17c478bd9Sstevel@tonic-gate /* 27c478bd9Sstevel@tonic-gate * CDDL HEADER START 37c478bd9Sstevel@tonic-gate * 47c478bd9Sstevel@tonic-gate * The contents of this file are subject to the terms of the 5a85a6733Sjosephb * Common Development and Distribution License (the "License"). 6a85a6733Sjosephb * You may not use this file except in compliance with the License. 77c478bd9Sstevel@tonic-gate * 87c478bd9Sstevel@tonic-gate * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 97c478bd9Sstevel@tonic-gate * or http://www.opensolaris.org/os/licensing. 107c478bd9Sstevel@tonic-gate * See the License for the specific language governing permissions 117c478bd9Sstevel@tonic-gate * and limitations under the License. 127c478bd9Sstevel@tonic-gate * 137c478bd9Sstevel@tonic-gate * When distributing Covered Code, include this CDDL HEADER in each 147c478bd9Sstevel@tonic-gate * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 157c478bd9Sstevel@tonic-gate * If applicable, add the following below this CDDL HEADER, with the 167c478bd9Sstevel@tonic-gate * fields enclosed by brackets "[]" replaced with your own identifying 177c478bd9Sstevel@tonic-gate * information: Portions Copyright [yyyy] [name of copyright owner] 187c478bd9Sstevel@tonic-gate * 197c478bd9Sstevel@tonic-gate * CDDL HEADER END 207c478bd9Sstevel@tonic-gate */ 21ae115bc7Smrj 227c478bd9Sstevel@tonic-gate /* 23ae115bc7Smrj * Copyright 2007 Sun Microsystems, Inc. All rights reserved. 247c478bd9Sstevel@tonic-gate * Use is subject to license terms. 257c478bd9Sstevel@tonic-gate */ 267c478bd9Sstevel@tonic-gate 277c478bd9Sstevel@tonic-gate #pragma ident "%Z%%M% %I% %E% SMI" 287c478bd9Sstevel@tonic-gate 297c478bd9Sstevel@tonic-gate #include <sys/types.h> 307c478bd9Sstevel@tonic-gate #include <sys/sysmacros.h> 317c478bd9Sstevel@tonic-gate #include <sys/kmem.h> 327c478bd9Sstevel@tonic-gate #include <sys/atomic.h> 337c478bd9Sstevel@tonic-gate #include <sys/bitmap.h> 347c478bd9Sstevel@tonic-gate #include <sys/machparam.h> 357c478bd9Sstevel@tonic-gate #include <sys/machsystm.h> 367c478bd9Sstevel@tonic-gate #include <sys/mman.h> 377c478bd9Sstevel@tonic-gate #include <sys/systm.h> 387c478bd9Sstevel@tonic-gate #include <sys/cpuvar.h> 397c478bd9Sstevel@tonic-gate #include <sys/thread.h> 407c478bd9Sstevel@tonic-gate #include <sys/proc.h> 417c478bd9Sstevel@tonic-gate #include <sys/cpu.h> 427c478bd9Sstevel@tonic-gate #include <sys/kmem.h> 437c478bd9Sstevel@tonic-gate #include <sys/disp.h> 447c478bd9Sstevel@tonic-gate #include <sys/vmem.h> 457c478bd9Sstevel@tonic-gate #include <sys/vmsystm.h> 467c478bd9Sstevel@tonic-gate #include <sys/promif.h> 477c478bd9Sstevel@tonic-gate #include <sys/var.h> 487c478bd9Sstevel@tonic-gate #include <sys/x86_archext.h> 49ae115bc7Smrj #include <sys/archsystm.h> 507c478bd9Sstevel@tonic-gate #include <sys/bootconf.h> 517c478bd9Sstevel@tonic-gate #include <sys/dumphdr.h> 527c478bd9Sstevel@tonic-gate #include <vm/seg_kmem.h> 537c478bd9Sstevel@tonic-gate #include <vm/seg_kpm.h> 547c478bd9Sstevel@tonic-gate #include <vm/hat.h> 557c478bd9Sstevel@tonic-gate #include <vm/hat_i86.h> 567c478bd9Sstevel@tonic-gate #include <sys/cmn_err.h> 577c478bd9Sstevel@tonic-gate 58ae115bc7Smrj #include <sys/bootinfo.h> 59ae115bc7Smrj #include <vm/kboot_mmu.h> 60ae115bc7Smrj 61ae115bc7Smrj static void x86pte_zero(htable_t *dest, uint_t entry, uint_t count); 62ae115bc7Smrj 637c478bd9Sstevel@tonic-gate kmem_cache_t *htable_cache; 647c478bd9Sstevel@tonic-gate 657c478bd9Sstevel@tonic-gate /* 667c478bd9Sstevel@tonic-gate * The variable htable_reserve_amount, rather than HTABLE_RESERVE_AMOUNT, 677c478bd9Sstevel@tonic-gate * is used in order to facilitate testing of the htable_steal() code. 687c478bd9Sstevel@tonic-gate * By resetting htable_reserve_amount to a lower value, we can force 697c478bd9Sstevel@tonic-gate * stealing to occur. The reserve amount is a guess to get us through boot. 707c478bd9Sstevel@tonic-gate */ 717c478bd9Sstevel@tonic-gate #define HTABLE_RESERVE_AMOUNT (200) 727c478bd9Sstevel@tonic-gate uint_t htable_reserve_amount = HTABLE_RESERVE_AMOUNT; 737c478bd9Sstevel@tonic-gate kmutex_t htable_reserve_mutex; 747c478bd9Sstevel@tonic-gate uint_t htable_reserve_cnt; 757c478bd9Sstevel@tonic-gate htable_t *htable_reserve_pool; 767c478bd9Sstevel@tonic-gate 777c478bd9Sstevel@tonic-gate /* 78a85a6733Sjosephb * Used to hand test htable_steal(). 797c478bd9Sstevel@tonic-gate */ 80a85a6733Sjosephb #ifdef DEBUG 81a85a6733Sjosephb ulong_t force_steal = 0; 82a85a6733Sjosephb ulong_t ptable_cnt = 0; 83a85a6733Sjosephb #endif 84a85a6733Sjosephb 85a85a6733Sjosephb /* 86a85a6733Sjosephb * This variable is so that we can tune this via /etc/system 87a85a6733Sjosephb * Any value works, but a power of two <= mmu.ptes_per_table is best. 88a85a6733Sjosephb */ 89a85a6733Sjosephb uint_t htable_steal_passes = 8; 907c478bd9Sstevel@tonic-gate 917c478bd9Sstevel@tonic-gate /* 927c478bd9Sstevel@tonic-gate * mutex stuff for access to htable hash 937c478bd9Sstevel@tonic-gate */ 947c478bd9Sstevel@tonic-gate #define NUM_HTABLE_MUTEX 128 957c478bd9Sstevel@tonic-gate kmutex_t htable_mutex[NUM_HTABLE_MUTEX]; 967c478bd9Sstevel@tonic-gate #define HTABLE_MUTEX_HASH(h) ((h) & (NUM_HTABLE_MUTEX - 1)) 977c478bd9Sstevel@tonic-gate 987c478bd9Sstevel@tonic-gate #define HTABLE_ENTER(h) mutex_enter(&htable_mutex[HTABLE_MUTEX_HASH(h)]); 997c478bd9Sstevel@tonic-gate #define HTABLE_EXIT(h) mutex_exit(&htable_mutex[HTABLE_MUTEX_HASH(h)]); 1007c478bd9Sstevel@tonic-gate 1017c478bd9Sstevel@tonic-gate /* 1027c478bd9Sstevel@tonic-gate * forward declarations 1037c478bd9Sstevel@tonic-gate */ 1047c478bd9Sstevel@tonic-gate static void link_ptp(htable_t *higher, htable_t *new, uintptr_t vaddr); 1057c478bd9Sstevel@tonic-gate static void unlink_ptp(htable_t *higher, htable_t *old, uintptr_t vaddr); 1067c478bd9Sstevel@tonic-gate static void htable_free(htable_t *ht); 107ae115bc7Smrj static x86pte_t *x86pte_access_pagetable(htable_t *ht, uint_t index); 1087c478bd9Sstevel@tonic-gate static void x86pte_release_pagetable(htable_t *ht); 1097c478bd9Sstevel@tonic-gate static x86pte_t x86pte_cas(htable_t *ht, uint_t entry, x86pte_t old, 1107c478bd9Sstevel@tonic-gate x86pte_t new); 1117c478bd9Sstevel@tonic-gate 1127c478bd9Sstevel@tonic-gate /* 1137c478bd9Sstevel@tonic-gate * A counter to track if we are stealing or reaping htables. When non-zero 1147c478bd9Sstevel@tonic-gate * htable_free() will directly free htables (either to the reserve or kmem) 1157c478bd9Sstevel@tonic-gate * instead of putting them in a hat's htable cache. 1167c478bd9Sstevel@tonic-gate */ 1177c478bd9Sstevel@tonic-gate uint32_t htable_dont_cache = 0; 1187c478bd9Sstevel@tonic-gate 1197c478bd9Sstevel@tonic-gate /* 1207c478bd9Sstevel@tonic-gate * Track the number of active pagetables, so we can know how many to reap 1217c478bd9Sstevel@tonic-gate */ 1227c478bd9Sstevel@tonic-gate static uint32_t active_ptables = 0; 1237c478bd9Sstevel@tonic-gate 1247c478bd9Sstevel@tonic-gate /* 1257c478bd9Sstevel@tonic-gate * Allocate a memory page for a hardware page table. 1267c478bd9Sstevel@tonic-gate * 127ae115bc7Smrj * A wrapper around page_get_physical(), with some extra checks. 1287c478bd9Sstevel@tonic-gate */ 129ae115bc7Smrj static pfn_t 130ae115bc7Smrj ptable_alloc(uintptr_t seed) 1317c478bd9Sstevel@tonic-gate { 1327c478bd9Sstevel@tonic-gate pfn_t pfn; 1337c478bd9Sstevel@tonic-gate page_t *pp; 1347c478bd9Sstevel@tonic-gate 135ae115bc7Smrj pfn = PFN_INVALID; 1367c478bd9Sstevel@tonic-gate atomic_add_32(&active_ptables, 1); 1377c478bd9Sstevel@tonic-gate 1387c478bd9Sstevel@tonic-gate /* 139ae115bc7Smrj * The first check is to see if there is memory in the system. If we 140ae115bc7Smrj * drop to throttlefree, then fail the ptable_alloc() and let the 141ae115bc7Smrj * stealing code kick in. Note that we have to do this test here, 142ae115bc7Smrj * since the test in page_create_throttle() would let the NOSLEEP 143ae115bc7Smrj * allocation go through and deplete the page reserves. 144a85a6733Sjosephb * 145a85a6733Sjosephb * The !NOMEMWAIT() lets pageout, fsflush, etc. skip this check. 1467c478bd9Sstevel@tonic-gate */ 147a85a6733Sjosephb if (!NOMEMWAIT() && freemem <= throttlefree + 1) 148ae115bc7Smrj return (PFN_INVALID); 1497c478bd9Sstevel@tonic-gate 150a85a6733Sjosephb #ifdef DEBUG 151a85a6733Sjosephb /* 152a85a6733Sjosephb * This code makes htable_steal() easier to test. By setting 153a85a6733Sjosephb * force_steal we force pagetable allocations to fall 154a85a6733Sjosephb * into the stealing code. Roughly 1 in ever "force_steal" 155a85a6733Sjosephb * page table allocations will fail. 156a85a6733Sjosephb */ 157ae115bc7Smrj if (proc_pageout != NULL && force_steal > 1 && 158a85a6733Sjosephb ++ptable_cnt > force_steal) { 159a85a6733Sjosephb ptable_cnt = 0; 160ae115bc7Smrj return (PFN_INVALID); 161a85a6733Sjosephb } 162a85a6733Sjosephb #endif /* DEBUG */ 163a85a6733Sjosephb 164ae115bc7Smrj pp = page_get_physical(seed); 1657c478bd9Sstevel@tonic-gate if (pp == NULL) 166ae115bc7Smrj return (PFN_INVALID); 1677c478bd9Sstevel@tonic-gate pfn = pp->p_pagenum; 1687c478bd9Sstevel@tonic-gate page_downgrade(pp); 1697c478bd9Sstevel@tonic-gate ASSERT(PAGE_SHARED(pp)); 1707c478bd9Sstevel@tonic-gate 1717c478bd9Sstevel@tonic-gate if (pfn == PFN_INVALID) 1727c478bd9Sstevel@tonic-gate panic("ptable_alloc(): Invalid PFN!!"); 173a85a6733Sjosephb HATSTAT_INC(hs_ptable_allocs); 174ae115bc7Smrj return (pfn); 1757c478bd9Sstevel@tonic-gate } 1767c478bd9Sstevel@tonic-gate 1777c478bd9Sstevel@tonic-gate /* 1787c478bd9Sstevel@tonic-gate * Free an htable's associated page table page. See the comments 1797c478bd9Sstevel@tonic-gate * for ptable_alloc(). 1807c478bd9Sstevel@tonic-gate */ 1817c478bd9Sstevel@tonic-gate static void 182ae115bc7Smrj ptable_free(pfn_t pfn) 1837c478bd9Sstevel@tonic-gate { 184ae115bc7Smrj page_t *pp = page_numtopp_nolock(pfn); 1857c478bd9Sstevel@tonic-gate 1867c478bd9Sstevel@tonic-gate /* 1877c478bd9Sstevel@tonic-gate * need to destroy the page used for the pagetable 1887c478bd9Sstevel@tonic-gate */ 1897c478bd9Sstevel@tonic-gate ASSERT(pfn != PFN_INVALID); 1907c478bd9Sstevel@tonic-gate HATSTAT_INC(hs_ptable_frees); 1917c478bd9Sstevel@tonic-gate atomic_add_32(&active_ptables, -1); 1927c478bd9Sstevel@tonic-gate if (pp == NULL) 1937c478bd9Sstevel@tonic-gate panic("ptable_free(): no page for pfn!"); 1947c478bd9Sstevel@tonic-gate ASSERT(PAGE_SHARED(pp)); 1957c478bd9Sstevel@tonic-gate ASSERT(pfn == pp->p_pagenum); 1967c478bd9Sstevel@tonic-gate 1977c478bd9Sstevel@tonic-gate /* 1987c478bd9Sstevel@tonic-gate * Get an exclusive lock, might have to wait for a kmem reader. 1997c478bd9Sstevel@tonic-gate */ 2007c478bd9Sstevel@tonic-gate if (!page_tryupgrade(pp)) { 2017c478bd9Sstevel@tonic-gate page_unlock(pp); 2027c478bd9Sstevel@tonic-gate /* 2037c478bd9Sstevel@tonic-gate * RFE: we could change this to not loop forever 2047c478bd9Sstevel@tonic-gate * George Cameron had some idea on how to do that. 2057c478bd9Sstevel@tonic-gate * For now looping works - it's just like sfmmu. 2067c478bd9Sstevel@tonic-gate */ 2077c478bd9Sstevel@tonic-gate while (!page_lock(pp, SE_EXCL, (kmutex_t *)NULL, P_RECLAIM)) 2087c478bd9Sstevel@tonic-gate continue; 2097c478bd9Sstevel@tonic-gate } 2107c478bd9Sstevel@tonic-gate page_free(pp, 1); 2117c478bd9Sstevel@tonic-gate page_unresv(1); 2127c478bd9Sstevel@tonic-gate } 2137c478bd9Sstevel@tonic-gate 2147c478bd9Sstevel@tonic-gate /* 2157c478bd9Sstevel@tonic-gate * Put one htable on the reserve list. 2167c478bd9Sstevel@tonic-gate */ 2177c478bd9Sstevel@tonic-gate static void 2187c478bd9Sstevel@tonic-gate htable_put_reserve(htable_t *ht) 2197c478bd9Sstevel@tonic-gate { 2207c478bd9Sstevel@tonic-gate ht->ht_hat = NULL; /* no longer tied to a hat */ 2217c478bd9Sstevel@tonic-gate ASSERT(ht->ht_pfn == PFN_INVALID); 2227c478bd9Sstevel@tonic-gate HATSTAT_INC(hs_htable_rputs); 2237c478bd9Sstevel@tonic-gate mutex_enter(&htable_reserve_mutex); 2247c478bd9Sstevel@tonic-gate ht->ht_next = htable_reserve_pool; 2257c478bd9Sstevel@tonic-gate htable_reserve_pool = ht; 2267c478bd9Sstevel@tonic-gate ++htable_reserve_cnt; 2277c478bd9Sstevel@tonic-gate mutex_exit(&htable_reserve_mutex); 2287c478bd9Sstevel@tonic-gate } 2297c478bd9Sstevel@tonic-gate 2307c478bd9Sstevel@tonic-gate /* 2317c478bd9Sstevel@tonic-gate * Take one htable from the reserve. 2327c478bd9Sstevel@tonic-gate */ 2337c478bd9Sstevel@tonic-gate static htable_t * 2347c478bd9Sstevel@tonic-gate htable_get_reserve(void) 2357c478bd9Sstevel@tonic-gate { 2367c478bd9Sstevel@tonic-gate htable_t *ht = NULL; 2377c478bd9Sstevel@tonic-gate 2387c478bd9Sstevel@tonic-gate mutex_enter(&htable_reserve_mutex); 2397c478bd9Sstevel@tonic-gate if (htable_reserve_cnt != 0) { 2407c478bd9Sstevel@tonic-gate ht = htable_reserve_pool; 2417c478bd9Sstevel@tonic-gate ASSERT(ht != NULL); 2427c478bd9Sstevel@tonic-gate ASSERT(ht->ht_pfn == PFN_INVALID); 2437c478bd9Sstevel@tonic-gate htable_reserve_pool = ht->ht_next; 2447c478bd9Sstevel@tonic-gate --htable_reserve_cnt; 2457c478bd9Sstevel@tonic-gate HATSTAT_INC(hs_htable_rgets); 2467c478bd9Sstevel@tonic-gate } 2477c478bd9Sstevel@tonic-gate mutex_exit(&htable_reserve_mutex); 2487c478bd9Sstevel@tonic-gate return (ht); 2497c478bd9Sstevel@tonic-gate } 2507c478bd9Sstevel@tonic-gate 2517c478bd9Sstevel@tonic-gate /* 252ae115bc7Smrj * Allocate initial htables and put them on the reserve list 2537c478bd9Sstevel@tonic-gate */ 2547c478bd9Sstevel@tonic-gate void 2557c478bd9Sstevel@tonic-gate htable_initial_reserve(uint_t count) 2567c478bd9Sstevel@tonic-gate { 2577c478bd9Sstevel@tonic-gate htable_t *ht; 2587c478bd9Sstevel@tonic-gate 2597c478bd9Sstevel@tonic-gate count += HTABLE_RESERVE_AMOUNT; 2607c478bd9Sstevel@tonic-gate while (count > 0) { 2617c478bd9Sstevel@tonic-gate ht = kmem_cache_alloc(htable_cache, KM_NOSLEEP); 2627c478bd9Sstevel@tonic-gate ASSERT(ht != NULL); 2637c478bd9Sstevel@tonic-gate 2647c478bd9Sstevel@tonic-gate ASSERT(use_boot_reserve); 265ae115bc7Smrj ht->ht_pfn = PFN_INVALID; 266ae115bc7Smrj htable_put_reserve(ht); 2677c478bd9Sstevel@tonic-gate --count; 2687c478bd9Sstevel@tonic-gate } 2697c478bd9Sstevel@tonic-gate } 2707c478bd9Sstevel@tonic-gate 2717c478bd9Sstevel@tonic-gate /* 2727c478bd9Sstevel@tonic-gate * Readjust the reserves after a thread finishes using them. 2737c478bd9Sstevel@tonic-gate */ 2747c478bd9Sstevel@tonic-gate void 2757c478bd9Sstevel@tonic-gate htable_adjust_reserve() 2767c478bd9Sstevel@tonic-gate { 2777c478bd9Sstevel@tonic-gate htable_t *ht; 2787c478bd9Sstevel@tonic-gate 2797c478bd9Sstevel@tonic-gate /* 2807c478bd9Sstevel@tonic-gate * Free any excess htables in the reserve list 2817c478bd9Sstevel@tonic-gate */ 282aac11643Sjosephb while (htable_reserve_cnt > htable_reserve_amount && 283aac11643Sjosephb !USE_HAT_RESERVES()) { 2847c478bd9Sstevel@tonic-gate ht = htable_get_reserve(); 2857c478bd9Sstevel@tonic-gate if (ht == NULL) 2867c478bd9Sstevel@tonic-gate return; 2877c478bd9Sstevel@tonic-gate ASSERT(ht->ht_pfn == PFN_INVALID); 2887c478bd9Sstevel@tonic-gate kmem_cache_free(htable_cache, ht); 2897c478bd9Sstevel@tonic-gate } 2907c478bd9Sstevel@tonic-gate } 2917c478bd9Sstevel@tonic-gate 2927c478bd9Sstevel@tonic-gate 2937c478bd9Sstevel@tonic-gate /* 2947c478bd9Sstevel@tonic-gate * This routine steals htables from user processes for htable_alloc() or 2957c478bd9Sstevel@tonic-gate * for htable_reap(). 2967c478bd9Sstevel@tonic-gate */ 2977c478bd9Sstevel@tonic-gate static htable_t * 2987c478bd9Sstevel@tonic-gate htable_steal(uint_t cnt) 2997c478bd9Sstevel@tonic-gate { 3007c478bd9Sstevel@tonic-gate hat_t *hat = kas.a_hat; /* list starts with khat */ 3017c478bd9Sstevel@tonic-gate htable_t *list = NULL; 3027c478bd9Sstevel@tonic-gate htable_t *ht; 3037c478bd9Sstevel@tonic-gate htable_t *higher; 3047c478bd9Sstevel@tonic-gate uint_t h; 305a85a6733Sjosephb uint_t h_start; 306a85a6733Sjosephb static uint_t h_seed = 0; 3077c478bd9Sstevel@tonic-gate uint_t e; 3087c478bd9Sstevel@tonic-gate uintptr_t va; 3097c478bd9Sstevel@tonic-gate x86pte_t pte; 3107c478bd9Sstevel@tonic-gate uint_t stolen = 0; 3117c478bd9Sstevel@tonic-gate uint_t pass; 312a85a6733Sjosephb uint_t threshold; 3137c478bd9Sstevel@tonic-gate 3147c478bd9Sstevel@tonic-gate /* 3157c478bd9Sstevel@tonic-gate * Limit htable_steal_passes to something reasonable 3167c478bd9Sstevel@tonic-gate */ 3177c478bd9Sstevel@tonic-gate if (htable_steal_passes == 0) 3187c478bd9Sstevel@tonic-gate htable_steal_passes = 1; 3197c478bd9Sstevel@tonic-gate if (htable_steal_passes > mmu.ptes_per_table) 3207c478bd9Sstevel@tonic-gate htable_steal_passes = mmu.ptes_per_table; 3217c478bd9Sstevel@tonic-gate 3227c478bd9Sstevel@tonic-gate /* 323a85a6733Sjosephb * Loop through all user hats. The 1st pass takes cached htables that 3247c478bd9Sstevel@tonic-gate * aren't in use. The later passes steal by removing mappings, too. 3257c478bd9Sstevel@tonic-gate */ 3267c478bd9Sstevel@tonic-gate atomic_add_32(&htable_dont_cache, 1); 327a85a6733Sjosephb for (pass = 0; pass <= htable_steal_passes && stolen < cnt; ++pass) { 328a85a6733Sjosephb threshold = pass * mmu.ptes_per_table / htable_steal_passes; 329a85a6733Sjosephb hat = kas.a_hat; 3307c478bd9Sstevel@tonic-gate for (;;) { 3317c478bd9Sstevel@tonic-gate 3327c478bd9Sstevel@tonic-gate /* 333a85a6733Sjosephb * Clear the victim flag and move to next hat 3347c478bd9Sstevel@tonic-gate */ 3357c478bd9Sstevel@tonic-gate mutex_enter(&hat_list_lock); 336a85a6733Sjosephb if (hat != kas.a_hat) { 3377c478bd9Sstevel@tonic-gate hat->hat_flags &= ~HAT_VICTIM; 3387c478bd9Sstevel@tonic-gate cv_broadcast(&hat_list_cv); 339a85a6733Sjosephb } 340a85a6733Sjosephb hat = hat->hat_next; 341a85a6733Sjosephb 342a85a6733Sjosephb /* 343a85a6733Sjosephb * Skip any hat that is already being stolen from. 344a85a6733Sjosephb * 345a85a6733Sjosephb * We skip SHARED hats, as these are dummy 346a85a6733Sjosephb * hats that host ISM shared page tables. 347a85a6733Sjosephb * 348a85a6733Sjosephb * We also skip if HAT_FREEING because hat_pte_unmap() 349a85a6733Sjosephb * won't zero out the PTE's. That would lead to hitting 350a85a6733Sjosephb * stale PTEs either here or under hat_unload() when we 351a85a6733Sjosephb * steal and unload the same page table in competing 352a85a6733Sjosephb * threads. 353a85a6733Sjosephb */ 354a85a6733Sjosephb while (hat != NULL && 355a85a6733Sjosephb (hat->hat_flags & 356a85a6733Sjosephb (HAT_VICTIM | HAT_SHARED | HAT_FREEING)) != 0) 357a85a6733Sjosephb hat = hat->hat_next; 358a85a6733Sjosephb 359a85a6733Sjosephb if (hat == NULL) { 3607c478bd9Sstevel@tonic-gate mutex_exit(&hat_list_lock); 3617c478bd9Sstevel@tonic-gate break; 3627c478bd9Sstevel@tonic-gate } 363a85a6733Sjosephb 364a85a6733Sjosephb /* 365a85a6733Sjosephb * Are we finished? 366a85a6733Sjosephb */ 367a85a6733Sjosephb if (stolen == cnt) { 368a85a6733Sjosephb /* 369a85a6733Sjosephb * Try to spread the pain of stealing, 370a85a6733Sjosephb * move victim HAT to the end of the HAT list. 371a85a6733Sjosephb */ 372a85a6733Sjosephb if (pass >= 1 && cnt == 1 && 373a85a6733Sjosephb kas.a_hat->hat_prev != hat) { 374a85a6733Sjosephb 375a85a6733Sjosephb /* unlink victim hat */ 376a85a6733Sjosephb if (hat->hat_prev) 377a85a6733Sjosephb hat->hat_prev->hat_next = 378a85a6733Sjosephb hat->hat_next; 379a85a6733Sjosephb else 380a85a6733Sjosephb kas.a_hat->hat_next = 381a85a6733Sjosephb hat->hat_next; 382a85a6733Sjosephb if (hat->hat_next) 383a85a6733Sjosephb hat->hat_next->hat_prev = 384a85a6733Sjosephb hat->hat_prev; 385a85a6733Sjosephb else 386a85a6733Sjosephb kas.a_hat->hat_prev = 387a85a6733Sjosephb hat->hat_prev; 388a85a6733Sjosephb 389a85a6733Sjosephb 390a85a6733Sjosephb /* relink at end of hat list */ 391a85a6733Sjosephb hat->hat_next = NULL; 392a85a6733Sjosephb hat->hat_prev = kas.a_hat->hat_prev; 393a85a6733Sjosephb if (hat->hat_prev) 394a85a6733Sjosephb hat->hat_prev->hat_next = hat; 395a85a6733Sjosephb else 396a85a6733Sjosephb kas.a_hat->hat_next = hat; 397a85a6733Sjosephb kas.a_hat->hat_prev = hat; 398a85a6733Sjosephb 399a85a6733Sjosephb } 400a85a6733Sjosephb 401a85a6733Sjosephb mutex_exit(&hat_list_lock); 402a85a6733Sjosephb break; 403a85a6733Sjosephb } 404a85a6733Sjosephb 405a85a6733Sjosephb /* 406a85a6733Sjosephb * Mark the HAT as a stealing victim. 407a85a6733Sjosephb */ 4087c478bd9Sstevel@tonic-gate hat->hat_flags |= HAT_VICTIM; 4097c478bd9Sstevel@tonic-gate mutex_exit(&hat_list_lock); 4107c478bd9Sstevel@tonic-gate 4117c478bd9Sstevel@tonic-gate /* 4127c478bd9Sstevel@tonic-gate * Take any htables from the hat's cached "free" list. 4137c478bd9Sstevel@tonic-gate */ 4147c478bd9Sstevel@tonic-gate hat_enter(hat); 4157c478bd9Sstevel@tonic-gate while ((ht = hat->hat_ht_cached) != NULL && 4167c478bd9Sstevel@tonic-gate stolen < cnt) { 4177c478bd9Sstevel@tonic-gate hat->hat_ht_cached = ht->ht_next; 4187c478bd9Sstevel@tonic-gate ht->ht_next = list; 4197c478bd9Sstevel@tonic-gate list = ht; 4207c478bd9Sstevel@tonic-gate ++stolen; 4217c478bd9Sstevel@tonic-gate } 4227c478bd9Sstevel@tonic-gate hat_exit(hat); 4237c478bd9Sstevel@tonic-gate 4247c478bd9Sstevel@tonic-gate /* 4257c478bd9Sstevel@tonic-gate * Don't steal on first pass. 4267c478bd9Sstevel@tonic-gate */ 427a85a6733Sjosephb if (pass == 0 || stolen == cnt) 4287c478bd9Sstevel@tonic-gate continue; 4297c478bd9Sstevel@tonic-gate 4307c478bd9Sstevel@tonic-gate /* 431a85a6733Sjosephb * Search the active htables for one to steal. 432a85a6733Sjosephb * Start at a different hash bucket every time to 433a85a6733Sjosephb * help spread the pain of stealing. 4347c478bd9Sstevel@tonic-gate */ 435a85a6733Sjosephb h = h_start = h_seed++ % hat->hat_num_hash; 436a85a6733Sjosephb do { 4377c478bd9Sstevel@tonic-gate higher = NULL; 4387c478bd9Sstevel@tonic-gate HTABLE_ENTER(h); 4397c478bd9Sstevel@tonic-gate for (ht = hat->hat_ht_hash[h]; ht; 4407c478bd9Sstevel@tonic-gate ht = ht->ht_next) { 4417c478bd9Sstevel@tonic-gate 4427c478bd9Sstevel@tonic-gate /* 4437c478bd9Sstevel@tonic-gate * Can we rule out reaping? 4447c478bd9Sstevel@tonic-gate */ 4457c478bd9Sstevel@tonic-gate if (ht->ht_busy != 0 || 4467c478bd9Sstevel@tonic-gate (ht->ht_flags & HTABLE_SHARED_PFN)|| 447a85a6733Sjosephb ht->ht_level > 0 || 448a85a6733Sjosephb ht->ht_valid_cnt > threshold || 4497c478bd9Sstevel@tonic-gate ht->ht_lock_cnt != 0) 4507c478bd9Sstevel@tonic-gate continue; 4517c478bd9Sstevel@tonic-gate 4527c478bd9Sstevel@tonic-gate /* 4537c478bd9Sstevel@tonic-gate * Increment busy so the htable can't 4547c478bd9Sstevel@tonic-gate * disappear. We drop the htable mutex 4557c478bd9Sstevel@tonic-gate * to avoid deadlocks with 4567c478bd9Sstevel@tonic-gate * hat_pageunload() and the hment mutex 4577c478bd9Sstevel@tonic-gate * while we call hat_pte_unmap() 4587c478bd9Sstevel@tonic-gate */ 4597c478bd9Sstevel@tonic-gate ++ht->ht_busy; 4607c478bd9Sstevel@tonic-gate HTABLE_EXIT(h); 4617c478bd9Sstevel@tonic-gate 4627c478bd9Sstevel@tonic-gate /* 4637c478bd9Sstevel@tonic-gate * Try stealing. 4647c478bd9Sstevel@tonic-gate * - unload and invalidate all PTEs 4657c478bd9Sstevel@tonic-gate */ 4667c478bd9Sstevel@tonic-gate for (e = 0, va = ht->ht_vaddr; 467ae115bc7Smrj e < HTABLE_NUM_PTES(ht) && 4687c478bd9Sstevel@tonic-gate ht->ht_valid_cnt > 0 && 4697c478bd9Sstevel@tonic-gate ht->ht_busy == 1 && 4707c478bd9Sstevel@tonic-gate ht->ht_lock_cnt == 0; 4717c478bd9Sstevel@tonic-gate ++e, va += MMU_PAGESIZE) { 4727c478bd9Sstevel@tonic-gate pte = x86pte_get(ht, e); 4737c478bd9Sstevel@tonic-gate if (!PTE_ISVALID(pte)) 4747c478bd9Sstevel@tonic-gate continue; 4757c478bd9Sstevel@tonic-gate hat_pte_unmap(ht, e, 4767c478bd9Sstevel@tonic-gate HAT_UNLOAD, pte, NULL); 4777c478bd9Sstevel@tonic-gate } 4787c478bd9Sstevel@tonic-gate 4797c478bd9Sstevel@tonic-gate /* 4807c478bd9Sstevel@tonic-gate * Reacquire htable lock. If we didn't 4817c478bd9Sstevel@tonic-gate * remove all mappings in the table, 4827c478bd9Sstevel@tonic-gate * or another thread added a new mapping 4837c478bd9Sstevel@tonic-gate * behind us, give up on this table. 4847c478bd9Sstevel@tonic-gate */ 4857c478bd9Sstevel@tonic-gate HTABLE_ENTER(h); 4867c478bd9Sstevel@tonic-gate if (ht->ht_busy != 1 || 4877c478bd9Sstevel@tonic-gate ht->ht_valid_cnt != 0 || 4887c478bd9Sstevel@tonic-gate ht->ht_lock_cnt != 0) { 4897c478bd9Sstevel@tonic-gate --ht->ht_busy; 4907c478bd9Sstevel@tonic-gate continue; 4917c478bd9Sstevel@tonic-gate } 4927c478bd9Sstevel@tonic-gate 4937c478bd9Sstevel@tonic-gate /* 4947c478bd9Sstevel@tonic-gate * Steal it and unlink the page table. 4957c478bd9Sstevel@tonic-gate */ 4967c478bd9Sstevel@tonic-gate higher = ht->ht_parent; 4977c478bd9Sstevel@tonic-gate unlink_ptp(higher, ht, ht->ht_vaddr); 4987c478bd9Sstevel@tonic-gate 4997c478bd9Sstevel@tonic-gate /* 5007c478bd9Sstevel@tonic-gate * remove from the hash list 5017c478bd9Sstevel@tonic-gate */ 5027c478bd9Sstevel@tonic-gate if (ht->ht_next) 5037c478bd9Sstevel@tonic-gate ht->ht_next->ht_prev = 5047c478bd9Sstevel@tonic-gate ht->ht_prev; 5057c478bd9Sstevel@tonic-gate 5067c478bd9Sstevel@tonic-gate if (ht->ht_prev) { 5077c478bd9Sstevel@tonic-gate ht->ht_prev->ht_next = 5087c478bd9Sstevel@tonic-gate ht->ht_next; 5097c478bd9Sstevel@tonic-gate } else { 5107c478bd9Sstevel@tonic-gate ASSERT(hat->hat_ht_hash[h] == 5117c478bd9Sstevel@tonic-gate ht); 5127c478bd9Sstevel@tonic-gate hat->hat_ht_hash[h] = 5137c478bd9Sstevel@tonic-gate ht->ht_next; 5147c478bd9Sstevel@tonic-gate } 5157c478bd9Sstevel@tonic-gate 5167c478bd9Sstevel@tonic-gate /* 5177c478bd9Sstevel@tonic-gate * Break to outer loop to release the 518ae115bc7Smrj * higher (ht_parent) pagetable. This 5197c478bd9Sstevel@tonic-gate * spreads out the pain caused by 5207c478bd9Sstevel@tonic-gate * pagefaults. 5217c478bd9Sstevel@tonic-gate */ 5227c478bd9Sstevel@tonic-gate ht->ht_next = list; 5237c478bd9Sstevel@tonic-gate list = ht; 5247c478bd9Sstevel@tonic-gate ++stolen; 5257c478bd9Sstevel@tonic-gate break; 5267c478bd9Sstevel@tonic-gate } 5277c478bd9Sstevel@tonic-gate HTABLE_EXIT(h); 5287c478bd9Sstevel@tonic-gate if (higher != NULL) 5297c478bd9Sstevel@tonic-gate htable_release(higher); 530a85a6733Sjosephb if (++h == hat->hat_num_hash) 531a85a6733Sjosephb h = 0; 532a85a6733Sjosephb } while (stolen < cnt && h != h_start); 5337c478bd9Sstevel@tonic-gate } 5347c478bd9Sstevel@tonic-gate } 5357c478bd9Sstevel@tonic-gate atomic_add_32(&htable_dont_cache, -1); 5367c478bd9Sstevel@tonic-gate return (list); 5377c478bd9Sstevel@tonic-gate } 5387c478bd9Sstevel@tonic-gate 5397c478bd9Sstevel@tonic-gate 5407c478bd9Sstevel@tonic-gate /* 5417c478bd9Sstevel@tonic-gate * This is invoked from kmem when the system is low on memory. We try 5427c478bd9Sstevel@tonic-gate * to free hments, htables, and ptables to improve the memory situation. 5437c478bd9Sstevel@tonic-gate */ 5447c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 5457c478bd9Sstevel@tonic-gate static void 5467c478bd9Sstevel@tonic-gate htable_reap(void *handle) 5477c478bd9Sstevel@tonic-gate { 5487c478bd9Sstevel@tonic-gate uint_t reap_cnt; 5497c478bd9Sstevel@tonic-gate htable_t *list; 5507c478bd9Sstevel@tonic-gate htable_t *ht; 5517c478bd9Sstevel@tonic-gate 5527c478bd9Sstevel@tonic-gate HATSTAT_INC(hs_reap_attempts); 5537c478bd9Sstevel@tonic-gate if (!can_steal_post_boot) 5547c478bd9Sstevel@tonic-gate return; 5557c478bd9Sstevel@tonic-gate 5567c478bd9Sstevel@tonic-gate /* 5577c478bd9Sstevel@tonic-gate * Try to reap 5% of the page tables bounded by a maximum of 5587c478bd9Sstevel@tonic-gate * 5% of physmem and a minimum of 10. 5597c478bd9Sstevel@tonic-gate */ 5607c478bd9Sstevel@tonic-gate reap_cnt = MIN(MAX(physmem / 20, active_ptables / 20), 10); 5617c478bd9Sstevel@tonic-gate 5627c478bd9Sstevel@tonic-gate /* 5637c478bd9Sstevel@tonic-gate * Let htable_steal() do the work, we just call htable_free() 5647c478bd9Sstevel@tonic-gate */ 5657c478bd9Sstevel@tonic-gate list = htable_steal(reap_cnt); 5667c478bd9Sstevel@tonic-gate while ((ht = list) != NULL) { 5677c478bd9Sstevel@tonic-gate list = ht->ht_next; 5687c478bd9Sstevel@tonic-gate HATSTAT_INC(hs_reaped); 5697c478bd9Sstevel@tonic-gate htable_free(ht); 5707c478bd9Sstevel@tonic-gate } 5717c478bd9Sstevel@tonic-gate 5727c478bd9Sstevel@tonic-gate /* 5737c478bd9Sstevel@tonic-gate * Free up excess reserves 5747c478bd9Sstevel@tonic-gate */ 5757c478bd9Sstevel@tonic-gate htable_adjust_reserve(); 5767c478bd9Sstevel@tonic-gate hment_adjust_reserve(); 5777c478bd9Sstevel@tonic-gate } 5787c478bd9Sstevel@tonic-gate 5797c478bd9Sstevel@tonic-gate /* 580ae115bc7Smrj * Allocate an htable, stealing one or using the reserve if necessary 5817c478bd9Sstevel@tonic-gate */ 5827c478bd9Sstevel@tonic-gate static htable_t * 5837c478bd9Sstevel@tonic-gate htable_alloc( 5847c478bd9Sstevel@tonic-gate hat_t *hat, 5857c478bd9Sstevel@tonic-gate uintptr_t vaddr, 5867c478bd9Sstevel@tonic-gate level_t level, 5877c478bd9Sstevel@tonic-gate htable_t *shared) 5887c478bd9Sstevel@tonic-gate { 5897c478bd9Sstevel@tonic-gate htable_t *ht = NULL; 5907c478bd9Sstevel@tonic-gate uint_t is_vlp; 5917c478bd9Sstevel@tonic-gate uint_t is_bare = 0; 5927c478bd9Sstevel@tonic-gate uint_t need_to_zero = 1; 5937c478bd9Sstevel@tonic-gate int kmflags = (can_steal_post_boot ? KM_NOSLEEP : KM_SLEEP); 5947c478bd9Sstevel@tonic-gate 5957c478bd9Sstevel@tonic-gate if (level < 0 || level > TOP_LEVEL(hat)) 5967c478bd9Sstevel@tonic-gate panic("htable_alloc(): level %d out of range\n", level); 5977c478bd9Sstevel@tonic-gate 5987c478bd9Sstevel@tonic-gate is_vlp = (hat->hat_flags & HAT_VLP) && level == VLP_LEVEL; 5997c478bd9Sstevel@tonic-gate if (is_vlp || shared != NULL) 6007c478bd9Sstevel@tonic-gate is_bare = 1; 6017c478bd9Sstevel@tonic-gate 6027c478bd9Sstevel@tonic-gate /* 6037c478bd9Sstevel@tonic-gate * First reuse a cached htable from the hat_ht_cached field, this 604ae115bc7Smrj * avoids unnecessary trips through kmem/page allocators. 6057c478bd9Sstevel@tonic-gate */ 6067c478bd9Sstevel@tonic-gate if (hat->hat_ht_cached != NULL && !is_bare) { 6077c478bd9Sstevel@tonic-gate hat_enter(hat); 6087c478bd9Sstevel@tonic-gate ht = hat->hat_ht_cached; 6097c478bd9Sstevel@tonic-gate if (ht != NULL) { 6107c478bd9Sstevel@tonic-gate hat->hat_ht_cached = ht->ht_next; 6117c478bd9Sstevel@tonic-gate need_to_zero = 0; 6127c478bd9Sstevel@tonic-gate /* XX64 ASSERT() they're all zero somehow */ 6137c478bd9Sstevel@tonic-gate ASSERT(ht->ht_pfn != PFN_INVALID); 6147c478bd9Sstevel@tonic-gate } 6157c478bd9Sstevel@tonic-gate hat_exit(hat); 6167c478bd9Sstevel@tonic-gate } 6177c478bd9Sstevel@tonic-gate 6187c478bd9Sstevel@tonic-gate if (ht == NULL) { 6197c478bd9Sstevel@tonic-gate /* 62097704650Sjosephb * Allocate an htable, possibly refilling the reserves. 6217c478bd9Sstevel@tonic-gate */ 62297704650Sjosephb if (USE_HAT_RESERVES()) { 6237c478bd9Sstevel@tonic-gate ht = htable_get_reserve(); 6247c478bd9Sstevel@tonic-gate } else { 6257c478bd9Sstevel@tonic-gate /* 6267c478bd9Sstevel@tonic-gate * Donate successful htable allocations to the reserve. 6277c478bd9Sstevel@tonic-gate */ 6287c478bd9Sstevel@tonic-gate for (;;) { 6297c478bd9Sstevel@tonic-gate ht = kmem_cache_alloc(htable_cache, kmflags); 6307c478bd9Sstevel@tonic-gate if (ht == NULL) 6317c478bd9Sstevel@tonic-gate break; 6327c478bd9Sstevel@tonic-gate ht->ht_pfn = PFN_INVALID; 63397704650Sjosephb if (USE_HAT_RESERVES() || 6347c478bd9Sstevel@tonic-gate htable_reserve_cnt >= htable_reserve_amount) 6357c478bd9Sstevel@tonic-gate break; 6367c478bd9Sstevel@tonic-gate htable_put_reserve(ht); 6377c478bd9Sstevel@tonic-gate } 6387c478bd9Sstevel@tonic-gate } 6397c478bd9Sstevel@tonic-gate 6407c478bd9Sstevel@tonic-gate /* 6417c478bd9Sstevel@tonic-gate * allocate a page for the hardware page table if needed 6427c478bd9Sstevel@tonic-gate */ 6437c478bd9Sstevel@tonic-gate if (ht != NULL && !is_bare) { 644a85a6733Sjosephb ht->ht_hat = hat; 645ae115bc7Smrj ht->ht_pfn = ptable_alloc((uintptr_t)ht); 6467c478bd9Sstevel@tonic-gate if (ht->ht_pfn == PFN_INVALID) { 64797704650Sjosephb if (USE_HAT_RESERVES()) 64897704650Sjosephb htable_put_reserve(ht); 64997704650Sjosephb else 6507c478bd9Sstevel@tonic-gate kmem_cache_free(htable_cache, ht); 6517c478bd9Sstevel@tonic-gate ht = NULL; 6527c478bd9Sstevel@tonic-gate } 6537c478bd9Sstevel@tonic-gate } 6547c478bd9Sstevel@tonic-gate } 6557c478bd9Sstevel@tonic-gate 6567c478bd9Sstevel@tonic-gate /* 657a85a6733Sjosephb * If allocations failed, kick off a kmem_reap() and resort to 658a85a6733Sjosephb * htable steal(). We may spin here if the system is very low on 659a85a6733Sjosephb * memory. If the kernel itself has consumed all memory and kmem_reap() 660a85a6733Sjosephb * can't free up anything, then we'll really get stuck here. 661a85a6733Sjosephb * That should only happen in a system where the administrator has 662a85a6733Sjosephb * misconfigured VM parameters via /etc/system. 6637c478bd9Sstevel@tonic-gate */ 664a85a6733Sjosephb while (ht == NULL && can_steal_post_boot) { 665a85a6733Sjosephb kmem_reap(); 6667c478bd9Sstevel@tonic-gate ht = htable_steal(1); 6677c478bd9Sstevel@tonic-gate HATSTAT_INC(hs_steals); 6687c478bd9Sstevel@tonic-gate 6697c478bd9Sstevel@tonic-gate /* 670a85a6733Sjosephb * If we stole for a bare htable, release the pagetable page. 6717c478bd9Sstevel@tonic-gate */ 672ae115bc7Smrj if (ht != NULL) { 673ae115bc7Smrj if (is_bare) { 674ae115bc7Smrj ptable_free(ht->ht_pfn); 675ae115bc7Smrj ht->ht_pfn = PFN_INVALID; 676ae115bc7Smrj } 677ae115bc7Smrj } 6787c478bd9Sstevel@tonic-gate } 6797c478bd9Sstevel@tonic-gate 6807c478bd9Sstevel@tonic-gate /* 681a85a6733Sjosephb * All attempts to allocate or steal failed. This should only happen 682a85a6733Sjosephb * if we run out of memory during boot, due perhaps to a huge 683a85a6733Sjosephb * boot_archive. At this point there's no way to continue. 6847c478bd9Sstevel@tonic-gate */ 6857c478bd9Sstevel@tonic-gate if (ht == NULL) 6867c478bd9Sstevel@tonic-gate panic("htable_alloc(): couldn't steal\n"); 6877c478bd9Sstevel@tonic-gate 6887c478bd9Sstevel@tonic-gate /* 6897c478bd9Sstevel@tonic-gate * Shared page tables have all entries locked and entries may not 6907c478bd9Sstevel@tonic-gate * be added or deleted. 6917c478bd9Sstevel@tonic-gate */ 6927c478bd9Sstevel@tonic-gate ht->ht_flags = 0; 6937c478bd9Sstevel@tonic-gate if (shared != NULL) { 6947c478bd9Sstevel@tonic-gate ASSERT(shared->ht_valid_cnt > 0); 6957c478bd9Sstevel@tonic-gate ht->ht_flags |= HTABLE_SHARED_PFN; 6967c478bd9Sstevel@tonic-gate ht->ht_pfn = shared->ht_pfn; 6977c478bd9Sstevel@tonic-gate ht->ht_lock_cnt = 0; 6987c478bd9Sstevel@tonic-gate ht->ht_valid_cnt = 0; /* updated in hat_share() */ 6997c478bd9Sstevel@tonic-gate ht->ht_shares = shared; 7007c478bd9Sstevel@tonic-gate need_to_zero = 0; 7017c478bd9Sstevel@tonic-gate } else { 7027c478bd9Sstevel@tonic-gate ht->ht_shares = NULL; 7037c478bd9Sstevel@tonic-gate ht->ht_lock_cnt = 0; 7047c478bd9Sstevel@tonic-gate ht->ht_valid_cnt = 0; 7057c478bd9Sstevel@tonic-gate } 7067c478bd9Sstevel@tonic-gate 7077c478bd9Sstevel@tonic-gate /* 7087c478bd9Sstevel@tonic-gate * setup flags, etc. for VLP htables 7097c478bd9Sstevel@tonic-gate */ 7107c478bd9Sstevel@tonic-gate if (is_vlp) { 7117c478bd9Sstevel@tonic-gate ht->ht_flags |= HTABLE_VLP; 7127c478bd9Sstevel@tonic-gate ASSERT(ht->ht_pfn == PFN_INVALID); 7137c478bd9Sstevel@tonic-gate need_to_zero = 0; 7147c478bd9Sstevel@tonic-gate } 7157c478bd9Sstevel@tonic-gate 7167c478bd9Sstevel@tonic-gate /* 7177c478bd9Sstevel@tonic-gate * fill in the htable 7187c478bd9Sstevel@tonic-gate */ 7197c478bd9Sstevel@tonic-gate ht->ht_hat = hat; 7207c478bd9Sstevel@tonic-gate ht->ht_parent = NULL; 7217c478bd9Sstevel@tonic-gate ht->ht_vaddr = vaddr; 7227c478bd9Sstevel@tonic-gate ht->ht_level = level; 7237c478bd9Sstevel@tonic-gate ht->ht_busy = 1; 7247c478bd9Sstevel@tonic-gate ht->ht_next = NULL; 7257c478bd9Sstevel@tonic-gate ht->ht_prev = NULL; 7267c478bd9Sstevel@tonic-gate 7277c478bd9Sstevel@tonic-gate /* 7287c478bd9Sstevel@tonic-gate * Zero out any freshly allocated page table 7297c478bd9Sstevel@tonic-gate */ 7307c478bd9Sstevel@tonic-gate if (need_to_zero) 7317c478bd9Sstevel@tonic-gate x86pte_zero(ht, 0, mmu.ptes_per_table); 732ae115bc7Smrj 7337c478bd9Sstevel@tonic-gate return (ht); 7347c478bd9Sstevel@tonic-gate } 7357c478bd9Sstevel@tonic-gate 7367c478bd9Sstevel@tonic-gate /* 7377c478bd9Sstevel@tonic-gate * Free up an htable, either to a hat's cached list, the reserves or 7387c478bd9Sstevel@tonic-gate * back to kmem. 7397c478bd9Sstevel@tonic-gate */ 7407c478bd9Sstevel@tonic-gate static void 7417c478bd9Sstevel@tonic-gate htable_free(htable_t *ht) 7427c478bd9Sstevel@tonic-gate { 7437c478bd9Sstevel@tonic-gate hat_t *hat = ht->ht_hat; 7447c478bd9Sstevel@tonic-gate 7457c478bd9Sstevel@tonic-gate /* 7467c478bd9Sstevel@tonic-gate * If the process isn't exiting, cache the free htable in the hat 7477c478bd9Sstevel@tonic-gate * structure. We always do this for the boot reserve. We don't 7487c478bd9Sstevel@tonic-gate * do this if the hat is exiting or we are stealing/reaping htables. 7497c478bd9Sstevel@tonic-gate */ 7507c478bd9Sstevel@tonic-gate if (hat != NULL && 7517c478bd9Sstevel@tonic-gate !(ht->ht_flags & HTABLE_SHARED_PFN) && 7527c478bd9Sstevel@tonic-gate (use_boot_reserve || 7537c478bd9Sstevel@tonic-gate (!(hat->hat_flags & HAT_FREEING) && !htable_dont_cache))) { 7547c478bd9Sstevel@tonic-gate ASSERT((ht->ht_flags & HTABLE_VLP) == 0); 7557c478bd9Sstevel@tonic-gate ASSERT(ht->ht_pfn != PFN_INVALID); 7567c478bd9Sstevel@tonic-gate hat_enter(hat); 7577c478bd9Sstevel@tonic-gate ht->ht_next = hat->hat_ht_cached; 7587c478bd9Sstevel@tonic-gate hat->hat_ht_cached = ht; 7597c478bd9Sstevel@tonic-gate hat_exit(hat); 7607c478bd9Sstevel@tonic-gate return; 7617c478bd9Sstevel@tonic-gate } 7627c478bd9Sstevel@tonic-gate 7637c478bd9Sstevel@tonic-gate /* 7647c478bd9Sstevel@tonic-gate * If we have a hardware page table, free it. 765ae115bc7Smrj * We don't free page tables that are accessed by sharing. 7667c478bd9Sstevel@tonic-gate */ 7677c478bd9Sstevel@tonic-gate if (ht->ht_flags & HTABLE_SHARED_PFN) { 7687c478bd9Sstevel@tonic-gate ASSERT(ht->ht_pfn != PFN_INVALID); 7697c478bd9Sstevel@tonic-gate } else if (!(ht->ht_flags & HTABLE_VLP)) { 770ae115bc7Smrj ptable_free(ht->ht_pfn); 7717c478bd9Sstevel@tonic-gate } 772ae115bc7Smrj ht->ht_pfn = PFN_INVALID; 7737c478bd9Sstevel@tonic-gate 7747c478bd9Sstevel@tonic-gate /* 77597704650Sjosephb * Free htables or put into reserves. 7767c478bd9Sstevel@tonic-gate */ 777aac11643Sjosephb if (USE_HAT_RESERVES() || htable_reserve_cnt < htable_reserve_amount) { 7787c478bd9Sstevel@tonic-gate htable_put_reserve(ht); 779aac11643Sjosephb } else { 7807c478bd9Sstevel@tonic-gate kmem_cache_free(htable_cache, ht); 781aac11643Sjosephb htable_adjust_reserve(); 782aac11643Sjosephb } 7837c478bd9Sstevel@tonic-gate } 7847c478bd9Sstevel@tonic-gate 7857c478bd9Sstevel@tonic-gate 7867c478bd9Sstevel@tonic-gate /* 7877c478bd9Sstevel@tonic-gate * This is called when a hat is being destroyed or swapped out. We reap all 7887c478bd9Sstevel@tonic-gate * the remaining htables in the hat cache. If destroying all left over 7897c478bd9Sstevel@tonic-gate * htables are also destroyed. 7907c478bd9Sstevel@tonic-gate * 7917c478bd9Sstevel@tonic-gate * We also don't need to invalidate any of the PTPs nor do any demapping. 7927c478bd9Sstevel@tonic-gate */ 7937c478bd9Sstevel@tonic-gate void 7947c478bd9Sstevel@tonic-gate htable_purge_hat(hat_t *hat) 7957c478bd9Sstevel@tonic-gate { 7967c478bd9Sstevel@tonic-gate htable_t *ht; 7977c478bd9Sstevel@tonic-gate int h; 7987c478bd9Sstevel@tonic-gate 7997c478bd9Sstevel@tonic-gate /* 8007c478bd9Sstevel@tonic-gate * Purge the htable cache if just reaping. 8017c478bd9Sstevel@tonic-gate */ 8027c478bd9Sstevel@tonic-gate if (!(hat->hat_flags & HAT_FREEING)) { 8037c478bd9Sstevel@tonic-gate atomic_add_32(&htable_dont_cache, 1); 8047c478bd9Sstevel@tonic-gate for (;;) { 8057c478bd9Sstevel@tonic-gate hat_enter(hat); 8067c478bd9Sstevel@tonic-gate ht = hat->hat_ht_cached; 8077c478bd9Sstevel@tonic-gate if (ht == NULL) { 8087c478bd9Sstevel@tonic-gate hat_exit(hat); 8097c478bd9Sstevel@tonic-gate break; 8107c478bd9Sstevel@tonic-gate } 8117c478bd9Sstevel@tonic-gate hat->hat_ht_cached = ht->ht_next; 8127c478bd9Sstevel@tonic-gate hat_exit(hat); 8137c478bd9Sstevel@tonic-gate htable_free(ht); 8147c478bd9Sstevel@tonic-gate } 8157c478bd9Sstevel@tonic-gate atomic_add_32(&htable_dont_cache, -1); 8167c478bd9Sstevel@tonic-gate return; 8177c478bd9Sstevel@tonic-gate } 8187c478bd9Sstevel@tonic-gate 8197c478bd9Sstevel@tonic-gate /* 8207c478bd9Sstevel@tonic-gate * if freeing, no locking is needed 8217c478bd9Sstevel@tonic-gate */ 8227c478bd9Sstevel@tonic-gate while ((ht = hat->hat_ht_cached) != NULL) { 8237c478bd9Sstevel@tonic-gate hat->hat_ht_cached = ht->ht_next; 8247c478bd9Sstevel@tonic-gate htable_free(ht); 8257c478bd9Sstevel@tonic-gate } 8267c478bd9Sstevel@tonic-gate 8277c478bd9Sstevel@tonic-gate /* 8287c478bd9Sstevel@tonic-gate * walk thru the htable hash table and free all the htables in it. 8297c478bd9Sstevel@tonic-gate */ 8307c478bd9Sstevel@tonic-gate for (h = 0; h < hat->hat_num_hash; ++h) { 8317c478bd9Sstevel@tonic-gate while ((ht = hat->hat_ht_hash[h]) != NULL) { 8327c478bd9Sstevel@tonic-gate if (ht->ht_next) 8337c478bd9Sstevel@tonic-gate ht->ht_next->ht_prev = ht->ht_prev; 8347c478bd9Sstevel@tonic-gate 8357c478bd9Sstevel@tonic-gate if (ht->ht_prev) { 8367c478bd9Sstevel@tonic-gate ht->ht_prev->ht_next = ht->ht_next; 8377c478bd9Sstevel@tonic-gate } else { 8387c478bd9Sstevel@tonic-gate ASSERT(hat->hat_ht_hash[h] == ht); 8397c478bd9Sstevel@tonic-gate hat->hat_ht_hash[h] = ht->ht_next; 8407c478bd9Sstevel@tonic-gate } 8417c478bd9Sstevel@tonic-gate htable_free(ht); 8427c478bd9Sstevel@tonic-gate } 8437c478bd9Sstevel@tonic-gate } 8447c478bd9Sstevel@tonic-gate } 8457c478bd9Sstevel@tonic-gate 8467c478bd9Sstevel@tonic-gate /* 8477c478bd9Sstevel@tonic-gate * Unlink an entry for a table at vaddr and level out of the existing table 8487c478bd9Sstevel@tonic-gate * one level higher. We are always holding the HASH_ENTER() when doing this. 8497c478bd9Sstevel@tonic-gate */ 8507c478bd9Sstevel@tonic-gate static void 8517c478bd9Sstevel@tonic-gate unlink_ptp(htable_t *higher, htable_t *old, uintptr_t vaddr) 8527c478bd9Sstevel@tonic-gate { 8537c478bd9Sstevel@tonic-gate uint_t entry = htable_va2entry(vaddr, higher); 8547c478bd9Sstevel@tonic-gate x86pte_t expect = MAKEPTP(old->ht_pfn, old->ht_level); 8557c478bd9Sstevel@tonic-gate x86pte_t found; 856935f8dd0Sjosephb hat_t *hat = old->ht_hat; 8577c478bd9Sstevel@tonic-gate 8587c478bd9Sstevel@tonic-gate ASSERT(higher->ht_busy > 0); 8597c478bd9Sstevel@tonic-gate ASSERT(higher->ht_valid_cnt > 0); 8607c478bd9Sstevel@tonic-gate ASSERT(old->ht_valid_cnt == 0); 8617c478bd9Sstevel@tonic-gate found = x86pte_cas(higher, entry, expect, 0); 8627c478bd9Sstevel@tonic-gate if (found != expect) 8637c478bd9Sstevel@tonic-gate panic("Bad PTP found=" FMT_PTE ", expected=" FMT_PTE, 8647c478bd9Sstevel@tonic-gate found, expect); 865935f8dd0Sjosephb 866935f8dd0Sjosephb /* 867935f8dd0Sjosephb * When any top level VLP page table entry changes, we must issue 868935f8dd0Sjosephb * a reload of cr3 on all processors. Also some CPU types require 869935f8dd0Sjosephb * invalidating when inner table entries are invalidated. 870935f8dd0Sjosephb */ 871935f8dd0Sjosephb if (!(hat->hat_flags & HAT_FREEING)) { 872935f8dd0Sjosephb if (higher->ht_flags & HTABLE_VLP) 873935f8dd0Sjosephb hat_tlb_inval(hat, DEMAP_ALL_ADDR); 874935f8dd0Sjosephb else if (mmu.inval_nonleaf) 875935f8dd0Sjosephb hat_tlb_inval(hat, old->ht_vaddr); 876935f8dd0Sjosephb } 877935f8dd0Sjosephb 8787c478bd9Sstevel@tonic-gate HTABLE_DEC(higher->ht_valid_cnt); 8797c478bd9Sstevel@tonic-gate } 8807c478bd9Sstevel@tonic-gate 8817c478bd9Sstevel@tonic-gate /* 8827c478bd9Sstevel@tonic-gate * Link an entry for a new table at vaddr and level into the existing table 8837c478bd9Sstevel@tonic-gate * one level higher. We are always holding the HASH_ENTER() when doing this. 8847c478bd9Sstevel@tonic-gate */ 8857c478bd9Sstevel@tonic-gate static void 8867c478bd9Sstevel@tonic-gate link_ptp(htable_t *higher, htable_t *new, uintptr_t vaddr) 8877c478bd9Sstevel@tonic-gate { 8887c478bd9Sstevel@tonic-gate uint_t entry = htable_va2entry(vaddr, higher); 8897c478bd9Sstevel@tonic-gate x86pte_t newptp = MAKEPTP(new->ht_pfn, new->ht_level); 8907c478bd9Sstevel@tonic-gate x86pte_t found; 8917c478bd9Sstevel@tonic-gate 8927c478bd9Sstevel@tonic-gate ASSERT(higher->ht_busy > 0); 8937c478bd9Sstevel@tonic-gate 8947c478bd9Sstevel@tonic-gate ASSERT(new->ht_level != mmu.max_level); 8957c478bd9Sstevel@tonic-gate 8967c478bd9Sstevel@tonic-gate HTABLE_INC(higher->ht_valid_cnt); 8977c478bd9Sstevel@tonic-gate 8987c478bd9Sstevel@tonic-gate found = x86pte_cas(higher, entry, 0, newptp); 899b4b46911Skchow if ((found & ~PT_REF) != 0) 9007c478bd9Sstevel@tonic-gate panic("HAT: ptp not 0, found=" FMT_PTE, found); 901935f8dd0Sjosephb 902935f8dd0Sjosephb /* 903935f8dd0Sjosephb * When any top level VLP page table entry changes, we must issue 904935f8dd0Sjosephb * a reload of cr3 on all processors using it. 9056b60931cSjosephb * We also need to do this for the kernel hat on PAE 32 bit kernel. 906935f8dd0Sjosephb */ 9076b60931cSjosephb if ( 9086b60931cSjosephb #ifdef __i386 9096b60931cSjosephb (higher->ht_hat == kas.a_hat && higher->ht_level == VLP_LEVEL) || 9106b60931cSjosephb #endif 9116b60931cSjosephb (higher->ht_flags & HTABLE_VLP)) 912935f8dd0Sjosephb hat_tlb_inval(higher->ht_hat, DEMAP_ALL_ADDR); 9137c478bd9Sstevel@tonic-gate } 9147c478bd9Sstevel@tonic-gate 9157c478bd9Sstevel@tonic-gate /* 916ae115bc7Smrj * Release of hold on an htable. If this is the last use and the pagetable 917ae115bc7Smrj * is empty we may want to free it, then recursively look at the pagetable 918ae115bc7Smrj * above it. The recursion is handled by the outer while() loop. 9197c478bd9Sstevel@tonic-gate */ 9207c478bd9Sstevel@tonic-gate void 9217c478bd9Sstevel@tonic-gate htable_release(htable_t *ht) 9227c478bd9Sstevel@tonic-gate { 9237c478bd9Sstevel@tonic-gate uint_t hashval; 9247c478bd9Sstevel@tonic-gate htable_t *shared; 9257c478bd9Sstevel@tonic-gate htable_t *higher; 9267c478bd9Sstevel@tonic-gate hat_t *hat; 9277c478bd9Sstevel@tonic-gate uintptr_t va; 9287c478bd9Sstevel@tonic-gate level_t level; 9297c478bd9Sstevel@tonic-gate 9307c478bd9Sstevel@tonic-gate while (ht != NULL) { 9317c478bd9Sstevel@tonic-gate shared = NULL; 9327c478bd9Sstevel@tonic-gate for (;;) { 9337c478bd9Sstevel@tonic-gate hat = ht->ht_hat; 9347c478bd9Sstevel@tonic-gate va = ht->ht_vaddr; 9357c478bd9Sstevel@tonic-gate level = ht->ht_level; 9367c478bd9Sstevel@tonic-gate hashval = HTABLE_HASH(hat, va, level); 9377c478bd9Sstevel@tonic-gate 9387c478bd9Sstevel@tonic-gate /* 9397c478bd9Sstevel@tonic-gate * The common case is that this isn't the last use of 9407c478bd9Sstevel@tonic-gate * an htable so we don't want to free the htable. 9417c478bd9Sstevel@tonic-gate */ 9427c478bd9Sstevel@tonic-gate HTABLE_ENTER(hashval); 9437c478bd9Sstevel@tonic-gate ASSERT(ht->ht_lock_cnt == 0 || ht->ht_valid_cnt > 0); 9447c478bd9Sstevel@tonic-gate ASSERT(ht->ht_valid_cnt >= 0); 9457c478bd9Sstevel@tonic-gate ASSERT(ht->ht_busy > 0); 9467c478bd9Sstevel@tonic-gate if (ht->ht_valid_cnt > 0) 9477c478bd9Sstevel@tonic-gate break; 9487c478bd9Sstevel@tonic-gate if (ht->ht_busy > 1) 9497c478bd9Sstevel@tonic-gate break; 9507c478bd9Sstevel@tonic-gate 9517c478bd9Sstevel@tonic-gate /* 9527c478bd9Sstevel@tonic-gate * we always release empty shared htables 9537c478bd9Sstevel@tonic-gate */ 9547c478bd9Sstevel@tonic-gate if (!(ht->ht_flags & HTABLE_SHARED_PFN)) { 9557c478bd9Sstevel@tonic-gate 9567c478bd9Sstevel@tonic-gate /* 9577c478bd9Sstevel@tonic-gate * don't release if in address space tear down 9587c478bd9Sstevel@tonic-gate */ 9597c478bd9Sstevel@tonic-gate if (hat->hat_flags & HAT_FREEING) 9607c478bd9Sstevel@tonic-gate break; 9617c478bd9Sstevel@tonic-gate 9627c478bd9Sstevel@tonic-gate /* 9637c478bd9Sstevel@tonic-gate * At and above max_page_level, free if it's for 9647c478bd9Sstevel@tonic-gate * a boot-time kernel mapping below kernelbase. 9657c478bd9Sstevel@tonic-gate */ 9667c478bd9Sstevel@tonic-gate if (level >= mmu.max_page_level && 9677c478bd9Sstevel@tonic-gate (hat != kas.a_hat || va >= kernelbase)) 9687c478bd9Sstevel@tonic-gate break; 9697c478bd9Sstevel@tonic-gate } 9707c478bd9Sstevel@tonic-gate 9717c478bd9Sstevel@tonic-gate /* 972ae115bc7Smrj * Remember if we destroy an htable that shares its PFN 973ae115bc7Smrj * from elsewhere. 9747c478bd9Sstevel@tonic-gate */ 9757c478bd9Sstevel@tonic-gate if (ht->ht_flags & HTABLE_SHARED_PFN) { 9767c478bd9Sstevel@tonic-gate ASSERT(shared == NULL); 9777c478bd9Sstevel@tonic-gate shared = ht->ht_shares; 9787c478bd9Sstevel@tonic-gate HATSTAT_INC(hs_htable_unshared); 9797c478bd9Sstevel@tonic-gate } 9807c478bd9Sstevel@tonic-gate 9817c478bd9Sstevel@tonic-gate /* 9827c478bd9Sstevel@tonic-gate * Handle release of a table and freeing the htable_t. 9837c478bd9Sstevel@tonic-gate * Unlink it from the table higher (ie. ht_parent). 9847c478bd9Sstevel@tonic-gate */ 9857c478bd9Sstevel@tonic-gate ASSERT(ht->ht_lock_cnt == 0); 9867c478bd9Sstevel@tonic-gate higher = ht->ht_parent; 9877c478bd9Sstevel@tonic-gate ASSERT(higher != NULL); 9887c478bd9Sstevel@tonic-gate 9897c478bd9Sstevel@tonic-gate /* 9907c478bd9Sstevel@tonic-gate * Unlink the pagetable. 9917c478bd9Sstevel@tonic-gate */ 9927c478bd9Sstevel@tonic-gate unlink_ptp(higher, ht, va); 9937c478bd9Sstevel@tonic-gate 9947c478bd9Sstevel@tonic-gate /* 9957c478bd9Sstevel@tonic-gate * remove this htable from its hash list 9967c478bd9Sstevel@tonic-gate */ 9977c478bd9Sstevel@tonic-gate if (ht->ht_next) 9987c478bd9Sstevel@tonic-gate ht->ht_next->ht_prev = ht->ht_prev; 9997c478bd9Sstevel@tonic-gate 10007c478bd9Sstevel@tonic-gate if (ht->ht_prev) { 10017c478bd9Sstevel@tonic-gate ht->ht_prev->ht_next = ht->ht_next; 10027c478bd9Sstevel@tonic-gate } else { 10037c478bd9Sstevel@tonic-gate ASSERT(hat->hat_ht_hash[hashval] == ht); 10047c478bd9Sstevel@tonic-gate hat->hat_ht_hash[hashval] = ht->ht_next; 10057c478bd9Sstevel@tonic-gate } 10067c478bd9Sstevel@tonic-gate HTABLE_EXIT(hashval); 10077c478bd9Sstevel@tonic-gate htable_free(ht); 10087c478bd9Sstevel@tonic-gate ht = higher; 10097c478bd9Sstevel@tonic-gate } 10107c478bd9Sstevel@tonic-gate 10117c478bd9Sstevel@tonic-gate ASSERT(ht->ht_busy >= 1); 10127c478bd9Sstevel@tonic-gate --ht->ht_busy; 10137c478bd9Sstevel@tonic-gate HTABLE_EXIT(hashval); 10147c478bd9Sstevel@tonic-gate 10157c478bd9Sstevel@tonic-gate /* 10167c478bd9Sstevel@tonic-gate * If we released a shared htable, do a release on the htable 10177c478bd9Sstevel@tonic-gate * from which it shared 10187c478bd9Sstevel@tonic-gate */ 10197c478bd9Sstevel@tonic-gate ht = shared; 10207c478bd9Sstevel@tonic-gate } 10217c478bd9Sstevel@tonic-gate } 10227c478bd9Sstevel@tonic-gate 10237c478bd9Sstevel@tonic-gate /* 10247c478bd9Sstevel@tonic-gate * Find the htable for the pagetable at the given level for the given address. 10257c478bd9Sstevel@tonic-gate * If found acquires a hold that eventually needs to be htable_release()d 10267c478bd9Sstevel@tonic-gate */ 10277c478bd9Sstevel@tonic-gate htable_t * 10287c478bd9Sstevel@tonic-gate htable_lookup(hat_t *hat, uintptr_t vaddr, level_t level) 10297c478bd9Sstevel@tonic-gate { 10307c478bd9Sstevel@tonic-gate uintptr_t base; 10317c478bd9Sstevel@tonic-gate uint_t hashval; 10327c478bd9Sstevel@tonic-gate htable_t *ht = NULL; 10337c478bd9Sstevel@tonic-gate 10347c478bd9Sstevel@tonic-gate ASSERT(level >= 0); 10357c478bd9Sstevel@tonic-gate ASSERT(level <= TOP_LEVEL(hat)); 10367c478bd9Sstevel@tonic-gate 10377c478bd9Sstevel@tonic-gate if (level == TOP_LEVEL(hat)) 10387c478bd9Sstevel@tonic-gate base = 0; 10397c478bd9Sstevel@tonic-gate else 10407c478bd9Sstevel@tonic-gate base = vaddr & LEVEL_MASK(level + 1); 10417c478bd9Sstevel@tonic-gate 10427c478bd9Sstevel@tonic-gate hashval = HTABLE_HASH(hat, base, level); 10437c478bd9Sstevel@tonic-gate HTABLE_ENTER(hashval); 10447c478bd9Sstevel@tonic-gate for (ht = hat->hat_ht_hash[hashval]; ht; ht = ht->ht_next) { 10457c478bd9Sstevel@tonic-gate if (ht->ht_hat == hat && 10467c478bd9Sstevel@tonic-gate ht->ht_vaddr == base && 10477c478bd9Sstevel@tonic-gate ht->ht_level == level) 10487c478bd9Sstevel@tonic-gate break; 10497c478bd9Sstevel@tonic-gate } 10507c478bd9Sstevel@tonic-gate if (ht) 10517c478bd9Sstevel@tonic-gate ++ht->ht_busy; 10527c478bd9Sstevel@tonic-gate 10537c478bd9Sstevel@tonic-gate HTABLE_EXIT(hashval); 10547c478bd9Sstevel@tonic-gate return (ht); 10557c478bd9Sstevel@tonic-gate } 10567c478bd9Sstevel@tonic-gate 10577c478bd9Sstevel@tonic-gate /* 10587c478bd9Sstevel@tonic-gate * Acquires a hold on a known htable (from a locked hment entry). 10597c478bd9Sstevel@tonic-gate */ 10607c478bd9Sstevel@tonic-gate void 10617c478bd9Sstevel@tonic-gate htable_acquire(htable_t *ht) 10627c478bd9Sstevel@tonic-gate { 10637c478bd9Sstevel@tonic-gate hat_t *hat = ht->ht_hat; 10647c478bd9Sstevel@tonic-gate level_t level = ht->ht_level; 10657c478bd9Sstevel@tonic-gate uintptr_t base = ht->ht_vaddr; 10667c478bd9Sstevel@tonic-gate uint_t hashval = HTABLE_HASH(hat, base, level); 10677c478bd9Sstevel@tonic-gate 10687c478bd9Sstevel@tonic-gate HTABLE_ENTER(hashval); 10697c478bd9Sstevel@tonic-gate #ifdef DEBUG 10707c478bd9Sstevel@tonic-gate /* 10717c478bd9Sstevel@tonic-gate * make sure the htable is there 10727c478bd9Sstevel@tonic-gate */ 10737c478bd9Sstevel@tonic-gate { 10747c478bd9Sstevel@tonic-gate htable_t *h; 10757c478bd9Sstevel@tonic-gate 10767c478bd9Sstevel@tonic-gate for (h = hat->hat_ht_hash[hashval]; 10777c478bd9Sstevel@tonic-gate h && h != ht; 10787c478bd9Sstevel@tonic-gate h = h->ht_next) 10797c478bd9Sstevel@tonic-gate ; 10807c478bd9Sstevel@tonic-gate ASSERT(h == ht); 10817c478bd9Sstevel@tonic-gate } 10827c478bd9Sstevel@tonic-gate #endif /* DEBUG */ 10837c478bd9Sstevel@tonic-gate ++ht->ht_busy; 10847c478bd9Sstevel@tonic-gate HTABLE_EXIT(hashval); 10857c478bd9Sstevel@tonic-gate } 10867c478bd9Sstevel@tonic-gate 10877c478bd9Sstevel@tonic-gate /* 10887c478bd9Sstevel@tonic-gate * Find the htable for the pagetable at the given level for the given address. 10897c478bd9Sstevel@tonic-gate * If found acquires a hold that eventually needs to be htable_release()d 10907c478bd9Sstevel@tonic-gate * If not found the table is created. 10917c478bd9Sstevel@tonic-gate * 10927c478bd9Sstevel@tonic-gate * Since we can't hold a hash table mutex during allocation, we have to 10937c478bd9Sstevel@tonic-gate * drop it and redo the search on a create. Then we may have to free the newly 10947c478bd9Sstevel@tonic-gate * allocated htable if another thread raced in and created it ahead of us. 10957c478bd9Sstevel@tonic-gate */ 10967c478bd9Sstevel@tonic-gate htable_t * 10977c478bd9Sstevel@tonic-gate htable_create( 10987c478bd9Sstevel@tonic-gate hat_t *hat, 10997c478bd9Sstevel@tonic-gate uintptr_t vaddr, 11007c478bd9Sstevel@tonic-gate level_t level, 11017c478bd9Sstevel@tonic-gate htable_t *shared) 11027c478bd9Sstevel@tonic-gate { 11037c478bd9Sstevel@tonic-gate uint_t h; 11047c478bd9Sstevel@tonic-gate level_t l; 11057c478bd9Sstevel@tonic-gate uintptr_t base; 11067c478bd9Sstevel@tonic-gate htable_t *ht; 11077c478bd9Sstevel@tonic-gate htable_t *higher = NULL; 11087c478bd9Sstevel@tonic-gate htable_t *new = NULL; 11097c478bd9Sstevel@tonic-gate 11107c478bd9Sstevel@tonic-gate if (level < 0 || level > TOP_LEVEL(hat)) 11117c478bd9Sstevel@tonic-gate panic("htable_create(): level %d out of range\n", level); 11127c478bd9Sstevel@tonic-gate 11137c478bd9Sstevel@tonic-gate /* 11147c478bd9Sstevel@tonic-gate * Create the page tables in top down order. 11157c478bd9Sstevel@tonic-gate */ 11167c478bd9Sstevel@tonic-gate for (l = TOP_LEVEL(hat); l >= level; --l) { 11177c478bd9Sstevel@tonic-gate new = NULL; 11187c478bd9Sstevel@tonic-gate if (l == TOP_LEVEL(hat)) 11197c478bd9Sstevel@tonic-gate base = 0; 11207c478bd9Sstevel@tonic-gate else 11217c478bd9Sstevel@tonic-gate base = vaddr & LEVEL_MASK(l + 1); 11227c478bd9Sstevel@tonic-gate 11237c478bd9Sstevel@tonic-gate h = HTABLE_HASH(hat, base, l); 11247c478bd9Sstevel@tonic-gate try_again: 11257c478bd9Sstevel@tonic-gate /* 11267c478bd9Sstevel@tonic-gate * look up the htable at this level 11277c478bd9Sstevel@tonic-gate */ 11287c478bd9Sstevel@tonic-gate HTABLE_ENTER(h); 11297c478bd9Sstevel@tonic-gate if (l == TOP_LEVEL(hat)) { 11307c478bd9Sstevel@tonic-gate ht = hat->hat_htable; 11317c478bd9Sstevel@tonic-gate } else { 11327c478bd9Sstevel@tonic-gate for (ht = hat->hat_ht_hash[h]; ht; ht = ht->ht_next) { 11337c478bd9Sstevel@tonic-gate ASSERT(ht->ht_hat == hat); 11347c478bd9Sstevel@tonic-gate if (ht->ht_vaddr == base && 11357c478bd9Sstevel@tonic-gate ht->ht_level == l) 11367c478bd9Sstevel@tonic-gate break; 11377c478bd9Sstevel@tonic-gate } 11387c478bd9Sstevel@tonic-gate } 11397c478bd9Sstevel@tonic-gate 11407c478bd9Sstevel@tonic-gate /* 11417c478bd9Sstevel@tonic-gate * if we found the htable, increment its busy cnt 11427c478bd9Sstevel@tonic-gate * and if we had allocated a new htable, free it. 11437c478bd9Sstevel@tonic-gate */ 11447c478bd9Sstevel@tonic-gate if (ht != NULL) { 11457c478bd9Sstevel@tonic-gate /* 11467c478bd9Sstevel@tonic-gate * If we find a pre-existing shared table, it must 11477c478bd9Sstevel@tonic-gate * share from the same place. 11487c478bd9Sstevel@tonic-gate */ 11497c478bd9Sstevel@tonic-gate if (l == level && shared && ht->ht_shares && 11507c478bd9Sstevel@tonic-gate ht->ht_shares != shared) { 11517c478bd9Sstevel@tonic-gate panic("htable shared from wrong place " 11527c478bd9Sstevel@tonic-gate "found htable=%p shared=%p", ht, shared); 11537c478bd9Sstevel@tonic-gate } 11547c478bd9Sstevel@tonic-gate ++ht->ht_busy; 11557c478bd9Sstevel@tonic-gate HTABLE_EXIT(h); 11567c478bd9Sstevel@tonic-gate if (new) 11577c478bd9Sstevel@tonic-gate htable_free(new); 11587c478bd9Sstevel@tonic-gate if (higher != NULL) 11597c478bd9Sstevel@tonic-gate htable_release(higher); 11607c478bd9Sstevel@tonic-gate higher = ht; 11617c478bd9Sstevel@tonic-gate 11627c478bd9Sstevel@tonic-gate /* 11637c478bd9Sstevel@tonic-gate * if we didn't find it on the first search 11647c478bd9Sstevel@tonic-gate * allocate a new one and search again 11657c478bd9Sstevel@tonic-gate */ 11667c478bd9Sstevel@tonic-gate } else if (new == NULL) { 11677c478bd9Sstevel@tonic-gate HTABLE_EXIT(h); 11687c478bd9Sstevel@tonic-gate new = htable_alloc(hat, base, l, 11697c478bd9Sstevel@tonic-gate l == level ? shared : NULL); 11707c478bd9Sstevel@tonic-gate goto try_again; 11717c478bd9Sstevel@tonic-gate 11727c478bd9Sstevel@tonic-gate /* 11737c478bd9Sstevel@tonic-gate * 2nd search and still not there, use "new" table 11747c478bd9Sstevel@tonic-gate * Link new table into higher, when not at top level. 11757c478bd9Sstevel@tonic-gate */ 11767c478bd9Sstevel@tonic-gate } else { 11777c478bd9Sstevel@tonic-gate ht = new; 11787c478bd9Sstevel@tonic-gate if (higher != NULL) { 11797c478bd9Sstevel@tonic-gate link_ptp(higher, ht, base); 11807c478bd9Sstevel@tonic-gate ht->ht_parent = higher; 11817c478bd9Sstevel@tonic-gate } 11827c478bd9Sstevel@tonic-gate ht->ht_next = hat->hat_ht_hash[h]; 11837c478bd9Sstevel@tonic-gate ASSERT(ht->ht_prev == NULL); 11847c478bd9Sstevel@tonic-gate if (hat->hat_ht_hash[h]) 11857c478bd9Sstevel@tonic-gate hat->hat_ht_hash[h]->ht_prev = ht; 11867c478bd9Sstevel@tonic-gate hat->hat_ht_hash[h] = ht; 11877c478bd9Sstevel@tonic-gate HTABLE_EXIT(h); 11887c478bd9Sstevel@tonic-gate 11897c478bd9Sstevel@tonic-gate /* 11907c478bd9Sstevel@tonic-gate * Note we don't do htable_release(higher). 11917c478bd9Sstevel@tonic-gate * That happens recursively when "new" is removed by 11927c478bd9Sstevel@tonic-gate * htable_release() or htable_steal(). 11937c478bd9Sstevel@tonic-gate */ 11947c478bd9Sstevel@tonic-gate higher = ht; 11957c478bd9Sstevel@tonic-gate 11967c478bd9Sstevel@tonic-gate /* 11977c478bd9Sstevel@tonic-gate * If we just created a new shared page table we 11987c478bd9Sstevel@tonic-gate * increment the shared htable's busy count, so that 11997c478bd9Sstevel@tonic-gate * it can't be the victim of a steal even if it's empty. 12007c478bd9Sstevel@tonic-gate */ 12017c478bd9Sstevel@tonic-gate if (l == level && shared) { 12027c478bd9Sstevel@tonic-gate (void) htable_lookup(shared->ht_hat, 12037c478bd9Sstevel@tonic-gate shared->ht_vaddr, shared->ht_level); 12047c478bd9Sstevel@tonic-gate HATSTAT_INC(hs_htable_shared); 12057c478bd9Sstevel@tonic-gate } 12067c478bd9Sstevel@tonic-gate } 12077c478bd9Sstevel@tonic-gate } 12087c478bd9Sstevel@tonic-gate 12097c478bd9Sstevel@tonic-gate return (ht); 12107c478bd9Sstevel@tonic-gate } 12117c478bd9Sstevel@tonic-gate 12127c478bd9Sstevel@tonic-gate /* 1213ae115bc7Smrj * Inherit initial pagetables from the boot program. 1214ae115bc7Smrj */ 1215ae115bc7Smrj void 1216ae115bc7Smrj htable_attach( 1217ae115bc7Smrj hat_t *hat, 1218ae115bc7Smrj uintptr_t base, 1219ae115bc7Smrj level_t level, 1220ae115bc7Smrj htable_t *parent, 1221ae115bc7Smrj pfn_t pfn) 1222ae115bc7Smrj { 1223ae115bc7Smrj htable_t *ht; 1224ae115bc7Smrj uint_t h; 1225ae115bc7Smrj uint_t i; 1226ae115bc7Smrj x86pte_t pte; 1227ae115bc7Smrj x86pte_t *ptep; 1228ae115bc7Smrj page_t *pp; 1229ae115bc7Smrj extern page_t *boot_claim_page(pfn_t); 1230ae115bc7Smrj 1231ae115bc7Smrj ht = htable_get_reserve(); 1232ae115bc7Smrj if (level == mmu.max_level) 1233ae115bc7Smrj kas.a_hat->hat_htable = ht; 1234ae115bc7Smrj ht->ht_hat = hat; 1235ae115bc7Smrj ht->ht_parent = parent; 1236ae115bc7Smrj ht->ht_vaddr = base; 1237ae115bc7Smrj ht->ht_level = level; 1238ae115bc7Smrj ht->ht_busy = 1; 1239ae115bc7Smrj ht->ht_next = NULL; 1240ae115bc7Smrj ht->ht_prev = NULL; 1241ae115bc7Smrj ht->ht_flags = 0; 1242ae115bc7Smrj ht->ht_pfn = pfn; 1243ae115bc7Smrj ht->ht_lock_cnt = 0; 1244ae115bc7Smrj ht->ht_valid_cnt = 0; 1245ae115bc7Smrj if (parent != NULL) 1246ae115bc7Smrj ++parent->ht_busy; 1247ae115bc7Smrj 1248ae115bc7Smrj h = HTABLE_HASH(hat, base, level); 1249ae115bc7Smrj HTABLE_ENTER(h); 1250ae115bc7Smrj ht->ht_next = hat->hat_ht_hash[h]; 1251ae115bc7Smrj ASSERT(ht->ht_prev == NULL); 1252ae115bc7Smrj if (hat->hat_ht_hash[h]) 1253ae115bc7Smrj hat->hat_ht_hash[h]->ht_prev = ht; 1254ae115bc7Smrj hat->hat_ht_hash[h] = ht; 1255ae115bc7Smrj HTABLE_EXIT(h); 1256ae115bc7Smrj 1257ae115bc7Smrj /* 1258ae115bc7Smrj * make sure the page table physical page is not FREE 1259ae115bc7Smrj */ 1260ae115bc7Smrj if (page_resv(1, KM_NOSLEEP) == 0) 1261ae115bc7Smrj panic("page_resv() failed in ptable alloc"); 1262ae115bc7Smrj 1263ae115bc7Smrj pp = boot_claim_page(pfn); 1264ae115bc7Smrj ASSERT(pp != NULL); 1265ae115bc7Smrj page_downgrade(pp); 1266ae115bc7Smrj /* 1267ae115bc7Smrj * Record in the page_t that is a pagetable for segkpm setup. 1268ae115bc7Smrj */ 1269ae115bc7Smrj if (kpm_vbase) 1270ae115bc7Smrj pp->p_index = 1; 1271ae115bc7Smrj 1272ae115bc7Smrj /* 1273ae115bc7Smrj * Count valid mappings and recursively attach lower level pagetables. 1274ae115bc7Smrj */ 1275ae115bc7Smrj ptep = kbm_remap_window(pfn_to_pa(pfn), 0); 1276ae115bc7Smrj for (i = 0; i < HTABLE_NUM_PTES(ht); ++i) { 1277ae115bc7Smrj if (mmu.pae_hat) 1278ae115bc7Smrj pte = ptep[i]; 1279ae115bc7Smrj else 1280ae115bc7Smrj pte = ((x86pte32_t *)ptep)[i]; 1281ae115bc7Smrj if (!IN_HYPERVISOR_VA(base) && PTE_ISVALID(pte)) { 1282ae115bc7Smrj ++ht->ht_valid_cnt; 1283ae115bc7Smrj if (!PTE_ISPAGE(pte, level)) { 1284ae115bc7Smrj htable_attach(hat, base, level - 1, 1285ae115bc7Smrj ht, PTE2PFN(pte, level)); 1286ae115bc7Smrj ptep = kbm_remap_window(pfn_to_pa(pfn), 0); 1287ae115bc7Smrj } 1288ae115bc7Smrj } 1289ae115bc7Smrj base += LEVEL_SIZE(level); 1290ae115bc7Smrj if (base == mmu.hole_start) 1291ae115bc7Smrj base = (mmu.hole_end + MMU_PAGEOFFSET) & MMU_PAGEMASK; 1292ae115bc7Smrj } 1293ae115bc7Smrj 1294ae115bc7Smrj /* 1295ae115bc7Smrj * As long as all the mappings we had were below kernel base 1296ae115bc7Smrj * we can release the htable. 1297ae115bc7Smrj */ 1298ae115bc7Smrj if (base < kernelbase) 1299ae115bc7Smrj htable_release(ht); 1300ae115bc7Smrj } 1301ae115bc7Smrj 1302ae115bc7Smrj /* 13037c478bd9Sstevel@tonic-gate * Walk through a given htable looking for the first valid entry. This 13047c478bd9Sstevel@tonic-gate * routine takes both a starting and ending address. The starting address 13057c478bd9Sstevel@tonic-gate * is required to be within the htable provided by the caller, but there is 13067c478bd9Sstevel@tonic-gate * no such restriction on the ending address. 13077c478bd9Sstevel@tonic-gate * 13087c478bd9Sstevel@tonic-gate * If the routine finds a valid entry in the htable (at or beyond the 13097c478bd9Sstevel@tonic-gate * starting address), the PTE (and its address) will be returned. 13107c478bd9Sstevel@tonic-gate * This PTE may correspond to either a page or a pagetable - it is the 13117c478bd9Sstevel@tonic-gate * caller's responsibility to determine which. If no valid entry is 13127c478bd9Sstevel@tonic-gate * found, 0 (and invalid PTE) and the next unexamined address will be 13137c478bd9Sstevel@tonic-gate * returned. 13147c478bd9Sstevel@tonic-gate * 13157c478bd9Sstevel@tonic-gate * The loop has been carefully coded for optimization. 13167c478bd9Sstevel@tonic-gate */ 13177c478bd9Sstevel@tonic-gate static x86pte_t 13187c478bd9Sstevel@tonic-gate htable_scan(htable_t *ht, uintptr_t *vap, uintptr_t eaddr) 13197c478bd9Sstevel@tonic-gate { 13207c478bd9Sstevel@tonic-gate uint_t e; 13217c478bd9Sstevel@tonic-gate x86pte_t found_pte = (x86pte_t)0; 1322ae115bc7Smrj caddr_t pte_ptr; 1323ae115bc7Smrj caddr_t end_pte_ptr; 13247c478bd9Sstevel@tonic-gate int l = ht->ht_level; 13257c478bd9Sstevel@tonic-gate uintptr_t va = *vap & LEVEL_MASK(l); 13267c478bd9Sstevel@tonic-gate size_t pgsize = LEVEL_SIZE(l); 13277c478bd9Sstevel@tonic-gate 13287c478bd9Sstevel@tonic-gate ASSERT(va >= ht->ht_vaddr); 13297c478bd9Sstevel@tonic-gate ASSERT(va <= HTABLE_LAST_PAGE(ht)); 13307c478bd9Sstevel@tonic-gate 13317c478bd9Sstevel@tonic-gate /* 13327c478bd9Sstevel@tonic-gate * Compute the starting index and ending virtual address 13337c478bd9Sstevel@tonic-gate */ 13347c478bd9Sstevel@tonic-gate e = htable_va2entry(va, ht); 13357c478bd9Sstevel@tonic-gate 13367c478bd9Sstevel@tonic-gate /* 13377c478bd9Sstevel@tonic-gate * The following page table scan code knows that the valid 13387c478bd9Sstevel@tonic-gate * bit of a PTE is in the lowest byte AND that x86 is little endian!! 13397c478bd9Sstevel@tonic-gate */ 1340ae115bc7Smrj pte_ptr = (caddr_t)x86pte_access_pagetable(ht, 0); 1341ae115bc7Smrj end_pte_ptr = (caddr_t)PT_INDEX_PTR(pte_ptr, HTABLE_NUM_PTES(ht)); 1342ae115bc7Smrj pte_ptr = (caddr_t)PT_INDEX_PTR((x86pte_t *)pte_ptr, e); 134330f7a194Skchow while (!PTE_ISVALID(*pte_ptr)) { 13447c478bd9Sstevel@tonic-gate va += pgsize; 13457c478bd9Sstevel@tonic-gate if (va >= eaddr) 13467c478bd9Sstevel@tonic-gate break; 13477c478bd9Sstevel@tonic-gate pte_ptr += mmu.pte_size; 13487c478bd9Sstevel@tonic-gate ASSERT(pte_ptr <= end_pte_ptr); 13497c478bd9Sstevel@tonic-gate if (pte_ptr == end_pte_ptr) 13507c478bd9Sstevel@tonic-gate break; 13517c478bd9Sstevel@tonic-gate } 13527c478bd9Sstevel@tonic-gate 13537c478bd9Sstevel@tonic-gate /* 13547c478bd9Sstevel@tonic-gate * if we found a valid PTE, load the entire PTE 13557c478bd9Sstevel@tonic-gate */ 1356ae115bc7Smrj if (va < eaddr && pte_ptr != end_pte_ptr) 1357ae115bc7Smrj found_pte = GET_PTE((x86pte_t *)pte_ptr); 13587c478bd9Sstevel@tonic-gate x86pte_release_pagetable(ht); 13597c478bd9Sstevel@tonic-gate 13607c478bd9Sstevel@tonic-gate #if defined(__amd64) 13617c478bd9Sstevel@tonic-gate /* 13627c478bd9Sstevel@tonic-gate * deal with VA hole on amd64 13637c478bd9Sstevel@tonic-gate */ 13647c478bd9Sstevel@tonic-gate if (l == mmu.max_level && va >= mmu.hole_start && va <= mmu.hole_end) 13657c478bd9Sstevel@tonic-gate va = mmu.hole_end + va - mmu.hole_start; 13667c478bd9Sstevel@tonic-gate #endif /* __amd64 */ 13677c478bd9Sstevel@tonic-gate 13687c478bd9Sstevel@tonic-gate *vap = va; 13697c478bd9Sstevel@tonic-gate return (found_pte); 13707c478bd9Sstevel@tonic-gate } 13717c478bd9Sstevel@tonic-gate 13727c478bd9Sstevel@tonic-gate /* 13737c478bd9Sstevel@tonic-gate * Find the address and htable for the first populated translation at or 13747c478bd9Sstevel@tonic-gate * above the given virtual address. The caller may also specify an upper 13757c478bd9Sstevel@tonic-gate * limit to the address range to search. Uses level information to quickly 13767c478bd9Sstevel@tonic-gate * skip unpopulated sections of virtual address spaces. 13777c478bd9Sstevel@tonic-gate * 13787c478bd9Sstevel@tonic-gate * If not found returns NULL. When found, returns the htable and virt addr 13797c478bd9Sstevel@tonic-gate * and has a hold on the htable. 13807c478bd9Sstevel@tonic-gate */ 13817c478bd9Sstevel@tonic-gate x86pte_t 13827c478bd9Sstevel@tonic-gate htable_walk( 13837c478bd9Sstevel@tonic-gate struct hat *hat, 13847c478bd9Sstevel@tonic-gate htable_t **htp, 13857c478bd9Sstevel@tonic-gate uintptr_t *vaddr, 13867c478bd9Sstevel@tonic-gate uintptr_t eaddr) 13877c478bd9Sstevel@tonic-gate { 13887c478bd9Sstevel@tonic-gate uintptr_t va = *vaddr; 13897c478bd9Sstevel@tonic-gate htable_t *ht; 13907c478bd9Sstevel@tonic-gate htable_t *prev = *htp; 13917c478bd9Sstevel@tonic-gate level_t l; 13927c478bd9Sstevel@tonic-gate level_t max_mapped_level; 13937c478bd9Sstevel@tonic-gate x86pte_t pte; 13947c478bd9Sstevel@tonic-gate 13957c478bd9Sstevel@tonic-gate ASSERT(eaddr > va); 13967c478bd9Sstevel@tonic-gate 13977c478bd9Sstevel@tonic-gate /* 13987c478bd9Sstevel@tonic-gate * If this is a user address, then we know we need not look beyond 13997c478bd9Sstevel@tonic-gate * kernelbase. 14007c478bd9Sstevel@tonic-gate */ 14017c478bd9Sstevel@tonic-gate ASSERT(hat == kas.a_hat || eaddr <= kernelbase || 14027c478bd9Sstevel@tonic-gate eaddr == HTABLE_WALK_TO_END); 14037c478bd9Sstevel@tonic-gate if (hat != kas.a_hat && eaddr == HTABLE_WALK_TO_END) 14047c478bd9Sstevel@tonic-gate eaddr = kernelbase; 14057c478bd9Sstevel@tonic-gate 14067c478bd9Sstevel@tonic-gate /* 14077c478bd9Sstevel@tonic-gate * If we're coming in with a previous page table, search it first 14087c478bd9Sstevel@tonic-gate * without doing an htable_lookup(), this should be frequent. 14097c478bd9Sstevel@tonic-gate */ 14107c478bd9Sstevel@tonic-gate if (prev) { 14117c478bd9Sstevel@tonic-gate ASSERT(prev->ht_busy > 0); 14127c478bd9Sstevel@tonic-gate ASSERT(prev->ht_vaddr <= va); 14137c478bd9Sstevel@tonic-gate l = prev->ht_level; 14147c478bd9Sstevel@tonic-gate if (va <= HTABLE_LAST_PAGE(prev)) { 14157c478bd9Sstevel@tonic-gate pte = htable_scan(prev, &va, eaddr); 14167c478bd9Sstevel@tonic-gate 14177c478bd9Sstevel@tonic-gate if (PTE_ISPAGE(pte, l)) { 14187c478bd9Sstevel@tonic-gate *vaddr = va; 14197c478bd9Sstevel@tonic-gate *htp = prev; 14207c478bd9Sstevel@tonic-gate return (pte); 14217c478bd9Sstevel@tonic-gate } 14227c478bd9Sstevel@tonic-gate } 14237c478bd9Sstevel@tonic-gate 14247c478bd9Sstevel@tonic-gate /* 14257c478bd9Sstevel@tonic-gate * We found nothing in the htable provided by the caller, 14267c478bd9Sstevel@tonic-gate * so fall through and do the full search 14277c478bd9Sstevel@tonic-gate */ 14287c478bd9Sstevel@tonic-gate htable_release(prev); 14297c478bd9Sstevel@tonic-gate } 14307c478bd9Sstevel@tonic-gate 14317c478bd9Sstevel@tonic-gate /* 14327c478bd9Sstevel@tonic-gate * Find the level of the largest pagesize used by this HAT. 14337c478bd9Sstevel@tonic-gate */ 14347c478bd9Sstevel@tonic-gate max_mapped_level = 0; 14357c478bd9Sstevel@tonic-gate for (l = 1; l <= mmu.max_page_level; ++l) 14367c478bd9Sstevel@tonic-gate if (hat->hat_pages_mapped[l] != 0) 14377c478bd9Sstevel@tonic-gate max_mapped_level = l; 14387c478bd9Sstevel@tonic-gate 14397c478bd9Sstevel@tonic-gate while (va < eaddr && va >= *vaddr) { 14407c478bd9Sstevel@tonic-gate ASSERT(!IN_VA_HOLE(va)); 14417c478bd9Sstevel@tonic-gate 14427c478bd9Sstevel@tonic-gate /* 14437c478bd9Sstevel@tonic-gate * Find lowest table with any entry for given address. 14447c478bd9Sstevel@tonic-gate */ 14457c478bd9Sstevel@tonic-gate for (l = 0; l <= TOP_LEVEL(hat); ++l) { 14467c478bd9Sstevel@tonic-gate ht = htable_lookup(hat, va, l); 14477c478bd9Sstevel@tonic-gate if (ht != NULL) { 14487c478bd9Sstevel@tonic-gate pte = htable_scan(ht, &va, eaddr); 14497c478bd9Sstevel@tonic-gate if (PTE_ISPAGE(pte, l)) { 14507c478bd9Sstevel@tonic-gate *vaddr = va; 14517c478bd9Sstevel@tonic-gate *htp = ht; 14527c478bd9Sstevel@tonic-gate return (pte); 14537c478bd9Sstevel@tonic-gate } 14547c478bd9Sstevel@tonic-gate htable_release(ht); 14557c478bd9Sstevel@tonic-gate break; 14567c478bd9Sstevel@tonic-gate } 14577c478bd9Sstevel@tonic-gate 14587c478bd9Sstevel@tonic-gate /* 1459*8b5842f9Sdm120769 * The ht is never NULL at the top level since 1460*8b5842f9Sdm120769 * the top level htable is created in hat_alloc(). 1461*8b5842f9Sdm120769 */ 1462*8b5842f9Sdm120769 ASSERT(l < TOP_LEVEL(hat)); 1463*8b5842f9Sdm120769 1464*8b5842f9Sdm120769 /* 14657c478bd9Sstevel@tonic-gate * No htable covers the address. If there is no 14667c478bd9Sstevel@tonic-gate * larger page size that could cover it, we 14677c478bd9Sstevel@tonic-gate * skip to the start of the next page table. 14687c478bd9Sstevel@tonic-gate */ 1469*8b5842f9Sdm120769 if (l >= max_mapped_level) { 14707c478bd9Sstevel@tonic-gate va = NEXT_ENTRY_VA(va, l + 1); 1471*8b5842f9Sdm120769 break; 1472*8b5842f9Sdm120769 } 14737c478bd9Sstevel@tonic-gate } 14747c478bd9Sstevel@tonic-gate } 14757c478bd9Sstevel@tonic-gate 14767c478bd9Sstevel@tonic-gate *vaddr = 0; 14777c478bd9Sstevel@tonic-gate *htp = NULL; 14787c478bd9Sstevel@tonic-gate return (0); 14797c478bd9Sstevel@tonic-gate } 14807c478bd9Sstevel@tonic-gate 14817c478bd9Sstevel@tonic-gate /* 14827c478bd9Sstevel@tonic-gate * Find the htable and page table entry index of the given virtual address 14837c478bd9Sstevel@tonic-gate * with pagesize at or below given level. 14847c478bd9Sstevel@tonic-gate * If not found returns NULL. When found, returns the htable, sets 14857c478bd9Sstevel@tonic-gate * entry, and has a hold on the htable. 14867c478bd9Sstevel@tonic-gate */ 14877c478bd9Sstevel@tonic-gate htable_t * 14887c478bd9Sstevel@tonic-gate htable_getpte( 14897c478bd9Sstevel@tonic-gate struct hat *hat, 14907c478bd9Sstevel@tonic-gate uintptr_t vaddr, 14917c478bd9Sstevel@tonic-gate uint_t *entry, 14927c478bd9Sstevel@tonic-gate x86pte_t *pte, 14937c478bd9Sstevel@tonic-gate level_t level) 14947c478bd9Sstevel@tonic-gate { 14957c478bd9Sstevel@tonic-gate htable_t *ht; 14967c478bd9Sstevel@tonic-gate level_t l; 14977c478bd9Sstevel@tonic-gate uint_t e; 14987c478bd9Sstevel@tonic-gate 14997c478bd9Sstevel@tonic-gate ASSERT(level <= mmu.max_page_level); 15007c478bd9Sstevel@tonic-gate 15017c478bd9Sstevel@tonic-gate for (l = 0; l <= level; ++l) { 15027c478bd9Sstevel@tonic-gate ht = htable_lookup(hat, vaddr, l); 15037c478bd9Sstevel@tonic-gate if (ht == NULL) 15047c478bd9Sstevel@tonic-gate continue; 15057c478bd9Sstevel@tonic-gate e = htable_va2entry(vaddr, ht); 15067c478bd9Sstevel@tonic-gate if (entry != NULL) 15077c478bd9Sstevel@tonic-gate *entry = e; 15087c478bd9Sstevel@tonic-gate if (pte != NULL) 15097c478bd9Sstevel@tonic-gate *pte = x86pte_get(ht, e); 15107c478bd9Sstevel@tonic-gate return (ht); 15117c478bd9Sstevel@tonic-gate } 15127c478bd9Sstevel@tonic-gate return (NULL); 15137c478bd9Sstevel@tonic-gate } 15147c478bd9Sstevel@tonic-gate 15157c478bd9Sstevel@tonic-gate /* 15167c478bd9Sstevel@tonic-gate * Find the htable and page table entry index of the given virtual address. 15177c478bd9Sstevel@tonic-gate * There must be a valid page mapped at the given address. 15187c478bd9Sstevel@tonic-gate * If not found returns NULL. When found, returns the htable, sets 15197c478bd9Sstevel@tonic-gate * entry, and has a hold on the htable. 15207c478bd9Sstevel@tonic-gate */ 15217c478bd9Sstevel@tonic-gate htable_t * 15227c478bd9Sstevel@tonic-gate htable_getpage(struct hat *hat, uintptr_t vaddr, uint_t *entry) 15237c478bd9Sstevel@tonic-gate { 15247c478bd9Sstevel@tonic-gate htable_t *ht; 15257c478bd9Sstevel@tonic-gate uint_t e; 15267c478bd9Sstevel@tonic-gate x86pte_t pte; 15277c478bd9Sstevel@tonic-gate 15287c478bd9Sstevel@tonic-gate ht = htable_getpte(hat, vaddr, &e, &pte, mmu.max_page_level); 15297c478bd9Sstevel@tonic-gate if (ht == NULL) 15307c478bd9Sstevel@tonic-gate return (NULL); 15317c478bd9Sstevel@tonic-gate 15327c478bd9Sstevel@tonic-gate if (entry) 15337c478bd9Sstevel@tonic-gate *entry = e; 15347c478bd9Sstevel@tonic-gate 15357c478bd9Sstevel@tonic-gate if (PTE_ISPAGE(pte, ht->ht_level)) 15367c478bd9Sstevel@tonic-gate return (ht); 15377c478bd9Sstevel@tonic-gate htable_release(ht); 15387c478bd9Sstevel@tonic-gate return (NULL); 15397c478bd9Sstevel@tonic-gate } 15407c478bd9Sstevel@tonic-gate 15417c478bd9Sstevel@tonic-gate 15427c478bd9Sstevel@tonic-gate void 15437c478bd9Sstevel@tonic-gate htable_init() 15447c478bd9Sstevel@tonic-gate { 15457c478bd9Sstevel@tonic-gate /* 15467c478bd9Sstevel@tonic-gate * To save on kernel VA usage, we avoid debug information in 32 bit 15477c478bd9Sstevel@tonic-gate * kernels. 15487c478bd9Sstevel@tonic-gate */ 15497c478bd9Sstevel@tonic-gate #if defined(__amd64) 15507c478bd9Sstevel@tonic-gate int kmem_flags = KMC_NOHASH; 15517c478bd9Sstevel@tonic-gate #elif defined(__i386) 15527c478bd9Sstevel@tonic-gate int kmem_flags = KMC_NOHASH | KMC_NODEBUG; 15537c478bd9Sstevel@tonic-gate #endif 15547c478bd9Sstevel@tonic-gate 15557c478bd9Sstevel@tonic-gate /* 15567c478bd9Sstevel@tonic-gate * initialize kmem caches 15577c478bd9Sstevel@tonic-gate */ 15587c478bd9Sstevel@tonic-gate htable_cache = kmem_cache_create("htable_t", 15597c478bd9Sstevel@tonic-gate sizeof (htable_t), 0, NULL, NULL, 15607c478bd9Sstevel@tonic-gate htable_reap, NULL, hat_memload_arena, kmem_flags); 15617c478bd9Sstevel@tonic-gate } 15627c478bd9Sstevel@tonic-gate 15637c478bd9Sstevel@tonic-gate /* 15647c478bd9Sstevel@tonic-gate * get the pte index for the virtual address in the given htable's pagetable 15657c478bd9Sstevel@tonic-gate */ 15667c478bd9Sstevel@tonic-gate uint_t 15677c478bd9Sstevel@tonic-gate htable_va2entry(uintptr_t va, htable_t *ht) 15687c478bd9Sstevel@tonic-gate { 15697c478bd9Sstevel@tonic-gate level_t l = ht->ht_level; 15707c478bd9Sstevel@tonic-gate 15717c478bd9Sstevel@tonic-gate ASSERT(va >= ht->ht_vaddr); 15727c478bd9Sstevel@tonic-gate ASSERT(va <= HTABLE_LAST_PAGE(ht)); 1573ae115bc7Smrj return ((va >> LEVEL_SHIFT(l)) & (HTABLE_NUM_PTES(ht) - 1)); 15747c478bd9Sstevel@tonic-gate } 15757c478bd9Sstevel@tonic-gate 15767c478bd9Sstevel@tonic-gate /* 15777c478bd9Sstevel@tonic-gate * Given an htable and the index of a pte in it, return the virtual address 15787c478bd9Sstevel@tonic-gate * of the page. 15797c478bd9Sstevel@tonic-gate */ 15807c478bd9Sstevel@tonic-gate uintptr_t 15817c478bd9Sstevel@tonic-gate htable_e2va(htable_t *ht, uint_t entry) 15827c478bd9Sstevel@tonic-gate { 15837c478bd9Sstevel@tonic-gate level_t l = ht->ht_level; 15847c478bd9Sstevel@tonic-gate uintptr_t va; 15857c478bd9Sstevel@tonic-gate 1586ae115bc7Smrj ASSERT(entry < HTABLE_NUM_PTES(ht)); 15877c478bd9Sstevel@tonic-gate va = ht->ht_vaddr + ((uintptr_t)entry << LEVEL_SHIFT(l)); 15887c478bd9Sstevel@tonic-gate 15897c478bd9Sstevel@tonic-gate /* 15907c478bd9Sstevel@tonic-gate * Need to skip over any VA hole in top level table 15917c478bd9Sstevel@tonic-gate */ 15927c478bd9Sstevel@tonic-gate #if defined(__amd64) 15937c478bd9Sstevel@tonic-gate if (ht->ht_level == mmu.max_level && va >= mmu.hole_start) 15947c478bd9Sstevel@tonic-gate va += ((mmu.hole_end - mmu.hole_start) + 1); 15957c478bd9Sstevel@tonic-gate #endif 15967c478bd9Sstevel@tonic-gate 15977c478bd9Sstevel@tonic-gate return (va); 15987c478bd9Sstevel@tonic-gate } 15997c478bd9Sstevel@tonic-gate 16007c478bd9Sstevel@tonic-gate /* 16017c478bd9Sstevel@tonic-gate * The code uses compare and swap instructions to read/write PTE's to 16027c478bd9Sstevel@tonic-gate * avoid atomicity problems, since PTEs can be 8 bytes on 32 bit systems. 16037c478bd9Sstevel@tonic-gate * will naturally be atomic. 16047c478bd9Sstevel@tonic-gate * 16057c478bd9Sstevel@tonic-gate * The combination of using kpreempt_disable()/_enable() and the hci_mutex 16067c478bd9Sstevel@tonic-gate * are used to ensure that an interrupt won't overwrite a temporary mapping 16077c478bd9Sstevel@tonic-gate * while it's in use. If an interrupt thread tries to access a PTE, it will 16087c478bd9Sstevel@tonic-gate * yield briefly back to the pinned thread which holds the cpu's hci_mutex. 16097c478bd9Sstevel@tonic-gate */ 16107c478bd9Sstevel@tonic-gate void 1611ae115bc7Smrj x86pte_cpu_init(cpu_t *cpu) 16127c478bd9Sstevel@tonic-gate { 16137c478bd9Sstevel@tonic-gate struct hat_cpu_info *hci; 16147c478bd9Sstevel@tonic-gate 1615ae115bc7Smrj hci = kmem_zalloc(sizeof (*hci), KM_SLEEP); 16167c478bd9Sstevel@tonic-gate mutex_init(&hci->hci_mutex, NULL, MUTEX_DEFAULT, NULL); 16177c478bd9Sstevel@tonic-gate cpu->cpu_hat_info = hci; 16187c478bd9Sstevel@tonic-gate } 16197c478bd9Sstevel@tonic-gate 1620ae115bc7Smrj void 1621ae115bc7Smrj x86pte_cpu_fini(cpu_t *cpu) 1622ae115bc7Smrj { 1623ae115bc7Smrj struct hat_cpu_info *hci = cpu->cpu_hat_info; 1624ae115bc7Smrj 1625ae115bc7Smrj kmem_free(hci, sizeof (*hci)); 1626ae115bc7Smrj cpu->cpu_hat_info = NULL; 16277c478bd9Sstevel@tonic-gate } 16287c478bd9Sstevel@tonic-gate 1629ae115bc7Smrj #ifdef __i386 1630ae115bc7Smrj /* 1631ae115bc7Smrj * On 32 bit kernels, loading a 64 bit PTE is a little tricky 1632ae115bc7Smrj */ 1633ae115bc7Smrj x86pte_t 1634ae115bc7Smrj get_pte64(x86pte_t *ptr) 1635ae115bc7Smrj { 1636ae115bc7Smrj volatile uint32_t *p = (uint32_t *)ptr; 1637ae115bc7Smrj x86pte_t t; 1638ae115bc7Smrj 1639ae115bc7Smrj ASSERT(mmu.pae_hat != 0); 1640ae115bc7Smrj for (;;) { 1641ae115bc7Smrj t = p[0]; 1642ae115bc7Smrj t |= (uint64_t)p[1] << 32; 1643ae115bc7Smrj if ((t & 0xffffffff) == p[0]) 1644ae115bc7Smrj return (t); 1645ae115bc7Smrj } 1646ae115bc7Smrj } 1647ae115bc7Smrj #endif /* __i386 */ 1648ae115bc7Smrj 16497c478bd9Sstevel@tonic-gate /* 16507c478bd9Sstevel@tonic-gate * Disable preemption and establish a mapping to the pagetable with the 16517c478bd9Sstevel@tonic-gate * given pfn. This is optimized for there case where it's the same 16527c478bd9Sstevel@tonic-gate * pfn as we last used referenced from this CPU. 16537c478bd9Sstevel@tonic-gate */ 16547c478bd9Sstevel@tonic-gate static x86pte_t * 1655ae115bc7Smrj x86pte_access_pagetable(htable_t *ht, uint_t index) 16567c478bd9Sstevel@tonic-gate { 16577c478bd9Sstevel@tonic-gate /* 16587c478bd9Sstevel@tonic-gate * VLP pagetables are contained in the hat_t 16597c478bd9Sstevel@tonic-gate */ 16607c478bd9Sstevel@tonic-gate if (ht->ht_flags & HTABLE_VLP) 1661ae115bc7Smrj return (PT_INDEX_PTR(ht->ht_hat->hat_vlp_ptes, index)); 1662ae115bc7Smrj return (x86pte_mapin(ht->ht_pfn, index, ht)); 1663ae115bc7Smrj } 16647c478bd9Sstevel@tonic-gate 16657c478bd9Sstevel@tonic-gate /* 1666ae115bc7Smrj * map the given pfn into the page table window. 16677c478bd9Sstevel@tonic-gate */ 1668ae115bc7Smrj /*ARGSUSED*/ 1669ae115bc7Smrj x86pte_t * 1670ae115bc7Smrj x86pte_mapin(pfn_t pfn, uint_t index, htable_t *ht) 1671ae115bc7Smrj { 1672ae115bc7Smrj x86pte_t *pteptr; 1673ae115bc7Smrj x86pte_t pte; 1674ae115bc7Smrj x86pte_t newpte; 1675ae115bc7Smrj int x; 1676ae115bc7Smrj 16777c478bd9Sstevel@tonic-gate ASSERT(pfn != PFN_INVALID); 16787c478bd9Sstevel@tonic-gate 16797c478bd9Sstevel@tonic-gate if (!khat_running) { 1680ae115bc7Smrj caddr_t va = kbm_remap_window(pfn_to_pa(pfn), 1); 1681ae115bc7Smrj return (PT_INDEX_PTR(va, index)); 16827c478bd9Sstevel@tonic-gate } 16837c478bd9Sstevel@tonic-gate 16847c478bd9Sstevel@tonic-gate /* 1685ae115bc7Smrj * If kpm is available, use it. 1686ae115bc7Smrj */ 1687ae115bc7Smrj if (kpm_vbase) 1688ae115bc7Smrj return (PT_INDEX_PTR(hat_kpm_pfn2va(pfn), index)); 1689ae115bc7Smrj 1690ae115bc7Smrj /* 1691ae115bc7Smrj * Disable preemption and grab the CPU's hci_mutex 16927c478bd9Sstevel@tonic-gate */ 16937c478bd9Sstevel@tonic-gate kpreempt_disable(); 1694ae115bc7Smrj ASSERT(CPU->cpu_hat_info != NULL); 1695ae115bc7Smrj mutex_enter(&CPU->cpu_hat_info->hci_mutex); 1696ae115bc7Smrj x = PWIN_TABLE(CPU->cpu_id); 1697ae115bc7Smrj pteptr = (x86pte_t *)PWIN_PTE_VA(x); 1698ae115bc7Smrj if (mmu.pae_hat) 1699ae115bc7Smrj pte = *pteptr; 1700ae115bc7Smrj else 1701ae115bc7Smrj pte = *(x86pte32_t *)pteptr; 1702ae115bc7Smrj 1703ae115bc7Smrj newpte = MAKEPTE(pfn, 0) | mmu.pt_global | mmu.pt_nx; 1704ae115bc7Smrj newpte |= PT_WRITABLE; 1705ae115bc7Smrj 1706ae115bc7Smrj if (!PTE_EQUIV(newpte, pte)) { 1707ae115bc7Smrj if (mmu.pae_hat) 1708ae115bc7Smrj *pteptr = newpte; 1709ae115bc7Smrj else 1710ae115bc7Smrj *(x86pte32_t *)pteptr = newpte; 1711ae115bc7Smrj mmu_tlbflush_entry((caddr_t)(PWIN_VA(x))); 17127c478bd9Sstevel@tonic-gate } 1713ae115bc7Smrj return (PT_INDEX_PTR(PWIN_VA(x), index)); 17147c478bd9Sstevel@tonic-gate } 17157c478bd9Sstevel@tonic-gate 17167c478bd9Sstevel@tonic-gate /* 17177c478bd9Sstevel@tonic-gate * Release access to a page table. 17187c478bd9Sstevel@tonic-gate */ 17197c478bd9Sstevel@tonic-gate static void 17207c478bd9Sstevel@tonic-gate x86pte_release_pagetable(htable_t *ht) 17217c478bd9Sstevel@tonic-gate { 17227c478bd9Sstevel@tonic-gate /* 17237c478bd9Sstevel@tonic-gate * nothing to do for VLP htables 17247c478bd9Sstevel@tonic-gate */ 17257c478bd9Sstevel@tonic-gate if (ht->ht_flags & HTABLE_VLP) 17267c478bd9Sstevel@tonic-gate return; 17277c478bd9Sstevel@tonic-gate 1728ae115bc7Smrj x86pte_mapout(); 17297c478bd9Sstevel@tonic-gate } 17307c478bd9Sstevel@tonic-gate 1731ae115bc7Smrj void 1732ae115bc7Smrj x86pte_mapout(void) 1733ae115bc7Smrj { 1734ae115bc7Smrj if (mmu.pwin_base == NULL || !khat_running) 1735ae115bc7Smrj return; 1736ae115bc7Smrj 17377c478bd9Sstevel@tonic-gate /* 1738ae115bc7Smrj * Drop the CPU's hci_mutex and restore preemption. 17397c478bd9Sstevel@tonic-gate */ 1740ae115bc7Smrj mutex_exit(&CPU->cpu_hat_info->hci_mutex); 17417c478bd9Sstevel@tonic-gate kpreempt_enable(); 17427c478bd9Sstevel@tonic-gate } 17437c478bd9Sstevel@tonic-gate 17447c478bd9Sstevel@tonic-gate /* 17457c478bd9Sstevel@tonic-gate * Atomic retrieval of a pagetable entry 17467c478bd9Sstevel@tonic-gate */ 17477c478bd9Sstevel@tonic-gate x86pte_t 17487c478bd9Sstevel@tonic-gate x86pte_get(htable_t *ht, uint_t entry) 17497c478bd9Sstevel@tonic-gate { 17507c478bd9Sstevel@tonic-gate x86pte_t pte; 1751aa2ed9e5Sjosephb x86pte_t *ptep; 17527c478bd9Sstevel@tonic-gate 17537c478bd9Sstevel@tonic-gate /* 1754aa2ed9e5Sjosephb * Be careful that loading PAE entries in 32 bit kernel is atomic. 17557c478bd9Sstevel@tonic-gate */ 1756ae115bc7Smrj ASSERT(entry < mmu.ptes_per_table); 1757ae115bc7Smrj ptep = x86pte_access_pagetable(ht, entry); 1758ae115bc7Smrj pte = GET_PTE(ptep); 17597c478bd9Sstevel@tonic-gate x86pte_release_pagetable(ht); 17607c478bd9Sstevel@tonic-gate return (pte); 17617c478bd9Sstevel@tonic-gate } 17627c478bd9Sstevel@tonic-gate 17637c478bd9Sstevel@tonic-gate /* 17647c478bd9Sstevel@tonic-gate * Atomic unconditional set of a page table entry, it returns the previous 1765ae115bc7Smrj * value. For pre-existing mappings if the PFN changes, then we don't care 1766ae115bc7Smrj * about the old pte's REF / MOD bits. If the PFN remains the same, we leave 1767ae115bc7Smrj * the MOD/REF bits unchanged. 1768ae115bc7Smrj * 1769ae115bc7Smrj * If asked to overwrite a link to a lower page table with a large page 1770ae115bc7Smrj * mapping, this routine returns the special value of LPAGE_ERROR. This 1771ae115bc7Smrj * allows the upper HAT layers to retry with a smaller mapping size. 17727c478bd9Sstevel@tonic-gate */ 17737c478bd9Sstevel@tonic-gate x86pte_t 17747c478bd9Sstevel@tonic-gate x86pte_set(htable_t *ht, uint_t entry, x86pte_t new, void *ptr) 17757c478bd9Sstevel@tonic-gate { 17767c478bd9Sstevel@tonic-gate x86pte_t old; 1777ae115bc7Smrj x86pte_t prev; 17787c478bd9Sstevel@tonic-gate x86pte_t *ptep; 1779ae115bc7Smrj level_t l = ht->ht_level; 1780ae115bc7Smrj x86pte_t pfn_mask = (l != 0) ? PT_PADDR_LGPG : PT_PADDR; 1781ae115bc7Smrj x86pte_t n; 1782ae115bc7Smrj uintptr_t addr = htable_e2va(ht, entry); 1783ae115bc7Smrj hat_t *hat = ht->ht_hat; 17847c478bd9Sstevel@tonic-gate 1785ae115bc7Smrj ASSERT(new != 0); /* don't use to invalidate a PTE, see x86pte_update */ 17867c478bd9Sstevel@tonic-gate ASSERT(!(ht->ht_flags & HTABLE_SHARED_PFN)); 1787ae115bc7Smrj if (ptr == NULL) 1788ae115bc7Smrj ptep = x86pte_access_pagetable(ht, entry); 1789ae115bc7Smrj else 17907c478bd9Sstevel@tonic-gate ptep = ptr; 17917c478bd9Sstevel@tonic-gate 1792b193e412Skchow /* 1793ae115bc7Smrj * Install the new PTE. If remapping the same PFN, then 1794ae115bc7Smrj * copy existing REF/MOD bits to new mapping. 1795b193e412Skchow */ 1796ae115bc7Smrj do { 1797ae115bc7Smrj prev = GET_PTE(ptep); 1798ae115bc7Smrj n = new; 1799ae115bc7Smrj if (PTE_ISVALID(n) && (prev & pfn_mask) == (new & pfn_mask)) 1800b193e412Skchow n |= prev & (PT_REF | PT_MOD); 1801ae115bc7Smrj 1802ae115bc7Smrj /* 1803ae115bc7Smrj * Another thread may have installed this mapping already, 1804ae115bc7Smrj * flush the local TLB and be done. 1805ae115bc7Smrj */ 1806b193e412Skchow if (prev == n) { 18077c478bd9Sstevel@tonic-gate old = new; 1808ae115bc7Smrj mmu_tlbflush_entry((caddr_t)addr); 1809ae115bc7Smrj goto done; 18107c478bd9Sstevel@tonic-gate } 1811ae115bc7Smrj 1812ae115bc7Smrj /* 1813ae115bc7Smrj * Detect if we have a collision of installing a large 1814ae115bc7Smrj * page mapping where there already is a lower page table. 1815ae115bc7Smrj */ 181697704650Sjosephb if (l > 0 && (prev & PT_VALID) && !(prev & PT_PAGESIZE)) { 181797704650Sjosephb old = LPAGE_ERROR; 181897704650Sjosephb goto done; 181997704650Sjosephb } 1820ae115bc7Smrj 1821ae115bc7Smrj old = CAS_PTE(ptep, prev, n); 1822ae115bc7Smrj } while (old != prev); 1823ae115bc7Smrj 1824ae115bc7Smrj /* 1825ae115bc7Smrj * Do a TLB demap if needed, ie. the old pte was valid. 1826ae115bc7Smrj * 1827ae115bc7Smrj * Note that a stale TLB writeback to the PTE here either can't happen 1828ae115bc7Smrj * or doesn't matter. The PFN can only change for NOSYNC|NOCONSIST 1829ae115bc7Smrj * mappings, but they were created with REF and MOD already set, so 1830ae115bc7Smrj * no stale writeback will happen. 1831ae115bc7Smrj * 1832ae115bc7Smrj * Segmap is the only place where remaps happen on the same pfn and for 1833ae115bc7Smrj * that we want to preserve the stale REF/MOD bits. 1834ae115bc7Smrj */ 1835ae115bc7Smrj if (old & PT_REF) 1836ae115bc7Smrj hat_tlb_inval(hat, addr); 1837ae115bc7Smrj 1838ae115bc7Smrj done: 18397c478bd9Sstevel@tonic-gate if (ptr == NULL) 18407c478bd9Sstevel@tonic-gate x86pte_release_pagetable(ht); 18417c478bd9Sstevel@tonic-gate return (old); 18427c478bd9Sstevel@tonic-gate } 18437c478bd9Sstevel@tonic-gate 18447c478bd9Sstevel@tonic-gate /* 1845ae115bc7Smrj * Atomic compare and swap of a page table entry. No TLB invalidates are done. 1846ae115bc7Smrj * This is used for links between pagetables of different levels. 1847ae115bc7Smrj * Note we always create these links with dirty/access set, so they should 1848ae115bc7Smrj * never change. 18497c478bd9Sstevel@tonic-gate */ 1850ae115bc7Smrj x86pte_t 18517c478bd9Sstevel@tonic-gate x86pte_cas(htable_t *ht, uint_t entry, x86pte_t old, x86pte_t new) 18527c478bd9Sstevel@tonic-gate { 18537c478bd9Sstevel@tonic-gate x86pte_t pte; 18547c478bd9Sstevel@tonic-gate x86pte_t *ptep; 18557c478bd9Sstevel@tonic-gate 1856ae115bc7Smrj ptep = x86pte_access_pagetable(ht, entry); 1857ae115bc7Smrj pte = CAS_PTE(ptep, old, new); 18587c478bd9Sstevel@tonic-gate x86pte_release_pagetable(ht); 18597c478bd9Sstevel@tonic-gate return (pte); 18607c478bd9Sstevel@tonic-gate } 18617c478bd9Sstevel@tonic-gate 18627c478bd9Sstevel@tonic-gate /* 1863ae115bc7Smrj * Invalidate a page table entry as long as it currently maps something that 1864ae115bc7Smrj * matches the value determined by expect. 18657c478bd9Sstevel@tonic-gate * 1866ae115bc7Smrj * Also invalidates any TLB entries and returns the previous value of the PTE. 18677c478bd9Sstevel@tonic-gate */ 18687c478bd9Sstevel@tonic-gate x86pte_t 1869ae115bc7Smrj x86pte_inval( 1870ae115bc7Smrj htable_t *ht, 1871ae115bc7Smrj uint_t entry, 1872ae115bc7Smrj x86pte_t expect, 1873ae115bc7Smrj x86pte_t *pte_ptr) 18747c478bd9Sstevel@tonic-gate { 18757c478bd9Sstevel@tonic-gate x86pte_t *ptep; 187695c0a3c8Sjosephb x86pte_t oldpte; 187795c0a3c8Sjosephb x86pte_t found; 18787c478bd9Sstevel@tonic-gate 18797c478bd9Sstevel@tonic-gate ASSERT(!(ht->ht_flags & HTABLE_SHARED_PFN)); 1880ae115bc7Smrj ASSERT(ht->ht_level != VLP_LEVEL); 188197704650Sjosephb 1882ae115bc7Smrj if (pte_ptr != NULL) 18837c478bd9Sstevel@tonic-gate ptep = pte_ptr; 1884ae115bc7Smrj else 1885ae115bc7Smrj ptep = x86pte_access_pagetable(ht, entry); 18867c478bd9Sstevel@tonic-gate 18877c478bd9Sstevel@tonic-gate /* 188897704650Sjosephb * Note that the loop is needed to handle changes due to h/w updating 188997704650Sjosephb * of PT_MOD/PT_REF. 18907c478bd9Sstevel@tonic-gate */ 1891ae115bc7Smrj do { 189295c0a3c8Sjosephb oldpte = GET_PTE(ptep); 189395c0a3c8Sjosephb if (expect != 0 && (oldpte & PT_PADDR) != (expect & PT_PADDR)) 189495c0a3c8Sjosephb goto done; 189595c0a3c8Sjosephb found = CAS_PTE(ptep, oldpte, 0); 189695c0a3c8Sjosephb } while (found != oldpte); 189795c0a3c8Sjosephb if (oldpte & (PT_REF | PT_MOD)) 189895c0a3c8Sjosephb hat_tlb_inval(ht->ht_hat, htable_e2va(ht, entry)); 18997c478bd9Sstevel@tonic-gate 190095c0a3c8Sjosephb done: 19017c478bd9Sstevel@tonic-gate if (pte_ptr == NULL) 19027c478bd9Sstevel@tonic-gate x86pte_release_pagetable(ht); 190395c0a3c8Sjosephb return (oldpte); 19047c478bd9Sstevel@tonic-gate } 19057c478bd9Sstevel@tonic-gate 19067c478bd9Sstevel@tonic-gate /* 1907ae115bc7Smrj * Change a page table entry af it currently matches the value in expect. 19087c478bd9Sstevel@tonic-gate */ 19097c478bd9Sstevel@tonic-gate x86pte_t 1910ae115bc7Smrj x86pte_update( 1911ae115bc7Smrj htable_t *ht, 1912ae115bc7Smrj uint_t entry, 1913ae115bc7Smrj x86pte_t expect, 1914ae115bc7Smrj x86pte_t new) 19157c478bd9Sstevel@tonic-gate { 19167c478bd9Sstevel@tonic-gate x86pte_t *ptep; 1917ae115bc7Smrj x86pte_t found; 19187c478bd9Sstevel@tonic-gate 1919ae115bc7Smrj ASSERT(new != 0); 19207c478bd9Sstevel@tonic-gate ASSERT(!(ht->ht_flags & HTABLE_SHARED_PFN)); 1921ae115bc7Smrj ASSERT(ht->ht_level != VLP_LEVEL); 1922ae115bc7Smrj 1923ae115bc7Smrj ptep = x86pte_access_pagetable(ht, entry); 1924ae115bc7Smrj found = CAS_PTE(ptep, expect, new); 1925ae115bc7Smrj if (found == expect) { 1926ae115bc7Smrj hat_tlb_inval(ht->ht_hat, htable_e2va(ht, entry)); 19277c478bd9Sstevel@tonic-gate 19287c478bd9Sstevel@tonic-gate /* 1929ae115bc7Smrj * When removing write permission *and* clearing the 1930ae115bc7Smrj * MOD bit, check if a write happened via a stale 1931ae115bc7Smrj * TLB entry before the TLB shootdown finished. 1932ae115bc7Smrj * 1933ae115bc7Smrj * If it did happen, simply re-enable write permission and 1934ae115bc7Smrj * act like the original CAS failed. 19357c478bd9Sstevel@tonic-gate */ 1936ae115bc7Smrj if ((expect & (PT_WRITABLE | PT_MOD)) == PT_WRITABLE && 1937ae115bc7Smrj (new & (PT_WRITABLE | PT_MOD)) == 0 && 1938ae115bc7Smrj (GET_PTE(ptep) & PT_MOD) != 0) { 1939ae115bc7Smrj do { 1940ae115bc7Smrj found = GET_PTE(ptep); 1941ae115bc7Smrj found = 1942ae115bc7Smrj CAS_PTE(ptep, found, found | PT_WRITABLE); 1943ae115bc7Smrj } while ((found & PT_WRITABLE) == 0); 1944ae115bc7Smrj } 1945ae115bc7Smrj } 19467c478bd9Sstevel@tonic-gate x86pte_release_pagetable(ht); 1947ae115bc7Smrj return (found); 19487c478bd9Sstevel@tonic-gate } 19497c478bd9Sstevel@tonic-gate 19507c478bd9Sstevel@tonic-gate /* 19517c478bd9Sstevel@tonic-gate * Copy page tables - this is just a little more complicated than the 19527c478bd9Sstevel@tonic-gate * previous routines. Note that it's also not atomic! It also is never 19537c478bd9Sstevel@tonic-gate * used for VLP pagetables. 19547c478bd9Sstevel@tonic-gate */ 19557c478bd9Sstevel@tonic-gate void 19567c478bd9Sstevel@tonic-gate x86pte_copy(htable_t *src, htable_t *dest, uint_t entry, uint_t count) 19577c478bd9Sstevel@tonic-gate { 19587c478bd9Sstevel@tonic-gate caddr_t src_va; 19597c478bd9Sstevel@tonic-gate caddr_t dst_va; 19607c478bd9Sstevel@tonic-gate size_t size; 1961ae115bc7Smrj x86pte_t *pteptr; 1962ae115bc7Smrj x86pte_t pte; 19637c478bd9Sstevel@tonic-gate 19647c478bd9Sstevel@tonic-gate ASSERT(khat_running); 19657c478bd9Sstevel@tonic-gate ASSERT(!(dest->ht_flags & HTABLE_VLP)); 19667c478bd9Sstevel@tonic-gate ASSERT(!(src->ht_flags & HTABLE_VLP)); 19677c478bd9Sstevel@tonic-gate ASSERT(!(src->ht_flags & HTABLE_SHARED_PFN)); 19687c478bd9Sstevel@tonic-gate ASSERT(!(dest->ht_flags & HTABLE_SHARED_PFN)); 19697c478bd9Sstevel@tonic-gate 19707c478bd9Sstevel@tonic-gate /* 1971ae115bc7Smrj * Acquire access to the CPU pagetable windows for the dest and source. 19727c478bd9Sstevel@tonic-gate */ 1973ae115bc7Smrj dst_va = (caddr_t)x86pte_access_pagetable(dest, entry); 1974ae115bc7Smrj if (kpm_vbase) { 1975ae115bc7Smrj src_va = (caddr_t) 1976ae115bc7Smrj PT_INDEX_PTR(hat_kpm_pfn2va(src->ht_pfn), entry); 19777c478bd9Sstevel@tonic-gate } else { 1978ae115bc7Smrj uint_t x = PWIN_SRC(CPU->cpu_id); 19797c478bd9Sstevel@tonic-gate 19807c478bd9Sstevel@tonic-gate /* 19817c478bd9Sstevel@tonic-gate * Finish defining the src pagetable mapping 19827c478bd9Sstevel@tonic-gate */ 1983ae115bc7Smrj src_va = (caddr_t)PT_INDEX_PTR(PWIN_VA(x), entry); 1984ae115bc7Smrj pte = MAKEPTE(src->ht_pfn, 0) | mmu.pt_global | mmu.pt_nx; 1985ae115bc7Smrj pteptr = (x86pte_t *)PWIN_PTE_VA(x); 1986ae115bc7Smrj if (mmu.pae_hat) 1987ae115bc7Smrj *pteptr = pte; 1988ae115bc7Smrj else 1989ae115bc7Smrj *(x86pte32_t *)pteptr = pte; 1990ae115bc7Smrj mmu_tlbflush_entry((caddr_t)(PWIN_VA(x))); 19917c478bd9Sstevel@tonic-gate } 19927c478bd9Sstevel@tonic-gate 19937c478bd9Sstevel@tonic-gate /* 19947c478bd9Sstevel@tonic-gate * now do the copy 19957c478bd9Sstevel@tonic-gate */ 19967c478bd9Sstevel@tonic-gate size = count << mmu.pte_size_shift; 19977c478bd9Sstevel@tonic-gate bcopy(src_va, dst_va, size); 19987c478bd9Sstevel@tonic-gate 19997c478bd9Sstevel@tonic-gate x86pte_release_pagetable(dest); 20007c478bd9Sstevel@tonic-gate } 20017c478bd9Sstevel@tonic-gate 20027c478bd9Sstevel@tonic-gate /* 20037c478bd9Sstevel@tonic-gate * Zero page table entries - Note this doesn't use atomic stores! 20047c478bd9Sstevel@tonic-gate */ 2005ae115bc7Smrj static void 20067c478bd9Sstevel@tonic-gate x86pte_zero(htable_t *dest, uint_t entry, uint_t count) 20077c478bd9Sstevel@tonic-gate { 20087c478bd9Sstevel@tonic-gate caddr_t dst_va; 20097c478bd9Sstevel@tonic-gate size_t size; 20107c478bd9Sstevel@tonic-gate 20117c478bd9Sstevel@tonic-gate /* 20127c478bd9Sstevel@tonic-gate * Map in the page table to be zeroed. 20137c478bd9Sstevel@tonic-gate */ 20147c478bd9Sstevel@tonic-gate ASSERT(!(dest->ht_flags & HTABLE_SHARED_PFN)); 20157c478bd9Sstevel@tonic-gate ASSERT(!(dest->ht_flags & HTABLE_VLP)); 2016ae115bc7Smrj 2017ae115bc7Smrj dst_va = (caddr_t)x86pte_access_pagetable(dest, entry); 2018ae115bc7Smrj 20197c478bd9Sstevel@tonic-gate size = count << mmu.pte_size_shift; 2020ae115bc7Smrj ASSERT(size > BLOCKZEROALIGN); 2021ae115bc7Smrj #ifdef __i386 2022ae115bc7Smrj if ((x86_feature & X86_SSE2) == 0) 20237c478bd9Sstevel@tonic-gate bzero(dst_va, size); 2024ae115bc7Smrj else 2025ae115bc7Smrj #endif 2026ae115bc7Smrj block_zero_no_xmm(dst_va, size); 2027ae115bc7Smrj 20287c478bd9Sstevel@tonic-gate x86pte_release_pagetable(dest); 20297c478bd9Sstevel@tonic-gate } 20307c478bd9Sstevel@tonic-gate 20317c478bd9Sstevel@tonic-gate /* 20327c478bd9Sstevel@tonic-gate * Called to ensure that all pagetables are in the system dump 20337c478bd9Sstevel@tonic-gate */ 20347c478bd9Sstevel@tonic-gate void 20357c478bd9Sstevel@tonic-gate hat_dump(void) 20367c478bd9Sstevel@tonic-gate { 20377c478bd9Sstevel@tonic-gate hat_t *hat; 20387c478bd9Sstevel@tonic-gate uint_t h; 20397c478bd9Sstevel@tonic-gate htable_t *ht; 20407c478bd9Sstevel@tonic-gate 20417c478bd9Sstevel@tonic-gate /* 2042a85a6733Sjosephb * Dump all page tables 20437c478bd9Sstevel@tonic-gate */ 2044a85a6733Sjosephb for (hat = kas.a_hat; hat != NULL; hat = hat->hat_next) { 20457c478bd9Sstevel@tonic-gate for (h = 0; h < hat->hat_num_hash; ++h) { 20467c478bd9Sstevel@tonic-gate for (ht = hat->hat_ht_hash[h]; ht; ht = ht->ht_next) { 2047a85a6733Sjosephb if ((ht->ht_flags & HTABLE_VLP) == 0) 20487c478bd9Sstevel@tonic-gate dump_page(ht->ht_pfn); 20497c478bd9Sstevel@tonic-gate } 20507c478bd9Sstevel@tonic-gate } 20517c478bd9Sstevel@tonic-gate } 20527c478bd9Sstevel@tonic-gate } 2053