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 /* 2225540de2SJakub Jermar * Copyright 2009 Sun Microsystems, Inc. All rights reserved. 237c478bd9Sstevel@tonic-gate * Use is subject to license terms. 247c478bd9Sstevel@tonic-gate */ 257c478bd9Sstevel@tonic-gate 267c478bd9Sstevel@tonic-gate 277c478bd9Sstevel@tonic-gate /* 287c478bd9Sstevel@tonic-gate * VM - Hardware Address Translation management for i386 and amd64 297c478bd9Sstevel@tonic-gate * 307c478bd9Sstevel@tonic-gate * Implementation of the interfaces described in <common/vm/hat.h> 317c478bd9Sstevel@tonic-gate * 327c478bd9Sstevel@tonic-gate * Nearly all the details of how the hardware is managed should not be 337c478bd9Sstevel@tonic-gate * visible outside this layer except for misc. machine specific functions 347c478bd9Sstevel@tonic-gate * that work in conjunction with this code. 357c478bd9Sstevel@tonic-gate * 367c478bd9Sstevel@tonic-gate * Routines used only inside of i86pc/vm start with hati_ for HAT Internal. 377c478bd9Sstevel@tonic-gate */ 387c478bd9Sstevel@tonic-gate 397c478bd9Sstevel@tonic-gate #include <sys/machparam.h> 407c478bd9Sstevel@tonic-gate #include <sys/machsystm.h> 417c478bd9Sstevel@tonic-gate #include <sys/mman.h> 427c478bd9Sstevel@tonic-gate #include <sys/types.h> 437c478bd9Sstevel@tonic-gate #include <sys/systm.h> 447c478bd9Sstevel@tonic-gate #include <sys/cpuvar.h> 457c478bd9Sstevel@tonic-gate #include <sys/thread.h> 467c478bd9Sstevel@tonic-gate #include <sys/proc.h> 477c478bd9Sstevel@tonic-gate #include <sys/cpu.h> 487c478bd9Sstevel@tonic-gate #include <sys/kmem.h> 497c478bd9Sstevel@tonic-gate #include <sys/disp.h> 507c478bd9Sstevel@tonic-gate #include <sys/shm.h> 517c478bd9Sstevel@tonic-gate #include <sys/sysmacros.h> 527c478bd9Sstevel@tonic-gate #include <sys/machparam.h> 537c478bd9Sstevel@tonic-gate #include <sys/vmem.h> 547c478bd9Sstevel@tonic-gate #include <sys/vmsystm.h> 557c478bd9Sstevel@tonic-gate #include <sys/promif.h> 567c478bd9Sstevel@tonic-gate #include <sys/var.h> 577c478bd9Sstevel@tonic-gate #include <sys/x86_archext.h> 587c478bd9Sstevel@tonic-gate #include <sys/atomic.h> 597c478bd9Sstevel@tonic-gate #include <sys/bitmap.h> 60ae115bc7Smrj #include <sys/controlregs.h> 61ae115bc7Smrj #include <sys/bootconf.h> 62ae115bc7Smrj #include <sys/bootsvcs.h> 63ae115bc7Smrj #include <sys/bootinfo.h> 6495c0a3c8Sjosephb #include <sys/archsystm.h> 657c478bd9Sstevel@tonic-gate 667c478bd9Sstevel@tonic-gate #include <vm/seg_kmem.h> 677c478bd9Sstevel@tonic-gate #include <vm/hat_i86.h> 687c478bd9Sstevel@tonic-gate #include <vm/as.h> 697c478bd9Sstevel@tonic-gate #include <vm/seg.h> 707c478bd9Sstevel@tonic-gate #include <vm/page.h> 717c478bd9Sstevel@tonic-gate #include <vm/seg_kp.h> 727c478bd9Sstevel@tonic-gate #include <vm/seg_kpm.h> 737c478bd9Sstevel@tonic-gate #include <vm/vm_dep.h> 74843e1988Sjohnlev #ifdef __xpv 75843e1988Sjohnlev #include <sys/hypervisor.h> 76843e1988Sjohnlev #endif 77ae115bc7Smrj #include <vm/kboot_mmu.h> 78250b7ff9Sjosephb #include <vm/seg_spt.h> 797c478bd9Sstevel@tonic-gate 807c478bd9Sstevel@tonic-gate #include <sys/cmn_err.h> 817c478bd9Sstevel@tonic-gate 827c478bd9Sstevel@tonic-gate /* 837c478bd9Sstevel@tonic-gate * Basic parameters for hat operation. 847c478bd9Sstevel@tonic-gate */ 857c478bd9Sstevel@tonic-gate struct hat_mmu_info mmu; 867c478bd9Sstevel@tonic-gate 877c478bd9Sstevel@tonic-gate /* 887c478bd9Sstevel@tonic-gate * The page that is the kernel's top level pagetable. 897c478bd9Sstevel@tonic-gate * 90843e1988Sjohnlev * For 32 bit PAE support on i86pc, the kernel hat will use the 1st 4 entries 917c478bd9Sstevel@tonic-gate * on this 4K page for its top level page table. The remaining groups of 927c478bd9Sstevel@tonic-gate * 4 entries are used for per processor copies of user VLP pagetables for 937c478bd9Sstevel@tonic-gate * running threads. See hat_switch() and reload_pae32() for details. 947c478bd9Sstevel@tonic-gate * 95843e1988Sjohnlev * vlp_page[0..3] - level==2 PTEs for kernel HAT 96843e1988Sjohnlev * vlp_page[4..7] - level==2 PTEs for user thread on cpu 0 97843e1988Sjohnlev * vlp_page[8..11] - level==2 PTE for user thread on cpu 1 98843e1988Sjohnlev * etc... 997c478bd9Sstevel@tonic-gate */ 1007c478bd9Sstevel@tonic-gate static x86pte_t *vlp_page; 1017c478bd9Sstevel@tonic-gate 1027c478bd9Sstevel@tonic-gate /* 1037c478bd9Sstevel@tonic-gate * forward declaration of internal utility routines 1047c478bd9Sstevel@tonic-gate */ 1057c478bd9Sstevel@tonic-gate static x86pte_t hati_update_pte(htable_t *ht, uint_t entry, x86pte_t expected, 1067c478bd9Sstevel@tonic-gate x86pte_t new); 1077c478bd9Sstevel@tonic-gate 1087c478bd9Sstevel@tonic-gate /* 1097c478bd9Sstevel@tonic-gate * The kernel address space exists in all HATs. To implement this the 110843e1988Sjohnlev * kernel reserves a fixed number of entries in the topmost level(s) of page 111843e1988Sjohnlev * tables. The values are setup during startup and then copied to every user 112843e1988Sjohnlev * hat created by hat_alloc(). This means that kernelbase must be: 1137c478bd9Sstevel@tonic-gate * 1147c478bd9Sstevel@tonic-gate * 4Meg aligned for 32 bit kernels 1157c478bd9Sstevel@tonic-gate * 512Gig aligned for x86_64 64 bit kernel 1167c478bd9Sstevel@tonic-gate * 117843e1988Sjohnlev * The hat_kernel_range_ts describe what needs to be copied from kernel hat 118843e1988Sjohnlev * to each user hat. 1197c478bd9Sstevel@tonic-gate */ 120843e1988Sjohnlev typedef struct hat_kernel_range { 121843e1988Sjohnlev level_t hkr_level; 122843e1988Sjohnlev uintptr_t hkr_start_va; 123843e1988Sjohnlev uintptr_t hkr_end_va; /* zero means to end of memory */ 124843e1988Sjohnlev } hat_kernel_range_t; 125843e1988Sjohnlev #define NUM_KERNEL_RANGE 2 126843e1988Sjohnlev static hat_kernel_range_t kernel_ranges[NUM_KERNEL_RANGE]; 127843e1988Sjohnlev static int num_kernel_ranges; 1287c478bd9Sstevel@tonic-gate 1297c478bd9Sstevel@tonic-gate uint_t use_boot_reserve = 1; /* cleared after early boot process */ 1307c478bd9Sstevel@tonic-gate uint_t can_steal_post_boot = 0; /* set late in boot to enable stealing */ 1317c478bd9Sstevel@tonic-gate 132512cf780Skchow /* 133512cf780Skchow * enable_1gpg: controls 1g page support for user applications. 134512cf780Skchow * By default, 1g pages are exported to user applications. enable_1gpg can 135512cf780Skchow * be set to 0 to not export. 136512cf780Skchow */ 13778b03d3aSkchow int enable_1gpg = 1; 13802bc52beSkchow 139512cf780Skchow /* 140512cf780Skchow * AMD shanghai processors provide better management of 1gb ptes in its tlb. 141*21584dbcSPavel Tatashin * By default, 1g page support will be disabled for pre-shanghai AMD 142512cf780Skchow * processors that don't have optimal tlb support for the 1g page size. 143512cf780Skchow * chk_optimal_1gtlb can be set to 0 to force 1g page support on sub-optimal 144512cf780Skchow * processors. 145512cf780Skchow */ 146512cf780Skchow int chk_optimal_1gtlb = 1; 147512cf780Skchow 148512cf780Skchow 14902bc52beSkchow #ifdef DEBUG 15002bc52beSkchow uint_t map1gcnt; 15102bc52beSkchow #endif 15202bc52beSkchow 15302bc52beSkchow 1547c478bd9Sstevel@tonic-gate /* 1557c478bd9Sstevel@tonic-gate * A cpuset for all cpus. This is used for kernel address cross calls, since 1567c478bd9Sstevel@tonic-gate * the kernel addresses apply to all cpus. 1577c478bd9Sstevel@tonic-gate */ 1587c478bd9Sstevel@tonic-gate cpuset_t khat_cpuset; 1597c478bd9Sstevel@tonic-gate 1607c478bd9Sstevel@tonic-gate /* 1617c478bd9Sstevel@tonic-gate * management stuff for hat structures 1627c478bd9Sstevel@tonic-gate */ 1637c478bd9Sstevel@tonic-gate kmutex_t hat_list_lock; 1647c478bd9Sstevel@tonic-gate kcondvar_t hat_list_cv; 1657c478bd9Sstevel@tonic-gate kmem_cache_t *hat_cache; 1667c478bd9Sstevel@tonic-gate kmem_cache_t *hat_hash_cache; 1677c478bd9Sstevel@tonic-gate kmem_cache_t *vlp_hash_cache; 1687c478bd9Sstevel@tonic-gate 1697c478bd9Sstevel@tonic-gate /* 1707c478bd9Sstevel@tonic-gate * Simple statistics 1717c478bd9Sstevel@tonic-gate */ 1727c478bd9Sstevel@tonic-gate struct hatstats hatstat; 1737c478bd9Sstevel@tonic-gate 174ab4a9bebSjohnlev /* 175ab4a9bebSjohnlev * Some earlier hypervisor versions do not emulate cmpxchg of PTEs 176ab4a9bebSjohnlev * correctly. For such hypervisors we must set PT_USER for kernel 177ab4a9bebSjohnlev * entries ourselves (normally the emulation would set PT_USER for 178ab4a9bebSjohnlev * kernel entries and PT_USER|PT_GLOBAL for user entries). pt_kern is 179ab4a9bebSjohnlev * thus set appropriately. Note that dboot/kbm is OK, as only the full 180ab4a9bebSjohnlev * HAT uses cmpxchg() and the other paths (hypercall etc.) were never 181ab4a9bebSjohnlev * incorrect. 182ab4a9bebSjohnlev */ 183ab4a9bebSjohnlev int pt_kern; 184ab4a9bebSjohnlev 1857c478bd9Sstevel@tonic-gate /* 1867c478bd9Sstevel@tonic-gate * useful stuff for atomic access/clearing/setting REF/MOD/RO bits in page_t's. 1877c478bd9Sstevel@tonic-gate */ 1887c478bd9Sstevel@tonic-gate extern void atomic_orb(uchar_t *addr, uchar_t val); 1897c478bd9Sstevel@tonic-gate extern void atomic_andb(uchar_t *addr, uchar_t val); 1907c478bd9Sstevel@tonic-gate 1917c478bd9Sstevel@tonic-gate #define PP_GETRM(pp, rmmask) (pp->p_nrm & rmmask) 1927c478bd9Sstevel@tonic-gate #define PP_ISMOD(pp) PP_GETRM(pp, P_MOD) 1937c478bd9Sstevel@tonic-gate #define PP_ISREF(pp) PP_GETRM(pp, P_REF) 1947c478bd9Sstevel@tonic-gate #define PP_ISRO(pp) PP_GETRM(pp, P_RO) 1957c478bd9Sstevel@tonic-gate 1967c478bd9Sstevel@tonic-gate #define PP_SETRM(pp, rm) atomic_orb(&(pp->p_nrm), rm) 1977c478bd9Sstevel@tonic-gate #define PP_SETMOD(pp) PP_SETRM(pp, P_MOD) 1987c478bd9Sstevel@tonic-gate #define PP_SETREF(pp) PP_SETRM(pp, P_REF) 1997c478bd9Sstevel@tonic-gate #define PP_SETRO(pp) PP_SETRM(pp, P_RO) 2007c478bd9Sstevel@tonic-gate 2017c478bd9Sstevel@tonic-gate #define PP_CLRRM(pp, rm) atomic_andb(&(pp->p_nrm), ~(rm)) 2027c478bd9Sstevel@tonic-gate #define PP_CLRMOD(pp) PP_CLRRM(pp, P_MOD) 2037c478bd9Sstevel@tonic-gate #define PP_CLRREF(pp) PP_CLRRM(pp, P_REF) 2047c478bd9Sstevel@tonic-gate #define PP_CLRRO(pp) PP_CLRRM(pp, P_RO) 2057c478bd9Sstevel@tonic-gate #define PP_CLRALL(pp) PP_CLRRM(pp, P_MOD | P_REF | P_RO) 2067c478bd9Sstevel@tonic-gate 2077c478bd9Sstevel@tonic-gate /* 2087c478bd9Sstevel@tonic-gate * kmem cache constructor for struct hat 2097c478bd9Sstevel@tonic-gate */ 2107c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 2117c478bd9Sstevel@tonic-gate static int 2127c478bd9Sstevel@tonic-gate hati_constructor(void *buf, void *handle, int kmflags) 2137c478bd9Sstevel@tonic-gate { 2147c478bd9Sstevel@tonic-gate hat_t *hat = buf; 2157c478bd9Sstevel@tonic-gate 2167c478bd9Sstevel@tonic-gate mutex_init(&hat->hat_mutex, NULL, MUTEX_DEFAULT, NULL); 2177c478bd9Sstevel@tonic-gate bzero(hat->hat_pages_mapped, 2187c478bd9Sstevel@tonic-gate sizeof (pgcnt_t) * (mmu.max_page_level + 1)); 219250b7ff9Sjosephb hat->hat_ism_pgcnt = 0; 2207c478bd9Sstevel@tonic-gate hat->hat_stats = 0; 2217c478bd9Sstevel@tonic-gate hat->hat_flags = 0; 2227c478bd9Sstevel@tonic-gate CPUSET_ZERO(hat->hat_cpus); 2237c478bd9Sstevel@tonic-gate hat->hat_htable = NULL; 2247c478bd9Sstevel@tonic-gate hat->hat_ht_hash = NULL; 2257c478bd9Sstevel@tonic-gate return (0); 2267c478bd9Sstevel@tonic-gate } 2277c478bd9Sstevel@tonic-gate 2287c478bd9Sstevel@tonic-gate /* 2297c478bd9Sstevel@tonic-gate * Allocate a hat structure for as. We also create the top level 2307c478bd9Sstevel@tonic-gate * htable and initialize it to contain the kernel hat entries. 2317c478bd9Sstevel@tonic-gate */ 2327c478bd9Sstevel@tonic-gate hat_t * 2337c478bd9Sstevel@tonic-gate hat_alloc(struct as *as) 2347c478bd9Sstevel@tonic-gate { 235843e1988Sjohnlev hat_t *hat; 236843e1988Sjohnlev htable_t *ht; /* top level htable */ 237843e1988Sjohnlev uint_t use_vlp; 238843e1988Sjohnlev uint_t r; 239843e1988Sjohnlev hat_kernel_range_t *rp; 240843e1988Sjohnlev uintptr_t va; 241843e1988Sjohnlev uintptr_t eva; 242843e1988Sjohnlev uint_t start; 243843e1988Sjohnlev uint_t cnt; 244843e1988Sjohnlev htable_t *src; 2457c478bd9Sstevel@tonic-gate 2467c478bd9Sstevel@tonic-gate /* 2477c478bd9Sstevel@tonic-gate * Once we start creating user process HATs we can enable 2487c478bd9Sstevel@tonic-gate * the htable_steal() code. 2497c478bd9Sstevel@tonic-gate */ 2507c478bd9Sstevel@tonic-gate if (can_steal_post_boot == 0) 2517c478bd9Sstevel@tonic-gate can_steal_post_boot = 1; 2527c478bd9Sstevel@tonic-gate 2537c478bd9Sstevel@tonic-gate ASSERT(AS_WRITE_HELD(as, &as->a_lock)); 2547c478bd9Sstevel@tonic-gate hat = kmem_cache_alloc(hat_cache, KM_SLEEP); 2557c478bd9Sstevel@tonic-gate hat->hat_as = as; 2567c478bd9Sstevel@tonic-gate mutex_init(&hat->hat_mutex, NULL, MUTEX_DEFAULT, NULL); 2577c478bd9Sstevel@tonic-gate ASSERT(hat->hat_flags == 0); 2587c478bd9Sstevel@tonic-gate 259843e1988Sjohnlev #if defined(__xpv) 2607c478bd9Sstevel@tonic-gate /* 261843e1988Sjohnlev * No VLP stuff on the hypervisor due to the 64-bit split top level 262843e1988Sjohnlev * page tables. On 32-bit it's not needed as the hypervisor takes 263843e1988Sjohnlev * care of copying the top level PTEs to a below 4Gig page. 2647c478bd9Sstevel@tonic-gate */ 265843e1988Sjohnlev use_vlp = 0; 266843e1988Sjohnlev #else /* __xpv */ 267843e1988Sjohnlev /* 32 bit processes uses a VLP style hat when running with PAE */ 2687c478bd9Sstevel@tonic-gate #if defined(__amd64) 2697c478bd9Sstevel@tonic-gate use_vlp = (ttoproc(curthread)->p_model == DATAMODEL_ILP32); 2707c478bd9Sstevel@tonic-gate #elif defined(__i386) 2717c478bd9Sstevel@tonic-gate use_vlp = mmu.pae_hat; 2727c478bd9Sstevel@tonic-gate #endif 273843e1988Sjohnlev #endif /* __xpv */ 2747c478bd9Sstevel@tonic-gate if (use_vlp) { 2757c478bd9Sstevel@tonic-gate hat->hat_flags = HAT_VLP; 2767c478bd9Sstevel@tonic-gate bzero(hat->hat_vlp_ptes, VLP_SIZE); 2777c478bd9Sstevel@tonic-gate } 2787c478bd9Sstevel@tonic-gate 2797c478bd9Sstevel@tonic-gate /* 2807c478bd9Sstevel@tonic-gate * Allocate the htable hash 2817c478bd9Sstevel@tonic-gate */ 2827c478bd9Sstevel@tonic-gate if ((hat->hat_flags & HAT_VLP)) { 2837c478bd9Sstevel@tonic-gate hat->hat_num_hash = mmu.vlp_hash_cnt; 2847c478bd9Sstevel@tonic-gate hat->hat_ht_hash = kmem_cache_alloc(vlp_hash_cache, KM_SLEEP); 2857c478bd9Sstevel@tonic-gate } else { 2867c478bd9Sstevel@tonic-gate hat->hat_num_hash = mmu.hash_cnt; 2877c478bd9Sstevel@tonic-gate hat->hat_ht_hash = kmem_cache_alloc(hat_hash_cache, KM_SLEEP); 2887c478bd9Sstevel@tonic-gate } 2897c478bd9Sstevel@tonic-gate bzero(hat->hat_ht_hash, hat->hat_num_hash * sizeof (htable_t *)); 2907c478bd9Sstevel@tonic-gate 2917c478bd9Sstevel@tonic-gate /* 2927c478bd9Sstevel@tonic-gate * Initialize Kernel HAT entries at the top of the top level page 293843e1988Sjohnlev * tables for the new hat. 2947c478bd9Sstevel@tonic-gate */ 2957c478bd9Sstevel@tonic-gate hat->hat_htable = NULL; 2967c478bd9Sstevel@tonic-gate hat->hat_ht_cached = NULL; 297843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 2987c478bd9Sstevel@tonic-gate ht = htable_create(hat, (uintptr_t)0, TOP_LEVEL(hat), NULL); 299843e1988Sjohnlev hat->hat_htable = ht; 300ae115bc7Smrj 301843e1988Sjohnlev #if defined(__amd64) 302843e1988Sjohnlev if (hat->hat_flags & HAT_VLP) 303843e1988Sjohnlev goto init_done; 3047c478bd9Sstevel@tonic-gate #endif 3057c478bd9Sstevel@tonic-gate 306843e1988Sjohnlev for (r = 0; r < num_kernel_ranges; ++r) { 307843e1988Sjohnlev rp = &kernel_ranges[r]; 308843e1988Sjohnlev for (va = rp->hkr_start_va; va != rp->hkr_end_va; 309843e1988Sjohnlev va += cnt * LEVEL_SIZE(rp->hkr_level)) { 310843e1988Sjohnlev 311843e1988Sjohnlev if (rp->hkr_level == TOP_LEVEL(hat)) 312843e1988Sjohnlev ht = hat->hat_htable; 313843e1988Sjohnlev else 314843e1988Sjohnlev ht = htable_create(hat, va, rp->hkr_level, 315843e1988Sjohnlev NULL); 316843e1988Sjohnlev 317843e1988Sjohnlev start = htable_va2entry(va, ht); 318843e1988Sjohnlev cnt = HTABLE_NUM_PTES(ht) - start; 319843e1988Sjohnlev eva = va + 320843e1988Sjohnlev ((uintptr_t)cnt << LEVEL_SHIFT(rp->hkr_level)); 321843e1988Sjohnlev if (rp->hkr_end_va != 0 && 322843e1988Sjohnlev (eva > rp->hkr_end_va || eva == 0)) 323843e1988Sjohnlev cnt = htable_va2entry(rp->hkr_end_va, ht) - 324843e1988Sjohnlev start; 325843e1988Sjohnlev 326843e1988Sjohnlev #if defined(__i386) && !defined(__xpv) 327843e1988Sjohnlev if (ht->ht_flags & HTABLE_VLP) { 328843e1988Sjohnlev bcopy(&vlp_page[start], 329843e1988Sjohnlev &hat->hat_vlp_ptes[start], 330843e1988Sjohnlev cnt * sizeof (x86pte_t)); 331843e1988Sjohnlev continue; 332843e1988Sjohnlev } 333843e1988Sjohnlev #endif 334843e1988Sjohnlev src = htable_lookup(kas.a_hat, va, rp->hkr_level); 335843e1988Sjohnlev ASSERT(src != NULL); 336843e1988Sjohnlev x86pte_copy(src, ht, start, cnt); 337843e1988Sjohnlev htable_release(src); 338843e1988Sjohnlev } 339843e1988Sjohnlev } 340843e1988Sjohnlev 341843e1988Sjohnlev init_done: 342843e1988Sjohnlev 343843e1988Sjohnlev #if defined(__xpv) 3447c478bd9Sstevel@tonic-gate /* 345843e1988Sjohnlev * Pin top level page tables after initializing them 3467c478bd9Sstevel@tonic-gate */ 347843e1988Sjohnlev xen_pin(hat->hat_htable->ht_pfn, mmu.max_level); 348843e1988Sjohnlev #if defined(__amd64) 349843e1988Sjohnlev xen_pin(hat->hat_user_ptable, mmu.max_level); 350843e1988Sjohnlev #endif 3517c478bd9Sstevel@tonic-gate #endif 352551bc2a6Smrj XPV_ALLOW_MIGRATE(); 3537c478bd9Sstevel@tonic-gate 3547c478bd9Sstevel@tonic-gate /* 355a85a6733Sjosephb * Put it at the start of the global list of all hats (used by stealing) 356a85a6733Sjosephb * 357a85a6733Sjosephb * kas.a_hat is not in the list but is instead used to find the 358a85a6733Sjosephb * first and last items in the list. 359a85a6733Sjosephb * 360a85a6733Sjosephb * - kas.a_hat->hat_next points to the start of the user hats. 361a85a6733Sjosephb * The list ends where hat->hat_next == NULL 362a85a6733Sjosephb * 363a85a6733Sjosephb * - kas.a_hat->hat_prev points to the last of the user hats. 364a85a6733Sjosephb * The list begins where hat->hat_prev == NULL 3657c478bd9Sstevel@tonic-gate */ 3667c478bd9Sstevel@tonic-gate mutex_enter(&hat_list_lock); 367a85a6733Sjosephb hat->hat_prev = NULL; 368a85a6733Sjosephb hat->hat_next = kas.a_hat->hat_next; 369a85a6733Sjosephb if (hat->hat_next) 370a85a6733Sjosephb hat->hat_next->hat_prev = hat; 371a85a6733Sjosephb else 372a85a6733Sjosephb kas.a_hat->hat_prev = hat; 3737c478bd9Sstevel@tonic-gate kas.a_hat->hat_next = hat; 3747c478bd9Sstevel@tonic-gate mutex_exit(&hat_list_lock); 3757c478bd9Sstevel@tonic-gate 3767c478bd9Sstevel@tonic-gate return (hat); 3777c478bd9Sstevel@tonic-gate } 3787c478bd9Sstevel@tonic-gate 3797c478bd9Sstevel@tonic-gate /* 3807c478bd9Sstevel@tonic-gate * process has finished executing but as has not been cleaned up yet. 3817c478bd9Sstevel@tonic-gate */ 3827c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 3837c478bd9Sstevel@tonic-gate void 3847c478bd9Sstevel@tonic-gate hat_free_start(hat_t *hat) 3857c478bd9Sstevel@tonic-gate { 3867c478bd9Sstevel@tonic-gate ASSERT(AS_WRITE_HELD(hat->hat_as, &hat->hat_as->a_lock)); 387a85a6733Sjosephb 388a85a6733Sjosephb /* 389a85a6733Sjosephb * If the hat is currently a stealing victim, wait for the stealing 390a85a6733Sjosephb * to finish. Once we mark it as HAT_FREEING, htable_steal() 391a85a6733Sjosephb * won't look at its pagetables anymore. 392a85a6733Sjosephb */ 3937c478bd9Sstevel@tonic-gate mutex_enter(&hat_list_lock); 394a85a6733Sjosephb while (hat->hat_flags & HAT_VICTIM) 395a85a6733Sjosephb cv_wait(&hat_list_cv, &hat_list_lock); 3967c478bd9Sstevel@tonic-gate hat->hat_flags |= HAT_FREEING; 3977c478bd9Sstevel@tonic-gate mutex_exit(&hat_list_lock); 3987c478bd9Sstevel@tonic-gate } 3997c478bd9Sstevel@tonic-gate 4007c478bd9Sstevel@tonic-gate /* 4017c478bd9Sstevel@tonic-gate * An address space is being destroyed, so we destroy the associated hat. 4027c478bd9Sstevel@tonic-gate */ 4037c478bd9Sstevel@tonic-gate void 4047c478bd9Sstevel@tonic-gate hat_free_end(hat_t *hat) 4057c478bd9Sstevel@tonic-gate { 4067c478bd9Sstevel@tonic-gate kmem_cache_t *cache; 4077c478bd9Sstevel@tonic-gate 4087c478bd9Sstevel@tonic-gate ASSERT(hat->hat_flags & HAT_FREEING); 4097c478bd9Sstevel@tonic-gate 4107c478bd9Sstevel@tonic-gate /* 4117c478bd9Sstevel@tonic-gate * must not be running on the given hat 4127c478bd9Sstevel@tonic-gate */ 4137c478bd9Sstevel@tonic-gate ASSERT(CPU->cpu_current_hat != hat); 4147c478bd9Sstevel@tonic-gate 4157c478bd9Sstevel@tonic-gate /* 416a85a6733Sjosephb * Remove it from the list of HATs 4177c478bd9Sstevel@tonic-gate */ 4187c478bd9Sstevel@tonic-gate mutex_enter(&hat_list_lock); 419a85a6733Sjosephb if (hat->hat_prev) 420a85a6733Sjosephb hat->hat_prev->hat_next = hat->hat_next; 421a85a6733Sjosephb else 4227c478bd9Sstevel@tonic-gate kas.a_hat->hat_next = hat->hat_next; 423a85a6733Sjosephb if (hat->hat_next) 424a85a6733Sjosephb hat->hat_next->hat_prev = hat->hat_prev; 425a85a6733Sjosephb else 426a85a6733Sjosephb kas.a_hat->hat_prev = hat->hat_prev; 4277c478bd9Sstevel@tonic-gate mutex_exit(&hat_list_lock); 428a85a6733Sjosephb hat->hat_next = hat->hat_prev = NULL; 4297c478bd9Sstevel@tonic-gate 430843e1988Sjohnlev #if defined(__xpv) 431843e1988Sjohnlev /* 432843e1988Sjohnlev * On the hypervisor, unpin top level page table(s) 433843e1988Sjohnlev */ 434843e1988Sjohnlev xen_unpin(hat->hat_htable->ht_pfn); 435843e1988Sjohnlev #if defined(__amd64) 436843e1988Sjohnlev xen_unpin(hat->hat_user_ptable); 437843e1988Sjohnlev #endif 438843e1988Sjohnlev #endif 439843e1988Sjohnlev 4407c478bd9Sstevel@tonic-gate /* 4417c478bd9Sstevel@tonic-gate * Make a pass through the htables freeing them all up. 4427c478bd9Sstevel@tonic-gate */ 4437c478bd9Sstevel@tonic-gate htable_purge_hat(hat); 4447c478bd9Sstevel@tonic-gate 4457c478bd9Sstevel@tonic-gate /* 4467c478bd9Sstevel@tonic-gate * Decide which kmem cache the hash table came from, then free it. 4477c478bd9Sstevel@tonic-gate */ 4487c478bd9Sstevel@tonic-gate if (hat->hat_flags & HAT_VLP) 4497c478bd9Sstevel@tonic-gate cache = vlp_hash_cache; 4507c478bd9Sstevel@tonic-gate else 4517c478bd9Sstevel@tonic-gate cache = hat_hash_cache; 4527c478bd9Sstevel@tonic-gate kmem_cache_free(cache, hat->hat_ht_hash); 4537c478bd9Sstevel@tonic-gate hat->hat_ht_hash = NULL; 4547c478bd9Sstevel@tonic-gate 4557c478bd9Sstevel@tonic-gate hat->hat_flags = 0; 4567c478bd9Sstevel@tonic-gate kmem_cache_free(hat_cache, hat); 4577c478bd9Sstevel@tonic-gate } 4587c478bd9Sstevel@tonic-gate 4597c478bd9Sstevel@tonic-gate /* 4607c478bd9Sstevel@tonic-gate * round kernelbase down to a supported value to use for _userlimit 4617c478bd9Sstevel@tonic-gate * 4627c478bd9Sstevel@tonic-gate * userlimit must be aligned down to an entry in the top level htable. 4637c478bd9Sstevel@tonic-gate * The one exception is for 32 bit HAT's running PAE. 4647c478bd9Sstevel@tonic-gate */ 4657c478bd9Sstevel@tonic-gate uintptr_t 4667c478bd9Sstevel@tonic-gate hat_kernelbase(uintptr_t va) 4677c478bd9Sstevel@tonic-gate { 4687c478bd9Sstevel@tonic-gate #if defined(__i386) 4697c478bd9Sstevel@tonic-gate va &= LEVEL_MASK(1); 4707c478bd9Sstevel@tonic-gate #endif 4717c478bd9Sstevel@tonic-gate if (IN_VA_HOLE(va)) 4727c478bd9Sstevel@tonic-gate panic("_userlimit %p will fall in VA hole\n", (void *)va); 4737c478bd9Sstevel@tonic-gate return (va); 4747c478bd9Sstevel@tonic-gate } 4757c478bd9Sstevel@tonic-gate 476512cf780Skchow /* 477512cf780Skchow * 478512cf780Skchow */ 479512cf780Skchow static void 480512cf780Skchow set_max_page_level() 481512cf780Skchow { 482512cf780Skchow level_t lvl; 483512cf780Skchow 484512cf780Skchow if (!kbm_largepage_support) { 485512cf780Skchow lvl = 0; 4866f0cf5c5Skchow } else { 4876f0cf5c5Skchow if (x86_feature & X86_1GPG) { 4886f0cf5c5Skchow lvl = 2; 4896f0cf5c5Skchow if (chk_optimal_1gtlb && 4906f0cf5c5Skchow cpuid_opteron_erratum(CPU, 6671130)) { 4916f0cf5c5Skchow lvl = 1; 4926f0cf5c5Skchow } 4936f0cf5c5Skchow if (plat_mnode_xcheck(LEVEL_SIZE(2) >> 4946f0cf5c5Skchow LEVEL_SHIFT(0))) { 4956f0cf5c5Skchow lvl = 1; 4966f0cf5c5Skchow } 4976f0cf5c5Skchow } else { 498512cf780Skchow lvl = 1; 499512cf780Skchow } 500512cf780Skchow } 501512cf780Skchow mmu.max_page_level = lvl; 502512cf780Skchow 503512cf780Skchow if ((lvl == 2) && (enable_1gpg == 0)) 504512cf780Skchow mmu.umax_page_level = 1; 505512cf780Skchow else 506512cf780Skchow mmu.umax_page_level = lvl; 507512cf780Skchow } 508512cf780Skchow 5097c478bd9Sstevel@tonic-gate /* 5107c478bd9Sstevel@tonic-gate * Initialize hat data structures based on processor MMU information. 5117c478bd9Sstevel@tonic-gate */ 5127c478bd9Sstevel@tonic-gate void 5137c478bd9Sstevel@tonic-gate mmu_init(void) 5147c478bd9Sstevel@tonic-gate { 5157c478bd9Sstevel@tonic-gate uint_t max_htables; 5167c478bd9Sstevel@tonic-gate uint_t pa_bits; 5177c478bd9Sstevel@tonic-gate uint_t va_bits; 5187c478bd9Sstevel@tonic-gate int i; 5197c478bd9Sstevel@tonic-gate 5207c478bd9Sstevel@tonic-gate /* 521ae115bc7Smrj * If CPU enabled the page table global bit, use it for the kernel 522ae115bc7Smrj * This is bit 7 in CR4 (PGE - Page Global Enable). 5237c478bd9Sstevel@tonic-gate */ 524ae115bc7Smrj if ((x86_feature & X86_PGE) != 0 && (getcr4() & CR4_PGE) != 0) 5257c478bd9Sstevel@tonic-gate mmu.pt_global = PT_GLOBAL; 5267c478bd9Sstevel@tonic-gate 5277c478bd9Sstevel@tonic-gate /* 528ae115bc7Smrj * Detect NX and PAE usage. 5297c478bd9Sstevel@tonic-gate */ 530ae115bc7Smrj mmu.pae_hat = kbm_pae_support; 531ae115bc7Smrj if (kbm_nx_support) 5327c478bd9Sstevel@tonic-gate mmu.pt_nx = PT_NX; 533ae115bc7Smrj else 5347c478bd9Sstevel@tonic-gate mmu.pt_nx = 0; 5357c478bd9Sstevel@tonic-gate 5367c478bd9Sstevel@tonic-gate /* 5377c478bd9Sstevel@tonic-gate * Use CPU info to set various MMU parameters 5387c478bd9Sstevel@tonic-gate */ 5397c478bd9Sstevel@tonic-gate cpuid_get_addrsize(CPU, &pa_bits, &va_bits); 5407c478bd9Sstevel@tonic-gate 5417c478bd9Sstevel@tonic-gate if (va_bits < sizeof (void *) * NBBY) { 5427c478bd9Sstevel@tonic-gate mmu.hole_start = (1ul << (va_bits - 1)); 5437c478bd9Sstevel@tonic-gate mmu.hole_end = 0ul - mmu.hole_start - 1; 5447c478bd9Sstevel@tonic-gate } else { 5457c478bd9Sstevel@tonic-gate mmu.hole_end = 0; 5467c478bd9Sstevel@tonic-gate mmu.hole_start = mmu.hole_end - 1; 5477c478bd9Sstevel@tonic-gate } 5487c478bd9Sstevel@tonic-gate #if defined(OPTERON_ERRATUM_121) 5497c478bd9Sstevel@tonic-gate /* 5507c478bd9Sstevel@tonic-gate * If erratum 121 has already been detected at this time, hole_start 5517c478bd9Sstevel@tonic-gate * contains the value to be subtracted from mmu.hole_start. 5527c478bd9Sstevel@tonic-gate */ 5537c478bd9Sstevel@tonic-gate ASSERT(hole_start == 0 || opteron_erratum_121 != 0); 5547c478bd9Sstevel@tonic-gate hole_start = mmu.hole_start - hole_start; 5557c478bd9Sstevel@tonic-gate #else 5567c478bd9Sstevel@tonic-gate hole_start = mmu.hole_start; 5577c478bd9Sstevel@tonic-gate #endif 5587c478bd9Sstevel@tonic-gate hole_end = mmu.hole_end; 5597c478bd9Sstevel@tonic-gate 5607c478bd9Sstevel@tonic-gate mmu.highest_pfn = mmu_btop((1ull << pa_bits) - 1); 5617c478bd9Sstevel@tonic-gate if (mmu.pae_hat == 0 && pa_bits > 32) 5627c478bd9Sstevel@tonic-gate mmu.highest_pfn = PFN_4G - 1; 5637c478bd9Sstevel@tonic-gate 5647c478bd9Sstevel@tonic-gate if (mmu.pae_hat) { 5657c478bd9Sstevel@tonic-gate mmu.pte_size = 8; /* 8 byte PTEs */ 5667c478bd9Sstevel@tonic-gate mmu.pte_size_shift = 3; 5677c478bd9Sstevel@tonic-gate } else { 5687c478bd9Sstevel@tonic-gate mmu.pte_size = 4; /* 4 byte PTEs */ 5697c478bd9Sstevel@tonic-gate mmu.pte_size_shift = 2; 5707c478bd9Sstevel@tonic-gate } 5717c478bd9Sstevel@tonic-gate 5727c478bd9Sstevel@tonic-gate if (mmu.pae_hat && (x86_feature & X86_PAE) == 0) 5737c478bd9Sstevel@tonic-gate panic("Processor does not support PAE"); 5747c478bd9Sstevel@tonic-gate 5757c478bd9Sstevel@tonic-gate if ((x86_feature & X86_CX8) == 0) 5767c478bd9Sstevel@tonic-gate panic("Processor does not support cmpxchg8b instruction"); 5777c478bd9Sstevel@tonic-gate 5787c478bd9Sstevel@tonic-gate #if defined(__amd64) 5797c478bd9Sstevel@tonic-gate 5807c478bd9Sstevel@tonic-gate mmu.num_level = 4; 5817c478bd9Sstevel@tonic-gate mmu.max_level = 3; 5827c478bd9Sstevel@tonic-gate mmu.ptes_per_table = 512; 5837c478bd9Sstevel@tonic-gate mmu.top_level_count = 512; 5847c478bd9Sstevel@tonic-gate 5857c478bd9Sstevel@tonic-gate mmu.level_shift[0] = 12; 5867c478bd9Sstevel@tonic-gate mmu.level_shift[1] = 21; 5877c478bd9Sstevel@tonic-gate mmu.level_shift[2] = 30; 5887c478bd9Sstevel@tonic-gate mmu.level_shift[3] = 39; 5897c478bd9Sstevel@tonic-gate 5907c478bd9Sstevel@tonic-gate #elif defined(__i386) 5917c478bd9Sstevel@tonic-gate 5927c478bd9Sstevel@tonic-gate if (mmu.pae_hat) { 5937c478bd9Sstevel@tonic-gate mmu.num_level = 3; 5947c478bd9Sstevel@tonic-gate mmu.max_level = 2; 5957c478bd9Sstevel@tonic-gate mmu.ptes_per_table = 512; 5967c478bd9Sstevel@tonic-gate mmu.top_level_count = 4; 5977c478bd9Sstevel@tonic-gate 5987c478bd9Sstevel@tonic-gate mmu.level_shift[0] = 12; 5997c478bd9Sstevel@tonic-gate mmu.level_shift[1] = 21; 6007c478bd9Sstevel@tonic-gate mmu.level_shift[2] = 30; 6017c478bd9Sstevel@tonic-gate 6027c478bd9Sstevel@tonic-gate } else { 6037c478bd9Sstevel@tonic-gate mmu.num_level = 2; 6047c478bd9Sstevel@tonic-gate mmu.max_level = 1; 6057c478bd9Sstevel@tonic-gate mmu.ptes_per_table = 1024; 6067c478bd9Sstevel@tonic-gate mmu.top_level_count = 1024; 6077c478bd9Sstevel@tonic-gate 6087c478bd9Sstevel@tonic-gate mmu.level_shift[0] = 12; 6097c478bd9Sstevel@tonic-gate mmu.level_shift[1] = 22; 6107c478bd9Sstevel@tonic-gate } 6117c478bd9Sstevel@tonic-gate 6127c478bd9Sstevel@tonic-gate #endif /* __i386 */ 6137c478bd9Sstevel@tonic-gate 6147c478bd9Sstevel@tonic-gate for (i = 0; i < mmu.num_level; ++i) { 6157c478bd9Sstevel@tonic-gate mmu.level_size[i] = 1UL << mmu.level_shift[i]; 6167c478bd9Sstevel@tonic-gate mmu.level_offset[i] = mmu.level_size[i] - 1; 6177c478bd9Sstevel@tonic-gate mmu.level_mask[i] = ~mmu.level_offset[i]; 6187c478bd9Sstevel@tonic-gate } 6197c478bd9Sstevel@tonic-gate 620512cf780Skchow set_max_page_level(); 621512cf780Skchow 622472714d6Skchow mmu_page_sizes = mmu.max_page_level + 1; 623472714d6Skchow mmu_exported_page_sizes = mmu.umax_page_level + 1; 624472714d6Skchow 625472714d6Skchow /* restrict legacy applications from using pagesizes 1g and above */ 626472714d6Skchow mmu_legacy_page_sizes = 627472714d6Skchow (mmu_exported_page_sizes > 2) ? 2 : mmu_exported_page_sizes; 628472714d6Skchow 629472714d6Skchow 630ae115bc7Smrj for (i = 0; i <= mmu.max_page_level; ++i) { 631ab4a9bebSjohnlev mmu.pte_bits[i] = PT_VALID | pt_kern; 632ae115bc7Smrj if (i > 0) 633ae115bc7Smrj mmu.pte_bits[i] |= PT_PAGESIZE; 634ae115bc7Smrj } 6357c478bd9Sstevel@tonic-gate 6367c478bd9Sstevel@tonic-gate /* 6377c478bd9Sstevel@tonic-gate * NOTE Legacy 32 bit PAE mode only has the P_VALID bit at top level. 6387c478bd9Sstevel@tonic-gate */ 6397c478bd9Sstevel@tonic-gate for (i = 1; i < mmu.num_level; ++i) 6407c478bd9Sstevel@tonic-gate mmu.ptp_bits[i] = PT_PTPBITS; 641ae115bc7Smrj 6427c478bd9Sstevel@tonic-gate #if defined(__i386) 6437c478bd9Sstevel@tonic-gate mmu.ptp_bits[2] = PT_VALID; 6447c478bd9Sstevel@tonic-gate #endif 6457c478bd9Sstevel@tonic-gate 6467c478bd9Sstevel@tonic-gate /* 6477c478bd9Sstevel@tonic-gate * Compute how many hash table entries to have per process for htables. 6487c478bd9Sstevel@tonic-gate * We start with 1 page's worth of entries. 6497c478bd9Sstevel@tonic-gate * 6507c478bd9Sstevel@tonic-gate * If physical memory is small, reduce the amount need to cover it. 6517c478bd9Sstevel@tonic-gate */ 6527c478bd9Sstevel@tonic-gate max_htables = physmax / mmu.ptes_per_table; 6537c478bd9Sstevel@tonic-gate mmu.hash_cnt = MMU_PAGESIZE / sizeof (htable_t *); 6547c478bd9Sstevel@tonic-gate while (mmu.hash_cnt > 16 && mmu.hash_cnt >= max_htables) 6557c478bd9Sstevel@tonic-gate mmu.hash_cnt >>= 1; 6567c478bd9Sstevel@tonic-gate mmu.vlp_hash_cnt = mmu.hash_cnt; 6577c478bd9Sstevel@tonic-gate 6587c478bd9Sstevel@tonic-gate #if defined(__amd64) 6597c478bd9Sstevel@tonic-gate /* 6607c478bd9Sstevel@tonic-gate * If running in 64 bits and physical memory is large, 6617c478bd9Sstevel@tonic-gate * increase the size of the cache to cover all of memory for 6627c478bd9Sstevel@tonic-gate * a 64 bit process. 6637c478bd9Sstevel@tonic-gate */ 6647c478bd9Sstevel@tonic-gate #define HASH_MAX_LENGTH 4 6657c478bd9Sstevel@tonic-gate while (mmu.hash_cnt * HASH_MAX_LENGTH < max_htables) 6667c478bd9Sstevel@tonic-gate mmu.hash_cnt <<= 1; 6677c478bd9Sstevel@tonic-gate #endif 6687c478bd9Sstevel@tonic-gate } 6697c478bd9Sstevel@tonic-gate 6707c478bd9Sstevel@tonic-gate 6717c478bd9Sstevel@tonic-gate /* 6727c478bd9Sstevel@tonic-gate * initialize hat data structures 6737c478bd9Sstevel@tonic-gate */ 6747c478bd9Sstevel@tonic-gate void 6757c478bd9Sstevel@tonic-gate hat_init() 6767c478bd9Sstevel@tonic-gate { 6777c478bd9Sstevel@tonic-gate #if defined(__i386) 6787c478bd9Sstevel@tonic-gate /* 6797c478bd9Sstevel@tonic-gate * _userlimit must be aligned correctly 6807c478bd9Sstevel@tonic-gate */ 6817c478bd9Sstevel@tonic-gate if ((_userlimit & LEVEL_MASK(1)) != _userlimit) { 6827c478bd9Sstevel@tonic-gate prom_printf("hat_init(): _userlimit=%p, not aligned at %p\n", 6837c478bd9Sstevel@tonic-gate (void *)_userlimit, (void *)LEVEL_SIZE(1)); 6847c478bd9Sstevel@tonic-gate halt("hat_init(): Unable to continue"); 6857c478bd9Sstevel@tonic-gate } 6867c478bd9Sstevel@tonic-gate #endif 6877c478bd9Sstevel@tonic-gate 6887c478bd9Sstevel@tonic-gate cv_init(&hat_list_cv, NULL, CV_DEFAULT, NULL); 6897c478bd9Sstevel@tonic-gate 6907c478bd9Sstevel@tonic-gate /* 6917c478bd9Sstevel@tonic-gate * initialize kmem caches 6927c478bd9Sstevel@tonic-gate */ 6937c478bd9Sstevel@tonic-gate htable_init(); 6947c478bd9Sstevel@tonic-gate hment_init(); 6957c478bd9Sstevel@tonic-gate 6967c478bd9Sstevel@tonic-gate hat_cache = kmem_cache_create("hat_t", 6977c478bd9Sstevel@tonic-gate sizeof (hat_t), 0, hati_constructor, NULL, NULL, 6987c478bd9Sstevel@tonic-gate NULL, 0, 0); 6997c478bd9Sstevel@tonic-gate 7007c478bd9Sstevel@tonic-gate hat_hash_cache = kmem_cache_create("HatHash", 7017c478bd9Sstevel@tonic-gate mmu.hash_cnt * sizeof (htable_t *), 0, NULL, NULL, NULL, 7027c478bd9Sstevel@tonic-gate NULL, 0, 0); 7037c478bd9Sstevel@tonic-gate 7047c478bd9Sstevel@tonic-gate /* 7057c478bd9Sstevel@tonic-gate * VLP hats can use a smaller hash table size on large memroy machines 7067c478bd9Sstevel@tonic-gate */ 7077c478bd9Sstevel@tonic-gate if (mmu.hash_cnt == mmu.vlp_hash_cnt) { 7087c478bd9Sstevel@tonic-gate vlp_hash_cache = hat_hash_cache; 7097c478bd9Sstevel@tonic-gate } else { 7107c478bd9Sstevel@tonic-gate vlp_hash_cache = kmem_cache_create("HatVlpHash", 7117c478bd9Sstevel@tonic-gate mmu.vlp_hash_cnt * sizeof (htable_t *), 0, NULL, NULL, NULL, 7127c478bd9Sstevel@tonic-gate NULL, 0, 0); 7137c478bd9Sstevel@tonic-gate } 7147c478bd9Sstevel@tonic-gate 7157c478bd9Sstevel@tonic-gate /* 7167c478bd9Sstevel@tonic-gate * Set up the kernel's hat 7177c478bd9Sstevel@tonic-gate */ 7187c478bd9Sstevel@tonic-gate AS_LOCK_ENTER(&kas, &kas.a_lock, RW_WRITER); 7197c478bd9Sstevel@tonic-gate kas.a_hat = kmem_cache_alloc(hat_cache, KM_NOSLEEP); 7207c478bd9Sstevel@tonic-gate mutex_init(&kas.a_hat->hat_mutex, NULL, MUTEX_DEFAULT, NULL); 7217c478bd9Sstevel@tonic-gate kas.a_hat->hat_as = &kas; 7227c478bd9Sstevel@tonic-gate kas.a_hat->hat_flags = 0; 7237c478bd9Sstevel@tonic-gate AS_LOCK_EXIT(&kas, &kas.a_lock); 7247c478bd9Sstevel@tonic-gate 7257c478bd9Sstevel@tonic-gate CPUSET_ZERO(khat_cpuset); 7267c478bd9Sstevel@tonic-gate CPUSET_ADD(khat_cpuset, CPU->cpu_id); 7277c478bd9Sstevel@tonic-gate 7287c478bd9Sstevel@tonic-gate /* 7297c478bd9Sstevel@tonic-gate * The kernel hat's next pointer serves as the head of the hat list . 730a85a6733Sjosephb * The kernel hat's prev pointer tracks the last hat on the list for 731a85a6733Sjosephb * htable_steal() to use. 7327c478bd9Sstevel@tonic-gate */ 7337c478bd9Sstevel@tonic-gate kas.a_hat->hat_next = NULL; 734a85a6733Sjosephb kas.a_hat->hat_prev = NULL; 7357c478bd9Sstevel@tonic-gate 7367c478bd9Sstevel@tonic-gate /* 7377c478bd9Sstevel@tonic-gate * Allocate an htable hash bucket for the kernel 7387c478bd9Sstevel@tonic-gate * XX64 - tune for 64 bit procs 7397c478bd9Sstevel@tonic-gate */ 7407c478bd9Sstevel@tonic-gate kas.a_hat->hat_num_hash = mmu.hash_cnt; 7417c478bd9Sstevel@tonic-gate kas.a_hat->hat_ht_hash = kmem_cache_alloc(hat_hash_cache, KM_NOSLEEP); 7427c478bd9Sstevel@tonic-gate bzero(kas.a_hat->hat_ht_hash, mmu.hash_cnt * sizeof (htable_t *)); 7437c478bd9Sstevel@tonic-gate 7447c478bd9Sstevel@tonic-gate /* 7457c478bd9Sstevel@tonic-gate * zero out the top level and cached htable pointers 7467c478bd9Sstevel@tonic-gate */ 7477c478bd9Sstevel@tonic-gate kas.a_hat->hat_ht_cached = NULL; 7487c478bd9Sstevel@tonic-gate kas.a_hat->hat_htable = NULL; 7499d9461f9Strevtom 7509d9461f9Strevtom /* 7519d9461f9Strevtom * Pre-allocate hrm_hashtab before enabling the collection of 7529d9461f9Strevtom * refmod statistics. Allocating on the fly would mean us 7539d9461f9Strevtom * running the risk of suffering recursive mutex enters or 7549d9461f9Strevtom * deadlocks. 7559d9461f9Strevtom */ 7569d9461f9Strevtom hrm_hashtab = kmem_zalloc(HRM_HASHSIZE * sizeof (struct hrmstat *), 7579d9461f9Strevtom KM_SLEEP); 7587c478bd9Sstevel@tonic-gate } 7597c478bd9Sstevel@tonic-gate 7607c478bd9Sstevel@tonic-gate /* 7617c478bd9Sstevel@tonic-gate * Prepare CPU specific pagetables for VLP processes on 64 bit kernels. 7627c478bd9Sstevel@tonic-gate * 7637c478bd9Sstevel@tonic-gate * Each CPU has a set of 2 pagetables that are reused for any 32 bit 7647c478bd9Sstevel@tonic-gate * process it runs. They are the top level pagetable, hci_vlp_l3ptes, and 7657c478bd9Sstevel@tonic-gate * the next to top level table for the bottom 512 Gig, hci_vlp_l2ptes. 7667c478bd9Sstevel@tonic-gate */ 7677c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 7687c478bd9Sstevel@tonic-gate static void 7697c478bd9Sstevel@tonic-gate hat_vlp_setup(struct cpu *cpu) 7707c478bd9Sstevel@tonic-gate { 771843e1988Sjohnlev #if defined(__amd64) && !defined(__xpv) 7727c478bd9Sstevel@tonic-gate struct hat_cpu_info *hci = cpu->cpu_hat_info; 7737c478bd9Sstevel@tonic-gate pfn_t pfn; 7747c478bd9Sstevel@tonic-gate 7757c478bd9Sstevel@tonic-gate /* 7767c478bd9Sstevel@tonic-gate * allocate the level==2 page table for the bottom most 7777c478bd9Sstevel@tonic-gate * 512Gig of address space (this is where 32 bit apps live) 7787c478bd9Sstevel@tonic-gate */ 7797c478bd9Sstevel@tonic-gate ASSERT(hci != NULL); 7807c478bd9Sstevel@tonic-gate hci->hci_vlp_l2ptes = kmem_zalloc(MMU_PAGESIZE, KM_SLEEP); 7817c478bd9Sstevel@tonic-gate 7827c478bd9Sstevel@tonic-gate /* 7837c478bd9Sstevel@tonic-gate * Allocate a top level pagetable and copy the kernel's 7847c478bd9Sstevel@tonic-gate * entries into it. Then link in hci_vlp_l2ptes in the 1st entry. 7857c478bd9Sstevel@tonic-gate */ 7867c478bd9Sstevel@tonic-gate hci->hci_vlp_l3ptes = kmem_zalloc(MMU_PAGESIZE, KM_SLEEP); 7877c478bd9Sstevel@tonic-gate hci->hci_vlp_pfn = 7887c478bd9Sstevel@tonic-gate hat_getpfnum(kas.a_hat, (caddr_t)hci->hci_vlp_l3ptes); 7897c478bd9Sstevel@tonic-gate ASSERT(hci->hci_vlp_pfn != PFN_INVALID); 790843e1988Sjohnlev bcopy(vlp_page, hci->hci_vlp_l3ptes, MMU_PAGESIZE); 7917c478bd9Sstevel@tonic-gate 7927c478bd9Sstevel@tonic-gate pfn = hat_getpfnum(kas.a_hat, (caddr_t)hci->hci_vlp_l2ptes); 7937c478bd9Sstevel@tonic-gate ASSERT(pfn != PFN_INVALID); 7947c478bd9Sstevel@tonic-gate hci->hci_vlp_l3ptes[0] = MAKEPTP(pfn, 2); 795843e1988Sjohnlev #endif /* __amd64 && !__xpv */ 7967c478bd9Sstevel@tonic-gate } 7977c478bd9Sstevel@tonic-gate 798ae115bc7Smrj /*ARGSUSED*/ 799ae115bc7Smrj static void 800ae115bc7Smrj hat_vlp_teardown(cpu_t *cpu) 801ae115bc7Smrj { 802843e1988Sjohnlev #if defined(__amd64) && !defined(__xpv) 803ae115bc7Smrj struct hat_cpu_info *hci; 804ae115bc7Smrj 805ae115bc7Smrj if ((hci = cpu->cpu_hat_info) == NULL) 806ae115bc7Smrj return; 807ae115bc7Smrj if (hci->hci_vlp_l2ptes) 808ae115bc7Smrj kmem_free(hci->hci_vlp_l2ptes, MMU_PAGESIZE); 809ae115bc7Smrj if (hci->hci_vlp_l3ptes) 810ae115bc7Smrj kmem_free(hci->hci_vlp_l3ptes, MMU_PAGESIZE); 811843e1988Sjohnlev #endif 812843e1988Sjohnlev } 813843e1988Sjohnlev 814843e1988Sjohnlev #define NEXT_HKR(r, l, s, e) { \ 815843e1988Sjohnlev kernel_ranges[r].hkr_level = l; \ 816843e1988Sjohnlev kernel_ranges[r].hkr_start_va = s; \ 817843e1988Sjohnlev kernel_ranges[r].hkr_end_va = e; \ 818843e1988Sjohnlev ++r; \ 819ae115bc7Smrj } 820ae115bc7Smrj 8217c478bd9Sstevel@tonic-gate /* 8227c478bd9Sstevel@tonic-gate * Finish filling in the kernel hat. 8237c478bd9Sstevel@tonic-gate * Pre fill in all top level kernel page table entries for the kernel's 8247c478bd9Sstevel@tonic-gate * part of the address range. From this point on we can't use any new 8257c478bd9Sstevel@tonic-gate * kernel large pages if they need PTE's at max_level 826ae115bc7Smrj * 827ae115bc7Smrj * create the kmap mappings. 8287c478bd9Sstevel@tonic-gate */ 8297c478bd9Sstevel@tonic-gate void 8307c478bd9Sstevel@tonic-gate hat_init_finish(void) 8317c478bd9Sstevel@tonic-gate { 832ae115bc7Smrj size_t size; 833843e1988Sjohnlev uint_t r = 0; 834843e1988Sjohnlev uintptr_t va; 835843e1988Sjohnlev hat_kernel_range_t *rp; 8367c478bd9Sstevel@tonic-gate 8377c478bd9Sstevel@tonic-gate 8387c478bd9Sstevel@tonic-gate /* 839843e1988Sjohnlev * We are now effectively running on the kernel hat. 840843e1988Sjohnlev * Clearing use_boot_reserve shuts off using the pre-allocated boot 841843e1988Sjohnlev * reserve for all HAT allocations. From here on, the reserves are 842843e1988Sjohnlev * only used when avoiding recursion in kmem_alloc(). 8437c478bd9Sstevel@tonic-gate */ 844843e1988Sjohnlev use_boot_reserve = 0; 845843e1988Sjohnlev htable_adjust_reserve(); 8467c478bd9Sstevel@tonic-gate 8477c478bd9Sstevel@tonic-gate /* 848843e1988Sjohnlev * User HATs are initialized with copies of all kernel mappings in 849843e1988Sjohnlev * higher level page tables. Ensure that those entries exist. 8507c478bd9Sstevel@tonic-gate */ 851843e1988Sjohnlev #if defined(__amd64) 852843e1988Sjohnlev 853843e1988Sjohnlev NEXT_HKR(r, 3, kernelbase, 0); 854843e1988Sjohnlev #if defined(__xpv) 855843e1988Sjohnlev NEXT_HKR(r, 3, HYPERVISOR_VIRT_START, HYPERVISOR_VIRT_END); 856843e1988Sjohnlev #endif 857843e1988Sjohnlev 858843e1988Sjohnlev #elif defined(__i386) 859843e1988Sjohnlev 860843e1988Sjohnlev #if !defined(__xpv) 861843e1988Sjohnlev if (mmu.pae_hat) { 862843e1988Sjohnlev va = kernelbase; 863843e1988Sjohnlev if ((va & LEVEL_MASK(2)) != va) { 864843e1988Sjohnlev va = P2ROUNDUP(va, LEVEL_SIZE(2)); 865843e1988Sjohnlev NEXT_HKR(r, 1, kernelbase, va); 866843e1988Sjohnlev } 867843e1988Sjohnlev if (va != 0) 868843e1988Sjohnlev NEXT_HKR(r, 2, va, 0); 869843e1988Sjohnlev } else 870843e1988Sjohnlev #endif /* __xpv */ 871843e1988Sjohnlev NEXT_HKR(r, 1, kernelbase, 0); 872843e1988Sjohnlev 873843e1988Sjohnlev #endif /* __i386 */ 874843e1988Sjohnlev 875843e1988Sjohnlev num_kernel_ranges = r; 8767c478bd9Sstevel@tonic-gate 8777c478bd9Sstevel@tonic-gate /* 878843e1988Sjohnlev * Create all the kernel pagetables that will have entries 879843e1988Sjohnlev * shared to user HATs. 8807c478bd9Sstevel@tonic-gate */ 881843e1988Sjohnlev for (r = 0; r < num_kernel_ranges; ++r) { 882843e1988Sjohnlev rp = &kernel_ranges[r]; 883843e1988Sjohnlev for (va = rp->hkr_start_va; va != rp->hkr_end_va; 884843e1988Sjohnlev va += LEVEL_SIZE(rp->hkr_level)) { 885843e1988Sjohnlev htable_t *ht; 886843e1988Sjohnlev 887843e1988Sjohnlev if (IN_HYPERVISOR_VA(va)) 888843e1988Sjohnlev continue; 889843e1988Sjohnlev 890843e1988Sjohnlev /* can/must skip if a page mapping already exists */ 891843e1988Sjohnlev if (rp->hkr_level <= mmu.max_page_level && 892843e1988Sjohnlev (ht = htable_getpage(kas.a_hat, va, NULL)) != 893843e1988Sjohnlev NULL) { 894843e1988Sjohnlev htable_release(ht); 895843e1988Sjohnlev continue; 896843e1988Sjohnlev } 897843e1988Sjohnlev 898843e1988Sjohnlev (void) htable_create(kas.a_hat, va, rp->hkr_level - 1, 899843e1988Sjohnlev NULL); 900843e1988Sjohnlev } 901843e1988Sjohnlev } 9027c478bd9Sstevel@tonic-gate 9037c478bd9Sstevel@tonic-gate /* 904843e1988Sjohnlev * 32 bit PAE metal kernels use only 4 of the 512 entries in the 905843e1988Sjohnlev * page holding the top level pagetable. We use the remainder for 906843e1988Sjohnlev * the "per CPU" page tables for VLP processes. 907843e1988Sjohnlev * Map the top level kernel pagetable into the kernel to make 908843e1988Sjohnlev * it easy to use bcopy access these tables. 9097c478bd9Sstevel@tonic-gate */ 9107c478bd9Sstevel@tonic-gate if (mmu.pae_hat) { 9117c478bd9Sstevel@tonic-gate vlp_page = vmem_alloc(heap_arena, MMU_PAGESIZE, VM_SLEEP); 9127c478bd9Sstevel@tonic-gate hat_devload(kas.a_hat, (caddr_t)vlp_page, MMU_PAGESIZE, 9137c478bd9Sstevel@tonic-gate kas.a_hat->hat_htable->ht_pfn, 914843e1988Sjohnlev #if !defined(__xpv) 915ae115bc7Smrj PROT_WRITE | 916843e1988Sjohnlev #endif 917ae115bc7Smrj PROT_READ | HAT_NOSYNC | HAT_UNORDERED_OK, 9187c478bd9Sstevel@tonic-gate HAT_LOAD | HAT_LOAD_NOCONSIST); 9197c478bd9Sstevel@tonic-gate } 9207c478bd9Sstevel@tonic-gate hat_vlp_setup(CPU); 921ae115bc7Smrj 922ae115bc7Smrj /* 923ae115bc7Smrj * Create kmap (cached mappings of kernel PTEs) 924ae115bc7Smrj * for 32 bit we map from segmap_start .. ekernelheap 925ae115bc7Smrj * for 64 bit we map from segmap_start .. segmap_start + segmapsize; 926ae115bc7Smrj */ 927ae115bc7Smrj #if defined(__i386) 928ae115bc7Smrj size = (uintptr_t)ekernelheap - segmap_start; 929ae115bc7Smrj #elif defined(__amd64) 930ae115bc7Smrj size = segmapsize; 931ae115bc7Smrj #endif 932ae115bc7Smrj hat_kmap_init((uintptr_t)segmap_start, size); 9337c478bd9Sstevel@tonic-gate } 9347c478bd9Sstevel@tonic-gate 9357c478bd9Sstevel@tonic-gate /* 9367c478bd9Sstevel@tonic-gate * On 32 bit PAE mode, PTE's are 64 bits, but ordinary atomic memory references 9377c478bd9Sstevel@tonic-gate * are 32 bit, so for safety we must use cas64() to install these. 9387c478bd9Sstevel@tonic-gate */ 9397c478bd9Sstevel@tonic-gate #ifdef __i386 9407c478bd9Sstevel@tonic-gate static void 9417c478bd9Sstevel@tonic-gate reload_pae32(hat_t *hat, cpu_t *cpu) 9427c478bd9Sstevel@tonic-gate { 9437c478bd9Sstevel@tonic-gate x86pte_t *src; 9447c478bd9Sstevel@tonic-gate x86pte_t *dest; 9457c478bd9Sstevel@tonic-gate x86pte_t pte; 9467c478bd9Sstevel@tonic-gate int i; 9477c478bd9Sstevel@tonic-gate 9487c478bd9Sstevel@tonic-gate /* 9497c478bd9Sstevel@tonic-gate * Load the 4 entries of the level 2 page table into this 9507c478bd9Sstevel@tonic-gate * cpu's range of the vlp_page and point cr3 at them. 9517c478bd9Sstevel@tonic-gate */ 9527c478bd9Sstevel@tonic-gate ASSERT(mmu.pae_hat); 9537c478bd9Sstevel@tonic-gate src = hat->hat_vlp_ptes; 9547c478bd9Sstevel@tonic-gate dest = vlp_page + (cpu->cpu_id + 1) * VLP_NUM_PTES; 9557c478bd9Sstevel@tonic-gate for (i = 0; i < VLP_NUM_PTES; ++i) { 9567c478bd9Sstevel@tonic-gate for (;;) { 9577c478bd9Sstevel@tonic-gate pte = dest[i]; 9587c478bd9Sstevel@tonic-gate if (pte == src[i]) 9597c478bd9Sstevel@tonic-gate break; 9607c478bd9Sstevel@tonic-gate if (cas64(dest + i, pte, src[i]) != src[i]) 9617c478bd9Sstevel@tonic-gate break; 9627c478bd9Sstevel@tonic-gate } 9637c478bd9Sstevel@tonic-gate } 9647c478bd9Sstevel@tonic-gate } 9657c478bd9Sstevel@tonic-gate #endif 9667c478bd9Sstevel@tonic-gate 9677c478bd9Sstevel@tonic-gate /* 9687c478bd9Sstevel@tonic-gate * Switch to a new active hat, maintaining bit masks to track active CPUs. 969843e1988Sjohnlev * 970843e1988Sjohnlev * On the 32-bit PAE hypervisor, %cr3 is a 64-bit value, on metal it 971843e1988Sjohnlev * remains a 32-bit value. 9727c478bd9Sstevel@tonic-gate */ 9737c478bd9Sstevel@tonic-gate void 9747c478bd9Sstevel@tonic-gate hat_switch(hat_t *hat) 9757c478bd9Sstevel@tonic-gate { 976843e1988Sjohnlev uint64_t newcr3; 9777c478bd9Sstevel@tonic-gate cpu_t *cpu = CPU; 9787c478bd9Sstevel@tonic-gate hat_t *old = cpu->cpu_current_hat; 9797c478bd9Sstevel@tonic-gate 9807c478bd9Sstevel@tonic-gate /* 9817c478bd9Sstevel@tonic-gate * set up this information first, so we don't miss any cross calls 9827c478bd9Sstevel@tonic-gate */ 9837c478bd9Sstevel@tonic-gate if (old != NULL) { 9847c478bd9Sstevel@tonic-gate if (old == hat) 9857c478bd9Sstevel@tonic-gate return; 9867c478bd9Sstevel@tonic-gate if (old != kas.a_hat) 9877c478bd9Sstevel@tonic-gate CPUSET_ATOMIC_DEL(old->hat_cpus, cpu->cpu_id); 9887c478bd9Sstevel@tonic-gate } 9897c478bd9Sstevel@tonic-gate 9907c478bd9Sstevel@tonic-gate /* 99195c0a3c8Sjosephb * Add this CPU to the active set for this HAT. 9927c478bd9Sstevel@tonic-gate */ 9937c478bd9Sstevel@tonic-gate if (hat != kas.a_hat) { 9947c478bd9Sstevel@tonic-gate CPUSET_ATOMIC_ADD(hat->hat_cpus, cpu->cpu_id); 9957c478bd9Sstevel@tonic-gate } 9967c478bd9Sstevel@tonic-gate cpu->cpu_current_hat = hat; 9977c478bd9Sstevel@tonic-gate 9987c478bd9Sstevel@tonic-gate /* 9997c478bd9Sstevel@tonic-gate * now go ahead and load cr3 10007c478bd9Sstevel@tonic-gate */ 10017c478bd9Sstevel@tonic-gate if (hat->hat_flags & HAT_VLP) { 10027c478bd9Sstevel@tonic-gate #if defined(__amd64) 10037c478bd9Sstevel@tonic-gate x86pte_t *vlpptep = cpu->cpu_hat_info->hci_vlp_l2ptes; 10047c478bd9Sstevel@tonic-gate 10057c478bd9Sstevel@tonic-gate VLP_COPY(hat->hat_vlp_ptes, vlpptep); 10067c478bd9Sstevel@tonic-gate newcr3 = MAKECR3(cpu->cpu_hat_info->hci_vlp_pfn); 10077c478bd9Sstevel@tonic-gate #elif defined(__i386) 10087c478bd9Sstevel@tonic-gate reload_pae32(hat, cpu); 10097c478bd9Sstevel@tonic-gate newcr3 = MAKECR3(kas.a_hat->hat_htable->ht_pfn) + 10107c478bd9Sstevel@tonic-gate (cpu->cpu_id + 1) * VLP_SIZE; 10117c478bd9Sstevel@tonic-gate #endif 10127c478bd9Sstevel@tonic-gate } else { 1013843e1988Sjohnlev newcr3 = MAKECR3((uint64_t)hat->hat_htable->ht_pfn); 10147c478bd9Sstevel@tonic-gate } 1015843e1988Sjohnlev #ifdef __xpv 1016843e1988Sjohnlev { 1017843e1988Sjohnlev struct mmuext_op t[2]; 1018843e1988Sjohnlev uint_t retcnt; 1019843e1988Sjohnlev uint_t opcnt = 1; 1020843e1988Sjohnlev 1021843e1988Sjohnlev t[0].cmd = MMUEXT_NEW_BASEPTR; 1022843e1988Sjohnlev t[0].arg1.mfn = mmu_btop(pa_to_ma(newcr3)); 1023843e1988Sjohnlev #if defined(__amd64) 1024843e1988Sjohnlev /* 1025843e1988Sjohnlev * There's an interesting problem here, as to what to 1026843e1988Sjohnlev * actually specify when switching to the kernel hat. 1027843e1988Sjohnlev * For now we'll reuse the kernel hat again. 1028843e1988Sjohnlev */ 1029843e1988Sjohnlev t[1].cmd = MMUEXT_NEW_USER_BASEPTR; 1030843e1988Sjohnlev if (hat == kas.a_hat) 1031843e1988Sjohnlev t[1].arg1.mfn = mmu_btop(pa_to_ma(newcr3)); 1032843e1988Sjohnlev else 1033843e1988Sjohnlev t[1].arg1.mfn = pfn_to_mfn(hat->hat_user_ptable); 1034843e1988Sjohnlev ++opcnt; 1035843e1988Sjohnlev #endif /* __amd64 */ 1036843e1988Sjohnlev if (HYPERVISOR_mmuext_op(t, opcnt, &retcnt, DOMID_SELF) < 0) 1037843e1988Sjohnlev panic("HYPERVISOR_mmu_update() failed"); 1038843e1988Sjohnlev ASSERT(retcnt == opcnt); 1039843e1988Sjohnlev 1040843e1988Sjohnlev } 1041843e1988Sjohnlev #else 10427c478bd9Sstevel@tonic-gate setcr3(newcr3); 1043843e1988Sjohnlev #endif 10447c478bd9Sstevel@tonic-gate ASSERT(cpu == CPU); 10457c478bd9Sstevel@tonic-gate } 10467c478bd9Sstevel@tonic-gate 10477c478bd9Sstevel@tonic-gate /* 10487c478bd9Sstevel@tonic-gate * Utility to return a valid x86pte_t from protections, pfn, and level number 10497c478bd9Sstevel@tonic-gate */ 10507c478bd9Sstevel@tonic-gate static x86pte_t 10517c478bd9Sstevel@tonic-gate hati_mkpte(pfn_t pfn, uint_t attr, level_t level, uint_t flags) 10527c478bd9Sstevel@tonic-gate { 10537c478bd9Sstevel@tonic-gate x86pte_t pte; 10547c478bd9Sstevel@tonic-gate uint_t cache_attr = attr & HAT_ORDER_MASK; 10557c478bd9Sstevel@tonic-gate 10567c478bd9Sstevel@tonic-gate pte = MAKEPTE(pfn, level); 10577c478bd9Sstevel@tonic-gate 10587c478bd9Sstevel@tonic-gate if (attr & PROT_WRITE) 10597c478bd9Sstevel@tonic-gate PTE_SET(pte, PT_WRITABLE); 10607c478bd9Sstevel@tonic-gate 10617c478bd9Sstevel@tonic-gate if (attr & PROT_USER) 10627c478bd9Sstevel@tonic-gate PTE_SET(pte, PT_USER); 10637c478bd9Sstevel@tonic-gate 10647c478bd9Sstevel@tonic-gate if (!(attr & PROT_EXEC)) 10657c478bd9Sstevel@tonic-gate PTE_SET(pte, mmu.pt_nx); 10667c478bd9Sstevel@tonic-gate 10677c478bd9Sstevel@tonic-gate /* 1068ae115bc7Smrj * Set the software bits used track ref/mod sync's and hments. 1069ae115bc7Smrj * If not using REF/MOD, set them to avoid h/w rewriting PTEs. 10707c478bd9Sstevel@tonic-gate */ 10717c478bd9Sstevel@tonic-gate if (flags & HAT_LOAD_NOCONSIST) 1072ae115bc7Smrj PTE_SET(pte, PT_NOCONSIST | PT_REF | PT_MOD); 1073ae115bc7Smrj else if (attr & HAT_NOSYNC) 1074ae115bc7Smrj PTE_SET(pte, PT_NOSYNC | PT_REF | PT_MOD); 10757c478bd9Sstevel@tonic-gate 10767c478bd9Sstevel@tonic-gate /* 10777c478bd9Sstevel@tonic-gate * Set the caching attributes in the PTE. The combination 10787c478bd9Sstevel@tonic-gate * of attributes are poorly defined, so we pay attention 10797c478bd9Sstevel@tonic-gate * to them in the given order. 10807c478bd9Sstevel@tonic-gate * 10817c478bd9Sstevel@tonic-gate * The test for HAT_STRICTORDER is different because it's defined 10827c478bd9Sstevel@tonic-gate * as "0" - which was a stupid thing to do, but is too late to change! 10837c478bd9Sstevel@tonic-gate */ 10847c478bd9Sstevel@tonic-gate if (cache_attr == HAT_STRICTORDER) { 10857c478bd9Sstevel@tonic-gate PTE_SET(pte, PT_NOCACHE); 10867c478bd9Sstevel@tonic-gate /*LINTED [Lint hates empty ifs, but it's the obvious way to do this] */ 10877c478bd9Sstevel@tonic-gate } else if (cache_attr & (HAT_UNORDERED_OK | HAT_STORECACHING_OK)) { 10887c478bd9Sstevel@tonic-gate /* nothing to set */; 10897c478bd9Sstevel@tonic-gate } else if (cache_attr & (HAT_MERGING_OK | HAT_LOADCACHING_OK)) { 10907c478bd9Sstevel@tonic-gate PTE_SET(pte, PT_NOCACHE); 10917c478bd9Sstevel@tonic-gate if (x86_feature & X86_PAT) 10927c478bd9Sstevel@tonic-gate PTE_SET(pte, (level == 0) ? PT_PAT_4K : PT_PAT_LARGE); 10937c478bd9Sstevel@tonic-gate else 10947c478bd9Sstevel@tonic-gate PTE_SET(pte, PT_WRITETHRU); 10957c478bd9Sstevel@tonic-gate } else { 10967c478bd9Sstevel@tonic-gate panic("hati_mkpte(): bad caching attributes: %x\n", cache_attr); 10977c478bd9Sstevel@tonic-gate } 10987c478bd9Sstevel@tonic-gate 10997c478bd9Sstevel@tonic-gate return (pte); 11007c478bd9Sstevel@tonic-gate } 11017c478bd9Sstevel@tonic-gate 11027c478bd9Sstevel@tonic-gate /* 11037c478bd9Sstevel@tonic-gate * Duplicate address translations of the parent to the child. 11047c478bd9Sstevel@tonic-gate * This function really isn't used anymore. 11057c478bd9Sstevel@tonic-gate */ 11067c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 11077c478bd9Sstevel@tonic-gate int 11087c478bd9Sstevel@tonic-gate hat_dup(hat_t *old, hat_t *new, caddr_t addr, size_t len, uint_t flag) 11097c478bd9Sstevel@tonic-gate { 11107c478bd9Sstevel@tonic-gate ASSERT((uintptr_t)addr < kernelbase); 11117c478bd9Sstevel@tonic-gate ASSERT(new != kas.a_hat); 11127c478bd9Sstevel@tonic-gate ASSERT(old != kas.a_hat); 11137c478bd9Sstevel@tonic-gate return (0); 11147c478bd9Sstevel@tonic-gate } 11157c478bd9Sstevel@tonic-gate 11167c478bd9Sstevel@tonic-gate /* 11177c478bd9Sstevel@tonic-gate * Allocate any hat resources required for a process being swapped in. 11187c478bd9Sstevel@tonic-gate */ 11197c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 11207c478bd9Sstevel@tonic-gate void 11217c478bd9Sstevel@tonic-gate hat_swapin(hat_t *hat) 11227c478bd9Sstevel@tonic-gate { 11237c478bd9Sstevel@tonic-gate /* do nothing - we let everything fault back in */ 11247c478bd9Sstevel@tonic-gate } 11257c478bd9Sstevel@tonic-gate 11267c478bd9Sstevel@tonic-gate /* 11277c478bd9Sstevel@tonic-gate * Unload all translations associated with an address space of a process 11287c478bd9Sstevel@tonic-gate * that is being swapped out. 11297c478bd9Sstevel@tonic-gate */ 11307c478bd9Sstevel@tonic-gate void 11317c478bd9Sstevel@tonic-gate hat_swapout(hat_t *hat) 11327c478bd9Sstevel@tonic-gate { 11337c478bd9Sstevel@tonic-gate uintptr_t vaddr = (uintptr_t)0; 11347c478bd9Sstevel@tonic-gate uintptr_t eaddr = _userlimit; 11357c478bd9Sstevel@tonic-gate htable_t *ht = NULL; 11367c478bd9Sstevel@tonic-gate level_t l; 11377c478bd9Sstevel@tonic-gate 1138843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 11397c478bd9Sstevel@tonic-gate /* 11407c478bd9Sstevel@tonic-gate * We can't just call hat_unload(hat, 0, _userlimit...) here, because 11417c478bd9Sstevel@tonic-gate * seg_spt and shared pagetables can't be swapped out. 11427c478bd9Sstevel@tonic-gate * Take a look at segspt_shmswapout() - it's a big no-op. 11437c478bd9Sstevel@tonic-gate * 11447c478bd9Sstevel@tonic-gate * Instead we'll walk through all the address space and unload 11457c478bd9Sstevel@tonic-gate * any mappings which we are sure are not shared, not locked. 11467c478bd9Sstevel@tonic-gate */ 11477c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(vaddr)); 11487c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(eaddr)); 11497c478bd9Sstevel@tonic-gate ASSERT(AS_LOCK_HELD(hat->hat_as, &hat->hat_as->a_lock)); 11507c478bd9Sstevel@tonic-gate if ((uintptr_t)hat->hat_as->a_userlimit < eaddr) 11517c478bd9Sstevel@tonic-gate eaddr = (uintptr_t)hat->hat_as->a_userlimit; 11527c478bd9Sstevel@tonic-gate 11537c478bd9Sstevel@tonic-gate while (vaddr < eaddr) { 11547c478bd9Sstevel@tonic-gate (void) htable_walk(hat, &ht, &vaddr, eaddr); 11557c478bd9Sstevel@tonic-gate if (ht == NULL) 11567c478bd9Sstevel@tonic-gate break; 11577c478bd9Sstevel@tonic-gate 11587c478bd9Sstevel@tonic-gate ASSERT(!IN_VA_HOLE(vaddr)); 11597c478bd9Sstevel@tonic-gate 11607c478bd9Sstevel@tonic-gate /* 11617c478bd9Sstevel@tonic-gate * If the page table is shared skip its entire range. 11627c478bd9Sstevel@tonic-gate */ 11637c478bd9Sstevel@tonic-gate l = ht->ht_level; 11647c478bd9Sstevel@tonic-gate if (ht->ht_flags & HTABLE_SHARED_PFN) { 1165ae320ee6Speterte vaddr = ht->ht_vaddr + LEVEL_SIZE(l + 1); 11667c478bd9Sstevel@tonic-gate htable_release(ht); 11677c478bd9Sstevel@tonic-gate ht = NULL; 11687c478bd9Sstevel@tonic-gate continue; 11697c478bd9Sstevel@tonic-gate } 11707c478bd9Sstevel@tonic-gate 11717c478bd9Sstevel@tonic-gate /* 11727c478bd9Sstevel@tonic-gate * If the page table has no locked entries, unload this one. 11737c478bd9Sstevel@tonic-gate */ 11747c478bd9Sstevel@tonic-gate if (ht->ht_lock_cnt == 0) 11757c478bd9Sstevel@tonic-gate hat_unload(hat, (caddr_t)vaddr, LEVEL_SIZE(l), 11767c478bd9Sstevel@tonic-gate HAT_UNLOAD_UNMAP); 11777c478bd9Sstevel@tonic-gate 11787c478bd9Sstevel@tonic-gate /* 11797c478bd9Sstevel@tonic-gate * If we have a level 0 page table with locked entries, 11807c478bd9Sstevel@tonic-gate * skip the entire page table, otherwise skip just one entry. 11817c478bd9Sstevel@tonic-gate */ 11827c478bd9Sstevel@tonic-gate if (ht->ht_lock_cnt > 0 && l == 0) 11837c478bd9Sstevel@tonic-gate vaddr = ht->ht_vaddr + LEVEL_SIZE(1); 11847c478bd9Sstevel@tonic-gate else 11857c478bd9Sstevel@tonic-gate vaddr += LEVEL_SIZE(l); 11867c478bd9Sstevel@tonic-gate } 11877c478bd9Sstevel@tonic-gate if (ht) 11887c478bd9Sstevel@tonic-gate htable_release(ht); 11897c478bd9Sstevel@tonic-gate 11907c478bd9Sstevel@tonic-gate /* 11917c478bd9Sstevel@tonic-gate * We're in swapout because the system is low on memory, so 11927c478bd9Sstevel@tonic-gate * go back and flush all the htables off the cached list. 11937c478bd9Sstevel@tonic-gate */ 11947c478bd9Sstevel@tonic-gate htable_purge_hat(hat); 1195843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 11967c478bd9Sstevel@tonic-gate } 11977c478bd9Sstevel@tonic-gate 11987c478bd9Sstevel@tonic-gate /* 11997c478bd9Sstevel@tonic-gate * returns number of bytes that have valid mappings in hat. 12007c478bd9Sstevel@tonic-gate */ 12017c478bd9Sstevel@tonic-gate size_t 12027c478bd9Sstevel@tonic-gate hat_get_mapped_size(hat_t *hat) 12037c478bd9Sstevel@tonic-gate { 12047c478bd9Sstevel@tonic-gate size_t total = 0; 12057c478bd9Sstevel@tonic-gate int l; 12067c478bd9Sstevel@tonic-gate 12077c478bd9Sstevel@tonic-gate for (l = 0; l <= mmu.max_page_level; l++) 12087c478bd9Sstevel@tonic-gate total += (hat->hat_pages_mapped[l] << LEVEL_SHIFT(l)); 1209250b7ff9Sjosephb total += hat->hat_ism_pgcnt; 12107c478bd9Sstevel@tonic-gate 12117c478bd9Sstevel@tonic-gate return (total); 12127c478bd9Sstevel@tonic-gate } 12137c478bd9Sstevel@tonic-gate 12147c478bd9Sstevel@tonic-gate /* 12157c478bd9Sstevel@tonic-gate * enable/disable collection of stats for hat. 12167c478bd9Sstevel@tonic-gate */ 12177c478bd9Sstevel@tonic-gate int 12187c478bd9Sstevel@tonic-gate hat_stats_enable(hat_t *hat) 12197c478bd9Sstevel@tonic-gate { 12207c478bd9Sstevel@tonic-gate atomic_add_32(&hat->hat_stats, 1); 12217c478bd9Sstevel@tonic-gate return (1); 12227c478bd9Sstevel@tonic-gate } 12237c478bd9Sstevel@tonic-gate 12247c478bd9Sstevel@tonic-gate void 12257c478bd9Sstevel@tonic-gate hat_stats_disable(hat_t *hat) 12267c478bd9Sstevel@tonic-gate { 12277c478bd9Sstevel@tonic-gate atomic_add_32(&hat->hat_stats, -1); 12287c478bd9Sstevel@tonic-gate } 12297c478bd9Sstevel@tonic-gate 12307c478bd9Sstevel@tonic-gate /* 12317c478bd9Sstevel@tonic-gate * Utility to sync the ref/mod bits from a page table entry to the page_t 12327c478bd9Sstevel@tonic-gate * We must be holding the mapping list lock when this is called. 12337c478bd9Sstevel@tonic-gate */ 12347c478bd9Sstevel@tonic-gate static void 12357c478bd9Sstevel@tonic-gate hati_sync_pte_to_page(page_t *pp, x86pte_t pte, level_t level) 12367c478bd9Sstevel@tonic-gate { 12377c478bd9Sstevel@tonic-gate uint_t rm = 0; 12387c478bd9Sstevel@tonic-gate pgcnt_t pgcnt; 12397c478bd9Sstevel@tonic-gate 1240ae115bc7Smrj if (PTE_GET(pte, PT_SOFTWARE) >= PT_NOSYNC) 12417c478bd9Sstevel@tonic-gate return; 12427c478bd9Sstevel@tonic-gate 12437c478bd9Sstevel@tonic-gate if (PTE_GET(pte, PT_REF)) 12447c478bd9Sstevel@tonic-gate rm |= P_REF; 12457c478bd9Sstevel@tonic-gate 12467c478bd9Sstevel@tonic-gate if (PTE_GET(pte, PT_MOD)) 12477c478bd9Sstevel@tonic-gate rm |= P_MOD; 12487c478bd9Sstevel@tonic-gate 12497c478bd9Sstevel@tonic-gate if (rm == 0) 12507c478bd9Sstevel@tonic-gate return; 12517c478bd9Sstevel@tonic-gate 12527c478bd9Sstevel@tonic-gate /* 12537c478bd9Sstevel@tonic-gate * sync to all constituent pages of a large page 12547c478bd9Sstevel@tonic-gate */ 12557c478bd9Sstevel@tonic-gate ASSERT(x86_hm_held(pp)); 12567c478bd9Sstevel@tonic-gate pgcnt = page_get_pagecnt(level); 12577c478bd9Sstevel@tonic-gate ASSERT(IS_P2ALIGNED(pp->p_pagenum, pgcnt)); 12587c478bd9Sstevel@tonic-gate for (; pgcnt > 0; --pgcnt) { 12597c478bd9Sstevel@tonic-gate /* 12607c478bd9Sstevel@tonic-gate * hat_page_demote() can't decrease 12617c478bd9Sstevel@tonic-gate * pszc below this mapping size 12627c478bd9Sstevel@tonic-gate * since this large mapping existed after we 12637c478bd9Sstevel@tonic-gate * took mlist lock. 12647c478bd9Sstevel@tonic-gate */ 12657c478bd9Sstevel@tonic-gate ASSERT(pp->p_szc >= level); 12667c478bd9Sstevel@tonic-gate hat_page_setattr(pp, rm); 12677c478bd9Sstevel@tonic-gate ++pp; 12687c478bd9Sstevel@tonic-gate } 12697c478bd9Sstevel@tonic-gate } 12707c478bd9Sstevel@tonic-gate 12717c478bd9Sstevel@tonic-gate /* 12727c478bd9Sstevel@tonic-gate * This the set of PTE bits for PFN, permissions and caching 1273843e1988Sjohnlev * that are allowed to change on a HAT_LOAD_REMAP 12747c478bd9Sstevel@tonic-gate */ 12757c478bd9Sstevel@tonic-gate #define PT_REMAP_BITS \ 12767c478bd9Sstevel@tonic-gate (PT_PADDR | PT_NX | PT_WRITABLE | PT_WRITETHRU | \ 1277843e1988Sjohnlev PT_NOCACHE | PT_PAT_4K | PT_PAT_LARGE | PT_IGNORE | PT_REF | PT_MOD) 12787c478bd9Sstevel@tonic-gate 1279b193e412Skchow #define REMAPASSERT(EX) if (!(EX)) panic("hati_pte_map: " #EX) 12807c478bd9Sstevel@tonic-gate /* 12817c478bd9Sstevel@tonic-gate * Do the low-level work to get a mapping entered into a HAT's pagetables 12827c478bd9Sstevel@tonic-gate * and in the mapping list of the associated page_t. 12837c478bd9Sstevel@tonic-gate */ 1284ae115bc7Smrj static int 12857c478bd9Sstevel@tonic-gate hati_pte_map( 12867c478bd9Sstevel@tonic-gate htable_t *ht, 12877c478bd9Sstevel@tonic-gate uint_t entry, 12887c478bd9Sstevel@tonic-gate page_t *pp, 12897c478bd9Sstevel@tonic-gate x86pte_t pte, 12907c478bd9Sstevel@tonic-gate int flags, 12917c478bd9Sstevel@tonic-gate void *pte_ptr) 12927c478bd9Sstevel@tonic-gate { 12937c478bd9Sstevel@tonic-gate hat_t *hat = ht->ht_hat; 12947c478bd9Sstevel@tonic-gate x86pte_t old_pte; 12957c478bd9Sstevel@tonic-gate level_t l = ht->ht_level; 12967c478bd9Sstevel@tonic-gate hment_t *hm; 12977c478bd9Sstevel@tonic-gate uint_t is_consist; 129825540de2SJakub Jermar uint_t is_locked; 1299ae115bc7Smrj int rv = 0; 13007c478bd9Sstevel@tonic-gate 13017c478bd9Sstevel@tonic-gate /* 1302*21584dbcSPavel Tatashin * Is this a consistent (ie. need mapping list lock) mapping? 13037c478bd9Sstevel@tonic-gate */ 13047c478bd9Sstevel@tonic-gate is_consist = (pp != NULL && (flags & HAT_LOAD_NOCONSIST) == 0); 13057c478bd9Sstevel@tonic-gate 13067c478bd9Sstevel@tonic-gate /* 13077c478bd9Sstevel@tonic-gate * Track locked mapping count in the htable. Do this first, 13087c478bd9Sstevel@tonic-gate * as we track locking even if there already is a mapping present. 13097c478bd9Sstevel@tonic-gate */ 131025540de2SJakub Jermar is_locked = (flags & HAT_LOAD_LOCK) != 0 && hat != kas.a_hat; 131125540de2SJakub Jermar if (is_locked) 13127c478bd9Sstevel@tonic-gate HTABLE_LOCK_INC(ht); 13137c478bd9Sstevel@tonic-gate 13147c478bd9Sstevel@tonic-gate /* 13157c478bd9Sstevel@tonic-gate * Acquire the page's mapping list lock and get an hment to use. 13167c478bd9Sstevel@tonic-gate * Note that hment_prepare() might return NULL. 13177c478bd9Sstevel@tonic-gate */ 13187c478bd9Sstevel@tonic-gate if (is_consist) { 13197c478bd9Sstevel@tonic-gate x86_hm_enter(pp); 13207c478bd9Sstevel@tonic-gate hm = hment_prepare(ht, entry, pp); 13217c478bd9Sstevel@tonic-gate } 13227c478bd9Sstevel@tonic-gate 13237c478bd9Sstevel@tonic-gate /* 13247c478bd9Sstevel@tonic-gate * Set the new pte, retrieving the old one at the same time. 13257c478bd9Sstevel@tonic-gate */ 13267c478bd9Sstevel@tonic-gate old_pte = x86pte_set(ht, entry, pte, pte_ptr); 13277c478bd9Sstevel@tonic-gate 13287c478bd9Sstevel@tonic-gate /* 132925540de2SJakub Jermar * Did we get a large page / page table collision? 13307c478bd9Sstevel@tonic-gate */ 1331ae115bc7Smrj if (old_pte == LPAGE_ERROR) { 133225540de2SJakub Jermar if (is_locked) 133325540de2SJakub Jermar HTABLE_LOCK_DEC(ht); 1334ae115bc7Smrj rv = -1; 1335ae115bc7Smrj goto done; 13367c478bd9Sstevel@tonic-gate } 13377c478bd9Sstevel@tonic-gate 1338ae115bc7Smrj /* 1339ae115bc7Smrj * If the mapping didn't change there is nothing more to do. 1340ae115bc7Smrj */ 1341ae115bc7Smrj if (PTE_EQUIV(pte, old_pte)) 1342ae115bc7Smrj goto done; 1343ae115bc7Smrj 13447c478bd9Sstevel@tonic-gate /* 13457c478bd9Sstevel@tonic-gate * Install a new mapping in the page's mapping list 13467c478bd9Sstevel@tonic-gate */ 13477c478bd9Sstevel@tonic-gate if (!PTE_ISVALID(old_pte)) { 13487c478bd9Sstevel@tonic-gate if (is_consist) { 13497c478bd9Sstevel@tonic-gate hment_assign(ht, entry, pp, hm); 13507c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 13517c478bd9Sstevel@tonic-gate } else { 13527c478bd9Sstevel@tonic-gate ASSERT(flags & HAT_LOAD_NOCONSIST); 13537c478bd9Sstevel@tonic-gate } 135402bc52beSkchow #if defined(__amd64) 135502bc52beSkchow if (ht->ht_flags & HTABLE_VLP) { 135602bc52beSkchow cpu_t *cpu = CPU; 135702bc52beSkchow x86pte_t *vlpptep = cpu->cpu_hat_info->hci_vlp_l2ptes; 135802bc52beSkchow VLP_COPY(hat->hat_vlp_ptes, vlpptep); 135902bc52beSkchow } 136002bc52beSkchow #endif 13617c478bd9Sstevel@tonic-gate HTABLE_INC(ht->ht_valid_cnt); 13627c478bd9Sstevel@tonic-gate PGCNT_INC(hat, l); 1363ae115bc7Smrj return (rv); 13647c478bd9Sstevel@tonic-gate } 13657c478bd9Sstevel@tonic-gate 13667c478bd9Sstevel@tonic-gate /* 13677c478bd9Sstevel@tonic-gate * Remap's are more complicated: 13687c478bd9Sstevel@tonic-gate * - HAT_LOAD_REMAP must be specified if changing the pfn. 13697c478bd9Sstevel@tonic-gate * We also require that NOCONSIST be specified. 13707c478bd9Sstevel@tonic-gate * - Otherwise only permission or caching bits may change. 13717c478bd9Sstevel@tonic-gate */ 13727c478bd9Sstevel@tonic-gate if (!PTE_ISPAGE(old_pte, l)) 13737c478bd9Sstevel@tonic-gate panic("non-null/page mapping pte=" FMT_PTE, old_pte); 13747c478bd9Sstevel@tonic-gate 13757c478bd9Sstevel@tonic-gate if (PTE2PFN(old_pte, l) != PTE2PFN(pte, l)) { 1376b193e412Skchow REMAPASSERT(flags & HAT_LOAD_REMAP); 1377b193e412Skchow REMAPASSERT(flags & HAT_LOAD_NOCONSIST); 1378ae115bc7Smrj REMAPASSERT(PTE_GET(old_pte, PT_SOFTWARE) >= PT_NOCONSIST); 1379b193e412Skchow REMAPASSERT(pf_is_memory(PTE2PFN(old_pte, l)) == 13807c478bd9Sstevel@tonic-gate pf_is_memory(PTE2PFN(pte, l))); 1381b193e412Skchow REMAPASSERT(!is_consist); 13827c478bd9Sstevel@tonic-gate } 13837c478bd9Sstevel@tonic-gate 13847c478bd9Sstevel@tonic-gate /* 1385843e1988Sjohnlev * We only let remaps change the certain bits in the PTE. 13867c478bd9Sstevel@tonic-gate */ 1387843e1988Sjohnlev if (PTE_GET(old_pte, ~PT_REMAP_BITS) != PTE_GET(pte, ~PT_REMAP_BITS)) 1388843e1988Sjohnlev panic("remap bits changed: old_pte="FMT_PTE", pte="FMT_PTE"\n", 1389843e1988Sjohnlev old_pte, pte); 13907c478bd9Sstevel@tonic-gate 13917c478bd9Sstevel@tonic-gate /* 13927c478bd9Sstevel@tonic-gate * We don't create any mapping list entries on a remap, so release 13937c478bd9Sstevel@tonic-gate * any allocated hment after we drop the mapping list lock. 13947c478bd9Sstevel@tonic-gate */ 1395ae115bc7Smrj done: 13967c478bd9Sstevel@tonic-gate if (is_consist) { 13977c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 13987c478bd9Sstevel@tonic-gate if (hm != NULL) 13997c478bd9Sstevel@tonic-gate hment_free(hm); 14007c478bd9Sstevel@tonic-gate } 1401ae115bc7Smrj return (rv); 14027c478bd9Sstevel@tonic-gate } 14037c478bd9Sstevel@tonic-gate 14047c478bd9Sstevel@tonic-gate /* 1405ae115bc7Smrj * Internal routine to load a single page table entry. This only fails if 1406ae115bc7Smrj * we attempt to overwrite a page table link with a large page. 14077c478bd9Sstevel@tonic-gate */ 1408ae115bc7Smrj static int 14097c478bd9Sstevel@tonic-gate hati_load_common( 14107c478bd9Sstevel@tonic-gate hat_t *hat, 14117c478bd9Sstevel@tonic-gate uintptr_t va, 14127c478bd9Sstevel@tonic-gate page_t *pp, 14137c478bd9Sstevel@tonic-gate uint_t attr, 14147c478bd9Sstevel@tonic-gate uint_t flags, 14157c478bd9Sstevel@tonic-gate level_t level, 14167c478bd9Sstevel@tonic-gate pfn_t pfn) 14177c478bd9Sstevel@tonic-gate { 14187c478bd9Sstevel@tonic-gate htable_t *ht; 14197c478bd9Sstevel@tonic-gate uint_t entry; 14207c478bd9Sstevel@tonic-gate x86pte_t pte; 1421ae115bc7Smrj int rv = 0; 14227c478bd9Sstevel@tonic-gate 1423aac11643Sjosephb /* 1424aac11643Sjosephb * The number 16 is arbitrary and here to catch a recursion problem 1425aac11643Sjosephb * early before we blow out the kernel stack. 1426aac11643Sjosephb */ 1427aac11643Sjosephb ++curthread->t_hatdepth; 1428aac11643Sjosephb ASSERT(curthread->t_hatdepth < 16); 1429aac11643Sjosephb 14307c478bd9Sstevel@tonic-gate ASSERT(hat == kas.a_hat || 14317c478bd9Sstevel@tonic-gate AS_LOCK_HELD(hat->hat_as, &hat->hat_as->a_lock)); 14327c478bd9Sstevel@tonic-gate 14337c478bd9Sstevel@tonic-gate if (flags & HAT_LOAD_SHARE) 14347c478bd9Sstevel@tonic-gate hat->hat_flags |= HAT_SHARED; 14357c478bd9Sstevel@tonic-gate 14367c478bd9Sstevel@tonic-gate /* 14377c478bd9Sstevel@tonic-gate * Find the page table that maps this page if it already exists. 14387c478bd9Sstevel@tonic-gate */ 14397c478bd9Sstevel@tonic-gate ht = htable_lookup(hat, va, level); 14407c478bd9Sstevel@tonic-gate 14417c478bd9Sstevel@tonic-gate /* 1442aac11643Sjosephb * We must have HAT_LOAD_NOCONSIST if page_t is NULL. 14437c478bd9Sstevel@tonic-gate */ 1444aac11643Sjosephb if (pp == NULL) 14457c478bd9Sstevel@tonic-gate flags |= HAT_LOAD_NOCONSIST; 14467c478bd9Sstevel@tonic-gate 14477c478bd9Sstevel@tonic-gate if (ht == NULL) { 14487c478bd9Sstevel@tonic-gate ht = htable_create(hat, va, level, NULL); 14497c478bd9Sstevel@tonic-gate ASSERT(ht != NULL); 14507c478bd9Sstevel@tonic-gate } 14517c478bd9Sstevel@tonic-gate entry = htable_va2entry(va, ht); 14527c478bd9Sstevel@tonic-gate 14537c478bd9Sstevel@tonic-gate /* 14547c478bd9Sstevel@tonic-gate * a bunch of paranoid error checking 14557c478bd9Sstevel@tonic-gate */ 14567c478bd9Sstevel@tonic-gate ASSERT(ht->ht_busy > 0); 14577c478bd9Sstevel@tonic-gate if (ht->ht_vaddr > va || va > HTABLE_LAST_PAGE(ht)) 1458903a11ebSrh panic("hati_load_common: bad htable %p, va %p", 1459903a11ebSrh (void *)ht, (void *)va); 14607c478bd9Sstevel@tonic-gate ASSERT(ht->ht_level == level); 14617c478bd9Sstevel@tonic-gate 14627c478bd9Sstevel@tonic-gate /* 14637c478bd9Sstevel@tonic-gate * construct the new PTE 14647c478bd9Sstevel@tonic-gate */ 14657c478bd9Sstevel@tonic-gate if (hat == kas.a_hat) 14667c478bd9Sstevel@tonic-gate attr &= ~PROT_USER; 14677c478bd9Sstevel@tonic-gate pte = hati_mkpte(pfn, attr, level, flags); 14687c478bd9Sstevel@tonic-gate if (hat == kas.a_hat && va >= kernelbase) 14697c478bd9Sstevel@tonic-gate PTE_SET(pte, mmu.pt_global); 14707c478bd9Sstevel@tonic-gate 14717c478bd9Sstevel@tonic-gate /* 14727c478bd9Sstevel@tonic-gate * establish the mapping 14737c478bd9Sstevel@tonic-gate */ 1474ae115bc7Smrj rv = hati_pte_map(ht, entry, pp, pte, flags, NULL); 14757c478bd9Sstevel@tonic-gate 14767c478bd9Sstevel@tonic-gate /* 14777c478bd9Sstevel@tonic-gate * release the htable and any reserves 14787c478bd9Sstevel@tonic-gate */ 14797c478bd9Sstevel@tonic-gate htable_release(ht); 1480aac11643Sjosephb --curthread->t_hatdepth; 1481ae115bc7Smrj return (rv); 14827c478bd9Sstevel@tonic-gate } 14837c478bd9Sstevel@tonic-gate 14847c478bd9Sstevel@tonic-gate /* 14857c478bd9Sstevel@tonic-gate * special case of hat_memload to deal with some kernel addrs for performance 14867c478bd9Sstevel@tonic-gate */ 14877c478bd9Sstevel@tonic-gate static void 14887c478bd9Sstevel@tonic-gate hat_kmap_load( 14897c478bd9Sstevel@tonic-gate caddr_t addr, 14907c478bd9Sstevel@tonic-gate page_t *pp, 14917c478bd9Sstevel@tonic-gate uint_t attr, 14927c478bd9Sstevel@tonic-gate uint_t flags) 14937c478bd9Sstevel@tonic-gate { 14947c478bd9Sstevel@tonic-gate uintptr_t va = (uintptr_t)addr; 14957c478bd9Sstevel@tonic-gate x86pte_t pte; 14967c478bd9Sstevel@tonic-gate pfn_t pfn = page_pptonum(pp); 14977c478bd9Sstevel@tonic-gate pgcnt_t pg_off = mmu_btop(va - mmu.kmap_addr); 14987c478bd9Sstevel@tonic-gate htable_t *ht; 14997c478bd9Sstevel@tonic-gate uint_t entry; 15007c478bd9Sstevel@tonic-gate void *pte_ptr; 15017c478bd9Sstevel@tonic-gate 15027c478bd9Sstevel@tonic-gate /* 15037c478bd9Sstevel@tonic-gate * construct the requested PTE 15047c478bd9Sstevel@tonic-gate */ 15057c478bd9Sstevel@tonic-gate attr &= ~PROT_USER; 15067c478bd9Sstevel@tonic-gate attr |= HAT_STORECACHING_OK; 15077c478bd9Sstevel@tonic-gate pte = hati_mkpte(pfn, attr, 0, flags); 15087c478bd9Sstevel@tonic-gate PTE_SET(pte, mmu.pt_global); 15097c478bd9Sstevel@tonic-gate 15107c478bd9Sstevel@tonic-gate /* 15117c478bd9Sstevel@tonic-gate * Figure out the pte_ptr and htable and use common code to finish up 15127c478bd9Sstevel@tonic-gate */ 15137c478bd9Sstevel@tonic-gate if (mmu.pae_hat) 15147c478bd9Sstevel@tonic-gate pte_ptr = mmu.kmap_ptes + pg_off; 15157c478bd9Sstevel@tonic-gate else 15167c478bd9Sstevel@tonic-gate pte_ptr = (x86pte32_t *)mmu.kmap_ptes + pg_off; 15177c478bd9Sstevel@tonic-gate ht = mmu.kmap_htables[(va - mmu.kmap_htables[0]->ht_vaddr) >> 15187c478bd9Sstevel@tonic-gate LEVEL_SHIFT(1)]; 15197c478bd9Sstevel@tonic-gate entry = htable_va2entry(va, ht); 1520aac11643Sjosephb ++curthread->t_hatdepth; 1521aac11643Sjosephb ASSERT(curthread->t_hatdepth < 16); 1522ae115bc7Smrj (void) hati_pte_map(ht, entry, pp, pte, flags, pte_ptr); 1523aac11643Sjosephb --curthread->t_hatdepth; 15247c478bd9Sstevel@tonic-gate } 15257c478bd9Sstevel@tonic-gate 15267c478bd9Sstevel@tonic-gate /* 15277c478bd9Sstevel@tonic-gate * hat_memload() - load a translation to the given page struct 15287c478bd9Sstevel@tonic-gate * 15297c478bd9Sstevel@tonic-gate * Flags for hat_memload/hat_devload/hat_*attr. 15307c478bd9Sstevel@tonic-gate * 15317c478bd9Sstevel@tonic-gate * HAT_LOAD Default flags to load a translation to the page. 15327c478bd9Sstevel@tonic-gate * 15337c478bd9Sstevel@tonic-gate * HAT_LOAD_LOCK Lock down mapping resources; hat_map(), hat_memload(), 15347c478bd9Sstevel@tonic-gate * and hat_devload(). 15357c478bd9Sstevel@tonic-gate * 15367c478bd9Sstevel@tonic-gate * HAT_LOAD_NOCONSIST Do not add mapping to page_t mapping list. 1537ae115bc7Smrj * sets PT_NOCONSIST 15387c478bd9Sstevel@tonic-gate * 15397c478bd9Sstevel@tonic-gate * HAT_LOAD_SHARE A flag to hat_memload() to indicate h/w page tables 15407c478bd9Sstevel@tonic-gate * that map some user pages (not kas) is shared by more 15417c478bd9Sstevel@tonic-gate * than one process (eg. ISM). 15427c478bd9Sstevel@tonic-gate * 15437c478bd9Sstevel@tonic-gate * HAT_LOAD_REMAP Reload a valid pte with a different page frame. 15447c478bd9Sstevel@tonic-gate * 15457c478bd9Sstevel@tonic-gate * HAT_NO_KALLOC Do not kmem_alloc while creating the mapping; at this 15467c478bd9Sstevel@tonic-gate * point, it's setting up mapping to allocate internal 15477c478bd9Sstevel@tonic-gate * hat layer data structures. This flag forces hat layer 15487c478bd9Sstevel@tonic-gate * to tap its reserves in order to prevent infinite 15497c478bd9Sstevel@tonic-gate * recursion. 15507c478bd9Sstevel@tonic-gate * 15517c478bd9Sstevel@tonic-gate * The following is a protection attribute (like PROT_READ, etc.) 15527c478bd9Sstevel@tonic-gate * 1553ae115bc7Smrj * HAT_NOSYNC set PT_NOSYNC - this mapping's ref/mod bits 15547c478bd9Sstevel@tonic-gate * are never cleared. 15557c478bd9Sstevel@tonic-gate * 15567c478bd9Sstevel@tonic-gate * Installing new valid PTE's and creation of the mapping list 15577c478bd9Sstevel@tonic-gate * entry are controlled under the same lock. It's derived from the 15587c478bd9Sstevel@tonic-gate * page_t being mapped. 15597c478bd9Sstevel@tonic-gate */ 15607c478bd9Sstevel@tonic-gate static uint_t supported_memload_flags = 15617c478bd9Sstevel@tonic-gate HAT_LOAD | HAT_LOAD_LOCK | HAT_LOAD_ADV | HAT_LOAD_NOCONSIST | 15627c478bd9Sstevel@tonic-gate HAT_LOAD_SHARE | HAT_NO_KALLOC | HAT_LOAD_REMAP | HAT_LOAD_TEXT; 15637c478bd9Sstevel@tonic-gate 15647c478bd9Sstevel@tonic-gate void 15657c478bd9Sstevel@tonic-gate hat_memload( 15667c478bd9Sstevel@tonic-gate hat_t *hat, 15677c478bd9Sstevel@tonic-gate caddr_t addr, 15687c478bd9Sstevel@tonic-gate page_t *pp, 15697c478bd9Sstevel@tonic-gate uint_t attr, 15707c478bd9Sstevel@tonic-gate uint_t flags) 15717c478bd9Sstevel@tonic-gate { 15727c478bd9Sstevel@tonic-gate uintptr_t va = (uintptr_t)addr; 15737c478bd9Sstevel@tonic-gate level_t level = 0; 15747c478bd9Sstevel@tonic-gate pfn_t pfn = page_pptonum(pp); 15757c478bd9Sstevel@tonic-gate 1576843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 15777c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(va)); 1578ae115bc7Smrj ASSERT(hat == kas.a_hat || va < _userlimit); 15797c478bd9Sstevel@tonic-gate ASSERT(hat == kas.a_hat || 15807c478bd9Sstevel@tonic-gate AS_LOCK_HELD(hat->hat_as, &hat->hat_as->a_lock)); 15817c478bd9Sstevel@tonic-gate ASSERT((flags & supported_memload_flags) == flags); 15827c478bd9Sstevel@tonic-gate 15837c478bd9Sstevel@tonic-gate ASSERT(!IN_VA_HOLE(va)); 15847c478bd9Sstevel@tonic-gate ASSERT(!PP_ISFREE(pp)); 15857c478bd9Sstevel@tonic-gate 15867c478bd9Sstevel@tonic-gate /* 15877c478bd9Sstevel@tonic-gate * kernel address special case for performance. 15887c478bd9Sstevel@tonic-gate */ 15897c478bd9Sstevel@tonic-gate if (mmu.kmap_addr <= va && va < mmu.kmap_eaddr) { 15907c478bd9Sstevel@tonic-gate ASSERT(hat == kas.a_hat); 15917c478bd9Sstevel@tonic-gate hat_kmap_load(addr, pp, attr, flags); 1592843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 15937c478bd9Sstevel@tonic-gate return; 15947c478bd9Sstevel@tonic-gate } 15957c478bd9Sstevel@tonic-gate 15967c478bd9Sstevel@tonic-gate /* 15977c478bd9Sstevel@tonic-gate * This is used for memory with normal caching enabled, so 15987c478bd9Sstevel@tonic-gate * always set HAT_STORECACHING_OK. 15997c478bd9Sstevel@tonic-gate */ 16007c478bd9Sstevel@tonic-gate attr |= HAT_STORECACHING_OK; 1601ae115bc7Smrj if (hati_load_common(hat, va, pp, attr, flags, level, pfn) != 0) 1602ae115bc7Smrj panic("unexpected hati_load_common() failure"); 1603843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 16047c478bd9Sstevel@tonic-gate } 16057c478bd9Sstevel@tonic-gate 160605d3dc4bSpaulsan /* ARGSUSED */ 160705d3dc4bSpaulsan void 160805d3dc4bSpaulsan hat_memload_region(struct hat *hat, caddr_t addr, struct page *pp, 160905d3dc4bSpaulsan uint_t attr, uint_t flags, hat_region_cookie_t rcookie) 161005d3dc4bSpaulsan { 161105d3dc4bSpaulsan hat_memload(hat, addr, pp, attr, flags); 161205d3dc4bSpaulsan } 161305d3dc4bSpaulsan 16147c478bd9Sstevel@tonic-gate /* 16157c478bd9Sstevel@tonic-gate * Load the given array of page structs using large pages when possible 16167c478bd9Sstevel@tonic-gate */ 16177c478bd9Sstevel@tonic-gate void 16187c478bd9Sstevel@tonic-gate hat_memload_array( 16197c478bd9Sstevel@tonic-gate hat_t *hat, 16207c478bd9Sstevel@tonic-gate caddr_t addr, 16217c478bd9Sstevel@tonic-gate size_t len, 16227c478bd9Sstevel@tonic-gate page_t **pages, 16237c478bd9Sstevel@tonic-gate uint_t attr, 16247c478bd9Sstevel@tonic-gate uint_t flags) 16257c478bd9Sstevel@tonic-gate { 16267c478bd9Sstevel@tonic-gate uintptr_t va = (uintptr_t)addr; 16277c478bd9Sstevel@tonic-gate uintptr_t eaddr = va + len; 16287c478bd9Sstevel@tonic-gate level_t level; 16297c478bd9Sstevel@tonic-gate size_t pgsize; 16307c478bd9Sstevel@tonic-gate pgcnt_t pgindx = 0; 16317c478bd9Sstevel@tonic-gate pfn_t pfn; 16327c478bd9Sstevel@tonic-gate pgcnt_t i; 16337c478bd9Sstevel@tonic-gate 1634843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 16357c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(va)); 1636ae115bc7Smrj ASSERT(hat == kas.a_hat || va + len <= _userlimit); 16377c478bd9Sstevel@tonic-gate ASSERT(hat == kas.a_hat || 16387c478bd9Sstevel@tonic-gate AS_LOCK_HELD(hat->hat_as, &hat->hat_as->a_lock)); 16397c478bd9Sstevel@tonic-gate ASSERT((flags & supported_memload_flags) == flags); 16407c478bd9Sstevel@tonic-gate 16417c478bd9Sstevel@tonic-gate /* 16427c478bd9Sstevel@tonic-gate * memload is used for memory with full caching enabled, so 16437c478bd9Sstevel@tonic-gate * set HAT_STORECACHING_OK. 16447c478bd9Sstevel@tonic-gate */ 16457c478bd9Sstevel@tonic-gate attr |= HAT_STORECACHING_OK; 16467c478bd9Sstevel@tonic-gate 16477c478bd9Sstevel@tonic-gate /* 16487c478bd9Sstevel@tonic-gate * handle all pages using largest possible pagesize 16497c478bd9Sstevel@tonic-gate */ 16507c478bd9Sstevel@tonic-gate while (va < eaddr) { 16517c478bd9Sstevel@tonic-gate /* 16527c478bd9Sstevel@tonic-gate * decide what level mapping to use (ie. pagesize) 16537c478bd9Sstevel@tonic-gate */ 16547c478bd9Sstevel@tonic-gate pfn = page_pptonum(pages[pgindx]); 16557c478bd9Sstevel@tonic-gate for (level = mmu.max_page_level; ; --level) { 16567c478bd9Sstevel@tonic-gate pgsize = LEVEL_SIZE(level); 16577c478bd9Sstevel@tonic-gate if (level == 0) 16587c478bd9Sstevel@tonic-gate break; 1659ae115bc7Smrj 16607c478bd9Sstevel@tonic-gate if (!IS_P2ALIGNED(va, pgsize) || 16617c478bd9Sstevel@tonic-gate (eaddr - va) < pgsize || 1662ae115bc7Smrj !IS_P2ALIGNED(pfn_to_pa(pfn), pgsize)) 16637c478bd9Sstevel@tonic-gate continue; 16647c478bd9Sstevel@tonic-gate 16657c478bd9Sstevel@tonic-gate /* 16667c478bd9Sstevel@tonic-gate * To use a large mapping of this size, all the 16677c478bd9Sstevel@tonic-gate * pages we are passed must be sequential subpages 16687c478bd9Sstevel@tonic-gate * of the large page. 16697c478bd9Sstevel@tonic-gate * hat_page_demote() can't change p_szc because 16707c478bd9Sstevel@tonic-gate * all pages are locked. 16717c478bd9Sstevel@tonic-gate */ 16727c478bd9Sstevel@tonic-gate if (pages[pgindx]->p_szc >= level) { 16737c478bd9Sstevel@tonic-gate for (i = 0; i < mmu_btop(pgsize); ++i) { 16747c478bd9Sstevel@tonic-gate if (pfn + i != 16757c478bd9Sstevel@tonic-gate page_pptonum(pages[pgindx + i])) 16767c478bd9Sstevel@tonic-gate break; 16777c478bd9Sstevel@tonic-gate ASSERT(pages[pgindx + i]->p_szc >= 16787c478bd9Sstevel@tonic-gate level); 16797c478bd9Sstevel@tonic-gate ASSERT(pages[pgindx] + i == 16807c478bd9Sstevel@tonic-gate pages[pgindx + i]); 16817c478bd9Sstevel@tonic-gate } 168202bc52beSkchow if (i == mmu_btop(pgsize)) { 168302bc52beSkchow #ifdef DEBUG 168402bc52beSkchow if (level == 2) 168502bc52beSkchow map1gcnt++; 168602bc52beSkchow #endif 16877c478bd9Sstevel@tonic-gate break; 168802bc52beSkchow } 16897c478bd9Sstevel@tonic-gate } 16907c478bd9Sstevel@tonic-gate } 16917c478bd9Sstevel@tonic-gate 16927c478bd9Sstevel@tonic-gate /* 1693ae115bc7Smrj * Load this page mapping. If the load fails, try a smaller 1694ae115bc7Smrj * pagesize. 16957c478bd9Sstevel@tonic-gate */ 16967c478bd9Sstevel@tonic-gate ASSERT(!IN_VA_HOLE(va)); 1697ae115bc7Smrj while (hati_load_common(hat, va, pages[pgindx], attr, 1698250b7ff9Sjosephb flags, level, pfn) != 0) { 1699ae115bc7Smrj if (level == 0) 1700ae115bc7Smrj panic("unexpected hati_load_common() failure"); 1701ae115bc7Smrj --level; 1702ae115bc7Smrj pgsize = LEVEL_SIZE(level); 1703ae115bc7Smrj } 17047c478bd9Sstevel@tonic-gate 17057c478bd9Sstevel@tonic-gate /* 17067c478bd9Sstevel@tonic-gate * move to next page 17077c478bd9Sstevel@tonic-gate */ 17087c478bd9Sstevel@tonic-gate va += pgsize; 17097c478bd9Sstevel@tonic-gate pgindx += mmu_btop(pgsize); 17107c478bd9Sstevel@tonic-gate } 1711843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 17127c478bd9Sstevel@tonic-gate } 17137c478bd9Sstevel@tonic-gate 171405d3dc4bSpaulsan /* ARGSUSED */ 171505d3dc4bSpaulsan void 171605d3dc4bSpaulsan hat_memload_array_region(struct hat *hat, caddr_t addr, size_t len, 171705d3dc4bSpaulsan struct page **pps, uint_t attr, uint_t flags, 171805d3dc4bSpaulsan hat_region_cookie_t rcookie) 171905d3dc4bSpaulsan { 172005d3dc4bSpaulsan hat_memload_array(hat, addr, len, pps, attr, flags); 172105d3dc4bSpaulsan } 172205d3dc4bSpaulsan 17237c478bd9Sstevel@tonic-gate /* 17247c478bd9Sstevel@tonic-gate * void hat_devload(hat, addr, len, pf, attr, flags) 17257c478bd9Sstevel@tonic-gate * load/lock the given page frame number 17267c478bd9Sstevel@tonic-gate * 17277c478bd9Sstevel@tonic-gate * Advisory ordering attributes. Apply only to device mappings. 17287c478bd9Sstevel@tonic-gate * 17297c478bd9Sstevel@tonic-gate * HAT_STRICTORDER: the CPU must issue the references in order, as the 17307c478bd9Sstevel@tonic-gate * programmer specified. This is the default. 17317c478bd9Sstevel@tonic-gate * HAT_UNORDERED_OK: the CPU may reorder the references (this is all kinds 17327c478bd9Sstevel@tonic-gate * of reordering; store or load with store or load). 17337c478bd9Sstevel@tonic-gate * HAT_MERGING_OK: merging and batching: the CPU may merge individual stores 17347c478bd9Sstevel@tonic-gate * to consecutive locations (for example, turn two consecutive byte 17357c478bd9Sstevel@tonic-gate * stores into one halfword store), and it may batch individual loads 17367c478bd9Sstevel@tonic-gate * (for example, turn two consecutive byte loads into one halfword load). 17377c478bd9Sstevel@tonic-gate * This also implies re-ordering. 17387c478bd9Sstevel@tonic-gate * HAT_LOADCACHING_OK: the CPU may cache the data it fetches and reuse it 17397c478bd9Sstevel@tonic-gate * until another store occurs. The default is to fetch new data 17407c478bd9Sstevel@tonic-gate * on every load. This also implies merging. 17417c478bd9Sstevel@tonic-gate * HAT_STORECACHING_OK: the CPU may keep the data in the cache and push it to 17427c478bd9Sstevel@tonic-gate * the device (perhaps with other data) at a later time. The default is 17437c478bd9Sstevel@tonic-gate * to push the data right away. This also implies load caching. 17447c478bd9Sstevel@tonic-gate * 17457c478bd9Sstevel@tonic-gate * Equivalent of hat_memload(), but can be used for device memory where 17467c478bd9Sstevel@tonic-gate * there are no page_t's and we support additional flags (write merging, etc). 17477c478bd9Sstevel@tonic-gate * Note that we can have large page mappings with this interface. 17487c478bd9Sstevel@tonic-gate */ 17497c478bd9Sstevel@tonic-gate int supported_devload_flags = HAT_LOAD | HAT_LOAD_LOCK | 17507c478bd9Sstevel@tonic-gate HAT_LOAD_NOCONSIST | HAT_STRICTORDER | HAT_UNORDERED_OK | 17517c478bd9Sstevel@tonic-gate HAT_MERGING_OK | HAT_LOADCACHING_OK | HAT_STORECACHING_OK; 17527c478bd9Sstevel@tonic-gate 17537c478bd9Sstevel@tonic-gate void 17547c478bd9Sstevel@tonic-gate hat_devload( 17557c478bd9Sstevel@tonic-gate hat_t *hat, 17567c478bd9Sstevel@tonic-gate caddr_t addr, 17577c478bd9Sstevel@tonic-gate size_t len, 17587c478bd9Sstevel@tonic-gate pfn_t pfn, 17597c478bd9Sstevel@tonic-gate uint_t attr, 17607c478bd9Sstevel@tonic-gate int flags) 17617c478bd9Sstevel@tonic-gate { 17627c478bd9Sstevel@tonic-gate uintptr_t va = ALIGN2PAGE(addr); 17637c478bd9Sstevel@tonic-gate uintptr_t eva = va + len; 17647c478bd9Sstevel@tonic-gate level_t level; 17657c478bd9Sstevel@tonic-gate size_t pgsize; 17667c478bd9Sstevel@tonic-gate page_t *pp; 17677c478bd9Sstevel@tonic-gate int f; /* per PTE copy of flags - maybe modified */ 17687c478bd9Sstevel@tonic-gate uint_t a; /* per PTE copy of attr */ 17697c478bd9Sstevel@tonic-gate 1770843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 17717c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(va)); 1772ae115bc7Smrj ASSERT(hat == kas.a_hat || eva <= _userlimit); 17737c478bd9Sstevel@tonic-gate ASSERT(hat == kas.a_hat || 17747c478bd9Sstevel@tonic-gate AS_LOCK_HELD(hat->hat_as, &hat->hat_as->a_lock)); 17757c478bd9Sstevel@tonic-gate ASSERT((flags & supported_devload_flags) == flags); 17767c478bd9Sstevel@tonic-gate 17777c478bd9Sstevel@tonic-gate /* 17787c478bd9Sstevel@tonic-gate * handle all pages 17797c478bd9Sstevel@tonic-gate */ 17807c478bd9Sstevel@tonic-gate while (va < eva) { 17817c478bd9Sstevel@tonic-gate 17827c478bd9Sstevel@tonic-gate /* 17837c478bd9Sstevel@tonic-gate * decide what level mapping to use (ie. pagesize) 17847c478bd9Sstevel@tonic-gate */ 17857c478bd9Sstevel@tonic-gate for (level = mmu.max_page_level; ; --level) { 17867c478bd9Sstevel@tonic-gate pgsize = LEVEL_SIZE(level); 17877c478bd9Sstevel@tonic-gate if (level == 0) 17887c478bd9Sstevel@tonic-gate break; 17897c478bd9Sstevel@tonic-gate if (IS_P2ALIGNED(va, pgsize) && 17907c478bd9Sstevel@tonic-gate (eva - va) >= pgsize && 179102bc52beSkchow IS_P2ALIGNED(pfn, mmu_btop(pgsize))) { 179202bc52beSkchow #ifdef DEBUG 179302bc52beSkchow if (level == 2) 179402bc52beSkchow map1gcnt++; 179502bc52beSkchow #endif 17967c478bd9Sstevel@tonic-gate break; 179702bc52beSkchow } 17987c478bd9Sstevel@tonic-gate } 17997c478bd9Sstevel@tonic-gate 18007c478bd9Sstevel@tonic-gate /* 1801ae115bc7Smrj * If this is just memory then allow caching (this happens 18027c478bd9Sstevel@tonic-gate * for the nucleus pages) - though HAT_PLAT_NOCACHE can be used 1803ae115bc7Smrj * to override that. If we don't have a page_t then make sure 18047c478bd9Sstevel@tonic-gate * NOCONSIST is set. 18057c478bd9Sstevel@tonic-gate */ 18067c478bd9Sstevel@tonic-gate a = attr; 18077c478bd9Sstevel@tonic-gate f = flags; 1808843e1988Sjohnlev if (!pf_is_memory(pfn)) 1809843e1988Sjohnlev f |= HAT_LOAD_NOCONSIST; 1810843e1988Sjohnlev else if (!(a & HAT_PLAT_NOCACHE)) 1811843e1988Sjohnlev a |= HAT_STORECACHING_OK; 18127c478bd9Sstevel@tonic-gate 1813843e1988Sjohnlev if (f & HAT_LOAD_NOCONSIST) 18147c478bd9Sstevel@tonic-gate pp = NULL; 1815843e1988Sjohnlev else 1816843e1988Sjohnlev pp = page_numtopp_nolock(pfn); 18177c478bd9Sstevel@tonic-gate 1818e803c1e6SPrakash Sangappa /* 1819e803c1e6SPrakash Sangappa * Check to make sure we are really trying to map a valid 1820e803c1e6SPrakash Sangappa * memory page. The caller wishing to intentionally map 1821e803c1e6SPrakash Sangappa * free memory pages will have passed the HAT_LOAD_NOCONSIST 1822e803c1e6SPrakash Sangappa * flag, then pp will be NULL. 1823e803c1e6SPrakash Sangappa */ 1824e803c1e6SPrakash Sangappa if (pp != NULL) { 1825e803c1e6SPrakash Sangappa if (PP_ISFREE(pp)) { 1826e803c1e6SPrakash Sangappa panic("hat_devload: loading " 1827e803c1e6SPrakash Sangappa "a mapping to free page %p", (void *)pp); 1828e803c1e6SPrakash Sangappa } 1829e803c1e6SPrakash Sangappa 1830e803c1e6SPrakash Sangappa if (!PAGE_LOCKED(pp) && !PP_ISNORELOC(pp)) { 1831e803c1e6SPrakash Sangappa panic("hat_devload: loading a mapping " 1832e803c1e6SPrakash Sangappa "to an unlocked page %p", 1833e803c1e6SPrakash Sangappa (void *)pp); 1834e803c1e6SPrakash Sangappa } 1835e803c1e6SPrakash Sangappa } 1836e803c1e6SPrakash Sangappa 18377c478bd9Sstevel@tonic-gate /* 18387c478bd9Sstevel@tonic-gate * load this page mapping 18397c478bd9Sstevel@tonic-gate */ 18407c478bd9Sstevel@tonic-gate ASSERT(!IN_VA_HOLE(va)); 1841ae115bc7Smrj while (hati_load_common(hat, va, pp, a, f, level, pfn) != 0) { 1842ae115bc7Smrj if (level == 0) 1843ae115bc7Smrj panic("unexpected hati_load_common() failure"); 1844ae115bc7Smrj --level; 1845ae115bc7Smrj pgsize = LEVEL_SIZE(level); 1846ae115bc7Smrj } 18477c478bd9Sstevel@tonic-gate 18487c478bd9Sstevel@tonic-gate /* 18497c478bd9Sstevel@tonic-gate * move to next page 18507c478bd9Sstevel@tonic-gate */ 18517c478bd9Sstevel@tonic-gate va += pgsize; 18527c478bd9Sstevel@tonic-gate pfn += mmu_btop(pgsize); 18537c478bd9Sstevel@tonic-gate } 1854843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 18557c478bd9Sstevel@tonic-gate } 18567c478bd9Sstevel@tonic-gate 18577c478bd9Sstevel@tonic-gate /* 18587c478bd9Sstevel@tonic-gate * void hat_unlock(hat, addr, len) 18597c478bd9Sstevel@tonic-gate * unlock the mappings to a given range of addresses 18607c478bd9Sstevel@tonic-gate * 18617c478bd9Sstevel@tonic-gate * Locks are tracked by ht_lock_cnt in the htable. 18627c478bd9Sstevel@tonic-gate */ 18637c478bd9Sstevel@tonic-gate void 18647c478bd9Sstevel@tonic-gate hat_unlock(hat_t *hat, caddr_t addr, size_t len) 18657c478bd9Sstevel@tonic-gate { 18667c478bd9Sstevel@tonic-gate uintptr_t vaddr = (uintptr_t)addr; 18677c478bd9Sstevel@tonic-gate uintptr_t eaddr = vaddr + len; 18687c478bd9Sstevel@tonic-gate htable_t *ht = NULL; 18697c478bd9Sstevel@tonic-gate 18707c478bd9Sstevel@tonic-gate /* 18717c478bd9Sstevel@tonic-gate * kernel entries are always locked, we don't track lock counts 18727c478bd9Sstevel@tonic-gate */ 1873ae115bc7Smrj ASSERT(hat == kas.a_hat || eaddr <= _userlimit); 18747c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(vaddr)); 18757c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(eaddr)); 18767c478bd9Sstevel@tonic-gate if (hat == kas.a_hat) 18777c478bd9Sstevel@tonic-gate return; 18787c478bd9Sstevel@tonic-gate if (eaddr > _userlimit) 18797c478bd9Sstevel@tonic-gate panic("hat_unlock() address out of range - above _userlimit"); 18807c478bd9Sstevel@tonic-gate 1881843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 18827c478bd9Sstevel@tonic-gate ASSERT(AS_LOCK_HELD(hat->hat_as, &hat->hat_as->a_lock)); 18837c478bd9Sstevel@tonic-gate while (vaddr < eaddr) { 18847c478bd9Sstevel@tonic-gate (void) htable_walk(hat, &ht, &vaddr, eaddr); 18857c478bd9Sstevel@tonic-gate if (ht == NULL) 18867c478bd9Sstevel@tonic-gate break; 18877c478bd9Sstevel@tonic-gate 18887c478bd9Sstevel@tonic-gate ASSERT(!IN_VA_HOLE(vaddr)); 18897c478bd9Sstevel@tonic-gate 18907c478bd9Sstevel@tonic-gate if (ht->ht_lock_cnt < 1) 18917c478bd9Sstevel@tonic-gate panic("hat_unlock(): lock_cnt < 1, " 1892903a11ebSrh "htable=%p, vaddr=%p\n", (void *)ht, (void *)vaddr); 18937c478bd9Sstevel@tonic-gate HTABLE_LOCK_DEC(ht); 18947c478bd9Sstevel@tonic-gate 18957c478bd9Sstevel@tonic-gate vaddr += LEVEL_SIZE(ht->ht_level); 18967c478bd9Sstevel@tonic-gate } 18977c478bd9Sstevel@tonic-gate if (ht) 18987c478bd9Sstevel@tonic-gate htable_release(ht); 1899843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 19007c478bd9Sstevel@tonic-gate } 19017c478bd9Sstevel@tonic-gate 190205d3dc4bSpaulsan /* ARGSUSED */ 190305d3dc4bSpaulsan void 19047dacfc44Spaulsan hat_unlock_region(struct hat *hat, caddr_t addr, size_t len, 190505d3dc4bSpaulsan hat_region_cookie_t rcookie) 190605d3dc4bSpaulsan { 190705d3dc4bSpaulsan panic("No shared region support on x86"); 190805d3dc4bSpaulsan } 190905d3dc4bSpaulsan 1910843e1988Sjohnlev #if !defined(__xpv) 19117c478bd9Sstevel@tonic-gate /* 19127c478bd9Sstevel@tonic-gate * Cross call service routine to demap a virtual page on 19137c478bd9Sstevel@tonic-gate * the current CPU or flush all mappings in TLB. 19147c478bd9Sstevel@tonic-gate */ 19157c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 19167c478bd9Sstevel@tonic-gate static int 19177c478bd9Sstevel@tonic-gate hati_demap_func(xc_arg_t a1, xc_arg_t a2, xc_arg_t a3) 19187c478bd9Sstevel@tonic-gate { 19197c478bd9Sstevel@tonic-gate hat_t *hat = (hat_t *)a1; 19207c478bd9Sstevel@tonic-gate caddr_t addr = (caddr_t)a2; 19217c478bd9Sstevel@tonic-gate 19227c478bd9Sstevel@tonic-gate /* 19237c478bd9Sstevel@tonic-gate * If the target hat isn't the kernel and this CPU isn't operating 19247c478bd9Sstevel@tonic-gate * in the target hat, we can ignore the cross call. 19257c478bd9Sstevel@tonic-gate */ 19267c478bd9Sstevel@tonic-gate if (hat != kas.a_hat && hat != CPU->cpu_current_hat) 19277c478bd9Sstevel@tonic-gate return (0); 19287c478bd9Sstevel@tonic-gate 19297c478bd9Sstevel@tonic-gate /* 19307c478bd9Sstevel@tonic-gate * For a normal address, we just flush one page mapping 19317c478bd9Sstevel@tonic-gate */ 19327c478bd9Sstevel@tonic-gate if ((uintptr_t)addr != DEMAP_ALL_ADDR) { 1933ae115bc7Smrj mmu_tlbflush_entry(addr); 19347c478bd9Sstevel@tonic-gate return (0); 19357c478bd9Sstevel@tonic-gate } 19367c478bd9Sstevel@tonic-gate 19377c478bd9Sstevel@tonic-gate /* 19387c478bd9Sstevel@tonic-gate * Otherwise we reload cr3 to effect a complete TLB flush. 19397c478bd9Sstevel@tonic-gate * 19407c478bd9Sstevel@tonic-gate * A reload of cr3 on a VLP process also means we must also recopy in 19417c478bd9Sstevel@tonic-gate * the pte values from the struct hat 19427c478bd9Sstevel@tonic-gate */ 19437c478bd9Sstevel@tonic-gate if (hat->hat_flags & HAT_VLP) { 19447c478bd9Sstevel@tonic-gate #if defined(__amd64) 19457c478bd9Sstevel@tonic-gate x86pte_t *vlpptep = CPU->cpu_hat_info->hci_vlp_l2ptes; 19467c478bd9Sstevel@tonic-gate 19477c478bd9Sstevel@tonic-gate VLP_COPY(hat->hat_vlp_ptes, vlpptep); 19487c478bd9Sstevel@tonic-gate #elif defined(__i386) 19497c478bd9Sstevel@tonic-gate reload_pae32(hat, CPU); 19507c478bd9Sstevel@tonic-gate #endif 19517c478bd9Sstevel@tonic-gate } 19527c478bd9Sstevel@tonic-gate reload_cr3(); 19537c478bd9Sstevel@tonic-gate return (0); 19547c478bd9Sstevel@tonic-gate } 19557c478bd9Sstevel@tonic-gate 195695c0a3c8Sjosephb /* 195795c0a3c8Sjosephb * Flush all TLB entries, including global (ie. kernel) ones. 195895c0a3c8Sjosephb */ 195995c0a3c8Sjosephb static void 196095c0a3c8Sjosephb flush_all_tlb_entries(void) 196195c0a3c8Sjosephb { 196295c0a3c8Sjosephb ulong_t cr4 = getcr4(); 196395c0a3c8Sjosephb 196495c0a3c8Sjosephb if (cr4 & CR4_PGE) { 196595c0a3c8Sjosephb setcr4(cr4 & ~(ulong_t)CR4_PGE); 196695c0a3c8Sjosephb setcr4(cr4); 196795c0a3c8Sjosephb 196895c0a3c8Sjosephb /* 196995c0a3c8Sjosephb * 32 bit PAE also needs to always reload_cr3() 197095c0a3c8Sjosephb */ 197195c0a3c8Sjosephb if (mmu.max_level == 2) 197295c0a3c8Sjosephb reload_cr3(); 197395c0a3c8Sjosephb } else { 197495c0a3c8Sjosephb reload_cr3(); 197595c0a3c8Sjosephb } 197695c0a3c8Sjosephb } 197795c0a3c8Sjosephb 197895c0a3c8Sjosephb #define TLB_CPU_HALTED (01ul) 197995c0a3c8Sjosephb #define TLB_INVAL_ALL (02ul) 198095c0a3c8Sjosephb #define CAS_TLB_INFO(cpu, old, new) \ 198195c0a3c8Sjosephb caslong((ulong_t *)&(cpu)->cpu_m.mcpu_tlb_info, (old), (new)) 198295c0a3c8Sjosephb 198395c0a3c8Sjosephb /* 198495c0a3c8Sjosephb * Record that a CPU is going idle 198595c0a3c8Sjosephb */ 198695c0a3c8Sjosephb void 198795c0a3c8Sjosephb tlb_going_idle(void) 198895c0a3c8Sjosephb { 198995c0a3c8Sjosephb atomic_or_long((ulong_t *)&CPU->cpu_m.mcpu_tlb_info, TLB_CPU_HALTED); 199095c0a3c8Sjosephb } 199195c0a3c8Sjosephb 199295c0a3c8Sjosephb /* 199395c0a3c8Sjosephb * Service a delayed TLB flush if coming out of being idle. 1994*21584dbcSPavel Tatashin * It will be called from cpu idle notification with interrupt disabled. 199595c0a3c8Sjosephb */ 199695c0a3c8Sjosephb void 199795c0a3c8Sjosephb tlb_service(void) 199895c0a3c8Sjosephb { 199995c0a3c8Sjosephb ulong_t tlb_info; 200095c0a3c8Sjosephb ulong_t found; 200195c0a3c8Sjosephb 200295c0a3c8Sjosephb /* 200395c0a3c8Sjosephb * We only have to do something if coming out of being idle. 200495c0a3c8Sjosephb */ 200595c0a3c8Sjosephb tlb_info = CPU->cpu_m.mcpu_tlb_info; 200695c0a3c8Sjosephb if (tlb_info & TLB_CPU_HALTED) { 200795c0a3c8Sjosephb ASSERT(CPU->cpu_current_hat == kas.a_hat); 200895c0a3c8Sjosephb 200995c0a3c8Sjosephb /* 201095c0a3c8Sjosephb * Atomic clear and fetch of old state. 201195c0a3c8Sjosephb */ 201295c0a3c8Sjosephb while ((found = CAS_TLB_INFO(CPU, tlb_info, 0)) != tlb_info) { 201395c0a3c8Sjosephb ASSERT(found & TLB_CPU_HALTED); 201495c0a3c8Sjosephb tlb_info = found; 201595c0a3c8Sjosephb SMT_PAUSE(); 201695c0a3c8Sjosephb } 201795c0a3c8Sjosephb if (tlb_info & TLB_INVAL_ALL) 201895c0a3c8Sjosephb flush_all_tlb_entries(); 201995c0a3c8Sjosephb } 202095c0a3c8Sjosephb } 2021843e1988Sjohnlev #endif /* !__xpv */ 202295c0a3c8Sjosephb 20237c478bd9Sstevel@tonic-gate /* 20247c478bd9Sstevel@tonic-gate * Internal routine to do cross calls to invalidate a range of pages on 20257c478bd9Sstevel@tonic-gate * all CPUs using a given hat. 20267c478bd9Sstevel@tonic-gate */ 20277c478bd9Sstevel@tonic-gate void 2028ae115bc7Smrj hat_tlb_inval(hat_t *hat, uintptr_t va) 20297c478bd9Sstevel@tonic-gate { 20307c478bd9Sstevel@tonic-gate extern int flushes_require_xcalls; /* from mp_startup.c */ 20317c478bd9Sstevel@tonic-gate cpuset_t justme; 2032ae115bc7Smrj cpuset_t cpus_to_shootdown; 2033843e1988Sjohnlev #ifndef __xpv 2034843e1988Sjohnlev cpuset_t check_cpus; 203595c0a3c8Sjosephb cpu_t *cpup; 203695c0a3c8Sjosephb int c; 2037843e1988Sjohnlev #endif 20387c478bd9Sstevel@tonic-gate 20397c478bd9Sstevel@tonic-gate /* 20407c478bd9Sstevel@tonic-gate * If the hat is being destroyed, there are no more users, so 20417c478bd9Sstevel@tonic-gate * demap need not do anything. 20427c478bd9Sstevel@tonic-gate */ 20437c478bd9Sstevel@tonic-gate if (hat->hat_flags & HAT_FREEING) 20447c478bd9Sstevel@tonic-gate return; 20457c478bd9Sstevel@tonic-gate 20467c478bd9Sstevel@tonic-gate /* 20477c478bd9Sstevel@tonic-gate * If demapping from a shared pagetable, we best demap the 20487c478bd9Sstevel@tonic-gate * entire set of user TLBs, since we don't know what addresses 20497c478bd9Sstevel@tonic-gate * these were shared at. 20507c478bd9Sstevel@tonic-gate */ 20517c478bd9Sstevel@tonic-gate if (hat->hat_flags & HAT_SHARED) { 20527c478bd9Sstevel@tonic-gate hat = kas.a_hat; 20537c478bd9Sstevel@tonic-gate va = DEMAP_ALL_ADDR; 20547c478bd9Sstevel@tonic-gate } 20557c478bd9Sstevel@tonic-gate 20567c478bd9Sstevel@tonic-gate /* 20577c478bd9Sstevel@tonic-gate * if not running with multiple CPUs, don't use cross calls 20587c478bd9Sstevel@tonic-gate */ 20597c478bd9Sstevel@tonic-gate if (panicstr || !flushes_require_xcalls) { 2060843e1988Sjohnlev #ifdef __xpv 2061843e1988Sjohnlev if (va == DEMAP_ALL_ADDR) 2062843e1988Sjohnlev xen_flush_tlb(); 2063843e1988Sjohnlev else 2064843e1988Sjohnlev xen_flush_va((caddr_t)va); 2065843e1988Sjohnlev #else 20667c478bd9Sstevel@tonic-gate (void) hati_demap_func((xc_arg_t)hat, (xc_arg_t)va, NULL); 2067843e1988Sjohnlev #endif 20687c478bd9Sstevel@tonic-gate return; 20697c478bd9Sstevel@tonic-gate } 20707c478bd9Sstevel@tonic-gate 20717c478bd9Sstevel@tonic-gate 20727c478bd9Sstevel@tonic-gate /* 2073ae115bc7Smrj * Determine CPUs to shootdown. Kernel changes always do all CPUs. 2074ae115bc7Smrj * Otherwise it's just CPUs currently executing in this hat. 20757c478bd9Sstevel@tonic-gate */ 20767c478bd9Sstevel@tonic-gate kpreempt_disable(); 20777c478bd9Sstevel@tonic-gate CPUSET_ONLY(justme, CPU->cpu_id); 2078ae115bc7Smrj if (hat == kas.a_hat) 2079ae115bc7Smrj cpus_to_shootdown = khat_cpuset; 20807c478bd9Sstevel@tonic-gate else 2081ae115bc7Smrj cpus_to_shootdown = hat->hat_cpus; 2082ae115bc7Smrj 2083843e1988Sjohnlev #ifndef __xpv 208495c0a3c8Sjosephb /* 208595c0a3c8Sjosephb * If any CPUs in the set are idle, just request a delayed flush 208695c0a3c8Sjosephb * and avoid waking them up. 208795c0a3c8Sjosephb */ 208895c0a3c8Sjosephb check_cpus = cpus_to_shootdown; 208995c0a3c8Sjosephb for (c = 0; c < NCPU && !CPUSET_ISNULL(check_cpus); ++c) { 209095c0a3c8Sjosephb ulong_t tlb_info; 209195c0a3c8Sjosephb 209295c0a3c8Sjosephb if (!CPU_IN_SET(check_cpus, c)) 209395c0a3c8Sjosephb continue; 209495c0a3c8Sjosephb CPUSET_DEL(check_cpus, c); 209595c0a3c8Sjosephb cpup = cpu[c]; 209695c0a3c8Sjosephb if (cpup == NULL) 209795c0a3c8Sjosephb continue; 209895c0a3c8Sjosephb 209995c0a3c8Sjosephb tlb_info = cpup->cpu_m.mcpu_tlb_info; 210095c0a3c8Sjosephb while (tlb_info == TLB_CPU_HALTED) { 210195c0a3c8Sjosephb (void) CAS_TLB_INFO(cpup, TLB_CPU_HALTED, 2102250b7ff9Sjosephb TLB_CPU_HALTED | TLB_INVAL_ALL); 210395c0a3c8Sjosephb SMT_PAUSE(); 210495c0a3c8Sjosephb tlb_info = cpup->cpu_m.mcpu_tlb_info; 210595c0a3c8Sjosephb } 210695c0a3c8Sjosephb if (tlb_info == (TLB_CPU_HALTED | TLB_INVAL_ALL)) { 210795c0a3c8Sjosephb HATSTAT_INC(hs_tlb_inval_delayed); 210895c0a3c8Sjosephb CPUSET_DEL(cpus_to_shootdown, c); 210995c0a3c8Sjosephb } 211095c0a3c8Sjosephb } 2111843e1988Sjohnlev #endif 211295c0a3c8Sjosephb 2113ae115bc7Smrj if (CPUSET_ISNULL(cpus_to_shootdown) || 2114ae115bc7Smrj CPUSET_ISEQUAL(cpus_to_shootdown, justme)) { 2115ae115bc7Smrj 2116843e1988Sjohnlev #ifdef __xpv 2117843e1988Sjohnlev if (va == DEMAP_ALL_ADDR) 2118843e1988Sjohnlev xen_flush_tlb(); 2119843e1988Sjohnlev else 2120843e1988Sjohnlev xen_flush_va((caddr_t)va); 2121843e1988Sjohnlev #else 2122ae115bc7Smrj (void) hati_demap_func((xc_arg_t)hat, (xc_arg_t)va, NULL); 2123843e1988Sjohnlev #endif 2124ae115bc7Smrj 2125ae115bc7Smrj } else { 2126ae115bc7Smrj 2127ae115bc7Smrj CPUSET_ADD(cpus_to_shootdown, CPU->cpu_id); 2128843e1988Sjohnlev #ifdef __xpv 2129843e1988Sjohnlev if (va == DEMAP_ALL_ADDR) 2130843e1988Sjohnlev xen_gflush_tlb(cpus_to_shootdown); 2131843e1988Sjohnlev else 2132843e1988Sjohnlev xen_gflush_va((caddr_t)va, cpus_to_shootdown); 2133843e1988Sjohnlev #else 2134f34a7178SJoe Bonasera xc_call((xc_arg_t)hat, (xc_arg_t)va, NULL, 2135f34a7178SJoe Bonasera CPUSET2BV(cpus_to_shootdown), hati_demap_func); 2136843e1988Sjohnlev #endif 2137ae115bc7Smrj 2138ae115bc7Smrj } 21397c478bd9Sstevel@tonic-gate kpreempt_enable(); 21407c478bd9Sstevel@tonic-gate } 21417c478bd9Sstevel@tonic-gate 21427c478bd9Sstevel@tonic-gate /* 21437c478bd9Sstevel@tonic-gate * Interior routine for HAT_UNLOADs from hat_unload_callback(), 21447c478bd9Sstevel@tonic-gate * hat_kmap_unload() OR from hat_steal() code. This routine doesn't 21457c478bd9Sstevel@tonic-gate * handle releasing of the htables. 21467c478bd9Sstevel@tonic-gate */ 21477c478bd9Sstevel@tonic-gate void 21487c478bd9Sstevel@tonic-gate hat_pte_unmap( 21497c478bd9Sstevel@tonic-gate htable_t *ht, 21507c478bd9Sstevel@tonic-gate uint_t entry, 21517c478bd9Sstevel@tonic-gate uint_t flags, 21527c478bd9Sstevel@tonic-gate x86pte_t old_pte, 21537c478bd9Sstevel@tonic-gate void *pte_ptr) 21547c478bd9Sstevel@tonic-gate { 21557c478bd9Sstevel@tonic-gate hat_t *hat = ht->ht_hat; 21567c478bd9Sstevel@tonic-gate hment_t *hm = NULL; 21577c478bd9Sstevel@tonic-gate page_t *pp = NULL; 21587c478bd9Sstevel@tonic-gate level_t l = ht->ht_level; 21597c478bd9Sstevel@tonic-gate pfn_t pfn; 21607c478bd9Sstevel@tonic-gate 21617c478bd9Sstevel@tonic-gate /* 21627c478bd9Sstevel@tonic-gate * We always track the locking counts, even if nothing is unmapped 21637c478bd9Sstevel@tonic-gate */ 21647c478bd9Sstevel@tonic-gate if ((flags & HAT_UNLOAD_UNLOCK) != 0 && hat != kas.a_hat) { 21657c478bd9Sstevel@tonic-gate ASSERT(ht->ht_lock_cnt > 0); 21667c478bd9Sstevel@tonic-gate HTABLE_LOCK_DEC(ht); 21677c478bd9Sstevel@tonic-gate } 21687c478bd9Sstevel@tonic-gate 21697c478bd9Sstevel@tonic-gate /* 21707c478bd9Sstevel@tonic-gate * Figure out which page's mapping list lock to acquire using the PFN 21717c478bd9Sstevel@tonic-gate * passed in "old" PTE. We then attempt to invalidate the PTE. 21727c478bd9Sstevel@tonic-gate * If another thread, probably a hat_pageunload, has asynchronously 21737c478bd9Sstevel@tonic-gate * unmapped/remapped this address we'll loop here. 21747c478bd9Sstevel@tonic-gate */ 21757c478bd9Sstevel@tonic-gate ASSERT(ht->ht_busy > 0); 21767c478bd9Sstevel@tonic-gate while (PTE_ISVALID(old_pte)) { 21777c478bd9Sstevel@tonic-gate pfn = PTE2PFN(old_pte, l); 2178ae115bc7Smrj if (PTE_GET(old_pte, PT_SOFTWARE) >= PT_NOCONSIST) { 21797c478bd9Sstevel@tonic-gate pp = NULL; 21807c478bd9Sstevel@tonic-gate } else { 2181843e1988Sjohnlev #ifdef __xpv 2182843e1988Sjohnlev if (pfn == PFN_INVALID) 2183843e1988Sjohnlev panic("Invalid PFN, but not PT_NOCONSIST"); 2184843e1988Sjohnlev #endif 21857c478bd9Sstevel@tonic-gate pp = page_numtopp_nolock(pfn); 2186aa2ed9e5Sjosephb if (pp == NULL) { 2187aa2ed9e5Sjosephb panic("no page_t, not NOCONSIST: old_pte=" 2188aa2ed9e5Sjosephb FMT_PTE " ht=%lx entry=0x%x pte_ptr=%lx", 2189aa2ed9e5Sjosephb old_pte, (uintptr_t)ht, entry, 2190aa2ed9e5Sjosephb (uintptr_t)pte_ptr); 2191aa2ed9e5Sjosephb } 21927c478bd9Sstevel@tonic-gate x86_hm_enter(pp); 21937c478bd9Sstevel@tonic-gate } 2194aa2ed9e5Sjosephb 2195aa2ed9e5Sjosephb /* 2196aa2ed9e5Sjosephb * If freeing the address space, check that the PTE 2197aa2ed9e5Sjosephb * hasn't changed, as the mappings are no longer in use by 2198aa2ed9e5Sjosephb * any thread, invalidation is unnecessary. 2199aa2ed9e5Sjosephb * If not freeing, do a full invalidate. 2200843e1988Sjohnlev * 2201843e1988Sjohnlev * On the hypervisor we must always remove mappings, as a 2202843e1988Sjohnlev * writable mapping left behind could cause a page table 2203843e1988Sjohnlev * allocation to fail. 2204aa2ed9e5Sjosephb */ 2205843e1988Sjohnlev #if !defined(__xpv) 2206aa2ed9e5Sjosephb if (hat->hat_flags & HAT_FREEING) 2207aa2ed9e5Sjosephb old_pte = x86pte_get(ht, entry); 2208aa2ed9e5Sjosephb else 2209843e1988Sjohnlev #endif 2210ae115bc7Smrj old_pte = x86pte_inval(ht, entry, old_pte, pte_ptr); 22117c478bd9Sstevel@tonic-gate 22127c478bd9Sstevel@tonic-gate /* 22137c478bd9Sstevel@tonic-gate * If the page hadn't changed we've unmapped it and can proceed 22147c478bd9Sstevel@tonic-gate */ 22157c478bd9Sstevel@tonic-gate if (PTE_ISVALID(old_pte) && PTE2PFN(old_pte, l) == pfn) 22167c478bd9Sstevel@tonic-gate break; 22177c478bd9Sstevel@tonic-gate 22187c478bd9Sstevel@tonic-gate /* 22197c478bd9Sstevel@tonic-gate * Otherwise, we'll have to retry with the current old_pte. 22207c478bd9Sstevel@tonic-gate * Drop the hment lock, since the pfn may have changed. 22217c478bd9Sstevel@tonic-gate */ 22227c478bd9Sstevel@tonic-gate if (pp != NULL) { 22237c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 22247c478bd9Sstevel@tonic-gate pp = NULL; 22257c478bd9Sstevel@tonic-gate } else { 2226ae115bc7Smrj ASSERT(PTE_GET(old_pte, PT_SOFTWARE) >= PT_NOCONSIST); 22277c478bd9Sstevel@tonic-gate } 22287c478bd9Sstevel@tonic-gate } 22297c478bd9Sstevel@tonic-gate 22307c478bd9Sstevel@tonic-gate /* 22317c478bd9Sstevel@tonic-gate * If the old mapping wasn't valid, there's nothing more to do 22327c478bd9Sstevel@tonic-gate */ 22337c478bd9Sstevel@tonic-gate if (!PTE_ISVALID(old_pte)) { 22347c478bd9Sstevel@tonic-gate if (pp != NULL) 22357c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 22367c478bd9Sstevel@tonic-gate return; 22377c478bd9Sstevel@tonic-gate } 22387c478bd9Sstevel@tonic-gate 22397c478bd9Sstevel@tonic-gate /* 22407c478bd9Sstevel@tonic-gate * Take care of syncing any MOD/REF bits and removing the hment. 22417c478bd9Sstevel@tonic-gate */ 22427c478bd9Sstevel@tonic-gate if (pp != NULL) { 22437c478bd9Sstevel@tonic-gate if (!(flags & HAT_UNLOAD_NOSYNC)) 22447c478bd9Sstevel@tonic-gate hati_sync_pte_to_page(pp, old_pte, l); 22457c478bd9Sstevel@tonic-gate hm = hment_remove(pp, ht, entry); 22467c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 22477c478bd9Sstevel@tonic-gate if (hm != NULL) 22487c478bd9Sstevel@tonic-gate hment_free(hm); 22497c478bd9Sstevel@tonic-gate } 22507c478bd9Sstevel@tonic-gate 22517c478bd9Sstevel@tonic-gate /* 22527c478bd9Sstevel@tonic-gate * Handle book keeping in the htable and hat 22537c478bd9Sstevel@tonic-gate */ 22547c478bd9Sstevel@tonic-gate ASSERT(ht->ht_valid_cnt > 0); 22557c478bd9Sstevel@tonic-gate HTABLE_DEC(ht->ht_valid_cnt); 22567c478bd9Sstevel@tonic-gate PGCNT_DEC(hat, l); 22577c478bd9Sstevel@tonic-gate } 22587c478bd9Sstevel@tonic-gate 22597c478bd9Sstevel@tonic-gate /* 22607c478bd9Sstevel@tonic-gate * very cheap unload implementation to special case some kernel addresses 22617c478bd9Sstevel@tonic-gate */ 22627c478bd9Sstevel@tonic-gate static void 22637c478bd9Sstevel@tonic-gate hat_kmap_unload(caddr_t addr, size_t len, uint_t flags) 22647c478bd9Sstevel@tonic-gate { 22657c478bd9Sstevel@tonic-gate uintptr_t va = (uintptr_t)addr; 22667c478bd9Sstevel@tonic-gate uintptr_t eva = va + len; 2267ae115bc7Smrj pgcnt_t pg_index; 22687c478bd9Sstevel@tonic-gate htable_t *ht; 22697c478bd9Sstevel@tonic-gate uint_t entry; 2270ae115bc7Smrj x86pte_t *pte_ptr; 22717c478bd9Sstevel@tonic-gate x86pte_t old_pte; 22727c478bd9Sstevel@tonic-gate 22737c478bd9Sstevel@tonic-gate for (; va < eva; va += MMU_PAGESIZE) { 22747c478bd9Sstevel@tonic-gate /* 22757c478bd9Sstevel@tonic-gate * Get the PTE 22767c478bd9Sstevel@tonic-gate */ 2277ae115bc7Smrj pg_index = mmu_btop(va - mmu.kmap_addr); 2278ae115bc7Smrj pte_ptr = PT_INDEX_PTR(mmu.kmap_ptes, pg_index); 2279ae115bc7Smrj old_pte = GET_PTE(pte_ptr); 22807c478bd9Sstevel@tonic-gate 22817c478bd9Sstevel@tonic-gate /* 22827c478bd9Sstevel@tonic-gate * get the htable / entry 22837c478bd9Sstevel@tonic-gate */ 22847c478bd9Sstevel@tonic-gate ht = mmu.kmap_htables[(va - mmu.kmap_htables[0]->ht_vaddr) 22857c478bd9Sstevel@tonic-gate >> LEVEL_SHIFT(1)]; 22867c478bd9Sstevel@tonic-gate entry = htable_va2entry(va, ht); 22877c478bd9Sstevel@tonic-gate 22887c478bd9Sstevel@tonic-gate /* 22897c478bd9Sstevel@tonic-gate * use mostly common code to unmap it. 22907c478bd9Sstevel@tonic-gate */ 22917c478bd9Sstevel@tonic-gate hat_pte_unmap(ht, entry, flags, old_pte, pte_ptr); 22927c478bd9Sstevel@tonic-gate } 22937c478bd9Sstevel@tonic-gate } 22947c478bd9Sstevel@tonic-gate 22957c478bd9Sstevel@tonic-gate 22967c478bd9Sstevel@tonic-gate /* 22977c478bd9Sstevel@tonic-gate * unload a range of virtual address space (no callback) 22987c478bd9Sstevel@tonic-gate */ 22997c478bd9Sstevel@tonic-gate void 23007c478bd9Sstevel@tonic-gate hat_unload(hat_t *hat, caddr_t addr, size_t len, uint_t flags) 23017c478bd9Sstevel@tonic-gate { 23027c478bd9Sstevel@tonic-gate uintptr_t va = (uintptr_t)addr; 2303ae115bc7Smrj 2304843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 2305ae115bc7Smrj ASSERT(hat == kas.a_hat || va + len <= _userlimit); 23067c478bd9Sstevel@tonic-gate 23077c478bd9Sstevel@tonic-gate /* 23087c478bd9Sstevel@tonic-gate * special case for performance. 23097c478bd9Sstevel@tonic-gate */ 23107c478bd9Sstevel@tonic-gate if (mmu.kmap_addr <= va && va < mmu.kmap_eaddr) { 23117c478bd9Sstevel@tonic-gate ASSERT(hat == kas.a_hat); 23127c478bd9Sstevel@tonic-gate hat_kmap_unload(addr, len, flags); 2313ae115bc7Smrj } else { 2314ae115bc7Smrj hat_unload_callback(hat, addr, len, flags, NULL); 23157c478bd9Sstevel@tonic-gate } 2316843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 23177c478bd9Sstevel@tonic-gate } 23187c478bd9Sstevel@tonic-gate 23197c478bd9Sstevel@tonic-gate /* 23207c478bd9Sstevel@tonic-gate * Do the callbacks for ranges being unloaded. 23217c478bd9Sstevel@tonic-gate */ 23227c478bd9Sstevel@tonic-gate typedef struct range_info { 23237c478bd9Sstevel@tonic-gate uintptr_t rng_va; 23247c478bd9Sstevel@tonic-gate ulong_t rng_cnt; 23257c478bd9Sstevel@tonic-gate level_t rng_level; 23267c478bd9Sstevel@tonic-gate } range_info_t; 23277c478bd9Sstevel@tonic-gate 23287c478bd9Sstevel@tonic-gate static void 23297c478bd9Sstevel@tonic-gate handle_ranges(hat_callback_t *cb, uint_t cnt, range_info_t *range) 23307c478bd9Sstevel@tonic-gate { 23317c478bd9Sstevel@tonic-gate /* 23327c478bd9Sstevel@tonic-gate * do callbacks to upper level VM system 23337c478bd9Sstevel@tonic-gate */ 23347c478bd9Sstevel@tonic-gate while (cb != NULL && cnt > 0) { 23357c478bd9Sstevel@tonic-gate --cnt; 23367c478bd9Sstevel@tonic-gate cb->hcb_start_addr = (caddr_t)range[cnt].rng_va; 23377c478bd9Sstevel@tonic-gate cb->hcb_end_addr = cb->hcb_start_addr; 23387c478bd9Sstevel@tonic-gate cb->hcb_end_addr += 23397c478bd9Sstevel@tonic-gate range[cnt].rng_cnt << LEVEL_SIZE(range[cnt].rng_level); 23407c478bd9Sstevel@tonic-gate cb->hcb_function(cb); 23417c478bd9Sstevel@tonic-gate } 23427c478bd9Sstevel@tonic-gate } 23437c478bd9Sstevel@tonic-gate 23447c478bd9Sstevel@tonic-gate /* 23457c478bd9Sstevel@tonic-gate * Unload a given range of addresses (has optional callback) 23467c478bd9Sstevel@tonic-gate * 23477c478bd9Sstevel@tonic-gate * Flags: 23487c478bd9Sstevel@tonic-gate * define HAT_UNLOAD 0x00 23497c478bd9Sstevel@tonic-gate * define HAT_UNLOAD_NOSYNC 0x02 23507c478bd9Sstevel@tonic-gate * define HAT_UNLOAD_UNLOCK 0x04 23517c478bd9Sstevel@tonic-gate * define HAT_UNLOAD_OTHER 0x08 - not used 23527c478bd9Sstevel@tonic-gate * define HAT_UNLOAD_UNMAP 0x10 - same as HAT_UNLOAD 23537c478bd9Sstevel@tonic-gate */ 23547c478bd9Sstevel@tonic-gate #define MAX_UNLOAD_CNT (8) 23557c478bd9Sstevel@tonic-gate void 23567c478bd9Sstevel@tonic-gate hat_unload_callback( 23577c478bd9Sstevel@tonic-gate hat_t *hat, 23587c478bd9Sstevel@tonic-gate caddr_t addr, 23597c478bd9Sstevel@tonic-gate size_t len, 23607c478bd9Sstevel@tonic-gate uint_t flags, 23617c478bd9Sstevel@tonic-gate hat_callback_t *cb) 23627c478bd9Sstevel@tonic-gate { 23637c478bd9Sstevel@tonic-gate uintptr_t vaddr = (uintptr_t)addr; 23647c478bd9Sstevel@tonic-gate uintptr_t eaddr = vaddr + len; 23657c478bd9Sstevel@tonic-gate htable_t *ht = NULL; 23667c478bd9Sstevel@tonic-gate uint_t entry; 2367aa2ed9e5Sjosephb uintptr_t contig_va = (uintptr_t)-1L; 23687c478bd9Sstevel@tonic-gate range_info_t r[MAX_UNLOAD_CNT]; 23697c478bd9Sstevel@tonic-gate uint_t r_cnt = 0; 23707c478bd9Sstevel@tonic-gate x86pte_t old_pte; 23717c478bd9Sstevel@tonic-gate 2372843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 2373ae115bc7Smrj ASSERT(hat == kas.a_hat || eaddr <= _userlimit); 23747c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(vaddr)); 23757c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(eaddr)); 23767c478bd9Sstevel@tonic-gate 2377ae115bc7Smrj /* 2378ae115bc7Smrj * Special case a single page being unloaded for speed. This happens 2379ae115bc7Smrj * quite frequently, COW faults after a fork() for example. 2380ae115bc7Smrj */ 2381ae115bc7Smrj if (cb == NULL && len == MMU_PAGESIZE) { 2382ae115bc7Smrj ht = htable_getpte(hat, vaddr, &entry, &old_pte, 0); 2383ae115bc7Smrj if (ht != NULL) { 2384ae115bc7Smrj if (PTE_ISVALID(old_pte)) 2385ae115bc7Smrj hat_pte_unmap(ht, entry, flags, old_pte, NULL); 2386ae115bc7Smrj htable_release(ht); 2387ae115bc7Smrj } 2388843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 2389ae115bc7Smrj return; 2390ae115bc7Smrj } 2391ae115bc7Smrj 23927c478bd9Sstevel@tonic-gate while (vaddr < eaddr) { 23937c478bd9Sstevel@tonic-gate old_pte = htable_walk(hat, &ht, &vaddr, eaddr); 23947c478bd9Sstevel@tonic-gate if (ht == NULL) 23957c478bd9Sstevel@tonic-gate break; 23967c478bd9Sstevel@tonic-gate 23977c478bd9Sstevel@tonic-gate ASSERT(!IN_VA_HOLE(vaddr)); 23987c478bd9Sstevel@tonic-gate 23997c478bd9Sstevel@tonic-gate if (vaddr < (uintptr_t)addr) 24007c478bd9Sstevel@tonic-gate panic("hat_unload_callback(): unmap inside large page"); 24017c478bd9Sstevel@tonic-gate 24027c478bd9Sstevel@tonic-gate /* 24037c478bd9Sstevel@tonic-gate * We'll do the call backs for contiguous ranges 24047c478bd9Sstevel@tonic-gate */ 2405aa2ed9e5Sjosephb if (vaddr != contig_va || 24067c478bd9Sstevel@tonic-gate (r_cnt > 0 && r[r_cnt - 1].rng_level != ht->ht_level)) { 24077c478bd9Sstevel@tonic-gate if (r_cnt == MAX_UNLOAD_CNT) { 24087c478bd9Sstevel@tonic-gate handle_ranges(cb, r_cnt, r); 24097c478bd9Sstevel@tonic-gate r_cnt = 0; 24107c478bd9Sstevel@tonic-gate } 24117c478bd9Sstevel@tonic-gate r[r_cnt].rng_va = vaddr; 24127c478bd9Sstevel@tonic-gate r[r_cnt].rng_cnt = 0; 24137c478bd9Sstevel@tonic-gate r[r_cnt].rng_level = ht->ht_level; 24147c478bd9Sstevel@tonic-gate ++r_cnt; 24157c478bd9Sstevel@tonic-gate } 24167c478bd9Sstevel@tonic-gate 24177c478bd9Sstevel@tonic-gate /* 24187c478bd9Sstevel@tonic-gate * Unload one mapping from the page tables. 24197c478bd9Sstevel@tonic-gate */ 24207c478bd9Sstevel@tonic-gate entry = htable_va2entry(vaddr, ht); 24217c478bd9Sstevel@tonic-gate hat_pte_unmap(ht, entry, flags, old_pte, NULL); 24227c478bd9Sstevel@tonic-gate ASSERT(ht->ht_level <= mmu.max_page_level); 24237c478bd9Sstevel@tonic-gate vaddr += LEVEL_SIZE(ht->ht_level); 2424aa2ed9e5Sjosephb contig_va = vaddr; 24257c478bd9Sstevel@tonic-gate ++r[r_cnt - 1].rng_cnt; 24267c478bd9Sstevel@tonic-gate } 24277c478bd9Sstevel@tonic-gate if (ht) 24287c478bd9Sstevel@tonic-gate htable_release(ht); 24297c478bd9Sstevel@tonic-gate 24307c478bd9Sstevel@tonic-gate /* 24317c478bd9Sstevel@tonic-gate * handle last range for callbacks 24327c478bd9Sstevel@tonic-gate */ 24337c478bd9Sstevel@tonic-gate if (r_cnt > 0) 24347c478bd9Sstevel@tonic-gate handle_ranges(cb, r_cnt, r); 2435843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 24367c478bd9Sstevel@tonic-gate } 24377c478bd9Sstevel@tonic-gate 24387c478bd9Sstevel@tonic-gate /* 24397c478bd9Sstevel@tonic-gate * synchronize mapping with software data structures 24407c478bd9Sstevel@tonic-gate * 24417c478bd9Sstevel@tonic-gate * This interface is currently only used by the working set monitor 24427c478bd9Sstevel@tonic-gate * driver. 24437c478bd9Sstevel@tonic-gate */ 24447c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 24457c478bd9Sstevel@tonic-gate void 24467c478bd9Sstevel@tonic-gate hat_sync(hat_t *hat, caddr_t addr, size_t len, uint_t flags) 24477c478bd9Sstevel@tonic-gate { 24487c478bd9Sstevel@tonic-gate uintptr_t vaddr = (uintptr_t)addr; 24497c478bd9Sstevel@tonic-gate uintptr_t eaddr = vaddr + len; 24507c478bd9Sstevel@tonic-gate htable_t *ht = NULL; 24517c478bd9Sstevel@tonic-gate uint_t entry; 24527c478bd9Sstevel@tonic-gate x86pte_t pte; 24537c478bd9Sstevel@tonic-gate x86pte_t save_pte; 24547c478bd9Sstevel@tonic-gate x86pte_t new; 24557c478bd9Sstevel@tonic-gate page_t *pp; 24567c478bd9Sstevel@tonic-gate 24577c478bd9Sstevel@tonic-gate ASSERT(!IN_VA_HOLE(vaddr)); 24587c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(vaddr)); 24597c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(eaddr)); 2460ae115bc7Smrj ASSERT(hat == kas.a_hat || eaddr <= _userlimit); 24617c478bd9Sstevel@tonic-gate 2462843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 24637c478bd9Sstevel@tonic-gate for (; vaddr < eaddr; vaddr += LEVEL_SIZE(ht->ht_level)) { 24647c478bd9Sstevel@tonic-gate try_again: 24657c478bd9Sstevel@tonic-gate pte = htable_walk(hat, &ht, &vaddr, eaddr); 24667c478bd9Sstevel@tonic-gate if (ht == NULL) 24677c478bd9Sstevel@tonic-gate break; 24687c478bd9Sstevel@tonic-gate entry = htable_va2entry(vaddr, ht); 24697c478bd9Sstevel@tonic-gate 2470ae115bc7Smrj if (PTE_GET(pte, PT_SOFTWARE) >= PT_NOSYNC || 24717c478bd9Sstevel@tonic-gate PTE_GET(pte, PT_REF | PT_MOD) == 0) 24727c478bd9Sstevel@tonic-gate continue; 24737c478bd9Sstevel@tonic-gate 24747c478bd9Sstevel@tonic-gate /* 24757c478bd9Sstevel@tonic-gate * We need to acquire the mapping list lock to protect 24767c478bd9Sstevel@tonic-gate * against hat_pageunload(), hat_unload(), etc. 24777c478bd9Sstevel@tonic-gate */ 24787c478bd9Sstevel@tonic-gate pp = page_numtopp_nolock(PTE2PFN(pte, ht->ht_level)); 24797c478bd9Sstevel@tonic-gate if (pp == NULL) 24807c478bd9Sstevel@tonic-gate break; 24817c478bd9Sstevel@tonic-gate x86_hm_enter(pp); 24827c478bd9Sstevel@tonic-gate save_pte = pte; 24837c478bd9Sstevel@tonic-gate pte = x86pte_get(ht, entry); 24847c478bd9Sstevel@tonic-gate if (pte != save_pte) { 24857c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 24867c478bd9Sstevel@tonic-gate goto try_again; 24877c478bd9Sstevel@tonic-gate } 2488ae115bc7Smrj if (PTE_GET(pte, PT_SOFTWARE) >= PT_NOSYNC || 24897c478bd9Sstevel@tonic-gate PTE_GET(pte, PT_REF | PT_MOD) == 0) { 24907c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 24917c478bd9Sstevel@tonic-gate continue; 24927c478bd9Sstevel@tonic-gate } 24937c478bd9Sstevel@tonic-gate 24947c478bd9Sstevel@tonic-gate /* 24957c478bd9Sstevel@tonic-gate * Need to clear ref or mod bits. We may compete with 24967c478bd9Sstevel@tonic-gate * hardware updating the R/M bits and have to try again. 24977c478bd9Sstevel@tonic-gate */ 24987c478bd9Sstevel@tonic-gate if (flags == HAT_SYNC_ZERORM) { 24997c478bd9Sstevel@tonic-gate new = pte; 25007c478bd9Sstevel@tonic-gate PTE_CLR(new, PT_REF | PT_MOD); 25017c478bd9Sstevel@tonic-gate pte = hati_update_pte(ht, entry, pte, new); 25027c478bd9Sstevel@tonic-gate if (pte != 0) { 25037c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 25047c478bd9Sstevel@tonic-gate goto try_again; 25057c478bd9Sstevel@tonic-gate } 25067c478bd9Sstevel@tonic-gate } else { 25077c478bd9Sstevel@tonic-gate /* 25087c478bd9Sstevel@tonic-gate * sync the PTE to the page_t 25097c478bd9Sstevel@tonic-gate */ 25107c478bd9Sstevel@tonic-gate hati_sync_pte_to_page(pp, save_pte, ht->ht_level); 25117c478bd9Sstevel@tonic-gate } 25127c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 25137c478bd9Sstevel@tonic-gate } 25147c478bd9Sstevel@tonic-gate if (ht) 25157c478bd9Sstevel@tonic-gate htable_release(ht); 2516843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 25177c478bd9Sstevel@tonic-gate } 25187c478bd9Sstevel@tonic-gate 25197c478bd9Sstevel@tonic-gate /* 25207c478bd9Sstevel@tonic-gate * void hat_map(hat, addr, len, flags) 25217c478bd9Sstevel@tonic-gate */ 25227c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 25237c478bd9Sstevel@tonic-gate void 25247c478bd9Sstevel@tonic-gate hat_map(hat_t *hat, caddr_t addr, size_t len, uint_t flags) 25257c478bd9Sstevel@tonic-gate { 25267c478bd9Sstevel@tonic-gate /* does nothing */ 25277c478bd9Sstevel@tonic-gate } 25287c478bd9Sstevel@tonic-gate 25297c478bd9Sstevel@tonic-gate /* 25307c478bd9Sstevel@tonic-gate * uint_t hat_getattr(hat, addr, *attr) 25317c478bd9Sstevel@tonic-gate * returns attr for <hat,addr> in *attr. returns 0 if there was a 25327c478bd9Sstevel@tonic-gate * mapping and *attr is valid, nonzero if there was no mapping and 25337c478bd9Sstevel@tonic-gate * *attr is not valid. 25347c478bd9Sstevel@tonic-gate */ 25357c478bd9Sstevel@tonic-gate uint_t 25367c478bd9Sstevel@tonic-gate hat_getattr(hat_t *hat, caddr_t addr, uint_t *attr) 25377c478bd9Sstevel@tonic-gate { 25387c478bd9Sstevel@tonic-gate uintptr_t vaddr = ALIGN2PAGE(addr); 25397c478bd9Sstevel@tonic-gate htable_t *ht = NULL; 25407c478bd9Sstevel@tonic-gate x86pte_t pte; 25417c478bd9Sstevel@tonic-gate 2542ae115bc7Smrj ASSERT(hat == kas.a_hat || vaddr <= _userlimit); 25437c478bd9Sstevel@tonic-gate 25447c478bd9Sstevel@tonic-gate if (IN_VA_HOLE(vaddr)) 25457c478bd9Sstevel@tonic-gate return ((uint_t)-1); 25467c478bd9Sstevel@tonic-gate 2547ae115bc7Smrj ht = htable_getpte(hat, vaddr, NULL, &pte, mmu.max_page_level); 25487c478bd9Sstevel@tonic-gate if (ht == NULL) 25497c478bd9Sstevel@tonic-gate return ((uint_t)-1); 25507c478bd9Sstevel@tonic-gate 25517c478bd9Sstevel@tonic-gate if (!PTE_ISVALID(pte) || !PTE_ISPAGE(pte, ht->ht_level)) { 25527c478bd9Sstevel@tonic-gate htable_release(ht); 25537c478bd9Sstevel@tonic-gate return ((uint_t)-1); 25547c478bd9Sstevel@tonic-gate } 25557c478bd9Sstevel@tonic-gate 25567c478bd9Sstevel@tonic-gate *attr = PROT_READ; 25577c478bd9Sstevel@tonic-gate if (PTE_GET(pte, PT_WRITABLE)) 25587c478bd9Sstevel@tonic-gate *attr |= PROT_WRITE; 25597c478bd9Sstevel@tonic-gate if (PTE_GET(pte, PT_USER)) 25607c478bd9Sstevel@tonic-gate *attr |= PROT_USER; 25617c478bd9Sstevel@tonic-gate if (!PTE_GET(pte, mmu.pt_nx)) 25627c478bd9Sstevel@tonic-gate *attr |= PROT_EXEC; 2563ae115bc7Smrj if (PTE_GET(pte, PT_SOFTWARE) >= PT_NOSYNC) 25647c478bd9Sstevel@tonic-gate *attr |= HAT_NOSYNC; 25657c478bd9Sstevel@tonic-gate htable_release(ht); 25667c478bd9Sstevel@tonic-gate return (0); 25677c478bd9Sstevel@tonic-gate } 25687c478bd9Sstevel@tonic-gate 25697c478bd9Sstevel@tonic-gate /* 25707c478bd9Sstevel@tonic-gate * hat_updateattr() applies the given attribute change to an existing mapping 25717c478bd9Sstevel@tonic-gate */ 25727c478bd9Sstevel@tonic-gate #define HAT_LOAD_ATTR 1 25737c478bd9Sstevel@tonic-gate #define HAT_SET_ATTR 2 25747c478bd9Sstevel@tonic-gate #define HAT_CLR_ATTR 3 25757c478bd9Sstevel@tonic-gate 25767c478bd9Sstevel@tonic-gate static void 25777c478bd9Sstevel@tonic-gate hat_updateattr(hat_t *hat, caddr_t addr, size_t len, uint_t attr, int what) 25787c478bd9Sstevel@tonic-gate { 25797c478bd9Sstevel@tonic-gate uintptr_t vaddr = (uintptr_t)addr; 25807c478bd9Sstevel@tonic-gate uintptr_t eaddr = (uintptr_t)addr + len; 25817c478bd9Sstevel@tonic-gate htable_t *ht = NULL; 25827c478bd9Sstevel@tonic-gate uint_t entry; 25837c478bd9Sstevel@tonic-gate x86pte_t oldpte, newpte; 25847c478bd9Sstevel@tonic-gate page_t *pp; 25857c478bd9Sstevel@tonic-gate 2586843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 25877c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(vaddr)); 25887c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(eaddr)); 25897c478bd9Sstevel@tonic-gate ASSERT(hat == kas.a_hat || 25907c478bd9Sstevel@tonic-gate AS_LOCK_HELD(hat->hat_as, &hat->hat_as->a_lock)); 25917c478bd9Sstevel@tonic-gate for (; vaddr < eaddr; vaddr += LEVEL_SIZE(ht->ht_level)) { 25927c478bd9Sstevel@tonic-gate try_again: 25937c478bd9Sstevel@tonic-gate oldpte = htable_walk(hat, &ht, &vaddr, eaddr); 25947c478bd9Sstevel@tonic-gate if (ht == NULL) 25957c478bd9Sstevel@tonic-gate break; 2596ae115bc7Smrj if (PTE_GET(oldpte, PT_SOFTWARE) >= PT_NOCONSIST) 25977c478bd9Sstevel@tonic-gate continue; 25987c478bd9Sstevel@tonic-gate 25997c478bd9Sstevel@tonic-gate pp = page_numtopp_nolock(PTE2PFN(oldpte, ht->ht_level)); 26007c478bd9Sstevel@tonic-gate if (pp == NULL) 26017c478bd9Sstevel@tonic-gate continue; 26027c478bd9Sstevel@tonic-gate x86_hm_enter(pp); 26037c478bd9Sstevel@tonic-gate 26047c478bd9Sstevel@tonic-gate newpte = oldpte; 26057c478bd9Sstevel@tonic-gate /* 26067c478bd9Sstevel@tonic-gate * We found a page table entry in the desired range, 26077c478bd9Sstevel@tonic-gate * figure out the new attributes. 26087c478bd9Sstevel@tonic-gate */ 26097c478bd9Sstevel@tonic-gate if (what == HAT_SET_ATTR || what == HAT_LOAD_ATTR) { 26107c478bd9Sstevel@tonic-gate if ((attr & PROT_WRITE) && 26117c478bd9Sstevel@tonic-gate !PTE_GET(oldpte, PT_WRITABLE)) 26127c478bd9Sstevel@tonic-gate newpte |= PT_WRITABLE; 26137c478bd9Sstevel@tonic-gate 2614ae115bc7Smrj if ((attr & HAT_NOSYNC) && 2615ae115bc7Smrj PTE_GET(oldpte, PT_SOFTWARE) < PT_NOSYNC) 26167c478bd9Sstevel@tonic-gate newpte |= PT_NOSYNC; 26177c478bd9Sstevel@tonic-gate 26187c478bd9Sstevel@tonic-gate if ((attr & PROT_EXEC) && PTE_GET(oldpte, mmu.pt_nx)) 26197c478bd9Sstevel@tonic-gate newpte &= ~mmu.pt_nx; 26207c478bd9Sstevel@tonic-gate } 26217c478bd9Sstevel@tonic-gate 26227c478bd9Sstevel@tonic-gate if (what == HAT_LOAD_ATTR) { 26237c478bd9Sstevel@tonic-gate if (!(attr & PROT_WRITE) && 26247c478bd9Sstevel@tonic-gate PTE_GET(oldpte, PT_WRITABLE)) 26257c478bd9Sstevel@tonic-gate newpte &= ~PT_WRITABLE; 26267c478bd9Sstevel@tonic-gate 2627ae115bc7Smrj if (!(attr & HAT_NOSYNC) && 2628ae115bc7Smrj PTE_GET(oldpte, PT_SOFTWARE) >= PT_NOSYNC) 2629ae115bc7Smrj newpte &= ~PT_SOFTWARE; 26307c478bd9Sstevel@tonic-gate 26317c478bd9Sstevel@tonic-gate if (!(attr & PROT_EXEC) && !PTE_GET(oldpte, mmu.pt_nx)) 26327c478bd9Sstevel@tonic-gate newpte |= mmu.pt_nx; 26337c478bd9Sstevel@tonic-gate } 26347c478bd9Sstevel@tonic-gate 26357c478bd9Sstevel@tonic-gate if (what == HAT_CLR_ATTR) { 26367c478bd9Sstevel@tonic-gate if ((attr & PROT_WRITE) && PTE_GET(oldpte, PT_WRITABLE)) 26377c478bd9Sstevel@tonic-gate newpte &= ~PT_WRITABLE; 26387c478bd9Sstevel@tonic-gate 2639ae115bc7Smrj if ((attr & HAT_NOSYNC) && 2640ae115bc7Smrj PTE_GET(oldpte, PT_SOFTWARE) >= PT_NOSYNC) 2641ae115bc7Smrj newpte &= ~PT_SOFTWARE; 26427c478bd9Sstevel@tonic-gate 26437c478bd9Sstevel@tonic-gate if ((attr & PROT_EXEC) && !PTE_GET(oldpte, mmu.pt_nx)) 26447c478bd9Sstevel@tonic-gate newpte |= mmu.pt_nx; 26457c478bd9Sstevel@tonic-gate } 26467c478bd9Sstevel@tonic-gate 2647ae115bc7Smrj /* 2648ae115bc7Smrj * Ensure NOSYNC/NOCONSIST mappings have REF and MOD set. 2649ae115bc7Smrj * x86pte_set() depends on this. 2650ae115bc7Smrj */ 2651ae115bc7Smrj if (PTE_GET(newpte, PT_SOFTWARE) >= PT_NOSYNC) 2652ae115bc7Smrj newpte |= PT_REF | PT_MOD; 2653ae115bc7Smrj 26547c478bd9Sstevel@tonic-gate /* 26557c478bd9Sstevel@tonic-gate * what about PROT_READ or others? this code only handles: 26567c478bd9Sstevel@tonic-gate * EXEC, WRITE, NOSYNC 26577c478bd9Sstevel@tonic-gate */ 26587c478bd9Sstevel@tonic-gate 26597c478bd9Sstevel@tonic-gate /* 26607c478bd9Sstevel@tonic-gate * If new PTE really changed, update the table. 26617c478bd9Sstevel@tonic-gate */ 26627c478bd9Sstevel@tonic-gate if (newpte != oldpte) { 26637c478bd9Sstevel@tonic-gate entry = htable_va2entry(vaddr, ht); 26647c478bd9Sstevel@tonic-gate oldpte = hati_update_pte(ht, entry, oldpte, newpte); 26657c478bd9Sstevel@tonic-gate if (oldpte != 0) { 26667c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 26677c478bd9Sstevel@tonic-gate goto try_again; 26687c478bd9Sstevel@tonic-gate } 26697c478bd9Sstevel@tonic-gate } 26707c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 26717c478bd9Sstevel@tonic-gate } 26727c478bd9Sstevel@tonic-gate if (ht) 26737c478bd9Sstevel@tonic-gate htable_release(ht); 2674843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 26757c478bd9Sstevel@tonic-gate } 26767c478bd9Sstevel@tonic-gate 26777c478bd9Sstevel@tonic-gate /* 26787c478bd9Sstevel@tonic-gate * Various wrappers for hat_updateattr() 26797c478bd9Sstevel@tonic-gate */ 26807c478bd9Sstevel@tonic-gate void 26817c478bd9Sstevel@tonic-gate hat_setattr(hat_t *hat, caddr_t addr, size_t len, uint_t attr) 26827c478bd9Sstevel@tonic-gate { 2683ae115bc7Smrj ASSERT(hat == kas.a_hat || (uintptr_t)addr + len <= _userlimit); 26847c478bd9Sstevel@tonic-gate hat_updateattr(hat, addr, len, attr, HAT_SET_ATTR); 26857c478bd9Sstevel@tonic-gate } 26867c478bd9Sstevel@tonic-gate 26877c478bd9Sstevel@tonic-gate void 26887c478bd9Sstevel@tonic-gate hat_clrattr(hat_t *hat, caddr_t addr, size_t len, uint_t attr) 26897c478bd9Sstevel@tonic-gate { 2690ae115bc7Smrj ASSERT(hat == kas.a_hat || (uintptr_t)addr + len <= _userlimit); 26917c478bd9Sstevel@tonic-gate hat_updateattr(hat, addr, len, attr, HAT_CLR_ATTR); 26927c478bd9Sstevel@tonic-gate } 26937c478bd9Sstevel@tonic-gate 26947c478bd9Sstevel@tonic-gate void 26957c478bd9Sstevel@tonic-gate hat_chgattr(hat_t *hat, caddr_t addr, size_t len, uint_t attr) 26967c478bd9Sstevel@tonic-gate { 2697ae115bc7Smrj ASSERT(hat == kas.a_hat || (uintptr_t)addr + len <= _userlimit); 26987c478bd9Sstevel@tonic-gate hat_updateattr(hat, addr, len, attr, HAT_LOAD_ATTR); 26997c478bd9Sstevel@tonic-gate } 27007c478bd9Sstevel@tonic-gate 27017c478bd9Sstevel@tonic-gate void 27027c478bd9Sstevel@tonic-gate hat_chgprot(hat_t *hat, caddr_t addr, size_t len, uint_t vprot) 27037c478bd9Sstevel@tonic-gate { 2704ae115bc7Smrj ASSERT(hat == kas.a_hat || (uintptr_t)addr + len <= _userlimit); 27057c478bd9Sstevel@tonic-gate hat_updateattr(hat, addr, len, vprot & HAT_PROT_MASK, HAT_LOAD_ATTR); 27067c478bd9Sstevel@tonic-gate } 27077c478bd9Sstevel@tonic-gate 27087c478bd9Sstevel@tonic-gate /* 27097c478bd9Sstevel@tonic-gate * size_t hat_getpagesize(hat, addr) 27107c478bd9Sstevel@tonic-gate * returns pagesize in bytes for <hat, addr>. returns -1 of there is 27117c478bd9Sstevel@tonic-gate * no mapping. This is an advisory call. 27127c478bd9Sstevel@tonic-gate */ 27137c478bd9Sstevel@tonic-gate ssize_t 27147c478bd9Sstevel@tonic-gate hat_getpagesize(hat_t *hat, caddr_t addr) 27157c478bd9Sstevel@tonic-gate { 27167c478bd9Sstevel@tonic-gate uintptr_t vaddr = ALIGN2PAGE(addr); 27177c478bd9Sstevel@tonic-gate htable_t *ht; 27187c478bd9Sstevel@tonic-gate size_t pagesize; 27197c478bd9Sstevel@tonic-gate 2720ae115bc7Smrj ASSERT(hat == kas.a_hat || vaddr <= _userlimit); 27217c478bd9Sstevel@tonic-gate if (IN_VA_HOLE(vaddr)) 27227c478bd9Sstevel@tonic-gate return (-1); 27237c478bd9Sstevel@tonic-gate ht = htable_getpage(hat, vaddr, NULL); 27247c478bd9Sstevel@tonic-gate if (ht == NULL) 27257c478bd9Sstevel@tonic-gate return (-1); 27267c478bd9Sstevel@tonic-gate pagesize = LEVEL_SIZE(ht->ht_level); 27277c478bd9Sstevel@tonic-gate htable_release(ht); 27287c478bd9Sstevel@tonic-gate return (pagesize); 27297c478bd9Sstevel@tonic-gate } 27307c478bd9Sstevel@tonic-gate 27317c478bd9Sstevel@tonic-gate 27327c478bd9Sstevel@tonic-gate 27337c478bd9Sstevel@tonic-gate /* 27347c478bd9Sstevel@tonic-gate * pfn_t hat_getpfnum(hat, addr) 27357c478bd9Sstevel@tonic-gate * returns pfn for <hat, addr> or PFN_INVALID if mapping is invalid. 27367c478bd9Sstevel@tonic-gate */ 27377c478bd9Sstevel@tonic-gate pfn_t 27387c478bd9Sstevel@tonic-gate hat_getpfnum(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 uint_t entry; 27437c478bd9Sstevel@tonic-gate pfn_t pfn = PFN_INVALID; 27447c478bd9Sstevel@tonic-gate 2745ae115bc7Smrj ASSERT(hat == kas.a_hat || vaddr <= _userlimit); 27467c478bd9Sstevel@tonic-gate if (khat_running == 0) 2747ae115bc7Smrj return (PFN_INVALID); 27487c478bd9Sstevel@tonic-gate 27497c478bd9Sstevel@tonic-gate if (IN_VA_HOLE(vaddr)) 27507c478bd9Sstevel@tonic-gate return (PFN_INVALID); 27517c478bd9Sstevel@tonic-gate 2752843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 27537c478bd9Sstevel@tonic-gate /* 27547c478bd9Sstevel@tonic-gate * A very common use of hat_getpfnum() is from the DDI for kernel pages. 27557c478bd9Sstevel@tonic-gate * Use the kmap_ptes (which also covers the 32 bit heap) to speed 27567c478bd9Sstevel@tonic-gate * this up. 27577c478bd9Sstevel@tonic-gate */ 27587c478bd9Sstevel@tonic-gate if (mmu.kmap_addr <= vaddr && vaddr < mmu.kmap_eaddr) { 27597c478bd9Sstevel@tonic-gate x86pte_t pte; 2760ae115bc7Smrj pgcnt_t pg_index; 27617c478bd9Sstevel@tonic-gate 2762ae115bc7Smrj pg_index = mmu_btop(vaddr - mmu.kmap_addr); 2763ae115bc7Smrj pte = GET_PTE(PT_INDEX_PTR(mmu.kmap_ptes, pg_index)); 2764843e1988Sjohnlev if (PTE_ISVALID(pte)) 2765843e1988Sjohnlev /*LINTED [use of constant 0 causes a lint warning] */ 2766843e1988Sjohnlev pfn = PTE2PFN(pte, 0); 2767843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 2768843e1988Sjohnlev return (pfn); 27697c478bd9Sstevel@tonic-gate } 27707c478bd9Sstevel@tonic-gate 27717c478bd9Sstevel@tonic-gate ht = htable_getpage(hat, vaddr, &entry); 2772843e1988Sjohnlev if (ht == NULL) { 2773843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 27747c478bd9Sstevel@tonic-gate return (PFN_INVALID); 2775843e1988Sjohnlev } 27767c478bd9Sstevel@tonic-gate ASSERT(vaddr >= ht->ht_vaddr); 27777c478bd9Sstevel@tonic-gate ASSERT(vaddr <= HTABLE_LAST_PAGE(ht)); 27787c478bd9Sstevel@tonic-gate pfn = PTE2PFN(x86pte_get(ht, entry), ht->ht_level); 27797c478bd9Sstevel@tonic-gate if (ht->ht_level > 0) 27807c478bd9Sstevel@tonic-gate pfn += mmu_btop(vaddr & LEVEL_OFFSET(ht->ht_level)); 27817c478bd9Sstevel@tonic-gate htable_release(ht); 2782843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 27837c478bd9Sstevel@tonic-gate return (pfn); 27847c478bd9Sstevel@tonic-gate } 27857c478bd9Sstevel@tonic-gate 27867c478bd9Sstevel@tonic-gate /* 27877c478bd9Sstevel@tonic-gate * hat_getkpfnum() is an obsolete DDI routine, and its use is discouraged. 27887c478bd9Sstevel@tonic-gate * Use hat_getpfnum(kas.a_hat, ...) instead. 27897c478bd9Sstevel@tonic-gate * 27907c478bd9Sstevel@tonic-gate * We'd like to return PFN_INVALID if the mappings have underlying page_t's 27917c478bd9Sstevel@tonic-gate * but can't right now due to the fact that some software has grown to use 27927c478bd9Sstevel@tonic-gate * this interface incorrectly. So for now when the interface is misused, 27937c478bd9Sstevel@tonic-gate * return a warning to the user that in the future it won't work in the 27947c478bd9Sstevel@tonic-gate * way they're abusing it, and carry on. 27957c478bd9Sstevel@tonic-gate * 27967c478bd9Sstevel@tonic-gate * Note that hat_getkpfnum() is never supported on amd64. 27977c478bd9Sstevel@tonic-gate */ 27987c478bd9Sstevel@tonic-gate #if !defined(__amd64) 27997c478bd9Sstevel@tonic-gate pfn_t 28007c478bd9Sstevel@tonic-gate hat_getkpfnum(caddr_t addr) 28017c478bd9Sstevel@tonic-gate { 28027c478bd9Sstevel@tonic-gate pfn_t pfn; 28037c478bd9Sstevel@tonic-gate int badcaller = 0; 28047c478bd9Sstevel@tonic-gate 28057c478bd9Sstevel@tonic-gate if (khat_running == 0) 28067c478bd9Sstevel@tonic-gate panic("hat_getkpfnum(): called too early\n"); 28077c478bd9Sstevel@tonic-gate if ((uintptr_t)addr < kernelbase) 28087c478bd9Sstevel@tonic-gate return (PFN_INVALID); 28097c478bd9Sstevel@tonic-gate 2810843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 28117c478bd9Sstevel@tonic-gate if (segkpm && IS_KPM_ADDR(addr)) { 28127c478bd9Sstevel@tonic-gate badcaller = 1; 28137c478bd9Sstevel@tonic-gate pfn = hat_kpm_va2pfn(addr); 28147c478bd9Sstevel@tonic-gate } else { 28157c478bd9Sstevel@tonic-gate pfn = hat_getpfnum(kas.a_hat, addr); 28167c478bd9Sstevel@tonic-gate badcaller = pf_is_memory(pfn); 28177c478bd9Sstevel@tonic-gate } 28187c478bd9Sstevel@tonic-gate 28197c478bd9Sstevel@tonic-gate if (badcaller) 28207c478bd9Sstevel@tonic-gate hat_getkpfnum_badcall(caller()); 2821843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 28227c478bd9Sstevel@tonic-gate return (pfn); 28237c478bd9Sstevel@tonic-gate } 28247c478bd9Sstevel@tonic-gate #endif /* __amd64 */ 28257c478bd9Sstevel@tonic-gate 28267c478bd9Sstevel@tonic-gate /* 28277c478bd9Sstevel@tonic-gate * int hat_probe(hat, addr) 28287c478bd9Sstevel@tonic-gate * return 0 if no valid mapping is present. Faster version 28297c478bd9Sstevel@tonic-gate * of hat_getattr in certain architectures. 28307c478bd9Sstevel@tonic-gate */ 28317c478bd9Sstevel@tonic-gate int 28327c478bd9Sstevel@tonic-gate hat_probe(hat_t *hat, caddr_t addr) 28337c478bd9Sstevel@tonic-gate { 28347c478bd9Sstevel@tonic-gate uintptr_t vaddr = ALIGN2PAGE(addr); 28357c478bd9Sstevel@tonic-gate uint_t entry; 28367c478bd9Sstevel@tonic-gate htable_t *ht; 28377c478bd9Sstevel@tonic-gate pgcnt_t pg_off; 28387c478bd9Sstevel@tonic-gate 2839ae115bc7Smrj ASSERT(hat == kas.a_hat || vaddr <= _userlimit); 28407c478bd9Sstevel@tonic-gate ASSERT(hat == kas.a_hat || 28417c478bd9Sstevel@tonic-gate AS_LOCK_HELD(hat->hat_as, &hat->hat_as->a_lock)); 28427c478bd9Sstevel@tonic-gate if (IN_VA_HOLE(vaddr)) 28437c478bd9Sstevel@tonic-gate return (0); 28447c478bd9Sstevel@tonic-gate 28457c478bd9Sstevel@tonic-gate /* 28467c478bd9Sstevel@tonic-gate * Most common use of hat_probe is from segmap. We special case it 28477c478bd9Sstevel@tonic-gate * for performance. 28487c478bd9Sstevel@tonic-gate */ 28497c478bd9Sstevel@tonic-gate if (mmu.kmap_addr <= vaddr && vaddr < mmu.kmap_eaddr) { 28507c478bd9Sstevel@tonic-gate pg_off = mmu_btop(vaddr - mmu.kmap_addr); 28517c478bd9Sstevel@tonic-gate if (mmu.pae_hat) 28527c478bd9Sstevel@tonic-gate return (PTE_ISVALID(mmu.kmap_ptes[pg_off])); 28537c478bd9Sstevel@tonic-gate else 28547c478bd9Sstevel@tonic-gate return (PTE_ISVALID( 28557c478bd9Sstevel@tonic-gate ((x86pte32_t *)mmu.kmap_ptes)[pg_off])); 28567c478bd9Sstevel@tonic-gate } 28577c478bd9Sstevel@tonic-gate 28587c478bd9Sstevel@tonic-gate ht = htable_getpage(hat, vaddr, &entry); 28597c478bd9Sstevel@tonic-gate htable_release(ht); 2860843e1988Sjohnlev return (ht != NULL); 28617c478bd9Sstevel@tonic-gate } 28627c478bd9Sstevel@tonic-gate 28637c478bd9Sstevel@tonic-gate /* 2864250b7ff9Sjosephb * Find out if the segment for hat_share()/hat_unshare() is DISM or locked ISM. 2865250b7ff9Sjosephb */ 2866250b7ff9Sjosephb static int 2867250b7ff9Sjosephb is_it_dism(hat_t *hat, caddr_t va) 2868250b7ff9Sjosephb { 2869250b7ff9Sjosephb struct seg *seg; 2870250b7ff9Sjosephb struct shm_data *shmd; 2871250b7ff9Sjosephb struct spt_data *sptd; 2872250b7ff9Sjosephb 2873250b7ff9Sjosephb seg = as_findseg(hat->hat_as, va, 0); 2874250b7ff9Sjosephb ASSERT(seg != NULL); 2875250b7ff9Sjosephb ASSERT(seg->s_base <= va); 2876250b7ff9Sjosephb shmd = (struct shm_data *)seg->s_data; 2877250b7ff9Sjosephb ASSERT(shmd != NULL); 2878250b7ff9Sjosephb sptd = (struct spt_data *)shmd->shm_sptseg->s_data; 2879250b7ff9Sjosephb ASSERT(sptd != NULL); 2880250b7ff9Sjosephb if (sptd->spt_flags & SHM_PAGEABLE) 2881250b7ff9Sjosephb return (1); 2882250b7ff9Sjosephb return (0); 2883250b7ff9Sjosephb } 2884250b7ff9Sjosephb 2885250b7ff9Sjosephb /* 2886250b7ff9Sjosephb * Simple implementation of ISM. hat_share() is similar to hat_memload_array(), 28877c478bd9Sstevel@tonic-gate * except that we use the ism_hat's existing mappings to determine the pages 2888250b7ff9Sjosephb * and protections to use for this hat. If we find a full properly aligned 2889250b7ff9Sjosephb * and sized pagetable, we will attempt to share the pagetable itself. 28907c478bd9Sstevel@tonic-gate */ 28917c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 28927c478bd9Sstevel@tonic-gate int 28937c478bd9Sstevel@tonic-gate hat_share( 28947c478bd9Sstevel@tonic-gate hat_t *hat, 28957c478bd9Sstevel@tonic-gate caddr_t addr, 28967c478bd9Sstevel@tonic-gate hat_t *ism_hat, 28977c478bd9Sstevel@tonic-gate caddr_t src_addr, 28987c478bd9Sstevel@tonic-gate size_t len, /* almost useless value, see below.. */ 28997c478bd9Sstevel@tonic-gate uint_t ismszc) 29007c478bd9Sstevel@tonic-gate { 29017c478bd9Sstevel@tonic-gate uintptr_t vaddr_start = (uintptr_t)addr; 29027c478bd9Sstevel@tonic-gate uintptr_t vaddr; 29037c478bd9Sstevel@tonic-gate uintptr_t eaddr = vaddr_start + len; 29047c478bd9Sstevel@tonic-gate uintptr_t ism_addr_start = (uintptr_t)src_addr; 29057c478bd9Sstevel@tonic-gate uintptr_t ism_addr = ism_addr_start; 29067c478bd9Sstevel@tonic-gate uintptr_t e_ism_addr = ism_addr + len; 29077c478bd9Sstevel@tonic-gate htable_t *ism_ht = NULL; 29087c478bd9Sstevel@tonic-gate htable_t *ht; 29097c478bd9Sstevel@tonic-gate x86pte_t pte; 29107c478bd9Sstevel@tonic-gate page_t *pp; 29117c478bd9Sstevel@tonic-gate pfn_t pfn; 29127c478bd9Sstevel@tonic-gate level_t l; 29137c478bd9Sstevel@tonic-gate pgcnt_t pgcnt; 29147c478bd9Sstevel@tonic-gate uint_t prot; 2915250b7ff9Sjosephb int is_dism; 2916250b7ff9Sjosephb int flags; 29177c478bd9Sstevel@tonic-gate 29187c478bd9Sstevel@tonic-gate /* 29197c478bd9Sstevel@tonic-gate * We might be asked to share an empty DISM hat by as_dup() 29207c478bd9Sstevel@tonic-gate */ 29217c478bd9Sstevel@tonic-gate ASSERT(hat != kas.a_hat); 2922ae115bc7Smrj ASSERT(eaddr <= _userlimit); 29237c478bd9Sstevel@tonic-gate if (!(ism_hat->hat_flags & HAT_SHARED)) { 29247c478bd9Sstevel@tonic-gate ASSERT(hat_get_mapped_size(ism_hat) == 0); 29257c478bd9Sstevel@tonic-gate return (0); 29267c478bd9Sstevel@tonic-gate } 2927843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 29287c478bd9Sstevel@tonic-gate 29297c478bd9Sstevel@tonic-gate /* 29307c478bd9Sstevel@tonic-gate * The SPT segment driver often passes us a size larger than there are 29317c478bd9Sstevel@tonic-gate * valid mappings. That's because it rounds the segment size up to a 29327c478bd9Sstevel@tonic-gate * large pagesize, even if the actual memory mapped by ism_hat is less. 29337c478bd9Sstevel@tonic-gate */ 29347c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(vaddr_start)); 29357c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(ism_addr_start)); 29367c478bd9Sstevel@tonic-gate ASSERT(ism_hat->hat_flags & HAT_SHARED); 2937250b7ff9Sjosephb is_dism = is_it_dism(hat, addr); 29387c478bd9Sstevel@tonic-gate while (ism_addr < e_ism_addr) { 29397c478bd9Sstevel@tonic-gate /* 29407c478bd9Sstevel@tonic-gate * use htable_walk to get the next valid ISM mapping 29417c478bd9Sstevel@tonic-gate */ 29427c478bd9Sstevel@tonic-gate pte = htable_walk(ism_hat, &ism_ht, &ism_addr, e_ism_addr); 29437c478bd9Sstevel@tonic-gate if (ism_ht == NULL) 29447c478bd9Sstevel@tonic-gate break; 29457c478bd9Sstevel@tonic-gate 29467c478bd9Sstevel@tonic-gate /* 2947250b7ff9Sjosephb * First check to see if we already share the page table. 29487c478bd9Sstevel@tonic-gate */ 2949250b7ff9Sjosephb l = ism_ht->ht_level; 29507c478bd9Sstevel@tonic-gate vaddr = vaddr_start + (ism_addr - ism_addr_start); 2951250b7ff9Sjosephb ht = htable_lookup(hat, vaddr, l); 2952250b7ff9Sjosephb if (ht != NULL) { 2953250b7ff9Sjosephb if (ht->ht_flags & HTABLE_SHARED_PFN) 2954250b7ff9Sjosephb goto shared; 2955250b7ff9Sjosephb htable_release(ht); 2956250b7ff9Sjosephb goto not_shared; 29577c478bd9Sstevel@tonic-gate } 29587c478bd9Sstevel@tonic-gate 29597c478bd9Sstevel@tonic-gate /* 2960250b7ff9Sjosephb * Can't ever share top table. 29617c478bd9Sstevel@tonic-gate */ 2962250b7ff9Sjosephb if (l == mmu.max_level) 2963250b7ff9Sjosephb goto not_shared; 29647c478bd9Sstevel@tonic-gate 2965250b7ff9Sjosephb /* 2966250b7ff9Sjosephb * Avoid level mismatches later due to DISM faults. 2967250b7ff9Sjosephb */ 2968250b7ff9Sjosephb if (is_dism && l > 0) 2969250b7ff9Sjosephb goto not_shared; 29707c478bd9Sstevel@tonic-gate 2971250b7ff9Sjosephb /* 2972250b7ff9Sjosephb * addresses and lengths must align 2973250b7ff9Sjosephb * table must be fully populated 2974250b7ff9Sjosephb * no lower level page tables 2975250b7ff9Sjosephb */ 2976250b7ff9Sjosephb if (ism_addr != ism_ht->ht_vaddr || 2977250b7ff9Sjosephb (vaddr & LEVEL_OFFSET(l + 1)) != 0) 2978250b7ff9Sjosephb goto not_shared; 29797c478bd9Sstevel@tonic-gate 2980250b7ff9Sjosephb /* 2981250b7ff9Sjosephb * The range of address space must cover a full table. 2982250b7ff9Sjosephb */ 29831d03c31eSjohnlev if (e_ism_addr - ism_addr < LEVEL_SIZE(l + 1)) 2984250b7ff9Sjosephb goto not_shared; 29857c478bd9Sstevel@tonic-gate 2986250b7ff9Sjosephb /* 2987250b7ff9Sjosephb * All entries in the ISM page table must be leaf PTEs. 2988250b7ff9Sjosephb */ 2989250b7ff9Sjosephb if (l > 0) { 2990250b7ff9Sjosephb int e; 29917c478bd9Sstevel@tonic-gate 2992250b7ff9Sjosephb /* 2993250b7ff9Sjosephb * We know the 0th is from htable_walk() above. 2994250b7ff9Sjosephb */ 2995250b7ff9Sjosephb for (e = 1; e < HTABLE_NUM_PTES(ism_ht); ++e) { 2996250b7ff9Sjosephb x86pte_t pte; 2997250b7ff9Sjosephb pte = x86pte_get(ism_ht, e); 2998250b7ff9Sjosephb if (!PTE_ISPAGE(pte, l)) 2999250b7ff9Sjosephb goto not_shared; 30007c478bd9Sstevel@tonic-gate } 30017c478bd9Sstevel@tonic-gate } 30027c478bd9Sstevel@tonic-gate 3003250b7ff9Sjosephb /* 3004250b7ff9Sjosephb * share the page table 3005250b7ff9Sjosephb */ 3006250b7ff9Sjosephb ht = htable_create(hat, vaddr, l, ism_ht); 3007250b7ff9Sjosephb shared: 3008250b7ff9Sjosephb ASSERT(ht->ht_flags & HTABLE_SHARED_PFN); 3009250b7ff9Sjosephb ASSERT(ht->ht_shares == ism_ht); 3010250b7ff9Sjosephb hat->hat_ism_pgcnt += 3011250b7ff9Sjosephb (ism_ht->ht_valid_cnt - ht->ht_valid_cnt) << 3012250b7ff9Sjosephb (LEVEL_SHIFT(ht->ht_level) - MMU_PAGESHIFT); 3013250b7ff9Sjosephb ht->ht_valid_cnt = ism_ht->ht_valid_cnt; 3014250b7ff9Sjosephb htable_release(ht); 3015250b7ff9Sjosephb ism_addr = ism_ht->ht_vaddr + LEVEL_SIZE(l + 1); 3016250b7ff9Sjosephb htable_release(ism_ht); 3017250b7ff9Sjosephb ism_ht = NULL; 3018250b7ff9Sjosephb continue; 3019250b7ff9Sjosephb 3020250b7ff9Sjosephb not_shared: 30217c478bd9Sstevel@tonic-gate /* 30227c478bd9Sstevel@tonic-gate * Unable to share the page table. Instead we will 30237c478bd9Sstevel@tonic-gate * create new mappings from the values in the ISM mappings. 3024250b7ff9Sjosephb * Figure out what level size mappings to use; 3025250b7ff9Sjosephb */ 3026250b7ff9Sjosephb for (l = ism_ht->ht_level; l > 0; --l) { 3027250b7ff9Sjosephb if (LEVEL_SIZE(l) <= eaddr - vaddr && 3028250b7ff9Sjosephb (vaddr & LEVEL_OFFSET(l)) == 0) 3029250b7ff9Sjosephb break; 3030250b7ff9Sjosephb } 3031250b7ff9Sjosephb 3032250b7ff9Sjosephb /* 30337c478bd9Sstevel@tonic-gate * The ISM mapping might be larger than the share area, 3034250b7ff9Sjosephb * be careful to truncate it if needed. 30357c478bd9Sstevel@tonic-gate */ 30367c478bd9Sstevel@tonic-gate if (eaddr - vaddr >= LEVEL_SIZE(ism_ht->ht_level)) { 30377c478bd9Sstevel@tonic-gate pgcnt = mmu_btop(LEVEL_SIZE(ism_ht->ht_level)); 30387c478bd9Sstevel@tonic-gate } else { 30397c478bd9Sstevel@tonic-gate pgcnt = mmu_btop(eaddr - vaddr); 30407c478bd9Sstevel@tonic-gate l = 0; 30417c478bd9Sstevel@tonic-gate } 30427c478bd9Sstevel@tonic-gate 30437c478bd9Sstevel@tonic-gate pfn = PTE2PFN(pte, ism_ht->ht_level); 30447c478bd9Sstevel@tonic-gate ASSERT(pfn != PFN_INVALID); 30457c478bd9Sstevel@tonic-gate while (pgcnt > 0) { 30467c478bd9Sstevel@tonic-gate /* 30477c478bd9Sstevel@tonic-gate * Make a new pte for the PFN for this level. 30487c478bd9Sstevel@tonic-gate * Copy protections for the pte from the ISM pte. 30497c478bd9Sstevel@tonic-gate */ 30507c478bd9Sstevel@tonic-gate pp = page_numtopp_nolock(pfn); 30517c478bd9Sstevel@tonic-gate ASSERT(pp != NULL); 30527c478bd9Sstevel@tonic-gate 30537c478bd9Sstevel@tonic-gate prot = PROT_USER | PROT_READ | HAT_UNORDERED_OK; 30547c478bd9Sstevel@tonic-gate if (PTE_GET(pte, PT_WRITABLE)) 30557c478bd9Sstevel@tonic-gate prot |= PROT_WRITE; 30567c478bd9Sstevel@tonic-gate if (!PTE_GET(pte, PT_NX)) 30577c478bd9Sstevel@tonic-gate prot |= PROT_EXEC; 30587c478bd9Sstevel@tonic-gate 3059250b7ff9Sjosephb flags = HAT_LOAD; 3060250b7ff9Sjosephb if (!is_dism) 3061250b7ff9Sjosephb flags |= HAT_LOAD_LOCK | HAT_LOAD_NOCONSIST; 3062250b7ff9Sjosephb while (hati_load_common(hat, vaddr, pp, prot, flags, 3063ae115bc7Smrj l, pfn) != 0) { 3064ae115bc7Smrj if (l == 0) 3065ae115bc7Smrj panic("hati_load_common() failure"); 3066ae115bc7Smrj --l; 3067ae115bc7Smrj } 30687c478bd9Sstevel@tonic-gate 30697c478bd9Sstevel@tonic-gate vaddr += LEVEL_SIZE(l); 30707c478bd9Sstevel@tonic-gate ism_addr += LEVEL_SIZE(l); 30717c478bd9Sstevel@tonic-gate pfn += mmu_btop(LEVEL_SIZE(l)); 30727c478bd9Sstevel@tonic-gate pgcnt -= mmu_btop(LEVEL_SIZE(l)); 30737c478bd9Sstevel@tonic-gate } 30747c478bd9Sstevel@tonic-gate } 30757c478bd9Sstevel@tonic-gate if (ism_ht != NULL) 30767c478bd9Sstevel@tonic-gate htable_release(ism_ht); 3077843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 30787c478bd9Sstevel@tonic-gate return (0); 30797c478bd9Sstevel@tonic-gate } 30807c478bd9Sstevel@tonic-gate 30817c478bd9Sstevel@tonic-gate 30827c478bd9Sstevel@tonic-gate /* 30837c478bd9Sstevel@tonic-gate * hat_unshare() is similar to hat_unload_callback(), but 30847c478bd9Sstevel@tonic-gate * we have to look for empty shared pagetables. Note that 30857c478bd9Sstevel@tonic-gate * hat_unshare() is always invoked against an entire segment. 30867c478bd9Sstevel@tonic-gate */ 30877c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 30887c478bd9Sstevel@tonic-gate void 30897c478bd9Sstevel@tonic-gate hat_unshare(hat_t *hat, caddr_t addr, size_t len, uint_t ismszc) 30907c478bd9Sstevel@tonic-gate { 30917173d045Sjosephb uint64_t vaddr = (uintptr_t)addr; 30927c478bd9Sstevel@tonic-gate uintptr_t eaddr = vaddr + len; 30937c478bd9Sstevel@tonic-gate htable_t *ht = NULL; 30947c478bd9Sstevel@tonic-gate uint_t need_demaps = 0; 3095250b7ff9Sjosephb int flags = HAT_UNLOAD_UNMAP; 3096250b7ff9Sjosephb level_t l; 30977c478bd9Sstevel@tonic-gate 30987c478bd9Sstevel@tonic-gate ASSERT(hat != kas.a_hat); 3099ae115bc7Smrj ASSERT(eaddr <= _userlimit); 31007c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(vaddr)); 31017c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(eaddr)); 3102843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 31037c478bd9Sstevel@tonic-gate 31047c478bd9Sstevel@tonic-gate /* 31057c478bd9Sstevel@tonic-gate * First go through and remove any shared pagetables. 31067c478bd9Sstevel@tonic-gate * 3107ae115bc7Smrj * Note that it's ok to delay the TLB shootdown till the entire range is 31087c478bd9Sstevel@tonic-gate * finished, because if hat_pageunload() were to unload a shared 3109ae115bc7Smrj * pagetable page, its hat_tlb_inval() will do a global TLB invalidate. 31107c478bd9Sstevel@tonic-gate */ 3111250b7ff9Sjosephb l = mmu.max_page_level; 3112250b7ff9Sjosephb if (l == mmu.max_level) 3113250b7ff9Sjosephb --l; 3114250b7ff9Sjosephb for (; l >= 0; --l) { 3115250b7ff9Sjosephb for (vaddr = (uintptr_t)addr; vaddr < eaddr; 3116250b7ff9Sjosephb vaddr = (vaddr & LEVEL_MASK(l + 1)) + LEVEL_SIZE(l + 1)) { 3117250b7ff9Sjosephb ASSERT(!IN_VA_HOLE(vaddr)); 3118250b7ff9Sjosephb /* 3119250b7ff9Sjosephb * find a pagetable that maps the current address 3120250b7ff9Sjosephb */ 3121250b7ff9Sjosephb ht = htable_lookup(hat, vaddr, l); 3122250b7ff9Sjosephb if (ht == NULL) 3123250b7ff9Sjosephb continue; 31247c478bd9Sstevel@tonic-gate if (ht->ht_flags & HTABLE_SHARED_PFN) { 31257c478bd9Sstevel@tonic-gate /* 3126250b7ff9Sjosephb * clear page count, set valid_cnt to 0, 3127250b7ff9Sjosephb * let htable_release() finish the job 31287c478bd9Sstevel@tonic-gate */ 3129250b7ff9Sjosephb hat->hat_ism_pgcnt -= ht->ht_valid_cnt << 3130250b7ff9Sjosephb (LEVEL_SHIFT(ht->ht_level) - MMU_PAGESHIFT); 31317c478bd9Sstevel@tonic-gate ht->ht_valid_cnt = 0; 31327c478bd9Sstevel@tonic-gate need_demaps = 1; 31337c478bd9Sstevel@tonic-gate } 31347c478bd9Sstevel@tonic-gate htable_release(ht); 31357c478bd9Sstevel@tonic-gate } 31367c478bd9Sstevel@tonic-gate } 31377c478bd9Sstevel@tonic-gate 31387c478bd9Sstevel@tonic-gate /* 31397c478bd9Sstevel@tonic-gate * flush the TLBs - since we're probably dealing with MANY mappings 31407c478bd9Sstevel@tonic-gate * we do just one CR3 reload. 31417c478bd9Sstevel@tonic-gate */ 31427c478bd9Sstevel@tonic-gate if (!(hat->hat_flags & HAT_FREEING) && need_demaps) 3143ae115bc7Smrj hat_tlb_inval(hat, DEMAP_ALL_ADDR); 31447c478bd9Sstevel@tonic-gate 31457c478bd9Sstevel@tonic-gate /* 31467c478bd9Sstevel@tonic-gate * Now go back and clean up any unaligned mappings that 31477c478bd9Sstevel@tonic-gate * couldn't share pagetables. 31487c478bd9Sstevel@tonic-gate */ 3149250b7ff9Sjosephb if (!is_it_dism(hat, addr)) 3150250b7ff9Sjosephb flags |= HAT_UNLOAD_UNLOCK; 3151250b7ff9Sjosephb hat_unload(hat, addr, len, flags); 3152843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 31537c478bd9Sstevel@tonic-gate } 31547c478bd9Sstevel@tonic-gate 31557c478bd9Sstevel@tonic-gate 31567c478bd9Sstevel@tonic-gate /* 31577c478bd9Sstevel@tonic-gate * hat_reserve() does nothing 31587c478bd9Sstevel@tonic-gate */ 31597c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 31607c478bd9Sstevel@tonic-gate void 31617c478bd9Sstevel@tonic-gate hat_reserve(struct as *as, caddr_t addr, size_t len) 31627c478bd9Sstevel@tonic-gate { 31637c478bd9Sstevel@tonic-gate } 31647c478bd9Sstevel@tonic-gate 31657c478bd9Sstevel@tonic-gate 31667c478bd9Sstevel@tonic-gate /* 31677c478bd9Sstevel@tonic-gate * Called when all mappings to a page should have write permission removed. 3168*21584dbcSPavel Tatashin * Mostly stolen from hat_pagesync() 31697c478bd9Sstevel@tonic-gate */ 31707c478bd9Sstevel@tonic-gate static void 31717c478bd9Sstevel@tonic-gate hati_page_clrwrt(struct page *pp) 31727c478bd9Sstevel@tonic-gate { 31737c478bd9Sstevel@tonic-gate hment_t *hm = NULL; 31747c478bd9Sstevel@tonic-gate htable_t *ht; 31757c478bd9Sstevel@tonic-gate uint_t entry; 31767c478bd9Sstevel@tonic-gate x86pte_t old; 31777c478bd9Sstevel@tonic-gate x86pte_t new; 31787c478bd9Sstevel@tonic-gate uint_t pszc = 0; 31797c478bd9Sstevel@tonic-gate 3180843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 31817c478bd9Sstevel@tonic-gate next_size: 31827c478bd9Sstevel@tonic-gate /* 31837c478bd9Sstevel@tonic-gate * walk thru the mapping list clearing write permission 31847c478bd9Sstevel@tonic-gate */ 31857c478bd9Sstevel@tonic-gate x86_hm_enter(pp); 31867c478bd9Sstevel@tonic-gate while ((hm = hment_walk(pp, &ht, &entry, hm)) != NULL) { 31877c478bd9Sstevel@tonic-gate if (ht->ht_level < pszc) 31887c478bd9Sstevel@tonic-gate continue; 31897c478bd9Sstevel@tonic-gate old = x86pte_get(ht, entry); 31907c478bd9Sstevel@tonic-gate 31917c478bd9Sstevel@tonic-gate for (;;) { 31927c478bd9Sstevel@tonic-gate /* 31937c478bd9Sstevel@tonic-gate * Is this mapping of interest? 31947c478bd9Sstevel@tonic-gate */ 31957c478bd9Sstevel@tonic-gate if (PTE2PFN(old, ht->ht_level) != pp->p_pagenum || 31967c478bd9Sstevel@tonic-gate PTE_GET(old, PT_WRITABLE) == 0) 31977c478bd9Sstevel@tonic-gate break; 31987c478bd9Sstevel@tonic-gate 31997c478bd9Sstevel@tonic-gate /* 32007c478bd9Sstevel@tonic-gate * Clear ref/mod writable bits. This requires cross 32017c478bd9Sstevel@tonic-gate * calls to ensure any executing TLBs see cleared bits. 32027c478bd9Sstevel@tonic-gate */ 32037c478bd9Sstevel@tonic-gate new = old; 32047c478bd9Sstevel@tonic-gate PTE_CLR(new, PT_REF | PT_MOD | PT_WRITABLE); 32057c478bd9Sstevel@tonic-gate old = hati_update_pte(ht, entry, old, new); 32067c478bd9Sstevel@tonic-gate if (old != 0) 32077c478bd9Sstevel@tonic-gate continue; 32087c478bd9Sstevel@tonic-gate 32097c478bd9Sstevel@tonic-gate break; 32107c478bd9Sstevel@tonic-gate } 32117c478bd9Sstevel@tonic-gate } 32127c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 32137c478bd9Sstevel@tonic-gate while (pszc < pp->p_szc) { 32147c478bd9Sstevel@tonic-gate page_t *tpp; 32157c478bd9Sstevel@tonic-gate pszc++; 32167c478bd9Sstevel@tonic-gate tpp = PP_GROUPLEADER(pp, pszc); 32177c478bd9Sstevel@tonic-gate if (pp != tpp) { 32187c478bd9Sstevel@tonic-gate pp = tpp; 32197c478bd9Sstevel@tonic-gate goto next_size; 32207c478bd9Sstevel@tonic-gate } 32217c478bd9Sstevel@tonic-gate } 3222843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 32237c478bd9Sstevel@tonic-gate } 32247c478bd9Sstevel@tonic-gate 32257c478bd9Sstevel@tonic-gate /* 32267c478bd9Sstevel@tonic-gate * void hat_page_setattr(pp, flag) 32277c478bd9Sstevel@tonic-gate * void hat_page_clrattr(pp, flag) 32287c478bd9Sstevel@tonic-gate * used to set/clr ref/mod bits. 32297c478bd9Sstevel@tonic-gate */ 32307c478bd9Sstevel@tonic-gate void 32317c478bd9Sstevel@tonic-gate hat_page_setattr(struct page *pp, uint_t flag) 32327c478bd9Sstevel@tonic-gate { 32337c478bd9Sstevel@tonic-gate vnode_t *vp = pp->p_vnode; 32347c478bd9Sstevel@tonic-gate kmutex_t *vphm = NULL; 32357c478bd9Sstevel@tonic-gate page_t **listp; 3236d9615970Sqiao int noshuffle; 3237d9615970Sqiao 3238d9615970Sqiao noshuffle = flag & P_NSH; 3239d9615970Sqiao flag &= ~P_NSH; 32407c478bd9Sstevel@tonic-gate 32417c478bd9Sstevel@tonic-gate if (PP_GETRM(pp, flag) == flag) 32427c478bd9Sstevel@tonic-gate return; 32437c478bd9Sstevel@tonic-gate 3244d9615970Sqiao if ((flag & P_MOD) != 0 && vp != NULL && IS_VMODSORT(vp) && 3245d9615970Sqiao !noshuffle) { 32467c478bd9Sstevel@tonic-gate vphm = page_vnode_mutex(vp); 32477c478bd9Sstevel@tonic-gate mutex_enter(vphm); 32487c478bd9Sstevel@tonic-gate } 32497c478bd9Sstevel@tonic-gate 32507c478bd9Sstevel@tonic-gate PP_SETRM(pp, flag); 32517c478bd9Sstevel@tonic-gate 32527c478bd9Sstevel@tonic-gate if (vphm != NULL) { 32537c478bd9Sstevel@tonic-gate 32547c478bd9Sstevel@tonic-gate /* 32557c478bd9Sstevel@tonic-gate * Some File Systems examine v_pages for NULL w/o 32567c478bd9Sstevel@tonic-gate * grabbing the vphm mutex. Must not let it become NULL when 32577c478bd9Sstevel@tonic-gate * pp is the only page on the list. 32587c478bd9Sstevel@tonic-gate */ 32597c478bd9Sstevel@tonic-gate if (pp->p_vpnext != pp) { 32607c478bd9Sstevel@tonic-gate page_vpsub(&vp->v_pages, pp); 32617c478bd9Sstevel@tonic-gate if (vp->v_pages != NULL) 32627c478bd9Sstevel@tonic-gate listp = &vp->v_pages->p_vpprev->p_vpnext; 32637c478bd9Sstevel@tonic-gate else 32647c478bd9Sstevel@tonic-gate listp = &vp->v_pages; 32657c478bd9Sstevel@tonic-gate page_vpadd(listp, pp); 32667c478bd9Sstevel@tonic-gate } 32677c478bd9Sstevel@tonic-gate mutex_exit(vphm); 32687c478bd9Sstevel@tonic-gate } 32697c478bd9Sstevel@tonic-gate } 32707c478bd9Sstevel@tonic-gate 32717c478bd9Sstevel@tonic-gate void 32727c478bd9Sstevel@tonic-gate hat_page_clrattr(struct page *pp, uint_t flag) 32737c478bd9Sstevel@tonic-gate { 32747c478bd9Sstevel@tonic-gate vnode_t *vp = pp->p_vnode; 32757c478bd9Sstevel@tonic-gate ASSERT(!(flag & ~(P_MOD | P_REF | P_RO))); 32767c478bd9Sstevel@tonic-gate 32777c478bd9Sstevel@tonic-gate /* 3278a71e32b6Sstans * Caller is expected to hold page's io lock for VMODSORT to work 3279a71e32b6Sstans * correctly with pvn_vplist_dirty() and pvn_getdirty() when mod 3280a71e32b6Sstans * bit is cleared. 3281a71e32b6Sstans * We don't have assert to avoid tripping some existing third party 3282a71e32b6Sstans * code. The dirty page is moved back to top of the v_page list 3283a71e32b6Sstans * after IO is done in pvn_write_done(). 32847c478bd9Sstevel@tonic-gate */ 32857c478bd9Sstevel@tonic-gate PP_CLRRM(pp, flag); 32867c478bd9Sstevel@tonic-gate 3287a71e32b6Sstans if ((flag & P_MOD) != 0 && vp != NULL && IS_VMODSORT(vp)) { 32887c478bd9Sstevel@tonic-gate 32897c478bd9Sstevel@tonic-gate /* 32907c478bd9Sstevel@tonic-gate * VMODSORT works by removing write permissions and getting 32917c478bd9Sstevel@tonic-gate * a fault when a page is made dirty. At this point 32927c478bd9Sstevel@tonic-gate * we need to remove write permission from all mappings 32937c478bd9Sstevel@tonic-gate * to this page. 32947c478bd9Sstevel@tonic-gate */ 32957c478bd9Sstevel@tonic-gate hati_page_clrwrt(pp); 32967c478bd9Sstevel@tonic-gate } 32977c478bd9Sstevel@tonic-gate } 32987c478bd9Sstevel@tonic-gate 32997c478bd9Sstevel@tonic-gate /* 33007c478bd9Sstevel@tonic-gate * If flag is specified, returns 0 if attribute is disabled 3301*21584dbcSPavel Tatashin * and non zero if enabled. If flag specifes multiple attributes 3302*21584dbcSPavel Tatashin * then returns 0 if ALL attributes are disabled. This is an advisory 33037c478bd9Sstevel@tonic-gate * call. 33047c478bd9Sstevel@tonic-gate */ 33057c478bd9Sstevel@tonic-gate uint_t 33067c478bd9Sstevel@tonic-gate hat_page_getattr(struct page *pp, uint_t flag) 33077c478bd9Sstevel@tonic-gate { 33087c478bd9Sstevel@tonic-gate return (PP_GETRM(pp, flag)); 33097c478bd9Sstevel@tonic-gate } 33107c478bd9Sstevel@tonic-gate 33117c478bd9Sstevel@tonic-gate 33127c478bd9Sstevel@tonic-gate /* 33137c478bd9Sstevel@tonic-gate * common code used by hat_pageunload() and hment_steal() 33147c478bd9Sstevel@tonic-gate */ 33157c478bd9Sstevel@tonic-gate hment_t * 33167c478bd9Sstevel@tonic-gate hati_page_unmap(page_t *pp, htable_t *ht, uint_t entry) 33177c478bd9Sstevel@tonic-gate { 33187c478bd9Sstevel@tonic-gate x86pte_t old_pte; 33197c478bd9Sstevel@tonic-gate pfn_t pfn = pp->p_pagenum; 33207c478bd9Sstevel@tonic-gate hment_t *hm; 33217c478bd9Sstevel@tonic-gate 33227c478bd9Sstevel@tonic-gate /* 33237c478bd9Sstevel@tonic-gate * We need to acquire a hold on the htable in order to 33247c478bd9Sstevel@tonic-gate * do the invalidate. We know the htable must exist, since 33257c478bd9Sstevel@tonic-gate * unmap's don't release the htable until after removing any 33267c478bd9Sstevel@tonic-gate * hment. Having x86_hm_enter() keeps that from proceeding. 33277c478bd9Sstevel@tonic-gate */ 33287c478bd9Sstevel@tonic-gate htable_acquire(ht); 33297c478bd9Sstevel@tonic-gate 33307c478bd9Sstevel@tonic-gate /* 33317c478bd9Sstevel@tonic-gate * Invalidate the PTE and remove the hment. 33327c478bd9Sstevel@tonic-gate */ 3333ae115bc7Smrj old_pte = x86pte_inval(ht, entry, 0, NULL); 3334aa2ed9e5Sjosephb if (PTE2PFN(old_pte, ht->ht_level) != pfn) { 3335ae115bc7Smrj panic("x86pte_inval() failure found PTE = " FMT_PTE 3336aa2ed9e5Sjosephb " pfn being unmapped is %lx ht=0x%lx entry=0x%x", 3337aa2ed9e5Sjosephb old_pte, pfn, (uintptr_t)ht, entry); 3338aa2ed9e5Sjosephb } 33397c478bd9Sstevel@tonic-gate 33407c478bd9Sstevel@tonic-gate /* 33417c478bd9Sstevel@tonic-gate * Clean up all the htable information for this mapping 33427c478bd9Sstevel@tonic-gate */ 33437c478bd9Sstevel@tonic-gate ASSERT(ht->ht_valid_cnt > 0); 33447c478bd9Sstevel@tonic-gate HTABLE_DEC(ht->ht_valid_cnt); 33457c478bd9Sstevel@tonic-gate PGCNT_DEC(ht->ht_hat, ht->ht_level); 33467c478bd9Sstevel@tonic-gate 33477c478bd9Sstevel@tonic-gate /* 33487c478bd9Sstevel@tonic-gate * sync ref/mod bits to the page_t 33497c478bd9Sstevel@tonic-gate */ 3350ae115bc7Smrj if (PTE_GET(old_pte, PT_SOFTWARE) < PT_NOSYNC) 33517c478bd9Sstevel@tonic-gate hati_sync_pte_to_page(pp, old_pte, ht->ht_level); 33527c478bd9Sstevel@tonic-gate 33537c478bd9Sstevel@tonic-gate /* 33547c478bd9Sstevel@tonic-gate * Remove the mapping list entry for this page. 33557c478bd9Sstevel@tonic-gate */ 33567c478bd9Sstevel@tonic-gate hm = hment_remove(pp, ht, entry); 33577c478bd9Sstevel@tonic-gate 33587c478bd9Sstevel@tonic-gate /* 33597c478bd9Sstevel@tonic-gate * drop the mapping list lock so that we might free the 33607c478bd9Sstevel@tonic-gate * hment and htable. 33617c478bd9Sstevel@tonic-gate */ 33627c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 33637c478bd9Sstevel@tonic-gate htable_release(ht); 33647c478bd9Sstevel@tonic-gate return (hm); 33657c478bd9Sstevel@tonic-gate } 33667c478bd9Sstevel@tonic-gate 3367a5652762Spraks extern int vpm_enable; 33687c478bd9Sstevel@tonic-gate /* 33697c478bd9Sstevel@tonic-gate * Unload all translations to a page. If the page is a subpage of a large 33707c478bd9Sstevel@tonic-gate * page, the large page mappings are also removed. 33717c478bd9Sstevel@tonic-gate * 33727c478bd9Sstevel@tonic-gate * The forceflags are unused. 33737c478bd9Sstevel@tonic-gate */ 33747c478bd9Sstevel@tonic-gate 33757c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 33767c478bd9Sstevel@tonic-gate static int 33777c478bd9Sstevel@tonic-gate hati_pageunload(struct page *pp, uint_t pg_szcd, uint_t forceflag) 33787c478bd9Sstevel@tonic-gate { 33797c478bd9Sstevel@tonic-gate page_t *cur_pp = pp; 33807c478bd9Sstevel@tonic-gate hment_t *hm; 33817c478bd9Sstevel@tonic-gate hment_t *prev; 33827c478bd9Sstevel@tonic-gate htable_t *ht; 33837c478bd9Sstevel@tonic-gate uint_t entry; 33847c478bd9Sstevel@tonic-gate level_t level; 33857c478bd9Sstevel@tonic-gate 3386843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 3387a5652762Spraks #if defined(__amd64) 3388a5652762Spraks /* 3389a5652762Spraks * clear the vpm ref. 3390a5652762Spraks */ 3391a5652762Spraks if (vpm_enable) { 3392a5652762Spraks pp->p_vpmref = 0; 3393a5652762Spraks } 3394a5652762Spraks #endif 33957c478bd9Sstevel@tonic-gate /* 33967c478bd9Sstevel@tonic-gate * The loop with next_size handles pages with multiple pagesize mappings 33977c478bd9Sstevel@tonic-gate */ 33987c478bd9Sstevel@tonic-gate next_size: 33997c478bd9Sstevel@tonic-gate for (;;) { 34007c478bd9Sstevel@tonic-gate 34017c478bd9Sstevel@tonic-gate /* 34027c478bd9Sstevel@tonic-gate * Get a mapping list entry 34037c478bd9Sstevel@tonic-gate */ 34047c478bd9Sstevel@tonic-gate x86_hm_enter(cur_pp); 34057c478bd9Sstevel@tonic-gate for (prev = NULL; ; prev = hm) { 34067c478bd9Sstevel@tonic-gate hm = hment_walk(cur_pp, &ht, &entry, prev); 34077c478bd9Sstevel@tonic-gate if (hm == NULL) { 34087c478bd9Sstevel@tonic-gate x86_hm_exit(cur_pp); 34097c478bd9Sstevel@tonic-gate 34107c478bd9Sstevel@tonic-gate /* 34117c478bd9Sstevel@tonic-gate * If not part of a larger page, we're done. 34127c478bd9Sstevel@tonic-gate */ 3413ae115bc7Smrj if (cur_pp->p_szc <= pg_szcd) { 3414843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 34157c478bd9Sstevel@tonic-gate return (0); 3416ae115bc7Smrj } 34177c478bd9Sstevel@tonic-gate 34187c478bd9Sstevel@tonic-gate /* 34197c478bd9Sstevel@tonic-gate * Else check the next larger page size. 34207c478bd9Sstevel@tonic-gate * hat_page_demote() may decrease p_szc 34217c478bd9Sstevel@tonic-gate * but that's ok we'll just take an extra 34227c478bd9Sstevel@tonic-gate * trip discover there're no larger mappings 34237c478bd9Sstevel@tonic-gate * and return. 34247c478bd9Sstevel@tonic-gate */ 34257c478bd9Sstevel@tonic-gate ++pg_szcd; 34267c478bd9Sstevel@tonic-gate cur_pp = PP_GROUPLEADER(cur_pp, pg_szcd); 34277c478bd9Sstevel@tonic-gate goto next_size; 34287c478bd9Sstevel@tonic-gate } 34297c478bd9Sstevel@tonic-gate 34307c478bd9Sstevel@tonic-gate /* 34317c478bd9Sstevel@tonic-gate * If this mapping size matches, remove it. 34327c478bd9Sstevel@tonic-gate */ 34337c478bd9Sstevel@tonic-gate level = ht->ht_level; 34347c478bd9Sstevel@tonic-gate if (level == pg_szcd) 34357c478bd9Sstevel@tonic-gate break; 34367c478bd9Sstevel@tonic-gate } 34377c478bd9Sstevel@tonic-gate 34387c478bd9Sstevel@tonic-gate /* 34397c478bd9Sstevel@tonic-gate * Remove the mapping list entry for this page. 34407c478bd9Sstevel@tonic-gate * Note this does the x86_hm_exit() for us. 34417c478bd9Sstevel@tonic-gate */ 34427c478bd9Sstevel@tonic-gate hm = hati_page_unmap(cur_pp, ht, entry); 34437c478bd9Sstevel@tonic-gate if (hm != NULL) 34447c478bd9Sstevel@tonic-gate hment_free(hm); 34457c478bd9Sstevel@tonic-gate } 34467c478bd9Sstevel@tonic-gate } 34477c478bd9Sstevel@tonic-gate 34487c478bd9Sstevel@tonic-gate int 34497c478bd9Sstevel@tonic-gate hat_pageunload(struct page *pp, uint_t forceflag) 34507c478bd9Sstevel@tonic-gate { 34517c478bd9Sstevel@tonic-gate ASSERT(PAGE_EXCL(pp)); 34527c478bd9Sstevel@tonic-gate return (hati_pageunload(pp, 0, forceflag)); 34537c478bd9Sstevel@tonic-gate } 34547c478bd9Sstevel@tonic-gate 34557c478bd9Sstevel@tonic-gate /* 34567c478bd9Sstevel@tonic-gate * Unload all large mappings to pp and reduce by 1 p_szc field of every large 34577c478bd9Sstevel@tonic-gate * page level that included pp. 34587c478bd9Sstevel@tonic-gate * 34597c478bd9Sstevel@tonic-gate * pp must be locked EXCL. Even though no other constituent pages are locked 34607c478bd9Sstevel@tonic-gate * it's legal to unload large mappings to pp because all constituent pages of 34617c478bd9Sstevel@tonic-gate * large locked mappings have to be locked SHARED. therefore if we have EXCL 34627c478bd9Sstevel@tonic-gate * lock on one of constituent pages none of the large mappings to pp are 34637c478bd9Sstevel@tonic-gate * locked. 34647c478bd9Sstevel@tonic-gate * 34657c478bd9Sstevel@tonic-gate * Change (always decrease) p_szc field starting from the last constituent 34667c478bd9Sstevel@tonic-gate * page and ending with root constituent page so that root's pszc always shows 34677c478bd9Sstevel@tonic-gate * the area where hat_page_demote() may be active. 34687c478bd9Sstevel@tonic-gate * 34697c478bd9Sstevel@tonic-gate * This mechanism is only used for file system pages where it's not always 34707c478bd9Sstevel@tonic-gate * possible to get EXCL locks on all constituent pages to demote the size code 34717c478bd9Sstevel@tonic-gate * (as is done for anonymous or kernel large pages). 34727c478bd9Sstevel@tonic-gate */ 34737c478bd9Sstevel@tonic-gate void 34747c478bd9Sstevel@tonic-gate hat_page_demote(page_t *pp) 34757c478bd9Sstevel@tonic-gate { 34767c478bd9Sstevel@tonic-gate uint_t pszc; 34777c478bd9Sstevel@tonic-gate uint_t rszc; 34787c478bd9Sstevel@tonic-gate uint_t szc; 34797c478bd9Sstevel@tonic-gate page_t *rootpp; 34807c478bd9Sstevel@tonic-gate page_t *firstpp; 34817c478bd9Sstevel@tonic-gate page_t *lastpp; 34827c478bd9Sstevel@tonic-gate pgcnt_t pgcnt; 34837c478bd9Sstevel@tonic-gate 34847c478bd9Sstevel@tonic-gate ASSERT(PAGE_EXCL(pp)); 34857c478bd9Sstevel@tonic-gate ASSERT(!PP_ISFREE(pp)); 34867c478bd9Sstevel@tonic-gate ASSERT(page_szc_lock_assert(pp)); 34877c478bd9Sstevel@tonic-gate 34887c478bd9Sstevel@tonic-gate if (pp->p_szc == 0) 34897c478bd9Sstevel@tonic-gate return; 34907c478bd9Sstevel@tonic-gate 34917c478bd9Sstevel@tonic-gate rootpp = PP_GROUPLEADER(pp, 1); 34927c478bd9Sstevel@tonic-gate (void) hati_pageunload(rootpp, 1, HAT_FORCE_PGUNLOAD); 34937c478bd9Sstevel@tonic-gate 34947c478bd9Sstevel@tonic-gate /* 34957c478bd9Sstevel@tonic-gate * all large mappings to pp are gone 34967c478bd9Sstevel@tonic-gate * and no new can be setup since pp is locked exclusively. 34977c478bd9Sstevel@tonic-gate * 34987c478bd9Sstevel@tonic-gate * Lock the root to make sure there's only one hat_page_demote() 34997c478bd9Sstevel@tonic-gate * outstanding within the area of this root's pszc. 35007c478bd9Sstevel@tonic-gate * 35017c478bd9Sstevel@tonic-gate * Second potential hat_page_demote() is already eliminated by upper 35027c478bd9Sstevel@tonic-gate * VM layer via page_szc_lock() but we don't rely on it and use our 35037c478bd9Sstevel@tonic-gate * own locking (so that upper layer locking can be changed without 35047c478bd9Sstevel@tonic-gate * assumptions that hat depends on upper layer VM to prevent multiple 35057c478bd9Sstevel@tonic-gate * hat_page_demote() to be issued simultaneously to the same large 35067c478bd9Sstevel@tonic-gate * page). 35077c478bd9Sstevel@tonic-gate */ 35087c478bd9Sstevel@tonic-gate again: 35097c478bd9Sstevel@tonic-gate pszc = pp->p_szc; 35107c478bd9Sstevel@tonic-gate if (pszc == 0) 35117c478bd9Sstevel@tonic-gate return; 35127c478bd9Sstevel@tonic-gate rootpp = PP_GROUPLEADER(pp, pszc); 35137c478bd9Sstevel@tonic-gate x86_hm_enter(rootpp); 35147c478bd9Sstevel@tonic-gate /* 35157c478bd9Sstevel@tonic-gate * If root's p_szc is different from pszc we raced with another 35167c478bd9Sstevel@tonic-gate * hat_page_demote(). Drop the lock and try to find the root again. 35177c478bd9Sstevel@tonic-gate * If root's p_szc is greater than pszc previous hat_page_demote() is 35187c478bd9Sstevel@tonic-gate * not done yet. Take and release mlist lock of root's root to wait 35197c478bd9Sstevel@tonic-gate * for previous hat_page_demote() to complete. 35207c478bd9Sstevel@tonic-gate */ 35217c478bd9Sstevel@tonic-gate if ((rszc = rootpp->p_szc) != pszc) { 35227c478bd9Sstevel@tonic-gate x86_hm_exit(rootpp); 35237c478bd9Sstevel@tonic-gate if (rszc > pszc) { 35247c478bd9Sstevel@tonic-gate /* p_szc of a locked non free page can't increase */ 35257c478bd9Sstevel@tonic-gate ASSERT(pp != rootpp); 35267c478bd9Sstevel@tonic-gate 35277c478bd9Sstevel@tonic-gate rootpp = PP_GROUPLEADER(rootpp, rszc); 35287c478bd9Sstevel@tonic-gate x86_hm_enter(rootpp); 35297c478bd9Sstevel@tonic-gate x86_hm_exit(rootpp); 35307c478bd9Sstevel@tonic-gate } 35317c478bd9Sstevel@tonic-gate goto again; 35327c478bd9Sstevel@tonic-gate } 35337c478bd9Sstevel@tonic-gate ASSERT(pp->p_szc == pszc); 35347c478bd9Sstevel@tonic-gate 35357c478bd9Sstevel@tonic-gate /* 35367c478bd9Sstevel@tonic-gate * Decrement by 1 p_szc of every constituent page of a region that 35377c478bd9Sstevel@tonic-gate * covered pp. For example if original szc is 3 it gets changed to 2 35387c478bd9Sstevel@tonic-gate * everywhere except in region 2 that covered pp. Region 2 that 35397c478bd9Sstevel@tonic-gate * covered pp gets demoted to 1 everywhere except in region 1 that 35407c478bd9Sstevel@tonic-gate * covered pp. The region 1 that covered pp is demoted to region 35417c478bd9Sstevel@tonic-gate * 0. It's done this way because from region 3 we removed level 3 35427c478bd9Sstevel@tonic-gate * mappings, from region 2 that covered pp we removed level 2 mappings 35437c478bd9Sstevel@tonic-gate * and from region 1 that covered pp we removed level 1 mappings. All 35447c478bd9Sstevel@tonic-gate * changes are done from from high pfn's to low pfn's so that roots 35457c478bd9Sstevel@tonic-gate * are changed last allowing one to know the largest region where 35467c478bd9Sstevel@tonic-gate * hat_page_demote() is stil active by only looking at the root page. 35477c478bd9Sstevel@tonic-gate * 35487c478bd9Sstevel@tonic-gate * This algorithm is implemented in 2 while loops. First loop changes 35497c478bd9Sstevel@tonic-gate * p_szc of pages to the right of pp's level 1 region and second 35507c478bd9Sstevel@tonic-gate * loop changes p_szc of pages of level 1 region that covers pp 35517c478bd9Sstevel@tonic-gate * and all pages to the left of level 1 region that covers pp. 35527c478bd9Sstevel@tonic-gate * In the first loop p_szc keeps dropping with every iteration 35537c478bd9Sstevel@tonic-gate * and in the second loop it keeps increasing with every iteration. 35547c478bd9Sstevel@tonic-gate * 35557c478bd9Sstevel@tonic-gate * First loop description: Demote pages to the right of pp outside of 35567c478bd9Sstevel@tonic-gate * level 1 region that covers pp. In every iteration of the while 35577c478bd9Sstevel@tonic-gate * loop below find the last page of szc region and the first page of 35587c478bd9Sstevel@tonic-gate * (szc - 1) region that is immediately to the right of (szc - 1) 35597c478bd9Sstevel@tonic-gate * region that covers pp. From last such page to first such page 35607c478bd9Sstevel@tonic-gate * change every page's szc to szc - 1. Decrement szc and continue 35617c478bd9Sstevel@tonic-gate * looping until szc is 1. If pp belongs to the last (szc - 1) region 35627c478bd9Sstevel@tonic-gate * of szc region skip to the next iteration. 35637c478bd9Sstevel@tonic-gate */ 35647c478bd9Sstevel@tonic-gate szc = pszc; 35657c478bd9Sstevel@tonic-gate while (szc > 1) { 35667c478bd9Sstevel@tonic-gate lastpp = PP_GROUPLEADER(pp, szc); 35677c478bd9Sstevel@tonic-gate pgcnt = page_get_pagecnt(szc); 35687c478bd9Sstevel@tonic-gate lastpp += pgcnt - 1; 35697c478bd9Sstevel@tonic-gate firstpp = PP_GROUPLEADER(pp, (szc - 1)); 35707c478bd9Sstevel@tonic-gate pgcnt = page_get_pagecnt(szc - 1); 35717c478bd9Sstevel@tonic-gate if (lastpp - firstpp < pgcnt) { 35727c478bd9Sstevel@tonic-gate szc--; 35737c478bd9Sstevel@tonic-gate continue; 35747c478bd9Sstevel@tonic-gate } 35757c478bd9Sstevel@tonic-gate firstpp += pgcnt; 35767c478bd9Sstevel@tonic-gate while (lastpp != firstpp) { 35777c478bd9Sstevel@tonic-gate ASSERT(lastpp->p_szc == pszc); 35787c478bd9Sstevel@tonic-gate lastpp->p_szc = szc - 1; 35797c478bd9Sstevel@tonic-gate lastpp--; 35807c478bd9Sstevel@tonic-gate } 35817c478bd9Sstevel@tonic-gate firstpp->p_szc = szc - 1; 35827c478bd9Sstevel@tonic-gate szc--; 35837c478bd9Sstevel@tonic-gate } 35847c478bd9Sstevel@tonic-gate 35857c478bd9Sstevel@tonic-gate /* 35867c478bd9Sstevel@tonic-gate * Second loop description: 35877c478bd9Sstevel@tonic-gate * First iteration changes p_szc to 0 of every 35887c478bd9Sstevel@tonic-gate * page of level 1 region that covers pp. 35897c478bd9Sstevel@tonic-gate * Subsequent iterations find last page of szc region 35907c478bd9Sstevel@tonic-gate * immediately to the left of szc region that covered pp 35917c478bd9Sstevel@tonic-gate * and first page of (szc + 1) region that covers pp. 35927c478bd9Sstevel@tonic-gate * From last to first page change p_szc of every page to szc. 35937c478bd9Sstevel@tonic-gate * Increment szc and continue looping until szc is pszc. 35947c478bd9Sstevel@tonic-gate * If pp belongs to the fist szc region of (szc + 1) region 35957c478bd9Sstevel@tonic-gate * skip to the next iteration. 35967c478bd9Sstevel@tonic-gate * 35977c478bd9Sstevel@tonic-gate */ 35987c478bd9Sstevel@tonic-gate szc = 0; 35997c478bd9Sstevel@tonic-gate while (szc < pszc) { 36007c478bd9Sstevel@tonic-gate firstpp = PP_GROUPLEADER(pp, (szc + 1)); 36017c478bd9Sstevel@tonic-gate if (szc == 0) { 36027c478bd9Sstevel@tonic-gate pgcnt = page_get_pagecnt(1); 36037c478bd9Sstevel@tonic-gate lastpp = firstpp + (pgcnt - 1); 36047c478bd9Sstevel@tonic-gate } else { 36057c478bd9Sstevel@tonic-gate lastpp = PP_GROUPLEADER(pp, szc); 36067c478bd9Sstevel@tonic-gate if (firstpp == lastpp) { 36077c478bd9Sstevel@tonic-gate szc++; 36087c478bd9Sstevel@tonic-gate continue; 36097c478bd9Sstevel@tonic-gate } 36107c478bd9Sstevel@tonic-gate lastpp--; 36117c478bd9Sstevel@tonic-gate pgcnt = page_get_pagecnt(szc); 36127c478bd9Sstevel@tonic-gate } 36137c478bd9Sstevel@tonic-gate while (lastpp != firstpp) { 36147c478bd9Sstevel@tonic-gate ASSERT(lastpp->p_szc == pszc); 36157c478bd9Sstevel@tonic-gate lastpp->p_szc = szc; 36167c478bd9Sstevel@tonic-gate lastpp--; 36177c478bd9Sstevel@tonic-gate } 36187c478bd9Sstevel@tonic-gate firstpp->p_szc = szc; 36197c478bd9Sstevel@tonic-gate if (firstpp == rootpp) 36207c478bd9Sstevel@tonic-gate break; 36217c478bd9Sstevel@tonic-gate szc++; 36227c478bd9Sstevel@tonic-gate } 36237c478bd9Sstevel@tonic-gate x86_hm_exit(rootpp); 36247c478bd9Sstevel@tonic-gate } 36257c478bd9Sstevel@tonic-gate 36267c478bd9Sstevel@tonic-gate /* 36277c478bd9Sstevel@tonic-gate * get hw stats from hardware into page struct and reset hw stats 36287c478bd9Sstevel@tonic-gate * returns attributes of page 36297c478bd9Sstevel@tonic-gate * Flags for hat_pagesync, hat_getstat, hat_sync 36307c478bd9Sstevel@tonic-gate * 36317c478bd9Sstevel@tonic-gate * define HAT_SYNC_ZERORM 0x01 36327c478bd9Sstevel@tonic-gate * 36337c478bd9Sstevel@tonic-gate * Additional flags for hat_pagesync 36347c478bd9Sstevel@tonic-gate * 36357c478bd9Sstevel@tonic-gate * define HAT_SYNC_STOPON_REF 0x02 36367c478bd9Sstevel@tonic-gate * define HAT_SYNC_STOPON_MOD 0x04 36377c478bd9Sstevel@tonic-gate * define HAT_SYNC_STOPON_RM 0x06 36387c478bd9Sstevel@tonic-gate * define HAT_SYNC_STOPON_SHARED 0x08 36397c478bd9Sstevel@tonic-gate */ 36407c478bd9Sstevel@tonic-gate uint_t 36417c478bd9Sstevel@tonic-gate hat_pagesync(struct page *pp, uint_t flags) 36427c478bd9Sstevel@tonic-gate { 36437c478bd9Sstevel@tonic-gate hment_t *hm = NULL; 36447c478bd9Sstevel@tonic-gate htable_t *ht; 36457c478bd9Sstevel@tonic-gate uint_t entry; 36467c478bd9Sstevel@tonic-gate x86pte_t old, save_old; 36477c478bd9Sstevel@tonic-gate x86pte_t new; 36487c478bd9Sstevel@tonic-gate uchar_t nrmbits = P_REF|P_MOD|P_RO; 36497c478bd9Sstevel@tonic-gate extern ulong_t po_share; 36507c478bd9Sstevel@tonic-gate page_t *save_pp = pp; 36517c478bd9Sstevel@tonic-gate uint_t pszc = 0; 36527c478bd9Sstevel@tonic-gate 36537c478bd9Sstevel@tonic-gate ASSERT(PAGE_LOCKED(pp) || panicstr); 36547c478bd9Sstevel@tonic-gate 36557c478bd9Sstevel@tonic-gate if (PP_ISRO(pp) && (flags & HAT_SYNC_STOPON_MOD)) 36567c478bd9Sstevel@tonic-gate return (pp->p_nrm & nrmbits); 36577c478bd9Sstevel@tonic-gate 36587c478bd9Sstevel@tonic-gate if ((flags & HAT_SYNC_ZERORM) == 0) { 36597c478bd9Sstevel@tonic-gate 36607c478bd9Sstevel@tonic-gate if ((flags & HAT_SYNC_STOPON_REF) != 0 && PP_ISREF(pp)) 36617c478bd9Sstevel@tonic-gate return (pp->p_nrm & nrmbits); 36627c478bd9Sstevel@tonic-gate 36637c478bd9Sstevel@tonic-gate if ((flags & HAT_SYNC_STOPON_MOD) != 0 && PP_ISMOD(pp)) 36647c478bd9Sstevel@tonic-gate return (pp->p_nrm & nrmbits); 36657c478bd9Sstevel@tonic-gate 36667c478bd9Sstevel@tonic-gate if ((flags & HAT_SYNC_STOPON_SHARED) != 0 && 36677c478bd9Sstevel@tonic-gate hat_page_getshare(pp) > po_share) { 36687c478bd9Sstevel@tonic-gate if (PP_ISRO(pp)) 36697c478bd9Sstevel@tonic-gate PP_SETREF(pp); 36707c478bd9Sstevel@tonic-gate return (pp->p_nrm & nrmbits); 36717c478bd9Sstevel@tonic-gate } 36727c478bd9Sstevel@tonic-gate } 36737c478bd9Sstevel@tonic-gate 3674843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 36757c478bd9Sstevel@tonic-gate next_size: 36767c478bd9Sstevel@tonic-gate /* 36777c478bd9Sstevel@tonic-gate * walk thru the mapping list syncing (and clearing) ref/mod bits. 36787c478bd9Sstevel@tonic-gate */ 36797c478bd9Sstevel@tonic-gate x86_hm_enter(pp); 36807c478bd9Sstevel@tonic-gate while ((hm = hment_walk(pp, &ht, &entry, hm)) != NULL) { 36817c478bd9Sstevel@tonic-gate if (ht->ht_level < pszc) 36827c478bd9Sstevel@tonic-gate continue; 36837c478bd9Sstevel@tonic-gate old = x86pte_get(ht, entry); 36847c478bd9Sstevel@tonic-gate try_again: 36857c478bd9Sstevel@tonic-gate 36867c478bd9Sstevel@tonic-gate ASSERT(PTE2PFN(old, ht->ht_level) == pp->p_pagenum); 36877c478bd9Sstevel@tonic-gate 36887c478bd9Sstevel@tonic-gate if (PTE_GET(old, PT_REF | PT_MOD) == 0) 36897c478bd9Sstevel@tonic-gate continue; 36907c478bd9Sstevel@tonic-gate 36917c478bd9Sstevel@tonic-gate save_old = old; 36927c478bd9Sstevel@tonic-gate if ((flags & HAT_SYNC_ZERORM) != 0) { 36937c478bd9Sstevel@tonic-gate 36947c478bd9Sstevel@tonic-gate /* 36957c478bd9Sstevel@tonic-gate * Need to clear ref or mod bits. Need to demap 36967c478bd9Sstevel@tonic-gate * to make sure any executing TLBs see cleared bits. 36977c478bd9Sstevel@tonic-gate */ 36987c478bd9Sstevel@tonic-gate new = old; 36997c478bd9Sstevel@tonic-gate PTE_CLR(new, PT_REF | PT_MOD); 37007c478bd9Sstevel@tonic-gate old = hati_update_pte(ht, entry, old, new); 37017c478bd9Sstevel@tonic-gate if (old != 0) 37027c478bd9Sstevel@tonic-gate goto try_again; 37037c478bd9Sstevel@tonic-gate 37047c478bd9Sstevel@tonic-gate old = save_old; 37057c478bd9Sstevel@tonic-gate } 37067c478bd9Sstevel@tonic-gate 37077c478bd9Sstevel@tonic-gate /* 37087c478bd9Sstevel@tonic-gate * Sync the PTE 37097c478bd9Sstevel@tonic-gate */ 3710ae115bc7Smrj if (!(flags & HAT_SYNC_ZERORM) && 3711ae115bc7Smrj PTE_GET(old, PT_SOFTWARE) <= PT_NOSYNC) 37127c478bd9Sstevel@tonic-gate hati_sync_pte_to_page(pp, old, ht->ht_level); 37137c478bd9Sstevel@tonic-gate 37147c478bd9Sstevel@tonic-gate /* 37157c478bd9Sstevel@tonic-gate * can stop short if we found a ref'd or mod'd page 37167c478bd9Sstevel@tonic-gate */ 37177c478bd9Sstevel@tonic-gate if ((flags & HAT_SYNC_STOPON_MOD) && PP_ISMOD(save_pp) || 37187c478bd9Sstevel@tonic-gate (flags & HAT_SYNC_STOPON_REF) && PP_ISREF(save_pp)) { 37197c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 3720ae115bc7Smrj goto done; 37217c478bd9Sstevel@tonic-gate } 37227c478bd9Sstevel@tonic-gate } 37237c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 37247c478bd9Sstevel@tonic-gate while (pszc < pp->p_szc) { 37257c478bd9Sstevel@tonic-gate page_t *tpp; 37267c478bd9Sstevel@tonic-gate pszc++; 37277c478bd9Sstevel@tonic-gate tpp = PP_GROUPLEADER(pp, pszc); 37287c478bd9Sstevel@tonic-gate if (pp != tpp) { 37297c478bd9Sstevel@tonic-gate pp = tpp; 37307c478bd9Sstevel@tonic-gate goto next_size; 37317c478bd9Sstevel@tonic-gate } 37327c478bd9Sstevel@tonic-gate } 3733ae115bc7Smrj done: 3734843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 37357c478bd9Sstevel@tonic-gate return (save_pp->p_nrm & nrmbits); 37367c478bd9Sstevel@tonic-gate } 37377c478bd9Sstevel@tonic-gate 37387c478bd9Sstevel@tonic-gate /* 37397c478bd9Sstevel@tonic-gate * returns approx number of mappings to this pp. A return of 0 implies 37407c478bd9Sstevel@tonic-gate * there are no mappings to the page. 37417c478bd9Sstevel@tonic-gate */ 37427c478bd9Sstevel@tonic-gate ulong_t 37437c478bd9Sstevel@tonic-gate hat_page_getshare(page_t *pp) 37447c478bd9Sstevel@tonic-gate { 37457c478bd9Sstevel@tonic-gate uint_t cnt; 37467c478bd9Sstevel@tonic-gate cnt = hment_mapcnt(pp); 3747a5652762Spraks #if defined(__amd64) 3748a5652762Spraks if (vpm_enable && pp->p_vpmref) { 3749a5652762Spraks cnt += 1; 3750a5652762Spraks } 3751a5652762Spraks #endif 37527c478bd9Sstevel@tonic-gate return (cnt); 37537c478bd9Sstevel@tonic-gate } 37547c478bd9Sstevel@tonic-gate 375505d3dc4bSpaulsan /* 375605d3dc4bSpaulsan * Return 1 the number of mappings exceeds sh_thresh. Return 0 375705d3dc4bSpaulsan * otherwise. 375805d3dc4bSpaulsan */ 375905d3dc4bSpaulsan int 376005d3dc4bSpaulsan hat_page_checkshare(page_t *pp, ulong_t sh_thresh) 376105d3dc4bSpaulsan { 376205d3dc4bSpaulsan return (hat_page_getshare(pp) > sh_thresh); 376305d3dc4bSpaulsan } 376405d3dc4bSpaulsan 37657c478bd9Sstevel@tonic-gate /* 37667c478bd9Sstevel@tonic-gate * hat_softlock isn't supported anymore 37677c478bd9Sstevel@tonic-gate */ 37687c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 37697c478bd9Sstevel@tonic-gate faultcode_t 37707c478bd9Sstevel@tonic-gate hat_softlock( 37717c478bd9Sstevel@tonic-gate hat_t *hat, 37727c478bd9Sstevel@tonic-gate caddr_t addr, 37737c478bd9Sstevel@tonic-gate size_t *len, 37747c478bd9Sstevel@tonic-gate struct page **page_array, 37757c478bd9Sstevel@tonic-gate uint_t flags) 37767c478bd9Sstevel@tonic-gate { 37777c478bd9Sstevel@tonic-gate return (FC_NOSUPPORT); 37787c478bd9Sstevel@tonic-gate } 37797c478bd9Sstevel@tonic-gate 37807c478bd9Sstevel@tonic-gate 37817c478bd9Sstevel@tonic-gate 37827c478bd9Sstevel@tonic-gate /* 37837c478bd9Sstevel@tonic-gate * Routine to expose supported HAT features to platform independent code. 37847c478bd9Sstevel@tonic-gate */ 37857c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 37867c478bd9Sstevel@tonic-gate int 37877c478bd9Sstevel@tonic-gate hat_supported(enum hat_features feature, void *arg) 37887c478bd9Sstevel@tonic-gate { 37897c478bd9Sstevel@tonic-gate switch (feature) { 37907c478bd9Sstevel@tonic-gate 37917c478bd9Sstevel@tonic-gate case HAT_SHARED_PT: /* this is really ISM */ 37927c478bd9Sstevel@tonic-gate return (1); 37937c478bd9Sstevel@tonic-gate 37947c478bd9Sstevel@tonic-gate case HAT_DYNAMIC_ISM_UNMAP: 37957c478bd9Sstevel@tonic-gate return (0); 37967c478bd9Sstevel@tonic-gate 37977c478bd9Sstevel@tonic-gate case HAT_VMODSORT: 37987c478bd9Sstevel@tonic-gate return (1); 37997c478bd9Sstevel@tonic-gate 380005d3dc4bSpaulsan case HAT_SHARED_REGIONS: 380105d3dc4bSpaulsan return (0); 380205d3dc4bSpaulsan 38037c478bd9Sstevel@tonic-gate default: 38047c478bd9Sstevel@tonic-gate panic("hat_supported() - unknown feature"); 38057c478bd9Sstevel@tonic-gate } 38067c478bd9Sstevel@tonic-gate return (0); 38077c478bd9Sstevel@tonic-gate } 38087c478bd9Sstevel@tonic-gate 38097c478bd9Sstevel@tonic-gate /* 38107c478bd9Sstevel@tonic-gate * Called when a thread is exiting and has been switched to the kernel AS 38117c478bd9Sstevel@tonic-gate */ 38127c478bd9Sstevel@tonic-gate void 38137c478bd9Sstevel@tonic-gate hat_thread_exit(kthread_t *thd) 38147c478bd9Sstevel@tonic-gate { 38157c478bd9Sstevel@tonic-gate ASSERT(thd->t_procp->p_as == &kas); 3816843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 38177c478bd9Sstevel@tonic-gate hat_switch(thd->t_procp->p_as->a_hat); 3818843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 38197c478bd9Sstevel@tonic-gate } 38207c478bd9Sstevel@tonic-gate 38217c478bd9Sstevel@tonic-gate /* 38227c478bd9Sstevel@tonic-gate * Setup the given brand new hat structure as the new HAT on this cpu's mmu. 38237c478bd9Sstevel@tonic-gate */ 38247c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 38257c478bd9Sstevel@tonic-gate void 38267c478bd9Sstevel@tonic-gate hat_setup(hat_t *hat, int flags) 38277c478bd9Sstevel@tonic-gate { 3828843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 38297c478bd9Sstevel@tonic-gate kpreempt_disable(); 38307c478bd9Sstevel@tonic-gate 38317c478bd9Sstevel@tonic-gate hat_switch(hat); 38327c478bd9Sstevel@tonic-gate 38337c478bd9Sstevel@tonic-gate kpreempt_enable(); 3834843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 38357c478bd9Sstevel@tonic-gate } 38367c478bd9Sstevel@tonic-gate 38377c478bd9Sstevel@tonic-gate /* 38387c478bd9Sstevel@tonic-gate * Prepare for a CPU private mapping for the given address. 38397c478bd9Sstevel@tonic-gate * 38407c478bd9Sstevel@tonic-gate * The address can only be used from a single CPU and can be remapped 38417c478bd9Sstevel@tonic-gate * using hat_mempte_remap(). Return the address of the PTE. 38427c478bd9Sstevel@tonic-gate * 38437c478bd9Sstevel@tonic-gate * We do the htable_create() if necessary and increment the valid count so 38447c478bd9Sstevel@tonic-gate * the htable can't disappear. We also hat_devload() the page table into 38457c478bd9Sstevel@tonic-gate * kernel so that the PTE is quickly accessed. 38467c478bd9Sstevel@tonic-gate */ 3847ae115bc7Smrj hat_mempte_t 3848ae115bc7Smrj hat_mempte_setup(caddr_t addr) 38497c478bd9Sstevel@tonic-gate { 38507c478bd9Sstevel@tonic-gate uintptr_t va = (uintptr_t)addr; 38517c478bd9Sstevel@tonic-gate htable_t *ht; 38527c478bd9Sstevel@tonic-gate uint_t entry; 38537c478bd9Sstevel@tonic-gate x86pte_t oldpte; 3854ae115bc7Smrj hat_mempte_t p; 38557c478bd9Sstevel@tonic-gate 38567c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(va)); 38577c478bd9Sstevel@tonic-gate ASSERT(!IN_VA_HOLE(va)); 3858aac11643Sjosephb ++curthread->t_hatdepth; 3859551bc2a6Smrj XPV_DISALLOW_MIGRATE(); 38607c478bd9Sstevel@tonic-gate ht = htable_getpte(kas.a_hat, va, &entry, &oldpte, 0); 38617c478bd9Sstevel@tonic-gate if (ht == NULL) { 38627c478bd9Sstevel@tonic-gate ht = htable_create(kas.a_hat, va, 0, NULL); 38637c478bd9Sstevel@tonic-gate entry = htable_va2entry(va, ht); 38647c478bd9Sstevel@tonic-gate ASSERT(ht->ht_level == 0); 38657c478bd9Sstevel@tonic-gate oldpte = x86pte_get(ht, entry); 38667c478bd9Sstevel@tonic-gate } 38677c478bd9Sstevel@tonic-gate if (PTE_ISVALID(oldpte)) 38687c478bd9Sstevel@tonic-gate panic("hat_mempte_setup(): address already mapped" 3869903a11ebSrh "ht=%p, entry=%d, pte=" FMT_PTE, (void *)ht, entry, oldpte); 38707c478bd9Sstevel@tonic-gate 38717c478bd9Sstevel@tonic-gate /* 38727c478bd9Sstevel@tonic-gate * increment ht_valid_cnt so that the pagetable can't disappear 38737c478bd9Sstevel@tonic-gate */ 38747c478bd9Sstevel@tonic-gate HTABLE_INC(ht->ht_valid_cnt); 38757c478bd9Sstevel@tonic-gate 38767c478bd9Sstevel@tonic-gate /* 3877ae115bc7Smrj * return the PTE physical address to the caller. 38787c478bd9Sstevel@tonic-gate */ 38797c478bd9Sstevel@tonic-gate htable_release(ht); 3880551bc2a6Smrj XPV_ALLOW_MIGRATE(); 3881ae115bc7Smrj p = PT_INDEX_PHYSADDR(pfn_to_pa(ht->ht_pfn), entry); 3882aac11643Sjosephb --curthread->t_hatdepth; 3883ae115bc7Smrj return (p); 38847c478bd9Sstevel@tonic-gate } 38857c478bd9Sstevel@tonic-gate 38867c478bd9Sstevel@tonic-gate /* 38877c478bd9Sstevel@tonic-gate * Release a CPU private mapping for the given address. 38887c478bd9Sstevel@tonic-gate * We decrement the htable valid count so it might be destroyed. 38897c478bd9Sstevel@tonic-gate */ 3890ae115bc7Smrj /*ARGSUSED1*/ 38917c478bd9Sstevel@tonic-gate void 3892ae115bc7Smrj hat_mempte_release(caddr_t addr, hat_mempte_t pte_pa) 38937c478bd9Sstevel@tonic-gate { 38947c478bd9Sstevel@tonic-gate htable_t *ht; 38957c478bd9Sstevel@tonic-gate 3896551bc2a6Smrj XPV_DISALLOW_MIGRATE(); 38977c478bd9Sstevel@tonic-gate /* 3898ae115bc7Smrj * invalidate any left over mapping and decrement the htable valid count 38997c478bd9Sstevel@tonic-gate */ 3900843e1988Sjohnlev #ifdef __xpv 3901843e1988Sjohnlev if (HYPERVISOR_update_va_mapping((uintptr_t)addr, 0, 3902843e1988Sjohnlev UVMF_INVLPG | UVMF_LOCAL)) 3903843e1988Sjohnlev panic("HYPERVISOR_update_va_mapping() failed"); 3904843e1988Sjohnlev #else 3905ae115bc7Smrj { 3906ae115bc7Smrj x86pte_t *pteptr; 3907ae115bc7Smrj 3908ae115bc7Smrj pteptr = x86pte_mapin(mmu_btop(pte_pa), 3909ae115bc7Smrj (pte_pa & MMU_PAGEOFFSET) >> mmu.pte_size_shift, NULL); 3910ae115bc7Smrj if (mmu.pae_hat) 3911ae115bc7Smrj *pteptr = 0; 3912ae115bc7Smrj else 3913ae115bc7Smrj *(x86pte32_t *)pteptr = 0; 3914ae115bc7Smrj mmu_tlbflush_entry(addr); 3915ae115bc7Smrj x86pte_mapout(); 3916ae115bc7Smrj } 3917843e1988Sjohnlev #endif 3918ae115bc7Smrj 39197c478bd9Sstevel@tonic-gate ht = htable_getpte(kas.a_hat, ALIGN2PAGE(addr), NULL, NULL, 0); 39207c478bd9Sstevel@tonic-gate if (ht == NULL) 39217c478bd9Sstevel@tonic-gate panic("hat_mempte_release(): invalid address"); 39227c478bd9Sstevel@tonic-gate ASSERT(ht->ht_level == 0); 39237c478bd9Sstevel@tonic-gate HTABLE_DEC(ht->ht_valid_cnt); 39247c478bd9Sstevel@tonic-gate htable_release(ht); 3925551bc2a6Smrj XPV_ALLOW_MIGRATE(); 39267c478bd9Sstevel@tonic-gate } 39277c478bd9Sstevel@tonic-gate 39287c478bd9Sstevel@tonic-gate /* 39297c478bd9Sstevel@tonic-gate * Apply a temporary CPU private mapping to a page. We flush the TLB only 39307c478bd9Sstevel@tonic-gate * on this CPU, so this ought to have been called with preemption disabled. 39317c478bd9Sstevel@tonic-gate */ 39327c478bd9Sstevel@tonic-gate void 39337c478bd9Sstevel@tonic-gate hat_mempte_remap( 3934ae115bc7Smrj pfn_t pfn, 3935ae115bc7Smrj caddr_t addr, 3936ae115bc7Smrj hat_mempte_t pte_pa, 3937ae115bc7Smrj uint_t attr, 3938ae115bc7Smrj uint_t flags) 39397c478bd9Sstevel@tonic-gate { 39407c478bd9Sstevel@tonic-gate uintptr_t va = (uintptr_t)addr; 39417c478bd9Sstevel@tonic-gate x86pte_t pte; 39427c478bd9Sstevel@tonic-gate 39437c478bd9Sstevel@tonic-gate /* 39447c478bd9Sstevel@tonic-gate * Remap the given PTE to the new page's PFN. Invalidate only 39457c478bd9Sstevel@tonic-gate * on this CPU. 39467c478bd9Sstevel@tonic-gate */ 39477c478bd9Sstevel@tonic-gate #ifdef DEBUG 39487c478bd9Sstevel@tonic-gate htable_t *ht; 39497c478bd9Sstevel@tonic-gate uint_t entry; 39507c478bd9Sstevel@tonic-gate 39517c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(va)); 39527c478bd9Sstevel@tonic-gate ASSERT(!IN_VA_HOLE(va)); 39537c478bd9Sstevel@tonic-gate ht = htable_getpte(kas.a_hat, va, &entry, NULL, 0); 39547c478bd9Sstevel@tonic-gate ASSERT(ht != NULL); 39557c478bd9Sstevel@tonic-gate ASSERT(ht->ht_level == 0); 39567c478bd9Sstevel@tonic-gate ASSERT(ht->ht_valid_cnt > 0); 3957ae115bc7Smrj ASSERT(ht->ht_pfn == mmu_btop(pte_pa)); 39587c478bd9Sstevel@tonic-gate htable_release(ht); 39597c478bd9Sstevel@tonic-gate #endif 3960843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 39617c478bd9Sstevel@tonic-gate pte = hati_mkpte(pfn, attr, 0, flags); 3962843e1988Sjohnlev #ifdef __xpv 3963843e1988Sjohnlev if (HYPERVISOR_update_va_mapping(va, pte, UVMF_INVLPG | UVMF_LOCAL)) 3964843e1988Sjohnlev panic("HYPERVISOR_update_va_mapping() failed"); 3965843e1988Sjohnlev #else 3966ae115bc7Smrj { 3967ae115bc7Smrj x86pte_t *pteptr; 3968ae115bc7Smrj 3969ae115bc7Smrj pteptr = x86pte_mapin(mmu_btop(pte_pa), 3970ae115bc7Smrj (pte_pa & MMU_PAGEOFFSET) >> mmu.pte_size_shift, NULL); 3971ae115bc7Smrj if (mmu.pae_hat) 3972ae115bc7Smrj *(x86pte_t *)pteptr = pte; 3973ae115bc7Smrj else 3974ae115bc7Smrj *(x86pte32_t *)pteptr = (x86pte32_t)pte; 3975ae115bc7Smrj mmu_tlbflush_entry(addr); 3976ae115bc7Smrj x86pte_mapout(); 3977ae115bc7Smrj } 3978843e1988Sjohnlev #endif 3979843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 39807c478bd9Sstevel@tonic-gate } 39817c478bd9Sstevel@tonic-gate 39827c478bd9Sstevel@tonic-gate 39837c478bd9Sstevel@tonic-gate 39847c478bd9Sstevel@tonic-gate /* 39857c478bd9Sstevel@tonic-gate * Hat locking functions 39867c478bd9Sstevel@tonic-gate * XXX - these two functions are currently being used by hatstats 39877c478bd9Sstevel@tonic-gate * they can be removed by using a per-as mutex for hatstats. 39887c478bd9Sstevel@tonic-gate */ 39897c478bd9Sstevel@tonic-gate void 39907c478bd9Sstevel@tonic-gate hat_enter(hat_t *hat) 39917c478bd9Sstevel@tonic-gate { 39927c478bd9Sstevel@tonic-gate mutex_enter(&hat->hat_mutex); 39937c478bd9Sstevel@tonic-gate } 39947c478bd9Sstevel@tonic-gate 39957c478bd9Sstevel@tonic-gate void 39967c478bd9Sstevel@tonic-gate hat_exit(hat_t *hat) 39977c478bd9Sstevel@tonic-gate { 39987c478bd9Sstevel@tonic-gate mutex_exit(&hat->hat_mutex); 39997c478bd9Sstevel@tonic-gate } 40007c478bd9Sstevel@tonic-gate 40017c478bd9Sstevel@tonic-gate /* 4002ae115bc7Smrj * HAT part of cpu initialization. 40037c478bd9Sstevel@tonic-gate */ 40047c478bd9Sstevel@tonic-gate void 4005ae115bc7Smrj hat_cpu_online(struct cpu *cpup) 40067c478bd9Sstevel@tonic-gate { 4007ae115bc7Smrj if (cpup != CPU) { 4008ae115bc7Smrj x86pte_cpu_init(cpup); 4009ae115bc7Smrj hat_vlp_setup(cpup); 40107c478bd9Sstevel@tonic-gate } 4011ae115bc7Smrj CPUSET_ATOMIC_ADD(khat_cpuset, cpup->cpu_id); 40127c478bd9Sstevel@tonic-gate } 40137c478bd9Sstevel@tonic-gate 40147c478bd9Sstevel@tonic-gate /* 4015ae115bc7Smrj * HAT part of cpu deletion. 4016ae115bc7Smrj * (currently, we only call this after the cpu is safely passivated.) 40177c478bd9Sstevel@tonic-gate */ 40187c478bd9Sstevel@tonic-gate void 4019ae115bc7Smrj hat_cpu_offline(struct cpu *cpup) 40207c478bd9Sstevel@tonic-gate { 4021ae115bc7Smrj ASSERT(cpup != CPU); 4022ae115bc7Smrj 4023ae115bc7Smrj CPUSET_ATOMIC_DEL(khat_cpuset, cpup->cpu_id); 4024ae115bc7Smrj x86pte_cpu_fini(cpup); 4025ae115bc7Smrj hat_vlp_teardown(cpup); 40267c478bd9Sstevel@tonic-gate } 40277c478bd9Sstevel@tonic-gate 40287c478bd9Sstevel@tonic-gate /* 40297c478bd9Sstevel@tonic-gate * Function called after all CPUs are brought online. 40307c478bd9Sstevel@tonic-gate * Used to remove low address boot mappings. 40317c478bd9Sstevel@tonic-gate */ 40327c478bd9Sstevel@tonic-gate void 40337c478bd9Sstevel@tonic-gate clear_boot_mappings(uintptr_t low, uintptr_t high) 40347c478bd9Sstevel@tonic-gate { 40357c478bd9Sstevel@tonic-gate uintptr_t vaddr = low; 40367c478bd9Sstevel@tonic-gate htable_t *ht = NULL; 40377c478bd9Sstevel@tonic-gate level_t level; 40387c478bd9Sstevel@tonic-gate uint_t entry; 40397c478bd9Sstevel@tonic-gate x86pte_t pte; 40407c478bd9Sstevel@tonic-gate 40417c478bd9Sstevel@tonic-gate /* 40427c478bd9Sstevel@tonic-gate * On 1st CPU we can unload the prom mappings, basically we blow away 4043ae115bc7Smrj * all virtual mappings under _userlimit. 40447c478bd9Sstevel@tonic-gate */ 40457c478bd9Sstevel@tonic-gate while (vaddr < high) { 40467c478bd9Sstevel@tonic-gate pte = htable_walk(kas.a_hat, &ht, &vaddr, high); 40477c478bd9Sstevel@tonic-gate if (ht == NULL) 40487c478bd9Sstevel@tonic-gate break; 40497c478bd9Sstevel@tonic-gate 40507c478bd9Sstevel@tonic-gate level = ht->ht_level; 40517c478bd9Sstevel@tonic-gate entry = htable_va2entry(vaddr, ht); 40527c478bd9Sstevel@tonic-gate ASSERT(level <= mmu.max_page_level); 40537c478bd9Sstevel@tonic-gate ASSERT(PTE_ISPAGE(pte, level)); 40547c478bd9Sstevel@tonic-gate 40557c478bd9Sstevel@tonic-gate /* 40567c478bd9Sstevel@tonic-gate * Unload the mapping from the page tables. 40577c478bd9Sstevel@tonic-gate */ 4058ae115bc7Smrj (void) x86pte_inval(ht, entry, 0, NULL); 40597c478bd9Sstevel@tonic-gate ASSERT(ht->ht_valid_cnt > 0); 40607c478bd9Sstevel@tonic-gate HTABLE_DEC(ht->ht_valid_cnt); 40617c478bd9Sstevel@tonic-gate PGCNT_DEC(ht->ht_hat, ht->ht_level); 40627c478bd9Sstevel@tonic-gate 40637c478bd9Sstevel@tonic-gate vaddr += LEVEL_SIZE(ht->ht_level); 40647c478bd9Sstevel@tonic-gate } 40657c478bd9Sstevel@tonic-gate if (ht) 40667c478bd9Sstevel@tonic-gate htable_release(ht); 40677c478bd9Sstevel@tonic-gate } 40687c478bd9Sstevel@tonic-gate 40697c478bd9Sstevel@tonic-gate /* 40707c478bd9Sstevel@tonic-gate * Atomically update a new translation for a single page. If the 40717c478bd9Sstevel@tonic-gate * currently installed PTE doesn't match the value we expect to find, 40727c478bd9Sstevel@tonic-gate * it's not updated and we return the PTE we found. 40737c478bd9Sstevel@tonic-gate * 40747c478bd9Sstevel@tonic-gate * If activating nosync or NOWRITE and the page was modified we need to sync 40757c478bd9Sstevel@tonic-gate * with the page_t. Also sync with page_t if clearing ref/mod bits. 40767c478bd9Sstevel@tonic-gate */ 40777c478bd9Sstevel@tonic-gate static x86pte_t 40787c478bd9Sstevel@tonic-gate hati_update_pte(htable_t *ht, uint_t entry, x86pte_t expected, x86pte_t new) 40797c478bd9Sstevel@tonic-gate { 40807c478bd9Sstevel@tonic-gate page_t *pp; 40817c478bd9Sstevel@tonic-gate uint_t rm = 0; 40827c478bd9Sstevel@tonic-gate x86pte_t replaced; 40837c478bd9Sstevel@tonic-gate 4084ae115bc7Smrj if (PTE_GET(expected, PT_SOFTWARE) < PT_NOSYNC && 40857c478bd9Sstevel@tonic-gate PTE_GET(expected, PT_MOD | PT_REF) && 40867c478bd9Sstevel@tonic-gate (PTE_GET(new, PT_NOSYNC) || !PTE_GET(new, PT_WRITABLE) || 4087250b7ff9Sjosephb !PTE_GET(new, PT_MOD | PT_REF))) { 40887c478bd9Sstevel@tonic-gate 4089ae115bc7Smrj ASSERT(!pfn_is_foreign(PTE2PFN(expected, ht->ht_level))); 40907c478bd9Sstevel@tonic-gate pp = page_numtopp_nolock(PTE2PFN(expected, ht->ht_level)); 40917c478bd9Sstevel@tonic-gate ASSERT(pp != NULL); 40927c478bd9Sstevel@tonic-gate if (PTE_GET(expected, PT_MOD)) 40937c478bd9Sstevel@tonic-gate rm |= P_MOD; 40947c478bd9Sstevel@tonic-gate if (PTE_GET(expected, PT_REF)) 40957c478bd9Sstevel@tonic-gate rm |= P_REF; 40967c478bd9Sstevel@tonic-gate PTE_CLR(new, PT_MOD | PT_REF); 40977c478bd9Sstevel@tonic-gate } 40987c478bd9Sstevel@tonic-gate 40997c478bd9Sstevel@tonic-gate replaced = x86pte_update(ht, entry, expected, new); 41007c478bd9Sstevel@tonic-gate if (replaced != expected) 41017c478bd9Sstevel@tonic-gate return (replaced); 41027c478bd9Sstevel@tonic-gate 41037c478bd9Sstevel@tonic-gate if (rm) { 41047c478bd9Sstevel@tonic-gate /* 41057c478bd9Sstevel@tonic-gate * sync to all constituent pages of a large page 41067c478bd9Sstevel@tonic-gate */ 41077c478bd9Sstevel@tonic-gate pgcnt_t pgcnt = page_get_pagecnt(ht->ht_level); 41087c478bd9Sstevel@tonic-gate ASSERT(IS_P2ALIGNED(pp->p_pagenum, pgcnt)); 41097c478bd9Sstevel@tonic-gate while (pgcnt-- > 0) { 41107c478bd9Sstevel@tonic-gate /* 41117c478bd9Sstevel@tonic-gate * hat_page_demote() can't decrease 41127c478bd9Sstevel@tonic-gate * pszc below this mapping size 41137c478bd9Sstevel@tonic-gate * since large mapping existed after we 41147c478bd9Sstevel@tonic-gate * took mlist lock. 41157c478bd9Sstevel@tonic-gate */ 41167c478bd9Sstevel@tonic-gate ASSERT(pp->p_szc >= ht->ht_level); 41177c478bd9Sstevel@tonic-gate hat_page_setattr(pp, rm); 41187c478bd9Sstevel@tonic-gate ++pp; 41197c478bd9Sstevel@tonic-gate } 41207c478bd9Sstevel@tonic-gate } 41217c478bd9Sstevel@tonic-gate 41227c478bd9Sstevel@tonic-gate return (0); 41237c478bd9Sstevel@tonic-gate } 41247c478bd9Sstevel@tonic-gate 412505d3dc4bSpaulsan /* ARGSUSED */ 412605d3dc4bSpaulsan void 41277dacfc44Spaulsan hat_join_srd(struct hat *hat, vnode_t *evp) 412805d3dc4bSpaulsan { 412905d3dc4bSpaulsan } 413005d3dc4bSpaulsan 413105d3dc4bSpaulsan /* ARGSUSED */ 413205d3dc4bSpaulsan hat_region_cookie_t 41337dacfc44Spaulsan hat_join_region(struct hat *hat, 413405d3dc4bSpaulsan caddr_t r_saddr, 413505d3dc4bSpaulsan size_t r_size, 413605d3dc4bSpaulsan void *r_obj, 413705d3dc4bSpaulsan u_offset_t r_objoff, 413805d3dc4bSpaulsan uchar_t r_perm, 413905d3dc4bSpaulsan uchar_t r_pgszc, 414005d3dc4bSpaulsan hat_rgn_cb_func_t r_cb_function, 414105d3dc4bSpaulsan uint_t flags) 414205d3dc4bSpaulsan { 414305d3dc4bSpaulsan panic("No shared region support on x86"); 414405d3dc4bSpaulsan return (HAT_INVALID_REGION_COOKIE); 414505d3dc4bSpaulsan } 414605d3dc4bSpaulsan 414705d3dc4bSpaulsan /* ARGSUSED */ 414805d3dc4bSpaulsan void 41497dacfc44Spaulsan hat_leave_region(struct hat *hat, hat_region_cookie_t rcookie, uint_t flags) 415005d3dc4bSpaulsan { 415105d3dc4bSpaulsan panic("No shared region support on x86"); 415205d3dc4bSpaulsan } 415305d3dc4bSpaulsan 415405d3dc4bSpaulsan /* ARGSUSED */ 415505d3dc4bSpaulsan void 41567dacfc44Spaulsan hat_dup_region(struct hat *hat, hat_region_cookie_t rcookie) 415705d3dc4bSpaulsan { 415805d3dc4bSpaulsan panic("No shared region support on x86"); 415905d3dc4bSpaulsan } 416005d3dc4bSpaulsan 416105d3dc4bSpaulsan 41627c478bd9Sstevel@tonic-gate /* 41637c478bd9Sstevel@tonic-gate * Kernel Physical Mapping (kpm) facility 41647c478bd9Sstevel@tonic-gate * 41657c478bd9Sstevel@tonic-gate * Most of the routines needed to support segkpm are almost no-ops on the 41667c478bd9Sstevel@tonic-gate * x86 platform. We map in the entire segment when it is created and leave 41677c478bd9Sstevel@tonic-gate * it mapped in, so there is no additional work required to set up and tear 41687c478bd9Sstevel@tonic-gate * down individual mappings. All of these routines were created to support 41697c478bd9Sstevel@tonic-gate * SPARC platforms that have to avoid aliasing in their virtually indexed 41707c478bd9Sstevel@tonic-gate * caches. 41717c478bd9Sstevel@tonic-gate * 41727c478bd9Sstevel@tonic-gate * Most of the routines have sanity checks in them (e.g. verifying that the 41737c478bd9Sstevel@tonic-gate * passed-in page is locked). We don't actually care about most of these 41747c478bd9Sstevel@tonic-gate * checks on x86, but we leave them in place to identify problems in the 41757c478bd9Sstevel@tonic-gate * upper levels. 41767c478bd9Sstevel@tonic-gate */ 41777c478bd9Sstevel@tonic-gate 41787c478bd9Sstevel@tonic-gate /* 41797c478bd9Sstevel@tonic-gate * Map in a locked page and return the vaddr. 41807c478bd9Sstevel@tonic-gate */ 41817c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 41827c478bd9Sstevel@tonic-gate caddr_t 41837c478bd9Sstevel@tonic-gate hat_kpm_mapin(struct page *pp, struct kpme *kpme) 41847c478bd9Sstevel@tonic-gate { 41857c478bd9Sstevel@tonic-gate caddr_t vaddr; 41867c478bd9Sstevel@tonic-gate 41877c478bd9Sstevel@tonic-gate #ifdef DEBUG 41887c478bd9Sstevel@tonic-gate if (kpm_enable == 0) { 41897c478bd9Sstevel@tonic-gate cmn_err(CE_WARN, "hat_kpm_mapin: kpm_enable not set\n"); 41907c478bd9Sstevel@tonic-gate return ((caddr_t)NULL); 41917c478bd9Sstevel@tonic-gate } 41927c478bd9Sstevel@tonic-gate 41937c478bd9Sstevel@tonic-gate if (pp == NULL || PAGE_LOCKED(pp) == 0) { 41947c478bd9Sstevel@tonic-gate cmn_err(CE_WARN, "hat_kpm_mapin: pp zero or not locked\n"); 41957c478bd9Sstevel@tonic-gate return ((caddr_t)NULL); 41967c478bd9Sstevel@tonic-gate } 41977c478bd9Sstevel@tonic-gate #endif 41987c478bd9Sstevel@tonic-gate 41997c478bd9Sstevel@tonic-gate vaddr = hat_kpm_page2va(pp, 1); 42007c478bd9Sstevel@tonic-gate 42017c478bd9Sstevel@tonic-gate return (vaddr); 42027c478bd9Sstevel@tonic-gate } 42037c478bd9Sstevel@tonic-gate 42047c478bd9Sstevel@tonic-gate /* 42057c478bd9Sstevel@tonic-gate * Mapout a locked page. 42067c478bd9Sstevel@tonic-gate */ 42077c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 42087c478bd9Sstevel@tonic-gate void 42097c478bd9Sstevel@tonic-gate hat_kpm_mapout(struct page *pp, struct kpme *kpme, caddr_t vaddr) 42107c478bd9Sstevel@tonic-gate { 42117c478bd9Sstevel@tonic-gate #ifdef DEBUG 42127c478bd9Sstevel@tonic-gate if (kpm_enable == 0) { 42137c478bd9Sstevel@tonic-gate cmn_err(CE_WARN, "hat_kpm_mapout: kpm_enable not set\n"); 42147c478bd9Sstevel@tonic-gate return; 42157c478bd9Sstevel@tonic-gate } 42167c478bd9Sstevel@tonic-gate 42177c478bd9Sstevel@tonic-gate if (IS_KPM_ADDR(vaddr) == 0) { 42187c478bd9Sstevel@tonic-gate cmn_err(CE_WARN, "hat_kpm_mapout: no kpm address\n"); 42197c478bd9Sstevel@tonic-gate return; 42207c478bd9Sstevel@tonic-gate } 42217c478bd9Sstevel@tonic-gate 42227c478bd9Sstevel@tonic-gate if (pp == NULL || PAGE_LOCKED(pp) == 0) { 42237c478bd9Sstevel@tonic-gate cmn_err(CE_WARN, "hat_kpm_mapout: page zero or not locked\n"); 42247c478bd9Sstevel@tonic-gate return; 42257c478bd9Sstevel@tonic-gate } 42267c478bd9Sstevel@tonic-gate #endif 42277c478bd9Sstevel@tonic-gate } 42287c478bd9Sstevel@tonic-gate 4229d20abfaaSPavel Tatashin /* 4230d20abfaaSPavel Tatashin * hat_kpm_mapin_pfn is used to obtain a kpm mapping for physical 4231d20abfaaSPavel Tatashin * memory addresses that are not described by a page_t. It can 4232d20abfaaSPavel Tatashin * also be used for normal pages that are not locked, but beware 4233d20abfaaSPavel Tatashin * this is dangerous - no locking is performed, so the identity of 4234d20abfaaSPavel Tatashin * the page could change. hat_kpm_mapin_pfn is not supported when 4235d20abfaaSPavel Tatashin * vac_colors > 1, because the chosen va depends on the page identity, 4236d20abfaaSPavel Tatashin * which could change. 4237d20abfaaSPavel Tatashin * The caller must only pass pfn's for valid physical addresses; violation 4238d20abfaaSPavel Tatashin * of this rule will cause panic. 4239d20abfaaSPavel Tatashin */ 4240d20abfaaSPavel Tatashin caddr_t 4241d20abfaaSPavel Tatashin hat_kpm_mapin_pfn(pfn_t pfn) 4242d20abfaaSPavel Tatashin { 4243d20abfaaSPavel Tatashin caddr_t paddr, vaddr; 4244d20abfaaSPavel Tatashin 4245d20abfaaSPavel Tatashin if (kpm_enable == 0) 4246d20abfaaSPavel Tatashin return ((caddr_t)NULL); 4247d20abfaaSPavel Tatashin 4248d20abfaaSPavel Tatashin paddr = (caddr_t)ptob(pfn); 4249d20abfaaSPavel Tatashin vaddr = (uintptr_t)kpm_vbase + paddr; 4250d20abfaaSPavel Tatashin 4251d20abfaaSPavel Tatashin return ((caddr_t)vaddr); 4252d20abfaaSPavel Tatashin } 4253d20abfaaSPavel Tatashin 4254d20abfaaSPavel Tatashin /*ARGSUSED*/ 4255d20abfaaSPavel Tatashin void 4256d20abfaaSPavel Tatashin hat_kpm_mapout_pfn(pfn_t pfn) 4257d20abfaaSPavel Tatashin { 4258d20abfaaSPavel Tatashin /* empty */ 4259d20abfaaSPavel Tatashin } 4260d20abfaaSPavel Tatashin 42617c478bd9Sstevel@tonic-gate /* 42627c478bd9Sstevel@tonic-gate * Return the kpm virtual address for a specific pfn 42637c478bd9Sstevel@tonic-gate */ 42647c478bd9Sstevel@tonic-gate caddr_t 42657c478bd9Sstevel@tonic-gate hat_kpm_pfn2va(pfn_t pfn) 42667c478bd9Sstevel@tonic-gate { 4267ae115bc7Smrj uintptr_t vaddr = (uintptr_t)kpm_vbase + mmu_ptob(pfn); 42687c478bd9Sstevel@tonic-gate 4269d2b85481Srscott ASSERT(!pfn_is_foreign(pfn)); 42707c478bd9Sstevel@tonic-gate return ((caddr_t)vaddr); 42717c478bd9Sstevel@tonic-gate } 42727c478bd9Sstevel@tonic-gate 42737c478bd9Sstevel@tonic-gate /* 42747c478bd9Sstevel@tonic-gate * Return the kpm virtual address for the page at pp. 42757c478bd9Sstevel@tonic-gate */ 42767c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 42777c478bd9Sstevel@tonic-gate caddr_t 42787c478bd9Sstevel@tonic-gate hat_kpm_page2va(struct page *pp, int checkswap) 42797c478bd9Sstevel@tonic-gate { 42807c478bd9Sstevel@tonic-gate return (hat_kpm_pfn2va(pp->p_pagenum)); 42817c478bd9Sstevel@tonic-gate } 42827c478bd9Sstevel@tonic-gate 42837c478bd9Sstevel@tonic-gate /* 42847c478bd9Sstevel@tonic-gate * Return the page frame number for the kpm virtual address vaddr. 42857c478bd9Sstevel@tonic-gate */ 42867c478bd9Sstevel@tonic-gate pfn_t 42877c478bd9Sstevel@tonic-gate hat_kpm_va2pfn(caddr_t vaddr) 42887c478bd9Sstevel@tonic-gate { 42897c478bd9Sstevel@tonic-gate pfn_t pfn; 42907c478bd9Sstevel@tonic-gate 42917c478bd9Sstevel@tonic-gate ASSERT(IS_KPM_ADDR(vaddr)); 42927c478bd9Sstevel@tonic-gate 42937c478bd9Sstevel@tonic-gate pfn = (pfn_t)btop(vaddr - kpm_vbase); 42947c478bd9Sstevel@tonic-gate 42957c478bd9Sstevel@tonic-gate return (pfn); 42967c478bd9Sstevel@tonic-gate } 42977c478bd9Sstevel@tonic-gate 42987c478bd9Sstevel@tonic-gate 42997c478bd9Sstevel@tonic-gate /* 43007c478bd9Sstevel@tonic-gate * Return the page for the kpm virtual address vaddr. 43017c478bd9Sstevel@tonic-gate */ 43027c478bd9Sstevel@tonic-gate page_t * 43037c478bd9Sstevel@tonic-gate hat_kpm_vaddr2page(caddr_t vaddr) 43047c478bd9Sstevel@tonic-gate { 43057c478bd9Sstevel@tonic-gate pfn_t pfn; 43067c478bd9Sstevel@tonic-gate 43077c478bd9Sstevel@tonic-gate ASSERT(IS_KPM_ADDR(vaddr)); 43087c478bd9Sstevel@tonic-gate 43097c478bd9Sstevel@tonic-gate pfn = hat_kpm_va2pfn(vaddr); 43107c478bd9Sstevel@tonic-gate 43117c478bd9Sstevel@tonic-gate return (page_numtopp_nolock(pfn)); 43127c478bd9Sstevel@tonic-gate } 43137c478bd9Sstevel@tonic-gate 43147c478bd9Sstevel@tonic-gate /* 43157c478bd9Sstevel@tonic-gate * hat_kpm_fault is called from segkpm_fault when we take a page fault on a 43167c478bd9Sstevel@tonic-gate * KPM page. This should never happen on x86 43177c478bd9Sstevel@tonic-gate */ 43187c478bd9Sstevel@tonic-gate int 43197c478bd9Sstevel@tonic-gate hat_kpm_fault(hat_t *hat, caddr_t vaddr) 43207c478bd9Sstevel@tonic-gate { 4321903a11ebSrh panic("pagefault in seg_kpm. hat: 0x%p vaddr: 0x%p", 4322903a11ebSrh (void *)hat, (void *)vaddr); 43237c478bd9Sstevel@tonic-gate 43247c478bd9Sstevel@tonic-gate return (0); 43257c478bd9Sstevel@tonic-gate } 43267c478bd9Sstevel@tonic-gate 43277c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 43287c478bd9Sstevel@tonic-gate void 43297c478bd9Sstevel@tonic-gate hat_kpm_mseghash_clear(int nentries) 43307c478bd9Sstevel@tonic-gate {} 43317c478bd9Sstevel@tonic-gate 43327c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 43337c478bd9Sstevel@tonic-gate void 43347c478bd9Sstevel@tonic-gate hat_kpm_mseghash_update(pgcnt_t inx, struct memseg *msp) 43357c478bd9Sstevel@tonic-gate {} 4336843e1988Sjohnlev 4337843e1988Sjohnlev #ifdef __xpv 4338843e1988Sjohnlev /* 4339843e1988Sjohnlev * There are specific Hypervisor calls to establish and remove mappings 4340843e1988Sjohnlev * to grant table references and the privcmd driver. We have to ensure 4341843e1988Sjohnlev * that a page table actually exists. 4342843e1988Sjohnlev */ 4343843e1988Sjohnlev void 43447eea693dSMark Johnson hat_prepare_mapping(hat_t *hat, caddr_t addr, uint64_t *pte_ma) 4345843e1988Sjohnlev { 43467eea693dSMark Johnson maddr_t base_ma; 43477eea693dSMark Johnson htable_t *ht; 43487eea693dSMark Johnson uint_t entry; 43497eea693dSMark Johnson 4350843e1988Sjohnlev ASSERT(IS_P2ALIGNED((uintptr_t)addr, MMU_PAGESIZE)); 4351551bc2a6Smrj XPV_DISALLOW_MIGRATE(); 43527eea693dSMark Johnson ht = htable_create(hat, (uintptr_t)addr, 0, NULL); 43537eea693dSMark Johnson 43547eea693dSMark Johnson /* 43557eea693dSMark Johnson * if an address for pte_ma is passed in, return the MA of the pte 43567eea693dSMark Johnson * for this specific address. This address is only valid as long 43577eea693dSMark Johnson * as the htable stays locked. 43587eea693dSMark Johnson */ 43597eea693dSMark Johnson if (pte_ma != NULL) { 43607eea693dSMark Johnson entry = htable_va2entry((uintptr_t)addr, ht); 43617eea693dSMark Johnson base_ma = pa_to_ma(ptob(ht->ht_pfn)); 43627eea693dSMark Johnson *pte_ma = base_ma + (entry << mmu.pte_size_shift); 43637eea693dSMark Johnson } 4364551bc2a6Smrj XPV_ALLOW_MIGRATE(); 4365843e1988Sjohnlev } 4366843e1988Sjohnlev 4367843e1988Sjohnlev void 4368843e1988Sjohnlev hat_release_mapping(hat_t *hat, caddr_t addr) 4369843e1988Sjohnlev { 4370843e1988Sjohnlev htable_t *ht; 4371843e1988Sjohnlev 4372843e1988Sjohnlev ASSERT(IS_P2ALIGNED((uintptr_t)addr, MMU_PAGESIZE)); 4373551bc2a6Smrj XPV_DISALLOW_MIGRATE(); 4374843e1988Sjohnlev ht = htable_lookup(hat, (uintptr_t)addr, 0); 4375843e1988Sjohnlev ASSERT(ht != NULL); 4376843e1988Sjohnlev ASSERT(ht->ht_busy >= 2); 4377843e1988Sjohnlev htable_release(ht); 4378843e1988Sjohnlev htable_release(ht); 4379551bc2a6Smrj XPV_ALLOW_MIGRATE(); 4380551bc2a6Smrj } 4381843e1988Sjohnlev #endif 4382