xref: /netbsd-src/sys/uvm/pmap/pmap.c (revision aef5eb5f59cdfe8314f1b5f78ac04eb144e44010)
1 /*	$NetBSD: pmap.c,v 1.67 2022/09/15 06:44:18 skrll 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 1.67 2022/09/15 06:44:18 skrll 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 #if defined(PMAP_PV_TRACK_ONLY_STUBS)
124 #undef	__HAVE_PMAP_PV_TRACK
125 #endif
126 
127 PMAP_COUNTER(remove_kernel_calls, "remove kernel calls");
128 PMAP_COUNTER(remove_kernel_pages, "kernel pages unmapped");
129 PMAP_COUNTER(remove_user_calls, "remove user calls");
130 PMAP_COUNTER(remove_user_pages, "user pages unmapped");
131 PMAP_COUNTER(remove_flushes, "remove cache flushes");
132 PMAP_COUNTER(remove_tlb_ops, "remove tlb ops");
133 PMAP_COUNTER(remove_pvfirst, "remove pv first");
134 PMAP_COUNTER(remove_pvsearch, "remove pv search");
135 
136 PMAP_COUNTER(prefer_requests, "prefer requests");
137 PMAP_COUNTER(prefer_adjustments, "prefer adjustments");
138 
139 PMAP_COUNTER(idlezeroed_pages, "pages idle zeroed");
140 
141 PMAP_COUNTER(kenter_pa, "kernel fast mapped pages");
142 PMAP_COUNTER(kenter_pa_bad, "kernel fast mapped pages (bad color)");
143 PMAP_COUNTER(kenter_pa_unmanaged, "kernel fast mapped unmanaged pages");
144 PMAP_COUNTER(kremove_pages, "kernel fast unmapped pages");
145 
146 PMAP_COUNTER(page_cache_evictions, "pages changed to uncacheable");
147 PMAP_COUNTER(page_cache_restorations, "pages changed to cacheable");
148 
149 PMAP_COUNTER(kernel_mappings_bad, "kernel pages mapped (bad color)");
150 PMAP_COUNTER(user_mappings_bad, "user pages mapped (bad color)");
151 PMAP_COUNTER(kernel_mappings, "kernel pages mapped");
152 PMAP_COUNTER(user_mappings, "user pages mapped");
153 PMAP_COUNTER(user_mappings_changed, "user mapping changed");
154 PMAP_COUNTER(kernel_mappings_changed, "kernel mapping changed");
155 PMAP_COUNTER(uncached_mappings, "uncached pages mapped");
156 PMAP_COUNTER(unmanaged_mappings, "unmanaged pages mapped");
157 PMAP_COUNTER(pvtracked_mappings, "pv-tracked unmanaged pages mapped");
158 PMAP_COUNTER(managed_mappings, "managed pages mapped");
159 PMAP_COUNTER(mappings, "pages mapped");
160 PMAP_COUNTER(remappings, "pages remapped");
161 PMAP_COUNTER(unmappings, "pages unmapped");
162 PMAP_COUNTER(primary_mappings, "page initial mappings");
163 PMAP_COUNTER(primary_unmappings, "page final unmappings");
164 PMAP_COUNTER(tlb_hit, "page mapping");
165 
166 PMAP_COUNTER(exec_mappings, "exec pages mapped");
167 PMAP_COUNTER(exec_synced_mappings, "exec pages synced");
168 PMAP_COUNTER(exec_synced_remove, "exec pages synced (PR)");
169 PMAP_COUNTER(exec_synced_clear_modify, "exec pages synced (CM)");
170 PMAP_COUNTER(exec_synced_page_protect, "exec pages synced (PP)");
171 PMAP_COUNTER(exec_synced_protect, "exec pages synced (P)");
172 PMAP_COUNTER(exec_uncached_page_protect, "exec pages uncached (PP)");
173 PMAP_COUNTER(exec_uncached_clear_modify, "exec pages uncached (CM)");
174 PMAP_COUNTER(exec_uncached_zero_page, "exec pages uncached (ZP)");
175 PMAP_COUNTER(exec_uncached_copy_page, "exec pages uncached (CP)");
176 PMAP_COUNTER(exec_uncached_remove, "exec pages uncached (PR)");
177 
178 PMAP_COUNTER(create, "creates");
179 PMAP_COUNTER(reference, "references");
180 PMAP_COUNTER(dereference, "dereferences");
181 PMAP_COUNTER(destroy, "destroyed");
182 PMAP_COUNTER(activate, "activations");
183 PMAP_COUNTER(deactivate, "deactivations");
184 PMAP_COUNTER(update, "updates");
185 #ifdef MULTIPROCESSOR
186 PMAP_COUNTER(shootdown_ipis, "shootdown IPIs");
187 #endif
188 PMAP_COUNTER(unwire, "unwires");
189 PMAP_COUNTER(copy, "copies");
190 PMAP_COUNTER(clear_modify, "clear_modifies");
191 PMAP_COUNTER(protect, "protects");
192 PMAP_COUNTER(page_protect, "page_protects");
193 
194 #define PMAP_ASID_RESERVED 0
195 CTASSERT(PMAP_ASID_RESERVED == 0);
196 
197 #ifndef PMAP_SEGTAB_ALIGN
198 #define PMAP_SEGTAB_ALIGN	/* nothing */
199 #endif
200 #ifdef _LP64
201 pmap_segtab_t	pmap_kstart_segtab PMAP_SEGTAB_ALIGN; /* first mid-level segtab for kernel */
202 #endif
203 pmap_segtab_t	pmap_kern_segtab PMAP_SEGTAB_ALIGN = { /* top level segtab for kernel */
204 #ifdef _LP64
205 	.seg_seg[(VM_MIN_KERNEL_ADDRESS & XSEGOFSET) >> SEGSHIFT] = &pmap_kstart_segtab,
206 #endif
207 };
208 
209 struct pmap_kernel kernel_pmap_store = {
210 	.kernel_pmap = {
211 		.pm_count = 1,
212 		.pm_segtab = &pmap_kern_segtab,
213 		.pm_minaddr = VM_MIN_KERNEL_ADDRESS,
214 		.pm_maxaddr = VM_MAX_KERNEL_ADDRESS,
215 	},
216 };
217 
218 struct pmap * const kernel_pmap_ptr = &kernel_pmap_store.kernel_pmap;
219 
220 /* The current top of kernel VM - gets updated by pmap_growkernel */
221 vaddr_t pmap_curmaxkvaddr;
222 
223 struct pmap_limits pmap_limits = {	/* VA and PA limits */
224 	.virtual_start = VM_MIN_KERNEL_ADDRESS,
225 	.virtual_end = VM_MAX_KERNEL_ADDRESS,
226 };
227 
228 #ifdef UVMHIST
229 static struct kern_history_ent pmapexechistbuf[10000];
230 static struct kern_history_ent pmaphistbuf[10000];
231 static struct kern_history_ent pmapsegtabhistbuf[1000];
232 UVMHIST_DEFINE(pmapexechist) = UVMHIST_INITIALIZER(pmapexechist, pmapexechistbuf);
233 UVMHIST_DEFINE(pmaphist) = UVMHIST_INITIALIZER(pmaphist, pmaphistbuf);
234 UVMHIST_DEFINE(pmapsegtabhist) = UVMHIST_INITIALIZER(pmapsegtabhist, pmapsegtabhistbuf);
235 #endif
236 
237 /*
238  * The pools from which pmap structures and sub-structures are allocated.
239  */
240 struct pool pmap_pmap_pool;
241 struct pool pmap_pv_pool;
242 
243 #ifndef PMAP_PV_LOWAT
244 #define	PMAP_PV_LOWAT	16
245 #endif
246 int	pmap_pv_lowat = PMAP_PV_LOWAT;
247 
248 bool	pmap_initialized = false;
249 #define	PMAP_PAGE_COLOROK_P(a, b) \
250 		((((int)(a) ^ (int)(b)) & pmap_page_colormask) == 0)
251 u_int	pmap_page_colormask;
252 
253 #define PAGE_IS_MANAGED(pa)	(pmap_initialized && uvm_pageismanaged(pa))
254 
255 #define PMAP_IS_ACTIVE(pm)						\
256 	((pm) == pmap_kernel() ||					\
257 	 (pm) == curlwp->l_proc->p_vmspace->vm_map.pmap)
258 
259 /* Forward function declarations */
260 void pmap_page_remove(struct vm_page_md *);
261 static void pmap_pvlist_check(struct vm_page_md *);
262 void pmap_remove_pv(pmap_t, vaddr_t, struct vm_page *, bool);
263 void pmap_enter_pv(pmap_t, vaddr_t, paddr_t, struct vm_page_md *, pt_entry_t *, u_int);
264 
265 /*
266  * PV table management functions.
267  */
268 void	*pmap_pv_page_alloc(struct pool *, int);
269 void	pmap_pv_page_free(struct pool *, void *);
270 
271 struct pool_allocator pmap_pv_page_allocator = {
272 	pmap_pv_page_alloc, pmap_pv_page_free, 0,
273 };
274 
275 #define	pmap_pv_alloc()		pool_get(&pmap_pv_pool, PR_NOWAIT)
276 #define	pmap_pv_free(pv)	pool_put(&pmap_pv_pool, (pv))
277 
278 #ifndef PMAP_NEED_TLB_MISS_LOCK
279 
280 #if defined(PMAP_MD_NEED_TLB_MISS_LOCK) || defined(DEBUG)
281 #define	PMAP_NEED_TLB_MISS_LOCK
282 #endif /* PMAP_MD_NEED_TLB_MISS_LOCK || DEBUG */
283 
284 #endif /* PMAP_NEED_TLB_MISS_LOCK */
285 
286 #ifdef PMAP_NEED_TLB_MISS_LOCK
287 
288 #ifdef PMAP_MD_NEED_TLB_MISS_LOCK
289 #define	pmap_tlb_miss_lock_init()	__nothing /* MD code deals with this */
290 #define	pmap_tlb_miss_lock_enter()	pmap_md_tlb_miss_lock_enter()
291 #define	pmap_tlb_miss_lock_exit()	pmap_md_tlb_miss_lock_exit()
292 #else
293 kmutex_t pmap_tlb_miss_lock		__cacheline_aligned;
294 
295 static void
296 pmap_tlb_miss_lock_init(void)
297 {
298 	mutex_init(&pmap_tlb_miss_lock, MUTEX_SPIN, IPL_HIGH);
299 }
300 
301 static inline void
302 pmap_tlb_miss_lock_enter(void)
303 {
304 	mutex_spin_enter(&pmap_tlb_miss_lock);
305 }
306 
307 static inline void
308 pmap_tlb_miss_lock_exit(void)
309 {
310 	mutex_spin_exit(&pmap_tlb_miss_lock);
311 }
312 #endif /* PMAP_MD_NEED_TLB_MISS_LOCK */
313 
314 #else
315 
316 #define	pmap_tlb_miss_lock_init()	__nothing
317 #define	pmap_tlb_miss_lock_enter()	__nothing
318 #define	pmap_tlb_miss_lock_exit()	__nothing
319 
320 #endif /* PMAP_NEED_TLB_MISS_LOCK */
321 
322 #ifndef MULTIPROCESSOR
323 kmutex_t pmap_pvlist_mutex	__cacheline_aligned;
324 #endif
325 
326 /*
327  * Debug functions.
328  */
329 
330 #ifdef DEBUG
331 static inline void
332 pmap_asid_check(pmap_t pm, const char *func)
333 {
334 	if (!PMAP_IS_ACTIVE(pm))
335 		return;
336 
337 	struct pmap_asid_info * const pai = PMAP_PAI(pm, cpu_tlb_info(curcpu()));
338 	tlb_asid_t asid = tlb_get_asid();
339 	if (asid != pai->pai_asid)
340 		panic("%s: inconsistency for active TLB update: %u <-> %u",
341 		    func, asid, pai->pai_asid);
342 }
343 #endif
344 
345 static void
346 pmap_addr_range_check(pmap_t pmap, vaddr_t sva, vaddr_t eva, const char *func)
347 {
348 #ifdef DEBUG
349 	if (pmap == pmap_kernel()) {
350 		if (sva < VM_MIN_KERNEL_ADDRESS)
351 			panic("%s: kva %#"PRIxVADDR" not in range",
352 			    func, sva);
353 		if (eva >= pmap_limits.virtual_end)
354 			panic("%s: kva %#"PRIxVADDR" not in range",
355 			    func, eva);
356 	} else {
357 		if (eva > VM_MAXUSER_ADDRESS)
358 			panic("%s: uva %#"PRIxVADDR" not in range",
359 			    func, eva);
360 		pmap_asid_check(pmap, func);
361 	}
362 #endif
363 }
364 
365 /*
366  * Misc. functions.
367  */
368 
369 bool
370 pmap_page_clear_attributes(struct vm_page_md *mdpg, u_int clear_attributes)
371 {
372 	volatile unsigned long * const attrp = &mdpg->mdpg_attrs;
373 #ifdef MULTIPROCESSOR
374 	for (;;) {
375 		u_int old_attr = *attrp;
376 		if ((old_attr & clear_attributes) == 0)
377 			return false;
378 		u_int new_attr = old_attr & ~clear_attributes;
379 		if (old_attr == atomic_cas_ulong(attrp, old_attr, new_attr))
380 			return true;
381 	}
382 #else
383 	unsigned long old_attr = *attrp;
384 	if ((old_attr & clear_attributes) == 0)
385 		return false;
386 	*attrp &= ~clear_attributes;
387 	return true;
388 #endif
389 }
390 
391 void
392 pmap_page_set_attributes(struct vm_page_md *mdpg, u_int set_attributes)
393 {
394 #ifdef MULTIPROCESSOR
395 	atomic_or_ulong(&mdpg->mdpg_attrs, set_attributes);
396 #else
397 	mdpg->mdpg_attrs |= set_attributes;
398 #endif
399 }
400 
401 static void
402 pmap_page_syncicache(struct vm_page *pg)
403 {
404 	UVMHIST_FUNC(__func__);
405 	UVMHIST_CALLED(pmaphist);
406 #ifndef MULTIPROCESSOR
407 	struct pmap * const curpmap = curlwp->l_proc->p_vmspace->vm_map.pmap;
408 #endif
409 	struct vm_page_md * const mdpg = VM_PAGE_TO_MD(pg);
410 	pv_entry_t pv = &mdpg->mdpg_first;
411 	kcpuset_t *onproc;
412 #ifdef MULTIPROCESSOR
413 	kcpuset_create(&onproc, true);
414 	KASSERT(onproc != NULL);
415 #else
416 	onproc = NULL;
417 #endif
418 	VM_PAGEMD_PVLIST_READLOCK(mdpg);
419 	pmap_pvlist_check(mdpg);
420 
421 	UVMHIST_LOG(pmaphist, "pv %#jx pv_pmap %#jx", (uintptr_t)pv,
422 	    (uintptr_t)pv->pv_pmap, 0, 0);
423 
424 	if (pv->pv_pmap != NULL) {
425 		for (; pv != NULL; pv = pv->pv_next) {
426 #ifdef MULTIPROCESSOR
427 			UVMHIST_LOG(pmaphist, "pv %#jx pv_pmap %#jx",
428 			    (uintptr_t)pv, (uintptr_t)pv->pv_pmap, 0, 0);
429 			kcpuset_merge(onproc, pv->pv_pmap->pm_onproc);
430 			if (kcpuset_match(onproc, kcpuset_running)) {
431 				break;
432 			}
433 #else
434 			if (pv->pv_pmap == curpmap) {
435 				onproc = curcpu()->ci_data.cpu_kcpuset;
436 				break;
437 			}
438 #endif
439 		}
440 	}
441 	pmap_pvlist_check(mdpg);
442 	VM_PAGEMD_PVLIST_UNLOCK(mdpg);
443 	kpreempt_disable();
444 	pmap_md_page_syncicache(mdpg, onproc);
445 	kpreempt_enable();
446 #ifdef MULTIPROCESSOR
447 	kcpuset_destroy(onproc);
448 #endif
449 }
450 
451 /*
452  * Define the initial bounds of the kernel virtual address space.
453  */
454 void
455 pmap_virtual_space(vaddr_t *vstartp, vaddr_t *vendp)
456 {
457 
458 	*vstartp = pmap_limits.virtual_start;
459 	*vendp = pmap_limits.virtual_end;
460 }
461 
462 vaddr_t
463 pmap_growkernel(vaddr_t maxkvaddr)
464 {
465 	UVMHIST_FUNC(__func__);
466 	UVMHIST_CALLARGS(pmaphist, "maxkvaddr=%#jx (%#jx)", maxkvaddr,
467 	    pmap_curmaxkvaddr, 0, 0);
468 
469 	vaddr_t virtual_end = pmap_curmaxkvaddr;
470 	maxkvaddr = pmap_round_seg(maxkvaddr) - 1;
471 
472 	/*
473 	 * Don't exceed VM_MAX_KERNEL_ADDRESS!
474 	 */
475 	if (maxkvaddr == 0 || maxkvaddr > VM_MAX_KERNEL_ADDRESS)
476 		maxkvaddr = VM_MAX_KERNEL_ADDRESS;
477 
478 	/*
479 	 * Reserve PTEs for the new KVA space.
480 	 */
481 	for (; virtual_end < maxkvaddr; virtual_end += NBSEG) {
482 		pmap_pte_reserve(pmap_kernel(), virtual_end, 0);
483 	}
484 
485 	kasan_shadow_map((void *)pmap_curmaxkvaddr,
486 	    (size_t)(virtual_end - pmap_curmaxkvaddr));
487 
488 	/*
489 	 * Update new end.
490 	 */
491 	pmap_curmaxkvaddr = virtual_end;
492 
493 	UVMHIST_LOG(pmaphist, " <-- done", 0, 0, 0, 0);
494 
495 	return virtual_end;
496 }
497 
498 /*
499  * Bootstrap memory allocator (alternative to vm_bootstrap_steal_memory()).
500  * This function allows for early dynamic memory allocation until the virtual
501  * memory system has been bootstrapped.  After that point, either kmem_alloc
502  * or malloc should be used.  This function works by stealing pages from the
503  * (to be) managed page pool, then implicitly mapping the pages (by using
504  * their direct mapped addresses) and zeroing them.
505  *
506  * It may be used once the physical memory segments have been pre-loaded
507  * into the vm_physmem[] array.  Early memory allocation MUST use this
508  * interface!  This cannot be used after vm_page_startup(), and will
509  * generate a panic if tried.
510  *
511  * Note that this memory will never be freed, and in essence it is wired
512  * down.
513  *
514  * We must adjust *vstartp and/or *vendp iff we use address space
515  * from the kernel virtual address range defined by pmap_virtual_space().
516  */
517 vaddr_t
518 pmap_steal_memory(vsize_t size, vaddr_t *vstartp, vaddr_t *vendp)
519 {
520 	size_t npgs;
521 	paddr_t pa;
522 	vaddr_t va;
523 
524 	uvm_physseg_t maybe_bank = UVM_PHYSSEG_TYPE_INVALID;
525 
526 	size = round_page(size);
527 	npgs = atop(size);
528 
529 	aprint_debug("%s: need %zu pages\n", __func__, npgs);
530 
531 	for (uvm_physseg_t bank = uvm_physseg_get_first();
532 	     uvm_physseg_valid_p(bank);
533 	     bank = uvm_physseg_get_next(bank)) {
534 
535 		if (uvm.page_init_done == true)
536 			panic("pmap_steal_memory: called _after_ bootstrap");
537 
538 		aprint_debug("%s: seg %"PRIxPHYSSEG": %#"PRIxPADDR" %#"PRIxPADDR" %#"PRIxPADDR" %#"PRIxPADDR"\n",
539 		    __func__, bank,
540 		    uvm_physseg_get_avail_start(bank), uvm_physseg_get_start(bank),
541 		    uvm_physseg_get_avail_end(bank), uvm_physseg_get_end(bank));
542 
543 		if (uvm_physseg_get_avail_start(bank) != uvm_physseg_get_start(bank)
544 		    || uvm_physseg_get_avail_start(bank) >= uvm_physseg_get_avail_end(bank)) {
545 			aprint_debug("%s: seg %"PRIxPHYSSEG": bad start\n", __func__, bank);
546 			continue;
547 		}
548 
549 		if (uvm_physseg_get_avail_end(bank) - uvm_physseg_get_avail_start(bank) < npgs) {
550 			aprint_debug("%s: seg %"PRIxPHYSSEG": too small for %zu pages\n",
551 			    __func__, bank, npgs);
552 			continue;
553 		}
554 
555 		if (!pmap_md_ok_to_steal_p(bank, npgs)) {
556 			continue;
557 		}
558 
559 		/*
560 		 * Always try to allocate from the segment with the least
561 		 * amount of space left.
562 		 */
563 #define VM_PHYSMEM_SPACE(b)	((uvm_physseg_get_avail_end(b)) - (uvm_physseg_get_avail_start(b)))
564 		if (uvm_physseg_valid_p(maybe_bank) == false
565 		    || VM_PHYSMEM_SPACE(bank) < VM_PHYSMEM_SPACE(maybe_bank)) {
566 			maybe_bank = bank;
567 		}
568 	}
569 
570 	if (uvm_physseg_valid_p(maybe_bank)) {
571 		const uvm_physseg_t bank = maybe_bank;
572 
573 		/*
574 		 * There are enough pages here; steal them!
575 		 */
576 		pa = ptoa(uvm_physseg_get_start(bank));
577 		uvm_physseg_unplug(atop(pa), npgs);
578 
579 		aprint_debug("%s: seg %"PRIxPHYSSEG": %zu pages stolen (%#"PRIxPADDR" left)\n",
580 		    __func__, bank, npgs, VM_PHYSMEM_SPACE(bank));
581 
582 		va = pmap_md_map_poolpage(pa, size);
583 		memset((void *)va, 0, size);
584 		return va;
585 	}
586 
587 	/*
588 	 * If we got here, there was no memory left.
589 	 */
590 	panic("pmap_steal_memory: no memory to steal %zu pages", npgs);
591 }
592 
593 /*
594  *	Bootstrap the system enough to run with virtual memory.
595  *	(Common routine called by machine-dependent bootstrap code.)
596  */
597 void
598 pmap_bootstrap_common(void)
599 {
600 	pmap_tlb_miss_lock_init();
601 }
602 
603 /*
604  *	Initialize the pmap module.
605  *	Called by vm_init, to initialize any structures that the pmap
606  *	system needs to map virtual memory.
607  */
608 void
609 pmap_init(void)
610 {
611 	UVMHIST_LINK_STATIC(pmapexechist);
612 	UVMHIST_LINK_STATIC(pmaphist);
613 	UVMHIST_LINK_STATIC(pmapsegtabhist);
614 
615 	UVMHIST_FUNC(__func__);
616 	UVMHIST_CALLED(pmaphist);
617 
618 	/*
619 	 * Initialize the segtab lock.
620 	 */
621 	mutex_init(&pmap_segtab_lock, MUTEX_DEFAULT, IPL_HIGH);
622 
623 	/*
624 	 * Set a low water mark on the pv_entry pool, so that we are
625 	 * more likely to have these around even in extreme memory
626 	 * starvation.
627 	 */
628 	pool_setlowat(&pmap_pv_pool, pmap_pv_lowat);
629 
630 	/*
631 	 * Set the page colormask but allow pmap_md_init to override it.
632 	 */
633 	pmap_page_colormask = ptoa(uvmexp.colormask);
634 
635 	pmap_md_init();
636 
637 	/*
638 	 * Now it is safe to enable pv entry recording.
639 	 */
640 	pmap_initialized = true;
641 }
642 
643 /*
644  *	Create and return a physical map.
645  *
646  *	If the size specified for the map
647  *	is zero, the map is an actual physical
648  *	map, and may be referenced by the
649  *	hardware.
650  *
651  *	If the size specified is non-zero,
652  *	the map will be used in software only, and
653  *	is bounded by that size.
654  */
655 pmap_t
656 pmap_create(void)
657 {
658 	UVMHIST_FUNC(__func__);
659 	UVMHIST_CALLED(pmaphist);
660 	PMAP_COUNT(create);
661 
662 	pmap_t pmap = pool_get(&pmap_pmap_pool, PR_WAITOK);
663 	memset(pmap, 0, PMAP_SIZE);
664 
665 	KASSERT(pmap->pm_pai[0].pai_link.le_prev == NULL);
666 
667 	pmap->pm_count = 1;
668 	pmap->pm_minaddr = VM_MIN_ADDRESS;
669 	pmap->pm_maxaddr = VM_MAXUSER_ADDRESS;
670 
671 	pmap_segtab_init(pmap);
672 
673 #ifdef MULTIPROCESSOR
674 	kcpuset_create(&pmap->pm_active, true);
675 	kcpuset_create(&pmap->pm_onproc, true);
676 	KASSERT(pmap->pm_active != NULL);
677 	KASSERT(pmap->pm_onproc != NULL);
678 #endif
679 
680 	UVMHIST_LOG(pmaphist, " <-- done (pmap=%#jx)", (uintptr_t)pmap,
681 	    0, 0, 0);
682 
683 	return pmap;
684 }
685 
686 /*
687  *	Retire the given physical map from service.
688  *	Should only be called if the map contains
689  *	no valid mappings.
690  */
691 void
692 pmap_destroy(pmap_t pmap)
693 {
694 	UVMHIST_FUNC(__func__);
695 	UVMHIST_CALLARGS(pmaphist, "(pmap=%#jx)", (uintptr_t)pmap, 0, 0, 0);
696 
697 	membar_release();
698 	if (atomic_dec_uint_nv(&pmap->pm_count) > 0) {
699 		PMAP_COUNT(dereference);
700 		UVMHIST_LOG(pmaphist, " <-- done (deref)", 0, 0, 0, 0);
701 		return;
702 	}
703 	membar_acquire();
704 
705 	PMAP_COUNT(destroy);
706 	KASSERT(pmap->pm_count == 0);
707 	kpreempt_disable();
708 	pmap_tlb_miss_lock_enter();
709 	pmap_tlb_asid_release_all(pmap);
710 	pmap_segtab_destroy(pmap, NULL, 0);
711 	pmap_tlb_miss_lock_exit();
712 
713 #ifdef MULTIPROCESSOR
714 	kcpuset_destroy(pmap->pm_active);
715 	kcpuset_destroy(pmap->pm_onproc);
716 	pmap->pm_active = NULL;
717 	pmap->pm_onproc = NULL;
718 #endif
719 
720 	pool_put(&pmap_pmap_pool, pmap);
721 	kpreempt_enable();
722 
723 	UVMHIST_LOG(pmaphist, " <-- done (freed)", 0, 0, 0, 0);
724 }
725 
726 /*
727  *	Add a reference to the specified pmap.
728  */
729 void
730 pmap_reference(pmap_t pmap)
731 {
732 	UVMHIST_FUNC(__func__);
733 	UVMHIST_CALLARGS(pmaphist, "(pmap=%#jx)", (uintptr_t)pmap, 0, 0, 0);
734 	PMAP_COUNT(reference);
735 
736 	if (pmap != NULL) {
737 		atomic_inc_uint(&pmap->pm_count);
738 	}
739 
740 	UVMHIST_LOG(pmaphist, " <-- done", 0, 0, 0, 0);
741 }
742 
743 /*
744  *	Make a new pmap (vmspace) active for the given process.
745  */
746 void
747 pmap_activate(struct lwp *l)
748 {
749 	pmap_t pmap = l->l_proc->p_vmspace->vm_map.pmap;
750 
751 	UVMHIST_FUNC(__func__);
752 	UVMHIST_CALLARGS(pmaphist, "(l=%#jx pmap=%#jx)", (uintptr_t)l,
753 	    (uintptr_t)pmap, 0, 0);
754 	PMAP_COUNT(activate);
755 
756 	kpreempt_disable();
757 	pmap_tlb_miss_lock_enter();
758 	pmap_tlb_asid_acquire(pmap, l);
759 	pmap_segtab_activate(pmap, l);
760 	pmap_tlb_miss_lock_exit();
761 	kpreempt_enable();
762 
763 	UVMHIST_LOG(pmaphist, " <-- done (%ju:%ju)", l->l_proc->p_pid,
764 	    l->l_lid, 0, 0);
765 }
766 
767 /*
768  * Remove this page from all physical maps in which it resides.
769  * Reflects back modify bits to the pager.
770  */
771 void
772 pmap_page_remove(struct vm_page_md *mdpg)
773 {
774 	kpreempt_disable();
775 	VM_PAGEMD_PVLIST_LOCK(mdpg);
776 	pmap_pvlist_check(mdpg);
777 
778 	struct vm_page * const pg =
779 	    VM_PAGEMD_VMPAGE_P(mdpg) ? VM_MD_TO_PAGE(mdpg) : NULL;
780 
781 	UVMHIST_FUNC(__func__);
782 	if (pg) {
783 		UVMHIST_CALLARGS(pmaphist, "mdpg %#jx pg %#jx (pa %#jx): "
784 		    "execpage cleared", (uintptr_t)mdpg, (uintptr_t)pg,
785 		    VM_PAGE_TO_PHYS(pg), 0);
786 	} else {
787 		UVMHIST_CALLARGS(pmaphist, "mdpg %#jx", (uintptr_t)mdpg, 0,
788 		    0, 0);
789 	}
790 
791 #ifdef PMAP_VIRTUAL_CACHE_ALIASES
792 	pmap_page_clear_attributes(mdpg, VM_PAGEMD_EXECPAGE|VM_PAGEMD_UNCACHED);
793 #else
794 	pmap_page_clear_attributes(mdpg, VM_PAGEMD_EXECPAGE);
795 #endif
796 	PMAP_COUNT(exec_uncached_remove);
797 
798 	pv_entry_t pv = &mdpg->mdpg_first;
799 	if (pv->pv_pmap == NULL) {
800 		VM_PAGEMD_PVLIST_UNLOCK(mdpg);
801 		kpreempt_enable();
802 		UVMHIST_LOG(pmaphist, " <-- done (empty)", 0, 0, 0, 0);
803 		return;
804 	}
805 
806 	pv_entry_t npv;
807 	pv_entry_t pvp = NULL;
808 
809 	for (; pv != NULL; pv = npv) {
810 		npv = pv->pv_next;
811 #ifdef PMAP_VIRTUAL_CACHE_ALIASES
812 		if (PV_ISKENTER_P(pv)) {
813 			UVMHIST_LOG(pmaphist, " pv %#jx pmap %#jx va %#jx"
814 			    " skip", (uintptr_t)pv, (uintptr_t)pv->pv_pmap,
815 			    pv->pv_va, 0);
816 
817 			KASSERT(pv->pv_pmap == pmap_kernel());
818 
819 			/* Assume no more - it'll get fixed if there are */
820 			pv->pv_next = NULL;
821 
822 			/*
823 			 * pvp is non-null when we already have a PV_KENTER
824 			 * pv in pvh_first; otherwise we haven't seen a
825 			 * PV_KENTER pv and we need to copy this one to
826 			 * pvh_first
827 			 */
828 			if (pvp) {
829 				/*
830 				 * The previous PV_KENTER pv needs to point to
831 				 * this PV_KENTER pv
832 				 */
833 				pvp->pv_next = pv;
834 			} else {
835 				pv_entry_t fpv = &mdpg->mdpg_first;
836 				*fpv = *pv;
837 				KASSERT(fpv->pv_pmap == pmap_kernel());
838 			}
839 			pvp = pv;
840 			continue;
841 		}
842 #endif
843 		const pmap_t pmap = pv->pv_pmap;
844 		vaddr_t va = trunc_page(pv->pv_va);
845 		pt_entry_t * const ptep = pmap_pte_lookup(pmap, va);
846 		KASSERTMSG(ptep != NULL, "%#"PRIxVADDR " %#"PRIxVADDR, va,
847 		    pmap_limits.virtual_end);
848 		pt_entry_t pte = *ptep;
849 		UVMHIST_LOG(pmaphist, " pv %#jx pmap %#jx va %#jx"
850 		    " pte %#jx", (uintptr_t)pv, (uintptr_t)pmap, va,
851 		    pte_value(pte));
852 		if (!pte_valid_p(pte))
853 			continue;
854 		const bool is_kernel_pmap_p = (pmap == pmap_kernel());
855 		if (is_kernel_pmap_p) {
856 			PMAP_COUNT(remove_kernel_pages);
857 		} else {
858 			PMAP_COUNT(remove_user_pages);
859 		}
860 		if (pte_wired_p(pte))
861 			pmap->pm_stats.wired_count--;
862 		pmap->pm_stats.resident_count--;
863 
864 		pmap_tlb_miss_lock_enter();
865 		const pt_entry_t npte = pte_nv_entry(is_kernel_pmap_p);
866 		pte_set(ptep, npte);
867 		if (__predict_true(!(pmap->pm_flags & PMAP_DEFERRED_ACTIVATE))) {
868 			/*
869 			 * Flush the TLB for the given address.
870 			 */
871 			pmap_tlb_invalidate_addr(pmap, va);
872 		}
873 		pmap_tlb_miss_lock_exit();
874 
875 		/*
876 		 * non-null means this is a non-pvh_first pv, so we should
877 		 * free it.
878 		 */
879 		if (pvp) {
880 			KASSERT(pvp->pv_pmap == pmap_kernel());
881 			KASSERT(pvp->pv_next == NULL);
882 			pmap_pv_free(pv);
883 		} else {
884 			pv->pv_pmap = NULL;
885 			pv->pv_next = NULL;
886 		}
887 	}
888 
889 	pmap_pvlist_check(mdpg);
890 	VM_PAGEMD_PVLIST_UNLOCK(mdpg);
891 	kpreempt_enable();
892 
893 	UVMHIST_LOG(pmaphist, " <-- done", 0, 0, 0, 0);
894 }
895 
896 #ifdef __HAVE_PMAP_PV_TRACK
897 /*
898  * pmap_pv_protect: change protection of an unmanaged pv-tracked page from
899  * all pmaps that map it
900  */
901 void
902 pmap_pv_protect(paddr_t pa, vm_prot_t prot)
903 {
904 
905 	/* the only case is remove at the moment */
906 	KASSERT(prot == VM_PROT_NONE);
907 	struct pmap_page *pp;
908 
909 	pp = pmap_pv_tracked(pa);
910 	if (pp == NULL)
911 		panic("pmap_pv_protect: page not pv-tracked: 0x%"PRIxPADDR,
912 		    pa);
913 
914 	struct vm_page_md *mdpg = PMAP_PAGE_TO_MD(pp);
915 	pmap_page_remove(mdpg);
916 }
917 #endif
918 
919 /*
920  *	Make a previously active pmap (vmspace) inactive.
921  */
922 void
923 pmap_deactivate(struct lwp *l)
924 {
925 	pmap_t pmap = l->l_proc->p_vmspace->vm_map.pmap;
926 
927 	UVMHIST_FUNC(__func__);
928 	UVMHIST_CALLARGS(pmaphist, "(l=%#jx pmap=%#jx)", (uintptr_t)l,
929 	    (uintptr_t)pmap, 0, 0);
930 	PMAP_COUNT(deactivate);
931 
932 	kpreempt_disable();
933 	KASSERT(l == curlwp || l->l_cpu == curlwp->l_cpu);
934 	pmap_tlb_miss_lock_enter();
935 	pmap_tlb_asid_deactivate(pmap);
936 	pmap_segtab_deactivate(pmap);
937 	pmap_tlb_miss_lock_exit();
938 	kpreempt_enable();
939 
940 	UVMHIST_LOG(pmaphist, " <-- done (%ju:%ju)", l->l_proc->p_pid,
941 	    l->l_lid, 0, 0);
942 }
943 
944 void
945 pmap_update(struct pmap *pmap)
946 {
947 	UVMHIST_FUNC(__func__);
948 	UVMHIST_CALLARGS(pmaphist, "(pmap=%#jx)", (uintptr_t)pmap, 0, 0, 0);
949 	PMAP_COUNT(update);
950 
951 	kpreempt_disable();
952 #if defined(MULTIPROCESSOR) && defined(PMAP_TLB_NEED_SHOOTDOWN)
953 	u_int pending = atomic_swap_uint(&pmap->pm_shootdown_pending, 0);
954 	if (pending && pmap_tlb_shootdown_bystanders(pmap))
955 		PMAP_COUNT(shootdown_ipis);
956 #endif
957 	pmap_tlb_miss_lock_enter();
958 #if defined(DEBUG) && !defined(MULTIPROCESSOR)
959 	pmap_tlb_check(pmap, pmap_md_tlb_check_entry);
960 #endif /* DEBUG */
961 
962 	/*
963 	 * If pmap_remove_all was called, we deactivated ourselves and nuked
964 	 * our ASID.  Now we have to reactivate ourselves.
965 	 */
966 	if (__predict_false(pmap->pm_flags & PMAP_DEFERRED_ACTIVATE)) {
967 		pmap->pm_flags ^= PMAP_DEFERRED_ACTIVATE;
968 		pmap_tlb_asid_acquire(pmap, curlwp);
969 		pmap_segtab_activate(pmap, curlwp);
970 	}
971 	pmap_tlb_miss_lock_exit();
972 	kpreempt_enable();
973 
974 	UVMHIST_LOG(pmaphist, " <-- done (kernel=%jd)",
975 		    (pmap == pmap_kernel() ? 1 : 0), 0, 0, 0);
976 }
977 
978 /*
979  *	Remove the given range of addresses from the specified map.
980  *
981  *	It is assumed that the start and end are properly
982  *	rounded to the page size.
983  */
984 
985 static bool
986 pmap_pte_remove(pmap_t pmap, vaddr_t sva, vaddr_t eva, pt_entry_t *ptep,
987     uintptr_t flags)
988 {
989 	const pt_entry_t npte = flags;
990 	const bool is_kernel_pmap_p = (pmap == pmap_kernel());
991 
992 	UVMHIST_FUNC(__func__);
993 	UVMHIST_CALLARGS(pmaphist, "(pmap=%#jx kernel=%jd va=%#jx..%#jx)",
994 	    (uintptr_t)pmap, (is_kernel_pmap_p ? 1 : 0), sva, eva);
995 	UVMHIST_LOG(pmaphist, "ptep=%#jx, flags(npte)=%#jx)",
996 	    (uintptr_t)ptep, flags, 0, 0);
997 
998 	KASSERT(kpreempt_disabled());
999 
1000 	for (; sva < eva; sva += NBPG, ptep++) {
1001 		const pt_entry_t pte = *ptep;
1002 		if (!pte_valid_p(pte))
1003 			continue;
1004 		if (is_kernel_pmap_p) {
1005 			PMAP_COUNT(remove_kernel_pages);
1006 		} else {
1007 			PMAP_COUNT(remove_user_pages);
1008 		}
1009 		if (pte_wired_p(pte))
1010 			pmap->pm_stats.wired_count--;
1011 		pmap->pm_stats.resident_count--;
1012 		struct vm_page * const pg = PHYS_TO_VM_PAGE(pte_to_paddr(pte));
1013 		if (__predict_true(pg != NULL)) {
1014 			pmap_remove_pv(pmap, sva, pg, pte_modified_p(pte));
1015 		}
1016 		pmap_tlb_miss_lock_enter();
1017 		pte_set(ptep, npte);
1018 		if (__predict_true(!(pmap->pm_flags & PMAP_DEFERRED_ACTIVATE))) {
1019 
1020 			/*
1021 			 * Flush the TLB for the given address.
1022 			 */
1023 			pmap_tlb_invalidate_addr(pmap, sva);
1024 		}
1025 		pmap_tlb_miss_lock_exit();
1026 	}
1027 
1028 	UVMHIST_LOG(pmaphist, " <-- done", 0, 0, 0, 0);
1029 
1030 	return false;
1031 }
1032 
1033 void
1034 pmap_remove(pmap_t pmap, vaddr_t sva, vaddr_t eva)
1035 {
1036 	const bool is_kernel_pmap_p = (pmap == pmap_kernel());
1037 	const pt_entry_t npte = pte_nv_entry(is_kernel_pmap_p);
1038 
1039 	UVMHIST_FUNC(__func__);
1040 	UVMHIST_CALLARGS(pmaphist, "(pmap=%#jx, va=%#jx..%#jx)",
1041 	    (uintptr_t)pmap, sva, eva, 0);
1042 
1043 	if (is_kernel_pmap_p) {
1044 		PMAP_COUNT(remove_kernel_calls);
1045 	} else {
1046 		PMAP_COUNT(remove_user_calls);
1047 	}
1048 #ifdef PMAP_FAULTINFO
1049 	curpcb->pcb_faultinfo.pfi_faultaddr = 0;
1050 	curpcb->pcb_faultinfo.pfi_repeats = 0;
1051 	curpcb->pcb_faultinfo.pfi_faultptep = NULL;
1052 #endif
1053 	kpreempt_disable();
1054 	pmap_addr_range_check(pmap, sva, eva, __func__);
1055 	pmap_pte_process(pmap, sva, eva, pmap_pte_remove, npte);
1056 	kpreempt_enable();
1057 
1058 	UVMHIST_LOG(pmaphist, " <-- done", 0, 0, 0, 0);
1059 }
1060 
1061 /*
1062  *	pmap_page_protect:
1063  *
1064  *	Lower the permission for all mappings to a given page.
1065  */
1066 void
1067 pmap_page_protect(struct vm_page *pg, vm_prot_t prot)
1068 {
1069 	struct vm_page_md * const mdpg = VM_PAGE_TO_MD(pg);
1070 	pv_entry_t pv;
1071 	vaddr_t va;
1072 
1073 	UVMHIST_FUNC(__func__);
1074 	UVMHIST_CALLARGS(pmaphist, "(pg=%#jx (pa %#jx) prot=%#jx)",
1075 	    (uintptr_t)pg, VM_PAGE_TO_PHYS(pg), prot, 0);
1076 	PMAP_COUNT(page_protect);
1077 
1078 	switch (prot) {
1079 	case VM_PROT_READ|VM_PROT_WRITE:
1080 	case VM_PROT_ALL:
1081 		break;
1082 
1083 	/* copy_on_write */
1084 	case VM_PROT_READ:
1085 	case VM_PROT_READ|VM_PROT_EXECUTE:
1086 		pv = &mdpg->mdpg_first;
1087 		kpreempt_disable();
1088 		VM_PAGEMD_PVLIST_READLOCK(mdpg);
1089 		pmap_pvlist_check(mdpg);
1090 		/*
1091 		 * Loop over all current mappings setting/clearing as
1092 		 * appropriate.
1093 		 */
1094 		if (pv->pv_pmap != NULL) {
1095 			while (pv != NULL) {
1096 #ifdef PMAP_VIRTUAL_CACHE_ALIASES
1097 				if (PV_ISKENTER_P(pv)) {
1098 					pv = pv->pv_next;
1099 					continue;
1100 				}
1101 #endif
1102 				const pmap_t pmap = pv->pv_pmap;
1103 				va = trunc_page(pv->pv_va);
1104 				const uintptr_t gen =
1105 				    VM_PAGEMD_PVLIST_UNLOCK(mdpg);
1106 				pmap_protect(pmap, va, va + PAGE_SIZE, prot);
1107 				KASSERT(pv->pv_pmap == pmap);
1108 				pmap_update(pmap);
1109 				if (gen != VM_PAGEMD_PVLIST_READLOCK(mdpg)) {
1110 					pv = &mdpg->mdpg_first;
1111 				} else {
1112 					pv = pv->pv_next;
1113 				}
1114 				pmap_pvlist_check(mdpg);
1115 			}
1116 		}
1117 		pmap_pvlist_check(mdpg);
1118 		VM_PAGEMD_PVLIST_UNLOCK(mdpg);
1119 		kpreempt_enable();
1120 		break;
1121 
1122 	/* remove_all */
1123 	default:
1124 		pmap_page_remove(mdpg);
1125 	}
1126 
1127 	UVMHIST_LOG(pmaphist, " <-- done", 0, 0, 0, 0);
1128 }
1129 
1130 static bool
1131 pmap_pte_protect(pmap_t pmap, vaddr_t sva, vaddr_t eva, pt_entry_t *ptep,
1132 	uintptr_t flags)
1133 {
1134 	const vm_prot_t prot = (flags & VM_PROT_ALL);
1135 
1136 	UVMHIST_FUNC(__func__);
1137 	UVMHIST_CALLARGS(pmaphist, "(pmap=%#jx kernel=%jd va=%#jx..%#jx)",
1138 	    (uintptr_t)pmap, (pmap == pmap_kernel() ? 1 : 0), sva, eva);
1139 	UVMHIST_LOG(pmaphist, "ptep=%#jx, flags(npte)=%#jx)",
1140 	    (uintptr_t)ptep, flags, 0, 0);
1141 
1142 	KASSERT(kpreempt_disabled());
1143 	/*
1144 	 * Change protection on every valid mapping within this segment.
1145 	 */
1146 	for (; sva < eva; sva += NBPG, ptep++) {
1147 		pt_entry_t pte = *ptep;
1148 		if (!pte_valid_p(pte))
1149 			continue;
1150 		struct vm_page * const pg = PHYS_TO_VM_PAGE(pte_to_paddr(pte));
1151 		if (pg != NULL && pte_modified_p(pte)) {
1152 			struct vm_page_md * const mdpg = VM_PAGE_TO_MD(pg);
1153 			if (VM_PAGEMD_EXECPAGE_P(mdpg)) {
1154 				KASSERT(!VM_PAGEMD_PVLIST_EMPTY_P(mdpg));
1155 #ifdef PMAP_VIRTUAL_CACHE_ALIASES
1156 				if (VM_PAGEMD_CACHED_P(mdpg)) {
1157 #endif
1158 					UVMHIST_LOG(pmapexechist,
1159 					    "pg %#jx (pa %#jx): "
1160 					    "syncicached performed",
1161 					    (uintptr_t)pg, VM_PAGE_TO_PHYS(pg),
1162 					    0, 0);
1163 					pmap_page_syncicache(pg);
1164 					PMAP_COUNT(exec_synced_protect);
1165 #ifdef PMAP_VIRTUAL_CACHE_ALIASES
1166 				}
1167 #endif
1168 			}
1169 		}
1170 		pte = pte_prot_downgrade(pte, prot);
1171 		if (*ptep != pte) {
1172 			pmap_tlb_miss_lock_enter();
1173 			pte_set(ptep, pte);
1174 			/*
1175 			 * Update the TLB if needed.
1176 			 */
1177 			pmap_tlb_update_addr(pmap, sva, pte, PMAP_TLB_NEED_IPI);
1178 			pmap_tlb_miss_lock_exit();
1179 		}
1180 	}
1181 
1182 	UVMHIST_LOG(pmaphist, " <-- done", 0, 0, 0, 0);
1183 
1184 	return false;
1185 }
1186 
1187 /*
1188  *	Set the physical protection on the
1189  *	specified range of this map as requested.
1190  */
1191 void
1192 pmap_protect(pmap_t pmap, vaddr_t sva, vaddr_t eva, vm_prot_t prot)
1193 {
1194 	UVMHIST_FUNC(__func__);
1195 	UVMHIST_CALLARGS(pmaphist, "(pmap=%#jx, va=%#jx..%#jx, prot=%ju)",
1196 	    (uintptr_t)pmap, sva, eva, prot);
1197 	PMAP_COUNT(protect);
1198 
1199 	if ((prot & VM_PROT_READ) == VM_PROT_NONE) {
1200 		pmap_remove(pmap, sva, eva);
1201 		UVMHIST_LOG(pmaphist, " <-- done", 0, 0, 0, 0);
1202 		return;
1203 	}
1204 
1205 	/*
1206 	 * Change protection on every valid mapping within this segment.
1207 	 */
1208 	kpreempt_disable();
1209 	pmap_addr_range_check(pmap, sva, eva, __func__);
1210 	pmap_pte_process(pmap, sva, eva, pmap_pte_protect, prot);
1211 	kpreempt_enable();
1212 
1213 	UVMHIST_LOG(pmaphist, " <-- done", 0, 0, 0, 0);
1214 }
1215 
1216 #if defined(PMAP_VIRTUAL_CACHE_ALIASES) && !defined(PMAP_NO_PV_UNCACHED)
1217 /*
1218  *	pmap_page_cache:
1219  *
1220  *	Change all mappings of a managed page to cached/uncached.
1221  */
1222 void
1223 pmap_page_cache(struct vm_page_md *mdpg, bool cached)
1224 {
1225 #ifdef UVMHIST
1226 	const bool vmpage_p = VM_PAGEMD_VMPAGE_P(mdpg);
1227 	struct vm_page * const pg = vmpage_p ? VM_MD_TO_PAGE(mdpg) : NULL;
1228 #endif
1229 
1230 	UVMHIST_FUNC(__func__);
1231 	UVMHIST_CALLARGS(pmaphist, "(mdpg=%#jx (pa %#jx) cached=%jd vmpage %jd)",
1232 	    (uintptr_t)mdpg, pg ? VM_PAGE_TO_PHYS(pg) : 0, cached, vmpage_p);
1233 
1234 	KASSERT(kpreempt_disabled());
1235 	KASSERT(VM_PAGEMD_PVLIST_LOCKED_P(mdpg));
1236 
1237 	if (cached) {
1238 		pmap_page_clear_attributes(mdpg, VM_PAGEMD_UNCACHED);
1239 		PMAP_COUNT(page_cache_restorations);
1240 	} else {
1241 		pmap_page_set_attributes(mdpg, VM_PAGEMD_UNCACHED);
1242 		PMAP_COUNT(page_cache_evictions);
1243 	}
1244 
1245 	for (pv_entry_t pv = &mdpg->mdpg_first; pv != NULL; pv = pv->pv_next) {
1246 		pmap_t pmap = pv->pv_pmap;
1247 		vaddr_t va = trunc_page(pv->pv_va);
1248 
1249 		KASSERT(pmap != NULL);
1250 		KASSERT(pmap != pmap_kernel() || !pmap_md_direct_mapped_vaddr_p(va));
1251 		pt_entry_t * const ptep = pmap_pte_lookup(pmap, va);
1252 		if (ptep == NULL)
1253 			continue;
1254 		pt_entry_t pte = *ptep;
1255 		if (pte_valid_p(pte)) {
1256 			pte = pte_cached_change(pte, cached);
1257 			pmap_tlb_miss_lock_enter();
1258 			pte_set(ptep, pte);
1259 			pmap_tlb_update_addr(pmap, va, pte, PMAP_TLB_NEED_IPI);
1260 			pmap_tlb_miss_lock_exit();
1261 		}
1262 	}
1263 
1264 	UVMHIST_LOG(pmaphist, " <-- done", 0, 0, 0, 0);
1265 }
1266 #endif	/* PMAP_VIRTUAL_CACHE_ALIASES && !PMAP_NO_PV_UNCACHED */
1267 
1268 /*
1269  *	Insert the given physical page (p) at
1270  *	the specified virtual address (v) in the
1271  *	target physical map with the protection requested.
1272  *
1273  *	If specified, the page will be wired down, meaning
1274  *	that the related pte can not be reclaimed.
1275  *
1276  *	NB:  This is the only routine which MAY NOT lazy-evaluate
1277  *	or lose information.  That is, this routine must actually
1278  *	insert this page into the given map NOW.
1279  */
1280 int
1281 pmap_enter(pmap_t pmap, vaddr_t va, paddr_t pa, vm_prot_t prot, u_int flags)
1282 {
1283 	const bool wired = (flags & PMAP_WIRED) != 0;
1284 	const bool is_kernel_pmap_p = (pmap == pmap_kernel());
1285 	u_int update_flags = (flags & VM_PROT_ALL) != 0 ? PMAP_TLB_INSERT : 0;
1286 #ifdef UVMHIST
1287 	struct kern_history * const histp =
1288 	    ((prot & VM_PROT_EXECUTE) ? &pmapexechist : &pmaphist);
1289 #endif
1290 
1291 	UVMHIST_FUNC(__func__);
1292 	UVMHIST_CALLARGS(*histp, "(pmap=%#jx, va=%#jx, pa=%#jx",
1293 	    (uintptr_t)pmap, va, pa, 0);
1294 	UVMHIST_LOG(*histp, "prot=%#jx flags=%#jx)", prot, flags, 0, 0);
1295 
1296 	const bool good_color = PMAP_PAGE_COLOROK_P(pa, va);
1297 	if (is_kernel_pmap_p) {
1298 		PMAP_COUNT(kernel_mappings);
1299 		if (!good_color)
1300 			PMAP_COUNT(kernel_mappings_bad);
1301 	} else {
1302 		PMAP_COUNT(user_mappings);
1303 		if (!good_color)
1304 			PMAP_COUNT(user_mappings_bad);
1305 	}
1306 	pmap_addr_range_check(pmap, va, va, __func__);
1307 
1308 	KASSERTMSG(prot & VM_PROT_READ, "no READ (%#x) in prot %#x",
1309 	    VM_PROT_READ, prot);
1310 
1311 	struct vm_page * const pg = PHYS_TO_VM_PAGE(pa);
1312 	struct vm_page_md * const mdpg = (pg ? VM_PAGE_TO_MD(pg) : NULL);
1313 
1314 	struct vm_page_md *mdpp = NULL;
1315 #ifdef __HAVE_PMAP_PV_TRACK
1316 	struct pmap_page *pp = pmap_pv_tracked(pa);
1317 	mdpp = pp ? PMAP_PAGE_TO_MD(pp) : NULL;
1318 #endif
1319 
1320 	if (mdpg) {
1321 		/* Set page referenced/modified status based on flags */
1322 		if (flags & VM_PROT_WRITE) {
1323 			pmap_page_set_attributes(mdpg, VM_PAGEMD_MODIFIED|VM_PAGEMD_REFERENCED);
1324 		} else if (flags & VM_PROT_ALL) {
1325 			pmap_page_set_attributes(mdpg, VM_PAGEMD_REFERENCED);
1326 		}
1327 
1328 #ifdef PMAP_VIRTUAL_CACHE_ALIASES
1329 		if (!VM_PAGEMD_CACHED_P(mdpg)) {
1330 			flags |= PMAP_NOCACHE;
1331 			PMAP_COUNT(uncached_mappings);
1332 		}
1333 #endif
1334 
1335 		PMAP_COUNT(managed_mappings);
1336 	} else if (mdpp) {
1337 #ifdef __HAVE_PMAP_PV_TRACK
1338 		pmap_page_set_attributes(mdpg, VM_PAGEMD_REFERENCED);
1339 
1340 		PMAP_COUNT(pvtracked_mappings);
1341 #endif
1342 	} else {
1343 		/*
1344 		 * Assumption: if it is not part of our managed memory
1345 		 * then it must be device memory which may be volatile.
1346 		 */
1347 		if ((flags & PMAP_CACHE_MASK) == 0)
1348 			flags |= PMAP_NOCACHE;
1349 		PMAP_COUNT(unmanaged_mappings);
1350 	}
1351 
1352 	KASSERTMSG(mdpg == NULL || mdpp == NULL, "mdpg %p mdpp %p", mdpg, mdpp);
1353 
1354 	struct vm_page_md *md = (mdpg != NULL) ? mdpg : mdpp;
1355 	pt_entry_t npte = pte_make_enter(pa, md, prot, flags,
1356 	    is_kernel_pmap_p);
1357 
1358 	kpreempt_disable();
1359 
1360 	pt_entry_t * const ptep = pmap_pte_reserve(pmap, va, flags);
1361 	if (__predict_false(ptep == NULL)) {
1362 		kpreempt_enable();
1363 		UVMHIST_LOG(*histp, " <-- ENOMEM", 0, 0, 0, 0);
1364 		return ENOMEM;
1365 	}
1366 	const pt_entry_t opte = *ptep;
1367 	const bool resident = pte_valid_p(opte);
1368 	bool remap = false;
1369 	if (resident) {
1370 		if (pte_to_paddr(opte) != pa) {
1371 			KASSERT(!is_kernel_pmap_p);
1372 			const pt_entry_t rpte = pte_nv_entry(false);
1373 
1374 			pmap_addr_range_check(pmap, va, va + NBPG, __func__);
1375 			pmap_pte_process(pmap, va, va + NBPG, pmap_pte_remove,
1376 			    rpte);
1377 			PMAP_COUNT(user_mappings_changed);
1378 			remap = true;
1379 		}
1380 		update_flags |= PMAP_TLB_NEED_IPI;
1381 	}
1382 
1383 	if (!resident || remap) {
1384 		pmap->pm_stats.resident_count++;
1385 	}
1386 
1387 	/* Done after case that may sleep/return. */
1388 	if (md)
1389 		pmap_enter_pv(pmap, va, pa, md, &npte, 0);
1390 
1391 	/*
1392 	 * Now validate mapping with desired protection/wiring.
1393 	 */
1394 	if (wired) {
1395 		pmap->pm_stats.wired_count++;
1396 		npte = pte_wire_entry(npte);
1397 	}
1398 
1399 	UVMHIST_LOG(*histp, "new pte %#jx (pa %#jx)",
1400 	    pte_value(npte), pa, 0, 0);
1401 
1402 	KASSERT(pte_valid_p(npte));
1403 
1404 	pmap_tlb_miss_lock_enter();
1405 	pte_set(ptep, npte);
1406 	pmap_tlb_update_addr(pmap, va, npte, update_flags);
1407 	pmap_tlb_miss_lock_exit();
1408 	kpreempt_enable();
1409 
1410 	if (pg != NULL && (prot == (VM_PROT_READ | VM_PROT_EXECUTE))) {
1411 		KASSERT(mdpg != NULL);
1412 		PMAP_COUNT(exec_mappings);
1413 		if (!VM_PAGEMD_EXECPAGE_P(mdpg) && pte_cached_p(npte)) {
1414 			if (!pte_deferred_exec_p(npte)) {
1415 				UVMHIST_LOG(*histp, "va=%#jx pg %#jx: "
1416 				    "immediate syncicache",
1417 				    va, (uintptr_t)pg, 0, 0);
1418 				pmap_page_syncicache(pg);
1419 				pmap_page_set_attributes(mdpg,
1420 				    VM_PAGEMD_EXECPAGE);
1421 				PMAP_COUNT(exec_synced_mappings);
1422 			} else {
1423 				UVMHIST_LOG(*histp, "va=%#jx pg %#jx: defer "
1424 				    "syncicache: pte %#jx",
1425 				    va, (uintptr_t)pg, npte, 0);
1426 			}
1427 		} else {
1428 			UVMHIST_LOG(*histp,
1429 			    "va=%#jx pg %#jx: no syncicache cached %jd",
1430 			    va, (uintptr_t)pg, pte_cached_p(npte), 0);
1431 		}
1432 	} else if (pg != NULL && (prot & VM_PROT_EXECUTE)) {
1433 		KASSERT(mdpg != NULL);
1434 		KASSERT(prot & VM_PROT_WRITE);
1435 		PMAP_COUNT(exec_mappings);
1436 		pmap_page_syncicache(pg);
1437 		pmap_page_clear_attributes(mdpg, VM_PAGEMD_EXECPAGE);
1438 		UVMHIST_LOG(*histp,
1439 		    "va=%#jx pg %#jx: immediate syncicache (writeable)",
1440 		    va, (uintptr_t)pg, 0, 0);
1441 	}
1442 
1443 	UVMHIST_LOG(*histp, " <-- 0 (OK)", 0, 0, 0, 0);
1444 	return 0;
1445 }
1446 
1447 void
1448 pmap_kenter_pa(vaddr_t va, paddr_t pa, vm_prot_t prot, u_int flags)
1449 {
1450 	pmap_t pmap = pmap_kernel();
1451 	struct vm_page * const pg = PHYS_TO_VM_PAGE(pa);
1452 	struct vm_page_md * const mdpg = (pg ? VM_PAGE_TO_MD(pg) : NULL);
1453 
1454 	UVMHIST_FUNC(__func__);
1455 	UVMHIST_CALLARGS(pmaphist, "(va=%#jx pa=%#jx prot=%ju, flags=%#jx)",
1456 	    va, pa, prot, flags);
1457 	PMAP_COUNT(kenter_pa);
1458 
1459 	if (mdpg == NULL) {
1460 		PMAP_COUNT(kenter_pa_unmanaged);
1461 		if ((flags & PMAP_CACHE_MASK) == 0)
1462 			flags |= PMAP_NOCACHE;
1463 	} else {
1464 		if ((flags & PMAP_NOCACHE) == 0 && !PMAP_PAGE_COLOROK_P(pa, va))
1465 			PMAP_COUNT(kenter_pa_bad);
1466 	}
1467 
1468 	pt_entry_t npte = pte_make_kenter_pa(pa, mdpg, prot, flags);
1469 	kpreempt_disable();
1470 	pt_entry_t * const ptep = pmap_pte_lookup(pmap, va);
1471 	KASSERTMSG(ptep != NULL, "%#"PRIxVADDR " %#"PRIxVADDR, va,
1472 	    pmap_limits.virtual_end);
1473 	KASSERT(!pte_valid_p(*ptep));
1474 
1475 	/*
1476 	 * No need to track non-managed pages or PMAP_KMPAGEs pages for aliases
1477 	 */
1478 #ifdef PMAP_VIRTUAL_CACHE_ALIASES
1479 	if (pg != NULL && (flags & PMAP_KMPAGE) == 0
1480 	    && pmap_md_virtual_cache_aliasing_p()) {
1481 		pmap_enter_pv(pmap, va, pa, mdpg, &npte, PV_KENTER);
1482 	}
1483 #endif
1484 
1485 	/*
1486 	 * We have the option to force this mapping into the TLB but we
1487 	 * don't.  Instead let the next reference to the page do it.
1488 	 */
1489 	pmap_tlb_miss_lock_enter();
1490 	pte_set(ptep, npte);
1491 	pmap_tlb_update_addr(pmap_kernel(), va, npte, 0);
1492 	pmap_tlb_miss_lock_exit();
1493 	kpreempt_enable();
1494 #if DEBUG > 1
1495 	for (u_int i = 0; i < PAGE_SIZE / sizeof(long); i++) {
1496 		if (((long *)va)[i] != ((long *)pa)[i])
1497 			panic("%s: contents (%lx) of va %#"PRIxVADDR
1498 			    " != contents (%lx) of pa %#"PRIxPADDR, __func__,
1499 			    ((long *)va)[i], va, ((long *)pa)[i], pa);
1500 	}
1501 #endif
1502 
1503 	UVMHIST_LOG(pmaphist, " <-- done (ptep=%#jx)", (uintptr_t)ptep, 0, 0,
1504 	    0);
1505 }
1506 
1507 /*
1508  *	Remove the given range of addresses from the kernel map.
1509  *
1510  *	It is assumed that the start and end are properly
1511  *	rounded to the page size.
1512  */
1513 
1514 static bool
1515 pmap_pte_kremove(pmap_t pmap, vaddr_t sva, vaddr_t eva, pt_entry_t *ptep,
1516 	uintptr_t flags)
1517 {
1518 	const pt_entry_t new_pte = pte_nv_entry(true);
1519 
1520 	UVMHIST_FUNC(__func__);
1521 	UVMHIST_CALLARGS(pmaphist, "(pmap=%#jx, sva=%#jx eva=%#jx ptep=%#jx)",
1522 	    (uintptr_t)pmap, sva, eva, (uintptr_t)ptep);
1523 
1524 	KASSERT(kpreempt_disabled());
1525 
1526 	for (; sva < eva; sva += NBPG, ptep++) {
1527 		pt_entry_t pte = *ptep;
1528 		if (!pte_valid_p(pte))
1529 			continue;
1530 
1531 		PMAP_COUNT(kremove_pages);
1532 #ifdef PMAP_VIRTUAL_CACHE_ALIASES
1533 		struct vm_page * const pg = PHYS_TO_VM_PAGE(pte_to_paddr(pte));
1534 		if (pg != NULL && pmap_md_virtual_cache_aliasing_p()) {
1535 			pmap_remove_pv(pmap, sva, pg, !pte_readonly_p(pte));
1536 		}
1537 #endif
1538 
1539 		pmap_tlb_miss_lock_enter();
1540 		pte_set(ptep, new_pte);
1541 		pmap_tlb_invalidate_addr(pmap, sva);
1542 		pmap_tlb_miss_lock_exit();
1543 	}
1544 
1545 	UVMHIST_LOG(pmaphist, " <-- done", 0, 0, 0, 0);
1546 
1547 	return false;
1548 }
1549 
1550 void
1551 pmap_kremove(vaddr_t va, vsize_t len)
1552 {
1553 	const vaddr_t sva = trunc_page(va);
1554 	const vaddr_t eva = round_page(va + len);
1555 
1556 	UVMHIST_FUNC(__func__);
1557 	UVMHIST_CALLARGS(pmaphist, "(va=%#jx len=%#jx)", va, len, 0, 0);
1558 
1559 	kpreempt_disable();
1560 	pmap_pte_process(pmap_kernel(), sva, eva, pmap_pte_kremove, 0);
1561 	kpreempt_enable();
1562 
1563 	UVMHIST_LOG(pmaphist, " <-- done", 0, 0, 0, 0);
1564 }
1565 
1566 bool
1567 pmap_remove_all(struct pmap *pmap)
1568 {
1569 	UVMHIST_FUNC(__func__);
1570 	UVMHIST_CALLARGS(pmaphist, "(pm=%#jx)", (uintptr_t)pmap, 0, 0, 0);
1571 
1572 	KASSERT(pmap != pmap_kernel());
1573 
1574 	kpreempt_disable();
1575 	/*
1576 	 * Free all of our ASIDs which means we can skip doing all the
1577 	 * tlb_invalidate_addrs().
1578 	 */
1579 	pmap_tlb_miss_lock_enter();
1580 #ifdef MULTIPROCESSOR
1581 	// This should be the last CPU with this pmap onproc
1582 	KASSERT(!kcpuset_isotherset(pmap->pm_onproc, cpu_index(curcpu())));
1583 	if (kcpuset_isset(pmap->pm_onproc, cpu_index(curcpu())))
1584 #endif
1585 		pmap_tlb_asid_deactivate(pmap);
1586 #ifdef MULTIPROCESSOR
1587 	KASSERT(kcpuset_iszero(pmap->pm_onproc));
1588 #endif
1589 	pmap_tlb_asid_release_all(pmap);
1590 	pmap_tlb_miss_lock_exit();
1591 	pmap->pm_flags |= PMAP_DEFERRED_ACTIVATE;
1592 
1593 #ifdef PMAP_FAULTINFO
1594 	curpcb->pcb_faultinfo.pfi_faultaddr = 0;
1595 	curpcb->pcb_faultinfo.pfi_repeats = 0;
1596 	curpcb->pcb_faultinfo.pfi_faultptep = NULL;
1597 #endif
1598 	kpreempt_enable();
1599 
1600 	UVMHIST_LOG(pmaphist, " <-- done", 0, 0, 0, 0);
1601 	return false;
1602 }
1603 
1604 /*
1605  *	Routine:	pmap_unwire
1606  *	Function:	Clear the wired attribute for a map/virtual-address
1607  *			pair.
1608  *	In/out conditions:
1609  *			The mapping must already exist in the pmap.
1610  */
1611 void
1612 pmap_unwire(pmap_t pmap, vaddr_t va)
1613 {
1614 	UVMHIST_FUNC(__func__);
1615 	UVMHIST_CALLARGS(pmaphist, "(pmap=%#jx, va=%#jx)", (uintptr_t)pmap, va,
1616 	    0, 0);
1617 	PMAP_COUNT(unwire);
1618 
1619 	/*
1620 	 * Don't need to flush the TLB since PG_WIRED is only in software.
1621 	 */
1622 	kpreempt_disable();
1623 	pmap_addr_range_check(pmap, va, va, __func__);
1624 	pt_entry_t * const ptep = pmap_pte_lookup(pmap, va);
1625 	KASSERTMSG(ptep != NULL, "pmap %p va %#"PRIxVADDR" invalid STE",
1626 	    pmap, va);
1627 	pt_entry_t pte = *ptep;
1628 	KASSERTMSG(pte_valid_p(pte),
1629 	    "pmap %p va %#" PRIxVADDR " invalid PTE %#" PRIxPTE " @ %p",
1630 	    pmap, va, pte_value(pte), ptep);
1631 
1632 	if (pte_wired_p(pte)) {
1633 		pmap_tlb_miss_lock_enter();
1634 		pte_set(ptep, pte_unwire_entry(pte));
1635 		pmap_tlb_miss_lock_exit();
1636 		pmap->pm_stats.wired_count--;
1637 	}
1638 #ifdef DIAGNOSTIC
1639 	else {
1640 		printf("%s: wiring for pmap %p va %#"PRIxVADDR" unchanged!\n",
1641 		    __func__, pmap, va);
1642 	}
1643 #endif
1644 	kpreempt_enable();
1645 
1646 	UVMHIST_LOG(pmaphist, " <-- done", 0, 0, 0, 0);
1647 }
1648 
1649 /*
1650  *	Routine:	pmap_extract
1651  *	Function:
1652  *		Extract the physical page address associated
1653  *		with the given map/virtual_address pair.
1654  */
1655 bool
1656 pmap_extract(pmap_t pmap, vaddr_t va, paddr_t *pap)
1657 {
1658 	paddr_t pa;
1659 
1660 	if (pmap == pmap_kernel()) {
1661 		if (pmap_md_direct_mapped_vaddr_p(va)) {
1662 			pa = pmap_md_direct_mapped_vaddr_to_paddr(va);
1663 			goto done;
1664 		}
1665 		if (pmap_md_io_vaddr_p(va))
1666 			panic("pmap_extract: io address %#"PRIxVADDR"", va);
1667 
1668 		if (va >= pmap_limits.virtual_end)
1669 			panic("%s: illegal kernel mapped address %#"PRIxVADDR,
1670 			    __func__, va);
1671 	}
1672 	kpreempt_disable();
1673 	const pt_entry_t * const ptep = pmap_pte_lookup(pmap, va);
1674 	if (ptep == NULL || !pte_valid_p(*ptep)) {
1675 		kpreempt_enable();
1676 		return false;
1677 	}
1678 	pa = pte_to_paddr(*ptep) | (va & PGOFSET);
1679 	kpreempt_enable();
1680 done:
1681 	if (pap != NULL) {
1682 		*pap = pa;
1683 	}
1684 	return true;
1685 }
1686 
1687 /*
1688  *	Copy the range specified by src_addr/len
1689  *	from the source map to the range dst_addr/len
1690  *	in the destination map.
1691  *
1692  *	This routine is only advisory and need not do anything.
1693  */
1694 void
1695 pmap_copy(pmap_t dst_pmap, pmap_t src_pmap, vaddr_t dst_addr, vsize_t len,
1696     vaddr_t src_addr)
1697 {
1698 	UVMHIST_FUNC(__func__);
1699 	UVMHIST_CALLED(pmaphist);
1700 	PMAP_COUNT(copy);
1701 }
1702 
1703 /*
1704  *	pmap_clear_reference:
1705  *
1706  *	Clear the reference bit on the specified physical page.
1707  */
1708 bool
1709 pmap_clear_reference(struct vm_page *pg)
1710 {
1711 	struct vm_page_md * const mdpg = VM_PAGE_TO_MD(pg);
1712 
1713 	UVMHIST_FUNC(__func__);
1714 	UVMHIST_CALLARGS(pmaphist, "(pg=%#jx (pa %#jx))",
1715 	   (uintptr_t)pg, VM_PAGE_TO_PHYS(pg), 0,0);
1716 
1717 	bool rv = pmap_page_clear_attributes(mdpg, VM_PAGEMD_REFERENCED);
1718 
1719 	UVMHIST_LOG(pmaphist, " <-- wasref %ju", rv, 0, 0, 0);
1720 
1721 	return rv;
1722 }
1723 
1724 /*
1725  *	pmap_is_referenced:
1726  *
1727  *	Return whether or not the specified physical page is referenced
1728  *	by any physical maps.
1729  */
1730 bool
1731 pmap_is_referenced(struct vm_page *pg)
1732 {
1733 	return VM_PAGEMD_REFERENCED_P(VM_PAGE_TO_MD(pg));
1734 }
1735 
1736 /*
1737  *	Clear the modify bits on the specified physical page.
1738  */
1739 bool
1740 pmap_clear_modify(struct vm_page *pg)
1741 {
1742 	struct vm_page_md * const mdpg = VM_PAGE_TO_MD(pg);
1743 	pv_entry_t pv = &mdpg->mdpg_first;
1744 	pv_entry_t pv_next;
1745 
1746 	UVMHIST_FUNC(__func__);
1747 	UVMHIST_CALLARGS(pmaphist, "(pg=%#jx (%#jx))",
1748 	    (uintptr_t)pg, VM_PAGE_TO_PHYS(pg), 0,0);
1749 	PMAP_COUNT(clear_modify);
1750 
1751 	if (VM_PAGEMD_EXECPAGE_P(mdpg)) {
1752 		if (pv->pv_pmap == NULL) {
1753 			UVMHIST_LOG(pmapexechist,
1754 			    "pg %#jx (pa %#jx): execpage cleared",
1755 			    (uintptr_t)pg, VM_PAGE_TO_PHYS(pg), 0, 0);
1756 			pmap_page_clear_attributes(mdpg, VM_PAGEMD_EXECPAGE);
1757 			PMAP_COUNT(exec_uncached_clear_modify);
1758 		} else {
1759 			UVMHIST_LOG(pmapexechist,
1760 			    "pg %#jx (pa %#jx): syncicache performed",
1761 			    (uintptr_t)pg, VM_PAGE_TO_PHYS(pg), 0, 0);
1762 			pmap_page_syncicache(pg);
1763 			PMAP_COUNT(exec_synced_clear_modify);
1764 		}
1765 	}
1766 	if (!pmap_page_clear_attributes(mdpg, VM_PAGEMD_MODIFIED)) {
1767 		UVMHIST_LOG(pmaphist, " <-- false", 0, 0, 0, 0);
1768 		return false;
1769 	}
1770 	if (pv->pv_pmap == NULL) {
1771 		UVMHIST_LOG(pmaphist, " <-- true (no mappings)", 0, 0, 0, 0);
1772 		return true;
1773 	}
1774 
1775 	/*
1776 	 * remove write access from any pages that are dirty
1777 	 * so we can tell if they are written to again later.
1778 	 * flush the VAC first if there is one.
1779 	 */
1780 	kpreempt_disable();
1781 	VM_PAGEMD_PVLIST_READLOCK(mdpg);
1782 	pmap_pvlist_check(mdpg);
1783 	for (; pv != NULL; pv = pv_next) {
1784 		pmap_t pmap = pv->pv_pmap;
1785 		vaddr_t va = trunc_page(pv->pv_va);
1786 
1787 		pv_next = pv->pv_next;
1788 #ifdef PMAP_VIRTUAL_CACHE_ALIASES
1789 		if (PV_ISKENTER_P(pv))
1790 			continue;
1791 #endif
1792 		pt_entry_t * const ptep = pmap_pte_lookup(pmap, va);
1793 		KASSERT(ptep);
1794 		pt_entry_t pte = pte_prot_nowrite(*ptep);
1795 		if (*ptep == pte) {
1796 			continue;
1797 		}
1798 		KASSERT(pte_valid_p(pte));
1799 		const uintptr_t gen = VM_PAGEMD_PVLIST_UNLOCK(mdpg);
1800 		pmap_tlb_miss_lock_enter();
1801 		pte_set(ptep, pte);
1802 		pmap_tlb_invalidate_addr(pmap, va);
1803 		pmap_tlb_miss_lock_exit();
1804 		pmap_update(pmap);
1805 		if (__predict_false(gen != VM_PAGEMD_PVLIST_READLOCK(mdpg))) {
1806 			/*
1807 			 * The list changed!  So restart from the beginning.
1808 			 */
1809 			pv_next = &mdpg->mdpg_first;
1810 			pmap_pvlist_check(mdpg);
1811 		}
1812 	}
1813 	pmap_pvlist_check(mdpg);
1814 	VM_PAGEMD_PVLIST_UNLOCK(mdpg);
1815 	kpreempt_enable();
1816 
1817 	UVMHIST_LOG(pmaphist, " <-- true (mappings changed)", 0, 0, 0, 0);
1818 	return true;
1819 }
1820 
1821 /*
1822  *	pmap_is_modified:
1823  *
1824  *	Return whether or not the specified physical page is modified
1825  *	by any physical maps.
1826  */
1827 bool
1828 pmap_is_modified(struct vm_page *pg)
1829 {
1830 	return VM_PAGEMD_MODIFIED_P(VM_PAGE_TO_MD(pg));
1831 }
1832 
1833 /*
1834  *	pmap_set_modified:
1835  *
1836  *	Sets the page modified reference bit for the specified page.
1837  */
1838 void
1839 pmap_set_modified(paddr_t pa)
1840 {
1841 	struct vm_page * const pg = PHYS_TO_VM_PAGE(pa);
1842 	struct vm_page_md * const mdpg = VM_PAGE_TO_MD(pg);
1843 	pmap_page_set_attributes(mdpg, VM_PAGEMD_MODIFIED|VM_PAGEMD_REFERENCED);
1844 }
1845 
1846 /******************** pv_entry management ********************/
1847 
1848 static void
1849 pmap_pvlist_check(struct vm_page_md *mdpg)
1850 {
1851 #ifdef DEBUG
1852 	pv_entry_t pv = &mdpg->mdpg_first;
1853 	if (pv->pv_pmap != NULL) {
1854 #ifdef PMAP_VIRTUAL_CACHE_ALIASES
1855 		const u_int colormask = uvmexp.colormask;
1856 		u_int colors = 0;
1857 #endif
1858 		for (; pv != NULL; pv = pv->pv_next) {
1859 			KASSERT(pv->pv_pmap != pmap_kernel() || !pmap_md_direct_mapped_vaddr_p(pv->pv_va));
1860 #ifdef PMAP_VIRTUAL_CACHE_ALIASES
1861 			colors |= __BIT(atop(pv->pv_va) & colormask);
1862 #endif
1863 		}
1864 #ifdef PMAP_VIRTUAL_CACHE_ALIASES
1865 		// Assert that if there is more than 1 color mapped, that the
1866 		// page is uncached.
1867 		KASSERTMSG(!pmap_md_virtual_cache_aliasing_p()
1868 		    || colors == 0 || (colors & (colors-1)) == 0
1869 		    || VM_PAGEMD_UNCACHED_P(mdpg), "colors=%#x uncached=%u",
1870 		    colors, VM_PAGEMD_UNCACHED_P(mdpg));
1871 #endif
1872 	} else {
1873 		KASSERT(pv->pv_next == NULL);
1874 	}
1875 #endif /* DEBUG */
1876 }
1877 
1878 /*
1879  * Enter the pmap and virtual address into the
1880  * physical to virtual map table.
1881  */
1882 void
1883 pmap_enter_pv(pmap_t pmap, vaddr_t va, paddr_t pa, struct vm_page_md *mdpg,
1884     pt_entry_t *nptep, u_int flags)
1885 {
1886 	pv_entry_t pv, npv, apv;
1887 #ifdef UVMHIST
1888 	bool first = false;
1889 	struct vm_page *pg = VM_PAGEMD_VMPAGE_P(mdpg) ? VM_MD_TO_PAGE(mdpg) :
1890 	    NULL;
1891 #endif
1892 
1893 	UVMHIST_FUNC(__func__);
1894 	UVMHIST_CALLARGS(pmaphist, "(pmap=%#jx va=%#jx pg=%#jx (%#jx)",
1895 	    (uintptr_t)pmap, va, (uintptr_t)pg, pa);
1896 	UVMHIST_LOG(pmaphist, "nptep=%#jx (%#jx))",
1897 	    (uintptr_t)nptep, pte_value(*nptep), 0, 0);
1898 
1899 	KASSERT(kpreempt_disabled());
1900 	KASSERT(pmap != pmap_kernel() || !pmap_md_direct_mapped_vaddr_p(va));
1901 	KASSERTMSG(pmap != pmap_kernel() || !pmap_md_io_vaddr_p(va),
1902 	    "va %#"PRIxVADDR, va);
1903 
1904 	apv = NULL;
1905 	VM_PAGEMD_PVLIST_LOCK(mdpg);
1906 again:
1907 	pv = &mdpg->mdpg_first;
1908 	pmap_pvlist_check(mdpg);
1909 	if (pv->pv_pmap == NULL) {
1910 		KASSERT(pv->pv_next == NULL);
1911 		/*
1912 		 * No entries yet, use header as the first entry
1913 		 */
1914 		PMAP_COUNT(primary_mappings);
1915 		PMAP_COUNT(mappings);
1916 #ifdef UVMHIST
1917 		first = true;
1918 #endif
1919 #ifdef PMAP_VIRTUAL_CACHE_ALIASES
1920 		KASSERT(VM_PAGEMD_CACHED_P(mdpg));
1921 		// If the new mapping has an incompatible color the last
1922 		// mapping of this page, clean the page before using it.
1923 		if (!PMAP_PAGE_COLOROK_P(va, pv->pv_va)) {
1924 			pmap_md_vca_clean(mdpg, PMAP_WBINV);
1925 		}
1926 #endif
1927 		pv->pv_pmap = pmap;
1928 		pv->pv_va = va | flags;
1929 	} else {
1930 #ifdef PMAP_VIRTUAL_CACHE_ALIASES
1931 		if (pmap_md_vca_add(mdpg, va, nptep)) {
1932 			goto again;
1933 		}
1934 #endif
1935 
1936 		/*
1937 		 * There is at least one other VA mapping this page.
1938 		 * Place this entry after the header.
1939 		 *
1940 		 * Note: the entry may already be in the table if
1941 		 * we are only changing the protection bits.
1942 		 */
1943 
1944 		for (npv = pv; npv; npv = npv->pv_next) {
1945 			if (pmap == npv->pv_pmap
1946 			    && va == trunc_page(npv->pv_va)) {
1947 #ifdef PARANOIADIAG
1948 				pt_entry_t *ptep = pmap_pte_lookup(pmap, va);
1949 				pt_entry_t pte = (ptep != NULL) ? *ptep : 0;
1950 				if (!pte_valid_p(pte) || pte_to_paddr(pte) != pa)
1951 					printf("%s: found va %#"PRIxVADDR
1952 					    " pa %#"PRIxPADDR
1953 					    " in pv_table but != %#"PRIxPTE"\n",
1954 					    __func__, va, pa, pte_value(pte));
1955 #endif
1956 				PMAP_COUNT(remappings);
1957 				VM_PAGEMD_PVLIST_UNLOCK(mdpg);
1958 				if (__predict_false(apv != NULL))
1959 					pmap_pv_free(apv);
1960 
1961 				UVMHIST_LOG(pmaphist,
1962 				    " <-- done pv=%#jx (reused)",
1963 				    (uintptr_t)pv, 0, 0, 0);
1964 				return;
1965 			}
1966 		}
1967 		if (__predict_true(apv == NULL)) {
1968 			/*
1969 			 * To allocate a PV, we have to release the PVLIST lock
1970 			 * so get the page generation.  We allocate the PV, and
1971 			 * then reacquire the lock.
1972 			 */
1973 			pmap_pvlist_check(mdpg);
1974 			const uintptr_t gen = VM_PAGEMD_PVLIST_UNLOCK(mdpg);
1975 
1976 			apv = (pv_entry_t)pmap_pv_alloc();
1977 			if (apv == NULL)
1978 				panic("pmap_enter_pv: pmap_pv_alloc() failed");
1979 
1980 			/*
1981 			 * If the generation has changed, then someone else
1982 			 * tinkered with this page so we should start over.
1983 			 */
1984 			if (gen != VM_PAGEMD_PVLIST_LOCK(mdpg))
1985 				goto again;
1986 		}
1987 		npv = apv;
1988 		apv = NULL;
1989 #ifdef PMAP_VIRTUAL_CACHE_ALIASES
1990 		/*
1991 		 * If need to deal with virtual cache aliases, keep mappings
1992 		 * in the kernel pmap at the head of the list.  This allows
1993 		 * the VCA code to easily use them for cache operations if
1994 		 * present.
1995 		 */
1996 		pmap_t kpmap = pmap_kernel();
1997 		if (pmap != kpmap) {
1998 			while (pv->pv_pmap == kpmap && pv->pv_next != NULL) {
1999 				pv = pv->pv_next;
2000 			}
2001 		}
2002 #endif
2003 		npv->pv_va = va | flags;
2004 		npv->pv_pmap = pmap;
2005 		npv->pv_next = pv->pv_next;
2006 		pv->pv_next = npv;
2007 		PMAP_COUNT(mappings);
2008 	}
2009 	pmap_pvlist_check(mdpg);
2010 	VM_PAGEMD_PVLIST_UNLOCK(mdpg);
2011 	if (__predict_false(apv != NULL))
2012 		pmap_pv_free(apv);
2013 
2014 	UVMHIST_LOG(pmaphist, " <-- done pv=%#jx (first %ju)", (uintptr_t)pv,
2015 	    first, 0, 0);
2016 }
2017 
2018 /*
2019  * Remove a physical to virtual address translation.
2020  * If cache was inhibited on this page, and there are no more cache
2021  * conflicts, restore caching.
2022  * Flush the cache if the last page is removed (should always be cached
2023  * at this point).
2024  */
2025 void
2026 pmap_remove_pv(pmap_t pmap, vaddr_t va, struct vm_page *pg, bool dirty)
2027 {
2028 	struct vm_page_md * const mdpg = VM_PAGE_TO_MD(pg);
2029 	pv_entry_t pv, npv;
2030 	bool last;
2031 
2032 	UVMHIST_FUNC(__func__);
2033 	UVMHIST_CALLARGS(pmaphist, "(pmap=%#jx, va=%#jx, pg=%#jx (pa %#jx)",
2034 	    (uintptr_t)pmap, va, (uintptr_t)pg, VM_PAGE_TO_PHYS(pg));
2035 	UVMHIST_LOG(pmaphist, "dirty=%ju)", dirty, 0, 0, 0);
2036 
2037 	KASSERT(kpreempt_disabled());
2038 	KASSERT((va & PAGE_MASK) == 0);
2039 	pv = &mdpg->mdpg_first;
2040 
2041 	VM_PAGEMD_PVLIST_LOCK(mdpg);
2042 	pmap_pvlist_check(mdpg);
2043 
2044 	/*
2045 	 * If it is the first entry on the list, it is actually
2046 	 * in the header and we must copy the following entry up
2047 	 * to the header.  Otherwise we must search the list for
2048 	 * the entry.  In either case we free the now unused entry.
2049 	 */
2050 
2051 	last = false;
2052 	if (pmap == pv->pv_pmap && va == trunc_page(pv->pv_va)) {
2053 		npv = pv->pv_next;
2054 		if (npv) {
2055 			*pv = *npv;
2056 			KASSERT(pv->pv_pmap != NULL);
2057 		} else {
2058 #ifdef PMAP_VIRTUAL_CACHE_ALIASES
2059 			pmap_page_clear_attributes(mdpg, VM_PAGEMD_UNCACHED);
2060 #endif
2061 			pv->pv_pmap = NULL;
2062 			last = true;	/* Last mapping removed */
2063 		}
2064 		PMAP_COUNT(remove_pvfirst);
2065 	} else {
2066 		for (npv = pv->pv_next; npv; pv = npv, npv = npv->pv_next) {
2067 			PMAP_COUNT(remove_pvsearch);
2068 			if (pmap == npv->pv_pmap && va == trunc_page(npv->pv_va))
2069 				break;
2070 		}
2071 		if (npv) {
2072 			pv->pv_next = npv->pv_next;
2073 		}
2074 	}
2075 
2076 	pmap_pvlist_check(mdpg);
2077 	VM_PAGEMD_PVLIST_UNLOCK(mdpg);
2078 
2079 #ifdef PMAP_VIRTUAL_CACHE_ALIASES
2080 	pmap_md_vca_remove(pg, va, dirty, last);
2081 #endif
2082 
2083 	/*
2084 	 * Free the pv_entry if needed.
2085 	 */
2086 	if (npv)
2087 		pmap_pv_free(npv);
2088 	if (VM_PAGEMD_EXECPAGE_P(mdpg) && dirty) {
2089 		if (last) {
2090 			/*
2091 			 * If this was the page's last mapping, we no longer
2092 			 * care about its execness.
2093 			 */
2094 			UVMHIST_LOG(pmapexechist,
2095 			    "pg %#jx (pa %#jx)last %ju: execpage cleared",
2096 			    (uintptr_t)pg, VM_PAGE_TO_PHYS(pg), last, 0);
2097 			pmap_page_clear_attributes(mdpg, VM_PAGEMD_EXECPAGE);
2098 			PMAP_COUNT(exec_uncached_remove);
2099 		} else {
2100 			/*
2101 			 * Someone still has it mapped as an executable page
2102 			 * so we must sync it.
2103 			 */
2104 			UVMHIST_LOG(pmapexechist,
2105 			    "pg %#jx (pa %#jx) last %ju: performed syncicache",
2106 			    (uintptr_t)pg, VM_PAGE_TO_PHYS(pg), last, 0);
2107 			pmap_page_syncicache(pg);
2108 			PMAP_COUNT(exec_synced_remove);
2109 		}
2110 	}
2111 
2112 	UVMHIST_LOG(pmaphist, " <-- done", 0, 0, 0, 0);
2113 }
2114 
2115 #if defined(MULTIPROCESSOR)
2116 struct pmap_pvlist_info {
2117 	kmutex_t *pli_locks[PAGE_SIZE / 32];
2118 	volatile u_int pli_lock_refs[PAGE_SIZE / 32];
2119 	volatile u_int pli_lock_index;
2120 	u_int pli_lock_mask;
2121 } pmap_pvlist_info;
2122 
2123 void
2124 pmap_pvlist_lock_init(size_t cache_line_size)
2125 {
2126 	struct pmap_pvlist_info * const pli = &pmap_pvlist_info;
2127 	const vaddr_t lock_page = uvm_pageboot_alloc(PAGE_SIZE);
2128 	vaddr_t lock_va = lock_page;
2129 	if (sizeof(kmutex_t) > cache_line_size) {
2130 		cache_line_size = roundup2(sizeof(kmutex_t), cache_line_size);
2131 	}
2132 	const size_t nlocks = PAGE_SIZE / cache_line_size;
2133 	KASSERT((nlocks & (nlocks - 1)) == 0);
2134 	/*
2135 	 * Now divide the page into a number of mutexes, one per cacheline.
2136 	 */
2137 	for (size_t i = 0; i < nlocks; lock_va += cache_line_size, i++) {
2138 		kmutex_t * const lock = (kmutex_t *)lock_va;
2139 		mutex_init(lock, MUTEX_DEFAULT, IPL_HIGH);
2140 		pli->pli_locks[i] = lock;
2141 	}
2142 	pli->pli_lock_mask = nlocks - 1;
2143 }
2144 
2145 kmutex_t *
2146 pmap_pvlist_lock_addr(struct vm_page_md *mdpg)
2147 {
2148 	struct pmap_pvlist_info * const pli = &pmap_pvlist_info;
2149 	kmutex_t *lock = mdpg->mdpg_lock;
2150 
2151 	/*
2152 	 * Allocate a lock on an as-needed basis.  This will hopefully give us
2153 	 * semi-random distribution not based on page color.
2154 	 */
2155 	if (__predict_false(lock == NULL)) {
2156 		size_t locknum = atomic_add_int_nv(&pli->pli_lock_index, 37);
2157 		size_t lockid = locknum & pli->pli_lock_mask;
2158 		kmutex_t * const new_lock = pli->pli_locks[lockid];
2159 		/*
2160 		 * Set the lock.  If some other thread already did, just use
2161 		 * the one they assigned.
2162 		 */
2163 		lock = atomic_cas_ptr(&mdpg->mdpg_lock, NULL, new_lock);
2164 		if (lock == NULL) {
2165 			lock = new_lock;
2166 			atomic_inc_uint(&pli->pli_lock_refs[lockid]);
2167 		}
2168 	}
2169 
2170 	/*
2171 	 * Now finally provide the lock.
2172 	 */
2173 	return lock;
2174 }
2175 #else /* !MULTIPROCESSOR */
2176 void
2177 pmap_pvlist_lock_init(size_t cache_line_size)
2178 {
2179 	mutex_init(&pmap_pvlist_mutex, MUTEX_DEFAULT, IPL_HIGH);
2180 }
2181 
2182 #ifdef MODULAR
2183 kmutex_t *
2184 pmap_pvlist_lock_addr(struct vm_page_md *mdpg)
2185 {
2186 	/*
2187 	 * We just use a global lock.
2188 	 */
2189 	if (__predict_false(mdpg->mdpg_lock == NULL)) {
2190 		mdpg->mdpg_lock = &pmap_pvlist_mutex;
2191 	}
2192 
2193 	/*
2194 	 * Now finally provide the lock.
2195 	 */
2196 	return mdpg->mdpg_lock;
2197 }
2198 #endif /* MODULAR */
2199 #endif /* !MULTIPROCESSOR */
2200 
2201 /*
2202  * pmap_pv_page_alloc:
2203  *
2204  *	Allocate a page for the pv_entry pool.
2205  */
2206 void *
2207 pmap_pv_page_alloc(struct pool *pp, int flags)
2208 {
2209 	struct vm_page * const pg = PMAP_ALLOC_POOLPAGE(UVM_PGA_USERESERVE);
2210 	if (pg == NULL)
2211 		return NULL;
2212 
2213 	return (void *)pmap_map_poolpage(VM_PAGE_TO_PHYS(pg));
2214 }
2215 
2216 /*
2217  * pmap_pv_page_free:
2218  *
2219  *	Free a pv_entry pool page.
2220  */
2221 void
2222 pmap_pv_page_free(struct pool *pp, void *v)
2223 {
2224 	vaddr_t va = (vaddr_t)v;
2225 
2226 	KASSERT(pmap_md_direct_mapped_vaddr_p(va));
2227 	const paddr_t pa = pmap_md_direct_mapped_vaddr_to_paddr(va);
2228 	struct vm_page * const pg = PHYS_TO_VM_PAGE(pa);
2229 	KASSERT(pg != NULL);
2230 #ifdef PMAP_VIRTUAL_CACHE_ALIASES
2231 	kpreempt_disable();
2232 	pmap_md_vca_remove(pg, va, true, true);
2233 	kpreempt_enable();
2234 #endif
2235 	pmap_page_clear_attributes(VM_PAGE_TO_MD(pg), VM_PAGEMD_POOLPAGE);
2236 	KASSERT(!VM_PAGEMD_EXECPAGE_P(VM_PAGE_TO_MD(pg)));
2237 	uvm_pagefree(pg);
2238 }
2239 
2240 #ifdef PMAP_PREFER
2241 /*
2242  * Find first virtual address >= *vap that doesn't cause
2243  * a cache alias conflict.
2244  */
2245 void
2246 pmap_prefer(vaddr_t foff, vaddr_t *vap, vsize_t sz, int td)
2247 {
2248 	vsize_t prefer_mask = ptoa(uvmexp.colormask);
2249 
2250 	PMAP_COUNT(prefer_requests);
2251 
2252 	prefer_mask |= pmap_md_cache_prefer_mask();
2253 
2254 	if (prefer_mask) {
2255 		vaddr_t	va = *vap;
2256 		vsize_t d = (foff - va) & prefer_mask;
2257 		if (d) {
2258 			if (td)
2259 				*vap = trunc_page(va - ((-d) & prefer_mask));
2260 			else
2261 				*vap = round_page(va + d);
2262 			PMAP_COUNT(prefer_adjustments);
2263 		}
2264 	}
2265 }
2266 #endif /* PMAP_PREFER */
2267 
2268 #ifdef PMAP_MAP_POOLPAGE
2269 vaddr_t
2270 pmap_map_poolpage(paddr_t pa)
2271 {
2272 	struct vm_page * const pg = PHYS_TO_VM_PAGE(pa);
2273 	KASSERT(pg);
2274 
2275 	struct vm_page_md * const mdpg = VM_PAGE_TO_MD(pg);
2276 	KASSERT(!VM_PAGEMD_EXECPAGE_P(mdpg));
2277 
2278 	pmap_page_set_attributes(mdpg, VM_PAGEMD_POOLPAGE);
2279 
2280 	return pmap_md_map_poolpage(pa, NBPG);
2281 }
2282 
2283 paddr_t
2284 pmap_unmap_poolpage(vaddr_t va)
2285 {
2286 	KASSERT(pmap_md_direct_mapped_vaddr_p(va));
2287 	paddr_t pa = pmap_md_direct_mapped_vaddr_to_paddr(va);
2288 
2289 	struct vm_page * const pg = PHYS_TO_VM_PAGE(pa);
2290 	KASSERT(pg != NULL);
2291 	KASSERT(!VM_PAGEMD_EXECPAGE_P(VM_PAGE_TO_MD(pg)));
2292 
2293 	pmap_page_clear_attributes(VM_PAGE_TO_MD(pg), VM_PAGEMD_POOLPAGE);
2294 	pmap_md_unmap_poolpage(va, NBPG);
2295 
2296 	return pa;
2297 }
2298 #endif /* PMAP_MAP_POOLPAGE */
2299