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 */ 21ae115bc7Smrj 227c478bd9Sstevel@tonic-gate /* 236c9930aeSJoe Bonasera * Copyright (c) 2004, 2010, Oracle and/or its affiliates. All rights reserved. 24*a6a74e0eSMatthew Ahrens * Copyright (c) 2014 by Delphix. All rights reserved. 257c478bd9Sstevel@tonic-gate */ 267c478bd9Sstevel@tonic-gate 277c478bd9Sstevel@tonic-gate #include <sys/types.h> 287c478bd9Sstevel@tonic-gate #include <sys/sysmacros.h> 297c478bd9Sstevel@tonic-gate #include <sys/kmem.h> 307c478bd9Sstevel@tonic-gate #include <sys/atomic.h> 317c478bd9Sstevel@tonic-gate #include <sys/bitmap.h> 327c478bd9Sstevel@tonic-gate #include <sys/machparam.h> 337c478bd9Sstevel@tonic-gate #include <sys/machsystm.h> 347c478bd9Sstevel@tonic-gate #include <sys/mman.h> 357c478bd9Sstevel@tonic-gate #include <sys/systm.h> 367c478bd9Sstevel@tonic-gate #include <sys/cpuvar.h> 377c478bd9Sstevel@tonic-gate #include <sys/thread.h> 387c478bd9Sstevel@tonic-gate #include <sys/proc.h> 397c478bd9Sstevel@tonic-gate #include <sys/cpu.h> 407c478bd9Sstevel@tonic-gate #include <sys/kmem.h> 417c478bd9Sstevel@tonic-gate #include <sys/disp.h> 427c478bd9Sstevel@tonic-gate #include <sys/vmem.h> 437c478bd9Sstevel@tonic-gate #include <sys/vmsystm.h> 447c478bd9Sstevel@tonic-gate #include <sys/promif.h> 457c478bd9Sstevel@tonic-gate #include <sys/var.h> 467c478bd9Sstevel@tonic-gate #include <sys/x86_archext.h> 47ae115bc7Smrj #include <sys/archsystm.h> 487c478bd9Sstevel@tonic-gate #include <sys/bootconf.h> 497c478bd9Sstevel@tonic-gate #include <sys/dumphdr.h> 507c478bd9Sstevel@tonic-gate #include <vm/seg_kmem.h> 517c478bd9Sstevel@tonic-gate #include <vm/seg_kpm.h> 527c478bd9Sstevel@tonic-gate #include <vm/hat.h> 537c478bd9Sstevel@tonic-gate #include <vm/hat_i86.h> 547c478bd9Sstevel@tonic-gate #include <sys/cmn_err.h> 55843e1988Sjohnlev #include <sys/panic.h> 56843e1988Sjohnlev 57843e1988Sjohnlev #ifdef __xpv 58843e1988Sjohnlev #include <sys/hypervisor.h> 59843e1988Sjohnlev #include <sys/xpv_panic.h> 60843e1988Sjohnlev #endif 617c478bd9Sstevel@tonic-gate 62ae115bc7Smrj #include <sys/bootinfo.h> 63ae115bc7Smrj #include <vm/kboot_mmu.h> 64ae115bc7Smrj 65ae115bc7Smrj static void x86pte_zero(htable_t *dest, uint_t entry, uint_t count); 66ae115bc7Smrj 677c478bd9Sstevel@tonic-gate kmem_cache_t *htable_cache; 687c478bd9Sstevel@tonic-gate 697c478bd9Sstevel@tonic-gate /* 707c478bd9Sstevel@tonic-gate * The variable htable_reserve_amount, rather than HTABLE_RESERVE_AMOUNT, 717c478bd9Sstevel@tonic-gate * is used in order to facilitate testing of the htable_steal() code. 727c478bd9Sstevel@tonic-gate * By resetting htable_reserve_amount to a lower value, we can force 737c478bd9Sstevel@tonic-gate * stealing to occur. The reserve amount is a guess to get us through boot. 747c478bd9Sstevel@tonic-gate */ 757c478bd9Sstevel@tonic-gate #define HTABLE_RESERVE_AMOUNT (200) 767c478bd9Sstevel@tonic-gate uint_t htable_reserve_amount = HTABLE_RESERVE_AMOUNT; 777c478bd9Sstevel@tonic-gate kmutex_t htable_reserve_mutex; 787c478bd9Sstevel@tonic-gate uint_t htable_reserve_cnt; 797c478bd9Sstevel@tonic-gate htable_t *htable_reserve_pool; 807c478bd9Sstevel@tonic-gate 817c478bd9Sstevel@tonic-gate /* 82a85a6733Sjosephb * Used to hand test htable_steal(). 837c478bd9Sstevel@tonic-gate */ 84a85a6733Sjosephb #ifdef DEBUG 85a85a6733Sjosephb ulong_t force_steal = 0; 86a85a6733Sjosephb ulong_t ptable_cnt = 0; 87a85a6733Sjosephb #endif 88a85a6733Sjosephb 89a85a6733Sjosephb /* 90a85a6733Sjosephb * This variable is so that we can tune this via /etc/system 91a85a6733Sjosephb * Any value works, but a power of two <= mmu.ptes_per_table is best. 92a85a6733Sjosephb */ 93a85a6733Sjosephb uint_t htable_steal_passes = 8; 947c478bd9Sstevel@tonic-gate 957c478bd9Sstevel@tonic-gate /* 967c478bd9Sstevel@tonic-gate * mutex stuff for access to htable hash 977c478bd9Sstevel@tonic-gate */ 987c478bd9Sstevel@tonic-gate #define NUM_HTABLE_MUTEX 128 997c478bd9Sstevel@tonic-gate kmutex_t htable_mutex[NUM_HTABLE_MUTEX]; 1007c478bd9Sstevel@tonic-gate #define HTABLE_MUTEX_HASH(h) ((h) & (NUM_HTABLE_MUTEX - 1)) 1017c478bd9Sstevel@tonic-gate 1027c478bd9Sstevel@tonic-gate #define HTABLE_ENTER(h) mutex_enter(&htable_mutex[HTABLE_MUTEX_HASH(h)]); 1037c478bd9Sstevel@tonic-gate #define HTABLE_EXIT(h) mutex_exit(&htable_mutex[HTABLE_MUTEX_HASH(h)]); 1047c478bd9Sstevel@tonic-gate 1057c478bd9Sstevel@tonic-gate /* 1067c478bd9Sstevel@tonic-gate * forward declarations 1077c478bd9Sstevel@tonic-gate */ 1087c478bd9Sstevel@tonic-gate static void link_ptp(htable_t *higher, htable_t *new, uintptr_t vaddr); 1097c478bd9Sstevel@tonic-gate static void unlink_ptp(htable_t *higher, htable_t *old, uintptr_t vaddr); 1107c478bd9Sstevel@tonic-gate static void htable_free(htable_t *ht); 111ae115bc7Smrj static x86pte_t *x86pte_access_pagetable(htable_t *ht, uint_t index); 1127c478bd9Sstevel@tonic-gate static void x86pte_release_pagetable(htable_t *ht); 1137c478bd9Sstevel@tonic-gate static x86pte_t x86pte_cas(htable_t *ht, uint_t entry, x86pte_t old, 1147c478bd9Sstevel@tonic-gate x86pte_t new); 1157c478bd9Sstevel@tonic-gate 1167c478bd9Sstevel@tonic-gate /* 1177c478bd9Sstevel@tonic-gate * A counter to track if we are stealing or reaping htables. When non-zero 1187c478bd9Sstevel@tonic-gate * htable_free() will directly free htables (either to the reserve or kmem) 1197c478bd9Sstevel@tonic-gate * instead of putting them in a hat's htable cache. 1207c478bd9Sstevel@tonic-gate */ 1217c478bd9Sstevel@tonic-gate uint32_t htable_dont_cache = 0; 1227c478bd9Sstevel@tonic-gate 1237c478bd9Sstevel@tonic-gate /* 1247c478bd9Sstevel@tonic-gate * Track the number of active pagetables, so we can know how many to reap 1257c478bd9Sstevel@tonic-gate */ 1267c478bd9Sstevel@tonic-gate static uint32_t active_ptables = 0; 1277c478bd9Sstevel@tonic-gate 128843e1988Sjohnlev #ifdef __xpv 129843e1988Sjohnlev /* 130843e1988Sjohnlev * Deal with hypervisor complications. 131843e1988Sjohnlev */ 132843e1988Sjohnlev void 133843e1988Sjohnlev xen_flush_va(caddr_t va) 134843e1988Sjohnlev { 135843e1988Sjohnlev struct mmuext_op t; 136843e1988Sjohnlev uint_t count; 137843e1988Sjohnlev 138843e1988Sjohnlev if (IN_XPV_PANIC()) { 139843e1988Sjohnlev mmu_tlbflush_entry((caddr_t)va); 140843e1988Sjohnlev } else { 141843e1988Sjohnlev t.cmd = MMUEXT_INVLPG_LOCAL; 142843e1988Sjohnlev t.arg1.linear_addr = (uintptr_t)va; 143843e1988Sjohnlev if (HYPERVISOR_mmuext_op(&t, 1, &count, DOMID_SELF) < 0) 144843e1988Sjohnlev panic("HYPERVISOR_mmuext_op() failed"); 145843e1988Sjohnlev ASSERT(count == 1); 146843e1988Sjohnlev } 147843e1988Sjohnlev } 148843e1988Sjohnlev 149843e1988Sjohnlev void 150843e1988Sjohnlev xen_gflush_va(caddr_t va, cpuset_t cpus) 151843e1988Sjohnlev { 152843e1988Sjohnlev struct mmuext_op t; 153843e1988Sjohnlev uint_t count; 154843e1988Sjohnlev 155843e1988Sjohnlev if (IN_XPV_PANIC()) { 156843e1988Sjohnlev mmu_tlbflush_entry((caddr_t)va); 157843e1988Sjohnlev return; 158843e1988Sjohnlev } 159843e1988Sjohnlev 160843e1988Sjohnlev t.cmd = MMUEXT_INVLPG_MULTI; 161843e1988Sjohnlev t.arg1.linear_addr = (uintptr_t)va; 162843e1988Sjohnlev /*LINTED: constant in conditional context*/ 163843e1988Sjohnlev set_xen_guest_handle(t.arg2.vcpumask, &cpus); 164843e1988Sjohnlev if (HYPERVISOR_mmuext_op(&t, 1, &count, DOMID_SELF) < 0) 165843e1988Sjohnlev panic("HYPERVISOR_mmuext_op() failed"); 166843e1988Sjohnlev ASSERT(count == 1); 167843e1988Sjohnlev } 168843e1988Sjohnlev 169843e1988Sjohnlev void 170843e1988Sjohnlev xen_flush_tlb() 171843e1988Sjohnlev { 172843e1988Sjohnlev struct mmuext_op t; 173843e1988Sjohnlev uint_t count; 174843e1988Sjohnlev 175843e1988Sjohnlev if (IN_XPV_PANIC()) { 176843e1988Sjohnlev xpv_panic_reload_cr3(); 177843e1988Sjohnlev } else { 178843e1988Sjohnlev t.cmd = MMUEXT_TLB_FLUSH_LOCAL; 179843e1988Sjohnlev if (HYPERVISOR_mmuext_op(&t, 1, &count, DOMID_SELF) < 0) 180843e1988Sjohnlev panic("HYPERVISOR_mmuext_op() failed"); 181843e1988Sjohnlev ASSERT(count == 1); 182843e1988Sjohnlev } 183843e1988Sjohnlev } 184843e1988Sjohnlev 185843e1988Sjohnlev void 186843e1988Sjohnlev xen_gflush_tlb(cpuset_t cpus) 187843e1988Sjohnlev { 188843e1988Sjohnlev struct mmuext_op t; 189843e1988Sjohnlev uint_t count; 190843e1988Sjohnlev 191843e1988Sjohnlev ASSERT(!IN_XPV_PANIC()); 192843e1988Sjohnlev t.cmd = MMUEXT_TLB_FLUSH_MULTI; 193843e1988Sjohnlev /*LINTED: constant in conditional context*/ 194843e1988Sjohnlev set_xen_guest_handle(t.arg2.vcpumask, &cpus); 195843e1988Sjohnlev if (HYPERVISOR_mmuext_op(&t, 1, &count, DOMID_SELF) < 0) 196843e1988Sjohnlev panic("HYPERVISOR_mmuext_op() failed"); 197843e1988Sjohnlev ASSERT(count == 1); 198843e1988Sjohnlev } 199843e1988Sjohnlev 200843e1988Sjohnlev /* 201843e1988Sjohnlev * Install/Adjust a kpm mapping under the hypervisor. 202843e1988Sjohnlev * Value of "how" should be: 203843e1988Sjohnlev * PT_WRITABLE | PT_VALID - regular kpm mapping 204843e1988Sjohnlev * PT_VALID - make mapping read-only 205843e1988Sjohnlev * 0 - remove mapping 206843e1988Sjohnlev * 207843e1988Sjohnlev * returns 0 on success. non-zero for failure. 208843e1988Sjohnlev */ 209843e1988Sjohnlev int 210843e1988Sjohnlev xen_kpm_page(pfn_t pfn, uint_t how) 211843e1988Sjohnlev { 212843e1988Sjohnlev paddr_t pa = mmu_ptob((paddr_t)pfn); 213843e1988Sjohnlev x86pte_t pte = PT_NOCONSIST | PT_REF | PT_MOD; 214843e1988Sjohnlev 215843e1988Sjohnlev if (kpm_vbase == NULL) 216843e1988Sjohnlev return (0); 217843e1988Sjohnlev 218843e1988Sjohnlev if (how) 219843e1988Sjohnlev pte |= pa_to_ma(pa) | how; 220843e1988Sjohnlev else 221843e1988Sjohnlev pte = 0; 222843e1988Sjohnlev return (HYPERVISOR_update_va_mapping((uintptr_t)kpm_vbase + pa, 223843e1988Sjohnlev pte, UVMF_INVLPG | UVMF_ALL)); 224843e1988Sjohnlev } 225843e1988Sjohnlev 226843e1988Sjohnlev void 227843e1988Sjohnlev xen_pin(pfn_t pfn, level_t lvl) 228843e1988Sjohnlev { 229843e1988Sjohnlev struct mmuext_op t; 230843e1988Sjohnlev uint_t count; 231843e1988Sjohnlev 232843e1988Sjohnlev t.cmd = MMUEXT_PIN_L1_TABLE + lvl; 233843e1988Sjohnlev t.arg1.mfn = pfn_to_mfn(pfn); 234843e1988Sjohnlev if (HYPERVISOR_mmuext_op(&t, 1, &count, DOMID_SELF) < 0) 235843e1988Sjohnlev panic("HYPERVISOR_mmuext_op() failed"); 236843e1988Sjohnlev ASSERT(count == 1); 237843e1988Sjohnlev } 238843e1988Sjohnlev 239843e1988Sjohnlev void 240843e1988Sjohnlev xen_unpin(pfn_t pfn) 241843e1988Sjohnlev { 242843e1988Sjohnlev struct mmuext_op t; 243843e1988Sjohnlev uint_t count; 244843e1988Sjohnlev 245843e1988Sjohnlev t.cmd = MMUEXT_UNPIN_TABLE; 246843e1988Sjohnlev t.arg1.mfn = pfn_to_mfn(pfn); 247843e1988Sjohnlev if (HYPERVISOR_mmuext_op(&t, 1, &count, DOMID_SELF) < 0) 248843e1988Sjohnlev panic("HYPERVISOR_mmuext_op() failed"); 249843e1988Sjohnlev ASSERT(count == 1); 250843e1988Sjohnlev } 251843e1988Sjohnlev 252843e1988Sjohnlev static void 253843e1988Sjohnlev xen_map(uint64_t pte, caddr_t va) 254843e1988Sjohnlev { 255843e1988Sjohnlev if (HYPERVISOR_update_va_mapping((uintptr_t)va, pte, 256843e1988Sjohnlev UVMF_INVLPG | UVMF_LOCAL)) 257843e1988Sjohnlev panic("HYPERVISOR_update_va_mapping() failed"); 258843e1988Sjohnlev } 259843e1988Sjohnlev #endif /* __xpv */ 260843e1988Sjohnlev 2617c478bd9Sstevel@tonic-gate /* 2627c478bd9Sstevel@tonic-gate * Allocate a memory page for a hardware page table. 2637c478bd9Sstevel@tonic-gate * 264ae115bc7Smrj * A wrapper around page_get_physical(), with some extra checks. 2657c478bd9Sstevel@tonic-gate */ 266ae115bc7Smrj static pfn_t 267a77271f8SVikram Hegde ptable_alloc(uintptr_t seed) 2687c478bd9Sstevel@tonic-gate { 2697c478bd9Sstevel@tonic-gate pfn_t pfn; 2707c478bd9Sstevel@tonic-gate page_t *pp; 2717c478bd9Sstevel@tonic-gate 272ae115bc7Smrj pfn = PFN_INVALID; 2737c478bd9Sstevel@tonic-gate 2747c478bd9Sstevel@tonic-gate /* 275ae115bc7Smrj * The first check is to see if there is memory in the system. If we 276ae115bc7Smrj * drop to throttlefree, then fail the ptable_alloc() and let the 277ae115bc7Smrj * stealing code kick in. Note that we have to do this test here, 278ae115bc7Smrj * since the test in page_create_throttle() would let the NOSLEEP 279ae115bc7Smrj * allocation go through and deplete the page reserves. 280a85a6733Sjosephb * 281a85a6733Sjosephb * The !NOMEMWAIT() lets pageout, fsflush, etc. skip this check. 2827c478bd9Sstevel@tonic-gate */ 283a85a6733Sjosephb if (!NOMEMWAIT() && freemem <= throttlefree + 1) 284ae115bc7Smrj return (PFN_INVALID); 2857c478bd9Sstevel@tonic-gate 286a85a6733Sjosephb #ifdef DEBUG 287a85a6733Sjosephb /* 288a85a6733Sjosephb * This code makes htable_steal() easier to test. By setting 289a85a6733Sjosephb * force_steal we force pagetable allocations to fall 290a85a6733Sjosephb * into the stealing code. Roughly 1 in ever "force_steal" 291a85a6733Sjosephb * page table allocations will fail. 292a85a6733Sjosephb */ 293ae115bc7Smrj if (proc_pageout != NULL && force_steal > 1 && 294a85a6733Sjosephb ++ptable_cnt > force_steal) { 295a85a6733Sjosephb ptable_cnt = 0; 296ae115bc7Smrj return (PFN_INVALID); 297a85a6733Sjosephb } 298a85a6733Sjosephb #endif /* DEBUG */ 299a85a6733Sjosephb 300a77271f8SVikram Hegde pp = page_get_physical(seed); 3017c478bd9Sstevel@tonic-gate if (pp == NULL) 302ae115bc7Smrj return (PFN_INVALID); 3037c478bd9Sstevel@tonic-gate ASSERT(PAGE_SHARED(pp)); 30486c1f4dcSVikram Hegde pfn = pp->p_pagenum; 3057c478bd9Sstevel@tonic-gate if (pfn == PFN_INVALID) 3067c478bd9Sstevel@tonic-gate panic("ptable_alloc(): Invalid PFN!!"); 3071a5e258fSJosef 'Jeff' Sipek atomic_inc_32(&active_ptables); 308a85a6733Sjosephb HATSTAT_INC(hs_ptable_allocs); 309ae115bc7Smrj return (pfn); 3107c478bd9Sstevel@tonic-gate } 3117c478bd9Sstevel@tonic-gate 3127c478bd9Sstevel@tonic-gate /* 3137c478bd9Sstevel@tonic-gate * Free an htable's associated page table page. See the comments 3147c478bd9Sstevel@tonic-gate * for ptable_alloc(). 3157c478bd9Sstevel@tonic-gate */ 3167c478bd9Sstevel@tonic-gate static void 317ae115bc7Smrj ptable_free(pfn_t pfn) 3187c478bd9Sstevel@tonic-gate { 319ae115bc7Smrj page_t *pp = page_numtopp_nolock(pfn); 3207c478bd9Sstevel@tonic-gate 3217c478bd9Sstevel@tonic-gate /* 3227c478bd9Sstevel@tonic-gate * need to destroy the page used for the pagetable 3237c478bd9Sstevel@tonic-gate */ 3247c478bd9Sstevel@tonic-gate ASSERT(pfn != PFN_INVALID); 3257c478bd9Sstevel@tonic-gate HATSTAT_INC(hs_ptable_frees); 3261a5e258fSJosef 'Jeff' Sipek atomic_dec_32(&active_ptables); 3277c478bd9Sstevel@tonic-gate if (pp == NULL) 3287c478bd9Sstevel@tonic-gate panic("ptable_free(): no page for pfn!"); 329a77271f8SVikram Hegde ASSERT(PAGE_SHARED(pp)); 3307c478bd9Sstevel@tonic-gate ASSERT(pfn == pp->p_pagenum); 331843e1988Sjohnlev ASSERT(!IN_XPV_PANIC()); 332a77271f8SVikram Hegde 333a77271f8SVikram Hegde /* 334a77271f8SVikram Hegde * Get an exclusive lock, might have to wait for a kmem reader. 335a77271f8SVikram Hegde */ 336a77271f8SVikram Hegde if (!page_tryupgrade(pp)) { 3376c9930aeSJoe Bonasera u_offset_t off = pp->p_offset; 338a77271f8SVikram Hegde page_unlock(pp); 3396c9930aeSJoe Bonasera pp = page_lookup(&kvp, off, SE_EXCL); 3406c9930aeSJoe Bonasera if (pp == NULL) 3416c9930aeSJoe Bonasera panic("page not found"); 342a77271f8SVikram Hegde } 343843e1988Sjohnlev #ifdef __xpv 344843e1988Sjohnlev if (kpm_vbase && xen_kpm_page(pfn, PT_VALID | PT_WRITABLE) < 0) 345843e1988Sjohnlev panic("failure making kpm r/w pfn=0x%lx", pfn); 346843e1988Sjohnlev #endif 3476c9930aeSJoe Bonasera page_hashout(pp, NULL); 348a77271f8SVikram Hegde page_free(pp, 1); 349a77271f8SVikram Hegde page_unresv(1); 3507c478bd9Sstevel@tonic-gate } 3517c478bd9Sstevel@tonic-gate 3527c478bd9Sstevel@tonic-gate /* 3537c478bd9Sstevel@tonic-gate * Put one htable on the reserve list. 3547c478bd9Sstevel@tonic-gate */ 3557c478bd9Sstevel@tonic-gate static void 3567c478bd9Sstevel@tonic-gate htable_put_reserve(htable_t *ht) 3577c478bd9Sstevel@tonic-gate { 3587c478bd9Sstevel@tonic-gate ht->ht_hat = NULL; /* no longer tied to a hat */ 3597c478bd9Sstevel@tonic-gate ASSERT(ht->ht_pfn == PFN_INVALID); 3607c478bd9Sstevel@tonic-gate HATSTAT_INC(hs_htable_rputs); 3617c478bd9Sstevel@tonic-gate mutex_enter(&htable_reserve_mutex); 3627c478bd9Sstevel@tonic-gate ht->ht_next = htable_reserve_pool; 3637c478bd9Sstevel@tonic-gate htable_reserve_pool = ht; 3647c478bd9Sstevel@tonic-gate ++htable_reserve_cnt; 3657c478bd9Sstevel@tonic-gate mutex_exit(&htable_reserve_mutex); 3667c478bd9Sstevel@tonic-gate } 3677c478bd9Sstevel@tonic-gate 3687c478bd9Sstevel@tonic-gate /* 3697c478bd9Sstevel@tonic-gate * Take one htable from the reserve. 3707c478bd9Sstevel@tonic-gate */ 3717c478bd9Sstevel@tonic-gate static htable_t * 3727c478bd9Sstevel@tonic-gate htable_get_reserve(void) 3737c478bd9Sstevel@tonic-gate { 3747c478bd9Sstevel@tonic-gate htable_t *ht = NULL; 3757c478bd9Sstevel@tonic-gate 3767c478bd9Sstevel@tonic-gate mutex_enter(&htable_reserve_mutex); 3777c478bd9Sstevel@tonic-gate if (htable_reserve_cnt != 0) { 3787c478bd9Sstevel@tonic-gate ht = htable_reserve_pool; 3797c478bd9Sstevel@tonic-gate ASSERT(ht != NULL); 3807c478bd9Sstevel@tonic-gate ASSERT(ht->ht_pfn == PFN_INVALID); 3817c478bd9Sstevel@tonic-gate htable_reserve_pool = ht->ht_next; 3827c478bd9Sstevel@tonic-gate --htable_reserve_cnt; 3837c478bd9Sstevel@tonic-gate HATSTAT_INC(hs_htable_rgets); 3847c478bd9Sstevel@tonic-gate } 3857c478bd9Sstevel@tonic-gate mutex_exit(&htable_reserve_mutex); 3867c478bd9Sstevel@tonic-gate return (ht); 3877c478bd9Sstevel@tonic-gate } 3887c478bd9Sstevel@tonic-gate 3897c478bd9Sstevel@tonic-gate /* 390ae115bc7Smrj * Allocate initial htables and put them on the reserve list 3917c478bd9Sstevel@tonic-gate */ 3927c478bd9Sstevel@tonic-gate void 3937c478bd9Sstevel@tonic-gate htable_initial_reserve(uint_t count) 3947c478bd9Sstevel@tonic-gate { 3957c478bd9Sstevel@tonic-gate htable_t *ht; 3967c478bd9Sstevel@tonic-gate 3977c478bd9Sstevel@tonic-gate count += HTABLE_RESERVE_AMOUNT; 3987c478bd9Sstevel@tonic-gate while (count > 0) { 3997c478bd9Sstevel@tonic-gate ht = kmem_cache_alloc(htable_cache, KM_NOSLEEP); 4007c478bd9Sstevel@tonic-gate ASSERT(ht != NULL); 4017c478bd9Sstevel@tonic-gate 4027c478bd9Sstevel@tonic-gate ASSERT(use_boot_reserve); 403ae115bc7Smrj ht->ht_pfn = PFN_INVALID; 404ae115bc7Smrj htable_put_reserve(ht); 4057c478bd9Sstevel@tonic-gate --count; 4067c478bd9Sstevel@tonic-gate } 4077c478bd9Sstevel@tonic-gate } 4087c478bd9Sstevel@tonic-gate 4097c478bd9Sstevel@tonic-gate /* 4107c478bd9Sstevel@tonic-gate * Readjust the reserves after a thread finishes using them. 4117c478bd9Sstevel@tonic-gate */ 4127c478bd9Sstevel@tonic-gate void 4137c478bd9Sstevel@tonic-gate htable_adjust_reserve() 4147c478bd9Sstevel@tonic-gate { 4157c478bd9Sstevel@tonic-gate htable_t *ht; 4167c478bd9Sstevel@tonic-gate 4177c478bd9Sstevel@tonic-gate /* 4187c478bd9Sstevel@tonic-gate * Free any excess htables in the reserve list 4197c478bd9Sstevel@tonic-gate */ 420aac11643Sjosephb while (htable_reserve_cnt > htable_reserve_amount && 421aac11643Sjosephb !USE_HAT_RESERVES()) { 4227c478bd9Sstevel@tonic-gate ht = htable_get_reserve(); 4237c478bd9Sstevel@tonic-gate if (ht == NULL) 4247c478bd9Sstevel@tonic-gate return; 4257c478bd9Sstevel@tonic-gate ASSERT(ht->ht_pfn == PFN_INVALID); 4267c478bd9Sstevel@tonic-gate kmem_cache_free(htable_cache, ht); 4277c478bd9Sstevel@tonic-gate } 4287c478bd9Sstevel@tonic-gate } 4297c478bd9Sstevel@tonic-gate 430b59c4a48SBoris Protopopov /* 431b59c4a48SBoris Protopopov * Search the active htables for one to steal. Start at a different hash 432b59c4a48SBoris Protopopov * bucket every time to help spread the pain of stealing 433b59c4a48SBoris Protopopov */ 434b59c4a48SBoris Protopopov static void 435b59c4a48SBoris Protopopov htable_steal_active(hat_t *hat, uint_t cnt, uint_t threshold, 436b59c4a48SBoris Protopopov uint_t *stolen, htable_t **list) 437b59c4a48SBoris Protopopov { 438b59c4a48SBoris Protopopov static uint_t h_seed = 0; 439b59c4a48SBoris Protopopov htable_t *higher, *ht; 440b59c4a48SBoris Protopopov uint_t h, e, h_start; 441b59c4a48SBoris Protopopov uintptr_t va; 442b59c4a48SBoris Protopopov x86pte_t pte; 443b59c4a48SBoris Protopopov 444b59c4a48SBoris Protopopov h = h_start = h_seed++ % hat->hat_num_hash; 445b59c4a48SBoris Protopopov do { 446b59c4a48SBoris Protopopov higher = NULL; 447b59c4a48SBoris Protopopov HTABLE_ENTER(h); 448b59c4a48SBoris Protopopov for (ht = hat->hat_ht_hash[h]; ht; ht = ht->ht_next) { 4497c478bd9Sstevel@tonic-gate 4507c478bd9Sstevel@tonic-gate /* 451b59c4a48SBoris Protopopov * Can we rule out reaping? 452b59c4a48SBoris Protopopov */ 453b59c4a48SBoris Protopopov if (ht->ht_busy != 0 || 454b59c4a48SBoris Protopopov (ht->ht_flags & HTABLE_SHARED_PFN) || 455b59c4a48SBoris Protopopov ht->ht_level > 0 || ht->ht_valid_cnt > threshold || 456b59c4a48SBoris Protopopov ht->ht_lock_cnt != 0) 457b59c4a48SBoris Protopopov continue; 458b59c4a48SBoris Protopopov 459b59c4a48SBoris Protopopov /* 460b59c4a48SBoris Protopopov * Increment busy so the htable can't disappear. We 461b59c4a48SBoris Protopopov * drop the htable mutex to avoid deadlocks with 462b59c4a48SBoris Protopopov * hat_pageunload() and the hment mutex while we 463b59c4a48SBoris Protopopov * call hat_pte_unmap() 464b59c4a48SBoris Protopopov */ 465b59c4a48SBoris Protopopov ++ht->ht_busy; 466b59c4a48SBoris Protopopov HTABLE_EXIT(h); 467b59c4a48SBoris Protopopov 468b59c4a48SBoris Protopopov /* 469b59c4a48SBoris Protopopov * Try stealing. 470b59c4a48SBoris Protopopov * - unload and invalidate all PTEs 471b59c4a48SBoris Protopopov */ 472b59c4a48SBoris Protopopov for (e = 0, va = ht->ht_vaddr; 473b59c4a48SBoris Protopopov e < HTABLE_NUM_PTES(ht) && ht->ht_valid_cnt > 0 && 474b59c4a48SBoris Protopopov ht->ht_busy == 1 && ht->ht_lock_cnt == 0; 475b59c4a48SBoris Protopopov ++e, va += MMU_PAGESIZE) { 476b59c4a48SBoris Protopopov pte = x86pte_get(ht, e); 477b59c4a48SBoris Protopopov if (!PTE_ISVALID(pte)) 478b59c4a48SBoris Protopopov continue; 479*a6a74e0eSMatthew Ahrens hat_pte_unmap(ht, e, HAT_UNLOAD, pte, NULL, 480*a6a74e0eSMatthew Ahrens B_TRUE); 481b59c4a48SBoris Protopopov } 482b59c4a48SBoris Protopopov 483b59c4a48SBoris Protopopov /* 484b59c4a48SBoris Protopopov * Reacquire htable lock. If we didn't remove all 485b59c4a48SBoris Protopopov * mappings in the table, or another thread added a new 486b59c4a48SBoris Protopopov * mapping behind us, give up on this table. 487b59c4a48SBoris Protopopov */ 488b59c4a48SBoris Protopopov HTABLE_ENTER(h); 489b59c4a48SBoris Protopopov if (ht->ht_busy != 1 || ht->ht_valid_cnt != 0 || 490b59c4a48SBoris Protopopov ht->ht_lock_cnt != 0) { 491b59c4a48SBoris Protopopov --ht->ht_busy; 492b59c4a48SBoris Protopopov continue; 493b59c4a48SBoris Protopopov } 494b59c4a48SBoris Protopopov 495b59c4a48SBoris Protopopov /* 496b59c4a48SBoris Protopopov * Steal it and unlink the page table. 497b59c4a48SBoris Protopopov */ 498b59c4a48SBoris Protopopov higher = ht->ht_parent; 499b59c4a48SBoris Protopopov unlink_ptp(higher, ht, ht->ht_vaddr); 500b59c4a48SBoris Protopopov 501b59c4a48SBoris Protopopov /* 502b59c4a48SBoris Protopopov * remove from the hash list 503b59c4a48SBoris Protopopov */ 504b59c4a48SBoris Protopopov if (ht->ht_next) 505b59c4a48SBoris Protopopov ht->ht_next->ht_prev = ht->ht_prev; 506b59c4a48SBoris Protopopov 507b59c4a48SBoris Protopopov if (ht->ht_prev) { 508b59c4a48SBoris Protopopov ht->ht_prev->ht_next = ht->ht_next; 509b59c4a48SBoris Protopopov } else { 510b59c4a48SBoris Protopopov ASSERT(hat->hat_ht_hash[h] == ht); 511b59c4a48SBoris Protopopov hat->hat_ht_hash[h] = ht->ht_next; 512b59c4a48SBoris Protopopov } 513b59c4a48SBoris Protopopov 514b59c4a48SBoris Protopopov /* 515b59c4a48SBoris Protopopov * Break to outer loop to release the 516b59c4a48SBoris Protopopov * higher (ht_parent) pagetable. This 517b59c4a48SBoris Protopopov * spreads out the pain caused by 518b59c4a48SBoris Protopopov * pagefaults. 519b59c4a48SBoris Protopopov */ 520b59c4a48SBoris Protopopov ht->ht_next = *list; 521b59c4a48SBoris Protopopov *list = ht; 522b59c4a48SBoris Protopopov ++*stolen; 523b59c4a48SBoris Protopopov break; 524b59c4a48SBoris Protopopov } 525b59c4a48SBoris Protopopov HTABLE_EXIT(h); 526b59c4a48SBoris Protopopov if (higher != NULL) 527b59c4a48SBoris Protopopov htable_release(higher); 528b59c4a48SBoris Protopopov if (++h == hat->hat_num_hash) 529b59c4a48SBoris Protopopov h = 0; 530b59c4a48SBoris Protopopov } while (*stolen < cnt && h != h_start); 531b59c4a48SBoris Protopopov } 532b59c4a48SBoris Protopopov 533b59c4a48SBoris Protopopov /* 534b59c4a48SBoris Protopopov * Move hat to the end of the kas list 535b59c4a48SBoris Protopopov */ 536b59c4a48SBoris Protopopov static void 537b59c4a48SBoris Protopopov move_victim(hat_t *hat) 538b59c4a48SBoris Protopopov { 539b59c4a48SBoris Protopopov ASSERT(MUTEX_HELD(&hat_list_lock)); 540b59c4a48SBoris Protopopov 541b59c4a48SBoris Protopopov /* unlink victim hat */ 542b59c4a48SBoris Protopopov if (hat->hat_prev) 543b59c4a48SBoris Protopopov hat->hat_prev->hat_next = hat->hat_next; 544b59c4a48SBoris Protopopov else 545b59c4a48SBoris Protopopov kas.a_hat->hat_next = hat->hat_next; 546b59c4a48SBoris Protopopov 547b59c4a48SBoris Protopopov if (hat->hat_next) 548b59c4a48SBoris Protopopov hat->hat_next->hat_prev = hat->hat_prev; 549b59c4a48SBoris Protopopov else 550b59c4a48SBoris Protopopov kas.a_hat->hat_prev = hat->hat_prev; 551b59c4a48SBoris Protopopov /* relink at end of hat list */ 552b59c4a48SBoris Protopopov hat->hat_next = NULL; 553b59c4a48SBoris Protopopov hat->hat_prev = kas.a_hat->hat_prev; 554b59c4a48SBoris Protopopov if (hat->hat_prev) 555b59c4a48SBoris Protopopov hat->hat_prev->hat_next = hat; 556b59c4a48SBoris Protopopov else 557b59c4a48SBoris Protopopov kas.a_hat->hat_next = hat; 558b59c4a48SBoris Protopopov 559b59c4a48SBoris Protopopov kas.a_hat->hat_prev = hat; 560b59c4a48SBoris Protopopov } 561b59c4a48SBoris Protopopov 562b59c4a48SBoris Protopopov /* 563b59c4a48SBoris Protopopov * This routine steals htables from user processes. Called by htable_reap 564b59c4a48SBoris Protopopov * (reap=TRUE) or htable_alloc (reap=FALSE). 5657c478bd9Sstevel@tonic-gate */ 5667c478bd9Sstevel@tonic-gate static htable_t * 567b59c4a48SBoris Protopopov htable_steal(uint_t cnt, boolean_t reap) 5687c478bd9Sstevel@tonic-gate { 5697c478bd9Sstevel@tonic-gate hat_t *hat = kas.a_hat; /* list starts with khat */ 5707c478bd9Sstevel@tonic-gate htable_t *list = NULL; 5717c478bd9Sstevel@tonic-gate htable_t *ht; 5727c478bd9Sstevel@tonic-gate uint_t stolen = 0; 5737c478bd9Sstevel@tonic-gate uint_t pass; 574a85a6733Sjosephb uint_t threshold; 5757c478bd9Sstevel@tonic-gate 5767c478bd9Sstevel@tonic-gate /* 5777c478bd9Sstevel@tonic-gate * Limit htable_steal_passes to something reasonable 5787c478bd9Sstevel@tonic-gate */ 5797c478bd9Sstevel@tonic-gate if (htable_steal_passes == 0) 5807c478bd9Sstevel@tonic-gate htable_steal_passes = 1; 5817c478bd9Sstevel@tonic-gate if (htable_steal_passes > mmu.ptes_per_table) 5827c478bd9Sstevel@tonic-gate htable_steal_passes = mmu.ptes_per_table; 5837c478bd9Sstevel@tonic-gate 5847c478bd9Sstevel@tonic-gate /* 585a85a6733Sjosephb * Loop through all user hats. The 1st pass takes cached htables that 5867c478bd9Sstevel@tonic-gate * aren't in use. The later passes steal by removing mappings, too. 5877c478bd9Sstevel@tonic-gate */ 5881a5e258fSJosef 'Jeff' Sipek atomic_inc_32(&htable_dont_cache); 589a85a6733Sjosephb for (pass = 0; pass <= htable_steal_passes && stolen < cnt; ++pass) { 590a85a6733Sjosephb threshold = pass * mmu.ptes_per_table / htable_steal_passes; 5917c478bd9Sstevel@tonic-gate 5927c478bd9Sstevel@tonic-gate mutex_enter(&hat_list_lock); 593a85a6733Sjosephb 594b59c4a48SBoris Protopopov /* skip the first hat (kernel) */ 595b59c4a48SBoris Protopopov hat = kas.a_hat->hat_next; 596b59c4a48SBoris Protopopov for (;;) { 597a85a6733Sjosephb /* 598a85a6733Sjosephb * Skip any hat that is already being stolen from. 599a85a6733Sjosephb * 600a85a6733Sjosephb * We skip SHARED hats, as these are dummy 601a85a6733Sjosephb * hats that host ISM shared page tables. 602a85a6733Sjosephb * 603a85a6733Sjosephb * We also skip if HAT_FREEING because hat_pte_unmap() 604a85a6733Sjosephb * won't zero out the PTE's. That would lead to hitting 605a85a6733Sjosephb * stale PTEs either here or under hat_unload() when we 606a85a6733Sjosephb * steal and unload the same page table in competing 607a85a6733Sjosephb * threads. 608a85a6733Sjosephb */ 609a85a6733Sjosephb while (hat != NULL && 610a85a6733Sjosephb (hat->hat_flags & 611a85a6733Sjosephb (HAT_VICTIM | HAT_SHARED | HAT_FREEING)) != 0) 612a85a6733Sjosephb hat = hat->hat_next; 613a85a6733Sjosephb 614b59c4a48SBoris Protopopov if (hat == NULL) 6157c478bd9Sstevel@tonic-gate break; 616a85a6733Sjosephb 617a85a6733Sjosephb /* 618b59c4a48SBoris Protopopov * Mark the HAT as a stealing victim so that it is 619b59c4a48SBoris Protopopov * not freed from under us, e.g. in as_free() 620a85a6733Sjosephb */ 6217c478bd9Sstevel@tonic-gate hat->hat_flags |= HAT_VICTIM; 6227c478bd9Sstevel@tonic-gate mutex_exit(&hat_list_lock); 6237c478bd9Sstevel@tonic-gate 6247c478bd9Sstevel@tonic-gate /* 6257c478bd9Sstevel@tonic-gate * Take any htables from the hat's cached "free" list. 6267c478bd9Sstevel@tonic-gate */ 6277c478bd9Sstevel@tonic-gate hat_enter(hat); 6287c478bd9Sstevel@tonic-gate while ((ht = hat->hat_ht_cached) != NULL && 6297c478bd9Sstevel@tonic-gate stolen < cnt) { 6307c478bd9Sstevel@tonic-gate hat->hat_ht_cached = ht->ht_next; 6317c478bd9Sstevel@tonic-gate ht->ht_next = list; 6327c478bd9Sstevel@tonic-gate list = ht; 6337c478bd9Sstevel@tonic-gate ++stolen; 6347c478bd9Sstevel@tonic-gate } 6357c478bd9Sstevel@tonic-gate hat_exit(hat); 6367c478bd9Sstevel@tonic-gate 6377c478bd9Sstevel@tonic-gate /* 638b59c4a48SBoris Protopopov * Don't steal active htables on first pass. 6397c478bd9Sstevel@tonic-gate */ 640b59c4a48SBoris Protopopov if (pass != 0 && (stolen < cnt)) 641b59c4a48SBoris Protopopov htable_steal_active(hat, cnt, threshold, 642b59c4a48SBoris Protopopov &stolen, &list); 643b59c4a48SBoris Protopopov 644b59c4a48SBoris Protopopov /* 645b59c4a48SBoris Protopopov * do synchronous teardown for the reap case so that 646b59c4a48SBoris Protopopov * we can forget hat; at this time, hat is 647b59c4a48SBoris Protopopov * guaranteed to be around because HAT_VICTIM is set 648b59c4a48SBoris Protopopov * (see htable_free() for similar code) 649b59c4a48SBoris Protopopov */ 650b59c4a48SBoris Protopopov for (ht = list; (ht) && (reap); ht = ht->ht_next) { 651b59c4a48SBoris Protopopov if (ht->ht_hat == NULL) 6527c478bd9Sstevel@tonic-gate continue; 653b59c4a48SBoris Protopopov ASSERT(ht->ht_hat == hat); 654b59c4a48SBoris Protopopov #if defined(__xpv) && defined(__amd64) 655b59c4a48SBoris Protopopov if (!(ht->ht_flags & HTABLE_VLP) && 656b59c4a48SBoris Protopopov ht->ht_level == mmu.max_level) { 657b59c4a48SBoris Protopopov ptable_free(hat->hat_user_ptable); 658b59c4a48SBoris Protopopov hat->hat_user_ptable = PFN_INVALID; 659b59c4a48SBoris Protopopov } 660b59c4a48SBoris Protopopov #endif 661b59c4a48SBoris Protopopov /* 662b59c4a48SBoris Protopopov * forget the hat 663b59c4a48SBoris Protopopov */ 664b59c4a48SBoris Protopopov ht->ht_hat = NULL; 665b59c4a48SBoris Protopopov } 666b59c4a48SBoris Protopopov 667b59c4a48SBoris Protopopov mutex_enter(&hat_list_lock); 6687c478bd9Sstevel@tonic-gate 6697c478bd9Sstevel@tonic-gate /* 670b59c4a48SBoris Protopopov * Are we finished? 6717c478bd9Sstevel@tonic-gate */ 672b59c4a48SBoris Protopopov if (stolen == cnt) { 6737c478bd9Sstevel@tonic-gate /* 674b59c4a48SBoris Protopopov * Try to spread the pain of stealing, 675b59c4a48SBoris Protopopov * move victim HAT to the end of the HAT list. 6767c478bd9Sstevel@tonic-gate */ 677b59c4a48SBoris Protopopov if (pass >= 1 && cnt == 1 && 678b59c4a48SBoris Protopopov kas.a_hat->hat_prev != hat) 679b59c4a48SBoris Protopopov move_victim(hat); 6807c478bd9Sstevel@tonic-gate /* 681b59c4a48SBoris Protopopov * We are finished 6827c478bd9Sstevel@tonic-gate */ 6837c478bd9Sstevel@tonic-gate } 6847c478bd9Sstevel@tonic-gate 6857c478bd9Sstevel@tonic-gate /* 686b59c4a48SBoris Protopopov * Clear the victim flag, hat can go away now (once 687b59c4a48SBoris Protopopov * the lock is dropped) 6887c478bd9Sstevel@tonic-gate */ 689b59c4a48SBoris Protopopov if (hat->hat_flags & HAT_VICTIM) { 690b59c4a48SBoris Protopopov ASSERT(hat != kas.a_hat); 691b59c4a48SBoris Protopopov hat->hat_flags &= ~HAT_VICTIM; 692b59c4a48SBoris Protopopov cv_broadcast(&hat_list_cv); 6937c478bd9Sstevel@tonic-gate } 6947c478bd9Sstevel@tonic-gate 695b59c4a48SBoris Protopopov /* move on to the next hat */ 696b59c4a48SBoris Protopopov hat = hat->hat_next; 6977c478bd9Sstevel@tonic-gate } 6987c478bd9Sstevel@tonic-gate 699b59c4a48SBoris Protopopov mutex_exit(&hat_list_lock); 700b59c4a48SBoris Protopopov 7017c478bd9Sstevel@tonic-gate } 702b59c4a48SBoris Protopopov ASSERT(!MUTEX_HELD(&hat_list_lock)); 703b59c4a48SBoris Protopopov 7041a5e258fSJosef 'Jeff' Sipek atomic_dec_32(&htable_dont_cache); 7057c478bd9Sstevel@tonic-gate return (list); 7067c478bd9Sstevel@tonic-gate } 7077c478bd9Sstevel@tonic-gate 7087c478bd9Sstevel@tonic-gate /* 7097c478bd9Sstevel@tonic-gate * This is invoked from kmem when the system is low on memory. We try 7107c478bd9Sstevel@tonic-gate * to free hments, htables, and ptables to improve the memory situation. 7117c478bd9Sstevel@tonic-gate */ 7127c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 7137c478bd9Sstevel@tonic-gate static void 7147c478bd9Sstevel@tonic-gate htable_reap(void *handle) 7157c478bd9Sstevel@tonic-gate { 7167c478bd9Sstevel@tonic-gate uint_t reap_cnt; 7177c478bd9Sstevel@tonic-gate htable_t *list; 7187c478bd9Sstevel@tonic-gate htable_t *ht; 7197c478bd9Sstevel@tonic-gate 7207c478bd9Sstevel@tonic-gate HATSTAT_INC(hs_reap_attempts); 7217c478bd9Sstevel@tonic-gate if (!can_steal_post_boot) 7227c478bd9Sstevel@tonic-gate return; 7237c478bd9Sstevel@tonic-gate 7247c478bd9Sstevel@tonic-gate /* 7257c478bd9Sstevel@tonic-gate * Try to reap 5% of the page tables bounded by a maximum of 7267c478bd9Sstevel@tonic-gate * 5% of physmem and a minimum of 10. 7277c478bd9Sstevel@tonic-gate */ 7285f661bbcSJakub Jermar reap_cnt = MAX(MIN(physmem / 20, active_ptables / 20), 10); 7297c478bd9Sstevel@tonic-gate 7307c478bd9Sstevel@tonic-gate /* 731b59c4a48SBoris Protopopov * Note: htable_dont_cache should be set at the time of 732b59c4a48SBoris Protopopov * invoking htable_free() 733b59c4a48SBoris Protopopov */ 734b59c4a48SBoris Protopopov atomic_inc_32(&htable_dont_cache); 735b59c4a48SBoris Protopopov /* 7367c478bd9Sstevel@tonic-gate * Let htable_steal() do the work, we just call htable_free() 7377c478bd9Sstevel@tonic-gate */ 738843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 739b59c4a48SBoris Protopopov list = htable_steal(reap_cnt, B_TRUE); 740843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 7417c478bd9Sstevel@tonic-gate while ((ht = list) != NULL) { 7427c478bd9Sstevel@tonic-gate list = ht->ht_next; 7437c478bd9Sstevel@tonic-gate HATSTAT_INC(hs_reaped); 7447c478bd9Sstevel@tonic-gate htable_free(ht); 7457c478bd9Sstevel@tonic-gate } 746b59c4a48SBoris Protopopov atomic_dec_32(&htable_dont_cache); 7477c478bd9Sstevel@tonic-gate 7487c478bd9Sstevel@tonic-gate /* 7497c478bd9Sstevel@tonic-gate * Free up excess reserves 7507c478bd9Sstevel@tonic-gate */ 7517c478bd9Sstevel@tonic-gate htable_adjust_reserve(); 7527c478bd9Sstevel@tonic-gate hment_adjust_reserve(); 7537c478bd9Sstevel@tonic-gate } 7547c478bd9Sstevel@tonic-gate 7557c478bd9Sstevel@tonic-gate /* 756ae115bc7Smrj * Allocate an htable, stealing one or using the reserve if necessary 7577c478bd9Sstevel@tonic-gate */ 7587c478bd9Sstevel@tonic-gate static htable_t * 7597c478bd9Sstevel@tonic-gate htable_alloc( 7607c478bd9Sstevel@tonic-gate hat_t *hat, 7617c478bd9Sstevel@tonic-gate uintptr_t vaddr, 7627c478bd9Sstevel@tonic-gate level_t level, 7637c478bd9Sstevel@tonic-gate htable_t *shared) 7647c478bd9Sstevel@tonic-gate { 7657c478bd9Sstevel@tonic-gate htable_t *ht = NULL; 7667c478bd9Sstevel@tonic-gate uint_t is_vlp; 7677c478bd9Sstevel@tonic-gate uint_t is_bare = 0; 7687c478bd9Sstevel@tonic-gate uint_t need_to_zero = 1; 7697c478bd9Sstevel@tonic-gate int kmflags = (can_steal_post_boot ? KM_NOSLEEP : KM_SLEEP); 7707c478bd9Sstevel@tonic-gate 7717c478bd9Sstevel@tonic-gate if (level < 0 || level > TOP_LEVEL(hat)) 7727c478bd9Sstevel@tonic-gate panic("htable_alloc(): level %d out of range\n", level); 7737c478bd9Sstevel@tonic-gate 7747c478bd9Sstevel@tonic-gate is_vlp = (hat->hat_flags & HAT_VLP) && level == VLP_LEVEL; 7757c478bd9Sstevel@tonic-gate if (is_vlp || shared != NULL) 7767c478bd9Sstevel@tonic-gate is_bare = 1; 7777c478bd9Sstevel@tonic-gate 7787c478bd9Sstevel@tonic-gate /* 7797c478bd9Sstevel@tonic-gate * First reuse a cached htable from the hat_ht_cached field, this 780ae115bc7Smrj * avoids unnecessary trips through kmem/page allocators. 7817c478bd9Sstevel@tonic-gate */ 7827c478bd9Sstevel@tonic-gate if (hat->hat_ht_cached != NULL && !is_bare) { 7837c478bd9Sstevel@tonic-gate hat_enter(hat); 7847c478bd9Sstevel@tonic-gate ht = hat->hat_ht_cached; 7857c478bd9Sstevel@tonic-gate if (ht != NULL) { 7867c478bd9Sstevel@tonic-gate hat->hat_ht_cached = ht->ht_next; 7877c478bd9Sstevel@tonic-gate need_to_zero = 0; 7887c478bd9Sstevel@tonic-gate /* XX64 ASSERT() they're all zero somehow */ 7897c478bd9Sstevel@tonic-gate ASSERT(ht->ht_pfn != PFN_INVALID); 7907c478bd9Sstevel@tonic-gate } 7917c478bd9Sstevel@tonic-gate hat_exit(hat); 7927c478bd9Sstevel@tonic-gate } 7937c478bd9Sstevel@tonic-gate 7947c478bd9Sstevel@tonic-gate if (ht == NULL) { 7957c478bd9Sstevel@tonic-gate /* 79697704650Sjosephb * Allocate an htable, possibly refilling the reserves. 7977c478bd9Sstevel@tonic-gate */ 79897704650Sjosephb if (USE_HAT_RESERVES()) { 7997c478bd9Sstevel@tonic-gate ht = htable_get_reserve(); 8007c478bd9Sstevel@tonic-gate } else { 8017c478bd9Sstevel@tonic-gate /* 8027c478bd9Sstevel@tonic-gate * Donate successful htable allocations to the reserve. 8037c478bd9Sstevel@tonic-gate */ 8047c478bd9Sstevel@tonic-gate for (;;) { 8057c478bd9Sstevel@tonic-gate ht = kmem_cache_alloc(htable_cache, kmflags); 8067c478bd9Sstevel@tonic-gate if (ht == NULL) 8077c478bd9Sstevel@tonic-gate break; 8087c478bd9Sstevel@tonic-gate ht->ht_pfn = PFN_INVALID; 80997704650Sjosephb if (USE_HAT_RESERVES() || 8107c478bd9Sstevel@tonic-gate htable_reserve_cnt >= htable_reserve_amount) 8117c478bd9Sstevel@tonic-gate break; 8127c478bd9Sstevel@tonic-gate htable_put_reserve(ht); 8137c478bd9Sstevel@tonic-gate } 8147c478bd9Sstevel@tonic-gate } 8157c478bd9Sstevel@tonic-gate 8167c478bd9Sstevel@tonic-gate /* 8177c478bd9Sstevel@tonic-gate * allocate a page for the hardware page table if needed 8187c478bd9Sstevel@tonic-gate */ 8197c478bd9Sstevel@tonic-gate if (ht != NULL && !is_bare) { 820a85a6733Sjosephb ht->ht_hat = hat; 821a77271f8SVikram Hegde ht->ht_pfn = ptable_alloc((uintptr_t)ht); 8227c478bd9Sstevel@tonic-gate if (ht->ht_pfn == PFN_INVALID) { 82397704650Sjosephb if (USE_HAT_RESERVES()) 82497704650Sjosephb htable_put_reserve(ht); 82597704650Sjosephb else 8267c478bd9Sstevel@tonic-gate kmem_cache_free(htable_cache, ht); 8277c478bd9Sstevel@tonic-gate ht = NULL; 8287c478bd9Sstevel@tonic-gate } 8297c478bd9Sstevel@tonic-gate } 8307c478bd9Sstevel@tonic-gate } 8317c478bd9Sstevel@tonic-gate 8327c478bd9Sstevel@tonic-gate /* 833a85a6733Sjosephb * If allocations failed, kick off a kmem_reap() and resort to 834a85a6733Sjosephb * htable steal(). We may spin here if the system is very low on 835a85a6733Sjosephb * memory. If the kernel itself has consumed all memory and kmem_reap() 836a85a6733Sjosephb * can't free up anything, then we'll really get stuck here. 837a85a6733Sjosephb * That should only happen in a system where the administrator has 838a85a6733Sjosephb * misconfigured VM parameters via /etc/system. 8397c478bd9Sstevel@tonic-gate */ 840a85a6733Sjosephb while (ht == NULL && can_steal_post_boot) { 841a85a6733Sjosephb kmem_reap(); 842b59c4a48SBoris Protopopov ht = htable_steal(1, B_FALSE); 8437c478bd9Sstevel@tonic-gate HATSTAT_INC(hs_steals); 8447c478bd9Sstevel@tonic-gate 8457c478bd9Sstevel@tonic-gate /* 846a85a6733Sjosephb * If we stole for a bare htable, release the pagetable page. 8477c478bd9Sstevel@tonic-gate */ 848ae115bc7Smrj if (ht != NULL) { 849ae115bc7Smrj if (is_bare) { 850ae115bc7Smrj ptable_free(ht->ht_pfn); 851ae115bc7Smrj ht->ht_pfn = PFN_INVALID; 852843e1988Sjohnlev #if defined(__xpv) && defined(__amd64) 853843e1988Sjohnlev /* 854843e1988Sjohnlev * make stolen page table writable again in kpm 855843e1988Sjohnlev */ 856843e1988Sjohnlev } else if (kpm_vbase && xen_kpm_page(ht->ht_pfn, 857843e1988Sjohnlev PT_VALID | PT_WRITABLE) < 0) { 858843e1988Sjohnlev panic("failure making kpm r/w pfn=0x%lx", 859843e1988Sjohnlev ht->ht_pfn); 860843e1988Sjohnlev #endif 861ae115bc7Smrj } 862ae115bc7Smrj } 8637c478bd9Sstevel@tonic-gate } 8647c478bd9Sstevel@tonic-gate 8657c478bd9Sstevel@tonic-gate /* 866a85a6733Sjosephb * All attempts to allocate or steal failed. This should only happen 867a85a6733Sjosephb * if we run out of memory during boot, due perhaps to a huge 868a85a6733Sjosephb * boot_archive. At this point there's no way to continue. 8697c478bd9Sstevel@tonic-gate */ 8707c478bd9Sstevel@tonic-gate if (ht == NULL) 8717c478bd9Sstevel@tonic-gate panic("htable_alloc(): couldn't steal\n"); 8727c478bd9Sstevel@tonic-gate 873843e1988Sjohnlev #if defined(__amd64) && defined(__xpv) 874843e1988Sjohnlev /* 875843e1988Sjohnlev * Under the 64-bit hypervisor, we have 2 top level page tables. 876843e1988Sjohnlev * If this allocation fails, we'll resort to stealing. 877843e1988Sjohnlev * We use the stolen page indirectly, by freeing the 878843e1988Sjohnlev * stolen htable first. 879843e1988Sjohnlev */ 880843e1988Sjohnlev if (level == mmu.max_level) { 881843e1988Sjohnlev for (;;) { 882843e1988Sjohnlev htable_t *stolen; 883843e1988Sjohnlev 884a77271f8SVikram Hegde hat->hat_user_ptable = ptable_alloc((uintptr_t)ht + 1); 885843e1988Sjohnlev if (hat->hat_user_ptable != PFN_INVALID) 886843e1988Sjohnlev break; 887b59c4a48SBoris Protopopov stolen = htable_steal(1, B_FALSE); 888843e1988Sjohnlev if (stolen == NULL) 889843e1988Sjohnlev panic("2nd steal ptable failed\n"); 890843e1988Sjohnlev htable_free(stolen); 891843e1988Sjohnlev } 892843e1988Sjohnlev block_zero_no_xmm(kpm_vbase + pfn_to_pa(hat->hat_user_ptable), 893843e1988Sjohnlev MMU_PAGESIZE); 894843e1988Sjohnlev } 895843e1988Sjohnlev #endif 896843e1988Sjohnlev 8977c478bd9Sstevel@tonic-gate /* 8987c478bd9Sstevel@tonic-gate * Shared page tables have all entries locked and entries may not 8997c478bd9Sstevel@tonic-gate * be added or deleted. 9007c478bd9Sstevel@tonic-gate */ 9017c478bd9Sstevel@tonic-gate ht->ht_flags = 0; 9027c478bd9Sstevel@tonic-gate if (shared != NULL) { 9037c478bd9Sstevel@tonic-gate ASSERT(shared->ht_valid_cnt > 0); 9047c478bd9Sstevel@tonic-gate ht->ht_flags |= HTABLE_SHARED_PFN; 9057c478bd9Sstevel@tonic-gate ht->ht_pfn = shared->ht_pfn; 9067c478bd9Sstevel@tonic-gate ht->ht_lock_cnt = 0; 9077c478bd9Sstevel@tonic-gate ht->ht_valid_cnt = 0; /* updated in hat_share() */ 9087c478bd9Sstevel@tonic-gate ht->ht_shares = shared; 9097c478bd9Sstevel@tonic-gate need_to_zero = 0; 9107c478bd9Sstevel@tonic-gate } else { 9117c478bd9Sstevel@tonic-gate ht->ht_shares = NULL; 9127c478bd9Sstevel@tonic-gate ht->ht_lock_cnt = 0; 9137c478bd9Sstevel@tonic-gate ht->ht_valid_cnt = 0; 9147c478bd9Sstevel@tonic-gate } 9157c478bd9Sstevel@tonic-gate 9167c478bd9Sstevel@tonic-gate /* 9177c478bd9Sstevel@tonic-gate * setup flags, etc. for VLP htables 9187c478bd9Sstevel@tonic-gate */ 9197c478bd9Sstevel@tonic-gate if (is_vlp) { 9207c478bd9Sstevel@tonic-gate ht->ht_flags |= HTABLE_VLP; 9217c478bd9Sstevel@tonic-gate ASSERT(ht->ht_pfn == PFN_INVALID); 9227c478bd9Sstevel@tonic-gate need_to_zero = 0; 9237c478bd9Sstevel@tonic-gate } 9247c478bd9Sstevel@tonic-gate 9257c478bd9Sstevel@tonic-gate /* 9267c478bd9Sstevel@tonic-gate * fill in the htable 9277c478bd9Sstevel@tonic-gate */ 9287c478bd9Sstevel@tonic-gate ht->ht_hat = hat; 9297c478bd9Sstevel@tonic-gate ht->ht_parent = NULL; 9307c478bd9Sstevel@tonic-gate ht->ht_vaddr = vaddr; 9317c478bd9Sstevel@tonic-gate ht->ht_level = level; 9327c478bd9Sstevel@tonic-gate ht->ht_busy = 1; 9337c478bd9Sstevel@tonic-gate ht->ht_next = NULL; 9347c478bd9Sstevel@tonic-gate ht->ht_prev = NULL; 9357c478bd9Sstevel@tonic-gate 9367c478bd9Sstevel@tonic-gate /* 9377c478bd9Sstevel@tonic-gate * Zero out any freshly allocated page table 9387c478bd9Sstevel@tonic-gate */ 9397c478bd9Sstevel@tonic-gate if (need_to_zero) 9407c478bd9Sstevel@tonic-gate x86pte_zero(ht, 0, mmu.ptes_per_table); 941ae115bc7Smrj 942843e1988Sjohnlev #if defined(__amd64) && defined(__xpv) 943843e1988Sjohnlev if (!is_bare && kpm_vbase) { 944843e1988Sjohnlev (void) xen_kpm_page(ht->ht_pfn, PT_VALID); 945843e1988Sjohnlev if (level == mmu.max_level) 946843e1988Sjohnlev (void) xen_kpm_page(hat->hat_user_ptable, PT_VALID); 947843e1988Sjohnlev } 948843e1988Sjohnlev #endif 949843e1988Sjohnlev 9507c478bd9Sstevel@tonic-gate return (ht); 9517c478bd9Sstevel@tonic-gate } 9527c478bd9Sstevel@tonic-gate 9537c478bd9Sstevel@tonic-gate /* 9547c478bd9Sstevel@tonic-gate * Free up an htable, either to a hat's cached list, the reserves or 9557c478bd9Sstevel@tonic-gate * back to kmem. 9567c478bd9Sstevel@tonic-gate */ 9577c478bd9Sstevel@tonic-gate static void 9587c478bd9Sstevel@tonic-gate htable_free(htable_t *ht) 9597c478bd9Sstevel@tonic-gate { 9607c478bd9Sstevel@tonic-gate hat_t *hat = ht->ht_hat; 9617c478bd9Sstevel@tonic-gate 9627c478bd9Sstevel@tonic-gate /* 9637c478bd9Sstevel@tonic-gate * If the process isn't exiting, cache the free htable in the hat 964843e1988Sjohnlev * structure. We always do this for the boot time reserve. We don't 9657c478bd9Sstevel@tonic-gate * do this if the hat is exiting or we are stealing/reaping htables. 9667c478bd9Sstevel@tonic-gate */ 9677c478bd9Sstevel@tonic-gate if (hat != NULL && 9687c478bd9Sstevel@tonic-gate !(ht->ht_flags & HTABLE_SHARED_PFN) && 9697c478bd9Sstevel@tonic-gate (use_boot_reserve || 9707c478bd9Sstevel@tonic-gate (!(hat->hat_flags & HAT_FREEING) && !htable_dont_cache))) { 9717c478bd9Sstevel@tonic-gate ASSERT((ht->ht_flags & HTABLE_VLP) == 0); 9727c478bd9Sstevel@tonic-gate ASSERT(ht->ht_pfn != PFN_INVALID); 9737c478bd9Sstevel@tonic-gate hat_enter(hat); 9747c478bd9Sstevel@tonic-gate ht->ht_next = hat->hat_ht_cached; 9757c478bd9Sstevel@tonic-gate hat->hat_ht_cached = ht; 9767c478bd9Sstevel@tonic-gate hat_exit(hat); 9777c478bd9Sstevel@tonic-gate return; 9787c478bd9Sstevel@tonic-gate } 9797c478bd9Sstevel@tonic-gate 9807c478bd9Sstevel@tonic-gate /* 9817c478bd9Sstevel@tonic-gate * If we have a hardware page table, free it. 982ae115bc7Smrj * We don't free page tables that are accessed by sharing. 9837c478bd9Sstevel@tonic-gate */ 9847c478bd9Sstevel@tonic-gate if (ht->ht_flags & HTABLE_SHARED_PFN) { 9857c478bd9Sstevel@tonic-gate ASSERT(ht->ht_pfn != PFN_INVALID); 9867c478bd9Sstevel@tonic-gate } else if (!(ht->ht_flags & HTABLE_VLP)) { 987ae115bc7Smrj ptable_free(ht->ht_pfn); 988843e1988Sjohnlev #if defined(__amd64) && defined(__xpv) 989b59c4a48SBoris Protopopov if (ht->ht_level == mmu.max_level && hat != NULL) { 990843e1988Sjohnlev ptable_free(hat->hat_user_ptable); 991843e1988Sjohnlev hat->hat_user_ptable = PFN_INVALID; 992843e1988Sjohnlev } 993843e1988Sjohnlev #endif 9947c478bd9Sstevel@tonic-gate } 995ae115bc7Smrj ht->ht_pfn = PFN_INVALID; 9967c478bd9Sstevel@tonic-gate 9977c478bd9Sstevel@tonic-gate /* 998843e1988Sjohnlev * Free it or put into reserves. 9997c478bd9Sstevel@tonic-gate */ 1000aac11643Sjosephb if (USE_HAT_RESERVES() || htable_reserve_cnt < htable_reserve_amount) { 10017c478bd9Sstevel@tonic-gate htable_put_reserve(ht); 1002aac11643Sjosephb } else { 10037c478bd9Sstevel@tonic-gate kmem_cache_free(htable_cache, ht); 1004aac11643Sjosephb htable_adjust_reserve(); 1005aac11643Sjosephb } 10067c478bd9Sstevel@tonic-gate } 10077c478bd9Sstevel@tonic-gate 10087c478bd9Sstevel@tonic-gate 10097c478bd9Sstevel@tonic-gate /* 10107c478bd9Sstevel@tonic-gate * This is called when a hat is being destroyed or swapped out. We reap all 10117c478bd9Sstevel@tonic-gate * the remaining htables in the hat cache. If destroying all left over 10127c478bd9Sstevel@tonic-gate * htables are also destroyed. 10137c478bd9Sstevel@tonic-gate * 10147c478bd9Sstevel@tonic-gate * We also don't need to invalidate any of the PTPs nor do any demapping. 10157c478bd9Sstevel@tonic-gate */ 10167c478bd9Sstevel@tonic-gate void 10177c478bd9Sstevel@tonic-gate htable_purge_hat(hat_t *hat) 10187c478bd9Sstevel@tonic-gate { 10197c478bd9Sstevel@tonic-gate htable_t *ht; 10207c478bd9Sstevel@tonic-gate int h; 10217c478bd9Sstevel@tonic-gate 10227c478bd9Sstevel@tonic-gate /* 10237c478bd9Sstevel@tonic-gate * Purge the htable cache if just reaping. 10247c478bd9Sstevel@tonic-gate */ 10257c478bd9Sstevel@tonic-gate if (!(hat->hat_flags & HAT_FREEING)) { 10261a5e258fSJosef 'Jeff' Sipek atomic_inc_32(&htable_dont_cache); 10277c478bd9Sstevel@tonic-gate for (;;) { 10287c478bd9Sstevel@tonic-gate hat_enter(hat); 10297c478bd9Sstevel@tonic-gate ht = hat->hat_ht_cached; 10307c478bd9Sstevel@tonic-gate if (ht == NULL) { 10317c478bd9Sstevel@tonic-gate hat_exit(hat); 10327c478bd9Sstevel@tonic-gate break; 10337c478bd9Sstevel@tonic-gate } 10347c478bd9Sstevel@tonic-gate hat->hat_ht_cached = ht->ht_next; 10357c478bd9Sstevel@tonic-gate hat_exit(hat); 10367c478bd9Sstevel@tonic-gate htable_free(ht); 10377c478bd9Sstevel@tonic-gate } 10381a5e258fSJosef 'Jeff' Sipek atomic_dec_32(&htable_dont_cache); 10397c478bd9Sstevel@tonic-gate return; 10407c478bd9Sstevel@tonic-gate } 10417c478bd9Sstevel@tonic-gate 10427c478bd9Sstevel@tonic-gate /* 10437c478bd9Sstevel@tonic-gate * if freeing, no locking is needed 10447c478bd9Sstevel@tonic-gate */ 10457c478bd9Sstevel@tonic-gate while ((ht = hat->hat_ht_cached) != NULL) { 10467c478bd9Sstevel@tonic-gate hat->hat_ht_cached = ht->ht_next; 10477c478bd9Sstevel@tonic-gate htable_free(ht); 10487c478bd9Sstevel@tonic-gate } 10497c478bd9Sstevel@tonic-gate 10507c478bd9Sstevel@tonic-gate /* 10517c478bd9Sstevel@tonic-gate * walk thru the htable hash table and free all the htables in it. 10527c478bd9Sstevel@tonic-gate */ 10537c478bd9Sstevel@tonic-gate for (h = 0; h < hat->hat_num_hash; ++h) { 10547c478bd9Sstevel@tonic-gate while ((ht = hat->hat_ht_hash[h]) != NULL) { 10557c478bd9Sstevel@tonic-gate if (ht->ht_next) 10567c478bd9Sstevel@tonic-gate ht->ht_next->ht_prev = ht->ht_prev; 10577c478bd9Sstevel@tonic-gate 10587c478bd9Sstevel@tonic-gate if (ht->ht_prev) { 10597c478bd9Sstevel@tonic-gate ht->ht_prev->ht_next = ht->ht_next; 10607c478bd9Sstevel@tonic-gate } else { 10617c478bd9Sstevel@tonic-gate ASSERT(hat->hat_ht_hash[h] == ht); 10627c478bd9Sstevel@tonic-gate hat->hat_ht_hash[h] = ht->ht_next; 10637c478bd9Sstevel@tonic-gate } 10647c478bd9Sstevel@tonic-gate htable_free(ht); 10657c478bd9Sstevel@tonic-gate } 10667c478bd9Sstevel@tonic-gate } 10677c478bd9Sstevel@tonic-gate } 10687c478bd9Sstevel@tonic-gate 10697c478bd9Sstevel@tonic-gate /* 10707c478bd9Sstevel@tonic-gate * Unlink an entry for a table at vaddr and level out of the existing table 10717c478bd9Sstevel@tonic-gate * one level higher. We are always holding the HASH_ENTER() when doing this. 10727c478bd9Sstevel@tonic-gate */ 10737c478bd9Sstevel@tonic-gate static void 10747c478bd9Sstevel@tonic-gate unlink_ptp(htable_t *higher, htable_t *old, uintptr_t vaddr) 10757c478bd9Sstevel@tonic-gate { 10767c478bd9Sstevel@tonic-gate uint_t entry = htable_va2entry(vaddr, higher); 10777c478bd9Sstevel@tonic-gate x86pte_t expect = MAKEPTP(old->ht_pfn, old->ht_level); 10787c478bd9Sstevel@tonic-gate x86pte_t found; 1079935f8dd0Sjosephb hat_t *hat = old->ht_hat; 10807c478bd9Sstevel@tonic-gate 10817c478bd9Sstevel@tonic-gate ASSERT(higher->ht_busy > 0); 10827c478bd9Sstevel@tonic-gate ASSERT(higher->ht_valid_cnt > 0); 10837c478bd9Sstevel@tonic-gate ASSERT(old->ht_valid_cnt == 0); 10847c478bd9Sstevel@tonic-gate found = x86pte_cas(higher, entry, expect, 0); 1085843e1988Sjohnlev #ifdef __xpv 1086843e1988Sjohnlev /* 1087843e1988Sjohnlev * This is weird, but Xen apparently automatically unlinks empty 1088843e1988Sjohnlev * pagetables from the upper page table. So allow PTP to be 0 already. 1089843e1988Sjohnlev */ 1090843e1988Sjohnlev if (found != expect && found != 0) 1091843e1988Sjohnlev #else 10927c478bd9Sstevel@tonic-gate if (found != expect) 1093843e1988Sjohnlev #endif 10947c478bd9Sstevel@tonic-gate panic("Bad PTP found=" FMT_PTE ", expected=" FMT_PTE, 10957c478bd9Sstevel@tonic-gate found, expect); 1096935f8dd0Sjosephb 1097935f8dd0Sjosephb /* 10987173d045Sjosephb * When a top level VLP page table entry changes, we must issue 10997173d045Sjosephb * a reload of cr3 on all processors. 11007173d045Sjosephb * 11017173d045Sjosephb * If we don't need do do that, then we still have to INVLPG against 11027173d045Sjosephb * an address covered by the inner page table, as the latest processors 11037173d045Sjosephb * have TLB-like caches for non-leaf page table entries. 1104935f8dd0Sjosephb */ 1105935f8dd0Sjosephb if (!(hat->hat_flags & HAT_FREEING)) { 11067173d045Sjosephb hat_tlb_inval(hat, (higher->ht_flags & HTABLE_VLP) ? 11077173d045Sjosephb DEMAP_ALL_ADDR : old->ht_vaddr); 1108935f8dd0Sjosephb } 1109935f8dd0Sjosephb 11107c478bd9Sstevel@tonic-gate HTABLE_DEC(higher->ht_valid_cnt); 11117c478bd9Sstevel@tonic-gate } 11127c478bd9Sstevel@tonic-gate 11137c478bd9Sstevel@tonic-gate /* 11147c478bd9Sstevel@tonic-gate * Link an entry for a new table at vaddr and level into the existing table 11157c478bd9Sstevel@tonic-gate * one level higher. We are always holding the HASH_ENTER() when doing this. 11167c478bd9Sstevel@tonic-gate */ 11177c478bd9Sstevel@tonic-gate static void 11187c478bd9Sstevel@tonic-gate link_ptp(htable_t *higher, htable_t *new, uintptr_t vaddr) 11197c478bd9Sstevel@tonic-gate { 11207c478bd9Sstevel@tonic-gate uint_t entry = htable_va2entry(vaddr, higher); 11217c478bd9Sstevel@tonic-gate x86pte_t newptp = MAKEPTP(new->ht_pfn, new->ht_level); 11227c478bd9Sstevel@tonic-gate x86pte_t found; 11237c478bd9Sstevel@tonic-gate 11247c478bd9Sstevel@tonic-gate ASSERT(higher->ht_busy > 0); 11257c478bd9Sstevel@tonic-gate 11267c478bd9Sstevel@tonic-gate ASSERT(new->ht_level != mmu.max_level); 11277c478bd9Sstevel@tonic-gate 11287c478bd9Sstevel@tonic-gate HTABLE_INC(higher->ht_valid_cnt); 11297c478bd9Sstevel@tonic-gate 11307c478bd9Sstevel@tonic-gate found = x86pte_cas(higher, entry, 0, newptp); 1131b4b46911Skchow if ((found & ~PT_REF) != 0) 11327c478bd9Sstevel@tonic-gate panic("HAT: ptp not 0, found=" FMT_PTE, found); 1133935f8dd0Sjosephb 1134935f8dd0Sjosephb /* 1135935f8dd0Sjosephb * When any top level VLP page table entry changes, we must issue 1136935f8dd0Sjosephb * a reload of cr3 on all processors using it. 11376b60931cSjosephb * We also need to do this for the kernel hat on PAE 32 bit kernel. 1138935f8dd0Sjosephb */ 11396b60931cSjosephb if ( 11406b60931cSjosephb #ifdef __i386 11416b60931cSjosephb (higher->ht_hat == kas.a_hat && higher->ht_level == VLP_LEVEL) || 11426b60931cSjosephb #endif 11436b60931cSjosephb (higher->ht_flags & HTABLE_VLP)) 1144935f8dd0Sjosephb hat_tlb_inval(higher->ht_hat, DEMAP_ALL_ADDR); 11457c478bd9Sstevel@tonic-gate } 11467c478bd9Sstevel@tonic-gate 11477c478bd9Sstevel@tonic-gate /* 1148ae115bc7Smrj * Release of hold on an htable. If this is the last use and the pagetable 1149ae115bc7Smrj * is empty we may want to free it, then recursively look at the pagetable 1150ae115bc7Smrj * above it. The recursion is handled by the outer while() loop. 1151843e1988Sjohnlev * 1152843e1988Sjohnlev * On the metal, during process exit, we don't bother unlinking the tables from 1153843e1988Sjohnlev * upper level pagetables. They are instead handled in bulk by hat_free_end(). 1154843e1988Sjohnlev * We can't do this on the hypervisor as we need the page table to be 1155843e1988Sjohnlev * implicitly unpinnned before it goes to the free page lists. This can't 1156843e1988Sjohnlev * happen unless we fully unlink it from the page table hierarchy. 11577c478bd9Sstevel@tonic-gate */ 11587c478bd9Sstevel@tonic-gate void 11597c478bd9Sstevel@tonic-gate htable_release(htable_t *ht) 11607c478bd9Sstevel@tonic-gate { 11617c478bd9Sstevel@tonic-gate uint_t hashval; 11627c478bd9Sstevel@tonic-gate htable_t *shared; 11637c478bd9Sstevel@tonic-gate htable_t *higher; 11647c478bd9Sstevel@tonic-gate hat_t *hat; 11657c478bd9Sstevel@tonic-gate uintptr_t va; 11667c478bd9Sstevel@tonic-gate level_t level; 11677c478bd9Sstevel@tonic-gate 11687c478bd9Sstevel@tonic-gate while (ht != NULL) { 11697c478bd9Sstevel@tonic-gate shared = NULL; 11707c478bd9Sstevel@tonic-gate for (;;) { 11717c478bd9Sstevel@tonic-gate hat = ht->ht_hat; 11727c478bd9Sstevel@tonic-gate va = ht->ht_vaddr; 11737c478bd9Sstevel@tonic-gate level = ht->ht_level; 11747c478bd9Sstevel@tonic-gate hashval = HTABLE_HASH(hat, va, level); 11757c478bd9Sstevel@tonic-gate 11767c478bd9Sstevel@tonic-gate /* 11777c478bd9Sstevel@tonic-gate * The common case is that this isn't the last use of 11787c478bd9Sstevel@tonic-gate * an htable so we don't want to free the htable. 11797c478bd9Sstevel@tonic-gate */ 11807c478bd9Sstevel@tonic-gate HTABLE_ENTER(hashval); 11817c478bd9Sstevel@tonic-gate ASSERT(ht->ht_valid_cnt >= 0); 11827c478bd9Sstevel@tonic-gate ASSERT(ht->ht_busy > 0); 11837c478bd9Sstevel@tonic-gate if (ht->ht_valid_cnt > 0) 11847c478bd9Sstevel@tonic-gate break; 11857c478bd9Sstevel@tonic-gate if (ht->ht_busy > 1) 11867c478bd9Sstevel@tonic-gate break; 11872ba723d8Smec ASSERT(ht->ht_lock_cnt == 0); 11887c478bd9Sstevel@tonic-gate 1189843e1988Sjohnlev #if !defined(__xpv) 11907c478bd9Sstevel@tonic-gate /* 11917c478bd9Sstevel@tonic-gate * we always release empty shared htables 11927c478bd9Sstevel@tonic-gate */ 11937c478bd9Sstevel@tonic-gate if (!(ht->ht_flags & HTABLE_SHARED_PFN)) { 11947c478bd9Sstevel@tonic-gate 11957c478bd9Sstevel@tonic-gate /* 11967c478bd9Sstevel@tonic-gate * don't release if in address space tear down 11977c478bd9Sstevel@tonic-gate */ 11987c478bd9Sstevel@tonic-gate if (hat->hat_flags & HAT_FREEING) 11997c478bd9Sstevel@tonic-gate break; 12007c478bd9Sstevel@tonic-gate 12017c478bd9Sstevel@tonic-gate /* 12027c478bd9Sstevel@tonic-gate * At and above max_page_level, free if it's for 12037c478bd9Sstevel@tonic-gate * a boot-time kernel mapping below kernelbase. 12047c478bd9Sstevel@tonic-gate */ 12057c478bd9Sstevel@tonic-gate if (level >= mmu.max_page_level && 12067c478bd9Sstevel@tonic-gate (hat != kas.a_hat || va >= kernelbase)) 12077c478bd9Sstevel@tonic-gate break; 12087c478bd9Sstevel@tonic-gate } 1209843e1988Sjohnlev #endif /* __xpv */ 12107c478bd9Sstevel@tonic-gate 12117c478bd9Sstevel@tonic-gate /* 1212ae115bc7Smrj * Remember if we destroy an htable that shares its PFN 1213ae115bc7Smrj * from elsewhere. 12147c478bd9Sstevel@tonic-gate */ 12157c478bd9Sstevel@tonic-gate if (ht->ht_flags & HTABLE_SHARED_PFN) { 12167c478bd9Sstevel@tonic-gate ASSERT(shared == NULL); 12177c478bd9Sstevel@tonic-gate shared = ht->ht_shares; 12187c478bd9Sstevel@tonic-gate HATSTAT_INC(hs_htable_unshared); 12197c478bd9Sstevel@tonic-gate } 12207c478bd9Sstevel@tonic-gate 12217c478bd9Sstevel@tonic-gate /* 12227c478bd9Sstevel@tonic-gate * Handle release of a table and freeing the htable_t. 12237c478bd9Sstevel@tonic-gate * Unlink it from the table higher (ie. ht_parent). 12247c478bd9Sstevel@tonic-gate */ 12257c478bd9Sstevel@tonic-gate higher = ht->ht_parent; 12267c478bd9Sstevel@tonic-gate ASSERT(higher != NULL); 12277c478bd9Sstevel@tonic-gate 12287c478bd9Sstevel@tonic-gate /* 12297c478bd9Sstevel@tonic-gate * Unlink the pagetable. 12307c478bd9Sstevel@tonic-gate */ 12317c478bd9Sstevel@tonic-gate unlink_ptp(higher, ht, va); 12327c478bd9Sstevel@tonic-gate 12337c478bd9Sstevel@tonic-gate /* 12347c478bd9Sstevel@tonic-gate * remove this htable from its hash list 12357c478bd9Sstevel@tonic-gate */ 12367c478bd9Sstevel@tonic-gate if (ht->ht_next) 12377c478bd9Sstevel@tonic-gate ht->ht_next->ht_prev = ht->ht_prev; 12387c478bd9Sstevel@tonic-gate 12397c478bd9Sstevel@tonic-gate if (ht->ht_prev) { 12407c478bd9Sstevel@tonic-gate ht->ht_prev->ht_next = ht->ht_next; 12417c478bd9Sstevel@tonic-gate } else { 12427c478bd9Sstevel@tonic-gate ASSERT(hat->hat_ht_hash[hashval] == ht); 12437c478bd9Sstevel@tonic-gate hat->hat_ht_hash[hashval] = ht->ht_next; 12447c478bd9Sstevel@tonic-gate } 12457c478bd9Sstevel@tonic-gate HTABLE_EXIT(hashval); 12467c478bd9Sstevel@tonic-gate htable_free(ht); 12477c478bd9Sstevel@tonic-gate ht = higher; 12487c478bd9Sstevel@tonic-gate } 12497c478bd9Sstevel@tonic-gate 12507c478bd9Sstevel@tonic-gate ASSERT(ht->ht_busy >= 1); 12517c478bd9Sstevel@tonic-gate --ht->ht_busy; 12527c478bd9Sstevel@tonic-gate HTABLE_EXIT(hashval); 12537c478bd9Sstevel@tonic-gate 12547c478bd9Sstevel@tonic-gate /* 12557c478bd9Sstevel@tonic-gate * If we released a shared htable, do a release on the htable 12567c478bd9Sstevel@tonic-gate * from which it shared 12577c478bd9Sstevel@tonic-gate */ 12587c478bd9Sstevel@tonic-gate ht = shared; 12597c478bd9Sstevel@tonic-gate } 12607c478bd9Sstevel@tonic-gate } 12617c478bd9Sstevel@tonic-gate 12627c478bd9Sstevel@tonic-gate /* 12637c478bd9Sstevel@tonic-gate * Find the htable for the pagetable at the given level for the given address. 12647c478bd9Sstevel@tonic-gate * If found acquires a hold that eventually needs to be htable_release()d 12657c478bd9Sstevel@tonic-gate */ 12667c478bd9Sstevel@tonic-gate htable_t * 12677c478bd9Sstevel@tonic-gate htable_lookup(hat_t *hat, uintptr_t vaddr, level_t level) 12687c478bd9Sstevel@tonic-gate { 12697c478bd9Sstevel@tonic-gate uintptr_t base; 12707c478bd9Sstevel@tonic-gate uint_t hashval; 12717c478bd9Sstevel@tonic-gate htable_t *ht = NULL; 12727c478bd9Sstevel@tonic-gate 12737c478bd9Sstevel@tonic-gate ASSERT(level >= 0); 12747c478bd9Sstevel@tonic-gate ASSERT(level <= TOP_LEVEL(hat)); 12757c478bd9Sstevel@tonic-gate 12767173d045Sjosephb if (level == TOP_LEVEL(hat)) { 12777173d045Sjosephb #if defined(__amd64) 12787173d045Sjosephb /* 12797173d045Sjosephb * 32 bit address spaces on 64 bit kernels need to check 12807173d045Sjosephb * for overflow of the 32 bit address space 12817173d045Sjosephb */ 12827173d045Sjosephb if ((hat->hat_flags & HAT_VLP) && vaddr >= ((uint64_t)1 << 32)) 12837173d045Sjosephb return (NULL); 12847173d045Sjosephb #endif 12857c478bd9Sstevel@tonic-gate base = 0; 12867173d045Sjosephb } else { 12877c478bd9Sstevel@tonic-gate base = vaddr & LEVEL_MASK(level + 1); 12887173d045Sjosephb } 12897c478bd9Sstevel@tonic-gate 12907c478bd9Sstevel@tonic-gate hashval = HTABLE_HASH(hat, base, level); 12917c478bd9Sstevel@tonic-gate HTABLE_ENTER(hashval); 12927c478bd9Sstevel@tonic-gate for (ht = hat->hat_ht_hash[hashval]; ht; ht = ht->ht_next) { 12937c478bd9Sstevel@tonic-gate if (ht->ht_hat == hat && 12947c478bd9Sstevel@tonic-gate ht->ht_vaddr == base && 12957c478bd9Sstevel@tonic-gate ht->ht_level == level) 12967c478bd9Sstevel@tonic-gate break; 12977c478bd9Sstevel@tonic-gate } 12987c478bd9Sstevel@tonic-gate if (ht) 12997c478bd9Sstevel@tonic-gate ++ht->ht_busy; 13007c478bd9Sstevel@tonic-gate 13017c478bd9Sstevel@tonic-gate HTABLE_EXIT(hashval); 13027c478bd9Sstevel@tonic-gate return (ht); 13037c478bd9Sstevel@tonic-gate } 13047c478bd9Sstevel@tonic-gate 13057c478bd9Sstevel@tonic-gate /* 13067c478bd9Sstevel@tonic-gate * Acquires a hold on a known htable (from a locked hment entry). 13077c478bd9Sstevel@tonic-gate */ 13087c478bd9Sstevel@tonic-gate void 13097c478bd9Sstevel@tonic-gate htable_acquire(htable_t *ht) 13107c478bd9Sstevel@tonic-gate { 13117c478bd9Sstevel@tonic-gate hat_t *hat = ht->ht_hat; 13127c478bd9Sstevel@tonic-gate level_t level = ht->ht_level; 13137c478bd9Sstevel@tonic-gate uintptr_t base = ht->ht_vaddr; 13147c478bd9Sstevel@tonic-gate uint_t hashval = HTABLE_HASH(hat, base, level); 13157c478bd9Sstevel@tonic-gate 13167c478bd9Sstevel@tonic-gate HTABLE_ENTER(hashval); 13177c478bd9Sstevel@tonic-gate #ifdef DEBUG 13187c478bd9Sstevel@tonic-gate /* 13197c478bd9Sstevel@tonic-gate * make sure the htable is there 13207c478bd9Sstevel@tonic-gate */ 13217c478bd9Sstevel@tonic-gate { 13227c478bd9Sstevel@tonic-gate htable_t *h; 13237c478bd9Sstevel@tonic-gate 13247c478bd9Sstevel@tonic-gate for (h = hat->hat_ht_hash[hashval]; 13257c478bd9Sstevel@tonic-gate h && h != ht; 13267c478bd9Sstevel@tonic-gate h = h->ht_next) 13277c478bd9Sstevel@tonic-gate ; 13287c478bd9Sstevel@tonic-gate ASSERT(h == ht); 13297c478bd9Sstevel@tonic-gate } 13307c478bd9Sstevel@tonic-gate #endif /* DEBUG */ 13317c478bd9Sstevel@tonic-gate ++ht->ht_busy; 13327c478bd9Sstevel@tonic-gate HTABLE_EXIT(hashval); 13337c478bd9Sstevel@tonic-gate } 13347c478bd9Sstevel@tonic-gate 13357c478bd9Sstevel@tonic-gate /* 13367c478bd9Sstevel@tonic-gate * Find the htable for the pagetable at the given level for the given address. 13377c478bd9Sstevel@tonic-gate * If found acquires a hold that eventually needs to be htable_release()d 13387c478bd9Sstevel@tonic-gate * If not found the table is created. 13397c478bd9Sstevel@tonic-gate * 13407c478bd9Sstevel@tonic-gate * Since we can't hold a hash table mutex during allocation, we have to 13417c478bd9Sstevel@tonic-gate * drop it and redo the search on a create. Then we may have to free the newly 13427c478bd9Sstevel@tonic-gate * allocated htable if another thread raced in and created it ahead of us. 13437c478bd9Sstevel@tonic-gate */ 13447c478bd9Sstevel@tonic-gate htable_t * 13457c478bd9Sstevel@tonic-gate htable_create( 13467c478bd9Sstevel@tonic-gate hat_t *hat, 13477c478bd9Sstevel@tonic-gate uintptr_t vaddr, 13487c478bd9Sstevel@tonic-gate level_t level, 13497c478bd9Sstevel@tonic-gate htable_t *shared) 13507c478bd9Sstevel@tonic-gate { 13517c478bd9Sstevel@tonic-gate uint_t h; 13527c478bd9Sstevel@tonic-gate level_t l; 13537c478bd9Sstevel@tonic-gate uintptr_t base; 13547c478bd9Sstevel@tonic-gate htable_t *ht; 13557c478bd9Sstevel@tonic-gate htable_t *higher = NULL; 13567c478bd9Sstevel@tonic-gate htable_t *new = NULL; 13577c478bd9Sstevel@tonic-gate 13587c478bd9Sstevel@tonic-gate if (level < 0 || level > TOP_LEVEL(hat)) 13597c478bd9Sstevel@tonic-gate panic("htable_create(): level %d out of range\n", level); 13607c478bd9Sstevel@tonic-gate 13617c478bd9Sstevel@tonic-gate /* 13627c478bd9Sstevel@tonic-gate * Create the page tables in top down order. 13637c478bd9Sstevel@tonic-gate */ 13647c478bd9Sstevel@tonic-gate for (l = TOP_LEVEL(hat); l >= level; --l) { 13657c478bd9Sstevel@tonic-gate new = NULL; 13667c478bd9Sstevel@tonic-gate if (l == TOP_LEVEL(hat)) 13677c478bd9Sstevel@tonic-gate base = 0; 13687c478bd9Sstevel@tonic-gate else 13697c478bd9Sstevel@tonic-gate base = vaddr & LEVEL_MASK(l + 1); 13707c478bd9Sstevel@tonic-gate 13717c478bd9Sstevel@tonic-gate h = HTABLE_HASH(hat, base, l); 13727c478bd9Sstevel@tonic-gate try_again: 13737c478bd9Sstevel@tonic-gate /* 13747c478bd9Sstevel@tonic-gate * look up the htable at this level 13757c478bd9Sstevel@tonic-gate */ 13767c478bd9Sstevel@tonic-gate HTABLE_ENTER(h); 13777c478bd9Sstevel@tonic-gate if (l == TOP_LEVEL(hat)) { 13787c478bd9Sstevel@tonic-gate ht = hat->hat_htable; 13797c478bd9Sstevel@tonic-gate } else { 13807c478bd9Sstevel@tonic-gate for (ht = hat->hat_ht_hash[h]; ht; ht = ht->ht_next) { 13817c478bd9Sstevel@tonic-gate ASSERT(ht->ht_hat == hat); 13827c478bd9Sstevel@tonic-gate if (ht->ht_vaddr == base && 13837c478bd9Sstevel@tonic-gate ht->ht_level == l) 13847c478bd9Sstevel@tonic-gate break; 13857c478bd9Sstevel@tonic-gate } 13867c478bd9Sstevel@tonic-gate } 13877c478bd9Sstevel@tonic-gate 13887c478bd9Sstevel@tonic-gate /* 13897c478bd9Sstevel@tonic-gate * if we found the htable, increment its busy cnt 13907c478bd9Sstevel@tonic-gate * and if we had allocated a new htable, free it. 13917c478bd9Sstevel@tonic-gate */ 13927c478bd9Sstevel@tonic-gate if (ht != NULL) { 13937c478bd9Sstevel@tonic-gate /* 13947c478bd9Sstevel@tonic-gate * If we find a pre-existing shared table, it must 13957c478bd9Sstevel@tonic-gate * share from the same place. 13967c478bd9Sstevel@tonic-gate */ 13977c478bd9Sstevel@tonic-gate if (l == level && shared && ht->ht_shares && 13987c478bd9Sstevel@tonic-gate ht->ht_shares != shared) { 13997c478bd9Sstevel@tonic-gate panic("htable shared from wrong place " 1400903a11ebSrh87107 "found htable=%p shared=%p", 1401903a11ebSrh87107 (void *)ht, (void *)shared); 14027c478bd9Sstevel@tonic-gate } 14037c478bd9Sstevel@tonic-gate ++ht->ht_busy; 14047c478bd9Sstevel@tonic-gate HTABLE_EXIT(h); 14057c478bd9Sstevel@tonic-gate if (new) 14067c478bd9Sstevel@tonic-gate htable_free(new); 14077c478bd9Sstevel@tonic-gate if (higher != NULL) 14087c478bd9Sstevel@tonic-gate htable_release(higher); 14097c478bd9Sstevel@tonic-gate higher = ht; 14107c478bd9Sstevel@tonic-gate 14117c478bd9Sstevel@tonic-gate /* 14127c478bd9Sstevel@tonic-gate * if we didn't find it on the first search 14137c478bd9Sstevel@tonic-gate * allocate a new one and search again 14147c478bd9Sstevel@tonic-gate */ 14157c478bd9Sstevel@tonic-gate } else if (new == NULL) { 14167c478bd9Sstevel@tonic-gate HTABLE_EXIT(h); 14177c478bd9Sstevel@tonic-gate new = htable_alloc(hat, base, l, 14187c478bd9Sstevel@tonic-gate l == level ? shared : NULL); 14197c478bd9Sstevel@tonic-gate goto try_again; 14207c478bd9Sstevel@tonic-gate 14217c478bd9Sstevel@tonic-gate /* 14227c478bd9Sstevel@tonic-gate * 2nd search and still not there, use "new" table 14237c478bd9Sstevel@tonic-gate * Link new table into higher, when not at top level. 14247c478bd9Sstevel@tonic-gate */ 14257c478bd9Sstevel@tonic-gate } else { 14267c478bd9Sstevel@tonic-gate ht = new; 14277c478bd9Sstevel@tonic-gate if (higher != NULL) { 14287c478bd9Sstevel@tonic-gate link_ptp(higher, ht, base); 14297c478bd9Sstevel@tonic-gate ht->ht_parent = higher; 14307c478bd9Sstevel@tonic-gate } 14317c478bd9Sstevel@tonic-gate ht->ht_next = hat->hat_ht_hash[h]; 14327c478bd9Sstevel@tonic-gate ASSERT(ht->ht_prev == NULL); 14337c478bd9Sstevel@tonic-gate if (hat->hat_ht_hash[h]) 14347c478bd9Sstevel@tonic-gate hat->hat_ht_hash[h]->ht_prev = ht; 14357c478bd9Sstevel@tonic-gate hat->hat_ht_hash[h] = ht; 14367c478bd9Sstevel@tonic-gate HTABLE_EXIT(h); 14377c478bd9Sstevel@tonic-gate 14387c478bd9Sstevel@tonic-gate /* 14397c478bd9Sstevel@tonic-gate * Note we don't do htable_release(higher). 14407c478bd9Sstevel@tonic-gate * That happens recursively when "new" is removed by 14417c478bd9Sstevel@tonic-gate * htable_release() or htable_steal(). 14427c478bd9Sstevel@tonic-gate */ 14437c478bd9Sstevel@tonic-gate higher = ht; 14447c478bd9Sstevel@tonic-gate 14457c478bd9Sstevel@tonic-gate /* 14467c478bd9Sstevel@tonic-gate * If we just created a new shared page table we 14477c478bd9Sstevel@tonic-gate * increment the shared htable's busy count, so that 14487c478bd9Sstevel@tonic-gate * it can't be the victim of a steal even if it's empty. 14497c478bd9Sstevel@tonic-gate */ 14507c478bd9Sstevel@tonic-gate if (l == level && shared) { 14517c478bd9Sstevel@tonic-gate (void) htable_lookup(shared->ht_hat, 14527c478bd9Sstevel@tonic-gate shared->ht_vaddr, shared->ht_level); 14537c478bd9Sstevel@tonic-gate HATSTAT_INC(hs_htable_shared); 14547c478bd9Sstevel@tonic-gate } 14557c478bd9Sstevel@tonic-gate } 14567c478bd9Sstevel@tonic-gate } 14577c478bd9Sstevel@tonic-gate 14587c478bd9Sstevel@tonic-gate return (ht); 14597c478bd9Sstevel@tonic-gate } 14607c478bd9Sstevel@tonic-gate 14617c478bd9Sstevel@tonic-gate /* 1462843e1988Sjohnlev * Inherit initial pagetables from the boot program. On the 64-bit 1463843e1988Sjohnlev * hypervisor we also temporarily mark the p_index field of page table 1464843e1988Sjohnlev * pages, so we know not to try making them writable in seg_kpm. 1465ae115bc7Smrj */ 1466ae115bc7Smrj void 1467ae115bc7Smrj htable_attach( 1468ae115bc7Smrj hat_t *hat, 1469ae115bc7Smrj uintptr_t base, 1470ae115bc7Smrj level_t level, 1471ae115bc7Smrj htable_t *parent, 1472ae115bc7Smrj pfn_t pfn) 1473ae115bc7Smrj { 1474ae115bc7Smrj htable_t *ht; 1475ae115bc7Smrj uint_t h; 1476ae115bc7Smrj uint_t i; 1477ae115bc7Smrj x86pte_t pte; 1478ae115bc7Smrj x86pte_t *ptep; 1479ae115bc7Smrj page_t *pp; 1480ae115bc7Smrj extern page_t *boot_claim_page(pfn_t); 1481ae115bc7Smrj 1482ae115bc7Smrj ht = htable_get_reserve(); 1483ae115bc7Smrj if (level == mmu.max_level) 1484ae115bc7Smrj kas.a_hat->hat_htable = ht; 1485ae115bc7Smrj ht->ht_hat = hat; 1486ae115bc7Smrj ht->ht_parent = parent; 1487ae115bc7Smrj ht->ht_vaddr = base; 1488ae115bc7Smrj ht->ht_level = level; 1489ae115bc7Smrj ht->ht_busy = 1; 1490ae115bc7Smrj ht->ht_next = NULL; 1491ae115bc7Smrj ht->ht_prev = NULL; 1492ae115bc7Smrj ht->ht_flags = 0; 1493ae115bc7Smrj ht->ht_pfn = pfn; 1494ae115bc7Smrj ht->ht_lock_cnt = 0; 1495ae115bc7Smrj ht->ht_valid_cnt = 0; 1496ae115bc7Smrj if (parent != NULL) 1497ae115bc7Smrj ++parent->ht_busy; 1498ae115bc7Smrj 1499ae115bc7Smrj h = HTABLE_HASH(hat, base, level); 1500ae115bc7Smrj HTABLE_ENTER(h); 1501ae115bc7Smrj ht->ht_next = hat->hat_ht_hash[h]; 1502ae115bc7Smrj ASSERT(ht->ht_prev == NULL); 1503ae115bc7Smrj if (hat->hat_ht_hash[h]) 1504ae115bc7Smrj hat->hat_ht_hash[h]->ht_prev = ht; 1505ae115bc7Smrj hat->hat_ht_hash[h] = ht; 1506ae115bc7Smrj HTABLE_EXIT(h); 1507ae115bc7Smrj 1508ae115bc7Smrj /* 1509ae115bc7Smrj * make sure the page table physical page is not FREE 1510ae115bc7Smrj */ 1511ae115bc7Smrj if (page_resv(1, KM_NOSLEEP) == 0) 1512ae115bc7Smrj panic("page_resv() failed in ptable alloc"); 1513ae115bc7Smrj 1514ae115bc7Smrj pp = boot_claim_page(pfn); 1515ae115bc7Smrj ASSERT(pp != NULL); 15162d44e974SJoe Bonasera 15172d44e974SJoe Bonasera /* 15182d44e974SJoe Bonasera * Page table pages that were allocated by dboot or 15192d44e974SJoe Bonasera * in very early startup didn't go through boot_mapin() 15202d44e974SJoe Bonasera * and so won't have vnode/offsets. Fix that here. 15212d44e974SJoe Bonasera */ 15222d44e974SJoe Bonasera if (pp->p_vnode == NULL) { 15232d44e974SJoe Bonasera /* match offset calculation in page_get_physical() */ 15242d44e974SJoe Bonasera u_offset_t offset = (uintptr_t)ht; 15252d44e974SJoe Bonasera if (offset > kernelbase) 15262d44e974SJoe Bonasera offset -= kernelbase; 15272d44e974SJoe Bonasera offset <<= MMU_PAGESHIFT; 15282d44e974SJoe Bonasera #if defined(__amd64) 15292d44e974SJoe Bonasera offset += mmu.hole_start; /* something in VA hole */ 15302d44e974SJoe Bonasera #else 15312d44e974SJoe Bonasera offset += 1ULL << 40; /* something > 4 Gig */ 15322d44e974SJoe Bonasera #endif 15332d44e974SJoe Bonasera ASSERT(page_exists(&kvp, offset) == NULL); 15342d44e974SJoe Bonasera (void) page_hashin(pp, &kvp, offset, NULL); 15352d44e974SJoe Bonasera } 1536ae115bc7Smrj page_downgrade(pp); 1537843e1988Sjohnlev #if defined(__xpv) && defined(__amd64) 1538ae115bc7Smrj /* 1539ae115bc7Smrj * Record in the page_t that is a pagetable for segkpm setup. 1540ae115bc7Smrj */ 1541ae115bc7Smrj if (kpm_vbase) 1542ae115bc7Smrj pp->p_index = 1; 1543843e1988Sjohnlev #endif 1544ae115bc7Smrj 1545ae115bc7Smrj /* 1546ae115bc7Smrj * Count valid mappings and recursively attach lower level pagetables. 1547ae115bc7Smrj */ 1548ae115bc7Smrj ptep = kbm_remap_window(pfn_to_pa(pfn), 0); 1549ae115bc7Smrj for (i = 0; i < HTABLE_NUM_PTES(ht); ++i) { 1550ae115bc7Smrj if (mmu.pae_hat) 1551ae115bc7Smrj pte = ptep[i]; 1552ae115bc7Smrj else 1553ae115bc7Smrj pte = ((x86pte32_t *)ptep)[i]; 1554ae115bc7Smrj if (!IN_HYPERVISOR_VA(base) && PTE_ISVALID(pte)) { 1555ae115bc7Smrj ++ht->ht_valid_cnt; 1556ae115bc7Smrj if (!PTE_ISPAGE(pte, level)) { 1557ae115bc7Smrj htable_attach(hat, base, level - 1, 1558ae115bc7Smrj ht, PTE2PFN(pte, level)); 1559ae115bc7Smrj ptep = kbm_remap_window(pfn_to_pa(pfn), 0); 1560ae115bc7Smrj } 1561ae115bc7Smrj } 1562ae115bc7Smrj base += LEVEL_SIZE(level); 1563ae115bc7Smrj if (base == mmu.hole_start) 1564ae115bc7Smrj base = (mmu.hole_end + MMU_PAGEOFFSET) & MMU_PAGEMASK; 1565ae115bc7Smrj } 1566ae115bc7Smrj 1567ae115bc7Smrj /* 1568ae115bc7Smrj * As long as all the mappings we had were below kernel base 1569ae115bc7Smrj * we can release the htable. 1570ae115bc7Smrj */ 1571ae115bc7Smrj if (base < kernelbase) 1572ae115bc7Smrj htable_release(ht); 1573ae115bc7Smrj } 1574ae115bc7Smrj 1575ae115bc7Smrj /* 15767c478bd9Sstevel@tonic-gate * Walk through a given htable looking for the first valid entry. This 15777c478bd9Sstevel@tonic-gate * routine takes both a starting and ending address. The starting address 15787c478bd9Sstevel@tonic-gate * is required to be within the htable provided by the caller, but there is 15797c478bd9Sstevel@tonic-gate * no such restriction on the ending address. 15807c478bd9Sstevel@tonic-gate * 15817c478bd9Sstevel@tonic-gate * If the routine finds a valid entry in the htable (at or beyond the 15827c478bd9Sstevel@tonic-gate * starting address), the PTE (and its address) will be returned. 15837c478bd9Sstevel@tonic-gate * This PTE may correspond to either a page or a pagetable - it is the 15847c478bd9Sstevel@tonic-gate * caller's responsibility to determine which. If no valid entry is 15857c478bd9Sstevel@tonic-gate * found, 0 (and invalid PTE) and the next unexamined address will be 15867c478bd9Sstevel@tonic-gate * returned. 15877c478bd9Sstevel@tonic-gate * 15887c478bd9Sstevel@tonic-gate * The loop has been carefully coded for optimization. 15897c478bd9Sstevel@tonic-gate */ 15907c478bd9Sstevel@tonic-gate static x86pte_t 15917c478bd9Sstevel@tonic-gate htable_scan(htable_t *ht, uintptr_t *vap, uintptr_t eaddr) 15927c478bd9Sstevel@tonic-gate { 15937c478bd9Sstevel@tonic-gate uint_t e; 15947c478bd9Sstevel@tonic-gate x86pte_t found_pte = (x86pte_t)0; 1595ae115bc7Smrj caddr_t pte_ptr; 1596ae115bc7Smrj caddr_t end_pte_ptr; 15977c478bd9Sstevel@tonic-gate int l = ht->ht_level; 15987c478bd9Sstevel@tonic-gate uintptr_t va = *vap & LEVEL_MASK(l); 15997c478bd9Sstevel@tonic-gate size_t pgsize = LEVEL_SIZE(l); 16007c478bd9Sstevel@tonic-gate 16017c478bd9Sstevel@tonic-gate ASSERT(va >= ht->ht_vaddr); 16027c478bd9Sstevel@tonic-gate ASSERT(va <= HTABLE_LAST_PAGE(ht)); 16037c478bd9Sstevel@tonic-gate 16047c478bd9Sstevel@tonic-gate /* 16057c478bd9Sstevel@tonic-gate * Compute the starting index and ending virtual address 16067c478bd9Sstevel@tonic-gate */ 16077c478bd9Sstevel@tonic-gate e = htable_va2entry(va, ht); 16087c478bd9Sstevel@tonic-gate 16097c478bd9Sstevel@tonic-gate /* 16107c478bd9Sstevel@tonic-gate * The following page table scan code knows that the valid 16117c478bd9Sstevel@tonic-gate * bit of a PTE is in the lowest byte AND that x86 is little endian!! 16127c478bd9Sstevel@tonic-gate */ 1613ae115bc7Smrj pte_ptr = (caddr_t)x86pte_access_pagetable(ht, 0); 1614ae115bc7Smrj end_pte_ptr = (caddr_t)PT_INDEX_PTR(pte_ptr, HTABLE_NUM_PTES(ht)); 1615ae115bc7Smrj pte_ptr = (caddr_t)PT_INDEX_PTR((x86pte_t *)pte_ptr, e); 161630f7a194Skchow while (!PTE_ISVALID(*pte_ptr)) { 16177c478bd9Sstevel@tonic-gate va += pgsize; 16187c478bd9Sstevel@tonic-gate if (va >= eaddr) 16197c478bd9Sstevel@tonic-gate break; 16207c478bd9Sstevel@tonic-gate pte_ptr += mmu.pte_size; 16217c478bd9Sstevel@tonic-gate ASSERT(pte_ptr <= end_pte_ptr); 16227c478bd9Sstevel@tonic-gate if (pte_ptr == end_pte_ptr) 16237c478bd9Sstevel@tonic-gate break; 16247c478bd9Sstevel@tonic-gate } 16257c478bd9Sstevel@tonic-gate 16267c478bd9Sstevel@tonic-gate /* 16277c478bd9Sstevel@tonic-gate * if we found a valid PTE, load the entire PTE 16287c478bd9Sstevel@tonic-gate */ 1629ae115bc7Smrj if (va < eaddr && pte_ptr != end_pte_ptr) 1630ae115bc7Smrj found_pte = GET_PTE((x86pte_t *)pte_ptr); 16317c478bd9Sstevel@tonic-gate x86pte_release_pagetable(ht); 16327c478bd9Sstevel@tonic-gate 16337c478bd9Sstevel@tonic-gate #if defined(__amd64) 16347c478bd9Sstevel@tonic-gate /* 16357c478bd9Sstevel@tonic-gate * deal with VA hole on amd64 16367c478bd9Sstevel@tonic-gate */ 16377c478bd9Sstevel@tonic-gate if (l == mmu.max_level && va >= mmu.hole_start && va <= mmu.hole_end) 16387c478bd9Sstevel@tonic-gate va = mmu.hole_end + va - mmu.hole_start; 16397c478bd9Sstevel@tonic-gate #endif /* __amd64 */ 16407c478bd9Sstevel@tonic-gate 16417c478bd9Sstevel@tonic-gate *vap = va; 16427c478bd9Sstevel@tonic-gate return (found_pte); 16437c478bd9Sstevel@tonic-gate } 16447c478bd9Sstevel@tonic-gate 16457c478bd9Sstevel@tonic-gate /* 16467c478bd9Sstevel@tonic-gate * Find the address and htable for the first populated translation at or 16477c478bd9Sstevel@tonic-gate * above the given virtual address. The caller may also specify an upper 16487c478bd9Sstevel@tonic-gate * limit to the address range to search. Uses level information to quickly 16497c478bd9Sstevel@tonic-gate * skip unpopulated sections of virtual address spaces. 16507c478bd9Sstevel@tonic-gate * 16517c478bd9Sstevel@tonic-gate * If not found returns NULL. When found, returns the htable and virt addr 16527c478bd9Sstevel@tonic-gate * and has a hold on the htable. 16537c478bd9Sstevel@tonic-gate */ 16547c478bd9Sstevel@tonic-gate x86pte_t 16557c478bd9Sstevel@tonic-gate htable_walk( 16567c478bd9Sstevel@tonic-gate struct hat *hat, 16577c478bd9Sstevel@tonic-gate htable_t **htp, 16587c478bd9Sstevel@tonic-gate uintptr_t *vaddr, 16597c478bd9Sstevel@tonic-gate uintptr_t eaddr) 16607c478bd9Sstevel@tonic-gate { 16617c478bd9Sstevel@tonic-gate uintptr_t va = *vaddr; 16627c478bd9Sstevel@tonic-gate htable_t *ht; 16637c478bd9Sstevel@tonic-gate htable_t *prev = *htp; 16647c478bd9Sstevel@tonic-gate level_t l; 16657c478bd9Sstevel@tonic-gate level_t max_mapped_level; 16667c478bd9Sstevel@tonic-gate x86pte_t pte; 16677c478bd9Sstevel@tonic-gate 16687c478bd9Sstevel@tonic-gate ASSERT(eaddr > va); 16697c478bd9Sstevel@tonic-gate 16707c478bd9Sstevel@tonic-gate /* 16717c478bd9Sstevel@tonic-gate * If this is a user address, then we know we need not look beyond 16727c478bd9Sstevel@tonic-gate * kernelbase. 16737c478bd9Sstevel@tonic-gate */ 16747c478bd9Sstevel@tonic-gate ASSERT(hat == kas.a_hat || eaddr <= kernelbase || 16757c478bd9Sstevel@tonic-gate eaddr == HTABLE_WALK_TO_END); 16767c478bd9Sstevel@tonic-gate if (hat != kas.a_hat && eaddr == HTABLE_WALK_TO_END) 16777c478bd9Sstevel@tonic-gate eaddr = kernelbase; 16787c478bd9Sstevel@tonic-gate 16797c478bd9Sstevel@tonic-gate /* 16807c478bd9Sstevel@tonic-gate * If we're coming in with a previous page table, search it first 16817c478bd9Sstevel@tonic-gate * without doing an htable_lookup(), this should be frequent. 16827c478bd9Sstevel@tonic-gate */ 16837c478bd9Sstevel@tonic-gate if (prev) { 16847c478bd9Sstevel@tonic-gate ASSERT(prev->ht_busy > 0); 16857c478bd9Sstevel@tonic-gate ASSERT(prev->ht_vaddr <= va); 16867c478bd9Sstevel@tonic-gate l = prev->ht_level; 16877c478bd9Sstevel@tonic-gate if (va <= HTABLE_LAST_PAGE(prev)) { 16887c478bd9Sstevel@tonic-gate pte = htable_scan(prev, &va, eaddr); 16897c478bd9Sstevel@tonic-gate 16907c478bd9Sstevel@tonic-gate if (PTE_ISPAGE(pte, l)) { 16917c478bd9Sstevel@tonic-gate *vaddr = va; 16927c478bd9Sstevel@tonic-gate *htp = prev; 16937c478bd9Sstevel@tonic-gate return (pte); 16947c478bd9Sstevel@tonic-gate } 16957c478bd9Sstevel@tonic-gate } 16967c478bd9Sstevel@tonic-gate 16977c478bd9Sstevel@tonic-gate /* 16987c478bd9Sstevel@tonic-gate * We found nothing in the htable provided by the caller, 16997c478bd9Sstevel@tonic-gate * so fall through and do the full search 17007c478bd9Sstevel@tonic-gate */ 17017c478bd9Sstevel@tonic-gate htable_release(prev); 17027c478bd9Sstevel@tonic-gate } 17037c478bd9Sstevel@tonic-gate 17047c478bd9Sstevel@tonic-gate /* 17057c478bd9Sstevel@tonic-gate * Find the level of the largest pagesize used by this HAT. 17067c478bd9Sstevel@tonic-gate */ 17077173d045Sjosephb if (hat->hat_ism_pgcnt > 0) { 170802bc52beSkchow max_mapped_level = mmu.umax_page_level; 17097173d045Sjosephb } else { 17107c478bd9Sstevel@tonic-gate max_mapped_level = 0; 17117c478bd9Sstevel@tonic-gate for (l = 1; l <= mmu.max_page_level; ++l) 17127c478bd9Sstevel@tonic-gate if (hat->hat_pages_mapped[l] != 0) 17137c478bd9Sstevel@tonic-gate max_mapped_level = l; 17147173d045Sjosephb } 17157c478bd9Sstevel@tonic-gate 17167c478bd9Sstevel@tonic-gate while (va < eaddr && va >= *vaddr) { 17177c478bd9Sstevel@tonic-gate ASSERT(!IN_VA_HOLE(va)); 17187c478bd9Sstevel@tonic-gate 17197c478bd9Sstevel@tonic-gate /* 17207c478bd9Sstevel@tonic-gate * Find lowest table with any entry for given address. 17217c478bd9Sstevel@tonic-gate */ 17227c478bd9Sstevel@tonic-gate for (l = 0; l <= TOP_LEVEL(hat); ++l) { 17237c478bd9Sstevel@tonic-gate ht = htable_lookup(hat, va, l); 17247c478bd9Sstevel@tonic-gate if (ht != NULL) { 17257c478bd9Sstevel@tonic-gate pte = htable_scan(ht, &va, eaddr); 17267c478bd9Sstevel@tonic-gate if (PTE_ISPAGE(pte, l)) { 17277c478bd9Sstevel@tonic-gate *vaddr = va; 17287c478bd9Sstevel@tonic-gate *htp = ht; 17297c478bd9Sstevel@tonic-gate return (pte); 17307c478bd9Sstevel@tonic-gate } 17317c478bd9Sstevel@tonic-gate htable_release(ht); 17327c478bd9Sstevel@tonic-gate break; 17337c478bd9Sstevel@tonic-gate } 17347c478bd9Sstevel@tonic-gate 17357c478bd9Sstevel@tonic-gate /* 17367173d045Sjosephb * No htable at this level for the address. If there 17377173d045Sjosephb * is no larger page size that could cover it, we can 17387173d045Sjosephb * skip right to the start of the next page table. 17398b5842f9Sdm120769 */ 17408b5842f9Sdm120769 ASSERT(l < TOP_LEVEL(hat)); 17418b5842f9Sdm120769 if (l >= max_mapped_level) { 17427c478bd9Sstevel@tonic-gate va = NEXT_ENTRY_VA(va, l + 1); 17437173d045Sjosephb if (va >= eaddr) 17448b5842f9Sdm120769 break; 17458b5842f9Sdm120769 } 17467c478bd9Sstevel@tonic-gate } 17477c478bd9Sstevel@tonic-gate } 17487c478bd9Sstevel@tonic-gate 17497c478bd9Sstevel@tonic-gate *vaddr = 0; 17507c478bd9Sstevel@tonic-gate *htp = NULL; 17517c478bd9Sstevel@tonic-gate return (0); 17527c478bd9Sstevel@tonic-gate } 17537c478bd9Sstevel@tonic-gate 17547c478bd9Sstevel@tonic-gate /* 17557c478bd9Sstevel@tonic-gate * Find the htable and page table entry index of the given virtual address 17567c478bd9Sstevel@tonic-gate * with pagesize at or below given level. 17577c478bd9Sstevel@tonic-gate * If not found returns NULL. When found, returns the htable, sets 17587c478bd9Sstevel@tonic-gate * entry, and has a hold on the htable. 17597c478bd9Sstevel@tonic-gate */ 17607c478bd9Sstevel@tonic-gate htable_t * 17617c478bd9Sstevel@tonic-gate htable_getpte( 17627c478bd9Sstevel@tonic-gate struct hat *hat, 17637c478bd9Sstevel@tonic-gate uintptr_t vaddr, 17647c478bd9Sstevel@tonic-gate uint_t *entry, 17657c478bd9Sstevel@tonic-gate x86pte_t *pte, 17667c478bd9Sstevel@tonic-gate level_t level) 17677c478bd9Sstevel@tonic-gate { 17687c478bd9Sstevel@tonic-gate htable_t *ht; 17697c478bd9Sstevel@tonic-gate level_t l; 17707c478bd9Sstevel@tonic-gate uint_t e; 17717c478bd9Sstevel@tonic-gate 17727c478bd9Sstevel@tonic-gate ASSERT(level <= mmu.max_page_level); 17737c478bd9Sstevel@tonic-gate 17747c478bd9Sstevel@tonic-gate for (l = 0; l <= level; ++l) { 17757c478bd9Sstevel@tonic-gate ht = htable_lookup(hat, vaddr, l); 17767c478bd9Sstevel@tonic-gate if (ht == NULL) 17777c478bd9Sstevel@tonic-gate continue; 17787c478bd9Sstevel@tonic-gate e = htable_va2entry(vaddr, ht); 17797c478bd9Sstevel@tonic-gate if (entry != NULL) 17807c478bd9Sstevel@tonic-gate *entry = e; 17817c478bd9Sstevel@tonic-gate if (pte != NULL) 17827c478bd9Sstevel@tonic-gate *pte = x86pte_get(ht, e); 17837c478bd9Sstevel@tonic-gate return (ht); 17847c478bd9Sstevel@tonic-gate } 17857c478bd9Sstevel@tonic-gate return (NULL); 17867c478bd9Sstevel@tonic-gate } 17877c478bd9Sstevel@tonic-gate 17887c478bd9Sstevel@tonic-gate /* 17897c478bd9Sstevel@tonic-gate * Find the htable and page table entry index of the given virtual address. 17907c478bd9Sstevel@tonic-gate * There must be a valid page mapped at the given address. 17917c478bd9Sstevel@tonic-gate * If not found returns NULL. When found, returns the htable, sets 17927c478bd9Sstevel@tonic-gate * entry, and has a hold on the htable. 17937c478bd9Sstevel@tonic-gate */ 17947c478bd9Sstevel@tonic-gate htable_t * 17957c478bd9Sstevel@tonic-gate htable_getpage(struct hat *hat, uintptr_t vaddr, uint_t *entry) 17967c478bd9Sstevel@tonic-gate { 17977c478bd9Sstevel@tonic-gate htable_t *ht; 17987c478bd9Sstevel@tonic-gate uint_t e; 17997c478bd9Sstevel@tonic-gate x86pte_t pte; 18007c478bd9Sstevel@tonic-gate 18017c478bd9Sstevel@tonic-gate ht = htable_getpte(hat, vaddr, &e, &pte, mmu.max_page_level); 18027c478bd9Sstevel@tonic-gate if (ht == NULL) 18037c478bd9Sstevel@tonic-gate return (NULL); 18047c478bd9Sstevel@tonic-gate 18057c478bd9Sstevel@tonic-gate if (entry) 18067c478bd9Sstevel@tonic-gate *entry = e; 18077c478bd9Sstevel@tonic-gate 18087c478bd9Sstevel@tonic-gate if (PTE_ISPAGE(pte, ht->ht_level)) 18097c478bd9Sstevel@tonic-gate return (ht); 18107c478bd9Sstevel@tonic-gate htable_release(ht); 18117c478bd9Sstevel@tonic-gate return (NULL); 18127c478bd9Sstevel@tonic-gate } 18137c478bd9Sstevel@tonic-gate 18147c478bd9Sstevel@tonic-gate 18157c478bd9Sstevel@tonic-gate void 18167c478bd9Sstevel@tonic-gate htable_init() 18177c478bd9Sstevel@tonic-gate { 18187c478bd9Sstevel@tonic-gate /* 18197c478bd9Sstevel@tonic-gate * To save on kernel VA usage, we avoid debug information in 32 bit 18207c478bd9Sstevel@tonic-gate * kernels. 18217c478bd9Sstevel@tonic-gate */ 18227c478bd9Sstevel@tonic-gate #if defined(__amd64) 18237c478bd9Sstevel@tonic-gate int kmem_flags = KMC_NOHASH; 18247c478bd9Sstevel@tonic-gate #elif defined(__i386) 18257c478bd9Sstevel@tonic-gate int kmem_flags = KMC_NOHASH | KMC_NODEBUG; 18267c478bd9Sstevel@tonic-gate #endif 18277c478bd9Sstevel@tonic-gate 18287c478bd9Sstevel@tonic-gate /* 18297c478bd9Sstevel@tonic-gate * initialize kmem caches 18307c478bd9Sstevel@tonic-gate */ 18317c478bd9Sstevel@tonic-gate htable_cache = kmem_cache_create("htable_t", 18327c478bd9Sstevel@tonic-gate sizeof (htable_t), 0, NULL, NULL, 18337c478bd9Sstevel@tonic-gate htable_reap, NULL, hat_memload_arena, kmem_flags); 18347c478bd9Sstevel@tonic-gate } 18357c478bd9Sstevel@tonic-gate 18367c478bd9Sstevel@tonic-gate /* 18377c478bd9Sstevel@tonic-gate * get the pte index for the virtual address in the given htable's pagetable 18387c478bd9Sstevel@tonic-gate */ 18397c478bd9Sstevel@tonic-gate uint_t 18407c478bd9Sstevel@tonic-gate htable_va2entry(uintptr_t va, htable_t *ht) 18417c478bd9Sstevel@tonic-gate { 18427c478bd9Sstevel@tonic-gate level_t l = ht->ht_level; 18437c478bd9Sstevel@tonic-gate 18447c478bd9Sstevel@tonic-gate ASSERT(va >= ht->ht_vaddr); 18457c478bd9Sstevel@tonic-gate ASSERT(va <= HTABLE_LAST_PAGE(ht)); 1846ae115bc7Smrj return ((va >> LEVEL_SHIFT(l)) & (HTABLE_NUM_PTES(ht) - 1)); 18477c478bd9Sstevel@tonic-gate } 18487c478bd9Sstevel@tonic-gate 18497c478bd9Sstevel@tonic-gate /* 18507c478bd9Sstevel@tonic-gate * Given an htable and the index of a pte in it, return the virtual address 18517c478bd9Sstevel@tonic-gate * of the page. 18527c478bd9Sstevel@tonic-gate */ 18537c478bd9Sstevel@tonic-gate uintptr_t 18547c478bd9Sstevel@tonic-gate htable_e2va(htable_t *ht, uint_t entry) 18557c478bd9Sstevel@tonic-gate { 18567c478bd9Sstevel@tonic-gate level_t l = ht->ht_level; 18577c478bd9Sstevel@tonic-gate uintptr_t va; 18587c478bd9Sstevel@tonic-gate 1859ae115bc7Smrj ASSERT(entry < HTABLE_NUM_PTES(ht)); 18607c478bd9Sstevel@tonic-gate va = ht->ht_vaddr + ((uintptr_t)entry << LEVEL_SHIFT(l)); 18617c478bd9Sstevel@tonic-gate 18627c478bd9Sstevel@tonic-gate /* 18637c478bd9Sstevel@tonic-gate * Need to skip over any VA hole in top level table 18647c478bd9Sstevel@tonic-gate */ 18657c478bd9Sstevel@tonic-gate #if defined(__amd64) 18667c478bd9Sstevel@tonic-gate if (ht->ht_level == mmu.max_level && va >= mmu.hole_start) 18677c478bd9Sstevel@tonic-gate va += ((mmu.hole_end - mmu.hole_start) + 1); 18687c478bd9Sstevel@tonic-gate #endif 18697c478bd9Sstevel@tonic-gate 18707c478bd9Sstevel@tonic-gate return (va); 18717c478bd9Sstevel@tonic-gate } 18727c478bd9Sstevel@tonic-gate 18737c478bd9Sstevel@tonic-gate /* 18747c478bd9Sstevel@tonic-gate * The code uses compare and swap instructions to read/write PTE's to 18757c478bd9Sstevel@tonic-gate * avoid atomicity problems, since PTEs can be 8 bytes on 32 bit systems. 18767c478bd9Sstevel@tonic-gate * will naturally be atomic. 18777c478bd9Sstevel@tonic-gate * 18787c478bd9Sstevel@tonic-gate * The combination of using kpreempt_disable()/_enable() and the hci_mutex 18797c478bd9Sstevel@tonic-gate * are used to ensure that an interrupt won't overwrite a temporary mapping 18807c478bd9Sstevel@tonic-gate * while it's in use. If an interrupt thread tries to access a PTE, it will 18817c478bd9Sstevel@tonic-gate * yield briefly back to the pinned thread which holds the cpu's hci_mutex. 18827c478bd9Sstevel@tonic-gate */ 18837c478bd9Sstevel@tonic-gate void 1884ae115bc7Smrj x86pte_cpu_init(cpu_t *cpu) 18857c478bd9Sstevel@tonic-gate { 18867c478bd9Sstevel@tonic-gate struct hat_cpu_info *hci; 18877c478bd9Sstevel@tonic-gate 1888ae115bc7Smrj hci = kmem_zalloc(sizeof (*hci), KM_SLEEP); 18897c478bd9Sstevel@tonic-gate mutex_init(&hci->hci_mutex, NULL, MUTEX_DEFAULT, NULL); 18907c478bd9Sstevel@tonic-gate cpu->cpu_hat_info = hci; 18917c478bd9Sstevel@tonic-gate } 18927c478bd9Sstevel@tonic-gate 1893ae115bc7Smrj void 1894ae115bc7Smrj x86pte_cpu_fini(cpu_t *cpu) 1895ae115bc7Smrj { 1896ae115bc7Smrj struct hat_cpu_info *hci = cpu->cpu_hat_info; 1897ae115bc7Smrj 1898ae115bc7Smrj kmem_free(hci, sizeof (*hci)); 1899ae115bc7Smrj cpu->cpu_hat_info = NULL; 19007c478bd9Sstevel@tonic-gate } 19017c478bd9Sstevel@tonic-gate 1902ae115bc7Smrj #ifdef __i386 1903ae115bc7Smrj /* 1904ae115bc7Smrj * On 32 bit kernels, loading a 64 bit PTE is a little tricky 1905ae115bc7Smrj */ 1906ae115bc7Smrj x86pte_t 1907ae115bc7Smrj get_pte64(x86pte_t *ptr) 1908ae115bc7Smrj { 1909ae115bc7Smrj volatile uint32_t *p = (uint32_t *)ptr; 1910ae115bc7Smrj x86pte_t t; 1911ae115bc7Smrj 1912ae115bc7Smrj ASSERT(mmu.pae_hat != 0); 1913ae115bc7Smrj for (;;) { 1914ae115bc7Smrj t = p[0]; 1915ae115bc7Smrj t |= (uint64_t)p[1] << 32; 1916ae115bc7Smrj if ((t & 0xffffffff) == p[0]) 1917ae115bc7Smrj return (t); 1918ae115bc7Smrj } 1919ae115bc7Smrj } 1920ae115bc7Smrj #endif /* __i386 */ 1921ae115bc7Smrj 19227c478bd9Sstevel@tonic-gate /* 19237c478bd9Sstevel@tonic-gate * Disable preemption and establish a mapping to the pagetable with the 19247c478bd9Sstevel@tonic-gate * given pfn. This is optimized for there case where it's the same 19257c478bd9Sstevel@tonic-gate * pfn as we last used referenced from this CPU. 19267c478bd9Sstevel@tonic-gate */ 19277c478bd9Sstevel@tonic-gate static x86pte_t * 1928ae115bc7Smrj x86pte_access_pagetable(htable_t *ht, uint_t index) 19297c478bd9Sstevel@tonic-gate { 19307c478bd9Sstevel@tonic-gate /* 19317c478bd9Sstevel@tonic-gate * VLP pagetables are contained in the hat_t 19327c478bd9Sstevel@tonic-gate */ 19337c478bd9Sstevel@tonic-gate if (ht->ht_flags & HTABLE_VLP) 1934ae115bc7Smrj return (PT_INDEX_PTR(ht->ht_hat->hat_vlp_ptes, index)); 1935ae115bc7Smrj return (x86pte_mapin(ht->ht_pfn, index, ht)); 1936ae115bc7Smrj } 19377c478bd9Sstevel@tonic-gate 19387c478bd9Sstevel@tonic-gate /* 1939ae115bc7Smrj * map the given pfn into the page table window. 19407c478bd9Sstevel@tonic-gate */ 1941ae115bc7Smrj /*ARGSUSED*/ 1942ae115bc7Smrj x86pte_t * 1943ae115bc7Smrj x86pte_mapin(pfn_t pfn, uint_t index, htable_t *ht) 1944ae115bc7Smrj { 1945ae115bc7Smrj x86pte_t *pteptr; 19468ea72728Sjosephb x86pte_t pte = 0; 1947ae115bc7Smrj x86pte_t newpte; 1948ae115bc7Smrj int x; 1949ae115bc7Smrj 19507c478bd9Sstevel@tonic-gate ASSERT(pfn != PFN_INVALID); 19517c478bd9Sstevel@tonic-gate 19527c478bd9Sstevel@tonic-gate if (!khat_running) { 1953ae115bc7Smrj caddr_t va = kbm_remap_window(pfn_to_pa(pfn), 1); 1954ae115bc7Smrj return (PT_INDEX_PTR(va, index)); 19557c478bd9Sstevel@tonic-gate } 19567c478bd9Sstevel@tonic-gate 19577c478bd9Sstevel@tonic-gate /* 1958ae115bc7Smrj * If kpm is available, use it. 1959ae115bc7Smrj */ 1960ae115bc7Smrj if (kpm_vbase) 1961ae115bc7Smrj return (PT_INDEX_PTR(hat_kpm_pfn2va(pfn), index)); 1962ae115bc7Smrj 1963ae115bc7Smrj /* 1964ae115bc7Smrj * Disable preemption and grab the CPU's hci_mutex 19657c478bd9Sstevel@tonic-gate */ 19667c478bd9Sstevel@tonic-gate kpreempt_disable(); 1967ae115bc7Smrj ASSERT(CPU->cpu_hat_info != NULL); 1968ae115bc7Smrj mutex_enter(&CPU->cpu_hat_info->hci_mutex); 1969ae115bc7Smrj x = PWIN_TABLE(CPU->cpu_id); 1970ae115bc7Smrj pteptr = (x86pte_t *)PWIN_PTE_VA(x); 19718ea72728Sjosephb #ifndef __xpv 1972ae115bc7Smrj if (mmu.pae_hat) 1973ae115bc7Smrj pte = *pteptr; 1974ae115bc7Smrj else 1975ae115bc7Smrj pte = *(x86pte32_t *)pteptr; 19768ea72728Sjosephb #endif 1977ae115bc7Smrj 1978ae115bc7Smrj newpte = MAKEPTE(pfn, 0) | mmu.pt_global | mmu.pt_nx; 1979843e1988Sjohnlev 1980843e1988Sjohnlev /* 1981843e1988Sjohnlev * For hardware we can use a writable mapping. 1982843e1988Sjohnlev */ 1983843e1988Sjohnlev #ifdef __xpv 1984843e1988Sjohnlev if (IN_XPV_PANIC()) 1985843e1988Sjohnlev #endif 1986ae115bc7Smrj newpte |= PT_WRITABLE; 1987ae115bc7Smrj 1988ae115bc7Smrj if (!PTE_EQUIV(newpte, pte)) { 1989843e1988Sjohnlev 1990843e1988Sjohnlev #ifdef __xpv 1991843e1988Sjohnlev if (!IN_XPV_PANIC()) { 1992843e1988Sjohnlev xen_map(newpte, PWIN_VA(x)); 1993843e1988Sjohnlev } else 1994843e1988Sjohnlev #endif 1995843e1988Sjohnlev { 1996843e1988Sjohnlev XPV_ALLOW_PAGETABLE_UPDATES(); 1997ae115bc7Smrj if (mmu.pae_hat) 1998ae115bc7Smrj *pteptr = newpte; 1999ae115bc7Smrj else 2000ae115bc7Smrj *(x86pte32_t *)pteptr = newpte; 2001843e1988Sjohnlev XPV_DISALLOW_PAGETABLE_UPDATES(); 2002ae115bc7Smrj mmu_tlbflush_entry((caddr_t)(PWIN_VA(x))); 20037c478bd9Sstevel@tonic-gate } 2004843e1988Sjohnlev } 2005ae115bc7Smrj return (PT_INDEX_PTR(PWIN_VA(x), index)); 20067c478bd9Sstevel@tonic-gate } 20077c478bd9Sstevel@tonic-gate 20087c478bd9Sstevel@tonic-gate /* 20097c478bd9Sstevel@tonic-gate * Release access to a page table. 20107c478bd9Sstevel@tonic-gate */ 20117c478bd9Sstevel@tonic-gate static void 20127c478bd9Sstevel@tonic-gate x86pte_release_pagetable(htable_t *ht) 20137c478bd9Sstevel@tonic-gate { 20147c478bd9Sstevel@tonic-gate /* 20157c478bd9Sstevel@tonic-gate * nothing to do for VLP htables 20167c478bd9Sstevel@tonic-gate */ 20177c478bd9Sstevel@tonic-gate if (ht->ht_flags & HTABLE_VLP) 20187c478bd9Sstevel@tonic-gate return; 20197c478bd9Sstevel@tonic-gate 2020ae115bc7Smrj x86pte_mapout(); 20217c478bd9Sstevel@tonic-gate } 20227c478bd9Sstevel@tonic-gate 2023ae115bc7Smrj void 2024ae115bc7Smrj x86pte_mapout(void) 2025ae115bc7Smrj { 2026843e1988Sjohnlev if (kpm_vbase != NULL || !khat_running) 2027ae115bc7Smrj return; 2028ae115bc7Smrj 20297c478bd9Sstevel@tonic-gate /* 2030ae115bc7Smrj * Drop the CPU's hci_mutex and restore preemption. 20317c478bd9Sstevel@tonic-gate */ 20328ea72728Sjosephb #ifdef __xpv 20338ea72728Sjosephb if (!IN_XPV_PANIC()) { 20348ea72728Sjosephb uintptr_t va; 20358ea72728Sjosephb 20368ea72728Sjosephb /* 20378ea72728Sjosephb * We need to always clear the mapping in case a page 20388ea72728Sjosephb * that was once a page table page is ballooned out. 20398ea72728Sjosephb */ 20408ea72728Sjosephb va = (uintptr_t)PWIN_VA(PWIN_TABLE(CPU->cpu_id)); 20418ea72728Sjosephb (void) HYPERVISOR_update_va_mapping(va, 0, 20428ea72728Sjosephb UVMF_INVLPG | UVMF_LOCAL); 20438ea72728Sjosephb } 20448ea72728Sjosephb #endif 2045ae115bc7Smrj mutex_exit(&CPU->cpu_hat_info->hci_mutex); 20467c478bd9Sstevel@tonic-gate kpreempt_enable(); 20477c478bd9Sstevel@tonic-gate } 20487c478bd9Sstevel@tonic-gate 20497c478bd9Sstevel@tonic-gate /* 20507c478bd9Sstevel@tonic-gate * Atomic retrieval of a pagetable entry 20517c478bd9Sstevel@tonic-gate */ 20527c478bd9Sstevel@tonic-gate x86pte_t 20537c478bd9Sstevel@tonic-gate x86pte_get(htable_t *ht, uint_t entry) 20547c478bd9Sstevel@tonic-gate { 20557c478bd9Sstevel@tonic-gate x86pte_t pte; 2056aa2ed9e5Sjosephb x86pte_t *ptep; 20577c478bd9Sstevel@tonic-gate 20587c478bd9Sstevel@tonic-gate /* 2059aa2ed9e5Sjosephb * Be careful that loading PAE entries in 32 bit kernel is atomic. 20607c478bd9Sstevel@tonic-gate */ 2061ae115bc7Smrj ASSERT(entry < mmu.ptes_per_table); 2062ae115bc7Smrj ptep = x86pte_access_pagetable(ht, entry); 2063ae115bc7Smrj pte = GET_PTE(ptep); 20647c478bd9Sstevel@tonic-gate x86pte_release_pagetable(ht); 20657c478bd9Sstevel@tonic-gate return (pte); 20667c478bd9Sstevel@tonic-gate } 20677c478bd9Sstevel@tonic-gate 20687c478bd9Sstevel@tonic-gate /* 20697c478bd9Sstevel@tonic-gate * Atomic unconditional set of a page table entry, it returns the previous 2070ae115bc7Smrj * value. For pre-existing mappings if the PFN changes, then we don't care 2071ae115bc7Smrj * about the old pte's REF / MOD bits. If the PFN remains the same, we leave 2072ae115bc7Smrj * the MOD/REF bits unchanged. 2073ae115bc7Smrj * 2074ae115bc7Smrj * If asked to overwrite a link to a lower page table with a large page 2075ae115bc7Smrj * mapping, this routine returns the special value of LPAGE_ERROR. This 2076ae115bc7Smrj * allows the upper HAT layers to retry with a smaller mapping size. 20777c478bd9Sstevel@tonic-gate */ 20787c478bd9Sstevel@tonic-gate x86pte_t 20797c478bd9Sstevel@tonic-gate x86pte_set(htable_t *ht, uint_t entry, x86pte_t new, void *ptr) 20807c478bd9Sstevel@tonic-gate { 20817c478bd9Sstevel@tonic-gate x86pte_t old; 2082ae115bc7Smrj x86pte_t prev; 20837c478bd9Sstevel@tonic-gate x86pte_t *ptep; 2084ae115bc7Smrj level_t l = ht->ht_level; 2085ae115bc7Smrj x86pte_t pfn_mask = (l != 0) ? PT_PADDR_LGPG : PT_PADDR; 2086ae115bc7Smrj x86pte_t n; 2087ae115bc7Smrj uintptr_t addr = htable_e2va(ht, entry); 2088ae115bc7Smrj hat_t *hat = ht->ht_hat; 20897c478bd9Sstevel@tonic-gate 2090ae115bc7Smrj ASSERT(new != 0); /* don't use to invalidate a PTE, see x86pte_update */ 20917c478bd9Sstevel@tonic-gate ASSERT(!(ht->ht_flags & HTABLE_SHARED_PFN)); 2092ae115bc7Smrj if (ptr == NULL) 2093ae115bc7Smrj ptep = x86pte_access_pagetable(ht, entry); 2094ae115bc7Smrj else 20957c478bd9Sstevel@tonic-gate ptep = ptr; 20967c478bd9Sstevel@tonic-gate 2097b193e412Skchow /* 2098ae115bc7Smrj * Install the new PTE. If remapping the same PFN, then 2099ae115bc7Smrj * copy existing REF/MOD bits to new mapping. 2100b193e412Skchow */ 2101ae115bc7Smrj do { 2102ae115bc7Smrj prev = GET_PTE(ptep); 2103ae115bc7Smrj n = new; 2104ae115bc7Smrj if (PTE_ISVALID(n) && (prev & pfn_mask) == (new & pfn_mask)) 2105b193e412Skchow n |= prev & (PT_REF | PT_MOD); 2106ae115bc7Smrj 2107ae115bc7Smrj /* 2108ae115bc7Smrj * Another thread may have installed this mapping already, 2109ae115bc7Smrj * flush the local TLB and be done. 2110ae115bc7Smrj */ 2111b193e412Skchow if (prev == n) { 21127c478bd9Sstevel@tonic-gate old = new; 2113843e1988Sjohnlev #ifdef __xpv 2114843e1988Sjohnlev if (!IN_XPV_PANIC()) 2115843e1988Sjohnlev xen_flush_va((caddr_t)addr); 2116843e1988Sjohnlev else 2117843e1988Sjohnlev #endif 2118ae115bc7Smrj mmu_tlbflush_entry((caddr_t)addr); 2119ae115bc7Smrj goto done; 21207c478bd9Sstevel@tonic-gate } 2121ae115bc7Smrj 2122ae115bc7Smrj /* 2123ae115bc7Smrj * Detect if we have a collision of installing a large 2124ae115bc7Smrj * page mapping where there already is a lower page table. 2125ae115bc7Smrj */ 212697704650Sjosephb if (l > 0 && (prev & PT_VALID) && !(prev & PT_PAGESIZE)) { 212797704650Sjosephb old = LPAGE_ERROR; 212897704650Sjosephb goto done; 212997704650Sjosephb } 2130ae115bc7Smrj 2131843e1988Sjohnlev XPV_ALLOW_PAGETABLE_UPDATES(); 2132ae115bc7Smrj old = CAS_PTE(ptep, prev, n); 2133843e1988Sjohnlev XPV_DISALLOW_PAGETABLE_UPDATES(); 2134ae115bc7Smrj } while (old != prev); 2135ae115bc7Smrj 2136ae115bc7Smrj /* 2137ae115bc7Smrj * Do a TLB demap if needed, ie. the old pte was valid. 2138ae115bc7Smrj * 2139ae115bc7Smrj * Note that a stale TLB writeback to the PTE here either can't happen 2140ae115bc7Smrj * or doesn't matter. The PFN can only change for NOSYNC|NOCONSIST 2141ae115bc7Smrj * mappings, but they were created with REF and MOD already set, so 2142ae115bc7Smrj * no stale writeback will happen. 2143ae115bc7Smrj * 2144ae115bc7Smrj * Segmap is the only place where remaps happen on the same pfn and for 2145ae115bc7Smrj * that we want to preserve the stale REF/MOD bits. 2146ae115bc7Smrj */ 2147ae115bc7Smrj if (old & PT_REF) 2148ae115bc7Smrj hat_tlb_inval(hat, addr); 2149ae115bc7Smrj 2150ae115bc7Smrj done: 21517c478bd9Sstevel@tonic-gate if (ptr == NULL) 21527c478bd9Sstevel@tonic-gate x86pte_release_pagetable(ht); 21537c478bd9Sstevel@tonic-gate return (old); 21547c478bd9Sstevel@tonic-gate } 21557c478bd9Sstevel@tonic-gate 21567c478bd9Sstevel@tonic-gate /* 2157ae115bc7Smrj * Atomic compare and swap of a page table entry. No TLB invalidates are done. 2158ae115bc7Smrj * This is used for links between pagetables of different levels. 2159ae115bc7Smrj * Note we always create these links with dirty/access set, so they should 2160ae115bc7Smrj * never change. 21617c478bd9Sstevel@tonic-gate */ 2162ae115bc7Smrj x86pte_t 21637c478bd9Sstevel@tonic-gate x86pte_cas(htable_t *ht, uint_t entry, x86pte_t old, x86pte_t new) 21647c478bd9Sstevel@tonic-gate { 21657c478bd9Sstevel@tonic-gate x86pte_t pte; 21667c478bd9Sstevel@tonic-gate x86pte_t *ptep; 2167843e1988Sjohnlev #ifdef __xpv 2168843e1988Sjohnlev /* 2169843e1988Sjohnlev * We can't use writable pagetables for upper level tables, so fake it. 2170843e1988Sjohnlev */ 2171843e1988Sjohnlev mmu_update_t t[2]; 2172843e1988Sjohnlev int cnt = 1; 2173843e1988Sjohnlev int count; 2174843e1988Sjohnlev maddr_t ma; 21757c478bd9Sstevel@tonic-gate 2176843e1988Sjohnlev if (!IN_XPV_PANIC()) { 2177843e1988Sjohnlev ASSERT(!(ht->ht_flags & HTABLE_VLP)); /* no VLP yet */ 2178843e1988Sjohnlev ma = pa_to_ma(PT_INDEX_PHYSADDR(pfn_to_pa(ht->ht_pfn), entry)); 2179843e1988Sjohnlev t[0].ptr = ma | MMU_NORMAL_PT_UPDATE; 2180843e1988Sjohnlev t[0].val = new; 2181843e1988Sjohnlev 2182843e1988Sjohnlev #if defined(__amd64) 2183843e1988Sjohnlev /* 2184843e1988Sjohnlev * On the 64-bit hypervisor we need to maintain the user mode 2185843e1988Sjohnlev * top page table too. 2186843e1988Sjohnlev */ 2187843e1988Sjohnlev if (ht->ht_level == mmu.max_level && ht->ht_hat != kas.a_hat) { 2188843e1988Sjohnlev ma = pa_to_ma(PT_INDEX_PHYSADDR(pfn_to_pa( 2189843e1988Sjohnlev ht->ht_hat->hat_user_ptable), entry)); 2190843e1988Sjohnlev t[1].ptr = ma | MMU_NORMAL_PT_UPDATE; 2191843e1988Sjohnlev t[1].val = new; 2192843e1988Sjohnlev ++cnt; 2193843e1988Sjohnlev } 2194843e1988Sjohnlev #endif /* __amd64 */ 2195843e1988Sjohnlev 2196843e1988Sjohnlev if (HYPERVISOR_mmu_update(t, cnt, &count, DOMID_SELF)) 2197843e1988Sjohnlev panic("HYPERVISOR_mmu_update() failed"); 2198843e1988Sjohnlev ASSERT(count == cnt); 2199843e1988Sjohnlev return (old); 2200843e1988Sjohnlev } 2201843e1988Sjohnlev #endif 2202ae115bc7Smrj ptep = x86pte_access_pagetable(ht, entry); 2203843e1988Sjohnlev XPV_ALLOW_PAGETABLE_UPDATES(); 2204ae115bc7Smrj pte = CAS_PTE(ptep, old, new); 2205843e1988Sjohnlev XPV_DISALLOW_PAGETABLE_UPDATES(); 22067c478bd9Sstevel@tonic-gate x86pte_release_pagetable(ht); 22077c478bd9Sstevel@tonic-gate return (pte); 22087c478bd9Sstevel@tonic-gate } 22097c478bd9Sstevel@tonic-gate 22107c478bd9Sstevel@tonic-gate /* 2211ae115bc7Smrj * Invalidate a page table entry as long as it currently maps something that 2212ae115bc7Smrj * matches the value determined by expect. 22137c478bd9Sstevel@tonic-gate * 2214*a6a74e0eSMatthew Ahrens * If tlb is set, also invalidates any TLB entries. 2215*a6a74e0eSMatthew Ahrens * 2216*a6a74e0eSMatthew Ahrens * Returns the previous value of the PTE. 22177c478bd9Sstevel@tonic-gate */ 22187c478bd9Sstevel@tonic-gate x86pte_t 2219ae115bc7Smrj x86pte_inval( 2220ae115bc7Smrj htable_t *ht, 2221ae115bc7Smrj uint_t entry, 2222ae115bc7Smrj x86pte_t expect, 2223*a6a74e0eSMatthew Ahrens x86pte_t *pte_ptr, 2224*a6a74e0eSMatthew Ahrens boolean_t tlb) 22257c478bd9Sstevel@tonic-gate { 22267c478bd9Sstevel@tonic-gate x86pte_t *ptep; 222795c0a3c8Sjosephb x86pte_t oldpte; 222895c0a3c8Sjosephb x86pte_t found; 22297c478bd9Sstevel@tonic-gate 22307c478bd9Sstevel@tonic-gate ASSERT(!(ht->ht_flags & HTABLE_SHARED_PFN)); 223102bc52beSkchow ASSERT(ht->ht_level <= mmu.max_page_level); 223297704650Sjosephb 2233ae115bc7Smrj if (pte_ptr != NULL) 22347c478bd9Sstevel@tonic-gate ptep = pte_ptr; 2235ae115bc7Smrj else 2236ae115bc7Smrj ptep = x86pte_access_pagetable(ht, entry); 22377c478bd9Sstevel@tonic-gate 2238843e1988Sjohnlev #if defined(__xpv) 2239843e1988Sjohnlev /* 2240843e1988Sjohnlev * If exit()ing just use HYPERVISOR_mmu_update(), as we can't be racing 2241843e1988Sjohnlev * with anything else. 2242843e1988Sjohnlev */ 2243843e1988Sjohnlev if ((ht->ht_hat->hat_flags & HAT_FREEING) && !IN_XPV_PANIC()) { 2244843e1988Sjohnlev int count; 2245843e1988Sjohnlev mmu_update_t t[1]; 2246843e1988Sjohnlev maddr_t ma; 2247843e1988Sjohnlev 2248843e1988Sjohnlev oldpte = GET_PTE(ptep); 2249843e1988Sjohnlev if (expect != 0 && (oldpte & PT_PADDR) != (expect & PT_PADDR)) 2250843e1988Sjohnlev goto done; 2251843e1988Sjohnlev ma = pa_to_ma(PT_INDEX_PHYSADDR(pfn_to_pa(ht->ht_pfn), entry)); 2252843e1988Sjohnlev t[0].ptr = ma | MMU_NORMAL_PT_UPDATE; 2253843e1988Sjohnlev t[0].val = 0; 2254843e1988Sjohnlev if (HYPERVISOR_mmu_update(t, 1, &count, DOMID_SELF)) 2255843e1988Sjohnlev panic("HYPERVISOR_mmu_update() failed"); 2256843e1988Sjohnlev ASSERT(count == 1); 2257843e1988Sjohnlev goto done; 2258843e1988Sjohnlev } 2259843e1988Sjohnlev #endif /* __xpv */ 2260843e1988Sjohnlev 22617c478bd9Sstevel@tonic-gate /* 226297704650Sjosephb * Note that the loop is needed to handle changes due to h/w updating 226397704650Sjosephb * of PT_MOD/PT_REF. 22647c478bd9Sstevel@tonic-gate */ 2265ae115bc7Smrj do { 226695c0a3c8Sjosephb oldpte = GET_PTE(ptep); 226795c0a3c8Sjosephb if (expect != 0 && (oldpte & PT_PADDR) != (expect & PT_PADDR)) 226895c0a3c8Sjosephb goto done; 2269843e1988Sjohnlev XPV_ALLOW_PAGETABLE_UPDATES(); 227095c0a3c8Sjosephb found = CAS_PTE(ptep, oldpte, 0); 2271843e1988Sjohnlev XPV_DISALLOW_PAGETABLE_UPDATES(); 227295c0a3c8Sjosephb } while (found != oldpte); 2273*a6a74e0eSMatthew Ahrens if (tlb && (oldpte & (PT_REF | PT_MOD))) 227495c0a3c8Sjosephb hat_tlb_inval(ht->ht_hat, htable_e2va(ht, entry)); 22757c478bd9Sstevel@tonic-gate 227695c0a3c8Sjosephb done: 22777c478bd9Sstevel@tonic-gate if (pte_ptr == NULL) 22787c478bd9Sstevel@tonic-gate x86pte_release_pagetable(ht); 227995c0a3c8Sjosephb return (oldpte); 22807c478bd9Sstevel@tonic-gate } 22817c478bd9Sstevel@tonic-gate 22827c478bd9Sstevel@tonic-gate /* 2283ae115bc7Smrj * Change a page table entry af it currently matches the value in expect. 22847c478bd9Sstevel@tonic-gate */ 22857c478bd9Sstevel@tonic-gate x86pte_t 2286ae115bc7Smrj x86pte_update( 2287ae115bc7Smrj htable_t *ht, 2288ae115bc7Smrj uint_t entry, 2289ae115bc7Smrj x86pte_t expect, 2290ae115bc7Smrj x86pte_t new) 22917c478bd9Sstevel@tonic-gate { 22927c478bd9Sstevel@tonic-gate x86pte_t *ptep; 2293ae115bc7Smrj x86pte_t found; 22947c478bd9Sstevel@tonic-gate 2295ae115bc7Smrj ASSERT(new != 0); 22967c478bd9Sstevel@tonic-gate ASSERT(!(ht->ht_flags & HTABLE_SHARED_PFN)); 229702bc52beSkchow ASSERT(ht->ht_level <= mmu.max_page_level); 2298ae115bc7Smrj 2299ae115bc7Smrj ptep = x86pte_access_pagetable(ht, entry); 2300843e1988Sjohnlev XPV_ALLOW_PAGETABLE_UPDATES(); 2301ae115bc7Smrj found = CAS_PTE(ptep, expect, new); 2302843e1988Sjohnlev XPV_DISALLOW_PAGETABLE_UPDATES(); 2303ae115bc7Smrj if (found == expect) { 2304ae115bc7Smrj hat_tlb_inval(ht->ht_hat, htable_e2va(ht, entry)); 23057c478bd9Sstevel@tonic-gate 23067c478bd9Sstevel@tonic-gate /* 2307ae115bc7Smrj * When removing write permission *and* clearing the 2308ae115bc7Smrj * MOD bit, check if a write happened via a stale 2309ae115bc7Smrj * TLB entry before the TLB shootdown finished. 2310ae115bc7Smrj * 2311ae115bc7Smrj * If it did happen, simply re-enable write permission and 2312ae115bc7Smrj * act like the original CAS failed. 23137c478bd9Sstevel@tonic-gate */ 2314ae115bc7Smrj if ((expect & (PT_WRITABLE | PT_MOD)) == PT_WRITABLE && 2315ae115bc7Smrj (new & (PT_WRITABLE | PT_MOD)) == 0 && 2316ae115bc7Smrj (GET_PTE(ptep) & PT_MOD) != 0) { 2317ae115bc7Smrj do { 2318ae115bc7Smrj found = GET_PTE(ptep); 2319843e1988Sjohnlev XPV_ALLOW_PAGETABLE_UPDATES(); 2320ae115bc7Smrj found = 2321ae115bc7Smrj CAS_PTE(ptep, found, found | PT_WRITABLE); 2322843e1988Sjohnlev XPV_DISALLOW_PAGETABLE_UPDATES(); 2323ae115bc7Smrj } while ((found & PT_WRITABLE) == 0); 2324ae115bc7Smrj } 2325ae115bc7Smrj } 23267c478bd9Sstevel@tonic-gate x86pte_release_pagetable(ht); 2327ae115bc7Smrj return (found); 23287c478bd9Sstevel@tonic-gate } 23297c478bd9Sstevel@tonic-gate 2330843e1988Sjohnlev #ifndef __xpv 23317c478bd9Sstevel@tonic-gate /* 23327c478bd9Sstevel@tonic-gate * Copy page tables - this is just a little more complicated than the 23337c478bd9Sstevel@tonic-gate * previous routines. Note that it's also not atomic! It also is never 23347c478bd9Sstevel@tonic-gate * used for VLP pagetables. 23357c478bd9Sstevel@tonic-gate */ 23367c478bd9Sstevel@tonic-gate void 23377c478bd9Sstevel@tonic-gate x86pte_copy(htable_t *src, htable_t *dest, uint_t entry, uint_t count) 23387c478bd9Sstevel@tonic-gate { 23397c478bd9Sstevel@tonic-gate caddr_t src_va; 23407c478bd9Sstevel@tonic-gate caddr_t dst_va; 23417c478bd9Sstevel@tonic-gate size_t size; 2342ae115bc7Smrj x86pte_t *pteptr; 2343ae115bc7Smrj x86pte_t pte; 23447c478bd9Sstevel@tonic-gate 23457c478bd9Sstevel@tonic-gate ASSERT(khat_running); 23467c478bd9Sstevel@tonic-gate ASSERT(!(dest->ht_flags & HTABLE_VLP)); 23477c478bd9Sstevel@tonic-gate ASSERT(!(src->ht_flags & HTABLE_VLP)); 23487c478bd9Sstevel@tonic-gate ASSERT(!(src->ht_flags & HTABLE_SHARED_PFN)); 23497c478bd9Sstevel@tonic-gate ASSERT(!(dest->ht_flags & HTABLE_SHARED_PFN)); 23507c478bd9Sstevel@tonic-gate 23517c478bd9Sstevel@tonic-gate /* 2352ae115bc7Smrj * Acquire access to the CPU pagetable windows for the dest and source. 23537c478bd9Sstevel@tonic-gate */ 2354ae115bc7Smrj dst_va = (caddr_t)x86pte_access_pagetable(dest, entry); 2355ae115bc7Smrj if (kpm_vbase) { 2356ae115bc7Smrj src_va = (caddr_t) 2357ae115bc7Smrj PT_INDEX_PTR(hat_kpm_pfn2va(src->ht_pfn), entry); 23587c478bd9Sstevel@tonic-gate } else { 2359ae115bc7Smrj uint_t x = PWIN_SRC(CPU->cpu_id); 23607c478bd9Sstevel@tonic-gate 23617c478bd9Sstevel@tonic-gate /* 23627c478bd9Sstevel@tonic-gate * Finish defining the src pagetable mapping 23637c478bd9Sstevel@tonic-gate */ 2364ae115bc7Smrj src_va = (caddr_t)PT_INDEX_PTR(PWIN_VA(x), entry); 2365ae115bc7Smrj pte = MAKEPTE(src->ht_pfn, 0) | mmu.pt_global | mmu.pt_nx; 2366ae115bc7Smrj pteptr = (x86pte_t *)PWIN_PTE_VA(x); 2367ae115bc7Smrj if (mmu.pae_hat) 2368ae115bc7Smrj *pteptr = pte; 2369ae115bc7Smrj else 2370ae115bc7Smrj *(x86pte32_t *)pteptr = pte; 2371ae115bc7Smrj mmu_tlbflush_entry((caddr_t)(PWIN_VA(x))); 23727c478bd9Sstevel@tonic-gate } 23737c478bd9Sstevel@tonic-gate 23747c478bd9Sstevel@tonic-gate /* 23757c478bd9Sstevel@tonic-gate * now do the copy 23767c478bd9Sstevel@tonic-gate */ 23777c478bd9Sstevel@tonic-gate size = count << mmu.pte_size_shift; 23787c478bd9Sstevel@tonic-gate bcopy(src_va, dst_va, size); 23797c478bd9Sstevel@tonic-gate 23807c478bd9Sstevel@tonic-gate x86pte_release_pagetable(dest); 23817c478bd9Sstevel@tonic-gate } 23827c478bd9Sstevel@tonic-gate 2383843e1988Sjohnlev #else /* __xpv */ 2384843e1988Sjohnlev 2385843e1988Sjohnlev /* 2386843e1988Sjohnlev * The hypervisor only supports writable pagetables at level 0, so we have 2387843e1988Sjohnlev * to install these 1 by 1 the slow way. 2388843e1988Sjohnlev */ 2389843e1988Sjohnlev void 2390843e1988Sjohnlev x86pte_copy(htable_t *src, htable_t *dest, uint_t entry, uint_t count) 2391843e1988Sjohnlev { 2392843e1988Sjohnlev caddr_t src_va; 2393843e1988Sjohnlev x86pte_t pte; 2394843e1988Sjohnlev 2395843e1988Sjohnlev ASSERT(!IN_XPV_PANIC()); 2396843e1988Sjohnlev src_va = (caddr_t)x86pte_access_pagetable(src, entry); 2397843e1988Sjohnlev while (count) { 2398843e1988Sjohnlev if (mmu.pae_hat) 2399843e1988Sjohnlev pte = *(x86pte_t *)src_va; 2400843e1988Sjohnlev else 2401843e1988Sjohnlev pte = *(x86pte32_t *)src_va; 2402843e1988Sjohnlev if (pte != 0) { 2403843e1988Sjohnlev set_pteval(pfn_to_pa(dest->ht_pfn), entry, 2404843e1988Sjohnlev dest->ht_level, pte); 2405843e1988Sjohnlev #ifdef __amd64 2406843e1988Sjohnlev if (dest->ht_level == mmu.max_level && 2407843e1988Sjohnlev htable_e2va(dest, entry) < HYPERVISOR_VIRT_END) 2408843e1988Sjohnlev set_pteval( 2409843e1988Sjohnlev pfn_to_pa(dest->ht_hat->hat_user_ptable), 2410843e1988Sjohnlev entry, dest->ht_level, pte); 2411843e1988Sjohnlev #endif 2412843e1988Sjohnlev } 2413843e1988Sjohnlev --count; 2414843e1988Sjohnlev ++entry; 2415843e1988Sjohnlev src_va += mmu.pte_size; 2416843e1988Sjohnlev } 2417843e1988Sjohnlev x86pte_release_pagetable(src); 2418843e1988Sjohnlev } 2419843e1988Sjohnlev #endif /* __xpv */ 2420843e1988Sjohnlev 24217c478bd9Sstevel@tonic-gate /* 24227c478bd9Sstevel@tonic-gate * Zero page table entries - Note this doesn't use atomic stores! 24237c478bd9Sstevel@tonic-gate */ 2424ae115bc7Smrj static void 24257c478bd9Sstevel@tonic-gate x86pte_zero(htable_t *dest, uint_t entry, uint_t count) 24267c478bd9Sstevel@tonic-gate { 24277c478bd9Sstevel@tonic-gate caddr_t dst_va; 24287c478bd9Sstevel@tonic-gate size_t size; 2429843e1988Sjohnlev #ifdef __xpv 2430843e1988Sjohnlev int x; 2431843e1988Sjohnlev x86pte_t newpte; 2432843e1988Sjohnlev #endif 24337c478bd9Sstevel@tonic-gate 24347c478bd9Sstevel@tonic-gate /* 24357c478bd9Sstevel@tonic-gate * Map in the page table to be zeroed. 24367c478bd9Sstevel@tonic-gate */ 24377c478bd9Sstevel@tonic-gate ASSERT(!(dest->ht_flags & HTABLE_SHARED_PFN)); 24387c478bd9Sstevel@tonic-gate ASSERT(!(dest->ht_flags & HTABLE_VLP)); 2439ae115bc7Smrj 2440843e1988Sjohnlev /* 2441843e1988Sjohnlev * On the hypervisor we don't use x86pte_access_pagetable() since 2442843e1988Sjohnlev * in this case the page is not pinned yet. 2443843e1988Sjohnlev */ 2444843e1988Sjohnlev #ifdef __xpv 2445843e1988Sjohnlev if (kpm_vbase == NULL) { 2446843e1988Sjohnlev kpreempt_disable(); 2447843e1988Sjohnlev ASSERT(CPU->cpu_hat_info != NULL); 2448843e1988Sjohnlev mutex_enter(&CPU->cpu_hat_info->hci_mutex); 2449843e1988Sjohnlev x = PWIN_TABLE(CPU->cpu_id); 2450843e1988Sjohnlev newpte = MAKEPTE(dest->ht_pfn, 0) | PT_WRITABLE; 2451843e1988Sjohnlev xen_map(newpte, PWIN_VA(x)); 2452843e1988Sjohnlev dst_va = (caddr_t)PT_INDEX_PTR(PWIN_VA(x), entry); 2453843e1988Sjohnlev } else 2454843e1988Sjohnlev #endif 2455ae115bc7Smrj dst_va = (caddr_t)x86pte_access_pagetable(dest, entry); 2456ae115bc7Smrj 24577c478bd9Sstevel@tonic-gate size = count << mmu.pte_size_shift; 2458ae115bc7Smrj ASSERT(size > BLOCKZEROALIGN); 2459ae115bc7Smrj #ifdef __i386 24607417cfdeSKuriakose Kuruvilla if (!is_x86_feature(x86_featureset, X86FSET_SSE2)) 24617c478bd9Sstevel@tonic-gate bzero(dst_va, size); 2462ae115bc7Smrj else 2463ae115bc7Smrj #endif 2464ae115bc7Smrj block_zero_no_xmm(dst_va, size); 2465ae115bc7Smrj 2466843e1988Sjohnlev #ifdef __xpv 2467843e1988Sjohnlev if (kpm_vbase == NULL) { 2468843e1988Sjohnlev xen_map(0, PWIN_VA(x)); 2469843e1988Sjohnlev mutex_exit(&CPU->cpu_hat_info->hci_mutex); 2470843e1988Sjohnlev kpreempt_enable(); 2471843e1988Sjohnlev } else 2472843e1988Sjohnlev #endif 24737c478bd9Sstevel@tonic-gate x86pte_release_pagetable(dest); 24747c478bd9Sstevel@tonic-gate } 24757c478bd9Sstevel@tonic-gate 24767c478bd9Sstevel@tonic-gate /* 24777c478bd9Sstevel@tonic-gate * Called to ensure that all pagetables are in the system dump 24787c478bd9Sstevel@tonic-gate */ 24797c478bd9Sstevel@tonic-gate void 24807c478bd9Sstevel@tonic-gate hat_dump(void) 24817c478bd9Sstevel@tonic-gate { 24827c478bd9Sstevel@tonic-gate hat_t *hat; 24837c478bd9Sstevel@tonic-gate uint_t h; 24847c478bd9Sstevel@tonic-gate htable_t *ht; 24857c478bd9Sstevel@tonic-gate 24867c478bd9Sstevel@tonic-gate /* 2487a85a6733Sjosephb * Dump all page tables 24887c478bd9Sstevel@tonic-gate */ 2489a85a6733Sjosephb for (hat = kas.a_hat; hat != NULL; hat = hat->hat_next) { 24907c478bd9Sstevel@tonic-gate for (h = 0; h < hat->hat_num_hash; ++h) { 24917c478bd9Sstevel@tonic-gate for (ht = hat->hat_ht_hash[h]; ht; ht = ht->ht_next) { 2492a85a6733Sjosephb if ((ht->ht_flags & HTABLE_VLP) == 0) 24937c478bd9Sstevel@tonic-gate dump_page(ht->ht_pfn); 24947c478bd9Sstevel@tonic-gate } 24957c478bd9Sstevel@tonic-gate } 24967c478bd9Sstevel@tonic-gate } 24977c478bd9Sstevel@tonic-gate } 2498