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 */ 28*c7c6ab2aSGarrett D'Amore /* 29*c7c6ab2aSGarrett D'Amore * Copyright 2011 Nexenta Systems, Inc. All rights reserved. 30*c7c6ab2aSGarrett D'Amore */ 317c478bd9Sstevel@tonic-gate 327c478bd9Sstevel@tonic-gate /* 337c478bd9Sstevel@tonic-gate * VM - Hardware Address Translation management for i386 and amd64 347c478bd9Sstevel@tonic-gate * 357c478bd9Sstevel@tonic-gate * Implementation of the interfaces described in <common/vm/hat.h> 367c478bd9Sstevel@tonic-gate * 377c478bd9Sstevel@tonic-gate * Nearly all the details of how the hardware is managed should not be 387c478bd9Sstevel@tonic-gate * visible outside this layer except for misc. machine specific functions 397c478bd9Sstevel@tonic-gate * that work in conjunction with this code. 407c478bd9Sstevel@tonic-gate * 417c478bd9Sstevel@tonic-gate * Routines used only inside of i86pc/vm start with hati_ for HAT Internal. 427c478bd9Sstevel@tonic-gate */ 437c478bd9Sstevel@tonic-gate 447c478bd9Sstevel@tonic-gate #include <sys/machparam.h> 457c478bd9Sstevel@tonic-gate #include <sys/machsystm.h> 467c478bd9Sstevel@tonic-gate #include <sys/mman.h> 477c478bd9Sstevel@tonic-gate #include <sys/types.h> 487c478bd9Sstevel@tonic-gate #include <sys/systm.h> 497c478bd9Sstevel@tonic-gate #include <sys/cpuvar.h> 507c478bd9Sstevel@tonic-gate #include <sys/thread.h> 517c478bd9Sstevel@tonic-gate #include <sys/proc.h> 527c478bd9Sstevel@tonic-gate #include <sys/cpu.h> 537c478bd9Sstevel@tonic-gate #include <sys/kmem.h> 547c478bd9Sstevel@tonic-gate #include <sys/disp.h> 557c478bd9Sstevel@tonic-gate #include <sys/shm.h> 567c478bd9Sstevel@tonic-gate #include <sys/sysmacros.h> 577c478bd9Sstevel@tonic-gate #include <sys/machparam.h> 587c478bd9Sstevel@tonic-gate #include <sys/vmem.h> 597c478bd9Sstevel@tonic-gate #include <sys/vmsystm.h> 607c478bd9Sstevel@tonic-gate #include <sys/promif.h> 617c478bd9Sstevel@tonic-gate #include <sys/var.h> 627c478bd9Sstevel@tonic-gate #include <sys/x86_archext.h> 637c478bd9Sstevel@tonic-gate #include <sys/atomic.h> 647c478bd9Sstevel@tonic-gate #include <sys/bitmap.h> 65ae115bc7Smrj #include <sys/controlregs.h> 66ae115bc7Smrj #include <sys/bootconf.h> 67ae115bc7Smrj #include <sys/bootsvcs.h> 68ae115bc7Smrj #include <sys/bootinfo.h> 6995c0a3c8Sjosephb #include <sys/archsystm.h> 707c478bd9Sstevel@tonic-gate 717c478bd9Sstevel@tonic-gate #include <vm/seg_kmem.h> 727c478bd9Sstevel@tonic-gate #include <vm/hat_i86.h> 737c478bd9Sstevel@tonic-gate #include <vm/as.h> 747c478bd9Sstevel@tonic-gate #include <vm/seg.h> 757c478bd9Sstevel@tonic-gate #include <vm/page.h> 767c478bd9Sstevel@tonic-gate #include <vm/seg_kp.h> 777c478bd9Sstevel@tonic-gate #include <vm/seg_kpm.h> 787c478bd9Sstevel@tonic-gate #include <vm/vm_dep.h> 79843e1988Sjohnlev #ifdef __xpv 80843e1988Sjohnlev #include <sys/hypervisor.h> 81843e1988Sjohnlev #endif 82ae115bc7Smrj #include <vm/kboot_mmu.h> 83250b7ff9Sjosephb #include <vm/seg_spt.h> 847c478bd9Sstevel@tonic-gate 857c478bd9Sstevel@tonic-gate #include <sys/cmn_err.h> 867c478bd9Sstevel@tonic-gate 877c478bd9Sstevel@tonic-gate /* 887c478bd9Sstevel@tonic-gate * Basic parameters for hat operation. 897c478bd9Sstevel@tonic-gate */ 907c478bd9Sstevel@tonic-gate struct hat_mmu_info mmu; 917c478bd9Sstevel@tonic-gate 927c478bd9Sstevel@tonic-gate /* 937c478bd9Sstevel@tonic-gate * The page that is the kernel's top level pagetable. 947c478bd9Sstevel@tonic-gate * 95843e1988Sjohnlev * For 32 bit PAE support on i86pc, the kernel hat will use the 1st 4 entries 967c478bd9Sstevel@tonic-gate * on this 4K page for its top level page table. The remaining groups of 977c478bd9Sstevel@tonic-gate * 4 entries are used for per processor copies of user VLP pagetables for 987c478bd9Sstevel@tonic-gate * running threads. See hat_switch() and reload_pae32() for details. 997c478bd9Sstevel@tonic-gate * 100843e1988Sjohnlev * vlp_page[0..3] - level==2 PTEs for kernel HAT 101843e1988Sjohnlev * vlp_page[4..7] - level==2 PTEs for user thread on cpu 0 102843e1988Sjohnlev * vlp_page[8..11] - level==2 PTE for user thread on cpu 1 103843e1988Sjohnlev * etc... 1047c478bd9Sstevel@tonic-gate */ 1057c478bd9Sstevel@tonic-gate static x86pte_t *vlp_page; 1067c478bd9Sstevel@tonic-gate 1077c478bd9Sstevel@tonic-gate /* 1087c478bd9Sstevel@tonic-gate * forward declaration of internal utility routines 1097c478bd9Sstevel@tonic-gate */ 1107c478bd9Sstevel@tonic-gate static x86pte_t hati_update_pte(htable_t *ht, uint_t entry, x86pte_t expected, 1117c478bd9Sstevel@tonic-gate x86pte_t new); 1127c478bd9Sstevel@tonic-gate 1137c478bd9Sstevel@tonic-gate /* 1147c478bd9Sstevel@tonic-gate * The kernel address space exists in all HATs. To implement this the 115843e1988Sjohnlev * kernel reserves a fixed number of entries in the topmost level(s) of page 116843e1988Sjohnlev * tables. The values are setup during startup and then copied to every user 117843e1988Sjohnlev * hat created by hat_alloc(). This means that kernelbase must be: 1187c478bd9Sstevel@tonic-gate * 1197c478bd9Sstevel@tonic-gate * 4Meg aligned for 32 bit kernels 1207c478bd9Sstevel@tonic-gate * 512Gig aligned for x86_64 64 bit kernel 1217c478bd9Sstevel@tonic-gate * 122843e1988Sjohnlev * The hat_kernel_range_ts describe what needs to be copied from kernel hat 123843e1988Sjohnlev * to each user hat. 1247c478bd9Sstevel@tonic-gate */ 125843e1988Sjohnlev typedef struct hat_kernel_range { 126843e1988Sjohnlev level_t hkr_level; 127843e1988Sjohnlev uintptr_t hkr_start_va; 128843e1988Sjohnlev uintptr_t hkr_end_va; /* zero means to end of memory */ 129843e1988Sjohnlev } hat_kernel_range_t; 130843e1988Sjohnlev #define NUM_KERNEL_RANGE 2 131843e1988Sjohnlev static hat_kernel_range_t kernel_ranges[NUM_KERNEL_RANGE]; 132843e1988Sjohnlev static int num_kernel_ranges; 1337c478bd9Sstevel@tonic-gate 1347c478bd9Sstevel@tonic-gate uint_t use_boot_reserve = 1; /* cleared after early boot process */ 1357c478bd9Sstevel@tonic-gate uint_t can_steal_post_boot = 0; /* set late in boot to enable stealing */ 1367c478bd9Sstevel@tonic-gate 137512cf780Skchow /* 138512cf780Skchow * enable_1gpg: controls 1g page support for user applications. 139512cf780Skchow * By default, 1g pages are exported to user applications. enable_1gpg can 140512cf780Skchow * be set to 0 to not export. 141512cf780Skchow */ 14278b03d3aSkchow int enable_1gpg = 1; 14302bc52beSkchow 144512cf780Skchow /* 145512cf780Skchow * AMD shanghai processors provide better management of 1gb ptes in its tlb. 14621584dbcSPavel Tatashin * By default, 1g page support will be disabled for pre-shanghai AMD 147512cf780Skchow * processors that don't have optimal tlb support for the 1g page size. 148512cf780Skchow * chk_optimal_1gtlb can be set to 0 to force 1g page support on sub-optimal 149512cf780Skchow * processors. 150512cf780Skchow */ 151512cf780Skchow int chk_optimal_1gtlb = 1; 152512cf780Skchow 153512cf780Skchow 15402bc52beSkchow #ifdef DEBUG 15502bc52beSkchow uint_t map1gcnt; 15602bc52beSkchow #endif 15702bc52beSkchow 15802bc52beSkchow 1597c478bd9Sstevel@tonic-gate /* 1607c478bd9Sstevel@tonic-gate * A cpuset for all cpus. This is used for kernel address cross calls, since 1617c478bd9Sstevel@tonic-gate * the kernel addresses apply to all cpus. 1627c478bd9Sstevel@tonic-gate */ 1637c478bd9Sstevel@tonic-gate cpuset_t khat_cpuset; 1647c478bd9Sstevel@tonic-gate 1657c478bd9Sstevel@tonic-gate /* 1667c478bd9Sstevel@tonic-gate * management stuff for hat structures 1677c478bd9Sstevel@tonic-gate */ 1687c478bd9Sstevel@tonic-gate kmutex_t hat_list_lock; 1697c478bd9Sstevel@tonic-gate kcondvar_t hat_list_cv; 1707c478bd9Sstevel@tonic-gate kmem_cache_t *hat_cache; 1717c478bd9Sstevel@tonic-gate kmem_cache_t *hat_hash_cache; 1727c478bd9Sstevel@tonic-gate kmem_cache_t *vlp_hash_cache; 1737c478bd9Sstevel@tonic-gate 1747c478bd9Sstevel@tonic-gate /* 1757c478bd9Sstevel@tonic-gate * Simple statistics 1767c478bd9Sstevel@tonic-gate */ 1777c478bd9Sstevel@tonic-gate struct hatstats hatstat; 1787c478bd9Sstevel@tonic-gate 179ab4a9bebSjohnlev /* 180ab4a9bebSjohnlev * Some earlier hypervisor versions do not emulate cmpxchg of PTEs 181ab4a9bebSjohnlev * correctly. For such hypervisors we must set PT_USER for kernel 182ab4a9bebSjohnlev * entries ourselves (normally the emulation would set PT_USER for 183ab4a9bebSjohnlev * kernel entries and PT_USER|PT_GLOBAL for user entries). pt_kern is 184ab4a9bebSjohnlev * thus set appropriately. Note that dboot/kbm is OK, as only the full 185ab4a9bebSjohnlev * HAT uses cmpxchg() and the other paths (hypercall etc.) were never 186ab4a9bebSjohnlev * incorrect. 187ab4a9bebSjohnlev */ 188ab4a9bebSjohnlev int pt_kern; 189ab4a9bebSjohnlev 1907c478bd9Sstevel@tonic-gate /* 1917c478bd9Sstevel@tonic-gate * useful stuff for atomic access/clearing/setting REF/MOD/RO bits in page_t's. 1927c478bd9Sstevel@tonic-gate */ 1937c478bd9Sstevel@tonic-gate extern void atomic_orb(uchar_t *addr, uchar_t val); 1947c478bd9Sstevel@tonic-gate extern void atomic_andb(uchar_t *addr, uchar_t val); 1957c478bd9Sstevel@tonic-gate 196a3114836SGerry Liu #ifndef __xpv 197a3114836SGerry Liu extern pfn_t memseg_get_start(struct memseg *); 198a3114836SGerry Liu #endif 199a3114836SGerry Liu 2007c478bd9Sstevel@tonic-gate #define PP_GETRM(pp, rmmask) (pp->p_nrm & rmmask) 2017c478bd9Sstevel@tonic-gate #define PP_ISMOD(pp) PP_GETRM(pp, P_MOD) 2027c478bd9Sstevel@tonic-gate #define PP_ISREF(pp) PP_GETRM(pp, P_REF) 2037c478bd9Sstevel@tonic-gate #define PP_ISRO(pp) PP_GETRM(pp, P_RO) 2047c478bd9Sstevel@tonic-gate 2057c478bd9Sstevel@tonic-gate #define PP_SETRM(pp, rm) atomic_orb(&(pp->p_nrm), rm) 2067c478bd9Sstevel@tonic-gate #define PP_SETMOD(pp) PP_SETRM(pp, P_MOD) 2077c478bd9Sstevel@tonic-gate #define PP_SETREF(pp) PP_SETRM(pp, P_REF) 2087c478bd9Sstevel@tonic-gate #define PP_SETRO(pp) PP_SETRM(pp, P_RO) 2097c478bd9Sstevel@tonic-gate 2107c478bd9Sstevel@tonic-gate #define PP_CLRRM(pp, rm) atomic_andb(&(pp->p_nrm), ~(rm)) 2117c478bd9Sstevel@tonic-gate #define PP_CLRMOD(pp) PP_CLRRM(pp, P_MOD) 2127c478bd9Sstevel@tonic-gate #define PP_CLRREF(pp) PP_CLRRM(pp, P_REF) 2137c478bd9Sstevel@tonic-gate #define PP_CLRRO(pp) PP_CLRRM(pp, P_RO) 2147c478bd9Sstevel@tonic-gate #define PP_CLRALL(pp) PP_CLRRM(pp, P_MOD | P_REF | P_RO) 2157c478bd9Sstevel@tonic-gate 2167c478bd9Sstevel@tonic-gate /* 2177c478bd9Sstevel@tonic-gate * kmem cache constructor for struct hat 2187c478bd9Sstevel@tonic-gate */ 2197c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 2207c478bd9Sstevel@tonic-gate static int 2217c478bd9Sstevel@tonic-gate hati_constructor(void *buf, void *handle, int kmflags) 2227c478bd9Sstevel@tonic-gate { 2237c478bd9Sstevel@tonic-gate hat_t *hat = buf; 2247c478bd9Sstevel@tonic-gate 2257c478bd9Sstevel@tonic-gate mutex_init(&hat->hat_mutex, NULL, MUTEX_DEFAULT, NULL); 2267c478bd9Sstevel@tonic-gate bzero(hat->hat_pages_mapped, 2277c478bd9Sstevel@tonic-gate sizeof (pgcnt_t) * (mmu.max_page_level + 1)); 228250b7ff9Sjosephb hat->hat_ism_pgcnt = 0; 2297c478bd9Sstevel@tonic-gate hat->hat_stats = 0; 2307c478bd9Sstevel@tonic-gate hat->hat_flags = 0; 2317c478bd9Sstevel@tonic-gate CPUSET_ZERO(hat->hat_cpus); 2327c478bd9Sstevel@tonic-gate hat->hat_htable = NULL; 2337c478bd9Sstevel@tonic-gate hat->hat_ht_hash = NULL; 2347c478bd9Sstevel@tonic-gate return (0); 2357c478bd9Sstevel@tonic-gate } 2367c478bd9Sstevel@tonic-gate 2377c478bd9Sstevel@tonic-gate /* 2387c478bd9Sstevel@tonic-gate * Allocate a hat structure for as. We also create the top level 2397c478bd9Sstevel@tonic-gate * htable and initialize it to contain the kernel hat entries. 2407c478bd9Sstevel@tonic-gate */ 2417c478bd9Sstevel@tonic-gate hat_t * 2427c478bd9Sstevel@tonic-gate hat_alloc(struct as *as) 2437c478bd9Sstevel@tonic-gate { 244843e1988Sjohnlev hat_t *hat; 245843e1988Sjohnlev htable_t *ht; /* top level htable */ 246843e1988Sjohnlev uint_t use_vlp; 247843e1988Sjohnlev uint_t r; 248843e1988Sjohnlev hat_kernel_range_t *rp; 249843e1988Sjohnlev uintptr_t va; 250843e1988Sjohnlev uintptr_t eva; 251843e1988Sjohnlev uint_t start; 252843e1988Sjohnlev uint_t cnt; 253843e1988Sjohnlev htable_t *src; 2547c478bd9Sstevel@tonic-gate 2557c478bd9Sstevel@tonic-gate /* 2567c478bd9Sstevel@tonic-gate * Once we start creating user process HATs we can enable 2577c478bd9Sstevel@tonic-gate * the htable_steal() code. 2587c478bd9Sstevel@tonic-gate */ 2597c478bd9Sstevel@tonic-gate if (can_steal_post_boot == 0) 2607c478bd9Sstevel@tonic-gate can_steal_post_boot = 1; 2617c478bd9Sstevel@tonic-gate 2627c478bd9Sstevel@tonic-gate ASSERT(AS_WRITE_HELD(as, &as->a_lock)); 2637c478bd9Sstevel@tonic-gate hat = kmem_cache_alloc(hat_cache, KM_SLEEP); 2647c478bd9Sstevel@tonic-gate hat->hat_as = as; 2657c478bd9Sstevel@tonic-gate mutex_init(&hat->hat_mutex, NULL, MUTEX_DEFAULT, NULL); 2667c478bd9Sstevel@tonic-gate ASSERT(hat->hat_flags == 0); 2677c478bd9Sstevel@tonic-gate 268843e1988Sjohnlev #if defined(__xpv) 2697c478bd9Sstevel@tonic-gate /* 270843e1988Sjohnlev * No VLP stuff on the hypervisor due to the 64-bit split top level 271843e1988Sjohnlev * page tables. On 32-bit it's not needed as the hypervisor takes 272843e1988Sjohnlev * care of copying the top level PTEs to a below 4Gig page. 2737c478bd9Sstevel@tonic-gate */ 274843e1988Sjohnlev use_vlp = 0; 275843e1988Sjohnlev #else /* __xpv */ 276843e1988Sjohnlev /* 32 bit processes uses a VLP style hat when running with PAE */ 2777c478bd9Sstevel@tonic-gate #if defined(__amd64) 2787c478bd9Sstevel@tonic-gate use_vlp = (ttoproc(curthread)->p_model == DATAMODEL_ILP32); 2797c478bd9Sstevel@tonic-gate #elif defined(__i386) 2807c478bd9Sstevel@tonic-gate use_vlp = mmu.pae_hat; 2817c478bd9Sstevel@tonic-gate #endif 282843e1988Sjohnlev #endif /* __xpv */ 2837c478bd9Sstevel@tonic-gate if (use_vlp) { 2847c478bd9Sstevel@tonic-gate hat->hat_flags = HAT_VLP; 2857c478bd9Sstevel@tonic-gate bzero(hat->hat_vlp_ptes, VLP_SIZE); 2867c478bd9Sstevel@tonic-gate } 2877c478bd9Sstevel@tonic-gate 2887c478bd9Sstevel@tonic-gate /* 2897c478bd9Sstevel@tonic-gate * Allocate the htable hash 2907c478bd9Sstevel@tonic-gate */ 2917c478bd9Sstevel@tonic-gate if ((hat->hat_flags & HAT_VLP)) { 2927c478bd9Sstevel@tonic-gate hat->hat_num_hash = mmu.vlp_hash_cnt; 2937c478bd9Sstevel@tonic-gate hat->hat_ht_hash = kmem_cache_alloc(vlp_hash_cache, KM_SLEEP); 2947c478bd9Sstevel@tonic-gate } else { 2957c478bd9Sstevel@tonic-gate hat->hat_num_hash = mmu.hash_cnt; 2967c478bd9Sstevel@tonic-gate hat->hat_ht_hash = kmem_cache_alloc(hat_hash_cache, KM_SLEEP); 2977c478bd9Sstevel@tonic-gate } 2987c478bd9Sstevel@tonic-gate bzero(hat->hat_ht_hash, hat->hat_num_hash * sizeof (htable_t *)); 2997c478bd9Sstevel@tonic-gate 3007c478bd9Sstevel@tonic-gate /* 3017c478bd9Sstevel@tonic-gate * Initialize Kernel HAT entries at the top of the top level page 302843e1988Sjohnlev * tables for the new hat. 3037c478bd9Sstevel@tonic-gate */ 3047c478bd9Sstevel@tonic-gate hat->hat_htable = NULL; 3057c478bd9Sstevel@tonic-gate hat->hat_ht_cached = NULL; 306843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 3077c478bd9Sstevel@tonic-gate ht = htable_create(hat, (uintptr_t)0, TOP_LEVEL(hat), NULL); 308843e1988Sjohnlev hat->hat_htable = ht; 309ae115bc7Smrj 310843e1988Sjohnlev #if defined(__amd64) 311843e1988Sjohnlev if (hat->hat_flags & HAT_VLP) 312843e1988Sjohnlev goto init_done; 3137c478bd9Sstevel@tonic-gate #endif 3147c478bd9Sstevel@tonic-gate 315843e1988Sjohnlev for (r = 0; r < num_kernel_ranges; ++r) { 316843e1988Sjohnlev rp = &kernel_ranges[r]; 317843e1988Sjohnlev for (va = rp->hkr_start_va; va != rp->hkr_end_va; 318843e1988Sjohnlev va += cnt * LEVEL_SIZE(rp->hkr_level)) { 319843e1988Sjohnlev 320843e1988Sjohnlev if (rp->hkr_level == TOP_LEVEL(hat)) 321843e1988Sjohnlev ht = hat->hat_htable; 322843e1988Sjohnlev else 323843e1988Sjohnlev ht = htable_create(hat, va, rp->hkr_level, 324843e1988Sjohnlev NULL); 325843e1988Sjohnlev 326843e1988Sjohnlev start = htable_va2entry(va, ht); 327843e1988Sjohnlev cnt = HTABLE_NUM_PTES(ht) - start; 328843e1988Sjohnlev eva = va + 329843e1988Sjohnlev ((uintptr_t)cnt << LEVEL_SHIFT(rp->hkr_level)); 330843e1988Sjohnlev if (rp->hkr_end_va != 0 && 331843e1988Sjohnlev (eva > rp->hkr_end_va || eva == 0)) 332843e1988Sjohnlev cnt = htable_va2entry(rp->hkr_end_va, ht) - 333843e1988Sjohnlev start; 334843e1988Sjohnlev 335843e1988Sjohnlev #if defined(__i386) && !defined(__xpv) 336843e1988Sjohnlev if (ht->ht_flags & HTABLE_VLP) { 337843e1988Sjohnlev bcopy(&vlp_page[start], 338843e1988Sjohnlev &hat->hat_vlp_ptes[start], 339843e1988Sjohnlev cnt * sizeof (x86pte_t)); 340843e1988Sjohnlev continue; 341843e1988Sjohnlev } 342843e1988Sjohnlev #endif 343843e1988Sjohnlev src = htable_lookup(kas.a_hat, va, rp->hkr_level); 344843e1988Sjohnlev ASSERT(src != NULL); 345843e1988Sjohnlev x86pte_copy(src, ht, start, cnt); 346843e1988Sjohnlev htable_release(src); 347843e1988Sjohnlev } 348843e1988Sjohnlev } 349843e1988Sjohnlev 350843e1988Sjohnlev init_done: 351843e1988Sjohnlev 352843e1988Sjohnlev #if defined(__xpv) 3537c478bd9Sstevel@tonic-gate /* 354843e1988Sjohnlev * Pin top level page tables after initializing them 3557c478bd9Sstevel@tonic-gate */ 356843e1988Sjohnlev xen_pin(hat->hat_htable->ht_pfn, mmu.max_level); 357843e1988Sjohnlev #if defined(__amd64) 358843e1988Sjohnlev xen_pin(hat->hat_user_ptable, mmu.max_level); 359843e1988Sjohnlev #endif 3607c478bd9Sstevel@tonic-gate #endif 361551bc2a6Smrj XPV_ALLOW_MIGRATE(); 3627c478bd9Sstevel@tonic-gate 3637c478bd9Sstevel@tonic-gate /* 364a85a6733Sjosephb * Put it at the start of the global list of all hats (used by stealing) 365a85a6733Sjosephb * 366a85a6733Sjosephb * kas.a_hat is not in the list but is instead used to find the 367a85a6733Sjosephb * first and last items in the list. 368a85a6733Sjosephb * 369a85a6733Sjosephb * - kas.a_hat->hat_next points to the start of the user hats. 370a85a6733Sjosephb * The list ends where hat->hat_next == NULL 371a85a6733Sjosephb * 372a85a6733Sjosephb * - kas.a_hat->hat_prev points to the last of the user hats. 373a85a6733Sjosephb * The list begins where hat->hat_prev == NULL 3747c478bd9Sstevel@tonic-gate */ 3757c478bd9Sstevel@tonic-gate mutex_enter(&hat_list_lock); 376a85a6733Sjosephb hat->hat_prev = NULL; 377a85a6733Sjosephb hat->hat_next = kas.a_hat->hat_next; 378a85a6733Sjosephb if (hat->hat_next) 379a85a6733Sjosephb hat->hat_next->hat_prev = hat; 380a85a6733Sjosephb else 381a85a6733Sjosephb kas.a_hat->hat_prev = hat; 3827c478bd9Sstevel@tonic-gate kas.a_hat->hat_next = hat; 3837c478bd9Sstevel@tonic-gate mutex_exit(&hat_list_lock); 3847c478bd9Sstevel@tonic-gate 3857c478bd9Sstevel@tonic-gate return (hat); 3867c478bd9Sstevel@tonic-gate } 3877c478bd9Sstevel@tonic-gate 3887c478bd9Sstevel@tonic-gate /* 3897c478bd9Sstevel@tonic-gate * process has finished executing but as has not been cleaned up yet. 3907c478bd9Sstevel@tonic-gate */ 3917c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 3927c478bd9Sstevel@tonic-gate void 3937c478bd9Sstevel@tonic-gate hat_free_start(hat_t *hat) 3947c478bd9Sstevel@tonic-gate { 3957c478bd9Sstevel@tonic-gate ASSERT(AS_WRITE_HELD(hat->hat_as, &hat->hat_as->a_lock)); 396a85a6733Sjosephb 397a85a6733Sjosephb /* 398a85a6733Sjosephb * If the hat is currently a stealing victim, wait for the stealing 399a85a6733Sjosephb * to finish. Once we mark it as HAT_FREEING, htable_steal() 400a85a6733Sjosephb * won't look at its pagetables anymore. 401a85a6733Sjosephb */ 4027c478bd9Sstevel@tonic-gate mutex_enter(&hat_list_lock); 403a85a6733Sjosephb while (hat->hat_flags & HAT_VICTIM) 404a85a6733Sjosephb cv_wait(&hat_list_cv, &hat_list_lock); 4057c478bd9Sstevel@tonic-gate hat->hat_flags |= HAT_FREEING; 4067c478bd9Sstevel@tonic-gate mutex_exit(&hat_list_lock); 4077c478bd9Sstevel@tonic-gate } 4087c478bd9Sstevel@tonic-gate 4097c478bd9Sstevel@tonic-gate /* 4107c478bd9Sstevel@tonic-gate * An address space is being destroyed, so we destroy the associated hat. 4117c478bd9Sstevel@tonic-gate */ 4127c478bd9Sstevel@tonic-gate void 4137c478bd9Sstevel@tonic-gate hat_free_end(hat_t *hat) 4147c478bd9Sstevel@tonic-gate { 4157c478bd9Sstevel@tonic-gate kmem_cache_t *cache; 4167c478bd9Sstevel@tonic-gate 4177c478bd9Sstevel@tonic-gate ASSERT(hat->hat_flags & HAT_FREEING); 4187c478bd9Sstevel@tonic-gate 4197c478bd9Sstevel@tonic-gate /* 4207c478bd9Sstevel@tonic-gate * must not be running on the given hat 4217c478bd9Sstevel@tonic-gate */ 4227c478bd9Sstevel@tonic-gate ASSERT(CPU->cpu_current_hat != hat); 4237c478bd9Sstevel@tonic-gate 4247c478bd9Sstevel@tonic-gate /* 425a85a6733Sjosephb * Remove it from the list of HATs 4267c478bd9Sstevel@tonic-gate */ 4277c478bd9Sstevel@tonic-gate mutex_enter(&hat_list_lock); 428a85a6733Sjosephb if (hat->hat_prev) 429a85a6733Sjosephb hat->hat_prev->hat_next = hat->hat_next; 430a85a6733Sjosephb else 4317c478bd9Sstevel@tonic-gate kas.a_hat->hat_next = hat->hat_next; 432a85a6733Sjosephb if (hat->hat_next) 433a85a6733Sjosephb hat->hat_next->hat_prev = hat->hat_prev; 434a85a6733Sjosephb else 435a85a6733Sjosephb kas.a_hat->hat_prev = hat->hat_prev; 4367c478bd9Sstevel@tonic-gate mutex_exit(&hat_list_lock); 437a85a6733Sjosephb hat->hat_next = hat->hat_prev = NULL; 4387c478bd9Sstevel@tonic-gate 439843e1988Sjohnlev #if defined(__xpv) 440843e1988Sjohnlev /* 441843e1988Sjohnlev * On the hypervisor, unpin top level page table(s) 442843e1988Sjohnlev */ 443843e1988Sjohnlev xen_unpin(hat->hat_htable->ht_pfn); 444843e1988Sjohnlev #if defined(__amd64) 445843e1988Sjohnlev xen_unpin(hat->hat_user_ptable); 446843e1988Sjohnlev #endif 447843e1988Sjohnlev #endif 448843e1988Sjohnlev 4497c478bd9Sstevel@tonic-gate /* 4507c478bd9Sstevel@tonic-gate * Make a pass through the htables freeing them all up. 4517c478bd9Sstevel@tonic-gate */ 4527c478bd9Sstevel@tonic-gate htable_purge_hat(hat); 4537c478bd9Sstevel@tonic-gate 4547c478bd9Sstevel@tonic-gate /* 4557c478bd9Sstevel@tonic-gate * Decide which kmem cache the hash table came from, then free it. 4567c478bd9Sstevel@tonic-gate */ 4577c478bd9Sstevel@tonic-gate if (hat->hat_flags & HAT_VLP) 4587c478bd9Sstevel@tonic-gate cache = vlp_hash_cache; 4597c478bd9Sstevel@tonic-gate else 4607c478bd9Sstevel@tonic-gate cache = hat_hash_cache; 4617c478bd9Sstevel@tonic-gate kmem_cache_free(cache, hat->hat_ht_hash); 4627c478bd9Sstevel@tonic-gate hat->hat_ht_hash = NULL; 4637c478bd9Sstevel@tonic-gate 4647c478bd9Sstevel@tonic-gate hat->hat_flags = 0; 4657c478bd9Sstevel@tonic-gate kmem_cache_free(hat_cache, hat); 4667c478bd9Sstevel@tonic-gate } 4677c478bd9Sstevel@tonic-gate 4687c478bd9Sstevel@tonic-gate /* 4697c478bd9Sstevel@tonic-gate * round kernelbase down to a supported value to use for _userlimit 4707c478bd9Sstevel@tonic-gate * 4717c478bd9Sstevel@tonic-gate * userlimit must be aligned down to an entry in the top level htable. 4727c478bd9Sstevel@tonic-gate * The one exception is for 32 bit HAT's running PAE. 4737c478bd9Sstevel@tonic-gate */ 4747c478bd9Sstevel@tonic-gate uintptr_t 4757c478bd9Sstevel@tonic-gate hat_kernelbase(uintptr_t va) 4767c478bd9Sstevel@tonic-gate { 4777c478bd9Sstevel@tonic-gate #if defined(__i386) 4787c478bd9Sstevel@tonic-gate va &= LEVEL_MASK(1); 4797c478bd9Sstevel@tonic-gate #endif 4807c478bd9Sstevel@tonic-gate if (IN_VA_HOLE(va)) 4817c478bd9Sstevel@tonic-gate panic("_userlimit %p will fall in VA hole\n", (void *)va); 4827c478bd9Sstevel@tonic-gate return (va); 4837c478bd9Sstevel@tonic-gate } 4847c478bd9Sstevel@tonic-gate 485512cf780Skchow /* 486512cf780Skchow * 487512cf780Skchow */ 488512cf780Skchow static void 489512cf780Skchow set_max_page_level() 490512cf780Skchow { 491512cf780Skchow level_t lvl; 492512cf780Skchow 493512cf780Skchow if (!kbm_largepage_support) { 494512cf780Skchow lvl = 0; 4956f0cf5c5Skchow } else { 4967417cfdeSKuriakose Kuruvilla if (is_x86_feature(x86_featureset, X86FSET_1GPG)) { 4976f0cf5c5Skchow lvl = 2; 4986f0cf5c5Skchow if (chk_optimal_1gtlb && 4996f0cf5c5Skchow cpuid_opteron_erratum(CPU, 6671130)) { 5006f0cf5c5Skchow lvl = 1; 5016f0cf5c5Skchow } 5026f0cf5c5Skchow if (plat_mnode_xcheck(LEVEL_SIZE(2) >> 5036f0cf5c5Skchow LEVEL_SHIFT(0))) { 5046f0cf5c5Skchow lvl = 1; 5056f0cf5c5Skchow } 5066f0cf5c5Skchow } else { 507512cf780Skchow lvl = 1; 508512cf780Skchow } 509512cf780Skchow } 510512cf780Skchow mmu.max_page_level = lvl; 511512cf780Skchow 512512cf780Skchow if ((lvl == 2) && (enable_1gpg == 0)) 513512cf780Skchow mmu.umax_page_level = 1; 514512cf780Skchow else 515512cf780Skchow mmu.umax_page_level = lvl; 516512cf780Skchow } 517512cf780Skchow 5187c478bd9Sstevel@tonic-gate /* 5197c478bd9Sstevel@tonic-gate * Initialize hat data structures based on processor MMU information. 5207c478bd9Sstevel@tonic-gate */ 5217c478bd9Sstevel@tonic-gate void 5227c478bd9Sstevel@tonic-gate mmu_init(void) 5237c478bd9Sstevel@tonic-gate { 5247c478bd9Sstevel@tonic-gate uint_t max_htables; 5257c478bd9Sstevel@tonic-gate uint_t pa_bits; 5267c478bd9Sstevel@tonic-gate uint_t va_bits; 5277c478bd9Sstevel@tonic-gate int i; 5287c478bd9Sstevel@tonic-gate 5297c478bd9Sstevel@tonic-gate /* 530ae115bc7Smrj * If CPU enabled the page table global bit, use it for the kernel 531ae115bc7Smrj * This is bit 7 in CR4 (PGE - Page Global Enable). 5327c478bd9Sstevel@tonic-gate */ 5337417cfdeSKuriakose Kuruvilla if (is_x86_feature(x86_featureset, X86FSET_PGE) && 5347417cfdeSKuriakose Kuruvilla (getcr4() & CR4_PGE) != 0) 5357c478bd9Sstevel@tonic-gate mmu.pt_global = PT_GLOBAL; 5367c478bd9Sstevel@tonic-gate 5377c478bd9Sstevel@tonic-gate /* 538ae115bc7Smrj * Detect NX and PAE usage. 5397c478bd9Sstevel@tonic-gate */ 540ae115bc7Smrj mmu.pae_hat = kbm_pae_support; 541ae115bc7Smrj if (kbm_nx_support) 5427c478bd9Sstevel@tonic-gate mmu.pt_nx = PT_NX; 543ae115bc7Smrj else 5447c478bd9Sstevel@tonic-gate mmu.pt_nx = 0; 5457c478bd9Sstevel@tonic-gate 5467c478bd9Sstevel@tonic-gate /* 5477c478bd9Sstevel@tonic-gate * Use CPU info to set various MMU parameters 5487c478bd9Sstevel@tonic-gate */ 5497c478bd9Sstevel@tonic-gate cpuid_get_addrsize(CPU, &pa_bits, &va_bits); 5507c478bd9Sstevel@tonic-gate 5517c478bd9Sstevel@tonic-gate if (va_bits < sizeof (void *) * NBBY) { 5527c478bd9Sstevel@tonic-gate mmu.hole_start = (1ul << (va_bits - 1)); 5537c478bd9Sstevel@tonic-gate mmu.hole_end = 0ul - mmu.hole_start - 1; 5547c478bd9Sstevel@tonic-gate } else { 5557c478bd9Sstevel@tonic-gate mmu.hole_end = 0; 5567c478bd9Sstevel@tonic-gate mmu.hole_start = mmu.hole_end - 1; 5577c478bd9Sstevel@tonic-gate } 5587c478bd9Sstevel@tonic-gate #if defined(OPTERON_ERRATUM_121) 5597c478bd9Sstevel@tonic-gate /* 5607c478bd9Sstevel@tonic-gate * If erratum 121 has already been detected at this time, hole_start 5617c478bd9Sstevel@tonic-gate * contains the value to be subtracted from mmu.hole_start. 5627c478bd9Sstevel@tonic-gate */ 5637c478bd9Sstevel@tonic-gate ASSERT(hole_start == 0 || opteron_erratum_121 != 0); 5647c478bd9Sstevel@tonic-gate hole_start = mmu.hole_start - hole_start; 5657c478bd9Sstevel@tonic-gate #else 5667c478bd9Sstevel@tonic-gate hole_start = mmu.hole_start; 5677c478bd9Sstevel@tonic-gate #endif 5687c478bd9Sstevel@tonic-gate hole_end = mmu.hole_end; 5697c478bd9Sstevel@tonic-gate 5707c478bd9Sstevel@tonic-gate mmu.highest_pfn = mmu_btop((1ull << pa_bits) - 1); 5717c478bd9Sstevel@tonic-gate if (mmu.pae_hat == 0 && pa_bits > 32) 5727c478bd9Sstevel@tonic-gate mmu.highest_pfn = PFN_4G - 1; 5737c478bd9Sstevel@tonic-gate 5747c478bd9Sstevel@tonic-gate if (mmu.pae_hat) { 5757c478bd9Sstevel@tonic-gate mmu.pte_size = 8; /* 8 byte PTEs */ 5767c478bd9Sstevel@tonic-gate mmu.pte_size_shift = 3; 5777c478bd9Sstevel@tonic-gate } else { 5787c478bd9Sstevel@tonic-gate mmu.pte_size = 4; /* 4 byte PTEs */ 5797c478bd9Sstevel@tonic-gate mmu.pte_size_shift = 2; 5807c478bd9Sstevel@tonic-gate } 5817c478bd9Sstevel@tonic-gate 5827417cfdeSKuriakose Kuruvilla if (mmu.pae_hat && !is_x86_feature(x86_featureset, X86FSET_PAE)) 5837c478bd9Sstevel@tonic-gate panic("Processor does not support PAE"); 5847c478bd9Sstevel@tonic-gate 5857417cfdeSKuriakose Kuruvilla if (!is_x86_feature(x86_featureset, X86FSET_CX8)) 5867c478bd9Sstevel@tonic-gate panic("Processor does not support cmpxchg8b instruction"); 5877c478bd9Sstevel@tonic-gate 5887c478bd9Sstevel@tonic-gate #if defined(__amd64) 5897c478bd9Sstevel@tonic-gate 5907c478bd9Sstevel@tonic-gate mmu.num_level = 4; 5917c478bd9Sstevel@tonic-gate mmu.max_level = 3; 5927c478bd9Sstevel@tonic-gate mmu.ptes_per_table = 512; 5937c478bd9Sstevel@tonic-gate mmu.top_level_count = 512; 5947c478bd9Sstevel@tonic-gate 5957c478bd9Sstevel@tonic-gate mmu.level_shift[0] = 12; 5967c478bd9Sstevel@tonic-gate mmu.level_shift[1] = 21; 5977c478bd9Sstevel@tonic-gate mmu.level_shift[2] = 30; 5987c478bd9Sstevel@tonic-gate mmu.level_shift[3] = 39; 5997c478bd9Sstevel@tonic-gate 6007c478bd9Sstevel@tonic-gate #elif defined(__i386) 6017c478bd9Sstevel@tonic-gate 6027c478bd9Sstevel@tonic-gate if (mmu.pae_hat) { 6037c478bd9Sstevel@tonic-gate mmu.num_level = 3; 6047c478bd9Sstevel@tonic-gate mmu.max_level = 2; 6057c478bd9Sstevel@tonic-gate mmu.ptes_per_table = 512; 6067c478bd9Sstevel@tonic-gate mmu.top_level_count = 4; 6077c478bd9Sstevel@tonic-gate 6087c478bd9Sstevel@tonic-gate mmu.level_shift[0] = 12; 6097c478bd9Sstevel@tonic-gate mmu.level_shift[1] = 21; 6107c478bd9Sstevel@tonic-gate mmu.level_shift[2] = 30; 6117c478bd9Sstevel@tonic-gate 6127c478bd9Sstevel@tonic-gate } else { 6137c478bd9Sstevel@tonic-gate mmu.num_level = 2; 6147c478bd9Sstevel@tonic-gate mmu.max_level = 1; 6157c478bd9Sstevel@tonic-gate mmu.ptes_per_table = 1024; 6167c478bd9Sstevel@tonic-gate mmu.top_level_count = 1024; 6177c478bd9Sstevel@tonic-gate 6187c478bd9Sstevel@tonic-gate mmu.level_shift[0] = 12; 6197c478bd9Sstevel@tonic-gate mmu.level_shift[1] = 22; 6207c478bd9Sstevel@tonic-gate } 6217c478bd9Sstevel@tonic-gate 6227c478bd9Sstevel@tonic-gate #endif /* __i386 */ 6237c478bd9Sstevel@tonic-gate 6247c478bd9Sstevel@tonic-gate for (i = 0; i < mmu.num_level; ++i) { 6257c478bd9Sstevel@tonic-gate mmu.level_size[i] = 1UL << mmu.level_shift[i]; 6267c478bd9Sstevel@tonic-gate mmu.level_offset[i] = mmu.level_size[i] - 1; 6277c478bd9Sstevel@tonic-gate mmu.level_mask[i] = ~mmu.level_offset[i]; 6287c478bd9Sstevel@tonic-gate } 6297c478bd9Sstevel@tonic-gate 630512cf780Skchow set_max_page_level(); 631512cf780Skchow 632472714d6Skchow mmu_page_sizes = mmu.max_page_level + 1; 633472714d6Skchow mmu_exported_page_sizes = mmu.umax_page_level + 1; 634472714d6Skchow 635472714d6Skchow /* restrict legacy applications from using pagesizes 1g and above */ 636472714d6Skchow mmu_legacy_page_sizes = 637472714d6Skchow (mmu_exported_page_sizes > 2) ? 2 : mmu_exported_page_sizes; 638472714d6Skchow 639472714d6Skchow 640ae115bc7Smrj for (i = 0; i <= mmu.max_page_level; ++i) { 641ab4a9bebSjohnlev mmu.pte_bits[i] = PT_VALID | pt_kern; 642ae115bc7Smrj if (i > 0) 643ae115bc7Smrj mmu.pte_bits[i] |= PT_PAGESIZE; 644ae115bc7Smrj } 6457c478bd9Sstevel@tonic-gate 6467c478bd9Sstevel@tonic-gate /* 6477c478bd9Sstevel@tonic-gate * NOTE Legacy 32 bit PAE mode only has the P_VALID bit at top level. 6487c478bd9Sstevel@tonic-gate */ 6497c478bd9Sstevel@tonic-gate for (i = 1; i < mmu.num_level; ++i) 6507c478bd9Sstevel@tonic-gate mmu.ptp_bits[i] = PT_PTPBITS; 651ae115bc7Smrj 6527c478bd9Sstevel@tonic-gate #if defined(__i386) 6537c478bd9Sstevel@tonic-gate mmu.ptp_bits[2] = PT_VALID; 6547c478bd9Sstevel@tonic-gate #endif 6557c478bd9Sstevel@tonic-gate 6567c478bd9Sstevel@tonic-gate /* 6577c478bd9Sstevel@tonic-gate * Compute how many hash table entries to have per process for htables. 6587c478bd9Sstevel@tonic-gate * We start with 1 page's worth of entries. 6597c478bd9Sstevel@tonic-gate * 6607c478bd9Sstevel@tonic-gate * If physical memory is small, reduce the amount need to cover it. 6617c478bd9Sstevel@tonic-gate */ 6627c478bd9Sstevel@tonic-gate max_htables = physmax / mmu.ptes_per_table; 6637c478bd9Sstevel@tonic-gate mmu.hash_cnt = MMU_PAGESIZE / sizeof (htable_t *); 6647c478bd9Sstevel@tonic-gate while (mmu.hash_cnt > 16 && mmu.hash_cnt >= max_htables) 6657c478bd9Sstevel@tonic-gate mmu.hash_cnt >>= 1; 6667c478bd9Sstevel@tonic-gate mmu.vlp_hash_cnt = mmu.hash_cnt; 6677c478bd9Sstevel@tonic-gate 6687c478bd9Sstevel@tonic-gate #if defined(__amd64) 6697c478bd9Sstevel@tonic-gate /* 6707c478bd9Sstevel@tonic-gate * If running in 64 bits and physical memory is large, 6717c478bd9Sstevel@tonic-gate * increase the size of the cache to cover all of memory for 6727c478bd9Sstevel@tonic-gate * a 64 bit process. 6737c478bd9Sstevel@tonic-gate */ 6747c478bd9Sstevel@tonic-gate #define HASH_MAX_LENGTH 4 6757c478bd9Sstevel@tonic-gate while (mmu.hash_cnt * HASH_MAX_LENGTH < max_htables) 6767c478bd9Sstevel@tonic-gate mmu.hash_cnt <<= 1; 6777c478bd9Sstevel@tonic-gate #endif 6787c478bd9Sstevel@tonic-gate } 6797c478bd9Sstevel@tonic-gate 6807c478bd9Sstevel@tonic-gate 6817c478bd9Sstevel@tonic-gate /* 6827c478bd9Sstevel@tonic-gate * initialize hat data structures 6837c478bd9Sstevel@tonic-gate */ 6847c478bd9Sstevel@tonic-gate void 6857c478bd9Sstevel@tonic-gate hat_init() 6867c478bd9Sstevel@tonic-gate { 6877c478bd9Sstevel@tonic-gate #if defined(__i386) 6887c478bd9Sstevel@tonic-gate /* 6897c478bd9Sstevel@tonic-gate * _userlimit must be aligned correctly 6907c478bd9Sstevel@tonic-gate */ 6917c478bd9Sstevel@tonic-gate if ((_userlimit & LEVEL_MASK(1)) != _userlimit) { 6927c478bd9Sstevel@tonic-gate prom_printf("hat_init(): _userlimit=%p, not aligned at %p\n", 6937c478bd9Sstevel@tonic-gate (void *)_userlimit, (void *)LEVEL_SIZE(1)); 6947c478bd9Sstevel@tonic-gate halt("hat_init(): Unable to continue"); 6957c478bd9Sstevel@tonic-gate } 6967c478bd9Sstevel@tonic-gate #endif 6977c478bd9Sstevel@tonic-gate 6987c478bd9Sstevel@tonic-gate cv_init(&hat_list_cv, NULL, CV_DEFAULT, NULL); 6997c478bd9Sstevel@tonic-gate 7007c478bd9Sstevel@tonic-gate /* 7017c478bd9Sstevel@tonic-gate * initialize kmem caches 7027c478bd9Sstevel@tonic-gate */ 7037c478bd9Sstevel@tonic-gate htable_init(); 7047c478bd9Sstevel@tonic-gate hment_init(); 7057c478bd9Sstevel@tonic-gate 7067c478bd9Sstevel@tonic-gate hat_cache = kmem_cache_create("hat_t", 7077c478bd9Sstevel@tonic-gate sizeof (hat_t), 0, hati_constructor, NULL, NULL, 7087c478bd9Sstevel@tonic-gate NULL, 0, 0); 7097c478bd9Sstevel@tonic-gate 7107c478bd9Sstevel@tonic-gate hat_hash_cache = kmem_cache_create("HatHash", 7117c478bd9Sstevel@tonic-gate mmu.hash_cnt * sizeof (htable_t *), 0, NULL, NULL, NULL, 7127c478bd9Sstevel@tonic-gate NULL, 0, 0); 7137c478bd9Sstevel@tonic-gate 7147c478bd9Sstevel@tonic-gate /* 7157c478bd9Sstevel@tonic-gate * VLP hats can use a smaller hash table size on large memroy machines 7167c478bd9Sstevel@tonic-gate */ 7177c478bd9Sstevel@tonic-gate if (mmu.hash_cnt == mmu.vlp_hash_cnt) { 7187c478bd9Sstevel@tonic-gate vlp_hash_cache = hat_hash_cache; 7197c478bd9Sstevel@tonic-gate } else { 7207c478bd9Sstevel@tonic-gate vlp_hash_cache = kmem_cache_create("HatVlpHash", 7217c478bd9Sstevel@tonic-gate mmu.vlp_hash_cnt * sizeof (htable_t *), 0, NULL, NULL, NULL, 7227c478bd9Sstevel@tonic-gate NULL, 0, 0); 7237c478bd9Sstevel@tonic-gate } 7247c478bd9Sstevel@tonic-gate 7257c478bd9Sstevel@tonic-gate /* 7267c478bd9Sstevel@tonic-gate * Set up the kernel's hat 7277c478bd9Sstevel@tonic-gate */ 7287c478bd9Sstevel@tonic-gate AS_LOCK_ENTER(&kas, &kas.a_lock, RW_WRITER); 7297c478bd9Sstevel@tonic-gate kas.a_hat = kmem_cache_alloc(hat_cache, KM_NOSLEEP); 7307c478bd9Sstevel@tonic-gate mutex_init(&kas.a_hat->hat_mutex, NULL, MUTEX_DEFAULT, NULL); 7317c478bd9Sstevel@tonic-gate kas.a_hat->hat_as = &kas; 7327c478bd9Sstevel@tonic-gate kas.a_hat->hat_flags = 0; 7337c478bd9Sstevel@tonic-gate AS_LOCK_EXIT(&kas, &kas.a_lock); 7347c478bd9Sstevel@tonic-gate 7357c478bd9Sstevel@tonic-gate CPUSET_ZERO(khat_cpuset); 7367c478bd9Sstevel@tonic-gate CPUSET_ADD(khat_cpuset, CPU->cpu_id); 7377c478bd9Sstevel@tonic-gate 7387c478bd9Sstevel@tonic-gate /* 7397c478bd9Sstevel@tonic-gate * The kernel hat's next pointer serves as the head of the hat list . 740a85a6733Sjosephb * The kernel hat's prev pointer tracks the last hat on the list for 741a85a6733Sjosephb * htable_steal() to use. 7427c478bd9Sstevel@tonic-gate */ 7437c478bd9Sstevel@tonic-gate kas.a_hat->hat_next = NULL; 744a85a6733Sjosephb kas.a_hat->hat_prev = NULL; 7457c478bd9Sstevel@tonic-gate 7467c478bd9Sstevel@tonic-gate /* 7477c478bd9Sstevel@tonic-gate * Allocate an htable hash bucket for the kernel 7487c478bd9Sstevel@tonic-gate * XX64 - tune for 64 bit procs 7497c478bd9Sstevel@tonic-gate */ 7507c478bd9Sstevel@tonic-gate kas.a_hat->hat_num_hash = mmu.hash_cnt; 7517c478bd9Sstevel@tonic-gate kas.a_hat->hat_ht_hash = kmem_cache_alloc(hat_hash_cache, KM_NOSLEEP); 7527c478bd9Sstevel@tonic-gate bzero(kas.a_hat->hat_ht_hash, mmu.hash_cnt * sizeof (htable_t *)); 7537c478bd9Sstevel@tonic-gate 7547c478bd9Sstevel@tonic-gate /* 7557c478bd9Sstevel@tonic-gate * zero out the top level and cached htable pointers 7567c478bd9Sstevel@tonic-gate */ 7577c478bd9Sstevel@tonic-gate kas.a_hat->hat_ht_cached = NULL; 7587c478bd9Sstevel@tonic-gate kas.a_hat->hat_htable = NULL; 7599d9461f9Strevtom 7609d9461f9Strevtom /* 7619d9461f9Strevtom * Pre-allocate hrm_hashtab before enabling the collection of 7629d9461f9Strevtom * refmod statistics. Allocating on the fly would mean us 7639d9461f9Strevtom * running the risk of suffering recursive mutex enters or 7649d9461f9Strevtom * deadlocks. 7659d9461f9Strevtom */ 7669d9461f9Strevtom hrm_hashtab = kmem_zalloc(HRM_HASHSIZE * sizeof (struct hrmstat *), 7679d9461f9Strevtom KM_SLEEP); 7687c478bd9Sstevel@tonic-gate } 7697c478bd9Sstevel@tonic-gate 7707c478bd9Sstevel@tonic-gate /* 7717c478bd9Sstevel@tonic-gate * Prepare CPU specific pagetables for VLP processes on 64 bit kernels. 7727c478bd9Sstevel@tonic-gate * 7737c478bd9Sstevel@tonic-gate * Each CPU has a set of 2 pagetables that are reused for any 32 bit 7747c478bd9Sstevel@tonic-gate * process it runs. They are the top level pagetable, hci_vlp_l3ptes, and 7757c478bd9Sstevel@tonic-gate * the next to top level table for the bottom 512 Gig, hci_vlp_l2ptes. 7767c478bd9Sstevel@tonic-gate */ 7777c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 7787c478bd9Sstevel@tonic-gate static void 7797c478bd9Sstevel@tonic-gate hat_vlp_setup(struct cpu *cpu) 7807c478bd9Sstevel@tonic-gate { 781843e1988Sjohnlev #if defined(__amd64) && !defined(__xpv) 7827c478bd9Sstevel@tonic-gate struct hat_cpu_info *hci = cpu->cpu_hat_info; 7837c478bd9Sstevel@tonic-gate pfn_t pfn; 7847c478bd9Sstevel@tonic-gate 7857c478bd9Sstevel@tonic-gate /* 7867c478bd9Sstevel@tonic-gate * allocate the level==2 page table for the bottom most 7877c478bd9Sstevel@tonic-gate * 512Gig of address space (this is where 32 bit apps live) 7887c478bd9Sstevel@tonic-gate */ 7897c478bd9Sstevel@tonic-gate ASSERT(hci != NULL); 7907c478bd9Sstevel@tonic-gate hci->hci_vlp_l2ptes = kmem_zalloc(MMU_PAGESIZE, KM_SLEEP); 7917c478bd9Sstevel@tonic-gate 7927c478bd9Sstevel@tonic-gate /* 7937c478bd9Sstevel@tonic-gate * Allocate a top level pagetable and copy the kernel's 7947c478bd9Sstevel@tonic-gate * entries into it. Then link in hci_vlp_l2ptes in the 1st entry. 7957c478bd9Sstevel@tonic-gate */ 7967c478bd9Sstevel@tonic-gate hci->hci_vlp_l3ptes = kmem_zalloc(MMU_PAGESIZE, KM_SLEEP); 7977c478bd9Sstevel@tonic-gate hci->hci_vlp_pfn = 7987c478bd9Sstevel@tonic-gate hat_getpfnum(kas.a_hat, (caddr_t)hci->hci_vlp_l3ptes); 7997c478bd9Sstevel@tonic-gate ASSERT(hci->hci_vlp_pfn != PFN_INVALID); 800843e1988Sjohnlev bcopy(vlp_page, hci->hci_vlp_l3ptes, MMU_PAGESIZE); 8017c478bd9Sstevel@tonic-gate 8027c478bd9Sstevel@tonic-gate pfn = hat_getpfnum(kas.a_hat, (caddr_t)hci->hci_vlp_l2ptes); 8037c478bd9Sstevel@tonic-gate ASSERT(pfn != PFN_INVALID); 8047c478bd9Sstevel@tonic-gate hci->hci_vlp_l3ptes[0] = MAKEPTP(pfn, 2); 805843e1988Sjohnlev #endif /* __amd64 && !__xpv */ 8067c478bd9Sstevel@tonic-gate } 8077c478bd9Sstevel@tonic-gate 808ae115bc7Smrj /*ARGSUSED*/ 809ae115bc7Smrj static void 810ae115bc7Smrj hat_vlp_teardown(cpu_t *cpu) 811ae115bc7Smrj { 812843e1988Sjohnlev #if defined(__amd64) && !defined(__xpv) 813ae115bc7Smrj struct hat_cpu_info *hci; 814ae115bc7Smrj 815ae115bc7Smrj if ((hci = cpu->cpu_hat_info) == NULL) 816ae115bc7Smrj return; 817ae115bc7Smrj if (hci->hci_vlp_l2ptes) 818ae115bc7Smrj kmem_free(hci->hci_vlp_l2ptes, MMU_PAGESIZE); 819ae115bc7Smrj if (hci->hci_vlp_l3ptes) 820ae115bc7Smrj kmem_free(hci->hci_vlp_l3ptes, MMU_PAGESIZE); 821843e1988Sjohnlev #endif 822843e1988Sjohnlev } 823843e1988Sjohnlev 824843e1988Sjohnlev #define NEXT_HKR(r, l, s, e) { \ 825843e1988Sjohnlev kernel_ranges[r].hkr_level = l; \ 826843e1988Sjohnlev kernel_ranges[r].hkr_start_va = s; \ 827843e1988Sjohnlev kernel_ranges[r].hkr_end_va = e; \ 828843e1988Sjohnlev ++r; \ 829ae115bc7Smrj } 830ae115bc7Smrj 8317c478bd9Sstevel@tonic-gate /* 8327c478bd9Sstevel@tonic-gate * Finish filling in the kernel hat. 8337c478bd9Sstevel@tonic-gate * Pre fill in all top level kernel page table entries for the kernel's 8347c478bd9Sstevel@tonic-gate * part of the address range. From this point on we can't use any new 8357c478bd9Sstevel@tonic-gate * kernel large pages if they need PTE's at max_level 836ae115bc7Smrj * 837ae115bc7Smrj * create the kmap mappings. 8387c478bd9Sstevel@tonic-gate */ 8397c478bd9Sstevel@tonic-gate void 8407c478bd9Sstevel@tonic-gate hat_init_finish(void) 8417c478bd9Sstevel@tonic-gate { 842ae115bc7Smrj size_t size; 843843e1988Sjohnlev uint_t r = 0; 844843e1988Sjohnlev uintptr_t va; 845843e1988Sjohnlev hat_kernel_range_t *rp; 8467c478bd9Sstevel@tonic-gate 8477c478bd9Sstevel@tonic-gate 8487c478bd9Sstevel@tonic-gate /* 849843e1988Sjohnlev * We are now effectively running on the kernel hat. 850843e1988Sjohnlev * Clearing use_boot_reserve shuts off using the pre-allocated boot 851843e1988Sjohnlev * reserve for all HAT allocations. From here on, the reserves are 852843e1988Sjohnlev * only used when avoiding recursion in kmem_alloc(). 8537c478bd9Sstevel@tonic-gate */ 854843e1988Sjohnlev use_boot_reserve = 0; 855843e1988Sjohnlev htable_adjust_reserve(); 8567c478bd9Sstevel@tonic-gate 8577c478bd9Sstevel@tonic-gate /* 858843e1988Sjohnlev * User HATs are initialized with copies of all kernel mappings in 859843e1988Sjohnlev * higher level page tables. Ensure that those entries exist. 8607c478bd9Sstevel@tonic-gate */ 861843e1988Sjohnlev #if defined(__amd64) 862843e1988Sjohnlev 863843e1988Sjohnlev NEXT_HKR(r, 3, kernelbase, 0); 864843e1988Sjohnlev #if defined(__xpv) 865843e1988Sjohnlev NEXT_HKR(r, 3, HYPERVISOR_VIRT_START, HYPERVISOR_VIRT_END); 866843e1988Sjohnlev #endif 867843e1988Sjohnlev 868843e1988Sjohnlev #elif defined(__i386) 869843e1988Sjohnlev 870843e1988Sjohnlev #if !defined(__xpv) 871843e1988Sjohnlev if (mmu.pae_hat) { 872843e1988Sjohnlev va = kernelbase; 873843e1988Sjohnlev if ((va & LEVEL_MASK(2)) != va) { 874843e1988Sjohnlev va = P2ROUNDUP(va, LEVEL_SIZE(2)); 875843e1988Sjohnlev NEXT_HKR(r, 1, kernelbase, va); 876843e1988Sjohnlev } 877843e1988Sjohnlev if (va != 0) 878843e1988Sjohnlev NEXT_HKR(r, 2, va, 0); 879843e1988Sjohnlev } else 880843e1988Sjohnlev #endif /* __xpv */ 881843e1988Sjohnlev NEXT_HKR(r, 1, kernelbase, 0); 882843e1988Sjohnlev 883843e1988Sjohnlev #endif /* __i386 */ 884843e1988Sjohnlev 885843e1988Sjohnlev num_kernel_ranges = r; 8867c478bd9Sstevel@tonic-gate 8877c478bd9Sstevel@tonic-gate /* 888843e1988Sjohnlev * Create all the kernel pagetables that will have entries 889843e1988Sjohnlev * shared to user HATs. 8907c478bd9Sstevel@tonic-gate */ 891843e1988Sjohnlev for (r = 0; r < num_kernel_ranges; ++r) { 892843e1988Sjohnlev rp = &kernel_ranges[r]; 893843e1988Sjohnlev for (va = rp->hkr_start_va; va != rp->hkr_end_va; 894843e1988Sjohnlev va += LEVEL_SIZE(rp->hkr_level)) { 895843e1988Sjohnlev htable_t *ht; 896843e1988Sjohnlev 897843e1988Sjohnlev if (IN_HYPERVISOR_VA(va)) 898843e1988Sjohnlev continue; 899843e1988Sjohnlev 900843e1988Sjohnlev /* can/must skip if a page mapping already exists */ 901843e1988Sjohnlev if (rp->hkr_level <= mmu.max_page_level && 902843e1988Sjohnlev (ht = htable_getpage(kas.a_hat, va, NULL)) != 903843e1988Sjohnlev NULL) { 904843e1988Sjohnlev htable_release(ht); 905843e1988Sjohnlev continue; 906843e1988Sjohnlev } 907843e1988Sjohnlev 908843e1988Sjohnlev (void) htable_create(kas.a_hat, va, rp->hkr_level - 1, 909843e1988Sjohnlev NULL); 910843e1988Sjohnlev } 911843e1988Sjohnlev } 9127c478bd9Sstevel@tonic-gate 9137c478bd9Sstevel@tonic-gate /* 914843e1988Sjohnlev * 32 bit PAE metal kernels use only 4 of the 512 entries in the 915843e1988Sjohnlev * page holding the top level pagetable. We use the remainder for 916843e1988Sjohnlev * the "per CPU" page tables for VLP processes. 917843e1988Sjohnlev * Map the top level kernel pagetable into the kernel to make 918843e1988Sjohnlev * it easy to use bcopy access these tables. 9197c478bd9Sstevel@tonic-gate */ 9207c478bd9Sstevel@tonic-gate if (mmu.pae_hat) { 9217c478bd9Sstevel@tonic-gate vlp_page = vmem_alloc(heap_arena, MMU_PAGESIZE, VM_SLEEP); 9227c478bd9Sstevel@tonic-gate hat_devload(kas.a_hat, (caddr_t)vlp_page, MMU_PAGESIZE, 9237c478bd9Sstevel@tonic-gate kas.a_hat->hat_htable->ht_pfn, 924843e1988Sjohnlev #if !defined(__xpv) 925ae115bc7Smrj PROT_WRITE | 926843e1988Sjohnlev #endif 927ae115bc7Smrj PROT_READ | HAT_NOSYNC | HAT_UNORDERED_OK, 9287c478bd9Sstevel@tonic-gate HAT_LOAD | HAT_LOAD_NOCONSIST); 9297c478bd9Sstevel@tonic-gate } 9307c478bd9Sstevel@tonic-gate hat_vlp_setup(CPU); 931ae115bc7Smrj 932ae115bc7Smrj /* 933ae115bc7Smrj * Create kmap (cached mappings of kernel PTEs) 934ae115bc7Smrj * for 32 bit we map from segmap_start .. ekernelheap 935ae115bc7Smrj * for 64 bit we map from segmap_start .. segmap_start + segmapsize; 936ae115bc7Smrj */ 937ae115bc7Smrj #if defined(__i386) 938ae115bc7Smrj size = (uintptr_t)ekernelheap - segmap_start; 939ae115bc7Smrj #elif defined(__amd64) 940ae115bc7Smrj size = segmapsize; 941ae115bc7Smrj #endif 942ae115bc7Smrj hat_kmap_init((uintptr_t)segmap_start, size); 9437c478bd9Sstevel@tonic-gate } 9447c478bd9Sstevel@tonic-gate 9457c478bd9Sstevel@tonic-gate /* 9467c478bd9Sstevel@tonic-gate * On 32 bit PAE mode, PTE's are 64 bits, but ordinary atomic memory references 9477c478bd9Sstevel@tonic-gate * are 32 bit, so for safety we must use cas64() to install these. 9487c478bd9Sstevel@tonic-gate */ 9497c478bd9Sstevel@tonic-gate #ifdef __i386 9507c478bd9Sstevel@tonic-gate static void 9517c478bd9Sstevel@tonic-gate reload_pae32(hat_t *hat, cpu_t *cpu) 9527c478bd9Sstevel@tonic-gate { 9537c478bd9Sstevel@tonic-gate x86pte_t *src; 9547c478bd9Sstevel@tonic-gate x86pte_t *dest; 9557c478bd9Sstevel@tonic-gate x86pte_t pte; 9567c478bd9Sstevel@tonic-gate int i; 9577c478bd9Sstevel@tonic-gate 9587c478bd9Sstevel@tonic-gate /* 9597c478bd9Sstevel@tonic-gate * Load the 4 entries of the level 2 page table into this 9607c478bd9Sstevel@tonic-gate * cpu's range of the vlp_page and point cr3 at them. 9617c478bd9Sstevel@tonic-gate */ 9627c478bd9Sstevel@tonic-gate ASSERT(mmu.pae_hat); 9637c478bd9Sstevel@tonic-gate src = hat->hat_vlp_ptes; 9647c478bd9Sstevel@tonic-gate dest = vlp_page + (cpu->cpu_id + 1) * VLP_NUM_PTES; 9657c478bd9Sstevel@tonic-gate for (i = 0; i < VLP_NUM_PTES; ++i) { 9667c478bd9Sstevel@tonic-gate for (;;) { 9677c478bd9Sstevel@tonic-gate pte = dest[i]; 9687c478bd9Sstevel@tonic-gate if (pte == src[i]) 9697c478bd9Sstevel@tonic-gate break; 9707c478bd9Sstevel@tonic-gate if (cas64(dest + i, pte, src[i]) != src[i]) 9717c478bd9Sstevel@tonic-gate break; 9727c478bd9Sstevel@tonic-gate } 9737c478bd9Sstevel@tonic-gate } 9747c478bd9Sstevel@tonic-gate } 9757c478bd9Sstevel@tonic-gate #endif 9767c478bd9Sstevel@tonic-gate 9777c478bd9Sstevel@tonic-gate /* 9787c478bd9Sstevel@tonic-gate * Switch to a new active hat, maintaining bit masks to track active CPUs. 979843e1988Sjohnlev * 980843e1988Sjohnlev * On the 32-bit PAE hypervisor, %cr3 is a 64-bit value, on metal it 981843e1988Sjohnlev * remains a 32-bit value. 9827c478bd9Sstevel@tonic-gate */ 9837c478bd9Sstevel@tonic-gate void 9847c478bd9Sstevel@tonic-gate hat_switch(hat_t *hat) 9857c478bd9Sstevel@tonic-gate { 986843e1988Sjohnlev uint64_t newcr3; 9877c478bd9Sstevel@tonic-gate cpu_t *cpu = CPU; 9887c478bd9Sstevel@tonic-gate hat_t *old = cpu->cpu_current_hat; 9897c478bd9Sstevel@tonic-gate 9907c478bd9Sstevel@tonic-gate /* 9917c478bd9Sstevel@tonic-gate * set up this information first, so we don't miss any cross calls 9927c478bd9Sstevel@tonic-gate */ 9937c478bd9Sstevel@tonic-gate if (old != NULL) { 9947c478bd9Sstevel@tonic-gate if (old == hat) 9957c478bd9Sstevel@tonic-gate return; 9967c478bd9Sstevel@tonic-gate if (old != kas.a_hat) 9977c478bd9Sstevel@tonic-gate CPUSET_ATOMIC_DEL(old->hat_cpus, cpu->cpu_id); 9987c478bd9Sstevel@tonic-gate } 9997c478bd9Sstevel@tonic-gate 10007c478bd9Sstevel@tonic-gate /* 100195c0a3c8Sjosephb * Add this CPU to the active set for this HAT. 10027c478bd9Sstevel@tonic-gate */ 10037c478bd9Sstevel@tonic-gate if (hat != kas.a_hat) { 10047c478bd9Sstevel@tonic-gate CPUSET_ATOMIC_ADD(hat->hat_cpus, cpu->cpu_id); 10057c478bd9Sstevel@tonic-gate } 10067c478bd9Sstevel@tonic-gate cpu->cpu_current_hat = hat; 10077c478bd9Sstevel@tonic-gate 10087c478bd9Sstevel@tonic-gate /* 10097c478bd9Sstevel@tonic-gate * now go ahead and load cr3 10107c478bd9Sstevel@tonic-gate */ 10117c478bd9Sstevel@tonic-gate if (hat->hat_flags & HAT_VLP) { 10127c478bd9Sstevel@tonic-gate #if defined(__amd64) 10137c478bd9Sstevel@tonic-gate x86pte_t *vlpptep = cpu->cpu_hat_info->hci_vlp_l2ptes; 10147c478bd9Sstevel@tonic-gate 10157c478bd9Sstevel@tonic-gate VLP_COPY(hat->hat_vlp_ptes, vlpptep); 10167c478bd9Sstevel@tonic-gate newcr3 = MAKECR3(cpu->cpu_hat_info->hci_vlp_pfn); 10177c478bd9Sstevel@tonic-gate #elif defined(__i386) 10187c478bd9Sstevel@tonic-gate reload_pae32(hat, cpu); 10197c478bd9Sstevel@tonic-gate newcr3 = MAKECR3(kas.a_hat->hat_htable->ht_pfn) + 10207c478bd9Sstevel@tonic-gate (cpu->cpu_id + 1) * VLP_SIZE; 10217c478bd9Sstevel@tonic-gate #endif 10227c478bd9Sstevel@tonic-gate } else { 1023843e1988Sjohnlev newcr3 = MAKECR3((uint64_t)hat->hat_htable->ht_pfn); 10247c478bd9Sstevel@tonic-gate } 1025843e1988Sjohnlev #ifdef __xpv 1026843e1988Sjohnlev { 1027843e1988Sjohnlev struct mmuext_op t[2]; 1028843e1988Sjohnlev uint_t retcnt; 1029843e1988Sjohnlev uint_t opcnt = 1; 1030843e1988Sjohnlev 1031843e1988Sjohnlev t[0].cmd = MMUEXT_NEW_BASEPTR; 1032843e1988Sjohnlev t[0].arg1.mfn = mmu_btop(pa_to_ma(newcr3)); 1033843e1988Sjohnlev #if defined(__amd64) 1034843e1988Sjohnlev /* 1035843e1988Sjohnlev * There's an interesting problem here, as to what to 1036843e1988Sjohnlev * actually specify when switching to the kernel hat. 1037843e1988Sjohnlev * For now we'll reuse the kernel hat again. 1038843e1988Sjohnlev */ 1039843e1988Sjohnlev t[1].cmd = MMUEXT_NEW_USER_BASEPTR; 1040843e1988Sjohnlev if (hat == kas.a_hat) 1041843e1988Sjohnlev t[1].arg1.mfn = mmu_btop(pa_to_ma(newcr3)); 1042843e1988Sjohnlev else 1043843e1988Sjohnlev t[1].arg1.mfn = pfn_to_mfn(hat->hat_user_ptable); 1044843e1988Sjohnlev ++opcnt; 1045843e1988Sjohnlev #endif /* __amd64 */ 1046843e1988Sjohnlev if (HYPERVISOR_mmuext_op(t, opcnt, &retcnt, DOMID_SELF) < 0) 1047843e1988Sjohnlev panic("HYPERVISOR_mmu_update() failed"); 1048843e1988Sjohnlev ASSERT(retcnt == opcnt); 1049843e1988Sjohnlev 1050843e1988Sjohnlev } 1051843e1988Sjohnlev #else 10527c478bd9Sstevel@tonic-gate setcr3(newcr3); 1053843e1988Sjohnlev #endif 10547c478bd9Sstevel@tonic-gate ASSERT(cpu == CPU); 10557c478bd9Sstevel@tonic-gate } 10567c478bd9Sstevel@tonic-gate 10577c478bd9Sstevel@tonic-gate /* 10587c478bd9Sstevel@tonic-gate * Utility to return a valid x86pte_t from protections, pfn, and level number 10597c478bd9Sstevel@tonic-gate */ 10607c478bd9Sstevel@tonic-gate static x86pte_t 10617c478bd9Sstevel@tonic-gate hati_mkpte(pfn_t pfn, uint_t attr, level_t level, uint_t flags) 10627c478bd9Sstevel@tonic-gate { 10637c478bd9Sstevel@tonic-gate x86pte_t pte; 10647c478bd9Sstevel@tonic-gate uint_t cache_attr = attr & HAT_ORDER_MASK; 10657c478bd9Sstevel@tonic-gate 10667c478bd9Sstevel@tonic-gate pte = MAKEPTE(pfn, level); 10677c478bd9Sstevel@tonic-gate 10687c478bd9Sstevel@tonic-gate if (attr & PROT_WRITE) 10697c478bd9Sstevel@tonic-gate PTE_SET(pte, PT_WRITABLE); 10707c478bd9Sstevel@tonic-gate 10717c478bd9Sstevel@tonic-gate if (attr & PROT_USER) 10727c478bd9Sstevel@tonic-gate PTE_SET(pte, PT_USER); 10737c478bd9Sstevel@tonic-gate 10747c478bd9Sstevel@tonic-gate if (!(attr & PROT_EXEC)) 10757c478bd9Sstevel@tonic-gate PTE_SET(pte, mmu.pt_nx); 10767c478bd9Sstevel@tonic-gate 10777c478bd9Sstevel@tonic-gate /* 1078ae115bc7Smrj * Set the software bits used track ref/mod sync's and hments. 1079ae115bc7Smrj * If not using REF/MOD, set them to avoid h/w rewriting PTEs. 10807c478bd9Sstevel@tonic-gate */ 10817c478bd9Sstevel@tonic-gate if (flags & HAT_LOAD_NOCONSIST) 1082ae115bc7Smrj PTE_SET(pte, PT_NOCONSIST | PT_REF | PT_MOD); 1083ae115bc7Smrj else if (attr & HAT_NOSYNC) 1084ae115bc7Smrj PTE_SET(pte, PT_NOSYNC | PT_REF | PT_MOD); 10857c478bd9Sstevel@tonic-gate 10867c478bd9Sstevel@tonic-gate /* 10877c478bd9Sstevel@tonic-gate * Set the caching attributes in the PTE. The combination 10887c478bd9Sstevel@tonic-gate * of attributes are poorly defined, so we pay attention 10897c478bd9Sstevel@tonic-gate * to them in the given order. 10907c478bd9Sstevel@tonic-gate * 10917c478bd9Sstevel@tonic-gate * The test for HAT_STRICTORDER is different because it's defined 10927c478bd9Sstevel@tonic-gate * as "0" - which was a stupid thing to do, but is too late to change! 10937c478bd9Sstevel@tonic-gate */ 10947c478bd9Sstevel@tonic-gate if (cache_attr == HAT_STRICTORDER) { 10957c478bd9Sstevel@tonic-gate PTE_SET(pte, PT_NOCACHE); 10967c478bd9Sstevel@tonic-gate /*LINTED [Lint hates empty ifs, but it's the obvious way to do this] */ 10977c478bd9Sstevel@tonic-gate } else if (cache_attr & (HAT_UNORDERED_OK | HAT_STORECACHING_OK)) { 10987c478bd9Sstevel@tonic-gate /* nothing to set */; 10997c478bd9Sstevel@tonic-gate } else if (cache_attr & (HAT_MERGING_OK | HAT_LOADCACHING_OK)) { 11007c478bd9Sstevel@tonic-gate PTE_SET(pte, PT_NOCACHE); 11017417cfdeSKuriakose Kuruvilla if (is_x86_feature(x86_featureset, X86FSET_PAT)) 11027c478bd9Sstevel@tonic-gate PTE_SET(pte, (level == 0) ? PT_PAT_4K : PT_PAT_LARGE); 11037c478bd9Sstevel@tonic-gate else 11047c478bd9Sstevel@tonic-gate PTE_SET(pte, PT_WRITETHRU); 11057c478bd9Sstevel@tonic-gate } else { 11067c478bd9Sstevel@tonic-gate panic("hati_mkpte(): bad caching attributes: %x\n", cache_attr); 11077c478bd9Sstevel@tonic-gate } 11087c478bd9Sstevel@tonic-gate 11097c478bd9Sstevel@tonic-gate return (pte); 11107c478bd9Sstevel@tonic-gate } 11117c478bd9Sstevel@tonic-gate 11127c478bd9Sstevel@tonic-gate /* 11137c478bd9Sstevel@tonic-gate * Duplicate address translations of the parent to the child. 11147c478bd9Sstevel@tonic-gate * This function really isn't used anymore. 11157c478bd9Sstevel@tonic-gate */ 11167c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 11177c478bd9Sstevel@tonic-gate int 11187c478bd9Sstevel@tonic-gate hat_dup(hat_t *old, hat_t *new, caddr_t addr, size_t len, uint_t flag) 11197c478bd9Sstevel@tonic-gate { 11207c478bd9Sstevel@tonic-gate ASSERT((uintptr_t)addr < kernelbase); 11217c478bd9Sstevel@tonic-gate ASSERT(new != kas.a_hat); 11227c478bd9Sstevel@tonic-gate ASSERT(old != kas.a_hat); 11237c478bd9Sstevel@tonic-gate return (0); 11247c478bd9Sstevel@tonic-gate } 11257c478bd9Sstevel@tonic-gate 11267c478bd9Sstevel@tonic-gate /* 11277c478bd9Sstevel@tonic-gate * Allocate any hat resources required for a process being swapped in. 11287c478bd9Sstevel@tonic-gate */ 11297c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 11307c478bd9Sstevel@tonic-gate void 11317c478bd9Sstevel@tonic-gate hat_swapin(hat_t *hat) 11327c478bd9Sstevel@tonic-gate { 11337c478bd9Sstevel@tonic-gate /* do nothing - we let everything fault back in */ 11347c478bd9Sstevel@tonic-gate } 11357c478bd9Sstevel@tonic-gate 11367c478bd9Sstevel@tonic-gate /* 11377c478bd9Sstevel@tonic-gate * Unload all translations associated with an address space of a process 11387c478bd9Sstevel@tonic-gate * that is being swapped out. 11397c478bd9Sstevel@tonic-gate */ 11407c478bd9Sstevel@tonic-gate void 11417c478bd9Sstevel@tonic-gate hat_swapout(hat_t *hat) 11427c478bd9Sstevel@tonic-gate { 11437c478bd9Sstevel@tonic-gate uintptr_t vaddr = (uintptr_t)0; 11447c478bd9Sstevel@tonic-gate uintptr_t eaddr = _userlimit; 11457c478bd9Sstevel@tonic-gate htable_t *ht = NULL; 11467c478bd9Sstevel@tonic-gate level_t l; 11477c478bd9Sstevel@tonic-gate 1148843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 11497c478bd9Sstevel@tonic-gate /* 11507c478bd9Sstevel@tonic-gate * We can't just call hat_unload(hat, 0, _userlimit...) here, because 11517c478bd9Sstevel@tonic-gate * seg_spt and shared pagetables can't be swapped out. 11527c478bd9Sstevel@tonic-gate * Take a look at segspt_shmswapout() - it's a big no-op. 11537c478bd9Sstevel@tonic-gate * 11547c478bd9Sstevel@tonic-gate * Instead we'll walk through all the address space and unload 11557c478bd9Sstevel@tonic-gate * any mappings which we are sure are not shared, not locked. 11567c478bd9Sstevel@tonic-gate */ 11577c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(vaddr)); 11587c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(eaddr)); 11597c478bd9Sstevel@tonic-gate ASSERT(AS_LOCK_HELD(hat->hat_as, &hat->hat_as->a_lock)); 11607c478bd9Sstevel@tonic-gate if ((uintptr_t)hat->hat_as->a_userlimit < eaddr) 11617c478bd9Sstevel@tonic-gate eaddr = (uintptr_t)hat->hat_as->a_userlimit; 11627c478bd9Sstevel@tonic-gate 11637c478bd9Sstevel@tonic-gate while (vaddr < eaddr) { 11647c478bd9Sstevel@tonic-gate (void) htable_walk(hat, &ht, &vaddr, eaddr); 11657c478bd9Sstevel@tonic-gate if (ht == NULL) 11667c478bd9Sstevel@tonic-gate break; 11677c478bd9Sstevel@tonic-gate 11687c478bd9Sstevel@tonic-gate ASSERT(!IN_VA_HOLE(vaddr)); 11697c478bd9Sstevel@tonic-gate 11707c478bd9Sstevel@tonic-gate /* 11717c478bd9Sstevel@tonic-gate * If the page table is shared skip its entire range. 11727c478bd9Sstevel@tonic-gate */ 11737c478bd9Sstevel@tonic-gate l = ht->ht_level; 11747c478bd9Sstevel@tonic-gate if (ht->ht_flags & HTABLE_SHARED_PFN) { 1175ae320ee6Speterte vaddr = ht->ht_vaddr + LEVEL_SIZE(l + 1); 11767c478bd9Sstevel@tonic-gate htable_release(ht); 11777c478bd9Sstevel@tonic-gate ht = NULL; 11787c478bd9Sstevel@tonic-gate continue; 11797c478bd9Sstevel@tonic-gate } 11807c478bd9Sstevel@tonic-gate 11817c478bd9Sstevel@tonic-gate /* 11827c478bd9Sstevel@tonic-gate * If the page table has no locked entries, unload this one. 11837c478bd9Sstevel@tonic-gate */ 11847c478bd9Sstevel@tonic-gate if (ht->ht_lock_cnt == 0) 11857c478bd9Sstevel@tonic-gate hat_unload(hat, (caddr_t)vaddr, LEVEL_SIZE(l), 11867c478bd9Sstevel@tonic-gate HAT_UNLOAD_UNMAP); 11877c478bd9Sstevel@tonic-gate 11887c478bd9Sstevel@tonic-gate /* 11897c478bd9Sstevel@tonic-gate * If we have a level 0 page table with locked entries, 11907c478bd9Sstevel@tonic-gate * skip the entire page table, otherwise skip just one entry. 11917c478bd9Sstevel@tonic-gate */ 11927c478bd9Sstevel@tonic-gate if (ht->ht_lock_cnt > 0 && l == 0) 11937c478bd9Sstevel@tonic-gate vaddr = ht->ht_vaddr + LEVEL_SIZE(1); 11947c478bd9Sstevel@tonic-gate else 11957c478bd9Sstevel@tonic-gate vaddr += LEVEL_SIZE(l); 11967c478bd9Sstevel@tonic-gate } 11977c478bd9Sstevel@tonic-gate if (ht) 11987c478bd9Sstevel@tonic-gate htable_release(ht); 11997c478bd9Sstevel@tonic-gate 12007c478bd9Sstevel@tonic-gate /* 12017c478bd9Sstevel@tonic-gate * We're in swapout because the system is low on memory, so 12027c478bd9Sstevel@tonic-gate * go back and flush all the htables off the cached list. 12037c478bd9Sstevel@tonic-gate */ 12047c478bd9Sstevel@tonic-gate htable_purge_hat(hat); 1205843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 12067c478bd9Sstevel@tonic-gate } 12077c478bd9Sstevel@tonic-gate 12087c478bd9Sstevel@tonic-gate /* 12097c478bd9Sstevel@tonic-gate * returns number of bytes that have valid mappings in hat. 12107c478bd9Sstevel@tonic-gate */ 12117c478bd9Sstevel@tonic-gate size_t 12127c478bd9Sstevel@tonic-gate hat_get_mapped_size(hat_t *hat) 12137c478bd9Sstevel@tonic-gate { 12147c478bd9Sstevel@tonic-gate size_t total = 0; 12157c478bd9Sstevel@tonic-gate int l; 12167c478bd9Sstevel@tonic-gate 12177c478bd9Sstevel@tonic-gate for (l = 0; l <= mmu.max_page_level; l++) 12187c478bd9Sstevel@tonic-gate total += (hat->hat_pages_mapped[l] << LEVEL_SHIFT(l)); 1219250b7ff9Sjosephb total += hat->hat_ism_pgcnt; 12207c478bd9Sstevel@tonic-gate 12217c478bd9Sstevel@tonic-gate return (total); 12227c478bd9Sstevel@tonic-gate } 12237c478bd9Sstevel@tonic-gate 12247c478bd9Sstevel@tonic-gate /* 12257c478bd9Sstevel@tonic-gate * enable/disable collection of stats for hat. 12267c478bd9Sstevel@tonic-gate */ 12277c478bd9Sstevel@tonic-gate int 12287c478bd9Sstevel@tonic-gate hat_stats_enable(hat_t *hat) 12297c478bd9Sstevel@tonic-gate { 12307c478bd9Sstevel@tonic-gate atomic_add_32(&hat->hat_stats, 1); 12317c478bd9Sstevel@tonic-gate return (1); 12327c478bd9Sstevel@tonic-gate } 12337c478bd9Sstevel@tonic-gate 12347c478bd9Sstevel@tonic-gate void 12357c478bd9Sstevel@tonic-gate hat_stats_disable(hat_t *hat) 12367c478bd9Sstevel@tonic-gate { 12377c478bd9Sstevel@tonic-gate atomic_add_32(&hat->hat_stats, -1); 12387c478bd9Sstevel@tonic-gate } 12397c478bd9Sstevel@tonic-gate 12407c478bd9Sstevel@tonic-gate /* 12417c478bd9Sstevel@tonic-gate * Utility to sync the ref/mod bits from a page table entry to the page_t 12427c478bd9Sstevel@tonic-gate * We must be holding the mapping list lock when this is called. 12437c478bd9Sstevel@tonic-gate */ 12447c478bd9Sstevel@tonic-gate static void 12457c478bd9Sstevel@tonic-gate hati_sync_pte_to_page(page_t *pp, x86pte_t pte, level_t level) 12467c478bd9Sstevel@tonic-gate { 12477c478bd9Sstevel@tonic-gate uint_t rm = 0; 12487c478bd9Sstevel@tonic-gate pgcnt_t pgcnt; 12497c478bd9Sstevel@tonic-gate 1250ae115bc7Smrj if (PTE_GET(pte, PT_SOFTWARE) >= PT_NOSYNC) 12517c478bd9Sstevel@tonic-gate return; 12527c478bd9Sstevel@tonic-gate 12537c478bd9Sstevel@tonic-gate if (PTE_GET(pte, PT_REF)) 12547c478bd9Sstevel@tonic-gate rm |= P_REF; 12557c478bd9Sstevel@tonic-gate 12567c478bd9Sstevel@tonic-gate if (PTE_GET(pte, PT_MOD)) 12577c478bd9Sstevel@tonic-gate rm |= P_MOD; 12587c478bd9Sstevel@tonic-gate 12597c478bd9Sstevel@tonic-gate if (rm == 0) 12607c478bd9Sstevel@tonic-gate return; 12617c478bd9Sstevel@tonic-gate 12627c478bd9Sstevel@tonic-gate /* 12637c478bd9Sstevel@tonic-gate * sync to all constituent pages of a large page 12647c478bd9Sstevel@tonic-gate */ 12657c478bd9Sstevel@tonic-gate ASSERT(x86_hm_held(pp)); 12667c478bd9Sstevel@tonic-gate pgcnt = page_get_pagecnt(level); 12677c478bd9Sstevel@tonic-gate ASSERT(IS_P2ALIGNED(pp->p_pagenum, pgcnt)); 12687c478bd9Sstevel@tonic-gate for (; pgcnt > 0; --pgcnt) { 12697c478bd9Sstevel@tonic-gate /* 12707c478bd9Sstevel@tonic-gate * hat_page_demote() can't decrease 12717c478bd9Sstevel@tonic-gate * pszc below this mapping size 12727c478bd9Sstevel@tonic-gate * since this large mapping existed after we 12737c478bd9Sstevel@tonic-gate * took mlist lock. 12747c478bd9Sstevel@tonic-gate */ 12757c478bd9Sstevel@tonic-gate ASSERT(pp->p_szc >= level); 12767c478bd9Sstevel@tonic-gate hat_page_setattr(pp, rm); 12777c478bd9Sstevel@tonic-gate ++pp; 12787c478bd9Sstevel@tonic-gate } 12797c478bd9Sstevel@tonic-gate } 12807c478bd9Sstevel@tonic-gate 12817c478bd9Sstevel@tonic-gate /* 12827c478bd9Sstevel@tonic-gate * This the set of PTE bits for PFN, permissions and caching 1283843e1988Sjohnlev * that are allowed to change on a HAT_LOAD_REMAP 12847c478bd9Sstevel@tonic-gate */ 12857c478bd9Sstevel@tonic-gate #define PT_REMAP_BITS \ 12867c478bd9Sstevel@tonic-gate (PT_PADDR | PT_NX | PT_WRITABLE | PT_WRITETHRU | \ 1287843e1988Sjohnlev PT_NOCACHE | PT_PAT_4K | PT_PAT_LARGE | PT_IGNORE | PT_REF | PT_MOD) 12887c478bd9Sstevel@tonic-gate 1289b193e412Skchow #define REMAPASSERT(EX) if (!(EX)) panic("hati_pte_map: " #EX) 12907c478bd9Sstevel@tonic-gate /* 12917c478bd9Sstevel@tonic-gate * Do the low-level work to get a mapping entered into a HAT's pagetables 12927c478bd9Sstevel@tonic-gate * and in the mapping list of the associated page_t. 12937c478bd9Sstevel@tonic-gate */ 1294ae115bc7Smrj static int 12957c478bd9Sstevel@tonic-gate hati_pte_map( 12967c478bd9Sstevel@tonic-gate htable_t *ht, 12977c478bd9Sstevel@tonic-gate uint_t entry, 12987c478bd9Sstevel@tonic-gate page_t *pp, 12997c478bd9Sstevel@tonic-gate x86pte_t pte, 13007c478bd9Sstevel@tonic-gate int flags, 13017c478bd9Sstevel@tonic-gate void *pte_ptr) 13027c478bd9Sstevel@tonic-gate { 13037c478bd9Sstevel@tonic-gate hat_t *hat = ht->ht_hat; 13047c478bd9Sstevel@tonic-gate x86pte_t old_pte; 13057c478bd9Sstevel@tonic-gate level_t l = ht->ht_level; 13067c478bd9Sstevel@tonic-gate hment_t *hm; 13077c478bd9Sstevel@tonic-gate uint_t is_consist; 130825540de2SJakub Jermar uint_t is_locked; 1309ae115bc7Smrj int rv = 0; 13107c478bd9Sstevel@tonic-gate 13117c478bd9Sstevel@tonic-gate /* 131221584dbcSPavel Tatashin * Is this a consistent (ie. need mapping list lock) mapping? 13137c478bd9Sstevel@tonic-gate */ 13147c478bd9Sstevel@tonic-gate is_consist = (pp != NULL && (flags & HAT_LOAD_NOCONSIST) == 0); 13157c478bd9Sstevel@tonic-gate 13167c478bd9Sstevel@tonic-gate /* 13177c478bd9Sstevel@tonic-gate * Track locked mapping count in the htable. Do this first, 13187c478bd9Sstevel@tonic-gate * as we track locking even if there already is a mapping present. 13197c478bd9Sstevel@tonic-gate */ 132025540de2SJakub Jermar is_locked = (flags & HAT_LOAD_LOCK) != 0 && hat != kas.a_hat; 132125540de2SJakub Jermar if (is_locked) 13227c478bd9Sstevel@tonic-gate HTABLE_LOCK_INC(ht); 13237c478bd9Sstevel@tonic-gate 13247c478bd9Sstevel@tonic-gate /* 13257c478bd9Sstevel@tonic-gate * Acquire the page's mapping list lock and get an hment to use. 13267c478bd9Sstevel@tonic-gate * Note that hment_prepare() might return NULL. 13277c478bd9Sstevel@tonic-gate */ 13287c478bd9Sstevel@tonic-gate if (is_consist) { 13297c478bd9Sstevel@tonic-gate x86_hm_enter(pp); 13307c478bd9Sstevel@tonic-gate hm = hment_prepare(ht, entry, pp); 13317c478bd9Sstevel@tonic-gate } 13327c478bd9Sstevel@tonic-gate 13337c478bd9Sstevel@tonic-gate /* 13347c478bd9Sstevel@tonic-gate * Set the new pte, retrieving the old one at the same time. 13357c478bd9Sstevel@tonic-gate */ 13367c478bd9Sstevel@tonic-gate old_pte = x86pte_set(ht, entry, pte, pte_ptr); 13377c478bd9Sstevel@tonic-gate 13387c478bd9Sstevel@tonic-gate /* 133925540de2SJakub Jermar * Did we get a large page / page table collision? 13407c478bd9Sstevel@tonic-gate */ 1341ae115bc7Smrj if (old_pte == LPAGE_ERROR) { 134225540de2SJakub Jermar if (is_locked) 134325540de2SJakub Jermar HTABLE_LOCK_DEC(ht); 1344ae115bc7Smrj rv = -1; 1345ae115bc7Smrj goto done; 13467c478bd9Sstevel@tonic-gate } 13477c478bd9Sstevel@tonic-gate 1348ae115bc7Smrj /* 1349ae115bc7Smrj * If the mapping didn't change there is nothing more to do. 1350ae115bc7Smrj */ 1351ae115bc7Smrj if (PTE_EQUIV(pte, old_pte)) 1352ae115bc7Smrj goto done; 1353ae115bc7Smrj 13547c478bd9Sstevel@tonic-gate /* 13557c478bd9Sstevel@tonic-gate * Install a new mapping in the page's mapping list 13567c478bd9Sstevel@tonic-gate */ 13577c478bd9Sstevel@tonic-gate if (!PTE_ISVALID(old_pte)) { 13587c478bd9Sstevel@tonic-gate if (is_consist) { 13597c478bd9Sstevel@tonic-gate hment_assign(ht, entry, pp, hm); 13607c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 13617c478bd9Sstevel@tonic-gate } else { 13627c478bd9Sstevel@tonic-gate ASSERT(flags & HAT_LOAD_NOCONSIST); 13637c478bd9Sstevel@tonic-gate } 136402bc52beSkchow #if defined(__amd64) 136502bc52beSkchow if (ht->ht_flags & HTABLE_VLP) { 136602bc52beSkchow cpu_t *cpu = CPU; 136702bc52beSkchow x86pte_t *vlpptep = cpu->cpu_hat_info->hci_vlp_l2ptes; 136802bc52beSkchow VLP_COPY(hat->hat_vlp_ptes, vlpptep); 136902bc52beSkchow } 137002bc52beSkchow #endif 13717c478bd9Sstevel@tonic-gate HTABLE_INC(ht->ht_valid_cnt); 13727c478bd9Sstevel@tonic-gate PGCNT_INC(hat, l); 1373ae115bc7Smrj return (rv); 13747c478bd9Sstevel@tonic-gate } 13757c478bd9Sstevel@tonic-gate 13767c478bd9Sstevel@tonic-gate /* 13777c478bd9Sstevel@tonic-gate * Remap's are more complicated: 13787c478bd9Sstevel@tonic-gate * - HAT_LOAD_REMAP must be specified if changing the pfn. 13797c478bd9Sstevel@tonic-gate * We also require that NOCONSIST be specified. 13807c478bd9Sstevel@tonic-gate * - Otherwise only permission or caching bits may change. 13817c478bd9Sstevel@tonic-gate */ 13827c478bd9Sstevel@tonic-gate if (!PTE_ISPAGE(old_pte, l)) 13837c478bd9Sstevel@tonic-gate panic("non-null/page mapping pte=" FMT_PTE, old_pte); 13847c478bd9Sstevel@tonic-gate 13857c478bd9Sstevel@tonic-gate if (PTE2PFN(old_pte, l) != PTE2PFN(pte, l)) { 1386b193e412Skchow REMAPASSERT(flags & HAT_LOAD_REMAP); 1387b193e412Skchow REMAPASSERT(flags & HAT_LOAD_NOCONSIST); 1388ae115bc7Smrj REMAPASSERT(PTE_GET(old_pte, PT_SOFTWARE) >= PT_NOCONSIST); 1389b193e412Skchow REMAPASSERT(pf_is_memory(PTE2PFN(old_pte, l)) == 13907c478bd9Sstevel@tonic-gate pf_is_memory(PTE2PFN(pte, l))); 1391b193e412Skchow REMAPASSERT(!is_consist); 13927c478bd9Sstevel@tonic-gate } 13937c478bd9Sstevel@tonic-gate 13947c478bd9Sstevel@tonic-gate /* 1395843e1988Sjohnlev * We only let remaps change the certain bits in the PTE. 13967c478bd9Sstevel@tonic-gate */ 1397843e1988Sjohnlev if (PTE_GET(old_pte, ~PT_REMAP_BITS) != PTE_GET(pte, ~PT_REMAP_BITS)) 1398843e1988Sjohnlev panic("remap bits changed: old_pte="FMT_PTE", pte="FMT_PTE"\n", 1399843e1988Sjohnlev old_pte, pte); 14007c478bd9Sstevel@tonic-gate 14017c478bd9Sstevel@tonic-gate /* 14027c478bd9Sstevel@tonic-gate * We don't create any mapping list entries on a remap, so release 14037c478bd9Sstevel@tonic-gate * any allocated hment after we drop the mapping list lock. 14047c478bd9Sstevel@tonic-gate */ 1405ae115bc7Smrj done: 14067c478bd9Sstevel@tonic-gate if (is_consist) { 14077c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 14087c478bd9Sstevel@tonic-gate if (hm != NULL) 14097c478bd9Sstevel@tonic-gate hment_free(hm); 14107c478bd9Sstevel@tonic-gate } 1411ae115bc7Smrj return (rv); 14127c478bd9Sstevel@tonic-gate } 14137c478bd9Sstevel@tonic-gate 14147c478bd9Sstevel@tonic-gate /* 1415ae115bc7Smrj * Internal routine to load a single page table entry. This only fails if 1416ae115bc7Smrj * we attempt to overwrite a page table link with a large page. 14177c478bd9Sstevel@tonic-gate */ 1418ae115bc7Smrj static int 14197c478bd9Sstevel@tonic-gate hati_load_common( 14207c478bd9Sstevel@tonic-gate hat_t *hat, 14217c478bd9Sstevel@tonic-gate uintptr_t va, 14227c478bd9Sstevel@tonic-gate page_t *pp, 14237c478bd9Sstevel@tonic-gate uint_t attr, 14247c478bd9Sstevel@tonic-gate uint_t flags, 14257c478bd9Sstevel@tonic-gate level_t level, 14267c478bd9Sstevel@tonic-gate pfn_t pfn) 14277c478bd9Sstevel@tonic-gate { 14287c478bd9Sstevel@tonic-gate htable_t *ht; 14297c478bd9Sstevel@tonic-gate uint_t entry; 14307c478bd9Sstevel@tonic-gate x86pte_t pte; 1431ae115bc7Smrj int rv = 0; 14327c478bd9Sstevel@tonic-gate 1433aac11643Sjosephb /* 1434aac11643Sjosephb * The number 16 is arbitrary and here to catch a recursion problem 1435aac11643Sjosephb * early before we blow out the kernel stack. 1436aac11643Sjosephb */ 1437aac11643Sjosephb ++curthread->t_hatdepth; 1438aac11643Sjosephb ASSERT(curthread->t_hatdepth < 16); 1439aac11643Sjosephb 14407c478bd9Sstevel@tonic-gate ASSERT(hat == kas.a_hat || 14417c478bd9Sstevel@tonic-gate AS_LOCK_HELD(hat->hat_as, &hat->hat_as->a_lock)); 14427c478bd9Sstevel@tonic-gate 14437c478bd9Sstevel@tonic-gate if (flags & HAT_LOAD_SHARE) 14447c478bd9Sstevel@tonic-gate hat->hat_flags |= HAT_SHARED; 14457c478bd9Sstevel@tonic-gate 14467c478bd9Sstevel@tonic-gate /* 14477c478bd9Sstevel@tonic-gate * Find the page table that maps this page if it already exists. 14487c478bd9Sstevel@tonic-gate */ 14497c478bd9Sstevel@tonic-gate ht = htable_lookup(hat, va, level); 14507c478bd9Sstevel@tonic-gate 14517c478bd9Sstevel@tonic-gate /* 1452aac11643Sjosephb * We must have HAT_LOAD_NOCONSIST if page_t is NULL. 14537c478bd9Sstevel@tonic-gate */ 1454aac11643Sjosephb if (pp == NULL) 14557c478bd9Sstevel@tonic-gate flags |= HAT_LOAD_NOCONSIST; 14567c478bd9Sstevel@tonic-gate 14577c478bd9Sstevel@tonic-gate if (ht == NULL) { 14587c478bd9Sstevel@tonic-gate ht = htable_create(hat, va, level, NULL); 14597c478bd9Sstevel@tonic-gate ASSERT(ht != NULL); 14607c478bd9Sstevel@tonic-gate } 14617c478bd9Sstevel@tonic-gate entry = htable_va2entry(va, ht); 14627c478bd9Sstevel@tonic-gate 14637c478bd9Sstevel@tonic-gate /* 14647c478bd9Sstevel@tonic-gate * a bunch of paranoid error checking 14657c478bd9Sstevel@tonic-gate */ 14667c478bd9Sstevel@tonic-gate ASSERT(ht->ht_busy > 0); 14677c478bd9Sstevel@tonic-gate if (ht->ht_vaddr > va || va > HTABLE_LAST_PAGE(ht)) 1468903a11ebSrh panic("hati_load_common: bad htable %p, va %p", 1469903a11ebSrh (void *)ht, (void *)va); 14707c478bd9Sstevel@tonic-gate ASSERT(ht->ht_level == level); 14717c478bd9Sstevel@tonic-gate 14727c478bd9Sstevel@tonic-gate /* 14737c478bd9Sstevel@tonic-gate * construct the new PTE 14747c478bd9Sstevel@tonic-gate */ 14757c478bd9Sstevel@tonic-gate if (hat == kas.a_hat) 14767c478bd9Sstevel@tonic-gate attr &= ~PROT_USER; 14777c478bd9Sstevel@tonic-gate pte = hati_mkpte(pfn, attr, level, flags); 14787c478bd9Sstevel@tonic-gate if (hat == kas.a_hat && va >= kernelbase) 14797c478bd9Sstevel@tonic-gate PTE_SET(pte, mmu.pt_global); 14807c478bd9Sstevel@tonic-gate 14817c478bd9Sstevel@tonic-gate /* 14827c478bd9Sstevel@tonic-gate * establish the mapping 14837c478bd9Sstevel@tonic-gate */ 1484ae115bc7Smrj rv = hati_pte_map(ht, entry, pp, pte, flags, NULL); 14857c478bd9Sstevel@tonic-gate 14867c478bd9Sstevel@tonic-gate /* 14877c478bd9Sstevel@tonic-gate * release the htable and any reserves 14887c478bd9Sstevel@tonic-gate */ 14897c478bd9Sstevel@tonic-gate htable_release(ht); 1490aac11643Sjosephb --curthread->t_hatdepth; 1491ae115bc7Smrj return (rv); 14927c478bd9Sstevel@tonic-gate } 14937c478bd9Sstevel@tonic-gate 14947c478bd9Sstevel@tonic-gate /* 14957c478bd9Sstevel@tonic-gate * special case of hat_memload to deal with some kernel addrs for performance 14967c478bd9Sstevel@tonic-gate */ 14977c478bd9Sstevel@tonic-gate static void 14987c478bd9Sstevel@tonic-gate hat_kmap_load( 14997c478bd9Sstevel@tonic-gate caddr_t addr, 15007c478bd9Sstevel@tonic-gate page_t *pp, 15017c478bd9Sstevel@tonic-gate uint_t attr, 15027c478bd9Sstevel@tonic-gate uint_t flags) 15037c478bd9Sstevel@tonic-gate { 15047c478bd9Sstevel@tonic-gate uintptr_t va = (uintptr_t)addr; 15057c478bd9Sstevel@tonic-gate x86pte_t pte; 15067c478bd9Sstevel@tonic-gate pfn_t pfn = page_pptonum(pp); 15077c478bd9Sstevel@tonic-gate pgcnt_t pg_off = mmu_btop(va - mmu.kmap_addr); 15087c478bd9Sstevel@tonic-gate htable_t *ht; 15097c478bd9Sstevel@tonic-gate uint_t entry; 15107c478bd9Sstevel@tonic-gate void *pte_ptr; 15117c478bd9Sstevel@tonic-gate 15127c478bd9Sstevel@tonic-gate /* 15137c478bd9Sstevel@tonic-gate * construct the requested PTE 15147c478bd9Sstevel@tonic-gate */ 15157c478bd9Sstevel@tonic-gate attr &= ~PROT_USER; 15167c478bd9Sstevel@tonic-gate attr |= HAT_STORECACHING_OK; 15177c478bd9Sstevel@tonic-gate pte = hati_mkpte(pfn, attr, 0, flags); 15187c478bd9Sstevel@tonic-gate PTE_SET(pte, mmu.pt_global); 15197c478bd9Sstevel@tonic-gate 15207c478bd9Sstevel@tonic-gate /* 15217c478bd9Sstevel@tonic-gate * Figure out the pte_ptr and htable and use common code to finish up 15227c478bd9Sstevel@tonic-gate */ 15237c478bd9Sstevel@tonic-gate if (mmu.pae_hat) 15247c478bd9Sstevel@tonic-gate pte_ptr = mmu.kmap_ptes + pg_off; 15257c478bd9Sstevel@tonic-gate else 15267c478bd9Sstevel@tonic-gate pte_ptr = (x86pte32_t *)mmu.kmap_ptes + pg_off; 15277c478bd9Sstevel@tonic-gate ht = mmu.kmap_htables[(va - mmu.kmap_htables[0]->ht_vaddr) >> 15287c478bd9Sstevel@tonic-gate LEVEL_SHIFT(1)]; 15297c478bd9Sstevel@tonic-gate entry = htable_va2entry(va, ht); 1530aac11643Sjosephb ++curthread->t_hatdepth; 1531aac11643Sjosephb ASSERT(curthread->t_hatdepth < 16); 1532ae115bc7Smrj (void) hati_pte_map(ht, entry, pp, pte, flags, pte_ptr); 1533aac11643Sjosephb --curthread->t_hatdepth; 15347c478bd9Sstevel@tonic-gate } 15357c478bd9Sstevel@tonic-gate 15367c478bd9Sstevel@tonic-gate /* 15377c478bd9Sstevel@tonic-gate * hat_memload() - load a translation to the given page struct 15387c478bd9Sstevel@tonic-gate * 15397c478bd9Sstevel@tonic-gate * Flags for hat_memload/hat_devload/hat_*attr. 15407c478bd9Sstevel@tonic-gate * 15417c478bd9Sstevel@tonic-gate * HAT_LOAD Default flags to load a translation to the page. 15427c478bd9Sstevel@tonic-gate * 15437c478bd9Sstevel@tonic-gate * HAT_LOAD_LOCK Lock down mapping resources; hat_map(), hat_memload(), 15447c478bd9Sstevel@tonic-gate * and hat_devload(). 15457c478bd9Sstevel@tonic-gate * 15467c478bd9Sstevel@tonic-gate * HAT_LOAD_NOCONSIST Do not add mapping to page_t mapping list. 1547ae115bc7Smrj * sets PT_NOCONSIST 15487c478bd9Sstevel@tonic-gate * 15497c478bd9Sstevel@tonic-gate * HAT_LOAD_SHARE A flag to hat_memload() to indicate h/w page tables 15507c478bd9Sstevel@tonic-gate * that map some user pages (not kas) is shared by more 15517c478bd9Sstevel@tonic-gate * than one process (eg. ISM). 15527c478bd9Sstevel@tonic-gate * 15537c478bd9Sstevel@tonic-gate * HAT_LOAD_REMAP Reload a valid pte with a different page frame. 15547c478bd9Sstevel@tonic-gate * 15557c478bd9Sstevel@tonic-gate * HAT_NO_KALLOC Do not kmem_alloc while creating the mapping; at this 15567c478bd9Sstevel@tonic-gate * point, it's setting up mapping to allocate internal 15577c478bd9Sstevel@tonic-gate * hat layer data structures. This flag forces hat layer 15587c478bd9Sstevel@tonic-gate * to tap its reserves in order to prevent infinite 15597c478bd9Sstevel@tonic-gate * recursion. 15607c478bd9Sstevel@tonic-gate * 15617c478bd9Sstevel@tonic-gate * The following is a protection attribute (like PROT_READ, etc.) 15627c478bd9Sstevel@tonic-gate * 1563ae115bc7Smrj * HAT_NOSYNC set PT_NOSYNC - this mapping's ref/mod bits 15647c478bd9Sstevel@tonic-gate * are never cleared. 15657c478bd9Sstevel@tonic-gate * 15667c478bd9Sstevel@tonic-gate * Installing new valid PTE's and creation of the mapping list 15677c478bd9Sstevel@tonic-gate * entry are controlled under the same lock. It's derived from the 15687c478bd9Sstevel@tonic-gate * page_t being mapped. 15697c478bd9Sstevel@tonic-gate */ 15707c478bd9Sstevel@tonic-gate static uint_t supported_memload_flags = 15717c478bd9Sstevel@tonic-gate HAT_LOAD | HAT_LOAD_LOCK | HAT_LOAD_ADV | HAT_LOAD_NOCONSIST | 15727c478bd9Sstevel@tonic-gate HAT_LOAD_SHARE | HAT_NO_KALLOC | HAT_LOAD_REMAP | HAT_LOAD_TEXT; 15737c478bd9Sstevel@tonic-gate 15747c478bd9Sstevel@tonic-gate void 15757c478bd9Sstevel@tonic-gate hat_memload( 15767c478bd9Sstevel@tonic-gate hat_t *hat, 15777c478bd9Sstevel@tonic-gate caddr_t addr, 15787c478bd9Sstevel@tonic-gate page_t *pp, 15797c478bd9Sstevel@tonic-gate uint_t attr, 15807c478bd9Sstevel@tonic-gate uint_t flags) 15817c478bd9Sstevel@tonic-gate { 15827c478bd9Sstevel@tonic-gate uintptr_t va = (uintptr_t)addr; 15837c478bd9Sstevel@tonic-gate level_t level = 0; 15847c478bd9Sstevel@tonic-gate pfn_t pfn = page_pptonum(pp); 15857c478bd9Sstevel@tonic-gate 1586843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 15877c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(va)); 1588ae115bc7Smrj ASSERT(hat == kas.a_hat || va < _userlimit); 15897c478bd9Sstevel@tonic-gate ASSERT(hat == kas.a_hat || 15907c478bd9Sstevel@tonic-gate AS_LOCK_HELD(hat->hat_as, &hat->hat_as->a_lock)); 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); 16477c478bd9Sstevel@tonic-gate ASSERT(hat == kas.a_hat || 16487c478bd9Sstevel@tonic-gate AS_LOCK_HELD(hat->hat_as, &hat->hat_as->a_lock)); 16497c478bd9Sstevel@tonic-gate ASSERT((flags & supported_memload_flags) == flags); 16507c478bd9Sstevel@tonic-gate 16517c478bd9Sstevel@tonic-gate /* 16527c478bd9Sstevel@tonic-gate * memload is used for memory with full caching enabled, so 16537c478bd9Sstevel@tonic-gate * set HAT_STORECACHING_OK. 16547c478bd9Sstevel@tonic-gate */ 16557c478bd9Sstevel@tonic-gate attr |= HAT_STORECACHING_OK; 16567c478bd9Sstevel@tonic-gate 16577c478bd9Sstevel@tonic-gate /* 16587c478bd9Sstevel@tonic-gate * handle all pages using largest possible pagesize 16597c478bd9Sstevel@tonic-gate */ 16607c478bd9Sstevel@tonic-gate while (va < eaddr) { 16617c478bd9Sstevel@tonic-gate /* 16627c478bd9Sstevel@tonic-gate * decide what level mapping to use (ie. pagesize) 16637c478bd9Sstevel@tonic-gate */ 16647c478bd9Sstevel@tonic-gate pfn = page_pptonum(pages[pgindx]); 16657c478bd9Sstevel@tonic-gate for (level = mmu.max_page_level; ; --level) { 16667c478bd9Sstevel@tonic-gate pgsize = LEVEL_SIZE(level); 16677c478bd9Sstevel@tonic-gate if (level == 0) 16687c478bd9Sstevel@tonic-gate break; 1669ae115bc7Smrj 16707c478bd9Sstevel@tonic-gate if (!IS_P2ALIGNED(va, pgsize) || 16717c478bd9Sstevel@tonic-gate (eaddr - va) < pgsize || 1672ae115bc7Smrj !IS_P2ALIGNED(pfn_to_pa(pfn), pgsize)) 16737c478bd9Sstevel@tonic-gate continue; 16747c478bd9Sstevel@tonic-gate 16757c478bd9Sstevel@tonic-gate /* 16767c478bd9Sstevel@tonic-gate * To use a large mapping of this size, all the 16777c478bd9Sstevel@tonic-gate * pages we are passed must be sequential subpages 16787c478bd9Sstevel@tonic-gate * of the large page. 16797c478bd9Sstevel@tonic-gate * hat_page_demote() can't change p_szc because 16807c478bd9Sstevel@tonic-gate * all pages are locked. 16817c478bd9Sstevel@tonic-gate */ 16827c478bd9Sstevel@tonic-gate if (pages[pgindx]->p_szc >= level) { 16837c478bd9Sstevel@tonic-gate for (i = 0; i < mmu_btop(pgsize); ++i) { 16847c478bd9Sstevel@tonic-gate if (pfn + i != 16857c478bd9Sstevel@tonic-gate page_pptonum(pages[pgindx + i])) 16867c478bd9Sstevel@tonic-gate break; 16877c478bd9Sstevel@tonic-gate ASSERT(pages[pgindx + i]->p_szc >= 16887c478bd9Sstevel@tonic-gate level); 16897c478bd9Sstevel@tonic-gate ASSERT(pages[pgindx] + i == 16907c478bd9Sstevel@tonic-gate pages[pgindx + i]); 16917c478bd9Sstevel@tonic-gate } 169202bc52beSkchow if (i == mmu_btop(pgsize)) { 169302bc52beSkchow #ifdef DEBUG 169402bc52beSkchow if (level == 2) 169502bc52beSkchow map1gcnt++; 169602bc52beSkchow #endif 16977c478bd9Sstevel@tonic-gate break; 169802bc52beSkchow } 16997c478bd9Sstevel@tonic-gate } 17007c478bd9Sstevel@tonic-gate } 17017c478bd9Sstevel@tonic-gate 17027c478bd9Sstevel@tonic-gate /* 1703ae115bc7Smrj * Load this page mapping. If the load fails, try a smaller 1704ae115bc7Smrj * pagesize. 17057c478bd9Sstevel@tonic-gate */ 17067c478bd9Sstevel@tonic-gate ASSERT(!IN_VA_HOLE(va)); 1707ae115bc7Smrj while (hati_load_common(hat, va, pages[pgindx], attr, 1708250b7ff9Sjosephb flags, level, pfn) != 0) { 1709ae115bc7Smrj if (level == 0) 1710ae115bc7Smrj panic("unexpected hati_load_common() failure"); 1711ae115bc7Smrj --level; 1712ae115bc7Smrj pgsize = LEVEL_SIZE(level); 1713ae115bc7Smrj } 17147c478bd9Sstevel@tonic-gate 17157c478bd9Sstevel@tonic-gate /* 17167c478bd9Sstevel@tonic-gate * move to next page 17177c478bd9Sstevel@tonic-gate */ 17187c478bd9Sstevel@tonic-gate va += pgsize; 17197c478bd9Sstevel@tonic-gate pgindx += mmu_btop(pgsize); 17207c478bd9Sstevel@tonic-gate } 1721843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 17227c478bd9Sstevel@tonic-gate } 17237c478bd9Sstevel@tonic-gate 172405d3dc4bSpaulsan /* ARGSUSED */ 172505d3dc4bSpaulsan void 172605d3dc4bSpaulsan hat_memload_array_region(struct hat *hat, caddr_t addr, size_t len, 172705d3dc4bSpaulsan struct page **pps, uint_t attr, uint_t flags, 172805d3dc4bSpaulsan hat_region_cookie_t rcookie) 172905d3dc4bSpaulsan { 173005d3dc4bSpaulsan hat_memload_array(hat, addr, len, pps, attr, flags); 173105d3dc4bSpaulsan } 173205d3dc4bSpaulsan 17337c478bd9Sstevel@tonic-gate /* 17347c478bd9Sstevel@tonic-gate * void hat_devload(hat, addr, len, pf, attr, flags) 17357c478bd9Sstevel@tonic-gate * load/lock the given page frame number 17367c478bd9Sstevel@tonic-gate * 17377c478bd9Sstevel@tonic-gate * Advisory ordering attributes. Apply only to device mappings. 17387c478bd9Sstevel@tonic-gate * 17397c478bd9Sstevel@tonic-gate * HAT_STRICTORDER: the CPU must issue the references in order, as the 17407c478bd9Sstevel@tonic-gate * programmer specified. This is the default. 17417c478bd9Sstevel@tonic-gate * HAT_UNORDERED_OK: the CPU may reorder the references (this is all kinds 17427c478bd9Sstevel@tonic-gate * of reordering; store or load with store or load). 17437c478bd9Sstevel@tonic-gate * HAT_MERGING_OK: merging and batching: the CPU may merge individual stores 17447c478bd9Sstevel@tonic-gate * to consecutive locations (for example, turn two consecutive byte 17457c478bd9Sstevel@tonic-gate * stores into one halfword store), and it may batch individual loads 17467c478bd9Sstevel@tonic-gate * (for example, turn two consecutive byte loads into one halfword load). 17477c478bd9Sstevel@tonic-gate * This also implies re-ordering. 17487c478bd9Sstevel@tonic-gate * HAT_LOADCACHING_OK: the CPU may cache the data it fetches and reuse it 17497c478bd9Sstevel@tonic-gate * until another store occurs. The default is to fetch new data 17507c478bd9Sstevel@tonic-gate * on every load. This also implies merging. 17517c478bd9Sstevel@tonic-gate * HAT_STORECACHING_OK: the CPU may keep the data in the cache and push it to 17527c478bd9Sstevel@tonic-gate * the device (perhaps with other data) at a later time. The default is 17537c478bd9Sstevel@tonic-gate * to push the data right away. This also implies load caching. 17547c478bd9Sstevel@tonic-gate * 17557c478bd9Sstevel@tonic-gate * Equivalent of hat_memload(), but can be used for device memory where 17567c478bd9Sstevel@tonic-gate * there are no page_t's and we support additional flags (write merging, etc). 17577c478bd9Sstevel@tonic-gate * Note that we can have large page mappings with this interface. 17587c478bd9Sstevel@tonic-gate */ 17597c478bd9Sstevel@tonic-gate int supported_devload_flags = HAT_LOAD | HAT_LOAD_LOCK | 17607c478bd9Sstevel@tonic-gate HAT_LOAD_NOCONSIST | HAT_STRICTORDER | HAT_UNORDERED_OK | 17617c478bd9Sstevel@tonic-gate HAT_MERGING_OK | HAT_LOADCACHING_OK | HAT_STORECACHING_OK; 17627c478bd9Sstevel@tonic-gate 17637c478bd9Sstevel@tonic-gate void 17647c478bd9Sstevel@tonic-gate hat_devload( 17657c478bd9Sstevel@tonic-gate hat_t *hat, 17667c478bd9Sstevel@tonic-gate caddr_t addr, 17677c478bd9Sstevel@tonic-gate size_t len, 17687c478bd9Sstevel@tonic-gate pfn_t pfn, 17697c478bd9Sstevel@tonic-gate uint_t attr, 17707c478bd9Sstevel@tonic-gate int flags) 17717c478bd9Sstevel@tonic-gate { 17727c478bd9Sstevel@tonic-gate uintptr_t va = ALIGN2PAGE(addr); 17737c478bd9Sstevel@tonic-gate uintptr_t eva = va + len; 17747c478bd9Sstevel@tonic-gate level_t level; 17757c478bd9Sstevel@tonic-gate size_t pgsize; 17767c478bd9Sstevel@tonic-gate page_t *pp; 17777c478bd9Sstevel@tonic-gate int f; /* per PTE copy of flags - maybe modified */ 17787c478bd9Sstevel@tonic-gate uint_t a; /* per PTE copy of attr */ 17797c478bd9Sstevel@tonic-gate 1780843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 17817c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(va)); 1782ae115bc7Smrj ASSERT(hat == kas.a_hat || eva <= _userlimit); 17837c478bd9Sstevel@tonic-gate ASSERT(hat == kas.a_hat || 17847c478bd9Sstevel@tonic-gate AS_LOCK_HELD(hat->hat_as, &hat->hat_as->a_lock)); 17857c478bd9Sstevel@tonic-gate ASSERT((flags & supported_devload_flags) == flags); 17867c478bd9Sstevel@tonic-gate 17877c478bd9Sstevel@tonic-gate /* 17887c478bd9Sstevel@tonic-gate * handle all pages 17897c478bd9Sstevel@tonic-gate */ 17907c478bd9Sstevel@tonic-gate while (va < eva) { 17917c478bd9Sstevel@tonic-gate 17927c478bd9Sstevel@tonic-gate /* 17937c478bd9Sstevel@tonic-gate * decide what level mapping to use (ie. pagesize) 17947c478bd9Sstevel@tonic-gate */ 17957c478bd9Sstevel@tonic-gate for (level = mmu.max_page_level; ; --level) { 17967c478bd9Sstevel@tonic-gate pgsize = LEVEL_SIZE(level); 17977c478bd9Sstevel@tonic-gate if (level == 0) 17987c478bd9Sstevel@tonic-gate break; 17997c478bd9Sstevel@tonic-gate if (IS_P2ALIGNED(va, pgsize) && 18007c478bd9Sstevel@tonic-gate (eva - va) >= pgsize && 180102bc52beSkchow IS_P2ALIGNED(pfn, mmu_btop(pgsize))) { 180202bc52beSkchow #ifdef DEBUG 180302bc52beSkchow if (level == 2) 180402bc52beSkchow map1gcnt++; 180502bc52beSkchow #endif 18067c478bd9Sstevel@tonic-gate break; 180702bc52beSkchow } 18087c478bd9Sstevel@tonic-gate } 18097c478bd9Sstevel@tonic-gate 18107c478bd9Sstevel@tonic-gate /* 1811ae115bc7Smrj * If this is just memory then allow caching (this happens 18127c478bd9Sstevel@tonic-gate * for the nucleus pages) - though HAT_PLAT_NOCACHE can be used 1813ae115bc7Smrj * to override that. If we don't have a page_t then make sure 18147c478bd9Sstevel@tonic-gate * NOCONSIST is set. 18157c478bd9Sstevel@tonic-gate */ 18167c478bd9Sstevel@tonic-gate a = attr; 18177c478bd9Sstevel@tonic-gate f = flags; 1818843e1988Sjohnlev if (!pf_is_memory(pfn)) 1819843e1988Sjohnlev f |= HAT_LOAD_NOCONSIST; 1820843e1988Sjohnlev else if (!(a & HAT_PLAT_NOCACHE)) 1821843e1988Sjohnlev a |= HAT_STORECACHING_OK; 18227c478bd9Sstevel@tonic-gate 1823843e1988Sjohnlev if (f & HAT_LOAD_NOCONSIST) 18247c478bd9Sstevel@tonic-gate pp = NULL; 1825843e1988Sjohnlev else 1826843e1988Sjohnlev pp = page_numtopp_nolock(pfn); 18277c478bd9Sstevel@tonic-gate 1828e803c1e6SPrakash Sangappa /* 1829e803c1e6SPrakash Sangappa * Check to make sure we are really trying to map a valid 1830e803c1e6SPrakash Sangappa * memory page. The caller wishing to intentionally map 1831e803c1e6SPrakash Sangappa * free memory pages will have passed the HAT_LOAD_NOCONSIST 1832e803c1e6SPrakash Sangappa * flag, then pp will be NULL. 1833e803c1e6SPrakash Sangappa */ 1834e803c1e6SPrakash Sangappa if (pp != NULL) { 1835e803c1e6SPrakash Sangappa if (PP_ISFREE(pp)) { 1836e803c1e6SPrakash Sangappa panic("hat_devload: loading " 1837e803c1e6SPrakash Sangappa "a mapping to free page %p", (void *)pp); 1838e803c1e6SPrakash Sangappa } 1839e803c1e6SPrakash Sangappa 1840e803c1e6SPrakash Sangappa if (!PAGE_LOCKED(pp) && !PP_ISNORELOC(pp)) { 1841e803c1e6SPrakash Sangappa panic("hat_devload: loading a mapping " 1842e803c1e6SPrakash Sangappa "to an unlocked page %p", 1843e803c1e6SPrakash Sangappa (void *)pp); 1844e803c1e6SPrakash Sangappa } 1845e803c1e6SPrakash Sangappa } 1846e803c1e6SPrakash Sangappa 18477c478bd9Sstevel@tonic-gate /* 18487c478bd9Sstevel@tonic-gate * load this page mapping 18497c478bd9Sstevel@tonic-gate */ 18507c478bd9Sstevel@tonic-gate ASSERT(!IN_VA_HOLE(va)); 1851ae115bc7Smrj while (hati_load_common(hat, va, pp, a, f, level, pfn) != 0) { 1852ae115bc7Smrj if (level == 0) 1853ae115bc7Smrj panic("unexpected hati_load_common() failure"); 1854ae115bc7Smrj --level; 1855ae115bc7Smrj pgsize = LEVEL_SIZE(level); 1856ae115bc7Smrj } 18577c478bd9Sstevel@tonic-gate 18587c478bd9Sstevel@tonic-gate /* 18597c478bd9Sstevel@tonic-gate * move to next page 18607c478bd9Sstevel@tonic-gate */ 18617c478bd9Sstevel@tonic-gate va += pgsize; 18627c478bd9Sstevel@tonic-gate pfn += mmu_btop(pgsize); 18637c478bd9Sstevel@tonic-gate } 1864843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 18657c478bd9Sstevel@tonic-gate } 18667c478bd9Sstevel@tonic-gate 18677c478bd9Sstevel@tonic-gate /* 18687c478bd9Sstevel@tonic-gate * void hat_unlock(hat, addr, len) 18697c478bd9Sstevel@tonic-gate * unlock the mappings to a given range of addresses 18707c478bd9Sstevel@tonic-gate * 18717c478bd9Sstevel@tonic-gate * Locks are tracked by ht_lock_cnt in the htable. 18727c478bd9Sstevel@tonic-gate */ 18737c478bd9Sstevel@tonic-gate void 18747c478bd9Sstevel@tonic-gate hat_unlock(hat_t *hat, caddr_t addr, size_t len) 18757c478bd9Sstevel@tonic-gate { 18767c478bd9Sstevel@tonic-gate uintptr_t vaddr = (uintptr_t)addr; 18777c478bd9Sstevel@tonic-gate uintptr_t eaddr = vaddr + len; 18787c478bd9Sstevel@tonic-gate htable_t *ht = NULL; 18797c478bd9Sstevel@tonic-gate 18807c478bd9Sstevel@tonic-gate /* 18817c478bd9Sstevel@tonic-gate * kernel entries are always locked, we don't track lock counts 18827c478bd9Sstevel@tonic-gate */ 1883ae115bc7Smrj ASSERT(hat == kas.a_hat || eaddr <= _userlimit); 18847c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(vaddr)); 18857c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(eaddr)); 18867c478bd9Sstevel@tonic-gate if (hat == kas.a_hat) 18877c478bd9Sstevel@tonic-gate return; 18887c478bd9Sstevel@tonic-gate if (eaddr > _userlimit) 18897c478bd9Sstevel@tonic-gate panic("hat_unlock() address out of range - above _userlimit"); 18907c478bd9Sstevel@tonic-gate 1891843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 18927c478bd9Sstevel@tonic-gate ASSERT(AS_LOCK_HELD(hat->hat_as, &hat->hat_as->a_lock)); 18937c478bd9Sstevel@tonic-gate while (vaddr < eaddr) { 18947c478bd9Sstevel@tonic-gate (void) htable_walk(hat, &ht, &vaddr, eaddr); 18957c478bd9Sstevel@tonic-gate if (ht == NULL) 18967c478bd9Sstevel@tonic-gate break; 18977c478bd9Sstevel@tonic-gate 18987c478bd9Sstevel@tonic-gate ASSERT(!IN_VA_HOLE(vaddr)); 18997c478bd9Sstevel@tonic-gate 19007c478bd9Sstevel@tonic-gate if (ht->ht_lock_cnt < 1) 19017c478bd9Sstevel@tonic-gate panic("hat_unlock(): lock_cnt < 1, " 1902903a11ebSrh "htable=%p, vaddr=%p\n", (void *)ht, (void *)vaddr); 19037c478bd9Sstevel@tonic-gate HTABLE_LOCK_DEC(ht); 19047c478bd9Sstevel@tonic-gate 19057c478bd9Sstevel@tonic-gate vaddr += LEVEL_SIZE(ht->ht_level); 19067c478bd9Sstevel@tonic-gate } 19077c478bd9Sstevel@tonic-gate if (ht) 19087c478bd9Sstevel@tonic-gate htable_release(ht); 1909843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 19107c478bd9Sstevel@tonic-gate } 19117c478bd9Sstevel@tonic-gate 191205d3dc4bSpaulsan /* ARGSUSED */ 191305d3dc4bSpaulsan void 19147dacfc44Spaulsan hat_unlock_region(struct hat *hat, caddr_t addr, size_t len, 191505d3dc4bSpaulsan hat_region_cookie_t rcookie) 191605d3dc4bSpaulsan { 191705d3dc4bSpaulsan panic("No shared region support on x86"); 191805d3dc4bSpaulsan } 191905d3dc4bSpaulsan 1920843e1988Sjohnlev #if !defined(__xpv) 19217c478bd9Sstevel@tonic-gate /* 19227c478bd9Sstevel@tonic-gate * Cross call service routine to demap a virtual page on 19237c478bd9Sstevel@tonic-gate * the current CPU or flush all mappings in TLB. 19247c478bd9Sstevel@tonic-gate */ 19257c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 19267c478bd9Sstevel@tonic-gate static int 19277c478bd9Sstevel@tonic-gate hati_demap_func(xc_arg_t a1, xc_arg_t a2, xc_arg_t a3) 19287c478bd9Sstevel@tonic-gate { 19297c478bd9Sstevel@tonic-gate hat_t *hat = (hat_t *)a1; 19307c478bd9Sstevel@tonic-gate caddr_t addr = (caddr_t)a2; 19317c478bd9Sstevel@tonic-gate 19327c478bd9Sstevel@tonic-gate /* 19337c478bd9Sstevel@tonic-gate * If the target hat isn't the kernel and this CPU isn't operating 19347c478bd9Sstevel@tonic-gate * in the target hat, we can ignore the cross call. 19357c478bd9Sstevel@tonic-gate */ 19367c478bd9Sstevel@tonic-gate if (hat != kas.a_hat && hat != CPU->cpu_current_hat) 19377c478bd9Sstevel@tonic-gate return (0); 19387c478bd9Sstevel@tonic-gate 19397c478bd9Sstevel@tonic-gate /* 19407c478bd9Sstevel@tonic-gate * For a normal address, we just flush one page mapping 19417c478bd9Sstevel@tonic-gate */ 19427c478bd9Sstevel@tonic-gate if ((uintptr_t)addr != DEMAP_ALL_ADDR) { 1943ae115bc7Smrj mmu_tlbflush_entry(addr); 19447c478bd9Sstevel@tonic-gate return (0); 19457c478bd9Sstevel@tonic-gate } 19467c478bd9Sstevel@tonic-gate 19477c478bd9Sstevel@tonic-gate /* 19487c478bd9Sstevel@tonic-gate * Otherwise we reload cr3 to effect a complete TLB flush. 19497c478bd9Sstevel@tonic-gate * 19507c478bd9Sstevel@tonic-gate * A reload of cr3 on a VLP process also means we must also recopy in 19517c478bd9Sstevel@tonic-gate * the pte values from the struct hat 19527c478bd9Sstevel@tonic-gate */ 19537c478bd9Sstevel@tonic-gate if (hat->hat_flags & HAT_VLP) { 19547c478bd9Sstevel@tonic-gate #if defined(__amd64) 19557c478bd9Sstevel@tonic-gate x86pte_t *vlpptep = CPU->cpu_hat_info->hci_vlp_l2ptes; 19567c478bd9Sstevel@tonic-gate 19577c478bd9Sstevel@tonic-gate VLP_COPY(hat->hat_vlp_ptes, vlpptep); 19587c478bd9Sstevel@tonic-gate #elif defined(__i386) 19597c478bd9Sstevel@tonic-gate reload_pae32(hat, CPU); 19607c478bd9Sstevel@tonic-gate #endif 19617c478bd9Sstevel@tonic-gate } 19627c478bd9Sstevel@tonic-gate reload_cr3(); 19637c478bd9Sstevel@tonic-gate return (0); 19647c478bd9Sstevel@tonic-gate } 19657c478bd9Sstevel@tonic-gate 196695c0a3c8Sjosephb /* 196795c0a3c8Sjosephb * Flush all TLB entries, including global (ie. kernel) ones. 196895c0a3c8Sjosephb */ 196995c0a3c8Sjosephb static void 197095c0a3c8Sjosephb flush_all_tlb_entries(void) 197195c0a3c8Sjosephb { 197295c0a3c8Sjosephb ulong_t cr4 = getcr4(); 197395c0a3c8Sjosephb 197495c0a3c8Sjosephb if (cr4 & CR4_PGE) { 197595c0a3c8Sjosephb setcr4(cr4 & ~(ulong_t)CR4_PGE); 197695c0a3c8Sjosephb setcr4(cr4); 197795c0a3c8Sjosephb 197895c0a3c8Sjosephb /* 197995c0a3c8Sjosephb * 32 bit PAE also needs to always reload_cr3() 198095c0a3c8Sjosephb */ 198195c0a3c8Sjosephb if (mmu.max_level == 2) 198295c0a3c8Sjosephb reload_cr3(); 198395c0a3c8Sjosephb } else { 198495c0a3c8Sjosephb reload_cr3(); 198595c0a3c8Sjosephb } 198695c0a3c8Sjosephb } 198795c0a3c8Sjosephb 198895c0a3c8Sjosephb #define TLB_CPU_HALTED (01ul) 198995c0a3c8Sjosephb #define TLB_INVAL_ALL (02ul) 199095c0a3c8Sjosephb #define CAS_TLB_INFO(cpu, old, new) \ 199195c0a3c8Sjosephb caslong((ulong_t *)&(cpu)->cpu_m.mcpu_tlb_info, (old), (new)) 199295c0a3c8Sjosephb 199395c0a3c8Sjosephb /* 199495c0a3c8Sjosephb * Record that a CPU is going idle 199595c0a3c8Sjosephb */ 199695c0a3c8Sjosephb void 199795c0a3c8Sjosephb tlb_going_idle(void) 199895c0a3c8Sjosephb { 199995c0a3c8Sjosephb atomic_or_long((ulong_t *)&CPU->cpu_m.mcpu_tlb_info, TLB_CPU_HALTED); 200095c0a3c8Sjosephb } 200195c0a3c8Sjosephb 200295c0a3c8Sjosephb /* 200395c0a3c8Sjosephb * Service a delayed TLB flush if coming out of being idle. 200421584dbcSPavel Tatashin * It will be called from cpu idle notification with interrupt disabled. 200595c0a3c8Sjosephb */ 200695c0a3c8Sjosephb void 200795c0a3c8Sjosephb tlb_service(void) 200895c0a3c8Sjosephb { 200995c0a3c8Sjosephb ulong_t tlb_info; 201095c0a3c8Sjosephb ulong_t found; 201195c0a3c8Sjosephb 201295c0a3c8Sjosephb /* 201395c0a3c8Sjosephb * We only have to do something if coming out of being idle. 201495c0a3c8Sjosephb */ 201595c0a3c8Sjosephb tlb_info = CPU->cpu_m.mcpu_tlb_info; 201695c0a3c8Sjosephb if (tlb_info & TLB_CPU_HALTED) { 201795c0a3c8Sjosephb ASSERT(CPU->cpu_current_hat == kas.a_hat); 201895c0a3c8Sjosephb 201995c0a3c8Sjosephb /* 202095c0a3c8Sjosephb * Atomic clear and fetch of old state. 202195c0a3c8Sjosephb */ 202295c0a3c8Sjosephb while ((found = CAS_TLB_INFO(CPU, tlb_info, 0)) != tlb_info) { 202395c0a3c8Sjosephb ASSERT(found & TLB_CPU_HALTED); 202495c0a3c8Sjosephb tlb_info = found; 202595c0a3c8Sjosephb SMT_PAUSE(); 202695c0a3c8Sjosephb } 202795c0a3c8Sjosephb if (tlb_info & TLB_INVAL_ALL) 202895c0a3c8Sjosephb flush_all_tlb_entries(); 202995c0a3c8Sjosephb } 203095c0a3c8Sjosephb } 2031843e1988Sjohnlev #endif /* !__xpv */ 203295c0a3c8Sjosephb 20337c478bd9Sstevel@tonic-gate /* 20347c478bd9Sstevel@tonic-gate * Internal routine to do cross calls to invalidate a range of pages on 20357c478bd9Sstevel@tonic-gate * all CPUs using a given hat. 20367c478bd9Sstevel@tonic-gate */ 20377c478bd9Sstevel@tonic-gate void 2038ae115bc7Smrj hat_tlb_inval(hat_t *hat, uintptr_t va) 20397c478bd9Sstevel@tonic-gate { 20407c478bd9Sstevel@tonic-gate extern int flushes_require_xcalls; /* from mp_startup.c */ 20417c478bd9Sstevel@tonic-gate cpuset_t justme; 2042ae115bc7Smrj cpuset_t cpus_to_shootdown; 2043843e1988Sjohnlev #ifndef __xpv 2044843e1988Sjohnlev cpuset_t check_cpus; 204595c0a3c8Sjosephb cpu_t *cpup; 204695c0a3c8Sjosephb int c; 2047843e1988Sjohnlev #endif 20487c478bd9Sstevel@tonic-gate 20497c478bd9Sstevel@tonic-gate /* 20507c478bd9Sstevel@tonic-gate * If the hat is being destroyed, there are no more users, so 20517c478bd9Sstevel@tonic-gate * demap need not do anything. 20527c478bd9Sstevel@tonic-gate */ 20537c478bd9Sstevel@tonic-gate if (hat->hat_flags & HAT_FREEING) 20547c478bd9Sstevel@tonic-gate return; 20557c478bd9Sstevel@tonic-gate 20567c478bd9Sstevel@tonic-gate /* 20577c478bd9Sstevel@tonic-gate * If demapping from a shared pagetable, we best demap the 20587c478bd9Sstevel@tonic-gate * entire set of user TLBs, since we don't know what addresses 20597c478bd9Sstevel@tonic-gate * these were shared at. 20607c478bd9Sstevel@tonic-gate */ 20617c478bd9Sstevel@tonic-gate if (hat->hat_flags & HAT_SHARED) { 20627c478bd9Sstevel@tonic-gate hat = kas.a_hat; 20637c478bd9Sstevel@tonic-gate va = DEMAP_ALL_ADDR; 20647c478bd9Sstevel@tonic-gate } 20657c478bd9Sstevel@tonic-gate 20667c478bd9Sstevel@tonic-gate /* 20677c478bd9Sstevel@tonic-gate * if not running with multiple CPUs, don't use cross calls 20687c478bd9Sstevel@tonic-gate */ 20697c478bd9Sstevel@tonic-gate if (panicstr || !flushes_require_xcalls) { 2070843e1988Sjohnlev #ifdef __xpv 2071843e1988Sjohnlev if (va == DEMAP_ALL_ADDR) 2072843e1988Sjohnlev xen_flush_tlb(); 2073843e1988Sjohnlev else 2074843e1988Sjohnlev xen_flush_va((caddr_t)va); 2075843e1988Sjohnlev #else 20767c478bd9Sstevel@tonic-gate (void) hati_demap_func((xc_arg_t)hat, (xc_arg_t)va, NULL); 2077843e1988Sjohnlev #endif 20787c478bd9Sstevel@tonic-gate return; 20797c478bd9Sstevel@tonic-gate } 20807c478bd9Sstevel@tonic-gate 20817c478bd9Sstevel@tonic-gate 20827c478bd9Sstevel@tonic-gate /* 2083ae115bc7Smrj * Determine CPUs to shootdown. Kernel changes always do all CPUs. 2084ae115bc7Smrj * Otherwise it's just CPUs currently executing in this hat. 20857c478bd9Sstevel@tonic-gate */ 20867c478bd9Sstevel@tonic-gate kpreempt_disable(); 20877c478bd9Sstevel@tonic-gate CPUSET_ONLY(justme, CPU->cpu_id); 2088ae115bc7Smrj if (hat == kas.a_hat) 2089ae115bc7Smrj cpus_to_shootdown = khat_cpuset; 20907c478bd9Sstevel@tonic-gate else 2091ae115bc7Smrj cpus_to_shootdown = hat->hat_cpus; 2092ae115bc7Smrj 2093843e1988Sjohnlev #ifndef __xpv 209495c0a3c8Sjosephb /* 209595c0a3c8Sjosephb * If any CPUs in the set are idle, just request a delayed flush 209695c0a3c8Sjosephb * and avoid waking them up. 209795c0a3c8Sjosephb */ 209895c0a3c8Sjosephb check_cpus = cpus_to_shootdown; 209995c0a3c8Sjosephb for (c = 0; c < NCPU && !CPUSET_ISNULL(check_cpus); ++c) { 210095c0a3c8Sjosephb ulong_t tlb_info; 210195c0a3c8Sjosephb 210295c0a3c8Sjosephb if (!CPU_IN_SET(check_cpus, c)) 210395c0a3c8Sjosephb continue; 210495c0a3c8Sjosephb CPUSET_DEL(check_cpus, c); 210595c0a3c8Sjosephb cpup = cpu[c]; 210695c0a3c8Sjosephb if (cpup == NULL) 210795c0a3c8Sjosephb continue; 210895c0a3c8Sjosephb 210995c0a3c8Sjosephb tlb_info = cpup->cpu_m.mcpu_tlb_info; 211095c0a3c8Sjosephb while (tlb_info == TLB_CPU_HALTED) { 211195c0a3c8Sjosephb (void) CAS_TLB_INFO(cpup, TLB_CPU_HALTED, 2112250b7ff9Sjosephb TLB_CPU_HALTED | TLB_INVAL_ALL); 211395c0a3c8Sjosephb SMT_PAUSE(); 211495c0a3c8Sjosephb tlb_info = cpup->cpu_m.mcpu_tlb_info; 211595c0a3c8Sjosephb } 211695c0a3c8Sjosephb if (tlb_info == (TLB_CPU_HALTED | TLB_INVAL_ALL)) { 211795c0a3c8Sjosephb HATSTAT_INC(hs_tlb_inval_delayed); 211895c0a3c8Sjosephb CPUSET_DEL(cpus_to_shootdown, c); 211995c0a3c8Sjosephb } 212095c0a3c8Sjosephb } 2121843e1988Sjohnlev #endif 212295c0a3c8Sjosephb 2123ae115bc7Smrj if (CPUSET_ISNULL(cpus_to_shootdown) || 2124ae115bc7Smrj CPUSET_ISEQUAL(cpus_to_shootdown, justme)) { 2125ae115bc7Smrj 2126843e1988Sjohnlev #ifdef __xpv 2127843e1988Sjohnlev if (va == DEMAP_ALL_ADDR) 2128843e1988Sjohnlev xen_flush_tlb(); 2129843e1988Sjohnlev else 2130843e1988Sjohnlev xen_flush_va((caddr_t)va); 2131843e1988Sjohnlev #else 2132ae115bc7Smrj (void) hati_demap_func((xc_arg_t)hat, (xc_arg_t)va, NULL); 2133843e1988Sjohnlev #endif 2134ae115bc7Smrj 2135ae115bc7Smrj } else { 2136ae115bc7Smrj 2137ae115bc7Smrj CPUSET_ADD(cpus_to_shootdown, CPU->cpu_id); 2138843e1988Sjohnlev #ifdef __xpv 2139843e1988Sjohnlev if (va == DEMAP_ALL_ADDR) 2140843e1988Sjohnlev xen_gflush_tlb(cpus_to_shootdown); 2141843e1988Sjohnlev else 2142843e1988Sjohnlev xen_gflush_va((caddr_t)va, cpus_to_shootdown); 2143843e1988Sjohnlev #else 2144f34a7178SJoe Bonasera xc_call((xc_arg_t)hat, (xc_arg_t)va, NULL, 2145f34a7178SJoe Bonasera CPUSET2BV(cpus_to_shootdown), hati_demap_func); 2146843e1988Sjohnlev #endif 2147ae115bc7Smrj 2148ae115bc7Smrj } 21497c478bd9Sstevel@tonic-gate kpreempt_enable(); 21507c478bd9Sstevel@tonic-gate } 21517c478bd9Sstevel@tonic-gate 21527c478bd9Sstevel@tonic-gate /* 21537c478bd9Sstevel@tonic-gate * Interior routine for HAT_UNLOADs from hat_unload_callback(), 21547c478bd9Sstevel@tonic-gate * hat_kmap_unload() OR from hat_steal() code. This routine doesn't 21557c478bd9Sstevel@tonic-gate * handle releasing of the htables. 21567c478bd9Sstevel@tonic-gate */ 21577c478bd9Sstevel@tonic-gate void 21587c478bd9Sstevel@tonic-gate hat_pte_unmap( 21597c478bd9Sstevel@tonic-gate htable_t *ht, 21607c478bd9Sstevel@tonic-gate uint_t entry, 21617c478bd9Sstevel@tonic-gate uint_t flags, 21627c478bd9Sstevel@tonic-gate x86pte_t old_pte, 21637c478bd9Sstevel@tonic-gate void *pte_ptr) 21647c478bd9Sstevel@tonic-gate { 21657c478bd9Sstevel@tonic-gate hat_t *hat = ht->ht_hat; 21667c478bd9Sstevel@tonic-gate hment_t *hm = NULL; 21677c478bd9Sstevel@tonic-gate page_t *pp = NULL; 21687c478bd9Sstevel@tonic-gate level_t l = ht->ht_level; 21697c478bd9Sstevel@tonic-gate pfn_t pfn; 21707c478bd9Sstevel@tonic-gate 21717c478bd9Sstevel@tonic-gate /* 21727c478bd9Sstevel@tonic-gate * We always track the locking counts, even if nothing is unmapped 21737c478bd9Sstevel@tonic-gate */ 21747c478bd9Sstevel@tonic-gate if ((flags & HAT_UNLOAD_UNLOCK) != 0 && hat != kas.a_hat) { 21757c478bd9Sstevel@tonic-gate ASSERT(ht->ht_lock_cnt > 0); 21767c478bd9Sstevel@tonic-gate HTABLE_LOCK_DEC(ht); 21777c478bd9Sstevel@tonic-gate } 21787c478bd9Sstevel@tonic-gate 21797c478bd9Sstevel@tonic-gate /* 21807c478bd9Sstevel@tonic-gate * Figure out which page's mapping list lock to acquire using the PFN 21817c478bd9Sstevel@tonic-gate * passed in "old" PTE. We then attempt to invalidate the PTE. 21827c478bd9Sstevel@tonic-gate * If another thread, probably a hat_pageunload, has asynchronously 21837c478bd9Sstevel@tonic-gate * unmapped/remapped this address we'll loop here. 21847c478bd9Sstevel@tonic-gate */ 21857c478bd9Sstevel@tonic-gate ASSERT(ht->ht_busy > 0); 21867c478bd9Sstevel@tonic-gate while (PTE_ISVALID(old_pte)) { 21877c478bd9Sstevel@tonic-gate pfn = PTE2PFN(old_pte, l); 2188ae115bc7Smrj if (PTE_GET(old_pte, PT_SOFTWARE) >= PT_NOCONSIST) { 21897c478bd9Sstevel@tonic-gate pp = NULL; 21907c478bd9Sstevel@tonic-gate } else { 2191843e1988Sjohnlev #ifdef __xpv 2192843e1988Sjohnlev if (pfn == PFN_INVALID) 2193843e1988Sjohnlev panic("Invalid PFN, but not PT_NOCONSIST"); 2194843e1988Sjohnlev #endif 21957c478bd9Sstevel@tonic-gate pp = page_numtopp_nolock(pfn); 2196aa2ed9e5Sjosephb if (pp == NULL) { 2197aa2ed9e5Sjosephb panic("no page_t, not NOCONSIST: old_pte=" 2198aa2ed9e5Sjosephb FMT_PTE " ht=%lx entry=0x%x pte_ptr=%lx", 2199aa2ed9e5Sjosephb old_pte, (uintptr_t)ht, entry, 2200aa2ed9e5Sjosephb (uintptr_t)pte_ptr); 2201aa2ed9e5Sjosephb } 22027c478bd9Sstevel@tonic-gate x86_hm_enter(pp); 22037c478bd9Sstevel@tonic-gate } 2204aa2ed9e5Sjosephb 22056c9930aeSJoe Bonasera old_pte = x86pte_inval(ht, entry, old_pte, pte_ptr); 22067c478bd9Sstevel@tonic-gate 22077c478bd9Sstevel@tonic-gate /* 22087c478bd9Sstevel@tonic-gate * If the page hadn't changed we've unmapped it and can proceed 22097c478bd9Sstevel@tonic-gate */ 22107c478bd9Sstevel@tonic-gate if (PTE_ISVALID(old_pte) && PTE2PFN(old_pte, l) == pfn) 22117c478bd9Sstevel@tonic-gate break; 22127c478bd9Sstevel@tonic-gate 22137c478bd9Sstevel@tonic-gate /* 22147c478bd9Sstevel@tonic-gate * Otherwise, we'll have to retry with the current old_pte. 22157c478bd9Sstevel@tonic-gate * Drop the hment lock, since the pfn may have changed. 22167c478bd9Sstevel@tonic-gate */ 22177c478bd9Sstevel@tonic-gate if (pp != NULL) { 22187c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 22197c478bd9Sstevel@tonic-gate pp = NULL; 22207c478bd9Sstevel@tonic-gate } else { 2221ae115bc7Smrj ASSERT(PTE_GET(old_pte, PT_SOFTWARE) >= PT_NOCONSIST); 22227c478bd9Sstevel@tonic-gate } 22237c478bd9Sstevel@tonic-gate } 22247c478bd9Sstevel@tonic-gate 22257c478bd9Sstevel@tonic-gate /* 22267c478bd9Sstevel@tonic-gate * If the old mapping wasn't valid, there's nothing more to do 22277c478bd9Sstevel@tonic-gate */ 22287c478bd9Sstevel@tonic-gate if (!PTE_ISVALID(old_pte)) { 22297c478bd9Sstevel@tonic-gate if (pp != NULL) 22307c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 22317c478bd9Sstevel@tonic-gate return; 22327c478bd9Sstevel@tonic-gate } 22337c478bd9Sstevel@tonic-gate 22347c478bd9Sstevel@tonic-gate /* 22357c478bd9Sstevel@tonic-gate * Take care of syncing any MOD/REF bits and removing the hment. 22367c478bd9Sstevel@tonic-gate */ 22377c478bd9Sstevel@tonic-gate if (pp != NULL) { 22387c478bd9Sstevel@tonic-gate if (!(flags & HAT_UNLOAD_NOSYNC)) 22397c478bd9Sstevel@tonic-gate hati_sync_pte_to_page(pp, old_pte, l); 22407c478bd9Sstevel@tonic-gate hm = hment_remove(pp, ht, entry); 22417c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 22427c478bd9Sstevel@tonic-gate if (hm != NULL) 22437c478bd9Sstevel@tonic-gate hment_free(hm); 22447c478bd9Sstevel@tonic-gate } 22457c478bd9Sstevel@tonic-gate 22467c478bd9Sstevel@tonic-gate /* 22477c478bd9Sstevel@tonic-gate * Handle book keeping in the htable and hat 22487c478bd9Sstevel@tonic-gate */ 22497c478bd9Sstevel@tonic-gate ASSERT(ht->ht_valid_cnt > 0); 22507c478bd9Sstevel@tonic-gate HTABLE_DEC(ht->ht_valid_cnt); 22517c478bd9Sstevel@tonic-gate PGCNT_DEC(hat, l); 22527c478bd9Sstevel@tonic-gate } 22537c478bd9Sstevel@tonic-gate 22547c478bd9Sstevel@tonic-gate /* 22557c478bd9Sstevel@tonic-gate * very cheap unload implementation to special case some kernel addresses 22567c478bd9Sstevel@tonic-gate */ 22577c478bd9Sstevel@tonic-gate static void 22587c478bd9Sstevel@tonic-gate hat_kmap_unload(caddr_t addr, size_t len, uint_t flags) 22597c478bd9Sstevel@tonic-gate { 22607c478bd9Sstevel@tonic-gate uintptr_t va = (uintptr_t)addr; 22617c478bd9Sstevel@tonic-gate uintptr_t eva = va + len; 2262ae115bc7Smrj pgcnt_t pg_index; 22637c478bd9Sstevel@tonic-gate htable_t *ht; 22647c478bd9Sstevel@tonic-gate uint_t entry; 2265ae115bc7Smrj x86pte_t *pte_ptr; 22667c478bd9Sstevel@tonic-gate x86pte_t old_pte; 22677c478bd9Sstevel@tonic-gate 22687c478bd9Sstevel@tonic-gate for (; va < eva; va += MMU_PAGESIZE) { 22697c478bd9Sstevel@tonic-gate /* 22707c478bd9Sstevel@tonic-gate * Get the PTE 22717c478bd9Sstevel@tonic-gate */ 2272ae115bc7Smrj pg_index = mmu_btop(va - mmu.kmap_addr); 2273ae115bc7Smrj pte_ptr = PT_INDEX_PTR(mmu.kmap_ptes, pg_index); 2274ae115bc7Smrj old_pte = GET_PTE(pte_ptr); 22757c478bd9Sstevel@tonic-gate 22767c478bd9Sstevel@tonic-gate /* 22777c478bd9Sstevel@tonic-gate * get the htable / entry 22787c478bd9Sstevel@tonic-gate */ 22797c478bd9Sstevel@tonic-gate ht = mmu.kmap_htables[(va - mmu.kmap_htables[0]->ht_vaddr) 22807c478bd9Sstevel@tonic-gate >> LEVEL_SHIFT(1)]; 22817c478bd9Sstevel@tonic-gate entry = htable_va2entry(va, ht); 22827c478bd9Sstevel@tonic-gate 22837c478bd9Sstevel@tonic-gate /* 22847c478bd9Sstevel@tonic-gate * use mostly common code to unmap it. 22857c478bd9Sstevel@tonic-gate */ 22867c478bd9Sstevel@tonic-gate hat_pte_unmap(ht, entry, flags, old_pte, pte_ptr); 22877c478bd9Sstevel@tonic-gate } 22887c478bd9Sstevel@tonic-gate } 22897c478bd9Sstevel@tonic-gate 22907c478bd9Sstevel@tonic-gate 22917c478bd9Sstevel@tonic-gate /* 22927c478bd9Sstevel@tonic-gate * unload a range of virtual address space (no callback) 22937c478bd9Sstevel@tonic-gate */ 22947c478bd9Sstevel@tonic-gate void 22957c478bd9Sstevel@tonic-gate hat_unload(hat_t *hat, caddr_t addr, size_t len, uint_t flags) 22967c478bd9Sstevel@tonic-gate { 22977c478bd9Sstevel@tonic-gate uintptr_t va = (uintptr_t)addr; 2298ae115bc7Smrj 2299843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 2300ae115bc7Smrj ASSERT(hat == kas.a_hat || va + len <= _userlimit); 23017c478bd9Sstevel@tonic-gate 23027c478bd9Sstevel@tonic-gate /* 23037c478bd9Sstevel@tonic-gate * special case for performance. 23047c478bd9Sstevel@tonic-gate */ 23057c478bd9Sstevel@tonic-gate if (mmu.kmap_addr <= va && va < mmu.kmap_eaddr) { 23067c478bd9Sstevel@tonic-gate ASSERT(hat == kas.a_hat); 23077c478bd9Sstevel@tonic-gate hat_kmap_unload(addr, len, flags); 2308ae115bc7Smrj } else { 2309ae115bc7Smrj hat_unload_callback(hat, addr, len, flags, NULL); 23107c478bd9Sstevel@tonic-gate } 2311843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 23127c478bd9Sstevel@tonic-gate } 23137c478bd9Sstevel@tonic-gate 23147c478bd9Sstevel@tonic-gate /* 23157c478bd9Sstevel@tonic-gate * Do the callbacks for ranges being unloaded. 23167c478bd9Sstevel@tonic-gate */ 23177c478bd9Sstevel@tonic-gate typedef struct range_info { 23187c478bd9Sstevel@tonic-gate uintptr_t rng_va; 23197c478bd9Sstevel@tonic-gate ulong_t rng_cnt; 23207c478bd9Sstevel@tonic-gate level_t rng_level; 23217c478bd9Sstevel@tonic-gate } range_info_t; 23227c478bd9Sstevel@tonic-gate 23237c478bd9Sstevel@tonic-gate static void 23247c478bd9Sstevel@tonic-gate handle_ranges(hat_callback_t *cb, uint_t cnt, range_info_t *range) 23257c478bd9Sstevel@tonic-gate { 23267c478bd9Sstevel@tonic-gate /* 23277c478bd9Sstevel@tonic-gate * do callbacks to upper level VM system 23287c478bd9Sstevel@tonic-gate */ 23297c478bd9Sstevel@tonic-gate while (cb != NULL && cnt > 0) { 23307c478bd9Sstevel@tonic-gate --cnt; 23317c478bd9Sstevel@tonic-gate cb->hcb_start_addr = (caddr_t)range[cnt].rng_va; 23327c478bd9Sstevel@tonic-gate cb->hcb_end_addr = cb->hcb_start_addr; 23337c478bd9Sstevel@tonic-gate cb->hcb_end_addr += 23347c478bd9Sstevel@tonic-gate range[cnt].rng_cnt << LEVEL_SIZE(range[cnt].rng_level); 23357c478bd9Sstevel@tonic-gate cb->hcb_function(cb); 23367c478bd9Sstevel@tonic-gate } 23377c478bd9Sstevel@tonic-gate } 23387c478bd9Sstevel@tonic-gate 23397c478bd9Sstevel@tonic-gate /* 23407c478bd9Sstevel@tonic-gate * Unload a given range of addresses (has optional callback) 23417c478bd9Sstevel@tonic-gate * 23427c478bd9Sstevel@tonic-gate * Flags: 23437c478bd9Sstevel@tonic-gate * define HAT_UNLOAD 0x00 23447c478bd9Sstevel@tonic-gate * define HAT_UNLOAD_NOSYNC 0x02 23457c478bd9Sstevel@tonic-gate * define HAT_UNLOAD_UNLOCK 0x04 23467c478bd9Sstevel@tonic-gate * define HAT_UNLOAD_OTHER 0x08 - not used 23477c478bd9Sstevel@tonic-gate * define HAT_UNLOAD_UNMAP 0x10 - same as HAT_UNLOAD 23487c478bd9Sstevel@tonic-gate */ 23497c478bd9Sstevel@tonic-gate #define MAX_UNLOAD_CNT (8) 23507c478bd9Sstevel@tonic-gate void 23517c478bd9Sstevel@tonic-gate hat_unload_callback( 23527c478bd9Sstevel@tonic-gate hat_t *hat, 23537c478bd9Sstevel@tonic-gate caddr_t addr, 23547c478bd9Sstevel@tonic-gate size_t len, 23557c478bd9Sstevel@tonic-gate uint_t flags, 23567c478bd9Sstevel@tonic-gate hat_callback_t *cb) 23577c478bd9Sstevel@tonic-gate { 23587c478bd9Sstevel@tonic-gate uintptr_t vaddr = (uintptr_t)addr; 23597c478bd9Sstevel@tonic-gate uintptr_t eaddr = vaddr + len; 23607c478bd9Sstevel@tonic-gate htable_t *ht = NULL; 23617c478bd9Sstevel@tonic-gate uint_t entry; 2362aa2ed9e5Sjosephb uintptr_t contig_va = (uintptr_t)-1L; 23637c478bd9Sstevel@tonic-gate range_info_t r[MAX_UNLOAD_CNT]; 23647c478bd9Sstevel@tonic-gate uint_t r_cnt = 0; 23657c478bd9Sstevel@tonic-gate x86pte_t old_pte; 23667c478bd9Sstevel@tonic-gate 2367843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 2368ae115bc7Smrj ASSERT(hat == kas.a_hat || eaddr <= _userlimit); 23697c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(vaddr)); 23707c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(eaddr)); 23717c478bd9Sstevel@tonic-gate 2372ae115bc7Smrj /* 2373ae115bc7Smrj * Special case a single page being unloaded for speed. This happens 2374ae115bc7Smrj * quite frequently, COW faults after a fork() for example. 2375ae115bc7Smrj */ 2376ae115bc7Smrj if (cb == NULL && len == MMU_PAGESIZE) { 2377ae115bc7Smrj ht = htable_getpte(hat, vaddr, &entry, &old_pte, 0); 2378ae115bc7Smrj if (ht != NULL) { 2379ae115bc7Smrj if (PTE_ISVALID(old_pte)) 2380ae115bc7Smrj hat_pte_unmap(ht, entry, flags, old_pte, NULL); 2381ae115bc7Smrj htable_release(ht); 2382ae115bc7Smrj } 2383843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 2384ae115bc7Smrj return; 2385ae115bc7Smrj } 2386ae115bc7Smrj 23877c478bd9Sstevel@tonic-gate while (vaddr < eaddr) { 23887c478bd9Sstevel@tonic-gate old_pte = htable_walk(hat, &ht, &vaddr, eaddr); 23897c478bd9Sstevel@tonic-gate if (ht == NULL) 23907c478bd9Sstevel@tonic-gate break; 23917c478bd9Sstevel@tonic-gate 23927c478bd9Sstevel@tonic-gate ASSERT(!IN_VA_HOLE(vaddr)); 23937c478bd9Sstevel@tonic-gate 23947c478bd9Sstevel@tonic-gate if (vaddr < (uintptr_t)addr) 23957c478bd9Sstevel@tonic-gate panic("hat_unload_callback(): unmap inside large page"); 23967c478bd9Sstevel@tonic-gate 23977c478bd9Sstevel@tonic-gate /* 23987c478bd9Sstevel@tonic-gate * We'll do the call backs for contiguous ranges 23997c478bd9Sstevel@tonic-gate */ 2400aa2ed9e5Sjosephb if (vaddr != contig_va || 24017c478bd9Sstevel@tonic-gate (r_cnt > 0 && r[r_cnt - 1].rng_level != ht->ht_level)) { 24027c478bd9Sstevel@tonic-gate if (r_cnt == MAX_UNLOAD_CNT) { 24037c478bd9Sstevel@tonic-gate handle_ranges(cb, r_cnt, r); 24047c478bd9Sstevel@tonic-gate r_cnt = 0; 24057c478bd9Sstevel@tonic-gate } 24067c478bd9Sstevel@tonic-gate r[r_cnt].rng_va = vaddr; 24077c478bd9Sstevel@tonic-gate r[r_cnt].rng_cnt = 0; 24087c478bd9Sstevel@tonic-gate r[r_cnt].rng_level = ht->ht_level; 24097c478bd9Sstevel@tonic-gate ++r_cnt; 24107c478bd9Sstevel@tonic-gate } 24117c478bd9Sstevel@tonic-gate 24127c478bd9Sstevel@tonic-gate /* 24137c478bd9Sstevel@tonic-gate * Unload one mapping from the page tables. 24147c478bd9Sstevel@tonic-gate */ 24157c478bd9Sstevel@tonic-gate entry = htable_va2entry(vaddr, ht); 24167c478bd9Sstevel@tonic-gate hat_pte_unmap(ht, entry, flags, old_pte, NULL); 24177c478bd9Sstevel@tonic-gate ASSERT(ht->ht_level <= mmu.max_page_level); 24187c478bd9Sstevel@tonic-gate vaddr += LEVEL_SIZE(ht->ht_level); 2419aa2ed9e5Sjosephb contig_va = vaddr; 24207c478bd9Sstevel@tonic-gate ++r[r_cnt - 1].rng_cnt; 24217c478bd9Sstevel@tonic-gate } 24227c478bd9Sstevel@tonic-gate if (ht) 24237c478bd9Sstevel@tonic-gate htable_release(ht); 24247c478bd9Sstevel@tonic-gate 24257c478bd9Sstevel@tonic-gate /* 24267c478bd9Sstevel@tonic-gate * handle last range for callbacks 24277c478bd9Sstevel@tonic-gate */ 24287c478bd9Sstevel@tonic-gate if (r_cnt > 0) 24297c478bd9Sstevel@tonic-gate handle_ranges(cb, r_cnt, r); 2430843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 24317c478bd9Sstevel@tonic-gate } 24327c478bd9Sstevel@tonic-gate 2433ca3e8d88SDave Plauger /* 24349e0b7c70SDave Plauger * Invalidate a virtual address translation on a slave CPU during 24359e0b7c70SDave Plauger * panic() dumps. 2436ca3e8d88SDave Plauger */ 2437ca3e8d88SDave Plauger void 2438ca3e8d88SDave Plauger hat_flush_range(hat_t *hat, caddr_t va, size_t size) 2439ca3e8d88SDave Plauger { 2440ca3e8d88SDave Plauger ssize_t sz; 2441ca3e8d88SDave Plauger caddr_t endva = va + size; 2442ca3e8d88SDave Plauger 2443ca3e8d88SDave Plauger while (va < endva) { 2444ca3e8d88SDave Plauger sz = hat_getpagesize(hat, va); 24459e0b7c70SDave Plauger if (sz < 0) { 2446ca3e8d88SDave Plauger #ifdef __xpv 2447ca3e8d88SDave Plauger xen_flush_tlb(); 2448ca3e8d88SDave Plauger #else 24499e0b7c70SDave Plauger flush_all_tlb_entries(); 2450ca3e8d88SDave Plauger #endif 2451ca3e8d88SDave Plauger break; 24529e0b7c70SDave Plauger } 24539e0b7c70SDave Plauger #ifdef __xpv 24549e0b7c70SDave Plauger xen_flush_va(va); 24559e0b7c70SDave Plauger #else 24569e0b7c70SDave Plauger mmu_tlbflush_entry(va); 24579e0b7c70SDave Plauger #endif 2458ca3e8d88SDave Plauger va += sz; 2459ca3e8d88SDave Plauger } 2460ca3e8d88SDave Plauger } 2461ca3e8d88SDave Plauger 24627c478bd9Sstevel@tonic-gate /* 24637c478bd9Sstevel@tonic-gate * synchronize mapping with software data structures 24647c478bd9Sstevel@tonic-gate * 24657c478bd9Sstevel@tonic-gate * This interface is currently only used by the working set monitor 24667c478bd9Sstevel@tonic-gate * driver. 24677c478bd9Sstevel@tonic-gate */ 24687c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 24697c478bd9Sstevel@tonic-gate void 24707c478bd9Sstevel@tonic-gate hat_sync(hat_t *hat, caddr_t addr, size_t len, uint_t flags) 24717c478bd9Sstevel@tonic-gate { 24727c478bd9Sstevel@tonic-gate uintptr_t vaddr = (uintptr_t)addr; 24737c478bd9Sstevel@tonic-gate uintptr_t eaddr = vaddr + len; 24747c478bd9Sstevel@tonic-gate htable_t *ht = NULL; 24757c478bd9Sstevel@tonic-gate uint_t entry; 24767c478bd9Sstevel@tonic-gate x86pte_t pte; 24777c478bd9Sstevel@tonic-gate x86pte_t save_pte; 24787c478bd9Sstevel@tonic-gate x86pte_t new; 24797c478bd9Sstevel@tonic-gate page_t *pp; 24807c478bd9Sstevel@tonic-gate 24817c478bd9Sstevel@tonic-gate ASSERT(!IN_VA_HOLE(vaddr)); 24827c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(vaddr)); 24837c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(eaddr)); 2484ae115bc7Smrj ASSERT(hat == kas.a_hat || eaddr <= _userlimit); 24857c478bd9Sstevel@tonic-gate 2486843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 24877c478bd9Sstevel@tonic-gate for (; vaddr < eaddr; vaddr += LEVEL_SIZE(ht->ht_level)) { 24887c478bd9Sstevel@tonic-gate try_again: 24897c478bd9Sstevel@tonic-gate pte = htable_walk(hat, &ht, &vaddr, eaddr); 24907c478bd9Sstevel@tonic-gate if (ht == NULL) 24917c478bd9Sstevel@tonic-gate break; 24927c478bd9Sstevel@tonic-gate entry = htable_va2entry(vaddr, ht); 24937c478bd9Sstevel@tonic-gate 2494ae115bc7Smrj if (PTE_GET(pte, PT_SOFTWARE) >= PT_NOSYNC || 24957c478bd9Sstevel@tonic-gate PTE_GET(pte, PT_REF | PT_MOD) == 0) 24967c478bd9Sstevel@tonic-gate continue; 24977c478bd9Sstevel@tonic-gate 24987c478bd9Sstevel@tonic-gate /* 24997c478bd9Sstevel@tonic-gate * We need to acquire the mapping list lock to protect 25007c478bd9Sstevel@tonic-gate * against hat_pageunload(), hat_unload(), etc. 25017c478bd9Sstevel@tonic-gate */ 25027c478bd9Sstevel@tonic-gate pp = page_numtopp_nolock(PTE2PFN(pte, ht->ht_level)); 25037c478bd9Sstevel@tonic-gate if (pp == NULL) 25047c478bd9Sstevel@tonic-gate break; 25057c478bd9Sstevel@tonic-gate x86_hm_enter(pp); 25067c478bd9Sstevel@tonic-gate save_pte = pte; 25077c478bd9Sstevel@tonic-gate pte = x86pte_get(ht, entry); 25087c478bd9Sstevel@tonic-gate if (pte != save_pte) { 25097c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 25107c478bd9Sstevel@tonic-gate goto try_again; 25117c478bd9Sstevel@tonic-gate } 2512ae115bc7Smrj if (PTE_GET(pte, PT_SOFTWARE) >= PT_NOSYNC || 25137c478bd9Sstevel@tonic-gate PTE_GET(pte, PT_REF | PT_MOD) == 0) { 25147c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 25157c478bd9Sstevel@tonic-gate continue; 25167c478bd9Sstevel@tonic-gate } 25177c478bd9Sstevel@tonic-gate 25187c478bd9Sstevel@tonic-gate /* 25197c478bd9Sstevel@tonic-gate * Need to clear ref or mod bits. We may compete with 25207c478bd9Sstevel@tonic-gate * hardware updating the R/M bits and have to try again. 25217c478bd9Sstevel@tonic-gate */ 25227c478bd9Sstevel@tonic-gate if (flags == HAT_SYNC_ZERORM) { 25237c478bd9Sstevel@tonic-gate new = pte; 25247c478bd9Sstevel@tonic-gate PTE_CLR(new, PT_REF | PT_MOD); 25257c478bd9Sstevel@tonic-gate pte = hati_update_pte(ht, entry, pte, new); 25267c478bd9Sstevel@tonic-gate if (pte != 0) { 25277c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 25287c478bd9Sstevel@tonic-gate goto try_again; 25297c478bd9Sstevel@tonic-gate } 25307c478bd9Sstevel@tonic-gate } else { 25317c478bd9Sstevel@tonic-gate /* 25327c478bd9Sstevel@tonic-gate * sync the PTE to the page_t 25337c478bd9Sstevel@tonic-gate */ 25347c478bd9Sstevel@tonic-gate hati_sync_pte_to_page(pp, save_pte, ht->ht_level); 25357c478bd9Sstevel@tonic-gate } 25367c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 25377c478bd9Sstevel@tonic-gate } 25387c478bd9Sstevel@tonic-gate if (ht) 25397c478bd9Sstevel@tonic-gate htable_release(ht); 2540843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 25417c478bd9Sstevel@tonic-gate } 25427c478bd9Sstevel@tonic-gate 25437c478bd9Sstevel@tonic-gate /* 25447c478bd9Sstevel@tonic-gate * void hat_map(hat, addr, len, flags) 25457c478bd9Sstevel@tonic-gate */ 25467c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 25477c478bd9Sstevel@tonic-gate void 25487c478bd9Sstevel@tonic-gate hat_map(hat_t *hat, caddr_t addr, size_t len, uint_t flags) 25497c478bd9Sstevel@tonic-gate { 25507c478bd9Sstevel@tonic-gate /* does nothing */ 25517c478bd9Sstevel@tonic-gate } 25527c478bd9Sstevel@tonic-gate 25537c478bd9Sstevel@tonic-gate /* 25547c478bd9Sstevel@tonic-gate * uint_t hat_getattr(hat, addr, *attr) 25557c478bd9Sstevel@tonic-gate * returns attr for <hat,addr> in *attr. returns 0 if there was a 25567c478bd9Sstevel@tonic-gate * mapping and *attr is valid, nonzero if there was no mapping and 25577c478bd9Sstevel@tonic-gate * *attr is not valid. 25587c478bd9Sstevel@tonic-gate */ 25597c478bd9Sstevel@tonic-gate uint_t 25607c478bd9Sstevel@tonic-gate hat_getattr(hat_t *hat, caddr_t addr, uint_t *attr) 25617c478bd9Sstevel@tonic-gate { 25627c478bd9Sstevel@tonic-gate uintptr_t vaddr = ALIGN2PAGE(addr); 25637c478bd9Sstevel@tonic-gate htable_t *ht = NULL; 25647c478bd9Sstevel@tonic-gate x86pte_t pte; 25657c478bd9Sstevel@tonic-gate 2566ae115bc7Smrj ASSERT(hat == kas.a_hat || vaddr <= _userlimit); 25677c478bd9Sstevel@tonic-gate 25687c478bd9Sstevel@tonic-gate if (IN_VA_HOLE(vaddr)) 25697c478bd9Sstevel@tonic-gate return ((uint_t)-1); 25707c478bd9Sstevel@tonic-gate 2571ae115bc7Smrj ht = htable_getpte(hat, vaddr, NULL, &pte, mmu.max_page_level); 25727c478bd9Sstevel@tonic-gate if (ht == NULL) 25737c478bd9Sstevel@tonic-gate return ((uint_t)-1); 25747c478bd9Sstevel@tonic-gate 25757c478bd9Sstevel@tonic-gate if (!PTE_ISVALID(pte) || !PTE_ISPAGE(pte, ht->ht_level)) { 25767c478bd9Sstevel@tonic-gate htable_release(ht); 25777c478bd9Sstevel@tonic-gate return ((uint_t)-1); 25787c478bd9Sstevel@tonic-gate } 25797c478bd9Sstevel@tonic-gate 25807c478bd9Sstevel@tonic-gate *attr = PROT_READ; 25817c478bd9Sstevel@tonic-gate if (PTE_GET(pte, PT_WRITABLE)) 25827c478bd9Sstevel@tonic-gate *attr |= PROT_WRITE; 25837c478bd9Sstevel@tonic-gate if (PTE_GET(pte, PT_USER)) 25847c478bd9Sstevel@tonic-gate *attr |= PROT_USER; 25857c478bd9Sstevel@tonic-gate if (!PTE_GET(pte, mmu.pt_nx)) 25867c478bd9Sstevel@tonic-gate *attr |= PROT_EXEC; 2587ae115bc7Smrj if (PTE_GET(pte, PT_SOFTWARE) >= PT_NOSYNC) 25887c478bd9Sstevel@tonic-gate *attr |= HAT_NOSYNC; 25897c478bd9Sstevel@tonic-gate htable_release(ht); 25907c478bd9Sstevel@tonic-gate return (0); 25917c478bd9Sstevel@tonic-gate } 25927c478bd9Sstevel@tonic-gate 25937c478bd9Sstevel@tonic-gate /* 25947c478bd9Sstevel@tonic-gate * hat_updateattr() applies the given attribute change to an existing mapping 25957c478bd9Sstevel@tonic-gate */ 25967c478bd9Sstevel@tonic-gate #define HAT_LOAD_ATTR 1 25977c478bd9Sstevel@tonic-gate #define HAT_SET_ATTR 2 25987c478bd9Sstevel@tonic-gate #define HAT_CLR_ATTR 3 25997c478bd9Sstevel@tonic-gate 26007c478bd9Sstevel@tonic-gate static void 26017c478bd9Sstevel@tonic-gate hat_updateattr(hat_t *hat, caddr_t addr, size_t len, uint_t attr, int what) 26027c478bd9Sstevel@tonic-gate { 26037c478bd9Sstevel@tonic-gate uintptr_t vaddr = (uintptr_t)addr; 26047c478bd9Sstevel@tonic-gate uintptr_t eaddr = (uintptr_t)addr + len; 26057c478bd9Sstevel@tonic-gate htable_t *ht = NULL; 26067c478bd9Sstevel@tonic-gate uint_t entry; 26077c478bd9Sstevel@tonic-gate x86pte_t oldpte, newpte; 26087c478bd9Sstevel@tonic-gate page_t *pp; 26097c478bd9Sstevel@tonic-gate 2610843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 26117c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(vaddr)); 26127c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(eaddr)); 26137c478bd9Sstevel@tonic-gate ASSERT(hat == kas.a_hat || 26147c478bd9Sstevel@tonic-gate AS_LOCK_HELD(hat->hat_as, &hat->hat_as->a_lock)); 26157c478bd9Sstevel@tonic-gate for (; vaddr < eaddr; vaddr += LEVEL_SIZE(ht->ht_level)) { 26167c478bd9Sstevel@tonic-gate try_again: 26177c478bd9Sstevel@tonic-gate oldpte = htable_walk(hat, &ht, &vaddr, eaddr); 26187c478bd9Sstevel@tonic-gate if (ht == NULL) 26197c478bd9Sstevel@tonic-gate break; 2620ae115bc7Smrj if (PTE_GET(oldpte, PT_SOFTWARE) >= PT_NOCONSIST) 26217c478bd9Sstevel@tonic-gate continue; 26227c478bd9Sstevel@tonic-gate 26237c478bd9Sstevel@tonic-gate pp = page_numtopp_nolock(PTE2PFN(oldpte, ht->ht_level)); 26247c478bd9Sstevel@tonic-gate if (pp == NULL) 26257c478bd9Sstevel@tonic-gate continue; 26267c478bd9Sstevel@tonic-gate x86_hm_enter(pp); 26277c478bd9Sstevel@tonic-gate 26287c478bd9Sstevel@tonic-gate newpte = oldpte; 26297c478bd9Sstevel@tonic-gate /* 26307c478bd9Sstevel@tonic-gate * We found a page table entry in the desired range, 26317c478bd9Sstevel@tonic-gate * figure out the new attributes. 26327c478bd9Sstevel@tonic-gate */ 26337c478bd9Sstevel@tonic-gate if (what == HAT_SET_ATTR || what == HAT_LOAD_ATTR) { 26347c478bd9Sstevel@tonic-gate if ((attr & PROT_WRITE) && 26357c478bd9Sstevel@tonic-gate !PTE_GET(oldpte, PT_WRITABLE)) 26367c478bd9Sstevel@tonic-gate newpte |= PT_WRITABLE; 26377c478bd9Sstevel@tonic-gate 2638ae115bc7Smrj if ((attr & HAT_NOSYNC) && 2639ae115bc7Smrj PTE_GET(oldpte, PT_SOFTWARE) < PT_NOSYNC) 26407c478bd9Sstevel@tonic-gate newpte |= PT_NOSYNC; 26417c478bd9Sstevel@tonic-gate 26427c478bd9Sstevel@tonic-gate if ((attr & PROT_EXEC) && PTE_GET(oldpte, mmu.pt_nx)) 26437c478bd9Sstevel@tonic-gate newpte &= ~mmu.pt_nx; 26447c478bd9Sstevel@tonic-gate } 26457c478bd9Sstevel@tonic-gate 26467c478bd9Sstevel@tonic-gate if (what == HAT_LOAD_ATTR) { 26477c478bd9Sstevel@tonic-gate if (!(attr & PROT_WRITE) && 26487c478bd9Sstevel@tonic-gate PTE_GET(oldpte, PT_WRITABLE)) 26497c478bd9Sstevel@tonic-gate newpte &= ~PT_WRITABLE; 26507c478bd9Sstevel@tonic-gate 2651ae115bc7Smrj if (!(attr & HAT_NOSYNC) && 2652ae115bc7Smrj PTE_GET(oldpte, PT_SOFTWARE) >= PT_NOSYNC) 2653ae115bc7Smrj newpte &= ~PT_SOFTWARE; 26547c478bd9Sstevel@tonic-gate 26557c478bd9Sstevel@tonic-gate if (!(attr & PROT_EXEC) && !PTE_GET(oldpte, mmu.pt_nx)) 26567c478bd9Sstevel@tonic-gate newpte |= mmu.pt_nx; 26577c478bd9Sstevel@tonic-gate } 26587c478bd9Sstevel@tonic-gate 26597c478bd9Sstevel@tonic-gate if (what == HAT_CLR_ATTR) { 26607c478bd9Sstevel@tonic-gate if ((attr & PROT_WRITE) && PTE_GET(oldpte, PT_WRITABLE)) 26617c478bd9Sstevel@tonic-gate newpte &= ~PT_WRITABLE; 26627c478bd9Sstevel@tonic-gate 2663ae115bc7Smrj if ((attr & HAT_NOSYNC) && 2664ae115bc7Smrj PTE_GET(oldpte, PT_SOFTWARE) >= PT_NOSYNC) 2665ae115bc7Smrj newpte &= ~PT_SOFTWARE; 26667c478bd9Sstevel@tonic-gate 26677c478bd9Sstevel@tonic-gate if ((attr & PROT_EXEC) && !PTE_GET(oldpte, mmu.pt_nx)) 26687c478bd9Sstevel@tonic-gate newpte |= mmu.pt_nx; 26697c478bd9Sstevel@tonic-gate } 26707c478bd9Sstevel@tonic-gate 2671ae115bc7Smrj /* 2672ae115bc7Smrj * Ensure NOSYNC/NOCONSIST mappings have REF and MOD set. 2673ae115bc7Smrj * x86pte_set() depends on this. 2674ae115bc7Smrj */ 2675ae115bc7Smrj if (PTE_GET(newpte, PT_SOFTWARE) >= PT_NOSYNC) 2676ae115bc7Smrj newpte |= PT_REF | PT_MOD; 2677ae115bc7Smrj 26787c478bd9Sstevel@tonic-gate /* 26797c478bd9Sstevel@tonic-gate * what about PROT_READ or others? this code only handles: 26807c478bd9Sstevel@tonic-gate * EXEC, WRITE, NOSYNC 26817c478bd9Sstevel@tonic-gate */ 26827c478bd9Sstevel@tonic-gate 26837c478bd9Sstevel@tonic-gate /* 26847c478bd9Sstevel@tonic-gate * If new PTE really changed, update the table. 26857c478bd9Sstevel@tonic-gate */ 26867c478bd9Sstevel@tonic-gate if (newpte != oldpte) { 26877c478bd9Sstevel@tonic-gate entry = htable_va2entry(vaddr, ht); 26887c478bd9Sstevel@tonic-gate oldpte = hati_update_pte(ht, entry, oldpte, newpte); 26897c478bd9Sstevel@tonic-gate if (oldpte != 0) { 26907c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 26917c478bd9Sstevel@tonic-gate goto try_again; 26927c478bd9Sstevel@tonic-gate } 26937c478bd9Sstevel@tonic-gate } 26947c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 26957c478bd9Sstevel@tonic-gate } 26967c478bd9Sstevel@tonic-gate if (ht) 26977c478bd9Sstevel@tonic-gate htable_release(ht); 2698843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 26997c478bd9Sstevel@tonic-gate } 27007c478bd9Sstevel@tonic-gate 27017c478bd9Sstevel@tonic-gate /* 27027c478bd9Sstevel@tonic-gate * Various wrappers for hat_updateattr() 27037c478bd9Sstevel@tonic-gate */ 27047c478bd9Sstevel@tonic-gate void 27057c478bd9Sstevel@tonic-gate hat_setattr(hat_t *hat, caddr_t addr, size_t len, uint_t attr) 27067c478bd9Sstevel@tonic-gate { 2707ae115bc7Smrj ASSERT(hat == kas.a_hat || (uintptr_t)addr + len <= _userlimit); 27087c478bd9Sstevel@tonic-gate hat_updateattr(hat, addr, len, attr, HAT_SET_ATTR); 27097c478bd9Sstevel@tonic-gate } 27107c478bd9Sstevel@tonic-gate 27117c478bd9Sstevel@tonic-gate void 27127c478bd9Sstevel@tonic-gate hat_clrattr(hat_t *hat, caddr_t addr, size_t len, uint_t attr) 27137c478bd9Sstevel@tonic-gate { 2714ae115bc7Smrj ASSERT(hat == kas.a_hat || (uintptr_t)addr + len <= _userlimit); 27157c478bd9Sstevel@tonic-gate hat_updateattr(hat, addr, len, attr, HAT_CLR_ATTR); 27167c478bd9Sstevel@tonic-gate } 27177c478bd9Sstevel@tonic-gate 27187c478bd9Sstevel@tonic-gate void 27197c478bd9Sstevel@tonic-gate hat_chgattr(hat_t *hat, caddr_t addr, size_t len, uint_t attr) 27207c478bd9Sstevel@tonic-gate { 2721ae115bc7Smrj ASSERT(hat == kas.a_hat || (uintptr_t)addr + len <= _userlimit); 27227c478bd9Sstevel@tonic-gate hat_updateattr(hat, addr, len, attr, HAT_LOAD_ATTR); 27237c478bd9Sstevel@tonic-gate } 27247c478bd9Sstevel@tonic-gate 27257c478bd9Sstevel@tonic-gate void 27267c478bd9Sstevel@tonic-gate hat_chgprot(hat_t *hat, caddr_t addr, size_t len, uint_t vprot) 27277c478bd9Sstevel@tonic-gate { 2728ae115bc7Smrj ASSERT(hat == kas.a_hat || (uintptr_t)addr + len <= _userlimit); 27297c478bd9Sstevel@tonic-gate hat_updateattr(hat, addr, len, vprot & HAT_PROT_MASK, HAT_LOAD_ATTR); 27307c478bd9Sstevel@tonic-gate } 27317c478bd9Sstevel@tonic-gate 27327c478bd9Sstevel@tonic-gate /* 27337c478bd9Sstevel@tonic-gate * size_t hat_getpagesize(hat, addr) 27347c478bd9Sstevel@tonic-gate * returns pagesize in bytes for <hat, addr>. returns -1 of there is 27357c478bd9Sstevel@tonic-gate * no mapping. This is an advisory call. 27367c478bd9Sstevel@tonic-gate */ 27377c478bd9Sstevel@tonic-gate ssize_t 27387c478bd9Sstevel@tonic-gate hat_getpagesize(hat_t *hat, caddr_t addr) 27397c478bd9Sstevel@tonic-gate { 27407c478bd9Sstevel@tonic-gate uintptr_t vaddr = ALIGN2PAGE(addr); 27417c478bd9Sstevel@tonic-gate htable_t *ht; 27427c478bd9Sstevel@tonic-gate size_t pagesize; 27437c478bd9Sstevel@tonic-gate 2744ae115bc7Smrj ASSERT(hat == kas.a_hat || vaddr <= _userlimit); 27457c478bd9Sstevel@tonic-gate if (IN_VA_HOLE(vaddr)) 27467c478bd9Sstevel@tonic-gate return (-1); 27477c478bd9Sstevel@tonic-gate ht = htable_getpage(hat, vaddr, NULL); 27487c478bd9Sstevel@tonic-gate if (ht == NULL) 27497c478bd9Sstevel@tonic-gate return (-1); 27507c478bd9Sstevel@tonic-gate pagesize = LEVEL_SIZE(ht->ht_level); 27517c478bd9Sstevel@tonic-gate htable_release(ht); 27527c478bd9Sstevel@tonic-gate return (pagesize); 27537c478bd9Sstevel@tonic-gate } 27547c478bd9Sstevel@tonic-gate 27557c478bd9Sstevel@tonic-gate 27567c478bd9Sstevel@tonic-gate 27577c478bd9Sstevel@tonic-gate /* 27587c478bd9Sstevel@tonic-gate * pfn_t hat_getpfnum(hat, addr) 27597c478bd9Sstevel@tonic-gate * returns pfn for <hat, addr> or PFN_INVALID if mapping is invalid. 27607c478bd9Sstevel@tonic-gate */ 27617c478bd9Sstevel@tonic-gate pfn_t 27627c478bd9Sstevel@tonic-gate hat_getpfnum(hat_t *hat, caddr_t addr) 27637c478bd9Sstevel@tonic-gate { 27647c478bd9Sstevel@tonic-gate uintptr_t vaddr = ALIGN2PAGE(addr); 27657c478bd9Sstevel@tonic-gate htable_t *ht; 27667c478bd9Sstevel@tonic-gate uint_t entry; 27677c478bd9Sstevel@tonic-gate pfn_t pfn = PFN_INVALID; 27687c478bd9Sstevel@tonic-gate 2769ae115bc7Smrj ASSERT(hat == kas.a_hat || vaddr <= _userlimit); 27707c478bd9Sstevel@tonic-gate if (khat_running == 0) 2771ae115bc7Smrj return (PFN_INVALID); 27727c478bd9Sstevel@tonic-gate 27737c478bd9Sstevel@tonic-gate if (IN_VA_HOLE(vaddr)) 27747c478bd9Sstevel@tonic-gate return (PFN_INVALID); 27757c478bd9Sstevel@tonic-gate 2776843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 27777c478bd9Sstevel@tonic-gate /* 27787c478bd9Sstevel@tonic-gate * A very common use of hat_getpfnum() is from the DDI for kernel pages. 27797c478bd9Sstevel@tonic-gate * Use the kmap_ptes (which also covers the 32 bit heap) to speed 27807c478bd9Sstevel@tonic-gate * this up. 27817c478bd9Sstevel@tonic-gate */ 27827c478bd9Sstevel@tonic-gate if (mmu.kmap_addr <= vaddr && vaddr < mmu.kmap_eaddr) { 27837c478bd9Sstevel@tonic-gate x86pte_t pte; 2784ae115bc7Smrj pgcnt_t pg_index; 27857c478bd9Sstevel@tonic-gate 2786ae115bc7Smrj pg_index = mmu_btop(vaddr - mmu.kmap_addr); 2787ae115bc7Smrj pte = GET_PTE(PT_INDEX_PTR(mmu.kmap_ptes, pg_index)); 2788843e1988Sjohnlev if (PTE_ISVALID(pte)) 2789843e1988Sjohnlev /*LINTED [use of constant 0 causes a lint warning] */ 2790843e1988Sjohnlev pfn = PTE2PFN(pte, 0); 2791843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 2792843e1988Sjohnlev return (pfn); 27937c478bd9Sstevel@tonic-gate } 27947c478bd9Sstevel@tonic-gate 27957c478bd9Sstevel@tonic-gate ht = htable_getpage(hat, vaddr, &entry); 2796843e1988Sjohnlev if (ht == NULL) { 2797843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 27987c478bd9Sstevel@tonic-gate return (PFN_INVALID); 2799843e1988Sjohnlev } 28007c478bd9Sstevel@tonic-gate ASSERT(vaddr >= ht->ht_vaddr); 28017c478bd9Sstevel@tonic-gate ASSERT(vaddr <= HTABLE_LAST_PAGE(ht)); 28027c478bd9Sstevel@tonic-gate pfn = PTE2PFN(x86pte_get(ht, entry), ht->ht_level); 28037c478bd9Sstevel@tonic-gate if (ht->ht_level > 0) 28047c478bd9Sstevel@tonic-gate pfn += mmu_btop(vaddr & LEVEL_OFFSET(ht->ht_level)); 28057c478bd9Sstevel@tonic-gate htable_release(ht); 2806843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 28077c478bd9Sstevel@tonic-gate return (pfn); 28087c478bd9Sstevel@tonic-gate } 28097c478bd9Sstevel@tonic-gate 28107c478bd9Sstevel@tonic-gate /* 28117c478bd9Sstevel@tonic-gate * int hat_probe(hat, addr) 28127c478bd9Sstevel@tonic-gate * return 0 if no valid mapping is present. Faster version 28137c478bd9Sstevel@tonic-gate * of hat_getattr in certain architectures. 28147c478bd9Sstevel@tonic-gate */ 28157c478bd9Sstevel@tonic-gate int 28167c478bd9Sstevel@tonic-gate hat_probe(hat_t *hat, caddr_t addr) 28177c478bd9Sstevel@tonic-gate { 28187c478bd9Sstevel@tonic-gate uintptr_t vaddr = ALIGN2PAGE(addr); 28197c478bd9Sstevel@tonic-gate uint_t entry; 28207c478bd9Sstevel@tonic-gate htable_t *ht; 28217c478bd9Sstevel@tonic-gate pgcnt_t pg_off; 28227c478bd9Sstevel@tonic-gate 2823ae115bc7Smrj ASSERT(hat == kas.a_hat || vaddr <= _userlimit); 28247c478bd9Sstevel@tonic-gate ASSERT(hat == kas.a_hat || 28257c478bd9Sstevel@tonic-gate AS_LOCK_HELD(hat->hat_as, &hat->hat_as->a_lock)); 28267c478bd9Sstevel@tonic-gate if (IN_VA_HOLE(vaddr)) 28277c478bd9Sstevel@tonic-gate return (0); 28287c478bd9Sstevel@tonic-gate 28297c478bd9Sstevel@tonic-gate /* 28307c478bd9Sstevel@tonic-gate * Most common use of hat_probe is from segmap. We special case it 28317c478bd9Sstevel@tonic-gate * for performance. 28327c478bd9Sstevel@tonic-gate */ 28337c478bd9Sstevel@tonic-gate if (mmu.kmap_addr <= vaddr && vaddr < mmu.kmap_eaddr) { 28347c478bd9Sstevel@tonic-gate pg_off = mmu_btop(vaddr - mmu.kmap_addr); 28357c478bd9Sstevel@tonic-gate if (mmu.pae_hat) 28367c478bd9Sstevel@tonic-gate return (PTE_ISVALID(mmu.kmap_ptes[pg_off])); 28377c478bd9Sstevel@tonic-gate else 28387c478bd9Sstevel@tonic-gate return (PTE_ISVALID( 28397c478bd9Sstevel@tonic-gate ((x86pte32_t *)mmu.kmap_ptes)[pg_off])); 28407c478bd9Sstevel@tonic-gate } 28417c478bd9Sstevel@tonic-gate 28427c478bd9Sstevel@tonic-gate ht = htable_getpage(hat, vaddr, &entry); 28437c478bd9Sstevel@tonic-gate htable_release(ht); 2844843e1988Sjohnlev return (ht != NULL); 28457c478bd9Sstevel@tonic-gate } 28467c478bd9Sstevel@tonic-gate 28477c478bd9Sstevel@tonic-gate /* 2848250b7ff9Sjosephb * Find out if the segment for hat_share()/hat_unshare() is DISM or locked ISM. 2849250b7ff9Sjosephb */ 2850250b7ff9Sjosephb static int 2851250b7ff9Sjosephb is_it_dism(hat_t *hat, caddr_t va) 2852250b7ff9Sjosephb { 2853250b7ff9Sjosephb struct seg *seg; 2854250b7ff9Sjosephb struct shm_data *shmd; 2855250b7ff9Sjosephb struct spt_data *sptd; 2856250b7ff9Sjosephb 2857250b7ff9Sjosephb seg = as_findseg(hat->hat_as, va, 0); 2858250b7ff9Sjosephb ASSERT(seg != NULL); 2859250b7ff9Sjosephb ASSERT(seg->s_base <= va); 2860250b7ff9Sjosephb shmd = (struct shm_data *)seg->s_data; 2861250b7ff9Sjosephb ASSERT(shmd != NULL); 2862250b7ff9Sjosephb sptd = (struct spt_data *)shmd->shm_sptseg->s_data; 2863250b7ff9Sjosephb ASSERT(sptd != NULL); 2864250b7ff9Sjosephb if (sptd->spt_flags & SHM_PAGEABLE) 2865250b7ff9Sjosephb return (1); 2866250b7ff9Sjosephb return (0); 2867250b7ff9Sjosephb } 2868250b7ff9Sjosephb 2869250b7ff9Sjosephb /* 2870250b7ff9Sjosephb * Simple implementation of ISM. hat_share() is similar to hat_memload_array(), 28717c478bd9Sstevel@tonic-gate * except that we use the ism_hat's existing mappings to determine the pages 2872250b7ff9Sjosephb * and protections to use for this hat. If we find a full properly aligned 2873250b7ff9Sjosephb * and sized pagetable, we will attempt to share the pagetable itself. 28747c478bd9Sstevel@tonic-gate */ 28757c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 28767c478bd9Sstevel@tonic-gate int 28777c478bd9Sstevel@tonic-gate hat_share( 28787c478bd9Sstevel@tonic-gate hat_t *hat, 28797c478bd9Sstevel@tonic-gate caddr_t addr, 28807c478bd9Sstevel@tonic-gate hat_t *ism_hat, 28817c478bd9Sstevel@tonic-gate caddr_t src_addr, 28827c478bd9Sstevel@tonic-gate size_t len, /* almost useless value, see below.. */ 28837c478bd9Sstevel@tonic-gate uint_t ismszc) 28847c478bd9Sstevel@tonic-gate { 28857c478bd9Sstevel@tonic-gate uintptr_t vaddr_start = (uintptr_t)addr; 28867c478bd9Sstevel@tonic-gate uintptr_t vaddr; 28877c478bd9Sstevel@tonic-gate uintptr_t eaddr = vaddr_start + len; 28887c478bd9Sstevel@tonic-gate uintptr_t ism_addr_start = (uintptr_t)src_addr; 28897c478bd9Sstevel@tonic-gate uintptr_t ism_addr = ism_addr_start; 28907c478bd9Sstevel@tonic-gate uintptr_t e_ism_addr = ism_addr + len; 28917c478bd9Sstevel@tonic-gate htable_t *ism_ht = NULL; 28927c478bd9Sstevel@tonic-gate htable_t *ht; 28937c478bd9Sstevel@tonic-gate x86pte_t pte; 28947c478bd9Sstevel@tonic-gate page_t *pp; 28957c478bd9Sstevel@tonic-gate pfn_t pfn; 28967c478bd9Sstevel@tonic-gate level_t l; 28977c478bd9Sstevel@tonic-gate pgcnt_t pgcnt; 28987c478bd9Sstevel@tonic-gate uint_t prot; 2899250b7ff9Sjosephb int is_dism; 2900250b7ff9Sjosephb int flags; 29017c478bd9Sstevel@tonic-gate 29027c478bd9Sstevel@tonic-gate /* 29037c478bd9Sstevel@tonic-gate * We might be asked to share an empty DISM hat by as_dup() 29047c478bd9Sstevel@tonic-gate */ 29057c478bd9Sstevel@tonic-gate ASSERT(hat != kas.a_hat); 2906ae115bc7Smrj ASSERT(eaddr <= _userlimit); 29077c478bd9Sstevel@tonic-gate if (!(ism_hat->hat_flags & HAT_SHARED)) { 29087c478bd9Sstevel@tonic-gate ASSERT(hat_get_mapped_size(ism_hat) == 0); 29097c478bd9Sstevel@tonic-gate return (0); 29107c478bd9Sstevel@tonic-gate } 2911843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 29127c478bd9Sstevel@tonic-gate 29137c478bd9Sstevel@tonic-gate /* 29147c478bd9Sstevel@tonic-gate * The SPT segment driver often passes us a size larger than there are 29157c478bd9Sstevel@tonic-gate * valid mappings. That's because it rounds the segment size up to a 29167c478bd9Sstevel@tonic-gate * large pagesize, even if the actual memory mapped by ism_hat is less. 29177c478bd9Sstevel@tonic-gate */ 29187c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(vaddr_start)); 29197c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(ism_addr_start)); 29207c478bd9Sstevel@tonic-gate ASSERT(ism_hat->hat_flags & HAT_SHARED); 2921250b7ff9Sjosephb is_dism = is_it_dism(hat, addr); 29227c478bd9Sstevel@tonic-gate while (ism_addr < e_ism_addr) { 29237c478bd9Sstevel@tonic-gate /* 29247c478bd9Sstevel@tonic-gate * use htable_walk to get the next valid ISM mapping 29257c478bd9Sstevel@tonic-gate */ 29267c478bd9Sstevel@tonic-gate pte = htable_walk(ism_hat, &ism_ht, &ism_addr, e_ism_addr); 29277c478bd9Sstevel@tonic-gate if (ism_ht == NULL) 29287c478bd9Sstevel@tonic-gate break; 29297c478bd9Sstevel@tonic-gate 29307c478bd9Sstevel@tonic-gate /* 2931250b7ff9Sjosephb * First check to see if we already share the page table. 29327c478bd9Sstevel@tonic-gate */ 2933250b7ff9Sjosephb l = ism_ht->ht_level; 29347c478bd9Sstevel@tonic-gate vaddr = vaddr_start + (ism_addr - ism_addr_start); 2935250b7ff9Sjosephb ht = htable_lookup(hat, vaddr, l); 2936250b7ff9Sjosephb if (ht != NULL) { 2937250b7ff9Sjosephb if (ht->ht_flags & HTABLE_SHARED_PFN) 2938250b7ff9Sjosephb goto shared; 2939250b7ff9Sjosephb htable_release(ht); 2940250b7ff9Sjosephb goto not_shared; 29417c478bd9Sstevel@tonic-gate } 29427c478bd9Sstevel@tonic-gate 29437c478bd9Sstevel@tonic-gate /* 2944250b7ff9Sjosephb * Can't ever share top table. 29457c478bd9Sstevel@tonic-gate */ 2946250b7ff9Sjosephb if (l == mmu.max_level) 2947250b7ff9Sjosephb goto not_shared; 29487c478bd9Sstevel@tonic-gate 2949250b7ff9Sjosephb /* 2950250b7ff9Sjosephb * Avoid level mismatches later due to DISM faults. 2951250b7ff9Sjosephb */ 2952250b7ff9Sjosephb if (is_dism && l > 0) 2953250b7ff9Sjosephb goto not_shared; 29547c478bd9Sstevel@tonic-gate 2955250b7ff9Sjosephb /* 2956250b7ff9Sjosephb * addresses and lengths must align 2957250b7ff9Sjosephb * table must be fully populated 2958250b7ff9Sjosephb * no lower level page tables 2959250b7ff9Sjosephb */ 2960250b7ff9Sjosephb if (ism_addr != ism_ht->ht_vaddr || 2961250b7ff9Sjosephb (vaddr & LEVEL_OFFSET(l + 1)) != 0) 2962250b7ff9Sjosephb goto not_shared; 29637c478bd9Sstevel@tonic-gate 2964250b7ff9Sjosephb /* 2965250b7ff9Sjosephb * The range of address space must cover a full table. 2966250b7ff9Sjosephb */ 29671d03c31eSjohnlev if (e_ism_addr - ism_addr < LEVEL_SIZE(l + 1)) 2968250b7ff9Sjosephb goto not_shared; 29697c478bd9Sstevel@tonic-gate 2970250b7ff9Sjosephb /* 2971250b7ff9Sjosephb * All entries in the ISM page table must be leaf PTEs. 2972250b7ff9Sjosephb */ 2973250b7ff9Sjosephb if (l > 0) { 2974250b7ff9Sjosephb int e; 29757c478bd9Sstevel@tonic-gate 2976250b7ff9Sjosephb /* 2977250b7ff9Sjosephb * We know the 0th is from htable_walk() above. 2978250b7ff9Sjosephb */ 2979250b7ff9Sjosephb for (e = 1; e < HTABLE_NUM_PTES(ism_ht); ++e) { 2980250b7ff9Sjosephb x86pte_t pte; 2981250b7ff9Sjosephb pte = x86pte_get(ism_ht, e); 2982250b7ff9Sjosephb if (!PTE_ISPAGE(pte, l)) 2983250b7ff9Sjosephb goto not_shared; 29847c478bd9Sstevel@tonic-gate } 29857c478bd9Sstevel@tonic-gate } 29867c478bd9Sstevel@tonic-gate 2987250b7ff9Sjosephb /* 2988250b7ff9Sjosephb * share the page table 2989250b7ff9Sjosephb */ 2990250b7ff9Sjosephb ht = htable_create(hat, vaddr, l, ism_ht); 2991250b7ff9Sjosephb shared: 2992250b7ff9Sjosephb ASSERT(ht->ht_flags & HTABLE_SHARED_PFN); 2993250b7ff9Sjosephb ASSERT(ht->ht_shares == ism_ht); 2994250b7ff9Sjosephb hat->hat_ism_pgcnt += 2995250b7ff9Sjosephb (ism_ht->ht_valid_cnt - ht->ht_valid_cnt) << 2996250b7ff9Sjosephb (LEVEL_SHIFT(ht->ht_level) - MMU_PAGESHIFT); 2997250b7ff9Sjosephb ht->ht_valid_cnt = ism_ht->ht_valid_cnt; 2998250b7ff9Sjosephb htable_release(ht); 2999250b7ff9Sjosephb ism_addr = ism_ht->ht_vaddr + LEVEL_SIZE(l + 1); 3000250b7ff9Sjosephb htable_release(ism_ht); 3001250b7ff9Sjosephb ism_ht = NULL; 3002250b7ff9Sjosephb continue; 3003250b7ff9Sjosephb 3004250b7ff9Sjosephb not_shared: 30057c478bd9Sstevel@tonic-gate /* 30067c478bd9Sstevel@tonic-gate * Unable to share the page table. Instead we will 30077c478bd9Sstevel@tonic-gate * create new mappings from the values in the ISM mappings. 3008250b7ff9Sjosephb * Figure out what level size mappings to use; 3009250b7ff9Sjosephb */ 3010250b7ff9Sjosephb for (l = ism_ht->ht_level; l > 0; --l) { 3011250b7ff9Sjosephb if (LEVEL_SIZE(l) <= eaddr - vaddr && 3012250b7ff9Sjosephb (vaddr & LEVEL_OFFSET(l)) == 0) 3013250b7ff9Sjosephb break; 3014250b7ff9Sjosephb } 3015250b7ff9Sjosephb 3016250b7ff9Sjosephb /* 30177c478bd9Sstevel@tonic-gate * The ISM mapping might be larger than the share area, 3018250b7ff9Sjosephb * be careful to truncate it if needed. 30197c478bd9Sstevel@tonic-gate */ 30207c478bd9Sstevel@tonic-gate if (eaddr - vaddr >= LEVEL_SIZE(ism_ht->ht_level)) { 30217c478bd9Sstevel@tonic-gate pgcnt = mmu_btop(LEVEL_SIZE(ism_ht->ht_level)); 30227c478bd9Sstevel@tonic-gate } else { 30237c478bd9Sstevel@tonic-gate pgcnt = mmu_btop(eaddr - vaddr); 30247c478bd9Sstevel@tonic-gate l = 0; 30257c478bd9Sstevel@tonic-gate } 30267c478bd9Sstevel@tonic-gate 30277c478bd9Sstevel@tonic-gate pfn = PTE2PFN(pte, ism_ht->ht_level); 30287c478bd9Sstevel@tonic-gate ASSERT(pfn != PFN_INVALID); 30297c478bd9Sstevel@tonic-gate while (pgcnt > 0) { 30307c478bd9Sstevel@tonic-gate /* 30317c478bd9Sstevel@tonic-gate * Make a new pte for the PFN for this level. 30327c478bd9Sstevel@tonic-gate * Copy protections for the pte from the ISM pte. 30337c478bd9Sstevel@tonic-gate */ 30347c478bd9Sstevel@tonic-gate pp = page_numtopp_nolock(pfn); 30357c478bd9Sstevel@tonic-gate ASSERT(pp != NULL); 30367c478bd9Sstevel@tonic-gate 30377c478bd9Sstevel@tonic-gate prot = PROT_USER | PROT_READ | HAT_UNORDERED_OK; 30387c478bd9Sstevel@tonic-gate if (PTE_GET(pte, PT_WRITABLE)) 30397c478bd9Sstevel@tonic-gate prot |= PROT_WRITE; 30407c478bd9Sstevel@tonic-gate if (!PTE_GET(pte, PT_NX)) 30417c478bd9Sstevel@tonic-gate prot |= PROT_EXEC; 30427c478bd9Sstevel@tonic-gate 3043250b7ff9Sjosephb flags = HAT_LOAD; 3044250b7ff9Sjosephb if (!is_dism) 3045250b7ff9Sjosephb flags |= HAT_LOAD_LOCK | HAT_LOAD_NOCONSIST; 3046250b7ff9Sjosephb while (hati_load_common(hat, vaddr, pp, prot, flags, 3047ae115bc7Smrj l, pfn) != 0) { 3048ae115bc7Smrj if (l == 0) 3049ae115bc7Smrj panic("hati_load_common() failure"); 3050ae115bc7Smrj --l; 3051ae115bc7Smrj } 30527c478bd9Sstevel@tonic-gate 30537c478bd9Sstevel@tonic-gate vaddr += LEVEL_SIZE(l); 30547c478bd9Sstevel@tonic-gate ism_addr += LEVEL_SIZE(l); 30557c478bd9Sstevel@tonic-gate pfn += mmu_btop(LEVEL_SIZE(l)); 30567c478bd9Sstevel@tonic-gate pgcnt -= mmu_btop(LEVEL_SIZE(l)); 30577c478bd9Sstevel@tonic-gate } 30587c478bd9Sstevel@tonic-gate } 30597c478bd9Sstevel@tonic-gate if (ism_ht != NULL) 30607c478bd9Sstevel@tonic-gate htable_release(ism_ht); 3061843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 30627c478bd9Sstevel@tonic-gate return (0); 30637c478bd9Sstevel@tonic-gate } 30647c478bd9Sstevel@tonic-gate 30657c478bd9Sstevel@tonic-gate 30667c478bd9Sstevel@tonic-gate /* 30677c478bd9Sstevel@tonic-gate * hat_unshare() is similar to hat_unload_callback(), but 30687c478bd9Sstevel@tonic-gate * we have to look for empty shared pagetables. Note that 30697c478bd9Sstevel@tonic-gate * hat_unshare() is always invoked against an entire segment. 30707c478bd9Sstevel@tonic-gate */ 30717c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 30727c478bd9Sstevel@tonic-gate void 30737c478bd9Sstevel@tonic-gate hat_unshare(hat_t *hat, caddr_t addr, size_t len, uint_t ismszc) 30747c478bd9Sstevel@tonic-gate { 30757173d045Sjosephb uint64_t vaddr = (uintptr_t)addr; 30767c478bd9Sstevel@tonic-gate uintptr_t eaddr = vaddr + len; 30777c478bd9Sstevel@tonic-gate htable_t *ht = NULL; 30787c478bd9Sstevel@tonic-gate uint_t need_demaps = 0; 3079250b7ff9Sjosephb int flags = HAT_UNLOAD_UNMAP; 3080250b7ff9Sjosephb level_t l; 30817c478bd9Sstevel@tonic-gate 30827c478bd9Sstevel@tonic-gate ASSERT(hat != kas.a_hat); 3083ae115bc7Smrj ASSERT(eaddr <= _userlimit); 30847c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(vaddr)); 30857c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(eaddr)); 3086843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 30877c478bd9Sstevel@tonic-gate 30887c478bd9Sstevel@tonic-gate /* 30897c478bd9Sstevel@tonic-gate * First go through and remove any shared pagetables. 30907c478bd9Sstevel@tonic-gate * 3091ae115bc7Smrj * Note that it's ok to delay the TLB shootdown till the entire range is 30927c478bd9Sstevel@tonic-gate * finished, because if hat_pageunload() were to unload a shared 3093ae115bc7Smrj * pagetable page, its hat_tlb_inval() will do a global TLB invalidate. 30947c478bd9Sstevel@tonic-gate */ 3095250b7ff9Sjosephb l = mmu.max_page_level; 3096250b7ff9Sjosephb if (l == mmu.max_level) 3097250b7ff9Sjosephb --l; 3098250b7ff9Sjosephb for (; l >= 0; --l) { 3099250b7ff9Sjosephb for (vaddr = (uintptr_t)addr; vaddr < eaddr; 3100250b7ff9Sjosephb vaddr = (vaddr & LEVEL_MASK(l + 1)) + LEVEL_SIZE(l + 1)) { 3101250b7ff9Sjosephb ASSERT(!IN_VA_HOLE(vaddr)); 3102250b7ff9Sjosephb /* 3103250b7ff9Sjosephb * find a pagetable that maps the current address 3104250b7ff9Sjosephb */ 3105250b7ff9Sjosephb ht = htable_lookup(hat, vaddr, l); 3106250b7ff9Sjosephb if (ht == NULL) 3107250b7ff9Sjosephb continue; 31087c478bd9Sstevel@tonic-gate if (ht->ht_flags & HTABLE_SHARED_PFN) { 31097c478bd9Sstevel@tonic-gate /* 3110250b7ff9Sjosephb * clear page count, set valid_cnt to 0, 3111250b7ff9Sjosephb * let htable_release() finish the job 31127c478bd9Sstevel@tonic-gate */ 3113250b7ff9Sjosephb hat->hat_ism_pgcnt -= ht->ht_valid_cnt << 3114250b7ff9Sjosephb (LEVEL_SHIFT(ht->ht_level) - MMU_PAGESHIFT); 31157c478bd9Sstevel@tonic-gate ht->ht_valid_cnt = 0; 31167c478bd9Sstevel@tonic-gate need_demaps = 1; 31177c478bd9Sstevel@tonic-gate } 31187c478bd9Sstevel@tonic-gate htable_release(ht); 31197c478bd9Sstevel@tonic-gate } 31207c478bd9Sstevel@tonic-gate } 31217c478bd9Sstevel@tonic-gate 31227c478bd9Sstevel@tonic-gate /* 31237c478bd9Sstevel@tonic-gate * flush the TLBs - since we're probably dealing with MANY mappings 31247c478bd9Sstevel@tonic-gate * we do just one CR3 reload. 31257c478bd9Sstevel@tonic-gate */ 31267c478bd9Sstevel@tonic-gate if (!(hat->hat_flags & HAT_FREEING) && need_demaps) 3127ae115bc7Smrj hat_tlb_inval(hat, DEMAP_ALL_ADDR); 31287c478bd9Sstevel@tonic-gate 31297c478bd9Sstevel@tonic-gate /* 31307c478bd9Sstevel@tonic-gate * Now go back and clean up any unaligned mappings that 31317c478bd9Sstevel@tonic-gate * couldn't share pagetables. 31327c478bd9Sstevel@tonic-gate */ 3133250b7ff9Sjosephb if (!is_it_dism(hat, addr)) 3134250b7ff9Sjosephb flags |= HAT_UNLOAD_UNLOCK; 3135250b7ff9Sjosephb hat_unload(hat, addr, len, flags); 3136843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 31377c478bd9Sstevel@tonic-gate } 31387c478bd9Sstevel@tonic-gate 31397c478bd9Sstevel@tonic-gate 31407c478bd9Sstevel@tonic-gate /* 31417c478bd9Sstevel@tonic-gate * hat_reserve() does nothing 31427c478bd9Sstevel@tonic-gate */ 31437c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 31447c478bd9Sstevel@tonic-gate void 31457c478bd9Sstevel@tonic-gate hat_reserve(struct as *as, caddr_t addr, size_t len) 31467c478bd9Sstevel@tonic-gate { 31477c478bd9Sstevel@tonic-gate } 31487c478bd9Sstevel@tonic-gate 31497c478bd9Sstevel@tonic-gate 31507c478bd9Sstevel@tonic-gate /* 31517c478bd9Sstevel@tonic-gate * Called when all mappings to a page should have write permission removed. 315221584dbcSPavel Tatashin * Mostly stolen from hat_pagesync() 31537c478bd9Sstevel@tonic-gate */ 31547c478bd9Sstevel@tonic-gate static void 31557c478bd9Sstevel@tonic-gate hati_page_clrwrt(struct page *pp) 31567c478bd9Sstevel@tonic-gate { 31577c478bd9Sstevel@tonic-gate hment_t *hm = NULL; 31587c478bd9Sstevel@tonic-gate htable_t *ht; 31597c478bd9Sstevel@tonic-gate uint_t entry; 31607c478bd9Sstevel@tonic-gate x86pte_t old; 31617c478bd9Sstevel@tonic-gate x86pte_t new; 31627c478bd9Sstevel@tonic-gate uint_t pszc = 0; 31637c478bd9Sstevel@tonic-gate 3164843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 31657c478bd9Sstevel@tonic-gate next_size: 31667c478bd9Sstevel@tonic-gate /* 31677c478bd9Sstevel@tonic-gate * walk thru the mapping list clearing write permission 31687c478bd9Sstevel@tonic-gate */ 31697c478bd9Sstevel@tonic-gate x86_hm_enter(pp); 31707c478bd9Sstevel@tonic-gate while ((hm = hment_walk(pp, &ht, &entry, hm)) != NULL) { 31717c478bd9Sstevel@tonic-gate if (ht->ht_level < pszc) 31727c478bd9Sstevel@tonic-gate continue; 31737c478bd9Sstevel@tonic-gate old = x86pte_get(ht, entry); 31747c478bd9Sstevel@tonic-gate 31757c478bd9Sstevel@tonic-gate for (;;) { 31767c478bd9Sstevel@tonic-gate /* 31777c478bd9Sstevel@tonic-gate * Is this mapping of interest? 31787c478bd9Sstevel@tonic-gate */ 31797c478bd9Sstevel@tonic-gate if (PTE2PFN(old, ht->ht_level) != pp->p_pagenum || 31807c478bd9Sstevel@tonic-gate PTE_GET(old, PT_WRITABLE) == 0) 31817c478bd9Sstevel@tonic-gate break; 31827c478bd9Sstevel@tonic-gate 31837c478bd9Sstevel@tonic-gate /* 31847c478bd9Sstevel@tonic-gate * Clear ref/mod writable bits. This requires cross 31857c478bd9Sstevel@tonic-gate * calls to ensure any executing TLBs see cleared bits. 31867c478bd9Sstevel@tonic-gate */ 31877c478bd9Sstevel@tonic-gate new = old; 31887c478bd9Sstevel@tonic-gate PTE_CLR(new, PT_REF | PT_MOD | PT_WRITABLE); 31897c478bd9Sstevel@tonic-gate old = hati_update_pte(ht, entry, old, new); 31907c478bd9Sstevel@tonic-gate if (old != 0) 31917c478bd9Sstevel@tonic-gate continue; 31927c478bd9Sstevel@tonic-gate 31937c478bd9Sstevel@tonic-gate break; 31947c478bd9Sstevel@tonic-gate } 31957c478bd9Sstevel@tonic-gate } 31967c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 31977c478bd9Sstevel@tonic-gate while (pszc < pp->p_szc) { 31987c478bd9Sstevel@tonic-gate page_t *tpp; 31997c478bd9Sstevel@tonic-gate pszc++; 32007c478bd9Sstevel@tonic-gate tpp = PP_GROUPLEADER(pp, pszc); 32017c478bd9Sstevel@tonic-gate if (pp != tpp) { 32027c478bd9Sstevel@tonic-gate pp = tpp; 32037c478bd9Sstevel@tonic-gate goto next_size; 32047c478bd9Sstevel@tonic-gate } 32057c478bd9Sstevel@tonic-gate } 3206843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 32077c478bd9Sstevel@tonic-gate } 32087c478bd9Sstevel@tonic-gate 32097c478bd9Sstevel@tonic-gate /* 32107c478bd9Sstevel@tonic-gate * void hat_page_setattr(pp, flag) 32117c478bd9Sstevel@tonic-gate * void hat_page_clrattr(pp, flag) 32127c478bd9Sstevel@tonic-gate * used to set/clr ref/mod bits. 32137c478bd9Sstevel@tonic-gate */ 32147c478bd9Sstevel@tonic-gate void 32157c478bd9Sstevel@tonic-gate hat_page_setattr(struct page *pp, uint_t flag) 32167c478bd9Sstevel@tonic-gate { 32177c478bd9Sstevel@tonic-gate vnode_t *vp = pp->p_vnode; 32187c478bd9Sstevel@tonic-gate kmutex_t *vphm = NULL; 32197c478bd9Sstevel@tonic-gate page_t **listp; 3220d9615970Sqiao int noshuffle; 3221d9615970Sqiao 3222d9615970Sqiao noshuffle = flag & P_NSH; 3223d9615970Sqiao flag &= ~P_NSH; 32247c478bd9Sstevel@tonic-gate 32257c478bd9Sstevel@tonic-gate if (PP_GETRM(pp, flag) == flag) 32267c478bd9Sstevel@tonic-gate return; 32277c478bd9Sstevel@tonic-gate 3228d9615970Sqiao if ((flag & P_MOD) != 0 && vp != NULL && IS_VMODSORT(vp) && 3229d9615970Sqiao !noshuffle) { 32307c478bd9Sstevel@tonic-gate vphm = page_vnode_mutex(vp); 32317c478bd9Sstevel@tonic-gate mutex_enter(vphm); 32327c478bd9Sstevel@tonic-gate } 32337c478bd9Sstevel@tonic-gate 32347c478bd9Sstevel@tonic-gate PP_SETRM(pp, flag); 32357c478bd9Sstevel@tonic-gate 32367c478bd9Sstevel@tonic-gate if (vphm != NULL) { 32377c478bd9Sstevel@tonic-gate 32387c478bd9Sstevel@tonic-gate /* 32397c478bd9Sstevel@tonic-gate * Some File Systems examine v_pages for NULL w/o 32407c478bd9Sstevel@tonic-gate * grabbing the vphm mutex. Must not let it become NULL when 32417c478bd9Sstevel@tonic-gate * pp is the only page on the list. 32427c478bd9Sstevel@tonic-gate */ 32437c478bd9Sstevel@tonic-gate if (pp->p_vpnext != pp) { 32447c478bd9Sstevel@tonic-gate page_vpsub(&vp->v_pages, pp); 32457c478bd9Sstevel@tonic-gate if (vp->v_pages != NULL) 32467c478bd9Sstevel@tonic-gate listp = &vp->v_pages->p_vpprev->p_vpnext; 32477c478bd9Sstevel@tonic-gate else 32487c478bd9Sstevel@tonic-gate listp = &vp->v_pages; 32497c478bd9Sstevel@tonic-gate page_vpadd(listp, pp); 32507c478bd9Sstevel@tonic-gate } 32517c478bd9Sstevel@tonic-gate mutex_exit(vphm); 32527c478bd9Sstevel@tonic-gate } 32537c478bd9Sstevel@tonic-gate } 32547c478bd9Sstevel@tonic-gate 32557c478bd9Sstevel@tonic-gate void 32567c478bd9Sstevel@tonic-gate hat_page_clrattr(struct page *pp, uint_t flag) 32577c478bd9Sstevel@tonic-gate { 32587c478bd9Sstevel@tonic-gate vnode_t *vp = pp->p_vnode; 32597c478bd9Sstevel@tonic-gate ASSERT(!(flag & ~(P_MOD | P_REF | P_RO))); 32607c478bd9Sstevel@tonic-gate 32617c478bd9Sstevel@tonic-gate /* 3262a71e32b6Sstans * Caller is expected to hold page's io lock for VMODSORT to work 3263a71e32b6Sstans * correctly with pvn_vplist_dirty() and pvn_getdirty() when mod 3264a71e32b6Sstans * bit is cleared. 3265a71e32b6Sstans * We don't have assert to avoid tripping some existing third party 3266a71e32b6Sstans * code. The dirty page is moved back to top of the v_page list 3267a71e32b6Sstans * after IO is done in pvn_write_done(). 32687c478bd9Sstevel@tonic-gate */ 32697c478bd9Sstevel@tonic-gate PP_CLRRM(pp, flag); 32707c478bd9Sstevel@tonic-gate 3271a71e32b6Sstans if ((flag & P_MOD) != 0 && vp != NULL && IS_VMODSORT(vp)) { 32727c478bd9Sstevel@tonic-gate 32737c478bd9Sstevel@tonic-gate /* 32747c478bd9Sstevel@tonic-gate * VMODSORT works by removing write permissions and getting 32757c478bd9Sstevel@tonic-gate * a fault when a page is made dirty. At this point 32767c478bd9Sstevel@tonic-gate * we need to remove write permission from all mappings 32777c478bd9Sstevel@tonic-gate * to this page. 32787c478bd9Sstevel@tonic-gate */ 32797c478bd9Sstevel@tonic-gate hati_page_clrwrt(pp); 32807c478bd9Sstevel@tonic-gate } 32817c478bd9Sstevel@tonic-gate } 32827c478bd9Sstevel@tonic-gate 32837c478bd9Sstevel@tonic-gate /* 32847c478bd9Sstevel@tonic-gate * If flag is specified, returns 0 if attribute is disabled 328521584dbcSPavel Tatashin * and non zero if enabled. If flag specifes multiple attributes 328621584dbcSPavel Tatashin * then returns 0 if ALL attributes are disabled. This is an advisory 32877c478bd9Sstevel@tonic-gate * call. 32887c478bd9Sstevel@tonic-gate */ 32897c478bd9Sstevel@tonic-gate uint_t 32907c478bd9Sstevel@tonic-gate hat_page_getattr(struct page *pp, uint_t flag) 32917c478bd9Sstevel@tonic-gate { 32927c478bd9Sstevel@tonic-gate return (PP_GETRM(pp, flag)); 32937c478bd9Sstevel@tonic-gate } 32947c478bd9Sstevel@tonic-gate 32957c478bd9Sstevel@tonic-gate 32967c478bd9Sstevel@tonic-gate /* 32977c478bd9Sstevel@tonic-gate * common code used by hat_pageunload() and hment_steal() 32987c478bd9Sstevel@tonic-gate */ 32997c478bd9Sstevel@tonic-gate hment_t * 33007c478bd9Sstevel@tonic-gate hati_page_unmap(page_t *pp, htable_t *ht, uint_t entry) 33017c478bd9Sstevel@tonic-gate { 33027c478bd9Sstevel@tonic-gate x86pte_t old_pte; 33037c478bd9Sstevel@tonic-gate pfn_t pfn = pp->p_pagenum; 33047c478bd9Sstevel@tonic-gate hment_t *hm; 33057c478bd9Sstevel@tonic-gate 33067c478bd9Sstevel@tonic-gate /* 33077c478bd9Sstevel@tonic-gate * We need to acquire a hold on the htable in order to 33087c478bd9Sstevel@tonic-gate * do the invalidate. We know the htable must exist, since 33097c478bd9Sstevel@tonic-gate * unmap's don't release the htable until after removing any 33107c478bd9Sstevel@tonic-gate * hment. Having x86_hm_enter() keeps that from proceeding. 33117c478bd9Sstevel@tonic-gate */ 33127c478bd9Sstevel@tonic-gate htable_acquire(ht); 33137c478bd9Sstevel@tonic-gate 33147c478bd9Sstevel@tonic-gate /* 33157c478bd9Sstevel@tonic-gate * Invalidate the PTE and remove the hment. 33167c478bd9Sstevel@tonic-gate */ 3317ae115bc7Smrj old_pte = x86pte_inval(ht, entry, 0, NULL); 3318aa2ed9e5Sjosephb if (PTE2PFN(old_pte, ht->ht_level) != pfn) { 3319ae115bc7Smrj panic("x86pte_inval() failure found PTE = " FMT_PTE 3320aa2ed9e5Sjosephb " pfn being unmapped is %lx ht=0x%lx entry=0x%x", 3321aa2ed9e5Sjosephb old_pte, pfn, (uintptr_t)ht, entry); 3322aa2ed9e5Sjosephb } 33237c478bd9Sstevel@tonic-gate 33247c478bd9Sstevel@tonic-gate /* 33257c478bd9Sstevel@tonic-gate * Clean up all the htable information for this mapping 33267c478bd9Sstevel@tonic-gate */ 33277c478bd9Sstevel@tonic-gate ASSERT(ht->ht_valid_cnt > 0); 33287c478bd9Sstevel@tonic-gate HTABLE_DEC(ht->ht_valid_cnt); 33297c478bd9Sstevel@tonic-gate PGCNT_DEC(ht->ht_hat, ht->ht_level); 33307c478bd9Sstevel@tonic-gate 33317c478bd9Sstevel@tonic-gate /* 33327c478bd9Sstevel@tonic-gate * sync ref/mod bits to the page_t 33337c478bd9Sstevel@tonic-gate */ 3334ae115bc7Smrj if (PTE_GET(old_pte, PT_SOFTWARE) < PT_NOSYNC) 33357c478bd9Sstevel@tonic-gate hati_sync_pte_to_page(pp, old_pte, ht->ht_level); 33367c478bd9Sstevel@tonic-gate 33377c478bd9Sstevel@tonic-gate /* 33387c478bd9Sstevel@tonic-gate * Remove the mapping list entry for this page. 33397c478bd9Sstevel@tonic-gate */ 33407c478bd9Sstevel@tonic-gate hm = hment_remove(pp, ht, entry); 33417c478bd9Sstevel@tonic-gate 33427c478bd9Sstevel@tonic-gate /* 33437c478bd9Sstevel@tonic-gate * drop the mapping list lock so that we might free the 33447c478bd9Sstevel@tonic-gate * hment and htable. 33457c478bd9Sstevel@tonic-gate */ 33467c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 33477c478bd9Sstevel@tonic-gate htable_release(ht); 33487c478bd9Sstevel@tonic-gate return (hm); 33497c478bd9Sstevel@tonic-gate } 33507c478bd9Sstevel@tonic-gate 3351a5652762Spraks extern int vpm_enable; 33527c478bd9Sstevel@tonic-gate /* 33537c478bd9Sstevel@tonic-gate * Unload all translations to a page. If the page is a subpage of a large 33547c478bd9Sstevel@tonic-gate * page, the large page mappings are also removed. 33557c478bd9Sstevel@tonic-gate * 33567c478bd9Sstevel@tonic-gate * The forceflags are unused. 33577c478bd9Sstevel@tonic-gate */ 33587c478bd9Sstevel@tonic-gate 33597c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 33607c478bd9Sstevel@tonic-gate static int 33617c478bd9Sstevel@tonic-gate hati_pageunload(struct page *pp, uint_t pg_szcd, uint_t forceflag) 33627c478bd9Sstevel@tonic-gate { 33637c478bd9Sstevel@tonic-gate page_t *cur_pp = pp; 33647c478bd9Sstevel@tonic-gate hment_t *hm; 33657c478bd9Sstevel@tonic-gate hment_t *prev; 33667c478bd9Sstevel@tonic-gate htable_t *ht; 33677c478bd9Sstevel@tonic-gate uint_t entry; 33687c478bd9Sstevel@tonic-gate level_t level; 33697c478bd9Sstevel@tonic-gate 3370843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 33716c9930aeSJoe Bonasera 33726c9930aeSJoe Bonasera /* 33736c9930aeSJoe Bonasera * prevent recursion due to kmem_free() 33746c9930aeSJoe Bonasera */ 33756c9930aeSJoe Bonasera ++curthread->t_hatdepth; 33766c9930aeSJoe Bonasera ASSERT(curthread->t_hatdepth < 16); 33776c9930aeSJoe Bonasera 3378a5652762Spraks #if defined(__amd64) 3379a5652762Spraks /* 3380a5652762Spraks * clear the vpm ref. 3381a5652762Spraks */ 3382a5652762Spraks if (vpm_enable) { 3383a5652762Spraks pp->p_vpmref = 0; 3384a5652762Spraks } 3385a5652762Spraks #endif 33867c478bd9Sstevel@tonic-gate /* 33877c478bd9Sstevel@tonic-gate * The loop with next_size handles pages with multiple pagesize mappings 33887c478bd9Sstevel@tonic-gate */ 33897c478bd9Sstevel@tonic-gate next_size: 33907c478bd9Sstevel@tonic-gate for (;;) { 33917c478bd9Sstevel@tonic-gate 33927c478bd9Sstevel@tonic-gate /* 33937c478bd9Sstevel@tonic-gate * Get a mapping list entry 33947c478bd9Sstevel@tonic-gate */ 33957c478bd9Sstevel@tonic-gate x86_hm_enter(cur_pp); 33967c478bd9Sstevel@tonic-gate for (prev = NULL; ; prev = hm) { 33977c478bd9Sstevel@tonic-gate hm = hment_walk(cur_pp, &ht, &entry, prev); 33987c478bd9Sstevel@tonic-gate if (hm == NULL) { 33997c478bd9Sstevel@tonic-gate x86_hm_exit(cur_pp); 34007c478bd9Sstevel@tonic-gate 34017c478bd9Sstevel@tonic-gate /* 34027c478bd9Sstevel@tonic-gate * If not part of a larger page, we're done. 34037c478bd9Sstevel@tonic-gate */ 3404ae115bc7Smrj if (cur_pp->p_szc <= pg_szcd) { 34056c9930aeSJoe Bonasera ASSERT(curthread->t_hatdepth > 0); 34066c9930aeSJoe Bonasera --curthread->t_hatdepth; 3407843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 34087c478bd9Sstevel@tonic-gate return (0); 3409ae115bc7Smrj } 34107c478bd9Sstevel@tonic-gate 34117c478bd9Sstevel@tonic-gate /* 34127c478bd9Sstevel@tonic-gate * Else check the next larger page size. 34137c478bd9Sstevel@tonic-gate * hat_page_demote() may decrease p_szc 34147c478bd9Sstevel@tonic-gate * but that's ok we'll just take an extra 34157c478bd9Sstevel@tonic-gate * trip discover there're no larger mappings 34167c478bd9Sstevel@tonic-gate * and return. 34177c478bd9Sstevel@tonic-gate */ 34187c478bd9Sstevel@tonic-gate ++pg_szcd; 34197c478bd9Sstevel@tonic-gate cur_pp = PP_GROUPLEADER(cur_pp, pg_szcd); 34207c478bd9Sstevel@tonic-gate goto next_size; 34217c478bd9Sstevel@tonic-gate } 34227c478bd9Sstevel@tonic-gate 34237c478bd9Sstevel@tonic-gate /* 34247c478bd9Sstevel@tonic-gate * If this mapping size matches, remove it. 34257c478bd9Sstevel@tonic-gate */ 34267c478bd9Sstevel@tonic-gate level = ht->ht_level; 34277c478bd9Sstevel@tonic-gate if (level == pg_szcd) 34287c478bd9Sstevel@tonic-gate break; 34297c478bd9Sstevel@tonic-gate } 34307c478bd9Sstevel@tonic-gate 34317c478bd9Sstevel@tonic-gate /* 34327c478bd9Sstevel@tonic-gate * Remove the mapping list entry for this page. 34337c478bd9Sstevel@tonic-gate * Note this does the x86_hm_exit() for us. 34347c478bd9Sstevel@tonic-gate */ 34357c478bd9Sstevel@tonic-gate hm = hati_page_unmap(cur_pp, ht, entry); 34367c478bd9Sstevel@tonic-gate if (hm != NULL) 34377c478bd9Sstevel@tonic-gate hment_free(hm); 34387c478bd9Sstevel@tonic-gate } 34397c478bd9Sstevel@tonic-gate } 34407c478bd9Sstevel@tonic-gate 34417c478bd9Sstevel@tonic-gate int 34427c478bd9Sstevel@tonic-gate hat_pageunload(struct page *pp, uint_t forceflag) 34437c478bd9Sstevel@tonic-gate { 34447c478bd9Sstevel@tonic-gate ASSERT(PAGE_EXCL(pp)); 34457c478bd9Sstevel@tonic-gate return (hati_pageunload(pp, 0, forceflag)); 34467c478bd9Sstevel@tonic-gate } 34477c478bd9Sstevel@tonic-gate 34487c478bd9Sstevel@tonic-gate /* 34497c478bd9Sstevel@tonic-gate * Unload all large mappings to pp and reduce by 1 p_szc field of every large 34507c478bd9Sstevel@tonic-gate * page level that included pp. 34517c478bd9Sstevel@tonic-gate * 34527c478bd9Sstevel@tonic-gate * pp must be locked EXCL. Even though no other constituent pages are locked 34537c478bd9Sstevel@tonic-gate * it's legal to unload large mappings to pp because all constituent pages of 34547c478bd9Sstevel@tonic-gate * large locked mappings have to be locked SHARED. therefore if we have EXCL 34557c478bd9Sstevel@tonic-gate * lock on one of constituent pages none of the large mappings to pp are 34567c478bd9Sstevel@tonic-gate * locked. 34577c478bd9Sstevel@tonic-gate * 34587c478bd9Sstevel@tonic-gate * Change (always decrease) p_szc field starting from the last constituent 34597c478bd9Sstevel@tonic-gate * page and ending with root constituent page so that root's pszc always shows 34607c478bd9Sstevel@tonic-gate * the area where hat_page_demote() may be active. 34617c478bd9Sstevel@tonic-gate * 34627c478bd9Sstevel@tonic-gate * This mechanism is only used for file system pages where it's not always 34637c478bd9Sstevel@tonic-gate * possible to get EXCL locks on all constituent pages to demote the size code 34647c478bd9Sstevel@tonic-gate * (as is done for anonymous or kernel large pages). 34657c478bd9Sstevel@tonic-gate */ 34667c478bd9Sstevel@tonic-gate void 34677c478bd9Sstevel@tonic-gate hat_page_demote(page_t *pp) 34687c478bd9Sstevel@tonic-gate { 34697c478bd9Sstevel@tonic-gate uint_t pszc; 34707c478bd9Sstevel@tonic-gate uint_t rszc; 34717c478bd9Sstevel@tonic-gate uint_t szc; 34727c478bd9Sstevel@tonic-gate page_t *rootpp; 34737c478bd9Sstevel@tonic-gate page_t *firstpp; 34747c478bd9Sstevel@tonic-gate page_t *lastpp; 34757c478bd9Sstevel@tonic-gate pgcnt_t pgcnt; 34767c478bd9Sstevel@tonic-gate 34777c478bd9Sstevel@tonic-gate ASSERT(PAGE_EXCL(pp)); 34787c478bd9Sstevel@tonic-gate ASSERT(!PP_ISFREE(pp)); 34797c478bd9Sstevel@tonic-gate ASSERT(page_szc_lock_assert(pp)); 34807c478bd9Sstevel@tonic-gate 34817c478bd9Sstevel@tonic-gate if (pp->p_szc == 0) 34827c478bd9Sstevel@tonic-gate return; 34837c478bd9Sstevel@tonic-gate 34847c478bd9Sstevel@tonic-gate rootpp = PP_GROUPLEADER(pp, 1); 34857c478bd9Sstevel@tonic-gate (void) hati_pageunload(rootpp, 1, HAT_FORCE_PGUNLOAD); 34867c478bd9Sstevel@tonic-gate 34877c478bd9Sstevel@tonic-gate /* 34887c478bd9Sstevel@tonic-gate * all large mappings to pp are gone 34897c478bd9Sstevel@tonic-gate * and no new can be setup since pp is locked exclusively. 34907c478bd9Sstevel@tonic-gate * 34917c478bd9Sstevel@tonic-gate * Lock the root to make sure there's only one hat_page_demote() 34927c478bd9Sstevel@tonic-gate * outstanding within the area of this root's pszc. 34937c478bd9Sstevel@tonic-gate * 34947c478bd9Sstevel@tonic-gate * Second potential hat_page_demote() is already eliminated by upper 34957c478bd9Sstevel@tonic-gate * VM layer via page_szc_lock() but we don't rely on it and use our 34967c478bd9Sstevel@tonic-gate * own locking (so that upper layer locking can be changed without 34977c478bd9Sstevel@tonic-gate * assumptions that hat depends on upper layer VM to prevent multiple 34987c478bd9Sstevel@tonic-gate * hat_page_demote() to be issued simultaneously to the same large 34997c478bd9Sstevel@tonic-gate * page). 35007c478bd9Sstevel@tonic-gate */ 35017c478bd9Sstevel@tonic-gate again: 35027c478bd9Sstevel@tonic-gate pszc = pp->p_szc; 35037c478bd9Sstevel@tonic-gate if (pszc == 0) 35047c478bd9Sstevel@tonic-gate return; 35057c478bd9Sstevel@tonic-gate rootpp = PP_GROUPLEADER(pp, pszc); 35067c478bd9Sstevel@tonic-gate x86_hm_enter(rootpp); 35077c478bd9Sstevel@tonic-gate /* 35087c478bd9Sstevel@tonic-gate * If root's p_szc is different from pszc we raced with another 35097c478bd9Sstevel@tonic-gate * hat_page_demote(). Drop the lock and try to find the root again. 35107c478bd9Sstevel@tonic-gate * If root's p_szc is greater than pszc previous hat_page_demote() is 35117c478bd9Sstevel@tonic-gate * not done yet. Take and release mlist lock of root's root to wait 35127c478bd9Sstevel@tonic-gate * for previous hat_page_demote() to complete. 35137c478bd9Sstevel@tonic-gate */ 35147c478bd9Sstevel@tonic-gate if ((rszc = rootpp->p_szc) != pszc) { 35157c478bd9Sstevel@tonic-gate x86_hm_exit(rootpp); 35167c478bd9Sstevel@tonic-gate if (rszc > pszc) { 35177c478bd9Sstevel@tonic-gate /* p_szc of a locked non free page can't increase */ 35187c478bd9Sstevel@tonic-gate ASSERT(pp != rootpp); 35197c478bd9Sstevel@tonic-gate 35207c478bd9Sstevel@tonic-gate rootpp = PP_GROUPLEADER(rootpp, rszc); 35217c478bd9Sstevel@tonic-gate x86_hm_enter(rootpp); 35227c478bd9Sstevel@tonic-gate x86_hm_exit(rootpp); 35237c478bd9Sstevel@tonic-gate } 35247c478bd9Sstevel@tonic-gate goto again; 35257c478bd9Sstevel@tonic-gate } 35267c478bd9Sstevel@tonic-gate ASSERT(pp->p_szc == pszc); 35277c478bd9Sstevel@tonic-gate 35287c478bd9Sstevel@tonic-gate /* 35297c478bd9Sstevel@tonic-gate * Decrement by 1 p_szc of every constituent page of a region that 35307c478bd9Sstevel@tonic-gate * covered pp. For example if original szc is 3 it gets changed to 2 35317c478bd9Sstevel@tonic-gate * everywhere except in region 2 that covered pp. Region 2 that 35327c478bd9Sstevel@tonic-gate * covered pp gets demoted to 1 everywhere except in region 1 that 35337c478bd9Sstevel@tonic-gate * covered pp. The region 1 that covered pp is demoted to region 35347c478bd9Sstevel@tonic-gate * 0. It's done this way because from region 3 we removed level 3 35357c478bd9Sstevel@tonic-gate * mappings, from region 2 that covered pp we removed level 2 mappings 35367c478bd9Sstevel@tonic-gate * and from region 1 that covered pp we removed level 1 mappings. All 35377c478bd9Sstevel@tonic-gate * changes are done from from high pfn's to low pfn's so that roots 35387c478bd9Sstevel@tonic-gate * are changed last allowing one to know the largest region where 35397c478bd9Sstevel@tonic-gate * hat_page_demote() is stil active by only looking at the root page. 35407c478bd9Sstevel@tonic-gate * 35417c478bd9Sstevel@tonic-gate * This algorithm is implemented in 2 while loops. First loop changes 35427c478bd9Sstevel@tonic-gate * p_szc of pages to the right of pp's level 1 region and second 35437c478bd9Sstevel@tonic-gate * loop changes p_szc of pages of level 1 region that covers pp 35447c478bd9Sstevel@tonic-gate * and all pages to the left of level 1 region that covers pp. 35457c478bd9Sstevel@tonic-gate * In the first loop p_szc keeps dropping with every iteration 35467c478bd9Sstevel@tonic-gate * and in the second loop it keeps increasing with every iteration. 35477c478bd9Sstevel@tonic-gate * 35487c478bd9Sstevel@tonic-gate * First loop description: Demote pages to the right of pp outside of 35497c478bd9Sstevel@tonic-gate * level 1 region that covers pp. In every iteration of the while 35507c478bd9Sstevel@tonic-gate * loop below find the last page of szc region and the first page of 35517c478bd9Sstevel@tonic-gate * (szc - 1) region that is immediately to the right of (szc - 1) 35527c478bd9Sstevel@tonic-gate * region that covers pp. From last such page to first such page 35537c478bd9Sstevel@tonic-gate * change every page's szc to szc - 1. Decrement szc and continue 35547c478bd9Sstevel@tonic-gate * looping until szc is 1. If pp belongs to the last (szc - 1) region 35557c478bd9Sstevel@tonic-gate * of szc region skip to the next iteration. 35567c478bd9Sstevel@tonic-gate */ 35577c478bd9Sstevel@tonic-gate szc = pszc; 35587c478bd9Sstevel@tonic-gate while (szc > 1) { 35597c478bd9Sstevel@tonic-gate lastpp = PP_GROUPLEADER(pp, szc); 35607c478bd9Sstevel@tonic-gate pgcnt = page_get_pagecnt(szc); 35617c478bd9Sstevel@tonic-gate lastpp += pgcnt - 1; 35627c478bd9Sstevel@tonic-gate firstpp = PP_GROUPLEADER(pp, (szc - 1)); 35637c478bd9Sstevel@tonic-gate pgcnt = page_get_pagecnt(szc - 1); 35647c478bd9Sstevel@tonic-gate if (lastpp - firstpp < pgcnt) { 35657c478bd9Sstevel@tonic-gate szc--; 35667c478bd9Sstevel@tonic-gate continue; 35677c478bd9Sstevel@tonic-gate } 35687c478bd9Sstevel@tonic-gate firstpp += pgcnt; 35697c478bd9Sstevel@tonic-gate while (lastpp != firstpp) { 35707c478bd9Sstevel@tonic-gate ASSERT(lastpp->p_szc == pszc); 35717c478bd9Sstevel@tonic-gate lastpp->p_szc = szc - 1; 35727c478bd9Sstevel@tonic-gate lastpp--; 35737c478bd9Sstevel@tonic-gate } 35747c478bd9Sstevel@tonic-gate firstpp->p_szc = szc - 1; 35757c478bd9Sstevel@tonic-gate szc--; 35767c478bd9Sstevel@tonic-gate } 35777c478bd9Sstevel@tonic-gate 35787c478bd9Sstevel@tonic-gate /* 35797c478bd9Sstevel@tonic-gate * Second loop description: 35807c478bd9Sstevel@tonic-gate * First iteration changes p_szc to 0 of every 35817c478bd9Sstevel@tonic-gate * page of level 1 region that covers pp. 35827c478bd9Sstevel@tonic-gate * Subsequent iterations find last page of szc region 35837c478bd9Sstevel@tonic-gate * immediately to the left of szc region that covered pp 35847c478bd9Sstevel@tonic-gate * and first page of (szc + 1) region that covers pp. 35857c478bd9Sstevel@tonic-gate * From last to first page change p_szc of every page to szc. 35867c478bd9Sstevel@tonic-gate * Increment szc and continue looping until szc is pszc. 35877c478bd9Sstevel@tonic-gate * If pp belongs to the fist szc region of (szc + 1) region 35887c478bd9Sstevel@tonic-gate * skip to the next iteration. 35897c478bd9Sstevel@tonic-gate * 35907c478bd9Sstevel@tonic-gate */ 35917c478bd9Sstevel@tonic-gate szc = 0; 35927c478bd9Sstevel@tonic-gate while (szc < pszc) { 35937c478bd9Sstevel@tonic-gate firstpp = PP_GROUPLEADER(pp, (szc + 1)); 35947c478bd9Sstevel@tonic-gate if (szc == 0) { 35957c478bd9Sstevel@tonic-gate pgcnt = page_get_pagecnt(1); 35967c478bd9Sstevel@tonic-gate lastpp = firstpp + (pgcnt - 1); 35977c478bd9Sstevel@tonic-gate } else { 35987c478bd9Sstevel@tonic-gate lastpp = PP_GROUPLEADER(pp, szc); 35997c478bd9Sstevel@tonic-gate if (firstpp == lastpp) { 36007c478bd9Sstevel@tonic-gate szc++; 36017c478bd9Sstevel@tonic-gate continue; 36027c478bd9Sstevel@tonic-gate } 36037c478bd9Sstevel@tonic-gate lastpp--; 36047c478bd9Sstevel@tonic-gate pgcnt = page_get_pagecnt(szc); 36057c478bd9Sstevel@tonic-gate } 36067c478bd9Sstevel@tonic-gate while (lastpp != firstpp) { 36077c478bd9Sstevel@tonic-gate ASSERT(lastpp->p_szc == pszc); 36087c478bd9Sstevel@tonic-gate lastpp->p_szc = szc; 36097c478bd9Sstevel@tonic-gate lastpp--; 36107c478bd9Sstevel@tonic-gate } 36117c478bd9Sstevel@tonic-gate firstpp->p_szc = szc; 36127c478bd9Sstevel@tonic-gate if (firstpp == rootpp) 36137c478bd9Sstevel@tonic-gate break; 36147c478bd9Sstevel@tonic-gate szc++; 36157c478bd9Sstevel@tonic-gate } 36167c478bd9Sstevel@tonic-gate x86_hm_exit(rootpp); 36177c478bd9Sstevel@tonic-gate } 36187c478bd9Sstevel@tonic-gate 36197c478bd9Sstevel@tonic-gate /* 36207c478bd9Sstevel@tonic-gate * get hw stats from hardware into page struct and reset hw stats 36217c478bd9Sstevel@tonic-gate * returns attributes of page 36227c478bd9Sstevel@tonic-gate * Flags for hat_pagesync, hat_getstat, hat_sync 36237c478bd9Sstevel@tonic-gate * 36247c478bd9Sstevel@tonic-gate * define HAT_SYNC_ZERORM 0x01 36257c478bd9Sstevel@tonic-gate * 36267c478bd9Sstevel@tonic-gate * Additional flags for hat_pagesync 36277c478bd9Sstevel@tonic-gate * 36287c478bd9Sstevel@tonic-gate * define HAT_SYNC_STOPON_REF 0x02 36297c478bd9Sstevel@tonic-gate * define HAT_SYNC_STOPON_MOD 0x04 36307c478bd9Sstevel@tonic-gate * define HAT_SYNC_STOPON_RM 0x06 36317c478bd9Sstevel@tonic-gate * define HAT_SYNC_STOPON_SHARED 0x08 36327c478bd9Sstevel@tonic-gate */ 36337c478bd9Sstevel@tonic-gate uint_t 36347c478bd9Sstevel@tonic-gate hat_pagesync(struct page *pp, uint_t flags) 36357c478bd9Sstevel@tonic-gate { 36367c478bd9Sstevel@tonic-gate hment_t *hm = NULL; 36377c478bd9Sstevel@tonic-gate htable_t *ht; 36387c478bd9Sstevel@tonic-gate uint_t entry; 36397c478bd9Sstevel@tonic-gate x86pte_t old, save_old; 36407c478bd9Sstevel@tonic-gate x86pte_t new; 36417c478bd9Sstevel@tonic-gate uchar_t nrmbits = P_REF|P_MOD|P_RO; 36427c478bd9Sstevel@tonic-gate extern ulong_t po_share; 36437c478bd9Sstevel@tonic-gate page_t *save_pp = pp; 36447c478bd9Sstevel@tonic-gate uint_t pszc = 0; 36457c478bd9Sstevel@tonic-gate 36467c478bd9Sstevel@tonic-gate ASSERT(PAGE_LOCKED(pp) || panicstr); 36477c478bd9Sstevel@tonic-gate 36487c478bd9Sstevel@tonic-gate if (PP_ISRO(pp) && (flags & HAT_SYNC_STOPON_MOD)) 36497c478bd9Sstevel@tonic-gate return (pp->p_nrm & nrmbits); 36507c478bd9Sstevel@tonic-gate 36517c478bd9Sstevel@tonic-gate if ((flags & HAT_SYNC_ZERORM) == 0) { 36527c478bd9Sstevel@tonic-gate 36537c478bd9Sstevel@tonic-gate if ((flags & HAT_SYNC_STOPON_REF) != 0 && PP_ISREF(pp)) 36547c478bd9Sstevel@tonic-gate return (pp->p_nrm & nrmbits); 36557c478bd9Sstevel@tonic-gate 36567c478bd9Sstevel@tonic-gate if ((flags & HAT_SYNC_STOPON_MOD) != 0 && PP_ISMOD(pp)) 36577c478bd9Sstevel@tonic-gate return (pp->p_nrm & nrmbits); 36587c478bd9Sstevel@tonic-gate 36597c478bd9Sstevel@tonic-gate if ((flags & HAT_SYNC_STOPON_SHARED) != 0 && 36607c478bd9Sstevel@tonic-gate hat_page_getshare(pp) > po_share) { 36617c478bd9Sstevel@tonic-gate if (PP_ISRO(pp)) 36627c478bd9Sstevel@tonic-gate PP_SETREF(pp); 36637c478bd9Sstevel@tonic-gate return (pp->p_nrm & nrmbits); 36647c478bd9Sstevel@tonic-gate } 36657c478bd9Sstevel@tonic-gate } 36667c478bd9Sstevel@tonic-gate 3667843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 36687c478bd9Sstevel@tonic-gate next_size: 36697c478bd9Sstevel@tonic-gate /* 36707c478bd9Sstevel@tonic-gate * walk thru the mapping list syncing (and clearing) ref/mod bits. 36717c478bd9Sstevel@tonic-gate */ 36727c478bd9Sstevel@tonic-gate x86_hm_enter(pp); 36737c478bd9Sstevel@tonic-gate while ((hm = hment_walk(pp, &ht, &entry, hm)) != NULL) { 36747c478bd9Sstevel@tonic-gate if (ht->ht_level < pszc) 36757c478bd9Sstevel@tonic-gate continue; 36767c478bd9Sstevel@tonic-gate old = x86pte_get(ht, entry); 36777c478bd9Sstevel@tonic-gate try_again: 36787c478bd9Sstevel@tonic-gate 36797c478bd9Sstevel@tonic-gate ASSERT(PTE2PFN(old, ht->ht_level) == pp->p_pagenum); 36807c478bd9Sstevel@tonic-gate 36817c478bd9Sstevel@tonic-gate if (PTE_GET(old, PT_REF | PT_MOD) == 0) 36827c478bd9Sstevel@tonic-gate continue; 36837c478bd9Sstevel@tonic-gate 36847c478bd9Sstevel@tonic-gate save_old = old; 36857c478bd9Sstevel@tonic-gate if ((flags & HAT_SYNC_ZERORM) != 0) { 36867c478bd9Sstevel@tonic-gate 36877c478bd9Sstevel@tonic-gate /* 36887c478bd9Sstevel@tonic-gate * Need to clear ref or mod bits. Need to demap 36897c478bd9Sstevel@tonic-gate * to make sure any executing TLBs see cleared bits. 36907c478bd9Sstevel@tonic-gate */ 36917c478bd9Sstevel@tonic-gate new = old; 36927c478bd9Sstevel@tonic-gate PTE_CLR(new, PT_REF | PT_MOD); 36937c478bd9Sstevel@tonic-gate old = hati_update_pte(ht, entry, old, new); 36947c478bd9Sstevel@tonic-gate if (old != 0) 36957c478bd9Sstevel@tonic-gate goto try_again; 36967c478bd9Sstevel@tonic-gate 36977c478bd9Sstevel@tonic-gate old = save_old; 36987c478bd9Sstevel@tonic-gate } 36997c478bd9Sstevel@tonic-gate 37007c478bd9Sstevel@tonic-gate /* 37017c478bd9Sstevel@tonic-gate * Sync the PTE 37027c478bd9Sstevel@tonic-gate */ 3703ae115bc7Smrj if (!(flags & HAT_SYNC_ZERORM) && 3704ae115bc7Smrj PTE_GET(old, PT_SOFTWARE) <= PT_NOSYNC) 37057c478bd9Sstevel@tonic-gate hati_sync_pte_to_page(pp, old, ht->ht_level); 37067c478bd9Sstevel@tonic-gate 37077c478bd9Sstevel@tonic-gate /* 37087c478bd9Sstevel@tonic-gate * can stop short if we found a ref'd or mod'd page 37097c478bd9Sstevel@tonic-gate */ 37107c478bd9Sstevel@tonic-gate if ((flags & HAT_SYNC_STOPON_MOD) && PP_ISMOD(save_pp) || 37117c478bd9Sstevel@tonic-gate (flags & HAT_SYNC_STOPON_REF) && PP_ISREF(save_pp)) { 37127c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 3713ae115bc7Smrj goto done; 37147c478bd9Sstevel@tonic-gate } 37157c478bd9Sstevel@tonic-gate } 37167c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 37177c478bd9Sstevel@tonic-gate while (pszc < pp->p_szc) { 37187c478bd9Sstevel@tonic-gate page_t *tpp; 37197c478bd9Sstevel@tonic-gate pszc++; 37207c478bd9Sstevel@tonic-gate tpp = PP_GROUPLEADER(pp, pszc); 37217c478bd9Sstevel@tonic-gate if (pp != tpp) { 37227c478bd9Sstevel@tonic-gate pp = tpp; 37237c478bd9Sstevel@tonic-gate goto next_size; 37247c478bd9Sstevel@tonic-gate } 37257c478bd9Sstevel@tonic-gate } 3726ae115bc7Smrj done: 3727843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 37287c478bd9Sstevel@tonic-gate return (save_pp->p_nrm & nrmbits); 37297c478bd9Sstevel@tonic-gate } 37307c478bd9Sstevel@tonic-gate 37317c478bd9Sstevel@tonic-gate /* 37327c478bd9Sstevel@tonic-gate * returns approx number of mappings to this pp. A return of 0 implies 37337c478bd9Sstevel@tonic-gate * there are no mappings to the page. 37347c478bd9Sstevel@tonic-gate */ 37357c478bd9Sstevel@tonic-gate ulong_t 37367c478bd9Sstevel@tonic-gate hat_page_getshare(page_t *pp) 37377c478bd9Sstevel@tonic-gate { 37387c478bd9Sstevel@tonic-gate uint_t cnt; 37397c478bd9Sstevel@tonic-gate cnt = hment_mapcnt(pp); 3740a5652762Spraks #if defined(__amd64) 3741a5652762Spraks if (vpm_enable && pp->p_vpmref) { 3742a5652762Spraks cnt += 1; 3743a5652762Spraks } 3744a5652762Spraks #endif 37457c478bd9Sstevel@tonic-gate return (cnt); 37467c478bd9Sstevel@tonic-gate } 37477c478bd9Sstevel@tonic-gate 374805d3dc4bSpaulsan /* 374905d3dc4bSpaulsan * Return 1 the number of mappings exceeds sh_thresh. Return 0 375005d3dc4bSpaulsan * otherwise. 375105d3dc4bSpaulsan */ 375205d3dc4bSpaulsan int 375305d3dc4bSpaulsan hat_page_checkshare(page_t *pp, ulong_t sh_thresh) 375405d3dc4bSpaulsan { 375505d3dc4bSpaulsan return (hat_page_getshare(pp) > sh_thresh); 375605d3dc4bSpaulsan } 375705d3dc4bSpaulsan 37587c478bd9Sstevel@tonic-gate /* 37597c478bd9Sstevel@tonic-gate * hat_softlock isn't supported anymore 37607c478bd9Sstevel@tonic-gate */ 37617c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 37627c478bd9Sstevel@tonic-gate faultcode_t 37637c478bd9Sstevel@tonic-gate hat_softlock( 37647c478bd9Sstevel@tonic-gate hat_t *hat, 37657c478bd9Sstevel@tonic-gate caddr_t addr, 37667c478bd9Sstevel@tonic-gate size_t *len, 37677c478bd9Sstevel@tonic-gate struct page **page_array, 37687c478bd9Sstevel@tonic-gate uint_t flags) 37697c478bd9Sstevel@tonic-gate { 37707c478bd9Sstevel@tonic-gate return (FC_NOSUPPORT); 37717c478bd9Sstevel@tonic-gate } 37727c478bd9Sstevel@tonic-gate 37737c478bd9Sstevel@tonic-gate 37747c478bd9Sstevel@tonic-gate 37757c478bd9Sstevel@tonic-gate /* 37767c478bd9Sstevel@tonic-gate * Routine to expose supported HAT features to platform independent code. 37777c478bd9Sstevel@tonic-gate */ 37787c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 37797c478bd9Sstevel@tonic-gate int 37807c478bd9Sstevel@tonic-gate hat_supported(enum hat_features feature, void *arg) 37817c478bd9Sstevel@tonic-gate { 37827c478bd9Sstevel@tonic-gate switch (feature) { 37837c478bd9Sstevel@tonic-gate 37847c478bd9Sstevel@tonic-gate case HAT_SHARED_PT: /* this is really ISM */ 37857c478bd9Sstevel@tonic-gate return (1); 37867c478bd9Sstevel@tonic-gate 37877c478bd9Sstevel@tonic-gate case HAT_DYNAMIC_ISM_UNMAP: 37887c478bd9Sstevel@tonic-gate return (0); 37897c478bd9Sstevel@tonic-gate 37907c478bd9Sstevel@tonic-gate case HAT_VMODSORT: 37917c478bd9Sstevel@tonic-gate return (1); 37927c478bd9Sstevel@tonic-gate 379305d3dc4bSpaulsan case HAT_SHARED_REGIONS: 379405d3dc4bSpaulsan return (0); 379505d3dc4bSpaulsan 37967c478bd9Sstevel@tonic-gate default: 37977c478bd9Sstevel@tonic-gate panic("hat_supported() - unknown feature"); 37987c478bd9Sstevel@tonic-gate } 37997c478bd9Sstevel@tonic-gate return (0); 38007c478bd9Sstevel@tonic-gate } 38017c478bd9Sstevel@tonic-gate 38027c478bd9Sstevel@tonic-gate /* 38037c478bd9Sstevel@tonic-gate * Called when a thread is exiting and has been switched to the kernel AS 38047c478bd9Sstevel@tonic-gate */ 38057c478bd9Sstevel@tonic-gate void 38067c478bd9Sstevel@tonic-gate hat_thread_exit(kthread_t *thd) 38077c478bd9Sstevel@tonic-gate { 38087c478bd9Sstevel@tonic-gate ASSERT(thd->t_procp->p_as == &kas); 3809843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 38107c478bd9Sstevel@tonic-gate hat_switch(thd->t_procp->p_as->a_hat); 3811843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 38127c478bd9Sstevel@tonic-gate } 38137c478bd9Sstevel@tonic-gate 38147c478bd9Sstevel@tonic-gate /* 38157c478bd9Sstevel@tonic-gate * Setup the given brand new hat structure as the new HAT on this cpu's mmu. 38167c478bd9Sstevel@tonic-gate */ 38177c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 38187c478bd9Sstevel@tonic-gate void 38197c478bd9Sstevel@tonic-gate hat_setup(hat_t *hat, int flags) 38207c478bd9Sstevel@tonic-gate { 3821843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 38227c478bd9Sstevel@tonic-gate kpreempt_disable(); 38237c478bd9Sstevel@tonic-gate 38247c478bd9Sstevel@tonic-gate hat_switch(hat); 38257c478bd9Sstevel@tonic-gate 38267c478bd9Sstevel@tonic-gate kpreempt_enable(); 3827843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 38287c478bd9Sstevel@tonic-gate } 38297c478bd9Sstevel@tonic-gate 38307c478bd9Sstevel@tonic-gate /* 38317c478bd9Sstevel@tonic-gate * Prepare for a CPU private mapping for the given address. 38327c478bd9Sstevel@tonic-gate * 38337c478bd9Sstevel@tonic-gate * The address can only be used from a single CPU and can be remapped 38347c478bd9Sstevel@tonic-gate * using hat_mempte_remap(). Return the address of the PTE. 38357c478bd9Sstevel@tonic-gate * 38367c478bd9Sstevel@tonic-gate * We do the htable_create() if necessary and increment the valid count so 38377c478bd9Sstevel@tonic-gate * the htable can't disappear. We also hat_devload() the page table into 38387c478bd9Sstevel@tonic-gate * kernel so that the PTE is quickly accessed. 38397c478bd9Sstevel@tonic-gate */ 3840ae115bc7Smrj hat_mempte_t 3841ae115bc7Smrj hat_mempte_setup(caddr_t addr) 38427c478bd9Sstevel@tonic-gate { 38437c478bd9Sstevel@tonic-gate uintptr_t va = (uintptr_t)addr; 38447c478bd9Sstevel@tonic-gate htable_t *ht; 38457c478bd9Sstevel@tonic-gate uint_t entry; 38467c478bd9Sstevel@tonic-gate x86pte_t oldpte; 3847ae115bc7Smrj hat_mempte_t p; 38487c478bd9Sstevel@tonic-gate 38497c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(va)); 38507c478bd9Sstevel@tonic-gate ASSERT(!IN_VA_HOLE(va)); 3851aac11643Sjosephb ++curthread->t_hatdepth; 3852551bc2a6Smrj XPV_DISALLOW_MIGRATE(); 38537c478bd9Sstevel@tonic-gate ht = htable_getpte(kas.a_hat, va, &entry, &oldpte, 0); 38547c478bd9Sstevel@tonic-gate if (ht == NULL) { 38557c478bd9Sstevel@tonic-gate ht = htable_create(kas.a_hat, va, 0, NULL); 38567c478bd9Sstevel@tonic-gate entry = htable_va2entry(va, ht); 38577c478bd9Sstevel@tonic-gate ASSERT(ht->ht_level == 0); 38587c478bd9Sstevel@tonic-gate oldpte = x86pte_get(ht, entry); 38597c478bd9Sstevel@tonic-gate } 38607c478bd9Sstevel@tonic-gate if (PTE_ISVALID(oldpte)) 38617c478bd9Sstevel@tonic-gate panic("hat_mempte_setup(): address already mapped" 3862903a11ebSrh "ht=%p, entry=%d, pte=" FMT_PTE, (void *)ht, entry, oldpte); 38637c478bd9Sstevel@tonic-gate 38647c478bd9Sstevel@tonic-gate /* 38657c478bd9Sstevel@tonic-gate * increment ht_valid_cnt so that the pagetable can't disappear 38667c478bd9Sstevel@tonic-gate */ 38677c478bd9Sstevel@tonic-gate HTABLE_INC(ht->ht_valid_cnt); 38687c478bd9Sstevel@tonic-gate 38697c478bd9Sstevel@tonic-gate /* 3870ae115bc7Smrj * return the PTE physical address to the caller. 38717c478bd9Sstevel@tonic-gate */ 38727c478bd9Sstevel@tonic-gate htable_release(ht); 3873551bc2a6Smrj XPV_ALLOW_MIGRATE(); 3874ae115bc7Smrj p = PT_INDEX_PHYSADDR(pfn_to_pa(ht->ht_pfn), entry); 3875aac11643Sjosephb --curthread->t_hatdepth; 3876ae115bc7Smrj return (p); 38777c478bd9Sstevel@tonic-gate } 38787c478bd9Sstevel@tonic-gate 38797c478bd9Sstevel@tonic-gate /* 38807c478bd9Sstevel@tonic-gate * Release a CPU private mapping for the given address. 38817c478bd9Sstevel@tonic-gate * We decrement the htable valid count so it might be destroyed. 38827c478bd9Sstevel@tonic-gate */ 3883ae115bc7Smrj /*ARGSUSED1*/ 38847c478bd9Sstevel@tonic-gate void 3885ae115bc7Smrj hat_mempte_release(caddr_t addr, hat_mempte_t pte_pa) 38867c478bd9Sstevel@tonic-gate { 38877c478bd9Sstevel@tonic-gate htable_t *ht; 38887c478bd9Sstevel@tonic-gate 3889551bc2a6Smrj XPV_DISALLOW_MIGRATE(); 38907c478bd9Sstevel@tonic-gate /* 3891ae115bc7Smrj * invalidate any left over mapping and decrement the htable valid count 38927c478bd9Sstevel@tonic-gate */ 3893843e1988Sjohnlev #ifdef __xpv 3894843e1988Sjohnlev if (HYPERVISOR_update_va_mapping((uintptr_t)addr, 0, 3895843e1988Sjohnlev UVMF_INVLPG | UVMF_LOCAL)) 3896843e1988Sjohnlev panic("HYPERVISOR_update_va_mapping() failed"); 3897843e1988Sjohnlev #else 3898ae115bc7Smrj { 3899ae115bc7Smrj x86pte_t *pteptr; 3900ae115bc7Smrj 3901ae115bc7Smrj pteptr = x86pte_mapin(mmu_btop(pte_pa), 3902ae115bc7Smrj (pte_pa & MMU_PAGEOFFSET) >> mmu.pte_size_shift, NULL); 3903ae115bc7Smrj if (mmu.pae_hat) 3904ae115bc7Smrj *pteptr = 0; 3905ae115bc7Smrj else 3906ae115bc7Smrj *(x86pte32_t *)pteptr = 0; 3907ae115bc7Smrj mmu_tlbflush_entry(addr); 3908ae115bc7Smrj x86pte_mapout(); 3909ae115bc7Smrj } 3910843e1988Sjohnlev #endif 3911ae115bc7Smrj 39127c478bd9Sstevel@tonic-gate ht = htable_getpte(kas.a_hat, ALIGN2PAGE(addr), NULL, NULL, 0); 39137c478bd9Sstevel@tonic-gate if (ht == NULL) 39147c478bd9Sstevel@tonic-gate panic("hat_mempte_release(): invalid address"); 39157c478bd9Sstevel@tonic-gate ASSERT(ht->ht_level == 0); 39167c478bd9Sstevel@tonic-gate HTABLE_DEC(ht->ht_valid_cnt); 39177c478bd9Sstevel@tonic-gate htable_release(ht); 3918551bc2a6Smrj XPV_ALLOW_MIGRATE(); 39197c478bd9Sstevel@tonic-gate } 39207c478bd9Sstevel@tonic-gate 39217c478bd9Sstevel@tonic-gate /* 39227c478bd9Sstevel@tonic-gate * Apply a temporary CPU private mapping to a page. We flush the TLB only 39237c478bd9Sstevel@tonic-gate * on this CPU, so this ought to have been called with preemption disabled. 39247c478bd9Sstevel@tonic-gate */ 39257c478bd9Sstevel@tonic-gate void 39267c478bd9Sstevel@tonic-gate hat_mempte_remap( 3927ae115bc7Smrj pfn_t pfn, 3928ae115bc7Smrj caddr_t addr, 3929ae115bc7Smrj hat_mempte_t pte_pa, 3930ae115bc7Smrj uint_t attr, 3931ae115bc7Smrj uint_t flags) 39327c478bd9Sstevel@tonic-gate { 39337c478bd9Sstevel@tonic-gate uintptr_t va = (uintptr_t)addr; 39347c478bd9Sstevel@tonic-gate x86pte_t pte; 39357c478bd9Sstevel@tonic-gate 39367c478bd9Sstevel@tonic-gate /* 39377c478bd9Sstevel@tonic-gate * Remap the given PTE to the new page's PFN. Invalidate only 39387c478bd9Sstevel@tonic-gate * on this CPU. 39397c478bd9Sstevel@tonic-gate */ 39407c478bd9Sstevel@tonic-gate #ifdef DEBUG 39417c478bd9Sstevel@tonic-gate htable_t *ht; 39427c478bd9Sstevel@tonic-gate uint_t entry; 39437c478bd9Sstevel@tonic-gate 39447c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(va)); 39457c478bd9Sstevel@tonic-gate ASSERT(!IN_VA_HOLE(va)); 39467c478bd9Sstevel@tonic-gate ht = htable_getpte(kas.a_hat, va, &entry, NULL, 0); 39477c478bd9Sstevel@tonic-gate ASSERT(ht != NULL); 39487c478bd9Sstevel@tonic-gate ASSERT(ht->ht_level == 0); 39497c478bd9Sstevel@tonic-gate ASSERT(ht->ht_valid_cnt > 0); 3950ae115bc7Smrj ASSERT(ht->ht_pfn == mmu_btop(pte_pa)); 39517c478bd9Sstevel@tonic-gate htable_release(ht); 39527c478bd9Sstevel@tonic-gate #endif 3953843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 39547c478bd9Sstevel@tonic-gate pte = hati_mkpte(pfn, attr, 0, flags); 3955843e1988Sjohnlev #ifdef __xpv 3956843e1988Sjohnlev if (HYPERVISOR_update_va_mapping(va, pte, UVMF_INVLPG | UVMF_LOCAL)) 3957843e1988Sjohnlev panic("HYPERVISOR_update_va_mapping() failed"); 3958843e1988Sjohnlev #else 3959ae115bc7Smrj { 3960ae115bc7Smrj x86pte_t *pteptr; 3961ae115bc7Smrj 3962ae115bc7Smrj pteptr = x86pte_mapin(mmu_btop(pte_pa), 3963ae115bc7Smrj (pte_pa & MMU_PAGEOFFSET) >> mmu.pte_size_shift, NULL); 3964ae115bc7Smrj if (mmu.pae_hat) 3965ae115bc7Smrj *(x86pte_t *)pteptr = pte; 3966ae115bc7Smrj else 3967ae115bc7Smrj *(x86pte32_t *)pteptr = (x86pte32_t)pte; 3968ae115bc7Smrj mmu_tlbflush_entry(addr); 3969ae115bc7Smrj x86pte_mapout(); 3970ae115bc7Smrj } 3971843e1988Sjohnlev #endif 3972843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 39737c478bd9Sstevel@tonic-gate } 39747c478bd9Sstevel@tonic-gate 39757c478bd9Sstevel@tonic-gate 39767c478bd9Sstevel@tonic-gate 39777c478bd9Sstevel@tonic-gate /* 39787c478bd9Sstevel@tonic-gate * Hat locking functions 39797c478bd9Sstevel@tonic-gate * XXX - these two functions are currently being used by hatstats 39807c478bd9Sstevel@tonic-gate * they can be removed by using a per-as mutex for hatstats. 39817c478bd9Sstevel@tonic-gate */ 39827c478bd9Sstevel@tonic-gate void 39837c478bd9Sstevel@tonic-gate hat_enter(hat_t *hat) 39847c478bd9Sstevel@tonic-gate { 39857c478bd9Sstevel@tonic-gate mutex_enter(&hat->hat_mutex); 39867c478bd9Sstevel@tonic-gate } 39877c478bd9Sstevel@tonic-gate 39887c478bd9Sstevel@tonic-gate void 39897c478bd9Sstevel@tonic-gate hat_exit(hat_t *hat) 39907c478bd9Sstevel@tonic-gate { 39917c478bd9Sstevel@tonic-gate mutex_exit(&hat->hat_mutex); 39927c478bd9Sstevel@tonic-gate } 39937c478bd9Sstevel@tonic-gate 39947c478bd9Sstevel@tonic-gate /* 3995ae115bc7Smrj * HAT part of cpu initialization. 39967c478bd9Sstevel@tonic-gate */ 39977c478bd9Sstevel@tonic-gate void 3998ae115bc7Smrj hat_cpu_online(struct cpu *cpup) 39997c478bd9Sstevel@tonic-gate { 4000ae115bc7Smrj if (cpup != CPU) { 4001ae115bc7Smrj x86pte_cpu_init(cpup); 4002ae115bc7Smrj hat_vlp_setup(cpup); 40037c478bd9Sstevel@tonic-gate } 4004ae115bc7Smrj CPUSET_ATOMIC_ADD(khat_cpuset, cpup->cpu_id); 40057c478bd9Sstevel@tonic-gate } 40067c478bd9Sstevel@tonic-gate 40077c478bd9Sstevel@tonic-gate /* 4008ae115bc7Smrj * HAT part of cpu deletion. 4009ae115bc7Smrj * (currently, we only call this after the cpu is safely passivated.) 40107c478bd9Sstevel@tonic-gate */ 40117c478bd9Sstevel@tonic-gate void 4012ae115bc7Smrj hat_cpu_offline(struct cpu *cpup) 40137c478bd9Sstevel@tonic-gate { 4014ae115bc7Smrj ASSERT(cpup != CPU); 4015ae115bc7Smrj 4016ae115bc7Smrj CPUSET_ATOMIC_DEL(khat_cpuset, cpup->cpu_id); 4017ae115bc7Smrj hat_vlp_teardown(cpup); 4018a3114836SGerry Liu x86pte_cpu_fini(cpup); 40197c478bd9Sstevel@tonic-gate } 40207c478bd9Sstevel@tonic-gate 40217c478bd9Sstevel@tonic-gate /* 40227c478bd9Sstevel@tonic-gate * Function called after all CPUs are brought online. 40237c478bd9Sstevel@tonic-gate * Used to remove low address boot mappings. 40247c478bd9Sstevel@tonic-gate */ 40257c478bd9Sstevel@tonic-gate void 40267c478bd9Sstevel@tonic-gate clear_boot_mappings(uintptr_t low, uintptr_t high) 40277c478bd9Sstevel@tonic-gate { 40287c478bd9Sstevel@tonic-gate uintptr_t vaddr = low; 40297c478bd9Sstevel@tonic-gate htable_t *ht = NULL; 40307c478bd9Sstevel@tonic-gate level_t level; 40317c478bd9Sstevel@tonic-gate uint_t entry; 40327c478bd9Sstevel@tonic-gate x86pte_t pte; 40337c478bd9Sstevel@tonic-gate 40347c478bd9Sstevel@tonic-gate /* 40357c478bd9Sstevel@tonic-gate * On 1st CPU we can unload the prom mappings, basically we blow away 4036ae115bc7Smrj * all virtual mappings under _userlimit. 40377c478bd9Sstevel@tonic-gate */ 40387c478bd9Sstevel@tonic-gate while (vaddr < high) { 40397c478bd9Sstevel@tonic-gate pte = htable_walk(kas.a_hat, &ht, &vaddr, high); 40407c478bd9Sstevel@tonic-gate if (ht == NULL) 40417c478bd9Sstevel@tonic-gate break; 40427c478bd9Sstevel@tonic-gate 40437c478bd9Sstevel@tonic-gate level = ht->ht_level; 40447c478bd9Sstevel@tonic-gate entry = htable_va2entry(vaddr, ht); 40457c478bd9Sstevel@tonic-gate ASSERT(level <= mmu.max_page_level); 40467c478bd9Sstevel@tonic-gate ASSERT(PTE_ISPAGE(pte, level)); 40477c478bd9Sstevel@tonic-gate 40487c478bd9Sstevel@tonic-gate /* 40497c478bd9Sstevel@tonic-gate * Unload the mapping from the page tables. 40507c478bd9Sstevel@tonic-gate */ 4051ae115bc7Smrj (void) x86pte_inval(ht, entry, 0, NULL); 40527c478bd9Sstevel@tonic-gate ASSERT(ht->ht_valid_cnt > 0); 40537c478bd9Sstevel@tonic-gate HTABLE_DEC(ht->ht_valid_cnt); 40547c478bd9Sstevel@tonic-gate PGCNT_DEC(ht->ht_hat, ht->ht_level); 40557c478bd9Sstevel@tonic-gate 40567c478bd9Sstevel@tonic-gate vaddr += LEVEL_SIZE(ht->ht_level); 40577c478bd9Sstevel@tonic-gate } 40587c478bd9Sstevel@tonic-gate if (ht) 40597c478bd9Sstevel@tonic-gate htable_release(ht); 40607c478bd9Sstevel@tonic-gate } 40617c478bd9Sstevel@tonic-gate 40627c478bd9Sstevel@tonic-gate /* 40637c478bd9Sstevel@tonic-gate * Atomically update a new translation for a single page. If the 40647c478bd9Sstevel@tonic-gate * currently installed PTE doesn't match the value we expect to find, 40657c478bd9Sstevel@tonic-gate * it's not updated and we return the PTE we found. 40667c478bd9Sstevel@tonic-gate * 40677c478bd9Sstevel@tonic-gate * If activating nosync or NOWRITE and the page was modified we need to sync 40687c478bd9Sstevel@tonic-gate * with the page_t. Also sync with page_t if clearing ref/mod bits. 40697c478bd9Sstevel@tonic-gate */ 40707c478bd9Sstevel@tonic-gate static x86pte_t 40717c478bd9Sstevel@tonic-gate hati_update_pte(htable_t *ht, uint_t entry, x86pte_t expected, x86pte_t new) 40727c478bd9Sstevel@tonic-gate { 40737c478bd9Sstevel@tonic-gate page_t *pp; 40747c478bd9Sstevel@tonic-gate uint_t rm = 0; 40757c478bd9Sstevel@tonic-gate x86pte_t replaced; 40767c478bd9Sstevel@tonic-gate 4077ae115bc7Smrj if (PTE_GET(expected, PT_SOFTWARE) < PT_NOSYNC && 40787c478bd9Sstevel@tonic-gate PTE_GET(expected, PT_MOD | PT_REF) && 40797c478bd9Sstevel@tonic-gate (PTE_GET(new, PT_NOSYNC) || !PTE_GET(new, PT_WRITABLE) || 4080250b7ff9Sjosephb !PTE_GET(new, PT_MOD | PT_REF))) { 40817c478bd9Sstevel@tonic-gate 4082ae115bc7Smrj ASSERT(!pfn_is_foreign(PTE2PFN(expected, ht->ht_level))); 40837c478bd9Sstevel@tonic-gate pp = page_numtopp_nolock(PTE2PFN(expected, ht->ht_level)); 40847c478bd9Sstevel@tonic-gate ASSERT(pp != NULL); 40857c478bd9Sstevel@tonic-gate if (PTE_GET(expected, PT_MOD)) 40867c478bd9Sstevel@tonic-gate rm |= P_MOD; 40877c478bd9Sstevel@tonic-gate if (PTE_GET(expected, PT_REF)) 40887c478bd9Sstevel@tonic-gate rm |= P_REF; 40897c478bd9Sstevel@tonic-gate PTE_CLR(new, PT_MOD | PT_REF); 40907c478bd9Sstevel@tonic-gate } 40917c478bd9Sstevel@tonic-gate 40927c478bd9Sstevel@tonic-gate replaced = x86pte_update(ht, entry, expected, new); 40937c478bd9Sstevel@tonic-gate if (replaced != expected) 40947c478bd9Sstevel@tonic-gate return (replaced); 40957c478bd9Sstevel@tonic-gate 40967c478bd9Sstevel@tonic-gate if (rm) { 40977c478bd9Sstevel@tonic-gate /* 40987c478bd9Sstevel@tonic-gate * sync to all constituent pages of a large page 40997c478bd9Sstevel@tonic-gate */ 41007c478bd9Sstevel@tonic-gate pgcnt_t pgcnt = page_get_pagecnt(ht->ht_level); 41017c478bd9Sstevel@tonic-gate ASSERT(IS_P2ALIGNED(pp->p_pagenum, pgcnt)); 41027c478bd9Sstevel@tonic-gate while (pgcnt-- > 0) { 41037c478bd9Sstevel@tonic-gate /* 41047c478bd9Sstevel@tonic-gate * hat_page_demote() can't decrease 41057c478bd9Sstevel@tonic-gate * pszc below this mapping size 41067c478bd9Sstevel@tonic-gate * since large mapping existed after we 41077c478bd9Sstevel@tonic-gate * took mlist lock. 41087c478bd9Sstevel@tonic-gate */ 41097c478bd9Sstevel@tonic-gate ASSERT(pp->p_szc >= ht->ht_level); 41107c478bd9Sstevel@tonic-gate hat_page_setattr(pp, rm); 41117c478bd9Sstevel@tonic-gate ++pp; 41127c478bd9Sstevel@tonic-gate } 41137c478bd9Sstevel@tonic-gate } 41147c478bd9Sstevel@tonic-gate 41157c478bd9Sstevel@tonic-gate return (0); 41167c478bd9Sstevel@tonic-gate } 41177c478bd9Sstevel@tonic-gate 411805d3dc4bSpaulsan /* ARGSUSED */ 411905d3dc4bSpaulsan void 41207dacfc44Spaulsan hat_join_srd(struct hat *hat, vnode_t *evp) 412105d3dc4bSpaulsan { 412205d3dc4bSpaulsan } 412305d3dc4bSpaulsan 412405d3dc4bSpaulsan /* ARGSUSED */ 412505d3dc4bSpaulsan hat_region_cookie_t 41267dacfc44Spaulsan hat_join_region(struct hat *hat, 412705d3dc4bSpaulsan caddr_t r_saddr, 412805d3dc4bSpaulsan size_t r_size, 412905d3dc4bSpaulsan void *r_obj, 413005d3dc4bSpaulsan u_offset_t r_objoff, 413105d3dc4bSpaulsan uchar_t r_perm, 413205d3dc4bSpaulsan uchar_t r_pgszc, 413305d3dc4bSpaulsan hat_rgn_cb_func_t r_cb_function, 413405d3dc4bSpaulsan uint_t flags) 413505d3dc4bSpaulsan { 413605d3dc4bSpaulsan panic("No shared region support on x86"); 413705d3dc4bSpaulsan return (HAT_INVALID_REGION_COOKIE); 413805d3dc4bSpaulsan } 413905d3dc4bSpaulsan 414005d3dc4bSpaulsan /* ARGSUSED */ 414105d3dc4bSpaulsan void 41427dacfc44Spaulsan hat_leave_region(struct hat *hat, hat_region_cookie_t rcookie, uint_t flags) 414305d3dc4bSpaulsan { 414405d3dc4bSpaulsan panic("No shared region support on x86"); 414505d3dc4bSpaulsan } 414605d3dc4bSpaulsan 414705d3dc4bSpaulsan /* ARGSUSED */ 414805d3dc4bSpaulsan void 41497dacfc44Spaulsan hat_dup_region(struct hat *hat, hat_region_cookie_t rcookie) 415005d3dc4bSpaulsan { 415105d3dc4bSpaulsan panic("No shared region support on x86"); 415205d3dc4bSpaulsan } 415305d3dc4bSpaulsan 415405d3dc4bSpaulsan 41557c478bd9Sstevel@tonic-gate /* 41567c478bd9Sstevel@tonic-gate * Kernel Physical Mapping (kpm) facility 41577c478bd9Sstevel@tonic-gate * 41587c478bd9Sstevel@tonic-gate * Most of the routines needed to support segkpm are almost no-ops on the 41597c478bd9Sstevel@tonic-gate * x86 platform. We map in the entire segment when it is created and leave 41607c478bd9Sstevel@tonic-gate * it mapped in, so there is no additional work required to set up and tear 41617c478bd9Sstevel@tonic-gate * down individual mappings. All of these routines were created to support 41627c478bd9Sstevel@tonic-gate * SPARC platforms that have to avoid aliasing in their virtually indexed 41637c478bd9Sstevel@tonic-gate * caches. 41647c478bd9Sstevel@tonic-gate * 41657c478bd9Sstevel@tonic-gate * Most of the routines have sanity checks in them (e.g. verifying that the 41667c478bd9Sstevel@tonic-gate * passed-in page is locked). We don't actually care about most of these 41677c478bd9Sstevel@tonic-gate * checks on x86, but we leave them in place to identify problems in the 41687c478bd9Sstevel@tonic-gate * upper levels. 41697c478bd9Sstevel@tonic-gate */ 41707c478bd9Sstevel@tonic-gate 41717c478bd9Sstevel@tonic-gate /* 41727c478bd9Sstevel@tonic-gate * Map in a locked page and return the vaddr. 41737c478bd9Sstevel@tonic-gate */ 41747c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 41757c478bd9Sstevel@tonic-gate caddr_t 41767c478bd9Sstevel@tonic-gate hat_kpm_mapin(struct page *pp, struct kpme *kpme) 41777c478bd9Sstevel@tonic-gate { 41787c478bd9Sstevel@tonic-gate caddr_t vaddr; 41797c478bd9Sstevel@tonic-gate 41807c478bd9Sstevel@tonic-gate #ifdef DEBUG 41817c478bd9Sstevel@tonic-gate if (kpm_enable == 0) { 41827c478bd9Sstevel@tonic-gate cmn_err(CE_WARN, "hat_kpm_mapin: kpm_enable not set\n"); 41837c478bd9Sstevel@tonic-gate return ((caddr_t)NULL); 41847c478bd9Sstevel@tonic-gate } 41857c478bd9Sstevel@tonic-gate 41867c478bd9Sstevel@tonic-gate if (pp == NULL || PAGE_LOCKED(pp) == 0) { 41877c478bd9Sstevel@tonic-gate cmn_err(CE_WARN, "hat_kpm_mapin: pp zero or not locked\n"); 41887c478bd9Sstevel@tonic-gate return ((caddr_t)NULL); 41897c478bd9Sstevel@tonic-gate } 41907c478bd9Sstevel@tonic-gate #endif 41917c478bd9Sstevel@tonic-gate 41927c478bd9Sstevel@tonic-gate vaddr = hat_kpm_page2va(pp, 1); 41937c478bd9Sstevel@tonic-gate 41947c478bd9Sstevel@tonic-gate return (vaddr); 41957c478bd9Sstevel@tonic-gate } 41967c478bd9Sstevel@tonic-gate 41977c478bd9Sstevel@tonic-gate /* 41987c478bd9Sstevel@tonic-gate * Mapout a locked page. 41997c478bd9Sstevel@tonic-gate */ 42007c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 42017c478bd9Sstevel@tonic-gate void 42027c478bd9Sstevel@tonic-gate hat_kpm_mapout(struct page *pp, struct kpme *kpme, caddr_t vaddr) 42037c478bd9Sstevel@tonic-gate { 42047c478bd9Sstevel@tonic-gate #ifdef DEBUG 42057c478bd9Sstevel@tonic-gate if (kpm_enable == 0) { 42067c478bd9Sstevel@tonic-gate cmn_err(CE_WARN, "hat_kpm_mapout: kpm_enable not set\n"); 42077c478bd9Sstevel@tonic-gate return; 42087c478bd9Sstevel@tonic-gate } 42097c478bd9Sstevel@tonic-gate 42107c478bd9Sstevel@tonic-gate if (IS_KPM_ADDR(vaddr) == 0) { 42117c478bd9Sstevel@tonic-gate cmn_err(CE_WARN, "hat_kpm_mapout: no kpm address\n"); 42127c478bd9Sstevel@tonic-gate return; 42137c478bd9Sstevel@tonic-gate } 42147c478bd9Sstevel@tonic-gate 42157c478bd9Sstevel@tonic-gate if (pp == NULL || PAGE_LOCKED(pp) == 0) { 42167c478bd9Sstevel@tonic-gate cmn_err(CE_WARN, "hat_kpm_mapout: page zero or not locked\n"); 42177c478bd9Sstevel@tonic-gate return; 42187c478bd9Sstevel@tonic-gate } 42197c478bd9Sstevel@tonic-gate #endif 42207c478bd9Sstevel@tonic-gate } 42217c478bd9Sstevel@tonic-gate 4222d20abfaaSPavel Tatashin /* 4223d20abfaaSPavel Tatashin * hat_kpm_mapin_pfn is used to obtain a kpm mapping for physical 4224d20abfaaSPavel Tatashin * memory addresses that are not described by a page_t. It can 4225d20abfaaSPavel Tatashin * also be used for normal pages that are not locked, but beware 4226d20abfaaSPavel Tatashin * this is dangerous - no locking is performed, so the identity of 4227d20abfaaSPavel Tatashin * the page could change. hat_kpm_mapin_pfn is not supported when 4228d20abfaaSPavel Tatashin * vac_colors > 1, because the chosen va depends on the page identity, 4229d20abfaaSPavel Tatashin * which could change. 4230d20abfaaSPavel Tatashin * The caller must only pass pfn's for valid physical addresses; violation 4231d20abfaaSPavel Tatashin * of this rule will cause panic. 4232d20abfaaSPavel Tatashin */ 4233d20abfaaSPavel Tatashin caddr_t 4234d20abfaaSPavel Tatashin hat_kpm_mapin_pfn(pfn_t pfn) 4235d20abfaaSPavel Tatashin { 4236d20abfaaSPavel Tatashin caddr_t paddr, vaddr; 4237d20abfaaSPavel Tatashin 4238d20abfaaSPavel Tatashin if (kpm_enable == 0) 4239d20abfaaSPavel Tatashin return ((caddr_t)NULL); 4240d20abfaaSPavel Tatashin 4241d20abfaaSPavel Tatashin paddr = (caddr_t)ptob(pfn); 4242d20abfaaSPavel Tatashin vaddr = (uintptr_t)kpm_vbase + paddr; 4243d20abfaaSPavel Tatashin 4244d20abfaaSPavel Tatashin return ((caddr_t)vaddr); 4245d20abfaaSPavel Tatashin } 4246d20abfaaSPavel Tatashin 4247d20abfaaSPavel Tatashin /*ARGSUSED*/ 4248d20abfaaSPavel Tatashin void 4249d20abfaaSPavel Tatashin hat_kpm_mapout_pfn(pfn_t pfn) 4250d20abfaaSPavel Tatashin { 4251d20abfaaSPavel Tatashin /* empty */ 4252d20abfaaSPavel Tatashin } 4253d20abfaaSPavel Tatashin 42547c478bd9Sstevel@tonic-gate /* 42557c478bd9Sstevel@tonic-gate * Return the kpm virtual address for a specific pfn 42567c478bd9Sstevel@tonic-gate */ 42577c478bd9Sstevel@tonic-gate caddr_t 42587c478bd9Sstevel@tonic-gate hat_kpm_pfn2va(pfn_t pfn) 42597c478bd9Sstevel@tonic-gate { 4260ae115bc7Smrj uintptr_t vaddr = (uintptr_t)kpm_vbase + mmu_ptob(pfn); 42617c478bd9Sstevel@tonic-gate 4262d2b85481Srscott ASSERT(!pfn_is_foreign(pfn)); 42637c478bd9Sstevel@tonic-gate return ((caddr_t)vaddr); 42647c478bd9Sstevel@tonic-gate } 42657c478bd9Sstevel@tonic-gate 42667c478bd9Sstevel@tonic-gate /* 42677c478bd9Sstevel@tonic-gate * Return the kpm virtual address for the page at pp. 42687c478bd9Sstevel@tonic-gate */ 42697c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 42707c478bd9Sstevel@tonic-gate caddr_t 42717c478bd9Sstevel@tonic-gate hat_kpm_page2va(struct page *pp, int checkswap) 42727c478bd9Sstevel@tonic-gate { 42737c478bd9Sstevel@tonic-gate return (hat_kpm_pfn2va(pp->p_pagenum)); 42747c478bd9Sstevel@tonic-gate } 42757c478bd9Sstevel@tonic-gate 42767c478bd9Sstevel@tonic-gate /* 42777c478bd9Sstevel@tonic-gate * Return the page frame number for the kpm virtual address vaddr. 42787c478bd9Sstevel@tonic-gate */ 42797c478bd9Sstevel@tonic-gate pfn_t 42807c478bd9Sstevel@tonic-gate hat_kpm_va2pfn(caddr_t vaddr) 42817c478bd9Sstevel@tonic-gate { 42827c478bd9Sstevel@tonic-gate pfn_t pfn; 42837c478bd9Sstevel@tonic-gate 42847c478bd9Sstevel@tonic-gate ASSERT(IS_KPM_ADDR(vaddr)); 42857c478bd9Sstevel@tonic-gate 42867c478bd9Sstevel@tonic-gate pfn = (pfn_t)btop(vaddr - kpm_vbase); 42877c478bd9Sstevel@tonic-gate 42887c478bd9Sstevel@tonic-gate return (pfn); 42897c478bd9Sstevel@tonic-gate } 42907c478bd9Sstevel@tonic-gate 42917c478bd9Sstevel@tonic-gate 42927c478bd9Sstevel@tonic-gate /* 42937c478bd9Sstevel@tonic-gate * Return the page for the kpm virtual address vaddr. 42947c478bd9Sstevel@tonic-gate */ 42957c478bd9Sstevel@tonic-gate page_t * 42967c478bd9Sstevel@tonic-gate hat_kpm_vaddr2page(caddr_t vaddr) 42977c478bd9Sstevel@tonic-gate { 42987c478bd9Sstevel@tonic-gate pfn_t pfn; 42997c478bd9Sstevel@tonic-gate 43007c478bd9Sstevel@tonic-gate ASSERT(IS_KPM_ADDR(vaddr)); 43017c478bd9Sstevel@tonic-gate 43027c478bd9Sstevel@tonic-gate pfn = hat_kpm_va2pfn(vaddr); 43037c478bd9Sstevel@tonic-gate 43047c478bd9Sstevel@tonic-gate return (page_numtopp_nolock(pfn)); 43057c478bd9Sstevel@tonic-gate } 43067c478bd9Sstevel@tonic-gate 43077c478bd9Sstevel@tonic-gate /* 43087c478bd9Sstevel@tonic-gate * hat_kpm_fault is called from segkpm_fault when we take a page fault on a 43097c478bd9Sstevel@tonic-gate * KPM page. This should never happen on x86 43107c478bd9Sstevel@tonic-gate */ 43117c478bd9Sstevel@tonic-gate int 43127c478bd9Sstevel@tonic-gate hat_kpm_fault(hat_t *hat, caddr_t vaddr) 43137c478bd9Sstevel@tonic-gate { 4314903a11ebSrh panic("pagefault in seg_kpm. hat: 0x%p vaddr: 0x%p", 4315903a11ebSrh (void *)hat, (void *)vaddr); 43167c478bd9Sstevel@tonic-gate 43177c478bd9Sstevel@tonic-gate return (0); 43187c478bd9Sstevel@tonic-gate } 43197c478bd9Sstevel@tonic-gate 43207c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 43217c478bd9Sstevel@tonic-gate void 43227c478bd9Sstevel@tonic-gate hat_kpm_mseghash_clear(int nentries) 43237c478bd9Sstevel@tonic-gate {} 43247c478bd9Sstevel@tonic-gate 43257c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 43267c478bd9Sstevel@tonic-gate void 43277c478bd9Sstevel@tonic-gate hat_kpm_mseghash_update(pgcnt_t inx, struct memseg *msp) 43287c478bd9Sstevel@tonic-gate {} 4329843e1988Sjohnlev 4330a3114836SGerry Liu #ifndef __xpv 4331a3114836SGerry Liu void 4332a3114836SGerry Liu hat_kpm_addmem_mseg_update(struct memseg *msp, pgcnt_t nkpmpgs, 4333a3114836SGerry Liu offset_t kpm_pages_off) 4334a3114836SGerry Liu { 4335a3114836SGerry Liu _NOTE(ARGUNUSED(nkpmpgs, kpm_pages_off)); 4336a3114836SGerry Liu pfn_t base, end; 4337a3114836SGerry Liu 4338a3114836SGerry Liu /* 4339a3114836SGerry Liu * kphysm_add_memory_dynamic() does not set nkpmpgs 4340a3114836SGerry Liu * when page_t memory is externally allocated. That 4341a3114836SGerry Liu * code must properly calculate nkpmpgs in all cases 4342a3114836SGerry Liu * if nkpmpgs needs to be used at some point. 4343a3114836SGerry Liu */ 4344a3114836SGerry Liu 4345a3114836SGerry Liu /* 4346a3114836SGerry Liu * The meta (page_t) pages for dynamically added memory are allocated 4347a3114836SGerry Liu * either from the incoming memory itself or from existing memory. 4348a3114836SGerry Liu * In the former case the base of the incoming pages will be different 4349a3114836SGerry Liu * than the base of the dynamic segment so call memseg_get_start() to 4350a3114836SGerry Liu * get the actual base of the incoming memory for each case. 4351a3114836SGerry Liu */ 4352a3114836SGerry Liu 4353a3114836SGerry Liu base = memseg_get_start(msp); 4354a3114836SGerry Liu end = msp->pages_end; 4355a3114836SGerry Liu 4356a3114836SGerry Liu hat_devload(kas.a_hat, kpm_vbase + mmu_ptob(base), 4357a3114836SGerry Liu mmu_ptob(end - base), base, PROT_READ | PROT_WRITE, 4358a3114836SGerry Liu HAT_LOAD | HAT_LOAD_LOCK | HAT_LOAD_NOCONSIST); 4359a3114836SGerry Liu } 4360a3114836SGerry Liu 4361a3114836SGerry Liu void 4362a3114836SGerry Liu hat_kpm_addmem_mseg_insert(struct memseg *msp) 4363a3114836SGerry Liu { 4364a3114836SGerry Liu _NOTE(ARGUNUSED(msp)); 4365a3114836SGerry Liu } 4366a3114836SGerry Liu 4367a3114836SGerry Liu void 4368a3114836SGerry Liu hat_kpm_addmem_memsegs_update(struct memseg *msp) 4369a3114836SGerry Liu { 4370a3114836SGerry Liu _NOTE(ARGUNUSED(msp)); 4371a3114836SGerry Liu } 4372a3114836SGerry Liu 4373a3114836SGerry Liu /* 4374a3114836SGerry Liu * Return end of metadata for an already setup memseg. 4375a3114836SGerry Liu * X86 platforms don't need per-page meta data to support kpm. 4376a3114836SGerry Liu */ 4377a3114836SGerry Liu caddr_t 4378a3114836SGerry Liu hat_kpm_mseg_reuse(struct memseg *msp) 4379a3114836SGerry Liu { 4380a3114836SGerry Liu return ((caddr_t)msp->epages); 4381a3114836SGerry Liu } 4382a3114836SGerry Liu 4383a3114836SGerry Liu void 4384a3114836SGerry Liu hat_kpm_delmem_mseg_update(struct memseg *msp, struct memseg **mspp) 4385a3114836SGerry Liu { 4386a3114836SGerry Liu _NOTE(ARGUNUSED(msp, mspp)); 4387a3114836SGerry Liu ASSERT(0); 4388a3114836SGerry Liu } 4389a3114836SGerry Liu 4390a3114836SGerry Liu void 4391a3114836SGerry Liu hat_kpm_split_mseg_update(struct memseg *msp, struct memseg **mspp, 4392a3114836SGerry Liu struct memseg *lo, struct memseg *mid, struct memseg *hi) 4393a3114836SGerry Liu { 4394a3114836SGerry Liu _NOTE(ARGUNUSED(msp, mspp, lo, mid, hi)); 4395a3114836SGerry Liu ASSERT(0); 4396a3114836SGerry Liu } 4397a3114836SGerry Liu 4398a3114836SGerry Liu /* 4399a3114836SGerry Liu * Walk the memsegs chain, applying func to each memseg span. 4400a3114836SGerry Liu */ 4401a3114836SGerry Liu void 4402a3114836SGerry Liu hat_kpm_walk(void (*func)(void *, void *, size_t), void *arg) 4403a3114836SGerry Liu { 4404a3114836SGerry Liu pfn_t pbase, pend; 4405a3114836SGerry Liu void *base; 4406a3114836SGerry Liu size_t size; 4407a3114836SGerry Liu struct memseg *msp; 4408a3114836SGerry Liu 4409a3114836SGerry Liu for (msp = memsegs; msp; msp = msp->next) { 4410a3114836SGerry Liu pbase = msp->pages_base; 4411a3114836SGerry Liu pend = msp->pages_end; 4412a3114836SGerry Liu base = ptob(pbase) + kpm_vbase; 4413a3114836SGerry Liu size = ptob(pend - pbase); 4414a3114836SGerry Liu func(arg, base, size); 4415a3114836SGerry Liu } 4416a3114836SGerry Liu } 4417a3114836SGerry Liu 4418a3114836SGerry Liu #else /* __xpv */ 4419a3114836SGerry Liu 4420843e1988Sjohnlev /* 4421843e1988Sjohnlev * There are specific Hypervisor calls to establish and remove mappings 4422843e1988Sjohnlev * to grant table references and the privcmd driver. We have to ensure 4423843e1988Sjohnlev * that a page table actually exists. 4424843e1988Sjohnlev */ 4425843e1988Sjohnlev void 44267eea693dSMark Johnson hat_prepare_mapping(hat_t *hat, caddr_t addr, uint64_t *pte_ma) 4427843e1988Sjohnlev { 44287eea693dSMark Johnson maddr_t base_ma; 44297eea693dSMark Johnson htable_t *ht; 44307eea693dSMark Johnson uint_t entry; 44317eea693dSMark Johnson 4432843e1988Sjohnlev ASSERT(IS_P2ALIGNED((uintptr_t)addr, MMU_PAGESIZE)); 4433551bc2a6Smrj XPV_DISALLOW_MIGRATE(); 44347eea693dSMark Johnson ht = htable_create(hat, (uintptr_t)addr, 0, NULL); 44357eea693dSMark Johnson 44367eea693dSMark Johnson /* 44377eea693dSMark Johnson * if an address for pte_ma is passed in, return the MA of the pte 44387eea693dSMark Johnson * for this specific address. This address is only valid as long 44397eea693dSMark Johnson * as the htable stays locked. 44407eea693dSMark Johnson */ 44417eea693dSMark Johnson if (pte_ma != NULL) { 44427eea693dSMark Johnson entry = htable_va2entry((uintptr_t)addr, ht); 44437eea693dSMark Johnson base_ma = pa_to_ma(ptob(ht->ht_pfn)); 44447eea693dSMark Johnson *pte_ma = base_ma + (entry << mmu.pte_size_shift); 44457eea693dSMark Johnson } 4446551bc2a6Smrj XPV_ALLOW_MIGRATE(); 4447843e1988Sjohnlev } 4448843e1988Sjohnlev 4449843e1988Sjohnlev void 4450843e1988Sjohnlev hat_release_mapping(hat_t *hat, caddr_t addr) 4451843e1988Sjohnlev { 4452843e1988Sjohnlev htable_t *ht; 4453843e1988Sjohnlev 4454843e1988Sjohnlev ASSERT(IS_P2ALIGNED((uintptr_t)addr, MMU_PAGESIZE)); 4455551bc2a6Smrj XPV_DISALLOW_MIGRATE(); 4456843e1988Sjohnlev ht = htable_lookup(hat, (uintptr_t)addr, 0); 4457843e1988Sjohnlev ASSERT(ht != NULL); 4458843e1988Sjohnlev ASSERT(ht->ht_busy >= 2); 4459843e1988Sjohnlev htable_release(ht); 4460843e1988Sjohnlev htable_release(ht); 4461551bc2a6Smrj XPV_ALLOW_MIGRATE(); 4462a3114836SGerry Liu } 4463a3114836SGerry Liu #endif /* __xpv */ 4464