
1087 		 */
1088 		if (pv->pv_pmap != NULL) {
1089 			while (pv != NULL) {
1090 #ifdef PMAP_VIRTUAL_CACHE_ALIASES
1091 				if (PV_ISKENTER_P(pv)) {
1092 					pv = pv->pv_next;
1093 					continue;
1094 				}
1095 #endif
1096 				const pmap_t pmap = pv->pv_pmap;
1097 				va = trunc_page(pv->pv_va);
1098 				const uintptr_t gen =
1099 				    VM_PAGEMD_PVLIST_UNLOCK(mdpg);
1100 				pmap_protect(pmap, va, va + PAGE_SIZE, prot);
1101 				KASSERT(pv->pv_pmap == pmap);
1102 				pmap_update(pmap);
1103 				if (gen != VM_PAGEMD_PVLIST_READLOCK(mdpg)) {
1104 					pv = &mdpg->mdpg_first;
1105 				} else {
1106 					pv = pv->pv_next;
1107 				}
1108 				pmap_pvlist_check(mdpg);
1109 			}
1110 		}
1111 		pmap_pvlist_check(mdpg);
1112 		VM_PAGEMD_PVLIST_UNLOCK(mdpg);
1113 		kpreempt_enable();
1114 		break;
1115 
1116 	/* remove_all */
1117 	default:
1118 		pmap_page_remove(mdpg);
1119 	}
1120 
1121 	UVMHIST_LOG(pmaphist, " <-- done", 0, 0, 0, 0);
1122 }
1123 
1124 static bool
1125 pmap_pte_protect(pmap_t pmap, vaddr_t sva, vaddr_t eva, pt_entry_t *ptep,
1126 	uintptr_t flags)
1127 {
1128 	const vm_prot_t prot = (flags & VM_PROT_ALL);
1129 
1130 	UVMHIST_FUNC(__func__);
1131 	UVMHIST_CALLARGS(pmaphist, "(pmap=%#jx kernel=%jd va=%#jx..%#jx)",
1132 	    (uintptr_t)pmap, (pmap == pmap_kernel() ? 1 : 0), sva, eva);
1133 	UVMHIST_LOG(pmaphist, "ptep=%#jx, flags(npte)=%#jx)",
1134 	    (uintptr_t)ptep, flags, 0, 0);
1135 
1136 	KASSERT(kpreempt_disabled());
1137 	/*
1138 	 * Change protection on every valid mapping within this segment.
1139 	 */
1140 	for (; sva < eva; sva += NBPG, ptep++) {
1141 		pt_entry_t pte = *ptep;
1142 		if (!pte_valid_p(pte))
1143 			continue;
1144 		struct vm_page * const pg = PHYS_TO_VM_PAGE(pte_to_paddr(pte));
1145 		if (pg != NULL && pte_modified_p(pte)) {
1146 			struct vm_page_md * const mdpg = VM_PAGE_TO_MD(pg);
1147 			if (VM_PAGEMD_EXECPAGE_P(mdpg)) {
1148 				KASSERT(!VM_PAGEMD_PVLIST_EMPTY_P(mdpg));
1149 #ifdef PMAP_VIRTUAL_CACHE_ALIASES
1150 				if (VM_PAGEMD_CACHED_P(mdpg)) {
1151 #endif
1152 					UVMHIST_LOG(pmapexechist,
1153 					    "pg %#jx (pa %#jx): "
1154 					    "syncicached performed",
1155 					    (uintptr_t)pg, VM_PAGE_TO_PHYS(pg),
1156 					    0, 0);
1157 					pmap_page_syncicache(pg);
1158 					PMAP_COUNT(exec_synced_protect);
1159 #ifdef PMAP_VIRTUAL_CACHE_ALIASES
1160 				}
1161 #endif
1162 			}
1163 		}
1164 		pte = pte_prot_downgrade(pte, prot);
1165 		if (*ptep != pte) {
1166 			pmap_tlb_miss_lock_enter();
1167 			pte_set(ptep, pte);
1168 			/*
1169 			 * Update the TLB if needed.
1170 			 */
1171 			pmap_tlb_update_addr(pmap, sva, pte, PMAP_TLB_NEED_IPI);
1172 			pmap_tlb_miss_lock_exit();
1173 		}
1174 	}
1175 
1176 	UVMHIST_LOG(pmaphist, " <-- done", 0, 0, 0, 0);
1177 
1178 	return false;
1179 }
1180 
1181 /*
1182  *	Set the physical protection on the
1183  *	specified range of this map as requested.
1184  */
1185 void
1186 pmap_protect(pmap_t pmap, vaddr_t sva, vaddr_t eva, vm_prot_t prot)
1187 {
1188 	UVMHIST_FUNC(__func__);
1189 	UVMHIST_CALLARGS(pmaphist, "(pmap=%#jx, va=%#jx..%#jx, prot=%ju)",
1190 	    (uintptr_t)pmap, sva, eva, prot);
1191 	PMAP_COUNT(protect);
1192 
1193 	if ((prot & VM_PROT_READ) == VM_PROT_NONE) {
1194 		pmap_remove(pmap, sva, eva);
1195 		UVMHIST_LOG(pmaphist, " <-- done", 0, 0, 0, 0);
1196 		return;
1197 	}
1198 
1199 	/*
1200 	 * Change protection on every valid mapping within this segment.
1201 	 */
1202 	kpreempt_disable();
1203 	pmap_addr_range_check(pmap, sva, eva, __func__);
1204 	pmap_pte_process(pmap, sva, eva, pmap_pte_protect, prot);
1205 	kpreempt_enable();
1206 
1207 	UVMHIST_LOG(pmaphist, " <-- done", 0, 0, 0, 0);
1208 }
1209 
1210 #if defined(PMAP_VIRTUAL_CACHE_ALIASES) && !defined(PMAP_NO_PV_UNCACHED)
1211 /*
1212  *	pmap_page_cache:
1213  *
1214  *	Change all mappings of a managed page to cached/uncached.
1215  */
1216 void
1217 pmap_page_cache(struct vm_page_md *mdpg, bool cached)
1218 {
1219 #ifdef UVMHIST
1220 	const bool vmpage_p = VM_PAGEMD_VMPAGE_P(mdpg);
1221 	struct vm_page * const pg = vmpage_p ? VM_MD_TO_PAGE(mdpg) : NULL;
1222 #endif
1223 
1224 	UVMHIST_FUNC(__func__);
1225 	UVMHIST_CALLARGS(pmaphist, "(mdpg=%#jx (pa %#jx) cached=%jd vmpage %jd)",
1226 	    (uintptr_t)mdpg, pg ? VM_PAGE_TO_PHYS(pg) : 0, cached, vmpage_p);
1227 
1228 	KASSERT(kpreempt_disabled());
1229 	KASSERT(VM_PAGEMD_PVLIST_LOCKED_P(mdpg));
1230 
1231 	if (cached) {
1232 		pmap_page_clear_attributes(mdpg, VM_PAGEMD_UNCACHED);
1233 		PMAP_COUNT(page_cache_restorations);
1234 	} else {
1235 		pmap_page_set_attributes(mdpg, VM_PAGEMD_UNCACHED);
1236 		PMAP_COUNT(page_cache_evictions);
1237 	}
1238 
1239 	for (pv_entry_t pv = &mdpg->mdpg_first; pv != NULL; pv = pv->pv_next) {
1240 		pmap_t pmap = pv->pv_pmap;
1241 		vaddr_t va = trunc_page(pv->pv_va);
1242 
1243 		KASSERT(pmap != NULL);
1244 		KASSERT(pmap != pmap_kernel() || !pmap_md_direct_mapped_vaddr_p(va));
1245 		pt_entry_t * const ptep = pmap_pte_lookup(pmap, va);
1246 		if (ptep == NULL)
1247 			continue;
1248 		pt_entry_t pte = *ptep;
1249 		if (pte_valid_p(pte)) {
1250 			pte = pte_cached_change(pte, cached);
1251 			pmap_tlb_miss_lock_enter();
1252 			pte_set(ptep, pte);
1253 			pmap_tlb_update_addr(pmap, va, pte, PMAP_TLB_NEED_IPI);
1254 			pmap_tlb_miss_lock_exit();
1255 		}
1256 	}
1257 
1258 	UVMHIST_LOG(pmaphist, " <-- done", 0, 0, 0, 0);
1259 }
1260 #endif	/* PMAP_VIRTUAL_CACHE_ALIASES && !PMAP_NO_PV_UNCACHED */
1261 
1262 /*
1263  *	Insert the given physical page (p) at
1264  *	the specified virtual address (v) in the
1265  *	target physical map with the protection requested.
1266  *
1267  *	If specified, the page will be wired down, meaning
1268  *	that the related pte can not be reclaimed.
1269  *
1270  *	NB:  This is the only routine which MAY NOT lazy-evaluate
1271  *	or lose information.  That is, this routine must actually
1272  *	insert this page into the given map NOW.
1273  */
1274 int
1275 pmap_enter(pmap_t pmap, vaddr_t va, paddr_t pa, vm_prot_t prot, u_int flags)
1276 {
1277 	const bool wired = (flags & PMAP_WIRED) != 0;
1278 	const bool is_kernel_pmap_p = (pmap == pmap_kernel());
1279 	u_int update_flags = (flags & VM_PROT_ALL) != 0 ? PMAP_TLB_INSERT : 0;
1280 #ifdef UVMHIST
1281 	struct kern_history * const histp =
1282 	    ((prot & VM_PROT_EXECUTE) ? &pmapexechist : &pmaphist);
1283 #endif
1284 
1285 	UVMHIST_FUNC(__func__);
1286 	UVMHIST_CALLARGS(*histp, "(pmap=%#jx, va=%#jx, pa=%#jx",
1287 	    (uintptr_t)pmap, va, pa, 0);
1288 	UVMHIST_LOG(*histp, "prot=%#jx flags=%#jx)", prot, flags, 0, 0);
1289 
1290 	const bool good_color = PMAP_PAGE_COLOROK_P(pa, va);
1291 	if (is_kernel_pmap_p) {
1292 		PMAP_COUNT(kernel_mappings);
1293 		if (!good_color)
1294 			PMAP_COUNT(kernel_mappings_bad);
1295 	} else {
1296 		PMAP_COUNT(user_mappings);
1297 		if (!good_color)
1298 			PMAP_COUNT(user_mappings_bad);
1299 	}
1300 	pmap_addr_range_check(pmap, va, va, __func__);
1301 
1302 	KASSERTMSG(prot & VM_PROT_READ, "no READ (%#x) in prot %#x",
1303 	    VM_PROT_READ, prot);
1304 
1305 	struct vm_page * const pg = PHYS_TO_VM_PAGE(pa);
1306 	struct vm_page_md * const mdpg = (pg ? VM_PAGE_TO_MD(pg) : NULL);
1307 
1308 	struct vm_page_md *mdpp = NULL;
1309 #ifdef __HAVE_PMAP_PV_TRACK
1310 	struct pmap_page *pp = pmap_pv_tracked(pa);
1311 	mdpp = pp ? PMAP_PAGE_TO_MD(pp) : NULL;
1312 #endif
1313 
1314 	if (mdpg) {
1315 		/* Set page referenced/modified status based on flags */
1316 		if (flags & VM_PROT_WRITE) {
1317 			pmap_page_set_attributes(mdpg, VM_PAGEMD_MODIFIED|VM_PAGEMD_REFERENCED);
1318 		} else if (flags & VM_PROT_ALL) {
1319 			pmap_page_set_attributes(mdpg, VM_PAGEMD_REFERENCED);
1320 		}
1321 
1322 #ifdef PMAP_VIRTUAL_CACHE_ALIASES
1323 		if (!VM_PAGEMD_CACHED_P(mdpg)) {
1324 			flags |= PMAP_NOCACHE;
1325 			PMAP_COUNT(uncached_mappings);
1326 		}
1327 #endif
1328 
1329 		PMAP_COUNT(managed_mappings);
1330 	} else if (mdpp) {
1331 #ifdef __HAVE_PMAP_PV_TRACK
1332 		pmap_page_set_attributes(mdpg, VM_PAGEMD_REFERENCED);
1333 
1334 		PMAP_COUNT(pvtracked_mappings);
1335 #endif
1336 	} else {
1337 		/*
1338 		 * Assumption: if it is not part of our managed memory
1339 		 * then it must be device memory which may be volatile.
1340 		 */
1341 		if ((flags & PMAP_CACHE_MASK) == 0)
1342 			flags |= PMAP_NOCACHE;
1343 		PMAP_COUNT(unmanaged_mappings);
1344 	}
1345 
1346 	KASSERTMSG(mdpg == NULL || mdpp == NULL, "mdpg %p mdpp %p", mdpg, mdpp);
1347 
1348 	struct vm_page_md *md = (mdpg != NULL) ? mdpg : mdpp;
1349 	pt_entry_t npte = pte_make_enter(pa, md, prot, flags,
1350 	    is_kernel_pmap_p);
1351 
1352 	kpreempt_disable();
1353 
1354 	pt_entry_t * const ptep = pmap_pte_reserve(pmap, va, flags);
1355 	if (__predict_false(ptep == NULL)) {
1356 		kpreempt_enable();
1357 		UVMHIST_LOG(*histp, " <-- ENOMEM", 0, 0, 0, 0);
1358 		return ENOMEM;
1359 	}
1360 	const pt_entry_t opte = *ptep;
1361 	const bool resident = pte_valid_p(opte);
1362 	bool remap = false;
1363 	if (resident) {
1364 		if (pte_to_paddr(opte) != pa) {
1365 			KASSERT(!is_kernel_pmap_p);
1366 		    	const pt_entry_t rpte = pte_nv_entry(false);
1367 
1368 			pmap_addr_range_check(pmap, va, va + NBPG, __func__);
1369 			pmap_pte_process(pmap, va, va + NBPG, pmap_pte_remove,
1370 			    rpte);
1371 			PMAP_COUNT(user_mappings_changed);
1372 			remap = true;
1373 		}
1374 		update_flags |= PMAP_TLB_NEED_IPI;
1375 	}
1376 
1377 	if (!resident || remap) {
1378 		pmap->pm_stats.resident_count++;
1379 	}
1380 
1381 	/* Done after case that may sleep/return. */
1382 	if (md)
1383 		pmap_enter_pv(pmap, va, pa, md, &npte, 0);
1384 
1385 	/*
1386 	 * Now validate mapping with desired protection/wiring.
1387 	 */
1388 	if (wired) {
1389 		pmap->pm_stats.wired_count++;
1390 		npte = pte_wire_entry(npte);
1391 	}
1392 
1393 	UVMHIST_LOG(*histp, "new pte %#jx (pa %#jx)",
1394 	    pte_value(npte), pa, 0, 0);
1395 
1396 	KASSERT(pte_valid_p(npte));
1397 
1398 	pmap_tlb_miss_lock_enter();
1399 	pte_set(ptep, npte);
1400 	pmap_tlb_update_addr(pmap, va, npte, update_flags);
1401 	pmap_tlb_miss_lock_exit();
1402 	kpreempt_enable();
1403 
1404 	if (pg != NULL && (prot == (VM_PROT_READ | VM_PROT_EXECUTE))) {
1405 		KASSERT(mdpg != NULL);
1406 		PMAP_COUNT(exec_mappings);
1407 		if (!VM_PAGEMD_EXECPAGE_P(mdpg) && pte_cached_p(npte)) {
1408 			if (!pte_deferred_exec_p(npte)) {
1409 				UVMHIST_LOG(*histp, "va=%#jx pg %#jx: "
1410 				    "immediate syncicache",
1411 				    va, (uintptr_t)pg, 0, 0);
1412 				pmap_page_syncicache(pg);
1413 				pmap_page_set_attributes(mdpg,
1414 				    VM_PAGEMD_EXECPAGE);
1415 				PMAP_COUNT(exec_synced_mappings);
1416 			} else {
1417 				UVMHIST_LOG(*histp, "va=%#jx pg %#jx: defer "
1418 				    "syncicache: pte %#jx",
1419 				    va, (uintptr_t)pg, npte, 0);
1420 			}
1421 		} else {
1422 			UVMHIST_LOG(*histp,
1423 			    "va=%#jx pg %#jx: no syncicache cached %jd",
1424 			    va, (uintptr_t)pg, pte_cached_p(npte), 0);
1425 		}
1426 	} else if (pg != NULL && (prot & VM_PROT_EXECUTE)) {
1427 		KASSERT(mdpg != NULL);
1428 		KASSERT(prot & VM_PROT_WRITE);
1429 		PMAP_COUNT(exec_mappings);
1430 		pmap_page_syncicache(pg);
1431 		pmap_page_clear_attributes(mdpg, VM_PAGEMD_EXECPAGE);
1432 		UVMHIST_LOG(*histp,
1433 		    "va=%#jx pg %#jx: immediate syncicache (writeable)",
1434 		    va, (uintptr_t)pg, 0, 0);
1435 	}
1436 
1437 	UVMHIST_LOG(*histp, " <-- 0 (OK)", 0, 0, 0, 0);
1438 	return 0;
1439 }
1440 
1441 void
1442 pmap_kenter_pa(vaddr_t va, paddr_t pa, vm_prot_t prot, u_int flags)
1443 {
1444 	pmap_t pmap = pmap_kernel();
1445 	struct vm_page * const pg = PHYS_TO_VM_PAGE(pa);
1446 	struct vm_page_md * const mdpg = (pg ? VM_PAGE_TO_MD(pg) : NULL);
1447 
1448 	UVMHIST_FUNC(__func__);
1449 	UVMHIST_CALLARGS(pmaphist, "(va=%#jx pa=%#jx prot=%ju, flags=%#jx)",
1450 	    va, pa, prot, flags);
1451 	PMAP_COUNT(kenter_pa);
1452 
1453 	if (mdpg == NULL) {
1454 		PMAP_COUNT(kenter_pa_unmanaged);
1455 		if ((flags & PMAP_CACHE_MASK) == 0)
1456 			flags |= PMAP_NOCACHE;
1457 	} else {
1458 		if ((flags & PMAP_NOCACHE) == 0 && !PMAP_PAGE_COLOROK_P(pa, va))
1459 			PMAP_COUNT(kenter_pa_bad);
1460 	}
1461 
1462 	pt_entry_t npte = pte_make_kenter_pa(pa, mdpg, prot, flags);
1463 	kpreempt_disable();
1464 	pt_entry_t * const ptep = pmap_pte_lookup(pmap, va);
1465 	KASSERTMSG(ptep != NULL, "%#"PRIxVADDR " %#"PRIxVADDR, va,
1466 	    pmap_limits.virtual_end);
1467 	KASSERT(!pte_valid_p(*ptep));
1468 
1469 	/*
1470 	 * No need to track non-managed pages or PMAP_KMPAGEs pages for aliases
1471 	 */
1472 #ifdef PMAP_VIRTUAL_CACHE_ALIASES
1473 	if (pg != NULL && (flags & PMAP_KMPAGE) == 0
1474 	    && pmap_md_virtual_cache_aliasing_p()) {
1475 		pmap_enter_pv(pmap, va, pa, mdpg, &npte, PV_KENTER);
1476 	}
1477 #endif
1478 
1479 	/*
1480 	 * We have the option to force this mapping into the TLB but we
1481 	 * don't.  Instead let the next reference to the page do it.
1482 	 */
1483 	pmap_tlb_miss_lock_enter();
1484 	pte_set(ptep, npte);
1485 	pmap_tlb_update_addr(pmap_kernel(), va, npte, 0);
1486 	pmap_tlb_miss_lock_exit();
1487 	kpreempt_enable();
1488 #if DEBUG > 1
1489 	for (u_int i = 0; i < PAGE_SIZE / sizeof(long); i++) {
1490 		if (((long *)va)[i] != ((long *)pa)[i])
1491 			panic("%s: contents (%lx) of va %#"PRIxVADDR
1492 			    " != contents (%lx) of pa %#"PRIxPADDR, __func__,
1493 			    ((long *)va)[i], va, ((long *)pa)[i], pa);
1494 	}
1495 #endif
1496 
1497 	UVMHIST_LOG(pmaphist, " <-- done (ptep=%#jx)", (uintptr_t)ptep, 0, 0,
1498 	    0);
1499 }
1500 
1501 /*
1502  *	Remove the given range of addresses from the kernel map.
1503  *
1504  *	It is assumed that the start and end are properly
1505  *	rounded to the page size.
1506  */
1507 
1508 static bool
1509 pmap_pte_kremove(pmap_t pmap, vaddr_t sva, vaddr_t eva, pt_entry_t *ptep,
1510 	uintptr_t flags)
1511 {
1512 	const pt_entry_t new_pte = pte_nv_entry(true);
1513 
1514 	UVMHIST_FUNC(__func__);
1515 	UVMHIST_CALLARGS(pmaphist, "(pmap=%#jx, sva=%#jx eva=%#jx ptep=%#jx)",
1516 	    (uintptr_t)pmap, sva, eva, (uintptr_t)ptep);
1517 
1518 	KASSERT(kpreempt_disabled());
1519 
1520 	for (; sva < eva; sva += NBPG, ptep++) {
1521 		pt_entry_t pte = *ptep;
1522 		if (!pte_valid_p(pte))
1523 			continue;
1524 
1525 		PMAP_COUNT(kremove_pages);
1526 #ifdef PMAP_VIRTUAL_CACHE_ALIASES
1527 		struct vm_page * const pg = PHYS_TO_VM_PAGE(pte_to_paddr(pte));
1528 		if (pg != NULL && pmap_md_virtual_cache_aliasing_p()) {
1529 			pmap_remove_pv(pmap, sva, pg, !pte_readonly_p(pte));
1530 		}
1531 #endif
1532 
1533 		pmap_tlb_miss_lock_enter();
1534 		pte_set(ptep, new_pte);
1535 		pmap_tlb_invalidate_addr(pmap, sva);
1536 		pmap_tlb_miss_lock_exit();
1537 	}
1538 
1539 	UVMHIST_LOG(pmaphist, " <-- done", 0, 0, 0, 0);
1540 
1541 	return false;
1542 }
1543 
1544 void
1545 pmap_kremove(vaddr_t va, vsize_t len)
1546 {
1547 	const vaddr_t sva = trunc_page(va);
1548 	const vaddr_t eva = round_page(va + len);
1549 
1550 	UVMHIST_FUNC(__func__);
1551 	UVMHIST_CALLARGS(pmaphist, "(va=%#jx len=%#jx)", va, len, 0, 0);
1552 
1553 	kpreempt_disable();
1554 	pmap_pte_process(pmap_kernel(), sva, eva, pmap_pte_kremove, 0);
1555 	kpreempt_enable();
1556 
1557 	UVMHIST_LOG(pmaphist, " <-- done", 0, 0, 0, 0);
1558 }
1559 
1560 bool
1561 pmap_remove_all(struct pmap *pmap)
1562 {
1563 	UVMHIST_FUNC(__func__);
1564 	UVMHIST_CALLARGS(pmaphist, "(pm=%#jx)", (uintptr_t)pmap, 0, 0, 0);
1565 
1566 	KASSERT(pmap != pmap_kernel());
1567 
1568 	kpreempt_disable();
1569 	/*
1570 	 * Free all of our ASIDs which means we can skip doing all the
1571 	 * tlb_invalidate_addrs().
1572 	 */
1573 	pmap_tlb_miss_lock_enter();
1574 #ifdef MULTIPROCESSOR
1575 	// This should be the last CPU with this pmap onproc
1576 	KASSERT(!kcpuset_isotherset(pmap->pm_onproc, cpu_index(curcpu())));
1577 	if (kcpuset_isset(pmap->pm_onproc, cpu_index(curcpu())))
1578 #endif
1579 		pmap_tlb_asid_deactivate(pmap);
1580 #ifdef MULTIPROCESSOR
1581 	KASSERT(kcpuset_iszero(pmap->pm_onproc));
1582 #endif
1583 	pmap_tlb_asid_release_all(pmap);
1584 	pmap_tlb_miss_lock_exit();
1585 	pmap->pm_flags |= PMAP_DEFERRED_ACTIVATE;
1586 
1587 #ifdef PMAP_FAULTINFO
1588 	curpcb->pcb_faultinfo.pfi_faultaddr = 0;
1589 	curpcb->pcb_faultinfo.pfi_repeats = 0;
1590 	curpcb->pcb_faultinfo.pfi_faultptep = NULL;
1591 #endif
1592 	kpreempt_enable();
1593 
1594 	UVMHIST_LOG(pmaphist, " <-- done", 0, 0, 0, 0);
1595 	return false;
1596 }
1597 
1598 /*
1599  *	Routine:	pmap_unwire
1600  *	Function:	Clear the wired attribute for a map/virtual-address
1601  *			pair.
1602  *	In/out conditions:
1603  *			The mapping must already exist in the pmap.
1604  */
1605 void
1606 pmap_unwire(pmap_t pmap, vaddr_t va)
1607 {
1608 	UVMHIST_FUNC(__func__);
1609 	UVMHIST_CALLARGS(pmaphist, "(pmap=%#jx, va=%#jx)", (uintptr_t)pmap, va,
1610 	    0, 0);
1611 	PMAP_COUNT(unwire);
1612 
1613 	/*
1614 	 * Don't need to flush the TLB since PG_WIRED is only in software.
1615 	 */
1616 	kpreempt_disable();
1617 	pmap_addr_range_check(pmap, va, va, __func__);
1618 	pt_entry_t * const ptep = pmap_pte_lookup(pmap, va);
1619 	KASSERTMSG(ptep != NULL, "pmap %p va %#"PRIxVADDR" invalid STE",
1620 	    pmap, va);
1621 	pt_entry_t pte = *ptep;
1622 	KASSERTMSG(pte_valid_p(pte),
1623 	    "pmap %p va %#"PRIxVADDR" invalid PTE %#"PRIxPTE" @ %p",
1624 	    pmap, va, pte_value(pte), ptep);
1625 
1626 	if (pte_wired_p(pte)) {
1627 		pmap_tlb_miss_lock_enter();
1628 		pte_set(ptep, pte_unwire_entry(pte));
1629 		pmap_tlb_miss_lock_exit();
1630 		pmap->pm_stats.wired_count--;
1631 	}
1632 #ifdef DIAGNOSTIC
1633 	else {
1634 		printf("%s: wiring for pmap %p va %#"PRIxVADDR" unchanged!\n",
1635 		    __func__, pmap, va);
1636 	}
1637 #endif
1638 	kpreempt_enable();
1639 
1640 	UVMHIST_LOG(pmaphist, " <-- done", 0, 0, 0, 0);
1641 }
1642 
1643 /*
1644  *	Routine:	pmap_extract
1645  *	Function:
1646  *		Extract the physical page address associated
1647  *		with the given map/virtual_address pair.
1648  */
1649 bool
1650 pmap_extract(pmap_t pmap, vaddr_t va, paddr_t *pap)
1651 {
1652 	paddr_t pa;
1653 
1654 	if (pmap == pmap_kernel()) {
1655 		if (pmap_md_direct_mapped_vaddr_p(va)) {
1656 			pa = pmap_md_direct_mapped_vaddr_to_paddr(va);
1657 			goto done;
1658 		}
1659 		if (pmap_md_io_vaddr_p(va))
1660 			panic("pmap_extract: io address %#"PRIxVADDR"", va);
1661 
1662 		if (va >= pmap_limits.virtual_end)
1663 			panic("%s: illegal kernel mapped address %#"PRIxVADDR,
1664 			    __func__, va);
1665 	}
1666 	kpreempt_disable();
1667 	const pt_entry_t * const ptep = pmap_pte_lookup(pmap, va);
1668 	if (ptep == NULL || !pte_valid_p(*ptep)) {
1669 		kpreempt_enable();
1670 		return false;
1671 	}
1672 	pa = pte_to_paddr(*ptep) | (va & PGOFSET);
1673 	kpreempt_enable();
1674 done:
1675 	if (pap != NULL) {
1676 		*pap = pa;
1677 	}
1678 	return true;
1679 }
1680 
1681 /*
1682  *	Copy the range specified by src_addr/len
1683  *	from the source map to the range dst_addr/len
1684  *	in the destination map.
1685  *
1686  *	This routine is only advisory and need not do anything.
1687  */
1688 void
1689 pmap_copy(pmap_t dst_pmap, pmap_t src_pmap, vaddr_t dst_addr, vsize_t len,
1690     vaddr_t src_addr)
1691 {
1692 	UVMHIST_FUNC(__func__);
1693 	UVMHIST_CALLED(pmaphist);
1694 	PMAP_COUNT(copy);
1695 }
1696 
1697 /*
1698  *	pmap_clear_reference:
1699  *
1700  *	Clear the reference bit on the specified physical page.
1701  */
1702 bool
1703 pmap_clear_reference(struct vm_page *pg)
1704 {
1705 	struct vm_page_md * const mdpg = VM_PAGE_TO_MD(pg);
1706 
1707 	UVMHIST_FUNC(__func__);
1708 	UVMHIST_CALLARGS(pmaphist, "(pg=%#jx (pa %#jx))",
1709 	   (uintptr_t)pg, VM_PAGE_TO_PHYS(pg), 0,0);
1710 
1711 	bool rv = pmap_page_clear_attributes(mdpg, VM_PAGEMD_REFERENCED);
1712 
1713 	UVMHIST_LOG(pmaphist, " <-- wasref %ju", rv, 0, 0, 0);
1714 
1715 	return rv;
1716 }
1717 
1718 /*
1719  *	pmap_is_referenced:
1720  *
1721  *	Return whether or not the specified physical page is referenced
1722  *	by any physical maps.
1723  */
1724 bool
1725 pmap_is_referenced(struct vm_page *pg)
1726 {
1727 	return VM_PAGEMD_REFERENCED_P(VM_PAGE_TO_MD(pg));
1728 }
1729 
1730 /*
1731  *	Clear the modify bits on the specified physical page.
1732  */
1733 bool
1734 pmap_clear_modify(struct vm_page *pg)
1735 {
1736 	struct vm_page_md * const mdpg = VM_PAGE_TO_MD(pg);
1737 	pv_entry_t pv = &mdpg->mdpg_first;
1738 	pv_entry_t pv_next;
1739 
1740 	UVMHIST_FUNC(__func__);
1741 	UVMHIST_CALLARGS(pmaphist, "(pg=%#jx (%#jx))",
1742 	    (uintptr_t)pg, VM_PAGE_TO_PHYS(pg), 0,0);
1743 	PMAP_COUNT(clear_modify);
1744 
1745 	if (VM_PAGEMD_EXECPAGE_P(mdpg)) {
1746 		if (pv->pv_pmap == NULL) {
1747 			UVMHIST_LOG(pmapexechist,
1748 			    "pg %#jx (pa %#jx): execpage cleared",
1749 			    (uintptr_t)pg, VM_PAGE_TO_PHYS(pg), 0, 0);
1750 			pmap_page_clear_attributes(mdpg, VM_PAGEMD_EXECPAGE);
1751 			PMAP_COUNT(exec_uncached_clear_modify);
1752 		} else {
1753 			UVMHIST_LOG(pmapexechist,
1754 			    "pg %#jx (pa %#jx): syncicache performed",
1755 			    (uintptr_t)pg, VM_PAGE_TO_PHYS(pg), 0, 0);
1756 			pmap_page_syncicache(pg);
1757 			PMAP_COUNT(exec_synced_clear_modify);
1758 		}
1759 	}
1760 	if (!pmap_page_clear_attributes(mdpg, VM_PAGEMD_MODIFIED)) {
1761 		UVMHIST_LOG(pmaphist, " <-- false", 0, 0, 0, 0);
1762 		return false;
1763 	}
1764 	if (pv->pv_pmap == NULL) {
1765 		UVMHIST_LOG(pmaphist, " <-- true (no mappings)", 0, 0, 0, 0);
1766 		return true;
1767 	}
1768 
1769 	/*
1770 	 * remove write access from any pages that are dirty
1771 	 * so we can tell if they are written to again later.
1772 	 * flush the VAC first if there is one.
1773 	 */
1774 	kpreempt_disable();
1775 	VM_PAGEMD_PVLIST_READLOCK(mdpg);
1776 	pmap_pvlist_check(mdpg);
1777 	for (; pv != NULL; pv = pv_next) {
1778 		pmap_t pmap = pv->pv_pmap;
1779 		vaddr_t va = trunc_page(pv->pv_va);
1780 
1781 		pv_next = pv->pv_next;
1782 #ifdef PMAP_VIRTUAL_CACHE_ALIASES
1783 		if (PV_ISKENTER_P(pv))
1784 			continue;
1785 #endif
1786 		pt_entry_t * const ptep = pmap_pte_lookup(pmap, va);
1787 		KASSERT(ptep);
1788 		pt_entry_t pte = pte_prot_nowrite(*ptep);
1789 		if (*ptep == pte) {
1790 			continue;
1791 		}
1792 		KASSERT(pte_valid_p(pte));
1793 		const uintptr_t gen = VM_PAGEMD_PVLIST_UNLOCK(mdpg);
1794 		pmap_tlb_miss_lock_enter();
1795 		pte_set(ptep, pte);
1796 		pmap_tlb_invalidate_addr(pmap, va);
1797 		pmap_tlb_miss_lock_exit();
1798 		pmap_update(pmap);
1799 		if (__predict_false(gen != VM_PAGEMD_PVLIST_READLOCK(mdpg))) {
1800 			/*
1801 			 * The list changed!  So restart from the beginning.
1802 			 */
1803 			pv_next = &mdpg->mdpg_first;
1804 			pmap_pvlist_check(mdpg);
1805 		}
1806 	}
1807 	pmap_pvlist_check(mdpg);
1808 	VM_PAGEMD_PVLIST_UNLOCK(mdpg);
1809 	kpreempt_enable();
1810 
1811 	UVMHIST_LOG(pmaphist, " <-- true (mappings changed)", 0, 0, 0, 0);
1812 	return true;
1813 }
1814 
1815 /*
1816  *	pmap_is_modified:
1817  *
1818  *	Return whether or not the specified physical page is modified
1819  *	by any physical maps.
1820  */
1821 bool
1822 pmap_is_modified(struct vm_page *pg)
1823 {
1824 	return VM_PAGEMD_MODIFIED_P(VM_PAGE_TO_MD(pg));
1825 }
1826 
1827 /*
1828  *	pmap_set_modified:
1829  *
1830  *	Sets the page modified reference bit for the specified page.
1831  */
1832 void
1833 pmap_set_modified(paddr_t pa)
1834 {
1835 	struct vm_page * const pg = PHYS_TO_VM_PAGE(pa);
1836 	struct vm_page_md * const mdpg = VM_PAGE_TO_MD(pg);
1837 	pmap_page_set_attributes(mdpg, VM_PAGEMD_MODIFIED|VM_PAGEMD_REFERENCED);
1838 }
1839 
1840 /******************** pv_entry management ********************/
1841 
1842 static void
1843 pmap_pvlist_check(struct vm_page_md *mdpg)
1844 {
1845 #ifdef DEBUG
1846 	pv_entry_t pv = &mdpg->mdpg_first;
1847 	if (pv->pv_pmap != NULL) {
1848 #ifdef PMAP_VIRTUAL_CACHE_ALIASES
1849 		const u_int colormask = uvmexp.colormask;
1850 		u_int colors = 0;
1851 #endif
1852 		for (; pv != NULL; pv = pv->pv_next) {
1853 			KASSERT(pv->pv_pmap != pmap_kernel() || !pmap_md_direct_mapped_vaddr_p(pv->pv_va));
1854 #ifdef PMAP_VIRTUAL_CACHE_ALIASES
1855 			colors |= __BIT(atop(pv->pv_va) & colormask);
1856 #endif
1857 		}
1858 #ifdef PMAP_VIRTUAL_CACHE_ALIASES
1859 		// Assert that if there is more than 1 color mapped, that the
1860 		// page is uncached.
1861 		KASSERTMSG(!pmap_md_virtual_cache_aliasing_p()
1862 		    || colors == 0 || (colors & (colors-1)) == 0
1863 		    || VM_PAGEMD_UNCACHED_P(mdpg), "colors=%#x uncached=%u",
1864 		    colors, VM_PAGEMD_UNCACHED_P(mdpg));
1865 #endif
1866 	} else {
1867     		KASSERT(pv->pv_next == NULL);
1868 	}
1869 #endif /* DEBUG */
1870 }
1871 
1872 /*
1873  * Enter the pmap and virtual address into the
1874  * physical to virtual map table.
1875  */
1876 void
1877 pmap_enter_pv(pmap_t pmap, vaddr_t va, paddr_t pa, struct vm_page_md *mdpg,
1878     pt_entry_t *nptep, u_int flags)
1879 {
1880 	pv_entry_t pv, npv, apv;
1881 #ifdef UVMHIST
1882 	bool first = false;
1883 	struct vm_page *pg = VM_PAGEMD_VMPAGE_P(mdpg) ? VM_MD_TO_PAGE(mdpg) :
1884 	    NULL;
1885 #endif
1886 
1887 	UVMHIST_FUNC(__func__);
1888 	UVMHIST_CALLARGS(pmaphist, "(pmap=%#jx va=%#jx pg=%#jx (%#jx)",
1889 	    (uintptr_t)pmap, va, (uintptr_t)pg, pa);
1890 	UVMHIST_LOG(pmaphist, "nptep=%#jx (%#jx))",
1891 	    (uintptr_t)nptep, pte_value(*nptep), 0, 0);
1892 
1893 	KASSERT(kpreempt_disabled());
1894 	KASSERT(pmap != pmap_kernel() || !pmap_md_direct_mapped_vaddr_p(va));
1895 	KASSERTMSG(pmap != pmap_kernel() || !pmap_md_io_vaddr_p(va),
1896 	    "va %#"PRIxVADDR, va);
1897 
1898 	apv = NULL;
1899 	VM_PAGEMD_PVLIST_LOCK(mdpg);
1900 again:
1901 	pv = &mdpg->mdpg_first;
1902 	pmap_pvlist_check(mdpg);
1903 	if (pv->pv_pmap == NULL) {
1904 		KASSERT(pv->pv_next == NULL);
1905 		/*
1906 		 * No entries yet, use header as the first entry
1907 		 */
1908 		PMAP_COUNT(primary_mappings);
1909 		PMAP_COUNT(mappings);
1910 #ifdef UVMHIST
1911 		first = true;
1912 #endif
1913 #ifdef PMAP_VIRTUAL_CACHE_ALIASES
1914 		KASSERT(VM_PAGEMD_CACHED_P(mdpg));
1915 		// If the new mapping has an incompatible color the last
1916 		// mapping of this page, clean the page before using it.
1917 		if (!PMAP_PAGE_COLOROK_P(va, pv->pv_va)) {
1918 			pmap_md_vca_clean(mdpg, PMAP_WBINV);
1919 		}
1920 #endif
1921 		pv->pv_pmap = pmap;
1922 		pv->pv_va = va | flags;
1923 	} else {
1924 #ifdef PMAP_VIRTUAL_CACHE_ALIASES
1925 		if (pmap_md_vca_add(mdpg, va, nptep)) {
1926 			goto again;
1927 		}
1928 #endif
1929 
1930 		/*
1931 		 * There is at least one other VA mapping this page.
1932 		 * Place this entry after the header.
1933 		 *
1934 		 * Note: the entry may already be in the table if
1935 		 * we are only changing the protection bits.
1936 		 */
1937 
1938 		for (npv = pv; npv; npv = npv->pv_next) {
1939 			if (pmap == npv->pv_pmap
1940 			    && va == trunc_page(npv->pv_va)) {
1941 #ifdef PARANOIADIAG
1942 				pt_entry_t *ptep = pmap_pte_lookup(pmap, va);
1943 				pt_entry_t pte = (ptep != NULL) ? *ptep : 0;
1944 				if (!pte_valid_p(pte) || pte_to_paddr(pte) != pa)
1945 					printf("%s: found va %#"PRIxVADDR
1946 					    " pa %#"PRIxPADDR
1947 					    " in pv_table but != %#"PRIxPTE"\n",
1948 					    __func__, va, pa, pte_value(pte));
1949 #endif
1950 				PMAP_COUNT(remappings);
1951 				VM_PAGEMD_PVLIST_UNLOCK(mdpg);
1952 				if (__predict_false(apv != NULL))
1953 					pmap_pv_free(apv);
1954 
1955 				UVMHIST_LOG(pmaphist,
1956 				    " <-- done pv=%#jx (reused)",
1957 				    (uintptr_t)pv, 0, 0, 0);
1958 				return;
1959 			}
1960 		}
1961 		if (__predict_true(apv == NULL)) {
1962 			/*
1963 			 * To allocate a PV, we have to release the PVLIST lock
1964 			 * so get the page generation.  We allocate the PV, and
1965 			 * then reacquire the lock.
1966 			 */
1967 			pmap_pvlist_check(mdpg);
1968 			const uintptr_t gen = VM_PAGEMD_PVLIST_UNLOCK(mdpg);
1969 
1970 			apv = (pv_entry_t)pmap_pv_alloc();
1971 			if (apv == NULL)
1972 				panic("pmap_enter_pv: pmap_pv_alloc() failed");
1973 
1974 			/*
1975 			 * If the generation has changed, then someone else
1976 			 * tinkered with this page so we should start over.
1977 			 */
1978 			if (gen != VM_PAGEMD_PVLIST_LOCK(mdpg))
1979 				goto again;
1980 		}
1981 		npv = apv;
1982 		apv = NULL;
1983 #ifdef PMAP_VIRTUAL_CACHE_ALIASES
1984 		/*
1985 		 * If need to deal with virtual cache aliases, keep mappings
1986 		 * in the kernel pmap at the head of the list.  This allows
1987 		 * the VCA code to easily use them for cache operations if
1988 		 * present.
1989 		 */
1990 		pmap_t kpmap = pmap_kernel();
1991 		if (pmap != kpmap) {
1992 			while (pv->pv_pmap == kpmap && pv->pv_next != NULL) {
1993 				pv = pv->pv_next;
1994 			}
1995 		}
1996 #endif
1997 		npv->pv_va = va | flags;
1998 		npv->pv_pmap = pmap;
1999 		npv->pv_next = pv->pv_next;
2000 		pv->pv_next = npv;
2001 		PMAP_COUNT(mappings);
2002 	}
2003 	pmap_pvlist_check(mdpg);
2004 	VM_PAGEMD_PVLIST_UNLOCK(mdpg);
2005 	if (__predict_false(apv != NULL))
2006 		pmap_pv_free(apv);
2007 
2008 	UVMHIST_LOG(pmaphist, " <-- done pv=%#jx (first %ju)", (uintptr_t)pv,
2009 	    first, 0, 0);
2010 }
2011 
2012 /*
2013  * Remove a physical to virtual address translation.
2014  * If cache was inhibited on this page, and there are no more cache
2015  * conflicts, restore caching.
2016  * Flush the cache if the last page is removed (should always be cached
2017  * at this point).
2018  */
2019 void
2020 pmap_remove_pv(pmap_t pmap, vaddr_t va, struct vm_page *pg, bool dirty)
2021 {
2022 	struct vm_page_md * const mdpg = VM_PAGE_TO_MD(pg);
2023 	pv_entry_t pv, npv;
2024 	bool last;
2025 
2026 	UVMHIST_FUNC(__func__);
2027 	UVMHIST_CALLARGS(pmaphist, "(pmap=%#jx, va=%#jx, pg=%#jx (pa %#jx)",
2028 	    (uintptr_t)pmap, va, (uintptr_t)pg, VM_PAGE_TO_PHYS(pg));
2029 	UVMHIST_LOG(pmaphist, "dirty=%ju)", dirty, 0, 0, 0);
2030 
2031 	KASSERT(kpreempt_disabled());
2032 	KASSERT((va & PAGE_MASK) == 0);
2033 	pv = &mdpg->mdpg_first;
2034 
2035 	VM_PAGEMD_PVLIST_LOCK(mdpg);
2036 	pmap_pvlist_check(mdpg);
2037 
2038 	/*
2039 	 * If it is the first entry on the list, it is actually
2040 	 * in the header and we must copy the following entry up
2041 	 * to the header.  Otherwise we must search the list for
2042 	 * the entry.  In either case we free the now unused entry.
2043 	 */
2044 
2045 	last = false;
2046 	if (pmap == pv->pv_pmap && va == trunc_page(pv->pv_va)) {
2047 		npv = pv->pv_next;
2048 		if (npv) {
2049 			*pv = *npv;
2050 			KASSERT(pv->pv_pmap != NULL);
2051 		} else {
2052 #ifdef PMAP_VIRTUAL_CACHE_ALIASES
2053 			pmap_page_clear_attributes(mdpg, VM_PAGEMD_UNCACHED);
2054 #endif
2055 			pv->pv_pmap = NULL;
2056 			last = true;	/* Last mapping removed */
2057 		}
2058 		PMAP_COUNT(remove_pvfirst);
2059 	} else {
2060 		for (npv = pv->pv_next; npv; pv = npv, npv = npv->pv_next) {
2061 			PMAP_COUNT(remove_pvsearch);
2062 			if (pmap == npv->pv_pmap && va == trunc_page(npv->pv_va))
2063 				break;
2064 		}
2065 		if (npv) {
2066 			pv->pv_next = npv->pv_next;
2067 		}
2068 	}
2069 
2070 	pmap_pvlist_check(mdpg);
2071 	VM_PAGEMD_PVLIST_UNLOCK(mdpg);
2072 
2073 #ifdef PMAP_VIRTUAL_CACHE_ALIASES
2074 	pmap_md_vca_remove(pg, va, dirty, last);
2075 #endif
2076 
2077 	/*
2078 	 * Free the pv_entry if needed.
2079 	 */
2080 	if (npv)
2081 		pmap_pv_free(npv);
2082 	if (VM_PAGEMD_EXECPAGE_P(mdpg) && dirty) {
2083 		if (last) {
2084 			/*
2085 			 * If this was the page's last mapping, we no longer
2086 			 * care about its execness.
2087 			 */
2088 			UVMHIST_LOG(pmapexechist,
2089 			    "pg %#jx (pa %#jx)last %ju: execpage cleared",
2090 			    (uintptr_t)pg, VM_PAGE_TO_PHYS(pg), last, 0);
2091 			pmap_page_clear_attributes(mdpg, VM_PAGEMD_EXECPAGE);
2092 			PMAP_COUNT(exec_uncached_remove);
2093 		} else {
2094 			/*
2095 			 * Someone still has it mapped as an executable page
2096 			 * so we must sync it.
2097 			 */
2098 			UVMHIST_LOG(pmapexechist,
2099 			    "pg %#jx (pa %#jx) last %ju: performed syncicache",
2100 			    (uintptr_t)pg, VM_PAGE_TO_PHYS(pg), last, 0);
2101 			pmap_page_syncicache(pg);
2102 			PMAP_COUNT(exec_synced_remove);
2103 		}
2104 	}
2105 
2106 	UVMHIST_LOG(pmaphist, " <-- done", 0, 0, 0, 0);
2107 }
2108 
2109 #if defined(MULTIPROCESSOR)
2110 struct pmap_pvlist_info {
2111 	kmutex_t *pli_locks[PAGE_SIZE / 32];
2112 	volatile u_int pli_lock_refs[PAGE_SIZE / 32];
2113 	volatile u_int pli_lock_index;
2114 	u_int pli_lock_mask;
2115 } pmap_pvlist_info;
2116 
2117 void
2118 pmap_pvlist_lock_init(size_t cache_line_size)
2119 {
2120 	struct pmap_pvlist_info * const pli = &pmap_pvlist_info;
2121 	const vaddr_t lock_page = uvm_pageboot_alloc(PAGE_SIZE);
2122 	vaddr_t lock_va = lock_page;
2123 	if (sizeof(kmutex_t) > cache_line_size) {
2124 		cache_line_size = roundup2(sizeof(kmutex_t), cache_line_size);
2125 	}
2126 	const size_t nlocks = PAGE_SIZE / cache_line_size;
2127 	KASSERT((nlocks & (nlocks - 1)) == 0);
2128 	/*
2129 	 * Now divide the page into a number of mutexes, one per cacheline.
2130 	 */
2131 	for (size_t i = 0; i < nlocks; lock_va += cache_line_size, i++) {
2132 		kmutex_t * const lock = (kmutex_t *)lock_va;
2133 		mutex_init(lock, MUTEX_DEFAULT, IPL_HIGH);
2134 		pli->pli_locks[i] = lock;
2135 	}
2136 	pli->pli_lock_mask = nlocks - 1;
2137 }
2138 
2139 kmutex_t *
2140 pmap_pvlist_lock_addr(struct vm_page_md *mdpg)
2141 {
2142 	struct pmap_pvlist_info * const pli = &pmap_pvlist_info;
2143 	kmutex_t *lock = mdpg->mdpg_lock;
2144 
2145 	/*
2146 	 * Allocate a lock on an as-needed basis.  This will hopefully give us
2147 	 * semi-random distribution not based on page color.
2148 	 */
2149 	if (__predict_false(lock == NULL)) {
2150 		size_t locknum = atomic_add_int_nv(&pli->pli_lock_index, 37);
2151 		size_t lockid = locknum & pli->pli_lock_mask;
2152 		kmutex_t * const new_lock = pli->pli_locks[lockid];
2153 		/*
2154 		 * Set the lock.  If some other thread already did, just use
2155 		 * the one they assigned.
2156 		 */
2157 		lock = atomic_cas_ptr(&mdpg->mdpg_lock, NULL, new_lock);
2158 		if (lock == NULL) {
2159 			lock = new_lock;
2160 			atomic_inc_uint(&pli->pli_lock_refs[lockid]);
2161 		}
2162 	}
2163 
2164 	/*
2165 	 * Now finally provide the lock.
2166 	 */
2167 	return lock;
2168 }
2169 #else /* !MULTIPROCESSOR */
2170 void
2171 pmap_pvlist_lock_init(size_t cache_line_size)
2172 {
2173 	mutex_init(&pmap_pvlist_mutex, MUTEX_DEFAULT, IPL_HIGH);
2174 }
2175 
2176 #ifdef MODULAR
2177 kmutex_t *
2178 pmap_pvlist_lock_addr(struct vm_page_md *mdpg)
2179 {
2180 	/*
2181 	 * We just use a global lock.
2182 	 */
2183 	if (__predict_false(mdpg->mdpg_lock == NULL)) {
2184 		mdpg->mdpg_lock = &pmap_pvlist_mutex;
2185 	}
2186 
2187 	/*
2188 	 * Now finally provide the lock.
2189 	 */
2190 	return mdpg->mdpg_lock;
2191 }
2192 #endif /* MODULAR */
2193 #endif /* !MULTIPROCESSOR */
2194 
2195 /*
2196  * pmap_pv_page_alloc:
2197  *
2198  *	Allocate a page for the pv_entry pool.
2199  */
2200 void *
2201 pmap_pv_page_alloc(struct pool *pp, int flags)
2202 {
2203 	struct vm_page * const pg = PMAP_ALLOC_POOLPAGE(UVM_PGA_USERESERVE);
2204 	if (pg == NULL)
2205 		return NULL;
2206 
2207 	return (void *)pmap_map_poolpage(VM_PAGE_TO_PHYS(pg));
2208 }
2209 
2210 /*
2211  * pmap_pv_page_free:
2212  *
2213  *	Free a pv_entry pool page.
2214  */
2215 void
2216 pmap_pv_page_free(struct pool *pp, void *v)
2217 {
2218 	vaddr_t va = (vaddr_t)v;
2219 
2220 	KASSERT(pmap_md_direct_mapped_vaddr_p(va));
2221 	const paddr_t pa = pmap_md_direct_mapped_vaddr_to_paddr(va);
2222 	struct vm_page * const pg = PHYS_TO_VM_PAGE(pa);
2223 	KASSERT(pg != NULL);
2224 #ifdef PMAP_VIRTUAL_CACHE_ALIASES
2225 	kpreempt_disable();
2226 	pmap_md_vca_remove(pg, va, true, true);
2227 	kpreempt_enable();
2228 #endif
2229 	pmap_page_clear_attributes(VM_PAGE_TO_MD(pg), VM_PAGEMD_POOLPAGE);
2230 	KASSERT(!VM_PAGEMD_EXECPAGE_P(VM_PAGE_TO_MD(pg)));
2231 	uvm_pagefree(pg);
2232 }
2233 
2234 #ifdef PMAP_PREFER
2235 /*
2236  * Find first virtual address >= *vap that doesn't cause
2237  * a cache alias conflict.
2238  */
2239 void
2240 pmap_prefer(vaddr_t foff, vaddr_t *vap, vsize_t sz, int td)
2241 {
2242 	vsize_t prefer_mask = ptoa(uvmexp.colormask);
2243 
2244 	PMAP_COUNT(prefer_requests);
2245 
2246 	prefer_mask |= pmap_md_cache_prefer_mask();
2247 
2248 	if (prefer_mask) {
2249 		vaddr_t	va = *vap;
2250 		vsize_t d = (foff - va) & prefer_mask;
2251 		if (d) {
2252 			if (td)
2253 				*vap = trunc_page(va - ((-d) & prefer_mask));
2254 			else
2255 				*vap = round_page(va + d);
2256 			PMAP_COUNT(prefer_adjustments);
2257 		}
2258 	}
2259 }
2260 #endif /* PMAP_PREFER */
2261 
2262 #ifdef PMAP_MAP_POOLPAGE
2263 vaddr_t
2264 pmap_map_poolpage(paddr_t pa)
2265 {
2266 	struct vm_page * const pg = PHYS_TO_VM_PAGE(pa);
2267 	KASSERT(pg);
2268 
2269 	struct vm_page_md * const mdpg = VM_PAGE_TO_MD(pg);
2270 	KASSERT(!VM_PAGEMD_EXECPAGE_P(mdpg));
2271 
2272 	pmap_page_set_attributes(mdpg, VM_PAGEMD_POOLPAGE);
2273 
2274 	return pmap_md_map_poolpage(pa, NBPG);
2275 }
2276 
2277 paddr_t
2278 pmap_unmap_poolpage(vaddr_t va)
2279 {
2280 	KASSERT(pmap_md_direct_mapped_vaddr_p(va));
2281 	paddr_t pa = pmap_md_direct_mapped_vaddr_to_paddr(va);
2282 
2283 	struct vm_page * const pg = PHYS_TO_VM_PAGE(pa);
2284 	KASSERT(pg != NULL);
2285 	KASSERT(!VM_PAGEMD_EXECPAGE_P(VM_PAGE_TO_MD(pg)));
2286 
2287 	pmap_page_clear_attributes(VM_PAGE_TO_MD(pg), VM_PAGEMD_POOLPAGE);
2288 	pmap_md_unmap_poolpage(va, NBPG);
2289 
2290 	return pa;
2291 }
2292 #endif /* PMAP_MAP_POOLPAGE */
2293