xref: /titanic_50/usr/src/uts/i86pc/vm/hat_i86.c (revision 1c42de6d020629af774dd9e9fc81be3f3ed9398e)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright 2007 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 #pragma ident	"%Z%%M%	%I%	%E% SMI"
27 
28 /*
29  * VM - Hardware Address Translation management for i386 and amd64
30  *
31  * Implementation of the interfaces described in <common/vm/hat.h>
32  *
33  * Nearly all the details of how the hardware is managed should not be
34  * visible outside this layer except for misc. machine specific functions
35  * that work in conjunction with this code.
36  *
37  * Routines used only inside of i86pc/vm start with hati_ for HAT Internal.
38  */
39 
40 #include <sys/machparam.h>
41 #include <sys/machsystm.h>
42 #include <sys/mman.h>
43 #include <sys/types.h>
44 #include <sys/systm.h>
45 #include <sys/cpuvar.h>
46 #include <sys/thread.h>
47 #include <sys/proc.h>
48 #include <sys/cpu.h>
49 #include <sys/kmem.h>
50 #include <sys/disp.h>
51 #include <sys/shm.h>
52 #include <sys/sysmacros.h>
53 #include <sys/machparam.h>
54 #include <sys/vmem.h>
55 #include <sys/vmsystm.h>
56 #include <sys/promif.h>
57 #include <sys/var.h>
58 #include <sys/x86_archext.h>
59 #include <sys/atomic.h>
60 #include <sys/bitmap.h>
61 #include <sys/controlregs.h>
62 #include <sys/bootconf.h>
63 #include <sys/bootsvcs.h>
64 #include <sys/bootinfo.h>
65 
66 #include <vm/seg_kmem.h>
67 #include <vm/hat_i86.h>
68 #include <vm/as.h>
69 #include <vm/seg.h>
70 #include <vm/page.h>
71 #include <vm/seg_kp.h>
72 #include <vm/seg_kpm.h>
73 #include <vm/vm_dep.h>
74 #include <vm/kboot_mmu.h>
75 
76 #include <sys/cmn_err.h>
77 
78 /*
79  * Basic parameters for hat operation.
80  */
81 struct hat_mmu_info mmu;
82 
83 /*
84  * The page that is the kernel's top level pagetable.
85  *
86  * For 32 bit VLP support, the kernel hat will use the 1st 4 entries
87  * on this 4K page for its top level page table. The remaining groups of
88  * 4 entries are used for per processor copies of user VLP pagetables for
89  * running threads.  See hat_switch() and reload_pae32() for details.
90  *
91  * vlp_page[0] - 0th level==2 PTE for kernel HAT (will be zero)
92  * vlp_page[1] - 1st level==2 PTE for kernel HAT (will be zero)
93  * vlp_page[2] - 2nd level==2 PTE for kernel HAT (zero for small memory)
94  * vlp_page[3] - 3rd level==2 PTE for kernel
95  *
96  * vlp_page[4] - 0th level==2 PTE for user thread on cpu 0
97  * vlp_page[5] - 1st level==2 PTE for user thread on cpu 0
98  * vlp_page[6] - 2nd level==2 PTE for user thread on cpu 0
99  * vlp_page[7] - probably copy of kernel PTE
100  *
101  * vlp_page[8]  - 0th level==2 PTE for user thread on cpu 1
102  * vlp_page[9]  - 1st level==2 PTE for user thread on cpu 1
103  * vlp_page[10] - 2nd level==2 PTE for user thread on cpu 1
104  * vlp_page[11] - probably copy of kernel PTE
105  * ...
106  *
107  * when / where the kernel PTE's are (entry 2 or 3 or none) depends
108  * on kernelbase.
109  */
110 static x86pte_t *vlp_page;
111 
112 /*
113  * forward declaration of internal utility routines
114  */
115 static x86pte_t hati_update_pte(htable_t *ht, uint_t entry, x86pte_t expected,
116 	x86pte_t new);
117 
118 /*
119  * The kernel address space exists in all HATs. To implement this the
120  * kernel reserves a fixed number of entries in every topmost level page
121  * table. The values are setup in hat_init() and then copied to every hat
122  * created by hat_alloc(). This means that kernelbase must be:
123  *
124  *	  4Meg aligned for 32 bit kernels
125  *	512Gig aligned for x86_64 64 bit kernel
126  *
127  * The PAE 32 bit hat is handled as a special case. Otherwise requiring 1Gig
128  * alignment would use too much VA for the kernel.
129  *
130  */
131 static uint_t	khat_start;	/* index of 1st entry in kernel's top ptable */
132 static uint_t	khat_entries;	/* number of entries in kernel's top ptable */
133 
134 #if defined(__i386)
135 
136 static htable_t	*khat_pae32_htable = NULL;
137 static uint_t	khat_pae32_start;
138 static uint_t	khat_pae32_entries;
139 
140 #endif
141 
142 uint_t use_boot_reserve = 1;	/* cleared after early boot process */
143 uint_t can_steal_post_boot = 0;	/* set late in boot to enable stealing */
144 
145 /*
146  * A cpuset for all cpus. This is used for kernel address cross calls, since
147  * the kernel addresses apply to all cpus.
148  */
149 cpuset_t khat_cpuset;
150 
151 /*
152  * management stuff for hat structures
153  */
154 kmutex_t	hat_list_lock;
155 kcondvar_t	hat_list_cv;
156 kmem_cache_t	*hat_cache;
157 kmem_cache_t	*hat_hash_cache;
158 kmem_cache_t	*vlp_hash_cache;
159 
160 /*
161  * Simple statistics
162  */
163 struct hatstats hatstat;
164 
165 /*
166  * useful stuff for atomic access/clearing/setting REF/MOD/RO bits in page_t's.
167  */
168 extern void atomic_orb(uchar_t *addr, uchar_t val);
169 extern void atomic_andb(uchar_t *addr, uchar_t val);
170 
171 #define	PP_GETRM(pp, rmmask)    (pp->p_nrm & rmmask)
172 #define	PP_ISMOD(pp)		PP_GETRM(pp, P_MOD)
173 #define	PP_ISREF(pp)		PP_GETRM(pp, P_REF)
174 #define	PP_ISRO(pp)		PP_GETRM(pp, P_RO)
175 
176 #define	PP_SETRM(pp, rm)	atomic_orb(&(pp->p_nrm), rm)
177 #define	PP_SETMOD(pp)		PP_SETRM(pp, P_MOD)
178 #define	PP_SETREF(pp)		PP_SETRM(pp, P_REF)
179 #define	PP_SETRO(pp)		PP_SETRM(pp, P_RO)
180 
181 #define	PP_CLRRM(pp, rm)	atomic_andb(&(pp->p_nrm), ~(rm))
182 #define	PP_CLRMOD(pp)   	PP_CLRRM(pp, P_MOD)
183 #define	PP_CLRREF(pp)   	PP_CLRRM(pp, P_REF)
184 #define	PP_CLRRO(pp)    	PP_CLRRM(pp, P_RO)
185 #define	PP_CLRALL(pp)		PP_CLRRM(pp, P_MOD | P_REF | P_RO)
186 
187 /*
188  * some useful tracing macros
189  */
190 
191 int hattrace = 0;
192 #ifdef DEBUG
193 
194 #define	HATIN(r, h, a, l)	\
195 	if (hattrace) prom_printf("->%s hat=%p, adr=%p, len=%lx\n", #r, h, a, l)
196 
197 #define	HATOUT(r, h, a)		\
198 	if (hattrace) prom_printf("<-%s hat=%p, adr=%p\n", #r, h, a)
199 #else
200 
201 #define	HATIN(r, h, a, l)
202 #define	HATOUT(r, h, a)
203 
204 #endif
205 
206 
207 /*
208  * kmem cache constructor for struct hat
209  */
210 /*ARGSUSED*/
211 static int
212 hati_constructor(void *buf, void *handle, int kmflags)
213 {
214 	hat_t	*hat = buf;
215 
216 	mutex_init(&hat->hat_mutex, NULL, MUTEX_DEFAULT, NULL);
217 	bzero(hat->hat_pages_mapped,
218 	    sizeof (pgcnt_t) * (mmu.max_page_level + 1));
219 	hat->hat_stats = 0;
220 	hat->hat_flags = 0;
221 	mutex_init(&hat->hat_switch_mutex, NULL, MUTEX_DRIVER,
222 	    (void *)ipltospl(DISP_LEVEL));
223 	CPUSET_ZERO(hat->hat_cpus);
224 	hat->hat_htable = NULL;
225 	hat->hat_ht_hash = NULL;
226 	return (0);
227 }
228 
229 /*
230  * Allocate a hat structure for as. We also create the top level
231  * htable and initialize it to contain the kernel hat entries.
232  */
233 hat_t *
234 hat_alloc(struct as *as)
235 {
236 	hat_t		*hat;
237 	htable_t	*ht;	/* top level htable */
238 	uint_t		use_vlp;
239 
240 	/*
241 	 * Once we start creating user process HATs we can enable
242 	 * the htable_steal() code.
243 	 */
244 	if (can_steal_post_boot == 0)
245 		can_steal_post_boot = 1;
246 
247 	ASSERT(AS_WRITE_HELD(as, &as->a_lock));
248 	hat = kmem_cache_alloc(hat_cache, KM_SLEEP);
249 	hat->hat_as = as;
250 	mutex_init(&hat->hat_mutex, NULL, MUTEX_DEFAULT, NULL);
251 	ASSERT(hat->hat_flags == 0);
252 
253 	/*
254 	 * a 32 bit process uses a VLP style hat when using PAE
255 	 */
256 #if defined(__amd64)
257 	use_vlp = (ttoproc(curthread)->p_model == DATAMODEL_ILP32);
258 #elif defined(__i386)
259 	use_vlp = mmu.pae_hat;
260 #endif
261 	if (use_vlp) {
262 		hat->hat_flags = HAT_VLP;
263 		bzero(hat->hat_vlp_ptes, VLP_SIZE);
264 	}
265 
266 	/*
267 	 * Allocate the htable hash
268 	 */
269 	if ((hat->hat_flags & HAT_VLP)) {
270 		hat->hat_num_hash = mmu.vlp_hash_cnt;
271 		hat->hat_ht_hash = kmem_cache_alloc(vlp_hash_cache, KM_SLEEP);
272 	} else {
273 		hat->hat_num_hash = mmu.hash_cnt;
274 		hat->hat_ht_hash = kmem_cache_alloc(hat_hash_cache, KM_SLEEP);
275 	}
276 	bzero(hat->hat_ht_hash, hat->hat_num_hash * sizeof (htable_t *));
277 
278 	/*
279 	 * Initialize Kernel HAT entries at the top of the top level page
280 	 * table for the new hat.
281 	 *
282 	 * Note that we don't call htable_release() for the top level, that
283 	 * happens when the hat is destroyed in hat_free_end()
284 	 */
285 	hat->hat_htable = NULL;
286 	hat->hat_ht_cached = NULL;
287 	ht = htable_create(hat, (uintptr_t)0, TOP_LEVEL(hat), NULL);
288 
289 	if (!(hat->hat_flags & HAT_VLP))
290 		x86pte_copy(kas.a_hat->hat_htable, ht, khat_start,
291 		    khat_entries);
292 #if defined(__i386)
293 	else if (khat_entries > 0)
294 		bcopy(vlp_page + khat_start, hat->hat_vlp_ptes + khat_start,
295 		    khat_entries * sizeof (x86pte_t));
296 #endif
297 	hat->hat_htable = ht;
298 
299 #if defined(__i386)
300 	/*
301 	 * PAE32 HAT alignment is less restrictive than the others to keep
302 	 * the kernel from using too much VA. Because of this we may need
303 	 * one layer further down when kernelbase isn't 1Gig aligned.
304 	 * See hat_free_end() for the htable_release() that goes with this
305 	 * htable_create()
306 	 */
307 	if (khat_pae32_htable != NULL) {
308 		ht = htable_create(hat, kernelbase,
309 		    khat_pae32_htable->ht_level, NULL);
310 		x86pte_copy(khat_pae32_htable, ht, khat_pae32_start,
311 		    khat_pae32_entries);
312 		ht->ht_valid_cnt = khat_pae32_entries;
313 	}
314 #endif
315 
316 	/*
317 	 * Put it at the start of the global list of all hats (used by stealing)
318 	 *
319 	 * kas.a_hat is not in the list but is instead used to find the
320 	 * first and last items in the list.
321 	 *
322 	 * - kas.a_hat->hat_next points to the start of the user hats.
323 	 *   The list ends where hat->hat_next == NULL
324 	 *
325 	 * - kas.a_hat->hat_prev points to the last of the user hats.
326 	 *   The list begins where hat->hat_prev == NULL
327 	 */
328 	mutex_enter(&hat_list_lock);
329 	hat->hat_prev = NULL;
330 	hat->hat_next = kas.a_hat->hat_next;
331 	if (hat->hat_next)
332 		hat->hat_next->hat_prev = hat;
333 	else
334 		kas.a_hat->hat_prev = hat;
335 	kas.a_hat->hat_next = hat;
336 	mutex_exit(&hat_list_lock);
337 
338 	return (hat);
339 }
340 
341 /*
342  * process has finished executing but as has not been cleaned up yet.
343  */
344 /*ARGSUSED*/
345 void
346 hat_free_start(hat_t *hat)
347 {
348 	ASSERT(AS_WRITE_HELD(hat->hat_as, &hat->hat_as->a_lock));
349 
350 	/*
351 	 * If the hat is currently a stealing victim, wait for the stealing
352 	 * to finish.  Once we mark it as HAT_FREEING, htable_steal()
353 	 * won't look at its pagetables anymore.
354 	 */
355 	mutex_enter(&hat_list_lock);
356 	while (hat->hat_flags & HAT_VICTIM)
357 		cv_wait(&hat_list_cv, &hat_list_lock);
358 	hat->hat_flags |= HAT_FREEING;
359 	mutex_exit(&hat_list_lock);
360 }
361 
362 /*
363  * An address space is being destroyed, so we destroy the associated hat.
364  */
365 void
366 hat_free_end(hat_t *hat)
367 {
368 	int i;
369 	kmem_cache_t *cache;
370 
371 #ifdef DEBUG
372 	for (i = 0; i <= mmu.max_page_level; i++)
373 		ASSERT(hat->hat_pages_mapped[i] == 0);
374 #endif
375 	ASSERT(hat->hat_flags & HAT_FREEING);
376 
377 	/*
378 	 * must not be running on the given hat
379 	 */
380 	ASSERT(CPU->cpu_current_hat != hat);
381 
382 	/*
383 	 * Remove it from the list of HATs
384 	 */
385 	mutex_enter(&hat_list_lock);
386 	if (hat->hat_prev)
387 		hat->hat_prev->hat_next = hat->hat_next;
388 	else
389 		kas.a_hat->hat_next = hat->hat_next;
390 	if (hat->hat_next)
391 		hat->hat_next->hat_prev = hat->hat_prev;
392 	else
393 		kas.a_hat->hat_prev = hat->hat_prev;
394 	mutex_exit(&hat_list_lock);
395 	hat->hat_next = hat->hat_prev = NULL;
396 
397 	/*
398 	 * Make a pass through the htables freeing them all up.
399 	 */
400 	htable_purge_hat(hat);
401 
402 	/*
403 	 * Decide which kmem cache the hash table came from, then free it.
404 	 */
405 	if (hat->hat_flags & HAT_VLP)
406 		cache = vlp_hash_cache;
407 	else
408 		cache = hat_hash_cache;
409 	kmem_cache_free(cache, hat->hat_ht_hash);
410 	hat->hat_ht_hash = NULL;
411 
412 	hat->hat_flags = 0;
413 	kmem_cache_free(hat_cache, hat);
414 }
415 
416 /*
417  * round kernelbase down to a supported value to use for _userlimit
418  *
419  * userlimit must be aligned down to an entry in the top level htable.
420  * The one exception is for 32 bit HAT's running PAE.
421  */
422 uintptr_t
423 hat_kernelbase(uintptr_t va)
424 {
425 #if defined(__i386)
426 	va &= LEVEL_MASK(1);
427 #endif
428 	if (IN_VA_HOLE(va))
429 		panic("_userlimit %p will fall in VA hole\n", (void *)va);
430 	return (va);
431 }
432 
433 /*
434  * Initialize hat data structures based on processor MMU information.
435  */
436 void
437 mmu_init(void)
438 {
439 	uint_t max_htables;
440 	uint_t pa_bits;
441 	uint_t va_bits;
442 	int i;
443 
444 	/*
445 	 * If CPU enabled the page table global bit, use it for the kernel
446 	 * This is bit 7 in CR4 (PGE - Page Global Enable).
447 	 */
448 	if ((x86_feature & X86_PGE) != 0 && (getcr4() & CR4_PGE) != 0)
449 		mmu.pt_global = PT_GLOBAL;
450 
451 	/*
452 	 * Detect NX and PAE usage.
453 	 */
454 	mmu.pae_hat = kbm_pae_support;
455 	if (kbm_nx_support)
456 		mmu.pt_nx = PT_NX;
457 	else
458 		mmu.pt_nx = 0;
459 
460 	/*
461 	 * Use CPU info to set various MMU parameters
462 	 */
463 	cpuid_get_addrsize(CPU, &pa_bits, &va_bits);
464 
465 	if (va_bits < sizeof (void *) * NBBY) {
466 		mmu.hole_start = (1ul << (va_bits - 1));
467 		mmu.hole_end = 0ul - mmu.hole_start - 1;
468 	} else {
469 		mmu.hole_end = 0;
470 		mmu.hole_start = mmu.hole_end - 1;
471 	}
472 #if defined(OPTERON_ERRATUM_121)
473 	/*
474 	 * If erratum 121 has already been detected at this time, hole_start
475 	 * contains the value to be subtracted from mmu.hole_start.
476 	 */
477 	ASSERT(hole_start == 0 || opteron_erratum_121 != 0);
478 	hole_start = mmu.hole_start - hole_start;
479 #else
480 	hole_start = mmu.hole_start;
481 #endif
482 	hole_end = mmu.hole_end;
483 
484 	mmu.highest_pfn = mmu_btop((1ull << pa_bits) - 1);
485 	if (mmu.pae_hat == 0 && pa_bits > 32)
486 		mmu.highest_pfn = PFN_4G - 1;
487 
488 	if (mmu.pae_hat) {
489 		mmu.pte_size = 8;	/* 8 byte PTEs */
490 		mmu.pte_size_shift = 3;
491 	} else {
492 		mmu.pte_size = 4;	/* 4 byte PTEs */
493 		mmu.pte_size_shift = 2;
494 	}
495 
496 	if (mmu.pae_hat && (x86_feature & X86_PAE) == 0)
497 		panic("Processor does not support PAE");
498 
499 	if ((x86_feature & X86_CX8) == 0)
500 		panic("Processor does not support cmpxchg8b instruction");
501 
502 	/*
503 	 * Initialize parameters based on the 64 or 32 bit kernels and
504 	 * for the 32 bit kernel decide if we should use PAE.
505 	 */
506 	if (kbm_largepage_support)
507 		mmu.max_page_level = 1;
508 	else
509 		mmu.max_page_level = 0;
510 	mmu_page_sizes = mmu.max_page_level + 1;
511 	mmu_exported_page_sizes = mmu_page_sizes;
512 
513 #if defined(__amd64)
514 
515 	mmu.num_level = 4;
516 	mmu.max_level = 3;
517 	mmu.ptes_per_table = 512;
518 	mmu.top_level_count = 512;
519 
520 	mmu.level_shift[0] = 12;
521 	mmu.level_shift[1] = 21;
522 	mmu.level_shift[2] = 30;
523 	mmu.level_shift[3] = 39;
524 
525 #elif defined(__i386)
526 
527 	if (mmu.pae_hat) {
528 		mmu.num_level = 3;
529 		mmu.max_level = 2;
530 		mmu.ptes_per_table = 512;
531 		mmu.top_level_count = 4;
532 
533 		mmu.level_shift[0] = 12;
534 		mmu.level_shift[1] = 21;
535 		mmu.level_shift[2] = 30;
536 
537 	} else {
538 		mmu.num_level = 2;
539 		mmu.max_level = 1;
540 		mmu.ptes_per_table = 1024;
541 		mmu.top_level_count = 1024;
542 
543 		mmu.level_shift[0] = 12;
544 		mmu.level_shift[1] = 22;
545 	}
546 
547 #endif	/* __i386 */
548 
549 	for (i = 0; i < mmu.num_level; ++i) {
550 		mmu.level_size[i] = 1UL << mmu.level_shift[i];
551 		mmu.level_offset[i] = mmu.level_size[i] - 1;
552 		mmu.level_mask[i] = ~mmu.level_offset[i];
553 	}
554 
555 	for (i = 0; i <= mmu.max_page_level; ++i) {
556 		mmu.pte_bits[i] = PT_VALID;
557 		if (i > 0)
558 			mmu.pte_bits[i] |= PT_PAGESIZE;
559 	}
560 
561 	/*
562 	 * NOTE Legacy 32 bit PAE mode only has the P_VALID bit at top level.
563 	 */
564 	for (i = 1; i < mmu.num_level; ++i)
565 		mmu.ptp_bits[i] = PT_PTPBITS;
566 
567 #if defined(__i386)
568 	mmu.ptp_bits[2] = PT_VALID;
569 #endif
570 
571 	/*
572 	 * Compute how many hash table entries to have per process for htables.
573 	 * We start with 1 page's worth of entries.
574 	 *
575 	 * If physical memory is small, reduce the amount need to cover it.
576 	 */
577 	max_htables = physmax / mmu.ptes_per_table;
578 	mmu.hash_cnt = MMU_PAGESIZE / sizeof (htable_t *);
579 	while (mmu.hash_cnt > 16 && mmu.hash_cnt >= max_htables)
580 		mmu.hash_cnt >>= 1;
581 	mmu.vlp_hash_cnt = mmu.hash_cnt;
582 
583 #if defined(__amd64)
584 	/*
585 	 * If running in 64 bits and physical memory is large,
586 	 * increase the size of the cache to cover all of memory for
587 	 * a 64 bit process.
588 	 */
589 #define	HASH_MAX_LENGTH 4
590 	while (mmu.hash_cnt * HASH_MAX_LENGTH < max_htables)
591 		mmu.hash_cnt <<= 1;
592 #endif
593 }
594 
595 
596 /*
597  * initialize hat data structures
598  */
599 void
600 hat_init()
601 {
602 #if defined(__i386)
603 	/*
604 	 * _userlimit must be aligned correctly
605 	 */
606 	if ((_userlimit & LEVEL_MASK(1)) != _userlimit) {
607 		prom_printf("hat_init(): _userlimit=%p, not aligned at %p\n",
608 		    (void *)_userlimit, (void *)LEVEL_SIZE(1));
609 		halt("hat_init(): Unable to continue");
610 	}
611 #endif
612 
613 	cv_init(&hat_list_cv, NULL, CV_DEFAULT, NULL);
614 
615 	/*
616 	 * initialize kmem caches
617 	 */
618 	htable_init();
619 	hment_init();
620 
621 	hat_cache = kmem_cache_create("hat_t",
622 	    sizeof (hat_t), 0, hati_constructor, NULL, NULL,
623 	    NULL, 0, 0);
624 
625 	hat_hash_cache = kmem_cache_create("HatHash",
626 	    mmu.hash_cnt * sizeof (htable_t *), 0, NULL, NULL, NULL,
627 	    NULL, 0, 0);
628 
629 	/*
630 	 * VLP hats can use a smaller hash table size on large memroy machines
631 	 */
632 	if (mmu.hash_cnt == mmu.vlp_hash_cnt) {
633 		vlp_hash_cache = hat_hash_cache;
634 	} else {
635 		vlp_hash_cache = kmem_cache_create("HatVlpHash",
636 		    mmu.vlp_hash_cnt * sizeof (htable_t *), 0, NULL, NULL, NULL,
637 		    NULL, 0, 0);
638 	}
639 
640 	/*
641 	 * Set up the kernel's hat
642 	 */
643 	AS_LOCK_ENTER(&kas, &kas.a_lock, RW_WRITER);
644 	kas.a_hat = kmem_cache_alloc(hat_cache, KM_NOSLEEP);
645 	mutex_init(&kas.a_hat->hat_mutex, NULL, MUTEX_DEFAULT, NULL);
646 	kas.a_hat->hat_as = &kas;
647 	kas.a_hat->hat_flags = 0;
648 	AS_LOCK_EXIT(&kas, &kas.a_lock);
649 
650 	CPUSET_ZERO(khat_cpuset);
651 	CPUSET_ADD(khat_cpuset, CPU->cpu_id);
652 
653 	/*
654 	 * The kernel hat's next pointer serves as the head of the hat list .
655 	 * The kernel hat's prev pointer tracks the last hat on the list for
656 	 * htable_steal() to use.
657 	 */
658 	kas.a_hat->hat_next = NULL;
659 	kas.a_hat->hat_prev = NULL;
660 
661 	/*
662 	 * Allocate an htable hash bucket for the kernel
663 	 * XX64 - tune for 64 bit procs
664 	 */
665 	kas.a_hat->hat_num_hash = mmu.hash_cnt;
666 	kas.a_hat->hat_ht_hash = kmem_cache_alloc(hat_hash_cache, KM_NOSLEEP);
667 	bzero(kas.a_hat->hat_ht_hash, mmu.hash_cnt * sizeof (htable_t *));
668 
669 	/*
670 	 * zero out the top level and cached htable pointers
671 	 */
672 	kas.a_hat->hat_ht_cached = NULL;
673 	kas.a_hat->hat_htable = NULL;
674 
675 	/*
676 	 * Pre-allocate hrm_hashtab before enabling the collection of
677 	 * refmod statistics.  Allocating on the fly would mean us
678 	 * running the risk of suffering recursive mutex enters or
679 	 * deadlocks.
680 	 */
681 	hrm_hashtab = kmem_zalloc(HRM_HASHSIZE * sizeof (struct hrmstat *),
682 	    KM_SLEEP);
683 }
684 
685 /*
686  * Prepare CPU specific pagetables for VLP processes on 64 bit kernels.
687  *
688  * Each CPU has a set of 2 pagetables that are reused for any 32 bit
689  * process it runs. They are the top level pagetable, hci_vlp_l3ptes, and
690  * the next to top level table for the bottom 512 Gig, hci_vlp_l2ptes.
691  */
692 /*ARGSUSED*/
693 static void
694 hat_vlp_setup(struct cpu *cpu)
695 {
696 #if defined(__amd64)
697 	struct hat_cpu_info *hci = cpu->cpu_hat_info;
698 	pfn_t pfn;
699 
700 	/*
701 	 * allocate the level==2 page table for the bottom most
702 	 * 512Gig of address space (this is where 32 bit apps live)
703 	 */
704 	ASSERT(hci != NULL);
705 	hci->hci_vlp_l2ptes = kmem_zalloc(MMU_PAGESIZE, KM_SLEEP);
706 
707 	/*
708 	 * Allocate a top level pagetable and copy the kernel's
709 	 * entries into it. Then link in hci_vlp_l2ptes in the 1st entry.
710 	 */
711 	hci->hci_vlp_l3ptes = kmem_zalloc(MMU_PAGESIZE, KM_SLEEP);
712 	hci->hci_vlp_pfn =
713 	    hat_getpfnum(kas.a_hat, (caddr_t)hci->hci_vlp_l3ptes);
714 	ASSERT(hci->hci_vlp_pfn != PFN_INVALID);
715 	bcopy(vlp_page + khat_start, hci->hci_vlp_l3ptes + khat_start,
716 	    khat_entries * sizeof (x86pte_t));
717 
718 	pfn = hat_getpfnum(kas.a_hat, (caddr_t)hci->hci_vlp_l2ptes);
719 	ASSERT(pfn != PFN_INVALID);
720 	hci->hci_vlp_l3ptes[0] = MAKEPTP(pfn, 2);
721 #endif /* __amd64 */
722 }
723 
724 /*ARGSUSED*/
725 static void
726 hat_vlp_teardown(cpu_t *cpu)
727 {
728 #if defined(__amd64)
729 	struct hat_cpu_info *hci;
730 
731 	if ((hci = cpu->cpu_hat_info) == NULL)
732 		return;
733 	if (hci->hci_vlp_l2ptes)
734 		kmem_free(hci->hci_vlp_l2ptes, MMU_PAGESIZE);
735 	if (hci->hci_vlp_l3ptes)
736 		kmem_free(hci->hci_vlp_l3ptes, MMU_PAGESIZE);
737 #endif	/* __amd64 */
738 }
739 
740 /*
741  * Finish filling in the kernel hat.
742  * Pre fill in all top level kernel page table entries for the kernel's
743  * part of the address range.  From this point on we can't use any new
744  * kernel large pages if they need PTE's at max_level
745  *
746  * create the kmap mappings.
747  */
748 void
749 hat_init_finish(void)
750 {
751 	htable_t	*top = kas.a_hat->hat_htable;
752 	htable_t	*ht;
753 	uint_t		e;
754 	x86pte_t	pte;
755 	uintptr_t	va = kernelbase;
756 	size_t		size;
757 
758 
759 #if defined(__i386)
760 	ASSERT((va & LEVEL_MASK(1)) == va);
761 
762 	/*
763 	 * Deal with kernelbase not 1Gig aligned for 32 bit PAE hats.
764 	 */
765 	if (!mmu.pae_hat || (va & LEVEL_OFFSET(mmu.max_level)) == 0) {
766 		khat_pae32_htable = NULL;
767 	} else {
768 		ASSERT(mmu.max_level == 2);
769 		ASSERT((va & LEVEL_OFFSET(mmu.max_level - 1)) == 0);
770 		khat_pae32_htable =
771 		    htable_create(kas.a_hat, va, mmu.max_level - 1, NULL);
772 		khat_pae32_start = htable_va2entry(va, khat_pae32_htable);
773 		khat_pae32_entries = mmu.ptes_per_table - khat_pae32_start;
774 		for (e = khat_pae32_start; e < mmu.ptes_per_table;
775 		    ++e, va += LEVEL_SIZE(mmu.max_level - 1)) {
776 			pte = x86pte_get(khat_pae32_htable, e);
777 			if (PTE_ISVALID(pte))
778 				continue;
779 			ht = htable_create(kas.a_hat, va, mmu.max_level - 2,
780 			    NULL);
781 			ASSERT(ht != NULL);
782 		}
783 	}
784 #endif
785 
786 	/*
787 	 * The kernel hat will need fixed values in the highest level
788 	 * ptable for copying to all other hat's. This implies
789 	 * alignment restrictions on _userlimit.
790 	 *
791 	 * Note we don't htable_release() these htables. This keeps them
792 	 * from ever being stolen or free'd.
793 	 *
794 	 * top_level_count is used instead of ptes_per_table, since
795 	 * on 32-bit PAE we only have 4 usable entries at the top level ptable.
796 	 */
797 	if (va == 0)
798 		khat_start = mmu.top_level_count;
799 	else
800 		khat_start = htable_va2entry(va, kas.a_hat->hat_htable);
801 	khat_entries = mmu.top_level_count - khat_start;
802 	for (e = khat_start; e < mmu.top_level_count;
803 	    ++e, va += LEVEL_SIZE(mmu.max_level)) {
804 		if (IN_HYPERVISOR_VA(va))
805 			continue;
806 		pte = x86pte_get(top, e);
807 		if (PTE_ISVALID(pte))
808 			continue;
809 		ht = htable_create(kas.a_hat, va, mmu.max_level - 1, NULL);
810 		ASSERT(ht != NULL);
811 	}
812 
813 	/*
814 	 * We are now effectively running on the kernel hat.
815 	 * Clearing use_boot_reserve shuts off using the pre-allocated boot
816 	 * reserve for all HAT allocations.  From here on, the reserves are
817 	 * only used when mapping in memory for the hat's own allocations.
818 	 */
819 	use_boot_reserve = 0;
820 	htable_adjust_reserve();
821 
822 	/*
823 	 * 32 bit kernels use only 4 of the 512 entries in its top level
824 	 * pagetable. We'll use the remainder for the "per CPU" page tables
825 	 * for VLP processes.
826 	 *
827 	 * We also map the top level kernel pagetable into the kernel to make
828 	 * it easy to use bcopy to initialize new address spaces.
829 	 */
830 	if (mmu.pae_hat) {
831 		vlp_page = vmem_alloc(heap_arena, MMU_PAGESIZE, VM_SLEEP);
832 		hat_devload(kas.a_hat, (caddr_t)vlp_page, MMU_PAGESIZE,
833 		    kas.a_hat->hat_htable->ht_pfn,
834 		    PROT_WRITE |
835 		    PROT_READ | HAT_NOSYNC | HAT_UNORDERED_OK,
836 		    HAT_LOAD | HAT_LOAD_NOCONSIST);
837 	}
838 	hat_vlp_setup(CPU);
839 
840 	/*
841 	 * Create kmap (cached mappings of kernel PTEs)
842 	 * for 32 bit we map from segmap_start .. ekernelheap
843 	 * for 64 bit we map from segmap_start .. segmap_start + segmapsize;
844 	 */
845 #if defined(__i386)
846 	size = (uintptr_t)ekernelheap - segmap_start;
847 #elif defined(__amd64)
848 	size = segmapsize;
849 #endif
850 	hat_kmap_init((uintptr_t)segmap_start, size);
851 }
852 
853 /*
854  * On 32 bit PAE mode, PTE's are 64 bits, but ordinary atomic memory references
855  * are 32 bit, so for safety we must use cas64() to install these.
856  */
857 #ifdef __i386
858 static void
859 reload_pae32(hat_t *hat, cpu_t *cpu)
860 {
861 	x86pte_t *src;
862 	x86pte_t *dest;
863 	x86pte_t pte;
864 	int i;
865 
866 	/*
867 	 * Load the 4 entries of the level 2 page table into this
868 	 * cpu's range of the vlp_page and point cr3 at them.
869 	 */
870 	ASSERT(mmu.pae_hat);
871 	src = hat->hat_vlp_ptes;
872 	dest = vlp_page + (cpu->cpu_id + 1) * VLP_NUM_PTES;
873 	for (i = 0; i < VLP_NUM_PTES; ++i) {
874 		for (;;) {
875 			pte = dest[i];
876 			if (pte == src[i])
877 				break;
878 			if (cas64(dest + i, pte, src[i]) != src[i])
879 				break;
880 		}
881 	}
882 }
883 #endif
884 
885 /*
886  * Switch to a new active hat, maintaining bit masks to track active CPUs.
887  */
888 void
889 hat_switch(hat_t *hat)
890 {
891 	uintptr_t	newcr3;
892 	cpu_t		*cpu = CPU;
893 	hat_t		*old = cpu->cpu_current_hat;
894 
895 	/*
896 	 * set up this information first, so we don't miss any cross calls
897 	 */
898 	if (old != NULL) {
899 		if (old == hat)
900 			return;
901 		if (old != kas.a_hat)
902 			CPUSET_ATOMIC_DEL(old->hat_cpus, cpu->cpu_id);
903 	}
904 
905 	/*
906 	 * Wait for any in flight pagetable invalidates on this hat to finish.
907 	 * This is a spin lock at DISP_LEVEL
908 	 */
909 	if (hat != kas.a_hat) {
910 		mutex_enter(&hat->hat_switch_mutex);
911 		CPUSET_ATOMIC_ADD(hat->hat_cpus, cpu->cpu_id);
912 		mutex_exit(&hat->hat_switch_mutex);
913 	}
914 	cpu->cpu_current_hat = hat;
915 
916 	/*
917 	 * now go ahead and load cr3
918 	 */
919 	if (hat->hat_flags & HAT_VLP) {
920 #if defined(__amd64)
921 		x86pte_t *vlpptep = cpu->cpu_hat_info->hci_vlp_l2ptes;
922 
923 		VLP_COPY(hat->hat_vlp_ptes, vlpptep);
924 		newcr3 = MAKECR3(cpu->cpu_hat_info->hci_vlp_pfn);
925 #elif defined(__i386)
926 		reload_pae32(hat, cpu);
927 		newcr3 = MAKECR3(kas.a_hat->hat_htable->ht_pfn) +
928 		    (cpu->cpu_id + 1) * VLP_SIZE;
929 #endif
930 	} else {
931 		newcr3 = MAKECR3(hat->hat_htable->ht_pfn);
932 	}
933 	setcr3(newcr3);
934 	ASSERT(cpu == CPU);
935 }
936 
937 /*
938  * Utility to return a valid x86pte_t from protections, pfn, and level number
939  */
940 static x86pte_t
941 hati_mkpte(pfn_t pfn, uint_t attr, level_t level, uint_t flags)
942 {
943 	x86pte_t	pte;
944 	uint_t		cache_attr = attr & HAT_ORDER_MASK;
945 
946 	pte = MAKEPTE(pfn, level);
947 
948 	if (attr & PROT_WRITE)
949 		PTE_SET(pte, PT_WRITABLE);
950 
951 	if (attr & PROT_USER)
952 		PTE_SET(pte, PT_USER);
953 
954 	if (!(attr & PROT_EXEC))
955 		PTE_SET(pte, mmu.pt_nx);
956 
957 	/*
958 	 * Set the software bits used track ref/mod sync's and hments.
959 	 * If not using REF/MOD, set them to avoid h/w rewriting PTEs.
960 	 */
961 	if (flags & HAT_LOAD_NOCONSIST)
962 		PTE_SET(pte, PT_NOCONSIST | PT_REF | PT_MOD);
963 	else if (attr & HAT_NOSYNC)
964 		PTE_SET(pte, PT_NOSYNC | PT_REF | PT_MOD);
965 
966 	/*
967 	 * Set the caching attributes in the PTE. The combination
968 	 * of attributes are poorly defined, so we pay attention
969 	 * to them in the given order.
970 	 *
971 	 * The test for HAT_STRICTORDER is different because it's defined
972 	 * as "0" - which was a stupid thing to do, but is too late to change!
973 	 */
974 	if (cache_attr == HAT_STRICTORDER) {
975 		PTE_SET(pte, PT_NOCACHE);
976 	/*LINTED [Lint hates empty ifs, but it's the obvious way to do this] */
977 	} else if (cache_attr & (HAT_UNORDERED_OK | HAT_STORECACHING_OK)) {
978 		/* nothing to set */;
979 	} else if (cache_attr & (HAT_MERGING_OK | HAT_LOADCACHING_OK)) {
980 		PTE_SET(pte, PT_NOCACHE);
981 		if (x86_feature & X86_PAT)
982 			PTE_SET(pte, (level == 0) ? PT_PAT_4K : PT_PAT_LARGE);
983 		else
984 			PTE_SET(pte, PT_WRITETHRU);
985 	} else {
986 		panic("hati_mkpte(): bad caching attributes: %x\n", cache_attr);
987 	}
988 
989 	return (pte);
990 }
991 
992 /*
993  * Duplicate address translations of the parent to the child.
994  * This function really isn't used anymore.
995  */
996 /*ARGSUSED*/
997 int
998 hat_dup(hat_t *old, hat_t *new, caddr_t addr, size_t len, uint_t flag)
999 {
1000 	ASSERT((uintptr_t)addr < kernelbase);
1001 	ASSERT(new != kas.a_hat);
1002 	ASSERT(old != kas.a_hat);
1003 	return (0);
1004 }
1005 
1006 /*
1007  * Allocate any hat resources required for a process being swapped in.
1008  */
1009 /*ARGSUSED*/
1010 void
1011 hat_swapin(hat_t *hat)
1012 {
1013 	/* do nothing - we let everything fault back in */
1014 }
1015 
1016 /*
1017  * Unload all translations associated with an address space of a process
1018  * that is being swapped out.
1019  */
1020 void
1021 hat_swapout(hat_t *hat)
1022 {
1023 	uintptr_t	vaddr = (uintptr_t)0;
1024 	uintptr_t	eaddr = _userlimit;
1025 	htable_t	*ht = NULL;
1026 	level_t		l;
1027 
1028 	/*
1029 	 * We can't just call hat_unload(hat, 0, _userlimit...)  here, because
1030 	 * seg_spt and shared pagetables can't be swapped out.
1031 	 * Take a look at segspt_shmswapout() - it's a big no-op.
1032 	 *
1033 	 * Instead we'll walk through all the address space and unload
1034 	 * any mappings which we are sure are not shared, not locked.
1035 	 */
1036 	ASSERT(IS_PAGEALIGNED(vaddr));
1037 	ASSERT(IS_PAGEALIGNED(eaddr));
1038 	ASSERT(AS_LOCK_HELD(hat->hat_as, &hat->hat_as->a_lock));
1039 	if ((uintptr_t)hat->hat_as->a_userlimit < eaddr)
1040 		eaddr = (uintptr_t)hat->hat_as->a_userlimit;
1041 
1042 	while (vaddr < eaddr) {
1043 		(void) htable_walk(hat, &ht, &vaddr, eaddr);
1044 		if (ht == NULL)
1045 			break;
1046 
1047 		ASSERT(!IN_VA_HOLE(vaddr));
1048 
1049 		/*
1050 		 * If the page table is shared skip its entire range.
1051 		 * This code knows that only level 0 page tables are shared
1052 		 */
1053 		l = ht->ht_level;
1054 		if (ht->ht_flags & HTABLE_SHARED_PFN) {
1055 			ASSERT(l == 0);
1056 			vaddr = ht->ht_vaddr + LEVEL_SIZE(1);
1057 			htable_release(ht);
1058 			ht = NULL;
1059 			continue;
1060 		}
1061 
1062 		/*
1063 		 * If the page table has no locked entries, unload this one.
1064 		 */
1065 		if (ht->ht_lock_cnt == 0)
1066 			hat_unload(hat, (caddr_t)vaddr, LEVEL_SIZE(l),
1067 			    HAT_UNLOAD_UNMAP);
1068 
1069 		/*
1070 		 * If we have a level 0 page table with locked entries,
1071 		 * skip the entire page table, otherwise skip just one entry.
1072 		 */
1073 		if (ht->ht_lock_cnt > 0 && l == 0)
1074 			vaddr = ht->ht_vaddr + LEVEL_SIZE(1);
1075 		else
1076 			vaddr += LEVEL_SIZE(l);
1077 	}
1078 	if (ht)
1079 		htable_release(ht);
1080 
1081 	/*
1082 	 * We're in swapout because the system is low on memory, so
1083 	 * go back and flush all the htables off the cached list.
1084 	 */
1085 	htable_purge_hat(hat);
1086 }
1087 
1088 /*
1089  * returns number of bytes that have valid mappings in hat.
1090  */
1091 size_t
1092 hat_get_mapped_size(hat_t *hat)
1093 {
1094 	size_t total = 0;
1095 	int l;
1096 
1097 	for (l = 0; l <= mmu.max_page_level; l++)
1098 		total += (hat->hat_pages_mapped[l] << LEVEL_SHIFT(l));
1099 
1100 	return (total);
1101 }
1102 
1103 /*
1104  * enable/disable collection of stats for hat.
1105  */
1106 int
1107 hat_stats_enable(hat_t *hat)
1108 {
1109 	atomic_add_32(&hat->hat_stats, 1);
1110 	return (1);
1111 }
1112 
1113 void
1114 hat_stats_disable(hat_t *hat)
1115 {
1116 	atomic_add_32(&hat->hat_stats, -1);
1117 }
1118 
1119 /*
1120  * Utility to sync the ref/mod bits from a page table entry to the page_t
1121  * We must be holding the mapping list lock when this is called.
1122  */
1123 static void
1124 hati_sync_pte_to_page(page_t *pp, x86pte_t pte, level_t level)
1125 {
1126 	uint_t	rm = 0;
1127 	pgcnt_t	pgcnt;
1128 
1129 	if (PTE_GET(pte, PT_SOFTWARE) >= PT_NOSYNC)
1130 		return;
1131 
1132 	if (PTE_GET(pte, PT_REF))
1133 		rm |= P_REF;
1134 
1135 	if (PTE_GET(pte, PT_MOD))
1136 		rm |= P_MOD;
1137 
1138 	if (rm == 0)
1139 		return;
1140 
1141 	/*
1142 	 * sync to all constituent pages of a large page
1143 	 */
1144 	ASSERT(x86_hm_held(pp));
1145 	pgcnt = page_get_pagecnt(level);
1146 	ASSERT(IS_P2ALIGNED(pp->p_pagenum, pgcnt));
1147 	for (; pgcnt > 0; --pgcnt) {
1148 		/*
1149 		 * hat_page_demote() can't decrease
1150 		 * pszc below this mapping size
1151 		 * since this large mapping existed after we
1152 		 * took mlist lock.
1153 		 */
1154 		ASSERT(pp->p_szc >= level);
1155 		hat_page_setattr(pp, rm);
1156 		++pp;
1157 	}
1158 }
1159 
1160 /*
1161  * This the set of PTE bits for PFN, permissions and caching
1162  * that require a TLB flush (hat_tlb_inval) if changed on a HAT_LOAD_REMAP
1163  */
1164 #define	PT_REMAP_BITS							\
1165 	(PT_PADDR | PT_NX | PT_WRITABLE | PT_WRITETHRU |		\
1166 	PT_NOCACHE | PT_PAT_4K | PT_PAT_LARGE)
1167 
1168 #define	REMAPASSERT(EX)	if (!(EX)) panic("hati_pte_map: " #EX)
1169 /*
1170  * Do the low-level work to get a mapping entered into a HAT's pagetables
1171  * and in the mapping list of the associated page_t.
1172  */
1173 static int
1174 hati_pte_map(
1175 	htable_t	*ht,
1176 	uint_t		entry,
1177 	page_t		*pp,
1178 	x86pte_t	pte,
1179 	int		flags,
1180 	void		*pte_ptr)
1181 {
1182 	hat_t		*hat = ht->ht_hat;
1183 	x86pte_t	old_pte;
1184 	level_t		l = ht->ht_level;
1185 	hment_t		*hm;
1186 	uint_t		is_consist;
1187 	int		rv = 0;
1188 
1189 	/*
1190 	 * Is this a consistant (ie. need mapping list lock) mapping?
1191 	 */
1192 	is_consist = (pp != NULL && (flags & HAT_LOAD_NOCONSIST) == 0);
1193 
1194 	/*
1195 	 * Track locked mapping count in the htable.  Do this first,
1196 	 * as we track locking even if there already is a mapping present.
1197 	 */
1198 	if ((flags & HAT_LOAD_LOCK) != 0 && hat != kas.a_hat)
1199 		HTABLE_LOCK_INC(ht);
1200 
1201 	/*
1202 	 * Acquire the page's mapping list lock and get an hment to use.
1203 	 * Note that hment_prepare() might return NULL.
1204 	 */
1205 	if (is_consist) {
1206 		x86_hm_enter(pp);
1207 		hm = hment_prepare(ht, entry, pp);
1208 	}
1209 
1210 	/*
1211 	 * Set the new pte, retrieving the old one at the same time.
1212 	 */
1213 	old_pte = x86pte_set(ht, entry, pte, pte_ptr);
1214 
1215 	/*
1216 	 * did we get a large page / page table collision?
1217 	 */
1218 	if (old_pte == LPAGE_ERROR) {
1219 		rv = -1;
1220 		goto done;
1221 	}
1222 
1223 	/*
1224 	 * If the mapping didn't change there is nothing more to do.
1225 	 */
1226 	if (PTE_EQUIV(pte, old_pte))
1227 		goto done;
1228 
1229 	/*
1230 	 * Install a new mapping in the page's mapping list
1231 	 */
1232 	if (!PTE_ISVALID(old_pte)) {
1233 		if (is_consist) {
1234 			hment_assign(ht, entry, pp, hm);
1235 			x86_hm_exit(pp);
1236 		} else {
1237 			ASSERT(flags & HAT_LOAD_NOCONSIST);
1238 		}
1239 		HTABLE_INC(ht->ht_valid_cnt);
1240 		PGCNT_INC(hat, l);
1241 		return (rv);
1242 	}
1243 
1244 	/*
1245 	 * Remap's are more complicated:
1246 	 *  - HAT_LOAD_REMAP must be specified if changing the pfn.
1247 	 *    We also require that NOCONSIST be specified.
1248 	 *  - Otherwise only permission or caching bits may change.
1249 	 */
1250 	if (!PTE_ISPAGE(old_pte, l))
1251 		panic("non-null/page mapping pte=" FMT_PTE, old_pte);
1252 
1253 	if (PTE2PFN(old_pte, l) != PTE2PFN(pte, l)) {
1254 		REMAPASSERT(flags & HAT_LOAD_REMAP);
1255 		REMAPASSERT(flags & HAT_LOAD_NOCONSIST);
1256 		REMAPASSERT(PTE_GET(old_pte, PT_SOFTWARE) >= PT_NOCONSIST);
1257 		REMAPASSERT(pf_is_memory(PTE2PFN(old_pte, l)) ==
1258 		    pf_is_memory(PTE2PFN(pte, l)));
1259 		REMAPASSERT(!is_consist);
1260 	}
1261 
1262 	/*
1263 	 * We only let remaps change the bits for PFNs, permissions
1264 	 * or caching type.
1265 	 */
1266 	ASSERT(PTE_GET(old_pte, ~(PT_REMAP_BITS | PT_REF | PT_MOD)) ==
1267 	    PTE_GET(pte, ~PT_REMAP_BITS));
1268 
1269 	/*
1270 	 * We don't create any mapping list entries on a remap, so release
1271 	 * any allocated hment after we drop the mapping list lock.
1272 	 */
1273 done:
1274 	if (is_consist) {
1275 		x86_hm_exit(pp);
1276 		if (hm != NULL)
1277 			hment_free(hm);
1278 	}
1279 	return (rv);
1280 }
1281 
1282 /*
1283  * Internal routine to load a single page table entry. This only fails if
1284  * we attempt to overwrite a page table link with a large page.
1285  */
1286 static int
1287 hati_load_common(
1288 	hat_t		*hat,
1289 	uintptr_t	va,
1290 	page_t		*pp,
1291 	uint_t		attr,
1292 	uint_t		flags,
1293 	level_t		level,
1294 	pfn_t		pfn)
1295 {
1296 	htable_t	*ht;
1297 	uint_t		entry;
1298 	x86pte_t	pte;
1299 	int		rv = 0;
1300 
1301 	/*
1302 	 * The number 16 is arbitrary and here to catch a recursion problem
1303 	 * early before we blow out the kernel stack.
1304 	 */
1305 	++curthread->t_hatdepth;
1306 	ASSERT(curthread->t_hatdepth < 16);
1307 
1308 	ASSERT(hat == kas.a_hat ||
1309 	    AS_LOCK_HELD(hat->hat_as, &hat->hat_as->a_lock));
1310 
1311 	if (flags & HAT_LOAD_SHARE)
1312 		hat->hat_flags |= HAT_SHARED;
1313 
1314 	/*
1315 	 * Find the page table that maps this page if it already exists.
1316 	 */
1317 	ht = htable_lookup(hat, va, level);
1318 
1319 	/*
1320 	 * We must have HAT_LOAD_NOCONSIST if page_t is NULL.
1321 	 */
1322 	if (pp == NULL)
1323 		flags |= HAT_LOAD_NOCONSIST;
1324 
1325 	if (ht == NULL) {
1326 		ht = htable_create(hat, va, level, NULL);
1327 		ASSERT(ht != NULL);
1328 	}
1329 	entry = htable_va2entry(va, ht);
1330 
1331 	/*
1332 	 * a bunch of paranoid error checking
1333 	 */
1334 	ASSERT(ht->ht_busy > 0);
1335 	if (ht->ht_vaddr > va || va > HTABLE_LAST_PAGE(ht))
1336 		panic("hati_load_common: bad htable %p, va %p", ht, (void *)va);
1337 	ASSERT(ht->ht_level == level);
1338 
1339 	/*
1340 	 * construct the new PTE
1341 	 */
1342 	if (hat == kas.a_hat)
1343 		attr &= ~PROT_USER;
1344 	pte = hati_mkpte(pfn, attr, level, flags);
1345 	if (hat == kas.a_hat && va >= kernelbase)
1346 		PTE_SET(pte, mmu.pt_global);
1347 
1348 	/*
1349 	 * establish the mapping
1350 	 */
1351 	rv = hati_pte_map(ht, entry, pp, pte, flags, NULL);
1352 
1353 	/*
1354 	 * release the htable and any reserves
1355 	 */
1356 	htable_release(ht);
1357 	--curthread->t_hatdepth;
1358 	return (rv);
1359 }
1360 
1361 /*
1362  * special case of hat_memload to deal with some kernel addrs for performance
1363  */
1364 static void
1365 hat_kmap_load(
1366 	caddr_t		addr,
1367 	page_t		*pp,
1368 	uint_t		attr,
1369 	uint_t		flags)
1370 {
1371 	uintptr_t	va = (uintptr_t)addr;
1372 	x86pte_t	pte;
1373 	pfn_t		pfn = page_pptonum(pp);
1374 	pgcnt_t		pg_off = mmu_btop(va - mmu.kmap_addr);
1375 	htable_t	*ht;
1376 	uint_t		entry;
1377 	void		*pte_ptr;
1378 
1379 	/*
1380 	 * construct the requested PTE
1381 	 */
1382 	attr &= ~PROT_USER;
1383 	attr |= HAT_STORECACHING_OK;
1384 	pte = hati_mkpte(pfn, attr, 0, flags);
1385 	PTE_SET(pte, mmu.pt_global);
1386 
1387 	/*
1388 	 * Figure out the pte_ptr and htable and use common code to finish up
1389 	 */
1390 	if (mmu.pae_hat)
1391 		pte_ptr = mmu.kmap_ptes + pg_off;
1392 	else
1393 		pte_ptr = (x86pte32_t *)mmu.kmap_ptes + pg_off;
1394 	ht = mmu.kmap_htables[(va - mmu.kmap_htables[0]->ht_vaddr) >>
1395 	    LEVEL_SHIFT(1)];
1396 	entry = htable_va2entry(va, ht);
1397 	++curthread->t_hatdepth;
1398 	ASSERT(curthread->t_hatdepth < 16);
1399 	(void) hati_pte_map(ht, entry, pp, pte, flags, pte_ptr);
1400 	--curthread->t_hatdepth;
1401 }
1402 
1403 /*
1404  * hat_memload() - load a translation to the given page struct
1405  *
1406  * Flags for hat_memload/hat_devload/hat_*attr.
1407  *
1408  * 	HAT_LOAD	Default flags to load a translation to the page.
1409  *
1410  * 	HAT_LOAD_LOCK	Lock down mapping resources; hat_map(), hat_memload(),
1411  *			and hat_devload().
1412  *
1413  *	HAT_LOAD_NOCONSIST Do not add mapping to page_t mapping list.
1414  *			sets PT_NOCONSIST
1415  *
1416  *	HAT_LOAD_SHARE	A flag to hat_memload() to indicate h/w page tables
1417  *			that map some user pages (not kas) is shared by more
1418  *			than one process (eg. ISM).
1419  *
1420  *	HAT_LOAD_REMAP	Reload a valid pte with a different page frame.
1421  *
1422  *	HAT_NO_KALLOC	Do not kmem_alloc while creating the mapping; at this
1423  *			point, it's setting up mapping to allocate internal
1424  *			hat layer data structures.  This flag forces hat layer
1425  *			to tap its reserves in order to prevent infinite
1426  *			recursion.
1427  *
1428  * The following is a protection attribute (like PROT_READ, etc.)
1429  *
1430  *	HAT_NOSYNC	set PT_NOSYNC - this mapping's ref/mod bits
1431  *			are never cleared.
1432  *
1433  * Installing new valid PTE's and creation of the mapping list
1434  * entry are controlled under the same lock. It's derived from the
1435  * page_t being mapped.
1436  */
1437 static uint_t supported_memload_flags =
1438 	HAT_LOAD | HAT_LOAD_LOCK | HAT_LOAD_ADV | HAT_LOAD_NOCONSIST |
1439 	HAT_LOAD_SHARE | HAT_NO_KALLOC | HAT_LOAD_REMAP | HAT_LOAD_TEXT;
1440 
1441 void
1442 hat_memload(
1443 	hat_t		*hat,
1444 	caddr_t		addr,
1445 	page_t		*pp,
1446 	uint_t		attr,
1447 	uint_t		flags)
1448 {
1449 	uintptr_t	va = (uintptr_t)addr;
1450 	level_t		level = 0;
1451 	pfn_t		pfn = page_pptonum(pp);
1452 
1453 	HATIN(hat_memload, hat, addr, (size_t)MMU_PAGESIZE);
1454 	ASSERT(IS_PAGEALIGNED(va));
1455 	ASSERT(hat == kas.a_hat || va < _userlimit);
1456 	ASSERT(hat == kas.a_hat ||
1457 	    AS_LOCK_HELD(hat->hat_as, &hat->hat_as->a_lock));
1458 	ASSERT((flags & supported_memload_flags) == flags);
1459 
1460 	ASSERT(!IN_VA_HOLE(va));
1461 	ASSERT(!PP_ISFREE(pp));
1462 
1463 	/*
1464 	 * kernel address special case for performance.
1465 	 */
1466 	if (mmu.kmap_addr <= va && va < mmu.kmap_eaddr) {
1467 		ASSERT(hat == kas.a_hat);
1468 		hat_kmap_load(addr, pp, attr, flags);
1469 		return;
1470 	}
1471 
1472 	/*
1473 	 * This is used for memory with normal caching enabled, so
1474 	 * always set HAT_STORECACHING_OK.
1475 	 */
1476 	attr |= HAT_STORECACHING_OK;
1477 	if (hati_load_common(hat, va, pp, attr, flags, level, pfn) != 0)
1478 		panic("unexpected hati_load_common() failure");
1479 	HATOUT(hat_memload, hat, addr);
1480 }
1481 
1482 /*
1483  * Load the given array of page structs using large pages when possible
1484  */
1485 void
1486 hat_memload_array(
1487 	hat_t		*hat,
1488 	caddr_t		addr,
1489 	size_t		len,
1490 	page_t		**pages,
1491 	uint_t		attr,
1492 	uint_t		flags)
1493 {
1494 	uintptr_t	va = (uintptr_t)addr;
1495 	uintptr_t	eaddr = va + len;
1496 	level_t		level;
1497 	size_t		pgsize;
1498 	pgcnt_t		pgindx = 0;
1499 	pfn_t		pfn;
1500 	pgcnt_t		i;
1501 
1502 	HATIN(hat_memload_array, hat, addr, len);
1503 	ASSERT(IS_PAGEALIGNED(va));
1504 	ASSERT(hat == kas.a_hat || va + len <= _userlimit);
1505 	ASSERT(hat == kas.a_hat ||
1506 	    AS_LOCK_HELD(hat->hat_as, &hat->hat_as->a_lock));
1507 	ASSERT((flags & supported_memload_flags) == flags);
1508 
1509 	/*
1510 	 * memload is used for memory with full caching enabled, so
1511 	 * set HAT_STORECACHING_OK.
1512 	 */
1513 	attr |= HAT_STORECACHING_OK;
1514 
1515 	/*
1516 	 * handle all pages using largest possible pagesize
1517 	 */
1518 	while (va < eaddr) {
1519 		/*
1520 		 * decide what level mapping to use (ie. pagesize)
1521 		 */
1522 		pfn = page_pptonum(pages[pgindx]);
1523 		for (level = mmu.max_page_level; ; --level) {
1524 			pgsize = LEVEL_SIZE(level);
1525 			if (level == 0)
1526 				break;
1527 
1528 			if (!IS_P2ALIGNED(va, pgsize) ||
1529 			    (eaddr - va) < pgsize ||
1530 			    !IS_P2ALIGNED(pfn_to_pa(pfn), pgsize))
1531 				continue;
1532 
1533 			/*
1534 			 * To use a large mapping of this size, all the
1535 			 * pages we are passed must be sequential subpages
1536 			 * of the large page.
1537 			 * hat_page_demote() can't change p_szc because
1538 			 * all pages are locked.
1539 			 */
1540 			if (pages[pgindx]->p_szc >= level) {
1541 				for (i = 0; i < mmu_btop(pgsize); ++i) {
1542 					if (pfn + i !=
1543 					    page_pptonum(pages[pgindx + i]))
1544 						break;
1545 					ASSERT(pages[pgindx + i]->p_szc >=
1546 					    level);
1547 					ASSERT(pages[pgindx] + i ==
1548 					    pages[pgindx + i]);
1549 				}
1550 				if (i == mmu_btop(pgsize))
1551 					break;
1552 			}
1553 		}
1554 
1555 		/*
1556 		 * Load this page mapping. If the load fails, try a smaller
1557 		 * pagesize.
1558 		 */
1559 		ASSERT(!IN_VA_HOLE(va));
1560 		while (hati_load_common(hat, va, pages[pgindx], attr,
1561 			    flags, level, pfn) != 0) {
1562 			if (level == 0)
1563 				panic("unexpected hati_load_common() failure");
1564 			--level;
1565 			pgsize = LEVEL_SIZE(level);
1566 		}
1567 
1568 		/*
1569 		 * move to next page
1570 		 */
1571 		va += pgsize;
1572 		pgindx += mmu_btop(pgsize);
1573 	}
1574 	HATOUT(hat_memload_array, hat, addr);
1575 }
1576 
1577 /*
1578  * void hat_devload(hat, addr, len, pf, attr, flags)
1579  *	load/lock the given page frame number
1580  *
1581  * Advisory ordering attributes. Apply only to device mappings.
1582  *
1583  * HAT_STRICTORDER: the CPU must issue the references in order, as the
1584  *	programmer specified.  This is the default.
1585  * HAT_UNORDERED_OK: the CPU may reorder the references (this is all kinds
1586  *	of reordering; store or load with store or load).
1587  * HAT_MERGING_OK: merging and batching: the CPU may merge individual stores
1588  *	to consecutive locations (for example, turn two consecutive byte
1589  *	stores into one halfword store), and it may batch individual loads
1590  *	(for example, turn two consecutive byte loads into one halfword load).
1591  *	This also implies re-ordering.
1592  * HAT_LOADCACHING_OK: the CPU may cache the data it fetches and reuse it
1593  *	until another store occurs.  The default is to fetch new data
1594  *	on every load.  This also implies merging.
1595  * HAT_STORECACHING_OK: the CPU may keep the data in the cache and push it to
1596  *	the device (perhaps with other data) at a later time.  The default is
1597  *	to push the data right away.  This also implies load caching.
1598  *
1599  * Equivalent of hat_memload(), but can be used for device memory where
1600  * there are no page_t's and we support additional flags (write merging, etc).
1601  * Note that we can have large page mappings with this interface.
1602  */
1603 int supported_devload_flags = HAT_LOAD | HAT_LOAD_LOCK |
1604 	HAT_LOAD_NOCONSIST | HAT_STRICTORDER | HAT_UNORDERED_OK |
1605 	HAT_MERGING_OK | HAT_LOADCACHING_OK | HAT_STORECACHING_OK;
1606 
1607 void
1608 hat_devload(
1609 	hat_t		*hat,
1610 	caddr_t		addr,
1611 	size_t		len,
1612 	pfn_t		pfn,
1613 	uint_t		attr,
1614 	int		flags)
1615 {
1616 	uintptr_t	va = ALIGN2PAGE(addr);
1617 	uintptr_t	eva = va + len;
1618 	level_t		level;
1619 	size_t		pgsize;
1620 	page_t		*pp;
1621 	int		f;	/* per PTE copy of flags  - maybe modified */
1622 	uint_t		a;	/* per PTE copy of attr */
1623 
1624 	HATIN(hat_devload, hat, addr, len);
1625 	ASSERT(IS_PAGEALIGNED(va));
1626 	ASSERT(hat == kas.a_hat || eva <= _userlimit);
1627 	ASSERT(hat == kas.a_hat ||
1628 	    AS_LOCK_HELD(hat->hat_as, &hat->hat_as->a_lock));
1629 	ASSERT((flags & supported_devload_flags) == flags);
1630 
1631 	/*
1632 	 * handle all pages
1633 	 */
1634 	while (va < eva) {
1635 
1636 		/*
1637 		 * decide what level mapping to use (ie. pagesize)
1638 		 */
1639 		for (level = mmu.max_page_level; ; --level) {
1640 			pgsize = LEVEL_SIZE(level);
1641 			if (level == 0)
1642 				break;
1643 			if (IS_P2ALIGNED(va, pgsize) &&
1644 			    (eva - va) >= pgsize &&
1645 			    IS_P2ALIGNED(pfn, mmu_btop(pgsize)))
1646 				break;
1647 		}
1648 
1649 		/*
1650 		 * If this is just memory then allow caching (this happens
1651 		 * for the nucleus pages) - though HAT_PLAT_NOCACHE can be used
1652 		 * to override that. If we don't have a page_t then make sure
1653 		 * NOCONSIST is set.
1654 		 */
1655 		a = attr;
1656 		f = flags;
1657 		if (pf_is_memory(pfn)) {
1658 			if (!(a & HAT_PLAT_NOCACHE))
1659 				a |= HAT_STORECACHING_OK;
1660 
1661 			if (f & HAT_LOAD_NOCONSIST)
1662 				pp = NULL;
1663 			else
1664 				pp = page_numtopp_nolock(pfn);
1665 		} else {
1666 			pp = NULL;
1667 			f |= HAT_LOAD_NOCONSIST;
1668 		}
1669 
1670 		/*
1671 		 * load this page mapping
1672 		 */
1673 		ASSERT(!IN_VA_HOLE(va));
1674 		while (hati_load_common(hat, va, pp, a, f, level, pfn) != 0) {
1675 			if (level == 0)
1676 				panic("unexpected hati_load_common() failure");
1677 			--level;
1678 			pgsize = LEVEL_SIZE(level);
1679 		}
1680 
1681 		/*
1682 		 * move to next page
1683 		 */
1684 		va += pgsize;
1685 		pfn += mmu_btop(pgsize);
1686 	}
1687 	HATOUT(hat_devload, hat, addr);
1688 }
1689 
1690 /*
1691  * void hat_unlock(hat, addr, len)
1692  *	unlock the mappings to a given range of addresses
1693  *
1694  * Locks are tracked by ht_lock_cnt in the htable.
1695  */
1696 void
1697 hat_unlock(hat_t *hat, caddr_t addr, size_t len)
1698 {
1699 	uintptr_t	vaddr = (uintptr_t)addr;
1700 	uintptr_t	eaddr = vaddr + len;
1701 	htable_t	*ht = NULL;
1702 
1703 	/*
1704 	 * kernel entries are always locked, we don't track lock counts
1705 	 */
1706 	ASSERT(hat == kas.a_hat || eaddr <= _userlimit);
1707 	ASSERT(IS_PAGEALIGNED(vaddr));
1708 	ASSERT(IS_PAGEALIGNED(eaddr));
1709 	if (hat == kas.a_hat)
1710 		return;
1711 	if (eaddr > _userlimit)
1712 		panic("hat_unlock() address out of range - above _userlimit");
1713 
1714 	ASSERT(AS_LOCK_HELD(hat->hat_as, &hat->hat_as->a_lock));
1715 	while (vaddr < eaddr) {
1716 		(void) htable_walk(hat, &ht, &vaddr, eaddr);
1717 		if (ht == NULL)
1718 			break;
1719 
1720 		ASSERT(!IN_VA_HOLE(vaddr));
1721 
1722 		if (ht->ht_lock_cnt < 1)
1723 			panic("hat_unlock(): lock_cnt < 1, "
1724 			    "htable=%p, vaddr=%p\n", ht, (caddr_t)vaddr);
1725 		HTABLE_LOCK_DEC(ht);
1726 
1727 		vaddr += LEVEL_SIZE(ht->ht_level);
1728 	}
1729 	if (ht)
1730 		htable_release(ht);
1731 }
1732 
1733 /*
1734  * Cross call service routine to demap a virtual page on
1735  * the current CPU or flush all mappings in TLB.
1736  */
1737 /*ARGSUSED*/
1738 static int
1739 hati_demap_func(xc_arg_t a1, xc_arg_t a2, xc_arg_t a3)
1740 {
1741 	hat_t	*hat = (hat_t *)a1;
1742 	caddr_t	addr = (caddr_t)a2;
1743 
1744 	/*
1745 	 * If the target hat isn't the kernel and this CPU isn't operating
1746 	 * in the target hat, we can ignore the cross call.
1747 	 */
1748 	if (hat != kas.a_hat && hat != CPU->cpu_current_hat)
1749 		return (0);
1750 
1751 	/*
1752 	 * For a normal address, we just flush one page mapping
1753 	 */
1754 	if ((uintptr_t)addr != DEMAP_ALL_ADDR) {
1755 		mmu_tlbflush_entry(addr);
1756 		return (0);
1757 	}
1758 
1759 	/*
1760 	 * Otherwise we reload cr3 to effect a complete TLB flush.
1761 	 *
1762 	 * A reload of cr3 on a VLP process also means we must also recopy in
1763 	 * the pte values from the struct hat
1764 	 */
1765 	if (hat->hat_flags & HAT_VLP) {
1766 #if defined(__amd64)
1767 		x86pte_t *vlpptep = CPU->cpu_hat_info->hci_vlp_l2ptes;
1768 
1769 		VLP_COPY(hat->hat_vlp_ptes, vlpptep);
1770 #elif defined(__i386)
1771 		reload_pae32(hat, CPU);
1772 #endif
1773 	}
1774 	reload_cr3();
1775 	return (0);
1776 }
1777 
1778 /*
1779  * Internal routine to do cross calls to invalidate a range of pages on
1780  * all CPUs using a given hat.
1781  */
1782 void
1783 hat_tlb_inval(hat_t *hat, uintptr_t va)
1784 {
1785 	extern int	flushes_require_xcalls;	/* from mp_startup.c */
1786 	cpuset_t	justme;
1787 	cpuset_t	cpus_to_shootdown;
1788 
1789 	/*
1790 	 * If the hat is being destroyed, there are no more users, so
1791 	 * demap need not do anything.
1792 	 */
1793 	if (hat->hat_flags & HAT_FREEING)
1794 		return;
1795 
1796 	/*
1797 	 * If demapping from a shared pagetable, we best demap the
1798 	 * entire set of user TLBs, since we don't know what addresses
1799 	 * these were shared at.
1800 	 */
1801 	if (hat->hat_flags & HAT_SHARED) {
1802 		hat = kas.a_hat;
1803 		va = DEMAP_ALL_ADDR;
1804 	}
1805 
1806 	/*
1807 	 * if not running with multiple CPUs, don't use cross calls
1808 	 */
1809 	if (panicstr || !flushes_require_xcalls) {
1810 		(void) hati_demap_func((xc_arg_t)hat, (xc_arg_t)va, NULL);
1811 		return;
1812 	}
1813 
1814 
1815 	/*
1816 	 * Determine CPUs to shootdown. Kernel changes always do all CPUs.
1817 	 * Otherwise it's just CPUs currently executing in this hat.
1818 	 */
1819 	kpreempt_disable();
1820 	CPUSET_ONLY(justme, CPU->cpu_id);
1821 	if (hat == kas.a_hat)
1822 		cpus_to_shootdown = khat_cpuset;
1823 	else
1824 		cpus_to_shootdown = hat->hat_cpus;
1825 
1826 	if (CPUSET_ISNULL(cpus_to_shootdown) ||
1827 	    CPUSET_ISEQUAL(cpus_to_shootdown, justme)) {
1828 
1829 		(void) hati_demap_func((xc_arg_t)hat, (xc_arg_t)va, NULL);
1830 
1831 	} else {
1832 
1833 		CPUSET_ADD(cpus_to_shootdown, CPU->cpu_id);
1834 		xc_call((xc_arg_t)hat, (xc_arg_t)va, NULL, X_CALL_HIPRI,
1835 		    cpus_to_shootdown, hati_demap_func);
1836 
1837 	}
1838 	kpreempt_enable();
1839 }
1840 
1841 /*
1842  * Interior routine for HAT_UNLOADs from hat_unload_callback(),
1843  * hat_kmap_unload() OR from hat_steal() code.  This routine doesn't
1844  * handle releasing of the htables.
1845  */
1846 void
1847 hat_pte_unmap(
1848 	htable_t	*ht,
1849 	uint_t		entry,
1850 	uint_t		flags,
1851 	x86pte_t	old_pte,
1852 	void		*pte_ptr)
1853 {
1854 	hat_t		*hat = ht->ht_hat;
1855 	hment_t		*hm = NULL;
1856 	page_t		*pp = NULL;
1857 	level_t		l = ht->ht_level;
1858 	pfn_t		pfn;
1859 
1860 	/*
1861 	 * We always track the locking counts, even if nothing is unmapped
1862 	 */
1863 	if ((flags & HAT_UNLOAD_UNLOCK) != 0 && hat != kas.a_hat) {
1864 		ASSERT(ht->ht_lock_cnt > 0);
1865 		HTABLE_LOCK_DEC(ht);
1866 	}
1867 
1868 	/*
1869 	 * Figure out which page's mapping list lock to acquire using the PFN
1870 	 * passed in "old" PTE. We then attempt to invalidate the PTE.
1871 	 * If another thread, probably a hat_pageunload, has asynchronously
1872 	 * unmapped/remapped this address we'll loop here.
1873 	 */
1874 	ASSERT(ht->ht_busy > 0);
1875 	while (PTE_ISVALID(old_pte)) {
1876 		pfn = PTE2PFN(old_pte, l);
1877 		if (PTE_GET(old_pte, PT_SOFTWARE) >= PT_NOCONSIST) {
1878 			pp = NULL;
1879 		} else {
1880 			pp = page_numtopp_nolock(pfn);
1881 			if (pp == NULL) {
1882 				panic("no page_t, not NOCONSIST: old_pte="
1883 				    FMT_PTE " ht=%lx entry=0x%x pte_ptr=%lx",
1884 				    old_pte, (uintptr_t)ht, entry,
1885 				    (uintptr_t)pte_ptr);
1886 			}
1887 			x86_hm_enter(pp);
1888 		}
1889 
1890 		/*
1891 		 * If freeing the address space, check that the PTE
1892 		 * hasn't changed, as the mappings are no longer in use by
1893 		 * any thread, invalidation is unnecessary.
1894 		 * If not freeing, do a full invalidate.
1895 		 */
1896 		if (hat->hat_flags & HAT_FREEING)
1897 			old_pte = x86pte_get(ht, entry);
1898 		else
1899 			old_pte = x86pte_inval(ht, entry, old_pte, pte_ptr);
1900 
1901 		/*
1902 		 * If the page hadn't changed we've unmapped it and can proceed
1903 		 */
1904 		if (PTE_ISVALID(old_pte) && PTE2PFN(old_pte, l) == pfn)
1905 			break;
1906 
1907 		/*
1908 		 * Otherwise, we'll have to retry with the current old_pte.
1909 		 * Drop the hment lock, since the pfn may have changed.
1910 		 */
1911 		if (pp != NULL) {
1912 			x86_hm_exit(pp);
1913 			pp = NULL;
1914 		} else {
1915 			ASSERT(PTE_GET(old_pte, PT_SOFTWARE) >= PT_NOCONSIST);
1916 		}
1917 	}
1918 
1919 	/*
1920 	 * If the old mapping wasn't valid, there's nothing more to do
1921 	 */
1922 	if (!PTE_ISVALID(old_pte)) {
1923 		if (pp != NULL)
1924 			x86_hm_exit(pp);
1925 		return;
1926 	}
1927 
1928 	/*
1929 	 * Take care of syncing any MOD/REF bits and removing the hment.
1930 	 */
1931 	if (pp != NULL) {
1932 		if (!(flags & HAT_UNLOAD_NOSYNC))
1933 			hati_sync_pte_to_page(pp, old_pte, l);
1934 		hm = hment_remove(pp, ht, entry);
1935 		x86_hm_exit(pp);
1936 		if (hm != NULL)
1937 			hment_free(hm);
1938 	}
1939 
1940 	/*
1941 	 * Handle book keeping in the htable and hat
1942 	 */
1943 	ASSERT(ht->ht_valid_cnt > 0);
1944 	HTABLE_DEC(ht->ht_valid_cnt);
1945 	PGCNT_DEC(hat, l);
1946 }
1947 
1948 /*
1949  * very cheap unload implementation to special case some kernel addresses
1950  */
1951 static void
1952 hat_kmap_unload(caddr_t addr, size_t len, uint_t flags)
1953 {
1954 	uintptr_t	va = (uintptr_t)addr;
1955 	uintptr_t	eva = va + len;
1956 	pgcnt_t		pg_index;
1957 	htable_t	*ht;
1958 	uint_t		entry;
1959 	x86pte_t	*pte_ptr;
1960 	x86pte_t	old_pte;
1961 
1962 	for (; va < eva; va += MMU_PAGESIZE) {
1963 		/*
1964 		 * Get the PTE
1965 		 */
1966 		pg_index = mmu_btop(va - mmu.kmap_addr);
1967 		pte_ptr = PT_INDEX_PTR(mmu.kmap_ptes, pg_index);
1968 		old_pte = GET_PTE(pte_ptr);
1969 
1970 		/*
1971 		 * get the htable / entry
1972 		 */
1973 		ht = mmu.kmap_htables[(va - mmu.kmap_htables[0]->ht_vaddr)
1974 		    >> LEVEL_SHIFT(1)];
1975 		entry = htable_va2entry(va, ht);
1976 
1977 		/*
1978 		 * use mostly common code to unmap it.
1979 		 */
1980 		hat_pte_unmap(ht, entry, flags, old_pte, pte_ptr);
1981 	}
1982 }
1983 
1984 
1985 /*
1986  * unload a range of virtual address space (no callback)
1987  */
1988 void
1989 hat_unload(hat_t *hat, caddr_t addr, size_t len, uint_t flags)
1990 {
1991 	uintptr_t va = (uintptr_t)addr;
1992 
1993 	ASSERT(hat == kas.a_hat || va + len <= _userlimit);
1994 
1995 	/*
1996 	 * special case for performance.
1997 	 */
1998 	if (mmu.kmap_addr <= va && va < mmu.kmap_eaddr) {
1999 		ASSERT(hat == kas.a_hat);
2000 		hat_kmap_unload(addr, len, flags);
2001 	} else {
2002 		hat_unload_callback(hat, addr, len, flags, NULL);
2003 	}
2004 }
2005 
2006 /*
2007  * Do the callbacks for ranges being unloaded.
2008  */
2009 typedef struct range_info {
2010 	uintptr_t	rng_va;
2011 	ulong_t		rng_cnt;
2012 	level_t		rng_level;
2013 } range_info_t;
2014 
2015 static void
2016 handle_ranges(hat_callback_t *cb, uint_t cnt, range_info_t *range)
2017 {
2018 	/*
2019 	 * do callbacks to upper level VM system
2020 	 */
2021 	while (cb != NULL && cnt > 0) {
2022 		--cnt;
2023 		cb->hcb_start_addr = (caddr_t)range[cnt].rng_va;
2024 		cb->hcb_end_addr = cb->hcb_start_addr;
2025 		cb->hcb_end_addr +=
2026 		    range[cnt].rng_cnt << LEVEL_SIZE(range[cnt].rng_level);
2027 		cb->hcb_function(cb);
2028 	}
2029 }
2030 
2031 /*
2032  * Unload a given range of addresses (has optional callback)
2033  *
2034  * Flags:
2035  * define	HAT_UNLOAD		0x00
2036  * define	HAT_UNLOAD_NOSYNC	0x02
2037  * define	HAT_UNLOAD_UNLOCK	0x04
2038  * define	HAT_UNLOAD_OTHER	0x08 - not used
2039  * define	HAT_UNLOAD_UNMAP	0x10 - same as HAT_UNLOAD
2040  */
2041 #define	MAX_UNLOAD_CNT (8)
2042 void
2043 hat_unload_callback(
2044 	hat_t		*hat,
2045 	caddr_t		addr,
2046 	size_t		len,
2047 	uint_t		flags,
2048 	hat_callback_t	*cb)
2049 {
2050 	uintptr_t	vaddr = (uintptr_t)addr;
2051 	uintptr_t	eaddr = vaddr + len;
2052 	htable_t	*ht = NULL;
2053 	uint_t		entry;
2054 	uintptr_t	contig_va = (uintptr_t)-1L;
2055 	range_info_t	r[MAX_UNLOAD_CNT];
2056 	uint_t		r_cnt = 0;
2057 	x86pte_t	old_pte;
2058 
2059 	HATIN(hat_unload_callback, hat, addr, len);
2060 	ASSERT(hat == kas.a_hat || eaddr <= _userlimit);
2061 	ASSERT(IS_PAGEALIGNED(vaddr));
2062 	ASSERT(IS_PAGEALIGNED(eaddr));
2063 
2064 	/*
2065 	 * Special case a single page being unloaded for speed. This happens
2066 	 * quite frequently, COW faults after a fork() for example.
2067 	 */
2068 	if (cb == NULL && len == MMU_PAGESIZE) {
2069 		ht = htable_getpte(hat, vaddr, &entry, &old_pte, 0);
2070 		if (ht != NULL) {
2071 			if (PTE_ISVALID(old_pte))
2072 				hat_pte_unmap(ht, entry, flags, old_pte, NULL);
2073 			htable_release(ht);
2074 		}
2075 		return;
2076 	}
2077 
2078 	while (vaddr < eaddr) {
2079 		old_pte = htable_walk(hat, &ht, &vaddr, eaddr);
2080 		if (ht == NULL)
2081 			break;
2082 
2083 		ASSERT(!IN_VA_HOLE(vaddr));
2084 
2085 		if (vaddr < (uintptr_t)addr)
2086 			panic("hat_unload_callback(): unmap inside large page");
2087 
2088 		/*
2089 		 * We'll do the call backs for contiguous ranges
2090 		 */
2091 		if (vaddr != contig_va ||
2092 		    (r_cnt > 0 && r[r_cnt - 1].rng_level != ht->ht_level)) {
2093 			if (r_cnt == MAX_UNLOAD_CNT) {
2094 				handle_ranges(cb, r_cnt, r);
2095 				r_cnt = 0;
2096 			}
2097 			r[r_cnt].rng_va = vaddr;
2098 			r[r_cnt].rng_cnt = 0;
2099 			r[r_cnt].rng_level = ht->ht_level;
2100 			++r_cnt;
2101 		}
2102 
2103 		/*
2104 		 * Unload one mapping from the page tables.
2105 		 */
2106 		entry = htable_va2entry(vaddr, ht);
2107 		hat_pte_unmap(ht, entry, flags, old_pte, NULL);
2108 		ASSERT(ht->ht_level <= mmu.max_page_level);
2109 		vaddr += LEVEL_SIZE(ht->ht_level);
2110 		contig_va = vaddr;
2111 		++r[r_cnt - 1].rng_cnt;
2112 	}
2113 	if (ht)
2114 		htable_release(ht);
2115 
2116 	/*
2117 	 * handle last range for callbacks
2118 	 */
2119 	if (r_cnt > 0)
2120 		handle_ranges(cb, r_cnt, r);
2121 
2122 	HATOUT(hat_unload_callback, hat, addr);
2123 }
2124 
2125 /*
2126  * synchronize mapping with software data structures
2127  *
2128  * This interface is currently only used by the working set monitor
2129  * driver.
2130  */
2131 /*ARGSUSED*/
2132 void
2133 hat_sync(hat_t *hat, caddr_t addr, size_t len, uint_t flags)
2134 {
2135 	uintptr_t	vaddr = (uintptr_t)addr;
2136 	uintptr_t	eaddr = vaddr + len;
2137 	htable_t	*ht = NULL;
2138 	uint_t		entry;
2139 	x86pte_t	pte;
2140 	x86pte_t	save_pte;
2141 	x86pte_t	new;
2142 	page_t		*pp;
2143 
2144 	ASSERT(!IN_VA_HOLE(vaddr));
2145 	ASSERT(IS_PAGEALIGNED(vaddr));
2146 	ASSERT(IS_PAGEALIGNED(eaddr));
2147 	ASSERT(hat == kas.a_hat || eaddr <= _userlimit);
2148 
2149 	for (; vaddr < eaddr; vaddr += LEVEL_SIZE(ht->ht_level)) {
2150 try_again:
2151 		pte = htable_walk(hat, &ht, &vaddr, eaddr);
2152 		if (ht == NULL)
2153 			break;
2154 		entry = htable_va2entry(vaddr, ht);
2155 
2156 		if (PTE_GET(pte, PT_SOFTWARE) >= PT_NOSYNC ||
2157 		    PTE_GET(pte, PT_REF | PT_MOD) == 0)
2158 			continue;
2159 
2160 		/*
2161 		 * We need to acquire the mapping list lock to protect
2162 		 * against hat_pageunload(), hat_unload(), etc.
2163 		 */
2164 		pp = page_numtopp_nolock(PTE2PFN(pte, ht->ht_level));
2165 		if (pp == NULL)
2166 			break;
2167 		x86_hm_enter(pp);
2168 		save_pte = pte;
2169 		pte = x86pte_get(ht, entry);
2170 		if (pte != save_pte) {
2171 			x86_hm_exit(pp);
2172 			goto try_again;
2173 		}
2174 		if (PTE_GET(pte, PT_SOFTWARE) >= PT_NOSYNC ||
2175 		    PTE_GET(pte, PT_REF | PT_MOD) == 0) {
2176 			x86_hm_exit(pp);
2177 			continue;
2178 		}
2179 
2180 		/*
2181 		 * Need to clear ref or mod bits. We may compete with
2182 		 * hardware updating the R/M bits and have to try again.
2183 		 */
2184 		if (flags == HAT_SYNC_ZERORM) {
2185 			new = pte;
2186 			PTE_CLR(new, PT_REF | PT_MOD);
2187 			pte = hati_update_pte(ht, entry, pte, new);
2188 			if (pte != 0) {
2189 				x86_hm_exit(pp);
2190 				goto try_again;
2191 			}
2192 		} else {
2193 			/*
2194 			 * sync the PTE to the page_t
2195 			 */
2196 			hati_sync_pte_to_page(pp, save_pte, ht->ht_level);
2197 		}
2198 		x86_hm_exit(pp);
2199 	}
2200 	if (ht)
2201 		htable_release(ht);
2202 }
2203 
2204 /*
2205  * void	hat_map(hat, addr, len, flags)
2206  */
2207 /*ARGSUSED*/
2208 void
2209 hat_map(hat_t *hat, caddr_t addr, size_t len, uint_t flags)
2210 {
2211 	/* does nothing */
2212 }
2213 
2214 /*
2215  * uint_t hat_getattr(hat, addr, *attr)
2216  *	returns attr for <hat,addr> in *attr.  returns 0 if there was a
2217  *	mapping and *attr is valid, nonzero if there was no mapping and
2218  *	*attr is not valid.
2219  */
2220 uint_t
2221 hat_getattr(hat_t *hat, caddr_t addr, uint_t *attr)
2222 {
2223 	uintptr_t	vaddr = ALIGN2PAGE(addr);
2224 	htable_t	*ht = NULL;
2225 	x86pte_t	pte;
2226 
2227 	ASSERT(hat == kas.a_hat || vaddr <= _userlimit);
2228 
2229 	if (IN_VA_HOLE(vaddr))
2230 		return ((uint_t)-1);
2231 
2232 	ht = htable_getpte(hat, vaddr, NULL, &pte, mmu.max_page_level);
2233 	if (ht == NULL)
2234 		return ((uint_t)-1);
2235 
2236 	if (!PTE_ISVALID(pte) || !PTE_ISPAGE(pte, ht->ht_level)) {
2237 		htable_release(ht);
2238 		return ((uint_t)-1);
2239 	}
2240 
2241 	*attr = PROT_READ;
2242 	if (PTE_GET(pte, PT_WRITABLE))
2243 		*attr |= PROT_WRITE;
2244 	if (PTE_GET(pte, PT_USER))
2245 		*attr |= PROT_USER;
2246 	if (!PTE_GET(pte, mmu.pt_nx))
2247 		*attr |= PROT_EXEC;
2248 	if (PTE_GET(pte, PT_SOFTWARE) >= PT_NOSYNC)
2249 		*attr |= HAT_NOSYNC;
2250 	htable_release(ht);
2251 	return (0);
2252 }
2253 
2254 /*
2255  * hat_updateattr() applies the given attribute change to an existing mapping
2256  */
2257 #define	HAT_LOAD_ATTR		1
2258 #define	HAT_SET_ATTR		2
2259 #define	HAT_CLR_ATTR		3
2260 
2261 static void
2262 hat_updateattr(hat_t *hat, caddr_t addr, size_t len, uint_t attr, int what)
2263 {
2264 	uintptr_t	vaddr = (uintptr_t)addr;
2265 	uintptr_t	eaddr = (uintptr_t)addr + len;
2266 	htable_t	*ht = NULL;
2267 	uint_t		entry;
2268 	x86pte_t	oldpte, newpte;
2269 	page_t		*pp;
2270 
2271 	ASSERT(IS_PAGEALIGNED(vaddr));
2272 	ASSERT(IS_PAGEALIGNED(eaddr));
2273 	ASSERT(hat == kas.a_hat ||
2274 	    AS_LOCK_HELD(hat->hat_as, &hat->hat_as->a_lock));
2275 	for (; vaddr < eaddr; vaddr += LEVEL_SIZE(ht->ht_level)) {
2276 try_again:
2277 		oldpte = htable_walk(hat, &ht, &vaddr, eaddr);
2278 		if (ht == NULL)
2279 			break;
2280 		if (PTE_GET(oldpte, PT_SOFTWARE) >= PT_NOCONSIST)
2281 			continue;
2282 
2283 		pp = page_numtopp_nolock(PTE2PFN(oldpte, ht->ht_level));
2284 		if (pp == NULL)
2285 			continue;
2286 		x86_hm_enter(pp);
2287 
2288 		newpte = oldpte;
2289 		/*
2290 		 * We found a page table entry in the desired range,
2291 		 * figure out the new attributes.
2292 		 */
2293 		if (what == HAT_SET_ATTR || what == HAT_LOAD_ATTR) {
2294 			if ((attr & PROT_WRITE) &&
2295 			    !PTE_GET(oldpte, PT_WRITABLE))
2296 				newpte |= PT_WRITABLE;
2297 
2298 			if ((attr & HAT_NOSYNC) &&
2299 			    PTE_GET(oldpte, PT_SOFTWARE) < PT_NOSYNC)
2300 				newpte |= PT_NOSYNC;
2301 
2302 			if ((attr & PROT_EXEC) && PTE_GET(oldpte, mmu.pt_nx))
2303 				newpte &= ~mmu.pt_nx;
2304 		}
2305 
2306 		if (what == HAT_LOAD_ATTR) {
2307 			if (!(attr & PROT_WRITE) &&
2308 			    PTE_GET(oldpte, PT_WRITABLE))
2309 				newpte &= ~PT_WRITABLE;
2310 
2311 			if (!(attr & HAT_NOSYNC) &&
2312 			    PTE_GET(oldpte, PT_SOFTWARE) >= PT_NOSYNC)
2313 				newpte &= ~PT_SOFTWARE;
2314 
2315 			if (!(attr & PROT_EXEC) && !PTE_GET(oldpte, mmu.pt_nx))
2316 				newpte |= mmu.pt_nx;
2317 		}
2318 
2319 		if (what == HAT_CLR_ATTR) {
2320 			if ((attr & PROT_WRITE) && PTE_GET(oldpte, PT_WRITABLE))
2321 				newpte &= ~PT_WRITABLE;
2322 
2323 			if ((attr & HAT_NOSYNC) &&
2324 			    PTE_GET(oldpte, PT_SOFTWARE) >= PT_NOSYNC)
2325 				newpte &= ~PT_SOFTWARE;
2326 
2327 			if ((attr & PROT_EXEC) && !PTE_GET(oldpte, mmu.pt_nx))
2328 				newpte |= mmu.pt_nx;
2329 		}
2330 
2331 		/*
2332 		 * Ensure NOSYNC/NOCONSIST mappings have REF and MOD set.
2333 		 * x86pte_set() depends on this.
2334 		 */
2335 		if (PTE_GET(newpte, PT_SOFTWARE) >= PT_NOSYNC)
2336 			newpte |= PT_REF | PT_MOD;
2337 
2338 		/*
2339 		 * what about PROT_READ or others? this code only handles:
2340 		 * EXEC, WRITE, NOSYNC
2341 		 */
2342 
2343 		/*
2344 		 * If new PTE really changed, update the table.
2345 		 */
2346 		if (newpte != oldpte) {
2347 			entry = htable_va2entry(vaddr, ht);
2348 			oldpte = hati_update_pte(ht, entry, oldpte, newpte);
2349 			if (oldpte != 0) {
2350 				x86_hm_exit(pp);
2351 				goto try_again;
2352 			}
2353 		}
2354 		x86_hm_exit(pp);
2355 	}
2356 	if (ht)
2357 		htable_release(ht);
2358 }
2359 
2360 /*
2361  * Various wrappers for hat_updateattr()
2362  */
2363 void
2364 hat_setattr(hat_t *hat, caddr_t addr, size_t len, uint_t attr)
2365 {
2366 	ASSERT(hat == kas.a_hat || (uintptr_t)addr + len <= _userlimit);
2367 	hat_updateattr(hat, addr, len, attr, HAT_SET_ATTR);
2368 }
2369 
2370 void
2371 hat_clrattr(hat_t *hat, caddr_t addr, size_t len, uint_t attr)
2372 {
2373 	ASSERT(hat == kas.a_hat || (uintptr_t)addr + len <= _userlimit);
2374 	hat_updateattr(hat, addr, len, attr, HAT_CLR_ATTR);
2375 }
2376 
2377 void
2378 hat_chgattr(hat_t *hat, caddr_t addr, size_t len, uint_t attr)
2379 {
2380 	ASSERT(hat == kas.a_hat || (uintptr_t)addr + len <= _userlimit);
2381 	hat_updateattr(hat, addr, len, attr, HAT_LOAD_ATTR);
2382 }
2383 
2384 void
2385 hat_chgprot(hat_t *hat, caddr_t addr, size_t len, uint_t vprot)
2386 {
2387 	ASSERT(hat == kas.a_hat || (uintptr_t)addr + len <= _userlimit);
2388 	hat_updateattr(hat, addr, len, vprot & HAT_PROT_MASK, HAT_LOAD_ATTR);
2389 }
2390 
2391 /*
2392  * size_t hat_getpagesize(hat, addr)
2393  *	returns pagesize in bytes for <hat, addr>. returns -1 of there is
2394  *	no mapping. This is an advisory call.
2395  */
2396 ssize_t
2397 hat_getpagesize(hat_t *hat, caddr_t addr)
2398 {
2399 	uintptr_t	vaddr = ALIGN2PAGE(addr);
2400 	htable_t	*ht;
2401 	size_t		pagesize;
2402 
2403 	ASSERT(hat == kas.a_hat || vaddr <= _userlimit);
2404 	if (IN_VA_HOLE(vaddr))
2405 		return (-1);
2406 	ht = htable_getpage(hat, vaddr, NULL);
2407 	if (ht == NULL)
2408 		return (-1);
2409 	pagesize = LEVEL_SIZE(ht->ht_level);
2410 	htable_release(ht);
2411 	return (pagesize);
2412 }
2413 
2414 
2415 
2416 /*
2417  * pfn_t hat_getpfnum(hat, addr)
2418  *	returns pfn for <hat, addr> or PFN_INVALID if mapping is invalid.
2419  */
2420 pfn_t
2421 hat_getpfnum(hat_t *hat, caddr_t addr)
2422 {
2423 	uintptr_t	vaddr = ALIGN2PAGE(addr);
2424 	htable_t	*ht;
2425 	uint_t		entry;
2426 	pfn_t		pfn = PFN_INVALID;
2427 
2428 	ASSERT(hat == kas.a_hat || vaddr <= _userlimit);
2429 	if (khat_running == 0)
2430 		return (PFN_INVALID);
2431 
2432 	if (IN_VA_HOLE(vaddr))
2433 		return (PFN_INVALID);
2434 
2435 	/*
2436 	 * A very common use of hat_getpfnum() is from the DDI for kernel pages.
2437 	 * Use the kmap_ptes (which also covers the 32 bit heap) to speed
2438 	 * this up.
2439 	 */
2440 	if (mmu.kmap_addr <= vaddr && vaddr < mmu.kmap_eaddr) {
2441 		x86pte_t pte;
2442 		pgcnt_t pg_index;
2443 
2444 		pg_index = mmu_btop(vaddr - mmu.kmap_addr);
2445 		pte = GET_PTE(PT_INDEX_PTR(mmu.kmap_ptes, pg_index));
2446 		if (!PTE_ISVALID(pte))
2447 			return (PFN_INVALID);
2448 		/*LINTED [use of constant 0 causes a silly lint warning] */
2449 		return (PTE2PFN(pte, 0));
2450 	}
2451 
2452 	ht = htable_getpage(hat, vaddr, &entry);
2453 	if (ht == NULL)
2454 		return (PFN_INVALID);
2455 	ASSERT(vaddr >= ht->ht_vaddr);
2456 	ASSERT(vaddr <= HTABLE_LAST_PAGE(ht));
2457 	pfn = PTE2PFN(x86pte_get(ht, entry), ht->ht_level);
2458 	if (ht->ht_level > 0)
2459 		pfn += mmu_btop(vaddr & LEVEL_OFFSET(ht->ht_level));
2460 	htable_release(ht);
2461 	return (pfn);
2462 }
2463 
2464 /*
2465  * hat_getkpfnum() is an obsolete DDI routine, and its use is discouraged.
2466  * Use hat_getpfnum(kas.a_hat, ...) instead.
2467  *
2468  * We'd like to return PFN_INVALID if the mappings have underlying page_t's
2469  * but can't right now due to the fact that some software has grown to use
2470  * this interface incorrectly. So for now when the interface is misused,
2471  * return a warning to the user that in the future it won't work in the
2472  * way they're abusing it, and carry on.
2473  *
2474  * Note that hat_getkpfnum() is never supported on amd64.
2475  */
2476 #if !defined(__amd64)
2477 pfn_t
2478 hat_getkpfnum(caddr_t addr)
2479 {
2480 	pfn_t	pfn;
2481 	int badcaller = 0;
2482 
2483 	if (khat_running == 0)
2484 		panic("hat_getkpfnum(): called too early\n");
2485 	if ((uintptr_t)addr < kernelbase)
2486 		return (PFN_INVALID);
2487 
2488 
2489 	if (segkpm && IS_KPM_ADDR(addr)) {
2490 		badcaller = 1;
2491 		pfn = hat_kpm_va2pfn(addr);
2492 	} else {
2493 		pfn = hat_getpfnum(kas.a_hat, addr);
2494 		badcaller = pf_is_memory(pfn);
2495 	}
2496 
2497 	if (badcaller)
2498 		hat_getkpfnum_badcall(caller());
2499 	return (pfn);
2500 }
2501 #endif /* __amd64 */
2502 
2503 /*
2504  * int hat_probe(hat, addr)
2505  *	return 0 if no valid mapping is present.  Faster version
2506  *	of hat_getattr in certain architectures.
2507  */
2508 int
2509 hat_probe(hat_t *hat, caddr_t addr)
2510 {
2511 	uintptr_t	vaddr = ALIGN2PAGE(addr);
2512 	uint_t		entry;
2513 	htable_t	*ht;
2514 	pgcnt_t		pg_off;
2515 
2516 	ASSERT(hat == kas.a_hat || vaddr <= _userlimit);
2517 	ASSERT(hat == kas.a_hat ||
2518 	    AS_LOCK_HELD(hat->hat_as, &hat->hat_as->a_lock));
2519 	if (IN_VA_HOLE(vaddr))
2520 		return (0);
2521 
2522 	/*
2523 	 * Most common use of hat_probe is from segmap. We special case it
2524 	 * for performance.
2525 	 */
2526 	if (mmu.kmap_addr <= vaddr && vaddr < mmu.kmap_eaddr) {
2527 		pg_off = mmu_btop(vaddr - mmu.kmap_addr);
2528 		if (mmu.pae_hat)
2529 			return (PTE_ISVALID(mmu.kmap_ptes[pg_off]));
2530 		else
2531 			return (PTE_ISVALID(
2532 			    ((x86pte32_t *)mmu.kmap_ptes)[pg_off]));
2533 	}
2534 
2535 	ht = htable_getpage(hat, vaddr, &entry);
2536 	if (ht == NULL)
2537 		return (0);
2538 	htable_release(ht);
2539 	return (1);
2540 }
2541 
2542 /*
2543  * Simple implementation of ISM. hat_share() is just like hat_memload_array(),
2544  * except that we use the ism_hat's existing mappings to determine the pages
2545  * and protections to use for this hat. In case we find a properly aligned
2546  * and sized pagetable of 4K mappings, we will attempt to share the pagetable
2547  * itself.
2548  */
2549 /*ARGSUSED*/
2550 int
2551 hat_share(
2552 	hat_t		*hat,
2553 	caddr_t		addr,
2554 	hat_t		*ism_hat,
2555 	caddr_t		src_addr,
2556 	size_t		len,	/* almost useless value, see below.. */
2557 	uint_t		ismszc)
2558 {
2559 	uintptr_t	vaddr_start = (uintptr_t)addr;
2560 	uintptr_t	vaddr;
2561 	uintptr_t	pt_vaddr;
2562 	uintptr_t	eaddr = vaddr_start + len;
2563 	uintptr_t	ism_addr_start = (uintptr_t)src_addr;
2564 	uintptr_t	ism_addr = ism_addr_start;
2565 	uintptr_t	e_ism_addr = ism_addr + len;
2566 	htable_t	*ism_ht = NULL;
2567 	htable_t	*ht;
2568 	x86pte_t	pte;
2569 	page_t		*pp;
2570 	pfn_t		pfn;
2571 	level_t		l;
2572 	pgcnt_t		pgcnt;
2573 	uint_t		prot;
2574 	uint_t		valid_cnt;
2575 
2576 	/*
2577 	 * We might be asked to share an empty DISM hat by as_dup()
2578 	 */
2579 	ASSERT(hat != kas.a_hat);
2580 	ASSERT(eaddr <= _userlimit);
2581 	if (!(ism_hat->hat_flags & HAT_SHARED)) {
2582 		ASSERT(hat_get_mapped_size(ism_hat) == 0);
2583 		return (0);
2584 	}
2585 
2586 	/*
2587 	 * The SPT segment driver often passes us a size larger than there are
2588 	 * valid mappings. That's because it rounds the segment size up to a
2589 	 * large pagesize, even if the actual memory mapped by ism_hat is less.
2590 	 */
2591 	HATIN(hat_share, hat, addr, len);
2592 	ASSERT(IS_PAGEALIGNED(vaddr_start));
2593 	ASSERT(IS_PAGEALIGNED(ism_addr_start));
2594 	ASSERT(ism_hat->hat_flags & HAT_SHARED);
2595 	while (ism_addr < e_ism_addr) {
2596 		/*
2597 		 * use htable_walk to get the next valid ISM mapping
2598 		 */
2599 		pte = htable_walk(ism_hat, &ism_ht, &ism_addr, e_ism_addr);
2600 		if (ism_ht == NULL)
2601 			break;
2602 
2603 		/*
2604 		 * Find the largest page size we can use, based on the
2605 		 * ISM mapping size, our address alignment and the remaining
2606 		 * map length.
2607 		 */
2608 		vaddr = vaddr_start + (ism_addr - ism_addr_start);
2609 		for (l = ism_ht->ht_level; l > 0; --l) {
2610 			if (LEVEL_SIZE(l) <= eaddr - vaddr &&
2611 			    (vaddr & LEVEL_OFFSET(l)) == 0)
2612 				break;
2613 		}
2614 
2615 		/*
2616 		 * attempt to share the pagetable
2617 		 *
2618 		 * - only 4K pagetables are shared (ie. level == 0)
2619 		 * - the hat_share() length must cover the whole pagetable
2620 		 * - the shared address must align at level 1
2621 		 * - a shared PTE for this address already exists OR
2622 		 * - no page table for this address exists yet
2623 		 */
2624 		pt_vaddr =
2625 		    vaddr_start + (ism_ht->ht_vaddr - ism_addr_start);
2626 		if (ism_ht->ht_level == 0 &&
2627 		    ism_ht->ht_vaddr + LEVEL_SIZE(1) <= e_ism_addr &&
2628 		    (pt_vaddr & LEVEL_OFFSET(1)) == 0) {
2629 
2630 			ht = htable_lookup(hat, pt_vaddr, 0);
2631 			if (ht == NULL)
2632 				ht = htable_create(hat, pt_vaddr, 0, ism_ht);
2633 
2634 			if (ht->ht_level > 0 ||
2635 			    !(ht->ht_flags & HTABLE_SHARED_PFN)) {
2636 
2637 				htable_release(ht);
2638 
2639 			} else {
2640 
2641 				/*
2642 				 * share the page table
2643 				 */
2644 				ASSERT(ht->ht_level == 0);
2645 				ASSERT(ht->ht_shares == ism_ht);
2646 				valid_cnt = ism_ht->ht_valid_cnt;
2647 				atomic_add_long(&hat->hat_pages_mapped[0],
2648 				    valid_cnt - ht->ht_valid_cnt);
2649 				ht->ht_valid_cnt = valid_cnt;
2650 				htable_release(ht);
2651 				ism_addr = ism_ht->ht_vaddr + LEVEL_SIZE(1);
2652 				htable_release(ism_ht);
2653 				ism_ht = NULL;
2654 				continue;
2655 			}
2656 		}
2657 
2658 		/*
2659 		 * Unable to share the page table. Instead we will
2660 		 * create new mappings from the values in the ISM mappings.
2661 		 *
2662 		 * The ISM mapping might be larger than the share area,
2663 		 * be careful to trunctate it if needed.
2664 		 */
2665 		if (eaddr - vaddr >= LEVEL_SIZE(ism_ht->ht_level)) {
2666 			pgcnt = mmu_btop(LEVEL_SIZE(ism_ht->ht_level));
2667 		} else {
2668 			pgcnt = mmu_btop(eaddr - vaddr);
2669 			l = 0;
2670 		}
2671 
2672 		pfn = PTE2PFN(pte, ism_ht->ht_level);
2673 		ASSERT(pfn != PFN_INVALID);
2674 		while (pgcnt > 0) {
2675 			/*
2676 			 * Make a new pte for the PFN for this level.
2677 			 * Copy protections for the pte from the ISM pte.
2678 			 */
2679 			pp = page_numtopp_nolock(pfn);
2680 			ASSERT(pp != NULL);
2681 
2682 			prot = PROT_USER | PROT_READ | HAT_UNORDERED_OK;
2683 			if (PTE_GET(pte, PT_WRITABLE))
2684 				prot |= PROT_WRITE;
2685 			if (!PTE_GET(pte, PT_NX))
2686 				prot |= PROT_EXEC;
2687 
2688 			/*
2689 			 * XX64 -- can shm ever be written to swap?
2690 			 * if not we could use HAT_NOSYNC here.
2691 			 */
2692 			while (hati_load_common(hat, vaddr, pp, prot, HAT_LOAD,
2693 			    l, pfn) != 0) {
2694 				if (l == 0)
2695 					panic("hati_load_common() failure");
2696 				--l;
2697 			}
2698 
2699 			vaddr += LEVEL_SIZE(l);
2700 			ism_addr += LEVEL_SIZE(l);
2701 			pfn += mmu_btop(LEVEL_SIZE(l));
2702 			pgcnt -= mmu_btop(LEVEL_SIZE(l));
2703 		}
2704 	}
2705 	if (ism_ht != NULL)
2706 		htable_release(ism_ht);
2707 
2708 	HATOUT(hat_share, hat, addr);
2709 	return (0);
2710 }
2711 
2712 
2713 /*
2714  * hat_unshare() is similar to hat_unload_callback(), but
2715  * we have to look for empty shared pagetables. Note that
2716  * hat_unshare() is always invoked against an entire segment.
2717  */
2718 /*ARGSUSED*/
2719 void
2720 hat_unshare(hat_t *hat, caddr_t addr, size_t len, uint_t ismszc)
2721 {
2722 	uintptr_t	vaddr = (uintptr_t)addr;
2723 	uintptr_t	eaddr = vaddr + len;
2724 	htable_t	*ht = NULL;
2725 	uint_t		need_demaps = 0;
2726 
2727 	ASSERT(hat != kas.a_hat);
2728 	ASSERT(eaddr <= _userlimit);
2729 	HATIN(hat_unshare, hat, addr, len);
2730 	ASSERT(IS_PAGEALIGNED(vaddr));
2731 	ASSERT(IS_PAGEALIGNED(eaddr));
2732 
2733 	/*
2734 	 * First go through and remove any shared pagetables.
2735 	 *
2736 	 * Note that it's ok to delay the TLB shootdown till the entire range is
2737 	 * finished, because if hat_pageunload() were to unload a shared
2738 	 * pagetable page, its hat_tlb_inval() will do a global TLB invalidate.
2739 	 */
2740 	while (vaddr < eaddr) {
2741 		ASSERT(!IN_VA_HOLE(vaddr));
2742 		/*
2743 		 * find the pagetable that would map the current address
2744 		 */
2745 		ht = htable_lookup(hat, vaddr, 0);
2746 		if (ht != NULL) {
2747 			if (ht->ht_flags & HTABLE_SHARED_PFN) {
2748 				/*
2749 				 * clear mapped pages count, set valid_cnt to 0
2750 				 * and let htable_release() finish the job
2751 				 */
2752 				atomic_add_long(&hat->hat_pages_mapped[0],
2753 				    -ht->ht_valid_cnt);
2754 				ht->ht_valid_cnt = 0;
2755 				need_demaps = 1;
2756 			}
2757 			htable_release(ht);
2758 		}
2759 		vaddr = (vaddr & LEVEL_MASK(1)) + LEVEL_SIZE(1);
2760 	}
2761 
2762 	/*
2763 	 * flush the TLBs - since we're probably dealing with MANY mappings
2764 	 * we do just one CR3 reload.
2765 	 */
2766 	if (!(hat->hat_flags & HAT_FREEING) && need_demaps)
2767 		hat_tlb_inval(hat, DEMAP_ALL_ADDR);
2768 
2769 	/*
2770 	 * Now go back and clean up any unaligned mappings that
2771 	 * couldn't share pagetables.
2772 	 */
2773 	hat_unload(hat, addr, len, HAT_UNLOAD_UNMAP);
2774 
2775 	HATOUT(hat_unshare, hat, addr);
2776 }
2777 
2778 
2779 /*
2780  * hat_reserve() does nothing
2781  */
2782 /*ARGSUSED*/
2783 void
2784 hat_reserve(struct as *as, caddr_t addr, size_t len)
2785 {
2786 }
2787 
2788 
2789 /*
2790  * Called when all mappings to a page should have write permission removed.
2791  * Mostly stolem from hat_pagesync()
2792  */
2793 static void
2794 hati_page_clrwrt(struct page *pp)
2795 {
2796 	hment_t		*hm = NULL;
2797 	htable_t	*ht;
2798 	uint_t		entry;
2799 	x86pte_t	old;
2800 	x86pte_t	new;
2801 	uint_t		pszc = 0;
2802 
2803 next_size:
2804 	/*
2805 	 * walk thru the mapping list clearing write permission
2806 	 */
2807 	x86_hm_enter(pp);
2808 	while ((hm = hment_walk(pp, &ht, &entry, hm)) != NULL) {
2809 		if (ht->ht_level < pszc)
2810 			continue;
2811 		old = x86pte_get(ht, entry);
2812 
2813 		for (;;) {
2814 			/*
2815 			 * Is this mapping of interest?
2816 			 */
2817 			if (PTE2PFN(old, ht->ht_level) != pp->p_pagenum ||
2818 			    PTE_GET(old, PT_WRITABLE) == 0)
2819 				break;
2820 
2821 			/*
2822 			 * Clear ref/mod writable bits. This requires cross
2823 			 * calls to ensure any executing TLBs see cleared bits.
2824 			 */
2825 			new = old;
2826 			PTE_CLR(new, PT_REF | PT_MOD | PT_WRITABLE);
2827 			old = hati_update_pte(ht, entry, old, new);
2828 			if (old != 0)
2829 				continue;
2830 
2831 			break;
2832 		}
2833 	}
2834 	x86_hm_exit(pp);
2835 	while (pszc < pp->p_szc) {
2836 		page_t *tpp;
2837 		pszc++;
2838 		tpp = PP_GROUPLEADER(pp, pszc);
2839 		if (pp != tpp) {
2840 			pp = tpp;
2841 			goto next_size;
2842 		}
2843 	}
2844 }
2845 
2846 /*
2847  * void hat_page_setattr(pp, flag)
2848  * void hat_page_clrattr(pp, flag)
2849  *	used to set/clr ref/mod bits.
2850  */
2851 void
2852 hat_page_setattr(struct page *pp, uint_t flag)
2853 {
2854 	vnode_t		*vp = pp->p_vnode;
2855 	kmutex_t	*vphm = NULL;
2856 	page_t		**listp;
2857 
2858 	if (PP_GETRM(pp, flag) == flag)
2859 		return;
2860 
2861 	if ((flag & P_MOD) != 0 && vp != NULL && IS_VMODSORT(vp)) {
2862 		vphm = page_vnode_mutex(vp);
2863 		mutex_enter(vphm);
2864 	}
2865 
2866 	PP_SETRM(pp, flag);
2867 
2868 	if (vphm != NULL) {
2869 
2870 		/*
2871 		 * Some File Systems examine v_pages for NULL w/o
2872 		 * grabbing the vphm mutex. Must not let it become NULL when
2873 		 * pp is the only page on the list.
2874 		 */
2875 		if (pp->p_vpnext != pp) {
2876 			page_vpsub(&vp->v_pages, pp);
2877 			if (vp->v_pages != NULL)
2878 				listp = &vp->v_pages->p_vpprev->p_vpnext;
2879 			else
2880 				listp = &vp->v_pages;
2881 			page_vpadd(listp, pp);
2882 		}
2883 		mutex_exit(vphm);
2884 	}
2885 }
2886 
2887 void
2888 hat_page_clrattr(struct page *pp, uint_t flag)
2889 {
2890 	vnode_t		*vp = pp->p_vnode;
2891 	ASSERT(!(flag & ~(P_MOD | P_REF | P_RO)));
2892 
2893 	/*
2894 	 * Caller is expected to hold page's io lock for VMODSORT to work
2895 	 * correctly with pvn_vplist_dirty() and pvn_getdirty() when mod
2896 	 * bit is cleared.
2897 	 * We don't have assert to avoid tripping some existing third party
2898 	 * code. The dirty page is moved back to top of the v_page list
2899 	 * after IO is done in pvn_write_done().
2900 	 */
2901 	PP_CLRRM(pp, flag);
2902 
2903 	if ((flag & P_MOD) != 0 && vp != NULL && IS_VMODSORT(vp)) {
2904 
2905 		/*
2906 		 * VMODSORT works by removing write permissions and getting
2907 		 * a fault when a page is made dirty. At this point
2908 		 * we need to remove write permission from all mappings
2909 		 * to this page.
2910 		 */
2911 		hati_page_clrwrt(pp);
2912 	}
2913 }
2914 
2915 /*
2916  *	If flag is specified, returns 0 if attribute is disabled
2917  *	and non zero if enabled.  If flag specifes multiple attributs
2918  *	then returns 0 if ALL atriibutes are disabled.  This is an advisory
2919  *	call.
2920  */
2921 uint_t
2922 hat_page_getattr(struct page *pp, uint_t flag)
2923 {
2924 	return (PP_GETRM(pp, flag));
2925 }
2926 
2927 
2928 /*
2929  * common code used by hat_pageunload() and hment_steal()
2930  */
2931 hment_t *
2932 hati_page_unmap(page_t *pp, htable_t *ht, uint_t entry)
2933 {
2934 	x86pte_t old_pte;
2935 	pfn_t pfn = pp->p_pagenum;
2936 	hment_t *hm;
2937 
2938 	/*
2939 	 * We need to acquire a hold on the htable in order to
2940 	 * do the invalidate. We know the htable must exist, since
2941 	 * unmap's don't release the htable until after removing any
2942 	 * hment. Having x86_hm_enter() keeps that from proceeding.
2943 	 */
2944 	htable_acquire(ht);
2945 
2946 	/*
2947 	 * Invalidate the PTE and remove the hment.
2948 	 */
2949 	old_pte = x86pte_inval(ht, entry, 0, NULL);
2950 	if (PTE2PFN(old_pte, ht->ht_level) != pfn) {
2951 		panic("x86pte_inval() failure found PTE = " FMT_PTE
2952 		    " pfn being unmapped is %lx ht=0x%lx entry=0x%x",
2953 		    old_pte, pfn, (uintptr_t)ht, entry);
2954 	}
2955 
2956 	/*
2957 	 * Clean up all the htable information for this mapping
2958 	 */
2959 	ASSERT(ht->ht_valid_cnt > 0);
2960 	HTABLE_DEC(ht->ht_valid_cnt);
2961 	PGCNT_DEC(ht->ht_hat, ht->ht_level);
2962 
2963 	/*
2964 	 * sync ref/mod bits to the page_t
2965 	 */
2966 	if (PTE_GET(old_pte, PT_SOFTWARE) < PT_NOSYNC)
2967 		hati_sync_pte_to_page(pp, old_pte, ht->ht_level);
2968 
2969 	/*
2970 	 * Remove the mapping list entry for this page.
2971 	 */
2972 	hm = hment_remove(pp, ht, entry);
2973 
2974 	/*
2975 	 * drop the mapping list lock so that we might free the
2976 	 * hment and htable.
2977 	 */
2978 	x86_hm_exit(pp);
2979 	htable_release(ht);
2980 	return (hm);
2981 }
2982 
2983 extern int	vpm_enable;
2984 /*
2985  * Unload all translations to a page. If the page is a subpage of a large
2986  * page, the large page mappings are also removed.
2987  *
2988  * The forceflags are unused.
2989  */
2990 
2991 /*ARGSUSED*/
2992 static int
2993 hati_pageunload(struct page *pp, uint_t pg_szcd, uint_t forceflag)
2994 {
2995 	page_t		*cur_pp = pp;
2996 	hment_t		*hm;
2997 	hment_t		*prev;
2998 	htable_t	*ht;
2999 	uint_t		entry;
3000 	level_t		level;
3001 
3002 #if defined(__amd64)
3003 	/*
3004 	 * clear the vpm ref.
3005 	 */
3006 	if (vpm_enable) {
3007 		pp->p_vpmref = 0;
3008 	}
3009 #endif
3010 	/*
3011 	 * The loop with next_size handles pages with multiple pagesize mappings
3012 	 */
3013 next_size:
3014 	for (;;) {
3015 
3016 		/*
3017 		 * Get a mapping list entry
3018 		 */
3019 		x86_hm_enter(cur_pp);
3020 		for (prev = NULL; ; prev = hm) {
3021 			hm = hment_walk(cur_pp, &ht, &entry, prev);
3022 			if (hm == NULL) {
3023 				x86_hm_exit(cur_pp);
3024 
3025 				/*
3026 				 * If not part of a larger page, we're done.
3027 				 */
3028 				if (cur_pp->p_szc <= pg_szcd) {
3029 					return (0);
3030 				}
3031 
3032 				/*
3033 				 * Else check the next larger page size.
3034 				 * hat_page_demote() may decrease p_szc
3035 				 * but that's ok we'll just take an extra
3036 				 * trip discover there're no larger mappings
3037 				 * and return.
3038 				 */
3039 				++pg_szcd;
3040 				cur_pp = PP_GROUPLEADER(cur_pp, pg_szcd);
3041 				goto next_size;
3042 			}
3043 
3044 			/*
3045 			 * If this mapping size matches, remove it.
3046 			 */
3047 			level = ht->ht_level;
3048 			if (level == pg_szcd)
3049 				break;
3050 		}
3051 
3052 		/*
3053 		 * Remove the mapping list entry for this page.
3054 		 * Note this does the x86_hm_exit() for us.
3055 		 */
3056 		hm = hati_page_unmap(cur_pp, ht, entry);
3057 		if (hm != NULL)
3058 			hment_free(hm);
3059 	}
3060 }
3061 
3062 int
3063 hat_pageunload(struct page *pp, uint_t forceflag)
3064 {
3065 	ASSERT(PAGE_EXCL(pp));
3066 	return (hati_pageunload(pp, 0, forceflag));
3067 }
3068 
3069 /*
3070  * Unload all large mappings to pp and reduce by 1 p_szc field of every large
3071  * page level that included pp.
3072  *
3073  * pp must be locked EXCL. Even though no other constituent pages are locked
3074  * it's legal to unload large mappings to pp because all constituent pages of
3075  * large locked mappings have to be locked SHARED.  therefore if we have EXCL
3076  * lock on one of constituent pages none of the large mappings to pp are
3077  * locked.
3078  *
3079  * Change (always decrease) p_szc field starting from the last constituent
3080  * page and ending with root constituent page so that root's pszc always shows
3081  * the area where hat_page_demote() may be active.
3082  *
3083  * This mechanism is only used for file system pages where it's not always
3084  * possible to get EXCL locks on all constituent pages to demote the size code
3085  * (as is done for anonymous or kernel large pages).
3086  */
3087 void
3088 hat_page_demote(page_t *pp)
3089 {
3090 	uint_t		pszc;
3091 	uint_t		rszc;
3092 	uint_t		szc;
3093 	page_t		*rootpp;
3094 	page_t		*firstpp;
3095 	page_t		*lastpp;
3096 	pgcnt_t		pgcnt;
3097 
3098 	ASSERT(PAGE_EXCL(pp));
3099 	ASSERT(!PP_ISFREE(pp));
3100 	ASSERT(page_szc_lock_assert(pp));
3101 
3102 	if (pp->p_szc == 0)
3103 		return;
3104 
3105 	rootpp = PP_GROUPLEADER(pp, 1);
3106 	(void) hati_pageunload(rootpp, 1, HAT_FORCE_PGUNLOAD);
3107 
3108 	/*
3109 	 * all large mappings to pp are gone
3110 	 * and no new can be setup since pp is locked exclusively.
3111 	 *
3112 	 * Lock the root to make sure there's only one hat_page_demote()
3113 	 * outstanding within the area of this root's pszc.
3114 	 *
3115 	 * Second potential hat_page_demote() is already eliminated by upper
3116 	 * VM layer via page_szc_lock() but we don't rely on it and use our
3117 	 * own locking (so that upper layer locking can be changed without
3118 	 * assumptions that hat depends on upper layer VM to prevent multiple
3119 	 * hat_page_demote() to be issued simultaneously to the same large
3120 	 * page).
3121 	 */
3122 again:
3123 	pszc = pp->p_szc;
3124 	if (pszc == 0)
3125 		return;
3126 	rootpp = PP_GROUPLEADER(pp, pszc);
3127 	x86_hm_enter(rootpp);
3128 	/*
3129 	 * If root's p_szc is different from pszc we raced with another
3130 	 * hat_page_demote().  Drop the lock and try to find the root again.
3131 	 * If root's p_szc is greater than pszc previous hat_page_demote() is
3132 	 * not done yet.  Take and release mlist lock of root's root to wait
3133 	 * for previous hat_page_demote() to complete.
3134 	 */
3135 	if ((rszc = rootpp->p_szc) != pszc) {
3136 		x86_hm_exit(rootpp);
3137 		if (rszc > pszc) {
3138 			/* p_szc of a locked non free page can't increase */
3139 			ASSERT(pp != rootpp);
3140 
3141 			rootpp = PP_GROUPLEADER(rootpp, rszc);
3142 			x86_hm_enter(rootpp);
3143 			x86_hm_exit(rootpp);
3144 		}
3145 		goto again;
3146 	}
3147 	ASSERT(pp->p_szc == pszc);
3148 
3149 	/*
3150 	 * Decrement by 1 p_szc of every constituent page of a region that
3151 	 * covered pp. For example if original szc is 3 it gets changed to 2
3152 	 * everywhere except in region 2 that covered pp. Region 2 that
3153 	 * covered pp gets demoted to 1 everywhere except in region 1 that
3154 	 * covered pp. The region 1 that covered pp is demoted to region
3155 	 * 0. It's done this way because from region 3 we removed level 3
3156 	 * mappings, from region 2 that covered pp we removed level 2 mappings
3157 	 * and from region 1 that covered pp we removed level 1 mappings.  All
3158 	 * changes are done from from high pfn's to low pfn's so that roots
3159 	 * are changed last allowing one to know the largest region where
3160 	 * hat_page_demote() is stil active by only looking at the root page.
3161 	 *
3162 	 * This algorithm is implemented in 2 while loops. First loop changes
3163 	 * p_szc of pages to the right of pp's level 1 region and second
3164 	 * loop changes p_szc of pages of level 1 region that covers pp
3165 	 * and all pages to the left of level 1 region that covers pp.
3166 	 * In the first loop p_szc keeps dropping with every iteration
3167 	 * and in the second loop it keeps increasing with every iteration.
3168 	 *
3169 	 * First loop description: Demote pages to the right of pp outside of
3170 	 * level 1 region that covers pp.  In every iteration of the while
3171 	 * loop below find the last page of szc region and the first page of
3172 	 * (szc - 1) region that is immediately to the right of (szc - 1)
3173 	 * region that covers pp.  From last such page to first such page
3174 	 * change every page's szc to szc - 1. Decrement szc and continue
3175 	 * looping until szc is 1. If pp belongs to the last (szc - 1) region
3176 	 * of szc region skip to the next iteration.
3177 	 */
3178 	szc = pszc;
3179 	while (szc > 1) {
3180 		lastpp = PP_GROUPLEADER(pp, szc);
3181 		pgcnt = page_get_pagecnt(szc);
3182 		lastpp += pgcnt - 1;
3183 		firstpp = PP_GROUPLEADER(pp, (szc - 1));
3184 		pgcnt = page_get_pagecnt(szc - 1);
3185 		if (lastpp - firstpp < pgcnt) {
3186 			szc--;
3187 			continue;
3188 		}
3189 		firstpp += pgcnt;
3190 		while (lastpp != firstpp) {
3191 			ASSERT(lastpp->p_szc == pszc);
3192 			lastpp->p_szc = szc - 1;
3193 			lastpp--;
3194 		}
3195 		firstpp->p_szc = szc - 1;
3196 		szc--;
3197 	}
3198 
3199 	/*
3200 	 * Second loop description:
3201 	 * First iteration changes p_szc to 0 of every
3202 	 * page of level 1 region that covers pp.
3203 	 * Subsequent iterations find last page of szc region
3204 	 * immediately to the left of szc region that covered pp
3205 	 * and first page of (szc + 1) region that covers pp.
3206 	 * From last to first page change p_szc of every page to szc.
3207 	 * Increment szc and continue looping until szc is pszc.
3208 	 * If pp belongs to the fist szc region of (szc + 1) region
3209 	 * skip to the next iteration.
3210 	 *
3211 	 */
3212 	szc = 0;
3213 	while (szc < pszc) {
3214 		firstpp = PP_GROUPLEADER(pp, (szc + 1));
3215 		if (szc == 0) {
3216 			pgcnt = page_get_pagecnt(1);
3217 			lastpp = firstpp + (pgcnt - 1);
3218 		} else {
3219 			lastpp = PP_GROUPLEADER(pp, szc);
3220 			if (firstpp == lastpp) {
3221 				szc++;
3222 				continue;
3223 			}
3224 			lastpp--;
3225 			pgcnt = page_get_pagecnt(szc);
3226 		}
3227 		while (lastpp != firstpp) {
3228 			ASSERT(lastpp->p_szc == pszc);
3229 			lastpp->p_szc = szc;
3230 			lastpp--;
3231 		}
3232 		firstpp->p_szc = szc;
3233 		if (firstpp == rootpp)
3234 			break;
3235 		szc++;
3236 	}
3237 	x86_hm_exit(rootpp);
3238 }
3239 
3240 /*
3241  * get hw stats from hardware into page struct and reset hw stats
3242  * returns attributes of page
3243  * Flags for hat_pagesync, hat_getstat, hat_sync
3244  *
3245  * define	HAT_SYNC_ZERORM		0x01
3246  *
3247  * Additional flags for hat_pagesync
3248  *
3249  * define	HAT_SYNC_STOPON_REF	0x02
3250  * define	HAT_SYNC_STOPON_MOD	0x04
3251  * define	HAT_SYNC_STOPON_RM	0x06
3252  * define	HAT_SYNC_STOPON_SHARED	0x08
3253  */
3254 uint_t
3255 hat_pagesync(struct page *pp, uint_t flags)
3256 {
3257 	hment_t		*hm = NULL;
3258 	htable_t	*ht;
3259 	uint_t		entry;
3260 	x86pte_t	old, save_old;
3261 	x86pte_t	new;
3262 	uchar_t		nrmbits = P_REF|P_MOD|P_RO;
3263 	extern ulong_t	po_share;
3264 	page_t		*save_pp = pp;
3265 	uint_t		pszc = 0;
3266 
3267 	ASSERT(PAGE_LOCKED(pp) || panicstr);
3268 
3269 	if (PP_ISRO(pp) && (flags & HAT_SYNC_STOPON_MOD))
3270 		return (pp->p_nrm & nrmbits);
3271 
3272 	if ((flags & HAT_SYNC_ZERORM) == 0) {
3273 
3274 		if ((flags & HAT_SYNC_STOPON_REF) != 0 && PP_ISREF(pp))
3275 			return (pp->p_nrm & nrmbits);
3276 
3277 		if ((flags & HAT_SYNC_STOPON_MOD) != 0 && PP_ISMOD(pp))
3278 			return (pp->p_nrm & nrmbits);
3279 
3280 		if ((flags & HAT_SYNC_STOPON_SHARED) != 0 &&
3281 		    hat_page_getshare(pp) > po_share) {
3282 			if (PP_ISRO(pp))
3283 				PP_SETREF(pp);
3284 			return (pp->p_nrm & nrmbits);
3285 		}
3286 	}
3287 
3288 next_size:
3289 	/*
3290 	 * walk thru the mapping list syncing (and clearing) ref/mod bits.
3291 	 */
3292 	x86_hm_enter(pp);
3293 	while ((hm = hment_walk(pp, &ht, &entry, hm)) != NULL) {
3294 		if (ht->ht_level < pszc)
3295 			continue;
3296 		old = x86pte_get(ht, entry);
3297 try_again:
3298 
3299 		ASSERT(PTE2PFN(old, ht->ht_level) == pp->p_pagenum);
3300 
3301 		if (PTE_GET(old, PT_REF | PT_MOD) == 0)
3302 			continue;
3303 
3304 		save_old = old;
3305 		if ((flags & HAT_SYNC_ZERORM) != 0) {
3306 
3307 			/*
3308 			 * Need to clear ref or mod bits. Need to demap
3309 			 * to make sure any executing TLBs see cleared bits.
3310 			 */
3311 			new = old;
3312 			PTE_CLR(new, PT_REF | PT_MOD);
3313 			old = hati_update_pte(ht, entry, old, new);
3314 			if (old != 0)
3315 				goto try_again;
3316 
3317 			old = save_old;
3318 		}
3319 
3320 		/*
3321 		 * Sync the PTE
3322 		 */
3323 		if (!(flags & HAT_SYNC_ZERORM) &&
3324 		    PTE_GET(old, PT_SOFTWARE) <= PT_NOSYNC)
3325 			hati_sync_pte_to_page(pp, old, ht->ht_level);
3326 
3327 		/*
3328 		 * can stop short if we found a ref'd or mod'd page
3329 		 */
3330 		if ((flags & HAT_SYNC_STOPON_MOD) && PP_ISMOD(save_pp) ||
3331 		    (flags & HAT_SYNC_STOPON_REF) && PP_ISREF(save_pp)) {
3332 			x86_hm_exit(pp);
3333 			goto done;
3334 		}
3335 	}
3336 	x86_hm_exit(pp);
3337 	while (pszc < pp->p_szc) {
3338 		page_t *tpp;
3339 		pszc++;
3340 		tpp = PP_GROUPLEADER(pp, pszc);
3341 		if (pp != tpp) {
3342 			pp = tpp;
3343 			goto next_size;
3344 		}
3345 	}
3346 done:
3347 	return (save_pp->p_nrm & nrmbits);
3348 }
3349 
3350 /*
3351  * returns approx number of mappings to this pp.  A return of 0 implies
3352  * there are no mappings to the page.
3353  */
3354 ulong_t
3355 hat_page_getshare(page_t *pp)
3356 {
3357 	uint_t cnt;
3358 	cnt = hment_mapcnt(pp);
3359 #if defined(__amd64)
3360 	if (vpm_enable && pp->p_vpmref) {
3361 		cnt += 1;
3362 	}
3363 #endif
3364 	return (cnt);
3365 }
3366 
3367 /*
3368  * hat_softlock isn't supported anymore
3369  */
3370 /*ARGSUSED*/
3371 faultcode_t
3372 hat_softlock(
3373 	hat_t *hat,
3374 	caddr_t addr,
3375 	size_t *len,
3376 	struct page **page_array,
3377 	uint_t flags)
3378 {
3379 	return (FC_NOSUPPORT);
3380 }
3381 
3382 
3383 
3384 /*
3385  * Routine to expose supported HAT features to platform independent code.
3386  */
3387 /*ARGSUSED*/
3388 int
3389 hat_supported(enum hat_features feature, void *arg)
3390 {
3391 	switch (feature) {
3392 
3393 	case HAT_SHARED_PT:	/* this is really ISM */
3394 		return (1);
3395 
3396 	case HAT_DYNAMIC_ISM_UNMAP:
3397 		return (0);
3398 
3399 	case HAT_VMODSORT:
3400 		return (1);
3401 
3402 	default:
3403 		panic("hat_supported() - unknown feature");
3404 	}
3405 	return (0);
3406 }
3407 
3408 /*
3409  * Called when a thread is exiting and has been switched to the kernel AS
3410  */
3411 void
3412 hat_thread_exit(kthread_t *thd)
3413 {
3414 	ASSERT(thd->t_procp->p_as == &kas);
3415 	hat_switch(thd->t_procp->p_as->a_hat);
3416 }
3417 
3418 /*
3419  * Setup the given brand new hat structure as the new HAT on this cpu's mmu.
3420  */
3421 /*ARGSUSED*/
3422 void
3423 hat_setup(hat_t *hat, int flags)
3424 {
3425 	kpreempt_disable();
3426 
3427 	hat_switch(hat);
3428 
3429 	kpreempt_enable();
3430 }
3431 
3432 /*
3433  * Prepare for a CPU private mapping for the given address.
3434  *
3435  * The address can only be used from a single CPU and can be remapped
3436  * using hat_mempte_remap().  Return the address of the PTE.
3437  *
3438  * We do the htable_create() if necessary and increment the valid count so
3439  * the htable can't disappear.  We also hat_devload() the page table into
3440  * kernel so that the PTE is quickly accessed.
3441  */
3442 hat_mempte_t
3443 hat_mempte_setup(caddr_t addr)
3444 {
3445 	uintptr_t	va = (uintptr_t)addr;
3446 	htable_t	*ht;
3447 	uint_t		entry;
3448 	x86pte_t	oldpte;
3449 	hat_mempte_t	p;
3450 
3451 	ASSERT(IS_PAGEALIGNED(va));
3452 	ASSERT(!IN_VA_HOLE(va));
3453 	++curthread->t_hatdepth;
3454 	ht = htable_getpte(kas.a_hat, va, &entry, &oldpte, 0);
3455 	if (ht == NULL) {
3456 		ht = htable_create(kas.a_hat, va, 0, NULL);
3457 		entry = htable_va2entry(va, ht);
3458 		ASSERT(ht->ht_level == 0);
3459 		oldpte = x86pte_get(ht, entry);
3460 	}
3461 	if (PTE_ISVALID(oldpte))
3462 		panic("hat_mempte_setup(): address already mapped"
3463 		    "ht=%p, entry=%d, pte=" FMT_PTE, ht, entry, oldpte);
3464 
3465 	/*
3466 	 * increment ht_valid_cnt so that the pagetable can't disappear
3467 	 */
3468 	HTABLE_INC(ht->ht_valid_cnt);
3469 
3470 	/*
3471 	 * return the PTE physical address to the caller.
3472 	 */
3473 	htable_release(ht);
3474 	p = PT_INDEX_PHYSADDR(pfn_to_pa(ht->ht_pfn), entry);
3475 	--curthread->t_hatdepth;
3476 	return (p);
3477 }
3478 
3479 /*
3480  * Release a CPU private mapping for the given address.
3481  * We decrement the htable valid count so it might be destroyed.
3482  */
3483 /*ARGSUSED1*/
3484 void
3485 hat_mempte_release(caddr_t addr, hat_mempte_t pte_pa)
3486 {
3487 	htable_t	*ht;
3488 
3489 	/*
3490 	 * invalidate any left over mapping and decrement the htable valid count
3491 	 */
3492 	{
3493 		x86pte_t *pteptr;
3494 
3495 		pteptr = x86pte_mapin(mmu_btop(pte_pa),
3496 		    (pte_pa & MMU_PAGEOFFSET) >> mmu.pte_size_shift, NULL);
3497 		if (mmu.pae_hat)
3498 			*pteptr = 0;
3499 		else
3500 			*(x86pte32_t *)pteptr = 0;
3501 		mmu_tlbflush_entry(addr);
3502 		x86pte_mapout();
3503 	}
3504 
3505 	ht = htable_getpte(kas.a_hat, ALIGN2PAGE(addr), NULL, NULL, 0);
3506 	if (ht == NULL)
3507 		panic("hat_mempte_release(): invalid address");
3508 	ASSERT(ht->ht_level == 0);
3509 	HTABLE_DEC(ht->ht_valid_cnt);
3510 	htable_release(ht);
3511 }
3512 
3513 /*
3514  * Apply a temporary CPU private mapping to a page. We flush the TLB only
3515  * on this CPU, so this ought to have been called with preemption disabled.
3516  */
3517 void
3518 hat_mempte_remap(
3519 	pfn_t		pfn,
3520 	caddr_t		addr,
3521 	hat_mempte_t	pte_pa,
3522 	uint_t		attr,
3523 	uint_t		flags)
3524 {
3525 	uintptr_t	va = (uintptr_t)addr;
3526 	x86pte_t	pte;
3527 
3528 	/*
3529 	 * Remap the given PTE to the new page's PFN. Invalidate only
3530 	 * on this CPU.
3531 	 */
3532 #ifdef DEBUG
3533 	htable_t	*ht;
3534 	uint_t		entry;
3535 
3536 	ASSERT(IS_PAGEALIGNED(va));
3537 	ASSERT(!IN_VA_HOLE(va));
3538 	ht = htable_getpte(kas.a_hat, va, &entry, NULL, 0);
3539 	ASSERT(ht != NULL);
3540 	ASSERT(ht->ht_level == 0);
3541 	ASSERT(ht->ht_valid_cnt > 0);
3542 	ASSERT(ht->ht_pfn == mmu_btop(pte_pa));
3543 	htable_release(ht);
3544 #endif
3545 	pte = hati_mkpte(pfn, attr, 0, flags);
3546 	{
3547 		x86pte_t *pteptr;
3548 
3549 		pteptr = x86pte_mapin(mmu_btop(pte_pa),
3550 		    (pte_pa & MMU_PAGEOFFSET) >> mmu.pte_size_shift, NULL);
3551 		if (mmu.pae_hat)
3552 			*(x86pte_t *)pteptr = pte;
3553 		else
3554 			*(x86pte32_t *)pteptr = (x86pte32_t)pte;
3555 		mmu_tlbflush_entry(addr);
3556 		x86pte_mapout();
3557 	}
3558 }
3559 
3560 
3561 
3562 /*
3563  * Hat locking functions
3564  * XXX - these two functions are currently being used by hatstats
3565  * 	they can be removed by using a per-as mutex for hatstats.
3566  */
3567 void
3568 hat_enter(hat_t *hat)
3569 {
3570 	mutex_enter(&hat->hat_mutex);
3571 }
3572 
3573 void
3574 hat_exit(hat_t *hat)
3575 {
3576 	mutex_exit(&hat->hat_mutex);
3577 }
3578 
3579 /*
3580  * HAT part of cpu initialization.
3581  */
3582 void
3583 hat_cpu_online(struct cpu *cpup)
3584 {
3585 	if (cpup != CPU) {
3586 		x86pte_cpu_init(cpup);
3587 		hat_vlp_setup(cpup);
3588 	}
3589 	CPUSET_ATOMIC_ADD(khat_cpuset, cpup->cpu_id);
3590 }
3591 
3592 /*
3593  * HAT part of cpu deletion.
3594  * (currently, we only call this after the cpu is safely passivated.)
3595  */
3596 void
3597 hat_cpu_offline(struct cpu *cpup)
3598 {
3599 	ASSERT(cpup != CPU);
3600 
3601 	CPUSET_ATOMIC_DEL(khat_cpuset, cpup->cpu_id);
3602 	x86pte_cpu_fini(cpup);
3603 	hat_vlp_teardown(cpup);
3604 }
3605 
3606 /*
3607  * Function called after all CPUs are brought online.
3608  * Used to remove low address boot mappings.
3609  */
3610 void
3611 clear_boot_mappings(uintptr_t low, uintptr_t high)
3612 {
3613 	uintptr_t vaddr = low;
3614 	htable_t *ht = NULL;
3615 	level_t level;
3616 	uint_t entry;
3617 	x86pte_t pte;
3618 
3619 	/*
3620 	 * On 1st CPU we can unload the prom mappings, basically we blow away
3621 	 * all virtual mappings under _userlimit.
3622 	 */
3623 	while (vaddr < high) {
3624 		pte = htable_walk(kas.a_hat, &ht, &vaddr, high);
3625 		if (ht == NULL)
3626 			break;
3627 
3628 		level = ht->ht_level;
3629 		entry = htable_va2entry(vaddr, ht);
3630 		ASSERT(level <= mmu.max_page_level);
3631 		ASSERT(PTE_ISPAGE(pte, level));
3632 
3633 		/*
3634 		 * Unload the mapping from the page tables.
3635 		 */
3636 		(void) x86pte_inval(ht, entry, 0, NULL);
3637 		ASSERT(ht->ht_valid_cnt > 0);
3638 		HTABLE_DEC(ht->ht_valid_cnt);
3639 		PGCNT_DEC(ht->ht_hat, ht->ht_level);
3640 
3641 		vaddr += LEVEL_SIZE(ht->ht_level);
3642 	}
3643 	if (ht)
3644 		htable_release(ht);
3645 }
3646 
3647 /*
3648  * Atomically update a new translation for a single page.  If the
3649  * currently installed PTE doesn't match the value we expect to find,
3650  * it's not updated and we return the PTE we found.
3651  *
3652  * If activating nosync or NOWRITE and the page was modified we need to sync
3653  * with the page_t. Also sync with page_t if clearing ref/mod bits.
3654  */
3655 static x86pte_t
3656 hati_update_pte(htable_t *ht, uint_t entry, x86pte_t expected, x86pte_t new)
3657 {
3658 	page_t		*pp;
3659 	uint_t		rm = 0;
3660 	x86pte_t	replaced;
3661 
3662 	if (PTE_GET(expected, PT_SOFTWARE) < PT_NOSYNC &&
3663 	    PTE_GET(expected, PT_MOD | PT_REF) &&
3664 	    (PTE_GET(new, PT_NOSYNC) || !PTE_GET(new, PT_WRITABLE) ||
3665 		!PTE_GET(new, PT_MOD | PT_REF))) {
3666 
3667 		ASSERT(!pfn_is_foreign(PTE2PFN(expected, ht->ht_level)));
3668 		pp = page_numtopp_nolock(PTE2PFN(expected, ht->ht_level));
3669 		ASSERT(pp != NULL);
3670 		if (PTE_GET(expected, PT_MOD))
3671 			rm |= P_MOD;
3672 		if (PTE_GET(expected, PT_REF))
3673 			rm |= P_REF;
3674 		PTE_CLR(new, PT_MOD | PT_REF);
3675 	}
3676 
3677 	replaced = x86pte_update(ht, entry, expected, new);
3678 	if (replaced != expected)
3679 		return (replaced);
3680 
3681 	if (rm) {
3682 		/*
3683 		 * sync to all constituent pages of a large page
3684 		 */
3685 		pgcnt_t pgcnt = page_get_pagecnt(ht->ht_level);
3686 		ASSERT(IS_P2ALIGNED(pp->p_pagenum, pgcnt));
3687 		while (pgcnt-- > 0) {
3688 			/*
3689 			 * hat_page_demote() can't decrease
3690 			 * pszc below this mapping size
3691 			 * since large mapping existed after we
3692 			 * took mlist lock.
3693 			 */
3694 			ASSERT(pp->p_szc >= ht->ht_level);
3695 			hat_page_setattr(pp, rm);
3696 			++pp;
3697 		}
3698 	}
3699 
3700 	return (0);
3701 }
3702 
3703 /*
3704  * Kernel Physical Mapping (kpm) facility
3705  *
3706  * Most of the routines needed to support segkpm are almost no-ops on the
3707  * x86 platform.  We map in the entire segment when it is created and leave
3708  * it mapped in, so there is no additional work required to set up and tear
3709  * down individual mappings.  All of these routines were created to support
3710  * SPARC platforms that have to avoid aliasing in their virtually indexed
3711  * caches.
3712  *
3713  * Most of the routines have sanity checks in them (e.g. verifying that the
3714  * passed-in page is locked).  We don't actually care about most of these
3715  * checks on x86, but we leave them in place to identify problems in the
3716  * upper levels.
3717  */
3718 
3719 /*
3720  * Map in a locked page and return the vaddr.
3721  */
3722 /*ARGSUSED*/
3723 caddr_t
3724 hat_kpm_mapin(struct page *pp, struct kpme *kpme)
3725 {
3726 	caddr_t		vaddr;
3727 
3728 #ifdef DEBUG
3729 	if (kpm_enable == 0) {
3730 		cmn_err(CE_WARN, "hat_kpm_mapin: kpm_enable not set\n");
3731 		return ((caddr_t)NULL);
3732 	}
3733 
3734 	if (pp == NULL || PAGE_LOCKED(pp) == 0) {
3735 		cmn_err(CE_WARN, "hat_kpm_mapin: pp zero or not locked\n");
3736 		return ((caddr_t)NULL);
3737 	}
3738 #endif
3739 
3740 	vaddr = hat_kpm_page2va(pp, 1);
3741 
3742 	return (vaddr);
3743 }
3744 
3745 /*
3746  * Mapout a locked page.
3747  */
3748 /*ARGSUSED*/
3749 void
3750 hat_kpm_mapout(struct page *pp, struct kpme *kpme, caddr_t vaddr)
3751 {
3752 #ifdef DEBUG
3753 	if (kpm_enable == 0) {
3754 		cmn_err(CE_WARN, "hat_kpm_mapout: kpm_enable not set\n");
3755 		return;
3756 	}
3757 
3758 	if (IS_KPM_ADDR(vaddr) == 0) {
3759 		cmn_err(CE_WARN, "hat_kpm_mapout: no kpm address\n");
3760 		return;
3761 	}
3762 
3763 	if (pp == NULL || PAGE_LOCKED(pp) == 0) {
3764 		cmn_err(CE_WARN, "hat_kpm_mapout: page zero or not locked\n");
3765 		return;
3766 	}
3767 #endif
3768 }
3769 
3770 /*
3771  * Return the kpm virtual address for a specific pfn
3772  */
3773 caddr_t
3774 hat_kpm_pfn2va(pfn_t pfn)
3775 {
3776 	uintptr_t vaddr = (uintptr_t)kpm_vbase + mmu_ptob(pfn);
3777 
3778 	return ((caddr_t)vaddr);
3779 }
3780 
3781 /*
3782  * Return the kpm virtual address for the page at pp.
3783  */
3784 /*ARGSUSED*/
3785 caddr_t
3786 hat_kpm_page2va(struct page *pp, int checkswap)
3787 {
3788 	return (hat_kpm_pfn2va(pp->p_pagenum));
3789 }
3790 
3791 /*
3792  * Return the page frame number for the kpm virtual address vaddr.
3793  */
3794 pfn_t
3795 hat_kpm_va2pfn(caddr_t vaddr)
3796 {
3797 	pfn_t		pfn;
3798 
3799 	ASSERT(IS_KPM_ADDR(vaddr));
3800 
3801 	pfn = (pfn_t)btop(vaddr - kpm_vbase);
3802 
3803 	return (pfn);
3804 }
3805 
3806 
3807 /*
3808  * Return the page for the kpm virtual address vaddr.
3809  */
3810 page_t *
3811 hat_kpm_vaddr2page(caddr_t vaddr)
3812 {
3813 	pfn_t		pfn;
3814 
3815 	ASSERT(IS_KPM_ADDR(vaddr));
3816 
3817 	pfn = hat_kpm_va2pfn(vaddr);
3818 
3819 	return (page_numtopp_nolock(pfn));
3820 }
3821 
3822 /*
3823  * hat_kpm_fault is called from segkpm_fault when we take a page fault on a
3824  * KPM page.  This should never happen on x86
3825  */
3826 int
3827 hat_kpm_fault(hat_t *hat, caddr_t vaddr)
3828 {
3829 	panic("pagefault in seg_kpm.  hat: 0x%p  vaddr: 0x%p", hat, vaddr);
3830 
3831 	return (0);
3832 }
3833 
3834 /*ARGSUSED*/
3835 void
3836 hat_kpm_mseghash_clear(int nentries)
3837 {}
3838 
3839 /*ARGSUSED*/
3840 void
3841 hat_kpm_mseghash_update(pgcnt_t inx, struct memseg *msp)
3842 {}
3843