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 */ 217c478bd9Sstevel@tonic-gate /* 22ae115bc7Smrj * Copyright 2007 Sun Microsystems, Inc. All rights reserved. 237c478bd9Sstevel@tonic-gate * Use is subject to license terms. 247c478bd9Sstevel@tonic-gate */ 257c478bd9Sstevel@tonic-gate 267c478bd9Sstevel@tonic-gate #pragma ident "%Z%%M% %I% %E% SMI" 277c478bd9Sstevel@tonic-gate 287c478bd9Sstevel@tonic-gate /* 297c478bd9Sstevel@tonic-gate * VM - Hardware Address Translation management for i386 and amd64 307c478bd9Sstevel@tonic-gate * 317c478bd9Sstevel@tonic-gate * Implementation of the interfaces described in <common/vm/hat.h> 327c478bd9Sstevel@tonic-gate * 337c478bd9Sstevel@tonic-gate * Nearly all the details of how the hardware is managed should not be 347c478bd9Sstevel@tonic-gate * visible outside this layer except for misc. machine specific functions 357c478bd9Sstevel@tonic-gate * that work in conjunction with this code. 367c478bd9Sstevel@tonic-gate * 377c478bd9Sstevel@tonic-gate * Routines used only inside of i86pc/vm start with hati_ for HAT Internal. 387c478bd9Sstevel@tonic-gate */ 397c478bd9Sstevel@tonic-gate 407c478bd9Sstevel@tonic-gate #include <sys/machparam.h> 417c478bd9Sstevel@tonic-gate #include <sys/machsystm.h> 427c478bd9Sstevel@tonic-gate #include <sys/mman.h> 437c478bd9Sstevel@tonic-gate #include <sys/types.h> 447c478bd9Sstevel@tonic-gate #include <sys/systm.h> 457c478bd9Sstevel@tonic-gate #include <sys/cpuvar.h> 467c478bd9Sstevel@tonic-gate #include <sys/thread.h> 477c478bd9Sstevel@tonic-gate #include <sys/proc.h> 487c478bd9Sstevel@tonic-gate #include <sys/cpu.h> 497c478bd9Sstevel@tonic-gate #include <sys/kmem.h> 507c478bd9Sstevel@tonic-gate #include <sys/disp.h> 517c478bd9Sstevel@tonic-gate #include <sys/shm.h> 527c478bd9Sstevel@tonic-gate #include <sys/sysmacros.h> 537c478bd9Sstevel@tonic-gate #include <sys/machparam.h> 547c478bd9Sstevel@tonic-gate #include <sys/vmem.h> 557c478bd9Sstevel@tonic-gate #include <sys/vmsystm.h> 567c478bd9Sstevel@tonic-gate #include <sys/promif.h> 577c478bd9Sstevel@tonic-gate #include <sys/var.h> 587c478bd9Sstevel@tonic-gate #include <sys/x86_archext.h> 597c478bd9Sstevel@tonic-gate #include <sys/atomic.h> 607c478bd9Sstevel@tonic-gate #include <sys/bitmap.h> 61ae115bc7Smrj #include <sys/controlregs.h> 62ae115bc7Smrj #include <sys/bootconf.h> 63ae115bc7Smrj #include <sys/bootsvcs.h> 64ae115bc7Smrj #include <sys/bootinfo.h> 65*95c0a3c8Sjosephb #include <sys/archsystm.h> 667c478bd9Sstevel@tonic-gate 677c478bd9Sstevel@tonic-gate #include <vm/seg_kmem.h> 687c478bd9Sstevel@tonic-gate #include <vm/hat_i86.h> 697c478bd9Sstevel@tonic-gate #include <vm/as.h> 707c478bd9Sstevel@tonic-gate #include <vm/seg.h> 717c478bd9Sstevel@tonic-gate #include <vm/page.h> 727c478bd9Sstevel@tonic-gate #include <vm/seg_kp.h> 737c478bd9Sstevel@tonic-gate #include <vm/seg_kpm.h> 747c478bd9Sstevel@tonic-gate #include <vm/vm_dep.h> 75ae115bc7Smrj #include <vm/kboot_mmu.h> 767c478bd9Sstevel@tonic-gate 777c478bd9Sstevel@tonic-gate #include <sys/cmn_err.h> 787c478bd9Sstevel@tonic-gate 797c478bd9Sstevel@tonic-gate /* 807c478bd9Sstevel@tonic-gate * Basic parameters for hat operation. 817c478bd9Sstevel@tonic-gate */ 827c478bd9Sstevel@tonic-gate struct hat_mmu_info mmu; 837c478bd9Sstevel@tonic-gate 847c478bd9Sstevel@tonic-gate /* 857c478bd9Sstevel@tonic-gate * The page that is the kernel's top level pagetable. 867c478bd9Sstevel@tonic-gate * 877c478bd9Sstevel@tonic-gate * For 32 bit VLP support, the kernel hat will use the 1st 4 entries 887c478bd9Sstevel@tonic-gate * on this 4K page for its top level page table. The remaining groups of 897c478bd9Sstevel@tonic-gate * 4 entries are used for per processor copies of user VLP pagetables for 907c478bd9Sstevel@tonic-gate * running threads. See hat_switch() and reload_pae32() for details. 917c478bd9Sstevel@tonic-gate * 927c478bd9Sstevel@tonic-gate * vlp_page[0] - 0th level==2 PTE for kernel HAT (will be zero) 937c478bd9Sstevel@tonic-gate * vlp_page[1] - 1st level==2 PTE for kernel HAT (will be zero) 947c478bd9Sstevel@tonic-gate * vlp_page[2] - 2nd level==2 PTE for kernel HAT (zero for small memory) 957c478bd9Sstevel@tonic-gate * vlp_page[3] - 3rd level==2 PTE for kernel 967c478bd9Sstevel@tonic-gate * 977c478bd9Sstevel@tonic-gate * vlp_page[4] - 0th level==2 PTE for user thread on cpu 0 987c478bd9Sstevel@tonic-gate * vlp_page[5] - 1st level==2 PTE for user thread on cpu 0 997c478bd9Sstevel@tonic-gate * vlp_page[6] - 2nd level==2 PTE for user thread on cpu 0 1007c478bd9Sstevel@tonic-gate * vlp_page[7] - probably copy of kernel PTE 1017c478bd9Sstevel@tonic-gate * 1027c478bd9Sstevel@tonic-gate * vlp_page[8] - 0th level==2 PTE for user thread on cpu 1 1037c478bd9Sstevel@tonic-gate * vlp_page[9] - 1st level==2 PTE for user thread on cpu 1 1047c478bd9Sstevel@tonic-gate * vlp_page[10] - 2nd level==2 PTE for user thread on cpu 1 1057c478bd9Sstevel@tonic-gate * vlp_page[11] - probably copy of kernel PTE 1067c478bd9Sstevel@tonic-gate * ... 1077c478bd9Sstevel@tonic-gate * 1087c478bd9Sstevel@tonic-gate * when / where the kernel PTE's are (entry 2 or 3 or none) depends 1097c478bd9Sstevel@tonic-gate * on kernelbase. 1107c478bd9Sstevel@tonic-gate */ 1117c478bd9Sstevel@tonic-gate static x86pte_t *vlp_page; 1127c478bd9Sstevel@tonic-gate 1137c478bd9Sstevel@tonic-gate /* 1147c478bd9Sstevel@tonic-gate * forward declaration of internal utility routines 1157c478bd9Sstevel@tonic-gate */ 1167c478bd9Sstevel@tonic-gate static x86pte_t hati_update_pte(htable_t *ht, uint_t entry, x86pte_t expected, 1177c478bd9Sstevel@tonic-gate x86pte_t new); 1187c478bd9Sstevel@tonic-gate 1197c478bd9Sstevel@tonic-gate /* 1207c478bd9Sstevel@tonic-gate * The kernel address space exists in all HATs. To implement this the 1217c478bd9Sstevel@tonic-gate * kernel reserves a fixed number of entries in every topmost level page 1227c478bd9Sstevel@tonic-gate * table. The values are setup in hat_init() and then copied to every hat 1237c478bd9Sstevel@tonic-gate * created by hat_alloc(). This means that kernelbase must be: 1247c478bd9Sstevel@tonic-gate * 1257c478bd9Sstevel@tonic-gate * 4Meg aligned for 32 bit kernels 1267c478bd9Sstevel@tonic-gate * 512Gig aligned for x86_64 64 bit kernel 1277c478bd9Sstevel@tonic-gate * 1287c478bd9Sstevel@tonic-gate * The PAE 32 bit hat is handled as a special case. Otherwise requiring 1Gig 1297c478bd9Sstevel@tonic-gate * alignment would use too much VA for the kernel. 1307c478bd9Sstevel@tonic-gate * 1317c478bd9Sstevel@tonic-gate */ 1327c478bd9Sstevel@tonic-gate static uint_t khat_start; /* index of 1st entry in kernel's top ptable */ 1337c478bd9Sstevel@tonic-gate static uint_t khat_entries; /* number of entries in kernel's top ptable */ 1347c478bd9Sstevel@tonic-gate 1357c478bd9Sstevel@tonic-gate #if defined(__i386) 1367c478bd9Sstevel@tonic-gate 1377c478bd9Sstevel@tonic-gate static htable_t *khat_pae32_htable = NULL; 1387c478bd9Sstevel@tonic-gate static uint_t khat_pae32_start; 1397c478bd9Sstevel@tonic-gate static uint_t khat_pae32_entries; 1407c478bd9Sstevel@tonic-gate 1417c478bd9Sstevel@tonic-gate #endif 1427c478bd9Sstevel@tonic-gate 1437c478bd9Sstevel@tonic-gate uint_t use_boot_reserve = 1; /* cleared after early boot process */ 1447c478bd9Sstevel@tonic-gate uint_t can_steal_post_boot = 0; /* set late in boot to enable stealing */ 1457c478bd9Sstevel@tonic-gate 1467c478bd9Sstevel@tonic-gate /* 1477c478bd9Sstevel@tonic-gate * A cpuset for all cpus. This is used for kernel address cross calls, since 1487c478bd9Sstevel@tonic-gate * the kernel addresses apply to all cpus. 1497c478bd9Sstevel@tonic-gate */ 1507c478bd9Sstevel@tonic-gate cpuset_t khat_cpuset; 1517c478bd9Sstevel@tonic-gate 1527c478bd9Sstevel@tonic-gate /* 1537c478bd9Sstevel@tonic-gate * management stuff for hat structures 1547c478bd9Sstevel@tonic-gate */ 1557c478bd9Sstevel@tonic-gate kmutex_t hat_list_lock; 1567c478bd9Sstevel@tonic-gate kcondvar_t hat_list_cv; 1577c478bd9Sstevel@tonic-gate kmem_cache_t *hat_cache; 1587c478bd9Sstevel@tonic-gate kmem_cache_t *hat_hash_cache; 1597c478bd9Sstevel@tonic-gate kmem_cache_t *vlp_hash_cache; 1607c478bd9Sstevel@tonic-gate 1617c478bd9Sstevel@tonic-gate /* 1627c478bd9Sstevel@tonic-gate * Simple statistics 1637c478bd9Sstevel@tonic-gate */ 1647c478bd9Sstevel@tonic-gate struct hatstats hatstat; 1657c478bd9Sstevel@tonic-gate 1667c478bd9Sstevel@tonic-gate /* 1677c478bd9Sstevel@tonic-gate * useful stuff for atomic access/clearing/setting REF/MOD/RO bits in page_t's. 1687c478bd9Sstevel@tonic-gate */ 1697c478bd9Sstevel@tonic-gate extern void atomic_orb(uchar_t *addr, uchar_t val); 1707c478bd9Sstevel@tonic-gate extern void atomic_andb(uchar_t *addr, uchar_t val); 1717c478bd9Sstevel@tonic-gate 1727c478bd9Sstevel@tonic-gate #define PP_GETRM(pp, rmmask) (pp->p_nrm & rmmask) 1737c478bd9Sstevel@tonic-gate #define PP_ISMOD(pp) PP_GETRM(pp, P_MOD) 1747c478bd9Sstevel@tonic-gate #define PP_ISREF(pp) PP_GETRM(pp, P_REF) 1757c478bd9Sstevel@tonic-gate #define PP_ISRO(pp) PP_GETRM(pp, P_RO) 1767c478bd9Sstevel@tonic-gate 1777c478bd9Sstevel@tonic-gate #define PP_SETRM(pp, rm) atomic_orb(&(pp->p_nrm), rm) 1787c478bd9Sstevel@tonic-gate #define PP_SETMOD(pp) PP_SETRM(pp, P_MOD) 1797c478bd9Sstevel@tonic-gate #define PP_SETREF(pp) PP_SETRM(pp, P_REF) 1807c478bd9Sstevel@tonic-gate #define PP_SETRO(pp) PP_SETRM(pp, P_RO) 1817c478bd9Sstevel@tonic-gate 1827c478bd9Sstevel@tonic-gate #define PP_CLRRM(pp, rm) atomic_andb(&(pp->p_nrm), ~(rm)) 1837c478bd9Sstevel@tonic-gate #define PP_CLRMOD(pp) PP_CLRRM(pp, P_MOD) 1847c478bd9Sstevel@tonic-gate #define PP_CLRREF(pp) PP_CLRRM(pp, P_REF) 1857c478bd9Sstevel@tonic-gate #define PP_CLRRO(pp) PP_CLRRM(pp, P_RO) 1867c478bd9Sstevel@tonic-gate #define PP_CLRALL(pp) PP_CLRRM(pp, P_MOD | P_REF | P_RO) 1877c478bd9Sstevel@tonic-gate 1887c478bd9Sstevel@tonic-gate /* 1897c478bd9Sstevel@tonic-gate * kmem cache constructor for struct hat 1907c478bd9Sstevel@tonic-gate */ 1917c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 1927c478bd9Sstevel@tonic-gate static int 1937c478bd9Sstevel@tonic-gate hati_constructor(void *buf, void *handle, int kmflags) 1947c478bd9Sstevel@tonic-gate { 1957c478bd9Sstevel@tonic-gate hat_t *hat = buf; 1967c478bd9Sstevel@tonic-gate 1977c478bd9Sstevel@tonic-gate mutex_init(&hat->hat_mutex, NULL, MUTEX_DEFAULT, NULL); 1987c478bd9Sstevel@tonic-gate bzero(hat->hat_pages_mapped, 1997c478bd9Sstevel@tonic-gate sizeof (pgcnt_t) * (mmu.max_page_level + 1)); 2007c478bd9Sstevel@tonic-gate hat->hat_stats = 0; 2017c478bd9Sstevel@tonic-gate hat->hat_flags = 0; 2027c478bd9Sstevel@tonic-gate CPUSET_ZERO(hat->hat_cpus); 2037c478bd9Sstevel@tonic-gate hat->hat_htable = NULL; 2047c478bd9Sstevel@tonic-gate hat->hat_ht_hash = NULL; 2057c478bd9Sstevel@tonic-gate return (0); 2067c478bd9Sstevel@tonic-gate } 2077c478bd9Sstevel@tonic-gate 2087c478bd9Sstevel@tonic-gate /* 2097c478bd9Sstevel@tonic-gate * Allocate a hat structure for as. We also create the top level 2107c478bd9Sstevel@tonic-gate * htable and initialize it to contain the kernel hat entries. 2117c478bd9Sstevel@tonic-gate */ 2127c478bd9Sstevel@tonic-gate hat_t * 2137c478bd9Sstevel@tonic-gate hat_alloc(struct as *as) 2147c478bd9Sstevel@tonic-gate { 2157c478bd9Sstevel@tonic-gate hat_t *hat; 2167c478bd9Sstevel@tonic-gate htable_t *ht; /* top level htable */ 2177c478bd9Sstevel@tonic-gate uint_t use_vlp; 2187c478bd9Sstevel@tonic-gate 2197c478bd9Sstevel@tonic-gate /* 2207c478bd9Sstevel@tonic-gate * Once we start creating user process HATs we can enable 2217c478bd9Sstevel@tonic-gate * the htable_steal() code. 2227c478bd9Sstevel@tonic-gate */ 2237c478bd9Sstevel@tonic-gate if (can_steal_post_boot == 0) 2247c478bd9Sstevel@tonic-gate can_steal_post_boot = 1; 2257c478bd9Sstevel@tonic-gate 2267c478bd9Sstevel@tonic-gate ASSERT(AS_WRITE_HELD(as, &as->a_lock)); 2277c478bd9Sstevel@tonic-gate hat = kmem_cache_alloc(hat_cache, KM_SLEEP); 2287c478bd9Sstevel@tonic-gate hat->hat_as = as; 2297c478bd9Sstevel@tonic-gate mutex_init(&hat->hat_mutex, NULL, MUTEX_DEFAULT, NULL); 2307c478bd9Sstevel@tonic-gate ASSERT(hat->hat_flags == 0); 2317c478bd9Sstevel@tonic-gate 2327c478bd9Sstevel@tonic-gate /* 2337c478bd9Sstevel@tonic-gate * a 32 bit process uses a VLP style hat when using PAE 2347c478bd9Sstevel@tonic-gate */ 2357c478bd9Sstevel@tonic-gate #if defined(__amd64) 2367c478bd9Sstevel@tonic-gate use_vlp = (ttoproc(curthread)->p_model == DATAMODEL_ILP32); 2377c478bd9Sstevel@tonic-gate #elif defined(__i386) 2387c478bd9Sstevel@tonic-gate use_vlp = mmu.pae_hat; 2397c478bd9Sstevel@tonic-gate #endif 2407c478bd9Sstevel@tonic-gate if (use_vlp) { 2417c478bd9Sstevel@tonic-gate hat->hat_flags = HAT_VLP; 2427c478bd9Sstevel@tonic-gate bzero(hat->hat_vlp_ptes, VLP_SIZE); 2437c478bd9Sstevel@tonic-gate } 2447c478bd9Sstevel@tonic-gate 2457c478bd9Sstevel@tonic-gate /* 2467c478bd9Sstevel@tonic-gate * Allocate the htable hash 2477c478bd9Sstevel@tonic-gate */ 2487c478bd9Sstevel@tonic-gate if ((hat->hat_flags & HAT_VLP)) { 2497c478bd9Sstevel@tonic-gate hat->hat_num_hash = mmu.vlp_hash_cnt; 2507c478bd9Sstevel@tonic-gate hat->hat_ht_hash = kmem_cache_alloc(vlp_hash_cache, KM_SLEEP); 2517c478bd9Sstevel@tonic-gate } else { 2527c478bd9Sstevel@tonic-gate hat->hat_num_hash = mmu.hash_cnt; 2537c478bd9Sstevel@tonic-gate hat->hat_ht_hash = kmem_cache_alloc(hat_hash_cache, KM_SLEEP); 2547c478bd9Sstevel@tonic-gate } 2557c478bd9Sstevel@tonic-gate bzero(hat->hat_ht_hash, hat->hat_num_hash * sizeof (htable_t *)); 2567c478bd9Sstevel@tonic-gate 2577c478bd9Sstevel@tonic-gate /* 2587c478bd9Sstevel@tonic-gate * Initialize Kernel HAT entries at the top of the top level page 2597c478bd9Sstevel@tonic-gate * table for the new hat. 2607c478bd9Sstevel@tonic-gate * 2617c478bd9Sstevel@tonic-gate * Note that we don't call htable_release() for the top level, that 2627c478bd9Sstevel@tonic-gate * happens when the hat is destroyed in hat_free_end() 2637c478bd9Sstevel@tonic-gate */ 2647c478bd9Sstevel@tonic-gate hat->hat_htable = NULL; 2657c478bd9Sstevel@tonic-gate hat->hat_ht_cached = NULL; 2667c478bd9Sstevel@tonic-gate ht = htable_create(hat, (uintptr_t)0, TOP_LEVEL(hat), NULL); 267ae115bc7Smrj 2687c478bd9Sstevel@tonic-gate if (!(hat->hat_flags & HAT_VLP)) 2697c478bd9Sstevel@tonic-gate x86pte_copy(kas.a_hat->hat_htable, ht, khat_start, 2707c478bd9Sstevel@tonic-gate khat_entries); 2717c478bd9Sstevel@tonic-gate #if defined(__i386) 2727c478bd9Sstevel@tonic-gate else if (khat_entries > 0) 2737c478bd9Sstevel@tonic-gate bcopy(vlp_page + khat_start, hat->hat_vlp_ptes + khat_start, 2747c478bd9Sstevel@tonic-gate khat_entries * sizeof (x86pte_t)); 2757c478bd9Sstevel@tonic-gate #endif 2767c478bd9Sstevel@tonic-gate hat->hat_htable = ht; 2777c478bd9Sstevel@tonic-gate 2787c478bd9Sstevel@tonic-gate #if defined(__i386) 2797c478bd9Sstevel@tonic-gate /* 2807c478bd9Sstevel@tonic-gate * PAE32 HAT alignment is less restrictive than the others to keep 2817c478bd9Sstevel@tonic-gate * the kernel from using too much VA. Because of this we may need 2827c478bd9Sstevel@tonic-gate * one layer further down when kernelbase isn't 1Gig aligned. 2837c478bd9Sstevel@tonic-gate * See hat_free_end() for the htable_release() that goes with this 2847c478bd9Sstevel@tonic-gate * htable_create() 2857c478bd9Sstevel@tonic-gate */ 2867c478bd9Sstevel@tonic-gate if (khat_pae32_htable != NULL) { 2877c478bd9Sstevel@tonic-gate ht = htable_create(hat, kernelbase, 2887c478bd9Sstevel@tonic-gate khat_pae32_htable->ht_level, NULL); 2897c478bd9Sstevel@tonic-gate x86pte_copy(khat_pae32_htable, ht, khat_pae32_start, 2907c478bd9Sstevel@tonic-gate khat_pae32_entries); 2917c478bd9Sstevel@tonic-gate ht->ht_valid_cnt = khat_pae32_entries; 2927c478bd9Sstevel@tonic-gate } 2937c478bd9Sstevel@tonic-gate #endif 2947c478bd9Sstevel@tonic-gate 2957c478bd9Sstevel@tonic-gate /* 296a85a6733Sjosephb * Put it at the start of the global list of all hats (used by stealing) 297a85a6733Sjosephb * 298a85a6733Sjosephb * kas.a_hat is not in the list but is instead used to find the 299a85a6733Sjosephb * first and last items in the list. 300a85a6733Sjosephb * 301a85a6733Sjosephb * - kas.a_hat->hat_next points to the start of the user hats. 302a85a6733Sjosephb * The list ends where hat->hat_next == NULL 303a85a6733Sjosephb * 304a85a6733Sjosephb * - kas.a_hat->hat_prev points to the last of the user hats. 305a85a6733Sjosephb * The list begins where hat->hat_prev == NULL 3067c478bd9Sstevel@tonic-gate */ 3077c478bd9Sstevel@tonic-gate mutex_enter(&hat_list_lock); 308a85a6733Sjosephb hat->hat_prev = NULL; 309a85a6733Sjosephb hat->hat_next = kas.a_hat->hat_next; 310a85a6733Sjosephb if (hat->hat_next) 311a85a6733Sjosephb hat->hat_next->hat_prev = hat; 312a85a6733Sjosephb else 313a85a6733Sjosephb kas.a_hat->hat_prev = hat; 3147c478bd9Sstevel@tonic-gate kas.a_hat->hat_next = hat; 3157c478bd9Sstevel@tonic-gate mutex_exit(&hat_list_lock); 3167c478bd9Sstevel@tonic-gate 3177c478bd9Sstevel@tonic-gate return (hat); 3187c478bd9Sstevel@tonic-gate } 3197c478bd9Sstevel@tonic-gate 3207c478bd9Sstevel@tonic-gate /* 3217c478bd9Sstevel@tonic-gate * process has finished executing but as has not been cleaned up yet. 3227c478bd9Sstevel@tonic-gate */ 3237c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 3247c478bd9Sstevel@tonic-gate void 3257c478bd9Sstevel@tonic-gate hat_free_start(hat_t *hat) 3267c478bd9Sstevel@tonic-gate { 3277c478bd9Sstevel@tonic-gate ASSERT(AS_WRITE_HELD(hat->hat_as, &hat->hat_as->a_lock)); 328a85a6733Sjosephb 329a85a6733Sjosephb /* 330a85a6733Sjosephb * If the hat is currently a stealing victim, wait for the stealing 331a85a6733Sjosephb * to finish. Once we mark it as HAT_FREEING, htable_steal() 332a85a6733Sjosephb * won't look at its pagetables anymore. 333a85a6733Sjosephb */ 3347c478bd9Sstevel@tonic-gate mutex_enter(&hat_list_lock); 335a85a6733Sjosephb while (hat->hat_flags & HAT_VICTIM) 336a85a6733Sjosephb cv_wait(&hat_list_cv, &hat_list_lock); 3377c478bd9Sstevel@tonic-gate hat->hat_flags |= HAT_FREEING; 3387c478bd9Sstevel@tonic-gate mutex_exit(&hat_list_lock); 3397c478bd9Sstevel@tonic-gate } 3407c478bd9Sstevel@tonic-gate 3417c478bd9Sstevel@tonic-gate /* 3427c478bd9Sstevel@tonic-gate * An address space is being destroyed, so we destroy the associated hat. 3437c478bd9Sstevel@tonic-gate */ 3447c478bd9Sstevel@tonic-gate void 3457c478bd9Sstevel@tonic-gate hat_free_end(hat_t *hat) 3467c478bd9Sstevel@tonic-gate { 3477c478bd9Sstevel@tonic-gate int i; 3487c478bd9Sstevel@tonic-gate kmem_cache_t *cache; 3497c478bd9Sstevel@tonic-gate 3507c478bd9Sstevel@tonic-gate #ifdef DEBUG 3517c478bd9Sstevel@tonic-gate for (i = 0; i <= mmu.max_page_level; i++) 3527c478bd9Sstevel@tonic-gate ASSERT(hat->hat_pages_mapped[i] == 0); 3537c478bd9Sstevel@tonic-gate #endif 3547c478bd9Sstevel@tonic-gate ASSERT(hat->hat_flags & HAT_FREEING); 3557c478bd9Sstevel@tonic-gate 3567c478bd9Sstevel@tonic-gate /* 3577c478bd9Sstevel@tonic-gate * must not be running on the given hat 3587c478bd9Sstevel@tonic-gate */ 3597c478bd9Sstevel@tonic-gate ASSERT(CPU->cpu_current_hat != hat); 3607c478bd9Sstevel@tonic-gate 3617c478bd9Sstevel@tonic-gate /* 362a85a6733Sjosephb * Remove it from the list of HATs 3637c478bd9Sstevel@tonic-gate */ 3647c478bd9Sstevel@tonic-gate mutex_enter(&hat_list_lock); 365a85a6733Sjosephb if (hat->hat_prev) 366a85a6733Sjosephb hat->hat_prev->hat_next = hat->hat_next; 367a85a6733Sjosephb else 3687c478bd9Sstevel@tonic-gate kas.a_hat->hat_next = hat->hat_next; 369a85a6733Sjosephb if (hat->hat_next) 370a85a6733Sjosephb hat->hat_next->hat_prev = hat->hat_prev; 371a85a6733Sjosephb else 372a85a6733Sjosephb kas.a_hat->hat_prev = hat->hat_prev; 3737c478bd9Sstevel@tonic-gate mutex_exit(&hat_list_lock); 374a85a6733Sjosephb hat->hat_next = hat->hat_prev = NULL; 3757c478bd9Sstevel@tonic-gate 3767c478bd9Sstevel@tonic-gate /* 3777c478bd9Sstevel@tonic-gate * Make a pass through the htables freeing them all up. 3787c478bd9Sstevel@tonic-gate */ 3797c478bd9Sstevel@tonic-gate htable_purge_hat(hat); 3807c478bd9Sstevel@tonic-gate 3817c478bd9Sstevel@tonic-gate /* 3827c478bd9Sstevel@tonic-gate * Decide which kmem cache the hash table came from, then free it. 3837c478bd9Sstevel@tonic-gate */ 3847c478bd9Sstevel@tonic-gate if (hat->hat_flags & HAT_VLP) 3857c478bd9Sstevel@tonic-gate cache = vlp_hash_cache; 3867c478bd9Sstevel@tonic-gate else 3877c478bd9Sstevel@tonic-gate cache = hat_hash_cache; 3887c478bd9Sstevel@tonic-gate kmem_cache_free(cache, hat->hat_ht_hash); 3897c478bd9Sstevel@tonic-gate hat->hat_ht_hash = NULL; 3907c478bd9Sstevel@tonic-gate 3917c478bd9Sstevel@tonic-gate hat->hat_flags = 0; 3927c478bd9Sstevel@tonic-gate kmem_cache_free(hat_cache, hat); 3937c478bd9Sstevel@tonic-gate } 3947c478bd9Sstevel@tonic-gate 3957c478bd9Sstevel@tonic-gate /* 3967c478bd9Sstevel@tonic-gate * round kernelbase down to a supported value to use for _userlimit 3977c478bd9Sstevel@tonic-gate * 3987c478bd9Sstevel@tonic-gate * userlimit must be aligned down to an entry in the top level htable. 3997c478bd9Sstevel@tonic-gate * The one exception is for 32 bit HAT's running PAE. 4007c478bd9Sstevel@tonic-gate */ 4017c478bd9Sstevel@tonic-gate uintptr_t 4027c478bd9Sstevel@tonic-gate hat_kernelbase(uintptr_t va) 4037c478bd9Sstevel@tonic-gate { 4047c478bd9Sstevel@tonic-gate #if defined(__i386) 4057c478bd9Sstevel@tonic-gate va &= LEVEL_MASK(1); 4067c478bd9Sstevel@tonic-gate #endif 4077c478bd9Sstevel@tonic-gate if (IN_VA_HOLE(va)) 4087c478bd9Sstevel@tonic-gate panic("_userlimit %p will fall in VA hole\n", (void *)va); 4097c478bd9Sstevel@tonic-gate return (va); 4107c478bd9Sstevel@tonic-gate } 4117c478bd9Sstevel@tonic-gate 4127c478bd9Sstevel@tonic-gate /* 4137c478bd9Sstevel@tonic-gate * Initialize hat data structures based on processor MMU information. 4147c478bd9Sstevel@tonic-gate */ 4157c478bd9Sstevel@tonic-gate void 4167c478bd9Sstevel@tonic-gate mmu_init(void) 4177c478bd9Sstevel@tonic-gate { 4187c478bd9Sstevel@tonic-gate uint_t max_htables; 4197c478bd9Sstevel@tonic-gate uint_t pa_bits; 4207c478bd9Sstevel@tonic-gate uint_t va_bits; 4217c478bd9Sstevel@tonic-gate int i; 4227c478bd9Sstevel@tonic-gate 4237c478bd9Sstevel@tonic-gate /* 424ae115bc7Smrj * If CPU enabled the page table global bit, use it for the kernel 425ae115bc7Smrj * This is bit 7 in CR4 (PGE - Page Global Enable). 4267c478bd9Sstevel@tonic-gate */ 427ae115bc7Smrj if ((x86_feature & X86_PGE) != 0 && (getcr4() & CR4_PGE) != 0) 4287c478bd9Sstevel@tonic-gate mmu.pt_global = PT_GLOBAL; 4297c478bd9Sstevel@tonic-gate 4307c478bd9Sstevel@tonic-gate /* 431ae115bc7Smrj * Detect NX and PAE usage. 4327c478bd9Sstevel@tonic-gate */ 433ae115bc7Smrj mmu.pae_hat = kbm_pae_support; 434ae115bc7Smrj if (kbm_nx_support) 4357c478bd9Sstevel@tonic-gate mmu.pt_nx = PT_NX; 436ae115bc7Smrj else 4377c478bd9Sstevel@tonic-gate mmu.pt_nx = 0; 4387c478bd9Sstevel@tonic-gate 439935f8dd0Sjosephb /* 440935f8dd0Sjosephb * Intel CPUs allow speculative caching (in TLB-like h/w) of 441935f8dd0Sjosephb * entries in upper page tables even though there may not be 442935f8dd0Sjosephb * any valid entries in lower tables. This implies we have to 443935f8dd0Sjosephb * re-INVLPG at every upper page table entry invalidation. 444935f8dd0Sjosephb */ 445935f8dd0Sjosephb if (cpuid_getvendor(CPU) == X86_VENDOR_Intel) 446935f8dd0Sjosephb mmu.inval_nonleaf = 1; 447935f8dd0Sjosephb else 448935f8dd0Sjosephb mmu.inval_nonleaf = 0; 4497c478bd9Sstevel@tonic-gate /* 4507c478bd9Sstevel@tonic-gate * Use CPU info to set various MMU parameters 4517c478bd9Sstevel@tonic-gate */ 4527c478bd9Sstevel@tonic-gate cpuid_get_addrsize(CPU, &pa_bits, &va_bits); 4537c478bd9Sstevel@tonic-gate 4547c478bd9Sstevel@tonic-gate if (va_bits < sizeof (void *) * NBBY) { 4557c478bd9Sstevel@tonic-gate mmu.hole_start = (1ul << (va_bits - 1)); 4567c478bd9Sstevel@tonic-gate mmu.hole_end = 0ul - mmu.hole_start - 1; 4577c478bd9Sstevel@tonic-gate } else { 4587c478bd9Sstevel@tonic-gate mmu.hole_end = 0; 4597c478bd9Sstevel@tonic-gate mmu.hole_start = mmu.hole_end - 1; 4607c478bd9Sstevel@tonic-gate } 4617c478bd9Sstevel@tonic-gate #if defined(OPTERON_ERRATUM_121) 4627c478bd9Sstevel@tonic-gate /* 4637c478bd9Sstevel@tonic-gate * If erratum 121 has already been detected at this time, hole_start 4647c478bd9Sstevel@tonic-gate * contains the value to be subtracted from mmu.hole_start. 4657c478bd9Sstevel@tonic-gate */ 4667c478bd9Sstevel@tonic-gate ASSERT(hole_start == 0 || opteron_erratum_121 != 0); 4677c478bd9Sstevel@tonic-gate hole_start = mmu.hole_start - hole_start; 4687c478bd9Sstevel@tonic-gate #else 4697c478bd9Sstevel@tonic-gate hole_start = mmu.hole_start; 4707c478bd9Sstevel@tonic-gate #endif 4717c478bd9Sstevel@tonic-gate hole_end = mmu.hole_end; 4727c478bd9Sstevel@tonic-gate 4737c478bd9Sstevel@tonic-gate mmu.highest_pfn = mmu_btop((1ull << pa_bits) - 1); 4747c478bd9Sstevel@tonic-gate if (mmu.pae_hat == 0 && pa_bits > 32) 4757c478bd9Sstevel@tonic-gate mmu.highest_pfn = PFN_4G - 1; 4767c478bd9Sstevel@tonic-gate 4777c478bd9Sstevel@tonic-gate if (mmu.pae_hat) { 4787c478bd9Sstevel@tonic-gate mmu.pte_size = 8; /* 8 byte PTEs */ 4797c478bd9Sstevel@tonic-gate mmu.pte_size_shift = 3; 4807c478bd9Sstevel@tonic-gate } else { 4817c478bd9Sstevel@tonic-gate mmu.pte_size = 4; /* 4 byte PTEs */ 4827c478bd9Sstevel@tonic-gate mmu.pte_size_shift = 2; 4837c478bd9Sstevel@tonic-gate } 4847c478bd9Sstevel@tonic-gate 4857c478bd9Sstevel@tonic-gate if (mmu.pae_hat && (x86_feature & X86_PAE) == 0) 4867c478bd9Sstevel@tonic-gate panic("Processor does not support PAE"); 4877c478bd9Sstevel@tonic-gate 4887c478bd9Sstevel@tonic-gate if ((x86_feature & X86_CX8) == 0) 4897c478bd9Sstevel@tonic-gate panic("Processor does not support cmpxchg8b instruction"); 4907c478bd9Sstevel@tonic-gate 4917c478bd9Sstevel@tonic-gate /* 4927c478bd9Sstevel@tonic-gate * Initialize parameters based on the 64 or 32 bit kernels and 4937c478bd9Sstevel@tonic-gate * for the 32 bit kernel decide if we should use PAE. 4947c478bd9Sstevel@tonic-gate */ 495ae115bc7Smrj if (kbm_largepage_support) 4967c478bd9Sstevel@tonic-gate mmu.max_page_level = 1; 4977c478bd9Sstevel@tonic-gate else 4987c478bd9Sstevel@tonic-gate mmu.max_page_level = 0; 4997c478bd9Sstevel@tonic-gate mmu_page_sizes = mmu.max_page_level + 1; 5007c478bd9Sstevel@tonic-gate mmu_exported_page_sizes = mmu_page_sizes; 5017c478bd9Sstevel@tonic-gate 5027c478bd9Sstevel@tonic-gate #if defined(__amd64) 5037c478bd9Sstevel@tonic-gate 5047c478bd9Sstevel@tonic-gate mmu.num_level = 4; 5057c478bd9Sstevel@tonic-gate mmu.max_level = 3; 5067c478bd9Sstevel@tonic-gate mmu.ptes_per_table = 512; 5077c478bd9Sstevel@tonic-gate mmu.top_level_count = 512; 5087c478bd9Sstevel@tonic-gate 5097c478bd9Sstevel@tonic-gate mmu.level_shift[0] = 12; 5107c478bd9Sstevel@tonic-gate mmu.level_shift[1] = 21; 5117c478bd9Sstevel@tonic-gate mmu.level_shift[2] = 30; 5127c478bd9Sstevel@tonic-gate mmu.level_shift[3] = 39; 5137c478bd9Sstevel@tonic-gate 5147c478bd9Sstevel@tonic-gate #elif defined(__i386) 5157c478bd9Sstevel@tonic-gate 5167c478bd9Sstevel@tonic-gate if (mmu.pae_hat) { 5177c478bd9Sstevel@tonic-gate mmu.num_level = 3; 5187c478bd9Sstevel@tonic-gate mmu.max_level = 2; 5197c478bd9Sstevel@tonic-gate mmu.ptes_per_table = 512; 5207c478bd9Sstevel@tonic-gate mmu.top_level_count = 4; 5217c478bd9Sstevel@tonic-gate 5227c478bd9Sstevel@tonic-gate mmu.level_shift[0] = 12; 5237c478bd9Sstevel@tonic-gate mmu.level_shift[1] = 21; 5247c478bd9Sstevel@tonic-gate mmu.level_shift[2] = 30; 5257c478bd9Sstevel@tonic-gate 5267c478bd9Sstevel@tonic-gate } else { 5277c478bd9Sstevel@tonic-gate mmu.num_level = 2; 5287c478bd9Sstevel@tonic-gate mmu.max_level = 1; 5297c478bd9Sstevel@tonic-gate mmu.ptes_per_table = 1024; 5307c478bd9Sstevel@tonic-gate mmu.top_level_count = 1024; 5317c478bd9Sstevel@tonic-gate 5327c478bd9Sstevel@tonic-gate mmu.level_shift[0] = 12; 5337c478bd9Sstevel@tonic-gate mmu.level_shift[1] = 22; 5347c478bd9Sstevel@tonic-gate } 5357c478bd9Sstevel@tonic-gate 5367c478bd9Sstevel@tonic-gate #endif /* __i386 */ 5377c478bd9Sstevel@tonic-gate 5387c478bd9Sstevel@tonic-gate for (i = 0; i < mmu.num_level; ++i) { 5397c478bd9Sstevel@tonic-gate mmu.level_size[i] = 1UL << mmu.level_shift[i]; 5407c478bd9Sstevel@tonic-gate mmu.level_offset[i] = mmu.level_size[i] - 1; 5417c478bd9Sstevel@tonic-gate mmu.level_mask[i] = ~mmu.level_offset[i]; 5427c478bd9Sstevel@tonic-gate } 5437c478bd9Sstevel@tonic-gate 544ae115bc7Smrj for (i = 0; i <= mmu.max_page_level; ++i) { 545ae115bc7Smrj mmu.pte_bits[i] = PT_VALID; 546ae115bc7Smrj if (i > 0) 547ae115bc7Smrj mmu.pte_bits[i] |= PT_PAGESIZE; 548ae115bc7Smrj } 5497c478bd9Sstevel@tonic-gate 5507c478bd9Sstevel@tonic-gate /* 5517c478bd9Sstevel@tonic-gate * NOTE Legacy 32 bit PAE mode only has the P_VALID bit at top level. 5527c478bd9Sstevel@tonic-gate */ 5537c478bd9Sstevel@tonic-gate for (i = 1; i < mmu.num_level; ++i) 5547c478bd9Sstevel@tonic-gate mmu.ptp_bits[i] = PT_PTPBITS; 555ae115bc7Smrj 5567c478bd9Sstevel@tonic-gate #if defined(__i386) 5577c478bd9Sstevel@tonic-gate mmu.ptp_bits[2] = PT_VALID; 5587c478bd9Sstevel@tonic-gate #endif 5597c478bd9Sstevel@tonic-gate 5607c478bd9Sstevel@tonic-gate /* 5617c478bd9Sstevel@tonic-gate * Compute how many hash table entries to have per process for htables. 5627c478bd9Sstevel@tonic-gate * We start with 1 page's worth of entries. 5637c478bd9Sstevel@tonic-gate * 5647c478bd9Sstevel@tonic-gate * If physical memory is small, reduce the amount need to cover it. 5657c478bd9Sstevel@tonic-gate */ 5667c478bd9Sstevel@tonic-gate max_htables = physmax / mmu.ptes_per_table; 5677c478bd9Sstevel@tonic-gate mmu.hash_cnt = MMU_PAGESIZE / sizeof (htable_t *); 5687c478bd9Sstevel@tonic-gate while (mmu.hash_cnt > 16 && mmu.hash_cnt >= max_htables) 5697c478bd9Sstevel@tonic-gate mmu.hash_cnt >>= 1; 5707c478bd9Sstevel@tonic-gate mmu.vlp_hash_cnt = mmu.hash_cnt; 5717c478bd9Sstevel@tonic-gate 5727c478bd9Sstevel@tonic-gate #if defined(__amd64) 5737c478bd9Sstevel@tonic-gate /* 5747c478bd9Sstevel@tonic-gate * If running in 64 bits and physical memory is large, 5757c478bd9Sstevel@tonic-gate * increase the size of the cache to cover all of memory for 5767c478bd9Sstevel@tonic-gate * a 64 bit process. 5777c478bd9Sstevel@tonic-gate */ 5787c478bd9Sstevel@tonic-gate #define HASH_MAX_LENGTH 4 5797c478bd9Sstevel@tonic-gate while (mmu.hash_cnt * HASH_MAX_LENGTH < max_htables) 5807c478bd9Sstevel@tonic-gate mmu.hash_cnt <<= 1; 5817c478bd9Sstevel@tonic-gate #endif 5827c478bd9Sstevel@tonic-gate } 5837c478bd9Sstevel@tonic-gate 5847c478bd9Sstevel@tonic-gate 5857c478bd9Sstevel@tonic-gate /* 5867c478bd9Sstevel@tonic-gate * initialize hat data structures 5877c478bd9Sstevel@tonic-gate */ 5887c478bd9Sstevel@tonic-gate void 5897c478bd9Sstevel@tonic-gate hat_init() 5907c478bd9Sstevel@tonic-gate { 5917c478bd9Sstevel@tonic-gate #if defined(__i386) 5927c478bd9Sstevel@tonic-gate /* 5937c478bd9Sstevel@tonic-gate * _userlimit must be aligned correctly 5947c478bd9Sstevel@tonic-gate */ 5957c478bd9Sstevel@tonic-gate if ((_userlimit & LEVEL_MASK(1)) != _userlimit) { 5967c478bd9Sstevel@tonic-gate prom_printf("hat_init(): _userlimit=%p, not aligned at %p\n", 5977c478bd9Sstevel@tonic-gate (void *)_userlimit, (void *)LEVEL_SIZE(1)); 5987c478bd9Sstevel@tonic-gate halt("hat_init(): Unable to continue"); 5997c478bd9Sstevel@tonic-gate } 6007c478bd9Sstevel@tonic-gate #endif 6017c478bd9Sstevel@tonic-gate 6027c478bd9Sstevel@tonic-gate cv_init(&hat_list_cv, NULL, CV_DEFAULT, NULL); 6037c478bd9Sstevel@tonic-gate 6047c478bd9Sstevel@tonic-gate /* 6057c478bd9Sstevel@tonic-gate * initialize kmem caches 6067c478bd9Sstevel@tonic-gate */ 6077c478bd9Sstevel@tonic-gate htable_init(); 6087c478bd9Sstevel@tonic-gate hment_init(); 6097c478bd9Sstevel@tonic-gate 6107c478bd9Sstevel@tonic-gate hat_cache = kmem_cache_create("hat_t", 6117c478bd9Sstevel@tonic-gate sizeof (hat_t), 0, hati_constructor, NULL, NULL, 6127c478bd9Sstevel@tonic-gate NULL, 0, 0); 6137c478bd9Sstevel@tonic-gate 6147c478bd9Sstevel@tonic-gate hat_hash_cache = kmem_cache_create("HatHash", 6157c478bd9Sstevel@tonic-gate mmu.hash_cnt * sizeof (htable_t *), 0, NULL, NULL, NULL, 6167c478bd9Sstevel@tonic-gate NULL, 0, 0); 6177c478bd9Sstevel@tonic-gate 6187c478bd9Sstevel@tonic-gate /* 6197c478bd9Sstevel@tonic-gate * VLP hats can use a smaller hash table size on large memroy machines 6207c478bd9Sstevel@tonic-gate */ 6217c478bd9Sstevel@tonic-gate if (mmu.hash_cnt == mmu.vlp_hash_cnt) { 6227c478bd9Sstevel@tonic-gate vlp_hash_cache = hat_hash_cache; 6237c478bd9Sstevel@tonic-gate } else { 6247c478bd9Sstevel@tonic-gate vlp_hash_cache = kmem_cache_create("HatVlpHash", 6257c478bd9Sstevel@tonic-gate mmu.vlp_hash_cnt * sizeof (htable_t *), 0, NULL, NULL, NULL, 6267c478bd9Sstevel@tonic-gate NULL, 0, 0); 6277c478bd9Sstevel@tonic-gate } 6287c478bd9Sstevel@tonic-gate 6297c478bd9Sstevel@tonic-gate /* 6307c478bd9Sstevel@tonic-gate * Set up the kernel's hat 6317c478bd9Sstevel@tonic-gate */ 6327c478bd9Sstevel@tonic-gate AS_LOCK_ENTER(&kas, &kas.a_lock, RW_WRITER); 6337c478bd9Sstevel@tonic-gate kas.a_hat = kmem_cache_alloc(hat_cache, KM_NOSLEEP); 6347c478bd9Sstevel@tonic-gate mutex_init(&kas.a_hat->hat_mutex, NULL, MUTEX_DEFAULT, NULL); 6357c478bd9Sstevel@tonic-gate kas.a_hat->hat_as = &kas; 6367c478bd9Sstevel@tonic-gate kas.a_hat->hat_flags = 0; 6377c478bd9Sstevel@tonic-gate AS_LOCK_EXIT(&kas, &kas.a_lock); 6387c478bd9Sstevel@tonic-gate 6397c478bd9Sstevel@tonic-gate CPUSET_ZERO(khat_cpuset); 6407c478bd9Sstevel@tonic-gate CPUSET_ADD(khat_cpuset, CPU->cpu_id); 6417c478bd9Sstevel@tonic-gate 6427c478bd9Sstevel@tonic-gate /* 6437c478bd9Sstevel@tonic-gate * The kernel hat's next pointer serves as the head of the hat list . 644a85a6733Sjosephb * The kernel hat's prev pointer tracks the last hat on the list for 645a85a6733Sjosephb * htable_steal() to use. 6467c478bd9Sstevel@tonic-gate */ 6477c478bd9Sstevel@tonic-gate kas.a_hat->hat_next = NULL; 648a85a6733Sjosephb kas.a_hat->hat_prev = NULL; 6497c478bd9Sstevel@tonic-gate 6507c478bd9Sstevel@tonic-gate /* 6517c478bd9Sstevel@tonic-gate * Allocate an htable hash bucket for the kernel 6527c478bd9Sstevel@tonic-gate * XX64 - tune for 64 bit procs 6537c478bd9Sstevel@tonic-gate */ 6547c478bd9Sstevel@tonic-gate kas.a_hat->hat_num_hash = mmu.hash_cnt; 6557c478bd9Sstevel@tonic-gate kas.a_hat->hat_ht_hash = kmem_cache_alloc(hat_hash_cache, KM_NOSLEEP); 6567c478bd9Sstevel@tonic-gate bzero(kas.a_hat->hat_ht_hash, mmu.hash_cnt * sizeof (htable_t *)); 6577c478bd9Sstevel@tonic-gate 6587c478bd9Sstevel@tonic-gate /* 6597c478bd9Sstevel@tonic-gate * zero out the top level and cached htable pointers 6607c478bd9Sstevel@tonic-gate */ 6617c478bd9Sstevel@tonic-gate kas.a_hat->hat_ht_cached = NULL; 6627c478bd9Sstevel@tonic-gate kas.a_hat->hat_htable = NULL; 6639d9461f9Strevtom 6649d9461f9Strevtom /* 6659d9461f9Strevtom * Pre-allocate hrm_hashtab before enabling the collection of 6669d9461f9Strevtom * refmod statistics. Allocating on the fly would mean us 6679d9461f9Strevtom * running the risk of suffering recursive mutex enters or 6689d9461f9Strevtom * deadlocks. 6699d9461f9Strevtom */ 6709d9461f9Strevtom hrm_hashtab = kmem_zalloc(HRM_HASHSIZE * sizeof (struct hrmstat *), 6719d9461f9Strevtom KM_SLEEP); 6727c478bd9Sstevel@tonic-gate } 6737c478bd9Sstevel@tonic-gate 6747c478bd9Sstevel@tonic-gate /* 6757c478bd9Sstevel@tonic-gate * Prepare CPU specific pagetables for VLP processes on 64 bit kernels. 6767c478bd9Sstevel@tonic-gate * 6777c478bd9Sstevel@tonic-gate * Each CPU has a set of 2 pagetables that are reused for any 32 bit 6787c478bd9Sstevel@tonic-gate * process it runs. They are the top level pagetable, hci_vlp_l3ptes, and 6797c478bd9Sstevel@tonic-gate * the next to top level table for the bottom 512 Gig, hci_vlp_l2ptes. 6807c478bd9Sstevel@tonic-gate */ 6817c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 6827c478bd9Sstevel@tonic-gate static void 6837c478bd9Sstevel@tonic-gate hat_vlp_setup(struct cpu *cpu) 6847c478bd9Sstevel@tonic-gate { 6857c478bd9Sstevel@tonic-gate #if defined(__amd64) 6867c478bd9Sstevel@tonic-gate struct hat_cpu_info *hci = cpu->cpu_hat_info; 6877c478bd9Sstevel@tonic-gate pfn_t pfn; 6887c478bd9Sstevel@tonic-gate 6897c478bd9Sstevel@tonic-gate /* 6907c478bd9Sstevel@tonic-gate * allocate the level==2 page table for the bottom most 6917c478bd9Sstevel@tonic-gate * 512Gig of address space (this is where 32 bit apps live) 6927c478bd9Sstevel@tonic-gate */ 6937c478bd9Sstevel@tonic-gate ASSERT(hci != NULL); 6947c478bd9Sstevel@tonic-gate hci->hci_vlp_l2ptes = kmem_zalloc(MMU_PAGESIZE, KM_SLEEP); 6957c478bd9Sstevel@tonic-gate 6967c478bd9Sstevel@tonic-gate /* 6977c478bd9Sstevel@tonic-gate * Allocate a top level pagetable and copy the kernel's 6987c478bd9Sstevel@tonic-gate * entries into it. Then link in hci_vlp_l2ptes in the 1st entry. 6997c478bd9Sstevel@tonic-gate */ 7007c478bd9Sstevel@tonic-gate hci->hci_vlp_l3ptes = kmem_zalloc(MMU_PAGESIZE, KM_SLEEP); 7017c478bd9Sstevel@tonic-gate hci->hci_vlp_pfn = 7027c478bd9Sstevel@tonic-gate hat_getpfnum(kas.a_hat, (caddr_t)hci->hci_vlp_l3ptes); 7037c478bd9Sstevel@tonic-gate ASSERT(hci->hci_vlp_pfn != PFN_INVALID); 7047c478bd9Sstevel@tonic-gate bcopy(vlp_page + khat_start, hci->hci_vlp_l3ptes + khat_start, 7057c478bd9Sstevel@tonic-gate khat_entries * sizeof (x86pte_t)); 7067c478bd9Sstevel@tonic-gate 7077c478bd9Sstevel@tonic-gate pfn = hat_getpfnum(kas.a_hat, (caddr_t)hci->hci_vlp_l2ptes); 7087c478bd9Sstevel@tonic-gate ASSERT(pfn != PFN_INVALID); 7097c478bd9Sstevel@tonic-gate hci->hci_vlp_l3ptes[0] = MAKEPTP(pfn, 2); 7107c478bd9Sstevel@tonic-gate #endif /* __amd64 */ 7117c478bd9Sstevel@tonic-gate } 7127c478bd9Sstevel@tonic-gate 713ae115bc7Smrj /*ARGSUSED*/ 714ae115bc7Smrj static void 715ae115bc7Smrj hat_vlp_teardown(cpu_t *cpu) 716ae115bc7Smrj { 717ae115bc7Smrj #if defined(__amd64) 718ae115bc7Smrj struct hat_cpu_info *hci; 719ae115bc7Smrj 720ae115bc7Smrj if ((hci = cpu->cpu_hat_info) == NULL) 721ae115bc7Smrj return; 722ae115bc7Smrj if (hci->hci_vlp_l2ptes) 723ae115bc7Smrj kmem_free(hci->hci_vlp_l2ptes, MMU_PAGESIZE); 724ae115bc7Smrj if (hci->hci_vlp_l3ptes) 725ae115bc7Smrj kmem_free(hci->hci_vlp_l3ptes, MMU_PAGESIZE); 726ae115bc7Smrj #endif /* __amd64 */ 727ae115bc7Smrj } 728ae115bc7Smrj 7297c478bd9Sstevel@tonic-gate /* 7307c478bd9Sstevel@tonic-gate * Finish filling in the kernel hat. 7317c478bd9Sstevel@tonic-gate * Pre fill in all top level kernel page table entries for the kernel's 7327c478bd9Sstevel@tonic-gate * part of the address range. From this point on we can't use any new 7337c478bd9Sstevel@tonic-gate * kernel large pages if they need PTE's at max_level 734ae115bc7Smrj * 735ae115bc7Smrj * create the kmap mappings. 7367c478bd9Sstevel@tonic-gate */ 7377c478bd9Sstevel@tonic-gate void 7387c478bd9Sstevel@tonic-gate hat_init_finish(void) 7397c478bd9Sstevel@tonic-gate { 7407c478bd9Sstevel@tonic-gate htable_t *top = kas.a_hat->hat_htable; 7417c478bd9Sstevel@tonic-gate htable_t *ht; 7427c478bd9Sstevel@tonic-gate uint_t e; 7437c478bd9Sstevel@tonic-gate x86pte_t pte; 7447c478bd9Sstevel@tonic-gate uintptr_t va = kernelbase; 745ae115bc7Smrj size_t size; 7467c478bd9Sstevel@tonic-gate 7477c478bd9Sstevel@tonic-gate 7487c478bd9Sstevel@tonic-gate #if defined(__i386) 7497c478bd9Sstevel@tonic-gate ASSERT((va & LEVEL_MASK(1)) == va); 7507c478bd9Sstevel@tonic-gate 7517c478bd9Sstevel@tonic-gate /* 7527c478bd9Sstevel@tonic-gate * Deal with kernelbase not 1Gig aligned for 32 bit PAE hats. 7537c478bd9Sstevel@tonic-gate */ 7547c478bd9Sstevel@tonic-gate if (!mmu.pae_hat || (va & LEVEL_OFFSET(mmu.max_level)) == 0) { 7557c478bd9Sstevel@tonic-gate khat_pae32_htable = NULL; 7567c478bd9Sstevel@tonic-gate } else { 7577c478bd9Sstevel@tonic-gate ASSERT(mmu.max_level == 2); 7587c478bd9Sstevel@tonic-gate ASSERT((va & LEVEL_OFFSET(mmu.max_level - 1)) == 0); 7597c478bd9Sstevel@tonic-gate khat_pae32_htable = 7607c478bd9Sstevel@tonic-gate htable_create(kas.a_hat, va, mmu.max_level - 1, NULL); 7617c478bd9Sstevel@tonic-gate khat_pae32_start = htable_va2entry(va, khat_pae32_htable); 7627c478bd9Sstevel@tonic-gate khat_pae32_entries = mmu.ptes_per_table - khat_pae32_start; 7637c478bd9Sstevel@tonic-gate for (e = khat_pae32_start; e < mmu.ptes_per_table; 7647c478bd9Sstevel@tonic-gate ++e, va += LEVEL_SIZE(mmu.max_level - 1)) { 7657c478bd9Sstevel@tonic-gate pte = x86pte_get(khat_pae32_htable, e); 7667c478bd9Sstevel@tonic-gate if (PTE_ISVALID(pte)) 7677c478bd9Sstevel@tonic-gate continue; 7687c478bd9Sstevel@tonic-gate ht = htable_create(kas.a_hat, va, mmu.max_level - 2, 7697c478bd9Sstevel@tonic-gate NULL); 7707c478bd9Sstevel@tonic-gate ASSERT(ht != NULL); 7717c478bd9Sstevel@tonic-gate } 7727c478bd9Sstevel@tonic-gate } 7737c478bd9Sstevel@tonic-gate #endif 7747c478bd9Sstevel@tonic-gate 7757c478bd9Sstevel@tonic-gate /* 7767c478bd9Sstevel@tonic-gate * The kernel hat will need fixed values in the highest level 7777c478bd9Sstevel@tonic-gate * ptable for copying to all other hat's. This implies 7787c478bd9Sstevel@tonic-gate * alignment restrictions on _userlimit. 7797c478bd9Sstevel@tonic-gate * 7807c478bd9Sstevel@tonic-gate * Note we don't htable_release() these htables. This keeps them 7817c478bd9Sstevel@tonic-gate * from ever being stolen or free'd. 7827c478bd9Sstevel@tonic-gate * 7837c478bd9Sstevel@tonic-gate * top_level_count is used instead of ptes_per_table, since 7847c478bd9Sstevel@tonic-gate * on 32-bit PAE we only have 4 usable entries at the top level ptable. 7857c478bd9Sstevel@tonic-gate */ 7867c478bd9Sstevel@tonic-gate if (va == 0) 7877c478bd9Sstevel@tonic-gate khat_start = mmu.top_level_count; 7887c478bd9Sstevel@tonic-gate else 7897c478bd9Sstevel@tonic-gate khat_start = htable_va2entry(va, kas.a_hat->hat_htable); 7907c478bd9Sstevel@tonic-gate khat_entries = mmu.top_level_count - khat_start; 7917c478bd9Sstevel@tonic-gate for (e = khat_start; e < mmu.top_level_count; 7927c478bd9Sstevel@tonic-gate ++e, va += LEVEL_SIZE(mmu.max_level)) { 793ae115bc7Smrj if (IN_HYPERVISOR_VA(va)) 794ae115bc7Smrj continue; 7957c478bd9Sstevel@tonic-gate pte = x86pte_get(top, e); 7967c478bd9Sstevel@tonic-gate if (PTE_ISVALID(pte)) 7977c478bd9Sstevel@tonic-gate continue; 7987c478bd9Sstevel@tonic-gate ht = htable_create(kas.a_hat, va, mmu.max_level - 1, NULL); 7997c478bd9Sstevel@tonic-gate ASSERT(ht != NULL); 8007c478bd9Sstevel@tonic-gate } 8017c478bd9Sstevel@tonic-gate 8027c478bd9Sstevel@tonic-gate /* 8037c478bd9Sstevel@tonic-gate * We are now effectively running on the kernel hat. 8047c478bd9Sstevel@tonic-gate * Clearing use_boot_reserve shuts off using the pre-allocated boot 8057c478bd9Sstevel@tonic-gate * reserve for all HAT allocations. From here on, the reserves are 8067c478bd9Sstevel@tonic-gate * only used when mapping in memory for the hat's own allocations. 8077c478bd9Sstevel@tonic-gate */ 8087c478bd9Sstevel@tonic-gate use_boot_reserve = 0; 8097c478bd9Sstevel@tonic-gate htable_adjust_reserve(); 8107c478bd9Sstevel@tonic-gate 8117c478bd9Sstevel@tonic-gate /* 8127c478bd9Sstevel@tonic-gate * 32 bit kernels use only 4 of the 512 entries in its top level 8137c478bd9Sstevel@tonic-gate * pagetable. We'll use the remainder for the "per CPU" page tables 8147c478bd9Sstevel@tonic-gate * for VLP processes. 8157c478bd9Sstevel@tonic-gate * 816ae115bc7Smrj * We also map the top level kernel pagetable into the kernel to make 817ae115bc7Smrj * it easy to use bcopy to initialize new address spaces. 8187c478bd9Sstevel@tonic-gate */ 8197c478bd9Sstevel@tonic-gate if (mmu.pae_hat) { 8207c478bd9Sstevel@tonic-gate vlp_page = vmem_alloc(heap_arena, MMU_PAGESIZE, VM_SLEEP); 8217c478bd9Sstevel@tonic-gate hat_devload(kas.a_hat, (caddr_t)vlp_page, MMU_PAGESIZE, 8227c478bd9Sstevel@tonic-gate kas.a_hat->hat_htable->ht_pfn, 823ae115bc7Smrj PROT_WRITE | 824ae115bc7Smrj PROT_READ | HAT_NOSYNC | HAT_UNORDERED_OK, 8257c478bd9Sstevel@tonic-gate HAT_LOAD | HAT_LOAD_NOCONSIST); 8267c478bd9Sstevel@tonic-gate } 8277c478bd9Sstevel@tonic-gate hat_vlp_setup(CPU); 828ae115bc7Smrj 829ae115bc7Smrj /* 830ae115bc7Smrj * Create kmap (cached mappings of kernel PTEs) 831ae115bc7Smrj * for 32 bit we map from segmap_start .. ekernelheap 832ae115bc7Smrj * for 64 bit we map from segmap_start .. segmap_start + segmapsize; 833ae115bc7Smrj */ 834ae115bc7Smrj #if defined(__i386) 835ae115bc7Smrj size = (uintptr_t)ekernelheap - segmap_start; 836ae115bc7Smrj #elif defined(__amd64) 837ae115bc7Smrj size = segmapsize; 838ae115bc7Smrj #endif 839ae115bc7Smrj hat_kmap_init((uintptr_t)segmap_start, size); 8407c478bd9Sstevel@tonic-gate } 8417c478bd9Sstevel@tonic-gate 8427c478bd9Sstevel@tonic-gate /* 8437c478bd9Sstevel@tonic-gate * On 32 bit PAE mode, PTE's are 64 bits, but ordinary atomic memory references 8447c478bd9Sstevel@tonic-gate * are 32 bit, so for safety we must use cas64() to install these. 8457c478bd9Sstevel@tonic-gate */ 8467c478bd9Sstevel@tonic-gate #ifdef __i386 8477c478bd9Sstevel@tonic-gate static void 8487c478bd9Sstevel@tonic-gate reload_pae32(hat_t *hat, cpu_t *cpu) 8497c478bd9Sstevel@tonic-gate { 8507c478bd9Sstevel@tonic-gate x86pte_t *src; 8517c478bd9Sstevel@tonic-gate x86pte_t *dest; 8527c478bd9Sstevel@tonic-gate x86pte_t pte; 8537c478bd9Sstevel@tonic-gate int i; 8547c478bd9Sstevel@tonic-gate 8557c478bd9Sstevel@tonic-gate /* 8567c478bd9Sstevel@tonic-gate * Load the 4 entries of the level 2 page table into this 8577c478bd9Sstevel@tonic-gate * cpu's range of the vlp_page and point cr3 at them. 8587c478bd9Sstevel@tonic-gate */ 8597c478bd9Sstevel@tonic-gate ASSERT(mmu.pae_hat); 8607c478bd9Sstevel@tonic-gate src = hat->hat_vlp_ptes; 8617c478bd9Sstevel@tonic-gate dest = vlp_page + (cpu->cpu_id + 1) * VLP_NUM_PTES; 8627c478bd9Sstevel@tonic-gate for (i = 0; i < VLP_NUM_PTES; ++i) { 8637c478bd9Sstevel@tonic-gate for (;;) { 8647c478bd9Sstevel@tonic-gate pte = dest[i]; 8657c478bd9Sstevel@tonic-gate if (pte == src[i]) 8667c478bd9Sstevel@tonic-gate break; 8677c478bd9Sstevel@tonic-gate if (cas64(dest + i, pte, src[i]) != src[i]) 8687c478bd9Sstevel@tonic-gate break; 8697c478bd9Sstevel@tonic-gate } 8707c478bd9Sstevel@tonic-gate } 8717c478bd9Sstevel@tonic-gate } 8727c478bd9Sstevel@tonic-gate #endif 8737c478bd9Sstevel@tonic-gate 8747c478bd9Sstevel@tonic-gate /* 8757c478bd9Sstevel@tonic-gate * Switch to a new active hat, maintaining bit masks to track active CPUs. 8767c478bd9Sstevel@tonic-gate */ 8777c478bd9Sstevel@tonic-gate void 8787c478bd9Sstevel@tonic-gate hat_switch(hat_t *hat) 8797c478bd9Sstevel@tonic-gate { 8807c478bd9Sstevel@tonic-gate uintptr_t newcr3; 8817c478bd9Sstevel@tonic-gate cpu_t *cpu = CPU; 8827c478bd9Sstevel@tonic-gate hat_t *old = cpu->cpu_current_hat; 8837c478bd9Sstevel@tonic-gate 8847c478bd9Sstevel@tonic-gate /* 8857c478bd9Sstevel@tonic-gate * set up this information first, so we don't miss any cross calls 8867c478bd9Sstevel@tonic-gate */ 8877c478bd9Sstevel@tonic-gate if (old != NULL) { 8887c478bd9Sstevel@tonic-gate if (old == hat) 8897c478bd9Sstevel@tonic-gate return; 8907c478bd9Sstevel@tonic-gate if (old != kas.a_hat) 8917c478bd9Sstevel@tonic-gate CPUSET_ATOMIC_DEL(old->hat_cpus, cpu->cpu_id); 8927c478bd9Sstevel@tonic-gate } 8937c478bd9Sstevel@tonic-gate 8947c478bd9Sstevel@tonic-gate /* 895*95c0a3c8Sjosephb * Add this CPU to the active set for this HAT. 8967c478bd9Sstevel@tonic-gate */ 8977c478bd9Sstevel@tonic-gate if (hat != kas.a_hat) { 8987c478bd9Sstevel@tonic-gate CPUSET_ATOMIC_ADD(hat->hat_cpus, cpu->cpu_id); 8997c478bd9Sstevel@tonic-gate } 9007c478bd9Sstevel@tonic-gate cpu->cpu_current_hat = hat; 9017c478bd9Sstevel@tonic-gate 9027c478bd9Sstevel@tonic-gate /* 9037c478bd9Sstevel@tonic-gate * now go ahead and load cr3 9047c478bd9Sstevel@tonic-gate */ 9057c478bd9Sstevel@tonic-gate if (hat->hat_flags & HAT_VLP) { 9067c478bd9Sstevel@tonic-gate #if defined(__amd64) 9077c478bd9Sstevel@tonic-gate x86pte_t *vlpptep = cpu->cpu_hat_info->hci_vlp_l2ptes; 9087c478bd9Sstevel@tonic-gate 9097c478bd9Sstevel@tonic-gate VLP_COPY(hat->hat_vlp_ptes, vlpptep); 9107c478bd9Sstevel@tonic-gate newcr3 = MAKECR3(cpu->cpu_hat_info->hci_vlp_pfn); 9117c478bd9Sstevel@tonic-gate #elif defined(__i386) 9127c478bd9Sstevel@tonic-gate reload_pae32(hat, cpu); 9137c478bd9Sstevel@tonic-gate newcr3 = MAKECR3(kas.a_hat->hat_htable->ht_pfn) + 9147c478bd9Sstevel@tonic-gate (cpu->cpu_id + 1) * VLP_SIZE; 9157c478bd9Sstevel@tonic-gate #endif 9167c478bd9Sstevel@tonic-gate } else { 9177c478bd9Sstevel@tonic-gate newcr3 = MAKECR3(hat->hat_htable->ht_pfn); 9187c478bd9Sstevel@tonic-gate } 9197c478bd9Sstevel@tonic-gate setcr3(newcr3); 9207c478bd9Sstevel@tonic-gate ASSERT(cpu == CPU); 9217c478bd9Sstevel@tonic-gate } 9227c478bd9Sstevel@tonic-gate 9237c478bd9Sstevel@tonic-gate /* 9247c478bd9Sstevel@tonic-gate * Utility to return a valid x86pte_t from protections, pfn, and level number 9257c478bd9Sstevel@tonic-gate */ 9267c478bd9Sstevel@tonic-gate static x86pte_t 9277c478bd9Sstevel@tonic-gate hati_mkpte(pfn_t pfn, uint_t attr, level_t level, uint_t flags) 9287c478bd9Sstevel@tonic-gate { 9297c478bd9Sstevel@tonic-gate x86pte_t pte; 9307c478bd9Sstevel@tonic-gate uint_t cache_attr = attr & HAT_ORDER_MASK; 9317c478bd9Sstevel@tonic-gate 9327c478bd9Sstevel@tonic-gate pte = MAKEPTE(pfn, level); 9337c478bd9Sstevel@tonic-gate 9347c478bd9Sstevel@tonic-gate if (attr & PROT_WRITE) 9357c478bd9Sstevel@tonic-gate PTE_SET(pte, PT_WRITABLE); 9367c478bd9Sstevel@tonic-gate 9377c478bd9Sstevel@tonic-gate if (attr & PROT_USER) 9387c478bd9Sstevel@tonic-gate PTE_SET(pte, PT_USER); 9397c478bd9Sstevel@tonic-gate 9407c478bd9Sstevel@tonic-gate if (!(attr & PROT_EXEC)) 9417c478bd9Sstevel@tonic-gate PTE_SET(pte, mmu.pt_nx); 9427c478bd9Sstevel@tonic-gate 9437c478bd9Sstevel@tonic-gate /* 944ae115bc7Smrj * Set the software bits used track ref/mod sync's and hments. 945ae115bc7Smrj * If not using REF/MOD, set them to avoid h/w rewriting PTEs. 9467c478bd9Sstevel@tonic-gate */ 9477c478bd9Sstevel@tonic-gate if (flags & HAT_LOAD_NOCONSIST) 948ae115bc7Smrj PTE_SET(pte, PT_NOCONSIST | PT_REF | PT_MOD); 949ae115bc7Smrj else if (attr & HAT_NOSYNC) 950ae115bc7Smrj PTE_SET(pte, PT_NOSYNC | PT_REF | PT_MOD); 9517c478bd9Sstevel@tonic-gate 9527c478bd9Sstevel@tonic-gate /* 9537c478bd9Sstevel@tonic-gate * Set the caching attributes in the PTE. The combination 9547c478bd9Sstevel@tonic-gate * of attributes are poorly defined, so we pay attention 9557c478bd9Sstevel@tonic-gate * to them in the given order. 9567c478bd9Sstevel@tonic-gate * 9577c478bd9Sstevel@tonic-gate * The test for HAT_STRICTORDER is different because it's defined 9587c478bd9Sstevel@tonic-gate * as "0" - which was a stupid thing to do, but is too late to change! 9597c478bd9Sstevel@tonic-gate */ 9607c478bd9Sstevel@tonic-gate if (cache_attr == HAT_STRICTORDER) { 9617c478bd9Sstevel@tonic-gate PTE_SET(pte, PT_NOCACHE); 9627c478bd9Sstevel@tonic-gate /*LINTED [Lint hates empty ifs, but it's the obvious way to do this] */ 9637c478bd9Sstevel@tonic-gate } else if (cache_attr & (HAT_UNORDERED_OK | HAT_STORECACHING_OK)) { 9647c478bd9Sstevel@tonic-gate /* nothing to set */; 9657c478bd9Sstevel@tonic-gate } else if (cache_attr & (HAT_MERGING_OK | HAT_LOADCACHING_OK)) { 9667c478bd9Sstevel@tonic-gate PTE_SET(pte, PT_NOCACHE); 9677c478bd9Sstevel@tonic-gate if (x86_feature & X86_PAT) 9687c478bd9Sstevel@tonic-gate PTE_SET(pte, (level == 0) ? PT_PAT_4K : PT_PAT_LARGE); 9697c478bd9Sstevel@tonic-gate else 9707c478bd9Sstevel@tonic-gate PTE_SET(pte, PT_WRITETHRU); 9717c478bd9Sstevel@tonic-gate } else { 9727c478bd9Sstevel@tonic-gate panic("hati_mkpte(): bad caching attributes: %x\n", cache_attr); 9737c478bd9Sstevel@tonic-gate } 9747c478bd9Sstevel@tonic-gate 9757c478bd9Sstevel@tonic-gate return (pte); 9767c478bd9Sstevel@tonic-gate } 9777c478bd9Sstevel@tonic-gate 9787c478bd9Sstevel@tonic-gate /* 9797c478bd9Sstevel@tonic-gate * Duplicate address translations of the parent to the child. 9807c478bd9Sstevel@tonic-gate * This function really isn't used anymore. 9817c478bd9Sstevel@tonic-gate */ 9827c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 9837c478bd9Sstevel@tonic-gate int 9847c478bd9Sstevel@tonic-gate hat_dup(hat_t *old, hat_t *new, caddr_t addr, size_t len, uint_t flag) 9857c478bd9Sstevel@tonic-gate { 9867c478bd9Sstevel@tonic-gate ASSERT((uintptr_t)addr < kernelbase); 9877c478bd9Sstevel@tonic-gate ASSERT(new != kas.a_hat); 9887c478bd9Sstevel@tonic-gate ASSERT(old != kas.a_hat); 9897c478bd9Sstevel@tonic-gate return (0); 9907c478bd9Sstevel@tonic-gate } 9917c478bd9Sstevel@tonic-gate 9927c478bd9Sstevel@tonic-gate /* 9937c478bd9Sstevel@tonic-gate * Allocate any hat resources required for a process being swapped in. 9947c478bd9Sstevel@tonic-gate */ 9957c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 9967c478bd9Sstevel@tonic-gate void 9977c478bd9Sstevel@tonic-gate hat_swapin(hat_t *hat) 9987c478bd9Sstevel@tonic-gate { 9997c478bd9Sstevel@tonic-gate /* do nothing - we let everything fault back in */ 10007c478bd9Sstevel@tonic-gate } 10017c478bd9Sstevel@tonic-gate 10027c478bd9Sstevel@tonic-gate /* 10037c478bd9Sstevel@tonic-gate * Unload all translations associated with an address space of a process 10047c478bd9Sstevel@tonic-gate * that is being swapped out. 10057c478bd9Sstevel@tonic-gate */ 10067c478bd9Sstevel@tonic-gate void 10077c478bd9Sstevel@tonic-gate hat_swapout(hat_t *hat) 10087c478bd9Sstevel@tonic-gate { 10097c478bd9Sstevel@tonic-gate uintptr_t vaddr = (uintptr_t)0; 10107c478bd9Sstevel@tonic-gate uintptr_t eaddr = _userlimit; 10117c478bd9Sstevel@tonic-gate htable_t *ht = NULL; 10127c478bd9Sstevel@tonic-gate level_t l; 10137c478bd9Sstevel@tonic-gate 10147c478bd9Sstevel@tonic-gate /* 10157c478bd9Sstevel@tonic-gate * We can't just call hat_unload(hat, 0, _userlimit...) here, because 10167c478bd9Sstevel@tonic-gate * seg_spt and shared pagetables can't be swapped out. 10177c478bd9Sstevel@tonic-gate * Take a look at segspt_shmswapout() - it's a big no-op. 10187c478bd9Sstevel@tonic-gate * 10197c478bd9Sstevel@tonic-gate * Instead we'll walk through all the address space and unload 10207c478bd9Sstevel@tonic-gate * any mappings which we are sure are not shared, not locked. 10217c478bd9Sstevel@tonic-gate */ 10227c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(vaddr)); 10237c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(eaddr)); 10247c478bd9Sstevel@tonic-gate ASSERT(AS_LOCK_HELD(hat->hat_as, &hat->hat_as->a_lock)); 10257c478bd9Sstevel@tonic-gate if ((uintptr_t)hat->hat_as->a_userlimit < eaddr) 10267c478bd9Sstevel@tonic-gate eaddr = (uintptr_t)hat->hat_as->a_userlimit; 10277c478bd9Sstevel@tonic-gate 10287c478bd9Sstevel@tonic-gate while (vaddr < eaddr) { 10297c478bd9Sstevel@tonic-gate (void) htable_walk(hat, &ht, &vaddr, eaddr); 10307c478bd9Sstevel@tonic-gate if (ht == NULL) 10317c478bd9Sstevel@tonic-gate break; 10327c478bd9Sstevel@tonic-gate 10337c478bd9Sstevel@tonic-gate ASSERT(!IN_VA_HOLE(vaddr)); 10347c478bd9Sstevel@tonic-gate 10357c478bd9Sstevel@tonic-gate /* 10367c478bd9Sstevel@tonic-gate * If the page table is shared skip its entire range. 10377c478bd9Sstevel@tonic-gate * This code knows that only level 0 page tables are shared 10387c478bd9Sstevel@tonic-gate */ 10397c478bd9Sstevel@tonic-gate l = ht->ht_level; 10407c478bd9Sstevel@tonic-gate if (ht->ht_flags & HTABLE_SHARED_PFN) { 10417c478bd9Sstevel@tonic-gate ASSERT(l == 0); 10427c478bd9Sstevel@tonic-gate vaddr = ht->ht_vaddr + LEVEL_SIZE(1); 10437c478bd9Sstevel@tonic-gate htable_release(ht); 10447c478bd9Sstevel@tonic-gate ht = NULL; 10457c478bd9Sstevel@tonic-gate continue; 10467c478bd9Sstevel@tonic-gate } 10477c478bd9Sstevel@tonic-gate 10487c478bd9Sstevel@tonic-gate /* 10497c478bd9Sstevel@tonic-gate * If the page table has no locked entries, unload this one. 10507c478bd9Sstevel@tonic-gate */ 10517c478bd9Sstevel@tonic-gate if (ht->ht_lock_cnt == 0) 10527c478bd9Sstevel@tonic-gate hat_unload(hat, (caddr_t)vaddr, LEVEL_SIZE(l), 10537c478bd9Sstevel@tonic-gate HAT_UNLOAD_UNMAP); 10547c478bd9Sstevel@tonic-gate 10557c478bd9Sstevel@tonic-gate /* 10567c478bd9Sstevel@tonic-gate * If we have a level 0 page table with locked entries, 10577c478bd9Sstevel@tonic-gate * skip the entire page table, otherwise skip just one entry. 10587c478bd9Sstevel@tonic-gate */ 10597c478bd9Sstevel@tonic-gate if (ht->ht_lock_cnt > 0 && l == 0) 10607c478bd9Sstevel@tonic-gate vaddr = ht->ht_vaddr + LEVEL_SIZE(1); 10617c478bd9Sstevel@tonic-gate else 10627c478bd9Sstevel@tonic-gate vaddr += LEVEL_SIZE(l); 10637c478bd9Sstevel@tonic-gate } 10647c478bd9Sstevel@tonic-gate if (ht) 10657c478bd9Sstevel@tonic-gate htable_release(ht); 10667c478bd9Sstevel@tonic-gate 10677c478bd9Sstevel@tonic-gate /* 10687c478bd9Sstevel@tonic-gate * We're in swapout because the system is low on memory, so 10697c478bd9Sstevel@tonic-gate * go back and flush all the htables off the cached list. 10707c478bd9Sstevel@tonic-gate */ 10717c478bd9Sstevel@tonic-gate htable_purge_hat(hat); 10727c478bd9Sstevel@tonic-gate } 10737c478bd9Sstevel@tonic-gate 10747c478bd9Sstevel@tonic-gate /* 10757c478bd9Sstevel@tonic-gate * returns number of bytes that have valid mappings in hat. 10767c478bd9Sstevel@tonic-gate */ 10777c478bd9Sstevel@tonic-gate size_t 10787c478bd9Sstevel@tonic-gate hat_get_mapped_size(hat_t *hat) 10797c478bd9Sstevel@tonic-gate { 10807c478bd9Sstevel@tonic-gate size_t total = 0; 10817c478bd9Sstevel@tonic-gate int l; 10827c478bd9Sstevel@tonic-gate 10837c478bd9Sstevel@tonic-gate for (l = 0; l <= mmu.max_page_level; l++) 10847c478bd9Sstevel@tonic-gate total += (hat->hat_pages_mapped[l] << LEVEL_SHIFT(l)); 10857c478bd9Sstevel@tonic-gate 10867c478bd9Sstevel@tonic-gate return (total); 10877c478bd9Sstevel@tonic-gate } 10887c478bd9Sstevel@tonic-gate 10897c478bd9Sstevel@tonic-gate /* 10907c478bd9Sstevel@tonic-gate * enable/disable collection of stats for hat. 10917c478bd9Sstevel@tonic-gate */ 10927c478bd9Sstevel@tonic-gate int 10937c478bd9Sstevel@tonic-gate hat_stats_enable(hat_t *hat) 10947c478bd9Sstevel@tonic-gate { 10957c478bd9Sstevel@tonic-gate atomic_add_32(&hat->hat_stats, 1); 10967c478bd9Sstevel@tonic-gate return (1); 10977c478bd9Sstevel@tonic-gate } 10987c478bd9Sstevel@tonic-gate 10997c478bd9Sstevel@tonic-gate void 11007c478bd9Sstevel@tonic-gate hat_stats_disable(hat_t *hat) 11017c478bd9Sstevel@tonic-gate { 11027c478bd9Sstevel@tonic-gate atomic_add_32(&hat->hat_stats, -1); 11037c478bd9Sstevel@tonic-gate } 11047c478bd9Sstevel@tonic-gate 11057c478bd9Sstevel@tonic-gate /* 11067c478bd9Sstevel@tonic-gate * Utility to sync the ref/mod bits from a page table entry to the page_t 11077c478bd9Sstevel@tonic-gate * We must be holding the mapping list lock when this is called. 11087c478bd9Sstevel@tonic-gate */ 11097c478bd9Sstevel@tonic-gate static void 11107c478bd9Sstevel@tonic-gate hati_sync_pte_to_page(page_t *pp, x86pte_t pte, level_t level) 11117c478bd9Sstevel@tonic-gate { 11127c478bd9Sstevel@tonic-gate uint_t rm = 0; 11137c478bd9Sstevel@tonic-gate pgcnt_t pgcnt; 11147c478bd9Sstevel@tonic-gate 1115ae115bc7Smrj if (PTE_GET(pte, PT_SOFTWARE) >= PT_NOSYNC) 11167c478bd9Sstevel@tonic-gate return; 11177c478bd9Sstevel@tonic-gate 11187c478bd9Sstevel@tonic-gate if (PTE_GET(pte, PT_REF)) 11197c478bd9Sstevel@tonic-gate rm |= P_REF; 11207c478bd9Sstevel@tonic-gate 11217c478bd9Sstevel@tonic-gate if (PTE_GET(pte, PT_MOD)) 11227c478bd9Sstevel@tonic-gate rm |= P_MOD; 11237c478bd9Sstevel@tonic-gate 11247c478bd9Sstevel@tonic-gate if (rm == 0) 11257c478bd9Sstevel@tonic-gate return; 11267c478bd9Sstevel@tonic-gate 11277c478bd9Sstevel@tonic-gate /* 11287c478bd9Sstevel@tonic-gate * sync to all constituent pages of a large page 11297c478bd9Sstevel@tonic-gate */ 11307c478bd9Sstevel@tonic-gate ASSERT(x86_hm_held(pp)); 11317c478bd9Sstevel@tonic-gate pgcnt = page_get_pagecnt(level); 11327c478bd9Sstevel@tonic-gate ASSERT(IS_P2ALIGNED(pp->p_pagenum, pgcnt)); 11337c478bd9Sstevel@tonic-gate for (; pgcnt > 0; --pgcnt) { 11347c478bd9Sstevel@tonic-gate /* 11357c478bd9Sstevel@tonic-gate * hat_page_demote() can't decrease 11367c478bd9Sstevel@tonic-gate * pszc below this mapping size 11377c478bd9Sstevel@tonic-gate * since this large mapping existed after we 11387c478bd9Sstevel@tonic-gate * took mlist lock. 11397c478bd9Sstevel@tonic-gate */ 11407c478bd9Sstevel@tonic-gate ASSERT(pp->p_szc >= level); 11417c478bd9Sstevel@tonic-gate hat_page_setattr(pp, rm); 11427c478bd9Sstevel@tonic-gate ++pp; 11437c478bd9Sstevel@tonic-gate } 11447c478bd9Sstevel@tonic-gate } 11457c478bd9Sstevel@tonic-gate 11467c478bd9Sstevel@tonic-gate /* 11477c478bd9Sstevel@tonic-gate * This the set of PTE bits for PFN, permissions and caching 1148ae115bc7Smrj * that require a TLB flush (hat_tlb_inval) if changed on a HAT_LOAD_REMAP 11497c478bd9Sstevel@tonic-gate */ 11507c478bd9Sstevel@tonic-gate #define PT_REMAP_BITS \ 11517c478bd9Sstevel@tonic-gate (PT_PADDR | PT_NX | PT_WRITABLE | PT_WRITETHRU | \ 11527c478bd9Sstevel@tonic-gate PT_NOCACHE | PT_PAT_4K | PT_PAT_LARGE) 11537c478bd9Sstevel@tonic-gate 1154b193e412Skchow #define REMAPASSERT(EX) if (!(EX)) panic("hati_pte_map: " #EX) 11557c478bd9Sstevel@tonic-gate /* 11567c478bd9Sstevel@tonic-gate * Do the low-level work to get a mapping entered into a HAT's pagetables 11577c478bd9Sstevel@tonic-gate * and in the mapping list of the associated page_t. 11587c478bd9Sstevel@tonic-gate */ 1159ae115bc7Smrj static int 11607c478bd9Sstevel@tonic-gate hati_pte_map( 11617c478bd9Sstevel@tonic-gate htable_t *ht, 11627c478bd9Sstevel@tonic-gate uint_t entry, 11637c478bd9Sstevel@tonic-gate page_t *pp, 11647c478bd9Sstevel@tonic-gate x86pte_t pte, 11657c478bd9Sstevel@tonic-gate int flags, 11667c478bd9Sstevel@tonic-gate void *pte_ptr) 11677c478bd9Sstevel@tonic-gate { 11687c478bd9Sstevel@tonic-gate hat_t *hat = ht->ht_hat; 11697c478bd9Sstevel@tonic-gate x86pte_t old_pte; 11707c478bd9Sstevel@tonic-gate level_t l = ht->ht_level; 11717c478bd9Sstevel@tonic-gate hment_t *hm; 11727c478bd9Sstevel@tonic-gate uint_t is_consist; 1173ae115bc7Smrj int rv = 0; 11747c478bd9Sstevel@tonic-gate 11757c478bd9Sstevel@tonic-gate /* 11767c478bd9Sstevel@tonic-gate * Is this a consistant (ie. need mapping list lock) mapping? 11777c478bd9Sstevel@tonic-gate */ 11787c478bd9Sstevel@tonic-gate is_consist = (pp != NULL && (flags & HAT_LOAD_NOCONSIST) == 0); 11797c478bd9Sstevel@tonic-gate 11807c478bd9Sstevel@tonic-gate /* 11817c478bd9Sstevel@tonic-gate * Track locked mapping count in the htable. Do this first, 11827c478bd9Sstevel@tonic-gate * as we track locking even if there already is a mapping present. 11837c478bd9Sstevel@tonic-gate */ 11847c478bd9Sstevel@tonic-gate if ((flags & HAT_LOAD_LOCK) != 0 && hat != kas.a_hat) 11857c478bd9Sstevel@tonic-gate HTABLE_LOCK_INC(ht); 11867c478bd9Sstevel@tonic-gate 11877c478bd9Sstevel@tonic-gate /* 11887c478bd9Sstevel@tonic-gate * Acquire the page's mapping list lock and get an hment to use. 11897c478bd9Sstevel@tonic-gate * Note that hment_prepare() might return NULL. 11907c478bd9Sstevel@tonic-gate */ 11917c478bd9Sstevel@tonic-gate if (is_consist) { 11927c478bd9Sstevel@tonic-gate x86_hm_enter(pp); 11937c478bd9Sstevel@tonic-gate hm = hment_prepare(ht, entry, pp); 11947c478bd9Sstevel@tonic-gate } 11957c478bd9Sstevel@tonic-gate 11967c478bd9Sstevel@tonic-gate /* 11977c478bd9Sstevel@tonic-gate * Set the new pte, retrieving the old one at the same time. 11987c478bd9Sstevel@tonic-gate */ 11997c478bd9Sstevel@tonic-gate old_pte = x86pte_set(ht, entry, pte, pte_ptr); 12007c478bd9Sstevel@tonic-gate 12017c478bd9Sstevel@tonic-gate /* 1202ae115bc7Smrj * did we get a large page / page table collision? 12037c478bd9Sstevel@tonic-gate */ 1204ae115bc7Smrj if (old_pte == LPAGE_ERROR) { 1205ae115bc7Smrj rv = -1; 1206ae115bc7Smrj goto done; 12077c478bd9Sstevel@tonic-gate } 12087c478bd9Sstevel@tonic-gate 1209ae115bc7Smrj /* 1210ae115bc7Smrj * If the mapping didn't change there is nothing more to do. 1211ae115bc7Smrj */ 1212ae115bc7Smrj if (PTE_EQUIV(pte, old_pte)) 1213ae115bc7Smrj goto done; 1214ae115bc7Smrj 12157c478bd9Sstevel@tonic-gate /* 12167c478bd9Sstevel@tonic-gate * Install a new mapping in the page's mapping list 12177c478bd9Sstevel@tonic-gate */ 12187c478bd9Sstevel@tonic-gate if (!PTE_ISVALID(old_pte)) { 12197c478bd9Sstevel@tonic-gate if (is_consist) { 12207c478bd9Sstevel@tonic-gate hment_assign(ht, entry, pp, hm); 12217c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 12227c478bd9Sstevel@tonic-gate } else { 12237c478bd9Sstevel@tonic-gate ASSERT(flags & HAT_LOAD_NOCONSIST); 12247c478bd9Sstevel@tonic-gate } 12257c478bd9Sstevel@tonic-gate HTABLE_INC(ht->ht_valid_cnt); 12267c478bd9Sstevel@tonic-gate PGCNT_INC(hat, l); 1227ae115bc7Smrj return (rv); 12287c478bd9Sstevel@tonic-gate } 12297c478bd9Sstevel@tonic-gate 12307c478bd9Sstevel@tonic-gate /* 12317c478bd9Sstevel@tonic-gate * Remap's are more complicated: 12327c478bd9Sstevel@tonic-gate * - HAT_LOAD_REMAP must be specified if changing the pfn. 12337c478bd9Sstevel@tonic-gate * We also require that NOCONSIST be specified. 12347c478bd9Sstevel@tonic-gate * - Otherwise only permission or caching bits may change. 12357c478bd9Sstevel@tonic-gate */ 12367c478bd9Sstevel@tonic-gate if (!PTE_ISPAGE(old_pte, l)) 12377c478bd9Sstevel@tonic-gate panic("non-null/page mapping pte=" FMT_PTE, old_pte); 12387c478bd9Sstevel@tonic-gate 12397c478bd9Sstevel@tonic-gate if (PTE2PFN(old_pte, l) != PTE2PFN(pte, l)) { 1240b193e412Skchow REMAPASSERT(flags & HAT_LOAD_REMAP); 1241b193e412Skchow REMAPASSERT(flags & HAT_LOAD_NOCONSIST); 1242ae115bc7Smrj REMAPASSERT(PTE_GET(old_pte, PT_SOFTWARE) >= PT_NOCONSIST); 1243b193e412Skchow REMAPASSERT(pf_is_memory(PTE2PFN(old_pte, l)) == 12447c478bd9Sstevel@tonic-gate pf_is_memory(PTE2PFN(pte, l))); 1245b193e412Skchow REMAPASSERT(!is_consist); 12467c478bd9Sstevel@tonic-gate } 12477c478bd9Sstevel@tonic-gate 12487c478bd9Sstevel@tonic-gate /* 12497c478bd9Sstevel@tonic-gate * We only let remaps change the bits for PFNs, permissions 12507c478bd9Sstevel@tonic-gate * or caching type. 12517c478bd9Sstevel@tonic-gate */ 12527c478bd9Sstevel@tonic-gate ASSERT(PTE_GET(old_pte, ~(PT_REMAP_BITS | PT_REF | PT_MOD)) == 12537c478bd9Sstevel@tonic-gate PTE_GET(pte, ~PT_REMAP_BITS)); 12547c478bd9Sstevel@tonic-gate 12557c478bd9Sstevel@tonic-gate /* 12567c478bd9Sstevel@tonic-gate * We don't create any mapping list entries on a remap, so release 12577c478bd9Sstevel@tonic-gate * any allocated hment after we drop the mapping list lock. 12587c478bd9Sstevel@tonic-gate */ 1259ae115bc7Smrj done: 12607c478bd9Sstevel@tonic-gate if (is_consist) { 12617c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 12627c478bd9Sstevel@tonic-gate if (hm != NULL) 12637c478bd9Sstevel@tonic-gate hment_free(hm); 12647c478bd9Sstevel@tonic-gate } 1265ae115bc7Smrj return (rv); 12667c478bd9Sstevel@tonic-gate } 12677c478bd9Sstevel@tonic-gate 12687c478bd9Sstevel@tonic-gate /* 1269ae115bc7Smrj * Internal routine to load a single page table entry. This only fails if 1270ae115bc7Smrj * we attempt to overwrite a page table link with a large page. 12717c478bd9Sstevel@tonic-gate */ 1272ae115bc7Smrj static int 12737c478bd9Sstevel@tonic-gate hati_load_common( 12747c478bd9Sstevel@tonic-gate hat_t *hat, 12757c478bd9Sstevel@tonic-gate uintptr_t va, 12767c478bd9Sstevel@tonic-gate page_t *pp, 12777c478bd9Sstevel@tonic-gate uint_t attr, 12787c478bd9Sstevel@tonic-gate uint_t flags, 12797c478bd9Sstevel@tonic-gate level_t level, 12807c478bd9Sstevel@tonic-gate pfn_t pfn) 12817c478bd9Sstevel@tonic-gate { 12827c478bd9Sstevel@tonic-gate htable_t *ht; 12837c478bd9Sstevel@tonic-gate uint_t entry; 12847c478bd9Sstevel@tonic-gate x86pte_t pte; 1285ae115bc7Smrj int rv = 0; 12867c478bd9Sstevel@tonic-gate 1287aac11643Sjosephb /* 1288aac11643Sjosephb * The number 16 is arbitrary and here to catch a recursion problem 1289aac11643Sjosephb * early before we blow out the kernel stack. 1290aac11643Sjosephb */ 1291aac11643Sjosephb ++curthread->t_hatdepth; 1292aac11643Sjosephb ASSERT(curthread->t_hatdepth < 16); 1293aac11643Sjosephb 12947c478bd9Sstevel@tonic-gate ASSERT(hat == kas.a_hat || 12957c478bd9Sstevel@tonic-gate AS_LOCK_HELD(hat->hat_as, &hat->hat_as->a_lock)); 12967c478bd9Sstevel@tonic-gate 12977c478bd9Sstevel@tonic-gate if (flags & HAT_LOAD_SHARE) 12987c478bd9Sstevel@tonic-gate hat->hat_flags |= HAT_SHARED; 12997c478bd9Sstevel@tonic-gate 13007c478bd9Sstevel@tonic-gate /* 13017c478bd9Sstevel@tonic-gate * Find the page table that maps this page if it already exists. 13027c478bd9Sstevel@tonic-gate */ 13037c478bd9Sstevel@tonic-gate ht = htable_lookup(hat, va, level); 13047c478bd9Sstevel@tonic-gate 13057c478bd9Sstevel@tonic-gate /* 1306aac11643Sjosephb * We must have HAT_LOAD_NOCONSIST if page_t is NULL. 13077c478bd9Sstevel@tonic-gate */ 1308aac11643Sjosephb if (pp == NULL) 13097c478bd9Sstevel@tonic-gate flags |= HAT_LOAD_NOCONSIST; 13107c478bd9Sstevel@tonic-gate 13117c478bd9Sstevel@tonic-gate if (ht == NULL) { 13127c478bd9Sstevel@tonic-gate ht = htable_create(hat, va, level, NULL); 13137c478bd9Sstevel@tonic-gate ASSERT(ht != NULL); 13147c478bd9Sstevel@tonic-gate } 13157c478bd9Sstevel@tonic-gate entry = htable_va2entry(va, ht); 13167c478bd9Sstevel@tonic-gate 13177c478bd9Sstevel@tonic-gate /* 13187c478bd9Sstevel@tonic-gate * a bunch of paranoid error checking 13197c478bd9Sstevel@tonic-gate */ 13207c478bd9Sstevel@tonic-gate ASSERT(ht->ht_busy > 0); 13217c478bd9Sstevel@tonic-gate if (ht->ht_vaddr > va || va > HTABLE_LAST_PAGE(ht)) 13227c478bd9Sstevel@tonic-gate panic("hati_load_common: bad htable %p, va %p", ht, (void *)va); 13237c478bd9Sstevel@tonic-gate ASSERT(ht->ht_level == level); 13247c478bd9Sstevel@tonic-gate 13257c478bd9Sstevel@tonic-gate /* 13267c478bd9Sstevel@tonic-gate * construct the new PTE 13277c478bd9Sstevel@tonic-gate */ 13287c478bd9Sstevel@tonic-gate if (hat == kas.a_hat) 13297c478bd9Sstevel@tonic-gate attr &= ~PROT_USER; 13307c478bd9Sstevel@tonic-gate pte = hati_mkpte(pfn, attr, level, flags); 13317c478bd9Sstevel@tonic-gate if (hat == kas.a_hat && va >= kernelbase) 13327c478bd9Sstevel@tonic-gate PTE_SET(pte, mmu.pt_global); 13337c478bd9Sstevel@tonic-gate 13347c478bd9Sstevel@tonic-gate /* 13357c478bd9Sstevel@tonic-gate * establish the mapping 13367c478bd9Sstevel@tonic-gate */ 1337ae115bc7Smrj rv = hati_pte_map(ht, entry, pp, pte, flags, NULL); 13387c478bd9Sstevel@tonic-gate 13397c478bd9Sstevel@tonic-gate /* 13407c478bd9Sstevel@tonic-gate * release the htable and any reserves 13417c478bd9Sstevel@tonic-gate */ 13427c478bd9Sstevel@tonic-gate htable_release(ht); 1343aac11643Sjosephb --curthread->t_hatdepth; 1344ae115bc7Smrj return (rv); 13457c478bd9Sstevel@tonic-gate } 13467c478bd9Sstevel@tonic-gate 13477c478bd9Sstevel@tonic-gate /* 13487c478bd9Sstevel@tonic-gate * special case of hat_memload to deal with some kernel addrs for performance 13497c478bd9Sstevel@tonic-gate */ 13507c478bd9Sstevel@tonic-gate static void 13517c478bd9Sstevel@tonic-gate hat_kmap_load( 13527c478bd9Sstevel@tonic-gate caddr_t addr, 13537c478bd9Sstevel@tonic-gate page_t *pp, 13547c478bd9Sstevel@tonic-gate uint_t attr, 13557c478bd9Sstevel@tonic-gate uint_t flags) 13567c478bd9Sstevel@tonic-gate { 13577c478bd9Sstevel@tonic-gate uintptr_t va = (uintptr_t)addr; 13587c478bd9Sstevel@tonic-gate x86pte_t pte; 13597c478bd9Sstevel@tonic-gate pfn_t pfn = page_pptonum(pp); 13607c478bd9Sstevel@tonic-gate pgcnt_t pg_off = mmu_btop(va - mmu.kmap_addr); 13617c478bd9Sstevel@tonic-gate htable_t *ht; 13627c478bd9Sstevel@tonic-gate uint_t entry; 13637c478bd9Sstevel@tonic-gate void *pte_ptr; 13647c478bd9Sstevel@tonic-gate 13657c478bd9Sstevel@tonic-gate /* 13667c478bd9Sstevel@tonic-gate * construct the requested PTE 13677c478bd9Sstevel@tonic-gate */ 13687c478bd9Sstevel@tonic-gate attr &= ~PROT_USER; 13697c478bd9Sstevel@tonic-gate attr |= HAT_STORECACHING_OK; 13707c478bd9Sstevel@tonic-gate pte = hati_mkpte(pfn, attr, 0, flags); 13717c478bd9Sstevel@tonic-gate PTE_SET(pte, mmu.pt_global); 13727c478bd9Sstevel@tonic-gate 13737c478bd9Sstevel@tonic-gate /* 13747c478bd9Sstevel@tonic-gate * Figure out the pte_ptr and htable and use common code to finish up 13757c478bd9Sstevel@tonic-gate */ 13767c478bd9Sstevel@tonic-gate if (mmu.pae_hat) 13777c478bd9Sstevel@tonic-gate pte_ptr = mmu.kmap_ptes + pg_off; 13787c478bd9Sstevel@tonic-gate else 13797c478bd9Sstevel@tonic-gate pte_ptr = (x86pte32_t *)mmu.kmap_ptes + pg_off; 13807c478bd9Sstevel@tonic-gate ht = mmu.kmap_htables[(va - mmu.kmap_htables[0]->ht_vaddr) >> 13817c478bd9Sstevel@tonic-gate LEVEL_SHIFT(1)]; 13827c478bd9Sstevel@tonic-gate entry = htable_va2entry(va, ht); 1383aac11643Sjosephb ++curthread->t_hatdepth; 1384aac11643Sjosephb ASSERT(curthread->t_hatdepth < 16); 1385ae115bc7Smrj (void) hati_pte_map(ht, entry, pp, pte, flags, pte_ptr); 1386aac11643Sjosephb --curthread->t_hatdepth; 13877c478bd9Sstevel@tonic-gate } 13887c478bd9Sstevel@tonic-gate 13897c478bd9Sstevel@tonic-gate /* 13907c478bd9Sstevel@tonic-gate * hat_memload() - load a translation to the given page struct 13917c478bd9Sstevel@tonic-gate * 13927c478bd9Sstevel@tonic-gate * Flags for hat_memload/hat_devload/hat_*attr. 13937c478bd9Sstevel@tonic-gate * 13947c478bd9Sstevel@tonic-gate * HAT_LOAD Default flags to load a translation to the page. 13957c478bd9Sstevel@tonic-gate * 13967c478bd9Sstevel@tonic-gate * HAT_LOAD_LOCK Lock down mapping resources; hat_map(), hat_memload(), 13977c478bd9Sstevel@tonic-gate * and hat_devload(). 13987c478bd9Sstevel@tonic-gate * 13997c478bd9Sstevel@tonic-gate * HAT_LOAD_NOCONSIST Do not add mapping to page_t mapping list. 1400ae115bc7Smrj * sets PT_NOCONSIST 14017c478bd9Sstevel@tonic-gate * 14027c478bd9Sstevel@tonic-gate * HAT_LOAD_SHARE A flag to hat_memload() to indicate h/w page tables 14037c478bd9Sstevel@tonic-gate * that map some user pages (not kas) is shared by more 14047c478bd9Sstevel@tonic-gate * than one process (eg. ISM). 14057c478bd9Sstevel@tonic-gate * 14067c478bd9Sstevel@tonic-gate * HAT_LOAD_REMAP Reload a valid pte with a different page frame. 14077c478bd9Sstevel@tonic-gate * 14087c478bd9Sstevel@tonic-gate * HAT_NO_KALLOC Do not kmem_alloc while creating the mapping; at this 14097c478bd9Sstevel@tonic-gate * point, it's setting up mapping to allocate internal 14107c478bd9Sstevel@tonic-gate * hat layer data structures. This flag forces hat layer 14117c478bd9Sstevel@tonic-gate * to tap its reserves in order to prevent infinite 14127c478bd9Sstevel@tonic-gate * recursion. 14137c478bd9Sstevel@tonic-gate * 14147c478bd9Sstevel@tonic-gate * The following is a protection attribute (like PROT_READ, etc.) 14157c478bd9Sstevel@tonic-gate * 1416ae115bc7Smrj * HAT_NOSYNC set PT_NOSYNC - this mapping's ref/mod bits 14177c478bd9Sstevel@tonic-gate * are never cleared. 14187c478bd9Sstevel@tonic-gate * 14197c478bd9Sstevel@tonic-gate * Installing new valid PTE's and creation of the mapping list 14207c478bd9Sstevel@tonic-gate * entry are controlled under the same lock. It's derived from the 14217c478bd9Sstevel@tonic-gate * page_t being mapped. 14227c478bd9Sstevel@tonic-gate */ 14237c478bd9Sstevel@tonic-gate static uint_t supported_memload_flags = 14247c478bd9Sstevel@tonic-gate HAT_LOAD | HAT_LOAD_LOCK | HAT_LOAD_ADV | HAT_LOAD_NOCONSIST | 14257c478bd9Sstevel@tonic-gate HAT_LOAD_SHARE | HAT_NO_KALLOC | HAT_LOAD_REMAP | HAT_LOAD_TEXT; 14267c478bd9Sstevel@tonic-gate 14277c478bd9Sstevel@tonic-gate void 14287c478bd9Sstevel@tonic-gate hat_memload( 14297c478bd9Sstevel@tonic-gate hat_t *hat, 14307c478bd9Sstevel@tonic-gate caddr_t addr, 14317c478bd9Sstevel@tonic-gate page_t *pp, 14327c478bd9Sstevel@tonic-gate uint_t attr, 14337c478bd9Sstevel@tonic-gate uint_t flags) 14347c478bd9Sstevel@tonic-gate { 14357c478bd9Sstevel@tonic-gate uintptr_t va = (uintptr_t)addr; 14367c478bd9Sstevel@tonic-gate level_t level = 0; 14377c478bd9Sstevel@tonic-gate pfn_t pfn = page_pptonum(pp); 14387c478bd9Sstevel@tonic-gate 14397c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(va)); 1440ae115bc7Smrj ASSERT(hat == kas.a_hat || va < _userlimit); 14417c478bd9Sstevel@tonic-gate ASSERT(hat == kas.a_hat || 14427c478bd9Sstevel@tonic-gate AS_LOCK_HELD(hat->hat_as, &hat->hat_as->a_lock)); 14437c478bd9Sstevel@tonic-gate ASSERT((flags & supported_memload_flags) == flags); 14447c478bd9Sstevel@tonic-gate 14457c478bd9Sstevel@tonic-gate ASSERT(!IN_VA_HOLE(va)); 14467c478bd9Sstevel@tonic-gate ASSERT(!PP_ISFREE(pp)); 14477c478bd9Sstevel@tonic-gate 14487c478bd9Sstevel@tonic-gate /* 14497c478bd9Sstevel@tonic-gate * kernel address special case for performance. 14507c478bd9Sstevel@tonic-gate */ 14517c478bd9Sstevel@tonic-gate if (mmu.kmap_addr <= va && va < mmu.kmap_eaddr) { 14527c478bd9Sstevel@tonic-gate ASSERT(hat == kas.a_hat); 14537c478bd9Sstevel@tonic-gate hat_kmap_load(addr, pp, attr, flags); 14547c478bd9Sstevel@tonic-gate return; 14557c478bd9Sstevel@tonic-gate } 14567c478bd9Sstevel@tonic-gate 14577c478bd9Sstevel@tonic-gate /* 14587c478bd9Sstevel@tonic-gate * This is used for memory with normal caching enabled, so 14597c478bd9Sstevel@tonic-gate * always set HAT_STORECACHING_OK. 14607c478bd9Sstevel@tonic-gate */ 14617c478bd9Sstevel@tonic-gate attr |= HAT_STORECACHING_OK; 1462ae115bc7Smrj if (hati_load_common(hat, va, pp, attr, flags, level, pfn) != 0) 1463ae115bc7Smrj panic("unexpected hati_load_common() failure"); 14647c478bd9Sstevel@tonic-gate } 14657c478bd9Sstevel@tonic-gate 14667c478bd9Sstevel@tonic-gate /* 14677c478bd9Sstevel@tonic-gate * Load the given array of page structs using large pages when possible 14687c478bd9Sstevel@tonic-gate */ 14697c478bd9Sstevel@tonic-gate void 14707c478bd9Sstevel@tonic-gate hat_memload_array( 14717c478bd9Sstevel@tonic-gate hat_t *hat, 14727c478bd9Sstevel@tonic-gate caddr_t addr, 14737c478bd9Sstevel@tonic-gate size_t len, 14747c478bd9Sstevel@tonic-gate page_t **pages, 14757c478bd9Sstevel@tonic-gate uint_t attr, 14767c478bd9Sstevel@tonic-gate uint_t flags) 14777c478bd9Sstevel@tonic-gate { 14787c478bd9Sstevel@tonic-gate uintptr_t va = (uintptr_t)addr; 14797c478bd9Sstevel@tonic-gate uintptr_t eaddr = va + len; 14807c478bd9Sstevel@tonic-gate level_t level; 14817c478bd9Sstevel@tonic-gate size_t pgsize; 14827c478bd9Sstevel@tonic-gate pgcnt_t pgindx = 0; 14837c478bd9Sstevel@tonic-gate pfn_t pfn; 14847c478bd9Sstevel@tonic-gate pgcnt_t i; 14857c478bd9Sstevel@tonic-gate 14867c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(va)); 1487ae115bc7Smrj ASSERT(hat == kas.a_hat || va + len <= _userlimit); 14887c478bd9Sstevel@tonic-gate ASSERT(hat == kas.a_hat || 14897c478bd9Sstevel@tonic-gate AS_LOCK_HELD(hat->hat_as, &hat->hat_as->a_lock)); 14907c478bd9Sstevel@tonic-gate ASSERT((flags & supported_memload_flags) == flags); 14917c478bd9Sstevel@tonic-gate 14927c478bd9Sstevel@tonic-gate /* 14937c478bd9Sstevel@tonic-gate * memload is used for memory with full caching enabled, so 14947c478bd9Sstevel@tonic-gate * set HAT_STORECACHING_OK. 14957c478bd9Sstevel@tonic-gate */ 14967c478bd9Sstevel@tonic-gate attr |= HAT_STORECACHING_OK; 14977c478bd9Sstevel@tonic-gate 14987c478bd9Sstevel@tonic-gate /* 14997c478bd9Sstevel@tonic-gate * handle all pages using largest possible pagesize 15007c478bd9Sstevel@tonic-gate */ 15017c478bd9Sstevel@tonic-gate while (va < eaddr) { 15027c478bd9Sstevel@tonic-gate /* 15037c478bd9Sstevel@tonic-gate * decide what level mapping to use (ie. pagesize) 15047c478bd9Sstevel@tonic-gate */ 15057c478bd9Sstevel@tonic-gate pfn = page_pptonum(pages[pgindx]); 15067c478bd9Sstevel@tonic-gate for (level = mmu.max_page_level; ; --level) { 15077c478bd9Sstevel@tonic-gate pgsize = LEVEL_SIZE(level); 15087c478bd9Sstevel@tonic-gate if (level == 0) 15097c478bd9Sstevel@tonic-gate break; 1510ae115bc7Smrj 15117c478bd9Sstevel@tonic-gate if (!IS_P2ALIGNED(va, pgsize) || 15127c478bd9Sstevel@tonic-gate (eaddr - va) < pgsize || 1513ae115bc7Smrj !IS_P2ALIGNED(pfn_to_pa(pfn), pgsize)) 15147c478bd9Sstevel@tonic-gate continue; 15157c478bd9Sstevel@tonic-gate 15167c478bd9Sstevel@tonic-gate /* 15177c478bd9Sstevel@tonic-gate * To use a large mapping of this size, all the 15187c478bd9Sstevel@tonic-gate * pages we are passed must be sequential subpages 15197c478bd9Sstevel@tonic-gate * of the large page. 15207c478bd9Sstevel@tonic-gate * hat_page_demote() can't change p_szc because 15217c478bd9Sstevel@tonic-gate * all pages are locked. 15227c478bd9Sstevel@tonic-gate */ 15237c478bd9Sstevel@tonic-gate if (pages[pgindx]->p_szc >= level) { 15247c478bd9Sstevel@tonic-gate for (i = 0; i < mmu_btop(pgsize); ++i) { 15257c478bd9Sstevel@tonic-gate if (pfn + i != 15267c478bd9Sstevel@tonic-gate page_pptonum(pages[pgindx + i])) 15277c478bd9Sstevel@tonic-gate break; 15287c478bd9Sstevel@tonic-gate ASSERT(pages[pgindx + i]->p_szc >= 15297c478bd9Sstevel@tonic-gate level); 15307c478bd9Sstevel@tonic-gate ASSERT(pages[pgindx] + i == 15317c478bd9Sstevel@tonic-gate pages[pgindx + i]); 15327c478bd9Sstevel@tonic-gate } 15337c478bd9Sstevel@tonic-gate if (i == mmu_btop(pgsize)) 15347c478bd9Sstevel@tonic-gate break; 15357c478bd9Sstevel@tonic-gate } 15367c478bd9Sstevel@tonic-gate } 15377c478bd9Sstevel@tonic-gate 15387c478bd9Sstevel@tonic-gate /* 1539ae115bc7Smrj * Load this page mapping. If the load fails, try a smaller 1540ae115bc7Smrj * pagesize. 15417c478bd9Sstevel@tonic-gate */ 15427c478bd9Sstevel@tonic-gate ASSERT(!IN_VA_HOLE(va)); 1543ae115bc7Smrj while (hati_load_common(hat, va, pages[pgindx], attr, 1544ae115bc7Smrj flags, level, pfn) != 0) { 1545ae115bc7Smrj if (level == 0) 1546ae115bc7Smrj panic("unexpected hati_load_common() failure"); 1547ae115bc7Smrj --level; 1548ae115bc7Smrj pgsize = LEVEL_SIZE(level); 1549ae115bc7Smrj } 15507c478bd9Sstevel@tonic-gate 15517c478bd9Sstevel@tonic-gate /* 15527c478bd9Sstevel@tonic-gate * move to next page 15537c478bd9Sstevel@tonic-gate */ 15547c478bd9Sstevel@tonic-gate va += pgsize; 15557c478bd9Sstevel@tonic-gate pgindx += mmu_btop(pgsize); 15567c478bd9Sstevel@tonic-gate } 15577c478bd9Sstevel@tonic-gate } 15587c478bd9Sstevel@tonic-gate 15597c478bd9Sstevel@tonic-gate /* 15607c478bd9Sstevel@tonic-gate * void hat_devload(hat, addr, len, pf, attr, flags) 15617c478bd9Sstevel@tonic-gate * load/lock the given page frame number 15627c478bd9Sstevel@tonic-gate * 15637c478bd9Sstevel@tonic-gate * Advisory ordering attributes. Apply only to device mappings. 15647c478bd9Sstevel@tonic-gate * 15657c478bd9Sstevel@tonic-gate * HAT_STRICTORDER: the CPU must issue the references in order, as the 15667c478bd9Sstevel@tonic-gate * programmer specified. This is the default. 15677c478bd9Sstevel@tonic-gate * HAT_UNORDERED_OK: the CPU may reorder the references (this is all kinds 15687c478bd9Sstevel@tonic-gate * of reordering; store or load with store or load). 15697c478bd9Sstevel@tonic-gate * HAT_MERGING_OK: merging and batching: the CPU may merge individual stores 15707c478bd9Sstevel@tonic-gate * to consecutive locations (for example, turn two consecutive byte 15717c478bd9Sstevel@tonic-gate * stores into one halfword store), and it may batch individual loads 15727c478bd9Sstevel@tonic-gate * (for example, turn two consecutive byte loads into one halfword load). 15737c478bd9Sstevel@tonic-gate * This also implies re-ordering. 15747c478bd9Sstevel@tonic-gate * HAT_LOADCACHING_OK: the CPU may cache the data it fetches and reuse it 15757c478bd9Sstevel@tonic-gate * until another store occurs. The default is to fetch new data 15767c478bd9Sstevel@tonic-gate * on every load. This also implies merging. 15777c478bd9Sstevel@tonic-gate * HAT_STORECACHING_OK: the CPU may keep the data in the cache and push it to 15787c478bd9Sstevel@tonic-gate * the device (perhaps with other data) at a later time. The default is 15797c478bd9Sstevel@tonic-gate * to push the data right away. This also implies load caching. 15807c478bd9Sstevel@tonic-gate * 15817c478bd9Sstevel@tonic-gate * Equivalent of hat_memload(), but can be used for device memory where 15827c478bd9Sstevel@tonic-gate * there are no page_t's and we support additional flags (write merging, etc). 15837c478bd9Sstevel@tonic-gate * Note that we can have large page mappings with this interface. 15847c478bd9Sstevel@tonic-gate */ 15857c478bd9Sstevel@tonic-gate int supported_devload_flags = HAT_LOAD | HAT_LOAD_LOCK | 15867c478bd9Sstevel@tonic-gate HAT_LOAD_NOCONSIST | HAT_STRICTORDER | HAT_UNORDERED_OK | 15877c478bd9Sstevel@tonic-gate HAT_MERGING_OK | HAT_LOADCACHING_OK | HAT_STORECACHING_OK; 15887c478bd9Sstevel@tonic-gate 15897c478bd9Sstevel@tonic-gate void 15907c478bd9Sstevel@tonic-gate hat_devload( 15917c478bd9Sstevel@tonic-gate hat_t *hat, 15927c478bd9Sstevel@tonic-gate caddr_t addr, 15937c478bd9Sstevel@tonic-gate size_t len, 15947c478bd9Sstevel@tonic-gate pfn_t pfn, 15957c478bd9Sstevel@tonic-gate uint_t attr, 15967c478bd9Sstevel@tonic-gate int flags) 15977c478bd9Sstevel@tonic-gate { 15987c478bd9Sstevel@tonic-gate uintptr_t va = ALIGN2PAGE(addr); 15997c478bd9Sstevel@tonic-gate uintptr_t eva = va + len; 16007c478bd9Sstevel@tonic-gate level_t level; 16017c478bd9Sstevel@tonic-gate size_t pgsize; 16027c478bd9Sstevel@tonic-gate page_t *pp; 16037c478bd9Sstevel@tonic-gate int f; /* per PTE copy of flags - maybe modified */ 16047c478bd9Sstevel@tonic-gate uint_t a; /* per PTE copy of attr */ 16057c478bd9Sstevel@tonic-gate 16067c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(va)); 1607ae115bc7Smrj ASSERT(hat == kas.a_hat || eva <= _userlimit); 16087c478bd9Sstevel@tonic-gate ASSERT(hat == kas.a_hat || 16097c478bd9Sstevel@tonic-gate AS_LOCK_HELD(hat->hat_as, &hat->hat_as->a_lock)); 16107c478bd9Sstevel@tonic-gate ASSERT((flags & supported_devload_flags) == flags); 16117c478bd9Sstevel@tonic-gate 16127c478bd9Sstevel@tonic-gate /* 16137c478bd9Sstevel@tonic-gate * handle all pages 16147c478bd9Sstevel@tonic-gate */ 16157c478bd9Sstevel@tonic-gate while (va < eva) { 16167c478bd9Sstevel@tonic-gate 16177c478bd9Sstevel@tonic-gate /* 16187c478bd9Sstevel@tonic-gate * decide what level mapping to use (ie. pagesize) 16197c478bd9Sstevel@tonic-gate */ 16207c478bd9Sstevel@tonic-gate for (level = mmu.max_page_level; ; --level) { 16217c478bd9Sstevel@tonic-gate pgsize = LEVEL_SIZE(level); 16227c478bd9Sstevel@tonic-gate if (level == 0) 16237c478bd9Sstevel@tonic-gate break; 16247c478bd9Sstevel@tonic-gate if (IS_P2ALIGNED(va, pgsize) && 16257c478bd9Sstevel@tonic-gate (eva - va) >= pgsize && 16267c478bd9Sstevel@tonic-gate IS_P2ALIGNED(pfn, mmu_btop(pgsize))) 16277c478bd9Sstevel@tonic-gate break; 16287c478bd9Sstevel@tonic-gate } 16297c478bd9Sstevel@tonic-gate 16307c478bd9Sstevel@tonic-gate /* 1631ae115bc7Smrj * If this is just memory then allow caching (this happens 16327c478bd9Sstevel@tonic-gate * for the nucleus pages) - though HAT_PLAT_NOCACHE can be used 1633ae115bc7Smrj * to override that. If we don't have a page_t then make sure 16347c478bd9Sstevel@tonic-gate * NOCONSIST is set. 16357c478bd9Sstevel@tonic-gate */ 16367c478bd9Sstevel@tonic-gate a = attr; 16377c478bd9Sstevel@tonic-gate f = flags; 16387c478bd9Sstevel@tonic-gate if (pf_is_memory(pfn)) { 16397c478bd9Sstevel@tonic-gate if (!(a & HAT_PLAT_NOCACHE)) 16407c478bd9Sstevel@tonic-gate a |= HAT_STORECACHING_OK; 16417c478bd9Sstevel@tonic-gate 16427c478bd9Sstevel@tonic-gate if (f & HAT_LOAD_NOCONSIST) 16437c478bd9Sstevel@tonic-gate pp = NULL; 16447c478bd9Sstevel@tonic-gate else 16457c478bd9Sstevel@tonic-gate pp = page_numtopp_nolock(pfn); 16467c478bd9Sstevel@tonic-gate } else { 16477c478bd9Sstevel@tonic-gate pp = NULL; 16487c478bd9Sstevel@tonic-gate f |= HAT_LOAD_NOCONSIST; 16497c478bd9Sstevel@tonic-gate } 16507c478bd9Sstevel@tonic-gate 16517c478bd9Sstevel@tonic-gate /* 16527c478bd9Sstevel@tonic-gate * load this page mapping 16537c478bd9Sstevel@tonic-gate */ 16547c478bd9Sstevel@tonic-gate ASSERT(!IN_VA_HOLE(va)); 1655ae115bc7Smrj while (hati_load_common(hat, va, pp, a, f, level, pfn) != 0) { 1656ae115bc7Smrj if (level == 0) 1657ae115bc7Smrj panic("unexpected hati_load_common() failure"); 1658ae115bc7Smrj --level; 1659ae115bc7Smrj pgsize = LEVEL_SIZE(level); 1660ae115bc7Smrj } 16617c478bd9Sstevel@tonic-gate 16627c478bd9Sstevel@tonic-gate /* 16637c478bd9Sstevel@tonic-gate * move to next page 16647c478bd9Sstevel@tonic-gate */ 16657c478bd9Sstevel@tonic-gate va += pgsize; 16667c478bd9Sstevel@tonic-gate pfn += mmu_btop(pgsize); 16677c478bd9Sstevel@tonic-gate } 16687c478bd9Sstevel@tonic-gate } 16697c478bd9Sstevel@tonic-gate 16707c478bd9Sstevel@tonic-gate /* 16717c478bd9Sstevel@tonic-gate * void hat_unlock(hat, addr, len) 16727c478bd9Sstevel@tonic-gate * unlock the mappings to a given range of addresses 16737c478bd9Sstevel@tonic-gate * 16747c478bd9Sstevel@tonic-gate * Locks are tracked by ht_lock_cnt in the htable. 16757c478bd9Sstevel@tonic-gate */ 16767c478bd9Sstevel@tonic-gate void 16777c478bd9Sstevel@tonic-gate hat_unlock(hat_t *hat, caddr_t addr, size_t len) 16787c478bd9Sstevel@tonic-gate { 16797c478bd9Sstevel@tonic-gate uintptr_t vaddr = (uintptr_t)addr; 16807c478bd9Sstevel@tonic-gate uintptr_t eaddr = vaddr + len; 16817c478bd9Sstevel@tonic-gate htable_t *ht = NULL; 16827c478bd9Sstevel@tonic-gate 16837c478bd9Sstevel@tonic-gate /* 16847c478bd9Sstevel@tonic-gate * kernel entries are always locked, we don't track lock counts 16857c478bd9Sstevel@tonic-gate */ 1686ae115bc7Smrj ASSERT(hat == kas.a_hat || eaddr <= _userlimit); 16877c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(vaddr)); 16887c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(eaddr)); 16897c478bd9Sstevel@tonic-gate if (hat == kas.a_hat) 16907c478bd9Sstevel@tonic-gate return; 16917c478bd9Sstevel@tonic-gate if (eaddr > _userlimit) 16927c478bd9Sstevel@tonic-gate panic("hat_unlock() address out of range - above _userlimit"); 16937c478bd9Sstevel@tonic-gate 16947c478bd9Sstevel@tonic-gate ASSERT(AS_LOCK_HELD(hat->hat_as, &hat->hat_as->a_lock)); 16957c478bd9Sstevel@tonic-gate while (vaddr < eaddr) { 16967c478bd9Sstevel@tonic-gate (void) htable_walk(hat, &ht, &vaddr, eaddr); 16977c478bd9Sstevel@tonic-gate if (ht == NULL) 16987c478bd9Sstevel@tonic-gate break; 16997c478bd9Sstevel@tonic-gate 17007c478bd9Sstevel@tonic-gate ASSERT(!IN_VA_HOLE(vaddr)); 17017c478bd9Sstevel@tonic-gate 17027c478bd9Sstevel@tonic-gate if (ht->ht_lock_cnt < 1) 17037c478bd9Sstevel@tonic-gate panic("hat_unlock(): lock_cnt < 1, " 17047c478bd9Sstevel@tonic-gate "htable=%p, vaddr=%p\n", ht, (caddr_t)vaddr); 17057c478bd9Sstevel@tonic-gate HTABLE_LOCK_DEC(ht); 17067c478bd9Sstevel@tonic-gate 17077c478bd9Sstevel@tonic-gate vaddr += LEVEL_SIZE(ht->ht_level); 17087c478bd9Sstevel@tonic-gate } 17097c478bd9Sstevel@tonic-gate if (ht) 17107c478bd9Sstevel@tonic-gate htable_release(ht); 17117c478bd9Sstevel@tonic-gate } 17127c478bd9Sstevel@tonic-gate 17137c478bd9Sstevel@tonic-gate /* 17147c478bd9Sstevel@tonic-gate * Cross call service routine to demap a virtual page on 17157c478bd9Sstevel@tonic-gate * the current CPU or flush all mappings in TLB. 17167c478bd9Sstevel@tonic-gate */ 17177c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 17187c478bd9Sstevel@tonic-gate static int 17197c478bd9Sstevel@tonic-gate hati_demap_func(xc_arg_t a1, xc_arg_t a2, xc_arg_t a3) 17207c478bd9Sstevel@tonic-gate { 17217c478bd9Sstevel@tonic-gate hat_t *hat = (hat_t *)a1; 17227c478bd9Sstevel@tonic-gate caddr_t addr = (caddr_t)a2; 17237c478bd9Sstevel@tonic-gate 17247c478bd9Sstevel@tonic-gate /* 17257c478bd9Sstevel@tonic-gate * If the target hat isn't the kernel and this CPU isn't operating 17267c478bd9Sstevel@tonic-gate * in the target hat, we can ignore the cross call. 17277c478bd9Sstevel@tonic-gate */ 17287c478bd9Sstevel@tonic-gate if (hat != kas.a_hat && hat != CPU->cpu_current_hat) 17297c478bd9Sstevel@tonic-gate return (0); 17307c478bd9Sstevel@tonic-gate 17317c478bd9Sstevel@tonic-gate /* 17327c478bd9Sstevel@tonic-gate * For a normal address, we just flush one page mapping 17337c478bd9Sstevel@tonic-gate */ 17347c478bd9Sstevel@tonic-gate if ((uintptr_t)addr != DEMAP_ALL_ADDR) { 1735ae115bc7Smrj mmu_tlbflush_entry(addr); 17367c478bd9Sstevel@tonic-gate return (0); 17377c478bd9Sstevel@tonic-gate } 17387c478bd9Sstevel@tonic-gate 17397c478bd9Sstevel@tonic-gate /* 17407c478bd9Sstevel@tonic-gate * Otherwise we reload cr3 to effect a complete TLB flush. 17417c478bd9Sstevel@tonic-gate * 17427c478bd9Sstevel@tonic-gate * A reload of cr3 on a VLP process also means we must also recopy in 17437c478bd9Sstevel@tonic-gate * the pte values from the struct hat 17447c478bd9Sstevel@tonic-gate */ 17457c478bd9Sstevel@tonic-gate if (hat->hat_flags & HAT_VLP) { 17467c478bd9Sstevel@tonic-gate #if defined(__amd64) 17477c478bd9Sstevel@tonic-gate x86pte_t *vlpptep = CPU->cpu_hat_info->hci_vlp_l2ptes; 17487c478bd9Sstevel@tonic-gate 17497c478bd9Sstevel@tonic-gate VLP_COPY(hat->hat_vlp_ptes, vlpptep); 17507c478bd9Sstevel@tonic-gate #elif defined(__i386) 17517c478bd9Sstevel@tonic-gate reload_pae32(hat, CPU); 17527c478bd9Sstevel@tonic-gate #endif 17537c478bd9Sstevel@tonic-gate } 17547c478bd9Sstevel@tonic-gate reload_cr3(); 17557c478bd9Sstevel@tonic-gate return (0); 17567c478bd9Sstevel@tonic-gate } 17577c478bd9Sstevel@tonic-gate 1758*95c0a3c8Sjosephb /* 1759*95c0a3c8Sjosephb * Flush all TLB entries, including global (ie. kernel) ones. 1760*95c0a3c8Sjosephb */ 1761*95c0a3c8Sjosephb static void 1762*95c0a3c8Sjosephb flush_all_tlb_entries(void) 1763*95c0a3c8Sjosephb { 1764*95c0a3c8Sjosephb ulong_t cr4 = getcr4(); 1765*95c0a3c8Sjosephb 1766*95c0a3c8Sjosephb if (cr4 & CR4_PGE) { 1767*95c0a3c8Sjosephb setcr4(cr4 & ~(ulong_t)CR4_PGE); 1768*95c0a3c8Sjosephb setcr4(cr4); 1769*95c0a3c8Sjosephb 1770*95c0a3c8Sjosephb /* 1771*95c0a3c8Sjosephb * 32 bit PAE also needs to always reload_cr3() 1772*95c0a3c8Sjosephb */ 1773*95c0a3c8Sjosephb if (mmu.max_level == 2) 1774*95c0a3c8Sjosephb reload_cr3(); 1775*95c0a3c8Sjosephb } else { 1776*95c0a3c8Sjosephb reload_cr3(); 1777*95c0a3c8Sjosephb } 1778*95c0a3c8Sjosephb } 1779*95c0a3c8Sjosephb 1780*95c0a3c8Sjosephb #define TLB_CPU_HALTED (01ul) 1781*95c0a3c8Sjosephb #define TLB_INVAL_ALL (02ul) 1782*95c0a3c8Sjosephb #define CAS_TLB_INFO(cpu, old, new) \ 1783*95c0a3c8Sjosephb caslong((ulong_t *)&(cpu)->cpu_m.mcpu_tlb_info, (old), (new)) 1784*95c0a3c8Sjosephb 1785*95c0a3c8Sjosephb /* 1786*95c0a3c8Sjosephb * Record that a CPU is going idle 1787*95c0a3c8Sjosephb */ 1788*95c0a3c8Sjosephb void 1789*95c0a3c8Sjosephb tlb_going_idle(void) 1790*95c0a3c8Sjosephb { 1791*95c0a3c8Sjosephb atomic_or_long((ulong_t *)&CPU->cpu_m.mcpu_tlb_info, TLB_CPU_HALTED); 1792*95c0a3c8Sjosephb } 1793*95c0a3c8Sjosephb 1794*95c0a3c8Sjosephb /* 1795*95c0a3c8Sjosephb * Service a delayed TLB flush if coming out of being idle. 1796*95c0a3c8Sjosephb */ 1797*95c0a3c8Sjosephb void 1798*95c0a3c8Sjosephb tlb_service(void) 1799*95c0a3c8Sjosephb { 1800*95c0a3c8Sjosephb ulong_t flags = getflags(); 1801*95c0a3c8Sjosephb ulong_t tlb_info; 1802*95c0a3c8Sjosephb ulong_t found; 1803*95c0a3c8Sjosephb 1804*95c0a3c8Sjosephb /* 1805*95c0a3c8Sjosephb * Be sure interrupts are off while doing this so that 1806*95c0a3c8Sjosephb * higher level interrupts correctly wait for flushes to finish. 1807*95c0a3c8Sjosephb */ 1808*95c0a3c8Sjosephb if (flags & PS_IE) 1809*95c0a3c8Sjosephb flags = intr_clear(); 1810*95c0a3c8Sjosephb 1811*95c0a3c8Sjosephb /* 1812*95c0a3c8Sjosephb * We only have to do something if coming out of being idle. 1813*95c0a3c8Sjosephb */ 1814*95c0a3c8Sjosephb tlb_info = CPU->cpu_m.mcpu_tlb_info; 1815*95c0a3c8Sjosephb if (tlb_info & TLB_CPU_HALTED) { 1816*95c0a3c8Sjosephb ASSERT(CPU->cpu_current_hat == kas.a_hat); 1817*95c0a3c8Sjosephb 1818*95c0a3c8Sjosephb /* 1819*95c0a3c8Sjosephb * Atomic clear and fetch of old state. 1820*95c0a3c8Sjosephb */ 1821*95c0a3c8Sjosephb while ((found = CAS_TLB_INFO(CPU, tlb_info, 0)) != tlb_info) { 1822*95c0a3c8Sjosephb ASSERT(found & TLB_CPU_HALTED); 1823*95c0a3c8Sjosephb tlb_info = found; 1824*95c0a3c8Sjosephb SMT_PAUSE(); 1825*95c0a3c8Sjosephb } 1826*95c0a3c8Sjosephb if (tlb_info & TLB_INVAL_ALL) 1827*95c0a3c8Sjosephb flush_all_tlb_entries(); 1828*95c0a3c8Sjosephb } 1829*95c0a3c8Sjosephb 1830*95c0a3c8Sjosephb /* 1831*95c0a3c8Sjosephb * Restore interrupt enable control bit. 1832*95c0a3c8Sjosephb */ 1833*95c0a3c8Sjosephb if (flags & PS_IE) 1834*95c0a3c8Sjosephb sti(); 1835*95c0a3c8Sjosephb } 1836*95c0a3c8Sjosephb 18377c478bd9Sstevel@tonic-gate /* 18387c478bd9Sstevel@tonic-gate * Internal routine to do cross calls to invalidate a range of pages on 18397c478bd9Sstevel@tonic-gate * all CPUs using a given hat. 18407c478bd9Sstevel@tonic-gate */ 18417c478bd9Sstevel@tonic-gate void 1842ae115bc7Smrj hat_tlb_inval(hat_t *hat, uintptr_t va) 18437c478bd9Sstevel@tonic-gate { 18447c478bd9Sstevel@tonic-gate extern int flushes_require_xcalls; /* from mp_startup.c */ 18457c478bd9Sstevel@tonic-gate cpuset_t justme; 1846*95c0a3c8Sjosephb cpuset_t check_cpus; 1847ae115bc7Smrj cpuset_t cpus_to_shootdown; 1848*95c0a3c8Sjosephb cpu_t *cpup; 1849*95c0a3c8Sjosephb int c; 18507c478bd9Sstevel@tonic-gate 18517c478bd9Sstevel@tonic-gate /* 18527c478bd9Sstevel@tonic-gate * If the hat is being destroyed, there are no more users, so 18537c478bd9Sstevel@tonic-gate * demap need not do anything. 18547c478bd9Sstevel@tonic-gate */ 18557c478bd9Sstevel@tonic-gate if (hat->hat_flags & HAT_FREEING) 18567c478bd9Sstevel@tonic-gate return; 18577c478bd9Sstevel@tonic-gate 18587c478bd9Sstevel@tonic-gate /* 18597c478bd9Sstevel@tonic-gate * If demapping from a shared pagetable, we best demap the 18607c478bd9Sstevel@tonic-gate * entire set of user TLBs, since we don't know what addresses 18617c478bd9Sstevel@tonic-gate * these were shared at. 18627c478bd9Sstevel@tonic-gate */ 18637c478bd9Sstevel@tonic-gate if (hat->hat_flags & HAT_SHARED) { 18647c478bd9Sstevel@tonic-gate hat = kas.a_hat; 18657c478bd9Sstevel@tonic-gate va = DEMAP_ALL_ADDR; 18667c478bd9Sstevel@tonic-gate } 18677c478bd9Sstevel@tonic-gate 18687c478bd9Sstevel@tonic-gate /* 18697c478bd9Sstevel@tonic-gate * if not running with multiple CPUs, don't use cross calls 18707c478bd9Sstevel@tonic-gate */ 18717c478bd9Sstevel@tonic-gate if (panicstr || !flushes_require_xcalls) { 18727c478bd9Sstevel@tonic-gate (void) hati_demap_func((xc_arg_t)hat, (xc_arg_t)va, NULL); 18737c478bd9Sstevel@tonic-gate return; 18747c478bd9Sstevel@tonic-gate } 18757c478bd9Sstevel@tonic-gate 18767c478bd9Sstevel@tonic-gate 18777c478bd9Sstevel@tonic-gate /* 1878ae115bc7Smrj * Determine CPUs to shootdown. Kernel changes always do all CPUs. 1879ae115bc7Smrj * Otherwise it's just CPUs currently executing in this hat. 18807c478bd9Sstevel@tonic-gate */ 18817c478bd9Sstevel@tonic-gate kpreempt_disable(); 18827c478bd9Sstevel@tonic-gate CPUSET_ONLY(justme, CPU->cpu_id); 1883ae115bc7Smrj if (hat == kas.a_hat) 1884ae115bc7Smrj cpus_to_shootdown = khat_cpuset; 18857c478bd9Sstevel@tonic-gate else 1886ae115bc7Smrj cpus_to_shootdown = hat->hat_cpus; 1887ae115bc7Smrj 1888*95c0a3c8Sjosephb /* 1889*95c0a3c8Sjosephb * If any CPUs in the set are idle, just request a delayed flush 1890*95c0a3c8Sjosephb * and avoid waking them up. 1891*95c0a3c8Sjosephb */ 1892*95c0a3c8Sjosephb check_cpus = cpus_to_shootdown; 1893*95c0a3c8Sjosephb for (c = 0; c < NCPU && !CPUSET_ISNULL(check_cpus); ++c) { 1894*95c0a3c8Sjosephb ulong_t tlb_info; 1895*95c0a3c8Sjosephb 1896*95c0a3c8Sjosephb if (!CPU_IN_SET(check_cpus, c)) 1897*95c0a3c8Sjosephb continue; 1898*95c0a3c8Sjosephb CPUSET_DEL(check_cpus, c); 1899*95c0a3c8Sjosephb cpup = cpu[c]; 1900*95c0a3c8Sjosephb if (cpup == NULL) 1901*95c0a3c8Sjosephb continue; 1902*95c0a3c8Sjosephb 1903*95c0a3c8Sjosephb tlb_info = cpup->cpu_m.mcpu_tlb_info; 1904*95c0a3c8Sjosephb while (tlb_info == TLB_CPU_HALTED) { 1905*95c0a3c8Sjosephb (void) CAS_TLB_INFO(cpup, TLB_CPU_HALTED, 1906*95c0a3c8Sjosephb TLB_CPU_HALTED | TLB_INVAL_ALL); 1907*95c0a3c8Sjosephb SMT_PAUSE(); 1908*95c0a3c8Sjosephb tlb_info = cpup->cpu_m.mcpu_tlb_info; 1909*95c0a3c8Sjosephb } 1910*95c0a3c8Sjosephb if (tlb_info == (TLB_CPU_HALTED | TLB_INVAL_ALL)) { 1911*95c0a3c8Sjosephb HATSTAT_INC(hs_tlb_inval_delayed); 1912*95c0a3c8Sjosephb CPUSET_DEL(cpus_to_shootdown, c); 1913*95c0a3c8Sjosephb } 1914*95c0a3c8Sjosephb } 1915*95c0a3c8Sjosephb 1916ae115bc7Smrj if (CPUSET_ISNULL(cpus_to_shootdown) || 1917ae115bc7Smrj CPUSET_ISEQUAL(cpus_to_shootdown, justme)) { 1918ae115bc7Smrj 1919ae115bc7Smrj (void) hati_demap_func((xc_arg_t)hat, (xc_arg_t)va, NULL); 1920ae115bc7Smrj 1921ae115bc7Smrj } else { 1922ae115bc7Smrj 1923ae115bc7Smrj CPUSET_ADD(cpus_to_shootdown, CPU->cpu_id); 1924ae115bc7Smrj xc_call((xc_arg_t)hat, (xc_arg_t)va, NULL, X_CALL_HIPRI, 1925ae115bc7Smrj cpus_to_shootdown, hati_demap_func); 1926ae115bc7Smrj 1927ae115bc7Smrj } 19287c478bd9Sstevel@tonic-gate kpreempt_enable(); 19297c478bd9Sstevel@tonic-gate } 19307c478bd9Sstevel@tonic-gate 19317c478bd9Sstevel@tonic-gate /* 19327c478bd9Sstevel@tonic-gate * Interior routine for HAT_UNLOADs from hat_unload_callback(), 19337c478bd9Sstevel@tonic-gate * hat_kmap_unload() OR from hat_steal() code. This routine doesn't 19347c478bd9Sstevel@tonic-gate * handle releasing of the htables. 19357c478bd9Sstevel@tonic-gate */ 19367c478bd9Sstevel@tonic-gate void 19377c478bd9Sstevel@tonic-gate hat_pte_unmap( 19387c478bd9Sstevel@tonic-gate htable_t *ht, 19397c478bd9Sstevel@tonic-gate uint_t entry, 19407c478bd9Sstevel@tonic-gate uint_t flags, 19417c478bd9Sstevel@tonic-gate x86pte_t old_pte, 19427c478bd9Sstevel@tonic-gate void *pte_ptr) 19437c478bd9Sstevel@tonic-gate { 19447c478bd9Sstevel@tonic-gate hat_t *hat = ht->ht_hat; 19457c478bd9Sstevel@tonic-gate hment_t *hm = NULL; 19467c478bd9Sstevel@tonic-gate page_t *pp = NULL; 19477c478bd9Sstevel@tonic-gate level_t l = ht->ht_level; 19487c478bd9Sstevel@tonic-gate pfn_t pfn; 19497c478bd9Sstevel@tonic-gate 19507c478bd9Sstevel@tonic-gate /* 19517c478bd9Sstevel@tonic-gate * We always track the locking counts, even if nothing is unmapped 19527c478bd9Sstevel@tonic-gate */ 19537c478bd9Sstevel@tonic-gate if ((flags & HAT_UNLOAD_UNLOCK) != 0 && hat != kas.a_hat) { 19547c478bd9Sstevel@tonic-gate ASSERT(ht->ht_lock_cnt > 0); 19557c478bd9Sstevel@tonic-gate HTABLE_LOCK_DEC(ht); 19567c478bd9Sstevel@tonic-gate } 19577c478bd9Sstevel@tonic-gate 19587c478bd9Sstevel@tonic-gate /* 19597c478bd9Sstevel@tonic-gate * Figure out which page's mapping list lock to acquire using the PFN 19607c478bd9Sstevel@tonic-gate * passed in "old" PTE. We then attempt to invalidate the PTE. 19617c478bd9Sstevel@tonic-gate * If another thread, probably a hat_pageunload, has asynchronously 19627c478bd9Sstevel@tonic-gate * unmapped/remapped this address we'll loop here. 19637c478bd9Sstevel@tonic-gate */ 19647c478bd9Sstevel@tonic-gate ASSERT(ht->ht_busy > 0); 19657c478bd9Sstevel@tonic-gate while (PTE_ISVALID(old_pte)) { 19667c478bd9Sstevel@tonic-gate pfn = PTE2PFN(old_pte, l); 1967ae115bc7Smrj if (PTE_GET(old_pte, PT_SOFTWARE) >= PT_NOCONSIST) { 19687c478bd9Sstevel@tonic-gate pp = NULL; 19697c478bd9Sstevel@tonic-gate } else { 19707c478bd9Sstevel@tonic-gate pp = page_numtopp_nolock(pfn); 1971aa2ed9e5Sjosephb if (pp == NULL) { 1972aa2ed9e5Sjosephb panic("no page_t, not NOCONSIST: old_pte=" 1973aa2ed9e5Sjosephb FMT_PTE " ht=%lx entry=0x%x pte_ptr=%lx", 1974aa2ed9e5Sjosephb old_pte, (uintptr_t)ht, entry, 1975aa2ed9e5Sjosephb (uintptr_t)pte_ptr); 1976aa2ed9e5Sjosephb } 19777c478bd9Sstevel@tonic-gate x86_hm_enter(pp); 19787c478bd9Sstevel@tonic-gate } 1979aa2ed9e5Sjosephb 1980aa2ed9e5Sjosephb /* 1981aa2ed9e5Sjosephb * If freeing the address space, check that the PTE 1982aa2ed9e5Sjosephb * hasn't changed, as the mappings are no longer in use by 1983aa2ed9e5Sjosephb * any thread, invalidation is unnecessary. 1984aa2ed9e5Sjosephb * If not freeing, do a full invalidate. 1985aa2ed9e5Sjosephb */ 1986aa2ed9e5Sjosephb if (hat->hat_flags & HAT_FREEING) 1987aa2ed9e5Sjosephb old_pte = x86pte_get(ht, entry); 1988aa2ed9e5Sjosephb else 1989ae115bc7Smrj old_pte = x86pte_inval(ht, entry, old_pte, pte_ptr); 19907c478bd9Sstevel@tonic-gate 19917c478bd9Sstevel@tonic-gate /* 19927c478bd9Sstevel@tonic-gate * If the page hadn't changed we've unmapped it and can proceed 19937c478bd9Sstevel@tonic-gate */ 19947c478bd9Sstevel@tonic-gate if (PTE_ISVALID(old_pte) && PTE2PFN(old_pte, l) == pfn) 19957c478bd9Sstevel@tonic-gate break; 19967c478bd9Sstevel@tonic-gate 19977c478bd9Sstevel@tonic-gate /* 19987c478bd9Sstevel@tonic-gate * Otherwise, we'll have to retry with the current old_pte. 19997c478bd9Sstevel@tonic-gate * Drop the hment lock, since the pfn may have changed. 20007c478bd9Sstevel@tonic-gate */ 20017c478bd9Sstevel@tonic-gate if (pp != NULL) { 20027c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 20037c478bd9Sstevel@tonic-gate pp = NULL; 20047c478bd9Sstevel@tonic-gate } else { 2005ae115bc7Smrj ASSERT(PTE_GET(old_pte, PT_SOFTWARE) >= PT_NOCONSIST); 20067c478bd9Sstevel@tonic-gate } 20077c478bd9Sstevel@tonic-gate } 20087c478bd9Sstevel@tonic-gate 20097c478bd9Sstevel@tonic-gate /* 20107c478bd9Sstevel@tonic-gate * If the old mapping wasn't valid, there's nothing more to do 20117c478bd9Sstevel@tonic-gate */ 20127c478bd9Sstevel@tonic-gate if (!PTE_ISVALID(old_pte)) { 20137c478bd9Sstevel@tonic-gate if (pp != NULL) 20147c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 20157c478bd9Sstevel@tonic-gate return; 20167c478bd9Sstevel@tonic-gate } 20177c478bd9Sstevel@tonic-gate 20187c478bd9Sstevel@tonic-gate /* 20197c478bd9Sstevel@tonic-gate * Take care of syncing any MOD/REF bits and removing the hment. 20207c478bd9Sstevel@tonic-gate */ 20217c478bd9Sstevel@tonic-gate if (pp != NULL) { 20227c478bd9Sstevel@tonic-gate if (!(flags & HAT_UNLOAD_NOSYNC)) 20237c478bd9Sstevel@tonic-gate hati_sync_pte_to_page(pp, old_pte, l); 20247c478bd9Sstevel@tonic-gate hm = hment_remove(pp, ht, entry); 20257c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 20267c478bd9Sstevel@tonic-gate if (hm != NULL) 20277c478bd9Sstevel@tonic-gate hment_free(hm); 20287c478bd9Sstevel@tonic-gate } 20297c478bd9Sstevel@tonic-gate 20307c478bd9Sstevel@tonic-gate /* 20317c478bd9Sstevel@tonic-gate * Handle book keeping in the htable and hat 20327c478bd9Sstevel@tonic-gate */ 20337c478bd9Sstevel@tonic-gate ASSERT(ht->ht_valid_cnt > 0); 20347c478bd9Sstevel@tonic-gate HTABLE_DEC(ht->ht_valid_cnt); 20357c478bd9Sstevel@tonic-gate PGCNT_DEC(hat, l); 20367c478bd9Sstevel@tonic-gate } 20377c478bd9Sstevel@tonic-gate 20387c478bd9Sstevel@tonic-gate /* 20397c478bd9Sstevel@tonic-gate * very cheap unload implementation to special case some kernel addresses 20407c478bd9Sstevel@tonic-gate */ 20417c478bd9Sstevel@tonic-gate static void 20427c478bd9Sstevel@tonic-gate hat_kmap_unload(caddr_t addr, size_t len, uint_t flags) 20437c478bd9Sstevel@tonic-gate { 20447c478bd9Sstevel@tonic-gate uintptr_t va = (uintptr_t)addr; 20457c478bd9Sstevel@tonic-gate uintptr_t eva = va + len; 2046ae115bc7Smrj pgcnt_t pg_index; 20477c478bd9Sstevel@tonic-gate htable_t *ht; 20487c478bd9Sstevel@tonic-gate uint_t entry; 2049ae115bc7Smrj x86pte_t *pte_ptr; 20507c478bd9Sstevel@tonic-gate x86pte_t old_pte; 20517c478bd9Sstevel@tonic-gate 20527c478bd9Sstevel@tonic-gate for (; va < eva; va += MMU_PAGESIZE) { 20537c478bd9Sstevel@tonic-gate /* 20547c478bd9Sstevel@tonic-gate * Get the PTE 20557c478bd9Sstevel@tonic-gate */ 2056ae115bc7Smrj pg_index = mmu_btop(va - mmu.kmap_addr); 2057ae115bc7Smrj pte_ptr = PT_INDEX_PTR(mmu.kmap_ptes, pg_index); 2058ae115bc7Smrj old_pte = GET_PTE(pte_ptr); 20597c478bd9Sstevel@tonic-gate 20607c478bd9Sstevel@tonic-gate /* 20617c478bd9Sstevel@tonic-gate * get the htable / entry 20627c478bd9Sstevel@tonic-gate */ 20637c478bd9Sstevel@tonic-gate ht = mmu.kmap_htables[(va - mmu.kmap_htables[0]->ht_vaddr) 20647c478bd9Sstevel@tonic-gate >> LEVEL_SHIFT(1)]; 20657c478bd9Sstevel@tonic-gate entry = htable_va2entry(va, ht); 20667c478bd9Sstevel@tonic-gate 20677c478bd9Sstevel@tonic-gate /* 20687c478bd9Sstevel@tonic-gate * use mostly common code to unmap it. 20697c478bd9Sstevel@tonic-gate */ 20707c478bd9Sstevel@tonic-gate hat_pte_unmap(ht, entry, flags, old_pte, pte_ptr); 20717c478bd9Sstevel@tonic-gate } 20727c478bd9Sstevel@tonic-gate } 20737c478bd9Sstevel@tonic-gate 20747c478bd9Sstevel@tonic-gate 20757c478bd9Sstevel@tonic-gate /* 20767c478bd9Sstevel@tonic-gate * unload a range of virtual address space (no callback) 20777c478bd9Sstevel@tonic-gate */ 20787c478bd9Sstevel@tonic-gate void 20797c478bd9Sstevel@tonic-gate hat_unload(hat_t *hat, caddr_t addr, size_t len, uint_t flags) 20807c478bd9Sstevel@tonic-gate { 20817c478bd9Sstevel@tonic-gate uintptr_t va = (uintptr_t)addr; 2082ae115bc7Smrj 2083ae115bc7Smrj ASSERT(hat == kas.a_hat || va + len <= _userlimit); 20847c478bd9Sstevel@tonic-gate 20857c478bd9Sstevel@tonic-gate /* 20867c478bd9Sstevel@tonic-gate * special case for performance. 20877c478bd9Sstevel@tonic-gate */ 20887c478bd9Sstevel@tonic-gate if (mmu.kmap_addr <= va && va < mmu.kmap_eaddr) { 20897c478bd9Sstevel@tonic-gate ASSERT(hat == kas.a_hat); 20907c478bd9Sstevel@tonic-gate hat_kmap_unload(addr, len, flags); 2091ae115bc7Smrj } else { 2092ae115bc7Smrj hat_unload_callback(hat, addr, len, flags, NULL); 20937c478bd9Sstevel@tonic-gate } 20947c478bd9Sstevel@tonic-gate } 20957c478bd9Sstevel@tonic-gate 20967c478bd9Sstevel@tonic-gate /* 20977c478bd9Sstevel@tonic-gate * Do the callbacks for ranges being unloaded. 20987c478bd9Sstevel@tonic-gate */ 20997c478bd9Sstevel@tonic-gate typedef struct range_info { 21007c478bd9Sstevel@tonic-gate uintptr_t rng_va; 21017c478bd9Sstevel@tonic-gate ulong_t rng_cnt; 21027c478bd9Sstevel@tonic-gate level_t rng_level; 21037c478bd9Sstevel@tonic-gate } range_info_t; 21047c478bd9Sstevel@tonic-gate 21057c478bd9Sstevel@tonic-gate static void 21067c478bd9Sstevel@tonic-gate handle_ranges(hat_callback_t *cb, uint_t cnt, range_info_t *range) 21077c478bd9Sstevel@tonic-gate { 21087c478bd9Sstevel@tonic-gate /* 21097c478bd9Sstevel@tonic-gate * do callbacks to upper level VM system 21107c478bd9Sstevel@tonic-gate */ 21117c478bd9Sstevel@tonic-gate while (cb != NULL && cnt > 0) { 21127c478bd9Sstevel@tonic-gate --cnt; 21137c478bd9Sstevel@tonic-gate cb->hcb_start_addr = (caddr_t)range[cnt].rng_va; 21147c478bd9Sstevel@tonic-gate cb->hcb_end_addr = cb->hcb_start_addr; 21157c478bd9Sstevel@tonic-gate cb->hcb_end_addr += 21167c478bd9Sstevel@tonic-gate range[cnt].rng_cnt << LEVEL_SIZE(range[cnt].rng_level); 21177c478bd9Sstevel@tonic-gate cb->hcb_function(cb); 21187c478bd9Sstevel@tonic-gate } 21197c478bd9Sstevel@tonic-gate } 21207c478bd9Sstevel@tonic-gate 21217c478bd9Sstevel@tonic-gate /* 21227c478bd9Sstevel@tonic-gate * Unload a given range of addresses (has optional callback) 21237c478bd9Sstevel@tonic-gate * 21247c478bd9Sstevel@tonic-gate * Flags: 21257c478bd9Sstevel@tonic-gate * define HAT_UNLOAD 0x00 21267c478bd9Sstevel@tonic-gate * define HAT_UNLOAD_NOSYNC 0x02 21277c478bd9Sstevel@tonic-gate * define HAT_UNLOAD_UNLOCK 0x04 21287c478bd9Sstevel@tonic-gate * define HAT_UNLOAD_OTHER 0x08 - not used 21297c478bd9Sstevel@tonic-gate * define HAT_UNLOAD_UNMAP 0x10 - same as HAT_UNLOAD 21307c478bd9Sstevel@tonic-gate */ 21317c478bd9Sstevel@tonic-gate #define MAX_UNLOAD_CNT (8) 21327c478bd9Sstevel@tonic-gate void 21337c478bd9Sstevel@tonic-gate hat_unload_callback( 21347c478bd9Sstevel@tonic-gate hat_t *hat, 21357c478bd9Sstevel@tonic-gate caddr_t addr, 21367c478bd9Sstevel@tonic-gate size_t len, 21377c478bd9Sstevel@tonic-gate uint_t flags, 21387c478bd9Sstevel@tonic-gate hat_callback_t *cb) 21397c478bd9Sstevel@tonic-gate { 21407c478bd9Sstevel@tonic-gate uintptr_t vaddr = (uintptr_t)addr; 21417c478bd9Sstevel@tonic-gate uintptr_t eaddr = vaddr + len; 21427c478bd9Sstevel@tonic-gate htable_t *ht = NULL; 21437c478bd9Sstevel@tonic-gate uint_t entry; 2144aa2ed9e5Sjosephb uintptr_t contig_va = (uintptr_t)-1L; 21457c478bd9Sstevel@tonic-gate range_info_t r[MAX_UNLOAD_CNT]; 21467c478bd9Sstevel@tonic-gate uint_t r_cnt = 0; 21477c478bd9Sstevel@tonic-gate x86pte_t old_pte; 21487c478bd9Sstevel@tonic-gate 2149ae115bc7Smrj ASSERT(hat == kas.a_hat || eaddr <= _userlimit); 21507c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(vaddr)); 21517c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(eaddr)); 21527c478bd9Sstevel@tonic-gate 2153ae115bc7Smrj /* 2154ae115bc7Smrj * Special case a single page being unloaded for speed. This happens 2155ae115bc7Smrj * quite frequently, COW faults after a fork() for example. 2156ae115bc7Smrj */ 2157ae115bc7Smrj if (cb == NULL && len == MMU_PAGESIZE) { 2158ae115bc7Smrj ht = htable_getpte(hat, vaddr, &entry, &old_pte, 0); 2159ae115bc7Smrj if (ht != NULL) { 2160ae115bc7Smrj if (PTE_ISVALID(old_pte)) 2161ae115bc7Smrj hat_pte_unmap(ht, entry, flags, old_pte, NULL); 2162ae115bc7Smrj htable_release(ht); 2163ae115bc7Smrj } 2164ae115bc7Smrj return; 2165ae115bc7Smrj } 2166ae115bc7Smrj 21677c478bd9Sstevel@tonic-gate while (vaddr < eaddr) { 21687c478bd9Sstevel@tonic-gate old_pte = htable_walk(hat, &ht, &vaddr, eaddr); 21697c478bd9Sstevel@tonic-gate if (ht == NULL) 21707c478bd9Sstevel@tonic-gate break; 21717c478bd9Sstevel@tonic-gate 21727c478bd9Sstevel@tonic-gate ASSERT(!IN_VA_HOLE(vaddr)); 21737c478bd9Sstevel@tonic-gate 21747c478bd9Sstevel@tonic-gate if (vaddr < (uintptr_t)addr) 21757c478bd9Sstevel@tonic-gate panic("hat_unload_callback(): unmap inside large page"); 21767c478bd9Sstevel@tonic-gate 21777c478bd9Sstevel@tonic-gate /* 21787c478bd9Sstevel@tonic-gate * We'll do the call backs for contiguous ranges 21797c478bd9Sstevel@tonic-gate */ 2180aa2ed9e5Sjosephb if (vaddr != contig_va || 21817c478bd9Sstevel@tonic-gate (r_cnt > 0 && r[r_cnt - 1].rng_level != ht->ht_level)) { 21827c478bd9Sstevel@tonic-gate if (r_cnt == MAX_UNLOAD_CNT) { 21837c478bd9Sstevel@tonic-gate handle_ranges(cb, r_cnt, r); 21847c478bd9Sstevel@tonic-gate r_cnt = 0; 21857c478bd9Sstevel@tonic-gate } 21867c478bd9Sstevel@tonic-gate r[r_cnt].rng_va = vaddr; 21877c478bd9Sstevel@tonic-gate r[r_cnt].rng_cnt = 0; 21887c478bd9Sstevel@tonic-gate r[r_cnt].rng_level = ht->ht_level; 21897c478bd9Sstevel@tonic-gate ++r_cnt; 21907c478bd9Sstevel@tonic-gate } 21917c478bd9Sstevel@tonic-gate 21927c478bd9Sstevel@tonic-gate /* 21937c478bd9Sstevel@tonic-gate * Unload one mapping from the page tables. 21947c478bd9Sstevel@tonic-gate */ 21957c478bd9Sstevel@tonic-gate entry = htable_va2entry(vaddr, ht); 21967c478bd9Sstevel@tonic-gate hat_pte_unmap(ht, entry, flags, old_pte, NULL); 21977c478bd9Sstevel@tonic-gate ASSERT(ht->ht_level <= mmu.max_page_level); 21987c478bd9Sstevel@tonic-gate vaddr += LEVEL_SIZE(ht->ht_level); 2199aa2ed9e5Sjosephb contig_va = vaddr; 22007c478bd9Sstevel@tonic-gate ++r[r_cnt - 1].rng_cnt; 22017c478bd9Sstevel@tonic-gate } 22027c478bd9Sstevel@tonic-gate if (ht) 22037c478bd9Sstevel@tonic-gate htable_release(ht); 22047c478bd9Sstevel@tonic-gate 22057c478bd9Sstevel@tonic-gate /* 22067c478bd9Sstevel@tonic-gate * handle last range for callbacks 22077c478bd9Sstevel@tonic-gate */ 22087c478bd9Sstevel@tonic-gate if (r_cnt > 0) 22097c478bd9Sstevel@tonic-gate handle_ranges(cb, r_cnt, r); 22107c478bd9Sstevel@tonic-gate } 22117c478bd9Sstevel@tonic-gate 22127c478bd9Sstevel@tonic-gate /* 22137c478bd9Sstevel@tonic-gate * synchronize mapping with software data structures 22147c478bd9Sstevel@tonic-gate * 22157c478bd9Sstevel@tonic-gate * This interface is currently only used by the working set monitor 22167c478bd9Sstevel@tonic-gate * driver. 22177c478bd9Sstevel@tonic-gate */ 22187c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 22197c478bd9Sstevel@tonic-gate void 22207c478bd9Sstevel@tonic-gate hat_sync(hat_t *hat, caddr_t addr, size_t len, uint_t flags) 22217c478bd9Sstevel@tonic-gate { 22227c478bd9Sstevel@tonic-gate uintptr_t vaddr = (uintptr_t)addr; 22237c478bd9Sstevel@tonic-gate uintptr_t eaddr = vaddr + len; 22247c478bd9Sstevel@tonic-gate htable_t *ht = NULL; 22257c478bd9Sstevel@tonic-gate uint_t entry; 22267c478bd9Sstevel@tonic-gate x86pte_t pte; 22277c478bd9Sstevel@tonic-gate x86pte_t save_pte; 22287c478bd9Sstevel@tonic-gate x86pte_t new; 22297c478bd9Sstevel@tonic-gate page_t *pp; 22307c478bd9Sstevel@tonic-gate 22317c478bd9Sstevel@tonic-gate ASSERT(!IN_VA_HOLE(vaddr)); 22327c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(vaddr)); 22337c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(eaddr)); 2234ae115bc7Smrj ASSERT(hat == kas.a_hat || eaddr <= _userlimit); 22357c478bd9Sstevel@tonic-gate 22367c478bd9Sstevel@tonic-gate for (; vaddr < eaddr; vaddr += LEVEL_SIZE(ht->ht_level)) { 22377c478bd9Sstevel@tonic-gate try_again: 22387c478bd9Sstevel@tonic-gate pte = htable_walk(hat, &ht, &vaddr, eaddr); 22397c478bd9Sstevel@tonic-gate if (ht == NULL) 22407c478bd9Sstevel@tonic-gate break; 22417c478bd9Sstevel@tonic-gate entry = htable_va2entry(vaddr, ht); 22427c478bd9Sstevel@tonic-gate 2243ae115bc7Smrj if (PTE_GET(pte, PT_SOFTWARE) >= PT_NOSYNC || 22447c478bd9Sstevel@tonic-gate PTE_GET(pte, PT_REF | PT_MOD) == 0) 22457c478bd9Sstevel@tonic-gate continue; 22467c478bd9Sstevel@tonic-gate 22477c478bd9Sstevel@tonic-gate /* 22487c478bd9Sstevel@tonic-gate * We need to acquire the mapping list lock to protect 22497c478bd9Sstevel@tonic-gate * against hat_pageunload(), hat_unload(), etc. 22507c478bd9Sstevel@tonic-gate */ 22517c478bd9Sstevel@tonic-gate pp = page_numtopp_nolock(PTE2PFN(pte, ht->ht_level)); 22527c478bd9Sstevel@tonic-gate if (pp == NULL) 22537c478bd9Sstevel@tonic-gate break; 22547c478bd9Sstevel@tonic-gate x86_hm_enter(pp); 22557c478bd9Sstevel@tonic-gate save_pte = pte; 22567c478bd9Sstevel@tonic-gate pte = x86pte_get(ht, entry); 22577c478bd9Sstevel@tonic-gate if (pte != save_pte) { 22587c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 22597c478bd9Sstevel@tonic-gate goto try_again; 22607c478bd9Sstevel@tonic-gate } 2261ae115bc7Smrj if (PTE_GET(pte, PT_SOFTWARE) >= PT_NOSYNC || 22627c478bd9Sstevel@tonic-gate PTE_GET(pte, PT_REF | PT_MOD) == 0) { 22637c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 22647c478bd9Sstevel@tonic-gate continue; 22657c478bd9Sstevel@tonic-gate } 22667c478bd9Sstevel@tonic-gate 22677c478bd9Sstevel@tonic-gate /* 22687c478bd9Sstevel@tonic-gate * Need to clear ref or mod bits. We may compete with 22697c478bd9Sstevel@tonic-gate * hardware updating the R/M bits and have to try again. 22707c478bd9Sstevel@tonic-gate */ 22717c478bd9Sstevel@tonic-gate if (flags == HAT_SYNC_ZERORM) { 22727c478bd9Sstevel@tonic-gate new = pte; 22737c478bd9Sstevel@tonic-gate PTE_CLR(new, PT_REF | PT_MOD); 22747c478bd9Sstevel@tonic-gate pte = hati_update_pte(ht, entry, pte, new); 22757c478bd9Sstevel@tonic-gate if (pte != 0) { 22767c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 22777c478bd9Sstevel@tonic-gate goto try_again; 22787c478bd9Sstevel@tonic-gate } 22797c478bd9Sstevel@tonic-gate } else { 22807c478bd9Sstevel@tonic-gate /* 22817c478bd9Sstevel@tonic-gate * sync the PTE to the page_t 22827c478bd9Sstevel@tonic-gate */ 22837c478bd9Sstevel@tonic-gate hati_sync_pte_to_page(pp, save_pte, ht->ht_level); 22847c478bd9Sstevel@tonic-gate } 22857c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 22867c478bd9Sstevel@tonic-gate } 22877c478bd9Sstevel@tonic-gate if (ht) 22887c478bd9Sstevel@tonic-gate htable_release(ht); 22897c478bd9Sstevel@tonic-gate } 22907c478bd9Sstevel@tonic-gate 22917c478bd9Sstevel@tonic-gate /* 22927c478bd9Sstevel@tonic-gate * void hat_map(hat, addr, len, flags) 22937c478bd9Sstevel@tonic-gate */ 22947c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 22957c478bd9Sstevel@tonic-gate void 22967c478bd9Sstevel@tonic-gate hat_map(hat_t *hat, caddr_t addr, size_t len, uint_t flags) 22977c478bd9Sstevel@tonic-gate { 22987c478bd9Sstevel@tonic-gate /* does nothing */ 22997c478bd9Sstevel@tonic-gate } 23007c478bd9Sstevel@tonic-gate 23017c478bd9Sstevel@tonic-gate /* 23027c478bd9Sstevel@tonic-gate * uint_t hat_getattr(hat, addr, *attr) 23037c478bd9Sstevel@tonic-gate * returns attr for <hat,addr> in *attr. returns 0 if there was a 23047c478bd9Sstevel@tonic-gate * mapping and *attr is valid, nonzero if there was no mapping and 23057c478bd9Sstevel@tonic-gate * *attr is not valid. 23067c478bd9Sstevel@tonic-gate */ 23077c478bd9Sstevel@tonic-gate uint_t 23087c478bd9Sstevel@tonic-gate hat_getattr(hat_t *hat, caddr_t addr, uint_t *attr) 23097c478bd9Sstevel@tonic-gate { 23107c478bd9Sstevel@tonic-gate uintptr_t vaddr = ALIGN2PAGE(addr); 23117c478bd9Sstevel@tonic-gate htable_t *ht = NULL; 23127c478bd9Sstevel@tonic-gate x86pte_t pte; 23137c478bd9Sstevel@tonic-gate 2314ae115bc7Smrj ASSERT(hat == kas.a_hat || vaddr <= _userlimit); 23157c478bd9Sstevel@tonic-gate 23167c478bd9Sstevel@tonic-gate if (IN_VA_HOLE(vaddr)) 23177c478bd9Sstevel@tonic-gate return ((uint_t)-1); 23187c478bd9Sstevel@tonic-gate 2319ae115bc7Smrj ht = htable_getpte(hat, vaddr, NULL, &pte, mmu.max_page_level); 23207c478bd9Sstevel@tonic-gate if (ht == NULL) 23217c478bd9Sstevel@tonic-gate return ((uint_t)-1); 23227c478bd9Sstevel@tonic-gate 23237c478bd9Sstevel@tonic-gate if (!PTE_ISVALID(pte) || !PTE_ISPAGE(pte, ht->ht_level)) { 23247c478bd9Sstevel@tonic-gate htable_release(ht); 23257c478bd9Sstevel@tonic-gate return ((uint_t)-1); 23267c478bd9Sstevel@tonic-gate } 23277c478bd9Sstevel@tonic-gate 23287c478bd9Sstevel@tonic-gate *attr = PROT_READ; 23297c478bd9Sstevel@tonic-gate if (PTE_GET(pte, PT_WRITABLE)) 23307c478bd9Sstevel@tonic-gate *attr |= PROT_WRITE; 23317c478bd9Sstevel@tonic-gate if (PTE_GET(pte, PT_USER)) 23327c478bd9Sstevel@tonic-gate *attr |= PROT_USER; 23337c478bd9Sstevel@tonic-gate if (!PTE_GET(pte, mmu.pt_nx)) 23347c478bd9Sstevel@tonic-gate *attr |= PROT_EXEC; 2335ae115bc7Smrj if (PTE_GET(pte, PT_SOFTWARE) >= PT_NOSYNC) 23367c478bd9Sstevel@tonic-gate *attr |= HAT_NOSYNC; 23377c478bd9Sstevel@tonic-gate htable_release(ht); 23387c478bd9Sstevel@tonic-gate return (0); 23397c478bd9Sstevel@tonic-gate } 23407c478bd9Sstevel@tonic-gate 23417c478bd9Sstevel@tonic-gate /* 23427c478bd9Sstevel@tonic-gate * hat_updateattr() applies the given attribute change to an existing mapping 23437c478bd9Sstevel@tonic-gate */ 23447c478bd9Sstevel@tonic-gate #define HAT_LOAD_ATTR 1 23457c478bd9Sstevel@tonic-gate #define HAT_SET_ATTR 2 23467c478bd9Sstevel@tonic-gate #define HAT_CLR_ATTR 3 23477c478bd9Sstevel@tonic-gate 23487c478bd9Sstevel@tonic-gate static void 23497c478bd9Sstevel@tonic-gate hat_updateattr(hat_t *hat, caddr_t addr, size_t len, uint_t attr, int what) 23507c478bd9Sstevel@tonic-gate { 23517c478bd9Sstevel@tonic-gate uintptr_t vaddr = (uintptr_t)addr; 23527c478bd9Sstevel@tonic-gate uintptr_t eaddr = (uintptr_t)addr + len; 23537c478bd9Sstevel@tonic-gate htable_t *ht = NULL; 23547c478bd9Sstevel@tonic-gate uint_t entry; 23557c478bd9Sstevel@tonic-gate x86pte_t oldpte, newpte; 23567c478bd9Sstevel@tonic-gate page_t *pp; 23577c478bd9Sstevel@tonic-gate 23587c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(vaddr)); 23597c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(eaddr)); 23607c478bd9Sstevel@tonic-gate ASSERT(hat == kas.a_hat || 23617c478bd9Sstevel@tonic-gate AS_LOCK_HELD(hat->hat_as, &hat->hat_as->a_lock)); 23627c478bd9Sstevel@tonic-gate for (; vaddr < eaddr; vaddr += LEVEL_SIZE(ht->ht_level)) { 23637c478bd9Sstevel@tonic-gate try_again: 23647c478bd9Sstevel@tonic-gate oldpte = htable_walk(hat, &ht, &vaddr, eaddr); 23657c478bd9Sstevel@tonic-gate if (ht == NULL) 23667c478bd9Sstevel@tonic-gate break; 2367ae115bc7Smrj if (PTE_GET(oldpte, PT_SOFTWARE) >= PT_NOCONSIST) 23687c478bd9Sstevel@tonic-gate continue; 23697c478bd9Sstevel@tonic-gate 23707c478bd9Sstevel@tonic-gate pp = page_numtopp_nolock(PTE2PFN(oldpte, ht->ht_level)); 23717c478bd9Sstevel@tonic-gate if (pp == NULL) 23727c478bd9Sstevel@tonic-gate continue; 23737c478bd9Sstevel@tonic-gate x86_hm_enter(pp); 23747c478bd9Sstevel@tonic-gate 23757c478bd9Sstevel@tonic-gate newpte = oldpte; 23767c478bd9Sstevel@tonic-gate /* 23777c478bd9Sstevel@tonic-gate * We found a page table entry in the desired range, 23787c478bd9Sstevel@tonic-gate * figure out the new attributes. 23797c478bd9Sstevel@tonic-gate */ 23807c478bd9Sstevel@tonic-gate if (what == HAT_SET_ATTR || what == HAT_LOAD_ATTR) { 23817c478bd9Sstevel@tonic-gate if ((attr & PROT_WRITE) && 23827c478bd9Sstevel@tonic-gate !PTE_GET(oldpte, PT_WRITABLE)) 23837c478bd9Sstevel@tonic-gate newpte |= PT_WRITABLE; 23847c478bd9Sstevel@tonic-gate 2385ae115bc7Smrj if ((attr & HAT_NOSYNC) && 2386ae115bc7Smrj PTE_GET(oldpte, PT_SOFTWARE) < PT_NOSYNC) 23877c478bd9Sstevel@tonic-gate newpte |= PT_NOSYNC; 23887c478bd9Sstevel@tonic-gate 23897c478bd9Sstevel@tonic-gate if ((attr & PROT_EXEC) && PTE_GET(oldpte, mmu.pt_nx)) 23907c478bd9Sstevel@tonic-gate newpte &= ~mmu.pt_nx; 23917c478bd9Sstevel@tonic-gate } 23927c478bd9Sstevel@tonic-gate 23937c478bd9Sstevel@tonic-gate if (what == HAT_LOAD_ATTR) { 23947c478bd9Sstevel@tonic-gate if (!(attr & PROT_WRITE) && 23957c478bd9Sstevel@tonic-gate PTE_GET(oldpte, PT_WRITABLE)) 23967c478bd9Sstevel@tonic-gate newpte &= ~PT_WRITABLE; 23977c478bd9Sstevel@tonic-gate 2398ae115bc7Smrj if (!(attr & HAT_NOSYNC) && 2399ae115bc7Smrj PTE_GET(oldpte, PT_SOFTWARE) >= PT_NOSYNC) 2400ae115bc7Smrj newpte &= ~PT_SOFTWARE; 24017c478bd9Sstevel@tonic-gate 24027c478bd9Sstevel@tonic-gate if (!(attr & PROT_EXEC) && !PTE_GET(oldpte, mmu.pt_nx)) 24037c478bd9Sstevel@tonic-gate newpte |= mmu.pt_nx; 24047c478bd9Sstevel@tonic-gate } 24057c478bd9Sstevel@tonic-gate 24067c478bd9Sstevel@tonic-gate if (what == HAT_CLR_ATTR) { 24077c478bd9Sstevel@tonic-gate if ((attr & PROT_WRITE) && PTE_GET(oldpte, PT_WRITABLE)) 24087c478bd9Sstevel@tonic-gate newpte &= ~PT_WRITABLE; 24097c478bd9Sstevel@tonic-gate 2410ae115bc7Smrj if ((attr & HAT_NOSYNC) && 2411ae115bc7Smrj PTE_GET(oldpte, PT_SOFTWARE) >= PT_NOSYNC) 2412ae115bc7Smrj newpte &= ~PT_SOFTWARE; 24137c478bd9Sstevel@tonic-gate 24147c478bd9Sstevel@tonic-gate if ((attr & PROT_EXEC) && !PTE_GET(oldpte, mmu.pt_nx)) 24157c478bd9Sstevel@tonic-gate newpte |= mmu.pt_nx; 24167c478bd9Sstevel@tonic-gate } 24177c478bd9Sstevel@tonic-gate 2418ae115bc7Smrj /* 2419ae115bc7Smrj * Ensure NOSYNC/NOCONSIST mappings have REF and MOD set. 2420ae115bc7Smrj * x86pte_set() depends on this. 2421ae115bc7Smrj */ 2422ae115bc7Smrj if (PTE_GET(newpte, PT_SOFTWARE) >= PT_NOSYNC) 2423ae115bc7Smrj newpte |= PT_REF | PT_MOD; 2424ae115bc7Smrj 24257c478bd9Sstevel@tonic-gate /* 24267c478bd9Sstevel@tonic-gate * what about PROT_READ or others? this code only handles: 24277c478bd9Sstevel@tonic-gate * EXEC, WRITE, NOSYNC 24287c478bd9Sstevel@tonic-gate */ 24297c478bd9Sstevel@tonic-gate 24307c478bd9Sstevel@tonic-gate /* 24317c478bd9Sstevel@tonic-gate * If new PTE really changed, update the table. 24327c478bd9Sstevel@tonic-gate */ 24337c478bd9Sstevel@tonic-gate if (newpte != oldpte) { 24347c478bd9Sstevel@tonic-gate entry = htable_va2entry(vaddr, ht); 24357c478bd9Sstevel@tonic-gate oldpte = hati_update_pte(ht, entry, oldpte, newpte); 24367c478bd9Sstevel@tonic-gate if (oldpte != 0) { 24377c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 24387c478bd9Sstevel@tonic-gate goto try_again; 24397c478bd9Sstevel@tonic-gate } 24407c478bd9Sstevel@tonic-gate } 24417c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 24427c478bd9Sstevel@tonic-gate } 24437c478bd9Sstevel@tonic-gate if (ht) 24447c478bd9Sstevel@tonic-gate htable_release(ht); 24457c478bd9Sstevel@tonic-gate } 24467c478bd9Sstevel@tonic-gate 24477c478bd9Sstevel@tonic-gate /* 24487c478bd9Sstevel@tonic-gate * Various wrappers for hat_updateattr() 24497c478bd9Sstevel@tonic-gate */ 24507c478bd9Sstevel@tonic-gate void 24517c478bd9Sstevel@tonic-gate hat_setattr(hat_t *hat, caddr_t addr, size_t len, uint_t attr) 24527c478bd9Sstevel@tonic-gate { 2453ae115bc7Smrj ASSERT(hat == kas.a_hat || (uintptr_t)addr + len <= _userlimit); 24547c478bd9Sstevel@tonic-gate hat_updateattr(hat, addr, len, attr, HAT_SET_ATTR); 24557c478bd9Sstevel@tonic-gate } 24567c478bd9Sstevel@tonic-gate 24577c478bd9Sstevel@tonic-gate void 24587c478bd9Sstevel@tonic-gate hat_clrattr(hat_t *hat, caddr_t addr, size_t len, uint_t attr) 24597c478bd9Sstevel@tonic-gate { 2460ae115bc7Smrj ASSERT(hat == kas.a_hat || (uintptr_t)addr + len <= _userlimit); 24617c478bd9Sstevel@tonic-gate hat_updateattr(hat, addr, len, attr, HAT_CLR_ATTR); 24627c478bd9Sstevel@tonic-gate } 24637c478bd9Sstevel@tonic-gate 24647c478bd9Sstevel@tonic-gate void 24657c478bd9Sstevel@tonic-gate hat_chgattr(hat_t *hat, caddr_t addr, size_t len, uint_t attr) 24667c478bd9Sstevel@tonic-gate { 2467ae115bc7Smrj ASSERT(hat == kas.a_hat || (uintptr_t)addr + len <= _userlimit); 24687c478bd9Sstevel@tonic-gate hat_updateattr(hat, addr, len, attr, HAT_LOAD_ATTR); 24697c478bd9Sstevel@tonic-gate } 24707c478bd9Sstevel@tonic-gate 24717c478bd9Sstevel@tonic-gate void 24727c478bd9Sstevel@tonic-gate hat_chgprot(hat_t *hat, caddr_t addr, size_t len, uint_t vprot) 24737c478bd9Sstevel@tonic-gate { 2474ae115bc7Smrj ASSERT(hat == kas.a_hat || (uintptr_t)addr + len <= _userlimit); 24757c478bd9Sstevel@tonic-gate hat_updateattr(hat, addr, len, vprot & HAT_PROT_MASK, HAT_LOAD_ATTR); 24767c478bd9Sstevel@tonic-gate } 24777c478bd9Sstevel@tonic-gate 24787c478bd9Sstevel@tonic-gate /* 24797c478bd9Sstevel@tonic-gate * size_t hat_getpagesize(hat, addr) 24807c478bd9Sstevel@tonic-gate * returns pagesize in bytes for <hat, addr>. returns -1 of there is 24817c478bd9Sstevel@tonic-gate * no mapping. This is an advisory call. 24827c478bd9Sstevel@tonic-gate */ 24837c478bd9Sstevel@tonic-gate ssize_t 24847c478bd9Sstevel@tonic-gate hat_getpagesize(hat_t *hat, caddr_t addr) 24857c478bd9Sstevel@tonic-gate { 24867c478bd9Sstevel@tonic-gate uintptr_t vaddr = ALIGN2PAGE(addr); 24877c478bd9Sstevel@tonic-gate htable_t *ht; 24887c478bd9Sstevel@tonic-gate size_t pagesize; 24897c478bd9Sstevel@tonic-gate 2490ae115bc7Smrj ASSERT(hat == kas.a_hat || vaddr <= _userlimit); 24917c478bd9Sstevel@tonic-gate if (IN_VA_HOLE(vaddr)) 24927c478bd9Sstevel@tonic-gate return (-1); 24937c478bd9Sstevel@tonic-gate ht = htable_getpage(hat, vaddr, NULL); 24947c478bd9Sstevel@tonic-gate if (ht == NULL) 24957c478bd9Sstevel@tonic-gate return (-1); 24967c478bd9Sstevel@tonic-gate pagesize = LEVEL_SIZE(ht->ht_level); 24977c478bd9Sstevel@tonic-gate htable_release(ht); 24987c478bd9Sstevel@tonic-gate return (pagesize); 24997c478bd9Sstevel@tonic-gate } 25007c478bd9Sstevel@tonic-gate 25017c478bd9Sstevel@tonic-gate 25027c478bd9Sstevel@tonic-gate 25037c478bd9Sstevel@tonic-gate /* 25047c478bd9Sstevel@tonic-gate * pfn_t hat_getpfnum(hat, addr) 25057c478bd9Sstevel@tonic-gate * returns pfn for <hat, addr> or PFN_INVALID if mapping is invalid. 25067c478bd9Sstevel@tonic-gate */ 25077c478bd9Sstevel@tonic-gate pfn_t 25087c478bd9Sstevel@tonic-gate hat_getpfnum(hat_t *hat, caddr_t addr) 25097c478bd9Sstevel@tonic-gate { 25107c478bd9Sstevel@tonic-gate uintptr_t vaddr = ALIGN2PAGE(addr); 25117c478bd9Sstevel@tonic-gate htable_t *ht; 25127c478bd9Sstevel@tonic-gate uint_t entry; 25137c478bd9Sstevel@tonic-gate pfn_t pfn = PFN_INVALID; 25147c478bd9Sstevel@tonic-gate 2515ae115bc7Smrj ASSERT(hat == kas.a_hat || vaddr <= _userlimit); 25167c478bd9Sstevel@tonic-gate if (khat_running == 0) 2517ae115bc7Smrj return (PFN_INVALID); 25187c478bd9Sstevel@tonic-gate 25197c478bd9Sstevel@tonic-gate if (IN_VA_HOLE(vaddr)) 25207c478bd9Sstevel@tonic-gate return (PFN_INVALID); 25217c478bd9Sstevel@tonic-gate 25227c478bd9Sstevel@tonic-gate /* 25237c478bd9Sstevel@tonic-gate * A very common use of hat_getpfnum() is from the DDI for kernel pages. 25247c478bd9Sstevel@tonic-gate * Use the kmap_ptes (which also covers the 32 bit heap) to speed 25257c478bd9Sstevel@tonic-gate * this up. 25267c478bd9Sstevel@tonic-gate */ 25277c478bd9Sstevel@tonic-gate if (mmu.kmap_addr <= vaddr && vaddr < mmu.kmap_eaddr) { 25287c478bd9Sstevel@tonic-gate x86pte_t pte; 2529ae115bc7Smrj pgcnt_t pg_index; 25307c478bd9Sstevel@tonic-gate 2531ae115bc7Smrj pg_index = mmu_btop(vaddr - mmu.kmap_addr); 2532ae115bc7Smrj pte = GET_PTE(PT_INDEX_PTR(mmu.kmap_ptes, pg_index)); 25337c478bd9Sstevel@tonic-gate if (!PTE_ISVALID(pte)) 25347c478bd9Sstevel@tonic-gate return (PFN_INVALID); 25357c478bd9Sstevel@tonic-gate /*LINTED [use of constant 0 causes a silly lint warning] */ 25367c478bd9Sstevel@tonic-gate return (PTE2PFN(pte, 0)); 25377c478bd9Sstevel@tonic-gate } 25387c478bd9Sstevel@tonic-gate 25397c478bd9Sstevel@tonic-gate ht = htable_getpage(hat, vaddr, &entry); 25407c478bd9Sstevel@tonic-gate if (ht == NULL) 25417c478bd9Sstevel@tonic-gate return (PFN_INVALID); 25427c478bd9Sstevel@tonic-gate ASSERT(vaddr >= ht->ht_vaddr); 25437c478bd9Sstevel@tonic-gate ASSERT(vaddr <= HTABLE_LAST_PAGE(ht)); 25447c478bd9Sstevel@tonic-gate pfn = PTE2PFN(x86pte_get(ht, entry), ht->ht_level); 25457c478bd9Sstevel@tonic-gate if (ht->ht_level > 0) 25467c478bd9Sstevel@tonic-gate pfn += mmu_btop(vaddr & LEVEL_OFFSET(ht->ht_level)); 25477c478bd9Sstevel@tonic-gate htable_release(ht); 25487c478bd9Sstevel@tonic-gate return (pfn); 25497c478bd9Sstevel@tonic-gate } 25507c478bd9Sstevel@tonic-gate 25517c478bd9Sstevel@tonic-gate /* 25527c478bd9Sstevel@tonic-gate * hat_getkpfnum() is an obsolete DDI routine, and its use is discouraged. 25537c478bd9Sstevel@tonic-gate * Use hat_getpfnum(kas.a_hat, ...) instead. 25547c478bd9Sstevel@tonic-gate * 25557c478bd9Sstevel@tonic-gate * We'd like to return PFN_INVALID if the mappings have underlying page_t's 25567c478bd9Sstevel@tonic-gate * but can't right now due to the fact that some software has grown to use 25577c478bd9Sstevel@tonic-gate * this interface incorrectly. So for now when the interface is misused, 25587c478bd9Sstevel@tonic-gate * return a warning to the user that in the future it won't work in the 25597c478bd9Sstevel@tonic-gate * way they're abusing it, and carry on. 25607c478bd9Sstevel@tonic-gate * 25617c478bd9Sstevel@tonic-gate * Note that hat_getkpfnum() is never supported on amd64. 25627c478bd9Sstevel@tonic-gate */ 25637c478bd9Sstevel@tonic-gate #if !defined(__amd64) 25647c478bd9Sstevel@tonic-gate pfn_t 25657c478bd9Sstevel@tonic-gate hat_getkpfnum(caddr_t addr) 25667c478bd9Sstevel@tonic-gate { 25677c478bd9Sstevel@tonic-gate pfn_t pfn; 25687c478bd9Sstevel@tonic-gate int badcaller = 0; 25697c478bd9Sstevel@tonic-gate 25707c478bd9Sstevel@tonic-gate if (khat_running == 0) 25717c478bd9Sstevel@tonic-gate panic("hat_getkpfnum(): called too early\n"); 25727c478bd9Sstevel@tonic-gate if ((uintptr_t)addr < kernelbase) 25737c478bd9Sstevel@tonic-gate return (PFN_INVALID); 25747c478bd9Sstevel@tonic-gate 25757c478bd9Sstevel@tonic-gate 25767c478bd9Sstevel@tonic-gate if (segkpm && IS_KPM_ADDR(addr)) { 25777c478bd9Sstevel@tonic-gate badcaller = 1; 25787c478bd9Sstevel@tonic-gate pfn = hat_kpm_va2pfn(addr); 25797c478bd9Sstevel@tonic-gate } else { 25807c478bd9Sstevel@tonic-gate pfn = hat_getpfnum(kas.a_hat, addr); 25817c478bd9Sstevel@tonic-gate badcaller = pf_is_memory(pfn); 25827c478bd9Sstevel@tonic-gate } 25837c478bd9Sstevel@tonic-gate 25847c478bd9Sstevel@tonic-gate if (badcaller) 25857c478bd9Sstevel@tonic-gate hat_getkpfnum_badcall(caller()); 25867c478bd9Sstevel@tonic-gate return (pfn); 25877c478bd9Sstevel@tonic-gate } 25887c478bd9Sstevel@tonic-gate #endif /* __amd64 */ 25897c478bd9Sstevel@tonic-gate 25907c478bd9Sstevel@tonic-gate /* 25917c478bd9Sstevel@tonic-gate * int hat_probe(hat, addr) 25927c478bd9Sstevel@tonic-gate * return 0 if no valid mapping is present. Faster version 25937c478bd9Sstevel@tonic-gate * of hat_getattr in certain architectures. 25947c478bd9Sstevel@tonic-gate */ 25957c478bd9Sstevel@tonic-gate int 25967c478bd9Sstevel@tonic-gate hat_probe(hat_t *hat, caddr_t addr) 25977c478bd9Sstevel@tonic-gate { 25987c478bd9Sstevel@tonic-gate uintptr_t vaddr = ALIGN2PAGE(addr); 25997c478bd9Sstevel@tonic-gate uint_t entry; 26007c478bd9Sstevel@tonic-gate htable_t *ht; 26017c478bd9Sstevel@tonic-gate pgcnt_t pg_off; 26027c478bd9Sstevel@tonic-gate 2603ae115bc7Smrj ASSERT(hat == kas.a_hat || vaddr <= _userlimit); 26047c478bd9Sstevel@tonic-gate ASSERT(hat == kas.a_hat || 26057c478bd9Sstevel@tonic-gate AS_LOCK_HELD(hat->hat_as, &hat->hat_as->a_lock)); 26067c478bd9Sstevel@tonic-gate if (IN_VA_HOLE(vaddr)) 26077c478bd9Sstevel@tonic-gate return (0); 26087c478bd9Sstevel@tonic-gate 26097c478bd9Sstevel@tonic-gate /* 26107c478bd9Sstevel@tonic-gate * Most common use of hat_probe is from segmap. We special case it 26117c478bd9Sstevel@tonic-gate * for performance. 26127c478bd9Sstevel@tonic-gate */ 26137c478bd9Sstevel@tonic-gate if (mmu.kmap_addr <= vaddr && vaddr < mmu.kmap_eaddr) { 26147c478bd9Sstevel@tonic-gate pg_off = mmu_btop(vaddr - mmu.kmap_addr); 26157c478bd9Sstevel@tonic-gate if (mmu.pae_hat) 26167c478bd9Sstevel@tonic-gate return (PTE_ISVALID(mmu.kmap_ptes[pg_off])); 26177c478bd9Sstevel@tonic-gate else 26187c478bd9Sstevel@tonic-gate return (PTE_ISVALID( 26197c478bd9Sstevel@tonic-gate ((x86pte32_t *)mmu.kmap_ptes)[pg_off])); 26207c478bd9Sstevel@tonic-gate } 26217c478bd9Sstevel@tonic-gate 26227c478bd9Sstevel@tonic-gate ht = htable_getpage(hat, vaddr, &entry); 26237c478bd9Sstevel@tonic-gate if (ht == NULL) 26247c478bd9Sstevel@tonic-gate return (0); 26257c478bd9Sstevel@tonic-gate htable_release(ht); 26267c478bd9Sstevel@tonic-gate return (1); 26277c478bd9Sstevel@tonic-gate } 26287c478bd9Sstevel@tonic-gate 26297c478bd9Sstevel@tonic-gate /* 26307c478bd9Sstevel@tonic-gate * Simple implementation of ISM. hat_share() is just like hat_memload_array(), 26317c478bd9Sstevel@tonic-gate * except that we use the ism_hat's existing mappings to determine the pages 26327c478bd9Sstevel@tonic-gate * and protections to use for this hat. In case we find a properly aligned 26337c478bd9Sstevel@tonic-gate * and sized pagetable of 4K mappings, we will attempt to share the pagetable 26347c478bd9Sstevel@tonic-gate * itself. 26357c478bd9Sstevel@tonic-gate */ 26367c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 26377c478bd9Sstevel@tonic-gate int 26387c478bd9Sstevel@tonic-gate hat_share( 26397c478bd9Sstevel@tonic-gate hat_t *hat, 26407c478bd9Sstevel@tonic-gate caddr_t addr, 26417c478bd9Sstevel@tonic-gate hat_t *ism_hat, 26427c478bd9Sstevel@tonic-gate caddr_t src_addr, 26437c478bd9Sstevel@tonic-gate size_t len, /* almost useless value, see below.. */ 26447c478bd9Sstevel@tonic-gate uint_t ismszc) 26457c478bd9Sstevel@tonic-gate { 26467c478bd9Sstevel@tonic-gate uintptr_t vaddr_start = (uintptr_t)addr; 26477c478bd9Sstevel@tonic-gate uintptr_t vaddr; 26487c478bd9Sstevel@tonic-gate uintptr_t pt_vaddr; 26497c478bd9Sstevel@tonic-gate uintptr_t eaddr = vaddr_start + len; 26507c478bd9Sstevel@tonic-gate uintptr_t ism_addr_start = (uintptr_t)src_addr; 26517c478bd9Sstevel@tonic-gate uintptr_t ism_addr = ism_addr_start; 26527c478bd9Sstevel@tonic-gate uintptr_t e_ism_addr = ism_addr + len; 26537c478bd9Sstevel@tonic-gate htable_t *ism_ht = NULL; 26547c478bd9Sstevel@tonic-gate htable_t *ht; 26557c478bd9Sstevel@tonic-gate x86pte_t pte; 26567c478bd9Sstevel@tonic-gate page_t *pp; 26577c478bd9Sstevel@tonic-gate pfn_t pfn; 26587c478bd9Sstevel@tonic-gate level_t l; 26597c478bd9Sstevel@tonic-gate pgcnt_t pgcnt; 26607c478bd9Sstevel@tonic-gate uint_t prot; 26617c478bd9Sstevel@tonic-gate uint_t valid_cnt; 26627c478bd9Sstevel@tonic-gate 26637c478bd9Sstevel@tonic-gate /* 26647c478bd9Sstevel@tonic-gate * We might be asked to share an empty DISM hat by as_dup() 26657c478bd9Sstevel@tonic-gate */ 26667c478bd9Sstevel@tonic-gate ASSERT(hat != kas.a_hat); 2667ae115bc7Smrj ASSERT(eaddr <= _userlimit); 26687c478bd9Sstevel@tonic-gate if (!(ism_hat->hat_flags & HAT_SHARED)) { 26697c478bd9Sstevel@tonic-gate ASSERT(hat_get_mapped_size(ism_hat) == 0); 26707c478bd9Sstevel@tonic-gate return (0); 26717c478bd9Sstevel@tonic-gate } 26727c478bd9Sstevel@tonic-gate 26737c478bd9Sstevel@tonic-gate /* 26747c478bd9Sstevel@tonic-gate * The SPT segment driver often passes us a size larger than there are 26757c478bd9Sstevel@tonic-gate * valid mappings. That's because it rounds the segment size up to a 26767c478bd9Sstevel@tonic-gate * large pagesize, even if the actual memory mapped by ism_hat is less. 26777c478bd9Sstevel@tonic-gate */ 26787c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(vaddr_start)); 26797c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(ism_addr_start)); 26807c478bd9Sstevel@tonic-gate ASSERT(ism_hat->hat_flags & HAT_SHARED); 26817c478bd9Sstevel@tonic-gate while (ism_addr < e_ism_addr) { 26827c478bd9Sstevel@tonic-gate /* 26837c478bd9Sstevel@tonic-gate * use htable_walk to get the next valid ISM mapping 26847c478bd9Sstevel@tonic-gate */ 26857c478bd9Sstevel@tonic-gate pte = htable_walk(ism_hat, &ism_ht, &ism_addr, e_ism_addr); 26867c478bd9Sstevel@tonic-gate if (ism_ht == NULL) 26877c478bd9Sstevel@tonic-gate break; 26887c478bd9Sstevel@tonic-gate 26897c478bd9Sstevel@tonic-gate /* 26907c478bd9Sstevel@tonic-gate * Find the largest page size we can use, based on the 26917c478bd9Sstevel@tonic-gate * ISM mapping size, our address alignment and the remaining 26927c478bd9Sstevel@tonic-gate * map length. 26937c478bd9Sstevel@tonic-gate */ 26947c478bd9Sstevel@tonic-gate vaddr = vaddr_start + (ism_addr - ism_addr_start); 26957c478bd9Sstevel@tonic-gate for (l = ism_ht->ht_level; l > 0; --l) { 26967c478bd9Sstevel@tonic-gate if (LEVEL_SIZE(l) <= eaddr - vaddr && 26977c478bd9Sstevel@tonic-gate (vaddr & LEVEL_OFFSET(l)) == 0) 26987c478bd9Sstevel@tonic-gate break; 26997c478bd9Sstevel@tonic-gate } 27007c478bd9Sstevel@tonic-gate 27017c478bd9Sstevel@tonic-gate /* 27027c478bd9Sstevel@tonic-gate * attempt to share the pagetable 27037c478bd9Sstevel@tonic-gate * 27047c478bd9Sstevel@tonic-gate * - only 4K pagetables are shared (ie. level == 0) 27057c478bd9Sstevel@tonic-gate * - the hat_share() length must cover the whole pagetable 27067c478bd9Sstevel@tonic-gate * - the shared address must align at level 1 27077c478bd9Sstevel@tonic-gate * - a shared PTE for this address already exists OR 27087c478bd9Sstevel@tonic-gate * - no page table for this address exists yet 27097c478bd9Sstevel@tonic-gate */ 27107c478bd9Sstevel@tonic-gate pt_vaddr = 27117c478bd9Sstevel@tonic-gate vaddr_start + (ism_ht->ht_vaddr - ism_addr_start); 27127c478bd9Sstevel@tonic-gate if (ism_ht->ht_level == 0 && 27137c478bd9Sstevel@tonic-gate ism_ht->ht_vaddr + LEVEL_SIZE(1) <= e_ism_addr && 27147c478bd9Sstevel@tonic-gate (pt_vaddr & LEVEL_OFFSET(1)) == 0) { 27157c478bd9Sstevel@tonic-gate 27167c478bd9Sstevel@tonic-gate ht = htable_lookup(hat, pt_vaddr, 0); 27177c478bd9Sstevel@tonic-gate if (ht == NULL) 27187c478bd9Sstevel@tonic-gate ht = htable_create(hat, pt_vaddr, 0, ism_ht); 27197c478bd9Sstevel@tonic-gate 27207c478bd9Sstevel@tonic-gate if (ht->ht_level > 0 || 27217c478bd9Sstevel@tonic-gate !(ht->ht_flags & HTABLE_SHARED_PFN)) { 27227c478bd9Sstevel@tonic-gate 27237c478bd9Sstevel@tonic-gate htable_release(ht); 27247c478bd9Sstevel@tonic-gate 27257c478bd9Sstevel@tonic-gate } else { 27267c478bd9Sstevel@tonic-gate 27277c478bd9Sstevel@tonic-gate /* 27287c478bd9Sstevel@tonic-gate * share the page table 27297c478bd9Sstevel@tonic-gate */ 27307c478bd9Sstevel@tonic-gate ASSERT(ht->ht_level == 0); 27317c478bd9Sstevel@tonic-gate ASSERT(ht->ht_shares == ism_ht); 27327c478bd9Sstevel@tonic-gate valid_cnt = ism_ht->ht_valid_cnt; 27337c478bd9Sstevel@tonic-gate atomic_add_long(&hat->hat_pages_mapped[0], 27347c478bd9Sstevel@tonic-gate valid_cnt - ht->ht_valid_cnt); 27357c478bd9Sstevel@tonic-gate ht->ht_valid_cnt = valid_cnt; 27367c478bd9Sstevel@tonic-gate htable_release(ht); 27377c478bd9Sstevel@tonic-gate ism_addr = ism_ht->ht_vaddr + LEVEL_SIZE(1); 27387c478bd9Sstevel@tonic-gate htable_release(ism_ht); 27397c478bd9Sstevel@tonic-gate ism_ht = NULL; 27407c478bd9Sstevel@tonic-gate continue; 27417c478bd9Sstevel@tonic-gate } 27427c478bd9Sstevel@tonic-gate } 27437c478bd9Sstevel@tonic-gate 27447c478bd9Sstevel@tonic-gate /* 27457c478bd9Sstevel@tonic-gate * Unable to share the page table. Instead we will 27467c478bd9Sstevel@tonic-gate * create new mappings from the values in the ISM mappings. 27477c478bd9Sstevel@tonic-gate * 27487c478bd9Sstevel@tonic-gate * The ISM mapping might be larger than the share area, 27497c478bd9Sstevel@tonic-gate * be careful to trunctate it if needed. 27507c478bd9Sstevel@tonic-gate */ 27517c478bd9Sstevel@tonic-gate if (eaddr - vaddr >= LEVEL_SIZE(ism_ht->ht_level)) { 27527c478bd9Sstevel@tonic-gate pgcnt = mmu_btop(LEVEL_SIZE(ism_ht->ht_level)); 27537c478bd9Sstevel@tonic-gate } else { 27547c478bd9Sstevel@tonic-gate pgcnt = mmu_btop(eaddr - vaddr); 27557c478bd9Sstevel@tonic-gate l = 0; 27567c478bd9Sstevel@tonic-gate } 27577c478bd9Sstevel@tonic-gate 27587c478bd9Sstevel@tonic-gate pfn = PTE2PFN(pte, ism_ht->ht_level); 27597c478bd9Sstevel@tonic-gate ASSERT(pfn != PFN_INVALID); 27607c478bd9Sstevel@tonic-gate while (pgcnt > 0) { 27617c478bd9Sstevel@tonic-gate /* 27627c478bd9Sstevel@tonic-gate * Make a new pte for the PFN for this level. 27637c478bd9Sstevel@tonic-gate * Copy protections for the pte from the ISM pte. 27647c478bd9Sstevel@tonic-gate */ 27657c478bd9Sstevel@tonic-gate pp = page_numtopp_nolock(pfn); 27667c478bd9Sstevel@tonic-gate ASSERT(pp != NULL); 27677c478bd9Sstevel@tonic-gate 27687c478bd9Sstevel@tonic-gate prot = PROT_USER | PROT_READ | HAT_UNORDERED_OK; 27697c478bd9Sstevel@tonic-gate if (PTE_GET(pte, PT_WRITABLE)) 27707c478bd9Sstevel@tonic-gate prot |= PROT_WRITE; 27717c478bd9Sstevel@tonic-gate if (!PTE_GET(pte, PT_NX)) 27727c478bd9Sstevel@tonic-gate prot |= PROT_EXEC; 27737c478bd9Sstevel@tonic-gate 27747c478bd9Sstevel@tonic-gate /* 27757c478bd9Sstevel@tonic-gate * XX64 -- can shm ever be written to swap? 27767c478bd9Sstevel@tonic-gate * if not we could use HAT_NOSYNC here. 27777c478bd9Sstevel@tonic-gate */ 2778ae115bc7Smrj while (hati_load_common(hat, vaddr, pp, prot, HAT_LOAD, 2779ae115bc7Smrj l, pfn) != 0) { 2780ae115bc7Smrj if (l == 0) 2781ae115bc7Smrj panic("hati_load_common() failure"); 2782ae115bc7Smrj --l; 2783ae115bc7Smrj } 27847c478bd9Sstevel@tonic-gate 27857c478bd9Sstevel@tonic-gate vaddr += LEVEL_SIZE(l); 27867c478bd9Sstevel@tonic-gate ism_addr += LEVEL_SIZE(l); 27877c478bd9Sstevel@tonic-gate pfn += mmu_btop(LEVEL_SIZE(l)); 27887c478bd9Sstevel@tonic-gate pgcnt -= mmu_btop(LEVEL_SIZE(l)); 27897c478bd9Sstevel@tonic-gate } 27907c478bd9Sstevel@tonic-gate } 27917c478bd9Sstevel@tonic-gate if (ism_ht != NULL) 27927c478bd9Sstevel@tonic-gate htable_release(ism_ht); 27937c478bd9Sstevel@tonic-gate return (0); 27947c478bd9Sstevel@tonic-gate } 27957c478bd9Sstevel@tonic-gate 27967c478bd9Sstevel@tonic-gate 27977c478bd9Sstevel@tonic-gate /* 27987c478bd9Sstevel@tonic-gate * hat_unshare() is similar to hat_unload_callback(), but 27997c478bd9Sstevel@tonic-gate * we have to look for empty shared pagetables. Note that 28007c478bd9Sstevel@tonic-gate * hat_unshare() is always invoked against an entire segment. 28017c478bd9Sstevel@tonic-gate */ 28027c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 28037c478bd9Sstevel@tonic-gate void 28047c478bd9Sstevel@tonic-gate hat_unshare(hat_t *hat, caddr_t addr, size_t len, uint_t ismszc) 28057c478bd9Sstevel@tonic-gate { 28067c478bd9Sstevel@tonic-gate uintptr_t vaddr = (uintptr_t)addr; 28077c478bd9Sstevel@tonic-gate uintptr_t eaddr = vaddr + len; 28087c478bd9Sstevel@tonic-gate htable_t *ht = NULL; 28097c478bd9Sstevel@tonic-gate uint_t need_demaps = 0; 28107c478bd9Sstevel@tonic-gate 28117c478bd9Sstevel@tonic-gate ASSERT(hat != kas.a_hat); 2812ae115bc7Smrj ASSERT(eaddr <= _userlimit); 28137c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(vaddr)); 28147c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(eaddr)); 28157c478bd9Sstevel@tonic-gate 28167c478bd9Sstevel@tonic-gate /* 28177c478bd9Sstevel@tonic-gate * First go through and remove any shared pagetables. 28187c478bd9Sstevel@tonic-gate * 2819ae115bc7Smrj * Note that it's ok to delay the TLB shootdown till the entire range is 28207c478bd9Sstevel@tonic-gate * finished, because if hat_pageunload() were to unload a shared 2821ae115bc7Smrj * pagetable page, its hat_tlb_inval() will do a global TLB invalidate. 28227c478bd9Sstevel@tonic-gate */ 28237c478bd9Sstevel@tonic-gate while (vaddr < eaddr) { 28247c478bd9Sstevel@tonic-gate ASSERT(!IN_VA_HOLE(vaddr)); 28257c478bd9Sstevel@tonic-gate /* 28267c478bd9Sstevel@tonic-gate * find the pagetable that would map the current address 28277c478bd9Sstevel@tonic-gate */ 28287c478bd9Sstevel@tonic-gate ht = htable_lookup(hat, vaddr, 0); 28297c478bd9Sstevel@tonic-gate if (ht != NULL) { 28307c478bd9Sstevel@tonic-gate if (ht->ht_flags & HTABLE_SHARED_PFN) { 28317c478bd9Sstevel@tonic-gate /* 28327c478bd9Sstevel@tonic-gate * clear mapped pages count, set valid_cnt to 0 28337c478bd9Sstevel@tonic-gate * and let htable_release() finish the job 28347c478bd9Sstevel@tonic-gate */ 28357c478bd9Sstevel@tonic-gate atomic_add_long(&hat->hat_pages_mapped[0], 28367c478bd9Sstevel@tonic-gate -ht->ht_valid_cnt); 28377c478bd9Sstevel@tonic-gate ht->ht_valid_cnt = 0; 28387c478bd9Sstevel@tonic-gate need_demaps = 1; 28397c478bd9Sstevel@tonic-gate } 28407c478bd9Sstevel@tonic-gate htable_release(ht); 28417c478bd9Sstevel@tonic-gate } 28427c478bd9Sstevel@tonic-gate vaddr = (vaddr & LEVEL_MASK(1)) + LEVEL_SIZE(1); 28437c478bd9Sstevel@tonic-gate } 28447c478bd9Sstevel@tonic-gate 28457c478bd9Sstevel@tonic-gate /* 28467c478bd9Sstevel@tonic-gate * flush the TLBs - since we're probably dealing with MANY mappings 28477c478bd9Sstevel@tonic-gate * we do just one CR3 reload. 28487c478bd9Sstevel@tonic-gate */ 28497c478bd9Sstevel@tonic-gate if (!(hat->hat_flags & HAT_FREEING) && need_demaps) 2850ae115bc7Smrj hat_tlb_inval(hat, DEMAP_ALL_ADDR); 28517c478bd9Sstevel@tonic-gate 28527c478bd9Sstevel@tonic-gate /* 28537c478bd9Sstevel@tonic-gate * Now go back and clean up any unaligned mappings that 28547c478bd9Sstevel@tonic-gate * couldn't share pagetables. 28557c478bd9Sstevel@tonic-gate */ 28567c478bd9Sstevel@tonic-gate hat_unload(hat, addr, len, HAT_UNLOAD_UNMAP); 28577c478bd9Sstevel@tonic-gate } 28587c478bd9Sstevel@tonic-gate 28597c478bd9Sstevel@tonic-gate 28607c478bd9Sstevel@tonic-gate /* 28617c478bd9Sstevel@tonic-gate * hat_reserve() does nothing 28627c478bd9Sstevel@tonic-gate */ 28637c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 28647c478bd9Sstevel@tonic-gate void 28657c478bd9Sstevel@tonic-gate hat_reserve(struct as *as, caddr_t addr, size_t len) 28667c478bd9Sstevel@tonic-gate { 28677c478bd9Sstevel@tonic-gate } 28687c478bd9Sstevel@tonic-gate 28697c478bd9Sstevel@tonic-gate 28707c478bd9Sstevel@tonic-gate /* 28717c478bd9Sstevel@tonic-gate * Called when all mappings to a page should have write permission removed. 28727c478bd9Sstevel@tonic-gate * Mostly stolem from hat_pagesync() 28737c478bd9Sstevel@tonic-gate */ 28747c478bd9Sstevel@tonic-gate static void 28757c478bd9Sstevel@tonic-gate hati_page_clrwrt(struct page *pp) 28767c478bd9Sstevel@tonic-gate { 28777c478bd9Sstevel@tonic-gate hment_t *hm = NULL; 28787c478bd9Sstevel@tonic-gate htable_t *ht; 28797c478bd9Sstevel@tonic-gate uint_t entry; 28807c478bd9Sstevel@tonic-gate x86pte_t old; 28817c478bd9Sstevel@tonic-gate x86pte_t new; 28827c478bd9Sstevel@tonic-gate uint_t pszc = 0; 28837c478bd9Sstevel@tonic-gate 28847c478bd9Sstevel@tonic-gate next_size: 28857c478bd9Sstevel@tonic-gate /* 28867c478bd9Sstevel@tonic-gate * walk thru the mapping list clearing write permission 28877c478bd9Sstevel@tonic-gate */ 28887c478bd9Sstevel@tonic-gate x86_hm_enter(pp); 28897c478bd9Sstevel@tonic-gate while ((hm = hment_walk(pp, &ht, &entry, hm)) != NULL) { 28907c478bd9Sstevel@tonic-gate if (ht->ht_level < pszc) 28917c478bd9Sstevel@tonic-gate continue; 28927c478bd9Sstevel@tonic-gate old = x86pte_get(ht, entry); 28937c478bd9Sstevel@tonic-gate 28947c478bd9Sstevel@tonic-gate for (;;) { 28957c478bd9Sstevel@tonic-gate /* 28967c478bd9Sstevel@tonic-gate * Is this mapping of interest? 28977c478bd9Sstevel@tonic-gate */ 28987c478bd9Sstevel@tonic-gate if (PTE2PFN(old, ht->ht_level) != pp->p_pagenum || 28997c478bd9Sstevel@tonic-gate PTE_GET(old, PT_WRITABLE) == 0) 29007c478bd9Sstevel@tonic-gate break; 29017c478bd9Sstevel@tonic-gate 29027c478bd9Sstevel@tonic-gate /* 29037c478bd9Sstevel@tonic-gate * Clear ref/mod writable bits. This requires cross 29047c478bd9Sstevel@tonic-gate * calls to ensure any executing TLBs see cleared bits. 29057c478bd9Sstevel@tonic-gate */ 29067c478bd9Sstevel@tonic-gate new = old; 29077c478bd9Sstevel@tonic-gate PTE_CLR(new, PT_REF | PT_MOD | PT_WRITABLE); 29087c478bd9Sstevel@tonic-gate old = hati_update_pte(ht, entry, old, new); 29097c478bd9Sstevel@tonic-gate if (old != 0) 29107c478bd9Sstevel@tonic-gate continue; 29117c478bd9Sstevel@tonic-gate 29127c478bd9Sstevel@tonic-gate break; 29137c478bd9Sstevel@tonic-gate } 29147c478bd9Sstevel@tonic-gate } 29157c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 29167c478bd9Sstevel@tonic-gate while (pszc < pp->p_szc) { 29177c478bd9Sstevel@tonic-gate page_t *tpp; 29187c478bd9Sstevel@tonic-gate pszc++; 29197c478bd9Sstevel@tonic-gate tpp = PP_GROUPLEADER(pp, pszc); 29207c478bd9Sstevel@tonic-gate if (pp != tpp) { 29217c478bd9Sstevel@tonic-gate pp = tpp; 29227c478bd9Sstevel@tonic-gate goto next_size; 29237c478bd9Sstevel@tonic-gate } 29247c478bd9Sstevel@tonic-gate } 29257c478bd9Sstevel@tonic-gate } 29267c478bd9Sstevel@tonic-gate 29277c478bd9Sstevel@tonic-gate /* 29287c478bd9Sstevel@tonic-gate * void hat_page_setattr(pp, flag) 29297c478bd9Sstevel@tonic-gate * void hat_page_clrattr(pp, flag) 29307c478bd9Sstevel@tonic-gate * used to set/clr ref/mod bits. 29317c478bd9Sstevel@tonic-gate */ 29327c478bd9Sstevel@tonic-gate void 29337c478bd9Sstevel@tonic-gate hat_page_setattr(struct page *pp, uint_t flag) 29347c478bd9Sstevel@tonic-gate { 29357c478bd9Sstevel@tonic-gate vnode_t *vp = pp->p_vnode; 29367c478bd9Sstevel@tonic-gate kmutex_t *vphm = NULL; 29377c478bd9Sstevel@tonic-gate page_t **listp; 29387c478bd9Sstevel@tonic-gate 29397c478bd9Sstevel@tonic-gate if (PP_GETRM(pp, flag) == flag) 29407c478bd9Sstevel@tonic-gate return; 29417c478bd9Sstevel@tonic-gate 29427c478bd9Sstevel@tonic-gate if ((flag & P_MOD) != 0 && vp != NULL && IS_VMODSORT(vp)) { 29437c478bd9Sstevel@tonic-gate vphm = page_vnode_mutex(vp); 29447c478bd9Sstevel@tonic-gate mutex_enter(vphm); 29457c478bd9Sstevel@tonic-gate } 29467c478bd9Sstevel@tonic-gate 29477c478bd9Sstevel@tonic-gate PP_SETRM(pp, flag); 29487c478bd9Sstevel@tonic-gate 29497c478bd9Sstevel@tonic-gate if (vphm != NULL) { 29507c478bd9Sstevel@tonic-gate 29517c478bd9Sstevel@tonic-gate /* 29527c478bd9Sstevel@tonic-gate * Some File Systems examine v_pages for NULL w/o 29537c478bd9Sstevel@tonic-gate * grabbing the vphm mutex. Must not let it become NULL when 29547c478bd9Sstevel@tonic-gate * pp is the only page on the list. 29557c478bd9Sstevel@tonic-gate */ 29567c478bd9Sstevel@tonic-gate if (pp->p_vpnext != pp) { 29577c478bd9Sstevel@tonic-gate page_vpsub(&vp->v_pages, pp); 29587c478bd9Sstevel@tonic-gate if (vp->v_pages != NULL) 29597c478bd9Sstevel@tonic-gate listp = &vp->v_pages->p_vpprev->p_vpnext; 29607c478bd9Sstevel@tonic-gate else 29617c478bd9Sstevel@tonic-gate listp = &vp->v_pages; 29627c478bd9Sstevel@tonic-gate page_vpadd(listp, pp); 29637c478bd9Sstevel@tonic-gate } 29647c478bd9Sstevel@tonic-gate mutex_exit(vphm); 29657c478bd9Sstevel@tonic-gate } 29667c478bd9Sstevel@tonic-gate } 29677c478bd9Sstevel@tonic-gate 29687c478bd9Sstevel@tonic-gate void 29697c478bd9Sstevel@tonic-gate hat_page_clrattr(struct page *pp, uint_t flag) 29707c478bd9Sstevel@tonic-gate { 29717c478bd9Sstevel@tonic-gate vnode_t *vp = pp->p_vnode; 29727c478bd9Sstevel@tonic-gate ASSERT(!(flag & ~(P_MOD | P_REF | P_RO))); 29737c478bd9Sstevel@tonic-gate 29747c478bd9Sstevel@tonic-gate /* 2975a71e32b6Sstans * Caller is expected to hold page's io lock for VMODSORT to work 2976a71e32b6Sstans * correctly with pvn_vplist_dirty() and pvn_getdirty() when mod 2977a71e32b6Sstans * bit is cleared. 2978a71e32b6Sstans * We don't have assert to avoid tripping some existing third party 2979a71e32b6Sstans * code. The dirty page is moved back to top of the v_page list 2980a71e32b6Sstans * after IO is done in pvn_write_done(). 29817c478bd9Sstevel@tonic-gate */ 29827c478bd9Sstevel@tonic-gate PP_CLRRM(pp, flag); 29837c478bd9Sstevel@tonic-gate 2984a71e32b6Sstans if ((flag & P_MOD) != 0 && vp != NULL && IS_VMODSORT(vp)) { 29857c478bd9Sstevel@tonic-gate 29867c478bd9Sstevel@tonic-gate /* 29877c478bd9Sstevel@tonic-gate * VMODSORT works by removing write permissions and getting 29887c478bd9Sstevel@tonic-gate * a fault when a page is made dirty. At this point 29897c478bd9Sstevel@tonic-gate * we need to remove write permission from all mappings 29907c478bd9Sstevel@tonic-gate * to this page. 29917c478bd9Sstevel@tonic-gate */ 29927c478bd9Sstevel@tonic-gate hati_page_clrwrt(pp); 29937c478bd9Sstevel@tonic-gate } 29947c478bd9Sstevel@tonic-gate } 29957c478bd9Sstevel@tonic-gate 29967c478bd9Sstevel@tonic-gate /* 29977c478bd9Sstevel@tonic-gate * If flag is specified, returns 0 if attribute is disabled 29987c478bd9Sstevel@tonic-gate * and non zero if enabled. If flag specifes multiple attributs 29997c478bd9Sstevel@tonic-gate * then returns 0 if ALL atriibutes are disabled. This is an advisory 30007c478bd9Sstevel@tonic-gate * call. 30017c478bd9Sstevel@tonic-gate */ 30027c478bd9Sstevel@tonic-gate uint_t 30037c478bd9Sstevel@tonic-gate hat_page_getattr(struct page *pp, uint_t flag) 30047c478bd9Sstevel@tonic-gate { 30057c478bd9Sstevel@tonic-gate return (PP_GETRM(pp, flag)); 30067c478bd9Sstevel@tonic-gate } 30077c478bd9Sstevel@tonic-gate 30087c478bd9Sstevel@tonic-gate 30097c478bd9Sstevel@tonic-gate /* 30107c478bd9Sstevel@tonic-gate * common code used by hat_pageunload() and hment_steal() 30117c478bd9Sstevel@tonic-gate */ 30127c478bd9Sstevel@tonic-gate hment_t * 30137c478bd9Sstevel@tonic-gate hati_page_unmap(page_t *pp, htable_t *ht, uint_t entry) 30147c478bd9Sstevel@tonic-gate { 30157c478bd9Sstevel@tonic-gate x86pte_t old_pte; 30167c478bd9Sstevel@tonic-gate pfn_t pfn = pp->p_pagenum; 30177c478bd9Sstevel@tonic-gate hment_t *hm; 30187c478bd9Sstevel@tonic-gate 30197c478bd9Sstevel@tonic-gate /* 30207c478bd9Sstevel@tonic-gate * We need to acquire a hold on the htable in order to 30217c478bd9Sstevel@tonic-gate * do the invalidate. We know the htable must exist, since 30227c478bd9Sstevel@tonic-gate * unmap's don't release the htable until after removing any 30237c478bd9Sstevel@tonic-gate * hment. Having x86_hm_enter() keeps that from proceeding. 30247c478bd9Sstevel@tonic-gate */ 30257c478bd9Sstevel@tonic-gate htable_acquire(ht); 30267c478bd9Sstevel@tonic-gate 30277c478bd9Sstevel@tonic-gate /* 30287c478bd9Sstevel@tonic-gate * Invalidate the PTE and remove the hment. 30297c478bd9Sstevel@tonic-gate */ 3030ae115bc7Smrj old_pte = x86pte_inval(ht, entry, 0, NULL); 3031aa2ed9e5Sjosephb if (PTE2PFN(old_pte, ht->ht_level) != pfn) { 3032ae115bc7Smrj panic("x86pte_inval() failure found PTE = " FMT_PTE 3033aa2ed9e5Sjosephb " pfn being unmapped is %lx ht=0x%lx entry=0x%x", 3034aa2ed9e5Sjosephb old_pte, pfn, (uintptr_t)ht, entry); 3035aa2ed9e5Sjosephb } 30367c478bd9Sstevel@tonic-gate 30377c478bd9Sstevel@tonic-gate /* 30387c478bd9Sstevel@tonic-gate * Clean up all the htable information for this mapping 30397c478bd9Sstevel@tonic-gate */ 30407c478bd9Sstevel@tonic-gate ASSERT(ht->ht_valid_cnt > 0); 30417c478bd9Sstevel@tonic-gate HTABLE_DEC(ht->ht_valid_cnt); 30427c478bd9Sstevel@tonic-gate PGCNT_DEC(ht->ht_hat, ht->ht_level); 30437c478bd9Sstevel@tonic-gate 30447c478bd9Sstevel@tonic-gate /* 30457c478bd9Sstevel@tonic-gate * sync ref/mod bits to the page_t 30467c478bd9Sstevel@tonic-gate */ 3047ae115bc7Smrj if (PTE_GET(old_pte, PT_SOFTWARE) < PT_NOSYNC) 30487c478bd9Sstevel@tonic-gate hati_sync_pte_to_page(pp, old_pte, ht->ht_level); 30497c478bd9Sstevel@tonic-gate 30507c478bd9Sstevel@tonic-gate /* 30517c478bd9Sstevel@tonic-gate * Remove the mapping list entry for this page. 30527c478bd9Sstevel@tonic-gate */ 30537c478bd9Sstevel@tonic-gate hm = hment_remove(pp, ht, entry); 30547c478bd9Sstevel@tonic-gate 30557c478bd9Sstevel@tonic-gate /* 30567c478bd9Sstevel@tonic-gate * drop the mapping list lock so that we might free the 30577c478bd9Sstevel@tonic-gate * hment and htable. 30587c478bd9Sstevel@tonic-gate */ 30597c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 30607c478bd9Sstevel@tonic-gate htable_release(ht); 30617c478bd9Sstevel@tonic-gate return (hm); 30627c478bd9Sstevel@tonic-gate } 30637c478bd9Sstevel@tonic-gate 3064a5652762Spraks extern int vpm_enable; 30657c478bd9Sstevel@tonic-gate /* 30667c478bd9Sstevel@tonic-gate * Unload all translations to a page. If the page is a subpage of a large 30677c478bd9Sstevel@tonic-gate * page, the large page mappings are also removed. 30687c478bd9Sstevel@tonic-gate * 30697c478bd9Sstevel@tonic-gate * The forceflags are unused. 30707c478bd9Sstevel@tonic-gate */ 30717c478bd9Sstevel@tonic-gate 30727c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 30737c478bd9Sstevel@tonic-gate static int 30747c478bd9Sstevel@tonic-gate hati_pageunload(struct page *pp, uint_t pg_szcd, uint_t forceflag) 30757c478bd9Sstevel@tonic-gate { 30767c478bd9Sstevel@tonic-gate page_t *cur_pp = pp; 30777c478bd9Sstevel@tonic-gate hment_t *hm; 30787c478bd9Sstevel@tonic-gate hment_t *prev; 30797c478bd9Sstevel@tonic-gate htable_t *ht; 30807c478bd9Sstevel@tonic-gate uint_t entry; 30817c478bd9Sstevel@tonic-gate level_t level; 30827c478bd9Sstevel@tonic-gate 3083a5652762Spraks #if defined(__amd64) 3084a5652762Spraks /* 3085a5652762Spraks * clear the vpm ref. 3086a5652762Spraks */ 3087a5652762Spraks if (vpm_enable) { 3088a5652762Spraks pp->p_vpmref = 0; 3089a5652762Spraks } 3090a5652762Spraks #endif 30917c478bd9Sstevel@tonic-gate /* 30927c478bd9Sstevel@tonic-gate * The loop with next_size handles pages with multiple pagesize mappings 30937c478bd9Sstevel@tonic-gate */ 30947c478bd9Sstevel@tonic-gate next_size: 30957c478bd9Sstevel@tonic-gate for (;;) { 30967c478bd9Sstevel@tonic-gate 30977c478bd9Sstevel@tonic-gate /* 30987c478bd9Sstevel@tonic-gate * Get a mapping list entry 30997c478bd9Sstevel@tonic-gate */ 31007c478bd9Sstevel@tonic-gate x86_hm_enter(cur_pp); 31017c478bd9Sstevel@tonic-gate for (prev = NULL; ; prev = hm) { 31027c478bd9Sstevel@tonic-gate hm = hment_walk(cur_pp, &ht, &entry, prev); 31037c478bd9Sstevel@tonic-gate if (hm == NULL) { 31047c478bd9Sstevel@tonic-gate x86_hm_exit(cur_pp); 31057c478bd9Sstevel@tonic-gate 31067c478bd9Sstevel@tonic-gate /* 31077c478bd9Sstevel@tonic-gate * If not part of a larger page, we're done. 31087c478bd9Sstevel@tonic-gate */ 3109ae115bc7Smrj if (cur_pp->p_szc <= pg_szcd) { 31107c478bd9Sstevel@tonic-gate return (0); 3111ae115bc7Smrj } 31127c478bd9Sstevel@tonic-gate 31137c478bd9Sstevel@tonic-gate /* 31147c478bd9Sstevel@tonic-gate * Else check the next larger page size. 31157c478bd9Sstevel@tonic-gate * hat_page_demote() may decrease p_szc 31167c478bd9Sstevel@tonic-gate * but that's ok we'll just take an extra 31177c478bd9Sstevel@tonic-gate * trip discover there're no larger mappings 31187c478bd9Sstevel@tonic-gate * and return. 31197c478bd9Sstevel@tonic-gate */ 31207c478bd9Sstevel@tonic-gate ++pg_szcd; 31217c478bd9Sstevel@tonic-gate cur_pp = PP_GROUPLEADER(cur_pp, pg_szcd); 31227c478bd9Sstevel@tonic-gate goto next_size; 31237c478bd9Sstevel@tonic-gate } 31247c478bd9Sstevel@tonic-gate 31257c478bd9Sstevel@tonic-gate /* 31267c478bd9Sstevel@tonic-gate * If this mapping size matches, remove it. 31277c478bd9Sstevel@tonic-gate */ 31287c478bd9Sstevel@tonic-gate level = ht->ht_level; 31297c478bd9Sstevel@tonic-gate if (level == pg_szcd) 31307c478bd9Sstevel@tonic-gate break; 31317c478bd9Sstevel@tonic-gate } 31327c478bd9Sstevel@tonic-gate 31337c478bd9Sstevel@tonic-gate /* 31347c478bd9Sstevel@tonic-gate * Remove the mapping list entry for this page. 31357c478bd9Sstevel@tonic-gate * Note this does the x86_hm_exit() for us. 31367c478bd9Sstevel@tonic-gate */ 31377c478bd9Sstevel@tonic-gate hm = hati_page_unmap(cur_pp, ht, entry); 31387c478bd9Sstevel@tonic-gate if (hm != NULL) 31397c478bd9Sstevel@tonic-gate hment_free(hm); 31407c478bd9Sstevel@tonic-gate } 31417c478bd9Sstevel@tonic-gate } 31427c478bd9Sstevel@tonic-gate 31437c478bd9Sstevel@tonic-gate int 31447c478bd9Sstevel@tonic-gate hat_pageunload(struct page *pp, uint_t forceflag) 31457c478bd9Sstevel@tonic-gate { 31467c478bd9Sstevel@tonic-gate ASSERT(PAGE_EXCL(pp)); 31477c478bd9Sstevel@tonic-gate return (hati_pageunload(pp, 0, forceflag)); 31487c478bd9Sstevel@tonic-gate } 31497c478bd9Sstevel@tonic-gate 31507c478bd9Sstevel@tonic-gate /* 31517c478bd9Sstevel@tonic-gate * Unload all large mappings to pp and reduce by 1 p_szc field of every large 31527c478bd9Sstevel@tonic-gate * page level that included pp. 31537c478bd9Sstevel@tonic-gate * 31547c478bd9Sstevel@tonic-gate * pp must be locked EXCL. Even though no other constituent pages are locked 31557c478bd9Sstevel@tonic-gate * it's legal to unload large mappings to pp because all constituent pages of 31567c478bd9Sstevel@tonic-gate * large locked mappings have to be locked SHARED. therefore if we have EXCL 31577c478bd9Sstevel@tonic-gate * lock on one of constituent pages none of the large mappings to pp are 31587c478bd9Sstevel@tonic-gate * locked. 31597c478bd9Sstevel@tonic-gate * 31607c478bd9Sstevel@tonic-gate * Change (always decrease) p_szc field starting from the last constituent 31617c478bd9Sstevel@tonic-gate * page and ending with root constituent page so that root's pszc always shows 31627c478bd9Sstevel@tonic-gate * the area where hat_page_demote() may be active. 31637c478bd9Sstevel@tonic-gate * 31647c478bd9Sstevel@tonic-gate * This mechanism is only used for file system pages where it's not always 31657c478bd9Sstevel@tonic-gate * possible to get EXCL locks on all constituent pages to demote the size code 31667c478bd9Sstevel@tonic-gate * (as is done for anonymous or kernel large pages). 31677c478bd9Sstevel@tonic-gate */ 31687c478bd9Sstevel@tonic-gate void 31697c478bd9Sstevel@tonic-gate hat_page_demote(page_t *pp) 31707c478bd9Sstevel@tonic-gate { 31717c478bd9Sstevel@tonic-gate uint_t pszc; 31727c478bd9Sstevel@tonic-gate uint_t rszc; 31737c478bd9Sstevel@tonic-gate uint_t szc; 31747c478bd9Sstevel@tonic-gate page_t *rootpp; 31757c478bd9Sstevel@tonic-gate page_t *firstpp; 31767c478bd9Sstevel@tonic-gate page_t *lastpp; 31777c478bd9Sstevel@tonic-gate pgcnt_t pgcnt; 31787c478bd9Sstevel@tonic-gate 31797c478bd9Sstevel@tonic-gate ASSERT(PAGE_EXCL(pp)); 31807c478bd9Sstevel@tonic-gate ASSERT(!PP_ISFREE(pp)); 31817c478bd9Sstevel@tonic-gate ASSERT(page_szc_lock_assert(pp)); 31827c478bd9Sstevel@tonic-gate 31837c478bd9Sstevel@tonic-gate if (pp->p_szc == 0) 31847c478bd9Sstevel@tonic-gate return; 31857c478bd9Sstevel@tonic-gate 31867c478bd9Sstevel@tonic-gate rootpp = PP_GROUPLEADER(pp, 1); 31877c478bd9Sstevel@tonic-gate (void) hati_pageunload(rootpp, 1, HAT_FORCE_PGUNLOAD); 31887c478bd9Sstevel@tonic-gate 31897c478bd9Sstevel@tonic-gate /* 31907c478bd9Sstevel@tonic-gate * all large mappings to pp are gone 31917c478bd9Sstevel@tonic-gate * and no new can be setup since pp is locked exclusively. 31927c478bd9Sstevel@tonic-gate * 31937c478bd9Sstevel@tonic-gate * Lock the root to make sure there's only one hat_page_demote() 31947c478bd9Sstevel@tonic-gate * outstanding within the area of this root's pszc. 31957c478bd9Sstevel@tonic-gate * 31967c478bd9Sstevel@tonic-gate * Second potential hat_page_demote() is already eliminated by upper 31977c478bd9Sstevel@tonic-gate * VM layer via page_szc_lock() but we don't rely on it and use our 31987c478bd9Sstevel@tonic-gate * own locking (so that upper layer locking can be changed without 31997c478bd9Sstevel@tonic-gate * assumptions that hat depends on upper layer VM to prevent multiple 32007c478bd9Sstevel@tonic-gate * hat_page_demote() to be issued simultaneously to the same large 32017c478bd9Sstevel@tonic-gate * page). 32027c478bd9Sstevel@tonic-gate */ 32037c478bd9Sstevel@tonic-gate again: 32047c478bd9Sstevel@tonic-gate pszc = pp->p_szc; 32057c478bd9Sstevel@tonic-gate if (pszc == 0) 32067c478bd9Sstevel@tonic-gate return; 32077c478bd9Sstevel@tonic-gate rootpp = PP_GROUPLEADER(pp, pszc); 32087c478bd9Sstevel@tonic-gate x86_hm_enter(rootpp); 32097c478bd9Sstevel@tonic-gate /* 32107c478bd9Sstevel@tonic-gate * If root's p_szc is different from pszc we raced with another 32117c478bd9Sstevel@tonic-gate * hat_page_demote(). Drop the lock and try to find the root again. 32127c478bd9Sstevel@tonic-gate * If root's p_szc is greater than pszc previous hat_page_demote() is 32137c478bd9Sstevel@tonic-gate * not done yet. Take and release mlist lock of root's root to wait 32147c478bd9Sstevel@tonic-gate * for previous hat_page_demote() to complete. 32157c478bd9Sstevel@tonic-gate */ 32167c478bd9Sstevel@tonic-gate if ((rszc = rootpp->p_szc) != pszc) { 32177c478bd9Sstevel@tonic-gate x86_hm_exit(rootpp); 32187c478bd9Sstevel@tonic-gate if (rszc > pszc) { 32197c478bd9Sstevel@tonic-gate /* p_szc of a locked non free page can't increase */ 32207c478bd9Sstevel@tonic-gate ASSERT(pp != rootpp); 32217c478bd9Sstevel@tonic-gate 32227c478bd9Sstevel@tonic-gate rootpp = PP_GROUPLEADER(rootpp, rszc); 32237c478bd9Sstevel@tonic-gate x86_hm_enter(rootpp); 32247c478bd9Sstevel@tonic-gate x86_hm_exit(rootpp); 32257c478bd9Sstevel@tonic-gate } 32267c478bd9Sstevel@tonic-gate goto again; 32277c478bd9Sstevel@tonic-gate } 32287c478bd9Sstevel@tonic-gate ASSERT(pp->p_szc == pszc); 32297c478bd9Sstevel@tonic-gate 32307c478bd9Sstevel@tonic-gate /* 32317c478bd9Sstevel@tonic-gate * Decrement by 1 p_szc of every constituent page of a region that 32327c478bd9Sstevel@tonic-gate * covered pp. For example if original szc is 3 it gets changed to 2 32337c478bd9Sstevel@tonic-gate * everywhere except in region 2 that covered pp. Region 2 that 32347c478bd9Sstevel@tonic-gate * covered pp gets demoted to 1 everywhere except in region 1 that 32357c478bd9Sstevel@tonic-gate * covered pp. The region 1 that covered pp is demoted to region 32367c478bd9Sstevel@tonic-gate * 0. It's done this way because from region 3 we removed level 3 32377c478bd9Sstevel@tonic-gate * mappings, from region 2 that covered pp we removed level 2 mappings 32387c478bd9Sstevel@tonic-gate * and from region 1 that covered pp we removed level 1 mappings. All 32397c478bd9Sstevel@tonic-gate * changes are done from from high pfn's to low pfn's so that roots 32407c478bd9Sstevel@tonic-gate * are changed last allowing one to know the largest region where 32417c478bd9Sstevel@tonic-gate * hat_page_demote() is stil active by only looking at the root page. 32427c478bd9Sstevel@tonic-gate * 32437c478bd9Sstevel@tonic-gate * This algorithm is implemented in 2 while loops. First loop changes 32447c478bd9Sstevel@tonic-gate * p_szc of pages to the right of pp's level 1 region and second 32457c478bd9Sstevel@tonic-gate * loop changes p_szc of pages of level 1 region that covers pp 32467c478bd9Sstevel@tonic-gate * and all pages to the left of level 1 region that covers pp. 32477c478bd9Sstevel@tonic-gate * In the first loop p_szc keeps dropping with every iteration 32487c478bd9Sstevel@tonic-gate * and in the second loop it keeps increasing with every iteration. 32497c478bd9Sstevel@tonic-gate * 32507c478bd9Sstevel@tonic-gate * First loop description: Demote pages to the right of pp outside of 32517c478bd9Sstevel@tonic-gate * level 1 region that covers pp. In every iteration of the while 32527c478bd9Sstevel@tonic-gate * loop below find the last page of szc region and the first page of 32537c478bd9Sstevel@tonic-gate * (szc - 1) region that is immediately to the right of (szc - 1) 32547c478bd9Sstevel@tonic-gate * region that covers pp. From last such page to first such page 32557c478bd9Sstevel@tonic-gate * change every page's szc to szc - 1. Decrement szc and continue 32567c478bd9Sstevel@tonic-gate * looping until szc is 1. If pp belongs to the last (szc - 1) region 32577c478bd9Sstevel@tonic-gate * of szc region skip to the next iteration. 32587c478bd9Sstevel@tonic-gate */ 32597c478bd9Sstevel@tonic-gate szc = pszc; 32607c478bd9Sstevel@tonic-gate while (szc > 1) { 32617c478bd9Sstevel@tonic-gate lastpp = PP_GROUPLEADER(pp, szc); 32627c478bd9Sstevel@tonic-gate pgcnt = page_get_pagecnt(szc); 32637c478bd9Sstevel@tonic-gate lastpp += pgcnt - 1; 32647c478bd9Sstevel@tonic-gate firstpp = PP_GROUPLEADER(pp, (szc - 1)); 32657c478bd9Sstevel@tonic-gate pgcnt = page_get_pagecnt(szc - 1); 32667c478bd9Sstevel@tonic-gate if (lastpp - firstpp < pgcnt) { 32677c478bd9Sstevel@tonic-gate szc--; 32687c478bd9Sstevel@tonic-gate continue; 32697c478bd9Sstevel@tonic-gate } 32707c478bd9Sstevel@tonic-gate firstpp += pgcnt; 32717c478bd9Sstevel@tonic-gate while (lastpp != firstpp) { 32727c478bd9Sstevel@tonic-gate ASSERT(lastpp->p_szc == pszc); 32737c478bd9Sstevel@tonic-gate lastpp->p_szc = szc - 1; 32747c478bd9Sstevel@tonic-gate lastpp--; 32757c478bd9Sstevel@tonic-gate } 32767c478bd9Sstevel@tonic-gate firstpp->p_szc = szc - 1; 32777c478bd9Sstevel@tonic-gate szc--; 32787c478bd9Sstevel@tonic-gate } 32797c478bd9Sstevel@tonic-gate 32807c478bd9Sstevel@tonic-gate /* 32817c478bd9Sstevel@tonic-gate * Second loop description: 32827c478bd9Sstevel@tonic-gate * First iteration changes p_szc to 0 of every 32837c478bd9Sstevel@tonic-gate * page of level 1 region that covers pp. 32847c478bd9Sstevel@tonic-gate * Subsequent iterations find last page of szc region 32857c478bd9Sstevel@tonic-gate * immediately to the left of szc region that covered pp 32867c478bd9Sstevel@tonic-gate * and first page of (szc + 1) region that covers pp. 32877c478bd9Sstevel@tonic-gate * From last to first page change p_szc of every page to szc. 32887c478bd9Sstevel@tonic-gate * Increment szc and continue looping until szc is pszc. 32897c478bd9Sstevel@tonic-gate * If pp belongs to the fist szc region of (szc + 1) region 32907c478bd9Sstevel@tonic-gate * skip to the next iteration. 32917c478bd9Sstevel@tonic-gate * 32927c478bd9Sstevel@tonic-gate */ 32937c478bd9Sstevel@tonic-gate szc = 0; 32947c478bd9Sstevel@tonic-gate while (szc < pszc) { 32957c478bd9Sstevel@tonic-gate firstpp = PP_GROUPLEADER(pp, (szc + 1)); 32967c478bd9Sstevel@tonic-gate if (szc == 0) { 32977c478bd9Sstevel@tonic-gate pgcnt = page_get_pagecnt(1); 32987c478bd9Sstevel@tonic-gate lastpp = firstpp + (pgcnt - 1); 32997c478bd9Sstevel@tonic-gate } else { 33007c478bd9Sstevel@tonic-gate lastpp = PP_GROUPLEADER(pp, szc); 33017c478bd9Sstevel@tonic-gate if (firstpp == lastpp) { 33027c478bd9Sstevel@tonic-gate szc++; 33037c478bd9Sstevel@tonic-gate continue; 33047c478bd9Sstevel@tonic-gate } 33057c478bd9Sstevel@tonic-gate lastpp--; 33067c478bd9Sstevel@tonic-gate pgcnt = page_get_pagecnt(szc); 33077c478bd9Sstevel@tonic-gate } 33087c478bd9Sstevel@tonic-gate while (lastpp != firstpp) { 33097c478bd9Sstevel@tonic-gate ASSERT(lastpp->p_szc == pszc); 33107c478bd9Sstevel@tonic-gate lastpp->p_szc = szc; 33117c478bd9Sstevel@tonic-gate lastpp--; 33127c478bd9Sstevel@tonic-gate } 33137c478bd9Sstevel@tonic-gate firstpp->p_szc = szc; 33147c478bd9Sstevel@tonic-gate if (firstpp == rootpp) 33157c478bd9Sstevel@tonic-gate break; 33167c478bd9Sstevel@tonic-gate szc++; 33177c478bd9Sstevel@tonic-gate } 33187c478bd9Sstevel@tonic-gate x86_hm_exit(rootpp); 33197c478bd9Sstevel@tonic-gate } 33207c478bd9Sstevel@tonic-gate 33217c478bd9Sstevel@tonic-gate /* 33227c478bd9Sstevel@tonic-gate * get hw stats from hardware into page struct and reset hw stats 33237c478bd9Sstevel@tonic-gate * returns attributes of page 33247c478bd9Sstevel@tonic-gate * Flags for hat_pagesync, hat_getstat, hat_sync 33257c478bd9Sstevel@tonic-gate * 33267c478bd9Sstevel@tonic-gate * define HAT_SYNC_ZERORM 0x01 33277c478bd9Sstevel@tonic-gate * 33287c478bd9Sstevel@tonic-gate * Additional flags for hat_pagesync 33297c478bd9Sstevel@tonic-gate * 33307c478bd9Sstevel@tonic-gate * define HAT_SYNC_STOPON_REF 0x02 33317c478bd9Sstevel@tonic-gate * define HAT_SYNC_STOPON_MOD 0x04 33327c478bd9Sstevel@tonic-gate * define HAT_SYNC_STOPON_RM 0x06 33337c478bd9Sstevel@tonic-gate * define HAT_SYNC_STOPON_SHARED 0x08 33347c478bd9Sstevel@tonic-gate */ 33357c478bd9Sstevel@tonic-gate uint_t 33367c478bd9Sstevel@tonic-gate hat_pagesync(struct page *pp, uint_t flags) 33377c478bd9Sstevel@tonic-gate { 33387c478bd9Sstevel@tonic-gate hment_t *hm = NULL; 33397c478bd9Sstevel@tonic-gate htable_t *ht; 33407c478bd9Sstevel@tonic-gate uint_t entry; 33417c478bd9Sstevel@tonic-gate x86pte_t old, save_old; 33427c478bd9Sstevel@tonic-gate x86pte_t new; 33437c478bd9Sstevel@tonic-gate uchar_t nrmbits = P_REF|P_MOD|P_RO; 33447c478bd9Sstevel@tonic-gate extern ulong_t po_share; 33457c478bd9Sstevel@tonic-gate page_t *save_pp = pp; 33467c478bd9Sstevel@tonic-gate uint_t pszc = 0; 33477c478bd9Sstevel@tonic-gate 33487c478bd9Sstevel@tonic-gate ASSERT(PAGE_LOCKED(pp) || panicstr); 33497c478bd9Sstevel@tonic-gate 33507c478bd9Sstevel@tonic-gate if (PP_ISRO(pp) && (flags & HAT_SYNC_STOPON_MOD)) 33517c478bd9Sstevel@tonic-gate return (pp->p_nrm & nrmbits); 33527c478bd9Sstevel@tonic-gate 33537c478bd9Sstevel@tonic-gate if ((flags & HAT_SYNC_ZERORM) == 0) { 33547c478bd9Sstevel@tonic-gate 33557c478bd9Sstevel@tonic-gate if ((flags & HAT_SYNC_STOPON_REF) != 0 && PP_ISREF(pp)) 33567c478bd9Sstevel@tonic-gate return (pp->p_nrm & nrmbits); 33577c478bd9Sstevel@tonic-gate 33587c478bd9Sstevel@tonic-gate if ((flags & HAT_SYNC_STOPON_MOD) != 0 && PP_ISMOD(pp)) 33597c478bd9Sstevel@tonic-gate return (pp->p_nrm & nrmbits); 33607c478bd9Sstevel@tonic-gate 33617c478bd9Sstevel@tonic-gate if ((flags & HAT_SYNC_STOPON_SHARED) != 0 && 33627c478bd9Sstevel@tonic-gate hat_page_getshare(pp) > po_share) { 33637c478bd9Sstevel@tonic-gate if (PP_ISRO(pp)) 33647c478bd9Sstevel@tonic-gate PP_SETREF(pp); 33657c478bd9Sstevel@tonic-gate return (pp->p_nrm & nrmbits); 33667c478bd9Sstevel@tonic-gate } 33677c478bd9Sstevel@tonic-gate } 33687c478bd9Sstevel@tonic-gate 33697c478bd9Sstevel@tonic-gate next_size: 33707c478bd9Sstevel@tonic-gate /* 33717c478bd9Sstevel@tonic-gate * walk thru the mapping list syncing (and clearing) ref/mod bits. 33727c478bd9Sstevel@tonic-gate */ 33737c478bd9Sstevel@tonic-gate x86_hm_enter(pp); 33747c478bd9Sstevel@tonic-gate while ((hm = hment_walk(pp, &ht, &entry, hm)) != NULL) { 33757c478bd9Sstevel@tonic-gate if (ht->ht_level < pszc) 33767c478bd9Sstevel@tonic-gate continue; 33777c478bd9Sstevel@tonic-gate old = x86pte_get(ht, entry); 33787c478bd9Sstevel@tonic-gate try_again: 33797c478bd9Sstevel@tonic-gate 33807c478bd9Sstevel@tonic-gate ASSERT(PTE2PFN(old, ht->ht_level) == pp->p_pagenum); 33817c478bd9Sstevel@tonic-gate 33827c478bd9Sstevel@tonic-gate if (PTE_GET(old, PT_REF | PT_MOD) == 0) 33837c478bd9Sstevel@tonic-gate continue; 33847c478bd9Sstevel@tonic-gate 33857c478bd9Sstevel@tonic-gate save_old = old; 33867c478bd9Sstevel@tonic-gate if ((flags & HAT_SYNC_ZERORM) != 0) { 33877c478bd9Sstevel@tonic-gate 33887c478bd9Sstevel@tonic-gate /* 33897c478bd9Sstevel@tonic-gate * Need to clear ref or mod bits. Need to demap 33907c478bd9Sstevel@tonic-gate * to make sure any executing TLBs see cleared bits. 33917c478bd9Sstevel@tonic-gate */ 33927c478bd9Sstevel@tonic-gate new = old; 33937c478bd9Sstevel@tonic-gate PTE_CLR(new, PT_REF | PT_MOD); 33947c478bd9Sstevel@tonic-gate old = hati_update_pte(ht, entry, old, new); 33957c478bd9Sstevel@tonic-gate if (old != 0) 33967c478bd9Sstevel@tonic-gate goto try_again; 33977c478bd9Sstevel@tonic-gate 33987c478bd9Sstevel@tonic-gate old = save_old; 33997c478bd9Sstevel@tonic-gate } 34007c478bd9Sstevel@tonic-gate 34017c478bd9Sstevel@tonic-gate /* 34027c478bd9Sstevel@tonic-gate * Sync the PTE 34037c478bd9Sstevel@tonic-gate */ 3404ae115bc7Smrj if (!(flags & HAT_SYNC_ZERORM) && 3405ae115bc7Smrj PTE_GET(old, PT_SOFTWARE) <= PT_NOSYNC) 34067c478bd9Sstevel@tonic-gate hati_sync_pte_to_page(pp, old, ht->ht_level); 34077c478bd9Sstevel@tonic-gate 34087c478bd9Sstevel@tonic-gate /* 34097c478bd9Sstevel@tonic-gate * can stop short if we found a ref'd or mod'd page 34107c478bd9Sstevel@tonic-gate */ 34117c478bd9Sstevel@tonic-gate if ((flags & HAT_SYNC_STOPON_MOD) && PP_ISMOD(save_pp) || 34127c478bd9Sstevel@tonic-gate (flags & HAT_SYNC_STOPON_REF) && PP_ISREF(save_pp)) { 34137c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 3414ae115bc7Smrj goto done; 34157c478bd9Sstevel@tonic-gate } 34167c478bd9Sstevel@tonic-gate } 34177c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 34187c478bd9Sstevel@tonic-gate while (pszc < pp->p_szc) { 34197c478bd9Sstevel@tonic-gate page_t *tpp; 34207c478bd9Sstevel@tonic-gate pszc++; 34217c478bd9Sstevel@tonic-gate tpp = PP_GROUPLEADER(pp, pszc); 34227c478bd9Sstevel@tonic-gate if (pp != tpp) { 34237c478bd9Sstevel@tonic-gate pp = tpp; 34247c478bd9Sstevel@tonic-gate goto next_size; 34257c478bd9Sstevel@tonic-gate } 34267c478bd9Sstevel@tonic-gate } 3427ae115bc7Smrj done: 34287c478bd9Sstevel@tonic-gate return (save_pp->p_nrm & nrmbits); 34297c478bd9Sstevel@tonic-gate } 34307c478bd9Sstevel@tonic-gate 34317c478bd9Sstevel@tonic-gate /* 34327c478bd9Sstevel@tonic-gate * returns approx number of mappings to this pp. A return of 0 implies 34337c478bd9Sstevel@tonic-gate * there are no mappings to the page. 34347c478bd9Sstevel@tonic-gate */ 34357c478bd9Sstevel@tonic-gate ulong_t 34367c478bd9Sstevel@tonic-gate hat_page_getshare(page_t *pp) 34377c478bd9Sstevel@tonic-gate { 34387c478bd9Sstevel@tonic-gate uint_t cnt; 34397c478bd9Sstevel@tonic-gate cnt = hment_mapcnt(pp); 3440a5652762Spraks #if defined(__amd64) 3441a5652762Spraks if (vpm_enable && pp->p_vpmref) { 3442a5652762Spraks cnt += 1; 3443a5652762Spraks } 3444a5652762Spraks #endif 34457c478bd9Sstevel@tonic-gate return (cnt); 34467c478bd9Sstevel@tonic-gate } 34477c478bd9Sstevel@tonic-gate 34487c478bd9Sstevel@tonic-gate /* 34497c478bd9Sstevel@tonic-gate * hat_softlock isn't supported anymore 34507c478bd9Sstevel@tonic-gate */ 34517c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 34527c478bd9Sstevel@tonic-gate faultcode_t 34537c478bd9Sstevel@tonic-gate hat_softlock( 34547c478bd9Sstevel@tonic-gate hat_t *hat, 34557c478bd9Sstevel@tonic-gate caddr_t addr, 34567c478bd9Sstevel@tonic-gate size_t *len, 34577c478bd9Sstevel@tonic-gate struct page **page_array, 34587c478bd9Sstevel@tonic-gate uint_t flags) 34597c478bd9Sstevel@tonic-gate { 34607c478bd9Sstevel@tonic-gate return (FC_NOSUPPORT); 34617c478bd9Sstevel@tonic-gate } 34627c478bd9Sstevel@tonic-gate 34637c478bd9Sstevel@tonic-gate 34647c478bd9Sstevel@tonic-gate 34657c478bd9Sstevel@tonic-gate /* 34667c478bd9Sstevel@tonic-gate * Routine to expose supported HAT features to platform independent code. 34677c478bd9Sstevel@tonic-gate */ 34687c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 34697c478bd9Sstevel@tonic-gate int 34707c478bd9Sstevel@tonic-gate hat_supported(enum hat_features feature, void *arg) 34717c478bd9Sstevel@tonic-gate { 34727c478bd9Sstevel@tonic-gate switch (feature) { 34737c478bd9Sstevel@tonic-gate 34747c478bd9Sstevel@tonic-gate case HAT_SHARED_PT: /* this is really ISM */ 34757c478bd9Sstevel@tonic-gate return (1); 34767c478bd9Sstevel@tonic-gate 34777c478bd9Sstevel@tonic-gate case HAT_DYNAMIC_ISM_UNMAP: 34787c478bd9Sstevel@tonic-gate return (0); 34797c478bd9Sstevel@tonic-gate 34807c478bd9Sstevel@tonic-gate case HAT_VMODSORT: 34817c478bd9Sstevel@tonic-gate return (1); 34827c478bd9Sstevel@tonic-gate 34837c478bd9Sstevel@tonic-gate default: 34847c478bd9Sstevel@tonic-gate panic("hat_supported() - unknown feature"); 34857c478bd9Sstevel@tonic-gate } 34867c478bd9Sstevel@tonic-gate return (0); 34877c478bd9Sstevel@tonic-gate } 34887c478bd9Sstevel@tonic-gate 34897c478bd9Sstevel@tonic-gate /* 34907c478bd9Sstevel@tonic-gate * Called when a thread is exiting and has been switched to the kernel AS 34917c478bd9Sstevel@tonic-gate */ 34927c478bd9Sstevel@tonic-gate void 34937c478bd9Sstevel@tonic-gate hat_thread_exit(kthread_t *thd) 34947c478bd9Sstevel@tonic-gate { 34957c478bd9Sstevel@tonic-gate ASSERT(thd->t_procp->p_as == &kas); 34967c478bd9Sstevel@tonic-gate hat_switch(thd->t_procp->p_as->a_hat); 34977c478bd9Sstevel@tonic-gate } 34987c478bd9Sstevel@tonic-gate 34997c478bd9Sstevel@tonic-gate /* 35007c478bd9Sstevel@tonic-gate * Setup the given brand new hat structure as the new HAT on this cpu's mmu. 35017c478bd9Sstevel@tonic-gate */ 35027c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 35037c478bd9Sstevel@tonic-gate void 35047c478bd9Sstevel@tonic-gate hat_setup(hat_t *hat, int flags) 35057c478bd9Sstevel@tonic-gate { 35067c478bd9Sstevel@tonic-gate kpreempt_disable(); 35077c478bd9Sstevel@tonic-gate 35087c478bd9Sstevel@tonic-gate hat_switch(hat); 35097c478bd9Sstevel@tonic-gate 35107c478bd9Sstevel@tonic-gate kpreempt_enable(); 35117c478bd9Sstevel@tonic-gate } 35127c478bd9Sstevel@tonic-gate 35137c478bd9Sstevel@tonic-gate /* 35147c478bd9Sstevel@tonic-gate * Prepare for a CPU private mapping for the given address. 35157c478bd9Sstevel@tonic-gate * 35167c478bd9Sstevel@tonic-gate * The address can only be used from a single CPU and can be remapped 35177c478bd9Sstevel@tonic-gate * using hat_mempte_remap(). Return the address of the PTE. 35187c478bd9Sstevel@tonic-gate * 35197c478bd9Sstevel@tonic-gate * We do the htable_create() if necessary and increment the valid count so 35207c478bd9Sstevel@tonic-gate * the htable can't disappear. We also hat_devload() the page table into 35217c478bd9Sstevel@tonic-gate * kernel so that the PTE is quickly accessed. 35227c478bd9Sstevel@tonic-gate */ 3523ae115bc7Smrj hat_mempte_t 3524ae115bc7Smrj hat_mempte_setup(caddr_t addr) 35257c478bd9Sstevel@tonic-gate { 35267c478bd9Sstevel@tonic-gate uintptr_t va = (uintptr_t)addr; 35277c478bd9Sstevel@tonic-gate htable_t *ht; 35287c478bd9Sstevel@tonic-gate uint_t entry; 35297c478bd9Sstevel@tonic-gate x86pte_t oldpte; 3530ae115bc7Smrj hat_mempte_t p; 35317c478bd9Sstevel@tonic-gate 35327c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(va)); 35337c478bd9Sstevel@tonic-gate ASSERT(!IN_VA_HOLE(va)); 3534aac11643Sjosephb ++curthread->t_hatdepth; 35357c478bd9Sstevel@tonic-gate ht = htable_getpte(kas.a_hat, va, &entry, &oldpte, 0); 35367c478bd9Sstevel@tonic-gate if (ht == NULL) { 35377c478bd9Sstevel@tonic-gate ht = htable_create(kas.a_hat, va, 0, NULL); 35387c478bd9Sstevel@tonic-gate entry = htable_va2entry(va, ht); 35397c478bd9Sstevel@tonic-gate ASSERT(ht->ht_level == 0); 35407c478bd9Sstevel@tonic-gate oldpte = x86pte_get(ht, entry); 35417c478bd9Sstevel@tonic-gate } 35427c478bd9Sstevel@tonic-gate if (PTE_ISVALID(oldpte)) 35437c478bd9Sstevel@tonic-gate panic("hat_mempte_setup(): address already mapped" 35447c478bd9Sstevel@tonic-gate "ht=%p, entry=%d, pte=" FMT_PTE, ht, entry, oldpte); 35457c478bd9Sstevel@tonic-gate 35467c478bd9Sstevel@tonic-gate /* 35477c478bd9Sstevel@tonic-gate * increment ht_valid_cnt so that the pagetable can't disappear 35487c478bd9Sstevel@tonic-gate */ 35497c478bd9Sstevel@tonic-gate HTABLE_INC(ht->ht_valid_cnt); 35507c478bd9Sstevel@tonic-gate 35517c478bd9Sstevel@tonic-gate /* 3552ae115bc7Smrj * return the PTE physical address to the caller. 35537c478bd9Sstevel@tonic-gate */ 35547c478bd9Sstevel@tonic-gate htable_release(ht); 3555ae115bc7Smrj p = PT_INDEX_PHYSADDR(pfn_to_pa(ht->ht_pfn), entry); 3556aac11643Sjosephb --curthread->t_hatdepth; 3557ae115bc7Smrj return (p); 35587c478bd9Sstevel@tonic-gate } 35597c478bd9Sstevel@tonic-gate 35607c478bd9Sstevel@tonic-gate /* 35617c478bd9Sstevel@tonic-gate * Release a CPU private mapping for the given address. 35627c478bd9Sstevel@tonic-gate * We decrement the htable valid count so it might be destroyed. 35637c478bd9Sstevel@tonic-gate */ 3564ae115bc7Smrj /*ARGSUSED1*/ 35657c478bd9Sstevel@tonic-gate void 3566ae115bc7Smrj hat_mempte_release(caddr_t addr, hat_mempte_t pte_pa) 35677c478bd9Sstevel@tonic-gate { 35687c478bd9Sstevel@tonic-gate htable_t *ht; 35697c478bd9Sstevel@tonic-gate 35707c478bd9Sstevel@tonic-gate /* 3571ae115bc7Smrj * invalidate any left over mapping and decrement the htable valid count 35727c478bd9Sstevel@tonic-gate */ 3573ae115bc7Smrj { 3574ae115bc7Smrj x86pte_t *pteptr; 3575ae115bc7Smrj 3576ae115bc7Smrj pteptr = x86pte_mapin(mmu_btop(pte_pa), 3577ae115bc7Smrj (pte_pa & MMU_PAGEOFFSET) >> mmu.pte_size_shift, NULL); 3578ae115bc7Smrj if (mmu.pae_hat) 3579ae115bc7Smrj *pteptr = 0; 3580ae115bc7Smrj else 3581ae115bc7Smrj *(x86pte32_t *)pteptr = 0; 3582ae115bc7Smrj mmu_tlbflush_entry(addr); 3583ae115bc7Smrj x86pte_mapout(); 3584ae115bc7Smrj } 3585ae115bc7Smrj 35867c478bd9Sstevel@tonic-gate ht = htable_getpte(kas.a_hat, ALIGN2PAGE(addr), NULL, NULL, 0); 35877c478bd9Sstevel@tonic-gate if (ht == NULL) 35887c478bd9Sstevel@tonic-gate panic("hat_mempte_release(): invalid address"); 35897c478bd9Sstevel@tonic-gate ASSERT(ht->ht_level == 0); 35907c478bd9Sstevel@tonic-gate HTABLE_DEC(ht->ht_valid_cnt); 35917c478bd9Sstevel@tonic-gate htable_release(ht); 35927c478bd9Sstevel@tonic-gate } 35937c478bd9Sstevel@tonic-gate 35947c478bd9Sstevel@tonic-gate /* 35957c478bd9Sstevel@tonic-gate * Apply a temporary CPU private mapping to a page. We flush the TLB only 35967c478bd9Sstevel@tonic-gate * on this CPU, so this ought to have been called with preemption disabled. 35977c478bd9Sstevel@tonic-gate */ 35987c478bd9Sstevel@tonic-gate void 35997c478bd9Sstevel@tonic-gate hat_mempte_remap( 3600ae115bc7Smrj pfn_t pfn, 3601ae115bc7Smrj caddr_t addr, 3602ae115bc7Smrj hat_mempte_t pte_pa, 3603ae115bc7Smrj uint_t attr, 3604ae115bc7Smrj uint_t flags) 36057c478bd9Sstevel@tonic-gate { 36067c478bd9Sstevel@tonic-gate uintptr_t va = (uintptr_t)addr; 36077c478bd9Sstevel@tonic-gate x86pte_t pte; 36087c478bd9Sstevel@tonic-gate 36097c478bd9Sstevel@tonic-gate /* 36107c478bd9Sstevel@tonic-gate * Remap the given PTE to the new page's PFN. Invalidate only 36117c478bd9Sstevel@tonic-gate * on this CPU. 36127c478bd9Sstevel@tonic-gate */ 36137c478bd9Sstevel@tonic-gate #ifdef DEBUG 36147c478bd9Sstevel@tonic-gate htable_t *ht; 36157c478bd9Sstevel@tonic-gate uint_t entry; 36167c478bd9Sstevel@tonic-gate 36177c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(va)); 36187c478bd9Sstevel@tonic-gate ASSERT(!IN_VA_HOLE(va)); 36197c478bd9Sstevel@tonic-gate ht = htable_getpte(kas.a_hat, va, &entry, NULL, 0); 36207c478bd9Sstevel@tonic-gate ASSERT(ht != NULL); 36217c478bd9Sstevel@tonic-gate ASSERT(ht->ht_level == 0); 36227c478bd9Sstevel@tonic-gate ASSERT(ht->ht_valid_cnt > 0); 3623ae115bc7Smrj ASSERT(ht->ht_pfn == mmu_btop(pte_pa)); 36247c478bd9Sstevel@tonic-gate htable_release(ht); 36257c478bd9Sstevel@tonic-gate #endif 36267c478bd9Sstevel@tonic-gate pte = hati_mkpte(pfn, attr, 0, flags); 3627ae115bc7Smrj { 3628ae115bc7Smrj x86pte_t *pteptr; 3629ae115bc7Smrj 3630ae115bc7Smrj pteptr = x86pte_mapin(mmu_btop(pte_pa), 3631ae115bc7Smrj (pte_pa & MMU_PAGEOFFSET) >> mmu.pte_size_shift, NULL); 3632ae115bc7Smrj if (mmu.pae_hat) 3633ae115bc7Smrj *(x86pte_t *)pteptr = pte; 3634ae115bc7Smrj else 3635ae115bc7Smrj *(x86pte32_t *)pteptr = (x86pte32_t)pte; 3636ae115bc7Smrj mmu_tlbflush_entry(addr); 3637ae115bc7Smrj x86pte_mapout(); 3638ae115bc7Smrj } 36397c478bd9Sstevel@tonic-gate } 36407c478bd9Sstevel@tonic-gate 36417c478bd9Sstevel@tonic-gate 36427c478bd9Sstevel@tonic-gate 36437c478bd9Sstevel@tonic-gate /* 36447c478bd9Sstevel@tonic-gate * Hat locking functions 36457c478bd9Sstevel@tonic-gate * XXX - these two functions are currently being used by hatstats 36467c478bd9Sstevel@tonic-gate * they can be removed by using a per-as mutex for hatstats. 36477c478bd9Sstevel@tonic-gate */ 36487c478bd9Sstevel@tonic-gate void 36497c478bd9Sstevel@tonic-gate hat_enter(hat_t *hat) 36507c478bd9Sstevel@tonic-gate { 36517c478bd9Sstevel@tonic-gate mutex_enter(&hat->hat_mutex); 36527c478bd9Sstevel@tonic-gate } 36537c478bd9Sstevel@tonic-gate 36547c478bd9Sstevel@tonic-gate void 36557c478bd9Sstevel@tonic-gate hat_exit(hat_t *hat) 36567c478bd9Sstevel@tonic-gate { 36577c478bd9Sstevel@tonic-gate mutex_exit(&hat->hat_mutex); 36587c478bd9Sstevel@tonic-gate } 36597c478bd9Sstevel@tonic-gate 36607c478bd9Sstevel@tonic-gate /* 3661ae115bc7Smrj * HAT part of cpu initialization. 36627c478bd9Sstevel@tonic-gate */ 36637c478bd9Sstevel@tonic-gate void 3664ae115bc7Smrj hat_cpu_online(struct cpu *cpup) 36657c478bd9Sstevel@tonic-gate { 3666ae115bc7Smrj if (cpup != CPU) { 3667ae115bc7Smrj x86pte_cpu_init(cpup); 3668ae115bc7Smrj hat_vlp_setup(cpup); 36697c478bd9Sstevel@tonic-gate } 3670ae115bc7Smrj CPUSET_ATOMIC_ADD(khat_cpuset, cpup->cpu_id); 36717c478bd9Sstevel@tonic-gate } 36727c478bd9Sstevel@tonic-gate 36737c478bd9Sstevel@tonic-gate /* 3674ae115bc7Smrj * HAT part of cpu deletion. 3675ae115bc7Smrj * (currently, we only call this after the cpu is safely passivated.) 36767c478bd9Sstevel@tonic-gate */ 36777c478bd9Sstevel@tonic-gate void 3678ae115bc7Smrj hat_cpu_offline(struct cpu *cpup) 36797c478bd9Sstevel@tonic-gate { 3680ae115bc7Smrj ASSERT(cpup != CPU); 3681ae115bc7Smrj 3682ae115bc7Smrj CPUSET_ATOMIC_DEL(khat_cpuset, cpup->cpu_id); 3683ae115bc7Smrj x86pte_cpu_fini(cpup); 3684ae115bc7Smrj hat_vlp_teardown(cpup); 36857c478bd9Sstevel@tonic-gate } 36867c478bd9Sstevel@tonic-gate 36877c478bd9Sstevel@tonic-gate /* 36887c478bd9Sstevel@tonic-gate * Function called after all CPUs are brought online. 36897c478bd9Sstevel@tonic-gate * Used to remove low address boot mappings. 36907c478bd9Sstevel@tonic-gate */ 36917c478bd9Sstevel@tonic-gate void 36927c478bd9Sstevel@tonic-gate clear_boot_mappings(uintptr_t low, uintptr_t high) 36937c478bd9Sstevel@tonic-gate { 36947c478bd9Sstevel@tonic-gate uintptr_t vaddr = low; 36957c478bd9Sstevel@tonic-gate htable_t *ht = NULL; 36967c478bd9Sstevel@tonic-gate level_t level; 36977c478bd9Sstevel@tonic-gate uint_t entry; 36987c478bd9Sstevel@tonic-gate x86pte_t pte; 36997c478bd9Sstevel@tonic-gate 37007c478bd9Sstevel@tonic-gate /* 37017c478bd9Sstevel@tonic-gate * On 1st CPU we can unload the prom mappings, basically we blow away 3702ae115bc7Smrj * all virtual mappings under _userlimit. 37037c478bd9Sstevel@tonic-gate */ 37047c478bd9Sstevel@tonic-gate while (vaddr < high) { 37057c478bd9Sstevel@tonic-gate pte = htable_walk(kas.a_hat, &ht, &vaddr, high); 37067c478bd9Sstevel@tonic-gate if (ht == NULL) 37077c478bd9Sstevel@tonic-gate break; 37087c478bd9Sstevel@tonic-gate 37097c478bd9Sstevel@tonic-gate level = ht->ht_level; 37107c478bd9Sstevel@tonic-gate entry = htable_va2entry(vaddr, ht); 37117c478bd9Sstevel@tonic-gate ASSERT(level <= mmu.max_page_level); 37127c478bd9Sstevel@tonic-gate ASSERT(PTE_ISPAGE(pte, level)); 37137c478bd9Sstevel@tonic-gate 37147c478bd9Sstevel@tonic-gate /* 37157c478bd9Sstevel@tonic-gate * Unload the mapping from the page tables. 37167c478bd9Sstevel@tonic-gate */ 3717ae115bc7Smrj (void) x86pte_inval(ht, entry, 0, NULL); 37187c478bd9Sstevel@tonic-gate ASSERT(ht->ht_valid_cnt > 0); 37197c478bd9Sstevel@tonic-gate HTABLE_DEC(ht->ht_valid_cnt); 37207c478bd9Sstevel@tonic-gate PGCNT_DEC(ht->ht_hat, ht->ht_level); 37217c478bd9Sstevel@tonic-gate 37227c478bd9Sstevel@tonic-gate vaddr += LEVEL_SIZE(ht->ht_level); 37237c478bd9Sstevel@tonic-gate } 37247c478bd9Sstevel@tonic-gate if (ht) 37257c478bd9Sstevel@tonic-gate htable_release(ht); 37267c478bd9Sstevel@tonic-gate } 37277c478bd9Sstevel@tonic-gate 37287c478bd9Sstevel@tonic-gate /* 37297c478bd9Sstevel@tonic-gate * Atomically update a new translation for a single page. If the 37307c478bd9Sstevel@tonic-gate * currently installed PTE doesn't match the value we expect to find, 37317c478bd9Sstevel@tonic-gate * it's not updated and we return the PTE we found. 37327c478bd9Sstevel@tonic-gate * 37337c478bd9Sstevel@tonic-gate * If activating nosync or NOWRITE and the page was modified we need to sync 37347c478bd9Sstevel@tonic-gate * with the page_t. Also sync with page_t if clearing ref/mod bits. 37357c478bd9Sstevel@tonic-gate */ 37367c478bd9Sstevel@tonic-gate static x86pte_t 37377c478bd9Sstevel@tonic-gate hati_update_pte(htable_t *ht, uint_t entry, x86pte_t expected, x86pte_t new) 37387c478bd9Sstevel@tonic-gate { 37397c478bd9Sstevel@tonic-gate page_t *pp; 37407c478bd9Sstevel@tonic-gate uint_t rm = 0; 37417c478bd9Sstevel@tonic-gate x86pte_t replaced; 37427c478bd9Sstevel@tonic-gate 3743ae115bc7Smrj if (PTE_GET(expected, PT_SOFTWARE) < PT_NOSYNC && 37447c478bd9Sstevel@tonic-gate PTE_GET(expected, PT_MOD | PT_REF) && 37457c478bd9Sstevel@tonic-gate (PTE_GET(new, PT_NOSYNC) || !PTE_GET(new, PT_WRITABLE) || 37467c478bd9Sstevel@tonic-gate !PTE_GET(new, PT_MOD | PT_REF))) { 37477c478bd9Sstevel@tonic-gate 3748ae115bc7Smrj ASSERT(!pfn_is_foreign(PTE2PFN(expected, ht->ht_level))); 37497c478bd9Sstevel@tonic-gate pp = page_numtopp_nolock(PTE2PFN(expected, ht->ht_level)); 37507c478bd9Sstevel@tonic-gate ASSERT(pp != NULL); 37517c478bd9Sstevel@tonic-gate if (PTE_GET(expected, PT_MOD)) 37527c478bd9Sstevel@tonic-gate rm |= P_MOD; 37537c478bd9Sstevel@tonic-gate if (PTE_GET(expected, PT_REF)) 37547c478bd9Sstevel@tonic-gate rm |= P_REF; 37557c478bd9Sstevel@tonic-gate PTE_CLR(new, PT_MOD | PT_REF); 37567c478bd9Sstevel@tonic-gate } 37577c478bd9Sstevel@tonic-gate 37587c478bd9Sstevel@tonic-gate replaced = x86pte_update(ht, entry, expected, new); 37597c478bd9Sstevel@tonic-gate if (replaced != expected) 37607c478bd9Sstevel@tonic-gate return (replaced); 37617c478bd9Sstevel@tonic-gate 37627c478bd9Sstevel@tonic-gate if (rm) { 37637c478bd9Sstevel@tonic-gate /* 37647c478bd9Sstevel@tonic-gate * sync to all constituent pages of a large page 37657c478bd9Sstevel@tonic-gate */ 37667c478bd9Sstevel@tonic-gate pgcnt_t pgcnt = page_get_pagecnt(ht->ht_level); 37677c478bd9Sstevel@tonic-gate ASSERT(IS_P2ALIGNED(pp->p_pagenum, pgcnt)); 37687c478bd9Sstevel@tonic-gate while (pgcnt-- > 0) { 37697c478bd9Sstevel@tonic-gate /* 37707c478bd9Sstevel@tonic-gate * hat_page_demote() can't decrease 37717c478bd9Sstevel@tonic-gate * pszc below this mapping size 37727c478bd9Sstevel@tonic-gate * since large mapping existed after we 37737c478bd9Sstevel@tonic-gate * took mlist lock. 37747c478bd9Sstevel@tonic-gate */ 37757c478bd9Sstevel@tonic-gate ASSERT(pp->p_szc >= ht->ht_level); 37767c478bd9Sstevel@tonic-gate hat_page_setattr(pp, rm); 37777c478bd9Sstevel@tonic-gate ++pp; 37787c478bd9Sstevel@tonic-gate } 37797c478bd9Sstevel@tonic-gate } 37807c478bd9Sstevel@tonic-gate 37817c478bd9Sstevel@tonic-gate return (0); 37827c478bd9Sstevel@tonic-gate } 37837c478bd9Sstevel@tonic-gate 37847c478bd9Sstevel@tonic-gate /* 37857c478bd9Sstevel@tonic-gate * Kernel Physical Mapping (kpm) facility 37867c478bd9Sstevel@tonic-gate * 37877c478bd9Sstevel@tonic-gate * Most of the routines needed to support segkpm are almost no-ops on the 37887c478bd9Sstevel@tonic-gate * x86 platform. We map in the entire segment when it is created and leave 37897c478bd9Sstevel@tonic-gate * it mapped in, so there is no additional work required to set up and tear 37907c478bd9Sstevel@tonic-gate * down individual mappings. All of these routines were created to support 37917c478bd9Sstevel@tonic-gate * SPARC platforms that have to avoid aliasing in their virtually indexed 37927c478bd9Sstevel@tonic-gate * caches. 37937c478bd9Sstevel@tonic-gate * 37947c478bd9Sstevel@tonic-gate * Most of the routines have sanity checks in them (e.g. verifying that the 37957c478bd9Sstevel@tonic-gate * passed-in page is locked). We don't actually care about most of these 37967c478bd9Sstevel@tonic-gate * checks on x86, but we leave them in place to identify problems in the 37977c478bd9Sstevel@tonic-gate * upper levels. 37987c478bd9Sstevel@tonic-gate */ 37997c478bd9Sstevel@tonic-gate 38007c478bd9Sstevel@tonic-gate /* 38017c478bd9Sstevel@tonic-gate * Map in a locked page and return the vaddr. 38027c478bd9Sstevel@tonic-gate */ 38037c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 38047c478bd9Sstevel@tonic-gate caddr_t 38057c478bd9Sstevel@tonic-gate hat_kpm_mapin(struct page *pp, struct kpme *kpme) 38067c478bd9Sstevel@tonic-gate { 38077c478bd9Sstevel@tonic-gate caddr_t vaddr; 38087c478bd9Sstevel@tonic-gate 38097c478bd9Sstevel@tonic-gate #ifdef DEBUG 38107c478bd9Sstevel@tonic-gate if (kpm_enable == 0) { 38117c478bd9Sstevel@tonic-gate cmn_err(CE_WARN, "hat_kpm_mapin: kpm_enable not set\n"); 38127c478bd9Sstevel@tonic-gate return ((caddr_t)NULL); 38137c478bd9Sstevel@tonic-gate } 38147c478bd9Sstevel@tonic-gate 38157c478bd9Sstevel@tonic-gate if (pp == NULL || PAGE_LOCKED(pp) == 0) { 38167c478bd9Sstevel@tonic-gate cmn_err(CE_WARN, "hat_kpm_mapin: pp zero or not locked\n"); 38177c478bd9Sstevel@tonic-gate return ((caddr_t)NULL); 38187c478bd9Sstevel@tonic-gate } 38197c478bd9Sstevel@tonic-gate #endif 38207c478bd9Sstevel@tonic-gate 38217c478bd9Sstevel@tonic-gate vaddr = hat_kpm_page2va(pp, 1); 38227c478bd9Sstevel@tonic-gate 38237c478bd9Sstevel@tonic-gate return (vaddr); 38247c478bd9Sstevel@tonic-gate } 38257c478bd9Sstevel@tonic-gate 38267c478bd9Sstevel@tonic-gate /* 38277c478bd9Sstevel@tonic-gate * Mapout a locked page. 38287c478bd9Sstevel@tonic-gate */ 38297c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 38307c478bd9Sstevel@tonic-gate void 38317c478bd9Sstevel@tonic-gate hat_kpm_mapout(struct page *pp, struct kpme *kpme, caddr_t vaddr) 38327c478bd9Sstevel@tonic-gate { 38337c478bd9Sstevel@tonic-gate #ifdef DEBUG 38347c478bd9Sstevel@tonic-gate if (kpm_enable == 0) { 38357c478bd9Sstevel@tonic-gate cmn_err(CE_WARN, "hat_kpm_mapout: kpm_enable not set\n"); 38367c478bd9Sstevel@tonic-gate return; 38377c478bd9Sstevel@tonic-gate } 38387c478bd9Sstevel@tonic-gate 38397c478bd9Sstevel@tonic-gate if (IS_KPM_ADDR(vaddr) == 0) { 38407c478bd9Sstevel@tonic-gate cmn_err(CE_WARN, "hat_kpm_mapout: no kpm address\n"); 38417c478bd9Sstevel@tonic-gate return; 38427c478bd9Sstevel@tonic-gate } 38437c478bd9Sstevel@tonic-gate 38447c478bd9Sstevel@tonic-gate if (pp == NULL || PAGE_LOCKED(pp) == 0) { 38457c478bd9Sstevel@tonic-gate cmn_err(CE_WARN, "hat_kpm_mapout: page zero or not locked\n"); 38467c478bd9Sstevel@tonic-gate return; 38477c478bd9Sstevel@tonic-gate } 38487c478bd9Sstevel@tonic-gate #endif 38497c478bd9Sstevel@tonic-gate } 38507c478bd9Sstevel@tonic-gate 38517c478bd9Sstevel@tonic-gate /* 38527c478bd9Sstevel@tonic-gate * Return the kpm virtual address for a specific pfn 38537c478bd9Sstevel@tonic-gate */ 38547c478bd9Sstevel@tonic-gate caddr_t 38557c478bd9Sstevel@tonic-gate hat_kpm_pfn2va(pfn_t pfn) 38567c478bd9Sstevel@tonic-gate { 3857ae115bc7Smrj uintptr_t vaddr = (uintptr_t)kpm_vbase + mmu_ptob(pfn); 38587c478bd9Sstevel@tonic-gate 38597c478bd9Sstevel@tonic-gate return ((caddr_t)vaddr); 38607c478bd9Sstevel@tonic-gate } 38617c478bd9Sstevel@tonic-gate 38627c478bd9Sstevel@tonic-gate /* 38637c478bd9Sstevel@tonic-gate * Return the kpm virtual address for the page at pp. 38647c478bd9Sstevel@tonic-gate */ 38657c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 38667c478bd9Sstevel@tonic-gate caddr_t 38677c478bd9Sstevel@tonic-gate hat_kpm_page2va(struct page *pp, int checkswap) 38687c478bd9Sstevel@tonic-gate { 38697c478bd9Sstevel@tonic-gate return (hat_kpm_pfn2va(pp->p_pagenum)); 38707c478bd9Sstevel@tonic-gate } 38717c478bd9Sstevel@tonic-gate 38727c478bd9Sstevel@tonic-gate /* 38737c478bd9Sstevel@tonic-gate * Return the page frame number for the kpm virtual address vaddr. 38747c478bd9Sstevel@tonic-gate */ 38757c478bd9Sstevel@tonic-gate pfn_t 38767c478bd9Sstevel@tonic-gate hat_kpm_va2pfn(caddr_t vaddr) 38777c478bd9Sstevel@tonic-gate { 38787c478bd9Sstevel@tonic-gate pfn_t pfn; 38797c478bd9Sstevel@tonic-gate 38807c478bd9Sstevel@tonic-gate ASSERT(IS_KPM_ADDR(vaddr)); 38817c478bd9Sstevel@tonic-gate 38827c478bd9Sstevel@tonic-gate pfn = (pfn_t)btop(vaddr - kpm_vbase); 38837c478bd9Sstevel@tonic-gate 38847c478bd9Sstevel@tonic-gate return (pfn); 38857c478bd9Sstevel@tonic-gate } 38867c478bd9Sstevel@tonic-gate 38877c478bd9Sstevel@tonic-gate 38887c478bd9Sstevel@tonic-gate /* 38897c478bd9Sstevel@tonic-gate * Return the page for the kpm virtual address vaddr. 38907c478bd9Sstevel@tonic-gate */ 38917c478bd9Sstevel@tonic-gate page_t * 38927c478bd9Sstevel@tonic-gate hat_kpm_vaddr2page(caddr_t vaddr) 38937c478bd9Sstevel@tonic-gate { 38947c478bd9Sstevel@tonic-gate pfn_t pfn; 38957c478bd9Sstevel@tonic-gate 38967c478bd9Sstevel@tonic-gate ASSERT(IS_KPM_ADDR(vaddr)); 38977c478bd9Sstevel@tonic-gate 38987c478bd9Sstevel@tonic-gate pfn = hat_kpm_va2pfn(vaddr); 38997c478bd9Sstevel@tonic-gate 39007c478bd9Sstevel@tonic-gate return (page_numtopp_nolock(pfn)); 39017c478bd9Sstevel@tonic-gate } 39027c478bd9Sstevel@tonic-gate 39037c478bd9Sstevel@tonic-gate /* 39047c478bd9Sstevel@tonic-gate * hat_kpm_fault is called from segkpm_fault when we take a page fault on a 39057c478bd9Sstevel@tonic-gate * KPM page. This should never happen on x86 39067c478bd9Sstevel@tonic-gate */ 39077c478bd9Sstevel@tonic-gate int 39087c478bd9Sstevel@tonic-gate hat_kpm_fault(hat_t *hat, caddr_t vaddr) 39097c478bd9Sstevel@tonic-gate { 39107c478bd9Sstevel@tonic-gate panic("pagefault in seg_kpm. hat: 0x%p vaddr: 0x%p", hat, vaddr); 39117c478bd9Sstevel@tonic-gate 39127c478bd9Sstevel@tonic-gate return (0); 39137c478bd9Sstevel@tonic-gate } 39147c478bd9Sstevel@tonic-gate 39157c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 39167c478bd9Sstevel@tonic-gate void 39177c478bd9Sstevel@tonic-gate hat_kpm_mseghash_clear(int nentries) 39187c478bd9Sstevel@tonic-gate {} 39197c478bd9Sstevel@tonic-gate 39207c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 39217c478bd9Sstevel@tonic-gate void 39227c478bd9Sstevel@tonic-gate hat_kpm_mseghash_update(pgcnt_t inx, struct memseg *msp) 39237c478bd9Sstevel@tonic-gate {} 3924