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> 6595c0a3c8Sjosephb #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> 76*250b7ff9Sjosephb #include <vm/seg_spt.h> 777c478bd9Sstevel@tonic-gate 787c478bd9Sstevel@tonic-gate #include <sys/cmn_err.h> 797c478bd9Sstevel@tonic-gate 807c478bd9Sstevel@tonic-gate /* 817c478bd9Sstevel@tonic-gate * Basic parameters for hat operation. 827c478bd9Sstevel@tonic-gate */ 837c478bd9Sstevel@tonic-gate struct hat_mmu_info mmu; 847c478bd9Sstevel@tonic-gate 857c478bd9Sstevel@tonic-gate /* 867c478bd9Sstevel@tonic-gate * The page that is the kernel's top level pagetable. 877c478bd9Sstevel@tonic-gate * 887c478bd9Sstevel@tonic-gate * For 32 bit VLP support, the kernel hat will use the 1st 4 entries 897c478bd9Sstevel@tonic-gate * on this 4K page for its top level page table. The remaining groups of 907c478bd9Sstevel@tonic-gate * 4 entries are used for per processor copies of user VLP pagetables for 917c478bd9Sstevel@tonic-gate * running threads. See hat_switch() and reload_pae32() for details. 927c478bd9Sstevel@tonic-gate * 937c478bd9Sstevel@tonic-gate * vlp_page[0] - 0th level==2 PTE for kernel HAT (will be zero) 947c478bd9Sstevel@tonic-gate * vlp_page[1] - 1st level==2 PTE for kernel HAT (will be zero) 957c478bd9Sstevel@tonic-gate * vlp_page[2] - 2nd level==2 PTE for kernel HAT (zero for small memory) 967c478bd9Sstevel@tonic-gate * vlp_page[3] - 3rd level==2 PTE for kernel 977c478bd9Sstevel@tonic-gate * 987c478bd9Sstevel@tonic-gate * vlp_page[4] - 0th level==2 PTE for user thread on cpu 0 997c478bd9Sstevel@tonic-gate * vlp_page[5] - 1st level==2 PTE for user thread on cpu 0 1007c478bd9Sstevel@tonic-gate * vlp_page[6] - 2nd level==2 PTE for user thread on cpu 0 1017c478bd9Sstevel@tonic-gate * vlp_page[7] - probably copy of kernel PTE 1027c478bd9Sstevel@tonic-gate * 1037c478bd9Sstevel@tonic-gate * vlp_page[8] - 0th level==2 PTE for user thread on cpu 1 1047c478bd9Sstevel@tonic-gate * vlp_page[9] - 1st level==2 PTE for user thread on cpu 1 1057c478bd9Sstevel@tonic-gate * vlp_page[10] - 2nd level==2 PTE for user thread on cpu 1 1067c478bd9Sstevel@tonic-gate * vlp_page[11] - probably copy of kernel PTE 1077c478bd9Sstevel@tonic-gate * ... 1087c478bd9Sstevel@tonic-gate * 1097c478bd9Sstevel@tonic-gate * when / where the kernel PTE's are (entry 2 or 3 or none) depends 1107c478bd9Sstevel@tonic-gate * on kernelbase. 1117c478bd9Sstevel@tonic-gate */ 1127c478bd9Sstevel@tonic-gate static x86pte_t *vlp_page; 1137c478bd9Sstevel@tonic-gate 1147c478bd9Sstevel@tonic-gate /* 1157c478bd9Sstevel@tonic-gate * forward declaration of internal utility routines 1167c478bd9Sstevel@tonic-gate */ 1177c478bd9Sstevel@tonic-gate static x86pte_t hati_update_pte(htable_t *ht, uint_t entry, x86pte_t expected, 1187c478bd9Sstevel@tonic-gate x86pte_t new); 1197c478bd9Sstevel@tonic-gate 1207c478bd9Sstevel@tonic-gate /* 1217c478bd9Sstevel@tonic-gate * The kernel address space exists in all HATs. To implement this the 1227c478bd9Sstevel@tonic-gate * kernel reserves a fixed number of entries in every topmost level page 1237c478bd9Sstevel@tonic-gate * table. The values are setup in hat_init() and then copied to every hat 1247c478bd9Sstevel@tonic-gate * created by hat_alloc(). This means that kernelbase must be: 1257c478bd9Sstevel@tonic-gate * 1267c478bd9Sstevel@tonic-gate * 4Meg aligned for 32 bit kernels 1277c478bd9Sstevel@tonic-gate * 512Gig aligned for x86_64 64 bit kernel 1287c478bd9Sstevel@tonic-gate * 1297c478bd9Sstevel@tonic-gate * The PAE 32 bit hat is handled as a special case. Otherwise requiring 1Gig 1307c478bd9Sstevel@tonic-gate * alignment would use too much VA for the kernel. 1317c478bd9Sstevel@tonic-gate * 1327c478bd9Sstevel@tonic-gate */ 1337c478bd9Sstevel@tonic-gate static uint_t khat_start; /* index of 1st entry in kernel's top ptable */ 1347c478bd9Sstevel@tonic-gate static uint_t khat_entries; /* number of entries in kernel's top ptable */ 1357c478bd9Sstevel@tonic-gate 1367c478bd9Sstevel@tonic-gate #if defined(__i386) 1377c478bd9Sstevel@tonic-gate 1387c478bd9Sstevel@tonic-gate static htable_t *khat_pae32_htable = NULL; 1397c478bd9Sstevel@tonic-gate static uint_t khat_pae32_start; 1407c478bd9Sstevel@tonic-gate static uint_t khat_pae32_entries; 1417c478bd9Sstevel@tonic-gate 1427c478bd9Sstevel@tonic-gate #endif 1437c478bd9Sstevel@tonic-gate 1447c478bd9Sstevel@tonic-gate uint_t use_boot_reserve = 1; /* cleared after early boot process */ 1457c478bd9Sstevel@tonic-gate uint_t can_steal_post_boot = 0; /* set late in boot to enable stealing */ 1467c478bd9Sstevel@tonic-gate 1477c478bd9Sstevel@tonic-gate /* 1487c478bd9Sstevel@tonic-gate * A cpuset for all cpus. This is used for kernel address cross calls, since 1497c478bd9Sstevel@tonic-gate * the kernel addresses apply to all cpus. 1507c478bd9Sstevel@tonic-gate */ 1517c478bd9Sstevel@tonic-gate cpuset_t khat_cpuset; 1527c478bd9Sstevel@tonic-gate 1537c478bd9Sstevel@tonic-gate /* 1547c478bd9Sstevel@tonic-gate * management stuff for hat structures 1557c478bd9Sstevel@tonic-gate */ 1567c478bd9Sstevel@tonic-gate kmutex_t hat_list_lock; 1577c478bd9Sstevel@tonic-gate kcondvar_t hat_list_cv; 1587c478bd9Sstevel@tonic-gate kmem_cache_t *hat_cache; 1597c478bd9Sstevel@tonic-gate kmem_cache_t *hat_hash_cache; 1607c478bd9Sstevel@tonic-gate kmem_cache_t *vlp_hash_cache; 1617c478bd9Sstevel@tonic-gate 1627c478bd9Sstevel@tonic-gate /* 1637c478bd9Sstevel@tonic-gate * Simple statistics 1647c478bd9Sstevel@tonic-gate */ 1657c478bd9Sstevel@tonic-gate struct hatstats hatstat; 1667c478bd9Sstevel@tonic-gate 1677c478bd9Sstevel@tonic-gate /* 1687c478bd9Sstevel@tonic-gate * useful stuff for atomic access/clearing/setting REF/MOD/RO bits in page_t's. 1697c478bd9Sstevel@tonic-gate */ 1707c478bd9Sstevel@tonic-gate extern void atomic_orb(uchar_t *addr, uchar_t val); 1717c478bd9Sstevel@tonic-gate extern void atomic_andb(uchar_t *addr, uchar_t val); 1727c478bd9Sstevel@tonic-gate 1737c478bd9Sstevel@tonic-gate #define PP_GETRM(pp, rmmask) (pp->p_nrm & rmmask) 1747c478bd9Sstevel@tonic-gate #define PP_ISMOD(pp) PP_GETRM(pp, P_MOD) 1757c478bd9Sstevel@tonic-gate #define PP_ISREF(pp) PP_GETRM(pp, P_REF) 1767c478bd9Sstevel@tonic-gate #define PP_ISRO(pp) PP_GETRM(pp, P_RO) 1777c478bd9Sstevel@tonic-gate 1787c478bd9Sstevel@tonic-gate #define PP_SETRM(pp, rm) atomic_orb(&(pp->p_nrm), rm) 1797c478bd9Sstevel@tonic-gate #define PP_SETMOD(pp) PP_SETRM(pp, P_MOD) 1807c478bd9Sstevel@tonic-gate #define PP_SETREF(pp) PP_SETRM(pp, P_REF) 1817c478bd9Sstevel@tonic-gate #define PP_SETRO(pp) PP_SETRM(pp, P_RO) 1827c478bd9Sstevel@tonic-gate 1837c478bd9Sstevel@tonic-gate #define PP_CLRRM(pp, rm) atomic_andb(&(pp->p_nrm), ~(rm)) 1847c478bd9Sstevel@tonic-gate #define PP_CLRMOD(pp) PP_CLRRM(pp, P_MOD) 1857c478bd9Sstevel@tonic-gate #define PP_CLRREF(pp) PP_CLRRM(pp, P_REF) 1867c478bd9Sstevel@tonic-gate #define PP_CLRRO(pp) PP_CLRRM(pp, P_RO) 1877c478bd9Sstevel@tonic-gate #define PP_CLRALL(pp) PP_CLRRM(pp, P_MOD | P_REF | P_RO) 1887c478bd9Sstevel@tonic-gate 1897c478bd9Sstevel@tonic-gate /* 1907c478bd9Sstevel@tonic-gate * kmem cache constructor for struct hat 1917c478bd9Sstevel@tonic-gate */ 1927c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 1937c478bd9Sstevel@tonic-gate static int 1947c478bd9Sstevel@tonic-gate hati_constructor(void *buf, void *handle, int kmflags) 1957c478bd9Sstevel@tonic-gate { 1967c478bd9Sstevel@tonic-gate hat_t *hat = buf; 1977c478bd9Sstevel@tonic-gate 1987c478bd9Sstevel@tonic-gate mutex_init(&hat->hat_mutex, NULL, MUTEX_DEFAULT, NULL); 1997c478bd9Sstevel@tonic-gate bzero(hat->hat_pages_mapped, 2007c478bd9Sstevel@tonic-gate sizeof (pgcnt_t) * (mmu.max_page_level + 1)); 201*250b7ff9Sjosephb hat->hat_ism_pgcnt = 0; 2027c478bd9Sstevel@tonic-gate hat->hat_stats = 0; 2037c478bd9Sstevel@tonic-gate hat->hat_flags = 0; 2047c478bd9Sstevel@tonic-gate CPUSET_ZERO(hat->hat_cpus); 2057c478bd9Sstevel@tonic-gate hat->hat_htable = NULL; 2067c478bd9Sstevel@tonic-gate hat->hat_ht_hash = NULL; 2077c478bd9Sstevel@tonic-gate return (0); 2087c478bd9Sstevel@tonic-gate } 2097c478bd9Sstevel@tonic-gate 2107c478bd9Sstevel@tonic-gate /* 2117c478bd9Sstevel@tonic-gate * Allocate a hat structure for as. We also create the top level 2127c478bd9Sstevel@tonic-gate * htable and initialize it to contain the kernel hat entries. 2137c478bd9Sstevel@tonic-gate */ 2147c478bd9Sstevel@tonic-gate hat_t * 2157c478bd9Sstevel@tonic-gate hat_alloc(struct as *as) 2167c478bd9Sstevel@tonic-gate { 2177c478bd9Sstevel@tonic-gate hat_t *hat; 2187c478bd9Sstevel@tonic-gate htable_t *ht; /* top level htable */ 2197c478bd9Sstevel@tonic-gate uint_t use_vlp; 2207c478bd9Sstevel@tonic-gate 2217c478bd9Sstevel@tonic-gate /* 2227c478bd9Sstevel@tonic-gate * Once we start creating user process HATs we can enable 2237c478bd9Sstevel@tonic-gate * the htable_steal() code. 2247c478bd9Sstevel@tonic-gate */ 2257c478bd9Sstevel@tonic-gate if (can_steal_post_boot == 0) 2267c478bd9Sstevel@tonic-gate can_steal_post_boot = 1; 2277c478bd9Sstevel@tonic-gate 2287c478bd9Sstevel@tonic-gate ASSERT(AS_WRITE_HELD(as, &as->a_lock)); 2297c478bd9Sstevel@tonic-gate hat = kmem_cache_alloc(hat_cache, KM_SLEEP); 2307c478bd9Sstevel@tonic-gate hat->hat_as = as; 2317c478bd9Sstevel@tonic-gate mutex_init(&hat->hat_mutex, NULL, MUTEX_DEFAULT, NULL); 2327c478bd9Sstevel@tonic-gate ASSERT(hat->hat_flags == 0); 2337c478bd9Sstevel@tonic-gate 2347c478bd9Sstevel@tonic-gate /* 2357c478bd9Sstevel@tonic-gate * a 32 bit process uses a VLP style hat when using PAE 2367c478bd9Sstevel@tonic-gate */ 2377c478bd9Sstevel@tonic-gate #if defined(__amd64) 2387c478bd9Sstevel@tonic-gate use_vlp = (ttoproc(curthread)->p_model == DATAMODEL_ILP32); 2397c478bd9Sstevel@tonic-gate #elif defined(__i386) 2407c478bd9Sstevel@tonic-gate use_vlp = mmu.pae_hat; 2417c478bd9Sstevel@tonic-gate #endif 2427c478bd9Sstevel@tonic-gate if (use_vlp) { 2437c478bd9Sstevel@tonic-gate hat->hat_flags = HAT_VLP; 2447c478bd9Sstevel@tonic-gate bzero(hat->hat_vlp_ptes, VLP_SIZE); 2457c478bd9Sstevel@tonic-gate } 2467c478bd9Sstevel@tonic-gate 2477c478bd9Sstevel@tonic-gate /* 2487c478bd9Sstevel@tonic-gate * Allocate the htable hash 2497c478bd9Sstevel@tonic-gate */ 2507c478bd9Sstevel@tonic-gate if ((hat->hat_flags & HAT_VLP)) { 2517c478bd9Sstevel@tonic-gate hat->hat_num_hash = mmu.vlp_hash_cnt; 2527c478bd9Sstevel@tonic-gate hat->hat_ht_hash = kmem_cache_alloc(vlp_hash_cache, KM_SLEEP); 2537c478bd9Sstevel@tonic-gate } else { 2547c478bd9Sstevel@tonic-gate hat->hat_num_hash = mmu.hash_cnt; 2557c478bd9Sstevel@tonic-gate hat->hat_ht_hash = kmem_cache_alloc(hat_hash_cache, KM_SLEEP); 2567c478bd9Sstevel@tonic-gate } 2577c478bd9Sstevel@tonic-gate bzero(hat->hat_ht_hash, hat->hat_num_hash * sizeof (htable_t *)); 2587c478bd9Sstevel@tonic-gate 2597c478bd9Sstevel@tonic-gate /* 2607c478bd9Sstevel@tonic-gate * Initialize Kernel HAT entries at the top of the top level page 2617c478bd9Sstevel@tonic-gate * table for the new hat. 2627c478bd9Sstevel@tonic-gate * 2637c478bd9Sstevel@tonic-gate * Note that we don't call htable_release() for the top level, that 2647c478bd9Sstevel@tonic-gate * happens when the hat is destroyed in hat_free_end() 2657c478bd9Sstevel@tonic-gate */ 2667c478bd9Sstevel@tonic-gate hat->hat_htable = NULL; 2677c478bd9Sstevel@tonic-gate hat->hat_ht_cached = NULL; 2687c478bd9Sstevel@tonic-gate ht = htable_create(hat, (uintptr_t)0, TOP_LEVEL(hat), NULL); 269ae115bc7Smrj 2707c478bd9Sstevel@tonic-gate if (!(hat->hat_flags & HAT_VLP)) 2717c478bd9Sstevel@tonic-gate x86pte_copy(kas.a_hat->hat_htable, ht, khat_start, 2727c478bd9Sstevel@tonic-gate khat_entries); 2737c478bd9Sstevel@tonic-gate #if defined(__i386) 2747c478bd9Sstevel@tonic-gate else if (khat_entries > 0) 2757c478bd9Sstevel@tonic-gate bcopy(vlp_page + khat_start, hat->hat_vlp_ptes + khat_start, 2767c478bd9Sstevel@tonic-gate khat_entries * sizeof (x86pte_t)); 2777c478bd9Sstevel@tonic-gate #endif 2787c478bd9Sstevel@tonic-gate hat->hat_htable = ht; 2797c478bd9Sstevel@tonic-gate 2807c478bd9Sstevel@tonic-gate #if defined(__i386) 2817c478bd9Sstevel@tonic-gate /* 2827c478bd9Sstevel@tonic-gate * PAE32 HAT alignment is less restrictive than the others to keep 2837c478bd9Sstevel@tonic-gate * the kernel from using too much VA. Because of this we may need 2847c478bd9Sstevel@tonic-gate * one layer further down when kernelbase isn't 1Gig aligned. 2857c478bd9Sstevel@tonic-gate * See hat_free_end() for the htable_release() that goes with this 2867c478bd9Sstevel@tonic-gate * htable_create() 2877c478bd9Sstevel@tonic-gate */ 2887c478bd9Sstevel@tonic-gate if (khat_pae32_htable != NULL) { 2897c478bd9Sstevel@tonic-gate ht = htable_create(hat, kernelbase, 2907c478bd9Sstevel@tonic-gate khat_pae32_htable->ht_level, NULL); 2917c478bd9Sstevel@tonic-gate x86pte_copy(khat_pae32_htable, ht, khat_pae32_start, 2927c478bd9Sstevel@tonic-gate khat_pae32_entries); 2937c478bd9Sstevel@tonic-gate ht->ht_valid_cnt = khat_pae32_entries; 2947c478bd9Sstevel@tonic-gate } 2957c478bd9Sstevel@tonic-gate #endif 2967c478bd9Sstevel@tonic-gate 2977c478bd9Sstevel@tonic-gate /* 298a85a6733Sjosephb * Put it at the start of the global list of all hats (used by stealing) 299a85a6733Sjosephb * 300a85a6733Sjosephb * kas.a_hat is not in the list but is instead used to find the 301a85a6733Sjosephb * first and last items in the list. 302a85a6733Sjosephb * 303a85a6733Sjosephb * - kas.a_hat->hat_next points to the start of the user hats. 304a85a6733Sjosephb * The list ends where hat->hat_next == NULL 305a85a6733Sjosephb * 306a85a6733Sjosephb * - kas.a_hat->hat_prev points to the last of the user hats. 307a85a6733Sjosephb * The list begins where hat->hat_prev == NULL 3087c478bd9Sstevel@tonic-gate */ 3097c478bd9Sstevel@tonic-gate mutex_enter(&hat_list_lock); 310a85a6733Sjosephb hat->hat_prev = NULL; 311a85a6733Sjosephb hat->hat_next = kas.a_hat->hat_next; 312a85a6733Sjosephb if (hat->hat_next) 313a85a6733Sjosephb hat->hat_next->hat_prev = hat; 314a85a6733Sjosephb else 315a85a6733Sjosephb kas.a_hat->hat_prev = hat; 3167c478bd9Sstevel@tonic-gate kas.a_hat->hat_next = hat; 3177c478bd9Sstevel@tonic-gate mutex_exit(&hat_list_lock); 3187c478bd9Sstevel@tonic-gate 3197c478bd9Sstevel@tonic-gate return (hat); 3207c478bd9Sstevel@tonic-gate } 3217c478bd9Sstevel@tonic-gate 3227c478bd9Sstevel@tonic-gate /* 3237c478bd9Sstevel@tonic-gate * process has finished executing but as has not been cleaned up yet. 3247c478bd9Sstevel@tonic-gate */ 3257c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 3267c478bd9Sstevel@tonic-gate void 3277c478bd9Sstevel@tonic-gate hat_free_start(hat_t *hat) 3287c478bd9Sstevel@tonic-gate { 3297c478bd9Sstevel@tonic-gate ASSERT(AS_WRITE_HELD(hat->hat_as, &hat->hat_as->a_lock)); 330a85a6733Sjosephb 331a85a6733Sjosephb /* 332a85a6733Sjosephb * If the hat is currently a stealing victim, wait for the stealing 333a85a6733Sjosephb * to finish. Once we mark it as HAT_FREEING, htable_steal() 334a85a6733Sjosephb * won't look at its pagetables anymore. 335a85a6733Sjosephb */ 3367c478bd9Sstevel@tonic-gate mutex_enter(&hat_list_lock); 337a85a6733Sjosephb while (hat->hat_flags & HAT_VICTIM) 338a85a6733Sjosephb cv_wait(&hat_list_cv, &hat_list_lock); 3397c478bd9Sstevel@tonic-gate hat->hat_flags |= HAT_FREEING; 3407c478bd9Sstevel@tonic-gate mutex_exit(&hat_list_lock); 3417c478bd9Sstevel@tonic-gate } 3427c478bd9Sstevel@tonic-gate 3437c478bd9Sstevel@tonic-gate /* 3447c478bd9Sstevel@tonic-gate * An address space is being destroyed, so we destroy the associated hat. 3457c478bd9Sstevel@tonic-gate */ 3467c478bd9Sstevel@tonic-gate void 3477c478bd9Sstevel@tonic-gate hat_free_end(hat_t *hat) 3487c478bd9Sstevel@tonic-gate { 3497c478bd9Sstevel@tonic-gate int i; 3507c478bd9Sstevel@tonic-gate kmem_cache_t *cache; 3517c478bd9Sstevel@tonic-gate 3527c478bd9Sstevel@tonic-gate #ifdef DEBUG 3537c478bd9Sstevel@tonic-gate for (i = 0; i <= mmu.max_page_level; i++) 3547c478bd9Sstevel@tonic-gate ASSERT(hat->hat_pages_mapped[i] == 0); 3557c478bd9Sstevel@tonic-gate #endif 3567c478bd9Sstevel@tonic-gate ASSERT(hat->hat_flags & HAT_FREEING); 3577c478bd9Sstevel@tonic-gate 3587c478bd9Sstevel@tonic-gate /* 3597c478bd9Sstevel@tonic-gate * must not be running on the given hat 3607c478bd9Sstevel@tonic-gate */ 3617c478bd9Sstevel@tonic-gate ASSERT(CPU->cpu_current_hat != hat); 3627c478bd9Sstevel@tonic-gate 3637c478bd9Sstevel@tonic-gate /* 364a85a6733Sjosephb * Remove it from the list of HATs 3657c478bd9Sstevel@tonic-gate */ 3667c478bd9Sstevel@tonic-gate mutex_enter(&hat_list_lock); 367a85a6733Sjosephb if (hat->hat_prev) 368a85a6733Sjosephb hat->hat_prev->hat_next = hat->hat_next; 369a85a6733Sjosephb else 3707c478bd9Sstevel@tonic-gate kas.a_hat->hat_next = hat->hat_next; 371a85a6733Sjosephb if (hat->hat_next) 372a85a6733Sjosephb hat->hat_next->hat_prev = hat->hat_prev; 373a85a6733Sjosephb else 374a85a6733Sjosephb kas.a_hat->hat_prev = hat->hat_prev; 3757c478bd9Sstevel@tonic-gate mutex_exit(&hat_list_lock); 376a85a6733Sjosephb hat->hat_next = hat->hat_prev = NULL; 3777c478bd9Sstevel@tonic-gate 3787c478bd9Sstevel@tonic-gate /* 3797c478bd9Sstevel@tonic-gate * Make a pass through the htables freeing them all up. 3807c478bd9Sstevel@tonic-gate */ 3817c478bd9Sstevel@tonic-gate htable_purge_hat(hat); 3827c478bd9Sstevel@tonic-gate 3837c478bd9Sstevel@tonic-gate /* 3847c478bd9Sstevel@tonic-gate * Decide which kmem cache the hash table came from, then free it. 3857c478bd9Sstevel@tonic-gate */ 3867c478bd9Sstevel@tonic-gate if (hat->hat_flags & HAT_VLP) 3877c478bd9Sstevel@tonic-gate cache = vlp_hash_cache; 3887c478bd9Sstevel@tonic-gate else 3897c478bd9Sstevel@tonic-gate cache = hat_hash_cache; 3907c478bd9Sstevel@tonic-gate kmem_cache_free(cache, hat->hat_ht_hash); 3917c478bd9Sstevel@tonic-gate hat->hat_ht_hash = NULL; 3927c478bd9Sstevel@tonic-gate 3937c478bd9Sstevel@tonic-gate hat->hat_flags = 0; 3947c478bd9Sstevel@tonic-gate kmem_cache_free(hat_cache, hat); 3957c478bd9Sstevel@tonic-gate } 3967c478bd9Sstevel@tonic-gate 3977c478bd9Sstevel@tonic-gate /* 3987c478bd9Sstevel@tonic-gate * round kernelbase down to a supported value to use for _userlimit 3997c478bd9Sstevel@tonic-gate * 4007c478bd9Sstevel@tonic-gate * userlimit must be aligned down to an entry in the top level htable. 4017c478bd9Sstevel@tonic-gate * The one exception is for 32 bit HAT's running PAE. 4027c478bd9Sstevel@tonic-gate */ 4037c478bd9Sstevel@tonic-gate uintptr_t 4047c478bd9Sstevel@tonic-gate hat_kernelbase(uintptr_t va) 4057c478bd9Sstevel@tonic-gate { 4067c478bd9Sstevel@tonic-gate #if defined(__i386) 4077c478bd9Sstevel@tonic-gate va &= LEVEL_MASK(1); 4087c478bd9Sstevel@tonic-gate #endif 4097c478bd9Sstevel@tonic-gate if (IN_VA_HOLE(va)) 4107c478bd9Sstevel@tonic-gate panic("_userlimit %p will fall in VA hole\n", (void *)va); 4117c478bd9Sstevel@tonic-gate return (va); 4127c478bd9Sstevel@tonic-gate } 4137c478bd9Sstevel@tonic-gate 4147c478bd9Sstevel@tonic-gate /* 4157c478bd9Sstevel@tonic-gate * Initialize hat data structures based on processor MMU information. 4167c478bd9Sstevel@tonic-gate */ 4177c478bd9Sstevel@tonic-gate void 4187c478bd9Sstevel@tonic-gate mmu_init(void) 4197c478bd9Sstevel@tonic-gate { 4207c478bd9Sstevel@tonic-gate uint_t max_htables; 4217c478bd9Sstevel@tonic-gate uint_t pa_bits; 4227c478bd9Sstevel@tonic-gate uint_t va_bits; 4237c478bd9Sstevel@tonic-gate int i; 4247c478bd9Sstevel@tonic-gate 4257c478bd9Sstevel@tonic-gate /* 426ae115bc7Smrj * If CPU enabled the page table global bit, use it for the kernel 427ae115bc7Smrj * This is bit 7 in CR4 (PGE - Page Global Enable). 4287c478bd9Sstevel@tonic-gate */ 429ae115bc7Smrj if ((x86_feature & X86_PGE) != 0 && (getcr4() & CR4_PGE) != 0) 4307c478bd9Sstevel@tonic-gate mmu.pt_global = PT_GLOBAL; 4317c478bd9Sstevel@tonic-gate 4327c478bd9Sstevel@tonic-gate /* 433ae115bc7Smrj * Detect NX and PAE usage. 4347c478bd9Sstevel@tonic-gate */ 435ae115bc7Smrj mmu.pae_hat = kbm_pae_support; 436ae115bc7Smrj if (kbm_nx_support) 4377c478bd9Sstevel@tonic-gate mmu.pt_nx = PT_NX; 438ae115bc7Smrj else 4397c478bd9Sstevel@tonic-gate mmu.pt_nx = 0; 4407c478bd9Sstevel@tonic-gate 441935f8dd0Sjosephb /* 442935f8dd0Sjosephb * Intel CPUs allow speculative caching (in TLB-like h/w) of 443935f8dd0Sjosephb * entries in upper page tables even though there may not be 444935f8dd0Sjosephb * any valid entries in lower tables. This implies we have to 445935f8dd0Sjosephb * re-INVLPG at every upper page table entry invalidation. 446935f8dd0Sjosephb */ 447935f8dd0Sjosephb if (cpuid_getvendor(CPU) == X86_VENDOR_Intel) 448935f8dd0Sjosephb mmu.inval_nonleaf = 1; 449935f8dd0Sjosephb else 450935f8dd0Sjosephb mmu.inval_nonleaf = 0; 4517c478bd9Sstevel@tonic-gate /* 4527c478bd9Sstevel@tonic-gate * Use CPU info to set various MMU parameters 4537c478bd9Sstevel@tonic-gate */ 4547c478bd9Sstevel@tonic-gate cpuid_get_addrsize(CPU, &pa_bits, &va_bits); 4557c478bd9Sstevel@tonic-gate 4567c478bd9Sstevel@tonic-gate if (va_bits < sizeof (void *) * NBBY) { 4577c478bd9Sstevel@tonic-gate mmu.hole_start = (1ul << (va_bits - 1)); 4587c478bd9Sstevel@tonic-gate mmu.hole_end = 0ul - mmu.hole_start - 1; 4597c478bd9Sstevel@tonic-gate } else { 4607c478bd9Sstevel@tonic-gate mmu.hole_end = 0; 4617c478bd9Sstevel@tonic-gate mmu.hole_start = mmu.hole_end - 1; 4627c478bd9Sstevel@tonic-gate } 4637c478bd9Sstevel@tonic-gate #if defined(OPTERON_ERRATUM_121) 4647c478bd9Sstevel@tonic-gate /* 4657c478bd9Sstevel@tonic-gate * If erratum 121 has already been detected at this time, hole_start 4667c478bd9Sstevel@tonic-gate * contains the value to be subtracted from mmu.hole_start. 4677c478bd9Sstevel@tonic-gate */ 4687c478bd9Sstevel@tonic-gate ASSERT(hole_start == 0 || opteron_erratum_121 != 0); 4697c478bd9Sstevel@tonic-gate hole_start = mmu.hole_start - hole_start; 4707c478bd9Sstevel@tonic-gate #else 4717c478bd9Sstevel@tonic-gate hole_start = mmu.hole_start; 4727c478bd9Sstevel@tonic-gate #endif 4737c478bd9Sstevel@tonic-gate hole_end = mmu.hole_end; 4747c478bd9Sstevel@tonic-gate 4757c478bd9Sstevel@tonic-gate mmu.highest_pfn = mmu_btop((1ull << pa_bits) - 1); 4767c478bd9Sstevel@tonic-gate if (mmu.pae_hat == 0 && pa_bits > 32) 4777c478bd9Sstevel@tonic-gate mmu.highest_pfn = PFN_4G - 1; 4787c478bd9Sstevel@tonic-gate 4797c478bd9Sstevel@tonic-gate if (mmu.pae_hat) { 4807c478bd9Sstevel@tonic-gate mmu.pte_size = 8; /* 8 byte PTEs */ 4817c478bd9Sstevel@tonic-gate mmu.pte_size_shift = 3; 4827c478bd9Sstevel@tonic-gate } else { 4837c478bd9Sstevel@tonic-gate mmu.pte_size = 4; /* 4 byte PTEs */ 4847c478bd9Sstevel@tonic-gate mmu.pte_size_shift = 2; 4857c478bd9Sstevel@tonic-gate } 4867c478bd9Sstevel@tonic-gate 4877c478bd9Sstevel@tonic-gate if (mmu.pae_hat && (x86_feature & X86_PAE) == 0) 4887c478bd9Sstevel@tonic-gate panic("Processor does not support PAE"); 4897c478bd9Sstevel@tonic-gate 4907c478bd9Sstevel@tonic-gate if ((x86_feature & X86_CX8) == 0) 4917c478bd9Sstevel@tonic-gate panic("Processor does not support cmpxchg8b instruction"); 4927c478bd9Sstevel@tonic-gate 4937c478bd9Sstevel@tonic-gate /* 4947c478bd9Sstevel@tonic-gate * Initialize parameters based on the 64 or 32 bit kernels and 4957c478bd9Sstevel@tonic-gate * for the 32 bit kernel decide if we should use PAE. 4967c478bd9Sstevel@tonic-gate */ 497ae115bc7Smrj if (kbm_largepage_support) 4987c478bd9Sstevel@tonic-gate mmu.max_page_level = 1; 4997c478bd9Sstevel@tonic-gate else 5007c478bd9Sstevel@tonic-gate mmu.max_page_level = 0; 5017c478bd9Sstevel@tonic-gate mmu_page_sizes = mmu.max_page_level + 1; 5027c478bd9Sstevel@tonic-gate mmu_exported_page_sizes = mmu_page_sizes; 5037c478bd9Sstevel@tonic-gate 5047c478bd9Sstevel@tonic-gate #if defined(__amd64) 5057c478bd9Sstevel@tonic-gate 5067c478bd9Sstevel@tonic-gate mmu.num_level = 4; 5077c478bd9Sstevel@tonic-gate mmu.max_level = 3; 5087c478bd9Sstevel@tonic-gate mmu.ptes_per_table = 512; 5097c478bd9Sstevel@tonic-gate mmu.top_level_count = 512; 5107c478bd9Sstevel@tonic-gate 5117c478bd9Sstevel@tonic-gate mmu.level_shift[0] = 12; 5127c478bd9Sstevel@tonic-gate mmu.level_shift[1] = 21; 5137c478bd9Sstevel@tonic-gate mmu.level_shift[2] = 30; 5147c478bd9Sstevel@tonic-gate mmu.level_shift[3] = 39; 5157c478bd9Sstevel@tonic-gate 5167c478bd9Sstevel@tonic-gate #elif defined(__i386) 5177c478bd9Sstevel@tonic-gate 5187c478bd9Sstevel@tonic-gate if (mmu.pae_hat) { 5197c478bd9Sstevel@tonic-gate mmu.num_level = 3; 5207c478bd9Sstevel@tonic-gate mmu.max_level = 2; 5217c478bd9Sstevel@tonic-gate mmu.ptes_per_table = 512; 5227c478bd9Sstevel@tonic-gate mmu.top_level_count = 4; 5237c478bd9Sstevel@tonic-gate 5247c478bd9Sstevel@tonic-gate mmu.level_shift[0] = 12; 5257c478bd9Sstevel@tonic-gate mmu.level_shift[1] = 21; 5267c478bd9Sstevel@tonic-gate mmu.level_shift[2] = 30; 5277c478bd9Sstevel@tonic-gate 5287c478bd9Sstevel@tonic-gate } else { 5297c478bd9Sstevel@tonic-gate mmu.num_level = 2; 5307c478bd9Sstevel@tonic-gate mmu.max_level = 1; 5317c478bd9Sstevel@tonic-gate mmu.ptes_per_table = 1024; 5327c478bd9Sstevel@tonic-gate mmu.top_level_count = 1024; 5337c478bd9Sstevel@tonic-gate 5347c478bd9Sstevel@tonic-gate mmu.level_shift[0] = 12; 5357c478bd9Sstevel@tonic-gate mmu.level_shift[1] = 22; 5367c478bd9Sstevel@tonic-gate } 5377c478bd9Sstevel@tonic-gate 5387c478bd9Sstevel@tonic-gate #endif /* __i386 */ 5397c478bd9Sstevel@tonic-gate 5407c478bd9Sstevel@tonic-gate for (i = 0; i < mmu.num_level; ++i) { 5417c478bd9Sstevel@tonic-gate mmu.level_size[i] = 1UL << mmu.level_shift[i]; 5427c478bd9Sstevel@tonic-gate mmu.level_offset[i] = mmu.level_size[i] - 1; 5437c478bd9Sstevel@tonic-gate mmu.level_mask[i] = ~mmu.level_offset[i]; 5447c478bd9Sstevel@tonic-gate } 5457c478bd9Sstevel@tonic-gate 546ae115bc7Smrj for (i = 0; i <= mmu.max_page_level; ++i) { 547ae115bc7Smrj mmu.pte_bits[i] = PT_VALID; 548ae115bc7Smrj if (i > 0) 549ae115bc7Smrj mmu.pte_bits[i] |= PT_PAGESIZE; 550ae115bc7Smrj } 5517c478bd9Sstevel@tonic-gate 5527c478bd9Sstevel@tonic-gate /* 5537c478bd9Sstevel@tonic-gate * NOTE Legacy 32 bit PAE mode only has the P_VALID bit at top level. 5547c478bd9Sstevel@tonic-gate */ 5557c478bd9Sstevel@tonic-gate for (i = 1; i < mmu.num_level; ++i) 5567c478bd9Sstevel@tonic-gate mmu.ptp_bits[i] = PT_PTPBITS; 557ae115bc7Smrj 5587c478bd9Sstevel@tonic-gate #if defined(__i386) 5597c478bd9Sstevel@tonic-gate mmu.ptp_bits[2] = PT_VALID; 5607c478bd9Sstevel@tonic-gate #endif 5617c478bd9Sstevel@tonic-gate 5627c478bd9Sstevel@tonic-gate /* 5637c478bd9Sstevel@tonic-gate * Compute how many hash table entries to have per process for htables. 5647c478bd9Sstevel@tonic-gate * We start with 1 page's worth of entries. 5657c478bd9Sstevel@tonic-gate * 5667c478bd9Sstevel@tonic-gate * If physical memory is small, reduce the amount need to cover it. 5677c478bd9Sstevel@tonic-gate */ 5687c478bd9Sstevel@tonic-gate max_htables = physmax / mmu.ptes_per_table; 5697c478bd9Sstevel@tonic-gate mmu.hash_cnt = MMU_PAGESIZE / sizeof (htable_t *); 5707c478bd9Sstevel@tonic-gate while (mmu.hash_cnt > 16 && mmu.hash_cnt >= max_htables) 5717c478bd9Sstevel@tonic-gate mmu.hash_cnt >>= 1; 5727c478bd9Sstevel@tonic-gate mmu.vlp_hash_cnt = mmu.hash_cnt; 5737c478bd9Sstevel@tonic-gate 5747c478bd9Sstevel@tonic-gate #if defined(__amd64) 5757c478bd9Sstevel@tonic-gate /* 5767c478bd9Sstevel@tonic-gate * If running in 64 bits and physical memory is large, 5777c478bd9Sstevel@tonic-gate * increase the size of the cache to cover all of memory for 5787c478bd9Sstevel@tonic-gate * a 64 bit process. 5797c478bd9Sstevel@tonic-gate */ 5807c478bd9Sstevel@tonic-gate #define HASH_MAX_LENGTH 4 5817c478bd9Sstevel@tonic-gate while (mmu.hash_cnt * HASH_MAX_LENGTH < max_htables) 5827c478bd9Sstevel@tonic-gate mmu.hash_cnt <<= 1; 5837c478bd9Sstevel@tonic-gate #endif 5847c478bd9Sstevel@tonic-gate } 5857c478bd9Sstevel@tonic-gate 5867c478bd9Sstevel@tonic-gate 5877c478bd9Sstevel@tonic-gate /* 5887c478bd9Sstevel@tonic-gate * initialize hat data structures 5897c478bd9Sstevel@tonic-gate */ 5907c478bd9Sstevel@tonic-gate void 5917c478bd9Sstevel@tonic-gate hat_init() 5927c478bd9Sstevel@tonic-gate { 5937c478bd9Sstevel@tonic-gate #if defined(__i386) 5947c478bd9Sstevel@tonic-gate /* 5957c478bd9Sstevel@tonic-gate * _userlimit must be aligned correctly 5967c478bd9Sstevel@tonic-gate */ 5977c478bd9Sstevel@tonic-gate if ((_userlimit & LEVEL_MASK(1)) != _userlimit) { 5987c478bd9Sstevel@tonic-gate prom_printf("hat_init(): _userlimit=%p, not aligned at %p\n", 5997c478bd9Sstevel@tonic-gate (void *)_userlimit, (void *)LEVEL_SIZE(1)); 6007c478bd9Sstevel@tonic-gate halt("hat_init(): Unable to continue"); 6017c478bd9Sstevel@tonic-gate } 6027c478bd9Sstevel@tonic-gate #endif 6037c478bd9Sstevel@tonic-gate 6047c478bd9Sstevel@tonic-gate cv_init(&hat_list_cv, NULL, CV_DEFAULT, NULL); 6057c478bd9Sstevel@tonic-gate 6067c478bd9Sstevel@tonic-gate /* 6077c478bd9Sstevel@tonic-gate * initialize kmem caches 6087c478bd9Sstevel@tonic-gate */ 6097c478bd9Sstevel@tonic-gate htable_init(); 6107c478bd9Sstevel@tonic-gate hment_init(); 6117c478bd9Sstevel@tonic-gate 6127c478bd9Sstevel@tonic-gate hat_cache = kmem_cache_create("hat_t", 6137c478bd9Sstevel@tonic-gate sizeof (hat_t), 0, hati_constructor, NULL, NULL, 6147c478bd9Sstevel@tonic-gate NULL, 0, 0); 6157c478bd9Sstevel@tonic-gate 6167c478bd9Sstevel@tonic-gate hat_hash_cache = kmem_cache_create("HatHash", 6177c478bd9Sstevel@tonic-gate mmu.hash_cnt * sizeof (htable_t *), 0, NULL, NULL, NULL, 6187c478bd9Sstevel@tonic-gate NULL, 0, 0); 6197c478bd9Sstevel@tonic-gate 6207c478bd9Sstevel@tonic-gate /* 6217c478bd9Sstevel@tonic-gate * VLP hats can use a smaller hash table size on large memroy machines 6227c478bd9Sstevel@tonic-gate */ 6237c478bd9Sstevel@tonic-gate if (mmu.hash_cnt == mmu.vlp_hash_cnt) { 6247c478bd9Sstevel@tonic-gate vlp_hash_cache = hat_hash_cache; 6257c478bd9Sstevel@tonic-gate } else { 6267c478bd9Sstevel@tonic-gate vlp_hash_cache = kmem_cache_create("HatVlpHash", 6277c478bd9Sstevel@tonic-gate mmu.vlp_hash_cnt * sizeof (htable_t *), 0, NULL, NULL, NULL, 6287c478bd9Sstevel@tonic-gate NULL, 0, 0); 6297c478bd9Sstevel@tonic-gate } 6307c478bd9Sstevel@tonic-gate 6317c478bd9Sstevel@tonic-gate /* 6327c478bd9Sstevel@tonic-gate * Set up the kernel's hat 6337c478bd9Sstevel@tonic-gate */ 6347c478bd9Sstevel@tonic-gate AS_LOCK_ENTER(&kas, &kas.a_lock, RW_WRITER); 6357c478bd9Sstevel@tonic-gate kas.a_hat = kmem_cache_alloc(hat_cache, KM_NOSLEEP); 6367c478bd9Sstevel@tonic-gate mutex_init(&kas.a_hat->hat_mutex, NULL, MUTEX_DEFAULT, NULL); 6377c478bd9Sstevel@tonic-gate kas.a_hat->hat_as = &kas; 6387c478bd9Sstevel@tonic-gate kas.a_hat->hat_flags = 0; 6397c478bd9Sstevel@tonic-gate AS_LOCK_EXIT(&kas, &kas.a_lock); 6407c478bd9Sstevel@tonic-gate 6417c478bd9Sstevel@tonic-gate CPUSET_ZERO(khat_cpuset); 6427c478bd9Sstevel@tonic-gate CPUSET_ADD(khat_cpuset, CPU->cpu_id); 6437c478bd9Sstevel@tonic-gate 6447c478bd9Sstevel@tonic-gate /* 6457c478bd9Sstevel@tonic-gate * The kernel hat's next pointer serves as the head of the hat list . 646a85a6733Sjosephb * The kernel hat's prev pointer tracks the last hat on the list for 647a85a6733Sjosephb * htable_steal() to use. 6487c478bd9Sstevel@tonic-gate */ 6497c478bd9Sstevel@tonic-gate kas.a_hat->hat_next = NULL; 650a85a6733Sjosephb kas.a_hat->hat_prev = NULL; 6517c478bd9Sstevel@tonic-gate 6527c478bd9Sstevel@tonic-gate /* 6537c478bd9Sstevel@tonic-gate * Allocate an htable hash bucket for the kernel 6547c478bd9Sstevel@tonic-gate * XX64 - tune for 64 bit procs 6557c478bd9Sstevel@tonic-gate */ 6567c478bd9Sstevel@tonic-gate kas.a_hat->hat_num_hash = mmu.hash_cnt; 6577c478bd9Sstevel@tonic-gate kas.a_hat->hat_ht_hash = kmem_cache_alloc(hat_hash_cache, KM_NOSLEEP); 6587c478bd9Sstevel@tonic-gate bzero(kas.a_hat->hat_ht_hash, mmu.hash_cnt * sizeof (htable_t *)); 6597c478bd9Sstevel@tonic-gate 6607c478bd9Sstevel@tonic-gate /* 6617c478bd9Sstevel@tonic-gate * zero out the top level and cached htable pointers 6627c478bd9Sstevel@tonic-gate */ 6637c478bd9Sstevel@tonic-gate kas.a_hat->hat_ht_cached = NULL; 6647c478bd9Sstevel@tonic-gate kas.a_hat->hat_htable = NULL; 6659d9461f9Strevtom 6669d9461f9Strevtom /* 6679d9461f9Strevtom * Pre-allocate hrm_hashtab before enabling the collection of 6689d9461f9Strevtom * refmod statistics. Allocating on the fly would mean us 6699d9461f9Strevtom * running the risk of suffering recursive mutex enters or 6709d9461f9Strevtom * deadlocks. 6719d9461f9Strevtom */ 6729d9461f9Strevtom hrm_hashtab = kmem_zalloc(HRM_HASHSIZE * sizeof (struct hrmstat *), 6739d9461f9Strevtom KM_SLEEP); 6747c478bd9Sstevel@tonic-gate } 6757c478bd9Sstevel@tonic-gate 6767c478bd9Sstevel@tonic-gate /* 6777c478bd9Sstevel@tonic-gate * Prepare CPU specific pagetables for VLP processes on 64 bit kernels. 6787c478bd9Sstevel@tonic-gate * 6797c478bd9Sstevel@tonic-gate * Each CPU has a set of 2 pagetables that are reused for any 32 bit 6807c478bd9Sstevel@tonic-gate * process it runs. They are the top level pagetable, hci_vlp_l3ptes, and 6817c478bd9Sstevel@tonic-gate * the next to top level table for the bottom 512 Gig, hci_vlp_l2ptes. 6827c478bd9Sstevel@tonic-gate */ 6837c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 6847c478bd9Sstevel@tonic-gate static void 6857c478bd9Sstevel@tonic-gate hat_vlp_setup(struct cpu *cpu) 6867c478bd9Sstevel@tonic-gate { 6877c478bd9Sstevel@tonic-gate #if defined(__amd64) 6887c478bd9Sstevel@tonic-gate struct hat_cpu_info *hci = cpu->cpu_hat_info; 6897c478bd9Sstevel@tonic-gate pfn_t pfn; 6907c478bd9Sstevel@tonic-gate 6917c478bd9Sstevel@tonic-gate /* 6927c478bd9Sstevel@tonic-gate * allocate the level==2 page table for the bottom most 6937c478bd9Sstevel@tonic-gate * 512Gig of address space (this is where 32 bit apps live) 6947c478bd9Sstevel@tonic-gate */ 6957c478bd9Sstevel@tonic-gate ASSERT(hci != NULL); 6967c478bd9Sstevel@tonic-gate hci->hci_vlp_l2ptes = kmem_zalloc(MMU_PAGESIZE, KM_SLEEP); 6977c478bd9Sstevel@tonic-gate 6987c478bd9Sstevel@tonic-gate /* 6997c478bd9Sstevel@tonic-gate * Allocate a top level pagetable and copy the kernel's 7007c478bd9Sstevel@tonic-gate * entries into it. Then link in hci_vlp_l2ptes in the 1st entry. 7017c478bd9Sstevel@tonic-gate */ 7027c478bd9Sstevel@tonic-gate hci->hci_vlp_l3ptes = kmem_zalloc(MMU_PAGESIZE, KM_SLEEP); 7037c478bd9Sstevel@tonic-gate hci->hci_vlp_pfn = 7047c478bd9Sstevel@tonic-gate hat_getpfnum(kas.a_hat, (caddr_t)hci->hci_vlp_l3ptes); 7057c478bd9Sstevel@tonic-gate ASSERT(hci->hci_vlp_pfn != PFN_INVALID); 7067c478bd9Sstevel@tonic-gate bcopy(vlp_page + khat_start, hci->hci_vlp_l3ptes + khat_start, 7077c478bd9Sstevel@tonic-gate khat_entries * sizeof (x86pte_t)); 7087c478bd9Sstevel@tonic-gate 7097c478bd9Sstevel@tonic-gate pfn = hat_getpfnum(kas.a_hat, (caddr_t)hci->hci_vlp_l2ptes); 7107c478bd9Sstevel@tonic-gate ASSERT(pfn != PFN_INVALID); 7117c478bd9Sstevel@tonic-gate hci->hci_vlp_l3ptes[0] = MAKEPTP(pfn, 2); 7127c478bd9Sstevel@tonic-gate #endif /* __amd64 */ 7137c478bd9Sstevel@tonic-gate } 7147c478bd9Sstevel@tonic-gate 715ae115bc7Smrj /*ARGSUSED*/ 716ae115bc7Smrj static void 717ae115bc7Smrj hat_vlp_teardown(cpu_t *cpu) 718ae115bc7Smrj { 719ae115bc7Smrj #if defined(__amd64) 720ae115bc7Smrj struct hat_cpu_info *hci; 721ae115bc7Smrj 722ae115bc7Smrj if ((hci = cpu->cpu_hat_info) == NULL) 723ae115bc7Smrj return; 724ae115bc7Smrj if (hci->hci_vlp_l2ptes) 725ae115bc7Smrj kmem_free(hci->hci_vlp_l2ptes, MMU_PAGESIZE); 726ae115bc7Smrj if (hci->hci_vlp_l3ptes) 727ae115bc7Smrj kmem_free(hci->hci_vlp_l3ptes, MMU_PAGESIZE); 728ae115bc7Smrj #endif /* __amd64 */ 729ae115bc7Smrj } 730ae115bc7Smrj 7317c478bd9Sstevel@tonic-gate /* 7327c478bd9Sstevel@tonic-gate * Finish filling in the kernel hat. 7337c478bd9Sstevel@tonic-gate * Pre fill in all top level kernel page table entries for the kernel's 7347c478bd9Sstevel@tonic-gate * part of the address range. From this point on we can't use any new 7357c478bd9Sstevel@tonic-gate * kernel large pages if they need PTE's at max_level 736ae115bc7Smrj * 737ae115bc7Smrj * create the kmap mappings. 7387c478bd9Sstevel@tonic-gate */ 7397c478bd9Sstevel@tonic-gate void 7407c478bd9Sstevel@tonic-gate hat_init_finish(void) 7417c478bd9Sstevel@tonic-gate { 7427c478bd9Sstevel@tonic-gate htable_t *top = kas.a_hat->hat_htable; 7437c478bd9Sstevel@tonic-gate htable_t *ht; 7447c478bd9Sstevel@tonic-gate uint_t e; 7457c478bd9Sstevel@tonic-gate x86pte_t pte; 7467c478bd9Sstevel@tonic-gate uintptr_t va = kernelbase; 747ae115bc7Smrj size_t size; 7487c478bd9Sstevel@tonic-gate 7497c478bd9Sstevel@tonic-gate 7507c478bd9Sstevel@tonic-gate #if defined(__i386) 7517c478bd9Sstevel@tonic-gate ASSERT((va & LEVEL_MASK(1)) == va); 7527c478bd9Sstevel@tonic-gate 7537c478bd9Sstevel@tonic-gate /* 7547c478bd9Sstevel@tonic-gate * Deal with kernelbase not 1Gig aligned for 32 bit PAE hats. 7557c478bd9Sstevel@tonic-gate */ 7567c478bd9Sstevel@tonic-gate if (!mmu.pae_hat || (va & LEVEL_OFFSET(mmu.max_level)) == 0) { 7577c478bd9Sstevel@tonic-gate khat_pae32_htable = NULL; 7587c478bd9Sstevel@tonic-gate } else { 7597c478bd9Sstevel@tonic-gate ASSERT(mmu.max_level == 2); 7607c478bd9Sstevel@tonic-gate ASSERT((va & LEVEL_OFFSET(mmu.max_level - 1)) == 0); 7617c478bd9Sstevel@tonic-gate khat_pae32_htable = 7627c478bd9Sstevel@tonic-gate htable_create(kas.a_hat, va, mmu.max_level - 1, NULL); 7637c478bd9Sstevel@tonic-gate khat_pae32_start = htable_va2entry(va, khat_pae32_htable); 7647c478bd9Sstevel@tonic-gate khat_pae32_entries = mmu.ptes_per_table - khat_pae32_start; 7657c478bd9Sstevel@tonic-gate for (e = khat_pae32_start; e < mmu.ptes_per_table; 7667c478bd9Sstevel@tonic-gate ++e, va += LEVEL_SIZE(mmu.max_level - 1)) { 7677c478bd9Sstevel@tonic-gate pte = x86pte_get(khat_pae32_htable, e); 7687c478bd9Sstevel@tonic-gate if (PTE_ISVALID(pte)) 7697c478bd9Sstevel@tonic-gate continue; 7707c478bd9Sstevel@tonic-gate ht = htable_create(kas.a_hat, va, mmu.max_level - 2, 7717c478bd9Sstevel@tonic-gate NULL); 7727c478bd9Sstevel@tonic-gate ASSERT(ht != NULL); 7737c478bd9Sstevel@tonic-gate } 7747c478bd9Sstevel@tonic-gate } 7757c478bd9Sstevel@tonic-gate #endif 7767c478bd9Sstevel@tonic-gate 7777c478bd9Sstevel@tonic-gate /* 7787c478bd9Sstevel@tonic-gate * The kernel hat will need fixed values in the highest level 7797c478bd9Sstevel@tonic-gate * ptable for copying to all other hat's. This implies 7807c478bd9Sstevel@tonic-gate * alignment restrictions on _userlimit. 7817c478bd9Sstevel@tonic-gate * 7827c478bd9Sstevel@tonic-gate * Note we don't htable_release() these htables. This keeps them 7837c478bd9Sstevel@tonic-gate * from ever being stolen or free'd. 7847c478bd9Sstevel@tonic-gate * 7857c478bd9Sstevel@tonic-gate * top_level_count is used instead of ptes_per_table, since 7867c478bd9Sstevel@tonic-gate * on 32-bit PAE we only have 4 usable entries at the top level ptable. 7877c478bd9Sstevel@tonic-gate */ 7887c478bd9Sstevel@tonic-gate if (va == 0) 7897c478bd9Sstevel@tonic-gate khat_start = mmu.top_level_count; 7907c478bd9Sstevel@tonic-gate else 7917c478bd9Sstevel@tonic-gate khat_start = htable_va2entry(va, kas.a_hat->hat_htable); 7927c478bd9Sstevel@tonic-gate khat_entries = mmu.top_level_count - khat_start; 7937c478bd9Sstevel@tonic-gate for (e = khat_start; e < mmu.top_level_count; 7947c478bd9Sstevel@tonic-gate ++e, va += LEVEL_SIZE(mmu.max_level)) { 795ae115bc7Smrj if (IN_HYPERVISOR_VA(va)) 796ae115bc7Smrj continue; 7977c478bd9Sstevel@tonic-gate pte = x86pte_get(top, e); 7987c478bd9Sstevel@tonic-gate if (PTE_ISVALID(pte)) 7997c478bd9Sstevel@tonic-gate continue; 8007c478bd9Sstevel@tonic-gate ht = htable_create(kas.a_hat, va, mmu.max_level - 1, NULL); 8017c478bd9Sstevel@tonic-gate ASSERT(ht != NULL); 8027c478bd9Sstevel@tonic-gate } 8037c478bd9Sstevel@tonic-gate 8047c478bd9Sstevel@tonic-gate /* 8057c478bd9Sstevel@tonic-gate * We are now effectively running on the kernel hat. 8067c478bd9Sstevel@tonic-gate * Clearing use_boot_reserve shuts off using the pre-allocated boot 8077c478bd9Sstevel@tonic-gate * reserve for all HAT allocations. From here on, the reserves are 8087c478bd9Sstevel@tonic-gate * only used when mapping in memory for the hat's own allocations. 8097c478bd9Sstevel@tonic-gate */ 8107c478bd9Sstevel@tonic-gate use_boot_reserve = 0; 8117c478bd9Sstevel@tonic-gate htable_adjust_reserve(); 8127c478bd9Sstevel@tonic-gate 8137c478bd9Sstevel@tonic-gate /* 8147c478bd9Sstevel@tonic-gate * 32 bit kernels use only 4 of the 512 entries in its top level 8157c478bd9Sstevel@tonic-gate * pagetable. We'll use the remainder for the "per CPU" page tables 8167c478bd9Sstevel@tonic-gate * for VLP processes. 8177c478bd9Sstevel@tonic-gate * 818ae115bc7Smrj * We also map the top level kernel pagetable into the kernel to make 819ae115bc7Smrj * it easy to use bcopy to initialize new address spaces. 8207c478bd9Sstevel@tonic-gate */ 8217c478bd9Sstevel@tonic-gate if (mmu.pae_hat) { 8227c478bd9Sstevel@tonic-gate vlp_page = vmem_alloc(heap_arena, MMU_PAGESIZE, VM_SLEEP); 8237c478bd9Sstevel@tonic-gate hat_devload(kas.a_hat, (caddr_t)vlp_page, MMU_PAGESIZE, 8247c478bd9Sstevel@tonic-gate kas.a_hat->hat_htable->ht_pfn, 825ae115bc7Smrj PROT_WRITE | 826ae115bc7Smrj PROT_READ | HAT_NOSYNC | HAT_UNORDERED_OK, 8277c478bd9Sstevel@tonic-gate HAT_LOAD | HAT_LOAD_NOCONSIST); 8287c478bd9Sstevel@tonic-gate } 8297c478bd9Sstevel@tonic-gate hat_vlp_setup(CPU); 830ae115bc7Smrj 831ae115bc7Smrj /* 832ae115bc7Smrj * Create kmap (cached mappings of kernel PTEs) 833ae115bc7Smrj * for 32 bit we map from segmap_start .. ekernelheap 834ae115bc7Smrj * for 64 bit we map from segmap_start .. segmap_start + segmapsize; 835ae115bc7Smrj */ 836ae115bc7Smrj #if defined(__i386) 837ae115bc7Smrj size = (uintptr_t)ekernelheap - segmap_start; 838ae115bc7Smrj #elif defined(__amd64) 839ae115bc7Smrj size = segmapsize; 840ae115bc7Smrj #endif 841ae115bc7Smrj hat_kmap_init((uintptr_t)segmap_start, size); 8427c478bd9Sstevel@tonic-gate } 8437c478bd9Sstevel@tonic-gate 8447c478bd9Sstevel@tonic-gate /* 8457c478bd9Sstevel@tonic-gate * On 32 bit PAE mode, PTE's are 64 bits, but ordinary atomic memory references 8467c478bd9Sstevel@tonic-gate * are 32 bit, so for safety we must use cas64() to install these. 8477c478bd9Sstevel@tonic-gate */ 8487c478bd9Sstevel@tonic-gate #ifdef __i386 8497c478bd9Sstevel@tonic-gate static void 8507c478bd9Sstevel@tonic-gate reload_pae32(hat_t *hat, cpu_t *cpu) 8517c478bd9Sstevel@tonic-gate { 8527c478bd9Sstevel@tonic-gate x86pte_t *src; 8537c478bd9Sstevel@tonic-gate x86pte_t *dest; 8547c478bd9Sstevel@tonic-gate x86pte_t pte; 8557c478bd9Sstevel@tonic-gate int i; 8567c478bd9Sstevel@tonic-gate 8577c478bd9Sstevel@tonic-gate /* 8587c478bd9Sstevel@tonic-gate * Load the 4 entries of the level 2 page table into this 8597c478bd9Sstevel@tonic-gate * cpu's range of the vlp_page and point cr3 at them. 8607c478bd9Sstevel@tonic-gate */ 8617c478bd9Sstevel@tonic-gate ASSERT(mmu.pae_hat); 8627c478bd9Sstevel@tonic-gate src = hat->hat_vlp_ptes; 8637c478bd9Sstevel@tonic-gate dest = vlp_page + (cpu->cpu_id + 1) * VLP_NUM_PTES; 8647c478bd9Sstevel@tonic-gate for (i = 0; i < VLP_NUM_PTES; ++i) { 8657c478bd9Sstevel@tonic-gate for (;;) { 8667c478bd9Sstevel@tonic-gate pte = dest[i]; 8677c478bd9Sstevel@tonic-gate if (pte == src[i]) 8687c478bd9Sstevel@tonic-gate break; 8697c478bd9Sstevel@tonic-gate if (cas64(dest + i, pte, src[i]) != src[i]) 8707c478bd9Sstevel@tonic-gate break; 8717c478bd9Sstevel@tonic-gate } 8727c478bd9Sstevel@tonic-gate } 8737c478bd9Sstevel@tonic-gate } 8747c478bd9Sstevel@tonic-gate #endif 8757c478bd9Sstevel@tonic-gate 8767c478bd9Sstevel@tonic-gate /* 8777c478bd9Sstevel@tonic-gate * Switch to a new active hat, maintaining bit masks to track active CPUs. 8787c478bd9Sstevel@tonic-gate */ 8797c478bd9Sstevel@tonic-gate void 8807c478bd9Sstevel@tonic-gate hat_switch(hat_t *hat) 8817c478bd9Sstevel@tonic-gate { 8827c478bd9Sstevel@tonic-gate uintptr_t newcr3; 8837c478bd9Sstevel@tonic-gate cpu_t *cpu = CPU; 8847c478bd9Sstevel@tonic-gate hat_t *old = cpu->cpu_current_hat; 8857c478bd9Sstevel@tonic-gate 8867c478bd9Sstevel@tonic-gate /* 8877c478bd9Sstevel@tonic-gate * set up this information first, so we don't miss any cross calls 8887c478bd9Sstevel@tonic-gate */ 8897c478bd9Sstevel@tonic-gate if (old != NULL) { 8907c478bd9Sstevel@tonic-gate if (old == hat) 8917c478bd9Sstevel@tonic-gate return; 8927c478bd9Sstevel@tonic-gate if (old != kas.a_hat) 8937c478bd9Sstevel@tonic-gate CPUSET_ATOMIC_DEL(old->hat_cpus, cpu->cpu_id); 8947c478bd9Sstevel@tonic-gate } 8957c478bd9Sstevel@tonic-gate 8967c478bd9Sstevel@tonic-gate /* 89795c0a3c8Sjosephb * Add this CPU to the active set for this HAT. 8987c478bd9Sstevel@tonic-gate */ 8997c478bd9Sstevel@tonic-gate if (hat != kas.a_hat) { 9007c478bd9Sstevel@tonic-gate CPUSET_ATOMIC_ADD(hat->hat_cpus, cpu->cpu_id); 9017c478bd9Sstevel@tonic-gate } 9027c478bd9Sstevel@tonic-gate cpu->cpu_current_hat = hat; 9037c478bd9Sstevel@tonic-gate 9047c478bd9Sstevel@tonic-gate /* 9057c478bd9Sstevel@tonic-gate * now go ahead and load cr3 9067c478bd9Sstevel@tonic-gate */ 9077c478bd9Sstevel@tonic-gate if (hat->hat_flags & HAT_VLP) { 9087c478bd9Sstevel@tonic-gate #if defined(__amd64) 9097c478bd9Sstevel@tonic-gate x86pte_t *vlpptep = cpu->cpu_hat_info->hci_vlp_l2ptes; 9107c478bd9Sstevel@tonic-gate 9117c478bd9Sstevel@tonic-gate VLP_COPY(hat->hat_vlp_ptes, vlpptep); 9127c478bd9Sstevel@tonic-gate newcr3 = MAKECR3(cpu->cpu_hat_info->hci_vlp_pfn); 9137c478bd9Sstevel@tonic-gate #elif defined(__i386) 9147c478bd9Sstevel@tonic-gate reload_pae32(hat, cpu); 9157c478bd9Sstevel@tonic-gate newcr3 = MAKECR3(kas.a_hat->hat_htable->ht_pfn) + 9167c478bd9Sstevel@tonic-gate (cpu->cpu_id + 1) * VLP_SIZE; 9177c478bd9Sstevel@tonic-gate #endif 9187c478bd9Sstevel@tonic-gate } else { 9197c478bd9Sstevel@tonic-gate newcr3 = MAKECR3(hat->hat_htable->ht_pfn); 9207c478bd9Sstevel@tonic-gate } 9217c478bd9Sstevel@tonic-gate setcr3(newcr3); 9227c478bd9Sstevel@tonic-gate ASSERT(cpu == CPU); 9237c478bd9Sstevel@tonic-gate } 9247c478bd9Sstevel@tonic-gate 9257c478bd9Sstevel@tonic-gate /* 9267c478bd9Sstevel@tonic-gate * Utility to return a valid x86pte_t from protections, pfn, and level number 9277c478bd9Sstevel@tonic-gate */ 9287c478bd9Sstevel@tonic-gate static x86pte_t 9297c478bd9Sstevel@tonic-gate hati_mkpte(pfn_t pfn, uint_t attr, level_t level, uint_t flags) 9307c478bd9Sstevel@tonic-gate { 9317c478bd9Sstevel@tonic-gate x86pte_t pte; 9327c478bd9Sstevel@tonic-gate uint_t cache_attr = attr & HAT_ORDER_MASK; 9337c478bd9Sstevel@tonic-gate 9347c478bd9Sstevel@tonic-gate pte = MAKEPTE(pfn, level); 9357c478bd9Sstevel@tonic-gate 9367c478bd9Sstevel@tonic-gate if (attr & PROT_WRITE) 9377c478bd9Sstevel@tonic-gate PTE_SET(pte, PT_WRITABLE); 9387c478bd9Sstevel@tonic-gate 9397c478bd9Sstevel@tonic-gate if (attr & PROT_USER) 9407c478bd9Sstevel@tonic-gate PTE_SET(pte, PT_USER); 9417c478bd9Sstevel@tonic-gate 9427c478bd9Sstevel@tonic-gate if (!(attr & PROT_EXEC)) 9437c478bd9Sstevel@tonic-gate PTE_SET(pte, mmu.pt_nx); 9447c478bd9Sstevel@tonic-gate 9457c478bd9Sstevel@tonic-gate /* 946ae115bc7Smrj * Set the software bits used track ref/mod sync's and hments. 947ae115bc7Smrj * If not using REF/MOD, set them to avoid h/w rewriting PTEs. 9487c478bd9Sstevel@tonic-gate */ 9497c478bd9Sstevel@tonic-gate if (flags & HAT_LOAD_NOCONSIST) 950ae115bc7Smrj PTE_SET(pte, PT_NOCONSIST | PT_REF | PT_MOD); 951ae115bc7Smrj else if (attr & HAT_NOSYNC) 952ae115bc7Smrj PTE_SET(pte, PT_NOSYNC | PT_REF | PT_MOD); 9537c478bd9Sstevel@tonic-gate 9547c478bd9Sstevel@tonic-gate /* 9557c478bd9Sstevel@tonic-gate * Set the caching attributes in the PTE. The combination 9567c478bd9Sstevel@tonic-gate * of attributes are poorly defined, so we pay attention 9577c478bd9Sstevel@tonic-gate * to them in the given order. 9587c478bd9Sstevel@tonic-gate * 9597c478bd9Sstevel@tonic-gate * The test for HAT_STRICTORDER is different because it's defined 9607c478bd9Sstevel@tonic-gate * as "0" - which was a stupid thing to do, but is too late to change! 9617c478bd9Sstevel@tonic-gate */ 9627c478bd9Sstevel@tonic-gate if (cache_attr == HAT_STRICTORDER) { 9637c478bd9Sstevel@tonic-gate PTE_SET(pte, PT_NOCACHE); 9647c478bd9Sstevel@tonic-gate /*LINTED [Lint hates empty ifs, but it's the obvious way to do this] */ 9657c478bd9Sstevel@tonic-gate } else if (cache_attr & (HAT_UNORDERED_OK | HAT_STORECACHING_OK)) { 9667c478bd9Sstevel@tonic-gate /* nothing to set */; 9677c478bd9Sstevel@tonic-gate } else if (cache_attr & (HAT_MERGING_OK | HAT_LOADCACHING_OK)) { 9687c478bd9Sstevel@tonic-gate PTE_SET(pte, PT_NOCACHE); 9697c478bd9Sstevel@tonic-gate if (x86_feature & X86_PAT) 9707c478bd9Sstevel@tonic-gate PTE_SET(pte, (level == 0) ? PT_PAT_4K : PT_PAT_LARGE); 9717c478bd9Sstevel@tonic-gate else 9727c478bd9Sstevel@tonic-gate PTE_SET(pte, PT_WRITETHRU); 9737c478bd9Sstevel@tonic-gate } else { 9747c478bd9Sstevel@tonic-gate panic("hati_mkpte(): bad caching attributes: %x\n", cache_attr); 9757c478bd9Sstevel@tonic-gate } 9767c478bd9Sstevel@tonic-gate 9777c478bd9Sstevel@tonic-gate return (pte); 9787c478bd9Sstevel@tonic-gate } 9797c478bd9Sstevel@tonic-gate 9807c478bd9Sstevel@tonic-gate /* 9817c478bd9Sstevel@tonic-gate * Duplicate address translations of the parent to the child. 9827c478bd9Sstevel@tonic-gate * This function really isn't used anymore. 9837c478bd9Sstevel@tonic-gate */ 9847c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 9857c478bd9Sstevel@tonic-gate int 9867c478bd9Sstevel@tonic-gate hat_dup(hat_t *old, hat_t *new, caddr_t addr, size_t len, uint_t flag) 9877c478bd9Sstevel@tonic-gate { 9887c478bd9Sstevel@tonic-gate ASSERT((uintptr_t)addr < kernelbase); 9897c478bd9Sstevel@tonic-gate ASSERT(new != kas.a_hat); 9907c478bd9Sstevel@tonic-gate ASSERT(old != kas.a_hat); 9917c478bd9Sstevel@tonic-gate return (0); 9927c478bd9Sstevel@tonic-gate } 9937c478bd9Sstevel@tonic-gate 9947c478bd9Sstevel@tonic-gate /* 9957c478bd9Sstevel@tonic-gate * Allocate any hat resources required for a process being swapped in. 9967c478bd9Sstevel@tonic-gate */ 9977c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 9987c478bd9Sstevel@tonic-gate void 9997c478bd9Sstevel@tonic-gate hat_swapin(hat_t *hat) 10007c478bd9Sstevel@tonic-gate { 10017c478bd9Sstevel@tonic-gate /* do nothing - we let everything fault back in */ 10027c478bd9Sstevel@tonic-gate } 10037c478bd9Sstevel@tonic-gate 10047c478bd9Sstevel@tonic-gate /* 10057c478bd9Sstevel@tonic-gate * Unload all translations associated with an address space of a process 10067c478bd9Sstevel@tonic-gate * that is being swapped out. 10077c478bd9Sstevel@tonic-gate */ 10087c478bd9Sstevel@tonic-gate void 10097c478bd9Sstevel@tonic-gate hat_swapout(hat_t *hat) 10107c478bd9Sstevel@tonic-gate { 10117c478bd9Sstevel@tonic-gate uintptr_t vaddr = (uintptr_t)0; 10127c478bd9Sstevel@tonic-gate uintptr_t eaddr = _userlimit; 10137c478bd9Sstevel@tonic-gate htable_t *ht = NULL; 10147c478bd9Sstevel@tonic-gate level_t l; 10157c478bd9Sstevel@tonic-gate 10167c478bd9Sstevel@tonic-gate /* 10177c478bd9Sstevel@tonic-gate * We can't just call hat_unload(hat, 0, _userlimit...) here, because 10187c478bd9Sstevel@tonic-gate * seg_spt and shared pagetables can't be swapped out. 10197c478bd9Sstevel@tonic-gate * Take a look at segspt_shmswapout() - it's a big no-op. 10207c478bd9Sstevel@tonic-gate * 10217c478bd9Sstevel@tonic-gate * Instead we'll walk through all the address space and unload 10227c478bd9Sstevel@tonic-gate * any mappings which we are sure are not shared, not locked. 10237c478bd9Sstevel@tonic-gate */ 10247c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(vaddr)); 10257c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(eaddr)); 10267c478bd9Sstevel@tonic-gate ASSERT(AS_LOCK_HELD(hat->hat_as, &hat->hat_as->a_lock)); 10277c478bd9Sstevel@tonic-gate if ((uintptr_t)hat->hat_as->a_userlimit < eaddr) 10287c478bd9Sstevel@tonic-gate eaddr = (uintptr_t)hat->hat_as->a_userlimit; 10297c478bd9Sstevel@tonic-gate 10307c478bd9Sstevel@tonic-gate while (vaddr < eaddr) { 10317c478bd9Sstevel@tonic-gate (void) htable_walk(hat, &ht, &vaddr, eaddr); 10327c478bd9Sstevel@tonic-gate if (ht == NULL) 10337c478bd9Sstevel@tonic-gate break; 10347c478bd9Sstevel@tonic-gate 10357c478bd9Sstevel@tonic-gate ASSERT(!IN_VA_HOLE(vaddr)); 10367c478bd9Sstevel@tonic-gate 10377c478bd9Sstevel@tonic-gate /* 10387c478bd9Sstevel@tonic-gate * If the page table is shared skip its entire range. 10397c478bd9Sstevel@tonic-gate * This code knows that only level 0 page tables are shared 10407c478bd9Sstevel@tonic-gate */ 10417c478bd9Sstevel@tonic-gate l = ht->ht_level; 10427c478bd9Sstevel@tonic-gate if (ht->ht_flags & HTABLE_SHARED_PFN) { 10437c478bd9Sstevel@tonic-gate ASSERT(l == 0); 10447c478bd9Sstevel@tonic-gate vaddr = ht->ht_vaddr + LEVEL_SIZE(1); 10457c478bd9Sstevel@tonic-gate htable_release(ht); 10467c478bd9Sstevel@tonic-gate ht = NULL; 10477c478bd9Sstevel@tonic-gate continue; 10487c478bd9Sstevel@tonic-gate } 10497c478bd9Sstevel@tonic-gate 10507c478bd9Sstevel@tonic-gate /* 10517c478bd9Sstevel@tonic-gate * If the page table has no locked entries, unload this one. 10527c478bd9Sstevel@tonic-gate */ 10537c478bd9Sstevel@tonic-gate if (ht->ht_lock_cnt == 0) 10547c478bd9Sstevel@tonic-gate hat_unload(hat, (caddr_t)vaddr, LEVEL_SIZE(l), 10557c478bd9Sstevel@tonic-gate HAT_UNLOAD_UNMAP); 10567c478bd9Sstevel@tonic-gate 10577c478bd9Sstevel@tonic-gate /* 10587c478bd9Sstevel@tonic-gate * If we have a level 0 page table with locked entries, 10597c478bd9Sstevel@tonic-gate * skip the entire page table, otherwise skip just one entry. 10607c478bd9Sstevel@tonic-gate */ 10617c478bd9Sstevel@tonic-gate if (ht->ht_lock_cnt > 0 && l == 0) 10627c478bd9Sstevel@tonic-gate vaddr = ht->ht_vaddr + LEVEL_SIZE(1); 10637c478bd9Sstevel@tonic-gate else 10647c478bd9Sstevel@tonic-gate vaddr += LEVEL_SIZE(l); 10657c478bd9Sstevel@tonic-gate } 10667c478bd9Sstevel@tonic-gate if (ht) 10677c478bd9Sstevel@tonic-gate htable_release(ht); 10687c478bd9Sstevel@tonic-gate 10697c478bd9Sstevel@tonic-gate /* 10707c478bd9Sstevel@tonic-gate * We're in swapout because the system is low on memory, so 10717c478bd9Sstevel@tonic-gate * go back and flush all the htables off the cached list. 10727c478bd9Sstevel@tonic-gate */ 10737c478bd9Sstevel@tonic-gate htable_purge_hat(hat); 10747c478bd9Sstevel@tonic-gate } 10757c478bd9Sstevel@tonic-gate 10767c478bd9Sstevel@tonic-gate /* 10777c478bd9Sstevel@tonic-gate * returns number of bytes that have valid mappings in hat. 10787c478bd9Sstevel@tonic-gate */ 10797c478bd9Sstevel@tonic-gate size_t 10807c478bd9Sstevel@tonic-gate hat_get_mapped_size(hat_t *hat) 10817c478bd9Sstevel@tonic-gate { 10827c478bd9Sstevel@tonic-gate size_t total = 0; 10837c478bd9Sstevel@tonic-gate int l; 10847c478bd9Sstevel@tonic-gate 10857c478bd9Sstevel@tonic-gate for (l = 0; l <= mmu.max_page_level; l++) 10867c478bd9Sstevel@tonic-gate total += (hat->hat_pages_mapped[l] << LEVEL_SHIFT(l)); 1087*250b7ff9Sjosephb total += hat->hat_ism_pgcnt; 10887c478bd9Sstevel@tonic-gate 10897c478bd9Sstevel@tonic-gate return (total); 10907c478bd9Sstevel@tonic-gate } 10917c478bd9Sstevel@tonic-gate 10927c478bd9Sstevel@tonic-gate /* 10937c478bd9Sstevel@tonic-gate * enable/disable collection of stats for hat. 10947c478bd9Sstevel@tonic-gate */ 10957c478bd9Sstevel@tonic-gate int 10967c478bd9Sstevel@tonic-gate hat_stats_enable(hat_t *hat) 10977c478bd9Sstevel@tonic-gate { 10987c478bd9Sstevel@tonic-gate atomic_add_32(&hat->hat_stats, 1); 10997c478bd9Sstevel@tonic-gate return (1); 11007c478bd9Sstevel@tonic-gate } 11017c478bd9Sstevel@tonic-gate 11027c478bd9Sstevel@tonic-gate void 11037c478bd9Sstevel@tonic-gate hat_stats_disable(hat_t *hat) 11047c478bd9Sstevel@tonic-gate { 11057c478bd9Sstevel@tonic-gate atomic_add_32(&hat->hat_stats, -1); 11067c478bd9Sstevel@tonic-gate } 11077c478bd9Sstevel@tonic-gate 11087c478bd9Sstevel@tonic-gate /* 11097c478bd9Sstevel@tonic-gate * Utility to sync the ref/mod bits from a page table entry to the page_t 11107c478bd9Sstevel@tonic-gate * We must be holding the mapping list lock when this is called. 11117c478bd9Sstevel@tonic-gate */ 11127c478bd9Sstevel@tonic-gate static void 11137c478bd9Sstevel@tonic-gate hati_sync_pte_to_page(page_t *pp, x86pte_t pte, level_t level) 11147c478bd9Sstevel@tonic-gate { 11157c478bd9Sstevel@tonic-gate uint_t rm = 0; 11167c478bd9Sstevel@tonic-gate pgcnt_t pgcnt; 11177c478bd9Sstevel@tonic-gate 1118ae115bc7Smrj if (PTE_GET(pte, PT_SOFTWARE) >= PT_NOSYNC) 11197c478bd9Sstevel@tonic-gate return; 11207c478bd9Sstevel@tonic-gate 11217c478bd9Sstevel@tonic-gate if (PTE_GET(pte, PT_REF)) 11227c478bd9Sstevel@tonic-gate rm |= P_REF; 11237c478bd9Sstevel@tonic-gate 11247c478bd9Sstevel@tonic-gate if (PTE_GET(pte, PT_MOD)) 11257c478bd9Sstevel@tonic-gate rm |= P_MOD; 11267c478bd9Sstevel@tonic-gate 11277c478bd9Sstevel@tonic-gate if (rm == 0) 11287c478bd9Sstevel@tonic-gate return; 11297c478bd9Sstevel@tonic-gate 11307c478bd9Sstevel@tonic-gate /* 11317c478bd9Sstevel@tonic-gate * sync to all constituent pages of a large page 11327c478bd9Sstevel@tonic-gate */ 11337c478bd9Sstevel@tonic-gate ASSERT(x86_hm_held(pp)); 11347c478bd9Sstevel@tonic-gate pgcnt = page_get_pagecnt(level); 11357c478bd9Sstevel@tonic-gate ASSERT(IS_P2ALIGNED(pp->p_pagenum, pgcnt)); 11367c478bd9Sstevel@tonic-gate for (; pgcnt > 0; --pgcnt) { 11377c478bd9Sstevel@tonic-gate /* 11387c478bd9Sstevel@tonic-gate * hat_page_demote() can't decrease 11397c478bd9Sstevel@tonic-gate * pszc below this mapping size 11407c478bd9Sstevel@tonic-gate * since this large mapping existed after we 11417c478bd9Sstevel@tonic-gate * took mlist lock. 11427c478bd9Sstevel@tonic-gate */ 11437c478bd9Sstevel@tonic-gate ASSERT(pp->p_szc >= level); 11447c478bd9Sstevel@tonic-gate hat_page_setattr(pp, rm); 11457c478bd9Sstevel@tonic-gate ++pp; 11467c478bd9Sstevel@tonic-gate } 11477c478bd9Sstevel@tonic-gate } 11487c478bd9Sstevel@tonic-gate 11497c478bd9Sstevel@tonic-gate /* 11507c478bd9Sstevel@tonic-gate * This the set of PTE bits for PFN, permissions and caching 1151ae115bc7Smrj * that require a TLB flush (hat_tlb_inval) if changed on a HAT_LOAD_REMAP 11527c478bd9Sstevel@tonic-gate */ 11537c478bd9Sstevel@tonic-gate #define PT_REMAP_BITS \ 11547c478bd9Sstevel@tonic-gate (PT_PADDR | PT_NX | PT_WRITABLE | PT_WRITETHRU | \ 11557c478bd9Sstevel@tonic-gate PT_NOCACHE | PT_PAT_4K | PT_PAT_LARGE) 11567c478bd9Sstevel@tonic-gate 1157b193e412Skchow #define REMAPASSERT(EX) if (!(EX)) panic("hati_pte_map: " #EX) 11587c478bd9Sstevel@tonic-gate /* 11597c478bd9Sstevel@tonic-gate * Do the low-level work to get a mapping entered into a HAT's pagetables 11607c478bd9Sstevel@tonic-gate * and in the mapping list of the associated page_t. 11617c478bd9Sstevel@tonic-gate */ 1162ae115bc7Smrj static int 11637c478bd9Sstevel@tonic-gate hati_pte_map( 11647c478bd9Sstevel@tonic-gate htable_t *ht, 11657c478bd9Sstevel@tonic-gate uint_t entry, 11667c478bd9Sstevel@tonic-gate page_t *pp, 11677c478bd9Sstevel@tonic-gate x86pte_t pte, 11687c478bd9Sstevel@tonic-gate int flags, 11697c478bd9Sstevel@tonic-gate void *pte_ptr) 11707c478bd9Sstevel@tonic-gate { 11717c478bd9Sstevel@tonic-gate hat_t *hat = ht->ht_hat; 11727c478bd9Sstevel@tonic-gate x86pte_t old_pte; 11737c478bd9Sstevel@tonic-gate level_t l = ht->ht_level; 11747c478bd9Sstevel@tonic-gate hment_t *hm; 11757c478bd9Sstevel@tonic-gate uint_t is_consist; 1176ae115bc7Smrj int rv = 0; 11777c478bd9Sstevel@tonic-gate 11787c478bd9Sstevel@tonic-gate /* 11797c478bd9Sstevel@tonic-gate * Is this a consistant (ie. need mapping list lock) mapping? 11807c478bd9Sstevel@tonic-gate */ 11817c478bd9Sstevel@tonic-gate is_consist = (pp != NULL && (flags & HAT_LOAD_NOCONSIST) == 0); 11827c478bd9Sstevel@tonic-gate 11837c478bd9Sstevel@tonic-gate /* 11847c478bd9Sstevel@tonic-gate * Track locked mapping count in the htable. Do this first, 11857c478bd9Sstevel@tonic-gate * as we track locking even if there already is a mapping present. 11867c478bd9Sstevel@tonic-gate */ 11877c478bd9Sstevel@tonic-gate if ((flags & HAT_LOAD_LOCK) != 0 && hat != kas.a_hat) 11887c478bd9Sstevel@tonic-gate HTABLE_LOCK_INC(ht); 11897c478bd9Sstevel@tonic-gate 11907c478bd9Sstevel@tonic-gate /* 11917c478bd9Sstevel@tonic-gate * Acquire the page's mapping list lock and get an hment to use. 11927c478bd9Sstevel@tonic-gate * Note that hment_prepare() might return NULL. 11937c478bd9Sstevel@tonic-gate */ 11947c478bd9Sstevel@tonic-gate if (is_consist) { 11957c478bd9Sstevel@tonic-gate x86_hm_enter(pp); 11967c478bd9Sstevel@tonic-gate hm = hment_prepare(ht, entry, pp); 11977c478bd9Sstevel@tonic-gate } 11987c478bd9Sstevel@tonic-gate 11997c478bd9Sstevel@tonic-gate /* 12007c478bd9Sstevel@tonic-gate * Set the new pte, retrieving the old one at the same time. 12017c478bd9Sstevel@tonic-gate */ 12027c478bd9Sstevel@tonic-gate old_pte = x86pte_set(ht, entry, pte, pte_ptr); 12037c478bd9Sstevel@tonic-gate 12047c478bd9Sstevel@tonic-gate /* 1205ae115bc7Smrj * did we get a large page / page table collision? 12067c478bd9Sstevel@tonic-gate */ 1207ae115bc7Smrj if (old_pte == LPAGE_ERROR) { 1208ae115bc7Smrj rv = -1; 1209ae115bc7Smrj goto done; 12107c478bd9Sstevel@tonic-gate } 12117c478bd9Sstevel@tonic-gate 1212ae115bc7Smrj /* 1213ae115bc7Smrj * If the mapping didn't change there is nothing more to do. 1214ae115bc7Smrj */ 1215ae115bc7Smrj if (PTE_EQUIV(pte, old_pte)) 1216ae115bc7Smrj goto done; 1217ae115bc7Smrj 12187c478bd9Sstevel@tonic-gate /* 12197c478bd9Sstevel@tonic-gate * Install a new mapping in the page's mapping list 12207c478bd9Sstevel@tonic-gate */ 12217c478bd9Sstevel@tonic-gate if (!PTE_ISVALID(old_pte)) { 12227c478bd9Sstevel@tonic-gate if (is_consist) { 12237c478bd9Sstevel@tonic-gate hment_assign(ht, entry, pp, hm); 12247c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 12257c478bd9Sstevel@tonic-gate } else { 12267c478bd9Sstevel@tonic-gate ASSERT(flags & HAT_LOAD_NOCONSIST); 12277c478bd9Sstevel@tonic-gate } 12287c478bd9Sstevel@tonic-gate HTABLE_INC(ht->ht_valid_cnt); 12297c478bd9Sstevel@tonic-gate PGCNT_INC(hat, l); 1230ae115bc7Smrj return (rv); 12317c478bd9Sstevel@tonic-gate } 12327c478bd9Sstevel@tonic-gate 12337c478bd9Sstevel@tonic-gate /* 12347c478bd9Sstevel@tonic-gate * Remap's are more complicated: 12357c478bd9Sstevel@tonic-gate * - HAT_LOAD_REMAP must be specified if changing the pfn. 12367c478bd9Sstevel@tonic-gate * We also require that NOCONSIST be specified. 12377c478bd9Sstevel@tonic-gate * - Otherwise only permission or caching bits may change. 12387c478bd9Sstevel@tonic-gate */ 12397c478bd9Sstevel@tonic-gate if (!PTE_ISPAGE(old_pte, l)) 12407c478bd9Sstevel@tonic-gate panic("non-null/page mapping pte=" FMT_PTE, old_pte); 12417c478bd9Sstevel@tonic-gate 12427c478bd9Sstevel@tonic-gate if (PTE2PFN(old_pte, l) != PTE2PFN(pte, l)) { 1243b193e412Skchow REMAPASSERT(flags & HAT_LOAD_REMAP); 1244b193e412Skchow REMAPASSERT(flags & HAT_LOAD_NOCONSIST); 1245ae115bc7Smrj REMAPASSERT(PTE_GET(old_pte, PT_SOFTWARE) >= PT_NOCONSIST); 1246b193e412Skchow REMAPASSERT(pf_is_memory(PTE2PFN(old_pte, l)) == 12477c478bd9Sstevel@tonic-gate pf_is_memory(PTE2PFN(pte, l))); 1248b193e412Skchow REMAPASSERT(!is_consist); 12497c478bd9Sstevel@tonic-gate } 12507c478bd9Sstevel@tonic-gate 12517c478bd9Sstevel@tonic-gate /* 12527c478bd9Sstevel@tonic-gate * We only let remaps change the bits for PFNs, permissions 12537c478bd9Sstevel@tonic-gate * or caching type. 12547c478bd9Sstevel@tonic-gate */ 12557c478bd9Sstevel@tonic-gate ASSERT(PTE_GET(old_pte, ~(PT_REMAP_BITS | PT_REF | PT_MOD)) == 12567c478bd9Sstevel@tonic-gate PTE_GET(pte, ~PT_REMAP_BITS)); 12577c478bd9Sstevel@tonic-gate 12587c478bd9Sstevel@tonic-gate /* 12597c478bd9Sstevel@tonic-gate * We don't create any mapping list entries on a remap, so release 12607c478bd9Sstevel@tonic-gate * any allocated hment after we drop the mapping list lock. 12617c478bd9Sstevel@tonic-gate */ 1262ae115bc7Smrj done: 12637c478bd9Sstevel@tonic-gate if (is_consist) { 12647c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 12657c478bd9Sstevel@tonic-gate if (hm != NULL) 12667c478bd9Sstevel@tonic-gate hment_free(hm); 12677c478bd9Sstevel@tonic-gate } 1268ae115bc7Smrj return (rv); 12697c478bd9Sstevel@tonic-gate } 12707c478bd9Sstevel@tonic-gate 12717c478bd9Sstevel@tonic-gate /* 1272ae115bc7Smrj * Internal routine to load a single page table entry. This only fails if 1273ae115bc7Smrj * we attempt to overwrite a page table link with a large page. 12747c478bd9Sstevel@tonic-gate */ 1275ae115bc7Smrj static int 12767c478bd9Sstevel@tonic-gate hati_load_common( 12777c478bd9Sstevel@tonic-gate hat_t *hat, 12787c478bd9Sstevel@tonic-gate uintptr_t va, 12797c478bd9Sstevel@tonic-gate page_t *pp, 12807c478bd9Sstevel@tonic-gate uint_t attr, 12817c478bd9Sstevel@tonic-gate uint_t flags, 12827c478bd9Sstevel@tonic-gate level_t level, 12837c478bd9Sstevel@tonic-gate pfn_t pfn) 12847c478bd9Sstevel@tonic-gate { 12857c478bd9Sstevel@tonic-gate htable_t *ht; 12867c478bd9Sstevel@tonic-gate uint_t entry; 12877c478bd9Sstevel@tonic-gate x86pte_t pte; 1288ae115bc7Smrj int rv = 0; 12897c478bd9Sstevel@tonic-gate 1290aac11643Sjosephb /* 1291aac11643Sjosephb * The number 16 is arbitrary and here to catch a recursion problem 1292aac11643Sjosephb * early before we blow out the kernel stack. 1293aac11643Sjosephb */ 1294aac11643Sjosephb ++curthread->t_hatdepth; 1295aac11643Sjosephb ASSERT(curthread->t_hatdepth < 16); 1296aac11643Sjosephb 12977c478bd9Sstevel@tonic-gate ASSERT(hat == kas.a_hat || 12987c478bd9Sstevel@tonic-gate AS_LOCK_HELD(hat->hat_as, &hat->hat_as->a_lock)); 12997c478bd9Sstevel@tonic-gate 13007c478bd9Sstevel@tonic-gate if (flags & HAT_LOAD_SHARE) 13017c478bd9Sstevel@tonic-gate hat->hat_flags |= HAT_SHARED; 13027c478bd9Sstevel@tonic-gate 13037c478bd9Sstevel@tonic-gate /* 13047c478bd9Sstevel@tonic-gate * Find the page table that maps this page if it already exists. 13057c478bd9Sstevel@tonic-gate */ 13067c478bd9Sstevel@tonic-gate ht = htable_lookup(hat, va, level); 13077c478bd9Sstevel@tonic-gate 13087c478bd9Sstevel@tonic-gate /* 1309aac11643Sjosephb * We must have HAT_LOAD_NOCONSIST if page_t is NULL. 13107c478bd9Sstevel@tonic-gate */ 1311aac11643Sjosephb if (pp == NULL) 13127c478bd9Sstevel@tonic-gate flags |= HAT_LOAD_NOCONSIST; 13137c478bd9Sstevel@tonic-gate 13147c478bd9Sstevel@tonic-gate if (ht == NULL) { 13157c478bd9Sstevel@tonic-gate ht = htable_create(hat, va, level, NULL); 13167c478bd9Sstevel@tonic-gate ASSERT(ht != NULL); 13177c478bd9Sstevel@tonic-gate } 13187c478bd9Sstevel@tonic-gate entry = htable_va2entry(va, ht); 13197c478bd9Sstevel@tonic-gate 13207c478bd9Sstevel@tonic-gate /* 13217c478bd9Sstevel@tonic-gate * a bunch of paranoid error checking 13227c478bd9Sstevel@tonic-gate */ 13237c478bd9Sstevel@tonic-gate ASSERT(ht->ht_busy > 0); 13247c478bd9Sstevel@tonic-gate if (ht->ht_vaddr > va || va > HTABLE_LAST_PAGE(ht)) 13257c478bd9Sstevel@tonic-gate panic("hati_load_common: bad htable %p, va %p", ht, (void *)va); 13267c478bd9Sstevel@tonic-gate ASSERT(ht->ht_level == level); 13277c478bd9Sstevel@tonic-gate 13287c478bd9Sstevel@tonic-gate /* 13297c478bd9Sstevel@tonic-gate * construct the new PTE 13307c478bd9Sstevel@tonic-gate */ 13317c478bd9Sstevel@tonic-gate if (hat == kas.a_hat) 13327c478bd9Sstevel@tonic-gate attr &= ~PROT_USER; 13337c478bd9Sstevel@tonic-gate pte = hati_mkpte(pfn, attr, level, flags); 13347c478bd9Sstevel@tonic-gate if (hat == kas.a_hat && va >= kernelbase) 13357c478bd9Sstevel@tonic-gate PTE_SET(pte, mmu.pt_global); 13367c478bd9Sstevel@tonic-gate 13377c478bd9Sstevel@tonic-gate /* 13387c478bd9Sstevel@tonic-gate * establish the mapping 13397c478bd9Sstevel@tonic-gate */ 1340ae115bc7Smrj rv = hati_pte_map(ht, entry, pp, pte, flags, NULL); 13417c478bd9Sstevel@tonic-gate 13427c478bd9Sstevel@tonic-gate /* 13437c478bd9Sstevel@tonic-gate * release the htable and any reserves 13447c478bd9Sstevel@tonic-gate */ 13457c478bd9Sstevel@tonic-gate htable_release(ht); 1346aac11643Sjosephb --curthread->t_hatdepth; 1347ae115bc7Smrj return (rv); 13487c478bd9Sstevel@tonic-gate } 13497c478bd9Sstevel@tonic-gate 13507c478bd9Sstevel@tonic-gate /* 13517c478bd9Sstevel@tonic-gate * special case of hat_memload to deal with some kernel addrs for performance 13527c478bd9Sstevel@tonic-gate */ 13537c478bd9Sstevel@tonic-gate static void 13547c478bd9Sstevel@tonic-gate hat_kmap_load( 13557c478bd9Sstevel@tonic-gate caddr_t addr, 13567c478bd9Sstevel@tonic-gate page_t *pp, 13577c478bd9Sstevel@tonic-gate uint_t attr, 13587c478bd9Sstevel@tonic-gate uint_t flags) 13597c478bd9Sstevel@tonic-gate { 13607c478bd9Sstevel@tonic-gate uintptr_t va = (uintptr_t)addr; 13617c478bd9Sstevel@tonic-gate x86pte_t pte; 13627c478bd9Sstevel@tonic-gate pfn_t pfn = page_pptonum(pp); 13637c478bd9Sstevel@tonic-gate pgcnt_t pg_off = mmu_btop(va - mmu.kmap_addr); 13647c478bd9Sstevel@tonic-gate htable_t *ht; 13657c478bd9Sstevel@tonic-gate uint_t entry; 13667c478bd9Sstevel@tonic-gate void *pte_ptr; 13677c478bd9Sstevel@tonic-gate 13687c478bd9Sstevel@tonic-gate /* 13697c478bd9Sstevel@tonic-gate * construct the requested PTE 13707c478bd9Sstevel@tonic-gate */ 13717c478bd9Sstevel@tonic-gate attr &= ~PROT_USER; 13727c478bd9Sstevel@tonic-gate attr |= HAT_STORECACHING_OK; 13737c478bd9Sstevel@tonic-gate pte = hati_mkpte(pfn, attr, 0, flags); 13747c478bd9Sstevel@tonic-gate PTE_SET(pte, mmu.pt_global); 13757c478bd9Sstevel@tonic-gate 13767c478bd9Sstevel@tonic-gate /* 13777c478bd9Sstevel@tonic-gate * Figure out the pte_ptr and htable and use common code to finish up 13787c478bd9Sstevel@tonic-gate */ 13797c478bd9Sstevel@tonic-gate if (mmu.pae_hat) 13807c478bd9Sstevel@tonic-gate pte_ptr = mmu.kmap_ptes + pg_off; 13817c478bd9Sstevel@tonic-gate else 13827c478bd9Sstevel@tonic-gate pte_ptr = (x86pte32_t *)mmu.kmap_ptes + pg_off; 13837c478bd9Sstevel@tonic-gate ht = mmu.kmap_htables[(va - mmu.kmap_htables[0]->ht_vaddr) >> 13847c478bd9Sstevel@tonic-gate LEVEL_SHIFT(1)]; 13857c478bd9Sstevel@tonic-gate entry = htable_va2entry(va, ht); 1386aac11643Sjosephb ++curthread->t_hatdepth; 1387aac11643Sjosephb ASSERT(curthread->t_hatdepth < 16); 1388ae115bc7Smrj (void) hati_pte_map(ht, entry, pp, pte, flags, pte_ptr); 1389aac11643Sjosephb --curthread->t_hatdepth; 13907c478bd9Sstevel@tonic-gate } 13917c478bd9Sstevel@tonic-gate 13927c478bd9Sstevel@tonic-gate /* 13937c478bd9Sstevel@tonic-gate * hat_memload() - load a translation to the given page struct 13947c478bd9Sstevel@tonic-gate * 13957c478bd9Sstevel@tonic-gate * Flags for hat_memload/hat_devload/hat_*attr. 13967c478bd9Sstevel@tonic-gate * 13977c478bd9Sstevel@tonic-gate * HAT_LOAD Default flags to load a translation to the page. 13987c478bd9Sstevel@tonic-gate * 13997c478bd9Sstevel@tonic-gate * HAT_LOAD_LOCK Lock down mapping resources; hat_map(), hat_memload(), 14007c478bd9Sstevel@tonic-gate * and hat_devload(). 14017c478bd9Sstevel@tonic-gate * 14027c478bd9Sstevel@tonic-gate * HAT_LOAD_NOCONSIST Do not add mapping to page_t mapping list. 1403ae115bc7Smrj * sets PT_NOCONSIST 14047c478bd9Sstevel@tonic-gate * 14057c478bd9Sstevel@tonic-gate * HAT_LOAD_SHARE A flag to hat_memload() to indicate h/w page tables 14067c478bd9Sstevel@tonic-gate * that map some user pages (not kas) is shared by more 14077c478bd9Sstevel@tonic-gate * than one process (eg. ISM). 14087c478bd9Sstevel@tonic-gate * 14097c478bd9Sstevel@tonic-gate * HAT_LOAD_REMAP Reload a valid pte with a different page frame. 14107c478bd9Sstevel@tonic-gate * 14117c478bd9Sstevel@tonic-gate * HAT_NO_KALLOC Do not kmem_alloc while creating the mapping; at this 14127c478bd9Sstevel@tonic-gate * point, it's setting up mapping to allocate internal 14137c478bd9Sstevel@tonic-gate * hat layer data structures. This flag forces hat layer 14147c478bd9Sstevel@tonic-gate * to tap its reserves in order to prevent infinite 14157c478bd9Sstevel@tonic-gate * recursion. 14167c478bd9Sstevel@tonic-gate * 14177c478bd9Sstevel@tonic-gate * The following is a protection attribute (like PROT_READ, etc.) 14187c478bd9Sstevel@tonic-gate * 1419ae115bc7Smrj * HAT_NOSYNC set PT_NOSYNC - this mapping's ref/mod bits 14207c478bd9Sstevel@tonic-gate * are never cleared. 14217c478bd9Sstevel@tonic-gate * 14227c478bd9Sstevel@tonic-gate * Installing new valid PTE's and creation of the mapping list 14237c478bd9Sstevel@tonic-gate * entry are controlled under the same lock. It's derived from the 14247c478bd9Sstevel@tonic-gate * page_t being mapped. 14257c478bd9Sstevel@tonic-gate */ 14267c478bd9Sstevel@tonic-gate static uint_t supported_memload_flags = 14277c478bd9Sstevel@tonic-gate HAT_LOAD | HAT_LOAD_LOCK | HAT_LOAD_ADV | HAT_LOAD_NOCONSIST | 14287c478bd9Sstevel@tonic-gate HAT_LOAD_SHARE | HAT_NO_KALLOC | HAT_LOAD_REMAP | HAT_LOAD_TEXT; 14297c478bd9Sstevel@tonic-gate 14307c478bd9Sstevel@tonic-gate void 14317c478bd9Sstevel@tonic-gate hat_memload( 14327c478bd9Sstevel@tonic-gate hat_t *hat, 14337c478bd9Sstevel@tonic-gate caddr_t addr, 14347c478bd9Sstevel@tonic-gate page_t *pp, 14357c478bd9Sstevel@tonic-gate uint_t attr, 14367c478bd9Sstevel@tonic-gate uint_t flags) 14377c478bd9Sstevel@tonic-gate { 14387c478bd9Sstevel@tonic-gate uintptr_t va = (uintptr_t)addr; 14397c478bd9Sstevel@tonic-gate level_t level = 0; 14407c478bd9Sstevel@tonic-gate pfn_t pfn = page_pptonum(pp); 14417c478bd9Sstevel@tonic-gate 14427c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(va)); 1443ae115bc7Smrj ASSERT(hat == kas.a_hat || va < _userlimit); 14447c478bd9Sstevel@tonic-gate ASSERT(hat == kas.a_hat || 14457c478bd9Sstevel@tonic-gate AS_LOCK_HELD(hat->hat_as, &hat->hat_as->a_lock)); 14467c478bd9Sstevel@tonic-gate ASSERT((flags & supported_memload_flags) == flags); 14477c478bd9Sstevel@tonic-gate 14487c478bd9Sstevel@tonic-gate ASSERT(!IN_VA_HOLE(va)); 14497c478bd9Sstevel@tonic-gate ASSERT(!PP_ISFREE(pp)); 14507c478bd9Sstevel@tonic-gate 14517c478bd9Sstevel@tonic-gate /* 14527c478bd9Sstevel@tonic-gate * kernel address special case for performance. 14537c478bd9Sstevel@tonic-gate */ 14547c478bd9Sstevel@tonic-gate if (mmu.kmap_addr <= va && va < mmu.kmap_eaddr) { 14557c478bd9Sstevel@tonic-gate ASSERT(hat == kas.a_hat); 14567c478bd9Sstevel@tonic-gate hat_kmap_load(addr, pp, attr, flags); 14577c478bd9Sstevel@tonic-gate return; 14587c478bd9Sstevel@tonic-gate } 14597c478bd9Sstevel@tonic-gate 14607c478bd9Sstevel@tonic-gate /* 14617c478bd9Sstevel@tonic-gate * This is used for memory with normal caching enabled, so 14627c478bd9Sstevel@tonic-gate * always set HAT_STORECACHING_OK. 14637c478bd9Sstevel@tonic-gate */ 14647c478bd9Sstevel@tonic-gate attr |= HAT_STORECACHING_OK; 1465ae115bc7Smrj if (hati_load_common(hat, va, pp, attr, flags, level, pfn) != 0) 1466ae115bc7Smrj panic("unexpected hati_load_common() failure"); 14677c478bd9Sstevel@tonic-gate } 14687c478bd9Sstevel@tonic-gate 14697c478bd9Sstevel@tonic-gate /* 14707c478bd9Sstevel@tonic-gate * Load the given array of page structs using large pages when possible 14717c478bd9Sstevel@tonic-gate */ 14727c478bd9Sstevel@tonic-gate void 14737c478bd9Sstevel@tonic-gate hat_memload_array( 14747c478bd9Sstevel@tonic-gate hat_t *hat, 14757c478bd9Sstevel@tonic-gate caddr_t addr, 14767c478bd9Sstevel@tonic-gate size_t len, 14777c478bd9Sstevel@tonic-gate page_t **pages, 14787c478bd9Sstevel@tonic-gate uint_t attr, 14797c478bd9Sstevel@tonic-gate uint_t flags) 14807c478bd9Sstevel@tonic-gate { 14817c478bd9Sstevel@tonic-gate uintptr_t va = (uintptr_t)addr; 14827c478bd9Sstevel@tonic-gate uintptr_t eaddr = va + len; 14837c478bd9Sstevel@tonic-gate level_t level; 14847c478bd9Sstevel@tonic-gate size_t pgsize; 14857c478bd9Sstevel@tonic-gate pgcnt_t pgindx = 0; 14867c478bd9Sstevel@tonic-gate pfn_t pfn; 14877c478bd9Sstevel@tonic-gate pgcnt_t i; 14887c478bd9Sstevel@tonic-gate 14897c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(va)); 1490ae115bc7Smrj ASSERT(hat == kas.a_hat || va + len <= _userlimit); 14917c478bd9Sstevel@tonic-gate ASSERT(hat == kas.a_hat || 14927c478bd9Sstevel@tonic-gate AS_LOCK_HELD(hat->hat_as, &hat->hat_as->a_lock)); 14937c478bd9Sstevel@tonic-gate ASSERT((flags & supported_memload_flags) == flags); 14947c478bd9Sstevel@tonic-gate 14957c478bd9Sstevel@tonic-gate /* 14967c478bd9Sstevel@tonic-gate * memload is used for memory with full caching enabled, so 14977c478bd9Sstevel@tonic-gate * set HAT_STORECACHING_OK. 14987c478bd9Sstevel@tonic-gate */ 14997c478bd9Sstevel@tonic-gate attr |= HAT_STORECACHING_OK; 15007c478bd9Sstevel@tonic-gate 15017c478bd9Sstevel@tonic-gate /* 15027c478bd9Sstevel@tonic-gate * handle all pages using largest possible pagesize 15037c478bd9Sstevel@tonic-gate */ 15047c478bd9Sstevel@tonic-gate while (va < eaddr) { 15057c478bd9Sstevel@tonic-gate /* 15067c478bd9Sstevel@tonic-gate * decide what level mapping to use (ie. pagesize) 15077c478bd9Sstevel@tonic-gate */ 15087c478bd9Sstevel@tonic-gate pfn = page_pptonum(pages[pgindx]); 15097c478bd9Sstevel@tonic-gate for (level = mmu.max_page_level; ; --level) { 15107c478bd9Sstevel@tonic-gate pgsize = LEVEL_SIZE(level); 15117c478bd9Sstevel@tonic-gate if (level == 0) 15127c478bd9Sstevel@tonic-gate break; 1513ae115bc7Smrj 15147c478bd9Sstevel@tonic-gate if (!IS_P2ALIGNED(va, pgsize) || 15157c478bd9Sstevel@tonic-gate (eaddr - va) < pgsize || 1516ae115bc7Smrj !IS_P2ALIGNED(pfn_to_pa(pfn), pgsize)) 15177c478bd9Sstevel@tonic-gate continue; 15187c478bd9Sstevel@tonic-gate 15197c478bd9Sstevel@tonic-gate /* 15207c478bd9Sstevel@tonic-gate * To use a large mapping of this size, all the 15217c478bd9Sstevel@tonic-gate * pages we are passed must be sequential subpages 15227c478bd9Sstevel@tonic-gate * of the large page. 15237c478bd9Sstevel@tonic-gate * hat_page_demote() can't change p_szc because 15247c478bd9Sstevel@tonic-gate * all pages are locked. 15257c478bd9Sstevel@tonic-gate */ 15267c478bd9Sstevel@tonic-gate if (pages[pgindx]->p_szc >= level) { 15277c478bd9Sstevel@tonic-gate for (i = 0; i < mmu_btop(pgsize); ++i) { 15287c478bd9Sstevel@tonic-gate if (pfn + i != 15297c478bd9Sstevel@tonic-gate page_pptonum(pages[pgindx + i])) 15307c478bd9Sstevel@tonic-gate break; 15317c478bd9Sstevel@tonic-gate ASSERT(pages[pgindx + i]->p_szc >= 15327c478bd9Sstevel@tonic-gate level); 15337c478bd9Sstevel@tonic-gate ASSERT(pages[pgindx] + i == 15347c478bd9Sstevel@tonic-gate pages[pgindx + i]); 15357c478bd9Sstevel@tonic-gate } 15367c478bd9Sstevel@tonic-gate if (i == mmu_btop(pgsize)) 15377c478bd9Sstevel@tonic-gate break; 15387c478bd9Sstevel@tonic-gate } 15397c478bd9Sstevel@tonic-gate } 15407c478bd9Sstevel@tonic-gate 15417c478bd9Sstevel@tonic-gate /* 1542ae115bc7Smrj * Load this page mapping. If the load fails, try a smaller 1543ae115bc7Smrj * pagesize. 15447c478bd9Sstevel@tonic-gate */ 15457c478bd9Sstevel@tonic-gate ASSERT(!IN_VA_HOLE(va)); 1546ae115bc7Smrj while (hati_load_common(hat, va, pages[pgindx], attr, 1547*250b7ff9Sjosephb flags, level, pfn) != 0) { 1548ae115bc7Smrj if (level == 0) 1549ae115bc7Smrj panic("unexpected hati_load_common() failure"); 1550ae115bc7Smrj --level; 1551ae115bc7Smrj pgsize = LEVEL_SIZE(level); 1552ae115bc7Smrj } 15537c478bd9Sstevel@tonic-gate 15547c478bd9Sstevel@tonic-gate /* 15557c478bd9Sstevel@tonic-gate * move to next page 15567c478bd9Sstevel@tonic-gate */ 15577c478bd9Sstevel@tonic-gate va += pgsize; 15587c478bd9Sstevel@tonic-gate pgindx += mmu_btop(pgsize); 15597c478bd9Sstevel@tonic-gate } 15607c478bd9Sstevel@tonic-gate } 15617c478bd9Sstevel@tonic-gate 15627c478bd9Sstevel@tonic-gate /* 15637c478bd9Sstevel@tonic-gate * void hat_devload(hat, addr, len, pf, attr, flags) 15647c478bd9Sstevel@tonic-gate * load/lock the given page frame number 15657c478bd9Sstevel@tonic-gate * 15667c478bd9Sstevel@tonic-gate * Advisory ordering attributes. Apply only to device mappings. 15677c478bd9Sstevel@tonic-gate * 15687c478bd9Sstevel@tonic-gate * HAT_STRICTORDER: the CPU must issue the references in order, as the 15697c478bd9Sstevel@tonic-gate * programmer specified. This is the default. 15707c478bd9Sstevel@tonic-gate * HAT_UNORDERED_OK: the CPU may reorder the references (this is all kinds 15717c478bd9Sstevel@tonic-gate * of reordering; store or load with store or load). 15727c478bd9Sstevel@tonic-gate * HAT_MERGING_OK: merging and batching: the CPU may merge individual stores 15737c478bd9Sstevel@tonic-gate * to consecutive locations (for example, turn two consecutive byte 15747c478bd9Sstevel@tonic-gate * stores into one halfword store), and it may batch individual loads 15757c478bd9Sstevel@tonic-gate * (for example, turn two consecutive byte loads into one halfword load). 15767c478bd9Sstevel@tonic-gate * This also implies re-ordering. 15777c478bd9Sstevel@tonic-gate * HAT_LOADCACHING_OK: the CPU may cache the data it fetches and reuse it 15787c478bd9Sstevel@tonic-gate * until another store occurs. The default is to fetch new data 15797c478bd9Sstevel@tonic-gate * on every load. This also implies merging. 15807c478bd9Sstevel@tonic-gate * HAT_STORECACHING_OK: the CPU may keep the data in the cache and push it to 15817c478bd9Sstevel@tonic-gate * the device (perhaps with other data) at a later time. The default is 15827c478bd9Sstevel@tonic-gate * to push the data right away. This also implies load caching. 15837c478bd9Sstevel@tonic-gate * 15847c478bd9Sstevel@tonic-gate * Equivalent of hat_memload(), but can be used for device memory where 15857c478bd9Sstevel@tonic-gate * there are no page_t's and we support additional flags (write merging, etc). 15867c478bd9Sstevel@tonic-gate * Note that we can have large page mappings with this interface. 15877c478bd9Sstevel@tonic-gate */ 15887c478bd9Sstevel@tonic-gate int supported_devload_flags = HAT_LOAD | HAT_LOAD_LOCK | 15897c478bd9Sstevel@tonic-gate HAT_LOAD_NOCONSIST | HAT_STRICTORDER | HAT_UNORDERED_OK | 15907c478bd9Sstevel@tonic-gate HAT_MERGING_OK | HAT_LOADCACHING_OK | HAT_STORECACHING_OK; 15917c478bd9Sstevel@tonic-gate 15927c478bd9Sstevel@tonic-gate void 15937c478bd9Sstevel@tonic-gate hat_devload( 15947c478bd9Sstevel@tonic-gate hat_t *hat, 15957c478bd9Sstevel@tonic-gate caddr_t addr, 15967c478bd9Sstevel@tonic-gate size_t len, 15977c478bd9Sstevel@tonic-gate pfn_t pfn, 15987c478bd9Sstevel@tonic-gate uint_t attr, 15997c478bd9Sstevel@tonic-gate int flags) 16007c478bd9Sstevel@tonic-gate { 16017c478bd9Sstevel@tonic-gate uintptr_t va = ALIGN2PAGE(addr); 16027c478bd9Sstevel@tonic-gate uintptr_t eva = va + len; 16037c478bd9Sstevel@tonic-gate level_t level; 16047c478bd9Sstevel@tonic-gate size_t pgsize; 16057c478bd9Sstevel@tonic-gate page_t *pp; 16067c478bd9Sstevel@tonic-gate int f; /* per PTE copy of flags - maybe modified */ 16077c478bd9Sstevel@tonic-gate uint_t a; /* per PTE copy of attr */ 16087c478bd9Sstevel@tonic-gate 16097c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(va)); 1610ae115bc7Smrj ASSERT(hat == kas.a_hat || eva <= _userlimit); 16117c478bd9Sstevel@tonic-gate ASSERT(hat == kas.a_hat || 16127c478bd9Sstevel@tonic-gate AS_LOCK_HELD(hat->hat_as, &hat->hat_as->a_lock)); 16137c478bd9Sstevel@tonic-gate ASSERT((flags & supported_devload_flags) == flags); 16147c478bd9Sstevel@tonic-gate 16157c478bd9Sstevel@tonic-gate /* 16167c478bd9Sstevel@tonic-gate * handle all pages 16177c478bd9Sstevel@tonic-gate */ 16187c478bd9Sstevel@tonic-gate while (va < eva) { 16197c478bd9Sstevel@tonic-gate 16207c478bd9Sstevel@tonic-gate /* 16217c478bd9Sstevel@tonic-gate * decide what level mapping to use (ie. pagesize) 16227c478bd9Sstevel@tonic-gate */ 16237c478bd9Sstevel@tonic-gate for (level = mmu.max_page_level; ; --level) { 16247c478bd9Sstevel@tonic-gate pgsize = LEVEL_SIZE(level); 16257c478bd9Sstevel@tonic-gate if (level == 0) 16267c478bd9Sstevel@tonic-gate break; 16277c478bd9Sstevel@tonic-gate if (IS_P2ALIGNED(va, pgsize) && 16287c478bd9Sstevel@tonic-gate (eva - va) >= pgsize && 16297c478bd9Sstevel@tonic-gate IS_P2ALIGNED(pfn, mmu_btop(pgsize))) 16307c478bd9Sstevel@tonic-gate break; 16317c478bd9Sstevel@tonic-gate } 16327c478bd9Sstevel@tonic-gate 16337c478bd9Sstevel@tonic-gate /* 1634ae115bc7Smrj * If this is just memory then allow caching (this happens 16357c478bd9Sstevel@tonic-gate * for the nucleus pages) - though HAT_PLAT_NOCACHE can be used 1636ae115bc7Smrj * to override that. If we don't have a page_t then make sure 16377c478bd9Sstevel@tonic-gate * NOCONSIST is set. 16387c478bd9Sstevel@tonic-gate */ 16397c478bd9Sstevel@tonic-gate a = attr; 16407c478bd9Sstevel@tonic-gate f = flags; 16417c478bd9Sstevel@tonic-gate if (pf_is_memory(pfn)) { 16427c478bd9Sstevel@tonic-gate if (!(a & HAT_PLAT_NOCACHE)) 16437c478bd9Sstevel@tonic-gate a |= HAT_STORECACHING_OK; 16447c478bd9Sstevel@tonic-gate 16457c478bd9Sstevel@tonic-gate if (f & HAT_LOAD_NOCONSIST) 16467c478bd9Sstevel@tonic-gate pp = NULL; 16477c478bd9Sstevel@tonic-gate else 16487c478bd9Sstevel@tonic-gate pp = page_numtopp_nolock(pfn); 16497c478bd9Sstevel@tonic-gate } else { 16507c478bd9Sstevel@tonic-gate pp = NULL; 16517c478bd9Sstevel@tonic-gate f |= HAT_LOAD_NOCONSIST; 16527c478bd9Sstevel@tonic-gate } 16537c478bd9Sstevel@tonic-gate 16547c478bd9Sstevel@tonic-gate /* 16557c478bd9Sstevel@tonic-gate * load this page mapping 16567c478bd9Sstevel@tonic-gate */ 16577c478bd9Sstevel@tonic-gate ASSERT(!IN_VA_HOLE(va)); 1658ae115bc7Smrj while (hati_load_common(hat, va, pp, a, f, level, pfn) != 0) { 1659ae115bc7Smrj if (level == 0) 1660ae115bc7Smrj panic("unexpected hati_load_common() failure"); 1661ae115bc7Smrj --level; 1662ae115bc7Smrj pgsize = LEVEL_SIZE(level); 1663ae115bc7Smrj } 16647c478bd9Sstevel@tonic-gate 16657c478bd9Sstevel@tonic-gate /* 16667c478bd9Sstevel@tonic-gate * move to next page 16677c478bd9Sstevel@tonic-gate */ 16687c478bd9Sstevel@tonic-gate va += pgsize; 16697c478bd9Sstevel@tonic-gate pfn += mmu_btop(pgsize); 16707c478bd9Sstevel@tonic-gate } 16717c478bd9Sstevel@tonic-gate } 16727c478bd9Sstevel@tonic-gate 16737c478bd9Sstevel@tonic-gate /* 16747c478bd9Sstevel@tonic-gate * void hat_unlock(hat, addr, len) 16757c478bd9Sstevel@tonic-gate * unlock the mappings to a given range of addresses 16767c478bd9Sstevel@tonic-gate * 16777c478bd9Sstevel@tonic-gate * Locks are tracked by ht_lock_cnt in the htable. 16787c478bd9Sstevel@tonic-gate */ 16797c478bd9Sstevel@tonic-gate void 16807c478bd9Sstevel@tonic-gate hat_unlock(hat_t *hat, caddr_t addr, size_t len) 16817c478bd9Sstevel@tonic-gate { 16827c478bd9Sstevel@tonic-gate uintptr_t vaddr = (uintptr_t)addr; 16837c478bd9Sstevel@tonic-gate uintptr_t eaddr = vaddr + len; 16847c478bd9Sstevel@tonic-gate htable_t *ht = NULL; 16857c478bd9Sstevel@tonic-gate 16867c478bd9Sstevel@tonic-gate /* 16877c478bd9Sstevel@tonic-gate * kernel entries are always locked, we don't track lock counts 16887c478bd9Sstevel@tonic-gate */ 1689ae115bc7Smrj ASSERT(hat == kas.a_hat || eaddr <= _userlimit); 16907c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(vaddr)); 16917c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(eaddr)); 16927c478bd9Sstevel@tonic-gate if (hat == kas.a_hat) 16937c478bd9Sstevel@tonic-gate return; 16947c478bd9Sstevel@tonic-gate if (eaddr > _userlimit) 16957c478bd9Sstevel@tonic-gate panic("hat_unlock() address out of range - above _userlimit"); 16967c478bd9Sstevel@tonic-gate 16977c478bd9Sstevel@tonic-gate ASSERT(AS_LOCK_HELD(hat->hat_as, &hat->hat_as->a_lock)); 16987c478bd9Sstevel@tonic-gate while (vaddr < eaddr) { 16997c478bd9Sstevel@tonic-gate (void) htable_walk(hat, &ht, &vaddr, eaddr); 17007c478bd9Sstevel@tonic-gate if (ht == NULL) 17017c478bd9Sstevel@tonic-gate break; 17027c478bd9Sstevel@tonic-gate 17037c478bd9Sstevel@tonic-gate ASSERT(!IN_VA_HOLE(vaddr)); 17047c478bd9Sstevel@tonic-gate 17057c478bd9Sstevel@tonic-gate if (ht->ht_lock_cnt < 1) 17067c478bd9Sstevel@tonic-gate panic("hat_unlock(): lock_cnt < 1, " 17077c478bd9Sstevel@tonic-gate "htable=%p, vaddr=%p\n", ht, (caddr_t)vaddr); 17087c478bd9Sstevel@tonic-gate HTABLE_LOCK_DEC(ht); 17097c478bd9Sstevel@tonic-gate 17107c478bd9Sstevel@tonic-gate vaddr += LEVEL_SIZE(ht->ht_level); 17117c478bd9Sstevel@tonic-gate } 17127c478bd9Sstevel@tonic-gate if (ht) 17137c478bd9Sstevel@tonic-gate htable_release(ht); 17147c478bd9Sstevel@tonic-gate } 17157c478bd9Sstevel@tonic-gate 17167c478bd9Sstevel@tonic-gate /* 17177c478bd9Sstevel@tonic-gate * Cross call service routine to demap a virtual page on 17187c478bd9Sstevel@tonic-gate * the current CPU or flush all mappings in TLB. 17197c478bd9Sstevel@tonic-gate */ 17207c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 17217c478bd9Sstevel@tonic-gate static int 17227c478bd9Sstevel@tonic-gate hati_demap_func(xc_arg_t a1, xc_arg_t a2, xc_arg_t a3) 17237c478bd9Sstevel@tonic-gate { 17247c478bd9Sstevel@tonic-gate hat_t *hat = (hat_t *)a1; 17257c478bd9Sstevel@tonic-gate caddr_t addr = (caddr_t)a2; 17267c478bd9Sstevel@tonic-gate 17277c478bd9Sstevel@tonic-gate /* 17287c478bd9Sstevel@tonic-gate * If the target hat isn't the kernel and this CPU isn't operating 17297c478bd9Sstevel@tonic-gate * in the target hat, we can ignore the cross call. 17307c478bd9Sstevel@tonic-gate */ 17317c478bd9Sstevel@tonic-gate if (hat != kas.a_hat && hat != CPU->cpu_current_hat) 17327c478bd9Sstevel@tonic-gate return (0); 17337c478bd9Sstevel@tonic-gate 17347c478bd9Sstevel@tonic-gate /* 17357c478bd9Sstevel@tonic-gate * For a normal address, we just flush one page mapping 17367c478bd9Sstevel@tonic-gate */ 17377c478bd9Sstevel@tonic-gate if ((uintptr_t)addr != DEMAP_ALL_ADDR) { 1738ae115bc7Smrj mmu_tlbflush_entry(addr); 17397c478bd9Sstevel@tonic-gate return (0); 17407c478bd9Sstevel@tonic-gate } 17417c478bd9Sstevel@tonic-gate 17427c478bd9Sstevel@tonic-gate /* 17437c478bd9Sstevel@tonic-gate * Otherwise we reload cr3 to effect a complete TLB flush. 17447c478bd9Sstevel@tonic-gate * 17457c478bd9Sstevel@tonic-gate * A reload of cr3 on a VLP process also means we must also recopy in 17467c478bd9Sstevel@tonic-gate * the pte values from the struct hat 17477c478bd9Sstevel@tonic-gate */ 17487c478bd9Sstevel@tonic-gate if (hat->hat_flags & HAT_VLP) { 17497c478bd9Sstevel@tonic-gate #if defined(__amd64) 17507c478bd9Sstevel@tonic-gate x86pte_t *vlpptep = CPU->cpu_hat_info->hci_vlp_l2ptes; 17517c478bd9Sstevel@tonic-gate 17527c478bd9Sstevel@tonic-gate VLP_COPY(hat->hat_vlp_ptes, vlpptep); 17537c478bd9Sstevel@tonic-gate #elif defined(__i386) 17547c478bd9Sstevel@tonic-gate reload_pae32(hat, CPU); 17557c478bd9Sstevel@tonic-gate #endif 17567c478bd9Sstevel@tonic-gate } 17577c478bd9Sstevel@tonic-gate reload_cr3(); 17587c478bd9Sstevel@tonic-gate return (0); 17597c478bd9Sstevel@tonic-gate } 17607c478bd9Sstevel@tonic-gate 176195c0a3c8Sjosephb /* 176295c0a3c8Sjosephb * Flush all TLB entries, including global (ie. kernel) ones. 176395c0a3c8Sjosephb */ 176495c0a3c8Sjosephb static void 176595c0a3c8Sjosephb flush_all_tlb_entries(void) 176695c0a3c8Sjosephb { 176795c0a3c8Sjosephb ulong_t cr4 = getcr4(); 176895c0a3c8Sjosephb 176995c0a3c8Sjosephb if (cr4 & CR4_PGE) { 177095c0a3c8Sjosephb setcr4(cr4 & ~(ulong_t)CR4_PGE); 177195c0a3c8Sjosephb setcr4(cr4); 177295c0a3c8Sjosephb 177395c0a3c8Sjosephb /* 177495c0a3c8Sjosephb * 32 bit PAE also needs to always reload_cr3() 177595c0a3c8Sjosephb */ 177695c0a3c8Sjosephb if (mmu.max_level == 2) 177795c0a3c8Sjosephb reload_cr3(); 177895c0a3c8Sjosephb } else { 177995c0a3c8Sjosephb reload_cr3(); 178095c0a3c8Sjosephb } 178195c0a3c8Sjosephb } 178295c0a3c8Sjosephb 178395c0a3c8Sjosephb #define TLB_CPU_HALTED (01ul) 178495c0a3c8Sjosephb #define TLB_INVAL_ALL (02ul) 178595c0a3c8Sjosephb #define CAS_TLB_INFO(cpu, old, new) \ 178695c0a3c8Sjosephb caslong((ulong_t *)&(cpu)->cpu_m.mcpu_tlb_info, (old), (new)) 178795c0a3c8Sjosephb 178895c0a3c8Sjosephb /* 178995c0a3c8Sjosephb * Record that a CPU is going idle 179095c0a3c8Sjosephb */ 179195c0a3c8Sjosephb void 179295c0a3c8Sjosephb tlb_going_idle(void) 179395c0a3c8Sjosephb { 179495c0a3c8Sjosephb atomic_or_long((ulong_t *)&CPU->cpu_m.mcpu_tlb_info, TLB_CPU_HALTED); 179595c0a3c8Sjosephb } 179695c0a3c8Sjosephb 179795c0a3c8Sjosephb /* 179895c0a3c8Sjosephb * Service a delayed TLB flush if coming out of being idle. 179995c0a3c8Sjosephb */ 180095c0a3c8Sjosephb void 180195c0a3c8Sjosephb tlb_service(void) 180295c0a3c8Sjosephb { 180395c0a3c8Sjosephb ulong_t flags = getflags(); 180495c0a3c8Sjosephb ulong_t tlb_info; 180595c0a3c8Sjosephb ulong_t found; 180695c0a3c8Sjosephb 180795c0a3c8Sjosephb /* 180895c0a3c8Sjosephb * Be sure interrupts are off while doing this so that 180995c0a3c8Sjosephb * higher level interrupts correctly wait for flushes to finish. 181095c0a3c8Sjosephb */ 181195c0a3c8Sjosephb if (flags & PS_IE) 181295c0a3c8Sjosephb flags = intr_clear(); 181395c0a3c8Sjosephb 181495c0a3c8Sjosephb /* 181595c0a3c8Sjosephb * We only have to do something if coming out of being idle. 181695c0a3c8Sjosephb */ 181795c0a3c8Sjosephb tlb_info = CPU->cpu_m.mcpu_tlb_info; 181895c0a3c8Sjosephb if (tlb_info & TLB_CPU_HALTED) { 181995c0a3c8Sjosephb ASSERT(CPU->cpu_current_hat == kas.a_hat); 182095c0a3c8Sjosephb 182195c0a3c8Sjosephb /* 182295c0a3c8Sjosephb * Atomic clear and fetch of old state. 182395c0a3c8Sjosephb */ 182495c0a3c8Sjosephb while ((found = CAS_TLB_INFO(CPU, tlb_info, 0)) != tlb_info) { 182595c0a3c8Sjosephb ASSERT(found & TLB_CPU_HALTED); 182695c0a3c8Sjosephb tlb_info = found; 182795c0a3c8Sjosephb SMT_PAUSE(); 182895c0a3c8Sjosephb } 182995c0a3c8Sjosephb if (tlb_info & TLB_INVAL_ALL) 183095c0a3c8Sjosephb flush_all_tlb_entries(); 183195c0a3c8Sjosephb } 183295c0a3c8Sjosephb 183395c0a3c8Sjosephb /* 183495c0a3c8Sjosephb * Restore interrupt enable control bit. 183595c0a3c8Sjosephb */ 183695c0a3c8Sjosephb if (flags & PS_IE) 183795c0a3c8Sjosephb sti(); 183895c0a3c8Sjosephb } 183995c0a3c8Sjosephb 18407c478bd9Sstevel@tonic-gate /* 18417c478bd9Sstevel@tonic-gate * Internal routine to do cross calls to invalidate a range of pages on 18427c478bd9Sstevel@tonic-gate * all CPUs using a given hat. 18437c478bd9Sstevel@tonic-gate */ 18447c478bd9Sstevel@tonic-gate void 1845ae115bc7Smrj hat_tlb_inval(hat_t *hat, uintptr_t va) 18467c478bd9Sstevel@tonic-gate { 18477c478bd9Sstevel@tonic-gate extern int flushes_require_xcalls; /* from mp_startup.c */ 18487c478bd9Sstevel@tonic-gate cpuset_t justme; 184995c0a3c8Sjosephb cpuset_t check_cpus; 1850ae115bc7Smrj cpuset_t cpus_to_shootdown; 185195c0a3c8Sjosephb cpu_t *cpup; 185295c0a3c8Sjosephb int c; 18537c478bd9Sstevel@tonic-gate 18547c478bd9Sstevel@tonic-gate /* 18557c478bd9Sstevel@tonic-gate * If the hat is being destroyed, there are no more users, so 18567c478bd9Sstevel@tonic-gate * demap need not do anything. 18577c478bd9Sstevel@tonic-gate */ 18587c478bd9Sstevel@tonic-gate if (hat->hat_flags & HAT_FREEING) 18597c478bd9Sstevel@tonic-gate return; 18607c478bd9Sstevel@tonic-gate 18617c478bd9Sstevel@tonic-gate /* 18627c478bd9Sstevel@tonic-gate * If demapping from a shared pagetable, we best demap the 18637c478bd9Sstevel@tonic-gate * entire set of user TLBs, since we don't know what addresses 18647c478bd9Sstevel@tonic-gate * these were shared at. 18657c478bd9Sstevel@tonic-gate */ 18667c478bd9Sstevel@tonic-gate if (hat->hat_flags & HAT_SHARED) { 18677c478bd9Sstevel@tonic-gate hat = kas.a_hat; 18687c478bd9Sstevel@tonic-gate va = DEMAP_ALL_ADDR; 18697c478bd9Sstevel@tonic-gate } 18707c478bd9Sstevel@tonic-gate 18717c478bd9Sstevel@tonic-gate /* 18727c478bd9Sstevel@tonic-gate * if not running with multiple CPUs, don't use cross calls 18737c478bd9Sstevel@tonic-gate */ 18747c478bd9Sstevel@tonic-gate if (panicstr || !flushes_require_xcalls) { 18757c478bd9Sstevel@tonic-gate (void) hati_demap_func((xc_arg_t)hat, (xc_arg_t)va, NULL); 18767c478bd9Sstevel@tonic-gate return; 18777c478bd9Sstevel@tonic-gate } 18787c478bd9Sstevel@tonic-gate 18797c478bd9Sstevel@tonic-gate 18807c478bd9Sstevel@tonic-gate /* 1881ae115bc7Smrj * Determine CPUs to shootdown. Kernel changes always do all CPUs. 1882ae115bc7Smrj * Otherwise it's just CPUs currently executing in this hat. 18837c478bd9Sstevel@tonic-gate */ 18847c478bd9Sstevel@tonic-gate kpreempt_disable(); 18857c478bd9Sstevel@tonic-gate CPUSET_ONLY(justme, CPU->cpu_id); 1886ae115bc7Smrj if (hat == kas.a_hat) 1887ae115bc7Smrj cpus_to_shootdown = khat_cpuset; 18887c478bd9Sstevel@tonic-gate else 1889ae115bc7Smrj cpus_to_shootdown = hat->hat_cpus; 1890ae115bc7Smrj 189195c0a3c8Sjosephb /* 189295c0a3c8Sjosephb * If any CPUs in the set are idle, just request a delayed flush 189395c0a3c8Sjosephb * and avoid waking them up. 189495c0a3c8Sjosephb */ 189595c0a3c8Sjosephb check_cpus = cpus_to_shootdown; 189695c0a3c8Sjosephb for (c = 0; c < NCPU && !CPUSET_ISNULL(check_cpus); ++c) { 189795c0a3c8Sjosephb ulong_t tlb_info; 189895c0a3c8Sjosephb 189995c0a3c8Sjosephb if (!CPU_IN_SET(check_cpus, c)) 190095c0a3c8Sjosephb continue; 190195c0a3c8Sjosephb CPUSET_DEL(check_cpus, c); 190295c0a3c8Sjosephb cpup = cpu[c]; 190395c0a3c8Sjosephb if (cpup == NULL) 190495c0a3c8Sjosephb continue; 190595c0a3c8Sjosephb 190695c0a3c8Sjosephb tlb_info = cpup->cpu_m.mcpu_tlb_info; 190795c0a3c8Sjosephb while (tlb_info == TLB_CPU_HALTED) { 190895c0a3c8Sjosephb (void) CAS_TLB_INFO(cpup, TLB_CPU_HALTED, 1909*250b7ff9Sjosephb TLB_CPU_HALTED | TLB_INVAL_ALL); 191095c0a3c8Sjosephb SMT_PAUSE(); 191195c0a3c8Sjosephb tlb_info = cpup->cpu_m.mcpu_tlb_info; 191295c0a3c8Sjosephb } 191395c0a3c8Sjosephb if (tlb_info == (TLB_CPU_HALTED | TLB_INVAL_ALL)) { 191495c0a3c8Sjosephb HATSTAT_INC(hs_tlb_inval_delayed); 191595c0a3c8Sjosephb CPUSET_DEL(cpus_to_shootdown, c); 191695c0a3c8Sjosephb } 191795c0a3c8Sjosephb } 191895c0a3c8Sjosephb 1919ae115bc7Smrj if (CPUSET_ISNULL(cpus_to_shootdown) || 1920ae115bc7Smrj CPUSET_ISEQUAL(cpus_to_shootdown, justme)) { 1921ae115bc7Smrj 1922ae115bc7Smrj (void) hati_demap_func((xc_arg_t)hat, (xc_arg_t)va, NULL); 1923ae115bc7Smrj 1924ae115bc7Smrj } else { 1925ae115bc7Smrj 1926ae115bc7Smrj CPUSET_ADD(cpus_to_shootdown, CPU->cpu_id); 1927ae115bc7Smrj xc_call((xc_arg_t)hat, (xc_arg_t)va, NULL, X_CALL_HIPRI, 1928ae115bc7Smrj cpus_to_shootdown, hati_demap_func); 1929ae115bc7Smrj 1930ae115bc7Smrj } 19317c478bd9Sstevel@tonic-gate kpreempt_enable(); 19327c478bd9Sstevel@tonic-gate } 19337c478bd9Sstevel@tonic-gate 19347c478bd9Sstevel@tonic-gate /* 19357c478bd9Sstevel@tonic-gate * Interior routine for HAT_UNLOADs from hat_unload_callback(), 19367c478bd9Sstevel@tonic-gate * hat_kmap_unload() OR from hat_steal() code. This routine doesn't 19377c478bd9Sstevel@tonic-gate * handle releasing of the htables. 19387c478bd9Sstevel@tonic-gate */ 19397c478bd9Sstevel@tonic-gate void 19407c478bd9Sstevel@tonic-gate hat_pte_unmap( 19417c478bd9Sstevel@tonic-gate htable_t *ht, 19427c478bd9Sstevel@tonic-gate uint_t entry, 19437c478bd9Sstevel@tonic-gate uint_t flags, 19447c478bd9Sstevel@tonic-gate x86pte_t old_pte, 19457c478bd9Sstevel@tonic-gate void *pte_ptr) 19467c478bd9Sstevel@tonic-gate { 19477c478bd9Sstevel@tonic-gate hat_t *hat = ht->ht_hat; 19487c478bd9Sstevel@tonic-gate hment_t *hm = NULL; 19497c478bd9Sstevel@tonic-gate page_t *pp = NULL; 19507c478bd9Sstevel@tonic-gate level_t l = ht->ht_level; 19517c478bd9Sstevel@tonic-gate pfn_t pfn; 19527c478bd9Sstevel@tonic-gate 19537c478bd9Sstevel@tonic-gate /* 19547c478bd9Sstevel@tonic-gate * We always track the locking counts, even if nothing is unmapped 19557c478bd9Sstevel@tonic-gate */ 19567c478bd9Sstevel@tonic-gate if ((flags & HAT_UNLOAD_UNLOCK) != 0 && hat != kas.a_hat) { 19577c478bd9Sstevel@tonic-gate ASSERT(ht->ht_lock_cnt > 0); 19587c478bd9Sstevel@tonic-gate HTABLE_LOCK_DEC(ht); 19597c478bd9Sstevel@tonic-gate } 19607c478bd9Sstevel@tonic-gate 19617c478bd9Sstevel@tonic-gate /* 19627c478bd9Sstevel@tonic-gate * Figure out which page's mapping list lock to acquire using the PFN 19637c478bd9Sstevel@tonic-gate * passed in "old" PTE. We then attempt to invalidate the PTE. 19647c478bd9Sstevel@tonic-gate * If another thread, probably a hat_pageunload, has asynchronously 19657c478bd9Sstevel@tonic-gate * unmapped/remapped this address we'll loop here. 19667c478bd9Sstevel@tonic-gate */ 19677c478bd9Sstevel@tonic-gate ASSERT(ht->ht_busy > 0); 19687c478bd9Sstevel@tonic-gate while (PTE_ISVALID(old_pte)) { 19697c478bd9Sstevel@tonic-gate pfn = PTE2PFN(old_pte, l); 1970ae115bc7Smrj if (PTE_GET(old_pte, PT_SOFTWARE) >= PT_NOCONSIST) { 19717c478bd9Sstevel@tonic-gate pp = NULL; 19727c478bd9Sstevel@tonic-gate } else { 19737c478bd9Sstevel@tonic-gate pp = page_numtopp_nolock(pfn); 1974aa2ed9e5Sjosephb if (pp == NULL) { 1975aa2ed9e5Sjosephb panic("no page_t, not NOCONSIST: old_pte=" 1976aa2ed9e5Sjosephb FMT_PTE " ht=%lx entry=0x%x pte_ptr=%lx", 1977aa2ed9e5Sjosephb old_pte, (uintptr_t)ht, entry, 1978aa2ed9e5Sjosephb (uintptr_t)pte_ptr); 1979aa2ed9e5Sjosephb } 19807c478bd9Sstevel@tonic-gate x86_hm_enter(pp); 19817c478bd9Sstevel@tonic-gate } 1982aa2ed9e5Sjosephb 1983aa2ed9e5Sjosephb /* 1984aa2ed9e5Sjosephb * If freeing the address space, check that the PTE 1985aa2ed9e5Sjosephb * hasn't changed, as the mappings are no longer in use by 1986aa2ed9e5Sjosephb * any thread, invalidation is unnecessary. 1987aa2ed9e5Sjosephb * If not freeing, do a full invalidate. 1988aa2ed9e5Sjosephb */ 1989aa2ed9e5Sjosephb if (hat->hat_flags & HAT_FREEING) 1990aa2ed9e5Sjosephb old_pte = x86pte_get(ht, entry); 1991aa2ed9e5Sjosephb else 1992ae115bc7Smrj old_pte = x86pte_inval(ht, entry, old_pte, pte_ptr); 19937c478bd9Sstevel@tonic-gate 19947c478bd9Sstevel@tonic-gate /* 19957c478bd9Sstevel@tonic-gate * If the page hadn't changed we've unmapped it and can proceed 19967c478bd9Sstevel@tonic-gate */ 19977c478bd9Sstevel@tonic-gate if (PTE_ISVALID(old_pte) && PTE2PFN(old_pte, l) == pfn) 19987c478bd9Sstevel@tonic-gate break; 19997c478bd9Sstevel@tonic-gate 20007c478bd9Sstevel@tonic-gate /* 20017c478bd9Sstevel@tonic-gate * Otherwise, we'll have to retry with the current old_pte. 20027c478bd9Sstevel@tonic-gate * Drop the hment lock, since the pfn may have changed. 20037c478bd9Sstevel@tonic-gate */ 20047c478bd9Sstevel@tonic-gate if (pp != NULL) { 20057c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 20067c478bd9Sstevel@tonic-gate pp = NULL; 20077c478bd9Sstevel@tonic-gate } else { 2008ae115bc7Smrj ASSERT(PTE_GET(old_pte, PT_SOFTWARE) >= PT_NOCONSIST); 20097c478bd9Sstevel@tonic-gate } 20107c478bd9Sstevel@tonic-gate } 20117c478bd9Sstevel@tonic-gate 20127c478bd9Sstevel@tonic-gate /* 20137c478bd9Sstevel@tonic-gate * If the old mapping wasn't valid, there's nothing more to do 20147c478bd9Sstevel@tonic-gate */ 20157c478bd9Sstevel@tonic-gate if (!PTE_ISVALID(old_pte)) { 20167c478bd9Sstevel@tonic-gate if (pp != NULL) 20177c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 20187c478bd9Sstevel@tonic-gate return; 20197c478bd9Sstevel@tonic-gate } 20207c478bd9Sstevel@tonic-gate 20217c478bd9Sstevel@tonic-gate /* 20227c478bd9Sstevel@tonic-gate * Take care of syncing any MOD/REF bits and removing the hment. 20237c478bd9Sstevel@tonic-gate */ 20247c478bd9Sstevel@tonic-gate if (pp != NULL) { 20257c478bd9Sstevel@tonic-gate if (!(flags & HAT_UNLOAD_NOSYNC)) 20267c478bd9Sstevel@tonic-gate hati_sync_pte_to_page(pp, old_pte, l); 20277c478bd9Sstevel@tonic-gate hm = hment_remove(pp, ht, entry); 20287c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 20297c478bd9Sstevel@tonic-gate if (hm != NULL) 20307c478bd9Sstevel@tonic-gate hment_free(hm); 20317c478bd9Sstevel@tonic-gate } 20327c478bd9Sstevel@tonic-gate 20337c478bd9Sstevel@tonic-gate /* 20347c478bd9Sstevel@tonic-gate * Handle book keeping in the htable and hat 20357c478bd9Sstevel@tonic-gate */ 20367c478bd9Sstevel@tonic-gate ASSERT(ht->ht_valid_cnt > 0); 20377c478bd9Sstevel@tonic-gate HTABLE_DEC(ht->ht_valid_cnt); 20387c478bd9Sstevel@tonic-gate PGCNT_DEC(hat, l); 20397c478bd9Sstevel@tonic-gate } 20407c478bd9Sstevel@tonic-gate 20417c478bd9Sstevel@tonic-gate /* 20427c478bd9Sstevel@tonic-gate * very cheap unload implementation to special case some kernel addresses 20437c478bd9Sstevel@tonic-gate */ 20447c478bd9Sstevel@tonic-gate static void 20457c478bd9Sstevel@tonic-gate hat_kmap_unload(caddr_t addr, size_t len, uint_t flags) 20467c478bd9Sstevel@tonic-gate { 20477c478bd9Sstevel@tonic-gate uintptr_t va = (uintptr_t)addr; 20487c478bd9Sstevel@tonic-gate uintptr_t eva = va + len; 2049ae115bc7Smrj pgcnt_t pg_index; 20507c478bd9Sstevel@tonic-gate htable_t *ht; 20517c478bd9Sstevel@tonic-gate uint_t entry; 2052ae115bc7Smrj x86pte_t *pte_ptr; 20537c478bd9Sstevel@tonic-gate x86pte_t old_pte; 20547c478bd9Sstevel@tonic-gate 20557c478bd9Sstevel@tonic-gate for (; va < eva; va += MMU_PAGESIZE) { 20567c478bd9Sstevel@tonic-gate /* 20577c478bd9Sstevel@tonic-gate * Get the PTE 20587c478bd9Sstevel@tonic-gate */ 2059ae115bc7Smrj pg_index = mmu_btop(va - mmu.kmap_addr); 2060ae115bc7Smrj pte_ptr = PT_INDEX_PTR(mmu.kmap_ptes, pg_index); 2061ae115bc7Smrj old_pte = GET_PTE(pte_ptr); 20627c478bd9Sstevel@tonic-gate 20637c478bd9Sstevel@tonic-gate /* 20647c478bd9Sstevel@tonic-gate * get the htable / entry 20657c478bd9Sstevel@tonic-gate */ 20667c478bd9Sstevel@tonic-gate ht = mmu.kmap_htables[(va - mmu.kmap_htables[0]->ht_vaddr) 20677c478bd9Sstevel@tonic-gate >> LEVEL_SHIFT(1)]; 20687c478bd9Sstevel@tonic-gate entry = htable_va2entry(va, ht); 20697c478bd9Sstevel@tonic-gate 20707c478bd9Sstevel@tonic-gate /* 20717c478bd9Sstevel@tonic-gate * use mostly common code to unmap it. 20727c478bd9Sstevel@tonic-gate */ 20737c478bd9Sstevel@tonic-gate hat_pte_unmap(ht, entry, flags, old_pte, pte_ptr); 20747c478bd9Sstevel@tonic-gate } 20757c478bd9Sstevel@tonic-gate } 20767c478bd9Sstevel@tonic-gate 20777c478bd9Sstevel@tonic-gate 20787c478bd9Sstevel@tonic-gate /* 20797c478bd9Sstevel@tonic-gate * unload a range of virtual address space (no callback) 20807c478bd9Sstevel@tonic-gate */ 20817c478bd9Sstevel@tonic-gate void 20827c478bd9Sstevel@tonic-gate hat_unload(hat_t *hat, caddr_t addr, size_t len, uint_t flags) 20837c478bd9Sstevel@tonic-gate { 20847c478bd9Sstevel@tonic-gate uintptr_t va = (uintptr_t)addr; 2085ae115bc7Smrj 2086ae115bc7Smrj ASSERT(hat == kas.a_hat || va + len <= _userlimit); 20877c478bd9Sstevel@tonic-gate 20887c478bd9Sstevel@tonic-gate /* 20897c478bd9Sstevel@tonic-gate * special case for performance. 20907c478bd9Sstevel@tonic-gate */ 20917c478bd9Sstevel@tonic-gate if (mmu.kmap_addr <= va && va < mmu.kmap_eaddr) { 20927c478bd9Sstevel@tonic-gate ASSERT(hat == kas.a_hat); 20937c478bd9Sstevel@tonic-gate hat_kmap_unload(addr, len, flags); 2094ae115bc7Smrj } else { 2095ae115bc7Smrj hat_unload_callback(hat, addr, len, flags, NULL); 20967c478bd9Sstevel@tonic-gate } 20977c478bd9Sstevel@tonic-gate } 20987c478bd9Sstevel@tonic-gate 20997c478bd9Sstevel@tonic-gate /* 21007c478bd9Sstevel@tonic-gate * Do the callbacks for ranges being unloaded. 21017c478bd9Sstevel@tonic-gate */ 21027c478bd9Sstevel@tonic-gate typedef struct range_info { 21037c478bd9Sstevel@tonic-gate uintptr_t rng_va; 21047c478bd9Sstevel@tonic-gate ulong_t rng_cnt; 21057c478bd9Sstevel@tonic-gate level_t rng_level; 21067c478bd9Sstevel@tonic-gate } range_info_t; 21077c478bd9Sstevel@tonic-gate 21087c478bd9Sstevel@tonic-gate static void 21097c478bd9Sstevel@tonic-gate handle_ranges(hat_callback_t *cb, uint_t cnt, range_info_t *range) 21107c478bd9Sstevel@tonic-gate { 21117c478bd9Sstevel@tonic-gate /* 21127c478bd9Sstevel@tonic-gate * do callbacks to upper level VM system 21137c478bd9Sstevel@tonic-gate */ 21147c478bd9Sstevel@tonic-gate while (cb != NULL && cnt > 0) { 21157c478bd9Sstevel@tonic-gate --cnt; 21167c478bd9Sstevel@tonic-gate cb->hcb_start_addr = (caddr_t)range[cnt].rng_va; 21177c478bd9Sstevel@tonic-gate cb->hcb_end_addr = cb->hcb_start_addr; 21187c478bd9Sstevel@tonic-gate cb->hcb_end_addr += 21197c478bd9Sstevel@tonic-gate range[cnt].rng_cnt << LEVEL_SIZE(range[cnt].rng_level); 21207c478bd9Sstevel@tonic-gate cb->hcb_function(cb); 21217c478bd9Sstevel@tonic-gate } 21227c478bd9Sstevel@tonic-gate } 21237c478bd9Sstevel@tonic-gate 21247c478bd9Sstevel@tonic-gate /* 21257c478bd9Sstevel@tonic-gate * Unload a given range of addresses (has optional callback) 21267c478bd9Sstevel@tonic-gate * 21277c478bd9Sstevel@tonic-gate * Flags: 21287c478bd9Sstevel@tonic-gate * define HAT_UNLOAD 0x00 21297c478bd9Sstevel@tonic-gate * define HAT_UNLOAD_NOSYNC 0x02 21307c478bd9Sstevel@tonic-gate * define HAT_UNLOAD_UNLOCK 0x04 21317c478bd9Sstevel@tonic-gate * define HAT_UNLOAD_OTHER 0x08 - not used 21327c478bd9Sstevel@tonic-gate * define HAT_UNLOAD_UNMAP 0x10 - same as HAT_UNLOAD 21337c478bd9Sstevel@tonic-gate */ 21347c478bd9Sstevel@tonic-gate #define MAX_UNLOAD_CNT (8) 21357c478bd9Sstevel@tonic-gate void 21367c478bd9Sstevel@tonic-gate hat_unload_callback( 21377c478bd9Sstevel@tonic-gate hat_t *hat, 21387c478bd9Sstevel@tonic-gate caddr_t addr, 21397c478bd9Sstevel@tonic-gate size_t len, 21407c478bd9Sstevel@tonic-gate uint_t flags, 21417c478bd9Sstevel@tonic-gate hat_callback_t *cb) 21427c478bd9Sstevel@tonic-gate { 21437c478bd9Sstevel@tonic-gate uintptr_t vaddr = (uintptr_t)addr; 21447c478bd9Sstevel@tonic-gate uintptr_t eaddr = vaddr + len; 21457c478bd9Sstevel@tonic-gate htable_t *ht = NULL; 21467c478bd9Sstevel@tonic-gate uint_t entry; 2147aa2ed9e5Sjosephb uintptr_t contig_va = (uintptr_t)-1L; 21487c478bd9Sstevel@tonic-gate range_info_t r[MAX_UNLOAD_CNT]; 21497c478bd9Sstevel@tonic-gate uint_t r_cnt = 0; 21507c478bd9Sstevel@tonic-gate x86pte_t old_pte; 21517c478bd9Sstevel@tonic-gate 2152ae115bc7Smrj ASSERT(hat == kas.a_hat || eaddr <= _userlimit); 21537c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(vaddr)); 21547c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(eaddr)); 21557c478bd9Sstevel@tonic-gate 2156ae115bc7Smrj /* 2157ae115bc7Smrj * Special case a single page being unloaded for speed. This happens 2158ae115bc7Smrj * quite frequently, COW faults after a fork() for example. 2159ae115bc7Smrj */ 2160ae115bc7Smrj if (cb == NULL && len == MMU_PAGESIZE) { 2161ae115bc7Smrj ht = htable_getpte(hat, vaddr, &entry, &old_pte, 0); 2162ae115bc7Smrj if (ht != NULL) { 2163ae115bc7Smrj if (PTE_ISVALID(old_pte)) 2164ae115bc7Smrj hat_pte_unmap(ht, entry, flags, old_pte, NULL); 2165ae115bc7Smrj htable_release(ht); 2166ae115bc7Smrj } 2167ae115bc7Smrj return; 2168ae115bc7Smrj } 2169ae115bc7Smrj 21707c478bd9Sstevel@tonic-gate while (vaddr < eaddr) { 21717c478bd9Sstevel@tonic-gate old_pte = htable_walk(hat, &ht, &vaddr, eaddr); 21727c478bd9Sstevel@tonic-gate if (ht == NULL) 21737c478bd9Sstevel@tonic-gate break; 21747c478bd9Sstevel@tonic-gate 21757c478bd9Sstevel@tonic-gate ASSERT(!IN_VA_HOLE(vaddr)); 21767c478bd9Sstevel@tonic-gate 21777c478bd9Sstevel@tonic-gate if (vaddr < (uintptr_t)addr) 21787c478bd9Sstevel@tonic-gate panic("hat_unload_callback(): unmap inside large page"); 21797c478bd9Sstevel@tonic-gate 21807c478bd9Sstevel@tonic-gate /* 21817c478bd9Sstevel@tonic-gate * We'll do the call backs for contiguous ranges 21827c478bd9Sstevel@tonic-gate */ 2183aa2ed9e5Sjosephb if (vaddr != contig_va || 21847c478bd9Sstevel@tonic-gate (r_cnt > 0 && r[r_cnt - 1].rng_level != ht->ht_level)) { 21857c478bd9Sstevel@tonic-gate if (r_cnt == MAX_UNLOAD_CNT) { 21867c478bd9Sstevel@tonic-gate handle_ranges(cb, r_cnt, r); 21877c478bd9Sstevel@tonic-gate r_cnt = 0; 21887c478bd9Sstevel@tonic-gate } 21897c478bd9Sstevel@tonic-gate r[r_cnt].rng_va = vaddr; 21907c478bd9Sstevel@tonic-gate r[r_cnt].rng_cnt = 0; 21917c478bd9Sstevel@tonic-gate r[r_cnt].rng_level = ht->ht_level; 21927c478bd9Sstevel@tonic-gate ++r_cnt; 21937c478bd9Sstevel@tonic-gate } 21947c478bd9Sstevel@tonic-gate 21957c478bd9Sstevel@tonic-gate /* 21967c478bd9Sstevel@tonic-gate * Unload one mapping from the page tables. 21977c478bd9Sstevel@tonic-gate */ 21987c478bd9Sstevel@tonic-gate entry = htable_va2entry(vaddr, ht); 21997c478bd9Sstevel@tonic-gate hat_pte_unmap(ht, entry, flags, old_pte, NULL); 22007c478bd9Sstevel@tonic-gate ASSERT(ht->ht_level <= mmu.max_page_level); 22017c478bd9Sstevel@tonic-gate vaddr += LEVEL_SIZE(ht->ht_level); 2202aa2ed9e5Sjosephb contig_va = vaddr; 22037c478bd9Sstevel@tonic-gate ++r[r_cnt - 1].rng_cnt; 22047c478bd9Sstevel@tonic-gate } 22057c478bd9Sstevel@tonic-gate if (ht) 22067c478bd9Sstevel@tonic-gate htable_release(ht); 22077c478bd9Sstevel@tonic-gate 22087c478bd9Sstevel@tonic-gate /* 22097c478bd9Sstevel@tonic-gate * handle last range for callbacks 22107c478bd9Sstevel@tonic-gate */ 22117c478bd9Sstevel@tonic-gate if (r_cnt > 0) 22127c478bd9Sstevel@tonic-gate handle_ranges(cb, r_cnt, r); 22137c478bd9Sstevel@tonic-gate } 22147c478bd9Sstevel@tonic-gate 22157c478bd9Sstevel@tonic-gate /* 22167c478bd9Sstevel@tonic-gate * synchronize mapping with software data structures 22177c478bd9Sstevel@tonic-gate * 22187c478bd9Sstevel@tonic-gate * This interface is currently only used by the working set monitor 22197c478bd9Sstevel@tonic-gate * driver. 22207c478bd9Sstevel@tonic-gate */ 22217c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 22227c478bd9Sstevel@tonic-gate void 22237c478bd9Sstevel@tonic-gate hat_sync(hat_t *hat, caddr_t addr, size_t len, uint_t flags) 22247c478bd9Sstevel@tonic-gate { 22257c478bd9Sstevel@tonic-gate uintptr_t vaddr = (uintptr_t)addr; 22267c478bd9Sstevel@tonic-gate uintptr_t eaddr = vaddr + len; 22277c478bd9Sstevel@tonic-gate htable_t *ht = NULL; 22287c478bd9Sstevel@tonic-gate uint_t entry; 22297c478bd9Sstevel@tonic-gate x86pte_t pte; 22307c478bd9Sstevel@tonic-gate x86pte_t save_pte; 22317c478bd9Sstevel@tonic-gate x86pte_t new; 22327c478bd9Sstevel@tonic-gate page_t *pp; 22337c478bd9Sstevel@tonic-gate 22347c478bd9Sstevel@tonic-gate ASSERT(!IN_VA_HOLE(vaddr)); 22357c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(vaddr)); 22367c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(eaddr)); 2237ae115bc7Smrj ASSERT(hat == kas.a_hat || eaddr <= _userlimit); 22387c478bd9Sstevel@tonic-gate 22397c478bd9Sstevel@tonic-gate for (; vaddr < eaddr; vaddr += LEVEL_SIZE(ht->ht_level)) { 22407c478bd9Sstevel@tonic-gate try_again: 22417c478bd9Sstevel@tonic-gate pte = htable_walk(hat, &ht, &vaddr, eaddr); 22427c478bd9Sstevel@tonic-gate if (ht == NULL) 22437c478bd9Sstevel@tonic-gate break; 22447c478bd9Sstevel@tonic-gate entry = htable_va2entry(vaddr, ht); 22457c478bd9Sstevel@tonic-gate 2246ae115bc7Smrj if (PTE_GET(pte, PT_SOFTWARE) >= PT_NOSYNC || 22477c478bd9Sstevel@tonic-gate PTE_GET(pte, PT_REF | PT_MOD) == 0) 22487c478bd9Sstevel@tonic-gate continue; 22497c478bd9Sstevel@tonic-gate 22507c478bd9Sstevel@tonic-gate /* 22517c478bd9Sstevel@tonic-gate * We need to acquire the mapping list lock to protect 22527c478bd9Sstevel@tonic-gate * against hat_pageunload(), hat_unload(), etc. 22537c478bd9Sstevel@tonic-gate */ 22547c478bd9Sstevel@tonic-gate pp = page_numtopp_nolock(PTE2PFN(pte, ht->ht_level)); 22557c478bd9Sstevel@tonic-gate if (pp == NULL) 22567c478bd9Sstevel@tonic-gate break; 22577c478bd9Sstevel@tonic-gate x86_hm_enter(pp); 22587c478bd9Sstevel@tonic-gate save_pte = pte; 22597c478bd9Sstevel@tonic-gate pte = x86pte_get(ht, entry); 22607c478bd9Sstevel@tonic-gate if (pte != save_pte) { 22617c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 22627c478bd9Sstevel@tonic-gate goto try_again; 22637c478bd9Sstevel@tonic-gate } 2264ae115bc7Smrj if (PTE_GET(pte, PT_SOFTWARE) >= PT_NOSYNC || 22657c478bd9Sstevel@tonic-gate PTE_GET(pte, PT_REF | PT_MOD) == 0) { 22667c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 22677c478bd9Sstevel@tonic-gate continue; 22687c478bd9Sstevel@tonic-gate } 22697c478bd9Sstevel@tonic-gate 22707c478bd9Sstevel@tonic-gate /* 22717c478bd9Sstevel@tonic-gate * Need to clear ref or mod bits. We may compete with 22727c478bd9Sstevel@tonic-gate * hardware updating the R/M bits and have to try again. 22737c478bd9Sstevel@tonic-gate */ 22747c478bd9Sstevel@tonic-gate if (flags == HAT_SYNC_ZERORM) { 22757c478bd9Sstevel@tonic-gate new = pte; 22767c478bd9Sstevel@tonic-gate PTE_CLR(new, PT_REF | PT_MOD); 22777c478bd9Sstevel@tonic-gate pte = hati_update_pte(ht, entry, pte, new); 22787c478bd9Sstevel@tonic-gate if (pte != 0) { 22797c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 22807c478bd9Sstevel@tonic-gate goto try_again; 22817c478bd9Sstevel@tonic-gate } 22827c478bd9Sstevel@tonic-gate } else { 22837c478bd9Sstevel@tonic-gate /* 22847c478bd9Sstevel@tonic-gate * sync the PTE to the page_t 22857c478bd9Sstevel@tonic-gate */ 22867c478bd9Sstevel@tonic-gate hati_sync_pte_to_page(pp, save_pte, ht->ht_level); 22877c478bd9Sstevel@tonic-gate } 22887c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 22897c478bd9Sstevel@tonic-gate } 22907c478bd9Sstevel@tonic-gate if (ht) 22917c478bd9Sstevel@tonic-gate htable_release(ht); 22927c478bd9Sstevel@tonic-gate } 22937c478bd9Sstevel@tonic-gate 22947c478bd9Sstevel@tonic-gate /* 22957c478bd9Sstevel@tonic-gate * void hat_map(hat, addr, len, flags) 22967c478bd9Sstevel@tonic-gate */ 22977c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 22987c478bd9Sstevel@tonic-gate void 22997c478bd9Sstevel@tonic-gate hat_map(hat_t *hat, caddr_t addr, size_t len, uint_t flags) 23007c478bd9Sstevel@tonic-gate { 23017c478bd9Sstevel@tonic-gate /* does nothing */ 23027c478bd9Sstevel@tonic-gate } 23037c478bd9Sstevel@tonic-gate 23047c478bd9Sstevel@tonic-gate /* 23057c478bd9Sstevel@tonic-gate * uint_t hat_getattr(hat, addr, *attr) 23067c478bd9Sstevel@tonic-gate * returns attr for <hat,addr> in *attr. returns 0 if there was a 23077c478bd9Sstevel@tonic-gate * mapping and *attr is valid, nonzero if there was no mapping and 23087c478bd9Sstevel@tonic-gate * *attr is not valid. 23097c478bd9Sstevel@tonic-gate */ 23107c478bd9Sstevel@tonic-gate uint_t 23117c478bd9Sstevel@tonic-gate hat_getattr(hat_t *hat, caddr_t addr, uint_t *attr) 23127c478bd9Sstevel@tonic-gate { 23137c478bd9Sstevel@tonic-gate uintptr_t vaddr = ALIGN2PAGE(addr); 23147c478bd9Sstevel@tonic-gate htable_t *ht = NULL; 23157c478bd9Sstevel@tonic-gate x86pte_t pte; 23167c478bd9Sstevel@tonic-gate 2317ae115bc7Smrj ASSERT(hat == kas.a_hat || vaddr <= _userlimit); 23187c478bd9Sstevel@tonic-gate 23197c478bd9Sstevel@tonic-gate if (IN_VA_HOLE(vaddr)) 23207c478bd9Sstevel@tonic-gate return ((uint_t)-1); 23217c478bd9Sstevel@tonic-gate 2322ae115bc7Smrj ht = htable_getpte(hat, vaddr, NULL, &pte, mmu.max_page_level); 23237c478bd9Sstevel@tonic-gate if (ht == NULL) 23247c478bd9Sstevel@tonic-gate return ((uint_t)-1); 23257c478bd9Sstevel@tonic-gate 23267c478bd9Sstevel@tonic-gate if (!PTE_ISVALID(pte) || !PTE_ISPAGE(pte, ht->ht_level)) { 23277c478bd9Sstevel@tonic-gate htable_release(ht); 23287c478bd9Sstevel@tonic-gate return ((uint_t)-1); 23297c478bd9Sstevel@tonic-gate } 23307c478bd9Sstevel@tonic-gate 23317c478bd9Sstevel@tonic-gate *attr = PROT_READ; 23327c478bd9Sstevel@tonic-gate if (PTE_GET(pte, PT_WRITABLE)) 23337c478bd9Sstevel@tonic-gate *attr |= PROT_WRITE; 23347c478bd9Sstevel@tonic-gate if (PTE_GET(pte, PT_USER)) 23357c478bd9Sstevel@tonic-gate *attr |= PROT_USER; 23367c478bd9Sstevel@tonic-gate if (!PTE_GET(pte, mmu.pt_nx)) 23377c478bd9Sstevel@tonic-gate *attr |= PROT_EXEC; 2338ae115bc7Smrj if (PTE_GET(pte, PT_SOFTWARE) >= PT_NOSYNC) 23397c478bd9Sstevel@tonic-gate *attr |= HAT_NOSYNC; 23407c478bd9Sstevel@tonic-gate htable_release(ht); 23417c478bd9Sstevel@tonic-gate return (0); 23427c478bd9Sstevel@tonic-gate } 23437c478bd9Sstevel@tonic-gate 23447c478bd9Sstevel@tonic-gate /* 23457c478bd9Sstevel@tonic-gate * hat_updateattr() applies the given attribute change to an existing mapping 23467c478bd9Sstevel@tonic-gate */ 23477c478bd9Sstevel@tonic-gate #define HAT_LOAD_ATTR 1 23487c478bd9Sstevel@tonic-gate #define HAT_SET_ATTR 2 23497c478bd9Sstevel@tonic-gate #define HAT_CLR_ATTR 3 23507c478bd9Sstevel@tonic-gate 23517c478bd9Sstevel@tonic-gate static void 23527c478bd9Sstevel@tonic-gate hat_updateattr(hat_t *hat, caddr_t addr, size_t len, uint_t attr, int what) 23537c478bd9Sstevel@tonic-gate { 23547c478bd9Sstevel@tonic-gate uintptr_t vaddr = (uintptr_t)addr; 23557c478bd9Sstevel@tonic-gate uintptr_t eaddr = (uintptr_t)addr + len; 23567c478bd9Sstevel@tonic-gate htable_t *ht = NULL; 23577c478bd9Sstevel@tonic-gate uint_t entry; 23587c478bd9Sstevel@tonic-gate x86pte_t oldpte, newpte; 23597c478bd9Sstevel@tonic-gate page_t *pp; 23607c478bd9Sstevel@tonic-gate 23617c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(vaddr)); 23627c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(eaddr)); 23637c478bd9Sstevel@tonic-gate ASSERT(hat == kas.a_hat || 23647c478bd9Sstevel@tonic-gate AS_LOCK_HELD(hat->hat_as, &hat->hat_as->a_lock)); 23657c478bd9Sstevel@tonic-gate for (; vaddr < eaddr; vaddr += LEVEL_SIZE(ht->ht_level)) { 23667c478bd9Sstevel@tonic-gate try_again: 23677c478bd9Sstevel@tonic-gate oldpte = htable_walk(hat, &ht, &vaddr, eaddr); 23687c478bd9Sstevel@tonic-gate if (ht == NULL) 23697c478bd9Sstevel@tonic-gate break; 2370ae115bc7Smrj if (PTE_GET(oldpte, PT_SOFTWARE) >= PT_NOCONSIST) 23717c478bd9Sstevel@tonic-gate continue; 23727c478bd9Sstevel@tonic-gate 23737c478bd9Sstevel@tonic-gate pp = page_numtopp_nolock(PTE2PFN(oldpte, ht->ht_level)); 23747c478bd9Sstevel@tonic-gate if (pp == NULL) 23757c478bd9Sstevel@tonic-gate continue; 23767c478bd9Sstevel@tonic-gate x86_hm_enter(pp); 23777c478bd9Sstevel@tonic-gate 23787c478bd9Sstevel@tonic-gate newpte = oldpte; 23797c478bd9Sstevel@tonic-gate /* 23807c478bd9Sstevel@tonic-gate * We found a page table entry in the desired range, 23817c478bd9Sstevel@tonic-gate * figure out the new attributes. 23827c478bd9Sstevel@tonic-gate */ 23837c478bd9Sstevel@tonic-gate if (what == HAT_SET_ATTR || what == HAT_LOAD_ATTR) { 23847c478bd9Sstevel@tonic-gate if ((attr & PROT_WRITE) && 23857c478bd9Sstevel@tonic-gate !PTE_GET(oldpte, PT_WRITABLE)) 23867c478bd9Sstevel@tonic-gate newpte |= PT_WRITABLE; 23877c478bd9Sstevel@tonic-gate 2388ae115bc7Smrj if ((attr & HAT_NOSYNC) && 2389ae115bc7Smrj PTE_GET(oldpte, PT_SOFTWARE) < PT_NOSYNC) 23907c478bd9Sstevel@tonic-gate newpte |= PT_NOSYNC; 23917c478bd9Sstevel@tonic-gate 23927c478bd9Sstevel@tonic-gate if ((attr & PROT_EXEC) && PTE_GET(oldpte, mmu.pt_nx)) 23937c478bd9Sstevel@tonic-gate newpte &= ~mmu.pt_nx; 23947c478bd9Sstevel@tonic-gate } 23957c478bd9Sstevel@tonic-gate 23967c478bd9Sstevel@tonic-gate if (what == HAT_LOAD_ATTR) { 23977c478bd9Sstevel@tonic-gate if (!(attr & PROT_WRITE) && 23987c478bd9Sstevel@tonic-gate PTE_GET(oldpte, PT_WRITABLE)) 23997c478bd9Sstevel@tonic-gate newpte &= ~PT_WRITABLE; 24007c478bd9Sstevel@tonic-gate 2401ae115bc7Smrj if (!(attr & HAT_NOSYNC) && 2402ae115bc7Smrj PTE_GET(oldpte, PT_SOFTWARE) >= PT_NOSYNC) 2403ae115bc7Smrj newpte &= ~PT_SOFTWARE; 24047c478bd9Sstevel@tonic-gate 24057c478bd9Sstevel@tonic-gate if (!(attr & PROT_EXEC) && !PTE_GET(oldpte, mmu.pt_nx)) 24067c478bd9Sstevel@tonic-gate newpte |= mmu.pt_nx; 24077c478bd9Sstevel@tonic-gate } 24087c478bd9Sstevel@tonic-gate 24097c478bd9Sstevel@tonic-gate if (what == HAT_CLR_ATTR) { 24107c478bd9Sstevel@tonic-gate if ((attr & PROT_WRITE) && PTE_GET(oldpte, PT_WRITABLE)) 24117c478bd9Sstevel@tonic-gate newpte &= ~PT_WRITABLE; 24127c478bd9Sstevel@tonic-gate 2413ae115bc7Smrj if ((attr & HAT_NOSYNC) && 2414ae115bc7Smrj PTE_GET(oldpte, PT_SOFTWARE) >= PT_NOSYNC) 2415ae115bc7Smrj newpte &= ~PT_SOFTWARE; 24167c478bd9Sstevel@tonic-gate 24177c478bd9Sstevel@tonic-gate if ((attr & PROT_EXEC) && !PTE_GET(oldpte, mmu.pt_nx)) 24187c478bd9Sstevel@tonic-gate newpte |= mmu.pt_nx; 24197c478bd9Sstevel@tonic-gate } 24207c478bd9Sstevel@tonic-gate 2421ae115bc7Smrj /* 2422ae115bc7Smrj * Ensure NOSYNC/NOCONSIST mappings have REF and MOD set. 2423ae115bc7Smrj * x86pte_set() depends on this. 2424ae115bc7Smrj */ 2425ae115bc7Smrj if (PTE_GET(newpte, PT_SOFTWARE) >= PT_NOSYNC) 2426ae115bc7Smrj newpte |= PT_REF | PT_MOD; 2427ae115bc7Smrj 24287c478bd9Sstevel@tonic-gate /* 24297c478bd9Sstevel@tonic-gate * what about PROT_READ or others? this code only handles: 24307c478bd9Sstevel@tonic-gate * EXEC, WRITE, NOSYNC 24317c478bd9Sstevel@tonic-gate */ 24327c478bd9Sstevel@tonic-gate 24337c478bd9Sstevel@tonic-gate /* 24347c478bd9Sstevel@tonic-gate * If new PTE really changed, update the table. 24357c478bd9Sstevel@tonic-gate */ 24367c478bd9Sstevel@tonic-gate if (newpte != oldpte) { 24377c478bd9Sstevel@tonic-gate entry = htable_va2entry(vaddr, ht); 24387c478bd9Sstevel@tonic-gate oldpte = hati_update_pte(ht, entry, oldpte, newpte); 24397c478bd9Sstevel@tonic-gate if (oldpte != 0) { 24407c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 24417c478bd9Sstevel@tonic-gate goto try_again; 24427c478bd9Sstevel@tonic-gate } 24437c478bd9Sstevel@tonic-gate } 24447c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 24457c478bd9Sstevel@tonic-gate } 24467c478bd9Sstevel@tonic-gate if (ht) 24477c478bd9Sstevel@tonic-gate htable_release(ht); 24487c478bd9Sstevel@tonic-gate } 24497c478bd9Sstevel@tonic-gate 24507c478bd9Sstevel@tonic-gate /* 24517c478bd9Sstevel@tonic-gate * Various wrappers for hat_updateattr() 24527c478bd9Sstevel@tonic-gate */ 24537c478bd9Sstevel@tonic-gate void 24547c478bd9Sstevel@tonic-gate hat_setattr(hat_t *hat, caddr_t addr, size_t len, uint_t attr) 24557c478bd9Sstevel@tonic-gate { 2456ae115bc7Smrj ASSERT(hat == kas.a_hat || (uintptr_t)addr + len <= _userlimit); 24577c478bd9Sstevel@tonic-gate hat_updateattr(hat, addr, len, attr, HAT_SET_ATTR); 24587c478bd9Sstevel@tonic-gate } 24597c478bd9Sstevel@tonic-gate 24607c478bd9Sstevel@tonic-gate void 24617c478bd9Sstevel@tonic-gate hat_clrattr(hat_t *hat, caddr_t addr, size_t len, uint_t attr) 24627c478bd9Sstevel@tonic-gate { 2463ae115bc7Smrj ASSERT(hat == kas.a_hat || (uintptr_t)addr + len <= _userlimit); 24647c478bd9Sstevel@tonic-gate hat_updateattr(hat, addr, len, attr, HAT_CLR_ATTR); 24657c478bd9Sstevel@tonic-gate } 24667c478bd9Sstevel@tonic-gate 24677c478bd9Sstevel@tonic-gate void 24687c478bd9Sstevel@tonic-gate hat_chgattr(hat_t *hat, caddr_t addr, size_t len, uint_t attr) 24697c478bd9Sstevel@tonic-gate { 2470ae115bc7Smrj ASSERT(hat == kas.a_hat || (uintptr_t)addr + len <= _userlimit); 24717c478bd9Sstevel@tonic-gate hat_updateattr(hat, addr, len, attr, HAT_LOAD_ATTR); 24727c478bd9Sstevel@tonic-gate } 24737c478bd9Sstevel@tonic-gate 24747c478bd9Sstevel@tonic-gate void 24757c478bd9Sstevel@tonic-gate hat_chgprot(hat_t *hat, caddr_t addr, size_t len, uint_t vprot) 24767c478bd9Sstevel@tonic-gate { 2477ae115bc7Smrj ASSERT(hat == kas.a_hat || (uintptr_t)addr + len <= _userlimit); 24787c478bd9Sstevel@tonic-gate hat_updateattr(hat, addr, len, vprot & HAT_PROT_MASK, HAT_LOAD_ATTR); 24797c478bd9Sstevel@tonic-gate } 24807c478bd9Sstevel@tonic-gate 24817c478bd9Sstevel@tonic-gate /* 24827c478bd9Sstevel@tonic-gate * size_t hat_getpagesize(hat, addr) 24837c478bd9Sstevel@tonic-gate * returns pagesize in bytes for <hat, addr>. returns -1 of there is 24847c478bd9Sstevel@tonic-gate * no mapping. This is an advisory call. 24857c478bd9Sstevel@tonic-gate */ 24867c478bd9Sstevel@tonic-gate ssize_t 24877c478bd9Sstevel@tonic-gate hat_getpagesize(hat_t *hat, caddr_t addr) 24887c478bd9Sstevel@tonic-gate { 24897c478bd9Sstevel@tonic-gate uintptr_t vaddr = ALIGN2PAGE(addr); 24907c478bd9Sstevel@tonic-gate htable_t *ht; 24917c478bd9Sstevel@tonic-gate size_t pagesize; 24927c478bd9Sstevel@tonic-gate 2493ae115bc7Smrj ASSERT(hat == kas.a_hat || vaddr <= _userlimit); 24947c478bd9Sstevel@tonic-gate if (IN_VA_HOLE(vaddr)) 24957c478bd9Sstevel@tonic-gate return (-1); 24967c478bd9Sstevel@tonic-gate ht = htable_getpage(hat, vaddr, NULL); 24977c478bd9Sstevel@tonic-gate if (ht == NULL) 24987c478bd9Sstevel@tonic-gate return (-1); 24997c478bd9Sstevel@tonic-gate pagesize = LEVEL_SIZE(ht->ht_level); 25007c478bd9Sstevel@tonic-gate htable_release(ht); 25017c478bd9Sstevel@tonic-gate return (pagesize); 25027c478bd9Sstevel@tonic-gate } 25037c478bd9Sstevel@tonic-gate 25047c478bd9Sstevel@tonic-gate 25057c478bd9Sstevel@tonic-gate 25067c478bd9Sstevel@tonic-gate /* 25077c478bd9Sstevel@tonic-gate * pfn_t hat_getpfnum(hat, addr) 25087c478bd9Sstevel@tonic-gate * returns pfn for <hat, addr> or PFN_INVALID if mapping is invalid. 25097c478bd9Sstevel@tonic-gate */ 25107c478bd9Sstevel@tonic-gate pfn_t 25117c478bd9Sstevel@tonic-gate hat_getpfnum(hat_t *hat, caddr_t addr) 25127c478bd9Sstevel@tonic-gate { 25137c478bd9Sstevel@tonic-gate uintptr_t vaddr = ALIGN2PAGE(addr); 25147c478bd9Sstevel@tonic-gate htable_t *ht; 25157c478bd9Sstevel@tonic-gate uint_t entry; 25167c478bd9Sstevel@tonic-gate pfn_t pfn = PFN_INVALID; 25177c478bd9Sstevel@tonic-gate 2518ae115bc7Smrj ASSERT(hat == kas.a_hat || vaddr <= _userlimit); 25197c478bd9Sstevel@tonic-gate if (khat_running == 0) 2520ae115bc7Smrj return (PFN_INVALID); 25217c478bd9Sstevel@tonic-gate 25227c478bd9Sstevel@tonic-gate if (IN_VA_HOLE(vaddr)) 25237c478bd9Sstevel@tonic-gate return (PFN_INVALID); 25247c478bd9Sstevel@tonic-gate 25257c478bd9Sstevel@tonic-gate /* 25267c478bd9Sstevel@tonic-gate * A very common use of hat_getpfnum() is from the DDI for kernel pages. 25277c478bd9Sstevel@tonic-gate * Use the kmap_ptes (which also covers the 32 bit heap) to speed 25287c478bd9Sstevel@tonic-gate * this up. 25297c478bd9Sstevel@tonic-gate */ 25307c478bd9Sstevel@tonic-gate if (mmu.kmap_addr <= vaddr && vaddr < mmu.kmap_eaddr) { 25317c478bd9Sstevel@tonic-gate x86pte_t pte; 2532ae115bc7Smrj pgcnt_t pg_index; 25337c478bd9Sstevel@tonic-gate 2534ae115bc7Smrj pg_index = mmu_btop(vaddr - mmu.kmap_addr); 2535ae115bc7Smrj pte = GET_PTE(PT_INDEX_PTR(mmu.kmap_ptes, pg_index)); 25367c478bd9Sstevel@tonic-gate if (!PTE_ISVALID(pte)) 25377c478bd9Sstevel@tonic-gate return (PFN_INVALID); 25387c478bd9Sstevel@tonic-gate /*LINTED [use of constant 0 causes a silly lint warning] */ 25397c478bd9Sstevel@tonic-gate return (PTE2PFN(pte, 0)); 25407c478bd9Sstevel@tonic-gate } 25417c478bd9Sstevel@tonic-gate 25427c478bd9Sstevel@tonic-gate ht = htable_getpage(hat, vaddr, &entry); 25437c478bd9Sstevel@tonic-gate if (ht == NULL) 25447c478bd9Sstevel@tonic-gate return (PFN_INVALID); 25457c478bd9Sstevel@tonic-gate ASSERT(vaddr >= ht->ht_vaddr); 25467c478bd9Sstevel@tonic-gate ASSERT(vaddr <= HTABLE_LAST_PAGE(ht)); 25477c478bd9Sstevel@tonic-gate pfn = PTE2PFN(x86pte_get(ht, entry), ht->ht_level); 25487c478bd9Sstevel@tonic-gate if (ht->ht_level > 0) 25497c478bd9Sstevel@tonic-gate pfn += mmu_btop(vaddr & LEVEL_OFFSET(ht->ht_level)); 25507c478bd9Sstevel@tonic-gate htable_release(ht); 25517c478bd9Sstevel@tonic-gate return (pfn); 25527c478bd9Sstevel@tonic-gate } 25537c478bd9Sstevel@tonic-gate 25547c478bd9Sstevel@tonic-gate /* 25557c478bd9Sstevel@tonic-gate * hat_getkpfnum() is an obsolete DDI routine, and its use is discouraged. 25567c478bd9Sstevel@tonic-gate * Use hat_getpfnum(kas.a_hat, ...) instead. 25577c478bd9Sstevel@tonic-gate * 25587c478bd9Sstevel@tonic-gate * We'd like to return PFN_INVALID if the mappings have underlying page_t's 25597c478bd9Sstevel@tonic-gate * but can't right now due to the fact that some software has grown to use 25607c478bd9Sstevel@tonic-gate * this interface incorrectly. So for now when the interface is misused, 25617c478bd9Sstevel@tonic-gate * return a warning to the user that in the future it won't work in the 25627c478bd9Sstevel@tonic-gate * way they're abusing it, and carry on. 25637c478bd9Sstevel@tonic-gate * 25647c478bd9Sstevel@tonic-gate * Note that hat_getkpfnum() is never supported on amd64. 25657c478bd9Sstevel@tonic-gate */ 25667c478bd9Sstevel@tonic-gate #if !defined(__amd64) 25677c478bd9Sstevel@tonic-gate pfn_t 25687c478bd9Sstevel@tonic-gate hat_getkpfnum(caddr_t addr) 25697c478bd9Sstevel@tonic-gate { 25707c478bd9Sstevel@tonic-gate pfn_t pfn; 25717c478bd9Sstevel@tonic-gate int badcaller = 0; 25727c478bd9Sstevel@tonic-gate 25737c478bd9Sstevel@tonic-gate if (khat_running == 0) 25747c478bd9Sstevel@tonic-gate panic("hat_getkpfnum(): called too early\n"); 25757c478bd9Sstevel@tonic-gate if ((uintptr_t)addr < kernelbase) 25767c478bd9Sstevel@tonic-gate return (PFN_INVALID); 25777c478bd9Sstevel@tonic-gate 25787c478bd9Sstevel@tonic-gate 25797c478bd9Sstevel@tonic-gate if (segkpm && IS_KPM_ADDR(addr)) { 25807c478bd9Sstevel@tonic-gate badcaller = 1; 25817c478bd9Sstevel@tonic-gate pfn = hat_kpm_va2pfn(addr); 25827c478bd9Sstevel@tonic-gate } else { 25837c478bd9Sstevel@tonic-gate pfn = hat_getpfnum(kas.a_hat, addr); 25847c478bd9Sstevel@tonic-gate badcaller = pf_is_memory(pfn); 25857c478bd9Sstevel@tonic-gate } 25867c478bd9Sstevel@tonic-gate 25877c478bd9Sstevel@tonic-gate if (badcaller) 25887c478bd9Sstevel@tonic-gate hat_getkpfnum_badcall(caller()); 25897c478bd9Sstevel@tonic-gate return (pfn); 25907c478bd9Sstevel@tonic-gate } 25917c478bd9Sstevel@tonic-gate #endif /* __amd64 */ 25927c478bd9Sstevel@tonic-gate 25937c478bd9Sstevel@tonic-gate /* 25947c478bd9Sstevel@tonic-gate * int hat_probe(hat, addr) 25957c478bd9Sstevel@tonic-gate * return 0 if no valid mapping is present. Faster version 25967c478bd9Sstevel@tonic-gate * of hat_getattr in certain architectures. 25977c478bd9Sstevel@tonic-gate */ 25987c478bd9Sstevel@tonic-gate int 25997c478bd9Sstevel@tonic-gate hat_probe(hat_t *hat, caddr_t addr) 26007c478bd9Sstevel@tonic-gate { 26017c478bd9Sstevel@tonic-gate uintptr_t vaddr = ALIGN2PAGE(addr); 26027c478bd9Sstevel@tonic-gate uint_t entry; 26037c478bd9Sstevel@tonic-gate htable_t *ht; 26047c478bd9Sstevel@tonic-gate pgcnt_t pg_off; 26057c478bd9Sstevel@tonic-gate 2606ae115bc7Smrj ASSERT(hat == kas.a_hat || vaddr <= _userlimit); 26077c478bd9Sstevel@tonic-gate ASSERT(hat == kas.a_hat || 26087c478bd9Sstevel@tonic-gate AS_LOCK_HELD(hat->hat_as, &hat->hat_as->a_lock)); 26097c478bd9Sstevel@tonic-gate if (IN_VA_HOLE(vaddr)) 26107c478bd9Sstevel@tonic-gate return (0); 26117c478bd9Sstevel@tonic-gate 26127c478bd9Sstevel@tonic-gate /* 26137c478bd9Sstevel@tonic-gate * Most common use of hat_probe is from segmap. We special case it 26147c478bd9Sstevel@tonic-gate * for performance. 26157c478bd9Sstevel@tonic-gate */ 26167c478bd9Sstevel@tonic-gate if (mmu.kmap_addr <= vaddr && vaddr < mmu.kmap_eaddr) { 26177c478bd9Sstevel@tonic-gate pg_off = mmu_btop(vaddr - mmu.kmap_addr); 26187c478bd9Sstevel@tonic-gate if (mmu.pae_hat) 26197c478bd9Sstevel@tonic-gate return (PTE_ISVALID(mmu.kmap_ptes[pg_off])); 26207c478bd9Sstevel@tonic-gate else 26217c478bd9Sstevel@tonic-gate return (PTE_ISVALID( 26227c478bd9Sstevel@tonic-gate ((x86pte32_t *)mmu.kmap_ptes)[pg_off])); 26237c478bd9Sstevel@tonic-gate } 26247c478bd9Sstevel@tonic-gate 26257c478bd9Sstevel@tonic-gate ht = htable_getpage(hat, vaddr, &entry); 26267c478bd9Sstevel@tonic-gate if (ht == NULL) 26277c478bd9Sstevel@tonic-gate return (0); 26287c478bd9Sstevel@tonic-gate htable_release(ht); 26297c478bd9Sstevel@tonic-gate return (1); 26307c478bd9Sstevel@tonic-gate } 26317c478bd9Sstevel@tonic-gate 26327c478bd9Sstevel@tonic-gate /* 2633*250b7ff9Sjosephb * Find out if the segment for hat_share()/hat_unshare() is DISM or locked ISM. 2634*250b7ff9Sjosephb */ 2635*250b7ff9Sjosephb static int 2636*250b7ff9Sjosephb is_it_dism(hat_t *hat, caddr_t va) 2637*250b7ff9Sjosephb { 2638*250b7ff9Sjosephb struct seg *seg; 2639*250b7ff9Sjosephb struct shm_data *shmd; 2640*250b7ff9Sjosephb struct spt_data *sptd; 2641*250b7ff9Sjosephb 2642*250b7ff9Sjosephb seg = as_findseg(hat->hat_as, va, 0); 2643*250b7ff9Sjosephb ASSERT(seg != NULL); 2644*250b7ff9Sjosephb ASSERT(seg->s_base <= va); 2645*250b7ff9Sjosephb shmd = (struct shm_data *)seg->s_data; 2646*250b7ff9Sjosephb ASSERT(shmd != NULL); 2647*250b7ff9Sjosephb sptd = (struct spt_data *)shmd->shm_sptseg->s_data; 2648*250b7ff9Sjosephb ASSERT(sptd != NULL); 2649*250b7ff9Sjosephb if (sptd->spt_flags & SHM_PAGEABLE) 2650*250b7ff9Sjosephb return (1); 2651*250b7ff9Sjosephb return (0); 2652*250b7ff9Sjosephb } 2653*250b7ff9Sjosephb 2654*250b7ff9Sjosephb /* 2655*250b7ff9Sjosephb * Simple implementation of ISM. hat_share() is similar to hat_memload_array(), 26567c478bd9Sstevel@tonic-gate * except that we use the ism_hat's existing mappings to determine the pages 2657*250b7ff9Sjosephb * and protections to use for this hat. If we find a full properly aligned 2658*250b7ff9Sjosephb * and sized pagetable, we will attempt to share the pagetable itself. 26597c478bd9Sstevel@tonic-gate */ 26607c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 26617c478bd9Sstevel@tonic-gate int 26627c478bd9Sstevel@tonic-gate hat_share( 26637c478bd9Sstevel@tonic-gate hat_t *hat, 26647c478bd9Sstevel@tonic-gate caddr_t addr, 26657c478bd9Sstevel@tonic-gate hat_t *ism_hat, 26667c478bd9Sstevel@tonic-gate caddr_t src_addr, 26677c478bd9Sstevel@tonic-gate size_t len, /* almost useless value, see below.. */ 26687c478bd9Sstevel@tonic-gate uint_t ismszc) 26697c478bd9Sstevel@tonic-gate { 26707c478bd9Sstevel@tonic-gate uintptr_t vaddr_start = (uintptr_t)addr; 26717c478bd9Sstevel@tonic-gate uintptr_t vaddr; 26727c478bd9Sstevel@tonic-gate uintptr_t eaddr = vaddr_start + len; 26737c478bd9Sstevel@tonic-gate uintptr_t ism_addr_start = (uintptr_t)src_addr; 26747c478bd9Sstevel@tonic-gate uintptr_t ism_addr = ism_addr_start; 26757c478bd9Sstevel@tonic-gate uintptr_t e_ism_addr = ism_addr + len; 26767c478bd9Sstevel@tonic-gate htable_t *ism_ht = NULL; 26777c478bd9Sstevel@tonic-gate htable_t *ht; 26787c478bd9Sstevel@tonic-gate x86pte_t pte; 26797c478bd9Sstevel@tonic-gate page_t *pp; 26807c478bd9Sstevel@tonic-gate pfn_t pfn; 26817c478bd9Sstevel@tonic-gate level_t l; 26827c478bd9Sstevel@tonic-gate pgcnt_t pgcnt; 26837c478bd9Sstevel@tonic-gate uint_t prot; 26847c478bd9Sstevel@tonic-gate uint_t valid_cnt; 2685*250b7ff9Sjosephb int is_dism; 2686*250b7ff9Sjosephb int flags; 26877c478bd9Sstevel@tonic-gate 26887c478bd9Sstevel@tonic-gate /* 26897c478bd9Sstevel@tonic-gate * We might be asked to share an empty DISM hat by as_dup() 26907c478bd9Sstevel@tonic-gate */ 26917c478bd9Sstevel@tonic-gate ASSERT(hat != kas.a_hat); 2692ae115bc7Smrj ASSERT(eaddr <= _userlimit); 26937c478bd9Sstevel@tonic-gate if (!(ism_hat->hat_flags & HAT_SHARED)) { 26947c478bd9Sstevel@tonic-gate ASSERT(hat_get_mapped_size(ism_hat) == 0); 26957c478bd9Sstevel@tonic-gate return (0); 26967c478bd9Sstevel@tonic-gate } 26977c478bd9Sstevel@tonic-gate 26987c478bd9Sstevel@tonic-gate /* 26997c478bd9Sstevel@tonic-gate * The SPT segment driver often passes us a size larger than there are 27007c478bd9Sstevel@tonic-gate * valid mappings. That's because it rounds the segment size up to a 27017c478bd9Sstevel@tonic-gate * large pagesize, even if the actual memory mapped by ism_hat is less. 27027c478bd9Sstevel@tonic-gate */ 27037c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(vaddr_start)); 27047c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(ism_addr_start)); 27057c478bd9Sstevel@tonic-gate ASSERT(ism_hat->hat_flags & HAT_SHARED); 2706*250b7ff9Sjosephb is_dism = is_it_dism(hat, addr); 27077c478bd9Sstevel@tonic-gate while (ism_addr < e_ism_addr) { 27087c478bd9Sstevel@tonic-gate /* 27097c478bd9Sstevel@tonic-gate * use htable_walk to get the next valid ISM mapping 27107c478bd9Sstevel@tonic-gate */ 27117c478bd9Sstevel@tonic-gate pte = htable_walk(ism_hat, &ism_ht, &ism_addr, e_ism_addr); 27127c478bd9Sstevel@tonic-gate if (ism_ht == NULL) 27137c478bd9Sstevel@tonic-gate break; 27147c478bd9Sstevel@tonic-gate 27157c478bd9Sstevel@tonic-gate /* 2716*250b7ff9Sjosephb * First check to see if we already share the page table. 27177c478bd9Sstevel@tonic-gate */ 2718*250b7ff9Sjosephb l = ism_ht->ht_level; 27197c478bd9Sstevel@tonic-gate vaddr = vaddr_start + (ism_addr - ism_addr_start); 2720*250b7ff9Sjosephb ht = htable_lookup(hat, vaddr, l); 2721*250b7ff9Sjosephb if (ht != NULL) { 2722*250b7ff9Sjosephb if (ht->ht_flags & HTABLE_SHARED_PFN) 2723*250b7ff9Sjosephb goto shared; 2724*250b7ff9Sjosephb htable_release(ht); 2725*250b7ff9Sjosephb goto not_shared; 27267c478bd9Sstevel@tonic-gate } 27277c478bd9Sstevel@tonic-gate 27287c478bd9Sstevel@tonic-gate /* 2729*250b7ff9Sjosephb * Can't ever share top table. 27307c478bd9Sstevel@tonic-gate */ 2731*250b7ff9Sjosephb if (l == mmu.max_level) 2732*250b7ff9Sjosephb goto not_shared; 27337c478bd9Sstevel@tonic-gate 2734*250b7ff9Sjosephb /* 2735*250b7ff9Sjosephb * Avoid level mismatches later due to DISM faults. 2736*250b7ff9Sjosephb */ 2737*250b7ff9Sjosephb if (is_dism && l > 0) 2738*250b7ff9Sjosephb goto not_shared; 27397c478bd9Sstevel@tonic-gate 2740*250b7ff9Sjosephb /* 2741*250b7ff9Sjosephb * addresses and lengths must align 2742*250b7ff9Sjosephb * table must be fully populated 2743*250b7ff9Sjosephb * no lower level page tables 2744*250b7ff9Sjosephb */ 2745*250b7ff9Sjosephb if (ism_addr != ism_ht->ht_vaddr || 2746*250b7ff9Sjosephb (vaddr & LEVEL_OFFSET(l + 1)) != 0) 2747*250b7ff9Sjosephb goto not_shared; 27487c478bd9Sstevel@tonic-gate 2749*250b7ff9Sjosephb /* 2750*250b7ff9Sjosephb * The range of address space must cover a full table. 2751*250b7ff9Sjosephb */ 2752*250b7ff9Sjosephb if (e_ism_addr - ism_addr < LEVEL_SIZE(1 + 1)) 2753*250b7ff9Sjosephb goto not_shared; 27547c478bd9Sstevel@tonic-gate 2755*250b7ff9Sjosephb /* 2756*250b7ff9Sjosephb * All entries in the ISM page table must be leaf PTEs. 2757*250b7ff9Sjosephb */ 2758*250b7ff9Sjosephb if (l > 0) { 2759*250b7ff9Sjosephb int e; 27607c478bd9Sstevel@tonic-gate 2761*250b7ff9Sjosephb /* 2762*250b7ff9Sjosephb * We know the 0th is from htable_walk() above. 2763*250b7ff9Sjosephb */ 2764*250b7ff9Sjosephb for (e = 1; e < HTABLE_NUM_PTES(ism_ht); ++e) { 2765*250b7ff9Sjosephb x86pte_t pte; 2766*250b7ff9Sjosephb pte = x86pte_get(ism_ht, e); 2767*250b7ff9Sjosephb if (!PTE_ISPAGE(pte, l)) 2768*250b7ff9Sjosephb goto not_shared; 27697c478bd9Sstevel@tonic-gate } 27707c478bd9Sstevel@tonic-gate } 27717c478bd9Sstevel@tonic-gate 2772*250b7ff9Sjosephb /* 2773*250b7ff9Sjosephb * share the page table 2774*250b7ff9Sjosephb */ 2775*250b7ff9Sjosephb ht = htable_create(hat, vaddr, l, ism_ht); 2776*250b7ff9Sjosephb shared: 2777*250b7ff9Sjosephb ASSERT(ht->ht_flags & HTABLE_SHARED_PFN); 2778*250b7ff9Sjosephb ASSERT(ht->ht_shares == ism_ht); 2779*250b7ff9Sjosephb hat->hat_ism_pgcnt += 2780*250b7ff9Sjosephb (ism_ht->ht_valid_cnt - ht->ht_valid_cnt) << 2781*250b7ff9Sjosephb (LEVEL_SHIFT(ht->ht_level) - MMU_PAGESHIFT); 2782*250b7ff9Sjosephb ht->ht_valid_cnt = ism_ht->ht_valid_cnt; 2783*250b7ff9Sjosephb htable_release(ht); 2784*250b7ff9Sjosephb ism_addr = ism_ht->ht_vaddr + LEVEL_SIZE(l + 1); 2785*250b7ff9Sjosephb htable_release(ism_ht); 2786*250b7ff9Sjosephb ism_ht = NULL; 2787*250b7ff9Sjosephb continue; 2788*250b7ff9Sjosephb 2789*250b7ff9Sjosephb not_shared: 27907c478bd9Sstevel@tonic-gate /* 27917c478bd9Sstevel@tonic-gate * Unable to share the page table. Instead we will 27927c478bd9Sstevel@tonic-gate * create new mappings from the values in the ISM mappings. 2793*250b7ff9Sjosephb * Figure out what level size mappings to use; 2794*250b7ff9Sjosephb */ 2795*250b7ff9Sjosephb for (l = ism_ht->ht_level; l > 0; --l) { 2796*250b7ff9Sjosephb if (LEVEL_SIZE(l) <= eaddr - vaddr && 2797*250b7ff9Sjosephb (vaddr & LEVEL_OFFSET(l)) == 0) 2798*250b7ff9Sjosephb break; 2799*250b7ff9Sjosephb } 2800*250b7ff9Sjosephb 2801*250b7ff9Sjosephb /* 28027c478bd9Sstevel@tonic-gate * The ISM mapping might be larger than the share area, 2803*250b7ff9Sjosephb * be careful to truncate it if needed. 28047c478bd9Sstevel@tonic-gate */ 28057c478bd9Sstevel@tonic-gate if (eaddr - vaddr >= LEVEL_SIZE(ism_ht->ht_level)) { 28067c478bd9Sstevel@tonic-gate pgcnt = mmu_btop(LEVEL_SIZE(ism_ht->ht_level)); 28077c478bd9Sstevel@tonic-gate } else { 28087c478bd9Sstevel@tonic-gate pgcnt = mmu_btop(eaddr - vaddr); 28097c478bd9Sstevel@tonic-gate l = 0; 28107c478bd9Sstevel@tonic-gate } 28117c478bd9Sstevel@tonic-gate 28127c478bd9Sstevel@tonic-gate pfn = PTE2PFN(pte, ism_ht->ht_level); 28137c478bd9Sstevel@tonic-gate ASSERT(pfn != PFN_INVALID); 28147c478bd9Sstevel@tonic-gate while (pgcnt > 0) { 28157c478bd9Sstevel@tonic-gate /* 28167c478bd9Sstevel@tonic-gate * Make a new pte for the PFN for this level. 28177c478bd9Sstevel@tonic-gate * Copy protections for the pte from the ISM pte. 28187c478bd9Sstevel@tonic-gate */ 28197c478bd9Sstevel@tonic-gate pp = page_numtopp_nolock(pfn); 28207c478bd9Sstevel@tonic-gate ASSERT(pp != NULL); 28217c478bd9Sstevel@tonic-gate 28227c478bd9Sstevel@tonic-gate prot = PROT_USER | PROT_READ | HAT_UNORDERED_OK; 28237c478bd9Sstevel@tonic-gate if (PTE_GET(pte, PT_WRITABLE)) 28247c478bd9Sstevel@tonic-gate prot |= PROT_WRITE; 28257c478bd9Sstevel@tonic-gate if (!PTE_GET(pte, PT_NX)) 28267c478bd9Sstevel@tonic-gate prot |= PROT_EXEC; 28277c478bd9Sstevel@tonic-gate 2828*250b7ff9Sjosephb flags = HAT_LOAD; 2829*250b7ff9Sjosephb if (!is_dism) 2830*250b7ff9Sjosephb flags |= HAT_LOAD_LOCK | HAT_LOAD_NOCONSIST; 2831*250b7ff9Sjosephb while (hati_load_common(hat, vaddr, pp, prot, flags, 2832ae115bc7Smrj l, pfn) != 0) { 2833ae115bc7Smrj if (l == 0) 2834ae115bc7Smrj panic("hati_load_common() failure"); 2835ae115bc7Smrj --l; 2836ae115bc7Smrj } 28377c478bd9Sstevel@tonic-gate 28387c478bd9Sstevel@tonic-gate vaddr += LEVEL_SIZE(l); 28397c478bd9Sstevel@tonic-gate ism_addr += LEVEL_SIZE(l); 28407c478bd9Sstevel@tonic-gate pfn += mmu_btop(LEVEL_SIZE(l)); 28417c478bd9Sstevel@tonic-gate pgcnt -= mmu_btop(LEVEL_SIZE(l)); 28427c478bd9Sstevel@tonic-gate } 28437c478bd9Sstevel@tonic-gate } 28447c478bd9Sstevel@tonic-gate if (ism_ht != NULL) 28457c478bd9Sstevel@tonic-gate htable_release(ism_ht); 28467c478bd9Sstevel@tonic-gate return (0); 28477c478bd9Sstevel@tonic-gate } 28487c478bd9Sstevel@tonic-gate 28497c478bd9Sstevel@tonic-gate 28507c478bd9Sstevel@tonic-gate /* 28517c478bd9Sstevel@tonic-gate * hat_unshare() is similar to hat_unload_callback(), but 28527c478bd9Sstevel@tonic-gate * we have to look for empty shared pagetables. Note that 28537c478bd9Sstevel@tonic-gate * hat_unshare() is always invoked against an entire segment. 28547c478bd9Sstevel@tonic-gate */ 28557c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 28567c478bd9Sstevel@tonic-gate void 28577c478bd9Sstevel@tonic-gate hat_unshare(hat_t *hat, caddr_t addr, size_t len, uint_t ismszc) 28587c478bd9Sstevel@tonic-gate { 28597c478bd9Sstevel@tonic-gate uintptr_t vaddr = (uintptr_t)addr; 28607c478bd9Sstevel@tonic-gate uintptr_t eaddr = vaddr + len; 28617c478bd9Sstevel@tonic-gate htable_t *ht = NULL; 28627c478bd9Sstevel@tonic-gate uint_t need_demaps = 0; 2863*250b7ff9Sjosephb int flags = HAT_UNLOAD_UNMAP; 2864*250b7ff9Sjosephb level_t l; 28657c478bd9Sstevel@tonic-gate 28667c478bd9Sstevel@tonic-gate ASSERT(hat != kas.a_hat); 2867ae115bc7Smrj ASSERT(eaddr <= _userlimit); 28687c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(vaddr)); 28697c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(eaddr)); 28707c478bd9Sstevel@tonic-gate 28717c478bd9Sstevel@tonic-gate /* 28727c478bd9Sstevel@tonic-gate * First go through and remove any shared pagetables. 28737c478bd9Sstevel@tonic-gate * 2874ae115bc7Smrj * Note that it's ok to delay the TLB shootdown till the entire range is 28757c478bd9Sstevel@tonic-gate * finished, because if hat_pageunload() were to unload a shared 2876ae115bc7Smrj * pagetable page, its hat_tlb_inval() will do a global TLB invalidate. 28777c478bd9Sstevel@tonic-gate */ 2878*250b7ff9Sjosephb l = mmu.max_page_level; 2879*250b7ff9Sjosephb if (l == mmu.max_level) 2880*250b7ff9Sjosephb --l; 2881*250b7ff9Sjosephb for (; l >= 0; --l) { 2882*250b7ff9Sjosephb for (vaddr = (uintptr_t)addr; vaddr < eaddr; 2883*250b7ff9Sjosephb vaddr = (vaddr & LEVEL_MASK(l + 1)) + LEVEL_SIZE(l + 1)) { 2884*250b7ff9Sjosephb ASSERT(!IN_VA_HOLE(vaddr)); 2885*250b7ff9Sjosephb /* 2886*250b7ff9Sjosephb * find a pagetable that maps the current address 2887*250b7ff9Sjosephb */ 2888*250b7ff9Sjosephb ht = htable_lookup(hat, vaddr, l); 2889*250b7ff9Sjosephb if (ht == NULL) 2890*250b7ff9Sjosephb continue; 28917c478bd9Sstevel@tonic-gate if (ht->ht_flags & HTABLE_SHARED_PFN) { 28927c478bd9Sstevel@tonic-gate /* 2893*250b7ff9Sjosephb * clear page count, set valid_cnt to 0, 2894*250b7ff9Sjosephb * let htable_release() finish the job 28957c478bd9Sstevel@tonic-gate */ 2896*250b7ff9Sjosephb hat->hat_ism_pgcnt -= ht->ht_valid_cnt << 2897*250b7ff9Sjosephb (LEVEL_SHIFT(ht->ht_level) - MMU_PAGESHIFT); 28987c478bd9Sstevel@tonic-gate ht->ht_valid_cnt = 0; 28997c478bd9Sstevel@tonic-gate need_demaps = 1; 29007c478bd9Sstevel@tonic-gate } 29017c478bd9Sstevel@tonic-gate htable_release(ht); 29027c478bd9Sstevel@tonic-gate } 29037c478bd9Sstevel@tonic-gate } 29047c478bd9Sstevel@tonic-gate 29057c478bd9Sstevel@tonic-gate /* 29067c478bd9Sstevel@tonic-gate * flush the TLBs - since we're probably dealing with MANY mappings 29077c478bd9Sstevel@tonic-gate * we do just one CR3 reload. 29087c478bd9Sstevel@tonic-gate */ 29097c478bd9Sstevel@tonic-gate if (!(hat->hat_flags & HAT_FREEING) && need_demaps) 2910ae115bc7Smrj hat_tlb_inval(hat, DEMAP_ALL_ADDR); 29117c478bd9Sstevel@tonic-gate 29127c478bd9Sstevel@tonic-gate /* 29137c478bd9Sstevel@tonic-gate * Now go back and clean up any unaligned mappings that 29147c478bd9Sstevel@tonic-gate * couldn't share pagetables. 29157c478bd9Sstevel@tonic-gate */ 2916*250b7ff9Sjosephb if (!is_it_dism(hat, addr)) 2917*250b7ff9Sjosephb flags |= HAT_UNLOAD_UNLOCK; 2918*250b7ff9Sjosephb hat_unload(hat, addr, len, flags); 29197c478bd9Sstevel@tonic-gate } 29207c478bd9Sstevel@tonic-gate 29217c478bd9Sstevel@tonic-gate 29227c478bd9Sstevel@tonic-gate /* 29237c478bd9Sstevel@tonic-gate * hat_reserve() does nothing 29247c478bd9Sstevel@tonic-gate */ 29257c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 29267c478bd9Sstevel@tonic-gate void 29277c478bd9Sstevel@tonic-gate hat_reserve(struct as *as, caddr_t addr, size_t len) 29287c478bd9Sstevel@tonic-gate { 29297c478bd9Sstevel@tonic-gate } 29307c478bd9Sstevel@tonic-gate 29317c478bd9Sstevel@tonic-gate 29327c478bd9Sstevel@tonic-gate /* 29337c478bd9Sstevel@tonic-gate * Called when all mappings to a page should have write permission removed. 29347c478bd9Sstevel@tonic-gate * Mostly stolem from hat_pagesync() 29357c478bd9Sstevel@tonic-gate */ 29367c478bd9Sstevel@tonic-gate static void 29377c478bd9Sstevel@tonic-gate hati_page_clrwrt(struct page *pp) 29387c478bd9Sstevel@tonic-gate { 29397c478bd9Sstevel@tonic-gate hment_t *hm = NULL; 29407c478bd9Sstevel@tonic-gate htable_t *ht; 29417c478bd9Sstevel@tonic-gate uint_t entry; 29427c478bd9Sstevel@tonic-gate x86pte_t old; 29437c478bd9Sstevel@tonic-gate x86pte_t new; 29447c478bd9Sstevel@tonic-gate uint_t pszc = 0; 29457c478bd9Sstevel@tonic-gate 29467c478bd9Sstevel@tonic-gate next_size: 29477c478bd9Sstevel@tonic-gate /* 29487c478bd9Sstevel@tonic-gate * walk thru the mapping list clearing write permission 29497c478bd9Sstevel@tonic-gate */ 29507c478bd9Sstevel@tonic-gate x86_hm_enter(pp); 29517c478bd9Sstevel@tonic-gate while ((hm = hment_walk(pp, &ht, &entry, hm)) != NULL) { 29527c478bd9Sstevel@tonic-gate if (ht->ht_level < pszc) 29537c478bd9Sstevel@tonic-gate continue; 29547c478bd9Sstevel@tonic-gate old = x86pte_get(ht, entry); 29557c478bd9Sstevel@tonic-gate 29567c478bd9Sstevel@tonic-gate for (;;) { 29577c478bd9Sstevel@tonic-gate /* 29587c478bd9Sstevel@tonic-gate * Is this mapping of interest? 29597c478bd9Sstevel@tonic-gate */ 29607c478bd9Sstevel@tonic-gate if (PTE2PFN(old, ht->ht_level) != pp->p_pagenum || 29617c478bd9Sstevel@tonic-gate PTE_GET(old, PT_WRITABLE) == 0) 29627c478bd9Sstevel@tonic-gate break; 29637c478bd9Sstevel@tonic-gate 29647c478bd9Sstevel@tonic-gate /* 29657c478bd9Sstevel@tonic-gate * Clear ref/mod writable bits. This requires cross 29667c478bd9Sstevel@tonic-gate * calls to ensure any executing TLBs see cleared bits. 29677c478bd9Sstevel@tonic-gate */ 29687c478bd9Sstevel@tonic-gate new = old; 29697c478bd9Sstevel@tonic-gate PTE_CLR(new, PT_REF | PT_MOD | PT_WRITABLE); 29707c478bd9Sstevel@tonic-gate old = hati_update_pte(ht, entry, old, new); 29717c478bd9Sstevel@tonic-gate if (old != 0) 29727c478bd9Sstevel@tonic-gate continue; 29737c478bd9Sstevel@tonic-gate 29747c478bd9Sstevel@tonic-gate break; 29757c478bd9Sstevel@tonic-gate } 29767c478bd9Sstevel@tonic-gate } 29777c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 29787c478bd9Sstevel@tonic-gate while (pszc < pp->p_szc) { 29797c478bd9Sstevel@tonic-gate page_t *tpp; 29807c478bd9Sstevel@tonic-gate pszc++; 29817c478bd9Sstevel@tonic-gate tpp = PP_GROUPLEADER(pp, pszc); 29827c478bd9Sstevel@tonic-gate if (pp != tpp) { 29837c478bd9Sstevel@tonic-gate pp = tpp; 29847c478bd9Sstevel@tonic-gate goto next_size; 29857c478bd9Sstevel@tonic-gate } 29867c478bd9Sstevel@tonic-gate } 29877c478bd9Sstevel@tonic-gate } 29887c478bd9Sstevel@tonic-gate 29897c478bd9Sstevel@tonic-gate /* 29907c478bd9Sstevel@tonic-gate * void hat_page_setattr(pp, flag) 29917c478bd9Sstevel@tonic-gate * void hat_page_clrattr(pp, flag) 29927c478bd9Sstevel@tonic-gate * used to set/clr ref/mod bits. 29937c478bd9Sstevel@tonic-gate */ 29947c478bd9Sstevel@tonic-gate void 29957c478bd9Sstevel@tonic-gate hat_page_setattr(struct page *pp, uint_t flag) 29967c478bd9Sstevel@tonic-gate { 29977c478bd9Sstevel@tonic-gate vnode_t *vp = pp->p_vnode; 29987c478bd9Sstevel@tonic-gate kmutex_t *vphm = NULL; 29997c478bd9Sstevel@tonic-gate page_t **listp; 3000d9615970Sqiao int noshuffle; 3001d9615970Sqiao 3002d9615970Sqiao noshuffle = flag & P_NSH; 3003d9615970Sqiao flag &= ~P_NSH; 30047c478bd9Sstevel@tonic-gate 30057c478bd9Sstevel@tonic-gate if (PP_GETRM(pp, flag) == flag) 30067c478bd9Sstevel@tonic-gate return; 30077c478bd9Sstevel@tonic-gate 3008d9615970Sqiao if ((flag & P_MOD) != 0 && vp != NULL && IS_VMODSORT(vp) && 3009d9615970Sqiao !noshuffle) { 30107c478bd9Sstevel@tonic-gate vphm = page_vnode_mutex(vp); 30117c478bd9Sstevel@tonic-gate mutex_enter(vphm); 30127c478bd9Sstevel@tonic-gate } 30137c478bd9Sstevel@tonic-gate 30147c478bd9Sstevel@tonic-gate PP_SETRM(pp, flag); 30157c478bd9Sstevel@tonic-gate 30167c478bd9Sstevel@tonic-gate if (vphm != NULL) { 30177c478bd9Sstevel@tonic-gate 30187c478bd9Sstevel@tonic-gate /* 30197c478bd9Sstevel@tonic-gate * Some File Systems examine v_pages for NULL w/o 30207c478bd9Sstevel@tonic-gate * grabbing the vphm mutex. Must not let it become NULL when 30217c478bd9Sstevel@tonic-gate * pp is the only page on the list. 30227c478bd9Sstevel@tonic-gate */ 30237c478bd9Sstevel@tonic-gate if (pp->p_vpnext != pp) { 30247c478bd9Sstevel@tonic-gate page_vpsub(&vp->v_pages, pp); 30257c478bd9Sstevel@tonic-gate if (vp->v_pages != NULL) 30267c478bd9Sstevel@tonic-gate listp = &vp->v_pages->p_vpprev->p_vpnext; 30277c478bd9Sstevel@tonic-gate else 30287c478bd9Sstevel@tonic-gate listp = &vp->v_pages; 30297c478bd9Sstevel@tonic-gate page_vpadd(listp, pp); 30307c478bd9Sstevel@tonic-gate } 30317c478bd9Sstevel@tonic-gate mutex_exit(vphm); 30327c478bd9Sstevel@tonic-gate } 30337c478bd9Sstevel@tonic-gate } 30347c478bd9Sstevel@tonic-gate 30357c478bd9Sstevel@tonic-gate void 30367c478bd9Sstevel@tonic-gate hat_page_clrattr(struct page *pp, uint_t flag) 30377c478bd9Sstevel@tonic-gate { 30387c478bd9Sstevel@tonic-gate vnode_t *vp = pp->p_vnode; 30397c478bd9Sstevel@tonic-gate ASSERT(!(flag & ~(P_MOD | P_REF | P_RO))); 30407c478bd9Sstevel@tonic-gate 30417c478bd9Sstevel@tonic-gate /* 3042a71e32b6Sstans * Caller is expected to hold page's io lock for VMODSORT to work 3043a71e32b6Sstans * correctly with pvn_vplist_dirty() and pvn_getdirty() when mod 3044a71e32b6Sstans * bit is cleared. 3045a71e32b6Sstans * We don't have assert to avoid tripping some existing third party 3046a71e32b6Sstans * code. The dirty page is moved back to top of the v_page list 3047a71e32b6Sstans * after IO is done in pvn_write_done(). 30487c478bd9Sstevel@tonic-gate */ 30497c478bd9Sstevel@tonic-gate PP_CLRRM(pp, flag); 30507c478bd9Sstevel@tonic-gate 3051a71e32b6Sstans if ((flag & P_MOD) != 0 && vp != NULL && IS_VMODSORT(vp)) { 30527c478bd9Sstevel@tonic-gate 30537c478bd9Sstevel@tonic-gate /* 30547c478bd9Sstevel@tonic-gate * VMODSORT works by removing write permissions and getting 30557c478bd9Sstevel@tonic-gate * a fault when a page is made dirty. At this point 30567c478bd9Sstevel@tonic-gate * we need to remove write permission from all mappings 30577c478bd9Sstevel@tonic-gate * to this page. 30587c478bd9Sstevel@tonic-gate */ 30597c478bd9Sstevel@tonic-gate hati_page_clrwrt(pp); 30607c478bd9Sstevel@tonic-gate } 30617c478bd9Sstevel@tonic-gate } 30627c478bd9Sstevel@tonic-gate 30637c478bd9Sstevel@tonic-gate /* 30647c478bd9Sstevel@tonic-gate * If flag is specified, returns 0 if attribute is disabled 30657c478bd9Sstevel@tonic-gate * and non zero if enabled. If flag specifes multiple attributs 30667c478bd9Sstevel@tonic-gate * then returns 0 if ALL atriibutes are disabled. This is an advisory 30677c478bd9Sstevel@tonic-gate * call. 30687c478bd9Sstevel@tonic-gate */ 30697c478bd9Sstevel@tonic-gate uint_t 30707c478bd9Sstevel@tonic-gate hat_page_getattr(struct page *pp, uint_t flag) 30717c478bd9Sstevel@tonic-gate { 30727c478bd9Sstevel@tonic-gate return (PP_GETRM(pp, flag)); 30737c478bd9Sstevel@tonic-gate } 30747c478bd9Sstevel@tonic-gate 30757c478bd9Sstevel@tonic-gate 30767c478bd9Sstevel@tonic-gate /* 30777c478bd9Sstevel@tonic-gate * common code used by hat_pageunload() and hment_steal() 30787c478bd9Sstevel@tonic-gate */ 30797c478bd9Sstevel@tonic-gate hment_t * 30807c478bd9Sstevel@tonic-gate hati_page_unmap(page_t *pp, htable_t *ht, uint_t entry) 30817c478bd9Sstevel@tonic-gate { 30827c478bd9Sstevel@tonic-gate x86pte_t old_pte; 30837c478bd9Sstevel@tonic-gate pfn_t pfn = pp->p_pagenum; 30847c478bd9Sstevel@tonic-gate hment_t *hm; 30857c478bd9Sstevel@tonic-gate 30867c478bd9Sstevel@tonic-gate /* 30877c478bd9Sstevel@tonic-gate * We need to acquire a hold on the htable in order to 30887c478bd9Sstevel@tonic-gate * do the invalidate. We know the htable must exist, since 30897c478bd9Sstevel@tonic-gate * unmap's don't release the htable until after removing any 30907c478bd9Sstevel@tonic-gate * hment. Having x86_hm_enter() keeps that from proceeding. 30917c478bd9Sstevel@tonic-gate */ 30927c478bd9Sstevel@tonic-gate htable_acquire(ht); 30937c478bd9Sstevel@tonic-gate 30947c478bd9Sstevel@tonic-gate /* 30957c478bd9Sstevel@tonic-gate * Invalidate the PTE and remove the hment. 30967c478bd9Sstevel@tonic-gate */ 3097ae115bc7Smrj old_pte = x86pte_inval(ht, entry, 0, NULL); 3098aa2ed9e5Sjosephb if (PTE2PFN(old_pte, ht->ht_level) != pfn) { 3099ae115bc7Smrj panic("x86pte_inval() failure found PTE = " FMT_PTE 3100aa2ed9e5Sjosephb " pfn being unmapped is %lx ht=0x%lx entry=0x%x", 3101aa2ed9e5Sjosephb old_pte, pfn, (uintptr_t)ht, entry); 3102aa2ed9e5Sjosephb } 31037c478bd9Sstevel@tonic-gate 31047c478bd9Sstevel@tonic-gate /* 31057c478bd9Sstevel@tonic-gate * Clean up all the htable information for this mapping 31067c478bd9Sstevel@tonic-gate */ 31077c478bd9Sstevel@tonic-gate ASSERT(ht->ht_valid_cnt > 0); 31087c478bd9Sstevel@tonic-gate HTABLE_DEC(ht->ht_valid_cnt); 31097c478bd9Sstevel@tonic-gate PGCNT_DEC(ht->ht_hat, ht->ht_level); 31107c478bd9Sstevel@tonic-gate 31117c478bd9Sstevel@tonic-gate /* 31127c478bd9Sstevel@tonic-gate * sync ref/mod bits to the page_t 31137c478bd9Sstevel@tonic-gate */ 3114ae115bc7Smrj if (PTE_GET(old_pte, PT_SOFTWARE) < PT_NOSYNC) 31157c478bd9Sstevel@tonic-gate hati_sync_pte_to_page(pp, old_pte, ht->ht_level); 31167c478bd9Sstevel@tonic-gate 31177c478bd9Sstevel@tonic-gate /* 31187c478bd9Sstevel@tonic-gate * Remove the mapping list entry for this page. 31197c478bd9Sstevel@tonic-gate */ 31207c478bd9Sstevel@tonic-gate hm = hment_remove(pp, ht, entry); 31217c478bd9Sstevel@tonic-gate 31227c478bd9Sstevel@tonic-gate /* 31237c478bd9Sstevel@tonic-gate * drop the mapping list lock so that we might free the 31247c478bd9Sstevel@tonic-gate * hment and htable. 31257c478bd9Sstevel@tonic-gate */ 31267c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 31277c478bd9Sstevel@tonic-gate htable_release(ht); 31287c478bd9Sstevel@tonic-gate return (hm); 31297c478bd9Sstevel@tonic-gate } 31307c478bd9Sstevel@tonic-gate 3131a5652762Spraks extern int vpm_enable; 31327c478bd9Sstevel@tonic-gate /* 31337c478bd9Sstevel@tonic-gate * Unload all translations to a page. If the page is a subpage of a large 31347c478bd9Sstevel@tonic-gate * page, the large page mappings are also removed. 31357c478bd9Sstevel@tonic-gate * 31367c478bd9Sstevel@tonic-gate * The forceflags are unused. 31377c478bd9Sstevel@tonic-gate */ 31387c478bd9Sstevel@tonic-gate 31397c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 31407c478bd9Sstevel@tonic-gate static int 31417c478bd9Sstevel@tonic-gate hati_pageunload(struct page *pp, uint_t pg_szcd, uint_t forceflag) 31427c478bd9Sstevel@tonic-gate { 31437c478bd9Sstevel@tonic-gate page_t *cur_pp = pp; 31447c478bd9Sstevel@tonic-gate hment_t *hm; 31457c478bd9Sstevel@tonic-gate hment_t *prev; 31467c478bd9Sstevel@tonic-gate htable_t *ht; 31477c478bd9Sstevel@tonic-gate uint_t entry; 31487c478bd9Sstevel@tonic-gate level_t level; 31497c478bd9Sstevel@tonic-gate 3150a5652762Spraks #if defined(__amd64) 3151a5652762Spraks /* 3152a5652762Spraks * clear the vpm ref. 3153a5652762Spraks */ 3154a5652762Spraks if (vpm_enable) { 3155a5652762Spraks pp->p_vpmref = 0; 3156a5652762Spraks } 3157a5652762Spraks #endif 31587c478bd9Sstevel@tonic-gate /* 31597c478bd9Sstevel@tonic-gate * The loop with next_size handles pages with multiple pagesize mappings 31607c478bd9Sstevel@tonic-gate */ 31617c478bd9Sstevel@tonic-gate next_size: 31627c478bd9Sstevel@tonic-gate for (;;) { 31637c478bd9Sstevel@tonic-gate 31647c478bd9Sstevel@tonic-gate /* 31657c478bd9Sstevel@tonic-gate * Get a mapping list entry 31667c478bd9Sstevel@tonic-gate */ 31677c478bd9Sstevel@tonic-gate x86_hm_enter(cur_pp); 31687c478bd9Sstevel@tonic-gate for (prev = NULL; ; prev = hm) { 31697c478bd9Sstevel@tonic-gate hm = hment_walk(cur_pp, &ht, &entry, prev); 31707c478bd9Sstevel@tonic-gate if (hm == NULL) { 31717c478bd9Sstevel@tonic-gate x86_hm_exit(cur_pp); 31727c478bd9Sstevel@tonic-gate 31737c478bd9Sstevel@tonic-gate /* 31747c478bd9Sstevel@tonic-gate * If not part of a larger page, we're done. 31757c478bd9Sstevel@tonic-gate */ 3176ae115bc7Smrj if (cur_pp->p_szc <= pg_szcd) { 31777c478bd9Sstevel@tonic-gate return (0); 3178ae115bc7Smrj } 31797c478bd9Sstevel@tonic-gate 31807c478bd9Sstevel@tonic-gate /* 31817c478bd9Sstevel@tonic-gate * Else check the next larger page size. 31827c478bd9Sstevel@tonic-gate * hat_page_demote() may decrease p_szc 31837c478bd9Sstevel@tonic-gate * but that's ok we'll just take an extra 31847c478bd9Sstevel@tonic-gate * trip discover there're no larger mappings 31857c478bd9Sstevel@tonic-gate * and return. 31867c478bd9Sstevel@tonic-gate */ 31877c478bd9Sstevel@tonic-gate ++pg_szcd; 31887c478bd9Sstevel@tonic-gate cur_pp = PP_GROUPLEADER(cur_pp, pg_szcd); 31897c478bd9Sstevel@tonic-gate goto next_size; 31907c478bd9Sstevel@tonic-gate } 31917c478bd9Sstevel@tonic-gate 31927c478bd9Sstevel@tonic-gate /* 31937c478bd9Sstevel@tonic-gate * If this mapping size matches, remove it. 31947c478bd9Sstevel@tonic-gate */ 31957c478bd9Sstevel@tonic-gate level = ht->ht_level; 31967c478bd9Sstevel@tonic-gate if (level == pg_szcd) 31977c478bd9Sstevel@tonic-gate break; 31987c478bd9Sstevel@tonic-gate } 31997c478bd9Sstevel@tonic-gate 32007c478bd9Sstevel@tonic-gate /* 32017c478bd9Sstevel@tonic-gate * Remove the mapping list entry for this page. 32027c478bd9Sstevel@tonic-gate * Note this does the x86_hm_exit() for us. 32037c478bd9Sstevel@tonic-gate */ 32047c478bd9Sstevel@tonic-gate hm = hati_page_unmap(cur_pp, ht, entry); 32057c478bd9Sstevel@tonic-gate if (hm != NULL) 32067c478bd9Sstevel@tonic-gate hment_free(hm); 32077c478bd9Sstevel@tonic-gate } 32087c478bd9Sstevel@tonic-gate } 32097c478bd9Sstevel@tonic-gate 32107c478bd9Sstevel@tonic-gate int 32117c478bd9Sstevel@tonic-gate hat_pageunload(struct page *pp, uint_t forceflag) 32127c478bd9Sstevel@tonic-gate { 32137c478bd9Sstevel@tonic-gate ASSERT(PAGE_EXCL(pp)); 32147c478bd9Sstevel@tonic-gate return (hati_pageunload(pp, 0, forceflag)); 32157c478bd9Sstevel@tonic-gate } 32167c478bd9Sstevel@tonic-gate 32177c478bd9Sstevel@tonic-gate /* 32187c478bd9Sstevel@tonic-gate * Unload all large mappings to pp and reduce by 1 p_szc field of every large 32197c478bd9Sstevel@tonic-gate * page level that included pp. 32207c478bd9Sstevel@tonic-gate * 32217c478bd9Sstevel@tonic-gate * pp must be locked EXCL. Even though no other constituent pages are locked 32227c478bd9Sstevel@tonic-gate * it's legal to unload large mappings to pp because all constituent pages of 32237c478bd9Sstevel@tonic-gate * large locked mappings have to be locked SHARED. therefore if we have EXCL 32247c478bd9Sstevel@tonic-gate * lock on one of constituent pages none of the large mappings to pp are 32257c478bd9Sstevel@tonic-gate * locked. 32267c478bd9Sstevel@tonic-gate * 32277c478bd9Sstevel@tonic-gate * Change (always decrease) p_szc field starting from the last constituent 32287c478bd9Sstevel@tonic-gate * page and ending with root constituent page so that root's pszc always shows 32297c478bd9Sstevel@tonic-gate * the area where hat_page_demote() may be active. 32307c478bd9Sstevel@tonic-gate * 32317c478bd9Sstevel@tonic-gate * This mechanism is only used for file system pages where it's not always 32327c478bd9Sstevel@tonic-gate * possible to get EXCL locks on all constituent pages to demote the size code 32337c478bd9Sstevel@tonic-gate * (as is done for anonymous or kernel large pages). 32347c478bd9Sstevel@tonic-gate */ 32357c478bd9Sstevel@tonic-gate void 32367c478bd9Sstevel@tonic-gate hat_page_demote(page_t *pp) 32377c478bd9Sstevel@tonic-gate { 32387c478bd9Sstevel@tonic-gate uint_t pszc; 32397c478bd9Sstevel@tonic-gate uint_t rszc; 32407c478bd9Sstevel@tonic-gate uint_t szc; 32417c478bd9Sstevel@tonic-gate page_t *rootpp; 32427c478bd9Sstevel@tonic-gate page_t *firstpp; 32437c478bd9Sstevel@tonic-gate page_t *lastpp; 32447c478bd9Sstevel@tonic-gate pgcnt_t pgcnt; 32457c478bd9Sstevel@tonic-gate 32467c478bd9Sstevel@tonic-gate ASSERT(PAGE_EXCL(pp)); 32477c478bd9Sstevel@tonic-gate ASSERT(!PP_ISFREE(pp)); 32487c478bd9Sstevel@tonic-gate ASSERT(page_szc_lock_assert(pp)); 32497c478bd9Sstevel@tonic-gate 32507c478bd9Sstevel@tonic-gate if (pp->p_szc == 0) 32517c478bd9Sstevel@tonic-gate return; 32527c478bd9Sstevel@tonic-gate 32537c478bd9Sstevel@tonic-gate rootpp = PP_GROUPLEADER(pp, 1); 32547c478bd9Sstevel@tonic-gate (void) hati_pageunload(rootpp, 1, HAT_FORCE_PGUNLOAD); 32557c478bd9Sstevel@tonic-gate 32567c478bd9Sstevel@tonic-gate /* 32577c478bd9Sstevel@tonic-gate * all large mappings to pp are gone 32587c478bd9Sstevel@tonic-gate * and no new can be setup since pp is locked exclusively. 32597c478bd9Sstevel@tonic-gate * 32607c478bd9Sstevel@tonic-gate * Lock the root to make sure there's only one hat_page_demote() 32617c478bd9Sstevel@tonic-gate * outstanding within the area of this root's pszc. 32627c478bd9Sstevel@tonic-gate * 32637c478bd9Sstevel@tonic-gate * Second potential hat_page_demote() is already eliminated by upper 32647c478bd9Sstevel@tonic-gate * VM layer via page_szc_lock() but we don't rely on it and use our 32657c478bd9Sstevel@tonic-gate * own locking (so that upper layer locking can be changed without 32667c478bd9Sstevel@tonic-gate * assumptions that hat depends on upper layer VM to prevent multiple 32677c478bd9Sstevel@tonic-gate * hat_page_demote() to be issued simultaneously to the same large 32687c478bd9Sstevel@tonic-gate * page). 32697c478bd9Sstevel@tonic-gate */ 32707c478bd9Sstevel@tonic-gate again: 32717c478bd9Sstevel@tonic-gate pszc = pp->p_szc; 32727c478bd9Sstevel@tonic-gate if (pszc == 0) 32737c478bd9Sstevel@tonic-gate return; 32747c478bd9Sstevel@tonic-gate rootpp = PP_GROUPLEADER(pp, pszc); 32757c478bd9Sstevel@tonic-gate x86_hm_enter(rootpp); 32767c478bd9Sstevel@tonic-gate /* 32777c478bd9Sstevel@tonic-gate * If root's p_szc is different from pszc we raced with another 32787c478bd9Sstevel@tonic-gate * hat_page_demote(). Drop the lock and try to find the root again. 32797c478bd9Sstevel@tonic-gate * If root's p_szc is greater than pszc previous hat_page_demote() is 32807c478bd9Sstevel@tonic-gate * not done yet. Take and release mlist lock of root's root to wait 32817c478bd9Sstevel@tonic-gate * for previous hat_page_demote() to complete. 32827c478bd9Sstevel@tonic-gate */ 32837c478bd9Sstevel@tonic-gate if ((rszc = rootpp->p_szc) != pszc) { 32847c478bd9Sstevel@tonic-gate x86_hm_exit(rootpp); 32857c478bd9Sstevel@tonic-gate if (rszc > pszc) { 32867c478bd9Sstevel@tonic-gate /* p_szc of a locked non free page can't increase */ 32877c478bd9Sstevel@tonic-gate ASSERT(pp != rootpp); 32887c478bd9Sstevel@tonic-gate 32897c478bd9Sstevel@tonic-gate rootpp = PP_GROUPLEADER(rootpp, rszc); 32907c478bd9Sstevel@tonic-gate x86_hm_enter(rootpp); 32917c478bd9Sstevel@tonic-gate x86_hm_exit(rootpp); 32927c478bd9Sstevel@tonic-gate } 32937c478bd9Sstevel@tonic-gate goto again; 32947c478bd9Sstevel@tonic-gate } 32957c478bd9Sstevel@tonic-gate ASSERT(pp->p_szc == pszc); 32967c478bd9Sstevel@tonic-gate 32977c478bd9Sstevel@tonic-gate /* 32987c478bd9Sstevel@tonic-gate * Decrement by 1 p_szc of every constituent page of a region that 32997c478bd9Sstevel@tonic-gate * covered pp. For example if original szc is 3 it gets changed to 2 33007c478bd9Sstevel@tonic-gate * everywhere except in region 2 that covered pp. Region 2 that 33017c478bd9Sstevel@tonic-gate * covered pp gets demoted to 1 everywhere except in region 1 that 33027c478bd9Sstevel@tonic-gate * covered pp. The region 1 that covered pp is demoted to region 33037c478bd9Sstevel@tonic-gate * 0. It's done this way because from region 3 we removed level 3 33047c478bd9Sstevel@tonic-gate * mappings, from region 2 that covered pp we removed level 2 mappings 33057c478bd9Sstevel@tonic-gate * and from region 1 that covered pp we removed level 1 mappings. All 33067c478bd9Sstevel@tonic-gate * changes are done from from high pfn's to low pfn's so that roots 33077c478bd9Sstevel@tonic-gate * are changed last allowing one to know the largest region where 33087c478bd9Sstevel@tonic-gate * hat_page_demote() is stil active by only looking at the root page. 33097c478bd9Sstevel@tonic-gate * 33107c478bd9Sstevel@tonic-gate * This algorithm is implemented in 2 while loops. First loop changes 33117c478bd9Sstevel@tonic-gate * p_szc of pages to the right of pp's level 1 region and second 33127c478bd9Sstevel@tonic-gate * loop changes p_szc of pages of level 1 region that covers pp 33137c478bd9Sstevel@tonic-gate * and all pages to the left of level 1 region that covers pp. 33147c478bd9Sstevel@tonic-gate * In the first loop p_szc keeps dropping with every iteration 33157c478bd9Sstevel@tonic-gate * and in the second loop it keeps increasing with every iteration. 33167c478bd9Sstevel@tonic-gate * 33177c478bd9Sstevel@tonic-gate * First loop description: Demote pages to the right of pp outside of 33187c478bd9Sstevel@tonic-gate * level 1 region that covers pp. In every iteration of the while 33197c478bd9Sstevel@tonic-gate * loop below find the last page of szc region and the first page of 33207c478bd9Sstevel@tonic-gate * (szc - 1) region that is immediately to the right of (szc - 1) 33217c478bd9Sstevel@tonic-gate * region that covers pp. From last such page to first such page 33227c478bd9Sstevel@tonic-gate * change every page's szc to szc - 1. Decrement szc and continue 33237c478bd9Sstevel@tonic-gate * looping until szc is 1. If pp belongs to the last (szc - 1) region 33247c478bd9Sstevel@tonic-gate * of szc region skip to the next iteration. 33257c478bd9Sstevel@tonic-gate */ 33267c478bd9Sstevel@tonic-gate szc = pszc; 33277c478bd9Sstevel@tonic-gate while (szc > 1) { 33287c478bd9Sstevel@tonic-gate lastpp = PP_GROUPLEADER(pp, szc); 33297c478bd9Sstevel@tonic-gate pgcnt = page_get_pagecnt(szc); 33307c478bd9Sstevel@tonic-gate lastpp += pgcnt - 1; 33317c478bd9Sstevel@tonic-gate firstpp = PP_GROUPLEADER(pp, (szc - 1)); 33327c478bd9Sstevel@tonic-gate pgcnt = page_get_pagecnt(szc - 1); 33337c478bd9Sstevel@tonic-gate if (lastpp - firstpp < pgcnt) { 33347c478bd9Sstevel@tonic-gate szc--; 33357c478bd9Sstevel@tonic-gate continue; 33367c478bd9Sstevel@tonic-gate } 33377c478bd9Sstevel@tonic-gate firstpp += pgcnt; 33387c478bd9Sstevel@tonic-gate while (lastpp != firstpp) { 33397c478bd9Sstevel@tonic-gate ASSERT(lastpp->p_szc == pszc); 33407c478bd9Sstevel@tonic-gate lastpp->p_szc = szc - 1; 33417c478bd9Sstevel@tonic-gate lastpp--; 33427c478bd9Sstevel@tonic-gate } 33437c478bd9Sstevel@tonic-gate firstpp->p_szc = szc - 1; 33447c478bd9Sstevel@tonic-gate szc--; 33457c478bd9Sstevel@tonic-gate } 33467c478bd9Sstevel@tonic-gate 33477c478bd9Sstevel@tonic-gate /* 33487c478bd9Sstevel@tonic-gate * Second loop description: 33497c478bd9Sstevel@tonic-gate * First iteration changes p_szc to 0 of every 33507c478bd9Sstevel@tonic-gate * page of level 1 region that covers pp. 33517c478bd9Sstevel@tonic-gate * Subsequent iterations find last page of szc region 33527c478bd9Sstevel@tonic-gate * immediately to the left of szc region that covered pp 33537c478bd9Sstevel@tonic-gate * and first page of (szc + 1) region that covers pp. 33547c478bd9Sstevel@tonic-gate * From last to first page change p_szc of every page to szc. 33557c478bd9Sstevel@tonic-gate * Increment szc and continue looping until szc is pszc. 33567c478bd9Sstevel@tonic-gate * If pp belongs to the fist szc region of (szc + 1) region 33577c478bd9Sstevel@tonic-gate * skip to the next iteration. 33587c478bd9Sstevel@tonic-gate * 33597c478bd9Sstevel@tonic-gate */ 33607c478bd9Sstevel@tonic-gate szc = 0; 33617c478bd9Sstevel@tonic-gate while (szc < pszc) { 33627c478bd9Sstevel@tonic-gate firstpp = PP_GROUPLEADER(pp, (szc + 1)); 33637c478bd9Sstevel@tonic-gate if (szc == 0) { 33647c478bd9Sstevel@tonic-gate pgcnt = page_get_pagecnt(1); 33657c478bd9Sstevel@tonic-gate lastpp = firstpp + (pgcnt - 1); 33667c478bd9Sstevel@tonic-gate } else { 33677c478bd9Sstevel@tonic-gate lastpp = PP_GROUPLEADER(pp, szc); 33687c478bd9Sstevel@tonic-gate if (firstpp == lastpp) { 33697c478bd9Sstevel@tonic-gate szc++; 33707c478bd9Sstevel@tonic-gate continue; 33717c478bd9Sstevel@tonic-gate } 33727c478bd9Sstevel@tonic-gate lastpp--; 33737c478bd9Sstevel@tonic-gate pgcnt = page_get_pagecnt(szc); 33747c478bd9Sstevel@tonic-gate } 33757c478bd9Sstevel@tonic-gate while (lastpp != firstpp) { 33767c478bd9Sstevel@tonic-gate ASSERT(lastpp->p_szc == pszc); 33777c478bd9Sstevel@tonic-gate lastpp->p_szc = szc; 33787c478bd9Sstevel@tonic-gate lastpp--; 33797c478bd9Sstevel@tonic-gate } 33807c478bd9Sstevel@tonic-gate firstpp->p_szc = szc; 33817c478bd9Sstevel@tonic-gate if (firstpp == rootpp) 33827c478bd9Sstevel@tonic-gate break; 33837c478bd9Sstevel@tonic-gate szc++; 33847c478bd9Sstevel@tonic-gate } 33857c478bd9Sstevel@tonic-gate x86_hm_exit(rootpp); 33867c478bd9Sstevel@tonic-gate } 33877c478bd9Sstevel@tonic-gate 33887c478bd9Sstevel@tonic-gate /* 33897c478bd9Sstevel@tonic-gate * get hw stats from hardware into page struct and reset hw stats 33907c478bd9Sstevel@tonic-gate * returns attributes of page 33917c478bd9Sstevel@tonic-gate * Flags for hat_pagesync, hat_getstat, hat_sync 33927c478bd9Sstevel@tonic-gate * 33937c478bd9Sstevel@tonic-gate * define HAT_SYNC_ZERORM 0x01 33947c478bd9Sstevel@tonic-gate * 33957c478bd9Sstevel@tonic-gate * Additional flags for hat_pagesync 33967c478bd9Sstevel@tonic-gate * 33977c478bd9Sstevel@tonic-gate * define HAT_SYNC_STOPON_REF 0x02 33987c478bd9Sstevel@tonic-gate * define HAT_SYNC_STOPON_MOD 0x04 33997c478bd9Sstevel@tonic-gate * define HAT_SYNC_STOPON_RM 0x06 34007c478bd9Sstevel@tonic-gate * define HAT_SYNC_STOPON_SHARED 0x08 34017c478bd9Sstevel@tonic-gate */ 34027c478bd9Sstevel@tonic-gate uint_t 34037c478bd9Sstevel@tonic-gate hat_pagesync(struct page *pp, uint_t flags) 34047c478bd9Sstevel@tonic-gate { 34057c478bd9Sstevel@tonic-gate hment_t *hm = NULL; 34067c478bd9Sstevel@tonic-gate htable_t *ht; 34077c478bd9Sstevel@tonic-gate uint_t entry; 34087c478bd9Sstevel@tonic-gate x86pte_t old, save_old; 34097c478bd9Sstevel@tonic-gate x86pte_t new; 34107c478bd9Sstevel@tonic-gate uchar_t nrmbits = P_REF|P_MOD|P_RO; 34117c478bd9Sstevel@tonic-gate extern ulong_t po_share; 34127c478bd9Sstevel@tonic-gate page_t *save_pp = pp; 34137c478bd9Sstevel@tonic-gate uint_t pszc = 0; 34147c478bd9Sstevel@tonic-gate 34157c478bd9Sstevel@tonic-gate ASSERT(PAGE_LOCKED(pp) || panicstr); 34167c478bd9Sstevel@tonic-gate 34177c478bd9Sstevel@tonic-gate if (PP_ISRO(pp) && (flags & HAT_SYNC_STOPON_MOD)) 34187c478bd9Sstevel@tonic-gate return (pp->p_nrm & nrmbits); 34197c478bd9Sstevel@tonic-gate 34207c478bd9Sstevel@tonic-gate if ((flags & HAT_SYNC_ZERORM) == 0) { 34217c478bd9Sstevel@tonic-gate 34227c478bd9Sstevel@tonic-gate if ((flags & HAT_SYNC_STOPON_REF) != 0 && PP_ISREF(pp)) 34237c478bd9Sstevel@tonic-gate return (pp->p_nrm & nrmbits); 34247c478bd9Sstevel@tonic-gate 34257c478bd9Sstevel@tonic-gate if ((flags & HAT_SYNC_STOPON_MOD) != 0 && PP_ISMOD(pp)) 34267c478bd9Sstevel@tonic-gate return (pp->p_nrm & nrmbits); 34277c478bd9Sstevel@tonic-gate 34287c478bd9Sstevel@tonic-gate if ((flags & HAT_SYNC_STOPON_SHARED) != 0 && 34297c478bd9Sstevel@tonic-gate hat_page_getshare(pp) > po_share) { 34307c478bd9Sstevel@tonic-gate if (PP_ISRO(pp)) 34317c478bd9Sstevel@tonic-gate PP_SETREF(pp); 34327c478bd9Sstevel@tonic-gate return (pp->p_nrm & nrmbits); 34337c478bd9Sstevel@tonic-gate } 34347c478bd9Sstevel@tonic-gate } 34357c478bd9Sstevel@tonic-gate 34367c478bd9Sstevel@tonic-gate next_size: 34377c478bd9Sstevel@tonic-gate /* 34387c478bd9Sstevel@tonic-gate * walk thru the mapping list syncing (and clearing) ref/mod bits. 34397c478bd9Sstevel@tonic-gate */ 34407c478bd9Sstevel@tonic-gate x86_hm_enter(pp); 34417c478bd9Sstevel@tonic-gate while ((hm = hment_walk(pp, &ht, &entry, hm)) != NULL) { 34427c478bd9Sstevel@tonic-gate if (ht->ht_level < pszc) 34437c478bd9Sstevel@tonic-gate continue; 34447c478bd9Sstevel@tonic-gate old = x86pte_get(ht, entry); 34457c478bd9Sstevel@tonic-gate try_again: 34467c478bd9Sstevel@tonic-gate 34477c478bd9Sstevel@tonic-gate ASSERT(PTE2PFN(old, ht->ht_level) == pp->p_pagenum); 34487c478bd9Sstevel@tonic-gate 34497c478bd9Sstevel@tonic-gate if (PTE_GET(old, PT_REF | PT_MOD) == 0) 34507c478bd9Sstevel@tonic-gate continue; 34517c478bd9Sstevel@tonic-gate 34527c478bd9Sstevel@tonic-gate save_old = old; 34537c478bd9Sstevel@tonic-gate if ((flags & HAT_SYNC_ZERORM) != 0) { 34547c478bd9Sstevel@tonic-gate 34557c478bd9Sstevel@tonic-gate /* 34567c478bd9Sstevel@tonic-gate * Need to clear ref or mod bits. Need to demap 34577c478bd9Sstevel@tonic-gate * to make sure any executing TLBs see cleared bits. 34587c478bd9Sstevel@tonic-gate */ 34597c478bd9Sstevel@tonic-gate new = old; 34607c478bd9Sstevel@tonic-gate PTE_CLR(new, PT_REF | PT_MOD); 34617c478bd9Sstevel@tonic-gate old = hati_update_pte(ht, entry, old, new); 34627c478bd9Sstevel@tonic-gate if (old != 0) 34637c478bd9Sstevel@tonic-gate goto try_again; 34647c478bd9Sstevel@tonic-gate 34657c478bd9Sstevel@tonic-gate old = save_old; 34667c478bd9Sstevel@tonic-gate } 34677c478bd9Sstevel@tonic-gate 34687c478bd9Sstevel@tonic-gate /* 34697c478bd9Sstevel@tonic-gate * Sync the PTE 34707c478bd9Sstevel@tonic-gate */ 3471ae115bc7Smrj if (!(flags & HAT_SYNC_ZERORM) && 3472ae115bc7Smrj PTE_GET(old, PT_SOFTWARE) <= PT_NOSYNC) 34737c478bd9Sstevel@tonic-gate hati_sync_pte_to_page(pp, old, ht->ht_level); 34747c478bd9Sstevel@tonic-gate 34757c478bd9Sstevel@tonic-gate /* 34767c478bd9Sstevel@tonic-gate * can stop short if we found a ref'd or mod'd page 34777c478bd9Sstevel@tonic-gate */ 34787c478bd9Sstevel@tonic-gate if ((flags & HAT_SYNC_STOPON_MOD) && PP_ISMOD(save_pp) || 34797c478bd9Sstevel@tonic-gate (flags & HAT_SYNC_STOPON_REF) && PP_ISREF(save_pp)) { 34807c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 3481ae115bc7Smrj goto done; 34827c478bd9Sstevel@tonic-gate } 34837c478bd9Sstevel@tonic-gate } 34847c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 34857c478bd9Sstevel@tonic-gate while (pszc < pp->p_szc) { 34867c478bd9Sstevel@tonic-gate page_t *tpp; 34877c478bd9Sstevel@tonic-gate pszc++; 34887c478bd9Sstevel@tonic-gate tpp = PP_GROUPLEADER(pp, pszc); 34897c478bd9Sstevel@tonic-gate if (pp != tpp) { 34907c478bd9Sstevel@tonic-gate pp = tpp; 34917c478bd9Sstevel@tonic-gate goto next_size; 34927c478bd9Sstevel@tonic-gate } 34937c478bd9Sstevel@tonic-gate } 3494ae115bc7Smrj done: 34957c478bd9Sstevel@tonic-gate return (save_pp->p_nrm & nrmbits); 34967c478bd9Sstevel@tonic-gate } 34977c478bd9Sstevel@tonic-gate 34987c478bd9Sstevel@tonic-gate /* 34997c478bd9Sstevel@tonic-gate * returns approx number of mappings to this pp. A return of 0 implies 35007c478bd9Sstevel@tonic-gate * there are no mappings to the page. 35017c478bd9Sstevel@tonic-gate */ 35027c478bd9Sstevel@tonic-gate ulong_t 35037c478bd9Sstevel@tonic-gate hat_page_getshare(page_t *pp) 35047c478bd9Sstevel@tonic-gate { 35057c478bd9Sstevel@tonic-gate uint_t cnt; 35067c478bd9Sstevel@tonic-gate cnt = hment_mapcnt(pp); 3507a5652762Spraks #if defined(__amd64) 3508a5652762Spraks if (vpm_enable && pp->p_vpmref) { 3509a5652762Spraks cnt += 1; 3510a5652762Spraks } 3511a5652762Spraks #endif 35127c478bd9Sstevel@tonic-gate return (cnt); 35137c478bd9Sstevel@tonic-gate } 35147c478bd9Sstevel@tonic-gate 35157c478bd9Sstevel@tonic-gate /* 35167c478bd9Sstevel@tonic-gate * hat_softlock isn't supported anymore 35177c478bd9Sstevel@tonic-gate */ 35187c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 35197c478bd9Sstevel@tonic-gate faultcode_t 35207c478bd9Sstevel@tonic-gate hat_softlock( 35217c478bd9Sstevel@tonic-gate hat_t *hat, 35227c478bd9Sstevel@tonic-gate caddr_t addr, 35237c478bd9Sstevel@tonic-gate size_t *len, 35247c478bd9Sstevel@tonic-gate struct page **page_array, 35257c478bd9Sstevel@tonic-gate uint_t flags) 35267c478bd9Sstevel@tonic-gate { 35277c478bd9Sstevel@tonic-gate return (FC_NOSUPPORT); 35287c478bd9Sstevel@tonic-gate } 35297c478bd9Sstevel@tonic-gate 35307c478bd9Sstevel@tonic-gate 35317c478bd9Sstevel@tonic-gate 35327c478bd9Sstevel@tonic-gate /* 35337c478bd9Sstevel@tonic-gate * Routine to expose supported HAT features to platform independent code. 35347c478bd9Sstevel@tonic-gate */ 35357c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 35367c478bd9Sstevel@tonic-gate int 35377c478bd9Sstevel@tonic-gate hat_supported(enum hat_features feature, void *arg) 35387c478bd9Sstevel@tonic-gate { 35397c478bd9Sstevel@tonic-gate switch (feature) { 35407c478bd9Sstevel@tonic-gate 35417c478bd9Sstevel@tonic-gate case HAT_SHARED_PT: /* this is really ISM */ 35427c478bd9Sstevel@tonic-gate return (1); 35437c478bd9Sstevel@tonic-gate 35447c478bd9Sstevel@tonic-gate case HAT_DYNAMIC_ISM_UNMAP: 35457c478bd9Sstevel@tonic-gate return (0); 35467c478bd9Sstevel@tonic-gate 35477c478bd9Sstevel@tonic-gate case HAT_VMODSORT: 35487c478bd9Sstevel@tonic-gate return (1); 35497c478bd9Sstevel@tonic-gate 35507c478bd9Sstevel@tonic-gate default: 35517c478bd9Sstevel@tonic-gate panic("hat_supported() - unknown feature"); 35527c478bd9Sstevel@tonic-gate } 35537c478bd9Sstevel@tonic-gate return (0); 35547c478bd9Sstevel@tonic-gate } 35557c478bd9Sstevel@tonic-gate 35567c478bd9Sstevel@tonic-gate /* 35577c478bd9Sstevel@tonic-gate * Called when a thread is exiting and has been switched to the kernel AS 35587c478bd9Sstevel@tonic-gate */ 35597c478bd9Sstevel@tonic-gate void 35607c478bd9Sstevel@tonic-gate hat_thread_exit(kthread_t *thd) 35617c478bd9Sstevel@tonic-gate { 35627c478bd9Sstevel@tonic-gate ASSERT(thd->t_procp->p_as == &kas); 35637c478bd9Sstevel@tonic-gate hat_switch(thd->t_procp->p_as->a_hat); 35647c478bd9Sstevel@tonic-gate } 35657c478bd9Sstevel@tonic-gate 35667c478bd9Sstevel@tonic-gate /* 35677c478bd9Sstevel@tonic-gate * Setup the given brand new hat structure as the new HAT on this cpu's mmu. 35687c478bd9Sstevel@tonic-gate */ 35697c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 35707c478bd9Sstevel@tonic-gate void 35717c478bd9Sstevel@tonic-gate hat_setup(hat_t *hat, int flags) 35727c478bd9Sstevel@tonic-gate { 35737c478bd9Sstevel@tonic-gate kpreempt_disable(); 35747c478bd9Sstevel@tonic-gate 35757c478bd9Sstevel@tonic-gate hat_switch(hat); 35767c478bd9Sstevel@tonic-gate 35777c478bd9Sstevel@tonic-gate kpreempt_enable(); 35787c478bd9Sstevel@tonic-gate } 35797c478bd9Sstevel@tonic-gate 35807c478bd9Sstevel@tonic-gate /* 35817c478bd9Sstevel@tonic-gate * Prepare for a CPU private mapping for the given address. 35827c478bd9Sstevel@tonic-gate * 35837c478bd9Sstevel@tonic-gate * The address can only be used from a single CPU and can be remapped 35847c478bd9Sstevel@tonic-gate * using hat_mempte_remap(). Return the address of the PTE. 35857c478bd9Sstevel@tonic-gate * 35867c478bd9Sstevel@tonic-gate * We do the htable_create() if necessary and increment the valid count so 35877c478bd9Sstevel@tonic-gate * the htable can't disappear. We also hat_devload() the page table into 35887c478bd9Sstevel@tonic-gate * kernel so that the PTE is quickly accessed. 35897c478bd9Sstevel@tonic-gate */ 3590ae115bc7Smrj hat_mempte_t 3591ae115bc7Smrj hat_mempte_setup(caddr_t addr) 35927c478bd9Sstevel@tonic-gate { 35937c478bd9Sstevel@tonic-gate uintptr_t va = (uintptr_t)addr; 35947c478bd9Sstevel@tonic-gate htable_t *ht; 35957c478bd9Sstevel@tonic-gate uint_t entry; 35967c478bd9Sstevel@tonic-gate x86pte_t oldpte; 3597ae115bc7Smrj hat_mempte_t p; 35987c478bd9Sstevel@tonic-gate 35997c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(va)); 36007c478bd9Sstevel@tonic-gate ASSERT(!IN_VA_HOLE(va)); 3601aac11643Sjosephb ++curthread->t_hatdepth; 36027c478bd9Sstevel@tonic-gate ht = htable_getpte(kas.a_hat, va, &entry, &oldpte, 0); 36037c478bd9Sstevel@tonic-gate if (ht == NULL) { 36047c478bd9Sstevel@tonic-gate ht = htable_create(kas.a_hat, va, 0, NULL); 36057c478bd9Sstevel@tonic-gate entry = htable_va2entry(va, ht); 36067c478bd9Sstevel@tonic-gate ASSERT(ht->ht_level == 0); 36077c478bd9Sstevel@tonic-gate oldpte = x86pte_get(ht, entry); 36087c478bd9Sstevel@tonic-gate } 36097c478bd9Sstevel@tonic-gate if (PTE_ISVALID(oldpte)) 36107c478bd9Sstevel@tonic-gate panic("hat_mempte_setup(): address already mapped" 36117c478bd9Sstevel@tonic-gate "ht=%p, entry=%d, pte=" FMT_PTE, ht, entry, oldpte); 36127c478bd9Sstevel@tonic-gate 36137c478bd9Sstevel@tonic-gate /* 36147c478bd9Sstevel@tonic-gate * increment ht_valid_cnt so that the pagetable can't disappear 36157c478bd9Sstevel@tonic-gate */ 36167c478bd9Sstevel@tonic-gate HTABLE_INC(ht->ht_valid_cnt); 36177c478bd9Sstevel@tonic-gate 36187c478bd9Sstevel@tonic-gate /* 3619ae115bc7Smrj * return the PTE physical address to the caller. 36207c478bd9Sstevel@tonic-gate */ 36217c478bd9Sstevel@tonic-gate htable_release(ht); 3622ae115bc7Smrj p = PT_INDEX_PHYSADDR(pfn_to_pa(ht->ht_pfn), entry); 3623aac11643Sjosephb --curthread->t_hatdepth; 3624ae115bc7Smrj return (p); 36257c478bd9Sstevel@tonic-gate } 36267c478bd9Sstevel@tonic-gate 36277c478bd9Sstevel@tonic-gate /* 36287c478bd9Sstevel@tonic-gate * Release a CPU private mapping for the given address. 36297c478bd9Sstevel@tonic-gate * We decrement the htable valid count so it might be destroyed. 36307c478bd9Sstevel@tonic-gate */ 3631ae115bc7Smrj /*ARGSUSED1*/ 36327c478bd9Sstevel@tonic-gate void 3633ae115bc7Smrj hat_mempte_release(caddr_t addr, hat_mempte_t pte_pa) 36347c478bd9Sstevel@tonic-gate { 36357c478bd9Sstevel@tonic-gate htable_t *ht; 36367c478bd9Sstevel@tonic-gate 36377c478bd9Sstevel@tonic-gate /* 3638ae115bc7Smrj * invalidate any left over mapping and decrement the htable valid count 36397c478bd9Sstevel@tonic-gate */ 3640ae115bc7Smrj { 3641ae115bc7Smrj x86pte_t *pteptr; 3642ae115bc7Smrj 3643ae115bc7Smrj pteptr = x86pte_mapin(mmu_btop(pte_pa), 3644ae115bc7Smrj (pte_pa & MMU_PAGEOFFSET) >> mmu.pte_size_shift, NULL); 3645ae115bc7Smrj if (mmu.pae_hat) 3646ae115bc7Smrj *pteptr = 0; 3647ae115bc7Smrj else 3648ae115bc7Smrj *(x86pte32_t *)pteptr = 0; 3649ae115bc7Smrj mmu_tlbflush_entry(addr); 3650ae115bc7Smrj x86pte_mapout(); 3651ae115bc7Smrj } 3652ae115bc7Smrj 36537c478bd9Sstevel@tonic-gate ht = htable_getpte(kas.a_hat, ALIGN2PAGE(addr), NULL, NULL, 0); 36547c478bd9Sstevel@tonic-gate if (ht == NULL) 36557c478bd9Sstevel@tonic-gate panic("hat_mempte_release(): invalid address"); 36567c478bd9Sstevel@tonic-gate ASSERT(ht->ht_level == 0); 36577c478bd9Sstevel@tonic-gate HTABLE_DEC(ht->ht_valid_cnt); 36587c478bd9Sstevel@tonic-gate htable_release(ht); 36597c478bd9Sstevel@tonic-gate } 36607c478bd9Sstevel@tonic-gate 36617c478bd9Sstevel@tonic-gate /* 36627c478bd9Sstevel@tonic-gate * Apply a temporary CPU private mapping to a page. We flush the TLB only 36637c478bd9Sstevel@tonic-gate * on this CPU, so this ought to have been called with preemption disabled. 36647c478bd9Sstevel@tonic-gate */ 36657c478bd9Sstevel@tonic-gate void 36667c478bd9Sstevel@tonic-gate hat_mempte_remap( 3667ae115bc7Smrj pfn_t pfn, 3668ae115bc7Smrj caddr_t addr, 3669ae115bc7Smrj hat_mempte_t pte_pa, 3670ae115bc7Smrj uint_t attr, 3671ae115bc7Smrj uint_t flags) 36727c478bd9Sstevel@tonic-gate { 36737c478bd9Sstevel@tonic-gate uintptr_t va = (uintptr_t)addr; 36747c478bd9Sstevel@tonic-gate x86pte_t pte; 36757c478bd9Sstevel@tonic-gate 36767c478bd9Sstevel@tonic-gate /* 36777c478bd9Sstevel@tonic-gate * Remap the given PTE to the new page's PFN. Invalidate only 36787c478bd9Sstevel@tonic-gate * on this CPU. 36797c478bd9Sstevel@tonic-gate */ 36807c478bd9Sstevel@tonic-gate #ifdef DEBUG 36817c478bd9Sstevel@tonic-gate htable_t *ht; 36827c478bd9Sstevel@tonic-gate uint_t entry; 36837c478bd9Sstevel@tonic-gate 36847c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(va)); 36857c478bd9Sstevel@tonic-gate ASSERT(!IN_VA_HOLE(va)); 36867c478bd9Sstevel@tonic-gate ht = htable_getpte(kas.a_hat, va, &entry, NULL, 0); 36877c478bd9Sstevel@tonic-gate ASSERT(ht != NULL); 36887c478bd9Sstevel@tonic-gate ASSERT(ht->ht_level == 0); 36897c478bd9Sstevel@tonic-gate ASSERT(ht->ht_valid_cnt > 0); 3690ae115bc7Smrj ASSERT(ht->ht_pfn == mmu_btop(pte_pa)); 36917c478bd9Sstevel@tonic-gate htable_release(ht); 36927c478bd9Sstevel@tonic-gate #endif 36937c478bd9Sstevel@tonic-gate pte = hati_mkpte(pfn, attr, 0, flags); 3694ae115bc7Smrj { 3695ae115bc7Smrj x86pte_t *pteptr; 3696ae115bc7Smrj 3697ae115bc7Smrj pteptr = x86pte_mapin(mmu_btop(pte_pa), 3698ae115bc7Smrj (pte_pa & MMU_PAGEOFFSET) >> mmu.pte_size_shift, NULL); 3699ae115bc7Smrj if (mmu.pae_hat) 3700ae115bc7Smrj *(x86pte_t *)pteptr = pte; 3701ae115bc7Smrj else 3702ae115bc7Smrj *(x86pte32_t *)pteptr = (x86pte32_t)pte; 3703ae115bc7Smrj mmu_tlbflush_entry(addr); 3704ae115bc7Smrj x86pte_mapout(); 3705ae115bc7Smrj } 37067c478bd9Sstevel@tonic-gate } 37077c478bd9Sstevel@tonic-gate 37087c478bd9Sstevel@tonic-gate 37097c478bd9Sstevel@tonic-gate 37107c478bd9Sstevel@tonic-gate /* 37117c478bd9Sstevel@tonic-gate * Hat locking functions 37127c478bd9Sstevel@tonic-gate * XXX - these two functions are currently being used by hatstats 37137c478bd9Sstevel@tonic-gate * they can be removed by using a per-as mutex for hatstats. 37147c478bd9Sstevel@tonic-gate */ 37157c478bd9Sstevel@tonic-gate void 37167c478bd9Sstevel@tonic-gate hat_enter(hat_t *hat) 37177c478bd9Sstevel@tonic-gate { 37187c478bd9Sstevel@tonic-gate mutex_enter(&hat->hat_mutex); 37197c478bd9Sstevel@tonic-gate } 37207c478bd9Sstevel@tonic-gate 37217c478bd9Sstevel@tonic-gate void 37227c478bd9Sstevel@tonic-gate hat_exit(hat_t *hat) 37237c478bd9Sstevel@tonic-gate { 37247c478bd9Sstevel@tonic-gate mutex_exit(&hat->hat_mutex); 37257c478bd9Sstevel@tonic-gate } 37267c478bd9Sstevel@tonic-gate 37277c478bd9Sstevel@tonic-gate /* 3728ae115bc7Smrj * HAT part of cpu initialization. 37297c478bd9Sstevel@tonic-gate */ 37307c478bd9Sstevel@tonic-gate void 3731ae115bc7Smrj hat_cpu_online(struct cpu *cpup) 37327c478bd9Sstevel@tonic-gate { 3733ae115bc7Smrj if (cpup != CPU) { 3734ae115bc7Smrj x86pte_cpu_init(cpup); 3735ae115bc7Smrj hat_vlp_setup(cpup); 37367c478bd9Sstevel@tonic-gate } 3737ae115bc7Smrj CPUSET_ATOMIC_ADD(khat_cpuset, cpup->cpu_id); 37387c478bd9Sstevel@tonic-gate } 37397c478bd9Sstevel@tonic-gate 37407c478bd9Sstevel@tonic-gate /* 3741ae115bc7Smrj * HAT part of cpu deletion. 3742ae115bc7Smrj * (currently, we only call this after the cpu is safely passivated.) 37437c478bd9Sstevel@tonic-gate */ 37447c478bd9Sstevel@tonic-gate void 3745ae115bc7Smrj hat_cpu_offline(struct cpu *cpup) 37467c478bd9Sstevel@tonic-gate { 3747ae115bc7Smrj ASSERT(cpup != CPU); 3748ae115bc7Smrj 3749ae115bc7Smrj CPUSET_ATOMIC_DEL(khat_cpuset, cpup->cpu_id); 3750ae115bc7Smrj x86pte_cpu_fini(cpup); 3751ae115bc7Smrj hat_vlp_teardown(cpup); 37527c478bd9Sstevel@tonic-gate } 37537c478bd9Sstevel@tonic-gate 37547c478bd9Sstevel@tonic-gate /* 37557c478bd9Sstevel@tonic-gate * Function called after all CPUs are brought online. 37567c478bd9Sstevel@tonic-gate * Used to remove low address boot mappings. 37577c478bd9Sstevel@tonic-gate */ 37587c478bd9Sstevel@tonic-gate void 37597c478bd9Sstevel@tonic-gate clear_boot_mappings(uintptr_t low, uintptr_t high) 37607c478bd9Sstevel@tonic-gate { 37617c478bd9Sstevel@tonic-gate uintptr_t vaddr = low; 37627c478bd9Sstevel@tonic-gate htable_t *ht = NULL; 37637c478bd9Sstevel@tonic-gate level_t level; 37647c478bd9Sstevel@tonic-gate uint_t entry; 37657c478bd9Sstevel@tonic-gate x86pte_t pte; 37667c478bd9Sstevel@tonic-gate 37677c478bd9Sstevel@tonic-gate /* 37687c478bd9Sstevel@tonic-gate * On 1st CPU we can unload the prom mappings, basically we blow away 3769ae115bc7Smrj * all virtual mappings under _userlimit. 37707c478bd9Sstevel@tonic-gate */ 37717c478bd9Sstevel@tonic-gate while (vaddr < high) { 37727c478bd9Sstevel@tonic-gate pte = htable_walk(kas.a_hat, &ht, &vaddr, high); 37737c478bd9Sstevel@tonic-gate if (ht == NULL) 37747c478bd9Sstevel@tonic-gate break; 37757c478bd9Sstevel@tonic-gate 37767c478bd9Sstevel@tonic-gate level = ht->ht_level; 37777c478bd9Sstevel@tonic-gate entry = htable_va2entry(vaddr, ht); 37787c478bd9Sstevel@tonic-gate ASSERT(level <= mmu.max_page_level); 37797c478bd9Sstevel@tonic-gate ASSERT(PTE_ISPAGE(pte, level)); 37807c478bd9Sstevel@tonic-gate 37817c478bd9Sstevel@tonic-gate /* 37827c478bd9Sstevel@tonic-gate * Unload the mapping from the page tables. 37837c478bd9Sstevel@tonic-gate */ 3784ae115bc7Smrj (void) x86pte_inval(ht, entry, 0, NULL); 37857c478bd9Sstevel@tonic-gate ASSERT(ht->ht_valid_cnt > 0); 37867c478bd9Sstevel@tonic-gate HTABLE_DEC(ht->ht_valid_cnt); 37877c478bd9Sstevel@tonic-gate PGCNT_DEC(ht->ht_hat, ht->ht_level); 37887c478bd9Sstevel@tonic-gate 37897c478bd9Sstevel@tonic-gate vaddr += LEVEL_SIZE(ht->ht_level); 37907c478bd9Sstevel@tonic-gate } 37917c478bd9Sstevel@tonic-gate if (ht) 37927c478bd9Sstevel@tonic-gate htable_release(ht); 37937c478bd9Sstevel@tonic-gate } 37947c478bd9Sstevel@tonic-gate 37957c478bd9Sstevel@tonic-gate /* 37967c478bd9Sstevel@tonic-gate * Atomically update a new translation for a single page. If the 37977c478bd9Sstevel@tonic-gate * currently installed PTE doesn't match the value we expect to find, 37987c478bd9Sstevel@tonic-gate * it's not updated and we return the PTE we found. 37997c478bd9Sstevel@tonic-gate * 38007c478bd9Sstevel@tonic-gate * If activating nosync or NOWRITE and the page was modified we need to sync 38017c478bd9Sstevel@tonic-gate * with the page_t. Also sync with page_t if clearing ref/mod bits. 38027c478bd9Sstevel@tonic-gate */ 38037c478bd9Sstevel@tonic-gate static x86pte_t 38047c478bd9Sstevel@tonic-gate hati_update_pte(htable_t *ht, uint_t entry, x86pte_t expected, x86pte_t new) 38057c478bd9Sstevel@tonic-gate { 38067c478bd9Sstevel@tonic-gate page_t *pp; 38077c478bd9Sstevel@tonic-gate uint_t rm = 0; 38087c478bd9Sstevel@tonic-gate x86pte_t replaced; 38097c478bd9Sstevel@tonic-gate 3810ae115bc7Smrj if (PTE_GET(expected, PT_SOFTWARE) < PT_NOSYNC && 38117c478bd9Sstevel@tonic-gate PTE_GET(expected, PT_MOD | PT_REF) && 38127c478bd9Sstevel@tonic-gate (PTE_GET(new, PT_NOSYNC) || !PTE_GET(new, PT_WRITABLE) || 3813*250b7ff9Sjosephb !PTE_GET(new, PT_MOD | PT_REF))) { 38147c478bd9Sstevel@tonic-gate 3815ae115bc7Smrj ASSERT(!pfn_is_foreign(PTE2PFN(expected, ht->ht_level))); 38167c478bd9Sstevel@tonic-gate pp = page_numtopp_nolock(PTE2PFN(expected, ht->ht_level)); 38177c478bd9Sstevel@tonic-gate ASSERT(pp != NULL); 38187c478bd9Sstevel@tonic-gate if (PTE_GET(expected, PT_MOD)) 38197c478bd9Sstevel@tonic-gate rm |= P_MOD; 38207c478bd9Sstevel@tonic-gate if (PTE_GET(expected, PT_REF)) 38217c478bd9Sstevel@tonic-gate rm |= P_REF; 38227c478bd9Sstevel@tonic-gate PTE_CLR(new, PT_MOD | PT_REF); 38237c478bd9Sstevel@tonic-gate } 38247c478bd9Sstevel@tonic-gate 38257c478bd9Sstevel@tonic-gate replaced = x86pte_update(ht, entry, expected, new); 38267c478bd9Sstevel@tonic-gate if (replaced != expected) 38277c478bd9Sstevel@tonic-gate return (replaced); 38287c478bd9Sstevel@tonic-gate 38297c478bd9Sstevel@tonic-gate if (rm) { 38307c478bd9Sstevel@tonic-gate /* 38317c478bd9Sstevel@tonic-gate * sync to all constituent pages of a large page 38327c478bd9Sstevel@tonic-gate */ 38337c478bd9Sstevel@tonic-gate pgcnt_t pgcnt = page_get_pagecnt(ht->ht_level); 38347c478bd9Sstevel@tonic-gate ASSERT(IS_P2ALIGNED(pp->p_pagenum, pgcnt)); 38357c478bd9Sstevel@tonic-gate while (pgcnt-- > 0) { 38367c478bd9Sstevel@tonic-gate /* 38377c478bd9Sstevel@tonic-gate * hat_page_demote() can't decrease 38387c478bd9Sstevel@tonic-gate * pszc below this mapping size 38397c478bd9Sstevel@tonic-gate * since large mapping existed after we 38407c478bd9Sstevel@tonic-gate * took mlist lock. 38417c478bd9Sstevel@tonic-gate */ 38427c478bd9Sstevel@tonic-gate ASSERT(pp->p_szc >= ht->ht_level); 38437c478bd9Sstevel@tonic-gate hat_page_setattr(pp, rm); 38447c478bd9Sstevel@tonic-gate ++pp; 38457c478bd9Sstevel@tonic-gate } 38467c478bd9Sstevel@tonic-gate } 38477c478bd9Sstevel@tonic-gate 38487c478bd9Sstevel@tonic-gate return (0); 38497c478bd9Sstevel@tonic-gate } 38507c478bd9Sstevel@tonic-gate 38517c478bd9Sstevel@tonic-gate /* 38527c478bd9Sstevel@tonic-gate * Kernel Physical Mapping (kpm) facility 38537c478bd9Sstevel@tonic-gate * 38547c478bd9Sstevel@tonic-gate * Most of the routines needed to support segkpm are almost no-ops on the 38557c478bd9Sstevel@tonic-gate * x86 platform. We map in the entire segment when it is created and leave 38567c478bd9Sstevel@tonic-gate * it mapped in, so there is no additional work required to set up and tear 38577c478bd9Sstevel@tonic-gate * down individual mappings. All of these routines were created to support 38587c478bd9Sstevel@tonic-gate * SPARC platforms that have to avoid aliasing in their virtually indexed 38597c478bd9Sstevel@tonic-gate * caches. 38607c478bd9Sstevel@tonic-gate * 38617c478bd9Sstevel@tonic-gate * Most of the routines have sanity checks in them (e.g. verifying that the 38627c478bd9Sstevel@tonic-gate * passed-in page is locked). We don't actually care about most of these 38637c478bd9Sstevel@tonic-gate * checks on x86, but we leave them in place to identify problems in the 38647c478bd9Sstevel@tonic-gate * upper levels. 38657c478bd9Sstevel@tonic-gate */ 38667c478bd9Sstevel@tonic-gate 38677c478bd9Sstevel@tonic-gate /* 38687c478bd9Sstevel@tonic-gate * Map in a locked page and return the vaddr. 38697c478bd9Sstevel@tonic-gate */ 38707c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 38717c478bd9Sstevel@tonic-gate caddr_t 38727c478bd9Sstevel@tonic-gate hat_kpm_mapin(struct page *pp, struct kpme *kpme) 38737c478bd9Sstevel@tonic-gate { 38747c478bd9Sstevel@tonic-gate caddr_t vaddr; 38757c478bd9Sstevel@tonic-gate 38767c478bd9Sstevel@tonic-gate #ifdef DEBUG 38777c478bd9Sstevel@tonic-gate if (kpm_enable == 0) { 38787c478bd9Sstevel@tonic-gate cmn_err(CE_WARN, "hat_kpm_mapin: kpm_enable not set\n"); 38797c478bd9Sstevel@tonic-gate return ((caddr_t)NULL); 38807c478bd9Sstevel@tonic-gate } 38817c478bd9Sstevel@tonic-gate 38827c478bd9Sstevel@tonic-gate if (pp == NULL || PAGE_LOCKED(pp) == 0) { 38837c478bd9Sstevel@tonic-gate cmn_err(CE_WARN, "hat_kpm_mapin: pp zero or not locked\n"); 38847c478bd9Sstevel@tonic-gate return ((caddr_t)NULL); 38857c478bd9Sstevel@tonic-gate } 38867c478bd9Sstevel@tonic-gate #endif 38877c478bd9Sstevel@tonic-gate 38887c478bd9Sstevel@tonic-gate vaddr = hat_kpm_page2va(pp, 1); 38897c478bd9Sstevel@tonic-gate 38907c478bd9Sstevel@tonic-gate return (vaddr); 38917c478bd9Sstevel@tonic-gate } 38927c478bd9Sstevel@tonic-gate 38937c478bd9Sstevel@tonic-gate /* 38947c478bd9Sstevel@tonic-gate * Mapout a locked page. 38957c478bd9Sstevel@tonic-gate */ 38967c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 38977c478bd9Sstevel@tonic-gate void 38987c478bd9Sstevel@tonic-gate hat_kpm_mapout(struct page *pp, struct kpme *kpme, caddr_t vaddr) 38997c478bd9Sstevel@tonic-gate { 39007c478bd9Sstevel@tonic-gate #ifdef DEBUG 39017c478bd9Sstevel@tonic-gate if (kpm_enable == 0) { 39027c478bd9Sstevel@tonic-gate cmn_err(CE_WARN, "hat_kpm_mapout: kpm_enable not set\n"); 39037c478bd9Sstevel@tonic-gate return; 39047c478bd9Sstevel@tonic-gate } 39057c478bd9Sstevel@tonic-gate 39067c478bd9Sstevel@tonic-gate if (IS_KPM_ADDR(vaddr) == 0) { 39077c478bd9Sstevel@tonic-gate cmn_err(CE_WARN, "hat_kpm_mapout: no kpm address\n"); 39087c478bd9Sstevel@tonic-gate return; 39097c478bd9Sstevel@tonic-gate } 39107c478bd9Sstevel@tonic-gate 39117c478bd9Sstevel@tonic-gate if (pp == NULL || PAGE_LOCKED(pp) == 0) { 39127c478bd9Sstevel@tonic-gate cmn_err(CE_WARN, "hat_kpm_mapout: page zero or not locked\n"); 39137c478bd9Sstevel@tonic-gate return; 39147c478bd9Sstevel@tonic-gate } 39157c478bd9Sstevel@tonic-gate #endif 39167c478bd9Sstevel@tonic-gate } 39177c478bd9Sstevel@tonic-gate 39187c478bd9Sstevel@tonic-gate /* 39197c478bd9Sstevel@tonic-gate * Return the kpm virtual address for a specific pfn 39207c478bd9Sstevel@tonic-gate */ 39217c478bd9Sstevel@tonic-gate caddr_t 39227c478bd9Sstevel@tonic-gate hat_kpm_pfn2va(pfn_t pfn) 39237c478bd9Sstevel@tonic-gate { 3924ae115bc7Smrj uintptr_t vaddr = (uintptr_t)kpm_vbase + mmu_ptob(pfn); 39257c478bd9Sstevel@tonic-gate 39267c478bd9Sstevel@tonic-gate return ((caddr_t)vaddr); 39277c478bd9Sstevel@tonic-gate } 39287c478bd9Sstevel@tonic-gate 39297c478bd9Sstevel@tonic-gate /* 39307c478bd9Sstevel@tonic-gate * Return the kpm virtual address for the page at pp. 39317c478bd9Sstevel@tonic-gate */ 39327c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 39337c478bd9Sstevel@tonic-gate caddr_t 39347c478bd9Sstevel@tonic-gate hat_kpm_page2va(struct page *pp, int checkswap) 39357c478bd9Sstevel@tonic-gate { 39367c478bd9Sstevel@tonic-gate return (hat_kpm_pfn2va(pp->p_pagenum)); 39377c478bd9Sstevel@tonic-gate } 39387c478bd9Sstevel@tonic-gate 39397c478bd9Sstevel@tonic-gate /* 39407c478bd9Sstevel@tonic-gate * Return the page frame number for the kpm virtual address vaddr. 39417c478bd9Sstevel@tonic-gate */ 39427c478bd9Sstevel@tonic-gate pfn_t 39437c478bd9Sstevel@tonic-gate hat_kpm_va2pfn(caddr_t vaddr) 39447c478bd9Sstevel@tonic-gate { 39457c478bd9Sstevel@tonic-gate pfn_t pfn; 39467c478bd9Sstevel@tonic-gate 39477c478bd9Sstevel@tonic-gate ASSERT(IS_KPM_ADDR(vaddr)); 39487c478bd9Sstevel@tonic-gate 39497c478bd9Sstevel@tonic-gate pfn = (pfn_t)btop(vaddr - kpm_vbase); 39507c478bd9Sstevel@tonic-gate 39517c478bd9Sstevel@tonic-gate return (pfn); 39527c478bd9Sstevel@tonic-gate } 39537c478bd9Sstevel@tonic-gate 39547c478bd9Sstevel@tonic-gate 39557c478bd9Sstevel@tonic-gate /* 39567c478bd9Sstevel@tonic-gate * Return the page for the kpm virtual address vaddr. 39577c478bd9Sstevel@tonic-gate */ 39587c478bd9Sstevel@tonic-gate page_t * 39597c478bd9Sstevel@tonic-gate hat_kpm_vaddr2page(caddr_t vaddr) 39607c478bd9Sstevel@tonic-gate { 39617c478bd9Sstevel@tonic-gate pfn_t pfn; 39627c478bd9Sstevel@tonic-gate 39637c478bd9Sstevel@tonic-gate ASSERT(IS_KPM_ADDR(vaddr)); 39647c478bd9Sstevel@tonic-gate 39657c478bd9Sstevel@tonic-gate pfn = hat_kpm_va2pfn(vaddr); 39667c478bd9Sstevel@tonic-gate 39677c478bd9Sstevel@tonic-gate return (page_numtopp_nolock(pfn)); 39687c478bd9Sstevel@tonic-gate } 39697c478bd9Sstevel@tonic-gate 39707c478bd9Sstevel@tonic-gate /* 39717c478bd9Sstevel@tonic-gate * hat_kpm_fault is called from segkpm_fault when we take a page fault on a 39727c478bd9Sstevel@tonic-gate * KPM page. This should never happen on x86 39737c478bd9Sstevel@tonic-gate */ 39747c478bd9Sstevel@tonic-gate int 39757c478bd9Sstevel@tonic-gate hat_kpm_fault(hat_t *hat, caddr_t vaddr) 39767c478bd9Sstevel@tonic-gate { 39777c478bd9Sstevel@tonic-gate panic("pagefault in seg_kpm. hat: 0x%p vaddr: 0x%p", hat, vaddr); 39787c478bd9Sstevel@tonic-gate 39797c478bd9Sstevel@tonic-gate return (0); 39807c478bd9Sstevel@tonic-gate } 39817c478bd9Sstevel@tonic-gate 39827c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 39837c478bd9Sstevel@tonic-gate void 39847c478bd9Sstevel@tonic-gate hat_kpm_mseghash_clear(int nentries) 39857c478bd9Sstevel@tonic-gate {} 39867c478bd9Sstevel@tonic-gate 39877c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 39887c478bd9Sstevel@tonic-gate void 39897c478bd9Sstevel@tonic-gate hat_kpm_mseghash_update(pgcnt_t inx, struct memseg *msp) 39907c478bd9Sstevel@tonic-gate {} 3991