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 /* 226c9930aeSJoe Bonasera * Copyright (c) 1992, 2010, Oracle and/or its affiliates. All rights reserved. 237c478bd9Sstevel@tonic-gate */ 24a3114836SGerry Liu /* 25a3114836SGerry Liu * Copyright (c) 2010, Intel Corporation. 26a3114836SGerry Liu * All rights reserved. 27a3114836SGerry Liu */ 28c7c6ab2aSGarrett D'Amore /* 29c7c6ab2aSGarrett D'Amore * Copyright 2011 Nexenta Systems, Inc. All rights reserved. 30*c2e5ad71SSam Gwydir * Copyright 2018 Joyent, Inc. All rights reserved. 31a6a74e0eSMatthew Ahrens * Copyright (c) 2014, 2015 by Delphix. All rights reserved. 32c7c6ab2aSGarrett D'Amore */ 337c478bd9Sstevel@tonic-gate 347c478bd9Sstevel@tonic-gate /* 357c478bd9Sstevel@tonic-gate * VM - Hardware Address Translation management for i386 and amd64 367c478bd9Sstevel@tonic-gate * 377c478bd9Sstevel@tonic-gate * Implementation of the interfaces described in <common/vm/hat.h> 387c478bd9Sstevel@tonic-gate * 397c478bd9Sstevel@tonic-gate * Nearly all the details of how the hardware is managed should not be 407c478bd9Sstevel@tonic-gate * visible outside this layer except for misc. machine specific functions 417c478bd9Sstevel@tonic-gate * that work in conjunction with this code. 427c478bd9Sstevel@tonic-gate * 437c478bd9Sstevel@tonic-gate * Routines used only inside of i86pc/vm start with hati_ for HAT Internal. 447c478bd9Sstevel@tonic-gate */ 457c478bd9Sstevel@tonic-gate 467c478bd9Sstevel@tonic-gate #include <sys/machparam.h> 477c478bd9Sstevel@tonic-gate #include <sys/machsystm.h> 487c478bd9Sstevel@tonic-gate #include <sys/mman.h> 497c478bd9Sstevel@tonic-gate #include <sys/types.h> 507c478bd9Sstevel@tonic-gate #include <sys/systm.h> 517c478bd9Sstevel@tonic-gate #include <sys/cpuvar.h> 527c478bd9Sstevel@tonic-gate #include <sys/thread.h> 537c478bd9Sstevel@tonic-gate #include <sys/proc.h> 547c478bd9Sstevel@tonic-gate #include <sys/cpu.h> 557c478bd9Sstevel@tonic-gate #include <sys/kmem.h> 567c478bd9Sstevel@tonic-gate #include <sys/disp.h> 577c478bd9Sstevel@tonic-gate #include <sys/shm.h> 587c478bd9Sstevel@tonic-gate #include <sys/sysmacros.h> 597c478bd9Sstevel@tonic-gate #include <sys/machparam.h> 607c478bd9Sstevel@tonic-gate #include <sys/vmem.h> 617c478bd9Sstevel@tonic-gate #include <sys/vmsystm.h> 627c478bd9Sstevel@tonic-gate #include <sys/promif.h> 637c478bd9Sstevel@tonic-gate #include <sys/var.h> 647c478bd9Sstevel@tonic-gate #include <sys/x86_archext.h> 657c478bd9Sstevel@tonic-gate #include <sys/atomic.h> 667c478bd9Sstevel@tonic-gate #include <sys/bitmap.h> 67ae115bc7Smrj #include <sys/controlregs.h> 68ae115bc7Smrj #include <sys/bootconf.h> 69ae115bc7Smrj #include <sys/bootsvcs.h> 70ae115bc7Smrj #include <sys/bootinfo.h> 7195c0a3c8Sjosephb #include <sys/archsystm.h> 727c478bd9Sstevel@tonic-gate 737c478bd9Sstevel@tonic-gate #include <vm/seg_kmem.h> 747c478bd9Sstevel@tonic-gate #include <vm/hat_i86.h> 757c478bd9Sstevel@tonic-gate #include <vm/as.h> 767c478bd9Sstevel@tonic-gate #include <vm/seg.h> 777c478bd9Sstevel@tonic-gate #include <vm/page.h> 787c478bd9Sstevel@tonic-gate #include <vm/seg_kp.h> 797c478bd9Sstevel@tonic-gate #include <vm/seg_kpm.h> 807c478bd9Sstevel@tonic-gate #include <vm/vm_dep.h> 81843e1988Sjohnlev #ifdef __xpv 82843e1988Sjohnlev #include <sys/hypervisor.h> 83843e1988Sjohnlev #endif 84ae115bc7Smrj #include <vm/kboot_mmu.h> 85250b7ff9Sjosephb #include <vm/seg_spt.h> 867c478bd9Sstevel@tonic-gate 877c478bd9Sstevel@tonic-gate #include <sys/cmn_err.h> 887c478bd9Sstevel@tonic-gate 897c478bd9Sstevel@tonic-gate /* 907c478bd9Sstevel@tonic-gate * Basic parameters for hat operation. 917c478bd9Sstevel@tonic-gate */ 927c478bd9Sstevel@tonic-gate struct hat_mmu_info mmu; 937c478bd9Sstevel@tonic-gate 947c478bd9Sstevel@tonic-gate /* 957c478bd9Sstevel@tonic-gate * The page that is the kernel's top level pagetable. 967c478bd9Sstevel@tonic-gate * 97843e1988Sjohnlev * For 32 bit PAE support on i86pc, the kernel hat will use the 1st 4 entries 987c478bd9Sstevel@tonic-gate * on this 4K page for its top level page table. The remaining groups of 997c478bd9Sstevel@tonic-gate * 4 entries are used for per processor copies of user VLP pagetables for 1007c478bd9Sstevel@tonic-gate * running threads. See hat_switch() and reload_pae32() for details. 1017c478bd9Sstevel@tonic-gate * 102843e1988Sjohnlev * vlp_page[0..3] - level==2 PTEs for kernel HAT 103843e1988Sjohnlev * vlp_page[4..7] - level==2 PTEs for user thread on cpu 0 104843e1988Sjohnlev * vlp_page[8..11] - level==2 PTE for user thread on cpu 1 105843e1988Sjohnlev * etc... 1067c478bd9Sstevel@tonic-gate */ 1077c478bd9Sstevel@tonic-gate static x86pte_t *vlp_page; 1087c478bd9Sstevel@tonic-gate 1097c478bd9Sstevel@tonic-gate /* 1107c478bd9Sstevel@tonic-gate * forward declaration of internal utility routines 1117c478bd9Sstevel@tonic-gate */ 1127c478bd9Sstevel@tonic-gate static x86pte_t hati_update_pte(htable_t *ht, uint_t entry, x86pte_t expected, 1137c478bd9Sstevel@tonic-gate x86pte_t new); 1147c478bd9Sstevel@tonic-gate 1157c478bd9Sstevel@tonic-gate /* 1167c478bd9Sstevel@tonic-gate * The kernel address space exists in all HATs. To implement this the 117843e1988Sjohnlev * kernel reserves a fixed number of entries in the topmost level(s) of page 118843e1988Sjohnlev * tables. The values are setup during startup and then copied to every user 119843e1988Sjohnlev * hat created by hat_alloc(). This means that kernelbase must be: 1207c478bd9Sstevel@tonic-gate * 1217c478bd9Sstevel@tonic-gate * 4Meg aligned for 32 bit kernels 1227c478bd9Sstevel@tonic-gate * 512Gig aligned for x86_64 64 bit kernel 1237c478bd9Sstevel@tonic-gate * 124843e1988Sjohnlev * The hat_kernel_range_ts describe what needs to be copied from kernel hat 125843e1988Sjohnlev * to each user hat. 1267c478bd9Sstevel@tonic-gate */ 127843e1988Sjohnlev typedef struct hat_kernel_range { 128843e1988Sjohnlev level_t hkr_level; 129843e1988Sjohnlev uintptr_t hkr_start_va; 130843e1988Sjohnlev uintptr_t hkr_end_va; /* zero means to end of memory */ 131843e1988Sjohnlev } hat_kernel_range_t; 132843e1988Sjohnlev #define NUM_KERNEL_RANGE 2 133843e1988Sjohnlev static hat_kernel_range_t kernel_ranges[NUM_KERNEL_RANGE]; 134843e1988Sjohnlev static int num_kernel_ranges; 1357c478bd9Sstevel@tonic-gate 1367c478bd9Sstevel@tonic-gate uint_t use_boot_reserve = 1; /* cleared after early boot process */ 1377c478bd9Sstevel@tonic-gate uint_t can_steal_post_boot = 0; /* set late in boot to enable stealing */ 1387c478bd9Sstevel@tonic-gate 139512cf780Skchow /* 140512cf780Skchow * enable_1gpg: controls 1g page support for user applications. 141512cf780Skchow * By default, 1g pages are exported to user applications. enable_1gpg can 142512cf780Skchow * be set to 0 to not export. 143512cf780Skchow */ 14478b03d3aSkchow int enable_1gpg = 1; 14502bc52beSkchow 146512cf780Skchow /* 147512cf780Skchow * AMD shanghai processors provide better management of 1gb ptes in its tlb. 14821584dbcSPavel Tatashin * By default, 1g page support will be disabled for pre-shanghai AMD 149512cf780Skchow * processors that don't have optimal tlb support for the 1g page size. 150512cf780Skchow * chk_optimal_1gtlb can be set to 0 to force 1g page support on sub-optimal 151512cf780Skchow * processors. 152512cf780Skchow */ 153512cf780Skchow int chk_optimal_1gtlb = 1; 154512cf780Skchow 155512cf780Skchow 15602bc52beSkchow #ifdef DEBUG 15702bc52beSkchow uint_t map1gcnt; 15802bc52beSkchow #endif 15902bc52beSkchow 16002bc52beSkchow 1617c478bd9Sstevel@tonic-gate /* 1627c478bd9Sstevel@tonic-gate * A cpuset for all cpus. This is used for kernel address cross calls, since 1637c478bd9Sstevel@tonic-gate * the kernel addresses apply to all cpus. 1647c478bd9Sstevel@tonic-gate */ 1657c478bd9Sstevel@tonic-gate cpuset_t khat_cpuset; 1667c478bd9Sstevel@tonic-gate 1677c478bd9Sstevel@tonic-gate /* 1687c478bd9Sstevel@tonic-gate * management stuff for hat structures 1697c478bd9Sstevel@tonic-gate */ 1707c478bd9Sstevel@tonic-gate kmutex_t hat_list_lock; 1717c478bd9Sstevel@tonic-gate kcondvar_t hat_list_cv; 1727c478bd9Sstevel@tonic-gate kmem_cache_t *hat_cache; 1737c478bd9Sstevel@tonic-gate kmem_cache_t *hat_hash_cache; 1747c478bd9Sstevel@tonic-gate kmem_cache_t *vlp_hash_cache; 1757c478bd9Sstevel@tonic-gate 1767c478bd9Sstevel@tonic-gate /* 1777c478bd9Sstevel@tonic-gate * Simple statistics 1787c478bd9Sstevel@tonic-gate */ 1797c478bd9Sstevel@tonic-gate struct hatstats hatstat; 1807c478bd9Sstevel@tonic-gate 181ab4a9bebSjohnlev /* 182ab4a9bebSjohnlev * Some earlier hypervisor versions do not emulate cmpxchg of PTEs 183ab4a9bebSjohnlev * correctly. For such hypervisors we must set PT_USER for kernel 184ab4a9bebSjohnlev * entries ourselves (normally the emulation would set PT_USER for 185ab4a9bebSjohnlev * kernel entries and PT_USER|PT_GLOBAL for user entries). pt_kern is 186ab4a9bebSjohnlev * thus set appropriately. Note that dboot/kbm is OK, as only the full 187ab4a9bebSjohnlev * HAT uses cmpxchg() and the other paths (hypercall etc.) were never 188ab4a9bebSjohnlev * incorrect. 189ab4a9bebSjohnlev */ 190ab4a9bebSjohnlev int pt_kern; 191ab4a9bebSjohnlev 1927c478bd9Sstevel@tonic-gate /* 1937c478bd9Sstevel@tonic-gate * useful stuff for atomic access/clearing/setting REF/MOD/RO bits in page_t's. 1947c478bd9Sstevel@tonic-gate */ 1957c478bd9Sstevel@tonic-gate extern void atomic_orb(uchar_t *addr, uchar_t val); 1967c478bd9Sstevel@tonic-gate extern void atomic_andb(uchar_t *addr, uchar_t val); 1977c478bd9Sstevel@tonic-gate 198a3114836SGerry Liu #ifndef __xpv 199a3114836SGerry Liu extern pfn_t memseg_get_start(struct memseg *); 200a3114836SGerry Liu #endif 201a3114836SGerry Liu 2027c478bd9Sstevel@tonic-gate #define PP_GETRM(pp, rmmask) (pp->p_nrm & rmmask) 2037c478bd9Sstevel@tonic-gate #define PP_ISMOD(pp) PP_GETRM(pp, P_MOD) 2047c478bd9Sstevel@tonic-gate #define PP_ISREF(pp) PP_GETRM(pp, P_REF) 2057c478bd9Sstevel@tonic-gate #define PP_ISRO(pp) PP_GETRM(pp, P_RO) 2067c478bd9Sstevel@tonic-gate 2077c478bd9Sstevel@tonic-gate #define PP_SETRM(pp, rm) atomic_orb(&(pp->p_nrm), rm) 2087c478bd9Sstevel@tonic-gate #define PP_SETMOD(pp) PP_SETRM(pp, P_MOD) 2097c478bd9Sstevel@tonic-gate #define PP_SETREF(pp) PP_SETRM(pp, P_REF) 2107c478bd9Sstevel@tonic-gate #define PP_SETRO(pp) PP_SETRM(pp, P_RO) 2117c478bd9Sstevel@tonic-gate 2127c478bd9Sstevel@tonic-gate #define PP_CLRRM(pp, rm) atomic_andb(&(pp->p_nrm), ~(rm)) 2137c478bd9Sstevel@tonic-gate #define PP_CLRMOD(pp) PP_CLRRM(pp, P_MOD) 2147c478bd9Sstevel@tonic-gate #define PP_CLRREF(pp) PP_CLRRM(pp, P_REF) 2157c478bd9Sstevel@tonic-gate #define PP_CLRRO(pp) PP_CLRRM(pp, P_RO) 2167c478bd9Sstevel@tonic-gate #define PP_CLRALL(pp) PP_CLRRM(pp, P_MOD | P_REF | P_RO) 2177c478bd9Sstevel@tonic-gate 2187c478bd9Sstevel@tonic-gate /* 2197c478bd9Sstevel@tonic-gate * kmem cache constructor for struct hat 2207c478bd9Sstevel@tonic-gate */ 2217c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 2227c478bd9Sstevel@tonic-gate static int 2237c478bd9Sstevel@tonic-gate hati_constructor(void *buf, void *handle, int kmflags) 2247c478bd9Sstevel@tonic-gate { 2257c478bd9Sstevel@tonic-gate hat_t *hat = buf; 2267c478bd9Sstevel@tonic-gate 2277c478bd9Sstevel@tonic-gate mutex_init(&hat->hat_mutex, NULL, MUTEX_DEFAULT, NULL); 2287c478bd9Sstevel@tonic-gate bzero(hat->hat_pages_mapped, 2297c478bd9Sstevel@tonic-gate sizeof (pgcnt_t) * (mmu.max_page_level + 1)); 230250b7ff9Sjosephb hat->hat_ism_pgcnt = 0; 2317c478bd9Sstevel@tonic-gate hat->hat_stats = 0; 2327c478bd9Sstevel@tonic-gate hat->hat_flags = 0; 2337c478bd9Sstevel@tonic-gate CPUSET_ZERO(hat->hat_cpus); 2347c478bd9Sstevel@tonic-gate hat->hat_htable = NULL; 2357c478bd9Sstevel@tonic-gate hat->hat_ht_hash = NULL; 2367c478bd9Sstevel@tonic-gate return (0); 2377c478bd9Sstevel@tonic-gate } 2387c478bd9Sstevel@tonic-gate 2397c478bd9Sstevel@tonic-gate /* 2407c478bd9Sstevel@tonic-gate * Allocate a hat structure for as. We also create the top level 2417c478bd9Sstevel@tonic-gate * htable and initialize it to contain the kernel hat entries. 2427c478bd9Sstevel@tonic-gate */ 2437c478bd9Sstevel@tonic-gate hat_t * 2447c478bd9Sstevel@tonic-gate hat_alloc(struct as *as) 2457c478bd9Sstevel@tonic-gate { 246843e1988Sjohnlev hat_t *hat; 247843e1988Sjohnlev htable_t *ht; /* top level htable */ 248843e1988Sjohnlev uint_t use_vlp; 249843e1988Sjohnlev uint_t r; 250843e1988Sjohnlev hat_kernel_range_t *rp; 251843e1988Sjohnlev uintptr_t va; 252843e1988Sjohnlev uintptr_t eva; 253843e1988Sjohnlev uint_t start; 254843e1988Sjohnlev uint_t cnt; 255843e1988Sjohnlev htable_t *src; 2567c478bd9Sstevel@tonic-gate 2577c478bd9Sstevel@tonic-gate /* 2587c478bd9Sstevel@tonic-gate * Once we start creating user process HATs we can enable 2597c478bd9Sstevel@tonic-gate * the htable_steal() code. 2607c478bd9Sstevel@tonic-gate */ 2617c478bd9Sstevel@tonic-gate if (can_steal_post_boot == 0) 2627c478bd9Sstevel@tonic-gate can_steal_post_boot = 1; 2637c478bd9Sstevel@tonic-gate 264dc32d872SJosef 'Jeff' Sipek ASSERT(AS_WRITE_HELD(as)); 2657c478bd9Sstevel@tonic-gate hat = kmem_cache_alloc(hat_cache, KM_SLEEP); 2667c478bd9Sstevel@tonic-gate hat->hat_as = as; 2677c478bd9Sstevel@tonic-gate mutex_init(&hat->hat_mutex, NULL, MUTEX_DEFAULT, NULL); 2687c478bd9Sstevel@tonic-gate ASSERT(hat->hat_flags == 0); 2697c478bd9Sstevel@tonic-gate 270843e1988Sjohnlev #if defined(__xpv) 2717c478bd9Sstevel@tonic-gate /* 272843e1988Sjohnlev * No VLP stuff on the hypervisor due to the 64-bit split top level 273843e1988Sjohnlev * page tables. On 32-bit it's not needed as the hypervisor takes 274843e1988Sjohnlev * care of copying the top level PTEs to a below 4Gig page. 2757c478bd9Sstevel@tonic-gate */ 276843e1988Sjohnlev use_vlp = 0; 277843e1988Sjohnlev #else /* __xpv */ 278843e1988Sjohnlev /* 32 bit processes uses a VLP style hat when running with PAE */ 2797c478bd9Sstevel@tonic-gate #if defined(__amd64) 2807c478bd9Sstevel@tonic-gate use_vlp = (ttoproc(curthread)->p_model == DATAMODEL_ILP32); 2817c478bd9Sstevel@tonic-gate #elif defined(__i386) 2827c478bd9Sstevel@tonic-gate use_vlp = mmu.pae_hat; 2837c478bd9Sstevel@tonic-gate #endif 284843e1988Sjohnlev #endif /* __xpv */ 2857c478bd9Sstevel@tonic-gate if (use_vlp) { 2867c478bd9Sstevel@tonic-gate hat->hat_flags = HAT_VLP; 2877c478bd9Sstevel@tonic-gate bzero(hat->hat_vlp_ptes, VLP_SIZE); 2887c478bd9Sstevel@tonic-gate } 2897c478bd9Sstevel@tonic-gate 2907c478bd9Sstevel@tonic-gate /* 2917c478bd9Sstevel@tonic-gate * Allocate the htable hash 2927c478bd9Sstevel@tonic-gate */ 2937c478bd9Sstevel@tonic-gate if ((hat->hat_flags & HAT_VLP)) { 2947c478bd9Sstevel@tonic-gate hat->hat_num_hash = mmu.vlp_hash_cnt; 2957c478bd9Sstevel@tonic-gate hat->hat_ht_hash = kmem_cache_alloc(vlp_hash_cache, KM_SLEEP); 2967c478bd9Sstevel@tonic-gate } else { 2977c478bd9Sstevel@tonic-gate hat->hat_num_hash = mmu.hash_cnt; 2987c478bd9Sstevel@tonic-gate hat->hat_ht_hash = kmem_cache_alloc(hat_hash_cache, KM_SLEEP); 2997c478bd9Sstevel@tonic-gate } 3007c478bd9Sstevel@tonic-gate bzero(hat->hat_ht_hash, hat->hat_num_hash * sizeof (htable_t *)); 3017c478bd9Sstevel@tonic-gate 3027c478bd9Sstevel@tonic-gate /* 3037c478bd9Sstevel@tonic-gate * Initialize Kernel HAT entries at the top of the top level page 304843e1988Sjohnlev * tables for the new hat. 3057c478bd9Sstevel@tonic-gate */ 3067c478bd9Sstevel@tonic-gate hat->hat_htable = NULL; 3077c478bd9Sstevel@tonic-gate hat->hat_ht_cached = NULL; 308843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 3097c478bd9Sstevel@tonic-gate ht = htable_create(hat, (uintptr_t)0, TOP_LEVEL(hat), NULL); 310843e1988Sjohnlev hat->hat_htable = ht; 311ae115bc7Smrj 312843e1988Sjohnlev #if defined(__amd64) 313843e1988Sjohnlev if (hat->hat_flags & HAT_VLP) 314843e1988Sjohnlev goto init_done; 3157c478bd9Sstevel@tonic-gate #endif 3167c478bd9Sstevel@tonic-gate 317843e1988Sjohnlev for (r = 0; r < num_kernel_ranges; ++r) { 318843e1988Sjohnlev rp = &kernel_ranges[r]; 319843e1988Sjohnlev for (va = rp->hkr_start_va; va != rp->hkr_end_va; 320843e1988Sjohnlev va += cnt * LEVEL_SIZE(rp->hkr_level)) { 321843e1988Sjohnlev 322843e1988Sjohnlev if (rp->hkr_level == TOP_LEVEL(hat)) 323843e1988Sjohnlev ht = hat->hat_htable; 324843e1988Sjohnlev else 325843e1988Sjohnlev ht = htable_create(hat, va, rp->hkr_level, 326843e1988Sjohnlev NULL); 327843e1988Sjohnlev 328843e1988Sjohnlev start = htable_va2entry(va, ht); 329843e1988Sjohnlev cnt = HTABLE_NUM_PTES(ht) - start; 330843e1988Sjohnlev eva = va + 331843e1988Sjohnlev ((uintptr_t)cnt << LEVEL_SHIFT(rp->hkr_level)); 332843e1988Sjohnlev if (rp->hkr_end_va != 0 && 333843e1988Sjohnlev (eva > rp->hkr_end_va || eva == 0)) 334843e1988Sjohnlev cnt = htable_va2entry(rp->hkr_end_va, ht) - 335843e1988Sjohnlev start; 336843e1988Sjohnlev 337843e1988Sjohnlev #if defined(__i386) && !defined(__xpv) 338843e1988Sjohnlev if (ht->ht_flags & HTABLE_VLP) { 339843e1988Sjohnlev bcopy(&vlp_page[start], 340843e1988Sjohnlev &hat->hat_vlp_ptes[start], 341843e1988Sjohnlev cnt * sizeof (x86pte_t)); 342843e1988Sjohnlev continue; 343843e1988Sjohnlev } 344843e1988Sjohnlev #endif 345843e1988Sjohnlev src = htable_lookup(kas.a_hat, va, rp->hkr_level); 346843e1988Sjohnlev ASSERT(src != NULL); 347843e1988Sjohnlev x86pte_copy(src, ht, start, cnt); 348843e1988Sjohnlev htable_release(src); 349843e1988Sjohnlev } 350843e1988Sjohnlev } 351843e1988Sjohnlev 352843e1988Sjohnlev init_done: 353843e1988Sjohnlev 354843e1988Sjohnlev #if defined(__xpv) 3557c478bd9Sstevel@tonic-gate /* 356843e1988Sjohnlev * Pin top level page tables after initializing them 3577c478bd9Sstevel@tonic-gate */ 358843e1988Sjohnlev xen_pin(hat->hat_htable->ht_pfn, mmu.max_level); 359843e1988Sjohnlev #if defined(__amd64) 360843e1988Sjohnlev xen_pin(hat->hat_user_ptable, mmu.max_level); 361843e1988Sjohnlev #endif 3627c478bd9Sstevel@tonic-gate #endif 363551bc2a6Smrj XPV_ALLOW_MIGRATE(); 3647c478bd9Sstevel@tonic-gate 3657c478bd9Sstevel@tonic-gate /* 366a85a6733Sjosephb * Put it at the start of the global list of all hats (used by stealing) 367a85a6733Sjosephb * 368a85a6733Sjosephb * kas.a_hat is not in the list but is instead used to find the 369a85a6733Sjosephb * first and last items in the list. 370a85a6733Sjosephb * 371a85a6733Sjosephb * - kas.a_hat->hat_next points to the start of the user hats. 372a85a6733Sjosephb * The list ends where hat->hat_next == NULL 373a85a6733Sjosephb * 374a85a6733Sjosephb * - kas.a_hat->hat_prev points to the last of the user hats. 375a85a6733Sjosephb * The list begins where hat->hat_prev == NULL 3767c478bd9Sstevel@tonic-gate */ 3777c478bd9Sstevel@tonic-gate mutex_enter(&hat_list_lock); 378a85a6733Sjosephb hat->hat_prev = NULL; 379a85a6733Sjosephb hat->hat_next = kas.a_hat->hat_next; 380a85a6733Sjosephb if (hat->hat_next) 381a85a6733Sjosephb hat->hat_next->hat_prev = hat; 382a85a6733Sjosephb else 383a85a6733Sjosephb kas.a_hat->hat_prev = hat; 3847c478bd9Sstevel@tonic-gate kas.a_hat->hat_next = hat; 3857c478bd9Sstevel@tonic-gate mutex_exit(&hat_list_lock); 3867c478bd9Sstevel@tonic-gate 3877c478bd9Sstevel@tonic-gate return (hat); 3887c478bd9Sstevel@tonic-gate } 3897c478bd9Sstevel@tonic-gate 3907c478bd9Sstevel@tonic-gate /* 3917c478bd9Sstevel@tonic-gate * process has finished executing but as has not been cleaned up yet. 3927c478bd9Sstevel@tonic-gate */ 3937c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 3947c478bd9Sstevel@tonic-gate void 3957c478bd9Sstevel@tonic-gate hat_free_start(hat_t *hat) 3967c478bd9Sstevel@tonic-gate { 397dc32d872SJosef 'Jeff' Sipek ASSERT(AS_WRITE_HELD(hat->hat_as)); 398a85a6733Sjosephb 399a85a6733Sjosephb /* 400a85a6733Sjosephb * If the hat is currently a stealing victim, wait for the stealing 401a85a6733Sjosephb * to finish. Once we mark it as HAT_FREEING, htable_steal() 402a85a6733Sjosephb * won't look at its pagetables anymore. 403a85a6733Sjosephb */ 4047c478bd9Sstevel@tonic-gate mutex_enter(&hat_list_lock); 405a85a6733Sjosephb while (hat->hat_flags & HAT_VICTIM) 406a85a6733Sjosephb cv_wait(&hat_list_cv, &hat_list_lock); 4077c478bd9Sstevel@tonic-gate hat->hat_flags |= HAT_FREEING; 4087c478bd9Sstevel@tonic-gate mutex_exit(&hat_list_lock); 4097c478bd9Sstevel@tonic-gate } 4107c478bd9Sstevel@tonic-gate 4117c478bd9Sstevel@tonic-gate /* 4127c478bd9Sstevel@tonic-gate * An address space is being destroyed, so we destroy the associated hat. 4137c478bd9Sstevel@tonic-gate */ 4147c478bd9Sstevel@tonic-gate void 4157c478bd9Sstevel@tonic-gate hat_free_end(hat_t *hat) 4167c478bd9Sstevel@tonic-gate { 4177c478bd9Sstevel@tonic-gate kmem_cache_t *cache; 4187c478bd9Sstevel@tonic-gate 4197c478bd9Sstevel@tonic-gate ASSERT(hat->hat_flags & HAT_FREEING); 4207c478bd9Sstevel@tonic-gate 4217c478bd9Sstevel@tonic-gate /* 4227c478bd9Sstevel@tonic-gate * must not be running on the given hat 4237c478bd9Sstevel@tonic-gate */ 4247c478bd9Sstevel@tonic-gate ASSERT(CPU->cpu_current_hat != hat); 4257c478bd9Sstevel@tonic-gate 4267c478bd9Sstevel@tonic-gate /* 427a85a6733Sjosephb * Remove it from the list of HATs 4287c478bd9Sstevel@tonic-gate */ 4297c478bd9Sstevel@tonic-gate mutex_enter(&hat_list_lock); 430a85a6733Sjosephb if (hat->hat_prev) 431a85a6733Sjosephb hat->hat_prev->hat_next = hat->hat_next; 432a85a6733Sjosephb else 4337c478bd9Sstevel@tonic-gate kas.a_hat->hat_next = hat->hat_next; 434a85a6733Sjosephb if (hat->hat_next) 435a85a6733Sjosephb hat->hat_next->hat_prev = hat->hat_prev; 436a85a6733Sjosephb else 437a85a6733Sjosephb kas.a_hat->hat_prev = hat->hat_prev; 4387c478bd9Sstevel@tonic-gate mutex_exit(&hat_list_lock); 439a85a6733Sjosephb hat->hat_next = hat->hat_prev = NULL; 4407c478bd9Sstevel@tonic-gate 441843e1988Sjohnlev #if defined(__xpv) 442843e1988Sjohnlev /* 443843e1988Sjohnlev * On the hypervisor, unpin top level page table(s) 444843e1988Sjohnlev */ 445843e1988Sjohnlev xen_unpin(hat->hat_htable->ht_pfn); 446843e1988Sjohnlev #if defined(__amd64) 447843e1988Sjohnlev xen_unpin(hat->hat_user_ptable); 448843e1988Sjohnlev #endif 449843e1988Sjohnlev #endif 450843e1988Sjohnlev 4517c478bd9Sstevel@tonic-gate /* 4527c478bd9Sstevel@tonic-gate * Make a pass through the htables freeing them all up. 4537c478bd9Sstevel@tonic-gate */ 4547c478bd9Sstevel@tonic-gate htable_purge_hat(hat); 4557c478bd9Sstevel@tonic-gate 4567c478bd9Sstevel@tonic-gate /* 4577c478bd9Sstevel@tonic-gate * Decide which kmem cache the hash table came from, then free it. 4587c478bd9Sstevel@tonic-gate */ 4597c478bd9Sstevel@tonic-gate if (hat->hat_flags & HAT_VLP) 4607c478bd9Sstevel@tonic-gate cache = vlp_hash_cache; 4617c478bd9Sstevel@tonic-gate else 4627c478bd9Sstevel@tonic-gate cache = hat_hash_cache; 4637c478bd9Sstevel@tonic-gate kmem_cache_free(cache, hat->hat_ht_hash); 4647c478bd9Sstevel@tonic-gate hat->hat_ht_hash = NULL; 4657c478bd9Sstevel@tonic-gate 4667c478bd9Sstevel@tonic-gate hat->hat_flags = 0; 4677c478bd9Sstevel@tonic-gate kmem_cache_free(hat_cache, hat); 4687c478bd9Sstevel@tonic-gate } 4697c478bd9Sstevel@tonic-gate 4707c478bd9Sstevel@tonic-gate /* 4717c478bd9Sstevel@tonic-gate * round kernelbase down to a supported value to use for _userlimit 4727c478bd9Sstevel@tonic-gate * 4737c478bd9Sstevel@tonic-gate * userlimit must be aligned down to an entry in the top level htable. 4747c478bd9Sstevel@tonic-gate * The one exception is for 32 bit HAT's running PAE. 4757c478bd9Sstevel@tonic-gate */ 4767c478bd9Sstevel@tonic-gate uintptr_t 4777c478bd9Sstevel@tonic-gate hat_kernelbase(uintptr_t va) 4787c478bd9Sstevel@tonic-gate { 4797c478bd9Sstevel@tonic-gate #if defined(__i386) 4807c478bd9Sstevel@tonic-gate va &= LEVEL_MASK(1); 4817c478bd9Sstevel@tonic-gate #endif 4827c478bd9Sstevel@tonic-gate if (IN_VA_HOLE(va)) 4837c478bd9Sstevel@tonic-gate panic("_userlimit %p will fall in VA hole\n", (void *)va); 4847c478bd9Sstevel@tonic-gate return (va); 4857c478bd9Sstevel@tonic-gate } 4867c478bd9Sstevel@tonic-gate 487512cf780Skchow /* 488512cf780Skchow * 489512cf780Skchow */ 490512cf780Skchow static void 491512cf780Skchow set_max_page_level() 492512cf780Skchow { 493512cf780Skchow level_t lvl; 494512cf780Skchow 495512cf780Skchow if (!kbm_largepage_support) { 496512cf780Skchow lvl = 0; 4976f0cf5c5Skchow } else { 4987417cfdeSKuriakose Kuruvilla if (is_x86_feature(x86_featureset, X86FSET_1GPG)) { 4996f0cf5c5Skchow lvl = 2; 5006f0cf5c5Skchow if (chk_optimal_1gtlb && 5016f0cf5c5Skchow cpuid_opteron_erratum(CPU, 6671130)) { 5026f0cf5c5Skchow lvl = 1; 5036f0cf5c5Skchow } 5046f0cf5c5Skchow if (plat_mnode_xcheck(LEVEL_SIZE(2) >> 5056f0cf5c5Skchow LEVEL_SHIFT(0))) { 5066f0cf5c5Skchow lvl = 1; 5076f0cf5c5Skchow } 5086f0cf5c5Skchow } else { 509512cf780Skchow lvl = 1; 510512cf780Skchow } 511512cf780Skchow } 512512cf780Skchow mmu.max_page_level = lvl; 513512cf780Skchow 514512cf780Skchow if ((lvl == 2) && (enable_1gpg == 0)) 515512cf780Skchow mmu.umax_page_level = 1; 516512cf780Skchow else 517512cf780Skchow mmu.umax_page_level = lvl; 518512cf780Skchow } 519512cf780Skchow 5207c478bd9Sstevel@tonic-gate /* 5217c478bd9Sstevel@tonic-gate * Initialize hat data structures based on processor MMU information. 5227c478bd9Sstevel@tonic-gate */ 5237c478bd9Sstevel@tonic-gate void 5247c478bd9Sstevel@tonic-gate mmu_init(void) 5257c478bd9Sstevel@tonic-gate { 5267c478bd9Sstevel@tonic-gate uint_t max_htables; 5277c478bd9Sstevel@tonic-gate uint_t pa_bits; 5287c478bd9Sstevel@tonic-gate uint_t va_bits; 5297c478bd9Sstevel@tonic-gate int i; 5307c478bd9Sstevel@tonic-gate 5317c478bd9Sstevel@tonic-gate /* 532ae115bc7Smrj * If CPU enabled the page table global bit, use it for the kernel 533ae115bc7Smrj * This is bit 7 in CR4 (PGE - Page Global Enable). 5347c478bd9Sstevel@tonic-gate */ 5357417cfdeSKuriakose Kuruvilla if (is_x86_feature(x86_featureset, X86FSET_PGE) && 5367417cfdeSKuriakose Kuruvilla (getcr4() & CR4_PGE) != 0) 5377c478bd9Sstevel@tonic-gate mmu.pt_global = PT_GLOBAL; 5387c478bd9Sstevel@tonic-gate 5397c478bd9Sstevel@tonic-gate /* 540ae115bc7Smrj * Detect NX and PAE usage. 5417c478bd9Sstevel@tonic-gate */ 542ae115bc7Smrj mmu.pae_hat = kbm_pae_support; 543ae115bc7Smrj if (kbm_nx_support) 5447c478bd9Sstevel@tonic-gate mmu.pt_nx = PT_NX; 545ae115bc7Smrj else 5467c478bd9Sstevel@tonic-gate mmu.pt_nx = 0; 5477c478bd9Sstevel@tonic-gate 5487c478bd9Sstevel@tonic-gate /* 5497c478bd9Sstevel@tonic-gate * Use CPU info to set various MMU parameters 5507c478bd9Sstevel@tonic-gate */ 5517c478bd9Sstevel@tonic-gate cpuid_get_addrsize(CPU, &pa_bits, &va_bits); 5527c478bd9Sstevel@tonic-gate 5537c478bd9Sstevel@tonic-gate if (va_bits < sizeof (void *) * NBBY) { 5547c478bd9Sstevel@tonic-gate mmu.hole_start = (1ul << (va_bits - 1)); 5557c478bd9Sstevel@tonic-gate mmu.hole_end = 0ul - mmu.hole_start - 1; 5567c478bd9Sstevel@tonic-gate } else { 5577c478bd9Sstevel@tonic-gate mmu.hole_end = 0; 5587c478bd9Sstevel@tonic-gate mmu.hole_start = mmu.hole_end - 1; 5597c478bd9Sstevel@tonic-gate } 5607c478bd9Sstevel@tonic-gate #if defined(OPTERON_ERRATUM_121) 5617c478bd9Sstevel@tonic-gate /* 5627c478bd9Sstevel@tonic-gate * If erratum 121 has already been detected at this time, hole_start 5637c478bd9Sstevel@tonic-gate * contains the value to be subtracted from mmu.hole_start. 5647c478bd9Sstevel@tonic-gate */ 5657c478bd9Sstevel@tonic-gate ASSERT(hole_start == 0 || opteron_erratum_121 != 0); 5667c478bd9Sstevel@tonic-gate hole_start = mmu.hole_start - hole_start; 5677c478bd9Sstevel@tonic-gate #else 5687c478bd9Sstevel@tonic-gate hole_start = mmu.hole_start; 5697c478bd9Sstevel@tonic-gate #endif 5707c478bd9Sstevel@tonic-gate hole_end = mmu.hole_end; 5717c478bd9Sstevel@tonic-gate 5727c478bd9Sstevel@tonic-gate mmu.highest_pfn = mmu_btop((1ull << pa_bits) - 1); 5737c478bd9Sstevel@tonic-gate if (mmu.pae_hat == 0 && pa_bits > 32) 5747c478bd9Sstevel@tonic-gate mmu.highest_pfn = PFN_4G - 1; 5757c478bd9Sstevel@tonic-gate 5767c478bd9Sstevel@tonic-gate if (mmu.pae_hat) { 5777c478bd9Sstevel@tonic-gate mmu.pte_size = 8; /* 8 byte PTEs */ 5787c478bd9Sstevel@tonic-gate mmu.pte_size_shift = 3; 5797c478bd9Sstevel@tonic-gate } else { 5807c478bd9Sstevel@tonic-gate mmu.pte_size = 4; /* 4 byte PTEs */ 5817c478bd9Sstevel@tonic-gate mmu.pte_size_shift = 2; 5827c478bd9Sstevel@tonic-gate } 5837c478bd9Sstevel@tonic-gate 5847417cfdeSKuriakose Kuruvilla if (mmu.pae_hat && !is_x86_feature(x86_featureset, X86FSET_PAE)) 5857c478bd9Sstevel@tonic-gate panic("Processor does not support PAE"); 5867c478bd9Sstevel@tonic-gate 5877417cfdeSKuriakose Kuruvilla if (!is_x86_feature(x86_featureset, X86FSET_CX8)) 5887c478bd9Sstevel@tonic-gate panic("Processor does not support cmpxchg8b instruction"); 5897c478bd9Sstevel@tonic-gate 5907c478bd9Sstevel@tonic-gate #if defined(__amd64) 5917c478bd9Sstevel@tonic-gate 5927c478bd9Sstevel@tonic-gate mmu.num_level = 4; 5937c478bd9Sstevel@tonic-gate mmu.max_level = 3; 5947c478bd9Sstevel@tonic-gate mmu.ptes_per_table = 512; 5957c478bd9Sstevel@tonic-gate mmu.top_level_count = 512; 5967c478bd9Sstevel@tonic-gate 5977c478bd9Sstevel@tonic-gate mmu.level_shift[0] = 12; 5987c478bd9Sstevel@tonic-gate mmu.level_shift[1] = 21; 5997c478bd9Sstevel@tonic-gate mmu.level_shift[2] = 30; 6007c478bd9Sstevel@tonic-gate mmu.level_shift[3] = 39; 6017c478bd9Sstevel@tonic-gate 6027c478bd9Sstevel@tonic-gate #elif defined(__i386) 6037c478bd9Sstevel@tonic-gate 6047c478bd9Sstevel@tonic-gate if (mmu.pae_hat) { 6057c478bd9Sstevel@tonic-gate mmu.num_level = 3; 6067c478bd9Sstevel@tonic-gate mmu.max_level = 2; 6077c478bd9Sstevel@tonic-gate mmu.ptes_per_table = 512; 6087c478bd9Sstevel@tonic-gate mmu.top_level_count = 4; 6097c478bd9Sstevel@tonic-gate 6107c478bd9Sstevel@tonic-gate mmu.level_shift[0] = 12; 6117c478bd9Sstevel@tonic-gate mmu.level_shift[1] = 21; 6127c478bd9Sstevel@tonic-gate mmu.level_shift[2] = 30; 6137c478bd9Sstevel@tonic-gate 6147c478bd9Sstevel@tonic-gate } else { 6157c478bd9Sstevel@tonic-gate mmu.num_level = 2; 6167c478bd9Sstevel@tonic-gate mmu.max_level = 1; 6177c478bd9Sstevel@tonic-gate mmu.ptes_per_table = 1024; 6187c478bd9Sstevel@tonic-gate mmu.top_level_count = 1024; 6197c478bd9Sstevel@tonic-gate 6207c478bd9Sstevel@tonic-gate mmu.level_shift[0] = 12; 6217c478bd9Sstevel@tonic-gate mmu.level_shift[1] = 22; 6227c478bd9Sstevel@tonic-gate } 6237c478bd9Sstevel@tonic-gate 6247c478bd9Sstevel@tonic-gate #endif /* __i386 */ 6257c478bd9Sstevel@tonic-gate 6267c478bd9Sstevel@tonic-gate for (i = 0; i < mmu.num_level; ++i) { 6277c478bd9Sstevel@tonic-gate mmu.level_size[i] = 1UL << mmu.level_shift[i]; 6287c478bd9Sstevel@tonic-gate mmu.level_offset[i] = mmu.level_size[i] - 1; 6297c478bd9Sstevel@tonic-gate mmu.level_mask[i] = ~mmu.level_offset[i]; 6307c478bd9Sstevel@tonic-gate } 6317c478bd9Sstevel@tonic-gate 632512cf780Skchow set_max_page_level(); 633512cf780Skchow 634472714d6Skchow mmu_page_sizes = mmu.max_page_level + 1; 635472714d6Skchow mmu_exported_page_sizes = mmu.umax_page_level + 1; 636472714d6Skchow 637472714d6Skchow /* restrict legacy applications from using pagesizes 1g and above */ 638472714d6Skchow mmu_legacy_page_sizes = 639472714d6Skchow (mmu_exported_page_sizes > 2) ? 2 : mmu_exported_page_sizes; 640472714d6Skchow 641472714d6Skchow 642ae115bc7Smrj for (i = 0; i <= mmu.max_page_level; ++i) { 643ab4a9bebSjohnlev mmu.pte_bits[i] = PT_VALID | pt_kern; 644ae115bc7Smrj if (i > 0) 645ae115bc7Smrj mmu.pte_bits[i] |= PT_PAGESIZE; 646ae115bc7Smrj } 6477c478bd9Sstevel@tonic-gate 6487c478bd9Sstevel@tonic-gate /* 6497c478bd9Sstevel@tonic-gate * NOTE Legacy 32 bit PAE mode only has the P_VALID bit at top level. 6507c478bd9Sstevel@tonic-gate */ 6517c478bd9Sstevel@tonic-gate for (i = 1; i < mmu.num_level; ++i) 6527c478bd9Sstevel@tonic-gate mmu.ptp_bits[i] = PT_PTPBITS; 653ae115bc7Smrj 6547c478bd9Sstevel@tonic-gate #if defined(__i386) 6557c478bd9Sstevel@tonic-gate mmu.ptp_bits[2] = PT_VALID; 6567c478bd9Sstevel@tonic-gate #endif 6577c478bd9Sstevel@tonic-gate 6587c478bd9Sstevel@tonic-gate /* 6597c478bd9Sstevel@tonic-gate * Compute how many hash table entries to have per process for htables. 6607c478bd9Sstevel@tonic-gate * We start with 1 page's worth of entries. 6617c478bd9Sstevel@tonic-gate * 6627c478bd9Sstevel@tonic-gate * If physical memory is small, reduce the amount need to cover it. 6637c478bd9Sstevel@tonic-gate */ 6647c478bd9Sstevel@tonic-gate max_htables = physmax / mmu.ptes_per_table; 6657c478bd9Sstevel@tonic-gate mmu.hash_cnt = MMU_PAGESIZE / sizeof (htable_t *); 6667c478bd9Sstevel@tonic-gate while (mmu.hash_cnt > 16 && mmu.hash_cnt >= max_htables) 6677c478bd9Sstevel@tonic-gate mmu.hash_cnt >>= 1; 6687c478bd9Sstevel@tonic-gate mmu.vlp_hash_cnt = mmu.hash_cnt; 6697c478bd9Sstevel@tonic-gate 6707c478bd9Sstevel@tonic-gate #if defined(__amd64) 6717c478bd9Sstevel@tonic-gate /* 6727c478bd9Sstevel@tonic-gate * If running in 64 bits and physical memory is large, 6737c478bd9Sstevel@tonic-gate * increase the size of the cache to cover all of memory for 6747c478bd9Sstevel@tonic-gate * a 64 bit process. 6757c478bd9Sstevel@tonic-gate */ 6767c478bd9Sstevel@tonic-gate #define HASH_MAX_LENGTH 4 6777c478bd9Sstevel@tonic-gate while (mmu.hash_cnt * HASH_MAX_LENGTH < max_htables) 6787c478bd9Sstevel@tonic-gate mmu.hash_cnt <<= 1; 6797c478bd9Sstevel@tonic-gate #endif 6807c478bd9Sstevel@tonic-gate } 6817c478bd9Sstevel@tonic-gate 6827c478bd9Sstevel@tonic-gate 6837c478bd9Sstevel@tonic-gate /* 6847c478bd9Sstevel@tonic-gate * initialize hat data structures 6857c478bd9Sstevel@tonic-gate */ 6867c478bd9Sstevel@tonic-gate void 6877c478bd9Sstevel@tonic-gate hat_init() 6887c478bd9Sstevel@tonic-gate { 6897c478bd9Sstevel@tonic-gate #if defined(__i386) 6907c478bd9Sstevel@tonic-gate /* 6917c478bd9Sstevel@tonic-gate * _userlimit must be aligned correctly 6927c478bd9Sstevel@tonic-gate */ 6937c478bd9Sstevel@tonic-gate if ((_userlimit & LEVEL_MASK(1)) != _userlimit) { 6947c478bd9Sstevel@tonic-gate prom_printf("hat_init(): _userlimit=%p, not aligned at %p\n", 6957c478bd9Sstevel@tonic-gate (void *)_userlimit, (void *)LEVEL_SIZE(1)); 6967c478bd9Sstevel@tonic-gate halt("hat_init(): Unable to continue"); 6977c478bd9Sstevel@tonic-gate } 6987c478bd9Sstevel@tonic-gate #endif 6997c478bd9Sstevel@tonic-gate 7007c478bd9Sstevel@tonic-gate cv_init(&hat_list_cv, NULL, CV_DEFAULT, NULL); 7017c478bd9Sstevel@tonic-gate 7027c478bd9Sstevel@tonic-gate /* 7037c478bd9Sstevel@tonic-gate * initialize kmem caches 7047c478bd9Sstevel@tonic-gate */ 7057c478bd9Sstevel@tonic-gate htable_init(); 7067c478bd9Sstevel@tonic-gate hment_init(); 7077c478bd9Sstevel@tonic-gate 7087c478bd9Sstevel@tonic-gate hat_cache = kmem_cache_create("hat_t", 7097c478bd9Sstevel@tonic-gate sizeof (hat_t), 0, hati_constructor, NULL, NULL, 7107c478bd9Sstevel@tonic-gate NULL, 0, 0); 7117c478bd9Sstevel@tonic-gate 7127c478bd9Sstevel@tonic-gate hat_hash_cache = kmem_cache_create("HatHash", 7137c478bd9Sstevel@tonic-gate mmu.hash_cnt * sizeof (htable_t *), 0, NULL, NULL, NULL, 7147c478bd9Sstevel@tonic-gate NULL, 0, 0); 7157c478bd9Sstevel@tonic-gate 7167c478bd9Sstevel@tonic-gate /* 7177c478bd9Sstevel@tonic-gate * VLP hats can use a smaller hash table size on large memroy machines 7187c478bd9Sstevel@tonic-gate */ 7197c478bd9Sstevel@tonic-gate if (mmu.hash_cnt == mmu.vlp_hash_cnt) { 7207c478bd9Sstevel@tonic-gate vlp_hash_cache = hat_hash_cache; 7217c478bd9Sstevel@tonic-gate } else { 7227c478bd9Sstevel@tonic-gate vlp_hash_cache = kmem_cache_create("HatVlpHash", 7237c478bd9Sstevel@tonic-gate mmu.vlp_hash_cnt * sizeof (htable_t *), 0, NULL, NULL, NULL, 7247c478bd9Sstevel@tonic-gate NULL, 0, 0); 7257c478bd9Sstevel@tonic-gate } 7267c478bd9Sstevel@tonic-gate 7277c478bd9Sstevel@tonic-gate /* 7287c478bd9Sstevel@tonic-gate * Set up the kernel's hat 7297c478bd9Sstevel@tonic-gate */ 730dc32d872SJosef 'Jeff' Sipek AS_LOCK_ENTER(&kas, RW_WRITER); 7317c478bd9Sstevel@tonic-gate kas.a_hat = kmem_cache_alloc(hat_cache, KM_NOSLEEP); 7327c478bd9Sstevel@tonic-gate mutex_init(&kas.a_hat->hat_mutex, NULL, MUTEX_DEFAULT, NULL); 7337c478bd9Sstevel@tonic-gate kas.a_hat->hat_as = &kas; 7347c478bd9Sstevel@tonic-gate kas.a_hat->hat_flags = 0; 735dc32d872SJosef 'Jeff' Sipek AS_LOCK_EXIT(&kas); 7367c478bd9Sstevel@tonic-gate 7377c478bd9Sstevel@tonic-gate CPUSET_ZERO(khat_cpuset); 7387c478bd9Sstevel@tonic-gate CPUSET_ADD(khat_cpuset, CPU->cpu_id); 7397c478bd9Sstevel@tonic-gate 7407c478bd9Sstevel@tonic-gate /* 7417c478bd9Sstevel@tonic-gate * The kernel hat's next pointer serves as the head of the hat list . 742a85a6733Sjosephb * The kernel hat's prev pointer tracks the last hat on the list for 743a85a6733Sjosephb * htable_steal() to use. 7447c478bd9Sstevel@tonic-gate */ 7457c478bd9Sstevel@tonic-gate kas.a_hat->hat_next = NULL; 746a85a6733Sjosephb kas.a_hat->hat_prev = NULL; 7477c478bd9Sstevel@tonic-gate 7487c478bd9Sstevel@tonic-gate /* 7497c478bd9Sstevel@tonic-gate * Allocate an htable hash bucket for the kernel 7507c478bd9Sstevel@tonic-gate * XX64 - tune for 64 bit procs 7517c478bd9Sstevel@tonic-gate */ 7527c478bd9Sstevel@tonic-gate kas.a_hat->hat_num_hash = mmu.hash_cnt; 7537c478bd9Sstevel@tonic-gate kas.a_hat->hat_ht_hash = kmem_cache_alloc(hat_hash_cache, KM_NOSLEEP); 7547c478bd9Sstevel@tonic-gate bzero(kas.a_hat->hat_ht_hash, mmu.hash_cnt * sizeof (htable_t *)); 7557c478bd9Sstevel@tonic-gate 7567c478bd9Sstevel@tonic-gate /* 7577c478bd9Sstevel@tonic-gate * zero out the top level and cached htable pointers 7587c478bd9Sstevel@tonic-gate */ 7597c478bd9Sstevel@tonic-gate kas.a_hat->hat_ht_cached = NULL; 7607c478bd9Sstevel@tonic-gate kas.a_hat->hat_htable = NULL; 7619d9461f9Strevtom 7629d9461f9Strevtom /* 7639d9461f9Strevtom * Pre-allocate hrm_hashtab before enabling the collection of 7649d9461f9Strevtom * refmod statistics. Allocating on the fly would mean us 7659d9461f9Strevtom * running the risk of suffering recursive mutex enters or 7669d9461f9Strevtom * deadlocks. 7679d9461f9Strevtom */ 7689d9461f9Strevtom hrm_hashtab = kmem_zalloc(HRM_HASHSIZE * sizeof (struct hrmstat *), 7699d9461f9Strevtom KM_SLEEP); 7707c478bd9Sstevel@tonic-gate } 7717c478bd9Sstevel@tonic-gate 7727c478bd9Sstevel@tonic-gate /* 7737c478bd9Sstevel@tonic-gate * Prepare CPU specific pagetables for VLP processes on 64 bit kernels. 7747c478bd9Sstevel@tonic-gate * 7757c478bd9Sstevel@tonic-gate * Each CPU has a set of 2 pagetables that are reused for any 32 bit 7767c478bd9Sstevel@tonic-gate * process it runs. They are the top level pagetable, hci_vlp_l3ptes, and 7777c478bd9Sstevel@tonic-gate * the next to top level table for the bottom 512 Gig, hci_vlp_l2ptes. 7787c478bd9Sstevel@tonic-gate */ 7797c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 7807c478bd9Sstevel@tonic-gate static void 7817c478bd9Sstevel@tonic-gate hat_vlp_setup(struct cpu *cpu) 7827c478bd9Sstevel@tonic-gate { 783843e1988Sjohnlev #if defined(__amd64) && !defined(__xpv) 7847c478bd9Sstevel@tonic-gate struct hat_cpu_info *hci = cpu->cpu_hat_info; 7857c478bd9Sstevel@tonic-gate pfn_t pfn; 7867c478bd9Sstevel@tonic-gate 7877c478bd9Sstevel@tonic-gate /* 7887c478bd9Sstevel@tonic-gate * allocate the level==2 page table for the bottom most 7897c478bd9Sstevel@tonic-gate * 512Gig of address space (this is where 32 bit apps live) 7907c478bd9Sstevel@tonic-gate */ 7917c478bd9Sstevel@tonic-gate ASSERT(hci != NULL); 7927c478bd9Sstevel@tonic-gate hci->hci_vlp_l2ptes = kmem_zalloc(MMU_PAGESIZE, KM_SLEEP); 7937c478bd9Sstevel@tonic-gate 7947c478bd9Sstevel@tonic-gate /* 7957c478bd9Sstevel@tonic-gate * Allocate a top level pagetable and copy the kernel's 7967c478bd9Sstevel@tonic-gate * entries into it. Then link in hci_vlp_l2ptes in the 1st entry. 7977c478bd9Sstevel@tonic-gate */ 7987c478bd9Sstevel@tonic-gate hci->hci_vlp_l3ptes = kmem_zalloc(MMU_PAGESIZE, KM_SLEEP); 7997c478bd9Sstevel@tonic-gate hci->hci_vlp_pfn = 8007c478bd9Sstevel@tonic-gate hat_getpfnum(kas.a_hat, (caddr_t)hci->hci_vlp_l3ptes); 8017c478bd9Sstevel@tonic-gate ASSERT(hci->hci_vlp_pfn != PFN_INVALID); 802843e1988Sjohnlev bcopy(vlp_page, hci->hci_vlp_l3ptes, MMU_PAGESIZE); 8037c478bd9Sstevel@tonic-gate 8047c478bd9Sstevel@tonic-gate pfn = hat_getpfnum(kas.a_hat, (caddr_t)hci->hci_vlp_l2ptes); 8057c478bd9Sstevel@tonic-gate ASSERT(pfn != PFN_INVALID); 8067c478bd9Sstevel@tonic-gate hci->hci_vlp_l3ptes[0] = MAKEPTP(pfn, 2); 807843e1988Sjohnlev #endif /* __amd64 && !__xpv */ 8087c478bd9Sstevel@tonic-gate } 8097c478bd9Sstevel@tonic-gate 810ae115bc7Smrj /*ARGSUSED*/ 811ae115bc7Smrj static void 812ae115bc7Smrj hat_vlp_teardown(cpu_t *cpu) 813ae115bc7Smrj { 814843e1988Sjohnlev #if defined(__amd64) && !defined(__xpv) 815ae115bc7Smrj struct hat_cpu_info *hci; 816ae115bc7Smrj 817ae115bc7Smrj if ((hci = cpu->cpu_hat_info) == NULL) 818ae115bc7Smrj return; 819ae115bc7Smrj if (hci->hci_vlp_l2ptes) 820ae115bc7Smrj kmem_free(hci->hci_vlp_l2ptes, MMU_PAGESIZE); 821ae115bc7Smrj if (hci->hci_vlp_l3ptes) 822ae115bc7Smrj kmem_free(hci->hci_vlp_l3ptes, MMU_PAGESIZE); 823843e1988Sjohnlev #endif 824843e1988Sjohnlev } 825843e1988Sjohnlev 826843e1988Sjohnlev #define NEXT_HKR(r, l, s, e) { \ 827843e1988Sjohnlev kernel_ranges[r].hkr_level = l; \ 828843e1988Sjohnlev kernel_ranges[r].hkr_start_va = s; \ 829843e1988Sjohnlev kernel_ranges[r].hkr_end_va = e; \ 830843e1988Sjohnlev ++r; \ 831ae115bc7Smrj } 832ae115bc7Smrj 8337c478bd9Sstevel@tonic-gate /* 8347c478bd9Sstevel@tonic-gate * Finish filling in the kernel hat. 8357c478bd9Sstevel@tonic-gate * Pre fill in all top level kernel page table entries for the kernel's 8367c478bd9Sstevel@tonic-gate * part of the address range. From this point on we can't use any new 8377c478bd9Sstevel@tonic-gate * kernel large pages if they need PTE's at max_level 838ae115bc7Smrj * 839ae115bc7Smrj * create the kmap mappings. 8407c478bd9Sstevel@tonic-gate */ 8417c478bd9Sstevel@tonic-gate void 8427c478bd9Sstevel@tonic-gate hat_init_finish(void) 8437c478bd9Sstevel@tonic-gate { 844ae115bc7Smrj size_t size; 845843e1988Sjohnlev uint_t r = 0; 846843e1988Sjohnlev uintptr_t va; 847843e1988Sjohnlev hat_kernel_range_t *rp; 8487c478bd9Sstevel@tonic-gate 8497c478bd9Sstevel@tonic-gate 8507c478bd9Sstevel@tonic-gate /* 851843e1988Sjohnlev * We are now effectively running on the kernel hat. 852843e1988Sjohnlev * Clearing use_boot_reserve shuts off using the pre-allocated boot 853843e1988Sjohnlev * reserve for all HAT allocations. From here on, the reserves are 854843e1988Sjohnlev * only used when avoiding recursion in kmem_alloc(). 8557c478bd9Sstevel@tonic-gate */ 856843e1988Sjohnlev use_boot_reserve = 0; 857843e1988Sjohnlev htable_adjust_reserve(); 8587c478bd9Sstevel@tonic-gate 8597c478bd9Sstevel@tonic-gate /* 860843e1988Sjohnlev * User HATs are initialized with copies of all kernel mappings in 861843e1988Sjohnlev * higher level page tables. Ensure that those entries exist. 8627c478bd9Sstevel@tonic-gate */ 863843e1988Sjohnlev #if defined(__amd64) 864843e1988Sjohnlev 865843e1988Sjohnlev NEXT_HKR(r, 3, kernelbase, 0); 866843e1988Sjohnlev #if defined(__xpv) 867843e1988Sjohnlev NEXT_HKR(r, 3, HYPERVISOR_VIRT_START, HYPERVISOR_VIRT_END); 868843e1988Sjohnlev #endif 869843e1988Sjohnlev 870843e1988Sjohnlev #elif defined(__i386) 871843e1988Sjohnlev 872843e1988Sjohnlev #if !defined(__xpv) 873843e1988Sjohnlev if (mmu.pae_hat) { 874843e1988Sjohnlev va = kernelbase; 875843e1988Sjohnlev if ((va & LEVEL_MASK(2)) != va) { 876843e1988Sjohnlev va = P2ROUNDUP(va, LEVEL_SIZE(2)); 877843e1988Sjohnlev NEXT_HKR(r, 1, kernelbase, va); 878843e1988Sjohnlev } 879843e1988Sjohnlev if (va != 0) 880843e1988Sjohnlev NEXT_HKR(r, 2, va, 0); 881843e1988Sjohnlev } else 882843e1988Sjohnlev #endif /* __xpv */ 883843e1988Sjohnlev NEXT_HKR(r, 1, kernelbase, 0); 884843e1988Sjohnlev 885843e1988Sjohnlev #endif /* __i386 */ 886843e1988Sjohnlev 887843e1988Sjohnlev num_kernel_ranges = r; 8887c478bd9Sstevel@tonic-gate 8897c478bd9Sstevel@tonic-gate /* 890843e1988Sjohnlev * Create all the kernel pagetables that will have entries 891843e1988Sjohnlev * shared to user HATs. 8927c478bd9Sstevel@tonic-gate */ 893843e1988Sjohnlev for (r = 0; r < num_kernel_ranges; ++r) { 894843e1988Sjohnlev rp = &kernel_ranges[r]; 895843e1988Sjohnlev for (va = rp->hkr_start_va; va != rp->hkr_end_va; 896843e1988Sjohnlev va += LEVEL_SIZE(rp->hkr_level)) { 897843e1988Sjohnlev htable_t *ht; 898843e1988Sjohnlev 899843e1988Sjohnlev if (IN_HYPERVISOR_VA(va)) 900843e1988Sjohnlev continue; 901843e1988Sjohnlev 902843e1988Sjohnlev /* can/must skip if a page mapping already exists */ 903843e1988Sjohnlev if (rp->hkr_level <= mmu.max_page_level && 904843e1988Sjohnlev (ht = htable_getpage(kas.a_hat, va, NULL)) != 905843e1988Sjohnlev NULL) { 906843e1988Sjohnlev htable_release(ht); 907843e1988Sjohnlev continue; 908843e1988Sjohnlev } 909843e1988Sjohnlev 910843e1988Sjohnlev (void) htable_create(kas.a_hat, va, rp->hkr_level - 1, 911843e1988Sjohnlev NULL); 912843e1988Sjohnlev } 913843e1988Sjohnlev } 9147c478bd9Sstevel@tonic-gate 9157c478bd9Sstevel@tonic-gate /* 916843e1988Sjohnlev * 32 bit PAE metal kernels use only 4 of the 512 entries in the 917843e1988Sjohnlev * page holding the top level pagetable. We use the remainder for 918843e1988Sjohnlev * the "per CPU" page tables for VLP processes. 919843e1988Sjohnlev * Map the top level kernel pagetable into the kernel to make 920843e1988Sjohnlev * it easy to use bcopy access these tables. 9217c478bd9Sstevel@tonic-gate */ 9227c478bd9Sstevel@tonic-gate if (mmu.pae_hat) { 9237c478bd9Sstevel@tonic-gate vlp_page = vmem_alloc(heap_arena, MMU_PAGESIZE, VM_SLEEP); 9247c478bd9Sstevel@tonic-gate hat_devload(kas.a_hat, (caddr_t)vlp_page, MMU_PAGESIZE, 9257c478bd9Sstevel@tonic-gate kas.a_hat->hat_htable->ht_pfn, 926843e1988Sjohnlev #if !defined(__xpv) 927ae115bc7Smrj PROT_WRITE | 928843e1988Sjohnlev #endif 929ae115bc7Smrj PROT_READ | HAT_NOSYNC | HAT_UNORDERED_OK, 9307c478bd9Sstevel@tonic-gate HAT_LOAD | HAT_LOAD_NOCONSIST); 9317c478bd9Sstevel@tonic-gate } 9327c478bd9Sstevel@tonic-gate hat_vlp_setup(CPU); 933ae115bc7Smrj 934ae115bc7Smrj /* 935ae115bc7Smrj * Create kmap (cached mappings of kernel PTEs) 936ae115bc7Smrj * for 32 bit we map from segmap_start .. ekernelheap 937ae115bc7Smrj * for 64 bit we map from segmap_start .. segmap_start + segmapsize; 938ae115bc7Smrj */ 939ae115bc7Smrj #if defined(__i386) 940ae115bc7Smrj size = (uintptr_t)ekernelheap - segmap_start; 941ae115bc7Smrj #elif defined(__amd64) 942ae115bc7Smrj size = segmapsize; 943ae115bc7Smrj #endif 944ae115bc7Smrj hat_kmap_init((uintptr_t)segmap_start, size); 9457c478bd9Sstevel@tonic-gate } 9467c478bd9Sstevel@tonic-gate 9477c478bd9Sstevel@tonic-gate /* 9487c478bd9Sstevel@tonic-gate * On 32 bit PAE mode, PTE's are 64 bits, but ordinary atomic memory references 94975d94465SJosef 'Jeff' Sipek * are 32 bit, so for safety we must use atomic_cas_64() to install these. 9507c478bd9Sstevel@tonic-gate */ 9517c478bd9Sstevel@tonic-gate #ifdef __i386 9527c478bd9Sstevel@tonic-gate static void 9537c478bd9Sstevel@tonic-gate reload_pae32(hat_t *hat, cpu_t *cpu) 9547c478bd9Sstevel@tonic-gate { 9557c478bd9Sstevel@tonic-gate x86pte_t *src; 9567c478bd9Sstevel@tonic-gate x86pte_t *dest; 9577c478bd9Sstevel@tonic-gate x86pte_t pte; 9587c478bd9Sstevel@tonic-gate int i; 9597c478bd9Sstevel@tonic-gate 9607c478bd9Sstevel@tonic-gate /* 9617c478bd9Sstevel@tonic-gate * Load the 4 entries of the level 2 page table into this 9627c478bd9Sstevel@tonic-gate * cpu's range of the vlp_page and point cr3 at them. 9637c478bd9Sstevel@tonic-gate */ 9647c478bd9Sstevel@tonic-gate ASSERT(mmu.pae_hat); 9657c478bd9Sstevel@tonic-gate src = hat->hat_vlp_ptes; 9667c478bd9Sstevel@tonic-gate dest = vlp_page + (cpu->cpu_id + 1) * VLP_NUM_PTES; 9677c478bd9Sstevel@tonic-gate for (i = 0; i < VLP_NUM_PTES; ++i) { 9687c478bd9Sstevel@tonic-gate for (;;) { 9697c478bd9Sstevel@tonic-gate pte = dest[i]; 9707c478bd9Sstevel@tonic-gate if (pte == src[i]) 9717c478bd9Sstevel@tonic-gate break; 97275d94465SJosef 'Jeff' Sipek if (atomic_cas_64(dest + i, pte, src[i]) != src[i]) 9737c478bd9Sstevel@tonic-gate break; 9747c478bd9Sstevel@tonic-gate } 9757c478bd9Sstevel@tonic-gate } 9767c478bd9Sstevel@tonic-gate } 9777c478bd9Sstevel@tonic-gate #endif 9787c478bd9Sstevel@tonic-gate 9797c478bd9Sstevel@tonic-gate /* 9807c478bd9Sstevel@tonic-gate * Switch to a new active hat, maintaining bit masks to track active CPUs. 981843e1988Sjohnlev * 982843e1988Sjohnlev * On the 32-bit PAE hypervisor, %cr3 is a 64-bit value, on metal it 983843e1988Sjohnlev * remains a 32-bit value. 9847c478bd9Sstevel@tonic-gate */ 9857c478bd9Sstevel@tonic-gate void 9867c478bd9Sstevel@tonic-gate hat_switch(hat_t *hat) 9877c478bd9Sstevel@tonic-gate { 988843e1988Sjohnlev uint64_t newcr3; 9897c478bd9Sstevel@tonic-gate cpu_t *cpu = CPU; 9907c478bd9Sstevel@tonic-gate hat_t *old = cpu->cpu_current_hat; 9917c478bd9Sstevel@tonic-gate 9927c478bd9Sstevel@tonic-gate /* 9937c478bd9Sstevel@tonic-gate * set up this information first, so we don't miss any cross calls 9947c478bd9Sstevel@tonic-gate */ 9957c478bd9Sstevel@tonic-gate if (old != NULL) { 9967c478bd9Sstevel@tonic-gate if (old == hat) 9977c478bd9Sstevel@tonic-gate return; 9987c478bd9Sstevel@tonic-gate if (old != kas.a_hat) 9997c478bd9Sstevel@tonic-gate CPUSET_ATOMIC_DEL(old->hat_cpus, cpu->cpu_id); 10007c478bd9Sstevel@tonic-gate } 10017c478bd9Sstevel@tonic-gate 10027c478bd9Sstevel@tonic-gate /* 100395c0a3c8Sjosephb * Add this CPU to the active set for this HAT. 10047c478bd9Sstevel@tonic-gate */ 10057c478bd9Sstevel@tonic-gate if (hat != kas.a_hat) { 10067c478bd9Sstevel@tonic-gate CPUSET_ATOMIC_ADD(hat->hat_cpus, cpu->cpu_id); 10077c478bd9Sstevel@tonic-gate } 10087c478bd9Sstevel@tonic-gate cpu->cpu_current_hat = hat; 10097c478bd9Sstevel@tonic-gate 10107c478bd9Sstevel@tonic-gate /* 10117c478bd9Sstevel@tonic-gate * now go ahead and load cr3 10127c478bd9Sstevel@tonic-gate */ 10137c478bd9Sstevel@tonic-gate if (hat->hat_flags & HAT_VLP) { 10147c478bd9Sstevel@tonic-gate #if defined(__amd64) 10157c478bd9Sstevel@tonic-gate x86pte_t *vlpptep = cpu->cpu_hat_info->hci_vlp_l2ptes; 10167c478bd9Sstevel@tonic-gate 10177c478bd9Sstevel@tonic-gate VLP_COPY(hat->hat_vlp_ptes, vlpptep); 10187c478bd9Sstevel@tonic-gate newcr3 = MAKECR3(cpu->cpu_hat_info->hci_vlp_pfn); 10197c478bd9Sstevel@tonic-gate #elif defined(__i386) 10207c478bd9Sstevel@tonic-gate reload_pae32(hat, cpu); 10217c478bd9Sstevel@tonic-gate newcr3 = MAKECR3(kas.a_hat->hat_htable->ht_pfn) + 10227c478bd9Sstevel@tonic-gate (cpu->cpu_id + 1) * VLP_SIZE; 10237c478bd9Sstevel@tonic-gate #endif 10247c478bd9Sstevel@tonic-gate } else { 1025843e1988Sjohnlev newcr3 = MAKECR3((uint64_t)hat->hat_htable->ht_pfn); 10267c478bd9Sstevel@tonic-gate } 1027843e1988Sjohnlev #ifdef __xpv 1028843e1988Sjohnlev { 1029843e1988Sjohnlev struct mmuext_op t[2]; 1030843e1988Sjohnlev uint_t retcnt; 1031843e1988Sjohnlev uint_t opcnt = 1; 1032843e1988Sjohnlev 1033843e1988Sjohnlev t[0].cmd = MMUEXT_NEW_BASEPTR; 1034843e1988Sjohnlev t[0].arg1.mfn = mmu_btop(pa_to_ma(newcr3)); 1035843e1988Sjohnlev #if defined(__amd64) 1036843e1988Sjohnlev /* 1037843e1988Sjohnlev * There's an interesting problem here, as to what to 1038843e1988Sjohnlev * actually specify when switching to the kernel hat. 1039843e1988Sjohnlev * For now we'll reuse the kernel hat again. 1040843e1988Sjohnlev */ 1041843e1988Sjohnlev t[1].cmd = MMUEXT_NEW_USER_BASEPTR; 1042843e1988Sjohnlev if (hat == kas.a_hat) 1043843e1988Sjohnlev t[1].arg1.mfn = mmu_btop(pa_to_ma(newcr3)); 1044843e1988Sjohnlev else 1045843e1988Sjohnlev t[1].arg1.mfn = pfn_to_mfn(hat->hat_user_ptable); 1046843e1988Sjohnlev ++opcnt; 1047843e1988Sjohnlev #endif /* __amd64 */ 1048843e1988Sjohnlev if (HYPERVISOR_mmuext_op(t, opcnt, &retcnt, DOMID_SELF) < 0) 1049843e1988Sjohnlev panic("HYPERVISOR_mmu_update() failed"); 1050843e1988Sjohnlev ASSERT(retcnt == opcnt); 1051843e1988Sjohnlev 1052843e1988Sjohnlev } 1053843e1988Sjohnlev #else 10547c478bd9Sstevel@tonic-gate setcr3(newcr3); 1055843e1988Sjohnlev #endif 10567c478bd9Sstevel@tonic-gate ASSERT(cpu == CPU); 10577c478bd9Sstevel@tonic-gate } 10587c478bd9Sstevel@tonic-gate 10597c478bd9Sstevel@tonic-gate /* 10607c478bd9Sstevel@tonic-gate * Utility to return a valid x86pte_t from protections, pfn, and level number 10617c478bd9Sstevel@tonic-gate */ 10627c478bd9Sstevel@tonic-gate static x86pte_t 10637c478bd9Sstevel@tonic-gate hati_mkpte(pfn_t pfn, uint_t attr, level_t level, uint_t flags) 10647c478bd9Sstevel@tonic-gate { 10657c478bd9Sstevel@tonic-gate x86pte_t pte; 10667c478bd9Sstevel@tonic-gate uint_t cache_attr = attr & HAT_ORDER_MASK; 10677c478bd9Sstevel@tonic-gate 10687c478bd9Sstevel@tonic-gate pte = MAKEPTE(pfn, level); 10697c478bd9Sstevel@tonic-gate 10707c478bd9Sstevel@tonic-gate if (attr & PROT_WRITE) 10717c478bd9Sstevel@tonic-gate PTE_SET(pte, PT_WRITABLE); 10727c478bd9Sstevel@tonic-gate 10737c478bd9Sstevel@tonic-gate if (attr & PROT_USER) 10747c478bd9Sstevel@tonic-gate PTE_SET(pte, PT_USER); 10757c478bd9Sstevel@tonic-gate 10767c478bd9Sstevel@tonic-gate if (!(attr & PROT_EXEC)) 10777c478bd9Sstevel@tonic-gate PTE_SET(pte, mmu.pt_nx); 10787c478bd9Sstevel@tonic-gate 10797c478bd9Sstevel@tonic-gate /* 1080ae115bc7Smrj * Set the software bits used track ref/mod sync's and hments. 1081ae115bc7Smrj * If not using REF/MOD, set them to avoid h/w rewriting PTEs. 10827c478bd9Sstevel@tonic-gate */ 10837c478bd9Sstevel@tonic-gate if (flags & HAT_LOAD_NOCONSIST) 1084ae115bc7Smrj PTE_SET(pte, PT_NOCONSIST | PT_REF | PT_MOD); 1085ae115bc7Smrj else if (attr & HAT_NOSYNC) 1086ae115bc7Smrj PTE_SET(pte, PT_NOSYNC | PT_REF | PT_MOD); 10877c478bd9Sstevel@tonic-gate 10887c478bd9Sstevel@tonic-gate /* 10897c478bd9Sstevel@tonic-gate * Set the caching attributes in the PTE. The combination 10907c478bd9Sstevel@tonic-gate * of attributes are poorly defined, so we pay attention 10917c478bd9Sstevel@tonic-gate * to them in the given order. 10927c478bd9Sstevel@tonic-gate * 10937c478bd9Sstevel@tonic-gate * The test for HAT_STRICTORDER is different because it's defined 10947c478bd9Sstevel@tonic-gate * as "0" - which was a stupid thing to do, but is too late to change! 10957c478bd9Sstevel@tonic-gate */ 10967c478bd9Sstevel@tonic-gate if (cache_attr == HAT_STRICTORDER) { 10977c478bd9Sstevel@tonic-gate PTE_SET(pte, PT_NOCACHE); 10987c478bd9Sstevel@tonic-gate /*LINTED [Lint hates empty ifs, but it's the obvious way to do this] */ 10997c478bd9Sstevel@tonic-gate } else if (cache_attr & (HAT_UNORDERED_OK | HAT_STORECACHING_OK)) { 11007c478bd9Sstevel@tonic-gate /* nothing to set */; 11017c478bd9Sstevel@tonic-gate } else if (cache_attr & (HAT_MERGING_OK | HAT_LOADCACHING_OK)) { 11027c478bd9Sstevel@tonic-gate PTE_SET(pte, PT_NOCACHE); 11037417cfdeSKuriakose Kuruvilla if (is_x86_feature(x86_featureset, X86FSET_PAT)) 11047c478bd9Sstevel@tonic-gate PTE_SET(pte, (level == 0) ? PT_PAT_4K : PT_PAT_LARGE); 11057c478bd9Sstevel@tonic-gate else 11067c478bd9Sstevel@tonic-gate PTE_SET(pte, PT_WRITETHRU); 11077c478bd9Sstevel@tonic-gate } else { 11087c478bd9Sstevel@tonic-gate panic("hati_mkpte(): bad caching attributes: %x\n", cache_attr); 11097c478bd9Sstevel@tonic-gate } 11107c478bd9Sstevel@tonic-gate 11117c478bd9Sstevel@tonic-gate return (pte); 11127c478bd9Sstevel@tonic-gate } 11137c478bd9Sstevel@tonic-gate 11147c478bd9Sstevel@tonic-gate /* 11157c478bd9Sstevel@tonic-gate * Duplicate address translations of the parent to the child. 11167c478bd9Sstevel@tonic-gate * This function really isn't used anymore. 11177c478bd9Sstevel@tonic-gate */ 11187c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 11197c478bd9Sstevel@tonic-gate int 11207c478bd9Sstevel@tonic-gate hat_dup(hat_t *old, hat_t *new, caddr_t addr, size_t len, uint_t flag) 11217c478bd9Sstevel@tonic-gate { 11227c478bd9Sstevel@tonic-gate ASSERT((uintptr_t)addr < kernelbase); 11237c478bd9Sstevel@tonic-gate ASSERT(new != kas.a_hat); 11247c478bd9Sstevel@tonic-gate ASSERT(old != kas.a_hat); 11257c478bd9Sstevel@tonic-gate return (0); 11267c478bd9Sstevel@tonic-gate } 11277c478bd9Sstevel@tonic-gate 11287c478bd9Sstevel@tonic-gate /* 11297c478bd9Sstevel@tonic-gate * Allocate any hat resources required for a process being swapped in. 11307c478bd9Sstevel@tonic-gate */ 11317c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 11327c478bd9Sstevel@tonic-gate void 11337c478bd9Sstevel@tonic-gate hat_swapin(hat_t *hat) 11347c478bd9Sstevel@tonic-gate { 11357c478bd9Sstevel@tonic-gate /* do nothing - we let everything fault back in */ 11367c478bd9Sstevel@tonic-gate } 11377c478bd9Sstevel@tonic-gate 11387c478bd9Sstevel@tonic-gate /* 11397c478bd9Sstevel@tonic-gate * Unload all translations associated with an address space of a process 11407c478bd9Sstevel@tonic-gate * that is being swapped out. 11417c478bd9Sstevel@tonic-gate */ 11427c478bd9Sstevel@tonic-gate void 11437c478bd9Sstevel@tonic-gate hat_swapout(hat_t *hat) 11447c478bd9Sstevel@tonic-gate { 11457c478bd9Sstevel@tonic-gate uintptr_t vaddr = (uintptr_t)0; 11467c478bd9Sstevel@tonic-gate uintptr_t eaddr = _userlimit; 11477c478bd9Sstevel@tonic-gate htable_t *ht = NULL; 11487c478bd9Sstevel@tonic-gate level_t l; 11497c478bd9Sstevel@tonic-gate 1150843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 11517c478bd9Sstevel@tonic-gate /* 11527c478bd9Sstevel@tonic-gate * We can't just call hat_unload(hat, 0, _userlimit...) here, because 11537c478bd9Sstevel@tonic-gate * seg_spt and shared pagetables can't be swapped out. 11547c478bd9Sstevel@tonic-gate * Take a look at segspt_shmswapout() - it's a big no-op. 11557c478bd9Sstevel@tonic-gate * 11567c478bd9Sstevel@tonic-gate * Instead we'll walk through all the address space and unload 11577c478bd9Sstevel@tonic-gate * any mappings which we are sure are not shared, not locked. 11587c478bd9Sstevel@tonic-gate */ 11597c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(vaddr)); 11607c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(eaddr)); 1161dc32d872SJosef 'Jeff' Sipek ASSERT(AS_LOCK_HELD(hat->hat_as)); 11627c478bd9Sstevel@tonic-gate if ((uintptr_t)hat->hat_as->a_userlimit < eaddr) 11637c478bd9Sstevel@tonic-gate eaddr = (uintptr_t)hat->hat_as->a_userlimit; 11647c478bd9Sstevel@tonic-gate 11657c478bd9Sstevel@tonic-gate while (vaddr < eaddr) { 11667c478bd9Sstevel@tonic-gate (void) htable_walk(hat, &ht, &vaddr, eaddr); 11677c478bd9Sstevel@tonic-gate if (ht == NULL) 11687c478bd9Sstevel@tonic-gate break; 11697c478bd9Sstevel@tonic-gate 11707c478bd9Sstevel@tonic-gate ASSERT(!IN_VA_HOLE(vaddr)); 11717c478bd9Sstevel@tonic-gate 11727c478bd9Sstevel@tonic-gate /* 11737c478bd9Sstevel@tonic-gate * If the page table is shared skip its entire range. 11747c478bd9Sstevel@tonic-gate */ 11757c478bd9Sstevel@tonic-gate l = ht->ht_level; 11767c478bd9Sstevel@tonic-gate if (ht->ht_flags & HTABLE_SHARED_PFN) { 1177ae320ee6Speterte vaddr = ht->ht_vaddr + LEVEL_SIZE(l + 1); 11787c478bd9Sstevel@tonic-gate htable_release(ht); 11797c478bd9Sstevel@tonic-gate ht = NULL; 11807c478bd9Sstevel@tonic-gate continue; 11817c478bd9Sstevel@tonic-gate } 11827c478bd9Sstevel@tonic-gate 11837c478bd9Sstevel@tonic-gate /* 11847c478bd9Sstevel@tonic-gate * If the page table has no locked entries, unload this one. 11857c478bd9Sstevel@tonic-gate */ 11867c478bd9Sstevel@tonic-gate if (ht->ht_lock_cnt == 0) 11877c478bd9Sstevel@tonic-gate hat_unload(hat, (caddr_t)vaddr, LEVEL_SIZE(l), 11887c478bd9Sstevel@tonic-gate HAT_UNLOAD_UNMAP); 11897c478bd9Sstevel@tonic-gate 11907c478bd9Sstevel@tonic-gate /* 11917c478bd9Sstevel@tonic-gate * If we have a level 0 page table with locked entries, 11927c478bd9Sstevel@tonic-gate * skip the entire page table, otherwise skip just one entry. 11937c478bd9Sstevel@tonic-gate */ 11947c478bd9Sstevel@tonic-gate if (ht->ht_lock_cnt > 0 && l == 0) 11957c478bd9Sstevel@tonic-gate vaddr = ht->ht_vaddr + LEVEL_SIZE(1); 11967c478bd9Sstevel@tonic-gate else 11977c478bd9Sstevel@tonic-gate vaddr += LEVEL_SIZE(l); 11987c478bd9Sstevel@tonic-gate } 11997c478bd9Sstevel@tonic-gate if (ht) 12007c478bd9Sstevel@tonic-gate htable_release(ht); 12017c478bd9Sstevel@tonic-gate 12027c478bd9Sstevel@tonic-gate /* 12037c478bd9Sstevel@tonic-gate * We're in swapout because the system is low on memory, so 12047c478bd9Sstevel@tonic-gate * go back and flush all the htables off the cached list. 12057c478bd9Sstevel@tonic-gate */ 12067c478bd9Sstevel@tonic-gate htable_purge_hat(hat); 1207843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 12087c478bd9Sstevel@tonic-gate } 12097c478bd9Sstevel@tonic-gate 12107c478bd9Sstevel@tonic-gate /* 12117c478bd9Sstevel@tonic-gate * returns number of bytes that have valid mappings in hat. 12127c478bd9Sstevel@tonic-gate */ 12137c478bd9Sstevel@tonic-gate size_t 12147c478bd9Sstevel@tonic-gate hat_get_mapped_size(hat_t *hat) 12157c478bd9Sstevel@tonic-gate { 12167c478bd9Sstevel@tonic-gate size_t total = 0; 12177c478bd9Sstevel@tonic-gate int l; 12187c478bd9Sstevel@tonic-gate 12197c478bd9Sstevel@tonic-gate for (l = 0; l <= mmu.max_page_level; l++) 12207c478bd9Sstevel@tonic-gate total += (hat->hat_pages_mapped[l] << LEVEL_SHIFT(l)); 1221250b7ff9Sjosephb total += hat->hat_ism_pgcnt; 12227c478bd9Sstevel@tonic-gate 12237c478bd9Sstevel@tonic-gate return (total); 12247c478bd9Sstevel@tonic-gate } 12257c478bd9Sstevel@tonic-gate 12267c478bd9Sstevel@tonic-gate /* 12277c478bd9Sstevel@tonic-gate * enable/disable collection of stats for hat. 12287c478bd9Sstevel@tonic-gate */ 12297c478bd9Sstevel@tonic-gate int 12307c478bd9Sstevel@tonic-gate hat_stats_enable(hat_t *hat) 12317c478bd9Sstevel@tonic-gate { 12321a5e258fSJosef 'Jeff' Sipek atomic_inc_32(&hat->hat_stats); 12337c478bd9Sstevel@tonic-gate return (1); 12347c478bd9Sstevel@tonic-gate } 12357c478bd9Sstevel@tonic-gate 12367c478bd9Sstevel@tonic-gate void 12377c478bd9Sstevel@tonic-gate hat_stats_disable(hat_t *hat) 12387c478bd9Sstevel@tonic-gate { 12391a5e258fSJosef 'Jeff' Sipek atomic_dec_32(&hat->hat_stats); 12407c478bd9Sstevel@tonic-gate } 12417c478bd9Sstevel@tonic-gate 12427c478bd9Sstevel@tonic-gate /* 12437c478bd9Sstevel@tonic-gate * Utility to sync the ref/mod bits from a page table entry to the page_t 12447c478bd9Sstevel@tonic-gate * We must be holding the mapping list lock when this is called. 12457c478bd9Sstevel@tonic-gate */ 12467c478bd9Sstevel@tonic-gate static void 12477c478bd9Sstevel@tonic-gate hati_sync_pte_to_page(page_t *pp, x86pte_t pte, level_t level) 12487c478bd9Sstevel@tonic-gate { 12497c478bd9Sstevel@tonic-gate uint_t rm = 0; 12507c478bd9Sstevel@tonic-gate pgcnt_t pgcnt; 12517c478bd9Sstevel@tonic-gate 1252ae115bc7Smrj if (PTE_GET(pte, PT_SOFTWARE) >= PT_NOSYNC) 12537c478bd9Sstevel@tonic-gate return; 12547c478bd9Sstevel@tonic-gate 12557c478bd9Sstevel@tonic-gate if (PTE_GET(pte, PT_REF)) 12567c478bd9Sstevel@tonic-gate rm |= P_REF; 12577c478bd9Sstevel@tonic-gate 12587c478bd9Sstevel@tonic-gate if (PTE_GET(pte, PT_MOD)) 12597c478bd9Sstevel@tonic-gate rm |= P_MOD; 12607c478bd9Sstevel@tonic-gate 12617c478bd9Sstevel@tonic-gate if (rm == 0) 12627c478bd9Sstevel@tonic-gate return; 12637c478bd9Sstevel@tonic-gate 12647c478bd9Sstevel@tonic-gate /* 12657c478bd9Sstevel@tonic-gate * sync to all constituent pages of a large page 12667c478bd9Sstevel@tonic-gate */ 12677c478bd9Sstevel@tonic-gate ASSERT(x86_hm_held(pp)); 12687c478bd9Sstevel@tonic-gate pgcnt = page_get_pagecnt(level); 12697c478bd9Sstevel@tonic-gate ASSERT(IS_P2ALIGNED(pp->p_pagenum, pgcnt)); 12707c478bd9Sstevel@tonic-gate for (; pgcnt > 0; --pgcnt) { 12717c478bd9Sstevel@tonic-gate /* 12727c478bd9Sstevel@tonic-gate * hat_page_demote() can't decrease 12737c478bd9Sstevel@tonic-gate * pszc below this mapping size 12747c478bd9Sstevel@tonic-gate * since this large mapping existed after we 12757c478bd9Sstevel@tonic-gate * took mlist lock. 12767c478bd9Sstevel@tonic-gate */ 12777c478bd9Sstevel@tonic-gate ASSERT(pp->p_szc >= level); 12787c478bd9Sstevel@tonic-gate hat_page_setattr(pp, rm); 12797c478bd9Sstevel@tonic-gate ++pp; 12807c478bd9Sstevel@tonic-gate } 12817c478bd9Sstevel@tonic-gate } 12827c478bd9Sstevel@tonic-gate 12837c478bd9Sstevel@tonic-gate /* 12847c478bd9Sstevel@tonic-gate * This the set of PTE bits for PFN, permissions and caching 1285843e1988Sjohnlev * that are allowed to change on a HAT_LOAD_REMAP 12867c478bd9Sstevel@tonic-gate */ 12877c478bd9Sstevel@tonic-gate #define PT_REMAP_BITS \ 12887c478bd9Sstevel@tonic-gate (PT_PADDR | PT_NX | PT_WRITABLE | PT_WRITETHRU | \ 1289843e1988Sjohnlev PT_NOCACHE | PT_PAT_4K | PT_PAT_LARGE | PT_IGNORE | PT_REF | PT_MOD) 12907c478bd9Sstevel@tonic-gate 1291b193e412Skchow #define REMAPASSERT(EX) if (!(EX)) panic("hati_pte_map: " #EX) 12927c478bd9Sstevel@tonic-gate /* 12937c478bd9Sstevel@tonic-gate * Do the low-level work to get a mapping entered into a HAT's pagetables 12947c478bd9Sstevel@tonic-gate * and in the mapping list of the associated page_t. 12957c478bd9Sstevel@tonic-gate */ 1296ae115bc7Smrj static int 12977c478bd9Sstevel@tonic-gate hati_pte_map( 12987c478bd9Sstevel@tonic-gate htable_t *ht, 12997c478bd9Sstevel@tonic-gate uint_t entry, 13007c478bd9Sstevel@tonic-gate page_t *pp, 13017c478bd9Sstevel@tonic-gate x86pte_t pte, 13027c478bd9Sstevel@tonic-gate int flags, 13037c478bd9Sstevel@tonic-gate void *pte_ptr) 13047c478bd9Sstevel@tonic-gate { 13057c478bd9Sstevel@tonic-gate hat_t *hat = ht->ht_hat; 13067c478bd9Sstevel@tonic-gate x86pte_t old_pte; 13077c478bd9Sstevel@tonic-gate level_t l = ht->ht_level; 13087c478bd9Sstevel@tonic-gate hment_t *hm; 13097c478bd9Sstevel@tonic-gate uint_t is_consist; 131025540de2SJakub Jermar uint_t is_locked; 1311ae115bc7Smrj int rv = 0; 13127c478bd9Sstevel@tonic-gate 13137c478bd9Sstevel@tonic-gate /* 131421584dbcSPavel Tatashin * Is this a consistent (ie. need mapping list lock) mapping? 13157c478bd9Sstevel@tonic-gate */ 13167c478bd9Sstevel@tonic-gate is_consist = (pp != NULL && (flags & HAT_LOAD_NOCONSIST) == 0); 13177c478bd9Sstevel@tonic-gate 13187c478bd9Sstevel@tonic-gate /* 13197c478bd9Sstevel@tonic-gate * Track locked mapping count in the htable. Do this first, 13207c478bd9Sstevel@tonic-gate * as we track locking even if there already is a mapping present. 13217c478bd9Sstevel@tonic-gate */ 132225540de2SJakub Jermar is_locked = (flags & HAT_LOAD_LOCK) != 0 && hat != kas.a_hat; 132325540de2SJakub Jermar if (is_locked) 13247c478bd9Sstevel@tonic-gate HTABLE_LOCK_INC(ht); 13257c478bd9Sstevel@tonic-gate 13267c478bd9Sstevel@tonic-gate /* 13277c478bd9Sstevel@tonic-gate * Acquire the page's mapping list lock and get an hment to use. 13287c478bd9Sstevel@tonic-gate * Note that hment_prepare() might return NULL. 13297c478bd9Sstevel@tonic-gate */ 13307c478bd9Sstevel@tonic-gate if (is_consist) { 13317c478bd9Sstevel@tonic-gate x86_hm_enter(pp); 13327c478bd9Sstevel@tonic-gate hm = hment_prepare(ht, entry, pp); 13337c478bd9Sstevel@tonic-gate } 13347c478bd9Sstevel@tonic-gate 13357c478bd9Sstevel@tonic-gate /* 13367c478bd9Sstevel@tonic-gate * Set the new pte, retrieving the old one at the same time. 13377c478bd9Sstevel@tonic-gate */ 13387c478bd9Sstevel@tonic-gate old_pte = x86pte_set(ht, entry, pte, pte_ptr); 13397c478bd9Sstevel@tonic-gate 13407c478bd9Sstevel@tonic-gate /* 134125540de2SJakub Jermar * Did we get a large page / page table collision? 13427c478bd9Sstevel@tonic-gate */ 1343ae115bc7Smrj if (old_pte == LPAGE_ERROR) { 134425540de2SJakub Jermar if (is_locked) 134525540de2SJakub Jermar HTABLE_LOCK_DEC(ht); 1346ae115bc7Smrj rv = -1; 1347ae115bc7Smrj goto done; 13487c478bd9Sstevel@tonic-gate } 13497c478bd9Sstevel@tonic-gate 1350ae115bc7Smrj /* 1351ae115bc7Smrj * If the mapping didn't change there is nothing more to do. 1352ae115bc7Smrj */ 1353ae115bc7Smrj if (PTE_EQUIV(pte, old_pte)) 1354ae115bc7Smrj goto done; 1355ae115bc7Smrj 13567c478bd9Sstevel@tonic-gate /* 13577c478bd9Sstevel@tonic-gate * Install a new mapping in the page's mapping list 13587c478bd9Sstevel@tonic-gate */ 13597c478bd9Sstevel@tonic-gate if (!PTE_ISVALID(old_pte)) { 13607c478bd9Sstevel@tonic-gate if (is_consist) { 13617c478bd9Sstevel@tonic-gate hment_assign(ht, entry, pp, hm); 13627c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 13637c478bd9Sstevel@tonic-gate } else { 13647c478bd9Sstevel@tonic-gate ASSERT(flags & HAT_LOAD_NOCONSIST); 13657c478bd9Sstevel@tonic-gate } 136602bc52beSkchow #if defined(__amd64) 136702bc52beSkchow if (ht->ht_flags & HTABLE_VLP) { 136802bc52beSkchow cpu_t *cpu = CPU; 136902bc52beSkchow x86pte_t *vlpptep = cpu->cpu_hat_info->hci_vlp_l2ptes; 137002bc52beSkchow VLP_COPY(hat->hat_vlp_ptes, vlpptep); 137102bc52beSkchow } 137202bc52beSkchow #endif 13737c478bd9Sstevel@tonic-gate HTABLE_INC(ht->ht_valid_cnt); 13747c478bd9Sstevel@tonic-gate PGCNT_INC(hat, l); 1375ae115bc7Smrj return (rv); 13767c478bd9Sstevel@tonic-gate } 13777c478bd9Sstevel@tonic-gate 13787c478bd9Sstevel@tonic-gate /* 13797c478bd9Sstevel@tonic-gate * Remap's are more complicated: 13807c478bd9Sstevel@tonic-gate * - HAT_LOAD_REMAP must be specified if changing the pfn. 13817c478bd9Sstevel@tonic-gate * We also require that NOCONSIST be specified. 13827c478bd9Sstevel@tonic-gate * - Otherwise only permission or caching bits may change. 13837c478bd9Sstevel@tonic-gate */ 13847c478bd9Sstevel@tonic-gate if (!PTE_ISPAGE(old_pte, l)) 13857c478bd9Sstevel@tonic-gate panic("non-null/page mapping pte=" FMT_PTE, old_pte); 13867c478bd9Sstevel@tonic-gate 13877c478bd9Sstevel@tonic-gate if (PTE2PFN(old_pte, l) != PTE2PFN(pte, l)) { 1388b193e412Skchow REMAPASSERT(flags & HAT_LOAD_REMAP); 1389b193e412Skchow REMAPASSERT(flags & HAT_LOAD_NOCONSIST); 1390ae115bc7Smrj REMAPASSERT(PTE_GET(old_pte, PT_SOFTWARE) >= PT_NOCONSIST); 1391b193e412Skchow REMAPASSERT(pf_is_memory(PTE2PFN(old_pte, l)) == 13927c478bd9Sstevel@tonic-gate pf_is_memory(PTE2PFN(pte, l))); 1393b193e412Skchow REMAPASSERT(!is_consist); 13947c478bd9Sstevel@tonic-gate } 13957c478bd9Sstevel@tonic-gate 13967c478bd9Sstevel@tonic-gate /* 1397843e1988Sjohnlev * We only let remaps change the certain bits in the PTE. 13987c478bd9Sstevel@tonic-gate */ 1399843e1988Sjohnlev if (PTE_GET(old_pte, ~PT_REMAP_BITS) != PTE_GET(pte, ~PT_REMAP_BITS)) 1400843e1988Sjohnlev panic("remap bits changed: old_pte="FMT_PTE", pte="FMT_PTE"\n", 1401843e1988Sjohnlev old_pte, pte); 14027c478bd9Sstevel@tonic-gate 14037c478bd9Sstevel@tonic-gate /* 14047c478bd9Sstevel@tonic-gate * We don't create any mapping list entries on a remap, so release 14057c478bd9Sstevel@tonic-gate * any allocated hment after we drop the mapping list lock. 14067c478bd9Sstevel@tonic-gate */ 1407ae115bc7Smrj done: 14087c478bd9Sstevel@tonic-gate if (is_consist) { 14097c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 14107c478bd9Sstevel@tonic-gate if (hm != NULL) 14117c478bd9Sstevel@tonic-gate hment_free(hm); 14127c478bd9Sstevel@tonic-gate } 1413ae115bc7Smrj return (rv); 14147c478bd9Sstevel@tonic-gate } 14157c478bd9Sstevel@tonic-gate 14167c478bd9Sstevel@tonic-gate /* 1417ae115bc7Smrj * Internal routine to load a single page table entry. This only fails if 1418ae115bc7Smrj * we attempt to overwrite a page table link with a large page. 14197c478bd9Sstevel@tonic-gate */ 1420ae115bc7Smrj static int 14217c478bd9Sstevel@tonic-gate hati_load_common( 14227c478bd9Sstevel@tonic-gate hat_t *hat, 14237c478bd9Sstevel@tonic-gate uintptr_t va, 14247c478bd9Sstevel@tonic-gate page_t *pp, 14257c478bd9Sstevel@tonic-gate uint_t attr, 14267c478bd9Sstevel@tonic-gate uint_t flags, 14277c478bd9Sstevel@tonic-gate level_t level, 14287c478bd9Sstevel@tonic-gate pfn_t pfn) 14297c478bd9Sstevel@tonic-gate { 14307c478bd9Sstevel@tonic-gate htable_t *ht; 14317c478bd9Sstevel@tonic-gate uint_t entry; 14327c478bd9Sstevel@tonic-gate x86pte_t pte; 1433ae115bc7Smrj int rv = 0; 14347c478bd9Sstevel@tonic-gate 1435aac11643Sjosephb /* 1436aac11643Sjosephb * The number 16 is arbitrary and here to catch a recursion problem 1437aac11643Sjosephb * early before we blow out the kernel stack. 1438aac11643Sjosephb */ 1439aac11643Sjosephb ++curthread->t_hatdepth; 1440aac11643Sjosephb ASSERT(curthread->t_hatdepth < 16); 1441aac11643Sjosephb 1442dc32d872SJosef 'Jeff' Sipek ASSERT(hat == kas.a_hat || AS_LOCK_HELD(hat->hat_as)); 14437c478bd9Sstevel@tonic-gate 14447c478bd9Sstevel@tonic-gate if (flags & HAT_LOAD_SHARE) 14457c478bd9Sstevel@tonic-gate hat->hat_flags |= HAT_SHARED; 14467c478bd9Sstevel@tonic-gate 14477c478bd9Sstevel@tonic-gate /* 14487c478bd9Sstevel@tonic-gate * Find the page table that maps this page if it already exists. 14497c478bd9Sstevel@tonic-gate */ 14507c478bd9Sstevel@tonic-gate ht = htable_lookup(hat, va, level); 14517c478bd9Sstevel@tonic-gate 14527c478bd9Sstevel@tonic-gate /* 1453aac11643Sjosephb * We must have HAT_LOAD_NOCONSIST if page_t is NULL. 14547c478bd9Sstevel@tonic-gate */ 1455aac11643Sjosephb if (pp == NULL) 14567c478bd9Sstevel@tonic-gate flags |= HAT_LOAD_NOCONSIST; 14577c478bd9Sstevel@tonic-gate 14587c478bd9Sstevel@tonic-gate if (ht == NULL) { 14597c478bd9Sstevel@tonic-gate ht = htable_create(hat, va, level, NULL); 14607c478bd9Sstevel@tonic-gate ASSERT(ht != NULL); 14617c478bd9Sstevel@tonic-gate } 14627c478bd9Sstevel@tonic-gate entry = htable_va2entry(va, ht); 14637c478bd9Sstevel@tonic-gate 14647c478bd9Sstevel@tonic-gate /* 14657c478bd9Sstevel@tonic-gate * a bunch of paranoid error checking 14667c478bd9Sstevel@tonic-gate */ 14677c478bd9Sstevel@tonic-gate ASSERT(ht->ht_busy > 0); 14687c478bd9Sstevel@tonic-gate if (ht->ht_vaddr > va || va > HTABLE_LAST_PAGE(ht)) 1469903a11ebSrh panic("hati_load_common: bad htable %p, va %p", 1470903a11ebSrh (void *)ht, (void *)va); 14717c478bd9Sstevel@tonic-gate ASSERT(ht->ht_level == level); 14727c478bd9Sstevel@tonic-gate 14737c478bd9Sstevel@tonic-gate /* 14747c478bd9Sstevel@tonic-gate * construct the new PTE 14757c478bd9Sstevel@tonic-gate */ 14767c478bd9Sstevel@tonic-gate if (hat == kas.a_hat) 14777c478bd9Sstevel@tonic-gate attr &= ~PROT_USER; 14787c478bd9Sstevel@tonic-gate pte = hati_mkpte(pfn, attr, level, flags); 14797c478bd9Sstevel@tonic-gate if (hat == kas.a_hat && va >= kernelbase) 14807c478bd9Sstevel@tonic-gate PTE_SET(pte, mmu.pt_global); 14817c478bd9Sstevel@tonic-gate 14827c478bd9Sstevel@tonic-gate /* 14837c478bd9Sstevel@tonic-gate * establish the mapping 14847c478bd9Sstevel@tonic-gate */ 1485ae115bc7Smrj rv = hati_pte_map(ht, entry, pp, pte, flags, NULL); 14867c478bd9Sstevel@tonic-gate 14877c478bd9Sstevel@tonic-gate /* 14887c478bd9Sstevel@tonic-gate * release the htable and any reserves 14897c478bd9Sstevel@tonic-gate */ 14907c478bd9Sstevel@tonic-gate htable_release(ht); 1491aac11643Sjosephb --curthread->t_hatdepth; 1492ae115bc7Smrj return (rv); 14937c478bd9Sstevel@tonic-gate } 14947c478bd9Sstevel@tonic-gate 14957c478bd9Sstevel@tonic-gate /* 14967c478bd9Sstevel@tonic-gate * special case of hat_memload to deal with some kernel addrs for performance 14977c478bd9Sstevel@tonic-gate */ 14987c478bd9Sstevel@tonic-gate static void 14997c478bd9Sstevel@tonic-gate hat_kmap_load( 15007c478bd9Sstevel@tonic-gate caddr_t addr, 15017c478bd9Sstevel@tonic-gate page_t *pp, 15027c478bd9Sstevel@tonic-gate uint_t attr, 15037c478bd9Sstevel@tonic-gate uint_t flags) 15047c478bd9Sstevel@tonic-gate { 15057c478bd9Sstevel@tonic-gate uintptr_t va = (uintptr_t)addr; 15067c478bd9Sstevel@tonic-gate x86pte_t pte; 15077c478bd9Sstevel@tonic-gate pfn_t pfn = page_pptonum(pp); 15087c478bd9Sstevel@tonic-gate pgcnt_t pg_off = mmu_btop(va - mmu.kmap_addr); 15097c478bd9Sstevel@tonic-gate htable_t *ht; 15107c478bd9Sstevel@tonic-gate uint_t entry; 15117c478bd9Sstevel@tonic-gate void *pte_ptr; 15127c478bd9Sstevel@tonic-gate 15137c478bd9Sstevel@tonic-gate /* 15147c478bd9Sstevel@tonic-gate * construct the requested PTE 15157c478bd9Sstevel@tonic-gate */ 15167c478bd9Sstevel@tonic-gate attr &= ~PROT_USER; 15177c478bd9Sstevel@tonic-gate attr |= HAT_STORECACHING_OK; 15187c478bd9Sstevel@tonic-gate pte = hati_mkpte(pfn, attr, 0, flags); 15197c478bd9Sstevel@tonic-gate PTE_SET(pte, mmu.pt_global); 15207c478bd9Sstevel@tonic-gate 15217c478bd9Sstevel@tonic-gate /* 15227c478bd9Sstevel@tonic-gate * Figure out the pte_ptr and htable and use common code to finish up 15237c478bd9Sstevel@tonic-gate */ 15247c478bd9Sstevel@tonic-gate if (mmu.pae_hat) 15257c478bd9Sstevel@tonic-gate pte_ptr = mmu.kmap_ptes + pg_off; 15267c478bd9Sstevel@tonic-gate else 15277c478bd9Sstevel@tonic-gate pte_ptr = (x86pte32_t *)mmu.kmap_ptes + pg_off; 15287c478bd9Sstevel@tonic-gate ht = mmu.kmap_htables[(va - mmu.kmap_htables[0]->ht_vaddr) >> 15297c478bd9Sstevel@tonic-gate LEVEL_SHIFT(1)]; 15307c478bd9Sstevel@tonic-gate entry = htable_va2entry(va, ht); 1531aac11643Sjosephb ++curthread->t_hatdepth; 1532aac11643Sjosephb ASSERT(curthread->t_hatdepth < 16); 1533ae115bc7Smrj (void) hati_pte_map(ht, entry, pp, pte, flags, pte_ptr); 1534aac11643Sjosephb --curthread->t_hatdepth; 15357c478bd9Sstevel@tonic-gate } 15367c478bd9Sstevel@tonic-gate 15377c478bd9Sstevel@tonic-gate /* 15387c478bd9Sstevel@tonic-gate * hat_memload() - load a translation to the given page struct 15397c478bd9Sstevel@tonic-gate * 15407c478bd9Sstevel@tonic-gate * Flags for hat_memload/hat_devload/hat_*attr. 15417c478bd9Sstevel@tonic-gate * 15427c478bd9Sstevel@tonic-gate * HAT_LOAD Default flags to load a translation to the page. 15437c478bd9Sstevel@tonic-gate * 15447c478bd9Sstevel@tonic-gate * HAT_LOAD_LOCK Lock down mapping resources; hat_map(), hat_memload(), 15457c478bd9Sstevel@tonic-gate * and hat_devload(). 15467c478bd9Sstevel@tonic-gate * 15477c478bd9Sstevel@tonic-gate * HAT_LOAD_NOCONSIST Do not add mapping to page_t mapping list. 1548ae115bc7Smrj * sets PT_NOCONSIST 15497c478bd9Sstevel@tonic-gate * 15507c478bd9Sstevel@tonic-gate * HAT_LOAD_SHARE A flag to hat_memload() to indicate h/w page tables 15517c478bd9Sstevel@tonic-gate * that map some user pages (not kas) is shared by more 15527c478bd9Sstevel@tonic-gate * than one process (eg. ISM). 15537c478bd9Sstevel@tonic-gate * 15547c478bd9Sstevel@tonic-gate * HAT_LOAD_REMAP Reload a valid pte with a different page frame. 15557c478bd9Sstevel@tonic-gate * 15567c478bd9Sstevel@tonic-gate * HAT_NO_KALLOC Do not kmem_alloc while creating the mapping; at this 15577c478bd9Sstevel@tonic-gate * point, it's setting up mapping to allocate internal 15587c478bd9Sstevel@tonic-gate * hat layer data structures. This flag forces hat layer 15597c478bd9Sstevel@tonic-gate * to tap its reserves in order to prevent infinite 15607c478bd9Sstevel@tonic-gate * recursion. 15617c478bd9Sstevel@tonic-gate * 15627c478bd9Sstevel@tonic-gate * The following is a protection attribute (like PROT_READ, etc.) 15637c478bd9Sstevel@tonic-gate * 1564ae115bc7Smrj * HAT_NOSYNC set PT_NOSYNC - this mapping's ref/mod bits 15657c478bd9Sstevel@tonic-gate * are never cleared. 15667c478bd9Sstevel@tonic-gate * 15677c478bd9Sstevel@tonic-gate * Installing new valid PTE's and creation of the mapping list 15687c478bd9Sstevel@tonic-gate * entry are controlled under the same lock. It's derived from the 15697c478bd9Sstevel@tonic-gate * page_t being mapped. 15707c478bd9Sstevel@tonic-gate */ 15717c478bd9Sstevel@tonic-gate static uint_t supported_memload_flags = 15727c478bd9Sstevel@tonic-gate HAT_LOAD | HAT_LOAD_LOCK | HAT_LOAD_ADV | HAT_LOAD_NOCONSIST | 15737c478bd9Sstevel@tonic-gate HAT_LOAD_SHARE | HAT_NO_KALLOC | HAT_LOAD_REMAP | HAT_LOAD_TEXT; 15747c478bd9Sstevel@tonic-gate 15757c478bd9Sstevel@tonic-gate void 15767c478bd9Sstevel@tonic-gate hat_memload( 15777c478bd9Sstevel@tonic-gate hat_t *hat, 15787c478bd9Sstevel@tonic-gate caddr_t addr, 15797c478bd9Sstevel@tonic-gate page_t *pp, 15807c478bd9Sstevel@tonic-gate uint_t attr, 15817c478bd9Sstevel@tonic-gate uint_t flags) 15827c478bd9Sstevel@tonic-gate { 15837c478bd9Sstevel@tonic-gate uintptr_t va = (uintptr_t)addr; 15847c478bd9Sstevel@tonic-gate level_t level = 0; 15857c478bd9Sstevel@tonic-gate pfn_t pfn = page_pptonum(pp); 15867c478bd9Sstevel@tonic-gate 1587843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 15887c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(va)); 1589ae115bc7Smrj ASSERT(hat == kas.a_hat || va < _userlimit); 1590dc32d872SJosef 'Jeff' Sipek ASSERT(hat == kas.a_hat || AS_LOCK_HELD(hat->hat_as)); 15917c478bd9Sstevel@tonic-gate ASSERT((flags & supported_memload_flags) == flags); 15927c478bd9Sstevel@tonic-gate 15937c478bd9Sstevel@tonic-gate ASSERT(!IN_VA_HOLE(va)); 15947c478bd9Sstevel@tonic-gate ASSERT(!PP_ISFREE(pp)); 15957c478bd9Sstevel@tonic-gate 15967c478bd9Sstevel@tonic-gate /* 15977c478bd9Sstevel@tonic-gate * kernel address special case for performance. 15987c478bd9Sstevel@tonic-gate */ 15997c478bd9Sstevel@tonic-gate if (mmu.kmap_addr <= va && va < mmu.kmap_eaddr) { 16007c478bd9Sstevel@tonic-gate ASSERT(hat == kas.a_hat); 16017c478bd9Sstevel@tonic-gate hat_kmap_load(addr, pp, attr, flags); 1602843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 16037c478bd9Sstevel@tonic-gate return; 16047c478bd9Sstevel@tonic-gate } 16057c478bd9Sstevel@tonic-gate 16067c478bd9Sstevel@tonic-gate /* 16077c478bd9Sstevel@tonic-gate * This is used for memory with normal caching enabled, so 16087c478bd9Sstevel@tonic-gate * always set HAT_STORECACHING_OK. 16097c478bd9Sstevel@tonic-gate */ 16107c478bd9Sstevel@tonic-gate attr |= HAT_STORECACHING_OK; 1611ae115bc7Smrj if (hati_load_common(hat, va, pp, attr, flags, level, pfn) != 0) 1612ae115bc7Smrj panic("unexpected hati_load_common() failure"); 1613843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 16147c478bd9Sstevel@tonic-gate } 16157c478bd9Sstevel@tonic-gate 161605d3dc4bSpaulsan /* ARGSUSED */ 161705d3dc4bSpaulsan void 161805d3dc4bSpaulsan hat_memload_region(struct hat *hat, caddr_t addr, struct page *pp, 161905d3dc4bSpaulsan uint_t attr, uint_t flags, hat_region_cookie_t rcookie) 162005d3dc4bSpaulsan { 162105d3dc4bSpaulsan hat_memload(hat, addr, pp, attr, flags); 162205d3dc4bSpaulsan } 162305d3dc4bSpaulsan 16247c478bd9Sstevel@tonic-gate /* 16257c478bd9Sstevel@tonic-gate * Load the given array of page structs using large pages when possible 16267c478bd9Sstevel@tonic-gate */ 16277c478bd9Sstevel@tonic-gate void 16287c478bd9Sstevel@tonic-gate hat_memload_array( 16297c478bd9Sstevel@tonic-gate hat_t *hat, 16307c478bd9Sstevel@tonic-gate caddr_t addr, 16317c478bd9Sstevel@tonic-gate size_t len, 16327c478bd9Sstevel@tonic-gate page_t **pages, 16337c478bd9Sstevel@tonic-gate uint_t attr, 16347c478bd9Sstevel@tonic-gate uint_t flags) 16357c478bd9Sstevel@tonic-gate { 16367c478bd9Sstevel@tonic-gate uintptr_t va = (uintptr_t)addr; 16377c478bd9Sstevel@tonic-gate uintptr_t eaddr = va + len; 16387c478bd9Sstevel@tonic-gate level_t level; 16397c478bd9Sstevel@tonic-gate size_t pgsize; 16407c478bd9Sstevel@tonic-gate pgcnt_t pgindx = 0; 16417c478bd9Sstevel@tonic-gate pfn_t pfn; 16427c478bd9Sstevel@tonic-gate pgcnt_t i; 16437c478bd9Sstevel@tonic-gate 1644843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 16457c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(va)); 1646ae115bc7Smrj ASSERT(hat == kas.a_hat || va + len <= _userlimit); 1647dc32d872SJosef 'Jeff' Sipek ASSERT(hat == kas.a_hat || AS_LOCK_HELD(hat->hat_as)); 16487c478bd9Sstevel@tonic-gate ASSERT((flags & supported_memload_flags) == flags); 16497c478bd9Sstevel@tonic-gate 16507c478bd9Sstevel@tonic-gate /* 16517c478bd9Sstevel@tonic-gate * memload is used for memory with full caching enabled, so 16527c478bd9Sstevel@tonic-gate * set HAT_STORECACHING_OK. 16537c478bd9Sstevel@tonic-gate */ 16547c478bd9Sstevel@tonic-gate attr |= HAT_STORECACHING_OK; 16557c478bd9Sstevel@tonic-gate 16567c478bd9Sstevel@tonic-gate /* 16577c478bd9Sstevel@tonic-gate * handle all pages using largest possible pagesize 16587c478bd9Sstevel@tonic-gate */ 16597c478bd9Sstevel@tonic-gate while (va < eaddr) { 16607c478bd9Sstevel@tonic-gate /* 16617c478bd9Sstevel@tonic-gate * decide what level mapping to use (ie. pagesize) 16627c478bd9Sstevel@tonic-gate */ 16637c478bd9Sstevel@tonic-gate pfn = page_pptonum(pages[pgindx]); 16647c478bd9Sstevel@tonic-gate for (level = mmu.max_page_level; ; --level) { 16657c478bd9Sstevel@tonic-gate pgsize = LEVEL_SIZE(level); 16667c478bd9Sstevel@tonic-gate if (level == 0) 16677c478bd9Sstevel@tonic-gate break; 1668ae115bc7Smrj 16697c478bd9Sstevel@tonic-gate if (!IS_P2ALIGNED(va, pgsize) || 16707c478bd9Sstevel@tonic-gate (eaddr - va) < pgsize || 1671ae115bc7Smrj !IS_P2ALIGNED(pfn_to_pa(pfn), pgsize)) 16727c478bd9Sstevel@tonic-gate continue; 16737c478bd9Sstevel@tonic-gate 16747c478bd9Sstevel@tonic-gate /* 16757c478bd9Sstevel@tonic-gate * To use a large mapping of this size, all the 16767c478bd9Sstevel@tonic-gate * pages we are passed must be sequential subpages 16777c478bd9Sstevel@tonic-gate * of the large page. 16787c478bd9Sstevel@tonic-gate * hat_page_demote() can't change p_szc because 16797c478bd9Sstevel@tonic-gate * all pages are locked. 16807c478bd9Sstevel@tonic-gate */ 16817c478bd9Sstevel@tonic-gate if (pages[pgindx]->p_szc >= level) { 16827c478bd9Sstevel@tonic-gate for (i = 0; i < mmu_btop(pgsize); ++i) { 16837c478bd9Sstevel@tonic-gate if (pfn + i != 16847c478bd9Sstevel@tonic-gate page_pptonum(pages[pgindx + i])) 16857c478bd9Sstevel@tonic-gate break; 16867c478bd9Sstevel@tonic-gate ASSERT(pages[pgindx + i]->p_szc >= 16877c478bd9Sstevel@tonic-gate level); 16887c478bd9Sstevel@tonic-gate ASSERT(pages[pgindx] + i == 16897c478bd9Sstevel@tonic-gate pages[pgindx + i]); 16907c478bd9Sstevel@tonic-gate } 169102bc52beSkchow if (i == mmu_btop(pgsize)) { 169202bc52beSkchow #ifdef DEBUG 169302bc52beSkchow if (level == 2) 169402bc52beSkchow map1gcnt++; 169502bc52beSkchow #endif 16967c478bd9Sstevel@tonic-gate break; 169702bc52beSkchow } 16987c478bd9Sstevel@tonic-gate } 16997c478bd9Sstevel@tonic-gate } 17007c478bd9Sstevel@tonic-gate 17017c478bd9Sstevel@tonic-gate /* 1702ae115bc7Smrj * Load this page mapping. If the load fails, try a smaller 1703ae115bc7Smrj * pagesize. 17047c478bd9Sstevel@tonic-gate */ 17057c478bd9Sstevel@tonic-gate ASSERT(!IN_VA_HOLE(va)); 1706ae115bc7Smrj while (hati_load_common(hat, va, pages[pgindx], attr, 1707250b7ff9Sjosephb flags, level, pfn) != 0) { 1708ae115bc7Smrj if (level == 0) 1709ae115bc7Smrj panic("unexpected hati_load_common() failure"); 1710ae115bc7Smrj --level; 1711ae115bc7Smrj pgsize = LEVEL_SIZE(level); 1712ae115bc7Smrj } 17137c478bd9Sstevel@tonic-gate 17147c478bd9Sstevel@tonic-gate /* 17157c478bd9Sstevel@tonic-gate * move to next page 17167c478bd9Sstevel@tonic-gate */ 17177c478bd9Sstevel@tonic-gate va += pgsize; 17187c478bd9Sstevel@tonic-gate pgindx += mmu_btop(pgsize); 17197c478bd9Sstevel@tonic-gate } 1720843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 17217c478bd9Sstevel@tonic-gate } 17227c478bd9Sstevel@tonic-gate 172305d3dc4bSpaulsan /* ARGSUSED */ 172405d3dc4bSpaulsan void 172505d3dc4bSpaulsan hat_memload_array_region(struct hat *hat, caddr_t addr, size_t len, 172605d3dc4bSpaulsan struct page **pps, uint_t attr, uint_t flags, 172705d3dc4bSpaulsan hat_region_cookie_t rcookie) 172805d3dc4bSpaulsan { 172905d3dc4bSpaulsan hat_memload_array(hat, addr, len, pps, attr, flags); 173005d3dc4bSpaulsan } 173105d3dc4bSpaulsan 17327c478bd9Sstevel@tonic-gate /* 17337c478bd9Sstevel@tonic-gate * void hat_devload(hat, addr, len, pf, attr, flags) 17347c478bd9Sstevel@tonic-gate * load/lock the given page frame number 17357c478bd9Sstevel@tonic-gate * 17367c478bd9Sstevel@tonic-gate * Advisory ordering attributes. Apply only to device mappings. 17377c478bd9Sstevel@tonic-gate * 17387c478bd9Sstevel@tonic-gate * HAT_STRICTORDER: the CPU must issue the references in order, as the 17397c478bd9Sstevel@tonic-gate * programmer specified. This is the default. 17407c478bd9Sstevel@tonic-gate * HAT_UNORDERED_OK: the CPU may reorder the references (this is all kinds 17417c478bd9Sstevel@tonic-gate * of reordering; store or load with store or load). 17427c478bd9Sstevel@tonic-gate * HAT_MERGING_OK: merging and batching: the CPU may merge individual stores 17437c478bd9Sstevel@tonic-gate * to consecutive locations (for example, turn two consecutive byte 17447c478bd9Sstevel@tonic-gate * stores into one halfword store), and it may batch individual loads 17457c478bd9Sstevel@tonic-gate * (for example, turn two consecutive byte loads into one halfword load). 17467c478bd9Sstevel@tonic-gate * This also implies re-ordering. 17477c478bd9Sstevel@tonic-gate * HAT_LOADCACHING_OK: the CPU may cache the data it fetches and reuse it 17487c478bd9Sstevel@tonic-gate * until another store occurs. The default is to fetch new data 17497c478bd9Sstevel@tonic-gate * on every load. This also implies merging. 17507c478bd9Sstevel@tonic-gate * HAT_STORECACHING_OK: the CPU may keep the data in the cache and push it to 17517c478bd9Sstevel@tonic-gate * the device (perhaps with other data) at a later time. The default is 17527c478bd9Sstevel@tonic-gate * to push the data right away. This also implies load caching. 17537c478bd9Sstevel@tonic-gate * 17547c478bd9Sstevel@tonic-gate * Equivalent of hat_memload(), but can be used for device memory where 17557c478bd9Sstevel@tonic-gate * there are no page_t's and we support additional flags (write merging, etc). 17567c478bd9Sstevel@tonic-gate * Note that we can have large page mappings with this interface. 17577c478bd9Sstevel@tonic-gate */ 17587c478bd9Sstevel@tonic-gate int supported_devload_flags = HAT_LOAD | HAT_LOAD_LOCK | 17597c478bd9Sstevel@tonic-gate HAT_LOAD_NOCONSIST | HAT_STRICTORDER | HAT_UNORDERED_OK | 17607c478bd9Sstevel@tonic-gate HAT_MERGING_OK | HAT_LOADCACHING_OK | HAT_STORECACHING_OK; 17617c478bd9Sstevel@tonic-gate 17627c478bd9Sstevel@tonic-gate void 17637c478bd9Sstevel@tonic-gate hat_devload( 17647c478bd9Sstevel@tonic-gate hat_t *hat, 17657c478bd9Sstevel@tonic-gate caddr_t addr, 17667c478bd9Sstevel@tonic-gate size_t len, 17677c478bd9Sstevel@tonic-gate pfn_t pfn, 17687c478bd9Sstevel@tonic-gate uint_t attr, 17697c478bd9Sstevel@tonic-gate int flags) 17707c478bd9Sstevel@tonic-gate { 17717c478bd9Sstevel@tonic-gate uintptr_t va = ALIGN2PAGE(addr); 17727c478bd9Sstevel@tonic-gate uintptr_t eva = va + len; 17737c478bd9Sstevel@tonic-gate level_t level; 17747c478bd9Sstevel@tonic-gate size_t pgsize; 17757c478bd9Sstevel@tonic-gate page_t *pp; 17767c478bd9Sstevel@tonic-gate int f; /* per PTE copy of flags - maybe modified */ 17777c478bd9Sstevel@tonic-gate uint_t a; /* per PTE copy of attr */ 17787c478bd9Sstevel@tonic-gate 1779843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 17807c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(va)); 1781ae115bc7Smrj ASSERT(hat == kas.a_hat || eva <= _userlimit); 1782dc32d872SJosef 'Jeff' Sipek ASSERT(hat == kas.a_hat || AS_LOCK_HELD(hat->hat_as)); 17837c478bd9Sstevel@tonic-gate ASSERT((flags & supported_devload_flags) == flags); 17847c478bd9Sstevel@tonic-gate 17857c478bd9Sstevel@tonic-gate /* 17867c478bd9Sstevel@tonic-gate * handle all pages 17877c478bd9Sstevel@tonic-gate */ 17887c478bd9Sstevel@tonic-gate while (va < eva) { 17897c478bd9Sstevel@tonic-gate 17907c478bd9Sstevel@tonic-gate /* 17917c478bd9Sstevel@tonic-gate * decide what level mapping to use (ie. pagesize) 17927c478bd9Sstevel@tonic-gate */ 17937c478bd9Sstevel@tonic-gate for (level = mmu.max_page_level; ; --level) { 17947c478bd9Sstevel@tonic-gate pgsize = LEVEL_SIZE(level); 17957c478bd9Sstevel@tonic-gate if (level == 0) 17967c478bd9Sstevel@tonic-gate break; 17977c478bd9Sstevel@tonic-gate if (IS_P2ALIGNED(va, pgsize) && 17987c478bd9Sstevel@tonic-gate (eva - va) >= pgsize && 179902bc52beSkchow IS_P2ALIGNED(pfn, mmu_btop(pgsize))) { 180002bc52beSkchow #ifdef DEBUG 180102bc52beSkchow if (level == 2) 180202bc52beSkchow map1gcnt++; 180302bc52beSkchow #endif 18047c478bd9Sstevel@tonic-gate break; 180502bc52beSkchow } 18067c478bd9Sstevel@tonic-gate } 18077c478bd9Sstevel@tonic-gate 18087c478bd9Sstevel@tonic-gate /* 1809ae115bc7Smrj * If this is just memory then allow caching (this happens 18107c478bd9Sstevel@tonic-gate * for the nucleus pages) - though HAT_PLAT_NOCACHE can be used 1811ae115bc7Smrj * to override that. If we don't have a page_t then make sure 18127c478bd9Sstevel@tonic-gate * NOCONSIST is set. 18137c478bd9Sstevel@tonic-gate */ 18147c478bd9Sstevel@tonic-gate a = attr; 18157c478bd9Sstevel@tonic-gate f = flags; 1816843e1988Sjohnlev if (!pf_is_memory(pfn)) 1817843e1988Sjohnlev f |= HAT_LOAD_NOCONSIST; 1818843e1988Sjohnlev else if (!(a & HAT_PLAT_NOCACHE)) 1819843e1988Sjohnlev a |= HAT_STORECACHING_OK; 18207c478bd9Sstevel@tonic-gate 1821843e1988Sjohnlev if (f & HAT_LOAD_NOCONSIST) 18227c478bd9Sstevel@tonic-gate pp = NULL; 1823843e1988Sjohnlev else 1824843e1988Sjohnlev pp = page_numtopp_nolock(pfn); 18257c478bd9Sstevel@tonic-gate 1826e803c1e6SPrakash Sangappa /* 1827e803c1e6SPrakash Sangappa * Check to make sure we are really trying to map a valid 1828e803c1e6SPrakash Sangappa * memory page. The caller wishing to intentionally map 1829e803c1e6SPrakash Sangappa * free memory pages will have passed the HAT_LOAD_NOCONSIST 1830e803c1e6SPrakash Sangappa * flag, then pp will be NULL. 1831e803c1e6SPrakash Sangappa */ 1832e803c1e6SPrakash Sangappa if (pp != NULL) { 1833e803c1e6SPrakash Sangappa if (PP_ISFREE(pp)) { 1834e803c1e6SPrakash Sangappa panic("hat_devload: loading " 1835e803c1e6SPrakash Sangappa "a mapping to free page %p", (void *)pp); 1836e803c1e6SPrakash Sangappa } 1837e803c1e6SPrakash Sangappa 1838e803c1e6SPrakash Sangappa if (!PAGE_LOCKED(pp) && !PP_ISNORELOC(pp)) { 1839e803c1e6SPrakash Sangappa panic("hat_devload: loading a mapping " 1840e803c1e6SPrakash Sangappa "to an unlocked page %p", 1841e803c1e6SPrakash Sangappa (void *)pp); 1842e803c1e6SPrakash Sangappa } 1843e803c1e6SPrakash Sangappa } 1844e803c1e6SPrakash Sangappa 18457c478bd9Sstevel@tonic-gate /* 18467c478bd9Sstevel@tonic-gate * load this page mapping 18477c478bd9Sstevel@tonic-gate */ 18487c478bd9Sstevel@tonic-gate ASSERT(!IN_VA_HOLE(va)); 1849ae115bc7Smrj while (hati_load_common(hat, va, pp, a, f, level, pfn) != 0) { 1850ae115bc7Smrj if (level == 0) 1851ae115bc7Smrj panic("unexpected hati_load_common() failure"); 1852ae115bc7Smrj --level; 1853ae115bc7Smrj pgsize = LEVEL_SIZE(level); 1854ae115bc7Smrj } 18557c478bd9Sstevel@tonic-gate 18567c478bd9Sstevel@tonic-gate /* 18577c478bd9Sstevel@tonic-gate * move to next page 18587c478bd9Sstevel@tonic-gate */ 18597c478bd9Sstevel@tonic-gate va += pgsize; 18607c478bd9Sstevel@tonic-gate pfn += mmu_btop(pgsize); 18617c478bd9Sstevel@tonic-gate } 1862843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 18637c478bd9Sstevel@tonic-gate } 18647c478bd9Sstevel@tonic-gate 18657c478bd9Sstevel@tonic-gate /* 18667c478bd9Sstevel@tonic-gate * void hat_unlock(hat, addr, len) 18677c478bd9Sstevel@tonic-gate * unlock the mappings to a given range of addresses 18687c478bd9Sstevel@tonic-gate * 18697c478bd9Sstevel@tonic-gate * Locks are tracked by ht_lock_cnt in the htable. 18707c478bd9Sstevel@tonic-gate */ 18717c478bd9Sstevel@tonic-gate void 18727c478bd9Sstevel@tonic-gate hat_unlock(hat_t *hat, caddr_t addr, size_t len) 18737c478bd9Sstevel@tonic-gate { 18747c478bd9Sstevel@tonic-gate uintptr_t vaddr = (uintptr_t)addr; 18757c478bd9Sstevel@tonic-gate uintptr_t eaddr = vaddr + len; 18767c478bd9Sstevel@tonic-gate htable_t *ht = NULL; 18777c478bd9Sstevel@tonic-gate 18787c478bd9Sstevel@tonic-gate /* 18797c478bd9Sstevel@tonic-gate * kernel entries are always locked, we don't track lock counts 18807c478bd9Sstevel@tonic-gate */ 1881ae115bc7Smrj ASSERT(hat == kas.a_hat || eaddr <= _userlimit); 18827c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(vaddr)); 18837c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(eaddr)); 18847c478bd9Sstevel@tonic-gate if (hat == kas.a_hat) 18857c478bd9Sstevel@tonic-gate return; 18867c478bd9Sstevel@tonic-gate if (eaddr > _userlimit) 18877c478bd9Sstevel@tonic-gate panic("hat_unlock() address out of range - above _userlimit"); 18887c478bd9Sstevel@tonic-gate 1889843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 1890dc32d872SJosef 'Jeff' Sipek ASSERT(AS_LOCK_HELD(hat->hat_as)); 18917c478bd9Sstevel@tonic-gate while (vaddr < eaddr) { 18927c478bd9Sstevel@tonic-gate (void) htable_walk(hat, &ht, &vaddr, eaddr); 18937c478bd9Sstevel@tonic-gate if (ht == NULL) 18947c478bd9Sstevel@tonic-gate break; 18957c478bd9Sstevel@tonic-gate 18967c478bd9Sstevel@tonic-gate ASSERT(!IN_VA_HOLE(vaddr)); 18977c478bd9Sstevel@tonic-gate 18987c478bd9Sstevel@tonic-gate if (ht->ht_lock_cnt < 1) 18997c478bd9Sstevel@tonic-gate panic("hat_unlock(): lock_cnt < 1, " 1900903a11ebSrh "htable=%p, vaddr=%p\n", (void *)ht, (void *)vaddr); 19017c478bd9Sstevel@tonic-gate HTABLE_LOCK_DEC(ht); 19027c478bd9Sstevel@tonic-gate 19037c478bd9Sstevel@tonic-gate vaddr += LEVEL_SIZE(ht->ht_level); 19047c478bd9Sstevel@tonic-gate } 19057c478bd9Sstevel@tonic-gate if (ht) 19067c478bd9Sstevel@tonic-gate htable_release(ht); 1907843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 19087c478bd9Sstevel@tonic-gate } 19097c478bd9Sstevel@tonic-gate 191005d3dc4bSpaulsan /* ARGSUSED */ 191105d3dc4bSpaulsan void 19127dacfc44Spaulsan hat_unlock_region(struct hat *hat, caddr_t addr, size_t len, 191305d3dc4bSpaulsan hat_region_cookie_t rcookie) 191405d3dc4bSpaulsan { 191505d3dc4bSpaulsan panic("No shared region support on x86"); 191605d3dc4bSpaulsan } 191705d3dc4bSpaulsan 1918*c2e5ad71SSam Gwydir /* 1919*c2e5ad71SSam Gwydir * A range of virtual pages for purposes of demapping. 1920*c2e5ad71SSam Gwydir */ 1921*c2e5ad71SSam Gwydir typedef struct range_info { 1922*c2e5ad71SSam Gwydir uintptr_t rng_va; /* address of page */ 1923*c2e5ad71SSam Gwydir ulong_t rng_cnt; /* number of pages in range */ 1924*c2e5ad71SSam Gwydir level_t rng_level; /* page table level */ 1925*c2e5ad71SSam Gwydir } range_info_t; 1926*c2e5ad71SSam Gwydir 1927843e1988Sjohnlev #if !defined(__xpv) 19287c478bd9Sstevel@tonic-gate /* 1929*c2e5ad71SSam Gwydir * Cross call service routine to demap a range of virtual 1930*c2e5ad71SSam Gwydir * pages on the current CPU or flush all mappings in TLB. 19317c478bd9Sstevel@tonic-gate */ 19327c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 19337c478bd9Sstevel@tonic-gate static int 19347c478bd9Sstevel@tonic-gate hati_demap_func(xc_arg_t a1, xc_arg_t a2, xc_arg_t a3) 19357c478bd9Sstevel@tonic-gate { 1936*c2e5ad71SSam Gwydir hat_t *hat = (hat_t *)a1; 1937*c2e5ad71SSam Gwydir range_info_t *range = (range_info_t *)a2; 1938*c2e5ad71SSam Gwydir size_t len = (size_t)a3; 1939*c2e5ad71SSam Gwydir caddr_t addr = (caddr_t)range->rng_va; 1940*c2e5ad71SSam Gwydir size_t pgsz = LEVEL_SIZE(range->rng_level); 19417c478bd9Sstevel@tonic-gate 19427c478bd9Sstevel@tonic-gate /* 19437c478bd9Sstevel@tonic-gate * If the target hat isn't the kernel and this CPU isn't operating 19447c478bd9Sstevel@tonic-gate * in the target hat, we can ignore the cross call. 19457c478bd9Sstevel@tonic-gate */ 19467c478bd9Sstevel@tonic-gate if (hat != kas.a_hat && hat != CPU->cpu_current_hat) 19477c478bd9Sstevel@tonic-gate return (0); 19487c478bd9Sstevel@tonic-gate 19497c478bd9Sstevel@tonic-gate /* 1950a6a74e0eSMatthew Ahrens * For a normal address, we flush a range of contiguous mappings 19517c478bd9Sstevel@tonic-gate */ 19527c478bd9Sstevel@tonic-gate if ((uintptr_t)addr != DEMAP_ALL_ADDR) { 1953*c2e5ad71SSam Gwydir for (size_t i = 0; i < len; i += pgsz) 1954a6a74e0eSMatthew Ahrens mmu_tlbflush_entry(addr + i); 19557c478bd9Sstevel@tonic-gate return (0); 19567c478bd9Sstevel@tonic-gate } 19577c478bd9Sstevel@tonic-gate 19587c478bd9Sstevel@tonic-gate /* 19597c478bd9Sstevel@tonic-gate * Otherwise we reload cr3 to effect a complete TLB flush. 19607c478bd9Sstevel@tonic-gate * 19617c478bd9Sstevel@tonic-gate * A reload of cr3 on a VLP process also means we must also recopy in 19627c478bd9Sstevel@tonic-gate * the pte values from the struct hat 19637c478bd9Sstevel@tonic-gate */ 19647c478bd9Sstevel@tonic-gate if (hat->hat_flags & HAT_VLP) { 19657c478bd9Sstevel@tonic-gate #if defined(__amd64) 19667c478bd9Sstevel@tonic-gate x86pte_t *vlpptep = CPU->cpu_hat_info->hci_vlp_l2ptes; 19677c478bd9Sstevel@tonic-gate 19687c478bd9Sstevel@tonic-gate VLP_COPY(hat->hat_vlp_ptes, vlpptep); 19697c478bd9Sstevel@tonic-gate #elif defined(__i386) 19707c478bd9Sstevel@tonic-gate reload_pae32(hat, CPU); 19717c478bd9Sstevel@tonic-gate #endif 19727c478bd9Sstevel@tonic-gate } 19737c478bd9Sstevel@tonic-gate reload_cr3(); 19747c478bd9Sstevel@tonic-gate return (0); 19757c478bd9Sstevel@tonic-gate } 19767c478bd9Sstevel@tonic-gate 197795c0a3c8Sjosephb /* 197895c0a3c8Sjosephb * Flush all TLB entries, including global (ie. kernel) ones. 197995c0a3c8Sjosephb */ 198095c0a3c8Sjosephb static void 198195c0a3c8Sjosephb flush_all_tlb_entries(void) 198295c0a3c8Sjosephb { 198395c0a3c8Sjosephb ulong_t cr4 = getcr4(); 198495c0a3c8Sjosephb 198595c0a3c8Sjosephb if (cr4 & CR4_PGE) { 198695c0a3c8Sjosephb setcr4(cr4 & ~(ulong_t)CR4_PGE); 198795c0a3c8Sjosephb setcr4(cr4); 198895c0a3c8Sjosephb 198995c0a3c8Sjosephb /* 199095c0a3c8Sjosephb * 32 bit PAE also needs to always reload_cr3() 199195c0a3c8Sjosephb */ 199295c0a3c8Sjosephb if (mmu.max_level == 2) 199395c0a3c8Sjosephb reload_cr3(); 199495c0a3c8Sjosephb } else { 199595c0a3c8Sjosephb reload_cr3(); 199695c0a3c8Sjosephb } 199795c0a3c8Sjosephb } 199895c0a3c8Sjosephb 199995c0a3c8Sjosephb #define TLB_CPU_HALTED (01ul) 200095c0a3c8Sjosephb #define TLB_INVAL_ALL (02ul) 200195c0a3c8Sjosephb #define CAS_TLB_INFO(cpu, old, new) \ 200275d94465SJosef 'Jeff' Sipek atomic_cas_ulong((ulong_t *)&(cpu)->cpu_m.mcpu_tlb_info, (old), (new)) 200395c0a3c8Sjosephb 200495c0a3c8Sjosephb /* 200595c0a3c8Sjosephb * Record that a CPU is going idle 200695c0a3c8Sjosephb */ 200795c0a3c8Sjosephb void 200895c0a3c8Sjosephb tlb_going_idle(void) 200995c0a3c8Sjosephb { 201075d94465SJosef 'Jeff' Sipek atomic_or_ulong((ulong_t *)&CPU->cpu_m.mcpu_tlb_info, TLB_CPU_HALTED); 201195c0a3c8Sjosephb } 201295c0a3c8Sjosephb 201395c0a3c8Sjosephb /* 201495c0a3c8Sjosephb * Service a delayed TLB flush if coming out of being idle. 201521584dbcSPavel Tatashin * It will be called from cpu idle notification with interrupt disabled. 201695c0a3c8Sjosephb */ 201795c0a3c8Sjosephb void 201895c0a3c8Sjosephb tlb_service(void) 201995c0a3c8Sjosephb { 202095c0a3c8Sjosephb ulong_t tlb_info; 202195c0a3c8Sjosephb ulong_t found; 202295c0a3c8Sjosephb 202395c0a3c8Sjosephb /* 202495c0a3c8Sjosephb * We only have to do something if coming out of being idle. 202595c0a3c8Sjosephb */ 202695c0a3c8Sjosephb tlb_info = CPU->cpu_m.mcpu_tlb_info; 202795c0a3c8Sjosephb if (tlb_info & TLB_CPU_HALTED) { 202895c0a3c8Sjosephb ASSERT(CPU->cpu_current_hat == kas.a_hat); 202995c0a3c8Sjosephb 203095c0a3c8Sjosephb /* 203195c0a3c8Sjosephb * Atomic clear and fetch of old state. 203295c0a3c8Sjosephb */ 203395c0a3c8Sjosephb while ((found = CAS_TLB_INFO(CPU, tlb_info, 0)) != tlb_info) { 203495c0a3c8Sjosephb ASSERT(found & TLB_CPU_HALTED); 203595c0a3c8Sjosephb tlb_info = found; 203695c0a3c8Sjosephb SMT_PAUSE(); 203795c0a3c8Sjosephb } 203895c0a3c8Sjosephb if (tlb_info & TLB_INVAL_ALL) 203995c0a3c8Sjosephb flush_all_tlb_entries(); 204095c0a3c8Sjosephb } 204195c0a3c8Sjosephb } 2042843e1988Sjohnlev #endif /* !__xpv */ 204395c0a3c8Sjosephb 20447c478bd9Sstevel@tonic-gate /* 20457c478bd9Sstevel@tonic-gate * Internal routine to do cross calls to invalidate a range of pages on 20467c478bd9Sstevel@tonic-gate * all CPUs using a given hat. 20477c478bd9Sstevel@tonic-gate */ 20487c478bd9Sstevel@tonic-gate void 2049*c2e5ad71SSam Gwydir hat_tlb_inval_range(hat_t *hat, range_info_t *range) 20507c478bd9Sstevel@tonic-gate { 20517c478bd9Sstevel@tonic-gate extern int flushes_require_xcalls; /* from mp_startup.c */ 20527c478bd9Sstevel@tonic-gate cpuset_t justme; 2053ae115bc7Smrj cpuset_t cpus_to_shootdown; 2054*c2e5ad71SSam Gwydir uintptr_t va = range->rng_va; 2055*c2e5ad71SSam Gwydir size_t len = range->rng_cnt << LEVEL_SHIFT(range->rng_level); 2056843e1988Sjohnlev #ifndef __xpv 2057843e1988Sjohnlev cpuset_t check_cpus; 205895c0a3c8Sjosephb cpu_t *cpup; 205995c0a3c8Sjosephb int c; 2060843e1988Sjohnlev #endif 20617c478bd9Sstevel@tonic-gate 20627c478bd9Sstevel@tonic-gate /* 20637c478bd9Sstevel@tonic-gate * If the hat is being destroyed, there are no more users, so 20647c478bd9Sstevel@tonic-gate * demap need not do anything. 20657c478bd9Sstevel@tonic-gate */ 20667c478bd9Sstevel@tonic-gate if (hat->hat_flags & HAT_FREEING) 20677c478bd9Sstevel@tonic-gate return; 20687c478bd9Sstevel@tonic-gate 20697c478bd9Sstevel@tonic-gate /* 20707c478bd9Sstevel@tonic-gate * If demapping from a shared pagetable, we best demap the 20717c478bd9Sstevel@tonic-gate * entire set of user TLBs, since we don't know what addresses 20727c478bd9Sstevel@tonic-gate * these were shared at. 20737c478bd9Sstevel@tonic-gate */ 20747c478bd9Sstevel@tonic-gate if (hat->hat_flags & HAT_SHARED) { 20757c478bd9Sstevel@tonic-gate hat = kas.a_hat; 20767c478bd9Sstevel@tonic-gate va = DEMAP_ALL_ADDR; 20777c478bd9Sstevel@tonic-gate } 20787c478bd9Sstevel@tonic-gate 20797c478bd9Sstevel@tonic-gate /* 20807c478bd9Sstevel@tonic-gate * if not running with multiple CPUs, don't use cross calls 20817c478bd9Sstevel@tonic-gate */ 20827c478bd9Sstevel@tonic-gate if (panicstr || !flushes_require_xcalls) { 2083843e1988Sjohnlev #ifdef __xpv 2084a6a74e0eSMatthew Ahrens if (va == DEMAP_ALL_ADDR) { 2085843e1988Sjohnlev xen_flush_tlb(); 2086a6a74e0eSMatthew Ahrens } else { 2087a6a74e0eSMatthew Ahrens for (size_t i = 0; i < len; i += MMU_PAGESIZE) 2088a6a74e0eSMatthew Ahrens xen_flush_va((caddr_t)(va + i)); 2089a6a74e0eSMatthew Ahrens } 2090843e1988Sjohnlev #else 2091a6a74e0eSMatthew Ahrens (void) hati_demap_func((xc_arg_t)hat, 2092*c2e5ad71SSam Gwydir (xc_arg_t)range, (xc_arg_t)len); 2093843e1988Sjohnlev #endif 20947c478bd9Sstevel@tonic-gate return; 20957c478bd9Sstevel@tonic-gate } 20967c478bd9Sstevel@tonic-gate 20977c478bd9Sstevel@tonic-gate 20987c478bd9Sstevel@tonic-gate /* 2099ae115bc7Smrj * Determine CPUs to shootdown. Kernel changes always do all CPUs. 2100ae115bc7Smrj * Otherwise it's just CPUs currently executing in this hat. 21017c478bd9Sstevel@tonic-gate */ 21027c478bd9Sstevel@tonic-gate kpreempt_disable(); 21037c478bd9Sstevel@tonic-gate CPUSET_ONLY(justme, CPU->cpu_id); 2104ae115bc7Smrj if (hat == kas.a_hat) 2105ae115bc7Smrj cpus_to_shootdown = khat_cpuset; 21067c478bd9Sstevel@tonic-gate else 2107ae115bc7Smrj cpus_to_shootdown = hat->hat_cpus; 2108ae115bc7Smrj 2109843e1988Sjohnlev #ifndef __xpv 211095c0a3c8Sjosephb /* 211195c0a3c8Sjosephb * If any CPUs in the set are idle, just request a delayed flush 211295c0a3c8Sjosephb * and avoid waking them up. 211395c0a3c8Sjosephb */ 211495c0a3c8Sjosephb check_cpus = cpus_to_shootdown; 211595c0a3c8Sjosephb for (c = 0; c < NCPU && !CPUSET_ISNULL(check_cpus); ++c) { 211695c0a3c8Sjosephb ulong_t tlb_info; 211795c0a3c8Sjosephb 211895c0a3c8Sjosephb if (!CPU_IN_SET(check_cpus, c)) 211995c0a3c8Sjosephb continue; 212095c0a3c8Sjosephb CPUSET_DEL(check_cpus, c); 212195c0a3c8Sjosephb cpup = cpu[c]; 212295c0a3c8Sjosephb if (cpup == NULL) 212395c0a3c8Sjosephb continue; 212495c0a3c8Sjosephb 212595c0a3c8Sjosephb tlb_info = cpup->cpu_m.mcpu_tlb_info; 212695c0a3c8Sjosephb while (tlb_info == TLB_CPU_HALTED) { 212795c0a3c8Sjosephb (void) CAS_TLB_INFO(cpup, TLB_CPU_HALTED, 2128250b7ff9Sjosephb TLB_CPU_HALTED | TLB_INVAL_ALL); 212995c0a3c8Sjosephb SMT_PAUSE(); 213095c0a3c8Sjosephb tlb_info = cpup->cpu_m.mcpu_tlb_info; 213195c0a3c8Sjosephb } 213295c0a3c8Sjosephb if (tlb_info == (TLB_CPU_HALTED | TLB_INVAL_ALL)) { 213395c0a3c8Sjosephb HATSTAT_INC(hs_tlb_inval_delayed); 213495c0a3c8Sjosephb CPUSET_DEL(cpus_to_shootdown, c); 213595c0a3c8Sjosephb } 213695c0a3c8Sjosephb } 2137843e1988Sjohnlev #endif 213895c0a3c8Sjosephb 2139ae115bc7Smrj if (CPUSET_ISNULL(cpus_to_shootdown) || 2140ae115bc7Smrj CPUSET_ISEQUAL(cpus_to_shootdown, justme)) { 2141ae115bc7Smrj 2142843e1988Sjohnlev #ifdef __xpv 2143a6a74e0eSMatthew Ahrens if (va == DEMAP_ALL_ADDR) { 2144843e1988Sjohnlev xen_flush_tlb(); 2145a6a74e0eSMatthew Ahrens } else { 2146a6a74e0eSMatthew Ahrens for (size_t i = 0; i < len; i += MMU_PAGESIZE) 2147a6a74e0eSMatthew Ahrens xen_flush_va((caddr_t)(va + i)); 2148a6a74e0eSMatthew Ahrens } 2149843e1988Sjohnlev #else 2150a6a74e0eSMatthew Ahrens (void) hati_demap_func((xc_arg_t)hat, 2151*c2e5ad71SSam Gwydir (xc_arg_t)range, (xc_arg_t)len); 2152843e1988Sjohnlev #endif 2153ae115bc7Smrj 2154ae115bc7Smrj } else { 2155ae115bc7Smrj 2156ae115bc7Smrj CPUSET_ADD(cpus_to_shootdown, CPU->cpu_id); 2157843e1988Sjohnlev #ifdef __xpv 2158a6a74e0eSMatthew Ahrens if (va == DEMAP_ALL_ADDR) { 2159843e1988Sjohnlev xen_gflush_tlb(cpus_to_shootdown); 2160a6a74e0eSMatthew Ahrens } else { 2161a6a74e0eSMatthew Ahrens for (size_t i = 0; i < len; i += MMU_PAGESIZE) { 2162a6a74e0eSMatthew Ahrens xen_gflush_va((caddr_t)(va + i), 2163a6a74e0eSMatthew Ahrens cpus_to_shootdown); 2164a6a74e0eSMatthew Ahrens } 2165a6a74e0eSMatthew Ahrens } 2166843e1988Sjohnlev #else 2167*c2e5ad71SSam Gwydir xc_call((xc_arg_t)hat, (xc_arg_t)range, (xc_arg_t)len, 2168f34a7178SJoe Bonasera CPUSET2BV(cpus_to_shootdown), hati_demap_func); 2169843e1988Sjohnlev #endif 2170ae115bc7Smrj 2171ae115bc7Smrj } 21727c478bd9Sstevel@tonic-gate kpreempt_enable(); 21737c478bd9Sstevel@tonic-gate } 21747c478bd9Sstevel@tonic-gate 2175a6a74e0eSMatthew Ahrens void 2176a6a74e0eSMatthew Ahrens hat_tlb_inval(hat_t *hat, uintptr_t va) 2177a6a74e0eSMatthew Ahrens { 2178*c2e5ad71SSam Gwydir /* 2179*c2e5ad71SSam Gwydir * Create range for a single page. 2180*c2e5ad71SSam Gwydir */ 2181*c2e5ad71SSam Gwydir range_info_t range; 2182*c2e5ad71SSam Gwydir range.rng_va = va; 2183*c2e5ad71SSam Gwydir range.rng_cnt = 1; /* one page */ 2184*c2e5ad71SSam Gwydir range.rng_level = MIN_PAGE_LEVEL; /* pages are MMU_PAGESIZE */ 2185*c2e5ad71SSam Gwydir 2186*c2e5ad71SSam Gwydir hat_tlb_inval_range(hat, &range); 2187a6a74e0eSMatthew Ahrens } 2188a6a74e0eSMatthew Ahrens 21897c478bd9Sstevel@tonic-gate /* 21907c478bd9Sstevel@tonic-gate * Interior routine for HAT_UNLOADs from hat_unload_callback(), 21917c478bd9Sstevel@tonic-gate * hat_kmap_unload() OR from hat_steal() code. This routine doesn't 21927c478bd9Sstevel@tonic-gate * handle releasing of the htables. 21937c478bd9Sstevel@tonic-gate */ 21947c478bd9Sstevel@tonic-gate void 21957c478bd9Sstevel@tonic-gate hat_pte_unmap( 21967c478bd9Sstevel@tonic-gate htable_t *ht, 21977c478bd9Sstevel@tonic-gate uint_t entry, 21987c478bd9Sstevel@tonic-gate uint_t flags, 21997c478bd9Sstevel@tonic-gate x86pte_t old_pte, 2200a6a74e0eSMatthew Ahrens void *pte_ptr, 2201a6a74e0eSMatthew Ahrens boolean_t tlb) 22027c478bd9Sstevel@tonic-gate { 22037c478bd9Sstevel@tonic-gate hat_t *hat = ht->ht_hat; 22047c478bd9Sstevel@tonic-gate hment_t *hm = NULL; 22057c478bd9Sstevel@tonic-gate page_t *pp = NULL; 22067c478bd9Sstevel@tonic-gate level_t l = ht->ht_level; 22077c478bd9Sstevel@tonic-gate pfn_t pfn; 22087c478bd9Sstevel@tonic-gate 22097c478bd9Sstevel@tonic-gate /* 22107c478bd9Sstevel@tonic-gate * We always track the locking counts, even if nothing is unmapped 22117c478bd9Sstevel@tonic-gate */ 22127c478bd9Sstevel@tonic-gate if ((flags & HAT_UNLOAD_UNLOCK) != 0 && hat != kas.a_hat) { 22137c478bd9Sstevel@tonic-gate ASSERT(ht->ht_lock_cnt > 0); 22147c478bd9Sstevel@tonic-gate HTABLE_LOCK_DEC(ht); 22157c478bd9Sstevel@tonic-gate } 22167c478bd9Sstevel@tonic-gate 22177c478bd9Sstevel@tonic-gate /* 22187c478bd9Sstevel@tonic-gate * Figure out which page's mapping list lock to acquire using the PFN 22197c478bd9Sstevel@tonic-gate * passed in "old" PTE. We then attempt to invalidate the PTE. 22207c478bd9Sstevel@tonic-gate * If another thread, probably a hat_pageunload, has asynchronously 22217c478bd9Sstevel@tonic-gate * unmapped/remapped this address we'll loop here. 22227c478bd9Sstevel@tonic-gate */ 22237c478bd9Sstevel@tonic-gate ASSERT(ht->ht_busy > 0); 22247c478bd9Sstevel@tonic-gate while (PTE_ISVALID(old_pte)) { 22257c478bd9Sstevel@tonic-gate pfn = PTE2PFN(old_pte, l); 2226ae115bc7Smrj if (PTE_GET(old_pte, PT_SOFTWARE) >= PT_NOCONSIST) { 22277c478bd9Sstevel@tonic-gate pp = NULL; 22287c478bd9Sstevel@tonic-gate } else { 2229843e1988Sjohnlev #ifdef __xpv 2230843e1988Sjohnlev if (pfn == PFN_INVALID) 2231843e1988Sjohnlev panic("Invalid PFN, but not PT_NOCONSIST"); 2232843e1988Sjohnlev #endif 22337c478bd9Sstevel@tonic-gate pp = page_numtopp_nolock(pfn); 2234aa2ed9e5Sjosephb if (pp == NULL) { 2235aa2ed9e5Sjosephb panic("no page_t, not NOCONSIST: old_pte=" 2236aa2ed9e5Sjosephb FMT_PTE " ht=%lx entry=0x%x pte_ptr=%lx", 2237aa2ed9e5Sjosephb old_pte, (uintptr_t)ht, entry, 2238aa2ed9e5Sjosephb (uintptr_t)pte_ptr); 2239aa2ed9e5Sjosephb } 22407c478bd9Sstevel@tonic-gate x86_hm_enter(pp); 22417c478bd9Sstevel@tonic-gate } 2242aa2ed9e5Sjosephb 2243a6a74e0eSMatthew Ahrens old_pte = x86pte_inval(ht, entry, old_pte, pte_ptr, tlb); 22447c478bd9Sstevel@tonic-gate 22457c478bd9Sstevel@tonic-gate /* 22467c478bd9Sstevel@tonic-gate * If the page hadn't changed we've unmapped it and can proceed 22477c478bd9Sstevel@tonic-gate */ 22487c478bd9Sstevel@tonic-gate if (PTE_ISVALID(old_pte) && PTE2PFN(old_pte, l) == pfn) 22497c478bd9Sstevel@tonic-gate break; 22507c478bd9Sstevel@tonic-gate 22517c478bd9Sstevel@tonic-gate /* 22527c478bd9Sstevel@tonic-gate * Otherwise, we'll have to retry with the current old_pte. 22537c478bd9Sstevel@tonic-gate * Drop the hment lock, since the pfn may have changed. 22547c478bd9Sstevel@tonic-gate */ 22557c478bd9Sstevel@tonic-gate if (pp != NULL) { 22567c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 22577c478bd9Sstevel@tonic-gate pp = NULL; 22587c478bd9Sstevel@tonic-gate } else { 2259ae115bc7Smrj ASSERT(PTE_GET(old_pte, PT_SOFTWARE) >= PT_NOCONSIST); 22607c478bd9Sstevel@tonic-gate } 22617c478bd9Sstevel@tonic-gate } 22627c478bd9Sstevel@tonic-gate 22637c478bd9Sstevel@tonic-gate /* 22647c478bd9Sstevel@tonic-gate * If the old mapping wasn't valid, there's nothing more to do 22657c478bd9Sstevel@tonic-gate */ 22667c478bd9Sstevel@tonic-gate if (!PTE_ISVALID(old_pte)) { 22677c478bd9Sstevel@tonic-gate if (pp != NULL) 22687c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 22697c478bd9Sstevel@tonic-gate return; 22707c478bd9Sstevel@tonic-gate } 22717c478bd9Sstevel@tonic-gate 22727c478bd9Sstevel@tonic-gate /* 22737c478bd9Sstevel@tonic-gate * Take care of syncing any MOD/REF bits and removing the hment. 22747c478bd9Sstevel@tonic-gate */ 22757c478bd9Sstevel@tonic-gate if (pp != NULL) { 22767c478bd9Sstevel@tonic-gate if (!(flags & HAT_UNLOAD_NOSYNC)) 22777c478bd9Sstevel@tonic-gate hati_sync_pte_to_page(pp, old_pte, l); 22787c478bd9Sstevel@tonic-gate hm = hment_remove(pp, ht, entry); 22797c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 22807c478bd9Sstevel@tonic-gate if (hm != NULL) 22817c478bd9Sstevel@tonic-gate hment_free(hm); 22827c478bd9Sstevel@tonic-gate } 22837c478bd9Sstevel@tonic-gate 22847c478bd9Sstevel@tonic-gate /* 22857c478bd9Sstevel@tonic-gate * Handle book keeping in the htable and hat 22867c478bd9Sstevel@tonic-gate */ 22877c478bd9Sstevel@tonic-gate ASSERT(ht->ht_valid_cnt > 0); 22887c478bd9Sstevel@tonic-gate HTABLE_DEC(ht->ht_valid_cnt); 22897c478bd9Sstevel@tonic-gate PGCNT_DEC(hat, l); 22907c478bd9Sstevel@tonic-gate } 22917c478bd9Sstevel@tonic-gate 22927c478bd9Sstevel@tonic-gate /* 22937c478bd9Sstevel@tonic-gate * very cheap unload implementation to special case some kernel addresses 22947c478bd9Sstevel@tonic-gate */ 22957c478bd9Sstevel@tonic-gate static void 22967c478bd9Sstevel@tonic-gate hat_kmap_unload(caddr_t addr, size_t len, uint_t flags) 22977c478bd9Sstevel@tonic-gate { 22987c478bd9Sstevel@tonic-gate uintptr_t va = (uintptr_t)addr; 22997c478bd9Sstevel@tonic-gate uintptr_t eva = va + len; 2300ae115bc7Smrj pgcnt_t pg_index; 23017c478bd9Sstevel@tonic-gate htable_t *ht; 23027c478bd9Sstevel@tonic-gate uint_t entry; 2303ae115bc7Smrj x86pte_t *pte_ptr; 23047c478bd9Sstevel@tonic-gate x86pte_t old_pte; 23057c478bd9Sstevel@tonic-gate 23067c478bd9Sstevel@tonic-gate for (; va < eva; va += MMU_PAGESIZE) { 23077c478bd9Sstevel@tonic-gate /* 23087c478bd9Sstevel@tonic-gate * Get the PTE 23097c478bd9Sstevel@tonic-gate */ 2310ae115bc7Smrj pg_index = mmu_btop(va - mmu.kmap_addr); 2311ae115bc7Smrj pte_ptr = PT_INDEX_PTR(mmu.kmap_ptes, pg_index); 2312ae115bc7Smrj old_pte = GET_PTE(pte_ptr); 23137c478bd9Sstevel@tonic-gate 23147c478bd9Sstevel@tonic-gate /* 23157c478bd9Sstevel@tonic-gate * get the htable / entry 23167c478bd9Sstevel@tonic-gate */ 23177c478bd9Sstevel@tonic-gate ht = mmu.kmap_htables[(va - mmu.kmap_htables[0]->ht_vaddr) 23187c478bd9Sstevel@tonic-gate >> LEVEL_SHIFT(1)]; 23197c478bd9Sstevel@tonic-gate entry = htable_va2entry(va, ht); 23207c478bd9Sstevel@tonic-gate 23217c478bd9Sstevel@tonic-gate /* 23227c478bd9Sstevel@tonic-gate * use mostly common code to unmap it. 23237c478bd9Sstevel@tonic-gate */ 2324a6a74e0eSMatthew Ahrens hat_pte_unmap(ht, entry, flags, old_pte, pte_ptr, B_TRUE); 23257c478bd9Sstevel@tonic-gate } 23267c478bd9Sstevel@tonic-gate } 23277c478bd9Sstevel@tonic-gate 23287c478bd9Sstevel@tonic-gate 23297c478bd9Sstevel@tonic-gate /* 23307c478bd9Sstevel@tonic-gate * unload a range of virtual address space (no callback) 23317c478bd9Sstevel@tonic-gate */ 23327c478bd9Sstevel@tonic-gate void 23337c478bd9Sstevel@tonic-gate hat_unload(hat_t *hat, caddr_t addr, size_t len, uint_t flags) 23347c478bd9Sstevel@tonic-gate { 23357c478bd9Sstevel@tonic-gate uintptr_t va = (uintptr_t)addr; 2336ae115bc7Smrj 2337843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 2338ae115bc7Smrj ASSERT(hat == kas.a_hat || va + len <= _userlimit); 23397c478bd9Sstevel@tonic-gate 23407c478bd9Sstevel@tonic-gate /* 23417c478bd9Sstevel@tonic-gate * special case for performance. 23427c478bd9Sstevel@tonic-gate */ 23437c478bd9Sstevel@tonic-gate if (mmu.kmap_addr <= va && va < mmu.kmap_eaddr) { 23447c478bd9Sstevel@tonic-gate ASSERT(hat == kas.a_hat); 23457c478bd9Sstevel@tonic-gate hat_kmap_unload(addr, len, flags); 2346ae115bc7Smrj } else { 2347ae115bc7Smrj hat_unload_callback(hat, addr, len, flags, NULL); 23487c478bd9Sstevel@tonic-gate } 2349843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 23507c478bd9Sstevel@tonic-gate } 23517c478bd9Sstevel@tonic-gate 2352a6a74e0eSMatthew Ahrens /* 2353a6a74e0eSMatthew Ahrens * Invalidate the TLB, and perform the callback to the upper level VM system, 2354a6a74e0eSMatthew Ahrens * for the specified ranges of contiguous pages. 2355a6a74e0eSMatthew Ahrens */ 23567c478bd9Sstevel@tonic-gate static void 2357a6a74e0eSMatthew Ahrens handle_ranges(hat_t *hat, hat_callback_t *cb, uint_t cnt, range_info_t *range) 23587c478bd9Sstevel@tonic-gate { 2359a6a74e0eSMatthew Ahrens while (cnt > 0) { 23607c478bd9Sstevel@tonic-gate --cnt; 2361*c2e5ad71SSam Gwydir hat_tlb_inval_range(hat, &range[cnt]); 2362a6a74e0eSMatthew Ahrens 2363a6a74e0eSMatthew Ahrens if (cb != NULL) { 2364a6a74e0eSMatthew Ahrens cb->hcb_start_addr = (caddr_t)range[cnt].rng_va; 2365a6a74e0eSMatthew Ahrens cb->hcb_end_addr = cb->hcb_start_addr; 2366*c2e5ad71SSam Gwydir cb->hcb_end_addr += range[cnt].rng_cnt << 2367*c2e5ad71SSam Gwydir LEVEL_SHIFT(range[cnt].rng_level); 2368a6a74e0eSMatthew Ahrens cb->hcb_function(cb); 2369a6a74e0eSMatthew Ahrens } 23707c478bd9Sstevel@tonic-gate } 23717c478bd9Sstevel@tonic-gate } 23727c478bd9Sstevel@tonic-gate 23737c478bd9Sstevel@tonic-gate /* 23747c478bd9Sstevel@tonic-gate * Unload a given range of addresses (has optional callback) 23757c478bd9Sstevel@tonic-gate * 23767c478bd9Sstevel@tonic-gate * Flags: 23777c478bd9Sstevel@tonic-gate * define HAT_UNLOAD 0x00 23787c478bd9Sstevel@tonic-gate * define HAT_UNLOAD_NOSYNC 0x02 23797c478bd9Sstevel@tonic-gate * define HAT_UNLOAD_UNLOCK 0x04 23807c478bd9Sstevel@tonic-gate * define HAT_UNLOAD_OTHER 0x08 - not used 23817c478bd9Sstevel@tonic-gate * define HAT_UNLOAD_UNMAP 0x10 - same as HAT_UNLOAD 23827c478bd9Sstevel@tonic-gate */ 23837c478bd9Sstevel@tonic-gate #define MAX_UNLOAD_CNT (8) 23847c478bd9Sstevel@tonic-gate void 23857c478bd9Sstevel@tonic-gate hat_unload_callback( 23867c478bd9Sstevel@tonic-gate hat_t *hat, 23877c478bd9Sstevel@tonic-gate caddr_t addr, 23887c478bd9Sstevel@tonic-gate size_t len, 23897c478bd9Sstevel@tonic-gate uint_t flags, 23907c478bd9Sstevel@tonic-gate hat_callback_t *cb) 23917c478bd9Sstevel@tonic-gate { 23927c478bd9Sstevel@tonic-gate uintptr_t vaddr = (uintptr_t)addr; 23937c478bd9Sstevel@tonic-gate uintptr_t eaddr = vaddr + len; 23947c478bd9Sstevel@tonic-gate htable_t *ht = NULL; 23957c478bd9Sstevel@tonic-gate uint_t entry; 2396aa2ed9e5Sjosephb uintptr_t contig_va = (uintptr_t)-1L; 23977c478bd9Sstevel@tonic-gate range_info_t r[MAX_UNLOAD_CNT]; 23987c478bd9Sstevel@tonic-gate uint_t r_cnt = 0; 23997c478bd9Sstevel@tonic-gate x86pte_t old_pte; 24007c478bd9Sstevel@tonic-gate 2401843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 2402ae115bc7Smrj ASSERT(hat == kas.a_hat || eaddr <= _userlimit); 24037c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(vaddr)); 24047c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(eaddr)); 24057c478bd9Sstevel@tonic-gate 2406ae115bc7Smrj /* 2407ae115bc7Smrj * Special case a single page being unloaded for speed. This happens 2408ae115bc7Smrj * quite frequently, COW faults after a fork() for example. 2409ae115bc7Smrj */ 2410ae115bc7Smrj if (cb == NULL && len == MMU_PAGESIZE) { 2411ae115bc7Smrj ht = htable_getpte(hat, vaddr, &entry, &old_pte, 0); 2412ae115bc7Smrj if (ht != NULL) { 2413a6a74e0eSMatthew Ahrens if (PTE_ISVALID(old_pte)) { 2414a6a74e0eSMatthew Ahrens hat_pte_unmap(ht, entry, flags, old_pte, 2415a6a74e0eSMatthew Ahrens NULL, B_TRUE); 2416a6a74e0eSMatthew Ahrens } 2417ae115bc7Smrj htable_release(ht); 2418ae115bc7Smrj } 2419843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 2420ae115bc7Smrj return; 2421ae115bc7Smrj } 2422ae115bc7Smrj 24237c478bd9Sstevel@tonic-gate while (vaddr < eaddr) { 24247c478bd9Sstevel@tonic-gate old_pte = htable_walk(hat, &ht, &vaddr, eaddr); 24257c478bd9Sstevel@tonic-gate if (ht == NULL) 24267c478bd9Sstevel@tonic-gate break; 24277c478bd9Sstevel@tonic-gate 24287c478bd9Sstevel@tonic-gate ASSERT(!IN_VA_HOLE(vaddr)); 24297c478bd9Sstevel@tonic-gate 24307c478bd9Sstevel@tonic-gate if (vaddr < (uintptr_t)addr) 24317c478bd9Sstevel@tonic-gate panic("hat_unload_callback(): unmap inside large page"); 24327c478bd9Sstevel@tonic-gate 24337c478bd9Sstevel@tonic-gate /* 24347c478bd9Sstevel@tonic-gate * We'll do the call backs for contiguous ranges 24357c478bd9Sstevel@tonic-gate */ 2436aa2ed9e5Sjosephb if (vaddr != contig_va || 24377c478bd9Sstevel@tonic-gate (r_cnt > 0 && r[r_cnt - 1].rng_level != ht->ht_level)) { 24387c478bd9Sstevel@tonic-gate if (r_cnt == MAX_UNLOAD_CNT) { 2439a6a74e0eSMatthew Ahrens handle_ranges(hat, cb, r_cnt, r); 24407c478bd9Sstevel@tonic-gate r_cnt = 0; 24417c478bd9Sstevel@tonic-gate } 24427c478bd9Sstevel@tonic-gate r[r_cnt].rng_va = vaddr; 24437c478bd9Sstevel@tonic-gate r[r_cnt].rng_cnt = 0; 24447c478bd9Sstevel@tonic-gate r[r_cnt].rng_level = ht->ht_level; 24457c478bd9Sstevel@tonic-gate ++r_cnt; 24467c478bd9Sstevel@tonic-gate } 24477c478bd9Sstevel@tonic-gate 24487c478bd9Sstevel@tonic-gate /* 2449a6a74e0eSMatthew Ahrens * Unload one mapping (for a single page) from the page tables. 2450a6a74e0eSMatthew Ahrens * Note that we do not remove the mapping from the TLB yet, 2451a6a74e0eSMatthew Ahrens * as indicated by the tlb=FALSE argument to hat_pte_unmap(). 2452a6a74e0eSMatthew Ahrens * handle_ranges() will clear the TLB entries with one call to 2453a6a74e0eSMatthew Ahrens * hat_tlb_inval_range() per contiguous range. This is 2454a6a74e0eSMatthew Ahrens * safe because the page can not be reused until the 2455a6a74e0eSMatthew Ahrens * callback is made (or we return). 24567c478bd9Sstevel@tonic-gate */ 24577c478bd9Sstevel@tonic-gate entry = htable_va2entry(vaddr, ht); 2458a6a74e0eSMatthew Ahrens hat_pte_unmap(ht, entry, flags, old_pte, NULL, B_FALSE); 24597c478bd9Sstevel@tonic-gate ASSERT(ht->ht_level <= mmu.max_page_level); 24607c478bd9Sstevel@tonic-gate vaddr += LEVEL_SIZE(ht->ht_level); 2461aa2ed9e5Sjosephb contig_va = vaddr; 24627c478bd9Sstevel@tonic-gate ++r[r_cnt - 1].rng_cnt; 24637c478bd9Sstevel@tonic-gate } 24647c478bd9Sstevel@tonic-gate if (ht) 24657c478bd9Sstevel@tonic-gate htable_release(ht); 24667c478bd9Sstevel@tonic-gate 24677c478bd9Sstevel@tonic-gate /* 24687c478bd9Sstevel@tonic-gate * handle last range for callbacks 24697c478bd9Sstevel@tonic-gate */ 24707c478bd9Sstevel@tonic-gate if (r_cnt > 0) 2471a6a74e0eSMatthew Ahrens handle_ranges(hat, cb, r_cnt, r); 2472843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 24737c478bd9Sstevel@tonic-gate } 24747c478bd9Sstevel@tonic-gate 2475ca3e8d88SDave Plauger /* 24769e0b7c70SDave Plauger * Invalidate a virtual address translation on a slave CPU during 24779e0b7c70SDave Plauger * panic() dumps. 2478ca3e8d88SDave Plauger */ 2479ca3e8d88SDave Plauger void 2480ca3e8d88SDave Plauger hat_flush_range(hat_t *hat, caddr_t va, size_t size) 2481ca3e8d88SDave Plauger { 2482ca3e8d88SDave Plauger ssize_t sz; 2483ca3e8d88SDave Plauger caddr_t endva = va + size; 2484ca3e8d88SDave Plauger 2485ca3e8d88SDave Plauger while (va < endva) { 2486ca3e8d88SDave Plauger sz = hat_getpagesize(hat, va); 24879e0b7c70SDave Plauger if (sz < 0) { 2488ca3e8d88SDave Plauger #ifdef __xpv 2489ca3e8d88SDave Plauger xen_flush_tlb(); 2490ca3e8d88SDave Plauger #else 24919e0b7c70SDave Plauger flush_all_tlb_entries(); 2492ca3e8d88SDave Plauger #endif 2493ca3e8d88SDave Plauger break; 24949e0b7c70SDave Plauger } 24959e0b7c70SDave Plauger #ifdef __xpv 24969e0b7c70SDave Plauger xen_flush_va(va); 24979e0b7c70SDave Plauger #else 24989e0b7c70SDave Plauger mmu_tlbflush_entry(va); 24999e0b7c70SDave Plauger #endif 2500ca3e8d88SDave Plauger va += sz; 2501ca3e8d88SDave Plauger } 2502ca3e8d88SDave Plauger } 2503ca3e8d88SDave Plauger 25047c478bd9Sstevel@tonic-gate /* 25057c478bd9Sstevel@tonic-gate * synchronize mapping with software data structures 25067c478bd9Sstevel@tonic-gate * 25077c478bd9Sstevel@tonic-gate * This interface is currently only used by the working set monitor 25087c478bd9Sstevel@tonic-gate * driver. 25097c478bd9Sstevel@tonic-gate */ 25107c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 25117c478bd9Sstevel@tonic-gate void 25127c478bd9Sstevel@tonic-gate hat_sync(hat_t *hat, caddr_t addr, size_t len, uint_t flags) 25137c478bd9Sstevel@tonic-gate { 25147c478bd9Sstevel@tonic-gate uintptr_t vaddr = (uintptr_t)addr; 25157c478bd9Sstevel@tonic-gate uintptr_t eaddr = vaddr + len; 25167c478bd9Sstevel@tonic-gate htable_t *ht = NULL; 25177c478bd9Sstevel@tonic-gate uint_t entry; 25187c478bd9Sstevel@tonic-gate x86pte_t pte; 25197c478bd9Sstevel@tonic-gate x86pte_t save_pte; 25207c478bd9Sstevel@tonic-gate x86pte_t new; 25217c478bd9Sstevel@tonic-gate page_t *pp; 25227c478bd9Sstevel@tonic-gate 25237c478bd9Sstevel@tonic-gate ASSERT(!IN_VA_HOLE(vaddr)); 25247c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(vaddr)); 25257c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(eaddr)); 2526ae115bc7Smrj ASSERT(hat == kas.a_hat || eaddr <= _userlimit); 25277c478bd9Sstevel@tonic-gate 2528843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 25297c478bd9Sstevel@tonic-gate for (; vaddr < eaddr; vaddr += LEVEL_SIZE(ht->ht_level)) { 25307c478bd9Sstevel@tonic-gate try_again: 25317c478bd9Sstevel@tonic-gate pte = htable_walk(hat, &ht, &vaddr, eaddr); 25327c478bd9Sstevel@tonic-gate if (ht == NULL) 25337c478bd9Sstevel@tonic-gate break; 25347c478bd9Sstevel@tonic-gate entry = htable_va2entry(vaddr, ht); 25357c478bd9Sstevel@tonic-gate 2536ae115bc7Smrj if (PTE_GET(pte, PT_SOFTWARE) >= PT_NOSYNC || 25377c478bd9Sstevel@tonic-gate PTE_GET(pte, PT_REF | PT_MOD) == 0) 25387c478bd9Sstevel@tonic-gate continue; 25397c478bd9Sstevel@tonic-gate 25407c478bd9Sstevel@tonic-gate /* 25417c478bd9Sstevel@tonic-gate * We need to acquire the mapping list lock to protect 25427c478bd9Sstevel@tonic-gate * against hat_pageunload(), hat_unload(), etc. 25437c478bd9Sstevel@tonic-gate */ 25447c478bd9Sstevel@tonic-gate pp = page_numtopp_nolock(PTE2PFN(pte, ht->ht_level)); 25457c478bd9Sstevel@tonic-gate if (pp == NULL) 25467c478bd9Sstevel@tonic-gate break; 25477c478bd9Sstevel@tonic-gate x86_hm_enter(pp); 25487c478bd9Sstevel@tonic-gate save_pte = pte; 25497c478bd9Sstevel@tonic-gate pte = x86pte_get(ht, entry); 25507c478bd9Sstevel@tonic-gate if (pte != save_pte) { 25517c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 25527c478bd9Sstevel@tonic-gate goto try_again; 25537c478bd9Sstevel@tonic-gate } 2554ae115bc7Smrj if (PTE_GET(pte, PT_SOFTWARE) >= PT_NOSYNC || 25557c478bd9Sstevel@tonic-gate PTE_GET(pte, PT_REF | PT_MOD) == 0) { 25567c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 25577c478bd9Sstevel@tonic-gate continue; 25587c478bd9Sstevel@tonic-gate } 25597c478bd9Sstevel@tonic-gate 25607c478bd9Sstevel@tonic-gate /* 25617c478bd9Sstevel@tonic-gate * Need to clear ref or mod bits. We may compete with 25627c478bd9Sstevel@tonic-gate * hardware updating the R/M bits and have to try again. 25637c478bd9Sstevel@tonic-gate */ 25647c478bd9Sstevel@tonic-gate if (flags == HAT_SYNC_ZERORM) { 25657c478bd9Sstevel@tonic-gate new = pte; 25667c478bd9Sstevel@tonic-gate PTE_CLR(new, PT_REF | PT_MOD); 25677c478bd9Sstevel@tonic-gate pte = hati_update_pte(ht, entry, pte, new); 25687c478bd9Sstevel@tonic-gate if (pte != 0) { 25697c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 25707c478bd9Sstevel@tonic-gate goto try_again; 25717c478bd9Sstevel@tonic-gate } 25727c478bd9Sstevel@tonic-gate } else { 25737c478bd9Sstevel@tonic-gate /* 25747c478bd9Sstevel@tonic-gate * sync the PTE to the page_t 25757c478bd9Sstevel@tonic-gate */ 25767c478bd9Sstevel@tonic-gate hati_sync_pte_to_page(pp, save_pte, ht->ht_level); 25777c478bd9Sstevel@tonic-gate } 25787c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 25797c478bd9Sstevel@tonic-gate } 25807c478bd9Sstevel@tonic-gate if (ht) 25817c478bd9Sstevel@tonic-gate htable_release(ht); 2582843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 25837c478bd9Sstevel@tonic-gate } 25847c478bd9Sstevel@tonic-gate 25857c478bd9Sstevel@tonic-gate /* 25867c478bd9Sstevel@tonic-gate * void hat_map(hat, addr, len, flags) 25877c478bd9Sstevel@tonic-gate */ 25887c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 25897c478bd9Sstevel@tonic-gate void 25907c478bd9Sstevel@tonic-gate hat_map(hat_t *hat, caddr_t addr, size_t len, uint_t flags) 25917c478bd9Sstevel@tonic-gate { 25927c478bd9Sstevel@tonic-gate /* does nothing */ 25937c478bd9Sstevel@tonic-gate } 25947c478bd9Sstevel@tonic-gate 25957c478bd9Sstevel@tonic-gate /* 25967c478bd9Sstevel@tonic-gate * uint_t hat_getattr(hat, addr, *attr) 25977c478bd9Sstevel@tonic-gate * returns attr for <hat,addr> in *attr. returns 0 if there was a 25987c478bd9Sstevel@tonic-gate * mapping and *attr is valid, nonzero if there was no mapping and 25997c478bd9Sstevel@tonic-gate * *attr is not valid. 26007c478bd9Sstevel@tonic-gate */ 26017c478bd9Sstevel@tonic-gate uint_t 26027c478bd9Sstevel@tonic-gate hat_getattr(hat_t *hat, caddr_t addr, uint_t *attr) 26037c478bd9Sstevel@tonic-gate { 26047c478bd9Sstevel@tonic-gate uintptr_t vaddr = ALIGN2PAGE(addr); 26057c478bd9Sstevel@tonic-gate htable_t *ht = NULL; 26067c478bd9Sstevel@tonic-gate x86pte_t pte; 26077c478bd9Sstevel@tonic-gate 2608ae115bc7Smrj ASSERT(hat == kas.a_hat || vaddr <= _userlimit); 26097c478bd9Sstevel@tonic-gate 26107c478bd9Sstevel@tonic-gate if (IN_VA_HOLE(vaddr)) 26117c478bd9Sstevel@tonic-gate return ((uint_t)-1); 26127c478bd9Sstevel@tonic-gate 2613ae115bc7Smrj ht = htable_getpte(hat, vaddr, NULL, &pte, mmu.max_page_level); 26147c478bd9Sstevel@tonic-gate if (ht == NULL) 26157c478bd9Sstevel@tonic-gate return ((uint_t)-1); 26167c478bd9Sstevel@tonic-gate 26177c478bd9Sstevel@tonic-gate if (!PTE_ISVALID(pte) || !PTE_ISPAGE(pte, ht->ht_level)) { 26187c478bd9Sstevel@tonic-gate htable_release(ht); 26197c478bd9Sstevel@tonic-gate return ((uint_t)-1); 26207c478bd9Sstevel@tonic-gate } 26217c478bd9Sstevel@tonic-gate 26227c478bd9Sstevel@tonic-gate *attr = PROT_READ; 26237c478bd9Sstevel@tonic-gate if (PTE_GET(pte, PT_WRITABLE)) 26247c478bd9Sstevel@tonic-gate *attr |= PROT_WRITE; 26257c478bd9Sstevel@tonic-gate if (PTE_GET(pte, PT_USER)) 26267c478bd9Sstevel@tonic-gate *attr |= PROT_USER; 26277c478bd9Sstevel@tonic-gate if (!PTE_GET(pte, mmu.pt_nx)) 26287c478bd9Sstevel@tonic-gate *attr |= PROT_EXEC; 2629ae115bc7Smrj if (PTE_GET(pte, PT_SOFTWARE) >= PT_NOSYNC) 26307c478bd9Sstevel@tonic-gate *attr |= HAT_NOSYNC; 26317c478bd9Sstevel@tonic-gate htable_release(ht); 26327c478bd9Sstevel@tonic-gate return (0); 26337c478bd9Sstevel@tonic-gate } 26347c478bd9Sstevel@tonic-gate 26357c478bd9Sstevel@tonic-gate /* 26367c478bd9Sstevel@tonic-gate * hat_updateattr() applies the given attribute change to an existing mapping 26377c478bd9Sstevel@tonic-gate */ 26387c478bd9Sstevel@tonic-gate #define HAT_LOAD_ATTR 1 26397c478bd9Sstevel@tonic-gate #define HAT_SET_ATTR 2 26407c478bd9Sstevel@tonic-gate #define HAT_CLR_ATTR 3 26417c478bd9Sstevel@tonic-gate 26427c478bd9Sstevel@tonic-gate static void 26437c478bd9Sstevel@tonic-gate hat_updateattr(hat_t *hat, caddr_t addr, size_t len, uint_t attr, int what) 26447c478bd9Sstevel@tonic-gate { 26457c478bd9Sstevel@tonic-gate uintptr_t vaddr = (uintptr_t)addr; 26467c478bd9Sstevel@tonic-gate uintptr_t eaddr = (uintptr_t)addr + len; 26477c478bd9Sstevel@tonic-gate htable_t *ht = NULL; 26487c478bd9Sstevel@tonic-gate uint_t entry; 26497c478bd9Sstevel@tonic-gate x86pte_t oldpte, newpte; 26507c478bd9Sstevel@tonic-gate page_t *pp; 26517c478bd9Sstevel@tonic-gate 2652843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 26537c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(vaddr)); 26547c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(eaddr)); 2655dc32d872SJosef 'Jeff' Sipek ASSERT(hat == kas.a_hat || AS_LOCK_HELD(hat->hat_as)); 26567c478bd9Sstevel@tonic-gate for (; vaddr < eaddr; vaddr += LEVEL_SIZE(ht->ht_level)) { 26577c478bd9Sstevel@tonic-gate try_again: 26587c478bd9Sstevel@tonic-gate oldpte = htable_walk(hat, &ht, &vaddr, eaddr); 26597c478bd9Sstevel@tonic-gate if (ht == NULL) 26607c478bd9Sstevel@tonic-gate break; 2661ae115bc7Smrj if (PTE_GET(oldpte, PT_SOFTWARE) >= PT_NOCONSIST) 26627c478bd9Sstevel@tonic-gate continue; 26637c478bd9Sstevel@tonic-gate 26647c478bd9Sstevel@tonic-gate pp = page_numtopp_nolock(PTE2PFN(oldpte, ht->ht_level)); 26657c478bd9Sstevel@tonic-gate if (pp == NULL) 26667c478bd9Sstevel@tonic-gate continue; 26677c478bd9Sstevel@tonic-gate x86_hm_enter(pp); 26687c478bd9Sstevel@tonic-gate 26697c478bd9Sstevel@tonic-gate newpte = oldpte; 26707c478bd9Sstevel@tonic-gate /* 26717c478bd9Sstevel@tonic-gate * We found a page table entry in the desired range, 26727c478bd9Sstevel@tonic-gate * figure out the new attributes. 26737c478bd9Sstevel@tonic-gate */ 26747c478bd9Sstevel@tonic-gate if (what == HAT_SET_ATTR || what == HAT_LOAD_ATTR) { 26757c478bd9Sstevel@tonic-gate if ((attr & PROT_WRITE) && 26767c478bd9Sstevel@tonic-gate !PTE_GET(oldpte, PT_WRITABLE)) 26777c478bd9Sstevel@tonic-gate newpte |= PT_WRITABLE; 26787c478bd9Sstevel@tonic-gate 2679ae115bc7Smrj if ((attr & HAT_NOSYNC) && 2680ae115bc7Smrj PTE_GET(oldpte, PT_SOFTWARE) < PT_NOSYNC) 26817c478bd9Sstevel@tonic-gate newpte |= PT_NOSYNC; 26827c478bd9Sstevel@tonic-gate 26837c478bd9Sstevel@tonic-gate if ((attr & PROT_EXEC) && PTE_GET(oldpte, mmu.pt_nx)) 26847c478bd9Sstevel@tonic-gate newpte &= ~mmu.pt_nx; 26857c478bd9Sstevel@tonic-gate } 26867c478bd9Sstevel@tonic-gate 26877c478bd9Sstevel@tonic-gate if (what == HAT_LOAD_ATTR) { 26887c478bd9Sstevel@tonic-gate if (!(attr & PROT_WRITE) && 26897c478bd9Sstevel@tonic-gate PTE_GET(oldpte, PT_WRITABLE)) 26907c478bd9Sstevel@tonic-gate newpte &= ~PT_WRITABLE; 26917c478bd9Sstevel@tonic-gate 2692ae115bc7Smrj if (!(attr & HAT_NOSYNC) && 2693ae115bc7Smrj PTE_GET(oldpte, PT_SOFTWARE) >= PT_NOSYNC) 2694ae115bc7Smrj newpte &= ~PT_SOFTWARE; 26957c478bd9Sstevel@tonic-gate 26967c478bd9Sstevel@tonic-gate if (!(attr & PROT_EXEC) && !PTE_GET(oldpte, mmu.pt_nx)) 26977c478bd9Sstevel@tonic-gate newpte |= mmu.pt_nx; 26987c478bd9Sstevel@tonic-gate } 26997c478bd9Sstevel@tonic-gate 27007c478bd9Sstevel@tonic-gate if (what == HAT_CLR_ATTR) { 27017c478bd9Sstevel@tonic-gate if ((attr & PROT_WRITE) && PTE_GET(oldpte, PT_WRITABLE)) 27027c478bd9Sstevel@tonic-gate newpte &= ~PT_WRITABLE; 27037c478bd9Sstevel@tonic-gate 2704ae115bc7Smrj if ((attr & HAT_NOSYNC) && 2705ae115bc7Smrj PTE_GET(oldpte, PT_SOFTWARE) >= PT_NOSYNC) 2706ae115bc7Smrj newpte &= ~PT_SOFTWARE; 27077c478bd9Sstevel@tonic-gate 27087c478bd9Sstevel@tonic-gate if ((attr & PROT_EXEC) && !PTE_GET(oldpte, mmu.pt_nx)) 27097c478bd9Sstevel@tonic-gate newpte |= mmu.pt_nx; 27107c478bd9Sstevel@tonic-gate } 27117c478bd9Sstevel@tonic-gate 2712ae115bc7Smrj /* 2713ae115bc7Smrj * Ensure NOSYNC/NOCONSIST mappings have REF and MOD set. 2714ae115bc7Smrj * x86pte_set() depends on this. 2715ae115bc7Smrj */ 2716ae115bc7Smrj if (PTE_GET(newpte, PT_SOFTWARE) >= PT_NOSYNC) 2717ae115bc7Smrj newpte |= PT_REF | PT_MOD; 2718ae115bc7Smrj 27197c478bd9Sstevel@tonic-gate /* 27207c478bd9Sstevel@tonic-gate * what about PROT_READ or others? this code only handles: 27217c478bd9Sstevel@tonic-gate * EXEC, WRITE, NOSYNC 27227c478bd9Sstevel@tonic-gate */ 27237c478bd9Sstevel@tonic-gate 27247c478bd9Sstevel@tonic-gate /* 27257c478bd9Sstevel@tonic-gate * If new PTE really changed, update the table. 27267c478bd9Sstevel@tonic-gate */ 27277c478bd9Sstevel@tonic-gate if (newpte != oldpte) { 27287c478bd9Sstevel@tonic-gate entry = htable_va2entry(vaddr, ht); 27297c478bd9Sstevel@tonic-gate oldpte = hati_update_pte(ht, entry, oldpte, newpte); 27307c478bd9Sstevel@tonic-gate if (oldpte != 0) { 27317c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 27327c478bd9Sstevel@tonic-gate goto try_again; 27337c478bd9Sstevel@tonic-gate } 27347c478bd9Sstevel@tonic-gate } 27357c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 27367c478bd9Sstevel@tonic-gate } 27377c478bd9Sstevel@tonic-gate if (ht) 27387c478bd9Sstevel@tonic-gate htable_release(ht); 2739843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 27407c478bd9Sstevel@tonic-gate } 27417c478bd9Sstevel@tonic-gate 27427c478bd9Sstevel@tonic-gate /* 27437c478bd9Sstevel@tonic-gate * Various wrappers for hat_updateattr() 27447c478bd9Sstevel@tonic-gate */ 27457c478bd9Sstevel@tonic-gate void 27467c478bd9Sstevel@tonic-gate hat_setattr(hat_t *hat, caddr_t addr, size_t len, uint_t attr) 27477c478bd9Sstevel@tonic-gate { 2748ae115bc7Smrj ASSERT(hat == kas.a_hat || (uintptr_t)addr + len <= _userlimit); 27497c478bd9Sstevel@tonic-gate hat_updateattr(hat, addr, len, attr, HAT_SET_ATTR); 27507c478bd9Sstevel@tonic-gate } 27517c478bd9Sstevel@tonic-gate 27527c478bd9Sstevel@tonic-gate void 27537c478bd9Sstevel@tonic-gate hat_clrattr(hat_t *hat, caddr_t addr, size_t len, uint_t attr) 27547c478bd9Sstevel@tonic-gate { 2755ae115bc7Smrj ASSERT(hat == kas.a_hat || (uintptr_t)addr + len <= _userlimit); 27567c478bd9Sstevel@tonic-gate hat_updateattr(hat, addr, len, attr, HAT_CLR_ATTR); 27577c478bd9Sstevel@tonic-gate } 27587c478bd9Sstevel@tonic-gate 27597c478bd9Sstevel@tonic-gate void 27607c478bd9Sstevel@tonic-gate hat_chgattr(hat_t *hat, caddr_t addr, size_t len, uint_t attr) 27617c478bd9Sstevel@tonic-gate { 2762ae115bc7Smrj ASSERT(hat == kas.a_hat || (uintptr_t)addr + len <= _userlimit); 27637c478bd9Sstevel@tonic-gate hat_updateattr(hat, addr, len, attr, HAT_LOAD_ATTR); 27647c478bd9Sstevel@tonic-gate } 27657c478bd9Sstevel@tonic-gate 27667c478bd9Sstevel@tonic-gate void 27677c478bd9Sstevel@tonic-gate hat_chgprot(hat_t *hat, caddr_t addr, size_t len, uint_t vprot) 27687c478bd9Sstevel@tonic-gate { 2769ae115bc7Smrj ASSERT(hat == kas.a_hat || (uintptr_t)addr + len <= _userlimit); 27707c478bd9Sstevel@tonic-gate hat_updateattr(hat, addr, len, vprot & HAT_PROT_MASK, HAT_LOAD_ATTR); 27717c478bd9Sstevel@tonic-gate } 27727c478bd9Sstevel@tonic-gate 27737c478bd9Sstevel@tonic-gate /* 27747c478bd9Sstevel@tonic-gate * size_t hat_getpagesize(hat, addr) 27757c478bd9Sstevel@tonic-gate * returns pagesize in bytes for <hat, addr>. returns -1 of there is 27767c478bd9Sstevel@tonic-gate * no mapping. This is an advisory call. 27777c478bd9Sstevel@tonic-gate */ 27787c478bd9Sstevel@tonic-gate ssize_t 27797c478bd9Sstevel@tonic-gate hat_getpagesize(hat_t *hat, caddr_t addr) 27807c478bd9Sstevel@tonic-gate { 27817c478bd9Sstevel@tonic-gate uintptr_t vaddr = ALIGN2PAGE(addr); 27827c478bd9Sstevel@tonic-gate htable_t *ht; 27837c478bd9Sstevel@tonic-gate size_t pagesize; 27847c478bd9Sstevel@tonic-gate 2785ae115bc7Smrj ASSERT(hat == kas.a_hat || vaddr <= _userlimit); 27867c478bd9Sstevel@tonic-gate if (IN_VA_HOLE(vaddr)) 27877c478bd9Sstevel@tonic-gate return (-1); 27887c478bd9Sstevel@tonic-gate ht = htable_getpage(hat, vaddr, NULL); 27897c478bd9Sstevel@tonic-gate if (ht == NULL) 27907c478bd9Sstevel@tonic-gate return (-1); 27917c478bd9Sstevel@tonic-gate pagesize = LEVEL_SIZE(ht->ht_level); 27927c478bd9Sstevel@tonic-gate htable_release(ht); 27937c478bd9Sstevel@tonic-gate return (pagesize); 27947c478bd9Sstevel@tonic-gate } 27957c478bd9Sstevel@tonic-gate 27967c478bd9Sstevel@tonic-gate 27977c478bd9Sstevel@tonic-gate 27987c478bd9Sstevel@tonic-gate /* 27997c478bd9Sstevel@tonic-gate * pfn_t hat_getpfnum(hat, addr) 28007c478bd9Sstevel@tonic-gate * returns pfn for <hat, addr> or PFN_INVALID if mapping is invalid. 28017c478bd9Sstevel@tonic-gate */ 28027c478bd9Sstevel@tonic-gate pfn_t 28037c478bd9Sstevel@tonic-gate hat_getpfnum(hat_t *hat, caddr_t addr) 28047c478bd9Sstevel@tonic-gate { 28057c478bd9Sstevel@tonic-gate uintptr_t vaddr = ALIGN2PAGE(addr); 28067c478bd9Sstevel@tonic-gate htable_t *ht; 28077c478bd9Sstevel@tonic-gate uint_t entry; 28087c478bd9Sstevel@tonic-gate pfn_t pfn = PFN_INVALID; 28097c478bd9Sstevel@tonic-gate 2810ae115bc7Smrj ASSERT(hat == kas.a_hat || vaddr <= _userlimit); 28117c478bd9Sstevel@tonic-gate if (khat_running == 0) 2812ae115bc7Smrj return (PFN_INVALID); 28137c478bd9Sstevel@tonic-gate 28147c478bd9Sstevel@tonic-gate if (IN_VA_HOLE(vaddr)) 28157c478bd9Sstevel@tonic-gate return (PFN_INVALID); 28167c478bd9Sstevel@tonic-gate 2817843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 28187c478bd9Sstevel@tonic-gate /* 28197c478bd9Sstevel@tonic-gate * A very common use of hat_getpfnum() is from the DDI for kernel pages. 28207c478bd9Sstevel@tonic-gate * Use the kmap_ptes (which also covers the 32 bit heap) to speed 28217c478bd9Sstevel@tonic-gate * this up. 28227c478bd9Sstevel@tonic-gate */ 28237c478bd9Sstevel@tonic-gate if (mmu.kmap_addr <= vaddr && vaddr < mmu.kmap_eaddr) { 28247c478bd9Sstevel@tonic-gate x86pte_t pte; 2825ae115bc7Smrj pgcnt_t pg_index; 28267c478bd9Sstevel@tonic-gate 2827ae115bc7Smrj pg_index = mmu_btop(vaddr - mmu.kmap_addr); 2828ae115bc7Smrj pte = GET_PTE(PT_INDEX_PTR(mmu.kmap_ptes, pg_index)); 2829843e1988Sjohnlev if (PTE_ISVALID(pte)) 2830843e1988Sjohnlev /*LINTED [use of constant 0 causes a lint warning] */ 2831843e1988Sjohnlev pfn = PTE2PFN(pte, 0); 2832843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 2833843e1988Sjohnlev return (pfn); 28347c478bd9Sstevel@tonic-gate } 28357c478bd9Sstevel@tonic-gate 28367c478bd9Sstevel@tonic-gate ht = htable_getpage(hat, vaddr, &entry); 2837843e1988Sjohnlev if (ht == NULL) { 2838843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 28397c478bd9Sstevel@tonic-gate return (PFN_INVALID); 2840843e1988Sjohnlev } 28417c478bd9Sstevel@tonic-gate ASSERT(vaddr >= ht->ht_vaddr); 28427c478bd9Sstevel@tonic-gate ASSERT(vaddr <= HTABLE_LAST_PAGE(ht)); 28437c478bd9Sstevel@tonic-gate pfn = PTE2PFN(x86pte_get(ht, entry), ht->ht_level); 28447c478bd9Sstevel@tonic-gate if (ht->ht_level > 0) 28457c478bd9Sstevel@tonic-gate pfn += mmu_btop(vaddr & LEVEL_OFFSET(ht->ht_level)); 28467c478bd9Sstevel@tonic-gate htable_release(ht); 2847843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 28487c478bd9Sstevel@tonic-gate return (pfn); 28497c478bd9Sstevel@tonic-gate } 28507c478bd9Sstevel@tonic-gate 28517c478bd9Sstevel@tonic-gate /* 28527c478bd9Sstevel@tonic-gate * int hat_probe(hat, addr) 28537c478bd9Sstevel@tonic-gate * return 0 if no valid mapping is present. Faster version 28547c478bd9Sstevel@tonic-gate * of hat_getattr in certain architectures. 28557c478bd9Sstevel@tonic-gate */ 28567c478bd9Sstevel@tonic-gate int 28577c478bd9Sstevel@tonic-gate hat_probe(hat_t *hat, caddr_t addr) 28587c478bd9Sstevel@tonic-gate { 28597c478bd9Sstevel@tonic-gate uintptr_t vaddr = ALIGN2PAGE(addr); 28607c478bd9Sstevel@tonic-gate uint_t entry; 28617c478bd9Sstevel@tonic-gate htable_t *ht; 28627c478bd9Sstevel@tonic-gate pgcnt_t pg_off; 28637c478bd9Sstevel@tonic-gate 2864ae115bc7Smrj ASSERT(hat == kas.a_hat || vaddr <= _userlimit); 2865dc32d872SJosef 'Jeff' Sipek ASSERT(hat == kas.a_hat || AS_LOCK_HELD(hat->hat_as)); 28667c478bd9Sstevel@tonic-gate if (IN_VA_HOLE(vaddr)) 28677c478bd9Sstevel@tonic-gate return (0); 28687c478bd9Sstevel@tonic-gate 28697c478bd9Sstevel@tonic-gate /* 28707c478bd9Sstevel@tonic-gate * Most common use of hat_probe is from segmap. We special case it 28717c478bd9Sstevel@tonic-gate * for performance. 28727c478bd9Sstevel@tonic-gate */ 28737c478bd9Sstevel@tonic-gate if (mmu.kmap_addr <= vaddr && vaddr < mmu.kmap_eaddr) { 28747c478bd9Sstevel@tonic-gate pg_off = mmu_btop(vaddr - mmu.kmap_addr); 28757c478bd9Sstevel@tonic-gate if (mmu.pae_hat) 28767c478bd9Sstevel@tonic-gate return (PTE_ISVALID(mmu.kmap_ptes[pg_off])); 28777c478bd9Sstevel@tonic-gate else 28787c478bd9Sstevel@tonic-gate return (PTE_ISVALID( 28797c478bd9Sstevel@tonic-gate ((x86pte32_t *)mmu.kmap_ptes)[pg_off])); 28807c478bd9Sstevel@tonic-gate } 28817c478bd9Sstevel@tonic-gate 28827c478bd9Sstevel@tonic-gate ht = htable_getpage(hat, vaddr, &entry); 28837c478bd9Sstevel@tonic-gate htable_release(ht); 2884843e1988Sjohnlev return (ht != NULL); 28857c478bd9Sstevel@tonic-gate } 28867c478bd9Sstevel@tonic-gate 28877c478bd9Sstevel@tonic-gate /* 2888250b7ff9Sjosephb * Find out if the segment for hat_share()/hat_unshare() is DISM or locked ISM. 2889250b7ff9Sjosephb */ 2890250b7ff9Sjosephb static int 2891250b7ff9Sjosephb is_it_dism(hat_t *hat, caddr_t va) 2892250b7ff9Sjosephb { 2893250b7ff9Sjosephb struct seg *seg; 2894250b7ff9Sjosephb struct shm_data *shmd; 2895250b7ff9Sjosephb struct spt_data *sptd; 2896250b7ff9Sjosephb 2897250b7ff9Sjosephb seg = as_findseg(hat->hat_as, va, 0); 2898250b7ff9Sjosephb ASSERT(seg != NULL); 2899250b7ff9Sjosephb ASSERT(seg->s_base <= va); 2900250b7ff9Sjosephb shmd = (struct shm_data *)seg->s_data; 2901250b7ff9Sjosephb ASSERT(shmd != NULL); 2902250b7ff9Sjosephb sptd = (struct spt_data *)shmd->shm_sptseg->s_data; 2903250b7ff9Sjosephb ASSERT(sptd != NULL); 2904250b7ff9Sjosephb if (sptd->spt_flags & SHM_PAGEABLE) 2905250b7ff9Sjosephb return (1); 2906250b7ff9Sjosephb return (0); 2907250b7ff9Sjosephb } 2908250b7ff9Sjosephb 2909250b7ff9Sjosephb /* 2910250b7ff9Sjosephb * Simple implementation of ISM. hat_share() is similar to hat_memload_array(), 29117c478bd9Sstevel@tonic-gate * except that we use the ism_hat's existing mappings to determine the pages 2912250b7ff9Sjosephb * and protections to use for this hat. If we find a full properly aligned 2913250b7ff9Sjosephb * and sized pagetable, we will attempt to share the pagetable itself. 29147c478bd9Sstevel@tonic-gate */ 29157c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 29167c478bd9Sstevel@tonic-gate int 29177c478bd9Sstevel@tonic-gate hat_share( 29187c478bd9Sstevel@tonic-gate hat_t *hat, 29197c478bd9Sstevel@tonic-gate caddr_t addr, 29207c478bd9Sstevel@tonic-gate hat_t *ism_hat, 29217c478bd9Sstevel@tonic-gate caddr_t src_addr, 29227c478bd9Sstevel@tonic-gate size_t len, /* almost useless value, see below.. */ 29237c478bd9Sstevel@tonic-gate uint_t ismszc) 29247c478bd9Sstevel@tonic-gate { 29257c478bd9Sstevel@tonic-gate uintptr_t vaddr_start = (uintptr_t)addr; 29267c478bd9Sstevel@tonic-gate uintptr_t vaddr; 29277c478bd9Sstevel@tonic-gate uintptr_t eaddr = vaddr_start + len; 29287c478bd9Sstevel@tonic-gate uintptr_t ism_addr_start = (uintptr_t)src_addr; 29297c478bd9Sstevel@tonic-gate uintptr_t ism_addr = ism_addr_start; 29307c478bd9Sstevel@tonic-gate uintptr_t e_ism_addr = ism_addr + len; 29317c478bd9Sstevel@tonic-gate htable_t *ism_ht = NULL; 29327c478bd9Sstevel@tonic-gate htable_t *ht; 29337c478bd9Sstevel@tonic-gate x86pte_t pte; 29347c478bd9Sstevel@tonic-gate page_t *pp; 29357c478bd9Sstevel@tonic-gate pfn_t pfn; 29367c478bd9Sstevel@tonic-gate level_t l; 29377c478bd9Sstevel@tonic-gate pgcnt_t pgcnt; 29387c478bd9Sstevel@tonic-gate uint_t prot; 2939250b7ff9Sjosephb int is_dism; 2940250b7ff9Sjosephb int flags; 29417c478bd9Sstevel@tonic-gate 29427c478bd9Sstevel@tonic-gate /* 29437c478bd9Sstevel@tonic-gate * We might be asked to share an empty DISM hat by as_dup() 29447c478bd9Sstevel@tonic-gate */ 29457c478bd9Sstevel@tonic-gate ASSERT(hat != kas.a_hat); 2946ae115bc7Smrj ASSERT(eaddr <= _userlimit); 29477c478bd9Sstevel@tonic-gate if (!(ism_hat->hat_flags & HAT_SHARED)) { 29487c478bd9Sstevel@tonic-gate ASSERT(hat_get_mapped_size(ism_hat) == 0); 29497c478bd9Sstevel@tonic-gate return (0); 29507c478bd9Sstevel@tonic-gate } 2951843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 29527c478bd9Sstevel@tonic-gate 29537c478bd9Sstevel@tonic-gate /* 29547c478bd9Sstevel@tonic-gate * The SPT segment driver often passes us a size larger than there are 29557c478bd9Sstevel@tonic-gate * valid mappings. That's because it rounds the segment size up to a 29567c478bd9Sstevel@tonic-gate * large pagesize, even if the actual memory mapped by ism_hat is less. 29577c478bd9Sstevel@tonic-gate */ 29587c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(vaddr_start)); 29597c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(ism_addr_start)); 29607c478bd9Sstevel@tonic-gate ASSERT(ism_hat->hat_flags & HAT_SHARED); 2961250b7ff9Sjosephb is_dism = is_it_dism(hat, addr); 29627c478bd9Sstevel@tonic-gate while (ism_addr < e_ism_addr) { 29637c478bd9Sstevel@tonic-gate /* 29647c478bd9Sstevel@tonic-gate * use htable_walk to get the next valid ISM mapping 29657c478bd9Sstevel@tonic-gate */ 29667c478bd9Sstevel@tonic-gate pte = htable_walk(ism_hat, &ism_ht, &ism_addr, e_ism_addr); 29677c478bd9Sstevel@tonic-gate if (ism_ht == NULL) 29687c478bd9Sstevel@tonic-gate break; 29697c478bd9Sstevel@tonic-gate 29707c478bd9Sstevel@tonic-gate /* 2971250b7ff9Sjosephb * First check to see if we already share the page table. 29727c478bd9Sstevel@tonic-gate */ 2973250b7ff9Sjosephb l = ism_ht->ht_level; 29747c478bd9Sstevel@tonic-gate vaddr = vaddr_start + (ism_addr - ism_addr_start); 2975250b7ff9Sjosephb ht = htable_lookup(hat, vaddr, l); 2976250b7ff9Sjosephb if (ht != NULL) { 2977250b7ff9Sjosephb if (ht->ht_flags & HTABLE_SHARED_PFN) 2978250b7ff9Sjosephb goto shared; 2979250b7ff9Sjosephb htable_release(ht); 2980250b7ff9Sjosephb goto not_shared; 29817c478bd9Sstevel@tonic-gate } 29827c478bd9Sstevel@tonic-gate 29837c478bd9Sstevel@tonic-gate /* 2984250b7ff9Sjosephb * Can't ever share top table. 29857c478bd9Sstevel@tonic-gate */ 2986250b7ff9Sjosephb if (l == mmu.max_level) 2987250b7ff9Sjosephb goto not_shared; 29887c478bd9Sstevel@tonic-gate 2989250b7ff9Sjosephb /* 2990250b7ff9Sjosephb * Avoid level mismatches later due to DISM faults. 2991250b7ff9Sjosephb */ 2992250b7ff9Sjosephb if (is_dism && l > 0) 2993250b7ff9Sjosephb goto not_shared; 29947c478bd9Sstevel@tonic-gate 2995250b7ff9Sjosephb /* 2996250b7ff9Sjosephb * addresses and lengths must align 2997250b7ff9Sjosephb * table must be fully populated 2998250b7ff9Sjosephb * no lower level page tables 2999250b7ff9Sjosephb */ 3000250b7ff9Sjosephb if (ism_addr != ism_ht->ht_vaddr || 3001250b7ff9Sjosephb (vaddr & LEVEL_OFFSET(l + 1)) != 0) 3002250b7ff9Sjosephb goto not_shared; 30037c478bd9Sstevel@tonic-gate 3004250b7ff9Sjosephb /* 3005250b7ff9Sjosephb * The range of address space must cover a full table. 3006250b7ff9Sjosephb */ 30071d03c31eSjohnlev if (e_ism_addr - ism_addr < LEVEL_SIZE(l + 1)) 3008250b7ff9Sjosephb goto not_shared; 30097c478bd9Sstevel@tonic-gate 3010250b7ff9Sjosephb /* 3011250b7ff9Sjosephb * All entries in the ISM page table must be leaf PTEs. 3012250b7ff9Sjosephb */ 3013250b7ff9Sjosephb if (l > 0) { 3014250b7ff9Sjosephb int e; 30157c478bd9Sstevel@tonic-gate 3016250b7ff9Sjosephb /* 3017250b7ff9Sjosephb * We know the 0th is from htable_walk() above. 3018250b7ff9Sjosephb */ 3019250b7ff9Sjosephb for (e = 1; e < HTABLE_NUM_PTES(ism_ht); ++e) { 3020250b7ff9Sjosephb x86pte_t pte; 3021250b7ff9Sjosephb pte = x86pte_get(ism_ht, e); 3022250b7ff9Sjosephb if (!PTE_ISPAGE(pte, l)) 3023250b7ff9Sjosephb goto not_shared; 30247c478bd9Sstevel@tonic-gate } 30257c478bd9Sstevel@tonic-gate } 30267c478bd9Sstevel@tonic-gate 3027250b7ff9Sjosephb /* 3028250b7ff9Sjosephb * share the page table 3029250b7ff9Sjosephb */ 3030250b7ff9Sjosephb ht = htable_create(hat, vaddr, l, ism_ht); 3031250b7ff9Sjosephb shared: 3032250b7ff9Sjosephb ASSERT(ht->ht_flags & HTABLE_SHARED_PFN); 3033250b7ff9Sjosephb ASSERT(ht->ht_shares == ism_ht); 3034250b7ff9Sjosephb hat->hat_ism_pgcnt += 3035250b7ff9Sjosephb (ism_ht->ht_valid_cnt - ht->ht_valid_cnt) << 3036250b7ff9Sjosephb (LEVEL_SHIFT(ht->ht_level) - MMU_PAGESHIFT); 3037250b7ff9Sjosephb ht->ht_valid_cnt = ism_ht->ht_valid_cnt; 3038250b7ff9Sjosephb htable_release(ht); 3039250b7ff9Sjosephb ism_addr = ism_ht->ht_vaddr + LEVEL_SIZE(l + 1); 3040250b7ff9Sjosephb htable_release(ism_ht); 3041250b7ff9Sjosephb ism_ht = NULL; 3042250b7ff9Sjosephb continue; 3043250b7ff9Sjosephb 3044250b7ff9Sjosephb not_shared: 30457c478bd9Sstevel@tonic-gate /* 30467c478bd9Sstevel@tonic-gate * Unable to share the page table. Instead we will 30477c478bd9Sstevel@tonic-gate * create new mappings from the values in the ISM mappings. 3048250b7ff9Sjosephb * Figure out what level size mappings to use; 3049250b7ff9Sjosephb */ 3050250b7ff9Sjosephb for (l = ism_ht->ht_level; l > 0; --l) { 3051250b7ff9Sjosephb if (LEVEL_SIZE(l) <= eaddr - vaddr && 3052250b7ff9Sjosephb (vaddr & LEVEL_OFFSET(l)) == 0) 3053250b7ff9Sjosephb break; 3054250b7ff9Sjosephb } 3055250b7ff9Sjosephb 3056250b7ff9Sjosephb /* 30577c478bd9Sstevel@tonic-gate * The ISM mapping might be larger than the share area, 3058250b7ff9Sjosephb * be careful to truncate it if needed. 30597c478bd9Sstevel@tonic-gate */ 30607c478bd9Sstevel@tonic-gate if (eaddr - vaddr >= LEVEL_SIZE(ism_ht->ht_level)) { 30617c478bd9Sstevel@tonic-gate pgcnt = mmu_btop(LEVEL_SIZE(ism_ht->ht_level)); 30627c478bd9Sstevel@tonic-gate } else { 30637c478bd9Sstevel@tonic-gate pgcnt = mmu_btop(eaddr - vaddr); 30647c478bd9Sstevel@tonic-gate l = 0; 30657c478bd9Sstevel@tonic-gate } 30667c478bd9Sstevel@tonic-gate 30677c478bd9Sstevel@tonic-gate pfn = PTE2PFN(pte, ism_ht->ht_level); 30687c478bd9Sstevel@tonic-gate ASSERT(pfn != PFN_INVALID); 30697c478bd9Sstevel@tonic-gate while (pgcnt > 0) { 30707c478bd9Sstevel@tonic-gate /* 30717c478bd9Sstevel@tonic-gate * Make a new pte for the PFN for this level. 30727c478bd9Sstevel@tonic-gate * Copy protections for the pte from the ISM pte. 30737c478bd9Sstevel@tonic-gate */ 30747c478bd9Sstevel@tonic-gate pp = page_numtopp_nolock(pfn); 30757c478bd9Sstevel@tonic-gate ASSERT(pp != NULL); 30767c478bd9Sstevel@tonic-gate 30777c478bd9Sstevel@tonic-gate prot = PROT_USER | PROT_READ | HAT_UNORDERED_OK; 30787c478bd9Sstevel@tonic-gate if (PTE_GET(pte, PT_WRITABLE)) 30797c478bd9Sstevel@tonic-gate prot |= PROT_WRITE; 30807c478bd9Sstevel@tonic-gate if (!PTE_GET(pte, PT_NX)) 30817c478bd9Sstevel@tonic-gate prot |= PROT_EXEC; 30827c478bd9Sstevel@tonic-gate 3083250b7ff9Sjosephb flags = HAT_LOAD; 3084250b7ff9Sjosephb if (!is_dism) 3085250b7ff9Sjosephb flags |= HAT_LOAD_LOCK | HAT_LOAD_NOCONSIST; 3086250b7ff9Sjosephb while (hati_load_common(hat, vaddr, pp, prot, flags, 3087ae115bc7Smrj l, pfn) != 0) { 3088ae115bc7Smrj if (l == 0) 3089ae115bc7Smrj panic("hati_load_common() failure"); 3090ae115bc7Smrj --l; 3091ae115bc7Smrj } 30927c478bd9Sstevel@tonic-gate 30937c478bd9Sstevel@tonic-gate vaddr += LEVEL_SIZE(l); 30947c478bd9Sstevel@tonic-gate ism_addr += LEVEL_SIZE(l); 30957c478bd9Sstevel@tonic-gate pfn += mmu_btop(LEVEL_SIZE(l)); 30967c478bd9Sstevel@tonic-gate pgcnt -= mmu_btop(LEVEL_SIZE(l)); 30977c478bd9Sstevel@tonic-gate } 30987c478bd9Sstevel@tonic-gate } 30997c478bd9Sstevel@tonic-gate if (ism_ht != NULL) 31007c478bd9Sstevel@tonic-gate htable_release(ism_ht); 3101843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 31027c478bd9Sstevel@tonic-gate return (0); 31037c478bd9Sstevel@tonic-gate } 31047c478bd9Sstevel@tonic-gate 31057c478bd9Sstevel@tonic-gate 31067c478bd9Sstevel@tonic-gate /* 31077c478bd9Sstevel@tonic-gate * hat_unshare() is similar to hat_unload_callback(), but 31087c478bd9Sstevel@tonic-gate * we have to look for empty shared pagetables. Note that 31097c478bd9Sstevel@tonic-gate * hat_unshare() is always invoked against an entire segment. 31107c478bd9Sstevel@tonic-gate */ 31117c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 31127c478bd9Sstevel@tonic-gate void 31137c478bd9Sstevel@tonic-gate hat_unshare(hat_t *hat, caddr_t addr, size_t len, uint_t ismszc) 31147c478bd9Sstevel@tonic-gate { 31157173d045Sjosephb uint64_t vaddr = (uintptr_t)addr; 31167c478bd9Sstevel@tonic-gate uintptr_t eaddr = vaddr + len; 31177c478bd9Sstevel@tonic-gate htable_t *ht = NULL; 31187c478bd9Sstevel@tonic-gate uint_t need_demaps = 0; 3119250b7ff9Sjosephb int flags = HAT_UNLOAD_UNMAP; 3120250b7ff9Sjosephb level_t l; 31217c478bd9Sstevel@tonic-gate 31227c478bd9Sstevel@tonic-gate ASSERT(hat != kas.a_hat); 3123ae115bc7Smrj ASSERT(eaddr <= _userlimit); 31247c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(vaddr)); 31257c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(eaddr)); 3126843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 31277c478bd9Sstevel@tonic-gate 31287c478bd9Sstevel@tonic-gate /* 31297c478bd9Sstevel@tonic-gate * First go through and remove any shared pagetables. 31307c478bd9Sstevel@tonic-gate * 3131ae115bc7Smrj * Note that it's ok to delay the TLB shootdown till the entire range is 31327c478bd9Sstevel@tonic-gate * finished, because if hat_pageunload() were to unload a shared 3133ae115bc7Smrj * pagetable page, its hat_tlb_inval() will do a global TLB invalidate. 31347c478bd9Sstevel@tonic-gate */ 3135250b7ff9Sjosephb l = mmu.max_page_level; 3136250b7ff9Sjosephb if (l == mmu.max_level) 3137250b7ff9Sjosephb --l; 3138250b7ff9Sjosephb for (; l >= 0; --l) { 3139250b7ff9Sjosephb for (vaddr = (uintptr_t)addr; vaddr < eaddr; 3140250b7ff9Sjosephb vaddr = (vaddr & LEVEL_MASK(l + 1)) + LEVEL_SIZE(l + 1)) { 3141250b7ff9Sjosephb ASSERT(!IN_VA_HOLE(vaddr)); 3142250b7ff9Sjosephb /* 3143250b7ff9Sjosephb * find a pagetable that maps the current address 3144250b7ff9Sjosephb */ 3145250b7ff9Sjosephb ht = htable_lookup(hat, vaddr, l); 3146250b7ff9Sjosephb if (ht == NULL) 3147250b7ff9Sjosephb continue; 31487c478bd9Sstevel@tonic-gate if (ht->ht_flags & HTABLE_SHARED_PFN) { 31497c478bd9Sstevel@tonic-gate /* 3150250b7ff9Sjosephb * clear page count, set valid_cnt to 0, 3151250b7ff9Sjosephb * let htable_release() finish the job 31527c478bd9Sstevel@tonic-gate */ 3153250b7ff9Sjosephb hat->hat_ism_pgcnt -= ht->ht_valid_cnt << 3154250b7ff9Sjosephb (LEVEL_SHIFT(ht->ht_level) - MMU_PAGESHIFT); 31557c478bd9Sstevel@tonic-gate ht->ht_valid_cnt = 0; 31567c478bd9Sstevel@tonic-gate need_demaps = 1; 31577c478bd9Sstevel@tonic-gate } 31587c478bd9Sstevel@tonic-gate htable_release(ht); 31597c478bd9Sstevel@tonic-gate } 31607c478bd9Sstevel@tonic-gate } 31617c478bd9Sstevel@tonic-gate 31627c478bd9Sstevel@tonic-gate /* 31637c478bd9Sstevel@tonic-gate * flush the TLBs - since we're probably dealing with MANY mappings 31647c478bd9Sstevel@tonic-gate * we do just one CR3 reload. 31657c478bd9Sstevel@tonic-gate */ 31667c478bd9Sstevel@tonic-gate if (!(hat->hat_flags & HAT_FREEING) && need_demaps) 3167ae115bc7Smrj hat_tlb_inval(hat, DEMAP_ALL_ADDR); 31687c478bd9Sstevel@tonic-gate 31697c478bd9Sstevel@tonic-gate /* 31707c478bd9Sstevel@tonic-gate * Now go back and clean up any unaligned mappings that 31717c478bd9Sstevel@tonic-gate * couldn't share pagetables. 31727c478bd9Sstevel@tonic-gate */ 3173250b7ff9Sjosephb if (!is_it_dism(hat, addr)) 3174250b7ff9Sjosephb flags |= HAT_UNLOAD_UNLOCK; 3175250b7ff9Sjosephb hat_unload(hat, addr, len, flags); 3176843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 31777c478bd9Sstevel@tonic-gate } 31787c478bd9Sstevel@tonic-gate 31797c478bd9Sstevel@tonic-gate 31807c478bd9Sstevel@tonic-gate /* 31817c478bd9Sstevel@tonic-gate * hat_reserve() does nothing 31827c478bd9Sstevel@tonic-gate */ 31837c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 31847c478bd9Sstevel@tonic-gate void 31857c478bd9Sstevel@tonic-gate hat_reserve(struct as *as, caddr_t addr, size_t len) 31867c478bd9Sstevel@tonic-gate { 31877c478bd9Sstevel@tonic-gate } 31887c478bd9Sstevel@tonic-gate 31897c478bd9Sstevel@tonic-gate 31907c478bd9Sstevel@tonic-gate /* 31917c478bd9Sstevel@tonic-gate * Called when all mappings to a page should have write permission removed. 319221584dbcSPavel Tatashin * Mostly stolen from hat_pagesync() 31937c478bd9Sstevel@tonic-gate */ 31947c478bd9Sstevel@tonic-gate static void 31957c478bd9Sstevel@tonic-gate hati_page_clrwrt(struct page *pp) 31967c478bd9Sstevel@tonic-gate { 31977c478bd9Sstevel@tonic-gate hment_t *hm = NULL; 31987c478bd9Sstevel@tonic-gate htable_t *ht; 31997c478bd9Sstevel@tonic-gate uint_t entry; 32007c478bd9Sstevel@tonic-gate x86pte_t old; 32017c478bd9Sstevel@tonic-gate x86pte_t new; 32027c478bd9Sstevel@tonic-gate uint_t pszc = 0; 32037c478bd9Sstevel@tonic-gate 3204843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 32057c478bd9Sstevel@tonic-gate next_size: 32067c478bd9Sstevel@tonic-gate /* 32077c478bd9Sstevel@tonic-gate * walk thru the mapping list clearing write permission 32087c478bd9Sstevel@tonic-gate */ 32097c478bd9Sstevel@tonic-gate x86_hm_enter(pp); 32107c478bd9Sstevel@tonic-gate while ((hm = hment_walk(pp, &ht, &entry, hm)) != NULL) { 32117c478bd9Sstevel@tonic-gate if (ht->ht_level < pszc) 32127c478bd9Sstevel@tonic-gate continue; 32137c478bd9Sstevel@tonic-gate old = x86pte_get(ht, entry); 32147c478bd9Sstevel@tonic-gate 32157c478bd9Sstevel@tonic-gate for (;;) { 32167c478bd9Sstevel@tonic-gate /* 32177c478bd9Sstevel@tonic-gate * Is this mapping of interest? 32187c478bd9Sstevel@tonic-gate */ 32197c478bd9Sstevel@tonic-gate if (PTE2PFN(old, ht->ht_level) != pp->p_pagenum || 32207c478bd9Sstevel@tonic-gate PTE_GET(old, PT_WRITABLE) == 0) 32217c478bd9Sstevel@tonic-gate break; 32227c478bd9Sstevel@tonic-gate 32237c478bd9Sstevel@tonic-gate /* 32247c478bd9Sstevel@tonic-gate * Clear ref/mod writable bits. This requires cross 32257c478bd9Sstevel@tonic-gate * calls to ensure any executing TLBs see cleared bits. 32267c478bd9Sstevel@tonic-gate */ 32277c478bd9Sstevel@tonic-gate new = old; 32287c478bd9Sstevel@tonic-gate PTE_CLR(new, PT_REF | PT_MOD | PT_WRITABLE); 32297c478bd9Sstevel@tonic-gate old = hati_update_pte(ht, entry, old, new); 32307c478bd9Sstevel@tonic-gate if (old != 0) 32317c478bd9Sstevel@tonic-gate continue; 32327c478bd9Sstevel@tonic-gate 32337c478bd9Sstevel@tonic-gate break; 32347c478bd9Sstevel@tonic-gate } 32357c478bd9Sstevel@tonic-gate } 32367c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 32377c478bd9Sstevel@tonic-gate while (pszc < pp->p_szc) { 32387c478bd9Sstevel@tonic-gate page_t *tpp; 32397c478bd9Sstevel@tonic-gate pszc++; 32407c478bd9Sstevel@tonic-gate tpp = PP_GROUPLEADER(pp, pszc); 32417c478bd9Sstevel@tonic-gate if (pp != tpp) { 32427c478bd9Sstevel@tonic-gate pp = tpp; 32437c478bd9Sstevel@tonic-gate goto next_size; 32447c478bd9Sstevel@tonic-gate } 32457c478bd9Sstevel@tonic-gate } 3246843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 32477c478bd9Sstevel@tonic-gate } 32487c478bd9Sstevel@tonic-gate 32497c478bd9Sstevel@tonic-gate /* 32507c478bd9Sstevel@tonic-gate * void hat_page_setattr(pp, flag) 32517c478bd9Sstevel@tonic-gate * void hat_page_clrattr(pp, flag) 32527c478bd9Sstevel@tonic-gate * used to set/clr ref/mod bits. 32537c478bd9Sstevel@tonic-gate */ 32547c478bd9Sstevel@tonic-gate void 32557c478bd9Sstevel@tonic-gate hat_page_setattr(struct page *pp, uint_t flag) 32567c478bd9Sstevel@tonic-gate { 32577c478bd9Sstevel@tonic-gate vnode_t *vp = pp->p_vnode; 32587c478bd9Sstevel@tonic-gate kmutex_t *vphm = NULL; 32597c478bd9Sstevel@tonic-gate page_t **listp; 3260d9615970Sqiao int noshuffle; 3261d9615970Sqiao 3262d9615970Sqiao noshuffle = flag & P_NSH; 3263d9615970Sqiao flag &= ~P_NSH; 32647c478bd9Sstevel@tonic-gate 32657c478bd9Sstevel@tonic-gate if (PP_GETRM(pp, flag) == flag) 32667c478bd9Sstevel@tonic-gate return; 32677c478bd9Sstevel@tonic-gate 3268d9615970Sqiao if ((flag & P_MOD) != 0 && vp != NULL && IS_VMODSORT(vp) && 3269d9615970Sqiao !noshuffle) { 32707c478bd9Sstevel@tonic-gate vphm = page_vnode_mutex(vp); 32717c478bd9Sstevel@tonic-gate mutex_enter(vphm); 32727c478bd9Sstevel@tonic-gate } 32737c478bd9Sstevel@tonic-gate 32747c478bd9Sstevel@tonic-gate PP_SETRM(pp, flag); 32757c478bd9Sstevel@tonic-gate 32767c478bd9Sstevel@tonic-gate if (vphm != NULL) { 32777c478bd9Sstevel@tonic-gate 32787c478bd9Sstevel@tonic-gate /* 32797c478bd9Sstevel@tonic-gate * Some File Systems examine v_pages for NULL w/o 32807c478bd9Sstevel@tonic-gate * grabbing the vphm mutex. Must not let it become NULL when 32817c478bd9Sstevel@tonic-gate * pp is the only page on the list. 32827c478bd9Sstevel@tonic-gate */ 32837c478bd9Sstevel@tonic-gate if (pp->p_vpnext != pp) { 32847c478bd9Sstevel@tonic-gate page_vpsub(&vp->v_pages, pp); 32857c478bd9Sstevel@tonic-gate if (vp->v_pages != NULL) 32867c478bd9Sstevel@tonic-gate listp = &vp->v_pages->p_vpprev->p_vpnext; 32877c478bd9Sstevel@tonic-gate else 32887c478bd9Sstevel@tonic-gate listp = &vp->v_pages; 32897c478bd9Sstevel@tonic-gate page_vpadd(listp, pp); 32907c478bd9Sstevel@tonic-gate } 32917c478bd9Sstevel@tonic-gate mutex_exit(vphm); 32927c478bd9Sstevel@tonic-gate } 32937c478bd9Sstevel@tonic-gate } 32947c478bd9Sstevel@tonic-gate 32957c478bd9Sstevel@tonic-gate void 32967c478bd9Sstevel@tonic-gate hat_page_clrattr(struct page *pp, uint_t flag) 32977c478bd9Sstevel@tonic-gate { 32987c478bd9Sstevel@tonic-gate vnode_t *vp = pp->p_vnode; 32997c478bd9Sstevel@tonic-gate ASSERT(!(flag & ~(P_MOD | P_REF | P_RO))); 33007c478bd9Sstevel@tonic-gate 33017c478bd9Sstevel@tonic-gate /* 3302a71e32b6Sstans * Caller is expected to hold page's io lock for VMODSORT to work 3303a71e32b6Sstans * correctly with pvn_vplist_dirty() and pvn_getdirty() when mod 3304a71e32b6Sstans * bit is cleared. 3305a71e32b6Sstans * We don't have assert to avoid tripping some existing third party 3306a71e32b6Sstans * code. The dirty page is moved back to top of the v_page list 3307a71e32b6Sstans * after IO is done in pvn_write_done(). 33087c478bd9Sstevel@tonic-gate */ 33097c478bd9Sstevel@tonic-gate PP_CLRRM(pp, flag); 33107c478bd9Sstevel@tonic-gate 3311a71e32b6Sstans if ((flag & P_MOD) != 0 && vp != NULL && IS_VMODSORT(vp)) { 33127c478bd9Sstevel@tonic-gate 33137c478bd9Sstevel@tonic-gate /* 33147c478bd9Sstevel@tonic-gate * VMODSORT works by removing write permissions and getting 33157c478bd9Sstevel@tonic-gate * a fault when a page is made dirty. At this point 33167c478bd9Sstevel@tonic-gate * we need to remove write permission from all mappings 33177c478bd9Sstevel@tonic-gate * to this page. 33187c478bd9Sstevel@tonic-gate */ 33197c478bd9Sstevel@tonic-gate hati_page_clrwrt(pp); 33207c478bd9Sstevel@tonic-gate } 33217c478bd9Sstevel@tonic-gate } 33227c478bd9Sstevel@tonic-gate 33237c478bd9Sstevel@tonic-gate /* 33247c478bd9Sstevel@tonic-gate * If flag is specified, returns 0 if attribute is disabled 332521584dbcSPavel Tatashin * and non zero if enabled. If flag specifes multiple attributes 332621584dbcSPavel Tatashin * then returns 0 if ALL attributes are disabled. This is an advisory 33277c478bd9Sstevel@tonic-gate * call. 33287c478bd9Sstevel@tonic-gate */ 33297c478bd9Sstevel@tonic-gate uint_t 33307c478bd9Sstevel@tonic-gate hat_page_getattr(struct page *pp, uint_t flag) 33317c478bd9Sstevel@tonic-gate { 33327c478bd9Sstevel@tonic-gate return (PP_GETRM(pp, flag)); 33337c478bd9Sstevel@tonic-gate } 33347c478bd9Sstevel@tonic-gate 33357c478bd9Sstevel@tonic-gate 33367c478bd9Sstevel@tonic-gate /* 33377c478bd9Sstevel@tonic-gate * common code used by hat_pageunload() and hment_steal() 33387c478bd9Sstevel@tonic-gate */ 33397c478bd9Sstevel@tonic-gate hment_t * 33407c478bd9Sstevel@tonic-gate hati_page_unmap(page_t *pp, htable_t *ht, uint_t entry) 33417c478bd9Sstevel@tonic-gate { 33427c478bd9Sstevel@tonic-gate x86pte_t old_pte; 33437c478bd9Sstevel@tonic-gate pfn_t pfn = pp->p_pagenum; 33447c478bd9Sstevel@tonic-gate hment_t *hm; 33457c478bd9Sstevel@tonic-gate 33467c478bd9Sstevel@tonic-gate /* 33477c478bd9Sstevel@tonic-gate * We need to acquire a hold on the htable in order to 33487c478bd9Sstevel@tonic-gate * do the invalidate. We know the htable must exist, since 33497c478bd9Sstevel@tonic-gate * unmap's don't release the htable until after removing any 33507c478bd9Sstevel@tonic-gate * hment. Having x86_hm_enter() keeps that from proceeding. 33517c478bd9Sstevel@tonic-gate */ 33527c478bd9Sstevel@tonic-gate htable_acquire(ht); 33537c478bd9Sstevel@tonic-gate 33547c478bd9Sstevel@tonic-gate /* 33557c478bd9Sstevel@tonic-gate * Invalidate the PTE and remove the hment. 33567c478bd9Sstevel@tonic-gate */ 3357a6a74e0eSMatthew Ahrens old_pte = x86pte_inval(ht, entry, 0, NULL, B_TRUE); 3358aa2ed9e5Sjosephb if (PTE2PFN(old_pte, ht->ht_level) != pfn) { 3359ae115bc7Smrj panic("x86pte_inval() failure found PTE = " FMT_PTE 3360aa2ed9e5Sjosephb " pfn being unmapped is %lx ht=0x%lx entry=0x%x", 3361aa2ed9e5Sjosephb old_pte, pfn, (uintptr_t)ht, entry); 3362aa2ed9e5Sjosephb } 33637c478bd9Sstevel@tonic-gate 33647c478bd9Sstevel@tonic-gate /* 33657c478bd9Sstevel@tonic-gate * Clean up all the htable information for this mapping 33667c478bd9Sstevel@tonic-gate */ 33677c478bd9Sstevel@tonic-gate ASSERT(ht->ht_valid_cnt > 0); 33687c478bd9Sstevel@tonic-gate HTABLE_DEC(ht->ht_valid_cnt); 33697c478bd9Sstevel@tonic-gate PGCNT_DEC(ht->ht_hat, ht->ht_level); 33707c478bd9Sstevel@tonic-gate 33717c478bd9Sstevel@tonic-gate /* 33727c478bd9Sstevel@tonic-gate * sync ref/mod bits to the page_t 33737c478bd9Sstevel@tonic-gate */ 3374ae115bc7Smrj if (PTE_GET(old_pte, PT_SOFTWARE) < PT_NOSYNC) 33757c478bd9Sstevel@tonic-gate hati_sync_pte_to_page(pp, old_pte, ht->ht_level); 33767c478bd9Sstevel@tonic-gate 33777c478bd9Sstevel@tonic-gate /* 33787c478bd9Sstevel@tonic-gate * Remove the mapping list entry for this page. 33797c478bd9Sstevel@tonic-gate */ 33807c478bd9Sstevel@tonic-gate hm = hment_remove(pp, ht, entry); 33817c478bd9Sstevel@tonic-gate 33827c478bd9Sstevel@tonic-gate /* 33837c478bd9Sstevel@tonic-gate * drop the mapping list lock so that we might free the 33847c478bd9Sstevel@tonic-gate * hment and htable. 33857c478bd9Sstevel@tonic-gate */ 33867c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 33877c478bd9Sstevel@tonic-gate htable_release(ht); 33887c478bd9Sstevel@tonic-gate return (hm); 33897c478bd9Sstevel@tonic-gate } 33907c478bd9Sstevel@tonic-gate 3391a5652762Spraks extern int vpm_enable; 33927c478bd9Sstevel@tonic-gate /* 33937c478bd9Sstevel@tonic-gate * Unload all translations to a page. If the page is a subpage of a large 33947c478bd9Sstevel@tonic-gate * page, the large page mappings are also removed. 33957c478bd9Sstevel@tonic-gate * 33967c478bd9Sstevel@tonic-gate * The forceflags are unused. 33977c478bd9Sstevel@tonic-gate */ 33987c478bd9Sstevel@tonic-gate 33997c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 34007c478bd9Sstevel@tonic-gate static int 34017c478bd9Sstevel@tonic-gate hati_pageunload(struct page *pp, uint_t pg_szcd, uint_t forceflag) 34027c478bd9Sstevel@tonic-gate { 34037c478bd9Sstevel@tonic-gate page_t *cur_pp = pp; 34047c478bd9Sstevel@tonic-gate hment_t *hm; 34057c478bd9Sstevel@tonic-gate hment_t *prev; 34067c478bd9Sstevel@tonic-gate htable_t *ht; 34077c478bd9Sstevel@tonic-gate uint_t entry; 34087c478bd9Sstevel@tonic-gate level_t level; 34097c478bd9Sstevel@tonic-gate 3410843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 34116c9930aeSJoe Bonasera 34126c9930aeSJoe Bonasera /* 34136c9930aeSJoe Bonasera * prevent recursion due to kmem_free() 34146c9930aeSJoe Bonasera */ 34156c9930aeSJoe Bonasera ++curthread->t_hatdepth; 34166c9930aeSJoe Bonasera ASSERT(curthread->t_hatdepth < 16); 34176c9930aeSJoe Bonasera 3418a5652762Spraks #if defined(__amd64) 3419a5652762Spraks /* 3420a5652762Spraks * clear the vpm ref. 3421a5652762Spraks */ 3422a5652762Spraks if (vpm_enable) { 3423a5652762Spraks pp->p_vpmref = 0; 3424a5652762Spraks } 3425a5652762Spraks #endif 34267c478bd9Sstevel@tonic-gate /* 34277c478bd9Sstevel@tonic-gate * The loop with next_size handles pages with multiple pagesize mappings 34287c478bd9Sstevel@tonic-gate */ 34297c478bd9Sstevel@tonic-gate next_size: 34307c478bd9Sstevel@tonic-gate for (;;) { 34317c478bd9Sstevel@tonic-gate 34327c478bd9Sstevel@tonic-gate /* 34337c478bd9Sstevel@tonic-gate * Get a mapping list entry 34347c478bd9Sstevel@tonic-gate */ 34357c478bd9Sstevel@tonic-gate x86_hm_enter(cur_pp); 34367c478bd9Sstevel@tonic-gate for (prev = NULL; ; prev = hm) { 34377c478bd9Sstevel@tonic-gate hm = hment_walk(cur_pp, &ht, &entry, prev); 34387c478bd9Sstevel@tonic-gate if (hm == NULL) { 34397c478bd9Sstevel@tonic-gate x86_hm_exit(cur_pp); 34407c478bd9Sstevel@tonic-gate 34417c478bd9Sstevel@tonic-gate /* 34427c478bd9Sstevel@tonic-gate * If not part of a larger page, we're done. 34437c478bd9Sstevel@tonic-gate */ 3444ae115bc7Smrj if (cur_pp->p_szc <= pg_szcd) { 34456c9930aeSJoe Bonasera ASSERT(curthread->t_hatdepth > 0); 34466c9930aeSJoe Bonasera --curthread->t_hatdepth; 3447843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 34487c478bd9Sstevel@tonic-gate return (0); 3449ae115bc7Smrj } 34507c478bd9Sstevel@tonic-gate 34517c478bd9Sstevel@tonic-gate /* 34527c478bd9Sstevel@tonic-gate * Else check the next larger page size. 34537c478bd9Sstevel@tonic-gate * hat_page_demote() may decrease p_szc 34547c478bd9Sstevel@tonic-gate * but that's ok we'll just take an extra 34557c478bd9Sstevel@tonic-gate * trip discover there're no larger mappings 34567c478bd9Sstevel@tonic-gate * and return. 34577c478bd9Sstevel@tonic-gate */ 34587c478bd9Sstevel@tonic-gate ++pg_szcd; 34597c478bd9Sstevel@tonic-gate cur_pp = PP_GROUPLEADER(cur_pp, pg_szcd); 34607c478bd9Sstevel@tonic-gate goto next_size; 34617c478bd9Sstevel@tonic-gate } 34627c478bd9Sstevel@tonic-gate 34637c478bd9Sstevel@tonic-gate /* 34647c478bd9Sstevel@tonic-gate * If this mapping size matches, remove it. 34657c478bd9Sstevel@tonic-gate */ 34667c478bd9Sstevel@tonic-gate level = ht->ht_level; 34677c478bd9Sstevel@tonic-gate if (level == pg_szcd) 34687c478bd9Sstevel@tonic-gate break; 34697c478bd9Sstevel@tonic-gate } 34707c478bd9Sstevel@tonic-gate 34717c478bd9Sstevel@tonic-gate /* 34727c478bd9Sstevel@tonic-gate * Remove the mapping list entry for this page. 34737c478bd9Sstevel@tonic-gate * Note this does the x86_hm_exit() for us. 34747c478bd9Sstevel@tonic-gate */ 34757c478bd9Sstevel@tonic-gate hm = hati_page_unmap(cur_pp, ht, entry); 34767c478bd9Sstevel@tonic-gate if (hm != NULL) 34777c478bd9Sstevel@tonic-gate hment_free(hm); 34787c478bd9Sstevel@tonic-gate } 34797c478bd9Sstevel@tonic-gate } 34807c478bd9Sstevel@tonic-gate 34817c478bd9Sstevel@tonic-gate int 34827c478bd9Sstevel@tonic-gate hat_pageunload(struct page *pp, uint_t forceflag) 34837c478bd9Sstevel@tonic-gate { 34847c478bd9Sstevel@tonic-gate ASSERT(PAGE_EXCL(pp)); 34857c478bd9Sstevel@tonic-gate return (hati_pageunload(pp, 0, forceflag)); 34867c478bd9Sstevel@tonic-gate } 34877c478bd9Sstevel@tonic-gate 34887c478bd9Sstevel@tonic-gate /* 34897c478bd9Sstevel@tonic-gate * Unload all large mappings to pp and reduce by 1 p_szc field of every large 34907c478bd9Sstevel@tonic-gate * page level that included pp. 34917c478bd9Sstevel@tonic-gate * 34927c478bd9Sstevel@tonic-gate * pp must be locked EXCL. Even though no other constituent pages are locked 34937c478bd9Sstevel@tonic-gate * it's legal to unload large mappings to pp because all constituent pages of 34947c478bd9Sstevel@tonic-gate * large locked mappings have to be locked SHARED. therefore if we have EXCL 34957c478bd9Sstevel@tonic-gate * lock on one of constituent pages none of the large mappings to pp are 34967c478bd9Sstevel@tonic-gate * locked. 34977c478bd9Sstevel@tonic-gate * 34987c478bd9Sstevel@tonic-gate * Change (always decrease) p_szc field starting from the last constituent 34997c478bd9Sstevel@tonic-gate * page and ending with root constituent page so that root's pszc always shows 35007c478bd9Sstevel@tonic-gate * the area where hat_page_demote() may be active. 35017c478bd9Sstevel@tonic-gate * 35027c478bd9Sstevel@tonic-gate * This mechanism is only used for file system pages where it's not always 35037c478bd9Sstevel@tonic-gate * possible to get EXCL locks on all constituent pages to demote the size code 35047c478bd9Sstevel@tonic-gate * (as is done for anonymous or kernel large pages). 35057c478bd9Sstevel@tonic-gate */ 35067c478bd9Sstevel@tonic-gate void 35077c478bd9Sstevel@tonic-gate hat_page_demote(page_t *pp) 35087c478bd9Sstevel@tonic-gate { 35097c478bd9Sstevel@tonic-gate uint_t pszc; 35107c478bd9Sstevel@tonic-gate uint_t rszc; 35117c478bd9Sstevel@tonic-gate uint_t szc; 35127c478bd9Sstevel@tonic-gate page_t *rootpp; 35137c478bd9Sstevel@tonic-gate page_t *firstpp; 35147c478bd9Sstevel@tonic-gate page_t *lastpp; 35157c478bd9Sstevel@tonic-gate pgcnt_t pgcnt; 35167c478bd9Sstevel@tonic-gate 35177c478bd9Sstevel@tonic-gate ASSERT(PAGE_EXCL(pp)); 35187c478bd9Sstevel@tonic-gate ASSERT(!PP_ISFREE(pp)); 35197c478bd9Sstevel@tonic-gate ASSERT(page_szc_lock_assert(pp)); 35207c478bd9Sstevel@tonic-gate 35217c478bd9Sstevel@tonic-gate if (pp->p_szc == 0) 35227c478bd9Sstevel@tonic-gate return; 35237c478bd9Sstevel@tonic-gate 35247c478bd9Sstevel@tonic-gate rootpp = PP_GROUPLEADER(pp, 1); 35257c478bd9Sstevel@tonic-gate (void) hati_pageunload(rootpp, 1, HAT_FORCE_PGUNLOAD); 35267c478bd9Sstevel@tonic-gate 35277c478bd9Sstevel@tonic-gate /* 35287c478bd9Sstevel@tonic-gate * all large mappings to pp are gone 35297c478bd9Sstevel@tonic-gate * and no new can be setup since pp is locked exclusively. 35307c478bd9Sstevel@tonic-gate * 35317c478bd9Sstevel@tonic-gate * Lock the root to make sure there's only one hat_page_demote() 35327c478bd9Sstevel@tonic-gate * outstanding within the area of this root's pszc. 35337c478bd9Sstevel@tonic-gate * 35347c478bd9Sstevel@tonic-gate * Second potential hat_page_demote() is already eliminated by upper 35357c478bd9Sstevel@tonic-gate * VM layer via page_szc_lock() but we don't rely on it and use our 35367c478bd9Sstevel@tonic-gate * own locking (so that upper layer locking can be changed without 35377c478bd9Sstevel@tonic-gate * assumptions that hat depends on upper layer VM to prevent multiple 35387c478bd9Sstevel@tonic-gate * hat_page_demote() to be issued simultaneously to the same large 35397c478bd9Sstevel@tonic-gate * page). 35407c478bd9Sstevel@tonic-gate */ 35417c478bd9Sstevel@tonic-gate again: 35427c478bd9Sstevel@tonic-gate pszc = pp->p_szc; 35437c478bd9Sstevel@tonic-gate if (pszc == 0) 35447c478bd9Sstevel@tonic-gate return; 35457c478bd9Sstevel@tonic-gate rootpp = PP_GROUPLEADER(pp, pszc); 35467c478bd9Sstevel@tonic-gate x86_hm_enter(rootpp); 35477c478bd9Sstevel@tonic-gate /* 35487c478bd9Sstevel@tonic-gate * If root's p_szc is different from pszc we raced with another 35497c478bd9Sstevel@tonic-gate * hat_page_demote(). Drop the lock and try to find the root again. 35507c478bd9Sstevel@tonic-gate * If root's p_szc is greater than pszc previous hat_page_demote() is 35517c478bd9Sstevel@tonic-gate * not done yet. Take and release mlist lock of root's root to wait 35527c478bd9Sstevel@tonic-gate * for previous hat_page_demote() to complete. 35537c478bd9Sstevel@tonic-gate */ 35547c478bd9Sstevel@tonic-gate if ((rszc = rootpp->p_szc) != pszc) { 35557c478bd9Sstevel@tonic-gate x86_hm_exit(rootpp); 35567c478bd9Sstevel@tonic-gate if (rszc > pszc) { 35577c478bd9Sstevel@tonic-gate /* p_szc of a locked non free page can't increase */ 35587c478bd9Sstevel@tonic-gate ASSERT(pp != rootpp); 35597c478bd9Sstevel@tonic-gate 35607c478bd9Sstevel@tonic-gate rootpp = PP_GROUPLEADER(rootpp, rszc); 35617c478bd9Sstevel@tonic-gate x86_hm_enter(rootpp); 35627c478bd9Sstevel@tonic-gate x86_hm_exit(rootpp); 35637c478bd9Sstevel@tonic-gate } 35647c478bd9Sstevel@tonic-gate goto again; 35657c478bd9Sstevel@tonic-gate } 35667c478bd9Sstevel@tonic-gate ASSERT(pp->p_szc == pszc); 35677c478bd9Sstevel@tonic-gate 35687c478bd9Sstevel@tonic-gate /* 35697c478bd9Sstevel@tonic-gate * Decrement by 1 p_szc of every constituent page of a region that 35707c478bd9Sstevel@tonic-gate * covered pp. For example if original szc is 3 it gets changed to 2 35717c478bd9Sstevel@tonic-gate * everywhere except in region 2 that covered pp. Region 2 that 35727c478bd9Sstevel@tonic-gate * covered pp gets demoted to 1 everywhere except in region 1 that 35737c478bd9Sstevel@tonic-gate * covered pp. The region 1 that covered pp is demoted to region 35747c478bd9Sstevel@tonic-gate * 0. It's done this way because from region 3 we removed level 3 35757c478bd9Sstevel@tonic-gate * mappings, from region 2 that covered pp we removed level 2 mappings 35767c478bd9Sstevel@tonic-gate * and from region 1 that covered pp we removed level 1 mappings. All 35777c478bd9Sstevel@tonic-gate * changes are done from from high pfn's to low pfn's so that roots 35787c478bd9Sstevel@tonic-gate * are changed last allowing one to know the largest region where 35797c478bd9Sstevel@tonic-gate * hat_page_demote() is stil active by only looking at the root page. 35807c478bd9Sstevel@tonic-gate * 35817c478bd9Sstevel@tonic-gate * This algorithm is implemented in 2 while loops. First loop changes 35827c478bd9Sstevel@tonic-gate * p_szc of pages to the right of pp's level 1 region and second 35837c478bd9Sstevel@tonic-gate * loop changes p_szc of pages of level 1 region that covers pp 35847c478bd9Sstevel@tonic-gate * and all pages to the left of level 1 region that covers pp. 35857c478bd9Sstevel@tonic-gate * In the first loop p_szc keeps dropping with every iteration 35867c478bd9Sstevel@tonic-gate * and in the second loop it keeps increasing with every iteration. 35877c478bd9Sstevel@tonic-gate * 35887c478bd9Sstevel@tonic-gate * First loop description: Demote pages to the right of pp outside of 35897c478bd9Sstevel@tonic-gate * level 1 region that covers pp. In every iteration of the while 35907c478bd9Sstevel@tonic-gate * loop below find the last page of szc region and the first page of 35917c478bd9Sstevel@tonic-gate * (szc - 1) region that is immediately to the right of (szc - 1) 35927c478bd9Sstevel@tonic-gate * region that covers pp. From last such page to first such page 35937c478bd9Sstevel@tonic-gate * change every page's szc to szc - 1. Decrement szc and continue 35947c478bd9Sstevel@tonic-gate * looping until szc is 1. If pp belongs to the last (szc - 1) region 35957c478bd9Sstevel@tonic-gate * of szc region skip to the next iteration. 35967c478bd9Sstevel@tonic-gate */ 35977c478bd9Sstevel@tonic-gate szc = pszc; 35987c478bd9Sstevel@tonic-gate while (szc > 1) { 35997c478bd9Sstevel@tonic-gate lastpp = PP_GROUPLEADER(pp, szc); 36007c478bd9Sstevel@tonic-gate pgcnt = page_get_pagecnt(szc); 36017c478bd9Sstevel@tonic-gate lastpp += pgcnt - 1; 36027c478bd9Sstevel@tonic-gate firstpp = PP_GROUPLEADER(pp, (szc - 1)); 36037c478bd9Sstevel@tonic-gate pgcnt = page_get_pagecnt(szc - 1); 36047c478bd9Sstevel@tonic-gate if (lastpp - firstpp < pgcnt) { 36057c478bd9Sstevel@tonic-gate szc--; 36067c478bd9Sstevel@tonic-gate continue; 36077c478bd9Sstevel@tonic-gate } 36087c478bd9Sstevel@tonic-gate firstpp += pgcnt; 36097c478bd9Sstevel@tonic-gate while (lastpp != firstpp) { 36107c478bd9Sstevel@tonic-gate ASSERT(lastpp->p_szc == pszc); 36117c478bd9Sstevel@tonic-gate lastpp->p_szc = szc - 1; 36127c478bd9Sstevel@tonic-gate lastpp--; 36137c478bd9Sstevel@tonic-gate } 36147c478bd9Sstevel@tonic-gate firstpp->p_szc = szc - 1; 36157c478bd9Sstevel@tonic-gate szc--; 36167c478bd9Sstevel@tonic-gate } 36177c478bd9Sstevel@tonic-gate 36187c478bd9Sstevel@tonic-gate /* 36197c478bd9Sstevel@tonic-gate * Second loop description: 36207c478bd9Sstevel@tonic-gate * First iteration changes p_szc to 0 of every 36217c478bd9Sstevel@tonic-gate * page of level 1 region that covers pp. 36227c478bd9Sstevel@tonic-gate * Subsequent iterations find last page of szc region 36237c478bd9Sstevel@tonic-gate * immediately to the left of szc region that covered pp 36247c478bd9Sstevel@tonic-gate * and first page of (szc + 1) region that covers pp. 36257c478bd9Sstevel@tonic-gate * From last to first page change p_szc of every page to szc. 36267c478bd9Sstevel@tonic-gate * Increment szc and continue looping until szc is pszc. 36277c478bd9Sstevel@tonic-gate * If pp belongs to the fist szc region of (szc + 1) region 36287c478bd9Sstevel@tonic-gate * skip to the next iteration. 36297c478bd9Sstevel@tonic-gate * 36307c478bd9Sstevel@tonic-gate */ 36317c478bd9Sstevel@tonic-gate szc = 0; 36327c478bd9Sstevel@tonic-gate while (szc < pszc) { 36337c478bd9Sstevel@tonic-gate firstpp = PP_GROUPLEADER(pp, (szc + 1)); 36347c478bd9Sstevel@tonic-gate if (szc == 0) { 36357c478bd9Sstevel@tonic-gate pgcnt = page_get_pagecnt(1); 36367c478bd9Sstevel@tonic-gate lastpp = firstpp + (pgcnt - 1); 36377c478bd9Sstevel@tonic-gate } else { 36387c478bd9Sstevel@tonic-gate lastpp = PP_GROUPLEADER(pp, szc); 36397c478bd9Sstevel@tonic-gate if (firstpp == lastpp) { 36407c478bd9Sstevel@tonic-gate szc++; 36417c478bd9Sstevel@tonic-gate continue; 36427c478bd9Sstevel@tonic-gate } 36437c478bd9Sstevel@tonic-gate lastpp--; 36447c478bd9Sstevel@tonic-gate pgcnt = page_get_pagecnt(szc); 36457c478bd9Sstevel@tonic-gate } 36467c478bd9Sstevel@tonic-gate while (lastpp != firstpp) { 36477c478bd9Sstevel@tonic-gate ASSERT(lastpp->p_szc == pszc); 36487c478bd9Sstevel@tonic-gate lastpp->p_szc = szc; 36497c478bd9Sstevel@tonic-gate lastpp--; 36507c478bd9Sstevel@tonic-gate } 36517c478bd9Sstevel@tonic-gate firstpp->p_szc = szc; 36527c478bd9Sstevel@tonic-gate if (firstpp == rootpp) 36537c478bd9Sstevel@tonic-gate break; 36547c478bd9Sstevel@tonic-gate szc++; 36557c478bd9Sstevel@tonic-gate } 36567c478bd9Sstevel@tonic-gate x86_hm_exit(rootpp); 36577c478bd9Sstevel@tonic-gate } 36587c478bd9Sstevel@tonic-gate 36597c478bd9Sstevel@tonic-gate /* 36607c478bd9Sstevel@tonic-gate * get hw stats from hardware into page struct and reset hw stats 36617c478bd9Sstevel@tonic-gate * returns attributes of page 36627c478bd9Sstevel@tonic-gate * Flags for hat_pagesync, hat_getstat, hat_sync 36637c478bd9Sstevel@tonic-gate * 36647c478bd9Sstevel@tonic-gate * define HAT_SYNC_ZERORM 0x01 36657c478bd9Sstevel@tonic-gate * 36667c478bd9Sstevel@tonic-gate * Additional flags for hat_pagesync 36677c478bd9Sstevel@tonic-gate * 36687c478bd9Sstevel@tonic-gate * define HAT_SYNC_STOPON_REF 0x02 36697c478bd9Sstevel@tonic-gate * define HAT_SYNC_STOPON_MOD 0x04 36707c478bd9Sstevel@tonic-gate * define HAT_SYNC_STOPON_RM 0x06 36717c478bd9Sstevel@tonic-gate * define HAT_SYNC_STOPON_SHARED 0x08 36727c478bd9Sstevel@tonic-gate */ 36737c478bd9Sstevel@tonic-gate uint_t 36747c478bd9Sstevel@tonic-gate hat_pagesync(struct page *pp, uint_t flags) 36757c478bd9Sstevel@tonic-gate { 36767c478bd9Sstevel@tonic-gate hment_t *hm = NULL; 36777c478bd9Sstevel@tonic-gate htable_t *ht; 36787c478bd9Sstevel@tonic-gate uint_t entry; 36797c478bd9Sstevel@tonic-gate x86pte_t old, save_old; 36807c478bd9Sstevel@tonic-gate x86pte_t new; 36817c478bd9Sstevel@tonic-gate uchar_t nrmbits = P_REF|P_MOD|P_RO; 36827c478bd9Sstevel@tonic-gate extern ulong_t po_share; 36837c478bd9Sstevel@tonic-gate page_t *save_pp = pp; 36847c478bd9Sstevel@tonic-gate uint_t pszc = 0; 36857c478bd9Sstevel@tonic-gate 36867c478bd9Sstevel@tonic-gate ASSERT(PAGE_LOCKED(pp) || panicstr); 36877c478bd9Sstevel@tonic-gate 36887c478bd9Sstevel@tonic-gate if (PP_ISRO(pp) && (flags & HAT_SYNC_STOPON_MOD)) 36897c478bd9Sstevel@tonic-gate return (pp->p_nrm & nrmbits); 36907c478bd9Sstevel@tonic-gate 36917c478bd9Sstevel@tonic-gate if ((flags & HAT_SYNC_ZERORM) == 0) { 36927c478bd9Sstevel@tonic-gate 36937c478bd9Sstevel@tonic-gate if ((flags & HAT_SYNC_STOPON_REF) != 0 && PP_ISREF(pp)) 36947c478bd9Sstevel@tonic-gate return (pp->p_nrm & nrmbits); 36957c478bd9Sstevel@tonic-gate 36967c478bd9Sstevel@tonic-gate if ((flags & HAT_SYNC_STOPON_MOD) != 0 && PP_ISMOD(pp)) 36977c478bd9Sstevel@tonic-gate return (pp->p_nrm & nrmbits); 36987c478bd9Sstevel@tonic-gate 36997c478bd9Sstevel@tonic-gate if ((flags & HAT_SYNC_STOPON_SHARED) != 0 && 37007c478bd9Sstevel@tonic-gate hat_page_getshare(pp) > po_share) { 37017c478bd9Sstevel@tonic-gate if (PP_ISRO(pp)) 37027c478bd9Sstevel@tonic-gate PP_SETREF(pp); 37037c478bd9Sstevel@tonic-gate return (pp->p_nrm & nrmbits); 37047c478bd9Sstevel@tonic-gate } 37057c478bd9Sstevel@tonic-gate } 37067c478bd9Sstevel@tonic-gate 3707843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 37087c478bd9Sstevel@tonic-gate next_size: 37097c478bd9Sstevel@tonic-gate /* 37107c478bd9Sstevel@tonic-gate * walk thru the mapping list syncing (and clearing) ref/mod bits. 37117c478bd9Sstevel@tonic-gate */ 37127c478bd9Sstevel@tonic-gate x86_hm_enter(pp); 37137c478bd9Sstevel@tonic-gate while ((hm = hment_walk(pp, &ht, &entry, hm)) != NULL) { 37147c478bd9Sstevel@tonic-gate if (ht->ht_level < pszc) 37157c478bd9Sstevel@tonic-gate continue; 37167c478bd9Sstevel@tonic-gate old = x86pte_get(ht, entry); 37177c478bd9Sstevel@tonic-gate try_again: 37187c478bd9Sstevel@tonic-gate 37197c478bd9Sstevel@tonic-gate ASSERT(PTE2PFN(old, ht->ht_level) == pp->p_pagenum); 37207c478bd9Sstevel@tonic-gate 37217c478bd9Sstevel@tonic-gate if (PTE_GET(old, PT_REF | PT_MOD) == 0) 37227c478bd9Sstevel@tonic-gate continue; 37237c478bd9Sstevel@tonic-gate 37247c478bd9Sstevel@tonic-gate save_old = old; 37257c478bd9Sstevel@tonic-gate if ((flags & HAT_SYNC_ZERORM) != 0) { 37267c478bd9Sstevel@tonic-gate 37277c478bd9Sstevel@tonic-gate /* 37287c478bd9Sstevel@tonic-gate * Need to clear ref or mod bits. Need to demap 37297c478bd9Sstevel@tonic-gate * to make sure any executing TLBs see cleared bits. 37307c478bd9Sstevel@tonic-gate */ 37317c478bd9Sstevel@tonic-gate new = old; 37327c478bd9Sstevel@tonic-gate PTE_CLR(new, PT_REF | PT_MOD); 37337c478bd9Sstevel@tonic-gate old = hati_update_pte(ht, entry, old, new); 37347c478bd9Sstevel@tonic-gate if (old != 0) 37357c478bd9Sstevel@tonic-gate goto try_again; 37367c478bd9Sstevel@tonic-gate 37377c478bd9Sstevel@tonic-gate old = save_old; 37387c478bd9Sstevel@tonic-gate } 37397c478bd9Sstevel@tonic-gate 37407c478bd9Sstevel@tonic-gate /* 37417c478bd9Sstevel@tonic-gate * Sync the PTE 37427c478bd9Sstevel@tonic-gate */ 3743ae115bc7Smrj if (!(flags & HAT_SYNC_ZERORM) && 3744ae115bc7Smrj PTE_GET(old, PT_SOFTWARE) <= PT_NOSYNC) 37457c478bd9Sstevel@tonic-gate hati_sync_pte_to_page(pp, old, ht->ht_level); 37467c478bd9Sstevel@tonic-gate 37477c478bd9Sstevel@tonic-gate /* 37487c478bd9Sstevel@tonic-gate * can stop short if we found a ref'd or mod'd page 37497c478bd9Sstevel@tonic-gate */ 37507c478bd9Sstevel@tonic-gate if ((flags & HAT_SYNC_STOPON_MOD) && PP_ISMOD(save_pp) || 37517c478bd9Sstevel@tonic-gate (flags & HAT_SYNC_STOPON_REF) && PP_ISREF(save_pp)) { 37527c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 3753ae115bc7Smrj goto done; 37547c478bd9Sstevel@tonic-gate } 37557c478bd9Sstevel@tonic-gate } 37567c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 37577c478bd9Sstevel@tonic-gate while (pszc < pp->p_szc) { 37587c478bd9Sstevel@tonic-gate page_t *tpp; 37597c478bd9Sstevel@tonic-gate pszc++; 37607c478bd9Sstevel@tonic-gate tpp = PP_GROUPLEADER(pp, pszc); 37617c478bd9Sstevel@tonic-gate if (pp != tpp) { 37627c478bd9Sstevel@tonic-gate pp = tpp; 37637c478bd9Sstevel@tonic-gate goto next_size; 37647c478bd9Sstevel@tonic-gate } 37657c478bd9Sstevel@tonic-gate } 3766ae115bc7Smrj done: 3767843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 37687c478bd9Sstevel@tonic-gate return (save_pp->p_nrm & nrmbits); 37697c478bd9Sstevel@tonic-gate } 37707c478bd9Sstevel@tonic-gate 37717c478bd9Sstevel@tonic-gate /* 37727c478bd9Sstevel@tonic-gate * returns approx number of mappings to this pp. A return of 0 implies 37737c478bd9Sstevel@tonic-gate * there are no mappings to the page. 37747c478bd9Sstevel@tonic-gate */ 37757c478bd9Sstevel@tonic-gate ulong_t 37767c478bd9Sstevel@tonic-gate hat_page_getshare(page_t *pp) 37777c478bd9Sstevel@tonic-gate { 37787c478bd9Sstevel@tonic-gate uint_t cnt; 37797c478bd9Sstevel@tonic-gate cnt = hment_mapcnt(pp); 3780a5652762Spraks #if defined(__amd64) 3781a5652762Spraks if (vpm_enable && pp->p_vpmref) { 3782a5652762Spraks cnt += 1; 3783a5652762Spraks } 3784a5652762Spraks #endif 37857c478bd9Sstevel@tonic-gate return (cnt); 37867c478bd9Sstevel@tonic-gate } 37877c478bd9Sstevel@tonic-gate 378805d3dc4bSpaulsan /* 378905d3dc4bSpaulsan * Return 1 the number of mappings exceeds sh_thresh. Return 0 379005d3dc4bSpaulsan * otherwise. 379105d3dc4bSpaulsan */ 379205d3dc4bSpaulsan int 379305d3dc4bSpaulsan hat_page_checkshare(page_t *pp, ulong_t sh_thresh) 379405d3dc4bSpaulsan { 379505d3dc4bSpaulsan return (hat_page_getshare(pp) > sh_thresh); 379605d3dc4bSpaulsan } 379705d3dc4bSpaulsan 37987c478bd9Sstevel@tonic-gate /* 37997c478bd9Sstevel@tonic-gate * hat_softlock isn't supported anymore 38007c478bd9Sstevel@tonic-gate */ 38017c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 38027c478bd9Sstevel@tonic-gate faultcode_t 38037c478bd9Sstevel@tonic-gate hat_softlock( 38047c478bd9Sstevel@tonic-gate hat_t *hat, 38057c478bd9Sstevel@tonic-gate caddr_t addr, 38067c478bd9Sstevel@tonic-gate size_t *len, 38077c478bd9Sstevel@tonic-gate struct page **page_array, 38087c478bd9Sstevel@tonic-gate uint_t flags) 38097c478bd9Sstevel@tonic-gate { 38107c478bd9Sstevel@tonic-gate return (FC_NOSUPPORT); 38117c478bd9Sstevel@tonic-gate } 38127c478bd9Sstevel@tonic-gate 38137c478bd9Sstevel@tonic-gate 38147c478bd9Sstevel@tonic-gate 38157c478bd9Sstevel@tonic-gate /* 38167c478bd9Sstevel@tonic-gate * Routine to expose supported HAT features to platform independent code. 38177c478bd9Sstevel@tonic-gate */ 38187c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 38197c478bd9Sstevel@tonic-gate int 38207c478bd9Sstevel@tonic-gate hat_supported(enum hat_features feature, void *arg) 38217c478bd9Sstevel@tonic-gate { 38227c478bd9Sstevel@tonic-gate switch (feature) { 38237c478bd9Sstevel@tonic-gate 38247c478bd9Sstevel@tonic-gate case HAT_SHARED_PT: /* this is really ISM */ 38257c478bd9Sstevel@tonic-gate return (1); 38267c478bd9Sstevel@tonic-gate 38277c478bd9Sstevel@tonic-gate case HAT_DYNAMIC_ISM_UNMAP: 38287c478bd9Sstevel@tonic-gate return (0); 38297c478bd9Sstevel@tonic-gate 38307c478bd9Sstevel@tonic-gate case HAT_VMODSORT: 38317c478bd9Sstevel@tonic-gate return (1); 38327c478bd9Sstevel@tonic-gate 383305d3dc4bSpaulsan case HAT_SHARED_REGIONS: 383405d3dc4bSpaulsan return (0); 383505d3dc4bSpaulsan 38367c478bd9Sstevel@tonic-gate default: 38377c478bd9Sstevel@tonic-gate panic("hat_supported() - unknown feature"); 38387c478bd9Sstevel@tonic-gate } 38397c478bd9Sstevel@tonic-gate return (0); 38407c478bd9Sstevel@tonic-gate } 38417c478bd9Sstevel@tonic-gate 38427c478bd9Sstevel@tonic-gate /* 38437c478bd9Sstevel@tonic-gate * Called when a thread is exiting and has been switched to the kernel AS 38447c478bd9Sstevel@tonic-gate */ 38457c478bd9Sstevel@tonic-gate void 38467c478bd9Sstevel@tonic-gate hat_thread_exit(kthread_t *thd) 38477c478bd9Sstevel@tonic-gate { 38487c478bd9Sstevel@tonic-gate ASSERT(thd->t_procp->p_as == &kas); 3849843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 38507c478bd9Sstevel@tonic-gate hat_switch(thd->t_procp->p_as->a_hat); 3851843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 38527c478bd9Sstevel@tonic-gate } 38537c478bd9Sstevel@tonic-gate 38547c478bd9Sstevel@tonic-gate /* 38557c478bd9Sstevel@tonic-gate * Setup the given brand new hat structure as the new HAT on this cpu's mmu. 38567c478bd9Sstevel@tonic-gate */ 38577c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 38587c478bd9Sstevel@tonic-gate void 38597c478bd9Sstevel@tonic-gate hat_setup(hat_t *hat, int flags) 38607c478bd9Sstevel@tonic-gate { 3861843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 38627c478bd9Sstevel@tonic-gate kpreempt_disable(); 38637c478bd9Sstevel@tonic-gate 38647c478bd9Sstevel@tonic-gate hat_switch(hat); 38657c478bd9Sstevel@tonic-gate 38667c478bd9Sstevel@tonic-gate kpreempt_enable(); 3867843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 38687c478bd9Sstevel@tonic-gate } 38697c478bd9Sstevel@tonic-gate 38707c478bd9Sstevel@tonic-gate /* 38717c478bd9Sstevel@tonic-gate * Prepare for a CPU private mapping for the given address. 38727c478bd9Sstevel@tonic-gate * 38737c478bd9Sstevel@tonic-gate * The address can only be used from a single CPU and can be remapped 38747c478bd9Sstevel@tonic-gate * using hat_mempte_remap(). Return the address of the PTE. 38757c478bd9Sstevel@tonic-gate * 38767c478bd9Sstevel@tonic-gate * We do the htable_create() if necessary and increment the valid count so 38777c478bd9Sstevel@tonic-gate * the htable can't disappear. We also hat_devload() the page table into 38787c478bd9Sstevel@tonic-gate * kernel so that the PTE is quickly accessed. 38797c478bd9Sstevel@tonic-gate */ 3880ae115bc7Smrj hat_mempte_t 3881ae115bc7Smrj hat_mempte_setup(caddr_t addr) 38827c478bd9Sstevel@tonic-gate { 38837c478bd9Sstevel@tonic-gate uintptr_t va = (uintptr_t)addr; 38847c478bd9Sstevel@tonic-gate htable_t *ht; 38857c478bd9Sstevel@tonic-gate uint_t entry; 38867c478bd9Sstevel@tonic-gate x86pte_t oldpte; 3887ae115bc7Smrj hat_mempte_t p; 38887c478bd9Sstevel@tonic-gate 38897c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(va)); 38907c478bd9Sstevel@tonic-gate ASSERT(!IN_VA_HOLE(va)); 3891aac11643Sjosephb ++curthread->t_hatdepth; 3892551bc2a6Smrj XPV_DISALLOW_MIGRATE(); 38937c478bd9Sstevel@tonic-gate ht = htable_getpte(kas.a_hat, va, &entry, &oldpte, 0); 38947c478bd9Sstevel@tonic-gate if (ht == NULL) { 38957c478bd9Sstevel@tonic-gate ht = htable_create(kas.a_hat, va, 0, NULL); 38967c478bd9Sstevel@tonic-gate entry = htable_va2entry(va, ht); 38977c478bd9Sstevel@tonic-gate ASSERT(ht->ht_level == 0); 38987c478bd9Sstevel@tonic-gate oldpte = x86pte_get(ht, entry); 38997c478bd9Sstevel@tonic-gate } 39007c478bd9Sstevel@tonic-gate if (PTE_ISVALID(oldpte)) 39017c478bd9Sstevel@tonic-gate panic("hat_mempte_setup(): address already mapped" 3902903a11ebSrh "ht=%p, entry=%d, pte=" FMT_PTE, (void *)ht, entry, oldpte); 39037c478bd9Sstevel@tonic-gate 39047c478bd9Sstevel@tonic-gate /* 39057c478bd9Sstevel@tonic-gate * increment ht_valid_cnt so that the pagetable can't disappear 39067c478bd9Sstevel@tonic-gate */ 39077c478bd9Sstevel@tonic-gate HTABLE_INC(ht->ht_valid_cnt); 39087c478bd9Sstevel@tonic-gate 39097c478bd9Sstevel@tonic-gate /* 3910ae115bc7Smrj * return the PTE physical address to the caller. 39117c478bd9Sstevel@tonic-gate */ 39127c478bd9Sstevel@tonic-gate htable_release(ht); 3913551bc2a6Smrj XPV_ALLOW_MIGRATE(); 3914ae115bc7Smrj p = PT_INDEX_PHYSADDR(pfn_to_pa(ht->ht_pfn), entry); 3915aac11643Sjosephb --curthread->t_hatdepth; 3916ae115bc7Smrj return (p); 39177c478bd9Sstevel@tonic-gate } 39187c478bd9Sstevel@tonic-gate 39197c478bd9Sstevel@tonic-gate /* 39207c478bd9Sstevel@tonic-gate * Release a CPU private mapping for the given address. 39217c478bd9Sstevel@tonic-gate * We decrement the htable valid count so it might be destroyed. 39227c478bd9Sstevel@tonic-gate */ 3923ae115bc7Smrj /*ARGSUSED1*/ 39247c478bd9Sstevel@tonic-gate void 3925ae115bc7Smrj hat_mempte_release(caddr_t addr, hat_mempte_t pte_pa) 39267c478bd9Sstevel@tonic-gate { 39277c478bd9Sstevel@tonic-gate htable_t *ht; 39287c478bd9Sstevel@tonic-gate 3929551bc2a6Smrj XPV_DISALLOW_MIGRATE(); 39307c478bd9Sstevel@tonic-gate /* 3931ae115bc7Smrj * invalidate any left over mapping and decrement the htable valid count 39327c478bd9Sstevel@tonic-gate */ 3933843e1988Sjohnlev #ifdef __xpv 3934843e1988Sjohnlev if (HYPERVISOR_update_va_mapping((uintptr_t)addr, 0, 3935843e1988Sjohnlev UVMF_INVLPG | UVMF_LOCAL)) 3936843e1988Sjohnlev panic("HYPERVISOR_update_va_mapping() failed"); 3937843e1988Sjohnlev #else 3938ae115bc7Smrj { 3939ae115bc7Smrj x86pte_t *pteptr; 3940ae115bc7Smrj 3941ae115bc7Smrj pteptr = x86pte_mapin(mmu_btop(pte_pa), 3942ae115bc7Smrj (pte_pa & MMU_PAGEOFFSET) >> mmu.pte_size_shift, NULL); 3943ae115bc7Smrj if (mmu.pae_hat) 3944ae115bc7Smrj *pteptr = 0; 3945ae115bc7Smrj else 3946ae115bc7Smrj *(x86pte32_t *)pteptr = 0; 3947ae115bc7Smrj mmu_tlbflush_entry(addr); 3948ae115bc7Smrj x86pte_mapout(); 3949ae115bc7Smrj } 3950843e1988Sjohnlev #endif 3951ae115bc7Smrj 39527c478bd9Sstevel@tonic-gate ht = htable_getpte(kas.a_hat, ALIGN2PAGE(addr), NULL, NULL, 0); 39537c478bd9Sstevel@tonic-gate if (ht == NULL) 39547c478bd9Sstevel@tonic-gate panic("hat_mempte_release(): invalid address"); 39557c478bd9Sstevel@tonic-gate ASSERT(ht->ht_level == 0); 39567c478bd9Sstevel@tonic-gate HTABLE_DEC(ht->ht_valid_cnt); 39577c478bd9Sstevel@tonic-gate htable_release(ht); 3958551bc2a6Smrj XPV_ALLOW_MIGRATE(); 39597c478bd9Sstevel@tonic-gate } 39607c478bd9Sstevel@tonic-gate 39617c478bd9Sstevel@tonic-gate /* 39627c478bd9Sstevel@tonic-gate * Apply a temporary CPU private mapping to a page. We flush the TLB only 39637c478bd9Sstevel@tonic-gate * on this CPU, so this ought to have been called with preemption disabled. 39647c478bd9Sstevel@tonic-gate */ 39657c478bd9Sstevel@tonic-gate void 39667c478bd9Sstevel@tonic-gate hat_mempte_remap( 3967ae115bc7Smrj pfn_t pfn, 3968ae115bc7Smrj caddr_t addr, 3969ae115bc7Smrj hat_mempte_t pte_pa, 3970ae115bc7Smrj uint_t attr, 3971ae115bc7Smrj uint_t flags) 39727c478bd9Sstevel@tonic-gate { 39737c478bd9Sstevel@tonic-gate uintptr_t va = (uintptr_t)addr; 39747c478bd9Sstevel@tonic-gate x86pte_t pte; 39757c478bd9Sstevel@tonic-gate 39767c478bd9Sstevel@tonic-gate /* 39777c478bd9Sstevel@tonic-gate * Remap the given PTE to the new page's PFN. Invalidate only 39787c478bd9Sstevel@tonic-gate * on this CPU. 39797c478bd9Sstevel@tonic-gate */ 39807c478bd9Sstevel@tonic-gate #ifdef DEBUG 39817c478bd9Sstevel@tonic-gate htable_t *ht; 39827c478bd9Sstevel@tonic-gate uint_t entry; 39837c478bd9Sstevel@tonic-gate 39847c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(va)); 39857c478bd9Sstevel@tonic-gate ASSERT(!IN_VA_HOLE(va)); 39867c478bd9Sstevel@tonic-gate ht = htable_getpte(kas.a_hat, va, &entry, NULL, 0); 39877c478bd9Sstevel@tonic-gate ASSERT(ht != NULL); 39887c478bd9Sstevel@tonic-gate ASSERT(ht->ht_level == 0); 39897c478bd9Sstevel@tonic-gate ASSERT(ht->ht_valid_cnt > 0); 3990ae115bc7Smrj ASSERT(ht->ht_pfn == mmu_btop(pte_pa)); 39917c478bd9Sstevel@tonic-gate htable_release(ht); 39927c478bd9Sstevel@tonic-gate #endif 3993843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 39947c478bd9Sstevel@tonic-gate pte = hati_mkpte(pfn, attr, 0, flags); 3995843e1988Sjohnlev #ifdef __xpv 3996843e1988Sjohnlev if (HYPERVISOR_update_va_mapping(va, pte, UVMF_INVLPG | UVMF_LOCAL)) 3997843e1988Sjohnlev panic("HYPERVISOR_update_va_mapping() failed"); 3998843e1988Sjohnlev #else 3999ae115bc7Smrj { 4000ae115bc7Smrj x86pte_t *pteptr; 4001ae115bc7Smrj 4002ae115bc7Smrj pteptr = x86pte_mapin(mmu_btop(pte_pa), 4003ae115bc7Smrj (pte_pa & MMU_PAGEOFFSET) >> mmu.pte_size_shift, NULL); 4004ae115bc7Smrj if (mmu.pae_hat) 4005ae115bc7Smrj *(x86pte_t *)pteptr = pte; 4006ae115bc7Smrj else 4007ae115bc7Smrj *(x86pte32_t *)pteptr = (x86pte32_t)pte; 4008ae115bc7Smrj mmu_tlbflush_entry(addr); 4009ae115bc7Smrj x86pte_mapout(); 4010ae115bc7Smrj } 4011843e1988Sjohnlev #endif 4012843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 40137c478bd9Sstevel@tonic-gate } 40147c478bd9Sstevel@tonic-gate 40157c478bd9Sstevel@tonic-gate 40167c478bd9Sstevel@tonic-gate 40177c478bd9Sstevel@tonic-gate /* 40187c478bd9Sstevel@tonic-gate * Hat locking functions 40197c478bd9Sstevel@tonic-gate * XXX - these two functions are currently being used by hatstats 40207c478bd9Sstevel@tonic-gate * they can be removed by using a per-as mutex for hatstats. 40217c478bd9Sstevel@tonic-gate */ 40227c478bd9Sstevel@tonic-gate void 40237c478bd9Sstevel@tonic-gate hat_enter(hat_t *hat) 40247c478bd9Sstevel@tonic-gate { 40257c478bd9Sstevel@tonic-gate mutex_enter(&hat->hat_mutex); 40267c478bd9Sstevel@tonic-gate } 40277c478bd9Sstevel@tonic-gate 40287c478bd9Sstevel@tonic-gate void 40297c478bd9Sstevel@tonic-gate hat_exit(hat_t *hat) 40307c478bd9Sstevel@tonic-gate { 40317c478bd9Sstevel@tonic-gate mutex_exit(&hat->hat_mutex); 40327c478bd9Sstevel@tonic-gate } 40337c478bd9Sstevel@tonic-gate 40347c478bd9Sstevel@tonic-gate /* 4035ae115bc7Smrj * HAT part of cpu initialization. 40367c478bd9Sstevel@tonic-gate */ 40377c478bd9Sstevel@tonic-gate void 4038ae115bc7Smrj hat_cpu_online(struct cpu *cpup) 40397c478bd9Sstevel@tonic-gate { 4040ae115bc7Smrj if (cpup != CPU) { 4041ae115bc7Smrj x86pte_cpu_init(cpup); 4042ae115bc7Smrj hat_vlp_setup(cpup); 40437c478bd9Sstevel@tonic-gate } 4044ae115bc7Smrj CPUSET_ATOMIC_ADD(khat_cpuset, cpup->cpu_id); 40457c478bd9Sstevel@tonic-gate } 40467c478bd9Sstevel@tonic-gate 40477c478bd9Sstevel@tonic-gate /* 4048ae115bc7Smrj * HAT part of cpu deletion. 4049ae115bc7Smrj * (currently, we only call this after the cpu is safely passivated.) 40507c478bd9Sstevel@tonic-gate */ 40517c478bd9Sstevel@tonic-gate void 4052ae115bc7Smrj hat_cpu_offline(struct cpu *cpup) 40537c478bd9Sstevel@tonic-gate { 4054ae115bc7Smrj ASSERT(cpup != CPU); 4055ae115bc7Smrj 4056ae115bc7Smrj CPUSET_ATOMIC_DEL(khat_cpuset, cpup->cpu_id); 4057ae115bc7Smrj hat_vlp_teardown(cpup); 4058a3114836SGerry Liu x86pte_cpu_fini(cpup); 40597c478bd9Sstevel@tonic-gate } 40607c478bd9Sstevel@tonic-gate 40617c478bd9Sstevel@tonic-gate /* 40627c478bd9Sstevel@tonic-gate * Function called after all CPUs are brought online. 40637c478bd9Sstevel@tonic-gate * Used to remove low address boot mappings. 40647c478bd9Sstevel@tonic-gate */ 40657c478bd9Sstevel@tonic-gate void 40667c478bd9Sstevel@tonic-gate clear_boot_mappings(uintptr_t low, uintptr_t high) 40677c478bd9Sstevel@tonic-gate { 40687c478bd9Sstevel@tonic-gate uintptr_t vaddr = low; 40697c478bd9Sstevel@tonic-gate htable_t *ht = NULL; 40707c478bd9Sstevel@tonic-gate level_t level; 40717c478bd9Sstevel@tonic-gate uint_t entry; 40727c478bd9Sstevel@tonic-gate x86pte_t pte; 40737c478bd9Sstevel@tonic-gate 40747c478bd9Sstevel@tonic-gate /* 40757c478bd9Sstevel@tonic-gate * On 1st CPU we can unload the prom mappings, basically we blow away 4076ae115bc7Smrj * all virtual mappings under _userlimit. 40777c478bd9Sstevel@tonic-gate */ 40787c478bd9Sstevel@tonic-gate while (vaddr < high) { 40797c478bd9Sstevel@tonic-gate pte = htable_walk(kas.a_hat, &ht, &vaddr, high); 40807c478bd9Sstevel@tonic-gate if (ht == NULL) 40817c478bd9Sstevel@tonic-gate break; 40827c478bd9Sstevel@tonic-gate 40837c478bd9Sstevel@tonic-gate level = ht->ht_level; 40847c478bd9Sstevel@tonic-gate entry = htable_va2entry(vaddr, ht); 40857c478bd9Sstevel@tonic-gate ASSERT(level <= mmu.max_page_level); 40867c478bd9Sstevel@tonic-gate ASSERT(PTE_ISPAGE(pte, level)); 40877c478bd9Sstevel@tonic-gate 40887c478bd9Sstevel@tonic-gate /* 40897c478bd9Sstevel@tonic-gate * Unload the mapping from the page tables. 40907c478bd9Sstevel@tonic-gate */ 4091a6a74e0eSMatthew Ahrens (void) x86pte_inval(ht, entry, 0, NULL, B_TRUE); 40927c478bd9Sstevel@tonic-gate ASSERT(ht->ht_valid_cnt > 0); 40937c478bd9Sstevel@tonic-gate HTABLE_DEC(ht->ht_valid_cnt); 40947c478bd9Sstevel@tonic-gate PGCNT_DEC(ht->ht_hat, ht->ht_level); 40957c478bd9Sstevel@tonic-gate 40967c478bd9Sstevel@tonic-gate vaddr += LEVEL_SIZE(ht->ht_level); 40977c478bd9Sstevel@tonic-gate } 40987c478bd9Sstevel@tonic-gate if (ht) 40997c478bd9Sstevel@tonic-gate htable_release(ht); 41007c478bd9Sstevel@tonic-gate } 41017c478bd9Sstevel@tonic-gate 41027c478bd9Sstevel@tonic-gate /* 41037c478bd9Sstevel@tonic-gate * Atomically update a new translation for a single page. If the 41047c478bd9Sstevel@tonic-gate * currently installed PTE doesn't match the value we expect to find, 41057c478bd9Sstevel@tonic-gate * it's not updated and we return the PTE we found. 41067c478bd9Sstevel@tonic-gate * 41077c478bd9Sstevel@tonic-gate * If activating nosync or NOWRITE and the page was modified we need to sync 41087c478bd9Sstevel@tonic-gate * with the page_t. Also sync with page_t if clearing ref/mod bits. 41097c478bd9Sstevel@tonic-gate */ 41107c478bd9Sstevel@tonic-gate static x86pte_t 41117c478bd9Sstevel@tonic-gate hati_update_pte(htable_t *ht, uint_t entry, x86pte_t expected, x86pte_t new) 41127c478bd9Sstevel@tonic-gate { 41137c478bd9Sstevel@tonic-gate page_t *pp; 41147c478bd9Sstevel@tonic-gate uint_t rm = 0; 41157c478bd9Sstevel@tonic-gate x86pte_t replaced; 41167c478bd9Sstevel@tonic-gate 4117ae115bc7Smrj if (PTE_GET(expected, PT_SOFTWARE) < PT_NOSYNC && 41187c478bd9Sstevel@tonic-gate PTE_GET(expected, PT_MOD | PT_REF) && 41197c478bd9Sstevel@tonic-gate (PTE_GET(new, PT_NOSYNC) || !PTE_GET(new, PT_WRITABLE) || 4120250b7ff9Sjosephb !PTE_GET(new, PT_MOD | PT_REF))) { 41217c478bd9Sstevel@tonic-gate 4122ae115bc7Smrj ASSERT(!pfn_is_foreign(PTE2PFN(expected, ht->ht_level))); 41237c478bd9Sstevel@tonic-gate pp = page_numtopp_nolock(PTE2PFN(expected, ht->ht_level)); 41247c478bd9Sstevel@tonic-gate ASSERT(pp != NULL); 41257c478bd9Sstevel@tonic-gate if (PTE_GET(expected, PT_MOD)) 41267c478bd9Sstevel@tonic-gate rm |= P_MOD; 41277c478bd9Sstevel@tonic-gate if (PTE_GET(expected, PT_REF)) 41287c478bd9Sstevel@tonic-gate rm |= P_REF; 41297c478bd9Sstevel@tonic-gate PTE_CLR(new, PT_MOD | PT_REF); 41307c478bd9Sstevel@tonic-gate } 41317c478bd9Sstevel@tonic-gate 41327c478bd9Sstevel@tonic-gate replaced = x86pte_update(ht, entry, expected, new); 41337c478bd9Sstevel@tonic-gate if (replaced != expected) 41347c478bd9Sstevel@tonic-gate return (replaced); 41357c478bd9Sstevel@tonic-gate 41367c478bd9Sstevel@tonic-gate if (rm) { 41377c478bd9Sstevel@tonic-gate /* 41387c478bd9Sstevel@tonic-gate * sync to all constituent pages of a large page 41397c478bd9Sstevel@tonic-gate */ 41407c478bd9Sstevel@tonic-gate pgcnt_t pgcnt = page_get_pagecnt(ht->ht_level); 41417c478bd9Sstevel@tonic-gate ASSERT(IS_P2ALIGNED(pp->p_pagenum, pgcnt)); 41427c478bd9Sstevel@tonic-gate while (pgcnt-- > 0) { 41437c478bd9Sstevel@tonic-gate /* 41447c478bd9Sstevel@tonic-gate * hat_page_demote() can't decrease 41457c478bd9Sstevel@tonic-gate * pszc below this mapping size 41467c478bd9Sstevel@tonic-gate * since large mapping existed after we 41477c478bd9Sstevel@tonic-gate * took mlist lock. 41487c478bd9Sstevel@tonic-gate */ 41497c478bd9Sstevel@tonic-gate ASSERT(pp->p_szc >= ht->ht_level); 41507c478bd9Sstevel@tonic-gate hat_page_setattr(pp, rm); 41517c478bd9Sstevel@tonic-gate ++pp; 41527c478bd9Sstevel@tonic-gate } 41537c478bd9Sstevel@tonic-gate } 41547c478bd9Sstevel@tonic-gate 41557c478bd9Sstevel@tonic-gate return (0); 41567c478bd9Sstevel@tonic-gate } 41577c478bd9Sstevel@tonic-gate 415805d3dc4bSpaulsan /* ARGSUSED */ 415905d3dc4bSpaulsan void 41607dacfc44Spaulsan hat_join_srd(struct hat *hat, vnode_t *evp) 416105d3dc4bSpaulsan { 416205d3dc4bSpaulsan } 416305d3dc4bSpaulsan 416405d3dc4bSpaulsan /* ARGSUSED */ 416505d3dc4bSpaulsan hat_region_cookie_t 41667dacfc44Spaulsan hat_join_region(struct hat *hat, 416705d3dc4bSpaulsan caddr_t r_saddr, 416805d3dc4bSpaulsan size_t r_size, 416905d3dc4bSpaulsan void *r_obj, 417005d3dc4bSpaulsan u_offset_t r_objoff, 417105d3dc4bSpaulsan uchar_t r_perm, 417205d3dc4bSpaulsan uchar_t r_pgszc, 417305d3dc4bSpaulsan hat_rgn_cb_func_t r_cb_function, 417405d3dc4bSpaulsan uint_t flags) 417505d3dc4bSpaulsan { 417605d3dc4bSpaulsan panic("No shared region support on x86"); 417705d3dc4bSpaulsan return (HAT_INVALID_REGION_COOKIE); 417805d3dc4bSpaulsan } 417905d3dc4bSpaulsan 418005d3dc4bSpaulsan /* ARGSUSED */ 418105d3dc4bSpaulsan void 41827dacfc44Spaulsan hat_leave_region(struct hat *hat, hat_region_cookie_t rcookie, uint_t flags) 418305d3dc4bSpaulsan { 418405d3dc4bSpaulsan panic("No shared region support on x86"); 418505d3dc4bSpaulsan } 418605d3dc4bSpaulsan 418705d3dc4bSpaulsan /* ARGSUSED */ 418805d3dc4bSpaulsan void 41897dacfc44Spaulsan hat_dup_region(struct hat *hat, hat_region_cookie_t rcookie) 419005d3dc4bSpaulsan { 419105d3dc4bSpaulsan panic("No shared region support on x86"); 419205d3dc4bSpaulsan } 419305d3dc4bSpaulsan 419405d3dc4bSpaulsan 41957c478bd9Sstevel@tonic-gate /* 41967c478bd9Sstevel@tonic-gate * Kernel Physical Mapping (kpm) facility 41977c478bd9Sstevel@tonic-gate * 41987c478bd9Sstevel@tonic-gate * Most of the routines needed to support segkpm are almost no-ops on the 41997c478bd9Sstevel@tonic-gate * x86 platform. We map in the entire segment when it is created and leave 42007c478bd9Sstevel@tonic-gate * it mapped in, so there is no additional work required to set up and tear 42017c478bd9Sstevel@tonic-gate * down individual mappings. All of these routines were created to support 42027c478bd9Sstevel@tonic-gate * SPARC platforms that have to avoid aliasing in their virtually indexed 42037c478bd9Sstevel@tonic-gate * caches. 42047c478bd9Sstevel@tonic-gate * 42057c478bd9Sstevel@tonic-gate * Most of the routines have sanity checks in them (e.g. verifying that the 42067c478bd9Sstevel@tonic-gate * passed-in page is locked). We don't actually care about most of these 42077c478bd9Sstevel@tonic-gate * checks on x86, but we leave them in place to identify problems in the 42087c478bd9Sstevel@tonic-gate * upper levels. 42097c478bd9Sstevel@tonic-gate */ 42107c478bd9Sstevel@tonic-gate 42117c478bd9Sstevel@tonic-gate /* 42127c478bd9Sstevel@tonic-gate * Map in a locked page and return the vaddr. 42137c478bd9Sstevel@tonic-gate */ 42147c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 42157c478bd9Sstevel@tonic-gate caddr_t 42167c478bd9Sstevel@tonic-gate hat_kpm_mapin(struct page *pp, struct kpme *kpme) 42177c478bd9Sstevel@tonic-gate { 42187c478bd9Sstevel@tonic-gate caddr_t vaddr; 42197c478bd9Sstevel@tonic-gate 42207c478bd9Sstevel@tonic-gate #ifdef DEBUG 42217c478bd9Sstevel@tonic-gate if (kpm_enable == 0) { 42227c478bd9Sstevel@tonic-gate cmn_err(CE_WARN, "hat_kpm_mapin: kpm_enable not set\n"); 42237c478bd9Sstevel@tonic-gate return ((caddr_t)NULL); 42247c478bd9Sstevel@tonic-gate } 42257c478bd9Sstevel@tonic-gate 42267c478bd9Sstevel@tonic-gate if (pp == NULL || PAGE_LOCKED(pp) == 0) { 42277c478bd9Sstevel@tonic-gate cmn_err(CE_WARN, "hat_kpm_mapin: pp zero or not locked\n"); 42287c478bd9Sstevel@tonic-gate return ((caddr_t)NULL); 42297c478bd9Sstevel@tonic-gate } 42307c478bd9Sstevel@tonic-gate #endif 42317c478bd9Sstevel@tonic-gate 42327c478bd9Sstevel@tonic-gate vaddr = hat_kpm_page2va(pp, 1); 42337c478bd9Sstevel@tonic-gate 42347c478bd9Sstevel@tonic-gate return (vaddr); 42357c478bd9Sstevel@tonic-gate } 42367c478bd9Sstevel@tonic-gate 42377c478bd9Sstevel@tonic-gate /* 42387c478bd9Sstevel@tonic-gate * Mapout a locked page. 42397c478bd9Sstevel@tonic-gate */ 42407c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 42417c478bd9Sstevel@tonic-gate void 42427c478bd9Sstevel@tonic-gate hat_kpm_mapout(struct page *pp, struct kpme *kpme, caddr_t vaddr) 42437c478bd9Sstevel@tonic-gate { 42447c478bd9Sstevel@tonic-gate #ifdef DEBUG 42457c478bd9Sstevel@tonic-gate if (kpm_enable == 0) { 42467c478bd9Sstevel@tonic-gate cmn_err(CE_WARN, "hat_kpm_mapout: kpm_enable not set\n"); 42477c478bd9Sstevel@tonic-gate return; 42487c478bd9Sstevel@tonic-gate } 42497c478bd9Sstevel@tonic-gate 42507c478bd9Sstevel@tonic-gate if (IS_KPM_ADDR(vaddr) == 0) { 42517c478bd9Sstevel@tonic-gate cmn_err(CE_WARN, "hat_kpm_mapout: no kpm address\n"); 42527c478bd9Sstevel@tonic-gate return; 42537c478bd9Sstevel@tonic-gate } 42547c478bd9Sstevel@tonic-gate 42557c478bd9Sstevel@tonic-gate if (pp == NULL || PAGE_LOCKED(pp) == 0) { 42567c478bd9Sstevel@tonic-gate cmn_err(CE_WARN, "hat_kpm_mapout: page zero or not locked\n"); 42577c478bd9Sstevel@tonic-gate return; 42587c478bd9Sstevel@tonic-gate } 42597c478bd9Sstevel@tonic-gate #endif 42607c478bd9Sstevel@tonic-gate } 42617c478bd9Sstevel@tonic-gate 4262d20abfaaSPavel Tatashin /* 4263d20abfaaSPavel Tatashin * hat_kpm_mapin_pfn is used to obtain a kpm mapping for physical 4264d20abfaaSPavel Tatashin * memory addresses that are not described by a page_t. It can 4265d20abfaaSPavel Tatashin * also be used for normal pages that are not locked, but beware 4266d20abfaaSPavel Tatashin * this is dangerous - no locking is performed, so the identity of 4267d20abfaaSPavel Tatashin * the page could change. hat_kpm_mapin_pfn is not supported when 4268d20abfaaSPavel Tatashin * vac_colors > 1, because the chosen va depends on the page identity, 4269d20abfaaSPavel Tatashin * which could change. 4270d20abfaaSPavel Tatashin * The caller must only pass pfn's for valid physical addresses; violation 4271d20abfaaSPavel Tatashin * of this rule will cause panic. 4272d20abfaaSPavel Tatashin */ 4273d20abfaaSPavel Tatashin caddr_t 4274d20abfaaSPavel Tatashin hat_kpm_mapin_pfn(pfn_t pfn) 4275d20abfaaSPavel Tatashin { 4276d20abfaaSPavel Tatashin caddr_t paddr, vaddr; 4277d20abfaaSPavel Tatashin 4278d20abfaaSPavel Tatashin if (kpm_enable == 0) 4279d20abfaaSPavel Tatashin return ((caddr_t)NULL); 4280d20abfaaSPavel Tatashin 4281d20abfaaSPavel Tatashin paddr = (caddr_t)ptob(pfn); 4282d20abfaaSPavel Tatashin vaddr = (uintptr_t)kpm_vbase + paddr; 4283d20abfaaSPavel Tatashin 4284d20abfaaSPavel Tatashin return ((caddr_t)vaddr); 4285d20abfaaSPavel Tatashin } 4286d20abfaaSPavel Tatashin 4287d20abfaaSPavel Tatashin /*ARGSUSED*/ 4288d20abfaaSPavel Tatashin void 4289d20abfaaSPavel Tatashin hat_kpm_mapout_pfn(pfn_t pfn) 4290d20abfaaSPavel Tatashin { 4291d20abfaaSPavel Tatashin /* empty */ 4292d20abfaaSPavel Tatashin } 4293d20abfaaSPavel Tatashin 42947c478bd9Sstevel@tonic-gate /* 42957c478bd9Sstevel@tonic-gate * Return the kpm virtual address for a specific pfn 42967c478bd9Sstevel@tonic-gate */ 42977c478bd9Sstevel@tonic-gate caddr_t 42987c478bd9Sstevel@tonic-gate hat_kpm_pfn2va(pfn_t pfn) 42997c478bd9Sstevel@tonic-gate { 4300ae115bc7Smrj uintptr_t vaddr = (uintptr_t)kpm_vbase + mmu_ptob(pfn); 43017c478bd9Sstevel@tonic-gate 4302d2b85481Srscott ASSERT(!pfn_is_foreign(pfn)); 43037c478bd9Sstevel@tonic-gate return ((caddr_t)vaddr); 43047c478bd9Sstevel@tonic-gate } 43057c478bd9Sstevel@tonic-gate 43067c478bd9Sstevel@tonic-gate /* 43077c478bd9Sstevel@tonic-gate * Return the kpm virtual address for the page at pp. 43087c478bd9Sstevel@tonic-gate */ 43097c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 43107c478bd9Sstevel@tonic-gate caddr_t 43117c478bd9Sstevel@tonic-gate hat_kpm_page2va(struct page *pp, int checkswap) 43127c478bd9Sstevel@tonic-gate { 43137c478bd9Sstevel@tonic-gate return (hat_kpm_pfn2va(pp->p_pagenum)); 43147c478bd9Sstevel@tonic-gate } 43157c478bd9Sstevel@tonic-gate 43167c478bd9Sstevel@tonic-gate /* 43177c478bd9Sstevel@tonic-gate * Return the page frame number for the kpm virtual address vaddr. 43187c478bd9Sstevel@tonic-gate */ 43197c478bd9Sstevel@tonic-gate pfn_t 43207c478bd9Sstevel@tonic-gate hat_kpm_va2pfn(caddr_t vaddr) 43217c478bd9Sstevel@tonic-gate { 43227c478bd9Sstevel@tonic-gate pfn_t pfn; 43237c478bd9Sstevel@tonic-gate 43247c478bd9Sstevel@tonic-gate ASSERT(IS_KPM_ADDR(vaddr)); 43257c478bd9Sstevel@tonic-gate 43267c478bd9Sstevel@tonic-gate pfn = (pfn_t)btop(vaddr - kpm_vbase); 43277c478bd9Sstevel@tonic-gate 43287c478bd9Sstevel@tonic-gate return (pfn); 43297c478bd9Sstevel@tonic-gate } 43307c478bd9Sstevel@tonic-gate 43317c478bd9Sstevel@tonic-gate 43327c478bd9Sstevel@tonic-gate /* 43337c478bd9Sstevel@tonic-gate * Return the page for the kpm virtual address vaddr. 43347c478bd9Sstevel@tonic-gate */ 43357c478bd9Sstevel@tonic-gate page_t * 43367c478bd9Sstevel@tonic-gate hat_kpm_vaddr2page(caddr_t vaddr) 43377c478bd9Sstevel@tonic-gate { 43387c478bd9Sstevel@tonic-gate pfn_t pfn; 43397c478bd9Sstevel@tonic-gate 43407c478bd9Sstevel@tonic-gate ASSERT(IS_KPM_ADDR(vaddr)); 43417c478bd9Sstevel@tonic-gate 43427c478bd9Sstevel@tonic-gate pfn = hat_kpm_va2pfn(vaddr); 43437c478bd9Sstevel@tonic-gate 43447c478bd9Sstevel@tonic-gate return (page_numtopp_nolock(pfn)); 43457c478bd9Sstevel@tonic-gate } 43467c478bd9Sstevel@tonic-gate 43477c478bd9Sstevel@tonic-gate /* 43487c478bd9Sstevel@tonic-gate * hat_kpm_fault is called from segkpm_fault when we take a page fault on a 43497c478bd9Sstevel@tonic-gate * KPM page. This should never happen on x86 43507c478bd9Sstevel@tonic-gate */ 43517c478bd9Sstevel@tonic-gate int 43527c478bd9Sstevel@tonic-gate hat_kpm_fault(hat_t *hat, caddr_t vaddr) 43537c478bd9Sstevel@tonic-gate { 4354903a11ebSrh panic("pagefault in seg_kpm. hat: 0x%p vaddr: 0x%p", 4355903a11ebSrh (void *)hat, (void *)vaddr); 43567c478bd9Sstevel@tonic-gate 43577c478bd9Sstevel@tonic-gate return (0); 43587c478bd9Sstevel@tonic-gate } 43597c478bd9Sstevel@tonic-gate 43607c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 43617c478bd9Sstevel@tonic-gate void 43627c478bd9Sstevel@tonic-gate hat_kpm_mseghash_clear(int nentries) 43637c478bd9Sstevel@tonic-gate {} 43647c478bd9Sstevel@tonic-gate 43657c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 43667c478bd9Sstevel@tonic-gate void 43677c478bd9Sstevel@tonic-gate hat_kpm_mseghash_update(pgcnt_t inx, struct memseg *msp) 43687c478bd9Sstevel@tonic-gate {} 4369843e1988Sjohnlev 4370a3114836SGerry Liu #ifndef __xpv 4371a3114836SGerry Liu void 4372a3114836SGerry Liu hat_kpm_addmem_mseg_update(struct memseg *msp, pgcnt_t nkpmpgs, 4373a3114836SGerry Liu offset_t kpm_pages_off) 4374a3114836SGerry Liu { 4375a3114836SGerry Liu _NOTE(ARGUNUSED(nkpmpgs, kpm_pages_off)); 4376a3114836SGerry Liu pfn_t base, end; 4377a3114836SGerry Liu 4378a3114836SGerry Liu /* 4379a3114836SGerry Liu * kphysm_add_memory_dynamic() does not set nkpmpgs 4380a3114836SGerry Liu * when page_t memory is externally allocated. That 4381a3114836SGerry Liu * code must properly calculate nkpmpgs in all cases 4382a3114836SGerry Liu * if nkpmpgs needs to be used at some point. 4383a3114836SGerry Liu */ 4384a3114836SGerry Liu 4385a3114836SGerry Liu /* 4386a3114836SGerry Liu * The meta (page_t) pages for dynamically added memory are allocated 4387a3114836SGerry Liu * either from the incoming memory itself or from existing memory. 4388a3114836SGerry Liu * In the former case the base of the incoming pages will be different 4389a3114836SGerry Liu * than the base of the dynamic segment so call memseg_get_start() to 4390a3114836SGerry Liu * get the actual base of the incoming memory for each case. 4391a3114836SGerry Liu */ 4392a3114836SGerry Liu 4393a3114836SGerry Liu base = memseg_get_start(msp); 4394a3114836SGerry Liu end = msp->pages_end; 4395a3114836SGerry Liu 4396a3114836SGerry Liu hat_devload(kas.a_hat, kpm_vbase + mmu_ptob(base), 4397a3114836SGerry Liu mmu_ptob(end - base), base, PROT_READ | PROT_WRITE, 4398a3114836SGerry Liu HAT_LOAD | HAT_LOAD_LOCK | HAT_LOAD_NOCONSIST); 4399a3114836SGerry Liu } 4400a3114836SGerry Liu 4401a3114836SGerry Liu void 4402a3114836SGerry Liu hat_kpm_addmem_mseg_insert(struct memseg *msp) 4403a3114836SGerry Liu { 4404a3114836SGerry Liu _NOTE(ARGUNUSED(msp)); 4405a3114836SGerry Liu } 4406a3114836SGerry Liu 4407a3114836SGerry Liu void 4408a3114836SGerry Liu hat_kpm_addmem_memsegs_update(struct memseg *msp) 4409a3114836SGerry Liu { 4410a3114836SGerry Liu _NOTE(ARGUNUSED(msp)); 4411a3114836SGerry Liu } 4412a3114836SGerry Liu 4413a3114836SGerry Liu /* 4414a3114836SGerry Liu * Return end of metadata for an already setup memseg. 4415a3114836SGerry Liu * X86 platforms don't need per-page meta data to support kpm. 4416a3114836SGerry Liu */ 4417a3114836SGerry Liu caddr_t 4418a3114836SGerry Liu hat_kpm_mseg_reuse(struct memseg *msp) 4419a3114836SGerry Liu { 4420a3114836SGerry Liu return ((caddr_t)msp->epages); 4421a3114836SGerry Liu } 4422a3114836SGerry Liu 4423a3114836SGerry Liu void 4424a3114836SGerry Liu hat_kpm_delmem_mseg_update(struct memseg *msp, struct memseg **mspp) 4425a3114836SGerry Liu { 4426a3114836SGerry Liu _NOTE(ARGUNUSED(msp, mspp)); 4427a3114836SGerry Liu ASSERT(0); 4428a3114836SGerry Liu } 4429a3114836SGerry Liu 4430a3114836SGerry Liu void 4431a3114836SGerry Liu hat_kpm_split_mseg_update(struct memseg *msp, struct memseg **mspp, 4432a3114836SGerry Liu struct memseg *lo, struct memseg *mid, struct memseg *hi) 4433a3114836SGerry Liu { 4434a3114836SGerry Liu _NOTE(ARGUNUSED(msp, mspp, lo, mid, hi)); 4435a3114836SGerry Liu ASSERT(0); 4436a3114836SGerry Liu } 4437a3114836SGerry Liu 4438a3114836SGerry Liu /* 4439a3114836SGerry Liu * Walk the memsegs chain, applying func to each memseg span. 4440a3114836SGerry Liu */ 4441a3114836SGerry Liu void 4442a3114836SGerry Liu hat_kpm_walk(void (*func)(void *, void *, size_t), void *arg) 4443a3114836SGerry Liu { 4444a3114836SGerry Liu pfn_t pbase, pend; 4445a3114836SGerry Liu void *base; 4446a3114836SGerry Liu size_t size; 4447a3114836SGerry Liu struct memseg *msp; 4448a3114836SGerry Liu 4449a3114836SGerry Liu for (msp = memsegs; msp; msp = msp->next) { 4450a3114836SGerry Liu pbase = msp->pages_base; 4451a3114836SGerry Liu pend = msp->pages_end; 4452a3114836SGerry Liu base = ptob(pbase) + kpm_vbase; 4453a3114836SGerry Liu size = ptob(pend - pbase); 4454a3114836SGerry Liu func(arg, base, size); 4455a3114836SGerry Liu } 4456a3114836SGerry Liu } 4457a3114836SGerry Liu 4458a3114836SGerry Liu #else /* __xpv */ 4459a3114836SGerry Liu 4460843e1988Sjohnlev /* 4461843e1988Sjohnlev * There are specific Hypervisor calls to establish and remove mappings 4462843e1988Sjohnlev * to grant table references and the privcmd driver. We have to ensure 4463843e1988Sjohnlev * that a page table actually exists. 4464843e1988Sjohnlev */ 4465843e1988Sjohnlev void 44667eea693dSMark Johnson hat_prepare_mapping(hat_t *hat, caddr_t addr, uint64_t *pte_ma) 4467843e1988Sjohnlev { 44687eea693dSMark Johnson maddr_t base_ma; 44697eea693dSMark Johnson htable_t *ht; 44707eea693dSMark Johnson uint_t entry; 44717eea693dSMark Johnson 4472843e1988Sjohnlev ASSERT(IS_P2ALIGNED((uintptr_t)addr, MMU_PAGESIZE)); 4473551bc2a6Smrj XPV_DISALLOW_MIGRATE(); 44747eea693dSMark Johnson ht = htable_create(hat, (uintptr_t)addr, 0, NULL); 44757eea693dSMark Johnson 44767eea693dSMark Johnson /* 44777eea693dSMark Johnson * if an address for pte_ma is passed in, return the MA of the pte 44787eea693dSMark Johnson * for this specific address. This address is only valid as long 44797eea693dSMark Johnson * as the htable stays locked. 44807eea693dSMark Johnson */ 44817eea693dSMark Johnson if (pte_ma != NULL) { 44827eea693dSMark Johnson entry = htable_va2entry((uintptr_t)addr, ht); 44837eea693dSMark Johnson base_ma = pa_to_ma(ptob(ht->ht_pfn)); 44847eea693dSMark Johnson *pte_ma = base_ma + (entry << mmu.pte_size_shift); 44857eea693dSMark Johnson } 4486551bc2a6Smrj XPV_ALLOW_MIGRATE(); 4487843e1988Sjohnlev } 4488843e1988Sjohnlev 4489843e1988Sjohnlev void 4490843e1988Sjohnlev hat_release_mapping(hat_t *hat, caddr_t addr) 4491843e1988Sjohnlev { 4492843e1988Sjohnlev htable_t *ht; 4493843e1988Sjohnlev 4494843e1988Sjohnlev ASSERT(IS_P2ALIGNED((uintptr_t)addr, MMU_PAGESIZE)); 4495551bc2a6Smrj XPV_DISALLOW_MIGRATE(); 4496843e1988Sjohnlev ht = htable_lookup(hat, (uintptr_t)addr, 0); 4497843e1988Sjohnlev ASSERT(ht != NULL); 4498843e1988Sjohnlev ASSERT(ht->ht_busy >= 2); 4499843e1988Sjohnlev htable_release(ht); 4500843e1988Sjohnlev htable_release(ht); 4501551bc2a6Smrj XPV_ALLOW_MIGRATE(); 4502a3114836SGerry Liu } 4503a3114836SGerry Liu #endif /* __xpv */ 4504